[PATCH] ntdll: Initialize USR1 handler XSTATE buffer (valgrind).

[Rémi Bernon]

Conditional jump or move depends on uninitialised value(s)
     at 0x46AC989: NtGetContextThread (signal_x86_64.c:2002)

Conditional jump or move depends on uninitialised value(s)
     at 0x46ABA1E: xstate_to_server (unix_private.h:342)
     by 0x46ABA1E: context_to_server (signal_x86_64.c:1698)

Conditional jump or move depends on uninitialised value(s)
     at 0x46AC18F: NtSetContextThread (signal_x86_64.c:1859)

There's still some uninitialized bytes being passed to writev, but that
seems to come from the pushfd instruction, and could probably be flagged
as false positive.

Signed-off-by: Rémi Bernon 
---
 dlls/ntdll/unix/signal_x86_64.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/dlls/ntdll/unix/signal_x86_64.c b/dlls/ntdll/unix/signal_x86_64.c
index 2a030372fb7..d0f3ece266c 100644
--- a/dlls/ntdll/unix/signal_x86_64.c
+++ b/dlls/ntdll/unix/signal_x86_64.c
@@ -2642,6 +2642,7 @@ static void usr1_handler( int signal, siginfo_t *siginfo, void *ucontext )
         DECLSPEC_ALIGN(64) XSTATE xs;
         context.c.ContextFlags = CONTEXT_FULL;
         context_init_xstate( &context.c, &xs );
+        memset( &xs, 0, offsetof(XSTATE, YmmContext) );

NtGetContextThread( GetCurrentThread(), &context.c );
         wait_suspend( &context.c );

[Jacek Caban]

Hi Rémi,

On 08.03.2021 09:14, Rémi Bernon wrote:
> diff --git a/dlls/ntdll/unix/signal_x86_64.c b/dlls/ntdll/unix/signal_x86_64.c
> index 2a030372fb7..d0f3ece266c 100644
> --- a/dlls/ntdll/unix/signal_x86_64.c
> +++ b/dlls/ntdll/unix/signal_x86_64.c
> @@ -2642,6 +2642,7 @@ static void usr1_handler( int signal, siginfo_t *siginfo, void *ucontext )
>           DECLSPEC_ALIGN(64) XSTATE xs;
>           context.c.ContextFlags = CONTEXT_FULL;
>           context_init_xstate( &context.c, &xs );
> -
>    memset( &xs, 0, offsetof(XSTATE, YmmContext) );
>
>     NtGetContextThread( GetCurrentThread(), &context.c );
>     wait_suspend( &context.c );

I think that the fix should go to NtGetContextThread. I have a patch for 
that, but it breaks our tests (and extended context tests seem generally 
fragile...). I need another look at that...

Thanks,

Jacek