easy-kernel 4.14.8-mc2: nouveau lock ups
|Reporter||A. Wilcox (awilfox)|
|Assignee||Horst Burkhardt (mc68030)|
|Reported||2016-08-14 00:20:57 -0500|
|Modified||2018-01-03 20:21:07 -0600|
|Hardware||Adélie Linux / Intel x86 (64-bit)|
|Importance||Normal / major|
Experiences running KDE Plasma 5 on a GeForce GT 730 (PCI-e) with current easy-kernel (mc3):
Screen locker has a quite groovy failure sometimes, where the clock will wind back and forth. ->
If there are many things using 3D acceleration at once, the card starts throwing errors like so in dmesg:
[68700.018469] nouveau 0000:01:00.0: gr: TRAP ch 7 [007f8d2000 kwin_x11]
[68700.018478] nouveau 0000:01:00.0: gr: GPC0/TPC0/MP trap: global 00000004 [MULTIPLE_WARP_ERRORS] warp 20009 [ILLEGAL_INSTR_ENCODING]
[68700.018491] nouveau 0000:01:00.0: gr: TRAP ch 7 [007f8d2000 kwin_x11]
[68700.018499] nouveau 0000:01:00.0: gr: GPC0/TPC0/MP trap: global 00000004 [MULTIPLE_WARP_ERRORS] warp 80009 [ILLEGAL_INSTR_ENCODING]
Sometimes these errors are recoverable, but occasionally, the display dies entirely. The mouse cursor continues to move, but no input is accepted, and all X clients are shown as 'stuck' in top (verified using ssh). Attempts to pkill them, even with -ABRT or -ILL, results in them becoming " "/Z state. At this point, it varies whether the system sits there unresponsive forever, reboots, or - in one instance - power off entirely.
Initially, I believed this may be an upstream issue in mesa. However, upon closer inspection, I am flagging this a kernel issue because the fix may lie somewhere in 4.7 or 4.8, as nouveau support for the NV1xx chips is progressing rapidly in those DRM trees.
It also did not prove to be an error in the KWin compositor, as even with composition disabled I was able to reproduce this issue.