fix gmc page fault on navi1X
Subsystems (2)
v3 cover fix gmc page fault on navi1X
We hit gmc page fault on navi1X.
UMR tells that the physical address of pte is bad.
Two issues:
1) we did not sync job schedule fence while update mapping. fix it by adding bo fence to resv after every job submit.
2) we might unref page table bo during update ptes, at the same time, there is job pending on bo. and submit a job in commit after free bo. We need free the bo after adding all fence to bo.
change from v2:
use the correct page table bo resv
change from v1:
fix rebase issue.
xinhui pan (2):
drm/amdgpu: Add job fence to resv conditionally
drm/amdgpu: unref the bo after job submit
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 42 +++++++++++++++------ drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 5 +++ drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c | 7 ++++ 3 files changed, 42 insertions(+), 12 deletions(-)
Signature · 6 lines
2.17.1 _______________________________________________ amd-gfx mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/amd-gfx
PATCH v3 1/2 drm/amdgpu: Add job fence to resv conditionally
Provide resv as a parameter for vm sdma commit.
Job fence on page table should be a shared one, so add it to the resv.
--- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 4 ++-- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 5 +++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c | 7 +++++++ 3 files changed, 14 insertions(+), 2 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +2 -2
@@ -1579,6 +1579,7 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,dma_addr_t *pages_addr, struct dma_fence **fence) { + struct amdgpu_bo *root = vm->root.base.bo;struct amdgpu_vm_update_params params;enum amdgpu_sync_mode sync_mode;int r;
@@ -1604,8 +1605,6 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev, }if (flags &AMDGPU_PTE_VALID ) { -struct amdgpu_bo *root = vm->root.base.bo; -if (!dma_fence_is_signaled (vm->last_direct))amdgpu_bo_fence (root, vm->last_direct, true);
@@ -1613,6 +1612,7 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,amdgpu_bo_fence (root, vm->last_delayed, true); } + params.resv = root->tbo.base.resv; r = vm->update_funcs->prepare (¶ms, resv, sync_mode);if (r) goto error_unlock;
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +5 -0
@@ -226,6 +226,11 @@ struct amdgpu_vm_update_params {* @num_dw_left: number of dw left for the IB */ unsigned int num_dw_left; + +/** +* @resv: sync the resv and add job fence to it conditionally. +*/ +struct dma_resv *resv; };struct amdgpu_vm_update_funcs {
drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c +7 -0
@@ -111,6 +111,13 @@ static int amdgpu_vm_sdma_commit(struct amdgpu_vm_update_params *p,swap (p->vm->last_delayed, tmp);dma_fence_put (tmp); +/* add job fence to resv. +* MM notifier path is an exception as we can not grab the +* resv lock. +*/ +if (!p->direct && p->resv) +dma_resv_add_shared_fence (p->resv, f); +if (fence && !p->direct)swap (*fence, f);dma_fence_put (f);
Signature · 6 lines
2.17.1 _______________________________________________ amd-gfx mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/amd-gfx
Christian König
maintainerRe: [PATCH v3 1/2] drm/amdgpu: Add job fence to resv conditionally
Am 13.03.20 um 15:07 schrieb xinhui pan:
Quoted 64 lines
Provide resv as a parameter for vm sdma commit.
Job fence on page table should be a shared one, so add it to the resv.Cc Christian König <[email protected]>Cc Alex Deucher <[email protected]>Cc Felix Kuehling <[email protected]>Suggested-by Christian König <[email protected]>Signed-off-by xinhui pan <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 4 ++-- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 5 +++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c | 7 +++++++ 3 files changed, 14 insertions(+), 2 deletions(-)drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +2
-2@@ -1579,6 +1579,7 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,dma_addr_t *pages_addr, struct dma_fence **fence) { + struct amdgpu_bo *root = vm->root.base.bo;struct amdgpu_vm_update_params params;enum amdgpu_sync_mode sync_mode;int r;@@ -1604,8 +1605,6 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev, }if (flags &AMDGPU_PTE_VALID ) { -struct amdgpu_bo *root = vm->root.base.bo; -if (!dma_fence_is_signaled (vm->last_direct))amdgpu_bo_fence (root, vm->last_direct, true);@@ -1613,6 +1612,7 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,amdgpu_bo_fence (root, vm->last_delayed, true); } + params.resv = root->tbo.base.resv; r = vm->update_funcs->prepare (¶ms, resv, sync_mode);if (r) goto error_unlock;drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +5
-0@@ -226,6 +226,11 @@ struct amdgpu_vm_update_params {* @num_dw_left: number of dw left for the IB */ unsigned int num_dw_left; + +/** +* @resv: sync the resv and add job fence to it conditionally. +*/ +struct dma_resv *resv; };struct amdgpu_vm_update_funcs {drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c +0
-0
drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c static int amdgpu_vm_sdma_commit(struct amdgpu_vm_update_params *p, · line 118
Quoted 8 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c +5
-0@@ -111,6 +111,13 @@ static int amdgpu_vm_sdma_commit(struct amdgpu_vm_update_params *p,swap (p->vm->last_delayed, tmp);dma_fence_put (tmp); +/* add job fence to resv. +* MM notifier path is an exception as we can not grab the +* resv lock. +*/ +if (!p->direct && p->resv)
You can just use p->vm->root.base.bo->tbo.base.resv here, no need for a new field in the paramater structure.
And it would probably be best to also remove the vm->last_delayed field entirely.
In other words use something like this here
if (p->direct) {
tmp = dma_fence_get(f);
swap(p->vm->last_direct, tmp);
dma_fence_put(tmp);
} else { dma_resv_add_shared_fence(p->vm->root.base.bo->tbo.base.resv, f); }
Regards,
Christian.
drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c static int amdgpu_vm_sdma_commit(struct amdgpu_vm_update_params *p, · line 123
Quoted 5 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c +2
-0+dma_resv_add_shared_fence (p->resv, f); +if (fence && !p->direct)swap (*fence, f);dma_fence_put (f);
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
Re: [PATCH v3 1/2] drm/amdgpu: Add job fence to resv conditionally
Quoted 66 lines
2020年3月13日 22:13,Koenig, Christian <[email protected]> 写道:
Am 13.03.20 um 15:07 schrieb xinhui pan:
Quoted 61 lines
Provide resv as a parameter for vm sdma commit.
Job fence on page table should be a shared one, so add it to the resv.Cc Christian König <[email protected]>Cc Alex Deucher <[email protected]>Cc Felix Kuehling <[email protected]>Suggested-by Christian König <[email protected]>Signed-off-by xinhui pan <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 4 ++-- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 5 +++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c | 7 +++++++ 3 files changed, 14 insertions(+), 2 deletions(-)drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +1
-2@@ -1579,6 +1579,7 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,dma_addr_t *pages_addr, struct dma_fence **fence) { + struct amdgpu_bo *root = vm->root.base.bo;struct amdgpu_vm_update_params params;enum amdgpu_sync_mode sync_mode;int r;@@ -1604,8 +1605,6 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev, }if (flags &AMDGPU_PTE_VALID ) { -struct amdgpu_bo *root = vm->root.base.bo; -if (!dma_fence_is_signaled (vm->last_direct))amdgpu_bo_fence (root, vm->last_direct, true); @@ -1613,6 +1612,7 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev, amdgpu_bo_fence(root, vm->last_delayed, true);drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +0
-0} + params.resv = root->tbo.base.resv; r = vm->update_funcs->prepare (¶ms, resv, sync_mode);if (r) goto error_unlock;drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +5
-0@@ -226,6 +226,11 @@ struct amdgpu_vm_update_params {* @num_dw_left: number of dw left for the IB */ unsigned int num_dw_left; + +/** +* @resv: sync the resv and add job fence to it conditionally. +*/ +struct dma_resv *resv; };struct amdgpu_vm_update_funcs {drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c +0
-0@@ -111,6 +111,13 @@ static int amdgpu_vm_sdma_commit(struct amdgpu_vm_update_params *p,swap (p->vm->last_delayed, tmp);
Quoted 9 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c static int amdgpu_vm_sdma_commit(struct amdgpu_vm_update_params *p, · line 118
Quoted 6 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c +4
-0dma_fence_put (tmp); +/* add job fence to resv. +* MM notifier path is an exception as we can not grab the +* resv lock. +*/ + if (!p->direct && p->resv)You can just use
p->vm->root.base.bo->tbo.base.resvhere, no need for a new field in the paramater structure.
