Bug 17004

it also happens on q965.

Created attachment 18151 [details]
xorg conf file

before running this demo ,you disable GEM by running "export INTEL_NO_GEM=1".

It is a new behavior with GEM,which uses drm-gem for all components except drm module and kernel from anholt's linux-2.6 tree drm-gem-merge branch.

the issue still exists with the latest master tip.

the issue still exists with the latest GEM.

Not reaching this point with classic on my 915gm -- something else is going wrong, pegging the cpu and leaving junk on the screen when the app is killed.

With mesa_7_2 branch, it work well.

But on 965 ,this bug also exists with mesa_7_2 branch.

libdrm:

commit 2db8e0c8ef8c7a66460fceda129533b364f6418c
Author: Eric Anholt <eric@anholt.net>
Date:   Tue Sep 23 17:06:01 2008 -0700

    intel: Allow up to 15 seconds chewing on one buffer before acknowledging -EB
    
    The gltestperf demo in some cases took over seven seconds to make it through
    one batchbuffer on a GM965.

with you fix,gltestperf still run abort,but without error info: "intelWaitIrq: drm_i915_irq_wait: -16"

So I'm removing these words from summary: with error "intelWaitIrq: drm_i915_irq_wait: -16"

Ok, so it aborts without the old message.  What's the new message?

Created attachment 19225 [details]
Xorg.0.log_new

Created attachment 19226 [details]
dmesg info

New message:
log :  Xorg.0.log_new 
  
dmesg:  dmesg info 

aborted message in gdb:
[Switching to Thread -1211320624 (LWP 7544)]
0xffffe424 in __kernel_vsyscall ()

back trace message:
#0  0xffffe424 in __kernel_vsyscall ()
#1  0x4fa84fa0 in raise () from /lib/libc.so.6
#2  0x4fa868b1 in abort () from /lib/libc.so.6
#3  0xb7a7c079 in _fence_wait_internal (bufmgr_fake=0x805a900, seq=1539997)
    at intel_bufmgr_fake.c:389
#4  0xb7a7c8f6 in dri_fake_reloc_and_validate_buffer (bo=0x832ad18)
    at intel_bufmgr_fake.c:1046
#5  0xb7a7cbdc in dri_fake_bo_exec (bo=0x832ad18, used=16,
    cliprects=0x8202750, num_cliprects=0, DR4=0) at intel_bufmgr_fake.c:1272
#6  0xb7a7af8e in dri_bo_exec (bo=0x832ad18, used=16, cliprects=0x8202750,
    num_cliprects=0, DR4=0) at intel_bufmgr.c:135
#7  0xb7aab71d in _intel_batchbuffer_flush (batch=0x807b4a8,
    file=0xb7c5ea85 "intel_context.c", line=384) at intel_batchbuffer.c:152
#8  0xb7ac4467 in intelFlush (ctx=0x805ac08) at intel_context.c:384
#9  0xb7ab655a in intelClearWithBlit (ctx=0x805ac08, mask=257)
    at intel_blit.c:417
#10 0xb7ab322e in intelClear (ctx=0x805ac08, mask=257) at intel_buffers.c:580
#11 0xb7c0d8c9 in _mesa_Clear (mask=0) at main/buffers.c:184
#12 0x0804945f in init_test03 () at gltestperf.c:110
#13 0x0804a22a in display () at gltestperf.c:499
#14 0xb7f71397 in processWindowWorkList (window=0x8051b00) at glut_event.c:1306
#15 0xb7f71ff8 in glutMainLoop () at glut_event.c:1353
#16 0x08049018 in main (ac=1, av=Cannot access memory at address 0x1d7c
) at gltestperf.c:578

I meant the output of the application.  The backtrace shows you in abort just after
      drmMsg("%s:%d: Error waiting for fence: %s.\n", __FILE__, __LINE__,
	     strerror(-ret));

Did the application really output nothing despite executing that code?

If the error is about -EBUSY, try bumping max busy_count from 5 to something huge, and see how long it actually takes (assuming that gltestperf isn't actually hanging).

The only output of the application when aborted is "Aborted".  and with the lastest gem-classic branch it is still so.

Eric, so gltestperf is working fine on your side? Which platforms have you tested?

Was a problem in the previous releases as well, and this release will be much better at it.  Real applications seem to no longer encounter the problem that this app was hitting.  It's also not a problem with GEM.  Not worth fixing.

(In reply to comment #20)
> Was a problem in the previous releases as well, and this release will be much
> better at it.  Real applications seem to no longer encounter the problem that
> this app was hitting.  It's also not a problem with GEM.  Not worth fixing.
> 
we test it with drm-intel-next kernel ,and it works fine.

agree to close it, since it works with GEM kernel.