summaryrefslogtreecommitdiff
path: root/drivers/gpu
diff options
context:
space:
mode:
authorDave Airlie <airlied@redhat.com>2024-04-05 05:25:28 +0300
committerDave Airlie <airlied@redhat.com>2024-04-05 05:25:35 +0300
commita5b5ab33d2e1f0f433fb33312923103e0b813aa7 (patch)
tree5ff235c5687f71f185920471a209fd5a15edfbb1 /drivers/gpu
parent4cf09f171f6f4e928ec2bd714b78928f1e2b2570 (diff)
parent77a011012d7d8b98368a763bf74317c6d5ce00f1 (diff)
downloadlinux-a5b5ab33d2e1f0f433fb33312923103e0b813aa7.tar.xz
Merge tag 'drm-xe-fixes-2024-04-04' of https://gitlab.freedesktop.org/drm/xe/kernel into drm-fixes
- Stop using system_unbound_wq for preempt fences, as this can cause starvation when reaching more than max_active defined by workqueue - Fix saving unordered rebinding fences by attaching them as kernel feces to the vm's resv - Fix TLB invalidation fences completing out of order - Move rebind TLB invalidation to the ring ops to reduce the latency Signed-off-by: Dave Airlie <airlied@redhat.com> From: Lucas De Marchi <lucas.demarchi@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/tizan6wdpxu4ayudeikjglxdgzmnhdzj3li3z2pgkierjtozzw@lbfddeg43a7h
Diffstat (limited to 'drivers/gpu')
-rw-r--r--drivers/gpu/drm/xe/xe_device.c11
-rw-r--r--drivers/gpu/drm/xe/xe_device_types.h3
-rw-r--r--drivers/gpu/drm/xe/xe_exec.c79
-rw-r--r--drivers/gpu/drm/xe/xe_exec_queue_types.h5
-rw-r--r--drivers/gpu/drm/xe/xe_gt_pagefault.c3
-rw-r--r--drivers/gpu/drm/xe/xe_gt_tlb_invalidation.c1
-rw-r--r--drivers/gpu/drm/xe/xe_gt_types.h7
-rw-r--r--drivers/gpu/drm/xe/xe_preempt_fence.c2
-rw-r--r--drivers/gpu/drm/xe/xe_pt.c25
-rw-r--r--drivers/gpu/drm/xe/xe_ring_ops.c11
-rw-r--r--drivers/gpu/drm/xe/xe_sched_job.c10
-rw-r--r--drivers/gpu/drm/xe/xe_sched_job_types.h2
-rw-r--r--drivers/gpu/drm/xe/xe_vm.c110
-rw-r--r--drivers/gpu/drm/xe/xe_vm.h8
-rw-r--r--drivers/gpu/drm/xe/xe_vm_types.h8
15 files changed, 140 insertions, 145 deletions
diff --git a/drivers/gpu/drm/xe/xe_device.c b/drivers/gpu/drm/xe/xe_device.c
index ca85e81fdb44..d32ff3857e65 100644
--- a/drivers/gpu/drm/xe/xe_device.c
+++ b/drivers/gpu/drm/xe/xe_device.c
@@ -193,6 +193,9 @@ static void xe_device_destroy(struct drm_device *dev, void *dummy)
{
struct xe_device *xe = to_xe_device(dev);
+ if (xe->preempt_fence_wq)
+ destroy_workqueue(xe->preempt_fence_wq);
+
if (xe->ordered_wq)
destroy_workqueue(xe->ordered_wq);
@@ -258,9 +261,15 @@ struct xe_device *xe_device_create(struct pci_dev *pdev,
INIT_LIST_HEAD(&xe->pinned.external_vram);
INIT_LIST_HEAD(&xe->pinned.evicted);
+ xe->preempt_fence_wq = alloc_ordered_workqueue("xe-preempt-fence-wq", 0);
xe->ordered_wq = alloc_ordered_workqueue("xe-ordered-wq", 0);
xe->unordered_wq = alloc_workqueue("xe-unordered-wq", 0, 0);
- if (!xe->ordered_wq || !xe->unordered_wq) {
+ if (!xe->ordered_wq || !xe->unordered_wq ||
+ !xe->preempt_fence_wq) {
+ /*
+ * Cleanup done in xe_device_destroy via
+ * drmm_add_action_or_reset register above
+ */
drm_err(&xe->drm, "Failed to allocate xe workqueues\n");
err = -ENOMEM;
goto err;
diff --git a/drivers/gpu/drm/xe/xe_device_types.h b/drivers/gpu/drm/xe/xe_device_types.h
index 9785eef2e5a4..8e3a222b41cf 100644
--- a/drivers/gpu/drm/xe/xe_device_types.h
+++ b/drivers/gpu/drm/xe/xe_device_types.h
