drm/amdgpu: make all per vm BOs list
Subsystems (2)
PATCH 1/3 drm/amdgpu: make all per vm BOs list
the list decides the lru order.
--- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 7 +++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 6 ++++++ 2 files changed, 13 insertions(+)
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +7 -0
@@ -1866,6 +1866,7 @@ struct amdgpu_bo_va *amdgpu_vm_bo_add(struct amdgpu_device *adev, } bo_va->base.vm = vm; bo_va->base.bo = bo; +INIT_LIST_HEAD (&bo_va->base.vm_bo);INIT_LIST_HEAD (&bo_va->base.bo_list);INIT_LIST_HEAD (&bo_va->base.vm_status);
@@ -1881,6 +1882,10 @@ struct amdgpu_bo_va *amdgpu_vm_bo_add(struct amdgpu_device *adev,if (bo->tbo.resv != vm->root.base.bo->tbo.resv)return bo_va; +spin_lock (&vm->status_lock); +list_add_tail (&bo_va->base.vm_bo, &vm->vm_bo_list); +spin_unlock (&vm->status_lock); +if (bo->preferred_domains &amdgpu_mem_type_to_domain (bo->tbo.mem.mem_type))return bo_va;
@@ -2237,6 +2242,7 @@ void amdgpu_vm_bo_rmv(struct amdgpu_device *adev,spin_lock (&vm->status_lock);list_del (&bo_va->base.vm_status); +list_del (&bo_va->base.vm_bo);spin_unlock (&vm->status_lock);list_for_each_entry_safe (mapping, next, &bo_va->valids, list) {
@@ -2409,6 +2415,7 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm,for (i =0 ; i <AMDGPU_MAX_VMHUBS ; i++) vm->reserved_vmid[i] =NULL ;spin_lock_init (&vm->status_lock); +INIT_LIST_HEAD (&vm->vm_bo_list);INIT_LIST_HEAD (&vm->evicted);INIT_LIST_HEAD (&vm->relocated);INIT_LIST_HEAD (&vm->moved);
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +6 -0
@@ -144,6 +144,9 @@ struct amdgpu_vm_bo_base {struct amdgpu_vm *vm;struct amdgpu_bo *bo; +/* protected by vm status lock */ +struct list_head vm_bo; +/* protected by bo being reserved */ struct list_head bo_list;
@@ -177,6 +180,9 @@ struct amdgpu_vm {/* protecting invalidated */ spinlock_t status_lock; +/* protected by status lock */ +struct list_head vm_bo_list; +/* BOs who needs a validation */ struct list_head evicted;
Signature · 6 lines
2.14.1 _______________________________________________ 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): [email protected]
PATCH 2/3 drm/amdgpu: refresh per vm bo lru
--- drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 2 ++ drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 15 +++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 1 + 3 files changed, 18 insertions(+)
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +2 -0
@@ -919,6 +919,8 @@ static int amdgpu_bo_vm_update_pte(struct amdgpu_cs_parser *p) } } +amdgpu_vm_refresh_lru (adev, vm); +return r; }
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +15 -0
@@ -1902,6 +1902,21 @@ struct amdgpu_bo_va *amdgpu_vm_bo_add(struct amdgpu_device *adev,return bo_va; } +void amdgpu_vm_refresh_lru (struct amdgpu_device *adev,struct amdgpu_vm *vm) +{ +struct ttm_bo_global *glob = adev->mman.bdev.glob; +struct amdgpu_vm_bo_base *bo_base; + +spin_lock (&vm->status_lock); +list_for_each_entry (bo_base, &vm->vm_bo_list, vm_bo) { +spin_lock (&glob->lru_lock); +ttm_bo_move_to_lru_tail (&bo_base->bo->tbo); +if (bo_base->bo->shadow) +ttm_bo_move_to_lru_tail (&bo_base->bo->shadow->tbo); +spin_unlock (&glob->lru_lock); + } +spin_unlock (&vm->status_lock); +}/** * amdgpu_vm_bo_insert_mapping - insert a new mapping
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +1 -0
@@ -285,6 +285,7 @@ int amdgpu_vm_clear_freed(struct amdgpu_device *adev, struct dma_fence **fence);int amdgpu_vm_handle_moved (struct amdgpu_device *adev,struct amdgpu_vm *vm); +void amdgpu_vm_refresh_lru (struct amdgpu_device *adev,struct amdgpu_vm *vm);int amdgpu_vm_bo_update (struct amdgpu_device *adev,struct amdgpu_bo_va *bo_va,bool clear);
Signature · 6 lines
2.14.1 _______________________________________________ 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): [email protected]
Christian König
NAKRe: [PATCH 2/3] drm/amdgpu: refresh per vm bo lru
NAK, we already tried that and it is really not a good idea because it massively increases the per submission overhead.
Christian.
