Commit 6ae7d6f0 authored by Linus Torvalds's avatar Linus Torvalds

Merge git://git.kernel.org/pub/scm/linux/kernel/git/rusty/linux-2.6-for-linus

* git://git.kernel.org/pub/scm/linux/kernel/git/rusty/linux-2.6-for-linus:
  lguest and virtio: cleanup struct definitions to Linux style.
  lguest: update commentry
  lguest: fix comment style
  virtio: refactor find_vqs
  virtio: delete vq from list
  virtio: fix memory leak on device removal
  lguest: fix descriptor corruption in example launcher
  lguest: dereferencing freed mem in add_eventfd()
parents ec30c5f3 1842f23c
/*P:100 This is the Launcher code, a simple program which lays out the
* "physical" memory for the new Guest by mapping the kernel image and
* the virtual devices, then opens /dev/lguest to tell the kernel
* about the Guest and control it. :*/
/*P:100
* This is the Launcher code, a simple program which lays out the "physical"
* memory for the new Guest by mapping the kernel image and the virtual
* devices, then opens /dev/lguest to tell the kernel about the Guest and
* control it.
:*/
#define _LARGEFILE64_SOURCE
#define _GNU_SOURCE
#include <stdio.h>
......@@ -46,13 +48,15 @@
#include "linux/virtio_rng.h"
#include "linux/virtio_ring.h"
#include "asm/bootparam.h"
/*L:110 We can ignore the 39 include files we need for this program, but I do
* want to draw attention to the use of kernel-style types.
/*L:110
* We can ignore the 42 include files we need for this program, but I do want
* to draw attention to the use of kernel-style types.
*
* As Linus said, "C is a Spartan language, and so should your naming be." I
* like these abbreviations, so we define them here. Note that u64 is always
* unsigned long long, which works on all Linux systems: this means that we can
* use %llu in printf for any u64. */
* use %llu in printf for any u64.
*/
typedef unsigned long long u64;
typedef uint32_t u32;
typedef uint16_t u16;
......@@ -69,8 +73,10 @@ typedef uint8_t u8;
/* This will occupy 3 pages: it must be a power of 2. */
#define VIRTQUEUE_NUM 256
/*L:120 verbose is both a global flag and a macro. The C preprocessor allows
* this, and although I wouldn't recommend it, it works quite nicely here. */
/*L:120
* verbose is both a global flag and a macro. The C preprocessor allows
* this, and although I wouldn't recommend it, it works quite nicely here.
*/
static bool verbose;
#define verbose(args...) \
do { if (verbose) printf(args); } while(0)
......@@ -87,8 +93,7 @@ static int lguest_fd;
static unsigned int __thread cpu_id;
/* This is our list of devices. */
struct device_list
{
struct device_list {
/* Counter to assign interrupt numbers. */
unsigned int next_irq;
......@@ -100,8 +105,7 @@ struct device_list
/* A single linked list of devices. */
struct device *dev;
/* And a pointer to the last device for easy append and also for
* configuration appending. */
/* And a pointer to the last device for easy append. */
struct device *lastdev;
};
......@@ -109,8 +113,7 @@ struct device_list
static struct device_list devices;
/* The device structure describes a single device. */
struct device
{
struct device {
/* The linked-list pointer. */
struct device *next;
......@@ -135,8 +138,7 @@ struct device
};
/* The virtqueue structure describes a queue attached to a device. */
struct virtqueue
{
struct virtqueue {
struct virtqueue *next;
/* Which device owns me. */
......@@ -168,20 +170,24 @@ static char **main_args;
/* The original tty settings to restore on exit. */
static struct termios orig_term;
/* We have to be careful with barriers: our devices are all run in separate
/*
* We have to be careful with barriers: our devices are all run in separate
* threads and so we need to make sure that changes visible to the Guest happen
* in precise order. */
* in precise order.
*/
#define wmb() __asm__ __volatile__("" : : : "memory")
#define mb() __asm__ __volatile__("" : : : "memory")
/* Convert an iovec element to the given type.
/*
* Convert an iovec element to the given type.
*
* This is a fairly ugly trick: we need to know the size of the type and
* alignment requirement to check the pointer is kosher. It's also nice to
* have the name of the type in case we report failure.
*
* Typing those three things all the time is cumbersome and error prone, so we
* have a macro which sets them all up and passes to the real function. */
* have a macro which sets them all up and passes to the real function.
*/
#define convert(iov, type) \
((type *)_convert((iov), sizeof(type), __alignof__(type), #type))
......@@ -198,8 +204,10 @@ static void *_convert(struct iovec *iov, size_t size, size_t align,
/* Wrapper for the last available index. Makes it easier to change. */
#define lg_last_avail(vq) ((vq)->last_avail_idx)
/* The virtio configuration space is defined to be little-endian. x86 is
* little-endian too, but it's nice to be explicit so we have these helpers. */
/*
* The virtio configuration space is defined to be little-endian. x86 is
* little-endian too, but it's nice to be explicit so we have these helpers.
*/
#define cpu_to_le16(v16) (v16)
#define cpu_to_le32(v32) (v32)
#define cpu_to_le64(v64) (v64)
......@@ -241,11 +249,12 @@ static u8 *get_feature_bits(struct device *dev)
+ dev->num_vq * sizeof(struct lguest_vqconfig);
}
/*L:100 The Launcher code itself takes us out into userspace, that scary place
* where pointers run wild and free! Unfortunately, like most userspace
* programs, it's quite boring (which is why everyone likes to hack on the
* kernel!). Perhaps if you make up an Lguest Drinking Game at this point, it
* will get you through this section. Or, maybe not.
/*L:100
* The Launcher code itself takes us out into userspace, that scary place where
* pointers run wild and free! Unfortunately, like most userspace programs,
* it's quite boring (which is why everyone likes to hack on the kernel!).
* Perhaps if you make up an Lguest Drinking Game at this point, it will get
* you through this section. Or, maybe not.
*
* The Launcher sets up a big chunk of memory to be the Guest's "physical"
* memory and stores it in "guest_base". In other words, Guest physical ==
......@@ -253,7 +262,8 @@ static u8 *get_feature_bits(struct device *dev)
*
* This can be tough to get your head around, but usually it just means that we
* use these trivial conversion functions when the Guest gives us it's
* "physical" addresses: */
* "physical" addresses:
*/
static void *from_guest_phys(unsigned long addr)
{
return guest_base + addr;
......@@ -268,7 +278,8 @@ static unsigned long to_guest_phys(const void *addr)
* Loading the Kernel.
*
* We start with couple of simple helper routines. open_or_die() avoids
* error-checking code cluttering the callers: */
* error-checking code cluttering the callers:
*/
static int open_or_die(const char *name, int flags)
{
int fd = open(name, flags);
......@@ -283,12 +294,19 @@ static void *map_zeroed_pages(unsigned int num)
int fd = open_or_die("/dev/zero", O_RDONLY);
void *addr;
/* We use a private mapping (ie. if we write to the page, it will be
* copied). */
/*
* We use a private mapping (ie. if we write to the page, it will be
* copied).
*/
addr = mmap(NULL, getpagesize() * num,
PROT_READ|PROT_WRITE|PROT_EXEC, MAP_PRIVATE, fd, 0);
if (addr == MAP_FAILED)
err(1, "Mmaping %u pages of /dev/zero", num);
/*
* One neat mmap feature is that you can close the fd, and it
* stays mapped.
*/
close(fd);
return addr;
......@@ -305,20 +323,24 @@ static void *get_pages(unsigned int num)
return addr;
}
/* This routine is used to load the kernel or initrd. It tries mmap, but if
/*
* This routine is used to load the kernel or initrd. It tries mmap, but if
* that fails (Plan 9's kernel file isn't nicely aligned on page boundaries),
* it falls back to reading the memory in. */
* it falls back to reading the memory in.
*/
static void map_at(int fd, void *addr, unsigned long offset, unsigned long len)
{
ssize_t r;
/* We map writable even though for some segments are marked read-only.
/*
* We map writable even though for some segments are marked read-only.
* The kernel really wants to be writable: it patches its own
* instructions.
*
* MAP_PRIVATE means that the page won't be copied until a write is
* done to it. This allows us to share untouched memory between
* Guests. */
* Guests.
*/
if (mmap(addr, len, PROT_READ|PROT_WRITE|PROT_EXEC,
MAP_FIXED|MAP_PRIVATE, fd, offset) != MAP_FAILED)
return;
......@@ -329,7 +351,8 @@ static void map_at(int fd, void *addr, unsigned long offset, unsigned long len)
err(1, "Reading offset %lu len %lu gave %zi", offset, len, r);
}
/* This routine takes an open vmlinux image, which is in ELF, and maps it into
/*
* This routine takes an open vmlinux image, which is in ELF, and maps it into
* the Guest memory. ELF = Embedded Linking Format, which is the format used
* by all modern binaries on Linux including the kernel.
*
......@@ -337,23 +360,28 @@ static void map_at(int fd, void *addr, unsigned long offset, unsigned long len)
* address. We use the physical address; the Guest will map itself to the
* virtual address.
*
* We return the starting address. */
* We return the starting address.
*/
static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr)
{
Elf32_Phdr phdr[ehdr->e_phnum];
unsigned int i;
/* Sanity checks on the main ELF header: an x86 executable with a
* reasonable number of correctly-sized program headers. */
/*
* Sanity checks on the main ELF header: an x86 executable with a
* reasonable number of correctly-sized program headers.
*/
if (ehdr->e_type != ET_EXEC
|| ehdr->e_machine != EM_386
|| ehdr->e_phentsize != sizeof(Elf32_Phdr)
|| ehdr->e_phnum < 1 || ehdr->e_phnum > 65536U/sizeof(Elf32_Phdr))
errx(1, "Malformed elf header");
/* An ELF executable contains an ELF header and a number of "program"
/*
* An ELF executable contains an ELF header and a number of "program"
* headers which indicate which parts ("segments") of the program to
* load where. */
* load where.
*/
/* We read in all the program headers at once: */
if (lseek(elf_fd, ehdr->e_phoff, SEEK_SET) < 0)
......@@ -361,8 +389,10 @@ static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr)
if (read(elf_fd, phdr, sizeof(phdr)) != sizeof(phdr))
err(1, "Reading program headers");
/* Try all the headers: there are usually only three. A read-only one,
* a read-write one, and a "note" section which we don't load. */
/*
* Try all the headers: there are usually only three. A read-only one,
* a read-write one, and a "note" section which we don't load.
*/
for (i = 0; i < ehdr->e_phnum; i++) {
/* If this isn't a loadable segment, we ignore it */
if (phdr[i].p_type != PT_LOAD)
......@@ -380,13 +410,15 @@ static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr)
return ehdr->e_entry;
}
/*L:150 A bzImage, unlike an ELF file, is not meant to be loaded. You're
* supposed to jump into it and it will unpack itself. We used to have to
* perform some hairy magic because the unpacking code scared me.
/*L:150
* A bzImage, unlike an ELF file, is not meant to be loaded. You're supposed
* to jump into it and it will unpack itself. We used to have to perform some
* hairy magic because the unpacking code scared me.
*
* Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote
* a small patch to jump over the tricky bits in the Guest, so now we just read
* the funky header so we know where in the file to load, and away we go! */
* the funky header so we know where in the file to load, and away we go!
*/
static unsigned long load_bzimage(int fd)
{
struct boot_params boot;
......@@ -394,8 +426,10 @@ static unsigned long load_bzimage(int fd)
/* Modern bzImages get loaded at 1M. */
void *p = from_guest_phys(0x100000);
/* Go back to the start of the file and read the header. It should be
* a Linux boot header (see Documentation/x86/i386/boot.txt) */
/*
* Go back to the start of the file and read the header. It should be
* a Linux boot header (see Documentation/x86/i386/boot.txt)
*/
lseek(fd, 0, SEEK_SET);
read(fd, &boot, sizeof(boot));
......@@ -414,9 +448,11 @@ static unsigned long load_bzimage(int fd)
return boot.hdr.code32_start;
}
/*L:140 Loading the kernel is easy when it's a "vmlinux", but most kernels
/*L:140
* Loading the kernel is easy when it's a "vmlinux", but most kernels
* come wrapped up in the self-decompressing "bzImage" format. With a little
* work, we can load those, too. */
* work, we can load those, too.
*/
static unsigned long load_kernel(int fd)
{
Elf32_Ehdr hdr;
......@@ -433,24 +469,28 @@ static unsigned long load_kernel(int fd)
return load_bzimage(fd);
}
/* This is a trivial little helper to align pages. Andi Kleen hated it because
/*
* This is a trivial little helper to align pages. Andi Kleen hated it because
* it calls getpagesize() twice: "it's dumb code."
*
* Kernel guys get really het up about optimization, even when it's not
* necessary. I leave this code as a reaction against that. */
* necessary. I leave this code as a reaction against that.
*/
static inline unsigned long page_align(unsigned long addr)
{
/* Add upwards and truncate downwards. */
return ((addr + getpagesize()-1) & ~(getpagesize()-1));
}
/*L:180 An "initial ram disk" is a disk image loaded into memory along with
* the kernel which the kernel can use to boot from without needing any
* drivers. Most distributions now use this as standard: the initrd contains
* the code to load the appropriate driver modules for the current machine.
/*L:180
* An "initial ram disk" is a disk image loaded into memory along with the
* kernel which the kernel can use to boot from without needing any drivers.
* Most distributions now use this as standard: the initrd contains the code to
* load the appropriate driver modules for the current machine.
*
* Importantly, James Morris works for RedHat, and Fedora uses initrds for its
* kernels. He sent me this (and tells me when I break it). */
* kernels. He sent me this (and tells me when I break it).
*/
static unsigned long load_initrd(const char *name, unsigned long mem)
{
int ifd;
......@@ -462,12 +502,16 @@ static unsigned long load_initrd(const char *name, unsigned long mem)
if (fstat(ifd, &st) < 0)
err(1, "fstat() on initrd '%s'", name);
/* We map the initrd at the top of memory, but mmap wants it to be
* page-aligned, so we round the size up for that. */
/*
* We map the initrd at the top of memory, but mmap wants it to be
* page-aligned, so we round the size up for that.
*/
len = page_align(st.st_size);
map_at(ifd, from_guest_phys(mem - len), 0, st.st_size);
/* Once a file is mapped, you can close the file descriptor. It's a
* little odd, but quite useful. */
/*
* Once a file is mapped, you can close the file descriptor. It's a
* little odd, but quite useful.
*/
close(ifd);
verbose("mapped initrd %s size=%lu @ %p\n", name, len, (void*)mem-len);
......@@ -476,8 +520,10 @@ static unsigned long load_initrd(const char *name, unsigned long mem)
}
/*:*/
/* Simple routine to roll all the commandline arguments together with spaces
* between them. */
/*
* Simple routine to roll all the commandline arguments together with spaces
* between them.
*/
static void concat(char *dst, char *args[])
{
unsigned int i, len = 0;
......@@ -494,10 +540,12 @@ static void concat(char *dst, char *args[])
dst[len] = '\0';
}
/*L:185 This is where we actually tell the kernel to initialize the Guest. We
/*L:185
* This is where we actually tell the kernel to initialize the Guest. We
* saw the arguments it expects when we looked at initialize() in lguest_user.c:
* the base of Guest "physical" memory, the top physical page to allow and the
* entry point for the Guest. */
* entry point for the Guest.
*/
static void tell_kernel(unsigned long start)
{
unsigned long args[] = { LHREQ_INITIALIZE,
......@@ -511,7 +559,7 @@ static void tell_kernel(unsigned long start)
}
/*:*/
/*
/*L:200
* Device Handling.
*
* When the Guest gives us a buffer, it sends an array of addresses and sizes.
......@@ -522,20 +570,26 @@ static void tell_kernel(unsigned long start)
static void *_check_pointer(unsigned long addr, unsigned int size,
unsigned int line)
{
/* We have to separately check addr and addr+size, because size could
* be huge and addr + size might wrap around. */
/*
* We have to separately check addr and addr+size, because size could
* be huge and addr + size might wrap around.
*/
if (addr >= guest_limit || addr + size >= guest_limit)
errx(1, "%s:%i: Invalid address %#lx", __FILE__, line, addr);
/* We return a pointer for the caller's convenience, now we know it's
* safe to use. */
/*
* We return a pointer for the caller's convenience, now we know it's
* safe to use.
*/
return from_guest_phys(addr);
}
/* A macro which transparently hands the line number to the real function. */
#define check_pointer(addr,size) _check_pointer(addr, size, __LINE__)
/* Each buffer in the virtqueues is actually a chain of descriptors. This
/*
* Each buffer in the virtqueues is actually a chain of descriptors. This
* function returns the next descriptor in the chain, or vq->vring.num if we're
* at the end. */
* at the end.
*/
static unsigned next_desc(struct vring_desc *desc,
unsigned int i, unsigned int max)
{
......@@ -556,7 +610,10 @@ static unsigned next_desc(struct vring_desc *desc,
return next;
}
/* This actually sends the interrupt for this virtqueue */
/*
* This actually sends the interrupt for this virtqueue, if we've used a
* buffer.
*/
static void trigger_irq(struct virtqueue *vq)
{
unsigned long buf[] = { LHREQ_IRQ, vq->config.irq };
......@@ -576,12 +633,14 @@ static void trigger_irq(struct virtqueue *vq)
err(1, "Triggering irq %i", vq->config.irq);
}
/* This looks in the virtqueue and for the first available buffer, and converts
/*
* This looks in the virtqueue for the first available buffer, and converts
* it to an iovec for convenient access. Since descriptors consist of some
* number of output then some number of input descriptors, it's actually two
* iovecs, but we pack them into one and note how many of each there were.
*
* This function returns the descriptor number found. */
* This function waits if necessary, and returns the descriptor number found.
*/
static unsigned wait_for_vq_desc(struct virtqueue *vq,
struct iovec iov[],
unsigned int *out_num, unsigned int *in_num)
......@@ -590,17 +649,23 @@ static unsigned wait_for_vq_desc(struct virtqueue *vq,
struct vring_desc *desc;
u16 last_avail = lg_last_avail(vq);
/* There's nothing available? */
while (last_avail == vq->vring.avail->idx) {
u64 event;
/* OK, tell Guest about progress up to now. */
/*
* Since we're about to sleep, now is a good time to tell the
* Guest about what we've used up to now.
*/
trigger_irq(vq);
/* OK, now we need to know about added descriptors. */
vq->vring.used->flags &= ~VRING_USED_F_NO_NOTIFY;
/* They could have slipped one in as we were doing that: make
* sure it's written, then check again. */
/*
* They could have slipped one in as we were doing that: make
* sure it's written, then check again.
*/
mb();
if (last_avail != vq->vring.avail->idx) {
vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY;
......@@ -620,8 +685,10 @@ static unsigned wait_for_vq_desc(struct virtqueue *vq,
errx(1, "Guest moved used index from %u to %u",
last_avail, vq->vring.avail->idx);
/* Grab the next descriptor number they're advertising, and increment
* the index we've seen. */
/*
* Grab the next descriptor number they're advertising, and increment
* the index we've seen.
*/
head = vq->vring.avail->ring[last_avail % vq->vring.num];
lg_last_avail(vq)++;
......@@ -636,8 +703,10 @@ static unsigned wait_for_vq_desc(struct virtqueue *vq,
desc = vq->vring.desc;
i = head;
/* If this is an indirect entry, then this buffer contains a descriptor
* table which we handle as if it's any normal descriptor chain. */
/*
* If this is an indirect entry, then this buffer contains a descriptor
* table which we handle as if it's any normal descriptor chain.
*/
if (desc[i].flags & VRING_DESC_F_INDIRECT) {
if (desc[i].len % sizeof(struct vring_desc))
errx(1, "Invalid size for indirect buffer table");
......@@ -656,8 +725,10 @@ static unsigned wait_for_vq_desc(struct virtqueue *vq,
if (desc[i].flags & VRING_DESC_F_WRITE)
(*in_num)++;
else {
/* If it's an output descriptor, they're all supposed
* to come before any input descriptors. */
/*
* If it's an output descriptor, they're all supposed
* to come before any input descriptors.
*/
if (*in_num)
errx(1, "Descriptor has out after in");
(*out_num)++;
......@@ -671,14 +742,19 @@ static unsigned wait_for_vq_desc(struct virtqueue *vq,
return head;
}
/* After we've used one of their buffers, we tell them about it. We'll then
* want to send them an interrupt, using trigger_irq(). */
/*
* After we've used one of their buffers, we tell the Guest about it. Sometime
* later we'll want to send them an interrupt using trigger_irq(); note that
* wait_for_vq_desc() does that for us if it has to wait.
*/
static void add_used(struct virtqueue *vq, unsigned int head, int len)
{
struct vring_used_elem *used;
/* The virtqueue contains a ring of used buffers. Get a pointer to the
* next entry in that used ring. */
/*
* The virtqueue contains a ring of used buffers. Get a pointer to the
* next entry in that used ring.
*/
used = &vq->vring.used->ring[vq->vring.used->idx % vq->vring.num];
used->id = head;
used->len = len;
......@@ -698,9 +774,9 @@ static void add_used_and_trigger(struct virtqueue *vq, unsigned head, int len)
/*
* The Console
*
* We associate some data with the console for our exit hack. */
struct console_abort
{
* We associate some data with the console for our exit hack.
*/
struct console_abort {
/* How many times have they hit ^C? */
int count;
/* When did they start? */
......@@ -715,30 +791,35 @@ static void console_input(struct virtqueue *vq)
struct console_abort *abort = vq->dev->priv;
struct iovec iov[vq->vring.num];
/* Make sure there's a descriptor waiting. */
/* Make sure there's a descriptor available. */
head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
if (out_num)
errx(1, "Output buffers in console in queue?");
/* Read it in. */
/* Read into it. This is where we usually wait. */
len = readv(STDIN_FILENO, iov, in_num);
if (len <= 0) {
/* Ran out of input? */
warnx("Failed to get console input, ignoring console.");
/* For simplicity, dying threads kill the whole Launcher. So
* just nap here. */
/*
* For simplicity, dying threads kill the whole Launcher. So
* just nap here.
*/
for (;;)
pause();
}
/* Tell the Guest we used a buffer. */
add_used_and_trigger(vq, head, len);
/* Three ^C within one second? Exit.
/*
* Three ^C within one second? Exit.
*
* This is such a hack, but works surprisingly well. Each ^C has to
* be in a buffer by itself, so they can't be too fast. But we check
* that we get three within about a second, so they can't be too
* slow. */
* slow.
*/
if (len != 1 || ((char *)iov[0].iov_base)[0] != 3) {
abort->count = 0;
return;
......@@ -763,15 +844,23 @@ static void console_output(struct virtqueue *vq)
unsigned int head, out, in;
struct iovec iov[vq->vring.num];
/* We usually wait in here, for the Guest to give us something. */
head = wait_for_vq_desc(vq, iov, &out, &in);
if (in)
errx(1, "Input buffers in console output queue?");
/* writev can return a partial write, so we loop here. */
while (!iov_empty(iov, out)) {
int len = writev(STDOUT_FILENO, iov, out);
if (len <= 0)
err(1, "Write to stdout gave %i", len);
iov_consume(iov, out, len);
}
/*
* We're finished with that buffer: if we're going to sleep,
* wait_for_vq_desc() will prod the Guest with an interrupt.
*/
add_used(vq, head, 0);
}
......@@ -791,15 +880,30 @@ static void net_output(struct virtqueue *vq)
unsigned int head, out, in;
struct iovec iov[vq->vring.num];
/* We usually wait in here for the Guest to give us a packet. */
head = wait_for_vq_desc(vq, iov, &out, &in);
if (in)
errx(1, "Input buffers in net output queue?");
/*
* Send the whole thing through to /dev/net/tun. It expects the exact
* same format: what a coincidence!
*/
if (writev(net_info->tunfd, iov, out) < 0)
errx(1, "Write to tun failed?");
/*
* Done with that one; wait_for_vq_desc() will send the interrupt if
* all packets are processed.
*/
add_used(vq, head, 0);
}
/* Will reading from this file descriptor block? */
/*
* Handling network input is a bit trickier, because I've tried to optimize it.
*
* First we have a helper routine which tells is if from this file descriptor
* (ie. the /dev/net/tun device) will block:
*/
static bool will_block(int fd)
{
fd_set fdset;
......@@ -809,8 +913,11 @@ static bool will_block(int fd)
return select(fd+1, &fdset, NULL, NULL, &zero) != 1;
}
/* This is where we handle packets coming in from the tun device to our
* Guest. */
/*
* This handles packets coming in from the tun device to our Guest. Like all
* service routines, it gets called again as soon as it returns, so you don't
* see a while(1) loop here.
*/
static void net_input(struct virtqueue *vq)
{
int len;
......@@ -818,21 +925,38 @@ static void net_input(struct virtqueue *vq)
struct iovec iov[vq->vring.num];
struct net_info *net_info = vq->dev->priv;
/*
* Get a descriptor to write an incoming packet into. This will also
* send an interrupt if they're out of descriptors.
*/
head = wait_for_vq_desc(vq, iov, &out, &in);
if (out)
errx(1, "Output buffers in net input queue?");
/* Deliver interrupt now, since we're about to sleep. */
/*
* If it looks like we'll block reading from the tun device, send them
* an interrupt.
*/
if (vq->pending_used && will_block(net_info->tunfd))
trigger_irq(vq);
/*
* Read in the packet. This is where we normally wait (when there's no
* incoming network traffic).
*/
len = readv(net_info->tunfd, iov, in);
if (len <= 0)
err(1, "Failed to read from tun.");
/*
* Mark that packet buffer as used, but don't interrupt here. We want
* to wait until we've done as much work as we can.
*/
add_used(vq, head, len);
}
/*:*/
/* This is the helper to create threads. */
/* This is the helper to create threads: run the service routine in a loop. */
static int do_thread(void *_vq)
{
struct virtqueue *vq = _vq;
......@@ -842,8 +966,10 @@ static int do_thread(void *_vq)
return 0;
}
/* When a child dies, we kill our entire process group with SIGTERM. This
* also has the side effect that the shell restores the console for us! */
/*
* When a child dies, we kill our entire process group with SIGTERM. This
* also has the side effect that the shell restores the console for us!
*/
static void kill_launcher(int signal)
{
kill(0, SIGTERM);
......@@ -878,11 +1004,15 @@ static void reset_device(struct device *dev)
signal(SIGCHLD, (void *)kill_launcher);
}
/*L:216
* This actually creates the thread which services the virtqueue for a device.
*/
static void create_thread(struct virtqueue *vq)
{
/* Create stack for thread and run it. Since stack grows
* upwards, we point the stack pointer to the end of this
* region. */
/*
* Create stack for thread. Since the stack grows upwards, we point
* the stack pointer to the end of this region.
*/
char *stack = malloc(32768);
unsigned long args[] = { LHREQ_EVENTFD,
vq->config.pfn*getpagesize(), 0 };
......@@ -893,17 +1023,22 @@ static void create_thread(struct virtqueue *vq)
err(1, "Creating eventfd");
args[2] = vq->eventfd;
/* Attach an eventfd to this virtqueue: it will go off
* when the Guest does an LHCALL_NOTIFY for this vq. */
/*
* Attach an eventfd to this virtqueue: it will go off when the Guest
* does an LHCALL_NOTIFY for this vq.
*/
if (write(lguest_fd, &args, sizeof(args)) != 0)
err(1, "Attaching eventfd");
/* CLONE_VM: because it has to access the Guest memory, and
* SIGCHLD so we get a signal if it dies. */
/*
* CLONE_VM: because it has to access the Guest memory, and SIGCHLD so
* we get a signal if it dies.
*/
vq->thread = clone(do_thread, stack + 32768, CLONE_VM | SIGCHLD, vq);
if (vq->thread == (pid_t)-1)
err(1, "Creating clone");
/* We close our local copy, now the child has it. */
/* We close our local copy now the child has it. */
close(vq->eventfd);
}
......@@ -955,7 +1090,10 @@ static void update_device_status(struct device *dev)
}
}
/* This is the generic routine we call when the Guest uses LHCALL_NOTIFY. */
/*L:215
* This is the generic routine we call when the Guest uses LHCALL_NOTIFY. In
* particular, it's used to notify us of device status changes during boot.
*/
static void handle_output(unsigned long addr)
{
struct device *i;
......@@ -964,25 +1102,42 @@ static void handle_output(unsigned long addr)
for (i = devices.dev; i; i = i->next) {
struct virtqueue *vq;
/* Notifications to device descriptors update device status. */
/*
* Notifications to device descriptors mean they updated the
* device status.
*/
if (from_guest_phys(addr) == i->desc) {
update_device_status(i);
return;
}
/* Devices *can* be used before status is set to DRIVER_OK. */
/*
* Devices *can* be used before status is set to DRIVER_OK.
* The original plan was that they would never do this: they
* would always finish setting up their status bits before
* actually touching the virtqueues. In practice, we allowed
* them to, and they do (eg. the disk probes for partition
* tables as part of initialization).
*
* If we see this, we start the device: once it's running, we
* expect the device to catch all the notifications.
*/
for (vq = i->vq; vq; vq = vq->next) {
if (addr != vq->config.pfn*getpagesize())
continue;
if (i->running)
errx(1, "Notification on running %s", i->name);
/* This just calls create_thread() for each virtqueue */
start_device(i);
return;
}
}
/* Early console write is done using notify on a nul-terminated string
* in Guest memory. */
/*
* Early console write is done using notify on a nul-terminated string
* in Guest memory. It's also great for hacking debugging messages
* into a Guest.
*/
if (addr >= guest_limit)
errx(1, "Bad NOTIFY %#lx", addr);
......@@ -998,10 +1153,12 @@ static void handle_output(unsigned long addr)
* routines to allocate and manage them.
*/
/* The layout of the device page is a "struct lguest_device_desc" followed by a
/*
* The layout of the device page is a "struct lguest_device_desc" followed by a
* number of virtqueue descriptors, then two sets of feature bits, then an
* array of configuration bytes. This routine returns the configuration
* pointer. */
* pointer.
*/
static u8 *device_config(const struct device *dev)
{
return (void *)(dev->desc + 1)
......@@ -1009,9 +1166,11 @@ static u8 *device_config(const struct device *dev)
+ dev->feature_len * 2;
}
/* This routine allocates a new "struct lguest_device_desc" from descriptor
/*
* This routine allocates a new "struct lguest_device_desc" from descriptor
* table page just above the Guest's normal memory. It returns a pointer to
* that descriptor. */
* that descriptor.
*/
static struct lguest_device_desc *new_dev_desc(u16 type)
{
struct lguest_device_desc d = { .type = type };
......@@ -1032,8 +1191,10 @@ static struct lguest_device_desc *new_dev_desc(u16 type)
return memcpy(p, &d, sizeof(d));
}
/* Each device descriptor is followed by the description of its virtqueues. We
* specify how many descriptors the virtqueue is to have. */
/*
* Each device descriptor is followed by the description of its virtqueues. We
* specify how many descriptors the virtqueue is to have.
*/
static void add_virtqueue(struct device *dev, unsigned int num_descs,
void (*service)(struct virtqueue *))
{
......@@ -1050,6 +1211,11 @@ static void add_virtqueue(struct device *dev, unsigned int num_descs,
vq->next = NULL;
vq->last_avail_idx = 0;
vq->dev = dev;
/*
* This is the routine the service thread will run, and its Process ID
* once it's running.
*/
vq->service = service;
vq->thread = (pid_t)-1;
......@@ -1061,10 +1227,12 @@ static void add_virtqueue(struct device *dev, unsigned int num_descs,
/* Initialize the vring. */
vring_init(&vq->vring, num_descs, p, LGUEST_VRING_ALIGN);
/* Append virtqueue to this device's descriptor. We use
/*
* Append virtqueue to this device's descriptor. We use
* device_config() to get the end of the device's current virtqueues;
* we check that we haven't added any config or feature information
* yet, otherwise we'd be overwriting them. */
* yet, otherwise we'd be overwriting them.
*/
assert(dev->desc->config_len == 0 && dev->desc->feature_len == 0);
memcpy(device_config(dev), &vq->config, sizeof(vq->config));
dev->num_vq++;
......@@ -1072,14 +1240,18 @@ static void add_virtqueue(struct device *dev, unsigned int num_descs,
verbose("Virtqueue page %#lx\n", to_guest_phys(p));
/* Add to tail of list, so dev->vq is first vq, dev->vq->next is
* second. */
/*
* Add to tail of list, so dev->vq is first vq, dev->vq->next is
* second.
*/
for (i = &dev->vq; *i; i = &(*i)->next);
*i = vq;
}
/* The first half of the feature bitmask is for us to advertise features. The
* second half is for the Guest to accept features. */
/*
* The first half of the feature bitmask is for us to advertise features. The
* second half is for the Guest to accept features.
*/
static void add_feature(struct device *dev, unsigned bit)
{
u8 *features = get_feature_bits(dev);
......@@ -1093,9 +1265,11 @@ static void add_feature(struct device *dev, unsigned bit)
features[bit / CHAR_BIT] |= (1 << (bit % CHAR_BIT));
}
/* This routine sets the configuration fields for an existing device's
/*
* This routine sets the configuration fields for an existing device's
* descriptor. It only works for the last device, but that's OK because that's
* how we use it. */
* how we use it.
*/
static void set_config(struct device *dev, unsigned len, const void *conf)
{
/* Check we haven't overflowed our single page. */
......@@ -1105,12 +1279,18 @@ static void set_config(struct device *dev, unsigned len, const void *conf)
/* Copy in the config information, and store the length. */
memcpy(device_config(dev), conf, len);
dev->desc->config_len = len;
/* Size must fit in config_len field (8 bits)! */
assert(dev->desc->config_len == len);
}
/* This routine does all the creation and setup of a new device, including
* calling new_dev_desc() to allocate the descriptor and device memory.
/*
* This routine does all the creation and setup of a new device, including
* calling new_dev_desc() to allocate the descriptor and device memory. We
* don't actually start the service threads until later.
*
* See what I mean about userspace being boring? */
* See what I mean about userspace being boring?
*/
static struct device *new_device(const char *name, u16 type)
{
struct device *dev = malloc(sizeof(*dev));
......@@ -1123,10 +1303,12 @@ static struct device *new_device(const char *name, u16 type)
dev->num_vq = 0;
dev->running = false;
/* Append to device list. Prepending to a single-linked list is
/*
* Append to device list. Prepending to a single-linked list is
* easier, but the user expects the devices to be arranged on the bus
* in command-line order. The first network device on the command line
* is eth0, the first block device /dev/vda, etc. */
* is eth0, the first block device /dev/vda, etc.
