Intel-XE Archive on lore.kernel.org
 help / color / mirror / Atom feed
From: "Christian König" <christian.koenig@amd.com>
To: "Thomas Hellström" <thomas.hellstrom@linux.intel.com>,
	"Nirmoy Das" <nirmoy.das@intel.com>,
	dri-devel@lists.freedesktop.org
Cc: intel-xe@lists.freedesktop.org, Matthew Auld <matthew.auld@intel.com>
Subject: Re: [PATCH v6 1/2] drm/ttm: Add a flag to allow drivers to skip clear-on-free
Date: Tue, 20 Aug 2024 17:47:28 +0200	[thread overview]
Message-ID: <b94e2ea6-569d-47bc-8e01-199b543b2018@amd.com> (raw)
In-Reply-To: <b48456effda2ac845e08318da404e2810c74bd64.camel@linux.intel.com>

Am 20.08.24 um 17:45 schrieb Thomas Hellström:
> On Tue, 2024-08-20 at 17:30 +0200, Christian König wrote:
>> Am 20.08.24 um 15:33 schrieb Thomas Hellström:
>>> Hi, Nirmoy, Christian
>>>
>>> On Fri, 2024-08-16 at 15:51 +0200, Nirmoy Das wrote:
>>>> Add TTM_TT_FLAG_CLEARED_ON_FREE, which DRM drivers can set before
>>>> releasing backing stores if they want to skip clear-on-free.
>>>>
>>>> Cc: Matthew Auld <matthew.auld@intel.com>
>>>> Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
>>>> Suggested-by: Christian König <christian.koenig@amd.com>
>>>> Signed-off-by: Nirmoy Das <nirmoy.das@intel.com>
>>>> Reviewed-by: Christian König <christian.koenig@amd.com>
>>> What happens if two devices share the same global TTM pool
>>>    type and one that does its own clearing. Wouldn't there be a
>>> pretty
>>> high chance that the the device that doesn't clear its own pages
>>> allocate non-cleared memory from the pool?
>> That's completely unproblematic. The flag indicates that the released
>> pages are already cleared, if that isn't the case then the flag
>> shouldn't be set on the TT object.
> Yeah, this patch is OK, but the way the follow-up xe patch uses it is
> problematic since, AFAICT, xe dma clears on alloc, meaning the pool
> pages are not cleared after use.

Yeah that is clearly invalid behavior.

Regards,
Christian.

