Date: Thu, 15 Nov 2001 14:56:31 -0500 (EST) From: Joe Clarke <marcus@marcuscom.com> To: Maxim Sobolev <sobomax@FreeBSD.org> Cc: freebsd-ports@FreeBSD.org, <freebsd-stable@FreeBSD.org>, <hackers@FreeBSD.org> Subject: Re: sigreturn: eflags creash (fixed!) Message-ID: <20011115145029.G47613-200000@shumai.marcuscom.com> In-Reply-To: <3BF41BE2.BD7D8EAA@FreeBSD.org>
next in thread | previous in thread | raw e-mail | index | archive | help
[-- Attachment #1 --]
On Thu, 15 Nov 2001, Maxim Sobolev wrote:
> Joe Clarke wrote:
> >
> > Sorry for the wide distribution, but I have sent email to both lists
> > regarding this problem in the past. It seems that while doing intensive
> > threaded operations in Gnome applications, crashes occur with the
> > following kernel message:
> >
> > sigreturn: eflags 0x280
> >
> > And, in .gnomerc-errors:
> >
> > Fatal error 'Thread has returned from sigreturn or longjump'
> >
> > The problem as I have found it is with libpng. libpng uses MMX
> > optimizations by default on FreeBSD. If you apply the attached patch to
> > patch-aa in /usr/ports/graphics/png, the problem goes away. You only need
> > to recompile and install libpng. Reinstalling Gnome isn't necessary. It
> > seems the MMX optimizations are corrupting eflags, and when a thread tries
> > to restore context after a signal, things go really wrong.
> >
> > The true fix will probably come in analyzing the MMX code in libpng.
> > Unfortunately, I don't know enough about x86 assembly to be of much use
> > here. Hopefully this will help other experiencing the same problem.
>
> Very interesting, and weird if true. I'll test this tomorrow. In the
> meantime, could some CPU guru confirm or reject theoretical
> possibility of MMX user-level code causing problems to the kernel?
I learned about this by reading through some of the -hackers archives.
One person complained of similar errors trying to get xine to work on
FreeBSD. Removing the MMX detection code fixed it. I remembered libpng
also used MMX, so I removed the pnggccrd.c source, and voila!
Based on core dumps, strace output, and a lot of code surfing, this makes
sense to me. Basically, any png-dependent app's thread that runs longer
than what ITIMER_PROF is set to gets hit with a SIGPROF. When that
happens, things context switch. eflags must have been corrupted by the
MMX code, thus sigreturn() bombs out, and causes uthread_kern to die as
well. Here's what strace looks like when balsa tries to read a 33 MB
mailbox:
...
74202 --- SIGPROF (Profiling timer expired) ---
74202 --- SIGPROF (Profiling timer expired) ---
74202 gettimeofday({1005789324, 257513}, NULL) = 0
74202 sigprocmask(SIG_SETMASK, [], NULL) = 0
74202 sigaltstack({ss_sp=0x811b000, ss_flags=0, ss_size=40960}, NULL) = 0
74202 poll([{fd=5, events=POLLIN}, {fd=7, events=POLLIN}, {fd=10,
events=POLLIN}, {fd=11, events=POLLIN|POLLPRI}, {fd=13, events=POLLIN},
{fd=12, events=POLLIN}], 6, 0) = 0
74202 sigreturn(0x81f2c64
When this happens, strace politely dies with a bus error.
Thanks for testing this, Maxim. Hopefully someone can find the problem
and fix it for good.
Joe
P.S. my original patch sucks I was in a hurry. The one attached is
better.
>
> -Maxim
>
>
[-- Attachment #2 --]
--- patch-aa.orig Thu Nov 15 14:55:43 2001
+++ patch-aa Thu Nov 15 14:55:57 2001
@@ -43,8 +43,8 @@
-all: libpng.a pngtest
+.if (${ARCH} == "i386")
-+CFLAGS+=-DPNG_USE_PNGGCCRD
-+OBJS+=pnggccrd.o
++#CFLAGS+=-DPNG_USE_PNGGCCRD
++#OBJS+=pnggccrd.o
+.endif
+
+.SUFFIXES: .c .so .o
@@ -91,5 +91,5 @@
pngwtran.o: png.h pngconf.h
pngwutil.o: png.h pngconf.h
pngpread.o: png.h pngconf.h
-+pnggccrd.o: png.h pngconf.h pngasmrd.h
++#pnggccrd.o: png.h pngconf.h pngasmrd.h
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20011115145029.G47613-200000>