All of lore.kernel.org
 help / color / mirror / Atom feed
From: "Christian König" <ckoenig.leichtzumerken@gmail.com>
To: "Thomas Hellström (VMware)" <thomas_os@shipmail.org>,
	kraxel@redhat.com, daniel@ffwll.ch,
	dri-devel@lists.freedesktop.org
Subject: Re: [PATCH 2/3] drm/ttm: always keep BOs on the LRU
Date: Wed, 16 Oct 2019 11:30:34 +0200	[thread overview]
Message-ID: <80abd72f-2336-135a-4304-b5ab8451e9c5@gmail.com> (raw)
In-Reply-To: <29539c40-1ea4-b60a-791f-385fb3468072@gmail.com>

Am 25.09.19 um 14:10 schrieb Christian König:
> Am 25.09.19 um 14:06 schrieb Thomas Hellström (VMware):
>> On 9/25/19 12:55 PM, Christian König wrote:
>>> This allows blocking for BOs to become available
>>> in the memory management.
>>>
>>> Amdgpu is doing this for quite a while now during CS. Now
>>> apply the new behavior to all drivers using TTM.
>>>
>>> Signed-off-by: Christian König <christian.koenig@amd.com>
>>
>> Got to test this to see that there are no regressions.

Did you got time to test this or did I just missed your response?

Thanks in advance,
Christian.

>>
>> There are some ugly cases in the vmwgfx driver (coded before we could 
>> create bos in the reserved state) that relies on creating a bo as 
>> NO_EVICT to make sure a following tryreserve will always succeed. So 
>> how are NO_EVICT bos handled after the below changes? Taken off the 
>> lru or just avoided when walking the lru?
>
> They are never added to the LRU in the first place, the NO_EVICT flag 
> is still checked during BO init.
>
>>
>> Also, does this mean we can in theory start using the drm_mm_scan() 
>> interface for VRAM evictions, relying on the ww_mutexes to resolve 
>> contention?
>
> It's the first step towards this, we still need some rather funky 
> ww_mutex dance to completely get this working.
>
> Christian.
>
>>
>> /Thomas
>>
>>
>>> ---
>>>   .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c  |  9 ++--
>>>   drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c        |  2 +-
>>>   drivers/gpu/drm/amd/amdgpu/amdgpu_csa.c       |  2 +-
>>>   drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c       |  4 +-
>>>   .../gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c |  2 +-
>>>   drivers/gpu/drm/qxl/qxl_release.c             |  4 +-
>>>   drivers/gpu/drm/radeon/radeon_gem.c           |  2 +-
>>>   drivers/gpu/drm/radeon/radeon_object.c        |  2 +-
>>>   drivers/gpu/drm/ttm/ttm_bo.c                  | 48 
>>> +++++--------------
>>>   drivers/gpu/drm/ttm/ttm_execbuf_util.c        | 25 ++--------
>>>   drivers/gpu/drm/vmwgfx/vmwgfx_resource.c      |  3 +-
>>>   drivers/gpu/drm/vmwgfx/vmwgfx_validation.h    |  2 +-
>>>   include/drm/ttm/ttm_bo_api.h                  | 24 ----------
>>>   include/drm/ttm/ttm_bo_driver.h               | 20 ++------
>>>   include/drm/ttm/ttm_execbuf_util.h            |  2 +-
>>>   15 files changed, 33 insertions(+), 118 deletions(-)
>>>
>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c 
>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
>>> index b9bb35d1699e..ad01c741caaf 100644
>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
>>> @@ -586,7 +586,7 @@ static int reserve_bo_and_vm(struct kgd_mem *mem,
>>>       amdgpu_vm_get_pd_bo(vm, &ctx->list, &ctx->vm_pd[0]);