yep, make sense.
Quoted 4 lines
And it would probably be best to also remove the
vm->last_delayedfield entirely.In other words use something like this here
Quoted 8 lines
if (p->direct) { tmp = dma_fence_get(f); swap(p->vm->last_direct, tmp); dma_fence_put(tmp); } else {
dma_resv_add_shared_fence(p->vm->root.base.bo->tbo.base.resv, f); }
I think we still need udpate the last_delayed. looks like eviction and some other ioctls test this field to determine the vm state. this should be done by a new patch if possible.
Quoted 9 lines
Regards,
Christian.drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c static int amdgpu_vm_sdma_commit(struct amdgpu_vm_update_params *p, · line 123
Quoted 5 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c +2
-0+dma_resv_add_shared_fence (p->resv, f); +if (fence && !p->direct)swap (*fence, f);dma_fence_put (f);
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
PATCH v3 2/2 drm/amdgpu: unref the bo after job submit
In reply to xinhui panOtherwise we might free BOs before job has completed. We add the fence to BO after commit, so free BOs after that.
--- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 38 +++++++++++++++++++------- 1 file changed, 28 insertions(+), 10 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +28 -10
@@ -942,13 +942,17 @@ static int amdgpu_vm_alloc_pts(struct amdgpu_device *adev,* * @entry: PDE to free */ -static void amdgpu_vm_free_table (struct amdgpu_vm_pt *entry) +static void amdgpu_vm_free_table (struct amdgpu_vm_pt *entry, +struct list_head *head) {if (entry->base.bo) { entry->base.bo->vm_bo =NULL ;list_del (&entry->base.vm_status); -amdgpu_bo_unref (&entry->base.bo->shadow); -amdgpu_bo_unref (&entry->base.bo); +if (!head) { +amdgpu_bo_unref (&entry->base.bo->shadow); +amdgpu_bo_unref (&entry->base.bo); + }else +list_add (&entry->base.vm_status, head); }kvfree (entry->entries); entry->entries =NULL ;
@@ -965,7 +969,8 @@ static void amdgpu_vm_free_table(struct amdgpu_vm_pt *entry)*/ static void amdgpu_vm_free_pts (struct amdgpu_device *adev,struct amdgpu_vm *vm, -struct amdgpu_vm_pt_cursor *start) +struct amdgpu_vm_pt_cursor *start, +struct list_head *head) {struct amdgpu_vm_pt_cursor cursor;struct amdgpu_vm_pt *entry;
@@ -973,10 +978,10 @@ static void amdgpu_vm_free_pts(struct amdgpu_device *adev, vm->bulk_moveable = false;for_each_amdgpu_vm_pt_dfs_safe (adev, vm, start, cursor, entry) -amdgpu_vm_free_table (entry); +amdgpu_vm_free_table (entry, head);if (start) -amdgpu_vm_free_table (start->entry); +amdgpu_vm_free_table (start->entry, head); }/**
@@ -1428,7 +1433,8 @@ static void amdgpu_vm_fragment(struct amdgpu_vm_update_params *params,*/ static int amdgpu_vm_update_ptes (struct amdgpu_vm_update_params *params,uint64_t start,uint64_t end, -uint64_t dst,uint64_t flags) +uint64_t dst,uint64_t flags, +struct list_head *head) {struct amdgpu_device *adev = params->adev;struct amdgpu_vm_pt_cursor cursor;
@@ -1539,7 +1545,7 @@ static int amdgpu_vm_update_ptes(struct amdgpu_vm_update_params *params,* completely covered by the range and so potentially still in use. */ while (cursor.pfn < frag_start) { -amdgpu_vm_free_pts (adev, params->vm, &cursor); +amdgpu_vm_free_pts (adev, params->vm, &cursor, head);amdgpu_vm_pt_next (adev, &cursor); }
