drm/i915: Replace execbuf vma ht with an idr
This was the competing idea long ago, but it was only with the rewrite
of the idr as an radixtree and using the radixtree directly ourselves,
along with the realisation that we can store the vma directly in the
radixtree and only need a list for the reverse mapping, that made the
patch performant enough to displace using a hashtable. Though the vma ht
is fast and doesn't require any extra allocation (as we can embed the node
inside the vma), it does require a thread for resizing and serialization
and will have the occasional slow lookup. That is hairy enough to
investigate alternatives and favour them if equivalent in peak performance.
One advantage of allocating an indirection entry is that we can support a
single shared bo between many clients, something that was done on a
first-come first-serve basis for shared GGTT vma previously. To offset
the extra allocations, we create yet another kmem_cache for them.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20170816085210.4199-5-chris@chris-wilson.co.uk
diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c
index 329fb36..48572b1 100644
--- a/drivers/gpu/drm/i915/i915_debugfs.c
+++ b/drivers/gpu/drm/i915/i915_debugfs.c
@@ -1954,12 +1954,6 @@
seq_putc(m, '\n');
}
- seq_printf(m,
- "\tvma hashtable size=%u (actual %lu), count=%u\n",
- ctx->vma_lut.ht_size,
- BIT(ctx->vma_lut.ht_bits),
- ctx->vma_lut.ht_count);
-
seq_putc(m, '\n');
}
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index a4ce8fb..5b70bb9 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -2156,6 +2156,7 @@
struct kmem_cache *objects;
struct kmem_cache *vmas;
+ struct kmem_cache *luts;
struct kmem_cache *requests;
struct kmem_cache *dependencies;
struct kmem_cache *priorities;
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 5a3f3bb..b9e8e0d 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -3242,25 +3242,33 @@
void i915_gem_close_object(struct drm_gem_object *gem, struct drm_file *file)
{
+ struct drm_i915_private *i915 = to_i915(gem->dev);
struct drm_i915_gem_object *obj = to_intel_bo(gem);
struct drm_i915_file_private *fpriv = file->driver_priv;
- struct i915_vma *vma, *vn;
+ struct i915_lut_handle *lut, *ln;
- mutex_lock(&obj->base.dev->struct_mutex);
- list_for_each_entry_safe(vma, vn, &obj->vma_list, obj_link)
- if (vma->vm->file == fpriv)
+ mutex_lock(&i915->drm.struct_mutex);
+
+ list_for_each_entry_safe(lut, ln, &obj->lut_list, obj_link) {
+ struct i915_gem_context *ctx = lut->ctx;
+ struct i915_vma *vma;
+
+ if (ctx->file_priv != fpriv)
+ continue;
+
+ vma = radix_tree_delete(&ctx->handles_vma, lut->handle);
+
+ if (!i915_vma_is_ggtt(vma))
i915_vma_close(vma);
- vma = obj->vma_hashed;
- if (vma && vma->ctx->file_priv == fpriv)
- i915_vma_unlink_ctx(vma);
+ list_del(&lut->obj_link);
+ list_del(&lut->ctx_link);
- if (i915_gem_object_is_active(obj) &&
- !i915_gem_object_has_active_reference(obj)) {
- i915_gem_object_set_active_reference(obj);
- i915_gem_object_get(obj);
+ kmem_cache_free(i915->luts, lut);
+ __i915_gem_object_release_unless_active(obj);
}
- mutex_unlock(&obj->base.dev->struct_mutex);
+
+ mutex_unlock(&i915->drm.struct_mutex);
}
static unsigned long to_wait_timeout(s64 timeout_ns)
@@ -4252,6 +4260,7 @@
INIT_LIST_HEAD(&obj->global_link);
INIT_LIST_HEAD(&obj->userfault_link);
