Skip to contentSkip to conversation

drm/amdgpu: return bo itself if userptr is cpu addr of bo (v3)

PATCH In review Junwei Zhang 5 participants 24 messages Jul 30, 2018 – Aug 2, 2018 20 messages not in the archive

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/2 drm/amdgpu: return bo itself if userptr is cpu addr of bo (v3)

From: Chunming Zhou <[email protected]>

v2: get original gem handle from gobj
v3: update find bo data structure as union(in, out)
simply some code logic

---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h     |  2 ++
 drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 63 +++++++++++++++++++++++++++++++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c |  3 +-
 include/uapi/drm/amdgpu_drm.h           | 21 +++++++++++
 4 files changed, 88 insertions(+), 1 deletion(-)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +2 -0
index 4cd20e7..46c370b 100644
@@ -1213,6 +1213,8 @@ int amdgpu_gem_info_ioctl(struct drm_device *dev, void *data,
 			  struct drm_file *filp);
 int amdgpu_gem_userptr_ioctl(struct drm_device *dev, void *data,
 			struct drm_file *filp);
+int amdgpu_gem_find_bo_by_cpu_mapping_ioctl(struct drm_device *dev, void *data,
+					    struct drm_file *filp);
 int amdgpu_gem_mmap_ioctl(struct drm_device *dev, void *data,
 			  struct drm_file *filp);
 int amdgpu_gem_wait_idle_ioctl(struct drm_device *dev, void *data,
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +63 -0

69 lines not on this page · raw message

drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c +2 -1
index 7956848..1bd2cc1 100644
@@ -1100,7 +1100,8 @@ void amdgpu_disable_vblank_kms(struct drm_device *dev, unsigned int pipe)
 	DRM_IOCTL_DEF_DRV(AMDGPU_GEM_VA, amdgpu_gem_va_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
 	DRM_IOCTL_DEF_DRV(AMDGPU_GEM_OP, amdgpu_gem_op_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
 	DRM_IOCTL_DEF_DRV(AMDGPU_GEM_USERPTR, amdgpu_gem_userptr_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
-	DRM_IOCTL_DEF_DRV(AMDGPU_FREESYNC, amdgpu_display_freesync_ioctl, DRM_MASTER)
+	DRM_IOCTL_DEF_DRV(AMDGPU_FREESYNC, amdgpu_display_freesync_ioctl, DRM_MASTER),
+	DRM_IOCTL_DEF_DRV(AMDGPU_GEM_FIND_BO, amdgpu_gem_find_bo_by_cpu_mapping_ioctl, DRM_AUTH|DRM_RENDER_ALLOW)
 };
 const int amdgpu_max_kms_ioctl = ARRAY_SIZE(amdgpu_ioctls_kms);
 
include/uapi/drm/amdgpu_drm.h +21 -0
index 94444ee..000c415 100644
@@ -54,6 +54,7 @@
 #define DRM_AMDGPU_VM			0x13
 #define DRM_AMDGPU_FENCE_TO_HANDLE	0x14
 #define DRM_AMDGPU_SCHED		0x15
+#define DRM_AMDGPU_GEM_FIND_BO		0x16
 /* not upstream */
 #define DRM_AMDGPU_FREESYNC	        0x5d
 
@@ -74,6 +75,7 @@
 #define DRM_IOCTL_AMDGPU_FENCE_TO_HANDLE DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_FENCE_TO_HANDLE, union drm_amdgpu_fence_to_handle)
 #define DRM_IOCTL_AMDGPU_SCHED		DRM_IOW(DRM_COMMAND_BASE + DRM_AMDGPU_SCHED, union drm_amdgpu_sched)
 #define DRM_IOCTL_AMDGPU_FREESYNC	DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_FREESYNC, struct drm_amdgpu_freesync)
+#define DRM_IOCTL_AMDGPU_GEM_FIND_BO	DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_GEM_FIND_BO, union drm_amdgpu_gem_find_bo)
 
 /**
  * DOC: memory domains
@@ -392,6 +394,25 @@ struct drm_amdgpu_gem_wait_idle_out {
 	struct drm_amdgpu_gem_wait_idle_out out;
 };
 
+struct drm_amdgpu_gem_find_bo_in {
+	/* CPU address */
+	__u64			addr;
+	/* memory size from CPU address */
+	__u64			size;
+};
+
+struct drm_amdgpu_gem_find_bo_out {
+	/* offset in bo */
+	__u64			offset;
+	/* resulting GEM handle */
+	__u32			handle;
+};
+
+union drm_amdgpu_gem_find_bo {
+	struct drm_amdgpu_gem_find_bo_in in;
+	struct drm_amdgpu_gem_find_bo_out out;
+};
+
 struct drm_amdgpu_wait_cs_in {
 	/* Command submission handle
          * handle equals 0 means none to wait for
Signature · 6 lines
1.9.1

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

PATCH 2/2 drm/amdgpu: bump version for API to find bo by cpu mapping

In reply to a message not in the archive

bump version for new API involved.

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +2 -1
index 8843a06..6b52551 100644
@@ -70,9 +70,10 @@
  * - 3.25.0 - Add support for sensor query info (stable pstate sclk/mclk).
  * - 3.26.0 - GFX9: Process AMDGPU_IB_FLAG_TC_WB_NOT_INVALIDATE.
  * - 3.27.0 - Add new chunk to to AMDGPU_CS to enable BO_LIST creation.
+ * - 3.28.0 - Add API to find bo by cpu mapping for UMD bo validation.
  */
 #define KMS_DRIVER_MAJOR	3
-#define KMS_DRIVER_MINOR	27
+#define KMS_DRIVER_MINOR	28
 #define KMS_DRIVER_PATCHLEVEL	0
 
 int amdgpu_vram_limit = 0;
Signature · 6 lines
1.9.1

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
#3 lore raw
In reply to a message not in the archive

Am 30.07.2018 um 12:02 schrieb Junwei Zhang:

Quoted 5 lines

From: Chunming Zhou <[email protected]>

v2: get original gem handle from gobj
v3: update find bo data structure as union(in, out)
simply some code logic

Do we now have an open source user for this, so that we can upstream it? One more point below.

Quoted 37 lines
---
  drivers/gpu/drm/amd/amdgpu/amdgpu.h     |  2 ++
  drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 63 +++++++++++++++++++++++++++++++++
  drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c |  3 +-
  include/uapi/drm/amdgpu_drm.h           | 21 +++++++++++
  4 files changed, 88 insertions(+), 1 deletion(-)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +2 -0
index 4cd20e7..46c370b 100644
@@ -1213,6 +1213,8 @@ int amdgpu_gem_info_ioctl(struct drm_device *dev, void *data,
  			  struct drm_file *filp);
  int amdgpu_gem_userptr_ioctl(struct drm_device *dev, void *data,
  			struct drm_file *filp);
+int amdgpu_gem_find_bo_by_cpu_mapping_ioctl(struct drm_device *dev, void *data,
+					    struct drm_file *filp);
  int amdgpu_gem_mmap_ioctl(struct drm_device *dev, void *data,
  			  struct drm_file *filp);
  int amdgpu_gem_wait_idle_ioctl(struct drm_device *dev, void *data,
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +4 -0
index 71792d8..bae8417 100644
@@ -288,6 +288,69 @@ int amdgpu_gem_create_ioctl(struct drm_device *dev, void *data,
  	return 0;
  }
 
+static int amdgpu_gem_get_handle_from_object(struct drm_file *filp,
+					     struct drm_gem_object *obj)
+{
+	int i;

drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c int amdgpu_gem_create_ioctl(struct drm_device *dev, void *data, · line 304

Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +10 -0
+	struct drm_gem_object *tmp;
+
+	spin_lock(&filp->table_lock);
+	idr_for_each_entry(&filp->object_idr, tmp, i) {
+		if (obj == tmp) {
+			drm_gem_object_reference(obj);
+			spin_unlock(&filp->table_lock);
+			return i;
+		}
+	}

Please double check if that is still up to date.

