[PATCH-RFC] x86: Remove last user of get_segment_eip

From: Harvey Harrison
Date: Fri Jan 11 2008 - 17:29:20 EST


Ingo, this probably shouldn't be applied without someone
with a much deeper understanding having a look. I think
I've properly understood how the segment limits were being
set/returned, but I am _not_ sure that probe_kernel_address()
will catch all the same cases, so I could be on crack here.

is_prefetch was the last user of get_segment_eip and only on
X86_32. This function returned the faulting instruction's
address and set the upper segment limit.

Instead, use the convert_ip_to_linear helper and rely on
probe_kernel_address to do the segment checks which was
already done everywhere the segment limit was being checked
on X86_32.

Remove get_segment_eip as well.

Signed-off-by: Harvey Harrison <harvey.harrison@xxxxxxxxx>
---
arch/x86/kernel/step.c | 84 ----------------------------------------------
arch/x86/mm/fault_32.c | 15 +-------
arch/x86/mm/fault_64.c | 15 +-------
include/asm-x86/ptrace.h | 6 ++-
4 files changed, 8 insertions(+), 112 deletions(-)

diff --git a/arch/x86/kernel/step.c b/arch/x86/kernel/step.c
index 923aee6..6685c7a 100644
--- a/arch/x86/kernel/step.c
+++ b/arch/x86/kernel/step.c
@@ -5,90 +5,6 @@
#include <linux/mm.h>
#include <linux/ptrace.h>

-#ifdef CONFIG_X86_32
-#include <linux/uaccess.h>
-
-#include <asm/desc.h>
-
-/*
- * 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.
- */
-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
-
-#ifdef CONFIG_X86_32
-static
-#endif
unsigned long convert_ip_to_linear(struct task_struct *child, struct pt_regs *regs)
{
unsigned long addr, seg;
diff --git a/arch/x86/mm/fault_32.c b/arch/x86/mm/fault_32.c
index f0bbcab..7bde8a8 100644
--- a/arch/x86/mm/fault_32.c
+++ b/arch/x86/mm/fault_32.c
@@ -81,7 +81,6 @@ static int is_prefetch(struct pt_regs *regs, unsigned long addr,
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. */
@@ -90,30 +89,23 @@ static int is_prefetch(struct pt_regs *regs, unsigned long addr,
} 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_ip_to_linear(current, regs);
#endif

+ instr = (unsigned char *)convert_ip_to_linear(current, regs);
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;

@@ -155,10 +147,7 @@ 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) &&
diff --git a/arch/x86/mm/fault_64.c b/arch/x86/mm/fault_64.c
index 7dedb6b..880ae66 100644
--- a/arch/x86/mm/fault_64.c
+++ b/arch/x86/mm/fault_64.c
@@ -84,7 +84,6 @@ static int is_prefetch(struct pt_regs *regs, unsigned long addr,
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. */
@@ -93,30 +92,23 @@ static int is_prefetch(struct pt_regs *regs, unsigned long addr,
} 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_ip_to_linear(current, regs);
#endif

+ instr = (unsigned char *)convert_ip_to_linear(current, regs);
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;

@@ -158,10 +150,7 @@ 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) &&
diff --git a/include/asm-x86/ptrace.h b/include/asm-x86/ptrace.h
index 35c1037..ee4b595 100644
--- a/include/asm-x86/ptrace.h
+++ b/include/asm-x86/ptrace.h
@@ -70,6 +70,10 @@ struct pt_regs {
#include <asm/segment.h>

struct task_struct;
+
+extern unsigned long
+convert_ip_to_linear(struct task_struct *child, struct pt_regs *regs);
+
extern void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs, int error_code);

/*
@@ -184,8 +188,6 @@ convert_ip_to_linear(struct task_struct *child, struct pt_regs *regs);

#ifdef __KERNEL__

-unsigned long get_segment_eip(struct pt_regs *regs, unsigned long *eip_limit);
-
/*
* These are defined as per linux/ptrace.h, which see.
*/
--
1.5.4.rc2.1164.g6451

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/