diff mbox series

fix segfault on x32, 64-bit time_t-related format strings

Message ID alpine.DEB.2.21.1911202147000.2011@tglase.lan.tarent.de (mailing list archive)
State New, archived
Headers show
Series fix segfault on x32, 64-bit time_t-related format strings | expand

Commit Message

Thorsten Glaser Nov. 20, 2019, 8:51 p.m. UTC
Hi everyone,

here are two patches I’ve been rebasing with each new upload
of alsa-lib to Debian to make things work for my x32 desktop.

The small one fixes it misdetecting x32 as amd64, which leads
to an immediate segfault.

The other one fixes issues related to printing time_t values
on platforms where time_t is 64 bit wide and long 32 bit wide
(various BSDs and all new 32-bit and *64ilp32 Linux platforms).

The Debian packager of alsa-lib hasn’t found it necessary to
forward them upstream, so I’m doing it with this. I hope they
can be included in the next releases. The patches were last
updated for 1.1.8 although I need to update them to 1.1.9 which
was recently introduced to Debian… usually they apply still.

Thanks in advance,
//mirabilos

Comments

Takashi Iwai Nov. 21, 2019, 8:11 a.m. UTC | #1
On Wed, 20 Nov 2019 21:51:14 +0100,
Thorsten Glaser wrote:
> 
> Hi everyone,
> 
> here are two patches I’ve been rebasing with each new upload
> of alsa-lib to Debian to make things work for my x32 desktop.
> 
> The small one fixes it misdetecting x32 as amd64, which leads
> to an immediate segfault.
> 
> The other one fixes issues related to printing time_t values
> on platforms where time_t is 64 bit wide and long 32 bit wide
> (various BSDs and all new 32-bit and *64ilp32 Linux platforms).
> 
> The Debian packager of alsa-lib hasn’t found it necessary to
> forward them upstream, so I’m doing it with this. I hope they
> can be included in the next releases. The patches were last
> updated for 1.1.8 although I need to update them to 1.1.9 which
> was recently introduced to Debian… usually they apply still.

Could you split these changes to separate patches?  They are for
different purposes.

Also, using time_t would be better if possible.  Unfortunately a cast
is needed for printf usage, but other than that, time_t would leave us
the right size.


thanks,

Takashi


> 
> Thanks in advance,
> //mirabilos
> -- 
> tarent solutions GmbH
> Rochusstraße 2-4, D-53123 Bonn • http://www.tarent.de/
> Tel: +49 228 54881-393 • Fax: +49 228 54881-235
> HRB 5168 (AG Bonn) • USt-ID (VAT): DE122264941
> Geschäftsführer: Dr. Stefan Barth, Kai Ebenrett, Boris Esser, Alexander Steeg
> 
> **********
> 
> Mit der tarent Academy bieten wir auch Trainings und Schulungen in den
> Bereichen Softwareentwicklung, Agiles Arbeiten und Zukunftstechnologien an.
> 
> Besuchen Sie uns auf www.tarent.de/academy. Wir freuen uns auf Ihren Kontakt.
> 
> **********
> # DP: fix long vs. long long confusion when there is a 64-bit time_t
> # DP: on a 32-bit long system, such as all newer 32-bit architectures
> 
> --- a/src/pcm/pcm.c
> +++ b/src/pcm/pcm.c
> @@ -2257,11 +2257,11 @@ int snd_pcm_status_dump(snd_pcm_status_t
>  {
>  	assert(status);
>  	snd_output_printf(out, "  state       : %s\n", snd_pcm_state_name((snd_pcm_state_t) status->state));
> -	snd_output_printf(out, "  trigger_time: %ld.%06ld\n",
> -			  status->trigger_tstamp.tv_sec,
> -			  status->trigger_tstamp.tv_nsec / 1000);
> -	snd_output_printf(out, "  tstamp      : %ld.%06ld\n",
> -		status->tstamp.tv_sec, status->tstamp.tv_nsec / 1000);
> +	snd_output_printf(out, "  trigger_time: %lld.%06ld\n",
> +			  (long long)status->trigger_tstamp.tv_sec,
> +			  (long)status->trigger_tstamp.tv_nsec / 1000L);
> +	snd_output_printf(out, "  tstamp      : %lld.%06ld\n",
> +		(long long)status->tstamp.tv_sec, (long)status->tstamp.tv_nsec / 1000L);
>  	snd_output_printf(out, "  delay       : %ld\n", (long)status->delay);
>  	snd_output_printf(out, "  avail       : %ld\n", (long)status->avail);
>  	snd_output_printf(out, "  avail_max   : %ld\n", (long)status->avail_max);
> --- a/test/latency.c
> +++ b/test/latency.c
> @@ -325,12 +325,12 @@ void setscheduler(void)
>  	printf("!!!Scheduler set to Round Robin with priority %i FAILED!!!\n", sched_param.sched_priority);
>  }
>  
> -long timediff(snd_timestamp_t t1, snd_timestamp_t t2)
> +long long timediff(snd_timestamp_t t1, snd_timestamp_t t2)
>  {
> -	signed long l;
> +	signed long long l;
>  
>  	t1.tv_sec -= t2.tv_sec;
> -	l = (signed long) t1.tv_usec - (signed long) t2.tv_usec;
> +	l = (signed long long) t1.tv_usec - (signed long long) t2.tv_usec;
>  	if (l < 0) {
>  		t1.tv_sec--;
>  		l = 1000000 + l;
> @@ -682,10 +682,10 @@ int main(int argc, char *argv[])
>  		snd_pcm_nonblock(phandle, !block ? 1 : 0);
>  		if (ok) {
>  #if 1
> -			printf("Playback time = %li.%i, Record time = %li.%i, diff = %li\n",
> -			       p_tstamp.tv_sec,
> +			printf("Playback time = %lli.%i, Record time = %lli.%i, diff = %lli\n",
> +			       (long long)p_tstamp.tv_sec,
>  			       (int)p_tstamp.tv_usec,
> -			       c_tstamp.tv_sec,
> +			       (long long)c_tstamp.tv_sec,
>  			       (int)c_tstamp.tv_usec,
>  			       timediff(p_tstamp, c_tstamp));
>  #endif
> --- a/test/queue_timer.c
> +++ b/test/queue_timer.c
> @@ -99,11 +99,11 @@ main(int argc ATTRIBUTE_UNUSED, char **a
>  	normalize(&diffdiff);
>  	prevdiff = diff;
>  
> -	fprintf(stderr, " real time: %12ld sec %8ld usec\nqueue time: %12ld sec %8ld usec\n      diff: %12ld sec %8ld usec\n  diffdiff: %12ld sec %8ld usec\n",
> -		tv.tv_sec, tv.tv_usec,
> -		(long)rtime->tv_sec, (long)rtime->tv_nsec / 1000,
> -		diff.tv_sec, diff.tv_usec,
> -		(long)diffdiff.tv_sec, (long)diffdiff.tv_usec);
> +	fprintf(stderr, " real time: %12lld sec %8ld usec\nqueue time: %12lld sec %8ld usec\n      diff: %12lld sec %8ld usec\n  diffdiff: %12lld sec %8ld usec\n",
> +		(long long)tv.tv_sec, (long)tv.tv_usec,
> +		(long long)rtime->tv_sec, (long)rtime->tv_nsec / 1000,
> +		(long long)diff.tv_sec, (long)diff.tv_usec,
> +		(long long)diffdiff.tv_sec, (long)diffdiff.tv_usec);
>  
>  	if (diffdiff.tv_usec >  5000 ||
>  	    diffdiff.tv_usec < -5000) {
> # DP: fix segmentation fault coming from this using amd64 assembly code
> # DP: on x32 systems
> 
> --- a/src/pcm/pcm_dmix.c
> +++ b/src/pcm/pcm_dmix.c
> @@ -145,7 +145,7 @@ static void dmix_server_free(snd_pcm_dir
>  #include "pcm_dmix_generic.c"
>  #if defined(__i386__)
>  #include "pcm_dmix_i386.c"
> -#elif defined(__x86_64__)
> +#elif defined(__x86_64__) && !defined(__ILP32__)
>  #include "pcm_dmix_x86_64.c"
>  #else
>  #ifndef DOC_HIDDEN
> _______________________________________________
> Alsa-devel mailing list
> Alsa-devel@alsa-project.org
> https://mailman.alsa-project.org/mailman/listinfo/alsa-devel
Thorsten Glaser Nov. 21, 2019, 2:37 p.m. UTC | #2
On Thu, 21 Nov 2019, Takashi Iwai wrote:

