* [PATCH 3/8] drm: implement top-down allocation method
@ 2021-10-25 13:00 ` Arunpravin
0 siblings, 0 replies; 58+ messages in thread
From: Arunpravin @ 2021-10-25 13:00 UTC (permalink / raw)
To: dri-devel, intel-gfx, amd-gfx
Cc: daniel, christian.koenig, alexander.deucher, matthew.auld,
tzimmermann, jani.nikula, Arunpravin
Implemented a function which walk through the order list,
compares the offset and returns the maximum offset block,
this method is unpredictable in obtaining the high range
address blocks which depends on allocation and deallocation.
for instance, if driver requests address at a low specific
range, allocator traverses from the root block and splits
the larger blocks until it reaches the specific block and
in the process of splitting, lower orders in the freelist
are occupied with low range address blocks and for the
subsequent TOPDOWN memory request we may return the low
range blocks.To overcome this issue, we may go with the
below approach.
The other approach, sorting each order list entries in
ascending order and compares the last entry of each
order list in the freelist and return the max block.
This creates sorting overhead on every drm_buddy_free()
request and split up of larger blocks for a single page
request.
Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
---
drivers/gpu/drm/drm_buddy.c | 42 +++++++++++++++++++++++++++++++------
include/drm/drm_buddy.h | 1 +
2 files changed, 37 insertions(+), 6 deletions(-)
diff --git a/drivers/gpu/drm/drm_buddy.c b/drivers/gpu/drm/drm_buddy.c
index 406e3d521903..9d3547bcc5da 100644
--- a/drivers/gpu/drm/drm_buddy.c
+++ b/drivers/gpu/drm/drm_buddy.c
@@ -362,6 +362,27 @@ alloc_range(struct drm_buddy_mm *mm,
return ERR_PTR(err);
}
+static struct drm_buddy_block *
+get_maxblock(struct list_head *head)
+{
+ struct drm_buddy_block *max_block = NULL, *node;
+
+ max_block = list_first_entry_or_null(head,
+ struct drm_buddy_block,
+ link);
+
+ if (!max_block)
+ return NULL;
+
+ list_for_each_entry(node, head, link) {
+ if (drm_buddy_block_offset(node) >
+ drm_buddy_block_offset(max_block))
+ max_block = node;
+ }
+
+ return max_block;
+}
+
static struct drm_buddy_block *
alloc_from_freelist(struct drm_buddy_mm *mm,
unsigned int order,
@@ -372,13 +393,22 @@ alloc_from_freelist(struct drm_buddy_mm *mm,
int err;
for (i = order; i <= mm->max_order; ++i) {
- if (!list_empty(&mm->free_list[i])) {
- block = list_first_entry_or_null(&mm->free_list[i],
- struct drm_buddy_block,
- link);
+ if (flags & DRM_BUDDY_TOPDOWN_ALLOCATION) {
+ if (!list_empty(&mm->free_list[i])) {
+ block = get_maxblock(&mm->free_list[i]);
- if (block)
- break;
+ if (block)
+ break;
+ }
+ } else {
+ if (!list_empty(&mm->free_list[i])) {
+ block = list_first_entry_or_null(&mm->free_list[i],
+ struct drm_buddy_block,
+ link);
+
+ if (block)
+ break;
+ }
}
}
diff --git a/include/drm/drm_buddy.h b/include/drm/drm_buddy.h
index c7bb5509a7ad..cd8021d2d6e7 100644
--- a/include/drm/drm_buddy.h
+++ b/include/drm/drm_buddy.h
@@ -28,6 +28,7 @@
})
#define DRM_BUDDY_RANGE_ALLOCATION (1 << 0)
+#define DRM_BUDDY_TOPDOWN_ALLOCATION (1 << 1)
struct drm_buddy_block {
#define DRM_BUDDY_HEADER_OFFSET GENMASK_ULL(63, 12)
--
2.25.1
^ permalink raw reply related [flat|nested] 58+ messages in thread
* [Intel-gfx] [PATCH 3/8] drm: implement top-down allocation method
@ 2021-10-25 13:00 ` Arunpravin
0 siblings, 0 replies; 58+ messages in thread
From: Arunpravin @ 2021-10-25 13:00 UTC (permalink / raw)
To: dri-devel, intel-gfx, amd-gfx
Cc: daniel, christian.koenig, alexander.deucher, matthew.auld,
tzimmermann, jani.nikula, Arunpravin
Implemented a function which walk through the order list,
compares the offset and returns the maximum offset block,
this method is unpredictable in obtaining the high range
address blocks which depends on allocation and deallocation.
for instance, if driver requests address at a low specific
range, allocator traverses from the root block and splits
the larger blocks until it reaches the specific block and
in the process of splitting, lower orders in the freelist
are occupied with low range address blocks and for the
subsequent TOPDOWN memory request we may return the low
range blocks.To overcome this issue, we may go with the
below approach.
The other approach, sorting each order list entries in
ascending order and compares the last entry of each
order list in the freelist and return the max block.
This creates sorting overhead on every drm_buddy_free()
request and split up of larger blocks for a single page
request.
Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
---
drivers/gpu/drm/drm_buddy.c | 42 +++++++++++++++++++++++++++++++------
include/drm/drm_buddy.h | 1 +
2 files changed, 37 insertions(+), 6 deletions(-)
diff --git a/drivers/gpu/drm/drm_buddy.c b/drivers/gpu/drm/drm_buddy.c
index 406e3d521903..9d3547bcc5da 100644
--- a/drivers/gpu/drm/drm_buddy.c
+++ b/drivers/gpu/drm/drm_buddy.c
@@ -362,6 +362,27 @@ alloc_range(struct drm_buddy_mm *mm,
return ERR_PTR(err);
}
+static struct drm_buddy_block *
+get_maxblock(struct list_head *head)
+{
+ struct drm_buddy_block *max_block = NULL, *node;
+
+ max_block = list_first_entry_or_null(head,
+ struct drm_buddy_block,
+ link);
+
+ if (!max_block)
+ return NULL;
+
+ list_for_each_entry(node, head, link) {
+ if (drm_buddy_block_offset(node) >
+ drm_buddy_block_offset(max_block))
+ max_block = node;
+ }
+
+ return max_block;
+}
+
static struct drm_buddy_block *
alloc_from_freelist(struct drm_buddy_mm *mm,
unsigned int order,
@@ -372,13 +393,22 @@ alloc_from_freelist(struct drm_buddy_mm *mm,
int err;
for (i = order; i <= mm->max_order; ++i) {
- if (!list_empty(&mm->free_list[i])) {
- block = list_first_entry_or_null(&mm->free_list[i],
- struct drm_buddy_block,
- link);
+ if (flags & DRM_BUDDY_TOPDOWN_ALLOCATION) {
+ if (!list_empty(&mm->free_list[i])) {
+ block = get_maxblock(&mm->free_list[i]);
- if (block)
- break;
+ if (block)
+ break;
+ }
+ } else {
+ if (!list_empty(&mm->free_list[i])) {
+ block = list_first_entry_or_null(&mm->free_list[i],
+ struct drm_buddy_block,
+ link);
+
+ if (block)
+ break;
+ }
}
}
diff --git a/include/drm/drm_buddy.h b/include/drm/drm_buddy.h
index c7bb5509a7ad..cd8021d2d6e7 100644
--- a/include/drm/drm_buddy.h
+++ b/include/drm/drm_buddy.h
@@ -28,6 +28,7 @@
})
#define DRM_BUDDY_RANGE_ALLOCATION (1 << 0)
+#define DRM_BUDDY_TOPDOWN_ALLOCATION (1 << 1)
struct drm_buddy_block {
#define DRM_BUDDY_HEADER_OFFSET GENMASK_ULL(63, 12)
--
2.25.1
^ permalink raw reply related [flat|nested] 58+ messages in thread
* [PATCH 4/8] drm/i915: add top-down alloc support to i915
2021-10-25 13:00 ` [Intel-gfx] " Arunpravin
@ 2021-10-25 13:00 ` Arunpravin
-1 siblings, 0 replies; 58+ messages in thread
From: Arunpravin @ 2021-10-25 13:00 UTC (permalink / raw)
To: dri-devel, intel-gfx, amd-gfx
Cc: daniel, christian.koenig, alexander.deucher, matthew.auld,
tzimmermann, jani.nikula, Arunpravin
add top down allocation support to i915 driver
Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
---
drivers/gpu/drm/i915/i915_ttm_buddy_manager.c | 3 +++
1 file changed, 3 insertions(+)
diff --git a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
index 75197ba6e40d..963468228392 100644
--- a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
+++ b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
@@ -53,6 +53,9 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
INIT_LIST_HEAD(&bman_res->blocks);
bman_res->mm = mm;
+ if (place->flags & TTM_PL_FLAG_TOPDOWN)
+ bman_res->flags |= DRM_BUDDY_TOPDOWN_ALLOCATION;
+
if (place->fpfn || lpfn != man->size)
bman_res->flags |= DRM_BUDDY_RANGE_ALLOCATION;
--
2.25.1
^ permalink raw reply related [flat|nested] 58+ messages in thread
* [Intel-gfx] [PATCH 4/8] drm/i915: add top-down alloc support to i915
@ 2021-10-25 13:00 ` Arunpravin
0 siblings, 0 replies; 58+ messages in thread
From: Arunpravin @ 2021-10-25 13:00 UTC (permalink / raw)
To: dri-devel, intel-gfx, amd-gfx
Cc: daniel, christian.koenig, alexander.deucher, matthew.auld,
tzimmermann, jani.nikula, Arunpravin
add top down allocation support to i915 driver
Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
---
drivers/gpu/drm/i915/i915_ttm_buddy_manager.c | 3 +++
1 file changed, 3 insertions(+)
diff --git a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
index 75197ba6e40d..963468228392 100644
--- a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
+++ b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
@@ -53,6 +53,9 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
INIT_LIST_HEAD(&bman_res->blocks);
bman_res->mm = mm;
+ if (place->flags & TTM_PL_FLAG_TOPDOWN)
+ bman_res->flags |= DRM_BUDDY_TOPDOWN_ALLOCATION;
+
if (place->fpfn || lpfn != man->size)
bman_res->flags |= DRM_BUDDY_RANGE_ALLOCATION;
--
2.25.1
^ permalink raw reply related [flat|nested] 58+ messages in thread
* [PATCH 5/8] drm: Implement method to free unused pages
2021-10-25 13:00 ` [Intel-gfx] " Arunpravin
@ 2021-10-25 13:00 ` Arunpravin
-1 siblings, 0 replies; 58+ messages in thread
From: Arunpravin @ 2021-10-25 13:00 UTC (permalink / raw)
To: dri-devel, intel-gfx, amd-gfx
Cc: daniel, christian.koenig, alexander.deucher, matthew.auld,
tzimmermann, jani.nikula, Arunpravin
On contiguous allocation, we round up the size
to the *next* power of 2, implement a function
to free the unused pages after the newly allocate block.
Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
---
drivers/gpu/drm/drm_buddy.c | 103 ++++++++++++++++++++++++++++++++++++
include/drm/drm_buddy.h | 4 ++
2 files changed, 107 insertions(+)
diff --git a/drivers/gpu/drm/drm_buddy.c b/drivers/gpu/drm/drm_buddy.c
index 9d3547bcc5da..0da8510736eb 100644
--- a/drivers/gpu/drm/drm_buddy.c
+++ b/drivers/gpu/drm/drm_buddy.c
@@ -284,6 +284,109 @@ static inline bool contains(u64 s1, u64 e1, u64 s2, u64 e2)
return s1 <= s2 && e1 >= e2;
}
+/**
+ * drm_buddy_free_unused_pages - free unused pages
+ *
+ * @mm: DRM buddy manager
+ * @actual_size: original size requested
+ * @blocks: output list head to add allocated blocks
+ *
+ * For contiguous allocation, we round up the size to the nearest
+ * power of two value, drivers consume *actual* size, so remaining
+ * portions are unused and it can be freed.
+ *
+ * Returns:
+ * 0 on success, error code on failure.
+ */
+int drm_buddy_free_unused_pages(struct drm_buddy_mm *mm,
+ u64 actual_size,
+ struct list_head *blocks)
+{
+ struct drm_buddy_block *block;
+ struct drm_buddy_block *buddy;
+ u64 actual_start;
+ u64 actual_end;
+ LIST_HEAD(dfs);
+ u64 count = 0;
+ int err;
+
+ if (!list_is_singular(blocks))
+ return -EINVAL;
+
+ block = list_first_entry_or_null(blocks,
+ struct drm_buddy_block,
+ link);
+
+ if (!block)
+ return -EINVAL;
+
+ if (actual_size > drm_buddy_block_size(mm, block))
+ return -EINVAL;
+
+ if (actual_size == drm_buddy_block_size(mm, block))
+ return 0;
+
+ list_del(&block->link);
+
+ actual_start = drm_buddy_block_offset(block);
+ actual_end = actual_start + actual_size - 1;
+
+ if (drm_buddy_block_is_allocated(block))
+ mark_free(mm, block);
+
+ list_add(&block->tmp_link, &dfs);
+
+ while (1) {
+ block = list_first_entry_or_null(&dfs,
+ struct drm_buddy_block,
+ tmp_link);
+
+ if (!block)
+ break;
+
+ list_del(&block->tmp_link);
+
+ if (count == actual_size)
+ return 0;
+
+ if (contains(actual_start, actual_end, drm_buddy_block_offset(block),
+ (drm_buddy_block_offset(block) + drm_buddy_block_size(mm, block) - 1))) {
+ BUG_ON(!drm_buddy_block_is_free(block));
+
+ /* Allocate only required blocks */
+ mark_allocated(block);
+ mm->avail -= drm_buddy_block_size(mm, block);
+ list_add_tail(&block->link, blocks);
+ count += drm_buddy_block_size(mm, block);
+ continue;
+ }
+
+ if (drm_buddy_block_order(block) == 0)
+ continue;
+
+ if (!drm_buddy_block_is_split(block)) {
+ err = split_block(mm, block);
+
+ if (unlikely(err))
+ goto err_undo;
+ }
+
+ list_add(&block->right->tmp_link, &dfs);
+ list_add(&block->left->tmp_link, &dfs);
+ }
+
+ return -ENOSPC;
+
+err_undo:
+ buddy = get_buddy(block);
+ if (buddy &&
+ (drm_buddy_block_is_free(block) &&
+ drm_buddy_block_is_free(buddy)))
+ __drm_buddy_free(mm, block);
+ return err;
+}
+EXPORT_SYMBOL(drm_buddy_free_unused_pages);
+
static struct drm_buddy_block *
alloc_range(struct drm_buddy_mm *mm,
u64 start, u64 end,
diff --git a/include/drm/drm_buddy.h b/include/drm/drm_buddy.h
index cd8021d2d6e7..1dfc80c88e1f 100644
--- a/include/drm/drm_buddy.h
+++ b/include/drm/drm_buddy.h
@@ -145,6 +145,10 @@ int drm_buddy_alloc(struct drm_buddy_mm *mm,
struct list_head *blocks,
unsigned long flags);
+int drm_buddy_free_unused_pages(struct drm_buddy_mm *mm,
+ u64 actual_size,
+ struct list_head *blocks);
+
void drm_buddy_free(struct drm_buddy_mm *mm, struct drm_buddy_block *block);
void drm_buddy_free_list(struct drm_buddy_mm *mm, struct list_head *objects);
--
2.25.1
^ permalink raw reply related [flat|nested] 58+ messages in thread
* [Intel-gfx] [PATCH 5/8] drm: Implement method to free unused pages
@ 2021-10-25 13:00 ` Arunpravin
0 siblings, 0 replies; 58+ messages in thread
From: Arunpravin @ 2021-10-25 13:00 UTC (permalink / raw)
To: dri-devel, intel-gfx, amd-gfx
Cc: daniel, christian.koenig, alexander.deucher, matthew.auld,
tzimmermann, jani.nikula, Arunpravin
On contiguous allocation, we round up the size
to the *next* power of 2, implement a function
to free the unused pages after the newly allocate block.
Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
---
drivers/gpu/drm/drm_buddy.c | 103 ++++++++++++++++++++++++++++++++++++
include/drm/drm_buddy.h | 4 ++
2 files changed, 107 insertions(+)
diff --git a/drivers/gpu/drm/drm_buddy.c b/drivers/gpu/drm/drm_buddy.c
index 9d3547bcc5da..0da8510736eb 100644
--- a/drivers/gpu/drm/drm_buddy.c
+++ b/drivers/gpu/drm/drm_buddy.c
@@ -284,6 +284,109 @@ static inline bool contains(u64 s1, u64 e1, u64 s2, u64 e2)
return s1 <= s2 && e1 >= e2;
}
+/**
+ * drm_buddy_free_unused_pages - free unused pages
+ *
+ * @mm: DRM buddy manager
+ * @actual_size: original size requested
+ * @blocks: output list head to add allocated blocks
+ *
+ * For contiguous allocation, we round up the size to the nearest
+ * power of two value, drivers consume *actual* size, so remaining
+ * portions are unused and it can be freed.
+ *
+ * Returns:
+ * 0 on success, error code on failure.
+ */
+int drm_buddy_free_unused_pages(struct drm_buddy_mm *mm,
+ u64 actual_size,
+ struct list_head *blocks)
+{
+ struct drm_buddy_block *block;
+ struct drm_buddy_block *buddy;
+ u64 actual_start;
+ u64 actual_end;
+ LIST_HEAD(dfs);
+ u64 count = 0;
+ int err;
+
+ if (!list_is_singular(blocks))
+ return -EINVAL;
+
+ block = list_first_entry_or_null(blocks,
+ struct drm_buddy_block,
+ link);
+
+ if (!block)
+ return -EINVAL;
+
+ if (actual_size > drm_buddy_block_size(mm, block))
+ return -EINVAL;
+
+ if (actual_size == drm_buddy_block_size(mm, block))
+ return 0;
+
+ list_del(&block->link);
+
+ actual_start = drm_buddy_block_offset(block);
+ actual_end = actual_start + actual_size - 1;
+
+ if (drm_buddy_block_is_allocated(block))
+ mark_free(mm, block);
+
+ list_add(&block->tmp_link, &dfs);
+
+ while (1) {
+ block = list_first_entry_or_null(&dfs,
+ struct drm_buddy_block,
+ tmp_link);
+
+ if (!block)
+ break;
+
+ list_del(&block->tmp_link);
+
+ if (count == actual_size)
+ return 0;
+
+ if (contains(actual_start, actual_end, drm_buddy_block_offset(block),
+ (drm_buddy_block_offset(block) + drm_buddy_block_size(mm, block) - 1))) {
+ BUG_ON(!drm_buddy_block_is_free(block));
+
+ /* Allocate only required blocks */
+ mark_allocated(block);
+ mm->avail -= drm_buddy_block_size(mm, block);
+ list_add_tail(&block->link, blocks);
+ count += drm_buddy_block_size(mm, block);
+ continue;
+ }
+
+ if (drm_buddy_block_order(block) == 0)
+ continue;
+
+ if (!drm_buddy_block_is_split(block)) {
+ err = split_block(mm, block);
+
+ if (unlikely(err))
+ goto err_undo;
+ }
+
+ list_add(&block->right->tmp_link, &dfs);
+ list_add(&block->left->tmp_link, &dfs);
+ }
+
+ return -ENOSPC;
+
+err_undo:
+ buddy = get_buddy(block);
+ if (buddy &&
+ (drm_buddy_block_is_free(block) &&
+ drm_buddy_block_is_free(buddy)))
+ __drm_buddy_free(mm, block);
+ return err;
+}
+EXPORT_SYMBOL(drm_buddy_free_unused_pages);
+
static struct drm_buddy_block *
alloc_range(struct drm_buddy_mm *mm,
u64 start, u64 end,
diff --git a/include/drm/drm_buddy.h b/include/drm/drm_buddy.h
index cd8021d2d6e7..1dfc80c88e1f 100644
--- a/include/drm/drm_buddy.h
+++ b/include/drm/drm_buddy.h
@@ -145,6 +145,10 @@ int drm_buddy_alloc(struct drm_buddy_mm *mm,
struct list_head *blocks,
unsigned long flags);
+int drm_buddy_free_unused_pages(struct drm_buddy_mm *mm,
+ u64 actual_size,
+ struct list_head *blocks);
+
void drm_buddy_free(struct drm_buddy_mm *mm, struct drm_buddy_block *block);
void drm_buddy_free_list(struct drm_buddy_mm *mm, struct list_head *objects);
--
2.25.1
^ permalink raw reply related [flat|nested] 58+ messages in thread
* [PATCH 6/8] drm/i915: add free_unused_pages support to i915
2021-10-25 13:00 ` [Intel-gfx] " Arunpravin
@ 2021-10-25 13:00 ` Arunpravin
-1 siblings, 0 replies; 58+ messages in thread
From: Arunpravin @ 2021-10-25 13:00 UTC (permalink / raw)
To: dri-devel, intel-gfx, amd-gfx
Cc: daniel, christian.koenig, alexander.deucher, matthew.auld,
tzimmermann, jani.nikula, Arunpravin
add drm_buddy_free_unused_pages() support on
contiguous allocation
Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
---
drivers/gpu/drm/i915/i915_ttm_buddy_manager.c | 8 ++++++++
1 file changed, 8 insertions(+)
diff --git a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
index 963468228392..162947af8e04 100644
--- a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
+++ b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
@@ -98,6 +98,14 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
if (unlikely(err))
goto err_free_blocks;
+ if (place->flags & TTM_PL_FLAG_CONTIGUOUS) {
+ err = drm_buddy_free_unused_pages(mm, (uint64_t)n_pages << PAGE_SHIFT,
+ &bman_res->blocks);
+
+ if (unlikely(err))
+ goto err_free_blocks;
+ }
+
*res = &bman_res->base;
return 0;
--
2.25.1
^ permalink raw reply related [flat|nested] 58+ messages in thread
* [Intel-gfx] [PATCH 6/8] drm/i915: add free_unused_pages support to i915
@ 2021-10-25 13:00 ` Arunpravin
0 siblings, 0 replies; 58+ messages in thread
From: Arunpravin @ 2021-10-25 13:00 UTC (permalink / raw)
To: dri-devel, intel-gfx, amd-gfx
Cc: daniel, christian.koenig, alexander.deucher, matthew.auld,
tzimmermann, jani.nikula, Arunpravin
add drm_buddy_free_unused_pages() support on
contiguous allocation
Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
---
drivers/gpu/drm/i915/i915_ttm_buddy_manager.c | 8 ++++++++
1 file changed, 8 insertions(+)
diff --git a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
index 963468228392..162947af8e04 100644
--- a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
+++ b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
@@ -98,6 +98,14 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
if (unlikely(err))
goto err_free_blocks;
+ if (place->flags & TTM_PL_FLAG_CONTIGUOUS) {
+ err = drm_buddy_free_unused_pages(mm, (uint64_t)n_pages << PAGE_SHIFT,
+ &bman_res->blocks);
+
+ if (unlikely(err))
+ goto err_free_blocks;
+ }
+
*res = &bman_res->base;
return 0;
--
2.25.1
^ permalink raw reply related [flat|nested] 58+ messages in thread
* [PATCH 7/8] drm/amdgpu: move vram inline functions into a header
2021-10-25 13:00 ` [Intel-gfx] " Arunpravin
@ 2021-10-25 13:00 ` Arunpravin
-1 siblings, 0 replies; 58+ messages in thread
From: Arunpravin @ 2021-10-25 13:00 UTC (permalink / raw)
To: dri-devel, intel-gfx, amd-gfx
Cc: daniel, christian.koenig, alexander.deucher, matthew.auld,
tzimmermann, jani.nikula, Arunpravin
Move shared vram inline functions and structs
into a header file
Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.h | 51 ++++++++++++++++++++
1 file changed, 51 insertions(+)
create mode 100644 drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.h
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.h
new file mode 100644
index 000000000000..32cd6f54b6f3
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.h
@@ -0,0 +1,51 @@
+/* SPDX-License-Identifier: MIT
+ * Copyright 2021 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#ifndef __AMDGPU_VRAM_MGR_H__
+#define __AMDGPU_VRAM_MGR_H__
+
+#include <drm/drm_buddy.h>
+
+struct amdgpu_vram_mgr_node {
+ struct ttm_resource base;
+ struct list_head blocks;
+ unsigned long flags;
+};
+
+static inline uint64_t amdgpu_node_start(struct drm_buddy_block *block)
+{
+ return drm_buddy_block_offset(block);
+}
+
+static inline uint64_t amdgpu_node_size(struct drm_buddy_block *block)
+{
+ return PAGE_SIZE << drm_buddy_block_order(block);
+}
+
+static inline struct amdgpu_vram_mgr_node *
+to_amdgpu_vram_mgr_node(struct ttm_resource *res)
+{
+ return container_of(res, struct amdgpu_vram_mgr_node, base);
+}
+
+#endif
--
2.25.1
^ permalink raw reply related [flat|nested] 58+ messages in thread
* [Intel-gfx] [PATCH 7/8] drm/amdgpu: move vram inline functions into a header
@ 2021-10-25 13:00 ` Arunpravin
0 siblings, 0 replies; 58+ messages in thread
From: Arunpravin @ 2021-10-25 13:00 UTC (permalink / raw)
To: dri-devel, intel-gfx, amd-gfx
Cc: daniel, christian.koenig, alexander.deucher, matthew.auld,
tzimmermann, jani.nikula, Arunpravin
Move shared vram inline functions and structs
into a header file
Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.h | 51 ++++++++++++++++++++
1 file changed, 51 insertions(+)
create mode 100644 drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.h
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.h
new file mode 100644
index 000000000000..32cd6f54b6f3
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.h
@@ -0,0 +1,51 @@
+/* SPDX-License-Identifier: MIT
+ * Copyright 2021 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#ifndef __AMDGPU_VRAM_MGR_H__
+#define __AMDGPU_VRAM_MGR_H__
+
+#include <drm/drm_buddy.h>
+
+struct amdgpu_vram_mgr_node {
+ struct ttm_resource base;
+ struct list_head blocks;
+ unsigned long flags;
+};
+
+static inline uint64_t amdgpu_node_start(struct drm_buddy_block *block)
+{
+ return drm_buddy_block_offset(block);
+}
+
+static inline uint64_t amdgpu_node_size(struct drm_buddy_block *block)
+{
+ return PAGE_SIZE << drm_buddy_block_order(block);
+}
+
+static inline struct amdgpu_vram_mgr_node *
+to_amdgpu_vram_mgr_node(struct ttm_resource *res)
+{
+ return container_of(res, struct amdgpu_vram_mgr_node, base);
+}
+
+#endif
--
2.25.1
^ permalink raw reply related [flat|nested] 58+ messages in thread
* [PATCH 8/8] drm/amdgpu: add drm buddy support to amdgpu
2021-10-25 13:00 ` [Intel-gfx] " Arunpravin
@ 2021-10-25 13:00 ` Arunpravin
-1 siblings, 0 replies; 58+ messages in thread
From: Arunpravin @ 2021-10-25 13:00 UTC (permalink / raw)
To: dri-devel, intel-gfx, amd-gfx
Cc: daniel, christian.koenig, alexander.deucher, matthew.auld,
tzimmermann, jani.nikula, Arunpravin
- Remove drm_mm references and replace with drm buddy functionalities
- Add res cursor support for drm buddy
Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
---
.../gpu/drm/amd/amdgpu/amdgpu_res_cursor.h | 97 +++++--
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h | 4 +-
drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c | 239 ++++++++++--------
3 files changed, 221 insertions(+), 119 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_res_cursor.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_res_cursor.h
index acfa207cf970..da12b4ff2e45 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_res_cursor.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_res_cursor.h
@@ -30,12 +30,15 @@
#include <drm/ttm/ttm_resource.h>
#include <drm/ttm/ttm_range_manager.h>
+#include "amdgpu_vram_mgr.h"
+
/* state back for walking over vram_mgr and gtt_mgr allocations */
struct amdgpu_res_cursor {
uint64_t start;
uint64_t size;
uint64_t remaining;
- struct drm_mm_node *node;
+ void *node;
+ uint32_t mem_type;
};
/**
@@ -52,27 +55,63 @@ static inline void amdgpu_res_first(struct ttm_resource *res,
uint64_t start, uint64_t size,
struct amdgpu_res_cursor *cur)
{
+ struct drm_buddy_block *block;
+ struct list_head *head, *next;
struct drm_mm_node *node;
- if (!res || res->mem_type == TTM_PL_SYSTEM) {
- cur->start = start;
- cur->size = size;
- cur->remaining = size;
- cur->node = NULL;
- WARN_ON(res && start + size > res->num_pages << PAGE_SHIFT);
- return;
- }
+ if (!res)
+ goto err_out;
BUG_ON(start + size > res->num_pages << PAGE_SHIFT);
- node = to_ttm_range_mgr_node(res)->mm_nodes;
- while (start >= node->size << PAGE_SHIFT)
- start -= node++->size << PAGE_SHIFT;
+ cur->mem_type = res->mem_type;
+
+ switch (cur->mem_type) {
+ case TTM_PL_VRAM:
+ head = &to_amdgpu_vram_mgr_node(res)->blocks;
+
+ block = list_first_entry_or_null(head,
+ struct drm_buddy_block,
+ link);
+ if (!block)
+ goto err_out;
+
+ while (start >= amdgpu_node_size(block)) {
+ start -= amdgpu_node_size(block);
+
+ next = block->link.next;
+ if (next != head)
+ block = list_entry(next, struct drm_buddy_block, link);
+ }
+
+ cur->start = amdgpu_node_start(block) + start;
+ cur->size = min(amdgpu_node_size(block) - start, size);
+ cur->remaining = size;
+ cur->node = block;
+ break;
+ case TTM_PL_TT:
+ node = to_ttm_range_mgr_node(res)->mm_nodes;
+ while (start >= node->size << PAGE_SHIFT)
+ start -= node++->size << PAGE_SHIFT;
+
+ cur->start = (node->start << PAGE_SHIFT) + start;
+ cur->size = min((node->size << PAGE_SHIFT) - start, size);
+ cur->remaining = size;
+ cur->node = node;
+ break;
+ default:
+ goto err_out;
+ }
- cur->start = (node->start << PAGE_SHIFT) + start;
- cur->size = min((node->size << PAGE_SHIFT) - start, size);
+ return;
+
+err_out:
+ cur->start = start;
+ cur->size = size;
cur->remaining = size;
- cur->node = node;
+ cur->node = NULL;
+ WARN_ON(res && start + size > res->num_pages << PAGE_SHIFT);
+ return;
}
/**
@@ -85,7 +124,9 @@ static inline void amdgpu_res_first(struct ttm_resource *res,
*/
static inline void amdgpu_res_next(struct amdgpu_res_cursor *cur, uint64_t size)
{
- struct drm_mm_node *node = cur->node;
+ struct drm_buddy_block *block;
+ struct drm_mm_node *node;
+ struct list_head *next;
BUG_ON(size > cur->remaining);
@@ -99,9 +140,27 @@ static inline void amdgpu_res_next(struct amdgpu_res_cursor *cur, uint64_t size)
return;
}
- cur->node = ++node;
- cur->start = node->start << PAGE_SHIFT;
- cur->size = min(node->size << PAGE_SHIFT, cur->remaining);
+ switch (cur->mem_type) {
+ case TTM_PL_VRAM:
+ block = cur->node;
+
+ next = block->link.next;
+ block = list_entry(next, struct drm_buddy_block, link);
+
+ cur->node = block;
+ cur->start = amdgpu_node_start(block);
+ cur->size = min(amdgpu_node_size(block), cur->remaining);
+ break;
+ case TTM_PL_TT:
+ node = cur->node;
+
+ cur->node = ++node;
+ cur->start = node->start << PAGE_SHIFT;
+ cur->size = min(node->size << PAGE_SHIFT, cur->remaining);
+ break;
+ default:
+ return;
+ }
}
#endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
index 639c7b41e30b..a8ac9902ab29 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
@@ -26,6 +26,7 @@
#include <linux/dma-direction.h>
#include <drm/gpu_scheduler.h>
+#include <drm/drm_buddy.h>
#include "amdgpu.h"
#define AMDGPU_PL_GDS (TTM_PL_PRIV + 0)
@@ -40,12 +41,13 @@
struct amdgpu_vram_mgr {
struct ttm_resource_manager manager;
- struct drm_mm mm;
+ struct drm_buddy_mm mm;
spinlock_t lock;
struct list_head reservations_pending;
struct list_head reserved_pages;
atomic64_t usage;
atomic64_t vis_usage;
+ uint64_t default_page_size;
};
struct amdgpu_gtt_mgr {
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c
index 7b2b0980ec41..9893ab583465 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c
@@ -32,8 +32,12 @@
#include "atom.h"
struct amdgpu_vram_reservation {
+ uint64_t start;
+ uint64_t size;
+ uint64_t min_size;
+ unsigned long flags;
+ struct list_head block;
struct list_head node;
- struct drm_mm_node mm_node;
};
static inline struct amdgpu_vram_mgr *
@@ -196,10 +200,10 @@ const struct attribute_group amdgpu_vram_mgr_attr_group = {
* Calculate how many bytes of the MM node are inside visible VRAM
*/
static u64 amdgpu_vram_mgr_vis_size(struct amdgpu_device *adev,
- struct drm_mm_node *node)
+ struct drm_buddy_block *block)
{
- uint64_t start = node->start << PAGE_SHIFT;
- uint64_t end = (node->size + node->start) << PAGE_SHIFT;
+ uint64_t start = amdgpu_node_start(block);
+ uint64_t end = start + amdgpu_node_size(block);
if (start >= adev->gmc.visible_vram_size)
return 0;
@@ -220,9 +224,9 @@ u64 amdgpu_vram_mgr_bo_visible_size(struct amdgpu_bo *bo)
{
struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
struct ttm_resource *res = bo->tbo.resource;
- unsigned pages = res->num_pages;
- struct drm_mm_node *mm;
- u64 usage;
+ struct amdgpu_vram_mgr_node *node = to_amdgpu_vram_mgr_node(res);
+ struct drm_buddy_block *block;
+ u64 usage = 0;
if (amdgpu_gmc_vram_full_visible(&adev->gmc))
return amdgpu_bo_size(bo);
@@ -230,9 +234,8 @@ u64 amdgpu_vram_mgr_bo_visible_size(struct amdgpu_bo *bo)
if (res->start >= adev->gmc.visible_vram_size >> PAGE_SHIFT)
return 0;
- mm = &container_of(res, struct ttm_range_mgr_node, base)->mm_nodes[0];
- for (usage = 0; pages; pages -= mm->size, mm++)
- usage += amdgpu_vram_mgr_vis_size(adev, mm);
+ list_for_each_entry(block, &node->blocks, link)
+ usage += amdgpu_vram_mgr_vis_size(adev, block);
return usage;
}
@@ -242,21 +245,30 @@ static void amdgpu_vram_mgr_do_reserve(struct ttm_resource_manager *man)
{
struct amdgpu_vram_mgr *mgr = to_vram_mgr(man);
struct amdgpu_device *adev = to_amdgpu_device(mgr);
- struct drm_mm *mm = &mgr->mm;
+ struct drm_buddy_mm *mm = &mgr->mm;
struct amdgpu_vram_reservation *rsv, *temp;
+ struct drm_buddy_block *block;
uint64_t vis_usage;
list_for_each_entry_safe(rsv, temp, &mgr->reservations_pending, node) {
- if (drm_mm_reserve_node(mm, &rsv->mm_node))
+ if (drm_buddy_alloc(mm, rsv->start, rsv->start + rsv->size,
+ rsv->size, rsv->min_size, &rsv->block,
+ rsv->flags))
continue;
- dev_dbg(adev->dev, "Reservation 0x%llx - %lld, Succeeded\n",
- rsv->mm_node.start, rsv->mm_node.size);
+ block = list_first_entry_or_null(&rsv->block,
+ struct drm_buddy_block,
+ link);
- vis_usage = amdgpu_vram_mgr_vis_size(adev, &rsv->mm_node);
- atomic64_add(vis_usage, &mgr->vis_usage);
- atomic64_add(rsv->mm_node.size << PAGE_SHIFT, &mgr->usage);
- list_move(&rsv->node, &mgr->reserved_pages);
+ if (block) {
+ dev_dbg(adev->dev, "Reservation 0x%llx - %lld, Succeeded\n",
+ rsv->start, rsv->size);
+
+ vis_usage = amdgpu_vram_mgr_vis_size(adev, block);
+ atomic64_add(vis_usage, &mgr->vis_usage);
+ atomic64_add(rsv->size, &mgr->usage);
+ list_move(&rsv->node, &mgr->reserved_pages);
+ }
}
}
@@ -280,11 +292,15 @@ int amdgpu_vram_mgr_reserve_range(struct ttm_resource_manager *man,
return -ENOMEM;
INIT_LIST_HEAD(&rsv->node);
- rsv->mm_node.start = start >> PAGE_SHIFT;
- rsv->mm_node.size = size >> PAGE_SHIFT;
+ INIT_LIST_HEAD(&rsv->block);
+
+ rsv->start = start;
+ rsv->size = size;
+ rsv->min_size = size;
+ rsv->flags |= DRM_BUDDY_RANGE_ALLOCATION;
spin_lock(&mgr->lock);
- list_add_tail(&mgr->reservations_pending, &rsv->node);
+ list_add_tail(&rsv->node, &mgr->reservations_pending);
amdgpu_vram_mgr_do_reserve(man);
spin_unlock(&mgr->lock);
@@ -312,16 +328,16 @@ int amdgpu_vram_mgr_query_page_status(struct ttm_resource_manager *man,
spin_lock(&mgr->lock);
list_for_each_entry(rsv, &mgr->reservations_pending, node) {
- if ((rsv->mm_node.start <= start) &&
- (start < (rsv->mm_node.start + rsv->mm_node.size))) {
+ if ((rsv->start <= start) &&
+ (start < (rsv->start + rsv->size))) {
ret = -EBUSY;
goto out;
}
}
list_for_each_entry(rsv, &mgr->reserved_pages, node) {
- if ((rsv->mm_node.start <= start) &&
- (start < (rsv->mm_node.start + rsv->mm_node.size))) {
+ if ((rsv->start <= start) &&
+ (start < (rsv->start + rsv->size))) {
ret = 0;
goto out;
}
@@ -333,28 +349,6 @@ int amdgpu_vram_mgr_query_page_status(struct ttm_resource_manager *man,
return ret;
}
-/**
- * amdgpu_vram_mgr_virt_start - update virtual start address
- *
- * @mem: ttm_resource to update
- * @node: just allocated node
- *
- * Calculate a virtual BO start address to easily check if everything is CPU
- * accessible.
- */
-static void amdgpu_vram_mgr_virt_start(struct ttm_resource *mem,
- struct drm_mm_node *node)
-{
- unsigned long start;
-
- start = node->start + node->size;
- if (start > mem->num_pages)
- start -= mem->num_pages;
- else
- start = 0;
- mem->start = max(mem->start, start);
-}
-
/**
* amdgpu_vram_mgr_new - allocate new ranges
*
@@ -370,13 +364,13 @@ static int amdgpu_vram_mgr_new(struct ttm_resource_manager *man,
const struct ttm_place *place,
struct ttm_resource **res)
{
- unsigned long lpfn, num_nodes, pages_per_node, pages_left, pages;
+ unsigned long lpfn, pages_per_node, pages_left, pages, n_pages;
+ uint64_t vis_usage = 0, mem_bytes, max_bytes, min_page_size;
struct amdgpu_vram_mgr *mgr = to_vram_mgr(man);
struct amdgpu_device *adev = to_amdgpu_device(mgr);
- uint64_t vis_usage = 0, mem_bytes, max_bytes;
- struct ttm_range_mgr_node *node;
- struct drm_mm *mm = &mgr->mm;
- enum drm_mm_insert_mode mode;
+ struct amdgpu_vram_mgr_node *node;
+ struct drm_buddy_mm *mm = &mgr->mm;
+ struct drm_buddy_block *block;
unsigned i;
int r;
@@ -395,10 +389,9 @@ static int amdgpu_vram_mgr_new(struct ttm_resource_manager *man,
goto error_sub;
}
- if (place->flags & TTM_PL_FLAG_CONTIGUOUS) {
+ if (place->flags & TTM_PL_FLAG_CONTIGUOUS)
pages_per_node = ~0ul;
- num_nodes = 1;
- } else {
+ else {
#ifdef CONFIG_TRANSPARENT_HUGEPAGE
pages_per_node = HPAGE_PMD_NR;
#else
@@ -407,11 +400,10 @@ static int amdgpu_vram_mgr_new(struct ttm_resource_manager *man,
#endif
pages_per_node = max_t(uint32_t, pages_per_node,
tbo->page_alignment);
- num_nodes = DIV_ROUND_UP_ULL(PFN_UP(mem_bytes), pages_per_node);
}
- node = kvmalloc(struct_size(node, mm_nodes, num_nodes),
- GFP_KERNEL | __GFP_ZERO);
+ node = kzalloc(sizeof(*node), GFP_KERNEL);
+
if (!node) {
r = -ENOMEM;
goto error_sub;
@@ -419,9 +411,17 @@ static int amdgpu_vram_mgr_new(struct ttm_resource_manager *man,
ttm_resource_init(tbo, place, &node->base);
- mode = DRM_MM_INSERT_BEST;
+ INIT_LIST_HEAD(&node->blocks);
+
if (place->flags & TTM_PL_FLAG_TOPDOWN)
- mode = DRM_MM_INSERT_HIGH;
+ node->flags |= DRM_BUDDY_TOPDOWN_ALLOCATION;
+
+ if (place->fpfn || lpfn != man->size)
+ /* Allocate blocks in desired range */
+ node->flags |= DRM_BUDDY_RANGE_ALLOCATION;
+
+ min_page_size = mgr->default_page_size;
+ BUG_ON(min_page_size < mm->chunk_size);
pages_left = node->base.num_pages;
@@ -429,36 +429,63 @@ static int amdgpu_vram_mgr_new(struct ttm_resource_manager *man,
pages = min(pages_left, 2UL << (30 - PAGE_SHIFT));
i = 0;
- spin_lock(&mgr->lock);
while (pages_left) {
- uint32_t alignment = tbo->page_alignment;
-
if (pages >= pages_per_node)
- alignment = pages_per_node;
-
- r = drm_mm_insert_node_in_range(mm, &node->mm_nodes[i], pages,
- alignment, 0, place->fpfn,
- lpfn, mode);
- if (unlikely(r)) {
- if (pages > pages_per_node) {
- if (is_power_of_2(pages))
- pages = pages / 2;
- else
- pages = rounddown_pow_of_two(pages);
- continue;
- }
- goto error_free;
+ pages = pages_per_node;
+
+ n_pages = pages;
+
+ if (place->flags & TTM_PL_FLAG_CONTIGUOUS) {
+ n_pages = roundup_pow_of_two(n_pages);
+ min_page_size = (uint64_t)n_pages << PAGE_SHIFT;
+
+ if (n_pages > lpfn)
+ lpfn = n_pages;
}
- vis_usage += amdgpu_vram_mgr_vis_size(adev, &node->mm_nodes[i]);
- amdgpu_vram_mgr_virt_start(&node->base, &node->mm_nodes[i]);
+ spin_lock(&mgr->lock);
+ r = drm_buddy_alloc(mm, (uint64_t)place->fpfn << PAGE_SHIFT,
+ (uint64_t)lpfn << PAGE_SHIFT,
+ (uint64_t)n_pages << PAGE_SHIFT,
+ min_page_size, &node->blocks,
+ node->flags);
+ spin_unlock(&mgr->lock);
+
+ if (unlikely(r))
+ goto error_free_blocks;
+
pages_left -= pages;
++i;
if (pages > pages_left)
pages = pages_left;
}
- spin_unlock(&mgr->lock);
+
+ /* Free unused pages for contiguous allocation */
+ if (place->flags & TTM_PL_FLAG_CONTIGUOUS) {
+ uint64_t actual_size = (uint64_t)node->base.num_pages << PAGE_SHIFT;
+
+ r = drm_buddy_free_unused_pages(mm,
+ actual_size,
+ &node->blocks);
+
+ if (unlikely(r))
+ goto error_free_blocks;
+ }
+
+ list_for_each_entry(block, &node->blocks, link)
+ vis_usage += amdgpu_vram_mgr_vis_size(adev, block);
+
+ block = list_first_entry_or_null(&node->blocks,
+ struct drm_buddy_block,
+ link);
+
+ if (!block) {
+ r = -ENOSPC;
+ goto error_free_res;
+ }
+
+ node->base.start = amdgpu_node_start(block) >> PAGE_SHIFT;
if (i == 1)
node->base.placement |= TTM_PL_FLAG_CONTIGUOUS;
@@ -472,12 +499,12 @@ static int amdgpu_vram_mgr_new(struct ttm_resource_manager *man,
*res = &node->base;
return 0;
-error_free:
- while (i--)
- drm_mm_remove_node(&node->mm_nodes[i]);
+error_free_blocks:
+ spin_lock(&mgr->lock);
+ drm_buddy_free_list(mm, &node->blocks);
spin_unlock(&mgr->lock);
- kvfree(node);
-
+error_free_res:
+ kfree(node);
error_sub:
atomic64_sub(mem_bytes, &mgr->usage);
return r;
@@ -494,28 +521,28 @@ static int amdgpu_vram_mgr_new(struct ttm_resource_manager *man,
static void amdgpu_vram_mgr_del(struct ttm_resource_manager *man,
struct ttm_resource *res)
{
- struct ttm_range_mgr_node *node = to_ttm_range_mgr_node(res);
+ struct amdgpu_vram_mgr_node *node = to_amdgpu_vram_mgr_node(res);
struct amdgpu_vram_mgr *mgr = to_vram_mgr(man);
struct amdgpu_device *adev = to_amdgpu_device(mgr);
+ struct drm_buddy_mm *mm = &mgr->mm;
+ struct drm_buddy_block *block;
uint64_t usage = 0, vis_usage = 0;
- unsigned i, pages;
spin_lock(&mgr->lock);
- for (i = 0, pages = res->num_pages; pages;
- pages -= node->mm_nodes[i].size, ++i) {
- struct drm_mm_node *mm = &node->mm_nodes[i];
-
- drm_mm_remove_node(mm);
- usage += mm->size << PAGE_SHIFT;
- vis_usage += amdgpu_vram_mgr_vis_size(adev, mm);
+ list_for_each_entry(block, &node->blocks, link) {
+ usage += amdgpu_node_size(block);
+ vis_usage += amdgpu_vram_mgr_vis_size(adev, block);
}
+
amdgpu_vram_mgr_do_reserve(man);
+
+ drm_buddy_free_list(mm, &node->blocks);
spin_unlock(&mgr->lock);
atomic64_sub(usage, &mgr->usage);
atomic64_sub(vis_usage, &mgr->vis_usage);
- kvfree(node);
+ kfree(node);
}
/**
@@ -669,9 +696,18 @@ static void amdgpu_vram_mgr_debug(struct ttm_resource_manager *man,
struct drm_printer *printer)
{
struct amdgpu_vram_mgr *mgr = to_vram_mgr(man);
+ struct drm_buddy_mm *mm = &mgr->mm;
+ struct drm_buddy_block *block;
spin_lock(&mgr->lock);
- drm_mm_print(&mgr->mm, printer);
+ drm_printf(printer, "default_page_size: %lluKiB\n",
+ mgr->default_page_size >> 10);
+
+ drm_buddy_print(mm, printer);
+
+ drm_printf(printer, "reserved:\n");
+ list_for_each_entry(block, &mgr->reserved_pages, link)
+ drm_buddy_block_print(mm, block, printer);
spin_unlock(&mgr->lock);
drm_printf(printer, "man size:%llu pages, ram usage:%lluMB, vis usage:%lluMB\n",
@@ -696,15 +732,20 @@ int amdgpu_vram_mgr_init(struct amdgpu_device *adev)
{
struct amdgpu_vram_mgr *mgr = &adev->mman.vram_mgr;
struct ttm_resource_manager *man = &mgr->manager;
+ int err;
ttm_resource_manager_init(man, adev->gmc.real_vram_size >> PAGE_SHIFT);
man->func = &amdgpu_vram_mgr_func;
- drm_mm_init(&mgr->mm, 0, man->size);
+ err = drm_buddy_init(&mgr->mm, man->size << PAGE_SHIFT, PAGE_SIZE);
+ if (err)
+ return err;
+
spin_lock_init(&mgr->lock);
INIT_LIST_HEAD(&mgr->reservations_pending);
INIT_LIST_HEAD(&mgr->reserved_pages);
+ mgr->default_page_size = PAGE_SIZE;
ttm_set_driver_manager(&adev->mman.bdev, TTM_PL_VRAM, &mgr->manager);
ttm_resource_manager_set_used(man, true);
@@ -737,10 +778,10 @@ void amdgpu_vram_mgr_fini(struct amdgpu_device *adev)
kfree(rsv);
list_for_each_entry_safe(rsv, temp, &mgr->reserved_pages, node) {
- drm_mm_remove_node(&rsv->mm_node);
+ drm_buddy_free_list(&mgr->mm, &rsv->block);
kfree(rsv);
}
- drm_mm_takedown(&mgr->mm);
+ drm_buddy_fini(&mgr->mm);
spin_unlock(&mgr->lock);
ttm_resource_manager_cleanup(man);
--
2.25.1
^ permalink raw reply related [flat|nested] 58+ messages in thread
* [Intel-gfx] [PATCH 8/8] drm/amdgpu: add drm buddy support to amdgpu
@ 2021-10-25 13:00 ` Arunpravin
0 siblings, 0 replies; 58+ messages in thread
From: Arunpravin @ 2021-10-25 13:00 UTC (permalink / raw)
To: dri-devel, intel-gfx, amd-gfx
Cc: daniel, christian.koenig, alexander.deucher, matthew.auld,
tzimmermann, jani.nikula, Arunpravin
- Remove drm_mm references and replace with drm buddy functionalities
- Add res cursor support for drm buddy
Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
---
.../gpu/drm/amd/amdgpu/amdgpu_res_cursor.h | 97 +++++--
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h | 4 +-
drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c | 239 ++++++++++--------
3 files changed, 221 insertions(+), 119 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_res_cursor.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_res_cursor.h
index acfa207cf970..da12b4ff2e45 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_res_cursor.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_res_cursor.h
@@ -30,12 +30,15 @@
#include <drm/ttm/ttm_resource.h>
#include <drm/ttm/ttm_range_manager.h>
+#include "amdgpu_vram_mgr.h"
+
/* state back for walking over vram_mgr and gtt_mgr allocations */
struct amdgpu_res_cursor {
uint64_t start;
uint64_t size;
uint64_t remaining;
- struct drm_mm_node *node;
+ void *node;
+ uint32_t mem_type;
};
/**
@@ -52,27 +55,63 @@ static inline void amdgpu_res_first(struct ttm_resource *res,
uint64_t start, uint64_t size,
struct amdgpu_res_cursor *cur)
{
+ struct drm_buddy_block *block;
+ struct list_head *head, *next;
struct drm_mm_node *node;
- if (!res || res->mem_type == TTM_PL_SYSTEM) {
- cur->start = start;
- cur->size = size;
- cur->remaining = size;
- cur->node = NULL;
- WARN_ON(res && start + size > res->num_pages << PAGE_SHIFT);
- return;
- }
+ if (!res)
+ goto err_out;
BUG_ON(start + size > res->num_pages << PAGE_SHIFT);
- node = to_ttm_range_mgr_node(res)->mm_nodes;
- while (start >= node->size << PAGE_SHIFT)
- start -= node++->size << PAGE_SHIFT;
+ cur->mem_type = res->mem_type;
+
+ switch (cur->mem_type) {
+ case TTM_PL_VRAM:
+ head = &to_amdgpu_vram_mgr_node(res)->blocks;
+
+ block = list_first_entry_or_null(head,
+ struct drm_buddy_block,
+ link);
+ if (!block)
+ goto err_out;
+
+ while (start >= amdgpu_node_size(block)) {
+ start -= amdgpu_node_size(block);
+
+ next = block->link.next;
+ if (next != head)
+ block = list_entry(next, struct drm_buddy_block, link);
+ }
+
+ cur->start = amdgpu_node_start(block) + start;
+ cur->size = min(amdgpu_node_size(block) - start, size);
+ cur->remaining = size;
+ cur->node = block;
+ break;
+ case TTM_PL_TT:
+ node = to_ttm_range_mgr_node(res)->mm_nodes;
+ while (start >= node->size << PAGE_SHIFT)
+ start -= node++->size << PAGE_SHIFT;
+
+ cur->start = (node->start << PAGE_SHIFT) + start;
+ cur->size = min((node->size << PAGE_SHIFT) - start, size);
+ cur->remaining = size;
+ cur->node = node;
+ break;
+ default:
+ goto err_out;
+ }
- cur->start = (node->start << PAGE_SHIFT) + start;
- cur->size = min((node->size << PAGE_SHIFT) - start, size);
+ return;
+
+err_out:
+ cur->start = start;
+ cur->size = size;
cur->remaining = size;
- cur->node = node;
+ cur->node = NULL;
+ WARN_ON(res && start + size > res->num_pages << PAGE_SHIFT);
+ return;
}
/**
@@ -85,7 +124,9 @@ static inline void amdgpu_res_first(struct ttm_resource *res,
*/
static inline void amdgpu_res_next(struct amdgpu_res_cursor *cur, uint64_t size)
{
- struct drm_mm_node *node = cur->node;
+ struct drm_buddy_block *block;
+ struct drm_mm_node *node;
+ struct list_head *next;
BUG_ON(size > cur->remaining);
@@ -99,9 +140,27 @@ static inline void amdgpu_res_next(struct amdgpu_res_cursor *cur, uint64_t size)
return;
}
- cur->node = ++node;
- cur->start = node->start << PAGE_SHIFT;
- cur->size = min(node->size << PAGE_SHIFT, cur->remaining);
+ switch (cur->mem_type) {
+ case TTM_PL_VRAM:
+ block = cur->node;
+
+ next = block->link.next;
+ block = list_entry(next, struct drm_buddy_block, link);
+
+ cur->node = block;
+ cur->start = amdgpu_node_start(block);
+ cur->size = min(amdgpu_node_size(block), cur->remaining);
+ break;
+ case TTM_PL_TT:
+ node = cur->node;
+
+ cur->node = ++node;
+ cur->start = node->start << PAGE_SHIFT;
+ cur->size = min(node->size << PAGE_SHIFT, cur->remaining);
+ break;
+ default:
+ return;
+ }
}
#endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
index 639c7b41e30b..a8ac9902ab29 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
@@ -26,6 +26,7 @@
#include <linux/dma-direction.h>
#include <drm/gpu_scheduler.h>
+#include <drm/drm_buddy.h>
#include "amdgpu.h"
#define AMDGPU_PL_GDS (TTM_PL_PRIV + 0)
@@ -40,12 +41,13 @@
struct amdgpu_vram_mgr {
struct ttm_resource_manager manager;
- struct drm_mm mm;
+ struct drm_buddy_mm mm;
spinlock_t lock;
struct list_head reservations_pending;
struct list_head reserved_pages;
atomic64_t usage;
atomic64_t vis_usage;
+ uint64_t default_page_size;
};
struct amdgpu_gtt_mgr {
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c
index 7b2b0980ec41..9893ab583465 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c
@@ -32,8 +32,12 @@
#include "atom.h"
struct amdgpu_vram_reservation {
+ uint64_t start;
+ uint64_t size;
+ uint64_t min_size;
+ unsigned long flags;
+ struct list_head block;
struct list_head node;
- struct drm_mm_node mm_node;
};
static inline struct amdgpu_vram_mgr *
@@ -196,10 +200,10 @@ const struct attribute_group amdgpu_vram_mgr_attr_group = {
* Calculate how many bytes of the MM node are inside visible VRAM
*/
static u64 amdgpu_vram_mgr_vis_size(struct amdgpu_device *adev,
- struct drm_mm_node *node)
+ struct drm_buddy_block *block)
{
- uint64_t start = node->start << PAGE_SHIFT;
- uint64_t end = (node->size + node->start) << PAGE_SHIFT;
+ uint64_t start = amdgpu_node_start(block);
+ uint64_t end = start + amdgpu_node_size(block);
if (start >= adev->gmc.visible_vram_size)
return 0;
@@ -220,9 +224,9 @@ u64 amdgpu_vram_mgr_bo_visible_size(struct amdgpu_bo *bo)
{
struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
struct ttm_resource *res = bo->tbo.resource;
- unsigned pages = res->num_pages;
- struct drm_mm_node *mm;
- u64 usage;
+ struct amdgpu_vram_mgr_node *node = to_amdgpu_vram_mgr_node(res);
+ struct drm_buddy_block *block;
+ u64 usage = 0;
if (amdgpu_gmc_vram_full_visible(&adev->gmc))
return amdgpu_bo_size(bo);
@@ -230,9 +234,8 @@ u64 amdgpu_vram_mgr_bo_visible_size(struct amdgpu_bo *bo)
if (res->start >= adev->gmc.visible_vram_size >> PAGE_SHIFT)
return 0;
- mm = &container_of(res, struct ttm_range_mgr_node, base)->mm_nodes[0];
- for (usage = 0; pages; pages -= mm->size, mm++)
- usage += amdgpu_vram_mgr_vis_size(adev, mm);
+ list_for_each_entry(block, &node->blocks, link)
+ usage += amdgpu_vram_mgr_vis_size(adev, block);
return usage;
}
@@ -242,21 +245,30 @@ static void amdgpu_vram_mgr_do_reserve(struct ttm_resource_manager *man)
{
struct amdgpu_vram_mgr *mgr = to_vram_mgr(man);
struct amdgpu_device *adev = to_amdgpu_device(mgr);
- struct drm_mm *mm = &mgr->mm;
+ struct drm_buddy_mm *mm = &mgr->mm;
struct amdgpu_vram_reservation *rsv, *temp;
+ struct drm_buddy_block *block;
uint64_t vis_usage;
list_for_each_entry_safe(rsv, temp, &mgr->reservations_pending, node) {
- if (drm_mm_reserve_node(mm, &rsv->mm_node))
+ if (drm_buddy_alloc(mm, rsv->start, rsv->start + rsv->size,
+ rsv->size, rsv->min_size, &rsv->block,
+ rsv->flags))
continue;
- dev_dbg(adev->dev, "Reservation 0x%llx - %lld, Succeeded\n",
- rsv->mm_node.start, rsv->mm_node.size);
+ block = list_first_entry_or_null(&rsv->block,
+ struct drm_buddy_block,
+ link);
- vis_usage = amdgpu_vram_mgr_vis_size(adev, &rsv->mm_node);
- atomic64_add(vis_usage, &mgr->vis_usage);
- atomic64_add(rsv->mm_node.size << PAGE_SHIFT, &mgr->usage);
- list_move(&rsv->node, &mgr->reserved_pages);
+ if (block) {
+ dev_dbg(adev->dev, "Reservation 0x%llx - %lld, Succeeded\n",
+ rsv->start, rsv->size);
+
+ vis_usage = amdgpu_vram_mgr_vis_size(adev, block);
+ atomic64_add(vis_usage, &mgr->vis_usage);
+ atomic64_add(rsv->size, &mgr->usage);
+ list_move(&rsv->node, &mgr->reserved_pages);
+ }
}
}
@@ -280,11 +292,15 @@ int amdgpu_vram_mgr_reserve_range(struct ttm_resource_manager *man,
return -ENOMEM;
INIT_LIST_HEAD(&rsv->node);
- rsv->mm_node.start = start >> PAGE_SHIFT;
- rsv->mm_node.size = size >> PAGE_SHIFT;
+ INIT_LIST_HEAD(&rsv->block);
+
+ rsv->start = start;
+ rsv->size = size;
+ rsv->min_size = size;
+ rsv->flags |= DRM_BUDDY_RANGE_ALLOCATION;
spin_lock(&mgr->lock);
- list_add_tail(&mgr->reservations_pending, &rsv->node);
+ list_add_tail(&rsv->node, &mgr->reservations_pending);
amdgpu_vram_mgr_do_reserve(man);
spin_unlock(&mgr->lock);
@@ -312,16 +328,16 @@ int amdgpu_vram_mgr_query_page_status(struct ttm_resource_manager *man,
spin_lock(&mgr->lock);
list_for_each_entry(rsv, &mgr->reservations_pending, node) {
- if ((rsv->mm_node.start <= start) &&
- (start < (rsv->mm_node.start + rsv->mm_node.size))) {
+ if ((rsv->start <= start) &&
+ (start < (rsv->start + rsv->size))) {
ret = -EBUSY;
goto out;
}
}
list_for_each_entry(rsv, &mgr->reserved_pages, node) {
- if ((rsv->mm_node.start <= start) &&
- (start < (rsv->mm_node.start + rsv->mm_node.size))) {
+ if ((rsv->start <= start) &&
+ (start < (rsv->start + rsv->size))) {
ret = 0;
goto out;
}
@@ -333,28 +349,6 @@ int amdgpu_vram_mgr_query_page_status(struct ttm_resource_manager *man,
return ret;
}
-/**
- * amdgpu_vram_mgr_virt_start - update virtual start address
- *
- * @mem: ttm_resource to update
- * @node: just allocated node
- *
- * Calculate a virtual BO start address to easily check if everything is CPU
- * accessible.