@@ -1583,6 +1589,7 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,struct amdgpu_vm_update_params params;enum amdgpu_sync_mode sync_mode;int r; +struct list_head head;memset (¶ms,0 ,sizeof (params)); params.adev = adev;
@@ -1590,6 +1597,8 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev, params.direct = direct; params.pages_addr = pages_addr; +INIT_LIST_HEAD (&head); +/* Implicitly sync to command submissions in the same VM before * unmapping. Sync to moving fences before mapping. */
@@ -1617,13 +1626,22 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,if (r) goto error_unlock; - r =amdgpu_vm_update_ptes (¶ms, start, last +1 , addr, flags); + r =amdgpu_vm_update_ptes (¶ms, start, last +1 , addr, flags, &head);if (r) goto error_unlock; r = vm->update_funcs->commit (¶ms, fence); error_unlock: +while (!list_empty (&head)) { +struct amdgpu_vm_pt *entry =list_first_entry (&head, + struct amdgpu_vm_pt, base.vm_status); +list_del (&entry->base.vm_status); + +amdgpu_bo_unref (&entry->base.bo->shadow); +amdgpu_bo_unref (&entry->base.bo); + } +amdgpu_vm_eviction_unlock (vm);return r; }
@@ -3118,7 +3136,7 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm)amdgpu_vm_free_mapping (adev, vm, mapping,NULL ); } -amdgpu_vm_free_pts (adev, vm,NULL ); +amdgpu_vm_free_pts (adev, vm,NULL ,NULL );amdgpu_bo_unreserve (root);amdgpu_bo_unref (&root);WARN_ON (vm->root.base.bo);
Signature · 6 lines
2.17.1 _______________________________________________ amd-gfx mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/amd-gfx
Christian König
maintainerRe: [PATCH v3 2/2] drm/amdgpu: unref the bo after job submit
Am 13.03.20 um 15:07 schrieb xinhui pan:
Quoted 2 lines
Otherwise we might free BOs before job has completed. We add the fence to BO after commit, so free BOs after that.
I'm not sure if this is necessary, but probably better save than sorry.
Some comments below.
Quoted 21 lines
Cc Christian König <[email protected]>Cc Alex Deucher <[email protected]>Cc Felix Kuehling <[email protected]>Signed-off-by xinhui pan <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 38 +++++++++++++++++++------- 1 file changed, 28 insertions(+), 10 deletions(-)drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +2
-1@@ -942,13 +942,17 @@ static int amdgpu_vm_alloc_pts(struct amdgpu_device *adev,* * @entry: PDE to free */ -static void amdgpu_vm_free_table (struct amdgpu_vm_pt *entry) +static void amdgpu_vm_free_table (struct amdgpu_vm_pt *entry, +struct list_head *head) {
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c static int amdgpu_vm_alloc_pts(struct amdgpu_device *adev, · line 955
Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +5
-2if (entry->base.bo) { entry->base.bo->vm_bo =NULL ;list_del (&entry->base.vm_status); -amdgpu_bo_unref (&entry->base.bo->shadow); -amdgpu_bo_unref (&entry->base.bo); +if (!head) { +amdgpu_bo_unref (&entry->base.bo->shadow); +amdgpu_bo_unref (&entry->base.bo); + }else +list_add (&entry->base.vm_status, head);
Instead of adding a parameter make this a new state in the VM. Something like vm->zombies or something similar.
Quoted 72 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +11
-6}kvfree (entry->entries); entry->entries =NULL ;@@ -965,7 +969,8 @@ static void amdgpu_vm_free_table(struct amdgpu_vm_pt *entry)*/ static void amdgpu_vm_free_pts (struct amdgpu_device *adev,struct amdgpu_vm *vm, -struct amdgpu_vm_pt_cursor *start) +struct amdgpu_vm_pt_cursor *start, +struct list_head *head) {struct amdgpu_vm_pt_cursor cursor;struct amdgpu_vm_pt *entry;@@ -973,10 +978,10 @@ static void amdgpu_vm_free_pts(struct amdgpu_device *adev, vm->bulk_moveable = false;for_each_amdgpu_vm_pt_dfs_safe (adev, vm, start, cursor, entry) -amdgpu_vm_free_table (entry); +amdgpu_vm_free_table (entry, head);if (start) -amdgpu_vm_free_table (start->entry); +amdgpu_vm_free_table (start->entry, head); }/** @@ -1428,7 +1433,8 @@ static void amdgpu_vm_fragment(struct amdgpu_vm_update_params *params,*/ static int amdgpu_vm_update_ptes (struct amdgpu_vm_update_params *params,uint64_t start,uint64_t end, -uint64_t dst,uint64_t flags) +uint64_t dst,uint64_t flags, +struct list_head *head) {struct amdgpu_device *adev = params->adev;struct amdgpu_vm_pt_cursor cursor;@@ -1539,7 +1545,7 @@ static int amdgpu_vm_update_ptes(struct amdgpu_vm_update_params *params,* completely covered by the range and so potentially still in use. */ while (cursor.pfn < frag_start) { -amdgpu_vm_free_pts (adev, params->vm, &cursor); +amdgpu_vm_free_pts (adev, params->vm, &cursor, head);amdgpu_vm_pt_next (adev, &cursor); }@@ -1583,6 +1589,7 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,struct amdgpu_vm_update_params params;enum amdgpu_sync_mode sync_mode;int r; +struct list_head head;memset (¶ms,0 ,sizeof (params)); params.adev = adev;@@ -1590,6 +1597,8 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev, params.direct = direct; params.pages_addr = pages_addr; +INIT_LIST_HEAD (&head); +/* Implicitly sync to command submissions in the same VM before * unmapping. Sync to moving fences before mapping. */ @@ -1617,13 +1626,22 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,if (r) goto error_unlock; - r =amdgpu_vm_update_ptes (¶ms, start, last +1 , addr, flags); + r =amdgpu_vm_update_ptes (¶ms, start, last +1 , addr, flags, &head);if (r) goto error_unlock; r = vm->update_funcs->commit (¶ms, fence);
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev, · line 1643
Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +8
-0error_unlock: +while (!list_empty (&head)) { +struct amdgpu_vm_pt *entry =list_first_entry (&head, + struct amdgpu_vm_pt, base.vm_status); +list_del (&entry->base.vm_status); + +amdgpu_bo_unref (&entry->base.bo->shadow); +amdgpu_bo_unref (&entry->base.bo); + }
Maybe put that into a separate function.
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm) · line 3139
Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +2
-1+amdgpu_vm_eviction_unlock (vm);return r; }@@ -3118,7 +3136,7 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm)amdgpu_vm_free_mapping (adev, vm, mapping,NULL ); } -amdgpu_vm_free_pts (adev, vm,NULL ); +amdgpu_vm_free_pts (adev, vm,NULL ,NULL );
And then also call it here.
Regards,
Christian.
