On Thu, Dec 22, 2016 at 08:36:04AM +0000, Chris Wilson wrote:
> When we evict from the GTT to make room for an object, the hole we
> create is put onto the MRU stack inside the drm_mm range manager. On the
> next search pass, we can speed up a PIN_HIGH allocation by referencing
> that stack for the new hole.
>
> v2: Pull together the 3 identical implements (ahem, a couple were
> outdated) into a common routine for allocating a node and evicting as
> necessary.
>
> Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
> Reviewed-by: Joonas Lahtinen <joonas.lahtinen at linux.intel.com>
Since it doesn't apply directly to drm-misc I'm leaving this one out for
now. I guess best to merge through drm-intel?
-Daniel
> ---
> drivers/gpu/drm/i915/gvt/aperture_gm.c | 33 +++++-----------
> drivers/gpu/drm/i915/i915_gem_gtt.c | 72
> ++++++++++++++++++++++++----------
> drivers/gpu/drm/i915/i915_gem_gtt.h | 5 +++
> drivers/gpu/drm/i915/i915_vma.c | 40 ++-----------------
> 4 files changed, 70 insertions(+), 80 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/gvt/aperture_gm.c
> b/drivers/gpu/drm/i915/gvt/aperture_gm.c
> index 7d33b607bc89..1bb7a5b80d47 100644
> --- a/drivers/gpu/drm/i915/gvt/aperture_gm.c
> +++ b/drivers/gpu/drm/i915/gvt/aperture_gm.c
> @@ -48,47 +48,34 @@ static int alloc_gm(struct intel_vgpu *vgpu, bool high_gm)
> {
> struct intel_gvt *gvt = vgpu->gvt;
> struct drm_i915_private *dev_priv = gvt->dev_priv;
> - u32 alloc_flag, search_flag;
> + unsigned int flags;
> u64 start, end, size;
> struct drm_mm_node *node;
> - int retried = 0;
> int ret;
>
> if (high_gm) {
> - search_flag = DRM_MM_SEARCH_BELOW;
> - alloc_flag = DRM_MM_CREATE_TOP;
> node = &vgpu->gm.high_gm_node;
> size = vgpu_hidden_sz(vgpu);
> start = gvt_hidden_gmadr_base(gvt);
> end = gvt_hidden_gmadr_end(gvt);
> + flags = PIN_HIGH;
> } else {
> - search_flag = DRM_MM_SEARCH_DEFAULT;
> - alloc_flag = DRM_MM_CREATE_DEFAULT;
> node = &vgpu->gm.low_gm_node;
> size = vgpu_aperture_sz(vgpu);
> start = gvt_aperture_gmadr_base(gvt);
> end = gvt_aperture_gmadr_end(gvt);
> + flags = PIN_MAPPABLE;
> }
>
> mutex_lock(&dev_priv->drm.struct_mutex);
> -search_again:
> - ret = drm_mm_insert_node_in_range_generic(&dev_priv->ggtt.base.mm,
> - node, size, 4096,
> - I915_COLOR_UNEVICTABLE,
> - start, end, search_flag,
> - alloc_flag);
> - if (ret) {
> - ret = i915_gem_evict_something(&dev_priv->ggtt.base,
> - size, 4096,
> - I915_COLOR_UNEVICTABLE,
> - start, end, 0);
> - if (ret == 0 && ++retried < 3)
> - goto search_again;
> -
> - gvt_err("fail to alloc %s gm space from host, retried %d\n",
> - high_gm ? "high" : "low", retried);
> - }
> + ret = i915_gem_gtt_insert(&dev_priv->ggtt.base, node,
> + size, 4096, I915_COLOR_UNEVICTABLE,
> + start, end, flags);
> mutex_unlock(&dev_priv->drm.struct_mutex);
> + if (ret)
> + gvt_err("fail to alloc %s gm space from host\n",
> + high_gm ? "high" : "low");
> +
> return ret;
> }
>
> diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c
> b/drivers/gpu/drm/i915/i915_gem_gtt.c
> index 6af9311f72f5..c8f1675852a7 100644
> --- a/drivers/gpu/drm/i915/i915_gem_gtt.c
> +++ b/drivers/gpu/drm/i915/i915_gem_gtt.c
> @@ -2044,7 +2044,6 @@ static int gen6_ppgtt_allocate_page_directories(struct
> i915_hw_ppgtt *ppgtt)
> struct i915_address_space *vm = &ppgtt->base;
> struct drm_i915_private *dev_priv = ppgtt->base.i915;
> struct i915_ggtt *ggtt = &dev_priv->ggtt;
> - bool retried = false;
> int ret;
>
> /* PPGTT PDEs reside in the GGTT and consists of 512 entries. The
> @@ -2057,29 +2056,14 @@ static int
> gen6_ppgtt_allocate_page_directories(struct i915_hw_ppgtt *ppgtt)
> if (ret)
> return ret;
>
> -alloc:
> - ret = drm_mm_insert_node_in_range_generic(&ggtt->base.mm, &ppgtt->node,
> - GEN6_PD_SIZE, GEN6_PD_ALIGN,
> - I915_COLOR_UNEVICTABLE,
> - 0, ggtt->base.total,
> - DRM_MM_TOPDOWN);
> - if (ret == -ENOSPC && !retried) {
> - ret = i915_gem_evict_something(&ggtt->base,
> - GEN6_PD_SIZE, GEN6_PD_ALIGN,
> - I915_COLOR_UNEVICTABLE,
> - 0, ggtt->base.total,
> - 0);
> - if (ret)
> - goto err_out;
> -
> - retried = true;
> - goto alloc;
> - }
> -
> + ret = i915_gem_gtt_insert(&ggtt->base, &ppgtt->node,
> + GEN6_PD_SIZE, GEN6_PD_ALIGN,
> + I915_COLOR_UNEVICTABLE,
> + 0, ggtt->base.total,
> + PIN_HIGH);
> if (ret)
> goto err_out;
>
> -
> if (ppgtt->node.start < ggtt->mappable_end)
> DRM_DEBUG("Forced to use aperture for PDEs\n");
>
> @@ -3553,3 +3537,49 @@ i915_get_ggtt_vma_pages(struct i915_vma *vma)
> return ret;
> }
>
> +int i915_gem_gtt_insert(struct i915_address_space *vm,
> + struct drm_mm_node *node,
> + u64 size, u64 alignment, unsigned long color,
> + u64 start, u64 end, unsigned int flags)
> +{
> + u32 search_flag, alloc_flag;
> + int err;
> +
> + lockdep_assert_held(&vm->i915->drm.struct_mutex);
> +
> + if (flags & PIN_HIGH) {
> + search_flag = DRM_MM_SEARCH_BELOW;
> + alloc_flag = DRM_MM_CREATE_TOP;
> + } else {
> + search_flag = DRM_MM_SEARCH_DEFAULT;
> + alloc_flag = DRM_MM_CREATE_DEFAULT;
> + }
> +
> + /* We only allocate in PAGE_SIZE/GTT_PAGE_SIZE (4096) chunks,
> + * so we know that we always have a minimum alignment of 4096.