- */
-static void amdgpu_vram_mgr_virt_start(struct ttm_resource *mem,
- struct drm_mm_node *node)
-{
- unsigned long start;
-
- start = node->start + node->size;
- if (start > mem->num_pages)
- start -= mem->num_pages;
- else
- start = 0;
- mem->start = max(mem->start, start);
-}
-
/**
* amdgpu_vram_mgr_new - allocate new ranges
*
@@ -370,13 +364,13 @@ static int amdgpu_vram_mgr_new(struct ttm_resource_manager *man,
const struct ttm_place *place,
struct ttm_resource **res)
{
- unsigned long lpfn, num_nodes, pages_per_node, pages_left, pages;
+ unsigned long lpfn, pages_per_node, pages_left, pages, n_pages;
+ uint64_t vis_usage = 0, mem_bytes, max_bytes, min_page_size;
struct amdgpu_vram_mgr *mgr = to_vram_mgr(man);
struct amdgpu_device *adev = to_amdgpu_device(mgr);
- uint64_t vis_usage = 0, mem_bytes, max_bytes;
- struct ttm_range_mgr_node *node;
- struct drm_mm *mm = &mgr->mm;
- enum drm_mm_insert_mode mode;
+ struct amdgpu_vram_mgr_node *node;
+ struct drm_buddy_mm *mm = &mgr->mm;
+ struct drm_buddy_block *block;
unsigned i;
int r;
@@ -395,10 +389,9 @@ static int amdgpu_vram_mgr_new(struct ttm_resource_manager *man,
goto error_sub;
}
- if (place->flags & TTM_PL_FLAG_CONTIGUOUS) {
+ if (place->flags & TTM_PL_FLAG_CONTIGUOUS)
pages_per_node = ~0ul;
- num_nodes = 1;
- } else {
+ else {
#ifdef CONFIG_TRANSPARENT_HUGEPAGE
pages_per_node = HPAGE_PMD_NR;
#else
@@ -407,11 +400,10 @@ static int amdgpu_vram_mgr_new(struct ttm_resource_manager *man,
#endif
pages_per_node = max_t(uint32_t, pages_per_node,
tbo->page_alignment);
- num_nodes = DIV_ROUND_UP_ULL(PFN_UP(mem_bytes), pages_per_node);
}
- node = kvmalloc(struct_size(node, mm_nodes, num_nodes),
- GFP_KERNEL | __GFP_ZERO);
+ node = kzalloc(sizeof(*node), GFP_KERNEL);
+
if (!node) {
r = -ENOMEM;
goto error_sub;
@@ -419,9 +411,17 @@ static int amdgpu_vram_mgr_new(struct ttm_resource_manager *man,
ttm_resource_init(tbo, place, &node->base);
- mode = DRM_MM_INSERT_BEST;
+ INIT_LIST_HEAD(&node->blocks);
+
if (place->flags & TTM_PL_FLAG_TOPDOWN)
- mode = DRM_MM_INSERT_HIGH;
+ node->flags |= DRM_BUDDY_TOPDOWN_ALLOCATION;
+
+ if (place->fpfn || lpfn != man->size)
+ /* Allocate blocks in desired range */
+ node->flags |= DRM_BUDDY_RANGE_ALLOCATION;
+
+ min_page_size = mgr->default_page_size;
+ BUG_ON(min_page_size < mm->chunk_size);
pages_left = node->base.num_pages;
@@ -429,36 +429,63 @@ static int amdgpu_vram_mgr_new(struct ttm_resource_manager *man,
pages = min(pages_left, 2UL << (30 - PAGE_SHIFT));
i = 0;
- spin_lock(&mgr->lock);
while (pages_left) {
- uint32_t alignment = tbo->page_alignment;
-
if (pages >= pages_per_node)
- alignment = pages_per_node;
-
- r = drm_mm_insert_node_in_range(mm, &node->mm_nodes[i], pages,
- alignment, 0, place->fpfn,
- lpfn, mode);
- if (unlikely(r)) {
- if (pages > pages_per_node) {
- if (is_power_of_2(pages))
- pages = pages / 2;
- else
- pages = rounddown_pow_of_two(pages);
- continue;
- }
- goto error_free;
+ pages = pages_per_node;
+
+ n_pages = pages;
+
+ if (place->flags & TTM_PL_FLAG_CONTIGUOUS) {
+ n_pages = roundup_pow_of_two(n_pages);
+ min_page_size = (uint64_t)n_pages << PAGE_SHIFT;
+
+ if (n_pages > lpfn)
+ lpfn = n_pages;
}
- vis_usage += amdgpu_vram_mgr_vis_size(adev, &node->mm_nodes[i]);
- amdgpu_vram_mgr_virt_start(&node->base, &node->mm_nodes[i]);
+ spin_lock(&mgr->lock);
+ r = drm_buddy_alloc(mm, (uint64_t)place->fpfn << PAGE_SHIFT,
+ (uint64_t)lpfn << PAGE_SHIFT,
+ (uint64_t)n_pages << PAGE_SHIFT,
+ min_page_size, &node->blocks,
+ node->flags);
+ spin_unlock(&mgr->lock);
+
+ if (unlikely(r))
+ goto error_free_blocks;
+
pages_left -= pages;
++i;
if (pages > pages_left)
pages = pages_left;
}
- spin_unlock(&mgr->lock);
+
+ /* Free unused pages for contiguous allocation */
+ if (place->flags & TTM_PL_FLAG_CONTIGUOUS) {
+ uint64_t actual_size = (uint64_t)node->base.num_pages << PAGE_SHIFT;
+
+ r = drm_buddy_free_unused_pages(mm,
+ actual_size,
+ &node->blocks);
+
+ if (unlikely(r))
+ goto error_free_blocks;
+ }
+
+ list_for_each_entry(block, &node->blocks, link)
+ vis_usage += amdgpu_vram_mgr_vis_size(adev, block);
+
+ block = list_first_entry_or_null(&node->blocks,
+ struct drm_buddy_block,
+ link);
+
+ if (!block) {
+ r = -ENOSPC;
+ goto error_free_res;
+ }
+
+ node->base.start = amdgpu_node_start(block) >> PAGE_SHIFT;
if (i == 1)
node->base.placement |= TTM_PL_FLAG_CONTIGUOUS;
@@ -472,12 +499,12 @@ static int amdgpu_vram_mgr_new(struct ttm_resource_manager *man,
*res = &node->base;
return 0;
-error_free:
- while (i--)
- drm_mm_remove_node(&node->mm_nodes[i]);
+error_free_blocks:
+ spin_lock(&mgr->lock);
+ drm_buddy_free_list(mm, &node->blocks);
spin_unlock(&mgr->lock);
- kvfree(node);
-
+error_free_res:
+ kfree(node);
error_sub:
atomic64_sub(mem_bytes, &mgr->usage);
return r;
@@ -494,28 +521,28 @@ static int amdgpu_vram_mgr_new(struct ttm_resource_manager *man,
static void amdgpu_vram_mgr_del(struct ttm_resource_manager *man,
struct ttm_resource *res)
{
- struct ttm_range_mgr_node *node = to_ttm_range_mgr_node(res);
+ struct amdgpu_vram_mgr_node *node = to_amdgpu_vram_mgr_node(res);
struct amdgpu_vram_mgr *mgr = to_vram_mgr(man);
struct amdgpu_device *adev = to_amdgpu_device(mgr);
+ struct drm_buddy_mm *mm = &mgr->mm;
+ struct drm_buddy_block *block;
uint64_t usage = 0, vis_usage = 0;
- unsigned i, pages;
spin_lock(&mgr->lock);
- for (i = 0, pages = res->num_pages; pages;
- pages -= node->mm_nodes[i].size, ++i) {
- struct drm_mm_node *mm = &node->mm_nodes[i];
-
- drm_mm_remove_node(mm);
- usage += mm->size << PAGE_SHIFT;
- vis_usage += amdgpu_vram_mgr_vis_size(adev, mm);
+ list_for_each_entry(block, &node->blocks, link) {
+ usage += amdgpu_node_size(block);
+ vis_usage += amdgpu_vram_mgr_vis_size(adev, block);
}
+
amdgpu_vram_mgr_do_reserve(man);
+
+ drm_buddy_free_list(mm, &node->blocks);
spin_unlock(&mgr->lock);
atomic64_sub(usage, &mgr->usage);
atomic64_sub(vis_usage, &mgr->vis_usage);
- kvfree(node);
+ kfree(node);
}
/**
@@ -669,9 +696,18 @@ static void amdgpu_vram_mgr_debug(struct ttm_resource_manager *man,
struct drm_printer *printer)
{
struct amdgpu_vram_mgr *mgr = to_vram_mgr(man);
+ struct drm_buddy_mm *mm = &mgr->mm;
+ struct drm_buddy_block *block;
spin_lock(&mgr->lock);
- drm_mm_print(&mgr->mm, printer);
+ drm_printf(printer, "default_page_size: %lluKiB\n",
+ mgr->default_page_size >> 10);
+
+ drm_buddy_print(mm, printer);
+
+ drm_printf(printer, "reserved:\n");
+ list_for_each_entry(block, &mgr->reserved_pages, link)
+ drm_buddy_block_print(mm, block, printer);
spin_unlock(&mgr->lock);
drm_printf(printer, "man size:%llu pages, ram usage:%lluMB, vis usage:%lluMB\n",
@@ -696,15 +732,20 @@ int amdgpu_vram_mgr_init(struct amdgpu_device *adev)
{
struct amdgpu_vram_mgr *mgr = &adev->mman.vram_mgr;
struct ttm_resource_manager *man = &mgr->manager;
+ int err;
ttm_resource_manager_init(man, adev->gmc.real_vram_size >> PAGE_SHIFT);
man->func = &amdgpu_vram_mgr_func;
- drm_mm_init(&mgr->mm, 0, man->size);
+ err = drm_buddy_init(&mgr->mm, man->size << PAGE_SHIFT, PAGE_SIZE);
+ if (err)
+ return err;
+
spin_lock_init(&mgr->lock);
INIT_LIST_HEAD(&mgr->reservations_pending);
INIT_LIST_HEAD(&mgr->reserved_pages);
+ mgr->default_page_size = PAGE_SIZE;
ttm_set_driver_manager(&adev->mman.bdev, TTM_PL_VRAM, &mgr->manager);
ttm_resource_manager_set_used(man, true);
@@ -737,10 +778,10 @@ void amdgpu_vram_mgr_fini(struct amdgpu_device *adev)
kfree(rsv);
list_for_each_entry_safe(rsv, temp, &mgr->reserved_pages, node) {
- drm_mm_remove_node(&rsv->mm_node);
+ drm_buddy_free_list(&mgr->mm, &rsv->block);
kfree(rsv);
}
- drm_mm_takedown(&mgr->mm);
+ drm_buddy_fini(&mgr->mm);
spin_unlock(&mgr->lock);
ttm_resource_manager_cleanup(man);
--
2.25.1
^ permalink raw reply related [flat|nested] 58+ messages in thread
* [PATCH v2 1/8] drm: move the buddy allocator from i915 into common drm
2021-10-25 13:00 ` [Intel-gfx] " Arunpravin
@ 2021-10-25 13:00 ` Arunpravin
-1 siblings, 0 replies; 58+ messages in thread
From: Arunpravin @ 2021-10-25 13:00 UTC (permalink / raw)
To: dri-devel, intel-gfx, amd-gfx
Cc: daniel, christian.koenig, alexander.deucher, matthew.auld,
tzimmermann, jani.nikula, Arunpravin
Move the base i915 buddy allocator code into drm
- Move i915_buddy.h to include/drm
- Move i915_buddy.c to drm root folder
- Rename "i915" string with "drm" string wherever applicable
- Rename "I915" string with "DRM" string wherever applicable
- Fix header file dependencies
- Fix alignment issues
- add Makefile support for drm buddy
- export functions and write kerneldoc description
- Remove i915 selftest config check condition as buddy selftest
will be moved to drm selftest folder
cleanup i915 buddy references in i915 driver module
and replace with drm buddy
V2:
- include header file in alphabetical order (Thomas)
- merged changes listed in the body section into a single patch
to keep the build intact (Christian, Jani)
Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
---
drivers/gpu/drm/Makefile | 2 +-
drivers/gpu/drm/drm_buddy.c | 521 ++++++++++++++++++
drivers/gpu/drm/drm_drv.c | 3 +
drivers/gpu/drm/i915/Makefile | 1 -
drivers/gpu/drm/i915/i915_buddy.c | 466 ----------------
drivers/gpu/drm/i915/i915_buddy.h | 143 -----
drivers/gpu/drm/i915/i915_module.c | 3 -
drivers/gpu/drm/i915/i915_scatterlist.c | 11 +-
drivers/gpu/drm/i915/i915_ttm_buddy_manager.c | 33 +-
drivers/gpu/drm/i915/i915_ttm_buddy_manager.h | 4 +-
include/drm/drm_buddy.h | 153 +++++
11 files changed, 702 insertions(+), 638 deletions(-)
create mode 100644 drivers/gpu/drm/drm_buddy.c
delete mode 100644 drivers/gpu/drm/i915/i915_buddy.c
delete mode 100644 drivers/gpu/drm/i915/i915_buddy.h
create mode 100644 include/drm/drm_buddy.h
diff --git a/drivers/gpu/drm/Makefile b/drivers/gpu/drm/Makefile
index 0dff40bb863c..dc61e91a3154 100644
--- a/drivers/gpu/drm/Makefile
+++ b/drivers/gpu/drm/Makefile
@@ -18,7 +18,7 @@ drm-y := drm_aperture.o drm_auth.o drm_cache.o \
drm_dumb_buffers.o drm_mode_config.o drm_vblank.o \
drm_syncobj.o drm_lease.o drm_writeback.o drm_client.o \
drm_client_modeset.o drm_atomic_uapi.o drm_hdcp.o \
- drm_managed.o drm_vblank_work.o
+ drm_managed.o drm_vblank_work.o drm_buddy.o
drm-$(CONFIG_DRM_LEGACY) += drm_agpsupport.o drm_bufs.o drm_context.o drm_dma.o \
drm_legacy_misc.o drm_lock.o drm_memory.o drm_scatter.o \
diff --git a/drivers/gpu/drm/drm_buddy.c b/drivers/gpu/drm/drm_buddy.c
new file mode 100644
index 000000000000..39eb1d224bec
--- /dev/null
+++ b/drivers/gpu/drm/drm_buddy.c
@@ -0,0 +1,521 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2021 Intel Corporation
+ */
+
+#include <linux/kmemleak.h>
+#include <linux/sizes.h>
+
+#include <drm/drm_buddy.h>
+
+static struct kmem_cache *slab_blocks;
+
+static struct drm_buddy_block *drm_block_alloc(struct drm_buddy_mm *mm,
+ struct drm_buddy_block *parent,
+ unsigned int order,
+ u64 offset)
+{
+ struct drm_buddy_block *block;
+
+ BUG_ON(order > DRM_BUDDY_MAX_ORDER);
+
+ block = kmem_cache_zalloc(slab_blocks, GFP_KERNEL);
+ if (!block)
+ return NULL;
+
+ block->header = offset;
+ block->header |= order;
+ block->parent = parent;
+
+ BUG_ON(block->header & DRM_BUDDY_HEADER_UNUSED);
+ return block;
+}
+
+static void drm_block_free(struct drm_buddy_mm *mm,
+ struct drm_buddy_block *block)
+{
+ kmem_cache_free(slab_blocks, block);
+}
+
+static void mark_allocated(struct drm_buddy_block *block)
+{
+ block->header &= ~DRM_BUDDY_HEADER_STATE;
+ block->header |= DRM_BUDDY_ALLOCATED;
+
+ list_del(&block->link);
+}
+
+static void mark_free(struct drm_buddy_mm *mm,
+ struct drm_buddy_block *block)
+{
+ block->header &= ~DRM_BUDDY_HEADER_STATE;
+ block->header |= DRM_BUDDY_FREE;
+
+ list_add(&block->link,
+ &mm->free_list[drm_buddy_block_order(block)]);
+}
+
+static void mark_split(struct drm_buddy_block *block)
+{
+ block->header &= ~DRM_BUDDY_HEADER_STATE;
+ block->header |= DRM_BUDDY_SPLIT;
+
+ list_del(&block->link);
+}
+
+/**
+ * drm_buddy_init - init memory manager
+ *
+ * @mm: DRM buddy manager to initialize
+ * @size: size in bytes to manage
+ * @chunk_size: minimum page size in bytes for our allocations
+ *
+ * Initializes the memory manager and its resources.
+ *
+ * Returns:
+ * 0 on success, error code on failure.
+ */
+int drm_buddy_init(struct drm_buddy_mm *mm, u64 size, u64 chunk_size)
+{
+ unsigned int i;
+ u64 offset;
+
+ if (size < chunk_size)
+ return -EINVAL;
+
+ if (chunk_size < PAGE_SIZE)
+ return -EINVAL;
+
+ if (!is_power_of_2(chunk_size))
+ return -EINVAL;
+
+ size = round_down(size, chunk_size);
+
+ mm->size = size;
+ mm->avail = size;
+ mm->chunk_size = chunk_size;
+ mm->max_order = ilog2(size) - ilog2(chunk_size);
+
+ BUG_ON(mm->max_order > DRM_BUDDY_MAX_ORDER);
+
+ mm->free_list = kmalloc_array(mm->max_order + 1,
+ sizeof(struct list_head),
+ GFP_KERNEL);
+ if (!mm->free_list)
+ return -ENOMEM;
+
+ for (i = 0; i <= mm->max_order; ++i)
+ INIT_LIST_HEAD(&mm->free_list[i]);
+
+ mm->n_roots = hweight64(size);
+
+ mm->roots = kmalloc_array(mm->n_roots,
+ sizeof(struct drm_buddy_block *),
+ GFP_KERNEL);
+ if (!mm->roots)
+ goto out_free_list;
+
+ offset = 0;
+ i = 0;
+
+ /*
+ * Split into power-of-two blocks, in case we are given a size that is
+ * not itself a power-of-two.
+ */
+ do {
+ struct drm_buddy_block *root;
+ unsigned int order;
+ u64 root_size;
+
+ root_size = rounddown_pow_of_two(size);
+ order = ilog2(root_size) - ilog2(chunk_size);
+
+ root = drm_block_alloc(mm, NULL, order, offset);
+ if (!root)
+ goto out_free_roots;
+
+ mark_free(mm, root);
+
+ BUG_ON(i > mm->max_order);
+ BUG_ON(drm_buddy_block_size(mm, root) < chunk_size);
+
+ mm->roots[i] = root;
+
+ offset += root_size;
+ size -= root_size;
+ i++;
+ } while (size);
+
+ return 0;
+
+out_free_roots:
+ while (i--)
+ drm_block_free(mm, mm->roots[i]);
+ kfree(mm->roots);
+out_free_list:
+ kfree(mm->free_list);
+ return -ENOMEM;
+}
+EXPORT_SYMBOL(drm_buddy_init);
+
+/**
+ * drm_buddy_fini - tear down the memory manager
+ *
+ * @mm: DRM buddy manager to free
+ *
+ * Cleanup memory manager resources and the freelist
+ */
+void drm_buddy_fini(struct drm_buddy_mm *mm)
+{
+ int i;
+
+ for (i = 0; i < mm->n_roots; ++i) {
+ WARN_ON(!drm_buddy_block_is_free(mm->roots[i]));
+ drm_block_free(mm, mm->roots[i]);
+ }
+
+ WARN_ON(mm->avail != mm->size);
+
+ kfree(mm->roots);
+ kfree(mm->free_list);
+}
+EXPORT_SYMBOL(drm_buddy_fini);
+
+static int split_block(struct drm_buddy_mm *mm,
+ struct drm_buddy_block *block)
+{
+ unsigned int block_order = drm_buddy_block_order(block) - 1;
+ u64 offset = drm_buddy_block_offset(block);
+
+ BUG_ON(!drm_buddy_block_is_free(block));
+ BUG_ON(!drm_buddy_block_order(block));
+
+ block->left = drm_block_alloc(mm, block, block_order, offset);
+ if (!block->left)
+ return -ENOMEM;
+
+ block->right = drm_block_alloc(mm, block, block_order,
+ offset + (mm->chunk_size << block_order));
+ if (!block->right) {
+ drm_block_free(mm, block->left);
+ return -ENOMEM;
+ }
+
+ mark_free(mm, block->left);
+ mark_free(mm, block->right);
+
+ mark_split(block);
+
+ return 0;
+}
+
+static struct drm_buddy_block *
+get_buddy(struct drm_buddy_block *block)
+{
+ struct drm_buddy_block *parent;
+
+ parent = block->parent;
+ if (!parent)
+ return NULL;
+
+ if (parent->left == block)
+ return parent->right;
+
+ return parent->left;
+}
+
+static void __drm_buddy_free(struct drm_buddy_mm *mm,
+ struct drm_buddy_block *block)
+{
+ struct drm_buddy_block *parent;
+
+ while ((parent = block->parent)) {
+ struct drm_buddy_block *buddy;
+
+ buddy = get_buddy(block);
+
+ if (!drm_buddy_block_is_free(buddy))
+ break;
+
+ list_del(&buddy->link);
+
+ drm_block_free(mm, block);
+ drm_block_free(mm, buddy);
+
+ block = parent;
+ }
+
+ mark_free(mm, block);
+}
+
+void drm_buddy_free(struct drm_buddy_mm *mm,
+ struct drm_buddy_block *block)
+{
+ BUG_ON(!drm_buddy_block_is_allocated(block));
+ mm->avail += drm_buddy_block_size(mm, block);
+ __drm_buddy_free(mm, block);
+}
+
+/**
+ * drm_buddy_free_list - free blocks
+ *
+ * @mm: DRM buddy manager
+ * @objects: input list head to free blocks
+ */
+void drm_buddy_free_list(struct drm_buddy_mm *mm, struct list_head *objects)
+{
+ struct drm_buddy_block *block, *on;
+
+ list_for_each_entry_safe(block, on, objects, link) {
+ drm_buddy_free(mm, block);
+ cond_resched();
+ }
+ INIT_LIST_HEAD(objects);
+}
+EXPORT_SYMBOL(drm_buddy_free_list);
+
+/**
+ * drm_buddy_alloc - allocate power-of-two blocks
+ *
+ * @mm: DRM buddy manager to allocate from
+ * @order: size of the allocation
+ *
+ * The order value here translates to:
+ *
+ * 0 = 2^0 * mm->chunk_size
+ * 1 = 2^1 * mm->chunk_size
+ * 2 = 2^2 * mm->chunk_size
+ *
+ * Returns:
+ * allocated ptr to the &drm_buddy_block on success
+ */
+struct drm_buddy_block *
+drm_buddy_alloc(struct drm_buddy_mm *mm, unsigned int order)
+{
+ struct drm_buddy_block *block = NULL;
+ unsigned int i;
+ int err;
+
+ for (i = order; i <= mm->max_order; ++i) {
+ block = list_first_entry_or_null(&mm->free_list[i],
+ struct drm_buddy_block,
+ link);
+ if (block)
+ break;
+ }
+
+ if (!block)
+ return ERR_PTR(-ENOSPC);
+
+ BUG_ON(!drm_buddy_block_is_free(block));
+
+ while (i != order) {
+ err = split_block(mm, block);
+ if (unlikely(err))
+ goto out_free;
+
+ /* Go low */
+ block = block->left;
+ i--;
+ }
+
+ mark_allocated(block);
+ mm->avail -= drm_buddy_block_size(mm, block);
+ kmemleak_update_trace(block);
+ return block;
+
+out_free:
+ if (i != order)
+ __drm_buddy_free(mm, block);
+ return ERR_PTR(err);
+}
+EXPORT_SYMBOL(drm_buddy_alloc);
+
+static inline bool overlaps(u64 s1, u64 e1, u64 s2, u64 e2)
+{
+ return s1 <= e2 && e1 >= s2;
+}
+
+static inline bool contains(u64 s1, u64 e1, u64 s2, u64 e2)
+{
+ return s1 <= s2 && e1 >= e2;
+}
+
+/**
+ * drm_buddy_alloc_range - allocate range
+ *
+ * @mm: DRM buddy manager to allocate from
+ * @blocks: output list head to add allocated blocks
+ * @start: start of the allowed range for this block
+ * @size: size of the allocation
+ *
+ * Intended for pre-allocating portions of the address space, for example to
+ * reserve a block for the initial framebuffer or similar, hence the expectation
+ * here is that drm_buddy_alloc() is still the main vehicle for
+ * allocations, so if that's not the case then the drm_mm range allocator is
+ * probably a much better fit, and so you should probably go use that instead.
+ *
+ * Note that it's safe to chain together multiple alloc_ranges
+ * with the same blocks list
+ *
+ * Returns:
+ * 0 on success, error code on failure.
+ */
+int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
+ struct list_head *blocks,
+ u64 start, u64 size)
+{
+ struct drm_buddy_block *block;
+ struct drm_buddy_block *buddy;
+ LIST_HEAD(allocated);
+ LIST_HEAD(dfs);
+ u64 end;
+ int err;
+ int i;
+
+ if (size < mm->chunk_size)
+ return -EINVAL;
+
+ if (!IS_ALIGNED(size | start, mm->chunk_size))
+ return -EINVAL;
+
+ if (range_overflows(start, size, mm->size))
+ return -EINVAL;
+
+ for (i = 0; i < mm->n_roots; ++i)
+ list_add_tail(&mm->roots[i]->tmp_link, &dfs);
+
+ end = start + size - 1;
+
+ do {
+ u64 block_start;
+ u64 block_end;
+
+ block = list_first_entry_or_null(&dfs,
+ struct drm_buddy_block,
+ tmp_link);
+ if (!block)
+ break;
+
+ list_del(&block->tmp_link);
+
+ block_start = drm_buddy_block_offset(block);
+ block_end = block_start + drm_buddy_block_size(mm, block) - 1;
+
+ if (!overlaps(start, end, block_start, block_end))
+ continue;
+
+ if (drm_buddy_block_is_allocated(block)) {
+ err = -ENOSPC;
+ goto err_free;
+ }
+
+ if (contains(start, end, block_start, block_end)) {
+ if (!drm_buddy_block_is_free(block)) {
+ err = -ENOSPC;
+ goto err_free;
+ }
+
+ mark_allocated(block);
+ mm->avail -= drm_buddy_block_size(mm, block);
+ list_add_tail(&block->link, &allocated);
+ continue;
+ }
+
+ if (!drm_buddy_block_is_split(block)) {
+ err = split_block(mm, block);
+ if (unlikely(err))
+ goto err_undo;
+ }
+
+ list_add(&block->right->tmp_link, &dfs);
+ list_add(&block->left->tmp_link, &dfs);
+ } while (1);
+
+ list_splice_tail(&allocated, blocks);
+ return 0;
+
+err_undo:
+ /*
+ * We really don't want to leave around a bunch of split blocks, since
+ * bigger is better, so make sure we merge everything back before we
+ * free the allocated blocks.
+ */
+ buddy = get_buddy(block);
+ if (buddy &&
+ (drm_buddy_block_is_free(block) &&
+ drm_buddy_block_is_free(buddy)))
+ __drm_buddy_free(mm, block);
+
+err_free:
+ drm_buddy_free_list(mm, &allocated);
+ return err;
+}
+EXPORT_SYMBOL(drm_buddy_alloc_range);
+
+/**
+ * drm_buddy_block_print - print block information
+ *
+ * @mm: DRM buddy manager
+ * @block: DRM buddy block
+ * @p: DRM printer to use
+ */
+void drm_buddy_block_print(struct drm_buddy_mm *mm,
+ struct drm_buddy_block *block,
+ struct drm_printer *p)
+{
+ u64 start = drm_buddy_block_offset(block);
+ u64 size = drm_buddy_block_size(mm, block);
+
+ drm_printf(p, "%#018llx-%#018llx: %llu\n", start, start + size, size);
+}
+EXPORT_SYMBOL(drm_buddy_block_print);
+
+/**
+ * drm_buddy_print - print allocator state
+ *
+ * @mm: DRM buddy manager
+ * @p: DRM printer to use
+ */
+void drm_buddy_print(struct drm_buddy_mm *mm, struct drm_printer *p)
+{
+ int order;
+
+ drm_printf(p, "chunk_size: %lluKiB, total: %lluMiB, free: %lluMiB\n",
+ mm->chunk_size >> 10, mm->size >> 20, mm->avail >> 20);
+
+ for (order = mm->max_order; order >= 0; order--) {
+ struct drm_buddy_block *block;
+ u64 count = 0, free;
+
+ list_for_each_entry(block, &mm->free_list[order], link) {
+ BUG_ON(!drm_buddy_block_is_free(block));
+ count++;
+ }
+
+ drm_printf(p, "order-%d ", order);
+
+ free = count * (mm->chunk_size << order);
+ if (free < SZ_1M)
+ drm_printf(p, "free: %lluKiB", free >> 10);
+ else
+ drm_printf(p, "free: %lluMiB", free >> 20);
+
+ drm_printf(p, ", pages: %llu\n", count);
+ }
+}
+EXPORT_SYMBOL(drm_buddy_print);
+
+void drm_buddy_module_exit(void)
+{
+ kmem_cache_destroy(slab_blocks);
+}
+
+int __init drm_buddy_module_init(void)
+{
+ slab_blocks = KMEM_CACHE(drm_buddy_block, 0);
+ if (!slab_blocks)
+ return -ENOMEM;
+
+ return 0;
+}
diff --git a/drivers/gpu/drm/drm_drv.c b/drivers/gpu/drm/drm_drv.c
index 7a5097467ba5..5d5e46e60586 100644
--- a/drivers/gpu/drm/drm_drv.c
+++ b/drivers/gpu/drm/drm_drv.c
@@ -35,6 +35,7 @@
#include <linux/slab.h>
#include <linux/srcu.h>
+#include <drm/drm_buddy.h>
#include <drm/drm_cache.h>
#include <drm/drm_client.h>
#include <drm/drm_color_mgmt.h>
@@ -1034,6 +1035,7 @@ static void drm_core_exit(void)
drm_sysfs_destroy();
idr_destroy(&drm_minors_idr);
drm_connector_ida_destroy();
+ drm_buddy_module_exit();
}
static int __init drm_core_init(void)
@@ -1043,6 +1045,7 @@ static int __init drm_core_init(void)
drm_connector_ida_init();
idr_init(&drm_minors_idr);
drm_memcpy_init_early();
+ drm_buddy_module_init();
ret = drm_sysfs_init();
if (ret < 0) {
diff --git a/drivers/gpu/drm/i915/Makefile b/drivers/gpu/drm/i915/Makefile
index 467872cca027..fc5ca8c4ccb2 100644
--- a/drivers/gpu/drm/i915/Makefile
+++ b/drivers/gpu/drm/i915/Makefile
@@ -161,7 +161,6 @@ gem-y += \
i915-y += \
$(gem-y) \
i915_active.o \
- i915_buddy.o \
i915_cmd_parser.o \
i915_gem_evict.o \
i915_gem_gtt.o \
diff --git a/drivers/gpu/drm/i915/i915_buddy.c b/drivers/gpu/drm/i915/i915_buddy.c
deleted file mode 100644
index 6e2ad68f8f3f..000000000000
--- a/drivers/gpu/drm/i915/i915_buddy.c
+++ /dev/null
@@ -1,466 +0,0 @@
-// SPDX-License-Identifier: MIT
-/*
- * Copyright © 2021 Intel Corporation
- */
-
-#include <linux/kmemleak.h>
-#include <linux/sizes.h>
-
-#include "i915_buddy.h"
-
-#include "i915_gem.h"
-#include "i915_utils.h"
-
-static struct kmem_cache *slab_blocks;
-
-static struct i915_buddy_block *i915_block_alloc(struct i915_buddy_mm *mm,
- struct i915_buddy_block *parent,
- unsigned int order,
- u64 offset)
-{
- struct i915_buddy_block *block;
-
- GEM_BUG_ON(order > I915_BUDDY_MAX_ORDER);
-
- block = kmem_cache_zalloc(slab_blocks, GFP_KERNEL);
- if (!block)
- return NULL;
-
- block->header = offset;
- block->header |= order;
- block->parent = parent;
-
- GEM_BUG_ON(block->header & I915_BUDDY_HEADER_UNUSED);
- return block;
-}
-
-static void i915_block_free(struct i915_buddy_mm *mm,
- struct i915_buddy_block *block)
-{
- kmem_cache_free(slab_blocks, block);
-}
-
-static void mark_allocated(struct i915_buddy_block *block)
-{
- block->header &= ~I915_BUDDY_HEADER_STATE;
- block->header |= I915_BUDDY_ALLOCATED;
-
- list_del(&block->link);
-}
-
-static void mark_free(struct i915_buddy_mm *mm,
- struct i915_buddy_block *block)
-{
- block->header &= ~I915_BUDDY_HEADER_STATE;
- block->header |= I915_BUDDY_FREE;
-
- list_add(&block->link,
- &mm->free_list[i915_buddy_block_order(block)]);
-}
-
-static void mark_split(struct i915_buddy_block *block)
-{
- block->header &= ~I915_BUDDY_HEADER_STATE;
- block->header |= I915_BUDDY_SPLIT;
-
- list_del(&block->link);
-}
-
-int i915_buddy_init(struct i915_buddy_mm *mm, u64 size, u64 chunk_size)
-{
- unsigned int i;
- u64 offset;
-
- if (size < chunk_size)
- return -EINVAL;
-
- if (chunk_size < PAGE_SIZE)
- return -EINVAL;
-
- if (!is_power_of_2(chunk_size))
- return -EINVAL;
-
- size = round_down(size, chunk_size);
-
- mm->size = size;
- mm->avail = size;
- mm->chunk_size = chunk_size;
- mm->max_order = ilog2(size) - ilog2(chunk_size);
-
- GEM_BUG_ON(mm->max_order > I915_BUDDY_MAX_ORDER);
-
- mm->free_list = kmalloc_array(mm->max_order + 1,
- sizeof(struct list_head),
- GFP_KERNEL);
- if (!mm->free_list)
- return -ENOMEM;
-
- for (i = 0; i <= mm->max_order; ++i)
- INIT_LIST_HEAD(&mm->free_list[i]);
-
- mm->n_roots = hweight64(size);
-
- mm->roots = kmalloc_array(mm->n_roots,
- sizeof(struct i915_buddy_block *),
- GFP_KERNEL);
- if (!mm->roots)
- goto out_free_list;
-
- offset = 0;
- i = 0;
-
- /*
- * Split into power-of-two blocks, in case we are given a size that is
- * not itself a power-of-two.
- */
- do {
- struct i915_buddy_block *root;
- unsigned int order;
- u64 root_size;
-
- root_size = rounddown_pow_of_two(size);
- order = ilog2(root_size) - ilog2(chunk_size);
-
- root = i915_block_alloc(mm, NULL, order, offset);
- if (!root)
- goto out_free_roots;
-
- mark_free(mm, root);
-
- GEM_BUG_ON(i > mm->max_order);
- GEM_BUG_ON(i915_buddy_block_size(mm, root) < chunk_size);
-
- mm->roots[i] = root;
-
- offset += root_size;
- size -= root_size;
- i++;
- } while (size);
-
- return 0;
-
-out_free_roots:
- while (i--)
- i915_block_free(mm, mm->roots[i]);
- kfree(mm->roots);
-out_free_list:
- kfree(mm->free_list);
- return -ENOMEM;
-}
-
-void i915_buddy_fini(struct i915_buddy_mm *mm)
-{
- int i;
-
- for (i = 0; i < mm->n_roots; ++i) {
- GEM_WARN_ON(!i915_buddy_block_is_free(mm->roots[i]));
- i915_block_free(mm, mm->roots[i]);
- }
-
- GEM_WARN_ON(mm->avail != mm->size);
-
- kfree(mm->roots);
- kfree(mm->free_list);
-}
-
-static int split_block(struct i915_buddy_mm *mm,
- struct i915_buddy_block *block)
-{
- unsigned int block_order = i915_buddy_block_order(block) - 1;
- u64 offset = i915_buddy_block_offset(block);
-
- GEM_BUG_ON(!i915_buddy_block_is_free(block));
- GEM_BUG_ON(!i915_buddy_block_order(block));
-
- block->left = i915_block_alloc(mm, block, block_order, offset);
- if (!block->left)
- return -ENOMEM;
-
- block->right = i915_block_alloc(mm, block, block_order,
- offset + (mm->chunk_size << block_order));
- if (!block->right) {
- i915_block_free(mm, block->left);
- return -ENOMEM;
- }
-
- mark_free(mm, block->left);
- mark_free(mm, block->right);
-
- mark_split(block);
-
- return 0;
-}
-
-static struct i915_buddy_block *
-get_buddy(struct i915_buddy_block *block)
-{
- struct i915_buddy_block *parent;
-
- parent = block->parent;
- if (!parent)
- return NULL;
-
- if (parent->left == block)
- return parent->right;
-
- return parent->left;
-}
-
-static void __i915_buddy_free(struct i915_buddy_mm *mm,
- struct i915_buddy_block *block)
-{
- struct i915_buddy_block *parent;
-
- while ((parent = block->parent)) {
- struct i915_buddy_block *buddy;
-
- buddy = get_buddy(block);
-
- if (!i915_buddy_block_is_free(buddy))
- break;
-
- list_del(&buddy->link);
-
- i915_block_free(mm, block);
- i915_block_free(mm, buddy);
-
- block = parent;
- }
-
- mark_free(mm, block);
-}
-
-void i915_buddy_free(struct i915_buddy_mm *mm,
- struct i915_buddy_block *block)
-{
- GEM_BUG_ON(!i915_buddy_block_is_allocated(block));
- mm->avail += i915_buddy_block_size(mm, block);
- __i915_buddy_free(mm, block);
-}
-
-void i915_buddy_free_list(struct i915_buddy_mm *mm, struct list_head *objects)
-{
- struct i915_buddy_block *block, *on;
-
- list_for_each_entry_safe(block, on, objects, link) {
- i915_buddy_free(mm, block);
- cond_resched();
- }
- INIT_LIST_HEAD(objects);
-}
-
-/*
- * Allocate power-of-two block. The order value here translates to:
- *
- * 0 = 2^0 * mm->chunk_size
- * 1 = 2^1 * mm->chunk_size
- * 2 = 2^2 * mm->chunk_size
- * ...
- */
-struct i915_buddy_block *
-i915_buddy_alloc(struct i915_buddy_mm *mm, unsigned int order)
-{
- struct i915_buddy_block *block = NULL;
- unsigned int i;
- int err;
-
- for (i = order; i <= mm->max_order; ++i) {
- block = list_first_entry_or_null(&mm->free_list[i],
- struct i915_buddy_block,
- link);
- if (block)
- break;
- }
-
- if (!block)
- return ERR_PTR(-ENOSPC);
-
- GEM_BUG_ON(!i915_buddy_block_is_free(block));
-
- while (i != order) {
- err = split_block(mm, block);
- if (unlikely(err))
- goto out_free;
-
- /* Go low */
- block = block->left;
- i--;
- }
-
- mark_allocated(block);
- mm->avail -= i915_buddy_block_size(mm, block);
- kmemleak_update_trace(block);
- return block;
-
-out_free:
- if (i != order)
- __i915_buddy_free(mm, block);
- return ERR_PTR(err);
-}
-
-static inline bool overlaps(u64 s1, u64 e1, u64 s2, u64 e2)
-{
- return s1 <= e2 && e1 >= s2;
-}
-
-static inline bool contains(u64 s1, u64 e1, u64 s2, u64 e2)
-{
- return s1 <= s2 && e1 >= e2;
-}
-
-/*
- * Allocate range. Note that it's safe to chain together multiple alloc_ranges
- * with the same blocks list.
- *
- * Intended for pre-allocating portions of the address space, for example to
- * reserve a block for the initial framebuffer or similar, hence the expectation
- * here is that i915_buddy_alloc() is still the main vehicle for
- * allocations, so if that's not the case then the drm_mm range allocator is
- * probably a much better fit, and so you should probably go use that instead.
- */
-int i915_buddy_alloc_range(struct i915_buddy_mm *mm,
- struct list_head *blocks,
- u64 start, u64 size)
-{
- struct i915_buddy_block *block;
- struct i915_buddy_block *buddy;
- LIST_HEAD(allocated);
- LIST_HEAD(dfs);
- u64 end;
- int err;
- int i;
-
- if (size < mm->chunk_size)
- return -EINVAL;
-
- if (!IS_ALIGNED(size | start, mm->chunk_size))
- return -EINVAL;
-
- if (range_overflows(start, size, mm->size))
- return -EINVAL;
-
- for (i = 0; i < mm->n_roots; ++i)
- list_add_tail(&mm->roots[i]->tmp_link, &dfs);
-
- end = start + size - 1;
-
- do {
- u64 block_start;
- u64 block_end;
-
- block = list_first_entry_or_null(&dfs,
- struct i915_buddy_block,
- tmp_link);
- if (!block)
- break;
-
- list_del(&block->tmp_link);
-
- block_start = i915_buddy_block_offset(block);
- block_end = block_start + i915_buddy_block_size(mm, block) - 1;
-
- if (!overlaps(start, end, block_start, block_end))
- continue;
-
- if (i915_buddy_block_is_allocated(block)) {
- err = -ENOSPC;
- goto err_free;
- }
-
- if (contains(start, end, block_start, block_end)) {
- if (!i915_buddy_block_is_free(block)) {
- err = -ENOSPC;
- goto err_free;
- }
-
- mark_allocated(block);
- mm->avail -= i915_buddy_block_size(mm, block);
- list_add_tail(&block->link, &allocated);
- continue;
- }
-
- if (!i915_buddy_block_is_split(block)) {
- err = split_block(mm, block);
- if (unlikely(err))
- goto err_undo;
- }
-
- list_add(&block->right->tmp_link, &dfs);
- list_add(&block->left->tmp_link, &dfs);
- } while (1);
-
- list_splice_tail(&allocated, blocks);
- return 0;
-
-err_undo:
- /*
- * We really don't want to leave around a bunch of split blocks, since
- * bigger is better, so make sure we merge everything back before we
- * free the allocated blocks.
- */
- buddy = get_buddy(block);
- if (buddy &&
- (i915_buddy_block_is_free(block) &&
- i915_buddy_block_is_free(buddy)))
- __i915_buddy_free(mm, block);
-
-err_free:
- i915_buddy_free_list(mm, &allocated);
- return err;
-}
-
-void i915_buddy_block_print(struct i915_buddy_mm *mm,
- struct i915_buddy_block *block,
- struct drm_printer *p)
-{
- u64 start = i915_buddy_block_offset(block);
- u64 size = i915_buddy_block_size(mm, block);
-
- drm_printf(p, "%#018llx-%#018llx: %llu\n", start, start + size, size);
-}
-
-void i915_buddy_print(struct i915_buddy_mm *mm, struct drm_printer *p)
-{
- int order;
-
- drm_printf(p, "chunk_size: %lluKiB, total: %lluMiB, free: %lluMiB\n",
- mm->chunk_size >> 10, mm->size >> 20, mm->avail >> 20);
-
- for (order = mm->max_order; order >= 0; order--) {
- struct i915_buddy_block *block;
- u64 count = 0, free;
-
- list_for_each_entry(block, &mm->free_list[order], link) {
- GEM_BUG_ON(!i915_buddy_block_is_free(block));
- count++;
- }
-
- drm_printf(p, "order-%d ", order);
-
- free = count * (mm->chunk_size << order);
- if (free < SZ_1M)
- drm_printf(p, "free: %lluKiB", free >> 10);
- else
- drm_printf(p, "free: %lluMiB", free >> 20);
-
- drm_printf(p, ", pages: %llu\n", count);
- }
-}
-
-#if IS_ENABLED(CONFIG_DRM_I915_SELFTEST)
-#include "selftests/i915_buddy.c"
-#endif
-
-void i915_buddy_module_exit(void)
-{
- kmem_cache_destroy(slab_blocks);
-}
-
-int __init i915_buddy_module_init(void)
-{
- slab_blocks = KMEM_CACHE(i915_buddy_block, 0);
- if (!slab_blocks)
- return -ENOMEM;
-
- return 0;
-}
diff --git a/drivers/gpu/drm/i915/i915_buddy.h b/drivers/gpu/drm/i915/i915_buddy.h
deleted file mode 100644
index 7077742112ac..000000000000
--- a/drivers/gpu/drm/i915/i915_buddy.h
+++ /dev/null
@@ -1,143 +0,0 @@
-/* SPDX-License-Identifier: MIT */
-/*
- * Copyright © 2021 Intel Corporation
- */
-
-#ifndef __I915_BUDDY_H__
-#define __I915_BUDDY_H__
-
-#include <linux/bitops.h>
-#include <linux/list.h>
-#include <linux/slab.h>
-
-#include <drm/drm_print.h>
-
-struct i915_buddy_block {
-#define I915_BUDDY_HEADER_OFFSET GENMASK_ULL(63, 12)
-#define I915_BUDDY_HEADER_STATE GENMASK_ULL(11, 10)
-#define I915_BUDDY_ALLOCATED (1 << 10)
-#define I915_BUDDY_FREE (2 << 10)
-#define I915_BUDDY_SPLIT (3 << 10)
-/* Free to be used, if needed in the future */
-#define I915_BUDDY_HEADER_UNUSED GENMASK_ULL(9, 6)
-#define I915_BUDDY_HEADER_ORDER GENMASK_ULL(5, 0)
- u64 header;
-
- struct i915_buddy_block *left;
- struct i915_buddy_block *right;
- struct i915_buddy_block *parent;
-
- void *private; /* owned by creator */
-
- /*
- * While the block is allocated by the user through i915_buddy_alloc*,
- * the user has ownership of the link, for example to maintain within
- * a list, if so desired. As soon as the block is freed with
- * i915_buddy_free* ownership is given back to the mm.
- */
- struct list_head link;
- struct list_head tmp_link;
-};
-
-/* Order-zero must be at least PAGE_SIZE */
-#define I915_BUDDY_MAX_ORDER (63 - PAGE_SHIFT)
-
-/*
- * Binary Buddy System.
- *
- * Locking should be handled by the user, a simple mutex around
- * i915_buddy_alloc* and i915_buddy_free* should suffice.
- */
-struct i915_buddy_mm {
- /* Maintain a free list for each order. */
- struct list_head *free_list;
-
- /*
- * Maintain explicit binary tree(s) to track the allocation of the
- * address space. This gives us a simple way of finding a buddy block
- * and performing the potentially recursive merge step when freeing a
- * block. Nodes are either allocated or free, in which case they will
- * also exist on the respective free list.
- */
- struct i915_buddy_block **roots;
-
- /*
- * Anything from here is public, and remains static for the lifetime of
- * the mm. Everything above is considered do-not-touch.
- */
- unsigned int n_roots;
- unsigned int max_order;
-
- /* Must be at least PAGE_SIZE */
- u64 chunk_size;
- u64 size;
- u64 avail;
-};
-
-static inline u64
-i915_buddy_block_offset(struct i915_buddy_block *block)
-{
- return block->header & I915_BUDDY_HEADER_OFFSET;
-}
-
-static inline unsigned int
-i915_buddy_block_order(struct i915_buddy_block *block)
-{
- return block->header & I915_BUDDY_HEADER_ORDER;
-}
-
-static inline unsigned int
-i915_buddy_block_state(struct i915_buddy_block *block)
-{
- return block->header & I915_BUDDY_HEADER_STATE;
-}
-
-static inline bool
-i915_buddy_block_is_allocated(struct i915_buddy_block *block)
-{
- return i915_buddy_block_state(block) == I915_BUDDY_ALLOCATED;
-}
-
-static inline bool
-i915_buddy_block_is_free(struct i915_buddy_block *block)
-{
- return i915_buddy_block_state(block) == I915_BUDDY_FREE;
-}
-
-static inline bool
-i915_buddy_block_is_split(struct i915_buddy_block *block)
-{
- return i915_buddy_block_state(block) == I915_BUDDY_SPLIT;
-}
-
-static inline u64
-i915_buddy_block_size(struct i915_buddy_mm *mm,
- struct i915_buddy_block *block)
-{
- return mm->chunk_size << i915_buddy_block_order(block);
-}
-
-int i915_buddy_init(struct i915_buddy_mm *mm, u64 size, u64 chunk_size);
-
-void i915_buddy_fini(struct i915_buddy_mm *mm);
-
-struct i915_buddy_block *
-i915_buddy_alloc(struct i915_buddy_mm *mm, unsigned int order);
-
-int i915_buddy_alloc_range(struct i915_buddy_mm *mm,
- struct list_head *blocks,
- u64 start, u64 size);
-
-void i915_buddy_free(struct i915_buddy_mm *mm, struct i915_buddy_block *block);
-
-void i915_buddy_free_list(struct i915_buddy_mm *mm, struct list_head *objects);
-
-void i915_buddy_print(struct i915_buddy_mm *mm, struct drm_printer *p);
-void i915_buddy_block_print(struct i915_buddy_mm *mm,
- struct i915_buddy_block *block,
- struct drm_printer *p);
-
-void i915_buddy_module_exit(void);
-int i915_buddy_module_init(void);
-
-#endif
diff --git a/drivers/gpu/drm/i915/i915_module.c b/drivers/gpu/drm/i915/i915_module.c
index ab2295dd4500..121b4178c5ca 100644
--- a/drivers/gpu/drm/i915/i915_module.c
+++ b/drivers/gpu/drm/i915/i915_module.c
@@ -9,7 +9,6 @@
#include "gem/i915_gem_context.h"
#include "gem/i915_gem_object.h"
#include "i915_active.h"
-#include "i915_buddy.h"
#include "i915_params.h"
#include "i915_pci.h"
#include "i915_perf.h"
@@ -50,8 +49,6 @@ static const struct {
{ .init = i915_check_nomodeset },
{ .init = i915_active_module_init,
.exit = i915_active_module_exit },
- { .init = i915_buddy_module_init,
- .exit = i915_buddy_module_exit },
{ .init = i915_context_module_init,
.exit = i915_context_module_exit },
{ .init = i915_gem_context_module_init,
diff --git a/drivers/gpu/drm/i915/i915_scatterlist.c b/drivers/gpu/drm/i915/i915_scatterlist.c
index 4a6712dca838..84d622aa32d2 100644
--- a/drivers/gpu/drm/i915/i915_scatterlist.c
+++ b/drivers/gpu/drm/i915/i915_scatterlist.c
@@ -5,10 +5,9 @@
*/
#include "i915_scatterlist.h"
-
-#include "i915_buddy.h"
#include "i915_ttm_buddy_manager.h"
+#include <drm/drm_buddy.h>
#include <drm/drm_mm.h>
#include <linux/slab.h>
@@ -126,9 +125,9 @@ struct sg_table *i915_sg_from_buddy_resource(struct ttm_resource *res,
struct i915_ttm_buddy_resource *bman_res = to_ttm_buddy_resource(res);
const u64 size = res->num_pages << PAGE_SHIFT;
const u64 max_segment = rounddown(UINT_MAX, PAGE_SIZE);
- struct i915_buddy_mm *mm = bman_res->mm;
+ struct drm_buddy_mm *mm = bman_res->mm;
struct list_head *blocks = &bman_res->blocks;
- struct i915_buddy_block *block;
+ struct drm_buddy_block *block;
struct scatterlist *sg;
struct sg_table *st;
resource_size_t prev_end;
@@ -151,8 +150,8 @@ struct sg_table *i915_sg_from_buddy_resource(struct ttm_resource *res,
list_for_each_entry(block, blocks, link) {
u64 block_size, offset;
- block_size = min_t(u64, size, i915_buddy_block_size(mm, block));
- offset = i915_buddy_block_offset(block);
+ block_size = min_t(u64, size, drm_buddy_block_size(mm, block));
+ offset = drm_buddy_block_offset(block);
while (block_size) {
u64 len;
diff --git a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
index d59fbb019032..c4b70cb8c248 100644
--- a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
+++ b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
@@ -8,14 +8,15 @@
#include <drm/ttm/ttm_bo_driver.h>
#include <drm/ttm/ttm_placement.h>
+#include <drm/drm_buddy.h>
+
#include "i915_ttm_buddy_manager.h"
-#include "i915_buddy.h"
#include "i915_gem.h"
struct i915_ttm_buddy_manager {
struct ttm_resource_manager manager;
- struct i915_buddy_mm mm;
+ struct drm_buddy_mm mm;
struct list_head reserved;
struct mutex lock;
u64 default_page_size;
@@ -34,7 +35,7 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
{
struct i915_ttm_buddy_manager *bman = to_buddy_manager(man);
struct i915_ttm_buddy_resource *bman_res;
- struct i915_buddy_mm *mm = &bman->mm;
+ struct drm_buddy_mm *mm = &bman->mm;
unsigned long n_pages;
unsigned int min_order;
u64 min_page_size;
@@ -73,7 +74,7 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
n_pages = size >> ilog2(mm->chunk_size);
do {
- struct i915_buddy_block *block;
+ struct drm_buddy_block *block;
unsigned int order;
order = fls(n_pages) - 1;
@@ -82,7 +83,7 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
do {
mutex_lock(&bman->lock);
- block = i915_buddy_alloc(mm, order);
+ block = drm_buddy_alloc(mm, order);
mutex_unlock(&bman->lock);
if (!IS_ERR(block))
break;
@@ -106,7 +107,7 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
err_free_blocks:
mutex_lock(&bman->lock);
- i915_buddy_free_list(mm, &bman_res->blocks);
+ drm_buddy_free_list(mm, &bman_res->blocks);
mutex_unlock(&bman->lock);
err_free_res:
kfree(bman_res);
@@ -120,7 +121,7 @@ static void i915_ttm_buddy_man_free(struct ttm_resource_manager *man,
struct i915_ttm_buddy_manager *bman = to_buddy_manager(man);
mutex_lock(&bman->lock);
- i915_buddy_free_list(&bman->mm, &bman_res->blocks);
+ drm_buddy_free_list(&bman->mm, &bman_res->blocks);
mutex_unlock(&bman->lock);
kfree(bman_res);
@@ -130,17 +131,17 @@ static void i915_ttm_buddy_man_debug(struct ttm_resource_manager *man,
struct drm_printer *printer)
{
struct i915_ttm_buddy_manager *bman = to_buddy_manager(man);
- struct i915_buddy_block *block;
+ struct drm_buddy_block *block;
mutex_lock(&bman->lock);
drm_printf(printer, "default_page_size: %lluKiB\n",
bman->default_page_size >> 10);
- i915_buddy_print(&bman->mm, printer);
+ drm_buddy_print(&bman->mm, printer);
drm_printf(printer, "reserved:\n");
list_for_each_entry(block, &bman->reserved, link)
- i915_buddy_block_print(&bman->mm, block, printer);
+ drm_buddy_block_print(&bman->mm, block, printer);
mutex_unlock(&bman->lock);
}
@@ -190,7 +191,7 @@ int i915_ttm_buddy_man_init(struct ttm_device *bdev,
if (!bman)
return -ENOMEM;
- err = i915_buddy_init(&bman->mm, size, chunk_size);
+ err = drm_buddy_init(&bman->mm, size, chunk_size);
if (err)
goto err_free_bman;
@@ -228,7 +229,7 @@ int i915_ttm_buddy_man_fini(struct ttm_device *bdev, unsigned int type)
{
struct ttm_resource_manager *man = ttm_manager_type(bdev, type);
struct i915_ttm_buddy_manager *bman = to_buddy_manager(man);
- struct i915_buddy_mm *mm = &bman->mm;
+ struct drm_buddy_mm *mm = &bman->mm;
int ret;
ttm_resource_manager_set_used(man, false);
@@ -240,8 +241,8 @@ int i915_ttm_buddy_man_fini(struct ttm_device *bdev, unsigned int type)
ttm_set_driver_manager(bdev, type, NULL);
mutex_lock(&bman->lock);
- i915_buddy_free_list(mm, &bman->reserved);
- i915_buddy_fini(mm);
+ drm_buddy_free_list(mm, &bman->reserved);
+ drm_buddy_fini(mm);
mutex_unlock(&bman->lock);
ttm_resource_manager_cleanup(man);
@@ -264,11 +265,11 @@ int i915_ttm_buddy_man_reserve(struct ttm_resource_manager *man,
u64 start, u64 size)
{
struct i915_ttm_buddy_manager *bman = to_buddy_manager(man);
- struct i915_buddy_mm *mm = &bman->mm;
+ struct drm_buddy_mm *mm = &bman->mm;
int ret;
mutex_lock(&bman->lock);
- ret = i915_buddy_alloc_range(mm, &bman->reserved, start, size);
+ ret = drm_buddy_alloc_range(mm, &bman->reserved, start, size);
mutex_unlock(&bman->lock);
return ret;
diff --git a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.h b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.h
index 0722d33f3e14..fa644b512c2e 100644
--- a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.h
+++ b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.h
@@ -13,7 +13,7 @@
struct ttm_device;
struct ttm_resource_manager;
-struct i915_buddy_mm;
+struct drm_buddy_mm;
/**
* struct i915_ttm_buddy_resource
@@ -28,7 +28,7 @@ struct i915_buddy_mm;
struct i915_ttm_buddy_resource {
struct ttm_resource base;
struct list_head blocks;
- struct i915_buddy_mm *mm;
+ struct drm_buddy_mm *mm;
};
/**
diff --git a/include/drm/drm_buddy.h b/include/drm/drm_buddy.h
new file mode 100644
index 000000000000..5ce3fc702f80
--- /dev/null
+++ b/include/drm/drm_buddy.h
@@ -0,0 +1,153 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2021 Intel Corporation
+ */
+
+#ifndef __DRM_BUDDY_H__
+#define __DRM_BUDDY_H__
+
+#include <linux/bitops.h>
+#include <linux/list.h>
+#include <linux/slab.h>
+#include <linux/sched.h>
+
+#include <drm/drm_print.h>
+
+#define range_overflows(start, size, max) ({ \
+ typeof(start) start__ = (start); \
+ typeof(size) size__ = (size); \
+ typeof(max) max__ = (max); \
+ (void)(&start__ == &size__); \
+ (void)(&start__ == &max__); \
+ start__ >= max__ || size__ > max__ - start__; \
+})
+
+struct drm_buddy_block {
+#define DRM_BUDDY_HEADER_OFFSET GENMASK_ULL(63, 12)
+#define DRM_BUDDY_HEADER_STATE GENMASK_ULL(11, 10)
+#define DRM_BUDDY_ALLOCATED (1 << 10)
+#define DRM_BUDDY_FREE (2 << 10)
+#define DRM_BUDDY_SPLIT (3 << 10)
+/* Free to be used, if needed in the future */
+#define DRM_BUDDY_HEADER_UNUSED GENMASK_ULL(9, 6)
+#define DRM_BUDDY_HEADER_ORDER GENMASK_ULL(5, 0)
+ u64 header;
+
+ struct drm_buddy_block *left;
+ struct drm_buddy_block *right;
+ struct drm_buddy_block *parent;
+
+ void *private; /* owned by creator */
+
+ /*
+ * While the block is allocated by the user through drm_buddy_alloc*,
+ * the user has ownership of the link, for example to maintain within
+ * a list, if so desired. As soon as the block is freed with
+ * drm_buddy_free* ownership is given back to the mm.