I think we could as well try to use the DMA-buf handle tree for that.

Christian.

Quoted 102 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +49 -0
+	spin_unlock(&filp->table_lock);
+
+	return 0;
+}
+
+int amdgpu_gem_find_bo_by_cpu_mapping_ioctl(struct drm_device *dev, void *data,
+					    struct drm_file *filp)
+{
+	union drm_amdgpu_gem_find_bo *args = data;
+	struct drm_gem_object *gobj;
+	struct amdgpu_bo *bo;
+	struct ttm_buffer_object *tbo;
+	struct vm_area_struct *vma;
+	uint32_t handle;
+	int r;
+
+	if (offset_in_page(args->in.addr | args->in.size))
+		return -EINVAL;
+
+	down_read(&current->mm->mmap_sem);
+	vma = find_vma(current->mm, args->in.addr);
+	if (!vma || vma->vm_file != filp->filp ||
+	    (args->in.size > (vma->vm_end - args->in.addr))) {
+		args->out.handle = 0;
+		up_read(&current->mm->mmap_sem);
+		return -EINVAL;
+	}
+	args->out.offset = args->in.addr - vma->vm_start;
+
+	tbo = vma->vm_private_data;
+	bo = container_of(tbo, struct amdgpu_bo, tbo);
+	amdgpu_bo_ref(bo);
+	gobj = &bo->gem_base;
+
+	handle = amdgpu_gem_get_handle_from_object(filp, gobj);
+	if (!handle) {
+		r = drm_gem_handle_create(filp, gobj, &handle);
+		if (r) {
+			DRM_ERROR("create gem handle failed\n");
+			up_read(&current->mm->mmap_sem);
+			return r;
+		}
+	}
+	args->out.handle = handle;
+	up_read(&current->mm->mmap_sem);
+
+	return 0;
+}
+
  int amdgpu_gem_userptr_ioctl(struct drm_device *dev, void *data,
  			     struct drm_file *filp)
  {
drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c +2 -1
index 7956848..1bd2cc1 100644
@@ -1100,7 +1100,8 @@ void amdgpu_disable_vblank_kms(struct drm_device *dev, unsigned int pipe)
  	DRM_IOCTL_DEF_DRV(AMDGPU_GEM_VA, amdgpu_gem_va_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
  	DRM_IOCTL_DEF_DRV(AMDGPU_GEM_OP, amdgpu_gem_op_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
  	DRM_IOCTL_DEF_DRV(AMDGPU_GEM_USERPTR, amdgpu_gem_userptr_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
-	DRM_IOCTL_DEF_DRV(AMDGPU_FREESYNC, amdgpu_display_freesync_ioctl, DRM_MASTER)
+	DRM_IOCTL_DEF_DRV(AMDGPU_FREESYNC, amdgpu_display_freesync_ioctl, DRM_MASTER),
+	DRM_IOCTL_DEF_DRV(AMDGPU_GEM_FIND_BO, amdgpu_gem_find_bo_by_cpu_mapping_ioctl, DRM_AUTH|DRM_RENDER_ALLOW)
  };
  const int amdgpu_max_kms_ioctl = ARRAY_SIZE(amdgpu_ioctls_kms);
 
include/uapi/drm/amdgpu_drm.h +14 -0
index 94444ee..000c415 100644
@@ -54,6 +54,7 @@
  #define DRM_AMDGPU_VM			0x13
  #define DRM_AMDGPU_FENCE_TO_HANDLE	0x14
  #define DRM_AMDGPU_SCHED		0x15
+#define DRM_AMDGPU_GEM_FIND_BO		0x16
  /* not upstream */
  #define DRM_AMDGPU_FREESYNC	        0x5d
 
@@ -74,6 +75,7 @@
  #define DRM_IOCTL_AMDGPU_FENCE_TO_HANDLE DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_FENCE_TO_HANDLE, union drm_amdgpu_fence_to_handle)
  #define DRM_IOCTL_AMDGPU_SCHED		DRM_IOW(DRM_COMMAND_BASE + DRM_AMDGPU_SCHED, union drm_amdgpu_sched)
  #define DRM_IOCTL_AMDGPU_FREESYNC	DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_FREESYNC, struct drm_amdgpu_freesync)
+#define DRM_IOCTL_AMDGPU_GEM_FIND_BO	DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_GEM_FIND_BO, union drm_amdgpu_gem_find_bo)
 
  /**
   * DOC: memory domains
@@ -392,6 +394,25 @@ struct drm_amdgpu_gem_wait_idle_out {
  	struct drm_amdgpu_gem_wait_idle_out out;
  };
 
+struct drm_amdgpu_gem_find_bo_in {
+	/* CPU address */
+	__u64			addr;
+	/* memory size from CPU address */
+	__u64			size;
+};
+
+struct drm_amdgpu_gem_find_bo_out {
+	/* offset in bo */
+	__u64			offset;
+	/* resulting GEM handle */
+	__u32			handle;

include/uapi/drm/amdgpu_drm.h struct drm_amdgpu_gem_wait_idle_out { · line 418

Quoted 10 lines
include/uapi/drm/amdgpu_drm.h +7 -0
+};
+
+union drm_amdgpu_gem_find_bo {
+	struct drm_amdgpu_gem_find_bo_in in;
+	struct drm_amdgpu_gem_find_bo_out out;
+};
+
  struct drm_amdgpu_wait_cs_in {
  	/* Command submission handle
           * handle equals 0 means none to wait for

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

#5 lore raw
In reply to a message not in the archive

Thanks for Jerry still remembers this series.

Hi Christian,

For upstream of this feature, seems we already had agreement long time ago. Two reasons for upstreaming:
1. this bug was found by an opengl game, so this bug also is in mesa driver in theory.
2. after upstream these patches, we can reduce pro specific patches, and close to open source.

Btw, an unit test is excepted when upstreaming, I remember Alex mentioned.

Thanks,
David Zhou
-----Original Message-----
From: Christian König <[email protected]>
Sent: Monday, July 30, 2018 6:48 PM
To: Zhang, Jerry <[email protected]>; [email protected]
Cc: Zhou, David(ChunMing) <[email protected]>; Koenig, Christian <[email protected]>
Subject: Re: [PATCH 1/2] drm/amdgpu: return bo itself if userptr is cpu addr of bo (v3)

Am 30.07.2018 um 12:02 schrieb Junwei Zhang:

Quoted 5 lines

From: Chunming Zhou <[email protected]>

v2: get original gem handle from gobj
v3: update find bo data structure as union(in, out)
simply some code logic

Do we now have an open source user for this, so that we can upstream it? One more point below.