> Could you split these changes to separate patches?  They are for
> different purposes.

I’ve sent them as two separate patch files attachments.

> Also, using time_t would be better if possible.  Unfortunately a cast
> is needed for printf usage, but other than that, time_t would leave us
> the right size.

In timediff() you mean? Hrm, indeed. I tried to be minimal-invasive
at first.

bye,
//mirabilos
Takashi Iwai Nov. 22, 2019, 1:30 p.m. UTC | #3
On Thu, 21 Nov 2019 15:37:46 +0100,
Thorsten Glaser wrote:
> 
> On Thu, 21 Nov 2019, Takashi Iwai wrote:
> 
> > Could you split these changes to separate patches?  They are for
> > different purposes.
> 
> I’ve sent them as two separate patch files attachments.

Then please make them cleanly applicable via git-am, with a proper
subject, a proper changelog and your sign-off, etc.
At best send one patch per mail (and with a cover letter for multiple
patches).

> > Also, using time_t would be better if possible.  Unfortunately a cast
> > is needed for printf usage, but other than that, time_t would leave us
> > the right size.
> 
> In timediff() you mean? Hrm, indeed. I tried to be minimal-invasive
> at first.

I meant using time_t as much as possible instead of long long.
For printf(), there is no other option, but other calculations can be
replaced with time_t.


thanks,

Takashi
Thorsten Glaser Jan. 23, 2020, 8:28 p.m. UTC | #4
On Fri, 22 Nov 2019, Takashi Iwai wrote:

> > I’ve sent them as two separate patch files attachments.
> 
> Then please make them cleanly applicable via git-am, with a proper
> subject, a proper changelog and your sign-off, etc.

I’ll resend them like that now. Sorry it took a bit, I was busy.

> At best send one patch per mail (and with a cover letter for multiple
> patches).

This is a bit tricky, as I can’t easily inject git format-patch
output into this mail setup. I will try sending them from home.

> > > Also, using time_t would be better if possible.  Unfortunately a cast
> > > is needed for printf usage, but other than that, time_t would leave us
> > > the right size.
> > 
> > In timediff() you mean? Hrm, indeed. I tried to be minimal-invasive
> > at first.
> 
> I meant using time_t as much as possible instead of long long.

I’ve done this now.

bye,
//mirabilos
diff mbox series

Patch

# DP: fix segmentation fault coming from this using amd64 assembly code
# DP: on x32 systems

--- a/src/pcm/pcm_dmix.c
+++ b/src/pcm/pcm_dmix.c
@@ -145,7 +145,7 @@  static void dmix_server_free(snd_pcm_dir
 #include "pcm_dmix_generic.c"
 #if defined(__i386__)
 #include "pcm_dmix_i386.c"
-#elif defined(__x86_64__)
+#elif defined(__x86_64__) && !defined(__ILP32__)
 #include "pcm_dmix_x86_64.c"
 #else
 #ifndef DOC_HIDDEN