INIT_LIST_HEAD(&obj->vma_list);
+ INIT_LIST_HEAD(&obj->lut_list);
INIT_LIST_HEAD(&obj->batch_pool_link);
obj->ops = ops;
@@ -4495,8 +4504,8 @@
{
lockdep_assert_held(&obj->base.dev->struct_mutex);
- GEM_BUG_ON(i915_gem_object_has_active_reference(obj));
- if (i915_gem_object_is_active(obj))
+ if (!i915_gem_object_has_active_reference(obj) &&
+ i915_gem_object_is_active(obj))
i915_gem_object_set_active_reference(obj);
else
i915_gem_object_put(obj);
@@ -4888,12 +4897,16 @@
if (!dev_priv->vmas)
goto err_objects;
+ dev_priv->luts = KMEM_CACHE(i915_lut_handle, 0);
+ if (!dev_priv->luts)
+ goto err_vmas;
+
dev_priv->requests = KMEM_CACHE(drm_i915_gem_request,
SLAB_HWCACHE_ALIGN |
SLAB_RECLAIM_ACCOUNT |
SLAB_TYPESAFE_BY_RCU);
if (!dev_priv->requests)
- goto err_vmas;
+ goto err_luts;
dev_priv->dependencies = KMEM_CACHE(i915_dependency,
SLAB_HWCACHE_ALIGN |
@@ -4937,6 +4950,8 @@
kmem_cache_destroy(dev_priv->dependencies);
err_requests:
kmem_cache_destroy(dev_priv->requests);
+err_luts:
+ kmem_cache_destroy(dev_priv->luts);
err_vmas:
kmem_cache_destroy(dev_priv->vmas);
err_objects:
@@ -4959,6 +4974,7 @@
kmem_cache_destroy(dev_priv->priorities);
kmem_cache_destroy(dev_priv->dependencies);
kmem_cache_destroy(dev_priv->requests);
+ kmem_cache_destroy(dev_priv->luts);
kmem_cache_destroy(dev_priv->vmas);
kmem_cache_destroy(dev_priv->objects);
diff --git a/drivers/gpu/drm/i915/i915_gem_context.c b/drivers/gpu/drm/i915/i915_gem_context.c
index 86ac74a..58a2a44 100644
--- a/drivers/gpu/drm/i915/i915_gem_context.c
+++ b/drivers/gpu/drm/i915/i915_gem_context.c
@@ -93,69 +93,28 @@
#define ALL_L3_SLICES(dev) (1 << NUM_L3_SLICES(dev)) - 1
-/* Initial size (as log2) to preallocate the handle->object hashtable */
-#define VMA_HT_BITS 2u /* 4 x 2 pointers, 64 bytes minimum */
-
-static void resize_vma_ht(struct work_struct *work)
+static void lut_close(struct i915_gem_context *ctx)
{
- struct i915_gem_context_vma_lut *lut =
- container_of(work, typeof(*lut), resize);
- unsigned int bits, new_bits, size, i;
- struct hlist_head *new_ht;
+ struct i915_lut_handle *lut, *ln;
+ struct radix_tree_iter iter;
+ void __rcu **slot;
- GEM_BUG_ON(!(lut->ht_size & I915_CTX_RESIZE_IN_PROGRESS));
-
- bits = 1 + ilog2(4*lut->ht_count/3 + 1);
- new_bits = min_t(unsigned int,
- max(bits, VMA_HT_BITS),
- sizeof(unsigned int) * BITS_PER_BYTE - 1);
- if (new_bits == lut->ht_bits)
- goto out;
-
- new_ht = kzalloc(sizeof(*new_ht)<<new_bits, GFP_KERNEL | __GFP_NOWARN);
- if (!new_ht)
- new_ht = vzalloc(sizeof(*new_ht)<<new_bits);
- if (!new_ht)
- /* Pretend resize succeeded and stop calling us for a bit! */
- goto out;
-
- size = BIT(lut->ht_bits);
- for (i = 0; i < size; i++) {
- struct i915_vma *vma;
- struct hlist_node *tmp;
-
- hlist_for_each_entry_safe(vma, tmp, &lut->ht[i], ctx_node)
- hlist_add_head(&vma->ctx_node,
- &new_ht[hash_32(vma->ctx_handle,
- new_bits)]);
+ list_for_each_entry_safe(lut, ln, &ctx->handles_list, ctx_link) {
+ list_del(&lut->obj_link);
+ kmem_cache_free(ctx->i915->luts, lut);
}
- kvfree(lut->ht);
- lut->ht = new_ht;
- lut->ht_bits = new_bits;
-out:
- smp_store_release(&lut->ht_size, BIT(bits));
- GEM_BUG_ON(lut->ht_size & I915_CTX_RESIZE_IN_PROGRESS);
-}
-static void vma_lut_free(struct i915_gem_context *ctx)
-{
- struct i915_gem_context_vma_lut *lut = &ctx->vma_lut;
