[PATCH] printllval: fix 64bit unpacking on mips/o32 and xtensa

[Dmitry V. Levin]

On Sun, May 05, 2013 at 10:31:15PM -0400, Mike Frysinger wrote:
> On Sunday 05 May 2013 19:38:01 Dmitry V. Levin wrote:
> > On Sun, May 05, 2013 at 06:30:21PM -0400, Mike Frysinger wrote:
> > > While ARM's EABI does aligning of 64bit values when making syscalls, its
> > > OABI does not.  Note that this doesn't work atm in case someone feels
> > > like fixing it :).  We don't keep tracking of the target ABI anywhere,
> > > so this will require tweaking a few core places.
> > > 
> > > The mips/o32 ABI and xtensa arch also do 64bit aligning, so add it to the
> > > printllval list for this.
> > > 
> > > Normally for sys_fadvise64_64 we'd handle the same list of arches, but
> > > mips/o32 ABI is funky and doesn't shift -- it has 7 args.  So just add
> > > xtensa to it.
> > > 
> > > * file.c (sys_fadvise64_64): Add XTENSA to the shifted list.
> > > * util.c (printllval): Add LINUX_MIPSO32 and XTENSA to the shifted list.
> > > ---
> > > 
> > >  file.c | 2 +-
> > >  util.c | 3 ++-
> > >  2 files changed, 3 insertions(+), 2 deletions(-)
> > > 
> > > diff --git a/file.c b/file.c
> > > index e5fc259..07f1ad9 100644
> > > --- a/file.c
> > > +++ b/file.c
> > > @@ -2677,7 +2677,7 @@ sys_fadvise64_64(struct tcb *tcp)
> > > 
> > >  		printfd(tcp, tcp->u_arg[0]);
> > >  		argn = printllval(tcp, ", %lld, ", 1);
> > >  		argn = printllval(tcp, "%lld, ", argn);
> > > 
> > > -#if defined ARM || defined AARCH64 || defined POWERPC
> > > +#if defined ARM || defined AARCH64 || defined POWERPC || defined XTENSA
> > > 
> > >  		printxval(advise, tcp->u_arg[1], "POSIX_FADV_???");
> > >  
> > >  #else
> > >  
> > >  		printxval(advise, tcp->u_arg[argn], "POSIX_FADV_???");
> > > 
> > > diff --git a/util.c b/util.c
> > > index 970faa4..f44c7b6 100644
> > > --- a/util.c
> > > +++ b/util.c
> > > @@ -207,8 +207,9 @@ printllval(struct tcb *tcp, const char *format, int
> > > arg_no)
> > > 
> > >  		arg_no += 2;
> > >  	
> > >  	}
> > >  
> > >  #else
> > > 
> > > -# if defined(ARM) || defined(POWERPC)
> > > +# if defined(ARM) || defined(POWERPC) || defined(LINUX_MIPSO32) ||
> > > defined(XTENSA)
> > >  	/* Align arg_no to the next even number. */
> > > +	/* Note: This is broken for ARM/OABI. */
> > 
> > We have "defined(__ARM_EABI__)" checks in few places, so we could add more
> > such checks if it helps.
> > All ARM systems I can reach are EABI, so I wouldn't be able to test changes
> > related to ARM OABI anyway.
> 
> it would help in the case where you have an OABI strace tracing OABI program, 
> or EABI strace tracing EABI program.  but if you try to cross use OABI to 
> trace EABI or vice versa, it still won't help.

BTW, this is a regression, the proper check for ARM EABI was introduced by
commit v4.6-11-g7a5b08f, but later it was lost in cleanup by commit
v4.6-239-ged720fd.

I've applied your LINUX_MIPSO32||XTENSA fix, and added an ARM OABI fix
on top of it.


-- 
ldv
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 198 bytes
Desc: not available
URL: <http://lists.strace.io/pipermail/strace-devel/attachments/20130507/2106e4c0/attachment.bin>