RE: [PATCH v4 4/6] drm: implement a method to free unused pages

["Paneer Selvam, Arunpravin" <Arunpravin.PaneerSelvam@amd.com>]

[Public]

Hi Matthew,

Ping?

Regards,
Arun
-----Original Message-----
From: Paneer Selvam, Arunpravin <Arunpravin.PaneerSelvam@amd.com> 
Sent: Wednesday, December 1, 2021 10:10 PM
To: dri-devel@lists.freedesktop.org; intel-gfx@lists.freedesktop.org; amd-gfx@lists.freedesktop.org
Cc: matthew.auld@intel.com; daniel@ffwll.ch; Koenig, Christian <Christian.Koenig@amd.com>; Deucher, Alexander <Alexander.Deucher@amd.com>; tzimmermann@suse.de; jani.nikula@linux.intel.com; Paneer Selvam, Arunpravin <Arunpravin.PaneerSelvam@amd.com>
Subject: [PATCH v4 4/6] drm: implement a method to free unused pages

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.

v2(Matthew Auld):
  - replace function name 'drm_buddy_free_unused_pages' with
    drm_buddy_block_trim
  - replace input argument name 'actual_size' with 'new_size'
  - add more validation checks for input arguments
  - add overlaps check to avoid needless searching and splitting
  - merged the below patch to see the feature in action
    - add free unused pages support to i915 driver
  - lock drm_buddy_block_trim() function as it calls mark_free/mark_split
    are all globally visible

v3:
  - remove drm_buddy_block_trim() error handling and
    print a warn message if it fails

Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
---
 drivers/gpu/drm/drm_buddy.c                   | 72 ++++++++++++++++++-
 drivers/gpu/drm/i915/i915_ttm_buddy_manager.c | 10 +++
 include/drm/drm_buddy.h                       |  4 ++
 3 files changed, 83 insertions(+), 3 deletions(-)

diff --git a/drivers/gpu/drm/drm_buddy.c b/drivers/gpu/drm/drm_buddy.c index eddc1eeda02e..707efc82216d 100644
--- a/drivers/gpu/drm/drm_buddy.c
+++ b/drivers/gpu/drm/drm_buddy.c
@@ -434,7 +434,8 @@ alloc_from_freelist(struct drm_buddy_mm *mm,  static int __alloc_range(struct drm_buddy_mm *mm,
 			 struct list_head *dfs,
 			 u64 start, u64 size,
-			 struct list_head *blocks)
+			 struct list_head *blocks,
+			 bool trim_path)
 {
 	struct drm_buddy_block *block;
 	struct drm_buddy_block *buddy;
@@ -480,8 +481,20 @@ static int __alloc_range(struct drm_buddy_mm *mm,
 
 		if (!drm_buddy_block_is_split(block)) {
 			err = split_block(mm, block);
-			if (unlikely(err))
+			if (unlikely(err)) {
+				if (trim_path)
+					/*
+					 * Here in case of trim, we return and dont goto
+					 * split failure path as it removes from the
+					 * original list and potentially also freeing
+					 * the block. so we could leave as it is,
+					 * worse case we get some internal fragmentation
+					 * and leave the decision to the user
+					 */
+					return err;
+
 				goto err_undo;
+			}
 		}
 
 		list_add(&block->right->tmp_link, dfs); @@ -535,8 +548,61 @@ static int __drm_buddy_alloc_range(struct drm_buddy_mm *mm,
 	for (i = 0; i < mm->n_roots; ++i)
 		list_add_tail(&mm->roots[i]->tmp_link, &dfs);
 
-	return __alloc_range(mm, &dfs, start, size, blocks);
+	return __alloc_range(mm, &dfs, start, size, blocks, 0); }
+
+/**
+ * drm_buddy_block_trim - free unused pages
+ *
+ * @mm: DRM buddy manager
+ * @new_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_block_trim(struct drm_buddy_mm *mm,
+			 u64 new_size,
+			 struct list_head *blocks)
+{
+	struct drm_buddy_block *block;
+	u64 new_start;
+	LIST_HEAD(dfs);
+
+	if (!list_is_singular(blocks))
+		return -EINVAL;
+
+	block = list_first_entry(blocks,
+				 struct drm_buddy_block,
+				 link);
+
+	if (!drm_buddy_block_is_allocated(block))
+		return -EINVAL;
+
+	if (new_size > drm_buddy_block_size(mm, block))
+		return -EINVAL;
+
+	if (!new_size && !IS_ALIGNED(new_size, mm->chunk_size))
+		return -EINVAL;
+
+	if (new_size == drm_buddy_block_size(mm, block))
+		return 0;
+
+	list_del(&block->link);
+
+	new_start = drm_buddy_block_offset(block);
+
+	mark_free(mm, block);
+
+	list_add(&block->tmp_link, &dfs);
+
+	return __alloc_range(mm, &dfs, new_start, new_size, blocks, 1);
 }
+EXPORT_SYMBOL(drm_buddy_block_trim);
 
 /**
  * drm_buddy_alloc - allocate power-of-two blocks diff --git a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
index 7c58efb60dba..c5831c27fe82 100644
--- a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
+++ b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
@@ -97,6 +97,16 @@ 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) {
+		mutex_lock(&bman->lock);
+		err = drm_buddy_block_trim(mm,
+				(u64)n_pages << PAGE_SHIFT,
+				&bman_res->blocks);
+		mutex_unlock(&bman->lock);
+		pr_warn("drm_buddy_block_trim failed returing %d for ttm_buffer_object(%p)\n",
+			err, bo);
+	}
+
 	*res = &bman_res->base;
 	return 0;
 
diff --git a/include/drm/drm_buddy.h b/include/drm/drm_buddy.h index 316ac0d25f08..90906d9dbbf0 100644
--- a/include/drm/drm_buddy.h
+++ b/include/drm/drm_buddy.h
@@ -146,6 +146,10 @@ int drm_buddy_alloc(struct drm_buddy_mm *mm,
 		    struct list_head *blocks,
 		    unsigned long flags);
 
+int drm_buddy_block_trim(struct drm_buddy_mm *mm,
+			 u64 new_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