Re: [Valgrind-developers] DARWIN branch regtests

[Greg P. <gp...@ap...>]

On Apr 21, 2009, at 3:33 PM, Filipe Cabecinhas wrote:
> This is a weird one. I copied the definition of alarm() from libc  
> but the error cause eludes me. Attached is a small test case for the  
> error.
> The output is:
> $ valgrind -q --track-origins=yes ./test-alarmallegedly  
> uninitialized value: 0xbffff6e8, start of oitv: 0xbffff6dc, end:  
> 0xbffff6ec
> ==84680== Conditional jump or move depends on uninitialised value(s)
> ==84680==    at 0x1F38: alrm (test-alarm.c:24)
> ==84680==    by 0x1FAC: main (test-alarm.c:39)
> ==84680==  Uninitialised value was created by a stack allocation
> ==84680==    at 0x1EAE: alrm (test-alarm.c:11)
> in handler: 14
>
> What's weird is that line 24 reads something that is inside the  
> struct itimerval oitv and setitimer writes to the address specified  
> for the third parameter for sizeof(struct itimerval) bytes, as we  
> can see in syswrap-generic.c:
>
> PRE(sys_setitimer)
> {
> ...
>   if (ARG3 != (Addr)NULL)
>      PRE_MEM_WRITE( "setitimer(ovalue)", ARG3, sizeof(struct  
> vki_itimerval));
> }
>
> POST(sys_setitimer)
> {
>   if (ARG3 != (Addr)NULL)
>      POST_MEM_WRITE(ARG3, sizeof(struct vki_itimerval));
> ...
>
> vki_itimerval is #defined to itimerval.
>
> What could be the error?

The error is the syscall_table entry in syswrap-darwin.c, at least if  
yours looks like mine:

     GENX_(__NR_setitimer, sys_setitimer),

That should be GENXY; as written it fails to run the POST handler.

Aside: struct itimerval is two struct timevals, and 64-bit struct  
timeval has an alignment pad in the middle, so you should use  
PRE_timeval_READ/PRE_timeval_WRITE/POST_timeval_WRITE. Using  
PRE_MEM_READ(sizeof(struct itimerval)) to check setitimer(ARG2) will  
cause false error reports.


-- 
Greg Parker     gp...@ap...     Runtime Wrangler