On 09/17/2015 02:40 PM, Paulo Zanoni wrote: > 2015-09-17 18:23 GMT-03:00 Jesse Barnes <[email protected]>: >> According to the PCI docs and Rafael, we need to be using >> pm_runtime_put_noidle() and pm_runtime_get_noresume() in our init and >> teardown routines, rather than using a direct enable/disable pair (and >> we didn't even have the enable side, so never autosuspended after an >> unload). >> >> This fixes one failure of the basic-pci-d3-state test on my BYT. I'm >> still debugging why the device never autosuspends. >> >> Signed-off-by: Jesse Barnes <[email protected]> >> --- >> drivers/gpu/drm/i915/intel_runtime_pm.c | 5 ++--- >> 1 file changed, 2 insertions(+), 3 deletions(-) >> >> diff --git a/drivers/gpu/drm/i915/intel_runtime_pm.c >> b/drivers/gpu/drm/i915/intel_runtime_pm.c >> index 85c35fd..1addb8a 100644 >> --- a/drivers/gpu/drm/i915/intel_runtime_pm.c >> +++ b/drivers/gpu/drm/i915/intel_runtime_pm.c >> @@ -1822,7 +1822,7 @@ static void intel_runtime_pm_disable(struct >> drm_i915_private *dev_priv) >> >> /* Make sure we're not suspended first. */ > > So is the comment above still valid? > > As far as I remember, we explicitly wake up the hardware after unload > because our driver is (was) not prepared to be loaded on a hardware > that is suspended. Did you try module reloading after this change?
I'm still debugging it; in fact I think this change may be wrong. > Also, basic-pci-d3-state should not be unloading/reloading the driver, > so it's not clear to me how this change helps passing that test. The BAT tests just happen to run the module reload test first. That's how I caught this in the first place... Jesse _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
