[RFC PATCH] drm/xe/lnl: Implement clear-on-free for pooled BOs
Matthew Auld
matthew.auld at intel.com
Tue Aug 27 09:04:39 UTC 2024
On 26/08/2024 09:36, Thomas Hellström wrote:
> On Mon, 2024-08-26 at 10:26 +0200, Nirmoy Das wrote:
>> Hi Thomas,
>>
>> On 8/23/2024 11:38 AM, Thomas Hellström wrote:
>>> Hi, Nirmoy,
>>>
>>> On Thu, 2024-08-22 at 14:42 +0200, Nirmoy Das wrote:
>>>> Implement GPU clear-on-free for pooled system pages in Xe.
>>>>
>>>> Ensure proper use of TTM_TT_FLAG_CLEARED_ON_FREE by leveraging
>>>> ttm_device_funcs.release_notify() for GPU clear-on-free. If GPU
>>>> clear
>>>> fails, xe_ttm_tt_unpopulate() will fallback to CPU clear.
>>>>
>>>> Clear-on-free is only relevant for pooled pages as driver needs
>>>> to
>>>> give
>>>> back those pages. So do clear-on-free only for such BOs and keep
>>>> doing
>>>> clear-on-alloc for ttm_cached type BOs
>>>>
>>>> Cc: Matthew Auld <matthew.auld at intel.com>
>>>> Cc: Matthew Brost <matthew.brost at intel.com>
>>>> Cc: Thomas Hellström <thomas.hellstrom at linux.intel.com>
>>>> Signed-off-by: Nirmoy Das <nirmoy.das at intel.com>
>>> While this would probably work, I don't immediately see the benefit
>>> over CPU clearing, since we have no way of combining this with the
>>> CCS
>>> clear, right?
>>
>>
>> If XE/ttm could do clear-on-free(data+CCS) with GPU all the time then
>> I
>> think we could
>>
>> skip ccs clearing on alloc, assuming only GPU access modifies a CCS
>> state and on boot CCS region
>>
>> is zeroed. I think that can't be guaranteed so we have to clear ccs
>> on
>> alloc. I agree, there won't be much
>>
>> latency benefit of doing clear-on-free for ccs devices. I will still
>> try
>> to run some tests to validate it, I have done that for this RFC.
>
> OK, yes this would probably work. Do we need to clear all CCS on module
> load or can we safely assume that no useful info is left in the CCS
> memory at that time?
Maybe another thing to check is if we can skip ccs clearing for wb,
assuming we don't already. I think compression is always coh_none (not
compatible with wb), and I think that to copy the ccs directly would
require mapping the src with compression enabled.
I'm pretty sure that is the case, otherwire userptr/imported external
dma-buf would be an issue today, assuming user could manually copy the
uncleared ccs.
>
>>
>>
>> I've discussed this with Ron and it seems there is on going
>> conversation
>> if there is a way to avoid ccs clearing if data is zeroed.
>>
>> Let's see how that goes.
>>
>>
>>> So the clearing latency will most probably be increased,
>>> but the bo releasing thread won't see that because the waiting for
>>> clear is offloaded to the TTM delayed destroy mechanism.
>>>
>>> Also, once we've dropped the gem refcount to zero, the gem members
>>> of
>>> the object, including bo_move, are strictly not valid anymore and
>>> shouldn't be used.
>>
>>
>> Could you please expand this? I am not seeing the connection between
>> bo_move and refcount.
>>
>> Are you saying release_notify is not the right place to do this ?
>
> Yes. At release_notify, the gem refcount has dropped to zero, and we
> don't allow calling bo_move at that point, as the driver might want to
> do some cleanup in the gem_release before putting the last ttm_bo
> reference.