+ */
+ struct list_head link;
+ struct list_head tmp_link;
+};
+
+/* Order-zero must be at least PAGE_SIZE */
+#define DRM_BUDDY_MAX_ORDER (63 - PAGE_SHIFT)
+
+/*
+ * Binary Buddy System.
+ *
+ * Locking should be handled by the user, a simple mutex around
+ * drm_buddy_alloc* and drm_buddy_free* should suffice.
+ */
+struct drm_buddy_mm {
+ /* Maintain a free list for each order. */
+ struct list_head *free_list;
+
+ /*
+ * Maintain explicit binary tree(s) to track the allocation of the
+ * address space. This gives us a simple way of finding a buddy block
+ * and performing the potentially recursive merge step when freeing a
+ * block. Nodes are either allocated or free, in which case they will
+ * also exist on the respective free list.
+ */
+ struct drm_buddy_block **roots;
+
+ /*
+ * Anything from here is public, and remains static for the lifetime of
+ * the mm. Everything above is considered do-not-touch.
+ */
+ unsigned int n_roots;
+ unsigned int max_order;
+
+ /* Must be at least PAGE_SIZE */
+ u64 chunk_size;
+ u64 size;
+ u64 avail;
+};
+
+static inline u64
+drm_buddy_block_offset(struct drm_buddy_block *block)
+{
+ return block->header & DRM_BUDDY_HEADER_OFFSET;
+}
+
+static inline unsigned int
+drm_buddy_block_order(struct drm_buddy_block *block)
+{
+ return block->header & DRM_BUDDY_HEADER_ORDER;
+}
+
+static inline unsigned int
+drm_buddy_block_state(struct drm_buddy_block *block)
+{
+ return block->header & DRM_BUDDY_HEADER_STATE;
+}
+
+static inline bool
+drm_buddy_block_is_allocated(struct drm_buddy_block *block)
+{
+ return drm_buddy_block_state(block) == DRM_BUDDY_ALLOCATED;
+}
+
+static inline bool
+drm_buddy_block_is_free(struct drm_buddy_block *block)
+{
+ return drm_buddy_block_state(block) == DRM_BUDDY_FREE;
+}
+
+static inline bool
+drm_buddy_block_is_split(struct drm_buddy_block *block)
+{
+ return drm_buddy_block_state(block) == DRM_BUDDY_SPLIT;
+}
+
+static inline u64
+drm_buddy_block_size(struct drm_buddy_mm *mm,
+ struct drm_buddy_block *block)
+{
+ return mm->chunk_size << drm_buddy_block_order(block);
+}
+
+int drm_buddy_init(struct drm_buddy_mm *mm, u64 size, u64 chunk_size);
+
+void drm_buddy_fini(struct drm_buddy_mm *mm);
+
+struct drm_buddy_block *
+drm_buddy_alloc(struct drm_buddy_mm *mm, unsigned int order);
+
+int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
+ struct list_head *blocks,
+ u64 start, u64 size);
+
+void drm_buddy_free(struct drm_buddy_mm *mm, struct drm_buddy_block *block);
+
+void drm_buddy_free_list(struct drm_buddy_mm *mm, struct list_head *objects);
+
+void drm_buddy_print(struct drm_buddy_mm *mm, struct drm_printer *p);
+void drm_buddy_block_print(struct drm_buddy_mm *mm,
+ struct drm_buddy_block *block,
+ struct drm_printer *p);
+
+void drm_buddy_module_exit(void);
+int drm_buddy_module_init(void);
+
+#endif
--
2.25.1
^ permalink raw reply related [flat|nested] 58+ messages in thread
* [Intel-gfx] [PATCH v2 1/8] drm: move the buddy allocator from i915 into common drm
@ 2021-10-25 13:00 ` Arunpravin
0 siblings, 0 replies; 58+ messages in thread
From: Arunpravin @ 2021-10-25 13:00 UTC (permalink / raw)
To: dri-devel, intel-gfx, amd-gfx
Cc: daniel, christian.koenig, alexander.deucher, matthew.auld,
tzimmermann, jani.nikula, Arunpravin
Move the base i915 buddy allocator code into drm
- Move i915_buddy.h to include/drm
- Move i915_buddy.c to drm root folder
- Rename "i915" string with "drm" string wherever applicable
- Rename "I915" string with "DRM" string wherever applicable
- Fix header file dependencies
- Fix alignment issues
- add Makefile support for drm buddy
- export functions and write kerneldoc description
- Remove i915 selftest config check condition as buddy selftest
will be moved to drm selftest folder
cleanup i915 buddy references in i915 driver module
and replace with drm buddy
V2:
- include header file in alphabetical order (Thomas)
- merged changes listed in the body section into a single patch
to keep the build intact (Christian, Jani)
Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
---
drivers/gpu/drm/Makefile | 2 +-
drivers/gpu/drm/drm_buddy.c | 521 ++++++++++++++++++
drivers/gpu/drm/drm_drv.c | 3 +
drivers/gpu/drm/i915/Makefile | 1 -
drivers/gpu/drm/i915/i915_buddy.c | 466 ----------------
drivers/gpu/drm/i915/i915_buddy.h | 143 -----
drivers/gpu/drm/i915/i915_module.c | 3 -
drivers/gpu/drm/i915/i915_scatterlist.c | 11 +-
drivers/gpu/drm/i915/i915_ttm_buddy_manager.c | 33 +-
drivers/gpu/drm/i915/i915_ttm_buddy_manager.h | 4 +-
include/drm/drm_buddy.h | 153 +++++
11 files changed, 702 insertions(+), 638 deletions(-)
create mode 100644 drivers/gpu/drm/drm_buddy.c
delete mode 100644 drivers/gpu/drm/i915/i915_buddy.c
delete mode 100644 drivers/gpu/drm/i915/i915_buddy.h
create mode 100644 include/drm/drm_buddy.h
diff --git a/drivers/gpu/drm/Makefile b/drivers/gpu/drm/Makefile
index 0dff40bb863c..dc61e91a3154 100644
--- a/drivers/gpu/drm/Makefile
+++ b/drivers/gpu/drm/Makefile
@@ -18,7 +18,7 @@ drm-y := drm_aperture.o drm_auth.o drm_cache.o \
drm_dumb_buffers.o drm_mode_config.o drm_vblank.o \
drm_syncobj.o drm_lease.o drm_writeback.o drm_client.o \
drm_client_modeset.o drm_atomic_uapi.o drm_hdcp.o \
- drm_managed.o drm_vblank_work.o
+ drm_managed.o drm_vblank_work.o drm_buddy.o
drm-$(CONFIG_DRM_LEGACY) += drm_agpsupport.o drm_bufs.o drm_context.o drm_dma.o \
drm_legacy_misc.o drm_lock.o drm_memory.o drm_scatter.o \
diff --git a/drivers/gpu/drm/drm_buddy.c b/drivers/gpu/drm/drm_buddy.c
new file mode 100644
index 000000000000..39eb1d224bec
--- /dev/null
+++ b/drivers/gpu/drm/drm_buddy.c
@@ -0,0 +1,521 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2021 Intel Corporation
+ */
+
+#include <linux/kmemleak.h>
+#include <linux/sizes.h>
+
+#include <drm/drm_buddy.h>
+
+static struct kmem_cache *slab_blocks;
+
+static struct drm_buddy_block *drm_block_alloc(struct drm_buddy_mm *mm,
+ struct drm_buddy_block *parent,
+ unsigned int order,
+ u64 offset)
+{
+ struct drm_buddy_block *block;
+
+ BUG_ON(order > DRM_BUDDY_MAX_ORDER);
+
+ block = kmem_cache_zalloc(slab_blocks, GFP_KERNEL);
+ if (!block)
+ return NULL;
+
+ block->header = offset;
+ block->header |= order;
+ block->parent = parent;
+
+ BUG_ON(block->header & DRM_BUDDY_HEADER_UNUSED);
+ return block;
+}
+
+static void drm_block_free(struct drm_buddy_mm *mm,
+ struct drm_buddy_block *block)
+{
+ kmem_cache_free(slab_blocks, block);
+}
+
+static void mark_allocated(struct drm_buddy_block *block)
+{
+ block->header &= ~DRM_BUDDY_HEADER_STATE;
+ block->header |= DRM_BUDDY_ALLOCATED;
+
+ list_del(&block->link);
+}
+
+static void mark_free(struct drm_buddy_mm *mm,
+ struct drm_buddy_block *block)
+{
+ block->header &= ~DRM_BUDDY_HEADER_STATE;
+ block->header |= DRM_BUDDY_FREE;
+
+ list_add(&block->link,
+ &mm->free_list[drm_buddy_block_order(block)]);
+}
+
+static void mark_split(struct drm_buddy_block *block)
+{
+ block->header &= ~DRM_BUDDY_HEADER_STATE;
+ block->header |= DRM_BUDDY_SPLIT;
+
+ list_del(&block->link);
+}
+
+/**
+ * drm_buddy_init - init memory manager
+ *
+ * @mm: DRM buddy manager to initialize
+ * @size: size in bytes to manage
+ * @chunk_size: minimum page size in bytes for our allocations
+ *
+ * Initializes the memory manager and its resources.
+ *
+ * Returns:
+ * 0 on success, error code on failure.
+ */
+int drm_buddy_init(struct drm_buddy_mm *mm, u64 size, u64 chunk_size)
+{
+ unsigned int i;
+ u64 offset;
+
+ if (size < chunk_size)
+ return -EINVAL;
+
+ if (chunk_size < PAGE_SIZE)
+ return -EINVAL;
+
+ if (!is_power_of_2(chunk_size))
+ return -EINVAL;
+
+ size = round_down(size, chunk_size);
+
+ mm->size = size;
+ mm->avail = size;
+ mm->chunk_size = chunk_size;
+ mm->max_order = ilog2(size) - ilog2(chunk_size);
+
+ BUG_ON(mm->max_order > DRM_BUDDY_MAX_ORDER);
+
+ mm->free_list = kmalloc_array(mm->max_order + 1,
+ sizeof(struct list_head),
+ GFP_KERNEL);
+ if (!mm->free_list)
+ return -ENOMEM;
+
+ for (i = 0; i <= mm->max_order; ++i)
+ INIT_LIST_HEAD(&mm->free_list[i]);
+
+ mm->n_roots = hweight64(size);
+
+ mm->roots = kmalloc_array(mm->n_roots,
+ sizeof(struct drm_buddy_block *),
+ GFP_KERNEL);
+ if (!mm->roots)
+ goto out_free_list;
+
+ offset = 0;
+ i = 0;
+
+ /*
+ * Split into power-of-two blocks, in case we are given a size that is
+ * not itself a power-of-two.
+ */
+ do {
+ struct drm_buddy_block *root;
+ unsigned int order;
+ u64 root_size;
+
+ root_size = rounddown_pow_of_two(size);
+ order = ilog2(root_size) - ilog2(chunk_size);
+
+ root = drm_block_alloc(mm, NULL, order, offset);
+ if (!root)
+ goto out_free_roots;
+
+ mark_free(mm, root);
+
+ BUG_ON(i > mm->max_order);
+ BUG_ON(drm_buddy_block_size(mm, root) < chunk_size);
+
+ mm->roots[i] = root;
+
+ offset += root_size;
+ size -= root_size;
+ i++;
+ } while (size);
+
+ return 0;
+
+out_free_roots:
+ while (i--)
+ drm_block_free(mm, mm->roots[i]);
+ kfree(mm->roots);
+out_free_list:
+ kfree(mm->free_list);
+ return -ENOMEM;
+}
+EXPORT_SYMBOL(drm_buddy_init);
+
+/**
+ * drm_buddy_fini - tear down the memory manager
+ *
+ * @mm: DRM buddy manager to free
+ *
+ * Cleanup memory manager resources and the freelist
+ */
+void drm_buddy_fini(struct drm_buddy_mm *mm)
+{
+ int i;
+
+ for (i = 0; i < mm->n_roots; ++i) {
+ WARN_ON(!drm_buddy_block_is_free(mm->roots[i]));
+ drm_block_free(mm, mm->roots[i]);
+ }
+
+ WARN_ON(mm->avail != mm->size);
+
+ kfree(mm->roots);
+ kfree(mm->free_list);
+}
+EXPORT_SYMBOL(drm_buddy_fini);
+
+static int split_block(struct drm_buddy_mm *mm,
+ struct drm_buddy_block *block)
+{
+ unsigned int block_order = drm_buddy_block_order(block) - 1;
+ u64 offset = drm_buddy_block_offset(block);
+
+ BUG_ON(!drm_buddy_block_is_free(block));
+ BUG_ON(!drm_buddy_block_order(block));
+
+ block->left = drm_block_alloc(mm, block, block_order, offset);
+ if (!block->left)
+ return -ENOMEM;
+
+ block->right = drm_block_alloc(mm, block, block_order,
+ offset + (mm->chunk_size << block_order));
+ if (!block->right) {
+ drm_block_free(mm, block->left);
+ return -ENOMEM;
+ }
+
+ mark_free(mm, block->left);
+ mark_free(mm, block->right);
+
+ mark_split(block);
+
+ return 0;
+}
+
+static struct drm_buddy_block *
+get_buddy(struct drm_buddy_block *block)
+{
+ struct drm_buddy_block *parent;
+
+ parent = block->parent;
+ if (!parent)
+ return NULL;
+
+ if (parent->left == block)
+ return parent->right;
+
+ return parent->left;
+}
+
+static void __drm_buddy_free(struct drm_buddy_mm *mm,
+ struct drm_buddy_block *block)
+{
+ struct drm_buddy_block *parent;
+
+ while ((parent = block->parent)) {
+ struct drm_buddy_block *buddy;
+
+ buddy = get_buddy(block);
+
+ if (!drm_buddy_block_is_free(buddy))
+ break;
+
+ list_del(&buddy->link);
+
+ drm_block_free(mm, block);
+ drm_block_free(mm, buddy);
+
+ block = parent;
+ }
+
+ mark_free(mm, block);
+}
+
+void drm_buddy_free(struct drm_buddy_mm *mm,
+ struct drm_buddy_block *block)
+{
+ BUG_ON(!drm_buddy_block_is_allocated(block));
+ mm->avail += drm_buddy_block_size(mm, block);
+ __drm_buddy_free(mm, block);
+}
+
+/**
+ * drm_buddy_free_list - free blocks
+ *
+ * @mm: DRM buddy manager
+ * @objects: input list head to free blocks
+ */
+void drm_buddy_free_list(struct drm_buddy_mm *mm, struct list_head *objects)
+{
+ struct drm_buddy_block *block, *on;
+
+ list_for_each_entry_safe(block, on, objects, link) {
+ drm_buddy_free(mm, block);
+ cond_resched();
+ }
+ INIT_LIST_HEAD(objects);
+}
+EXPORT_SYMBOL(drm_buddy_free_list);
+
+/**
+ * drm_buddy_alloc - allocate power-of-two blocks
+ *
+ * @mm: DRM buddy manager to allocate from
+ * @order: size of the allocation
+ *
+ * The order value here translates to:
+ *
+ * 0 = 2^0 * mm->chunk_size
+ * 1 = 2^1 * mm->chunk_size
+ * 2 = 2^2 * mm->chunk_size
+ *
+ * Returns:
+ * allocated ptr to the &drm_buddy_block on success
+ */
+struct drm_buddy_block *
+drm_buddy_alloc(struct drm_buddy_mm *mm, unsigned int order)
+{
+ struct drm_buddy_block *block = NULL;
+ unsigned int i;
+ int err;
+
+ for (i = order; i <= mm->max_order; ++i) {
+ block = list_first_entry_or_null(&mm->free_list[i],
+ struct drm_buddy_block,
+ link);
+ if (block)
+ break;
+ }
+
+ if (!block)
+ return ERR_PTR(-ENOSPC);
+
+ BUG_ON(!drm_buddy_block_is_free(block));
+
+ while (i != order) {
+ err = split_block(mm, block);
+ if (unlikely(err))
+ goto out_free;
+
+ /* Go low */
+ block = block->left;
+ i--;
+ }
+
+ mark_allocated(block);
+ mm->avail -= drm_buddy_block_size(mm, block);
+ kmemleak_update_trace(block);
+ return block;
+
+out_free:
+ if (i != order)
+ __drm_buddy_free(mm, block);
+ return ERR_PTR(err);
+}
+EXPORT_SYMBOL(drm_buddy_alloc);
+
+static inline bool overlaps(u64 s1, u64 e1, u64 s2, u64 e2)
+{
+ return s1 <= e2 && e1 >= s2;
+}
+
+static inline bool contains(u64 s1, u64 e1, u64 s2, u64 e2)
+{
+ return s1 <= s2 && e1 >= e2;
+}
+
+/**
+ * drm_buddy_alloc_range - allocate range
+ *
+ * @mm: DRM buddy manager to allocate from
+ * @blocks: output list head to add allocated blocks
+ * @start: start of the allowed range for this block
+ * @size: size of the allocation
+ *
+ * Intended for pre-allocating portions of the address space, for example to
+ * reserve a block for the initial framebuffer or similar, hence the expectation
+ * here is that drm_buddy_alloc() is still the main vehicle for
+ * allocations, so if that's not the case then the drm_mm range allocator is
+ * probably a much better fit, and so you should probably go use that instead.
+ *
+ * Note that it's safe to chain together multiple alloc_ranges
+ * with the same blocks list
+ *
+ * Returns:
+ * 0 on success, error code on failure.
+ */
+int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
+ struct list_head *blocks,
+ u64 start, u64 size)
+{
+ struct drm_buddy_block *block;
+ struct drm_buddy_block *buddy;
+ LIST_HEAD(allocated);
+ LIST_HEAD(dfs);
+ u64 end;
+ int err;
+ int i;
+
+ if (size < mm->chunk_size)
+ return -EINVAL;
+
+ if (!IS_ALIGNED(size | start, mm->chunk_size))
+ return -EINVAL;
+
+ if (range_overflows(start, size, mm->size))
+ return -EINVAL;
+
+ for (i = 0; i < mm->n_roots; ++i)
+ list_add_tail(&mm->roots[i]->tmp_link, &dfs);
+
+ end = start + size - 1;
+
+ do {
+ u64 block_start;
+ u64 block_end;
+
+ block = list_first_entry_or_null(&dfs,
+ struct drm_buddy_block,
+ tmp_link);
+ if (!block)
+ break;
+
+ list_del(&block->tmp_link);
+
+ block_start = drm_buddy_block_offset(block);
+ block_end = block_start + drm_buddy_block_size(mm, block) - 1;
+
+ if (!overlaps(start, end, block_start, block_end))
+ continue;
+
+ if (drm_buddy_block_is_allocated(block)) {
+ err = -ENOSPC;
+ goto err_free;
+ }
+
+ if (contains(start, end, block_start, block_end)) {
+ if (!drm_buddy_block_is_free(block)) {
+ err = -ENOSPC;
+ goto err_free;
+ }
+
+ mark_allocated(block);
+ mm->avail -= drm_buddy_block_size(mm, block);
+ list_add_tail(&block->link, &allocated);
+ continue;
+ }
+
+ if (!drm_buddy_block_is_split(block)) {
+ err = split_block(mm, block);
+ if (unlikely(err))
+ goto err_undo;
+ }
+
+ list_add(&block->right->tmp_link, &dfs);
+ list_add(&block->left->tmp_link, &dfs);
+ } while (1);
+
+ list_splice_tail(&allocated, blocks);
+ return 0;
+
+err_undo:
+ /*
+ * We really don't want to leave around a bunch of split blocks, since
+ * bigger is better, so make sure we merge everything back before we
+ * free the allocated blocks.
+ */
+ buddy = get_buddy(block);
+ if (buddy &&
+ (drm_buddy_block_is_free(block) &&
+ drm_buddy_block_is_free(buddy)))
+ __drm_buddy_free(mm, block);
+
+err_free:
+ drm_buddy_free_list(mm, &allocated);
+ return err;
+}
+EXPORT_SYMBOL(drm_buddy_alloc_range);
+
+/**
+ * drm_buddy_block_print - print block information
+ *
+ * @mm: DRM buddy manager
+ * @block: DRM buddy block
+ * @p: DRM printer to use
+ */
+void drm_buddy_block_print(struct drm_buddy_mm *mm,
+ struct drm_buddy_block *block,
+ struct drm_printer *p)
+{
+ u64 start = drm_buddy_block_offset(block);
+ u64 size = drm_buddy_block_size(mm, block);
+
+ drm_printf(p, "%#018llx-%#018llx: %llu\n", start, start + size, size);
+}
+EXPORT_SYMBOL(drm_buddy_block_print);
+
+/**
+ * drm_buddy_print - print allocator state
+ *
+ * @mm: DRM buddy manager
+ * @p: DRM printer to use
+ */
+void drm_buddy_print(struct drm_buddy_mm *mm, struct drm_printer *p)
+{
+ int order;
+
+ drm_printf(p, "chunk_size: %lluKiB, total: %lluMiB, free: %lluMiB\n",
+ mm->chunk_size >> 10, mm->size >> 20, mm->avail >> 20);
+
+ for (order = mm->max_order; order >= 0; order--) {
+ struct drm_buddy_block *block;
+ u64 count = 0, free;
+
+ list_for_each_entry(block, &mm->free_list[order], link) {
+ BUG_ON(!drm_buddy_block_is_free(block));
+ count++;
+ }
+
+ drm_printf(p, "order-%d ", order);
+
+ free = count * (mm->chunk_size << order);
+ if (free < SZ_1M)
+ drm_printf(p, "free: %lluKiB", free >> 10);
+ else
+ drm_printf(p, "free: %lluMiB", free >> 20);
+
+ drm_printf(p, ", pages: %llu\n", count);
+ }
+}
+EXPORT_SYMBOL(drm_buddy_print);
+
+void drm_buddy_module_exit(void)
+{
+ kmem_cache_destroy(slab_blocks);
+}
+
+int __init drm_buddy_module_init(void)
+{
+ slab_blocks = KMEM_CACHE(drm_buddy_block, 0);
+ if (!slab_blocks)
+ return -ENOMEM;
+
+ return 0;
+}
diff --git a/drivers/gpu/drm/drm_drv.c b/drivers/gpu/drm/drm_drv.c
index 7a5097467ba5..5d5e46e60586 100644
--- a/drivers/gpu/drm/drm_drv.c
+++ b/drivers/gpu/drm/drm_drv.c
@@ -35,6 +35,7 @@
#include <linux/slab.h>
#include <linux/srcu.h>
+#include <drm/drm_buddy.h>
#include <drm/drm_cache.h>
#include <drm/drm_client.h>
#include <drm/drm_color_mgmt.h>
@@ -1034,6 +1035,7 @@ static void drm_core_exit(void)
drm_sysfs_destroy();
idr_destroy(&drm_minors_idr);
drm_connector_ida_destroy();
+ drm_buddy_module_exit();
}
static int __init drm_core_init(void)
@@ -1043,6 +1045,7 @@ static int __init drm_core_init(void)
drm_connector_ida_init();
idr_init(&drm_minors_idr);
drm_memcpy_init_early();
+ drm_buddy_module_init();
ret = drm_sysfs_init();
if (ret < 0) {
diff --git a/drivers/gpu/drm/i915/Makefile b/drivers/gpu/drm/i915/Makefile
index 467872cca027..fc5ca8c4ccb2 100644
--- a/drivers/gpu/drm/i915/Makefile
+++ b/drivers/gpu/drm/i915/Makefile
@@ -161,7 +161,6 @@ gem-y += \
i915-y += \
$(gem-y) \
i915_active.o \
- i915_buddy.o \
i915_cmd_parser.o \
i915_gem_evict.o \
i915_gem_gtt.o \
diff --git a/drivers/gpu/drm/i915/i915_buddy.c b/drivers/gpu/drm/i915/i915_buddy.c
deleted file mode 100644
index 6e2ad68f8f3f..000000000000
--- a/drivers/gpu/drm/i915/i915_buddy.c
+++ /dev/null
@@ -1,466 +0,0 @@
-// SPDX-License-Identifier: MIT
-/*
- * Copyright © 2021 Intel Corporation
- */
-
-#include <linux/kmemleak.h>
-#include <linux/sizes.h>
-
-#include "i915_buddy.h"
-
-#include "i915_gem.h"
-#include "i915_utils.h"
-
-static struct kmem_cache *slab_blocks;
-
-static struct i915_buddy_block *i915_block_alloc(struct i915_buddy_mm *mm,
- struct i915_buddy_block *parent,
- unsigned int order,
- u64 offset)
-{
- struct i915_buddy_block *block;
-
- GEM_BUG_ON(order > I915_BUDDY_MAX_ORDER);
-
- block = kmem_cache_zalloc(slab_blocks, GFP_KERNEL);
- if (!block)
- return NULL;
-
- block->header = offset;
- block->header |= order;
- block->parent = parent;
-
- GEM_BUG_ON(block->header & I915_BUDDY_HEADER_UNUSED);
- return block;
-}
-
-static void i915_block_free(struct i915_buddy_mm *mm,
- struct i915_buddy_block *block)
-{
- kmem_cache_free(slab_blocks, block);
-}
-
-static void mark_allocated(struct i915_buddy_block *block)
-{
- block->header &= ~I915_BUDDY_HEADER_STATE;
- block->header |= I915_BUDDY_ALLOCATED;
-
- list_del(&block->link);
-}
-
-static void mark_free(struct i915_buddy_mm *mm,
- struct i915_buddy_block *block)
-{
- block->header &= ~I915_BUDDY_HEADER_STATE;
- block->header |= I915_BUDDY_FREE;
-
- list_add(&block->link,
- &mm->free_list[i915_buddy_block_order(block)]);
-}
-
-static void mark_split(struct i915_buddy_block *block)
-{
- block->header &= ~I915_BUDDY_HEADER_STATE;
- block->header |= I915_BUDDY_SPLIT;
-
- list_del(&block->link);
-}
-
-int i915_buddy_init(struct i915_buddy_mm *mm, u64 size, u64 chunk_size)
-{
- unsigned int i;
- u64 offset;
-
- if (size < chunk_size)
- return -EINVAL;
-
- if (chunk_size < PAGE_SIZE)
- return -EINVAL;
-
- if (!is_power_of_2(chunk_size))
- return -EINVAL;
-
- size = round_down(size, chunk_size);
-
- mm->size = size;
- mm->avail = size;
- mm->chunk_size = chunk_size;
- mm->max_order = ilog2(size) - ilog2(chunk_size);
-
- GEM_BUG_ON(mm->max_order > I915_BUDDY_MAX_ORDER);
-
- mm->free_list = kmalloc_array(mm->max_order + 1,
- sizeof(struct list_head),
- GFP_KERNEL);
- if (!mm->free_list)
- return -ENOMEM;
-
- for (i = 0; i <= mm->max_order; ++i)
- INIT_LIST_HEAD(&mm->free_list[i]);
-
- mm->n_roots = hweight64(size);
-
- mm->roots = kmalloc_array(mm->n_roots,
- sizeof(struct i915_buddy_block *),
- GFP_KERNEL);
- if (!mm->roots)
- goto out_free_list;
-
- offset = 0;
- i = 0;
-
- /*
- * Split into power-of-two blocks, in case we are given a size that is
- * not itself a power-of-two.
- */
- do {
- struct i915_buddy_block *root;
- unsigned int order;
- u64 root_size;
-
- root_size = rounddown_pow_of_two(size);
- order = ilog2(root_size) - ilog2(chunk_size);
-
- root = i915_block_alloc(mm, NULL, order, offset);
- if (!root)
- goto out_free_roots;
-
- mark_free(mm, root);
-
- GEM_BUG_ON(i > mm->max_order);
- GEM_BUG_ON(i915_buddy_block_size(mm, root) < chunk_size);
-
- mm->roots[i] = root;
-
- offset += root_size;
- size -= root_size;
- i++;
- } while (size);
-
- return 0;
-
-out_free_roots:
- while (i--)
- i915_block_free(mm, mm->roots[i]);
- kfree(mm->roots);
-out_free_list:
- kfree(mm->free_list);
- return -ENOMEM;
-}
-
-void i915_buddy_fini(struct i915_buddy_mm *mm)
-{
- int i;
-
- for (i = 0; i < mm->n_roots; ++i) {
- GEM_WARN_ON(!i915_buddy_block_is_free(mm->roots[i]));
- i915_block_free(mm, mm->roots[i]);
- }
-
- GEM_WARN_ON(mm->avail != mm->size);
-
- kfree(mm->roots);
- kfree(mm->free_list);
-}
-
-static int split_block(struct i915_buddy_mm *mm,
- struct i915_buddy_block *block)
-{
- unsigned int block_order = i915_buddy_block_order(block) - 1;
- u64 offset = i915_buddy_block_offset(block);
-
- GEM_BUG_ON(!i915_buddy_block_is_free(block));
- GEM_BUG_ON(!i915_buddy_block_order(block));
-
- block->left = i915_block_alloc(mm, block, block_order, offset);
- if (!block->left)
- return -ENOMEM;
-
- block->right = i915_block_alloc(mm, block, block_order,
- offset + (mm->chunk_size << block_order));
- if (!block->right) {
- i915_block_free(mm, block->left);
- return -ENOMEM;
- }
-
- mark_free(mm, block->left);
- mark_free(mm, block->right);
-
- mark_split(block);
-
- return 0;
-}
-
-static struct i915_buddy_block *
-get_buddy(struct i915_buddy_block *block)
-{
- struct i915_buddy_block *parent;
-
- parent = block->parent;
- if (!parent)
- return NULL;
-
- if (parent->left == block)
- return parent->right;
-
- return parent->left;
-}
-
-static void __i915_buddy_free(struct i915_buddy_mm *mm,
- struct i915_buddy_block *block)
-{
- struct i915_buddy_block *parent;
-
- while ((parent = block->parent)) {
- struct i915_buddy_block *buddy;
-
- buddy = get_buddy(block);
-
- if (!i915_buddy_block_is_free(buddy))
- break;
-
- list_del(&buddy->link);
-
- i915_block_free(mm, block);
- i915_block_free(mm, buddy);
-
- block = parent;
- }
-
- mark_free(mm, block);
-}
-
-void i915_buddy_free(struct i915_buddy_mm *mm,
- struct i915_buddy_block *block)
-{
- GEM_BUG_ON(!i915_buddy_block_is_allocated(block));
- mm->avail += i915_buddy_block_size(mm, block);
- __i915_buddy_free(mm, block);
-}
-
-void i915_buddy_free_list(struct i915_buddy_mm *mm, struct list_head *objects)
-{
- struct i915_buddy_block *block, *on;
-
- list_for_each_entry_safe(block, on, objects, link) {
- i915_buddy_free(mm, block);
- cond_resched();
- }
- INIT_LIST_HEAD(objects);
-}
-
-/*
- * Allocate power-of-two block. The order value here translates to:
- *
- * 0 = 2^0 * mm->chunk_size
- * 1 = 2^1 * mm->chunk_size
- * 2 = 2^2 * mm->chunk_size
- * ...
- */
-struct i915_buddy_block *
-i915_buddy_alloc(struct i915_buddy_mm *mm, unsigned int order)
-{
- struct i915_buddy_block *block = NULL;
- unsigned int i;
- int err;
-
- for (i = order; i <= mm->max_order; ++i) {
- block = list_first_entry_or_null(&mm->free_list[i],
- struct i915_buddy_block,
- link);
- if (block)
- break;
- }
-
- if (!block)
- return ERR_PTR(-ENOSPC);
-
- GEM_BUG_ON(!i915_buddy_block_is_free(block));
-
- while (i != order) {
- err = split_block(mm, block);
- if (unlikely(err))
- goto out_free;
-
- /* Go low */
- block = block->left;
- i--;
- }
-
- mark_allocated(block);
- mm->avail -= i915_buddy_block_size(mm, block);
- kmemleak_update_trace(block);
- return block;
-
-out_free:
- if (i != order)
- __i915_buddy_free(mm, block);
- return ERR_PTR(err);
-}
-
-static inline bool overlaps(u64 s1, u64 e1, u64 s2, u64 e2)
-{
- return s1 <= e2 && e1 >= s2;
-}
-
-static inline bool contains(u64 s1, u64 e1, u64 s2, u64 e2)
-{
- return s1 <= s2 && e1 >= e2;
-}
-
-/*
- * Allocate range. Note that it's safe to chain together multiple alloc_ranges
- * with the same blocks list.
- *
- * Intended for pre-allocating portions of the address space, for example to
- * reserve a block for the initial framebuffer or similar, hence the expectation
- * here is that i915_buddy_alloc() is still the main vehicle for
- * allocations, so if that's not the case then the drm_mm range allocator is
- * probably a much better fit, and so you should probably go use that instead.
- */
-int i915_buddy_alloc_range(struct i915_buddy_mm *mm,
- struct list_head *blocks,
- u64 start, u64 size)
-{
- struct i915_buddy_block *block;
- struct i915_buddy_block *buddy;
- LIST_HEAD(allocated);
- LIST_HEAD(dfs);
- u64 end;
- int err;
- int i;
-
- if (size < mm->chunk_size)
- return -EINVAL;
-
- if (!IS_ALIGNED(size | start, mm->chunk_size))
- return -EINVAL;
-
- if (range_overflows(start, size, mm->size))
- return -EINVAL;
-
- for (i = 0; i < mm->n_roots; ++i)
- list_add_tail(&mm->roots[i]->tmp_link, &dfs);
-
- end = start + size - 1;
-
- do {
- u64 block_start;
- u64 block_end;
-
- block = list_first_entry_or_null(&dfs,
- struct i915_buddy_block,
- tmp_link);
- if (!block)
- break;
-
- list_del(&block->tmp_link);
-
- block_start = i915_buddy_block_offset(block);
- block_end = block_start + i915_buddy_block_size(mm, block) - 1;
-
- if (!overlaps(start, end, block_start, block_end))
- continue;
-
- if (i915_buddy_block_is_allocated(block)) {
- err = -ENOSPC;
- goto err_free;
- }
-
- if (contains(start, end, block_start, block_end)) {
- if (!i915_buddy_block_is_free(block)) {
- err = -ENOSPC;
- goto err_free;
- }
-
- mark_allocated(block);
- mm->avail -= i915_buddy_block_size(mm, block);
- list_add_tail(&block->link, &allocated);
- continue;
- }
-
- if (!i915_buddy_block_is_split(block)) {
- err = split_block(mm, block);
- if (unlikely(err))
- goto err_undo;
- }
-
- list_add(&block->right->tmp_link, &dfs);
- list_add(&block->left->tmp_link, &dfs);
- } while (1);
-
- list_splice_tail(&allocated, blocks);
- return 0;
-
-err_undo:
- /*
- * We really don't want to leave around a bunch of split blocks, since
- * bigger is better, so make sure we merge everything back before we
- * free the allocated blocks.
- */
- buddy = get_buddy(block);
- if (buddy &&
- (i915_buddy_block_is_free(block) &&
- i915_buddy_block_is_free(buddy)))
- __i915_buddy_free(mm, block);
-
-err_free:
- i915_buddy_free_list(mm, &allocated);
- return err;
-}
-
-void i915_buddy_block_print(struct i915_buddy_mm *mm,
- struct i915_buddy_block *block,
- struct drm_printer *p)
-{
- u64 start = i915_buddy_block_offset(block);
- u64 size = i915_buddy_block_size(mm, block);
-
- drm_printf(p, "%#018llx-%#018llx: %llu\n", start, start + size, size);
-}
-
-void i915_buddy_print(struct i915_buddy_mm *mm, struct drm_printer *p)
-{
- int order;
-
- drm_printf(p, "chunk_size: %lluKiB, total: %lluMiB, free: %lluMiB\n",
- mm->chunk_size >> 10, mm->size >> 20, mm->avail >> 20);
-
- for (order = mm->max_order; order >= 0; order--) {
- struct i915_buddy_block *block;
- u64 count = 0, free;
-
- list_for_each_entry(block, &mm->free_list[order], link) {
- GEM_BUG_ON(!i915_buddy_block_is_free(block));
- count++;
- }
-
- drm_printf(p, "order-%d ", order);
-
- free = count * (mm->chunk_size << order);
- if (free < SZ_1M)
- drm_printf(p, "free: %lluKiB", free >> 10);
- else
- drm_printf(p, "free: %lluMiB", free >> 20);
-
- drm_printf(p, ", pages: %llu\n", count);
- }
-}
-
-#if IS_ENABLED(CONFIG_DRM_I915_SELFTEST)
-#include "selftests/i915_buddy.c"
-#endif
-
-void i915_buddy_module_exit(void)
-{
- kmem_cache_destroy(slab_blocks);
-}
-
-int __init i915_buddy_module_init(void)
-{
- slab_blocks = KMEM_CACHE(i915_buddy_block, 0);
- if (!slab_blocks)
- return -ENOMEM;
-
- return 0;
-}
diff --git a/drivers/gpu/drm/i915/i915_buddy.h b/drivers/gpu/drm/i915/i915_buddy.h
deleted file mode 100644
index 7077742112ac..000000000000
--- a/drivers/gpu/drm/i915/i915_buddy.h
+++ /dev/null
@@ -1,143 +0,0 @@
-/* SPDX-License-Identifier: MIT */
-/*
- * Copyright © 2021 Intel Corporation
- */
-
-#ifndef __I915_BUDDY_H__
-#define __I915_BUDDY_H__
-
-#include <linux/bitops.h>
-#include <linux/list.h>
-#include <linux/slab.h>
-
-#include <drm/drm_print.h>
-
-struct i915_buddy_block {
-#define I915_BUDDY_HEADER_OFFSET GENMASK_ULL(63, 12)
-#define I915_BUDDY_HEADER_STATE GENMASK_ULL(11, 10)
-#define I915_BUDDY_ALLOCATED (1 << 10)
-#define I915_BUDDY_FREE (2 << 10)
-#define I915_BUDDY_SPLIT (3 << 10)
-/* Free to be used, if needed in the future */
-#define I915_BUDDY_HEADER_UNUSED GENMASK_ULL(9, 6)
-#define I915_BUDDY_HEADER_ORDER GENMASK_ULL(5, 0)
- u64 header;
-
- struct i915_buddy_block *left;
- struct i915_buddy_block *right;
- struct i915_buddy_block *parent;
-
- void *private; /* owned by creator */
-
- /*
- * While the block is allocated by the user through i915_buddy_alloc*,
- * the user has ownership of the link, for example to maintain within
- * a list, if so desired. As soon as the block is freed with
- * i915_buddy_free* ownership is given back to the mm.
- */
- struct list_head link;
- struct list_head tmp_link;
-};
-
-/* Order-zero must be at least PAGE_SIZE */
-#define I915_BUDDY_MAX_ORDER (63 - PAGE_SHIFT)
-
-/*
- * Binary Buddy System.
- *
- * Locking should be handled by the user, a simple mutex around
- * i915_buddy_alloc* and i915_buddy_free* should suffice.
- */
-struct i915_buddy_mm {
- /* Maintain a free list for each order. */
- struct list_head *free_list;
-
- /*
- * Maintain explicit binary tree(s) to track the allocation of the
- * address space. This gives us a simple way of finding a buddy block
- * and performing the potentially recursive merge step when freeing a
- * block. Nodes are either allocated or free, in which case they will
- * also exist on the respective free list.
- */
- struct i915_buddy_block **roots;
-
- /*
- * Anything from here is public, and remains static for the lifetime of
- * the mm. Everything above is considered do-not-touch.
- */
- unsigned int n_roots;
- unsigned int max_order;
-
- /* Must be at least PAGE_SIZE */
- u64 chunk_size;
- u64 size;
- u64 avail;
-};
-
-static inline u64
-i915_buddy_block_offset(struct i915_buddy_block *block)
-{
- return block->header & I915_BUDDY_HEADER_OFFSET;
-}
-
-static inline unsigned int
-i915_buddy_block_order(struct i915_buddy_block *block)
-{
- return block->header & I915_BUDDY_HEADER_ORDER;
-}
-
-static inline unsigned int
-i915_buddy_block_state(struct i915_buddy_block *block)
-{
- return block->header & I915_BUDDY_HEADER_STATE;
-}
-
-static inline bool
-i915_buddy_block_is_allocated(struct i915_buddy_block *block)
-{
- return i915_buddy_block_state(block) == I915_BUDDY_ALLOCATED;
-}
-
-static inline bool
-i915_buddy_block_is_free(struct i915_buddy_block *block)
-{
- return i915_buddy_block_state(block) == I915_BUDDY_FREE;
-}
-
-static inline bool
-i915_buddy_block_is_split(struct i915_buddy_block *block)
-{
- return i915_buddy_block_state(block) == I915_BUDDY_SPLIT;
-}
-
-static inline u64
-i915_buddy_block_size(struct i915_buddy_mm *mm,
- struct i915_buddy_block *block)
-{
- return mm->chunk_size << i915_buddy_block_order(block);
-}
-
-int i915_buddy_init(struct i915_buddy_mm *mm, u64 size, u64 chunk_size);
-
-void i915_buddy_fini(struct i915_buddy_mm *mm);
-
-struct i915_buddy_block *
-i915_buddy_alloc(struct i915_buddy_mm *mm, unsigned int order);
-
-int i915_buddy_alloc_range(struct i915_buddy_mm *mm,
- struct list_head *blocks,
- u64 start, u64 size);
-
-void i915_buddy_free(struct i915_buddy_mm *mm, struct i915_buddy_block *block);
-
-void i915_buddy_free_list(struct i915_buddy_mm *mm, struct list_head *objects);
-
-void i915_buddy_print(struct i915_buddy_mm *mm, struct drm_printer *p);
-void i915_buddy_block_print(struct i915_buddy_mm *mm,
- struct i915_buddy_block *block,
- struct drm_printer *p);
-
-void i915_buddy_module_exit(void);
-int i915_buddy_module_init(void);
-
-#endif
diff --git a/drivers/gpu/drm/i915/i915_module.c b/drivers/gpu/drm/i915/i915_module.c
index ab2295dd4500..121b4178c5ca 100644
--- a/drivers/gpu/drm/i915/i915_module.c
+++ b/drivers/gpu/drm/i915/i915_module.c
@@ -9,7 +9,6 @@
#include "gem/i915_gem_context.h"
#include "gem/i915_gem_object.h"
#include "i915_active.h"
-#include "i915_buddy.h"
#include "i915_params.h"
#include "i915_pci.h"
#include "i915_perf.h"
@@ -50,8 +49,6 @@ static const struct {
{ .init = i915_check_nomodeset },
{ .init = i915_active_module_init,
.exit = i915_active_module_exit },
- { .init = i915_buddy_module_init,
- .exit = i915_buddy_module_exit },
{ .init = i915_context_module_init,
.exit = i915_context_module_exit },
{ .init = i915_gem_context_module_init,
diff --git a/drivers/gpu/drm/i915/i915_scatterlist.c b/drivers/gpu/drm/i915/i915_scatterlist.c
index 4a6712dca838..84d622aa32d2 100644
--- a/drivers/gpu/drm/i915/i915_scatterlist.c
+++ b/drivers/gpu/drm/i915/i915_scatterlist.c
@@ -5,10 +5,9 @@
*/
#include "i915_scatterlist.h"
-
-#include "i915_buddy.h"
#include "i915_ttm_buddy_manager.h"
+#include <drm/drm_buddy.h>
#include <drm/drm_mm.h>
#include <linux/slab.h>
@@ -126,9 +125,9 @@ struct sg_table *i915_sg_from_buddy_resource(struct ttm_resource *res,
struct i915_ttm_buddy_resource *bman_res = to_ttm_buddy_resource(res);
const u64 size = res->num_pages << PAGE_SHIFT;
const u64 max_segment = rounddown(UINT_MAX, PAGE_SIZE);
- struct i915_buddy_mm *mm = bman_res->mm;
+ struct drm_buddy_mm *mm = bman_res->mm;
struct list_head *blocks = &bman_res->blocks;
- struct i915_buddy_block *block;
+ struct drm_buddy_block *block;
struct scatterlist *sg;
struct sg_table *st;
resource_size_t prev_end;
@@ -151,8 +150,8 @@ struct sg_table *i915_sg_from_buddy_resource(struct ttm_resource *res,
list_for_each_entry(block, blocks, link) {
u64 block_size, offset;
- block_size = min_t(u64, size, i915_buddy_block_size(mm, block));
- offset = i915_buddy_block_offset(block);
+ block_size = min_t(u64, size, drm_buddy_block_size(mm, block));
+ offset = drm_buddy_block_offset(block);
while (block_size) {
u64 len;
diff --git a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
index d59fbb019032..c4b70cb8c248 100644
--- a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
+++ b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
@@ -8,14 +8,15 @@
#include <drm/ttm/ttm_bo_driver.h>
#include <drm/ttm/ttm_placement.h>
+#include <drm/drm_buddy.h>
+
#include "i915_ttm_buddy_manager.h"
-#include "i915_buddy.h"
#include "i915_gem.h"
struct i915_ttm_buddy_manager {
struct ttm_resource_manager manager;
- struct i915_buddy_mm mm;
+ struct drm_buddy_mm mm;
struct list_head reserved;
struct mutex lock;
u64 default_page_size;
@@ -34,7 +35,7 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
{
struct i915_ttm_buddy_manager *bman = to_buddy_manager(man);
struct i915_ttm_buddy_resource *bman_res;
- struct i915_buddy_mm *mm = &bman->mm;
+ struct drm_buddy_mm *mm = &bman->mm;
unsigned long n_pages;
unsigned int min_order;
u64 min_page_size;
@@ -73,7 +74,7 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
n_pages = size >> ilog2(mm->chunk_size);
do {
- struct i915_buddy_block *block;
+ struct drm_buddy_block *block;
unsigned int order;
order = fls(n_pages) - 1;
@@ -82,7 +83,7 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
do {
mutex_lock(&bman->lock);
- block = i915_buddy_alloc(mm, order);
+ block = drm_buddy_alloc(mm, order);
mutex_unlock(&bman->lock);
if (!IS_ERR(block))
break;
@@ -106,7 +107,7 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
err_free_blocks:
mutex_lock(&bman->lock);
- i915_buddy_free_list(mm, &bman_res->blocks);
+ drm_buddy_free_list(mm, &bman_res->blocks);
mutex_unlock(&bman->lock);
err_free_res:
kfree(bman_res);
@@ -120,7 +121,7 @@ static void i915_ttm_buddy_man_free(struct ttm_resource_manager *man,
struct i915_ttm_buddy_manager *bman = to_buddy_manager(man);
mutex_lock(&bman->lock);
- i915_buddy_free_list(&bman->mm, &bman_res->blocks);
+ drm_buddy_free_list(&bman->mm, &bman_res->blocks);
mutex_unlock(&bman->lock);
kfree(bman_res);
@@ -130,17 +131,17 @@ static void i915_ttm_buddy_man_debug(struct ttm_resource_manager *man,
struct drm_printer *printer)
{
struct i915_ttm_buddy_manager *bman = to_buddy_manager(man);
- struct i915_buddy_block *block;
+ struct drm_buddy_block *block;
mutex_lock(&bman->lock);
drm_printf(printer, "default_page_size: %lluKiB\n",
bman->default_page_size >> 10);
- i915_buddy_print(&bman->mm, printer);
+ drm_buddy_print(&bman->mm, printer);
drm_printf(printer, "reserved:\n");
list_for_each_entry(block, &bman->reserved, link)
- i915_buddy_block_print(&bman->mm, block, printer);
+ drm_buddy_block_print(&bman->mm, block, printer);
mutex_unlock(&bman->lock);
}
@@ -190,7 +191,7 @@ int i915_ttm_buddy_man_init(struct ttm_device *bdev,
if (!bman)
return -ENOMEM;
- err = i915_buddy_init(&bman->mm, size, chunk_size);
+ err = drm_buddy_init(&bman->mm, size, chunk_size);
if (err)
goto err_free_bman;
@@ -228,7 +229,7 @@ int i915_ttm_buddy_man_fini(struct ttm_device *bdev, unsigned int type)
{
struct ttm_resource_manager *man = ttm_manager_type(bdev, type);
struct i915_ttm_buddy_manager *bman = to_buddy_manager(man);
- struct i915_buddy_mm *mm = &bman->mm;
+ struct drm_buddy_mm *mm = &bman->mm;
int ret;
ttm_resource_manager_set_used(man, false);
@@ -240,8 +241,8 @@ int i915_ttm_buddy_man_fini(struct ttm_device *bdev, unsigned int type)
ttm_set_driver_manager(bdev, type, NULL);
mutex_lock(&bman->lock);
- i915_buddy_free_list(mm, &bman->reserved);
- i915_buddy_fini(mm);
+ drm_buddy_free_list(mm, &bman->reserved);
+ drm_buddy_fini(mm);
mutex_unlock(&bman->lock);
ttm_resource_manager_cleanup(man);
@@ -264,11 +265,11 @@ int i915_ttm_buddy_man_reserve(struct ttm_resource_manager *man,
u64 start, u64 size)
{
struct i915_ttm_buddy_manager *bman = to_buddy_manager(man);
- struct i915_buddy_mm *mm = &bman->mm;
+ struct drm_buddy_mm *mm = &bman->mm;
int ret;
mutex_lock(&bman->lock);
- ret = i915_buddy_alloc_range(mm, &bman->reserved, start, size);
+ ret = drm_buddy_alloc_range(mm, &bman->reserved, start, size);
mutex_unlock(&bman->lock);
return ret;
diff --git a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.h b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.h
index 0722d33f3e14..fa644b512c2e 100644
--- a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.h
+++ b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.h
@@ -13,7 +13,7 @@
struct ttm_device;
struct ttm_resource_manager;
-struct i915_buddy_mm;
+struct drm_buddy_mm;
/**
* struct i915_ttm_buddy_resource
@@ -28,7 +28,7 @@ struct i915_buddy_mm;
struct i915_ttm_buddy_resource {
struct ttm_resource base;
struct list_head blocks;
- struct i915_buddy_mm *mm;
+ struct drm_buddy_mm *mm;
};
/**
diff --git a/include/drm/drm_buddy.h b/include/drm/drm_buddy.h
new file mode 100644
index 000000000000..5ce3fc702f80
--- /dev/null
+++ b/include/drm/drm_buddy.h
@@ -0,0 +1,153 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2021 Intel Corporation
+ */
+
+#ifndef __DRM_BUDDY_H__
+#define __DRM_BUDDY_H__
+
+#include <linux/bitops.h>
+#include <linux/list.h>
+#include <linux/slab.h>
+#include <linux/sched.h>
+
+#include <drm/drm_print.h>
+
+#define range_overflows(start, size, max) ({ \
+ typeof(start) start__ = (start); \
+ typeof(size) size__ = (size); \
+ typeof(max) max__ = (max); \
+ (void)(&start__ == &size__); \
+ (void)(&start__ == &max__); \
+ start__ >= max__ || size__ > max__ - start__; \
+})
+
+struct drm_buddy_block {
+#define DRM_BUDDY_HEADER_OFFSET GENMASK_ULL(63, 12)
+#define DRM_BUDDY_HEADER_STATE GENMASK_ULL(11, 10)
+#define DRM_BUDDY_ALLOCATED (1 << 10)
+#define DRM_BUDDY_FREE (2 << 10)
+#define DRM_BUDDY_SPLIT (3 << 10)
+/* Free to be used, if needed in the future */
+#define DRM_BUDDY_HEADER_UNUSED GENMASK_ULL(9, 6)
+#define DRM_BUDDY_HEADER_ORDER GENMASK_ULL(5, 0)
+ u64 header;
+
+ struct drm_buddy_block *left;
+ struct drm_buddy_block *right;
+ struct drm_buddy_block *parent;
+
+ void *private; /* owned by creator */
+
+ /*
+ * While the block is allocated by the user through drm_buddy_alloc*,
+ * the user has ownership of the link, for example to maintain within
+ * a list, if so desired. As soon as the block is freed with
+ * drm_buddy_free* ownership is given back to the mm.
+ */
+ struct list_head link;
+ struct list_head tmp_link;
+};
+
+/* Order-zero must be at least PAGE_SIZE */
+#define DRM_BUDDY_MAX_ORDER (63 - PAGE_SHIFT)
+
+/*
+ * Binary Buddy System.
+ *
+ * Locking should be handled by the user, a simple mutex around
+ * drm_buddy_alloc* and drm_buddy_free* should suffice.
+ */
+struct drm_buddy_mm {
+ /* Maintain a free list for each order. */
+ struct list_head *free_list;
+
+ /*
+ * Maintain explicit binary tree(s) to track the allocation of the
+ * address space. This gives us a simple way of finding a buddy block
+ * and performing the potentially recursive merge step when freeing a
+ * block. Nodes are either allocated or free, in which case they will
+ * also exist on the respective free list.
+ */
+ struct drm_buddy_block **roots;
+
+ /*
+ * Anything from here is public, and remains static for the lifetime of
+ * the mm. Everything above is considered do-not-touch.
