From: Chris Wilson Date: Wed, 11 Jan 2017 11:23:10 +0000 (+0000) Subject: drm/i915: Use the MRU stack search after evicting X-Git-Url: https://git.karo-electronics.de/?a=commitdiff_plain;h=e007b19d7ba7424735fd4f17a355b145ae153e4c;p=linux-beck.git drm/i915: Use the MRU stack search after evicting 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 v2: Pull together the 3 identical implements (ahem, a couple were outdated) into a common routine for allocating a node and evicting as necessary. v3: Detect invalid calls to i915_gem_gtt_insert() v4: kerneldoc Signed-off-by: Chris Wilson Reviewed-by: Joonas Lahtinen Link: http://patchwork.freedesktop.org/patch/msgid/20170111112312.31493-1-chris@chris-wilson.co.uk --- 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 8aca11f5f446..f0f17b2b7215 100644 --- a/drivers/gpu/drm/i915/i915_gem_gtt.c +++ b/drivers/gpu/drm/i915/i915_gem_gtt.c @@ -23,10 +23,13 @@ * */ +#include #include #include + #include #include + #include "i915_drv.h" #include "i915_vgpu.h" #include "i915_trace.h" @@ -2032,7 +2035,6 @@ 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 @@ -2045,29 +2047,14 @@ 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"); @@ -3567,3 +3554,95 @@ i915_get_ggtt_vma_pages(struct i915_vma *vma) return ret; } +/** + * i915_gem_gtt_insert - insert a node into an address_space (GTT) + * @vm - the &struct i915_address_space + * @node - the &struct drm_mm_node (typically i915_vma.node) + * @size - how much space to allocate inside the GTT, + * must be #I915_GTT_PAGE_SIZE aligned + * @alignment - required alignment of starting offset, may be 0 but + * if specified, this must be a power-of-two and at least + * #I915_GTT_MIN_ALIGNMENT + * @color - color to apply to node + * @start - start of any range restriction inside GTT (0 for all), + * must be #I915_GTT_PAGE_SIZE aligned + * @end - end of any range restriction inside GTT (U64_MAX for all), + * must be #I915_GTT_PAGE_SIZE aligned if not U64_MAX + * @flags - control search and eviction behaviour + * + * i915_gem_gtt_insert() first searches for an available hole into which + * is can insert the node. The hole address is aligned to @alignment and + * its @size must then fit entirely within the [@start, @end] bounds. The + * nodes on either side of the hole must match @color, or else a guard page + * will be inserted between the two nodes (or the node evicted). If no + * suitable hole is found, then the LRU list of objects within the GTT + * is scanned to find the first set of replacement nodes to create the hole. + * Those old overlapping nodes are evicted from the GTT (and so must be + * rebound before any future use). Any node that is currently pinned cannot + * be evicted (see i915_vma_pin()). Similar if the node's VMA is currently + * active and #PIN_NONBLOCK is specified, that node is also skipped when + * searching for an eviction candidate. See i915_gem_evict_something() for + * the gory details on the eviction algorithm. + * + * Returns: 0 on success, -ENOSPC if no suitable hole is found, -EINTR if + * asked to wait for eviction and interrupted. + */ +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); + GEM_BUG_ON(!size); + GEM_BUG_ON(!IS_ALIGNED(size, I915_GTT_PAGE_SIZE)); + GEM_BUG_ON(alignment && !is_power_of_2(alignment)); + GEM_BUG_ON(alignment && !IS_ALIGNED(alignment, I915_GTT_MIN_ALIGNMENT)); + GEM_BUG_ON(start >= end); + GEM_BUG_ON(start > 0 && !IS_ALIGNED(start, I915_GTT_PAGE_SIZE)); + GEM_BUG_ON(end < U64_MAX && !IS_ALIGNED(end, I915_GTT_PAGE_SIZE)); + + if (unlikely(range_overflows(start, size, end))) + return -ENOSPC; + + if (unlikely(round_up(start, alignment) > round_down(end - size, alignment))) + return -ENOSPC; + + 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. + */ + BUILD_BUG_ON(I915_GTT_MIN_ALIGNMENT > I915_GTT_PAGE_SIZE); + if (alignment <= I915_GTT_MIN_ALIGNMENT) + 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 34a4fd560fa2..79198352a491 100644 --- a/drivers/gpu/drm/i915/i915_gem_gtt.h +++ b/drivers/gpu/drm/i915/i915_gem_gtt.h @@ -532,6 +532,11 @@ 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 490914f89663..df3750d4c907 100644 --- a/drivers/gpu/drm/i915/i915_vma.c +++ b/drivers/gpu/drm/i915/i915_vma.c @@ -431,43 +431,11 @@ 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 <= I915_GTT_MIN_ALIGNMENT) - 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);