amdgpu: fix performance drop
Subsystems (10)
Hi Roger,
as promised the proper fix for the performance drop during command submission.
Please test the attached set of patches with your OpenCL test case.
Regards,
Christian.
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
A message that is not in the archive (sent off-list or not yet received): 1484301071-21703-1-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org
PATCH 1/6 drm/ttm: remove allow_errors parameter from ttm_bo_force_list_clean
From: Christian König <[email protected]>
Not allowing errors here is completely pointless and actually dangerous cause trying to continue on an error can cause an endless loop.
--- drivers/gpu/drm/ttm/ttm_bo.c | 28 +++++++++++----------------- 1 file changed, 11 insertions(+), 17 deletions(-)
drivers/gpu/drm/ttm/ttm_bo.c +11 -17
@@ -1291,7 +1291,7 @@ int ttm_bo_create(struct ttm_bo_device *bdev,EXPORT_SYMBOL (ttm_bo_create);static int ttm_bo_force_list_clean (struct ttm_bo_device *bdev, -unsigned mem_type,bool allow_errors) +unsigned mem_type) {struct ttm_mem_type_manager *man = &bdev->man[mem_type];struct ttm_bo_global *glob = bdev->glob;
@@ -1306,13 +1306,8 @@ static int ttm_bo_force_list_clean(struct ttm_bo_device *bdev,while (!list_empty (&man->lru)) {spin_unlock (&glob->lru_lock); ret =ttm_mem_evict_first (bdev, mem_type,NULL , false, false); -if (ret) { -if (allow_errors) { -return ret; - }else { -pr_err ("Cleanup eviction failed\n" ); - } - } +if (ret) +return ret;spin_lock (&glob->lru_lock); }spin_unlock (&glob->lru_lock);
@@ -1324,13 +1319,8 @@ static int ttm_bo_force_list_clean(struct ttm_bo_device *bdev,if (fence) { ret =fence_wait (fence, false);fence_put (fence); -if (ret) { -if (allow_errors) { -return ret; - }else { -pr_err ("Cleanup eviction failed\n" ); - } - } +if (ret) +return ret; }return 0 ;
@@ -1359,7 +1349,11 @@ int ttm_bo_clean_mm(struct ttm_bo_device *bdev, unsigned mem_type) ret =0 ;if (mem_type >0 ) { -ttm_bo_force_list_clean (bdev, mem_type, false); + ret =ttm_bo_force_list_clean (bdev, mem_type); +if (ret) { +pr_err ("Cleanup eviction failed\n" ); +return ret; + } ret = (*man->func->takedown)(man); }
@@ -1382,7 +1376,7 @@ int ttm_bo_evict_mm(struct ttm_bo_device *bdev, unsigned mem_type)return 0 ; } -return ttm_bo_force_list_clean (bdev, mem_type, true); +return ttm_bo_force_list_clean (bdev, mem_type); }EXPORT_SYMBOL (ttm_bo_evict_mm);
Signature · 6 lines
2.7.4 _______________________________________________ amd-gfx mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/amd-gfx
A message that is not in the archive (sent off-list or not yet received): 1484301071-21703-2-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org
RE: [PATCH 1/6] drm/ttm: remove allow_errors parameter from ttm_bo_force_list_clean
Looks good to me, patch is Reviewed-by: Roger.He <[email protected]>
-----Original Message-----
From: amd-gfx [mailto:[email protected]] On Behalf Of Christian K?nig
Sent: Friday, January 13, 2017 5:51 PM
To: He, Hongbo <[email protected]>
Cc: [email protected]; [email protected]
Subject: [PATCH 1/6] drm/ttm: remove allow_errors parameter from ttm_bo_force_list_clean
From: Christian König <[email protected]>
Not allowing errors here is completely pointless and actually dangerous cause trying to continue on an error can cause an endless loop.
--- drivers/gpu/drm/ttm/ttm_bo.c | 28 +++++++++++----------------- 1 file changed, 11 insertions(+), 17 deletions(-)
drivers/gpu/drm/ttm/ttm_bo.c +1 -1
@@ -1291,7 +1291,7 @@ int ttm_bo_create(struct ttm_bo_device *bdev, EXPORT_SYMBOL(ttm_bo_create);static int ttm_bo_force_list_clean (struct ttm_bo_device *bdev, -unsigned mem_type,bool allow_errors) +unsigned mem_type) {struct ttm_mem_type_manager *man = &bdev->man[mem_type]; struct ttm_bo_global *glob = bdev->glob; @@ -1306,13 +1306,8 @@ static int ttm_bo_force_list_clean(struct ttm_bo_device *bdev, while (!list_empty(&man->lru)) {
spin_unlock(&glob->lru_lock);
ret = ttm_mem_evict_first(bdev, mem_type, NULL, false, false);
- if (ret) {
- if (allow_errors) {
- return ret;
- } else {
- pr_err("Cleanup eviction failed\n");
- }
- }
+ if (ret)
+ return ret;
spin_lock(&glob->lru_lock);
}
spin_unlock(&glob->lru_lock);
@@ -1324,13 +1319,8 @@ static int ttm_bo_force_list_clean(struct ttm_bo_device *bdev,
if (fence) {
ret = fence_wait(fence, false);
fence_put(fence);
- if (ret) {
- if (allow_errors) {
- return ret;
- } else {
- pr_err("Cleanup eviction failed\n");
- }
- }
+ if (ret)
+ return ret;
}
return 0;
@@ -1359,7 +1349,11 @@ int ttm_bo_clean_mm(struct ttm_bo_device *bdev, unsigned mem_type)
ret = 0;
if (mem_type > 0) {
- ttm_bo_force_list_clean(bdev, mem_type, false);
+ ret = ttm_bo_force_list_clean(bdev, mem_type);
+ if (ret) {
+ pr_err("Cleanup eviction failed\n");
+ return ret;
+ }
ret = (*man->func->takedown)(man);
}
@@ -1382,7 +1376,7 @@ int ttm_bo_evict_mm(struct ttm_bo_device *bdev, unsigned mem_type)
return 0;
}
- return ttm_bo_force_list_clean(bdev, mem_type, true);
+ return ttm_bo_force_list_clean(bdev, mem_type);
}
EXPORT_SYMBOL(ttm_bo_evict_mm);
--
2.7.4
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
PATCH 2/6 drm/ttm: add BO priorities for the LRUs
In reply to a message not in the archiveFrom: Christian König <[email protected]>
This way the driver can specify a priority for a BO which has the effect that a BO is only evicted when all other BOs with a lower priority are evicted first.
--- drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 4 +- drivers/gpu/drm/ttm/ttm_bo.c | 67 ++++++++++++++++++++++----------- include/drm/ttm/ttm_bo_api.h | 2 + include/drm/ttm/ttm_bo_driver.h | 6 ++- 4 files changed, 52 insertions(+), 27 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +2 -2
@@ -1166,8 +1166,8 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)struct amdgpu_mman_lru *lru = &adev->mman.log2_size[i];for (j =0 ; j <TTM_NUM_MEM_TYPES ; ++j) - lru->lru[j] = &adev->mman.bdev.man[j].lru; - lru->swap_lru = &adev->mman.bdev.glob->swap_lru; + lru->lru[j] = &adev->mman.bdev.man[j].lru[0 ]; + lru->swap_lru = &adev->mman.bdev.glob->swap_lru[0 ]; }for (j =0 ; j <TTM_NUM_MEM_TYPES ; ++j)
drivers/gpu/drm/ttm/ttm_bo.c +44 -23
145 lines not on this page · raw message
include/drm/ttm/ttm_bo_api.h +2 -0
@@ -215,6 +215,8 @@ struct ttm_buffer_object {struct drm_vma_offset_node vma_node; +unsigned priority; +/** * Special members that are protected by the reserve lock * and the bo::lock when written to. Can be read with
include/drm/ttm/ttm_bo_driver.h +4 -2
@@ -42,6 +42,8 @@#include <linux/spinlock.h> #include <linux/reservation.h> +#define TTM_MAX_BO_PRIORITY 16 +struct ttm_backend_func {/** * struct ttm_backend_func member bind
@@ -298,7 +300,7 @@ struct ttm_mem_type_manager {* Protected by the global->lru_lock. */ -struct list_head lru; +struct list_head lru[TTM_MAX_BO_PRIORITY ];/* * Protected by @move_lock.