Quoted 38 lines
---
  drivers/gpu/drm/amd/amdgpu/amdgpu.h     |  2 ++
  drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 63 +++++++++++++++++++++++++++++++++
  drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c |  3 +-
  include/uapi/drm/amdgpu_drm.h           | 21 +++++++++++
  4 files changed, 88 insertions(+), 1 deletion(-)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +2 -0
index 4cd20e7..46c370b 100644
@@ -1213,6 +1213,8 @@ int amdgpu_gem_info_ioctl(struct drm_device *dev, void *data,
  			  struct drm_file *filp);
  int amdgpu_gem_userptr_ioctl(struct drm_device *dev, void *data,
  			struct drm_file *filp);
+int amdgpu_gem_find_bo_by_cpu_mapping_ioctl(struct drm_device *dev, void *data,
+					    struct drm_file *filp);
  int amdgpu_gem_mmap_ioctl(struct drm_device *dev, void *data,
  			  struct drm_file *filp);
  int amdgpu_gem_wait_idle_ioctl(struct drm_device *dev, void *data, 
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +3 -0
index 71792d8..bae8417 100644
@@ -288,6 +288,69 @@ int amdgpu_gem_create_ioctl(struct drm_device *dev, void *data,
  	return 0;
  }
 
+static int amdgpu_gem_get_handle_from_object(struct drm_file *filp,
+					     struct drm_gem_object *obj) {
+	int i;

drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c int amdgpu_gem_create_ioctl(struct drm_device *dev, void *data, · line 303

Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +10 -0
+	struct drm_gem_object *tmp;
+
+	spin_lock(&filp->table_lock);
+	idr_for_each_entry(&filp->object_idr, tmp, i) {
+		if (obj == tmp) {
+			drm_gem_object_reference(obj);
+			spin_unlock(&filp->table_lock);
+			return i;
+		}
+	}

Please double check if that is still up to date.

I think we could as well try to use the DMA-buf handle tree for that.

Christian.

Quoted 103 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +49 -0
+	spin_unlock(&filp->table_lock);
+
+	return 0;
+}
+
+int amdgpu_gem_find_bo_by_cpu_mapping_ioctl(struct drm_device *dev, void *data,
+					    struct drm_file *filp)
+{
+	union drm_amdgpu_gem_find_bo *args = data;
+	struct drm_gem_object *gobj;
+	struct amdgpu_bo *bo;
+	struct ttm_buffer_object *tbo;
+	struct vm_area_struct *vma;
+	uint32_t handle;
+	int r;
+
+	if (offset_in_page(args->in.addr | args->in.size))
+		return -EINVAL;
+
+	down_read(&current->mm->mmap_sem);
+	vma = find_vma(current->mm, args->in.addr);
+	if (!vma || vma->vm_file != filp->filp ||
+	    (args->in.size > (vma->vm_end - args->in.addr))) {
+		args->out.handle = 0;
+		up_read(&current->mm->mmap_sem);
+		return -EINVAL;
+	}
+	args->out.offset = args->in.addr - vma->vm_start;
+
+	tbo = vma->vm_private_data;
+	bo = container_of(tbo, struct amdgpu_bo, tbo);
+	amdgpu_bo_ref(bo);
+	gobj = &bo->gem_base;
+
+	handle = amdgpu_gem_get_handle_from_object(filp, gobj);
+	if (!handle) {
+		r = drm_gem_handle_create(filp, gobj, &handle);
+		if (r) {
+			DRM_ERROR("create gem handle failed\n");
+			up_read(&current->mm->mmap_sem);
+			return r;
+		}
+	}
+	args->out.handle = handle;
+	up_read(&current->mm->mmap_sem);
+
+	return 0;
+}
+
  int amdgpu_gem_userptr_ioctl(struct drm_device *dev, void *data,
  			     struct drm_file *filp)
  {
drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c +3 -1
index 7956848..1bd2cc1 100644
@@ -1100,7 +1100,8 @@ void amdgpu_disable_vblank_kms(struct drm_device *dev, unsigned int pipe)
  	DRM_IOCTL_DEF_DRV(AMDGPU_GEM_VA, amdgpu_gem_va_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
  	DRM_IOCTL_DEF_DRV(AMDGPU_GEM_OP, amdgpu_gem_op_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
  	DRM_IOCTL_DEF_DRV(AMDGPU_GEM_USERPTR, amdgpu_gem_userptr_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
-	DRM_IOCTL_DEF_DRV(AMDGPU_FREESYNC, amdgpu_display_freesync_ioctl, DRM_MASTER)
+	DRM_IOCTL_DEF_DRV(AMDGPU_FREESYNC, amdgpu_display_freesync_ioctl, DRM_MASTER),
+	DRM_IOCTL_DEF_DRV(AMDGPU_GEM_FIND_BO, 
+amdgpu_gem_find_bo_by_cpu_mapping_ioctl, DRM_AUTH|DRM_RENDER_ALLOW)
  };
  const int amdgpu_max_kms_ioctl = ARRAY_SIZE(amdgpu_ioctls_kms);
(quoted diff) +0 -0

b/include/uapi/drm/amdgpu_drm.h index 94444ee..000c415 100644

include/uapi/drm/amdgpu_drm.h +13 -0
@@ -54,6 +54,7 @@
  #define DRM_AMDGPU_VM			0x13
  #define DRM_AMDGPU_FENCE_TO_HANDLE	0x14
  #define DRM_AMDGPU_SCHED		0x15
+#define DRM_AMDGPU_GEM_FIND_BO		0x16
  /* not upstream */
  #define DRM_AMDGPU_FREESYNC	        0x5d
 
@@ -74,6 +75,7 @@
  #define DRM_IOCTL_AMDGPU_FENCE_TO_HANDLE DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_FENCE_TO_HANDLE, union drm_amdgpu_fence_to_handle)
  #define DRM_IOCTL_AMDGPU_SCHED		DRM_IOW(DRM_COMMAND_BASE + DRM_AMDGPU_SCHED, union drm_amdgpu_sched)
  #define DRM_IOCTL_AMDGPU_FREESYNC	DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_FREESYNC, struct drm_amdgpu_freesync)
+#define DRM_IOCTL_AMDGPU_GEM_FIND_BO	DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_GEM_FIND_BO, union drm_amdgpu_gem_find_bo)
 
  /**
   * DOC: memory domains
@@ -392,6 +394,25 @@ struct drm_amdgpu_gem_wait_idle_out {
  	struct drm_amdgpu_gem_wait_idle_out out;
  };
 
+struct drm_amdgpu_gem_find_bo_in {
+	/* CPU address */
+	__u64			addr;
+	/* memory size from CPU address */
+	__u64			size;
+};
+
+struct drm_amdgpu_gem_find_bo_out {
+	/* offset in bo */
+	__u64			offset;
+	/* resulting GEM handle */

include/uapi/drm/amdgpu_drm.h struct drm_amdgpu_gem_wait_idle_out { · line 417

Quoted 10 lines
include/uapi/drm/amdgpu_drm.h +7 -0
+	__u32			handle;
+};
+
+union drm_amdgpu_gem_find_bo {
+	struct drm_amdgpu_gem_find_bo_in in;
+	struct drm_amdgpu_gem_find_bo_out out; };
+
  struct drm_amdgpu_wait_cs_in {
  	/* Command submission handle
           * handle equals 0 means none to wait for

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

#7 lore raw
In reply to a message not in the archive

Typo, excepted -> expected

-----Original Message-----
From: amd-gfx <[email protected]> On Behalf Of Zhou, David(ChunMing)
Sent: Tuesday, July 31, 2018 9:41 AM
To: Koenig, Christian <[email protected]>; Zhang, Jerry <[email protected]>; [email protected]
Subject: RE: [PATCH 1/2] drm/amdgpu: return bo itself if userptr is cpu addr of bo (v3)

Thanks for Jerry still remembers this series.

