Allow num_binds/num_batch_buffer to be zero IOCTLs
Subsystems (2)
Matthew Brost
maintainercover Allow num_binds/num_batch_buffer to be zero IOCTLs
Easy way to flush exec queues. A semantic that VK will likely use.
Matthew Brost (5):
drm/xe: Allow num_binds == 0 in VM bind IOCTL
drm/xe: Use a flags field instead of bools for VMA create drm/xe: Use a flags field instead of bools for sync parse rm/xe: Allow num_batch_buffer == 0 in exec IOCTL
drm/xe: Take in-syncs into account when num_execs or num_binds == 0
drivers/gpu/drm/xe/xe_exec.c | 32 ++++++- drivers/gpu/drm/xe/xe_exec_queue.c | 5 +- drivers/gpu/drm/xe/xe_exec_queue_types.h | 5 +- drivers/gpu/drm/xe/xe_sync.c | 86 ++++++++++++++++- drivers/gpu/drm/xe/xe_sync.h | 11 ++- drivers/gpu/drm/xe/xe_vm.c | 117 ++++++++++++++--------- 6 files changed, 199 insertions(+), 57 deletions(-)
Signature · 1 line
2.34.1
Matthew Brost
maintainerPATCH 1/5 drm/xe: Allow num_binds == 0 in VM bind IOCTL
The idea being out-syncs can signal indicating all previous operations on the bind queue are complete. An example use case of this would be support for implementing vkQueueWaitForIdle easily.
--- drivers/gpu/drm/xe/xe_vm.c | 30 ++++++++++++++++++------------ 1 file changed, 18 insertions(+), 12 deletions(-)
drivers/gpu/drm/xe/xe_vm.c +18 -12
@@ -2850,7 +2850,6 @@ static int vm_bind_ioctl_check_args(struct xe_device *xe,int i;if (XE_IOCTL_DBG (xe, args->extensions) || -XE_IOCTL_DBG (xe, !args->num_binds) ||XE_IOCTL_DBG (xe, args->num_binds >MAX_BINDS ))return -EINVAL ;
@@ -2977,7 +2976,7 @@ int xe_vm_bind_ioctl(struct drm_device *dev, void *data, struct drm_file *file) goto put_exec_queue; } -if (XE_IOCTL_DBG (xe, async != +if (XE_IOCTL_DBG (xe, args->num_binds && async != !!(q->flags &EXEC_QUEUE_FLAG_VM_ASYNC ))) { err = -EINVAL ; goto put_exec_queue;
@@ -2991,7 +2990,7 @@ int xe_vm_bind_ioctl(struct drm_device *dev, void *data, struct drm_file *file) }if (!args->exec_queue_id) { -if (XE_IOCTL_DBG (xe, async != +if (XE_IOCTL_DBG (xe, args->num_binds && async != !!(vm->flags &XE_VM_FLAG_ASYNC_DEFAULT ))) { err = -EINVAL ; goto put_vm;
@@ -3028,16 +3027,18 @@ int xe_vm_bind_ioctl(struct drm_device *dev, void *data, struct drm_file *file) } } - bos =kzalloc (sizeof (*bos) * args->num_binds,GFP_KERNEL ); -if (!bos) { - err = -ENOMEM ; - goto release_vm_lock; - } +if (args->num_binds) { + bos =kzalloc (sizeof (*bos) * args->num_binds,GFP_KERNEL ); +if (!bos) { + err = -ENOMEM ; + goto release_vm_lock; + } - ops =kzalloc (sizeof (*ops) * args->num_binds,GFP_KERNEL ); -if (!ops) { - err = -ENOMEM ; - goto release_vm_lock; + ops =kzalloc (sizeof (*ops) * args->num_binds,GFP_KERNEL ); +if (!ops) { + err = -ENOMEM ; + goto release_vm_lock; + } }for (i =0 ; i < args->num_binds; ++i) {
@@ -3092,6 +3093,11 @@ int xe_vm_bind_ioctl(struct drm_device *dev, void *data, struct drm_file *file) goto free_syncs; } +if (!args->num_binds) { + err = -ENODATA ; + goto free_syncs; + } +for (i =0 ; i < args->num_binds; ++i) {u64 range = bind_ops[i].range;u64 addr = bind_ops[i].addr;
Signature · 1 line
2.34.1
Re: [Intel-xe] [PATCH 1/5] drm/xe: Allow num_binds == 0 in VM bind IOCTL
Quoted 3 lines · Matthew Brost
The idea being out-syncs can signal indicating all previous operations on the bind queue are complete. An example use case of this would be support for implementing vkQueueWaitForIdle easily.
s/vkQueueWaitForIdle/vkQueueWaitIdle
So this is just for the bind queue.
Not sure if it is useful, Mesa will eventually move to async vm bind and we will use a single TIMELINE_SYNCOBJ to track what bind operations are completed.
Paulo do you see any use with sparse?
Quoted 77 lines
Signed-off-by Matthew Brost <[email protected]>Reviewed-by Thomas Hellström <[email protected]>--- drivers/gpu/drm/xe/xe_vm.c | 30 ++++++++++++++++++------------ 1 file changed, 18 insertions(+), 12 deletions(-)drivers/gpu/drm/xe/xe_vm.c +18
-12@@ -2850,7 +2850,6 @@ static int vm_bind_ioctl_check_args(struct xe_device *xe,int i;if (XE_IOCTL_DBG (xe, args->extensions) || -XE_IOCTL_DBG (xe, !args->num_binds) ||XE_IOCTL_DBG (xe, args->num_binds >MAX_BINDS ))return -EINVAL ;@@ -2977,7 +2976,7 @@ int xe_vm_bind_ioctl(struct drm_device *dev, void *data, struct drm_file *file) goto put_exec_queue; } -if (XE_IOCTL_DBG (xe, async != +if (XE_IOCTL_DBG (xe, args->num_binds && async != !!(q->flags &EXEC_QUEUE_FLAG_VM_ASYNC ))) { err = -EINVAL ; goto put_exec_queue;@@ -2991,7 +2990,7 @@ int xe_vm_bind_ioctl(struct drm_device *dev, void *data, struct drm_file *file) }if (!args->exec_queue_id) { -if (XE_IOCTL_DBG (xe, async != +if (XE_IOCTL_DBG (xe, args->num_binds && async != !!(vm->flags &XE_VM_FLAG_ASYNC_DEFAULT ))) { err = -EINVAL ; goto put_vm;@@ -3028,16 +3027,18 @@ int xe_vm_bind_ioctl(struct drm_device *dev, void *data, struct drm_file *file) } } - bos =kzalloc (sizeof (*bos) * args->num_binds,GFP_KERNEL ); -if (!bos) { - err = -ENOMEM ; - goto release_vm_lock; - } +if (args->num_binds) { + bos =kzalloc (sizeof (*bos) * args->num_binds,GFP_KERNEL ); +if (!bos) { + err = -ENOMEM ; + goto release_vm_lock; + } - ops =kzalloc (sizeof (*ops) * args->num_binds,GFP_KERNEL ); -if (!ops) { - err = -ENOMEM ; - goto release_vm_lock; + ops =kzalloc (sizeof (*ops) * args->num_binds,GFP_KERNEL ); +if (!ops) { + err = -ENOMEM ; + goto release_vm_lock; + } }for (i =0 ; i < args->num_binds; ++i) {@@ -3092,6 +3093,11 @@ int xe_vm_bind_ioctl(struct drm_device *dev, void *data, struct drm_file *file) goto free_syncs; } +if (!args->num_binds) { + err = -ENODATA ; + goto free_syncs; + } +for (i =0 ; i < args->num_binds; ++i) {u64 range = bind_ops[i].range;u64 addr = bind_ops[i].addr;
Re: [Intel-xe] [PATCH 1/5] drm/xe: Allow num_binds == 0 in VM bind IOCTL
Quoted 11 lines · Jose
Quoted 3 lines · Matthew Brost
The idea being out-syncs can signal indicating all previous operations on the bind queue are complete. An example use case of this would be support for implementing vkQueueWaitForIdle easily.
s/vkQueueWaitForIdle/vkQueueWaitIdle
So this is just for the bind queue.
Not sure if it is useful, Mesa will eventually move to async vm bind and we will use a singleTIMELINE_SYNCOBJto track what bind operations are completed.
Paulo do you see any use with sparse?
We don't have plans to submit stuff to the Kernel with num_binds=0, but we do need the Kernel to support the case where we're trying to bind something that's already bound (so there's nothing for the Kernel to do, but num_binds > 0).
And yes to track operations that are already complete it's usually better to just check the syncobj (timeline or not) of the last thing you submitted.
Currently, due to the xe.ko bugs, our vm_bind code is completely synchronous, but the plan is to move it to async like we're doing with TR-TT.
I'm not discarding the possibility that we may have to do something extra for vkQueueWaitIdle, especially after we make it async, but I don't think submitting num_binds=0 will be the right choice.
Quoted 79 lines
Quoted 78 lines
Signed-off-by Matthew Brost <[email protected]>Reviewed-by Thomas Hellström <[email protected]>--- drivers/gpu/drm/xe/xe_vm.c | 30 ++++++++++++++++++------------ 1 file changed, 18 insertions(+), 12 deletions(-)drivers/gpu/drm/xe/xe_vm.c +18
-12@@ -2850,7 +2850,6 @@ static int vm_bind_ioctl_check_args(struct xe_device *xe,int i;if (XE_IOCTL_DBG (xe, args->extensions) || -XE_IOCTL_DBG (xe, !args->num_binds) ||XE_IOCTL_DBG (xe, args->num_binds >MAX_BINDS ))return -EINVAL ;@@ -2977,7 +2976,7 @@ int xe_vm_bind_ioctl(struct drm_device *dev, void *data, struct drm_file *file) goto put_exec_queue; } -if (XE_IOCTL_DBG (xe, async != +if (XE_IOCTL_DBG (xe, args->num_binds && async != !!(q->flags &EXEC_QUEUE_FLAG_VM_ASYNC ))) { err = -EINVAL ; goto put_exec_queue;@@ -2991,7 +2990,7 @@ int xe_vm_bind_ioctl(struct drm_device *dev, void *data, struct drm_file *file) }if (!args->exec_queue_id) { -if (XE_IOCTL_DBG (xe, async != +if (XE_IOCTL_DBG (xe, args->num_binds && async != !!(vm->flags &XE_VM_FLAG_ASYNC_DEFAULT ))) { err = -EINVAL ; goto put_vm;@@ -3028,16 +3027,18 @@ int xe_vm_bind_ioctl(struct drm_device *dev, void *data, struct drm_file *file) } } - bos =kzalloc (sizeof (*bos) * args->num_binds,GFP_KERNEL ); -if (!bos) { - err = -ENOMEM ; - goto release_vm_lock; - } +if (args->num_binds) { + bos =kzalloc (sizeof (*bos) * args->num_binds,GFP_KERNEL ); +if (!bos) { + err = -ENOMEM ; + goto release_vm_lock; + } - ops =kzalloc (sizeof (*ops) * args->num_binds,GFP_KERNEL ); -if (!ops) { - err = -ENOMEM ; - goto release_vm_lock; + ops =kzalloc (sizeof (*ops) * args->num_binds,GFP_KERNEL ); +if (!ops) { + err = -ENOMEM ; + goto release_vm_lock; + } }for (i =0 ; i < args->num_binds; ++i) {@@ -3092,6 +3093,11 @@ int xe_vm_bind_ioctl(struct drm_device *dev, void *data, struct drm_file *file) goto free_syncs; } +if (!args->num_binds) { + err = -ENODATA ; + goto free_syncs; + } +for (i =0 ; i < args->num_binds; ++i) {u64 range = bind_ops[i].range;u64 addr = bind_ops[i].addr;
Matthew Brost
maintainerRe: [Intel-xe] [PATCH 1/5] drm/xe: Allow num_binds == 0 in VM bind IOCTL
Quoted 13 lines · Paulo R
Quoted 11 lines · Jose
Quoted 3 lines · Matthew Brost
The idea being out-syncs can signal indicating all previous operations on the bind queue are complete. An example use case of this would be support for implementing vkQueueWaitForIdle easily.
s/vkQueueWaitForIdle/vkQueueWaitIdle
So this is just for the bind queue.
Not sure if it is useful, Mesa will eventually move to async vm bind and we will use a singleTIMELINE_SYNCOBJto track what bind operations are completed.
Paulo do you see any use with sparse?
Quoted 5 lines
We don't have plans to submit stuff to the Kernel with num_binds=0, but we do need the Kernel to support the case where we're trying to bind something that's already bound (so there's nothing for the Kernel to do, but
num_binds> 0).
This series fixes an issue with num_binds > 0 but nothing to bind too.
Quoted 4 lines
And yes to track operations that are already complete it's usually better to just check the syncobj (timeline or not) of the last thing you submitted.
Quoted 8 lines
Currently, due to the xe.ko bugs, our
vm_bindcode is completely synchronous, but the plan is to move it to async like we're doing with TR-TT.I'm not discarding the possibility that we may have to do something extra for vkQueueWaitIdle, especially after we make it async, but I don't think submitting num_binds=0 will be the right choice.
I'd like leave this option in the kernel as it is pretty easy to implement and matches the behavior for execs with num batches == 0.
Matt
Quoted 80 lines
Quoted 79 lines
Quoted 78 lines
Signed-off-by Matthew Brost <[email protected]>Reviewed-by Thomas Hellström <[email protected]>--- drivers/gpu/drm/xe/xe_vm.c | 30 ++++++++++++++++++------------ 1 file changed, 18 insertions(+), 12 deletions(-)drivers/gpu/drm/xe/xe_vm.c +18
-12@@ -2850,7 +2850,6 @@ static int vm_bind_ioctl_check_args(struct xe_device *xe,int i;if (XE_IOCTL_DBG (xe, args->extensions) || -XE_IOCTL_DBG (xe, !args->num_binds) ||XE_IOCTL_DBG (xe, args->num_binds >MAX_BINDS ))return -EINVAL ;@@ -2977,7 +2976,7 @@ int xe_vm_bind_ioctl(struct drm_device *dev, void *data, struct drm_file *file) goto put_exec_queue; } -if (XE_IOCTL_DBG (xe, async != +if (XE_IOCTL_DBG (xe, args->num_binds && async != !!(q->flags &EXEC_QUEUE_FLAG_VM_ASYNC ))) { err = -EINVAL ; goto put_exec_queue;@@ -2991,7 +2990,7 @@ int xe_vm_bind_ioctl(struct drm_device *dev, void *data, struct drm_file *file) }if (!args->exec_queue_id) { -if (XE_IOCTL_DBG (xe, async != +if (XE_IOCTL_DBG (xe, args->num_binds && async != !!(vm->flags &XE_VM_FLAG_ASYNC_DEFAULT ))) { err = -EINVAL ; goto put_vm;@@ -3028,16 +3027,18 @@ int xe_vm_bind_ioctl(struct drm_device *dev, void *data, struct drm_file *file) } } - bos =kzalloc (sizeof (*bos) * args->num_binds,GFP_KERNEL ); -if (!bos) { - err = -ENOMEM ; - goto release_vm_lock; - } +if (args->num_binds) { + bos =kzalloc (sizeof (*bos) * args->num_binds,GFP_KERNEL ); +if (!bos) { + err = -ENOMEM ; + goto release_vm_lock; + } - ops =kzalloc (sizeof (*ops) * args->num_binds,GFP_KERNEL ); -if (!ops) { - err = -ENOMEM ; - goto release_vm_lock; + ops =kzalloc (sizeof (*ops) * args->num_binds,GFP_KERNEL ); +if (!ops) { + err = -ENOMEM ; + goto release_vm_lock; + } }for (i =0 ; i < args->num_binds; ++i) {@@ -3092,6 +3093,11 @@ int xe_vm_bind_ioctl(struct drm_device *dev, void *data, struct drm_file *file) goto free_syncs; } +if (!args->num_binds) { + err = -ENODATA ; + goto free_syncs; + } +for (i =0 ; i < args->num_binds; ++i) {u64 range = bind_ops[i].range;u64 addr = bind_ops[i].addr;
Matthew Brost
maintainerPATCH 2/5 drm/xe: Use a flags field instead of bools for VMA create
In reply to Matthew BrostUse a flags field instead of severval bools for VMA create as it is easier to read and less bug prone.