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm) · line 3142
Quoted 3 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +0
-0amdgpu_bo_unreserve (root);amdgpu_bo_unref (&root);WARN_ON (vm->root.base.bo);
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
Re: [PATCH v3 2/2] drm/amdgpu: unref the bo after job submit
Quoted 9 lines
2020年3月13日 22:20,Koenig, Christian <[email protected]> 写道:
Am 13.03.20 um 15:07 schrieb xinhui pan:
Quoted 2 lines
Otherwise we might free BOs before job has completed. We add the fence to BO after commit, so free BOs after that.
I'm not sure if this is necessary, but probably better save than sorry.
without this patch, we hit gmc page fault.
we have individualized bo resv during bo releasing.
so any fences added to root PT bo resv after that is actually untested.
Quoted 124 lines
Some comments below.
Quoted 21 lines
Cc Christian König <[email protected]>Cc Alex Deucher <[email protected]>Cc Felix Kuehling <[email protected]>Signed-off-by xinhui pan <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 38 +++++++++++++++++++------- 1 file changed, 28 insertions(+), 10 deletions(-)drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +2
-1@@ -942,13 +942,17 @@ static int amdgpu_vm_alloc_pts(struct amdgpu_device *adev,* * @entry: PDE to free */ -static void amdgpu_vm_free_table (struct amdgpu_vm_pt *entry) +static void amdgpu_vm_free_table (struct amdgpu_vm_pt *entry, +struct list_head *head) {drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c static int amdgpu_vm_alloc_pts(struct amdgpu_device *adev, · line 955
Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +5
-2if (entry->base.bo) { entry->base.bo->vm_bo =NULL ;list_del (&entry->base.vm_status); -amdgpu_bo_unref (&entry->base.bo->shadow); -amdgpu_bo_unref (&entry->base.bo); +if (!head) { +amdgpu_bo_unref (&entry->base.bo->shadow); +amdgpu_bo_unref (&entry->base.bo); + }else +list_add (&entry->base.vm_status, head);Instead of adding a parameter make this a new state in the VM. Something like
vm->zombiesor something similar.Quoted 63 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +7
-5}kvfree (entry->entries); entry->entries =NULL ;@@ -965,7 +969,8 @@ static void amdgpu_vm_free_table(struct amdgpu_vm_pt *entry)*/ static void amdgpu_vm_free_pts (struct amdgpu_device *adev,struct amdgpu_vm *vm, -struct amdgpu_vm_pt_cursor *start) +struct amdgpu_vm_pt_cursor *start, +struct list_head *head) {struct amdgpu_vm_pt_cursor cursor;struct amdgpu_vm_pt *entry;@@ -973,10 +978,10 @@ static void amdgpu_vm_free_pts(struct amdgpu_device *adev, vm->bulk_moveable = false;for_each_amdgpu_vm_pt_dfs_safe (adev, vm, start, cursor, entry) -amdgpu_vm_free_table (entry); +amdgpu_vm_free_table (entry, head);if (start) -amdgpu_vm_free_table (start->entry); +amdgpu_vm_free_table (start->entry, head); }/** @@ -1428,7 +1433,8 @@ static void amdgpu_vm_fragment(struct amdgpu_vm_update_params *params,*/ static int amdgpu_vm_update_ptes (struct amdgpu_vm_update_params *params,uint64_t start,uint64_t end, -uint64_t dst,uint64_t flags) +uint64_t dst,uint64_t flags, +struct list_head *head) {struct amdgpu_device *adev = params->adev;struct amdgpu_vm_pt_cursor cursor;@@ -1539,7 +1545,7 @@ static int amdgpu_vm_update_ptes(struct amdgpu_vm_update_params *params,* completely covered by the range and so potentially still in use. */ while (cursor.pfn < frag_start) { -amdgpu_vm_free_pts (adev, params->vm, &cursor); +amdgpu_vm_free_pts (adev, params->vm, &cursor, head);amdgpu_vm_pt_next (adev, &cursor); } @@ -1583,6 +1589,7 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +3
-0struct amdgpu_vm_update_params params;enum amdgpu_sync_mode sync_mode;int r; +struct list_head head;memset (¶ms,0 ,sizeof (params)); params.adev = adev;@@ -1590,6 +1597,8 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev, params.direct = direct; params.pages_addr = pages_addr; +INIT_LIST_HEAD (&head); +/* Implicitly sync to command submissions in the same VM before * unmapping. Sync to moving fences before mapping. */ @@ -1617,13 +1626,22 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,if (r) goto error_unlock; - r =amdgpu_vm_update_ptes (¶ms, start, last +1 , addr, flags); + r =amdgpu_vm_update_ptes (¶ms, start, last +1 , addr, flags, &head);if (r) goto error_unlock;drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev, · line 1641
Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +8
-0r = vm->update_funcs->commit (¶ms, fence); error_unlock: +while (!list_empty (&head)) { +struct amdgpu_vm_pt *entry =list_first_entry (&head, + struct amdgpu_vm_pt, base.vm_status); +list_del (&entry->base.vm_status); + +amdgpu_bo_unref (&entry->base.bo->shadow); +amdgpu_bo_unref (&entry->base.bo); + }Maybe put that into a separate function.