+ */
+ unsigned int n_roots;
+ unsigned int max_order;
+
+ /* Must be at least PAGE_SIZE */
+ u64 chunk_size;
+ u64 size;
+ u64 avail;
+};
+
+static inline u64
+drm_buddy_block_offset(struct drm_buddy_block *block)
+{
+ return block->header & DRM_BUDDY_HEADER_OFFSET;
+}
+
+static inline unsigned int
+drm_buddy_block_order(struct drm_buddy_block *block)
+{
+ return block->header & DRM_BUDDY_HEADER_ORDER;
+}
+
+static inline unsigned int
+drm_buddy_block_state(struct drm_buddy_block *block)
+{
+ return block->header & DRM_BUDDY_HEADER_STATE;
+}
+
+static inline bool
+drm_buddy_block_is_allocated(struct drm_buddy_block *block)
+{
+ return drm_buddy_block_state(block) == DRM_BUDDY_ALLOCATED;
+}
+
+static inline bool
+drm_buddy_block_is_free(struct drm_buddy_block *block)
+{
+ return drm_buddy_block_state(block) == DRM_BUDDY_FREE;
+}
+
+static inline bool
+drm_buddy_block_is_split(struct drm_buddy_block *block)
+{
+ return drm_buddy_block_state(block) == DRM_BUDDY_SPLIT;
+}
+
+static inline u64
+drm_buddy_block_size(struct drm_buddy_mm *mm,
+ struct drm_buddy_block *block)
+{
+ return mm->chunk_size << drm_buddy_block_order(block);
+}
+
+int drm_buddy_init(struct drm_buddy_mm *mm, u64 size, u64 chunk_size);
+
+void drm_buddy_fini(struct drm_buddy_mm *mm);
+
+struct drm_buddy_block *
+drm_buddy_alloc(struct drm_buddy_mm *mm, unsigned int order);
+
+int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
+ struct list_head *blocks,
+ u64 start, u64 size);
+
+void drm_buddy_free(struct drm_buddy_mm *mm, struct drm_buddy_block *block);
+
+void drm_buddy_free_list(struct drm_buddy_mm *mm, struct list_head *objects);
+
+void drm_buddy_print(struct drm_buddy_mm *mm, struct drm_printer *p);
+void drm_buddy_block_print(struct drm_buddy_mm *mm,
+ struct drm_buddy_block *block,
+ struct drm_printer *p);
+
+void drm_buddy_module_exit(void);
+int drm_buddy_module_init(void);
+
+#endif
--
2.25.1
^ permalink raw reply related [flat|nested] 58+ messages in thread
* [PATCH v2 2/8] drm: improve drm_buddy_alloc function
2021-10-25 13:00 ` [Intel-gfx] " Arunpravin
@ 2021-10-25 13:00 ` Arunpravin
-1 siblings, 0 replies; 58+ messages in thread
From: Arunpravin @ 2021-10-25 13:00 UTC (permalink / raw)
To: dri-devel, intel-gfx, amd-gfx
Cc: daniel, christian.koenig, alexander.deucher, matthew.auld,
tzimmermann, jani.nikula, Arunpravin
- Make drm_buddy_alloc a single function to handle
range allocation and non-range allocation demands
- Implemented a new function alloc_range() which allocates
the requested power-of-two block comply with range limitations
- Moved order computation and memory alignment logic from
i915 driver to drm buddy
V2:
merged below changes to keep the build unbroken
- drm_buddy_alloc_range() becomes obsolete and may be removed
- enable ttm range allocation (fpfn / lpfn) support in i915 driver
- apply enhanced drm_buddy_alloc() function to i915 driver
Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
---
drivers/gpu/drm/drm_buddy.c | 265 +++++++++++-------
drivers/gpu/drm/i915/i915_ttm_buddy_manager.c | 67 ++---
drivers/gpu/drm/i915/i915_ttm_buddy_manager.h | 2 +
include/drm/drm_buddy.h | 22 +-
4 files changed, 207 insertions(+), 149 deletions(-)
diff --git a/drivers/gpu/drm/drm_buddy.c b/drivers/gpu/drm/drm_buddy.c
index 39eb1d224bec..406e3d521903 100644
--- a/drivers/gpu/drm/drm_buddy.c
+++ b/drivers/gpu/drm/drm_buddy.c
@@ -274,63 +274,6 @@ void drm_buddy_free_list(struct drm_buddy_mm *mm, struct list_head *objects)
}
EXPORT_SYMBOL(drm_buddy_free_list);
-/**
- * drm_buddy_alloc - allocate power-of-two blocks
- *
- * @mm: DRM buddy manager to allocate from
- * @order: size of the allocation
- *
- * The order value here translates to:
- *
- * 0 = 2^0 * mm->chunk_size
- * 1 = 2^1 * mm->chunk_size
- * 2 = 2^2 * mm->chunk_size
- *
- * Returns:
- * allocated ptr to the &drm_buddy_block on success
- */
-struct drm_buddy_block *
-drm_buddy_alloc(struct drm_buddy_mm *mm, unsigned int order)
-{
- struct drm_buddy_block *block = NULL;
- unsigned int i;
- int err;
-
- for (i = order; i <= mm->max_order; ++i) {
- block = list_first_entry_or_null(&mm->free_list[i],
- struct drm_buddy_block,
- link);
- if (block)
- break;
- }
-
- if (!block)
- return ERR_PTR(-ENOSPC);
-
- BUG_ON(!drm_buddy_block_is_free(block));
-
- while (i != order) {
- err = split_block(mm, block);
- if (unlikely(err))
- goto out_free;
-
- /* Go low */
- block = block->left;
- i--;
- }
-
- mark_allocated(block);
- mm->avail -= drm_buddy_block_size(mm, block);
- kmemleak_update_trace(block);
- return block;
-
-out_free:
- if (i != order)
- __drm_buddy_free(mm, block);
- return ERR_PTR(err);
-}
-EXPORT_SYMBOL(drm_buddy_alloc);
-
static inline bool overlaps(u64 s1, u64 e1, u64 s2, u64 e2)
{
return s1 <= e2 && e1 >= s2;
@@ -341,52 +284,22 @@ static inline bool contains(u64 s1, u64 e1, u64 s2, u64 e2)
return s1 <= s2 && e1 >= e2;
}
-/**
- * drm_buddy_alloc_range - allocate range
- *
- * @mm: DRM buddy manager to allocate from
- * @blocks: output list head to add allocated blocks
- * @start: start of the allowed range for this block
- * @size: size of the allocation
- *
- * Intended for pre-allocating portions of the address space, for example to
- * reserve a block for the initial framebuffer or similar, hence the expectation
- * here is that drm_buddy_alloc() is still the main vehicle for
- * allocations, so if that's not the case then the drm_mm range allocator is
- * probably a much better fit, and so you should probably go use that instead.
- *
- * Note that it's safe to chain together multiple alloc_ranges
- * with the same blocks list
- *
- * Returns:
- * 0 on success, error code on failure.
- */
-int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
- struct list_head *blocks,
- u64 start, u64 size)
+static struct drm_buddy_block *
+alloc_range(struct drm_buddy_mm *mm,
+ u64 start, u64 end,
+ unsigned int order)
{
struct drm_buddy_block *block;
struct drm_buddy_block *buddy;
- LIST_HEAD(allocated);
LIST_HEAD(dfs);
- u64 end;
int err;
int i;
- if (size < mm->chunk_size)
- return -EINVAL;
-
- if (!IS_ALIGNED(size | start, mm->chunk_size))
- return -EINVAL;
-
- if (range_overflows(start, size, mm->size))
- return -EINVAL;
+ end = end - 1;
for (i = 0; i < mm->n_roots; ++i)
list_add_tail(&mm->roots[i]->tmp_link, &dfs);
- end = start + size - 1;
-
do {
u64 block_start;
u64 block_end;
@@ -394,31 +307,32 @@ int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
block = list_first_entry_or_null(&dfs,
struct drm_buddy_block,
tmp_link);
+
if (!block)
break;
list_del(&block->tmp_link);
+ if (drm_buddy_block_order(block) < order)
+ continue;
+
block_start = drm_buddy_block_offset(block);
block_end = block_start + drm_buddy_block_size(mm, block) - 1;
if (!overlaps(start, end, block_start, block_end))
continue;
- if (drm_buddy_block_is_allocated(block)) {
- err = -ENOSPC;
- goto err_free;
- }
+ if (drm_buddy_block_is_allocated(block))
+ continue;
- if (contains(start, end, block_start, block_end)) {
- if (!drm_buddy_block_is_free(block)) {
- err = -ENOSPC;
- goto err_free;
- }
+ if (contains(start, end, block_start, block_end)
+ && order == drm_buddy_block_order(block)) {
+ /*
+ * Find the free block within the range.
+ */
+ if (drm_buddy_block_is_free(block))
+ return block;
- mark_allocated(block);
- mm->avail -= drm_buddy_block_size(mm, block);
- list_add_tail(&block->link, &allocated);
continue;
}
@@ -428,12 +342,11 @@ int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
goto err_undo;
}
- list_add(&block->right->tmp_link, &dfs);
list_add(&block->left->tmp_link, &dfs);
+ list_add(&block->right->tmp_link, &dfs);
} while (1);
- list_splice_tail(&allocated, blocks);
- return 0;
+ return ERR_PTR(-ENOSPC);
err_undo:
/*
@@ -446,12 +359,148 @@ int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
(drm_buddy_block_is_free(block) &&
drm_buddy_block_is_free(buddy)))
__drm_buddy_free(mm, block);
+ return ERR_PTR(err);
+}
+
+static struct drm_buddy_block *
+alloc_from_freelist(struct drm_buddy_mm *mm,
+ unsigned int order,
+ unsigned long flags)
+{
+ struct drm_buddy_block *block = NULL;
+ unsigned int i;
+ int err;
+
+ for (i = order; i <= mm->max_order; ++i) {
+ if (!list_empty(&mm->free_list[i])) {
+ block = list_first_entry_or_null(&mm->free_list[i],
+ struct drm_buddy_block,
+ link);
+
+ if (block)
+ break;
+ }
+ }
+
+ if (!block)
+ return ERR_PTR(-ENOSPC);
+
+ BUG_ON(!drm_buddy_block_is_free(block));
+
+ while (i != order) {
+ err = split_block(mm, block);
+
+ if (unlikely(err))
+ goto err_undo;
+
+ block = block->right;
+ i--;
+ }
+ return block;
+
+err_undo:
+ if (i != order)
+ __drm_buddy_free(mm, block);
+ return ERR_PTR(err);
+}
+
+/**
+ * drm_buddy_alloc - allocate power-of-two blocks
+ *
+ * @mm: DRM buddy manager to allocate from
+ * @start: start of the allowed range for this block
+ * @end: end of the allowed range for this block
+ * @size: size of the allocation
+ * @min_page_size: alignment of the allocation
+ * @blocks: output list head to add allocated blocks
+ * @flags: DRM_BUDDY_*_ALLOCATION flags
+ *
+ * alloc_range() called on range limitations, which traverses
+ * the tree and returns the desired block.
+ *
+ * alloc_from_freelist() called when *no* range restrictions
+ * are enforced, which picks the block from the freelist.
+ *
+ * blocks are allocated in order, the order value here translates to:
+ *
+ * 0 = 2^0 * mm->chunk_size
+ * 1 = 2^1 * mm->chunk_size
+ * 2 = 2^2 * mm->chunk_size
+ *
+ * Returns:
+ * 0 on success, error code on failure.
+ */
+int drm_buddy_alloc(struct drm_buddy_mm *mm,
+ u64 start, u64 end, u64 size,
+ u64 min_page_size,
+ struct list_head *blocks,
+ unsigned long flags)
+{
+ struct drm_buddy_block *block = NULL;
+ unsigned int min_order, order;
+ unsigned long pages;
+ LIST_HEAD(allocated);
+ int err;
+
+ if (size < mm->chunk_size)
+ return -EINVAL;
+
+ if (!IS_ALIGNED(start, mm->chunk_size))
+ return -EINVAL;
+
+ if (!IS_ALIGNED(end, mm->chunk_size))
+ return -EINVAL;
+
+ if (!IS_ALIGNED(size, mm->chunk_size))
+ return -EINVAL;
+
+ if (check_range_overflow(start, end, size, mm->size))
+ return -EINVAL;
+
+ pages = size >> ilog2(mm->chunk_size);
+ min_order = ilog2(min_page_size) - ilog2(mm->chunk_size);
+
+ do {
+ order = fls(pages) - 1;
+ BUG_ON(order > mm->max_order);
+ BUG_ON(order < min_order);
+
+ do {
+ if (flags & DRM_BUDDY_RANGE_ALLOCATION)
+ /* Allocate traversing within the range */
+ block = alloc_range(mm, start, end, order);
+ else
+ /* Allocate from freelist */
+ block = alloc_from_freelist(mm, order, flags);
+
+ if (!IS_ERR(block))
+ break;
+
+ if (order-- == min_order) {
+ err = -ENOSPC;
+ goto err_free;
+ }
+ } while (1);
+
+ mark_allocated(block);
+ mm->avail -= drm_buddy_block_size(mm, block);
+ kmemleak_update_trace(block);
+ list_add_tail(&block->link, &allocated);
+
+ pages -= BIT(order);
+
+ if (!pages)
+ break;
+ } while (1);
+
+ list_splice_tail(&allocated, blocks);
+ return 0;
err_free:
drm_buddy_free_list(mm, &allocated);
return err;
}
-EXPORT_SYMBOL(drm_buddy_alloc_range);
+EXPORT_SYMBOL(drm_buddy_alloc);
/**
* drm_buddy_block_print - print block information
diff --git a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
index c4b70cb8c248..75197ba6e40d 100644
--- a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
+++ b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
@@ -36,13 +36,14 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
struct i915_ttm_buddy_manager *bman = to_buddy_manager(man);
struct i915_ttm_buddy_resource *bman_res;
struct drm_buddy_mm *mm = &bman->mm;
- unsigned long n_pages;
- unsigned int min_order;
+ unsigned long n_pages, lpfn;
u64 min_page_size;
u64 size;
int err;
- GEM_BUG_ON(place->fpfn || place->lpfn);
+ lpfn = place->lpfn;
+ if (!lpfn)
+ lpfn = man->size;
bman_res = kzalloc(sizeof(*bman_res), GFP_KERNEL);
if (!bman_res)
@@ -52,6 +53,9 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
INIT_LIST_HEAD(&bman_res->blocks);
bman_res->mm = mm;
+ if (place->fpfn || lpfn != man->size)
+ bman_res->flags |= DRM_BUDDY_RANGE_ALLOCATION;
+
GEM_BUG_ON(!bman_res->base.num_pages);
size = bman_res->base.num_pages << PAGE_SHIFT;
@@ -60,10 +64,16 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
min_page_size = bo->page_alignment << PAGE_SHIFT;
GEM_BUG_ON(min_page_size < mm->chunk_size);
- min_order = ilog2(min_page_size) - ilog2(mm->chunk_size);
+
if (place->flags & TTM_PL_FLAG_CONTIGUOUS) {
+ unsigned long pages;
+
size = roundup_pow_of_two(size);
- min_order = ilog2(size) - ilog2(mm->chunk_size);
+ min_page_size = size;
+
+ pages = size >> ilog2(mm->chunk_size);
+ if (pages > lpfn)
+ lpfn = pages;
}
if (size > mm->size) {
@@ -73,34 +83,17 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
n_pages = size >> ilog2(mm->chunk_size);
- do {
- struct drm_buddy_block *block;
- unsigned int order;
-
- order = fls(n_pages) - 1;
- GEM_BUG_ON(order > mm->max_order);
- GEM_BUG_ON(order < min_order);
-
- do {
- mutex_lock(&bman->lock);
- block = drm_buddy_alloc(mm, order);
- mutex_unlock(&bman->lock);
- if (!IS_ERR(block))
- break;
-
- if (order-- == min_order) {
- err = -ENOSPC;
- goto err_free_blocks;
- }
- } while (1);
-
- n_pages -= BIT(order);
-
- list_add_tail(&block->link, &bman_res->blocks);
+ mutex_lock(&bman->lock);
+ err = drm_buddy_alloc(mm, (uint64_t)place->fpfn << PAGE_SHIFT,
+ (uint64_t)place->lpfn << PAGE_SHIFT,
+ (uint64_t)n_pages << PAGE_SHIFT,
+ min_page_size,
+ &bman_res->blocks,
+ bman_res->flags);
+ mutex_unlock(&bman->lock);
- if (!n_pages)
- break;
- } while (1);
+ if (unlikely(err))
+ goto err_free_blocks;
*res = &bman_res->base;
return 0;
@@ -266,10 +259,18 @@ int i915_ttm_buddy_man_reserve(struct ttm_resource_manager *man,
{
struct i915_ttm_buddy_manager *bman = to_buddy_manager(man);
struct drm_buddy_mm *mm = &bman->mm;
+ unsigned long flags = 0;
+ u64 min_size;
int ret;
+ min_size = size;
+ flags |= DRM_BUDDY_RANGE_ALLOCATION;
+
mutex_lock(&bman->lock);
- ret = drm_buddy_alloc_range(mm, &bman->reserved, start, size);
+ ret = drm_buddy_alloc(mm, start, start + size,
+ size, min_size,
+ &bman->reserved,
+ flags);
mutex_unlock(&bman->lock);
return ret;
diff --git a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.h b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.h
index fa644b512c2e..5ba490875f66 100644
--- a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.h
+++ b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.h
@@ -20,6 +20,7 @@ struct drm_buddy_mm;
*
* @base: struct ttm_resource base class we extend
* @blocks: the list of struct i915_buddy_block for this resource/allocation
+ * @flags: DRM_BUDDY_*_ALLOCATION flags
* @mm: the struct i915_buddy_mm for this resource
*
* Extends the struct ttm_resource to manage an address space allocation with
@@ -28,6 +29,7 @@ struct drm_buddy_mm;
struct i915_ttm_buddy_resource {
struct ttm_resource base;
struct list_head blocks;
+ unsigned long flags;
struct drm_buddy_mm *mm;
};
diff --git a/include/drm/drm_buddy.h b/include/drm/drm_buddy.h
index 5ce3fc702f80..c7bb5509a7ad 100644
--- a/include/drm/drm_buddy.h
+++ b/include/drm/drm_buddy.h
@@ -13,15 +13,22 @@
#include <drm/drm_print.h>
-#define range_overflows(start, size, max) ({ \
+#define check_range_overflow(start, end, size, max) ({ \
typeof(start) start__ = (start); \
+ typeof(end) end__ = (end);\
typeof(size) size__ = (size); \
typeof(max) max__ = (max); \
(void)(&start__ == &size__); \
(void)(&start__ == &max__); \
- start__ >= max__ || size__ > max__ - start__; \
+ (void)(&start__ == &end__); \
+ (void)(&end__ == &size__); \
+ (void)(&end__ == &max__); \
+ start__ >= max__ || end__ > max__ \
+ || size__ > end__ - start__; \
})
+#define DRM_BUDDY_RANGE_ALLOCATION (1 << 0)
+
struct drm_buddy_block {
#define DRM_BUDDY_HEADER_OFFSET GENMASK_ULL(63, 12)
#define DRM_BUDDY_HEADER_STATE GENMASK_ULL(11, 10)
@@ -131,12 +138,11 @@ int drm_buddy_init(struct drm_buddy_mm *mm, u64 size, u64 chunk_size);
void drm_buddy_fini(struct drm_buddy_mm *mm);
-struct drm_buddy_block *
-drm_buddy_alloc(struct drm_buddy_mm *mm, unsigned int order);
-
-int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
- struct list_head *blocks,
- u64 start, u64 size);
+int drm_buddy_alloc(struct drm_buddy_mm *mm,
+ u64 start, u64 end, u64 size,
+ u64 min_page_size,
+ struct list_head *blocks,
+ unsigned long flags);
void drm_buddy_free(struct drm_buddy_mm *mm, struct drm_buddy_block *block);
--
2.25.1
^ permalink raw reply related [flat|nested] 58+ messages in thread
* [Intel-gfx] [PATCH v2 2/8] drm: improve drm_buddy_alloc function
@ 2021-10-25 13:00 ` Arunpravin
0 siblings, 0 replies; 58+ messages in thread
From: Arunpravin @ 2021-10-25 13:00 UTC (permalink / raw)
To: dri-devel, intel-gfx, amd-gfx
Cc: daniel, christian.koenig, alexander.deucher, matthew.auld,
tzimmermann, jani.nikula, Arunpravin
- Make drm_buddy_alloc a single function to handle
range allocation and non-range allocation demands
- Implemented a new function alloc_range() which allocates
the requested power-of-two block comply with range limitations
- Moved order computation and memory alignment logic from
i915 driver to drm buddy
V2:
merged below changes to keep the build unbroken
- drm_buddy_alloc_range() becomes obsolete and may be removed
- enable ttm range allocation (fpfn / lpfn) support in i915 driver
- apply enhanced drm_buddy_alloc() function to i915 driver
Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
---
drivers/gpu/drm/drm_buddy.c | 265 +++++++++++-------
drivers/gpu/drm/i915/i915_ttm_buddy_manager.c | 67 ++---
drivers/gpu/drm/i915/i915_ttm_buddy_manager.h | 2 +
include/drm/drm_buddy.h | 22 +-
4 files changed, 207 insertions(+), 149 deletions(-)
diff --git a/drivers/gpu/drm/drm_buddy.c b/drivers/gpu/drm/drm_buddy.c
index 39eb1d224bec..406e3d521903 100644
--- a/drivers/gpu/drm/drm_buddy.c
+++ b/drivers/gpu/drm/drm_buddy.c
@@ -274,63 +274,6 @@ void drm_buddy_free_list(struct drm_buddy_mm *mm, struct list_head *objects)
}
EXPORT_SYMBOL(drm_buddy_free_list);
-/**
- * drm_buddy_alloc - allocate power-of-two blocks
- *
- * @mm: DRM buddy manager to allocate from
- * @order: size of the allocation
- *
- * The order value here translates to:
- *
- * 0 = 2^0 * mm->chunk_size
- * 1 = 2^1 * mm->chunk_size
- * 2 = 2^2 * mm->chunk_size
- *
- * Returns:
- * allocated ptr to the &drm_buddy_block on success
- */
-struct drm_buddy_block *
-drm_buddy_alloc(struct drm_buddy_mm *mm, unsigned int order)
-{
- struct drm_buddy_block *block = NULL;
- unsigned int i;
- int err;
-
- for (i = order; i <= mm->max_order; ++i) {
- block = list_first_entry_or_null(&mm->free_list[i],
- struct drm_buddy_block,
- link);
- if (block)
- break;
- }
-
- if (!block)
- return ERR_PTR(-ENOSPC);
-
- BUG_ON(!drm_buddy_block_is_free(block));
-
- while (i != order) {
- err = split_block(mm, block);
- if (unlikely(err))
- goto out_free;
-
- /* Go low */
- block = block->left;
- i--;
- }
-
- mark_allocated(block);
- mm->avail -= drm_buddy_block_size(mm, block);
- kmemleak_update_trace(block);
- return block;
-
-out_free:
- if (i != order)
- __drm_buddy_free(mm, block);
- return ERR_PTR(err);
-}
-EXPORT_SYMBOL(drm_buddy_alloc);
-
static inline bool overlaps(u64 s1, u64 e1, u64 s2, u64 e2)
{
return s1 <= e2 && e1 >= s2;
@@ -341,52 +284,22 @@ static inline bool contains(u64 s1, u64 e1, u64 s2, u64 e2)
return s1 <= s2 && e1 >= e2;
}
-/**
- * drm_buddy_alloc_range - allocate range
- *
- * @mm: DRM buddy manager to allocate from
- * @blocks: output list head to add allocated blocks
- * @start: start of the allowed range for this block
- * @size: size of the allocation
- *
- * Intended for pre-allocating portions of the address space, for example to
- * reserve a block for the initial framebuffer or similar, hence the expectation
- * here is that drm_buddy_alloc() is still the main vehicle for
- * allocations, so if that's not the case then the drm_mm range allocator is
- * probably a much better fit, and so you should probably go use that instead.
- *
- * Note that it's safe to chain together multiple alloc_ranges
- * with the same blocks list
- *
- * Returns:
- * 0 on success, error code on failure.
- */
-int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
- struct list_head *blocks,
- u64 start, u64 size)
+static struct drm_buddy_block *
+alloc_range(struct drm_buddy_mm *mm,
+ u64 start, u64 end,
+ unsigned int order)
{
struct drm_buddy_block *block;
struct drm_buddy_block *buddy;
- LIST_HEAD(allocated);
LIST_HEAD(dfs);
- u64 end;
int err;
int i;
- if (size < mm->chunk_size)
- return -EINVAL;
-
- if (!IS_ALIGNED(size | start, mm->chunk_size))
- return -EINVAL;
-
- if (range_overflows(start, size, mm->size))
- return -EINVAL;
+ end = end - 1;
for (i = 0; i < mm->n_roots; ++i)
list_add_tail(&mm->roots[i]->tmp_link, &dfs);
- end = start + size - 1;
-
do {
u64 block_start;
u64 block_end;
@@ -394,31 +307,32 @@ int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
block = list_first_entry_or_null(&dfs,
struct drm_buddy_block,
tmp_link);
+
if (!block)
break;
list_del(&block->tmp_link);
+ if (drm_buddy_block_order(block) < order)
+ continue;
+
block_start = drm_buddy_block_offset(block);
block_end = block_start + drm_buddy_block_size(mm, block) - 1;
if (!overlaps(start, end, block_start, block_end))
continue;
- if (drm_buddy_block_is_allocated(block)) {
- err = -ENOSPC;
- goto err_free;
- }
+ if (drm_buddy_block_is_allocated(block))
+ continue;
- if (contains(start, end, block_start, block_end)) {
- if (!drm_buddy_block_is_free(block)) {
- err = -ENOSPC;
- goto err_free;
- }
+ if (contains(start, end, block_start, block_end)
+ && order == drm_buddy_block_order(block)) {
+ /*
+ * Find the free block within the range.
+ */
+ if (drm_buddy_block_is_free(block))
+ return block;
- mark_allocated(block);
- mm->avail -= drm_buddy_block_size(mm, block);
- list_add_tail(&block->link, &allocated);
continue;
}
@@ -428,12 +342,11 @@ int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
goto err_undo;
}
- list_add(&block->right->tmp_link, &dfs);
list_add(&block->left->tmp_link, &dfs);
+ list_add(&block->right->tmp_link, &dfs);
} while (1);
- list_splice_tail(&allocated, blocks);
- return 0;
+ return ERR_PTR(-ENOSPC);
err_undo:
/*
@@ -446,12 +359,148 @@ int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
(drm_buddy_block_is_free(block) &&
drm_buddy_block_is_free(buddy)))
__drm_buddy_free(mm, block);
+ return ERR_PTR(err);
+}
+
+static struct drm_buddy_block *
+alloc_from_freelist(struct drm_buddy_mm *mm,
+ unsigned int order,
+ unsigned long flags)
+{
+ struct drm_buddy_block *block = NULL;
+ unsigned int i;
+ int err;
+
+ for (i = order; i <= mm->max_order; ++i) {
+ if (!list_empty(&mm->free_list[i])) {
+ block = list_first_entry_or_null(&mm->free_list[i],
+ struct drm_buddy_block,
+ link);
+
+ if (block)
+ break;
+ }
+ }
+
+ if (!block)
+ return ERR_PTR(-ENOSPC);
+
+ BUG_ON(!drm_buddy_block_is_free(block));
+
+ while (i != order) {
+ err = split_block(mm, block);
+
+ if (unlikely(err))
+ goto err_undo;
+
+ block = block->right;
+ i--;
+ }
+ return block;
+
+err_undo:
+ if (i != order)
+ __drm_buddy_free(mm, block);
+ return ERR_PTR(err);
+}
+
+/**
+ * drm_buddy_alloc - allocate power-of-two blocks
+ *
+ * @mm: DRM buddy manager to allocate from
+ * @start: start of the allowed range for this block
+ * @end: end of the allowed range for this block
+ * @size: size of the allocation
+ * @min_page_size: alignment of the allocation
+ * @blocks: output list head to add allocated blocks
+ * @flags: DRM_BUDDY_*_ALLOCATION flags
+ *
+ * alloc_range() called on range limitations, which traverses
+ * the tree and returns the desired block.
+ *
+ * alloc_from_freelist() called when *no* range restrictions
+ * are enforced, which picks the block from the freelist.
+ *
+ * blocks are allocated in order, the order value here translates to:
+ *
+ * 0 = 2^0 * mm->chunk_size
+ * 1 = 2^1 * mm->chunk_size
+ * 2 = 2^2 * mm->chunk_size
+ *
+ * Returns:
+ * 0 on success, error code on failure.
+ */
+int drm_buddy_alloc(struct drm_buddy_mm *mm,
+ u64 start, u64 end, u64 size,
+ u64 min_page_size,
+ struct list_head *blocks,
+ unsigned long flags)
+{
+ struct drm_buddy_block *block = NULL;
+ unsigned int min_order, order;
+ unsigned long pages;
+ LIST_HEAD(allocated);
+ int err;
+
+ if (size < mm->chunk_size)
+ return -EINVAL;
+
+ if (!IS_ALIGNED(start, mm->chunk_size))
+ return -EINVAL;
+
+ if (!IS_ALIGNED(end, mm->chunk_size))
+ return -EINVAL;
+
+ if (!IS_ALIGNED(size, mm->chunk_size))
+ return -EINVAL;
+
+ if (check_range_overflow(start, end, size, mm->size))
+ return -EINVAL;
+
+ pages = size >> ilog2(mm->chunk_size);
+ min_order = ilog2(min_page_size) - ilog2(mm->chunk_size);
+
+ do {
+ order = fls(pages) - 1;
+ BUG_ON(order > mm->max_order);
+ BUG_ON(order < min_order);
+
+ do {
+ if (flags & DRM_BUDDY_RANGE_ALLOCATION)
+ /* Allocate traversing within the range */
+ block = alloc_range(mm, start, end, order);
+ else
+ /* Allocate from freelist */
+ block = alloc_from_freelist(mm, order, flags);
+
+ if (!IS_ERR(block))
+ break;
+
+ if (order-- == min_order) {
+ err = -ENOSPC;
+ goto err_free;
+ }
+ } while (1);
+
+ mark_allocated(block);
+ mm->avail -= drm_buddy_block_size(mm, block);
+ kmemleak_update_trace(block);
+ list_add_tail(&block->link, &allocated);
+
+ pages -= BIT(order);
+
+ if (!pages)
+ break;
+ } while (1);
+
+ list_splice_tail(&allocated, blocks);
+ return 0;
err_free:
drm_buddy_free_list(mm, &allocated);
return err;
}
-EXPORT_SYMBOL(drm_buddy_alloc_range);
+EXPORT_SYMBOL(drm_buddy_alloc);
/**
* drm_buddy_block_print - print block information
diff --git a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
index c4b70cb8c248..75197ba6e40d 100644
--- a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
+++ b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
@@ -36,13 +36,14 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
struct i915_ttm_buddy_manager *bman = to_buddy_manager(man);
struct i915_ttm_buddy_resource *bman_res;
struct drm_buddy_mm *mm = &bman->mm;
- unsigned long n_pages;
- unsigned int min_order;
+ unsigned long n_pages, lpfn;
u64 min_page_size;
u64 size;
int err;
- GEM_BUG_ON(place->fpfn || place->lpfn);
+ lpfn = place->lpfn;
+ if (!lpfn)
+ lpfn = man->size;
bman_res = kzalloc(sizeof(*bman_res), GFP_KERNEL);
if (!bman_res)
@@ -52,6 +53,9 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
INIT_LIST_HEAD(&bman_res->blocks);
bman_res->mm = mm;
+ if (place->fpfn || lpfn != man->size)
+ bman_res->flags |= DRM_BUDDY_RANGE_ALLOCATION;
+
GEM_BUG_ON(!bman_res->base.num_pages);
size = bman_res->base.num_pages << PAGE_SHIFT;
@@ -60,10 +64,16 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
min_page_size = bo->page_alignment << PAGE_SHIFT;
GEM_BUG_ON(min_page_size < mm->chunk_size);
- min_order = ilog2(min_page_size) - ilog2(mm->chunk_size);
+
if (place->flags & TTM_PL_FLAG_CONTIGUOUS) {
+ unsigned long pages;
+
size = roundup_pow_of_two(size);
- min_order = ilog2(size) - ilog2(mm->chunk_size);
+ min_page_size = size;
+
+ pages = size >> ilog2(mm->chunk_size);
+ if (pages > lpfn)
+ lpfn = pages;
}
if (size > mm->size) {
@@ -73,34 +83,17 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
n_pages = size >> ilog2(mm->chunk_size);
- do {
- struct drm_buddy_block *block;
- unsigned int order;
-
- order = fls(n_pages) - 1;
- GEM_BUG_ON(order > mm->max_order);
- GEM_BUG_ON(order < min_order);
-
- do {
- mutex_lock(&bman->lock);
- block = drm_buddy_alloc(mm, order);
- mutex_unlock(&bman->lock);
- if (!IS_ERR(block))
- break;
-
- if (order-- == min_order) {
- err = -ENOSPC;
- goto err_free_blocks;
- }
- } while (1);
-
- n_pages -= BIT(order);
-
- list_add_tail(&block->link, &bman_res->blocks);
+ mutex_lock(&bman->lock);
+ err = drm_buddy_alloc(mm, (uint64_t)place->fpfn << PAGE_SHIFT,
+ (uint64_t)place->lpfn << PAGE_SHIFT,
+ (uint64_t)n_pages << PAGE_SHIFT,
+ min_page_size,
+ &bman_res->blocks,
+ bman_res->flags);
+ mutex_unlock(&bman->lock);
- if (!n_pages)
- break;
- } while (1);
+ if (unlikely(err))
+ goto err_free_blocks;
*res = &bman_res->base;
return 0;
@@ -266,10 +259,18 @@ int i915_ttm_buddy_man_reserve(struct ttm_resource_manager *man,
{
struct i915_ttm_buddy_manager *bman = to_buddy_manager(man);
struct drm_buddy_mm *mm = &bman->mm;
+ unsigned long flags = 0;
+ u64 min_size;
int ret;
+ min_size = size;
+ flags |= DRM_BUDDY_RANGE_ALLOCATION;
+
mutex_lock(&bman->lock);
- ret = drm_buddy_alloc_range(mm, &bman->reserved, start, size);
+ ret = drm_buddy_alloc(mm, start, start + size,
+ size, min_size,
+ &bman->reserved,
+ flags);
mutex_unlock(&bman->lock);
return ret;
diff --git a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.h b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.h
index fa644b512c2e..5ba490875f66 100644
--- a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.h
+++ b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.h
@@ -20,6 +20,7 @@ struct drm_buddy_mm;
*
* @base: struct ttm_resource base class we extend
* @blocks: the list of struct i915_buddy_block for this resource/allocation
+ * @flags: DRM_BUDDY_*_ALLOCATION flags
* @mm: the struct i915_buddy_mm for this resource
*
* Extends the struct ttm_resource to manage an address space allocation with
@@ -28,6 +29,7 @@ struct drm_buddy_mm;
struct i915_ttm_buddy_resource {
struct ttm_resource base;
struct list_head blocks;
+ unsigned long flags;
struct drm_buddy_mm *mm;
};
diff --git a/include/drm/drm_buddy.h b/include/drm/drm_buddy.h
index 5ce3fc702f80..c7bb5509a7ad 100644
--- a/include/drm/drm_buddy.h
+++ b/include/drm/drm_buddy.h
@@ -13,15 +13,22 @@
#include <drm/drm_print.h>
-#define range_overflows(start, size, max) ({ \
+#define check_range_overflow(start, end, size, max) ({ \
typeof(start) start__ = (start); \
+ typeof(end) end__ = (end);\
typeof(size) size__ = (size); \
typeof(max) max__ = (max); \
(void)(&start__ == &size__); \
(void)(&start__ == &max__); \
- start__ >= max__ || size__ > max__ - start__; \
+ (void)(&start__ == &end__); \
+ (void)(&end__ == &size__); \
+ (void)(&end__ == &max__); \
+ start__ >= max__ || end__ > max__ \
+ || size__ > end__ - start__; \
})
+#define DRM_BUDDY_RANGE_ALLOCATION (1 << 0)
+
struct drm_buddy_block {
#define DRM_BUDDY_HEADER_OFFSET GENMASK_ULL(63, 12)
#define DRM_BUDDY_HEADER_STATE GENMASK_ULL(11, 10)
@@ -131,12 +138,11 @@ int drm_buddy_init(struct drm_buddy_mm *mm, u64 size, u64 chunk_size);
void drm_buddy_fini(struct drm_buddy_mm *mm);
-struct drm_buddy_block *
-drm_buddy_alloc(struct drm_buddy_mm *mm, unsigned int order);
-
-int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
- struct list_head *blocks,
- u64 start, u64 size);
+int drm_buddy_alloc(struct drm_buddy_mm *mm,
+ u64 start, u64 end, u64 size,
+ u64 min_page_size,
+ struct list_head *blocks,
+ unsigned long flags);
void drm_buddy_free(struct drm_buddy_mm *mm, struct drm_buddy_block *block);
--
2.25.1
^ permalink raw reply related [flat|nested] 58+ messages in thread
* [Intel-gfx] ✗ Fi.CI.BUILD: failure for series starting with [v2,1/8] drm: move the buddy allocator from i915 into common drm
2021-10-25 13:00 ` [Intel-gfx] " Arunpravin
` (7 preceding siblings ...)
(?)
@ 2021-10-25 13:56 ` Patchwork
-1 siblings, 0 replies; 58+ messages in thread
From: Patchwork @ 2021-10-25 13:56 UTC (permalink / raw)
To: Arunpravin; +Cc: intel-gfx
== Series Details ==
Series: series starting with [v2,1/8] drm: move the buddy allocator from i915 into common drm
URL : https://patchwork.freedesktop.org/series/96245/
State : failure
== Summary ==
Applying: drm: move the buddy allocator from i915 into common drm
Using index info to reconstruct a base tree...
M drivers/gpu/drm/Makefile
M drivers/gpu/drm/drm_drv.c
Falling back to patching base and 3-way merge...
Removing drivers/gpu/drm/i915/i915_buddy.h
Removing drivers/gpu/drm/i915/i915_buddy.c
Auto-merging drivers/gpu/drm/drm_drv.c
Auto-merging drivers/gpu/drm/Makefile
CONFLICT (content): Merge conflict in drivers/gpu/drm/Makefile
error: Failed to merge in the changes.
hint: Use 'git am --show-current-patch=diff' to see the failed patch
Patch failed at 0001 drm: move the buddy allocator from i915 into common drm
When you have resolved this problem, run "git am --continue".
If you prefer to skip this patch, run "git am --skip" instead.
To restore the original branch and stop patching, run "git am --abort".
^ permalink raw reply [flat|nested] 58+ messages in thread
* Re: [PATCH v2 2/8] drm: improve drm_buddy_alloc function
2021-10-25 13:00 ` [Intel-gfx] " Arunpravin
(?)
@ 2021-11-03 18:41 ` Matthew Auld
-1 siblings, 0 replies; 58+ messages in thread
From: Matthew Auld @ 2021-11-03 18:41 UTC (permalink / raw)
To: Arunpravin, dri-devel, intel-gfx, amd-gfx
Cc: alexander.deucher, tzimmermann, christian.koenig
On 25/10/2021 14:00, Arunpravin wrote:
> - Make drm_buddy_alloc a single function to handle
> range allocation and non-range allocation demands
>
> - Implemented a new function alloc_range() which allocates
> the requested power-of-two block comply with range limitations
>
> - Moved order computation and memory alignment logic from
> i915 driver to drm buddy
>
> V2:
> merged below changes to keep the build unbroken
> - drm_buddy_alloc_range() becomes obsolete and may be removed
> - enable ttm range allocation (fpfn / lpfn) support in i915 driver
> - apply enhanced drm_buddy_alloc() function to i915 driver
>
> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
> ---
> drivers/gpu/drm/drm_buddy.c | 265 +++++++++++-------
> drivers/gpu/drm/i915/i915_ttm_buddy_manager.c | 67 ++---
> drivers/gpu/drm/i915/i915_ttm_buddy_manager.h | 2 +
> include/drm/drm_buddy.h | 22 +-
> 4 files changed, 207 insertions(+), 149 deletions(-)
>
> diff --git a/drivers/gpu/drm/drm_buddy.c b/drivers/gpu/drm/drm_buddy.c
> index 39eb1d224bec..406e3d521903 100644
> --- a/drivers/gpu/drm/drm_buddy.c
> +++ b/drivers/gpu/drm/drm_buddy.c
> @@ -274,63 +274,6 @@ void drm_buddy_free_list(struct drm_buddy_mm *mm, struct list_head *objects)
> }
> EXPORT_SYMBOL(drm_buddy_free_list);
>
> -/**
> - * drm_buddy_alloc - allocate power-of-two blocks
> - *
> - * @mm: DRM buddy manager to allocate from
> - * @order: size of the allocation
> - *
> - * The order value here translates to:
> - *
> - * 0 = 2^0 * mm->chunk_size
> - * 1 = 2^1 * mm->chunk_size
> - * 2 = 2^2 * mm->chunk_size
> - *
> - * Returns:
> - * allocated ptr to the &drm_buddy_block on success
> - */
> -struct drm_buddy_block *
> -drm_buddy_alloc(struct drm_buddy_mm *mm, unsigned int order)
> -{
> - struct drm_buddy_block *block = NULL;
> - unsigned int i;
> - int err;
> -
> - for (i = order; i <= mm->max_order; ++i) {
> - block = list_first_entry_or_null(&mm->free_list[i],
> - struct drm_buddy_block,
> - link);
> - if (block)
> - break;
> - }
> -
> - if (!block)
> - return ERR_PTR(-ENOSPC);
> -
> - BUG_ON(!drm_buddy_block_is_free(block));
> -
> - while (i != order) {
> - err = split_block(mm, block);
> - if (unlikely(err))
> - goto out_free;
> -
> - /* Go low */
> - block = block->left;
> - i--;
> - }
> -
> - mark_allocated(block);
> - mm->avail -= drm_buddy_block_size(mm, block);
> - kmemleak_update_trace(block);
> - return block;
> -
> -out_free:
> - if (i != order)
> - __drm_buddy_free(mm, block);
> - return ERR_PTR(err);
> -}
> -EXPORT_SYMBOL(drm_buddy_alloc);
> -
> static inline bool overlaps(u64 s1, u64 e1, u64 s2, u64 e2)
> {
> return s1 <= e2 && e1 >= s2;
> @@ -341,52 +284,22 @@ static inline bool contains(u64 s1, u64 e1, u64 s2, u64 e2)
> return s1 <= s2 && e1 >= e2;
> }
>
> -/**
> - * drm_buddy_alloc_range - allocate range
> - *
> - * @mm: DRM buddy manager to allocate from
> - * @blocks: output list head to add allocated blocks
> - * @start: start of the allowed range for this block
> - * @size: size of the allocation
> - *
> - * Intended for pre-allocating portions of the address space, for example to
> - * reserve a block for the initial framebuffer or similar, hence the expectation
> - * here is that drm_buddy_alloc() is still the main vehicle for
> - * allocations, so if that's not the case then the drm_mm range allocator is
> - * probably a much better fit, and so you should probably go use that instead.
> - *
> - * Note that it's safe to chain together multiple alloc_ranges
> - * with the same blocks list
> - *
> - * Returns:
> - * 0 on success, error code on failure.
> - */
> -int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
> - struct list_head *blocks,
> - u64 start, u64 size)
> +static struct drm_buddy_block *
> +alloc_range(struct drm_buddy_mm *mm,
> + u64 start, u64 end,
> + unsigned int order)
> {
> struct drm_buddy_block *block;
> struct drm_buddy_block *buddy;
> - LIST_HEAD(allocated);
> LIST_HEAD(dfs);
> - u64 end;
> int err;
> int i;
>
> - if (size < mm->chunk_size)
> - return -EINVAL;
> -
> - if (!IS_ALIGNED(size | start, mm->chunk_size))
> - return -EINVAL;
> -
> - if (range_overflows(start, size, mm->size))
> - return -EINVAL;
> + end = end - 1;
>
> for (i = 0; i < mm->n_roots; ++i)
> list_add_tail(&mm->roots[i]->tmp_link, &dfs);
>
> - end = start + size - 1;
> -
> do {
> u64 block_start;
> u64 block_end;
> @@ -394,31 +307,32 @@ int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
> block = list_first_entry_or_null(&dfs,
> struct drm_buddy_block,
> tmp_link);
> +
No need for the newline.
> if (!block)
> break;
>
> list_del(&block->tmp_link);
>
> + if (drm_buddy_block_order(block) < order)
> + continue;
> +
> block_start = drm_buddy_block_offset(block);
> block_end = block_start + drm_buddy_block_size(mm, block) - 1;
>
> if (!overlaps(start, end, block_start, block_end))
> continue;
>
> - if (drm_buddy_block_is_allocated(block)) {
> - err = -ENOSPC;
> - goto err_free;
> - }
> + if (drm_buddy_block_is_allocated(block))
> + continue;
>
> - if (contains(start, end, block_start, block_end)) {
> - if (!drm_buddy_block_is_free(block)) {
> - err = -ENOSPC;
> - goto err_free;
> - }
> + if (contains(start, end, block_start, block_end)
> + && order == drm_buddy_block_order(block)) {
Alignment looks off, also && should be on the line above.
> + /*
> + * Find the free block within the range.
> + */
> + if (drm_buddy_block_is_free(block))
> + return block;
Would it make sense to keep searching here, rather than restarting the
search from scratch every time? Would it work if we pass in the total
size and min order?
I take it this will now also be used for allocating CPU mappable memory
using an end bias?
>
> - mark_allocated(block);
> - mm->avail -= drm_buddy_block_size(mm, block);
> - list_add_tail(&block->link, &allocated);
> continue;
> }
>
> @@ -428,12 +342,11 @@ int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
> goto err_undo;
> }
>
> - list_add(&block->right->tmp_link, &dfs);
> list_add(&block->left->tmp_link, &dfs);
> + list_add(&block->right->tmp_link, &dfs);
So this is now top-down? Do we need this? Also I guess this now changes
the ordering of the allocated blocks? Ideally they would appear in
order, starting from the lowest address(at least for the actual range
allocation, and not just the bias thing). I think some places might only
check the first block to get the device address, where underneath we
have a contiguous allocation.
> } while (1);
>
> - list_splice_tail(&allocated, blocks);
> - return 0;
> + return ERR_PTR(-ENOSPC);
>
> err_undo:
> /*
> @@ -446,12 +359,148 @@ int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
> (drm_buddy_block_is_free(block) &&
> drm_buddy_block_is_free(buddy)))
> __drm_buddy_free(mm, block);
> + return ERR_PTR(err);
> +}
> +
> +static struct drm_buddy_block *
> +alloc_from_freelist(struct drm_buddy_mm *mm,
> + unsigned int order,
> + unsigned long flags)
> +{
> + struct drm_buddy_block *block = NULL;
> + unsigned int i;
> + int err;
> +
> + for (i = order; i <= mm->max_order; ++i) {
> + if (!list_empty(&mm->free_list[i])) {
list_first_entry_or_null already handles that.
> + block = list_first_entry_or_null(&mm->free_list[i],
> + struct drm_buddy_block,
> + link);
> +
> + if (block)
> + break;
> + }
> + }
> +
> + if (!block)
> + return ERR_PTR(-ENOSPC);
> +
> + BUG_ON(!drm_buddy_block_is_free(block));
> +
> + while (i != order) {
> + err = split_block(mm, block);
> +
No need for the newline.
> + if (unlikely(err))
> + goto err_undo;
> +
> + block = block->right;
> + i--;
> + }
> + return block;
> +
> +err_undo:
> + if (i != order)
> + __drm_buddy_free(mm, block);
> + return ERR_PTR(err);
> +}
> +
> +/**
> + * drm_buddy_alloc - allocate power-of-two blocks
> + *
> + * @mm: DRM buddy manager to allocate from
> + * @start: start of the allowed range for this block
> + * @end: end of the allowed range for this block
> + * @size: size of the allocation
> + * @min_page_size: alignment of the allocation
> + * @blocks: output list head to add allocated blocks
> + * @flags: DRM_BUDDY_*_ALLOCATION flags
> + *
> + * alloc_range() called on range limitations, which traverses
> + * the tree and returns the desired block.
> + *
> + * alloc_from_freelist() called when *no* range restrictions
> + * are enforced, which picks the block from the freelist.
> + *
> + * blocks are allocated in order, the order value here translates to:
> + *
> + * 0 = 2^0 * mm->chunk_size
> + * 1 = 2^1 * mm->chunk_size
> + * 2 = 2^2 * mm->chunk_size
> + *
> + * Returns:
> + * 0 on success, error code on failure.
> + */
> +int drm_buddy_alloc(struct drm_buddy_mm *mm,
> + u64 start, u64 end, u64 size,
> + u64 min_page_size,
> + struct list_head *blocks,
> + unsigned long flags)
Do we need to validate the flags somewhere?
> +{
> + struct drm_buddy_block *block = NULL;
> + unsigned int min_order, order;
> + unsigned long pages;
> + LIST_HEAD(allocated);
> + int err;
> +
> + if (size < mm->chunk_size)
> + return -EINVAL;
Also:
if (min_page_size < mm->chunk_size)
return -EINVAL;
if (!is_power_of_2_u64(min_page_size))
return -EINVAL;
?
> +
> + if (!IS_ALIGNED(start, mm->chunk_size))
> + return -EINVAL;
> +
> + if (!IS_ALIGNED(end, mm->chunk_size))
> + return -EINVAL;
> +
> + if (!IS_ALIGNED(size, mm->chunk_size))
> + return -EINVAL;
Maybe:
if (!IS_ALIGNED(start | end | size, mm->chunk_size))
return -EINVAL;
?
> +
> + if (check_range_overflow(start, end, size, mm->size))
> + return -EINVAL;
> +
> + pages = size >> ilog2(mm->chunk_size);
> + min_order = ilog2(min_page_size) - ilog2(mm->chunk_size);
> +
> + do {
> + order = fls(pages) - 1;
Maybe:
order = min(order, fls(pages) - 1);
We shouldn't need to reset the order here all the way back to the number
of pages, I think. We are still holding the lock, and so we still get
the same answer as before, assuming we needed to decrement the order
below. Otherwise that might be slightly wasteful AFAIK.
> + BUG_ON(order > mm->max_order);
> + BUG_ON(order < min_order);
> +
> + do {
> + if (flags & DRM_BUDDY_RANGE_ALLOCATION)
> + /* Allocate traversing within the range */
> + block = alloc_range(mm, start, end, order);
Ok, so blocks might be in a random order, which is a slight concern for
actual range allocations(not the bias thing). Can we somehow make
alloc_range just do the old behaviour when end - start == size? Not the
end of the world though if not.
> + else
> + /* Allocate from freelist */
> + block = alloc_from_freelist(mm, order, flags);
> +
> + if (!IS_ERR(block))
> + break;
> +
> + if (order-- == min_order) {
> + err = -ENOSPC;
> + goto err_free;
> + }
> + } while (1);
> +
> + mark_allocated(block);
> + mm->avail -= drm_buddy_block_size(mm, block);
> + kmemleak_update_trace(block);
> + list_add_tail(&block->link, &allocated);
> +
> + pages -= BIT(order);
> +
> + if (!pages)
> + break;
> + } while (1);
> +
> + list_splice_tail(&allocated, blocks);
> + return 0;
>
> err_free:
> drm_buddy_free_list(mm, &allocated);
> return err;
> }
> -EXPORT_SYMBOL(drm_buddy_alloc_range);
> +EXPORT_SYMBOL(drm_buddy_alloc);
>
> /**
> * drm_buddy_block_print - print block information
> diff --git a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
> index c4b70cb8c248..75197ba6e40d 100644
> --- a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
> +++ b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
> @@ -36,13 +36,14 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
> struct i915_ttm_buddy_manager *bman = to_buddy_manager(man);
> struct i915_ttm_buddy_resource *bman_res;
> struct drm_buddy_mm *mm = &bman->mm;
> - unsigned long n_pages;
> - unsigned int min_order;
> + unsigned long n_pages, lpfn;
> u64 min_page_size;
> u64 size;
> int err;
>
> - GEM_BUG_ON(place->fpfn || place->lpfn);
> + lpfn = place->lpfn;
> + if (!lpfn)
> + lpfn = man->size;
>
> bman_res = kzalloc(sizeof(*bman_res), GFP_KERNEL);
> if (!bman_res)
> @@ -52,6 +53,9 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
> INIT_LIST_HEAD(&bman_res->blocks);
> bman_res->mm = mm;
>
> + if (place->fpfn || lpfn != man->size)
> + bman_res->flags |= DRM_BUDDY_RANGE_ALLOCATION;
> +
> GEM_BUG_ON(!bman_res->base.num_pages);
> size = bman_res->base.num_pages << PAGE_SHIFT;
>
> @@ -60,10 +64,16 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
> min_page_size = bo->page_alignment << PAGE_SHIFT;
>
> GEM_BUG_ON(min_page_size < mm->chunk_size);
> - min_order = ilog2(min_page_size) - ilog2(mm->chunk_size);
> +
> if (place->flags & TTM_PL_FLAG_CONTIGUOUS) {
> + unsigned long pages;
> +
> size = roundup_pow_of_two(size);
> - min_order = ilog2(size) - ilog2(mm->chunk_size);
> + min_page_size = size;
> +
> + pages = size >> ilog2(mm->chunk_size);
> + if (pages > lpfn)
> + lpfn = pages;
> }
>
> if (size > mm->size) {
> @@ -73,34 +83,17 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
>
> n_pages = size >> ilog2(mm->chunk_size);
>
> - do {
> - struct drm_buddy_block *block;
> - unsigned int order;
> -
> - order = fls(n_pages) - 1;
> - GEM_BUG_ON(order > mm->max_order);
> - GEM_BUG_ON(order < min_order);
> -
> - do {
> - mutex_lock(&bman->lock);
> - block = drm_buddy_alloc(mm, order);
> - mutex_unlock(&bman->lock);
> - if (!IS_ERR(block))
> - break;
> -
> - if (order-- == min_order) {
> - err = -ENOSPC;
> - goto err_free_blocks;
> - }
> - } while (1);
> -
> - n_pages -= BIT(order);
> -
> - list_add_tail(&block->link, &bman_res->blocks);
> + mutex_lock(&bman->lock);
> + err = drm_buddy_alloc(mm, (uint64_t)place->fpfn << PAGE_SHIFT,
> + (uint64_t)place->lpfn << PAGE_SHIFT,
> + (uint64_t)n_pages << PAGE_SHIFT,
s/uint64_t/u64/
Some other places also. AFAIK that is preferred in the kernel, outside
of maybe uapi headers.
> + min_page_size,
> + &bman_res->blocks,
> + bman_res->flags);
> + mutex_unlock(&bman->lock);
>
> - if (!n_pages)
> - break;
> - } while (1);
> + if (unlikely(err))
> + goto err_free_blocks;
>
> *res = &bman_res->base;
> return 0;
> @@ -266,10 +259,18 @@ int i915_ttm_buddy_man_reserve(struct ttm_resource_manager *man,
> {
> struct i915_ttm_buddy_manager *bman = to_buddy_manager(man);
> struct drm_buddy_mm *mm = &bman->mm;
> + unsigned long flags = 0;
> + u64 min_size;
> int ret;
>
> + min_size = size;
> + flags |= DRM_BUDDY_RANGE_ALLOCATION;
> +
> mutex_lock(&bman->lock);
> - ret = drm_buddy_alloc_range(mm, &bman->reserved, start, size);
> + ret = drm_buddy_alloc(mm, start, start + size,
> + size, min_size,
> + &bman->reserved,
> + flags);
> mutex_unlock(&bman->lock);
>
> return ret;
> diff --git a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.h b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.h
> index fa644b512c2e..5ba490875f66 100644
> --- a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.h
> +++ b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.h
> @@ -20,6 +20,7 @@ struct drm_buddy_mm;
> *
> * @base: struct ttm_resource base class we extend
> * @blocks: the list of struct i915_buddy_block for this resource/allocation
> + * @flags: DRM_BUDDY_*_ALLOCATION flags
> * @mm: the struct i915_buddy_mm for this resource
> *
> * Extends the struct ttm_resource to manage an address space allocation with
> @@ -28,6 +29,7 @@ struct drm_buddy_mm;
> struct i915_ttm_buddy_resource {
> struct ttm_resource base;
> struct list_head blocks;
> + unsigned long flags;
> struct drm_buddy_mm *mm;
> };
>
> diff --git a/include/drm/drm_buddy.h b/include/drm/drm_buddy.h
> index 5ce3fc702f80..c7bb5509a7ad 100644
> --- a/include/drm/drm_buddy.h
> +++ b/include/drm/drm_buddy.h
> @@ -13,15 +13,22 @@
>
> #include <drm/drm_print.h>
>
> -#define range_overflows(start, size, max) ({ \
> +#define check_range_overflow(start, end, size, max) ({ \
> typeof(start) start__ = (start); \
> + typeof(end) end__ = (end);\
> typeof(size) size__ = (size); \
> typeof(max) max__ = (max); \
> (void)(&start__ == &size__); \
> (void)(&start__ == &max__); \
> - start__ >= max__ || size__ > max__ - start__; \
> + (void)(&start__ == &end__); \
> + (void)(&end__ == &size__); \
> + (void)(&end__ == &max__); \
> + start__ >= max__ || end__ > max__ \
> + || size__ > end__ - start__; \
> })
>
> +#define DRM_BUDDY_RANGE_ALLOCATION (1 << 0)
Would it make sense to make this an enum or something?
enum drm_buddy_alloc_mode {
DRM_BUDDY_ALLOC_FIRST = 0,
DRM_BUDDY_ALLOC_TOPDOWN,
DRM_BUDDY_ALLOC_RANGE,
};
That way can more easily see where this is actually used, rather than
having an opaque 'unsigned long flags'? Or are we expecting to combine
allocation strategies or something? Anyway, just a thought.