- unsigned int i, size;
+ radix_tree_for_each_slot(slot, &ctx->handles_vma, &iter, 0) {
+ struct i915_vma *vma = rcu_dereference_raw(*slot);
+ struct drm_i915_gem_object *obj = vma->obj;
- if (lut->ht_size & I915_CTX_RESIZE_IN_PROGRESS)
- cancel_work_sync(&lut->resize);
+ radix_tree_iter_delete(&ctx->handles_vma, &iter, slot);
- size = BIT(lut->ht_bits);
- for (i = 0; i < size; i++) {
- struct i915_vma *vma;
+ if (!i915_vma_is_ggtt(vma))
+ i915_vma_close(vma);
- hlist_for_each_entry(vma, &lut->ht[i], ctx_node) {
- vma->obj->vma_hashed = NULL;
- vma->ctx = NULL;
- i915_vma_put(vma);
- }
+ __i915_gem_object_release_unless_active(obj);
}
- kvfree(lut->ht);
}
static void i915_gem_context_free(struct i915_gem_context *ctx)
@@ -165,7 +124,6 @@
lockdep_assert_held(&ctx->i915->drm.struct_mutex);
GEM_BUG_ON(!i915_gem_context_is_closed(ctx));
- vma_lut_free(ctx);
i915_ppgtt_put(ctx->ppgtt);
for (i = 0; i < I915_NUM_ENGINES; i++) {
@@ -239,8 +197,11 @@
static void context_close(struct i915_gem_context *ctx)
{
i915_gem_context_set_closed(ctx);
+
+ lut_close(ctx);
if (ctx->ppgtt)
i915_ppgtt_close(&ctx->ppgtt->base);
+
ctx->file_priv = ERR_PTR(-EBADF);
i915_gem_context_put(ctx);
}
@@ -313,16 +274,8 @@
ctx->i915 = dev_priv;
ctx->priority = I915_PRIORITY_NORMAL;
- ctx->vma_lut.ht_bits = VMA_HT_BITS;
- ctx->vma_lut.ht_size = BIT(VMA_HT_BITS);
- BUILD_BUG_ON(BIT(VMA_HT_BITS) == I915_CTX_RESIZE_IN_PROGRESS);
- ctx->vma_lut.ht = kcalloc(ctx->vma_lut.ht_size,
- sizeof(*ctx->vma_lut.ht),
- GFP_KERNEL);
- if (!ctx->vma_lut.ht)
- goto err_out;
-
- INIT_WORK(&ctx->vma_lut.resize, resize_vma_ht);
+ INIT_RADIX_TREE(&ctx->handles_vma, GFP_KERNEL);
+ INIT_LIST_HEAD(&ctx->handles_list);
/* Default context will never have a file_priv */
ret = DEFAULT_CONTEXT_HANDLE;
@@ -372,8 +325,6 @@
put_pid(ctx->pid);
idr_remove(&file_priv->context_idr, ctx->user_handle);
err_lut:
- kvfree(ctx->vma_lut.ht);
-err_out:
context_close(ctx);
return ERR_PTR(ret);
}
diff --git a/drivers/gpu/drm/i915/i915_gem_context.h b/drivers/gpu/drm/i915/i915_gem_context.h
index 2d02918..44688e2 100644
--- a/drivers/gpu/drm/i915/i915_gem_context.h
+++ b/drivers/gpu/drm/i915/i915_gem_context.h
@@ -27,6 +27,7 @@
#include <linux/bitops.h>
#include <linux/list.h>
+#include <linux/radix-tree.h>
struct pid;
@@ -149,32 +150,6 @@
/** ggtt_offset_bias: placement restriction for context objects */
u32 ggtt_offset_bias;
- struct i915_gem_context_vma_lut {
- /** ht_size: last request size to allocate the hashtable for. */
- unsigned int ht_size;
-#define I915_CTX_RESIZE_IN_PROGRESS BIT(0)
- /** ht_bits: real log2(size) of hashtable. */
- unsigned int ht_bits;
- /** ht_count: current number of entries inside the hashtable */
- unsigned int ht_count;
-
- /** ht: the array of buckets comprising the simple hashtable */
- struct hlist_head *ht;
-
- /**
- * resize: After an execbuf completes, we check the load factor
- * of the hashtable. If the hashtable is too full, or too empty,
- * we schedule a task to resize the hashtable. During the
- * resize, the entries are moved between different buckets and
- * so we cannot simultaneously read the hashtable as it is
- * being resized (unlike rhashtable). Therefore we treat the
- * active work as a strong barrier, pausing a subsequent
- * execbuf to wait for the resize worker to complete, if
- * required.