@@ -518,7 +520,7 @@ struct ttm_bo_global {/** * Protected by the lru_lock. */ -struct list_head swap_lru; +struct list_head swap_lru[TTM_MAX_BO_PRIORITY ];/** * Internal protection.
Signature · 6 lines
2.7.4 _______________________________________________ amd-gfx mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/amd-gfx
A message that is not in the archive (sent off-list or not yet received): 1484301071-21703-3-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org
He, Hongbo
Reviewed-by Roger.HeRE: [PATCH 2/6] drm/ttm: add BO priorities for the LRUs
-----Original Message-----
From: amd-gfx [mailto:[email protected]] On Behalf Of Christian K?nig
Sent: Friday, January 13, 2017 5:51 PM
To: He, Hongbo <[email protected]>
Cc: [email protected]; [email protected]
Subject: [PATCH 2/6] drm/ttm: add BO priorities for the LRUs
From: Christian König <[email protected]>
This way the driver can specify a priority for a BO which has the effect that a BO is only evicted when all other BOs with a lower priority are evicted first.
--- drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 4 +- drivers/gpu/drm/ttm/ttm_bo.c | 67 ++++++++++++++++++++++----------- include/drm/ttm/ttm_bo_api.h | 2 + include/drm/ttm/ttm_bo_driver.h | 6 ++- 4 files changed, 52 insertions(+), 27 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +2 -2
@@ -1166,8 +1166,8 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)struct amdgpu_mman_lru *lru = &adev->mman.log2_size[i];for (j =0 ; j <TTM_NUM_MEM_TYPES ; ++j) - lru->lru[j] = &adev->mman.bdev.man[j].lru; - lru->swap_lru = &adev->mman.bdev.glob->swap_lru; + lru->lru[j] = &adev->mman.bdev.man[j].lru[0 ]; + lru->swap_lru = &adev->mman.bdev.glob->swap_lru[0 ]; } for (j = 0; j < TTM_NUM_MEM_TYPES; ++j) diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c index 6683399..f078b43 100644
drivers/gpu/drm/ttm/ttm_bo.c +19 -11
@@ -242,13 +242,13 @@ EXPORT_SYMBOL(ttm_bo_move_to_lru_tail);struct list_head *ttm_bo_default_lru_tail (struct ttm_buffer_object *bo) { -return bo->bdev->man[bo->mem.mem_type].lru.prev; +return bo->bdev->man[bo->mem.mem_type].lru[bo->priority].prev; }EXPORT_SYMBOL (ttm_bo_default_lru_tail);struct list_head *ttm_bo_default_swap_lru_tail (struct ttm_buffer_object *bo) { -return bo->glob->swap_lru.prev; +return bo->glob->swap_lru[bo->priority].prev; }EXPORT_SYMBOL (ttm_bo_default_swap_lru_tail);
@@ -741,20 +741,27 @@ static int ttm_mem_evict_first(struct ttm_bo_device *bdev,struct ttm_mem_type_manager *man = &bdev->man[mem_type];struct ttm_buffer_object *bo;int ret = -EBUSY , put_count; +unsigned i;spin_lock (&glob->lru_lock); -list_for_each_entry (bo, &man->lru, lru) { - ret =__ttm_bo_reserve (bo, false, true,NULL ); -if (ret) -continue ; +for (i =0 ; i <TTM_MAX_BO_PRIORITY ; ++i) { +list_for_each_entry (bo, &man->lru[i], lru) { + ret =__ttm_bo_reserve (bo, false, true,NULL ); +if (ret) +continue ; -if (place && !bdev->driver->eviction_valuable (bo, place)) { -__ttm_bo_unreserve (bo); - ret = -EBUSY ; -continue ; +if (place && !bdev->driver->eviction_valuable (bo, + place)) { +__ttm_bo_unreserve (bo); + ret = -EBUSY ; +continue ; + } + +break ; } -break ; +if (!ret) +break ; }if (ret) {
@@ -1197,6 +1204,7 @@ int ttm_bo_init(struct ttm_bo_device *bdev, }atomic_inc (&bo->glob->bo_count);drm_vma_node_reset (&bo->vma_node); + bo->priority =0 ;/* * For ttm_bo_type_device buffers, allocate @@ -1297,18 +1305,21 @@ static int ttm_bo_force_list_clean(struct ttm_bo_device *bdev,
struct ttm_bo_global *glob = bdev->glob;
struct fence *fence;
int ret;
+ unsigned i;
/*
* Can't use standard list traversal since we're unlocking. */
spin_lock(&glob->lru_lock);
- while (!list_empty(&man->lru)) {
- spin_unlock(&glob->lru_lock);
- ret = ttm_mem_evict_first(bdev, mem_type, NULL, false, false);
- if (ret)
- return ret;
- spin_lock(&glob->lru_lock);
+ for (i = 0; i < TTM_MAX_BO_PRIORITY; ++i) {
+ while (!list_empty(&man->lru[i])) {
+ spin_unlock(&glob->lru_lock);
+ ret = ttm_mem_evict_first(bdev, mem_type, NULL, false, false);
+ if (ret)
+ return ret;
+ spin_lock(&glob->lru_lock);
+ }
}
spin_unlock(&glob->lru_lock);
@@ -1385,6 +1396,7 @@ int ttm_bo_init_mm(struct ttm_bo_device *bdev, unsigned type, {
int ret = -EINVAL;
struct ttm_mem_type_manager *man;
+ unsigned i;
BUG_ON(type >= TTM_NUM_MEM_TYPES);
man = &bdev->man[type];
@@ -1410,7 +1422,8 @@ int ttm_bo_init_mm(struct ttm_bo_device *bdev, unsigned type,
man->use_type = true;
man->size = p_size;
- INIT_LIST_HEAD(&man->lru);
+ for (i = 0; i < TTM_MAX_BO_PRIORITY; ++i)
+ INIT_LIST_HEAD(&man->lru[i]);
man->move = NULL;
return 0;
@@ -1442,6 +1455,7 @@ int ttm_bo_global_init(struct drm_global_reference *ref)
container_of(ref, struct ttm_bo_global_ref, ref);
struct ttm_bo_global *glob = ref->object;
int ret;
+ unsigned i;
mutex_init(&glob->device_list_mutex); spin_lock_init(&glob->lru_lock);
@@ -1453,7 +1467,8 @@ int ttm_bo_global_init(struct drm_global_reference *ref)
goto out_no_drp;
}
- INIT_LIST_HEAD(&glob->swap_lru);
+ for (i = 0; i < TTM_MAX_BO_PRIORITY; ++i)
+ INIT_LIST_HEAD(&glob->swap_lru[i]);
INIT_LIST_HEAD(&glob->device_list);
ttm_mem_init_shrink(&glob->shrink, ttm_bo_swapout); @@ -1512,8 +1527,9 @@ int ttm_bo_device_release(struct ttm_bo_device *bdev) if (list_empty(&bdev->ddestroy))
TTM_DEBUG("Delayed destroy list was clean\n");
- if (list_empty(&bdev->man[0].lru))
- TTM_DEBUG("Swap list was clean\n");
+ for (i = 0; i < TTM_MAX_BO_PRIORITY; ++i)
+ if (list_empty(&bdev->man[0].lru[0]))
+ TTM_DEBUG("Swap list %d was clean\n", i);
spin_unlock(&glob->lru_lock);
drm_vma_offset_manager_destroy(&bdev->vma_manager);
@@ -1665,10 +1681,15 @@ static int ttm_bo_swapout(struct ttm_mem_shrink *shrink)
int ret = -EBUSY;
int put_count;
uint32_t swap_placement = (TTM_PL_FLAG_CACHED | TTM_PL_FLAG_SYSTEM);
+ unsigned i;
spin_lock(&glob->lru_lock);
- list_for_each_entry(bo, &glob->swap_lru, swap) {
- ret = __ttm_bo_reserve(bo, false, true, NULL);
+ for (i = 0; i < TTM_MAX_BO_PRIORITY; ++i) {
+ list_for_each_entry(bo, &glob->swap_lru[i], swap) {
+ ret = __ttm_bo_reserve(bo, false, true, NULL);
+ if (!ret)
+ break;
+ }
if (!ret)
break;
}
include/drm/ttm/ttm_bo_api.h +2 -0
@@ -215,6 +215,8 @@ struct ttm_buffer_object {struct drm_vma_offset_node vma_node; +unsigned priority; +/** * Special members that are protected by the reserve lock * and the bo::lock when written to. Can be read with diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h index 12aa155..90dd5e9 100644
include/drm/ttm/ttm_bo_driver.h +2 -0
@@ -42,6 +42,8 @@#include <linux/spinlock.h> #include <linux/reservation.h> +#define TTM_MAX_BO_PRIORITY 16 +struct ttm_backend_func {/** * struct ttm_backend_func member bind @@ -298,7 +300,7 @@ struct ttm_mem_type_manager {
* Protected by the global->lru_lock.
*/
- struct list_head lru;
+ struct list_head lru[TTM_MAX_BO_PRIORITY];
/*
* Protected by @move_lock.
@@ -518,7 +520,7 @@ struct ttm_bo_global {
/**
* Protected by the lru_lock.
*/
- struct list_head swap_lru;
+ struct list_head swap_lru[TTM_MAX_BO_PRIORITY];
/**
* Internal protection.
--
2.7.4
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
PATCH 5/6 drm/ttm: revert "implement LRU add callbacks v2"
In reply to a message not in the archiveFrom: Christian König <[email protected]>
The additional housekeeping had to much CPU overhead, let's use the BO priorities instead.
--- drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 2 -- drivers/gpu/drm/ast/ast_ttm.c | 2 -- drivers/gpu/drm/bochs/bochs_mm.c | 2 -- drivers/gpu/drm/cirrus/cirrus_ttm.c | 2 -- drivers/gpu/drm/mgag200/mgag200_ttm.c | 2 -- drivers/gpu/drm/nouveau/nouveau_bo.c | 2 -- drivers/gpu/drm/qxl/qxl_ttm.c | 2 -- drivers/gpu/drm/radeon/radeon_ttm.c | 2 -- drivers/gpu/drm/ttm/ttm_bo.c | 19 +++++-------------- drivers/gpu/drm/virtio/virtgpu_ttm.c | 2 -- drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c | 2 -- include/drm/ttm/ttm_bo_driver.h | 9 --------- 12 files changed, 5 insertions(+), 43 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +0 -2
@@ -1086,8 +1086,6 @@ static struct ttm_bo_driver amdgpu_bo_driver = { .fault_reserve_notify = &amdgpu_bo_fault_reserve_notify, .io_mem_reserve = &amdgpu_ttm_io_mem_reserve, .io_mem_free = &amdgpu_ttm_io_mem_free, - .lru_tail = &ttm_bo_default_lru_tail, - .swap_lru_tail = &ttm_bo_default_swap_lru_tail, };int amdgpu_ttm_init (struct amdgpu_device *adev)
drivers/gpu/drm/ast/ast_ttm.c +0 -2
@@ -236,8 +236,6 @@ struct ttm_bo_driver ast_bo_driver = { .verify_access = ast_bo_verify_access, .io_mem_reserve = &ast_ttm_io_mem_reserve, .io_mem_free = &ast_ttm_io_mem_free, - .lru_tail = &ttm_bo_default_lru_tail, - .swap_lru_tail = &ttm_bo_default_swap_lru_tail, };int ast_mm_init (struct ast_private *ast)
drivers/gpu/drm/bochs/bochs_mm.c +0 -2
@@ -205,8 +205,6 @@ struct ttm_bo_driver bochs_bo_driver = { .verify_access = bochs_bo_verify_access, .io_mem_reserve = &bochs_ttm_io_mem_reserve, .io_mem_free = &bochs_ttm_io_mem_free, - .lru_tail = &ttm_bo_default_lru_tail, - .swap_lru_tail = &ttm_bo_default_swap_lru_tail, };int bochs_mm_init (struct bochs_device *bochs)
drivers/gpu/drm/cirrus/cirrus_ttm.c +0 -2
@@ -236,8 +236,6 @@ struct ttm_bo_driver cirrus_bo_driver = { .verify_access = cirrus_bo_verify_access, .io_mem_reserve = &cirrus_ttm_io_mem_reserve, .io_mem_free = &cirrus_ttm_io_mem_free, - .lru_tail = &ttm_bo_default_lru_tail, - .swap_lru_tail = &ttm_bo_default_swap_lru_tail, };int cirrus_mm_init (struct cirrus_device *cirrus)
drivers/gpu/drm/mgag200/mgag200_ttm.c +0 -2
@@ -236,8 +236,6 @@ struct ttm_bo_driver mgag200_bo_driver = { .verify_access = mgag200_bo_verify_access, .io_mem_reserve = &mgag200_ttm_io_mem_reserve, .io_mem_free = &mgag200_ttm_io_mem_free, - .lru_tail = &ttm_bo_default_lru_tail, - .swap_lru_tail = &ttm_bo_default_swap_lru_tail, };int mgag200_mm_init (struct mga_device *mdev)
drivers/gpu/drm/nouveau/nouveau_bo.c +0 -2
@@ -1570,8 +1570,6 @@ struct ttm_bo_driver nouveau_bo_driver = { .fault_reserve_notify = &nouveau_ttm_fault_reserve_notify, .io_mem_reserve = &nouveau_ttm_io_mem_reserve, .io_mem_free = &nouveau_ttm_io_mem_free, - .lru_tail = &ttm_bo_default_lru_tail, - .swap_lru_tail = &ttm_bo_default_swap_lru_tail, }; struct nvkm_vma *
drivers/gpu/drm/qxl/qxl_ttm.c +0 -2
@@ -395,8 +395,6 @@ static struct ttm_bo_driver qxl_bo_driver = { .io_mem_reserve = &qxl_ttm_io_mem_reserve, .io_mem_free = &qxl_ttm_io_mem_free, .move_notify = &qxl_bo_move_notify, - .lru_tail = &ttm_bo_default_lru_tail, - .swap_lru_tail = &ttm_bo_default_swap_lru_tail, };int qxl_ttm_init (struct qxl_device *qdev)
drivers/gpu/drm/radeon/radeon_ttm.c +0 -2
@@ -871,8 +871,6 @@ static struct ttm_bo_driver radeon_bo_driver = { .fault_reserve_notify = &radeon_bo_fault_reserve_notify, .io_mem_reserve = &radeon_ttm_io_mem_reserve, .io_mem_free = &radeon_ttm_io_mem_free, - .lru_tail = &ttm_bo_default_lru_tail, - .swap_lru_tail = &ttm_bo_default_swap_lru_tail, };int radeon_ttm_init (struct radeon_device *rdev)
drivers/gpu/drm/ttm/ttm_bo.c +5 -14
@@ -163,6 +163,7 @@ static void ttm_bo_release_list(struct kref *list_kref)void ttm_bo_add_to_lru (struct ttm_buffer_object *bo) {struct ttm_bo_device *bdev = bo->bdev; +struct ttm_mem_type_manager *man;lockdep_assert_held (&bo->resv->lock.base);
@@ -170,11 +171,13 @@ void ttm_bo_add_to_lru(struct ttm_buffer_object *bo)BUG_ON (!list_empty (&bo->lru)); -list_add (&bo->lru, bdev->driver->lru_tail (bo)); + man = &bdev->man[bo->mem.mem_type]; +list_add_tail (&bo->lru, &man->lru[bo->priority]);kref_get (&bo->list_kref);if (bo->ttm && !(bo->ttm->page_flags &TTM_PAGE_FLAG_SG )) { -list_add (&bo->swap, bdev->driver->swap_lru_tail (bo)); +list_add_tail (&bo->swap, + &bo->glob->swap_lru[bo->priority]);kref_get (&bo->list_kref); } }
