* [PATCH] Lazy FPU handling in ptrace
@ 2003-03-14 20:56 Andi Kleen
0 siblings, 0 replies; 2+ messages in thread
From: Andi Kleen @ 2003-03-14 20:56 UTC (permalink / raw)
To: linux-kernel
While working on some x86-64 ptrace problems I noticed that i386
has bugs in this area too.
Before doing PTRACE_SETFPREGS or PTRACE_SETFPXREGS you need to do
an unlazy_fpu(), otherwise there is no guarantee that the changed
state will be picked up.
Patch for 2.4, but 2.5 seems to have it too.
(untested, but obviously correct ;-)
-Andi
--- linux-work/arch/i386/kernel/ptrace.c-o 2002-08-08 10:27:42.000000000 +0200
+++ linux-work/arch/i386/kernel/ptrace.c 2003-03-14 21:51:21.000000000 +0100
@@ -381,6 +381,7 @@
ret = -EIO;
break;
}
+ unlazy_fpu(child);
child->used_math = 1;
set_fpregs(child, (struct user_i387_struct *)data);
ret = 0;
@@ -405,6 +406,7 @@
ret = -EIO;
break;
}
+ unlazy_fpu(child);
child->used_math = 1;
ret = set_fpxregs(child, (struct user_fxsr_struct *)data);
break;
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: [PATCH] Lazy FPU handling in ptrace
@ 2003-03-14 21:11 Manfred Spraul
0 siblings, 0 replies; 2+ messages in thread
From: Manfred Spraul @ 2003-03-14 21:11 UTC (permalink / raw)
To: Andi Kleen; +Cc: linux-kernel
Andi wrote:
>(untested, but obviously correct ;-)
>-Andi
>
Famous last words.
The patch is not needed: __switch_to unlazies the fpu state of the
previous thread. ptrace PTRACE_SETFPREGS and PTRACE_SETFPXREGS only
operate on stopped threads, thus it's guaranteed that the fpu state is
stored in the task structure.
--
Manfred
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2003-03-14 21:01 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2003-03-14 20:56 [PATCH] Lazy FPU handling in ptrace Andi Kleen
-- strict thread matches above, loose matches on Subject: below --
2003-03-14 21:11 Manfred Spraul
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox