Hi,
Maybe I should not send patches this late at night as I end up messing things up and spamming sleeping people. Sorry for this version 4.
The real fix is in patch 2. The rest is a helper that adds the with_intel_gt_pm_if_awake() (from Chris) and one more check on the status of the engine before accessing it for clearing the TLB.
Andi
Changelog ========= v3 -> v4 - Fix e-mail mismatch in author and in SoB (Sorry!).
v2 -> v3 (v2: https://patchwork.freedesktop.org/series/103835/) - Add missing header file that was causing a compmile error. - Fix wrong patch formatting.
v1 -> v2 (v1: https://patchwork.freedesktop.org/series/103831/) - Add with_intel_gt_pm_if_awake() macro for gt specific wakeref. - Check if an engine is awake before invalidating its TLB.
Andi Shyti (2): drm/i915/gem: Flush TLBs for all the tiles when clearing an obj drm/i915/gt: Skip TLB invalidation if the engine is not awake
Chris Wilson (1): drm/i915/gt: Ignore TLB invalidations on idle engines
drivers/gpu/drm/i915/gem/i915_gem_pages.c | 12 +++++++++--- drivers/gpu/drm/i915/gt/intel_gt.c | 4 ++++ drivers/gpu/drm/i915/gt/intel_gt_pm.h | 4 ++++ 3 files changed, 17 insertions(+), 3 deletions(-)
From: Chris Wilson chris@chris-wilson.co.uk
As an extension of the current skip TLB invalidations if the device is powered down, we recognised that prior to any engine activity, all the TLBs are explicitly invalidated. Thus anytime we know the engine is asleep, we can skip invalidating the TLBs on that engine.
Signed-off-by: Chris Wilson chris@chris-wilson.co.uk Signed-off-by: Andi Shyti andi.shyti@linux.intel.com --- drivers/gpu/drm/i915/gt/intel_gt_pm.h | 4 ++++ 1 file changed, 4 insertions(+)
diff --git a/drivers/gpu/drm/i915/gt/intel_gt_pm.h b/drivers/gpu/drm/i915/gt/intel_gt_pm.h index bc898df7a48cc..3b1fbce7ea369 100644 --- a/drivers/gpu/drm/i915/gt/intel_gt_pm.h +++ b/drivers/gpu/drm/i915/gt/intel_gt_pm.h @@ -55,6 +55,10 @@ static inline void intel_gt_pm_might_put(struct intel_gt *gt) for (tmp = 1, intel_gt_pm_get(gt); tmp; \ intel_gt_pm_put(gt), tmp = 0)
+#define with_intel_gt_pm_if_awake(gt, tmp) \ + for (tmp = 1, intel_gt_pm_get_if_awake(gt); tmp; \ + intel_gt_pm_put(gt), tmp = 0) + static inline int intel_gt_pm_wait_for_idle(struct intel_gt *gt) { return intel_wakeref_wait_for_idle(>->wakeref);
On 11/05/2022 02:11, Andi Shyti wrote:
From: Chris Wilson chris@chris-wilson.co.uk
As an extension of the current skip TLB invalidations if the device is powered down, we recognised that prior to any engine activity, all the TLBs are explicitly invalidated. Thus anytime we know the engine is asleep, we can skip invalidating the TLBs on that engine.
I've only noticed this after looking at the single patch version.
The commit message here does not match the patch really and CI also says something is not quite right.
Otherwise the idea I think is a good one. I think it can work together with the optimisation I mentioned in my reply to the single patch version.
But there is no rush, for_each_gt flow is unreachable in upstream anyway.
Regards,
Tvrtko
Signed-off-by: Chris Wilson chris@chris-wilson.co.uk Signed-off-by: Andi Shyti andi.shyti@linux.intel.com
drivers/gpu/drm/i915/gt/intel_gt_pm.h | 4 ++++ 1 file changed, 4 insertions(+)
diff --git a/drivers/gpu/drm/i915/gt/intel_gt_pm.h b/drivers/gpu/drm/i915/gt/intel_gt_pm.h index bc898df7a48cc..3b1fbce7ea369 100644 --- a/drivers/gpu/drm/i915/gt/intel_gt_pm.h +++ b/drivers/gpu/drm/i915/gt/intel_gt_pm.h @@ -55,6 +55,10 @@ static inline void intel_gt_pm_might_put(struct intel_gt *gt) for (tmp = 1, intel_gt_pm_get(gt); tmp; \ intel_gt_pm_put(gt), tmp = 0)
+#define with_intel_gt_pm_if_awake(gt, tmp) \
- for (tmp = 1, intel_gt_pm_get_if_awake(gt); tmp; \
intel_gt_pm_put(gt), tmp = 0)- static inline int intel_gt_pm_wait_for_idle(struct intel_gt *gt) { return intel_wakeref_wait_for_idle(>->wakeref);
During object cleanup we invalidate the TLBs but we do it only for gt0. Invalidate the caches for all the tiles.
Reported-by: Chris Wilson chris@chris-wilson.co.uk Signed-off-by: Andi Shyti andi.shyti@linux.intel.com --- drivers/gpu/drm/i915/gem/i915_gem_pages.c | 12 +++++++++--- 1 file changed, 9 insertions(+), 3 deletions(-)
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_pages.c b/drivers/gpu/drm/i915/gem/i915_gem_pages.c index 97c820eee115a..37d23e328bd0c 100644 --- a/drivers/gpu/drm/i915/gem/i915_gem_pages.c +++ b/drivers/gpu/drm/i915/gem/i915_gem_pages.c @@ -13,6 +13,7 @@ #include "i915_gem_mman.h"
#include "gt/intel_gt.h" +#include "gt/intel_gt_pm.h"
void __i915_gem_object_set_pages(struct drm_i915_gem_object *obj, struct sg_table *pages, @@ -217,10 +218,15 @@ __i915_gem_object_unset_pages(struct drm_i915_gem_object *obj)
if (test_and_clear_bit(I915_BO_WAS_BOUND_BIT, &obj->flags)) { struct drm_i915_private *i915 = to_i915(obj->base.dev); - intel_wakeref_t wakeref; + struct intel_gt *gt; + int i;
- with_intel_runtime_pm_if_active(&i915->runtime_pm, wakeref) - intel_gt_invalidate_tlbs(to_gt(i915)); + for_each_gt(gt, i915, i) { + int tmp; + + with_intel_gt_pm_if_awake(gt, tmp) + intel_gt_invalidate_tlbs(gt); + } }
return pages;
We want to check if the engine is awake first before invalidating its cache.
Suggested-by: Chris Wilson chris@chris-wilson.co.uk Signed-off-by: Andi Shyti andi.shyti@linux.intel.com --- drivers/gpu/drm/i915/gt/intel_gt.c | 4 ++++ 1 file changed, 4 insertions(+)
diff --git a/drivers/gpu/drm/i915/gt/intel_gt.c b/drivers/gpu/drm/i915/gt/intel_gt.c index 034182f85501b..a1dc9f4203c2b 100644 --- a/drivers/gpu/drm/i915/gt/intel_gt.c +++ b/drivers/gpu/drm/i915/gt/intel_gt.c @@ -12,6 +12,7 @@ #include "i915_drv.h" #include "intel_context.h" #include "intel_engine_regs.h" +#include "intel_engine_pm.h" #include "intel_gt.h" #include "intel_gt_buffer_pool.h" #include "intel_gt_clock_utils.h" @@ -1219,6 +1220,9 @@ void intel_gt_invalidate_tlbs(struct intel_gt *gt) const unsigned int timeout_ms = 4; struct reg_and_bit rb;
+ if (!intel_engine_pm_is_awake(engine)) + continue; + rb = get_reg_and_bit(engine, regs == gen8_regs, regs, num); if (!i915_mmio_reg_offset(rb.reg)) continue;
dri-devel@lists.freedesktop.org
-
Andi Shyti -
Tvrtko Ursulin