Am 27.03.2018 um 12:16 schrieb Chunming Zhou:
Quoted 52 lines
Change-Id Ibad84ed585b0746867a5f4cd1eadc2273e7cf596Signed-off-by Chunming Zhou <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 2 ++ drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 15 +++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 1 + 3 files changed, 18 insertions(+)drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +2
-0@@ -919,6 +919,8 @@ static int amdgpu_bo_vm_update_pte(struct amdgpu_cs_parser *p) } } +amdgpu_vm_refresh_lru (adev, vm); +return r; }drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +15
-0@@ -1902,6 +1902,21 @@ struct amdgpu_bo_va *amdgpu_vm_bo_add(struct amdgpu_device *adev,return bo_va; } +void amdgpu_vm_refresh_lru (struct amdgpu_device *adev,struct amdgpu_vm *vm) +{ +struct ttm_bo_global *glob = adev->mman.bdev.glob; +struct amdgpu_vm_bo_base *bo_base; + +spin_lock (&vm->status_lock); +list_for_each_entry (bo_base, &vm->vm_bo_list, vm_bo) { +spin_lock (&glob->lru_lock); +ttm_bo_move_to_lru_tail (&bo_base->bo->tbo); +if (bo_base->bo->shadow) +ttm_bo_move_to_lru_tail (&bo_base->bo->shadow->tbo); +spin_unlock (&glob->lru_lock); + } +spin_unlock (&vm->status_lock); +}/** * amdgpu_vm_bo_insert_mapping - insert a new mapping drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +0
-0
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h int amdgpu_vm_clear_freed(struct amdgpu_device *adev, · line 291
Quoted 7 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +1
-0@@ -285,6 +285,7 @@ int amdgpu_vm_clear_freed(struct amdgpu_device *adev, struct dma_fence **fence);int amdgpu_vm_handle_moved (struct amdgpu_device *adev,struct amdgpu_vm *vm); +void amdgpu_vm_refresh_lru (struct amdgpu_device *adev,struct amdgpu_vm *vm);int amdgpu_vm_bo_update (struct amdgpu_device *adev,struct amdgpu_bo_va *bo_va,bool clear);
_______________________________________________
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): 6afa6cca-d59c-9e64-f891-c43a741069d4-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org
Re: [PATCH 2/3] drm/amdgpu: refresh per vm bo lru
then how to keep unique lru order? any ideas?
To stable performance, we have to keep unique lru order, otherwise like the issue I look into, sometimes F1game is 40fps, sometimes 28fps...even re-validate allowed domains BO.
The left root cause is the moved BOs are not same.
send from Smartisan Pro
Christian K鰊ig <[email protected]> 于 2018年3月27日 下午6:50写道:
NAK, we already tried that and it is really not a good idea because it massively increases the per submission overhead.
Christian.
Am 27.03.2018 um 12:16 schrieb Chunming Zhou:
Quoted 52 lines
Change-Id Ibad84ed585b0746867a5f4cd1eadc2273e7cf596Signed-off-by Chunming Zhou <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 2 ++ drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 15 +++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 1 + 3 files changed, 18 insertions(+)drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +2
-0@@ -919,6 +919,8 @@ static int amdgpu_bo_vm_update_pte(struct amdgpu_cs_parser *p) } } +amdgpu_vm_refresh_lru (adev, vm); +return r; }drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +15
-0@@ -1902,6 +1902,21 @@ struct amdgpu_bo_va *amdgpu_vm_bo_add(struct amdgpu_device *adev,return bo_va; } +void amdgpu_vm_refresh_lru (struct amdgpu_device *adev,struct amdgpu_vm *vm) +{ +struct ttm_bo_global *glob = adev->mman.bdev.glob; +struct amdgpu_vm_bo_base *bo_base; + +spin_lock (&vm->status_lock); +list_for_each_entry (bo_base, &vm->vm_bo_list, vm_bo) { +spin_lock (&glob->lru_lock); +ttm_bo_move_to_lru_tail (&bo_base->bo->tbo); +if (bo_base->bo->shadow) +ttm_bo_move_to_lru_tail (&bo_base->bo->shadow->tbo); +spin_unlock (&glob->lru_lock); + } +spin_unlock (&vm->status_lock); +}/** * amdgpu_vm_bo_insert_mapping - insert a new mapping drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +0
-0
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h int amdgpu_vm_clear_freed(struct amdgpu_device *adev, · line 291
Quoted 7 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +1
-0@@ -285,6 +285,7 @@ int amdgpu_vm_clear_freed(struct amdgpu_device *adev, struct dma_fence **fence);int amdgpu_vm_handle_moved (struct amdgpu_device *adev,struct amdgpu_vm *vm); +void amdgpu_vm_refresh_lru (struct amdgpu_device *adev,struct amdgpu_vm *vm);int amdgpu_vm_bo_update (struct amdgpu_device *adev,struct amdgpu_bo_va *bo_va,bool clear);
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
Re: [PATCH 2/3] drm/amdgpu: refresh per vm bo lru
In reply to a message not in the archiveHow about we update the LRU only when we need to re-validate at least one BO?