*/
if (devices.lastdev)
devices.lastdev->next = dev;
else
......@@ -1136,8 +1318,10 @@ static struct device *new_device(const char *name, u16 type)
return dev;
}
/* Our first setup routine is the console. It's a fairly simple device, but
* UNIX tty handling makes it uglier than it could be. */
/*
* Our first setup routine is the console. It's a fairly simple device, but
* UNIX tty handling makes it uglier than it could be.
*/
static void setup_console(void)
{
struct device *dev;
......@@ -1145,8 +1329,10 @@ static void setup_console(void)
/* If we can save the initial standard input settings... */
if (tcgetattr(STDIN_FILENO, &orig_term) == 0) {
struct termios term = orig_term;
/* Then we turn off echo, line buffering and ^C etc. We want a
* raw input stream to the Guest. */
/*
* Then we turn off echo, line buffering and ^C etc: We want a
* raw input stream to the Guest.
*/
term.c_lflag &= ~(ISIG|ICANON|ECHO);
tcsetattr(STDIN_FILENO, TCSANOW, &term);
}
......@@ -1157,10 +1343,12 @@ static void setup_console(void)
dev->priv = malloc(sizeof(struct console_abort));
((struct console_abort *)dev->priv)->count = 0;
/* The console needs two virtqueues: the input then the output. When
/*
* The console needs two virtqueues: the input then the output. When
* they put something the input queue, we make sure we're listening to
* stdin. When they put something in the output queue, we write it to
* stdout. */
* stdout.
*/
add_virtqueue(dev, VIRTQUEUE_NUM, console_input);
add_virtqueue(dev, VIRTQUEUE_NUM, console_output);
......@@ -1168,7 +1356,8 @@ static void setup_console(void)
}
/*:*/
/*M:010 Inter-guest networking is an interesting area. Simplest is to have a
/*M:010
* Inter-guest networking is an interesting area. Simplest is to have a
* --sharenet=<name> option which opens or creates a named pipe. This can be
* used to send packets to another guest in a 1:1 manner.
*
......@@ -1182,7 +1371,8 @@ static void setup_console(void)
* multiple inter-guest channels behind one interface, although it would
* require some manner of hotplugging new virtio channels.
*
* Finally, we could implement a virtio network switch in the kernel. :*/
* Finally, we could implement a virtio network switch in the kernel.
:*/
static u32 str2ip(const char *ipaddr)
{
......@@ -1207,11 +1397,13 @@ static void str2mac(const char *macaddr, unsigned char mac[6])
mac[5] = m[5];
}
/* This code is "adapted" from libbridge: it attaches the Host end of the
/*
* This code is "adapted" from libbridge: it attaches the Host end of the
* network device to the bridge device specified by the command line.
*
* This is yet another James Morris contribution (I'm an IP-level guy, so I
* dislike bridging), and I just try not to break it. */
* dislike bridging), and I just try not to break it.
*/
static void add_to_bridge(int fd, const char *if_name, const char *br_name)
{
int ifidx;
......@@ -1231,9 +1423,11 @@ static void add_to_bridge(int fd, const char *if_name, const char *br_name)
err(1, "can't add %s to bridge %s", if_name, br_name);
}
/* This sets up the Host end of the network device with an IP address, brings
/*
* This sets up the Host end of the network device with an IP address, brings
* it up so packets will flow, the copies the MAC address into the hwaddr
* pointer. */
* pointer.
*/
static void configure_device(int fd, const char *tapif, u32 ipaddr)
{
struct ifreq ifr;
......@@ -1260,10 +1454,12 @@ static int get_tun_device(char tapif[IFNAMSIZ])
/* Start with this zeroed. Messy but sure. */
memset(&ifr, 0, sizeof(ifr));
/* We open the /dev/net/tun device and tell it we want a tap device. A
/*
* We open the /dev/net/tun device and tell it we want a tap device. A
* tap device is like a tun device, only somehow different. To tell
* the truth, I completely blundered my way through this code, but it
* works now! */
* works now!
*/
netfd = open_or_die("/dev/net/tun", O_RDWR);
ifr.ifr_flags = IFF_TAP | IFF_NO_PI | IFF_VNET_HDR;
strcpy(ifr.ifr_name, "tap%d");
......@@ -1274,18 +1470,22 @@ static int get_tun_device(char tapif[IFNAMSIZ])
TUN_F_CSUM|TUN_F_TSO4|TUN_F_TSO6|TUN_F_TSO_ECN) != 0)
err(1, "Could not set features for tun device");
/* We don't need checksums calculated for packets coming in this
* device: trust us! */
/*
* We don't need checksums calculated for packets coming in this
* device: trust us!
*/
ioctl(netfd, TUNSETNOCSUM, 1);
memcpy(tapif, ifr.ifr_name, IFNAMSIZ);
return netfd;
}
/*L:195 Our network is a Host<->Guest network. This can either use bridging or
/*L:195
* Our network is a Host<->Guest network. This can either use bridging or
* routing, but the principle is the same: it uses the "tun" device to inject
* packets into the Host as if they came in from a normal network card. We
* just shunt packets between the Guest and the tun device. */
* just shunt packets between the Guest and the tun device.
*/
static void setup_tun_net(char *arg)
{
struct device *dev;
......@@ -1302,13 +1502,14 @@ static void setup_tun_net(char *arg)
dev = new_device("net", VIRTIO_ID_NET);
dev->priv = net_info;
/* Network devices need a receive and a send queue, just like
* console. */
/* Network devices need a recv and a send queue, just like console. */
add_virtqueue(dev, VIRTQUEUE_NUM, net_input);
add_virtqueue(dev, VIRTQUEUE_NUM, net_output);
/* We need a socket to perform the magic network ioctls to bring up the
* tap interface, connect to the bridge etc. Any socket will do! */
/*
* We need a socket to perform the magic network ioctls to bring up the
* tap interface, connect to the bridge etc. Any socket will do!
*/
ipfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
if (ipfd < 0)
err(1, "opening IP socket");
......@@ -1362,39 +1563,31 @@ static void setup_tun_net(char *arg)
verbose("device %u: tun %s: %s\n",
devices.device_num, tapif, arg);
}
/* Our block (disk) device should be really simple: the Guest asks for a block
* number and we read or write that position in the file. Unfortunately, that
* was amazingly slow: the Guest waits until the read is finished before
* running anything else, even if it could have been doing useful work.
*
* We could use async I/O, except it's reputed to suck so hard that characters
* actually go missing from your code when you try to use it.
*
* So we farm the I/O out to thread, and communicate with it via a pipe. */
/*:*/
/* This hangs off device->priv. */
struct vblk_info
{
struct vblk_info {
/* The size of the file. */
off64_t len;
/* The file descriptor for the file. */
int fd;
/* IO thread listens on this file descriptor [0]. */
int workpipe[2];
/* IO thread writes to this file descriptor to mark it done, then
* Launcher triggers interrupt to Guest. */
int done_fd;
};
/*L:210
* The Disk
*
* Remember that the block device is handled by a separate I/O thread. We head
* straight into the core of that thread here:
* The disk only has one virtqueue, so it only has one thread. It is really
* simple: the Guest asks for a block number and we read or write that position
* in the file.
*
* Before we serviced each virtqueue in a separate thread, that was unacceptably
* slow: the Guest waits until the read is finished before running anything
* else, even if it could have been doing useful work.
*
* We could have used async I/O, except it's reputed to suck so hard that
* characters actually go missing from your code when you try to use it.
*/
static void blk_request(struct virtqueue *vq)
{
......@@ -1406,47 +1599,64 @@ static void blk_request(struct virtqueue *vq)
struct iovec iov[vq->vring.num];
off64_t off;
/* Get the next request. */
/*
* Get the next request, where we normally wait. It triggers the
* interrupt to acknowledge previously serviced requests (if any).
*/
head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
/* Every block request should contain at least one output buffer
/*
* Every block request should contain at least one output buffer
* (detailing the location on disk and the type of request) and one
* input buffer (to hold the result). */
* input buffer (to hold the result).
*/
if (out_num == 0 || in_num == 0)
errx(1, "Bad virtblk cmd %u out=%u in=%u",
head, out_num, in_num);
out = convert(&iov[0], struct virtio_blk_outhdr);
in = convert(&iov[out_num+in_num-1], u8);
/*
* For historical reasons, block operations are expressed in 512 byte
* "sectors".
*/
off = out->sector * 512;
/* The block device implements "barriers", where the Guest indicates
/*
* The block device implements "barriers", where the Guest indicates
* that it wants all previous writes to occur before this write. We
* don't have a way of asking our kernel to do a barrier, so we just
* synchronize all the data in the file. Pretty poor, no? */
* synchronize all the data in the file. Pretty poor, no?
*/
if (out->type & VIRTIO_BLK_T_BARRIER)
fdatasync(vblk->fd);
/* In general the virtio block driver is allowed to try SCSI commands.
* It'd be nice if we supported eject, for example, but we don't. */
/*
* In general the virtio block driver is allowed to try SCSI commands.
* It'd be nice if we supported eject, for example, but we don't.
*/
if (out->type & VIRTIO_BLK_T_SCSI_CMD) {
fprintf(stderr, "Scsi commands unsupported\n");
*in = VIRTIO_BLK_S_UNSUPP;
wlen = sizeof(*in);
} else if (out->type & VIRTIO_BLK_T_OUT) {
/* Write */
/* Move to the right location in the block file. This can fail
* if they try to write past end. */
/*
* Write
*
* Move to the right location in the block file. This can fail
* if they try to write past end.
*/
if (lseek64(vblk->fd, off, SEEK_SET) != off)
err(1, "Bad seek to sector %llu", out->sector);
ret = writev(vblk->fd, iov+1, out_num-1);
verbose("WRITE to sector %llu: %i\n", out->sector, ret);
/* Grr... Now we know how long the descriptor they sent was, we
/*
* Grr... Now we know how long the descriptor they sent was, we
* make sure they didn't try to write over the end of the block
* file (possibly extending it). */
* file (possibly extending it).
*/
if (ret > 0 && off + ret > vblk->len) {
/* Trim it back to the correct length */
ftruncate64(vblk->fd, vblk->len);
......@@ -1456,10 +1666,12 @@ static void blk_request(struct virtqueue *vq)
wlen = sizeof(*in);
*in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR);
} else {
/* Read */
/* Move to the right location in the block file. This can fail
* if they try to read past end. */
/*
* Read
*
* Move to the right location in the block file. This can fail
* if they try to read past end.
*/
if (lseek64(vblk->fd, off, SEEK_SET) != off)
err(1, "Bad seek to sector %llu", out->sector);
......@@ -1474,13 +1686,16 @@ static void blk_request(struct virtqueue *vq)
}
}
/* OK, so we noted that it was pretty poor to use an fdatasync as a
/*
* OK, so we noted that it was pretty poor to use an fdatasync as a
* barrier. But Christoph Hellwig points out that we need a sync
* *afterwards* as well: "Barriers specify no reordering to the front
* or the back." And Jens Axboe confirmed it, so here we are: */
* or the back." And Jens Axboe confirmed it, so here we are:
*/
if (out->type & VIRTIO_BLK_T_BARRIER)
fdatasync(vblk->fd);
/* Finished that request. */
add_used(vq, head, wlen);
}
......@@ -1491,7 +1706,7 @@ static void setup_block_file(const char *filename)
struct vblk_info *vblk;
struct virtio_blk_config conf;
/* The device responds to return from I/O thread. */
/* Creat the device. */
dev = new_device("block", VIRTIO_ID_BLOCK);
/* The device has one virtqueue, where the Guest places requests. */
......@@ -1510,27 +1725,32 @@ static void setup_block_file(const char *filename)
/* Tell Guest how many sectors this device has. */
conf.capacity = cpu_to_le64(vblk->len / 512);
/* Tell Guest not to put in too many descriptors at once: two are used
* for the in and out elements. */
/*
* Tell Guest not to put in too many descriptors at once: two are used
* for the in and out elements.
*/
add_feature(dev, VIRTIO_BLK_F_SEG_MAX);
conf.seg_max = cpu_to_le32(VIRTQUEUE_NUM - 2);
set_config(dev, sizeof(conf), &conf);
/* Don't try to put whole struct: we have 8 bit limit. */
set_config(dev, offsetof(struct virtio_blk_config, geometry), &conf);
verbose("device %u: virtblock %llu sectors\n",
++devices.device_num, le64_to_cpu(conf.capacity));
}
struct rng_info {
int rfd;
};
/* Our random number generator device reads from /dev/random into the Guest's
/*L:211
* Our random number generator device reads from /dev/random into the Guest's
* input buffers. The usual case is that the Guest doesn't want random numbers
* and so has no buffers although /dev/random is still readable, whereas
* console is the reverse.
*
* The same logic applies, however. */
* The same logic applies, however.
*/
struct rng_info {
int rfd;
};
static void rng_input(struct virtqueue *vq)
{
int len;
......@@ -1543,9 +1763,10 @@ static void rng_input(struct virtqueue *vq)
if (out_num)
errx(1, "Output buffers in rng?");
/* This is why we convert to iovecs: the readv() call uses them, and so
* it reads straight into the Guest's buffer. We loop to make sure we
* fill it. */
/*
* Just like the console write, we loop to cover the whole iovec.
* In this case, short reads actually happen quite a bit.
*/
while (!iov_empty(iov, in_num)) {
len = readv(rng_info->rfd, iov, in_num);
if (len <= 0)
......@@ -1558,15 +1779,18 @@ static void rng_input(struct virtqueue *vq)
add_used(vq, head, totlen);
}
/* And this creates a "hardware" random number device for the Guest. */
/*L:199
* This creates a "hardware" random number device for the Guest.
*/
static void setup_rng(void)
{
struct device *dev;
struct rng_info *rng_info = malloc(sizeof(*rng_info));
/* Our device's privat info simply contains the /dev/random fd. */
rng_info->rfd = open_or_die("/dev/random", O_RDONLY);
/* The device responds to return from I/O thread. */
/* Create the new device. */
dev = new_device("rng", VIRTIO_ID_RNG);
dev->priv = rng_info;
......@@ -1582,8 +1806,10 @@ static void __attribute__((noreturn)) restart_guest(void)
{
unsigned int i;
/* Since we don't track all open fds, we simply close everything beyond
* stderr. */
/*
* Since we don't track all open fds, we simply close everything beyond
* stderr.
*/
for (i = 3; i < FD_SETSIZE; i++)
close(i);
......@@ -1594,8 +1820,10 @@ static void __attribute__((noreturn)) restart_guest(void)
err(1, "Could not exec %s", main_args[0]);
}
/*L:220 Finally we reach the core of the Launcher which runs the Guest, serves
* its input and output, and finally, lays it to rest. */
/*L:220
* Finally we reach the core of the Launcher which runs the Guest, serves
* its input and output, and finally, lays it to rest.
*/
static void __attribute__((noreturn)) run_guest(void)
{
for (;;) {
......@@ -1630,7 +1858,7 @@ static void __attribute__((noreturn)) run_guest(void)
*
* Are you ready? Take a deep breath and join me in the core of the Host, in
* "make Host".
:*/
:*/
static struct option opts[] = {
{ "verbose", 0, NULL, 'v' },
......@@ -1651,8 +1879,7 @@ static void usage(void)
/*L:105 The main routine is where the real work begins: */
int main(int argc, char *argv[])
{
/* Memory, top-level pagetable, code startpoint and size of the
* (optional) initrd. */
/* Memory, code startpoint and size of the (optional) initrd. */
unsigned long mem = 0, start, initrd_size = 0;
/* Two temporaries. */
int i, c;
......@@ -1664,24 +1891,32 @@ int main(int argc, char *argv[])
/* Save the args: we "reboot" by execing ourselves again. */
main_args = argv;
/* First we initialize the device list. We keep a pointer to the last
/*
* First we initialize the device list. We keep a pointer to the last
* device, and the next interrupt number to use for devices (1:
* remember that 0 is used by the timer). */
* remember that 0 is used by the timer).
*/
devices.lastdev = NULL;
devices.next_irq = 1;
/* We're CPU 0. In fact, that's the only CPU possible right now. */
cpu_id = 0;
/* We need to know how much memory so we can set up the device
/*
* We need to know how much memory so we can set up the device
* descriptor and memory pages for the devices as we parse the command
* line. So we quickly look through the arguments to find the amount
* of memory now. */
* of memory now.
*/
for (i = 1; i < argc; i++) {
if (argv[i][0] != '-') {
mem = atoi(argv[i]) * 1024 * 1024;
/* We start by mapping anonymous pages over all of
/*
* We start by mapping anonymous pages over all of
* guest-physical memory range. This fills it with 0,
* and ensures that the Guest won't be killed when it
* tries to access it. */
* tries to access it.
*/
guest_base = map_zeroed_pages(mem / getpagesize()
+ DEVICE_PAGES);
guest_limit = mem;
......@@ -1714,8 +1949,10 @@ int main(int argc, char *argv[])
usage();
}
}
/* After the other arguments we expect memory and kernel image name,
* followed by command line arguments for the kernel. */
/*
* After the other arguments we expect memory and kernel image name,
* followed by command line arguments for the kernel.
*/
if (optind + 2 > argc)
usage();
......@@ -1733,20 +1970,26 @@ int main(int argc, char *argv[])
/* Map the initrd image if requested (at top of physical memory) */
if (initrd_name) {
initrd_size = load_initrd(initrd_name, mem);
/* These are the location in the Linux boot header where the
* start and size of the initrd are expected to be found. */
/*
* These are the location in the Linux boot header where the
* start and size of the initrd are expected to be found.
*/
boot->hdr.ramdisk_image = mem - initrd_size;
boot->hdr.ramdisk_size = initrd_size;
/* The bootloader type 0xFF means "unknown"; that's OK. */
boot->hdr.type_of_loader = 0xFF;
}
/* The Linux boot header contains an "E820" memory map: ours is a
* simple, single region. */
/*
* The Linux boot header contains an "E820" memory map: ours is a
* simple, single region.
*/
boot->e820_entries = 1;
boot->e820_map[0] = ((struct e820entry) { 0, mem, E820_RAM });
/* The boot header contains a command line pointer: we put the command
* line after the boot header. */
/*
* The boot header contains a command line pointer: we put the command
* line after the boot header.
*/
boot->hdr.cmd_line_ptr = to_guest_phys(boot + 1);
/* We use a simple helper to copy the arguments separated by spaces. */
concat((char *)(boot + 1), argv+optind+2);
......@@ -1760,11 +2003,13 @@ int main(int argc, char *argv[])
/* Tell the entry path not to try to reload segment registers. */
boot->hdr.loadflags |= KEEP_SEGMENTS;
/* We tell the kernel to initialize the Guest: this returns the open
* /dev/lguest file descriptor. */
/*
* We tell the kernel to initialize the Guest: this returns the open
* /dev/lguest file descriptor.
*/
tell_kernel(start);
/* Ensure that we terminate if a child dies. */
/* Ensure that we terminate if a device-servicing child dies. */
signal(SIGCHLD, kill_launcher);
/* If we exit via err(), this kills all the threads, restores tty. */
......
......@@ -17,8 +17,7 @@
/* Pages for switcher itself, then two pages per cpu */
#define TOTAL_SWITCHER_PAGES (SHARED_SWITCHER_PAGES + 2 * nr_cpu_ids)
/* We map at -4M (-2M when PAE is activated) for ease of mapping
* into the guest (one PTE page). */
/* We map at -4M (-2M for PAE) for ease of mapping (one PTE page). */
#ifdef CONFIG_X86_PAE
#define SWITCHER_ADDR 0xFFE00000
#else
......
......@@ -30,27 +30,27 @@
#include <asm/hw_irq.h>
#include <asm/kvm_para.h>
/*G:030 But first, how does our Guest contact the Host to ask for privileged
/*G:030
* But first, how does our Guest contact the Host to ask for privileged
* operations? There are two ways: the direct way is to make a "hypercall",
* to make requests of the Host Itself.
*
* We use the KVM hypercall mechanism. Seventeen hypercalls are
* available: the hypercall number is put in the %eax register, and the
* arguments (when required) are placed in %ebx, %ecx, %edx and %esi.
* If a return value makes sense, it's returned in %eax.
* We use the KVM hypercall mechanism, though completely different hypercall
* numbers. Seventeen hypercalls are available: the hypercall number is put in
* the %eax register, and the arguments (when required) are placed in %ebx,
* %ecx, %edx and %esi. If a return value makes sense, it's returned in %eax.
*
* Grossly invalid calls result in Sudden Death at the hands of the vengeful
* Host, rather than returning failure. This reflects Winston Churchill's
* definition of a gentleman: "someone who is only rude intentionally". */
/*:*/
* definition of a gentleman: "someone who is only rude intentionally".
:*/
/* Can't use our min() macro here: needs to be a constant */
#define LGUEST_IRQS (NR_IRQS < 32 ? NR_IRQS: 32)
#define LHCALL_RING_SIZE 64
struct hcall_args {
/* These map directly onto eax, ebx, ecx, edx and esi
* in struct lguest_regs */
/* These map directly onto eax/ebx/ecx/edx/esi in struct lguest_regs */
unsigned long arg0, arg1, arg2, arg3, arg4;
};
......
......@@ -22,7 +22,8 @@
*
* So how does the kernel know it's a Guest? We'll see that later, but let's
* just say that we end up here where we replace the native functions various
* "paravirt" structures with our Guest versions, then boot like normal. :*/
* "paravirt" structures with our Guest versions, then boot like normal.
:*/
/*
* Copyright (C) 2006, Rusty Russell <rusty@rustcorp.com.au> IBM Corporation.
......@@ -74,7 +75,8 @@
*
* The Guest in our tale is a simple creature: identical to the Host but
* behaving in simplified but equivalent ways. In particular, the Guest is the
* same kernel as the Host (or at least, built from the same source code). :*/
* same kernel as the Host (or at least, built from the same source code).
:*/
struct lguest_data lguest_data = {
.hcall_status = { [0 ... LHCALL_RING_SIZE-1] = 0xFF },
......@@ -85,7 +87,8 @@ struct lguest_data lguest_data = {
.syscall_vec = SYSCALL_VECTOR,
};
/*G:037 async_hcall() is pretty simple: I'm quite proud of it really. We have a
/*G:037
* async_hcall() is pretty simple: I'm quite proud of it really. We have a
* ring buffer of stored hypercalls which the Host will run though next time we
* do a normal hypercall. Each entry in the ring has 5 slots for the hypercall
* arguments, and a "hcall_status" word which is 0 if the call is ready to go,
......@@ -94,7 +97,8 @@ struct lguest_data lguest_data = {
* If we come around to a slot which hasn't been finished, then the table is
* full and we just make the hypercall directly. This has the nice side
* effect of causing the Host to run all the stored calls in the ring buffer
* which empties it for next time! */
* which empties it for next time!
*/
static void async_hcall(unsigned long call, unsigned long arg1,
unsigned long arg2, unsigned long arg3,
unsigned long arg4)
......@@ -103,9 +107,11 @@ static void async_hcall(unsigned long call, unsigned long arg1,
static unsigned int next_call;
unsigned long flags;
/* Disable interrupts if not already disabled: we don't want an
/*
* Disable interrupts if not already disabled: we don't want an
* interrupt handler making a hypercall while we're already doing
* one! */
* one!
*/
local_irq_save(flags);
if (lguest_data.hcall_status[next_call] != 0xFF) {
/* Table full, so do normal hcall which will flush table. */
......@@ -125,8 +131,9 @@ static void async_hcall(unsigned long call, unsigned long arg1,
local_irq_restore(flags);
}
/*G:035 Notice the lazy_hcall() above, rather than hcall(). This is our first
* real optimization trick!
/*G:035
* Notice the lazy_hcall() above, rather than hcall(). This is our first real
* optimization trick!
*
* When lazy_mode is set, it means we're allowed to defer all hypercalls and do
* them as a batch when lazy_mode is eventually turned off. Because hypercalls
......@@ -136,7 +143,8 @@ static void async_hcall(unsigned long call, unsigned long arg1,
* lguest_leave_lazy_mode().
*
* So, when we're in lazy mode, we call async_hcall() to store the call for
* future processing: */
* future processing:
*/
static void lazy_hcall1(unsigned long call,
unsigned long arg1)
{
......@@ -146,6 +154,7 @@ static void lazy_hcall1(unsigned long call,
async_hcall(call, arg1, 0, 0, 0);
}
/* You can imagine what lazy_hcall2, 3 and 4 look like. :*/
static void lazy_hcall2(unsigned long call,
unsigned long arg1,
unsigned long arg2)
......@@ -181,8 +190,10 @@ static void lazy_hcall4(unsigned long call,
}
#endif
/* When lazy mode is turned off reset the per-cpu lazy mode variable and then
* issue the do-nothing hypercall to flush any stored calls. */
/*G:036
* When lazy mode is turned off reset the per-cpu lazy mode variable and then
* issue the do-nothing hypercall to flush any stored calls.
:*/
static void lguest_leave_lazy_mmu_mode(void)
{
kvm_hypercall0(LHCALL_FLUSH_ASYNC);
......@@ -208,9 +219,11 @@ static void lguest_end_context_switch(struct task_struct *next)
* check there before it tries to deliver an interrupt.
*/
/* save_flags() is expected to return the processor state (ie. "flags"). The
/*
* save_flags() is expected to return the processor state (ie. "flags"). The
* flags word contains all kind of stuff, but in practice Linux only cares
* about the interrupt flag. Our "save_flags()" just returns that. */
* about the interrupt flag. Our "save_flags()" just returns that.
*/
static unsigned long save_fl(void)
{
return lguest_data.irq_enabled;
......@@ -222,13 +235,15 @@ static void irq_disable(void)
lguest_data.irq_enabled = 0;
}
/* Let's pause a moment. Remember how I said these are called so often?
/*
* Let's pause a moment. Remember how I said these are called so often?
* Jeremy Fitzhardinge optimized them so hard early in 2009 that he had to
* break some rules. In particular, these functions are assumed to save their
* own registers if they need to: normal C functions assume they can trash the
* eax register. To use normal C functions, we use
* PV_CALLEE_SAVE_REGS_THUNK(), which pushes %eax onto the stack, calls the
* C function, then restores it. */
* C function, then restores it.
*/
PV_CALLEE_SAVE_REGS_THUNK(save_fl);
PV_CALLEE_SAVE_REGS_THUNK(irq_disable);
/*:*/
......@@ -237,18 +252,18 @@ PV_CALLEE_SAVE_REGS_THUNK(irq_disable);
extern void lg_irq_enable(void);
extern void lg_restore_fl(unsigned long flags);
/*M:003 Note that we don't check for outstanding interrupts when we re-enable
* them (or when we unmask an interrupt). This seems to work for the moment,
* since interrupts are rare and we'll just get the interrupt on the next timer
* tick, but now we can run with CONFIG_NO_HZ, we should revisit this. One way
* would be to put the "irq_enabled" field in a page by itself, and have the
* Host write-protect it when an interrupt comes in when irqs are disabled.
* There will then be a page fault as soon as interrupts are re-enabled.
/*M:003
* We could be more efficient in our checking of outstanding interrupts, rather
* than using a branch. One way would be to put the "irq_enabled" field in a
* page by itself, and have the Host write-protect it when an interrupt comes
* in when irqs are disabled. There will then be a page fault as soon as
* interrupts are re-enabled.
*
* A better method is to implement soft interrupt disable generally for x86:
* instead of disabling interrupts, we set a flag. If an interrupt does come
* in, we then disable them for real. This is uncommon, so we could simply use
* a hypercall for interrupt control and not worry about efficiency. :*/
* a hypercall for interrupt control and not worry about efficiency.
:*/
/*G:034
* The Interrupt Descriptor Table (IDT).
......@@ -261,10 +276,12 @@ extern void lg_restore_fl(unsigned long flags);
static void lguest_write_idt_entry(gate_desc *dt,
int entrynum, const gate_desc *g)
{
/* The gate_desc structure is 8 bytes long: we hand it to the Host in
/*
* The gate_desc structure is 8 bytes long: we hand it to the Host in
* two 32-bit chunks. The whole 32-bit kernel used to hand descriptors
* around like this; typesafety wasn't a big concern in Linux's early
* years. */
* years.
*/
u32 *desc = (u32 *)g;
/* Keep the local copy up to date. */
native_write_idt_entry(dt, entrynum, g);
......@@ -272,9 +289,11 @@ static void lguest_write_idt_entry(gate_desc *dt,
kvm_hypercall3(LHCALL_LOAD_IDT_ENTRY, entrynum, desc[0], desc[1]);
}
/* Changing to a different IDT is very rare: we keep the IDT up-to-date every
/*
* Changing to a different IDT is very rare: we keep the IDT up-to-date every
* time it is written, so we can simply loop through all entries and tell the
* Host about them. */
* Host about them.
*/
static void lguest_load_idt(const struct desc_ptr *desc)
{
unsigned int i;
......@@ -305,9 +324,11 @@ static void lguest_load_gdt(const struct desc_ptr *desc)
kvm_hypercall3(LHCALL_LOAD_GDT_ENTRY, i, gdt[i].a, gdt[i].b);
}
/* For a single GDT entry which changes, we do the lazy thing: alter our GDT,
/*
* For a single GDT entry which changes, we do the lazy thing: alter our GDT,
* then tell the Host to reload the entire thing. This operation is so rare
* that this naive implementation is reasonable. */
* that this naive implementation is reasonable.
*/
static void lguest_write_gdt_entry(struct desc_struct *dt, int entrynum,
const void *desc, int type)
{
......@@ -317,29 +338,36 @@ static void lguest_write_gdt_entry(struct desc_struct *dt, int entrynum,
dt[entrynum].a, dt[entrynum].b);
}
/* OK, I lied. There are three "thread local storage" GDT entries which change
/*
* OK, I lied. There are three "thread local storage" GDT entries which change
* on every context switch (these three entries are how glibc implements
* __thread variables). So we have a hypercall specifically for this case. */
* __thread variables). So we have a hypercall specifically for this case.
*/
static void lguest_load_tls(struct thread_struct *t, unsigned int cpu)
{
/* There's one problem which normal hardware doesn't have: the Host
/*
* There's one problem which normal hardware doesn't have: the Host
* can't handle us removing entries we're currently using. So we clear
* the GS register here: if it's needed it'll be reloaded anyway. */
* the GS register here: if it's needed it'll be reloaded anyway.
*/
lazy_load_gs(0);
lazy_hcall2(LHCALL_LOAD_TLS, __pa(&t->tls_array), cpu);
}
/*G:038 That's enough excitement for now, back to ploughing through each of
* the different pv_ops structures (we're about 1/3 of the way through).
/*G:038
* That's enough excitement for now, back to ploughing through each of the
* different pv_ops structures (we're about 1/3 of the way through).
*
* This is the Local Descriptor Table, another weird Intel thingy. Linux only
* uses this for some strange applications like Wine. We don't do anything
* here, so they'll get an informative and friendly Segmentation Fault. */
* here, so they'll get an informative and friendly Segmentation Fault.
*/
static void lguest_set_ldt(const void *addr, unsigned entries)
{
}
/* This loads a GDT entry into the "Task Register": that entry points to a
/*
* This loads a GDT entry into the "Task Register": that entry points to a
* structure called the Task State Segment. Some comments scattered though the
* kernel code indicate that this used for task switching in ages past, along
* with blood sacrifice and astrology.
......@@ -347,19 +375,21 @@ static void lguest_set_ldt(const void *addr, unsigned entries)
* Now there's nothing interesting in here that we don't get told elsewhere.
* But the native version uses the "ltr" instruction, which makes the Host
* complain to the Guest about a Segmentation Fault and it'll oops. So we
* override the native version with a do-nothing version. */
* override the native version with a do-nothing version.
*/
static void lguest_load_tr_desc(void)
{
}
/* The "cpuid" instruction is a way of querying both the CPU identity
/*
* The "cpuid" instruction is a way of querying both the CPU identity
* (manufacturer, model, etc) and its features. It was introduced before the
* Pentium in 1993 and keeps getting extended by both Intel, AMD and others.
* As you might imagine, after a decade and a half this treatment, it is now a
* giant ball of hair. Its entry in the current Intel manual runs to 28 pages.
*
* This instruction even it has its own Wikipedia entry. The Wikipedia entry
* has been translated into 4 languages. I am not making this up!
* has been translated into 5 languages. I am not making this up!
*
* We could get funky here and identify ourselves as "GenuineLguest", but
* instead we just use the real "cpuid" instruction. Then I pretty much turned
......@@ -371,7 +401,8 @@ static void lguest_load_tr_desc(void)
* Replacing the cpuid so we can turn features off is great for the kernel, but
* anyone (including userspace) can just use the raw "cpuid" instruction and
* the Host won't even notice since it isn't privileged. So we try not to get
* too worked up about it. */
* too worked up about it.
*/
static void lguest_cpuid(unsigned int *ax, unsigned int *bx,
unsigned int *cx, unsigned int *dx)
{
......@@ -379,43 +410,63 @@ static void lguest_cpuid(unsigned int *ax, unsigned int *bx,
native_cpuid(ax, bx, cx, dx);
switch (function) {
case 0: /* ID and highest CPUID. Futureproof a little by sticking to
* older ones. */
/*
* CPUID 0 gives the highest legal CPUID number (and the ID string).
* We futureproof our code a little by sticking to known CPUID values.
*/
case 0:
if (*ax > 5)
*ax = 5;
break;
case 1: /* Basic feature request. */
/* We only allow kernel to see SSE3, CMPXCHG16B and SSSE3 */
/*
* CPUID 1 is a basic feature request.
*
* CX: we only allow kernel to see SSE3, CMPXCHG16B and SSSE3
* DX: SSE, SSE2, FXSR, MMX, CMOV, CMPXCHG8B, TSC, FPU and PAE.
*/
case 1:
*cx &= 0x00002201;
/* SSE, SSE2, FXSR, MMX, CMOV, CMPXCHG8B, TSC, FPU, PAE. */
*dx &= 0x07808151;
/* The Host can do a nice optimization if it knows that the
/*
* The Host can do a nice optimization if it knows that the
* kernel mappings (addresses above 0xC0000000 or whatever
* PAGE_OFFSET is set to) haven't changed. But Linux calls
* flush_tlb_user() for both user and kernel mappings unless
* the Page Global Enable (PGE) feature bit is set. */
* the Page Global Enable (PGE) feature bit is set.
*/
*dx |= 0x00002000;
/* We also lie, and say we're family id 5. 6 or greater
/*
* We also lie, and say we're family id 5. 6 or greater
* leads to a rdmsr in early_init_intel which we can't handle.
* Family ID is returned as bits 8-12 in ax. */
* Family ID is returned as bits 8-12 in ax.
*/
*ax &= 0xFFFFF0FF;
*ax |= 0x00000500;
break;
/*
* 0x80000000 returns the highest Extended Function, so we futureproof
* like we do above by limiting it to known fields.
