Skip to contentSkip to conversation

drm/i915/gem: Introduce a migrate interface

PATCH Superseded Thomas Hellström 5 participants 29 messages Jun 28, 2021 – Jun 29, 2021

Subsystems (3)

v3 of 5 Superseded Latest is v5 →

lore
Share & tools
lore
#0 lore raw

v3 cover drm/i915/gem: Introduce a migrate interface

We want to be able to explicitly migrate objects between gem memory regions, initially for display and dma-buf, but there might be more use-cases coming up.

Introduce a gem migrate interface, add a selftest and use it for display fb pinning and dma-buf mapping.

This series should make accelerated desktop work on DG1 with DG1-enabled OpenGL.

v2:
- Address review comments by Matthew Auld on patch 1/5. More details on the patch commit message.
- Address a dma-buf locking issue pointed out by Michael Ruhl, and add a selftest to catch that issue moving forward.
- Rebase the dma-buf migration patch on the above-mentioned fix.

v3:
- Fix i915_gem_object_can_migrate() to return true if object is already in the correct region, even if the object ops doesn't have a migrate() callback.
- Update typo in commit message.

Matthew Auld (1):
drm/i915/gem: Introduce a selftest for the gem object migrate
functionality

Thomas Hellström (4):
drm/i915/gem: Implement object migration drm/i915/display: Migrate objects to LMEM if possible for display drm/i915/gem: Fix same-driver-another-instance dma-buf export drm/i915/gem: Migrate to system at dma-buf map time

 drivers/gpu/drm/i915/display/intel_display.c  |   5 +-
 drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c    |  36 ++-
 drivers/gpu/drm/i915/gem/i915_gem_domain.c    |   2 +-
 drivers/gpu/drm/i915/gem/i915_gem_lmem.c      |  21 --
 drivers/gpu/drm/i915/gem/i915_gem_object.c    |  97 +++++++
 drivers/gpu/drm/i915/gem/i915_gem_object.h    |  12 +-
 .../gpu/drm/i915/gem/i915_gem_object_types.h  |   9 +
 drivers/gpu/drm/i915/gem/i915_gem_ttm.c       |  69 +++--
 drivers/gpu/drm/i915/gem/i915_gem_wait.c      |  19 ++
 .../drm/i915/gem/selftests/i915_gem_dmabuf.c  |  83 +++++-
 .../drm/i915/gem/selftests/i915_gem_migrate.c | 237 ++++++++++++++++++
 .../drm/i915/selftests/i915_live_selftests.h  |   1 +
 12 files changed, 546 insertions(+), 45 deletions(-)
 create mode 100644 drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c
Signature · 1 line
2.31.1
#1 lore raw

PATCH v3 1/5 drm/i915/gem: Implement object migration

Introduce an interface to migrate objects between regions. This is primarily intended to migrate objects to LMEM for display and to SYSTEM for dma-buf, but might be reused in one form or another for performance-based migration.

v2:
- Verify that the memory region given as an id really exists. (Reported by Matthew Auld)
- Call i915_gem_object_{init,release}_memory_region() when switching region to handle also switching region lists. (Reported by Matthew Auld)
v3:
- Fix i915_gem_object_can_migrate() to return true if object is already in the correct region, even if the object ops doesn't have a migrate() callback.
- Update typo in commit message.
- Fix kerneldoc of i915_gem_object_wait_migration().

---
 drivers/gpu/drm/i915/gem/i915_gem_object.c    | 96 +++++++++++++++++++
 drivers/gpu/drm/i915/gem/i915_gem_object.h    | 12 +++
 .../gpu/drm/i915/gem/i915_gem_object_types.h  |  9 ++
 drivers/gpu/drm/i915/gem/i915_gem_ttm.c       | 69 +++++++++----
 drivers/gpu/drm/i915/gem/i915_gem_wait.c      | 19 ++++
 5 files changed, 188 insertions(+), 17 deletions(-)
drivers/gpu/drm/i915/gem/i915_gem_object.c +96 -0
index 07e8ff9a8aae..1c18be067b58 100644
@@ -513,6 +513,102 @@ bool i915_gem_object_has_iomem(const struct drm_i915_gem_object *obj)
 	return obj->mem_flags & I915_BO_FLAG_IOMEM;
 }
 
+/**
+ * i915_gem_object_can_migrate - Whether an object likely can be migrated
+ *
+ * @obj: The object to migrate
+ * @id: The region intended to migrate to
+ *
+ * Check whether the object backend supports migration to the
+ * given region. Note that pinning may affect the ability to migrate.
+ *
+ * Return: true if migration is possible, false otherwise.
+ */
+bool i915_gem_object_can_migrate(struct drm_i915_gem_object *obj,
+				 enum intel_region_id id)
+{
+	struct drm_i915_private *i915 = to_i915(obj->base.dev);
+	unsigned int num_allowed = obj->mm.n_placements;
+	struct intel_memory_region *mr;
+	unsigned int i;
+
+	GEM_BUG_ON(id >= INTEL_REGION_UNKNOWN);
+	GEM_BUG_ON(obj->mm.madv != I915_MADV_WILLNEED);
+
+	mr = i915->mm.regions[id];
+	if (!mr)
+		return false;
+
+	if (obj->mm.region == mr)
+		return true;
+
+	if (!i915_gem_object_evictable(obj))
+		return false;
+
+	if (!obj->ops->migrate)
+		return false;
+
+	if (!(obj->flags & I915_BO_ALLOC_USER))
+		return true;
+
+	if (num_allowed == 0)
+		return false;
+
+	for (i = 0; i < num_allowed; ++i) {
+		if (mr == obj->mm.placements[i])
+			return true;
+	}
+
+	return false;
+}
+
+/**
+ * i915_gem_object_migrate - Migrate an object to the desired region id
+ * @obj: The object to migrate.
+ * @ww: An optional struct i915_gem_ww_ctx. If NULL, the backend may
+ * not be successful in evicting other objects to make room for this object.
+ * @id: The region id to migrate to.
+ *
+ * Attempt to migrate the object to the desired memory region. The
+ * object backend must support migration and the object may not be
+ * pinned, (explicitly pinned pages or pinned vmas). The object must
+ * be locked.
+ * On successful completion, the object will have pages pointing to
+ * memory in the new region, but an async migration task may not have
+ * completed yet, and to accomplish that, i915_gem_object_wait_migration()
+ * must be called.
+ *
+ * Return: 0 on success. Negative error code on failure. In particular may
+ * return -ENXIO on lack of region space, -EDEADLK for deadlock avoidance
+ * if @ww is set, -EINTR or -ERESTARTSYS if signal pending, and
+ * -EBUSY if the object is pinned.
+ */
+int i915_gem_object_migrate(struct drm_i915_gem_object *obj,
+			    struct i915_gem_ww_ctx *ww,
+			    enum intel_region_id id)
+{
+	struct drm_i915_private *i915 = to_i915(obj->base.dev);
+	struct intel_memory_region *mr;
+
+	GEM_BUG_ON(id >= INTEL_REGION_UNKNOWN);
+	GEM_BUG_ON(obj->mm.madv != I915_MADV_WILLNEED);
+	assert_object_held(obj);
+
+	mr = i915->mm.regions[id];
+	GEM_BUG_ON(!mr);
+
+	if (obj->mm.region == mr)
+		return 0;
+
+	if (!i915_gem_object_evictable(obj))
+		return -EBUSY;
+
+	if (!obj->ops->migrate)
+		return -EOPNOTSUPP;
+
+	return obj->ops->migrate(obj, mr);
+}
+
 void i915_gem_init__objects(struct drm_i915_private *i915)
 {
 	INIT_WORK(&i915->mm.free_work, __i915_gem_free_work);
drivers/gpu/drm/i915/gem/i915_gem_object.h +12 -0
index ea3224a480c4..8cbd7a5334e2 100644
@@ -17,6 +17,8 @@
 #include "i915_gem_ww.h"
 #include "i915_vma_types.h"
 
+enum intel_region_id;
+
 /*
  * XXX: There is a prevalence of the assumption that we fit the
  * object's page count inside a 32bit _signed_ variable. Let's document
@@ -597,6 +599,16 @@ bool i915_gem_object_migratable(struct drm_i915_gem_object *obj);
 
 bool i915_gem_object_validates_to_lmem(struct drm_i915_gem_object *obj);
 
+int i915_gem_object_migrate(struct drm_i915_gem_object *obj,
+			    struct i915_gem_ww_ctx *ww,
+			    enum intel_region_id id);
+
+bool i915_gem_object_can_migrate(struct drm_i915_gem_object *obj,
+				 enum intel_region_id id);
+
+int i915_gem_object_wait_migration(struct drm_i915_gem_object *obj,
+				   unsigned int flags);
+
 #ifdef CONFIG_MMU_NOTIFIER
 static inline bool
 i915_gem_object_is_userptr(struct drm_i915_gem_object *obj)
drivers/gpu/drm/i915/gem/i915_gem_object_types.h +9 -0
index 441f913c87e6..ef3de2ae9723 100644
@@ -18,6 +18,7 @@
 
 struct drm_i915_gem_object;
 struct intel_fronbuffer;
+struct intel_memory_region;
 
 /*
  * struct i915_lut_handle tracks the fast lookups from handle to vma used
@@ -77,6 +78,14 @@ struct drm_i915_gem_object_ops {
 	 * delayed_free - Override the default delayed free implementation
 	 */
 	void (*delayed_free)(struct drm_i915_gem_object *obj);
+
+	/**
+	 * migrate - Migrate object to a different region either for
+	 * pinning or for as long as the object lock is held.
+	 */
+	int (*migrate)(struct drm_i915_gem_object *obj,
+		       struct intel_memory_region *mr);
+
 	void (*release)(struct drm_i915_gem_object *obj);
 
 	const struct vm_operations_struct *mmap_ops;
drivers/gpu/drm/i915/gem/i915_gem_ttm.c +52 -17

108 lines not on this page · raw message

drivers/gpu/drm/i915/gem/i915_gem_wait.c +19 -0
index 1070d3afdce7..f909aaa09d9c 100644
@@ -290,3 +290,22 @@ i915_gem_wait_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
 	i915_gem_object_put(obj);
 	return ret;
 }
+
+/**
+ * i915_gem_object_wait_migration - Sync an accelerated migration operation
+ * @obj: The migrating object.
+ * @flags: waiting flags. Currently supports only I915_WAIT_INTERRUPTIBLE.
+ *
+ * Wait for any pending async migration operation on the object,
+ * whether it's explicitly (i915_gem_object_migrate()) or implicitly
+ * (swapin, initial clearing) initiated.
+ *
+ * Return: 0 if successful, -ERESTARTSYS if a signal was hit during waiting.
+ */
+int i915_gem_object_wait_migration(struct drm_i915_gem_object *obj,
+				   unsigned int flags)
+{
+	might_sleep();
+	/* NOP for now. */
+	return 0;
+}
Signature · 6 lines
2.31.1

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
#10 lore raw

PATCH v3 2/5 drm/i915/gem: Introduce a selftest for the gem object migrate functionality

In reply to Thomas Hellström

From: Matthew Auld <[email protected]>

A selftest for the gem object migrate functionality. Slightly adapted from the original by Matthew to the new interface and new fill blit code.

---
 drivers/gpu/drm/i915/gem/i915_gem_object.c    |   1 +
 .../drm/i915/gem/selftests/i915_gem_migrate.c | 237 ++++++++++++++++++
 .../drm/i915/selftests/i915_live_selftests.h  |   1 +
 3 files changed, 239 insertions(+)
 create mode 100644 drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c
drivers/gpu/drm/i915/gem/i915_gem_object.c +1 -0
index 1c18be067b58..ff147fd59874 100644
@@ -649,6 +649,7 @@ static const struct drm_gem_object_funcs i915_gem_object_funcs = {
 #if IS_ENABLED(CONFIG_DRM_I915_SELFTEST)
 #include "selftests/huge_gem_object.c"
 #include "selftests/huge_pages.c"
+#include "selftests/i915_gem_migrate.c"
 #include "selftests/i915_gem_object.c"
 #include "selftests/i915_gem_coherency.c"
 #endif
drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c new file +237 -0

237 lines not on this page · raw message

drivers/gpu/drm/i915/selftests/i915_live_selftests.h +1 -0
index a68197cf1044..e2fd1b61af71 100644
@@ -40,6 +40,7 @@ selftest(hugepages, i915_gem_huge_page_live_selftests)
 selftest(gem_contexts, i915_gem_context_live_selftests)
 selftest(gem_execbuf, i915_gem_execbuffer_live_selftests)
 selftest(client, i915_gem_client_blt_live_selftests)
+selftest(gem_migrate, i915_gem_migrate_live_selftests)
 selftest(reset, intel_reset_live_selftests)
 selftest(memory_region, intel_memory_region_live_selftests)
 selftest(hangcheck, intel_hangcheck_live_selftests)
Signature · 6 lines
2.31.1

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
#16 lore raw

PATCH v3 3/5 drm/i915/display: Migrate objects to LMEM if possible for display

In reply to Thomas Hellström

Objects intended to be used as display framebuffers must reside in LMEM for discrete. If they happen to not do that, migrate them to LMEM before pinning.

---
 drivers/gpu/drm/i915/display/intel_display.c |  5 ++++-
 drivers/gpu/drm/i915/gem/i915_gem_domain.c   |  2 +-
 drivers/gpu/drm/i915/gem/i915_gem_lmem.c     | 21 --------------------
 drivers/gpu/drm/i915/gem/i915_gem_object.h   |  2 --
 4 files changed, 5 insertions(+), 25 deletions(-)
drivers/gpu/drm/i915/display/intel_display.c +4 -1
index eec6c9e9cda7..026c28c612f0 100644
@@ -1331,6 +1331,9 @@ intel_pin_and_fence_fb_obj(struct drm_framebuffer *fb,
 	ret = i915_gem_object_lock(obj, &ww);
 	if (!ret && phys_cursor)
 		ret = i915_gem_object_attach_phys(obj, alignment);
+	else if (!ret && HAS_LMEM(dev_priv))
+		ret = i915_gem_object_migrate(obj, &ww, INTEL_REGION_LMEM);
+	/* TODO: Do we need to sync when migration becomes async? */
 	if (!ret)
 		ret = i915_gem_object_pin_pages(obj);
 	if (ret)
@@ -11778,7 +11781,7 @@ intel_user_framebuffer_create(struct drm_device *dev,
 
 	/* object is backed with LMEM for discrete */
 	i915 = to_i915(obj->base.dev);
-	if (HAS_LMEM(i915) && !i915_gem_object_validates_to_lmem(obj)) {
+	if (HAS_LMEM(i915) && !i915_gem_object_can_migrate(obj, INTEL_REGION_LMEM)) {
 		/* object is "remote", not in local memory */
 		i915_gem_object_put(obj);
 		return ERR_PTR(-EREMOTE);
drivers/gpu/drm/i915/gem/i915_gem_domain.c +1 -1
index 073822100da7..7d1400b13429 100644
@@ -375,7 +375,7 @@ i915_gem_object_pin_to_display_plane(struct drm_i915_gem_object *obj,
 	struct i915_vma *vma;
 	int ret;
 
-	/* Frame buffer must be in LMEM (no migration yet) */
+	/* Frame buffer must be in LMEM */
 	if (HAS_LMEM(i915) && !i915_gem_object_is_lmem(obj))
 		return ERR_PTR(-EINVAL);
 
drivers/gpu/drm/i915/gem/i915_gem_lmem.c +0 -21
index 41d5182cd367..be1d122574af 100644
@@ -23,27 +23,6 @@ i915_gem_object_lmem_io_map(struct drm_i915_gem_object *obj,
 	return io_mapping_map_wc(&obj->mm.region->iomap, offset, size);
 }
 
-/**
- * i915_gem_object_validates_to_lmem - Whether the object is resident in
- * lmem when pages are present.
- * @obj: The object to check.
- *
- * Migratable objects residency may change from under us if the object is
- * not pinned or locked. This function is intended to be used to check whether
- * the object can only reside in lmem when pages are present.
- *
- * Return: Whether the object is always resident in lmem when pages are
- * present.
- */
-bool i915_gem_object_validates_to_lmem(struct drm_i915_gem_object *obj)
-{
-	struct intel_memory_region *mr = READ_ONCE(obj->mm.region);
-
-	return !i915_gem_object_migratable(obj) &&
-		mr && (mr->type == INTEL_MEMORY_LOCAL ||
-		       mr->type == INTEL_MEMORY_STOLEN_LOCAL);
-}
-
 /**
  * i915_gem_object_is_lmem - Whether the object is resident in
  * lmem
drivers/gpu/drm/i915/gem/i915_gem_object.h +0 -2
index 8cbd7a5334e2..d423d8cac4f2 100644
@@ -597,8 +597,6 @@ bool i915_gem_object_evictable(struct drm_i915_gem_object *obj);
 
 bool i915_gem_object_migratable(struct drm_i915_gem_object *obj);
 
-bool i915_gem_object_validates_to_lmem(struct drm_i915_gem_object *obj);
-
 int i915_gem_object_migrate(struct drm_i915_gem_object *obj,
 			    struct i915_gem_ww_ctx *ww,
 			    enum intel_region_id id);
Signature · 1 line
2.31.1
#18 lore raw

PATCH v3 4/5 drm/i915/gem: Fix same-driver-another-instance dma-buf export

In reply to Thomas Hellström

If our exported dma-bufs are imported by another instance of our driver, that instance will typically have the imported dma-bufs locked during map_attachment(). But the exporter also locks the same reservation object in the map_dma_buf() callback, which leads to recursive locking.

Add a live selftest to catch this case, and as a workaround until we fully support dynamic import and export, declare the exporter dynamic by providing NOP pin() and unpin() functions. This means our map_dma_buf() callback will *always* get called locked, and by pinning unconditionally in i915_gem_map_dma_buf() we make sure we don't need to use the move_notify() functionality which is not yet implemented.

---
 drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c    | 31 ++++++-
 .../drm/i915/gem/selftests/i915_gem_dmabuf.c  | 81 ++++++++++++++++++-
 2 files changed, 108 insertions(+), 4 deletions(-)
drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c +29 -2
index 616c3a2f1baf..1d1eeb167d28 100644
@@ -12,6 +12,8 @@
 #include "i915_gem_object.h"
 #include "i915_scatterlist.h"
 
+I915_SELFTEST_DECLARE(static bool force_different_devices;)
+
 static struct drm_i915_gem_object *dma_buf_to_obj(struct dma_buf *buf)
 {
 	return to_intel_bo(buf->priv);
@@ -25,7 +27,9 @@ static struct sg_table *i915_gem_map_dma_buf(struct dma_buf_attachment *attachme
 	struct scatterlist *src, *dst;
 	int ret, i;
 
-	ret = i915_gem_object_pin_pages_unlocked(obj);
+	assert_object_held(obj);
+
+	ret = i915_gem_object_pin_pages(obj);
 	if (ret)
 		goto err;
 
@@ -168,6 +172,26 @@ static int i915_gem_end_cpu_access(struct dma_buf *dma_buf, enum dma_data_direct
 	return err;
 }
 
+/*
+ * As a workaround until we fully support dynamic import and export,
+ * declare the exporter dynamic by providing NOP pin() and unpin() functions.
+ * This means our i915_gem_map_dma_buf() callback will *always* get called
+ * locked, and by pinning unconditionally in i915_gem_map_dma_buf() we make
+ * sure we don't need to use the move_notify() functionality which is
+ * not yet implemented. Typically for the same-driver-another-instance case,
+ * i915_gem_map_dma_buf() will be called at importer attach time and the
+ * mapped sg_list will be cached by the dma-buf core for the
+ * duration of the attachment.
+ */
+static int i915_gem_dmabuf_pin(struct dma_buf_attachment *attach)
+{
+	return 0;
+}
+
+static void i915_gem_dmabuf_unpin(struct dma_buf_attachment *attach)
+{
+}
+
 static const struct dma_buf_ops i915_dmabuf_ops =  {
 	.map_dma_buf = i915_gem_map_dma_buf,
 	.unmap_dma_buf = i915_gem_unmap_dma_buf,
@@ -177,6 +201,8 @@ static const struct dma_buf_ops i915_dmabuf_ops =  {
 	.vunmap = i915_gem_dmabuf_vunmap,
 	.begin_cpu_access = i915_gem_begin_cpu_access,
 	.end_cpu_access = i915_gem_end_cpu_access,
+	.pin = i915_gem_dmabuf_pin,
+	.unpin = i915_gem_dmabuf_unpin,
 };
 
 struct dma_buf *i915_gem_prime_export(struct drm_gem_object *gem_obj, int flags)
@@ -241,7 +267,8 @@ struct drm_gem_object *i915_gem_prime_import(struct drm_device *dev,
 	if (dma_buf->ops == &i915_dmabuf_ops) {
 		obj = dma_buf_to_obj(dma_buf);
 		/* is it from our device? */
-		if (obj->base.dev == dev) {
+		if (obj->base.dev == dev &&
+		    !I915_SELFTEST_ONLY(force_different_devices)) {
 			/*
 			 * Importing dmabuf exported from out own gem increases
 			 * refcount on gem itself instead of f_count of dmabuf.
drivers/gpu/drm/i915/gem/selftests/i915_gem_dmabuf.c +79 -2

106 lines not on this page · raw message

Signature · 1 line
2.31.1
#22 lore raw

PATCH v3 5/5 drm/i915/gem: Migrate to system at dma-buf map time

In reply to Thomas Hellström

Until we support p2p dma or as a complement to that, migrate data to system memory at dma-buf map time if possible.

v2:
- Rebase on dynamic exporter. Update the igt_dmabuf_import_same_driver selftest to migrate if we are LMEM capable.

---
 drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c           | 7 ++++++-
 drivers/gpu/drm/i915/gem/selftests/i915_gem_dmabuf.c | 4 +++-
 2 files changed, 9 insertions(+), 2 deletions(-)
drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c +6 -1
index 1d1eeb167d28..5207aa3af009 100644
@@ -29,7 +29,12 @@ static struct sg_table *i915_gem_map_dma_buf(struct dma_buf_attachment *attachme
 
 	assert_object_held(obj);
 
-	ret = i915_gem_object_pin_pages(obj);
+	if (!i915_gem_object_can_migrate(obj, INTEL_REGION_SMEM))
+		return ERR_PTR(-EOPNOTSUPP);
+
+	ret = i915_gem_object_migrate(obj, NULL, INTEL_REGION_SMEM);
+	if (!ret)
+		ret = i915_gem_object_pin_pages(obj);
 	if (ret)
 		goto err;
 
drivers/gpu/drm/i915/gem/selftests/i915_gem_dmabuf.c +3 -1
index 24735d6c12a2..b6278dba98d0 100644
@@ -94,7 +94,9 @@ static int igt_dmabuf_import_same_driver(void *arg)
 	int err;
 
 	force_different_devices = true;
-	obj = i915_gem_object_create_shmem(i915, PAGE_SIZE);
+	obj = i915_gem_object_create_lmem(i915, PAGE_SIZE, 0);
+	if (IS_ERR(obj))
+		obj = i915_gem_object_create_shmem(i915, PAGE_SIZE);
 	if (IS_ERR(obj))
 		goto out_ret;
 
Signature · 1 line
2.31.1
#11 lore raw

Re: [PATCH v3 2/5] drm/i915/gem: Introduce a selftest for the gem object migrate functionality

In reply to Thomas Hellström
Quoted 254 lines · Thomas Hellström

From: Matthew Auld <[email protected]>

A selftest for the gem object migrate functionality. Slightly adapted from the original by Matthew to the new interface and new fill blit code.

---
  drivers/gpu/drm/i915/gem/i915_gem_object.c    |   1 +
  .../drm/i915/gem/selftests/i915_gem_migrate.c | 237 ++++++++++++++++++
  .../drm/i915/selftests/i915_live_selftests.h  |   1 +
  3 files changed, 239 insertions(+)

create mode 100644 drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c

drivers/gpu/drm/i915/gem/i915_gem_object.c +1 -0
index 1c18be067b58..ff147fd59874 100644
@@ -649,6 +649,7 @@ static const struct drm_gem_object_funcs i915_gem_object_funcs = {
  #if IS_ENABLED(CONFIG_DRM_I915_SELFTEST)
  #include "selftests/huge_gem_object.c"
  #include "selftests/huge_pages.c"
+#include "selftests/i915_gem_migrate.c"
  #include "selftests/i915_gem_object.c"
  #include "selftests/i915_gem_coherency.c"
  #endif
drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c new file +220 -0
new file mode 100644
index 000000000000..a437b66f64d9
@@ -0,0 +1,237 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2020-2021 Intel Corporation
+ */
+
+#include "gt/intel_migrate.h"
+
+static int igt_smem_create_migrate(void *arg)
+{
+	struct intel_gt *gt = arg;
+	struct drm_i915_private *i915 = gt->i915;
+	struct drm_i915_gem_object *obj;
+	struct i915_gem_ww_ctx ww;
+	int err = 0;
+
+	/* Switch object backing-store on create */
+	obj = i915_gem_object_create_lmem(i915, PAGE_SIZE, 0);
+	if (IS_ERR(obj))
+		return PTR_ERR(obj);
+
+	for_i915_gem_ww(&ww, err, true) {
+		err = i915_gem_object_lock(obj, &ww);
+		if (err)
+			continue;
+
+		if (!i915_gem_object_can_migrate(obj, INTEL_REGION_SMEM)) {
+			err = -EINVAL;
+			continue;
+		}
+
+		err = i915_gem_object_migrate(obj, &ww, INTEL_REGION_SMEM);
+		if (err)
+			continue;
+
+		err = i915_gem_object_pin_pages(obj);
+		if (err)
+			continue;
+
+		if (i915_gem_object_can_migrate(obj, INTEL_REGION_LMEM))
+			err = -EINVAL;
+
+		i915_gem_object_unpin_pages(obj);
+	}
+	i915_gem_object_put(obj);
+
+	return err;
+}
+
+static int igt_lmem_create_migrate(void *arg)
+{
+	struct intel_gt *gt = arg;
+	struct drm_i915_private *i915 = gt->i915;
+	struct drm_i915_gem_object *obj;
+	struct i915_gem_ww_ctx ww;
+	int err = 0;
+
+	/* Switch object backing-store on create */
+	obj = i915_gem_object_create_shmem(i915, PAGE_SIZE);
+	if (IS_ERR(obj))
+		return PTR_ERR(obj);
+
+	for_i915_gem_ww(&ww, err, true) {
+		err = i915_gem_object_lock(obj, &ww);
+		if (err)
+			continue;
+
+		if (!i915_gem_object_can_migrate(obj, INTEL_REGION_LMEM)) {
+			err = -EINVAL;
+			continue;
+		}
+
+		err = i915_gem_object_migrate(obj, &ww, INTEL_REGION_LMEM);
+		if (err)
+			continue;
+
+		err = i915_gem_object_pin_pages(obj);
+		if (err)
+			continue;
+
+		if (i915_gem_object_can_migrate(obj, INTEL_REGION_SMEM))
+			err = -EINVAL;
+
+		i915_gem_object_unpin_pages(obj);
+	}
+	i915_gem_object_put(obj);
+
+	return err;
+}
+
+static int lmem_pages_migrate_one(struct i915_gem_ww_ctx *ww,
+				  struct drm_i915_gem_object *obj)
+{
+	int err;
+
+	err = i915_gem_object_lock(obj, ww);
+	if (err)
+		return err;
+
+	err = i915_gem_object_wait(obj,
+				   I915_WAIT_INTERRUPTIBLE |
+				   I915_WAIT_PRIORITY |
+				   I915_WAIT_ALL,
+				   MAX_SCHEDULE_TIMEOUT);
+	if (err)
+		return err;
+
+	if (i915_gem_object_is_lmem(obj)) {
+		if (!i915_gem_object_can_migrate(obj, INTEL_REGION_SMEM)) {
+			pr_err("object can't migrate to smem.\n");
+			return -EINVAL;
+		}
+
+		err = i915_gem_object_migrate(obj, ww, INTEL_REGION_SMEM);
+		if (err) {
+			pr_err("Object failed migration to smem\n");
+			if (err)
+				return err;
+		}
+
+		if (i915_gem_object_is_lmem(obj)) {
+			pr_err("object still backed by lmem\n");
+			err = -EINVAL;
+		}
+
+		if (!i915_gem_object_has_struct_page(obj)) {
+			pr_err("object not backed by struct page\n");
+			err = -EINVAL;
+		}
+
+	} else {
+		if (!i915_gem_object_can_migrate(obj, INTEL_REGION_LMEM)) {
+			pr_err("object can't migrate to lmem.\n");
+			return -EINVAL;
+		}
+
+		err = i915_gem_object_migrate(obj, ww, INTEL_REGION_LMEM);
+		if (err) {
+			pr_err("Object failed migration to lmem\n");
+			if (err)
+				return err;
+		}
+
+		if (i915_gem_object_has_struct_page(obj)) {
+			pr_err("object still backed by struct page\n");
+			err = -EINVAL;
+		}
+
+		if (!i915_gem_object_is_lmem(obj)) {
+			pr_err("object not backed by lmem\n");
+			err = -EINVAL;
+		}
+	}
+
+	return err;
+}
+
+static int igt_lmem_pages_migrate(void *arg)
+{
+	struct intel_gt *gt = arg;
+	struct drm_i915_private *i915 = gt->i915;
+	struct drm_i915_gem_object *obj;
+	struct i915_gem_ww_ctx ww;
+	struct i915_request *rq;
+	int err;
+	int i;
+
+	/* From LMEM to shmem and back again */
+
+	obj = i915_gem_object_create_lmem(i915, SZ_2M, 0);
+	if (IS_ERR(obj))
+		return PTR_ERR(obj);
+
+	err = i915_gem_object_lock(obj, NULL);
+	if (err)
+		goto out_put;
+
+	err = ____i915_gem_object_get_pages(obj);
+	if (err) {
+		i915_gem_object_unlock(obj);
+		goto out_put;
+	}
+
+	err = intel_context_migrate_clear(gt->migrate.context, NULL,
+					  obj->mm.pages->sgl, obj->cache_level,
+					  i915_gem_object_is_lmem(obj),
+					  0, &rq);
+	if (rq) {
+		dma_resv_add_excl_fence(obj->base.resv, &rq->fence);
+		i915_request_put(rq);
+	}
+	i915_gem_object_unlock(obj);
+	if (err)
+		goto out_put;
+
+	for (i = 1; i <= 4; ++i) {
+		for_i915_gem_ww(&ww, err, true) {
+			err = lmem_pages_migrate_one(&ww, obj);
+			if (err)
+				continue;
+
+			err = i915_gem_object_wait_migration(obj, true);
+			if (err)
+				continue;
+
+			err = intel_migrate_clear(&gt->migrate, &ww, NULL,
+						  obj->mm.pages->sgl,
+						  obj->cache_level,
+						  i915_gem_object_is_lmem(obj),
+						  0xdeadbeaf, &rq);
+			if (rq) {
+				dma_resv_add_excl_fence(obj->base.resv,
+							&rq->fence);
+				i915_request_put(rq);
+			}
+		}
+		if (err)
+			break;
+	}
+out_put:
+	i915_gem_object_put(obj);

drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c line 230

Quoted 10 lines
drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c new file +10 -0
+
+	return err;
+}
+
+int i915_gem_migrate_live_selftests(struct drm_i915_private *i915)
+{
+	static const struct i915_subtest tests[] = {
+		SUBTEST(igt_smem_create_migrate),
+		SUBTEST(igt_lmem_create_migrate),
+		SUBTEST(igt_lmem_pages_migrate),

Might be good to dirty the pages before the migration, and then verify they still match at the end?

Quoted 20 lines
drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c new file +7 -0
+	};
+
+	if (!HAS_LMEM(i915))
+		return 0;
+
+	return intel_gt_live_subtests(tests, &i915->gt);
+}
drivers/gpu/drm/i915/selftests/i915_live_selftests.h +1 -0
index a68197cf1044..e2fd1b61af71 100644
@@ -40,6 +40,7 @@ selftest(hugepages, i915_gem_huge_page_live_selftests)
  selftest(gem_contexts, i915_gem_context_live_selftests)
  selftest(gem_execbuf, i915_gem_execbuffer_live_selftests)
  selftest(client, i915_gem_client_blt_live_selftests)
+selftest(gem_migrate, i915_gem_migrate_live_selftests)
  selftest(reset, intel_reset_live_selftests)
  selftest(memory_region, intel_memory_region_live_selftests)
  selftest(hangcheck, intel_hangcheck_live_selftests)
#2 lore raw

Re: [PATCH v3 1/5] drm/i915/gem: Implement object migration

In reply to Thomas Hellström
Quoted 239 lines · Thomas Hellström

Introduce an interface to migrate objects between regions. This is primarily intended to migrate objects to LMEM for display and to SYSTEM for dma-buf, but might be reused in one form or another for performance-based migration.

v2:
- Verify that the memory region given as an id really exists. (Reported by Matthew Auld)
- Call i915_gem_object_{init,release}_memory_region() when switching region to handle also switching region lists. (Reported by Matthew Auld)
v3:
- Fix i915_gem_object_can_migrate() to return true if object is already in the correct region, even if the object ops doesn't have a migrate() callback.
- Update typo in commit message.
- Fix kerneldoc of i915_gem_object_wait_migration().

