On 08/12/2016 22:29, Chris Wilson wrote:
On Thu, Dec 08, 2016 at 04:52:24PM +0000, Tvrtko Ursulin wrote:
Idea for another late test:
for (offset = 0; offset < 0x40000; offset++) {
fw_domain = intel_uncore_forcewake_for_reg(dev_priv, { .reg =
offset }, FW_REG_READ | FW_REG_WRITE);
if (WARN_ON(fw_domain & ~dev_priv->uncore.fw_domains))
return -EINVAL;
}
And then we could convert the existing related WARNs in the live
code base to GEM_WARN_ONs.
I liked it enough to type something up... However, isn't the argument
for skipping some fw_domains that the associated register banks are
invalid/absent on those platforms? i.e. we can't simply walk
for (offset = 0; NEEDS_FORCE_WAKE(offset); offset++)
and expect every offset to correspond to a register (and so need a
covering fw_domain)?
I thought it won't be a problem because it would return 0 in those
cases. So only those offsets that need a forcewake in a platform can
trigger the fail.
Oh I see. We would need another layer before the condition checking, like:
fw_domains = intel_uncore_forcewake_for_reg(...)
if ((fw_domains & FORCEWAKE_MEDIA) && !HAS_ENGINE(VCS))
fw_domains &= ~FORCEWAKE_MEDIA;
...
WARN_ON(fw_domains & ~...);
Would that work?
Regards,
Tvrtko
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