*/
case 0x80000000:
/* Futureproof this a little: if they ask how much extended
* processor information there is, limit it to known fields. */
if (*ax > 0x80000008)
*ax = 0x80000008;
break;
/*
* PAE systems can mark pages as non-executable. Linux calls this the
* NX bit. Intel calls it XD (eXecute Disable), AMD EVP (Enhanced
* Virus Protection). We just switch turn if off here, since we don't
* support it.
*/
case 0x80000001:
/* Here we should fix nx cap depending on host. */
/* For this version of PAE, we just clear NX bit. */
*dx &= ~(1 << 20);
break;
}
}
/* Intel has four control registers, imaginatively named cr0, cr2, cr3 and cr4.
/*
* Intel has four control registers, imaginatively named cr0, cr2, cr3 and cr4.
* I assume there's a cr1, but it hasn't bothered us yet, so we'll not bother
* it. The Host needs to know when the Guest wants to change them, so we have
* a whole series of functions like read_cr0() and write_cr0().
......@@ -430,7 +481,8 @@ static void lguest_cpuid(unsigned int *ax, unsigned int *bx,
* name like "FPUTRAP bit" be a little less cryptic?
*
* We store cr0 locally because the Host never changes it. The Guest sometimes
* wants to read it and we'd prefer not to bother the Host unnecessarily. */
* wants to read it and we'd prefer not to bother the Host unnecessarily.
*/
static unsigned long current_cr0;
static void lguest_write_cr0(unsigned long val)
{
......@@ -443,18 +495,22 @@ static unsigned long lguest_read_cr0(void)
return current_cr0;
}
/* Intel provided a special instruction to clear the TS bit for people too cool
/*
* Intel provided a special instruction to clear the TS bit for people too cool
* to use write_cr0() to do it. This "clts" instruction is faster, because all
* the vowels have been optimized out. */
* the vowels have been optimized out.
*/
static void lguest_clts(void)
{
lazy_hcall1(LHCALL_TS, 0);
current_cr0 &= ~X86_CR0_TS;
}
/* cr2 is the virtual address of the last page fault, which the Guest only ever
/*
* cr2 is the virtual address of the last page fault, which the Guest only ever
* reads. The Host kindly writes this into our "struct lguest_data", so we
* just read it out of there. */
* just read it out of there.
*/
static unsigned long lguest_read_cr2(void)
{
return lguest_data.cr2;
......@@ -463,10 +519,12 @@ static unsigned long lguest_read_cr2(void)
/* See lguest_set_pte() below. */
static bool cr3_changed = false;
/* cr3 is the current toplevel pagetable page: the principle is the same as
/*
* cr3 is the current toplevel pagetable page: the principle is the same as
* cr0. Keep a local copy, and tell the Host when it changes. The only
* difference is that our local copy is in lguest_data because the Host needs
* to set it upon our initial hypercall. */
* to set it upon our initial hypercall.
*/
static void lguest_write_cr3(unsigned long cr3)
{
lguest_data.pgdir = cr3;
......@@ -511,7 +569,7 @@ static void lguest_write_cr4(unsigned long val)
* cr3 ---> +---------+
* | --------->+---------+
* | | | PADDR1 |
* Top-level | | PADDR2 |
* Mid-level | | PADDR2 |
* (PMD) page | | |
* | | Lower-level |
* | | (PTE) page |
......@@ -531,21 +589,62 @@ static void lguest_write_cr4(unsigned long val)
* Index into top Index into second Offset within page
* page directory page pagetable page
*
* The kernel spends a lot of time changing both the top-level page directory
* and lower-level pagetable pages. The Guest doesn't know physical addresses,
* so while it maintains these page tables exactly like normal, it also needs
* to keep the Host informed whenever it makes a change: the Host will create
* the real page tables based on the Guests'.
* Now, unfortunately, this isn't the whole story: Intel added Physical Address
* Extension (PAE) to allow 32 bit systems to use 64GB of memory (ie. 36 bits).
* These are held in 64-bit page table entries, so we can now only fit 512
* entries in a page, and the neat three-level tree breaks down.
*
* The result is a four level page table:
*
* cr3 --> [ 4 Upper ]
* [ Level ]
* [ Entries ]
* [(PUD Page)]---> +---------+
* | --------->+---------+
* | | | PADDR1 |
* Mid-level | | PADDR2 |
* (PMD) page | | |
* | | Lower-level |
* | | (PTE) page |
* | | | |
* .... ....
*
*
* And the virtual address is decoded as:
*
* 1 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
* |<-2->|<--- 9 bits ---->|<---- 9 bits --->|<------ 12 bits ------>|
* Index into Index into mid Index into lower Offset within page
* top entries directory page pagetable page
*
* It's too hard to switch between these two formats at runtime, so Linux only
* supports one or the other depending on whether CONFIG_X86_PAE is set. Many
* distributions turn it on, and not just for people with silly amounts of
* memory: the larger PTE entries allow room for the NX bit, which lets the
* kernel disable execution of pages and increase security.
*
* This was a problem for lguest, which couldn't run on these distributions;
* then Matias Zabaljauregui figured it all out and implemented it, and only a
* handful of puppies were crushed in the process!
*
* Back to our point: the kernel spends a lot of time changing both the
* top-level page directory and lower-level pagetable pages. The Guest doesn't
* know physical addresses, so while it maintains these page tables exactly
* like normal, it also needs to keep the Host informed whenever it makes a
* change: the Host will create the real page tables based on the Guests'.
*/
/* The Guest calls this to set a second-level entry (pte), ie. to map a page
* into a process' address space. We set the entry then tell the Host the
* toplevel and address this corresponds to. The Guest uses one pagetable per
* process, so we need to tell the Host which one we're changing (mm->pgd). */
/*
* The Guest calls this after it has set a second-level entry (pte), ie. to map
* a page into a process' address space. Wetell the Host the toplevel and
* address this corresponds to. The Guest uses one pagetable per process, so
* we need to tell the Host which one we're changing (mm->pgd).
*/
static void lguest_pte_update(struct mm_struct *mm, unsigned long addr,
pte_t *ptep)
{
#ifdef CONFIG_X86_PAE
/* PAE needs to hand a 64 bit page table entry, so it uses two args. */
lazy_hcall4(LHCALL_SET_PTE, __pa(mm->pgd), addr,
ptep->pte_low, ptep->pte_high);
#else
......@@ -553,6 +652,7 @@ static void lguest_pte_update(struct mm_struct *mm, unsigned long addr,
#endif
}
/* This is the "set and update" combo-meal-deal version. */
static void lguest_set_pte_at(struct mm_struct *mm, unsigned long addr,
pte_t *ptep, pte_t pteval)
{
......@@ -560,10 +660,13 @@ static void lguest_set_pte_at(struct mm_struct *mm, unsigned long addr,
lguest_pte_update(mm, addr, ptep);
}
/* The Guest calls lguest_set_pud to set a top-level entry and lguest_set_pmd
/*
* The Guest calls lguest_set_pud to set a top-level entry and lguest_set_pmd
* to set a middle-level entry when PAE is activated.
*
* Again, we set the entry then tell the Host which page we changed,
* and the index of the entry we changed. */
* and the index of the entry we changed.
*/
#ifdef CONFIG_X86_PAE
static void lguest_set_pud(pud_t *pudp, pud_t pudval)
{
......@@ -582,8 +685,7 @@ static void lguest_set_pmd(pmd_t *pmdp, pmd_t pmdval)
}
#else
/* The Guest calls lguest_set_pmd to set a top-level entry when PAE is not
* activated. */
/* The Guest calls lguest_set_pmd to set a top-level entry when !PAE. */
static void lguest_set_pmd(pmd_t *pmdp, pmd_t pmdval)
{
native_set_pmd(pmdp, pmdval);
......@@ -592,7 +694,8 @@ static void lguest_set_pmd(pmd_t *pmdp, pmd_t pmdval)
}
#endif
/* There are a couple of legacy places where the kernel sets a PTE, but we
/*
* There are a couple of legacy places where the kernel sets a PTE, but we
* don't know the top level any more. This is useless for us, since we don't
* know which pagetable is changing or what address, so we just tell the Host
* to forget all of them. Fortunately, this is very rare.
......@@ -600,7 +703,8 @@ static void lguest_set_pmd(pmd_t *pmdp, pmd_t pmdval)
* ... except in early boot when the kernel sets up the initial pagetables,
* which makes booting astonishingly slow: 1.83 seconds! So we don't even tell
* the Host anything changed until we've done the first page table switch,
* which brings boot back to 0.25 seconds. */
* which brings boot back to 0.25 seconds.
*/
static void lguest_set_pte(pte_t *ptep, pte_t pteval)
{
native_set_pte(ptep, pteval);
......@@ -609,6 +713,11 @@ static void lguest_set_pte(pte_t *ptep, pte_t pteval)
}
#ifdef CONFIG_X86_PAE
/*
* With 64-bit PTE values, we need to be careful setting them: if we set 32
* bits at a time, the hardware could see a weird half-set entry. These
* versions ensure we update all 64 bits at once.
*/
static void lguest_set_pte_atomic(pte_t *ptep, pte_t pte)
{
native_set_pte_atomic(ptep, pte);
......@@ -616,19 +725,21 @@ static void lguest_set_pte_atomic(pte_t *ptep, pte_t pte)
lazy_hcall1(LHCALL_FLUSH_TLB, 1);
}
void lguest_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
static void lguest_pte_clear(struct mm_struct *mm, unsigned long addr,
pte_t *ptep)
{
native_pte_clear(mm, addr, ptep);
lguest_pte_update(mm, addr, ptep);
}
void lguest_pmd_clear(pmd_t *pmdp)
static void lguest_pmd_clear(pmd_t *pmdp)
{
lguest_set_pmd(pmdp, __pmd(0));
}
#endif
/* Unfortunately for Lguest, the pv_mmu_ops for page tables were based on
/*
* Unfortunately for Lguest, the pv_mmu_ops for page tables were based on
* native page table operations. On native hardware you can set a new page
* table entry whenever you want, but if you want to remove one you have to do
* a TLB flush (a TLB is a little cache of page table entries kept by the CPU).
......@@ -637,24 +748,29 @@ void lguest_pmd_clear(pmd_t *pmdp)
* called when a valid entry is written, not when it's removed (ie. marked not
* present). Instead, this is where we come when the Guest wants to remove a
* page table entry: we tell the Host to set that entry to 0 (ie. the present
* bit is zero). */
* bit is zero).
*/
static void lguest_flush_tlb_single(unsigned long addr)
{
/* Simply set it to zero: if it was not, it will fault back in. */
lazy_hcall3(LHCALL_SET_PTE, lguest_data.pgdir, addr, 0);
}
/* This is what happens after the Guest has removed a large number of entries.
/*
* This is what happens after the Guest has removed a large number of entries.
* This tells the Host that any of the page table entries for userspace might
* have changed, ie. virtual addresses below PAGE_OFFSET. */
* have changed, ie. virtual addresses below PAGE_OFFSET.
*/
static void lguest_flush_tlb_user(void)
{
lazy_hcall1(LHCALL_FLUSH_TLB, 0);
}
/* This is called when the kernel page tables have changed. That's not very
/*
* This is called when the kernel page tables have changed. That's not very
* common (unless the Guest is using highmem, which makes the Guest extremely
* slow), so it's worth separating this from the user flushing above. */
* slow), so it's worth separating this from the user flushing above.
*/
static void lguest_flush_tlb_kernel(void)
{
lazy_hcall1(LHCALL_FLUSH_TLB, 1);
......@@ -691,26 +807,38 @@ static struct irq_chip lguest_irq_controller = {
.unmask = enable_lguest_irq,
};
/* This sets up the Interrupt Descriptor Table (IDT) entry for each hardware
/*
* This sets up the Interrupt Descriptor Table (IDT) entry for each hardware
* interrupt (except 128, which is used for system calls), and then tells the
* Linux infrastructure that each interrupt is controlled by our level-based
* lguest interrupt controller. */
* lguest interrupt controller.
*/
static void __init lguest_init_IRQ(void)
{
unsigned int i;
for (i = FIRST_EXTERNAL_VECTOR; i < NR_VECTORS; i++) {
/* Some systems map "vectors" to interrupts weirdly. Lguest has
* a straightforward 1 to 1 mapping, so force that here. */
/* Some systems map "vectors" to interrupts weirdly. Not us! */
__get_cpu_var(vector_irq)[i] = i - FIRST_EXTERNAL_VECTOR;
if (i != SYSCALL_VECTOR)
set_intr_gate(i, interrupt[i - FIRST_EXTERNAL_VECTOR]);
}
/* This call is required to set up for 4k stacks, where we have
* separate stacks for hard and soft interrupts. */
/*
* This call is required to set up for 4k stacks, where we have
* separate stacks for hard and soft interrupts.
*/
irq_ctx_init(smp_processor_id());
}
/*
* With CONFIG_SPARSE_IRQ, interrupt descriptors are allocated as-needed, so
* rather than set them in lguest_init_IRQ we are called here every time an
* lguest device needs an interrupt.
*
* FIXME: irq_to_desc_alloc_node() can fail due to lack of memory, we should
* pass that up!
*/
void lguest_setup_irq(unsigned int irq)
{
irq_to_desc_alloc_node(irq, 0);
......@@ -729,31 +857,39 @@ static unsigned long lguest_get_wallclock(void)
return lguest_data.time.tv_sec;
}
/* The TSC is an Intel thing called the Time Stamp Counter. The Host tells us
/*
* The TSC is an Intel thing called the Time Stamp Counter. The Host tells us
* what speed it runs at, or 0 if it's unusable as a reliable clock source.
* This matches what we want here: if we return 0 from this function, the x86
* TSC clock will give up and not register itself. */
* TSC clock will give up and not register itself.
*/
static unsigned long lguest_tsc_khz(void)
{
return lguest_data.tsc_khz;
}
/* If we can't use the TSC, the kernel falls back to our lower-priority
* "lguest_clock", where we read the time value given to us by the Host. */
/*
* If we can't use the TSC, the kernel falls back to our lower-priority
* "lguest_clock", where we read the time value given to us by the Host.
*/
static cycle_t lguest_clock_read(struct clocksource *cs)
{
unsigned long sec, nsec;
/* Since the time is in two parts (seconds and nanoseconds), we risk
/*
* Since the time is in two parts (seconds and nanoseconds), we risk
* reading it just as it's changing from 99 & 0.999999999 to 100 and 0,
* and getting 99 and 0. As Linux tends to come apart under the stress
* of time travel, we must be careful: */
* of time travel, we must be careful:
*/
do {
/* First we read the seconds part. */
sec = lguest_data.time.tv_sec;
/* This read memory barrier tells the compiler and the CPU that
/*
* This read memory barrier tells the compiler and the CPU that
* this can't be reordered: we have to complete the above
* before going on. */
* before going on.
*/
rmb();
/* Now we read the nanoseconds part. */
nsec = lguest_data.time.tv_nsec;
......@@ -777,9 +913,11 @@ static struct clocksource lguest_clock = {
.flags = CLOCK_SOURCE_IS_CONTINUOUS,
};
/* We also need a "struct clock_event_device": Linux asks us to set it to go
/*
* We also need a "struct clock_event_device": Linux asks us to set it to go
* off some time in the future. Actually, James Morris figured all this out, I
* just applied the patch. */
* just applied the patch.
*/
static int lguest_clockevent_set_next_event(unsigned long delta,
struct clock_event_device *evt)
{
......@@ -829,8 +967,10 @@ static struct clock_event_device lguest_clockevent = {
.max_delta_ns = LG_CLOCK_MAX_DELTA,
};
/* This is the Guest timer interrupt handler (hardware interrupt 0). We just
* call the clockevent infrastructure and it does whatever needs doing. */
/*
* This is the Guest timer interrupt handler (hardware interrupt 0). We just
* call the clockevent infrastructure and it does whatever needs doing.
*/
static void lguest_time_irq(unsigned int irq, struct irq_desc *desc)
{
unsigned long flags;
......@@ -841,10 +981,12 @@ static void lguest_time_irq(unsigned int irq, struct irq_desc *desc)
local_irq_restore(flags);
}
/* At some point in the boot process, we get asked to set up our timing
/*
* At some point in the boot process, we get asked to set up our timing
* infrastructure. The kernel doesn't expect timer interrupts before this, but
* we cleverly initialized the "blocked_interrupts" field of "struct
* lguest_data" so that timer interrupts were blocked until now. */
* lguest_data" so that timer interrupts were blocked until now.
*/
static void lguest_time_init(void)
{
/* Set up the timer interrupt (0) to go to our simple timer routine */
......@@ -868,14 +1010,16 @@ static void lguest_time_init(void)
* to work. They're pretty simple.
*/
/* The Guest needs to tell the Host what stack it expects traps to use. For
/*
* The Guest needs to tell the Host what stack it expects traps to use. For
* native hardware, this is part of the Task State Segment mentioned above in
* lguest_load_tr_desc(), but to help hypervisors there's this special call.
*
* We tell the Host the segment we want to use (__KERNEL_DS is the kernel data
* segment), the privilege level (we're privilege level 1, the Host is 0 and
* will not tolerate us trying to use that), the stack pointer, and the number
* of pages in the stack. */
* of pages in the stack.
*/
static void lguest_load_sp0(struct tss_struct *tss,
struct thread_struct *thread)
{
......@@ -889,7 +1033,8 @@ static void lguest_set_debugreg(int regno, unsigned long value)
/* FIXME: Implement */
}
/* There are times when the kernel wants to make sure that no memory writes are
/*
* There are times when the kernel wants to make sure that no memory writes are
* caught in the cache (that they've all reached real hardware devices). This
* doesn't matter for the Guest which has virtual hardware.
*
......@@ -903,11 +1048,13 @@ static void lguest_wbinvd(void)
{
}
/* If the Guest expects to have an Advanced Programmable Interrupt Controller,
/*
* If the Guest expects to have an Advanced Programmable Interrupt Controller,
* we play dumb by ignoring writes and returning 0 for reads. So it's no
* longer Programmable nor Controlling anything, and I don't think 8 lines of
* code qualifies for Advanced. It will also never interrupt anything. It
* does, however, allow us to get through the Linux boot code. */
* does, however, allow us to get through the Linux boot code.
*/
#ifdef CONFIG_X86_LOCAL_APIC
static void lguest_apic_write(u32 reg, u32 v)
{
......@@ -956,11 +1103,13 @@ static void lguest_safe_halt(void)
kvm_hypercall0(LHCALL_HALT);
}
/* The SHUTDOWN hypercall takes a string to describe what's happening, and
/*
* The SHUTDOWN hypercall takes a string to describe what's happening, and
* an argument which says whether this to restart (reboot) the Guest or not.
*
* Note that the Host always prefers that the Guest speak in physical addresses
* rather than virtual addresses, so we use __pa() here. */
* rather than virtual addresses, so we use __pa() here.
*/
static void lguest_power_off(void)
{
kvm_hypercall2(LHCALL_SHUTDOWN, __pa("Power down"),
......@@ -991,8 +1140,10 @@ static __init char *lguest_memory_setup(void)
* nice to move it back to lguest_init. Patch welcome... */
atomic_notifier_chain_register(&panic_notifier_list, &paniced);
/* The Linux bootloader header contains an "e820" memory map: the
* Launcher populated the first entry with our memory limit. */
/*
*The Linux bootloader header contains an "e820" memory map: the
* Launcher populated the first entry with our memory limit.
*/
e820_add_region(boot_params.e820_map[0].addr,
boot_params.e820_map[0].size,
boot_params.e820_map[0].type);
......@@ -1001,16 +1152,17 @@ static __init char *lguest_memory_setup(void)
return "LGUEST";
}
/* We will eventually use the virtio console device to produce console output,
/*
* We will eventually use the virtio console device to produce console output,
* but before that is set up we use LHCALL_NOTIFY on normal memory to produce
* console output. */
* console output.
*/
static __init int early_put_chars(u32 vtermno, const char *buf, int count)
{
char scratch[17];
unsigned int len = count;
/* We use a nul-terminated string, so we have to make a copy. Icky,
* huh? */
/* We use a nul-terminated string, so we make a copy. Icky, huh? */
if (len > sizeof(scratch) - 1)
len = sizeof(scratch) - 1;
scratch[len] = '\0';
......@@ -1021,8 +1173,10 @@ static __init int early_put_chars(u32 vtermno, const char *buf, int count)
return len;
}
/* Rebooting also tells the Host we're finished, but the RESTART flag tells the
* Launcher to reboot us. */
/*
* Rebooting also tells the Host we're finished, but the RESTART flag tells the
* Launcher to reboot us.
*/
static void lguest_restart(char *reason)
{
kvm_hypercall2(LHCALL_SHUTDOWN, __pa(reason), LGUEST_SHUTDOWN_RESTART);
......@@ -1049,7 +1203,8 @@ static void lguest_restart(char *reason)
* fit comfortably.
*
* First we need assembly templates of each of the patchable Guest operations,
* and these are in i386_head.S. */
* and these are in i386_head.S.
*/
/*G:060 We construct a table from the assembler templates: */
static const struct lguest_insns
......@@ -1060,9 +1215,11 @@ static const struct lguest_insns
[PARAVIRT_PATCH(pv_irq_ops.save_fl)] = { lgstart_pushf, lgend_pushf },
};
/* Now our patch routine is fairly simple (based on the native one in
/*
* Now our patch routine is fairly simple (based on the native one in
* paravirt.c). If we have a replacement, we copy it in and return how much of
* the available space we used. */
* the available space we used.
*/
static unsigned lguest_patch(u8 type, u16 clobber, void *ibuf,
unsigned long addr, unsigned len)
{
......@@ -1074,8 +1231,7 @@ static unsigned lguest_patch(u8 type, u16 clobber, void *ibuf,
insn_len = lguest_insns[type].end - lguest_insns[type].start;
/* Similarly if we can't fit replacement (shouldn't happen, but let's
* be thorough). */
/* Similarly if it can't fit (doesn't happen, but let's be thorough). */
if (len < insn_len)
return paravirt_patch_default(type, clobber, ibuf, addr, len);
......@@ -1084,22 +1240,28 @@ static unsigned lguest_patch(u8 type, u16 clobber, void *ibuf,
return insn_len;
}
/*G:029 Once we get to lguest_init(), we know we're a Guest. The various
/*G:029
* Once we get to lguest_init(), we know we're a Guest. The various
* pv_ops structures in the kernel provide points for (almost) every routine we
* have to override to avoid privileged instructions. */
* have to override to avoid privileged instructions.
*/
__init void lguest_init(void)
{
/* We're under lguest, paravirt is enabled, and we're running at
* privilege level 1, not 0 as normal. */
/* We're under lguest. */
pv_info.name = "lguest";
/* Paravirt is enabled. */
pv_info.paravirt_enabled = 1;
/* We're running at privilege level 1, not 0 as normal. */
pv_info.kernel_rpl = 1;
/* Everyone except Xen runs with this set. */
pv_info.shared_kernel_pmd = 1;
/* We set up all the lguest overrides for sensitive operations. These
* are detailed with the operations themselves. */
/*
* We set up all the lguest overrides for sensitive operations. These
* are detailed with the operations themselves.
*/
/* interrupt-related operations */
/* Interrupt-related operations */
pv_irq_ops.init_IRQ = lguest_init_IRQ;
pv_irq_ops.save_fl = PV_CALLEE_SAVE(save_fl);
pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(lg_restore_fl);
......@@ -1107,11 +1269,11 @@ __init void lguest_init(void)
pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(lg_irq_enable);
pv_irq_ops.safe_halt = lguest_safe_halt;
/* init-time operations */
/* Setup operations */
pv_init_ops.memory_setup = lguest_memory_setup;
pv_init_ops.patch = lguest_patch;
/* Intercepts of various cpu instructions */
/* Intercepts of various CPU instructions */
pv_cpu_ops.load_gdt = lguest_load_gdt;
pv_cpu_ops.cpuid = lguest_cpuid;
pv_cpu_ops.load_idt = lguest_load_idt;
......@@ -1132,7 +1294,7 @@ __init void lguest_init(void)
pv_cpu_ops.start_context_switch = paravirt_start_context_switch;
pv_cpu_ops.end_context_switch = lguest_end_context_switch;
/* pagetable management */
/* Pagetable management */
pv_mmu_ops.write_cr3 = lguest_write_cr3;
pv_mmu_ops.flush_tlb_user = lguest_flush_tlb_user;
pv_mmu_ops.flush_tlb_single = lguest_flush_tlb_single;
......@@ -1154,54 +1316,71 @@ __init void lguest_init(void)
pv_mmu_ops.pte_update_defer = lguest_pte_update;
#ifdef CONFIG_X86_LOCAL_APIC
/* apic read/write intercepts */
/* APIC read/write intercepts */
set_lguest_basic_apic_ops();
#endif
/* time operations */
/* Time operations */
pv_time_ops.get_wallclock = lguest_get_wallclock;
pv_time_ops.time_init = lguest_time_init;
pv_time_ops.get_tsc_khz = lguest_tsc_khz;
/* Now is a good time to look at the implementations of these functions
* before returning to the rest of lguest_init(). */
/*
* Now is a good time to look at the implementations of these functions
* before returning to the rest of lguest_init().
*/
/*G:070 Now we've seen all the paravirt_ops, we return to
/*G:070
* Now we've seen all the paravirt_ops, we return to
* lguest_init() where the rest of the fairly chaotic boot setup
* occurs. */
* occurs.
*/
/* The stack protector is a weird thing where gcc places a canary
/*
* The stack protector is a weird thing where gcc places a canary
* value on the stack and then checks it on return. This file is
* compiled with -fno-stack-protector it, so we got this far without
* problems. The value of the canary is kept at offset 20 from the
* %gs register, so we need to set that up before calling C functions
* in other files. */
* in other files.
*/
setup_stack_canary_segment(0);
/* We could just call load_stack_canary_segment(), but we might as
* call switch_to_new_gdt() which loads the whole table and sets up
* the per-cpu segment descriptor register %fs as well. */
/*
* We could just call load_stack_canary_segment(), but we might as well
* call switch_to_new_gdt() which loads the whole table and sets up the
* per-cpu segment descriptor register %fs as well.
*/
switch_to_new_gdt(0);
/* As described in head_32.S, we map the first 128M of memory. */
/* We actually boot with all memory mapped, but let's say 128MB. */
max_pfn_mapped = (128*1024*1024) >> PAGE_SHIFT;
/* The Host<->Guest Switcher lives at the top of our address space, and
/*
* The Host<->Guest Switcher lives at the top of our address space, and
* the Host told us how big it is when we made LGUEST_INIT hypercall:
* it put the answer in lguest_data.reserve_mem */
* it put the answer in lguest_data.reserve_mem
*/
reserve_top_address(lguest_data.reserve_mem);
/* If we don't initialize the lock dependency checker now, it crashes
* paravirt_disable_iospace. */
/*
* If we don't initialize the lock dependency checker now, it crashes
* paravirt_disable_iospace.
*/
lockdep_init();
/* The IDE code spends about 3 seconds probing for disks: if we reserve
/*
* The IDE code spends about 3 seconds probing for disks: if we reserve
* all the I/O ports up front it can't get them and so doesn't probe.
* Other device drivers are similar (but less severe). This cuts the
* kernel boot time on my machine from 4.1 seconds to 0.45 seconds. */
* kernel boot time on my machine from 4.1 seconds to 0.45 seconds.
*/
paravirt_disable_iospace();
/* This is messy CPU setup stuff which the native boot code does before
* start_kernel, so we have to do, too: */
/*
* This is messy CPU setup stuff which the native boot code does before
* start_kernel, so we have to do, too:
*/
cpu_detect(&new_cpu_data);
/* head.S usually sets up the first capability word, so do it here. */
new_cpu_data.x86_capability[0] = cpuid_edx(1);
......@@ -1218,22 +1397,28 @@ __init void lguest_init(void)
acpi_ht = 0;
#endif
/* We set the preferred console to "hvc". This is the "hypervisor
/*
* We set the preferred console to "hvc". This is the "hypervisor
* virtual console" driver written by the PowerPC people, which we also
* adapted for lguest's use. */
* adapted for lguest's use.
*/
add_preferred_console("hvc", 0, NULL);
/* Register our very early console. */
virtio_cons_early_init(early_put_chars);
/* Last of all, we set the power management poweroff hook to point to
/*
* Last of all, we set the power management poweroff hook to point to
* the Guest routine to power off, and the reboot hook to our restart
* routine. */
* routine.
*/
pm_power_off = lguest_power_off;
machine_ops.restart = lguest_restart;
/* Now we're set up, call i386_start_kernel() in head32.c and we proceed
* to boot as normal. It never returns. */
/*
* Now we're set up, call i386_start_kernel() in head32.c and we proceed
* to boot as normal. It never returns.
*/
i386_start_kernel();
}
/*
......
......@@ -5,7 +5,8 @@
#include <asm/thread_info.h>
#include <asm/processor-flags.h>
/*G:020 Our story starts with the kernel booting into startup_32 in
/*G:020
* Our story starts with the kernel booting into startup_32 in
* arch/x86/kernel/head_32.S. It expects a boot header, which is created by
* the bootloader (the Launcher in our case).
*
......@@ -21,11 +22,14 @@
* data without remembering to subtract __PAGE_OFFSET!
*
* The .section line puts this code in .init.text so it will be discarded after
* boot. */
* boot.
*/
.section .init.text, "ax", @progbits
ENTRY(lguest_entry)
/* We make the "initialization" hypercall now to tell the Host about
* us, and also find out where it put our page tables. */
/*
* We make the "initialization" hypercall now to tell the Host about
* us, and also find out where it put our page tables.
*/
movl $LHCALL_LGUEST_INIT, %eax
movl $lguest_data - __PAGE_OFFSET, %ebx
.byte 0x0f,0x01,0xc1 /* KVM_HYPERCALL */
......@@ -33,13 +37,14 @@ ENTRY(lguest_entry)
/* Set up the initial stack so we can run C code. */
movl $(init_thread_union+THREAD_SIZE),%esp
/* Jumps are relative, and we're running __PAGE_OFFSET too low at the
* moment. */
/* Jumps are relative: we're running __PAGE_OFFSET too low. */
jmp lguest_init+__PAGE_OFFSET
/*G:055 We create a macro which puts the assembler code between lgstart_ and
* lgend_ markers. These templates are put in the .text section: they can't be
* discarded after boot as we may need to patch modules, too. */
/*G:055
* We create a macro which puts the assembler code between lgstart_ and lgend_
* markers. These templates are put in the .text section: they can't be
* discarded after boot as we may need to patch modules, too.
*/
.text
#define LGUEST_PATCH(name, insns...) \
lgstart_##name: insns; lgend_##name:; \
......@@ -48,83 +53,103 @@ ENTRY(lguest_entry)
LGUEST_PATCH(cli, movl $0, lguest_data+LGUEST_DATA_irq_enabled)
LGUEST_PATCH(pushf, movl lguest_data+LGUEST_DATA_irq_enabled, %eax)
/*G:033 But using those wrappers is inefficient (we'll see why that doesn't
* matter for save_fl and irq_disable later). If we write our routines
* carefully in assembler, we can avoid clobbering any registers and avoid
* jumping through the wrapper functions.
/*G:033
* But using those wrappers is inefficient (we'll see why that doesn't matter
* for save_fl and irq_disable later). If we write our routines carefully in
* assembler, we can avoid clobbering any registers and avoid jumping through
* the wrapper functions.
*
* I skipped over our first piece of assembler, but this one is worth studying
* in a bit more detail so I'll describe in easy stages. First, the routine
* to enable interrupts: */
* in a bit more detail so I'll describe in easy stages. First, the routine to
* enable interrupts:
*/
ENTRY(lg_irq_enable)
/* The reverse of irq_disable, this sets lguest_data.irq_enabled to
* X86_EFLAGS_IF (ie. "Interrupts enabled"). */
/*
* The reverse of irq_disable, this sets lguest_data.irq_enabled to
* X86_EFLAGS_IF (ie. "Interrupts enabled").
*/
movl $X86_EFLAGS_IF, lguest_data+LGUEST_DATA_irq_enabled
/* But now we need to check if the Host wants to know: there might have
/*
* But now we need to check if the Host wants to know: there might have
* been interrupts waiting to be delivered, in which case it will have
* set lguest_data.irq_pending to X86_EFLAGS_IF. If it's not zero, we
* jump to send_interrupts, otherwise we're done. */
* jump to send_interrupts, otherwise we're done.
*/
testl $0, lguest_data+LGUEST_DATA_irq_pending
jnz send_interrupts
/* One cool thing about x86 is that you can do many things without using
/*
* One cool thing about x86 is that you can do many things without using
* a register. In this case, the normal path hasn't needed to save or
* restore any registers at all! */
* restore any registers at all!
*/
ret
send_interrupts:
/* OK, now we need a register: eax is used for the hypercall number,
/*
* OK, now we need a register: eax is used for the hypercall number,
* which is LHCALL_SEND_INTERRUPTS.
*
* We used not to bother with this pending detection at all, which was
* much simpler. Sooner or later the Host would realize it had to
* send us an interrupt. But that turns out to make performance 7
* times worse on a simple tcp benchmark. So now we do this the hard
* way. */
* way.
*/
pushl %eax
movl $LHCALL_SEND_INTERRUPTS, %eax
/* This is a vmcall instruction (same thing that KVM uses). Older
/*
* This is a vmcall instruction (same thing that KVM uses). Older
* assembler versions might not know the "vmcall" instruction, so we
* create one manually here. */
* create one manually here.
*/
.byte 0x0f,0x01,0xc1 /* KVM_HYPERCALL */
/* Put eax back the way we found it. */
popl %eax
ret
/* Finally, the "popf" or "restore flags" routine. The %eax register holds the
/*
* Finally, the "popf" or "restore flags" routine. The %eax register holds the
* flags (in practice, either X86_EFLAGS_IF or 0): if it's X86_EFLAGS_IF we're
* enabling interrupts again, if it's 0 we're leaving them off. */
* enabling interrupts again, if it's 0 we're leaving them off.
*/
ENTRY(lg_restore_fl)
/* This is just "lguest_data.irq_enabled = flags;" */
movl %eax, lguest_data+LGUEST_DATA_irq_enabled
/* Now, if the %eax value has enabled interrupts and
/*
* Now, if the %eax value has enabled interrupts and
* lguest_data.irq_pending is set, we want to tell the Host so it can
* deliver any outstanding interrupts. Fortunately, both values will
* be X86_EFLAGS_IF (ie. 512) in that case, and the "testl"
* instruction will AND them together for us. If both are set, we
* jump to send_interrupts. */
* jump to send_interrupts.
