On 04/30/2015 12:21 PM, Joonas Lahtinen wrote:
Use partial view for huge BOs (bigger than the mappable aperture) in fault handler so that they can be accessed without trying to make room for them by evicting other objects. v2: - Only use partial views in the case where early rejection was previously done. - Account variable type changes from previous reroll. Cc: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx> Signed-off-by: Joonas Lahtinen <joonas.lahtinen@xxxxxxxxxxxxxxx> --- drivers/gpu/drm/i915/i915_gem.c | 69 +++++++++++++++++++++++++++-------------- 1 file changed, 46 insertions(+), 23 deletions(-) diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c index a020836..2f3fa0b 100644 --- a/drivers/gpu/drm/i915/i915_gem.c +++ b/drivers/gpu/drm/i915/i915_gem.c @@ -1635,6 +1635,7 @@ int i915_gem_fault(struct vm_area_struct *vma, struct vm_fault *vmf) struct drm_i915_gem_object *obj = to_intel_bo(vma->vm_private_data); struct drm_device *dev = obj->base.dev; struct drm_i915_private *dev_priv = dev->dev_private; + struct i915_ggtt_view view = i915_ggtt_view_normal; pgoff_t page_offset; unsigned long pfn; int ret = 0; @@ -1667,8 +1668,21 @@ int i915_gem_fault(struct vm_area_struct *vma, struct vm_fault *vmf) goto unlock; } - /* Now bind it into the GTT if needed */ - ret = i915_gem_obj_ggtt_pin(obj, 0, PIN_MAPPABLE); + /* Use a partial view if the object is bigger than the aperture. */ + if (obj->base.size >= dev_priv->gtt.mappable_end) { + static const unsigned int chunk_size = 256; // 1 MiB + memset(&view, 0, sizeof(view)); + view.type = I915_GGTT_VIEW_PARTIAL; + view.params.partial.offset = rounddown(page_offset, chunk_size); + view.params.partial.size = + min_t(unsigned int, + chunk_size, + (vma->vm_end - vma->vm_start)/PAGE_SIZE - + view.params.partial.offset); + } + + /* Now pin it into the GTT if needed */ + ret = i915_gem_object_ggtt_pin(obj, &view, 0, PIN_MAPPABLE); if (ret) goto unlock; @@ -1681,30 +1695,44 @@ int i915_gem_fault(struct vm_area_struct *vma, struct vm_fault *vmf) goto unpin; /* Finally, remap it using the new GTT offset */ - pfn = dev_priv->gtt.mappable_base + i915_gem_obj_ggtt_offset(obj); + pfn = dev_priv->gtt.mappable_base + + i915_gem_obj_ggtt_offset_view(obj, &view); pfn >>= PAGE_SHIFT; - if (!obj->fault_mappable) { - unsigned long size = min_t(unsigned long, - vma->vm_end - vma->vm_start, - obj->base.size); - int i; + if (unlikely(view.type == I915_GGTT_VIEW_PARTIAL)) { + unsigned long base = vma->vm_start + + (view.params.partial.offset << PAGE_SHIFT); + unsigned int i; - for (i = 0; i < size >> PAGE_SHIFT; i++) { - ret = vm_insert_pfn(vma, - (unsigned long)vma->vm_start + i * PAGE_SIZE, - pfn + i); + for (i = 0; i < view.params.partial.size; i++) { + ret = vm_insert_pfn(vma, base + i * PAGE_SIZE, pfn + i); if (ret) break; } - obj->fault_mappable = true; - } else - ret = vm_insert_pfn(vma, - (unsigned long)vmf->virtual_address, - pfn + page_offset);
If I read the diff correctly you don't have equivalent handling (as the normal view) for when the case when the pre-fault fails somewhere in the middle?
That means if that happens you'll call vm_insert_pfn on the whole range again - is that OK?
Regards, Tvrtko _______________________________________________ Intel-gfx mailing list Intel-gfx@xxxxxxxxxxxxxxxxxxxxx http://lists.freedesktop.org/mailman/listinfo/intel-gfx