glmark2 [desktop] effect=shadow:windows=4 failure on mx3

Bug #1347828 reported by Kevin DuBois
6
This bug affects 1 person
Affects Status Importance Assigned to Milestone
glmark2
Opinion
Undecided
Unassigned

Bug Description

./glmark2-es2-mir --reuse-context --fullscreen
running against mir_demo_server_basic

once glmark2 gets to the test:
[desktop] effect=shadow:windows=4

on the mx3, the driver starts complaining, and the server/client locks up.

The device's driver kicks out some cryptic messages:
E/IMGSRV ( 3519): :0: FlushRenderSurface: ScheduleTA did not work properly on the current surface
E/IMGSRV ( 3519): :0: SGXKickSubmit: failed acquire CCB space!
E/IMGSRV ( 3519): :0: DoKickTA: SGXKickTA() failed with error 9
E/IMGSRV ( 3519): :0: FlushRenderSurface: ScheduleTA did not work properly on the attachment

E/IMGSRV ( 3519): :0: KEGL_SGXDestroyRenderSurface: Timeout failed on waiting for previous render op
E/IMGSRV ( 3519): :0: FreePDSFragBuffers: PDS fragment buffer for render surface still in use
E/IMGSRV ( 3519): :0: Failed to flush HW render context (24)
E/IMGSRV ( 3519): :0: KEGLDestroyRenderSurface: Couldn't remove render target
E/IMGSRV ( 3519): :0: FBOAttachableHasBeenModified: Couldn't destroy render surface
E/IMGSRV ( 3519): :0: SGXKickSubmit: failed acquire CCB space!
E/IMGSRV ( 3519): :0: DoKickTA: SGXKickTA() failed with error 9
E/IMGSRV ( 3519): :0: FreeListOfVertexUSEVariants: Kicking the TA failed
E/IMGSRV ( 3519):
E/IMGSRV ( 3497): :0: DeferFlushOpWorker: Timed out waiting for ops to flush

Tags: mx3
Kevin DuBois (kdub)
description: updated
Revision history for this message
Kevin DuBois (kdub) wrote :

seems like something is triggering a page fault:

 (dmesg)
[11482.742] c0@A15 kworker/0:1 PVR_K: Found MMU context for page fault 0x00000000
[11482.750] c2@A15 migration/2 IRQ154 no longer affine to CPU2
[11482.752] c0@A15 kworker/0:1 PVR_K: GPU memory context is for PID=5371 (glmark2-es2-mir)

Revision history for this message
kevin gunn (kgunn72) wrote :
Changed in mir:
status: New → Opinion
Revision history for this message
Daniel van Vugt (vanvugt) wrote :

I'm not sure how it's possible a client/server lock-up can be "Opinion". Also, I notice the function "glBindFramebuffer" is not used in Mir's compositing at all. It's only used in Mir's screencasting system (so probably not relevant?).

Changed in mir:
status: Opinion → Incomplete
Revision history for this message
kevin gunn (kgunn72) wrote :

ah, duh...wrong spot.
glmark2 calls glBindBuffer....its not Mir

no longer affects: mir
Changed in glmark2:
status: New → Opinion
Revision history for this message
kevin gunn (kgunn72) wrote :

also using opinion, so this doesn't get auto closed in 90 days....so we don't have people submitting duplicates.
and i say opinion, b/c it seems to be a known driver issue based on IMG's tech boards

To post a comment you must log in.
This report contains Public information  
Everyone can see this information.

Other bug subscribers

Remote bug watches

Bug watches keep track of this bug in other bug trackers.