--- drivers/gpu/drm/xe/xe_vm.c | 58 ++++++++++++++++++++------------------ 1 file changed, 30 insertions(+), 28 deletions(-)
drivers/gpu/drm/xe/xe_vm.c +30 -28
@@ -858,17 +858,21 @@ struct dma_fence *xe_vm_rebind(struct xe_vm *vm, bool rebind_worker)return fence; } +#define VMA_CREATE_FLAG_READ_ONLY BIT(0) +#define VMA_CREATE_FLAG_IS_NULL BIT(1) +static struct xe_vma *xe_vma_create (struct xe_vm *vm,struct xe_bo *bo,u64 bo_offset_or_userptr,u64 start,u64 end, -bool read_only, -bool is_null, -u8 tile_mask) +u8 tile_mask, +unsigned int flags) {struct xe_vma *vma;struct xe_tile *tile;u8 id; +bool read_only = (flags &VMA_CREATE_FLAG_READ_ONLY ); +bool is_null = (flags &VMA_CREATE_FLAG_IS_NULL );xe_assert (vm->xe, start < end);xe_assert (vm->xe, end < vm->size);
@@ -2263,7 +2267,7 @@ vm_bind_ioctl_ops_create(struct xe_vm *vm, struct xe_bo *bo, }static struct xe_vma *new_vma (struct xe_vm *vm,struct drm_gpuva_op_map *op, -u8 tile_mask,bool read_only,bool is_null) +u8 tile_mask,unsigned int flags) {struct xe_bo *bo = op->gem.obj ?gem_to_xe_bo (op->gem.obj) :NULL ;struct xe_vma *vma;
@@ -2278,8 +2282,7 @@ static struct xe_vma *new_vma(struct xe_vm *vm, struct drm_gpuva_op_map *op, } vma =xe_vma_create (vm, bo, op->gem.offset, op->va.addr, op->va.addr + - op->va.range -1 , read_only, is_null, - tile_mask); + op->va.range -1 , tile_mask, flags);if (bo)xe_bo_unlock (bo);
@@ -2405,7 +2408,9 @@ static int vm_bind_ioctl_ops_parse(struct xe_vm *vm, struct xe_exec_queue *q,drm_gpuva_for_each_op (__op, ops) {struct xe_vma_op *op =gpuva_op_to_vma_op (__op); +struct xe_vma *vma;bool first =list_empty (ops_list); +unsigned int flags =0 ;INIT_LIST_HEAD (&op->link);list_add_tail (&op->link, ops_list);
@@ -2421,11 +2426,13 @@ static int vm_bind_ioctl_ops_parse(struct xe_vm *vm, struct xe_exec_queue *q,switch (op->base.op) {case DRM_GPUVA_OP_MAP : { -struct xe_vma *vma; + flags |= op->map.read_only ? +VMA_CREATE_FLAG_READ_ONLY :0 ; + flags |= op->map.is_null ? +VMA_CREATE_FLAG_IS_NULL :0 ; vma =new_vma (vm, &op->base.map, - op->tile_mask, op->map.read_only, - op->map.is_null); + op->tile_mask, flags);if (IS_ERR (vma))return PTR_ERR (vma);
@@ -2441,17 +2448,15 @@ static int vm_bind_ioctl_ops_parse(struct xe_vm *vm, struct xe_exec_queue *q, op->remap.range =xe_vma_size (old);if (op->base.remap.prev) { -struct xe_vma *vma; -bool read_only = - op->base.remap.unmap->va->flags & -XE_VMA_READ_ONLY ; -bool is_null = - op->base.remap.unmap->va->flags & -DRM_GPUVA_SPARSE ; + flags |= op->base.remap.unmap->va->flags & +XE_VMA_READ_ONLY ? +VMA_CREATE_FLAG_READ_ONLY :0 ; + flags |= op->base.remap.unmap->va->flags & +DRM_GPUVA_SPARSE ? +VMA_CREATE_FLAG_IS_NULL :0 ; vma =new_vma (vm, op->base.remap.prev, - op->tile_mask, read_only, - is_null); + op->tile_mask, flags);if (IS_ERR (vma))return PTR_ERR (vma);
@@ -2474,18 +2479,15 @@ static int vm_bind_ioctl_ops_parse(struct xe_vm *vm, struct xe_exec_queue *q, }if (op->base.remap.next) { -struct xe_vma *vma; -bool read_only = - op->base.remap.unmap->va->flags & -XE_VMA_READ_ONLY ; - -bool is_null = - op->base.remap.unmap->va->flags & -DRM_GPUVA_SPARSE ; + flags |= op->base.remap.unmap->va->flags & +XE_VMA_READ_ONLY ? +VMA_CREATE_FLAG_READ_ONLY :0 ; + flags |= op->base.remap.unmap->va->flags & +DRM_GPUVA_SPARSE ? +VMA_CREATE_FLAG_IS_NULL :0 ; vma =new_vma (vm, op->base.remap.next, - op->tile_mask, read_only, - is_null); + op->tile_mask, flags);if (IS_ERR (vma))return PTR_ERR (vma);
Signature · 1 line
2.34.1
Re: [Intel-xe] [PATCH 2/5] drm/xe: Use a flags field instead of bools for VMA create
Quoted 5 lines · Matthew Brost
Use a flags field instead of severval bools for VMA create as it is easier to read and less bug prone.
Suggested-by Thomas Hellström <[email protected]>Signed-off-by Matthew Brost <[email protected]>
Hmm.
If you feel that this was not really an improvement feel free to disregard. Anyway,
Quoted 130 lines
--- drivers/gpu/drm/xe/xe_vm.c | 58 ++++++++++++++++++++------------------ 1 file changed, 30 insertions(+), 28 deletions(-)drivers/gpu/drm/xe/xe_vm.c +30
-28@@ -858,17 +858,21 @@ struct dma_fence *xe_vm_rebind(struct xe_vm *vm, bool rebind_worker)return fence; } +#define VMA_CREATE_FLAG_READ_ONLY BIT(0) +#define VMA_CREATE_FLAG_IS_NULL BIT(1) +static struct xe_vma *xe_vma_create (struct xe_vm *vm,struct xe_bo *bo,u64 bo_offset_or_userptr,u64 start,u64 end, -bool read_only, -bool is_null, -u8 tile_mask) +u8 tile_mask, +unsigned int flags) {struct xe_vma *vma;struct xe_tile *tile;u8 id; +bool read_only = (flags &VMA_CREATE_FLAG_READ_ONLY ); +bool is_null = (flags &VMA_CREATE_FLAG_IS_NULL );xe_assert (vm->xe, start < end);xe_assert (vm->xe, end < vm->size);@@ -2263,7 +2267,7 @@ vm_bind_ioctl_ops_create(struct xe_vm *vm, struct xe_bo *bo, }static struct xe_vma *new_vma (struct xe_vm *vm,struct drm_gpuva_op_map *op, -u8 tile_mask,bool read_only,bool is_null) +u8 tile_mask,unsigned int flags) {struct xe_bo *bo = op->gem.obj ?gem_to_xe_bo (op->gem.obj) :NULL ;struct xe_vma *vma;@@ -2278,8 +2282,7 @@ static struct xe_vma *new_vma(struct xe_vm *vm, struct drm_gpuva_op_map *op, } vma =xe_vma_create (vm, bo, op->gem.offset, op->va.addr, op->va.addr + - op->va.range -1 , read_only, is_null, - tile_mask); + op->va.range -1 , tile_mask, flags);if (bo)xe_bo_unlock (bo);@@ -2405,7 +2408,9 @@ static int vm_bind_ioctl_ops_parse(struct xe_vm *vm, struct xe_exec_queue *q,drm_gpuva_for_each_op (__op, ops) {struct xe_vma_op *op =gpuva_op_to_vma_op (__op); +struct xe_vma *vma;bool first =list_empty (ops_list); +unsigned int flags =0 ;INIT_LIST_HEAD (&op->link);list_add_tail (&op->link, ops_list);@@ -2421,11 +2426,13 @@ static int vm_bind_ioctl_ops_parse(struct xe_vm *vm, struct xe_exec_queue *q,switch (op->base.op) {case DRM_GPUVA_OP_MAP : { -struct xe_vma *vma; + flags |= op->map.read_only ? +VMA_CREATE_FLAG_READ_ONLY :0 ; + flags |= op->map.is_null ? +VMA_CREATE_FLAG_IS_NULL :0 ; vma =new_vma (vm, &op->base.map, - op->tile_mask, op->map.read_only, - op->map.is_null); + op->tile_mask, flags);if (IS_ERR (vma))return PTR_ERR (vma);@@ -2441,17 +2448,15 @@ static int vm_bind_ioctl_ops_parse(struct xe_vm *vm, struct xe_exec_queue *q, op->remap.range =xe_vma_size (old);if (op->base.remap.prev) { -struct xe_vma *vma; -bool read_only = - op->base.remap.unmap->va->flags & -XE_VMA_READ_ONLY ; -bool is_null = - op->base.remap.unmap->va->flags & -DRM_GPUVA_SPARSE ; + flags |= op->base.remap.unmap->va->flags & +XE_VMA_READ_ONLY ? +VMA_CREATE_FLAG_READ_ONLY :0 ; + flags |= op->base.remap.unmap->va->flags & +DRM_GPUVA_SPARSE ? +VMA_CREATE_FLAG_IS_NULL :0 ; vma =new_vma (vm, op->base.remap.prev, - op->tile_mask, read_only, - is_null); + op->tile_mask, flags);if (IS_ERR (vma))return PTR_ERR (vma);@@ -2474,18 +2479,15 @@ static int vm_bind_ioctl_ops_parse(struct xe_vm *vm, struct xe_exec_queue *q, }if (op->base.remap.next) { -struct xe_vma *vma; -bool read_only = - op->base.remap.unmap->va->flags & -XE_VMA_READ_ONLY ; - -bool is_null = - op->base.remap.unmap->va->flags & -DRM_GPUVA_SPARSE ; + flags |= op->base.remap.unmap->va->flags & +XE_VMA_READ_ONLY ? +VMA_CREATE_FLAG_READ_ONLY :0 ; + flags |= op->base.remap.unmap->va->flags & +DRM_GPUVA_SPARSE ? +VMA_CREATE_FLAG_IS_NULL :0 ; vma =new_vma (vm, op->base.remap.next, - op->tile_mask, read_only, - is_null); + op->tile_mask, flags);if (IS_ERR (vma))return PTR_ERR (vma);
Matthew Brost
maintainerPATCH 3/5 drm/xe: Use a flags field instead of bools for sync parse
In reply to Matthew BrostUse a flags field instead of severval bools for sync parse as it is easier to read and less bug prone.
--- drivers/gpu/drm/xe/xe_exec.c | 5 +++-- drivers/gpu/drm/xe/xe_sync.c | 8 +++++--- drivers/gpu/drm/xe/xe_vm.c | 5 +++-- 3 files changed, 11 insertions(+), 7 deletions(-)
drivers/gpu/drm/xe/xe_exec.c +3 -2
@@ -181,8 +181,9 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file)for (i =0 ; i < args->num_syncs; i++) { err =xe_sync_entry_parse (xe, xef, &syncs[num_syncs++], - &syncs_user[i], true, -xe_vm_no_dma_fences (vm)); + &syncs_user[i],SYNC_PARSE_FLAG_EXEC | + (xe_vm_no_dma_fences (vm) ? +SYNC_PARSE_FLAG_LR_MODE :0 ));if (err) goto err_syncs; }
drivers/gpu/drm/xe/xe_sync.c +5 -3
@@ -100,10 +100,12 @@ static void user_fence_cb(struct dma_fence *fence, struct dma_fence_cb *cb)int xe_sync_entry_parse (struct xe_device *xe,struct xe_file *xef,struct xe_sync_entry *sync,struct drm_xe_sync __user *sync_user, -bool exec,bool no_dma_fences) +unsigned int flags) {struct drm_xe_sync sync_in;int err; +bool exec = flags &SYNC_PARSE_FLAG_EXEC ; +bool lr_mode = flags &SYNC_PARSE_FLAG_LR_MODE ;bool signal;if (copy_from_user (&sync_in, sync_user,sizeof (*sync_user)))
@@ -118,7 +120,7 @@ int xe_sync_entry_parse(struct xe_device *xe, struct xe_file *xef, signal = sync_in.flags &DRM_XE_SYNC_SIGNAL ;switch (sync_in.flags &SYNC_FLAGS_TYPE_MASK ) {case DRM_XE_SYNC_SYNCOBJ : -if (XE_IOCTL_DBG (xe, no_dma_fences && signal)) +if (XE_IOCTL_DBG (xe, lr_mode && signal))return -EOPNOTSUPP ;if (XE_IOCTL_DBG (xe,upper_32_bits (sync_in.addr)))
@@ -136,7 +138,7 @@ int xe_sync_entry_parse(struct xe_device *xe, struct xe_file *xef,break ;case DRM_XE_SYNC_TIMELINE_SYNCOBJ : -if (XE_IOCTL_DBG (xe, no_dma_fences && signal)) +if (XE_IOCTL_DBG (xe, lr_mode && signal))return -EOPNOTSUPP ;if (XE_IOCTL_DBG (xe,upper_32_bits (sync_in.addr)))
drivers/gpu/drm/xe/xe_vm.c +3 -2
@@ -3089,8 +3089,9 @@ int xe_vm_bind_ioctl(struct drm_device *dev, void *data, struct drm_file *file) syncs_user =u64_to_user_ptr (args->syncs);for (num_syncs =0 ; num_syncs < args->num_syncs; num_syncs++) { err =xe_sync_entry_parse (xe, xef, &syncs[num_syncs], - &syncs_user[num_syncs], false, -xe_vm_no_dma_fences (vm)); + &syncs_user[num_syncs], +xe_vm_no_dma_fences (vm) ? +SYNC_PARSE_FLAG_LR_MODE :0 );if (err) goto free_syncs; }
Signature · 1 line
2.34.1
Re: [Intel-xe] [PATCH 3/5] drm/xe: Use a flags field instead of bools for sync parse
Quoted 5 lines · Matthew Brost
Use a flags field instead of severval bools for sync parse as it is easier to read and less bug prone.
Suggested-by Thomas Hellström <[email protected]>Signed-off-by Matthew Brost <[email protected]>
Quoted 74 lines
--- drivers/gpu/drm/xe/xe_exec.c | 5 +++-- drivers/gpu/drm/xe/xe_sync.c | 8 +++++--- drivers/gpu/drm/xe/xe_vm.c | 5 +++-- 3 files changed, 11 insertions(+), 7 deletions(-)drivers/gpu/drm/xe/xe_exec.c +3
-2@@ -181,8 +181,9 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file)for (i =0 ; i < args->num_syncs; i++) { err =xe_sync_entry_parse (xe, xef, &syncs[num_syncs++], - &syncs_user[i], true, -xe_vm_no_dma_fences (vm)); + &syncs_user[i],SYNC_PARSE_FLAG_EXEC | + (xe_vm_no_dma_fences (vm) ? +SYNC_PARSE_FLAG_LR_MODE :0 ));if (err) goto err_syncs; }drivers/gpu/drm/xe/xe_sync.c +5
-3@@ -100,10 +100,12 @@ static void user_fence_cb(struct dma_fence *fence, struct dma_fence_cb *cb)int xe_sync_entry_parse (struct xe_device *xe,struct xe_file *xef,struct xe_sync_entry *sync,struct drm_xe_sync __user *sync_user, -bool exec,bool no_dma_fences) +unsigned int flags) {struct drm_xe_sync sync_in;int err; +bool exec = flags &SYNC_PARSE_FLAG_EXEC ; +bool lr_mode = flags &SYNC_PARSE_FLAG_LR_MODE ;bool signal;if (copy_from_user (&sync_in, sync_user,sizeof (*sync_user)))@@ -118,7 +120,7 @@ int xe_sync_entry_parse(struct xe_device *xe, struct xe_file *xef, signal = sync_in.flags &DRM_XE_SYNC_SIGNAL ;switch (sync_in.flags &SYNC_FLAGS_TYPE_MASK ) {case DRM_XE_SYNC_SYNCOBJ : -if (XE_IOCTL_DBG (xe, no_dma_fences && signal)) +if (XE_IOCTL_DBG (xe, lr_mode && signal))return -EOPNOTSUPP ;if (XE_IOCTL_DBG (xe,upper_32_bits (sync_in.addr)))@@ -136,7 +138,7 @@ int xe_sync_entry_parse(struct xe_device *xe, struct xe_file *xef,break ;case DRM_XE_SYNC_TIMELINE_SYNCOBJ : -if (XE_IOCTL_DBG (xe, no_dma_fences && signal)) +if (XE_IOCTL_DBG (xe, lr_mode && signal))return -EOPNOTSUPP ;if (XE_IOCTL_DBG (xe,upper_32_bits (sync_in.addr)))drivers/gpu/drm/xe/xe_vm.c +3
-2@@ -3089,8 +3089,9 @@ int xe_vm_bind_ioctl(struct drm_device *dev, void *data, struct drm_file *file) syncs_user =u64_to_user_ptr (args->syncs);for (num_syncs =0 ; num_syncs < args->num_syncs; num_syncs++) { err =xe_sync_entry_parse (xe, xef, &syncs[num_syncs], - &syncs_user[num_syncs], false, -xe_vm_no_dma_fences (vm)); + &syncs_user[num_syncs], +xe_vm_no_dma_fences (vm) ? +SYNC_PARSE_FLAG_LR_MODE :0 );if (err) goto free_syncs; }
Thomas Hellström
maintainerRe: [Intel-xe] [PATCH 3/5] drm/xe: Use a flags field instead of bools for sync parse
Quoted 10 lines · Thomas Hellström
Quoted 5 lines · Matthew Brost
Use a flags field instead of severval bools for sync parse as it is easier to read and less bug prone.
Suggested-by Thomas Hellström <[email protected]>Signed-off-by Matthew Brost <[email protected]>Reviewed-by Thomas Hellström <[email protected]>
Hmm. It looks, though, that the header update slipped into the next patch.
/Thomas
Quoted 75 lines
Quoted 75 lines
---
drivers/gpu/drm/xe/xe_exec.c| 5 +++--
drivers/gpu/drm/xe/xe_sync.c| 8 +++++---
drivers/gpu/drm/xe/xe_vm.c | 5 +++--3 files changed, 11 insertions(+), 7 deletions(-)drivers/gpu/drm/xe/xe_exec.c +3
-2@@ -181,8 +181,9 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file) for (i = 0; i < args->num_syncs; i++) { err = xe_sync_entry_parse(xe, xef, &syncs[num_syncs++], - &syncs_user[i], true, - xe_vm_no_dma_fences(vm)); + &syncs_user[i], SYNC_PARSE_FLAG_EXEC | + (xe_vm_no_dma_fences(vm) ? + SYNC_PARSE_FLAG_LR_MODE : 0)); if (err) goto err_syncs; }drivers/gpu/drm/xe/xe_sync.c +3
-1@@ -100,10 +100,12 @@ static void user_fence_cb(struct dma_fence *fence, struct dma_fence_cb *cb) int xe_sync_entry_parse(struct xe_device *xe, struct xe_file *xef, struct xe_sync_entry *sync, struct drm_xe_sync __user *sync_user, - bool exec, bool no_dma_fences) + unsigned int flags) { struct drm_xe_sync sync_in; int err; + bool exec = flags & SYNC_PARSE_FLAG_EXEC; + bool lr_mode = flags & SYNC_PARSE_FLAG_LR_MODE; bool signal; if (copy_from_user(&sync_in, sync_user, sizeof(*sync_user)))@@ -118,7 +120,7 @@ int xe_sync_entry_parse(struct xe_device *xe,struct xe_file *xef, signal = sync_in.flags & DRM_XE_SYNC_SIGNAL; switch (sync_in.flags & SYNC_FLAGS_TYPE_MASK) {case
DRM_XE_SYNC_SYNCOBJ:
- if (XE_IOCTL_DBG(xe, no_dma_fences && signal)) + if (XE_IOCTL_DBG(xe, lr_mode && signal)) return-EOPNOTSUPP;
if (XE_IOCTL_DBG(xe, upper_32_bits(sync_in.addr)))drivers/gpu/drm/xe/xe_sync.c +0
-0@@ -136,7 +138,7 @@ int xe_sync_entry_parse(struct xe_device *xe, struct xe_file *xef, break;case
DRM_XE_SYNC_TIMELINE_SYNCOBJ:
- if (XE_IOCTL_DBG(xe, no_dma_fences && signal)) + if (XE_IOCTL_DBG(xe, lr_mode && signal)) return-EOPNOTSUPP;
if (XE_IOCTL_DBG(xe, upper_32_bits(sync_in.addr)))drivers/gpu/drm/xe/xe_vm.c +3
-2@@ -3089,8 +3089,9 @@ int xe_vm_bind_ioctl(struct drm_device *dev, void *data, struct drm_file *file) syncs_user = u64_to_user_ptr(args->syncs); for (num_syncs = 0; num_syncs < args->num_syncs; num_syncs++) { err = xe_sync_entry_parse(xe, xef, &syncs[num_syncs], - &syncs_user[num_syncs], false, - xe_vm_no_dma_fences(vm)); + &syncs_user[num_syncs], + xe_vm_no_dma_fences(vm) ? + SYNC_PARSE_FLAG_LR_MODE : 0); if (err) goto free_syncs;}
Matthew Brost
maintainerPATCH 4/5 rm/xe: Allow num_batch_buffer == 0 in exec IOCTL
In reply to Matthew BrostThe idea being out-syncs can signal indicating all previous operations on the exec queue are complete. An example use case of this would be support for implementing vkQueueWaitForIdle easily.
v2: Don't add last_fence for VM's that do not support dma fences
v3: Use a flags field instead of severval bools in sync parse (Thomas)
--- drivers/gpu/drm/xe/xe_exec.c | 23 ++++++++++++++++++++--- drivers/gpu/drm/xe/xe_exec_queue.c | 5 ++++- drivers/gpu/drm/xe/xe_exec_queue_types.h | 5 +++-- drivers/gpu/drm/xe/xe_sync.c | 4 ++++ drivers/gpu/drm/xe/xe_sync.h | 6 +++++- 5 files changed, 36 insertions(+), 7 deletions(-)
drivers/gpu/drm/xe/xe_exec.c +20 -3
@@ -161,7 +161,8 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file)if (XE_IOCTL_DBG (xe, q->flags &EXEC_QUEUE_FLAG_VM ))return -EINVAL ; -if (XE_IOCTL_DBG (xe, q->width != args->num_batch_buffer)) +if (XE_IOCTL_DBG (xe, args->num_batch_buffer && + q->width != args->num_batch_buffer))return -EINVAL ;if (XE_IOCTL_DBG (xe, q->flags &EXEC_QUEUE_FLAG_BANNED )) {
@@ -183,12 +184,14 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file) err =xe_sync_entry_parse (xe, xef, &syncs[num_syncs++], &syncs_user[i],SYNC_PARSE_FLAG_EXEC | (xe_vm_no_dma_fences (vm) ? -SYNC_PARSE_FLAG_LR_MODE :0 )); +SYNC_PARSE_FLAG_LR_MODE :0 ) | + (!args->num_batch_buffer ? +SYNC_PARSE_FLAG_DISALLOW_USER_FENCE :0 ));if (err) goto err_syncs; } -if (xe_exec_queue_is_parallel (q)) { +if (args->num_batch_buffer &&xe_exec_queue_is_parallel (q)) { err =__copy_from_user (addresses, addresses_user,sizeof (u64) * q->width);if (err) {
@@ -235,6 +238,18 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file) goto err_exec; } +if (!args->num_batch_buffer) { +if (!xe_vm_no_dma_fences (vm)) { +struct dma_fence *fence = +xe_exec_queue_last_fence_get (q, vm); + +for (i =0 ; i < num_syncs; i++) +xe_sync_entry_signal (&syncs[i],NULL , fence); + } + + goto err_exec; + } +if (xe_exec_queue_is_lr (q) &&xe_exec_queue_ring_full (q)) { err = -EWOULDBLOCK ; goto err_exec;
@@ -328,6 +343,8 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file)if (xe_exec_queue_is_lr (q)) q->ring_ops->emit_job (job); +if (!xe_vm_no_dma_fences (vm)) +xe_exec_queue_last_fence_set (q, vm, &job->drm.s_fence->finished);xe_sched_job_push (job);xe_vm_reactivate_rebind (vm);
drivers/gpu/drm/xe/xe_exec_queue.c +4 -1
@@ -924,7 +924,10 @@ int xe_exec_queue_set_property_ioctl(struct drm_device *dev, void *data,static void xe_exec_queue_last_fence_lockdep_assert (struct xe_exec_queue *q,struct xe_vm *vm) { -lockdep_assert_held_write (&vm->lock); +if (q->flags &EXEC_QUEUE_FLAG_VM ) +lockdep_assert_held_write (&vm->lock); +else +xe_vm_assert_held (vm); }/**
drivers/gpu/drm/xe/xe_exec_queue_types.h +3 -2
@@ -56,8 +56,9 @@ struct xe_exec_queue {struct xe_hw_fence_irq *fence_irq;/** -* @last_fence: last fence on engine, protected by vm->lock in write -* mode if bind engine +* @last_fence: last fence on exec queue, protected by vm->lock in write +* mode if bind exec queue, protected by dma resv lock if non-bind exec +* queue */ struct dma_fence *last_fence;
drivers/gpu/drm/xe/xe_sync.c +4 -0
@@ -106,6 +106,7 @@ int xe_sync_entry_parse(struct xe_device *xe, struct xe_file *xef,int err;bool exec = flags &SYNC_PARSE_FLAG_EXEC ;bool lr_mode = flags &SYNC_PARSE_FLAG_LR_MODE ; +bool disallow_user_fence = flags &SYNC_PARSE_FLAG_DISALLOW_USER_FENCE ;bool signal;if (copy_from_user (&sync_in, sync_user,sizeof (*sync_user)))
@@ -173,6 +174,9 @@ int xe_sync_entry_parse(struct xe_device *xe, struct xe_file *xef,break ;case DRM_XE_SYNC_USER_FENCE : +if (XE_IOCTL_DBG (xe, disallow_user_fence)) +return -EOPNOTSUPP ; +if (XE_IOCTL_DBG (xe, !signal))return -EOPNOTSUPP ;
drivers/gpu/drm/xe/xe_sync.h +5 -1
@@ -12,10 +12,14 @@ struct xe_device;struct xe_file ;struct xe_sched_job ; +#define SYNC_PARSE_FLAG_EXEC BIT(0) +#define SYNC_PARSE_FLAG_LR_MODE BIT(1) +#define SYNC_PARSE_FLAG_DISALLOW_USER_FENCE BIT(2) +int xe_sync_entry_parse (struct xe_device *xe,struct xe_file *xef,struct xe_sync_entry *sync,struct drm_xe_sync __user *sync_user, -bool exec,bool compute_mode); +unsigned int flags);int xe_sync_entry_wait (struct xe_sync_entry *sync);int xe_sync_entry_add_deps (struct xe_sync_entry *sync,struct xe_sched_job *job);
Signature · 1 line
2.34.1
Re: [Intel-xe] [PATCH 4/5] rm/xe: Allow num_batch_buffer == 0 in exec IOCTL
Quoted 3 lines · Matthew Brost
The idea being out-syncs can signal indicating all previous operations on the exec queue are complete. An example use case of this would be support for implementing vkQueueWaitForIdle easily.
vkQueueWaitIdle
That is useful for Mesa.
Quoted 146 lines
v2: Don't add
last_fencefor VM's that do not support dma fences
v3: Use a flags field instead of severval bools in sync parse (Thomas)Signed-off-by Matthew Brost <[email protected]>--- drivers/gpu/drm/xe/xe_exec.c | 23 ++++++++++++++++++++--- drivers/gpu/drm/xe/xe_exec_queue.c | 5 ++++- drivers/gpu/drm/xe/xe_exec_queue_types.h | 5 +++-- drivers/gpu/drm/xe/xe_sync.c | 4 ++++ drivers/gpu/drm/xe/xe_sync.h | 6 +++++- 5 files changed, 36 insertions(+), 7 deletions(-)drivers/gpu/drm/xe/xe_exec.c +20
-3@@ -161,7 +161,8 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file)if (XE_IOCTL_DBG (xe, q->flags &EXEC_QUEUE_FLAG_VM ))return -EINVAL ; -if (XE_IOCTL_DBG (xe, q->width != args->num_batch_buffer)) +if (XE_IOCTL_DBG (xe, args->num_batch_buffer && + q->width != args->num_batch_buffer))return -EINVAL ;if (XE_IOCTL_DBG (xe, q->flags &EXEC_QUEUE_FLAG_BANNED )) {@@ -183,12 +184,14 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file) err =xe_sync_entry_parse (xe, xef, &syncs[num_syncs++], &syncs_user[i],SYNC_PARSE_FLAG_EXEC | (xe_vm_no_dma_fences (vm) ? -SYNC_PARSE_FLAG_LR_MODE :0 )); +SYNC_PARSE_FLAG_LR_MODE :0 ) | + (!args->num_batch_buffer ? +SYNC_PARSE_FLAG_DISALLOW_USER_FENCE :0 ));if (err) goto err_syncs; } -if (xe_exec_queue_is_parallel (q)) { +if (args->num_batch_buffer &&xe_exec_queue_is_parallel (q)) { err =__copy_from_user (addresses, addresses_user,sizeof (u64) * q->width);if (err) {@@ -235,6 +238,18 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file) goto err_exec; } +if (!args->num_batch_buffer) { +if (!xe_vm_no_dma_fences (vm)) { +struct dma_fence *fence = +xe_exec_queue_last_fence_get (q, vm); + +for (i =0 ; i < num_syncs; i++) +xe_sync_entry_signal (&syncs[i],NULL , fence); + } + + goto err_exec; + } +if (xe_exec_queue_is_lr (q) &&xe_exec_queue_ring_full (q)) { err = -EWOULDBLOCK ; goto err_exec;@@ -328,6 +343,8 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file)if (xe_exec_queue_is_lr (q)) q->ring_ops->emit_job (job); +if (!xe_vm_no_dma_fences (vm)) +xe_exec_queue_last_fence_set (q, vm, &job->drm.s_fence->finished);xe_sched_job_push (job);xe_vm_reactivate_rebind (vm);drivers/gpu/drm/xe/xe_exec_queue.c +4
-1@@ -924,7 +924,10 @@ int xe_exec_queue_set_property_ioctl(struct drm_device *dev, void *data,static void xe_exec_queue_last_fence_lockdep_assert (struct xe_exec_queue *q,struct xe_vm *vm) { -lockdep_assert_held_write (&vm->lock); +if (q->flags &EXEC_QUEUE_FLAG_VM ) +lockdep_assert_held_write (&vm->lock); +else +xe_vm_assert_held (vm); }/** drivers/gpu/drm/xe/xe_exec_queue_types.h +3
-2@@ -56,8 +56,9 @@ struct xe_exec_queue {struct xe_hw_fence_irq *fence_irq;/** -* @last_fence: last fence on engine, protected by vm->lock in write -* mode if bind engine +* @last_fence: last fence on exec queue, protected by vm->lock in write +* mode if bind exec queue, protected by dma resv lock if non-bind exec +* queue */ struct dma_fence *last_fence;drivers/gpu/drm/xe/xe_sync.c +4
-0@@ -106,6 +106,7 @@ int xe_sync_entry_parse(struct xe_device *xe, struct xe_file *xef,int err;bool exec = flags &SYNC_PARSE_FLAG_EXEC ;bool lr_mode = flags &SYNC_PARSE_FLAG_LR_MODE ; +bool disallow_user_fence = flags &SYNC_PARSE_FLAG_DISALLOW_USER_FENCE ;bool signal;if (copy_from_user (&sync_in, sync_user,sizeof (*sync_user)))@@ -173,6 +174,9 @@ int xe_sync_entry_parse(struct xe_device *xe, struct xe_file *xef,break ;case DRM_XE_SYNC_USER_FENCE : +if (XE_IOCTL_DBG (xe, disallow_user_fence)) +return -EOPNOTSUPP ; +if (XE_IOCTL_DBG (xe, !signal))return -EOPNOTSUPP ;drivers/gpu/drm/xe/xe_sync.h +5
-1@@ -12,10 +12,14 @@ struct xe_device;struct xe_file ;struct xe_sched_job ; +#define SYNC_PARSE_FLAG_EXEC BIT(0) +#define SYNC_PARSE_FLAG_LR_MODE BIT(1) +#define SYNC_PARSE_FLAG_DISALLOW_USER_FENCE BIT(2) +int xe_sync_entry_parse (struct xe_device *xe,struct xe_file *xef,struct xe_sync_entry *sync,struct drm_xe_sync __user *sync_user, -bool exec,bool compute_mode); +unsigned int flags);int xe_sync_entry_wait (struct xe_sync_entry *sync);int xe_sync_entry_add_deps (struct xe_sync_entry *sync,struct xe_sched_job *job);
Thomas Hellström
maintainerRe: [Intel-xe] [PATCH 4/5] rm/xe: Allow num_batch_buffer == 0 in exec IOCTL
In reply to Matthew BrostQuoted 31 lines · Matthew Brost
The idea being out-syncs can signal indicating all previous operations on the exec queue are complete. An example use case of this would be support for implementing vkQueueWaitForIdle easily.
v2: Don't add
last_fencefor VM's that do not support dma fences
v3: Use a flags field instead of severval bools in sync parse (Thomas)Signed-off-by Matthew Brost <[email protected]>--- drivers/gpu/drm/xe/xe_exec.c | 23 ++++++++++++++++++++--- drivers/gpu/drm/xe/xe_exec_queue.c | 5 ++++- drivers/gpu/drm/xe/xe_exec_queue_types.h | 5 +++-- drivers/gpu/drm/xe/xe_sync.c | 4 ++++ drivers/gpu/drm/xe/xe_sync.h | 6 +++++- 5 files changed, 36 insertions(+), 7 deletions(-)drivers/gpu/drm/xe/xe_exec.c +2
-1@@ -161,7 +161,8 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file)if (XE_IOCTL_DBG (xe, q->flags &EXEC_QUEUE_FLAG_VM ))return -EINVAL ; -if (XE_IOCTL_DBG (xe, q->width != args->num_batch_buffer)) +if (XE_IOCTL_DBG (xe, args->num_batch_buffer && + q->width != args->num_batch_buffer))return -EINVAL ;if (XE_IOCTL_DBG (xe, q->flags &EXEC_QUEUE_FLAG_BANNED )) {
drivers/gpu/drm/xe/xe_exec.c int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file) · line 189
Quoted 7 lines
drivers/gpu/drm/xe/xe_exec.c +3
-1@@ -183,12 +184,14 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file) err =xe_sync_entry_parse (xe, xef, &syncs[num_syncs++], &syncs_user[i],SYNC_PARSE_FLAG_EXEC | (xe_vm_no_dma_fences (vm) ? -SYNC_PARSE_FLAG_LR_MODE :0 )); +SYNC_PARSE_FLAG_LR_MODE :0 ) | + (!args->num_batch_buffer ? +SYNC_PARSE_FLAG_DISALLOW_USER_FENCE :0 ));
Q: Why do we disallow user fences in this situation?
Quoted 98 lines
drivers/gpu/drm/xe/xe_exec.c +15
-1if (err) goto err_syncs; } -if (xe_exec_queue_is_parallel (q)) { +if (args->num_batch_buffer &&xe_exec_queue_is_parallel (q)) { err =__copy_from_user (addresses, addresses_user,sizeof (u64) * q->width);if (err) {@@ -235,6 +238,18 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file) goto err_exec; } +if (!args->num_batch_buffer) { +if (!xe_vm_no_dma_fences (vm)) { +struct dma_fence *fence = +xe_exec_queue_last_fence_get (q, vm); + +for (i =0 ; i < num_syncs; i++) +xe_sync_entry_signal (&syncs[i],NULL , fence); + } + + goto err_exec; + } +if (xe_exec_queue_is_lr (q) &&xe_exec_queue_ring_full (q)) { err = -EWOULDBLOCK ; goto err_exec;@@ -328,6 +343,8 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file)if (xe_exec_queue_is_lr (q)) q->ring_ops->emit_job (job); +if (!xe_vm_no_dma_fences (vm)) +xe_exec_queue_last_fence_set (q, vm, &job->drm.s_fence->finished);xe_sched_job_push (job);xe_vm_reactivate_rebind (vm);drivers/gpu/drm/xe/xe_exec_queue.c +4
-1@@ -924,7 +924,10 @@ int xe_exec_queue_set_property_ioctl(struct drm_device *dev, void *data,static void xe_exec_queue_last_fence_lockdep_assert (struct xe_exec_queue *q,struct xe_vm *vm) { -lockdep_assert_held_write (&vm->lock); +if (q->flags &EXEC_QUEUE_FLAG_VM ) +lockdep_assert_held_write (&vm->lock); +else +xe_vm_assert_held (vm); }/** drivers/gpu/drm/xe/xe_exec_queue_types.h +3
-2@@ -56,8 +56,9 @@ struct xe_exec_queue {struct xe_hw_fence_irq *fence_irq;/** -* @last_fence: last fence on engine, protected by vm->lock in write -* mode if bind engine +* @last_fence: last fence on exec queue, protected by vm->lock in write +* mode if bind exec queue, protected by dma resv lock if non-bind exec +* queue */ struct dma_fence *last_fence;drivers/gpu/drm/xe/xe_sync.c +4
-0@@ -106,6 +106,7 @@ int xe_sync_entry_parse(struct xe_device *xe, struct xe_file *xef,int err;bool exec = flags &SYNC_PARSE_FLAG_EXEC ;bool lr_mode = flags &SYNC_PARSE_FLAG_LR_MODE ; +bool disallow_user_fence = flags &SYNC_PARSE_FLAG_DISALLOW_USER_FENCE ;bool signal;if (copy_from_user (&sync_in, sync_user,sizeof (*sync_user)))@@ -173,6 +174,9 @@ int xe_sync_entry_parse(struct xe_device *xe, struct xe_file *xef,break ;case DRM_XE_SYNC_USER_FENCE : +if (XE_IOCTL_DBG (xe, disallow_user_fence)) +return -EOPNOTSUPP ; +if (XE_IOCTL_DBG (xe, !signal))return -EOPNOTSUPP ;drivers/gpu/drm/xe/xe_sync.h +0
-0@@ -12,10 +12,14 @@ struct xe_device;struct xe_file ;struct xe_sched_job ;
drivers/gpu/drm/xe/xe_sync.h struct xe_device; · line 22
Quoted 10 lines
drivers/gpu/drm/xe/xe_sync.h +5
-1+#define SYNC_PARSE_FLAG_EXEC BIT(0) +#define SYNC_PARSE_FLAG_LR_MODE BIT(1) +#define SYNC_PARSE_FLAG_DISALLOW_USER_FENCE BIT(2) +int xe_sync_entry_parse (struct xe_device *xe,struct xe_file *xef,struct xe_sync_entry *sync,struct drm_xe_sync __user *sync_user, -bool exec,bool compute_mode); +unsigned int flags);
Should this have been in the previous patch?
Thomas
Quoted 3 lines
drivers/gpu/drm/xe/xe_sync.h +0
-0int xe_sync_entry_wait (struct xe_sync_entry *sync);int xe_sync_entry_add_deps (struct xe_sync_entry *sync,struct xe_sched_job *job);
Matthew Brost
maintainerRe: [Intel-xe] [PATCH 4/5] rm/xe: Allow num_batch_buffer == 0 in exec IOCTL
Quoted 32 lines · Thomas Hellström
Quoted 30 lines · Matthew Brost
The idea being out-syncs can signal indicating all previous operations on the exec queue are complete. An example use case of this would be support for implementing vkQueueWaitForIdle easily.
v2: Don't add
last_fencefor VM's that do not support dma fences
v3: Use a flags field instead of severval bools in sync parse (Thomas)Signed-off-by Matthew Brost <[email protected]>--- drivers/gpu/drm/xe/xe_exec.c | 23 ++++++++++++++++++++--- drivers/gpu/drm/xe/xe_exec_queue.c | 5 ++++- drivers/gpu/drm/xe/xe_exec_queue_types.h | 5 +++-- drivers/gpu/drm/xe/xe_sync.c | 4 ++++ drivers/gpu/drm/xe/xe_sync.h | 6 +++++- 5 files changed, 36 insertions(+), 7 deletions(-)drivers/gpu/drm/xe/xe_exec.c +2
-1@@ -161,7 +161,8 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file)if (XE_IOCTL_DBG (xe, q->flags &EXEC_QUEUE_FLAG_VM ))return -EINVAL ; -if (XE_IOCTL_DBG (xe, q->width != args->num_batch_buffer)) +if (XE_IOCTL_DBG (xe, args->num_batch_buffer && + q->width != args->num_batch_buffer))return -EINVAL ;if (XE_IOCTL_DBG (xe, q->flags &EXEC_QUEUE_FLAG_BANNED )) {@@ -183,12 +184,14 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file) err = xe_sync_entry_parse(xe, xef, &syncs[num_syncs++],
Quoted 10 lines
drivers/gpu/drm/xe/xe_exec.c int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file) · line 189
Quoted 6 lines
drivers/gpu/drm/xe/xe_exec.c +3
-1&syncs_user[i],SYNC_PARSE_FLAG_EXEC | (xe_vm_no_dma_fences (vm) ? - SYNC_PARSE_FLAG_LR_MODE : 0)); +SYNC_PARSE_FLAG_LR_MODE :0 ) | + (!args->num_batch_buffer ? + SYNC_PARSE_FLAG_DISALLOW_USER_FENCE : 0));Q: Why do we disallow user fences in this situation?
A user fence is inserted into the ring so unless we issue a NULL batch or something this is not easy to support. I don't think we have use case for this either. If this is blocker can fix change this. Also I think we can add support this later if needed too.
Quoted 91 lines
Quoted 91 lines
drivers/gpu/drm/xe/xe_exec.c +15
-1if (err) goto err_syncs; } -if (xe_exec_queue_is_parallel (q)) { +if (args->num_batch_buffer &&xe_exec_queue_is_parallel (q)) { err =__copy_from_user (addresses, addresses_user,sizeof (u64) * q->width);if (err) {@@ -235,6 +238,18 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file) goto err_exec; } +if (!args->num_batch_buffer) { +if (!xe_vm_no_dma_fences (vm)) { +struct dma_fence *fence = +xe_exec_queue_last_fence_get (q, vm); + +for (i =0 ; i < num_syncs; i++) +xe_sync_entry_signal (&syncs[i],NULL , fence); + } + + goto err_exec; + } +if (xe_exec_queue_is_lr (q) &&xe_exec_queue_ring_full (q)) { err = -EWOULDBLOCK ; goto err_exec;@@ -328,6 +343,8 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file)if (xe_exec_queue_is_lr (q)) q->ring_ops->emit_job (job); +if (!xe_vm_no_dma_fences (vm)) +xe_exec_queue_last_fence_set (q, vm, &job->drm.s_fence->finished);xe_sched_job_push (job);xe_vm_reactivate_rebind (vm);drivers/gpu/drm/xe/xe_exec_queue.c +4
-1@@ -924,7 +924,10 @@ int xe_exec_queue_set_property_ioctl(struct drm_device *dev, void *data,static void xe_exec_queue_last_fence_lockdep_assert (struct xe_exec_queue *q,struct xe_vm *vm) { -lockdep_assert_held_write (&vm->lock); +if (q->flags &EXEC_QUEUE_FLAG_VM ) +lockdep_assert_held_write (&vm->lock); +else +xe_vm_assert_held (vm); }/** drivers/gpu/drm/xe/xe_exec_queue_types.h +3
-2@@ -56,8 +56,9 @@ struct xe_exec_queue {struct xe_hw_fence_irq *fence_irq;/** -* @last_fence: last fence on engine, protected by vm->lock in write -* mode if bind engine +* @last_fence: last fence on exec queue, protected by vm->lock in write +* mode if bind exec queue, protected by dma resv lock if non-bind exec +* queue */ struct dma_fence *last_fence;drivers/gpu/drm/xe/xe_sync.c +4
-0@@ -106,6 +106,7 @@ int xe_sync_entry_parse(struct xe_device *xe, struct xe_file *xef,int err;bool exec = flags &SYNC_PARSE_FLAG_EXEC ;bool lr_mode = flags &SYNC_PARSE_FLAG_LR_MODE ; +bool disallow_user_fence = flags &SYNC_PARSE_FLAG_DISALLOW_USER_FENCE ;bool signal;if (copy_from_user (&sync_in, sync_user,sizeof (*sync_user)))@@ -173,6 +174,9 @@ int xe_sync_entry_parse(struct xe_device *xe, struct xe_file *xef,break ;case DRM_XE_SYNC_USER_FENCE : +if (XE_IOCTL_DBG (xe, disallow_user_fence)) +return -EOPNOTSUPP ; +if (XE_IOCTL_DBG (xe, !signal))return -EOPNOTSUPP ;drivers/gpu/drm/xe/xe_sync.h +3
-0@@ -12,10 +12,14 @@ struct xe_device;struct xe_file ;struct xe_sched_job ; +#define SYNC_PARSE_FLAG_EXEC BIT(0) +#define SYNC_PARSE_FLAG_LR_MODE BIT(1) +#define SYNC_PARSE_FLAG_DISALLOW_USER_FENCE BIT(2)
Quoted 9 lines
drivers/gpu/drm/xe/xe_sync.h struct xe_device; · line 22
Quoted 6 lines
drivers/gpu/drm/xe/xe_sync.h +2
-1+int xe_sync_entry_parse (struct xe_device *xe,struct xe_file *xef,struct xe_sync_entry *sync,struct drm_xe_sync __user *sync_user, -bool exec,bool compute_mode); +unsigned int flags);Should this have been in the previous patch?
Yes, good catch. Will fix.
Matt
Quoted 6 lines
Thomas
Quoted 3 lines
drivers/gpu/drm/xe/xe_sync.h +0
-0int xe_sync_entry_wait (struct xe_sync_entry *sync);int xe_sync_entry_add_deps (struct xe_sync_entry *sync,struct xe_sched_job *job);
Matthew Brost
maintainerPATCH 5/5 drm/xe: Take in-syncs into account when num_execs or num_binds == 0
In reply to Matthew BrostWait on in-syncs before signaling out-syncs if num_execs or num_binds == 0 in execbuf IOCTL or VM bind IOCTL respectfully.
v2: Wait on last fence in addition to in-fences (Thomas)
--- drivers/gpu/drm/xe/xe_exec.c | 10 ++++- drivers/gpu/drm/xe/xe_sync.c | 74 ++++++++++++++++++++++++++++++++++++ drivers/gpu/drm/xe/xe_sync.h | 5 +++ drivers/gpu/drm/xe/xe_vm.c | 24 ++++++++++-- 4 files changed, 107 insertions(+), 6 deletions(-)
drivers/gpu/drm/xe/xe_exec.c +8 -2
@@ -240,11 +240,17 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file)if (!args->num_batch_buffer) {if (!xe_vm_no_dma_fences (vm)) { -struct dma_fence *fence = -xe_exec_queue_last_fence_get (q, vm); +struct dma_fence *fence; + fence =xe_sync_in_fence_get (syncs, num_syncs, q, vm); +if (IS_ERR (fence)) { + err =PTR_ERR (fence); + goto err_exec; + }for (i =0 ; i < num_syncs; i++)xe_sync_entry_signal (&syncs[i],NULL , fence); +xe_exec_queue_last_fence_set (q, vm, fence); +dma_fence_put (fence); } goto err_exec;
drivers/gpu/drm/xe/xe_sync.c +74 -0
89 lines not on this page · raw message
drivers/gpu/drm/xe/xe_sync.h +5 -0
@@ -9,8 +9,10 @@#include "xe_sync_types.h" struct xe_device ; +struct xe_exec_queue ;struct xe_file ;struct xe_sched_job ; +struct xe_vm ;#define SYNC_PARSE_FLAG_EXEC BIT(0)#define SYNC_PARSE_FLAG_LR_MODE BIT(1)
@@ -27,5 +29,8 @@ void xe_sync_entry_signal(struct xe_sync_entry *sync,struct xe_sched_job *job, struct dma_fence *fence);void xe_sync_entry_cleanup (struct xe_sync_entry *sync); +struct dma_fence * +xe_sync_in_fence_get (struct xe_sync_entry *sync,int num_sync, +struct xe_exec_queue *q,struct xe_vm *vm);#endif
drivers/gpu/drm/xe/xe_vm.c +20 -4
@@ -3158,12 +3158,28 @@ int xe_vm_bind_ioctl(struct drm_device *dev, void *data, struct drm_file *file) unwind_ops:vm_bind_ioctl_ops_unwind (vm, ops, args->num_binds); free_syncs: -for (i =0 ; err == -ENODATA && i < num_syncs; i++) { -struct dma_fence *fence = -xe_exec_queue_last_fence_get (to_wait_exec_queue (vm, q), vm); +if (err == -ENODATA ) { +struct dma_fence *fence; + + fence =xe_sync_in_fence_get (syncs, num_syncs, +to_wait_exec_queue (vm, q), vm); +if (IS_ERR (fence)) { + err =PTR_ERR (fence); + goto cleanup_syncs; + } +for (i =0 ; i < num_syncs; i++) +xe_sync_entry_signal (&syncs[i],NULL , fence); +if (!async) { +long timeout =dma_fence_wait (fence, true); -xe_sync_entry_signal (&syncs[i],NULL , fence); +if (timeout <0 ) + err = -EINTR ; + } +xe_exec_queue_last_fence_set (to_wait_exec_queue (vm, q), vm, + fence); +dma_fence_put (fence); } +cleanup_syncs:while (num_syncs--)xe_sync_entry_cleanup (&syncs[num_syncs]);
Signature · 1 line
2.34.1
Patchwork
CI[Intel-xe] ✗ CI.Patch_applied: failure for Allow num_binds/num_batch_buffer to be zero IOCTLs
In reply to Matthew Brost== Series Details ==
== Series Details ==
Series: Allow num_binds/num_batch_buffer to be zero IOCTLs
URL : https://patchwork.freedesktop.org/series/126546/
State : failure
== Summary ==
=== Applying kernel patches on branch 'drm-xe-next' with base: === Base commit: 6253cfa4c drm/xe/pmu: Drop interrupt pmu event
=== git am output follows ===
Applying: drm/xe: Allow num_binds == 0 in VM bind IOCTL
Applying: drm/xe: Use a flags field instead of bools for VMA create
Applying: drm/xe: Use a flags field instead of bools for sync parse
Applying: rm/xe: Allow num_batch_buffer == 0 in exec IOCTL
Applying: drm/xe: Take in-syncs into account when num_execs or num_binds == 0
Patchwork
CI[Intel-xe] ✗ CI.Patch_applied: failure for Allow num_binds/num_batch_buffer to be zero IOCTLs
In reply to Matthew Brost== Series Details ==
== Series Details ==
Series: Allow num_binds/num_batch_buffer to be zero IOCTLs
URL : https://patchwork.freedesktop.org/series/126546/
State : failure
== Summary ==
=== Applying kernel patches on branch 'drm-xe-next' with base: === Base commit: cb9b702d4 drm/xe/dg2: Wa_18028616096 now applies to all DG2 === git am output follows ===
error: patch failed: drivers/gpu/drm/xe/xe_sync.c:118
error: drivers/gpu/drm/xe/xe_sync.c: patch does not apply
hint: Use 'git am --show-current-patch' to see the failed patch
Applying: drm/xe: Allow num_binds == 0 in VM bind IOCTL
Applying: drm/xe: Use a flags field instead of bools for VMA create
Applying: drm/xe: Use a flags field instead of bools for sync parse Patch failed at 0003 drm/xe: Use a flags field instead of bools for sync parse When you have resolved this problem, run "git am --continue".
If you prefer to skip this patch, run "git am --skip" instead.
To restore the original branch and stop patching, run "git am --abort".