Re: [PATCH v2 1/2] drm/buddy: Add kernel-doc for allocator structures and flags

[Arunpravin Paneer Selvam <arunpravin.paneerselvam@amd.com>]

On 2/12/2026 1:08 PM, Arunpravin Paneer Selvam wrote:
>
>
> On 2/11/2026 11:01 AM, Sanjay Yadav wrote:
>> Add missing kernel-doc for GPU buddy allocator flags,
>> gpu_buddy_block, and gpu_buddy. The documentation covers block
>> header fields, allocator roots, free trees, and allocation flags
>> such as RANGE, TOPDOWN, CONTIGUOUS, CLEAR, and TRIM_DISABLE.
>> Private members are marked with kernel-doc private markers
>> and documented with regular comments.
>>
>> No functional changes.
>>
>> v2:
>> - Corrected GPU_BUDDY_CLEAR_TREE and GPU_BUDDY_DIRTY_TREE index values (Arun)
>> - Rebased after DRM buddy allocator moved to drivers/gpu/
>> - Updated commit message
>>
>> Cc: Christian König<christian.koenig@amd.com>
>> Cc: Arunpravin Paneer Selvam<Arunpravin.PaneerSelvam@amd.com>
>> Suggested-by: Matthew Auld<matthew.auld@intel.com>
>> Signed-off-by: Sanjay Yadav<sanjay.kumar.yadav@intel.com>
>> ---
>>   include/linux/gpu_buddy.h | 122 +++++++++++++++++++++++++++++++-------
>>   1 file changed, 102 insertions(+), 20 deletions(-)
>>
>> diff --git a/include/linux/gpu_buddy.h b/include/linux/gpu_buddy.h
>> index 07ac65db6d2e..b06241c78437 100644
>> --- a/include/linux/gpu_buddy.h
>> +++ b/include/linux/gpu_buddy.h
>> @@ -12,11 +12,58 @@
>>   #include <linux/sched.h>
>>   #include <linux/rbtree.h>
>>   
>> +/**
>> + * GPU_BUDDY_RANGE_ALLOCATION - Allocate within a specific address range
>> + *
>> + * When set, allocation is restricted to the range [start, end) specified
>> + * in gpu_buddy_alloc_blocks(). Without this flag, start/end are ignored
>> + * and allocation can use any free space.
>> + */
>>   #define GPU_BUDDY_RANGE_ALLOCATION		BIT(0)
>> +
>> +/**
>> + * GPU_BUDDY_TOPDOWN_ALLOCATION - Allocate from top of address space
>> + *
>> + * Allocate starting from high addresses and working down. Useful for
>> + * separating different allocation types (e.g., kernel vs userspace)
>> + * to reduce fragmentation.
>> + */
>>   #define GPU_BUDDY_TOPDOWN_ALLOCATION		BIT(1)
>> +
>> +/**
>> + * GPU_BUDDY_CONTIGUOUS_ALLOCATION - Require physically contiguous blocks
>> + *
>> + * The allocation must be satisfied with a single contiguous block.
>> + * If the requested size cannot be allocated contiguously, the
>> + * allocation fails with -ENOSPC.
>> + */
>>   #define GPU_BUDDY_CONTIGUOUS_ALLOCATION		BIT(2)
>> +
>> +/**
>> + * GPU_BUDDY_CLEAR_ALLOCATION - Prefer pre-cleared (zeroed) memory
>> + *
>> + * Attempt to allocate from the clear tree first. If insufficient clear
>> + * memory is available, falls back to dirty memory. Useful when the
>> + * caller needs zeroed memory and wants to avoid GPU clear operations.
>> + */
>>   #define GPU_BUDDY_CLEAR_ALLOCATION		BIT(3)
>> +
>> +/**
>> + * GPU_BUDDY_CLEARED - Mark returned blocks as cleared
>> + *
>> + * Used with gpu_buddy_free_list() to indicate that the memory being
>> + * freed has been cleared (zeroed). The blocks will be placed in the
>> + * clear tree for future GPU_BUDDY_CLEAR_ALLOCATION requests.
>> + */
>>   #define GPU_BUDDY_CLEARED			BIT(4)
>> +
>> +/**
>> + * GPU_BUDDY_TRIM_DISABLE - Disable automatic block trimming
>> + *
>> + * By default, if an allocation is smaller than the allocated block,
>> + * excess memory is trimmed and returned to the free pool. This flag
>> + * disables trimming, keeping the full power-of-two block size.
>> + */
>>   #define GPU_BUDDY_TRIM_DISABLE			BIT(5)
>>   
>>   enum gpu_buddy_free_tree {
>> @@ -28,7 +75,27 @@ enum gpu_buddy_free_tree {
>>   #define for_each_free_tree(tree) \
>>   	for ((tree) = 0; (tree) < GPU_BUDDY_MAX_FREE_TREES; (tree)++)
>>   
>> +/**
>> + * struct gpu_buddy_block - Block within a buddy allocator
>> + *
>> + * Each block in the buddy allocator is represented by this structure.