>>>         ret = ttm_eu_reserve_buffers(&ctx->ticket, &ctx->list,
>>> -                     false, &ctx->duplicates, true);
>>> +                     false, &ctx->duplicates);
>>>       if (!ret)
>>>           ctx->reserved = true;
>>>       else {
>>> @@ -659,7 +659,7 @@ static int reserve_bo_and_cond_vms(struct 
>>> kgd_mem *mem,
>>>       }
>>>         ret = ttm_eu_reserve_buffers(&ctx->ticket, &ctx->list,
>>> -                     false, &ctx->duplicates, true);
>>> +                     false, &ctx->duplicates);
>>>       if (!ret)
>>>           ctx->reserved = true;
>>>       else
>>> @@ -1796,8 +1796,7 @@ static int validate_invalid_user_pages(struct 
>>> amdkfd_process_info *process_info)
>>>       }
>>>         /* Reserve all BOs and page tables for validation */
>>> -    ret = ttm_eu_reserve_buffers(&ticket, &resv_list, false, 
>>> &duplicates,
>>> -                     true);
>>> +    ret = ttm_eu_reserve_buffers(&ticket, &resv_list, false, 
>>> &duplicates);
>>>       WARN(!list_empty(&duplicates), "Duplicates should be empty");
>>>       if (ret)
>>>           goto out_free;
>>> @@ -1995,7 +1994,7 @@ int 
>>> amdgpu_amdkfd_gpuvm_restore_process_bos(void *info, struct dma_fence 
>>> **ef)
>>>       }
>>>         ret = ttm_eu_reserve_buffers(&ctx.ticket, &ctx.list,
>>> -                     false, &duplicate_save, true);
>>> +                     false, &duplicate_save);
>>>       if (ret) {
>>>           pr_debug("Memory eviction: TTM Reserve Failed. Try again\n");
>>>           goto ttm_reserve_fail;
>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c 
>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
>>> index 95ec965fcc2d..82f239ac4050 100644
>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
>>> @@ -650,7 +650,7 @@ static int amdgpu_cs_parser_bos(struct 
>>> amdgpu_cs_parser *p,
>>>       }
>>>         r = ttm_eu_reserve_buffers(&p->ticket, &p->validated, true,
>>> -                   &duplicates, false);
>>> +                   &duplicates);
>>>       if (unlikely(r != 0)) {
>>>           if (r != -ERESTARTSYS)
>>>               DRM_ERROR("ttm_eu_reserve_buffers failed.\n");
>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_csa.c 
>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_csa.c
>>> index 605f83046039..b1608d47508f 100644
>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_csa.c
>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_csa.c
>>> @@ -80,7 +80,7 @@ int amdgpu_map_static_csa(struct amdgpu_device 
>>> *adev, struct amdgpu_vm *vm,
>>>       list_add(&csa_tv.head, &list);
>>>       amdgpu_vm_get_pd_bo(vm, &list, &pd);
>>>   -    r = ttm_eu_reserve_buffers(&ticket, &list, true, NULL, false);
>>> +    r = ttm_eu_reserve_buffers(&ticket, &list, true, NULL);
>>>       if (r) {
>>>           DRM_ERROR("failed to reserve CSA,PD BOs: err=%d\n", r);
>>>           return r;
>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c 
>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
>>> index b25a59c4bec6..cbfb3e84f352 100644
>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
>>> @@ -175,7 +175,7 @@ void amdgpu_gem_object_close(struct 
>>> drm_gem_object *obj,
>>>         amdgpu_vm_get_pd_bo(vm, &list, &vm_pd);
>>>   -    r = ttm_eu_reserve_buffers(&ticket, &list, false, 
>>> &duplicates, false);
>>> +    r = ttm_eu_reserve_buffers(&ticket, &list, false, &duplicates);
>>>       if (r) {
>>>           dev_err(adev->dev, "leaking bo va because "
>>>               "we fail to reserve bo (%d)\n", r);
>>> @@ -611,7 +611,7 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, 
>>> void *data,
>>>         amdgpu_vm_get_pd_bo(&fpriv->vm, &list, &vm_pd);
>>>   -    r = ttm_eu_reserve_buffers(&ticket, &list, true, &duplicates, 
>>> false);
>>> +    r = ttm_eu_reserve_buffers(&ticket, &list, true, &duplicates);
>>>       if (r)
>>>           goto error_unref;
>>>   diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c 
>>> b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
>>> index 5df2ee1e10d8..760078b9925b 100644
>>> --- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
>>> +++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
>>> @@ -4439,7 +4439,7 @@ static int dm_plane_helper_prepare_fb(struct 
>>> drm_plane *plane,
>>>       tv.num_shared = 1;
>>>       list_add(&tv.head, &list);
>>>   -    r = ttm_eu_reserve_buffers(&ticket, &list, false, NULL, true);
>>> +    r = ttm_eu_reserve_buffers(&ticket, &list, false, NULL);
>>>       if (r) {
>>>           dev_err(adev->dev, "fail to reserve bo (%d)\n", r);
>>>           return r;
>>> diff --git a/drivers/gpu/drm/qxl/qxl_release.c 
>>> b/drivers/gpu/drm/qxl/qxl_release.c
>>> index aa7a28795645..71f1b51a1350 100644
>>> --- a/drivers/gpu/drm/qxl/qxl_release.c
>>> +++ b/drivers/gpu/drm/qxl/qxl_release.c
>>> @@ -260,7 +260,7 @@ int qxl_release_reserve_list(struct qxl_release 
>>> *release, bool no_intr)
>>>           return 0;
>>>         ret = ttm_eu_reserve_buffers(&release->ticket, &release->bos,
>>> -                     !no_intr, NULL, true);
>>> +                     !no_intr, NULL);
>>>       if (ret)
>>>           return ret;
>>>   @@ -459,7 +459,7 @@ void qxl_release_fence_buffer_objects(struct 
>>> qxl_release *release)
>>>           bo = entry->bo;
>>>             dma_resv_add_shared_fence(bo->base.resv, &release->base);
>>> -        ttm_bo_add_to_lru(bo);
>>> +        ttm_bo_move_to_lru_tail(bo, NULL);
>>>           dma_resv_unlock(bo->base.resv);
>>>       }
>>>       spin_unlock(&glob->lru_lock);
>>> diff --git a/drivers/gpu/drm/radeon/radeon_gem.c 
>>> b/drivers/gpu/drm/radeon/radeon_gem.c
>>> index c48c2fb35456..879eac67cdd7 100644
>>> --- a/drivers/gpu/drm/radeon/radeon_gem.c
>>> +++ b/drivers/gpu/drm/radeon/radeon_gem.c
>>> @@ -564,7 +564,7 @@ static void radeon_gem_va_update_vm(struct 
>>> radeon_device *rdev,
>>>       if (!vm_bos)
>>>           return;
>>>   -    r = ttm_eu_reserve_buffers(&ticket, &list, true, NULL, true);
>>> +    r = ttm_eu_reserve_buffers(&ticket, &list, true, NULL);
>>>       if (r)
>>>           goto error_free;
>>>   diff --git a/drivers/gpu/drm/radeon/radeon_object.c 
>>> b/drivers/gpu/drm/radeon/radeon_object.c
>>> index 653fd7937b39..a94d827ba976 100644
>>> --- a/drivers/gpu/drm/radeon/radeon_object.c
>>> +++ b/drivers/gpu/drm/radeon/radeon_object.c
>>> @@ -542,7 +542,7 @@ int radeon_bo_list_validate(struct radeon_device 
>>> *rdev,
>>>       u64 bytes_moved_threshold = 
>>> radeon_bo_get_threshold_for_moves(rdev);
>>>         INIT_LIST_HEAD(&duplicates);
>>> -    r = ttm_eu_reserve_buffers(ticket, head, true, &duplicates, true);
>>> +    r = ttm_eu_reserve_buffers(ticket, head, true, &duplicates);
>>>       if (unlikely(r != 0)) {
>>>           return r;
>>>       }
>>> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c 
>>> b/drivers/gpu/drm/ttm/ttm_bo.c
>>> index 02f301e1ff29..ab9967b84e0b 100644
>>> --- a/drivers/gpu/drm/ttm/ttm_bo.c
>>> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
>>> @@ -192,18 +192,12 @@ static void ttm_bo_add_mem_to_lru(struct 
>>> ttm_buffer_object *bo,
>>>       }
>>>   }
>>>   -void ttm_bo_add_to_lru(struct ttm_buffer_object *bo)
>>> -{
>>> -    ttm_bo_add_mem_to_lru(bo, &bo->mem);
>>> -}
>>> -EXPORT_SYMBOL(ttm_bo_add_to_lru);
>>> -
>>>   static void ttm_bo_ref_bug(struct kref *list_kref)
>>>   {
>>>       BUG();
>>>   }
>>>   -void ttm_bo_del_from_lru(struct ttm_buffer_object *bo)
>>> +static void ttm_bo_del_from_lru(struct ttm_buffer_object *bo)
>>>   {
>>>       struct ttm_bo_device *bdev = bo->bdev;
>>>       bool notify = false;
>>> @@ -223,16 +217,6 @@ void ttm_bo_del_from_lru(struct 
>>> ttm_buffer_object *bo)
>>>           bdev->driver->del_from_lru_notify(bo);
>>>   }
>>>   -void ttm_bo_del_sub_from_lru(struct ttm_buffer_object *bo)
>>> -{
>>> -    struct ttm_bo_global *glob = bo->bdev->glob;
>>> -
>>> -    spin_lock(&glob->lru_lock);
>>> -    ttm_bo_del_from_lru(bo);
>>> -    spin_unlock(&glob->lru_lock);
>>> -}
>>> -EXPORT_SYMBOL(ttm_bo_del_sub_from_lru);
>>> -
>>>   static void ttm_bo_bulk_move_set_pos(struct ttm_lru_bulk_move_pos 
>>> *pos,
>>>                        struct ttm_buffer_object *bo)
>>>   {
>>> @@ -247,7 +231,7 @@ void ttm_bo_move_to_lru_tail(struct 
>>> ttm_buffer_object *bo,
>>>       dma_resv_assert_held(bo->base.resv);
>>>         ttm_bo_del_from_lru(bo);
>>> -    ttm_bo_add_to_lru(bo);
>>> +    ttm_bo_add_mem_to_lru(bo, &bo->mem);
>>>         if (bulk && !(bo->mem.placement & TTM_PL_FLAG_NO_EVICT)) {
>>>           switch (bo->mem.mem_type) {
>>> @@ -511,7 +495,7 @@ static void ttm_bo_cleanup_refs_or_queue(struct 
>>> ttm_buffer_object *bo)
>>>            */
>>>           if (bo->mem.placement & TTM_PL_FLAG_NO_EVICT) {
>>>               bo->mem.placement &= ~TTM_PL_FLAG_NO_EVICT;
>>> -            ttm_bo_add_to_lru(bo);
>>> +            ttm_bo_move_to_lru_tail(bo, NULL);
>>>           }
>>>             dma_resv_unlock(bo->base.resv);
>>> @@ -892,17 +876,11 @@ static int ttm_mem_evict_first(struct 
>>> ttm_bo_device *bdev,
>>>           return ret;
>>>       }
>>>   -    ttm_bo_del_from_lru(bo);
>>>       spin_unlock(&glob->lru_lock);
>>>         ret = ttm_bo_evict(bo, ctx);
>>> -    if (locked) {
>>> +    if (locked)
>>>           ttm_bo_unreserve(bo);
>>> -    } else {
>>> -        spin_lock(&glob->lru_lock);
>>> -        ttm_bo_add_to_lru(bo);
>>> -        spin_unlock(&glob->lru_lock);
>>> -    }
>>>         kref_put(&bo->list_kref, ttm_bo_release_list);
>>>       return ret;
>>> @@ -1064,12 +1042,10 @@ static int ttm_bo_mem_placement(struct 
>>> ttm_buffer_object *bo,
>>>       mem->mem_type = mem_type;
>>>       mem->placement = cur_flags;
>>>   -    if (bo->mem.mem_type < mem_type && !list_empty(&bo->lru)) {
>>> -        spin_lock(&bo->bdev->glob->lru_lock);
>>> -        ttm_bo_del_from_lru(bo);
>>> -        ttm_bo_add_mem_to_lru(bo, mem);
>>> -        spin_unlock(&bo->bdev->glob->lru_lock);
>>> -    }
>>> +    spin_lock(&bo->bdev->glob->lru_lock);
>>> +    ttm_bo_del_from_lru(bo);
>>> +    ttm_bo_add_mem_to_lru(bo, mem);
>>> +    spin_unlock(&bo->bdev->glob->lru_lock);
>>>         return 0;
>>>   }
>>> @@ -1375,11 +1351,9 @@ int ttm_bo_init_reserved(struct ttm_bo_device 
>>> *bdev,
>>>           return ret;
>>>       }
>>>   -    if (resv && !(bo->mem.placement & TTM_PL_FLAG_NO_EVICT)) {
>>> -        spin_lock(&bdev->glob->lru_lock);
>>> -        ttm_bo_add_to_lru(bo);
>>> -        spin_unlock(&bdev->glob->lru_lock);
>>> -    }
>>> +    spin_lock(&bdev->glob->lru_lock);
>>> +    ttm_bo_move_to_lru_tail(bo, NULL);
>>> +    spin_unlock(&bdev->glob->lru_lock);
>>>         return ret;
>>>   }
>>> diff --git a/drivers/gpu/drm/ttm/ttm_execbuf_util.c 
>>> b/drivers/gpu/drm/ttm/ttm_execbuf_util.c
>>> index 71148c83cc4f..b09c2c8caf13 100644
>>> --- a/drivers/gpu/drm/ttm/ttm_execbuf_util.c
>>> +++ b/drivers/gpu/drm/ttm/ttm_execbuf_util.c
>>> @@ -33,16 +33,6 @@
>>>   #include <linux/sched.h>
>>>   #include <linux/module.h>
>>>   -static void ttm_eu_del_from_lru_locked(struct list_head *list)
>>> -{
>>> -    struct ttm_validate_buffer *entry;
>>> -
>>> -    list_for_each_entry(entry, list, head) {
>>> -        struct ttm_buffer_object *bo = entry->bo;
>>> -        ttm_bo_del_from_lru(bo);
>>> -    }
>>> -}
>>> -
>>>   void ttm_eu_backoff_reservation(struct dma_resv_ctx *ticket,
>>>                   struct list_head *list)
>>>   {
>>> @@ -59,8 +49,7 @@ void ttm_eu_backoff_reservation(struct 
>>> dma_resv_ctx *ticket,
>>>       list_for_each_entry(entry, list, head) {
>>>           struct ttm_buffer_object *bo = entry->bo;
>>>   -        if (list_empty(&bo->lru))
>>> -            ttm_bo_add_to_lru(bo);
>>> +        ttm_bo_move_to_lru_tail(bo, NULL);
>>>           if (!ticket)
>>>               dma_resv_unlock(bo->base.resv);
>>>       }
>>> @@ -87,7 +76,7 @@ EXPORT_SYMBOL(ttm_eu_backoff_reservation);
>>>     int ttm_eu_reserve_buffers(struct dma_resv_ctx *ticket,
>>>                  struct list_head *list, bool intr,
>>> -               struct list_head *dups, bool del_lru)
>>> +               struct list_head *dups)
>>>   {
>>>       struct ttm_validate_buffer *entry;
>>>       struct ttm_bo_global *glob;
>>> @@ -142,11 +131,6 @@ int ttm_eu_reserve_buffers(struct dma_resv_ctx 
>>> *ticket,
>>>           }
>>>       }
>>>   -    if (del_lru) {
>>> -        spin_lock(&glob->lru_lock);
>>> -        ttm_eu_del_from_lru_locked(list);
>>> -        spin_unlock(&glob->lru_lock);
>>> -    }
>>>       return 0;
>>>     error:
>>> @@ -186,10 +170,7 @@ void ttm_eu_fence_buffer_objects(struct 
>>> dma_resv_ctx *ticket,
>>>               dma_resv_add_shared_fence(bo->base.resv, fence);
>>>           else
>>>               dma_resv_add_excl_fence(bo->base.resv, fence);
>>> -        if (list_empty(&bo->lru))
>>> -            ttm_bo_add_to_lru(bo);
>>> -        else
>>> -            ttm_bo_move_to_lru_tail(bo, NULL);
>>> +        ttm_bo_move_to_lru_tail(bo, NULL);
>>>           if (!ticket)
>>>               dma_resv_unlock(bo->base.resv);
>>>       }
>>> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c 
>>> b/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c
>>> index 2d7c5ad25359..1d53a69819e5 100644
>>> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c
>>> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c
>>> @@ -464,8 +464,7 @@ vmw_resource_check_buffer(struct dma_resv_ctx 
>>> *ticket,
>>>       val_buf->bo = &res->backup->base;
>>>       val_buf->num_shared = 0;
>>>       list_add_tail(&val_buf->head, &val_list);
>>> -    ret = ttm_eu_reserve_buffers(ticket, &val_list, interruptible, 
>>> NULL,
>>> -                     true);
>>> +    ret = ttm_eu_reserve_buffers(ticket, &val_list, interruptible, 
>>> NULL);
>>>       if (unlikely(ret != 0))
>>>           goto out_no_reserve;
>>>   diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_validation.h 
>>> b/drivers/gpu/drm/vmwgfx/vmwgfx_validation.h
>>> index 43f48df3844f..ebb5b68b5c1f 100644
>>> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_validation.h
>>> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_validation.h
>>> @@ -169,7 +169,7 @@ vmw_validation_bo_reserve(struct 
>>> vmw_validation_context *ctx,
>>>                 bool intr)
>>>   {
>>>       return ttm_eu_reserve_buffers(&ctx->ticket, &ctx->bo_list, intr,
>>> -                      NULL, true);
>>> +                      NULL);
>>>   }
>>>     /**
>>> diff --git a/include/drm/ttm/ttm_bo_api.h 
>>> b/include/drm/ttm/ttm_bo_api.h
>>> index 43c4929a2171..288fd64b7219 100644
>>> --- a/include/drm/ttm/ttm_bo_api.h
>>> +++ b/include/drm/ttm/ttm_bo_api.h
>>> @@ -367,30 +367,6 @@ int ttm_bo_validate(struct ttm_buffer_object *bo,
>>>    */
>>>   void ttm_bo_put(struct ttm_buffer_object *bo);
>>>   -/**
>>> - * ttm_bo_add_to_lru
>>> - *
>>> - * @bo: The buffer object.
>>> - *
>>> - * Add this bo to the relevant mem type lru and, if it's backed by
>>> - * system pages (ttms) to the swap list.
>>> - * This function must be called with struct ttm_bo_global::lru_lock 
>>> held, and
>>> - * is typically called immediately prior to unreserving a bo.
>>> - */
>>> -void ttm_bo_add_to_lru(struct ttm_buffer_object *bo);
>>> -
>>> -/**
>>> - * ttm_bo_del_from_lru
>>> - *
>>> - * @bo: The buffer object.
>>> - *
>>> - * Remove this bo from all lru lists used to lookup and reserve an 
>>> object.
>>> - * This function must be called with struct ttm_bo_global::lru_lock 
>>> held,
>>> - * and is usually called just immediately after the bo has been 
>>> reserved to
>>> - * avoid recursive reservation from lru lists.
>>> - */
>>> -void ttm_bo_del_from_lru(struct ttm_buffer_object *bo);
>>> -
>>>   /**
>>>    * ttm_bo_move_to_lru_tail
>>>    *
>>> diff --git a/include/drm/ttm/ttm_bo_driver.h 
>>> b/include/drm/ttm/ttm_bo_driver.h
>>> index 4e307f65f497..4332cc036483 100644
>>> --- a/include/drm/ttm/ttm_bo_driver.h
>>> +++ b/include/drm/ttm/ttm_bo_driver.h
>>> @@ -621,9 +621,6 @@ void ttm_mem_io_free_vm(struct ttm_buffer_object 
>>> *bo);
>>>   int ttm_mem_io_lock(struct ttm_mem_type_manager *man, bool 
>>> interruptible);
>>>   void ttm_mem_io_unlock(struct ttm_mem_type_manager *man);
>>>   -void ttm_bo_del_sub_from_lru(struct ttm_buffer_object *bo);
>>> -void ttm_bo_add_to_lru(struct ttm_buffer_object *bo);
>>> -
>>>   /**
>>>    * __ttm_bo_reserve:
>>>    *
>>> @@ -717,15 +714,9 @@ static inline int ttm_bo_reserve(struct 
>>> ttm_buffer_object *bo,
>>>                    bool interruptible, bool no_wait,
>>>                    struct ww_acquire_ctx *ticket)
>>>   {
>>> -    int ret;
>>> -
>>>       WARN_ON(!kref_read(&bo->kref));
>>>   -    ret = __ttm_bo_reserve(bo, interruptible, no_wait, ticket);
>>> -    if (likely(ret == 0))
>>> -        ttm_bo_del_sub_from_lru(bo);
>>> -
>>> -    return ret;
>>> +    return __ttm_bo_reserve(bo, interruptible, no_wait, ticket);
>>>   }
>>>     /**
>>> @@ -752,9 +743,7 @@ static inline int ttm_bo_reserve_slowpath(struct 
>>> ttm_buffer_object *bo,
>>>       else
>>>           dma_resv_lock_slow(bo->base.resv, ticket);
>>>   -    if (likely(ret == 0))
>>> -        ttm_bo_del_sub_from_lru(bo);
>>> -    else if (ret == -EINTR)
>>> +    if (ret == -EINTR)
>>>           ret = -ERESTARTSYS;
>>>         return ret;
>>> @@ -770,10 +759,7 @@ static inline int 
>>> ttm_bo_reserve_slowpath(struct ttm_buffer_object *bo,
>>>   static inline void ttm_bo_unreserve(struct ttm_buffer_object *bo)
>>>   {
>>>       spin_lock(&bo->bdev->glob->lru_lock);
>>> -    if (list_empty(&bo->lru))
>>> -        ttm_bo_add_to_lru(bo);
>>> -    else
>>> -        ttm_bo_move_to_lru_tail(bo, NULL);
>>> +    ttm_bo_move_to_lru_tail(bo, NULL);
>>>       spin_unlock(&bo->bdev->glob->lru_lock);
>>>       dma_resv_unlock(bo->base.resv);
>>>   }
>>> diff --git a/include/drm/ttm/ttm_execbuf_util.h 
>>> b/include/drm/ttm/ttm_execbuf_util.h
>>> index 4e86b6fd6c57..1a3553ac53b1 100644
>>> --- a/include/drm/ttm/ttm_execbuf_util.h
>>> +++ b/include/drm/ttm/ttm_execbuf_util.h
>>> @@ -100,7 +100,7 @@ extern void ttm_eu_backoff_reservation(struct 
>>> dma_resv_ctx *ticket,
>>>     extern int ttm_eu_reserve_buffers(struct dma_resv_ctx *ticket,
>>>                     struct list_head *list, bool intr,
>>> -                  struct list_head *dups, bool del_lru);
>>> +                  struct list_head *dups);
>>>     /**
>>>    * function ttm_eu_fence_buffer_objects.
>>
>>
>

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

  reply	other threads:[~2019-10-16  9:30 UTC|newest]

Thread overview: 10+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2019-09-25 10:55 [PATCH 1/3] drm/ttm: fix busy reference in ttm_mem_evict_first Christian König
2019-09-25 10:55 ` [PATCH 2/3] drm/ttm: always keep BOs on the LRU Christian König
2019-09-25 12:06   ` Thomas Hellström (VMware)
2019-09-25 12:10     ` Christian König
2019-10-16  9:30       ` Christian König [this message]
2019-10-18 13:49         ` Thomas Hellström (VMware)
2019-10-21 12:27           ` Thomas Hellström (VMware)
2019-09-25 10:55 ` [PATCH 3/3] drm/ttm: remove pointers to globals Christian König
2019-09-25 11:56   ` Thomas Hellström (VMware)
2019-09-25 11:55 ` [PATCH 1/3] drm/ttm: fix busy reference in ttm_mem_evict_first Thomas Hellström (VMware)

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=80abd72f-2336-135a-4304-b5ab8451e9c5@gmail.com \
    --to=ckoenig.leichtzumerken@gmail.com \
    --cc=christian.koenig@amd.com \
    --cc=daniel@ffwll.ch \
    --cc=dri-devel@lists.freedesktop.org \
    --cc=kraxel@redhat.com \
    --cc=thomas_os@shipmail.org \
    /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 an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.