>
> Thanks,
> Thomas
>
>
>>
>>> If we want to try to improve freeing latency by offloading the
>>> clearing
>>> on free to a separate CPU thread, though, maybe we could discuss
>>> with
>>> Christian to always (or if a flag in the ttm device requests it)
>>> take
>>> the TTM delayed destruction path for bos with pooled pages, rather
>>> than
>>> to free them sync, something along the lines of:
>>>
>>> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c
>>> b/drivers/gpu/drm/ttm/ttm_bo.c
>>> index 320592435252..fca69ec1740d 100644
>>> --- a/drivers/gpu/drm/ttm/ttm_bo.c
>>> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
>>> @@ -271,7 +271,7 @@ static void ttm_bo_release(struct kref *kref)
>>>
>>> if (!dma_resv_test_signaled(bo->base.resv,
>>>
>>> DMA_RESV_USAGE_BOOKKEEP) ||
>>> - (want_init_on_free() && (bo->ttm != NULL)) ||
>>> + (bo->ttm && (want_init_on_free() || bo->ttm-
>>>> caching != ttm_cached)) ||
>>> bo->type == ttm_bo_type_sg ||
>>> !dma_resv_trylock(bo->base.resv)) {
>>> /* The BO is not idle, resurrect it for
>>> delayed
>>> destroy */
>>>
>>> Would ofc require some substantial proven latency gain, though.
>>> Overall
>>> system cpu usage would probably not improve.
>>
>>
>> I will run some tests with the above change and get back.
>>
>>
>> Thanks,
>>
>> Nirmoy
>>
>>>
>>> /Thomas
>>>
>>>
>>>> ---
>>>> drivers/gpu/drm/xe/xe_bo.c | 101
>>>> +++++++++++++++++++++++++++++++++--
>>>> --
>>>> 1 file changed, 91 insertions(+), 10 deletions(-)
>>>>
>>>> diff --git a/drivers/gpu/drm/xe/xe_bo.c
>>>> b/drivers/gpu/drm/xe/xe_bo.c
>>>> index 6ed0e1955215..e7bc74f8ae82 100644
>>>> --- a/drivers/gpu/drm/xe/xe_bo.c
>>>> +++ b/drivers/gpu/drm/xe/xe_bo.c
>>>> @@ -283,6 +283,8 @@ struct xe_ttm_tt {
>>>> struct device *dev;
>>>> struct sg_table sgt;
>>>> struct sg_table *sg;
>>>> + bool sys_clear_on_free;
>>>> + bool sys_clear_on_alloc;
>>>> };
>>>>
>>>> static int xe_tt_map_sg(struct ttm_tt *tt)
>>>> @@ -401,8 +403,23 @@ static struct ttm_tt
>>>> *xe_ttm_tt_create(struct
>>>> ttm_buffer_object *ttm_bo,
>>>> * flag. Zeroed pages are only required for
>>>> ttm_bo_type_device so
>>>> * unwanted data is not leaked to userspace.
>>>> */
>>>> - if (ttm_bo->type == ttm_bo_type_device && xe-
>>>>> mem.gpu_page_clear_sys)
>>>> - page_flags |= TTM_TT_FLAG_CLEARED_ON_FREE;
>>>> + if (ttm_bo->type == ttm_bo_type_device && xe-
>>>>> mem.gpu_page_clear_sys) {
>>>> + /*
>>>> + * Non-pooled BOs are always clear on alloc when
>>>> possible.
>>>> + * clear-on-free is not needed as there is no
>>>> pool
>>>> to give pages back.
>>>> + */
>>>> + if (caching == ttm_cached) {
>>>> + tt->sys_clear_on_alloc = true;
>>>> + tt->sys_clear_on_free = false;
>>>> + } else {
>>>> + /*
>>>> + * For pooled BO, clear-on-alloc is done by the
>>>> CPU
>>>> for now and
>>>> + * GPU will do clear on free when releasing the
>>>> BO.
>>>> + */
>>>> + tt->sys_clear_on_alloc = false;
>>>> + tt->sys_clear_on_free = true;
>>>> + }
>>>> + }
>>>>
>>>> err = ttm_tt_init(&tt->ttm, &bo->ttm, page_flags,
>>>> caching,
>>>> extra_pages);
>>>> if (err) {
>>>> @@ -416,8 +433,10 @@ static struct ttm_tt
>>>> *xe_ttm_tt_create(struct
>>>> ttm_buffer_object *ttm_bo,
>>>> static int xe_ttm_tt_populate(struct ttm_device *ttm_dev,
>>>> struct
>>>> ttm_tt *tt,
>>>> struct ttm_operation_ctx *ctx)
>>>> {
>>>> + struct xe_ttm_tt *xe_tt;
>>>> int err;
>>>>
>>>> + xe_tt = container_of(tt, struct xe_ttm_tt, ttm);
>>>> /*
>>>> * dma-bufs are not populated with pages, and the dma-
>>>> * addresses are set up when moved to XE_PL_TT.
>>>> @@ -426,7 +445,7 @@ static int xe_ttm_tt_populate(struct
>>>> ttm_device
>>>> *ttm_dev, struct ttm_tt *tt,
>>>> return 0;
>>>>
>>>> /* Clear TTM_TT_FLAG_ZERO_ALLOC when GPU is set to clear
>>>> system pages */
>>>> - if (tt->page_flags & TTM_TT_FLAG_CLEARED_ON_FREE)
>>>> + if (xe_tt->sys_clear_on_alloc)
>>>> tt->page_flags &= ~TTM_TT_FLAG_ZERO_ALLOC;
>>>>
>>>> err = ttm_pool_alloc(&ttm_dev->pool, tt, ctx);
>>>> @@ -438,11 +457,19 @@ static int xe_ttm_tt_populate(struct
>>>> ttm_device
>>>> *ttm_dev, struct ttm_tt *tt,
>>>>
>>>> static void xe_ttm_tt_unpopulate(struct ttm_device *ttm_dev,
>>>> struct
>>>> ttm_tt *tt)
>>>> {
>>>> + struct xe_ttm_tt *xe_tt;
>>>> +
>>>> + xe_tt = container_of(tt, struct xe_ttm_tt, ttm);
>>>> +
>>>> if (tt->page_flags & TTM_TT_FLAG_EXTERNAL)
>>>> return;
>>>>
>>>> xe_tt_unmap_sg(tt);
>>>>
>>>> + /* Hint TTM pool that pages are already cleared */
>>>> + if (xe_tt->sys_clear_on_free)
>>>> + tt->page_flags |= TTM_TT_FLAG_CLEARED_ON_FREE;
>>>> +
>>>> return ttm_pool_free(&ttm_dev->pool, tt);
>>>> }
>>>>
>>>> @@ -664,6 +691,7 @@ static int xe_bo_move(struct
>>>> ttm_buffer_object
>>>> *ttm_bo, bool evict,
>>>> struct ttm_resource *old_mem = ttm_bo->resource;
>>>> u32 old_mem_type = old_mem ? old_mem->mem_type :
>>>> XE_PL_SYSTEM;
>>>> struct ttm_tt *ttm = ttm_bo->ttm;
>>>> + struct xe_ttm_tt *xe_tt;
>>>> struct xe_migrate *migrate = NULL;
>>>> struct dma_fence *fence;
>>>> bool move_lacks_source;
>>>> @@ -674,12 +702,13 @@ static int xe_bo_move(struct
>>>> ttm_buffer_object
>>>> *ttm_bo, bool evict,
>>>> bool clear_system_pages;
>>>> int ret = 0;
>>>>
>>>> + xe_tt = container_of(ttm_bo->ttm, struct xe_ttm_tt,
>>>> ttm);
>>>> /*
>>>> * Clear TTM_TT_FLAG_CLEARED_ON_FREE on bo creation path
>>>> when
>>>> * moving to system as the bo doesn't have dma_mapping.
>>>> */
>>>> if (!old_mem && ttm && !ttm_tt_is_populated(ttm))
>>>> - ttm->page_flags &= ~TTM_TT_FLAG_CLEARED_ON_FREE;
>>>> + xe_tt->sys_clear_on_alloc = false;
>>>>
>>>> /* Bo creation path, moving to system or TT. */
>>>> if ((!old_mem && ttm) && !handle_system_ccs) {
>>>> @@ -703,10 +732,9 @@ static int xe_bo_move(struct
>>>> ttm_buffer_object
>>>> *ttm_bo, bool evict,
>>>> move_lacks_source = handle_system_ccs ? (!bo-
>>>>> ccs_cleared)
>>>> :
>>>> (!mem_type_is_vr
>>>> am(o
>>>> ld_mem_type) && !tt_has_data);
>>>>
>>>> - clear_system_pages = ttm && (ttm->page_flags &
>>>> TTM_TT_FLAG_CLEARED_ON_FREE);
>>>> + clear_system_pages = ttm && xe_tt->sys_clear_on_alloc;
>>>> needs_clear = (ttm && ttm->page_flags &
>>>> TTM_TT_FLAG_ZERO_ALLOC) ||
>>>> - (!ttm && ttm_bo->type == ttm_bo_type_device) ||
>>>> - clear_system_pages;
>>>> + (!ttm && ttm_bo->type == ttm_bo_type_device) ||
>>>> clear_system_pages;
>>>>
>>>> if (new_mem->mem_type == XE_PL_TT) {
>>>> ret = xe_tt_map_sg(ttm);
>>>> @@ -1028,10 +1056,47 @@ static bool
>>>> xe_ttm_bo_lock_in_destructor(struct ttm_buffer_object *ttm_bo)
>>>> return locked;
>>>> }
>>>>
>>>> +static struct dma_fence *xe_ttm_bo_clear_on_free(struct
>>>> ttm_buffer_object *ttm_bo)
>>>> +{
>>>> + struct xe_bo *bo = ttm_to_xe_bo(ttm_bo);
>>>> + struct xe_device *xe = xe_bo_device(bo);
>>>> + struct xe_migrate *migrate;
>>>> + struct xe_ttm_tt *xe_tt;
>>>> + struct dma_fence *clear_fence;
>>>> +
>>>> + /* return early if nothing to clear */
>>>> + if (!ttm_bo->ttm)
>>>> + return NULL;
>>>> +
>>>> + xe_tt = container_of(ttm_bo->ttm, struct xe_ttm_tt,
>>>> ttm);
>>>> + /* return early if nothing to clear */
>>>> + if (!xe_tt->sys_clear_on_free || !bo->ttm.resource)
>>>> + return NULL;
>>>> +
>>>> + if (XE_WARN_ON(!xe_tt->sg))
>>>> + return NULL;
>>>> +
>>>> + if (bo->tile)
>>>> + migrate = bo->tile->migrate;
>>>> + else
>>>> + migrate = xe->tiles[0].migrate;
>>>> +
>>>> + xe_assert(xe, migrate);
>>>> +
>>>> + clear_fence = xe_migrate_clear(migrate, bo, bo-
>>>>> ttm.resource,
>>>> +
>>>> XE_MIGRATE_CLEAR_FLAG_FULL);
>>>> + if (IS_ERR(clear_fence))
>>>> + return NULL;
>>>> +
>>>> + xe_tt->sys_clear_on_free = false;
>>>> +
>>>> + return clear_fence;
>>>> +}
>>>> +
>>>> static void xe_ttm_bo_release_notify(struct ttm_buffer_object
>>>> *ttm_bo)
>>>> {
>>>> struct dma_resv_iter cursor;
>>>> - struct dma_fence *fence;
>>>> + struct dma_fence *clear_fence, *fence;
>>>> struct dma_fence *replacement = NULL;
>>>> struct xe_bo *bo;
>>>>
>>>> @@ -1041,15 +1106,31 @@ static void
>>>> xe_ttm_bo_release_notify(struct
>>>> ttm_buffer_object *ttm_bo)
>>>> bo = ttm_to_xe_bo(ttm_bo);
>>>> xe_assert(xe_bo_device(bo), !(bo->created &&
>>>> kref_read(&ttm_bo->base.refcount)));
>>>>
>>>> + clear_fence = xe_ttm_bo_clear_on_free(ttm_bo);
>>>> +
>>>> /*
>>>> * Corner case where TTM fails to allocate memory and
>>>> this
>>>> BOs resv
>>>> * still points the VMs resv
>>>> */
>>>> - if (ttm_bo->base.resv != &ttm_bo->base._resv)
>>>> + if (ttm_bo->base.resv != &ttm_bo->base._resv) {
>>>> + if (clear_fence)
>>>> + dma_fence_wait(clear_fence, false);
>>>> return;
>>>> + }
>>>>
>>>> - if (!xe_ttm_bo_lock_in_destructor(ttm_bo))
>>>> + if (!xe_ttm_bo_lock_in_destructor(ttm_bo)) {
>>>> + if (clear_fence)
>>>> + dma_fence_wait(clear_fence, false);
>>>> return;
>>>> + }
>>>> +
>>>> + if (clear_fence) {
>>>> + if (dma_resv_reserve_fences(ttm_bo->base.resv,
>>>> 1))
>>>> + dma_fence_wait(clear_fence, false);
>>>> + else
>>>> + dma_resv_add_fence(ttm_bo->base.resv,
>>>> clear_fence,
>>>> +
>>>> DMA_RESV_USAGE_KERNEL);
>>>> + }
>>>>
>>>> /*
>>>> * Scrub the preempt fences if any. The unbind fence is
>>>> already
>
More information about the Intel-xe
mailing list