x86: begin fault_{32|64}.c unification

Move X86_32 only get_segment_eip to X86_64
Move X86_64 only is_errata93 to X86_32

Change X86_32 loop in is_prefetch to highlight the differences
between them.  Fold the logic from __is_prefetch in as well on
X86_32.

Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
This commit is contained in:
Harvey Harrison 2008-01-30 13:32:35 +01:00 committed by Ingo Molnar
parent b6795e65f1
commit 1dc85be087
2 changed files with 196 additions and 25 deletions

View File

@ -61,6 +61,7 @@ static inline int notify_page_fault(struct pt_regs *regs)
#endif
}
#ifdef CONFIG_X86_32
/*
* Return EIP plus the CS segment base. The segment limit is also
* adjusted, clamped to the kernel/user address space (whichever is
@ -135,26 +136,61 @@ static inline unsigned long get_segment_eip(struct pt_regs *regs,
*eip_limit = seg_limit;
return ip + base;
}
#endif
/*
* X86_32
* Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch.
* Check that here and ignore it.
*
* X86_64
* Sometimes the CPU reports invalid exceptions on prefetch.
* Check that here and ignore it.
*
* Opcode checker based on code by Richard Brunner
*/
static int __is_prefetch(struct pt_regs *regs, unsigned long addr)
static int is_prefetch(struct pt_regs *regs, unsigned long addr,
unsigned long error_code)
{
unsigned long limit;
unsigned char *instr = (unsigned char *)get_segment_eip(regs, &limit);
unsigned char *instr;
int scan_more = 1;
int prefetch = 0;
int i;
unsigned char *max_instr;
for (i = 0; scan_more && i < 15; i++) {
#ifdef CONFIG_X86_32
unsigned long limit;
if (unlikely(boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
boot_cpu_data.x86 >= 6)) {
/* Catch an obscure case of prefetch inside an NX page. */
if (nx_enabled && (error_code & PF_INSTR))
return 0;
} else {
return 0;
}
instr = (unsigned char *)get_segment_eip(regs, &limit);
#else
/* If it was a exec fault ignore */
if (error_code & PF_INSTR)
return 0;
instr = (unsigned char __user *)convert_rip_to_linear(current, regs);
#endif
max_instr = instr + 15;
#ifdef CONFIG_X86_64
if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE)
return 0;
#endif
while (scan_more && instr < max_instr) {
unsigned char opcode;
unsigned char instr_hi;
unsigned char instr_lo;
#ifdef CONFIG_X86_32
if (instr > (unsigned char *)limit)
break;
#endif
if (probe_kernel_address(instr, opcode))
break;
@ -196,8 +232,10 @@ static int __is_prefetch(struct pt_regs *regs, unsigned long addr)
case 0x00:
/* Prefetch instruction is 0x0F0D or 0x0F18 */
scan_more = 0;
#ifdef CONFIG_X86_32
if (instr > (unsigned char *)limit)
break;
#endif
if (probe_kernel_address(instr, opcode))
break;
prefetch = (instr_lo == 0xF) &&
@ -211,19 +249,6 @@ static int __is_prefetch(struct pt_regs *regs, unsigned long addr)
return prefetch;
}
static inline int is_prefetch(struct pt_regs *regs, unsigned long addr,
unsigned long error_code)
{
if (unlikely(boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
boot_cpu_data.x86 >= 6)) {
/* Catch an obscure case of prefetch inside an NX page. */
if (nx_enabled && (error_code & 16))
return 0;
return __is_prefetch(regs, addr);
}
return 0;
}
static noinline void force_sig_info_fault(int si_signo, int si_code,
unsigned long address, struct task_struct *tsk)
{
@ -274,6 +299,42 @@ static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address)
return pmd_k;
}
#ifdef CONFIG_X86_64
static const char errata93_warning[] =
KERN_ERR "******* Your BIOS seems to not contain a fix for K8 errata #93\n"
KERN_ERR "******* Working around it, but it may cause SEGVs or burn power.\n"
KERN_ERR "******* Please consider a BIOS update.\n"
KERN_ERR "******* Disabling USB legacy in the BIOS may also help.\n";
/* Workaround for K8 erratum #93 & buggy BIOS.
BIOS SMM functions are required to use a specific workaround
to avoid corruption of the 64bit RIP register on C stepping K8.
A lot of BIOS that didn't get tested properly miss this.
The OS sees this as a page fault with the upper 32bits of RIP cleared.
Try to work around it here.
Note we only handle faults in kernel here. */
static int is_errata93(struct pt_regs *regs, unsigned long address)
{
static int warned;
if (address != regs->ip)
return 0;
if ((address >> 32) != 0)
return 0;
address |= 0xffffffffUL << 32;
if ((address >= (u64)_stext && address <= (u64)_etext) ||
(address >= MODULES_VADDR && address <= MODULES_END)) {
if (!warned) {
printk(errata93_warning);
warned = 1;
}
regs->ip = address;
return 1;
}
return 0;
}
#endif
/*
* Handle a fault on the vmalloc or module mapping area
*

