Hi Alex, Timur, Christian,
Thanks Alex for picking these up, and Christian for the Trinity
background.
Timur, yes, that is possible. The only place where Trinity/Richland
gets the interrupt enabled is the family check in vce_v1_0_start() in
patch 2. Today it reads "rdev->family < CHIP_BONAIRE", which also
matches CHIP_ARUBA. Narrowing it to
if (rdev->family >= CHIP_TAHITI && rdev->family < CHIP_BONAIRE)
enables it on Tahiti, Pitcairn, Verde, Oland and Hainan only, and
leaves Trinity/Richland (and CIK) as they were. I built vce_v1_0.o
with that change against 6.12 and checked the gate against the
radeon_family enum: SI on, ARUBA and CIK off. On SI the generated
code is the same register write I tested on the Cape Verde card, so
the earlier results still apply there. I cannot test the Trinity side.
The handling added to evergreen_irq_process() by patch 1 stays
harmless on Trinity: with the trap interrupt left disabled it is
never raised.
Since the series is already applied, I did not want to send a v3 on
top of it unasked. If Christian would still prefer to keep Trinity
out of the picture, especially for stable, I can send this as a
small follow-up against the current tree right away. Otherwise I
will leave it as it is.
Thanks,
Matheus
