[Intel-gfx] [PATCH v9 40/70] drm/i915: Use a single page table lock for each gtt.
Maarten Lankhorst
maarten.lankhorst at linux.intel.com
Tue Mar 23 15:50:29 UTC 2021
We may create page table objects on the fly, but we may need to
wait with the ww lock held. Instead of waiting on a freed obj
lock, ensure we have the same lock for each object to keep
-EDEADLK working. This ensures that i915_vma_pin_ww can lock
the page tables when required.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst at linux.intel.com>
Reviewed-by: Thomas Hellström <thomas.hellstrom at linux.intel.com>
---
drivers/gpu/drm/i915/gt/intel_ggtt.c | 8 +++++-
drivers/gpu/drm/i915/gt/intel_gtt.c | 38 ++++++++++++++++++++++++++-
drivers/gpu/drm/i915/gt/intel_gtt.h | 5 ++++
drivers/gpu/drm/i915/gt/intel_ppgtt.c | 3 ++-
drivers/gpu/drm/i915/i915_vma.c | 5 ++++
5 files changed, 56 insertions(+), 3 deletions(-)
diff --git a/drivers/gpu/drm/i915/gt/intel_ggtt.c b/drivers/gpu/drm/i915/gt/intel_ggtt.c
index c8eb034c806a..c2fc49495029 100644
--- a/drivers/gpu/drm/i915/gt/intel_ggtt.c
+++ b/drivers/gpu/drm/i915/gt/intel_ggtt.c
@@ -656,7 +656,9 @@ static int init_aliasing_ppgtt(struct i915_ggtt *ggtt)
if (err)
goto err_ppgtt;
+ i915_gem_object_lock(ppgtt->vm.scratch[0], NULL);
err = i915_vm_pin_pt_stash(&ppgtt->vm, &stash);
+ i915_gem_object_unlock(ppgtt->vm.scratch[0]);
if (err)
goto err_stash;
@@ -743,6 +745,7 @@ static void ggtt_cleanup_hw(struct i915_ggtt *ggtt)
mutex_unlock(&ggtt->vm.mutex);
i915_address_space_fini(&ggtt->vm);
+ dma_resv_fini(&ggtt->vm.resv);
arch_phys_wc_del(ggtt->mtrr);
@@ -1129,6 +1132,7 @@ static int ggtt_probe_hw(struct i915_ggtt *ggtt, struct intel_gt *gt)
ggtt->vm.gt = gt;
ggtt->vm.i915 = i915;
ggtt->vm.dma = i915->drm.dev;
+ dma_resv_init(&ggtt->vm.resv);
if (INTEL_GEN(i915) <= 5)
ret = i915_gmch_probe(ggtt);
@@ -1136,8 +1140,10 @@ static int ggtt_probe_hw(struct i915_ggtt *ggtt, struct intel_gt *gt)
ret = gen6_gmch_probe(ggtt);
else
ret = gen8_gmch_probe(ggtt);
- if (ret)
+ if (ret) {
+ dma_resv_fini(&ggtt->vm.resv);
return ret;
+ }
if ((ggtt->vm.total - 1) >> 32) {
drm_err(&i915->drm,
diff --git a/drivers/gpu/drm/i915/gt/intel_gtt.c b/drivers/gpu/drm/i915/gt/intel_gtt.c
index 1b532a2791ea..994e4ea28903 100644
--- a/drivers/gpu/drm/i915/gt/intel_gtt.c
+++ b/drivers/gpu/drm/i915/gt/intel_gtt.c
@@ -13,16 +13,36 @@
struct drm_i915_gem_object *alloc_pt_dma(struct i915_address_space *vm, int sz)
{
+ struct drm_i915_gem_object *obj;
+
if (I915_SELFTEST_ONLY(should_fail(&vm->fault_attr, 1)))
i915_gem_shrink_all(vm->i915);
- return i915_gem_object_create_internal(vm->i915, sz);
+ obj = i915_gem_object_create_internal(vm->i915, sz);
+ /* ensure all dma objects have the same reservation class */
+ if (!IS_ERR(obj))
+ obj->base.resv = &vm->resv;
+ return obj;
}
int pin_pt_dma(struct i915_address_space *vm, struct drm_i915_gem_object *obj)
{
int err;
+ i915_gem_object_lock(obj, NULL);
+ err = i915_gem_object_pin_pages(obj);
+ i915_gem_object_unlock(obj);
+ if (err)
+ return err;
+
+ i915_gem_object_make_unshrinkable(obj);
+ return 0;
+}
+
+int pin_pt_dma_locked(struct i915_address_space *vm, struct drm_i915_gem_object *obj)
+{
+ int err;
+
err = i915_gem_object_pin_pages(obj);
if (err)
return err;
@@ -56,6 +76,20 @@ void __i915_vm_close(struct i915_address_space *vm)
mutex_unlock(&vm->mutex);
}
+/* lock the vm into the current ww, if we lock one, we lock all */
+int i915_vm_lock_objects(struct i915_address_space *vm,
+ struct i915_gem_ww_ctx *ww)
+{
+ if (vm->scratch[0]->base.resv == &vm->resv) {
+ return i915_gem_object_lock(vm->scratch[0], ww);
+ } else {
+ struct i915_ppgtt *ppgtt = i915_vm_to_ppgtt(vm);
+
+ /* We borrowed the scratch page from ggtt, take the top level object */
+ return i915_gem_object_lock(ppgtt->pd->pt.base, ww);
+ }
+}
+
void i915_address_space_fini(struct i915_address_space *vm)
{
drm_mm_takedown(&vm->mm);
@@ -69,6 +103,7 @@ static void __i915_vm_release(struct work_struct *work)
vm->cleanup(vm);
i915_address_space_fini(vm);
+ dma_resv_fini(&vm->resv);
kfree(vm);
}
@@ -98,6 +133,7 @@ void i915_address_space_init(struct i915_address_space *vm, int subclass)
mutex_init(&vm->mutex);
lockdep_set_subclass(&vm->mutex, subclass);
fs_reclaim_taints_mutex(&vm->mutex);
+ dma_resv_init(&vm->resv);
GEM_BUG_ON(!vm->total);
drm_mm_init(&vm->mm, 0, vm->total);
diff --git a/drivers/gpu/drm/i915/gt/intel_gtt.h b/drivers/gpu/drm/i915/gt/intel_gtt.h
index 784c4372b405..e67e34e17913 100644
--- a/drivers/gpu/drm/i915/gt/intel_gtt.h
+++ b/drivers/gpu/drm/i915/gt/intel_gtt.h
@@ -242,6 +242,7 @@ struct i915_address_space {
atomic_t open;
struct mutex mutex; /* protects vma and our lists */
+ struct dma_resv resv; /* reservation lock for all pd objects, and buffer pool */
#define VM_CLASS_GGTT 0
#define VM_CLASS_PPGTT 1
@@ -351,6 +352,9 @@ struct i915_ppgtt {
#define i915_is_ggtt(vm) ((vm)->is_ggtt)
+int __must_check
+i915_vm_lock_objects(struct i915_address_space *vm, struct i915_gem_ww_ctx *ww);
+
static inline bool
i915_vm_is_4lvl(const struct i915_address_space *vm)
{
@@ -527,6 +531,7 @@ struct i915_page_directory *alloc_pd(struct i915_address_space *vm);
struct i915_page_directory *__alloc_pd(int npde);
int pin_pt_dma(struct i915_address_space *vm, struct drm_i915_gem_object *obj);
+int pin_pt_dma_locked(struct i915_address_space *vm, struct drm_i915_gem_object *obj);
void free_px(struct i915_address_space *vm,
struct i915_page_table *pt, int lvl);
diff --git a/drivers/gpu/drm/i915/gt/intel_ppgtt.c b/drivers/gpu/drm/i915/gt/intel_ppgtt.c
index a91955af50a6..014ae8ac4480 100644
--- a/drivers/gpu/drm/i915/gt/intel_ppgtt.c
+++ b/drivers/gpu/drm/i915/gt/intel_ppgtt.c
@@ -266,7 +266,7 @@ int i915_vm_pin_pt_stash(struct i915_address_space *vm,
for (n = 0; n < ARRAY_SIZE(stash->pt); n++) {
for (pt = stash->pt[n]; pt; pt = pt->stash) {
- err = pin_pt_dma(vm, pt->base);
+ err = pin_pt_dma_locked(vm, pt->base);
if (err)
return err;
}
@@ -308,6 +308,7 @@ void ppgtt_init(struct i915_ppgtt *ppgtt, struct intel_gt *gt)
ppgtt->vm.dma = i915->drm.dev;
ppgtt->vm.total = BIT_ULL(INTEL_INFO(i915)->ppgtt_size);
+ dma_resv_init(&ppgtt->vm.resv);
i915_address_space_init(&ppgtt->vm, VM_CLASS_PPGTT);
ppgtt->vm.vma_ops.bind_vma = ppgtt_bind_vma;
diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
index 265e3a3079e2..c5b9f30ac0a3 100644
--- a/drivers/gpu/drm/i915/i915_vma.c
+++ b/drivers/gpu/drm/i915/i915_vma.c
@@ -884,6 +884,11 @@ int i915_vma_pin_ww(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,
wakeref = intel_runtime_pm_get(&vma->vm->i915->runtime_pm);
if (flags & vma->vm->bind_async_flags) {
+ /* lock VM */
+ err = i915_vm_lock_objects(vma->vm, ww);
+ if (err)
+ goto err_rpm;
+
work = i915_vma_work();
if (!work) {
err = -ENOMEM;
--
2.31.0
More information about the Intel-gfx
mailing list