Hi Christian,

For upstream of this feature, seems we already had agreement long time ago. Two reasons for upstreaming:
1. this bug was found by an opengl game, so this bug also is in mesa driver in theory.
2. after upstream these patches, we can reduce pro specific patches, and close to open source.

Btw, an unit test is excepted when upstreaming, I remember Alex mentioned.

Thanks,
David Zhou
-----Original Message-----
From: Christian König <[email protected]>
Sent: Monday, July 30, 2018 6:48 PM
To: Zhang, Jerry <[email protected]>; [email protected]
Cc: Zhou, David(ChunMing) <[email protected]>; Koenig, Christian <[email protected]>
Subject: Re: [PATCH 1/2] drm/amdgpu: return bo itself if userptr is cpu addr of bo (v3)

Am 30.07.2018 um 12:02 schrieb Junwei Zhang:

Quoted 5 lines

From: Chunming Zhou <[email protected]>

v2: get original gem handle from gobj
v3: update find bo data structure as union(in, out)
simply some code logic

Do we now have an open source user for this, so that we can upstream it? One more point below.

Quoted 38 lines
---
  drivers/gpu/drm/amd/amdgpu/amdgpu.h     |  2 ++
  drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 63 +++++++++++++++++++++++++++++++++
  drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c |  3 +-
  include/uapi/drm/amdgpu_drm.h           | 21 +++++++++++
  4 files changed, 88 insertions(+), 1 deletion(-)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +2 -0
index 4cd20e7..46c370b 100644
@@ -1213,6 +1213,8 @@ int amdgpu_gem_info_ioctl(struct drm_device *dev, void *data,
  			  struct drm_file *filp);
  int amdgpu_gem_userptr_ioctl(struct drm_device *dev, void *data,
  			struct drm_file *filp);
+int amdgpu_gem_find_bo_by_cpu_mapping_ioctl(struct drm_device *dev, void *data,
+					    struct drm_file *filp);
  int amdgpu_gem_mmap_ioctl(struct drm_device *dev, void *data,
  			  struct drm_file *filp);
  int amdgpu_gem_wait_idle_ioctl(struct drm_device *dev, void *data, 
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +3 -0
index 71792d8..bae8417 100644
@@ -288,6 +288,69 @@ int amdgpu_gem_create_ioctl(struct drm_device *dev, void *data,
  	return 0;
  }
 
+static int amdgpu_gem_get_handle_from_object(struct drm_file *filp,
+					     struct drm_gem_object *obj) {
+	int i;

drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c int amdgpu_gem_create_ioctl(struct drm_device *dev, void *data, · line 303

Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +10 -0
+	struct drm_gem_object *tmp;
+
+	spin_lock(&filp->table_lock);
+	idr_for_each_entry(&filp->object_idr, tmp, i) {
+		if (obj == tmp) {
+			drm_gem_object_reference(obj);
+			spin_unlock(&filp->table_lock);
+			return i;
+		}
+	}

Please double check if that is still up to date.

I think we could as well try to use the DMA-buf handle tree for that.

Christian.

Quoted 103 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +49 -0
+	spin_unlock(&filp->table_lock);
+
+	return 0;
+}
+
+int amdgpu_gem_find_bo_by_cpu_mapping_ioctl(struct drm_device *dev, void *data,
+					    struct drm_file *filp)
+{
+	union drm_amdgpu_gem_find_bo *args = data;
+	struct drm_gem_object *gobj;
+	struct amdgpu_bo *bo;
+	struct ttm_buffer_object *tbo;
+	struct vm_area_struct *vma;
+	uint32_t handle;
+	int r;
+
+	if (offset_in_page(args->in.addr | args->in.size))
+		return -EINVAL;
+
+	down_read(&current->mm->mmap_sem);
+	vma = find_vma(current->mm, args->in.addr);
+	if (!vma || vma->vm_file != filp->filp ||
+	    (args->in.size > (vma->vm_end - args->in.addr))) {
+		args->out.handle = 0;
+		up_read(&current->mm->mmap_sem);
+		return -EINVAL;
+	}
+	args->out.offset = args->in.addr - vma->vm_start;
+
+	tbo = vma->vm_private_data;
+	bo = container_of(tbo, struct amdgpu_bo, tbo);
+	amdgpu_bo_ref(bo);
+	gobj = &bo->gem_base;
+
+	handle = amdgpu_gem_get_handle_from_object(filp, gobj);
+	if (!handle) {
+		r = drm_gem_handle_create(filp, gobj, &handle);
+		if (r) {
+			DRM_ERROR("create gem handle failed\n");
+			up_read(&current->mm->mmap_sem);
+			return r;
+		}
+	}
+	args->out.handle = handle;
+	up_read(&current->mm->mmap_sem);
+
+	return 0;
+}
+
  int amdgpu_gem_userptr_ioctl(struct drm_device *dev, void *data,
  			     struct drm_file *filp)
  {
drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c +3 -1
index 7956848..1bd2cc1 100644
@@ -1100,7 +1100,8 @@ void amdgpu_disable_vblank_kms(struct drm_device *dev, unsigned int pipe)
  	DRM_IOCTL_DEF_DRV(AMDGPU_GEM_VA, amdgpu_gem_va_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
  	DRM_IOCTL_DEF_DRV(AMDGPU_GEM_OP, amdgpu_gem_op_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
  	DRM_IOCTL_DEF_DRV(AMDGPU_GEM_USERPTR, amdgpu_gem_userptr_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
-	DRM_IOCTL_DEF_DRV(AMDGPU_FREESYNC, amdgpu_display_freesync_ioctl, DRM_MASTER)
+	DRM_IOCTL_DEF_DRV(AMDGPU_FREESYNC, amdgpu_display_freesync_ioctl, DRM_MASTER),
+	DRM_IOCTL_DEF_DRV(AMDGPU_GEM_FIND_BO,
+amdgpu_gem_find_bo_by_cpu_mapping_ioctl, DRM_AUTH|DRM_RENDER_ALLOW)
  };
  const int amdgpu_max_kms_ioctl = ARRAY_SIZE(amdgpu_ioctls_kms);
(quoted diff) +0 -0

b/include/uapi/drm/amdgpu_drm.h index 94444ee..000c415 100644

include/uapi/drm/amdgpu_drm.h +13 -0
@@ -54,6 +54,7 @@
  #define DRM_AMDGPU_VM			0x13
  #define DRM_AMDGPU_FENCE_TO_HANDLE	0x14
  #define DRM_AMDGPU_SCHED		0x15
+#define DRM_AMDGPU_GEM_FIND_BO		0x16
  /* not upstream */
  #define DRM_AMDGPU_FREESYNC	        0x5d
 
@@ -74,6 +75,7 @@
  #define DRM_IOCTL_AMDGPU_FENCE_TO_HANDLE DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_FENCE_TO_HANDLE, union drm_amdgpu_fence_to_handle)
  #define DRM_IOCTL_AMDGPU_SCHED		DRM_IOW(DRM_COMMAND_BASE + DRM_AMDGPU_SCHED, union drm_amdgpu_sched)
  #define DRM_IOCTL_AMDGPU_FREESYNC	DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_FREESYNC, struct drm_amdgpu_freesync)
+#define DRM_IOCTL_AMDGPU_GEM_FIND_BO	DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_GEM_FIND_BO, union drm_amdgpu_gem_find_bo)
 
  /**
   * DOC: memory domains
@@ -392,6 +394,25 @@ struct drm_amdgpu_gem_wait_idle_out {
  	struct drm_amdgpu_gem_wait_idle_out out;
  };
 