BTW: We can easily walk all BOs which belong to a VM, skipping over the few which aren't per VM BOs should be trivial.
Christian.
Am 27.03.2018 um 13:56 schrieb Zhou, David(ChunMing):
Quoted 75 lines
then how to keep unique lru order? any ideas?
To stable performance, we have to keep unique lru order, otherwise like the issue I look into, sometimes F1game is 40fps, sometimes 28fps...even re-validate allowed domains BO.
The left root cause is the moved BOs are not same.
send from Smartisan Pro
Christian K鰊ig <[email protected]> 于 2018年3月27日 下午6:50写道:
NAK, we already tried that and it is really not a good idea because it massively increases the per submission overhead.
Christian.
Am 27.03.2018 um 12:16 schrieb Chunming Zhou:
Review
Quoted 9 lines
Change-Id Ibad84ed585b0746867a5f4cd1eadc2273e7cf596Signed-off-by Chunming Zhou <[email protected]>---
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 2 ++ drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 15 +++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 1 + 3 files changed, 18 insertions(+)(quoted diff) +0
-0
b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.cdrivers/gpu/drm/amd/amdgpu/amdgpu_cs.c static int amdgpu_bo_vm_update_pte(struct
Quoted 4 lines
index
383bf2d31c92..414e61799236100644drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +0
-0@@ -919,6 +919,8 @@ static int amdgpu_bo_vm_update_pte(struct
amdgpu_cs_parser*p)Review
Quoted 9 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +2
-0} } + amdgpu_vm_refresh_lru(adev, vm); + return r; }(quoted diff) +0
-0
b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.cdrivers/gpu/drm/amd/amdgpu/amdgpu_vm.c struct amdgpu_bo_va *amdgpu_vm_bo_add(struct
Quoted 4 lines
index
5e35e23511cf..8ad2bb705765100644drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +0
-0@@ -1902,6 +1902,21 @@ struct amdgpu_bo_va *amdgpu_vm_bo_add(struct
amdgpu_device*adev,drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c struct amdgpu_bo_va *amdgpu_vm_bo_add(struct amdgpu_device *adev,
Quoted 4 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +1
-0return bo_va; } +void amdgpu_vm_refresh_lru(struct amdgpu_device *adev, struct
amdgpu_vm*vm)Review
Quoted 18 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +14
-0+{ + struct ttm_bo_global *glob = adev->mman.bdev.glob; + struct amdgpu_vm_bo_base *bo_base; + + spin_lock(&vm->status_lock); + list_for_each_entry(bo_base, &vm->vm_bo_list, vm_bo) { + spin_lock(&glob->lru_lock); +ttm_bo_move_to_lru_tail (&bo_base->bo->tbo); + if (bo_base->bo->shadow) +ttm_bo_move_to_lru_tail (&bo_base->bo->shadow->tbo); + spin_unlock(&glob->lru_lock); + } + spin_unlock(&vm->status_lock); +} /*** amdgpu_vm_bo_insert_mapping - insert a new mapping (quoted diff) +0
-0
b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.hdrivers/gpu/drm/amd/amdgpu/amdgpu_vm.h int amdgpu_vm_clear_freed(struct amdgpu_device
Quoted 4 lines
index
1886a561c84e..e01895581489100644drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +0
-0@@ -285,6 +285,7 @@ int amdgpu_vm_clear_freed(struct amdgpu_device*adev,
Quoted 9 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h int amdgpu_vm_clear_freed(struct amdgpu_device *adev,
Quoted 4 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +1
-0struct dma_fence **fence); int amdgpu_vm_handle_moved(struct amdgpu_device *adev, struct amdgpu_vm *vm); +void amdgpu_vm_refresh_lru(struct amdgpu_device *adev, struct
amdgpu_vm*vm);Quoted 3 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +0
-0int amdgpu_vm_bo_update(struct amdgpu_device *adev, struct amdgpu_bo_va *bo_va, bool clear);
_______________________________________________
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): a9d42a8a-b9b5-1d89-f95e-e678829a8260-5C7GfCeVMHo@public.gmane.org
Re: [PATCH 2/3] drm/amdgpu: refresh per vm bo lru
Quoted 1 line · Christian König
How about we update the LRU only when we need to re-validate at least one BO?
I tried this just now, performance still isn't stable, sometime drop to 28fps by accident.
I also tried to check num_evictions, if eviction happens, then update LRU, also sometime drop to 28fps by accident.
When BOs change, we not only need keep LRU order, but also validation order in vm->evicted list. Any other ideas which can keep these order but not increase submission overhead?
Regards,
David Zhou
Quoted 91 lines
BTW: We can easily walk all BOs which belong to a VM, skipping over the few which aren't per VM BOs should be trivial.
Christian.
Am 27.03.2018 um 13:56 schrieb Zhou, David(ChunMing):
Quoted 84 lines
then how to keep unique lru order? any ideas?
To stable performance, we have to keep unique lru order, otherwise like the issue I look into, sometimes F1game is 40fps, sometimes 28fps...even re-validate allowed domains BO.
The left root cause is the moved BOs are not same.
send from Smartisan Pro
Christian K鰊ig <[email protected]> 于 2018年3月27日 下午6:50写道:
NAK, we already tried that and it is really not a good idea because it massively increases the per submission overhead.
Christian.
Am 27.03.2018 um 12:16 schrieb Chunming Zhou:
Review
Quoted 9 lines
Change-Id Ibad84ed585b0746867a5f4cd1eadc2273e7cf596Signed-off-by Chunming Zhou <[email protected]>---
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 2 ++ drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 15 +++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 1 + 3 files changed, 18 insertions(+)(quoted diff) +0
-0
b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.cdrivers/gpu/drm/amd/amdgpu/amdgpu_cs.c static int amdgpu_bo_vm_update_pte(struct
Quoted 4 lines
index
383bf2d31c92..414e61799236100644drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +0
-0@@ -919,6 +919,8 @@ static int amdgpu_bo_vm_update_pte(struct
amdgpu_cs_parser*p)Review
Quoted 9 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +2
-0} } + amdgpu_vm_refresh_lru(adev, vm); + return r; }(quoted diff) +0
-0
b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.cdrivers/gpu/drm/amd/amdgpu/amdgpu_vm.c struct amdgpu_bo_va *amdgpu_vm_bo_add(struct
Quoted 4 lines
index
5e35e23511cf..8ad2bb705765100644drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +0
-0@@ -1902,6 +1902,21 @@ struct amdgpu_bo_va *amdgpu_vm_bo_add(struct
amdgpu_device*adev,drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c struct amdgpu_bo_va *amdgpu_vm_bo_add(struct amdgpu_device *adev,
Quoted 4 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +1
-0return bo_va; } +void amdgpu_vm_refresh_lru(struct amdgpu_device *adev, struct
amdgpu_vm*vm)Review
Quoted 18 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +14
-0+{ + struct ttm_bo_global *glob = adev->mman.bdev.glob; + struct amdgpu_vm_bo_base *bo_base; + + spin_lock(&vm->status_lock); + list_for_each_entry(bo_base, &vm->vm_bo_list, vm_bo) { + spin_lock(&glob->lru_lock); +ttm_bo_move_to_lru_tail (&bo_base->bo->tbo); + if (bo_base->bo->shadow) +ttm_bo_move_to_lru_tail (&bo_base->bo->shadow->tbo); + spin_unlock(&glob->lru_lock); + } + spin_unlock(&vm->status_lock); +} /*** amdgpu_vm_bo_insert_mapping - insert a new mapping (quoted diff) +0
-0
b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.hdrivers/gpu/drm/amd/amdgpu/amdgpu_vm.h int amdgpu_vm_clear_freed(struct amdgpu_device
Quoted 4 lines
index
1886a561c84e..e01895581489100644drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +0
-0@@ -285,6 +285,7 @@ int amdgpu_vm_clear_freed(struct amdgpu_device*adev,
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h int amdgpu_vm_clear_freed(struct amdgpu_device *adev,
Quoted 4 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +1
-0struct dma_fence **fence); int amdgpu_vm_handle_moved(struct amdgpu_device *adev, struct amdgpu_vm *vm); +void amdgpu_vm_refresh_lru(struct amdgpu_device *adev, struct
amdgpu_vm*vm);Quoted 3 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +0
-0int amdgpu_vm_bo_update(struct amdgpu_device *adev, struct amdgpu_bo_va *bo_va, bool clear);
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
Re: [PATCH 2/3] drm/amdgpu: refresh per vm bo lru
Quoted 11 lines · On 2018年03月28日 16:13, zhoucm1 wrote:
Quoted 1 line · Christian König
How about we update the LRU only when we need to re-validate at least one BO?
I tried this just now, performance still isn't stable, sometime drop to 28fps by accident.
I also tried to check
num_evictions, if eviction happens, then update LRU, also sometime drop to 28fps by accident.When BOs change, we not only need keep LRU order, but also validation order in
vm->evictedlist. Any other ideas which can keep these order but not increase submission overhead?
With more thinking, we need to add new LRU design for per vm bo, we need to make sure the order when adding to LRU. How about the below idea:
0. separate traditional bo list lru and per-vm-bo lru. Traditional lru keeps old way, per-vm-lru follows below design.
1. TTM bdev maintains a vm/process list.
2. Every vm_list node contains its own per-vm-bo LRU[priority]
3. To manage the vm_list lru in specific driver, we will need add callback for it.
4. We will add an order for every per-vm-bo in that vm/process.
5. To speed up per-vm-lru sort, we will introduce RB tree for it in callback. The RB tree key is order.
This way, we will be able to keep the per-vm-bo LRU order.
What do you think of it?
Regards,
David Zhou
Quoted 95 lines
Regards,
David ZhouQuoted 91 lines
BTW: We can easily walk all BOs which belong to a VM, skipping over the few which aren't per VM BOs should be trivial.
Christian.
Am 27.03.2018 um 13:56 schrieb Zhou, David(ChunMing):
Quoted 84 lines
then how to keep unique lru order? any ideas?
To stable performance, we have to keep unique lru order, otherwise like the issue I look into, sometimes F1game is 40fps, sometimes 28fps...even re-validate allowed domains BO.
The left root cause is the moved BOs are not same.
send from Smartisan Pro
Christian K鰊ig <[email protected]> 于 2018年3月27日 下午6:50写道:
NAK, we already tried that and it is really not a good idea because it massively increases the per submission overhead.
Christian.
Am 27.03.2018 um 12:16 schrieb Chunming Zhou:
Review
Quoted 9 lines
Change-Id Ibad84ed585b0746867a5f4cd1eadc2273e7cf596Signed-off-by Chunming Zhou <[email protected]>---
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 2 ++ drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 15 +++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 1 + 3 files changed, 18 insertions(+)(quoted diff) +0
-0
b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.cdrivers/gpu/drm/amd/amdgpu/amdgpu_cs.c static int amdgpu_bo_vm_update_pte(struct
Quoted 4 lines
index
383bf2d31c92..414e61799236100644drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +0
-0@@ -919,6 +919,8 @@ static int amdgpu_bo_vm_update_pte(struct
amdgpu_cs_parser*p)Review
Quoted 9 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +2
-0} } + amdgpu_vm_refresh_lru(adev, vm); + return r; }(quoted diff) +0
-0
b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.cdrivers/gpu/drm/amd/amdgpu/amdgpu_vm.c struct amdgpu_bo_va
Quoted 4 lines
index
5e35e23511cf..8ad2bb705765100644drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +0
-0@@ -1902,6 +1902,21 @@ struct amdgpu_bo_va*amdgpu_vm_bo_add(struct
amdgpu_device*adev,drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c struct amdgpu_bo_va · line 1905
Quoted 4 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +1
-0return bo_va; } +void amdgpu_vm_refresh_lru(struct amdgpu_device *adev, struct
amdgpu_vm*vm)Review
Quoted 18 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +14
-0+{ + struct ttm_bo_global *glob = adev->mman.bdev.glob; + struct amdgpu_vm_bo_base *bo_base; + + spin_lock(&vm->status_lock); + list_for_each_entry(bo_base, &vm->vm_bo_list, vm_bo) { + spin_lock(&glob->lru_lock); +ttm_bo_move_to_lru_tail (&bo_base->bo->tbo); + if (bo_base->bo->shadow) +ttm_bo_move_to_lru_tail (&bo_base->bo->shadow->tbo); + spin_unlock(&glob->lru_lock); + } + spin_unlock(&vm->status_lock); +} /*** amdgpu_vm_bo_insert_mapping - insert a new mapping (quoted diff) +0
-0
b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.hdrivers/gpu/drm/amd/amdgpu/amdgpu_vm.h int amdgpu_vm_clear_freed(struct amdgpu_device
Quoted 4 lines
index
1886a561c84e..e01895581489100644drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +0
-0@@ -285,6 +285,7 @@ int amdgpu_vm_clear_freed(struct amdgpu_device*adev,
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h int amdgpu_vm_clear_freed(struct amdgpu_device · line 288
Quoted 4 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +1
-0struct dma_fence **fence); int amdgpu_vm_handle_moved(struct amdgpu_device *adev, struct amdgpu_vm *vm); +void amdgpu_vm_refresh_lru(struct amdgpu_device *adev, struct
amdgpu_vm*vm);Quoted 3 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +0
-0int amdgpu_vm_bo_update(struct amdgpu_device *adev, struct amdgpu_bo_va *bo_va, bool clear);
_______________________________________________
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): 14672700-82d7-c9d8-1086-84a4d8a711bc-5C7GfCeVMHo@public.gmane.org
Re: [PATCH 2/3] drm/amdgpu: refresh per vm bo lru
Am 29.03.2018 um 10:37 schrieb zhoucm1:
Quoted 11 lines · On 2018年03月28日 16:13, zhoucm1 wrote:
Quoted 7 lines · Christian König
Quoted 2 lines
How about we update the LRU only when we need to re-validate at least one BO?
I tried this just now, performance still isn't stable, sometime drop to 28fps by accident.
Can you give me the code for that? I probably can't work this week on that, but I can take a look next week.
Quoted 13 lines
Quoted 4 lines
I also tried to check
num_evictions, if eviction happens, then update LRU, also sometime drop to 28fps by accident.When BOs change, we not only need keep LRU order, but also validation order in
vm->evictedlist. Any other ideas which can keep these order but not increase submission overhead?With more thinking, we need to add new LRU design for per vm bo, we need to make sure the order when adding to LRU. How about the below idea:
0. separate traditional bo list lru and per-vm-bo lru. Traditional lru keeps old way, per-vm-lru follows below design.
1. TTM bdev maintains a vm/process list.
2. Everyvm_listnode contains its own per-vm-bo LRU[priority]
3. To manage thevm_listlru in specific driver, we will need add callback for it.
4. We will add an order for every per-vm-bo in that vm/process.
5. To speed up per-vm-lru sort, we will introduce RB tree for it in callback. The RB tree key is order.
Quoted 3 lines
This way, we will be able to keep the per-vm-bo LRU order.
What do you think of it?
No, we need a single LRU for per VM and not per VM BOs to maintain eviction fairness, so we don't really win anything with that.
Regards,
Christian.
Quoted 99 lines
Regards,
David ZhouQuoted 95 lines
Regards,
David ZhouQuoted 91 lines
BTW: We can easily walk all BOs which belong to a VM, skipping over the few which aren't per VM BOs should be trivial.
Christian.
Am 27.03.2018 um 13:56 schrieb Zhou, David(ChunMing):
Quoted 84 lines
then how to keep unique lru order? any ideas?
To stable performance, we have to keep unique lru order, otherwise like the issue I look into, sometimes F1game is 40fps, sometimes 28fps...even re-validate allowed domains BO.
The left root cause is the moved BOs are not same.
send from Smartisan Pro
Christian K鰊ig <[email protected]> 于 2018年3月27日 下午6:50写道:
NAK, we already tried that and it is really not a good idea because it massively increases the per submission overhead.
Christian.
Am 27.03.2018 um 12:16 schrieb Chunming Zhou:
Review
Quoted 9 lines
Change-Id Ibad84ed585b0746867a5f4cd1eadc2273e7cf596Signed-off-by Chunming Zhou <[email protected]>---
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 2 ++ drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 15 +++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 1 + 3 files changed, 18 insertions(+)(quoted diff) +0
-0
b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.cdrivers/gpu/drm/amd/amdgpu/amdgpu_cs.c static int amdgpu_bo_vm_update_pte(struct
Quoted 4 lines
index
383bf2d31c92..414e61799236100644drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +0
-0@@ -919,6 +919,8 @@ static int amdgpu_bo_vm_update_pte(struct
amdgpu_cs_parser*p)Review
Quoted 9 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +2
-0} } + amdgpu_vm_refresh_lru(adev, vm); + return r; }(quoted diff) +0
-0
b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.cdrivers/gpu/drm/amd/amdgpu/amdgpu_vm.c struct amdgpu_bo_va
Quoted 4 lines
index
5e35e23511cf..8ad2bb705765100644drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +0
-0@@ -1902,6 +1902,21 @@ struct amdgpu_bo_va*amdgpu_vm_bo_add(struct
amdgpu_device*adev,drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c struct amdgpu_bo_va · line 1905
Quoted 4 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +1
-0return bo_va; } +void amdgpu_vm_refresh_lru(struct amdgpu_device *adev, struct
amdgpu_vm*vm)Review
Quoted 18 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +14
-0+{ + struct ttm_bo_global *glob = adev->mman.bdev.glob; + struct amdgpu_vm_bo_base *bo_base; + + spin_lock(&vm->status_lock); + list_for_each_entry(bo_base, &vm->vm_bo_list, vm_bo) { + spin_lock(&glob->lru_lock); +ttm_bo_move_to_lru_tail (&bo_base->bo->tbo); + if (bo_base->bo->shadow) +ttm_bo_move_to_lru_tail (&bo_base->bo->shadow->tbo); +spin_unlock (&glob->lru_lock); + } + spin_unlock(&vm->status_lock); +} /*** amdgpu_vm_bo_insert_mapping - insert a new mapping (quoted diff) +0
-0
b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.hdrivers/gpu/drm/amd/amdgpu/amdgpu_vm.h int amdgpu_vm_clear_freed(struct
Quoted 4 lines
index
1886a561c84e..e01895581489100644drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +0
-0@@ -285,6 +285,7 @@ int amdgpu_vm_clear_freed(struct
amdgpu_device*adev,drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h int amdgpu_vm_clear_freed(struct · line 288
Quoted 4 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +1
-0struct dma_fence **fence); int amdgpu_vm_handle_moved(struct amdgpu_device *adev, struct amdgpu_vm *vm); +void amdgpu_vm_refresh_lru(struct amdgpu_device *adev, struct
amdgpu_vm*vm);Quoted 3 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +0
-0int amdgpu_vm_bo_update(struct amdgpu_device *adev, struct amdgpu_bo_va *bo_va, bool clear);
_______________________________________________
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): d7dad640-edd0-5aa3-9165-b19d44525a5b-5C7GfCeVMHo@public.gmane.org
Re: [PATCH 2/3] drm/amdgpu: refresh per vm bo lru
Quoted 8 lines · Christian König
Am 29.03.2018 um 10:37 schrieb zhoucm1:
Quoted 7 lines · On 2018年03月28日 16:13, zhoucm1 wrote:
Quoted 3 lines
Quoted 6 lines
Quoted 4 lines
Quoted 4 lines · Christian König
Quoted 2 lines
How about we update the LRU only when we need to re-validate at least one BO?
