Re: [patch] x86_64: fix is_at_popf() for compat tasks

Re: [patch] x86_64: fix is_at_popf() for compat tasks

[Chuck Ebbert]

In-Reply-To: <200607311054.38585.ak@suse.de>

On Mon, 31 Jul 2006 10:54:38 +0200, Andi Kleen wrote:
>
> > When testing for the REX instruction prefix, first check
> > for a 32-bit task because in compat mode the REX prefix is an
> > increment instruction.
> 
> is_compat_task doesn't actually say that a task is in compat mode
> (it refers to the Linux compat layer, not x86-64 compat mode)
> 
> A better test would be regs->cs == __USER32_CS, but in theory
> there could be other code segments in LDT. I guess that can 
> be ignored though.

How about checking for regs->cs != __USER_CS instead?  In 64-bit mode
a program shouldn't have any other value there while in 32-bit mode
it could be using LDT segments.



From: Chuck Ebbert <76306.1226@compuserve.com>

When testing for the REX instruction prefix, first check
for 32-bit mode because in compat mode the REX prefix is an
increment instruction.

Signed-off-by: Chuck Ebbert <76306.1226@compuserve.com>

--- 2.6.18-rc2-64.orig/arch/x86_64/kernel/ptrace.c
+++ 2.6.18-rc2-64/arch/x86_64/kernel/ptrace.c
@@ -141,8 +141,11 @@ static int is_at_popf(struct task_struct
 		case 0xf0: case 0xf2: case 0xf3:
 			continue;
 
-		/* REX prefixes */
 		case 0x40 ... 0x4f:
+			if (regs->cs != __USER_CS)
+				/* 32-bit mode: register increment */
+				return 0;
+			/* 64-bit mode: REX prefix */
 			continue;
 
 			/* CHECKME: f0, f2, f3 */
-- 
Chuck

Re: [patch] x86_64: fix is_at_popf() for compat tasks

[Andi Kleen]

 
> How about checking for regs->cs != __USER_CS instead?  In 64-bit mode
> a program shouldn't have any other value there while in 32-bit mode
> it could be using LDT segments.


It could in theory, but it shouldn't agreed.  Added thanks.
-Andi

[patch] x86_64: fix is_at_popf() for compat tasks

[Chuck Ebbert]

When testing for the REX instruction prefix, first check
for a 32-bit task because in compat mode the REX prefix is an
increment instruction.

Signed-off-by: Chuck Ebbert <76306.1226@compuserve.com>

---

Compiled and booted but needs a test case...

--- 2.6.18-rc2-64.orig/arch/x86_64/kernel/ptrace.c
+++ 2.6.18-rc2-64/arch/x86_64/kernel/ptrace.c
@@ -141,8 +141,11 @@ static int is_at_popf(struct task_struct
 		case 0xf0: case 0xf2: case 0xf3:
 			continue;
 
-		/* REX prefixes */
 		case 0x40 ... 0x4f:
+			if (is_compat_task())
+				/* register increment */
+				return 0;
+			/* REX prefix */
 			continue;
 
 			/* CHECKME: f0, f2, f3 */
-- 
Chuck

Re: [patch] x86_64: fix is_at_popf() for compat tasks

[Andi Kleen]

On Monday 31 July 2006 09:22, Chuck Ebbert wrote:
> When testing for the REX instruction prefix, first check
> for a 32-bit task because in compat mode the REX prefix is an
> increment instruction.

is_compat_task doesn't actually say that a task is in compat mode
(it refers to the Linux compat layer, not x86-64 compat mode)

A better test would be regs->cs == __USER32_CS, but in theory
there could be other code segments in LDT. I guess that can 
be ignored though.

-Andi