Re: [Intel-gfx] [PATCH v3 4/6] drm: implement a method to free unused pages

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

On 18/11/21 12:32 am, Matthew Auld wrote:
> On 16/11/2021 20:18, 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.
>>
>> 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
>>
>> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
>> ---
>>   drivers/gpu/drm/drm_buddy.c                   | 108 ++++++++++++++++++
>>   drivers/gpu/drm/i915/i915_ttm_buddy_manager.c |  10 ++
>>   include/drm/drm_buddy.h                       |   4 +
>>   3 files changed, 122 insertions(+)
>>
>> diff --git a/drivers/gpu/drm/drm_buddy.c b/drivers/gpu/drm/drm_buddy.c
>> index 0a9db2981188..943fe2ad27bf 100644
>> --- a/drivers/gpu/drm/drm_buddy.c
>> +++ b/drivers/gpu/drm/drm_buddy.c
>> @@ -284,6 +284,114 @@ static inline bool contains(u64 s1, u64 e1, u64 s2, u64 e2)
>>   	return s1 <= s2 && e1 >= e2;
>>   }
>>   
>> +/**
>> + * 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;
>> +	struct drm_buddy_block *buddy;
>> +	u64 new_start;
>> +	u64 new_end;
>> +	LIST_HEAD(dfs);
>> +	u64 count = 0;
>> +	int err;
>> +
>> +	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);
>> +	new_end = new_start + new_size - 1;
>> +
>> +	mark_free(mm, block);
>> +
>> +	list_add(&block->tmp_link, &dfs);
>> +
>> +	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);
>> +
>> +		if (count == new_size)
>> +			return 0;
>> +
>> +		block_start = drm_buddy_block_offset(block);
>> +		block_end = block_start + drm_buddy_block_size(mm, block) - 1;
>> +
>> +		if (!overlaps(new_start, new_end, block_start, block_end))
>> +			continue;
>> +
>> +		if (contains(new_start, new_end, block_start, block_end)) {
>> +			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_is_split(block)) {
> 
> Should always be true, right? But I guess depends if we want to re-use 
> this for generic range allocation...
yes, since we re-use this for generic range allocation I think we can
keep this check
> 
>> +			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);
>> +
>> +	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;
> 
> Looking at the split_block failure path. The user allocated some block, 
> and then tried to trim it, but this then marks it as free and removes it 
> from their original list(potentially also freeing it), if we fail here. 
> Would it be better to leave that decision to the user? If the trim() 
> fails, worse case we get some internal fragmentation, and the user might 
> be totally fine with that? I guess we could leave as-is, but for sure 
> needs some big comment somewhere.

Agreed

would it make sense to add a bool type input argument, so that we can
skip the split_block failure path in case of block_trim().

__alloc_range(mm, &dfs, start, size, ... bool trim)

err = split_block(mm, block);
if (unlikely(err)) {
	/* I think we can add big comment here for trim case */
	if (trim)
		return err;

goto err_undo;
}

> 
>> +}
>> +EXPORT_SYMBOL(drm_buddy_block_trim);
>> +
>>   static struct drm_buddy_block *
>>   alloc_range(struct drm_buddy_mm *mm,
>>   	    u64 start, u64 end,
>> diff --git a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
>> index b6ed5b37cf18..5e1f8c443058 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);
>> +		if (unlikely(err))
>> +			goto err_free_blocks;
> 
> Yeah, so maybe we could in theory treat this as best effort? But I guess 
> unlikely to ever hit this anyway, so meh.
so I understood that we remove the below 2 lines
if (unlikely(err))
	goto err_free_blocks;

would it make sense to print a WARN msg
WARN(err < 0, "Trim failed returning %d for ttm_resource(%llu)\n",
     err, &bman_res->base);
> 
>> +	}
>> +
>>   	*res = &bman_res->base;
>>   	return 0;
>>   
>> diff --git a/include/drm/drm_buddy.h b/include/drm/drm_buddy.h
>> index cbe5e648440a..36e8f548acf7 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_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);
>>