@@ -240,18 +243,6 @@ void ttm_bo_move_to_lru_tail(struct ttm_buffer_object *bo) }EXPORT_SYMBOL (ttm_bo_move_to_lru_tail); -struct list_head *ttm_bo_default_lru_tail (struct ttm_buffer_object *bo) -{ -return bo->bdev->man[bo->mem.mem_type].lru[bo->priority].prev; -} -EXPORT_SYMBOL (ttm_bo_default_lru_tail); - -struct list_head *ttm_bo_default_swap_lru_tail (struct ttm_buffer_object *bo) -{ -return bo->glob->swap_lru[bo->priority].prev; -} -EXPORT_SYMBOL (ttm_bo_default_swap_lru_tail); -/* * Call bo->mutex locked. */
drivers/gpu/drm/virtio/virtgpu_ttm.c +0 -2
@@ -434,8 +434,6 @@ static struct ttm_bo_driver virtio_gpu_bo_driver = { .io_mem_free = &virtio_gpu_ttm_io_mem_free, .move_notify = &virtio_gpu_bo_move_notify, .swap_notify = &virtio_gpu_bo_swap_notify, - .lru_tail = &ttm_bo_default_lru_tail, - .swap_lru_tail = &ttm_bo_default_swap_lru_tail, };int virtio_gpu_ttm_init (struct virtio_gpu_device *vgdev)
drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c +0 -2
@@ -859,6 +859,4 @@ struct ttm_bo_driver vmw_bo_driver = { .fault_reserve_notify = &vmw_ttm_fault_reserve_notify, .io_mem_reserve = &vmw_ttm_io_mem_reserve, .io_mem_free = &vmw_ttm_io_mem_free, - .lru_tail = &ttm_bo_default_lru_tail, - .swap_lru_tail = &ttm_bo_default_swap_lru_tail, };
include/drm/ttm/ttm_bo_driver.h +0 -9
@@ -468,12 +468,6 @@ struct ttm_bo_driver {* Called with LRU lock held immediately before the removal. */ void (*lru_removal)(struct ttm_buffer_object *bo); - -/** -* Return the list_head after which a BO should be inserted in the LRU. -*/ -struct list_head *(*lru_tail)(struct ttm_buffer_object *bo); -struct list_head *(*swap_lru_tail)(struct ttm_buffer_object *bo); };/**
@@ -788,9 +782,6 @@ extern void ttm_mem_io_unlock(struct ttm_mem_type_manager *man);extern void ttm_bo_del_sub_from_lru (struct ttm_buffer_object *bo);extern void ttm_bo_add_to_lru (struct ttm_buffer_object *bo); -struct list_head *ttm_bo_default_lru_tail (struct ttm_buffer_object *bo); -struct list_head *ttm_bo_default_swap_lru_tail (struct ttm_buffer_object *bo); -/** * __ttm_bo_reserve: *
Signature · 6 lines
2.7.4 _______________________________________________ amd-gfx mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/amd-gfx
RE: [PATCH 5/6] drm/ttm: revert "implement LRU add callbacks v2"
Series patches:
and Test-by Roger.He
-----Original Message-----
From: Christian König [mailto:[email protected]]
Sent: Friday, January 13, 2017 5:51 PM
To: He, Hongbo <[email protected]>
Cc: [email protected]; [email protected]
Subject: [PATCH 5/6] drm/ttm: revert "implement LRU add callbacks v2"
From: Christian König <[email protected]>
The additional housekeeping had to much CPU overhead, let's use the BO priorities instead.
--- drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 2 -- drivers/gpu/drm/ast/ast_ttm.c | 2 -- drivers/gpu/drm/bochs/bochs_mm.c | 2 -- drivers/gpu/drm/cirrus/cirrus_ttm.c | 2 -- drivers/gpu/drm/mgag200/mgag200_ttm.c | 2 -- drivers/gpu/drm/nouveau/nouveau_bo.c | 2 -- drivers/gpu/drm/qxl/qxl_ttm.c | 2 -- drivers/gpu/drm/radeon/radeon_ttm.c | 2 -- drivers/gpu/drm/ttm/ttm_bo.c | 19 +++++-------------- drivers/gpu/drm/virtio/virtgpu_ttm.c | 2 -- drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c | 2 -- include/drm/ttm/ttm_bo_driver.h | 9 --------- 12 files changed, 5 insertions(+), 43 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +0 -2
@@ -1086,8 +1086,6 @@ static struct ttm_bo_driver amdgpu_bo_driver = { .fault_reserve_notify = &amdgpu_bo_fault_reserve_notify, .io_mem_reserve = &amdgpu_ttm_io_mem_reserve, .io_mem_free = &amdgpu_ttm_io_mem_free, - .lru_tail = &ttm_bo_default_lru_tail, - .swap_lru_tail = &ttm_bo_default_swap_lru_tail, }; int amdgpu_ttm_init(struct amdgpu_device *adev) diff --git a/drivers/gpu/drm/ast/ast_ttm.c b/drivers/gpu/drm/ast/ast_ttm.c index 2a1368f..50c910e 100644
drivers/gpu/drm/ast/ast_ttm.c +0 -2
@@ -236,8 +236,6 @@ struct ttm_bo_driver ast_bo_driver = { .verify_access = ast_bo_verify_access, .io_mem_reserve = &ast_ttm_io_mem_reserve, .io_mem_free = &ast_ttm_io_mem_free, - .lru_tail = &ttm_bo_default_lru_tail, - .swap_lru_tail = &ttm_bo_default_swap_lru_tail, }; int ast_mm_init(struct ast_private *ast) diff --git a/drivers/gpu/drm/bochs/bochs_mm.c b/drivers/gpu/drm/bochs/bochs_mm.c
index 099a3c6..e4c1125 100644
drivers/gpu/drm/bochs/bochs_mm.c +0 -2
@@ -205,8 +205,6 @@ struct ttm_bo_driver bochs_bo_driver = { .verify_access = bochs_bo_verify_access, .io_mem_reserve = &bochs_ttm_io_mem_reserve, .io_mem_free = &bochs_ttm_io_mem_free, - .lru_tail = &ttm_bo_default_lru_tail, - .swap_lru_tail = &ttm_bo_default_swap_lru_tail, }; int bochs_mm_init(struct bochs_device *bochs) diff --git a/drivers/gpu/drm/cirrus/cirrus_ttm.c b/drivers/gpu/drm/cirrus/cirrus_ttm.c
index d6da848..f53aa8f 100644
drivers/gpu/drm/cirrus/cirrus_ttm.c +0 -2
@@ -236,8 +236,6 @@ struct ttm_bo_driver cirrus_bo_driver = { .verify_access = cirrus_bo_verify_access, .io_mem_reserve = &cirrus_ttm_io_mem_reserve, .io_mem_free = &cirrus_ttm_io_mem_free, - .lru_tail = &ttm_bo_default_lru_tail, - .swap_lru_tail = &ttm_bo_default_swap_lru_tail, }; int cirrus_mm_init(struct cirrus_device *cirrus) diff --git a/drivers/gpu/drm/mgag200/mgag200_ttm.c b/drivers/gpu/drm/mgag200/mgag200_ttm.c
index 5e20220..657598b 100644
drivers/gpu/drm/mgag200/mgag200_ttm.c +0 -2
