No problemo, maybe we should split that BUG() statement to catch the
two halves of the PDE address so next time it'll be more obvious.
For ref though your 4 patches have applied cleanly on top of drm-next
every day and work fine on my navi10/raven1 system. So it seems like
just gmc8 is
OH, thanks for that hint! I was staring at the code for a day now
without having a clue what's going wrong.
Haven't realized that something could have been sign extended!
Probably going to figure it out by tomorrow now,
Christian.
Am 27.01.20 um 18:37 schrieb Tom St Denis:
The offending PDE
The offending PDE address is: "3eff60"
Which looks like it was sign extended into the "reserved" section
between bits 40:58 (0x3fff) hence triggering the BUG() in the gmc_v8 driver.
Tom
On 2020-01-27 9:57 a.m., Tom St Denis wrote:
Reverting this patch avoids the gmc_v8 errors (I
Reverting this patch avoids the gmc_v8 errors (I previously sent kernel
logs, here's one for convenience):
[ 358.554335] [ cut here ]
[ 358.554338] kernel BUG at drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c:725!
[ 358.554351] invalid opcode: [#1] SMP NOPTI
[
Allow partial invalidation on unallocated PDs. This is useful when we
need to silence faults to stop interrupt floods on Vega.
Signed-off-by: Christian König
---
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 23 ++-
1 file changed, 18 insertions(+), 5 deletions(-)
diff --git