*/
testl lguest_data+LGUEST_DATA_irq_pending, %eax
jnz send_interrupts
/* Again, the normal path has used no extra registers. Clever, huh? */
ret
/*:*/
/* These demark the EIP range where host should never deliver interrupts. */
.global lguest_noirq_start
.global lguest_noirq_end
/*M:004 When the Host reflects a trap or injects an interrupt into the Guest,
* it sets the eflags interrupt bit on the stack based on
* lguest_data.irq_enabled, so the Guest iret logic does the right thing when
* restoring it. However, when the Host sets the Guest up for direct traps,
* such as system calls, the processor is the one to push eflags onto the
* stack, and the interrupt bit will be 1 (in reality, interrupts are always
* enabled in the Guest).
/*M:004
* When the Host reflects a trap or injects an interrupt into the Guest, it
* sets the eflags interrupt bit on the stack based on lguest_data.irq_enabled,
* so the Guest iret logic does the right thing when restoring it. However,
* when the Host sets the Guest up for direct traps, such as system calls, the
* processor is the one to push eflags onto the stack, and the interrupt bit
* will be 1 (in reality, interrupts are always enabled in the Guest).
*
* This turns out to be harmless: the only trap which should happen under Linux
* with interrupts disabled is Page Fault (due to our lazy mapping of vmalloc
* regions), which has to be reflected through the Host anyway. If another
* trap *does* go off when interrupts are disabled, the Guest will panic, and
* we'll never get to this iret! :*/
* we'll never get to this iret!
:*/
/*G:045 There is one final paravirt_op that the Guest implements, and glancing
* at it you can see why I left it to last. It's *cool*! It's in *assembler*!
/*G:045
* There is one final paravirt_op that the Guest implements, and glancing at it
* you can see why I left it to last. It's *cool*! It's in *assembler*!
*
* The "iret" instruction is used to return from an interrupt or trap. The
* stack looks like this:
......@@ -148,15 +173,18 @@ ENTRY(lg_restore_fl)
* return to userspace or wherever. Our solution to this is to surround the
* code with lguest_noirq_start: and lguest_noirq_end: labels. We tell the
* Host that it is *never* to interrupt us there, even if interrupts seem to be
* enabled. */
* enabled.
*/
ENTRY(lguest_iret)
pushl %eax
movl 12(%esp), %eax
lguest_noirq_start:
/* Note the %ss: segment prefix here. Normal data accesses use the
/*
* Note the %ss: segment prefix here. Normal data accesses use the
* "ds" segment, but that will have already been restored for whatever
* we're returning to (such as userspace): we can't trust it. The %ss:
* prefix makes sure we use the stack segment, which is still valid. */
* prefix makes sure we use the stack segment, which is still valid.
*/
movl %eax,%ss:lguest_data+LGUEST_DATA_irq_enabled
popl %eax
iret
......
/*P:400 This contains run_guest() which actually calls into the Host<->Guest
/*P:400
* This contains run_guest() which actually calls into the Host<->Guest
* Switcher and analyzes the return, such as determining if the Guest wants the
* Host to do something. This file also contains useful helper routines. :*/
* Host to do something. This file also contains useful helper routines.
:*/
#include <linux/module.h>
#include <linux/stringify.h>
#include <linux/stddef.h>
......@@ -24,7 +26,8 @@ static struct page **switcher_page;
/* This One Big lock protects all inter-guest data structures. */
DEFINE_MUTEX(lguest_lock);
/*H:010 We need to set up the Switcher at a high virtual address. Remember the
/*H:010
* We need to set up the Switcher at a high virtual address. Remember the
* Switcher is a few hundred bytes of assembler code which actually changes the
* CPU to run the Guest, and then changes back to the Host when a trap or
* interrupt happens.
......@@ -33,7 +36,8 @@ DEFINE_MUTEX(lguest_lock);
* Host since it will be running as the switchover occurs.
*
* Trying to map memory at a particular address is an unusual thing to do, so
* it's not a simple one-liner. */
* it's not a simple one-liner.
*/
static __init int map_switcher(void)
{
int i, err;
......@@ -47,8 +51,10 @@ static __init int map_switcher(void)
* easy.
*/
/* We allocate an array of struct page pointers. map_vm_area() wants
* this, rather than just an array of pages. */
/*
* We allocate an array of struct page pointers. map_vm_area() wants
* this, rather than just an array of pages.
*/
switcher_page = kmalloc(sizeof(switcher_page[0])*TOTAL_SWITCHER_PAGES,
GFP_KERNEL);
if (!switcher_page) {
......@@ -56,8 +62,10 @@ static __init int map_switcher(void)
goto out;
}
/* Now we actually allocate the pages. The Guest will see these pages,
* so we make sure they're zeroed. */
/*
* Now we actually allocate the pages. The Guest will see these pages,
* so we make sure they're zeroed.
*/
for (i = 0; i < TOTAL_SWITCHER_PAGES; i++) {
unsigned long addr = get_zeroed_page(GFP_KERNEL);
if (!addr) {
......@@ -67,19 +75,23 @@ static __init int map_switcher(void)
switcher_page[i] = virt_to_page(addr);
}
/* First we check that the Switcher won't overlap the fixmap area at
/*
* First we check that the Switcher won't overlap the fixmap area at
* the top of memory. It's currently nowhere near, but it could have
* very strange effects if it ever happened. */
* very strange effects if it ever happened.
*/
if (SWITCHER_ADDR + (TOTAL_SWITCHER_PAGES+1)*PAGE_SIZE > FIXADDR_START){
err = -ENOMEM;
printk("lguest: mapping switcher would thwack fixmap\n");
goto free_pages;
}
/* Now we reserve the "virtual memory area" we want: 0xFFC00000
/*
* Now we reserve the "virtual memory area" we want: 0xFFC00000
* (SWITCHER_ADDR). We might not get it in theory, but in practice
* it's worked so far. The end address needs +1 because __get_vm_area
* allocates an extra guard page, so we need space for that. */
* allocates an extra guard page, so we need space for that.
*/
switcher_vma = __get_vm_area(TOTAL_SWITCHER_PAGES * PAGE_SIZE,
VM_ALLOC, SWITCHER_ADDR, SWITCHER_ADDR
+ (TOTAL_SWITCHER_PAGES+1) * PAGE_SIZE);
......@@ -89,11 +101,13 @@ static __init int map_switcher(void)
goto free_pages;
}
/* This code actually sets up the pages we've allocated to appear at
/*
* This code actually sets up the pages we've allocated to appear at
* SWITCHER_ADDR. map_vm_area() takes the vma we allocated above, the
* kind of pages we're mapping (kernel pages), and a pointer to our
* array of struct pages. It increments that pointer, but we don't
* care. */
* care.
*/
pagep = switcher_page;
err = map_vm_area(switcher_vma, PAGE_KERNEL_EXEC, &pagep);
if (err) {
......@@ -101,8 +115,10 @@ static __init int map_switcher(void)
goto free_vma;
}
/* Now the Switcher is mapped at the right address, we can't fail!
* Copy in the compiled-in Switcher code (from <arch>_switcher.S). */
/*
* Now the Switcher is mapped at the right address, we can't fail!
* Copy in the compiled-in Switcher code (from <arch>_switcher.S).
*/
memcpy(switcher_vma->addr, start_switcher_text,
end_switcher_text - start_switcher_text);
......@@ -124,8 +140,7 @@ out:
}
/*:*/
/* Cleaning up the mapping when the module is unloaded is almost...
* too easy. */
/* Cleaning up the mapping when the module is unloaded is almost... too easy. */
static void unmap_switcher(void)
{
unsigned int i;
......@@ -151,16 +166,19 @@ static void unmap_switcher(void)
* But we can't trust the Guest: it might be trying to access the Launcher
* code. We have to check that the range is below the pfn_limit the Launcher
* gave us. We have to make sure that addr + len doesn't give us a false
* positive by overflowing, too. */
* positive by overflowing, too.
*/
bool lguest_address_ok(const struct lguest *lg,
unsigned long addr, unsigned long len)
{
return (addr+len) / PAGE_SIZE < lg->pfn_limit && (addr+len >= addr);
}
/* This routine copies memory from the Guest. Here we can see how useful the
/*
* This routine copies memory from the Guest. Here we can see how useful the
* kill_lguest() routine we met in the Launcher can be: we return a random
* value (all zeroes) instead of needing to return an error. */
* value (all zeroes) instead of needing to return an error.
*/
void __lgread(struct lg_cpu *cpu, void *b, unsigned long addr, unsigned bytes)
{
if (!lguest_address_ok(cpu->lg, addr, bytes)
......@@ -181,9 +199,11 @@ void __lgwrite(struct lg_cpu *cpu, unsigned long addr, const void *b,
}
/*:*/
/*H:030 Let's jump straight to the the main loop which runs the Guest.
/*H:030
* Let's jump straight to the the main loop which runs the Guest.
* Remember, this is called by the Launcher reading /dev/lguest, and we keep
* going around and around until something interesting happens. */
* going around and around until something interesting happens.
*/
int run_guest(struct lg_cpu *cpu, unsigned long __user *user)
{
/* We stop running once the Guest is dead. */
......@@ -195,10 +215,17 @@ int run_guest(struct lg_cpu *cpu, unsigned long __user *user)
if (cpu->hcall)
do_hypercalls(cpu);
/* It's possible the Guest did a NOTIFY hypercall to the
* Launcher, in which case we return from the read() now. */
/*
* It's possible the Guest did a NOTIFY hypercall to the
* Launcher.
*/
if (cpu->pending_notify) {
/*
* Does it just needs to write to a registered
* eventfd (ie. the appropriate virtqueue thread)?
*/
if (!send_notify_to_eventfd(cpu)) {
/* OK, we tell the main Laucher. */
if (put_user(cpu->pending_notify, user))
return -EFAULT;
return sizeof(cpu->pending_notify);
......@@ -209,29 +236,39 @@ int run_guest(struct lg_cpu *cpu, unsigned long __user *user)
if (signal_pending(current))
return -ERESTARTSYS;
/* Check if there are any interrupts which can be delivered now:
/*
* Check if there are any interrupts which can be delivered now:
* if so, this sets up the hander to be executed when we next
* run the Guest. */
* run the Guest.
*/
irq = interrupt_pending(cpu, &more);
if (irq < LGUEST_IRQS)
try_deliver_interrupt(cpu, irq, more);
/* All long-lived kernel loops need to check with this horrible
/*
* All long-lived kernel loops need to check with this horrible
* thing called the freezer. If the Host is trying to suspend,
* it stops us. */
* it stops us.
*/
try_to_freeze();
/* Just make absolutely sure the Guest is still alive. One of
* those hypercalls could have been fatal, for example. */
/*
* Just make absolutely sure the Guest is still alive. One of
* those hypercalls could have been fatal, for example.
*/
if (cpu->lg->dead)
break;
/* If the Guest asked to be stopped, we sleep. The Guest's
* clock timer will wake us. */
/*
* If the Guest asked to be stopped, we sleep. The Guest's
* clock timer will wake us.
*/
if (cpu->halted) {
set_current_state(TASK_INTERRUPTIBLE);
/* Just before we sleep, make sure no interrupt snuck in
* which we should be doing. */
/*
* Just before we sleep, make sure no interrupt snuck in
* which we should be doing.
*/
if (interrupt_pending(cpu, &more) < LGUEST_IRQS)
set_current_state(TASK_RUNNING);
else
......@@ -239,8 +276,10 @@ int run_guest(struct lg_cpu *cpu, unsigned long __user *user)
continue;
}
/* OK, now we're ready to jump into the Guest. First we put up
* the "Do Not Disturb" sign: */
/*
* OK, now we're ready to jump into the Guest. First we put up
* the "Do Not Disturb" sign:
*/
local_irq_disable();
/* Actually run the Guest until something happens. */
......@@ -327,8 +366,10 @@ static void __exit fini(void)
}
/*:*/
/* The Host side of lguest can be a module. This is a nice way for people to
* play with it. */
/*
* The Host side of lguest can be a module. This is a nice way for people to
* play with it.
*/
module_init(init);
module_exit(fini);
MODULE_LICENSE("GPL");
......
/*P:500 Just as userspace programs request kernel operations through a system
/*P:500
* Just as userspace programs request kernel operations through a system
* call, the Guest requests Host operations through a "hypercall". You might
* notice this nomenclature doesn't really follow any logic, but the name has
* been around for long enough that we're stuck with it. As you'd expect, this
* code is basically a one big switch statement. :*/
* code is basically a one big switch statement.
:*/
/* Copyright (C) 2006 Rusty Russell IBM Corporation
......@@ -28,30 +30,41 @@
#include <asm/pgtable.h>
#include "lg.h"
/*H:120 This is the core hypercall routine: where the Guest gets what it wants.
* Or gets killed. Or, in the case of LHCALL_SHUTDOWN, both. */
/*H:120
* This is the core hypercall routine: where the Guest gets what it wants.
* Or gets killed. Or, in the case of LHCALL_SHUTDOWN, both.
*/
static void do_hcall(struct lg_cpu *cpu, struct hcall_args *args)
{
switch (args->arg0) {
case LHCALL_FLUSH_ASYNC:
/* This call does nothing, except by breaking out of the Guest
* it makes us process all the asynchronous hypercalls. */
/*
* This call does nothing, except by breaking out of the Guest
* it makes us process all the asynchronous hypercalls.
*/
break;
case LHCALL_SEND_INTERRUPTS:
/* This call does nothing too, but by breaking out of the Guest
* it makes us process any pending interrupts. */
/*
* This call does nothing too, but by breaking out of the Guest
* it makes us process any pending interrupts.
*/
break;
case LHCALL_LGUEST_INIT:
/* You can't get here unless you're already initialized. Don't
* do that. */
/*
* You can't get here unless you're already initialized. Don't
* do that.
*/
kill_guest(cpu, "already have lguest_data");
break;
case LHCALL_SHUTDOWN: {
/* Shutdown is such a trivial hypercall that we do it in four
* lines right here. */
char msg[128];
/* If the lgread fails, it will call kill_guest() itself; the
* kill_guest() with the message will be ignored. */
/*
* Shutdown is such a trivial hypercall that we do it in five
* lines right here.
*
* If the lgread fails, it will call kill_guest() itself; the
* kill_guest() with the message will be ignored.
*/
__lgread(cpu, msg, args->arg1, sizeof(msg));
msg[sizeof(msg)-1] = '\0';
kill_guest(cpu, "CRASH: %s", msg);
......@@ -60,16 +73,17 @@ static void do_hcall(struct lg_cpu *cpu, struct hcall_args *args)
break;
}
case LHCALL_FLUSH_TLB:
/* FLUSH_TLB comes in two flavors, depending on the
* argument: */
/* FLUSH_TLB comes in two flavors, depending on the argument: */
if (args->arg1)
guest_pagetable_clear_all(cpu);
else
guest_pagetable_flush_user(cpu);
break;
/* All these calls simply pass the arguments through to the right
* routines. */
/*
* All these calls simply pass the arguments through to the right
* routines.
*/
case LHCALL_NEW_PGTABLE:
guest_new_pagetable(cpu, args->arg1);
break;
......@@ -112,15 +126,16 @@ static void do_hcall(struct lg_cpu *cpu, struct hcall_args *args)
kill_guest(cpu, "Bad hypercall %li\n", args->arg0);
}
}
/*:*/
/*H:124 Asynchronous hypercalls are easy: we just look in the array in the
/*H:124
* Asynchronous hypercalls are easy: we just look in the array in the
* Guest's "struct lguest_data" to see if any new ones are marked "ready".
*
* We are careful to do these in order: obviously we respect the order the
* Guest put them in the ring, but we also promise the Guest that they will
* happen before any normal hypercall (which is why we check this before
* checking for a normal hcall). */
* checking for a normal hcall).
*/
static void do_async_hcalls(struct lg_cpu *cpu)
{
unsigned int i;
......@@ -133,22 +148,28 @@ static void do_async_hcalls(struct lg_cpu *cpu)
/* We process "struct lguest_data"s hcalls[] ring once. */
for (i = 0; i < ARRAY_SIZE(st); i++) {
struct hcall_args args;
/* We remember where we were up to from last time. This makes
/*
* We remember where we were up to from last time. This makes
* sure that the hypercalls are done in the order the Guest
* places them in the ring. */
* places them in the ring.
*/
unsigned int n = cpu->next_hcall;
/* 0xFF means there's no call here (yet). */
if (st[n] == 0xFF)
break;
/* OK, we have hypercall. Increment the "next_hcall" cursor,
* and wrap back to 0 if we reach the end. */
/*
* OK, we have hypercall. Increment the "next_hcall" cursor,
* and wrap back to 0 if we reach the end.
*/
if (++cpu->next_hcall == LHCALL_RING_SIZE)
cpu->next_hcall = 0;
/* Copy the hypercall arguments into a local copy of
* the hcall_args struct. */
/*
* Copy the hypercall arguments into a local copy of the
* hcall_args struct.
*/
if (copy_from_user(&args, &cpu->lg->lguest_data->hcalls[n],
sizeof(struct hcall_args))) {
kill_guest(cpu, "Fetching async hypercalls");
......@@ -164,19 +185,25 @@ static void do_async_hcalls(struct lg_cpu *cpu)
break;
}
/* Stop doing hypercalls if they want to notify the Launcher:
* it needs to service this first. */
/*
* Stop doing hypercalls if they want to notify the Launcher:
* it needs to service this first.
*/
if (cpu->pending_notify)
break;
}
}
/* Last of all, we look at what happens first of all. The very first time the
* Guest makes a hypercall, we end up here to set things up: */
/*
* Last of all, we look at what happens first of all. The very first time the
* Guest makes a hypercall, we end up here to set things up:
*/
static void initialize(struct lg_cpu *cpu)
{
/* You can't do anything until you're initialized. The Guest knows the
* rules, so we're unforgiving here. */
/*
* You can't do anything until you're initialized. The Guest knows the
* rules, so we're unforgiving here.
*/
if (cpu->hcall->arg0 != LHCALL_LGUEST_INIT) {
kill_guest(cpu, "hypercall %li before INIT", cpu->hcall->arg0);
return;
......@@ -185,32 +212,44 @@ static void initialize(struct lg_cpu *cpu)
if (lguest_arch_init_hypercalls(cpu))
kill_guest(cpu, "bad guest page %p", cpu->lg->lguest_data);
/* The Guest tells us where we're not to deliver interrupts by putting
* the range of addresses into "struct lguest_data". */
/*
* The Guest tells us where we're not to deliver interrupts by putting
* the range of addresses into "struct lguest_data".
*/
if (get_user(cpu->lg->noirq_start, &cpu->lg->lguest_data->noirq_start)
|| get_user(cpu->lg->noirq_end, &cpu->lg->lguest_data->noirq_end))
kill_guest(cpu, "bad guest page %p", cpu->lg->lguest_data);
/* We write the current time into the Guest's data page once so it can
* set its clock. */
/*
* We write the current time into the Guest's data page once so it can
* set its clock.
*/
write_timestamp(cpu);
/* page_tables.c will also do some setup. */
page_table_guest_data_init(cpu);
/* This is the one case where the above accesses might have been the
/*
* This is the one case where the above accesses might have been the
* first write to a Guest page. This may have caused a copy-on-write
* fault, but the old page might be (read-only) in the Guest
* pagetable. */
* pagetable.
*/
guest_pagetable_clear_all(cpu);
}
/*:*/
/*M:013 If a Guest reads from a page (so creates a mapping) that it has never
/*M:013
* If a Guest reads from a page (so creates a mapping) that it has never
* written to, and then the Launcher writes to it (ie. the output of a virtual
* device), the Guest will still see the old page. In practice, this never
* happens: why would the Guest read a page which it has never written to? But
* a similar scenario might one day bite us, so it's worth mentioning. :*/
* a similar scenario might one day bite us, so it's worth mentioning.
*
* Note that if we used a shared anonymous mapping in the Launcher instead of
* mapping /dev/zero private, we wouldn't worry about cop-on-write. And we
* need that to switch the Launcher to processes (away from threads) anyway.
:*/
/*H:100
* Hypercalls
......@@ -229,17 +268,22 @@ void do_hypercalls(struct lg_cpu *cpu)
return;
}
/* The Guest has initialized.
/*
* The Guest has initialized.
*
* Look in the hypercall ring for the async hypercalls: */
* Look in the hypercall ring for the async hypercalls:
*/
do_async_hcalls(cpu);
/* If we stopped reading the hypercall ring because the Guest did a
/*
* If we stopped reading the hypercall ring because the Guest did a
* NOTIFY to the Launcher, we want to return now. Otherwise we do
* the hypercall. */
* the hypercall.
*/
if (!cpu->pending_notify) {
do_hcall(cpu, cpu->hcall);
/* Tricky point: we reset the hcall pointer to mark the
/*
* Tricky point: we reset the hcall pointer to mark the
* hypercall as "done". We use the hcall pointer rather than
* the trap number to indicate a hypercall is pending.
* Normally it doesn't matter: the Guest will run again and
......@@ -248,13 +292,16 @@ void do_hypercalls(struct lg_cpu *cpu)
* However, if we are signalled or the Guest sends I/O to the
* Launcher, the run_guest() loop will exit without running the
* Guest. When it comes back it would try to re-run the
* hypercall. Finding that bug sucked. */
* hypercall. Finding that bug sucked.
*/
cpu->hcall = NULL;
}
}
/* This routine supplies the Guest with time: it's used for wallclock time at
* initial boot and as a rough time source if the TSC isn't available. */
/*
* This routine supplies the Guest with time: it's used for wallclock time at
* initial boot and as a rough time source if the TSC isn't available.
*/
void write_timestamp(struct lg_cpu *cpu)
{
struct timespec now;
......
/*P:800 Interrupts (traps) are complicated enough to earn their own file.
/*P:800
* Interrupts (traps) are complicated enough to earn their own file.
* There are three classes of interrupts:
*
* 1) Real hardware interrupts which occur while we're running the Guest,
......@@ -10,7 +11,8 @@
* just like real hardware would deliver them. Traps from the Guest can be set
* up to go directly back into the Guest, but sometimes the Host wants to see
* them first, so we also have a way of "reflecting" them into the Guest as if
* they had been delivered to it directly. :*/
* they had been delivered to it directly.
:*/
#include <linux/uaccess.h>
#include <linux/interrupt.h>
#include <linux/module.h>
......@@ -26,8 +28,10 @@ static unsigned long idt_address(u32 lo, u32 hi)
return (lo & 0x0000FFFF) | (hi & 0xFFFF0000);
}
/* The "type" of the interrupt handler is a 4 bit field: we only support a
* couple of types. */
/*
* The "type" of the interrupt handler is a 4 bit field: we only support a
* couple of types.
*/
static int idt_type(u32 lo, u32 hi)
{
return (hi >> 8) & 0xF;
......@@ -39,8 +43,10 @@ static bool idt_present(u32 lo, u32 hi)
return (hi & 0x8000);
}
/* We need a helper to "push" a value onto the Guest's stack, since that's a
* big part of what delivering an interrupt does. */
/*
* We need a helper to "push" a value onto the Guest's stack, since that's a
* big part of what delivering an interrupt does.
*/
static void push_guest_stack(struct lg_cpu *cpu, unsigned long *gstack, u32 val)
{
/* Stack grows upwards: move stack then write value. */
......@@ -48,7 +54,8 @@ static void push_guest_stack(struct lg_cpu *cpu, unsigned long *gstack, u32 val)
lgwrite(cpu, *gstack, u32, val);
}
/*H:210 The set_guest_interrupt() routine actually delivers the interrupt or
/*H:210
* The set_guest_interrupt() routine actually delivers the interrupt or
* trap. The mechanics of delivering traps and interrupts to the Guest are the
* same, except some traps have an "error code" which gets pushed onto the
* stack as well: the caller tells us if this is one.
......@@ -59,7 +66,8 @@ static void push_guest_stack(struct lg_cpu *cpu, unsigned long *gstack, u32 val)
*
* We set up the stack just like the CPU does for a real interrupt, so it's
* identical for the Guest (and the standard "iret" instruction will undo
* it). */
* it).
*/
static void set_guest_interrupt(struct lg_cpu *cpu, u32 lo, u32 hi,
bool has_err)
{
......@@ -67,20 +75,26 @@ static void set_guest_interrupt(struct lg_cpu *cpu, u32 lo, u32 hi,
u32 eflags, ss, irq_enable;
unsigned long virtstack;
/* There are two cases for interrupts: one where the Guest is already
/*
* There are two cases for interrupts: one where the Guest is already
* in the kernel, and a more complex one where the Guest is in
* userspace. We check the privilege level to find out. */
* userspace. We check the privilege level to find out.
*/
if ((cpu->regs->ss&0x3) != GUEST_PL) {
/* The Guest told us their kernel stack with the SET_STACK
* hypercall: both the virtual address and the segment */
/*
* The Guest told us their kernel stack with the SET_STACK
* hypercall: both the virtual address and the segment.
*/
virtstack = cpu->esp1;
ss = cpu->ss1;
origstack = gstack = guest_pa(cpu, virtstack);
/* We push the old stack segment and pointer onto the new
/*
* We push the old stack segment and pointer onto the new
* stack: when the Guest does an "iret" back from the interrupt
* handler the CPU will notice they're dropping privilege
* levels and expect these here. */
* levels and expect these here.
*/
push_guest_stack(cpu, &gstack, cpu->regs->ss);
push_guest_stack(cpu, &gstack, cpu->regs->esp);
} else {
......@@ -91,18 +105,22 @@ static void set_guest_interrupt(struct lg_cpu *cpu, u32 lo, u32 hi,
origstack = gstack = guest_pa(cpu, virtstack);
}
/* Remember that we never let the Guest actually disable interrupts, so
/*
* Remember that we never let the Guest actually disable interrupts, so
* the "Interrupt Flag" bit is always set. We copy that bit from the
* Guest's "irq_enabled" field into the eflags word: we saw the Guest
* copy it back in "lguest_iret". */
* copy it back in "lguest_iret".
*/
eflags = cpu->regs->eflags;
if (get_user(irq_enable, &cpu->lg->lguest_data->irq_enabled) == 0
&& !(irq_enable & X86_EFLAGS_IF))
eflags &= ~X86_EFLAGS_IF;
/* An interrupt is expected to push three things on the stack: the old
/*
* An interrupt is expected to push three things on the stack: the old
* "eflags" word, the old code segment, and the old instruction
* pointer. */
* pointer.
*/
push_guest_stack(cpu, &gstack, eflags);
push_guest_stack(cpu, &gstack, cpu->regs->cs);
push_guest_stack(cpu, &gstack, cpu->regs->eip);
......@@ -111,15 +129,19 @@ static void set_guest_interrupt(struct lg_cpu *cpu, u32 lo, u32 hi,
if (has_err)
push_guest_stack(cpu, &gstack, cpu->regs->errcode);
/* Now we've pushed all the old state, we change the stack, the code
* segment and the address to execute. */
/*
* Now we've pushed all the old state, we change the stack, the code
* segment and the address to execute.
*/
cpu->regs->ss = ss;
cpu->regs->esp = virtstack + (gstack - origstack);
cpu->regs->cs = (__KERNEL_CS|GUEST_PL);
cpu->regs->eip = idt_address(lo, hi);
/* There are two kinds of interrupt handlers: 0xE is an "interrupt
* gate" which expects interrupts to be disabled on entry. */
/*
* There are two kinds of interrupt handlers: 0xE is an "interrupt
* gate" which expects interrupts to be disabled on entry.
*/
if (idt_type(lo, hi) == 0xE)
if (put_user(0, &cpu->lg->lguest_data->irq_enabled))
kill_guest(cpu, "Disabling interrupts");
......@@ -130,7 +152,8 @@ static void set_guest_interrupt(struct lg_cpu *cpu, u32 lo, u32 hi,
*
* interrupt_pending() returns the first pending interrupt which isn't blocked
* by the Guest. It is called before every entry to the Guest, and just before
* we go to sleep when the Guest has halted itself. */
* we go to sleep when the Guest has halted itself.
*/
unsigned int interrupt_pending(struct lg_cpu *cpu, bool *more)
{
unsigned int irq;
......@@ -140,8 +163,10 @@ unsigned int interrupt_pending(struct lg_cpu *cpu, bool *more)
if (!cpu->lg->lguest_data)
return LGUEST_IRQS;
/* Take our "irqs_pending" array and remove any interrupts the Guest
* wants blocked: the result ends up in "blk". */
/*
* Take our "irqs_pending" array and remove any interrupts the Guest
* wants blocked: the result ends up in "blk".
*/
if (copy_from_user(&blk, cpu->lg->lguest_data->blocked_interrupts,
sizeof(blk)))
return LGUEST_IRQS;
......@@ -154,16 +179,20 @@ unsigned int interrupt_pending(struct lg_cpu *cpu, bool *more)
return irq;
}
/* This actually diverts the Guest to running an interrupt handler, once an
* interrupt has been identified by interrupt_pending(). */
/*
* This actually diverts the Guest to running an interrupt handler, once an
* interrupt has been identified by interrupt_pending().
*/
void try_deliver_interrupt(struct lg_cpu *cpu, unsigned int irq, bool more)
{
struct desc_struct *idt;
BUG_ON(irq >= LGUEST_IRQS);
/* They may be in the middle of an iret, where they asked us never to
* deliver interrupts. */
/*
* They may be in the middle of an iret, where they asked us never to
* deliver interrupts.
*/
if (cpu->regs->eip >= cpu->lg->noirq_start &&
(cpu->regs->eip < cpu->lg->noirq_end))
return;
......@@ -187,29 +216,37 @@ void try_deliver_interrupt(struct lg_cpu *cpu, unsigned int irq, bool more)
}
}
/* Look at the IDT entry the Guest gave us for this interrupt. The
/*
* Look at the IDT entry the Guest gave us for this interrupt. The
* first 32 (FIRST_EXTERNAL_VECTOR) entries are for traps, so we skip
* over them. */
* over them.
*/
idt = &cpu->arch.idt[FIRST_EXTERNAL_VECTOR+irq];
/* If they don't have a handler (yet?), we just ignore it */
if (idt_present(idt->a, idt->b)) {
/* OK, mark it no longer pending and deliver it. */
clear_bit(irq, cpu->irqs_pending);
/* set_guest_interrupt() takes the interrupt descriptor and a
/*
* set_guest_interrupt() takes the interrupt descriptor and a
* flag to say whether this interrupt pushes an error code onto
* the stack as well: virtual interrupts never do. */
* the stack as well: virtual interrupts never do.
*/
set_guest_interrupt(cpu, idt->a, idt->b, false);
}
/* Every time we deliver an interrupt, we update the timestamp in the
/*
* Every time we deliver an interrupt, we update the timestamp in the
* Guest's lguest_data struct. It would be better for the Guest if we
* did this more often, but it can actually be quite slow: doing it
* here is a compromise which means at least it gets updated every
* timer interrupt. */
* timer interrupt.
*/
write_timestamp(cpu);
/* If there are no other interrupts we want to deliver, clear
* the pending flag. */
/*
* If there are no other interrupts we want to deliver, clear
* the pending flag.
*/
if (!more)
put_user(0, &cpu->lg->lguest_data->irq_pending);
}
......@@ -217,24 +254,29 @@ void try_deliver_interrupt(struct lg_cpu *cpu, unsigned int irq, bool more)
/* And this is the routine when we want to set an interrupt for the Guest. */
void set_interrupt(struct lg_cpu *cpu, unsigned int irq)
{
/* Next time the Guest runs, the core code will see if it can deliver
* this interrupt. */
/*
* Next time the Guest runs, the core code will see if it can deliver
* this interrupt.
*/
set_bit(irq, cpu->irqs_pending);
/* Make sure it sees it; it might be asleep (eg. halted), or
* running the Guest right now, in which case kick_process()
* will knock it out. */
/*
* Make sure it sees it; it might be asleep (eg. halted), or running
* the Guest right now, in which case kick_process() will knock it out.
*/
if (!wake_up_process(cpu->tsk))
kick_process(cpu->tsk);
}
/*:*/
/* Linux uses trap 128 for system calls. Plan9 uses 64, and Ron Minnich sent
/*
* Linux uses trap 128 for system calls. Plan9 uses 64, and Ron Minnich sent
* me a patch, so we support that too. It'd be a big step for lguest if half
* the Plan 9 user base were to start using it.
*
* Actually now I think of it, it's possible that Ron *is* half the Plan 9
* userbase. Oh well. */
* userbase. Oh well.
*/
static bool could_be_syscall(unsigned int num)
{
/* Normal Linux SYSCALL_VECTOR or reserved vector? */
......@@ -274,9 +316,11 @@ void free_interrupts(void)
clear_bit(syscall_vector, used_vectors);
}
/*H:220 Now we've got the routines to deliver interrupts, delivering traps like
/*H:220
* Now we've got the routines to deliver interrupts, delivering traps like
* page fault is easy. The only trick is that Intel decided that some traps
* should have error codes: */
* should have error codes:
*/
static bool has_err(unsigned int trap)
{
return (trap == 8 || (trap >= 10 && trap <= 14) || trap == 17);
......@@ -285,13 +329,17 @@ static bool has_err(unsigned int trap)
/* deliver_trap() returns true if it could deliver the trap. */
bool deliver_trap(struct lg_cpu *cpu, unsigned int num)
{
/* Trap numbers are always 8 bit, but we set an impossible trap number
* for traps inside the Switcher, so check that here. */
/*
* Trap numbers are always 8 bit, but we set an impossible trap number
* for traps inside the Switcher, so check that here.
*/
if (num >= ARRAY_SIZE(cpu->arch.idt))
return false;
/* Early on the Guest hasn't set the IDT entries (or maybe it put a
* bogus one in): if we fail here, the Guest will be killed. */
/*
* Early on the Guest hasn't set the IDT entries (or maybe it put a
* bogus one in): if we fail here, the Guest will be killed.
*/
if (!idt_present(cpu->arch.idt[num].a, cpu->arch.idt[num].b))
return false;
set_guest_interrupt(cpu, cpu->arch.idt[num].a,
......@@ -299,7 +347,8 @@ bool deliver_trap(struct lg_cpu *cpu, unsigned int num)
return true;
}
/*H:250 Here's the hard part: returning to the Host every time a trap happens
/*H:250
* Here's the hard part: returning to the Host every time a trap happens
* and then calling deliver_trap() and re-entering the Guest is slow.
* Particularly because Guest userspace system calls are traps (usually trap
* 128).
......@@ -311,69 +360,87 @@ bool deliver_trap(struct lg_cpu *cpu, unsigned int num)
* the other hypervisors would beat it up at lunchtime.
*
* This routine indicates if a particular trap number could be delivered
* directly. */
* directly.
*/
static bool direct_trap(unsigned int num)
{
/* Hardware interrupts don't go to the Guest at all (except system
* call). */
/*
* Hardware interrupts don't go to the Guest at all (except system
* call).
*/
if (num >= FIRST_EXTERNAL_VECTOR && !could_be_syscall(num))
return false;
/* The Host needs to see page faults (for shadow paging and to save the
/*
* The Host needs to see page faults (for shadow paging and to save the
* fault address), general protection faults (in/out emulation) and
* device not available (TS handling), invalid opcode fault (kvm hcall),
* and of course, the hypercall trap. */
* and of course, the hypercall trap.
*/
return num != 14 && num != 13 && num != 7 &&
num != 6 && num != LGUEST_TRAP_ENTRY;
}
/*:*/
/*M:005 The Guest has the ability to turn its interrupt gates into trap gates,
/*M:005
* The Guest has the ability to turn its interrupt gates into trap gates,
* if it is careful. The Host will let trap gates can go directly to the
* Guest, but the Guest needs the interrupts atomically disabled for an
* interrupt gate. It can do this by pointing the trap gate at instructions
* within noirq_start and noirq_end, where it can safely disable interrupts. */
* within noirq_start and noirq_end, where it can safely disable interrupts.
*/
/*M:006 The Guests do not use the sysenter (fast system call) instruction,
/*M:006
* The Guests do not use the sysenter (fast system call) instruction,
* because it's hardcoded to enter privilege level 0 and so can't go direct.
* It's about twice as fast as the older "int 0x80" system call, so it might
* still be worthwhile to handle it in the Switcher and lcall down to the
* Guest. The sysenter semantics are hairy tho: search for that keyword in
* entry.S :*/
* entry.S
:*/
/*H:260 When we make traps go directly into the Guest, we need to make sure
/*H:260
* When we make traps go directly into the Guest, we need to make sure
* the kernel stack is valid (ie. mapped in the page tables). Otherwise, the
* CPU trying to deliver the trap will fault while trying to push the interrupt
* words on the stack: this is called a double fault, and it forces us to kill
* the Guest.
*
* Which is deeply unfair, because (literally!) it wasn't the Guests' fault. */
* Which is deeply unfair, because (literally!) it wasn't the Guests' fault.
*/
void pin_stack_pages(struct lg_cpu *cpu)
{
unsigned int i;
/* Depending on the CONFIG_4KSTACKS option, the Guest can have one or
* two pages of stack space. */
/*
* Depending on the CONFIG_4KSTACKS option, the Guest can have one or
* two pages of stack space.
*/
for (i = 0; i < cpu->lg->stack_pages; i++)
/* The stack grows *upwards*, so the address we're given is the
/*
* The stack grows *upwards*, so the address we're given is the
* start of the page after the kernel stack. Subtract one to
* get back onto the first stack page, and keep subtracting to
* get to the rest of the stack pages. */
* get to the rest of the stack pages.
*/
pin_page(cpu, cpu->esp1 - 1 - i * PAGE_SIZE);
}
/* Direct traps also mean that we need to know whenever the Guest wants to use
/*
* Direct traps also mean that we need to know whenever the Guest wants to use
* a different kernel stack, so we can change the IDT entries to use that
* stack. The IDT entries expect a virtual address, so unlike most addresses
* the Guest gives us, the "esp" (stack pointer) value here is virtual, not
* physical.
*
* In Linux each process has its own kernel stack, so this happens a lot: we
* change stacks on each context switch. */
* change stacks on each context switch.
*/
void guest_set_stack(struct lg_cpu *cpu, u32 seg, u32 esp, unsigned int pages)
{
/* You are not allowed have a stack segment with privilege level 0: bad
* Guest! */
/*
* You're not allowed a stack segment with privilege level 0: bad Guest!
*/
if ((seg & 0x3) != GUEST_PL)
kill_guest(cpu, "bad stack segment %i", seg);
/* We only expect one or two stack pages. */
......@@ -387,11 +454,15 @@ void guest_set_stack(struct lg_cpu *cpu, u32 seg, u32 esp, unsigned int pages)
pin_stack_pages(cpu);
}
/* All this reference to mapping stacks leads us neatly into the other complex
* part of the Host: page table handling. */
/*
* All this reference to mapping stacks leads us neatly into the other complex
* part of the Host: page table handling.
*/
/*H:235 This is the routine which actually checks the Guest's IDT entry and
* transfers it into the entry in "struct lguest": */
/*H:235
* This is the routine which actually checks the Guest's IDT entry and
* transfers it into the entry in "struct lguest":
*/
static void set_trap(struct lg_cpu *cpu, struct desc_struct *trap,
unsigned int num, u32 lo, u32 hi)
{
......@@ -407,30 +478,38 @@ static void set_trap(struct lg_cpu *cpu, struct desc_struct *trap,
if (type != 0xE && type != 0xF)
kill_guest(cpu, "bad IDT type %i", type);
/* We only copy the handler address, present bit, privilege level and
/*
* We only copy the handler address, present bit, privilege level and
* type. The privilege level controls where the trap can be triggered
* manually with an "int" instruction. This is usually GUEST_PL,
* except for system calls which userspace can use. */
* except for system calls which userspace can use.
*/
trap->a = ((__KERNEL_CS|GUEST_PL)<<16) | (lo&0x0000FFFF);
trap->b = (hi&0xFFFFEF00);
}
/*H:230 While we're here, dealing with delivering traps and interrupts to the
/*H:230
* While we're here, dealing with delivering traps and interrupts to the
* Guest, we might as well complete the picture: how the Guest tells us where
* it wants them to go. This would be simple, except making traps fast
* requires some tricks.
*
* We saw the Guest setting Interrupt Descriptor Table (IDT) entries with the
* LHCALL_LOAD_IDT_ENTRY hypercall before: that comes here. */
* LHCALL_LOAD_IDT_ENTRY hypercall before: that comes here.
*/
void load_guest_idt_entry(struct lg_cpu *cpu, unsigned int num, u32 lo, u32 hi)
{
/* Guest never handles: NMI, doublefault, spurious interrupt or
* hypercall. We ignore when it tries to set them. */
/*
* Guest never handles: NMI, doublefault, spurious interrupt or
* hypercall. We ignore when it tries to set them.
*/
if (num == 2 || num == 8 || num == 15 || num == LGUEST_TRAP_ENTRY)
return;
/* Mark the IDT as changed: next time the Guest runs we'll know we have
* to copy this again. */
/*
* Mark the IDT as changed: next time the Guest runs we'll know we have
* to copy this again.
*/
cpu->changed |= CHANGED_IDT;
/* Check that the Guest doesn't try to step outside the bounds. */
......@@ -440,9 +519,11 @@ void load_guest_idt_entry(struct lg_cpu *cpu, unsigned int num, u32 lo, u32 hi)
set_trap(cpu, &cpu->arch.idt[num], num, lo, hi);
}
/* The default entry for each interrupt points into the Switcher routines which
/*
* The default entry for each interrupt points into the Switcher routines which
* simply return to the Host. The run_guest() loop will then call
* deliver_trap() to bounce it back into the Guest. */
* deliver_trap() to bounce it back into the Guest.
*/
static void default_idt_entry(struct desc_struct *idt,
int trap,
const unsigned long handler,
......@@ -451,13 +532,17 @@ static void default_idt_entry(struct desc_struct *idt,
/* A present interrupt gate. */
u32 flags = 0x8e00;
/* Set the privilege level on the entry for the hypercall: this allows
* the Guest to use the "int" instruction to trigger it. */
/*
* Set the privilege level on the entry for the hypercall: this allows
* the Guest to use the "int" instruction to trigger it.
*/
if (trap == LGUEST_TRAP_ENTRY)
flags |= (GUEST_PL << 13);
else if (base)
/* Copy priv. level from what Guest asked for. This allows
* debug (int 3) traps from Guest userspace, for example. */
/*
* Copy privilege level from what Guest asked for. This allows
* debug (int 3) traps from Guest userspace, for example.
*/
flags |= (base->b & 0x6000);
/* Now pack it into the IDT entry in its weird format. */
......@@ -475,16 +560,20 @@ void setup_default_idt_entries(struct lguest_ro_state *state,
default_idt_entry(&state->guest_idt[i], i, def[i], NULL);
}
/*H:240 We don't use the IDT entries in the "struct lguest" directly, instead
/*H:240
* We don't use the IDT entries in the "struct lguest" directly, instead
* we copy them into the IDT which we've set up for Guests on this CPU, just
* before we run the Guest. This routine does that copy. */
* before we run the Guest. This routine does that copy.
*/
void copy_traps(const struct lg_cpu *cpu, struct desc_struct *idt,
const unsigned long *def)
{
unsigned int i;
/* We can simply copy the direct traps, otherwise we use the default
* ones in the Switcher: they will return to the Host. */
/*
* We can simply copy the direct traps, otherwise we use the default
* ones in the Switcher: they will return to the Host.
*/
for (i = 0; i < ARRAY_SIZE(cpu->arch.idt); i++) {
const struct desc_struct *gidt = &cpu->arch.idt[i];
......@@ -492,14 +581,16 @@ void copy_traps(const struct lg_cpu *cpu, struct desc_struct *idt,
if (!direct_trap(i))
continue;
/* Only trap gates (type 15) can go direct to the Guest.
/*
* Only trap gates (type 15) can go direct to the Guest.
* Interrupt gates (type 14) disable interrupts as they are
* entered, which we never let the Guest do. Not present
* entries (type 0x0) also can't go direct, of course.
*
* If it can't go direct, we still need to copy the priv. level:
* they might want to give userspace access to a software
* interrupt. */
* interrupt.
*/
if (idt_type(gidt->a, gidt->b) == 0xF)
idt[i] = *gidt;
else
......@@ -518,7 +609,8 @@ void copy_traps(const struct lg_cpu *cpu, struct desc_struct *idt,
* the next timer interrupt (in nanoseconds). We use the high-resolution timer
* infrastructure to set a callback at that time.
*
* 0 means "turn off the clock". */
* 0 means "turn off the clock".
*/
void guest_set_clockevent(struct lg_cpu *cpu, unsigned long delta)
{
ktime_t expires;
......@@ -529,9 +621,11 @@ void guest_set_clockevent(struct lg_cpu *cpu, unsigned long delta)
return;
}
/* We use wallclock time here, so the Guest might not be running for
/*
* We use wallclock time here, so the Guest might not be running for
* all the time between now and the timer interrupt it asked for. This
* is almost always the right thing to do. */
* is almost always the right thing to do.
*/
expires = ktime_add_ns(ktime_get_real(), delta);
hrtimer_start(&cpu->hrt, expires, HRTIMER_MODE_ABS);
}
......
......@@ -16,15 +16,13 @@
void free_pagetables(void);
int init_pagetables(struct page **switcher_page, unsigned int pages);
struct pgdir
{
struct pgdir {
unsigned long gpgdir;
pgd_t *pgdir;
};
/* We have two pages shared with guests, per cpu. */
struct lguest_pages
{
struct lguest_pages {
/* This is the stack page mapped rw in guest */
char spare[PAGE_SIZE - sizeof(struct lguest_regs)];
struct lguest_regs regs;
......@@ -54,13 +52,13 @@ struct lg_cpu {
unsigned long pending_notify; /* pfn from LHCALL_NOTIFY */
/* At end of a page shared mapped over lguest_pages in guest. */
/* At end of a page shared mapped over lguest_pages in guest. */
unsigned long regs_page;
struct lguest_regs *regs;
struct lguest_pages *last_pages;
int cpu_pgd; /* which pgd this cpu is currently using */
int cpu_pgd; /* Which pgd this cpu is currently using */
/* If a hypercall was asked for, this points to the arguments. */
struct hcall_args *hcall;
......@@ -89,15 +87,17 @@ struct lg_eventfd_map {
};
/* The private info the thread maintains about the guest. */
struct lguest
{
struct lguest {
struct lguest_data __user *lguest_data;
struct lg_cpu cpus[NR_CPUS];
unsigned int nr_cpus;
u32 pfn_limit;
/* This provides the offset to the base of guest-physical
* memory in the Launcher. */
/*
* This provides the offset to the base of guest-physical memory in the
* Launcher.
*/
void __user *mem_base;
unsigned long kernel_address;
......@@ -122,11 +122,13 @@ bool lguest_address_ok(const struct lguest *lg,
void __lgread(struct lg_cpu *, void *, unsigned long, unsigned);
void __lgwrite(struct lg_cpu *, unsigned long, const void *, unsigned);
/*H:035 Using memory-copy operations like that is usually inconvient, so we
/*H:035
* Using memory-copy operations like that is usually inconvient, so we
* have the following helper macros which read and write a specific type (often
* an unsigned long).
*
* This reads into a variable of the given type then returns that. */
* This reads into a variable of the given type then returns that.
*/
#define lgread(cpu, addr, type) \
({ type _v; __lgread((cpu), &_v, (addr), sizeof(_v)); _v; })
......@@ -140,9 +142,11 @@ void __lgwrite(struct lg_cpu *, unsigned long, const void *, unsigned);
int run_guest(struct lg_cpu *cpu, unsigned long __user *user);
/* Helper macros to obtain the first 12 or the last 20 bits, this is only the
/*
* Helper macros to obtain the first 12 or the last 20 bits, this is only the
* first step in the migration to the kernel types. pte_pfn is already defined
* in the kernel. */
* in the kernel.
*/
#define pgd_flags(x) (pgd_val(x) & ~PAGE_MASK)
#define pgd_pfn(x) (pgd_val(x) >> PAGE_SHIFT)
#define pmd_flags(x) (pmd_val(x) & ~PAGE_MASK)
......
/*P:050 Lguest guests use a very simple method to describe devices. It's a
/*P:050
* Lguest guests use a very simple method to describe devices. It's a
* series of device descriptors contained just above the top of normal Guest
* memory.
*
* We use the standard "virtio" device infrastructure, which provides us with a
* console, a network and a block driver. Each one expects some configuration
* information and a "virtqueue" or two to send and receive data. :*/
* information and a "virtqueue" or two to send and receive data.
:*/
#include <linux/init.h>
#include <linux/bootmem.h>
#include <linux/lguest_launcher.h>
......@@ -20,8 +22,10 @@
/* The pointer to our (page) of device descriptions. */
static void *lguest_devices;
/* For Guests, device memory can be used as normal memory, so we cast away the
* __iomem to quieten sparse. */
/*
* For Guests, device memory can be used as normal memory, so we cast away the
* __iomem to quieten sparse.
*/
static inline void *lguest_map(unsigned long phys_addr, unsigned long pages)
{
return (__force void *)ioremap_cache(phys_addr, PAGE_SIZE*pages);
......@@ -32,8 +36,10 @@ static inline void lguest_unmap(void *addr)
iounmap((__force void __iomem *)addr);
}
/*D:100 Each lguest device is just a virtio device plus a pointer to its entry
* in the lguest_devices page. */
/*D:100
* Each lguest device is just a virtio device plus a pointer to its entry
* in the lguest_devices page.
*/
struct lguest_device {
struct virtio_device vdev;
......@@ -41,9 +47,11 @@ struct lguest_device {
struct lguest_device_desc *desc;
};
/* Since the virtio infrastructure hands us a pointer to the virtio_device all
/*
* Since the virtio infrastructure hands us a pointer to the virtio_device all
* the time, it helps to have a curt macro to get a pointer to the struct
* lguest_device it's enclosed in. */
* lguest_device it's enclosed in.
*/
#define to_lgdev(vd) container_of(vd, struct lguest_device, vdev)
/*D:130
......@@ -55,7 +63,8 @@ struct lguest_device {
* the driver will look at them during setup.
*
* A convenient routine to return the device's virtqueue config array:
* immediately after the descriptor. */
* immediately after the descriptor.
*/
static struct lguest_vqconfig *lg_vq(const struct lguest_device_desc *desc)
{
return (void *)(desc + 1);
......@@ -98,10 +107,12 @@ static u32 lg_get_features(struct virtio_device *vdev)
return features;
}
/* The virtio core takes the features the Host offers, and copies the
* ones supported by the driver into the vdev->features array. Once
* that's all sorted out, this routine is called so we can tell the
* Host which features we understand and accept. */
/*
* The virtio core takes the features the Host offers, and copies the ones
* supported by the driver into the vdev->features array. Once that's all
* sorted out, this routine is called so we can tell the Host which features we
* understand and accept.
*/
static void lg_finalize_features(struct virtio_device *vdev)
{
unsigned int i, bits;
......@@ -112,10 +123,11 @@ static void lg_finalize_features(struct virtio_device *vdev)
/* Give virtio_ring a chance to accept features. */
vring_transport_features(vdev);
/* The vdev->feature array is a Linux bitmask: this isn't the
* same as a the simple array of bits used by lguest devices
* for features. So we do this slow, manual conversion which is
* completely general. */
/*
* The vdev->feature array is a Linux bitmask: this isn't the same as a
* the simple array of bits used by lguest devices for features. So we
* do this slow, manual conversion which is completely general.
*/
memset(out_features, 0, desc->feature_len);
bits = min_t(unsigned, desc->feature_len, sizeof(vdev->features)) * 8;
for (i = 0; i < bits; i++) {
......@@ -146,15 +158,19 @@ static void lg_set(struct virtio_device *vdev, unsigned int offset,
memcpy(lg_config(desc) + offset, buf, len);
}
/* The operations to get and set the status word just access the status field
* of the device descriptor. */
/*
* The operations to get and set the status word just access the status field
* of the device descriptor.
*/
static u8 lg_get_status(struct virtio_device *vdev)
{
return to_lgdev(vdev)->desc->status;
}
/* To notify on status updates, we (ab)use the NOTIFY hypercall, with the
* descriptor address of the device. A zero status means "reset". */
/*
* To notify on status updates, we (ab)use the NOTIFY hypercall, with the
* descriptor address of the device. A zero status means "reset".
*/
static void set_status(struct virtio_device *vdev, u8 status)
{
unsigned long offset = (void *)to_lgdev(vdev)->desc - lguest_devices;
......@@ -191,8 +207,7 @@ static void lg_reset(struct virtio_device *vdev)
*/
/*D:140 This is the information we remember about each virtqueue. */
struct lguest_vq_info
{
struct lguest_vq_info {
/* A copy of the information contained in the device config. */
struct lguest_vqconfig config;
......@@ -200,13 +215,17 @@ struct lguest_vq_info
void *pages;
};
/* When the virtio_ring code wants to prod the Host, it calls us here and we
/*
* When the virtio_ring code wants to prod the Host, it calls us here and we
* make a hypercall. We hand the physical address of the virtqueue so the Host
* knows which virtqueue we're talking about. */
* knows which virtqueue we're talking about.
*/
static void lg_notify(struct virtqueue *vq)
{
/* We store our virtqueue information in the "priv" pointer of the
* virtqueue structure. */
/*
* We store our virtqueue information in the "priv" pointer of the
* virtqueue structure.
*/
struct lguest_vq_info *lvq = vq->priv;
kvm_hypercall1(LHCALL_NOTIFY, lvq->config.pfn << PAGE_SHIFT);
......@@ -215,7 +234,8 @@ static void lg_notify(struct virtqueue *vq)
/* An extern declaration inside a C file is bad form. Don't do it. */
extern void lguest_setup_irq(unsigned int irq);
/* This routine finds the first virtqueue described in the configuration of
/*
* This routine finds the Nth virtqueue described in the configuration of
* this device and sets it up.
*
* This is kind of an ugly duckling. It'd be nicer to have a standard
......@@ -223,9 +243,7 @@ extern void lguest_setup_irq(unsigned int irq);
* everyone wants to do it differently. The KVM coders want the Guest to
* allocate its own pages and tell the Host where they are, but for lguest it's
* simpler for the Host to simply tell us where the pages are.
*
* So we provide drivers with a "find the Nth virtqueue and set it up"
* function. */
*/
static struct virtqueue *lg_find_vq(struct virtio_device *vdev,
unsigned index,
void (*callback)(struct virtqueue *vq),
......@@ -244,9 +262,11 @@ static struct virtqueue *lg_find_vq(struct virtio_device *vdev,
if (!lvq)
return ERR_PTR(-ENOMEM);
/* Make a copy of the "struct lguest_vqconfig" entry, which sits after
/*
* Make a copy of the "struct lguest_vqconfig" entry, which sits after
* the descriptor. We need a copy because the config space might not
* be aligned correctly. */
* be aligned correctly.
*/
memcpy(&lvq->config, lg_vq(ldev->desc)+index, sizeof(lvq->config));
printk("Mapping virtqueue %i addr %lx\n", index,
......@@ -261,8 +281,10 @@ static struct virtqueue *lg_find_vq(struct virtio_device *vdev,
goto free_lvq;
}
/* OK, tell virtio_ring.c to set up a virtqueue now we know its size
* and we've got a pointer to its pages. */
/*
* OK, tell virtio_ring.c to set up a virtqueue now we know its size
* and we've got a pointer to its pages.
*/
vq = vring_new_virtqueue(lvq->config.num, LGUEST_VRING_ALIGN,
vdev, lvq->pages, lg_notify, callback, name);
if (!vq) {
......@@ -273,18 +295,23 @@ static struct virtqueue *lg_find_vq(struct virtio_device *vdev,
/* Make sure the interrupt is allocated. */
lguest_setup_irq(lvq->config.irq);
/* Tell the interrupt for this virtqueue to go to the virtio_ring
* interrupt handler. */
/* FIXME: We used to have a flag for the Host to tell us we could use
/*
* Tell the interrupt for this virtqueue to go to the virtio_ring
* interrupt handler.
*
* FIXME: We used to have a flag for the Host to tell us we could use
* the interrupt as a source of randomness: it'd be nice to have that
* back.. */
* back.
*/
err = request_irq(lvq->config.irq, vring_interrupt, IRQF_SHARED,
dev_name(&vdev->dev), vq);
if (err)
goto destroy_vring;
/* Last of all we hook up our 'struct lguest_vq_info" to the
* virtqueue's priv pointer. */
/*
* Last of all we hook up our 'struct lguest_vq_info" to the
* virtqueue's priv pointer.
*/
vq->priv = lvq;
return vq;
......@@ -358,11 +385,14 @@ static struct virtio_config_ops lguest_config_ops = {
.del_vqs = lg_del_vqs,
};
/* The root device for the lguest virtio devices. This makes them appear as
* /sys/devices/lguest/0,1,2 not /sys/devices/0,1,2. */
/*
* The root device for the lguest virtio devices. This makes them appear as
* /sys/devices/lguest/0,1,2 not /sys/devices/0,1,2.
*/
static struct device *lguest_root;
/*D:120 This is the core of the lguest bus: actually adding a new device.
/*D:120
* This is the core of the lguest bus: actually adding a new device.
* It's a separate function because it's neater that way, and because an
* earlier version of the code supported hotplug and unplug. They were removed
* early on because they were never used.
......@@ -371,14 +401,14 @@ static struct device *lguest_root;
*
* It's worth reading this carefully: we start with a pointer to the new device
* descriptor in the "lguest_devices" page, and the offset into the device
* descriptor page so we can uniquely identify it if things go badly wrong. */
* descriptor page so we can uniquely identify it if things go badly wrong.
*/
static void add_lguest_device(struct lguest_device_desc *d,
unsigned int offset)
{
struct lguest_device *ldev;
/* Start with zeroed memory; Linux's device layer seems to count on
* it. */
/* Start with zeroed memory; Linux's device layer counts on it. */
ldev = kzalloc(sizeof(*ldev), GFP_KERNEL);
if (!ldev) {
printk(KERN_EMERG "Cannot allocate lguest dev %u type %u\n",
......@@ -388,17 +418,25 @@ static void add_lguest_device(struct lguest_device_desc *d,
/* This devices' parent is the lguest/ dir. */
ldev->vdev.dev.parent = lguest_root;
/* We have a unique device index thanks to the dev_index counter. */
/*
* The device type comes straight from the descriptor. There's also a
* device vendor field in the virtio_device struct, which we leave as
* 0.
*/
ldev->vdev.id.device = d->type;
/* We have a simple set of routines for querying the device's
* configuration information and setting its status. */
/*
* We have a simple set of routines for querying the device's
* configuration information and setting its status.
*/
ldev->vdev.config = &lguest_config_ops;
/* And we remember the device's descriptor for lguest_config_ops. */
ldev->desc = d;
/* register_virtio_device() sets up the generic fields for the struct
/*
* register_virtio_device() sets up the generic fields for the struct
* virtio_device and calls device_register(). This makes the bus
* infrastructure look for a matching driver. */
* infrastructure look for a matching driver.
*/
if (register_virtio_device(&ldev->vdev) != 0) {
printk(KERN_ERR "Failed to register lguest dev %u type %u\n",
offset, d->type);
......@@ -406,8 +444,10 @@ static void add_lguest_device(struct lguest_device_desc *d,
}
}
/*D:110 scan_devices() simply iterates through the device page. The type 0 is
* reserved to mean "end of devices". */
/*D:110
* scan_devices() simply iterates through the device page. The type 0 is
* reserved to mean "end of devices".
*/
static void scan_devices(void)
{
unsigned int i;
......@@ -426,7 +466,8 @@ static void scan_devices(void)
}
}
/*D:105 Fairly early in boot, lguest_devices_init() is called to set up the
/*D:105
* Fairly early in boot, lguest_devices_init() is called to set up the
* lguest device infrastructure. We check that we are a Guest by checking
* pv_info.name: there are other ways of checking, but this seems most
* obvious to me.
......@@ -437,7 +478,8 @@ static void scan_devices(void)
* correct sysfs incantation).
*
* Finally we call scan_devices() which adds all the devices found in the
* lguest_devices page. */
* lguest_devices page.
*/
static int __init lguest_devices_init(void)
{
if (strcmp(pv_info.name, "lguest") != 0)
......@@ -456,11 +498,13 @@ static int __init lguest_devices_init(void)
/* We do this after core stuff, but before the drivers. */
postcore_initcall(lguest_devices_init);
/*D:150 At this point in the journey we used to now wade through the lguest
/*D:150
* At this point in the journey we used to now wade through the lguest
* devices themselves: net, block and console. Since they're all now virtio
* devices rather than lguest-specific, I've decided to ignore them. Mostly,
* they're kind of boring. But this does mean you'll never experience the
* thrill of reading the forbidden love scene buried deep in the block driver.
*
* "make Launcher" beckons, where we answer questions like "Where do Guests
* come from?", and "What do you do when someone asks for optimization?". */
* come from?", and "What do you do when someone asks for optimization?".
*/
/*P:200 This contains all the /dev/lguest code, whereby the userspace launcher
* controls and communicates with the Guest. For example, the first write will
* tell us the Guest's memory layout, pagetable, entry point and kernel address
* offset. A read will run the Guest until something happens, such as a signal
* or the Guest doing a NOTIFY out to the Launcher. :*/
* tell us the Guest's memory layout and entry point. A read will run the
* Guest until something happens, such as a signal or the Guest doing a NOTIFY
* out to the Launcher.
:*/
#include <linux/uaccess.h>
#include <linux/miscdevice.h>
#include <linux/fs.h>
......@@ -11,14 +12,41 @@
#include <linux/file.h>
#include "lg.h"
/*L:056
* Before we move on, let's jump ahead and look at what the kernel does when
* it needs to look up the eventfds. That will complete our picture of how we
* use RCU.
*
* The notification value is in cpu->pending_notify: we return true if it went
* to an eventfd.
*/
bool send_notify_to_eventfd(struct lg_cpu *cpu)
{
unsigned int i;
struct lg_eventfd_map *map;
/* lg->eventfds is RCU-protected */
/*
* This "rcu_read_lock()" helps track when someone is still looking at
* the (RCU-using) eventfds array. It's not actually a lock at all;
* indeed it's a noop in many configurations. (You didn't expect me to
* explain all the RCU secrets here, did you?)
*/
rcu_read_lock();
/*
* rcu_dereference is the counter-side of rcu_assign_pointer(); it
* makes sure we don't access the memory pointed to by
* cpu->lg->eventfds before cpu->lg->eventfds is set. Sounds crazy,
* but Alpha allows this! Paul McKenney points out that a really
* aggressive compiler could have the same effect:
* http://lists.ozlabs.org/pipermail/lguest/2009-July/001560.html
*
* So play safe, use rcu_dereference to get the rcu-protected pointer:
*/
map = rcu_dereference(cpu->lg->eventfds);
/*
* Simple array search: even if they add an eventfd while we do this,
* we'll continue to use the old array and just won't see the new one.
*/
for (i = 0; i < map->num; i++) {
if (map->map[i].addr == cpu->pending_notify) {
eventfd_signal(map->map[i].event, 1);
......@@ -26,19 +54,50 @@ bool send_notify_to_eventfd(struct lg_cpu *cpu)
break;
}
}
/* We're done with the rcu-protected variable cpu->lg->eventfds. */
rcu_read_unlock();
/* If we cleared the notification, it's because we found a match. */
return cpu->pending_notify == 0;
}
/*L:055
* One of the more tricksy tricks in the Linux Kernel is a technique called
* Read Copy Update. Since one point of lguest is to teach lguest journeyers
* about kernel coding, I use it here. (In case you're curious, other purposes
* include learning about virtualization and instilling a deep appreciation for
* simplicity and puppies).
*
* We keep a simple array which maps LHCALL_NOTIFY values to eventfds, but we
* add new eventfds without ever blocking readers from accessing the array.
* The current Launcher only does this during boot, so that never happens. But
* Read Copy Update is cool, and adding a lock risks damaging even more puppies
* than this code does.
*
* We allocate a brand new one-larger array, copy the old one and add our new
* element. Then we make the lg eventfd pointer point to the new array.
* That's the easy part: now we need to free the old one, but we need to make
* sure no slow CPU somewhere is still looking at it. That's what
* synchronize_rcu does for us: waits until every CPU has indicated that it has
* moved on to know it's no longer using the old one.
*
* If that's unclear, see http://en.wikipedia.org/wiki/Read-copy-update.
*/
static int add_eventfd(struct lguest *lg, unsigned long addr, int fd)
{
struct lg_eventfd_map *new, *old = lg->eventfds;
/*
* We don't allow notifications on value 0 anyway (pending_notify of
* 0 means "nothing pending").
*/
if (!addr)
return -EINVAL;
/* Replace the old array with the new one, carefully: others can
* be accessing it at the same time */
/*
* Replace the old array with the new one, carefully: others can
* be accessing it at the same time.
*/
new = kmalloc(sizeof(*new) + sizeof(new->map[0]) * (old->num + 1),
GFP_KERNEL);
if (!new)
......@@ -52,22 +111,41 @@ static int add_eventfd(struct lguest *lg, unsigned long addr, int fd)
new->map[new->num].addr = addr;
new->map[new->num].event = eventfd_ctx_fdget(fd);
if (IS_ERR(new->map[new->num].event)) {
int err = PTR_ERR(new->map[new->num].event);
kfree(new);
return PTR_ERR(new->map[new->num].event);
return err;
}
new->num++;
/* Now put new one in place. */
/*
* Now put new one in place: rcu_assign_pointer() is a fancy way of
* doing "lg->eventfds = new", but it uses memory barriers to make
* absolutely sure that the contents of "new" written above is nailed
* down before we actually do the assignment.
*
* We have to think about these kinds of things when we're operating on
* live data without locks.
*/
rcu_assign_pointer(lg->eventfds, new);
/* We're not in a big hurry. Wait until noone's looking at old
* version, then delete it. */
/*
* We're not in a big hurry. Wait until noone's looking at old
* version, then free it.
*/
synchronize_rcu();
kfree(old);
return 0;
}
/*L:052
* Receiving notifications from the Guest is usually done by attaching a
* particular LHCALL_NOTIFY value to an event filedescriptor. The eventfd will
* become readable when the Guest does an LHCALL_NOTIFY with that value.
*
* This is really convenient for processing each virtqueue in a separate
* thread.
*/
static int attach_eventfd(struct lguest *lg, const unsigned long __user *input)
{
unsigned long addr, fd;
......@@ -79,15 +157,22 @@ static int attach_eventfd(struct lguest *lg, const unsigned long __user *input)
if (get_user(fd, input) != 0)
return -EFAULT;
/*
* Just make sure two callers don't add eventfds at once. We really
* only need to lock against callers adding to the same Guest, so using
* the Big Lguest Lock is overkill. But this is setup, not a fast path.
*/
mutex_lock(&lguest_lock);
err = add_eventfd(lg, addr, fd);
mutex_unlock(&lguest_lock);
return 0;
return err;
}
/*L:050 Sending an interrupt is done by writing LHREQ_IRQ and an interrupt
* number to /dev/lguest. */
/*L:050
* Sending an interrupt is done by writing LHREQ_IRQ and an interrupt
* number to /dev/lguest.
*/
static int user_send_irq(struct lg_cpu *cpu, const unsigned long __user *input)
{
unsigned long irq;
......@@ -97,12 +182,18 @@ static int user_send_irq(struct lg_cpu *cpu, const unsigned long __user *input)
if (irq >= LGUEST_IRQS)
return -EINVAL;
/*
* Next time the Guest runs, the core code will see if it can deliver
* this interrupt.
*/
set_interrupt(cpu, irq);
return 0;
}
/*L:040 Once our Guest is initialized, the Launcher makes it run by reading
* from /dev/lguest. */
/*L:040
* Once our Guest is initialized, the Launcher makes it run by reading
* from /dev/lguest.
*/
static ssize_t read(struct file *file, char __user *user, size_t size,loff_t*o)
{
struct lguest *lg = file->private_data;
......@@ -138,8 +229,10 @@ static ssize_t read(struct file *file, char __user *user, size_t size,loff_t*o)
return len;
}
/* If we returned from read() last time because the Guest sent I/O,
* clear the flag. */
/*
* If we returned from read() last time because the Guest sent I/O,
* clear the flag.
*/
if (cpu->pending_notify)
cpu->pending_notify = 0;
......@@ -147,8 +240,10 @@ static ssize_t read(struct file *file, char __user *user, size_t size,loff_t*o)
return run_guest(cpu, (unsigned long __user *)user);
}
/*L:025 This actually initializes a CPU. For the moment, a Guest is only
* uniprocessor, so "id" is always 0. */
/*L:025
* This actually initializes a CPU. For the moment, a Guest is only
* uniprocessor, so "id" is always 0.
*/
static int lg_cpu_start(struct lg_cpu *cpu, unsigned id, unsigned long start_ip)
{
/* We have a limited number the number of CPUs in the lguest struct. */
......@@ -163,8 +258,10 @@ static int lg_cpu_start(struct lg_cpu *cpu, unsigned id, unsigned long start_ip)
/* Each CPU has a timer it can set. */
init_clockdev(cpu);
/* We need a complete page for the Guest registers: they are accessible
* to the Guest and we can only grant it access to whole pages. */
/*
* We need a complete page for the Guest registers: they are accessible
* to the Guest and we can only grant it access to whole pages.
*/
cpu->regs_page = get_zeroed_page(GFP_KERNEL);
if (!cpu->regs_page)
return -ENOMEM;
......@@ -172,29 +269,38 @@ static int lg_cpu_start(struct lg_cpu *cpu, unsigned id, unsigned long start_ip)
/* We actually put the registers at the bottom of the page. */
cpu->regs = (void *)cpu->regs_page + PAGE_SIZE - sizeof(*cpu->regs);
/* Now we initialize the Guest's registers, handing it the start
* address. */
/*
* Now we initialize the Guest's registers, handing it the start
* address.