@@ -236,8 +236,6 @@ struct ttm_bo_driver mgag200_bo_driver = { .verify_access = mgag200_bo_verify_access, .io_mem_reserve = &mgag200_ttm_io_mem_reserve, .io_mem_free = &mgag200_ttm_io_mem_free, - .lru_tail = &ttm_bo_default_lru_tail, - .swap_lru_tail = &ttm_bo_default_swap_lru_tail, }; int mgag200_mm_init(struct mga_device *mdev) diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
index 5b0a28b..3949a74 100644
drivers/gpu/drm/nouveau/nouveau_bo.c +0 -2
@@ -1570,8 +1570,6 @@ struct ttm_bo_driver nouveau_bo_driver = { .fault_reserve_notify = &nouveau_ttm_fault_reserve_notify, .io_mem_reserve = &nouveau_ttm_io_mem_reserve, .io_mem_free = &nouveau_ttm_io_mem_free, - .lru_tail = &ttm_bo_default_lru_tail, - .swap_lru_tail = &ttm_bo_default_swap_lru_tail, }; struct nvkm_vma *
drivers/gpu/drm/qxl/qxl_ttm.c +0 -2
@@ -395,8 +395,6 @@ static struct ttm_bo_driver qxl_bo_driver = { .io_mem_reserve = &qxl_ttm_io_mem_reserve, .io_mem_free = &qxl_ttm_io_mem_free, .move_notify = &qxl_bo_move_notify, - .lru_tail = &ttm_bo_default_lru_tail, - .swap_lru_tail = &ttm_bo_default_swap_lru_tail, }; int qxl_ttm_init(struct qxl_device *qdev) diff --git a/drivers/gpu/drm/radeon/radeon_ttm.c b/drivers/gpu/drm/radeon/radeon_ttm.c
index 0cf03cc..d610481 100644
drivers/gpu/drm/radeon/radeon_ttm.c +0 -2
@@ -871,8 +871,6 @@ static struct ttm_bo_driver radeon_bo_driver = { .fault_reserve_notify = &radeon_bo_fault_reserve_notify, .io_mem_reserve = &radeon_ttm_io_mem_reserve, .io_mem_free = &radeon_ttm_io_mem_free, - .lru_tail = &ttm_bo_default_lru_tail, - .swap_lru_tail = &ttm_bo_default_swap_lru_tail, }; int radeon_ttm_init(struct radeon_device *rdev) diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c index f078b43..59fac2f 100644
drivers/gpu/drm/ttm/ttm_bo.c +5 -12
@@ -163,6 +163,7 @@ static void ttm_bo_release_list(struct kref *list_kref) void ttm_bo_add_to_lru(struct ttm_buffer_object *bo) {struct ttm_bo_device *bdev = bo->bdev; +struct ttm_mem_type_manager *man;lockdep_assert_held (&bo->resv->lock.base);
@@ -170,11 +171,13 @@ void ttm_bo_add_to_lru(struct ttm_buffer_object *bo)BUG_ON (!list_empty (&bo->lru)); -list_add (&bo->lru, bdev->driver->lru_tail (bo)); + man = &bdev->man[bo->mem.mem_type]; +list_add_tail (&bo->lru, &man->lru[bo->priority]);kref_get (&bo->list_kref);if (bo->ttm && !(bo->ttm->page_flags &TTM_PAGE_FLAG_SG )) { -list_add (&bo->swap, bdev->driver->swap_lru_tail (bo)); +list_add_tail (&bo->swap, + &bo->glob->swap_lru[bo->priority]);kref_get (&bo->list_kref); } }
@@ -240,18 +243,6 @@ void ttm_bo_move_to_lru_tail(struct ttm_buffer_object *bo) } EXPORT_SYMBOL(ttm_bo_move_to_lru_tail); -struct list_head *ttm_bo_default_lru_tail(struct ttm_buffer_object *bo) -{ -return bo->bdev->man[bo->mem.mem_type].lru[bo->priority].prev; -} -EXPORT_SYMBOL (ttm_bo_default_lru_tail); - -struct list_head *ttm_bo_default_swap_lru_tail(struct ttm_buffer_object *bo) -{ -return bo->glob->swap_lru[bo->priority].prev; -} -EXPORT_SYMBOL (ttm_bo_default_swap_lru_tail); -/* * Call bo->mutex locked. */
drivers/gpu/drm/virtio/virtgpu_ttm.c +0 -2
@@ -434,8 +434,6 @@ static struct ttm_bo_driver virtio_gpu_bo_driver = { .io_mem_free = &virtio_gpu_ttm_io_mem_free, .move_notify = &virtio_gpu_bo_move_notify, .swap_notify = &virtio_gpu_bo_swap_notify, - .lru_tail = &ttm_bo_default_lru_tail, - .swap_lru_tail = &ttm_bo_default_swap_lru_tail, }; int virtio_gpu_ttm_init(struct virtio_gpu_device *vgdev) diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
index caa279b..1de9669 100644
drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c +0 -2
@@ -859,6 +859,4 @@ struct ttm_bo_driver vmw_bo_driver = { .fault_reserve_notify = &vmw_ttm_fault_reserve_notify, .io_mem_reserve = &vmw_ttm_io_mem_reserve, .io_mem_free = &vmw_ttm_io_mem_free, - .lru_tail = &ttm_bo_default_lru_tail, - .swap_lru_tail = &ttm_bo_default_swap_lru_tail, };
include/drm/ttm/ttm_bo_driver.h +0 -9
@@ -468,12 +468,6 @@ struct ttm_bo_driver {* Called with LRU lock held immediately before the removal. */ void (*lru_removal)(struct ttm_buffer_object *bo); - -/** -* Return the list_head after which a BO should be inserted in the LRU. -*/ -struct list_head *(*lru_tail)(struct ttm_buffer_object *bo); -struct list_head *(*swap_lru_tail)(struct ttm_buffer_object *bo); };/**
@@ -788,9 +782,6 @@ extern void ttm_mem_io_unlock(struct ttm_mem_type_manager *man); extern void ttm_bo_del_sub_from_lru(struct ttm_buffer_object *bo); extern void ttm_bo_add_to_lru(struct ttm_buffer_object *bo); -struct list_head *ttm_bo_default_lru_tail(struct ttm_buffer_object *bo); -struct list_head *ttm_bo_default_swap_lru_tail(struct ttm_buffer_object *bo); -/** * __ttm_bo_reserve: * --
2.7.4
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
A message that is not in the archive (sent off-list or not yet received): 1484301071-21703-6-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org
Re: [PATCH 5/6] drm/ttm: revert "implement LRU add callbacks v2"
Minor typo
Quoted 3 lines · Christian König
From: Christian König <[email protected]>
The additional housekeeping had to much CPU overhead,
^
Quoted 227 lines
let's use the BO priorities instead.
Signed-off-by Christian König <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 2 -- drivers/gpu/drm/ast/ast_ttm.c | 2 -- drivers/gpu/drm/bochs/bochs_mm.c | 2 -- drivers/gpu/drm/cirrus/cirrus_ttm.c | 2 -- drivers/gpu/drm/mgag200/mgag200_ttm.c | 2 -- drivers/gpu/drm/nouveau/nouveau_bo.c | 2 -- drivers/gpu/drm/qxl/qxl_ttm.c | 2 -- drivers/gpu/drm/radeon/radeon_ttm.c | 2 -- drivers/gpu/drm/ttm/ttm_bo.c | 19 +++++-------------- drivers/gpu/drm/virtio/virtgpu_ttm.c | 2 -- drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c | 2 -- include/drm/ttm/ttm_bo_driver.h | 9 --------- 12 files changed, 5 insertions(+), 43 deletions(-)drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +0
-2@@ -1086,8 +1086,6 @@ static struct ttm_bo_driver amdgpu_bo_driver = { .fault_reserve_notify = &amdgpu_bo_fault_reserve_notify, .io_mem_reserve = &amdgpu_ttm_io_mem_reserve, .io_mem_free = &amdgpu_ttm_io_mem_free, - .lru_tail = &ttm_bo_default_lru_tail, - .swap_lru_tail = &ttm_bo_default_swap_lru_tail, };int amdgpu_ttm_init (struct amdgpu_device *adev)drivers/gpu/drm/ast/ast_ttm.c +0
-2@@ -236,8 +236,6 @@ struct ttm_bo_driver ast_bo_driver = { .verify_access = ast_bo_verify_access, .io_mem_reserve = &ast_ttm_io_mem_reserve, .io_mem_free = &ast_ttm_io_mem_free, - .lru_tail = &ttm_bo_default_lru_tail, - .swap_lru_tail = &ttm_bo_default_swap_lru_tail, };int ast_mm_init (struct ast_private *ast)drivers/gpu/drm/bochs/bochs_mm.c +0
-2@@ -205,8 +205,6 @@ struct ttm_bo_driver bochs_bo_driver = { .verify_access = bochs_bo_verify_access, .io_mem_reserve = &bochs_ttm_io_mem_reserve, .io_mem_free = &bochs_ttm_io_mem_free, - .lru_tail = &ttm_bo_default_lru_tail, - .swap_lru_tail = &ttm_bo_default_swap_lru_tail, };int bochs_mm_init (struct bochs_device *bochs)drivers/gpu/drm/cirrus/cirrus_ttm.c +0
-2@@ -236,8 +236,6 @@ struct ttm_bo_driver cirrus_bo_driver = { .verify_access = cirrus_bo_verify_access, .io_mem_reserve = &cirrus_ttm_io_mem_reserve, .io_mem_free = &cirrus_ttm_io_mem_free, - .lru_tail = &ttm_bo_default_lru_tail, - .swap_lru_tail = &ttm_bo_default_swap_lru_tail, };int cirrus_mm_init (struct cirrus_device *cirrus)drivers/gpu/drm/mgag200/mgag200_ttm.c +0
-2@@ -236,8 +236,6 @@ struct ttm_bo_driver mgag200_bo_driver = { .verify_access = mgag200_bo_verify_access, .io_mem_reserve = &mgag200_ttm_io_mem_reserve, .io_mem_free = &mgag200_ttm_io_mem_free, - .lru_tail = &ttm_bo_default_lru_tail, - .swap_lru_tail = &ttm_bo_default_swap_lru_tail, };int mgag200_mm_init (struct mga_device *mdev)drivers/gpu/drm/nouveau/nouveau_bo.c +0
-2@@ -1570,8 +1570,6 @@ struct ttm_bo_driver nouveau_bo_driver = { .fault_reserve_notify = &nouveau_ttm_fault_reserve_notify, .io_mem_reserve = &nouveau_ttm_io_mem_reserve, .io_mem_free = &nouveau_ttm_io_mem_free, - .lru_tail = &ttm_bo_default_lru_tail, - .swap_lru_tail = &ttm_bo_default_swap_lru_tail, }; struct nvkm_vma *drivers/gpu/drm/qxl/qxl_ttm.c +0
-2@@ -395,8 +395,6 @@ static struct ttm_bo_driver qxl_bo_driver = { .io_mem_reserve = &qxl_ttm_io_mem_reserve, .io_mem_free = &qxl_ttm_io_mem_free, .move_notify = &qxl_bo_move_notify, - .lru_tail = &ttm_bo_default_lru_tail, - .swap_lru_tail = &ttm_bo_default_swap_lru_tail, };int qxl_ttm_init (struct qxl_device *qdev)drivers/gpu/drm/radeon/radeon_ttm.c +0
-2@@ -871,8 +871,6 @@ static struct ttm_bo_driver radeon_bo_driver = { .fault_reserve_notify = &radeon_bo_fault_reserve_notify, .io_mem_reserve = &radeon_ttm_io_mem_reserve, .io_mem_free = &radeon_ttm_io_mem_free, - .lru_tail = &ttm_bo_default_lru_tail, - .swap_lru_tail = &ttm_bo_default_swap_lru_tail, };int radeon_ttm_init (struct radeon_device *rdev)drivers/gpu/drm/ttm/ttm_bo.c +5
-14@@ -163,6 +163,7 @@ static void ttm_bo_release_list(struct kref *list_kref)void ttm_bo_add_to_lru (struct ttm_buffer_object *bo) {struct ttm_bo_device *bdev = bo->bdev; +struct ttm_mem_type_manager *man;lockdep_assert_held (&bo->resv->lock.base);@@ -170,11 +171,13 @@ void ttm_bo_add_to_lru(struct ttm_buffer_object *bo)BUG_ON (!list_empty (&bo->lru)); -list_add (&bo->lru, bdev->driver->lru_tail (bo)); + man = &bdev->man[bo->mem.mem_type]; +list_add_tail (&bo->lru, &man->lru[bo->priority]);kref_get (&bo->list_kref);if (bo->ttm && !(bo->ttm->page_flags &TTM_PAGE_FLAG_SG )) { -list_add (&bo->swap, bdev->driver->swap_lru_tail (bo)); +list_add_tail (&bo->swap, + &bo->glob->swap_lru[bo->priority]);kref_get (&bo->list_kref); } }@@ -240,18 +243,6 @@ void ttm_bo_move_to_lru_tail(struct ttm_buffer_object *bo) }EXPORT_SYMBOL (ttm_bo_move_to_lru_tail); -struct list_head *ttm_bo_default_lru_tail (struct ttm_buffer_object *bo) -{ -return bo->bdev->man[bo->mem.mem_type].lru[bo->priority].prev; -} -EXPORT_SYMBOL (ttm_bo_default_lru_tail); - -struct list_head *ttm_bo_default_swap_lru_tail (struct ttm_buffer_object *bo) -{ -return bo->glob->swap_lru[bo->priority].prev; -} -EXPORT_SYMBOL (ttm_bo_default_swap_lru_tail); -/* * Call bo->mutex locked. */ drivers/gpu/drm/virtio/virtgpu_ttm.c +0
-2@@ -434,8 +434,6 @@ static struct ttm_bo_driver virtio_gpu_bo_driver = { .io_mem_free = &virtio_gpu_ttm_io_mem_free, .move_notify = &virtio_gpu_bo_move_notify, .swap_notify = &virtio_gpu_bo_swap_notify, - .lru_tail = &ttm_bo_default_lru_tail, - .swap_lru_tail = &ttm_bo_default_swap_lru_tail, };int virtio_gpu_ttm_init (struct virtio_gpu_device *vgdev)drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c +0
-2@@ -859,6 +859,4 @@ struct ttm_bo_driver vmw_bo_driver = { .fault_reserve_notify = &vmw_ttm_fault_reserve_notify, .io_mem_reserve = &vmw_ttm_io_mem_reserve, .io_mem_free = &vmw_ttm_io_mem_free, - .lru_tail = &ttm_bo_default_lru_tail, - .swap_lru_tail = &ttm_bo_default_swap_lru_tail, };include/drm/ttm/ttm_bo_driver.h +0
-9@@ -468,12 +468,6 @@ struct ttm_bo_driver {* Called with LRU lock held immediately before the removal. */ void (*lru_removal)(struct ttm_buffer_object *bo); - -/** -* Return the list_head after which a BO should be inserted in the LRU. -*/ -struct list_head *(*lru_tail)(struct ttm_buffer_object *bo); -struct list_head *(*swap_lru_tail)(struct ttm_buffer_object *bo); };/** @@ -788,9 +782,6 @@ extern void ttm_mem_io_unlock(struct ttm_mem_type_manager *man);extern void ttm_bo_del_sub_from_lru (struct ttm_buffer_object *bo);extern void ttm_bo_add_to_lru (struct ttm_buffer_object *bo); -struct list_head *ttm_bo_default_lru_tail (struct ttm_buffer_object *bo); -struct list_head *ttm_bo_default_swap_lru_tail (struct ttm_buffer_object *bo); -/** * __ttm_bo_reserve: * Signature · 6 lines
2.7.4 _______________________________________________ dri-devel mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/dri-devel
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
PATCH 6/6 drm/ttm: revert "add optional LRU removal callback v2"
In reply to a message not in the archiveFrom: Christian König <[email protected]>
Without the custom LRU management the callback is not used any more.
--- drivers/gpu/drm/ttm/ttm_bo.c | 12 +++++------- include/drm/ttm/ttm_bo_driver.h | 6 ------ 2 files changed, 5 insertions(+), 13 deletions(-)
drivers/gpu/drm/ttm/ttm_bo.c +5 -7
@@ -186,12 +186,8 @@ EXPORT_SYMBOL(ttm_bo_add_to_lru);int ttm_bo_del_from_lru (struct ttm_buffer_object *bo) { -struct ttm_bo_device *bdev = bo->bdev;int put_count =0 ; -if (bdev->driver->lru_removal) - bdev->driver->lru_removal (bo); -if (!list_empty (&bo->swap)) {list_del_init (&bo->swap); ++put_count;
@@ -201,6 +197,11 @@ int ttm_bo_del_from_lru(struct ttm_buffer_object *bo) ++put_count; } +/* +* TODO: Add a driver hook to delete from +* driver-specific LRU's here. +*/ +return put_count; }
@@ -234,9 +235,6 @@ void ttm_bo_move_to_lru_tail(struct ttm_buffer_object *bo)lockdep_assert_held (&bo->resv->lock.base); -if (bdev->driver->lru_removal) - bdev->driver->lru_removal (bo); - put_count =ttm_bo_del_from_lru (bo);ttm_bo_list_ref_sub (bo, put_count, true);ttm_bo_add_to_lru (bo);
include/drm/ttm/ttm_bo_driver.h +0 -6
@@ -462,12 +462,6 @@ struct ttm_bo_driver { struct ttm_mem_reg *mem);void (*io_mem_free)(struct ttm_bo_device *bdev,struct ttm_mem_reg *mem); - -/** -* Optional driver callback for when BO is removed from the LRU. -* Called with LRU lock held immediately before the removal. -*/ -void (*lru_removal)(struct ttm_buffer_object *bo); };/**
Signature · 6 lines
2.7.4 _______________________________________________ amd-gfx mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/amd-gfx
RE: [PATCH 6/6] drm/ttm: revert "add optional LRU removal callback v2"
Good way to keep original idea and remove CPU overhead.
-----Original Message-----
From: Christian König [mailto:[email protected]]
Sent: Friday, January 13, 2017 5:51 PM
To: He, Hongbo <[email protected]>
Cc: [email protected]; [email protected]
Subject: [PATCH 6/6] drm/ttm: revert "add optional LRU removal callback v2"
From: Christian König <[email protected]>
Without the custom LRU management the callback is not used any more.
--- drivers/gpu/drm/ttm/ttm_bo.c | 12 +++++------- include/drm/ttm/ttm_bo_driver.h | 6 ------ 2 files changed, 5 insertions(+), 13 deletions(-)
drivers/gpu/drm/ttm/ttm_bo.c +5 -7
@@ -186,12 +186,8 @@ EXPORT_SYMBOL(ttm_bo_add_to_lru);int ttm_bo_del_from_lru (struct ttm_buffer_object *bo) { -struct ttm_bo_device *bdev = bo->bdev;int put_count =0 ; -if (bdev->driver->lru_removal) - bdev->driver->lru_removal (bo); -if (!list_empty (&bo->swap)) {list_del_init (&bo->swap); ++put_count;
@@ -201,6 +197,11 @@ int ttm_bo_del_from_lru(struct ttm_buffer_object *bo) ++put_count; } +/* +* TODO: Add a driver hook to delete from +* driver-specific LRU's here. +*/ +return put_count; }
@@ -234,9 +235,6 @@ void ttm_bo_move_to_lru_tail(struct ttm_buffer_object *bo)lockdep_assert_held (&bo->resv->lock.base); -if (bdev->driver->lru_removal) - bdev->driver->lru_removal (bo); - put_count =ttm_bo_del_from_lru (bo);ttm_bo_list_ref_sub (bo, put_count, true);ttm_bo_add_to_lru (bo);
include/drm/ttm/ttm_bo_driver.h +0 -6
@@ -462,12 +462,6 @@ struct ttm_bo_driver { struct ttm_mem_reg *mem);void (*io_mem_free)(struct ttm_bo_device *bdev,struct ttm_mem_reg *mem); - -/** -* Optional driver callback for when BO is removed from the LRU. -* Called with LRU lock held immediately before the removal. -*/ -void (*lru_removal)(struct ttm_buffer_object *bo); };/**
--
2.7.4
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
Re: [PATCH 6/6] drm/ttm: revert "add optional LRU removal callback v2"
In reply to Christian KönigPathces 1, 2, 5, 6:
Quoted 73 lines · Christian König
From: Christian König <[email protected]>
Without the custom LRU management the callback is not used any more.
Signed-off-by Christian König <[email protected]>--- drivers/gpu/drm/ttm/ttm_bo.c | 12 +++++------- include/drm/ttm/ttm_bo_driver.h | 6 ------ 2 files changed, 5 insertions(+), 13 deletions(-)drivers/gpu/drm/ttm/ttm_bo.c +5
-7@@ -186,12 +186,8 @@ EXPORT_SYMBOL(ttm_bo_add_to_lru);int ttm_bo_del_from_lru (struct ttm_buffer_object *bo) { -struct ttm_bo_device *bdev = bo->bdev;int put_count =0 ; -if (bdev->driver->lru_removal) - bdev->driver->lru_removal (bo); -if (!list_empty (&bo->swap)) {list_del_init (&bo->swap); ++put_count;@@ -201,6 +197,11 @@ int ttm_bo_del_from_lru(struct ttm_buffer_object *bo) ++put_count; } +/* +* TODO: Add a driver hook to delete from +* driver-specific LRU's here. +*/ +return put_count; }@@ -234,9 +235,6 @@ void ttm_bo_move_to_lru_tail(struct ttm_buffer_object *bo)lockdep_assert_held (&bo->resv->lock.base); -if (bdev->driver->lru_removal) - bdev->driver->lru_removal (bo); - put_count =ttm_bo_del_from_lru (bo);ttm_bo_list_ref_sub (bo, put_count, true);ttm_bo_add_to_lru (bo);include/drm/ttm/ttm_bo_driver.h +0
-6@@ -462,12 +462,6 @@ struct ttm_bo_driver { struct ttm_mem_reg *mem);void (*io_mem_free)(struct ttm_bo_device *bdev,struct ttm_mem_reg *mem); - -/** -* Optional driver callback for when BO is removed from the LRU. -* Called with LRU lock held immediately before the removal. -*/ -void (*lru_removal)(struct ttm_buffer_object *bo); };/** Signature · 6 lines
2.7.4 _______________________________________________ dri-devel mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/dri-devel
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
Hi Sinclair,
you probably have missed this set of patches I've send to the dri-devel mailing list a while ago.
It basically replaces some work for TTM I've did back in April last year and you reviewed with a different implementation.
The original callback design turned out to have to much overhead which made this change necessary. Could you take a look at the new implementation as well?
Thanks in advance,
Christian.
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
PATCH 3/6 drm/amdgpu: user BO priority instead of self coding it
In reply to Christian KönigFrom: Christian König <[email protected]>
Keeping groups of BOs on the LRU is to time consuming on command submission.
Instead use the newly added BO priority to give a certain eviction order.
--- drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 3 ++ drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 69 +----------------------------- drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h | 12 ------ 3 files changed, 5 insertions(+), 79 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_object.c +3 -0
@@ -386,6 +386,9 @@ int amdgpu_bo_create_restricted(struct amdgpu_device *adev,if (unlikely (r !=0 ))return r; + bo->tbo.priority =ilog2 (bo->tbo.num_pages); + bo->tbo.priority =min (bo->tbo.priority,TTM_MAX_BO_PRIORITY -1 ); +if (flags &AMDGPU_GEM_CREATE_VRAM_CLEARED && bo->tbo.mem.placement &TTM_PL_FLAG_VRAM ) {struct fence *fence;
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +2 -67
91 lines not on this page · raw message
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h +0 -12
@@ -34,13 +34,6 @@#define AMDGPU_PL_FLAG_GWS (TTM_PL_FLAG_PRIV << 1)#define AMDGPU_PL_FLAG_OA (TTM_PL_FLAG_PRIV << 2) -#define AMDGPU_TTM_LRU_SIZE 20 - -struct amdgpu_mman_lru { -struct list_head *lru[TTM_NUM_MEM_TYPES ]; -struct list_head *swap_lru; -}; -struct amdgpu_mman {struct ttm_bo_global_ref bo_global_ref;struct drm_global_reference mem_global_ref;
@@ -58,11 +51,6 @@ struct amdgpu_mman {struct amdgpu_ring *buffer_funcs_ring;/* Scheduler entity for buffer moves */ struct amd_sched_entity entity; - -/* custom LRU management */ -struct amdgpu_mman_lru log2_size[AMDGPU_TTM_LRU_SIZE ]; -/* guard for log2_size array, don't add anything in between */ -struct amdgpu_mman_lru guard; };extern const struct ttm_mem_type_manager_func amdgpu_gtt_mgr_func;
Signature · 6 lines
2.7.4 _______________________________________________ dri-devel mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/dri-devel
RE: [PATCH 3/6] drm/amdgpu: user BO priority instead of self coding it
-----Original Message-----
From: Christian König [mailto:[email protected]]
Sent: Friday, January 13, 2017 5:51 PM
To: He, Hongbo <[email protected]>
Cc: [email protected]; [email protected]
Subject: [PATCH 3/6] drm/amdgpu: user BO priority instead of self coding it
From: Christian König <[email protected]>
Keeping groups of BOs on the LRU is to time consuming on command submission.
Instead use the newly added BO priority to give a certain eviction order.
--- drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 3 ++ drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 69 +----------------------------- drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h | 12 ------ 3 files changed, 5 insertions(+), 79 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_object.c +3 -0
@@ -386,6 +386,9 @@ int amdgpu_bo_create_restricted(struct amdgpu_device *adev,if (unlikely (r !=0 ))return r; + bo->tbo.priority =ilog2 (bo->tbo.num_pages); + bo->tbo.priority =min (bo->tbo.priority,TTM_MAX_BO_PRIORITY -1 ); +if (flags &AMDGPU_GEM_CREATE_VRAM_CLEARED && bo->tbo.mem.placement &TTM_PL_FLAG_VRAM ) {struct fence *fence;
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +0 -46
52 lines not on this page · raw message
.io_mem_reserve = &amdgpu_ttm_io_mem_reserve, .io_mem_free = &amdgpu_ttm_io_mem_free,
- .lru_removal = &amdgpu_ttm_lru_removal,
- .lru_tail = &amdgpu_ttm_lru_tail,
- .swap_lru_tail = &amdgpu_ttm_swap_lru_tail,
+ .lru_tail = &ttm_bo_default_lru_tail,
+ .swap_lru_tail = &ttm_bo_default_swap_lru_tail,
};
int amdgpu_ttm_init(struct amdgpu_device *adev) {
- unsigned i, j;
int r;
r = amdgpu_ttm_global_init(adev);
@@ -1161,19 +1109,6 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)
DRM_ERROR("failed initializing buffer object driver(%d).\n", r);
return r;
}
-
- for (i = 0; i < AMDGPU_TTM_LRU_SIZE; ++i) {
- struct amdgpu_mman_lru *lru = &adev->mman.log2_size[i];
-
- for (j = 0; j < TTM_NUM_MEM_TYPES; ++j)
- lru->lru[j] = &adev->mman.bdev.man[j].lru[0];
- lru->swap_lru = &adev->mman.bdev.glob->swap_lru[0];
- }
-
- for (j = 0; j < TTM_NUM_MEM_TYPES; ++j)
- adev->mman.guard.lru[j] = NULL;
- adev->mman.guard.swap_lru = NULL;
-
adev->mman.initialized = true;
r = ttm_bo_init_mm(&adev->mman.bdev, TTM_PL_VRAM,
adev->mc.real_vram_size >> PAGE_SHIFT); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
index d1c00c0..0eabbb2 100644
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h +0 -12
@@ -34,13 +34,6 @@#define AMDGPU_PL_FLAG_GWS (TTM_PL_FLAG_PRIV << 1)#define AMDGPU_PL_FLAG_OA (TTM_PL_FLAG_PRIV << 2) -#define AMDGPU_TTM_LRU_SIZE 20 - -struct amdgpu_mman_lru { -struct list_head *lru[TTM_NUM_MEM_TYPES ]; -struct list_head *swap_lru; -}; -struct amdgpu_mman {struct ttm_bo_global_ref bo_global_ref;struct drm_global_reference mem_global_ref;
@@ -58,11 +51,6 @@ struct amdgpu_mman {struct amdgpu_ring *buffer_funcs_ring;/* Scheduler entity for buffer moves */ struct amd_sched_entity entity; - -/* custom LRU management */ -struct amdgpu_mman_lru log2_size[AMDGPU_TTM_LRU_SIZE ]; -/* guard for log2_size array, don't add anything in between */ -struct amdgpu_mman_lru guard; };extern const struct ttm_mem_type_manager_func amdgpu_gtt_mgr_func;
--
2.7.4
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
PATCH 4/6 drm/amdgpu: double the priority of kernel allocations
In reply to Christian KönigFrom: Christian König <[email protected]>
Give kernel allocations a higher priority cause it is often more work to swap them back in.
--- drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 2 ++ 1 file changed, 2 insertions(+)
drivers/gpu/drm/amd/amdgpu/amdgpu_object.c +2 -0
@@ -387,6 +387,8 @@ int amdgpu_bo_create_restricted(struct amdgpu_device *adev,return r; bo->tbo.priority =ilog2 (bo->tbo.num_pages); +if (kernel) + bo->tbo.priority *=2 ; bo->tbo.priority =min (bo->tbo.priority,TTM_MAX_BO_PRIORITY -1 ); if (flags & AMDGPU_GEM_CREATE_VRAM_CLEARED &&
Signature · 6 lines
2.7.4 _______________________________________________ dri-devel mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/dri-devel
A message that is not in the archive (sent off-list or not yet received): 1484301071-21703-5-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org
He, Hongbo
Reviewed-by Roger.HeRE: [PATCH 4/6] drm/amdgpu: double the priority of kernel allocations
-----Original Message-----
From: Christian König [mailto:[email protected]]
Sent: Friday, January 13, 2017 5:51 PM
To: He, Hongbo <[email protected]>
Cc: [email protected]; [email protected]
Subject: [PATCH 4/6] drm/amdgpu: double the priority of kernel allocations
From: Christian König <[email protected]>
Give kernel allocations a higher priority cause it is often more work to swap them back in.
--- drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 2 ++ 1 file changed, 2 insertions(+)
drivers/gpu/drm/amd/amdgpu/amdgpu_object.c +2 -0
@@ -387,6 +387,8 @@ int amdgpu_bo_create_restricted(struct amdgpu_device *adev,return r; bo->tbo.priority =ilog2 (bo->tbo.num_pages); +if (kernel) + bo->tbo.priority *=2 ; bo->tbo.priority =min (bo->tbo.priority,TTM_MAX_BO_PRIORITY -1 ); if (flags & AMDGPU_GEM_CREATE_VRAM_CLEARED &&
--
2.7.4
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx