Closed jennycao closed 6 years ago
Interesting, this timeout shows up on APL but not SKL. Here are the relevant messages I see in the logs:
08-24 14:59:11.553 6061 6079 I dEQP :
08-24 14:59:11.553 6061 6079 I dEQP : Test case 'dEQP-VK.ubo.random.all_shared_buffer.48'..
08-24 14:59:11.691 3095 4039 D vndksupport: Loading /vendor/lib64/hw/gralloc.project-celadon.so from current namespace instead of sphal namespace.
08-24 14:59:11.709 3095 4039 I ConfigStore: android::hardware::configstore::V1_0::ISurfaceFlingerConfigs::hasWideColorDisplay retrieved: 0
08-24 14:59:11.709 3095 4039 I ConfigStore: android::hardware::configstore::V1_0::ISurfaceFlingerConfigs::hasHDRDisplay retrieved: 0
08-24 14:59:11.709 3095 4039 E libEGL : eglSetBlobCacheFuncsANDROID resulted in an error: 0x300c
08-24 14:59:11.709 3095 4039 I OpenGLRenderer: Initialized EGL, version 1.4
08-24 14:59:11.709 3095 4039 D OpenGLRenderer: Swap behavior 2
08-24 14:59:12.642 3095 3618 D DhcpClient: Broadcasting DHCPDISCOVER
08-24 14:59:12.890 3095 4190 V BackupManagerConstants: getFullBackupRequiredNetworkType(...) returns 2
08-24 14:59:12.890 3095 4190 V BackupManagerConstants: getFullBackupRequireCharging(...) returns true
08-24 14:59:42.692 6061 6078 F dEQP : Watchdog timer timeout for touch interval
08-24 14:59:42.716 3095 3170 W InputDispatcher: channel 'fc4708a com.drawelements.deqp/android.app.NativeActivity (server)' ~ Consumer closed input channel or an error occurred. events=0x9
08-24 14:59:42.716 3095 3170 E InputDispatcher: channel 'fc4708a com.drawelements.deqp/android.app.NativeActivity (server)' ~ Channel is unrecoverably broken and will be disposed!
08-24 14:59:42.717 3095 3579 I WindowManager: WIN DEATH: Window{fc4708a u0 com.drawelements.deqp/android.app.NativeActivity}
08-24 14:59:42.717 3095 3579 W InputDispatcher: Attempted to unregister already unregistered input channel 'fc4708a com.drawelements.deqp/android.app.NativeActivity (server)'
08-24 14:59:42.717 3095 3113 W libprocessgroup: kill(-6061, 9) failed: No such process
08-24 14:59:42.717 3095 3113 I libprocessgroup: Successfully killed process cgroup uid 10061 pid 6061 in 0ms
08-24 14:59:42.717 2942 2942 I Zygote : Process 6061 exited due to signal (9)
08-24 14:59:42.717 3095 5042 I ActivityManager: Process com.drawelements.deqp:testercore (pid 6061) has died: fore TOP
08-24 14:59:42.718 3095 5042 W ActivityManager: Force removing ActivityRecord{50708b8 u0 com.drawelements.deqp/android.app.NativeActivity t7}: app died, no saved state
08-24 14:59:42.719 3669 3669 D StatusBar: disable<e i a s b h r c s > disable2<q i n >
08-24 14:59:42.723 2962 3081 W SurfaceFlinger: Attempting to destroy on removed layer: AppWindowToken{36d59f6 token=Token{ef6f691 ActivityRecord{50708b8 u0 com.drawelements.deqp/android.app.NativeActivity t7}}}#0
08-24 14:59:42.734 3095 3118 W ActivityManager: setHasOverlayUi called on unknown pid: 6061
08-24 14:59:43.121 2987 3021 E storaged: getDiskStats failed with result NOT_SUPPORTED and size 0
08-24 14:59:47.828 3095 3579 I ActivityManager: Force stopping com.drawelements.deqp appid=10061 user=0: finished inst
08-24 14:59:47.829 3095 6097 W Binder : Outgoing transactions from this process must be FLAG_ONEWAY
08-24 14:59:47.829 3095 6097 W Binder : java.lang.Throwable
08-24 14:59:47.829 3095 6097 W Binder : at android.os.BinderProxy.transact(Binder.java:1114)
08-24 14:59:47.829 3095 6097 W Binder : at android.app.IInstrumentationWatcher$Stub$Proxy.instrumentationFinished(IInstrumentationWatcher.java:164)
08-24 14:59:47.829 3095 6097 W Binder : at com.android.server.am.InstrumentationReporter$MyThread.run(InstrumentationReporter.java:86)
08-24 14:59:47.829 3095 3579 I ActivityManager: Killing 6044:com.drawelements.deqp/u0a61 (adj 0): stop com.drawelements.deqp
I'll have to add some more printing to see what call it is hanging on.
With the same image, this issue could be reproduced on APL, but not on KBL.
I was able to determine that the test is hanging during pipeline creation - not a big surprise given how intensive that operation is. Tomorrow I will dive into the driver to find out exactly where we are spending time.
This might be shader compile/link, KBL is much more powerful so it does not timeout.
Right, looks like we timeout while the vertex shader is getting compiled. It doesn't look like there is anything wrong going on here, this is just a particularly nasty shader. So we are either going to need to increase the timeout value or figure out how to optimize the compiler.
But this seems one regression since last mesa rebase, @jennycao please help fix me.
Sorry for the noise, this is not a regression. The regression should be dEQP-GLES31.functional.ssbo.layout.random.all_shared_buffer#23.
this test are dropped from CTS 9.0_R2, so close this bug.
CTS version: 9.0_R1 Following test failed. dEQP-VK.ubo.random.all_shared_buffer#48| fail (timeout)