---
  drivers/gpu/drm/i915/gem/i915_gem_object.c    | 96 +++++++++++++++++++
  drivers/gpu/drm/i915/gem/i915_gem_object.h    | 12 +++
  .../gpu/drm/i915/gem/i915_gem_object_types.h  |  9 ++
  drivers/gpu/drm/i915/gem/i915_gem_ttm.c       | 69 +++++++++----
  drivers/gpu/drm/i915/gem/i915_gem_wait.c      | 19 ++++
  5 files changed, 188 insertions(+), 17 deletions(-)
drivers/gpu/drm/i915/gem/i915_gem_object.c +96 -0
index 07e8ff9a8aae..1c18be067b58 100644
@@ -513,6 +513,102 @@ bool i915_gem_object_has_iomem(const struct drm_i915_gem_object *obj)
  	return obj->mem_flags & I915_BO_FLAG_IOMEM;
  }
 
+/**
+ * i915_gem_object_can_migrate - Whether an object likely can be migrated
+ *
+ * @obj: The object to migrate
+ * @id: The region intended to migrate to
+ *
+ * Check whether the object backend supports migration to the
+ * given region. Note that pinning may affect the ability to migrate.
+ *
+ * Return: true if migration is possible, false otherwise.
+ */
+bool i915_gem_object_can_migrate(struct drm_i915_gem_object *obj,
+				 enum intel_region_id id)
+{
+	struct drm_i915_private *i915 = to_i915(obj->base.dev);
+	unsigned int num_allowed = obj->mm.n_placements;
+	struct intel_memory_region *mr;
+	unsigned int i;
+
+	GEM_BUG_ON(id >= INTEL_REGION_UNKNOWN);
+	GEM_BUG_ON(obj->mm.madv != I915_MADV_WILLNEED);
+
+	mr = i915->mm.regions[id];
+	if (!mr)
+		return false;
+
+	if (obj->mm.region == mr)
+		return true;
+
+	if (!i915_gem_object_evictable(obj))
+		return false;
+
+	if (!obj->ops->migrate)
+		return false;
+
+	if (!(obj->flags & I915_BO_ALLOC_USER))
+		return true;
+
+	if (num_allowed == 0)
+		return false;
+
+	for (i = 0; i < num_allowed; ++i) {
+		if (mr == obj->mm.placements[i])
+			return true;
+	}
+
+	return false;
+}
+
+/**
+ * i915_gem_object_migrate - Migrate an object to the desired region id
+ * @obj: The object to migrate.
+ * @ww: An optional struct i915_gem_ww_ctx. If NULL, the backend may
+ * not be successful in evicting other objects to make room for this object.
+ * @id: The region id to migrate to.
+ *
+ * Attempt to migrate the object to the desired memory region. The
+ * object backend must support migration and the object may not be
+ * pinned, (explicitly pinned pages or pinned vmas). The object must
+ * be locked.
+ * On successful completion, the object will have pages pointing to
+ * memory in the new region, but an async migration task may not have
+ * completed yet, and to accomplish that, i915_gem_object_wait_migration()
+ * must be called.
+ *
+ * Return: 0 on success. Negative error code on failure. In particular may
+ * return -ENXIO on lack of region space, -EDEADLK for deadlock avoidance
+ * if @ww is set, -EINTR or -ERESTARTSYS if signal pending, and
+ * -EBUSY if the object is pinned.
+ */
+int i915_gem_object_migrate(struct drm_i915_gem_object *obj,
+			    struct i915_gem_ww_ctx *ww,
+			    enum intel_region_id id)
+{
+	struct drm_i915_private *i915 = to_i915(obj->base.dev);
+	struct intel_memory_region *mr;
+
+	GEM_BUG_ON(id >= INTEL_REGION_UNKNOWN);
+	GEM_BUG_ON(obj->mm.madv != I915_MADV_WILLNEED);
+	assert_object_held(obj);
+
+	mr = i915->mm.regions[id];
+	GEM_BUG_ON(!mr);
+
+	if (obj->mm.region == mr)
+		return 0;
+
+	if (!i915_gem_object_evictable(obj))
+		return -EBUSY;
+
+	if (!obj->ops->migrate)
+		return -EOPNOTSUPP;
+
+	return obj->ops->migrate(obj, mr);
+}
+
  void i915_gem_init__objects(struct drm_i915_private *i915)
  {
  	INIT_WORK(&i915->mm.free_work, __i915_gem_free_work);
drivers/gpu/drm/i915/gem/i915_gem_object.h +12 -0
index ea3224a480c4..8cbd7a5334e2 100644
@@ -17,6 +17,8 @@
  #include "i915_gem_ww.h"
  #include "i915_vma_types.h"
 
+enum intel_region_id;
+
  /*
   * XXX: There is a prevalence of the assumption that we fit the
   * object's page count inside a 32bit _signed_ variable. Let's document
@@ -597,6 +599,16 @@ bool i915_gem_object_migratable(struct drm_i915_gem_object *obj);
 
  bool i915_gem_object_validates_to_lmem(struct drm_i915_gem_object *obj);
 
+int i915_gem_object_migrate(struct drm_i915_gem_object *obj,
+			    struct i915_gem_ww_ctx *ww,
+			    enum intel_region_id id);
+
+bool i915_gem_object_can_migrate(struct drm_i915_gem_object *obj,
+				 enum intel_region_id id);
+
+int i915_gem_object_wait_migration(struct drm_i915_gem_object *obj,
+				   unsigned int flags);
+
  #ifdef CONFIG_MMU_NOTIFIER
  static inline bool
  i915_gem_object_is_userptr(struct drm_i915_gem_object *obj)
drivers/gpu/drm/i915/gem/i915_gem_object_types.h +9 -0
index 441f913c87e6..ef3de2ae9723 100644
@@ -18,6 +18,7 @@
 
  struct drm_i915_gem_object;
  struct intel_fronbuffer;
+struct intel_memory_region;
 
  /*
   * struct i915_lut_handle tracks the fast lookups from handle to vma used
@@ -77,6 +78,14 @@ struct drm_i915_gem_object_ops {
  	 * delayed_free - Override the default delayed free implementation
  	 */
  	void (*delayed_free)(struct drm_i915_gem_object *obj);
+
+	/**
+	 * migrate - Migrate object to a different region either for
+	 * pinning or for as long as the object lock is held.
+	 */
+	int (*migrate)(struct drm_i915_gem_object *obj,
+		       struct intel_memory_region *mr);
+
  	void (*release)(struct drm_i915_gem_object *obj);
 
  	const struct vm_operations_struct *mmap_ops;
drivers/gpu/drm/i915/gem/i915_gem_ttm.c +6 -12
index c39d982c4fa6..8f89185b6507 100644
@@ -617,7 +617,8 @@ struct ttm_device_funcs *i915_ttm_driver(void)
  	return &i915_ttm_bo_driver;
  }
 
-static int i915_ttm_get_pages(struct drm_i915_gem_object *obj)
+static int __i915_ttm_get_pages(struct drm_i915_gem_object *obj,
+				struct ttm_placement *placement)
  {
  	struct ttm_buffer_object *bo = i915_gem_to_ttm(obj);
  	struct ttm_operation_ctx ctx = {
@@ -625,19 +626,12 @@ static int i915_ttm_get_pages(struct drm_i915_gem_object *obj)
  		.no_wait_gpu = false,
  	};
  	struct sg_table *st;
-	struct ttm_place requested, busy[I915_TTM_MAX_PLACEMENTS];
-	struct ttm_placement placement;
  	int real_num_busy;
  	int ret;
 
-	GEM_BUG_ON(obj->mm.n_placements > I915_TTM_MAX_PLACEMENTS);
-
-	/* Move to the requested placement. */
-	i915_ttm_placement_from_obj(obj, &requested, busy, &placement);
-
  	/* First try only the requested placement. No eviction. */
-	real_num_busy = fetch_and_zero(&placement.num_busy_placement);
-	ret = ttm_bo_validate(bo, &placement, &ctx);
+	real_num_busy = fetch_and_zero(&placement->num_busy_placement);
+	ret = ttm_bo_validate(bo, placement, &ctx);
  	if (ret) {
  		ret = i915_ttm_err_to_gem(ret);
  		/*
@@ -652,8 +646,8 @@ static int i915_ttm_get_pages(struct drm_i915_gem_object *obj)
  		 * If the initial attempt fails, allow all accepted placements,
  		 * evicting if necessary.
  		 */
-		placement.num_busy_placement = real_num_busy;
-		ret = ttm_bo_validate(bo, &placement, &ctx);
+		placement->num_busy_placement = real_num_busy;
+		ret = ttm_bo_validate(bo, placement, &ctx);
  		if (ret)
  			return i915_ttm_err_to_gem(ret);
  	}

drivers/gpu/drm/i915/gem/i915_gem_ttm.c static int i915_ttm_get_pages(struct drm_i915_gem_object *obj) · line 665

Quoted 8 lines
drivers/gpu/drm/i915/gem/i915_gem_ttm.c +1 -4
@@ -668,16 +662,56 @@ static int i915_ttm_get_pages(struct drm_i915_gem_object *obj)
  		i915_ttm_adjust_gem_after_move(obj);
  	}
 
-	/* Object either has a page vector or is an iomem object */
-	st = bo->ttm ? i915_ttm_tt_get_st(bo->ttm) : obj->ttm.cached_io_st;
-	if (IS_ERR(st))
-		return PTR_ERR(st);
+	if (!obj->mm.pages) {

Is this for the migration case? Where are we updating the mm.pages, assuming the migration step has to take care of pre-existing pages? What am I missing?

Quoted 35 lines
drivers/gpu/drm/i915/gem/i915_gem_ttm.c +29 -1
+		/* Object either has a page vector or is an iomem object */
+		st = bo->ttm ? i915_ttm_tt_get_st(bo->ttm) : obj->ttm.cached_io_st;
+		if (IS_ERR(st))
+			return PTR_ERR(st);
 
-	__i915_gem_object_set_pages(obj, st, i915_sg_dma_sizes(st->sgl));
+		__i915_gem_object_set_pages(obj, st, i915_sg_dma_sizes(st->sgl));
+	}
 
  	return ret;
  }
 
+static int i915_ttm_get_pages(struct drm_i915_gem_object *obj)
+{
+	struct ttm_place requested, busy[I915_TTM_MAX_PLACEMENTS];
+	struct ttm_placement placement;
+
+	GEM_BUG_ON(obj->mm.n_placements > I915_TTM_MAX_PLACEMENTS);
+
+	/* Move to the requested placement. */
+	i915_ttm_placement_from_obj(obj, &requested, busy, &placement);
+
+	return __i915_ttm_get_pages(obj, &placement);
+}
+
+static int i915_ttm_migrate(struct drm_i915_gem_object *obj,
+			    struct intel_memory_region *mr)
+{
+	struct ttm_place requested;
+	struct ttm_placement placement;
+	int ret;
+
+	i915_ttm_place_from_region(mr, &requested, obj->flags);
+	placement.num_placement = 1;
+	placement.num_busy_placement = 1;

drivers/gpu/drm/i915/gem/i915_gem_ttm.c static int i915_ttm_get_pages(struct drm_i915_gem_object *obj) · line 709

Quoted 10 lines
drivers/gpu/drm/i915/gem/i915_gem_ttm.c +10 -0
+	placement.placement = &requested;
+	placement.busy_placement = &requested;
+
+	ret = __i915_ttm_get_pages(obj, &placement);
+	if (ret)
+		return ret;
+
+	if (obj->mm.region != mr) {
+		i915_gem_object_release_memory_region(obj);
+		i915_gem_object_init_memory_region(obj, mr);

Hmm, so is this not already handled by adjust_gem_after_move()?

Quoted 34 lines
drivers/gpu/drm/i915/gem/i915_gem_ttm.c +6 -0
+	}
+
+	return 0;
+}
+
  static void i915_ttm_put_pages(struct drm_i915_gem_object *obj,
  			       struct sg_table *st)
  {
@@ -814,6 +848,7 @@ static const struct drm_i915_gem_object_ops i915_gem_ttm_obj_ops = {
  	.truncate = i915_ttm_purge,
  	.adjust_lru = i915_ttm_adjust_lru,
  	.delayed_free = i915_ttm_delayed_free,
+	.migrate = i915_ttm_migrate,
  	.mmap_offset = i915_ttm_mmap_offset,
  	.mmap_ops = &vm_ops_ttm,
  };
drivers/gpu/drm/i915/gem/i915_gem_wait.c +9 -0
index 1070d3afdce7..f909aaa09d9c 100644
@@ -290,3 +290,22 @@ i915_gem_wait_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
  	i915_gem_object_put(obj);
  	return ret;
  }
+
+/**
+ * i915_gem_object_wait_migration - Sync an accelerated migration operation
+ * @obj: The migrating object.
+ * @flags: waiting flags. Currently supports only I915_WAIT_INTERRUPTIBLE.
+ *
+ * Wait for any pending async migration operation on the object,
+ * whether it's explicitly (i915_gem_object_migrate()) or implicitly
+ * (swapin, initial clearing) initiated.

drivers/gpu/drm/i915/gem/i915_gem_wait.c i915_gem_wait_ioctl(struct drm_device *dev, void *data, struct drm_file *file) · line 311

Quoted 10 lines
drivers/gpu/drm/i915/gem/i915_gem_wait.c +10 -0
+ *
+ * Return: 0 if successful, -ERESTARTSYS if a signal was hit during waiting.
+ */
+int i915_gem_object_wait_migration(struct drm_i915_gem_object *obj,
+				   unsigned int flags)
+{
+	might_sleep();
+	/* NOP for now. */
+	return 0;
+}

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#25 lore raw

[Intel-gfx] ✗ Fi.CI.CHECKPATCH: warning for drm/i915/gem: Introduce a migrate interface (rev3)

In reply to Thomas Hellström
== Series Details ==

== Series Details ==

Series: drm/i915/gem: Introduce a migrate interface (rev3)
URL : https://patchwork.freedesktop.org/series/91890/
State : warning

== Summary ==

$ dim checkpatch origin/drm-tip
673d9747bd21 drm/i915/gem: Implement object migration
fb5a2bfc3766 drm/i915/gem: Introduce a selftest for the gem object migrate functionality
-:31: WARNING:FILE_PATH_CHANGES: added, moved or deleted file(s), does MAINTAINERS need updating?
#31: 
new file mode 100644

total: 0 errors, 1 warnings, 0 checks, 251 lines checked 65e3941fa259 drm/i915/display: Migrate objects to LMEM if possible for display 00ac06b936ba drm/i915/gem: Fix same-driver-another-instance dma-buf export effbc92a43a9 drm/i915/gem: Migrate to system at dma-buf map time

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#26 lore raw

[Intel-gfx] ✗ Fi.CI.SPARSE: warning for drm/i915/gem: Introduce a migrate interface (rev3)

In reply to Thomas Hellström
== Series Details ==

== Series Details ==

Series: drm/i915/gem: Introduce a migrate interface (rev3)
URL : https://patchwork.freedesktop.org/series/91890/
State : warning

== Summary ==

$ dim sparse --fast origin/drm-tip
Sparse version: v0.6.2
Fast mode used, each commit won't be checked separately.
-
+drivers/gpu/drm/i915/display/intel_display.c:1896:21:    expected struct i915_vma *[assigned] vma
+drivers/gpu/drm/i915/display/intel_display.c:1896:21:    got void [noderef] __iomem *[assigned] iomem
+drivers/gpu/drm/i915/display/intel_display.c:1896:21: warning: incorrect type in assignment (different address spaces)
+drivers/gpu/drm/i915/gt/intel_engine_stats.h:27:9: warning: trying to copy expression type 31
+drivers/gpu/drm/i915/gt/intel_engine_stats.h:27:9: warning: trying to copy expression type 31
+drivers/gpu/drm/i915/gt/intel_engine_stats.h:27:9: warning: trying to copy expression type 31
+drivers/gpu/drm/i915/gt/intel_engine_stats.h:32:9: warning: trying to copy expression type 31
+drivers/gpu/drm/i915/gt/intel_engine_stats.h:32:9: warning: trying to copy expression type 31
+drivers/gpu/drm/i915/gt/intel_engine_stats.h:49:9: warning: trying to copy expression type 31
+drivers/gpu/drm/i915/gt/intel_engine_stats.h:49:9: warning: trying to copy expression type 31
+drivers/gpu/drm/i915/gt/intel_engine_stats.h:49:9: warning: trying to copy expression type 31
+drivers/gpu/drm/i915/gt/intel_engine_stats.h:56:9: warning: trying to copy expression type 31
+drivers/gpu/drm/i915/gt/intel_engine_stats.h:56:9: warning: trying to copy expression type 31
+drivers/gpu/drm/i915/gt/intel_reset.c:1396:5: warning: context imbalance in 'intel_gt_reset_trylock' - different lock contexts for basic block
+drivers/gpu/drm/i915/gt/intel_ring_submission.c:1207:24: warning: Using plain integer as NULL pointer
+drivers/gpu/drm/i915/i915_perf.c:1434:15: warning: memset with byte count of 16777216
+drivers/gpu/drm/i915/i915_perf.c:1488:15: warning: memset with byte count of 16777216
+drivers/gpu/drm/i915/selftests/i915_syncmap.c:80:54: warning: dubious: x | !y
+./include/asm-generic/bitops/find.h:112:45: warning: shift count is negative (-262080)
+./include/asm-generic/bitops/find.h:32:31: warning: shift count is negative (-262080)
+./include/linux/spinlock.h:409:9: warning: context imbalance in 'fwtable_read16' - different lock contexts for basic block
+./include/linux/spinlock.h:409:9: warning: context imbalance in 'fwtable_read32' - different lock contexts for basic block
+./include/linux/spinlock.h:409:9: warning: context imbalance in 'fwtable_read64' - different lock contexts for basic block
+./include/linux/spinlock.h:409:9: warning: context imbalance in 'fwtable_read8' - different lock contexts for basic block
+./include/linux/spinlock.h:409:9: warning: context imbalance in 'fwtable_write16' - different lock contexts for basic block
+./include/linux/spinlock.h:409:9: warning: context imbalance in 'fwtable_write32' - different lock contexts for basic block
+./include/linux/spinlock.h:409:9: warning: context imbalance in 'fwtable_write8' - different lock contexts for basic block
+./include/linux/spinlock.h:409:9: warning: context imbalance in 'gen11_fwtable_read16' - different lock contexts for basic block
+./include/linux/spinlock.h:409:9: warning: context imbalance in 'gen11_fwtable_read32' - different lock contexts for basic block
+./include/linux/spinlock.h:409:9: warning: context imbalance in 'gen11_fwtable_read64' - different lock contexts for basic block
+./include/linux/spinlock.h:409:9: warning: context imbalance in 'gen11_fwtable_read8' - different lock contexts for basic block
+./include/linux/spinlock.h:409:9: warning: context imbalance in 'gen11_fwtable_write16' - different lock contexts for basic block
+./include/linux/spinlock.h:409:9: warning: context imbalance in 'gen11_fwtable_write32' - different lock contexts for basic block
+./include/linux/spinlock.h:409:9: warning: context imbalance in 'gen11_fwtable_write8' - different lock contexts for basic block
+./include/linux/spinlock.h:409:9: warning: context imbalance in 'gen12_fwtable_read16' - different lock contexts for basic block
+./include/linux/spinlock.h:409:9: warning: context imbalance in 'gen12_fwtable_read32' - different lock contexts for basic block
+./include/linux/spinlock.h:409:9: warning: context imbalance in 'gen12_fwtable_read64' - different lock contexts for basic block
+./include/linux/spinlock.h:409:9: warning: context imbalance in 'gen12_fwtable_read8' - different lock contexts for basic block
+./include/linux/spinlock.h:409:9: warning: context imbalance in 'gen12_fwtable_write16' - different lock contexts for basic block
+./include/linux/spinlock.h:409:9: warning: context imbalance in 'gen12_fwtable_write32' - different lock contexts for basic block
+./include/linux/spinlock.h:409:9: warning: context imbalance in 'gen12_fwtable_write8' - different lock contexts for basic block
+./include/linux/spinlock.h:409:9: warning: context imbalance in 'gen6_read16' - different lock contexts for basic block
+./include/linux/spinlock.h:409:9: warning: context imbalance in 'gen6_read32' - different lock contexts for basic block
+./include/linux/spinlock.h:409:9: warning: context imbalance in 'gen6_read64' - different lock contexts for basic block
+./include/linux/spinlock.h:409:9: warning: context imbalance in 'gen6_read8' - different lock contexts for basic block
+./include/linux/spinlock.h:409:9: warning: context imbalance in 'gen6_write16' - different lock contexts for basic block
+./include/linux/spinlock.h:409:9: warning: context imbalance in 'gen6_write32' - different lock contexts for basic block
+./include/linux/spinlock.h:409:9: warning: context imbalance in 'gen6_write8' - different lock contexts for basic block
+./include/linux/spinlock.h:409:9: warning: context imbalance in 'gen8_write16' - different lock contexts for basic block
+./include/linux/spinlock.h:409:9: warning: context imbalance in 'gen8_write32' - different lock contexts for basic block
+./include/linux/spinlock.h:409:9: warning: context imbalance in 'gen8_write8' - different lock contexts for basic block

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#27 lore raw

[Intel-gfx] ✓ Fi.CI.BAT: success for drm/i915/gem: Introduce a migrate interface (rev3)

In reply to Thomas Hellström
== Series Details ==

== Series Details ==

Series: drm/i915/gem: Introduce a migrate interface (rev3)
URL : https://patchwork.freedesktop.org/series/91890/
State : success

== Summary ==

CI Bug Log - changes from CI_DRM_10284 -> Patchwork_20479
====================================================

Summary
-------

**SUCCESS**

No regressions found.

External URL: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/index.html

New tests
---------

New tests have been introduced between CI_DRM_10284 and Patchwork_20479:

### New IGT tests (1) ###

* igt@i915_selftest@live@gem_migrate:
- Statuses : 33 pass(s)
- Exec time: [0.43, 5.28] s

Known issues
------------

Here are the changes found in Patchwork_20479 that come from known issues:

### IGT changes ###

#### Issues hit ####

* igt@amdgpu/amd_cs_nop@sync-fork-compute0:
- fi-snb-2600: NOTRUN -> [SKIP][1] ([fdo#109271]) +17 similar issues
[1]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/fi-snb-2600/igt@amdgpu/[email protected]

* igt@i915_selftest@live@execlists:
- fi-bsw-nick: [PASS][2] -> [INCOMPLETE][3] ([i915#2782] / [i915#2940])
[2]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10284/fi-bsw-nick/igt@i915_selftest@[email protected]
[3]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/fi-bsw-nick/igt@i915_selftest@[email protected]

* igt@runner@aborted:
- fi-bsw-nick: NOTRUN -> [FAIL][4] ([fdo#109271] / [i915#1436])
[4]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/fi-bsw-nick/igt@[email protected]

#### Possible fixes ####

* igt@i915_selftest@live@hangcheck:
- fi-snb-2600: [INCOMPLETE][5] ([i915#2782]) -> [PASS][6]
[5]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10284/fi-snb-2600/igt@i915_selftest@[email protected]
[6]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/fi-snb-2600/igt@i915_selftest@[email protected]

[fdo#109271]: https://bugs.freedesktop.org/show_bug.cgi?id=109271
[i915#1436]: https://gitlab.freedesktop.org/drm/intel/issues/1436
[i915#2782]: https://gitlab.freedesktop.org/drm/intel/issues/2782
[i915#2940]: https://gitlab.freedesktop.org/drm/intel/issues/2940

Participating hosts (40 -> 36)
------------------------------

Missing (4): fi-bsw-cyan fi-hsw-4200u fi-bdw-samus fi-bsw-n3050

Build changes
-------------

* Linux: CI_DRM_10284 -> Patchwork_20479

CI-20190529: 20190529
CI_DRM_10284: c4cc46175b38e554eab7ac3d1b95a85c79963d4a @ git://anongit.freedesktop.org/gfx-ci/linux
IGT_6120: c45c6b727c1efaced0b53620bd41c8e4facfb31f @ https://gitlab.freedesktop.org/drm/igt-gpu-tools.git
Patchwork_20479: effbc92a43a968b385dca5695e50487386c8d33c @ git://anongit.freedesktop.org/gfx-ci/linux

== Linux commits ==

effbc92a43a9 drm/i915/gem: Migrate to system at dma-buf map time 00ac06b936ba drm/i915/gem: Fix same-driver-another-instance dma-buf export 65e3941fa259 drm/i915/display: Migrate objects to LMEM if possible for display fb5a2bfc3766 drm/i915/gem: Introduce a selftest for the gem object migrate functionality 673d9747bd21 drm/i915/gem: Implement object migration

== Logs ==

For more details see: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/index.html

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#3 lore raw

Re: [PATCH v3 1/5] drm/i915/gem: Implement object migration

In reply to Matthew Auld
Quoted 244 lines · Matthew Auld
Quoted 243 lines · Thomas Hellström

Introduce an interface to migrate objects between regions. This is primarily intended to migrate objects to LMEM for display and to SYSTEM for dma-buf, but might be reused in one form or another for performance-based migration.

v2:
- Verify that the memory region given as an id really exists. (Reported by Matthew Auld)
- Call i915_gem_object_{init,release}_memory_region() when switching region
   to handle also switching region lists. (Reported by Matthew Auld) v3:
- Fix i915_gem_object_can_migrate() to return true if object is already in
   the correct region, even if the object ops doesn't have a migrate() callback.
- Update typo in commit message.
- Fix kerneldoc of i915_gem_object_wait_migration().

---
  drivers/gpu/drm/i915/gem/i915_gem_object.c    | 96 +++++++++++++++++++ drivers/gpu/drm/i915/gem/i915_gem_object.h    | 12 +++ .../gpu/drm/i915/gem/i915_gem_object_types.h  |  9 ++ drivers/gpu/drm/i915/gem/i915_gem_ttm.c       | 69 +++++++++---- drivers/gpu/drm/i915/gem/i915_gem_wait.c      | 19 ++++

  5 files changed, 188 insertions(+), 17 deletions(-)
drivers/gpu/drm/i915/gem/i915_gem_object.c +0 -0
index 07e8ff9a8aae..1c18be067b58 100644
@@ -513,6 +513,102 @@ bool i915_gem_object_has_iomem(const struct 
 drm_i915_gem_object *obj)
       return obj->mem_flags & I915_BO_FLAG_IOMEM;
   }

  +/**
+ * i915_gem_object_can_migrate - Whether an object likely can be migrated
+ *
+ * @obj: The object to migrate
+ * @id: The region intended to migrate to
+ *
+ * Check whether the object backend supports migration to the
+ * given region. Note that pinning may affect the ability to migrate.
+ *
+ * Return: true if migration is possible, false otherwise.
+ */

drivers/gpu/drm/i915/gem/i915_gem_object.c +43 -0
+bool i915_gem_object_can_migrate(struct drm_i915_gem_object *obj,
+                 enum intel_region_id id)
+{
+    struct drm_i915_private *i915 = to_i915(obj->base.dev);
+    unsigned int num_allowed = obj->mm.n_placements;
+    struct intel_memory_region *mr;
+    unsigned int i;
+
+    GEM_BUG_ON(id >= INTEL_REGION_UNKNOWN);
+    GEM_BUG_ON(obj->mm.madv != I915_MADV_WILLNEED);
+
+    mr = i915->mm.regions[id];
+    if (!mr)
+        return false;
+
+    if (obj->mm.region == mr)
+        return true;
+
+    if (!i915_gem_object_evictable(obj))
+        return false;
+
+    if (!obj->ops->migrate)
+        return false;
+
+    if (!(obj->flags & I915_BO_ALLOC_USER))
+        return true;
+
+    if (num_allowed == 0)
+        return false;
+
+    for (i = 0; i < num_allowed; ++i) {
+        if (mr == obj->mm.placements[i])
+            return true;
+    }
+
+    return false;
+}
+
+/**
+ * i915_gem_object_migrate - Migrate an object to the desired region id
+ * @obj: The object to migrate.
+ * @ww: An optional struct i915_gem_ww_ctx. If NULL, the backend may
+ * not be successful in evicting other objects to make room for this 

object.
+ * @id: The region id to migrate to.
+ *
+ * Attempt to migrate the object to the desired memory region. The
+ * object backend must support migration and the object may not be
+ * pinned, (explicitly pinned pages or pinned vmas). The object must
+ * be locked.
+ * On successful completion, the object will have pages pointing to
+ * memory in the new region, but an async migration task may not have
+ * completed yet, and to accomplish that, i915_gem_object_wait_migration()
+ * must be called.
+ *
+ * Return: 0 on success. Negative error code on failure. In particular may
+ * return -ENXIO on lack of region space, -EDEADLK for deadlock avoidance
+ * if @ww is set, -EINTR or -ERESTARTSYS if signal pending, and
+ * -EBUSY if the object is pinned.
+ */

drivers/gpu/drm/i915/gem/i915_gem_object.c +26 -0
+int i915_gem_object_migrate(struct drm_i915_gem_object *obj,
+                struct i915_gem_ww_ctx *ww,
+                enum intel_region_id id)
+{
+    struct drm_i915_private *i915 = to_i915(obj->base.dev);
+    struct intel_memory_region *mr;
+
+    GEM_BUG_ON(id >= INTEL_REGION_UNKNOWN);
+    GEM_BUG_ON(obj->mm.madv != I915_MADV_WILLNEED);
+    assert_object_held(obj);
+
+    mr = i915->mm.regions[id];
+    GEM_BUG_ON(!mr);
+
+    if (obj->mm.region == mr)
+        return 0;
+
+    if (!i915_gem_object_evictable(obj))
+        return -EBUSY;
+
+    if (!obj->ops->migrate)
+        return -EOPNOTSUPP;
+
+    return obj->ops->migrate(obj, mr);
+}
+

  void i915_gem_init__objects(struct drm_i915_private *i915) {
      INIT_WORK(&i915->mm.free_work, __i915_gem_free_work);

drivers/gpu/drm/i915/gem/i915_gem_object.h +10 -0
index ea3224a480c4..8cbd7a5334e2 100644
@@ -17,6 +17,8 @@
   #include "i915_gem_ww.h"
   #include "i915_vma_types.h"
   +enum intel_region_id;
+
   /*
    * XXX: There is a prevalence of the assumption that we fit the
    * object's page count inside a 32bit _signed_ variable. Let's document
@@ -597,6 +599,16 @@ bool i915_gem_object_migratable(struct 
 drm_i915_gem_object *obj);
     bool i915_gem_object_validates_to_lmem(struct drm_i915_gem_object *obj);
   +int i915_gem_object_migrate(struct drm_i915_gem_object *obj,
+                struct i915_gem_ww_ctx *ww,
+                enum intel_region_id id);
+
+bool i915_gem_object_can_migrate(struct drm_i915_gem_object *obj,
+                 enum intel_region_id id);
+
+int i915_gem_object_wait_migration(struct drm_i915_gem_object *obj,
+                   unsigned int flags);
+
   #ifdef CONFIG_MMU_NOTIFIER

  static inline bool i915_gem_object_is_userptr(struct drm_i915_gem_object *obj)

drivers/gpu/drm/i915/gem/i915_gem_object_types.h +9 -0
index 441f913c87e6..ef3de2ae9723 100644
@@ -18,6 +18,7 @@
     struct drm_i915_gem_object;
   struct intel_fronbuffer;
+struct intel_memory_region;
     /*
    * struct i915_lut_handle tracks the fast lookups from handle to vma used
@@ -77,6 +78,14 @@ struct drm_i915_gem_object_ops {
        * delayed_free - Override the default delayed free implementation
        */
       void (*delayed_free)(struct drm_i915_gem_object *obj);
+
+    /**
+     * migrate - Migrate object to a different region either for
+     * pinning or for as long as the object lock is held.
+     */
+    int (*migrate)(struct drm_i915_gem_object *obj,
+               struct intel_memory_region *mr);
+
       void (*release)(struct drm_i915_gem_object *obj);
         const struct vm_operations_struct *mmap_ops;
drivers/gpu/drm/i915/gem/i915_gem_ttm.c +4 -8
index c39d982c4fa6..8f89185b6507 100644
@@ -617,7 +617,8 @@ struct ttm_device_funcs *i915_ttm_driver(void)
       return &i915_ttm_bo_driver;
   }
   -static int i915_ttm_get_pages(struct drm_i915_gem_object *obj)
+static int __i915_ttm_get_pages(struct drm_i915_gem_object *obj,
+                struct ttm_placement *placement)
   {
       struct ttm_buffer_object *bo = i915_gem_to_ttm(obj);
       struct ttm_operation_ctx ctx = {
@@ -625,19 +626,12 @@ static int i915_ttm_get_pages(struct 
 drm_i915_gem_object *obj)
           .no_wait_gpu = false,
       };
       struct sg_table *st;
-    struct ttm_place requested, busy[I915_TTM_MAX_PLACEMENTS];
-    struct ttm_placement placement;
       int real_num_busy;
       int ret;
   -    GEM_BUG_ON(obj->mm.n_placements > I915_TTM_MAX_PLACEMENTS);
-
-    /* Move to the requested placement. */
-    i915_ttm_placement_from_obj(obj, &requested, busy, &placement);
-
       /* First try only the requested placement. No eviction. */
-    real_num_busy = fetch_and_zero(&placement.num_busy_placement);
-    ret = ttm_bo_validate(bo, &placement, &ctx);
+    real_num_busy = fetch_and_zero(&placement->num_busy_placement);
+    ret = ttm_bo_validate(bo, placement, &ctx);
       if (ret) {
           ret = i915_ttm_err_to_gem(ret);

          /*

drivers/gpu/drm/i915/gem/i915_gem_ttm.c +2 -2
@@ -652,8 +646,8 @@ static int i915_ttm_get_pages(struct 
 drm_i915_gem_object *obj)
            * If the initial attempt fails, allow all accepted placements,
            * evicting if necessary.
            */
-        placement.num_busy_placement = real_num_busy;
-        ret = ttm_bo_validate(bo, &placement, &ctx);
+        placement->num_busy_placement = real_num_busy;
+        ret = ttm_bo_validate(bo, placement, &ctx);
           if (ret)
               return i915_ttm_err_to_gem(ret);

      }

drivers/gpu/drm/i915/gem/i915_gem_ttm.c +0 -0
@@ -668,16 +662,56 @@ static int i915_ttm_get_pages(struct 
 drm_i915_gem_object *obj)
           i915_ttm_adjust_gem_after_move(obj);
Quoted 8 lines
Quoted 6 lines
drivers/gpu/drm/i915/gem/i915_gem_ttm.c +0 -0
       }

  -    /* Object either has a page vector or is an iomem object */ -    st = bo->ttm ? i915_ttm_tt_get_st(bo->ttm) : obj->ttm.cached_io_st; -    if (IS_ERR(st))
-        return PTR_ERR(st);
+    if (!obj->mm.pages) {

Is this for the migration case? Where are we updating the mm.pages, assuming the migration step has to take care of pre-existing pages? What am I missing?

