| Message ID | 20230925134938.12043-1-nirmoy.das@intel.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | drm/i915/gem: Make i915_gem_shrinker multi-gt aware | expand |
Hi Nirmoy, you forgot the v2 here. On Mon, Sep 25, 2023 at 03:49:38PM +0200, 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. > > v2: Pass correct GT to intel_gt_retire_requests(Andrzej). > > Signed-off-by: Jonathan Cavitt <jonathan.cavitt@intel.com> > Signed-off-by: Nirmoy Das <nirmoy.das@intel.com> > Reviewed-by: Andrzej Hajda <andrzej.hajda@intel.com> [...] > - 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(gt); > + } These two brackets are not needed. > > /* > * 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; the trend is to use 'unsigned int' here and I've seen it reviewed. Personally, if I really have to express a preference, I prefer 'int' because it's a bit safer, generally I don't really mind :) The rest looks good. Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com> Andi
On Mon, 25 Sep 2023, Andi Shyti <andi.shyti@linux.intel.com> wrote: > Hi Nirmoy, > > you forgot the v2 here. > > On Mon, Sep 25, 2023 at 03:49:38PM +0200, 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. >> >> v2: Pass correct GT to intel_gt_retire_requests(Andrzej). >> >> Signed-off-by: Jonathan Cavitt <jonathan.cavitt@intel.com> >> Signed-off-by: Nirmoy Das <nirmoy.das@intel.com> >> Reviewed-by: Andrzej Hajda <andrzej.hajda@intel.com> > > [...] > >> - 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(gt); >> + } > > These two brackets are not needed. > >> >> /* >> * 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; > > the trend is to use 'unsigned int' here and I've seen it > reviewed. Personally, if I really have to express a preference, I > prefer 'int' because it's a bit safer, generally I don't really > mind :) Always use int over unsigned int if you don't have a specific reason not to. ("It can't be negative" is not a good reason.) BR, Jani. > > The rest looks good. > > Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com> > > Andi
Hi Jani, > >> struct i915_vma *vma, *next; > >> unsigned long freed_pages = 0; > >> intel_wakeref_t wakeref; > >> + struct intel_gt *gt; > >> + int i; > > > > the trend is to use 'unsigned int' here and I've seen it > > reviewed. Personally, if I really have to express a preference, I > > prefer 'int' because it's a bit safer, generally I don't really > > mind :) > > Always use int over unsigned int if you don't have a specific reason not > to. ("It can't be negative" is not a good reason.) Finally someone! I totally agree! Andi
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_shrinker.c b/drivers/gpu/drm/i915/gem/i915_gem_shrinker.c index 214763942aa2..9cb7bbfb4278 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(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); *(unsigned long *)ptr += freed_pages; return NOTIFY_DONE;