> +
> struct drm_buddy_block {
> #define DRM_BUDDY_HEADER_OFFSET GENMASK_ULL(63, 12)
> #define DRM_BUDDY_HEADER_STATE GENMASK_ULL(11, 10)
> @@ -131,12 +138,11 @@ int drm_buddy_init(struct drm_buddy_mm *mm, u64 size, u64 chunk_size);
>
> void drm_buddy_fini(struct drm_buddy_mm *mm);
>
> -struct drm_buddy_block *
> -drm_buddy_alloc(struct drm_buddy_mm *mm, unsigned int order);
> -
> -int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
> - struct list_head *blocks,
> - u64 start, u64 size);
> +int drm_buddy_alloc(struct drm_buddy_mm *mm,
> + u64 start, u64 end, u64 size,
> + u64 min_page_size,
> + struct list_head *blocks,
> + unsigned long flags);
>
> void drm_buddy_free(struct drm_buddy_mm *mm, struct drm_buddy_block *block);
>
>
^ permalink raw reply [flat|nested] 58+ messages in thread
* Re: [Intel-gfx] [PATCH v2 2/8] drm: improve drm_buddy_alloc function
@ 2021-11-03 18:41 ` Matthew Auld
0 siblings, 0 replies; 58+ messages in thread
From: Matthew Auld @ 2021-11-03 18:41 UTC (permalink / raw)
To: Arunpravin, dri-devel, intel-gfx, amd-gfx
Cc: alexander.deucher, tzimmermann, christian.koenig
On 25/10/2021 14:00, Arunpravin wrote:
> - Make drm_buddy_alloc a single function to handle
> range allocation and non-range allocation demands
>
> - Implemented a new function alloc_range() which allocates
> the requested power-of-two block comply with range limitations
>
> - Moved order computation and memory alignment logic from
> i915 driver to drm buddy
>
> V2:
> merged below changes to keep the build unbroken
> - drm_buddy_alloc_range() becomes obsolete and may be removed
> - enable ttm range allocation (fpfn / lpfn) support in i915 driver
> - apply enhanced drm_buddy_alloc() function to i915 driver
>
> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
> ---
> drivers/gpu/drm/drm_buddy.c | 265 +++++++++++-------
> drivers/gpu/drm/i915/i915_ttm_buddy_manager.c | 67 ++---
> drivers/gpu/drm/i915/i915_ttm_buddy_manager.h | 2 +
> include/drm/drm_buddy.h | 22 +-
> 4 files changed, 207 insertions(+), 149 deletions(-)
>
> diff --git a/drivers/gpu/drm/drm_buddy.c b/drivers/gpu/drm/drm_buddy.c
> index 39eb1d224bec..406e3d521903 100644
> --- a/drivers/gpu/drm/drm_buddy.c
> +++ b/drivers/gpu/drm/drm_buddy.c
> @@ -274,63 +274,6 @@ void drm_buddy_free_list(struct drm_buddy_mm *mm, struct list_head *objects)
> }
> EXPORT_SYMBOL(drm_buddy_free_list);
>
> -/**
> - * drm_buddy_alloc - allocate power-of-two blocks
> - *
> - * @mm: DRM buddy manager to allocate from
> - * @order: size of the allocation
> - *
> - * The order value here translates to:
> - *
> - * 0 = 2^0 * mm->chunk_size
> - * 1 = 2^1 * mm->chunk_size
> - * 2 = 2^2 * mm->chunk_size
> - *
> - * Returns:
> - * allocated ptr to the &drm_buddy_block on success
> - */
> -struct drm_buddy_block *
> -drm_buddy_alloc(struct drm_buddy_mm *mm, unsigned int order)
> -{
> - struct drm_buddy_block *block = NULL;
> - unsigned int i;
> - int err;
> -
> - for (i = order; i <= mm->max_order; ++i) {
> - block = list_first_entry_or_null(&mm->free_list[i],
> - struct drm_buddy_block,
> - link);
> - if (block)
> - break;
> - }
> -
> - if (!block)
> - return ERR_PTR(-ENOSPC);
> -
> - BUG_ON(!drm_buddy_block_is_free(block));
> -
> - while (i != order) {
> - err = split_block(mm, block);
> - if (unlikely(err))
> - goto out_free;
> -
> - /* Go low */
> - block = block->left;
> - i--;
> - }
> -
> - mark_allocated(block);
> - mm->avail -= drm_buddy_block_size(mm, block);
> - kmemleak_update_trace(block);
> - return block;
> -
> -out_free:
> - if (i != order)
> - __drm_buddy_free(mm, block);
> - return ERR_PTR(err);
> -}
> -EXPORT_SYMBOL(drm_buddy_alloc);
> -
> static inline bool overlaps(u64 s1, u64 e1, u64 s2, u64 e2)
> {
> return s1 <= e2 && e1 >= s2;
> @@ -341,52 +284,22 @@ static inline bool contains(u64 s1, u64 e1, u64 s2, u64 e2)
> return s1 <= s2 && e1 >= e2;
> }
>
> -/**
> - * drm_buddy_alloc_range - allocate range
> - *
> - * @mm: DRM buddy manager to allocate from
> - * @blocks: output list head to add allocated blocks
> - * @start: start of the allowed range for this block
> - * @size: size of the allocation
> - *
> - * Intended for pre-allocating portions of the address space, for example to
> - * reserve a block for the initial framebuffer or similar, hence the expectation
> - * here is that drm_buddy_alloc() is still the main vehicle for
> - * allocations, so if that's not the case then the drm_mm range allocator is
> - * probably a much better fit, and so you should probably go use that instead.
> - *
> - * Note that it's safe to chain together multiple alloc_ranges
> - * with the same blocks list
> - *
> - * Returns:
> - * 0 on success, error code on failure.
> - */
> -int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
> - struct list_head *blocks,
> - u64 start, u64 size)
> +static struct drm_buddy_block *
> +alloc_range(struct drm_buddy_mm *mm,
> + u64 start, u64 end,
> + unsigned int order)
> {
> struct drm_buddy_block *block;
> struct drm_buddy_block *buddy;
> - LIST_HEAD(allocated);
> LIST_HEAD(dfs);
> - u64 end;
> int err;
> int i;
>
> - if (size < mm->chunk_size)
> - return -EINVAL;
> -
> - if (!IS_ALIGNED(size | start, mm->chunk_size))
> - return -EINVAL;
> -
> - if (range_overflows(start, size, mm->size))
> - return -EINVAL;
> + end = end - 1;
>
> for (i = 0; i < mm->n_roots; ++i)
> list_add_tail(&mm->roots[i]->tmp_link, &dfs);
>
> - end = start + size - 1;
> -
> do {
> u64 block_start;
> u64 block_end;
> @@ -394,31 +307,32 @@ int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
> block = list_first_entry_or_null(&dfs,
> struct drm_buddy_block,
> tmp_link);
> +
No need for the newline.
> if (!block)
> break;
>
> list_del(&block->tmp_link);
>
> + if (drm_buddy_block_order(block) < order)
> + continue;
> +
> block_start = drm_buddy_block_offset(block);
> block_end = block_start + drm_buddy_block_size(mm, block) - 1;
>
> if (!overlaps(start, end, block_start, block_end))
> continue;
>
> - if (drm_buddy_block_is_allocated(block)) {
> - err = -ENOSPC;
> - goto err_free;
> - }
> + if (drm_buddy_block_is_allocated(block))
> + continue;
>
> - if (contains(start, end, block_start, block_end)) {
> - if (!drm_buddy_block_is_free(block)) {
> - err = -ENOSPC;
> - goto err_free;
> - }
> + if (contains(start, end, block_start, block_end)
> + && order == drm_buddy_block_order(block)) {
Alignment looks off, also && should be on the line above.
> + /*
> + * Find the free block within the range.
> + */
> + if (drm_buddy_block_is_free(block))
> + return block;
Would it make sense to keep searching here, rather than restarting the
search from scratch every time? Would it work if we pass in the total
size and min order?
I take it this will now also be used for allocating CPU mappable memory
using an end bias?
>
> - mark_allocated(block);
> - mm->avail -= drm_buddy_block_size(mm, block);
> - list_add_tail(&block->link, &allocated);
> continue;
> }
>
> @@ -428,12 +342,11 @@ int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
> goto err_undo;
> }
>
> - list_add(&block->right->tmp_link, &dfs);
> list_add(&block->left->tmp_link, &dfs);
> + list_add(&block->right->tmp_link, &dfs);
So this is now top-down? Do we need this? Also I guess this now changes
the ordering of the allocated blocks? Ideally they would appear in
order, starting from the lowest address(at least for the actual range
allocation, and not just the bias thing). I think some places might only
check the first block to get the device address, where underneath we
have a contiguous allocation.
> } while (1);
>
> - list_splice_tail(&allocated, blocks);
> - return 0;
> + return ERR_PTR(-ENOSPC);
>
> err_undo:
> /*
> @@ -446,12 +359,148 @@ int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
> (drm_buddy_block_is_free(block) &&
> drm_buddy_block_is_free(buddy)))
> __drm_buddy_free(mm, block);
> + return ERR_PTR(err);
> +}
> +
> +static struct drm_buddy_block *
> +alloc_from_freelist(struct drm_buddy_mm *mm,
> + unsigned int order,
> + unsigned long flags)
> +{
> + struct drm_buddy_block *block = NULL;
> + unsigned int i;
> + int err;
> +
> + for (i = order; i <= mm->max_order; ++i) {
> + if (!list_empty(&mm->free_list[i])) {
list_first_entry_or_null already handles that.
> + block = list_first_entry_or_null(&mm->free_list[i],
> + struct drm_buddy_block,
> + link);
> +
> + if (block)
> + break;
> + }
> + }
> +
> + if (!block)
> + return ERR_PTR(-ENOSPC);
> +
> + BUG_ON(!drm_buddy_block_is_free(block));
> +
> + while (i != order) {
> + err = split_block(mm, block);
> +
No need for the newline.
> + if (unlikely(err))
> + goto err_undo;
> +
> + block = block->right;
> + i--;
> + }
> + return block;
> +
> +err_undo:
> + if (i != order)
> + __drm_buddy_free(mm, block);
> + return ERR_PTR(err);
> +}
> +
> +/**
> + * drm_buddy_alloc - allocate power-of-two blocks
> + *
> + * @mm: DRM buddy manager to allocate from
> + * @start: start of the allowed range for this block
> + * @end: end of the allowed range for this block
> + * @size: size of the allocation
> + * @min_page_size: alignment of the allocation
> + * @blocks: output list head to add allocated blocks
> + * @flags: DRM_BUDDY_*_ALLOCATION flags
> + *
> + * alloc_range() called on range limitations, which traverses
> + * the tree and returns the desired block.
> + *
> + * alloc_from_freelist() called when *no* range restrictions
> + * are enforced, which picks the block from the freelist.
> + *
> + * blocks are allocated in order, the order value here translates to:
> + *
> + * 0 = 2^0 * mm->chunk_size
> + * 1 = 2^1 * mm->chunk_size
> + * 2 = 2^2 * mm->chunk_size
> + *
> + * Returns:
> + * 0 on success, error code on failure.
> + */
> +int drm_buddy_alloc(struct drm_buddy_mm *mm,
> + u64 start, u64 end, u64 size,
> + u64 min_page_size,
> + struct list_head *blocks,
> + unsigned long flags)
Do we need to validate the flags somewhere?
> +{
> + struct drm_buddy_block *block = NULL;
> + unsigned int min_order, order;
> + unsigned long pages;
> + LIST_HEAD(allocated);
> + int err;
> +
> + if (size < mm->chunk_size)
> + return -EINVAL;
Also:
if (min_page_size < mm->chunk_size)
return -EINVAL;
if (!is_power_of_2_u64(min_page_size))
return -EINVAL;
?
> +
> + if (!IS_ALIGNED(start, mm->chunk_size))
> + return -EINVAL;
> +
> + if (!IS_ALIGNED(end, mm->chunk_size))
> + return -EINVAL;
> +
> + if (!IS_ALIGNED(size, mm->chunk_size))
> + return -EINVAL;
Maybe:
if (!IS_ALIGNED(start | end | size, mm->chunk_size))
return -EINVAL;
?
> +
> + if (check_range_overflow(start, end, size, mm->size))
> + return -EINVAL;
> +
> + pages = size >> ilog2(mm->chunk_size);
> + min_order = ilog2(min_page_size) - ilog2(mm->chunk_size);
> +
> + do {
> + order = fls(pages) - 1;
Maybe:
order = min(order, fls(pages) - 1);
We shouldn't need to reset the order here all the way back to the number
of pages, I think. We are still holding the lock, and so we still get
the same answer as before, assuming we needed to decrement the order
below. Otherwise that might be slightly wasteful AFAIK.
> + BUG_ON(order > mm->max_order);
> + BUG_ON(order < min_order);
> +
> + do {
> + if (flags & DRM_BUDDY_RANGE_ALLOCATION)
> + /* Allocate traversing within the range */
> + block = alloc_range(mm, start, end, order);
Ok, so blocks might be in a random order, which is a slight concern for
actual range allocations(not the bias thing). Can we somehow make
alloc_range just do the old behaviour when end - start == size? Not the
end of the world though if not.
> + else
> + /* Allocate from freelist */
> + block = alloc_from_freelist(mm, order, flags);
> +
> + if (!IS_ERR(block))
> + break;
> +
> + if (order-- == min_order) {
> + err = -ENOSPC;
> + goto err_free;
> + }
> + } while (1);
> +
> + mark_allocated(block);
> + mm->avail -= drm_buddy_block_size(mm, block);
> + kmemleak_update_trace(block);
> + list_add_tail(&block->link, &allocated);
> +
> + pages -= BIT(order);
> +
> + if (!pages)
> + break;
> + } while (1);
> +
> + list_splice_tail(&allocated, blocks);
> + return 0;
>
> err_free:
> drm_buddy_free_list(mm, &allocated);
> return err;
> }
> -EXPORT_SYMBOL(drm_buddy_alloc_range);
> +EXPORT_SYMBOL(drm_buddy_alloc);
>
> /**
> * drm_buddy_block_print - print block information
> diff --git a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
> index c4b70cb8c248..75197ba6e40d 100644
> --- a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
> +++ b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
> @@ -36,13 +36,14 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
> struct i915_ttm_buddy_manager *bman = to_buddy_manager(man);
> struct i915_ttm_buddy_resource *bman_res;
> struct drm_buddy_mm *mm = &bman->mm;
> - unsigned long n_pages;
> - unsigned int min_order;
> + unsigned long n_pages, lpfn;
> u64 min_page_size;
> u64 size;
> int err;
>
> - GEM_BUG_ON(place->fpfn || place->lpfn);
> + lpfn = place->lpfn;
> + if (!lpfn)
> + lpfn = man->size;
>
> bman_res = kzalloc(sizeof(*bman_res), GFP_KERNEL);
> if (!bman_res)
> @@ -52,6 +53,9 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
> INIT_LIST_HEAD(&bman_res->blocks);
> bman_res->mm = mm;
>
> + if (place->fpfn || lpfn != man->size)
> + bman_res->flags |= DRM_BUDDY_RANGE_ALLOCATION;
> +
> GEM_BUG_ON(!bman_res->base.num_pages);
> size = bman_res->base.num_pages << PAGE_SHIFT;
>
> @@ -60,10 +64,16 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
> min_page_size = bo->page_alignment << PAGE_SHIFT;
>
> GEM_BUG_ON(min_page_size < mm->chunk_size);
> - min_order = ilog2(min_page_size) - ilog2(mm->chunk_size);
> +
> if (place->flags & TTM_PL_FLAG_CONTIGUOUS) {
> + unsigned long pages;
> +
> size = roundup_pow_of_two(size);
> - min_order = ilog2(size) - ilog2(mm->chunk_size);
> + min_page_size = size;
> +
> + pages = size >> ilog2(mm->chunk_size);
> + if (pages > lpfn)
> + lpfn = pages;
> }
>
> if (size > mm->size) {
> @@ -73,34 +83,17 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
>
> n_pages = size >> ilog2(mm->chunk_size);
>
> - do {
> - struct drm_buddy_block *block;
> - unsigned int order;
> -
> - order = fls(n_pages) - 1;
> - GEM_BUG_ON(order > mm->max_order);
> - GEM_BUG_ON(order < min_order);
> -
> - do {
> - mutex_lock(&bman->lock);
> - block = drm_buddy_alloc(mm, order);
> - mutex_unlock(&bman->lock);
> - if (!IS_ERR(block))
> - break;
> -
> - if (order-- == min_order) {
> - err = -ENOSPC;
> - goto err_free_blocks;
> - }
> - } while (1);
> -
> - n_pages -= BIT(order);
> -
> - list_add_tail(&block->link, &bman_res->blocks);
> + mutex_lock(&bman->lock);
> + err = drm_buddy_alloc(mm, (uint64_t)place->fpfn << PAGE_SHIFT,
> + (uint64_t)place->lpfn << PAGE_SHIFT,
> + (uint64_t)n_pages << PAGE_SHIFT,
s/uint64_t/u64/
Some other places also. AFAIK that is preferred in the kernel, outside
of maybe uapi headers.
> + min_page_size,
> + &bman_res->blocks,
> + bman_res->flags);
> + mutex_unlock(&bman->lock);
>
> - if (!n_pages)
> - break;
> - } while (1);
> + if (unlikely(err))
> + goto err_free_blocks;
>
> *res = &bman_res->base;
> return 0;
> @@ -266,10 +259,18 @@ int i915_ttm_buddy_man_reserve(struct ttm_resource_manager *man,
> {
> struct i915_ttm_buddy_manager *bman = to_buddy_manager(man);
> struct drm_buddy_mm *mm = &bman->mm;
> + unsigned long flags = 0;
> + u64 min_size;
> int ret;
>
> + min_size = size;
> + flags |= DRM_BUDDY_RANGE_ALLOCATION;
> +
> mutex_lock(&bman->lock);
> - ret = drm_buddy_alloc_range(mm, &bman->reserved, start, size);
> + ret = drm_buddy_alloc(mm, start, start + size,
> + size, min_size,
> + &bman->reserved,
> + flags);
> mutex_unlock(&bman->lock);
>
> return ret;
> diff --git a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.h b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.h
> index fa644b512c2e..5ba490875f66 100644
> --- a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.h
> +++ b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.h
> @@ -20,6 +20,7 @@ struct drm_buddy_mm;
> *
> * @base: struct ttm_resource base class we extend
> * @blocks: the list of struct i915_buddy_block for this resource/allocation
> + * @flags: DRM_BUDDY_*_ALLOCATION flags
> * @mm: the struct i915_buddy_mm for this resource
> *
> * Extends the struct ttm_resource to manage an address space allocation with
> @@ -28,6 +29,7 @@ struct drm_buddy_mm;
> struct i915_ttm_buddy_resource {
> struct ttm_resource base;
> struct list_head blocks;
> + unsigned long flags;
> struct drm_buddy_mm *mm;
> };
>
> diff --git a/include/drm/drm_buddy.h b/include/drm/drm_buddy.h
> index 5ce3fc702f80..c7bb5509a7ad 100644
> --- a/include/drm/drm_buddy.h
> +++ b/include/drm/drm_buddy.h
> @@ -13,15 +13,22 @@
>
> #include <drm/drm_print.h>
>
> -#define range_overflows(start, size, max) ({ \
> +#define check_range_overflow(start, end, size, max) ({ \
> typeof(start) start__ = (start); \
> + typeof(end) end__ = (end);\
> typeof(size) size__ = (size); \
> typeof(max) max__ = (max); \
> (void)(&start__ == &size__); \
> (void)(&start__ == &max__); \
> - start__ >= max__ || size__ > max__ - start__; \
> + (void)(&start__ == &end__); \
> + (void)(&end__ == &size__); \
> + (void)(&end__ == &max__); \
> + start__ >= max__ || end__ > max__ \
> + || size__ > end__ - start__; \
> })
>
> +#define DRM_BUDDY_RANGE_ALLOCATION (1 << 0)
Would it make sense to make this an enum or something?
enum drm_buddy_alloc_mode {
DRM_BUDDY_ALLOC_FIRST = 0,
DRM_BUDDY_ALLOC_TOPDOWN,
DRM_BUDDY_ALLOC_RANGE,
};
That way can more easily see where this is actually used, rather than
having an opaque 'unsigned long flags'? Or are we expecting to combine
allocation strategies or something? Anyway, just a thought.
> +
> struct drm_buddy_block {
> #define DRM_BUDDY_HEADER_OFFSET GENMASK_ULL(63, 12)
> #define DRM_BUDDY_HEADER_STATE GENMASK_ULL(11, 10)
> @@ -131,12 +138,11 @@ int drm_buddy_init(struct drm_buddy_mm *mm, u64 size, u64 chunk_size);
>
> void drm_buddy_fini(struct drm_buddy_mm *mm);
>
> -struct drm_buddy_block *
> -drm_buddy_alloc(struct drm_buddy_mm *mm, unsigned int order);
> -
> -int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
> - struct list_head *blocks,
> - u64 start, u64 size);
> +int drm_buddy_alloc(struct drm_buddy_mm *mm,
> + u64 start, u64 end, u64 size,
> + u64 min_page_size,
> + struct list_head *blocks,
> + unsigned long flags);
>
> void drm_buddy_free(struct drm_buddy_mm *mm, struct drm_buddy_block *block);
>
>
^ permalink raw reply [flat|nested] 58+ messages in thread
* Re: [PATCH v2 2/8] drm: improve drm_buddy_alloc function
@ 2021-11-03 18:41 ` Matthew Auld
0 siblings, 0 replies; 58+ messages in thread
From: Matthew Auld @ 2021-11-03 18:41 UTC (permalink / raw)
To: Arunpravin, dri-devel, intel-gfx, amd-gfx
Cc: alexander.deucher, tzimmermann, jani.nikula, christian.koenig, daniel
On 25/10/2021 14:00, Arunpravin wrote:
> - Make drm_buddy_alloc a single function to handle
> range allocation and non-range allocation demands
>
> - Implemented a new function alloc_range() which allocates
> the requested power-of-two block comply with range limitations
>
> - Moved order computation and memory alignment logic from
> i915 driver to drm buddy
>
> V2:
> merged below changes to keep the build unbroken
> - drm_buddy_alloc_range() becomes obsolete and may be removed
> - enable ttm range allocation (fpfn / lpfn) support in i915 driver
> - apply enhanced drm_buddy_alloc() function to i915 driver
>
> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
> ---
> drivers/gpu/drm/drm_buddy.c | 265 +++++++++++-------
> drivers/gpu/drm/i915/i915_ttm_buddy_manager.c | 67 ++---
> drivers/gpu/drm/i915/i915_ttm_buddy_manager.h | 2 +
> include/drm/drm_buddy.h | 22 +-
> 4 files changed, 207 insertions(+), 149 deletions(-)
>
> diff --git a/drivers/gpu/drm/drm_buddy.c b/drivers/gpu/drm/drm_buddy.c
> index 39eb1d224bec..406e3d521903 100644
> --- a/drivers/gpu/drm/drm_buddy.c
> +++ b/drivers/gpu/drm/drm_buddy.c
> @@ -274,63 +274,6 @@ void drm_buddy_free_list(struct drm_buddy_mm *mm, struct list_head *objects)
> }
> EXPORT_SYMBOL(drm_buddy_free_list);
>
> -/**
> - * drm_buddy_alloc - allocate power-of-two blocks
> - *
> - * @mm: DRM buddy manager to allocate from
> - * @order: size of the allocation
> - *
> - * The order value here translates to:
> - *
> - * 0 = 2^0 * mm->chunk_size
> - * 1 = 2^1 * mm->chunk_size
> - * 2 = 2^2 * mm->chunk_size
> - *
> - * Returns:
> - * allocated ptr to the &drm_buddy_block on success
> - */
> -struct drm_buddy_block *
> -drm_buddy_alloc(struct drm_buddy_mm *mm, unsigned int order)
> -{
> - struct drm_buddy_block *block = NULL;
> - unsigned int i;
> - int err;
> -
> - for (i = order; i <= mm->max_order; ++i) {
> - block = list_first_entry_or_null(&mm->free_list[i],
> - struct drm_buddy_block,
> - link);
> - if (block)
> - break;
> - }
> -
> - if (!block)
> - return ERR_PTR(-ENOSPC);
> -
> - BUG_ON(!drm_buddy_block_is_free(block));
> -
> - while (i != order) {
> - err = split_block(mm, block);
> - if (unlikely(err))
> - goto out_free;
> -
> - /* Go low */
> - block = block->left;
> - i--;
> - }
> -
> - mark_allocated(block);
> - mm->avail -= drm_buddy_block_size(mm, block);
> - kmemleak_update_trace(block);
> - return block;
> -
> -out_free:
> - if (i != order)
> - __drm_buddy_free(mm, block);
> - return ERR_PTR(err);
> -}
> -EXPORT_SYMBOL(drm_buddy_alloc);
> -
> static inline bool overlaps(u64 s1, u64 e1, u64 s2, u64 e2)
> {
> return s1 <= e2 && e1 >= s2;
> @@ -341,52 +284,22 @@ static inline bool contains(u64 s1, u64 e1, u64 s2, u64 e2)
> return s1 <= s2 && e1 >= e2;
> }
>
> -/**
> - * drm_buddy_alloc_range - allocate range
> - *
> - * @mm: DRM buddy manager to allocate from
> - * @blocks: output list head to add allocated blocks
> - * @start: start of the allowed range for this block
> - * @size: size of the allocation
> - *
> - * Intended for pre-allocating portions of the address space, for example to
> - * reserve a block for the initial framebuffer or similar, hence the expectation
> - * here is that drm_buddy_alloc() is still the main vehicle for
> - * allocations, so if that's not the case then the drm_mm range allocator is
> - * probably a much better fit, and so you should probably go use that instead.
> - *
> - * Note that it's safe to chain together multiple alloc_ranges
> - * with the same blocks list
> - *
> - * Returns:
> - * 0 on success, error code on failure.
> - */
> -int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
> - struct list_head *blocks,
> - u64 start, u64 size)
> +static struct drm_buddy_block *
> +alloc_range(struct drm_buddy_mm *mm,
> + u64 start, u64 end,
> + unsigned int order)
> {
> struct drm_buddy_block *block;
> struct drm_buddy_block *buddy;
> - LIST_HEAD(allocated);
> LIST_HEAD(dfs);
> - u64 end;
> int err;
> int i;
>
> - if (size < mm->chunk_size)
> - return -EINVAL;
> -
> - if (!IS_ALIGNED(size | start, mm->chunk_size))
> - return -EINVAL;
> -
> - if (range_overflows(start, size, mm->size))
> - return -EINVAL;
> + end = end - 1;
>
> for (i = 0; i < mm->n_roots; ++i)
> list_add_tail(&mm->roots[i]->tmp_link, &dfs);
>
> - end = start + size - 1;
> -
> do {
> u64 block_start;
> u64 block_end;
> @@ -394,31 +307,32 @@ int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
> block = list_first_entry_or_null(&dfs,
> struct drm_buddy_block,
> tmp_link);
> +
No need for the newline.
> if (!block)
> break;
>
> list_del(&block->tmp_link);
>
> + if (drm_buddy_block_order(block) < order)
> + continue;
> +
> block_start = drm_buddy_block_offset(block);
> block_end = block_start + drm_buddy_block_size(mm, block) - 1;
>
> if (!overlaps(start, end, block_start, block_end))
> continue;
>
> - if (drm_buddy_block_is_allocated(block)) {
> - err = -ENOSPC;
> - goto err_free;
> - }
> + if (drm_buddy_block_is_allocated(block))
> + continue;
>
> - if (contains(start, end, block_start, block_end)) {
> - if (!drm_buddy_block_is_free(block)) {
> - err = -ENOSPC;
> - goto err_free;
> - }
> + if (contains(start, end, block_start, block_end)
> + && order == drm_buddy_block_order(block)) {
Alignment looks off, also && should be on the line above.
> + /*
> + * Find the free block within the range.
> + */
> + if (drm_buddy_block_is_free(block))
> + return block;
Would it make sense to keep searching here, rather than restarting the
search from scratch every time? Would it work if we pass in the total
size and min order?
I take it this will now also be used for allocating CPU mappable memory
using an end bias?
>
> - mark_allocated(block);
> - mm->avail -= drm_buddy_block_size(mm, block);
> - list_add_tail(&block->link, &allocated);
> continue;
> }
>
> @@ -428,12 +342,11 @@ int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
> goto err_undo;
> }
>
> - list_add(&block->right->tmp_link, &dfs);
> list_add(&block->left->tmp_link, &dfs);
> + list_add(&block->right->tmp_link, &dfs);
So this is now top-down? Do we need this? Also I guess this now changes
the ordering of the allocated blocks? Ideally they would appear in
order, starting from the lowest address(at least for the actual range
allocation, and not just the bias thing). I think some places might only
check the first block to get the device address, where underneath we
have a contiguous allocation.
> } while (1);
>
> - list_splice_tail(&allocated, blocks);
> - return 0;
> + return ERR_PTR(-ENOSPC);
>
> err_undo:
> /*
> @@ -446,12 +359,148 @@ int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
> (drm_buddy_block_is_free(block) &&
> drm_buddy_block_is_free(buddy)))
> __drm_buddy_free(mm, block);
> + return ERR_PTR(err);
> +}
> +
> +static struct drm_buddy_block *
> +alloc_from_freelist(struct drm_buddy_mm *mm,
> + unsigned int order,
> + unsigned long flags)
> +{
> + struct drm_buddy_block *block = NULL;
> + unsigned int i;
> + int err;
> +
> + for (i = order; i <= mm->max_order; ++i) {
> + if (!list_empty(&mm->free_list[i])) {
list_first_entry_or_null already handles that.
> + block = list_first_entry_or_null(&mm->free_list[i],
> + struct drm_buddy_block,
> + link);
> +
> + if (block)
> + break;
> + }
> + }
> +
> + if (!block)
> + return ERR_PTR(-ENOSPC);
> +
> + BUG_ON(!drm_buddy_block_is_free(block));
> +
> + while (i != order) {
> + err = split_block(mm, block);
> +
No need for the newline.
> + if (unlikely(err))
> + goto err_undo;
> +
> + block = block->right;
> + i--;
> + }
> + return block;
> +
> +err_undo:
> + if (i != order)
> + __drm_buddy_free(mm, block);
> + return ERR_PTR(err);
> +}
> +
> +/**
> + * drm_buddy_alloc - allocate power-of-two blocks
> + *
> + * @mm: DRM buddy manager to allocate from
> + * @start: start of the allowed range for this block
> + * @end: end of the allowed range for this block
> + * @size: size of the allocation
> + * @min_page_size: alignment of the allocation
> + * @blocks: output list head to add allocated blocks
> + * @flags: DRM_BUDDY_*_ALLOCATION flags
> + *
> + * alloc_range() called on range limitations, which traverses
> + * the tree and returns the desired block.
> + *
> + * alloc_from_freelist() called when *no* range restrictions
> + * are enforced, which picks the block from the freelist.
> + *
> + * blocks are allocated in order, the order value here translates to:
> + *
> + * 0 = 2^0 * mm->chunk_size
> + * 1 = 2^1 * mm->chunk_size
> + * 2 = 2^2 * mm->chunk_size
> + *
> + * Returns:
> + * 0 on success, error code on failure.
> + */
> +int drm_buddy_alloc(struct drm_buddy_mm *mm,
> + u64 start, u64 end, u64 size,
> + u64 min_page_size,
> + struct list_head *blocks,
> + unsigned long flags)
Do we need to validate the flags somewhere?
> +{
> + struct drm_buddy_block *block = NULL;
> + unsigned int min_order, order;
> + unsigned long pages;
> + LIST_HEAD(allocated);
> + int err;
> +
> + if (size < mm->chunk_size)
> + return -EINVAL;
Also:
if (min_page_size < mm->chunk_size)
return -EINVAL;
if (!is_power_of_2_u64(min_page_size))
return -EINVAL;
?
> +
> + if (!IS_ALIGNED(start, mm->chunk_size))
> + return -EINVAL;
> +
> + if (!IS_ALIGNED(end, mm->chunk_size))
> + return -EINVAL;
> +
> + if (!IS_ALIGNED(size, mm->chunk_size))
> + return -EINVAL;
Maybe:
if (!IS_ALIGNED(start | end | size, mm->chunk_size))
return -EINVAL;
?
> +
> + if (check_range_overflow(start, end, size, mm->size))
> + return -EINVAL;
> +
> + pages = size >> ilog2(mm->chunk_size);
> + min_order = ilog2(min_page_size) - ilog2(mm->chunk_size);
> +
> + do {
> + order = fls(pages) - 1;
Maybe:
order = min(order, fls(pages) - 1);
We shouldn't need to reset the order here all the way back to the number
of pages, I think. We are still holding the lock, and so we still get
the same answer as before, assuming we needed to decrement the order
below. Otherwise that might be slightly wasteful AFAIK.
> + BUG_ON(order > mm->max_order);
> + BUG_ON(order < min_order);
> +
> + do {
> + if (flags & DRM_BUDDY_RANGE_ALLOCATION)
> + /* Allocate traversing within the range */
> + block = alloc_range(mm, start, end, order);
Ok, so blocks might be in a random order, which is a slight concern for
actual range allocations(not the bias thing). Can we somehow make
alloc_range just do the old behaviour when end - start == size? Not the
end of the world though if not.
> + else
> + /* Allocate from freelist */
> + block = alloc_from_freelist(mm, order, flags);
> +
> + if (!IS_ERR(block))
> + break;
> +
> + if (order-- == min_order) {
> + err = -ENOSPC;
> + goto err_free;
> + }
> + } while (1);
> +
> + mark_allocated(block);
> + mm->avail -= drm_buddy_block_size(mm, block);
> + kmemleak_update_trace(block);
> + list_add_tail(&block->link, &allocated);
> +
> + pages -= BIT(order);
> +
> + if (!pages)
> + break;
> + } while (1);
> +
> + list_splice_tail(&allocated, blocks);
> + return 0;
>
> err_free:
> drm_buddy_free_list(mm, &allocated);
> return err;
> }
> -EXPORT_SYMBOL(drm_buddy_alloc_range);
> +EXPORT_SYMBOL(drm_buddy_alloc);
>
> /**
> * drm_buddy_block_print - print block information
> diff --git a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
> index c4b70cb8c248..75197ba6e40d 100644
> --- a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
> +++ b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
> @@ -36,13 +36,14 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
> struct i915_ttm_buddy_manager *bman = to_buddy_manager(man);
> struct i915_ttm_buddy_resource *bman_res;
> struct drm_buddy_mm *mm = &bman->mm;
> - unsigned long n_pages;
> - unsigned int min_order;
> + unsigned long n_pages, lpfn;
> u64 min_page_size;
> u64 size;
> int err;
>
> - GEM_BUG_ON(place->fpfn || place->lpfn);
> + lpfn = place->lpfn;
> + if (!lpfn)
> + lpfn = man->size;
>
> bman_res = kzalloc(sizeof(*bman_res), GFP_KERNEL);
> if (!bman_res)
> @@ -52,6 +53,9 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
> INIT_LIST_HEAD(&bman_res->blocks);
> bman_res->mm = mm;
>
> + if (place->fpfn || lpfn != man->size)
> + bman_res->flags |= DRM_BUDDY_RANGE_ALLOCATION;
> +
> GEM_BUG_ON(!bman_res->base.num_pages);
> size = bman_res->base.num_pages << PAGE_SHIFT;
>
> @@ -60,10 +64,16 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
> min_page_size = bo->page_alignment << PAGE_SHIFT;
>
> GEM_BUG_ON(min_page_size < mm->chunk_size);
> - min_order = ilog2(min_page_size) - ilog2(mm->chunk_size);
> +
> if (place->flags & TTM_PL_FLAG_CONTIGUOUS) {
> + unsigned long pages;
> +
> size = roundup_pow_of_two(size);
> - min_order = ilog2(size) - ilog2(mm->chunk_size);
> + min_page_size = size;
> +
> + pages = size >> ilog2(mm->chunk_size);
> + if (pages > lpfn)
> + lpfn = pages;
> }
>
> if (size > mm->size) {
> @@ -73,34 +83,17 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
>
> n_pages = size >> ilog2(mm->chunk_size);
>
> - do {
> - struct drm_buddy_block *block;
> - unsigned int order;
> -
> - order = fls(n_pages) - 1;
> - GEM_BUG_ON(order > mm->max_order);
> - GEM_BUG_ON(order < min_order);
> -
> - do {
> - mutex_lock(&bman->lock);
> - block = drm_buddy_alloc(mm, order);
> - mutex_unlock(&bman->lock);
> - if (!IS_ERR(block))
> - break;
> -
> - if (order-- == min_order) {
> - err = -ENOSPC;
> - goto err_free_blocks;
> - }
> - } while (1);
> -
> - n_pages -= BIT(order);
> -
> - list_add_tail(&block->link, &bman_res->blocks);
> + mutex_lock(&bman->lock);
> + err = drm_buddy_alloc(mm, (uint64_t)place->fpfn << PAGE_SHIFT,
> + (uint64_t)place->lpfn << PAGE_SHIFT,
> + (uint64_t)n_pages << PAGE_SHIFT,
s/uint64_t/u64/
Some other places also. AFAIK that is preferred in the kernel, outside
of maybe uapi headers.
> + min_page_size,
> + &bman_res->blocks,
> + bman_res->flags);
> + mutex_unlock(&bman->lock);
>
> - if (!n_pages)
> - break;
> - } while (1);
> + if (unlikely(err))
> + goto err_free_blocks;
>
> *res = &bman_res->base;
> return 0;
> @@ -266,10 +259,18 @@ int i915_ttm_buddy_man_reserve(struct ttm_resource_manager *man,
> {
> struct i915_ttm_buddy_manager *bman = to_buddy_manager(man);
> struct drm_buddy_mm *mm = &bman->mm;
> + unsigned long flags = 0;
> + u64 min_size;
> int ret;
>
> + min_size = size;
> + flags |= DRM_BUDDY_RANGE_ALLOCATION;
> +
> mutex_lock(&bman->lock);
> - ret = drm_buddy_alloc_range(mm, &bman->reserved, start, size);
> + ret = drm_buddy_alloc(mm, start, start + size,
> + size, min_size,
> + &bman->reserved,
> + flags);
> mutex_unlock(&bman->lock);
>
> return ret;
> diff --git a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.h b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.h
> index fa644b512c2e..5ba490875f66 100644
> --- a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.h
> +++ b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.h
> @@ -20,6 +20,7 @@ struct drm_buddy_mm;
> *
> * @base: struct ttm_resource base class we extend
> * @blocks: the list of struct i915_buddy_block for this resource/allocation
> + * @flags: DRM_BUDDY_*_ALLOCATION flags
> * @mm: the struct i915_buddy_mm for this resource
> *
> * Extends the struct ttm_resource to manage an address space allocation with
> @@ -28,6 +29,7 @@ struct drm_buddy_mm;
> struct i915_ttm_buddy_resource {
> struct ttm_resource base;
> struct list_head blocks;
> + unsigned long flags;
> struct drm_buddy_mm *mm;
> };
>
> diff --git a/include/drm/drm_buddy.h b/include/drm/drm_buddy.h
> index 5ce3fc702f80..c7bb5509a7ad 100644
> --- a/include/drm/drm_buddy.h
> +++ b/include/drm/drm_buddy.h
> @@ -13,15 +13,22 @@
>
> #include <drm/drm_print.h>
>
> -#define range_overflows(start, size, max) ({ \
> +#define check_range_overflow(start, end, size, max) ({ \
> typeof(start) start__ = (start); \
> + typeof(end) end__ = (end);\
> typeof(size) size__ = (size); \
> typeof(max) max__ = (max); \
> (void)(&start__ == &size__); \
> (void)(&start__ == &max__); \
> - start__ >= max__ || size__ > max__ - start__; \
> + (void)(&start__ == &end__); \
> + (void)(&end__ == &size__); \
> + (void)(&end__ == &max__); \
> + start__ >= max__ || end__ > max__ \
> + || size__ > end__ - start__; \
> })
>
> +#define DRM_BUDDY_RANGE_ALLOCATION (1 << 0)
Would it make sense to make this an enum or something?
enum drm_buddy_alloc_mode {
DRM_BUDDY_ALLOC_FIRST = 0,
DRM_BUDDY_ALLOC_TOPDOWN,
DRM_BUDDY_ALLOC_RANGE,
};
That way can more easily see where this is actually used, rather than
having an opaque 'unsigned long flags'? Or are we expecting to combine
allocation strategies or something? Anyway, just a thought.
> +
> struct drm_buddy_block {
> #define DRM_BUDDY_HEADER_OFFSET GENMASK_ULL(63, 12)
> #define DRM_BUDDY_HEADER_STATE GENMASK_ULL(11, 10)
> @@ -131,12 +138,11 @@ int drm_buddy_init(struct drm_buddy_mm *mm, u64 size, u64 chunk_size);
>
> void drm_buddy_fini(struct drm_buddy_mm *mm);
>
> -struct drm_buddy_block *
> -drm_buddy_alloc(struct drm_buddy_mm *mm, unsigned int order);
> -
> -int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
> - struct list_head *blocks,
> - u64 start, u64 size);
> +int drm_buddy_alloc(struct drm_buddy_mm *mm,
> + u64 start, u64 end, u64 size,
> + u64 min_page_size,
> + struct list_head *blocks,
> + unsigned long flags);
>
> void drm_buddy_free(struct drm_buddy_mm *mm, struct drm_buddy_block *block);
>
>
^ permalink raw reply [flat|nested] 58+ messages in thread
* Re: [PATCH 3/8] drm: implement top-down allocation method
2021-10-25 13:00 ` [Intel-gfx] " Arunpravin
(?)
@ 2021-11-03 18:44 ` Matthew Auld
-1 siblings, 0 replies; 58+ messages in thread
From: Matthew Auld @ 2021-11-03 18:44 UTC (permalink / raw)
To: Arunpravin, dri-devel, intel-gfx, amd-gfx
Cc: alexander.deucher, tzimmermann, christian.koenig
On 25/10/2021 14:00, Arunpravin wrote:
> Implemented a function which walk through the order list,
> compares the offset and returns the maximum offset block,
> this method is unpredictable in obtaining the high range
> address blocks which depends on allocation and deallocation.
> for instance, if driver requests address at a low specific
> range, allocator traverses from the root block and splits
> the larger blocks until it reaches the specific block and
> in the process of splitting, lower orders in the freelist
> are occupied with low range address blocks and for the
> subsequent TOPDOWN memory request we may return the low
> range blocks.To overcome this issue, we may go with the
> below approach.
>
> The other approach, sorting each order list entries in
> ascending order and compares the last entry of each
> order list in the freelist and return the max block.
> This creates sorting overhead on every drm_buddy_free()
> request and split up of larger blocks for a single page
> request.
>
> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
> ---
> drivers/gpu/drm/drm_buddy.c | 42 +++++++++++++++++++++++++++++++------
> include/drm/drm_buddy.h | 1 +
> 2 files changed, 37 insertions(+), 6 deletions(-)
>
> diff --git a/drivers/gpu/drm/drm_buddy.c b/drivers/gpu/drm/drm_buddy.c
> index 406e3d521903..9d3547bcc5da 100644
> --- a/drivers/gpu/drm/drm_buddy.c
> +++ b/drivers/gpu/drm/drm_buddy.c
> @@ -362,6 +362,27 @@ alloc_range(struct drm_buddy_mm *mm,
> return ERR_PTR(err);
> }
>
> +static struct drm_buddy_block *
> +get_maxblock(struct list_head *head)
> +{
> + struct drm_buddy_block *max_block = NULL, *node;
> +
> + max_block = list_first_entry_or_null(head,
> + struct drm_buddy_block,
> + link);
> +
> + if (!max_block)
> + return NULL;
> +
> + list_for_each_entry(node, head, link) {
> + if (drm_buddy_block_offset(node) >
> + drm_buddy_block_offset(max_block))
Alignment.
> + max_block = node;
> + }
I suppose there will be pathological cases where this will unnecessarily
steal the mappable portion? But in practice maybe this is good enough?
> +
> + return max_block;
> +}
> +
> static struct drm_buddy_block *
> alloc_from_freelist(struct drm_buddy_mm *mm,
> unsigned int order,
> @@ -372,13 +393,22 @@ alloc_from_freelist(struct drm_buddy_mm *mm,
> int err;
>
> for (i = order; i <= mm->max_order; ++i) {
> - if (!list_empty(&mm->free_list[i])) {
> - block = list_first_entry_or_null(&mm->free_list[i],
> - struct drm_buddy_block,
> - link);
> + if (flags & DRM_BUDDY_TOPDOWN_ALLOCATION) {
> + if (!list_empty(&mm->free_list[i])) {
AFAIK no need to keep checking list_empty(), below also.
> + block = get_maxblock(&mm->free_list[i]);
>
> - if (block)
> - break;
> + if (block)
> + break;
> + }
> + } else {
> + if (!list_empty(&mm->free_list[i])) {
> + block = list_first_entry_or_null(&mm->free_list[i],
> + struct drm_buddy_block,
> + link);
> +
> + if (block)
> + break;
> + }
> }
> }
>
> diff --git a/include/drm/drm_buddy.h b/include/drm/drm_buddy.h
> index c7bb5509a7ad..cd8021d2d6e7 100644
> --- a/include/drm/drm_buddy.h
> +++ b/include/drm/drm_buddy.h
> @@ -28,6 +28,7 @@
> })
>
> #define DRM_BUDDY_RANGE_ALLOCATION (1 << 0)
> +#define DRM_BUDDY_TOPDOWN_ALLOCATION (1 << 1)
>
> struct drm_buddy_block {
> #define DRM_BUDDY_HEADER_OFFSET GENMASK_ULL(63, 12)
>
^ permalink raw reply [flat|nested] 58+ messages in thread
* Re: [Intel-gfx] [PATCH 3/8] drm: implement top-down allocation method
@ 2021-11-03 18:44 ` Matthew Auld
0 siblings, 0 replies; 58+ messages in thread
From: Matthew Auld @ 2021-11-03 18:44 UTC (permalink / raw)
To: Arunpravin, dri-devel, intel-gfx, amd-gfx
Cc: alexander.deucher, tzimmermann, christian.koenig
On 25/10/2021 14:00, Arunpravin wrote:
> Implemented a function which walk through the order list,
> compares the offset and returns the maximum offset block,
> this method is unpredictable in obtaining the high range
> address blocks which depends on allocation and deallocation.
> for instance, if driver requests address at a low specific
> range, allocator traverses from the root block and splits
> the larger blocks until it reaches the specific block and
> in the process of splitting, lower orders in the freelist
> are occupied with low range address blocks and for the
> subsequent TOPDOWN memory request we may return the low
> range blocks.To overcome this issue, we may go with the
> below approach.
>
> The other approach, sorting each order list entries in
> ascending order and compares the last entry of each
> order list in the freelist and return the max block.
> This creates sorting overhead on every drm_buddy_free()
> request and split up of larger blocks for a single page
> request.
>
> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
> ---
> drivers/gpu/drm/drm_buddy.c | 42 +++++++++++++++++++++++++++++++------
> include/drm/drm_buddy.h | 1 +
> 2 files changed, 37 insertions(+), 6 deletions(-)
>
> diff --git a/drivers/gpu/drm/drm_buddy.c b/drivers/gpu/drm/drm_buddy.c
> index 406e3d521903..9d3547bcc5da 100644
> --- a/drivers/gpu/drm/drm_buddy.c
> +++ b/drivers/gpu/drm/drm_buddy.c
> @@ -362,6 +362,27 @@ alloc_range(struct drm_buddy_mm *mm,
> return ERR_PTR(err);
> }
>
> +static struct drm_buddy_block *
> +get_maxblock(struct list_head *head)
> +{
> + struct drm_buddy_block *max_block = NULL, *node;
> +
> + max_block = list_first_entry_or_null(head,
> + struct drm_buddy_block,
> + link);
> +
> + if (!max_block)
> + return NULL;
> +
> + list_for_each_entry(node, head, link) {
> + if (drm_buddy_block_offset(node) >
> + drm_buddy_block_offset(max_block))
Alignment.
> + max_block = node;
> + }
I suppose there will be pathological cases where this will unnecessarily
steal the mappable portion? But in practice maybe this is good enough?
> +
> + return max_block;
> +}
> +
> static struct drm_buddy_block *
> alloc_from_freelist(struct drm_buddy_mm *mm,
> unsigned int order,
> @@ -372,13 +393,22 @@ alloc_from_freelist(struct drm_buddy_mm *mm,
> int err;
>
> for (i = order; i <= mm->max_order; ++i) {
> - if (!list_empty(&mm->free_list[i])) {
> - block = list_first_entry_or_null(&mm->free_list[i],
> - struct drm_buddy_block,
> - link);
> + if (flags & DRM_BUDDY_TOPDOWN_ALLOCATION) {
> + if (!list_empty(&mm->free_list[i])) {
AFAIK no need to keep checking list_empty(), below also.
> + block = get_maxblock(&mm->free_list[i]);
>
> - if (block)
> - break;
> + if (block)
> + break;
> + }
> + } else {
> + if (!list_empty(&mm->free_list[i])) {
> + block = list_first_entry_or_null(&mm->free_list[i],
> + struct drm_buddy_block,
> + link);
> +
> + if (block)
> + break;
> + }
> }
> }
>
> diff --git a/include/drm/drm_buddy.h b/include/drm/drm_buddy.h
> index c7bb5509a7ad..cd8021d2d6e7 100644
> --- a/include/drm/drm_buddy.h
> +++ b/include/drm/drm_buddy.h
> @@ -28,6 +28,7 @@
> })
>
> #define DRM_BUDDY_RANGE_ALLOCATION (1 << 0)
> +#define DRM_BUDDY_TOPDOWN_ALLOCATION (1 << 1)
>
> struct drm_buddy_block {
> #define DRM_BUDDY_HEADER_OFFSET GENMASK_ULL(63, 12)
>
^ permalink raw reply [flat|nested] 58+ messages in thread
* Re: [PATCH 3/8] drm: implement top-down allocation method
@ 2021-11-03 18:44 ` Matthew Auld
0 siblings, 0 replies; 58+ messages in thread
From: Matthew Auld @ 2021-11-03 18:44 UTC (permalink / raw)
To: Arunpravin, dri-devel, intel-gfx, amd-gfx
Cc: alexander.deucher, tzimmermann, jani.nikula, christian.koenig, daniel
On 25/10/2021 14:00, Arunpravin wrote:
> Implemented a function which walk through the order list,
> compares the offset and returns the maximum offset block,
> this method is unpredictable in obtaining the high range
> address blocks which depends on allocation and deallocation.
> for instance, if driver requests address at a low specific
> range, allocator traverses from the root block and splits
> the larger blocks until it reaches the specific block and
> in the process of splitting, lower orders in the freelist
> are occupied with low range address blocks and for the
> subsequent TOPDOWN memory request we may return the low
> range blocks.To overcome this issue, we may go with the
> below approach.
>
> The other approach, sorting each order list entries in
> ascending order and compares the last entry of each
> order list in the freelist and return the max block.
> This creates sorting overhead on every drm_buddy_free()
> request and split up of larger blocks for a single page
> request.
>
> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
> ---
> drivers/gpu/drm/drm_buddy.c | 42 +++++++++++++++++++++++++++++++------
> include/drm/drm_buddy.h | 1 +
> 2 files changed, 37 insertions(+), 6 deletions(-)
>
> diff --git a/drivers/gpu/drm/drm_buddy.c b/drivers/gpu/drm/drm_buddy.c
> index 406e3d521903..9d3547bcc5da 100644
> --- a/drivers/gpu/drm/drm_buddy.c
> +++ b/drivers/gpu/drm/drm_buddy.c
> @@ -362,6 +362,27 @@ alloc_range(struct drm_buddy_mm *mm,
> return ERR_PTR(err);
> }
>
> +static struct drm_buddy_block *
> +get_maxblock(struct list_head *head)
> +{
> + struct drm_buddy_block *max_block = NULL, *node;
> +
> + max_block = list_first_entry_or_null(head,
> + struct drm_buddy_block,
> + link);
> +
> + if (!max_block)
> + return NULL;
> +
> + list_for_each_entry(node, head, link) {
> + if (drm_buddy_block_offset(node) >
> + drm_buddy_block_offset(max_block))
Alignment.
> + max_block = node;
> + }
I suppose there will be pathological cases where this will unnecessarily
steal the mappable portion? But in practice maybe this is good enough?
> +
> + return max_block;
> +}
> +
> static struct drm_buddy_block *
> alloc_from_freelist(struct drm_buddy_mm *mm,
> unsigned int order,
> @@ -372,13 +393,22 @@ alloc_from_freelist(struct drm_buddy_mm *mm,
> int err;
>
> for (i = order; i <= mm->max_order; ++i) {
> - if (!list_empty(&mm->free_list[i])) {
> - block = list_first_entry_or_null(&mm->free_list[i],
> - struct drm_buddy_block,
> - link);
> + if (flags & DRM_BUDDY_TOPDOWN_ALLOCATION) {
> + if (!list_empty(&mm->free_list[i])) {
AFAIK no need to keep checking list_empty(), below also.
> + block = get_maxblock(&mm->free_list[i]);
>
> - if (block)
> - break;
> + if (block)
> + break;
> + }
> + } else {
> + if (!list_empty(&mm->free_list[i])) {
> + block = list_first_entry_or_null(&mm->free_list[i],
> + struct drm_buddy_block,
> + link);
> +
> + if (block)
> + break;
> + }
> }
> }
>
> diff --git a/include/drm/drm_buddy.h b/include/drm/drm_buddy.h
> index c7bb5509a7ad..cd8021d2d6e7 100644
> --- a/include/drm/drm_buddy.h
> +++ b/include/drm/drm_buddy.h
> @@ -28,6 +28,7 @@
> })
>
> #define DRM_BUDDY_RANGE_ALLOCATION (1 << 0)
> +#define DRM_BUDDY_TOPDOWN_ALLOCATION (1 << 1)
>
> struct drm_buddy_block {
> #define DRM_BUDDY_HEADER_OFFSET GENMASK_ULL(63, 12)
>
^ permalink raw reply [flat|nested] 58+ messages in thread
* Re: [PATCH 5/8] drm: Implement method to free unused pages
2021-10-25 13:00 ` [Intel-gfx] " Arunpravin
(?)
@ 2021-11-03 19:16 ` Matthew Auld
-1 siblings, 0 replies; 58+ messages in thread
From: Matthew Auld @ 2021-11-03 19:16 UTC (permalink / raw)
To: Arunpravin, dri-devel, intel-gfx, amd-gfx
Cc: alexander.deucher, tzimmermann, christian.koenig
On 25/10/2021 14:00, Arunpravin wrote:
> On contiguous allocation, we round up the size
> to the *next* power of 2, implement a function
> to free the unused pages after the newly allocate block.
>
> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
Ideally this gets added with some user, so we can see it in action?
Maybe squash the next patch here?
> ---
> drivers/gpu/drm/drm_buddy.c | 103 ++++++++++++++++++++++++++++++++++++
> include/drm/drm_buddy.h | 4 ++
> 2 files changed, 107 insertions(+)
>
> diff --git a/drivers/gpu/drm/drm_buddy.c b/drivers/gpu/drm/drm_buddy.c
> index 9d3547bcc5da..0da8510736eb 100644
> --- a/drivers/gpu/drm/drm_buddy.c
> +++ b/drivers/gpu/drm/drm_buddy.c
> @@ -284,6 +284,109 @@ static inline bool contains(u64 s1, u64 e1, u64 s2, u64 e2)
> return s1 <= s2 && e1 >= e2;
> }
>
> +/**
> + * drm_buddy_free_unused_pages - free unused pages
> + *
> + * @mm: DRM buddy manager
> + * @actual_size: original size requested
> + * @blocks: output list head to add allocated blocks
> + *
> + * For contiguous allocation, we round up the size to the nearest
> + * power of two value, drivers consume *actual* size, so remaining
> + * portions are unused and it can be freed.
> + *
> + * Returns:
> + * 0 on success, error code on failure.
> + */
> +int drm_buddy_free_unused_pages(struct drm_buddy_mm *mm,
drm_buddy_block_trim?
> + u64 actual_size,
new_size?
> + struct list_head *blocks)
> +{
> + struct drm_buddy_block *block;
> + struct drm_buddy_block *buddy;
> + u64 actual_start;
> + u64 actual_end;
> + LIST_HEAD(dfs);
> + u64 count = 0;
> + int err;
> +
> + if (!list_is_singular(blocks))
> + return -EINVAL;
> +
> + block = list_first_entry_or_null(blocks,
> + struct drm_buddy_block,
> + link);
> +
> + if (!block)
> + return -EINVAL;
list_is_singular() already ensures that I guess?
> +
> + if (actual_size > drm_buddy_block_size(mm, block))
> + return -EINVAL;
> +
> + if (actual_size == drm_buddy_block_size(mm, block))
> + return 0;
Probably need to check the alignment of the actual_size, and also check
that it is non-zero?
> +
> + list_del(&block->link);
> +
> + actual_start = drm_buddy_block_offset(block);
> + actual_end = actual_start + actual_size - 1;
> +
> + if (drm_buddy_block_is_allocated(block))
That should rather be a programmer error.
> + mark_free(mm, block);
> +
> + list_add(&block->tmp_link, &dfs);
> +
> + while (1) {
> + block = list_first_entry_or_null(&dfs,
> + struct drm_buddy_block,
> + tmp_link);
> +
> + if (!block)
> + break;
> +
> + list_del(&block->tmp_link);
> +
> + if (count == actual_size)
> + return 0;
Check for overlaps somewhere here to avoid needless searching and splitting?
> +
> + if (contains(actual_start, actual_end, drm_buddy_block_offset(block),
> + (drm_buddy_block_offset(block) + drm_buddy_block_size(mm, block) - 1))) {
Could maybe record the start/end for better readability?
> + BUG_ON(!drm_buddy_block_is_free(block));
> +
> + /* Allocate only required blocks */
> + mark_allocated(block);
> + mm->avail -= drm_buddy_block_size(mm, block);
> + list_add_tail(&block->link, blocks);
> + count += drm_buddy_block_size(mm, block);
> + continue;
> + }
> +
> + if (drm_buddy_block_order(block) == 0)
> + continue;
Should be impossible with overlaps check added.
> +
> + if (!drm_buddy_block_is_split(block)) {
That should always be true.
> + err = split_block(mm, block);
> +
> + if (unlikely(err))
> + goto err_undo;
> + }
> +
> + list_add(&block->right->tmp_link, &dfs);
> + list_add(&block->left->tmp_link, &dfs);
> + }
> +
> + return -ENOSPC;
Would it make sense to factor out part of the alloc_range for this? It
looks roughly the same.
> +
> +err_undo:
> + buddy = get_buddy(block);
> + if (buddy &&
> + (drm_buddy_block_is_free(block) &&
> + drm_buddy_block_is_free(buddy)))
> + __drm_buddy_free(mm, block);
> + return err;
Where do we add the block back to the original list? Did we not just
leak it?
