| Message ID | 20230922123541.23822-1-nirmoy.das@intel.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | drm/i915/gem: Make i915_gem_shrinker multi-gt aware | expand |
On 22.09.2023 14:35, Nirmoy Das wrote: > From: Jonathan Cavitt <jonathan.cavitt@intel.com> > > Where applicable, use for_each_gt instead of to_gt in the > i915_gem_shrinker functions to make them apply to more than just the > primary GT. Specifically, this ensure i915_gem_shrink_all retires all > requests across all GTs, and this makes i915_gem_shrinker_vmap unmap > VMAs from all GTs. > > Signed-off-by: Jonathan Cavitt <jonathan.cavitt@intel.com> > Signed-off-by: Nirmoy Das <nirmoy.das@intel.com> > --- > drivers/gpu/drm/i915/gem/i915_gem_shrinker.c | 44 ++++++++++++-------- > 1 file changed, 26 insertions(+), 18 deletions(-) > > diff --git a/drivers/gpu/drm/i915/gem/i915_gem_shrinker.c b/drivers/gpu/drm/i915/gem/i915_gem_shrinker.c > index 214763942aa2..3ef1fd32f80a 100644 > --- a/drivers/gpu/drm/i915/gem/i915_gem_shrinker.c > +++ b/drivers/gpu/drm/i915/gem/i915_gem_shrinker.c > @@ -14,6 +14,7 @@ > #include <linux/vmalloc.h> > > #include "gt/intel_gt_requests.h" > +#include "gt/intel_gt.h" > > #include "i915_trace.h" > > @@ -119,7 +120,8 @@ i915_gem_shrink(struct i915_gem_ww_ctx *ww, > intel_wakeref_t wakeref = 0; > unsigned long count = 0; > unsigned long scanned = 0; > - int err = 0; > + int err = 0, i = 0; > + struct intel_gt *gt; > > /* CHV + VTD workaround use stop_machine(); need to trylock vm->mutex */ > bool trylock_vm = !ww && intel_vm_no_concurrent_access_wa(i915); > @@ -147,9 +149,11 @@ i915_gem_shrink(struct i915_gem_ww_ctx *ww, > * what we can do is give them a kick so that we do not keep idle > * contexts around longer than is necessary. > */ > - if (shrink & I915_SHRINK_ACTIVE) > - /* Retire requests to unpin all idle contexts */ > - intel_gt_retire_requests(to_gt(i915)); > + if (shrink & I915_SHRINK_ACTIVE) { > + for_each_gt(gt, i915, i) > + /* Retire requests to unpin all idle contexts */ > + intel_gt_retire_requests(to_gt(i915)); to_gt(...) -> gt ? > + } > > /* > * As we may completely rewrite the (un)bound list whilst unbinding > @@ -389,6 +393,8 @@ i915_gem_shrinker_vmap(struct notifier_block *nb, unsigned long event, void *ptr > struct i915_vma *vma, *next; > unsigned long freed_pages = 0; > intel_wakeref_t wakeref; > + struct intel_gt *gt; > + int i; > > with_intel_runtime_pm(&i915->runtime_pm, wakeref) > freed_pages += i915_gem_shrink(NULL, i915, -1UL, NULL, > @@ -397,24 +403,26 @@ i915_gem_shrinker_vmap(struct notifier_block *nb, unsigned long event, void *ptr > I915_SHRINK_VMAPS); > > /* We also want to clear any cached iomaps as they wrap vmap */ > - mutex_lock(&to_gt(i915)->ggtt->vm.mutex); > - list_for_each_entry_safe(vma, next, > - &to_gt(i915)->ggtt->vm.bound_list, vm_link) { > - unsigned long count = i915_vma_size(vma) >> PAGE_SHIFT; > - struct drm_i915_gem_object *obj = vma->obj; > - > - if (!vma->iomap || i915_vma_is_active(vma)) > - continue; > + for_each_gt(gt, i915, i) { > + mutex_lock(>->ggtt->vm.mutex); > + list_for_each_entry_safe(vma, next, > + >->ggtt->vm.bound_list, vm_link) { > + unsigned long count = i915_vma_size(vma) >> PAGE_SHIFT; > + struct drm_i915_gem_object *obj = vma->obj; > + > + if (!vma->iomap || i915_vma_is_active(vma)) > + continue; > > - if (!i915_gem_object_trylock(obj, NULL)) > - continue; > + if (!i915_gem_object_trylock(obj, NULL)) > + continue; > > - if (__i915_vma_unbind(vma) == 0) > - freed_pages += count; > + if (__i915_vma_unbind(vma) == 0) > + freed_pages += count; > > - i915_gem_object_unlock(obj); > + i915_gem_object_unlock(obj); > + } > + mutex_unlock(>->ggtt->vm.mutex); > } > - mutex_unlock(&to_gt(i915)->ggtt->vm.mutex); This seems correct. With 1st stanza fixed: Reviewed-by: Andrzej Hajda <andrzej.hajda@intel.com> Regards Andrzej > > *(unsigned long *)ptr += freed_pages; > return NOTIFY_DONE;
On 9/25/2023 3:23 PM, Andrzej Hajda wrote: > On 22.09.2023 14:35, Nirmoy Das wrote: >> From: Jonathan Cavitt <jonathan.cavitt@intel.com> >> >> Where applicable, use for_each_gt instead of to_gt in the >> i915_gem_shrinker functions to make them apply to more than just the >> primary GT. Specifically, this ensure i915_gem_shrink_all retires all >> requests across all GTs, and this makes i915_gem_shrinker_vmap unmap >> VMAs from all GTs. >> >> Signed-off-by: Jonathan Cavitt <jonathan.cavitt@intel.com> >> Signed-off-by: Nirmoy Das <nirmoy.das@intel.com> >> --- >> drivers/gpu/drm/i915/gem/i915_gem_shrinker.c | 44 ++++++++++++-------- >> 1 file changed, 26 insertions(+), 18 deletions(-) >> >> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_shrinker.c >> b/drivers/gpu/drm/i915/gem/i915_gem_shrinker.c >> index 214763942aa2..3ef1fd32f80a 100644 >> --- a/drivers/gpu/drm/i915/gem/i915_gem_shrinker.c >> +++ b/drivers/gpu/drm/i915/gem/i915_gem_shrinker.c >> @@ -14,6 +14,7 @@ >> #include <linux/vmalloc.h> >> #include "gt/intel_gt_requests.h" >> +#include "gt/intel_gt.h" >> #include "i915_trace.h" >> @@ -119,7 +120,8 @@ i915_gem_shrink(struct i915_gem_ww_ctx *ww, >> intel_wakeref_t wakeref = 0; >> unsigned long count = 0; >> unsigned long scanned = 0; >> - int err = 0; >> + int err = 0, i = 0; >> + struct intel_gt *gt; >> /* CHV + VTD workaround use stop_machine(); need to trylock >> vm->mutex */ >> bool trylock_vm = !ww && intel_vm_no_concurrent_access_wa(i915); >> @@ -147,9 +149,11 @@ i915_gem_shrink(struct i915_gem_ww_ctx *ww, >> * what we can do is give them a kick so that we do not keep idle >> * contexts around longer than is necessary. >> */ >> - if (shrink & I915_SHRINK_ACTIVE) >> - /* Retire requests to unpin all idle contexts */ >> - intel_gt_retire_requests(to_gt(i915)); >> + if (shrink & I915_SHRINK_ACTIVE) { >> + for_each_gt(gt, i915, i) >> + /* Retire requests to unpin all idle contexts */ >> + intel_gt_retire_requests(to_gt(i915)); > > > to_gt(...) -> gt ? Wow, a huge miss. Thanks will resend! > > >> + } >> /* >> * As we may completely rewrite the (un)bound list whilst >> unbinding >> @@ -389,6 +393,8 @@ i915_gem_shrinker_vmap(struct notifier_block *nb, >> unsigned long event, void *ptr >> struct i915_vma *vma, *next; >> unsigned long freed_pages = 0; >> intel_wakeref_t wakeref; >> + struct intel_gt *gt; >> + int i; >> with_intel_runtime_pm(&i915->runtime_pm, wakeref) >> freed_pages += i915_gem_shrink(NULL, i915, -1UL, NULL, >> @@ -397,24 +403,26 @@ i915_gem_shrinker_vmap(struct notifier_block >> *nb, unsigned long event, void *ptr >> I915_SHRINK_VMAPS); >> /* We also want to clear any cached iomaps as they wrap vmap */ >> - mutex_lock(&to_gt(i915)->ggtt->vm.mutex); >> - list_for_each_entry_safe(vma, next, >> - &to_gt(i915)->ggtt->vm.bound_list, vm_link) { >> - unsigned long count = i915_vma_size(vma) >> PAGE_SHIFT; >> - struct drm_i915_gem_object *obj = vma->obj; >> - >> - if (!vma->iomap || i915_vma_is_active(vma)) >> - continue; >> + for_each_gt(gt, i915, i) { >> + mutex_lock(>->ggtt->vm.mutex); >> + list_for_each_entry_safe(vma, next, >> + >->ggtt->vm.bound_list, vm_link) { >> + unsigned long count = i915_vma_size(vma) >> PAGE_SHIFT; >> + struct drm_i915_gem_object *obj = vma->obj; >> + >> + if (!vma->iomap || i915_vma_is_active(vma)) >> + continue; >> - if (!i915_gem_object_trylock(obj, NULL)) >> - continue; >> + if (!i915_gem_object_trylock(obj, NULL)) >> + continue; >> - if (__i915_vma_unbind(vma) == 0) >> - freed_pages += count; >> + if (__i915_vma_unbind(vma) == 0) >> + freed_pages += count; >> - i915_gem_object_unlock(obj); >> + i915_gem_object_unlock(obj); >> + } >> + mutex_unlock(>->ggtt->vm.mutex); >> } >> - mutex_unlock(&to_gt(i915)->ggtt->vm.mutex); > > > This seems correct. > > With 1st stanza fixed: > Reviewed-by: Andrzej Hajda <andrzej.hajda@intel.com> Thanks, Nirmoy > > Regards > Andrzej > > >> *(unsigned long *)ptr += freed_pages; >> return NOTIFY_DONE; >
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_shrinker.c b/drivers/gpu/drm/i915/gem/i915_gem_shrinker.c index 214763942aa2..3ef1fd32f80a 100644 --- a/drivers/gpu/drm/i915/gem/i915_gem_shrinker.c +++ b/drivers/gpu/drm/i915/gem/i915_gem_shrinker.c @@ -14,6 +14,7 @@ #include <linux/vmalloc.h> #include "gt/intel_gt_requests.h" +#include "gt/intel_gt.h" #include "i915_trace.h" @@ -119,7 +120,8 @@ i915_gem_shrink(struct i915_gem_ww_ctx *ww, intel_wakeref_t wakeref = 0; unsigned long count = 0; unsigned long scanned = 0; - int err = 0; + int err = 0, i = 0; + struct intel_gt *gt; /* CHV + VTD workaround use stop_machine(); need to trylock vm->mutex */ bool trylock_vm = !ww && intel_vm_no_concurrent_access_wa(i915); @@ -147,9 +149,11 @@ i915_gem_shrink(struct i915_gem_ww_ctx *ww, * what we can do is give them a kick so that we do not keep idle * contexts around longer than is necessary. */ - if (shrink & I915_SHRINK_ACTIVE) - /* Retire requests to unpin all idle contexts */ - intel_gt_retire_requests(to_gt(i915)); + if (shrink & I915_SHRINK_ACTIVE) { + for_each_gt(gt, i915, i) + /* Retire requests to unpin all idle contexts */ + intel_gt_retire_requests(to_gt(i915)); + } /* * As we may completely rewrite the (un)bound list whilst unbinding @@ -389,6 +393,8 @@ i915_gem_shrinker_vmap(struct notifier_block *nb, unsigned long event, void *ptr struct i915_vma *vma, *next; unsigned long freed_pages = 0; intel_wakeref_t wakeref; + struct intel_gt *gt; + int i; with_intel_runtime_pm(&i915->runtime_pm, wakeref) freed_pages += i915_gem_shrink(NULL, i915, -1UL, NULL, @@ -397,24 +403,26 @@ i915_gem_shrinker_vmap(struct notifier_block *nb, unsigned long event, void *ptr I915_SHRINK_VMAPS); /* We also want to clear any cached iomaps as they wrap vmap */ - mutex_lock(&to_gt(i915)->ggtt->vm.mutex); - list_for_each_entry_safe(vma, next, - &to_gt(i915)->ggtt->vm.bound_list, vm_link) { - unsigned long count = i915_vma_size(vma) >> PAGE_SHIFT; - struct drm_i915_gem_object *obj = vma->obj; - - if (!vma->iomap || i915_vma_is_active(vma)) - continue; + for_each_gt(gt, i915, i) { + mutex_lock(>->ggtt->vm.mutex); + list_for_each_entry_safe(vma, next, + >->ggtt->vm.bound_list, vm_link) { + unsigned long count = i915_vma_size(vma) >> PAGE_SHIFT; + struct drm_i915_gem_object *obj = vma->obj; + + if (!vma->iomap || i915_vma_is_active(vma)) + continue; - if (!i915_gem_object_trylock(obj, NULL)) - continue; + if (!i915_gem_object_trylock(obj, NULL)) + continue; - if (__i915_vma_unbind(vma) == 0) - freed_pages += count; + if (__i915_vma_unbind(vma) == 0) + freed_pages += count; - i915_gem_object_unlock(obj); + i915_gem_object_unlock(obj); + } + mutex_unlock(>->ggtt->vm.mutex); } - mutex_unlock(&to_gt(i915)->ggtt->vm.mutex); *(unsigned long *)ptr += freed_pages; return NOTIFY_DONE;