On 17/02/18 06:17, Sagar Arun Kamble wrote:
On 2/17/2018 5:48 PM, Chris Wilson wrote:
Quoting Sagar Arun Kamble (2018-02-17 12:10:32)
On 2/17/2018 2:34 PM, Chris Wilson wrote:
Quoting Sagar Arun Kamble (2018-02-17 08:51:44)
Earlier I had thought of calling ASSIGN_FW_DOMAINS_TABLE,
ASSIGN_*_MMIO_VFUNCS before intel_uncore_fw_domains_init
and use I915_READ here for reading the fuse. But that approach
seems to
expose the vfuncs and forcewake table before fw domains
are initialized. Although we can get to know invalid access to
read/write accessors before fw_domains get initialized, current
ordering
of fw_domains init followed by fw domain table/read/write vfuncs init
seems right. So using fw_domains_get/put as suggested above
should be the way.
Chris, Tvrtko, Mika, do you agree?
What's the complication with the fw_domains? Do the additional
powerwell
depend on the fused status of the extra engines? Does it matter if the
fw_domain are prepped if they are never used?
Yes. To discover the available VD/VE engines/power domains, fuse needs
to be read under blitter forcewake as
RC6 will be enabled by BIOS. We do have usage of forcewake in IVB to
discover FORCEWAKE_MT availability in fw_domains_init.
It should not be a problem if they are prepped but never used.
Imo, I would have placed the fused discovery in
intel_engines_init_mmio() (where we do the setup and can take
forcewake).
Then adding something like intel_uncore_reinit_mmio() (which would just
prune the uncore->fw_domains) after checking fused status with
commentary
doesn't seem that horrible.
Yes. This approach looks good too. But, we might want to optimize the
driver_load to avoid this setup at first place
instead of pruning later.
How many cycles does it take to run through all domains and set up the
register offsets? No mmio access required right, we are just moving
memory around without even hitting locked instructions?
Yes. Latency is minimal. We can go with your suggestion. Will need to
maintain separation of engine_cs/device_info/uncore update
through separate functions.
We do currently call fw_domain_reset on all domains at initialization
time, which could write to a register that doesn't exist. We don't wait
for the ack and I assume the write would just be dropped if the power
well is not there so it shouldn't be an issue, but it might be worth
adding a comment in fw_domain_reset to remind us that we can't start
waiting for the ack in there unless we move the fuse read to an earlier
point.
Daniele
Another related setup that can be avoided/pruned is the fw_range for the
engine fused off as it can improve the fw lookup.
Which is a bsearch on register range, I doubt that's going to be
substantially impacted by removing a few ranges. Where it matters, we
should be looking to precalculate the result anyway.
Yes. seems not so worth.
-Chris
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
