Bug 105871

Summary: UVD decoder fails after hibernate
Product: Drivers Reporter: Jose (jjorge)
Component: Video(DRI - non Intel)Assignee: drivers_video-dri
Status: NEW ---    
Severity: normal    
Priority: P1    
Hardware: x86-64   
OS: Linux   
Kernel Version: 4.1.8 Subsystem:
Regression: No Bisected commit-id:
Attachments: Full hibernate/resume log

Description Jose 2015-10-13 06:22:14 UTC 
[AMD/ATI] RS780 [Radeon HD 3200] works nicely, even after suspend/resume.

But after a hibernate/resume, the UVD decoder fails, while OpenGL apps run ok.
Full dmesg joined, here are the interesting lines :

[  258.694723] ACPI: Waking up from system sleep state S4
[  258.696487] PM: noirq restore of devices complete after 0.724 msecs
[  258.696783] PM: early restore of devices complete after 0.260 msecs
[  258.735733] rtc_cmos 00:01: System wakeup disabled by ACPI
[  258.744344] [drm] PCIE GART of 512M enabled (table at 0x00000000C0258000).
[  258.744394] radeon 0000:01:05.0: WB enabled
[  258.744397] radeon 0000:01:05.0: fence driver on ring 0 use gpu addr 0x00000000a0000c00 and cpu addr 0xffff88003758dc00
[  258.750826] radeon 0000:01:05.0: fence driver on ring 5 use gpu addr 0x00000000c0056038 and cpu addr 0xffffc90000816038
[  258.783182] [drm] ring test on 0 succeeded in 0 usecs
[  258.959618] [drm] ring test on 5 succeeded in 1 usecs
[  258.959622] [drm] UVD initialized successfully.
[  258.959645] [drm] ib test on ring 0 succeeded in 0 usecs
[  269.631608] radeon 0000:01:05.0: ring 5 stalled for more than 10020msec
[  269.631610] radeon 0000:01:05.0: GPU lockup (current fence id 0x0000000000000862 last fence id 0x0000000000000864 on ring 5)
[  269.631693] [drm:uvd_v1_0_ib_test [radeon]] *ERROR* radeon: fence wait failed (-35).
[  269.631719] [drm:radeon_ib_ring_tests [radeon]] *ERROR* radeon: failed testing IB on ring 5 (-35).
Comment 1 Jose 2015-10-13 06:23:43 UTC 
Created attachment 190111 [details]
Full hibernate/resume log
Comment 2 Jose 2015-10-13 06:28:37 UTC 
I forgot to mention that as workaround a suspend/resume works, so it is only the hibernate resume that triggers the bug.