> +}
> +EXPORT_SYMBOL(drm_buddy_free_unused_pages);
> +
> static struct drm_buddy_block *
> alloc_range(struct drm_buddy_mm *mm,
> u64 start, u64 end,
> diff --git a/include/drm/drm_buddy.h b/include/drm/drm_buddy.h
> index cd8021d2d6e7..1dfc80c88e1f 100644
> --- a/include/drm/drm_buddy.h
> +++ b/include/drm/drm_buddy.h
> @@ -145,6 +145,10 @@ int drm_buddy_alloc(struct drm_buddy_mm *mm,
> struct list_head *blocks,
> unsigned long flags);
>
> +int drm_buddy_free_unused_pages(struct drm_buddy_mm *mm,
> + u64 actual_size,
> + struct list_head *blocks);
> +
> void drm_buddy_free(struct drm_buddy_mm *mm, struct drm_buddy_block *block);
>
> void drm_buddy_free_list(struct drm_buddy_mm *mm, struct list_head *objects);
>
^ permalink raw reply [flat|nested] 58+ messages in thread
* Re: [Intel-gfx] [PATCH 5/8] drm: Implement method to free unused pages
@ 2021-11-03 19:16 ` Matthew Auld
0 siblings, 0 replies; 58+ messages in thread
From: Matthew Auld @ 2021-11-03 19:16 UTC (permalink / raw)
To: Arunpravin, dri-devel, intel-gfx, amd-gfx
Cc: alexander.deucher, tzimmermann, christian.koenig
On 25/10/2021 14:00, Arunpravin wrote:
> On contiguous allocation, we round up the size
> to the *next* power of 2, implement a function
> to free the unused pages after the newly allocate block.
>
> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
Ideally this gets added with some user, so we can see it in action?
Maybe squash the next patch here?
> ---
> drivers/gpu/drm/drm_buddy.c | 103 ++++++++++++++++++++++++++++++++++++
> include/drm/drm_buddy.h | 4 ++
> 2 files changed, 107 insertions(+)
>
> diff --git a/drivers/gpu/drm/drm_buddy.c b/drivers/gpu/drm/drm_buddy.c
> index 9d3547bcc5da..0da8510736eb 100644
> --- a/drivers/gpu/drm/drm_buddy.c
> +++ b/drivers/gpu/drm/drm_buddy.c
> @@ -284,6 +284,109 @@ static inline bool contains(u64 s1, u64 e1, u64 s2, u64 e2)
> return s1 <= s2 && e1 >= e2;
> }
>
> +/**
> + * drm_buddy_free_unused_pages - free unused pages
> + *
> + * @mm: DRM buddy manager
> + * @actual_size: original size requested
> + * @blocks: output list head to add allocated blocks
> + *
> + * For contiguous allocation, we round up the size to the nearest
> + * power of two value, drivers consume *actual* size, so remaining
> + * portions are unused and it can be freed.
> + *
> + * Returns:
> + * 0 on success, error code on failure.
> + */
> +int drm_buddy_free_unused_pages(struct drm_buddy_mm *mm,
drm_buddy_block_trim?
> + u64 actual_size,
new_size?
> + struct list_head *blocks)
> +{
> + struct drm_buddy_block *block;
> + struct drm_buddy_block *buddy;
> + u64 actual_start;
> + u64 actual_end;
> + LIST_HEAD(dfs);
> + u64 count = 0;
> + int err;
> +
> + if (!list_is_singular(blocks))
> + return -EINVAL;
> +
> + block = list_first_entry_or_null(blocks,
> + struct drm_buddy_block,
> + link);
> +
> + if (!block)
> + return -EINVAL;
list_is_singular() already ensures that I guess?
> +
> + if (actual_size > drm_buddy_block_size(mm, block))
> + return -EINVAL;
> +
> + if (actual_size == drm_buddy_block_size(mm, block))
> + return 0;
Probably need to check the alignment of the actual_size, and also check
that it is non-zero?
> +
> + list_del(&block->link);
> +
> + actual_start = drm_buddy_block_offset(block);
> + actual_end = actual_start + actual_size - 1;
> +
> + if (drm_buddy_block_is_allocated(block))
That should rather be a programmer error.
> + mark_free(mm, block);
> +
> + list_add(&block->tmp_link, &dfs);
> +
> + while (1) {
> + block = list_first_entry_or_null(&dfs,
> + struct drm_buddy_block,
> + tmp_link);
> +
> + if (!block)
> + break;
> +
> + list_del(&block->tmp_link);
> +
> + if (count == actual_size)
> + return 0;
Check for overlaps somewhere here to avoid needless searching and splitting?
> +
> + if (contains(actual_start, actual_end, drm_buddy_block_offset(block),
> + (drm_buddy_block_offset(block) + drm_buddy_block_size(mm, block) - 1))) {
Could maybe record the start/end for better readability?
> + BUG_ON(!drm_buddy_block_is_free(block));
> +
> + /* Allocate only required blocks */
> + mark_allocated(block);
> + mm->avail -= drm_buddy_block_size(mm, block);
> + list_add_tail(&block->link, blocks);
> + count += drm_buddy_block_size(mm, block);
> + continue;
> + }
> +
> + if (drm_buddy_block_order(block) == 0)
> + continue;
Should be impossible with overlaps check added.
> +
> + if (!drm_buddy_block_is_split(block)) {
That should always be true.
> + err = split_block(mm, block);
> +
> + if (unlikely(err))
> + goto err_undo;
> + }
> +
> + list_add(&block->right->tmp_link, &dfs);
> + list_add(&block->left->tmp_link, &dfs);
> + }
> +
> + return -ENOSPC;
Would it make sense to factor out part of the alloc_range for this? It
looks roughly the same.
> +
> +err_undo:
> + buddy = get_buddy(block);
> + if (buddy &&
> + (drm_buddy_block_is_free(block) &&
> + drm_buddy_block_is_free(buddy)))
> + __drm_buddy_free(mm, block);
> + return err;
Where do we add the block back to the original list? Did we not just
leak it?
> +}
> +EXPORT_SYMBOL(drm_buddy_free_unused_pages);
> +
> static struct drm_buddy_block *
> alloc_range(struct drm_buddy_mm *mm,
> u64 start, u64 end,
> diff --git a/include/drm/drm_buddy.h b/include/drm/drm_buddy.h
> index cd8021d2d6e7..1dfc80c88e1f 100644
> --- a/include/drm/drm_buddy.h
> +++ b/include/drm/drm_buddy.h
> @@ -145,6 +145,10 @@ int drm_buddy_alloc(struct drm_buddy_mm *mm,
> struct list_head *blocks,
> unsigned long flags);
>
> +int drm_buddy_free_unused_pages(struct drm_buddy_mm *mm,
> + u64 actual_size,
> + struct list_head *blocks);
> +
> void drm_buddy_free(struct drm_buddy_mm *mm, struct drm_buddy_block *block);
>
> void drm_buddy_free_list(struct drm_buddy_mm *mm, struct list_head *objects);
>
^ permalink raw reply [flat|nested] 58+ messages in thread
* Re: [PATCH 5/8] drm: Implement method to free unused pages
@ 2021-11-03 19:16 ` Matthew Auld
0 siblings, 0 replies; 58+ messages in thread
From: Matthew Auld @ 2021-11-03 19:16 UTC (permalink / raw)
To: Arunpravin, dri-devel, intel-gfx, amd-gfx
Cc: alexander.deucher, tzimmermann, jani.nikula, christian.koenig, daniel
On 25/10/2021 14:00, Arunpravin wrote:
> On contiguous allocation, we round up the size
> to the *next* power of 2, implement a function
> to free the unused pages after the newly allocate block.
>
> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
Ideally this gets added with some user, so we can see it in action?
Maybe squash the next patch here?
> ---
> drivers/gpu/drm/drm_buddy.c | 103 ++++++++++++++++++++++++++++++++++++
> include/drm/drm_buddy.h | 4 ++
> 2 files changed, 107 insertions(+)
>
> diff --git a/drivers/gpu/drm/drm_buddy.c b/drivers/gpu/drm/drm_buddy.c
> index 9d3547bcc5da..0da8510736eb 100644
> --- a/drivers/gpu/drm/drm_buddy.c
> +++ b/drivers/gpu/drm/drm_buddy.c
> @@ -284,6 +284,109 @@ static inline bool contains(u64 s1, u64 e1, u64 s2, u64 e2)
> return s1 <= s2 && e1 >= e2;
> }
>
> +/**
> + * drm_buddy_free_unused_pages - free unused pages
> + *
> + * @mm: DRM buddy manager
> + * @actual_size: original size requested
> + * @blocks: output list head to add allocated blocks
> + *
> + * For contiguous allocation, we round up the size to the nearest
> + * power of two value, drivers consume *actual* size, so remaining
> + * portions are unused and it can be freed.
> + *
> + * Returns:
> + * 0 on success, error code on failure.
> + */
> +int drm_buddy_free_unused_pages(struct drm_buddy_mm *mm,
drm_buddy_block_trim?
> + u64 actual_size,
new_size?
> + struct list_head *blocks)
> +{
> + struct drm_buddy_block *block;
> + struct drm_buddy_block *buddy;
> + u64 actual_start;
> + u64 actual_end;
> + LIST_HEAD(dfs);
> + u64 count = 0;
> + int err;
> +
> + if (!list_is_singular(blocks))
> + return -EINVAL;
> +
> + block = list_first_entry_or_null(blocks,
> + struct drm_buddy_block,
> + link);
> +
> + if (!block)
> + return -EINVAL;
list_is_singular() already ensures that I guess?
> +
> + if (actual_size > drm_buddy_block_size(mm, block))
> + return -EINVAL;
> +
> + if (actual_size == drm_buddy_block_size(mm, block))
> + return 0;
Probably need to check the alignment of the actual_size, and also check
that it is non-zero?
> +
> + list_del(&block->link);
> +
> + actual_start = drm_buddy_block_offset(block);
> + actual_end = actual_start + actual_size - 1;
> +
> + if (drm_buddy_block_is_allocated(block))
That should rather be a programmer error.
> + mark_free(mm, block);
> +
> + list_add(&block->tmp_link, &dfs);
> +
> + while (1) {
> + block = list_first_entry_or_null(&dfs,
> + struct drm_buddy_block,
> + tmp_link);
> +
> + if (!block)
> + break;
> +
> + list_del(&block->tmp_link);
> +
> + if (count == actual_size)
> + return 0;
Check for overlaps somewhere here to avoid needless searching and splitting?
> +
> + if (contains(actual_start, actual_end, drm_buddy_block_offset(block),
> + (drm_buddy_block_offset(block) + drm_buddy_block_size(mm, block) - 1))) {
Could maybe record the start/end for better readability?
> + BUG_ON(!drm_buddy_block_is_free(block));
> +
> + /* Allocate only required blocks */
> + mark_allocated(block);
> + mm->avail -= drm_buddy_block_size(mm, block);
> + list_add_tail(&block->link, blocks);
> + count += drm_buddy_block_size(mm, block);
> + continue;
> + }
> +
> + if (drm_buddy_block_order(block) == 0)
> + continue;
Should be impossible with overlaps check added.
> +
> + if (!drm_buddy_block_is_split(block)) {
That should always be true.
> + err = split_block(mm, block);
> +
> + if (unlikely(err))
> + goto err_undo;
> + }
> +
> + list_add(&block->right->tmp_link, &dfs);
> + list_add(&block->left->tmp_link, &dfs);
> + }
> +
> + return -ENOSPC;
Would it make sense to factor out part of the alloc_range for this? It
looks roughly the same.
> +
> +err_undo:
> + buddy = get_buddy(block);
> + if (buddy &&
> + (drm_buddy_block_is_free(block) &&
> + drm_buddy_block_is_free(buddy)))
> + __drm_buddy_free(mm, block);
> + return err;
Where do we add the block back to the original list? Did we not just
leak it?
> +}
> +EXPORT_SYMBOL(drm_buddy_free_unused_pages);
> +
> static struct drm_buddy_block *
> alloc_range(struct drm_buddy_mm *mm,
> u64 start, u64 end,
> diff --git a/include/drm/drm_buddy.h b/include/drm/drm_buddy.h
> index cd8021d2d6e7..1dfc80c88e1f 100644
> --- a/include/drm/drm_buddy.h
> +++ b/include/drm/drm_buddy.h
> @@ -145,6 +145,10 @@ int drm_buddy_alloc(struct drm_buddy_mm *mm,
> struct list_head *blocks,
> unsigned long flags);
>
> +int drm_buddy_free_unused_pages(struct drm_buddy_mm *mm,
> + u64 actual_size,
> + struct list_head *blocks);
> +
> void drm_buddy_free(struct drm_buddy_mm *mm, struct drm_buddy_block *block);
>
> void drm_buddy_free_list(struct drm_buddy_mm *mm, struct list_head *objects);
>
^ permalink raw reply [flat|nested] 58+ messages in thread
* Re: [PATCH 6/8] drm/i915: add free_unused_pages support to i915
2021-10-25 13:00 ` [Intel-gfx] " Arunpravin
(?)
@ 2021-11-03 19:18 ` Matthew Auld
-1 siblings, 0 replies; 58+ messages in thread
From: Matthew Auld @ 2021-11-03 19:18 UTC (permalink / raw)
To: Arunpravin, dri-devel, intel-gfx, amd-gfx
Cc: alexander.deucher, tzimmermann, christian.koenig
On 25/10/2021 14:00, Arunpravin wrote:
> add drm_buddy_free_unused_pages() support on
> contiguous allocation
>
> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
> ---
> drivers/gpu/drm/i915/i915_ttm_buddy_manager.c | 8 ++++++++
> 1 file changed, 8 insertions(+)
>
> diff --git a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
> index 963468228392..162947af8e04 100644
> --- a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
> +++ b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
> @@ -98,6 +98,14 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
> if (unlikely(err))
> goto err_free_blocks;
>
> + if (place->flags & TTM_PL_FLAG_CONTIGUOUS) {
> + err = drm_buddy_free_unused_pages(mm, (uint64_t)n_pages << PAGE_SHIFT,
> + &bman_res->blocks);
> +
> + if (unlikely(err))
> + goto err_free_blocks;
That needs some locking, mark_free/mark_split are all globally visible.
Some concurrent user might steal the block, or worse.
> + }
> +
> *res = &bman_res->base;
> return 0;
>
>
^ permalink raw reply [flat|nested] 58+ messages in thread
* Re: [Intel-gfx] [PATCH 6/8] drm/i915: add free_unused_pages support to i915
@ 2021-11-03 19:18 ` Matthew Auld
0 siblings, 0 replies; 58+ messages in thread
From: Matthew Auld @ 2021-11-03 19:18 UTC (permalink / raw)
To: Arunpravin, dri-devel, intel-gfx, amd-gfx
Cc: alexander.deucher, tzimmermann, christian.koenig
On 25/10/2021 14:00, Arunpravin wrote:
> add drm_buddy_free_unused_pages() support on
> contiguous allocation
>
> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
> ---
> drivers/gpu/drm/i915/i915_ttm_buddy_manager.c | 8 ++++++++
> 1 file changed, 8 insertions(+)
>
> diff --git a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
> index 963468228392..162947af8e04 100644
> --- a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
> +++ b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
> @@ -98,6 +98,14 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
> if (unlikely(err))
> goto err_free_blocks;
>
> + if (place->flags & TTM_PL_FLAG_CONTIGUOUS) {
> + err = drm_buddy_free_unused_pages(mm, (uint64_t)n_pages << PAGE_SHIFT,
> + &bman_res->blocks);
> +
> + if (unlikely(err))
> + goto err_free_blocks;
That needs some locking, mark_free/mark_split are all globally visible.
Some concurrent user might steal the block, or worse.
> + }
> +
> *res = &bman_res->base;
> return 0;
>
>
^ permalink raw reply [flat|nested] 58+ messages in thread
* Re: [PATCH 6/8] drm/i915: add free_unused_pages support to i915
@ 2021-11-03 19:18 ` Matthew Auld
0 siblings, 0 replies; 58+ messages in thread
From: Matthew Auld @ 2021-11-03 19:18 UTC (permalink / raw)
To: Arunpravin, dri-devel, intel-gfx, amd-gfx
Cc: alexander.deucher, tzimmermann, jani.nikula, christian.koenig, daniel
On 25/10/2021 14:00, Arunpravin wrote:
> add drm_buddy_free_unused_pages() support on
> contiguous allocation
>
> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
> ---
> drivers/gpu/drm/i915/i915_ttm_buddy_manager.c | 8 ++++++++
> 1 file changed, 8 insertions(+)
>
> diff --git a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
> index 963468228392..162947af8e04 100644
> --- a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
> +++ b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
> @@ -98,6 +98,14 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
> if (unlikely(err))
> goto err_free_blocks;
>
> + if (place->flags & TTM_PL_FLAG_CONTIGUOUS) {
> + err = drm_buddy_free_unused_pages(mm, (uint64_t)n_pages << PAGE_SHIFT,
> + &bman_res->blocks);
> +
> + if (unlikely(err))
> + goto err_free_blocks;
That needs some locking, mark_free/mark_split are all globally visible.
Some concurrent user might steal the block, or worse.
> + }
> +
> *res = &bman_res->base;
> return 0;
>
>
^ permalink raw reply [flat|nested] 58+ messages in thread
* Re: [PATCH 8/8] drm/amdgpu: add drm buddy support to amdgpu
2021-10-25 13:00 ` [Intel-gfx] " Arunpravin
(?)
@ 2021-11-03 19:25 ` Matthew Auld
-1 siblings, 0 replies; 58+ messages in thread
From: Matthew Auld @ 2021-11-03 19:25 UTC (permalink / raw)
To: Arunpravin, dri-devel, intel-gfx, amd-gfx
Cc: alexander.deucher, tzimmermann, christian.koenig
On 25/10/2021 14:00, Arunpravin wrote:
> - Remove drm_mm references and replace with drm buddy functionalities
> - Add res cursor support for drm buddy
>
> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
<snip>
> + spin_lock(&mgr->lock);
> + r = drm_buddy_alloc(mm, (uint64_t)place->fpfn << PAGE_SHIFT,
> + (uint64_t)lpfn << PAGE_SHIFT,
> + (uint64_t)n_pages << PAGE_SHIFT,
> + min_page_size, &node->blocks,
> + node->flags);
Is spinlock + GFP_KERNEL allowed?
> + spin_unlock(&mgr->lock);
> +
> + if (unlikely(r))
> + goto error_free_blocks;
> +
> pages_left -= pages;
> ++i;
>
> if (pages > pages_left)
> pages = pages_left;
> }
> - spin_unlock(&mgr->lock);
> +
> + /* Free unused pages for contiguous allocation */
> + if (place->flags & TTM_PL_FLAG_CONTIGUOUS) {
> + uint64_t actual_size = (uint64_t)node->base.num_pages << PAGE_SHIFT;
> +
> + r = drm_buddy_free_unused_pages(mm,
> + actual_size,
> + &node->blocks);
Needs some locking.
^ permalink raw reply [flat|nested] 58+ messages in thread
* Re: [Intel-gfx] [PATCH 8/8] drm/amdgpu: add drm buddy support to amdgpu
@ 2021-11-03 19:25 ` Matthew Auld
0 siblings, 0 replies; 58+ messages in thread
From: Matthew Auld @ 2021-11-03 19:25 UTC (permalink / raw)
To: Arunpravin, dri-devel, intel-gfx, amd-gfx
Cc: alexander.deucher, tzimmermann, christian.koenig
On 25/10/2021 14:00, Arunpravin wrote:
> - Remove drm_mm references and replace with drm buddy functionalities
> - Add res cursor support for drm buddy
>
> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
<snip>
> + spin_lock(&mgr->lock);
> + r = drm_buddy_alloc(mm, (uint64_t)place->fpfn << PAGE_SHIFT,
> + (uint64_t)lpfn << PAGE_SHIFT,
> + (uint64_t)n_pages << PAGE_SHIFT,
> + min_page_size, &node->blocks,
> + node->flags);
Is spinlock + GFP_KERNEL allowed?
> + spin_unlock(&mgr->lock);
> +
> + if (unlikely(r))
> + goto error_free_blocks;
> +
> pages_left -= pages;
> ++i;
>
> if (pages > pages_left)
> pages = pages_left;
> }
> - spin_unlock(&mgr->lock);
> +
> + /* Free unused pages for contiguous allocation */
> + if (place->flags & TTM_PL_FLAG_CONTIGUOUS) {
> + uint64_t actual_size = (uint64_t)node->base.num_pages << PAGE_SHIFT;
> +
> + r = drm_buddy_free_unused_pages(mm,
> + actual_size,
> + &node->blocks);
Needs some locking.
^ permalink raw reply [flat|nested] 58+ messages in thread
* Re: [PATCH 8/8] drm/amdgpu: add drm buddy support to amdgpu
@ 2021-11-03 19:25 ` Matthew Auld
0 siblings, 0 replies; 58+ messages in thread
From: Matthew Auld @ 2021-11-03 19:25 UTC (permalink / raw)
To: Arunpravin, dri-devel, intel-gfx, amd-gfx
Cc: alexander.deucher, tzimmermann, jani.nikula, christian.koenig, daniel
On 25/10/2021 14:00, Arunpravin wrote:
> - Remove drm_mm references and replace with drm buddy functionalities
> - Add res cursor support for drm buddy
>
> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
<snip>
> + spin_lock(&mgr->lock);
> + r = drm_buddy_alloc(mm, (uint64_t)place->fpfn << PAGE_SHIFT,
> + (uint64_t)lpfn << PAGE_SHIFT,
> + (uint64_t)n_pages << PAGE_SHIFT,
> + min_page_size, &node->blocks,
> + node->flags);
Is spinlock + GFP_KERNEL allowed?
> + spin_unlock(&mgr->lock);
> +
> + if (unlikely(r))
> + goto error_free_blocks;
> +
> pages_left -= pages;
> ++i;
>
> if (pages > pages_left)
> pages = pages_left;
> }
> - spin_unlock(&mgr->lock);
> +
> + /* Free unused pages for contiguous allocation */
> + if (place->flags & TTM_PL_FLAG_CONTIGUOUS) {
> + uint64_t actual_size = (uint64_t)node->base.num_pages << PAGE_SHIFT;
> +
> + r = drm_buddy_free_unused_pages(mm,
> + actual_size,
> + &node->blocks);
Needs some locking.
^ permalink raw reply [flat|nested] 58+ messages in thread
* Re: [PATCH 8/8] drm/amdgpu: add drm buddy support to amdgpu
2021-11-03 19:25 ` [Intel-gfx] " Matthew Auld
(?)
@ 2021-11-04 7:34 ` Christian König
-1 siblings, 0 replies; 58+ messages in thread
From: Christian König @ 2021-11-04 7:34 UTC (permalink / raw)
To: Matthew Auld, Arunpravin, dri-devel, intel-gfx, amd-gfx
Cc: alexander.deucher, tzimmermann
Am 03.11.21 um 20:25 schrieb Matthew Auld:
> On 25/10/2021 14:00, Arunpravin wrote:
>> - Remove drm_mm references and replace with drm buddy functionalities
>> - Add res cursor support for drm buddy
>>
>> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
>
> <snip>
>
>> + spin_lock(&mgr->lock);
>> + r = drm_buddy_alloc(mm, (uint64_t)place->fpfn << PAGE_SHIFT,
>> + (uint64_t)lpfn << PAGE_SHIFT,
>> + (uint64_t)n_pages << PAGE_SHIFT,
>> + min_page_size, &node->blocks,
>> + node->flags);
>
>
> Is spinlock + GFP_KERNEL allowed?
Nope it isn't, but does that function really calls kmalloc()?
Christian.
>
>> + spin_unlock(&mgr->lock);
>> +
>> + if (unlikely(r))
>> + goto error_free_blocks;
>> +
>> pages_left -= pages;
>> ++i;
>> if (pages > pages_left)
>> pages = pages_left;
>> }
>> - spin_unlock(&mgr->lock);
>> +
>> + /* Free unused pages for contiguous allocation */
>> + if (place->flags & TTM_PL_FLAG_CONTIGUOUS) {
>> + uint64_t actual_size = (uint64_t)node->base.num_pages <<
>> PAGE_SHIFT;
>> +
>> + r = drm_buddy_free_unused_pages(mm,
>> + actual_size,
>> + &node->blocks);
>
> Needs some locking.
^ permalink raw reply [flat|nested] 58+ messages in thread
* Re: [Intel-gfx] [PATCH 8/8] drm/amdgpu: add drm buddy support to amdgpu
@ 2021-11-04 7:34 ` Christian König
0 siblings, 0 replies; 58+ messages in thread
From: Christian König @ 2021-11-04 7:34 UTC (permalink / raw)
To: Matthew Auld, Arunpravin, dri-devel, intel-gfx, amd-gfx
Cc: alexander.deucher, tzimmermann
Am 03.11.21 um 20:25 schrieb Matthew Auld:
> On 25/10/2021 14:00, Arunpravin wrote:
>> - Remove drm_mm references and replace with drm buddy functionalities
>> - Add res cursor support for drm buddy
>>
>> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
>
> <snip>
>
>> + spin_lock(&mgr->lock);
>> + r = drm_buddy_alloc(mm, (uint64_t)place->fpfn << PAGE_SHIFT,
>> + (uint64_t)lpfn << PAGE_SHIFT,
>> + (uint64_t)n_pages << PAGE_SHIFT,
>> + min_page_size, &node->blocks,
>> + node->flags);
>
>
> Is spinlock + GFP_KERNEL allowed?
Nope it isn't, but does that function really calls kmalloc()?
Christian.
>
>> + spin_unlock(&mgr->lock);
>> +
>> + if (unlikely(r))
>> + goto error_free_blocks;
>> +
>> pages_left -= pages;
>> ++i;
>> if (pages > pages_left)
>> pages = pages_left;
>> }
>> - spin_unlock(&mgr->lock);
>> +
>> + /* Free unused pages for contiguous allocation */
>> + if (place->flags & TTM_PL_FLAG_CONTIGUOUS) {
>> + uint64_t actual_size = (uint64_t)node->base.num_pages <<
>> PAGE_SHIFT;
>> +
>> + r = drm_buddy_free_unused_pages(mm,
>> + actual_size,
>> + &node->blocks);
>
> Needs some locking.
^ permalink raw reply [flat|nested] 58+ messages in thread
* Re: [PATCH 8/8] drm/amdgpu: add drm buddy support to amdgpu
@ 2021-11-04 7:34 ` Christian König
0 siblings, 0 replies; 58+ messages in thread
From: Christian König @ 2021-11-04 7:34 UTC (permalink / raw)
To: Matthew Auld, Arunpravin, dri-devel, intel-gfx, amd-gfx
Cc: alexander.deucher, jani.nikula, tzimmermann, daniel
Am 03.11.21 um 20:25 schrieb Matthew Auld:
> On 25/10/2021 14:00, Arunpravin wrote:
>> - Remove drm_mm references and replace with drm buddy functionalities
>> - Add res cursor support for drm buddy
>>
>> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
>
> <snip>
>
>> + spin_lock(&mgr->lock);
>> + r = drm_buddy_alloc(mm, (uint64_t)place->fpfn << PAGE_SHIFT,
>> + (uint64_t)lpfn << PAGE_SHIFT,
>> + (uint64_t)n_pages << PAGE_SHIFT,
>> + min_page_size, &node->blocks,
>> + node->flags);
>
>
> Is spinlock + GFP_KERNEL allowed?
Nope it isn't, but does that function really calls kmalloc()?
Christian.
>
>> + spin_unlock(&mgr->lock);
>> +
>> + if (unlikely(r))
>> + goto error_free_blocks;
>> +
>> pages_left -= pages;
>> ++i;
>> if (pages > pages_left)
>> pages = pages_left;
>> }
>> - spin_unlock(&mgr->lock);
>> +
>> + /* Free unused pages for contiguous allocation */
>> + if (place->flags & TTM_PL_FLAG_CONTIGUOUS) {
>> + uint64_t actual_size = (uint64_t)node->base.num_pages <<
>> PAGE_SHIFT;
>> +
>> + r = drm_buddy_free_unused_pages(mm,
>> + actual_size,
>> + &node->blocks);
>
> Needs some locking.
^ permalink raw reply [flat|nested] 58+ messages in thread
* Re: [PATCH 8/8] drm/amdgpu: add drm buddy support to amdgpu
2021-11-04 7:34 ` [Intel-gfx] " Christian König
(?)
@ 2021-11-04 8:49 ` Matthew Auld
-1 siblings, 0 replies; 58+ messages in thread
From: Matthew Auld @ 2021-11-04 8:49 UTC (permalink / raw)
To: Christian König, Arunpravin, dri-devel, intel-gfx, amd-gfx
Cc: alexander.deucher, tzimmermann
On 04/11/2021 07:34, Christian König wrote:
>
>
> Am 03.11.21 um 20:25 schrieb Matthew Auld:
>> On 25/10/2021 14:00, Arunpravin wrote:
>>> - Remove drm_mm references and replace with drm buddy functionalities
>>> - Add res cursor support for drm buddy
>>>
>>> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
>>
>> <snip>
>>
>>> + spin_lock(&mgr->lock);
>>> + r = drm_buddy_alloc(mm, (uint64_t)place->fpfn << PAGE_SHIFT,
>>> + (uint64_t)lpfn << PAGE_SHIFT,
>>> + (uint64_t)n_pages << PAGE_SHIFT,
>>> + min_page_size, &node->blocks,
>>> + node->flags);
>>
>>
>> Is spinlock + GFP_KERNEL allowed?
>
> Nope it isn't, but does that function really calls kmalloc()?
It calls kmem_cache_zalloc(..., GFP_KERNEL)
>
> Christian.
>
>>
>>> + spin_unlock(&mgr->lock);
>>> +
>>> + if (unlikely(r))
>>> + goto error_free_blocks;
>>> +
>>> pages_left -= pages;
>>> ++i;
>>> if (pages > pages_left)
>>> pages = pages_left;
>>> }
>>> - spin_unlock(&mgr->lock);
>>> +
>>> + /* Free unused pages for contiguous allocation */
>>> + if (place->flags & TTM_PL_FLAG_CONTIGUOUS) {
>>> + uint64_t actual_size = (uint64_t)node->base.num_pages <<
>>> PAGE_SHIFT;
>>> +
>>> + r = drm_buddy_free_unused_pages(mm,
>>> + actual_size,
>>> + &node->blocks);
>>
>> Needs some locking.
>
^ permalink raw reply [flat|nested] 58+ messages in thread
* Re: [Intel-gfx] [PATCH 8/8] drm/amdgpu: add drm buddy support to amdgpu
@ 2021-11-04 8:49 ` Matthew Auld
0 siblings, 0 replies; 58+ messages in thread
From: Matthew Auld @ 2021-11-04 8:49 UTC (permalink / raw)
To: Christian König, Arunpravin, dri-devel, intel-gfx, amd-gfx
Cc: alexander.deucher, tzimmermann
On 04/11/2021 07:34, Christian König wrote:
>
>
> Am 03.11.21 um 20:25 schrieb Matthew Auld:
>> On 25/10/2021 14:00, Arunpravin wrote:
>>> - Remove drm_mm references and replace with drm buddy functionalities
>>> - Add res cursor support for drm buddy
>>>
>>> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
>>
>> <snip>
>>
>>> + spin_lock(&mgr->lock);
>>> + r = drm_buddy_alloc(mm, (uint64_t)place->fpfn << PAGE_SHIFT,
>>> + (uint64_t)lpfn << PAGE_SHIFT,
>>> + (uint64_t)n_pages << PAGE_SHIFT,
>>> + min_page_size, &node->blocks,
>>> + node->flags);
>>
>>
>> Is spinlock + GFP_KERNEL allowed?
>
> Nope it isn't, but does that function really calls kmalloc()?
It calls kmem_cache_zalloc(..., GFP_KERNEL)
>
> Christian.
>
>>
>>> + spin_unlock(&mgr->lock);
>>> +
>>> + if (unlikely(r))
>>> + goto error_free_blocks;
>>> +
>>> pages_left -= pages;
>>> ++i;
>>> if (pages > pages_left)
>>> pages = pages_left;
>>> }
>>> - spin_unlock(&mgr->lock);
>>> +
>>> + /* Free unused pages for contiguous allocation */
>>> + if (place->flags & TTM_PL_FLAG_CONTIGUOUS) {
>>> + uint64_t actual_size = (uint64_t)node->base.num_pages <<
>>> PAGE_SHIFT;
>>> +
>>> + r = drm_buddy_free_unused_pages(mm,
>>> + actual_size,
>>> + &node->blocks);
>>
>> Needs some locking.
>
^ permalink raw reply [flat|nested] 58+ messages in thread
* Re: [PATCH 8/8] drm/amdgpu: add drm buddy support to amdgpu
@ 2021-11-04 8:49 ` Matthew Auld
0 siblings, 0 replies; 58+ messages in thread
From: Matthew Auld @ 2021-11-04 8:49 UTC (permalink / raw)
To: Christian König, Arunpravin, dri-devel, intel-gfx, amd-gfx
Cc: alexander.deucher, jani.nikula, tzimmermann, daniel
On 04/11/2021 07:34, Christian König wrote:
>
>
> Am 03.11.21 um 20:25 schrieb Matthew Auld:
>> On 25/10/2021 14:00, Arunpravin wrote:
>>> - Remove drm_mm references and replace with drm buddy functionalities
>>> - Add res cursor support for drm buddy
>>>
>>> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
>>
>> <snip>
>>
>>> + spin_lock(&mgr->lock);
>>> + r = drm_buddy_alloc(mm, (uint64_t)place->fpfn << PAGE_SHIFT,
>>> + (uint64_t)lpfn << PAGE_SHIFT,
>>> + (uint64_t)n_pages << PAGE_SHIFT,
>>> + min_page_size, &node->blocks,
>>> + node->flags);
>>
>>
>> Is spinlock + GFP_KERNEL allowed?
>
> Nope it isn't, but does that function really calls kmalloc()?
It calls kmem_cache_zalloc(..., GFP_KERNEL)
>
> Christian.
>
>>
>>> + spin_unlock(&mgr->lock);
>>> +
>>> + if (unlikely(r))
>>> + goto error_free_blocks;
>>> +
>>> pages_left -= pages;
>>> ++i;
>>> if (pages > pages_left)
>>> pages = pages_left;
>>> }
>>> - spin_unlock(&mgr->lock);
>>> +
>>> + /* Free unused pages for contiguous allocation */
>>> + if (place->flags & TTM_PL_FLAG_CONTIGUOUS) {
>>> + uint64_t actual_size = (uint64_t)node->base.num_pages <<
>>> PAGE_SHIFT;
>>> +
>>> + r = drm_buddy_free_unused_pages(mm,
>>> + actual_size,
>>> + &node->blocks);
>>
>> Needs some locking.
>
^ permalink raw reply [flat|nested] 58+ messages in thread
* Re: [PATCH 8/8] drm/amdgpu: add drm buddy support to amdgpu
2021-11-04 8:49 ` [Intel-gfx] " Matthew Auld
(?)
@ 2021-11-04 8:54 ` Christian König
-1 siblings, 0 replies; 58+ messages in thread
From: Christian König @ 2021-11-04 8:54 UTC (permalink / raw)
To: Matthew Auld, Arunpravin, dri-devel, intel-gfx, amd-gfx
Cc: alexander.deucher, tzimmermann
Am 04.11.21 um 09:49 schrieb Matthew Auld:
> On 04/11/2021 07:34, Christian König wrote:
>>
>>
>> Am 03.11.21 um 20:25 schrieb Matthew Auld:
>>> On 25/10/2021 14:00, Arunpravin wrote:
>>>> - Remove drm_mm references and replace with drm buddy functionalities
>>>> - Add res cursor support for drm buddy
>>>>
>>>> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
>>>
>>> <snip>
>>>
>>>> + spin_lock(&mgr->lock);
>>>> + r = drm_buddy_alloc(mm, (uint64_t)place->fpfn << PAGE_SHIFT,
>>>> + (uint64_t)lpfn << PAGE_SHIFT,
>>>> + (uint64_t)n_pages << PAGE_SHIFT,
>>>> + min_page_size, &node->blocks,
>>>> + node->flags);
>>>
>>>
>>> Is spinlock + GFP_KERNEL allowed?
>>
>> Nope it isn't, but does that function really calls kmalloc()?
>
> It calls kmem_cache_zalloc(..., GFP_KERNEL)
Oh that's bad. In this case we either need a mutex here or some other
approach to avoid the allocation.
Christian.
>
>>
>> Christian.
>>
>>>
>>>> + spin_unlock(&mgr->lock);
>>>> +
>>>> + if (unlikely(r))
>>>> + goto error_free_blocks;
>>>> +
>>>> pages_left -= pages;
>>>> ++i;
>>>> if (pages > pages_left)
>>>> pages = pages_left;
>>>> }
>>>> - spin_unlock(&mgr->lock);
>>>> +
>>>> + /* Free unused pages for contiguous allocation */
>>>> + if (place->flags & TTM_PL_FLAG_CONTIGUOUS) {
>>>> + uint64_t actual_size = (uint64_t)node->base.num_pages <<
>>>> PAGE_SHIFT;
>>>> +
>>>> + r = drm_buddy_free_unused_pages(mm,
>>>> + actual_size,
>>>> + &node->blocks);
>>>
>>> Needs some locking.
>>
^ permalink raw reply [flat|nested] 58+ messages in thread
* Re: [Intel-gfx] [PATCH 8/8] drm/amdgpu: add drm buddy support to amdgpu
@ 2021-11-04 8:54 ` Christian König
0 siblings, 0 replies; 58+ messages in thread
From: Christian König @ 2021-11-04 8:54 UTC (permalink / raw)
To: Matthew Auld, Arunpravin, dri-devel, intel-gfx, amd-gfx
Cc: alexander.deucher, tzimmermann
Am 04.11.21 um 09:49 schrieb Matthew Auld:
> On 04/11/2021 07:34, Christian König wrote:
>>
>>
>> Am 03.11.21 um 20:25 schrieb Matthew Auld:
>>> On 25/10/2021 14:00, Arunpravin wrote:
>>>> - Remove drm_mm references and replace with drm buddy functionalities
>>>> - Add res cursor support for drm buddy
>>>>
>>>> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
>>>
>>> <snip>
>>>
>>>> + spin_lock(&mgr->lock);
>>>> + r = drm_buddy_alloc(mm, (uint64_t)place->fpfn << PAGE_SHIFT,
>>>> + (uint64_t)lpfn << PAGE_SHIFT,
>>>> + (uint64_t)n_pages << PAGE_SHIFT,
>>>> + min_page_size, &node->blocks,
>>>> + node->flags);
>>>
>>>
>>> Is spinlock + GFP_KERNEL allowed?
>>
>> Nope it isn't, but does that function really calls kmalloc()?
>
> It calls kmem_cache_zalloc(..., GFP_KERNEL)
Oh that's bad. In this case we either need a mutex here or some other
approach to avoid the allocation.
Christian.
>
>>
>> Christian.
>>
>>>
>>>> + spin_unlock(&mgr->lock);
>>>> +
>>>> + if (unlikely(r))
>>>> + goto error_free_blocks;
>>>> +
>>>> pages_left -= pages;
>>>> ++i;
>>>> if (pages > pages_left)
>>>> pages = pages_left;
>>>> }
>>>> - spin_unlock(&mgr->lock);
>>>> +
>>>> + /* Free unused pages for contiguous allocation */
>>>> + if (place->flags & TTM_PL_FLAG_CONTIGUOUS) {
>>>> + uint64_t actual_size = (uint64_t)node->base.num_pages <<
>>>> PAGE_SHIFT;
>>>> +
>>>> + r = drm_buddy_free_unused_pages(mm,
>>>> + actual_size,
>>>> + &node->blocks);
>>>
>>> Needs some locking.
>>
^ permalink raw reply [flat|nested] 58+ messages in thread
* Re: [PATCH 8/8] drm/amdgpu: add drm buddy support to amdgpu
@ 2021-11-04 8:54 ` Christian König
0 siblings, 0 replies; 58+ messages in thread
From: Christian König @ 2021-11-04 8:54 UTC (permalink / raw)
To: Matthew Auld, Arunpravin, dri-devel, intel-gfx, amd-gfx
Cc: alexander.deucher, jani.nikula, tzimmermann, daniel
Am 04.11.21 um 09:49 schrieb Matthew Auld:
> On 04/11/2021 07:34, Christian König wrote:
>>
>>
>> Am 03.11.21 um 20:25 schrieb Matthew Auld:
>>> On 25/10/2021 14:00, Arunpravin wrote:
>>>> - Remove drm_mm references and replace with drm buddy functionalities
>>>> - Add res cursor support for drm buddy
>>>>
>>>> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
>>>
>>> <snip>
>>>
>>>> + spin_lock(&mgr->lock);
>>>> + r = drm_buddy_alloc(mm, (uint64_t)place->fpfn << PAGE_SHIFT,
>>>> + (uint64_t)lpfn << PAGE_SHIFT,
>>>> + (uint64_t)n_pages << PAGE_SHIFT,
>>>> + min_page_size, &node->blocks,
>>>> + node->flags);
>>>
>>>
>>> Is spinlock + GFP_KERNEL allowed?
>>
>> Nope it isn't, but does that function really calls kmalloc()?
>
> It calls kmem_cache_zalloc(..., GFP_KERNEL)
Oh that's bad. In this case we either need a mutex here or some other
approach to avoid the allocation.
Christian.
>
>>
>> Christian.
>>
>>>
>>>> + spin_unlock(&mgr->lock);
>>>> +
>>>> + if (unlikely(r))
>>>> + goto error_free_blocks;
>>>> +
>>>> pages_left -= pages;
>>>> ++i;
>>>> if (pages > pages_left)
>>>> pages = pages_left;
>>>> }
>>>> - spin_unlock(&mgr->lock);
>>>> +
>>>> + /* Free unused pages for contiguous allocation */
>>>> + if (place->flags & TTM_PL_FLAG_CONTIGUOUS) {
>>>> + uint64_t actual_size = (uint64_t)node->base.num_pages <<
>>>> PAGE_SHIFT;
>>>> +
>>>> + r = drm_buddy_free_unused_pages(mm,
>>>> + actual_size,
>>>> + &node->blocks);
>>>
>>> Needs some locking.
>>
^ permalink raw reply [flat|nested] 58+ messages in thread
* Re: [PATCH v2 2/8] drm: improve drm_buddy_alloc function
2021-11-03 18:41 ` [Intel-gfx] " Matthew Auld
(?)
@ 2021-11-08 19:50 ` Arunpravin
-1 siblings, 0 replies; 58+ messages in thread
From: Arunpravin @ 2021-11-08 19:50 UTC (permalink / raw)
To: Matthew Auld, dri-devel, intel-gfx, amd-gfx
Cc: alexander.deucher, tzimmermann, christian.koenig
Hi Matthew,
Thanks for the review, Please find my comments
On 04/11/21 12:11 am, Matthew Auld wrote:
> On 25/10/2021 14:00, Arunpravin wrote:
>> - Make drm_buddy_alloc a single function to handle
>> range allocation and non-range allocation demands
>>
>> - Implemented a new function alloc_range() which allocates
>> the requested power-of-two block comply with range limitations
>>
>> - Moved order computation and memory alignment logic from
>> i915 driver to drm buddy
>>
>> V2:
>> merged below changes to keep the build unbroken
>> - drm_buddy_alloc_range() becomes obsolete and may be removed
>> - enable ttm range allocation (fpfn / lpfn) support in i915 driver
>> - apply enhanced drm_buddy_alloc() function to i915 driver
>>
>> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
>> ---
>> drivers/gpu/drm/drm_buddy.c | 265 +++++++++++-------
>> drivers/gpu/drm/i915/i915_ttm_buddy_manager.c | 67 ++---
>> drivers/gpu/drm/i915/i915_ttm_buddy_manager.h | 2 +
>> include/drm/drm_buddy.h | 22 +-
>> 4 files changed, 207 insertions(+), 149 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/drm_buddy.c b/drivers/gpu/drm/drm_buddy.c
>> index 39eb1d224bec..406e3d521903 100644
>> --- a/drivers/gpu/drm/drm_buddy.c
>> +++ b/drivers/gpu/drm/drm_buddy.c
>> @@ -274,63 +274,6 @@ void drm_buddy_free_list(struct drm_buddy_mm *mm, struct list_head *objects)
>> }
>> EXPORT_SYMBOL(drm_buddy_free_list);
>>
>> -/**
>> - * drm_buddy_alloc - allocate power-of-two blocks
>> - *
>> - * @mm: DRM buddy manager to allocate from
>> - * @order: size of the allocation
>> - *
>> - * The order value here translates to:
>> - *
>> - * 0 = 2^0 * mm->chunk_size
>> - * 1 = 2^1 * mm->chunk_size
>> - * 2 = 2^2 * mm->chunk_size
>> - *
>> - * Returns:
>> - * allocated ptr to the &drm_buddy_block on success
>> - */
>> -struct drm_buddy_block *
>> -drm_buddy_alloc(struct drm_buddy_mm *mm, unsigned int order)
>> -{
>> - struct drm_buddy_block *block = NULL;
>> - unsigned int i;
>> - int err;
>> -
>> - for (i = order; i <= mm->max_order; ++i) {
>> - block = list_first_entry_or_null(&mm->free_list[i],
>> - struct drm_buddy_block,
>> - link);
>> - if (block)
>> - break;
>> - }
>> -
>> - if (!block)
>> - return ERR_PTR(-ENOSPC);
>> -
>> - BUG_ON(!drm_buddy_block_is_free(block));
>> -
>> - while (i != order) {
>> - err = split_block(mm, block);
>> - if (unlikely(err))
>> - goto out_free;
>> -
>> - /* Go low */
>> - block = block->left;
>> - i--;
>> - }
>> -
>> - mark_allocated(block);
>> - mm->avail -= drm_buddy_block_size(mm, block);
>> - kmemleak_update_trace(block);
>> - return block;
>> -
>> -out_free:
>> - if (i != order)
>> - __drm_buddy_free(mm, block);
>> - return ERR_PTR(err);
>> -}
>> -EXPORT_SYMBOL(drm_buddy_alloc);
>> -
>> static inline bool overlaps(u64 s1, u64 e1, u64 s2, u64 e2)
>> {
>> return s1 <= e2 && e1 >= s2;
>> @@ -341,52 +284,22 @@ static inline bool contains(u64 s1, u64 e1, u64 s2, u64 e2)
>> return s1 <= s2 && e1 >= e2;
>> }
>>
>> -/**
>> - * drm_buddy_alloc_range - allocate range
>> - *
>> - * @mm: DRM buddy manager to allocate from
>> - * @blocks: output list head to add allocated blocks
>> - * @start: start of the allowed range for this block
>> - * @size: size of the allocation
>> - *
>> - * Intended for pre-allocating portions of the address space, for example to
>> - * reserve a block for the initial framebuffer or similar, hence the expectation
>> - * here is that drm_buddy_alloc() is still the main vehicle for
>> - * allocations, so if that's not the case then the drm_mm range allocator is
>> - * probably a much better fit, and so you should probably go use that instead.
>> - *
>> - * Note that it's safe to chain together multiple alloc_ranges
>> - * with the same blocks list
>> - *
>> - * Returns:
>> - * 0 on success, error code on failure.
>> - */
>> -int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
>> - struct list_head *blocks,
>> - u64 start, u64 size)
>> +static struct drm_buddy_block *
>> +alloc_range(struct drm_buddy_mm *mm,
>> + u64 start, u64 end,
>> + unsigned int order)
>> {
>> struct drm_buddy_block *block;
>> struct drm_buddy_block *buddy;
>> - LIST_HEAD(allocated);
>> LIST_HEAD(dfs);
>> - u64 end;
>> int err;
>> int i;
>>
>> - if (size < mm->chunk_size)
>> - return -EINVAL;
>> -
>> - if (!IS_ALIGNED(size | start, mm->chunk_size))
>> - return -EINVAL;
>> -
>> - if (range_overflows(start, size, mm->size))
>> - return -EINVAL;
>> + end = end - 1;
>>
>> for (i = 0; i < mm->n_roots; ++i)
>> list_add_tail(&mm->roots[i]->tmp_link, &dfs);
>>
>> - end = start + size - 1;
>> -
>> do {
>> u64 block_start;
>> u64 block_end;
>> @@ -394,31 +307,32 @@ int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
>> block = list_first_entry_or_null(&dfs,
>> struct drm_buddy_block,
>> tmp_link);
>> +
>
> No need for the newline.
[Arun] ok
>
>> if (!block)
>> break;
>>
>> list_del(&block->tmp_link);
>>
>> + if (drm_buddy_block_order(block) < order)
>> + continue;
>> +
>> block_start = drm_buddy_block_offset(block);
>> block_end = block_start + drm_buddy_block_size(mm, block) - 1;
>>
>> if (!overlaps(start, end, block_start, block_end))
>> continue;
>>
>> - if (drm_buddy_block_is_allocated(block)) {
>> - err = -ENOSPC;
>> - goto err_free;
>> - }
>> + if (drm_buddy_block_is_allocated(block))
>> + continue;
>>
>> - if (contains(start, end, block_start, block_end)) {
>> - if (!drm_buddy_block_is_free(block)) {
>> - err = -ENOSPC;
>> - goto err_free;
>> - }
>> + if (contains(start, end, block_start, block_end)
>> + && order == drm_buddy_block_order(block)) {
>
> Alignment looks off, also && should be on the line above.
[Arun] ok
>
>> + /*
>> + * Find the free block within the range.
>> + */
>> + if (drm_buddy_block_is_free(block))
>> + return block;
>
> Would it make sense to keep searching here, rather than restarting the
> search from scratch every time? Would it work if we pass in the total
> size and min order?
[Arun] yes, I will rewrite this function
>
> I take it this will now also be used for allocating CPU mappable memory
> using an end bias?
[Arun] yes, this function used for allocating CPU mappable memory using
an end bias and for actual range allocation as well.
>
>>
>> - mark_allocated(block);
>> - mm->avail -= drm_buddy_block_size(mm, block);
>> - list_add_tail(&block->link, &allocated);
>> continue;
>> }
>>
>> @@ -428,12 +342,11 @@ int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
>> goto err_undo;
>> }
>>
>> - list_add(&block->right->tmp_link, &dfs);
>> list_add(&block->left->tmp_link, &dfs);
>> + list_add(&block->right->tmp_link, &dfs);
>
> So this is now top-down? Do we need this? Also I guess this now changes
> the ordering of the allocated blocks? Ideally they would appear in
> order, starting from the lowest address(at least for the actual range
> allocation, and not just the bias thing). I think some places might only
> check the first block to get the device address, where underneath we
> have a contiguous allocation.
[Arun] True, I will keep bottom-up
>
>
>> } while (1);
>>
>> - list_splice_tail(&allocated, blocks);
>> - return 0;
>> + return ERR_PTR(-ENOSPC);
>>
>> err_undo:
>> /*
>> @@ -446,12 +359,148 @@ int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
>> (drm_buddy_block_is_free(block) &&
>> drm_buddy_block_is_free(buddy)))
>> __drm_buddy_free(mm, block);
>> + return ERR_PTR(err);
>> +}
>> +
>> +static struct drm_buddy_block *
>> +alloc_from_freelist(struct drm_buddy_mm *mm,
>> + unsigned int order,
>> + unsigned long flags)
>> +{
>> + struct drm_buddy_block *block = NULL;
>> + unsigned int i;
>> + int err;
>> +
>> + for (i = order; i <= mm->max_order; ++i) {
>> + if (!list_empty(&mm->free_list[i])) {
>
> list_first_entry_or_null already handles that.
[Arun] I will remove list_empty()
>
>> + block = list_first_entry_or_null(&mm->free_list[i],
>> + struct drm_buddy_block,
>> + link);
>> +
>> + if (block)
>> + break;
>> + }
>> + }
>> +
>> + if (!block)
>> + return ERR_PTR(-ENOSPC);
>> +
>> + BUG_ON(!drm_buddy_block_is_free(block));
>> +
>> + while (i != order) {
>> + err = split_block(mm, block);
>> +
>
> No need for the newline.
[Arun] ok
>
>> + if (unlikely(err))
>> + goto err_undo;
>> +
>> + block = block->right;
>> + i--;
>> + }
>> + return block;
>> +
>> +err_undo:
>> + if (i != order)
>> + __drm_buddy_free(mm, block);
>> + return ERR_PTR(err);
>> +}
>> +
>> +/**
>> + * drm_buddy_alloc - allocate power-of-two blocks
>> + *
>> + * @mm: DRM buddy manager to allocate from
>> + * @start: start of the allowed range for this block
>> + * @end: end of the allowed range for this block
>> + * @size: size of the allocation
>> + * @min_page_size: alignment of the allocation
>> + * @blocks: output list head to add allocated blocks
>> + * @flags: DRM_BUDDY_*_ALLOCATION flags
>> + *
>> + * alloc_range() called on range limitations, which traverses
>> + * the tree and returns the desired block.
>> + *
>> + * alloc_from_freelist() called when *no* range restrictions
>> + * are enforced, which picks the block from the freelist.
>> + *
>> + * blocks are allocated in order, the order value here translates to:
>> + *
>> + * 0 = 2^0 * mm->chunk_size
>> + * 1 = 2^1 * mm->chunk_size
>> + * 2 = 2^2 * mm->chunk_size
>> + *
>> + * Returns:
>> + * 0 on success, error code on failure.
>> + */
>> +int drm_buddy_alloc(struct drm_buddy_mm *mm,
>> + u64 start, u64 end, u64 size,
>> + u64 min_page_size,
>> + struct list_head *blocks,
>> + unsigned long flags)
>
> Do we need to validate the flags somewhere?
[Arun] I will move 'unsigned long flags' to enum type declaration
>
>> +{
>> + struct drm_buddy_block *block = NULL;
>> + unsigned int min_order, order;
>> + unsigned long pages;
>> + LIST_HEAD(allocated);
>> + int err;
>> +
>> + if (size < mm->chunk_size)
>> + return -EINVAL;
>
> Also:
>
> if (min_page_size < mm->chunk_size)
> return -EINVAL;
>
> if (!is_power_of_2_u64(min_page_size))
> return -EINVAL;
>
> ?
[Arun] ok
>
>> +
>> + if (!IS_ALIGNED(start, mm->chunk_size))
>> + return -EINVAL;
>> +
>> + if (!IS_ALIGNED(end, mm->chunk_size))
>> + return -EINVAL;
>> +
>> + if (!IS_ALIGNED(size, mm->chunk_size))
>> + return -EINVAL;
>
> Maybe:
>
> if (!IS_ALIGNED(start | end | size, mm->chunk_size))
> return -EINVAL;
>
> ?
[Arun] ok
>
>> +
>> + if (check_range_overflow(start, end, size, mm->size))
>> + return -EINVAL;
>> +
>> + pages = size >> ilog2(mm->chunk_size);
>> + min_order = ilog2(min_page_size) - ilog2(mm->chunk_size);
>> +
>> + do {
>> + order = fls(pages) - 1;
>
> Maybe:
>
> order = min(order, fls(pages) - 1);
[Arun] ok
>
> We shouldn't need to reset the order here all the way back to the number
> of pages, I think. We are still holding the lock, and so we still get
> the same answer as before, assuming we needed to decrement the order
> below. Otherwise that might be slightly wasteful AFAIK.
[Arun] right, I will modify
>
>> + BUG_ON(order > mm->max_order);
>> + BUG_ON(order < min_order);
>> +
>> + do {
>> + if (flags & DRM_BUDDY_RANGE_ALLOCATION)
>> + /* Allocate traversing within the range */
>> + block = alloc_range(mm, start, end, order);
>
> Ok, so blocks might be in a random order, which is a slight concern for
> actual range allocations(not the bias thing). Can we somehow make
> alloc_range just do the old behaviour when end - start == size? Not the
> end of the world though if not.
[Arun] I will change the alloc_range() block allocations to bottom-up,
so both actual range allocation and end bias allocation blocks will
start from lowest address. And, since we are traversing the tree from
left to right, blocks will be in order.
And, alloc_range() handles actual range allocation demands the same way
as in the old i915_buddy_alloc_range() function except alloc_range()
make use of order value to find the blocks within the actual range
allocation.
>
>> + else
>> + /* Allocate from freelist */
>> + block = alloc_from_freelist(mm, order, flags);
>> +
>> + if (!IS_ERR(block))
>> + break;
>> +
>> + if (order-- == min_order) {
>> + err = -ENOSPC;
>> + goto err_free;
>> + }
>> + } while (1);
>> +
>> + mark_allocated(block);
>> + mm->avail -= drm_buddy_block_size(mm, block);
>> + kmemleak_update_trace(block);
>> + list_add_tail(&block->link, &allocated);
>> +
>> + pages -= BIT(order);
>> +
>> + if (!pages)
>> + break;
>> + } while (1);
>> +
>> + list_splice_tail(&allocated, blocks);
>> + return 0;
>>
>> err_free:
>> drm_buddy_free_list(mm, &allocated);
>> return err;
>> }
>> -EXPORT_SYMBOL(drm_buddy_alloc_range);
>> +EXPORT_SYMBOL(drm_buddy_alloc);
>>
>> /**
>> * drm_buddy_block_print - print block information
>> diff --git a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
>> index c4b70cb8c248..75197ba6e40d 100644
>> --- a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
>> +++ b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
>> @@ -36,13 +36,14 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
>> struct i915_ttm_buddy_manager *bman = to_buddy_manager(man);
>> struct i915_ttm_buddy_resource *bman_res;
>> struct drm_buddy_mm *mm = &bman->mm;
>> - unsigned long n_pages;
>> - unsigned int min_order;
>> + unsigned long n_pages, lpfn;
>> u64 min_page_size;
>> u64 size;
>> int err;
>>
>> - GEM_BUG_ON(place->fpfn || place->lpfn);
>> + lpfn = place->lpfn;
>> + if (!lpfn)
>> + lpfn = man->size;
>>
>> bman_res = kzalloc(sizeof(*bman_res), GFP_KERNEL);
>> if (!bman_res)
>> @@ -52,6 +53,9 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
>> INIT_LIST_HEAD(&bman_res->blocks);
>> bman_res->mm = mm;
>>
>> + if (place->fpfn || lpfn != man->size)
>> + bman_res->flags |= DRM_BUDDY_RANGE_ALLOCATION;
>> +
>> GEM_BUG_ON(!bman_res->base.num_pages);
>> size = bman_res->base.num_pages << PAGE_SHIFT;
>>
>> @@ -60,10 +64,16 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
>> min_page_size = bo->page_alignment << PAGE_SHIFT;
>>
>> GEM_BUG_ON(min_page_size < mm->chunk_size);
>> - min_order = ilog2(min_page_size) - ilog2(mm->chunk_size);
>> +
>> if (place->flags & TTM_PL_FLAG_CONTIGUOUS) {
>> + unsigned long pages;
>> +
>> size = roundup_pow_of_two(size);
>> - min_order = ilog2(size) - ilog2(mm->chunk_size);
>> + min_page_size = size;
>> +
>> + pages = size >> ilog2(mm->chunk_size);
>> + if (pages > lpfn)
>> + lpfn = pages;
>> }
>>
>> if (size > mm->size) {
>> @@ -73,34 +83,17 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
>>
>> n_pages = size >> ilog2(mm->chunk_size);
>>
>> - do {
>> - struct drm_buddy_block *block;
>> - unsigned int order;
>> -
>> - order = fls(n_pages) - 1;
>> - GEM_BUG_ON(order > mm->max_order);
>> - GEM_BUG_ON(order < min_order);
>> -
>> - do {
>> - mutex_lock(&bman->lock);
>> - block = drm_buddy_alloc(mm, order);
>> - mutex_unlock(&bman->lock);
>> - if (!IS_ERR(block))
>> - break;
>> -
>> - if (order-- == min_order) {
>> - err = -ENOSPC;
>> - goto err_free_blocks;
>> - }
>> - } while (1);
>> -
>> - n_pages -= BIT(order);
>> -
>> - list_add_tail(&block->link, &bman_res->blocks);
>> + mutex_lock(&bman->lock);
>> + err = drm_buddy_alloc(mm, (uint64_t)place->fpfn << PAGE_SHIFT,
>> + (uint64_t)place->lpfn << PAGE_SHIFT,
>> + (uint64_t)n_pages << PAGE_SHIFT,
>
> s/uint64_t/u64/
>
> Some other places also. AFAIK that is preferred in the kernel, outside
> of maybe uapi headers.