Hmm, yes this is for a hypotetical migration case where ttm_bo_validate() decided to not change placement of the object, meaning we migrate to the very same region that we're already in. That should be blocked at the gem level, so I guess we could replace this with a GEM_BUG_ON().

Quoted 39 lines
Quoted 39 lines

+        /* Object either has a page vector or is an iomem object */ +        st = bo->ttm ? i915_ttm_tt_get_st(bo->ttm) :
obj->ttm.cached_io_st;
+        if (IS_ERR(st))
+            return PTR_ERR(st);
  -    __i915_gem_object_set_pages(obj, st, i915_sg_dma_sizes(st->sgl)); +        __i915_gem_object_set_pages(obj, st, i915_sg_dma_sizes(st->sgl));
+    }
        return ret;
  }
  +static int i915_ttm_get_pages(struct drm_i915_gem_object *obj)

drivers/gpu/drm/i915/gem/i915_gem_ttm.c +26 -0
+{
+    struct ttm_place requested, busy[I915_TTM_MAX_PLACEMENTS];
+    struct ttm_placement placement;
+
+    GEM_BUG_ON(obj->mm.n_placements > I915_TTM_MAX_PLACEMENTS);
+
+    /* Move to the requested placement. */
+    i915_ttm_placement_from_obj(obj, &requested, busy, &placement);
+
+    return __i915_ttm_get_pages(obj, &placement);
+}
+
+static int i915_ttm_migrate(struct drm_i915_gem_object *obj,
+                struct intel_memory_region *mr)
+{
+    struct ttm_place requested;
+    struct ttm_placement placement;
+    int ret;
+
+    i915_ttm_place_from_region(mr, &requested, obj->flags);
+    placement.num_placement = 1;
+    placement.num_busy_placement = 1;
+    placement.placement = &requested;
+    placement.busy_placement = &requested;
+
+    ret = __i915_ttm_get_pages(obj, &placement);
Quoted 8 lines
Quoted 6 lines

+    if (ret)
+        return ret;
+
+    if (obj->mm.region != mr) {
+        i915_gem_object_release_memory_region(obj); +        i915_gem_object_init_memory_region(obj, mr);

Hmm, so is this not already handled by adjust_gem_after_move()?

No, not always. adjust_gem_after_move() only considers an object migrated and flips region if the new region is in one of the allowed placements. Otherwise it's considered evicted and the next get_pages() will revalidate into the preferred placement.

the gem_migrate interface is more flexible, and offers a possibility to migrate to whatever it's told, so as long as we want that behaviour, we need the above.

/Thomas

Quoted 42 lines
Quoted 42 lines

+    }

drivers/gpu/drm/i915/gem/i915_gem_ttm.c +4 -0
+
+    return 0;
+}
+

  static void i915_ttm_put_pages(struct drm_i915_gem_object *obj, struct sg_table *st)
  {

drivers/gpu/drm/i915/gem/i915_gem_ttm.c +1 -0
@@ -814,6 +848,7 @@ static const struct drm_i915_gem_object_ops 
 i915_gem_ttm_obj_ops = {
       .truncate = i915_ttm_purge,
       .adjust_lru = i915_ttm_adjust_lru,
       .delayed_free = i915_ttm_delayed_free,
+    .migrate = i915_ttm_migrate,
       .mmap_offset = i915_ttm_mmap_offset,
       .mmap_ops = &vm_ops_ttm,

  };

drivers/gpu/drm/i915/gem/i915_gem_wait.c +0 -0
index 1070d3afdce7..f909aaa09d9c 100644
@@ -290,3 +290,22 @@ i915_gem_wait_ioctl(struct drm_device *dev, void 
 *data, struct drm_file *file)
       i915_gem_object_put(obj);
       return ret;

  }

drivers/gpu/drm/i915/gem/i915_gem_wait.c +3 -0
+
+/**
+ * i915_gem_object_wait_migration - Sync an accelerated migration 

operation
+ * @obj: The migrating object.
+ * @flags: waiting flags. Currently supports only I915_WAIT_INTERRUPTIBLE.
+ *
+ * Wait for any pending async migration operation on the object,
+ * whether it's explicitly (i915_gem_object_migrate()) or implicitly
+ * (swapin, initial clearing) initiated.
+ *
+ * Return: 0 if successful, -ERESTARTSYS if a signal was hit during waiting.
+ */

Quoted 8 lines

drivers/gpu/drm/i915/gem/i915_gem_wait.c i915_gem_wait_ioctl(struct drm_device *dev, void *data, struct drm_file *file) · line 311

Quoted 8 lines
drivers/gpu/drm/i915/gem/i915_gem_wait.c +7 -0
+int i915_gem_object_wait_migration(struct drm_i915_gem_object *obj,
+                   unsigned int flags)
+{
+    might_sleep();
+    /* NOP for now. */
+    return 0;
+}

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#4 lore raw

RE: [PATCH v3 1/5] drm/i915/gem: Implement object migration

In reply to Thomas Hellström
Quoted 84 lines

-----Original Message-----
From: dri-devel <[email protected]> On Behalf Of Thomas Hellström
Sent: Monday, June 28, 2021 10:46 AM
To: [email protected]; [email protected]
Cc: Thomas Hellström <[email protected]>; Auld, Matthew <[email protected]>; lkp <[email protected]>
Subject: [PATCH v3 1/5] drm/i915/gem: Implement object migration

Introduce an interface to migrate objects between regions. This is primarily intended to migrate objects to LMEM for display and to SYSTEM for dma-buf, but might be reused in one form or another for performance-based migration.

v2:
- Verify that the memory region given as an id really exists. (Reported by Matthew Auld)
- Call i915_gem_object_{init,release}_memory_region() when switching region
to handle also switching region lists. (Reported by Matthew Auld)
v3:
- Fix i915_gem_object_can_migrate() to return true if object is already in the correct region, even if the object ops doesn't have a migrate() callback.
- Update typo in commit message.
- Fix kerneldoc of i915_gem_object_wait_migration().

---
drivers/gpu/drm/i915/gem/i915_gem_object.c | 96
+++++++++++++++++++

drivers/gpu/drm/i915/gem/i915_gem_object.h    | 12 +++
.../gpu/drm/i915/gem/i915_gem_object_types.h  |  9 ++
drivers/gpu/drm/i915/gem/i915_gem_ttm.c       | 69 +++++++++----
drivers/gpu/drm/i915/gem/i915_gem_wait.c      | 19 ++++
5 files changed, 188 insertions(+), 17 deletions(-)
drivers/gpu/drm/i915/gem/i915_gem_object.c +35 -0
index 07e8ff9a8aae..1c18be067b58 100644
@@ -513,6 +513,102 @@ bool i915_gem_object_has_iomem(const struct
 drm_i915_gem_object *obj)
 	return obj->mem_flags & I915_BO_FLAG_IOMEM;
 }
 
+/**
+ * i915_gem_object_can_migrate - Whether an object likely can be migrated
+ *
+ * @obj: The object to migrate
+ * @id: The region intended to migrate to
+ *
+ * Check whether the object backend supports migration to the
+ * given region. Note that pinning may affect the ability to migrate.
+ *
+ * Return: true if migration is possible, false otherwise.
+ */
+bool i915_gem_object_can_migrate(struct drm_i915_gem_object *obj,
+				 enum intel_region_id id)
+{
+	struct drm_i915_private *i915 = to_i915(obj->base.dev);
+	unsigned int num_allowed = obj->mm.n_placements;
+	struct intel_memory_region *mr;
+	unsigned int i;
+
+	GEM_BUG_ON(id >= INTEL_REGION_UNKNOWN);
+	GEM_BUG_ON(obj->mm.madv != I915_MADV_WILLNEED);
+
+	mr = i915->mm.regions[id];
+	if (!mr)
+		return false;
+
+	if (obj->mm.region == mr)
+		return true;
+
+	if (!i915_gem_object_evictable(obj))
+		return false;
+
+	if (!obj->ops->migrate)
+		return false;
+

drivers/gpu/drm/i915/gem/i915_gem_object.c bool i915_gem_object_has_iomem(const struct · line 561

Quoted 10 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +10 -0
+	if (!(obj->flags & I915_BO_ALLOC_USER))
+		return true;
+
+	if (num_allowed == 0)
+		return false;
+
+	for (i = 0; i < num_allowed; ++i) {
+		if (mr == obj->mm.placements[i])
+			return true;
+	}

Hi Thomas,

I am a little confused over the difference in checks between this function and i915_gem_object_migrate().

Why is the lack of an mr a BUG_ON in _object_migrate(), but here it just false?

So that means that under certain circumstances, you could not have a mr?

If that is the case, when?

Would that be when the I915_BO_ALLOC_USER is set?

If so, should there be a check for "non" user vs user?

Or maybe document this function pointing out why there are differences and why?

drivers/gpu/drm/i915/gem/i915_gem_object.c bool i915_gem_object_has_iomem(const struct drm_i915_gem_object *obj) · line 569

Quoted 8 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +8 -0
+	return false;
+}
+
+/**
+ * i915_gem_object_migrate - Migrate an object to the desired region id
+ * @obj: The object to migrate.
+ * @ww: An optional struct i915_gem_ww_ctx. If NULL, the backend may
+ * not be successful in evicting other objects to make room for this object.

Is the ww for future consideration? (I don't see any use of it in the patch).

Quoted 29 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +28 -0
+ * @id: The region id to migrate to.
+ *
+ * Attempt to migrate the object to the desired memory region. The
+ * object backend must support migration and the object may not be
+ * pinned, (explicitly pinned pages or pinned vmas). The object must
+ * be locked.
+ * On successful completion, the object will have pages pointing to
+ * memory in the new region, but an async migration task may not have
+ * completed yet, and to accomplish that,
 i915_gem_object_wait_migration()
+ * must be called.
+ *
+ * Return: 0 on success. Negative error code on failure. In particular may
+ * return -ENXIO on lack of region space, -EDEADLK for deadlock avoidance
+ * if @ww is set, -EINTR or -ERESTARTSYS if signal pending, and
+ * -EBUSY if the object is pinned.
+ */
+int i915_gem_object_migrate(struct drm_i915_gem_object *obj,
+			    struct i915_gem_ww_ctx *ww,
+			    enum intel_region_id id)
+{
+	struct drm_i915_private *i915 = to_i915(obj->base.dev);
+	struct intel_memory_region *mr;
+
+	GEM_BUG_ON(id >= INTEL_REGION_UNKNOWN);
+	GEM_BUG_ON(obj->mm.madv != I915_MADV_WILLNEED);
+	assert_object_held(obj);
+
+	mr = i915->mm.regions[id];

drivers/gpu/drm/i915/gem/i915_gem_object.c bool i915_gem_object_has_iomem(const struct drm_i915_gem_object *obj) · line 607

Quoted 10 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +10 -0
+	GEM_BUG_ON(!mr);
+
+	if (obj->mm.region == mr)
+		return 0;
+
+	if (!i915_gem_object_evictable(obj))
+		return -EBUSY;
+
+	if (!obj->ops->migrate)
+		return -EOPNOTSUPP;

Why aren't you using _can_migrate here?

Quoted 169 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +3 -0
+	return obj->ops->migrate(obj, mr);
+}
+
 void i915_gem_init__objects(struct drm_i915_private *i915)
{
	INIT_WORK(&i915->mm.free_work, __i915_gem_free_work);
drivers/gpu/drm/i915/gem/i915_gem_object.h +2 -0
index ea3224a480c4..8cbd7a5334e2 100644
@@ -17,6 +17,8 @@
 #include "i915_gem_ww.h"
 #include "i915_vma_types.h"
 
+enum intel_region_id;
+
 /*
 * XXX: There is a prevalence of the assumption that we fit the
 * object's page count inside a 32bit _signed_ variable. Let's document
@@ -597,6 +599,16 @@ bool i915_gem_object_migratable(struct
 drm_i915_gem_object *obj);
 

bool i915_gem_object_validates_to_lmem(struct drm_i915_gem_object *obj);

drivers/gpu/drm/i915/gem/i915_gem_object.h +10 -0
+int i915_gem_object_migrate(struct drm_i915_gem_object *obj,
+			    struct i915_gem_ww_ctx *ww,
+			    enum intel_region_id id);
+
+bool i915_gem_object_can_migrate(struct drm_i915_gem_object *obj,
+				 enum intel_region_id id);
+
+int i915_gem_object_wait_migration(struct drm_i915_gem_object *obj,
+				   unsigned int flags);
+

#ifdef CONFIG_MMU_NOTIFIER
static inline bool i915_gem_object_is_userptr(struct drm_i915_gem_object *obj)

drivers/gpu/drm/i915/gem/i915_gem_object_types.h +9 -0
index 441f913c87e6..ef3de2ae9723 100644
@@ -18,6 +18,7 @@
 
 struct drm_i915_gem_object;
 struct intel_fronbuffer;
+struct intel_memory_region;
 
 /*
 * struct i915_lut_handle tracks the fast lookups from handle to vma used
@@ -77,6 +78,14 @@ struct drm_i915_gem_object_ops {
 	 * delayed_free - Override the default delayed free implementation
 	 */
 	void (*delayed_free)(struct drm_i915_gem_object *obj);
+
+	/**
+	 * migrate - Migrate object to a different region either for
+	 * pinning or for as long as the object lock is held.
+	 */
+	int (*migrate)(struct drm_i915_gem_object *obj,
+		       struct intel_memory_region *mr);
+
 	void (*release)(struct drm_i915_gem_object *obj);
 
 	const struct vm_operations_struct *mmap_ops;
drivers/gpu/drm/i915/gem/i915_gem_ttm.c +3 -10
index c39d982c4fa6..8f89185b6507 100644
@@ -617,7 +617,8 @@ struct ttm_device_funcs *i915_ttm_driver(void)
 	return &i915_ttm_bo_driver;
 }
 
-static int i915_ttm_get_pages(struct drm_i915_gem_object *obj)
+static int __i915_ttm_get_pages(struct drm_i915_gem_object *obj,
+				struct ttm_placement *placement)
 {
 	struct ttm_buffer_object *bo = i915_gem_to_ttm(obj);
 	struct ttm_operation_ctx ctx = {
@@ -625,19 +626,12 @@ static int i915_ttm_get_pages(struct
 drm_i915_gem_object *obj)
 		.no_wait_gpu = false,
 	};
 	struct sg_table *st;
-	struct ttm_place requested, busy[I915_TTM_MAX_PLACEMENTS];
-	struct ttm_placement placement;
 	int real_num_busy;
 	int ret;
 
-	GEM_BUG_ON(obj->mm.n_placements >
 I915_TTM_MAX_PLACEMENTS);
-
-	/* Move to the requested placement. */
-	i915_ttm_placement_from_obj(obj, &requested, busy, &placement);
-
 	/* First try only the requested placement. No eviction. */
-	real_num_busy =
 fetch_and_zero(&placement.num_busy_placement);
-	ret = ttm_bo_validate(bo, &placement, &ctx);
+	real_num_busy = fetch_and_zero(&placement-
Quoted 1 line

num_busy_placement);

drivers/gpu/drm/i915/gem/i915_gem_ttm.c +1 -0
+	ret = ttm_bo_validate(bo, placement, &ctx);
	if (ret) {
		ret = i915_ttm_err_to_gem(ret);
		/*
drivers/gpu/drm/i915/gem/i915_gem_ttm.c +2 -2
@@ -652,8 +646,8 @@ static int i915_ttm_get_pages(struct
 drm_i915_gem_object *obj)
 		 * If the initial attempt fails, allow all accepted placements,
 		 * evicting if necessary.
 		 */
-		placement.num_busy_placement = real_num_busy;
-		ret = ttm_bo_validate(bo, &placement, &ctx);
+		placement->num_busy_placement = real_num_busy;
+		ret = ttm_bo_validate(bo, placement, &ctx);
 		if (ret)
 			return i915_ttm_err_to_gem(ret);
	}
drivers/gpu/drm/i915/gem/i915_gem_ttm.c +13 -5
@@ -668,16 +662,56 @@ static int i915_ttm_get_pages(struct
 drm_i915_gem_object *obj)
 		i915_ttm_adjust_gem_after_move(obj);
 	}
 
-	/* Object either has a page vector or is an iomem object */
-	st = bo->ttm ? i915_ttm_tt_get_st(bo->ttm) : obj->ttm.cached_io_st;
-	if (IS_ERR(st))
-		return PTR_ERR(st);
+	if (!obj->mm.pages) {
+		/* Object either has a page vector or is an iomem object */
+		st = bo->ttm ? i915_ttm_tt_get_st(bo->ttm) : obj-
 >ttm.cached_io_st;
+		if (IS_ERR(st))
+			return PTR_ERR(st);
 
-	__i915_gem_object_set_pages(obj, st, i915_sg_dma_sizes(st->sgl));
+		__i915_gem_object_set_pages(obj, st,
 i915_sg_dma_sizes(st->sgl));
+	}
 
 	return ret;
 }
 
+static int i915_ttm_get_pages(struct drm_i915_gem_object *obj)
+{
+	struct ttm_place requested, busy[I915_TTM_MAX_PLACEMENTS];
+	struct ttm_placement placement;
+
+	GEM_BUG_ON(obj->mm.n_placements >

I915_TTM_MAX_PLACEMENTS);

drivers/gpu/drm/i915/gem/i915_gem_ttm.c +18 -0
+
+	/* Move to the requested placement. */
+	i915_ttm_placement_from_obj(obj, &requested, busy, &placement);
+
+	return __i915_ttm_get_pages(obj, &placement);
+}
+
+static int i915_ttm_migrate(struct drm_i915_gem_object *obj,
+			    struct intel_memory_region *mr)
+{
+	struct ttm_place requested;
+	struct ttm_placement placement;
+	int ret;
+
+	i915_ttm_place_from_region(mr, &requested, obj->flags);
+	placement.num_placement = 1;
+	placement.num_busy_placement = 1;
+	placement.placement = &requested;

drivers/gpu/drm/i915/gem/i915_gem_ttm.c static int i915_ttm_get_pages(struct drm_i915_gem_object *obj) · line 710

Quoted 10 lines
drivers/gpu/drm/i915/gem/i915_gem_ttm.c +10 -0
+	placement.busy_placement = &requested;
+
+	ret = __i915_ttm_get_pages(obj, &placement);
+	if (ret)
+		return ret;
+
+	if (obj->mm.region != mr) {
+		i915_gem_object_release_memory_region(obj);
+		i915_gem_object_init_memory_region(obj, mr);
+	}

Perhaps a minor nit:

Doing this after we have done the _get_pages() just doesn't seem right.

I.e. we do work on the object, and then we init some portion of it.

Do we need to do this incase the migration/placement fails? If so, maybe a comment to that effect?

Thanks,

Mike

Quoted 46 lines
drivers/gpu/drm/i915/gem/i915_gem_ttm.c +3 -0
+	return 0;
+}
+

static void i915_ttm_put_pages(struct drm_i915_gem_object *obj,
struct sg_table *st)
{

drivers/gpu/drm/i915/gem/i915_gem_ttm.c +1 -0
@@ -814,6 +848,7 @@ static const struct drm_i915_gem_object_ops
 i915_gem_ttm_obj_ops = {
 	.truncate = i915_ttm_purge,
 	.adjust_lru = i915_ttm_adjust_lru,
 	.delayed_free = i915_ttm_delayed_free,
+	.migrate = i915_ttm_migrate,
 	.mmap_offset = i915_ttm_mmap_offset,
 	.mmap_ops = &vm_ops_ttm,

};

drivers/gpu/drm/i915/gem/i915_gem_wait.c +0 -0
index 1070d3afdce7..f909aaa09d9c 100644
@@ -290,3 +290,22 @@ i915_gem_wait_ioctl(struct drm_device *dev, void
 *data, struct drm_file *file)
 	i915_gem_object_put(obj);
 	return ret;

}

drivers/gpu/drm/i915/gem/i915_gem_wait.c +3 -0
+
+/**
+ * i915_gem_object_wait_migration - Sync an accelerated migration

operation
+ * @obj: The migrating object.
+ * @flags: waiting flags. Currently supports only I915_WAIT_INTERRUPTIBLE.
+ *
+ * Wait for any pending async migration operation on the object,
+ * whether it's explicitly (i915_gem_object_migrate()) or implicitly
+ * (swapin, initial clearing) initiated.
+ *
+ * Return: 0 if successful, -ERESTARTSYS if a signal was hit during waiting.
+ */

drivers/gpu/drm/i915/gem/i915_gem_wait.c +7 -1
+int i915_gem_object_wait_migration(struct drm_i915_gem_object *obj,
+				   unsigned int flags)
+{
+	might_sleep();
+	/* NOP for now. */
+	return 0;
+}
--
Quoted 1 line

2.31.1

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#12 lore raw

RE: [Intel-gfx] [PATCH v3 2/5] drm/i915/gem: Introduce a selftest for the gem object migrate functionality

In reply to Thomas Hellström
Quoted 163 lines

-----Original Message-----
From: Intel-gfx <[email protected]> On Behalf Of Thomas Hellström
Sent: Monday, June 28, 2021 10:46 AM
To: [email protected]; [email protected]
Cc: Thomas Hellström <[email protected]>; Auld, Matthew <[email protected]>
Subject: [Intel-gfx] [PATCH v3 2/5] drm/i915/gem: Introduce a selftest for the gem object migrate functionality

From: Matthew Auld <[email protected]>

A selftest for the gem object migrate functionality. Slightly adapted from the original by Matthew to the new interface and new fill blit code.

---
drivers/gpu/drm/i915/gem/i915_gem_object.c | 1 +
.../drm/i915/gem/selftests/i915_gem_migrate.c | 237
++++++++++++++++++
.../drm/i915/selftests/i915_live_selftests.h | 1 +

3 files changed, 239 insertions(+)

create mode 100644 drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c

drivers/gpu/drm/i915/gem/i915_gem_object.c +1 -0
index 1c18be067b58..ff147fd59874 100644
@@ -649,6 +649,7 @@ static const struct drm_gem_object_funcs
 i915_gem_object_funcs = {
 #if IS_ENABLED(CONFIG_DRM_I915_SELFTEST)
 #include "selftests/huge_gem_object.c"
 #include "selftests/huge_pages.c"
+#include "selftests/i915_gem_migrate.c"
 #include "selftests/i915_gem_object.c"
 #include "selftests/i915_gem_coherency.c"

#endif

drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c new file +26 -0
new file mode 100644
index 000000000000..a437b66f64d9
@@ -0,0 +1,237 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2020-2021 Intel Corporation
+ */
+
+#include "gt/intel_migrate.h"
+
+static int igt_smem_create_migrate(void *arg)
+{
+	struct intel_gt *gt = arg;
+	struct drm_i915_private *i915 = gt->i915;
+	struct drm_i915_gem_object *obj;
+	struct i915_gem_ww_ctx ww;
+	int err = 0;
+
+	/* Switch object backing-store on create */
+	obj = i915_gem_object_create_lmem(i915, PAGE_SIZE, 0);
+	if (IS_ERR(obj))
+		return PTR_ERR(obj);
+
+	for_i915_gem_ww(&ww, err, true) {
+		err = i915_gem_object_lock(obj, &ww);
+		if (err)
+			continue;
+
+		if (!i915_gem_object_can_migrate(obj,

INTEL_REGION_SMEM)) {

drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c new file +5 -0
+			err = -EINVAL;
+			continue;
+		}
+
+		err = i915_gem_object_migrate(obj, &ww,

INTEL_REGION_SMEM);

drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c new file +8 -0
+		if (err)
+			continue;
+
+		err = i915_gem_object_pin_pages(obj);
+		if (err)
+			continue;
+
+		if (i915_gem_object_can_migrate(obj,

INTEL_REGION_LMEM))

drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c new file +28 -0
+			err = -EINVAL;
+
+		i915_gem_object_unpin_pages(obj);
+	}
+	i915_gem_object_put(obj);
+
+	return err;
+}
+
+static int igt_lmem_create_migrate(void *arg)
+{
+	struct intel_gt *gt = arg;
+	struct drm_i915_private *i915 = gt->i915;
+	struct drm_i915_gem_object *obj;
+	struct i915_gem_ww_ctx ww;
+	int err = 0;
+
+	/* Switch object backing-store on create */
+	obj = i915_gem_object_create_shmem(i915, PAGE_SIZE);
+	if (IS_ERR(obj))
+		return PTR_ERR(obj);
+
+	for_i915_gem_ww(&ww, err, true) {
+		err = i915_gem_object_lock(obj, &ww);
+		if (err)
+			continue;
+
+		if (!i915_gem_object_can_migrate(obj,

INTEL_REGION_LMEM)) {

drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c new file +5 -0
+			err = -EINVAL;
+			continue;
+		}
+
+		err = i915_gem_object_migrate(obj, &ww,

INTEL_REGION_LMEM);

drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c new file +8 -0
+		if (err)
+			continue;
+
+		err = i915_gem_object_pin_pages(obj);
+		if (err)
+			continue;
+
+		if (i915_gem_object_can_migrate(obj,

INTEL_REGION_SMEM))

drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c new file +28 -0
+			err = -EINVAL;
+
+		i915_gem_object_unpin_pages(obj);
+	}
+	i915_gem_object_put(obj);
+
+	return err;
+}
+
+static int lmem_pages_migrate_one(struct i915_gem_ww_ctx *ww,
+				  struct drm_i915_gem_object *obj)
+{
+	int err;
+
+	err = i915_gem_object_lock(obj, ww);
+	if (err)
+		return err;
+
+	err = i915_gem_object_wait(obj,
+				   I915_WAIT_INTERRUPTIBLE |
+				   I915_WAIT_PRIORITY |
+				   I915_WAIT_ALL,
+				   MAX_SCHEDULE_TIMEOUT);
+	if (err)
+		return err;
+
+	if (i915_gem_object_is_lmem(obj)) {
+		if (!i915_gem_object_can_migrate(obj,
Quoted 1 line

INTEL_REGION_SMEM)) {

I don't see any testing of the parameter num_allowed.

Is that done somewhere else?

Mike

Quoted 151 lines
drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c new file +5 -0
+			pr_err("object can't migrate to smem.\n");
+			return -EINVAL;
+		}
+
+		err = i915_gem_object_migrate(obj, ww,

INTEL_REGION_SMEM);

drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c new file +18 -0
+		if (err) {
+			pr_err("Object failed migration to smem\n");
+			if (err)
+				return err;
+		}
+
+		if (i915_gem_object_is_lmem(obj)) {
+			pr_err("object still backed by lmem\n");
+			err = -EINVAL;
+		}
+
+		if (!i915_gem_object_has_struct_page(obj)) {
+			pr_err("object not backed by struct page\n");
+			err = -EINVAL;
+		}
+
+	} else {
+		if (!i915_gem_object_can_migrate(obj,

INTEL_REGION_LMEM)) {

drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c new file +5 -0
+			pr_err("object can't migrate to lmem.\n");
+			return -EINVAL;
+		}
+
+		err = i915_gem_object_migrate(obj, ww,

INTEL_REGION_LMEM);

drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c new file +48 -0
+		if (err) {
+			pr_err("Object failed migration to lmem\n");
+			if (err)
+				return err;
+		}
+
+		if (i915_gem_object_has_struct_page(obj)) {
+			pr_err("object still backed by struct page\n");
+			err = -EINVAL;
+		}
+
+		if (!i915_gem_object_is_lmem(obj)) {
+			pr_err("object not backed by lmem\n");
+			err = -EINVAL;
+		}
+	}
+
+	return err;
+}
+
+static int igt_lmem_pages_migrate(void *arg)
+{
+	struct intel_gt *gt = arg;
+	struct drm_i915_private *i915 = gt->i915;
+	struct drm_i915_gem_object *obj;
+	struct i915_gem_ww_ctx ww;
+	struct i915_request *rq;
+	int err;
+	int i;
+
+	/* From LMEM to shmem and back again */
+
+	obj = i915_gem_object_create_lmem(i915, SZ_2M, 0);
+	if (IS_ERR(obj))
+		return PTR_ERR(obj);
+
+	err = i915_gem_object_lock(obj, NULL);
+	if (err)
+		goto out_put;
+
+	err = ____i915_gem_object_get_pages(obj);
+	if (err) {
+		i915_gem_object_unlock(obj);
+		goto out_put;
+	}
+
+	err = intel_context_migrate_clear(gt->migrate.context, NULL,
+					  obj->mm.pages->sgl, obj-
Quoted 1 line

cache_level,

drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c new file +24 -0
+					  i915_gem_object_is_lmem(obj),
+					  0, &rq);
+	if (rq) {
+		dma_resv_add_excl_fence(obj->base.resv, &rq->fence);
+		i915_request_put(rq);
+	}
+	i915_gem_object_unlock(obj);
+	if (err)
+		goto out_put;
+
+	for (i = 1; i <= 4; ++i) {
+		for_i915_gem_ww(&ww, err, true) {
+			err = lmem_pages_migrate_one(&ww, obj);
+			if (err)
+				continue;
+
+			err = i915_gem_object_wait_migration(obj, true);
+			if (err)
+				continue;
+
+			err = intel_migrate_clear(&gt->migrate, &ww, NULL,
+						  obj->mm.pages->sgl,
+						  obj->cache_level,
+

i915_gem_object_is_lmem(obj),

drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c new file +29 -0
+						  0xdeadbeaf, &rq);
+			if (rq) {
+				dma_resv_add_excl_fence(obj->base.resv,
+							&rq->fence);
+				i915_request_put(rq);
+			}
+		}
+		if (err)
+			break;
+	}
+out_put:
+	i915_gem_object_put(obj);
+
+	return err;
+}
+
+int i915_gem_migrate_live_selftests(struct drm_i915_private *i915)
+{
+	static const struct i915_subtest tests[] = {
+		SUBTEST(igt_smem_create_migrate),
+		SUBTEST(igt_lmem_create_migrate),
+		SUBTEST(igt_lmem_pages_migrate),
+	};
+
+	if (!HAS_LMEM(i915))
+		return 0;
+
+	return intel_gt_live_subtests(tests, &i915->gt);
+}
drivers/gpu/drm/i915/selftests/i915_live_selftests.h +1 -0
index a68197cf1044..e2fd1b61af71 100644
@@ -40,6 +40,7 @@ selftest(hugepages,
 i915_gem_huge_page_live_selftests)
 selftest(gem_contexts, i915_gem_context_live_selftests)
 selftest(gem_execbuf, i915_gem_execbuffer_live_selftests)
 selftest(client, i915_gem_client_blt_live_selftests)
+selftest(gem_migrate, i915_gem_migrate_live_selftests)
 selftest(reset, intel_reset_live_selftests)
 selftest(memory_region, intel_memory_region_live_selftests)

selftest(hangcheck, intel_hangcheck_live_selftests) --
2.31.1

Quoted 4 lines

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#5 lore raw

Re: [PATCH v3 1/5] drm/i915/gem: Implement object migration

In reply to Ruhl, Michael J
Quoted 103 lines · Michael J
Quoted 83 lines

-----Original Message-----
From: dri-devel <[email protected]> On Behalf Of Thomas Hellström
Sent: Monday, June 28, 2021 10:46 AM
To: [email protected]; [email protected]
Cc: Thomas Hellström <[email protected]>; Auld, Matthew <[email protected]>; lkp <[email protected]>
Subject: [PATCH v3 1/5] drm/i915/gem: Implement object migration

Introduce an interface to migrate objects between regions. This is primarily intended to migrate objects to LMEM for display and to SYSTEM for dma-buf, but might be reused in one form or another for performance-based migration.

v2:
- Verify that the memory region given as an id really exists. (Reported by Matthew Auld)
- Call i915_gem_object_{init,release}_memory_region() when switching region
to handle also switching region lists. (Reported by Matthew Auld)
v3:
- Fix i915_gem_object_can_migrate() to return true if object is already in the correct region, even if the object ops doesn't have a migrate() callback.
- Update typo in commit message.
- Fix kerneldoc of i915_gem_object_wait_migration().

---
drivers/gpu/drm/i915/gem/i915_gem_object.c | 96
+++++++++++++++++++

drivers/gpu/drm/i915/gem/i915_gem_object.h    | 12 +++
.../gpu/drm/i915/gem/i915_gem_object_types.h  |  9 ++
drivers/gpu/drm/i915/gem/i915_gem_ttm.c       | 69 +++++++++----
drivers/gpu/drm/i915/gem/i915_gem_wait.c      | 19 ++++
5 files changed, 188 insertions(+), 17 deletions(-)
drivers/gpu/drm/i915/gem/i915_gem_object.c +35 -0
index 07e8ff9a8aae..1c18be067b58 100644
@@ -513,6 +513,102 @@ bool i915_gem_object_has_iomem(const struct
 drm_i915_gem_object *obj)
 	return obj->mem_flags & I915_BO_FLAG_IOMEM;
 }
 
+/**
+ * i915_gem_object_can_migrate - Whether an object likely can be migrated
+ *
+ * @obj: The object to migrate
+ * @id: The region intended to migrate to
+ *
+ * Check whether the object backend supports migration to the
+ * given region. Note that pinning may affect the ability to migrate.
+ *
+ * Return: true if migration is possible, false otherwise.
+ */
+bool i915_gem_object_can_migrate(struct drm_i915_gem_object *obj,
+				 enum intel_region_id id)
+{
+	struct drm_i915_private *i915 = to_i915(obj->base.dev);
+	unsigned int num_allowed = obj->mm.n_placements;
+	struct intel_memory_region *mr;
+	unsigned int i;
+
+	GEM_BUG_ON(id >= INTEL_REGION_UNKNOWN);
+	GEM_BUG_ON(obj->mm.madv != I915_MADV_WILLNEED);
+
+	mr = i915->mm.regions[id];
+	if (!mr)
+		return false;
+
+	if (obj->mm.region == mr)
+		return true;
+
+	if (!i915_gem_object_evictable(obj))
+		return false;
+
+	if (!obj->ops->migrate)
+		return false;
+

drivers/gpu/drm/i915/gem/i915_gem_object.c bool i915_gem_object_has_iomem(const struct · line 561

Quoted 10 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +10 -0
+	if (!(obj->flags & I915_BO_ALLOC_USER))
+		return true;
+
+	if (num_allowed == 0)
+		return false;
+
+	for (i = 0; i < num_allowed; ++i) {
+		if (mr == obj->mm.placements[i])
+			return true;
+	}

Hi Thomas,

I am a little confused over the difference in checks between this function and i915_gem_object_migrate().

Why is the lack of an mr a BUG_ON in _object_migrate(), but here it just false?

So that means that under certain circumstances, you could not have a mr?

Quoted 8 lines

If that is the case, when?

Would that be when the I915_BO_ALLOC_USER is set?

If so, should there be a check for "non" user vs user?

Or maybe document this function pointing out why there are differences and why?

Yes, I'll give it some more documentation. The basic idea is that the above function also could be
used to validate user supplied data, whereas there might be cases where we want to use the gem_object_migrate() function and override the above.

Quoted 10 lines

drivers/gpu/drm/i915/gem/i915_gem_object.c bool i915_gem_object_has_iomem(const struct drm_i915_gem_object *obj) · line 569

Quoted 9 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +8 -0
+	return false;
+}
+
+/**
+ * i915_gem_object_migrate - Migrate an object to the desired region id
+ * @obj: The object to migrate.
+ * @ww: An optional struct i915_gem_ww_ctx. If NULL, the backend may
+ * not be successful in evicting other objects to make room for this object.

Is the ww for future consideration? (I don't see any use of it in the patch).

Yes, but it will remain optional.

Quoted 33 lines
Quoted 33 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +31 -0
+ * @id: The region id to migrate to.
+ *
+ * Attempt to migrate the object to the desired memory region. The
+ * object backend must support migration and the object may not be
+ * pinned, (explicitly pinned pages or pinned vmas). The object must
+ * be locked.
+ * On successful completion, the object will have pages pointing to
+ * memory in the new region, but an async migration task may not have
+ * completed yet, and to accomplish that,
 i915_gem_object_wait_migration()
+ * must be called.
+ *
+ * Return: 0 on success. Negative error code on failure. In particular may
+ * return -ENXIO on lack of region space, -EDEADLK for deadlock avoidance
+ * if @ww is set, -EINTR or -ERESTARTSYS if signal pending, and
+ * -EBUSY if the object is pinned.
+ */
+int i915_gem_object_migrate(struct drm_i915_gem_object *obj,
+			    struct i915_gem_ww_ctx *ww,
+			    enum intel_region_id id)
+{
+	struct drm_i915_private *i915 = to_i915(obj->base.dev);
+	struct intel_memory_region *mr;
+
+	GEM_BUG_ON(id >= INTEL_REGION_UNKNOWN);
+	GEM_BUG_ON(obj->mm.madv != I915_MADV_WILLNEED);
+	assert_object_held(obj);
+
+	mr = i915->mm.regions[id];
+	GEM_BUG_ON(!mr);
+
+	if (obj->mm.region == mr)
Quoted 8 lines

drivers/gpu/drm/i915/gem/i915_gem_object.c bool i915_gem_object_has_iomem(const struct drm_i915_gem_object *obj) · line 607

Quoted 7 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +7 -0
+		return 0;
+
+	if (!i915_gem_object_evictable(obj))
+		return -EBUSY;
+
+	if (!obj->ops->migrate)
+		return -EOPNOTSUPP;

Why aren't you using _can_migrate here?

It's just in case we want to override. I'll add some more comments about this.

Quoted 182 lines
Quoted 170 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +3 -0
+	return obj->ops->migrate(obj, mr);
+}
+
 void i915_gem_init__objects(struct drm_i915_private *i915)
{
	INIT_WORK(&i915->mm.free_work, __i915_gem_free_work);
drivers/gpu/drm/i915/gem/i915_gem_object.h +2 -0
index ea3224a480c4..8cbd7a5334e2 100644
@@ -17,6 +17,8 @@
 #include "i915_gem_ww.h"
 #include "i915_vma_types.h"
 
+enum intel_region_id;
+
 /*
  * XXX: There is a prevalence of the assumption that we fit the
  * object's page count inside a 32bit _signed_ variable. Let's document
@@ -597,6 +599,16 @@ bool i915_gem_object_migratable(struct
 drm_i915_gem_object *obj);
 

bool i915_gem_object_validates_to_lmem(struct drm_i915_gem_object *obj);

drivers/gpu/drm/i915/gem/i915_gem_object.h +10 -0
+int i915_gem_object_migrate(struct drm_i915_gem_object *obj,
+			    struct i915_gem_ww_ctx *ww,
+			    enum intel_region_id id);
+
+bool i915_gem_object_can_migrate(struct drm_i915_gem_object *obj,
+				 enum intel_region_id id);
+
+int i915_gem_object_wait_migration(struct drm_i915_gem_object *obj,
+				   unsigned int flags);
+

#ifdef CONFIG_MMU_NOTIFIER
static inline bool i915_gem_object_is_userptr(struct drm_i915_gem_object *obj)

drivers/gpu/drm/i915/gem/i915_gem_object_types.h +9 -0
index 441f913c87e6..ef3de2ae9723 100644
@@ -18,6 +18,7 @@
 
 struct drm_i915_gem_object;
 struct intel_fronbuffer;
+struct intel_memory_region;
 
 /*
  * struct i915_lut_handle tracks the fast lookups from handle to vma used
@@ -77,6 +78,14 @@ struct drm_i915_gem_object_ops {
 	 * delayed_free - Override the default delayed free implementation
 	 */
 	void (*delayed_free)(struct drm_i915_gem_object *obj);
+
+	/**
+	 * migrate - Migrate object to a different region either for
+	 * pinning or for as long as the object lock is held.
+	 */
+	int (*migrate)(struct drm_i915_gem_object *obj,
+		       struct intel_memory_region *mr);
+
 	void (*release)(struct drm_i915_gem_object *obj);
 
 	const struct vm_operations_struct *mmap_ops;
drivers/gpu/drm/i915/gem/i915_gem_ttm.c +3 -10
index c39d982c4fa6..8f89185b6507 100644
@@ -617,7 +617,8 @@ struct ttm_device_funcs *i915_ttm_driver(void)
 	return &i915_ttm_bo_driver;
 }
 
-static int i915_ttm_get_pages(struct drm_i915_gem_object *obj)
+static int __i915_ttm_get_pages(struct drm_i915_gem_object *obj,
+				struct ttm_placement *placement)
 {
 	struct ttm_buffer_object *bo = i915_gem_to_ttm(obj);
 	struct ttm_operation_ctx ctx = {
@@ -625,19 +626,12 @@ static int i915_ttm_get_pages(struct
 drm_i915_gem_object *obj)
 		.no_wait_gpu = false,
 	};
 	struct sg_table *st;
-	struct ttm_place requested, busy[I915_TTM_MAX_PLACEMENTS];
-	struct ttm_placement placement;
 	int real_num_busy;
 	int ret;
 
-	GEM_BUG_ON(obj->mm.n_placements >
 I915_TTM_MAX_PLACEMENTS);
-
-	/* Move to the requested placement. */
-	i915_ttm_placement_from_obj(obj, &requested, busy, &placement);
-
 	/* First try only the requested placement. No eviction. */
-	real_num_busy =
 fetch_and_zero(&placement.num_busy_placement);
-	ret = ttm_bo_validate(bo, &placement, &ctx);
+	real_num_busy = fetch_and_zero(&placement-
Quoted 1 line

num_busy_placement);

drivers/gpu/drm/i915/gem/i915_gem_ttm.c +1 -0
+	ret = ttm_bo_validate(bo, placement, &ctx);
	if (ret) {
		ret = i915_ttm_err_to_gem(ret);
		/*
drivers/gpu/drm/i915/gem/i915_gem_ttm.c +2 -2
@@ -652,8 +646,8 @@ static int i915_ttm_get_pages(struct
 drm_i915_gem_object *obj)
 		 * If the initial attempt fails, allow all accepted placements,
 		 * evicting if necessary.
 		 */
-		placement.num_busy_placement = real_num_busy;
-		ret = ttm_bo_validate(bo, &placement, &ctx);
+		placement->num_busy_placement = real_num_busy;
+		ret = ttm_bo_validate(bo, placement, &ctx);
 		if (ret)
 			return i915_ttm_err_to_gem(ret);
	}
drivers/gpu/drm/i915/gem/i915_gem_ttm.c +13 -5
@@ -668,16 +662,56 @@ static int i915_ttm_get_pages(struct
 drm_i915_gem_object *obj)
 		i915_ttm_adjust_gem_after_move(obj);
 	}
 
-	/* Object either has a page vector or is an iomem object */
-	st = bo->ttm ? i915_ttm_tt_get_st(bo->ttm) : obj->ttm.cached_io_st;
-	if (IS_ERR(st))
-		return PTR_ERR(st);
+	if (!obj->mm.pages) {
+		/* Object either has a page vector or is an iomem object */
+		st = bo->ttm ? i915_ttm_tt_get_st(bo->ttm) : obj-
 > ttm.cached_io_st;
+		if (IS_ERR(st))
+			return PTR_ERR(st);
 
-	__i915_gem_object_set_pages(obj, st, i915_sg_dma_sizes(st->sgl));
+		__i915_gem_object_set_pages(obj, st,
 i915_sg_dma_sizes(st->sgl));
+	}
 
 	return ret;
 }
 
+static int i915_ttm_get_pages(struct drm_i915_gem_object *obj)
+{
+	struct ttm_place requested, busy[I915_TTM_MAX_PLACEMENTS];
+	struct ttm_placement placement;
+
+	GEM_BUG_ON(obj->mm.n_placements >

I915_TTM_MAX_PLACEMENTS);

drivers/gpu/drm/i915/gem/i915_gem_ttm.c +18 -0
+
+	/* Move to the requested placement. */
+	i915_ttm_placement_from_obj(obj, &requested, busy, &placement);
+
+	return __i915_ttm_get_pages(obj, &placement);
+}
+
+static int i915_ttm_migrate(struct drm_i915_gem_object *obj,
+			    struct intel_memory_region *mr)
+{
+	struct ttm_place requested;
+	struct ttm_placement placement;
+	int ret;
+
+	i915_ttm_place_from_region(mr, &requested, obj->flags);
+	placement.num_placement = 1;
+	placement.num_busy_placement = 1;
+	placement.placement = &requested;

drivers/gpu/drm/i915/gem/i915_gem_ttm.c static int i915_ttm_get_pages(struct drm_i915_gem_object *obj) · line 710

Quoted 10 lines
drivers/gpu/drm/i915/gem/i915_gem_ttm.c +10 -0
+	placement.busy_placement = &requested;
+
+	ret = __i915_ttm_get_pages(obj, &placement);
+	if (ret)
+		return ret;
+
+	if (obj->mm.region != mr) {
+		i915_gem_object_release_memory_region(obj);
+		i915_gem_object_init_memory_region(obj, mr);
+	}

Perhaps a minor nit:

Quoted 6 lines

Doing this after we have done the _get_pages() just doesn't seem right.

I.e. we do work on the object, and then we init some portion of it.

Do we need to do this incase the migration/placement fails? If so, maybe a comment to that effect?

This is simply switching memory region under the lock, and to also move to another memory region list. Is it the naming _release_ and _init_ that is confusing?

/Thomas

Quoted 44 lines

Thanks,

Mike

Quoted 39 lines
drivers/gpu/drm/i915/gem/i915_gem_ttm.c +3 -0
+	return 0;
+}
+

static void i915_ttm_put_pages(struct drm_i915_gem_object *obj,
struct sg_table *st)
{

drivers/gpu/drm/i915/gem/i915_gem_ttm.c +1 -0
@@ -814,6 +848,7 @@ static const struct drm_i915_gem_object_ops
 i915_gem_ttm_obj_ops = {
 	.truncate = i915_ttm_purge,
 	.adjust_lru = i915_ttm_adjust_lru,
 	.delayed_free = i915_ttm_delayed_free,
+	.migrate = i915_ttm_migrate,
 	.mmap_offset = i915_ttm_mmap_offset,
 	.mmap_ops = &vm_ops_ttm,

};

drivers/gpu/drm/i915/gem/i915_gem_wait.c +0 -0
index 1070d3afdce7..f909aaa09d9c 100644
@@ -290,3 +290,22 @@ i915_gem_wait_ioctl(struct drm_device *dev, void
 *data, struct drm_file *file)
 	i915_gem_object_put(obj);
 	return ret;

}

drivers/gpu/drm/i915/gem/i915_gem_wait.c +3 -0
+
+/**
+ * i915_gem_object_wait_migration - Sync an accelerated migration

operation
+ * @obj: The migrating object.
+ * @flags: waiting flags. Currently supports only I915_WAIT_INTERRUPTIBLE.
+ *
+ * Wait for any pending async migration operation on the object,
+ * whether it's explicitly (i915_gem_object_migrate()) or implicitly
+ * (swapin, initial clearing) initiated.
+ *
+ * Return: 0 if successful, -ERESTARTSYS if a signal was hit during waiting.
+ */
+int i915_gem_object_wait_migration(struct drm_i915_gem_object *obj,

Quoted 8 lines
Quoted 8 lines
drivers/gpu/drm/i915/gem/i915_gem_wait.c +6 -1
+				   unsigned int flags)
+{
+	might_sleep();
+	/* NOP for now. */
+	return 0;
+}
--

2.31.1

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#13 lore raw

Re: [Intel-gfx] [PATCH v3 2/5] drm/i915/gem: Introduce a selftest for the gem object migrate functionality

In reply to Ruhl, Michael J
Quoted 173 lines · Michael J
Quoted 169 lines

-----Original Message-----
From: Intel-gfx <[email protected]> On Behalf Of
Thomas Hellström
Sent: Monday, June 28, 2021 10:46 AM
To: [email protected]; [email protected]
Cc: Thomas Hellström <[email protected]>; Auld, Matthew
<[email protected]>
Subject: [Intel-gfx] [PATCH v3 2/5] drm/i915/gem: Introduce a selftest for the
gem object migrate functionality

From: Matthew Auld <[email protected]>

A selftest for the gem object migrate functionality. Slightly adapted
from the original by Matthew to the new interface and new fill blit code.

Co-developed-by: Thomas Hellström
<[email protected]>

---
drivers/gpu/drm/i915/gem/i915_gem_object.c    |   1 + .../drm/i915/gem/selftests/i915_gem_migrate.c | 237
++++++++++++++++++
.../drm/i915/selftests/i915_live_selftests.h  |   1 +

3 files changed, 239 insertions(+)

create mode 100644 drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c

drivers/gpu/drm/i915/gem/i915_gem_object.c +1 -0
index 1c18be067b58..ff147fd59874 100644
@@ -649,6 +649,7 @@ static const struct drm_gem_object_funcs
 i915_gem_object_funcs = {
 #if IS_ENABLED(CONFIG_DRM_I915_SELFTEST)
 #include "selftests/huge_gem_object.c"
 #include "selftests/huge_pages.c"
+#include "selftests/i915_gem_migrate.c"
 #include "selftests/i915_gem_object.c"
 #include "selftests/i915_gem_coherency.c"

#endif

drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c new file +26 -0
new file mode 100644
index 000000000000..a437b66f64d9
@@ -0,0 +1,237 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2020-2021 Intel Corporation
+ */
+
+#include "gt/intel_migrate.h"
+
+static int igt_smem_create_migrate(void *arg)
+{
+       struct intel_gt *gt = arg;
+       struct drm_i915_private *i915 = gt->i915;
+       struct drm_i915_gem_object *obj;
+       struct i915_gem_ww_ctx ww;
+       int err = 0;
+
+       /* Switch object backing-store on create */
+       obj = i915_gem_object_create_lmem(i915, PAGE_SIZE, 0);
+       if (IS_ERR(obj))
+               return PTR_ERR(obj);
+
+       for_i915_gem_ww(&ww, err, true) {
+               err = i915_gem_object_lock(obj, &ww);
+               if (err)
+                       continue;
+
+               if (!i915_gem_object_can_migrate(obj,

INTEL_REGION_SMEM)) {
+                       err = -EINVAL;
+                       continue;
+               }
+
+               err = i915_gem_object_migrate(obj, &ww, INTEL_REGION_SMEM);
+               if (err)
+                       continue;

drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c new file +6 -0
+
+               err = i915_gem_object_pin_pages(obj);
+               if (err)
+                       continue;
+
+               if (i915_gem_object_can_migrate(obj,

INTEL_REGION_LMEM))
+                       err = -EINVAL;

drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c new file +27 -0
+
+               i915_gem_object_unpin_pages(obj);
+       }
+       i915_gem_object_put(obj);
+
+       return err;
+}
+
+static int igt_lmem_create_migrate(void *arg)
+{
+       struct intel_gt *gt = arg;
+       struct drm_i915_private *i915 = gt->i915;
+       struct drm_i915_gem_object *obj;
+       struct i915_gem_ww_ctx ww;
+       int err = 0;
+
+       /* Switch object backing-store on create */
+       obj = i915_gem_object_create_shmem(i915, PAGE_SIZE);
+       if (IS_ERR(obj))
+               return PTR_ERR(obj);
+
+       for_i915_gem_ww(&ww, err, true) {
+               err = i915_gem_object_lock(obj, &ww);
+               if (err)
+                       continue;
+
+               if (!i915_gem_object_can_migrate(obj,

INTEL_REGION_LMEM)) {
+                       err = -EINVAL;
+                       continue;
+               }
+
+               err = i915_gem_object_migrate(obj, &ww, INTEL_REGION_LMEM);
+               if (err)
+                       continue;

drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c new file +6 -0
+
+               err = i915_gem_object_pin_pages(obj);
+               if (err)
+                       continue;
+
+               if (i915_gem_object_can_migrate(obj,

INTEL_REGION_SMEM))
+                       err = -EINVAL;

drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c new file +27 -0
+
+               i915_gem_object_unpin_pages(obj);
+       }
+       i915_gem_object_put(obj);
+
+       return err;
+}
+
+static int lmem_pages_migrate_one(struct i915_gem_ww_ctx *ww,
+                                 struct drm_i915_gem_object *obj)
+{
+       int err;
+
+       err = i915_gem_object_lock(obj, ww);
+       if (err)
+               return err;
+
+       err = i915_gem_object_wait(obj,
+                                  I915_WAIT_INTERRUPTIBLE |
+                                  I915_WAIT_PRIORITY |
+                                  I915_WAIT_ALL,
+                                  MAX_SCHEDULE_TIMEOUT);
+       if (err)
+               return err;
+
+       if (i915_gem_object_is_lmem(obj)) {
+               if (!i915_gem_object_can_migrate(obj,
Quoted 1 line

INTEL_REGION_SMEM)) {

I don't see any  testing of the parameter num_allowed.

Quoted 3 lines

Is that done somewhere else?

Mike

That's a user-space submitted parameter only, dictating what region the object is allowed in when bound to the GPU. It's not exercised in this selftest.

Thanks,
Thomas

Reviewed-by
#14 lore raw

RE: [Intel-gfx] [PATCH v3 2/5] drm/i915/gem: Introduce a selftest for the gem object migrate functionality

Quoted 183 lines

-----Original Message-----
From: Thomas Hellström <[email protected]>
Sent: Monday, June 28, 2021 3:15 PM
To: Ruhl, Michael J <[email protected]>; intel- [email protected]; [email protected]
Cc: Auld, Matthew <[email protected]>
Subject: Re: [Intel-gfx] [PATCH v3 2/5] drm/i915/gem: Introduce a selftest for the gem object migrate functionality

Quoted 173 lines · Michael J
Quoted 169 lines

-----Original Message-----
From: Intel-gfx <[email protected]> On Behalf Of
Thomas Hellström
Sent: Monday, June 28, 2021 10:46 AM
To: [email protected]; [email protected]
Cc: Thomas Hellström <[email protected]>; Auld, Matthew
<[email protected]>
Subject: [Intel-gfx] [PATCH v3 2/5] drm/i915/gem: Introduce a selftest for the
gem object migrate functionality

From: Matthew Auld <[email protected]>

A selftest for the gem object migrate functionality. Slightly adapted
from the original by Matthew to the new interface and new fill blit code.

Co-developed-by: Thomas Hellström
<[email protected]>

---
drivers/gpu/drm/i915/gem/i915_gem_object.c    |   1 + .../drm/i915/gem/selftests/i915_gem_migrate.c | 237
++++++++++++++++++
.../drm/i915/selftests/i915_live_selftests.h  |   1 +

3 files changed, 239 insertions(+)

create mode 100644 drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c

drivers/gpu/drm/i915/gem/i915_gem_object.c +1 -0
index 1c18be067b58..ff147fd59874 100644
@@ -649,6 +649,7 @@ static const struct drm_gem_object_funcs
 i915_gem_object_funcs = {
 #if IS_ENABLED(CONFIG_DRM_I915_SELFTEST)
 #include "selftests/huge_gem_object.c"
 #include "selftests/huge_pages.c"
+#include "selftests/i915_gem_migrate.c"
 #include "selftests/i915_gem_object.c"
 #include "selftests/i915_gem_coherency.c"

#endif

drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c new file +26 -0
new file mode 100644
index 000000000000..a437b66f64d9
@@ -0,0 +1,237 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2020-2021 Intel Corporation
+ */
+
+#include "gt/intel_migrate.h"
+
+static int igt_smem_create_migrate(void *arg)
+{
+       struct intel_gt *gt = arg;
+       struct drm_i915_private *i915 = gt->i915;
+       struct drm_i915_gem_object *obj;
+       struct i915_gem_ww_ctx ww;
+       int err = 0;
+
+       /* Switch object backing-store on create */
+       obj = i915_gem_object_create_lmem(i915, PAGE_SIZE, 0);
+       if (IS_ERR(obj))
+               return PTR_ERR(obj);
+
+       for_i915_gem_ww(&ww, err, true) {
+               err = i915_gem_object_lock(obj, &ww);
+               if (err)
+                       continue;
+
+               if (!i915_gem_object_can_migrate(obj,

INTEL_REGION_SMEM)) {
+                       err = -EINVAL;
+                       continue;
+               }
+
+               err = i915_gem_object_migrate(obj, &ww, INTEL_REGION_SMEM);
+               if (err)
+                       continue;

drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c new file +6 -0
+
+               err = i915_gem_object_pin_pages(obj);
+               if (err)
+                       continue;
+
+               if (i915_gem_object_can_migrate(obj,

INTEL_REGION_LMEM))
+                       err = -EINVAL;

drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c new file +27 -0
+
+               i915_gem_object_unpin_pages(obj);
+       }
+       i915_gem_object_put(obj);
+
+       return err;
+}
+
+static int igt_lmem_create_migrate(void *arg)
+{
+       struct intel_gt *gt = arg;
+       struct drm_i915_private *i915 = gt->i915;
+       struct drm_i915_gem_object *obj;
+       struct i915_gem_ww_ctx ww;
+       int err = 0;
+
+       /* Switch object backing-store on create */
+       obj = i915_gem_object_create_shmem(i915, PAGE_SIZE);
+       if (IS_ERR(obj))
+               return PTR_ERR(obj);
+
+       for_i915_gem_ww(&ww, err, true) {
+               err = i915_gem_object_lock(obj, &ww);
+               if (err)
+                       continue;
+
+               if (!i915_gem_object_can_migrate(obj,

INTEL_REGION_LMEM)) {
+                       err = -EINVAL;
+                       continue;
+               }
+
+               err = i915_gem_object_migrate(obj, &ww, INTEL_REGION_LMEM);
+               if (err)
+                       continue;

drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c new file +6 -0
+
+               err = i915_gem_object_pin_pages(obj);
+               if (err)
+                       continue;
+
+               if (i915_gem_object_can_migrate(obj,

INTEL_REGION_SMEM))
+                       err = -EINVAL;

drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c new file +27 -0
+
+               i915_gem_object_unpin_pages(obj);
+       }
+       i915_gem_object_put(obj);
+
+       return err;
+}
+
+static int lmem_pages_migrate_one(struct i915_gem_ww_ctx *ww,
+                                 struct drm_i915_gem_object *obj)
+{
+       int err;
+
+       err = i915_gem_object_lock(obj, ww);
+       if (err)
+               return err;
+
+       err = i915_gem_object_wait(obj,
+                                  I915_WAIT_INTERRUPTIBLE |
+                                  I915_WAIT_PRIORITY |
+                                  I915_WAIT_ALL,
+                                  MAX_SCHEDULE_TIMEOUT);
+       if (err)
+               return err;
+
+       if (i915_gem_object_is_lmem(obj)) {
+               if (!i915_gem_object_can_migrate(obj,
Quoted 1 line

INTEL_REGION_SMEM)) {

I don't see any  testing of the parameter num_allowed.

Quoted 7 lines
Quoted 3 lines

Is that done somewhere else?

Mike

That's a user-space submitted parameter only, dictating what region the object is allowed in when bound to the GPU. It's not exercised in this selftest.

Got it.

M

Quoted 3 lines

Thanks,
Thomas

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#15 lore raw

Re: [Intel-gfx] [PATCH v3 2/5] drm/i915/gem: Introduce a selftest for the gem object migrate functionality

Quoted 185 lines · Michael J
Quoted 185 lines

-----Original Message-----
From: Thomas Hellström <[email protected]>
Sent: Monday, June 28, 2021 3:15 PM
To: Ruhl, Michael J <[email protected]>; intel- [email protected]; [email protected]
Cc: Auld, Matthew <[email protected]>
Subject: Re: [Intel-gfx] [PATCH v3 2/5] drm/i915/gem: Introduce a selftest for the gem object migrate functionality

Quoted 175 lines · Michael J
Quoted 169 lines

-----Original Message-----
From: Intel-gfx <[email protected]> On Behalf Of
Thomas Hellström
Sent: Monday, June 28, 2021 10:46 AM
To: [email protected]; [email protected]
Cc: Thomas Hellström <[email protected]>; Auld, Matthew
<[email protected]>
Subject: [Intel-gfx] [PATCH v3 2/5] drm/i915/gem: Introduce a selftest for the
gem object migrate functionality

From: Matthew Auld <[email protected]>

A selftest for the gem object migrate functionality. Slightly adapted
from the original by Matthew to the new interface and new fill blit code.

Co-developed-by: Thomas Hellström
<[email protected]>

---
drivers/gpu/drm/i915/gem/i915_gem_object.c    |   1 + .../drm/i915/gem/selftests/i915_gem_migrate.c | 237
++++++++++++++++++
.../drm/i915/selftests/i915_live_selftests.h  |   1 +

3 files changed, 239 insertions(+)

create mode 100644 drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c

drivers/gpu/drm/i915/gem/i915_gem_object.c +1 -0
index 1c18be067b58..ff147fd59874 100644
@@ -649,6 +649,7 @@ static const struct drm_gem_object_funcs
 i915_gem_object_funcs = {
 #if IS_ENABLED(CONFIG_DRM_I915_SELFTEST)
 #include "selftests/huge_gem_object.c"
 #include "selftests/huge_pages.c"
+#include "selftests/i915_gem_migrate.c"
 #include "selftests/i915_gem_object.c"
 #include "selftests/i915_gem_coherency.c"

#endif

drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c new file +26 -0
new file mode 100644
index 000000000000..a437b66f64d9
@@ -0,0 +1,237 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2020-2021 Intel Corporation
+ */
+
+#include "gt/intel_migrate.h"
+
+static int igt_smem_create_migrate(void *arg)
+{
+       struct intel_gt *gt = arg;
+       struct drm_i915_private *i915 = gt->i915;
+       struct drm_i915_gem_object *obj;
+       struct i915_gem_ww_ctx ww;
+       int err = 0;
+
+       /* Switch object backing-store on create */
+       obj = i915_gem_object_create_lmem(i915, PAGE_SIZE, 0);
+       if (IS_ERR(obj))
+               return PTR_ERR(obj);
+
+       for_i915_gem_ww(&ww, err, true) {
+               err = i915_gem_object_lock(obj, &ww);
+               if (err)
+                       continue;
+
+               if (!i915_gem_object_can_migrate(obj,

INTEL_REGION_SMEM)) {
+                       err = -EINVAL;
+                       continue;
+               }
+
+               err = i915_gem_object_migrate(obj, &ww, INTEL_REGION_SMEM);
+               if (err)
+                       continue;

drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c new file +6 -0
+
+               err = i915_gem_object_pin_pages(obj);
+               if (err)
+                       continue;
+
+               if (i915_gem_object_can_migrate(obj,

INTEL_REGION_LMEM))
+                       err = -EINVAL;

drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c new file +27 -0
+
+               i915_gem_object_unpin_pages(obj);
+       }
+       i915_gem_object_put(obj);
+
+       return err;
+}
+
+static int igt_lmem_create_migrate(void *arg)
+{
+       struct intel_gt *gt = arg;
+       struct drm_i915_private *i915 = gt->i915;
+       struct drm_i915_gem_object *obj;
+       struct i915_gem_ww_ctx ww;
+       int err = 0;
+
+       /* Switch object backing-store on create */
+       obj = i915_gem_object_create_shmem(i915, PAGE_SIZE);
+       if (IS_ERR(obj))
+               return PTR_ERR(obj);
+
+       for_i915_gem_ww(&ww, err, true) {
+               err = i915_gem_object_lock(obj, &ww);
+               if (err)
+                       continue;
+
+               if (!i915_gem_object_can_migrate(obj,

INTEL_REGION_LMEM)) {
+                       err = -EINVAL;
+                       continue;
+               }
+
+               err = i915_gem_object_migrate(obj, &ww, INTEL_REGION_LMEM);
+               if (err)
+                       continue;

drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c new file +6 -0
+
+               err = i915_gem_object_pin_pages(obj);
+               if (err)
+                       continue;
+
+               if (i915_gem_object_can_migrate(obj,

INTEL_REGION_SMEM))
+                       err = -EINVAL;

drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c new file +27 -0
+
+               i915_gem_object_unpin_pages(obj);
+       }
+       i915_gem_object_put(obj);
+
+       return err;
+}
+
+static int lmem_pages_migrate_one(struct i915_gem_ww_ctx *ww,
+                                 struct drm_i915_gem_object *obj)
+{
+       int err;
+
+       err = i915_gem_object_lock(obj, ww);
+       if (err)
+               return err;
+
+       err = i915_gem_object_wait(obj,
+                                  I915_WAIT_INTERRUPTIBLE |
+                                  I915_WAIT_PRIORITY |
+                                  I915_WAIT_ALL,
+                                  MAX_SCHEDULE_TIMEOUT);
+       if (err)
+               return err;
+
+       if (i915_gem_object_is_lmem(obj)) {
+               if (!i915_gem_object_can_migrate(obj,
Quoted 1 line

INTEL_REGION_SMEM)) {

I don't see any  testing of the parameter num_allowed.

Is that done somewhere else?

Mike

Quoted 8 lines
Quoted 3 lines

That's a user-space submitted parameter only, dictating what region the object is allowed in when bound to the GPU. It's not exercised in this selftest.

Got it.

M

Thanks!

Thomas

Quoted 3 lines
Quoted 3 lines

Thanks,
Thomas

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Reviewed-by on v3 4/5
#19 lore raw

RE: [PATCH v3 4/5] drm/i915/gem: Fix same-driver-another-instance dma-buf export

In reply to Thomas Hellström
Quoted 16 lines

-----Original Message-----
From: Thomas Hellström <[email protected]>
Sent: Monday, June 28, 2021 10:46 AM
To: [email protected]; [email protected]
Cc: Auld, Matthew <[email protected]>; [email protected]; Thomas Hellström <[email protected]>; Ruhl; Ruhl, Michael J <[email protected]>
Subject: [PATCH v3 4/5] drm/i915/gem: Fix same-driver-another-instance dma-buf export

If our exported dma-bufs are imported by another instance of our driver, that instance will typically have the imported dma-bufs locked during map_attachment(). But the exporter also locks the same reservation object in the map_dma_buf() callback, which leads to recursive locking.

Quoted 6 lines

Add a live selftest to catch this case, and as a workaround until we fully support dynamic import and export, declare the exporter dynamic by providing NOP pin() and unpin() functions. This means our map_dma_buf() callback will *always* get called locked, and by pinning unconditionally in i915_gem_map_dma_buf() we make sure we don't need to use the move_notify() functionality which is not yet implemented.

An interesting abuse of the interface, but it seems reasonable (for now) to me.

Mike

Quoted 206 lines

---
drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c | 31 ++++++-
.../drm/i915/gem/selftests/i915_gem_dmabuf.c | 81
++++++++++++++++++-

2 files changed, 108 insertions(+), 4 deletions(-)
drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c +2 -0
index 616c3a2f1baf..1d1eeb167d28 100644
@@ -12,6 +12,8 @@
 #include "i915_gem_object.h"
 #include "i915_scatterlist.h"
 
+I915_SELFTEST_DECLARE(static bool force_different_devices;)
+
 static struct drm_i915_gem_object *dma_buf_to_obj(struct dma_buf *buf)
 {
 	return to_intel_bo(buf->priv);
@@ -25,7 +27,9 @@ static struct sg_table *i915_gem_map_dma_buf(struct

dma_buf_attachment *attachme

	struct scatterlist *src, *dst;
	int ret, i;
drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c +3 -1
-	ret = i915_gem_object_pin_pages_unlocked(obj);
+	assert_object_held(obj);
+
+	ret = i915_gem_object_pin_pages(obj);
	if (ret)
		goto err;
drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c +3 -0
@@ -168,6 +172,26 @@ static int i915_gem_end_cpu_access(struct dma_buf
 *dma_buf, enum dma_data_direct
 	return err;
 }
 
+/*
+ * As a workaround until we fully support dynamic import and export,
+ * declare the exporter dynamic by providing NOP pin() and unpin()

functions.
+ * This means our i915_gem_map_dma_buf() callback will *always* get called
+ * locked, and by pinning unconditionally in i915_gem_map_dma_buf() we make
+ * sure we don't need to use the move_notify() functionality which is
+ * not yet implemented. Typically for the same-driver-another-instance case,
+ * i915_gem_map_dma_buf() will be called at importer attach time and the
+ * mapped sg_list will be cached by the dma-buf core for the
+ * duration of the attachment.
+ */

drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c +9 -0
+static int i915_gem_dmabuf_pin(struct dma_buf_attachment *attach)
+{
+	return 0;
+}
+
+static void i915_gem_dmabuf_unpin(struct dma_buf_attachment *attach)
+{
+}
+
static const struct dma_buf_ops i915_dmabuf_ops =  {
	.map_dma_buf = i915_gem_map_dma_buf,
	.unmap_dma_buf = i915_gem_unmap_dma_buf,
drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c +2 -0
@@ -177,6 +201,8 @@ static const struct dma_buf_ops i915_dmabuf_ops =  {
 	.vunmap = i915_gem_dmabuf_vunmap,
 	.begin_cpu_access = i915_gem_begin_cpu_access,
 	.end_cpu_access = i915_gem_end_cpu_access,
+	.pin = i915_gem_dmabuf_pin,
+	.unpin = i915_gem_dmabuf_unpin,
 };
 

struct dma_buf *i915_gem_prime_export(struct drm_gem_object *gem_obj, int flags)

drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c +2 -1
@@ -241,7 +267,8 @@ struct drm_gem_object
 *i915_gem_prime_import(struct drm_device *dev,
 	if (dma_buf->ops == &i915_dmabuf_ops) {
 		obj = dma_buf_to_obj(dma_buf);
 		/* is it from our device? */
-		if (obj->base.dev == dev) {
+		if (obj->base.dev == dev &&
+		    !I915_SELFTEST_ONLY(force_different_devices)) {
 			/*
 			 * Importing dmabuf exported from out own gem

increases
* refcount on gem itself instead of f_count of
dmabuf.

drivers/gpu/drm/i915/gem/selftests/i915_gem_dmabuf.c +10 -1
index dd74bc09ec88..24735d6c12a2 100644
@@ -35,7 +35,7 @@ static int igt_dmabuf_export(void *arg)
 static int igt_dmabuf_import_self(void *arg)
 {
 	struct drm_i915_private *i915 = arg;
-	struct drm_i915_gem_object *obj;
+	struct drm_i915_gem_object *obj, *import_obj;
 	struct drm_gem_object *import;
 	struct dma_buf *dmabuf;
 	int err;
@@ -65,14 +65,90 @@ static int igt_dmabuf_import_self(void *arg)
 		err = -EINVAL;
 		goto out_import;
 	}
+	import_obj = to_intel_bo(import);
+
+	i915_gem_object_lock(import_obj, NULL);
+	err = ____i915_gem_object_get_pages(import_obj);
+	i915_gem_object_unlock(import_obj);
+	if (err) {
+		pr_err("Same object dma-buf get_pages failed!\n");
+		goto out_import;
+	}
 
 	err = 0;

out_import:

drivers/gpu/drm/i915/gem/selftests/i915_gem_dmabuf.c +66 -1
-	i915_gem_object_put(to_intel_bo(import));
+	i915_gem_object_put(import_obj);
+out_dmabuf:
+	dma_buf_put(dmabuf);
+out:
+	i915_gem_object_put(obj);
+	return err;
+}
+
+static int igt_dmabuf_import_same_driver(void *arg)
+{
+	struct drm_i915_private *i915 = arg;
+	struct drm_i915_gem_object *obj, *import_obj;
+	struct drm_gem_object *import;
+	struct dma_buf *dmabuf;
+	int err;
+
+	force_different_devices = true;
+	obj = i915_gem_object_create_shmem(i915, PAGE_SIZE);
+	if (IS_ERR(obj))
+		goto out_ret;
+
+	dmabuf = i915_gem_prime_export(&obj->base, 0);
+	if (IS_ERR(dmabuf)) {
+		pr_err("i915_gem_prime_export failed with err=%d\n",
+		       (int)PTR_ERR(dmabuf));
+		err = PTR_ERR(dmabuf);
+		goto out;
+	}
+
+	import = i915_gem_prime_import(&i915->drm, dmabuf);
+	if (IS_ERR(import)) {
+		pr_err("i915_gem_prime_import failed with err=%d\n",
+		       (int)PTR_ERR(import));
+		err = PTR_ERR(import);
+		goto out_dmabuf;
+	}
+
+	if (import == &obj->base) {
+		pr_err("i915_gem_prime_import reused gem object!\n");
+		err = -EINVAL;
+		goto out_import;
+	}
+
+	import_obj = to_intel_bo(import);
+
+	i915_gem_object_lock(import_obj, NULL);
+	err = ____i915_gem_object_get_pages(import_obj);
+	if (err) {
+		pr_err("Different objects dma-buf get_pages failed!\n");
+		i915_gem_object_unlock(import_obj);
+		goto out_import;
+	}
+
+	/*
+	 * If the exported object is not in system memory, something
+	 * weird is going on. TODO: When p2p is supported, this is no
+	 * longer considered weird.
+	 */
+	if (obj->mm.region != i915->mm.regions[INTEL_REGION_SMEM]) {
+		pr_err("Exported dma-buf is not in system memory\n");
+		err = -EINVAL;
+	}
+	i915_gem_object_unlock(import_obj);
+
+out_import:
+	i915_gem_object_put(import_obj);

out_dmabuf:

	dma_buf_put(dmabuf);
out:
	i915_gem_object_put(obj);
drivers/gpu/drm/i915/gem/selftests/i915_gem_dmabuf.c +2 -0
+out_ret:
+	force_different_devices = false;
	return err;
}
drivers/gpu/drm/i915/gem/selftests/i915_gem_dmabuf.c +1 -0
@@ -286,6 +362,7 @@ int i915_gem_dmabuf_live_selftests(struct
 drm_i915_private *i915)
 {
 	static const struct i915_subtest tests[] = {
 		SUBTEST(igt_dmabuf_export),
+		SUBTEST(igt_dmabuf_import_same_driver),
 	};
 
	return i915_subtests(tests, i915);

--
2.31.1

Reviewed-by on v3 5/5
#23 lore raw

RE: [PATCH v3 5/5] drm/i915/gem: Migrate to system at dma-buf map time

In reply to Thomas Hellström
Quoted 40 lines

-----Original Message-----
From: dri-devel <[email protected]> On Behalf Of Thomas Hellström
Sent: Monday, June 28, 2021 10:46 AM
To: [email protected]; [email protected]
Cc: Thomas Hellström <[email protected]>; Auld, Matthew <[email protected]>
Subject: [PATCH v3 5/5] drm/i915/gem: Migrate to system at dma-buf map time

Until we support p2p dma or as a complement to that, migrate data to system memory at dma-buf map time if possible.

v2:
- Rebase on dynamic exporter. Update the igt_dmabuf_import_same_driver selftest to migrate if we are LMEM capable.

---
drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c | 7 ++++++-
drivers/gpu/drm/i915/gem/selftests/i915_gem_dmabuf.c | 4 +++-

2 files changed, 9 insertions(+), 2 deletions(-)
drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c +0 -0
index 1d1eeb167d28..5207aa3af009 100644
@@ -29,7 +29,12 @@ static struct sg_table *i915_gem_map_dma_buf(struct

dma_buf_attachment *attachme

	assert_object_held(obj);
drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c +6 -1
-	ret = i915_gem_object_pin_pages(obj);
+	if (!i915_gem_object_can_migrate(obj, INTEL_REGION_SMEM))
+		return ERR_PTR(-EOPNOTSUPP);
+
+	ret = i915_gem_object_migrate(obj, NULL, INTEL_REGION_SMEM);
+	if (!ret)
+		ret = i915_gem_object_pin_pages(obj);
Quoted 2 lines
	if (ret)
		goto err;

With the caveats of the previous patches this looks good.

Mike

Quoted 17 lines
drivers/gpu/drm/i915/gem/selftests/i915_gem_dmabuf.c +3 -2
index 24735d6c12a2..b6278dba98d0 100644
@@ -94,7 +94,9 @@ static int igt_dmabuf_import_same_driver(void *arg)
 	int err;
 
 	force_different_devices = true;
-	obj = i915_gem_object_create_shmem(i915, PAGE_SIZE);
+	obj = i915_gem_object_create_lmem(i915, PAGE_SIZE, 0);
+	if (IS_ERR(obj))
+		obj = i915_gem_object_create_shmem(i915, PAGE_SIZE);
 	if (IS_ERR(obj))
 		goto out_ret;
 
--
Quoted 1 line

2.31.1

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#6 lore raw

RE: [PATCH v3 1/5] drm/i915/gem: Implement object migration

In reply to Thomas Hellström
Quoted 374 lines

-----Original Message-----
From: Thomas Hellström <[email protected]>
Sent: Monday, June 28, 2021 3:03 PM
To: Ruhl, Michael J <[email protected]>; intel- [email protected]; [email protected]
Cc: Auld, Matthew <[email protected]>; lkp <[email protected]>
Subject: Re: [PATCH v3 1/5] drm/i915/gem: Implement object migration

Quoted 6 lines · Michael J
Quoted 6 lines

-----Original Message-----
From: dri-devel <[email protected]> On Behalf Of Thomas Hellström
Sent: Monday, June 28, 2021 10:46 AM
To: [email protected]; [email protected]
Cc: Thomas Hellström <[email protected]>; Auld,

Matthew

Quoted 42 lines
Quoted 42 lines

<[email protected]>; lkp <[email protected]>
Subject: [PATCH v3 1/5] drm/i915/gem: Implement object migration

Introduce an interface to migrate objects between regions. This is primarily intended to migrate objects to LMEM for display and to SYSTEM for dma-buf, but might be reused in one form or another for performance-based migration.

v2:
- Verify that the memory region given as an id really exists. (Reported by Matthew Auld)
- Call i915_gem_object_{init,release}_memory_region() when switching region
to handle also switching region lists. (Reported by Matthew Auld)
v3:
- Fix i915_gem_object_can_migrate() to return true if object is already in the correct region, even if the object ops doesn't have a migrate() callback.
- Update typo in commit message.
- Fix kerneldoc of i915_gem_object_wait_migration().

---
drivers/gpu/drm/i915/gem/i915_gem_object.c | 96
+++++++++++++++++++

drivers/gpu/drm/i915/gem/i915_gem_object.h    | 12 +++
.../gpu/drm/i915/gem/i915_gem_object_types.h  |  9 ++
drivers/gpu/drm/i915/gem/i915_gem_ttm.c       | 69 +++++++++----
drivers/gpu/drm/i915/gem/i915_gem_wait.c      | 19 ++++
5 files changed, 188 insertions(+), 17 deletions(-)
drivers/gpu/drm/i915/gem/i915_gem_object.c +0 -0
index 07e8ff9a8aae..1c18be067b58 100644
@@ -513,6 +513,102 @@ bool i915_gem_object_has_iomem(const struct
 drm_i915_gem_object *obj)
 	return obj->mem_flags & I915_BO_FLAG_IOMEM;
 }
 
Quoted 2 lines

drivers/gpu/drm/i915/gem/i915_gem_object.c bool i915_gem_object_has_iomem(const struct · line 518

Quoted 2 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +2 -0
+/**
+ * i915_gem_object_can_migrate - Whether an object likely can be

migrated

Quoted 53 lines
Quoted 33 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +33 -0
+ *
+ * @obj: The object to migrate
+ * @id: The region intended to migrate to
+ *
+ * Check whether the object backend supports migration to the
+ * given region. Note that pinning may affect the ability to migrate.
+ *
+ * Return: true if migration is possible, false otherwise.
+ */
+bool i915_gem_object_can_migrate(struct drm_i915_gem_object *obj,
+				 enum intel_region_id id)
+{
+	struct drm_i915_private *i915 = to_i915(obj->base.dev);
+	unsigned int num_allowed = obj->mm.n_placements;
+	struct intel_memory_region *mr;
+	unsigned int i;
+
+	GEM_BUG_ON(id >= INTEL_REGION_UNKNOWN);
+	GEM_BUG_ON(obj->mm.madv != I915_MADV_WILLNEED);
+
+	mr = i915->mm.regions[id];
+	if (!mr)
+		return false;
+
+	if (obj->mm.region == mr)
+		return true;
+
+	if (!i915_gem_object_evictable(obj))
+		return false;
+
+	if (!obj->ops->migrate)
+		return false;
+

drivers/gpu/drm/i915/gem/i915_gem_object.c bool i915_gem_object_has_iomem(const struct drm_i915_gem_object *obj) · line 560

Quoted 10 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +10 -0
+	if (!(obj->flags & I915_BO_ALLOC_USER))
+		return true;
+
+	if (num_allowed == 0)
+		return false;
+
+	for (i = 0; i < num_allowed; ++i) {
+		if (mr == obj->mm.placements[i])
+			return true;
+	}

Hi Thomas,

I am a little confused over the difference in checks between this function and i915_gem_object_migrate().

Why is the lack of an mr a BUG_ON in _object_migrate(), but here it just false?

So that means that under certain circumstances, you could not have a mr?

Quoted 8 lines

If that is the case, when?

Would that be when the I915_BO_ALLOC_USER is set?

If so, should there be a check for "non" user vs user?

Or maybe document this function pointing out why there are differences and why?

Yes, I'll give it some more documentation. The basic idea is that the above function also could be
used to validate user supplied data, whereas there might be cases where we want to use the gem_object_migrate() function and override the above.

Quoted 10 lines

drivers/gpu/drm/i915/gem/i915_gem_object.c bool i915_gem_object_has_iomem(const struct drm_i915_gem_object *obj) · line 569

Quoted 9 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +8 -0
+	return false;
+}
+
+/**
+ * i915_gem_object_migrate - Migrate an object to the desired region id
+ * @obj: The object to migrate.
+ * @ww: An optional struct i915_gem_ww_ctx. If NULL, the backend may
+ * not be successful in evicting other objects to make room for this object.

Is the ww for future consideration? (I don't see any use of it in the patch).

Yes, but it will remain optional.

Quoted 15 lines
Quoted 5 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +4 -0
+ * @id: The region id to migrate to.
+ *
+ * Attempt to migrate the object to the desired memory region. The
+ * object backend must support migration and the object may not be

drivers/gpu/drm/i915/gem/i915_gem_object.c bool i915_gem_object_has_iomem(const struct drm_i915_gem_object *obj)

Quoted 10 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +9 -0
+ * pinned, (explicitly pinned pages or pinned vmas). The object must
+ * be locked.
+ * On successful completion, the object will have pages pointing to
+ * memory in the new region, but an async migration task may not have
+ * completed yet, and to accomplish that,
 i915_gem_object_wait_migration()
+ * must be called.
+ *
+ * Return: 0 on success. Negative error code on failure. In particular may
+ * return -ENXIO on lack of region space, -EDEADLK for deadlock

avoidance

Quoted 18 lines
Quoted 18 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +18 -0
+ * if @ww is set, -EINTR or -ERESTARTSYS if signal pending, and
+ * -EBUSY if the object is pinned.
+ */
+int i915_gem_object_migrate(struct drm_i915_gem_object *obj,
+			    struct i915_gem_ww_ctx *ww,
+			    enum intel_region_id id)
+{
+	struct drm_i915_private *i915 = to_i915(obj->base.dev);
+	struct intel_memory_region *mr;
+
+	GEM_BUG_ON(id >= INTEL_REGION_UNKNOWN);
+	GEM_BUG_ON(obj->mm.madv != I915_MADV_WILLNEED);
+	assert_object_held(obj);
+
+	mr = i915->mm.regions[id];
+	GEM_BUG_ON(!mr);
+
+	if (obj->mm.region == mr)
Quoted 8 lines

drivers/gpu/drm/i915/gem/i915_gem_object.c bool i915_gem_object_has_iomem(const struct drm_i915_gem_object *obj) · line 607

Quoted 7 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +7 -0
+		return 0;
+
+	if (!i915_gem_object_evictable(obj))
+		return -EBUSY;
+
+	if (!obj->ops->migrate)
+		return -EOPNOTSUPP;

Why aren't you using _can_migrate here?

It's just in case we want to override. I'll add some more comments about this.

Quoted 184 lines
Quoted 170 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +3 -0
+	return obj->ops->migrate(obj, mr);
+}
+
 void i915_gem_init__objects(struct drm_i915_private *i915)
{
	INIT_WORK(&i915->mm.free_work, __i915_gem_free_work);
drivers/gpu/drm/i915/gem/i915_gem_object.h +2 -0
index ea3224a480c4..8cbd7a5334e2 100644
@@ -17,6 +17,8 @@
 #include "i915_gem_ww.h"
 #include "i915_vma_types.h"
 
+enum intel_region_id;
+
 /*
  * XXX: There is a prevalence of the assumption that we fit the
  * object's page count inside a 32bit _signed_ variable. Let's document
@@ -597,6 +599,16 @@ bool i915_gem_object_migratable(struct
 drm_i915_gem_object *obj);
 

bool i915_gem_object_validates_to_lmem(struct drm_i915_gem_object *obj);

drivers/gpu/drm/i915/gem/i915_gem_object.h +10 -0
+int i915_gem_object_migrate(struct drm_i915_gem_object *obj,
+			    struct i915_gem_ww_ctx *ww,
+			    enum intel_region_id id);
+
+bool i915_gem_object_can_migrate(struct drm_i915_gem_object *obj,
+				 enum intel_region_id id);
+
+int i915_gem_object_wait_migration(struct drm_i915_gem_object *obj,
+				   unsigned int flags);
+

#ifdef CONFIG_MMU_NOTIFIER
static inline bool i915_gem_object_is_userptr(struct drm_i915_gem_object *obj)

drivers/gpu/drm/i915/gem/i915_gem_object_types.h +9 -0
index 441f913c87e6..ef3de2ae9723 100644
@@ -18,6 +18,7 @@
 
 struct drm_i915_gem_object;
 struct intel_fronbuffer;
+struct intel_memory_region;
 
 /*
  * struct i915_lut_handle tracks the fast lookups from handle to vma used
@@ -77,6 +78,14 @@ struct drm_i915_gem_object_ops {
 	 * delayed_free - Override the default delayed free implementation
 	 */
 	void (*delayed_free)(struct drm_i915_gem_object *obj);
+
+	/**
+	 * migrate - Migrate object to a different region either for
+	 * pinning or for as long as the object lock is held.
+	 */
+	int (*migrate)(struct drm_i915_gem_object *obj,
+		       struct intel_memory_region *mr);
+
 	void (*release)(struct drm_i915_gem_object *obj);
 
 	const struct vm_operations_struct *mmap_ops;
drivers/gpu/drm/i915/gem/i915_gem_ttm.c +3 -10
index c39d982c4fa6..8f89185b6507 100644
@@ -617,7 +617,8 @@ struct ttm_device_funcs *i915_ttm_driver(void)
 	return &i915_ttm_bo_driver;
 }
 
-static int i915_ttm_get_pages(struct drm_i915_gem_object *obj)
+static int __i915_ttm_get_pages(struct drm_i915_gem_object *obj,
+				struct ttm_placement *placement)
 {
 	struct ttm_buffer_object *bo = i915_gem_to_ttm(obj);
 	struct ttm_operation_ctx ctx = {
@@ -625,19 +626,12 @@ static int i915_ttm_get_pages(struct
 drm_i915_gem_object *obj)
 		.no_wait_gpu = false,
 	};
 	struct sg_table *st;
-	struct ttm_place requested, busy[I915_TTM_MAX_PLACEMENTS];
-	struct ttm_placement placement;
 	int real_num_busy;
 	int ret;
 
-	GEM_BUG_ON(obj->mm.n_placements >
 I915_TTM_MAX_PLACEMENTS);
-
-	/* Move to the requested placement. */
-	i915_ttm_placement_from_obj(obj, &requested, busy, &placement);
-
 	/* First try only the requested placement. No eviction. */
-	real_num_busy =
 fetch_and_zero(&placement.num_busy_placement);
-	ret = ttm_bo_validate(bo, &placement, &ctx);
+	real_num_busy = fetch_and_zero(&placement-
Quoted 1 line

num_busy_placement);

drivers/gpu/drm/i915/gem/i915_gem_ttm.c +1 -0
+	ret = ttm_bo_validate(bo, placement, &ctx);
	if (ret) {
		ret = i915_ttm_err_to_gem(ret);
		/*
drivers/gpu/drm/i915/gem/i915_gem_ttm.c +2 -2
@@ -652,8 +646,8 @@ static int i915_ttm_get_pages(struct
 drm_i915_gem_object *obj)
 		 * If the initial attempt fails, allow all accepted placements,
 		 * evicting if necessary.
 		 */
-		placement.num_busy_placement = real_num_busy;
-		ret = ttm_bo_validate(bo, &placement, &ctx);
+		placement->num_busy_placement = real_num_busy;
+		ret = ttm_bo_validate(bo, placement, &ctx);
 		if (ret)
 			return i915_ttm_err_to_gem(ret);
	}
drivers/gpu/drm/i915/gem/i915_gem_ttm.c +13 -5
@@ -668,16 +662,56 @@ static int i915_ttm_get_pages(struct
 drm_i915_gem_object *obj)
 		i915_ttm_adjust_gem_after_move(obj);
 	}
 
-	/* Object either has a page vector or is an iomem object */
-	st = bo->ttm ? i915_ttm_tt_get_st(bo->ttm) : obj->ttm.cached_io_st;
-	if (IS_ERR(st))
-		return PTR_ERR(st);
+	if (!obj->mm.pages) {
+		/* Object either has a page vector or is an iomem object */
+		st = bo->ttm ? i915_ttm_tt_get_st(bo->ttm) : obj-
 > ttm.cached_io_st;
+		if (IS_ERR(st))
+			return PTR_ERR(st);
 
-	__i915_gem_object_set_pages(obj, st, i915_sg_dma_sizes(st->sgl));
+		__i915_gem_object_set_pages(obj, st,
 i915_sg_dma_sizes(st->sgl));
+	}
 
 	return ret;
 }
 
+static int i915_ttm_get_pages(struct drm_i915_gem_object *obj)
+{
+	struct ttm_place requested, busy[I915_TTM_MAX_PLACEMENTS];
+	struct ttm_placement placement;
+
+	GEM_BUG_ON(obj->mm.n_placements >

I915_TTM_MAX_PLACEMENTS);

drivers/gpu/drm/i915/gem/i915_gem_ttm.c +18 -0
+
+	/* Move to the requested placement. */
+	i915_ttm_placement_from_obj(obj, &requested, busy, &placement);
+
+	return __i915_ttm_get_pages(obj, &placement);
+}
+
+static int i915_ttm_migrate(struct drm_i915_gem_object *obj,
+			    struct intel_memory_region *mr)
+{
+	struct ttm_place requested;
+	struct ttm_placement placement;
+	int ret;
+
+	i915_ttm_place_from_region(mr, &requested, obj->flags);
+	placement.num_placement = 1;
+	placement.num_busy_placement = 1;
+	placement.placement = &requested;

drivers/gpu/drm/i915/gem/i915_gem_ttm.c static int i915_ttm_get_pages(struct drm_i915_gem_object *obj) · line 710

Quoted 10 lines
drivers/gpu/drm/i915/gem/i915_gem_ttm.c +10 -0
+	placement.busy_placement = &requested;
+
+	ret = __i915_ttm_get_pages(obj, &placement);
+	if (ret)
+		return ret;
+
+	if (obj->mm.region != mr) {
+		i915_gem_object_release_memory_region(obj);
+		i915_gem_object_init_memory_region(obj, mr);
+	}

Perhaps a minor nit:

Doing this after we have done the _get_pages() just doesn't seem right.

Quoted 8 lines
Quoted 4 lines

I.e. we do work on the object, and then we init some portion of it.

Do we need to do this incase the migration/placement fails? If so, maybe a comment to that effect?

This is simply switching memory region under the lock, and to also move to another memory region list. Is it the naming _release_ and _init_ that is confusing?

Hmm, re-reading my question, I am wondering if I was unclear.

My "real" question was, can the release/init occur before the _get_pages()?

But looking at this some more, I can see answer is probably no.

I was going to suggest calling _init_ _set_, but when I looked at it, it was doing init things as well as setting things

Maybe just a comment like:

/* Complete the migration by updating the memory region info. */ if (object->mm...)

?

M

Quoted 40 lines

/Thomas

Quoted 28 lines

Thanks,

Mike

Quoted 23 lines
drivers/gpu/drm/i915/gem/i915_gem_ttm.c +3 -0
+	return 0;
+}
+

static void i915_ttm_put_pages(struct drm_i915_gem_object *obj,
struct sg_table *st)
{

drivers/gpu/drm/i915/gem/i915_gem_ttm.c +1 -0
@@ -814,6 +848,7 @@ static const struct drm_i915_gem_object_ops
 i915_gem_ttm_obj_ops = {
 	.truncate = i915_ttm_purge,
 	.adjust_lru = i915_ttm_adjust_lru,
 	.delayed_free = i915_ttm_delayed_free,
+	.migrate = i915_ttm_migrate,
 	.mmap_offset = i915_ttm_mmap_offset,
 	.mmap_ops = &vm_ops_ttm,

};

drivers/gpu/drm/i915/gem/i915_gem_wait.c +0 -0
index 1070d3afdce7..f909aaa09d9c 100644
@@ -290,3 +290,22 @@ i915_gem_wait_ioctl(struct drm_device *dev, void
 *data, struct drm_file *file)
 	i915_gem_object_put(obj);
Quoted 8 lines
Quoted 8 lines
drivers/gpu/drm/i915/gem/i915_gem_wait.c +0 -0
 	return ret;

}

drivers/gpu/drm/i915/gem/i915_gem_wait.c +3 -0
+
+/**
+ * i915_gem_object_wait_migration - Sync an accelerated migration

operation
+ * @obj: The migrating object.
+ * @flags: waiting flags. Currently supports only

I915_WAIT_INTERRUPTIBLE.

Quoted 16 lines
Quoted 8 lines
Quoted 8 lines

+ *
+ * Wait for any pending async migration operation on the object,
+ * whether it's explicitly (i915_gem_object_migrate()) or implicitly
+ * (swapin, initial clearing) initiated.
+ *
+ * Return: 0 if successful, -ERESTARTSYS if a signal was hit during waiting.
+ */
+int i915_gem_object_wait_migration(struct drm_i915_gem_object *obj,

Quoted 8 lines
Quoted 8 lines
drivers/gpu/drm/i915/gem/i915_gem_wait.c +6 -1
+				   unsigned int flags)
+{
+	might_sleep();
+	/* NOP for now. */
+	return 0;
+}
--

2.31.1

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#24 lore raw

Re: [PATCH v3 5/5] drm/i915/gem: Migrate to system at dma-buf map time

In reply to Ruhl, Michael J
Quoted 33 lines · Michael J
Quoted 33 lines

-----Original Message-----
From: dri-devel <[email protected]> On Behalf Of Thomas Hellström
Sent: Monday, June 28, 2021 10:46 AM
To: [email protected]; [email protected]
Cc: Thomas Hellström <[email protected]>; Auld, Matthew <[email protected]>
Subject: [PATCH v3 5/5] drm/i915/gem: Migrate to system at dma-buf map time

Until we support p2p dma or as a complement to that, migrate data to system memory at dma-buf map time if possible.

v2:
- Rebase on dynamic exporter. Update the igt_dmabuf_import_same_driver selftest to migrate if we are LMEM capable.

---
drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c | 7 ++++++-
drivers/gpu/drm/i915/gem/selftests/i915_gem_dmabuf.c | 4 +++-

2 files changed, 9 insertions(+), 2 deletions(-)
drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c +0 -0
index 1d1eeb167d28..5207aa3af009 100644
@@ -29,7 +29,12 @@ static struct sg_table *i915_gem_map_dma_buf(struct

dma_buf_attachment *attachme

	assert_object_held(obj);
Quoted 10 lines
Quoted 9 lines
drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c +6 -1
-	ret = i915_gem_object_pin_pages(obj);
+	if (!i915_gem_object_can_migrate(obj, INTEL_REGION_SMEM))
+		return ERR_PTR(-EOPNOTSUPP);
+
+	ret = i915_gem_object_migrate(obj, NULL, INTEL_REGION_SMEM);
+	if (!ret)
+		ret = i915_gem_object_pin_pages(obj);
	if (ret)
		goto err;

With the caveats of the previous patches this looks good.

Yes, this sort of sucks, but knowing that we have a correct implementation in the works...

Quoted 4 lines

Mike

Thanks for the reviews!

Thomas

#7 lore raw

Re: [PATCH v3 1/5] drm/i915/gem: Implement object migration

In reply to Ruhl, Michael J
Quoted 381 lines · Michael J
Quoted 368 lines

-----Original Message-----
From: Thomas Hellström <[email protected]>
Sent: Monday, June 28, 2021 3:03 PM
To: Ruhl, Michael J <[email protected]>; intel- [email protected]; [email protected]
Cc: Auld, Matthew <[email protected]>; lkp <[email protected]>
Subject: Re: [PATCH v3 1/5] drm/i915/gem: Implement object migration

Quoted 6 lines · Michael J
Quoted 6 lines

-----Original Message-----
From: dri-devel <[email protected]> On Behalf Of Thomas Hellström
Sent: Monday, June 28, 2021 10:46 AM
To: [email protected]; [email protected]
Cc: Thomas Hellström <[email protected]>; Auld,

Matthew

Quoted 42 lines
Quoted 42 lines

<[email protected]>; lkp <[email protected]>
Subject: [PATCH v3 1/5] drm/i915/gem: Implement object migration

Introduce an interface to migrate objects between regions. This is primarily intended to migrate objects to LMEM for display and to SYSTEM for dma-buf, but might be reused in one form or another for performance-based migration.

v2:
- Verify that the memory region given as an id really exists. (Reported by Matthew Auld)
- Call i915_gem_object_{init,release}_memory_region() when switching region
to handle also switching region lists. (Reported by Matthew Auld)
v3:
- Fix i915_gem_object_can_migrate() to return true if object is already in the correct region, even if the object ops doesn't have a migrate() callback.
- Update typo in commit message.
- Fix kerneldoc of i915_gem_object_wait_migration().

---
drivers/gpu/drm/i915/gem/i915_gem_object.c | 96
+++++++++++++++++++

drivers/gpu/drm/i915/gem/i915_gem_object.h    | 12 +++
.../gpu/drm/i915/gem/i915_gem_object_types.h  |  9 ++
drivers/gpu/drm/i915/gem/i915_gem_ttm.c       | 69 +++++++++----
drivers/gpu/drm/i915/gem/i915_gem_wait.c      | 19 ++++
5 files changed, 188 insertions(+), 17 deletions(-)
drivers/gpu/drm/i915/gem/i915_gem_object.c +0 -0
index 07e8ff9a8aae..1c18be067b58 100644
@@ -513,6 +513,102 @@ bool i915_gem_object_has_iomem(const struct
 drm_i915_gem_object *obj)
 	return obj->mem_flags & I915_BO_FLAG_IOMEM;
 }
 
Quoted 2 lines

drivers/gpu/drm/i915/gem/i915_gem_object.c bool i915_gem_object_has_iomem(const struct · line 518

Quoted 2 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +2 -0
+/**
+ * i915_gem_object_can_migrate - Whether an object likely can be

migrated

Quoted 53 lines
Quoted 33 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +33 -0
+ *
+ * @obj: The object to migrate
+ * @id: The region intended to migrate to
+ *
+ * Check whether the object backend supports migration to the
+ * given region. Note that pinning may affect the ability to migrate.
+ *
+ * Return: true if migration is possible, false otherwise.
+ */
+bool i915_gem_object_can_migrate(struct drm_i915_gem_object *obj,
+				 enum intel_region_id id)
+{
+	struct drm_i915_private *i915 = to_i915(obj->base.dev);
+	unsigned int num_allowed = obj->mm.n_placements;
+	struct intel_memory_region *mr;
+	unsigned int i;
+
+	GEM_BUG_ON(id >= INTEL_REGION_UNKNOWN);
+	GEM_BUG_ON(obj->mm.madv != I915_MADV_WILLNEED);
+
+	mr = i915->mm.regions[id];
+	if (!mr)
+		return false;
+
+	if (obj->mm.region == mr)
+		return true;
+
+	if (!i915_gem_object_evictable(obj))
+		return false;
+
+	if (!obj->ops->migrate)
+		return false;
+

drivers/gpu/drm/i915/gem/i915_gem_object.c bool i915_gem_object_has_iomem(const struct drm_i915_gem_object *obj) · line 560

Quoted 10 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +10 -0
+	if (!(obj->flags & I915_BO_ALLOC_USER))
+		return true;
+
+	if (num_allowed == 0)
+		return false;
+
+	for (i = 0; i < num_allowed; ++i) {
+		if (mr == obj->mm.placements[i])
+			return true;
+	}

Hi Thomas,

I am a little confused over the difference in checks between this function and i915_gem_object_migrate().

Why is the lack of an mr a BUG_ON in _object_migrate(), but here it just false?

So that means that under certain circumstances, you could not have a mr?

Quoted 8 lines

If that is the case, when?

Would that be when the I915_BO_ALLOC_USER is set?

If so, should there be a check for "non" user vs user?

Or maybe document this function pointing out why there are differences and why?

Yes, I'll give it some more documentation. The basic idea is that the above function also could be
used to validate user supplied data, whereas there might be cases where we want to use the gem_object_migrate() function and override the above.

Quoted 9 lines

drivers/gpu/drm/i915/gem/i915_gem_object.c bool i915_gem_object_has_iomem(const struct drm_i915_gem_object *obj) · line 569

Quoted 8 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +8 -0
+	return false;
+}
+
+/**
+ * i915_gem_object_migrate - Migrate an object to the desired region id
+ * @obj: The object to migrate.
+ * @ww: An optional struct i915_gem_ww_ctx. If NULL, the backend may
+ * not be successful in evicting other objects to make room for this object.

Is the ww for future consideration? (I don't see any use of it in the patch).

Yes, but it will remain optional.

Quoted 6 lines
Quoted 6 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +6 -0
+ * @id: The region id to migrate to.
+ *
+ * Attempt to migrate the object to the desired memory region. The
+ * object backend must support migration and the object may not be
+ * pinned, (explicitly pinned pages or pinned vmas). The object must
+ * be locked.
Quoted 8 lines

drivers/gpu/drm/i915/gem/i915_gem_object.c bool i915_gem_object_has_iomem(const struct drm_i915_gem_object *obj)

Quoted 8 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +7 -0
+ * On successful completion, the object will have pages pointing to
+ * memory in the new region, but an async migration task may not have
+ * completed yet, and to accomplish that,
 i915_gem_object_wait_migration()
+ * must be called.
+ *
+ * Return: 0 on success. Negative error code on failure. In particular may
+ * return -ENXIO on lack of region space, -EDEADLK for deadlock

avoidance

Quoted 18 lines
Quoted 18 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +18 -0
+ * if @ww is set, -EINTR or -ERESTARTSYS if signal pending, and
+ * -EBUSY if the object is pinned.
+ */
+int i915_gem_object_migrate(struct drm_i915_gem_object *obj,
+			    struct i915_gem_ww_ctx *ww,
+			    enum intel_region_id id)
+{
+	struct drm_i915_private *i915 = to_i915(obj->base.dev);
+	struct intel_memory_region *mr;
+
+	GEM_BUG_ON(id >= INTEL_REGION_UNKNOWN);
+	GEM_BUG_ON(obj->mm.madv != I915_MADV_WILLNEED);
+	assert_object_held(obj);
+
+	mr = i915->mm.regions[id];
+	GEM_BUG_ON(!mr);
+
+	if (obj->mm.region == mr)
Quoted 8 lines

drivers/gpu/drm/i915/gem/i915_gem_object.c bool i915_gem_object_has_iomem(const struct drm_i915_gem_object *obj) · line 607

Quoted 7 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +7 -0
+		return 0;
+
+	if (!i915_gem_object_evictable(obj))
+		return -EBUSY;
+
+	if (!obj->ops->migrate)
+		return -EOPNOTSUPP;

Why aren't you using _can_migrate here?

It's just in case we want to override. I'll add some more comments about this.

Quoted 183 lines
Quoted 169 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +3 -0
+	return obj->ops->migrate(obj, mr);
+}
+
 void i915_gem_init__objects(struct drm_i915_private *i915)
{
	INIT_WORK(&i915->mm.free_work, __i915_gem_free_work);
drivers/gpu/drm/i915/gem/i915_gem_object.h +2 -0
index ea3224a480c4..8cbd7a5334e2 100644
@@ -17,6 +17,8 @@
 #include "i915_gem_ww.h"
 #include "i915_vma_types.h"
 
+enum intel_region_id;
+
 /*
   * XXX: There is a prevalence of the assumption that we fit the
   * object's page count inside a 32bit _signed_ variable. Let's document
@@ -597,6 +599,16 @@ bool i915_gem_object_migratable(struct
 drm_i915_gem_object *obj);
 

bool i915_gem_object_validates_to_lmem(struct drm_i915_gem_object *obj);

drivers/gpu/drm/i915/gem/i915_gem_object.h +10 -0
+int i915_gem_object_migrate(struct drm_i915_gem_object *obj,
+			    struct i915_gem_ww_ctx *ww,
+			    enum intel_region_id id);
+
+bool i915_gem_object_can_migrate(struct drm_i915_gem_object *obj,
+				 enum intel_region_id id);
+
+int i915_gem_object_wait_migration(struct drm_i915_gem_object *obj,
+				   unsigned int flags);
+

#ifdef CONFIG_MMU_NOTIFIER
static inline bool i915_gem_object_is_userptr(struct drm_i915_gem_object *obj)

drivers/gpu/drm/i915/gem/i915_gem_object_types.h +9 -0
index 441f913c87e6..ef3de2ae9723 100644
@@ -18,6 +18,7 @@
 
 struct drm_i915_gem_object;
 struct intel_fronbuffer;
+struct intel_memory_region;
 
 /*
   * struct i915_lut_handle tracks the fast lookups from handle to vma used
@@ -77,6 +78,14 @@ struct drm_i915_gem_object_ops {
 	 * delayed_free - Override the default delayed free implementation
 	 */
 	void (*delayed_free)(struct drm_i915_gem_object *obj);
+
+	/**
+	 * migrate - Migrate object to a different region either for
+	 * pinning or for as long as the object lock is held.
+	 */
+	int (*migrate)(struct drm_i915_gem_object *obj,
+		       struct intel_memory_region *mr);
+
 	void (*release)(struct drm_i915_gem_object *obj);
 
 	const struct vm_operations_struct *mmap_ops;
drivers/gpu/drm/i915/gem/i915_gem_ttm.c +3 -10
index c39d982c4fa6..8f89185b6507 100644
@@ -617,7 +617,8 @@ struct ttm_device_funcs *i915_ttm_driver(void)
 	return &i915_ttm_bo_driver;
 }
 
-static int i915_ttm_get_pages(struct drm_i915_gem_object *obj)
+static int __i915_ttm_get_pages(struct drm_i915_gem_object *obj,
+				struct ttm_placement *placement)
 {
 	struct ttm_buffer_object *bo = i915_gem_to_ttm(obj);
 	struct ttm_operation_ctx ctx = {
@@ -625,19 +626,12 @@ static int i915_ttm_get_pages(struct
 drm_i915_gem_object *obj)
 		.no_wait_gpu = false,
 	};
 	struct sg_table *st;
-	struct ttm_place requested, busy[I915_TTM_MAX_PLACEMENTS];
-	struct ttm_placement placement;
 	int real_num_busy;
 	int ret;
 
-	GEM_BUG_ON(obj->mm.n_placements >
 I915_TTM_MAX_PLACEMENTS);
-
-	/* Move to the requested placement. */
-	i915_ttm_placement_from_obj(obj, &requested, busy, &placement);
-
 	/* First try only the requested placement. No eviction. */
-	real_num_busy =
 fetch_and_zero(&placement.num_busy_placement);
-	ret = ttm_bo_validate(bo, &placement, &ctx);
+	real_num_busy = fetch_and_zero(&placement-
Quoted 1 line

num_busy_placement);

drivers/gpu/drm/i915/gem/i915_gem_ttm.c +1 -0
+	ret = ttm_bo_validate(bo, placement, &ctx);
	if (ret) {
		ret = i915_ttm_err_to_gem(ret);
		/*
drivers/gpu/drm/i915/gem/i915_gem_ttm.c +2 -2
@@ -652,8 +646,8 @@ static int i915_ttm_get_pages(struct
 drm_i915_gem_object *obj)
 		 * If the initial attempt fails, allow all accepted placements,
 		 * evicting if necessary.
 		 */
-		placement.num_busy_placement = real_num_busy;
-		ret = ttm_bo_validate(bo, &placement, &ctx);
+		placement->num_busy_placement = real_num_busy;
+		ret = ttm_bo_validate(bo, placement, &ctx);
 		if (ret)
 			return i915_ttm_err_to_gem(ret);
	}
drivers/gpu/drm/i915/gem/i915_gem_ttm.c +13 -5
@@ -668,16 +662,56 @@ static int i915_ttm_get_pages(struct
 drm_i915_gem_object *obj)
 		i915_ttm_adjust_gem_after_move(obj);
 	}
 
-	/* Object either has a page vector or is an iomem object */
-	st = bo->ttm ? i915_ttm_tt_get_st(bo->ttm) : obj->ttm.cached_io_st;
-	if (IS_ERR(st))
-		return PTR_ERR(st);
+	if (!obj->mm.pages) {
+		/* Object either has a page vector or is an iomem object */
+		st = bo->ttm ? i915_ttm_tt_get_st(bo->ttm) : obj-
 > ttm.cached_io_st;
+		if (IS_ERR(st))
+			return PTR_ERR(st);
 
-	__i915_gem_object_set_pages(obj, st, i915_sg_dma_sizes(st->sgl));
+		__i915_gem_object_set_pages(obj, st,
 i915_sg_dma_sizes(st->sgl));
+	}
 
 	return ret;
 }
 
+static int i915_ttm_get_pages(struct drm_i915_gem_object *obj)
+{
+	struct ttm_place requested, busy[I915_TTM_MAX_PLACEMENTS];
+	struct ttm_placement placement;
+
+	GEM_BUG_ON(obj->mm.n_placements >

I915_TTM_MAX_PLACEMENTS);

drivers/gpu/drm/i915/gem/i915_gem_ttm.c +18 -0
+
+	/* Move to the requested placement. */
+	i915_ttm_placement_from_obj(obj, &requested, busy, &placement);
+
+	return __i915_ttm_get_pages(obj, &placement);
+}
+
+static int i915_ttm_migrate(struct drm_i915_gem_object *obj,
+			    struct intel_memory_region *mr)
+{
+	struct ttm_place requested;
+	struct ttm_placement placement;
+	int ret;
+
+	i915_ttm_place_from_region(mr, &requested, obj->flags);
+	placement.num_placement = 1;
+	placement.num_busy_placement = 1;
+	placement.placement = &requested;

drivers/gpu/drm/i915/gem/i915_gem_ttm.c static int i915_ttm_get_pages(struct drm_i915_gem_object *obj) · line 710

Quoted 10 lines
drivers/gpu/drm/i915/gem/i915_gem_ttm.c +10 -0
+	placement.busy_placement = &requested;
+
+	ret = __i915_ttm_get_pages(obj, &placement);
+	if (ret)
+		return ret;
+
+	if (obj->mm.region != mr) {
+		i915_gem_object_release_memory_region(obj);
+		i915_gem_object_init_memory_region(obj, mr);
+	}

Perhaps a minor nit:

Doing this after we have done the _get_pages() just doesn't seem right.

Quoted 7 lines
Quoted 4 lines

I.e. we do work on the object, and then we init some portion of it.

Do we need to do this incase the migration/placement fails? If so, maybe a comment to that effect?

This is simply switching memory region under the lock, and to also move to another memory region list. Is it the naming _release_ and _init_ that is confusing?

Hmm, re-reading my question, I am wondering if I was unclear.

My "real" question was, can the release/init occur before the _get_pages()?

But looking at this some more, I can see answer is probably no.

Quoted 7 lines

I was going to suggest calling _init_ _set_, but when I looked at it, it was doing init things as well as setting things

Maybe just a comment like:

/* Complete the migration by updating the memory region info. */ if (object->mm...)

Sure, I'll add that. Note that another thread would have to take the object lock anyway to inspect both the region and the pages, so it shouldn't be confused by the order here.

/Thomas

Reviewed-by
#8 lore raw

RE: [PATCH v3 1/5] drm/i915/gem: Implement object migration

Quoted 411 lines

-----Original Message-----
From: Thomas Hellström <[email protected]>
Sent: Monday, June 28, 2021 3:54 PM
To: Ruhl, Michael J <[email protected]>; intel- [email protected]; [email protected]
Cc: Auld, Matthew <[email protected]>; lkp <[email protected]>
Subject: Re: [PATCH v3 1/5] drm/i915/gem: Implement object migration

Quoted 9 lines · Michael J
Quoted 9 lines

-----Original Message-----
From: Thomas Hellström <[email protected]>
Sent: Monday, June 28, 2021 3:03 PM
To: Ruhl, Michael J <[email protected]>; intel- [email protected]; [email protected]
Cc: Auld, Matthew <[email protected]>; lkp <[email protected]>
Subject: Re: [PATCH v3 1/5] drm/i915/gem: Implement object migration

Quoted 3 lines
Quoted 3 lines · Michael J
Quoted 2 lines
Quoted 2 lines

-----Original Message-----
From: dri-devel <[email protected]> On Behalf

Of

Quoted 13 lines
Quoted 13 lines
Quoted 4 lines
Quoted 4 lines

Thomas Hellström
Sent: Monday, June 28, 2021 10:46 AM
To: [email protected]; [email protected]
Cc: Thomas Hellström <[email protected]>; Auld,

Matthew

Quoted 8 lines
Quoted 8 lines

<[email protected]>; lkp <[email protected]>
Subject: [PATCH v3 1/5] drm/i915/gem: Implement object migration

Introduce an interface to migrate objects between regions. This is primarily intended to migrate objects to LMEM for display and to SYSTEM for dma-buf, but might be reused in one form or another for performance-based migration.

Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines

v2:
- Verify that the memory region given as an id really exists. (Reported by Matthew Auld)
- Call i915_gem_object_{init,release}_memory_region() when switching region
to handle also switching region lists. (Reported by Matthew Auld)
v3:
- Fix i915_gem_object_can_migrate() to return true if object is already

in

Quoted 16 lines
Quoted 16 lines
Quoted 16 lines
Quoted 16 lines

the correct region, even if the object ops doesn't have a migrate() callback.
- Update typo in commit message.
- Fix kerneldoc of i915_gem_object_wait_migration().

---
drivers/gpu/drm/i915/gem/i915_gem_object.c | 96
+++++++++++++++++++

drivers/gpu/drm/i915/gem/i915_gem_object.h    | 12 +++
.../gpu/drm/i915/gem/i915_gem_object_types.h  |  9 ++
drivers/gpu/drm/i915/gem/i915_gem_ttm.c       | 69 +++++++++----
drivers/gpu/drm/i915/gem/i915_gem_wait.c      | 19 ++++
5 files changed, 188 insertions(+), 17 deletions(-)
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines

drivers/gpu/drm/i915/gem/i915_gem_object.c bool i915_gem_object_has_iomem(const

Quoted 6 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +0 -0
index 07e8ff9a8aae..1c18be067b58 100644
@@ -513,6 +513,102 @@ bool i915_gem_object_has_iomem(const

struct

Quoted 7 lines
Quoted 7 lines
Quoted 6 lines

drivers/gpu/drm/i915/gem/i915_gem_object.c bool i915_gem_object_has_iomem(const struct drm_i915_gem_object *obj)

Quoted 6 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +2 -0
 drm_i915_gem_object *obj)
 	return obj->mem_flags & I915_BO_FLAG_IOMEM;
 }
 
+/**
+ * i915_gem_object_can_migrate - Whether an object likely can be

migrated

Quoted 10 lines
Quoted 10 lines
Quoted 10 lines

drivers/gpu/drm/i915/gem/i915_gem_object.c bool i915_gem_object_has_iomem(const struct drm_i915_gem_object *obj)

Quoted 10 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +10 -0
+ *
+ * @obj: The object to migrate
+ * @id: The region intended to migrate to
+ *
+ * Check whether the object backend supports migration to the
+ * given region. Note that pinning may affect the ability to migrate.
+ *
+ * Return: true if migration is possible, false otherwise.
+ */
+bool i915_gem_object_can_migrate(struct drm_i915_gem_object

*obj,

Quoted 47 lines
Quoted 47 lines
Quoted 47 lines
Quoted 23 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +23 -0
+				 enum intel_region_id id)
+{
+	struct drm_i915_private *i915 = to_i915(obj->base.dev);
+	unsigned int num_allowed = obj->mm.n_placements;
+	struct intel_memory_region *mr;
+	unsigned int i;
+
+	GEM_BUG_ON(id >= INTEL_REGION_UNKNOWN);
+	GEM_BUG_ON(obj->mm.madv != I915_MADV_WILLNEED);
+
+	mr = i915->mm.regions[id];
+	if (!mr)
+		return false;
+
+	if (obj->mm.region == mr)
+		return true;
+
+	if (!i915_gem_object_evictable(obj))
+		return false;
+
+	if (!obj->ops->migrate)
+		return false;
+

drivers/gpu/drm/i915/gem/i915_gem_object.c bool i915_gem_object_has_iomem(const struct drm_i915_gem_object *obj) · line 560

Quoted 10 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +10 -0
+	if (!(obj->flags & I915_BO_ALLOC_USER))
+		return true;
+
+	if (num_allowed == 0)
+		return false;
+
+	for (i = 0; i < num_allowed; ++i) {
+		if (mr == obj->mm.placements[i])
+			return true;
+	}

Hi Thomas,

I am a little confused over the difference in checks between this function and i915_gem_object_migrate().

Why is the lack of an mr a BUG_ON in _object_migrate(), but here it just false?

So that means that under certain circumstances, you could not have a mr?

If that is the case, when?

Would that be when the I915_BO_ALLOC_USER is set?

Quoted 8 lines
Quoted 8 lines
Quoted 4 lines

If so, should there be a check for "non" user vs user?

Or maybe document this function pointing out why there are differences and why?

Yes, I'll give it some more documentation. The basic idea is that the above function also could be
used to validate user supplied data, whereas there might be cases where we want to use the gem_object_migrate() function and override the

above.

Quoted 7 lines
Quoted 7 lines
Quoted 7 lines

drivers/gpu/drm/i915/gem/i915_gem_object.c bool i915_gem_object_has_iomem(const struct drm_i915_gem_object *obj)

Quoted 7 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +5 -0
+	return false;
+}
+
+/**
+ * i915_gem_object_migrate - Migrate an object to the desired region

id

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

drivers/gpu/drm/i915/gem/i915_gem_object.c bool i915_gem_object_has_iomem(const · line 568

Quoted 2 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +2 -0
+ * @obj: The object to migrate.
+ * @ww: An optional struct i915_gem_ww_ctx. If NULL, the backend

may

Quoted 1 line
Quoted 1 line
Quoted 1 line

drivers/gpu/drm/i915/gem/i915_gem_object.c bool i915_gem_object_has_iomem(const · line 569

Quoted 1 line
drivers/gpu/drm/i915/gem/i915_gem_object.c +1 -0
+ * not be successful in evicting other objects to make room for this

object.

Quoted 1 line
Quoted 1 line
Quoted 1 line

Is the ww for future consideration? (I don't see any use of it in the

patch).

Quoted 11 lines
Quoted 11 lines

Yes, but it will remain optional.

Quoted 8 lines
Quoted 8 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +8 -0
+ * @id: The region id to migrate to.
+ *
+ * Attempt to migrate the object to the desired memory region. The
+ * object backend must support migration and the object may not be
+ * pinned, (explicitly pinned pages or pinned vmas). The object must
+ * be locked.
+ * On successful completion, the object will have pages pointing to
+ * memory in the new region, but an async migration task may not

have

Quoted 5 lines
Quoted 5 lines
Quoted 5 lines

drivers/gpu/drm/i915/gem/i915_gem_object.c bool i915_gem_object_has_iomem(const · line 582

Quoted 5 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +4 -0
+ * completed yet, and to accomplish that,
 i915_gem_object_wait_migration()
+ * must be called.
+ *
+ * Return: 0 on success. Negative error code on failure. In particular

may

Quoted 54 lines
Quoted 54 lines
Quoted 1 line

drivers/gpu/drm/i915/gem/i915_gem_object.c bool i915_gem_object_has_iomem(const · line 583

Quoted 1 line
drivers/gpu/drm/i915/gem/i915_gem_object.c +1 -0
+ * return -ENXIO on lack of region space, -EDEADLK for deadlock

avoidance

Quoted 18 lines
Quoted 18 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +18 -0
+ * if @ww is set, -EINTR or -ERESTARTSYS if signal pending, and
+ * -EBUSY if the object is pinned.
+ */
+int i915_gem_object_migrate(struct drm_i915_gem_object *obj,
+			    struct i915_gem_ww_ctx *ww,
+			    enum intel_region_id id)
+{
+	struct drm_i915_private *i915 = to_i915(obj->base.dev);
+	struct intel_memory_region *mr;
+
+	GEM_BUG_ON(id >= INTEL_REGION_UNKNOWN);
+	GEM_BUG_ON(obj->mm.madv != I915_MADV_WILLNEED);
+	assert_object_held(obj);
+
+	mr = i915->mm.regions[id];
+	GEM_BUG_ON(!mr);
+
+	if (obj->mm.region == mr)
Quoted 8 lines

drivers/gpu/drm/i915/gem/i915_gem_object.c bool i915_gem_object_has_iomem(const struct drm_i915_gem_object *obj) · line 607

Quoted 7 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +7 -0
+		return 0;
+
+	if (!i915_gem_object_evictable(obj))
+		return -EBUSY;
+
+	if (!obj->ops->migrate)
+		return -EOPNOTSUPP;

Why aren't you using _can_migrate here?

It's just in case we want to override. I'll add some more comments about this.

Quoted 23 lines
Quoted 23 lines
drivers/gpu/drm/i915/gem/i915_gem_object.c +3 -0
+	return obj->ops->migrate(obj, mr);
+}
+
 void i915_gem_init__objects(struct drm_i915_private *i915)
{
	INIT_WORK(&i915->mm.free_work, __i915_gem_free_work);
drivers/gpu/drm/i915/gem/i915_gem_object.h +2 -0
index ea3224a480c4..8cbd7a5334e2 100644
@@ -17,6 +17,8 @@
 #include "i915_gem_ww.h"
 #include "i915_vma_types.h"
 
+enum intel_region_id;
+
 /*
   * XXX: There is a prevalence of the assumption that we fit the
   * object's page count inside a 32bit _signed_ variable. Let's document
@@ -597,6 +599,16 @@ bool i915_gem_object_migratable(struct
 drm_i915_gem_object *obj);
 
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

bool i915_gem_object_validates_to_lmem(struct

drm_i915_gem_object

Quoted 7 lines
Quoted 7 lines
Quoted 7 lines

drivers/gpu/drm/i915/gem/i915_gem_object.h bool i915_gem_object_migratable(struct drm_i915_gem_object *obj);

Quoted 7 lines

*obj);

drivers/gpu/drm/i915/gem/i915_gem_object.h +5 -0
+int i915_gem_object_migrate(struct drm_i915_gem_object *obj,
+			    struct i915_gem_ww_ctx *ww,
+			    enum intel_region_id id);
+
+bool i915_gem_object_can_migrate(struct drm_i915_gem_object

*obj,

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines

drivers/gpu/drm/i915/gem/i915_gem_object.h bool i915_gem_object_migratable(struct drm_i915_gem_object *obj);

Quoted 3 lines
drivers/gpu/drm/i915/gem/i915_gem_object.h +3 -0
+				 enum intel_region_id id);
+
+int i915_gem_object_wait_migration(struct drm_i915_gem_object

*obj,

Quoted 12 lines
Quoted 12 lines
Quoted 12 lines
Quoted 12 lines
drivers/gpu/drm/i915/gem/i915_gem_object.h +2 -0
+				   unsigned int flags);
+

#ifdef CONFIG_MMU_NOTIFIER
static inline bool i915_gem_object_is_userptr(struct drm_i915_gem_object *obj)

drivers/gpu/drm/i915/gem/i915_gem_object_types.h +0 -0
index 441f913c87e6..ef3de2ae9723 100644
@@ -18,6 +18,7 @@
 
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines

drivers/gpu/drm/i915/gem/i915_gem_object_types.h

Quoted 6 lines
drivers/gpu/drm/i915/gem/i915_gem_object_types.h +1 -0
 struct drm_i915_gem_object;
 struct intel_fronbuffer;
+struct intel_memory_region;
 
 /*
   * struct i915_lut_handle tracks the fast lookups from handle to vma

used

Quoted 147 lines
Quoted 131 lines
Quoted 131 lines
Quoted 117 lines
drivers/gpu/drm/i915/gem/i915_gem_object_types.h +8 -0
@@ -77,6 +78,14 @@ struct drm_i915_gem_object_ops {
 	 * delayed_free - Override the default delayed free implementation
 	 */
 	void (*delayed_free)(struct drm_i915_gem_object *obj);
+
+	/**
+	 * migrate - Migrate object to a different region either for
+	 * pinning or for as long as the object lock is held.
+	 */
+	int (*migrate)(struct drm_i915_gem_object *obj,
+		       struct intel_memory_region *mr);
+
 	void (*release)(struct drm_i915_gem_object *obj);
 
 	const struct vm_operations_struct *mmap_ops;
drivers/gpu/drm/i915/gem/i915_gem_ttm.c +3 -10
index c39d982c4fa6..8f89185b6507 100644
@@ -617,7 +617,8 @@ struct ttm_device_funcs *i915_ttm_driver(void)
 	return &i915_ttm_bo_driver;
 }
 
-static int i915_ttm_get_pages(struct drm_i915_gem_object *obj)
+static int __i915_ttm_get_pages(struct drm_i915_gem_object *obj,
+				struct ttm_placement *placement)
 {
 	struct ttm_buffer_object *bo = i915_gem_to_ttm(obj);
 	struct ttm_operation_ctx ctx = {
@@ -625,19 +626,12 @@ static int i915_ttm_get_pages(struct
 drm_i915_gem_object *obj)
 		.no_wait_gpu = false,
 	};
 	struct sg_table *st;
-	struct ttm_place requested, busy[I915_TTM_MAX_PLACEMENTS];
-	struct ttm_placement placement;
 	int real_num_busy;
 	int ret;
 
-	GEM_BUG_ON(obj->mm.n_placements >
 I915_TTM_MAX_PLACEMENTS);
-
-	/* Move to the requested placement. */
-	i915_ttm_placement_from_obj(obj, &requested, busy, &placement);
-
 	/* First try only the requested placement. No eviction. */
-	real_num_busy =
 fetch_and_zero(&placement.num_busy_placement);
-	ret = ttm_bo_validate(bo, &placement, &ctx);
+	real_num_busy = fetch_and_zero(&placement-
Quoted 1 line

num_busy_placement);

drivers/gpu/drm/i915/gem/i915_gem_ttm.c +1 -0
+	ret = ttm_bo_validate(bo, placement, &ctx);
	if (ret) {
		ret = i915_ttm_err_to_gem(ret);
		/*
drivers/gpu/drm/i915/gem/i915_gem_ttm.c +2 -2
@@ -652,8 +646,8 @@ static int i915_ttm_get_pages(struct
 drm_i915_gem_object *obj)
 		 * If the initial attempt fails, allow all accepted placements,
 		 * evicting if necessary.
 		 */
-		placement.num_busy_placement = real_num_busy;
-		ret = ttm_bo_validate(bo, &placement, &ctx);
+		placement->num_busy_placement = real_num_busy;
+		ret = ttm_bo_validate(bo, placement, &ctx);
 		if (ret)
 			return i915_ttm_err_to_gem(ret);
	}
drivers/gpu/drm/i915/gem/i915_gem_ttm.c +13 -5
@@ -668,16 +662,56 @@ static int i915_ttm_get_pages(struct
 drm_i915_gem_object *obj)
 		i915_ttm_adjust_gem_after_move(obj);
 	}
 
-	/* Object either has a page vector or is an iomem object */
-	st = bo->ttm ? i915_ttm_tt_get_st(bo->ttm) : obj->ttm.cached_io_st;
-	if (IS_ERR(st))
-		return PTR_ERR(st);
+	if (!obj->mm.pages) {
+		/* Object either has a page vector or is an iomem object */
+		st = bo->ttm ? i915_ttm_tt_get_st(bo->ttm) : obj-
 > ttm.cached_io_st;
+		if (IS_ERR(st))
+			return PTR_ERR(st);
 
-	__i915_gem_object_set_pages(obj, st, i915_sg_dma_sizes(st->sgl));
+		__i915_gem_object_set_pages(obj, st,
 i915_sg_dma_sizes(st->sgl));
+	}
 
 	return ret;
 }
 
+static int i915_ttm_get_pages(struct drm_i915_gem_object *obj)
+{
+	struct ttm_place requested, busy[I915_TTM_MAX_PLACEMENTS];
+	struct ttm_placement placement;
+
+	GEM_BUG_ON(obj->mm.n_placements >

I915_TTM_MAX_PLACEMENTS);

drivers/gpu/drm/i915/gem/i915_gem_ttm.c +18 -0
+
+	/* Move to the requested placement. */
+	i915_ttm_placement_from_obj(obj, &requested, busy, &placement);
+
+	return __i915_ttm_get_pages(obj, &placement);
+}
+
+static int i915_ttm_migrate(struct drm_i915_gem_object *obj,
+			    struct intel_memory_region *mr)
+{
+	struct ttm_place requested;
+	struct ttm_placement placement;
+	int ret;
+
+	i915_ttm_place_from_region(mr, &requested, obj->flags);
+	placement.num_placement = 1;
+	placement.num_busy_placement = 1;
+	placement.placement = &requested;

drivers/gpu/drm/i915/gem/i915_gem_ttm.c static int i915_ttm_get_pages(struct drm_i915_gem_object *obj) · line 710

Quoted 10 lines
drivers/gpu/drm/i915/gem/i915_gem_ttm.c +10 -0
+	placement.busy_placement = &requested;
+
+	ret = __i915_ttm_get_pages(obj, &placement);
+	if (ret)
+		return ret;
+
+	if (obj->mm.region != mr) {
+		i915_gem_object_release_memory_region(obj);
+		i915_gem_object_init_memory_region(obj, mr);
+	}

Perhaps a minor nit:

Doing this after we have done the _get_pages() just doesn't seem right.

Quoted 7 lines
Quoted 4 lines

I.e. we do work on the object, and then we init some portion of it.

Do we need to do this incase the migration/placement fails? If so, maybe a comment to that effect?

This is simply switching memory region under the lock, and to also move to another memory region list. Is it the naming _release_ and _init_ that is confusing?

Hmm, re-reading my question, I am wondering if I was unclear.

My "real" question was, can the release/init occur before the _get_pages()?

But looking at this some more, I can see answer is probably no.

I was going to suggest calling _init_ _set_, but when I looked at it, it was doing init things as well as setting things

Quoted 8 lines
Quoted 4 lines

Maybe just a comment like:

/* Complete the migration by updating the memory region info. */ if (object->mm...)

Sure, I'll add that. Note that another thread would have to take the object lock anyway to inspect both the region and the pages, so it shouldn't be confused by the order here.

Understood.

I think I was just disoriented by the pattern. :)

With the added documentation, I think this looks good.

Mike

Quoted 2 lines

/Thomas

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#28 lore raw

[Intel-gfx] ✓ Fi.CI.IGT: success for drm/i915/gem: Introduce a migrate interface (rev3)

In reply to Thomas Hellström
== Series Details ==

== Series Details ==

Series: drm/i915/gem: Introduce a migrate interface (rev3)
URL : https://patchwork.freedesktop.org/series/91890/
State : success

== Summary ==

CI Bug Log - changes from CI_DRM_10284_full -> Patchwork_20479_full
====================================================

Summary
-------

**SUCCESS**

No regressions found.

Possible new issues
-------------------

Here are the unknown changes that may have been introduced in Patchwork_20479_full:

### IGT changes ###

#### Suppressed ####

The following results come from untrusted machines, tests, or statuses. They do not affect the overall result.

* igt@gem_eio@hibernate:
- {shard-rkl-6}: [PASS][1] -> [FAIL][2] +1 similar issue
[1]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10284/shard-rkl-6/igt@[email protected]
[2]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-rkl-6/igt@[email protected]

* {igt@kms_ccs@pipe-a-bad-aux-stride-y_tiled_gen12_mc_ccs}:
- {shard-rkl-2}: NOTRUN -> [FAIL][3] +38 similar issues
[3]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-rkl-2/igt@kms_ccs@pipe-a-bad-aux-stride-y_tiled_gen12_mc_ccs.html

* {igt@kms_ccs@pipe-a-crc-primary-rotation-180-y_tiled_gen12_mc_ccs}:
- {shard-rkl-6}: NOTRUN -> [SKIP][4] +2 similar issues
[4]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-rkl-6/igt@kms_ccs@pipe-a-crc-primary-rotation-180-y_tiled_gen12_mc_ccs.html

* {igt@kms_ccs@pipe-c-ccs-on-another-bo-y_tiled_gen12_mc_ccs}:
- {shard-rkl-5}: NOTRUN -> [FAIL][5] +40 similar issues
[5]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-rkl-5/igt@kms_ccs@pipe-c-ccs-on-another-bo-y_tiled_gen12_mc_ccs.html

* {igt@kms_ccs@pipe-c-crc-primary-rotation-180-y_tiled_gen12_mc_ccs}:
- {shard-rkl-1}: NOTRUN -> [FAIL][6] +48 similar issues
[6]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-rkl-1/igt@kms_ccs@pipe-c-crc-primary-rotation-180-y_tiled_gen12_mc_ccs.html

* igt@kms_cursor_crc@pipe-c-cursor-32x10-sliding:
- {shard-rkl-5}: NOTRUN -> [SKIP][7] +75 similar issues
[7]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-rkl-5/igt@[email protected]

* igt@kms_flip@modeset-vs-vblank-race-interruptible:
- {shard-rkl-2}: NOTRUN -> [SKIP][8] +85 similar issues
[8]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-rkl-2/igt@[email protected]

* igt@kms_flip_scaled_crc@flip-32bpp-ytile-to-32bpp-ytilegen12rcccs:
- {shard-rkl-5}: NOTRUN -> [INCOMPLETE][9]
[9]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-rkl-5/igt@kms_flip_scaled_crc@flip-32bpp-ytile-to-32bpp-ytilegen12rcccs.html

* igt@kms_flip_scaled_crc@flip-64bpp-ytile-to-16bpp-ytile:
- {shard-rkl-2}: NOTRUN -> [INCOMPLETE][10] +2 similar issues
[10]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-rkl-2/igt@[email protected]

* igt@kms_frontbuffer_tracking@fbc-2p-scndscrn-spr-indfb-draw-blt:
- {shard-rkl-1}: NOTRUN -> [SKIP][11] +222 similar issues
[11]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-rkl-1/igt@[email protected]

* igt@runner@aborted:
- {shard-rkl-1}: [FAIL][12] ([i915#3002]) -> [FAIL][13]
[12]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10284/shard-rkl-1/igt@[email protected]
[13]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-rkl-1/igt@[email protected]

New tests
---------

New tests have been introduced between CI_DRM_10284_full and Patchwork_20479_full:

### New IGT tests (1) ###

* igt@i915_selftest@live@gem_migrate:
- Statuses : 8 pass(s)
- Exec time: [0.49, 4.97] s

Known issues
------------

Here are the changes found in Patchwork_20479_full that come from known issues:

### IGT changes ###

#### Issues hit ####

* igt@gem_ctx_persistence@smoketest:
- shard-snb: NOTRUN -> [SKIP][14] ([fdo#109271] / [i915#1099]) +5 similar issues
[14]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-snb7/igt@[email protected]

* igt@gem_eio@unwedge-stress:
- shard-snb: NOTRUN -> [FAIL][15] ([i915#3354])
[15]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-snb7/igt@[email protected]

* igt@gem_exec_fair@basic-none@vcs0:
- shard-kbl: [PASS][16] -> [FAIL][17] ([i915#2842]) +1 similar issue
[16]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10284/shard-kbl1/igt@gem_exec_fair@[email protected]
[17]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-kbl4/igt@gem_exec_fair@[email protected]

* igt@gem_exec_fair@basic-none@vcs1:
- shard-iclb: NOTRUN -> [FAIL][18] ([i915#2842]) +4 similar issues
[18]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-iclb4/igt@gem_exec_fair@[email protected]

* igt@gem_exec_fair@basic-pace-solo@rcs0:
- shard-tglb: [PASS][19] -> [FAIL][20] ([i915#2842]) +1 similar issue
[19]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10284/shard-tglb7/igt@gem_exec_fair@[email protected]
[20]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-tglb6/igt@gem_exec_fair@[email protected]

* igt@gem_exec_reloc@basic-wide-active@bcs0:
- shard-skl: NOTRUN -> [FAIL][21] ([i915#3633]) +3 similar issues
[21]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-skl8/igt@gem_exec_reloc@[email protected]

* igt@gem_pread@exhaustion:
- shard-apl: NOTRUN -> [WARN][22] ([i915#2658])
[22]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-apl6/igt@[email protected]

* igt@gem_render_copy@y-tiled-to-vebox-linear:
- shard-iclb: NOTRUN -> [SKIP][23] ([i915#768])
[23]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-iclb4/igt@[email protected]

* igt@gen9_exec_parse@bb-large:
- shard-apl: NOTRUN -> [FAIL][24] ([i915#3296])
[24]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-apl3/igt@[email protected]

* igt@i915_pm_dc@dc3co-vpb-simulation:
- shard-skl: NOTRUN -> [SKIP][25] ([fdo#109271] / [i915#658])
[25]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-skl4/igt@[email protected]

* igt@i915_pm_dc@dc6-dpms:
- shard-skl: NOTRUN -> [FAIL][26] ([i915#454])
[26]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-skl8/igt@[email protected]

* igt@i915_pm_dc@dc9-dpms:
- shard-iclb: NOTRUN -> [INCOMPLETE][27] ([i915#3698])
[27]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-iclb4/igt@[email protected]

* igt@i915_pm_lpsp@kms-lpsp@kms-lpsp-dp:
- shard-apl: NOTRUN -> [SKIP][28] ([fdo#109271] / [i915#1937])
[28]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-apl8/igt@i915_pm_lpsp@[email protected]

* igt@i915_pm_rpm@modeset-lpsp-stress:
- shard-apl: NOTRUN -> [SKIP][29] ([fdo#109271]) +297 similar issues
[29]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-apl2/igt@[email protected]

* igt@i915_suspend@forcewake:
- shard-apl: [PASS][30] -> [DMESG-WARN][31] ([i915#180])
[30]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10284/shard-apl2/igt@[email protected]
[31]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-apl8/igt@[email protected]

* igt@kms_big_fb@linear-8bpp-rotate-270:
- shard-iclb: NOTRUN -> [SKIP][32] ([fdo#110725] / [fdo#111614])
[32]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-iclb4/igt@[email protected]

* igt@kms_big_fb@y-tiled-32bpp-rotate-0:
- shard-glk: [PASS][33] -> [DMESG-WARN][34] ([i915#118] / [i915#95]) +2 similar issues
[33]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10284/shard-glk5/igt@[email protected]
[34]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-glk5/igt@[email protected]

* igt@kms_chamelium@dp-crc-multiple:
- shard-skl: NOTRUN -> [SKIP][35] ([fdo#109271] / [fdo#111827]) +7 similar issues
[35]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-skl9/igt@[email protected]

* igt@kms_chamelium@hdmi-hpd:
- shard-iclb: NOTRUN -> [SKIP][36] ([fdo#109284] / [fdo#111827])
[36]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-iclb4/igt@[email protected]

* igt@kms_chamelium@hdmi-hpd-storm-disable:
- shard-kbl: NOTRUN -> [SKIP][37] ([fdo#109271] / [fdo#111827]) +16 similar issues
[37]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-kbl3/igt@[email protected]

* igt@kms_chamelium@hdmi-mode-timings:
- shard-snb: NOTRUN -> [SKIP][38] ([fdo#109271] / [fdo#111827]) +23 similar issues
[38]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-snb6/igt@[email protected]

* igt@kms_color@pipe-c-ctm-0-5:
- shard-skl: NOTRUN -> [DMESG-WARN][39] ([i915#1982])
[39]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-skl4/igt@[email protected]

* igt@kms_color_chamelium@pipe-a-ctm-limited-range:
- shard-apl: NOTRUN -> [SKIP][40] ([fdo#109271] / [fdo#111827]) +33 similar issues
[40]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-apl6/igt@[email protected]

* igt@kms_content_protection@atomic-dpms:
- shard-apl: NOTRUN -> [TIMEOUT][41] ([i915#1319]) +2 similar issues
[41]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-apl2/igt@[email protected]
- shard-kbl: NOTRUN -> [TIMEOUT][42] ([i915#1319])
[42]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-kbl6/igt@[email protected]

* igt@kms_content_protection@uevent:
- shard-kbl: NOTRUN -> [FAIL][43] ([i915#2105])
[43]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-kbl3/igt@[email protected]
- shard-apl: NOTRUN -> [FAIL][44] ([i915#2105])
[44]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-apl3/igt@[email protected]

* igt@kms_cursor_crc@pipe-a-cursor-32x32-onscreen:
- shard-iclb: NOTRUN -> [SKIP][45] ([fdo#109278]) +2 similar issues
[45]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-iclb4/igt@[email protected]

* igt@kms_draw_crc@draw-method-xrgb8888-render-ytiled:
- shard-skl: [PASS][46] -> [DMESG-WARN][47] ([i915#1982]) +1 similar issue
[46]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10284/shard-skl1/igt@[email protected]
[47]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-skl10/igt@[email protected]

* igt@kms_flip@2x-flip-vs-dpms:
- shard-iclb: NOTRUN -> [SKIP][48] ([fdo#109274])
[48]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-iclb4/igt@[email protected]

* igt@kms_flip@flip-vs-expired-vblank@a-hdmi-a2:
- shard-glk: [PASS][49] -> [FAIL][50] ([i915#79])
[49]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10284/shard-glk8/igt@kms_flip@[email protected]
[50]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-glk2/igt@kms_flip@[email protected]

* igt@kms_flip@flip-vs-suspend-interruptible@a-dp1:
- shard-kbl: [PASS][51] -> [DMESG-WARN][52] ([i915#180]) +8 similar issues
[51]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10284/shard-kbl2/igt@kms_flip@[email protected]
[52]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-kbl6/igt@kms_flip@[email protected]

* igt@kms_flip@flip-vs-suspend@a-dp1:
- shard-kbl: NOTRUN -> [DMESG-WARN][53] ([i915#180]) +1 similar issue
[53]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-kbl1/igt@kms_flip@[email protected]

* igt@kms_flip@plain-flip-fb-recreate@a-edp1:
- shard-skl: [PASS][54] -> [FAIL][55] ([i915#2122]) +1 similar issue
[54]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10284/shard-skl3/igt@kms_flip@[email protected]
[55]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-skl5/igt@kms_flip@[email protected]

* igt@kms_flip_scaled_crc@flip-64bpp-ytile-to-32bpp-ytilercccs:
- shard-iclb: NOTRUN -> [SKIP][56] ([i915#2587])
[56]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-iclb4/igt@[email protected]

* igt@kms_frontbuffer_tracking@fbc-2p-primscrn-pri-indfb-draw-render:
- shard-iclb: NOTRUN -> [SKIP][57] ([fdo#109280]) +2 similar issues
[57]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-iclb4/igt@kms_frontbuffer_tracking@fbc-2p-primscrn-pri-indfb-draw-render.html

* igt@kms_frontbuffer_tracking@fbc-2p-scndscrn-pri-indfb-draw-blt:
- shard-skl: NOTRUN -> [SKIP][58] ([fdo#109271]) +89 similar issues
[58]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-skl9/igt@[email protected]

* igt@kms_frontbuffer_tracking@fbcpsr-2p-primscrn-cur-indfb-draw-mmap-wc:
- shard-kbl: NOTRUN -> [SKIP][59] ([fdo#109271]) +138 similar issues
[59]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-kbl3/igt@kms_frontbuffer_tracking@fbcpsr-2p-primscrn-cur-indfb-draw-mmap-wc.html

* igt@kms_frontbuffer_tracking@psr-rgb101010-draw-pwrite:
- shard-skl: [PASS][60] -> [FAIL][61] ([i915#49])
[60]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10284/shard-skl10/igt@[email protected]
[61]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-skl7/igt@[email protected]

* igt@kms_hdr@bpc-switch:
- shard-skl: NOTRUN -> [FAIL][62] ([i915#1188])
[62]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-skl9/igt@[email protected]

* igt@kms_hdr@bpc-switch-suspend:
- shard-skl: [PASS][63] -> [FAIL][64] ([i915#1188])
[63]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10284/shard-skl4/igt@[email protected]
[64]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-skl6/igt@[email protected]

* igt@kms_pipe_crc_basic@disable-crc-after-crtc-pipe-d:
- shard-kbl: NOTRUN -> [SKIP][65] ([fdo#109271] / [i915#533])
[65]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-kbl3/igt@[email protected]

* igt@kms_pipe_crc_basic@hang-read-crc-pipe-d:
- shard-skl: NOTRUN -> [SKIP][66] ([fdo#109271] / [i915#533]) +1 similar issue
[66]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-skl9/igt@[email protected]

* igt@kms_plane_alpha_blend@pipe-a-alpha-opaque-fb:
- shard-apl: NOTRUN -> [FAIL][67] ([fdo#108145] / [i915#265]) +6 similar issues
[67]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-apl2/igt@[email protected]

* igt@kms_plane_alpha_blend@pipe-a-alpha-transparent-fb:
- shard-skl: NOTRUN -> [FAIL][68] ([i915#265])
[68]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-skl4/igt@[email protected]

* igt@kms_plane_alpha_blend@pipe-a-constant-alpha-max:
- shard-skl: NOTRUN -> [FAIL][69] ([fdo#108145] / [i915#265])
[69]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-skl9/igt@[email protected]

* igt@kms_plane_alpha_blend@pipe-b-alpha-opaque-fb:
- shard-kbl: NOTRUN -> [FAIL][70] ([fdo#108145] / [i915#265]) +1 similar issue
[70]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-kbl3/igt@[email protected]

* igt@kms_plane_alpha_blend@pipe-c-alpha-transparent-fb:
- shard-kbl: NOTRUN -> [FAIL][71] ([i915#265])
[71]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-kbl4/igt@[email protected]

* igt@kms_plane_alpha_blend@pipe-c-coverage-7efc:
- shard-skl: [PASS][72] -> [FAIL][73] ([fdo#108145] / [i915#265]) +1 similar issue
[72]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10284/shard-skl9/igt@[email protected]
[73]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-skl4/igt@[email protected]

* igt@kms_psr2_sf@overlay-plane-update-sf-dmg-area-2:
- shard-apl: NOTRUN -> [SKIP][74] ([fdo#109271] / [i915#658]) +5 similar issues
[74]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-apl6/igt@[email protected]

* igt@kms_psr2_sf@overlay-primary-update-sf-dmg-area-3:
- shard-kbl: NOTRUN -> [SKIP][75] ([fdo#109271] / [i915#658]) +3 similar issues
[75]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-kbl6/igt@[email protected]

* igt@kms_psr@psr2_cursor_plane_onoff:
- shard-iclb: [PASS][76] -> [SKIP][77] ([fdo#109441]) +2 similar issues
[76]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10284/shard-iclb2/igt@kms_psr@psr2_cursor_plane_onoff.html
[77]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-iclb6/igt@kms_psr@psr2_cursor_plane_onoff.html

* igt@kms_vblank@pipe-a-ts-continuation-suspend:
- shard-kbl: [PASS][78] -> [DMESG-WARN][79] ([i915#180] / [i915#295])
[78]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10284/shard-kbl4/igt@[email protected]
[79]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-kbl1/igt@[email protected]

* igt@kms_vblank@pipe-d-query-forked-hang:
- shard-snb: NOTRUN -> [SKIP][80] ([fdo#109271]) +385 similar issues
[80]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-snb7/igt@[email protected]

* igt@kms_vblank@pipe-d-wait-idle:
- shard-apl: NOTRUN -> [SKIP][81] ([fdo#109271] / [i915#533]) +1 similar issue
[81]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-apl3/igt@[email protected]

* igt@kms_writeback@writeback-fb-id:
- shard-kbl: NOTRUN -> [SKIP][82] ([fdo#109271] / [i915#2437]) +1 similar issue
[82]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-kbl4/igt@[email protected]

* igt@perf@polling-parameterized:
- shard-skl: [PASS][83] -> [FAIL][84] ([i915#1542])
[83]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10284/shard-skl8/igt@[email protected]
[84]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-skl1/igt@[email protected]

* igt@sysfs_clients@fair-1:
- shard-apl: NOTRUN -> [SKIP][85] ([fdo#109271] / [i915#2994]) +4 similar issues
[85]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-apl3/igt@[email protected]

* igt@sysfs_clients@sema-10:
- shard-skl: NOTRUN -> [SKIP][86] ([fdo#109271] / [i915#2994]) +1 similar issue
[86]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-skl4/igt@[email protected]

* igt@sysfs_clients@split-50:
- shard-kbl: NOTRUN -> [SKIP][87] ([fdo#109271] / [i915#2994]) +1 similar issue
[87]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-kbl6/igt@[email protected]

#### Possible fixes ####

* igt@gem_eio@in-flight-contexts-10ms:
- shard-tglb: [TIMEOUT][88] ([i915#3063]) -> [PASS][89]
[88]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10284/shard-tglb2/igt@[email protected]
[89]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-tglb5/igt@[email protected]

* igt@gem_eio@in-flight-suspend:
- shard-skl: [INCOMPLETE][90] ([i915#198]) -> [PASS][91]
[90]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10284/shard-skl2/igt@[email protected]
[91]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-skl4/igt@[email protected]

* igt@gem_exec_fair@basic-deadline:
- shard-glk: [FAIL][92] ([i915#2846]) -> [PASS][93]
[92]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10284/shard-glk8/igt@[email protected]
[93]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-glk2/igt@[email protected]

* igt@gem_exec_fair@basic-none@vcs1:
- shard-kbl: [FAIL][94] ([i915#2842]) -> [PASS][95]
[94]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10284/shard-kbl1/igt@gem_exec_fair@[email protected]
[95]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-kbl4/igt@gem_exec_fair@[email protected]

* igt@gem_exec_fair@basic-pace@bcs0:
- shard-tglb: [FAIL][96] ([i915#2842]) -> [PASS][97]
[96]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10284/shard-tglb5/igt@gem_exec_fair@[email protected]
[97]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-tglb3/igt@gem_exec_fair@[email protected]
- shard-iclb: [FAIL][98] ([i915#2842]) -> [PASS][99]
[98]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10284/shard-iclb7/igt@gem_exec_fair@[email protected]
[99]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-iclb8/igt@gem_exec_fair@[email protected]

* igt@gem_exec_whisper@basic-queues-forked:
- shard-iclb: [INCOMPLETE][100] ([i915#1895] / [i915#2405]) -> [PASS][101]
[100]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10284/shard-iclb7/igt@[email protected]
[101]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-iclb4/igt@[email protected]

* igt@gem_mmap_gtt@big-copy-xy:
- shard-skl: [FAIL][102] ([i915#307]) -> [PASS][103]
[102]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10284/shard-skl4/igt@[email protected]
[103]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-skl6/igt@[email protected]

* igt@gem_mmap_gtt@cpuset-big-copy:
- shard-iclb: [FAIL][104] ([i915#307]) -> [PASS][105]
[104]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10284/shard-iclb5/igt@[email protected]
[105]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-iclb2/igt@[email protected]

* igt@gen9_exec_parse@allowed-single:
- shard-skl: [DMESG-WARN][106] ([i915#1436] / [i915#716]) -> [PASS][107]
[106]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10284/shard-skl10/igt@[email protected]
[107]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-skl9/igt@[email protected]

* igt@kms_big_fb@x-tiled-32bpp-rotate-0:
- shard-glk: [DMESG-WARN][108] ([i915#118] / [i915#95]) -> [PASS][109]
[108]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10284/shard-glk7/igt@[email protected]
[109]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-glk7/igt@[email protected]

* igt@kms_cursor_crc@pipe-a-cursor-suspend:
- shard-kbl: [DMESG-WARN][110] ([i915#180]) -> [PASS][111] +4 similar issues
[110]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10284/shard-kbl1/igt@[email protected]
[111]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-kbl1/igt@[email protected]

* igt@kms_cursor_legacy@flip-vs-cursor-atomic-transitions:
- shard-skl: [FAIL][112] ([i915#2346]) -> [PASS][113]
[112]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10284/shard-skl10/igt@[email protected]
[113]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-skl6/igt@[email protected]

* igt@kms_cursor_legacy@flip-vs-cursor-atomic-transitions-varying-size:
- shard-skl: [FAIL][114] ([i915#2346] / [i915#533]) -> [PASS][115]
[114]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10284/shard-skl7/igt@kms_cursor_legacy@flip-vs-cursor-atomic-transitions-varying-size.html
[115]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-skl2/igt@kms_cursor_legacy@flip-vs-cursor-atomic-transitions-varying-size.html

* igt@kms_fbcon_fbt@fbc-suspend:
- shard-kbl: [INCOMPLETE][116] ([i915#155] / [i915#180] / [i915#636]) -> [PASS][117]
[116]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10284/shard-kbl3/igt@[email protected]
[117]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-kbl6/igt@[email protected]

* igt@kms_flip@flip-vs-expired-vblank-interruptible@a-hdmi-a2:
- shard-glk: [FAIL][118] ([i915#79]) -> [PASS][119]
[118]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10284/shard-glk7/igt@kms_flip@[email protected]
[119]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-glk7/igt@kms_flip@[email protected]

* igt@kms_flip@flip-vs-suspend-interruptible@b-edp1:
- shard-skl: [INCOMPLETE][120] ([i915#146] / [i915#198]) -> [PASS][121]
[120]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10284/shard-skl2/igt@kms_flip@[email protected]
[121]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-skl8/igt@kms_flip@[email protected]

* igt@kms_plane_alpha_blend@pipe-b-coverage-7efc:
- shard-skl: [FAIL][122] ([fdo#108145] / [i915#265]) -> [PASS][123] +1 similar issue
[122]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10284/shard-skl2/igt@[email protected]
[123]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-skl2/igt@[email protected]

* igt@kms_psr@psr2_sprite_plane_move:
- shard-iclb: [SKIP][124] ([fdo#109441]) -> [PASS][125] +3 similar issues
[124]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10284/shard-iclb7/igt@kms_psr@psr2_sprite_plane_move.html
[125]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-iclb2/igt@kms_psr@psr2_sprite_plane_move.html

* igt@perf@blocking:
- shard-skl: [FAIL][126] ([i915#1542]) -> [PASS][127]
[126]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10284/shard-skl6/igt@[email protected]
[127]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-skl3/igt@[email protected]

#### Warnings ####

* igt@i915_pm_dc@dc3co-vpb-simulation:
- shard-iclb: [SKIP][128] ([i915#658]) -> [SKIP][129] ([i915#588])
[128]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10284/shard-iclb7/igt@[email protected]
[129]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-iclb2/igt@[email protected]

* igt@i915_pm_rc6_residency@rc6-idle:
- shard-iclb: [WARN][130] ([i915#2684]) -> [WARN][131] ([i915#1804] / [i915#2684])
[130]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10284/shard-iclb1/igt@[email protected]
[131]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-iclb7/igt@[email protected]

* igt@kms_dp_dsc@basic-dsc-enable-edp:
- shard-iclb: [SKIP][132] ([fdo#109349]) -> [DMESG-WARN][133] ([i915#1226])
[132]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10284/shard-iclb5/igt@[email protected]
[133]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-iclb2/igt@[email protected]

* igt@kms_psr2_sf@overlay-plane-update-sf-dmg-area-1:
- shard-iclb: [SKIP][134] ([i915#2920]) -> [SKIP][135] ([i915#658]) +2 similar issues
[134]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10284/shard-iclb2/igt@[email protected]
[135]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-iclb6/igt@[email protected]

* igt@kms_psr2_sf@plane-move-sf-dmg-area-1:
- shard-iclb: [SKIP][136] ([i915#658]) -> [SKIP][137] ([i915#2920])
[136]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10284/shard-iclb7/igt@[email protected]
[137]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/shard-iclb2/igt@[email protected]

* igt@runner@aborted:
- shard-kbl: ([FAIL][138], [FAIL][139], [FAIL][140], [FAIL][141], [FAIL][142], [FAIL][143], [FAIL][144], [FAIL][145]) ([i915#1436] / [i915#180] / [i915#1814] / [i915#2505] / [i915#3002] / [i915#3363] / [i915#92]) -> ([FAIL][146], [FAIL][147], [FAIL][148], [FAIL][149], [FAIL][150], [FAIL][151], [FAIL][152], [FAIL][153], [FAIL][154]) ([fdo#109271] / [i915#1436] / [i915#180] / [i915#1814] / [i915#2505] / [i915#3002] / [i915#3363] / [i915#602])
[138]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_1

== Logs ==

For more details see: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_20479/index.html

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#20 lore raw

Re: [PATCH v3 4/5] drm/i915/gem: Fix same-driver-another-instance dma-buf export

In reply to Ruhl, Michael J
Quoted 16 lines · Michael J
Quoted 16 lines

-----Original Message-----
From: Thomas Hellström <[email protected]>
Sent: Monday, June 28, 2021 10:46 AM
To: [email protected]; [email protected]
Cc: Auld, Matthew <[email protected]>; [email protected]; Thomas Hellström <[email protected]>; Ruhl; Ruhl, Michael J <[email protected]>
Subject: [PATCH v3 4/5] drm/i915/gem: Fix same-driver-another-instance dma-buf export

If our exported dma-bufs are imported by another instance of our driver, that instance will typically have the imported dma-bufs locked during map_attachment(). But the exporter also locks the same reservation object in the map_dma_buf() callback, which leads to recursive locking.

Quoted 8 lines
Quoted 6 lines

Add a live selftest to catch this case, and as a workaround until we fully support dynamic import and export, declare the exporter dynamic by providing NOP pin() and unpin() functions. This means our map_dma_buf() callback will *always* get called locked, and by pinning unconditionally in i915_gem_map_dma_buf() we make sure we don't need to use the move_notify() functionality which is not yet implemented.

An interesting abuse of the interface, but it seems reasonable (for now) to me.

Hm I'm not sure this is the best interface abuse, since if we combine this with amdgpu it goes boom. Also I thought the dynamic stuff is optional (or is that only for the importer).

What I discussed a bit with Maarten on irc is to essentially emulate the rules of what a dynamic exporter would end up with with a non-dynamic
importer: pin/unpin the buffer at attach/detach time. We could fake this in our attach/detach callbacks.

At least I don't think it's the locking changes that saves us here, but the caching of the sgt list in attach/detach. As long as we hand-roll that we should be fine. So hand-rolling that feels like the best option to make sure we're not making this worse, as long as we haven't fully validated the true dynamic importer _and_ exporter case.

Cheers, Daniel

Quoted 211 lines

Mike

Quoted 206 lines

---
drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c | 31 ++++++-
.../drm/i915/gem/selftests/i915_gem_dmabuf.c | 81
++++++++++++++++++-

2 files changed, 108 insertions(+), 4 deletions(-)
drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c +2 -0
index 616c3a2f1baf..1d1eeb167d28 100644
@@ -12,6 +12,8 @@
 #include "i915_gem_object.h"
 #include "i915_scatterlist.h"
 
+I915_SELFTEST_DECLARE(static bool force_different_devices;)
+
 static struct drm_i915_gem_object *dma_buf_to_obj(struct dma_buf *buf)
 {
 	return to_intel_bo(buf->priv);
@@ -25,7 +27,9 @@ static struct sg_table *i915_gem_map_dma_buf(struct

dma_buf_attachment *attachme

	struct scatterlist *src, *dst;
	int ret, i;
drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c +3 -1
-	ret = i915_gem_object_pin_pages_unlocked(obj);
+	assert_object_held(obj);
+
+	ret = i915_gem_object_pin_pages(obj);
	if (ret)
		goto err;
drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c +3 -0
@@ -168,6 +172,26 @@ static int i915_gem_end_cpu_access(struct dma_buf
 *dma_buf, enum dma_data_direct
 	return err;
 }
 
+/*
+ * As a workaround until we fully support dynamic import and export,
+ * declare the exporter dynamic by providing NOP pin() and unpin()

functions.
+ * This means our i915_gem_map_dma_buf() callback will *always* get called
+ * locked, and by pinning unconditionally in i915_gem_map_dma_buf() we make
+ * sure we don't need to use the move_notify() functionality which is
+ * not yet implemented. Typically for the same-driver-another-instance case,
+ * i915_gem_map_dma_buf() will be called at importer attach time and the
+ * mapped sg_list will be cached by the dma-buf core for the
+ * duration of the attachment.
+ */

drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c +9 -0
+static int i915_gem_dmabuf_pin(struct dma_buf_attachment *attach)
+{
+	return 0;
+}
+
+static void i915_gem_dmabuf_unpin(struct dma_buf_attachment *attach)
+{
+}
+
static const struct dma_buf_ops i915_dmabuf_ops =  {
	.map_dma_buf = i915_gem_map_dma_buf,
	.unmap_dma_buf = i915_gem_unmap_dma_buf,
drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c +2 -0
@@ -177,6 +201,8 @@ static const struct dma_buf_ops i915_dmabuf_ops =  {
 	.vunmap = i915_gem_dmabuf_vunmap,
 	.begin_cpu_access = i915_gem_begin_cpu_access,
 	.end_cpu_access = i915_gem_end_cpu_access,
+	.pin = i915_gem_dmabuf_pin,
+	.unpin = i915_gem_dmabuf_unpin,
 };
 

struct dma_buf *i915_gem_prime_export(struct drm_gem_object *gem_obj, int flags)

drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c +2 -1
@@ -241,7 +267,8 @@ struct drm_gem_object
 *i915_gem_prime_import(struct drm_device *dev,
 	if (dma_buf->ops == &i915_dmabuf_ops) {
 		obj = dma_buf_to_obj(dma_buf);
 		/* is it from our device? */
-		if (obj->base.dev == dev) {
+		if (obj->base.dev == dev &&
+		    !I915_SELFTEST_ONLY(force_different_devices)) {
 			/*
 			 * Importing dmabuf exported from out own gem

increases
* refcount on gem itself instead of f_count of
dmabuf.

drivers/gpu/drm/i915/gem/selftests/i915_gem_dmabuf.c +10 -1
index dd74bc09ec88..24735d6c12a2 100644
@@ -35,7 +35,7 @@ static int igt_dmabuf_export(void *arg)
 static int igt_dmabuf_import_self(void *arg)
 {
 	struct drm_i915_private *i915 = arg;
-	struct drm_i915_gem_object *obj;
+	struct drm_i915_gem_object *obj, *import_obj;
 	struct drm_gem_object *import;
 	struct dma_buf *dmabuf;
 	int err;
@@ -65,14 +65,90 @@ static int igt_dmabuf_import_self(void *arg)
 		err = -EINVAL;
 		goto out_import;
 	}
+	import_obj = to_intel_bo(import);
+
+	i915_gem_object_lock(import_obj, NULL);
+	err = ____i915_gem_object_get_pages(import_obj);
+	i915_gem_object_unlock(import_obj);
+	if (err) {
+		pr_err("Same object dma-buf get_pages failed!\n");
+		goto out_import;
+	}
 
 	err = 0;

out_import:

drivers/gpu/drm/i915/gem/selftests/i915_gem_dmabuf.c +66 -1
-	i915_gem_object_put(to_intel_bo(import));
+	i915_gem_object_put(import_obj);
+out_dmabuf:
+	dma_buf_put(dmabuf);
+out:
+	i915_gem_object_put(obj);
+	return err;
+}
+
+static int igt_dmabuf_import_same_driver(void *arg)
+{
+	struct drm_i915_private *i915 = arg;
+	struct drm_i915_gem_object *obj, *import_obj;
+	struct drm_gem_object *import;
+	struct dma_buf *dmabuf;
+	int err;
+
+	force_different_devices = true;
+	obj = i915_gem_object_create_shmem(i915, PAGE_SIZE);
+	if (IS_ERR(obj))
+		goto out_ret;
+
+	dmabuf = i915_gem_prime_export(&obj->base, 0);
+	if (IS_ERR(dmabuf)) {
+		pr_err("i915_gem_prime_export failed with err=%d\n",
+		       (int)PTR_ERR(dmabuf));
+		err = PTR_ERR(dmabuf);
+		goto out;
+	}
+
+	import = i915_gem_prime_import(&i915->drm, dmabuf);
+	if (IS_ERR(import)) {
+		pr_err("i915_gem_prime_import failed with err=%d\n",
+		       (int)PTR_ERR(import));
+		err = PTR_ERR(import);
+		goto out_dmabuf;
+	}
+
+	if (import == &obj->base) {
+		pr_err("i915_gem_prime_import reused gem object!\n");
+		err = -EINVAL;
+		goto out_import;
+	}
+
+	import_obj = to_intel_bo(import);
+
+	i915_gem_object_lock(import_obj, NULL);
+	err = ____i915_gem_object_get_pages(import_obj);
+	if (err) {
+		pr_err("Different objects dma-buf get_pages failed!\n");
+		i915_gem_object_unlock(import_obj);
+		goto out_import;
+	}
+
+	/*
+	 * If the exported object is not in system memory, something
+	 * weird is going on. TODO: When p2p is supported, this is no
+	 * longer considered weird.
+	 */
+	if (obj->mm.region != i915->mm.regions[INTEL_REGION_SMEM]) {
+		pr_err("Exported dma-buf is not in system memory\n");
+		err = -EINVAL;
+	}
+	i915_gem_object_unlock(import_obj);
+
+out_import:
+	i915_gem_object_put(import_obj);

out_dmabuf:

	dma_buf_put(dmabuf);
out:
	i915_gem_object_put(obj);
drivers/gpu/drm/i915/gem/selftests/i915_gem_dmabuf.c +2 -0
+out_ret:
+	force_different_devices = false;
	return err;
}
drivers/gpu/drm/i915/gem/selftests/i915_gem_dmabuf.c +1 -0
@@ -286,6 +362,7 @@ int i915_gem_dmabuf_live_selftests(struct
 drm_i915_private *i915)
 {
 	static const struct i915_subtest tests[] = {
 		SUBTEST(igt_dmabuf_export),
+		SUBTEST(igt_dmabuf_import_same_driver),
 	};
 
	return i915_subtests(tests, i915);

--
2.31.1

Signature · 3 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
Acked-by on v3 1/5
#9 lore raw

Re: [Intel-gfx] [PATCH v3 1/5] drm/i915/gem: Implement object migration

In reply to Thomas Hellström
Quoted 17 lines · Thomas Hellström

Introduce an interface to migrate objects between regions. This is primarily intended to migrate objects to LMEM for display and to SYSTEM for dma-buf, but might be reused in one form or another for performance-based migration.

v2:
- Verify that the memory region given as an id really exists. (Reported by Matthew Auld)
- Call i915_gem_object_{init,release}_memory_region() when switching region to handle also switching region lists. (Reported by Matthew Auld)
v3:
- Fix i915_gem_object_can_migrate() to return true if object is already in the correct region, even if the object ops doesn't have a migrate() callback.
- Update typo in commit message.
- Fix kerneldoc of i915_gem_object_wait_migration().

Quoted 2 lines

Looks reasonable to me as the main interface. I'm a bit on the fence on hiding everything behind obj->ops, since we're not going to have another implementation of migrate than the ttm one. But also while everything is in-flight it's probably good to at least try and maintain some boundaries, for otherwise the messiness might become unmanageable

But don't count this as real review :-)

One suggestion at the very bottom.

Quoted 304 lines
---
 drivers/gpu/drm/i915/gem/i915_gem_object.c    | 96 +++++++++++++++++++
 drivers/gpu/drm/i915/gem/i915_gem_object.h    | 12 +++
 .../gpu/drm/i915/gem/i915_gem_object_types.h  |  9 ++
 drivers/gpu/drm/i915/gem/i915_gem_ttm.c       | 69 +++++++++----
 drivers/gpu/drm/i915/gem/i915_gem_wait.c      | 19 ++++
 5 files changed, 188 insertions(+), 17 deletions(-)
drivers/gpu/drm/i915/gem/i915_gem_object.c +96 -0
index 07e8ff9a8aae..1c18be067b58 100644
@@ -513,6 +513,102 @@ bool i915_gem_object_has_iomem(const struct drm_i915_gem_object *obj)
 	return obj->mem_flags & I915_BO_FLAG_IOMEM;
 }
 
+/**
+ * i915_gem_object_can_migrate - Whether an object likely can be migrated
+ *
+ * @obj: The object to migrate
+ * @id: The region intended to migrate to
+ *
+ * Check whether the object backend supports migration to the
+ * given region. Note that pinning may affect the ability to migrate.
+ *
+ * Return: true if migration is possible, false otherwise.
+ */
+bool i915_gem_object_can_migrate(struct drm_i915_gem_object *obj,
+				 enum intel_region_id id)
+{
+	struct drm_i915_private *i915 = to_i915(obj->base.dev);
+	unsigned int num_allowed = obj->mm.n_placements;
+	struct intel_memory_region *mr;
+	unsigned int i;
+
+	GEM_BUG_ON(id >= INTEL_REGION_UNKNOWN);
+	GEM_BUG_ON(obj->mm.madv != I915_MADV_WILLNEED);
+
+	mr = i915->mm.regions[id];
+	if (!mr)
+		return false;
+
+	if (obj->mm.region == mr)
+		return true;
+
+	if (!i915_gem_object_evictable(obj))
+		return false;
+
+	if (!obj->ops->migrate)
+		return false;
+
+	if (!(obj->flags & I915_BO_ALLOC_USER))
+		return true;
+
+	if (num_allowed == 0)
+		return false;
+
+	for (i = 0; i < num_allowed; ++i) {
+		if (mr == obj->mm.placements[i])
+			return true;
+	}
+
+	return false;
+}
+
+/**
+ * i915_gem_object_migrate - Migrate an object to the desired region id
+ * @obj: The object to migrate.
+ * @ww: An optional struct i915_gem_ww_ctx. If NULL, the backend may
+ * not be successful in evicting other objects to make room for this object.
+ * @id: The region id to migrate to.
+ *
+ * Attempt to migrate the object to the desired memory region. The
+ * object backend must support migration and the object may not be
+ * pinned, (explicitly pinned pages or pinned vmas). The object must
+ * be locked.
+ * On successful completion, the object will have pages pointing to
+ * memory in the new region, but an async migration task may not have
+ * completed yet, and to accomplish that, i915_gem_object_wait_migration()
+ * must be called.
+ *
+ * Return: 0 on success. Negative error code on failure. In particular may
+ * return -ENXIO on lack of region space, -EDEADLK for deadlock avoidance
+ * if @ww is set, -EINTR or -ERESTARTSYS if signal pending, and
+ * -EBUSY if the object is pinned.
+ */
+int i915_gem_object_migrate(struct drm_i915_gem_object *obj,
+			    struct i915_gem_ww_ctx *ww,
+			    enum intel_region_id id)
+{
+	struct drm_i915_private *i915 = to_i915(obj->base.dev);
+	struct intel_memory_region *mr;
+
+	GEM_BUG_ON(id >= INTEL_REGION_UNKNOWN);
+	GEM_BUG_ON(obj->mm.madv != I915_MADV_WILLNEED);
+	assert_object_held(obj);
+
+	mr = i915->mm.regions[id];
+	GEM_BUG_ON(!mr);
+
+	if (obj->mm.region == mr)
+		return 0;
+
+	if (!i915_gem_object_evictable(obj))
+		return -EBUSY;
+
+	if (!obj->ops->migrate)
+		return -EOPNOTSUPP;
+
+	return obj->ops->migrate(obj, mr);
+}
+
 void i915_gem_init__objects(struct drm_i915_private *i915)
 {
 	INIT_WORK(&i915->mm.free_work, __i915_gem_free_work);
drivers/gpu/drm/i915/gem/i915_gem_object.h +12 -0
index ea3224a480c4..8cbd7a5334e2 100644
@@ -17,6 +17,8 @@
 #include "i915_gem_ww.h"
 #include "i915_vma_types.h"
 
+enum intel_region_id;
+
 /*
  * XXX: There is a prevalence of the assumption that we fit the
  * object's page count inside a 32bit _signed_ variable. Let's document
@@ -597,6 +599,16 @@ bool i915_gem_object_migratable(struct drm_i915_gem_object *obj);
 
 bool i915_gem_object_validates_to_lmem(struct drm_i915_gem_object *obj);
 
+int i915_gem_object_migrate(struct drm_i915_gem_object *obj,
+			    struct i915_gem_ww_ctx *ww,
+			    enum intel_region_id id);
+
+bool i915_gem_object_can_migrate(struct drm_i915_gem_object *obj,
+				 enum intel_region_id id);
+
+int i915_gem_object_wait_migration(struct drm_i915_gem_object *obj,
+				   unsigned int flags);
+
 #ifdef CONFIG_MMU_NOTIFIER
 static inline bool
 i915_gem_object_is_userptr(struct drm_i915_gem_object *obj)
drivers/gpu/drm/i915/gem/i915_gem_object_types.h +9 -0
index 441f913c87e6..ef3de2ae9723 100644
@@ -18,6 +18,7 @@
 
 struct drm_i915_gem_object;
 struct intel_fronbuffer;
+struct intel_memory_region;
 
 /*
  * struct i915_lut_handle tracks the fast lookups from handle to vma used
@@ -77,6 +78,14 @@ struct drm_i915_gem_object_ops {
 	 * delayed_free - Override the default delayed free implementation
 	 */
 	void (*delayed_free)(struct drm_i915_gem_object *obj);
+
+	/**
+	 * migrate - Migrate object to a different region either for
+	 * pinning or for as long as the object lock is held.
+	 */
+	int (*migrate)(struct drm_i915_gem_object *obj,
+		       struct intel_memory_region *mr);
+
 	void (*release)(struct drm_i915_gem_object *obj);
 
 	const struct vm_operations_struct *mmap_ops;
drivers/gpu/drm/i915/gem/i915_gem_ttm.c +52 -17
index c39d982c4fa6..8f89185b6507 100644
@@ -617,7 +617,8 @@ struct ttm_device_funcs *i915_ttm_driver(void)
 	return &i915_ttm_bo_driver;
 }
 
-static int i915_ttm_get_pages(struct drm_i915_gem_object *obj)
+static int __i915_ttm_get_pages(struct drm_i915_gem_object *obj,
+				struct ttm_placement *placement)
 {
 	struct ttm_buffer_object *bo = i915_gem_to_ttm(obj);
 	struct ttm_operation_ctx ctx = {
@@ -625,19 +626,12 @@ static int i915_ttm_get_pages(struct drm_i915_gem_object *obj)
 		.no_wait_gpu = false,
 	};
 	struct sg_table *st;
-	struct ttm_place requested, busy[I915_TTM_MAX_PLACEMENTS];
-	struct ttm_placement placement;
 	int real_num_busy;
 	int ret;
 
-	GEM_BUG_ON(obj->mm.n_placements > I915_TTM_MAX_PLACEMENTS);
-
-	/* Move to the requested placement. */
-	i915_ttm_placement_from_obj(obj, &requested, busy, &placement);
-
 	/* First try only the requested placement. No eviction. */
-	real_num_busy = fetch_and_zero(&placement.num_busy_placement);
-	ret = ttm_bo_validate(bo, &placement, &ctx);
+	real_num_busy = fetch_and_zero(&placement->num_busy_placement);
+	ret = ttm_bo_validate(bo, placement, &ctx);
 	if (ret) {
 		ret = i915_ttm_err_to_gem(ret);
 		/*
@@ -652,8 +646,8 @@ static int i915_ttm_get_pages(struct drm_i915_gem_object *obj)
 		 * If the initial attempt fails, allow all accepted placements,
 		 * evicting if necessary.
 		 */
-		placement.num_busy_placement = real_num_busy;
-		ret = ttm_bo_validate(bo, &placement, &ctx);
+		placement->num_busy_placement = real_num_busy;
+		ret = ttm_bo_validate(bo, placement, &ctx);
 		if (ret)
 			return i915_ttm_err_to_gem(ret);
 	}
@@ -668,16 +662,56 @@ static int i915_ttm_get_pages(struct drm_i915_gem_object *obj)
 		i915_ttm_adjust_gem_after_move(obj);
 	}
 
-	/* Object either has a page vector or is an iomem object */
-	st = bo->ttm ? i915_ttm_tt_get_st(bo->ttm) : obj->ttm.cached_io_st;
-	if (IS_ERR(st))
-		return PTR_ERR(st);
+	if (!obj->mm.pages) {
+		/* Object either has a page vector or is an iomem object */
+		st = bo->ttm ? i915_ttm_tt_get_st(bo->ttm) : obj->ttm.cached_io_st;
+		if (IS_ERR(st))
+			return PTR_ERR(st);
 
-	__i915_gem_object_set_pages(obj, st, i915_sg_dma_sizes(st->sgl));
+		__i915_gem_object_set_pages(obj, st, i915_sg_dma_sizes(st->sgl));
+	}
 
 	return ret;
 }
 
+static int i915_ttm_get_pages(struct drm_i915_gem_object *obj)
+{
+	struct ttm_place requested, busy[I915_TTM_MAX_PLACEMENTS];
+	struct ttm_placement placement;
+
+	GEM_BUG_ON(obj->mm.n_placements > I915_TTM_MAX_PLACEMENTS);
+
+	/* Move to the requested placement. */
+	i915_ttm_placement_from_obj(obj, &requested, busy, &placement);
+
+	return __i915_ttm_get_pages(obj, &placement);
+}
+
+static int i915_ttm_migrate(struct drm_i915_gem_object *obj,
+			    struct intel_memory_region *mr)
+{
+	struct ttm_place requested;
+	struct ttm_placement placement;
+	int ret;
+
+	i915_ttm_place_from_region(mr, &requested, obj->flags);
+	placement.num_placement = 1;
+	placement.num_busy_placement = 1;
+	placement.placement = &requested;
+	placement.busy_placement = &requested;
+
+	ret = __i915_ttm_get_pages(obj, &placement);
+	if (ret)
+		return ret;
+
+	if (obj->mm.region != mr) {
+		i915_gem_object_release_memory_region(obj);
+		i915_gem_object_init_memory_region(obj, mr);
+	}
+
+	return 0;
+}
+
 static void i915_ttm_put_pages(struct drm_i915_gem_object *obj,
 			       struct sg_table *st)
 {
@@ -814,6 +848,7 @@ static const struct drm_i915_gem_object_ops i915_gem_ttm_obj_ops = {
 	.truncate = i915_ttm_purge,
 	.adjust_lru = i915_ttm_adjust_lru,
 	.delayed_free = i915_ttm_delayed_free,
+	.migrate = i915_ttm_migrate,
 	.mmap_offset = i915_ttm_mmap_offset,
 	.mmap_ops = &vm_ops_ttm,
 };
drivers/gpu/drm/i915/gem/i915_gem_wait.c +7 -0
index 1070d3afdce7..f909aaa09d9c 100644
@@ -290,3 +290,22 @@ i915_gem_wait_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
 	i915_gem_object_put(obj);
 	return ret;
 }
+
+/**
+ * i915_gem_object_wait_migration - Sync an accelerated migration operation
+ * @obj: The migrating object.
+ * @flags: waiting flags. Currently supports only I915_WAIT_INTERRUPTIBLE.
+ *
+ * Wait for any pending async migration operation on the object,

drivers/gpu/drm/i915/gem/i915_gem_wait.c i915_gem_wait_ioctl(struct drm_device *dev, void *data, struct drm_file *file) · line 309

Quoted 10 lines
drivers/gpu/drm/i915/gem/i915_gem_wait.c +10 -0
+ * whether it's explicitly (i915_gem_object_migrate()) or implicitly
+ * (swapin, initial clearing) initiated.
+ *
+ * Return: 0 if successful, -ERESTARTSYS if a signal was hit during waiting.
+ */
+int i915_gem_object_wait_migration(struct drm_i915_gem_object *obj,
+				   unsigned int flags)
+{
+	might_sleep();
+	/* NOP for now. */

Maybe dma_fence_might_wait() here, since that's a notch more nasty?

Cheers, Daniel

Quoted 9 lines
drivers/gpu/drm/i915/gem/i915_gem_wait.c +2 -0
+	return 0;
+}
Signature · 6 lines
2.31.1

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 7 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
#21 lore raw

Re: [PATCH v3 4/5] drm/i915/gem: Fix same-driver-another-instance dma-buf export

In reply to Daniel Vetter
Quoted 23 lines · Daniel Vetter
Quoted 22 lines · Michael J
Quoted 22 lines

-----Original Message-----
From: Thomas Hellström <[email protected]>
Sent: Monday, June 28, 2021 10:46 AM
To: [email protected]; [email protected]
Cc: Auld, Matthew <[email protected]>; [email protected]; Thomas Hellström <[email protected]>; Ruhl; Ruhl, Michael J <[email protected]>
Subject: [PATCH v3 4/5] drm/i915/gem: Fix same-driver-another-instance dma-buf export

If our exported dma-bufs are imported by another instance of our driver, that instance will typically have the imported dma-bufs locked during map_attachment(). But the exporter also locks the same reservation object in the map_dma_buf() callback, which leads to recursive locking.

Add a live selftest to catch this case, and as a workaround until we fully support dynamic import and export, declare the exporter dynamic by providing NOP pin() and unpin() functions. This means our map_dma_buf() callback will *always* get called locked, and by pinning unconditionally in i915_gem_map_dma_buf() we make sure we don't need to use the move_notify() functionality which is not yet implemented.

Quoted 4 lines
Quoted 1 line

An interesting abuse of the interface, but it seems reasonable (for now) to me.

Hm I'm not sure this is the best interface abuse, since if we combine this with amdgpu it goes boom. Also I thought the dynamic stuff is optional (or is that only for the importer).

I'm not sure what would go wrong here when combined with amdgpu? I figure an amdgpu importer being dynamic, would expect to get notified using move_notify() on move, but that never happens since the exported bo is pinned. If it matters for interface abuse then I could add real implementations of pin() and unpin(). But choosing to not evict a mapped dma-buf must remain at the exporter's discretion and is not an interface abuse IMO.

Could you point me to a case that would not work with this code?

Quoted 4 lines

What I discussed a bit with Maarten on irc is to essentially emulate the rules of what a dynamic exporter would end up with with a non-dynamic
importer: pin/unpin the buffer at attach/detach time. We could fake this in our attach/detach callbacks.

Yes, but that would only reimplement what's already in the dma-buf core? Since we're about to add a real and correct implementation of this, that sounds like a waste of time IMHO.

Quoted 3 lines

At least I don't think it's the locking changes that saves us here, but the caching of the sgt list in attach/detach.

Yes that saves us for the case of a non-locking non-dynamic importer, but for same-driver-another-instance, it's indeed the locking changes.

Quoted 4 lines

As long as we hand-roll that
we should be fine. So hand-rolling that feels like the best option to make sure we're not making this worse, as long as we haven't fully validated the true dynamic importer _and_ exporter case.

/Thomas

Quoted 213 lines

Cheers, Daniel

Quoted 210 lines

Mike

Quoted 206 lines

---
drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c | 31 ++++++-
.../drm/i915/gem/selftests/i915_gem_dmabuf.c | 81
++++++++++++++++++-

2 files changed, 108 insertions(+), 4 deletions(-)
drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c +2 -0
index 616c3a2f1baf..1d1eeb167d28 100644
@@ -12,6 +12,8 @@
 #include "i915_gem_object.h"
 #include "i915_scatterlist.h"
 
+I915_SELFTEST_DECLARE(static bool force_different_devices;)
+
 static struct drm_i915_gem_object *dma_buf_to_obj(struct dma_buf *buf)
 {
 	return to_intel_bo(buf->priv);
@@ -25,7 +27,9 @@ static struct sg_table *i915_gem_map_dma_buf(struct

dma_buf_attachment *attachme

	struct scatterlist *src, *dst;
	int ret, i;
drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c +3 -1
-	ret = i915_gem_object_pin_pages_unlocked(obj);
+	assert_object_held(obj);
+
+	ret = i915_gem_object_pin_pages(obj);
	if (ret)
		goto err;
drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c +3 -0
@@ -168,6 +172,26 @@ static int i915_gem_end_cpu_access(struct dma_buf
 *dma_buf, enum dma_data_direct
 	return err;
 }
 
+/*
+ * As a workaround until we fully support dynamic import and export,
+ * declare the exporter dynamic by providing NOP pin() and unpin()

functions.
+ * This means our i915_gem_map_dma_buf() callback will *always* get called
+ * locked, and by pinning unconditionally in i915_gem_map_dma_buf() we make
+ * sure we don't need to use the move_notify() functionality which is
+ * not yet implemented. Typically for the same-driver-another-instance case,
+ * i915_gem_map_dma_buf() will be called at importer attach time and the
+ * mapped sg_list will be cached by the dma-buf core for the
+ * duration of the attachment.
+ */

drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c +9 -0
+static int i915_gem_dmabuf_pin(struct dma_buf_attachment *attach)
+{
+	return 0;
+}
+
+static void i915_gem_dmabuf_unpin(struct dma_buf_attachment *attach)
+{
+}
+
static const struct dma_buf_ops i915_dmabuf_ops =  {
	.map_dma_buf = i915_gem_map_dma_buf,
	.unmap_dma_buf = i915_gem_unmap_dma_buf,
drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c +2 -0
@@ -177,6 +201,8 @@ static const struct dma_buf_ops i915_dmabuf_ops =  {
 	.vunmap = i915_gem_dmabuf_vunmap,
 	.begin_cpu_access = i915_gem_begin_cpu_access,
 	.end_cpu_access = i915_gem_end_cpu_access,
+	.pin = i915_gem_dmabuf_pin,
+	.unpin = i915_gem_dmabuf_unpin,
 };
 

struct dma_buf *i915_gem_prime_export(struct drm_gem_object *gem_obj, int flags)

drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c +2 -1
@@ -241,7 +267,8 @@ struct drm_gem_object
 *i915_gem_prime_import(struct drm_device *dev,
 	if (dma_buf->ops == &i915_dmabuf_ops) {
 		obj = dma_buf_to_obj(dma_buf);
 		/* is it from our device? */
-		if (obj->base.dev == dev) {
+		if (obj->base.dev == dev &&
+		    !I915_SELFTEST_ONLY(force_different_devices)) {
 			/*
 			 * Importing dmabuf exported from out own gem

increases
* refcount on gem itself instead of f_count of
dmabuf.

drivers/gpu/drm/i915/gem/selftests/i915_gem_dmabuf.c +10 -1
index dd74bc09ec88..24735d6c12a2 100644
@@ -35,7 +35,7 @@ static int igt_dmabuf_export(void *arg)
 static int igt_dmabuf_import_self(void *arg)
 {
 	struct drm_i915_private *i915 = arg;
-	struct drm_i915_gem_object *obj;
+	struct drm_i915_gem_object *obj, *import_obj;
 	struct drm_gem_object *import;
 	struct dma_buf *dmabuf;
 	int err;
@@ -65,14 +65,90 @@ static int igt_dmabuf_import_self(void *arg)
 		err = -EINVAL;
 		goto out_import;
 	}
+	import_obj = to_intel_bo(import);
+
+	i915_gem_object_lock(import_obj, NULL);
+	err = ____i915_gem_object_get_pages(import_obj);
+	i915_gem_object_unlock(import_obj);
+	if (err) {
+		pr_err("Same object dma-buf get_pages failed!\n");
+		goto out_import;
+	}
 
 	err = 0;

out_import:

drivers/gpu/drm/i915/gem/selftests/i915_gem_dmabuf.c +66 -1
-	i915_gem_object_put(to_intel_bo(import));
+	i915_gem_object_put(import_obj);
+out_dmabuf:
+	dma_buf_put(dmabuf);
+out:
+	i915_gem_object_put(obj);
+	return err;
+}
+
+static int igt_dmabuf_import_same_driver(void *arg)
+{
+	struct drm_i915_private *i915 = arg;
+	struct drm_i915_gem_object *obj, *import_obj;
+	struct drm_gem_object *import;
+	struct dma_buf *dmabuf;
+	int err;
+
+	force_different_devices = true;
+	obj = i915_gem_object_create_shmem(i915, PAGE_SIZE);
+	if (IS_ERR(obj))
+		goto out_ret;
+
+	dmabuf = i915_gem_prime_export(&obj->base, 0);
+	if (IS_ERR(dmabuf)) {
+		pr_err("i915_gem_prime_export failed with err=%d\n",
+		       (int)PTR_ERR(dmabuf));
+		err = PTR_ERR(dmabuf);
+		goto out;
+	}
+
+	import = i915_gem_prime_import(&i915->drm, dmabuf);
+	if (IS_ERR(import)) {
+		pr_err("i915_gem_prime_import failed with err=%d\n",
+		       (int)PTR_ERR(import));
+		err = PTR_ERR(import);
+		goto out_dmabuf;
+	}
+
+	if (import == &obj->base) {
+		pr_err("i915_gem_prime_import reused gem object!\n");
+		err = -EINVAL;
+		goto out_import;
+	}
+
+	import_obj = to_intel_bo(import);
+
+	i915_gem_object_lock(import_obj, NULL);
+	err = ____i915_gem_object_get_pages(import_obj);
+	if (err) {
+		pr_err("Different objects dma-buf get_pages failed!\n");
+		i915_gem_object_unlock(import_obj);
+		goto out_import;
+	}
+
+	/*
+	 * If the exported object is not in system memory, something
+	 * weird is going on. TODO: When p2p is supported, this is no
+	 * longer considered weird.
+	 */
+	if (obj->mm.region != i915->mm.regions[INTEL_REGION_SMEM]) {
+		pr_err("Exported dma-buf is not in system memory\n");
+		err = -EINVAL;
+	}
+	i915_gem_object_unlock(import_obj);
+
+out_import:
+	i915_gem_object_put(import_obj);

out_dmabuf:

	dma_buf_put(dmabuf);
out:
	i915_gem_object_put(obj);
drivers/gpu/drm/i915/gem/selftests/i915_gem_dmabuf.c +2 -0
+out_ret:
+	force_different_devices = false;
	return err;
}
drivers/gpu/drm/i915/gem/selftests/i915_gem_dmabuf.c +1 -0
@@ -286,6 +362,7 @@ int i915_gem_dmabuf_live_selftests(struct
 drm_i915_private *i915)
 {
 	static const struct i915_subtest tests[] = {
 		SUBTEST(igt_dmabuf_export),
+		SUBTEST(igt_dmabuf_import_same_driver),
 	};
 
	return i915_subtests(tests, i915);

--
2.31.1