I tried this just now, performance still isn't stable, sometime drop to 28fps by accident.
Can you give me the code for that? I probably can't work this week on that, but I can take a look next week.
just git send-mail to your amd mail.
Quoted 14 lines
Quoted 14 lines
Quoted 5 lines
I also tried to check
num_evictions, if eviction happens, then update LRU, also sometime drop to 28fps by accident.When BOs change, we not only need keep LRU order, but also validation order in
vm->evictedlist. Any other ideas which can keep these order but not increase submission overhead?With more thinking, we need to add new LRU design for per vm bo, we need to make sure the order when adding to LRU. How about the below idea:
0. separate traditional bo list lru and per-vm-bo lru. Traditional lru keeps old way, per-vm-lru follows below design.
1. TTM bdev maintains a vm/process list.
2. Everyvm_listnode contains its own per-vm-bo LRU[priority]
3. To manage thevm_listlru in specific driver, we will need add callback for it.
4. We will add an order for every per-vm-bo in that vm/process.
5. To speed up per-vm-lru sort, we will introduce RB tree for it in callback. The RB tree key is order.
Quoted 5 lines
Quoted 3 lines
This way, we will be able to keep the per-vm-bo LRU order.
What do you think of it?
No, we need a single LRU for per VM and not per VM BOs to maintain eviction fairness, so we don't really win anything with that.
If following original LRU design, the bo should be moved to lru tail when bo is used, so that keep the last used bo is in lru tail.
All per vm BOs are used for every command submission, then after every CS, we should refresh the lru, that is required by original LRU design, but as your NAK on it, it will introduce much CPU overhead for CS, they are inconsistencies.
For per vm case, if we don't want to introduce extra overhead, the per-vm-bo order shoud fixed in lru to avoid refresh LRU for every CS. so my thinking for lru is:
VM1-BO1---->BO2--->BO3--->BOn--->VM2-BO1--->BO2--->BO3--->BOn--->VM3-BO...
Regards,
David
Quoted 104 lines
Regards,
Christian.Quoted 99 lines
Regards,
David ZhouQuoted 95 lines
Regards,
David ZhouQuoted 91 lines
BTW: We can easily walk all BOs which belong to a VM, skipping over the few which aren't per VM BOs should be trivial.
Christian.
Am 27.03.2018 um 13:56 schrieb Zhou, David(ChunMing):
Quoted 84 lines
then how to keep unique lru order? any ideas?
To stable performance, we have to keep unique lru order, otherwise like the issue I look into, sometimes F1game is 40fps, sometimes 28fps...even re-validate allowed domains BO.
The left root cause is the moved BOs are not same.
send from Smartisan Pro
Christian K鰊ig <[email protected]> 于 2018年3月27日 下午6:50写道:
NAK, we already tried that and it is really not a good idea because it massively increases the per submission overhead.
Christian.
Am 27.03.2018 um 12:16 schrieb Chunming Zhou:
Review
Quoted 9 lines
Change-Id Ibad84ed585b0746867a5f4cd1eadc2273e7cf596Signed-off-by Chunming Zhou <[email protected]>---
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 2 ++ drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 15 +++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 1 + 3 files changed, 18 insertions(+)(quoted diff) +0
-0
b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.cdrivers/gpu/drm/amd/amdgpu/amdgpu_cs.c static int amdgpu_bo_vm_update_pte(struct
Quoted 4 lines
index
383bf2d31c92..414e61799236100644drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +0
-0@@ -919,6 +919,8 @@ static int amdgpu_bo_vm_update_pte(struct
amdgpu_cs_parser*p)Review
Quoted 9 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +2
-0} } + amdgpu_vm_refresh_lru(adev, vm); + return r; }(quoted diff) +0
-0
b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.cdrivers/gpu/drm/amd/amdgpu/amdgpu_vm.c struct amdgpu_bo_va
Quoted 4 lines
index
5e35e23511cf..8ad2bb705765100644drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +0
-0@@ -1902,6 +1902,21 @@ struct amdgpu_bo_va*amdgpu_vm_bo_add(struct
amdgpu_device*adev,drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c struct amdgpu_bo_va · line 1905
Quoted 4 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +1
-0return bo_va; } +void amdgpu_vm_refresh_lru(struct amdgpu_device *adev, struct
amdgpu_vm*vm)Review
Quoted 18 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +14
-0+{ + struct ttm_bo_global *glob = adev->mman.bdev.glob; + struct amdgpu_vm_bo_base *bo_base; + + spin_lock(&vm->status_lock); + list_for_each_entry(bo_base, &vm->vm_bo_list, vm_bo) { +spin_lock (&glob->lru_lock); +ttm_bo_move_to_lru_tail (&bo_base->bo->tbo); + if (bo_base->bo->shadow) +ttm_bo_move_to_lru_tail (&bo_base->bo->shadow->tbo); +spin_unlock (&glob->lru_lock); + } + spin_unlock(&vm->status_lock); +} /*** amdgpu_vm_bo_insert_mapping - insert a new mapping (quoted diff) +0
-0
b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.hdrivers/gpu/drm/amd/amdgpu/amdgpu_vm.h int amdgpu_vm_clear_freed(struct
Quoted 4 lines
index
1886a561c84e..e01895581489100644drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +0
-0@@ -285,6 +285,7 @@ int amdgpu_vm_clear_freed(struct
amdgpu_device*adev,drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h int amdgpu_vm_clear_freed(struct · line 288
Quoted 4 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +1
-0struct dma_fence **fence); int amdgpu_vm_handle_moved(struct amdgpu_device *adev, struct amdgpu_vm *vm); +void amdgpu_vm_refresh_lru(struct amdgpu_device *adev, struct
amdgpu_vm*vm);Quoted 3 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +0
-0int amdgpu_vm_bo_update(struct amdgpu_device *adev, struct amdgpu_bo_va *bo_va, bool clear);
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
PATCH 3/3 drm/amdgpu: re-validate per VM BOs if required v2
In reply to a message not in the archiveFrom: Christian König <[email protected]>
If a per VM BO ends up in a allowed domain it never moves back into the prefered domain.