- */
- struct work_struct resize;
- } vma_lut;
-
/** engine: per-engine logical HW state */
struct intel_context {
struct i915_vma *state;
@@ -205,6 +180,18 @@
/** remap_slice: Bitmask of cache lines that need remapping */
u8 remap_slice;
+
+ /** handles_vma: rbtree to look up our context specific obj/vma for
+ * the user handle. (user handles are per fd, but the binding is
+ * per vm, which may be one per context or shared with the global GTT)
+ */
+ struct radix_tree_root handles_vma;
+
+ /** handles_list: reverse list of all the rbtree entries in use for
+ * this context, which allows us to free all the allocations on
+ * context close.
+ */
+ struct list_head handles_list;
};
static inline bool i915_gem_context_is_closed(const struct i915_gem_context *ctx)
diff --git a/drivers/gpu/drm/i915/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
index 95e4612..9f1057c 100644
--- a/drivers/gpu/drm/i915/i915_gem_execbuffer.c
+++ b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
@@ -450,9 +450,7 @@
}
static int
-eb_add_vma(struct i915_execbuffer *eb,
- unsigned int i, struct i915_vma *vma,
- unsigned int flags)
+eb_add_vma(struct i915_execbuffer *eb, unsigned int i, struct i915_vma *vma)
{
struct drm_i915_gem_exec_object2 *entry = &eb->exec[i];
int err;
@@ -482,7 +480,7 @@
* to find the right target VMA when doing relocations.
*/
eb->vma[i] = vma;
- eb->flags[i] = entry->flags | flags;
+ eb->flags[i] = entry->flags;
vma->exec_flags = &eb->flags[i];
err = 0;
@@ -647,19 +645,6 @@
} while (1);
}
-static inline struct hlist_head *
-ht_head(const struct i915_gem_context_vma_lut *lut, u32 handle)
-{
- return &lut->ht[hash_32(handle, lut->ht_bits)];
-}
-
-static inline bool
-ht_needs_resize(const struct i915_gem_context_vma_lut *lut)
-{
- return (4*lut->ht_count > 3*lut->ht_size ||
- 4*lut->ht_count + 1 < lut->ht_size);
-}
-
static unsigned int eb_batch_index(const struct i915_execbuffer *eb)
{
if (eb->args->flags & I915_EXEC_BATCH_FIRST)
@@ -688,7 +673,7 @@
static int eb_lookup_vmas(struct i915_execbuffer *eb)
{
- struct i915_gem_context_vma_lut *lut = &eb->ctx->vma_lut;
+ struct radix_tree_root *handles_vma = &eb->ctx->handles_vma;
struct drm_i915_gem_object *uninitialized_var(obj);
unsigned int i;
int err;
@@ -702,24 +687,14 @@
INIT_LIST_HEAD(&eb->relocs);
INIT_LIST_HEAD(&eb->unbound);
- if (unlikely(lut->ht_size & I915_CTX_RESIZE_IN_PROGRESS))
- flush_work(&lut->resize);
- GEM_BUG_ON(lut->ht_size & I915_CTX_RESIZE_IN_PROGRESS);
-
for (i = 0; i < eb->buffer_count; i++) {
u32 handle = eb->exec[i].handle;
- struct hlist_head *hl = ht_head(lut, handle);
- unsigned int flags = 0;
+ struct i915_lut_handle *lut;
struct i915_vma *vma;
- hlist_for_each_entry(vma, hl, ctx_node) {
- GEM_BUG_ON(vma->ctx != eb->ctx);
-
- if (vma->ctx_handle != handle)
- continue;
-
+ vma = radix_tree_lookup(handles_vma, handle);
+ if (likely(vma))
goto add_vma;
- }
obj = i915_gem_object_lookup(eb->file, handle);
if (unlikely(!obj)) {
@@ -733,26 +708,28 @@
goto err_obj;
}
- /* First come, first served */
- if (!vma->ctx) {
- vma->ctx = eb->ctx;
- vma->ctx_handle = handle;
- hlist_add_head(&vma->ctx_node, hl);
- lut->ht_count++;
- lut->ht_size |= I915_CTX_RESIZE_IN_PROGRESS;
- if (i915_vma_is_ggtt(vma)) {
- GEM_BUG_ON(obj->vma_hashed);
- obj->vma_hashed = vma;
- }
-
- /* transfer ref to ctx */
- obj = NULL;
- } else {
- flags = __EXEC_OBJECT_HAS_REF;
+ lut = kmem_cache_alloc(eb->i915->luts, GFP_KERNEL);
+ if (unlikely(!lut)) {
+ err = -ENOMEM;
+ goto err_obj;
}
+ err = radix_tree_insert(handles_vma, handle, vma);
+ if (unlikely(err)) {
+ kfree(lut);
+ goto err_obj;
+ }
+
+ list_add(&lut->obj_link, &obj->lut_list);
+ list_add(&lut->ctx_link, &eb->ctx->handles_list);
+ lut->ctx = eb->ctx;
+ lut->handle = handle;
+
+ /* transfer ref to ctx */
+ obj = NULL;
+
add_vma:
- err = eb_add_vma(eb, i, vma, flags);
+ err = eb_add_vma(eb, i, vma);
if (unlikely(err))
goto err_obj;
@@ -760,13 +737,6 @@
GEM_BUG_ON(vma->exec_flags != &eb->flags[i]);
}
- if (lut->ht_size & I915_CTX_RESIZE_IN_PROGRESS) {
- if (ht_needs_resize(lut))
- queue_work(system_highpri_wq, &lut->resize);
- else
- lut->ht_size &= ~I915_CTX_RESIZE_IN_PROGRESS;
- }
-
/* take note of the batch buffer before we might reorder the lists */
i = eb_batch_index(eb);
eb->batch = eb->vma[i];
@@ -794,7 +764,6 @@
i915_gem_object_put(obj);
err_vma:
eb->vma[i] = NULL;
- lut->ht_size &= ~I915_CTX_RESIZE_IN_PROGRESS;
return err;
}
diff --git a/drivers/gpu/drm/i915/i915_gem_object.h b/drivers/gpu/drm/i915/i915_gem_object.h
index 3baa341..c30d8f8 100644
--- a/drivers/gpu/drm/i915/i915_gem_object.h
+++ b/drivers/gpu/drm/i915/i915_gem_object.h
@@ -38,6 +38,19 @@
struct drm_i915_gem_object;
+/*
+ * struct i915_lut_handle tracks the fast lookups from handle to vma used
+ * for execbuf. Although we use a radixtree for that mapping, in order to
+ * remove them as the object or context is closed, we need a secondary list
+ * and a translation entry (i915_lut_handle).
+ */
+struct i915_lut_handle {
+ struct list_head obj_link;
+ struct list_head ctx_link;
+ struct i915_gem_context *ctx;
+ u32 handle;
+};
+
struct drm_i915_gem_object_ops {
unsigned int flags;
#define I915_GEM_OBJECT_HAS_STRUCT_PAGE BIT(0)
@@ -89,7 +102,15 @@
* They are also added to @vma_list for easy iteration.
*/
struct rb_root vma_tree;
- struct i915_vma *vma_hashed;
+
+ /**
+ * @lut_list: List of vma lookup entries in use for this object.
+ *
+ * If this object is closed, we need to remove all of its VMA from
+ * the fast lookup index in associated contexts; @lut_list provides
+ * this translation from object to context->handles_vma.
+ */
+ struct list_head lut_list;
/** Stolen memory for this object, instead of being backed by shmem. */
struct drm_mm_node *stolen;
diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
index 958be0a..02d1a5e 100644
--- a/drivers/gpu/drm/i915/i915_vma.c
+++ b/drivers/gpu/drm/i915/i915_vma.c
@@ -597,33 +597,11 @@
kmem_cache_free(to_i915(vma->obj->base.dev)->vmas, vma);
}
-void i915_vma_unlink_ctx(struct i915_vma *vma)
-{
- struct i915_gem_context *ctx = vma->ctx;
-
- if (ctx->vma_lut.ht_size & I915_CTX_RESIZE_IN_PROGRESS) {
- cancel_work_sync(&ctx->vma_lut.resize);
- ctx->vma_lut.ht_size &= ~I915_CTX_RESIZE_IN_PROGRESS;
- }
-
- __hlist_del(&vma->ctx_node);
- ctx->vma_lut.ht_count--;
-
- if (i915_vma_is_ggtt(vma))
- vma->obj->vma_hashed = NULL;
- vma->ctx = NULL;
-
- i915_vma_put(vma);
-}
-
void i915_vma_close(struct i915_vma *vma)
{
GEM_BUG_ON(i915_vma_is_closed(vma));
vma->flags |= I915_VMA_CLOSED;
- if (vma->ctx)
- i915_vma_unlink_ctx(vma);
-
list_del(&vma->obj_link);
rb_erase(&vma->obj_node, &vma->obj->vma_tree);
diff --git a/drivers/gpu/drm/i915/i915_vma.h b/drivers/gpu/drm/i915/i915_vma.h
index 5c49506..1fd61e8 100644
--- a/drivers/gpu/drm/i915/i915_vma.h
+++ b/drivers/gpu/drm/i915/i915_vma.h
@@ -115,10 +115,6 @@
unsigned int *exec_flags;
struct hlist_node exec_node;
u32 exec_handle;
-
- struct i915_gem_context *ctx;
- struct hlist_node ctx_node;
- u32 ctx_handle;
};
struct i915_vma *
diff --git a/drivers/gpu/drm/i915/selftests/mock_context.c b/drivers/gpu/drm/i915/selftests/mock_context.c
index d436f2d..098ce64 100644
--- a/drivers/gpu/drm/i915/selftests/mock_context.c
+++ b/drivers/gpu/drm/i915/selftests/mock_context.c
@@ -40,18 +40,13 @@
INIT_LIST_HEAD(&ctx->link);
ctx->i915 = i915;
- ctx->vma_lut.ht_bits = VMA_HT_BITS;
- ctx->vma_lut.ht_size = BIT(VMA_HT_BITS);
- ctx->vma_lut.ht = kcalloc(ctx->vma_lut.ht_size,
- sizeof(*ctx->vma_lut.ht),
- GFP_KERNEL);
- if (!ctx->vma_lut.ht)
- goto err_free;
+ INIT_RADIX_TREE(&ctx->handles_vma, GFP_KERNEL);
+ INIT_LIST_HEAD(&ctx->handles_list);
ret = ida_simple_get(&i915->contexts.hw_ida,
0, MAX_CONTEXT_HW_ID, GFP_KERNEL);
if (ret < 0)
- goto err_vma_ht;
+ goto err_handles;
ctx->hw_id = ret;
if (name) {
@@ -66,9 +61,7 @@
return ctx;
-err_vma_ht:
- kvfree(ctx->vma_lut.ht);
-err_free:
+err_handles:
kfree(ctx);
return NULL;