Comment # 23 on bug 87278 from Daniel Scharrer

(Some of) the GPU faults are back with Mesa git-8a71fd8 and LLVM r229671:

 [11047.892869] radeon 0000:01:00.0: GPU fault detected: 147 0x04088801
 [11047.892875] radeon 0000:01:00.0:   VM_CONTEXT1_PROTECTION_FAULT_ADDR  
0x0FF00820
 [11047.892878] radeon 0000:01:00.0:   VM_CONTEXT1_PROTECTION_FAULT_STATUS
0x08088001
 [11047.892881] VM fault (0x01, vmid 4) at page 267388960, read from TC (136)
 [...]

There are also plenty of Packet0 errors still/again.

This happens after 4db985a5fa9ea985616a726b1770727309502d81 which reverts
0e9cdedd2e3943bdb7f3543a3508b883b167e427 "radeon/llvm: enable unsafe math for
graphics shaders" as mentioned in bug 89069 comment 21. Unlike before this bug
was closed, now there are only GPU faults after actually loading a level, which
is not covered in the above trace. Here is the new, longer trace from the other
bug report - maybe it will also allow others to better reproduce the Packet0
errors:

 http://constexpr.org/tmp/TalosDemo-radeonsi.2.trace.xz (83 MiB)