@@ -363,6 +363,9 @@ struct xe_device {
/** @ufence_wq: user fence wait queue */
wait_queue_head_t ufence_wq;
+ /** @preempt_fence_wq: used to serialize preempt fences */
+ struct workqueue_struct *preempt_fence_wq;
+
/** @ordered_wq: used to serialize compute mode resume */
struct workqueue_struct *ordered_wq;
diff --git a/drivers/gpu/drm/xe/xe_exec.c b/drivers/gpu/drm/xe/xe_exec.c
index 826c8b389672..cc5e0f75de3c 100644
--- a/drivers/gpu/drm/xe/xe_exec.c
+++ b/drivers/gpu/drm/xe/xe_exec.c
@@ -94,48 +94,16 @@
* Unlock all
*/
+/*
+ * Add validation and rebinding to the drm_exec locking loop, since both can
+ * trigger eviction which may require sleeping dma_resv locks.
+ */
static int xe_exec_fn(struct drm_gpuvm_exec *vm_exec)
{
struct xe_vm *vm = container_of(vm_exec->vm, struct xe_vm, gpuvm);
- struct drm_gem_object *obj;
- unsigned long index;
- int num_fences;
- int ret;
-
- ret = drm_gpuvm_validate(vm_exec->vm, &vm_exec->exec);
- if (ret)
- return ret;
-
- /*
- * 1 fence slot for the final submit, and 1 more for every per-tile for
- * GPU bind and 1 extra for CPU bind. Note that there are potentially
- * many vma per object/dma-resv, however the fence slot will just be
- * re-used, since they are largely the same timeline and the seqno
- * should be in order. In the case of CPU bind there is dummy fence used
- * for all CPU binds, so no need to have a per-tile slot for that.
- */
- num_fences = 1 + 1 + vm->xe->info.tile_count;
- /*
- * We don't know upfront exactly how many fence slots we will need at
- * the start of the exec, since the TTM bo_validate above can consume
- * numerous fence slots. Also due to how the dma_resv_reserve_fences()
- * works it only ensures that at least that many fence slots are
- * available i.e if there are already 10 slots available and we reserve
- * two more, it can just noop without reserving anything. With this it
- * is quite possible that TTM steals some of the fence slots and then
- * when it comes time to do the vma binding and final exec stage we are
- * lacking enough fence slots, leading to some nasty BUG_ON() when
- * adding the fences. Hence just add our own fences here, after the
- * validate stage.
- */
- drm_exec_for_each_locked_object(&vm_exec->exec, index, obj) {
- ret = dma_resv_reserve_fences(obj->resv, num_fences);
- if (ret)
- return ret;
- }
-
- return 0;
+ /* The fence slot added here is intended for the exec sched job. */
+ return xe_vm_validate_rebind(vm, &vm_exec->exec, 1);
}
int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
@@ -152,7 +120,6 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
struct drm_exec *exec = &vm_exec.exec;
u32 i, num_syncs = 0, num_ufence = 0;
struct xe_sched_job *job;
- struct dma_fence *rebind_fence;
struct xe_vm *vm;
bool write_locked, skip_retry = false;
ktime_t end = 0;
@@ -290,39 +257,7 @@ retry:
goto err_exec;
}
- /*
- * Rebind any invalidated userptr or evicted BOs in the VM, non-compute
- * VM mode only.
- */
- rebind_fence = xe_vm_rebind(vm, false);
- if (IS_ERR(rebind_fence)) {
- err = PTR_ERR(rebind_fence);
- goto err_put_job;
- }
-
- /*
- * We store the rebind_fence in the VM so subsequent execs don't get
- * scheduled before the rebinds of userptrs / evicted BOs is complete.
- */
- if (rebind_fence) {
- dma_fence_put(vm->rebind_fence);
- vm->rebind_fence = rebind_fence;
- }
- if (vm->rebind_fence) {
- if (test_bit(DMA_FENCE_FLAG_SIGNALED_BIT,
- &vm->rebind_fence->flags)) {
- dma_fence_put(vm->rebind_fence);
- vm->rebind_fence = NULL;
- } else {
- dma_fence_get(vm->rebind_fence);
- err = drm_sched_job_add_dependency(&job->drm,
- vm->rebind_fence);
- if (err)
- goto err_put_job;
- }
- }
-
- /* Wait behind munmap style rebinds */
+ /* Wait behind rebinds */
if (!xe_vm_in_lr_mode(vm)) {
err = drm_sched_job_add_resv_dependencies(&job->drm,
xe_vm_resv(vm),
diff --git a/drivers/gpu/drm/xe/xe_exec_queue_types.h b/drivers/gpu/drm/xe/xe_exec_queue_types.h
index 62b3d9d1d7cd..462b33195032 100644
--- a/drivers/gpu/drm/xe/xe_exec_queue_types.h
+++ b/drivers/gpu/drm/xe/xe_exec_queue_types.h
@@ -148,6 +148,11 @@ struct xe_exec_queue {
const struct xe_ring_ops *ring_ops;
/** @entity: DRM sched entity for this exec queue (1 to 1 relationship) */
struct drm_sched_entity *entity;
+ /**
+ * @tlb_flush_seqno: The seqno of the last rebind tlb flush performed
+ * Protected by @vm's resv. Unused if @vm == NULL.
+ */
+ u64 tlb_flush_seqno;
/** @lrc: logical ring context for this exec queue */
struct xe_lrc lrc[];
};
diff --git a/drivers/gpu/drm/xe/xe_gt_pagefault.c b/drivers/gpu/drm/xe/xe_gt_pagefault.c
index 241c294270d9..fa9e9853c53b 100644
--- a/drivers/gpu/drm/xe/xe_gt_pagefault.c
+++ b/drivers/gpu/drm/xe/xe_gt_pagefault.c
@@ -100,10 +100,9 @@ static int xe_pf_begin(struct drm_exec *exec, struct xe_vma *vma,
{
struct xe_bo *bo = xe_vma_bo(vma);
struct xe_vm *vm = xe_vma_vm(vma);
- unsigned int num_shared = 2; /* slots for bind + move */
int err;
- err = xe_vm_prepare_vma(exec, vma, num_shared);
+ err = xe_vm_lock_vma(exec, vma);
if (err)
return err;
diff --git a/drivers/gpu/drm/xe/xe_gt_tlb_invalidation.c b/drivers/gpu/drm/xe/xe_gt_tlb_invalidation.c
index f03e077f81a0..e598a4363d01 100644
--- a/drivers/gpu/drm/xe/xe_gt_tlb_invalidation.c
+++ b/drivers/gpu/drm/xe/xe_gt_tlb_invalidation.c
@@ -61,7 +61,6 @@ int xe_gt_tlb_invalidation_init(struct xe_gt *gt)
INIT_LIST_HEAD(&gt->tlb_invalidation.pending_fences);
spin_lock_init(&gt->tlb_invalidation.pending_lock);
spin_lock_init(&gt->tlb_invalidation.lock);
- gt->tlb_invalidation.fence_context = dma_fence_context_alloc(1);
INIT_DELAYED_WORK(&gt->tlb_invalidation.fence_tdr,
xe_gt_tlb_fence_timeout);
diff --git a/drivers/gpu/drm/xe/xe_gt_types.h b/drivers/gpu/drm/xe/xe_gt_types.h
index 70c615dd1498..07b2f724ec45 100644
--- a/drivers/gpu/drm/xe/xe_gt_types.h
+++ b/drivers/gpu/drm/xe/xe_gt_types.h
@@ -177,13 +177,6 @@ struct xe_gt {
* xe_gt_tlb_fence_timeout after the timeut interval is over.