> + * The drm_mm range manager is optimised to return results
> + * with zero alignment, so where possible use the optimal
> + * path.
> + */
> + GEM_BUG_ON(size & 4095);
> + if (alignment <= 4096)
> + alignment = 0;
> +
> + err = drm_mm_insert_node_in_range_generic(&vm->mm, node,
> + size, alignment, color,
> + start, end,
> + search_flag, alloc_flag);
> + if (err != -ENOSPC)
> + return err;
> +
> + err = i915_gem_evict_something(vm, size, alignment, color,
> + start, end, flags);
> + if (err)
> + return err;
> +
> + search_flag = DRM_MM_SEARCH_DEFAULT;
> + return drm_mm_insert_node_in_range_generic(&vm->mm, node,
> + size, alignment, color,
> + start, end,
> + search_flag, alloc_flag);
> +}
> diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.h
> b/drivers/gpu/drm/i915/i915_gem_gtt.h
> index 0055b8567a43..4c7bef07e38a 100644
> --- a/drivers/gpu/drm/i915/i915_gem_gtt.h
> +++ b/drivers/gpu/drm/i915/i915_gem_gtt.h
> @@ -528,6 +528,11 @@ int __must_check i915_gem_gtt_prepare_pages(struct
> drm_i915_gem_object *obj,
> void i915_gem_gtt_finish_pages(struct drm_i915_gem_object *obj,
> struct sg_table *pages);
>
> +int i915_gem_gtt_insert(struct i915_address_space *vm,
> + struct drm_mm_node *node,
> + u64 size, u64 alignment, unsigned long color,
> + u64 start, u64 end, unsigned int flags);
> +
> /* Flags used by pin/bind&friends. */
> #define PIN_NONBLOCK BIT(0)
> #define PIN_MAPPABLE BIT(1)
> diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
> index fd75d5704287..608008d2d999 100644
> --- a/drivers/gpu/drm/i915/i915_vma.c
> +++ b/drivers/gpu/drm/i915/i915_vma.c
> @@ -415,43 +415,11 @@ i915_vma_insert(struct i915_vma *vma, u64 size, u64
> alignment, u64 flags)
> goto err_unpin;
> }
> } else {
> - u32 search_flag, alloc_flag;
> -
> - if (flags & PIN_HIGH) {
> - search_flag = DRM_MM_SEARCH_BELOW;
> - alloc_flag = DRM_MM_CREATE_TOP;
> - } else {
> - search_flag = DRM_MM_SEARCH_DEFAULT;
> - alloc_flag = DRM_MM_CREATE_DEFAULT;
> - }
> -
> - /* We only allocate in PAGE_SIZE/GTT_PAGE_SIZE (4096) chunks,
> - * so we know that we always have a minimum alignment of 4096.
> - * The drm_mm range manager is optimised to return results
> - * with zero alignment, so where possible use the optimal
> - * path.
> - */
> - if (alignment <= 4096)
> - alignment = 0;
> -
> -search_free:
> - ret = drm_mm_insert_node_in_range_generic(&vma->vm->mm,
> - &vma->node,
> - size, alignment,
> - obj->cache_level,
> - start, end,
> - search_flag,
> - alloc_flag);
> - if (ret) {
> - ret = i915_gem_evict_something(vma->vm, size, alignment,
> - obj->cache_level,
> - start, end,
> - flags);
> - if (ret == 0)
> - goto search_free;
> -
> + ret = i915_gem_gtt_insert(vma->vm, &vma->node,
> + size, alignment, obj->cache_level,
> + start, end, flags);
> + if (ret)
> goto err_unpin;
> - }
>
> GEM_BUG_ON(vma->node.start < start);
> GEM_BUG_ON(vma->node.start + vma->node.size > end);
> --
> 2.11.0
>
> _______________________________________________
> Intel-gfx mailing list
> Intel-gfx at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/intel-gfx
--
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
