Bug 96246

Not the first time I see such a backtrace, iirc, mclasen had reported on irc a similar backtrace but could not reproduce either.

Assuming the backtrace is not corrupted

$ rpm -qf /usr/lib/debug/usr/bin/Xwayland.debug
xorg-x11-server-debuginfo-1.18.3-4.fc25.x86_64

$ addr2line -fe /usr/lib/debug/usr/bin/Xwayland.debug 0x424e7a 0x4276ec 0x549c93 0x4f5b67 0x549c93 0x587ef0 0x55034d 0x55638f 0x55a3c3

gives:

InitOutput
/usr/src/debug/xorg-server-1.18.3/hw/xwayland/xwayland.c:751
xwl_screen_init_output
:?
FreeCursor
/usr/src/debug/xorg-server-1.18.3/dix/cursor.c:120 (discriminator 3)
dixGetPrivateAddr
/usr/src/debug/xorg-server-1.18.3/miext/sync/../../include/privates.h:123
FreeCursor
/usr/src/debug/xorg-server-1.18.3/dix/cursor.c:120 (discriminator 3)
ChangeWindowAttributes
/usr/src/debug/xorg-server-1.18.3/dix/window.c:1567
ProcDestroyWindow
/usr/src/debug/xorg-server-1.18.3/dix/dispatch.c:712
Dispatch
/usr/src/debug/xorg-server-1.18.3/dix/dispatch.c:353
dix_main
/usr/src/debug/xorg-server-1.18.3/dix/main.c:340

Not sure I make much sense out of that...

I tried to get more info, but not sure how much it helps... I logged in and attached to Xwayland and waited for the crash and got: 

Thread 1 "Xwayland" received signal SIGSEGV, Segmentation fault.
xwl_seat_set_cursor (xwl_seat=0x1839ce0) at xwayland-cursor.c:127
127	        memcpy(pixmap->devPrivate.ptr,
(gdb) 
Continuing.
[Thread 0x7fde28eb5700 (LWP 2002) exited]
[Thread 0x7fde296b6700 (LWP 2001) exited]
[Thread 0x7fde29eb7700 (LWP 2000) exited]
[Thread 0x7fde36125ec0 (LWP 1998) exited]
[Inferior 1 (process 1998) exited with code 01]

(In reply to Kevin Fenzi from comment #3)
> I tried to get more info, but not sure how much it helps... I logged in and
> attached to Xwayland and waited for the crash and got: 
> 
> Thread 1 "Xwayland" received signal SIGSEGV, Segmentation fault.
> xwl_seat_set_cursor (xwl_seat=0x1839ce0) at xwayland-cursor.c:127
> 127	        memcpy(pixmap->devPrivate.ptr,
> (gdb) 

Oh, that's interesting because we had commit 1815540 cherry-picked in 1.18.x so that might be related, not sure exactly how yet, though...

https://cgit.freedesktop.org/xorg/xserver/commit/?id=1815540

Can you try to reproduce and capture a bit more data then:

 - a backtrace:

 (gdb) bt full

 - print the values of xwl_seat, pixmap, cursor and cursor->bits in xwl_seat_set_cursor() where it crashes:

 (gdb) p *xwl_seat
 (gdb) p *pixmap
 (gdb) p *cursor
 (gdb) p *cursor->bits

(Note, some value might  be optimized out, in which case we won't get their content)

Thanks for your help!

PS: Still no idea how to reproduce? I don't understand why I don't see this at all...

I wonder, is it possible to get a pointer_handle_enter() after a unrealize_cursor()?

If that happens, we would now end up calling xwl_seat_set_cursor() from pointer_handle_enter() with xwl_seat->cursor_frame_cb == NULL so we wouldn't return in xwl_seat_set_cursor() and do the memcpy() onto pixmap->devPrivate.ptr even though it would have been destroyed previously in xwl_unrealize_cursor()

Well, it's just a theory at this point...

Created attachment 124226 [details] [review]
[PATCH RFC xserver] wayland: clear resource for pixmap on unrealize

Once you've captured the data as described in commit 4, you may want to try this patch, I wonder if that would make any difference...

So, I waited for it to crash all yesterday and this morning and of course it didn't. ;) 

But finally it just crashed, but the trace was somewhat different: 