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm) · line 3139
Quoted 9 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +2
-0+amdgpu_vm_eviction_unlock (vm);return r; }@@ -3118,7 +3136,7 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm)amdgpu_vm_free_mapping (adev, vm, mapping,NULL ); } -amdgpu_vm_free_pts (adev, vm,NULL ); +amdgpu_vm_free_pts (adev, vm,NULL ,NULL );And then also call it here.
Quoted 7 lines
Regards,
Christian.Quoted 3 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +0
-0amdgpu_bo_unreserve (root);amdgpu_bo_unref (&root);WARN_ON (vm->root.base.bo);
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
Christian König
maintainerRe: [PATCH v3 2/2] drm/amdgpu: unref the bo after job submit
Am 13.03.20 um 15:54 schrieb Pan, Xinhui:
Quoted 8 lines
Quoted 7 lines
2020年3月13日 22:20,Koenig, Christian <[email protected]> 写道:
Am 13.03.20 um 15:07 schrieb xinhui pan:
Quoted 2 lines
Otherwise we might free BOs before job has completed. We add the fence to BO after commit, so free BOs after that.
I'm not sure if this is necessary, but probably better save than sorry.
without this patch, we hit gmc page fault.
Then this is not the root cause.
See freeing page tables here is save since they can only be reused after all jobs completed.
This only fixed a very small window where the PDE would still point to the freed PT which might already be reused.
But this can't cause a GMC page fault, it would rather be a security problem.
Regards,
Christian.
Quoted 130 lines
we have individualized bo resv during bo releasing.
so any fences added to root PT bo resv after that is actually untested.Quoted 126 lines
Some comments below.
Quoted 20 lines
Cc Christian König <[email protected]>Cc Alex Deucher <[email protected]>Cc Felix Kuehling <[email protected]>Signed-off-by xinhui pan <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 38 +++++++++++++++++++------- 1 file changed, 28 insertions(+), 10 deletions(-)drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +2
-1@@ -942,13 +942,17 @@ static int amdgpu_vm_alloc_pts(struct amdgpu_device *adev,* * @entry: PDE to free */ -static void amdgpu_vm_free_table (struct amdgpu_vm_pt *entry) +static void amdgpu_vm_free_table (struct amdgpu_vm_pt *entry, +struct list_head *head) {drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c static int amdgpu_vm_alloc_pts(struct amdgpu_device *adev, · line 955
Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +5
-2if (entry->base.bo) { entry->base.bo->vm_bo =NULL ;list_del (&entry->base.vm_status); -amdgpu_bo_unref (&entry->base.bo->shadow); -amdgpu_bo_unref (&entry->base.bo); +if (!head) { +amdgpu_bo_unref (&entry->base.bo->shadow); +amdgpu_bo_unref (&entry->base.bo); + }else +list_add (&entry->base.vm_status, head);Instead of adding a parameter make this a new state in the VM. Something like
vm->zombiesor something similar.Quoted 63 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +7
-5}kvfree (entry->entries); entry->entries =NULL ;@@ -965,7 +969,8 @@ static void amdgpu_vm_free_table(struct amdgpu_vm_pt *entry)*/ static void amdgpu_vm_free_pts (struct amdgpu_device *adev,struct amdgpu_vm *vm, -struct amdgpu_vm_pt_cursor *start) +struct amdgpu_vm_pt_cursor *start, +struct list_head *head) {struct amdgpu_vm_pt_cursor cursor;struct amdgpu_vm_pt *entry;@@ -973,10 +978,10 @@ static void amdgpu_vm_free_pts(struct amdgpu_device *adev, vm->bulk_moveable = false;for_each_amdgpu_vm_pt_dfs_safe (adev, vm, start, cursor, entry) -amdgpu_vm_free_table (entry); +amdgpu_vm_free_table (entry, head);if (start) -amdgpu_vm_free_table (start->entry); +amdgpu_vm_free_table (start->entry, head); }/** @@ -1428,7 +1433,8 @@ static void amdgpu_vm_fragment(struct amdgpu_vm_update_params *params,*/ static int amdgpu_vm_update_ptes (struct amdgpu_vm_update_params *params,uint64_t start,uint64_t end, -uint64_t dst,uint64_t flags) +uint64_t dst,uint64_t flags, +struct list_head *head) {struct amdgpu_device *adev = params->adev;struct amdgpu_vm_pt_cursor cursor;@@ -1539,7 +1545,7 @@ static int amdgpu_vm_update_ptes(struct amdgpu_vm_update_params *params,* completely covered by the range and so potentially still in use. */ while (cursor.pfn < frag_start) { -amdgpu_vm_free_pts (adev, params->vm, &cursor); +amdgpu_vm_free_pts (adev, params->vm, &cursor, head);amdgpu_vm_pt_next (adev, &cursor); } @@ -1583,6 +1589,7 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,struct
amdgpu_vm_update_paramsparams; enumamdgpu_sync_modesync_mode;
int r;drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +3
-0+struct list_head head;memset (¶ms,0 ,sizeof (params)); params.adev = adev;@@ -1590,6 +1597,8 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev, params.direct = direct; params.pages_addr = pages_addr; +INIT_LIST_HEAD (&head); +/* Implicitly sync to command submissions in the same VM before * unmapping. Sync to moving fences before mapping. */ @@ -1617,13 +1626,22 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,if (r) goto error_unlock; - r =amdgpu_vm_update_ptes (¶ms, start, last +1 , addr, flags); + r =amdgpu_vm_update_ptes (¶ms, start, last +1 , addr, flags, &head);if (r) goto error_unlock;drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev, · line 1641
Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +8
-0r = vm->update_funcs->commit (¶ms, fence); error_unlock: +while (!list_empty (&head)) { +struct amdgpu_vm_pt *entry =list_first_entry (&head, + struct amdgpu_vm_pt, base.vm_status); +list_del (&entry->base.vm_status); + +amdgpu_bo_unref (&entry->base.bo->shadow); +amdgpu_bo_unref (&entry->base.bo); + }Maybe put that into a separate function.
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm) · line 3139
Quoted 9 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +2
-0+amdgpu_vm_eviction_unlock (vm);return r; }@@ -3118,7 +3136,7 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm)amdgpu_vm_free_mapping (adev, vm, mapping,NULL ); } -amdgpu_vm_free_pts (adev, vm,NULL ); +amdgpu_vm_free_pts (adev, vm,NULL ,NULL );And then also call it here.
Regards,
Christian.Quoted 3 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +0
-0amdgpu_bo_unreserve (root);amdgpu_bo_unref (&root);WARN_ON (vm->root.base.bo);
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx