* [PATCH] drm/radeon: stop using addr to check for BO move
@ 2015-05-23 18:58 Christian König
2015-05-23 19:06 ` Christian König
0 siblings, 1 reply; 5+ messages in thread
From: Christian König @ 2015-05-23 18:58 UTC (permalink / raw)
To: alexdeucher; +Cc: dri-devel
From: Christian König <christian.koenig@amd.com>
It is theoretically possible that a swapped out BO gets the
same GTT address, but different backing pages while being swapped in.
Instead just use another VA state to note updated areas.
Signed-off-by: Christian König <christian.koenig@amd.com>
---
drivers/gpu/drm/radeon/radeon.h | 4 ++-
drivers/gpu/drm/radeon/radeon_vm.c | 53 +++++++++++++++++++++-----------------
2 files changed, 32 insertions(+), 25 deletions(-)
diff --git a/drivers/gpu/drm/radeon/radeon.h b/drivers/gpu/drm/radeon/radeon.h
index 46eb0fa..ef7df51 100644
--- a/drivers/gpu/drm/radeon/radeon.h
+++ b/drivers/gpu/drm/radeon/radeon.h
@@ -467,7 +467,6 @@ struct radeon_bo_va {
/* protected by bo being reserved */
struct list_head bo_list;
uint32_t flags;
- uint64_t addr;
struct radeon_fence *last_pt_update;
unsigned ref_count;
@@ -941,6 +940,9 @@ struct radeon_vm {
/* BOs freed, but not yet updated in the PT */
struct list_head freed;
+ /* BOs cleared in the PT */
+ struct list_head cleared;
+
/* contains the page directory */
struct radeon_bo *page_directory;
unsigned max_pde_used;
diff --git a/drivers/gpu/drm/radeon/radeon_vm.c b/drivers/gpu/drm/radeon/radeon_vm.c
index de42fc4..9739ded 100644
--- a/drivers/gpu/drm/radeon/radeon_vm.c
+++ b/drivers/gpu/drm/radeon/radeon_vm.c
@@ -331,7 +331,6 @@ struct radeon_bo_va *radeon_vm_bo_add(struct radeon_device *rdev,
bo_va->it.start = 0;
bo_va->it.last = 0;
bo_va->flags = 0;
- bo_va->addr = 0;
bo_va->ref_count = 1;
INIT_LIST_HEAD(&bo_va->bo_list);
INIT_LIST_HEAD(&bo_va->vm_status);
@@ -491,9 +490,11 @@ int radeon_vm_bo_set_addr(struct radeon_device *rdev,
}
if (bo_va->it.start || bo_va->it.last) {
- if (bo_va->addr) {
+ spin_lock(&vm->status_lock);
+ if (list_empty(&bo_va->vm_status)) {
/* add a clone of the bo_va to clear the old address */
struct radeon_bo_va *tmp;
+ spin_unlock(&vm->status_lock);
tmp = kzalloc(sizeof(struct radeon_bo_va), GFP_KERNEL);
if (!tmp) {
mutex_unlock(&vm->mutex);
@@ -502,14 +503,11 @@ int radeon_vm_bo_set_addr(struct radeon_device *rdev,
tmp->it.start = bo_va->it.start;
tmp->it.last = bo_va->it.last;
tmp->vm = vm;
- tmp->addr = bo_va->addr;
tmp->bo = radeon_bo_ref(bo_va->bo);
spin_lock(&vm->status_lock);
list_add(&tmp->vm_status, &vm->freed);
- spin_unlock(&vm->status_lock);
-
- bo_va->addr = 0;
}
+ spin_unlock(&vm->status_lock);
interval_tree_remove(&bo_va->it, &vm->va);
bo_va->it.start = 0;
@@ -520,10 +518,12 @@ int radeon_vm_bo_set_addr(struct radeon_device *rdev,
bo_va->it.start = soffset;
bo_va->it.last = eoffset - 1;
interval_tree_insert(&bo_va->it, &vm->va);
+ spin_lock(&vm->status_lock);
+ list_add(&bo_va->vm_status, &vm->cleared);
+ spin_unlock(&vm->status_lock);
}
bo_va->flags = flags;
- bo_va->addr = 0;
soffset >>= radeon_vm_block_size;
eoffset >>= radeon_vm_block_size;
@@ -921,7 +921,16 @@ int radeon_vm_bo_update(struct radeon_device *rdev,
}
spin_lock(&vm->status_lock);
- list_del_init(&bo_va->vm_status);
+ if (mem) {
+ if (list_empty(&bo_va->vm_status)) {
+ spin_unlock(&vm->status_lock);
+ return 0;
+ }
+ list_del_init(&bo_va->vm_status);
+ } else {
+ list_del(&bo_va->vm_status);
+ list_add(&bo_va->vm_status, &vm->cleared);
+ }
spin_unlock(&vm->status_lock);
bo_va->flags &= ~RADEON_VM_PAGE_VALID;
@@ -947,10 +956,6 @@ int radeon_vm_bo_update(struct radeon_device *rdev,
addr = 0;
}
- if (addr == bo_va->addr)
- return 0;
- bo_va->addr = addr;
-
trace_radeon_vm_bo_update(bo_va);
nptes = bo_va->it.last - bo_va->it.start + 1;
@@ -1038,7 +1043,7 @@ int radeon_vm_clear_freed(struct radeon_device *rdev,
struct radeon_vm *vm)
{
struct radeon_bo_va *bo_va;
- int r;
+ int r = 0;
spin_lock(&vm->status_lock);
while (!list_empty(&vm->freed)) {
@@ -1049,14 +1054,15 @@ int radeon_vm_clear_freed(struct radeon_device *rdev,
r = radeon_vm_bo_update(rdev, bo_va, NULL);
radeon_bo_unref(&bo_va->bo);
radeon_fence_unref(&bo_va->last_pt_update);
+ spin_lock(&vm->status_lock);
+ list_del(&bo_va->vm_status);
kfree(bo_va);
if (r)
- return r;
+ break;
- spin_lock(&vm->status_lock);
}
spin_unlock(&vm->status_lock);
- return 0;
+ return r;
}
@@ -1114,14 +1120,14 @@ void radeon_vm_bo_rmv(struct radeon_device *rdev,
mutex_lock(&vm->mutex);
if (bo_va->it.start || bo_va->it.last)
interval_tree_remove(&bo_va->it, &vm->va);
- spin_lock(&vm->status_lock);
- list_del(&bo_va->vm_status);
- if (bo_va->addr) {
+ spin_lock(&vm->status_lock);
+ if (list_empty(&bo_va->vm_status)) {
bo_va->bo = radeon_bo_ref(bo_va->bo);
list_add(&bo_va->vm_status, &vm->freed);
} else {
radeon_fence_unref(&bo_va->last_pt_update);
+ list_del(&bo_va->vm_status);
kfree(bo_va);
}
spin_unlock(&vm->status_lock);
@@ -1144,12 +1150,10 @@ void radeon_vm_bo_invalidate(struct radeon_device *rdev,
struct radeon_bo_va *bo_va;
list_for_each_entry(bo_va, &bo->va, bo_list) {
- if (bo_va->addr) {
- spin_lock(&bo_va->vm->status_lock);
- list_del(&bo_va->vm_status);
+ spin_lock(&bo_va->vm->status_lock);
+ if (list_empty(&bo_va->vm_status))
list_add(&bo_va->vm_status, &bo_va->vm->invalidated);
- spin_unlock(&bo_va->vm->status_lock);
- }
+ spin_unlock(&bo_va->vm->status_lock);
}
}
@@ -1179,6 +1183,7 @@ int radeon_vm_init(struct radeon_device *rdev, struct radeon_vm *vm)
spin_lock_init(&vm->status_lock);
INIT_LIST_HEAD(&vm->invalidated);
INIT_LIST_HEAD(&vm->freed);
+ INIT_LIST_HEAD(&vm->cleared);
pd_size = radeon_vm_directory_size(rdev);
pd_entries = radeon_vm_num_pdes(rdev);
--
1.9.1
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/dri-devel
^ permalink raw reply related [flat|nested] 5+ messages in thread* Re: [PATCH] drm/radeon: stop using addr to check for BO move
2015-05-23 18:58 [PATCH] drm/radeon: stop using addr to check for BO move Christian König
@ 2015-05-23 19:06 ` Christian König
2015-05-26 10:26 ` Christian König
0 siblings, 1 reply; 5+ messages in thread
From: Christian König @ 2015-05-23 19:06 UTC (permalink / raw)
To: alexdeucher; +Cc: dri-devel
On 23.05.2015 20:58, Christian König wrote:
> From: Christian König <christian.koenig@amd.com>
>
> It is theoretically possible that a swapped out BO gets the
> same GTT address, but different backing pages while being swapped in.
>
> Instead just use another VA state to note updated areas.
Alex, please ignore this one. I've send it out preliminary and there are
still bugs in it.
Regards,
Christian.
>
> Signed-off-by: Christian König <christian.koenig@amd.com>
> ---
> drivers/gpu/drm/radeon/radeon.h | 4 ++-
> drivers/gpu/drm/radeon/radeon_vm.c | 53 +++++++++++++++++++++-----------------
> 2 files changed, 32 insertions(+), 25 deletions(-)
>
> diff --git a/drivers/gpu/drm/radeon/radeon.h b/drivers/gpu/drm/radeon/radeon.h
> index 46eb0fa..ef7df51 100644
> --- a/drivers/gpu/drm/radeon/radeon.h
> +++ b/drivers/gpu/drm/radeon/radeon.h
> @@ -467,7 +467,6 @@ struct radeon_bo_va {
> /* protected by bo being reserved */
> struct list_head bo_list;
> uint32_t flags;
> - uint64_t addr;
> struct radeon_fence *last_pt_update;
> unsigned ref_count;
>
> @@ -941,6 +940,9 @@ struct radeon_vm {
> /* BOs freed, but not yet updated in the PT */
> struct list_head freed;
>
> + /* BOs cleared in the PT */
> + struct list_head cleared;
> +
> /* contains the page directory */
> struct radeon_bo *page_directory;
> unsigned max_pde_used;
> diff --git a/drivers/gpu/drm/radeon/radeon_vm.c b/drivers/gpu/drm/radeon/radeon_vm.c
> index de42fc4..9739ded 100644
> --- a/drivers/gpu/drm/radeon/radeon_vm.c
> +++ b/drivers/gpu/drm/radeon/radeon_vm.c
> @@ -331,7 +331,6 @@ struct radeon_bo_va *radeon_vm_bo_add(struct radeon_device *rdev,
> bo_va->it.start = 0;
> bo_va->it.last = 0;
> bo_va->flags = 0;
> - bo_va->addr = 0;
> bo_va->ref_count = 1;
> INIT_LIST_HEAD(&bo_va->bo_list);
> INIT_LIST_HEAD(&bo_va->vm_status);
> @@ -491,9 +490,11 @@ int radeon_vm_bo_set_addr(struct radeon_device *rdev,
> }
>
> if (bo_va->it.start || bo_va->it.last) {
> - if (bo_va->addr) {
> + spin_lock(&vm->status_lock);
> + if (list_empty(&bo_va->vm_status)) {
> /* add a clone of the bo_va to clear the old address */
> struct radeon_bo_va *tmp;
> + spin_unlock(&vm->status_lock);
> tmp = kzalloc(sizeof(struct radeon_bo_va), GFP_KERNEL);
> if (!tmp) {
> mutex_unlock(&vm->mutex);
> @@ -502,14 +503,11 @@ int radeon_vm_bo_set_addr(struct radeon_device *rdev,
> tmp->it.start = bo_va->it.start;
> tmp->it.last = bo_va->it.last;
> tmp->vm = vm;
> - tmp->addr = bo_va->addr;
> tmp->bo = radeon_bo_ref(bo_va->bo);
> spin_lock(&vm->status_lock);
> list_add(&tmp->vm_status, &vm->freed);
> - spin_unlock(&vm->status_lock);
> -
> - bo_va->addr = 0;
> }
> + spin_unlock(&vm->status_lock);
>
> interval_tree_remove(&bo_va->it, &vm->va);
> bo_va->it.start = 0;
> @@ -520,10 +518,12 @@ int radeon_vm_bo_set_addr(struct radeon_device *rdev,
> bo_va->it.start = soffset;
> bo_va->it.last = eoffset - 1;
> interval_tree_insert(&bo_va->it, &vm->va);
> + spin_lock(&vm->status_lock);
> + list_add(&bo_va->vm_status, &vm->cleared);
> + spin_unlock(&vm->status_lock);
> }
>
> bo_va->flags = flags;
> - bo_va->addr = 0;
>
> soffset >>= radeon_vm_block_size;
> eoffset >>= radeon_vm_block_size;
> @@ -921,7 +921,16 @@ int radeon_vm_bo_update(struct radeon_device *rdev,
> }
>
> spin_lock(&vm->status_lock);
> - list_del_init(&bo_va->vm_status);
> + if (mem) {
> + if (list_empty(&bo_va->vm_status)) {
> + spin_unlock(&vm->status_lock);
> + return 0;
> + }
> + list_del_init(&bo_va->vm_status);
> + } else {
> + list_del(&bo_va->vm_status);
> + list_add(&bo_va->vm_status, &vm->cleared);
> + }
> spin_unlock(&vm->status_lock);
>
> bo_va->flags &= ~RADEON_VM_PAGE_VALID;
> @@ -947,10 +956,6 @@ int radeon_vm_bo_update(struct radeon_device *rdev,
> addr = 0;
> }
>
> - if (addr == bo_va->addr)
> - return 0;
> - bo_va->addr = addr;
> -
> trace_radeon_vm_bo_update(bo_va);
>
> nptes = bo_va->it.last - bo_va->it.start + 1;
> @@ -1038,7 +1043,7 @@ int radeon_vm_clear_freed(struct radeon_device *rdev,
> struct radeon_vm *vm)
> {
> struct radeon_bo_va *bo_va;
> - int r;
> + int r = 0;
>
> spin_lock(&vm->status_lock);
> while (!list_empty(&vm->freed)) {
> @@ -1049,14 +1054,15 @@ int radeon_vm_clear_freed(struct radeon_device *rdev,
> r = radeon_vm_bo_update(rdev, bo_va, NULL);
> radeon_bo_unref(&bo_va->bo);
> radeon_fence_unref(&bo_va->last_pt_update);
> + spin_lock(&vm->status_lock);
> + list_del(&bo_va->vm_status);
> kfree(bo_va);
> if (r)
> - return r;
> + break;
>
> - spin_lock(&vm->status_lock);
> }
> spin_unlock(&vm->status_lock);
> - return 0;
> + return r;
>
> }
>
> @@ -1114,14 +1120,14 @@ void radeon_vm_bo_rmv(struct radeon_device *rdev,
> mutex_lock(&vm->mutex);
> if (bo_va->it.start || bo_va->it.last)
> interval_tree_remove(&bo_va->it, &vm->va);
> - spin_lock(&vm->status_lock);
> - list_del(&bo_va->vm_status);
>
> - if (bo_va->addr) {
> + spin_lock(&vm->status_lock);
> + if (list_empty(&bo_va->vm_status)) {
> bo_va->bo = radeon_bo_ref(bo_va->bo);
> list_add(&bo_va->vm_status, &vm->freed);
> } else {
> radeon_fence_unref(&bo_va->last_pt_update);
> + list_del(&bo_va->vm_status);
> kfree(bo_va);
> }
> spin_unlock(&vm->status_lock);
> @@ -1144,12 +1150,10 @@ void radeon_vm_bo_invalidate(struct radeon_device *rdev,
> struct radeon_bo_va *bo_va;
>
> list_for_each_entry(bo_va, &bo->va, bo_list) {
> - if (bo_va->addr) {
> - spin_lock(&bo_va->vm->status_lock);
> - list_del(&bo_va->vm_status);
> + spin_lock(&bo_va->vm->status_lock);
> + if (list_empty(&bo_va->vm_status))
> list_add(&bo_va->vm_status, &bo_va->vm->invalidated);
> - spin_unlock(&bo_va->vm->status_lock);
> - }
> + spin_unlock(&bo_va->vm->status_lock);
> }
> }
>
> @@ -1179,6 +1183,7 @@ int radeon_vm_init(struct radeon_device *rdev, struct radeon_vm *vm)
> spin_lock_init(&vm->status_lock);
> INIT_LIST_HEAD(&vm->invalidated);
> INIT_LIST_HEAD(&vm->freed);
> + INIT_LIST_HEAD(&vm->cleared);
>
> pd_size = radeon_vm_directory_size(rdev);
> pd_entries = radeon_vm_num_pdes(rdev);
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/dri-devel
^ permalink raw reply [flat|nested] 5+ messages in thread* Re: [PATCH] drm/radeon: stop using addr to check for BO move
2015-05-23 19:06 ` Christian König
@ 2015-05-26 10:26 ` Christian König
0 siblings, 0 replies; 5+ messages in thread
From: Christian König @ 2015-05-26 10:26 UTC (permalink / raw)
To: alexdeucher; +Cc: dri-devel
On 23.05.2015 21:06, Christian König wrote:
> On 23.05.2015 20:58, Christian König wrote:
>> From: Christian König <christian.koenig@amd.com>
>>
>> It is theoretically possible that a swapped out BO gets the
>> same GTT address, but different backing pages while being swapped in.
>>
>> Instead just use another VA state to note updated areas.
>
> Alex, please ignore this one. I've send it out preliminary and there
> are still bugs in it.
More testing showed that the problem wasn't in this patch and this one
actually works perfectly fine.
I've just send it out one more time, please include it in your 4.2
branch since I want to give it a bit more testing before it goes upstream.
Regards,
Christian.
>
> Regards,
> Christian.
>
>>
>> Signed-off-by: Christian König <christian.koenig@amd.com>
>> ---
>> drivers/gpu/drm/radeon/radeon.h | 4 ++-
>> drivers/gpu/drm/radeon/radeon_vm.c | 53
>> +++++++++++++++++++++-----------------
>> 2 files changed, 32 insertions(+), 25 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/radeon/radeon.h
>> b/drivers/gpu/drm/radeon/radeon.h
>> index 46eb0fa..ef7df51 100644
>> --- a/drivers/gpu/drm/radeon/radeon.h
>> +++ b/drivers/gpu/drm/radeon/radeon.h
>> @@ -467,7 +467,6 @@ struct radeon_bo_va {
>> /* protected by bo being reserved */
>> struct list_head bo_list;
>> uint32_t flags;
>> - uint64_t addr;
>> struct radeon_fence *last_pt_update;
>> unsigned ref_count;
>> @@ -941,6 +940,9 @@ struct radeon_vm {
>> /* BOs freed, but not yet updated in the PT */
>> struct list_head freed;
>> + /* BOs cleared in the PT */
>> + struct list_head cleared;
>> +
>> /* contains the page directory */
>> struct radeon_bo *page_directory;
>> unsigned max_pde_used;
>> diff --git a/drivers/gpu/drm/radeon/radeon_vm.c
>> b/drivers/gpu/drm/radeon/radeon_vm.c
>> index de42fc4..9739ded 100644
>> --- a/drivers/gpu/drm/radeon/radeon_vm.c
>> +++ b/drivers/gpu/drm/radeon/radeon_vm.c
>> @@ -331,7 +331,6 @@ struct radeon_bo_va *radeon_vm_bo_add(struct
>> radeon_device *rdev,
>> bo_va->it.start = 0;
>> bo_va->it.last = 0;
>> bo_va->flags = 0;
>> - bo_va->addr = 0;
>> bo_va->ref_count = 1;
>> INIT_LIST_HEAD(&bo_va->bo_list);
>> INIT_LIST_HEAD(&bo_va->vm_status);
>> @@ -491,9 +490,11 @@ int radeon_vm_bo_set_addr(struct radeon_device
>> *rdev,
>> }
>> if (bo_va->it.start || bo_va->it.last) {
>> - if (bo_va->addr) {
>> + spin_lock(&vm->status_lock);
>> + if (list_empty(&bo_va->vm_status)) {
>> /* add a clone of the bo_va to clear the old address */
>> struct radeon_bo_va *tmp;
>> + spin_unlock(&vm->status_lock);
>> tmp = kzalloc(sizeof(struct radeon_bo_va), GFP_KERNEL);
>> if (!tmp) {
>> mutex_unlock(&vm->mutex);
>> @@ -502,14 +503,11 @@ int radeon_vm_bo_set_addr(struct radeon_device
>> *rdev,
>> tmp->it.start = bo_va->it.start;
>> tmp->it.last = bo_va->it.last;
>> tmp->vm = vm;
>> - tmp->addr = bo_va->addr;
>> tmp->bo = radeon_bo_ref(bo_va->bo);
>> spin_lock(&vm->status_lock);
>> list_add(&tmp->vm_status, &vm->freed);
>> - spin_unlock(&vm->status_lock);
>> -
>> - bo_va->addr = 0;
>> }
>> + spin_unlock(&vm->status_lock);
>> interval_tree_remove(&bo_va->it, &vm->va);
>> bo_va->it.start = 0;
>> @@ -520,10 +518,12 @@ int radeon_vm_bo_set_addr(struct radeon_device
>> *rdev,
>> bo_va->it.start = soffset;
>> bo_va->it.last = eoffset - 1;
>> interval_tree_insert(&bo_va->it, &vm->va);
>> + spin_lock(&vm->status_lock);
>> + list_add(&bo_va->vm_status, &vm->cleared);
>> + spin_unlock(&vm->status_lock);
>> }
>> bo_va->flags = flags;
>> - bo_va->addr = 0;
>> soffset >>= radeon_vm_block_size;
>> eoffset >>= radeon_vm_block_size;
>> @@ -921,7 +921,16 @@ int radeon_vm_bo_update(struct radeon_device *rdev,
>> }
>> spin_lock(&vm->status_lock);
>> - list_del_init(&bo_va->vm_status);
>> + if (mem) {
>> + if (list_empty(&bo_va->vm_status)) {
>> + spin_unlock(&vm->status_lock);
>> + return 0;
>> + }
>> + list_del_init(&bo_va->vm_status);
>> + } else {
>> + list_del(&bo_va->vm_status);
>> + list_add(&bo_va->vm_status, &vm->cleared);
>> + }
>> spin_unlock(&vm->status_lock);
>> bo_va->flags &= ~RADEON_VM_PAGE_VALID;
>> @@ -947,10 +956,6 @@ int radeon_vm_bo_update(struct radeon_device *rdev,
>> addr = 0;
>> }
>> - if (addr == bo_va->addr)
>> - return 0;
>> - bo_va->addr = addr;
>> -
>> trace_radeon_vm_bo_update(bo_va);
>> nptes = bo_va->it.last - bo_va->it.start + 1;
>> @@ -1038,7 +1043,7 @@ int radeon_vm_clear_freed(struct radeon_device
>> *rdev,
>> struct radeon_vm *vm)
>> {
>> struct radeon_bo_va *bo_va;
>> - int r;
>> + int r = 0;
>> spin_lock(&vm->status_lock);
>> while (!list_empty(&vm->freed)) {
>> @@ -1049,14 +1054,15 @@ int radeon_vm_clear_freed(struct
>> radeon_device *rdev,
>> r = radeon_vm_bo_update(rdev, bo_va, NULL);
>> radeon_bo_unref(&bo_va->bo);
>> radeon_fence_unref(&bo_va->last_pt_update);
>> + spin_lock(&vm->status_lock);
>> + list_del(&bo_va->vm_status);
>> kfree(bo_va);
>> if (r)
>> - return r;
>> + break;
>> - spin_lock(&vm->status_lock);
>> }
>> spin_unlock(&vm->status_lock);
>> - return 0;
>> + return r;
>> }
>> @@ -1114,14 +1120,14 @@ void radeon_vm_bo_rmv(struct radeon_device
>> *rdev,
>> mutex_lock(&vm->mutex);
>> if (bo_va->it.start || bo_va->it.last)
>> interval_tree_remove(&bo_va->it, &vm->va);
>> - spin_lock(&vm->status_lock);
>> - list_del(&bo_va->vm_status);
>> - if (bo_va->addr) {
>> + spin_lock(&vm->status_lock);
>> + if (list_empty(&bo_va->vm_status)) {
>> bo_va->bo = radeon_bo_ref(bo_va->bo);
>> list_add(&bo_va->vm_status, &vm->freed);
>> } else {
>> radeon_fence_unref(&bo_va->last_pt_update);
>> + list_del(&bo_va->vm_status);
>> kfree(bo_va);
>> }
>> spin_unlock(&vm->status_lock);
>> @@ -1144,12 +1150,10 @@ void radeon_vm_bo_invalidate(struct
>> radeon_device *rdev,
>> struct radeon_bo_va *bo_va;
>> list_for_each_entry(bo_va, &bo->va, bo_list) {
>> - if (bo_va->addr) {
>> - spin_lock(&bo_va->vm->status_lock);
>> - list_del(&bo_va->vm_status);
>> + spin_lock(&bo_va->vm->status_lock);
>> + if (list_empty(&bo_va->vm_status))
>> list_add(&bo_va->vm_status, &bo_va->vm->invalidated);
>> - spin_unlock(&bo_va->vm->status_lock);
>> - }
>> + spin_unlock(&bo_va->vm->status_lock);
>> }
>> }
>> @@ -1179,6 +1183,7 @@ int radeon_vm_init(struct radeon_device
>> *rdev, struct radeon_vm *vm)
>> spin_lock_init(&vm->status_lock);
>> INIT_LIST_HEAD(&vm->invalidated);
>> INIT_LIST_HEAD(&vm->freed);
>> + INIT_LIST_HEAD(&vm->cleared);
>> pd_size = radeon_vm_directory_size(rdev);
>> pd_entries = radeon_vm_num_pdes(rdev);
>
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/dri-devel
^ permalink raw reply [flat|nested] 5+ messages in thread
* [PATCH] drm/radeon: stop using addr to check for BO move
@ 2015-05-26 10:24 Christian König
2015-05-26 14:35 ` Alex Deucher
0 siblings, 1 reply; 5+ messages in thread
From: Christian König @ 2015-05-26 10:24 UTC (permalink / raw)
To: alexdeucher; +Cc: dri-devel
From: Christian König <christian.koenig@amd.com>
It is theoretically possible that a swapped out BO gets the
same GTT address, but different backing pages while being swapped in.
Instead just use another VA state to note updated areas.
Signed-off-by: Christian König <christian.koenig@amd.com>
---
drivers/gpu/drm/radeon/radeon.h | 4 ++-
drivers/gpu/drm/radeon/radeon_vm.c | 53 +++++++++++++++++++++-----------------
2 files changed, 32 insertions(+), 25 deletions(-)
diff --git a/drivers/gpu/drm/radeon/radeon.h b/drivers/gpu/drm/radeon/radeon.h
index 46eb0fa..ef7df51 100644
--- a/drivers/gpu/drm/radeon/radeon.h
+++ b/drivers/gpu/drm/radeon/radeon.h
@@ -467,7 +467,6 @@ struct radeon_bo_va {
/* protected by bo being reserved */
struct list_head bo_list;
uint32_t flags;
- uint64_t addr;
struct radeon_fence *last_pt_update;
unsigned ref_count;
@@ -941,6 +940,9 @@ struct radeon_vm {
/* BOs freed, but not yet updated in the PT */
struct list_head freed;
+ /* BOs cleared in the PT */
+ struct list_head cleared;
+
/* contains the page directory */
struct radeon_bo *page_directory;
unsigned max_pde_used;
diff --git a/drivers/gpu/drm/radeon/radeon_vm.c b/drivers/gpu/drm/radeon/radeon_vm.c
index de42fc4..9739ded 100644
--- a/drivers/gpu/drm/radeon/radeon_vm.c
+++ b/drivers/gpu/drm/radeon/radeon_vm.c
@@ -331,7 +331,6 @@ struct radeon_bo_va *radeon_vm_bo_add(struct radeon_device *rdev,
bo_va->it.start = 0;
bo_va->it.last = 0;
bo_va->flags = 0;
- bo_va->addr = 0;
bo_va->ref_count = 1;
INIT_LIST_HEAD(&bo_va->bo_list);
INIT_LIST_HEAD(&bo_va->vm_status);
@@ -491,9 +490,11 @@ int radeon_vm_bo_set_addr(struct radeon_device *rdev,
}
if (bo_va->it.start || bo_va->it.last) {
- if (bo_va->addr) {
+ spin_lock(&vm->status_lock);
+ if (list_empty(&bo_va->vm_status)) {
/* add a clone of the bo_va to clear the old address */
struct radeon_bo_va *tmp;
+ spin_unlock(&vm->status_lock);
tmp = kzalloc(sizeof(struct radeon_bo_va), GFP_KERNEL);
if (!tmp) {
mutex_unlock(&vm->mutex);
@@ -502,14 +503,11 @@ int radeon_vm_bo_set_addr(struct radeon_device *rdev,
tmp->it.start = bo_va->it.start;
tmp->it.last = bo_va->it.last;
tmp->vm = vm;
- tmp->addr = bo_va->addr;
tmp->bo = radeon_bo_ref(bo_va->bo);
spin_lock(&vm->status_lock);
list_add(&tmp->vm_status, &vm->freed);
- spin_unlock(&vm->status_lock);
-
- bo_va->addr = 0;
}
+ spin_unlock(&vm->status_lock);
interval_tree_remove(&bo_va->it, &vm->va);
bo_va->it.start = 0;
@@ -520,10 +518,12 @@ int radeon_vm_bo_set_addr(struct radeon_device *rdev,
bo_va->it.start = soffset;
bo_va->it.last = eoffset - 1;
interval_tree_insert(&bo_va->it, &vm->va);
+ spin_lock(&vm->status_lock);
+ list_add(&bo_va->vm_status, &vm->cleared);
+ spin_unlock(&vm->status_lock);
}
bo_va->flags = flags;
- bo_va->addr = 0;
soffset >>= radeon_vm_block_size;
eoffset >>= radeon_vm_block_size;
@@ -921,7 +921,16 @@ int radeon_vm_bo_update(struct radeon_device *rdev,
}
spin_lock(&vm->status_lock);
- list_del_init(&bo_va->vm_status);
+ if (mem) {
+ if (list_empty(&bo_va->vm_status)) {
+ spin_unlock(&vm->status_lock);
+ return 0;
+ }
+ list_del_init(&bo_va->vm_status);
+ } else {
+ list_del(&bo_va->vm_status);
+ list_add(&bo_va->vm_status, &vm->cleared);
+ }
spin_unlock(&vm->status_lock);
bo_va->flags &= ~RADEON_VM_PAGE_VALID;
@@ -947,10 +956,6 @@ int radeon_vm_bo_update(struct radeon_device *rdev,
addr = 0;
}
- if (addr == bo_va->addr)
- return 0;
- bo_va->addr = addr;
-
trace_radeon_vm_bo_update(bo_va);
nptes = bo_va->it.last - bo_va->it.start + 1;
@@ -1038,7 +1043,7 @@ int radeon_vm_clear_freed(struct radeon_device *rdev,
struct radeon_vm *vm)
{
struct radeon_bo_va *bo_va;
- int r;
+ int r = 0;
spin_lock(&vm->status_lock);
while (!list_empty(&vm->freed)) {
@@ -1049,14 +1054,15 @@ int radeon_vm_clear_freed(struct radeon_device *rdev,
r = radeon_vm_bo_update(rdev, bo_va, NULL);
radeon_bo_unref(&bo_va->bo);
radeon_fence_unref(&bo_va->last_pt_update);
+ spin_lock(&vm->status_lock);
+ list_del(&bo_va->vm_status);
kfree(bo_va);
if (r)
- return r;
+ break;
- spin_lock(&vm->status_lock);
}
spin_unlock(&vm->status_lock);
- return 0;
+ return r;
}
@@ -1114,14 +1120,14 @@ void radeon_vm_bo_rmv(struct radeon_device *rdev,
mutex_lock(&vm->mutex);
if (bo_va->it.start || bo_va->it.last)
interval_tree_remove(&bo_va->it, &vm->va);
- spin_lock(&vm->status_lock);
- list_del(&bo_va->vm_status);
- if (bo_va->addr) {
+ spin_lock(&vm->status_lock);
+ if (list_empty(&bo_va->vm_status)) {
bo_va->bo = radeon_bo_ref(bo_va->bo);
list_add(&bo_va->vm_status, &vm->freed);
} else {
radeon_fence_unref(&bo_va->last_pt_update);
+ list_del(&bo_va->vm_status);
kfree(bo_va);
}
spin_unlock(&vm->status_lock);
@@ -1144,12 +1150,10 @@ void radeon_vm_bo_invalidate(struct radeon_device *rdev,
struct radeon_bo_va *bo_va;
list_for_each_entry(bo_va, &bo->va, bo_list) {
- if (bo_va->addr) {
- spin_lock(&bo_va->vm->status_lock);
- list_del(&bo_va->vm_status);
+ spin_lock(&bo_va->vm->status_lock);
+ if (list_empty(&bo_va->vm_status))
list_add(&bo_va->vm_status, &bo_va->vm->invalidated);
- spin_unlock(&bo_va->vm->status_lock);
- }
+ spin_unlock(&bo_va->vm->status_lock);
}
}
@@ -1179,6 +1183,7 @@ int radeon_vm_init(struct radeon_device *rdev, struct radeon_vm *vm)
spin_lock_init(&vm->status_lock);
INIT_LIST_HEAD(&vm->invalidated);
INIT_LIST_HEAD(&vm->freed);
+ INIT_LIST_HEAD(&vm->cleared);
pd_size = radeon_vm_directory_size(rdev);
pd_entries = radeon_vm_num_pdes(rdev);
--
1.9.1
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/dri-devel
^ permalink raw reply related [flat|nested] 5+ messages in thread* Re: [PATCH] drm/radeon: stop using addr to check for BO move
2015-05-26 10:24 Christian König
@ 2015-05-26 14:35 ` Alex Deucher
0 siblings, 0 replies; 5+ messages in thread
From: Alex Deucher @ 2015-05-26 14:35 UTC (permalink / raw)
To: Christian König; +Cc: Maling list - DRI developers
On Tue, May 26, 2015 at 6:24 AM, Christian König
<deathsimple@vodafone.de> wrote:
> From: Christian König <christian.koenig@amd.com>
>
> It is theoretically possible that a swapped out BO gets the
> same GTT address, but different backing pages while being swapped in.
>
> Instead just use another VA state to note updated areas.
>
> Signed-off-by: Christian König <christian.koenig@amd.com>
Applied to my -next tree.
Thanks!
Alex
> ---
> drivers/gpu/drm/radeon/radeon.h | 4 ++-
> drivers/gpu/drm/radeon/radeon_vm.c | 53 +++++++++++++++++++++-----------------
> 2 files changed, 32 insertions(+), 25 deletions(-)
>
> diff --git a/drivers/gpu/drm/radeon/radeon.h b/drivers/gpu/drm/radeon/radeon.h
> index 46eb0fa..ef7df51 100644
> --- a/drivers/gpu/drm/radeon/radeon.h
> +++ b/drivers/gpu/drm/radeon/radeon.h
> @@ -467,7 +467,6 @@ struct radeon_bo_va {
> /* protected by bo being reserved */
> struct list_head bo_list;
> uint32_t flags;
> - uint64_t addr;
> struct radeon_fence *last_pt_update;
> unsigned ref_count;
>
> @@ -941,6 +940,9 @@ struct radeon_vm {
> /* BOs freed, but not yet updated in the PT */
> struct list_head freed;
>
> + /* BOs cleared in the PT */
> + struct list_head cleared;
> +
> /* contains the page directory */
> struct radeon_bo *page_directory;
> unsigned max_pde_used;
> diff --git a/drivers/gpu/drm/radeon/radeon_vm.c b/drivers/gpu/drm/radeon/radeon_vm.c
> index de42fc4..9739ded 100644
> --- a/drivers/gpu/drm/radeon/radeon_vm.c
> +++ b/drivers/gpu/drm/radeon/radeon_vm.c
> @@ -331,7 +331,6 @@ struct radeon_bo_va *radeon_vm_bo_add(struct radeon_device *rdev,
> bo_va->it.start = 0;
> bo_va->it.last = 0;
> bo_va->flags = 0;
> - bo_va->addr = 0;
> bo_va->ref_count = 1;
> INIT_LIST_HEAD(&bo_va->bo_list);
> INIT_LIST_HEAD(&bo_va->vm_status);
> @@ -491,9 +490,11 @@ int radeon_vm_bo_set_addr(struct radeon_device *rdev,
> }
>
> if (bo_va->it.start || bo_va->it.last) {
> - if (bo_va->addr) {
> + spin_lock(&vm->status_lock);
> + if (list_empty(&bo_va->vm_status)) {
> /* add a clone of the bo_va to clear the old address */
> struct radeon_bo_va *tmp;
> + spin_unlock(&vm->status_lock);
> tmp = kzalloc(sizeof(struct radeon_bo_va), GFP_KERNEL);
> if (!tmp) {
> mutex_unlock(&vm->mutex);
> @@ -502,14 +503,11 @@ int radeon_vm_bo_set_addr(struct radeon_device *rdev,
> tmp->it.start = bo_va->it.start;
> tmp->it.last = bo_va->it.last;
> tmp->vm = vm;
> - tmp->addr = bo_va->addr;
> tmp->bo = radeon_bo_ref(bo_va->bo);
> spin_lock(&vm->status_lock);
> list_add(&tmp->vm_status, &vm->freed);
> - spin_unlock(&vm->status_lock);
> -
> - bo_va->addr = 0;
> }
> + spin_unlock(&vm->status_lock);
>
> interval_tree_remove(&bo_va->it, &vm->va);
> bo_va->it.start = 0;
> @@ -520,10 +518,12 @@ int radeon_vm_bo_set_addr(struct radeon_device *rdev,
> bo_va->it.start = soffset;
> bo_va->it.last = eoffset - 1;
> interval_tree_insert(&bo_va->it, &vm->va);
> + spin_lock(&vm->status_lock);
> + list_add(&bo_va->vm_status, &vm->cleared);
> + spin_unlock(&vm->status_lock);
> }
>
> bo_va->flags = flags;
> - bo_va->addr = 0;
>
> soffset >>= radeon_vm_block_size;
> eoffset >>= radeon_vm_block_size;
> @@ -921,7 +921,16 @@ int radeon_vm_bo_update(struct radeon_device *rdev,
> }
>
> spin_lock(&vm->status_lock);
> - list_del_init(&bo_va->vm_status);
> + if (mem) {
> + if (list_empty(&bo_va->vm_status)) {
> + spin_unlock(&vm->status_lock);
> + return 0;
> + }
> + list_del_init(&bo_va->vm_status);
> + } else {
> + list_del(&bo_va->vm_status);
> + list_add(&bo_va->vm_status, &vm->cleared);
> + }
> spin_unlock(&vm->status_lock);
>
> bo_va->flags &= ~RADEON_VM_PAGE_VALID;
> @@ -947,10 +956,6 @@ int radeon_vm_bo_update(struct radeon_device *rdev,
> addr = 0;
> }
>
> - if (addr == bo_va->addr)
> - return 0;
> - bo_va->addr = addr;
> -
> trace_radeon_vm_bo_update(bo_va);
>
> nptes = bo_va->it.last - bo_va->it.start + 1;
> @@ -1038,7 +1043,7 @@ int radeon_vm_clear_freed(struct radeon_device *rdev,
> struct radeon_vm *vm)
> {
> struct radeon_bo_va *bo_va;
> - int r;
> + int r = 0;
>
> spin_lock(&vm->status_lock);
> while (!list_empty(&vm->freed)) {
> @@ -1049,14 +1054,15 @@ int radeon_vm_clear_freed(struct radeon_device *rdev,
> r = radeon_vm_bo_update(rdev, bo_va, NULL);
> radeon_bo_unref(&bo_va->bo);
> radeon_fence_unref(&bo_va->last_pt_update);
> + spin_lock(&vm->status_lock);
> + list_del(&bo_va->vm_status);
> kfree(bo_va);
> if (r)
> - return r;
> + break;
>
> - spin_lock(&vm->status_lock);
> }
> spin_unlock(&vm->status_lock);
> - return 0;
> + return r;
>
> }
>
> @@ -1114,14 +1120,14 @@ void radeon_vm_bo_rmv(struct radeon_device *rdev,
> mutex_lock(&vm->mutex);
> if (bo_va->it.start || bo_va->it.last)
> interval_tree_remove(&bo_va->it, &vm->va);
> - spin_lock(&vm->status_lock);
> - list_del(&bo_va->vm_status);
>
> - if (bo_va->addr) {
> + spin_lock(&vm->status_lock);
> + if (list_empty(&bo_va->vm_status)) {
> bo_va->bo = radeon_bo_ref(bo_va->bo);
> list_add(&bo_va->vm_status, &vm->freed);
> } else {
> radeon_fence_unref(&bo_va->last_pt_update);
> + list_del(&bo_va->vm_status);
> kfree(bo_va);
> }
> spin_unlock(&vm->status_lock);
> @@ -1144,12 +1150,10 @@ void radeon_vm_bo_invalidate(struct radeon_device *rdev,
> struct radeon_bo_va *bo_va;
>
> list_for_each_entry(bo_va, &bo->va, bo_list) {
> - if (bo_va->addr) {
> - spin_lock(&bo_va->vm->status_lock);
> - list_del(&bo_va->vm_status);
> + spin_lock(&bo_va->vm->status_lock);
> + if (list_empty(&bo_va->vm_status))
> list_add(&bo_va->vm_status, &bo_va->vm->invalidated);
> - spin_unlock(&bo_va->vm->status_lock);
> - }
> + spin_unlock(&bo_va->vm->status_lock);
> }
> }
>
> @@ -1179,6 +1183,7 @@ int radeon_vm_init(struct radeon_device *rdev, struct radeon_vm *vm)
> spin_lock_init(&vm->status_lock);
> INIT_LIST_HEAD(&vm->invalidated);
> INIT_LIST_HEAD(&vm->freed);
> + INIT_LIST_HEAD(&vm->cleared);
>
> pd_size = radeon_vm_directory_size(rdev);
> pd_entries = radeon_vm_num_pdes(rdev);
> --
> 1.9.1
>
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/dri-devel
^ permalink raw reply [flat|nested] 5+ messages in thread
end of thread, other threads:[~2015-05-26 14:35 UTC | newest]
Thread overview: 5+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2015-05-23 18:58 [PATCH] drm/radeon: stop using addr to check for BO move Christian König
2015-05-23 19:06 ` Christian König
2015-05-26 10:26 ` Christian König
-- strict thread matches above, loose matches on Subject: below --
2015-05-26 10:24 Christian König
2015-05-26 14:35 ` Alex Deucher
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.