[Arun] ok
>
>> + min_page_size,
>> + &bman_res->blocks,
>> + bman_res->flags);
>> + mutex_unlock(&bman->lock);
>>
>> - if (!n_pages)
>> - break;
>> - } while (1);
>> + if (unlikely(err))
>> + goto err_free_blocks;
>>
>> *res = &bman_res->base;
>> return 0;
>> @@ -266,10 +259,18 @@ int i915_ttm_buddy_man_reserve(struct ttm_resource_manager *man,
>> {
>> struct i915_ttm_buddy_manager *bman = to_buddy_manager(man);
>> struct drm_buddy_mm *mm = &bman->mm;
>> + unsigned long flags = 0;
>> + u64 min_size;
>> int ret;
>>
>> + min_size = size;
>> + flags |= DRM_BUDDY_RANGE_ALLOCATION;
>> +
>> mutex_lock(&bman->lock);
>> - ret = drm_buddy_alloc_range(mm, &bman->reserved, start, size);
>> + ret = drm_buddy_alloc(mm, start, start + size,
>> + size, min_size,
>> + &bman->reserved,
>> + flags);
>> mutex_unlock(&bman->lock);
>>
>> return ret;
>> diff --git a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.h b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.h
>> index fa644b512c2e..5ba490875f66 100644
>> --- a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.h
>> +++ b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.h
>> @@ -20,6 +20,7 @@ struct drm_buddy_mm;
>> *
>> * @base: struct ttm_resource base class we extend
>> * @blocks: the list of struct i915_buddy_block for this resource/allocation
>> + * @flags: DRM_BUDDY_*_ALLOCATION flags
>> * @mm: the struct i915_buddy_mm for this resource
>> *
>> * Extends the struct ttm_resource to manage an address space allocation with
>> @@ -28,6 +29,7 @@ struct drm_buddy_mm;
>> struct i915_ttm_buddy_resource {
>> struct ttm_resource base;
>> struct list_head blocks;
>> + unsigned long flags;
>> struct drm_buddy_mm *mm;
>> };
>>
>> diff --git a/include/drm/drm_buddy.h b/include/drm/drm_buddy.h
>> index 5ce3fc702f80..c7bb5509a7ad 100644
>> --- a/include/drm/drm_buddy.h
>> +++ b/include/drm/drm_buddy.h
>> @@ -13,15 +13,22 @@
>>
>> #include <drm/drm_print.h>
>>
>> -#define range_overflows(start, size, max) ({ \
>> +#define check_range_overflow(start, end, size, max) ({ \
>> typeof(start) start__ = (start); \
>> + typeof(end) end__ = (end);\
>> typeof(size) size__ = (size); \
>> typeof(max) max__ = (max); \
>> (void)(&start__ == &size__); \
>> (void)(&start__ == &max__); \
>> - start__ >= max__ || size__ > max__ - start__; \
>> + (void)(&start__ == &end__); \
>> + (void)(&end__ == &size__); \
>> + (void)(&end__ == &max__); \
>> + start__ >= max__ || end__ > max__ \
>> + || size__ > end__ - start__; \
>> })
>>
>> +#define DRM_BUDDY_RANGE_ALLOCATION (1 << 0)
>
> Would it make sense to make this an enum or something?
>
> enum drm_buddy_alloc_mode {
> DRM_BUDDY_ALLOC_FIRST = 0,
> DRM_BUDDY_ALLOC_TOPDOWN,
> DRM_BUDDY_ALLOC_RANGE,
> };
>
> That way can more easily see where this is actually used, rather than
> having an opaque 'unsigned long flags'? Or are we expecting to combine
> allocation strategies or something? Anyway, just a thought.
>
[Arun] we can, initially I used 'unsigned long flags' keeping in mind to
combine the allocation strategies for end bias demands, but I dont see
any usecase. I will move to enum type declaration.
>> +
>> struct drm_buddy_block {
>> #define DRM_BUDDY_HEADER_OFFSET GENMASK_ULL(63, 12)
>> #define DRM_BUDDY_HEADER_STATE GENMASK_ULL(11, 10)
>> @@ -131,12 +138,11 @@ int drm_buddy_init(struct drm_buddy_mm *mm, u64 size, u64 chunk_size);
>>
>> void drm_buddy_fini(struct drm_buddy_mm *mm);
>>
>> -struct drm_buddy_block *
>> -drm_buddy_alloc(struct drm_buddy_mm *mm, unsigned int order);
>> -
>> -int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
>> - struct list_head *blocks,
>> - u64 start, u64 size);
>> +int drm_buddy_alloc(struct drm_buddy_mm *mm,
>> + u64 start, u64 end, u64 size,
>> + u64 min_page_size,
>> + struct list_head *blocks,
>> + unsigned long flags);
>>
>> void drm_buddy_free(struct drm_buddy_mm *mm, struct drm_buddy_block *block);
>>
>>
^ permalink raw reply [flat|nested] 58+ messages in thread
* Re: [PATCH v2 2/8] drm: improve drm_buddy_alloc function
@ 2021-11-08 19:50 ` Arunpravin
0 siblings, 0 replies; 58+ messages in thread
From: Arunpravin @ 2021-11-08 19:50 UTC (permalink / raw)
To: Matthew Auld, dri-devel, intel-gfx, amd-gfx
Cc: alexander.deucher, tzimmermann, jani.nikula, christian.koenig, daniel
Hi Matthew,
Thanks for the review, Please find my comments
On 04/11/21 12:11 am, Matthew Auld wrote:
> On 25/10/2021 14:00, Arunpravin wrote:
>> - Make drm_buddy_alloc a single function to handle
>> range allocation and non-range allocation demands
>>
>> - Implemented a new function alloc_range() which allocates
>> the requested power-of-two block comply with range limitations
>>
>> - Moved order computation and memory alignment logic from
>> i915 driver to drm buddy
>>
>> V2:
>> merged below changes to keep the build unbroken
>> - drm_buddy_alloc_range() becomes obsolete and may be removed
>> - enable ttm range allocation (fpfn / lpfn) support in i915 driver
>> - apply enhanced drm_buddy_alloc() function to i915 driver
>>
>> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
>> ---
>> drivers/gpu/drm/drm_buddy.c | 265 +++++++++++-------
>> drivers/gpu/drm/i915/i915_ttm_buddy_manager.c | 67 ++---
>> drivers/gpu/drm/i915/i915_ttm_buddy_manager.h | 2 +
>> include/drm/drm_buddy.h | 22 +-
>> 4 files changed, 207 insertions(+), 149 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/drm_buddy.c b/drivers/gpu/drm/drm_buddy.c
>> index 39eb1d224bec..406e3d521903 100644
>> --- a/drivers/gpu/drm/drm_buddy.c
>> +++ b/drivers/gpu/drm/drm_buddy.c
>> @@ -274,63 +274,6 @@ void drm_buddy_free_list(struct drm_buddy_mm *mm, struct list_head *objects)
>> }
>> EXPORT_SYMBOL(drm_buddy_free_list);
>>
>> -/**
>> - * drm_buddy_alloc - allocate power-of-two blocks
>> - *
>> - * @mm: DRM buddy manager to allocate from
>> - * @order: size of the allocation
>> - *
>> - * The order value here translates to:
>> - *
>> - * 0 = 2^0 * mm->chunk_size
>> - * 1 = 2^1 * mm->chunk_size
>> - * 2 = 2^2 * mm->chunk_size
>> - *
>> - * Returns:
>> - * allocated ptr to the &drm_buddy_block on success
>> - */
>> -struct drm_buddy_block *
>> -drm_buddy_alloc(struct drm_buddy_mm *mm, unsigned int order)
>> -{
>> - struct drm_buddy_block *block = NULL;
>> - unsigned int i;
>> - int err;
>> -
>> - for (i = order; i <= mm->max_order; ++i) {
>> - block = list_first_entry_or_null(&mm->free_list[i],
>> - struct drm_buddy_block,
>> - link);
>> - if (block)
>> - break;
>> - }
>> -
>> - if (!block)
>> - return ERR_PTR(-ENOSPC);
>> -
>> - BUG_ON(!drm_buddy_block_is_free(block));
>> -
>> - while (i != order) {
>> - err = split_block(mm, block);
>> - if (unlikely(err))
>> - goto out_free;
>> -
>> - /* Go low */
>> - block = block->left;
>> - i--;
>> - }
>> -
>> - mark_allocated(block);
>> - mm->avail -= drm_buddy_block_size(mm, block);
>> - kmemleak_update_trace(block);
>> - return block;
>> -
>> -out_free:
>> - if (i != order)
>> - __drm_buddy_free(mm, block);
>> - return ERR_PTR(err);
>> -}
>> -EXPORT_SYMBOL(drm_buddy_alloc);
>> -
>> static inline bool overlaps(u64 s1, u64 e1, u64 s2, u64 e2)
>> {
>> return s1 <= e2 && e1 >= s2;
>> @@ -341,52 +284,22 @@ static inline bool contains(u64 s1, u64 e1, u64 s2, u64 e2)
>> return s1 <= s2 && e1 >= e2;
>> }
>>
>> -/**
>> - * drm_buddy_alloc_range - allocate range
>> - *
>> - * @mm: DRM buddy manager to allocate from
>> - * @blocks: output list head to add allocated blocks
>> - * @start: start of the allowed range for this block
>> - * @size: size of the allocation
>> - *
>> - * Intended for pre-allocating portions of the address space, for example to
>> - * reserve a block for the initial framebuffer or similar, hence the expectation
>> - * here is that drm_buddy_alloc() is still the main vehicle for
>> - * allocations, so if that's not the case then the drm_mm range allocator is
>> - * probably a much better fit, and so you should probably go use that instead.
>> - *
>> - * Note that it's safe to chain together multiple alloc_ranges
>> - * with the same blocks list
>> - *
>> - * Returns:
>> - * 0 on success, error code on failure.
>> - */
>> -int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
>> - struct list_head *blocks,
>> - u64 start, u64 size)
>> +static struct drm_buddy_block *
>> +alloc_range(struct drm_buddy_mm *mm,
>> + u64 start, u64 end,
>> + unsigned int order)
>> {
>> struct drm_buddy_block *block;
>> struct drm_buddy_block *buddy;
>> - LIST_HEAD(allocated);
>> LIST_HEAD(dfs);
>> - u64 end;
>> int err;
>> int i;
>>
>> - if (size < mm->chunk_size)
>> - return -EINVAL;
>> -
>> - if (!IS_ALIGNED(size | start, mm->chunk_size))
>> - return -EINVAL;
>> -
>> - if (range_overflows(start, size, mm->size))
>> - return -EINVAL;
>> + end = end - 1;
>>
>> for (i = 0; i < mm->n_roots; ++i)
>> list_add_tail(&mm->roots[i]->tmp_link, &dfs);
>>
>> - end = start + size - 1;
>> -
>> do {
>> u64 block_start;
>> u64 block_end;
>> @@ -394,31 +307,32 @@ int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
>> block = list_first_entry_or_null(&dfs,
>> struct drm_buddy_block,
>> tmp_link);
>> +
>
> No need for the newline.
[Arun] ok
>
>> if (!block)
>> break;
>>
>> list_del(&block->tmp_link);
>>
>> + if (drm_buddy_block_order(block) < order)
>> + continue;
>> +
>> block_start = drm_buddy_block_offset(block);
>> block_end = block_start + drm_buddy_block_size(mm, block) - 1;
>>
>> if (!overlaps(start, end, block_start, block_end))
>> continue;
>>
>> - if (drm_buddy_block_is_allocated(block)) {
>> - err = -ENOSPC;
>> - goto err_free;
>> - }
>> + if (drm_buddy_block_is_allocated(block))
>> + continue;
>>
>> - if (contains(start, end, block_start, block_end)) {
>> - if (!drm_buddy_block_is_free(block)) {
>> - err = -ENOSPC;
>> - goto err_free;
>> - }
>> + if (contains(start, end, block_start, block_end)
>> + && order == drm_buddy_block_order(block)) {
>
> Alignment looks off, also && should be on the line above.
[Arun] ok
>
>> + /*
>> + * Find the free block within the range.
>> + */
>> + if (drm_buddy_block_is_free(block))
>> + return block;
>
> Would it make sense to keep searching here, rather than restarting the
> search from scratch every time? Would it work if we pass in the total
> size and min order?
[Arun] yes, I will rewrite this function
>
> I take it this will now also be used for allocating CPU mappable memory
> using an end bias?
[Arun] yes, this function used for allocating CPU mappable memory using
an end bias and for actual range allocation as well.
>
>>
>> - mark_allocated(block);
>> - mm->avail -= drm_buddy_block_size(mm, block);
>> - list_add_tail(&block->link, &allocated);
>> continue;
>> }
>>
>> @@ -428,12 +342,11 @@ int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
>> goto err_undo;
>> }
>>
>> - list_add(&block->right->tmp_link, &dfs);
>> list_add(&block->left->tmp_link, &dfs);
>> + list_add(&block->right->tmp_link, &dfs);
>
> So this is now top-down? Do we need this? Also I guess this now changes
> the ordering of the allocated blocks? Ideally they would appear in
> order, starting from the lowest address(at least for the actual range
> allocation, and not just the bias thing). I think some places might only
> check the first block to get the device address, where underneath we
> have a contiguous allocation.
[Arun] True, I will keep bottom-up
>
>
>> } while (1);
>>
>> - list_splice_tail(&allocated, blocks);
>> - return 0;
>> + return ERR_PTR(-ENOSPC);
>>
>> err_undo:
>> /*
>> @@ -446,12 +359,148 @@ int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
>> (drm_buddy_block_is_free(block) &&
>> drm_buddy_block_is_free(buddy)))
>> __drm_buddy_free(mm, block);
>> + return ERR_PTR(err);
>> +}
>> +
>> +static struct drm_buddy_block *
>> +alloc_from_freelist(struct drm_buddy_mm *mm,
>> + unsigned int order,
>> + unsigned long flags)
>> +{
>> + struct drm_buddy_block *block = NULL;
>> + unsigned int i;
>> + int err;
>> +
>> + for (i = order; i <= mm->max_order; ++i) {
>> + if (!list_empty(&mm->free_list[i])) {
>
> list_first_entry_or_null already handles that.
[Arun] I will remove list_empty()
>
>> + block = list_first_entry_or_null(&mm->free_list[i],
>> + struct drm_buddy_block,
>> + link);
>> +
>> + if (block)
>> + break;
>> + }
>> + }
>> +
>> + if (!block)
>> + return ERR_PTR(-ENOSPC);
>> +
>> + BUG_ON(!drm_buddy_block_is_free(block));
>> +
>> + while (i != order) {
>> + err = split_block(mm, block);
>> +
>
> No need for the newline.
[Arun] ok
>
>> + if (unlikely(err))
>> + goto err_undo;
>> +
>> + block = block->right;
>> + i--;
>> + }
>> + return block;
>> +
>> +err_undo:
>> + if (i != order)
>> + __drm_buddy_free(mm, block);
>> + return ERR_PTR(err);
>> +}
>> +
>> +/**
>> + * drm_buddy_alloc - allocate power-of-two blocks
>> + *
>> + * @mm: DRM buddy manager to allocate from
>> + * @start: start of the allowed range for this block
>> + * @end: end of the allowed range for this block
>> + * @size: size of the allocation
>> + * @min_page_size: alignment of the allocation
>> + * @blocks: output list head to add allocated blocks
>> + * @flags: DRM_BUDDY_*_ALLOCATION flags
>> + *
>> + * alloc_range() called on range limitations, which traverses
>> + * the tree and returns the desired block.
>> + *
>> + * alloc_from_freelist() called when *no* range restrictions
>> + * are enforced, which picks the block from the freelist.
>> + *
>> + * blocks are allocated in order, the order value here translates to:
>> + *
>> + * 0 = 2^0 * mm->chunk_size
>> + * 1 = 2^1 * mm->chunk_size
>> + * 2 = 2^2 * mm->chunk_size
>> + *
>> + * Returns:
>> + * 0 on success, error code on failure.
>> + */
>> +int drm_buddy_alloc(struct drm_buddy_mm *mm,
>> + u64 start, u64 end, u64 size,
>> + u64 min_page_size,
>> + struct list_head *blocks,
>> + unsigned long flags)
>
> Do we need to validate the flags somewhere?
[Arun] I will move 'unsigned long flags' to enum type declaration
>
>> +{
>> + struct drm_buddy_block *block = NULL;
>> + unsigned int min_order, order;
>> + unsigned long pages;
>> + LIST_HEAD(allocated);
>> + int err;
>> +
>> + if (size < mm->chunk_size)
>> + return -EINVAL;
>
> Also:
>
> if (min_page_size < mm->chunk_size)
> return -EINVAL;
>
> if (!is_power_of_2_u64(min_page_size))
> return -EINVAL;
>
> ?
[Arun] ok
>
>> +
>> + if (!IS_ALIGNED(start, mm->chunk_size))
>> + return -EINVAL;
>> +
>> + if (!IS_ALIGNED(end, mm->chunk_size))
>> + return -EINVAL;
>> +
>> + if (!IS_ALIGNED(size, mm->chunk_size))
>> + return -EINVAL;
>
> Maybe:
>
> if (!IS_ALIGNED(start | end | size, mm->chunk_size))
> return -EINVAL;
>
> ?
[Arun] ok
>
>> +
>> + if (check_range_overflow(start, end, size, mm->size))
>> + return -EINVAL;
>> +
>> + pages = size >> ilog2(mm->chunk_size);
>> + min_order = ilog2(min_page_size) - ilog2(mm->chunk_size);
>> +
>> + do {
>> + order = fls(pages) - 1;
>
> Maybe:
>
> order = min(order, fls(pages) - 1);
[Arun] ok
>
> We shouldn't need to reset the order here all the way back to the number
> of pages, I think. We are still holding the lock, and so we still get
> the same answer as before, assuming we needed to decrement the order
> below. Otherwise that might be slightly wasteful AFAIK.
[Arun] right, I will modify
>
>> + BUG_ON(order > mm->max_order);
>> + BUG_ON(order < min_order);
>> +
>> + do {
>> + if (flags & DRM_BUDDY_RANGE_ALLOCATION)
>> + /* Allocate traversing within the range */
>> + block = alloc_range(mm, start, end, order);
>
> Ok, so blocks might be in a random order, which is a slight concern for
> actual range allocations(not the bias thing). Can we somehow make
> alloc_range just do the old behaviour when end - start == size? Not the
> end of the world though if not.
[Arun] I will change the alloc_range() block allocations to bottom-up,
so both actual range allocation and end bias allocation blocks will
start from lowest address. And, since we are traversing the tree from
left to right, blocks will be in order.
And, alloc_range() handles actual range allocation demands the same way
as in the old i915_buddy_alloc_range() function except alloc_range()
make use of order value to find the blocks within the actual range
allocation.
>
>> + else
>> + /* Allocate from freelist */
>> + block = alloc_from_freelist(mm, order, flags);
>> +
>> + if (!IS_ERR(block))
>> + break;
>> +
>> + if (order-- == min_order) {
>> + err = -ENOSPC;
>> + goto err_free;
>> + }
>> + } while (1);
>> +
>> + mark_allocated(block);
>> + mm->avail -= drm_buddy_block_size(mm, block);
>> + kmemleak_update_trace(block);
>> + list_add_tail(&block->link, &allocated);
>> +
>> + pages -= BIT(order);
>> +
>> + if (!pages)
>> + break;
>> + } while (1);
>> +
>> + list_splice_tail(&allocated, blocks);
>> + return 0;
>>
>> err_free:
>> drm_buddy_free_list(mm, &allocated);
>> return err;
>> }
>> -EXPORT_SYMBOL(drm_buddy_alloc_range);
>> +EXPORT_SYMBOL(drm_buddy_alloc);
>>
>> /**
>> * drm_buddy_block_print - print block information
>> diff --git a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
>> index c4b70cb8c248..75197ba6e40d 100644
>> --- a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
>> +++ b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
>> @@ -36,13 +36,14 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
>> struct i915_ttm_buddy_manager *bman = to_buddy_manager(man);
>> struct i915_ttm_buddy_resource *bman_res;
>> struct drm_buddy_mm *mm = &bman->mm;
>> - unsigned long n_pages;
>> - unsigned int min_order;
>> + unsigned long n_pages, lpfn;
>> u64 min_page_size;
>> u64 size;
>> int err;
>>
>> - GEM_BUG_ON(place->fpfn || place->lpfn);
>> + lpfn = place->lpfn;
>> + if (!lpfn)
>> + lpfn = man->size;
>>
>> bman_res = kzalloc(sizeof(*bman_res), GFP_KERNEL);
>> if (!bman_res)
>> @@ -52,6 +53,9 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
>> INIT_LIST_HEAD(&bman_res->blocks);
>> bman_res->mm = mm;
>>
>> + if (place->fpfn || lpfn != man->size)
>> + bman_res->flags |= DRM_BUDDY_RANGE_ALLOCATION;
>> +
>> GEM_BUG_ON(!bman_res->base.num_pages);
>> size = bman_res->base.num_pages << PAGE_SHIFT;
>>
>> @@ -60,10 +64,16 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
>> min_page_size = bo->page_alignment << PAGE_SHIFT;
>>
>> GEM_BUG_ON(min_page_size < mm->chunk_size);
>> - min_order = ilog2(min_page_size) - ilog2(mm->chunk_size);
>> +
>> if (place->flags & TTM_PL_FLAG_CONTIGUOUS) {
>> + unsigned long pages;
>> +
>> size = roundup_pow_of_two(size);
>> - min_order = ilog2(size) - ilog2(mm->chunk_size);
>> + min_page_size = size;
>> +
>> + pages = size >> ilog2(mm->chunk_size);
>> + if (pages > lpfn)
>> + lpfn = pages;
>> }
>>
>> if (size > mm->size) {
>> @@ -73,34 +83,17 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
>>
>> n_pages = size >> ilog2(mm->chunk_size);
>>
>> - do {
>> - struct drm_buddy_block *block;
>> - unsigned int order;
>> -
>> - order = fls(n_pages) - 1;
>> - GEM_BUG_ON(order > mm->max_order);
>> - GEM_BUG_ON(order < min_order);
>> -
>> - do {
>> - mutex_lock(&bman->lock);
>> - block = drm_buddy_alloc(mm, order);
>> - mutex_unlock(&bman->lock);
>> - if (!IS_ERR(block))
>> - break;
>> -
>> - if (order-- == min_order) {
>> - err = -ENOSPC;
>> - goto err_free_blocks;
>> - }
>> - } while (1);
>> -
>> - n_pages -= BIT(order);
>> -
>> - list_add_tail(&block->link, &bman_res->blocks);
>> + mutex_lock(&bman->lock);
>> + err = drm_buddy_alloc(mm, (uint64_t)place->fpfn << PAGE_SHIFT,
>> + (uint64_t)place->lpfn << PAGE_SHIFT,
>> + (uint64_t)n_pages << PAGE_SHIFT,
>
> s/uint64_t/u64/
>
> Some other places also. AFAIK that is preferred in the kernel, outside
> of maybe uapi headers.
[Arun] ok
>
>> + min_page_size,
>> + &bman_res->blocks,
>> + bman_res->flags);
>> + mutex_unlock(&bman->lock);
>>
>> - if (!n_pages)
>> - break;
>> - } while (1);
>> + if (unlikely(err))
>> + goto err_free_blocks;
>>
>> *res = &bman_res->base;
>> return 0;
>> @@ -266,10 +259,18 @@ int i915_ttm_buddy_man_reserve(struct ttm_resource_manager *man,
>> {
>> struct i915_ttm_buddy_manager *bman = to_buddy_manager(man);
>> struct drm_buddy_mm *mm = &bman->mm;
>> + unsigned long flags = 0;
>> + u64 min_size;
>> int ret;
>>
>> + min_size = size;
>> + flags |= DRM_BUDDY_RANGE_ALLOCATION;
>> +
>> mutex_lock(&bman->lock);
>> - ret = drm_buddy_alloc_range(mm, &bman->reserved, start, size);
>> + ret = drm_buddy_alloc(mm, start, start + size,
>> + size, min_size,
>> + &bman->reserved,
>> + flags);
>> mutex_unlock(&bman->lock);
>>
>> return ret;
>> diff --git a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.h b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.h
>> index fa644b512c2e..5ba490875f66 100644
>> --- a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.h
>> +++ b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.h
>> @@ -20,6 +20,7 @@ struct drm_buddy_mm;
>> *
>> * @base: struct ttm_resource base class we extend
>> * @blocks: the list of struct i915_buddy_block for this resource/allocation
>> + * @flags: DRM_BUDDY_*_ALLOCATION flags
>> * @mm: the struct i915_buddy_mm for this resource
>> *
>> * Extends the struct ttm_resource to manage an address space allocation with
>> @@ -28,6 +29,7 @@ struct drm_buddy_mm;
>> struct i915_ttm_buddy_resource {
>> struct ttm_resource base;
>> struct list_head blocks;
>> + unsigned long flags;
>> struct drm_buddy_mm *mm;
>> };
>>
>> diff --git a/include/drm/drm_buddy.h b/include/drm/drm_buddy.h
>> index 5ce3fc702f80..c7bb5509a7ad 100644
>> --- a/include/drm/drm_buddy.h
>> +++ b/include/drm/drm_buddy.h
>> @@ -13,15 +13,22 @@
>>
>> #include <drm/drm_print.h>
>>
>> -#define range_overflows(start, size, max) ({ \
>> +#define check_range_overflow(start, end, size, max) ({ \
>> typeof(start) start__ = (start); \
>> + typeof(end) end__ = (end);\
>> typeof(size) size__ = (size); \
>> typeof(max) max__ = (max); \
>> (void)(&start__ == &size__); \
>> (void)(&start__ == &max__); \
>> - start__ >= max__ || size__ > max__ - start__; \
>> + (void)(&start__ == &end__); \
>> + (void)(&end__ == &size__); \
>> + (void)(&end__ == &max__); \
>> + start__ >= max__ || end__ > max__ \
>> + || size__ > end__ - start__; \
>> })
>>
>> +#define DRM_BUDDY_RANGE_ALLOCATION (1 << 0)
>
> Would it make sense to make this an enum or something?
>
> enum drm_buddy_alloc_mode {
> DRM_BUDDY_ALLOC_FIRST = 0,
> DRM_BUDDY_ALLOC_TOPDOWN,
> DRM_BUDDY_ALLOC_RANGE,
> };
>
> That way can more easily see where this is actually used, rather than
> having an opaque 'unsigned long flags'? Or are we expecting to combine
> allocation strategies or something? Anyway, just a thought.
>
[Arun] we can, initially I used 'unsigned long flags' keeping in mind to
combine the allocation strategies for end bias demands, but I dont see
any usecase. I will move to enum type declaration.
>> +
>> struct drm_buddy_block {
>> #define DRM_BUDDY_HEADER_OFFSET GENMASK_ULL(63, 12)
>> #define DRM_BUDDY_HEADER_STATE GENMASK_ULL(11, 10)
>> @@ -131,12 +138,11 @@ int drm_buddy_init(struct drm_buddy_mm *mm, u64 size, u64 chunk_size);
>>
>> void drm_buddy_fini(struct drm_buddy_mm *mm);
>>
>> -struct drm_buddy_block *
>> -drm_buddy_alloc(struct drm_buddy_mm *mm, unsigned int order);
>> -
>> -int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
>> - struct list_head *blocks,
>> - u64 start, u64 size);
>> +int drm_buddy_alloc(struct drm_buddy_mm *mm,
>> + u64 start, u64 end, u64 size,
>> + u64 min_page_size,
>> + struct list_head *blocks,
>> + unsigned long flags);
>>
>> void drm_buddy_free(struct drm_buddy_mm *mm, struct drm_buddy_block *block);
>>
>>
^ permalink raw reply [flat|nested] 58+ messages in thread
* Re: [Intel-gfx] [PATCH v2 2/8] drm: improve drm_buddy_alloc function
@ 2021-11-08 19:50 ` Arunpravin
0 siblings, 0 replies; 58+ messages in thread
From: Arunpravin @ 2021-11-08 19:50 UTC (permalink / raw)
To: Matthew Auld, dri-devel, intel-gfx, amd-gfx
Cc: alexander.deucher, tzimmermann, christian.koenig
Hi Matthew,
Thanks for the review, Please find my comments
On 04/11/21 12:11 am, Matthew Auld wrote:
> On 25/10/2021 14:00, Arunpravin wrote:
>> - Make drm_buddy_alloc a single function to handle
>> range allocation and non-range allocation demands
>>
>> - Implemented a new function alloc_range() which allocates
>> the requested power-of-two block comply with range limitations
>>
>> - Moved order computation and memory alignment logic from
>> i915 driver to drm buddy
>>
>> V2:
>> merged below changes to keep the build unbroken
>> - drm_buddy_alloc_range() becomes obsolete and may be removed
>> - enable ttm range allocation (fpfn / lpfn) support in i915 driver
>> - apply enhanced drm_buddy_alloc() function to i915 driver
>>
>> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
>> ---
>> drivers/gpu/drm/drm_buddy.c | 265 +++++++++++-------
>> drivers/gpu/drm/i915/i915_ttm_buddy_manager.c | 67 ++---
>> drivers/gpu/drm/i915/i915_ttm_buddy_manager.h | 2 +
>> include/drm/drm_buddy.h | 22 +-
>> 4 files changed, 207 insertions(+), 149 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/drm_buddy.c b/drivers/gpu/drm/drm_buddy.c
>> index 39eb1d224bec..406e3d521903 100644
>> --- a/drivers/gpu/drm/drm_buddy.c
>> +++ b/drivers/gpu/drm/drm_buddy.c
>> @@ -274,63 +274,6 @@ void drm_buddy_free_list(struct drm_buddy_mm *mm, struct list_head *objects)
>> }
>> EXPORT_SYMBOL(drm_buddy_free_list);
>>
>> -/**
>> - * drm_buddy_alloc - allocate power-of-two blocks
>> - *
>> - * @mm: DRM buddy manager to allocate from
>> - * @order: size of the allocation
>> - *
>> - * The order value here translates to:
>> - *
>> - * 0 = 2^0 * mm->chunk_size
>> - * 1 = 2^1 * mm->chunk_size
>> - * 2 = 2^2 * mm->chunk_size
>> - *
>> - * Returns:
>> - * allocated ptr to the &drm_buddy_block on success
>> - */
>> -struct drm_buddy_block *
>> -drm_buddy_alloc(struct drm_buddy_mm *mm, unsigned int order)
>> -{
>> - struct drm_buddy_block *block = NULL;
>> - unsigned int i;
>> - int err;
>> -
>> - for (i = order; i <= mm->max_order; ++i) {
>> - block = list_first_entry_or_null(&mm->free_list[i],
>> - struct drm_buddy_block,
>> - link);
>> - if (block)
>> - break;
>> - }
>> -
>> - if (!block)
>> - return ERR_PTR(-ENOSPC);
>> -
>> - BUG_ON(!drm_buddy_block_is_free(block));
>> -
>> - while (i != order) {
>> - err = split_block(mm, block);
>> - if (unlikely(err))
>> - goto out_free;
>> -
>> - /* Go low */
>> - block = block->left;
>> - i--;
>> - }
>> -
>> - mark_allocated(block);
>> - mm->avail -= drm_buddy_block_size(mm, block);
>> - kmemleak_update_trace(block);
>> - return block;
>> -
>> -out_free:
>> - if (i != order)
>> - __drm_buddy_free(mm, block);
>> - return ERR_PTR(err);
>> -}
>> -EXPORT_SYMBOL(drm_buddy_alloc);
>> -
>> static inline bool overlaps(u64 s1, u64 e1, u64 s2, u64 e2)
>> {
>> return s1 <= e2 && e1 >= s2;
>> @@ -341,52 +284,22 @@ static inline bool contains(u64 s1, u64 e1, u64 s2, u64 e2)
>> return s1 <= s2 && e1 >= e2;
>> }
>>
>> -/**
>> - * drm_buddy_alloc_range - allocate range
>> - *
>> - * @mm: DRM buddy manager to allocate from
>> - * @blocks: output list head to add allocated blocks
>> - * @start: start of the allowed range for this block
>> - * @size: size of the allocation
>> - *
>> - * Intended for pre-allocating portions of the address space, for example to
>> - * reserve a block for the initial framebuffer or similar, hence the expectation
>> - * here is that drm_buddy_alloc() is still the main vehicle for
>> - * allocations, so if that's not the case then the drm_mm range allocator is
>> - * probably a much better fit, and so you should probably go use that instead.
>> - *
>> - * Note that it's safe to chain together multiple alloc_ranges
>> - * with the same blocks list
>> - *
>> - * Returns:
>> - * 0 on success, error code on failure.
>> - */
>> -int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
>> - struct list_head *blocks,
>> - u64 start, u64 size)
>> +static struct drm_buddy_block *
>> +alloc_range(struct drm_buddy_mm *mm,
>> + u64 start, u64 end,
>> + unsigned int order)
>> {
>> struct drm_buddy_block *block;
>> struct drm_buddy_block *buddy;
>> - LIST_HEAD(allocated);
>> LIST_HEAD(dfs);
>> - u64 end;
>> int err;
>> int i;
>>
>> - if (size < mm->chunk_size)
>> - return -EINVAL;
>> -
>> - if (!IS_ALIGNED(size | start, mm->chunk_size))
>> - return -EINVAL;
>> -
>> - if (range_overflows(start, size, mm->size))
>> - return -EINVAL;
>> + end = end - 1;
>>
>> for (i = 0; i < mm->n_roots; ++i)
>> list_add_tail(&mm->roots[i]->tmp_link, &dfs);
>>
>> - end = start + size - 1;
>> -
>> do {
>> u64 block_start;
>> u64 block_end;
>> @@ -394,31 +307,32 @@ int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
>> block = list_first_entry_or_null(&dfs,
>> struct drm_buddy_block,
>> tmp_link);
>> +
>
> No need for the newline.
[Arun] ok
>
>> if (!block)
>> break;
>>
>> list_del(&block->tmp_link);
>>
>> + if (drm_buddy_block_order(block) < order)
>> + continue;
>> +
>> block_start = drm_buddy_block_offset(block);
>> block_end = block_start + drm_buddy_block_size(mm, block) - 1;
>>
>> if (!overlaps(start, end, block_start, block_end))
>> continue;
>>
>> - if (drm_buddy_block_is_allocated(block)) {
>> - err = -ENOSPC;
>> - goto err_free;
>> - }
>> + if (drm_buddy_block_is_allocated(block))
>> + continue;
>>
>> - if (contains(start, end, block_start, block_end)) {
>> - if (!drm_buddy_block_is_free(block)) {
>> - err = -ENOSPC;
>> - goto err_free;
>> - }
>> + if (contains(start, end, block_start, block_end)
>> + && order == drm_buddy_block_order(block)) {
>
> Alignment looks off, also && should be on the line above.
[Arun] ok
>
>> + /*
>> + * Find the free block within the range.
>> + */
>> + if (drm_buddy_block_is_free(block))
>> + return block;
>
> Would it make sense to keep searching here, rather than restarting the
> search from scratch every time? Would it work if we pass in the total
> size and min order?
[Arun] yes, I will rewrite this function
>
> I take it this will now also be used for allocating CPU mappable memory
> using an end bias?
[Arun] yes, this function used for allocating CPU mappable memory using
an end bias and for actual range allocation as well.
>
>>
>> - mark_allocated(block);
>> - mm->avail -= drm_buddy_block_size(mm, block);
>> - list_add_tail(&block->link, &allocated);
>> continue;
>> }
>>
>> @@ -428,12 +342,11 @@ int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
>> goto err_undo;
>> }
>>
>> - list_add(&block->right->tmp_link, &dfs);
>> list_add(&block->left->tmp_link, &dfs);
>> + list_add(&block->right->tmp_link, &dfs);
>
> So this is now top-down? Do we need this? Also I guess this now changes
> the ordering of the allocated blocks? Ideally they would appear in
> order, starting from the lowest address(at least for the actual range
> allocation, and not just the bias thing). I think some places might only
> check the first block to get the device address, where underneath we
> have a contiguous allocation.
[Arun] True, I will keep bottom-up
>
>
>> } while (1);
>>
>> - list_splice_tail(&allocated, blocks);
>> - return 0;
>> + return ERR_PTR(-ENOSPC);
>>
>> err_undo:
>> /*
>> @@ -446,12 +359,148 @@ int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
>> (drm_buddy_block_is_free(block) &&
>> drm_buddy_block_is_free(buddy)))
>> __drm_buddy_free(mm, block);
>> + return ERR_PTR(err);
>> +}
>> +
>> +static struct drm_buddy_block *
>> +alloc_from_freelist(struct drm_buddy_mm *mm,
>> + unsigned int order,
>> + unsigned long flags)
>> +{
>> + struct drm_buddy_block *block = NULL;
>> + unsigned int i;
>> + int err;
>> +
>> + for (i = order; i <= mm->max_order; ++i) {
>> + if (!list_empty(&mm->free_list[i])) {
>
> list_first_entry_or_null already handles that.
[Arun] I will remove list_empty()
>
>> + block = list_first_entry_or_null(&mm->free_list[i],
>> + struct drm_buddy_block,
>> + link);
>> +
>> + if (block)
>> + break;
>> + }
>> + }
>> +
>> + if (!block)
>> + return ERR_PTR(-ENOSPC);
>> +
>> + BUG_ON(!drm_buddy_block_is_free(block));
>> +
>> + while (i != order) {
>> + err = split_block(mm, block);
>> +
>
> No need for the newline.
[Arun] ok
>
>> + if (unlikely(err))
>> + goto err_undo;
>> +
>> + block = block->right;
>> + i--;
>> + }
>> + return block;
>> +
>> +err_undo:
>> + if (i != order)
>> + __drm_buddy_free(mm, block);
>> + return ERR_PTR(err);
>> +}
>> +
>> +/**
>> + * drm_buddy_alloc - allocate power-of-two blocks
>> + *
>> + * @mm: DRM buddy manager to allocate from
>> + * @start: start of the allowed range for this block
>> + * @end: end of the allowed range for this block
>> + * @size: size of the allocation
>> + * @min_page_size: alignment of the allocation
>> + * @blocks: output list head to add allocated blocks
>> + * @flags: DRM_BUDDY_*_ALLOCATION flags
>> + *
>> + * alloc_range() called on range limitations, which traverses
>> + * the tree and returns the desired block.
>> + *
>> + * alloc_from_freelist() called when *no* range restrictions
>> + * are enforced, which picks the block from the freelist.
>> + *
>> + * blocks are allocated in order, the order value here translates to:
>> + *
>> + * 0 = 2^0 * mm->chunk_size
>> + * 1 = 2^1 * mm->chunk_size
>> + * 2 = 2^2 * mm->chunk_size
>> + *
>> + * Returns:
>> + * 0 on success, error code on failure.
>> + */
>> +int drm_buddy_alloc(struct drm_buddy_mm *mm,
>> + u64 start, u64 end, u64 size,
>> + u64 min_page_size,
>> + struct list_head *blocks,
>> + unsigned long flags)
>
> Do we need to validate the flags somewhere?
[Arun] I will move 'unsigned long flags' to enum type declaration
>
>> +{
>> + struct drm_buddy_block *block = NULL;
>> + unsigned int min_order, order;
>> + unsigned long pages;
>> + LIST_HEAD(allocated);
>> + int err;
>> +
>> + if (size < mm->chunk_size)
>> + return -EINVAL;
>
> Also:
>
> if (min_page_size < mm->chunk_size)
> return -EINVAL;
>
> if (!is_power_of_2_u64(min_page_size))
> return -EINVAL;
>
> ?
[Arun] ok
>
>> +
>> + if (!IS_ALIGNED(start, mm->chunk_size))
>> + return -EINVAL;
>> +
>> + if (!IS_ALIGNED(end, mm->chunk_size))
>> + return -EINVAL;
>> +
>> + if (!IS_ALIGNED(size, mm->chunk_size))
>> + return -EINVAL;
>
> Maybe:
>
> if (!IS_ALIGNED(start | end | size, mm->chunk_size))
> return -EINVAL;
>
> ?
[Arun] ok
>
>> +
>> + if (check_range_overflow(start, end, size, mm->size))
>> + return -EINVAL;
>> +
>> + pages = size >> ilog2(mm->chunk_size);
>> + min_order = ilog2(min_page_size) - ilog2(mm->chunk_size);
>> +
>> + do {
>> + order = fls(pages) - 1;
>
> Maybe:
>
> order = min(order, fls(pages) - 1);
[Arun] ok
>
> We shouldn't need to reset the order here all the way back to the number
> of pages, I think. We are still holding the lock, and so we still get
> the same answer as before, assuming we needed to decrement the order
> below. Otherwise that might be slightly wasteful AFAIK.
[Arun] right, I will modify
>
>> + BUG_ON(order > mm->max_order);
>> + BUG_ON(order < min_order);
>> +
>> + do {
>> + if (flags & DRM_BUDDY_RANGE_ALLOCATION)
>> + /* Allocate traversing within the range */
>> + block = alloc_range(mm, start, end, order);
>
> Ok, so blocks might be in a random order, which is a slight concern for
> actual range allocations(not the bias thing). Can we somehow make
> alloc_range just do the old behaviour when end - start == size? Not the
> end of the world though if not.
[Arun] I will change the alloc_range() block allocations to bottom-up,
so both actual range allocation and end bias allocation blocks will
start from lowest address. And, since we are traversing the tree from
left to right, blocks will be in order.
And, alloc_range() handles actual range allocation demands the same way
as in the old i915_buddy_alloc_range() function except alloc_range()
make use of order value to find the blocks within the actual range
allocation.
>
>> + else
>> + /* Allocate from freelist */
>> + block = alloc_from_freelist(mm, order, flags);
>> +
>> + if (!IS_ERR(block))
>> + break;
>> +
>> + if (order-- == min_order) {
>> + err = -ENOSPC;
>> + goto err_free;
>> + }
>> + } while (1);
>> +
>> + mark_allocated(block);
>> + mm->avail -= drm_buddy_block_size(mm, block);
>> + kmemleak_update_trace(block);
>> + list_add_tail(&block->link, &allocated);
>> +
>> + pages -= BIT(order);
>> +
>> + if (!pages)
>> + break;
>> + } while (1);
>> +
>> + list_splice_tail(&allocated, blocks);
>> + return 0;
>>
>> err_free:
>> drm_buddy_free_list(mm, &allocated);
>> return err;
>> }
>> -EXPORT_SYMBOL(drm_buddy_alloc_range);
>> +EXPORT_SYMBOL(drm_buddy_alloc);
>>
>> /**
>> * drm_buddy_block_print - print block information
>> diff --git a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
>> index c4b70cb8c248..75197ba6e40d 100644
>> --- a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
>> +++ b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
>> @@ -36,13 +36,14 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
>> struct i915_ttm_buddy_manager *bman = to_buddy_manager(man);
>> struct i915_ttm_buddy_resource *bman_res;
>> struct drm_buddy_mm *mm = &bman->mm;
>> - unsigned long n_pages;
>> - unsigned int min_order;
>> + unsigned long n_pages, lpfn;
>> u64 min_page_size;
>> u64 size;
>> int err;
>>
>> - GEM_BUG_ON(place->fpfn || place->lpfn);
>> + lpfn = place->lpfn;
>> + if (!lpfn)
>> + lpfn = man->size;
>>
>> bman_res = kzalloc(sizeof(*bman_res), GFP_KERNEL);
>> if (!bman_res)
>> @@ -52,6 +53,9 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
>> INIT_LIST_HEAD(&bman_res->blocks);
>> bman_res->mm = mm;
>>
>> + if (place->fpfn || lpfn != man->size)
>> + bman_res->flags |= DRM_BUDDY_RANGE_ALLOCATION;
>> +
>> GEM_BUG_ON(!bman_res->base.num_pages);
>> size = bman_res->base.num_pages << PAGE_SHIFT;
>>
>> @@ -60,10 +64,16 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
>> min_page_size = bo->page_alignment << PAGE_SHIFT;
>>
>> GEM_BUG_ON(min_page_size < mm->chunk_size);
>> - min_order = ilog2(min_page_size) - ilog2(mm->chunk_size);
>> +
>> if (place->flags & TTM_PL_FLAG_CONTIGUOUS) {
>> + unsigned long pages;
>> +
>> size = roundup_pow_of_two(size);
>> - min_order = ilog2(size) - ilog2(mm->chunk_size);
>> + min_page_size = size;
>> +
>> + pages = size >> ilog2(mm->chunk_size);
>> + if (pages > lpfn)
>> + lpfn = pages;
>> }
>>
>> if (size > mm->size) {
>> @@ -73,34 +83,17 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
>>
>> n_pages = size >> ilog2(mm->chunk_size);
>>
>> - do {
>> - struct drm_buddy_block *block;
>> - unsigned int order;
>> -
>> - order = fls(n_pages) - 1;
>> - GEM_BUG_ON(order > mm->max_order);
>> - GEM_BUG_ON(order < min_order);
>> -
>> - do {
>> - mutex_lock(&bman->lock);
>> - block = drm_buddy_alloc(mm, order);
>> - mutex_unlock(&bman->lock);
>> - if (!IS_ERR(block))
>> - break;
>> -
>> - if (order-- == min_order) {
>> - err = -ENOSPC;
>> - goto err_free_blocks;
>> - }
>> - } while (1);
>> -
>> - n_pages -= BIT(order);
>> -
>> - list_add_tail(&block->link, &bman_res->blocks);
>> + mutex_lock(&bman->lock);
>> + err = drm_buddy_alloc(mm, (uint64_t)place->fpfn << PAGE_SHIFT,
>> + (uint64_t)place->lpfn << PAGE_SHIFT,
>> + (uint64_t)n_pages << PAGE_SHIFT,
>
> s/uint64_t/u64/
>
> Some other places also. AFAIK that is preferred in the kernel, outside
> of maybe uapi headers.
[Arun] ok
>
>> + min_page_size,
>> + &bman_res->blocks,
>> + bman_res->flags);
>> + mutex_unlock(&bman->lock);
>>
>> - if (!n_pages)
>> - break;
>> - } while (1);
>> + if (unlikely(err))
>> + goto err_free_blocks;
>>
>> *res = &bman_res->base;
>> return 0;
>> @@ -266,10 +259,18 @@ int i915_ttm_buddy_man_reserve(struct ttm_resource_manager *man,
>> {
>> struct i915_ttm_buddy_manager *bman = to_buddy_manager(man);
>> struct drm_buddy_mm *mm = &bman->mm;
>> + unsigned long flags = 0;
>> + u64 min_size;
>> int ret;
>>
>> + min_size = size;
>> + flags |= DRM_BUDDY_RANGE_ALLOCATION;
>> +
>> mutex_lock(&bman->lock);
>> - ret = drm_buddy_alloc_range(mm, &bman->reserved, start, size);
>> + ret = drm_buddy_alloc(mm, start, start + size,
>> + size, min_size,
>> + &bman->reserved,
>> + flags);
>> mutex_unlock(&bman->lock);
>>
>> return ret;
>> diff --git a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.h b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.h
>> index fa644b512c2e..5ba490875f66 100644
>> --- a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.h
>> +++ b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.h
>> @@ -20,6 +20,7 @@ struct drm_buddy_mm;
>> *
>> * @base: struct ttm_resource base class we extend
>> * @blocks: the list of struct i915_buddy_block for this resource/allocation
>> + * @flags: DRM_BUDDY_*_ALLOCATION flags
>> * @mm: the struct i915_buddy_mm for this resource
>> *
>> * Extends the struct ttm_resource to manage an address space allocation with
>> @@ -28,6 +29,7 @@ struct drm_buddy_mm;
>> struct i915_ttm_buddy_resource {
>> struct ttm_resource base;
>> struct list_head blocks;
>> + unsigned long flags;
>> struct drm_buddy_mm *mm;
>> };
>>
>> diff --git a/include/drm/drm_buddy.h b/include/drm/drm_buddy.h
>> index 5ce3fc702f80..c7bb5509a7ad 100644
>> --- a/include/drm/drm_buddy.h
>> +++ b/include/drm/drm_buddy.h
>> @@ -13,15 +13,22 @@
>>
>> #include <drm/drm_print.h>
>>
>> -#define range_overflows(start, size, max) ({ \
>> +#define check_range_overflow(start, end, size, max) ({ \
>> typeof(start) start__ = (start); \
>> + typeof(end) end__ = (end);\
>> typeof(size) size__ = (size); \
>> typeof(max) max__ = (max); \
>> (void)(&start__ == &size__); \
>> (void)(&start__ == &max__); \
>> - start__ >= max__ || size__ > max__ - start__; \
>> + (void)(&start__ == &end__); \
>> + (void)(&end__ == &size__); \
>> + (void)(&end__ == &max__); \
>> + start__ >= max__ || end__ > max__ \
>> + || size__ > end__ - start__; \
>> })
>>
>> +#define DRM_BUDDY_RANGE_ALLOCATION (1 << 0)
>
> Would it make sense to make this an enum or something?
>
> enum drm_buddy_alloc_mode {
> DRM_BUDDY_ALLOC_FIRST = 0,
> DRM_BUDDY_ALLOC_TOPDOWN,
> DRM_BUDDY_ALLOC_RANGE,
> };
>
> That way can more easily see where this is actually used, rather than
> having an opaque 'unsigned long flags'? Or are we expecting to combine
> allocation strategies or something? Anyway, just a thought.
>
[Arun] we can, initially I used 'unsigned long flags' keeping in mind to
combine the allocation strategies for end bias demands, but I dont see
any usecase. I will move to enum type declaration.
>> +
>> struct drm_buddy_block {
>> #define DRM_BUDDY_HEADER_OFFSET GENMASK_ULL(63, 12)
>> #define DRM_BUDDY_HEADER_STATE GENMASK_ULL(11, 10)
>> @@ -131,12 +138,11 @@ int drm_buddy_init(struct drm_buddy_mm *mm, u64 size, u64 chunk_size);
>>
>> void drm_buddy_fini(struct drm_buddy_mm *mm);
>>
>> -struct drm_buddy_block *
>> -drm_buddy_alloc(struct drm_buddy_mm *mm, unsigned int order);
>> -
>> -int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
>> - struct list_head *blocks,
>> - u64 start, u64 size);
>> +int drm_buddy_alloc(struct drm_buddy_mm *mm,
>> + u64 start, u64 end, u64 size,
>> + u64 min_page_size,
>> + struct list_head *blocks,
>> + unsigned long flags);
>>
>> void drm_buddy_free(struct drm_buddy_mm *mm, struct drm_buddy_block *block);
>>
>>
^ permalink raw reply [flat|nested] 58+ messages in thread
* Re: [PATCH 3/8] drm: implement top-down allocation method
2021-11-03 18:44 ` [Intel-gfx] " Matthew Auld
(?)
@ 2021-11-08 21:45 ` Arunpravin
-1 siblings, 0 replies; 58+ messages in thread
From: Arunpravin @ 2021-11-08 21:45 UTC (permalink / raw)
To: Matthew Auld, dri-devel, intel-gfx, amd-gfx
Cc: alexander.deucher, tzimmermann, christian.koenig
On 04/11/21 12:14 am, Matthew Auld wrote:
> On 25/10/2021 14:00, Arunpravin wrote:
>> Implemented a function which walk through the order list,
>> compares the offset and returns the maximum offset block,
>> this method is unpredictable in obtaining the high range
>> address blocks which depends on allocation and deallocation.
>> for instance, if driver requests address at a low specific
>> range, allocator traverses from the root block and splits
>> the larger blocks until it reaches the specific block and
>> in the process of splitting, lower orders in the freelist
>> are occupied with low range address blocks and for the
>> subsequent TOPDOWN memory request we may return the low
>> range blocks.To overcome this issue, we may go with the
>> below approach.
>>
>> The other approach, sorting each order list entries in
>> ascending order and compares the last entry of each
>> order list in the freelist and return the max block.
>> This creates sorting overhead on every drm_buddy_free()
>> request and split up of larger blocks for a single page
>> request.
>>
>> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
>> ---
>> drivers/gpu/drm/drm_buddy.c | 42 +++++++++++++++++++++++++++++++------
>> include/drm/drm_buddy.h | 1 +
>> 2 files changed, 37 insertions(+), 6 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/drm_buddy.c b/drivers/gpu/drm/drm_buddy.c
>> index 406e3d521903..9d3547bcc5da 100644
>> --- a/drivers/gpu/drm/drm_buddy.c
>> +++ b/drivers/gpu/drm/drm_buddy.c
>> @@ -362,6 +362,27 @@ alloc_range(struct drm_buddy_mm *mm,
>> return ERR_PTR(err);
>> }
>>
>> +static struct drm_buddy_block *
>> +get_maxblock(struct list_head *head)
>> +{
>> + struct drm_buddy_block *max_block = NULL, *node;
>> +
>> + max_block = list_first_entry_or_null(head,
>> + struct drm_buddy_block,
>> + link);
>> +
>> + if (!max_block)
>> + return NULL;
>> +
>> + list_for_each_entry(node, head, link) {
>> + if (drm_buddy_block_offset(node) >
>> + drm_buddy_block_offset(max_block))
>
> Alignment.
[Arun] ok
>
>> + max_block = node;
>> + }
>
> I suppose there will be pathological cases where this will unnecessarily
> steal the mappable portion? But in practice maybe this is good enough?
[Arun] we can go with the other approach, sorting each order list
entries in ascending order and compares the last entry of each order
list in the freelist and return the max block. I think this creates
sorting overhead on every drm_buddy_free() request and (max_order -
requested_order) iterations on every top_down allocation request.
With this method, there will be no unnecessary steal of the mappable
portion, but I guess there might be a performance hit.
>
>> +
>> + return max_block;
>> +}
>> +
>> static struct drm_buddy_block *
>> alloc_from_freelist(struct drm_buddy_mm *mm,
>> unsigned int order,
>> @@ -372,13 +393,22 @@ alloc_from_freelist(struct drm_buddy_mm *mm,
>> int err;
>>
>> for (i = order; i <= mm->max_order; ++i) {
>> - if (!list_empty(&mm->free_list[i])) {
>> - block = list_first_entry_or_null(&mm->free_list[i],
>> - struct drm_buddy_block,
>> - link);
>> + if (flags & DRM_BUDDY_TOPDOWN_ALLOCATION) {
>> + if (!list_empty(&mm->free_list[i])) {
>
> AFAIK no need to keep checking list_empty(), below also.
[Arun] ok
>
>> + block = get_maxblock(&mm->free_list[i]);
>>
>> - if (block)
>> - break;
>> + if (block)
>> + break;
>> + }
>> + } else {
>> + if (!list_empty(&mm->free_list[i])) {
>> + block = list_first_entry_or_null(&mm->free_list[i],
>> + struct drm_buddy_block,
>> + link);
>> +
>> + if (block)
>> + break;
>> + }
>> }
>> }
>>
>> diff --git a/include/drm/drm_buddy.h b/include/drm/drm_buddy.h
>> index c7bb5509a7ad..cd8021d2d6e7 100644
>> --- a/include/drm/drm_buddy.h
>> +++ b/include/drm/drm_buddy.h
>> @@ -28,6 +28,7 @@
>> })
>>
>> #define DRM_BUDDY_RANGE_ALLOCATION (1 << 0)
>> +#define DRM_BUDDY_TOPDOWN_ALLOCATION (1 << 1)
>>
>> struct drm_buddy_block {
>> #define DRM_BUDDY_HEADER_OFFSET GENMASK_ULL(63, 12)
>>
^ permalink raw reply [flat|nested] 58+ messages in thread
* Re: [PATCH 3/8] drm: implement top-down allocation method
@ 2021-11-08 21:45 ` Arunpravin
0 siblings, 0 replies; 58+ messages in thread
From: Arunpravin @ 2021-11-08 21:45 UTC (permalink / raw)
To: Matthew Auld, dri-devel, intel-gfx, amd-gfx
Cc: alexander.deucher, tzimmermann, jani.nikula, christian.koenig, daniel
On 04/11/21 12:14 am, Matthew Auld wrote:
> On 25/10/2021 14:00, Arunpravin wrote:
>> Implemented a function which walk through the order list,
>> compares the offset and returns the maximum offset block,
>> this method is unpredictable in obtaining the high range
>> address blocks which depends on allocation and deallocation.
>> for instance, if driver requests address at a low specific
>> range, allocator traverses from the root block and splits
>> the larger blocks until it reaches the specific block and
>> in the process of splitting, lower orders in the freelist
>> are occupied with low range address blocks and for the
>> subsequent TOPDOWN memory request we may return the low
>> range blocks.To overcome this issue, we may go with the
>> below approach.
>>
>> The other approach, sorting each order list entries in
>> ascending order and compares the last entry of each
>> order list in the freelist and return the max block.
>> This creates sorting overhead on every drm_buddy_free()
>> request and split up of larger blocks for a single page
>> request.
>>
>> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
>> ---
>> drivers/gpu/drm/drm_buddy.c | 42 +++++++++++++++++++++++++++++++------
>> include/drm/drm_buddy.h | 1 +
>> 2 files changed, 37 insertions(+), 6 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/drm_buddy.c b/drivers/gpu/drm/drm_buddy.c
>> index 406e3d521903..9d3547bcc5da 100644
>> --- a/drivers/gpu/drm/drm_buddy.c
>> +++ b/drivers/gpu/drm/drm_buddy.c
>> @@ -362,6 +362,27 @@ alloc_range(struct drm_buddy_mm *mm,
>> return ERR_PTR(err);
>> }
>>
>> +static struct drm_buddy_block *
>> +get_maxblock(struct list_head *head)
>> +{
>> + struct drm_buddy_block *max_block = NULL, *node;
>> +
>> + max_block = list_first_entry_or_null(head,
>> + struct drm_buddy_block,
>> + link);
>> +
>> + if (!max_block)
>> + return NULL;
>> +
>> + list_for_each_entry(node, head, link) {
>> + if (drm_buddy_block_offset(node) >
>> + drm_buddy_block_offset(max_block))
>
> Alignment.
[Arun] ok
>
>> + max_block = node;
>> + }
>
> I suppose there will be pathological cases where this will unnecessarily
> steal the mappable portion? But in practice maybe this is good enough?
[Arun] we can go with the other approach, sorting each order list
entries in ascending order and compares the last entry of each order
list in the freelist and return the max block. I think this creates
sorting overhead on every drm_buddy_free() request and (max_order -
requested_order) iterations on every top_down allocation request.
With this method, there will be no unnecessary steal of the mappable
portion, but I guess there might be a performance hit.
>
>> +
>> + return max_block;
>> +}
>> +
>> static struct drm_buddy_block *
>> alloc_from_freelist(struct drm_buddy_mm *mm,
>> unsigned int order,
>> @@ -372,13 +393,22 @@ alloc_from_freelist(struct drm_buddy_mm *mm,
>> int err;
>>
>> for (i = order; i <= mm->max_order; ++i) {
>> - if (!list_empty(&mm->free_list[i])) {
>> - block = list_first_entry_or_null(&mm->free_list[i],
>> - struct drm_buddy_block,
>> - link);
>> + if (flags & DRM_BUDDY_TOPDOWN_ALLOCATION) {
>> + if (!list_empty(&mm->free_list[i])) {
>
> AFAIK no need to keep checking list_empty(), below also.
[Arun] ok
>
>> + block = get_maxblock(&mm->free_list[i]);
>>
>> - if (block)
>> - break;
>> + if (block)
>> + break;
>> + }
>> + } else {
>> + if (!list_empty(&mm->free_list[i])) {
>> + block = list_first_entry_or_null(&mm->free_list[i],
>> + struct drm_buddy_block,
>> + link);
>> +
>> + if (block)
>> + break;
>> + }
>> }
>> }
>>
>> diff --git a/include/drm/drm_buddy.h b/include/drm/drm_buddy.h
>> index c7bb5509a7ad..cd8021d2d6e7 100644
>> --- a/include/drm/drm_buddy.h
>> +++ b/include/drm/drm_buddy.h
>> @@ -28,6 +28,7 @@
>> })
>>
>> #define DRM_BUDDY_RANGE_ALLOCATION (1 << 0)
>> +#define DRM_BUDDY_TOPDOWN_ALLOCATION (1 << 1)
>>
>> struct drm_buddy_block {
>> #define DRM_BUDDY_HEADER_OFFSET GENMASK_ULL(63, 12)
>>
^ permalink raw reply [flat|nested] 58+ messages in thread
* Re: [Intel-gfx] [PATCH 3/8] drm: implement top-down allocation method
@ 2021-11-08 21:45 ` Arunpravin
0 siblings, 0 replies; 58+ messages in thread
From: Arunpravin @ 2021-11-08 21:45 UTC (permalink / raw)
To: Matthew Auld, dri-devel, intel-gfx, amd-gfx
Cc: alexander.deucher, tzimmermann, christian.koenig
On 04/11/21 12:14 am, Matthew Auld wrote:
> On 25/10/2021 14:00, Arunpravin wrote:
>> Implemented a function which walk through the order list,
>> compares the offset and returns the maximum offset block,
>> this method is unpredictable in obtaining the high range
>> address blocks which depends on allocation and deallocation.
>> for instance, if driver requests address at a low specific
>> range, allocator traverses from the root block and splits
>> the larger blocks until it reaches the specific block and
>> in the process of splitting, lower orders in the freelist
>> are occupied with low range address blocks and for the
>> subsequent TOPDOWN memory request we may return the low
>> range blocks.To overcome this issue, we may go with the
>> below approach.
>>
>> The other approach, sorting each order list entries in
>> ascending order and compares the last entry of each
>> order list in the freelist and return the max block.
>> This creates sorting overhead on every drm_buddy_free()
>> request and split up of larger blocks for a single page
>> request.
>>
>> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
>> ---
>> drivers/gpu/drm/drm_buddy.c | 42 +++++++++++++++++++++++++++++++------
>> include/drm/drm_buddy.h | 1 +
>> 2 files changed, 37 insertions(+), 6 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/drm_buddy.c b/drivers/gpu/drm/drm_buddy.c
>> index 406e3d521903..9d3547bcc5da 100644
>> --- a/drivers/gpu/drm/drm_buddy.c
>> +++ b/drivers/gpu/drm/drm_buddy.c
>> @@ -362,6 +362,27 @@ alloc_range(struct drm_buddy_mm *mm,
>> return ERR_PTR(err);
>> }
>>
>> +static struct drm_buddy_block *
>> +get_maxblock(struct list_head *head)
>> +{
>> + struct drm_buddy_block *max_block = NULL, *node;
>> +
>> + max_block = list_first_entry_or_null(head,
>> + struct drm_buddy_block,
>> + link);
>> +
>> + if (!max_block)
>> + return NULL;
>> +
>> + list_for_each_entry(node, head, link) {
>> + if (drm_buddy_block_offset(node) >
>> + drm_buddy_block_offset(max_block))
>
> Alignment.
[Arun] ok
>
>> + max_block = node;
>> + }
>
> I suppose there will be pathological cases where this will unnecessarily
> steal the mappable portion? But in practice maybe this is good enough?
[Arun] we can go with the other approach, sorting each order list
entries in ascending order and compares the last entry of each order
list in the freelist and return the max block. I think this creates
sorting overhead on every drm_buddy_free() request and (max_order -
requested_order) iterations on every top_down allocation request.
With this method, there will be no unnecessary steal of the mappable
portion, but I guess there might be a performance hit.
>
>> +
>> + return max_block;
>> +}
>> +
>> static struct drm_buddy_block *
>> alloc_from_freelist(struct drm_buddy_mm *mm,
>> unsigned int order,
>> @@ -372,13 +393,22 @@ alloc_from_freelist(struct drm_buddy_mm *mm,
>> int err;
>>
>> for (i = order; i <= mm->max_order; ++i) {
>> - if (!list_empty(&mm->free_list[i])) {
>> - block = list_first_entry_or_null(&mm->free_list[i],
>> - struct drm_buddy_block,
>> - link);
>> + if (flags & DRM_BUDDY_TOPDOWN_ALLOCATION) {
>> + if (!list_empty(&mm->free_list[i])) {
>
> AFAIK no need to keep checking list_empty(), below also.
[Arun] ok
>
>> + block = get_maxblock(&mm->free_list[i]);
>>
>> - if (block)
>> - break;
>> + if (block)
>> + break;
>> + }
>> + } else {
>> + if (!list_empty(&mm->free_list[i])) {
>> + block = list_first_entry_or_null(&mm->free_list[i],
>> + struct drm_buddy_block,
>> + link);
>> +
>> + if (block)
>> + break;
>> + }
>> }
>> }
>>
>> diff --git a/include/drm/drm_buddy.h b/include/drm/drm_buddy.h
>> index c7bb5509a7ad..cd8021d2d6e7 100644
>> --- a/include/drm/drm_buddy.h
>> +++ b/include/drm/drm_buddy.h
>> @@ -28,6 +28,7 @@
>> })
>>
>> #define DRM_BUDDY_RANGE_ALLOCATION (1 << 0)
>> +#define DRM_BUDDY_TOPDOWN_ALLOCATION (1 << 1)
>>
>> struct drm_buddy_block {
>> #define DRM_BUDDY_HEADER_OFFSET GENMASK_ULL(63, 12)
>>
^ permalink raw reply [flat|nested] 58+ messages in thread
* Re: [PATCH 6/8] drm/i915: add free_unused_pages support to i915
2021-11-03 19:18 ` [Intel-gfx] " Matthew Auld
(?)
@ 2021-11-08 22:04 ` Arunpravin
-1 siblings, 0 replies; 58+ messages in thread
From: Arunpravin @ 2021-11-08 22:04 UTC (permalink / raw)
To: Matthew Auld, dri-devel, intel-gfx, amd-gfx
Cc: alexander.deucher, tzimmermann, christian.koenig
On 04/11/21 12:48 am, Matthew Auld wrote:
> On 25/10/2021 14:00, Arunpravin wrote:
>> add drm_buddy_free_unused_pages() support on
>> contiguous allocation
>>
>> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
>> ---
>> drivers/gpu/drm/i915/i915_ttm_buddy_manager.c | 8 ++++++++
>> 1 file changed, 8 insertions(+)
>>
>> diff --git a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
>> index 963468228392..162947af8e04 100644
>> --- a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
>> +++ b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
>> @@ -98,6 +98,14 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
>> if (unlikely(err))
>> goto err_free_blocks;
>>
>> + if (place->flags & TTM_PL_FLAG_CONTIGUOUS) {
>> + err = drm_buddy_free_unused_pages(mm, (uint64_t)n_pages << PAGE_SHIFT,
>> + &bman_res->blocks);
>> +
>> + if (unlikely(err))
>> + goto err_free_blocks;
>
> That needs some locking, mark_free/mark_split are all globally visible.
> Some concurrent user might steal the block, or worse.
[Arun] I will lock using mutex
>
>> + }
>> +
>> *res = &bman_res->base;
>> return 0;
>>
>>
^ permalink raw reply [flat|nested] 58+ messages in thread
* Re: [PATCH 6/8] drm/i915: add free_unused_pages support to i915
@ 2021-11-08 22:04 ` Arunpravin
0 siblings, 0 replies; 58+ messages in thread
From: Arunpravin @ 2021-11-08 22:04 UTC (permalink / raw)
To: Matthew Auld, dri-devel, intel-gfx, amd-gfx
Cc: alexander.deucher, tzimmermann, jani.nikula, christian.koenig, daniel
On 04/11/21 12:48 am, Matthew Auld wrote:
> On 25/10/2021 14:00, Arunpravin wrote:
>> add drm_buddy_free_unused_pages() support on
>> contiguous allocation
>>
>> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
>> ---
>> drivers/gpu/drm/i915/i915_ttm_buddy_manager.c | 8 ++++++++
>> 1 file changed, 8 insertions(+)
>>
>> diff --git a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
>> index 963468228392..162947af8e04 100644
>> --- a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
>> +++ b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
>> @@ -98,6 +98,14 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
>> if (unlikely(err))
>> goto err_free_blocks;
>>
>> + if (place->flags & TTM_PL_FLAG_CONTIGUOUS) {
>> + err = drm_buddy_free_unused_pages(mm, (uint64_t)n_pages << PAGE_SHIFT,
>> + &bman_res->blocks);
>> +
>> + if (unlikely(err))
>> + goto err_free_blocks;
>
> That needs some locking, mark_free/mark_split are all globally visible.
> Some concurrent user might steal the block, or worse.
[Arun] I will lock using mutex
>
>> + }
>> +
>> *res = &bman_res->base;
>> return 0;
>>
>>
^ permalink raw reply [flat|nested] 58+ messages in thread
* Re: [Intel-gfx] [PATCH 6/8] drm/i915: add free_unused_pages support to i915
@ 2021-11-08 22:04 ` Arunpravin
0 siblings, 0 replies; 58+ messages in thread
From: Arunpravin @ 2021-11-08 22:04 UTC (permalink / raw)
To: Matthew Auld, dri-devel, intel-gfx, amd-gfx
Cc: alexander.deucher, tzimmermann, christian.koenig
On 04/11/21 12:48 am, Matthew Auld wrote:
> On 25/10/2021 14:00, Arunpravin wrote:
>> add drm_buddy_free_unused_pages() support on
>> contiguous allocation
>>
>> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
>> ---
>> drivers/gpu/drm/i915/i915_ttm_buddy_manager.c | 8 ++++++++
>> 1 file changed, 8 insertions(+)
>>
>> diff --git a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
>> index 963468228392..162947af8e04 100644
>> --- a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
>> +++ b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
>> @@ -98,6 +98,14 @@ static int i915_ttm_buddy_man_alloc(struct ttm_resource_manager *man,
>> if (unlikely(err))
>> goto err_free_blocks;
>>
>> + if (place->flags & TTM_PL_FLAG_CONTIGUOUS) {
>> + err = drm_buddy_free_unused_pages(mm, (uint64_t)n_pages << PAGE_SHIFT,
>> + &bman_res->blocks);
>> +
>> + if (unlikely(err))
>> + goto err_free_blocks;
>
> That needs some locking, mark_free/mark_split are all globally visible.
> Some concurrent user might steal the block, or worse.
[Arun] I will lock using mutex
>
>> + }
>> +
>> *res = &bman_res->base;
>> return 0;
>>
>>
^ permalink raw reply [flat|nested] 58+ messages in thread
* Re: [PATCH 8/8] drm/amdgpu: add drm buddy support to amdgpu
2021-11-04 8:54 ` [Intel-gfx] " Christian König
(?)
@ 2021-11-08 22:08 ` Arunpravin
-1 siblings, 0 replies; 58+ messages in thread
From: Arunpravin @ 2021-11-08 22:08 UTC (permalink / raw)
To: Christian König, Matthew Auld, dri-devel, intel-gfx, amd-gfx
Cc: alexander.deucher, tzimmermann
On 04/11/21 2:24 pm, Christian König wrote:
> Am 04.11.21 um 09:49 schrieb Matthew Auld:
>> On 04/11/2021 07:34, Christian König wrote:
>>>
>>>
>>> Am 03.11.21 um 20:25 schrieb Matthew Auld:
>>>> On 25/10/2021 14:00, Arunpravin wrote:
>>>>> - Remove drm_mm references and replace with drm buddy functionalities
>>>>> - Add res cursor support for drm buddy
>>>>>
>>>>> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
>>>>
>>>> <snip>
>>>>
>>>>> + spin_lock(&mgr->lock);
>>>>> + r = drm_buddy_alloc(mm, (uint64_t)place->fpfn << PAGE_SHIFT,
>>>>> + (uint64_t)lpfn << PAGE_SHIFT,
>>>>> + (uint64_t)n_pages << PAGE_SHIFT,
>>>>> + min_page_size, &node->blocks,
>>>>> + node->flags);
>>>>
>>>>
>>>> Is spinlock + GFP_KERNEL allowed?
>>>
>>> Nope it isn't, but does that function really calls kmalloc()?
>>
>> It calls kmem_cache_zalloc(..., GFP_KERNEL)
>
> Oh that's bad. In this case we either need a mutex here or some other
> approach to avoid the allocation.
>
> Christian.
[Arun] I will replace spinlock with mutex
>
>>
>>>
>>> Christian.
>>>
>>>>
>>>>> + spin_unlock(&mgr->lock);
>>>>> +
>>>>> + if (unlikely(r))
>>>>> + goto error_free_blocks;
>>>>> +
>>>>> pages_left -= pages;
>>>>> ++i;
>>>>> if (pages > pages_left)
>>>>> pages = pages_left;
>>>>> }
>>>>> - spin_unlock(&mgr->lock);
>>>>> +
>>>>> + /* Free unused pages for contiguous allocation */
>>>>> + if (place->flags & TTM_PL_FLAG_CONTIGUOUS) {
>>>>> + uint64_t actual_size = (uint64_t)node->base.num_pages <<
>>>>> PAGE_SHIFT;
>>>>> +
>>>>> + r = drm_buddy_free_unused_pages(mm,
>>>>> + actual_size,
>>>>> + &node->blocks);
>>>>
>>>> Needs some locking.
[Arun] ok
>>>
>
^ permalink raw reply [flat|nested] 58+ messages in thread
* Re: [PATCH 8/8] drm/amdgpu: add drm buddy support to amdgpu
@ 2021-11-08 22:08 ` Arunpravin
0 siblings, 0 replies; 58+ messages in thread
From: Arunpravin @ 2021-11-08 22:08 UTC (permalink / raw)
To: Christian König, Matthew Auld, dri-devel, intel-gfx, amd-gfx
Cc: alexander.deucher, jani.nikula, tzimmermann, daniel
On 04/11/21 2:24 pm, Christian König wrote:
> Am 04.11.21 um 09:49 schrieb Matthew Auld:
>> On 04/11/2021 07:34, Christian König wrote:
>>>
>>>
>>> Am 03.11.21 um 20:25 schrieb Matthew Auld:
>>>> On 25/10/2021 14:00, Arunpravin wrote:
>>>>> - Remove drm_mm references and replace with drm buddy functionalities
>>>>> - Add res cursor support for drm buddy
>>>>>
>>>>> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
>>>>
>>>> <snip>
>>>>
>>>>> + spin_lock(&mgr->lock);
>>>>> + r = drm_buddy_alloc(mm, (uint64_t)place->fpfn << PAGE_SHIFT,
>>>>> + (uint64_t)lpfn << PAGE_SHIFT,
>>>>> + (uint64_t)n_pages << PAGE_SHIFT,
>>>>> + min_page_size, &node->blocks,
>>>>> + node->flags);
>>>>
>>>>
>>>> Is spinlock + GFP_KERNEL allowed?
>>>
>>> Nope it isn't, but does that function really calls kmalloc()?
>>
>> It calls kmem_cache_zalloc(..., GFP_KERNEL)
>
> Oh that's bad. In this case we either need a mutex here or some other
> approach to avoid the allocation.
>
> Christian.
[Arun] I will replace spinlock with mutex
>
>>
>>>
>>> Christian.
>>>
>>>>
>>>>> + spin_unlock(&mgr->lock);
>>>>> +
>>>>> + if (unlikely(r))
>>>>> + goto error_free_blocks;
>>>>> +
>>>>> pages_left -= pages;
>>>>> ++i;
>>>>> if (pages > pages_left)
>>>>> pages = pages_left;
>>>>> }
>>>>> - spin_unlock(&mgr->lock);
>>>>> +
>>>>> + /* Free unused pages for contiguous allocation */
>>>>> + if (place->flags & TTM_PL_FLAG_CONTIGUOUS) {
>>>>> + uint64_t actual_size = (uint64_t)node->base.num_pages <<
>>>>> PAGE_SHIFT;
>>>>> +
>>>>> + r = drm_buddy_free_unused_pages(mm,
>>>>> + actual_size,
>>>>> + &node->blocks);
>>>>
>>>> Needs some locking.
[Arun] ok
>>>
>
^ permalink raw reply [flat|nested] 58+ messages in thread
* Re: [Intel-gfx] [PATCH 8/8] drm/amdgpu: add drm buddy support to amdgpu
@ 2021-11-08 22:08 ` Arunpravin
0 siblings, 0 replies; 58+ messages in thread
From: Arunpravin @ 2021-11-08 22:08 UTC (permalink / raw)
To: Christian König, Matthew Auld, dri-devel, intel-gfx, amd-gfx
Cc: alexander.deucher, tzimmermann
On 04/11/21 2:24 pm, Christian König wrote:
> Am 04.11.21 um 09:49 schrieb Matthew Auld:
>> On 04/11/2021 07:34, Christian König wrote:
>>>
>>>
>>> Am 03.11.21 um 20:25 schrieb Matthew Auld:
>>>> On 25/10/2021 14:00, Arunpravin wrote:
>>>>> - Remove drm_mm references and replace with drm buddy functionalities
>>>>> - Add res cursor support for drm buddy
>>>>>
>>>>> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
>>>>
>>>> <snip>
>>>>
>>>>> + spin_lock(&mgr->lock);
>>>>> + r = drm_buddy_alloc(mm, (uint64_t)place->fpfn << PAGE_SHIFT,
>>>>> + (uint64_t)lpfn << PAGE_SHIFT,
>>>>> + (uint64_t)n_pages << PAGE_SHIFT,
>>>>> + min_page_size, &node->blocks,
>>>>> + node->flags);
>>>>
>>>>
>>>> Is spinlock + GFP_KERNEL allowed?
>>>
>>> Nope it isn't, but does that function really calls kmalloc()?
>>
>> It calls kmem_cache_zalloc(..., GFP_KERNEL)
>
> Oh that's bad. In this case we either need a mutex here or some other
> approach to avoid the allocation.
>
> Christian.
[Arun] I will replace spinlock with mutex
>
>>
>>>
>>> Christian.
>>>
>>>>
>>>>> + spin_unlock(&mgr->lock);
>>>>> +
>>>>> + if (unlikely(r))
>>>>> + goto error_free_blocks;
>>>>> +
>>>>> pages_left -= pages;
>>>>> ++i;
>>>>> if (pages > pages_left)
>>>>> pages = pages_left;
>>>>> }
>>>>> - spin_unlock(&mgr->lock);
>>>>> +
>>>>> + /* Free unused pages for contiguous allocation */
>>>>> + if (place->flags & TTM_PL_FLAG_CONTIGUOUS) {
>>>>> + uint64_t actual_size = (uint64_t)node->base.num_pages <<
>>>>> PAGE_SHIFT;
>>>>> +
>>>>> + r = drm_buddy_free_unused_pages(mm,
>>>>> + actual_size,
>>>>> + &node->blocks);
>>>>
>>>> Needs some locking.
[Arun] ok
>>>
>
^ permalink raw reply [flat|nested] 58+ messages in thread
* Re: [PATCH 5/8] drm: Implement method to free unused pages
2021-11-03 19:16 ` [Intel-gfx] " Matthew Auld
(?)
@ 2021-11-09 14:06 ` Arunpravin
-1 siblings, 0 replies; 58+ messages in thread
From: Arunpravin @ 2021-11-09 14:06 UTC (permalink / raw)
To: Matthew Auld, dri-devel, intel-gfx, amd-gfx
Cc: alexander.deucher, tzimmermann, christian.koenig
On 04/11/21 12:46 am, Matthew Auld wrote:
> On 25/10/2021 14:00, Arunpravin wrote:
>> On contiguous allocation, we round up the size
>> to the *next* power of 2, implement a function
>> to free the unused pages after the newly allocate block.
>>
>> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
>
> Ideally this gets added with some user, so we can see it in action?
> Maybe squash the next patch here?
[Arun] ok
>
>> ---
>> drivers/gpu/drm/drm_buddy.c | 103 ++++++++++++++++++++++++++++++++++++
>> include/drm/drm_buddy.h | 4 ++
>> 2 files changed, 107 insertions(+)
>>
>> diff --git a/drivers/gpu/drm/drm_buddy.c b/drivers/gpu/drm/drm_buddy.c
>> index 9d3547bcc5da..0da8510736eb 100644
>> --- a/drivers/gpu/drm/drm_buddy.c
>> +++ b/drivers/gpu/drm/drm_buddy.c
>> @@ -284,6 +284,109 @@ static inline bool contains(u64 s1, u64 e1, u64 s2, u64 e2)
>> return s1 <= s2 && e1 >= e2;
>> }
>>
>> +/**
>> + * drm_buddy_free_unused_pages - free unused pages
>> + *
>> + * @mm: DRM buddy manager
>> + * @actual_size: original size requested
>> + * @blocks: output list head to add allocated blocks
>> + *
>> + * For contiguous allocation, we round up the size to the nearest
>> + * power of two value, drivers consume *actual* size, so remaining
>> + * portions are unused and it can be freed.
>> + *
>> + * Returns:
>> + * 0 on success, error code on failure.
>> + */
>> +int drm_buddy_free_unused_pages(struct drm_buddy_mm *mm,
>
> drm_buddy_block_trim?
[Arun] ok
>
>> + u64 actual_size,
>
> new_size?
[Arun] ok
>
>> + struct list_head *blocks)
>> +{
>> + struct drm_buddy_block *block;
>> + struct drm_buddy_block *buddy;
>> + u64 actual_start;
>> + u64 actual_end;
>> + LIST_HEAD(dfs);
>> + u64 count = 0;
>> + int err;
>> +
>> + if (!list_is_singular(blocks))
>> + return -EINVAL;
>> +
>> + block = list_first_entry_or_null(blocks,
>> + struct drm_buddy_block,
>> + link);
>> +
>> + if (!block)
>> + return -EINVAL;
>
> list_is_singular() already ensures that I guess?
[Arun] yes it checks the list empty status, I will remove 'if (!block)'
check
>
>
>> +
>> + if (actual_size > drm_buddy_block_size(mm, block))
>> + return -EINVAL;
>> +
>> + if (actual_size == drm_buddy_block_size(mm, block))
>> + return 0;
>
> Probably need to check the alignment of the actual_size, and also check
> that it is non-zero?
[Arun] ok
>
>> +
>> + list_del(&block->link);
>> +
>> + actual_start = drm_buddy_block_offset(block);
>> + actual_end = actual_start + actual_size - 1;
>> +
>> + if (drm_buddy_block_is_allocated(block))
>
> That should rather be a programmer error.
[Arun] ok, I will check for the allocation status and return -EINVAL if
the block is not allocated.
>
>> + mark_free(mm, block);
>> +
>> + list_add(&block->tmp_link, &dfs);
>> +
>> + while (1) {
>> + block = list_first_entry_or_null(&dfs,
>> + struct drm_buddy_block,
>> + tmp_link);
>> +
>> + if (!block)
>> + break;
>> +
>> + list_del(&block->tmp_link);
>> +
>> + if (count == actual_size)
>> + return 0;
>
>
> Check for overlaps somewhere here to avoid needless searching and splitting?
[Arun] ok
>
>> +
>> + if (contains(actual_start, actual_end, drm_buddy_block_offset(block),
>> + (drm_buddy_block_offset(block) + drm_buddy_block_size(mm, block) - 1))) {
>
> Could maybe record the start/end for better readability?
[Arun] ok
>
>> + BUG_ON(!drm_buddy_block_is_free(block));
>> +
>> + /* Allocate only required blocks */
>> + mark_allocated(block);
>> + mm->avail -= drm_buddy_block_size(mm, block);
>> + list_add_tail(&block->link, blocks);
>> + count += drm_buddy_block_size(mm, block);
>> + continue;
>> + }
>> +
>> + if (drm_buddy_block_order(block) == 0)
>> + continue;
>
> Should be impossible with overlaps check added.
[Arun] yes, I will remove
>
>> +
>> + if (!drm_buddy_block_is_split(block)) {
>
> That should always be true.
[Arun] ok
>
>> + err = split_block(mm, block);
>> +
>> + if (unlikely(err))
>> + goto err_undo;
>> + }
>> +
>> + list_add(&block->right->tmp_link, &dfs);
>> + list_add(&block->left->tmp_link, &dfs);
>> + }
>> +
>> + return -ENOSPC;
>
>
> Would it make sense to factor out part of the alloc_range for this? It
> looks roughly the same.
[Arun] This function gets called for non-range allocations (0..max_size)
as well on contiguous allocation. alloc_range() is called only for range
allocations.
>
>> +
>> +err_undo:
>> + buddy = get_buddy(block);
>> + if (buddy &&
>> + (drm_buddy_block_is_free(block) &&
>> + drm_buddy_block_is_free(buddy)))
>> + __drm_buddy_free(mm, block);
>> + return err;
>
>
> Where do we add the block back to the original list? Did we not just
> leak it?
[Arun] we are adding back to the original list if contains() check
becomes true. we are adding all the blocks within the actual_start and
actual_end, and remaining blocks are freed (added to free list).
>
>
>> +}
>> +EXPORT_SYMBOL(drm_buddy_free_unused_pages);
>> +
>> static struct drm_buddy_block *
>> alloc_range(struct drm_buddy_mm *mm,
>> u64 start, u64 end,
>> diff --git a/include/drm/drm_buddy.h b/include/drm/drm_buddy.h
>> index cd8021d2d6e7..1dfc80c88e1f 100644
>> --- a/include/drm/drm_buddy.h
>> +++ b/include/drm/drm_buddy.h
>> @@ -145,6 +145,10 @@ int drm_buddy_alloc(struct drm_buddy_mm *mm,
>> struct list_head *blocks,
>> unsigned long flags);
>>
>> +int drm_buddy_free_unused_pages(struct drm_buddy_mm *mm,
>> + u64 actual_size,
>> + struct list_head *blocks);
>> +
>> void drm_buddy_free(struct drm_buddy_mm *mm, struct drm_buddy_block *block);
>>
>> void drm_buddy_free_list(struct drm_buddy_mm *mm, struct list_head *objects);
>>
^ permalink raw reply [flat|nested] 58+ messages in thread
* Re: [PATCH 5/8] drm: Implement method to free unused pages
@ 2021-11-09 14:06 ` Arunpravin
0 siblings, 0 replies; 58+ messages in thread
From: Arunpravin @ 2021-11-09 14:06 UTC (permalink / raw)
To: Matthew Auld, dri-devel, intel-gfx, amd-gfx
Cc: alexander.deucher, tzimmermann, jani.nikula, christian.koenig, daniel
On 04/11/21 12:46 am, Matthew Auld wrote:
> On 25/10/2021 14:00, Arunpravin wrote:
>> On contiguous allocation, we round up the size
>> to the *next* power of 2, implement a function
>> to free the unused pages after the newly allocate block.
>>
>> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
>
> Ideally this gets added with some user, so we can see it in action?
> Maybe squash the next patch here?
[Arun] ok
>
>> ---
>> drivers/gpu/drm/drm_buddy.c | 103 ++++++++++++++++++++++++++++++++++++
>> include/drm/drm_buddy.h | 4 ++
>> 2 files changed, 107 insertions(+)
>>
>> diff --git a/drivers/gpu/drm/drm_buddy.c b/drivers/gpu/drm/drm_buddy.c
>> index 9d3547bcc5da..0da8510736eb 100644
>> --- a/drivers/gpu/drm/drm_buddy.c
>> +++ b/drivers/gpu/drm/drm_buddy.c
>> @@ -284,6 +284,109 @@ static inline bool contains(u64 s1, u64 e1, u64 s2, u64 e2)
>> return s1 <= s2 && e1 >= e2;
>> }
>>
>> +/**
>> + * drm_buddy_free_unused_pages - free unused pages
>> + *
>> + * @mm: DRM buddy manager
>> + * @actual_size: original size requested
>> + * @blocks: output list head to add allocated blocks
>> + *
>> + * For contiguous allocation, we round up the size to the nearest
>> + * power of two value, drivers consume *actual* size, so remaining
>> + * portions are unused and it can be freed.
>> + *
>> + * Returns:
>> + * 0 on success, error code on failure.
>> + */
>> +int drm_buddy_free_unused_pages(struct drm_buddy_mm *mm,
>
> drm_buddy_block_trim?
[Arun] ok
>
>> + u64 actual_size,
>
> new_size?
[Arun] ok
>
>> + struct list_head *blocks)
>> +{
>> + struct drm_buddy_block *block;
>> + struct drm_buddy_block *buddy;
>> + u64 actual_start;
>> + u64 actual_end;
>> + LIST_HEAD(dfs);
>> + u64 count = 0;
>> + int err;
>> +
>> + if (!list_is_singular(blocks))
>> + return -EINVAL;
>> +
>> + block = list_first_entry_or_null(blocks,
>> + struct drm_buddy_block,
>> + link);
>> +
>> + if (!block)
>> + return -EINVAL;
>
> list_is_singular() already ensures that I guess?
[Arun] yes it checks the list empty status, I will remove 'if (!block)'
check
>
>
>> +
>> + if (actual_size > drm_buddy_block_size(mm, block))
>> + return -EINVAL;
>> +
>> + if (actual_size == drm_buddy_block_size(mm, block))
>> + return 0;
>
> Probably need to check the alignment of the actual_size, and also check
> that it is non-zero?
[Arun] ok
>
>> +
>> + list_del(&block->link);
>> +
>> + actual_start = drm_buddy_block_offset(block);
>> + actual_end = actual_start + actual_size - 1;
>> +
>> + if (drm_buddy_block_is_allocated(block))
>
> That should rather be a programmer error.
[Arun] ok, I will check for the allocation status and return -EINVAL if
the block is not allocated.
>
>> + mark_free(mm, block);
>> +
>> + list_add(&block->tmp_link, &dfs);
>> +
>> + while (1) {
>> + block = list_first_entry_or_null(&dfs,
>> + struct drm_buddy_block,
>> + tmp_link);
>> +
>> + if (!block)
>> + break;
>> +
>> + list_del(&block->tmp_link);
>> +
>> + if (count == actual_size)
>> + return 0;
>
>
> Check for overlaps somewhere here to avoid needless searching and splitting?
[Arun] ok
>
>> +
>> + if (contains(actual_start, actual_end, drm_buddy_block_offset(block),
>> + (drm_buddy_block_offset(block) + drm_buddy_block_size(mm, block) - 1))) {
>
> Could maybe record the start/end for better readability?
[Arun] ok
>
>> + BUG_ON(!drm_buddy_block_is_free(block));
>> +
>> + /* Allocate only required blocks */
>> + mark_allocated(block);
>> + mm->avail -= drm_buddy_block_size(mm, block);
>> + list_add_tail(&block->link, blocks);
>> + count += drm_buddy_block_size(mm, block);
>> + continue;
>> + }
>> +
>> + if (drm_buddy_block_order(block) == 0)
>> + continue;
>
> Should be impossible with overlaps check added.
[Arun] yes, I will remove
>
>> +
>> + if (!drm_buddy_block_is_split(block)) {
>
> That should always be true.
[Arun] ok
>
>> + err = split_block(mm, block);
>> +
>> + if (unlikely(err))
>> + goto err_undo;
>> + }
>> +
>> + list_add(&block->right->tmp_link, &dfs);
>> + list_add(&block->left->tmp_link, &dfs);
>> + }
>> +
>> + return -ENOSPC;
>
>
> Would it make sense to factor out part of the alloc_range for this? It
> looks roughly the same.
[Arun] This function gets called for non-range allocations (0..max_size)
as well on contiguous allocation. alloc_range() is called only for range
allocations.
>
>> +
>> +err_undo:
>> + buddy = get_buddy(block);
>> + if (buddy &&
>> + (drm_buddy_block_is_free(block) &&
>> + drm_buddy_block_is_free(buddy)))
>> + __drm_buddy_free(mm, block);
>> + return err;
>
>
> Where do we add the block back to the original list? Did we not just
> leak it?
[Arun] we are adding back to the original list if contains() check
becomes true. we are adding all the blocks within the actual_start and
actual_end, and remaining blocks are freed (added to free list).
>
>
>> +}
>> +EXPORT_SYMBOL(drm_buddy_free_unused_pages);
>> +
>> static struct drm_buddy_block *
>> alloc_range(struct drm_buddy_mm *mm,
>> u64 start, u64 end,
>> diff --git a/include/drm/drm_buddy.h b/include/drm/drm_buddy.h
>> index cd8021d2d6e7..1dfc80c88e1f 100644
>> --- a/include/drm/drm_buddy.h
>> +++ b/include/drm/drm_buddy.h
>> @@ -145,6 +145,10 @@ int drm_buddy_alloc(struct drm_buddy_mm *mm,
>> struct list_head *blocks,
>> unsigned long flags);
>>
>> +int drm_buddy_free_unused_pages(struct drm_buddy_mm *mm,
>> + u64 actual_size,
>> + struct list_head *blocks);
>> +
>> void drm_buddy_free(struct drm_buddy_mm *mm, struct drm_buddy_block *block);
>>
>> void drm_buddy_free_list(struct drm_buddy_mm *mm, struct list_head *objects);
>>
^ permalink raw reply [flat|nested] 58+ messages in thread
* Re: [Intel-gfx] [PATCH 5/8] drm: Implement method to free unused pages
@ 2021-11-09 14:06 ` Arunpravin
0 siblings, 0 replies; 58+ messages in thread
From: Arunpravin @ 2021-11-09 14:06 UTC (permalink / raw)
To: Matthew Auld, dri-devel, intel-gfx, amd-gfx
Cc: alexander.deucher, tzimmermann, christian.koenig
On 04/11/21 12:46 am, Matthew Auld wrote:
> On 25/10/2021 14:00, Arunpravin wrote:
>> On contiguous allocation, we round up the size
>> to the *next* power of 2, implement a function
>> to free the unused pages after the newly allocate block.
>>
>> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
>
> Ideally this gets added with some user, so we can see it in action?
> Maybe squash the next patch here?
[Arun] ok
>
>> ---
>> drivers/gpu/drm/drm_buddy.c | 103 ++++++++++++++++++++++++++++++++++++
>> include/drm/drm_buddy.h | 4 ++
>> 2 files changed, 107 insertions(+)
>>
>> diff --git a/drivers/gpu/drm/drm_buddy.c b/drivers/gpu/drm/drm_buddy.c
>> index 9d3547bcc5da..0da8510736eb 100644
>> --- a/drivers/gpu/drm/drm_buddy.c
>> +++ b/drivers/gpu/drm/drm_buddy.c
>> @@ -284,6 +284,109 @@ static inline bool contains(u64 s1, u64 e1, u64 s2, u64 e2)
>> return s1 <= s2 && e1 >= e2;
>> }
>>
>> +/**
>> + * drm_buddy_free_unused_pages - free unused pages
>> + *
>> + * @mm: DRM buddy manager
>> + * @actual_size: original size requested
>> + * @blocks: output list head to add allocated blocks
>> + *
>> + * For contiguous allocation, we round up the size to the nearest
>> + * power of two value, drivers consume *actual* size, so remaining
>> + * portions are unused and it can be freed.
>> + *
>> + * Returns:
>> + * 0 on success, error code on failure.
>> + */
>> +int drm_buddy_free_unused_pages(struct drm_buddy_mm *mm,
>
> drm_buddy_block_trim?
[Arun] ok
>
>> + u64 actual_size,
>
> new_size?
[Arun] ok
>
>> + struct list_head *blocks)
>> +{
>> + struct drm_buddy_block *block;
>> + struct drm_buddy_block *buddy;
>> + u64 actual_start;
>> + u64 actual_end;
>> + LIST_HEAD(dfs);
>> + u64 count = 0;
>> + int err;
>> +
>> + if (!list_is_singular(blocks))
>> + return -EINVAL;
>> +
>> + block = list_first_entry_or_null(blocks,
>> + struct drm_buddy_block,
>> + link);
>> +
>> + if (!block)
>> + return -EINVAL;
>
> list_is_singular() already ensures that I guess?
[Arun] yes it checks the list empty status, I will remove 'if (!block)'
check
>
>
>> +
>> + if (actual_size > drm_buddy_block_size(mm, block))
>> + return -EINVAL;
>> +
>> + if (actual_size == drm_buddy_block_size(mm, block))
>> + return 0;
>
> Probably need to check the alignment of the actual_size, and also check
> that it is non-zero?
[Arun] ok
>
>> +
>> + list_del(&block->link);
>> +
>> + actual_start = drm_buddy_block_offset(block);
>> + actual_end = actual_start + actual_size - 1;
>> +
>> + if (drm_buddy_block_is_allocated(block))
>
> That should rather be a programmer error.
[Arun] ok, I will check for the allocation status and return -EINVAL if
the block is not allocated.
>
>> + mark_free(mm, block);
>> +
>> + list_add(&block->tmp_link, &dfs);
>> +
>> + while (1) {
>> + block = list_first_entry_or_null(&dfs,
>> + struct drm_buddy_block,
>> + tmp_link);
>> +
>> + if (!block)
>> + break;
>> +
>> + list_del(&block->tmp_link);
>> +
>> + if (count == actual_size)
>> + return 0;
>
>
> Check for overlaps somewhere here to avoid needless searching and splitting?
[Arun] ok
>
>> +
>> + if (contains(actual_start, actual_end, drm_buddy_block_offset(block),
>> + (drm_buddy_block_offset(block) + drm_buddy_block_size(mm, block) - 1))) {
>
> Could maybe record the start/end for better readability?
[Arun] ok
>
>> + BUG_ON(!drm_buddy_block_is_free(block));
>> +
>> + /* Allocate only required blocks */
>> + mark_allocated(block);
>> + mm->avail -= drm_buddy_block_size(mm, block);
>> + list_add_tail(&block->link, blocks);
>> + count += drm_buddy_block_size(mm, block);
>> + continue;
>> + }
>> +
>> + if (drm_buddy_block_order(block) == 0)
>> + continue;
>
> Should be impossible with overlaps check added.
[Arun] yes, I will remove
>
>> +
>> + if (!drm_buddy_block_is_split(block)) {
>
> That should always be true.
[Arun] ok
>
>> + err = split_block(mm, block);
>> +
>> + if (unlikely(err))
>> + goto err_undo;
>> + }
>> +
>> + list_add(&block->right->tmp_link, &dfs);
>> + list_add(&block->left->tmp_link, &dfs);
>> + }
>> +
>> + return -ENOSPC;
>
>
> Would it make sense to factor out part of the alloc_range for this? It
> looks roughly the same.
[Arun] This function gets called for non-range allocations (0..max_size)
as well on contiguous allocation. alloc_range() is called only for range
allocations.
>
>> +
>> +err_undo:
>> + buddy = get_buddy(block);
>> + if (buddy &&
>> + (drm_buddy_block_is_free(block) &&
>> + drm_buddy_block_is_free(buddy)))
>> + __drm_buddy_free(mm, block);
>> + return err;
>
>
> Where do we add the block back to the original list? Did we not just
> leak it?
[Arun] we are adding back to the original list if contains() check
becomes true. we are adding all the blocks within the actual_start and
actual_end, and remaining blocks are freed (added to free list).
>
>
>> +}
>> +EXPORT_SYMBOL(drm_buddy_free_unused_pages);
>> +
>> static struct drm_buddy_block *
>> alloc_range(struct drm_buddy_mm *mm,
>> u64 start, u64 end,
>> diff --git a/include/drm/drm_buddy.h b/include/drm/drm_buddy.h
>> index cd8021d2d6e7..1dfc80c88e1f 100644
>> --- a/include/drm/drm_buddy.h
>> +++ b/include/drm/drm_buddy.h
>> @@ -145,6 +145,10 @@ int drm_buddy_alloc(struct drm_buddy_mm *mm,
>> struct list_head *blocks,
>> unsigned long flags);
>>
>> +int drm_buddy_free_unused_pages(struct drm_buddy_mm *mm,
>> + u64 actual_size,
>> + struct list_head *blocks);
>> +
>> void drm_buddy_free(struct drm_buddy_mm *mm, struct drm_buddy_block *block);
>>
>> void drm_buddy_free_list(struct drm_buddy_mm *mm, struct list_head *objects);
>>
^ permalink raw reply [flat|nested] 58+ messages in thread
* Re: [PATCH v2 2/8] drm: improve drm_buddy_alloc function
2021-11-08 19:50 ` Arunpravin
@ 2021-11-15 21:52 ` Arunpravin
-1 siblings, 0 replies; 58+ messages in thread
From: Arunpravin @ 2021-11-15 21:52 UTC (permalink / raw)
To: Matthew Auld, dri-devel, intel-gfx, amd-gfx
Hi Matthew,
I am preparing version3 of the buddy allocator,
Please find the updated comments.
SNIP
>>> -int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
>>> - struct list_head *blocks,
>>> - u64 start, u64 size)
>>> +static struct drm_buddy_block *
>>> +alloc_range(struct drm_buddy_mm *mm,
>>> + u64 start, u64 end,
>>> + unsigned int order)
>>> {
>>> struct drm_buddy_block *block;
>>> struct drm_buddy_block *buddy;
>>> - LIST_HEAD(allocated);
>>> LIST_HEAD(dfs);
>>> - u64 end;
>>> int err;
>>> int i;
>>>
>>
>>> if (!block)
>>> break;
>>>
>>> list_del(&block->tmp_link);
>>>
>>> + if (drm_buddy_block_order(block) < order)
>>> + continue;
>>> +
>>> block_start = drm_buddy_block_offset(block);
>>> block_end = block_start + drm_buddy_block_size(mm, block) - 1;
>>>
>>> if (!overlaps(start, end, block_start, block_end))
>>> continue;
>>>
>>> - if (drm_buddy_block_is_allocated(block)) {
>>> - err = -ENOSPC;
>>> - goto err_free;
>>> - }
>>> + if (drm_buddy_block_is_allocated(block))
>>> + continue;
>>>
>>> - if (contains(start, end, block_start, block_end)) {
>>> - if (!drm_buddy_block_is_free(block)) {
>>> - err = -ENOSPC;
>>> - goto err_free;
>>> - }
>>> + if (contains(start, end, block_start, block_end)
>>> + && order == drm_buddy_block_order(block)) {
>>
>> Alignment looks off, also && should be on the line above.
>
> [Arun] ok
>>
>>> + /*
>>> + * Find the free block within the range.
>>> + */
>>> + if (drm_buddy_block_is_free(block))
>>> + return block;
>>
>> Would it make sense to keep searching here, rather than restarting the
>> search from scratch every time? Would it work if we pass in the total
>> size and min order?
> [Arun] yes, I will rewrite this function
I tried to rewrite the function, AFAIK, in case of end bias allocation,
we have to restart the search on every new order computed value from the
requested total size since we have to find a free node in the required
order level traversing from left to right, here continuing the search
for the subsequent order value would skip the free nodes present in the
beginning of the tree.
In case of actual range allocation, as handled at
i915_buddy_alloc_range, we can continue the search from where the
previous allocation happened since we allocate all the blocks
progressively within the start and end address values.
alloc_range() handles both the cases, having a penalty of restarting the
search in case of actual range allocation. Please let me know if any
suggestions?
>>> +int drm_buddy_alloc(struct drm_buddy_mm *mm,
>>> + u64 start, u64 end, u64 size,
>>> + u64 min_page_size,
>>> + struct list_head *blocks,
>>> + unsigned long flags)
>>
>> Do we need to validate the flags somewhere?
> [Arun] I will move 'unsigned long flags' to enum type declaration
I tried to move 'unsigned long flags' to enum type declaration, it
creates an ambiguity in i915 driver as both DRM_BUDDY_ALLOC_TOPDOWN and
DRM_BUDDY_ALLOC_RANGE are mutually non-exclusive. So I think its better
to have 'unsigned long flags'.
AFAIK, we don't need to validate the flags since we check flags using
'flags & DRM_BUDDY_RANGE_ALLOCATION'
>>
>>> + BUG_ON(order > mm->max_order);
>>> + BUG_ON(order < min_order);
>>> +
>>> + do {
>>> + if (flags & DRM_BUDDY_RANGE_ALLOCATION)
>>> + /* Allocate traversing within the range */
>>> + block = alloc_range(mm, start, end, order);
>>
>> Ok, so blocks might be in a random order, which is a slight concern for
>> actual range allocations(not the bias thing). Can we somehow make
>> alloc_range just do the old behaviour when end - start == size? Not the
>> end of the world though if not.
> [Arun] I will change the alloc_range() block allocations to bottom-up,
> so both actual range allocation and end bias allocation blocks will
> start from lowest address. And, since we are traversing the tree from
> left to right, blocks will be in order.
>
> And, alloc_range() handles actual range allocation demands the same way
> as in the old i915_buddy_alloc_range() function except alloc_range()
> make use of order value to find the blocks within the actual range
> allocation.
Correction - I will change alloc_range() block allocations to bottom-up,
so actual range allocation blocks will start from lowest address (not
the bias thing), and since we are traversing the tree from left to right
progressively within the required range, blocks will be in order (not
the bias thing)
alloc_range() handles actual range allocation demands the same way as in
the old i915_buddy_alloc_range() function except it restarts the search
(since we are handling both end bias and actual range allocations in the
same function) and make use of order value to find the blocks within
start and end value.
Regards,
Arun
^ permalink raw reply [flat|nested] 58+ messages in thread
* Re: [Intel-gfx] [PATCH v2 2/8] drm: improve drm_buddy_alloc function
@ 2021-11-15 21:52 ` Arunpravin
0 siblings, 0 replies; 58+ messages in thread
From: Arunpravin @ 2021-11-15 21:52 UTC (permalink / raw)
To: Matthew Auld, dri-devel, intel-gfx, amd-gfx
Hi Matthew,
I am preparing version3 of the buddy allocator,
Please find the updated comments.
SNIP
>>> -int drm_buddy_alloc_range(struct drm_buddy_mm *mm,
>>> - struct list_head *blocks,
>>> - u64 start, u64 size)
>>> +static struct drm_buddy_block *
>>> +alloc_range(struct drm_buddy_mm *mm,
>>> + u64 start, u64 end,
>>> + unsigned int order)
>>> {
>>> struct drm_buddy_block *block;
>>> struct drm_buddy_block *buddy;
>>> - LIST_HEAD(allocated);
>>> LIST_HEAD(dfs);
>>> - u64 end;
>>> int err;
>>> int i;
>>>
>>
>>> if (!block)
>>> break;
>>>
>>> list_del(&block->tmp_link);
>>>
>>> + if (drm_buddy_block_order(block) < order)
>>> + continue;
>>> +
>>> block_start = drm_buddy_block_offset(block);
>>> block_end = block_start + drm_buddy_block_size(mm, block) - 1;
>>>
>>> if (!overlaps(start, end, block_start, block_end))
>>> continue;
>>>
>>> - if (drm_buddy_block_is_allocated(block)) {
>>> - err = -ENOSPC;
>>> - goto err_free;
>>> - }
>>> + if (drm_buddy_block_is_allocated(block))
>>> + continue;
>>>
>>> - if (contains(start, end, block_start, block_end)) {
>>> - if (!drm_buddy_block_is_free(block)) {
>>> - err = -ENOSPC;
>>> - goto err_free;
>>> - }
>>> + if (contains(start, end, block_start, block_end)
>>> + && order == drm_buddy_block_order(block)) {
>>
>> Alignment looks off, also && should be on the line above.
>
> [Arun] ok
>>
>>> + /*
>>> + * Find the free block within the range.
>>> + */
>>> + if (drm_buddy_block_is_free(block))
>>> + return block;
>>
>> Would it make sense to keep searching here, rather than restarting the
>> search from scratch every time? Would it work if we pass in the total
>> size and min order?
> [Arun] yes, I will rewrite this function
I tried to rewrite the function, AFAIK, in case of end bias allocation,
we have to restart the search on every new order computed value from the
requested total size since we have to find a free node in the required
order level traversing from left to right, here continuing the search
for the subsequent order value would skip the free nodes present in the
beginning of the tree.
In case of actual range allocation, as handled at
i915_buddy_alloc_range, we can continue the search from where the
previous allocation happened since we allocate all the blocks
progressively within the start and end address values.
alloc_range() handles both the cases, having a penalty of restarting the
search in case of actual range allocation. Please let me know if any
suggestions?
>>> +int drm_buddy_alloc(struct drm_buddy_mm *mm,
>>> + u64 start, u64 end, u64 size,
>>> + u64 min_page_size,
>>> + struct list_head *blocks,
>>> + unsigned long flags)
>>
>> Do we need to validate the flags somewhere?
> [Arun] I will move 'unsigned long flags' to enum type declaration
I tried to move 'unsigned long flags' to enum type declaration, it
creates an ambiguity in i915 driver as both DRM_BUDDY_ALLOC_TOPDOWN and
DRM_BUDDY_ALLOC_RANGE are mutually non-exclusive. So I think its better
to have 'unsigned long flags'.
AFAIK, we don't need to validate the flags since we check flags using
'flags & DRM_BUDDY_RANGE_ALLOCATION'
>>
>>> + BUG_ON(order > mm->max_order);
>>> + BUG_ON(order < min_order);
>>> +
>>> + do {
>>> + if (flags & DRM_BUDDY_RANGE_ALLOCATION)
>>> + /* Allocate traversing within the range */
>>> + block = alloc_range(mm, start, end, order);
>>
>> Ok, so blocks might be in a random order, which is a slight concern for
>> actual range allocations(not the bias thing). Can we somehow make
>> alloc_range just do the old behaviour when end - start == size? Not the
>> end of the world though if not.
> [Arun] I will change the alloc_range() block allocations to bottom-up,
> so both actual range allocation and end bias allocation blocks will
> start from lowest address. And, since we are traversing the tree from
> left to right, blocks will be in order.
>
> And, alloc_range() handles actual range allocation demands the same way
> as in the old i915_buddy_alloc_range() function except alloc_range()
> make use of order value to find the blocks within the actual range
> allocation.
Correction - I will change alloc_range() block allocations to bottom-up,
so actual range allocation blocks will start from lowest address (not
the bias thing), and since we are traversing the tree from left to right
progressively within the required range, blocks will be in order (not
the bias thing)
alloc_range() handles actual range allocation demands the same way as in
the old i915_buddy_alloc_range() function except it restarts the search
(since we are handling both end bias and actual range allocations in the
same function) and make use of order value to find the blocks within
start and end value.
Regards,
Arun
^ permalink raw reply [flat|nested] 58+ messages in thread
end of thread, other threads:[~2021-11-15 22:10 UTC | newest]
Thread overview: 58+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-10-25 13:00 [PATCH 3/8] drm: implement top-down allocation method Arunpravin
2021-10-25 13:00 ` [Intel-gfx] " Arunpravin
2021-10-25 13:00 ` [PATCH 4/8] drm/i915: add top-down alloc support to i915 Arunpravin
2021-10-25 13:00 ` [Intel-gfx] " Arunpravin
2021-10-25 13:00 ` [PATCH 5/8] drm: Implement method to free unused pages Arunpravin
2021-10-25 13:00 ` [Intel-gfx] " Arunpravin
2021-11-03 19:16 ` Matthew Auld
2021-11-03 19:16 ` Matthew Auld
2021-11-03 19:16 ` [Intel-gfx] " Matthew Auld
2021-11-09 14:06 ` Arunpravin
2021-11-09 14:06 ` [Intel-gfx] " Arunpravin
2021-11-09 14:06 ` Arunpravin
2021-10-25 13:00 ` [PATCH 6/8] drm/i915: add free_unused_pages support to i915 Arunpravin
2021-10-25 13:00 ` [Intel-gfx] " Arunpravin
2021-11-03 19:18 ` Matthew Auld
2021-11-03 19:18 ` Matthew Auld
2021-11-03 19:18 ` [Intel-gfx] " Matthew Auld
2021-11-08 22:04 ` Arunpravin
2021-11-08 22:04 ` [Intel-gfx] " Arunpravin
2021-11-08 22:04 ` Arunpravin
2021-10-25 13:00 ` [PATCH 7/8] drm/amdgpu: move vram inline functions into a header Arunpravin
2021-10-25 13:00 ` [Intel-gfx] " Arunpravin
2021-10-25 13:00 ` [PATCH 8/8] drm/amdgpu: add drm buddy support to amdgpu Arunpravin
2021-10-25 13:00 ` [Intel-gfx] " Arunpravin
2021-11-03 19:25 ` Matthew Auld
2021-11-03 19:25 ` Matthew Auld
2021-11-03 19:25 ` [Intel-gfx] " Matthew Auld
2021-11-04 7:34 ` Christian König
2021-11-04 7:34 ` Christian König
2021-11-04 7:34 ` [Intel-gfx] " Christian König
2021-11-04 8:49 ` Matthew Auld
2021-11-04 8:49 ` Matthew Auld
2021-11-04 8:49 ` [Intel-gfx] " Matthew Auld
2021-11-04 8:54 ` Christian König
2021-11-04 8:54 ` Christian König
2021-11-04 8:54 ` [Intel-gfx] " Christian König
2021-11-08 22:08 ` Arunpravin
2021-11-08 22:08 ` [Intel-gfx] " Arunpravin
2021-11-08 22:08 ` Arunpravin
2021-10-25 13:00 ` [PATCH v2 1/8] drm: move the buddy allocator from i915 into common drm Arunpravin
2021-10-25 13:00 ` [Intel-gfx] " Arunpravin
2021-10-25 13:00 ` [PATCH v2 2/8] drm: improve drm_buddy_alloc function Arunpravin
2021-10-25 13:00 ` [Intel-gfx] " Arunpravin
2021-11-03 18:41 ` Matthew Auld
2021-11-03 18:41 ` Matthew Auld
2021-11-03 18:41 ` [Intel-gfx] " Matthew Auld
2021-11-08 19:50 ` Arunpravin
2021-11-08 19:50 ` [Intel-gfx] " Arunpravin
2021-11-08 19:50 ` Arunpravin
2021-11-15 21:52 ` Arunpravin
2021-11-15 21:52 ` [Intel-gfx] " Arunpravin
2021-10-25 13:56 ` [Intel-gfx] ✗ Fi.CI.BUILD: failure for series starting with [v2,1/8] drm: move the buddy allocator from i915 into common drm Patchwork
2021-11-03 18:44 ` [PATCH 3/8] drm: implement top-down allocation method Matthew Auld
2021-11-03 18:44 ` Matthew Auld
2021-11-03 18:44 ` [Intel-gfx] " Matthew Auld
2021-11-08 21:45 ` Arunpravin
2021-11-08 21:45 ` [Intel-gfx] " Arunpravin
2021-11-08 21:45 ` Arunpravin
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.