Re: [PATCH drm-misc-next v6 1/6] drm/gpuvm: add common dma-resv per struct drm_gpuvm

From: Danilo Krummrich
Date: Tue Oct 17 2023 - 09:19:29 EST


On 10/13/23 15:00, Thomas Hellström wrote:
On Fri, 2023-10-13 at 13:51 +0200, Danilo Krummrich wrote:
On 10/13/23 13:38, Thomas Hellström wrote:
On Mon, 2023-10-09 at 01:32 +0200, Danilo Krummrich wrote:
Provide a common dma-resv for GEM objects not being used outside
of
this
GPU-VM. This is used in a subsequent patch to generalize dma-
resv,
external and evicted object handling and GEM validation.

Signed-off-by: Danilo Krummrich <dakr@xxxxxxxxxx>
---
  drivers/gpu/drm/drm_gpuvm.c            | 56
+++++++++++++++++++++++++-
  drivers/gpu/drm/nouveau/nouveau_uvmm.c | 13 +++++-
  include/drm/drm_gpuvm.h                | 35 +++++++++++++++-
  3 files changed, 99 insertions(+), 5 deletions(-)

diff --git a/drivers/gpu/drm/drm_gpuvm.c
b/drivers/gpu/drm/drm_gpuvm.c
index 02ecb45a2544..ebda9d594165 100644
--- a/drivers/gpu/drm/drm_gpuvm.c
+++ b/drivers/gpu/drm/drm_gpuvm.c
@@ -61,6 +61,15 @@
   * contained within struct drm_gpuva already. Hence, for
inserting
&drm_gpuva
   * entries from within dma-fence signalling critical sections
it is
enough to
   * pre-allocate the &drm_gpuva structures.
+ *
+ * &drm_gem_objects which are private to a single VM can share a
common
+ * &dma_resv in order to improve locking efficiency (e.g. with
&drm_exec).
+ * For this purpose drivers must pass a &drm_gem_object to
drm_gpuvm_init(), in
+ * the following called 'root object', which serves as the
container

Nit: Perhaps resv object altough it might typically be the root
page-
table object, that doesn't have any meaning to drm_gpuvm, which
uses it
solely as a container for the resv?

With "root" I didn't want to refer to the object representing the
root
page-table object, but being *the* object every other (internal)
object
needs to keep a reference to.

OK, yes but the reason they need a reference is because of the shared
resv, so IMO resv_object is a good fit. (I later noticed there's even
the function name drm_gpuvm_resv_obj()). And it will probably get
confused with the driver's "root" page table object, but up to you.

Yeah, maybe 'resv object' is better, gonna change it.


Maybe I should be more explicit here and say
that drivers need to make sure every internal object requires a
reference
to take a reference to this root object.


of the
+ * GPUVM's shared &dma_resv. This root object can be a driver
specific
+ * &drm_gem_object, such as the &drm_gem_object containing the
root
page table,
+ * but it can also be a 'dummy' object, which can be allocated
with
+ * drm_gpuvm_root_object_alloc().
   */
  /**
@@ -652,9 +661,47 @@ drm_gpuvm_range_valid(struct drm_gpuvm
*gpuvm,
                !drm_gpuvm_in_kernel_node(gpuvm, addr, range);
  }
+static void
+drm_gpuvm_gem_object_free(struct drm_gem_object *obj)
+{
+       drm_gem_object_release(obj);
+       kfree(obj);
+}
+
+static const struct drm_gem_object_funcs drm_gpuvm_object_funcs
= {
+       .free = drm_gpuvm_gem_object_free,
+};
+
+/**
+ * drm_gpuvm_root_object_alloc() - allocate a dummy
&drm_gem_object
+ * @drm: the drivers &drm_device
+ *
+ * Allocates a dummy &drm_gem_object which can be passed to
drm_gpuvm_init() in
+ * order to serve as root GEM object providing the &drm_resv
shared
across
+ * &drm_gem_objects local to a single GPUVM.
+ *
+ * Returns: the &drm_gem_object on success, NULL on failure
+ */
+struct drm_gem_object *
+drm_gpuvm_root_object_alloc(struct drm_device *drm)
+{
+       struct drm_gem_object *obj;
+
+       obj = kzalloc(sizeof(*obj), GFP_KERNEL);
+       if (!obj)
+               return NULL;
+
+       obj->funcs = &drm_gpuvm_object_funcs;
+       drm_gem_private_object_init(drm, obj, 0);
+
+       return obj;
+}
+EXPORT_SYMBOL_GPL(drm_gpuvm_root_object_alloc);
+
  /**
   * drm_gpuvm_init() - initialize a &drm_gpuvm
   * @gpuvm: pointer to the &drm_gpuvm to initialize
+ * @r_obj: the root &drm_gem_object providing the GPUVM's common
&dma_resv
   * @name: the name of the GPU VA space
   * @start_offset: the start offset of the GPU VA space
   * @range: the size of the GPU VA space
@@ -668,7 +715,7 @@ drm_gpuvm_range_valid(struct drm_gpuvm
*gpuvm,
   * &name is expected to be managed by the surrounding driver
structures.
   */
  void
-drm_gpuvm_init(struct drm_gpuvm *gpuvm,
+drm_gpuvm_init(struct drm_gpuvm *gpuvm, struct drm_gem_object
*r_obj,
                const char *name,
                u64 start_offset, u64 range,
                u64 reserve_offset, u64 reserve_range,
@@ -683,6 +730,9 @@ drm_gpuvm_init(struct drm_gpuvm *gpuvm,
         gpuvm->name = name ? name : "unknown";
         gpuvm->ops = ops;
+       gpuvm->r_obj = r_obj;
+
+       drm_gem_object_get(r_obj);
         memset(&gpuvm->kernel_alloc_node, 0, sizeof(struct
drm_gpuva));
@@ -713,7 +763,9 @@ drm_gpuvm_destroy(struct drm_gpuvm *gpuvm)
                 __drm_gpuva_remove(&gpuvm->kernel_alloc_node);
         WARN(!RB_EMPTY_ROOT(&gpuvm->rb.tree.rb_root),
-            "GPUVA tree is not empty, potentially leaking
memory.");
+            "GPUVA tree is not empty, potentially leaking
memory.\n");

Should we cache the drm device in struct drm_gpuvm and use
drm_warn()
here instead of WARN?

I'd guess the additional backtrace of WARN() isn't overly useful in
this
case. However, it might be a bit more obvious in dmesg due to its
verboseness. Not a strong opinion on that, though.

My bad. I meant drm_WARN(). In a multi-gpu environment it's nice to
have the extra device info.

Sure, gonna add a separate patch to change that in all places.


/Thomas




+
+       drm_gem_object_put(gpuvm->r_obj);
  }
  EXPORT_SYMBOL_GPL(drm_gpuvm_destroy);
diff --git a/drivers/gpu/drm/nouveau/nouveau_uvmm.c
b/drivers/gpu/drm/nouveau/nouveau_uvmm.c
index 5cf892c50f43..4dea847ef989 100644
--- a/drivers/gpu/drm/nouveau/nouveau_uvmm.c
+++ b/drivers/gpu/drm/nouveau/nouveau_uvmm.c
@@ -1808,8 +1808,9 @@ int
  nouveau_uvmm_init(struct nouveau_uvmm *uvmm, struct nouveau_cli
*cli,
                   u64 kernel_managed_addr, u64
kernel_managed_size)
  {
-       int ret;
+       struct drm_gem_object *r_obj;
         u64 kernel_managed_end = kernel_managed_addr +
kernel_managed_size;
+       int ret;
         mutex_init(&uvmm->mutex);
         dma_resv_init(&uvmm->resv);
@@ -1833,14 +1834,22 @@ nouveau_uvmm_init(struct nouveau_uvmm
*uvmm,
struct nouveau_cli *cli,
                 goto out_unlock;
         }
+       r_obj = drm_gpuvm_root_object_alloc(cli->drm->dev);
+       if (!r_obj) {
+               ret = -ENOMEM;
+               goto out_unlock;
+       }
+
         uvmm->kernel_managed_addr = kernel_managed_addr;
         uvmm->kernel_managed_size = kernel_managed_size;
-       drm_gpuvm_init(&uvmm->base, cli->name,
+       drm_gpuvm_init(&uvmm->base, r_obj, cli->name,
                        NOUVEAU_VA_SPACE_START,
                        NOUVEAU_VA_SPACE_END,
                        kernel_managed_addr, kernel_managed_size,
                        NULL);
+       /* GPUVM takes care from here on. */
+       drm_gem_object_put(r_obj);
         ret = nvif_vmm_ctor(&cli->mmu, "uvmm",
                             cli->vmm.vmm.object.oclass, RAW,
diff --git a/include/drm/drm_gpuvm.h b/include/drm/drm_gpuvm.h
index c7ed6bf441d4..0aec14d8b259 100644
--- a/include/drm/drm_gpuvm.h
+++ b/include/drm/drm_gpuvm.h
@@ -238,9 +238,15 @@ struct drm_gpuvm {
          * @ops: &drm_gpuvm_ops providing the split/merge steps
to
drivers
          */
         const struct drm_gpuvm_ops *ops;
+
+       /**
+        * @r_obj: Root GEM object; representing the GPUVM's
common
&dma_resv.
+        */
+       struct drm_gem_object *r_obj;
  };
-void drm_gpuvm_init(struct drm_gpuvm *gpuvm, const char *name,
+void drm_gpuvm_init(struct drm_gpuvm *gpuvm, struct
drm_gem_object
*r_obj,
+                   const char *name,
                     u64 start_offset, u64 range,
                     u64 reserve_offset, u64 reserve_range,
                     const struct drm_gpuvm_ops *ops);
@@ -248,6 +254,33 @@ void drm_gpuvm_destroy(struct drm_gpuvm
*gpuvm);
  bool drm_gpuvm_interval_empty(struct drm_gpuvm *gpuvm, u64
addr, u64
range);
+struct drm_gem_object *
+drm_gpuvm_root_object_alloc(struct drm_device *drm);
+
+/**
+ * drm_gpuvm_resv() - returns the &drm_gpuvm's &dma_resv
+ * @gpuvm__: the &drm_gpuvm
+ *
+ * Returns: a pointer to the &drm_gpuvm's shared &dma_resv
+ */
+#define drm_gpuvm_resv(gpuvm__) ((gpuvm__)->r_obj->resv)
+
+/**
+ * drm_gpuvm_resv_obj() - returns the &drm_gem_object holding
the
&drm_gpuvm's
+ * &dma_resv
+ * @gpuvm__: the &drm_gpuvm
+ *
+ * Returns: a pointer to the &drm_gem_object holding the
&drm_gpuvm's shared
+ * &dma_resv
+ */
+#define drm_gpuvm_resv_obj(gpuvm__) ((gpuvm__)->r_obj)
+
+#define drm_gpuvm_resv_held(gpuvm__) \
+       dma_resv_held(drm_gpuvm_resv(gpuvm__))
+
+#define drm_gpuvm_resv_assert_held(gpuvm__) \
+       dma_resv_assert_held(drm_gpuvm_resv(gpuvm__))
+
  static inline struct drm_gpuva *
  __drm_gpuva_next(struct drm_gpuva *va)
  {

Reviewed-by: Thomas Hellström <thomas.hellstrom@xxxxxxxxxxxxxxx>