View File

@ -64,10 +64,95 @@ static inline int notify_page_fault(struct pt_regs *regs)
#endif
}
/* Sometimes the CPU reports invalid exceptions on prefetch.
Check that here and ignore.
Opcode checker based on code by Richard Brunner */
static noinline int is_prefetch(struct pt_regs *regs, unsigned long addr,
#ifdef CONFIG_X86_32
/*
* Return EIP plus the CS segment base. The segment limit is also
* adjusted, clamped to the kernel/user address space (whichever is
* appropriate), and returned in *eip_limit.
*
* The segment is checked, because it might have been changed by another
* task between the original faulting instruction and here.
*
* If CS is no longer a valid code segment, or if EIP is beyond the
* limit, or if it is a kernel address when CS is not a kernel segment,
* then the returned value will be greater than *eip_limit.
*
* This is slow, but is very rarely executed.
*/
static inline unsigned long get_segment_eip(struct pt_regs *regs,
unsigned long *eip_limit)
{
unsigned long ip = regs->ip;
unsigned seg = regs->cs & 0xffff;
u32 seg_ar, seg_limit, base, *desc;
/* Unlikely, but must come before segment checks. */
if (unlikely(regs->flags & VM_MASK)) {
base = seg << 4;
*eip_limit = base + 0xffff;
return base + (ip & 0xffff);
}
/* The standard kernel/user address space limit. */
*eip_limit = user_mode(regs) ? USER_DS.seg : KERNEL_DS.seg;
/* By far the most common cases. */
if (likely(SEGMENT_IS_FLAT_CODE(seg)))
return ip;
/* Check the segment exists, is within the current LDT/GDT size,
that kernel/user (ring 0..3) has the appropriate privilege,
that it's a code segment, and get the limit. */
__asm__("larl %3,%0; lsll %3,%1"
: "=&r" (seg_ar), "=r" (seg_limit) : "0" (0), "rm" (seg));
if ((~seg_ar & 0x9800) || ip > seg_limit) {
*eip_limit = 0;
return 1; /* So that returned ip > *eip_limit. */
}
/* Get the GDT/LDT descriptor base.
When you look for races in this code remember that
LDT and other horrors are only used in user space. */
if (seg & (1<<2)) {
/* Must lock the LDT while reading it. */
mutex_lock(&current->mm->context.lock);
desc = current->mm->context.ldt;
desc = (void *)desc + (seg & ~7);
} else {
/* Must disable preemption while reading the GDT. */
desc = (u32 *)get_cpu_gdt_table(get_cpu());
desc = (void *)desc + (seg & ~7);
}
/* Decode the code segment base from the descriptor */
base = get_desc_base((struct desc_struct *)desc);
if (seg & (1<<2))
mutex_unlock(&current->mm->context.lock);
else
put_cpu();
/* Adjust EIP and segment limit, and clamp at the kernel limit.
It's legitimate for segments to wrap at 0xffffffff. */
seg_limit += base;
if (seg_limit < *eip_limit && seg_limit >= base)
*eip_limit = seg_limit;
return ip + base;
}
#endif
/*
* X86_32
* Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch.
* Check that here and ignore it.
*
* X86_64
* Sometimes the CPU reports invalid exceptions on prefetch.
* Check that here and ignore it.
*
* Opcode checker based on code by Richard Brunner
*/
static int is_prefetch(struct pt_regs *regs, unsigned long addr,
unsigned long error_code)
{
unsigned char *instr;
@ -75,21 +160,40 @@ static noinline int is_prefetch(struct pt_regs *regs, unsigned long addr,
int prefetch = 0;
unsigned char *max_instr;
#ifdef CONFIG_X86_32
unsigned long limit;
if (unlikely(boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
boot_cpu_data.x86 >= 6)) {
/* Catch an obscure case of prefetch inside an NX page. */
if (nx_enabled && (error_code & PF_INSTR))
return 0;
} else {
return 0;
}
instr = (unsigned char *)get_segment_eip(regs, &limit);
#else
/* If it was a exec fault ignore */
if (error_code & PF_INSTR)
return 0;
instr = (unsigned char __user *)convert_rip_to_linear(current, regs);
#endif
max_instr = instr + 15;
#ifdef CONFIG_X86_64
if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE)
return 0;
#endif
while (scan_more && instr < max_instr) {
unsigned char opcode;
unsigned char instr_hi;
unsigned char instr_lo;
#ifdef CONFIG_X86_32
if (instr > (unsigned char *)limit)
break;
#endif
if (probe_kernel_address(instr, opcode))
break;
@ -125,12 +229,16 @@ static noinline int is_prefetch(struct pt_regs *regs, unsigned long addr,
scan_more = (instr_lo & 0xC) == 0x4;
break;
case 0xF0:
/* 0xF0, 0xF2, and 0xF3 are valid prefixes in all modes. */
/* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */
scan_more = !instr_lo || (instr_lo>>1) == 1;
break;
case 0x00:
/* Prefetch instruction is 0x0F0D or 0x0F18 */
scan_more = 0;
#ifdef CONFIG_X86_32
if (instr > (unsigned char *)limit)
break;
#endif
if (probe_kernel_address(instr, opcode))
break;
prefetch = (instr_lo == 0xF) &&
@ -185,6 +293,7 @@ bad:
printk("BAD\n");
}
#ifdef CONFIG_X86_64
static const char errata93_warning[] =
KERN_ERR "******* Your BIOS seems to not contain a fix for K8 errata #93\n"
KERN_ERR "******* Working around it, but it may cause SEGVs or burn power.\n"
@ -218,6 +327,7 @@ static int is_errata93(struct pt_regs *regs, unsigned long address)
}
return 0;
}
#endif
static noinline void pgtable_bad(unsigned long address, struct pt_regs *regs,
unsigned long error_code)