>
> /Thomas
>
>> If one device clear it's pages and another device doesn't clear it's
>> pages then we would just clear the pages of the device which doesn't
>> do
>> it with a hardware DMA.
>>
>> Regards,
>> Christian.
>>
>>> /Thomas
>>>
>>>> ---
>>>>    drivers/gpu/drm/ttm/ttm_pool.c | 18 +++++++++++-------
>>>>    include/drm/ttm/ttm_tt.h       |  6 +++++-
>>>>    2 files changed, 16 insertions(+), 8 deletions(-)
>>>>
>>>> diff --git a/drivers/gpu/drm/ttm/ttm_pool.c
>>>> b/drivers/gpu/drm/ttm/ttm_pool.c
>>>> index 8504dbe19c1a..935ab3cfd046 100644
>>>> --- a/drivers/gpu/drm/ttm/ttm_pool.c
>>>> +++ b/drivers/gpu/drm/ttm/ttm_pool.c
>>>> @@ -222,15 +222,18 @@ static void ttm_pool_unmap(struct ttm_pool
>>>> *pool, dma_addr_t dma_addr,
>>>>    }
>>>>    
>>>>    /* Give pages into a specific pool_type */
>>>> -static void ttm_pool_type_give(struct ttm_pool_type *pt, struct
>>>> page
>>>> *p)
>>>> +static void ttm_pool_type_give(struct ttm_pool_type *pt, struct
>>>> page
>>>> *p,
>>>> +			       bool cleared)
>>>>    {
>>>>    	unsigned int i, num_pages = 1 << pt->order;
>>>>    
>>>> -	for (i = 0; i < num_pages; ++i) {
>>>> -		if (PageHighMem(p))
>>>> -			clear_highpage(p + i);
>>>> -		else
>>>> -			clear_page(page_address(p + i));
>>>> +	if (!cleared) {
>>>> +		for (i = 0; i < num_pages; ++i) {
>>>> +			if (PageHighMem(p))
>>>> +				clear_highpage(p + i);
>>>> +			else
>>>> +				clear_page(page_address(p + i));
>>>> +		}
>>>>    	}
>>>>    
>>>>    	spin_lock(&pt->lock);
>>>> @@ -394,6 +397,7 @@ static void ttm_pool_free_range(struct
>>>> ttm_pool
>>>> *pool, struct ttm_tt *tt,
>>>>    				pgoff_t start_page, pgoff_t
>>>> end_page)
>>>>    {
>>>>    	struct page **pages = &tt->pages[start_page];
>>>> +	bool cleared = tt->page_flags &
>>>> TTM_TT_FLAG_CLEARED_ON_FREE;
>>>>    	unsigned int order;
>>>>    	pgoff_t i, nr;
>>>>    
>>>> @@ -407,7 +411,7 @@ static void ttm_pool_free_range(struct
>>>> ttm_pool
>>>> *pool, struct ttm_tt *tt,
>>>>    
>>>>    		pt = ttm_pool_select_type(pool, caching, order);
>>>>    		if (pt)
>>>> -			ttm_pool_type_give(pt, *pages);
>>>> +			ttm_pool_type_give(pt, *pages, cleared);
>>>>    		else
>>>>    			ttm_pool_free_page(pool, caching, order,
>>>> *pages);
>>>>    	}
>>>> diff --git a/include/drm/ttm/ttm_tt.h b/include/drm/ttm/ttm_tt.h
>>>> index 2b9d856ff388..cfaf49de2419 100644
>>>> --- a/include/drm/ttm/ttm_tt.h
>>>> +++ b/include/drm/ttm/ttm_tt.h
>>>> @@ -85,6 +85,9 @@ struct ttm_tt {
>>>>    	 * fault handling abuses the DMA api a bit and
>>>> dma_map_attrs
>>>> can't be
>>>>    	 * used to assure pgprot always matches.
>>>>    	 *
>>>> +	 * TTM_TT_FLAG_CLEARED_ON_FREE: Set this if a drm driver
>>>> handles
>>>> +	 * clearing backing store
>>>> +	 *
>>>>    	 * TTM_TT_FLAG_PRIV_POPULATED: TTM internal only. DO NOT
>>>> USE. This is
>>>>    	 * set by TTM after ttm_tt_populate() has successfully
>>>> returned, and is
>>>>    	 * then unset when TTM calls ttm_tt_unpopulate().
>>>> @@ -94,8 +97,9 @@ struct ttm_tt {
>>>>    #define TTM_TT_FLAG_EXTERNAL		BIT(2)
>>>>    #define TTM_TT_FLAG_EXTERNAL_MAPPABLE	BIT(3)
>>>>    #define TTM_TT_FLAG_DECRYPTED		BIT(4)
>>>> +#define TTM_TT_FLAG_CLEARED_ON_FREE	BIT(5)
>>>>    
>>>> -#define TTM_TT_FLAG_PRIV_POPULATED	BIT(5)
>>>> +#define TTM_TT_FLAG_PRIV_POPULATED	BIT(6)
>>>>    	uint32_t page_flags;
>>>>    	/** @num_pages: Number of pages in the page array. */
>>>>    	uint32_t num_pages;


  reply	other threads:[~2024-08-20 15:47 UTC|newest]

Thread overview: 22+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2024-08-16 13:51 [PATCH v6 1/2] drm/ttm: Add a flag to allow drivers to skip clear-on-free Nirmoy Das
2024-08-16 13:51 ` [PATCH v6 2/2] drm/xe/lnl: Offload system clear page activity to GPU Nirmoy Das
2024-08-19 11:05   ` Matthew Auld
2024-08-19 16:01     ` Nirmoy Das
2024-08-20 13:36       ` Thomas Hellström
2024-08-16 14:25 ` ✓ CI.Patch_applied: success for series starting with [v6,1/2] drm/ttm: Add a flag to allow drivers to skip clear-on-free Patchwork
2024-08-16 14:25 ` ✓ CI.checkpatch: " Patchwork
2024-08-16 14:27 ` ✓ CI.KUnit: " Patchwork
2024-08-16 14:38 ` ✓ CI.Build: " Patchwork
2024-08-16 14:40 ` ✓ CI.Hooks: " Patchwork
2024-08-16 14:42 ` ✗ CI.checksparse: warning " Patchwork
2024-08-16 14:57 ` ✓ CI.BAT: success " Patchwork
2024-08-17  0:39 ` ✓ CI.FULL: " Patchwork
2024-08-20 13:33 ` [PATCH v6 1/2] " Thomas Hellström
2024-08-20 14:06   ` Nirmoy Das
2024-08-20 15:30   ` Christian König
2024-08-20 15:45     ` Thomas Hellström
2024-08-20 15:47       ` Christian König [this message]
2024-08-20 16:46         ` Nirmoy Das
2024-08-21  7:47           ` Christian König
2024-08-21  8:08             ` Thomas Hellström
2024-08-21 10:22               ` Nirmoy Das

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=b94e2ea6-569d-47bc-8e01-199b543b2018@amd.com \
    --to=christian.koenig@amd.com \
    --cc=dri-devel@lists.freedesktop.org \
    --cc=intel-xe@lists.freedesktop.org \
    --cc=matthew.auld@intel.com \
    --cc=nirmoy.das@intel.com \
    --cc=thomas.hellstrom@linux.intel.com \
    /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