>> + * Blocks are organized in a binary tree where each parent block can be
>> + * split into two children (left and right buddies). The allocator manages
>> + * blocks at various orders (power-of-2 sizes) from chunk_size up to the
>> + * largest contiguous region.
>> + *
>> + * @private: Private data owned by the allocator user (e.g., driver-specific data)
>> + * @link: List node for user ownership while block is allocated
>> + */
>>   struct gpu_buddy_block {
>> +/* private: */
>> +	/*
>> +	 * Header bit layout:
>> +	 * - Bits 63:12: block offset within the address space
>> +	 * - Bits 11:10: state (ALLOCATED, FREE, or SPLIT)
>> +	 * - Bit 9: clear bit (1 if memory is zeroed)
> I think we should add
> - Bits 8:9: reserved
it should be -Bits 8:6: reserved
>
> Apart from that, looks good to me.
> Reviewed-by: Arunpravin Paneer Selvam <Arunpravin.PaneerSelvam@amd.com>
>> +	 * - Bits 5:0: order (log2 of size relative to chunk_size)
>> +	 */
>>   #define GPU_BUDDY_HEADER_OFFSET GENMASK_ULL(63, 12)
>>   #define GPU_BUDDY_HEADER_STATE  GENMASK_ULL(11, 10)
>>   #define   GPU_BUDDY_ALLOCATED	   (1 << 10)
>> @@ -43,7 +110,7 @@ struct gpu_buddy_block {
>>   	struct gpu_buddy_block *left;
>>   	struct gpu_buddy_block *right;
>>   	struct gpu_buddy_block *parent;
>> -
>> +/* public: */
>>   	void *private; /* owned by creator */
>>   
>>   	/*
>> @@ -53,43 +120,58 @@ struct gpu_buddy_block {
>>   	 * gpu_buddy_free* ownership is given back to the mm.
>>   	 */
>>   	union {
>> +/* private: */
>>   		struct rb_node rb;
>> +/* public: */
>>   		struct list_head link;
>>   	};
>> -
>> +/* private: */
>>   	struct list_head tmp_link;
>>   };
>>   
>>   /* Order-zero must be at least SZ_4K */
>>   #define GPU_BUDDY_MAX_ORDER (63 - 12)
>>   
>> -/*
>> - * Binary Buddy System.
>> +/**
>> + * struct gpu_buddy - GPU binary buddy allocator
>> + *
>> + * The buddy allocator provides efficient power-of-two memory allocation
>> + * with fast allocation and free operations. It is commonly used for GPU
>> + * memory management where allocations can be split into power-of-two
>> + * block sizes.
>>    *
>> - * Locking should be handled by the user, a simple mutex around
>> - * gpu_buddy_alloc* and gpu_buddy_free* should suffice.
>> + * Locking should be handled by the user; a simple mutex around
>> + * gpu_buddy_alloc_blocks() and gpu_buddy_free_block()/gpu_buddy_free_list()
>> + * should suffice.
>> + *
>> + * @n_roots: Number of root blocks in the roots array.
>> + * @max_order: Maximum block order (log2 of largest block size / chunk_size).
>> + * @chunk_size: Minimum allocation granularity in bytes. Must be at least SZ_4K.
>> + * @size: Total size of the address space managed by this allocator in bytes.
>> + * @avail: Total free space currently available for allocation in bytes.
>> + * @clear_avail: Free space available in the clear tree (zeroed memory) in bytes.
>> + *               This is a subset of @avail.
>>    */
>>   struct gpu_buddy {
>> -	/* Maintain a free list for each order. */
>> -	struct rb_root **free_trees;
>> -
>> +/* private: */
>>   	/*
>> -	 * 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.
>> +	 * Array of red-black trees for free block management.
>> +	 * Indexed as free_trees[clear/dirty][order] where:
>> +	 * - Index 0 (GPU_BUDDY_CLEAR_TREE): blocks with zeroed content
>> +	 * - Index 1 (GPU_BUDDY_DIRTY_TREE): blocks with unknown content
>> +	 * Each tree holds free blocks of the corresponding order.
>>   	 */
>> -	struct gpu_buddy_block **roots;
>> -
>> +	struct rb_root **free_trees;
>>   	/*
>> -	 * Anything from here is public, and remains static for the lifetime of
>> -	 * the mm. Everything above is considered do-not-touch.
>> +	 * Array of root blocks representing the top-level blocks of the
>> +	 * binary tree(s). Multiple roots exist when the total size is not
>> +	 * a power of two, with each root being the largest power-of-two
>> +	 * that fits in the remaining space.
>>   	 */
>> +	struct gpu_buddy_block **roots;
>> +/* public: */
>>   	unsigned int n_roots;
>>   	unsigned int max_order;
>> -
>> -	/* Must be at least SZ_4K */
>>   	u64 chunk_size;
>>   	u64 size;
>>   	u64 avail;
>