Hi Nirmoy, please ignore this series... I'm going to send Tvrtko's original work. The changes in patch 1 are not required. Thank you! Andi On Thu, Sep 07, 2023 at 11:04:33AM +0200, Nirmoy Das wrote: > > On 9/5/2023 9:36 PM, Andi Shyti wrote: > > In preparation for multi-gt cache flushing debugfs interface, > > split the cache dropping function and gt idling. > > > > Based on a patch by Tvrtko. > > > > Signed-off-by: Andi Shyti <andi.shyti@xxxxxxxxxxxxxxx> > > Cc: Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxx> > > --- > > drivers/gpu/drm/i915/i915_debugfs.c | 32 +++++++++++++++++++++-------- > > 1 file changed, 24 insertions(+), 8 deletions(-) > > > > diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c > > index 7a90a2e32c9f1..3dfe8a8b7cdfe 100644 > > --- a/drivers/gpu/drm/i915/i915_debugfs.c > > +++ b/drivers/gpu/drm/i915/i915_debugfs.c > > @@ -703,11 +703,25 @@ i915_drop_caches_get(void *data, u64 *val) > > return 0; > > } > > +static int gt_idle(struct intel_gt *gt, u64 val) > > +{ > > + if (val & (DROP_RETIRE | DROP_IDLE)) > > + intel_gt_retire_requests(gt); > > + > > + if (val & DROP_IDLE) { > > + int ret; > > + > > + ret = intel_gt_pm_wait_for_idle(gt); > return intel_gt_pm_wait_for_idle(gt) should work fine. > > + if (ret) > > + return ret; > > + } > > + > > + return 0; > > +} > > + > > static int > > gt_drop_caches(struct intel_gt *gt, u64 val) > > { > > - int ret; > > - > > if (val & DROP_RESET_ACTIVE && > > wait_for(intel_engines_are_idle(gt), 200)) > > intel_gt_set_wedged(gt); > > @@ -716,13 +730,9 @@ gt_drop_caches(struct intel_gt *gt, u64 val) > > intel_gt_retire_requests(gt); > > if (val & (DROP_IDLE | DROP_ACTIVE)) { > > - ret = intel_gt_wait_for_idle(gt, MAX_SCHEDULE_TIMEOUT); > > - if (ret) > > - return ret; > > - } > > + int ret; > > - if (val & DROP_IDLE) { > > - ret = intel_gt_pm_wait_for_idle(gt); > > + ret = intel_gt_wait_for_idle(gt, MAX_SCHEDULE_TIMEOUT); > > if (ret) > > return ret; > > } > > @@ -746,10 +756,16 @@ i915_drop_caches_set(void *data, u64 val) > > drm_dbg(&i915->drm, "Dropping caches: 0x%08llx [0x%08llx]\n", > > val, val & DROP_ALL); > > + /* Flush all the active requests across both GT ... */ > For now it is just GT0 so this comment should be in next patch ? > > ret = gt_drop_caches(to_gt(i915), val); > > if (ret) > > return ret; > > + /* ... then wait for idle as there may be cross-gt wakerefs. */ > > same as above. > > Regards, > > Nirmoy > > > + ret = gt_idle(to_gt(i915), val); > > + if (ret) > > + return ret; > > + > > fs_reclaim_acquire(GFP_KERNEL); > > flags = memalloc_noreclaim_save(); > > if (val & DROP_BOUND)