- 07 May, 2007 40 commits
-
-
Jeff Dike authored
When doing a full address space flush, only look at areas covered by a VMA. Signed-off-by:
Jeff Dike <jdike@linux.intel.com> Cc: Paolo 'Blaisorblade' Giarrusso <blaisorblade@yahoo.it> Signed-off-by:
Andrew Morton <akpm@linux-foundation.org> Signed-off-by:
Linus Torvalds <torvalds@linux-foundation.org>
-
Jeff Dike authored
More trimming of the page fault path. Permissions are passed around in a single int rather than one bit per int. The permission values are copied from libc so that they can be passed to mmap and mprotect without any further conversion. The register sets used by do_syscall_stub and copy_context_skas0 are initialized once, at boot time, rather than once per call. wait_stub_done checks whether it is getting the signals it expects by comparing the wait status to a mask containing bits for the signals of interest rather than comparing individually to the signal numbers. It also has one check for a wait failure instead of two. The caller is expected to do the initial continue of the stub. This gets rid of an argument and some logic. The fname argument is gone, as that can be had from a stack trace. user_signal() is collapsed into userspace() as it is basically one or two lines of code afterwards. The physical memory remapping stuff is gone, as it is unused. flush_tlb_page is inlined. Signed-off-by:
-
Jeff Dike authored
I missed removing another piece of debugging in an earlier patch. Signed-off-by:
-
Jeff Dike authored
Give the page fault code a specialized path. There is only one page to look at, so there's no point in going into the general page table walking code. There's only going to be one host operation, so there are no opportunities for merging. So, we go straight to the pte we want, figure out what needs doing, and do it. While I was in here, I fixed the wart where the address passed to unmap was a void *, but an unsigned long to map and protect. This gives me just under 10% on a kernel build. Signed-off-by:
-
Jeff Dike authored
Allow deadlocks to be avoided in the AIO code by setting the pipe to the I/O thread non-blocking. Signed-off-by:
-
Jeff Dike authored
Rename os_{read_write}_file_k back to os_{read_write}_file, delete the originals and their bogus infrastructure, and fix all the callers. Signed-off-by: -
Jeff Dike authored
I accidentally left the remnants of some debugging in an earlier patch. Signed-off-by:
-
Jeff Dike authored
Formatting fixes ahead of renaming os_{read_write}_file_k to os_{read_write}_file and fixing all the callers. Signed-off-by: -
Jeff Dike authored
Convert all remaining os_{read_write}_file users to use the simple {read,write} wrappers, os_{read_write}_file_k. Signed-off-by: -
Jeff Dike authored
Code running on the initial UML stack can't receive or process signals since current must be valid when IRQs are handled, and there is no current for this stack. So, instead of using UML_LONGJMP and UML_SETJMP, which are careful to save and restore signal state, and, as a side-effect, handle any deferred signals, start_idle_thread must use the bare equivalents, which don't do anything with signals. Signed-off-by:
-
Jeff Dike authored
Dump core after a panic. This will provide better debugging information than is currently available. Signed-off-by:
-
Peter Zijlstra authored
Sanitise gfp flags; it actually is an atomic context, so drop the GFP_KERNEL part. Signed-off-by:
Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by:
Jeff Dike <jdike@addtoit.com> Signed-off-by:
-
Jeff Dike authored
Instead of writing entire structures between UML and the I/O thread, we send pointers. This cuts down on the amount of data being copied and possibly allows more requests to be pending between the two. This requires that the requests be kmalloced and freed instead of living on the stack. Signed-off-by:
-
Jeff Dike authored
Send as many I/O requests to the I/O thread as possible, even though it will still only handle one at a time. This provides an opportunity to reduce latency by starting one request before the previous one has been finished in the driver. Request handling is somewhat modernized by requesting sg pieces of a request and handling them separately, finishing off the entire request after all the pieces are done. When a request queue stalls, normally because its pipe to the I/O thread is full, it is put on the restart list. This list is processed by starting up the queues on it whenever there is some indication that progress might be possible again. Currently, this happens in the driver interrupt routine. Some requests have been finished, so there is likely to be room in the pipe again. This almost doubles throughput when copying data between devices, but made no noticable difference on anything else I tried. Signed-off-by:
-
Jeff Dike authored
This patch converts calls in the os layer to os_{read,write}_file to calls directly to libc read() and write() where it is clear that the I/O buffer is in the kernel. We can do that here instead of calling os_{read,write}_file_k since we are in libc code and can call libc directly. With the change in the calls, error handling needs to be changed to refer to errno directly rather than the return value of the call. CATCH_EINTR wrappers were also added where needed. Signed-off-by: -
Jeff Dike authored
This patch lays some groundwork for the next one, which converts calls to os_{read,write}_file into {read,write}, by doing some tidying in the affected areas. do_not_aio gets restructured to make the final result a bit cleaner. There are also whitespace and other formatting fixes, fixes in error messages, and a typo fix. Signed-off-by: -
Jeff Dike authored
This patch starts the removal of a very old, very broken piece of code. This stems from the problem of passing a userspace buffer into read() or write() on the host. If that buffer had not yet been faulted in, read and write will return -EFAULT. To avoid this problem, the solution was to fault the buffer in before the system call by touching the pages that hold the buffer by doing a copy-user of a byte to each page. This is obviously bogus, but it does usually work, in tt mode, since the kernel and process are in the same address space and userspace addresses can be accessed directly in the kernel. In skas mode, where the kernel and process are in separate address spaces, it is completely bogus because the userspace address, which is invalid in the kernel, is passed into the system call instead of the corresponding physical address, which would be valid. Here, it appears that this code, on every host read() or write(), tries to fault in a random process page. This doesn't seem to cause any correctness problems, but there is a performance impact. This patch, and the ones following, result in a 10-15% performance gain on a kernel build. This code can't be immediately tossed out because when it is, you can't log in. Apparently, there is some code in the console driver which depends on this somehow. However, we can start removing it by switching the code which does I/O using kernel addresses to using plain read() and write(). This patch introduces os_read_file_k and os_write_file_k for use with kernel buffers and converts all call locations which use obvious kernel buffers to use them. These include I/O using buffers which are local variables which are on the stack or kmalloc-ed. Later patches will handle the less obvious cases, followed by a mass conversion back to the original interface. Signed-off-by:
-
Jeff Dike authored
It turns out that essentially none of the x86_64 bugs.c is needed. So, we can delete most of it. Signed-off-by:
-
Jeff Dike authored
The previous page table walking code was horribly inefficient. This patch replaces it with code taken from elsewhere in the kernel. Forking from bash is now ~5% faster and page faults are handled ~10% faster. Signed-off-by:
-
Jeff Dike authored
Provide a register dump if handle_trap fails. Abstract out ptrace_dump_regs since it now has two callers. Signed-off-by:
-
Jeff Dike authored
Define release methods for the ubd and net drivers. They contain as much of the remove methods as make sense. All error checking must have already been done as well as anything else that might be holding a reference on the device kobject. Signed-off-by:
-
Jeff Dike authored
HOST_FRAME_SIZE isn't used any more. It has been replaced with MAX_REG_NR. Signed-off-by:
-
Jeff Dike authored
Locking commentary. Signed-off-by:
-
Jeff Dike authored
Commentary about missing locking. Also got rid of uml_start because it was pointless. Signed-off-by:
-
Jeff Dike authored
