From: Arunpravin <arunpravin.paneerselvam@amd.com>
To: Matthew Auld <matthew.auld@intel.com>,
dri-devel@lists.freedesktop.org, intel-gfx@lists.freedesktop.org,
amd-gfx@lists.freedesktop.org
Cc: alexander.deucher@amd.com, tzimmermann@suse.de, christian.koenig@amd.com
Subject: Re: [PATCH v6 4/6] drm: implement a method to free unused pages
Date: Thu, 6 Jan 2022 23:21:04 +0530 [thread overview]
Message-ID: <aae8b9f0-c5c1-b303-1bc0-5b9c14eb6765@amd.com> (raw)
In-Reply-To: <9350846b-26bb-cfd6-6c9e-af109ad8b7bb@intel.com>
On 04/01/22 7:41 pm, Matthew Auld wrote:
> On 26/12/2021 22:24, 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
>>
>> v3(Matthew Auld):
>> - remove trim method error handling as we address the failure case
>> at drm_buddy_block_trim() function
>>
>> v4:
>> - in case of trim, at __alloc_range() split_block failure path
>> marks the block as free and removes it from the original list,
>> potentially also freeing it, to overcome this problem, we turn
>> the drm_buddy_block_trim() input node into a temporary node to
>> prevent recursively freeing itself, but still retain the
>> un-splitting/freeing of the other nodes(Matthew Auld)
>>
>> - modify the drm_buddy_block_trim() function return type
>>
>> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
>> ---
>> drivers/gpu/drm/drm_buddy.c | 61 +++++++++++++++++++
>> drivers/gpu/drm/i915/i915_ttm_buddy_manager.c | 8 +++
>> include/drm/drm_buddy.h | 4 ++
>> 3 files changed, 73 insertions(+)
>>
>> diff --git a/drivers/gpu/drm/drm_buddy.c b/drivers/gpu/drm/drm_buddy.c
>> index eddc1eeda02e..855afcaf7edd 100644
>> --- a/drivers/gpu/drm/drm_buddy.c
>> +++ b/drivers/gpu/drm/drm_buddy.c
>> @@ -538,6 +538,67 @@ static int __drm_buddy_alloc_range(struct drm_buddy_mm *mm,
>> return __alloc_range(mm, &dfs, start, size, blocks);
>> }
>>
>> +/**
>> + * 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.
>> + */
>> +void drm_buddy_block_trim(struct drm_buddy_mm *mm,
>> + u64 new_size,
>> + struct list_head *blocks)
>
> It might be better to just return the error, and let the user decide if
> they want to ignore it? Also we might want some kind of unit test for
> this, so having an actual return value might be useful there.
>
sure will revert it
>> +{
>> + struct drm_buddy_block *parent;
>> + struct drm_buddy_block *block;
>> + LIST_HEAD(dfs);
>> + u64 new_start;
>> + int err;
>> +
>> + if (!list_is_singular(blocks))
>> + return;
>> +
>> + block = list_first_entry(blocks,
>> + struct drm_buddy_block,
>> + link);
>> +
>> + if (!drm_buddy_block_is_allocated(block))
>> + return;
>> +
>> + if (new_size > drm_buddy_block_size(mm, block))
>> + return;
>> +
>> + if (!new_size && !IS_ALIGNED(new_size, mm->chunk_size))
>> + return;
>> +
>> + if (new_size == drm_buddy_block_size(mm, block))
>> + return;
>> +
>> + list_del(&block->link);
>> + mark_free(mm, block);
>> + mm->avail += drm_buddy_block_size(mm, block);
>> +
>> + /* Prevent recursively freeing this node */
>> + parent = block->parent;
>> + block->parent = NULL;
>> +
>> + new_start = drm_buddy_block_offset(block);
>> + list_add(&block->tmp_link, &dfs);
>> + err = __alloc_range(mm, &dfs, new_start, new_size, blocks);
>> + if (err) {
>> + mark_allocated(block);
>> + mm->avail -= drm_buddy_block_size(mm, block);
>> + list_add(&block->link, blocks);
>> + }
>> +
>> + block->parent = parent;
>> +}
>> +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..05f924f32e96 100644
>> --- a/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
>> +++ b/drivers/gpu/drm/i915/i915_ttm_buddy_manager.c
>> @@ -97,6 +97,14 @@ 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) {
>> + mutex_lock(&bman->lock);
>> + drm_buddy_block_trim(mm,
>> + (u64)n_pages << PAGE_SHIFT,
>
> AFAIK, n_pages has already been rounded up to the next power-of-two
> here, so this becomes a noop. I assume we need to use the "original"
> size here?
>
ah yes..missed it, will change as below
u64 original_size = (u64)bman_res->base.num_pages << PAGE_SHIFT;
mutex_lock(&bman->lock);
drm_buddy_block_trim(mm,
original_size,
&bman_res->blocks);
>> + &bman_res->blocks);
>> + mutex_unlock(&bman->lock);
>> + }
>> +
>> *res = &bman_res->base;
>> return 0;
>>
>> diff --git a/include/drm/drm_buddy.h b/include/drm/drm_buddy.h
>> index f573b02304f4..703866a87939 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);
>>
>> +void 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);
>>
next prev parent reply other threads:[~2022-01-06 17:42 UTC|newest]
Thread overview: 13+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-12-26 22:24 [PATCH v6 1/6] drm: move the buddy allocator from i915 into common drm Arunpravin
2021-12-26 22:24 ` [PATCH v6 2/6] drm: improve drm_buddy_alloc function Arunpravin
2022-01-04 14:01 ` Matthew Auld
[not found] ` <MN2PR12MB4342A3191114E4D3F2824441E44C9@MN2PR12MB4342.namprd12.prod.outlook.com>
2022-01-06 21:23 ` Arunpravin
2021-12-26 22:24 ` [PATCH v6 3/6] drm: implement top-down allocation method Arunpravin
2021-12-26 22:24 ` [PATCH v6 4/6] drm: implement a method to free unused pages Arunpravin
2022-01-04 14:11 ` Matthew Auld
2022-01-06 17:51 ` Arunpravin [this message]
2021-12-26 22:24 ` [PATCH v6 5/6] drm/amdgpu: move vram inline functions into a header Arunpravin
2021-12-26 22:24 ` [PATCH v6 6/6] drm/amdgpu: add drm buddy support to amdgpu Arunpravin
2022-01-07 15:49 ` [PATCH v6 1/6] drm: move the buddy allocator from i915 into common drm Matthew Auld
2022-01-07 15:57 ` Christian König
2022-01-09 10:22 ` Arunpravin
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=aae8b9f0-c5c1-b303-1bc0-5b9c14eb6765@amd.com \
--to=arunpravin.paneerselvam@amd.com \
--cc=alexander.deucher@amd.com \
--cc=amd-gfx@lists.freedesktop.org \
--cc=christian.koenig@amd.com \
--cc=dri-devel@lists.freedesktop.org \
--cc=intel-gfx@lists.freedesktop.org \
--cc=matthew.auld@intel.com \
--cc=tzimmermann@suse.de \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).