*/
struct delayed_work fence_tdr;
- /** @tlb_invalidation.fence_context: context for TLB invalidation fences */
- u64 fence_context;
- /**
- * @tlb_invalidation.fence_seqno: seqno to TLB invalidation fences, protected by
- * tlb_invalidation.lock
- */
- u32 fence_seqno;
/** @tlb_invalidation.lock: protects TLB invalidation fences */
spinlock_t lock;
} tlb_invalidation;
diff --git a/drivers/gpu/drm/xe/xe_preempt_fence.c b/drivers/gpu/drm/xe/xe_preempt_fence.c
index 7bce2a332603..7d50c6e89d8e 100644
--- a/drivers/gpu/drm/xe/xe_preempt_fence.c
+++ b/drivers/gpu/drm/xe/xe_preempt_fence.c
@@ -49,7 +49,7 @@ static bool preempt_fence_enable_signaling(struct dma_fence *fence)
struct xe_exec_queue *q = pfence->q;
pfence->error = q->ops->suspend(q);
- queue_work(system_unbound_wq, &pfence->preempt_work);
+ queue_work(q->vm->xe->preempt_fence_wq, &pfence->preempt_work);
return true;
}
diff --git a/drivers/gpu/drm/xe/xe_pt.c b/drivers/gpu/drm/xe/xe_pt.c
index 7f54bc3e389d..4efc8c1a3d7a 100644
--- a/drivers/gpu/drm/xe/xe_pt.c
+++ b/drivers/gpu/drm/xe/xe_pt.c
@@ -1135,8 +1135,7 @@ static int invalidation_fence_init(struct xe_gt *gt,
spin_lock_irq(&gt->tlb_invalidation.lock);
dma_fence_init(&ifence->base.base, &invalidation_fence_ops,
&gt->tlb_invalidation.lock,
- gt->tlb_invalidation.fence_context,
- ++gt->tlb_invalidation.fence_seqno);
+ dma_fence_context_alloc(1), 1);
spin_unlock_irq(&gt->tlb_invalidation.lock);
INIT_LIST_HEAD(&ifence->base.link);
@@ -1236,6 +1235,13 @@ __xe_pt_bind_vma(struct xe_tile *tile, struct xe_vma *vma, struct xe_exec_queue
err = xe_pt_prepare_bind(tile, vma, entries, &num_entries);
if (err)
goto err;
+
+ err = dma_resv_reserve_fences(xe_vm_resv(vm), 1);
+ if (!err && !xe_vma_has_no_bo(vma) && !xe_vma_bo(vma)->vm)
+ err = dma_resv_reserve_fences(xe_vma_bo(vma)->ttm.base.resv, 1);
+ if (err)
+ goto err;
+
xe_tile_assert(tile, num_entries <= ARRAY_SIZE(entries));
xe_vm_dbg_print_entries(tile_to_xe(tile), entries, num_entries);
@@ -1254,11 +1260,13 @@ __xe_pt_bind_vma(struct xe_tile *tile, struct xe_vma *vma, struct xe_exec_queue
* non-faulting LR, in particular on user-space batch buffer chaining,
* it needs to be done here.
*/
- if ((rebind && !xe_vm_in_lr_mode(vm) && !vm->batch_invalidate_tlb) ||
- (!rebind && xe_vm_has_scratch(vm) && xe_vm_in_preempt_fence_mode(vm))) {
+ if ((!rebind && xe_vm_has_scratch(vm) && xe_vm_in_preempt_fence_mode(vm))) {
ifence = kzalloc(sizeof(*ifence), GFP_KERNEL);
if (!ifence)
return ERR_PTR(-ENOMEM);
+ } else if (rebind && !xe_vm_in_lr_mode(vm)) {
+ /* We bump also if batch_invalidate_tlb is true */
+ vm->tlb_flush_seqno++;
}
rfence = kzalloc(sizeof(*rfence), GFP_KERNEL);
@@ -1297,7 +1305,7 @@ __xe_pt_bind_vma(struct xe_tile *tile, struct xe_vma *vma, struct xe_exec_queue
}
/* add shared fence now for pagetable delayed destroy */
- dma_resv_add_fence(xe_vm_resv(vm), fence, !rebind &&
+ dma_resv_add_fence(xe_vm_resv(vm), fence, rebind ||
last_munmap_rebind ?