*/
lguest_arch_setup_regs(cpu, start_ip);
/* We keep a pointer to the Launcher task (ie. current task) for when
* other Guests want to wake this one (eg. console input). */
/*
* We keep a pointer to the Launcher task (ie. current task) for when
* other Guests want to wake this one (eg. console input).
*/
cpu->tsk = current;
/* We need to keep a pointer to the Launcher's memory map, because if
/*
* We need to keep a pointer to the Launcher's memory map, because if
* the Launcher dies we need to clean it up. If we don't keep a
* reference, it is destroyed before close() is called. */
* reference, it is destroyed before close() is called.
*/
cpu->mm = get_task_mm(cpu->tsk);
/* We remember which CPU's pages this Guest used last, for optimization
* when the same Guest runs on the same CPU twice. */
/*
* We remember which CPU's pages this Guest used last, for optimization
* when the same Guest runs on the same CPU twice.
*/
cpu->last_pages = NULL;
/* No error == success. */
return 0;
}
/*L:020 The initialization write supplies 3 pointer sized (32 or 64 bit)
* values (in addition to the LHREQ_INITIALIZE value). These are:
/*L:020
* The initialization write supplies 3 pointer sized (32 or 64 bit) values (in
* addition to the LHREQ_INITIALIZE value). These are:
*
* base: The start of the Guest-physical memory inside the Launcher memory.
*
......@@ -206,14 +312,15 @@ static int lg_cpu_start(struct lg_cpu *cpu, unsigned id, unsigned long start_ip)
*/
static int initialize(struct file *file, const unsigned long __user *input)
{
/* "struct lguest" contains everything we (the Host) know about a
* Guest. */
/* "struct lguest" contains all we (the Host) know about a Guest. */
struct lguest *lg;
int err;
unsigned long args[3];
/* We grab the Big Lguest lock, which protects against multiple
* simultaneous initializations. */
/*
* We grab the Big Lguest lock, which protects against multiple
* simultaneous initializations.
*/
mutex_lock(&lguest_lock);
/* You can't initialize twice! Close the device and start again... */
if (file->private_data) {
......@@ -248,8 +355,10 @@ static int initialize(struct file *file, const unsigned long __user *input)
if (err)
goto free_eventfds;
/* Initialize the Guest's shadow page tables, using the toplevel
* address the Launcher gave us. This allocates memory, so can fail. */
/*
* Initialize the Guest's shadow page tables, using the toplevel
* address the Launcher gave us. This allocates memory, so can fail.
*/
err = init_guest_pagetable(lg);
if (err)
goto free_regs;
......@@ -274,20 +383,24 @@ unlock:
return err;
}
/*L:010 The first operation the Launcher does must be a write. All writes
/*L:010
* The first operation the Launcher does must be a write. All writes
* start with an unsigned long number: for the first write this must be
* LHREQ_INITIALIZE to set up the Guest. After that the Launcher can use
* writes of other values to send interrupts.
* writes of other values to send interrupts or set up receipt of notifications.
*
* Note that we overload the "offset" in the /dev/lguest file to indicate what
* CPU number we're dealing with. Currently this is always 0, since we only
* CPU number we're dealing with. Currently this is always 0 since we only
* support uniprocessor Guests, but you can see the beginnings of SMP support
* here. */
* here.
*/
static ssize_t write(struct file *file, const char __user *in,
size_t size, loff_t *off)
{
/* Once the Guest is initialized, we hold the "struct lguest" in the
* file private data. */
/*
* Once the Guest is initialized, we hold the "struct lguest" in the
* file private data.
*/
struct lguest *lg = file->private_data;
const unsigned long __user *input = (const unsigned long __user *)in;
unsigned long req;
......@@ -322,13 +435,15 @@ static ssize_t write(struct file *file, const char __user *in,
}
}
/*L:060 The final piece of interface code is the close() routine. It reverses
/*L:060
* The final piece of interface code is the close() routine. It reverses
* everything done in initialize(). This is usually called because the
* Launcher exited.
*
* Note that the close routine returns 0 or a negative error number: it can't
* really fail, but it can whine. I blame Sun for this wart, and K&R C for
* letting them do it. :*/
* letting them do it.
:*/
static int close(struct inode *inode, struct file *file)
{
struct lguest *lg = file->private_data;
......@@ -338,8 +453,10 @@ static int close(struct inode *inode, struct file *file)
if (!lg)
return 0;
/* We need the big lock, to protect from inter-guest I/O and other
* Launchers initializing guests. */
/*
* We need the big lock, to protect from inter-guest I/O and other
* Launchers initializing guests.
*/
mutex_lock(&lguest_lock);
/* Free up the shadow page tables for the Guest. */
......@@ -350,8 +467,10 @@ static int close(struct inode *inode, struct file *file)
hrtimer_cancel(&lg->cpus[i].hrt);
/* We can free up the register page we allocated. */
free_page(lg->cpus[i].regs_page);
/* Now all the memory cleanups are done, it's safe to release
* the Launcher's memory management structure. */
/*
* Now all the memory cleanups are done, it's safe to release
* the Launcher's memory management structure.
*/
mmput(lg->cpus[i].mm);
}
......@@ -360,8 +479,10 @@ static int close(struct inode *inode, struct file *file)
eventfd_ctx_put(lg->eventfds->map[i].event);
kfree(lg->eventfds);
/* If lg->dead doesn't contain an error code it will be NULL or a
* kmalloc()ed string, either of which is ok to hand to kfree(). */
/*
* If lg->dead doesn't contain an error code it will be NULL or a
* kmalloc()ed string, either of which is ok to hand to kfree().
*/
if (!IS_ERR(lg->dead))
kfree(lg->dead);
/* Free the memory allocated to the lguest_struct */
......@@ -385,7 +506,8 @@ static int close(struct inode *inode, struct file *file)
*
* We begin our understanding with the Host kernel interface which the Launcher
* uses: reading and writing a character device called /dev/lguest. All the
* work happens in the read(), write() and close() routines: */
* work happens in the read(), write() and close() routines:
*/
static struct file_operations lguest_fops = {
.owner = THIS_MODULE,
.release = close,
......@@ -393,8 +515,10 @@ static struct file_operations lguest_fops = {
.read = read,
};
/* This is a textbook example of a "misc" character device. Populate a "struct
* miscdevice" and register it with misc_register(). */
/*
* This is a textbook example of a "misc" character device. Populate a "struct
* miscdevice" and register it with misc_register().
*/
static struct miscdevice lguest_dev = {
.minor = MISC_DYNAMIC_MINOR,
.name = "lguest",
......
/*P:700 The pagetable code, on the other hand, still shows the scars of
/*P:700
* The pagetable code, on the other hand, still shows the scars of
* previous encounters. It's functional, and as neat as it can be in the
* circumstances, but be wary, for these things are subtle and break easily.
* The Guest provides a virtual to physical mapping, but we can neither trust
* it nor use it: we verify and convert it here then point the CPU to the
* converted Guest pages when running the Guest. :*/
* converted Guest pages when running the Guest.
:*/
/* Copyright (C) Rusty Russell IBM Corporation 2006.
* GPL v2 and any later version */
......@@ -17,18 +19,20 @@
#include <asm/bootparam.h>
#include "lg.h"
/*M:008 We hold reference to pages, which prevents them from being swapped.
/*M:008
* We hold reference to pages, which prevents them from being swapped.
* It'd be nice to have a callback in the "struct mm_struct" when Linux wants
* to swap out. If we had this, and a shrinker callback to trim PTE pages, we
* could probably consider launching Guests as non-root. :*/
* could probably consider launching Guests as non-root.
:*/
/*H:300
* The Page Table Code
*
* We use two-level page tables for the Guest. If you're not entirely
* comfortable with virtual addresses, physical addresses and page tables then
* I recommend you review arch/x86/lguest/boot.c's "Page Table Handling" (with
* diagrams!).
* We use two-level page tables for the Guest, or three-level with PAE. If
* you're not entirely comfortable with virtual addresses, physical addresses
* and page tables then I recommend you review arch/x86/lguest/boot.c's "Page
* Table Handling" (with diagrams!).
*
* The Guest keeps page tables, but we maintain the actual ones here: these are
* called "shadow" page tables. Which is a very Guest-centric name: these are
......@@ -45,16 +49,18 @@
* (v) Flushing (throwing away) page tables,
* (vi) Mapping the Switcher when the Guest is about to run,
* (vii) Setting up the page tables initially.
:*/
:*/
/* 1024 entries in a page table page maps 1024 pages: 4MB. The Switcher is
* conveniently placed at the top 4MB, so it uses a separate, complete PTE
* page. */
/*
* The Switcher uses the complete top PTE page. That's 1024 PTE entries (4MB)
* or 512 PTE entries with PAE (2MB).
*/
#define SWITCHER_PGD_INDEX (PTRS_PER_PGD - 1)
/* For PAE we need the PMD index as well. We use the last 2MB, so we
* will need the last pmd entry of the last pmd page. */
/*
* For PAE we need the PMD index as well. We use the last 2MB, so we
* will need the last pmd entry of the last pmd page.
*/
#ifdef CONFIG_X86_PAE
#define SWITCHER_PMD_INDEX (PTRS_PER_PMD - 1)
#define RESERVE_MEM 2U
......@@ -64,14 +70,18 @@
#define CHECK_GPGD_MASK _PAGE_TABLE
#endif
/* We actually need a separate PTE page for each CPU. Remember that after the
/*
* We actually need a separate PTE page for each CPU. Remember that after the
* Switcher code itself comes two pages for each CPU, and we don't want this
* CPU's guest to see the pages of any other CPU. */
* CPU's guest to see the pages of any other CPU.
*/
static DEFINE_PER_CPU(pte_t *, switcher_pte_pages);
#define switcher_pte_page(cpu) per_cpu(switcher_pte_pages, cpu)
/*H:320 The page table code is curly enough to need helper functions to keep it
* clear and clean.
/*H:320
* The page table code is curly enough to need helper functions to keep it
* clear and clean. The kernel itself provides many of them; one advantage
* of insisting that the Guest and Host use the same CONFIG_PAE setting.
*
* There are two functions which return pointers to the shadow (aka "real")
* page tables.
......@@ -79,7 +89,8 @@ static DEFINE_PER_CPU(pte_t *, switcher_pte_pages);
* spgd_addr() takes the virtual address and returns a pointer to the top-level
* page directory entry (PGD) for that address. Since we keep track of several
* page tables, the "i" argument tells us which one we're interested in (it's
* usually the current one). */
* usually the current one).
*/
static pgd_t *spgd_addr(struct lg_cpu *cpu, u32 i, unsigned long vaddr)
{
unsigned int index = pgd_index(vaddr);
......@@ -96,9 +107,11 @@ static pgd_t *spgd_addr(struct lg_cpu *cpu, u32 i, unsigned long vaddr)
}
#ifdef CONFIG_X86_PAE
/* This routine then takes the PGD entry given above, which contains the
/*
* This routine then takes the PGD entry given above, which contains the
* address of the PMD page. It then returns a pointer to the PMD entry for the
* given address. */
* given address.
*/
static pmd_t *spmd_addr(struct lg_cpu *cpu, pgd_t spgd, unsigned long vaddr)
{
unsigned int index = pmd_index(vaddr);
......@@ -119,9 +132,11 @@ static pmd_t *spmd_addr(struct lg_cpu *cpu, pgd_t spgd, unsigned long vaddr)
}
#endif
/* This routine then takes the page directory entry returned above, which
/*
* This routine then takes the page directory entry returned above, which
* contains the address of the page table entry (PTE) page. It then returns a
* pointer to the PTE entry for the given address. */
* pointer to the PTE entry for the given address.
*/
static pte_t *spte_addr(struct lg_cpu *cpu, pgd_t spgd, unsigned long vaddr)
{
#ifdef CONFIG_X86_PAE
......@@ -139,8 +154,10 @@ static pte_t *spte_addr(struct lg_cpu *cpu, pgd_t spgd, unsigned long vaddr)
return &page[pte_index(vaddr)];
}
/* These two functions just like the above two, except they access the Guest
* page tables. Hence they return a Guest address. */
/*
* These functions are just like the above two, except they access the Guest
* page tables. Hence they return a Guest address.
*/
static unsigned long gpgd_addr(struct lg_cpu *cpu, unsigned long vaddr)
{
unsigned int index = vaddr >> (PGDIR_SHIFT);
......@@ -148,6 +165,7 @@ static unsigned long gpgd_addr(struct lg_cpu *cpu, unsigned long vaddr)
}
#ifdef CONFIG_X86_PAE
/* Follow the PGD to the PMD. */
static unsigned long gpmd_addr(pgd_t gpgd, unsigned long vaddr)
{
unsigned long gpage = pgd_pfn(gpgd) << PAGE_SHIFT;
......@@ -155,6 +173,7 @@ static unsigned long gpmd_addr(pgd_t gpgd, unsigned long vaddr)
return gpage + pmd_index(vaddr) * sizeof(pmd_t);
}
/* Follow the PMD to the PTE. */
static unsigned long gpte_addr(struct lg_cpu *cpu,
pmd_t gpmd, unsigned long vaddr)
{
......@@ -164,6 +183,7 @@ static unsigned long gpte_addr(struct lg_cpu *cpu,
return gpage + pte_index(vaddr) * sizeof(pte_t);
}
#else
/* Follow the PGD to the PTE (no mid-level for !PAE). */
static unsigned long gpte_addr(struct lg_cpu *cpu,
pgd_t gpgd, unsigned long vaddr)
{
......@@ -175,17 +195,21 @@ static unsigned long gpte_addr(struct lg_cpu *cpu,
#endif
/*:*/
/*M:014 get_pfn is slow: we could probably try to grab batches of pages here as
* an optimization (ie. pre-faulting). :*/
/*M:014
* get_pfn is slow: we could probably try to grab batches of pages here as
* an optimization (ie. pre-faulting).
:*/
/*H:350 This routine takes a page number given by the Guest and converts it to
/*H:350
* This routine takes a page number given by the Guest and converts it to
* an actual, physical page number. It can fail for several reasons: the
* virtual address might not be mapped by the Launcher, the write flag is set
* and the page is read-only, or the write flag was set and the page was
* shared so had to be copied, but we ran out of memory.
*
* This holds a reference to the page, so release_pte() is careful to put that
* back. */
* back.
*/
static unsigned long get_pfn(unsigned long virtpfn, int write)
{
struct page *page;
......@@ -198,33 +222,41 @@ static unsigned long get_pfn(unsigned long virtpfn, int write)
return -1UL;
}
/*H:340 Converting a Guest page table entry to a shadow (ie. real) page table
/*H:340
* Converting a Guest page table entry to a shadow (ie. real) page table
* entry can be a little tricky. The flags are (almost) the same, but the
* Guest PTE contains a virtual page number: the CPU needs the real page
* number. */
* number.
*/
static pte_t gpte_to_spte(struct lg_cpu *cpu, pte_t gpte, int write)
{
unsigned long pfn, base, flags;
/* The Guest sets the global flag, because it thinks that it is using
/*
* The Guest sets the global flag, because it thinks that it is using
* PGE. We only told it to use PGE so it would tell us whether it was
* flushing a kernel mapping or a userspace mapping. We don't actually
* use the global bit, so throw it away. */
* use the global bit, so throw it away.
*/
flags = (pte_flags(gpte) & ~_PAGE_GLOBAL);
/* The Guest's pages are offset inside the Launcher. */
base = (unsigned long)cpu->lg->mem_base / PAGE_SIZE;
/* We need a temporary "unsigned long" variable to hold the answer from
/*
* We need a temporary "unsigned long" variable to hold the answer from
* get_pfn(), because it returns 0xFFFFFFFF on failure, which wouldn't
* fit in spte.pfn. get_pfn() finds the real physical number of the
* page, given the virtual number. */
* page, given the virtual number.
*/
pfn = get_pfn(base + pte_pfn(gpte), write);
if (pfn == -1UL) {
kill_guest(cpu, "failed to get page %lu", pte_pfn(gpte));
/* When we destroy the Guest, we'll go through the shadow page
/*
* When we destroy the Guest, we'll go through the shadow page
* tables and release_pte() them. Make sure we don't think
* this one is valid! */
* this one is valid!
*/
flags = 0;
}
/* Now we assemble our shadow PTE from the page number and flags. */
......@@ -234,8 +266,10 @@ static pte_t gpte_to_spte(struct lg_cpu *cpu, pte_t gpte, int write)
/*H:460 And to complete the chain, release_pte() looks like this: */
static void release_pte(pte_t pte)
{
/* Remember that get_user_pages_fast() took a reference to the page, in
* get_pfn()? We have to put it back now. */
/*
* Remember that get_user_pages_fast() took a reference to the page, in
* get_pfn()? We have to put it back now.
*/
if (pte_flags(pte) & _PAGE_PRESENT)
put_page(pte_page(pte));
}
......@@ -273,7 +307,8 @@ static void check_gpmd(struct lg_cpu *cpu, pmd_t gpmd)
* and return to the Guest without it knowing.
*
* If we fixed up the fault (ie. we mapped the address), this routine returns
* true. Otherwise, it was a real fault and we need to tell the Guest. */
* true. Otherwise, it was a real fault and we need to tell the Guest.
*/
bool demand_page(struct lg_cpu *cpu, unsigned long vaddr, int errcode)
{
pgd_t gpgd;
......@@ -282,6 +317,7 @@ bool demand_page(struct lg_cpu *cpu, unsigned long vaddr, int errcode)
pte_t gpte;
pte_t *spte;
/* Mid level for PAE. */
#ifdef CONFIG_X86_PAE
pmd_t *spmd;
pmd_t gpmd;
......@@ -298,22 +334,26 @@ bool demand_page(struct lg_cpu *cpu, unsigned long vaddr, int errcode)
if (!(pgd_flags(*spgd) & _PAGE_PRESENT)) {
/* No shadow entry: allocate a new shadow PTE page. */
unsigned long ptepage = get_zeroed_page(GFP_KERNEL);
/* This is not really the Guest's fault, but killing it is
* simple for this corner case. */
/*
* This is not really the Guest's fault, but killing it is
* simple for this corner case.
*/
if (!ptepage) {
kill_guest(cpu, "out of memory allocating pte page");
return false;
}
/* We check that the Guest pgd is OK. */
check_gpgd(cpu, gpgd);
/* And we copy the flags to the shadow PGD entry. The page
* number in the shadow PGD is the page we just allocated. */
/*
* And we copy the flags to the shadow PGD entry. The page
* number in the shadow PGD is the page we just allocated.
*/
set_pgd(spgd, __pgd(__pa(ptepage) | pgd_flags(gpgd)));
}
#ifdef CONFIG_X86_PAE
gpmd = lgread(cpu, gpmd_addr(gpgd, vaddr), pmd_t);
/* middle level not present? We can't map it in. */
/* Middle level not present? We can't map it in. */
if (!(pmd_flags(gpmd) & _PAGE_PRESENT))
return false;
......@@ -324,8 +364,10 @@ bool demand_page(struct lg_cpu *cpu, unsigned long vaddr, int errcode)
/* No shadow entry: allocate a new shadow PTE page. */
unsigned long ptepage = get_zeroed_page(GFP_KERNEL);
/* This is not really the Guest's fault, but killing it is
* simple for this corner case. */
/*
* This is not really the Guest's fault, but killing it is
* simple for this corner case.
*/
if (!ptepage) {
kill_guest(cpu, "out of memory allocating pte page");
return false;
......@@ -334,27 +376,37 @@ bool demand_page(struct lg_cpu *cpu, unsigned long vaddr, int errcode)
/* We check that the Guest pmd is OK. */
check_gpmd(cpu, gpmd);
/* And we copy the flags to the shadow PMD entry. The page
* number in the shadow PMD is the page we just allocated. */
/*
* And we copy the flags to the shadow PMD entry. The page
* number in the shadow PMD is the page we just allocated.
*/
native_set_pmd(spmd, __pmd(__pa(ptepage) | pmd_flags(gpmd)));
}
/* OK, now we look at the lower level in the Guest page table: keep its
* address, because we might update it later. */
/*
* OK, now we look at the lower level in the Guest page table: keep its
* address, because we might update it later.
*/
gpte_ptr = gpte_addr(cpu, gpmd, vaddr);
#else
/* OK, now we look at the lower level in the Guest page table: keep its
* address, because we might update it later. */
/*
* OK, now we look at the lower level in the Guest page table: keep its
* address, because we might update it later.
*/
gpte_ptr = gpte_addr(cpu, gpgd, vaddr);
#endif
/* Read the actual PTE value. */
gpte = lgread(cpu, gpte_ptr, pte_t);
/* If this page isn't in the Guest page tables, we can't page it in. */
if (!(pte_flags(gpte) & _PAGE_PRESENT))
return false;
/* Check they're not trying to write to a page the Guest wants
* read-only (bit 2 of errcode == write). */
/*
* Check they're not trying to write to a page the Guest wants
* read-only (bit 2 of errcode == write).
*/
if ((errcode & 2) && !(pte_flags(gpte) & _PAGE_RW))
return false;
......@@ -362,8 +414,10 @@ bool demand_page(struct lg_cpu *cpu, unsigned long vaddr, int errcode)
if ((errcode & 4) && !(pte_flags(gpte) & _PAGE_USER))
return false;
/* Check that the Guest PTE flags are OK, and the page number is below
* the pfn_limit (ie. not mapping the Launcher binary). */
/*
* Check that the Guest PTE flags are OK, and the page number is below
* the pfn_limit (ie. not mapping the Launcher binary).
*/
check_gpte(cpu, gpte);
/* Add the _PAGE_ACCESSED and (for a write) _PAGE_DIRTY flag */
......@@ -373,29 +427,40 @@ bool demand_page(struct lg_cpu *cpu, unsigned long vaddr, int errcode)
/* Get the pointer to the shadow PTE entry we're going to set. */
spte = spte_addr(cpu, *spgd, vaddr);
/* If there was a valid shadow PTE entry here before, we release it.
* This can happen with a write to a previously read-only entry. */
/*
* If there was a valid shadow PTE entry here before, we release it.
* This can happen with a write to a previously read-only entry.
*/
release_pte(*spte);
/* If this is a write, we insist that the Guest page is writable (the
* final arg to gpte_to_spte()). */
/*
* If this is a write, we insist that the Guest page is writable (the
* final arg to gpte_to_spte()).
*/
if (pte_dirty(gpte))
*spte = gpte_to_spte(cpu, gpte, 1);
else
/* If this is a read, don't set the "writable" bit in the page
/*
* If this is a read, don't set the "writable" bit in the page
* table entry, even if the Guest says it's writable. That way
* we will come back here when a write does actually occur, so
* we can update the Guest's _PAGE_DIRTY flag. */
* we can update the Guest's _PAGE_DIRTY flag.
*/
native_set_pte(spte, gpte_to_spte(cpu, pte_wrprotect(gpte), 0));
/* Finally, we write the Guest PTE entry back: we've set the
* _PAGE_ACCESSED and maybe the _PAGE_DIRTY flags. */
/*
* Finally, we write the Guest PTE entry back: we've set the
* _PAGE_ACCESSED and maybe the _PAGE_DIRTY flags.
*/
lgwrite(cpu, gpte_ptr, pte_t, gpte);
/* The fault is fixed, the page table is populated, the mapping
/*
* The fault is fixed, the page table is populated, the mapping
* manipulated, the result returned and the code complete. A small
* delay and a trace of alliteration are the only indications the Guest
* has that a page fault occurred at all. */
* has that a page fault occurred at all.
*/
return true;
}
......@@ -408,7 +473,8 @@ bool demand_page(struct lg_cpu *cpu, unsigned long vaddr, int errcode)
* mapped, so it's overkill.
*
* This is a quick version which answers the question: is this virtual address
* mapped by the shadow page tables, and is it writable? */
* mapped by the shadow page tables, and is it writable?
*/
static bool page_writable(struct lg_cpu *cpu, unsigned long vaddr)
{
pgd_t *spgd;
......@@ -428,21 +494,26 @@ static bool page_writable(struct lg_cpu *cpu, unsigned long vaddr)
return false;
#endif
/* Check the flags on the pte entry itself: it must be present and
* writable. */
/*
* Check the flags on the pte entry itself: it must be present and
* writable.
*/
flags = pte_flags(*(spte_addr(cpu, *spgd, vaddr)));
return (flags & (_PAGE_PRESENT|_PAGE_RW)) == (_PAGE_PRESENT|_PAGE_RW);
}
/* So, when pin_stack_pages() asks us to pin a page, we check if it's already
/*
* So, when pin_stack_pages() asks us to pin a page, we check if it's already
* in the page tables, and if not, we call demand_page() with error code 2
* (meaning "write"). */
* (meaning "write").
*/
void pin_page(struct lg_cpu *cpu, unsigned long vaddr)
{
if (!page_writable(cpu, vaddr) && !demand_page(cpu, vaddr, 2))
kill_guest(cpu, "bad stack page %#lx", vaddr);
}
/*:*/
#ifdef CONFIG_X86_PAE
static void release_pmd(pmd_t *spmd)
......@@ -479,15 +550,21 @@ static void release_pgd(pgd_t *spgd)
}
#else /* !CONFIG_X86_PAE */
/*H:450 If we chase down the release_pgd() code, it looks like this: */
/*H:450
* If we chase down the release_pgd() code, the non-PAE version looks like
* this. The PAE version is almost identical, but instead of calling
* release_pte it calls release_pmd(), which looks much like this.
*/
static void release_pgd(pgd_t *spgd)
{
/* If the entry's not present, there's nothing to release. */
if (pgd_flags(*spgd) & _PAGE_PRESENT) {
unsigned int i;
/* Converting the pfn to find the actual PTE page is easy: turn
/*
* Converting the pfn to find the actual PTE page is easy: turn
* the page number into a physical address, then convert to a
* virtual address (easy for kernel pages like this one). */
* virtual address (easy for kernel pages like this one).
*/
pte_t *ptepage = __va(pgd_pfn(*spgd) << PAGE_SHIFT);
/* For each entry in the page, we might need to release it. */
for (i = 0; i < PTRS_PER_PTE; i++)
......@@ -499,9 +576,12 @@ static void release_pgd(pgd_t *spgd)
}
}
#endif
/*H:445 We saw flush_user_mappings() twice: once from the flush_user_mappings()
/*H:445
* We saw flush_user_mappings() twice: once from the flush_user_mappings()
* hypercall and once in new_pgdir() when we re-used a top-level pgdir page.
* It simply releases every PTE page from 0 up to the Guest's kernel address. */
* It simply releases every PTE page from 0 up to the Guest's kernel address.
*/
static void flush_user_mappings(struct lguest *lg, int idx)
{
unsigned int i;
......@@ -510,10 +590,12 @@ static void flush_user_mappings(struct lguest *lg, int idx)
release_pgd(lg->pgdirs[idx].pgdir + i);
}
/*H:440 (v) Flushing (throwing away) page tables,
/*H:440
* (v) Flushing (throwing away) page tables,
*
* The Guest has a hypercall to throw away the page tables: it's used when a
* large number of mappings have been changed. */
* large number of mappings have been changed.
*/
void guest_pagetable_flush_user(struct lg_cpu *cpu)
{
/* Drop the userspace part of the current page table. */
......@@ -551,9 +633,11 @@ unsigned long guest_pa(struct lg_cpu *cpu, unsigned long vaddr)
return pte_pfn(gpte) * PAGE_SIZE | (vaddr & ~PAGE_MASK);
}
/* We keep several page tables. This is a simple routine to find the page
/*
* We keep several page tables. This is a simple routine to find the page
* table (if any) corresponding to this top-level address the Guest has given
* us. */
* us.
*/
static unsigned int find_pgdir(struct lguest *lg, unsigned long pgtable)
{
unsigned int i;
......@@ -563,9 +647,11 @@ static unsigned int find_pgdir(struct lguest *lg, unsigned long pgtable)
return i;
}
/*H:435 And this is us, creating the new page directory. If we really do
/*H:435
* And this is us, creating the new page directory. If we really do
* allocate a new one (and so the kernel parts are not there), we set
* blank_pgdir. */
* blank_pgdir.
*/
static unsigned int new_pgdir(struct lg_cpu *cpu,
unsigned long gpgdir,
int *blank_pgdir)
......@@ -575,8 +661,10 @@ static unsigned int new_pgdir(struct lg_cpu *cpu,
pmd_t *pmd_table;
#endif
/* We pick one entry at random to throw out. Choosing the Least
* Recently Used might be better, but this is easy. */
/*
* We pick one entry at random to throw out. Choosing the Least
* Recently Used might be better, but this is easy.
*/
next = random32() % ARRAY_SIZE(cpu->lg->pgdirs);
/* If it's never been allocated at all before, try now. */
if (!cpu->lg->pgdirs[next].pgdir) {
......@@ -587,8 +675,10 @@ static unsigned int new_pgdir(struct lg_cpu *cpu,
next = cpu->cpu_pgd;
else {
#ifdef CONFIG_X86_PAE
/* In PAE mode, allocate a pmd page and populate the
* last pgd entry. */
/*
* In PAE mode, allocate a pmd page and populate the
* last pgd entry.
*/
pmd_table = (pmd_t *)get_zeroed_page(GFP_KERNEL);
if (!pmd_table) {
free_page((long)cpu->lg->pgdirs[next].pgdir);
......@@ -598,8 +688,10 @@ static unsigned int new_pgdir(struct lg_cpu *cpu,
set_pgd(cpu->lg->pgdirs[next].pgdir +
SWITCHER_PGD_INDEX,
__pgd(__pa(pmd_table) | _PAGE_PRESENT));
/* This is a blank page, so there are no kernel
* mappings: caller must map the stack! */
/*
* This is a blank page, so there are no kernel
* mappings: caller must map the stack!
*/
*blank_pgdir = 1;
}
#else
......@@ -615,19 +707,23 @@ static unsigned int new_pgdir(struct lg_cpu *cpu,
return next;
}
/*H:430 (iv) Switching page tables
/*H:430
* (iv) Switching page tables
*
* Now we've seen all the page table setting and manipulation, let's see
* what happens when the Guest changes page tables (ie. changes the top-level
* pgdir). This occurs on almost every context switch. */
* pgdir). This occurs on almost every context switch.
*/
void guest_new_pagetable(struct lg_cpu *cpu, unsigned long pgtable)
{
int newpgdir, repin = 0;
/* Look to see if we have this one already. */
newpgdir = find_pgdir(cpu->lg, pgtable);
/* If not, we allocate or mug an existing one: if it's a fresh one,
* repin gets set to 1. */
/*
* If not, we allocate or mug an existing one: if it's a fresh one,
* repin gets set to 1.
*/
if (newpgdir == ARRAY_SIZE(cpu->lg->pgdirs))
newpgdir = new_pgdir(cpu, pgtable, &repin);
/* Change the current pgd index to the new one. */
......@@ -637,9 +733,11 @@ void guest_new_pagetable(struct lg_cpu *cpu, unsigned long pgtable)
pin_stack_pages(cpu);
}
/*H:470 Finally, a routine which throws away everything: all PGD entries in all
/*H:470
* Finally, a routine which throws away everything: all PGD entries in all
* the shadow page tables, including the Guest's kernel mappings. This is used
* when we destroy the Guest. */
* when we destroy the Guest.
*/
static void release_all_pagetables(struct lguest *lg)
{
unsigned int i, j;
......@@ -656,8 +754,10 @@ static void release_all_pagetables(struct lguest *lg)
spgd = lg->pgdirs[i].pgdir + SWITCHER_PGD_INDEX;
pmdpage = __va(pgd_pfn(*spgd) << PAGE_SHIFT);
/* And release the pmd entries of that pmd page,
* except for the switcher pmd. */
/*
* And release the pmd entries of that pmd page,
* except for the switcher pmd.
*/
for (k = 0; k < SWITCHER_PMD_INDEX; k++)
release_pmd(&pmdpage[k]);
#endif
......@@ -667,10 +767,12 @@ static void release_all_pagetables(struct lguest *lg)
}
}
/* We also throw away everything when a Guest tells us it's changed a kernel
/*
* We also throw away everything when a Guest tells us it's changed a kernel
* mapping. Since kernel mappings are in every page table, it's easiest to
* throw them all away. This traps the Guest in amber for a while as
* everything faults back in, but it's rare. */
* everything faults back in, but it's rare.
*/
void guest_pagetable_clear_all(struct lg_cpu *cpu)
{
release_all_pagetables(cpu->lg);
......@@ -678,15 +780,19 @@ void guest_pagetable_clear_all(struct lg_cpu *cpu)
pin_stack_pages(cpu);
}
/*:*/
/*M:009 Since we throw away all mappings when a kernel mapping changes, our
/*M:009
* Since we throw away all mappings when a kernel mapping changes, our
* performance sucks for guests using highmem. In fact, a guest with
* PAGE_OFFSET 0xc0000000 (the default) and more than about 700MB of RAM is
* usually slower than a Guest with less memory.
*
* This, of course, cannot be fixed. It would take some kind of... well, I
* don't know, but the term "puissant code-fu" comes to mind. :*/
* don't know, but the term "puissant code-fu" comes to mind.
:*/
/*H:420 This is the routine which actually sets the page table entry for then
/*H:420
* This is the routine which actually sets the page table entry for then
* "idx"'th shadow page table.
*
* Normally, we can just throw out the old entry and replace it with 0: if they
......@@ -715,31 +821,36 @@ static void do_set_pte(struct lg_cpu *cpu, int idx,
spmd = spmd_addr(cpu, *spgd, vaddr);
if (pmd_flags(*spmd) & _PAGE_PRESENT) {
#endif
/* Otherwise, we start by releasing
* the existing entry. */
/* Otherwise, start by releasing the existing entry. */
pte_t *spte = spte_addr(cpu, *spgd, vaddr);
release_pte(*spte);
/* If they're setting this entry as dirty or accessed,
* we might as well put that entry they've given us
* in now. This shaves 10% off a
* copy-on-write micro-benchmark. */
/*
* If they're setting this entry as dirty or accessed,
* we might as well put that entry they've given us in
* now. This shaves 10% off a copy-on-write
* micro-benchmark.
*/
if (pte_flags(gpte) & (_PAGE_DIRTY | _PAGE_ACCESSED)) {
check_gpte(cpu, gpte);
native_set_pte(spte,
gpte_to_spte(cpu, gpte,
pte_flags(gpte) & _PAGE_DIRTY));
} else
/* Otherwise kill it and we can demand_page()
* it in later. */
} else {
/*
* Otherwise kill it and we can demand_page()
* it in later.
*/
native_set_pte(spte, __pte(0));
}
#ifdef CONFIG_X86_PAE
}
#endif
}
}
/*H:410 Updating a PTE entry is a little trickier.
/*H:410
* Updating a PTE entry is a little trickier.
*
* We keep track of several different page tables (the Guest uses one for each
* process, so it makes sense to cache at least a few). Each of these have
......@@ -748,12 +859,15 @@ static void do_set_pte(struct lg_cpu *cpu, int idx,
* all the page tables, not just the current one. This is rare.
*
* The benefit is that when we have to track a new page table, we can keep all
* the kernel mappings. This speeds up context switch immensely. */
* the kernel mappings. This speeds up context switch immensely.
*/
void guest_set_pte(struct lg_cpu *cpu,
unsigned long gpgdir, unsigned long vaddr, pte_t gpte)
{
/* Kernel mappings must be changed on all top levels. Slow, but doesn't
* happen often. */
/*
* Kernel mappings must be changed on all top levels. Slow, but doesn't
* happen often.
*/
if (vaddr >= cpu->lg->kernel_address) {
unsigned int i;
for (i = 0; i < ARRAY_SIZE(cpu->lg->pgdirs); i++)
......@@ -795,19 +909,25 @@ void guest_set_pgd(struct lguest *lg, unsigned long gpgdir, u32 idx)
/* ... throw it away. */
release_pgd(lg->pgdirs[pgdir].pgdir + idx);
}
#ifdef CONFIG_X86_PAE
/* For setting a mid-level, we just throw everything away. It's easy. */
void guest_set_pmd(struct lguest *lg, unsigned long pmdp, u32 idx)
{
guest_pagetable_clear_all(&lg->cpus[0]);
}
#endif
/* Once we know how much memory we have we can construct simple identity
* (which set virtual == physical) and linear mappings
* which will get the Guest far enough into the boot to create its own.
/*H:505
* To get through boot, we construct simple identity page mappings (which
* set virtual == physical) and linear mappings which will get the Guest far
* enough into the boot to create its own. The linear mapping means we
* simplify the Guest boot, but it makes assumptions about their PAGE_OFFSET,
* as you'll see.
*
* We lay them out of the way, just below the initrd (which is why we need to
* know its size here). */
* know its size here).
*/
static unsigned long setup_pagetables(struct lguest *lg,
unsigned long mem,
unsigned long initrd_size)
......@@ -825,8 +945,10 @@ static unsigned long setup_pagetables(struct lguest *lg,
unsigned int phys_linear;
#endif
/* We have mapped_pages frames to map, so we need
* linear_pages page tables to map them. */
/*
* We have mapped_pages frames to map, so we need linear_pages page
* tables to map them.
*/
mapped_pages = mem / PAGE_SIZE;
linear_pages = (mapped_pages + PTRS_PER_PTE - 1) / PTRS_PER_PTE;
......@@ -837,10 +959,16 @@ static unsigned long setup_pagetables(struct lguest *lg,
linear = (void *)pgdir - linear_pages * PAGE_SIZE;
#ifdef CONFIG_X86_PAE
/*
* And the single mid page goes below that. We only use one, but
* that's enough to map 1G, which definitely gets us through boot.
*/
pmds = (void *)linear - PAGE_SIZE;
#endif
/* Linear mapping is easy: put every page's address into the
* mapping in order. */
/*
* Linear mapping is easy: put every page's address into the
* mapping in order.
*/
for (i = 0; i < mapped_pages; i++) {
pte_t pte;
pte = pfn_pte(i, __pgprot(_PAGE_PRESENT|_PAGE_RW|_PAGE_USER));
......@@ -848,11 +976,14 @@ static unsigned long setup_pagetables(struct lguest *lg,
return -EFAULT;
}
/* The top level points to the linear page table pages above.
* We setup the identity and linear mappings here. */
#ifdef CONFIG_X86_PAE
/*
* Make the Guest PMD entries point to the corresponding place in the
* linear mapping (up to one page worth of PMD).
*/
for (i = j = 0; i < mapped_pages && j < PTRS_PER_PMD;
i += PTRS_PER_PTE, j++) {
/* FIXME: native_set_pmd is overkill here. */
native_set_pmd(&pmd, __pmd(((unsigned long)(linear + i)
- mem_base) | _PAGE_PRESENT | _PAGE_RW | _PAGE_USER));
......@@ -860,18 +991,36 @@ static unsigned long setup_pagetables(struct lguest *lg,
return -EFAULT;
}
/* One PGD entry, pointing to that PMD page. */
set_pgd(&pgd, __pgd(((u32)pmds - mem_base) | _PAGE_PRESENT));
/* Copy it in as the first PGD entry (ie. addresses 0-1G). */
if (copy_to_user(&pgdir[0], &pgd, sizeof(pgd)) != 0)
return -EFAULT;
/*
* And the third PGD entry (ie. addresses 3G-4G).
*
* FIXME: This assumes that PAGE_OFFSET for the Guest is 0xC0000000.
*/
if (copy_to_user(&pgdir[3], &pgd, sizeof(pgd)) != 0)
return -EFAULT;
#else
/*
* The top level points to the linear page table pages above.
* We setup the identity and linear mappings here.
*/
phys_linear = (unsigned long)linear - mem_base;
for (i = 0; i < mapped_pages; i += PTRS_PER_PTE) {
pgd_t pgd;
/*
* Create a PGD entry which points to the right part of the
* linear PTE pages.
*/
pgd = __pgd((phys_linear + i * sizeof(pte_t)) |
(_PAGE_PRESENT | _PAGE_RW | _PAGE_USER));
/*
* Copy it into the PGD page at 0 and PAGE_OFFSET.
*/
if (copy_to_user(&pgdir[i / PTRS_PER_PTE], &pgd, sizeof(pgd))
|| copy_to_user(&pgdir[pgd_index(PAGE_OFFSET)
+ i / PTRS_PER_PTE],
......@@ -880,15 +1029,19 @@ static unsigned long setup_pagetables(struct lguest *lg,
}
#endif
/* We return the top level (guest-physical) address: remember where
* this is. */
/*
* We return the top level (guest-physical) address: we remember where
* this is to write it into lguest_data when the Guest initializes.
*/
return (unsigned long)pgdir - mem_base;
}
/*H:500 (vii) Setting up the page tables initially.
/*H:500
* (vii) Setting up the page tables initially.
*
* When a Guest is first created, the Launcher tells us where the toplevel of
* its first page table is. We set some things up here: */
* its first page table is. We set some things up here:
*/
int init_guest_pagetable(struct lguest *lg)
{
u64 mem;
......@@ -898,21 +1051,27 @@ int init_guest_pagetable(struct lguest *lg)
pgd_t *pgd;
pmd_t *pmd_table;
#endif
/* Get the Guest memory size and the ramdisk size from the boot header
* located at lg->mem_base (Guest address 0). */
/*
* Get the Guest memory size and the ramdisk size from the boot header
* located at lg->mem_base (Guest address 0).
*/
if (copy_from_user(&mem, &boot->e820_map[0].size, sizeof(mem))
|| get_user(initrd_size, &boot->hdr.ramdisk_size))
return -EFAULT;
/* We start on the first shadow page table, and give it a blank PGD
* page. */
/*
* We start on the first shadow page table, and give it a blank PGD
* page.
*/
lg->pgdirs[0].gpgdir = setup_pagetables(lg, mem, initrd_size);
if (IS_ERR_VALUE(lg->pgdirs[0].gpgdir))
return lg->pgdirs[0].gpgdir;
lg->pgdirs[0].pgdir = (pgd_t *)get_zeroed_page(GFP_KERNEL);
if (!lg->pgdirs[0].pgdir)
return -ENOMEM;
#ifdef CONFIG_X86_PAE
/* For PAE, we also create the initial mid-level. */
pgd = lg->pgdirs[0].pgdir;
pmd_table = (pmd_t *) get_zeroed_page(GFP_KERNEL);
if (!pmd_table)
......@@ -921,27 +1080,33 @@ int init_guest_pagetable(struct lguest *lg)
set_pgd(pgd + SWITCHER_PGD_INDEX,
__pgd(__pa(pmd_table) | _PAGE_PRESENT));
#endif
/* This is the current page table. */
lg->cpus[0].cpu_pgd = 0;
return 0;
}
/* When the Guest calls LHCALL_LGUEST_INIT we do more setup. */
/*H:508 When the Guest calls LHCALL_LGUEST_INIT we do more setup. */
void page_table_guest_data_init(struct lg_cpu *cpu)
{
/* We get the kernel address: above this is all kernel memory. */
if (get_user(cpu->lg->kernel_address,
&cpu->lg->lguest_data->kernel_address)
/* We tell the Guest that it can't use the top 2 or 4 MB
* of virtual addresses used by the Switcher. */
/*
* We tell the Guest that it can't use the top 2 or 4 MB
* of virtual addresses used by the Switcher.
*/
|| put_user(RESERVE_MEM * 1024 * 1024,
&cpu->lg->lguest_data->reserve_mem)
|| put_user(cpu->lg->pgdirs[0].gpgdir,
&cpu->lg->lguest_data->pgdir))
kill_guest(cpu, "bad guest page %p", cpu->lg->lguest_data);
/* In flush_user_mappings() we loop from 0 to
/*
* In flush_user_mappings() we loop from 0 to
* "pgd_index(lg->kernel_address)". This assumes it won't hit the
* Switcher mappings, so check that now. */
* Switcher mappings, so check that now.
*/
#ifdef CONFIG_X86_PAE
if (pgd_index(cpu->lg->kernel_address) == SWITCHER_PGD_INDEX &&
pmd_index(cpu->lg->kernel_address) == SWITCHER_PMD_INDEX)
......@@ -964,12 +1129,14 @@ void free_guest_pagetable(struct lguest *lg)
free_page((long)lg->pgdirs[i].pgdir);
}
/*H:480 (vi) Mapping the Switcher when the Guest is about to run.
/*H:480
* (vi) Mapping the Switcher when the Guest is about to run.
*
* The Switcher and the two pages for this CPU need to be visible in the
* Guest (and not the pages for other CPUs). We have the appropriate PTE pages
* for each CPU already set up, we just need to hook them in now we know which
* Guest is about to run on this CPU. */
* Guest is about to run on this CPU.
*/
void map_switcher_in_guest(struct lg_cpu *cpu, struct lguest_pages *pages)
{
pte_t *switcher_pte_page = __get_cpu_var(switcher_pte_pages);
......@@ -980,30 +1147,38 @@ void map_switcher_in_guest(struct lg_cpu *cpu, struct lguest_pages *pages)
pmd_t switcher_pmd;
pmd_t *pmd_table;
/* FIXME: native_set_pmd is overkill here. */
native_set_pmd(&switcher_pmd, pfn_pmd(__pa(switcher_pte_page) >>
PAGE_SHIFT, PAGE_KERNEL_EXEC));
/* Figure out where the pmd page is, by reading the PGD, and converting
* it to a virtual address. */
pmd_table = __va(pgd_pfn(cpu->lg->
pgdirs[cpu->cpu_pgd].pgdir[SWITCHER_PGD_INDEX])
<< PAGE_SHIFT);
/* Now write it into the shadow page table. */
native_set_pmd(&pmd_table[SWITCHER_PMD_INDEX], switcher_pmd);
#else
pgd_t switcher_pgd;
/* Make the last PGD entry for this Guest point to the Switcher's PTE
* page for this CPU (with appropriate flags). */
/*
* Make the last PGD entry for this Guest point to the Switcher's PTE
* page for this CPU (with appropriate flags).
*/
switcher_pgd = __pgd(__pa(switcher_pte_page) | __PAGE_KERNEL_EXEC);
cpu->lg->pgdirs[cpu->cpu_pgd].pgdir[SWITCHER_PGD_INDEX] = switcher_pgd;
#endif
/* We also change the Switcher PTE page. When we're running the Guest,
/*
* We also change the Switcher PTE page. When we're running the Guest,
* we want the Guest's "regs" page to appear where the first Switcher
* page for this CPU is. This is an optimization: when the Switcher
* saves the Guest registers, it saves them into the first page of this
* CPU's "struct lguest_pages": if we make sure the Guest's register
* page is already mapped there, we don't have to copy them out
* again. */
* again.
*/
pfn = __pa(cpu->regs_page) >> PAGE_SHIFT;
native_set_pte(&regs_pte, pfn_pte(pfn, PAGE_KERNEL));
native_set_pte(&switcher_pte_page[pte_index((unsigned long)pages)],
......@@ -1019,10 +1194,12 @@ static void free_switcher_pte_pages(void)
free_page((long)switcher_pte_page(i));
}
/*H:520 Setting up the Switcher PTE page for given CPU is fairly easy, given
/*H:520
* Setting up the Switcher PTE page for given CPU is fairly easy, given
* the CPU number and the "struct page"s for the Switcher code itself.
*
* Currently the Switcher is less than a page long, so "pages" is always 1. */
* Currently the Switcher is less than a page long, so "pages" is always 1.
*/
static __init void populate_switcher_pte_page(unsigned int cpu,
struct page *switcher_page[],
unsigned int pages)
......@@ -1043,13 +1220,16 @@ static __init void populate_switcher_pte_page(unsigned int cpu,
native_set_pte(&pte[i], pfn_pte(page_to_pfn(switcher_page[i]),
__pgprot(_PAGE_PRESENT|_PAGE_ACCESSED|_PAGE_RW)));
/* The second page contains the "struct lguest_ro_state", and is
* read-only. */
/*
* The second page contains the "struct lguest_ro_state", and is
* read-only.
*/
native_set_pte(&pte[i+1], pfn_pte(page_to_pfn(switcher_page[i+1]),
__pgprot(_PAGE_PRESENT|_PAGE_ACCESSED)));
}
/* We've made it through the page table code. Perhaps our tired brains are
/*
* We've made it through the page table code. Perhaps our tired brains are
* still processing the details, or perhaps we're simply glad it's over.
*
* If nothing else, note that all this complexity in juggling shadow page tables
......@@ -1058,10 +1238,13 @@ static __init void populate_switcher_pte_page(unsigned int cpu,
* uses exotic direct Guest pagetable manipulation, and why both Intel and AMD
* have implemented shadow page table support directly into hardware.
*
* There is just one file remaining in the Host. */
* There is just one file remaining in the Host.
*/
/*H:510 At boot or module load time, init_pagetables() allocates and populates
* the Switcher PTE page for each CPU. */
/*H:510
* At boot or module load time, init_pagetables() allocates and populates
* the Switcher PTE page for each CPU.
*/
__init int init_pagetables(struct page **switcher_page, unsigned int pages)
{
unsigned int i;
......
/*P:600 The x86 architecture has segments, which involve a table of descriptors
/*P:600
* The x86 architecture has segments, which involve a table of descriptors
* which can be used to do funky things with virtual address interpretation.
* We originally used to use segments so the Guest couldn't alter the
* Guest<->Host Switcher, and then we had to trim Guest segments, and restore
......@@ -8,7 +9,8 @@
*
* In these modern times, the segment handling code consists of simple sanity
* checks, and the worst you'll experience reading this code is butterfly-rash
* from frolicking through its parklike serenity. :*/
* from frolicking through its parklike serenity.
:*/
#include "lg.h"
/*H:600
......@@ -41,10 +43,12 @@
* begin.
*/
/* There are several entries we don't let the Guest set. The TSS entry is the
/*
* There are several entries we don't let the Guest set. The TSS entry is the
* "Task State Segment" which controls all kinds of delicate things. The
* LGUEST_CS and LGUEST_DS entries are reserved for the Switcher, and the
* the Guest can't be trusted to deal with double faults. */
* the Guest can't be trusted to deal with double faults.
*/
static bool ignored_gdt(unsigned int num)
{
return (num == GDT_ENTRY_TSS
......@@ -53,42 +57,52 @@ static bool ignored_gdt(unsigned int num)
|| num == GDT_ENTRY_DOUBLEFAULT_TSS);
}
/*H:630 Once the Guest gave us new GDT entries, we fix them up a little. We
/*H:630
* Once the Guest gave us new GDT entries, we fix them up a little. We
* don't care if they're invalid: the worst that can happen is a General
* Protection Fault in the Switcher when it restores a Guest segment register
* which tries to use that entry. Then we kill the Guest for causing such a
* mess: the message will be "unhandled trap 256". */
* mess: the message will be "unhandled trap 256".
*/
static void fixup_gdt_table(struct lg_cpu *cpu, unsigned start, unsigned end)
{
unsigned int i;
for (i = start; i < end; i++) {
/* We never copy these ones to real GDT, so we don't care what
* they say */
/*
* We never copy these ones to real GDT, so we don't care what
* they say
*/
if (ignored_gdt(i))
continue;
/* Segment descriptors contain a privilege level: the Guest is
/*
* Segment descriptors contain a privilege level: the Guest is
* sometimes careless and leaves this as 0, even though it's
* running at privilege level 1. If so, we fix it here. */
* running at privilege level 1. If so, we fix it here.
*/
if ((cpu->arch.gdt[i].b & 0x00006000) == 0)
cpu->arch.gdt[i].b |= (GUEST_PL << 13);
/* Each descriptor has an "accessed" bit. If we don't set it
/*
* Each descriptor has an "accessed" bit. If we don't set it
* now, the CPU will try to set it when the Guest first loads
* that entry into a segment register. But the GDT isn't
* writable by the Guest, so bad things can happen. */
* writable by the Guest, so bad things can happen.
*/
cpu->arch.gdt[i].b |= 0x00000100;
}
}
/*H:610 Like the IDT, we never simply use the GDT the Guest gives us. We keep
/*H:610
* Like the IDT, we never simply use the GDT the Guest gives us. We keep
* a GDT for each CPU, and copy across the Guest's entries each time we want to
* run the Guest on that CPU.
*
* This routine is called at boot or modprobe time for each CPU to set up the
* constant GDT entries: the ones which are the same no matter what Guest we're
* running. */
* running.
*/
void setup_default_gdt_entries(struct lguest_ro_state *state)
{
struct desc_struct *gdt = state->guest_gdt;
......@@ -98,30 +112,37 @@ void setup_default_gdt_entries(struct lguest_ro_state *state)
gdt[GDT_ENTRY_LGUEST_CS] = FULL_EXEC_SEGMENT;
gdt[GDT_ENTRY_LGUEST_DS] = FULL_SEGMENT;
/* The TSS segment refers to the TSS entry for this particular CPU.
/*
* The TSS segment refers to the TSS entry for this particular CPU.
* Forgive the magic flags: the 0x8900 means the entry is Present, it's
* privilege level 0 Available 386 TSS system segment, and the 0x67
* means Saturn is eclipsed by Mercury in the twelfth house. */
* means Saturn is eclipsed by Mercury in the twelfth house.
*/
gdt[GDT_ENTRY_TSS].a = 0x00000067 | (tss << 16);
gdt[GDT_ENTRY_TSS].b = 0x00008900 | (tss & 0xFF000000)
| ((tss >> 16) & 0x000000FF);
}
/* This routine sets up the initial Guest GDT for booting. All entries start
* as 0 (unusable). */
/*
* This routine sets up the initial Guest GDT for booting. All entries start
* as 0 (unusable).
*/
void setup_guest_gdt(struct lg_cpu *cpu)
{
/* Start with full 0-4G segments... */
/*
* Start with full 0-4G segments...except the Guest is allowed to use
* them, so set the privilege level appropriately in the flags.
*/
cpu->arch.gdt[GDT_ENTRY_KERNEL_CS] = FULL_EXEC_SEGMENT;
cpu->arch.gdt[GDT_ENTRY_KERNEL_DS] = FULL_SEGMENT;
/* ...except the Guest is allowed to use them, so set the privilege
* level appropriately in the flags. */
cpu->arch.gdt[GDT_ENTRY_KERNEL_CS].b |= (GUEST_PL << 13);
cpu->arch.gdt[GDT_ENTRY_KERNEL_DS].b |= (GUEST_PL << 13);
}
/*H:650 An optimization of copy_gdt(), for just the three "thead-local storage"
* entries. */
/*H:650
* An optimization of copy_gdt(), for just the three "thead-local storage"
* entries.
*/
void copy_gdt_tls(const struct lg_cpu *cpu, struct desc_struct *gdt)
{
unsigned int i;
......@@ -130,26 +151,34 @@ void copy_gdt_tls(const struct lg_cpu *cpu, struct desc_struct *gdt)
gdt[i] = cpu->arch.gdt[i];
}
/*H:640 When the Guest is run on a different CPU, or the GDT entries have
* changed, copy_gdt() is called to copy the Guest's GDT entries across to this
* CPU's GDT. */
/*H:640
* When the Guest is run on a different CPU, or the GDT entries have changed,
* copy_gdt() is called to copy the Guest's GDT entries across to this CPU's
* GDT.
*/
void copy_gdt(const struct lg_cpu *cpu, struct desc_struct *gdt)
{
unsigned int i;
/* The default entries from setup_default_gdt_entries() are not
* replaced. See ignored_gdt() above. */
/*
* The default entries from setup_default_gdt_entries() are not
* replaced. See ignored_gdt() above.
*/
for (i = 0; i < GDT_ENTRIES; i++)
if (!ignored_gdt(i))
gdt[i] = cpu->arch.gdt[i];
}
/*H:620 This is where the Guest asks us to load a new GDT entry
* (LHCALL_LOAD_GDT_ENTRY). We tweak the entry and copy it in. */
/*H:620
* This is where the Guest asks us to load a new GDT entry
* (LHCALL_LOAD_GDT_ENTRY). We tweak the entry and copy it in.
*/
void load_guest_gdt_entry(struct lg_cpu *cpu, u32 num, u32 lo, u32 hi)
{
/* We assume the Guest has the same number of GDT entries as the
* Host, otherwise we'd have to dynamically allocate the Guest GDT. */
/*
* We assume the Guest has the same number of GDT entries as the
* Host, otherwise we'd have to dynamically allocate the Guest GDT.
*/
if (num >= ARRAY_SIZE(cpu->arch.gdt))
kill_guest(cpu, "too many gdt entries %i", num);
......@@ -157,15 +186,19 @@ void load_guest_gdt_entry(struct lg_cpu *cpu, u32 num, u32 lo, u32 hi)
cpu->arch.gdt[num].a = lo;
cpu->arch.gdt[num].b = hi;
fixup_gdt_table(cpu, num, num+1);
/* Mark that the GDT changed so the core knows it has to copy it again,
* even if the Guest is run on the same CPU. */
/*
* Mark that the GDT changed so the core knows it has to copy it again,
* even if the Guest is run on the same CPU.
*/
cpu->changed |= CHANGED_GDT;
}
/* This is the fast-track version for just changing the three TLS entries.
/*
* This is the fast-track version for just changing the three TLS entries.
* Remember that this happens on every context switch, so it's worth
* optimizing. But wouldn't it be neater to have a single hypercall to cover
* both cases? */
* both cases?
*/
void guest_load_tls(struct lg_cpu *cpu, unsigned long gtls)
{
struct desc_struct *tls = &cpu->arch.gdt[GDT_ENTRY_TLS_MIN];
......@@ -175,7 +208,6 @@ void guest_load_tls(struct lg_cpu *cpu, unsigned long gtls)
/* Note that just the TLS entries have changed. */
cpu->changed |= CHANGED_GDT_TLS;
}
/*:*/
/*H:660
* With this, we have finished the Host.
......
......@@ -17,13 +17,15 @@
* along with this program; if not, write to the Free Software
* Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
*/
/*P:450 This file contains the x86-specific lguest code. It used to be all
/*P:450
* This file contains the x86-specific lguest code. It used to be all
* mixed in with drivers/lguest/core.c but several foolhardy code slashers
* wrestled most of the dependencies out to here in preparation for porting
* lguest to other architectures (see what I mean by foolhardy?).
*
* This also contains a couple of non-obvious setup and teardown pieces which
* were implemented after days of debugging pain. :*/
* were implemented after days of debugging pain.
:*/
#include <linux/kernel.h>
#include <linux/start_kernel.h>
#include <linux/string.h>
......@@ -82,25 +84,33 @@ static DEFINE_PER_CPU(struct lg_cpu *, last_cpu);
*/
static void copy_in_guest_info(struct lg_cpu *cpu, struct lguest_pages *pages)
{
/* Copying all this data can be quite expensive. We usually run the
/*
* Copying all this data can be quite expensive. We usually run the
* same Guest we ran last time (and that Guest hasn't run anywhere else
* meanwhile). If that's not the case, we pretend everything in the
* Guest has changed. */
* Guest has changed.
*/
if (__get_cpu_var(last_cpu) != cpu || cpu->last_pages != pages) {
__get_cpu_var(last_cpu) = cpu;
cpu->last_pages = pages;
cpu->changed = CHANGED_ALL;
}
/* These copies are pretty cheap, so we do them unconditionally: */
/* Save the current Host top-level page directory. */
/*
* These copies are pretty cheap, so we do them unconditionally: */
/* Save the current Host top-level page directory.
*/
pages->state.host_cr3 = __pa(current->mm->pgd);
/* Set up the Guest's page tables to see this CPU's pages (and no
* other CPU's pages). */
/*
* Set up the Guest's page tables to see this CPU's pages (and no
* other CPU's pages).
*/
map_switcher_in_guest(cpu, pages);
/* Set up the two "TSS" members which tell the CPU what stack to use
/*
* Set up the two "TSS" members which tell the CPU what stack to use
* for traps which do directly into the Guest (ie. traps at privilege
* level 1). */
* level 1).
*/
pages->state.guest_tss.sp1 = cpu->esp1;
pages->state.guest_tss.ss1 = cpu->ss1;
......@@ -125,97 +135,126 @@ static void run_guest_once(struct lg_cpu *cpu, struct lguest_pages *pages)
/* This is a dummy value we need for GCC's sake. */
unsigned int clobber;
/* Copy the guest-specific information into this CPU's "struct
* lguest_pages". */
/*
* Copy the guest-specific information into this CPU's "struct
* lguest_pages".
*/
copy_in_guest_info(cpu, pages);
/* Set the trap number to 256 (impossible value). If we fault while
/*
* Set the trap number to 256 (impossible value). If we fault while
* switching to the Guest (bad segment registers or bug), this will
* cause us to abort the Guest. */
* cause us to abort the Guest.
*/
cpu->regs->trapnum = 256;
/* Now: we push the "eflags" register on the stack, then do an "lcall".
/*
* Now: we push the "eflags" register on the stack, then do an "lcall".
* This is how we change from using the kernel code segment to using
* the dedicated lguest code segment, as well as jumping into the
* Switcher.
*
* The lcall also pushes the old code segment (KERNEL_CS) onto the
* stack, then the address of this call. This stack layout happens to
* exactly match the stack layout created by an interrupt... */
* exactly match the stack layout created by an interrupt...
*/
asm volatile("pushf; lcall *lguest_entry"
/* This is how we tell GCC that %eax ("a") and %ebx ("b")
* are changed by this routine. The "=" means output. */
/*
* This is how we tell GCC that %eax ("a") and %ebx ("b")
* are changed by this routine. The "=" means output.
*/
: "=a"(clobber), "=b"(clobber)
/* %eax contains the pages pointer. ("0" refers to the
/*
* %eax contains the pages pointer. ("0" refers to the
* 0-th argument above, ie "a"). %ebx contains the
* physical address of the Guest's top-level page
* directory. */
* directory.
*/
: "0"(pages), "1"(__pa(cpu->lg->pgdirs[cpu->cpu_pgd].pgdir))
/* We tell gcc that all these registers could change,
/*
* We tell gcc that all these registers could change,
* which means we don't have to save and restore them in
* the Switcher. */
* the Switcher.
*/
: "memory", "%edx", "%ecx", "%edi", "%esi");
}
/*:*/
/*M:002 There are hooks in the scheduler which we can register to tell when we
/*M:002
* There are hooks in the scheduler which we can register to tell when we
* get kicked off the CPU (preempt_notifier_register()). This would allow us
* to lazily disable SYSENTER which would regain some performance, and should
* also simplify copy_in_guest_info(). Note that we'd still need to restore
* things when we exit to Launcher userspace, but that's fairly easy.
*
* We could also try using this hooks for PGE, but that might be too expensive.
* We could also try using these hooks for PGE, but that might be too expensive.
*
* The hooks were designed for KVM, but we can also put them to good use. :*/
* The hooks were designed for KVM, but we can also put them to good use.
:*/
/*H:040 This is the i386-specific code to setup and run the Guest. Interrupts
* are disabled: we own the CPU. */
/*H:040
* This is the i386-specific code to setup and run the Guest. Interrupts
* are disabled: we own the CPU.
*/
void lguest_arch_run_guest(struct lg_cpu *cpu)
{
/* Remember the awfully-named TS bit? If the Guest has asked to set it
/*
* Remember the awfully-named TS bit? If the Guest has asked to set it
* we set it now, so we can trap and pass that trap to the Guest if it
* uses the FPU. */
* uses the FPU.
*/
if (cpu->ts)
unlazy_fpu(current);
/* SYSENTER is an optimized way of doing system calls. We can't allow
/*
* SYSENTER is an optimized way of doing system calls. We can't allow
* it because it always jumps to privilege level 0. A normal Guest
* won't try it because we don't advertise it in CPUID, but a malicious
* Guest (or malicious Guest userspace program) could, so we tell the
* CPU to disable it before running the Guest. */
* CPU to disable it before running the Guest.
*/
if (boot_cpu_has(X86_FEATURE_SEP))
wrmsr(MSR_IA32_SYSENTER_CS, 0, 0);
/* Now we actually run the Guest. It will return when something
/*
* Now we actually run the Guest. It will return when something
* interesting happens, and we can examine its registers to see what it
* was doing. */
* was doing.
*/
run_guest_once(cpu, lguest_pages(raw_smp_processor_id()));
/* Note that the "regs" structure contains two extra entries which are
/*
* Note that the "regs" structure contains two extra entries which are
* not really registers: a trap number which says what interrupt or
* trap made the switcher code come back, and an error code which some
* traps set. */
* traps set.
*/
/* Restore SYSENTER if it's supposed to be on. */
if (boot_cpu_has(X86_FEATURE_SEP))
wrmsr(MSR_IA32_SYSENTER_CS, __KERNEL_CS, 0);
/* If the Guest page faulted, then the cr2 register will tell us the
/*
* If the Guest page faulted, then the cr2 register will tell us the
* bad virtual address. We have to grab this now, because once we
* re-enable interrupts an interrupt could fault and thus overwrite
* cr2, or we could even move off to a different CPU. */
* cr2, or we could even move off to a different CPU.
*/
if (cpu->regs->trapnum == 14)
cpu->arch.last_pagefault = read_cr2();
/* Similarly, if we took a trap because the Guest used the FPU,
/*
* Similarly, if we took a trap because the Guest used the FPU,
* we have to restore the FPU it expects to see.
* math_state_restore() may sleep and we may even move off to
* a different CPU. So all the critical stuff should be done
* before this. */
* before this.
*/
else if (cpu->regs->trapnum == 7)
math_state_restore();
}
/*H:130 Now we've examined the hypercall code; our Guest can make requests.
/*H:130
* Now we've examined the hypercall code; our Guest can make requests.
* Our Guest is usually so well behaved; it never tries to do things it isn't
* allowed to, and uses hypercalls instead. Unfortunately, Linux's paravirtual
* infrastructure isn't quite complete, because it doesn't contain replacements
......@@ -225,26 +264,33 @@ void lguest_arch_run_guest(struct lg_cpu *cpu)
*
* When the Guest uses one of these instructions, we get a trap (General
* Protection Fault) and come here. We see if it's one of those troublesome
* instructions and skip over it. We return true if we did. */
* instructions and skip over it. We return true if we did.
*/
static int emulate_insn(struct lg_cpu *cpu)
{
u8 insn;
unsigned int insnlen = 0, in = 0, shift = 0;
/* The eip contains the *virtual* address of the Guest's instruction:
* guest_pa just subtracts the Guest's page_offset. */
/*
* The eip contains the *virtual* address of the Guest's instruction:
* guest_pa just subtracts the Guest's page_offset.
*/
unsigned long physaddr = guest_pa(cpu, cpu->regs->eip);
/* This must be the Guest kernel trying to do something, not userspace!
/*
* This must be the Guest kernel trying to do something, not userspace!
* The bottom two bits of the CS segment register are the privilege
* level. */
* level.
*/
if ((cpu->regs->cs & 3) != GUEST_PL)
return 0;
/* Decoding x86 instructions is icky. */
insn = lgread(cpu, physaddr, u8);
/* 0x66 is an "operand prefix". It means it's using the upper 16 bits
of the eax register. */
/*
* 0x66 is an "operand prefix". It means it's using the upper 16 bits
* of the eax register.
*/
if (insn == 0x66) {
shift = 16;
/* The instruction is 1 byte so far, read the next byte. */
......@@ -252,8 +298,10 @@ static int emulate_insn(struct lg_cpu *cpu)
insn = lgread(cpu, physaddr + insnlen, u8);
}
/* We can ignore the lower bit for the moment and decode the 4 opcodes
* we need to emulate. */
/*
* We can ignore the lower bit for the moment and decode the 4 opcodes
* we need to emulate.
*/
switch (insn & 0xFE) {
case 0xE4: /* in <next byte>,%al */
insnlen += 2;
......@@ -274,9 +322,11 @@ static int emulate_insn(struct lg_cpu *cpu)
return 0;
}
/* If it was an "IN" instruction, they expect the result to be read
/*
* If it was an "IN" instruction, they expect the result to be read
* into %eax, so we change %eax. We always return all-ones, which
* traditionally means "there's nothing there". */
* traditionally means "there's nothing there".
*/
if (in) {
/* Lower bit tells is whether it's a 16 or 32 bit access */
if (insn & 0x1)
......@@ -290,7 +340,8 @@ static int emulate_insn(struct lg_cpu *cpu)
return 1;
}
/* Our hypercalls mechanism used to be based on direct software interrupts.
/*
* Our hypercalls mechanism used to be based on direct software interrupts.
* After Anthony's "Refactor hypercall infrastructure" kvm patch, we decided to
* change over to using kvm hypercalls.
*
......@@ -318,16 +369,20 @@ static int emulate_insn(struct lg_cpu *cpu)
*/
static void rewrite_hypercall(struct lg_cpu *cpu)
{
/* This are the opcodes we use to patch the Guest. The opcode for "int
/*
* This are the opcodes we use to patch the Guest. The opcode for "int
* $0x1f" is "0xcd 0x1f" but vmcall instruction is 3 bytes long, so we
* complete the sequence with a NOP (0x90). */
* complete the sequence with a NOP (0x90).
*/
u8 insn[3] = {0xcd, 0x1f, 0x90};
__lgwrite(cpu, guest_pa(cpu, cpu->regs->eip), insn, sizeof(insn));
/* The above write might have caused a copy of that page to be made
/*
* The above write might have caused a copy of that page to be made
* (if it was read-only). We need to make sure the Guest has
* up-to-date pagetables. As this doesn't happen often, we can just
* drop them all. */
* drop them all.
*/
guest_pagetable_clear_all(cpu);
}
......@@ -335,9 +390,11 @@ static bool is_hypercall(struct lg_cpu *cpu)
{
u8 insn[3];
/* This must be the Guest kernel trying to do something.
/*
* This must be the Guest kernel trying to do something.
* The bottom two bits of the CS segment register are the privilege
* level. */
* level.
*/
if ((cpu->regs->cs & 3) != GUEST_PL)
return false;
......@@ -351,86 +408,105 @@ void lguest_arch_handle_trap(struct lg_cpu *cpu)
{
switch (cpu->regs->trapnum) {
case 13: /* We've intercepted a General Protection Fault. */
/* Check if this was one of those annoying IN or OUT
/*
* Check if this was one of those annoying IN or OUT
* instructions which we need to emulate. If so, we just go
* back into the Guest after we've done it. */
* back into the Guest after we've done it.
*/
if (cpu->regs->errcode == 0) {
if (emulate_insn(cpu))
return;
}
/* If KVM is active, the vmcall instruction triggers a
* General Protection Fault. Normally it triggers an
* invalid opcode fault (6): */
/*
* If KVM is active, the vmcall instruction triggers a General
* Protection Fault. Normally it triggers an invalid opcode
* fault (6):
*/
case 6:
/* We need to check if ring == GUEST_PL and
* faulting instruction == vmcall. */
/*
* We need to check if ring == GUEST_PL and faulting
* instruction == vmcall.
*/
if (is_hypercall(cpu)) {
rewrite_hypercall(cpu);
return;
}
break;
case 14: /* We've intercepted a Page Fault. */
/* The Guest accessed a virtual address that wasn't mapped.
/*
* The Guest accessed a virtual address that wasn't mapped.
* This happens a lot: we don't actually set up most of the page
* tables for the Guest at all when we start: as it runs it asks
* for more and more, and we set them up as required. In this
* case, we don't even tell the Guest that the fault happened.
*
* The errcode tells whether this was a read or a write, and
* whether kernel or userspace code. */
* whether kernel or userspace code.
*/
if (demand_page(cpu, cpu->arch.last_pagefault,
cpu->regs->errcode))
return;
/* OK, it's really not there (or not OK): the Guest needs to
/*
* OK, it's really not there (or not OK): the Guest needs to
* know. We write out the cr2 value so it knows where the
* fault occurred.
*
* Note that if the Guest were really messed up, this could
* happen before it's done the LHCALL_LGUEST_INIT hypercall, so
* lg->lguest_data could be NULL */
* lg->lguest_data could be NULL
*/
if (cpu->lg->lguest_data &&
put_user(cpu->arch.last_pagefault,
&cpu->lg->lguest_data->cr2))
kill_guest(cpu, "Writing cr2");
break;
case 7: /* We've intercepted a Device Not Available fault. */
/* If the Guest doesn't want to know, we already restored the
* Floating Point Unit, so we just continue without telling
* it. */
/*
* If the Guest doesn't want to know, we already restored the
* Floating Point Unit, so we just continue without telling it.
*/
if (!cpu->ts)
return;
break;
case 32 ... 255:
/* These values mean a real interrupt occurred, in which case
/*
* These values mean a real interrupt occurred, in which case
* the Host handler has already been run. We just do a
* friendly check if another process should now be run, then
* return to run the Guest again */
* return to run the Guest again
*/
cond_resched();
return;
case LGUEST_TRAP_ENTRY:
/* Our 'struct hcall_args' maps directly over our regs: we set
* up the pointer now to indicate a hypercall is pending. */
/*
* Our 'struct hcall_args' maps directly over our regs: we set
* up the pointer now to indicate a hypercall is pending.
*/
cpu->hcall = (struct hcall_args *)cpu->regs;
return;
}
/* We didn't handle the trap, so it needs to go to the Guest. */
if (!deliver_trap(cpu, cpu->regs->trapnum))
/* If the Guest doesn't have a handler (either it hasn't
/*
* If the Guest doesn't have a handler (either it hasn't
* registered any yet, or it's one of the faults we don't let
* it handle), it dies with this cryptic error message. */
* it handle), it dies with this cryptic error message.
*/
kill_guest(cpu, "unhandled trap %li at %#lx (%#lx)",
cpu->regs->trapnum, cpu->regs->eip,
cpu->regs->trapnum == 14 ? cpu->arch.last_pagefault
: cpu->regs->errcode);
}
/* Now we can look at each of the routines this calls, in increasing order of
/*
* Now we can look at each of the routines this calls, in increasing order of
* complexity: do_hypercalls(), emulate_insn(), maybe_do_interrupt(),
* deliver_trap() and demand_page(). After all those, we'll be ready to
* examine the Switcher, and our philosophical understanding of the Host/Guest
* duality will be complete. :*/
* duality will be complete.
:*/
static void adjust_pge(void *on)
{
if (on)
......@@ -439,13 +515,16 @@ static void adjust_pge(void *on)
write_cr4(read_cr4() & ~X86_CR4_PGE);
}
/*H:020 Now the Switcher is mapped and every thing else is ready, we need to do
* some more i386-specific initialization. */
/*H:020
* Now the Switcher is mapped and every thing else is ready, we need to do
* some more i386-specific initialization.
*/
void __init lguest_arch_host_init(void)
{
int i;
/* Most of the i386/switcher.S doesn't care that it's been moved; on
/*
* Most of the i386/switcher.S doesn't care that it's been moved; on
* Intel, jumps are relative, and it doesn't access any references to
* external code or data.
*
......@@ -453,7 +532,8 @@ void __init lguest_arch_host_init(void)
* addresses are placed in a table (default_idt_entries), so we need to
* update the table with the new addresses. switcher_offset() is a
* convenience function which returns the distance between the
* compiled-in switcher code and the high-mapped copy we just made. */
* compiled-in switcher code and the high-mapped copy we just made.
*/
for (i = 0; i < IDT_ENTRIES; i++)
default_idt_entries[i] += switcher_offset();
......@@ -468,63 +548,81 @@ void __init lguest_arch_host_init(void)
for_each_possible_cpu(i) {
/* lguest_pages() returns this CPU's two pages. */
struct lguest_pages *pages = lguest_pages(i);
/* This is a convenience pointer to make the code fit one
* statement to a line. */
/* This is a convenience pointer to make the code neater. */
struct lguest_ro_state *state = &pages->state;
/* The Global Descriptor Table: the Host has a different one
/*
* The Global Descriptor Table: the Host has a different one
* for each CPU. We keep a descriptor for the GDT which says
* where it is and how big it is (the size is actually the last
* byte, not the size, hence the "-1"). */
* byte, not the size, hence the "-1").
*/
state->host_gdt_desc.size = GDT_SIZE-1;
state->host_gdt_desc.address = (long)get_cpu_gdt_table(i);
/* All CPUs on the Host use the same Interrupt Descriptor
/*
* All CPUs on the Host use the same Interrupt Descriptor
* Table, so we just use store_idt(), which gets this CPU's IDT
* descriptor. */
* descriptor.
*/
store_idt(&state->host_idt_desc);
/* The descriptors for the Guest's GDT and IDT can be filled
/*
* The descriptors for the Guest's GDT and IDT can be filled
* out now, too. We copy the GDT & IDT into ->guest_gdt and
* ->guest_idt before actually running the Guest. */
* ->guest_idt before actually running the Guest.
*/
state->guest_idt_desc.size = sizeof(state->guest_idt)-1;
state->guest_idt_desc.address = (long)&state->guest_idt;
state->guest_gdt_desc.size = sizeof(state->guest_gdt)-1;
state->guest_gdt_desc.address = (long)&state->guest_gdt;
/* We know where we want the stack to be when the Guest enters
/*
* We know where we want the stack to be when the Guest enters
* the Switcher: in pages->regs. The stack grows upwards, so
* we start it at the end of that structure. */
* we start it at the end of that structure.
*/
state->guest_tss.sp0 = (long)(&pages->regs + 1);
/* And this is the GDT entry to use for the stack: we keep a
* couple of special LGUEST entries. */
/*
* And this is the GDT entry to use for the stack: we keep a
* couple of special LGUEST entries.
*/
state->guest_tss.ss0 = LGUEST_DS;
/* x86 can have a finegrained bitmap which indicates what I/O
/*
* x86 can have a finegrained bitmap which indicates what I/O
* ports the process can use. We set it to the end of our
* structure, meaning "none". */
* structure, meaning "none".
*/
state->guest_tss.io_bitmap_base = sizeof(state->guest_tss);
/* Some GDT entries are the same across all Guests, so we can
* set them up now. */
/*
* Some GDT entries are the same across all Guests, so we can
* set them up now.
*/
setup_default_gdt_entries(state);
/* Most IDT entries are the same for all Guests, too.*/
setup_default_idt_entries(state, default_idt_entries);
/* The Host needs to be able to use the LGUEST segments on this
* CPU, too, so put them in the Host GDT. */
/*
* The Host needs to be able to use the LGUEST segments on this
* CPU, too, so put them in the Host GDT.
*/
get_cpu_gdt_table(i)[GDT_ENTRY_LGUEST_CS] = FULL_EXEC_SEGMENT;
get_cpu_gdt_table(i)[GDT_ENTRY_LGUEST_DS] = FULL_SEGMENT;
}
/* In the Switcher, we want the %cs segment register to use the
/*
* In the Switcher, we want the %cs segment register to use the
* LGUEST_CS GDT entry: we've put that in the Host and Guest GDTs, so
* it will be undisturbed when we switch. To change %cs and jump we
* need this structure to feed to Intel's "lcall" instruction. */
* need this structure to feed to Intel's "lcall" instruction.
*/
lguest_entry.offset = (long)switch_to_guest + switcher_offset();
lguest_entry.segment = LGUEST_CS;
/* Finally, we need to turn off "Page Global Enable". PGE is an
/*
* Finally, we need to turn off "Page Global Enable". PGE is an
* optimization where page table entries are specially marked to show
* they never change. The Host kernel marks all the kernel pages this
* way because it's always present, even when userspace is running.
......@@ -534,16 +632,21 @@ void __init lguest_arch_host_init(void)
* you'll get really weird bugs that you'll chase for two days.
*
* I used to turn PGE off every time we switched to the Guest and back
* on when we return, but that slowed the Switcher down noticibly. */
* on when we return, but that slowed the Switcher down noticibly.
*/
/* We don't need the complexity of CPUs coming and going while we're
* doing this. */
/*
* We don't need the complexity of CPUs coming and going while we're
* doing this.
*/
get_online_cpus();
if (cpu_has_pge) { /* We have a broader idea of "global". */
/* Remember that this was originally set (for cleanup). */
cpu_had_pge = 1;
/* adjust_pge is a helper function which sets or unsets the PGE
* bit on its CPU, depending on the argument (0 == unset). */
/*
* adjust_pge is a helper function which sets or unsets the PGE
* bit on its CPU, depending on the argument (0 == unset).
*/
on_each_cpu(adjust_pge, (void *)0, 1);
/* Turn off the feature in the global feature set. */
clear_cpu_cap(&boot_cpu_data, X86_FEATURE_PGE);
......@@ -590,26 +693,32 @@ int lguest_arch_init_hypercalls(struct lg_cpu *cpu)
{
u32 tsc_speed;
/* The pointer to the Guest's "struct lguest_data" is the only argument.
* We check that address now. */
/*
* The pointer to the Guest's "struct lguest_data" is the only argument.
* We check that address now.
*/
if (!lguest_address_ok(cpu->lg, cpu->hcall->arg1,
sizeof(*cpu->lg->lguest_data)))
return -EFAULT;
/* Having checked it, we simply set lg->lguest_data to point straight
/*
* Having checked it, we simply set lg->lguest_data to point straight
* into the Launcher's memory at the right place and then use
* copy_to_user/from_user from now on, instead of lgread/write. I put
* this in to show that I'm not immune to writing stupid
* optimizations. */
* optimizations.
*/
cpu->lg->lguest_data = cpu->lg->mem_base + cpu->hcall->arg1;
/* We insist that the Time Stamp Counter exist and doesn't change with
/*
* We insist that the Time Stamp Counter exist and doesn't change with
* cpu frequency. Some devious chip manufacturers decided that TSC
* changes could be handled in software. I decided that time going
* backwards might be good for benchmarks, but it's bad for users.
*
* We also insist that the TSC be stable: the kernel detects unreliable
* TSCs for its own purposes, and we use that here. */
* TSCs for its own purposes, and we use that here.
*/
if (boot_cpu_has(X86_FEATURE_CONSTANT_TSC) && !check_tsc_unstable())
tsc_speed = tsc_khz;
else
......@@ -625,38 +734,47 @@ int lguest_arch_init_hypercalls(struct lg_cpu *cpu)
}
/*:*/
/*L:030 lguest_arch_setup_regs()
/*L:030
* lguest_arch_setup_regs()
*
* Most of the Guest's registers are left alone: we used get_zeroed_page() to
* allocate the structure, so they will be 0. */
* allocate the structure, so they will be 0.
*/
void lguest_arch_setup_regs(struct lg_cpu *cpu, unsigned long start)
{
struct lguest_regs *regs = cpu->regs;
/* There are four "segment" registers which the Guest needs to boot:
/*
* There are four "segment" registers which the Guest needs to boot:
* The "code segment" register (cs) refers to the kernel code segment
* __KERNEL_CS, and the "data", "extra" and "stack" segment registers
* refer to the kernel data segment __KERNEL_DS.
*
* The privilege level is packed into the lower bits. The Guest runs
* at privilege level 1 (GUEST_PL).*/
* at privilege level 1 (GUEST_PL).
*/
regs->ds = regs->es = regs->ss = __KERNEL_DS|GUEST_PL;
regs->cs = __KERNEL_CS|GUEST_PL;
/* The "eflags" register contains miscellaneous flags. Bit 1 (0x002)
/*
* The "eflags" register contains miscellaneous flags. Bit 1 (0x002)
* is supposed to always be "1". Bit 9 (0x200) controls whether
* interrupts are enabled. We always leave interrupts enabled while
* running the Guest. */
* running the Guest.
*/
regs->eflags = X86_EFLAGS_IF | 0x2;
/* The "Extended Instruction Pointer" register says where the Guest is
* running. */
/*
* The "Extended Instruction Pointer" register says where the Guest is
* running.
*/
regs->eip = start;
/* %esi points to our boot information, at physical address 0, so don't
* touch it. */
/*
* %esi points to our boot information, at physical address 0, so don't
* touch it.
*/
/* There are a couple of GDT entries the Guest expects when first
* booting. */
/* There are a couple of GDT entries the Guest expects at boot. */
setup_guest_gdt(cpu);
}
/*P:900 This is the Switcher: code which sits at 0xFFC00000 astride both the
* Host and Guest to do the low-level Guest<->Host switch. It is as simple as
* it can be made, but it's naturally very specific to x86.
/*P:900
* This is the Switcher: code which sits at 0xFFC00000 (or 0xFFE00000) astride
* both the Host and Guest to do the low-level Guest<->Host switch. It is as
* simple as it can be made, but it's naturally very specific to x86.
*
* You have now completed Preparation. If this has whet your appetite; if you
* are feeling invigorated and refreshed then the next, more challenging stage
* can be found in "make Guest". :*/
* can be found in "make Guest".
:*/
/*M:012 Lguest is meant to be simple: my rule of thumb is that 1% more LOC must
/*M:012
* Lguest is meant to be simple: my rule of thumb is that 1% more LOC must
* gain at least 1% more performance. Since neither LOC nor performance can be
* measured beforehand, it generally means implementing a feature then deciding
* if it's worth it. And once it's implemented, who can say no?
......@@ -31,11 +34,14 @@
* Host (which is actually really easy).
*
* Two questions remain. Would the performance gain outweigh the complexity?
* And who would write the verse documenting it? :*/
* And who would write the verse documenting it?
:*/
/*M:011 Lguest64 handles NMI. This gave me NMI envy (until I looked at their
/*M:011
* Lguest64 handles NMI. This gave me NMI envy (until I looked at their
* code). It's worth doing though, since it would let us use oprofile in the
* Host when a Guest is running. :*/
* Host when a Guest is running.
:*/
/*S:100
* Welcome to the Switcher itself!
......
......@@ -52,8 +52,10 @@ struct virtio_pci_device
char (*msix_names)[256];
/* Number of available vectors */
unsigned msix_vectors;
/* Vectors allocated */
/* Vectors allocated, excluding per-vq vectors if any */
unsigned msix_used_vectors;
/* Whether we have vector per vq */
bool per_vq_vectors;
};
/* Constants for MSI-X */
......@@ -258,7 +260,6 @@ static void vp_free_vectors(struct virtio_device *vdev)
for (i = 0; i < vp_dev->msix_used_vectors; ++i)
free_irq(vp_dev->msix_entries[i].vector, vp_dev);
vp_dev->msix_used_vectors = 0;
if (vp_dev->msix_enabled) {
/* Disable the vector used for configuration */
......@@ -267,80 +268,77 @@ static void vp_free_vectors(struct virtio_device *vdev)
/* Flush the write out to device */
ioread16(vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
vp_dev->msix_enabled = 0;
pci_disable_msix(vp_dev->pci_dev);
vp_dev->msix_enabled = 0;
vp_dev->msix_vectors = 0;
}
}
static int vp_enable_msix(struct pci_dev *dev, struct msix_entry *entries,
int *options, int noptions)
{
int i;
for (i = 0; i < noptions; ++i)
if (!pci_enable_msix(dev, entries, options[i]))
return options[i];
return -EBUSY;
vp_dev->msix_used_vectors = 0;
kfree(vp_dev->msix_names);
vp_dev->msix_names = NULL;
kfree(vp_dev->msix_entries);
vp_dev->msix_entries = NULL;
}
static int vp_request_vectors(struct virtio_device *vdev, unsigned max_vqs)
static int vp_request_vectors(struct virtio_device *vdev, int nvectors,
bool per_vq_vectors)
{
struct virtio_pci_device *vp_dev = to_vp_device(vdev);
const char *name = dev_name(&vp_dev->vdev.dev);
unsigned i, v;
int err = -ENOMEM;
/* We want at most one vector per queue and one for config changes.
* Fallback to separate vectors for config and a shared for queues.
* Finally fall back to regular interrupts. */
int options[] = { max_vqs + 1, 2 };
int nvectors = max(options[0], options[1]);
if (!nvectors) {
/* Can't allocate MSI-X vectors, use regular interrupt */
vp_dev->msix_vectors = 0;
err = request_irq(vp_dev->pci_dev->irq, vp_interrupt,
IRQF_SHARED, name, vp_dev);
if (err)
return err;
vp_dev->intx_enabled = 1;
return 0;
}
vp_dev->msix_entries = kmalloc(nvectors * sizeof *vp_dev->msix_entries,
GFP_KERNEL);
if (!vp_dev->msix_entries)
goto error_entries;
goto error;
vp_dev->msix_names = kmalloc(nvectors * sizeof *vp_dev->msix_names,
GFP_KERNEL);
if (!vp_dev->msix_names)
goto error_names;
goto error;
for (i = 0; i < nvectors; ++i)
vp_dev->msix_entries[i].entry = i;
err = vp_enable_msix(vp_dev->pci_dev, vp_dev->msix_entries,
options, ARRAY_SIZE(options));
if (err < 0) {
/* Can't allocate enough MSI-X vectors, use regular interrupt */
vp_dev->msix_vectors = 0;
err = request_irq(vp_dev->pci_dev->irq, vp_interrupt,
IRQF_SHARED, name, vp_dev);
if (err)
goto error_irq;
vp_dev->intx_enabled = 1;
} else {
vp_dev->msix_vectors = err;
vp_dev->msix_enabled = 1;
/* Set the vector used for configuration */
v = vp_dev->msix_used_vectors;
snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
"%s-config", name);
err = request_irq(vp_dev->msix_entries[v].vector,
vp_config_changed, 0, vp_dev->msix_names[v],
vp_dev);
if (err)
goto error_irq;
++vp_dev->msix_used_vectors;
iowrite16(v, vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
/* Verify we had enough resources to assign the vector */
v = ioread16(vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
if (v == VIRTIO_MSI_NO_VECTOR) {
err = -EBUSY;
goto error_irq;
}
err = pci_enable_msix(vp_dev->pci_dev, vp_dev->msix_entries, nvectors);
if (err > 0)
err = -ENOSPC;
if (err)
goto error;
vp_dev->msix_vectors = nvectors;
vp_dev->msix_enabled = 1;
/* Set the vector used for configuration */
v = vp_dev->msix_used_vectors;
snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
"%s-config", name);
err = request_irq(vp_dev->msix_entries[v].vector,
vp_config_changed, 0, vp_dev->msix_names[v],
vp_dev);
if (err)
goto error;
++vp_dev->msix_used_vectors;
iowrite16(v, vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
/* Verify we had enough resources to assign the vector */
v = ioread16(vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
if (v == VIRTIO_MSI_NO_VECTOR) {
err = -EBUSY;
goto error;
}
if (vp_dev->msix_vectors && vp_dev->msix_vectors != max_vqs + 1) {
if (!per_vq_vectors) {
/* Shared vector for all VQs */
v = vp_dev->msix_used_vectors;
snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
......@@ -349,28 +347,25 @@ static int vp_request_vectors(struct virtio_device *vdev, unsigned max_vqs)
vp_vring_interrupt, 0, vp_dev->msix_names[v],
vp_dev);
if (err)
goto error_irq;
goto error;
++vp_dev->msix_used_vectors;
}
return 0;
error_irq:
error:
vp_free_vectors(vdev);
kfree(vp_dev->msix_names);
error_names:
kfree(vp_dev->msix_entries);
error_entries:
return err;
}
static struct virtqueue *vp_find_vq(struct virtio_device *vdev, unsigned index,
void (*callback)(struct virtqueue *vq),
const char *name)
const char *name,
u16 vector)
{
struct virtio_pci_device *vp_dev = to_vp_device(vdev);
struct virtio_pci_vq_info *info;
struct virtqueue *vq;
unsigned long flags, size;
u16 num, vector;
u16 num;
int err;
/* Select the queue we're interested in */
......@@ -389,7 +384,7 @@ static struct virtqueue *vp_find_vq(struct virtio_device *vdev, unsigned index,
info->queue_index = index;
info->num = num;
info->vector = VIRTIO_MSI_NO_VECTOR;
info->vector = vector;
size = PAGE_ALIGN(vring_size(num, VIRTIO_PCI_VRING_ALIGN));
info->queue = alloc_pages_exact(size, GFP_KERNEL|__GFP_ZERO);
......@@ -413,22 +408,7 @@ static struct virtqueue *vp_find_vq(struct virtio_device *vdev, unsigned index,
vq->priv = info;
info->vq = vq;
/* allocate per-vq vector if available and necessary */
if (callback && vp_dev->msix_used_vectors < vp_dev->msix_vectors) {
vector = vp_dev->msix_used_vectors;
snprintf(vp_dev->msix_names[vector], sizeof *vp_dev->msix_names,
"%s-%s", dev_name(&vp_dev->vdev.dev), name);
err = request_irq(vp_dev->msix_entries[vector].vector,
vring_interrupt, 0,
vp_dev->msix_names[vector], vq);
if (err)
goto out_request_irq;
info->vector = vector;
++vp_dev->msix_used_vectors;
} else
vector = VP_MSIX_VQ_VECTOR;
if (callback && vp_dev->msix_enabled) {
if (vector != VIRTIO_MSI_NO_VECTOR) {
iowrite16(vector, vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR);
vector = ioread16(vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR);
if (vector == VIRTIO_MSI_NO_VECTOR) {
......@@ -444,11 +424,6 @@ static struct virtqueue *vp_find_vq(struct virtio_device *vdev, unsigned index,
return vq;
out_assign:
if (info->vector != VIRTIO_MSI_NO_VECTOR) {
free_irq(vp_dev->msix_entries[info->vector].vector, vq);
--vp_dev->msix_used_vectors;
}
out_request_irq:
vring_del_virtqueue(vq);
out_activate_queue:
iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
......@@ -462,12 +437,13 @@ static void vp_del_vq(struct virtqueue *vq)
{
struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
struct virtio_pci_vq_info *info = vq->priv;
unsigned long size;
unsigned long flags, size;
iowrite16(info->queue_index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL);
spin_lock_irqsave(&vp_dev->lock, flags);
list_del(&info->node);
spin_unlock_irqrestore(&vp_dev->lock, flags);
if (info->vector != VIRTIO_MSI_NO_VECTOR)
free_irq(vp_dev->msix_entries[info->vector].vector, vq);
iowrite16(info->queue_index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL);
if (vp_dev->msix_enabled) {
iowrite16(VIRTIO_MSI_NO_VECTOR,
......@@ -489,36 +465,62 @@ static void vp_del_vq(struct virtqueue *vq)
/* the config->del_vqs() implementation */
static void vp_del_vqs(struct virtio_device *vdev)
{
struct virtio_pci_device *vp_dev = to_vp_device(vdev);
struct virtqueue *vq, *n;
struct virtio_pci_vq_info *info;
list_for_each_entry_safe(vq, n, &vdev->vqs, list)
list_for_each_entry_safe(vq, n, &vdev->vqs, list) {
info = vq->priv;
if (vp_dev->per_vq_vectors)
free_irq(vp_dev->msix_entries[info->vector].vector, vq);
vp_del_vq(vq);
}
vp_dev->per_vq_vectors = false;
vp_free_vectors(vdev);
}
/* the config->find_vqs() implementation */
static int vp_find_vqs(struct virtio_device *vdev, unsigned nvqs,
struct virtqueue *vqs[],
vq_callback_t *callbacks[],
const char *names[])
static int vp_try_to_find_vqs(struct virtio_device *vdev, unsigned nvqs,
struct virtqueue *vqs[],
vq_callback_t *callbacks[],
const char *names[],
int nvectors,
bool per_vq_vectors)
{
int vectors = 0;
int i, err;
/* How many vectors would we like? */
for (i = 0; i < nvqs; ++i)
if (callbacks[i])
++vectors;
struct virtio_pci_device *vp_dev = to_vp_device(vdev);
u16 vector;
int i, err, allocated_vectors;
err = vp_request_vectors(vdev, vectors);
err = vp_request_vectors(vdev, nvectors, per_vq_vectors);
if (err)
goto error_request;
vp_dev->per_vq_vectors = per_vq_vectors;
allocated_vectors = vp_dev->msix_used_vectors;
for (i = 0; i < nvqs; ++i) {
vqs[i] = vp_find_vq(vdev, i, callbacks[i], names[i]);
if (IS_ERR(vqs[i]))
if (!callbacks[i] || !vp_dev->msix_enabled)
vector = VIRTIO_MSI_NO_VECTOR;
else if (vp_dev->per_vq_vectors)
vector = allocated_vectors++;
else
vector = VP_MSIX_VQ_VECTOR;
vqs[i] = vp_find_vq(vdev, i, callbacks[i], names[i], vector);
if (IS_ERR(vqs[i])) {
err = PTR_ERR(vqs[i]);
goto error_find;
}
/* allocate per-vq irq if available and necessary */
if (vp_dev->per_vq_vectors && vector != VIRTIO_MSI_NO_VECTOR) {
snprintf(vp_dev->msix_names[vector], sizeof *vp_dev->msix_names,
"%s-%s", dev_name(&vp_dev->vdev.dev), names[i]);
err = request_irq(vp_dev->msix_entries[vector].vector,
vring_interrupt, 0,
vp_dev->msix_names[vector], vqs[i]);
if (err) {
vp_del_vq(vqs[i]);
goto error_find;
}
}
}
return 0;
......@@ -526,7 +528,37 @@ error_find:
vp_del_vqs(vdev);
error_request:
return PTR_ERR(vqs[i]);
return err;
}
/* the config->find_vqs() implementation */
static int vp_find_vqs(struct virtio_device *vdev, unsigned nvqs,
struct virtqueue *vqs[],
vq_callback_t *callbacks[],
const char *names[])
{
int vectors = 0;
int i, uninitialized_var(err);
/* How many vectors would we like? */
for (i = 0; i < nvqs; ++i)
if (callbacks[i])
++vectors;
/* We want at most one vector per queue and one for config changes. */
err = vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names,
vectors + 1, true);
if (!err)
return 0;
/* Fallback to separate vectors for config and a shared for queues. */
err = vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names,
2, false);
if (!err)
return 0;
/* Finally fall back to regular interrupts. */
err = vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names,
0, false);
return err;
}
static struct virtio_config_ops virtio_pci_config_ops = {
......
/* Things the lguest guest needs to know. Note: like all lguest interfaces,
* this is subject to wild and random change between versions. */
/*
* Things the lguest guest needs to know. Note: like all lguest interfaces,
* this is subject to wild and random change between versions.
*/
#ifndef _LINUX_LGUEST_H
#define _LINUX_LGUEST_H
......@@ -11,32 +13,41 @@
#define LG_CLOCK_MIN_DELTA 100UL
#define LG_CLOCK_MAX_DELTA ULONG_MAX
/*G:031 The second method of communicating with the Host is to via "struct
/*G:031
* The second method of communicating with the Host is to via "struct
* lguest_data". Once the Guest's initialization hypercall tells the Host where
* this is, the Guest and Host both publish information in it. :*/
struct lguest_data
{
/* 512 == enabled (same as eflags in normal hardware). The Guest
* changes interrupts so often that a hypercall is too slow. */
* this is, the Guest and Host both publish information in it.
:*/
struct lguest_data {
/*
* 512 == enabled (same as eflags in normal hardware). The Guest
* changes interrupts so often that a hypercall is too slow.
*/
unsigned int irq_enabled;
/* Fine-grained interrupt disabling by the Guest */
DECLARE_BITMAP(blocked_interrupts, LGUEST_IRQS);
/* The Host writes the virtual address of the last page fault here,
/*
* The Host writes the virtual address of the last page fault here,
* which saves the Guest a hypercall. CR2 is the native register where
* this address would normally be found. */
* this address would normally be found.
*/
unsigned long cr2;
/* Wallclock time set by the Host. */
struct timespec time;
/* Interrupt pending set by the Host. The Guest should do a hypercall
* if it re-enables interrupts and sees this set (to X86_EFLAGS_IF). */
/*
* Interrupt pending set by the Host. The Guest should do a hypercall
* if it re-enables interrupts and sees this set (to X86_EFLAGS_IF).
*/
int irq_pending;
/* Async hypercall ring. Instead of directly making hypercalls, we can
/*
* Async hypercall ring. Instead of directly making hypercalls, we can
* place them in here for processing the next time the Host wants.
* This batching can be quite efficient. */
* This batching can be quite efficient.
*/
/* 0xFF == done (set by Host), 0 == pending (set by Guest). */
u8 hcall_status[LHCALL_RING_SIZE];
......
......@@ -29,8 +29,10 @@ struct lguest_device_desc {
__u8 type;
/* The number of virtqueues (first in config array) */
__u8 num_vq;
/* The number of bytes of feature bits. Multiply by 2: one for host
* features and one for Guest acknowledgements. */
/*
* The number of bytes of feature bits. Multiply by 2: one for host
* features and one for Guest acknowledgements.
*/
__u8 feature_len;
/* The number of bytes of the config array after virtqueues. */
__u8 config_len;
......@@ -39,8 +41,10 @@ struct lguest_device_desc {
__u8 config[0];
};
/*D:135 This is how we expect the device configuration field for a virtqueue
* to be laid out in config space. */
/*D:135
* This is how we expect the device configuration field for a virtqueue
* to be laid out in config space.
*/
struct lguest_vqconfig {
/* The number of entries in the virtio_ring */
__u16 num;
......@@ -61,7 +65,9 @@ enum lguest_req
LHREQ_EVENTFD, /* + address, fd. */
};
/* The alignment to use between consumer and producer parts of vring.
* x86 pagesize for historical reasons. */
/*
* The alignment to use between consumer and producer parts of vring.
* x86 pagesize for historical reasons.
*/
#define LGUEST_VRING_ALIGN 4096
#endif /* _LINUX_LGUEST_LAUNCHER */
......@@ -20,8 +20,7 @@
#define VIRTIO_BLK_ID_BYTES (sizeof(__u16[256])) /* IDENTIFY DATA */
struct virtio_blk_config
{
struct virtio_blk_config {
/* The capacity (in 512-byte sectors). */
__u64 capacity;
/* The maximum segment size (if VIRTIO_BLK_F_SIZE_MAX) */
......@@ -50,8 +49,7 @@ struct virtio_blk_config
#define VIRTIO_BLK_T_BARRIER 0x80000000
/* This is the first element of the read scatter-gather list. */
struct virtio_blk_outhdr
{
struct virtio_blk_outhdr {
/* VIRTIO_BLK_T* */
__u32 type;
/* io priority. */
......
......@@ -79,8 +79,7 @@
* the dev->feature bits if it wants.
*/
typedef void vq_callback_t(struct virtqueue *);
struct virtio_config_ops
{
struct virtio_config_ops {
void (*get)(struct virtio_device *vdev, unsigned offset,
void *buf, unsigned len);
void (*set)(struct virtio_device *vdev, unsigned offset,
......
......@@ -31,8 +31,7 @@
#define VIRTIO_NET_S_LINK_UP 1 /* Link is up */
struct virtio_net_config
{
struct virtio_net_config {
/* The config defining mac address (if VIRTIO_NET_F_MAC) */
__u8 mac[6];
/* See VIRTIO_NET_F_STATUS and VIRTIO_NET_S_* above */
......@@ -41,8 +40,7 @@ struct virtio_net_config
/* This is the first element of the scatter-gather list. If you don't
* specify GSO or CSUM features, you can simply ignore the header. */
struct virtio_net_hdr
{
struct virtio_net_hdr {
#define VIRTIO_NET_HDR_F_NEEDS_CSUM 1 // Use csum_start, csum_offset
__u8 flags;
#define VIRTIO_NET_HDR_GSO_NONE 0 // Not a GSO frame
......
......@@ -30,8 +30,7 @@
#define VIRTIO_RING_F_INDIRECT_DESC 28
/* Virtio ring descriptors: 16 bytes. These can chain together via "next". */
struct vring_desc
{
struct vring_desc {
/* Address (guest-physical). */
__u64 addr;
/* Length. */
......@@ -42,24 +41,21 @@ struct vring_desc
__u16 next;
};
struct vring_avail
{
struct vring_avail {
__u16 flags;
__u16 idx;
__u16 ring[];
};
/* u32 is used here for ids for padding reasons. */
struct vring_used_elem
{
struct vring_used_elem {
/* Index of start of used descriptor chain. */
__u32 id;
/* Total length of the descriptor chain which was used (written to) */
__u32 len;
};
struct vring_used
{
struct vring_used {
__u16 flags;
__u16 idx;
struct vring_used_elem ring[];
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment