On 14/06/2019 19:22, Chris Wilson wrote:
Quoting John Harrison (2019-06-14 19:17:25)
On 6/14/2019 10:19, Tvrtko Ursulin wrote:
Read-back of 0x20ec (GEN9_CS_DEBUG_MODE1) seems to be causing issues.
It could be put on the wo_register list if appropriate. And/or made
RING_FORCE_TO_NONPRIV_RW.
However now I realize that handling of new whitelisting modes is
missing from the selftest altogether. All RING_FORCE_TO_NONPRIV_RD and
RING_FORCE_TO_NONPRIV_WR probably need explicit handling there - first
ones to check they cannot be written to, and skip read-back on the
second one.
Regards,
Tvrtko
It looks like there are multiple issues.
I think the quickest/simplest option right now is to drop a bunch of the
ICL updates (because there are too many for the current 12 slot maximum)
and to drop the problematic CFL updates.
The only changes that are urgently needed are the HUC status registers.
Unfortunately, they are read-only entries. So the self-test will need to
be updated to cope with those. Not sure what it can do though. Do we
just skip read-only entries? The test can't know what a valid value to
read back is.
Plan sounds good to me. (Drop everything apart from HuC.)
There's an exception list [wo_register] for registers that don't
conform to the test expectations.
But that's not appropriate for read-only ones. I think for these we
modify the selftest to read before and after writing and check value is
unchanged. Is that doable today or Monday at latest?
Regards,
Tvrtko
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