Thread 1 "Xwayland" received signal SIGSEGV, Segmentation fault.
dixGetPrivate (key=0x816720 <xwl_pixmap_private_key>, privates=0x20)
    at ../../include/privates.h:137
137	    return *(void **) dixGetPrivateAddr(privates, key);
(gdb) where
#0  dixGetPrivate (key=0x816720 <xwl_pixmap_private_key>, privates=0x20)
    at ../../include/privates.h:137
#1  dixLookupPrivate (key=0x816720 <xwl_pixmap_private_key>, privates=0x20)
    at ../../include/privates.h:167
#2  xwl_pixmap_get (pixmap=pixmap@entry=0x0) at xwayland.c:179
#3  0x0000000000426f3c in xwl_shm_destroy_pixmap (pixmap=0x0) at xwayland-shm.c:234
#4  0x0000000000549c93 in FreeCursor (value=0x51bc390, cid=cid@entry=0)
    at cursor.c:122
#5  0x00000000004ee637 in AnimCurUnrealizeCursor (pDev=<optimized out>, 
    pScreen=0xb083c0, pCursor=0x5232180) at animcur.c:278
#6  0x0000000000549c93 in FreeCursor (value=value@entry=0x5232180, cid=cid@entry=0)
    at cursor.c:122
#7  0x0000000000587ef0 in ChangeWindowAttributes (pWin=0x13b8d10, 
    vmask=<optimized out>, vlist=vlist@entry=0x51c1fb0, 
    client=client@entry=0xb01410) at window.c:1567
#8  0x00000000005501ed in ProcChangeWindowAttributes (client=0xb01410)
    at dispatch.c:677
#9  0x000000000055630f in Dispatch () at dispatch.c:430
#10 0x000000000055a333 in dix_main (argc=10, argv=0x7ffd07acbd68, 
    envp=<optimized out>) at main.c:300
#11 0x00007f748e393231 in __libc_start_main (main=0x423840 <main>, argc=10, 
    argv=0x7ffd07acbd68, init=<optimized out>, fini=<optimized out>, 
    rtld_fini=<optimized out>, stack_end=0x7ffd07acbd58) at ../csu/libc-start.c:289
#12 0x0000000000423879 in _start ()
(gdb) p *xwl_seat
No symbol "xwl_seat" in current context.
(gdb) p *pixmap
No symbol "pixmap" in current context.

I'll setup to regather...

Just got another of the same ones:

Thread 1 "Xwayland" received signal SIGSEGV, Segmentation fault.
dixGetPrivate (key=0x816720 <xwl_pixmap_private_key>, privates=0x20)
    at ../../include/privates.h:137
137	    return *(void **) dixGetPrivateAddr(privates, key);
(gdb) where
#0  dixGetPrivate (key=0x816720 <xwl_pixmap_private_key>, privates=0x20)
    at ../../include/privates.h:137
#1  dixLookupPrivate (key=0x816720 <xwl_pixmap_private_key>, privates=0x20)
    at ../../include/privates.h:167
#2  xwl_pixmap_get (pixmap=pixmap@entry=0x0) at xwayland.c:179
#3  0x0000000000426f7c in xwl_shm_destroy_pixmap (pixmap=0x0) at xwayland-shm.c:234
#4  0x0000000000549cd3 in FreeCursor (value=0x35c6fc0, cid=cid@entry=0)
    at cursor.c:122
#5  0x00000000004ee677 in AnimCurUnrealizeCursor (pDev=<optimized out>, 
    pScreen=0x17373c0, pCursor=0x3633010) at animcur.c:278
#6  0x0000000000549cd3 in FreeCursor (value=value@entry=0x3633010, cid=cid@entry=0)
    at cursor.c:122
#7  0x0000000000587f30 in ChangeWindowAttributes (pWin=0x1fe2d70, 
    vmask=<optimized out>, vlist=vlist@entry=0x35d5060, 
    client=client@entry=0x1730410) at window.c:1567
#8  0x000000000055022d in ProcChangeWindowAttributes (client=0x1730410)
    at dispatch.c:677
#9  0x000000000055634f in Dispatch () at dispatch.c:430
#10 0x000000000055a373 in dix_main (argc=10, argv=0x7fff76b49e18, 
    envp=<optimized out>) at main.c:300
#11 0x00007f41a518f461 in __libc_start_main () from /lib64/libc.so.6
#12 0x00000000004238b9 in _start ()

Created attachment 124254 [details] [review]
[PATCH xserver v2] wayland: clear resource for pixmap on unrealize

Can you try this patch instead?

It seems to me that FreeCursor() can be called more than once (FreeCursor() -> (*pscr->UnrealizeCursor) -> AnimCurUnrealizeCursor() -> FreeCursor()) so that we would end up calling xwl_unrealize_cursor() on a cursor with a pixmap that was already freed previously.

ok. Running with that patch now. ;)

So, it's not crashed yet with that patch... but I have seen an odddity: 

twice now when I have been in a X app, the app and cursor have stopped responding at all. Switching to a vt and back and everything returns to normal. 

In my gdb session I see: 

Detaching after fork from child process 25619.
Detaching after fork from child process 439.

Will keep watching for a crash...

(In reply to Kevin Fenzi from comment #11)
> So, it's not crashed yet with that patch... but I have seen an odddity: 
> 
> twice now when I have been in a X app, the app and cursor have stopped
> responding at all. Switching to a vt and back and everything returns to
> normal. 
> [...]

Thanks! Is that not responding to cursor only or any input (including keyboard)? Was that the same X11 application in both cases (and possibly which app)?

I suspect my patch might fix the effect but not the root cause of the issue...

It seemed to be any input (but it did take the control-alt-f3 to switch to vty3). 

It was claws-mail in all cases so far I think.

(In reply to Kevin Fenzi from comment #13)
> It seemed to be any input (but it did take the control-alt-f3 to switch to
> vty3). 

OK, so it's probably unrelated to this patch ...
 
> It was claws-mail in all cases so far I think.

... and we cannot possibly rule out an application or even compositor (gnome-shell) bug for this.

I'll send the patch to the ML for further comments then.

I'll keep trying to get it to crash again. 

Last night/this morning so far there have been 0 input issues, but I haven't yet setup gdb from my phone. Perhaps gdm was interfering somehow there and causing those input issues I was seeing?

FYI, I have seen no crashes since I applied the patch in comment 9. ;)

I have submitted a slightly different version including Rui's suggestion to xorg-devel ML here:

https://patchwork.freedesktop.org/series/8234/

The patch has landed in master, closing.