DMA_RESV_USAGE_KERNEL :
DMA_RESV_USAGE_BOOKKEEP);
@@ -1576,6 +1584,7 @@ __xe_pt_unbind_vma(struct xe_tile *tile, struct xe_vma *vma, struct xe_exec_queu
struct dma_fence *fence = NULL;
struct invalidation_fence *ifence;
struct xe_range_fence *rfence;
+ int err;
LLIST_HEAD(deferred);
@@ -1593,6 +1602,12 @@ __xe_pt_unbind_vma(struct xe_tile *tile, struct xe_vma *vma, struct xe_exec_queu
xe_pt_calc_rfence_interval(vma, &unbind_pt_update, entries,
num_entries);
+ err = dma_resv_reserve_fences(xe_vm_resv(vm), 1);
+ if (!err && !xe_vma_has_no_bo(vma) && !xe_vma_bo(vma)->vm)
+ err = dma_resv_reserve_fences(xe_vma_bo(vma)->ttm.base.resv, 1);
+ if (err)
+ return ERR_PTR(err);
+
ifence = kzalloc(sizeof(*ifence), GFP_KERNEL);
if (!ifence)
return ERR_PTR(-ENOMEM);
diff --git a/drivers/gpu/drm/xe/xe_ring_ops.c b/drivers/gpu/drm/xe/xe_ring_ops.c
index c4edffcd4a32..5b2b37b59813 100644
--- a/drivers/gpu/drm/xe/xe_ring_ops.c
+++ b/drivers/gpu/drm/xe/xe_ring_ops.c
@@ -219,10 +219,9 @@ static void __emit_job_gen12_simple(struct xe_sched_job *job, struct xe_lrc *lrc
{
u32 dw[MAX_JOB_SIZE_DW], i = 0;
u32 ppgtt_flag = get_ppgtt_flag(job);
- struct xe_vm *vm = job->q->vm;
struct xe_gt *gt = job->q->gt;
- if (vm && vm->batch_invalidate_tlb) {
+ if (job->ring_ops_flush_tlb) {
dw[i++] = preparser_disable(true);
i = emit_flush_imm_ggtt(xe_lrc_start_seqno_ggtt_addr(lrc),
seqno, true, dw, i);
@@ -270,7 +269,6 @@ static void __emit_job_gen12_video(struct xe_sched_job *job, struct xe_lrc *lrc,
struct xe_gt *gt = job->q->gt;
struct xe_device *xe = gt_to_xe(gt);
bool decode = job->q->class == XE_ENGINE_CLASS_VIDEO_DECODE;
- struct xe_vm *vm = job->q->vm;
dw[i++] = preparser_disable(true);
@@ -282,13 +280,13 @@ static void __emit_job_gen12_video(struct xe_sched_job *job, struct xe_lrc *lrc,
i = emit_aux_table_inv(gt, VE0_AUX_INV, dw, i);
}
- if (vm && vm->batch_invalidate_tlb)
+ if (job->ring_ops_flush_tlb)
i = emit_flush_imm_ggtt(xe_lrc_start_seqno_ggtt_addr(lrc),
seqno, true, dw, i);
dw[i++] = preparser_disable(false);
- if (!vm || !vm->batch_invalidate_tlb)
+ if (!job->ring_ops_flush_tlb)
i = emit_store_imm_ggtt(xe_lrc_start_seqno_ggtt_addr(lrc),
seqno, dw, i);
@@ -317,7 +315,6 @@ static void __emit_job_gen12_render_compute(struct xe_sched_job *job,
struct xe_gt *gt = job->q->gt;
struct xe_device *xe = gt_to_xe(gt);
bool lacks_render = !(gt->info.engine_mask & XE_HW_ENGINE_RCS_MASK);
- struct xe_vm *vm = job->q->vm;
u32 mask_flags = 0;
dw[i++] = preparser_disable(true);
@@ -327,7 +324,7 @@ static void __emit_job_gen12_render_compute(struct xe_sched_job *job,
mask_flags = PIPE_CONTROL_3D_ENGINE_FLAGS;
/* See __xe_pt_bind_vma() for a discussion on TLB invalidations. */
- i = emit_pipe_invalidate(mask_flags, vm && vm->batch_invalidate_tlb, dw, i);
+ i = emit_pipe_invalidate(mask_flags, job->ring_ops_flush_tlb, dw, i);
/* hsdes: 1809175790 */
if (has_aux_ccs(xe))
diff --git a/drivers/gpu/drm/xe/xe_sched_job.c b/drivers/gpu/drm/xe/xe_sched_job.c
index 8151ddafb940..b0c7fa4693cf 100644
--- a/drivers/gpu/drm/xe/xe_sched_job.c
+++ b/drivers/gpu/drm/xe/xe_sched_job.c
@@ -250,6 +250,16 @@ bool xe_sched_job_completed(struct xe_sched_job *job)
void xe_sched_job_arm(struct xe_sched_job *job)
{
+ struct xe_exec_queue *q = job->q;
+ struct xe_vm *vm = q->vm;
+
+ if (vm && !xe_sched_job_is_migration(q) && !xe_vm_in_lr_mode(vm) &&
+ (vm->batch_invalidate_tlb || vm->tlb_flush_seqno != q->tlb_flush_seqno)) {
+ xe_vm_assert_held(vm);
+ q->tlb_flush_seqno = vm->tlb_flush_seqno;
+ job->ring_ops_flush_tlb = true;
+ }
+
drm_sched_job_arm(&job->drm);
}
diff --git a/drivers/gpu/drm/xe/xe_sched_job_types.h b/drivers/gpu/drm/xe/xe_sched_job_types.h
index b1d83da50a53..5e12724219fd 100644
--- a/drivers/gpu/drm/xe/xe_sched_job_types.h
+++ b/drivers/gpu/drm/xe/xe_sched_job_types.h
@@ -39,6 +39,8 @@ struct xe_sched_job {
} user_fence;
/** @migrate_flush_flags: Additional flush flags for migration jobs */
u32 migrate_flush_flags;
+ /** @ring_ops_flush_tlb: The ring ops need to flush TLB before payload. */
+ bool ring_ops_flush_tlb;
/** @batch_addr: batch buffer address of job */
u64 batch_addr[];
};
diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c
index f88faef4142b..62d1ef8867a8 100644
--- a/drivers/gpu/drm/xe/xe_vm.c
+++ b/drivers/gpu/drm/xe/xe_vm.c
@@ -482,17 +482,53 @@ static int xe_gpuvm_validate(struct drm_gpuvm_bo *vm_bo, struct drm_exec *exec)
return 0;
}
+/**
+ * xe_vm_validate_rebind() - Validate buffer objects and rebind vmas
+ * @vm: The vm for which we are rebinding.
+ * @exec: The struct drm_exec with the locked GEM objects.
+ * @num_fences: The number of fences to reserve for the operation, not
+ * including rebinds and validations.
+ *
+ * Validates all evicted gem objects and rebinds their vmas. Note that
+ * rebindings may cause evictions and hence the validation-rebind
+ * sequence is rerun until there are no more objects to validate.
+ *
+ * Return: 0 on success, negative error code on error. In particular,
+ * may return -EINTR or -ERESTARTSYS if interrupted, and -EDEADLK if
+ * the drm_exec transaction needs to be restarted.
+ */
+int xe_vm_validate_rebind(struct xe_vm *vm, struct drm_exec *exec,
+ unsigned int num_fences)
+{
+ struct drm_gem_object *obj;
+ unsigned long index;
+ int ret;
+
+ do {
+ ret = drm_gpuvm_validate(&vm->gpuvm, exec);
+ if (ret)
+ return ret;
+
+ ret = xe_vm_rebind(vm, false);
+ if (ret)
+ return ret;
+ } while (!list_empty(&vm->gpuvm.evict.list));
+
+ drm_exec_for_each_locked_object(exec, index, obj) {
+ ret = dma_resv_reserve_fences(obj->resv, num_fences);
+ if (ret)
+ return ret;
+ }
+
+ return 0;
+}
+
static int xe_preempt_work_begin(struct drm_exec *exec, struct xe_vm *vm,
bool *done)
{
int err;
- /*
- * 1 fence for each preempt fence plus a fence for each tile from a
- * possible rebind
- */
- err = drm_gpuvm_prepare_vm(&vm->gpuvm, exec, vm->preempt.num_exec_queues +
- vm->xe->info.tile_count);
+ err = drm_gpuvm_prepare_vm(&vm->gpuvm, exec, 0);
if (err)
return err;
@@ -507,7 +543,7 @@ static int xe_preempt_work_begin(struct drm_exec *exec, struct xe_vm *vm,
return 0;
}
- err = drm_gpuvm_prepare_objects(&vm->gpuvm, exec, vm->preempt.num_exec_queues);
+ err = drm_gpuvm_prepare_objects(&vm->gpuvm, exec, 0);
if (err)
return err;
@@ -515,14 +551,19 @@ static int xe_preempt_work_begin(struct drm_exec *exec, struct xe_vm *vm,
if (err)
return err;
- return drm_gpuvm_validate(&vm->gpuvm, exec);
+ /*
+ * Add validation and rebinding to the locking loop since both can
+ * cause evictions which may require blocing dma_resv locks.
+ * The fence reservation here is intended for the new preempt fences
+ * we attach at the end of the rebind work.
+ */
+ return xe_vm_validate_rebind(vm, exec, vm->preempt.num_exec_queues);
}
static void preempt_rebind_work_func(struct work_struct *w)
{
struct xe_vm *vm = container_of(w, struct xe_vm, preempt.rebind_work);
struct drm_exec exec;
- struct dma_fence *rebind_fence;
unsigned int fence_count = 0;
LIST_HEAD(preempt_fences);
ktime_t end = 0;
@@ -568,18 +609,11 @@ retry:
if (err)
goto out_unlock;
- rebind_fence = xe_vm_rebind(vm, true);
- if (IS_ERR(rebind_fence)) {
- err = PTR_ERR(rebind_fence);
+ err = xe_vm_rebind(vm, true);
+ if (err)
goto out_unlock;
- }
- if (rebind_fence) {
- dma_fence_wait(rebind_fence, false);
- dma_fence_put(rebind_fence);
- }
-
- /* Wait on munmap style VM unbinds */
+ /* Wait on rebinds and munmap style VM unbinds */
wait = dma_resv_wait_timeout(xe_vm_resv(vm),
DMA_RESV_USAGE_KERNEL,
false, MAX_SCHEDULE_TIMEOUT);
@@ -773,14 +807,14 @@ xe_vm_bind_vma(struct xe_vma *vma, struct xe_exec_queue *q,
struct xe_sync_entry *syncs, u32 num_syncs,
bool first_op, bool last_op);
-struct dma_fence *xe_vm_rebind(struct xe_vm *vm, bool rebind_worker)
+int xe_vm_rebind(struct xe_vm *vm, bool rebind_worker)
{
- struct dma_fence *fence = NULL;
+ struct dma_fence *fence;
struct xe_vma *vma, *next;
lockdep_assert_held(&vm->lock);
if (xe_vm_in_lr_mode(vm) && !rebind_worker)
- return NULL;
+ return 0;
xe_vm_assert_held(vm);
list_for_each_entry_safe(vma, next, &vm->rebind_list,
@@ -788,17 +822,17 @@ struct dma_fence *xe_vm_rebind(struct xe_vm *vm, bool rebind_worker)
xe_assert(vm->xe, vma->tile_present);
list_del_init(&vma->combined_links.rebind);
- dma_fence_put(fence);
if (rebind_worker)
trace_xe_vma_rebind_worker(vma);
else
trace_xe_vma_rebind_exec(vma);
fence = xe_vm_bind_vma(vma, NULL, NULL, 0, false, false);
if (IS_ERR(fence))
- return fence;
+ return PTR_ERR(fence);
+ dma_fence_put(fence);
}
- return fence;
+ return 0;
}
static void xe_vma_free(struct xe_vma *vma)
@@ -1004,35 +1038,26 @@ static void xe_vma_destroy(struct xe_vma *vma, struct dma_fence *fence)
}
/**
- * xe_vm_prepare_vma() - drm_exec utility to lock a vma
+ * xe_vm_lock_vma() - drm_exec utility to lock a vma
* @exec: The drm_exec object we're currently locking for.
* @vma: The vma for witch we want to lock the vm resv and any attached
* object's resv.
- * @num_shared: The number of dma-fence slots to pre-allocate in the
- * objects' reservation objects.
*
* Return: 0 on success, negative error code on error. In particular
* may return -EDEADLK on WW transaction contention and -EINTR if
* an interruptible wait is terminated by a signal.
*/
-int xe_vm_prepare_vma(struct drm_exec *exec, struct xe_vma *vma,
- unsigned int num_shared)
+int xe_vm_lock_vma(struct drm_exec *exec, struct xe_vma *vma)
{
struct xe_vm *vm = xe_vma_vm(vma);
struct xe_bo *bo = xe_vma_bo(vma);
int err;
XE_WARN_ON(!vm);
- if (num_shared)
- err = drm_exec_prepare_obj(exec, xe_vm_obj(vm), num_shared);
- else
- err = drm_exec_lock_obj(exec, xe_vm_obj(vm));
- if (!err && bo && !bo->vm) {
- if (num_shared)
- err = drm_exec_prepare_obj(exec, &bo->ttm.base, num_shared);
- else
- err = drm_exec_lock_obj(exec, &bo->ttm.base);
- }
+
+ err = drm_exec_lock_obj(exec, xe_vm_obj(vm));
+ if (!err && bo && !bo->vm)
+ err = drm_exec_lock_obj(exec, &bo->ttm.base);
return err;
}
@@ -1044,7 +1069,7 @@ static void xe_vma_destroy_unlocked(struct xe_vma *vma)
drm_exec_init(&exec, 0, 0);
drm_exec_until_all_locked(&exec) {
- err = xe_vm_prepare_vma(&exec, vma, 0);
+ err = xe_vm_lock_vma(&exec, vma);
drm_exec_retry_on_contention(&exec);
if (XE_WARN_ON(err))
break;
@@ -1589,7 +1614,6 @@ static void vm_destroy_work_func(struct work_struct *w)
XE_WARN_ON(vm->pt_root[id]);
trace_xe_vm_free(vm);
- dma_fence_put(vm->rebind_fence);
kfree(vm);
}
@@ -2512,7 +2536,7 @@ static int op_execute(struct drm_exec *exec, struct xe_vm *vm,
lockdep_assert_held_write(&vm->lock);
- err = xe_vm_prepare_vma(exec, vma, 1);
+ err = xe_vm_lock_vma(exec, vma);
if (err)
return err;
diff --git a/drivers/gpu/drm/xe/xe_vm.h b/drivers/gpu/drm/xe/xe_vm.h
index 6df1f1c7f85d..306cd0934a19 100644
--- a/drivers/gpu/drm/xe/xe_vm.h
+++ b/drivers/gpu/drm/xe/xe_vm.h
@@ -207,7 +207,7 @@ int __xe_vm_userptr_needs_repin(struct xe_vm *vm);
int xe_vm_userptr_check_repin(struct xe_vm *vm);
-struct dma_fence *xe_vm_rebind(struct xe_vm *vm, bool rebind_worker);
+int xe_vm_rebind(struct xe_vm *vm, bool rebind_worker);
int xe_vm_invalidate_vma(struct xe_vma *vma);
@@ -242,8 +242,10 @@ bool xe_vm_validate_should_retry(struct drm_exec *exec, int err, ktime_t *end);
int xe_analyze_vm(struct drm_printer *p, struct xe_vm *vm, int gt_id);
-int xe_vm_prepare_vma(struct drm_exec *exec, struct xe_vma *vma,
- unsigned int num_shared);
+int xe_vm_lock_vma(struct drm_exec *exec, struct xe_vma *vma);
+
+int xe_vm_validate_rebind(struct xe_vm *vm, struct drm_exec *exec,
+ unsigned int num_fences);
/**
* xe_vm_resv() - Return's the vm's reservation object
diff --git a/drivers/gpu/drm/xe/xe_vm_types.h b/drivers/gpu/drm/xe/xe_vm_types.h
index ae5fb565f6bf..badf3945083d 100644
--- a/drivers/gpu/drm/xe/xe_vm_types.h
+++ b/drivers/gpu/drm/xe/xe_vm_types.h
@@ -177,9 +177,6 @@ struct xe_vm {
*/
struct list_head rebind_list;
- /** @rebind_fence: rebind fence from execbuf */
- struct dma_fence *rebind_fence;
-
/**
* @destroy_work: worker to destroy VM, needed as a dma_fence signaling
* from an irq context can be last put and the destroy needs to be able
@@ -264,6 +261,11 @@ struct xe_vm {
bool capture_once;
} error_capture;
+ /**
+ * @tlb_flush_seqno: Required TLB flush seqno for the next exec.
+ * protected by the vm resv.
+ */
+ u64 tlb_flush_seqno;
/** @batch_invalidate_tlb: Always invalidate TLB before batch start */
bool batch_invalidate_tlb;
/** @xef: XE file handle for tracking this VM's drm client */