v2: move to refresh function, the validation order also keeps same as lru
--- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 7 +++++++ 1 file changed, 7 insertions(+)
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +7 -0
@@ -1914,6 +1914,13 @@ void amdgpu_vm_refresh_lru(struct amdgpu_device *adev, struct amdgpu_vm *vm)if (bo_base->bo->shadow)ttm_bo_move_to_lru_tail (&bo_base->bo->shadow->tbo);spin_unlock (&glob->lru_lock); +/* If the BO prefers to be in VRAM, but currently isn't add it +* back to the evicted list so that it gets validated again on +* the next command submission. +*/ +if (bo_base->bo->preferred_domains ==AMDGPU_GEM_DOMAIN_VRAM && + bo_base->bo->tbo.mem.mem_type !=TTM_PL_VRAM ) +list_move_tail (&bo_base->vm_status, &vm->evicted); }spin_unlock (&vm->status_lock); }
Signature · 6 lines
2.14.1 _______________________________________________ amd-gfx mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/amd-gfx
Christian König
NAKNAK, that increases the memory footprint way to much and there is no description on what that should be good for.
Christian.
Am 27.03.2018 um 12:16 schrieb Chunming Zhou:
Quoted 63 lines
the list decides the lru order.
Change-Id I8baf85aefd5781501599ff672949a9b71099a30eSigned-off-by Chunming Zhou <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 7 +++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 6 ++++++ 2 files changed, 13 insertions(+)drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +7
-0@@ -1866,6 +1866,7 @@ struct amdgpu_bo_va *amdgpu_vm_bo_add(struct amdgpu_device *adev, } bo_va->base.vm = vm; bo_va->base.bo = bo; +INIT_LIST_HEAD (&bo_va->base.vm_bo);INIT_LIST_HEAD (&bo_va->base.bo_list);INIT_LIST_HEAD (&bo_va->base.vm_status);@@ -1881,6 +1882,10 @@ struct amdgpu_bo_va *amdgpu_vm_bo_add(struct amdgpu_device *adev,if (bo->tbo.resv != vm->root.base.bo->tbo.resv)return bo_va; +spin_lock (&vm->status_lock); +list_add_tail (&bo_va->base.vm_bo, &vm->vm_bo_list); +spin_unlock (&vm->status_lock); +if (bo->preferred_domains &amdgpu_mem_type_to_domain (bo->tbo.mem.mem_type))return bo_va;@@ -2237,6 +2242,7 @@ void amdgpu_vm_bo_rmv(struct amdgpu_device *adev,spin_lock (&vm->status_lock);list_del (&bo_va->base.vm_status); +list_del (&bo_va->base.vm_bo);spin_unlock (&vm->status_lock);list_for_each_entry_safe (mapping, next, &bo_va->valids, list) {@@ -2409,6 +2415,7 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm,for (i =0 ; i <AMDGPU_MAX_VMHUBS ; i++) vm->reserved_vmid[i] =NULL ;spin_lock_init (&vm->status_lock); +INIT_LIST_HEAD (&vm->vm_bo_list);INIT_LIST_HEAD (&vm->evicted);INIT_LIST_HEAD (&vm->relocated);INIT_LIST_HEAD (&vm->moved);drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +3
-0@@ -144,6 +144,9 @@ struct amdgpu_vm_bo_base {struct amdgpu_vm *vm;struct amdgpu_bo *bo; +/* protected by vm status lock */ +struct list_head vm_bo; +/* protected by bo being reserved */ struct list_head bo_list;
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h struct amdgpu_vm { · line 188
Quoted 9 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +3
-0@@ -177,6 +180,9 @@ struct amdgpu_vm {/* protecting invalidated */ spinlock_t status_lock; +/* protected by status lock */ +struct list_head vm_bo_list; +/* BOs who needs a validation */ struct list_head evicted;
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx