[PATCH] i386: Use %gs as the PDA base-segment in the kernel

This patch is the meat of the PDA change.  This patch makes several related
changes:

1: Most significantly, %gs is now used in the kernel.  This means that on
   entry, the old value of %gs is saved away, and it is reloaded with
   __KERNEL_PDA.

2: entry.S constructs the stack in the shape of struct pt_regs, and this
   is passed around the kernel so that the process's saved register
   state can be accessed.

   Unfortunately struct pt_regs doesn't currently have space for %gs
   (or %fs). This patch extends pt_regs to add space for gs (no space
   is allocated for %fs, since it won't be used, and it would just
   complicate the code in entry.S to work around the space).

3: Because %gs is now saved on the stack like %ds, %es and the integer
   registers, there are a number of places where it no longer needs to
   be handled specially; namely context switch, and saving/restoring the
   register state in a signal context.

4: And since kernel threads run in kernel space and call normal kernel
   code, they need to be created with their %gs == __KERNEL_PDA.
Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Chuck Ebbert <76306.1226@compuserve.com>
Cc: Zachary Amsden <zach@vmware.com>
Cc: Jan Beulich <jbeulich@novell.com>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>

arch/i386/kernel/asm-offsets.c

@@ -72,6 +72,7 @@ void foo(void)
 	OFFSET(PT_EAX, pt_regs, eax);
 	OFFSET(PT_DS,  pt_regs, xds);
 	OFFSET(PT_ES,  pt_regs, xes);
+	OFFSET(PT_GS,  pt_regs, xgs);
 	OFFSET(PT_ORIG_EAX, pt_regs, orig_eax);
 	OFFSET(PT_EIP, pt_regs, eip);
 	OFFSET(PT_CS,  pt_regs, xcs);

arch/i386/kernel/cpu/common.c

@@ -593,6 +593,14 @@ void __init early_cpu_init(void)
 #endif
 }
 
+/* Make sure %gs is initialized properly in idle threads */
+struct pt_regs * __devinit idle_regs(struct pt_regs *regs)
+{
+	memset(regs, 0, sizeof(struct pt_regs));
+	regs->xgs = __KERNEL_PDA;
+	return regs;
+}
+
 __cpuinit int alloc_gdt(int cpu)
 {
 	struct Xgt_desc_struct *cpu_gdt_descr = &per_cpu(cpu_gdt_descr, cpu);
@@ -644,6 +652,14 @@ struct i386_pda boot_pda = {
 	._pda = &boot_pda,
 };
 
+static inline void set_kernel_gs(void)
+{
+	/* Set %gs for this CPU's PDA.  Memory clobber is to create a
+	   barrier with respect to any PDA operations, so the compiler
+	   doesn't move any before here. */
+	asm volatile ("mov %0, %%gs" : : "r" (__KERNEL_PDA) : "memory");
+}
+
 /* Initialize the CPU's GDT and PDA.  The boot CPU does this for
    itself, but secondaries find this done for them. */
 __cpuinit int init_gdt(int cpu, struct task_struct *idle)
@@ -693,6 +709,7 @@ static void __cpuinit _cpu_init(int cpu, struct task_struct *curr)
 	   the boot CPU, this will transition from the boot gdt+pda to
 	   the real ones). */
 	load_gdt(cpu_gdt_descr);
+	set_kernel_gs();
 
 	if (cpu_test_and_set(cpu, cpu_initialized)) {
 		printk(KERN_WARNING "CPU#%d already initialized!\n", cpu);
@@ -731,8 +748,8 @@ static void __cpuinit _cpu_init(int cpu, struct task_struct *curr)
 	__set_tss_desc(cpu, GDT_ENTRY_DOUBLEFAULT_TSS, &doublefault_tss);
 #endif
 
-	/* Clear %fs and %gs. */
-	asm volatile ("movl %0, %%fs; movl %0, %%gs" : : "r" (0));
+	/* Clear %fs. */
+	asm volatile ("mov %0, %%fs" : : "r" (0));
 
 	/* Clear all 6 debug registers: */
 	set_debugreg(0, 0);

arch/i386/kernel/entry.S

@@ -30,12 +30,13 @@
  *	18(%esp) - %eax
  *	1C(%esp) - %ds
  *	20(%esp) - %es
- *	24(%esp) - orig_eax
- *	28(%esp) - %eip
- *	2C(%esp) - %cs
- *	30(%esp) - %eflags
- *	34(%esp) - %oldesp
- *	38(%esp) - %oldss
+ *	24(%esp) - %gs
+ *	28(%esp) - orig_eax
+ *	2C(%esp) - %eip
+ *	30(%esp) - %cs
+ *	34(%esp) - %eflags
+ *	38(%esp) - %oldesp
+ *	3C(%esp) - %oldss
  *
  * "current" is in register %ebx during any slow entries.
  */
@@ -92,6 +93,9 @@ VM_MASK		= 0x00020000
 
 #define SAVE_ALL \
 	cld; \
+	pushl %gs; \
+	CFI_ADJUST_CFA_OFFSET 4;\
+	/*CFI_REL_OFFSET gs, 0;*/\
 	pushl %es; \
 	CFI_ADJUST_CFA_OFFSET 4;\
 	/*CFI_REL_OFFSET es, 0;*/\
@@ -121,7 +125,9 @@ VM_MASK		= 0x00020000
 	CFI_REL_OFFSET ebx, 0;\
 	movl $(__USER_DS), %edx; \
 	movl %edx, %ds; \
-	movl %edx, %es;
+	movl %edx, %es; \
+	movl $(__KERNEL_PDA), %edx; \
+	movl %edx, %gs
 
 #define RESTORE_INT_REGS \
 	popl %ebx;	\
@@ -154,17 +160,22 @@ VM_MASK		= 0x00020000
 2:	popl %es;	\
 	CFI_ADJUST_CFA_OFFSET -4;\
 	/*CFI_RESTORE es;*/\
-.section .fixup,"ax";	\
-3:	movl $0,(%esp);	\
-	jmp 1b;		\
+3:	popl %gs;	\
+	CFI_ADJUST_CFA_OFFSET -4;\
+	/*CFI_RESTORE gs;*/\
+.pushsection .fixup,"ax";	\
 4:	movl $0,(%esp);	\
+	jmp 1b;		\
+5:	movl $0,(%esp);	\
 	jmp 2b;		\
-.previous;		\
+6:	movl $0,(%esp);	\
+	jmp 3b;		\
 .section __ex_table,"a";\
 	.align 4;	\
-	.long 1b,3b;	\
-	.long 2b,4b;	\
-.previous
+	.long 1b,4b;	\
+	.long 2b,5b;	\
+	.long 3b,6b;	\
+.popsection
 
 #define RING0_INT_FRAME \
 	CFI_STARTPROC simple;\
@@ -231,6 +242,7 @@ check_userspace:
 	andl $(VM_MASK | SEGMENT_RPL_MASK), %eax
 	cmpl $USER_RPL, %eax
 	jb resume_kernel		# not returning to v8086 or userspace
+
 ENTRY(resume_userspace)
  	DISABLE_INTERRUPTS		# make sure we don't miss an interrupt
 					# setting need_resched or sigpending
@@ -327,9 +339,16 @@ sysenter_past_esp:
 	movl PT_OLDESP(%esp), %ecx
 	xorl %ebp,%ebp
 	TRACE_IRQS_ON
+1:	mov  PT_GS(%esp), %gs
 	ENABLE_INTERRUPTS_SYSEXIT
 	CFI_ENDPROC
-
+.pushsection .fixup,"ax"
+2:	movl $0,PT_GS(%esp)
+	jmp 1b
+.section __ex_table,"a"
+	.align 4
+	.long 1b,2b
+.popsection
 
 	# system call handler stub
 ENTRY(system_call)
@@ -375,7 +394,7 @@ restore_nocheck:
 	TRACE_IRQS_IRET
 restore_nocheck_notrace:
 	RESTORE_REGS
-	addl $4, %esp
+	addl $4, %esp			# skip orig_eax/error_code
 	CFI_ADJUST_CFA_OFFSET -4
 1:	INTERRUPT_RETURN
 .section .fixup,"ax"
@@ -588,6 +607,10 @@ KPROBE_ENTRY(page_fault)
 	CFI_ADJUST_CFA_OFFSET 4
 	ALIGN
 error_code:
+	/* the function address is in %gs's slot on the stack */
+	pushl %es
+	CFI_ADJUST_CFA_OFFSET 4
+	/*CFI_REL_OFFSET es, 0*/
 	pushl %ds
 	CFI_ADJUST_CFA_OFFSET 4
 	/*CFI_REL_OFFSET ds, 0*/
@@ -613,18 +636,20 @@ error_code:
 	CFI_ADJUST_CFA_OFFSET 4
 	CFI_REL_OFFSET ebx, 0
 	cld
-	pushl %es
+	pushl %gs
 	CFI_ADJUST_CFA_OFFSET 4
-	/*CFI_REL_OFFSET es, 0*/
+	/*CFI_REL_OFFSET gs, 0*/
+	movl $(__KERNEL_PDA), %ecx
+	movl %ecx, %gs
 	UNWIND_ESPFIX_STACK
 	popl %ecx
 	CFI_ADJUST_CFA_OFFSET -4
 	/*CFI_REGISTER es, ecx*/
-	movl PT_ES(%esp), %edi		# get the function address
+	movl PT_GS(%esp), %edi		# get the function address
 	movl PT_ORIG_EAX(%esp), %edx	# get the error code
-	movl $-1, PT_ORIG_EAX(%esp)
-	movl %ecx, PT_ES(%esp)
-	/*CFI_REL_OFFSET es, ES*/
+	movl $-1, PT_ORIG_EAX(%esp)	# no syscall to restart
+	mov  %ecx, PT_GS(%esp)
+	/*CFI_REL_OFFSET gs, ES*/
 	movl $(__USER_DS), %ecx
 	movl %ecx, %ds
 	movl %ecx, %es
@@ -936,6 +961,7 @@ ENTRY(arch_unwind_init_running)
 	movl	%ebx, PT_EAX(%edx)
 	movl	$__USER_DS, PT_DS(%edx)
 	movl	$__USER_DS, PT_ES(%edx)
+	movl	$0, PT_GS(%edx)
 	movl	%ebx, PT_ORIG_EAX(%edx)
 	movl	%ecx, PT_EIP(%edx)
 	movl	12(%esp), %ecx

arch/i386/kernel/head.S

@@ -302,6 +302,7 @@ is386:	movl $2,%ecx		# set MP
 	movl %eax,%cr0
 
 	call check_x87
+	call setup_pda
 	lgdt cpu_gdt_descr
 	lidt idt_descr
 	ljmp $(__KERNEL_CS),$1f
@@ -312,10 +313,13 @@ is386:	movl $2,%ecx		# set MP
 	movl %eax,%ds
 	movl %eax,%es
 
-	xorl %eax,%eax			# Clear FS/GS and LDT
+	xorl %eax,%eax			# Clear FS and LDT
 	movl %eax,%fs
-	movl %eax,%gs
 	lldt %ax
+
+	movl $(__KERNEL_PDA),%eax
+	mov  %eax,%gs
+
 	cld			# gcc2 wants the direction flag cleared at all times
 	pushl $0		# fake return address for unwinder
 #ifdef CONFIG_SMP
@@ -345,6 +349,23 @@ check_x87:
 	.byte 0xDB,0xE4		/* fsetpm for 287, ignored by 387 */
 	ret
 
+/*
+ * Point the GDT at this CPU's PDA.  On boot this will be
+ * cpu_gdt_table and boot_pda; for secondary CPUs, these will be
+ * that CPU's GDT and PDA.
+ */
+setup_pda:
+	/* get the PDA pointer */
+	movl start_pda, %eax
+
+	/* slot the PDA address into the GDT */
+	mov cpu_gdt_descr+2, %ecx
+	mov %ax, (__KERNEL_PDA+0+2)(%ecx)		/* base & 0x0000ffff */
+	shr $16, %eax
+	mov %al, (__KERNEL_PDA+4+0)(%ecx)		/* base & 0x00ff0000 */
+	mov %ah, (__KERNEL_PDA+4+3)(%ecx)		/* base & 0xff000000 */
+	ret
+
 /*
  *  setup_idt
  *
@@ -484,6 +505,8 @@ ENTRY(empty_zero_page)
  * This starts the data section.
  */
 .data
+ENTRY(start_pda)
+	.long boot_pda
 
 ENTRY(stack_start)
 	.long init_thread_union+THREAD_SIZE
@@ -525,7 +548,7 @@ idt_descr:
 
 # boot GDT descriptor (later on used by CPU#0):
 	.word 0				# 32 bit align gdt_desc.address
-cpu_gdt_descr:
+ENTRY(cpu_gdt_descr)
 	.word GDT_ENTRIES*8-1
 	.long cpu_gdt_table
 
@@ -585,7 +608,7 @@ ENTRY(cpu_gdt_table)
 	.quad 0x004092000000ffff	/* 0xc8 APM DS    data */
 
 	.quad 0x00c0920000000000	/* 0xd0 - ESPFIX SS */
-	.quad 0x0000000000000000	/* 0xd8 - PDA */
+	.quad 0x00cf92000000ffff	/* 0xd8 - PDA */
 	.quad 0x0000000000000000	/* 0xe0 - unused */
 	.quad 0x0000000000000000	/* 0xe8 - unused */
 	.quad 0x0000000000000000	/* 0xf0 - unused */

arch/i386/kernel/process.c

@@ -56,6 +56,7 @@
 
 #include <asm/tlbflush.h>
 #include <asm/cpu.h>
+#include <asm/pda.h>
 
 asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
 
@@ -346,6 +347,7 @@ int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
 
 	regs.xds = __USER_DS;
 	regs.xes = __USER_DS;
+	regs.xgs = __KERNEL_PDA;
 	regs.orig_eax = -1;
 	regs.eip = (unsigned long) kernel_thread_helper;
 	regs.xcs = __KERNEL_CS | get_kernel_rpl();
@@ -431,7 +433,6 @@ int copy_thread(int nr, unsigned long clone_flags, unsigned long esp,
 	p->thread.eip = (unsigned long) ret_from_fork;
 
 	savesegment(fs,p->thread.fs);
-	savesegment(gs,p->thread.gs);
 
 	tsk = current;
 	if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
@@ -659,16 +660,16 @@ struct task_struct fastcall * __switch_to(struct task_struct *prev_p, struct tas
 	load_esp0(tss, next);
 
 	/*
-	 * Save away %fs and %gs. No need to save %es and %ds, as
-	 * those are always kernel segments while inside the kernel.
-	 * Doing this before setting the new TLS descriptors avoids
-	 * the situation where we temporarily have non-reloadable
-	 * segments in %fs and %gs.  This could be an issue if the
-	 * NMI handler ever used %fs or %gs (it does not today), or
-	 * if the kernel is running inside of a hypervisor layer.
+	 * Save away %fs. No need to save %gs, as it was saved on the
+	 * stack on entry.  No need to save %es and %ds, as those are
+	 * always kernel segments while inside the kernel.  Doing this
+	 * before setting the new TLS descriptors avoids the situation
+	 * where we temporarily have non-reloadable segments in %fs
+	 * and %gs.  This could be an issue if the NMI handler ever
+	 * used %fs or %gs (it does not today), or if the kernel is
+	 * running inside of a hypervisor layer.
 	 */
 	savesegment(fs, prev->fs);
-	savesegment(gs, prev->gs);
 
 	/*
 	 * Load the per-thread Thread-Local Storage descriptor.
@@ -676,16 +677,13 @@ struct task_struct fastcall * __switch_to(struct task_struct *prev_p, struct tas
 	load_TLS(next, cpu);
 
 	/*
-	 * Restore %fs and %gs if needed.
+	 * Restore %fs if needed.
 	 *
-	 * Glibc normally makes %fs be zero, and %gs is one of
-	 * the TLS segments.
+	 * Glibc normally makes %fs be zero.
 	 */
 	if (unlikely(prev->fs | next->fs))
 		loadsegment(fs, next->fs);
 
-	if (prev->gs | next->gs)
-		loadsegment(gs, next->gs);
 
 	/*
 	 * Restore IOPL if needed.

arch/i386/kernel/signal.c

@@ -128,7 +128,7 @@ restore_sigcontext(struct pt_regs *regs, struct sigcontext __user *sc, int *peax
 			 X86_EFLAGS_TF | X86_EFLAGS_SF | X86_EFLAGS_ZF | \
 			 X86_EFLAGS_AF | X86_EFLAGS_PF | X86_EFLAGS_CF)
 
-	GET_SEG(gs);
+	COPY_SEG(gs);
 	GET_SEG(fs);
 	COPY_SEG(es);
 	COPY_SEG(ds);
@@ -244,9 +244,7 @@ setup_sigcontext(struct sigcontext __user *sc, struct _fpstate __user *fpstate,
 {
 	int tmp, err = 0;
 
-	tmp = 0;
-	savesegment(gs, tmp);
-	err |= __put_user(tmp, (unsigned int __user *)&sc->gs);
+	err |= __put_user(regs->xgs, (unsigned int __user *)&sc->gs);
 	savesegment(fs, tmp);
 	err |= __put_user(tmp, (unsigned int __user *)&sc->fs);
 

include/asm-i386/mmu_context.h

@@ -62,8 +62,8 @@ static inline void switch_mm(struct mm_struct *prev,
 #endif
 }
 
-#define deactivate_mm(tsk, mm) \
-	asm("movl %0,%%fs ; movl %0,%%gs": :"r" (0))
+#define deactivate_mm(tsk, mm)			\
+	asm("movl %0,%%fs": :"r" (0));
 
 #define activate_mm(prev, next) \
 	switch_mm((prev),(next),NULL)

include/asm-i386/processor.h

@@ -473,6 +473,7 @@ struct thread_struct {
 	.vm86_info = NULL,						\
 	.sysenter_cs = __KERNEL_CS,					\
 	.io_bitmap_ptr = NULL,						\
+	.gs = __KERNEL_PDA,						\
 }
 
 /*
@@ -500,7 +501,8 @@ static inline void load_esp0(struct tss_struct *tss, struct thread_struct *threa
 }
 
 #define start_thread(regs, new_eip, new_esp) do {		\
-	__asm__("movl %0,%%fs ; movl %0,%%gs": :"r" (0));	\
+	__asm__("movl %0,%%fs": :"r" (0));			\
+	regs->xgs = 0;						\
 	set_fs(USER_DS);					\
 	regs->xds = __USER_DS;					\
 	regs->xes = __USER_DS;					\

include/asm-i386/ptrace.h

@@ -16,6 +16,8 @@ struct pt_regs {
 	long eax;
 	int  xds;
 	int  xes;
+	/* int  xfs; */
+	int  xgs;
 	long orig_eax;
 	long eip;
 	int  xcs;

kernel/fork.c

@@ -1303,7 +1303,7 @@ fork_out:
 	return ERR_PTR(retval);
 }
 
-struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
+noinline struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
 {
 	memset(regs, 0, sizeof(struct pt_regs));
 	return regs;