If there's a segfault inside the kernel, we want a dump of the registers at the point of the segfault, not the registers at the point of calling panic or the last userspace registers. sig_handler_common_skas now uses a static register set in the case of a SIGSEGV to avoid messing up the process registers if the segfault turns out to be non-fatal. The architecture sigcontext-to-pt_regs copying code was repurposed to copy data out of the SEGV stack frame. Signed-off-by:
-
Jeff Dike authored
Tidying in preparation for the segfault register dumping patch which follows. void * pointers are changed to union uml_pt_regs *. This makes the types match reality, except in arch_fixup, which is changed to operate on a union uml_pt_regs. This fixes a bug in the call from segv_handler, which passes a union uml_pt_regs, to segv, which expects to pass a struct sigcontext to arch_fixup. Whitespace and other style fixes. There's also a errno printk fix. Signed-off-by:
-
Jeff Dike authored
kernel_thread() should just return an error value on do_fork failure, not panic. Signed-off-by:
-
Jeff Dike authored
userspace code used to have to call the kernelspace function page_size() in order to determine the value of the kernel's PAGE_SIZE. Since this is now available directly from kern_constants.h as UM_KERN_PAGE_SIZE, page_size() can be deleted and calls changed to use the constant. Signed-off-by:
-
Jeff Dike authored
Clean up arch/um/kernel/process.c: - lots of return(x); -> return x; conversions - a number of the small functions are either unused, in which case they are gone, along any declarations in a header, or could be made static. - current_pid is ifdefed on CONFIG_MODE_TT and its declaration is ifdefed on both CONFIG_MODE_TT and UML_CONFIG_MODE_TT because we don't know whether it's being used in a userspace or kernel file. Signed-off-by:
-
Jeff Dike authored
Comment the lack of locking on a couple of globals. Also fix the formatting of __setup_host_supports_tls. Signed-off-by:
-
Jeff Dike authored
flush_thread doesn't need to do a full page table walk in order to clear the address space. It knows what the end result needs to be, so it can call unmap directly. This results in a 10-20% speedup in an exec from bash. Signed-off-by:
-
Davide Brini authored
Calls lines_init() *after* xterm_title is modified to include umid. Signed-off-by:
Davide Brini <davide.brini@unibo.it> Signed-off-by:
Paolo 'Blaisorblade' Giarrusso <blaisorblade@yahoo.it> Signed-off-by:
-
Paolo 'Blaisorblade' Giarrusso authored
To look at users I did: $ find arch/um/ include/asm-um -name '*.[ch]'|xargs grep -r 'net_kern\.h' +-l|xargs grep '\<user\>' Most users just cast user to the appropriate pointer, the remaining ones are fixed here. In net_kern.c, I'm almost sure that save trick is not needed anymore, but I've not verified it. Signed-off-by:
-
Paolo 'Blaisorblade' Giarrusso authored
Avoid using the temporary buffer introduced by previous patch to hold the device name. Btw, avoid leaking device on an error path. Other error paths may need cleanup. Signed-off-by:
-
Paolo 'Blaisorblade' Giarrusso authored
Improve checking and diagnostics for broadcast and multicast Ethernet MAC addresses, and distinguish between those cases in output; also make sure the device is assigned a MAC address valid only locally to avoid collisions. Signed-off-by:
-
Robert P. J. Day authored
Signed-off-by:
Robert P. J. Day <rpjday@mindspring.com> Acked-by:
-
Jeff Dike authored
The build started finding calls from non-init to init functions. These are just cases of init functions not being properly marked, so this patch fixes that. Signed-off-by:
-
Jeff Dike authored
user_util.h isn't needed any more, so delete it and remove all includes of it. Signed-off-by:
-
Jeff Dike authored
Rescue the useful contents of the soon-to-be-gone user-util.h. pty.c now gets ptsname from stdlib.h like it should have always done. CATCH_EINTR is now in os.h, although perhaps all usage should be under os-Linux at some point. get_pty is also in os.h. This patch restores the old definition of ARRAY_SIZE in user.h. This file is included only in userspace files, so there will be no conflict with the kernel's new ARRAY_SIZE. The copy of the kernel's ARRAY_SIZE and associated infrastructure is now gone. Signed-off-by:
-
Jeff Dike authored
This patch moves all the the symbols defined in um_arch.c, which are mostly boundaries between different parts of the UML kernel address space, to a new header, as-layout.h. There are also a few things here which aren't really related to address space layout, but which don't really have a better place to go. Signed-off-by:
-