+struct drm_amdgpu_gem_find_bo_in {
+	/* CPU address */
+	__u64			addr;
+	/* memory size from CPU address */
+	__u64			size;
+};
+
+struct drm_amdgpu_gem_find_bo_out {
+	/* offset in bo */
+	__u64			offset;
+	/* resulting GEM handle */

include/uapi/drm/amdgpu_drm.h struct drm_amdgpu_gem_wait_idle_out { · line 417

Quoted 10 lines
include/uapi/drm/amdgpu_drm.h +7 -0
+	__u32			handle;
+};
+
+union drm_amdgpu_gem_find_bo {
+	struct drm_amdgpu_gem_find_bo_in in;
+	struct drm_amdgpu_gem_find_bo_out out; };
+
  struct drm_amdgpu_wait_cs_in {
  	/* Command submission handle
           * handle equals 0 means none to wait for

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

#9 lore raw
In reply to a message not in the archive
Quoted 8 lines · David(ChunMing)

Typo, excepted -> expected

-----Original Message-----
From: amd-gfx <[email protected]> On Behalf Of Zhou, David(ChunMing)
Sent: Tuesday, July 31, 2018 9:41 AM
To: Koenig, Christian <[email protected]>; Zhang, Jerry <[email protected]>; [email protected]
Subject: RE: [PATCH 1/2] drm/amdgpu: return bo itself if userptr is cpu addr of bo (v3)

Quoted 7 lines

Thanks for Jerry still remembers this series.

Hi Christian,

For upstream of this feature, seems we already had agreement long time ago. Two reasons for upstreaming:
1. this bug was found by an opengl game, so this bug also is in mesa driver in theory.
2. after upstream these patches, we can reduce pro specific patches, and close to open source.

Thanks for explanation.

Quoted 2 lines

Btw, an unit test is excepted when upstreaming, I remember Alex mentioned.

Yes, I will.

Quoted 55 lines

Thanks,
David Zhou
-----Original Message-----
From: Christian König <[email protected]>
Sent: Monday, July 30, 2018 6:48 PM
To: Zhang, Jerry <[email protected]>; [email protected]
Cc: Zhou, David(ChunMing) <[email protected]>; Koenig, Christian <[email protected]>
Subject: Re: [PATCH 1/2] drm/amdgpu: return bo itself if userptr is cpu addr of bo (v3)

Am 30.07.2018 um 12:02 schrieb Junwei Zhang:

Quoted 5 lines

From: Chunming Zhou <[email protected]>

v2: get original gem handle from gobj
v3: update find bo data structure as union(in, out)
simply some code logic

Do we now have an open source user for this, so that we can upstream it? One more point below.

Quoted 35 lines
---
   drivers/gpu/drm/amd/amdgpu/amdgpu.h     |  2 ++
   drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 63 +++++++++++++++++++++++++++++++++
   drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c |  3 +-
   include/uapi/drm/amdgpu_drm.h           | 21 +++++++++++
   4 files changed, 88 insertions(+), 1 deletion(-)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +2 -0
index 4cd20e7..46c370b 100644
@@ -1213,6 +1213,8 @@ int amdgpu_gem_info_ioctl(struct drm_device *dev, void *data,
   			  struct drm_file *filp);
   int amdgpu_gem_userptr_ioctl(struct drm_device *dev, void *data,
   			struct drm_file *filp);
+int amdgpu_gem_find_bo_by_cpu_mapping_ioctl(struct drm_device *dev, void *data,
+					    struct drm_file *filp);
   int amdgpu_gem_mmap_ioctl(struct drm_device *dev, void *data,
   			  struct drm_file *filp);
   int amdgpu_gem_wait_idle_ioctl(struct drm_device *dev, void *data,
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +0 -0
index 71792d8..bae8417 100644
@@ -288,6 +288,69 @@ int amdgpu_gem_create_ioctl(struct drm_device *dev, void *data,
   	return 0;
   }
 
Quoted 15 lines

drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c int amdgpu_gem_create_ioctl(struct drm_device *dev, void *data, · line 303

Quoted 13 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +13 -0
+static int amdgpu_gem_get_handle_from_object(struct drm_file *filp,
+					     struct drm_gem_object *obj) {
+	int i;
+	struct drm_gem_object *tmp;
+
+	spin_lock(&filp->table_lock);
+	idr_for_each_entry(&filp->object_idr, tmp, i) {
+		if (obj == tmp) {
+			drm_gem_object_reference(obj);
+			spin_unlock(&filp->table_lock);
+			return i;
+		}
+	}

Please double check if that is still up to date.

We may have to replace drm_gem_object_reference() with drm_gem_object_get().

On 2nd thought, do we really need to do reference every time?
if UMD find the same gem object for 3 times, it also need to explicitly free(put) that object for 3 times?

Quoted 2 lines

I think we could as well try to use the DMA-buf handle tree for that.

If the bo is not import/export, DMA-buf will be NULL always.

Regards,
Jerry

Quoted 113 lines

Christian.

Quoted 110 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +49 -0
+	spin_unlock(&filp->table_lock);
+
+	return 0;
+}
+
+int amdgpu_gem_find_bo_by_cpu_mapping_ioctl(struct drm_device *dev, void *data,
+					    struct drm_file *filp)
+{
+	union drm_amdgpu_gem_find_bo *args = data;
+	struct drm_gem_object *gobj;
+	struct amdgpu_bo *bo;
+	struct ttm_buffer_object *tbo;
+	struct vm_area_struct *vma;
+	uint32_t handle;
+	int r;
+
+	if (offset_in_page(args->in.addr | args->in.size))
+		return -EINVAL;
+
+	down_read(&current->mm->mmap_sem);
+	vma = find_vma(current->mm, args->in.addr);
+	if (!vma || vma->vm_file != filp->filp ||
+	    (args->in.size > (vma->vm_end - args->in.addr))) {
+		args->out.handle = 0;
+		up_read(&current->mm->mmap_sem);
+		return -EINVAL;
+	}
+	args->out.offset = args->in.addr - vma->vm_start;
+
+	tbo = vma->vm_private_data;
+	bo = container_of(tbo, struct amdgpu_bo, tbo);
+	amdgpu_bo_ref(bo);
+	gobj = &bo->gem_base;
+
+	handle = amdgpu_gem_get_handle_from_object(filp, gobj);
+	if (!handle) {
+		r = drm_gem_handle_create(filp, gobj, &handle);
+		if (r) {
+			DRM_ERROR("create gem handle failed\n");
+			up_read(&current->mm->mmap_sem);
+			return r;
+		}
+	}
+	args->out.handle = handle;
+	up_read(&current->mm->mmap_sem);
+
+	return 0;
+}
+
   int amdgpu_gem_userptr_ioctl(struct drm_device *dev, void *data,
   			     struct drm_file *filp)
   {
drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c +3 -1
index 7956848..1bd2cc1 100644
@@ -1100,7 +1100,8 @@ void amdgpu_disable_vblank_kms(struct drm_device *dev, unsigned int pipe)
   	DRM_IOCTL_DEF_DRV(AMDGPU_GEM_VA, amdgpu_gem_va_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
   	DRM_IOCTL_DEF_DRV(AMDGPU_GEM_OP, amdgpu_gem_op_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
   	DRM_IOCTL_DEF_DRV(AMDGPU_GEM_USERPTR, amdgpu_gem_userptr_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
-	DRM_IOCTL_DEF_DRV(AMDGPU_FREESYNC, amdgpu_display_freesync_ioctl, DRM_MASTER)
+	DRM_IOCTL_DEF_DRV(AMDGPU_FREESYNC, amdgpu_display_freesync_ioctl, DRM_MASTER),
+	DRM_IOCTL_DEF_DRV(AMDGPU_GEM_FIND_BO,
+amdgpu_gem_find_bo_by_cpu_mapping_ioctl, DRM_AUTH|DRM_RENDER_ALLOW)
   };
   const int amdgpu_max_kms_ioctl = ARRAY_SIZE(amdgpu_ioctls_kms);
(quoted diff) +0 -0

b/include/uapi/drm/amdgpu_drm.h index 94444ee..000c415 100644

include/uapi/drm/amdgpu_drm.h +20 -0
@@ -54,6 +54,7 @@
   #define DRM_AMDGPU_VM			0x13
   #define DRM_AMDGPU_FENCE_TO_HANDLE	0x14
   #define DRM_AMDGPU_SCHED		0x15
+#define DRM_AMDGPU_GEM_FIND_BO		0x16
   /* not upstream */
   #define DRM_AMDGPU_FREESYNC	        0x5d
 
@@ -74,6 +75,7 @@
   #define DRM_IOCTL_AMDGPU_FENCE_TO_HANDLE DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_FENCE_TO_HANDLE, union drm_amdgpu_fence_to_handle)
   #define DRM_IOCTL_AMDGPU_SCHED		DRM_IOW(DRM_COMMAND_BASE + DRM_AMDGPU_SCHED, union drm_amdgpu_sched)
   #define DRM_IOCTL_AMDGPU_FREESYNC	DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_FREESYNC, struct drm_amdgpu_freesync)
+#define DRM_IOCTL_AMDGPU_GEM_FIND_BO	DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_GEM_FIND_BO, union drm_amdgpu_gem_find_bo)
 
   /**
    * DOC: memory domains
@@ -392,6 +394,25 @@ struct drm_amdgpu_gem_wait_idle_out {
   	struct drm_amdgpu_gem_wait_idle_out out;
   };
 
+struct drm_amdgpu_gem_find_bo_in {
+	/* CPU address */
+	__u64			addr;
+	/* memory size from CPU address */
+	__u64			size;
+};
+
+struct drm_amdgpu_gem_find_bo_out {
+	/* offset in bo */
+	__u64			offset;
+	/* resulting GEM handle */
+	__u32			handle;
+};
+
+union drm_amdgpu_gem_find_bo {
+	struct drm_amdgpu_gem_find_bo_in in;
+	struct drm_amdgpu_gem_find_bo_out out; };
+
Quoted 9 lines

include/uapi/drm/amdgpu_drm.h struct drm_amdgpu_gem_wait_idle_out { · line 417

Quoted 3 lines
include/uapi/drm/amdgpu_drm.h +0 -0
   struct drm_amdgpu_wait_cs_in {
   	/* Command submission handle
            * handle equals 0 means none to wait for

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

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

#10 lore raw
In reply to a message not in the archive

Am 31.07.2018 um 03:40 schrieb Zhou, David(ChunMing):

Quoted 7 lines

Thanks for Jerry still remembers this series.

Hi Christian,

For upstream of this feature, seems we already had agreement long time ago. Two reasons for upstreaming:
1. this bug was found by an opengl game, so this bug also is in mesa driver in theory.
2. after upstream these patches, we can reduce pro specific patches, and close to open source.

I mean the functionality is actually rather simple, so I don't see much of an issue with that.

Quoted 1 line

Btw, an unit test is excepted when upstreaming, I remember Alex mentioned.

Yeah, agree an unit test is really a must have for that.

Christian.

Quoted 181 lines

Thanks,
David Zhou
-----Original Message-----
From: Christian König <[email protected]>
Sent: Monday, July 30, 2018 6:48 PM
To: Zhang, Jerry <[email protected]>; [email protected]
Cc: Zhou, David(ChunMing) <[email protected]>; Koenig, Christian <[email protected]>
Subject: Re: [PATCH 1/2] drm/amdgpu: return bo itself if userptr is cpu addr of bo (v3)

Am 30.07.2018 um 12:02 schrieb Junwei Zhang:

Quoted 5 lines

From: Chunming Zhou <[email protected]>

v2: get original gem handle from gobj
v3: update find bo data structure as union(in, out)
simply some code logic

Do we now have an open source user for this, so that we can upstream it? One more point below.

Quoted 37 lines
---
   drivers/gpu/drm/amd/amdgpu/amdgpu.h     |  2 ++
   drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 63 +++++++++++++++++++++++++++++++++
   drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c |  3 +-
   include/uapi/drm/amdgpu_drm.h           | 21 +++++++++++
   4 files changed, 88 insertions(+), 1 deletion(-)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +2 -0
index 4cd20e7..46c370b 100644
@@ -1213,6 +1213,8 @@ int amdgpu_gem_info_ioctl(struct drm_device *dev, void *data,
   			  struct drm_file *filp);
   int amdgpu_gem_userptr_ioctl(struct drm_device *dev, void *data,
   			struct drm_file *filp);
+int amdgpu_gem_find_bo_by_cpu_mapping_ioctl(struct drm_device *dev, void *data,
+					    struct drm_file *filp);
   int amdgpu_gem_mmap_ioctl(struct drm_device *dev, void *data,
   			  struct drm_file *filp);
   int amdgpu_gem_wait_idle_ioctl(struct drm_device *dev, void *data,
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +3 -0
index 71792d8..bae8417 100644
@@ -288,6 +288,69 @@ int amdgpu_gem_create_ioctl(struct drm_device *dev, void *data,
   	return 0;
   }
 
+static int amdgpu_gem_get_handle_from_object(struct drm_file *filp,
+					     struct drm_gem_object *obj) {
+	int i;

drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c int amdgpu_gem_create_ioctl(struct drm_device *dev, void *data, · line 303

Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +10 -0
+	struct drm_gem_object *tmp;
+
+	spin_lock(&filp->table_lock);
+	idr_for_each_entry(&filp->object_idr, tmp, i) {
+		if (obj == tmp) {
+			drm_gem_object_reference(obj);
+			spin_unlock(&filp->table_lock);
+			return i;
+		}
+	}

Please double check if that is still up to date.

I think we could as well try to use the DMA-buf handle tree for that.

Christian.

Quoted 109 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +49 -0
+	spin_unlock(&filp->table_lock);
+
+	return 0;
+}
+
+int amdgpu_gem_find_bo_by_cpu_mapping_ioctl(struct drm_device *dev, void *data,
+					    struct drm_file *filp)
+{
+	union drm_amdgpu_gem_find_bo *args = data;
+	struct drm_gem_object *gobj;
+	struct amdgpu_bo *bo;
+	struct ttm_buffer_object *tbo;
+	struct vm_area_struct *vma;
+	uint32_t handle;
+	int r;
+
+	if (offset_in_page(args->in.addr | args->in.size))
+		return -EINVAL;
+
+	down_read(&current->mm->mmap_sem);
+	vma = find_vma(current->mm, args->in.addr);
+	if (!vma || vma->vm_file != filp->filp ||
+	    (args->in.size > (vma->vm_end - args->in.addr))) {
+		args->out.handle = 0;
+		up_read(&current->mm->mmap_sem);
+		return -EINVAL;
+	}
+	args->out.offset = args->in.addr - vma->vm_start;
+
+	tbo = vma->vm_private_data;
+	bo = container_of(tbo, struct amdgpu_bo, tbo);
+	amdgpu_bo_ref(bo);
+	gobj = &bo->gem_base;
+
+	handle = amdgpu_gem_get_handle_from_object(filp, gobj);
+	if (!handle) {
+		r = drm_gem_handle_create(filp, gobj, &handle);
+		if (r) {
+			DRM_ERROR("create gem handle failed\n");
+			up_read(&current->mm->mmap_sem);
+			return r;
+		}
+	}
+	args->out.handle = handle;
+	up_read(&current->mm->mmap_sem);
+
+	return 0;
+}
+
   int amdgpu_gem_userptr_ioctl(struct drm_device *dev, void *data,
   			     struct drm_file *filp)
   {
drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c +3 -1
index 7956848..1bd2cc1 100644
@@ -1100,7 +1100,8 @@ void amdgpu_disable_vblank_kms(struct drm_device *dev, unsigned int pipe)
   	DRM_IOCTL_DEF_DRV(AMDGPU_GEM_VA, amdgpu_gem_va_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
   	DRM_IOCTL_DEF_DRV(AMDGPU_GEM_OP, amdgpu_gem_op_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
   	DRM_IOCTL_DEF_DRV(AMDGPU_GEM_USERPTR, amdgpu_gem_userptr_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
-	DRM_IOCTL_DEF_DRV(AMDGPU_FREESYNC, amdgpu_display_freesync_ioctl, DRM_MASTER)
+	DRM_IOCTL_DEF_DRV(AMDGPU_FREESYNC, amdgpu_display_freesync_ioctl, DRM_MASTER),
+	DRM_IOCTL_DEF_DRV(AMDGPU_GEM_FIND_BO,
+amdgpu_gem_find_bo_by_cpu_mapping_ioctl, DRM_AUTH|DRM_RENDER_ALLOW)
   };
   const int amdgpu_max_kms_ioctl = ARRAY_SIZE(amdgpu_ioctls_kms);
(quoted diff) +0 -0

b/include/uapi/drm/amdgpu_drm.h index 94444ee..000c415 100644

include/uapi/drm/amdgpu_drm.h +19 -0
@@ -54,6 +54,7 @@
   #define DRM_AMDGPU_VM			0x13
   #define DRM_AMDGPU_FENCE_TO_HANDLE	0x14
   #define DRM_AMDGPU_SCHED		0x15
+#define DRM_AMDGPU_GEM_FIND_BO		0x16
   /* not upstream */
   #define DRM_AMDGPU_FREESYNC	        0x5d
 
@@ -74,6 +75,7 @@
   #define DRM_IOCTL_AMDGPU_FENCE_TO_HANDLE DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_FENCE_TO_HANDLE, union drm_amdgpu_fence_to_handle)
   #define DRM_IOCTL_AMDGPU_SCHED		DRM_IOW(DRM_COMMAND_BASE + DRM_AMDGPU_SCHED, union drm_amdgpu_sched)
   #define DRM_IOCTL_AMDGPU_FREESYNC	DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_FREESYNC, struct drm_amdgpu_freesync)
+#define DRM_IOCTL_AMDGPU_GEM_FIND_BO	DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_GEM_FIND_BO, union drm_amdgpu_gem_find_bo)
 
   /**
    * DOC: memory domains
@@ -392,6 +394,25 @@ struct drm_amdgpu_gem_wait_idle_out {
   	struct drm_amdgpu_gem_wait_idle_out out;
   };
 
+struct drm_amdgpu_gem_find_bo_in {
+	/* CPU address */
+	__u64			addr;
+	/* memory size from CPU address */
+	__u64			size;
+};
+
+struct drm_amdgpu_gem_find_bo_out {
+	/* offset in bo */
+	__u64			offset;
+	/* resulting GEM handle */
+	__u32			handle;
+};
+
+union drm_amdgpu_gem_find_bo {
+	struct drm_amdgpu_gem_find_bo_in in;
+	struct drm_amdgpu_gem_find_bo_out out; };
Quoted 8 lines

include/uapi/drm/amdgpu_drm.h struct drm_amdgpu_gem_wait_idle_out { · line 418

Quoted 4 lines
include/uapi/drm/amdgpu_drm.h +1 -0
+
   struct drm_amdgpu_wait_cs_in {
   	/* Command submission handle
            * handle equals 0 means none to wait for

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

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

#11 lore raw
In reply to a message not in the archive
Quoted 49 lines · Christian König

Am 30.07.2018 um 12:02 schrieb Junwei Zhang:

Quoted 5 lines

From: Chunming Zhou <[email protected]>

v2: get original gem handle from gobj
v3: update find bo data structure as union(in, out)
simply some code logic

Do we now have an open source user for this, so that we can upstream it? One more point below.

Quoted 40 lines
---
  drivers/gpu/drm/amd/amdgpu/amdgpu.h     |  2 ++
  drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 63 +++++++++++++++++++++++++++++++++
  drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c |  3 +-
  include/uapi/drm/amdgpu_drm.h           | 21 +++++++++++
  4 files changed, 88 insertions(+), 1 deletion(-)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +2 -0
index 4cd20e7..46c370b 100644
@@ -1213,6 +1213,8 @@ int amdgpu_gem_info_ioctl(struct drm_device *dev, void *data,
                struct drm_file *filp);
  int amdgpu_gem_userptr_ioctl(struct drm_device *dev, void *data,
              struct drm_file *filp);
+int amdgpu_gem_find_bo_by_cpu_mapping_ioctl(struct drm_device *dev, void *data,
+                        struct drm_file *filp);
  int amdgpu_gem_mmap_ioctl(struct drm_device *dev, void *data,
                struct drm_file *filp);
  int amdgpu_gem_wait_idle_ioctl(struct drm_device *dev, void *data,
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +8 -0
index 71792d8..bae8417 100644
@@ -288,6 +288,69 @@ int amdgpu_gem_create_ioctl(struct drm_device *dev, void *data,
      return 0;
  }
+static int amdgpu_gem_get_handle_from_object(struct drm_file *filp,
+                         struct drm_gem_object *obj)
+{
+    int i;
+    struct drm_gem_object *tmp;
+
+    spin_lock(&filp->table_lock);
+    idr_for_each_entry(&filp->object_idr, tmp, i) {
Quoted 8 lines

drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c int amdgpu_gem_create_ioctl(struct drm_device *dev, void *data, · line 304

Quoted 6 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +6 -0
+        if (obj == tmp) {
+            drm_gem_object_reference(obj);
+            spin_unlock(&filp->table_lock);
+            return i;
+        }
+    }

Please double check if that is still up to date.

We may have to replace drm_gem_object_reference() with drm_gem_object_get().

On 2nd thought, do we really need to do reference every time?
if UMD find the same gem object for 3 times, it also need to explicitly free(put) that object for 3 times?

Quoted 2 lines

I think we could as well try to use the DMA-buf handle tree for that.

If the bo is not import/export, DMA-buf will be NULL always.

Regards,
Jerry

Quoted 103 lines

Christian.

Quoted 100 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +49 -0
+    spin_unlock(&filp->table_lock);
+
+    return 0;
+}
+
+int amdgpu_gem_find_bo_by_cpu_mapping_ioctl(struct drm_device *dev, void *data,
+                        struct drm_file *filp)
+{
+    union drm_amdgpu_gem_find_bo *args = data;
+    struct drm_gem_object *gobj;
+    struct amdgpu_bo *bo;
+    struct ttm_buffer_object *tbo;
+    struct vm_area_struct *vma;
+    uint32_t handle;
+    int r;
+
+    if (offset_in_page(args->in.addr | args->in.size))
+        return -EINVAL;
+
+    down_read(&current->mm->mmap_sem);
+    vma = find_vma(current->mm, args->in.addr);
+    if (!vma || vma->vm_file != filp->filp ||
+        (args->in.size > (vma->vm_end - args->in.addr))) {
+        args->out.handle = 0;
+        up_read(&current->mm->mmap_sem);
+        return -EINVAL;
+    }
+    args->out.offset = args->in.addr - vma->vm_start;
+
+    tbo = vma->vm_private_data;
+    bo = container_of(tbo, struct amdgpu_bo, tbo);
+    amdgpu_bo_ref(bo);
+    gobj = &bo->gem_base;
+
+    handle = amdgpu_gem_get_handle_from_object(filp, gobj);
+    if (!handle) {
+        r = drm_gem_handle_create(filp, gobj, &handle);
+        if (r) {
+            DRM_ERROR("create gem handle failed\n");
+            up_read(&current->mm->mmap_sem);
+            return r;
+        }
+    }
+    args->out.handle = handle;
+    up_read(&current->mm->mmap_sem);
+
+    return 0;
+}
+
  int amdgpu_gem_userptr_ioctl(struct drm_device *dev, void *data,
                   struct drm_file *filp)
  {
drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c +2 -1
index 7956848..1bd2cc1 100644
@@ -1100,7 +1100,8 @@ void amdgpu_disable_vblank_kms(struct drm_device *dev, unsigned int pipe)
      DRM_IOCTL_DEF_DRV(AMDGPU_GEM_VA, amdgpu_gem_va_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
      DRM_IOCTL_DEF_DRV(AMDGPU_GEM_OP, amdgpu_gem_op_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
      DRM_IOCTL_DEF_DRV(AMDGPU_GEM_USERPTR, amdgpu_gem_userptr_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
-    DRM_IOCTL_DEF_DRV(AMDGPU_FREESYNC, amdgpu_display_freesync_ioctl, DRM_MASTER)
+    DRM_IOCTL_DEF_DRV(AMDGPU_FREESYNC, amdgpu_display_freesync_ioctl, DRM_MASTER),
+    DRM_IOCTL_DEF_DRV(AMDGPU_GEM_FIND_BO, amdgpu_gem_find_bo_by_cpu_mapping_ioctl, DRM_AUTH|DRM_RENDER_ALLOW)
  };
  const int amdgpu_max_kms_ioctl = ARRAY_SIZE(amdgpu_ioctls_kms);
include/uapi/drm/amdgpu_drm.h +16 -0
index 94444ee..000c415 100644
@@ -54,6 +54,7 @@
  #define DRM_AMDGPU_VM            0x13
  #define DRM_AMDGPU_FENCE_TO_HANDLE    0x14
  #define DRM_AMDGPU_SCHED        0x15
+#define DRM_AMDGPU_GEM_FIND_BO        0x16
  /* not upstream */
  #define DRM_AMDGPU_FREESYNC            0x5d
@@ -74,6 +75,7 @@
  #define DRM_IOCTL_AMDGPU_FENCE_TO_HANDLE DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_FENCE_TO_HANDLE, union drm_amdgpu_fence_to_handle)
  #define DRM_IOCTL_AMDGPU_SCHED        DRM_IOW(DRM_COMMAND_BASE + DRM_AMDGPU_SCHED, union drm_amdgpu_sched)
  #define DRM_IOCTL_AMDGPU_FREESYNC    DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_FREESYNC, struct drm_amdgpu_freesync)
+#define DRM_IOCTL_AMDGPU_GEM_FIND_BO    DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_GEM_FIND_BO, union drm_amdgpu_gem_find_bo)
  /**
   * DOC: memory domains
@@ -392,6 +394,25 @@ struct drm_amdgpu_gem_wait_idle_out {
      struct drm_amdgpu_gem_wait_idle_out out;
  };
+struct drm_amdgpu_gem_find_bo_in {
+    /* CPU address */
+    __u64            addr;
+    /* memory size from CPU address */
+    __u64            size;
+};
+
+struct drm_amdgpu_gem_find_bo_out {
+    /* offset in bo */
+    __u64            offset;
+    /* resulting GEM handle */
+    __u32            handle;
+};
+
Quoted 9 lines

include/uapi/drm/amdgpu_drm.h struct drm_amdgpu_gem_wait_idle_out { · line 418

Quoted 8 lines
include/uapi/drm/amdgpu_drm.h +5 -0
+union drm_amdgpu_gem_find_bo {
+    struct drm_amdgpu_gem_find_bo_in in;
+    struct drm_amdgpu_gem_find_bo_out out;
+};
+
  struct drm_amdgpu_wait_cs_in {
      /* Command submission handle
           * handle equals 0 means none to wait for

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

#13 lore raw
In reply to a message not in the archive

Am 31.07.2018 um 08:58 schrieb Zhang, Jerry (Junwei):

Quoted 63 lines · Christian König
Quoted 53 lines

Am 30.07.2018 um 12:02 schrieb Junwei Zhang:

Quoted 5 lines

From: Chunming Zhou <[email protected]>

v2: get original gem handle from gobj
v3: update find bo data structure as union(in, out) simply some code logic

Do we now have an open source user for this, so that we can upstream it? One more point below.

Quoted 44 lines

---
  drivers/gpu/drm/amd/amdgpu/amdgpu.h     |  2 ++ drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 63
+++++++++++++++++++++++++++++++++
  drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c |  3 +- include/uapi/drm/amdgpu_drm.h           | 21 +++++++++++

  4 files changed, 88 insertions(+), 1 deletion(-)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +2 -0
index 4cd20e7..46c370b 100644
@@ -1213,6 +1213,8 @@ int amdgpu_gem_info_ioctl(struct drm_device 
 *dev, void *data,
                 struct drm_file *filp);
   int amdgpu_gem_userptr_ioctl(struct drm_device *dev, void *data,
               struct drm_file *filp);
+int amdgpu_gem_find_bo_by_cpu_mapping_ioctl(struct drm_device *dev, 
 void *data,
+                        struct drm_file *filp);
   int amdgpu_gem_mmap_ioctl(struct drm_device *dev, void *data,

                struct drm_file *filp);
  int amdgpu_gem_wait_idle_ioctl(struct drm_device *dev, void *data,

drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +8 -0
index 71792d8..bae8417 100644
@@ -288,6 +288,69 @@ int amdgpu_gem_create_ioctl(struct drm_device 
 *dev, void *data,
       return 0;
   }
+static int amdgpu_gem_get_handle_from_object(struct drm_file *filp,
+                         struct drm_gem_object *obj)
+{
+    int i;
+    struct drm_gem_object *tmp;
+
+    spin_lock(&filp->table_lock);
+    idr_for_each_entry(&filp->object_idr, tmp, i) {
Quoted 8 lines

drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c int amdgpu_gem_create_ioctl(struct drm_device *dev, void *data, · line 304

Quoted 6 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +6 -0
+        if (obj == tmp) {
+            drm_gem_object_reference(obj);
+            spin_unlock(&filp->table_lock);
+            return i;
+        }
+    }

Please double check if that is still up to date.

Quoted 3 lines

We may have to replace drm_gem_object_reference() with drm_gem_object_get().

On 2nd thought, do we really need to do reference every time?

Yes, that's a must have. Otherwise the handle could be freed and reused already when we return.

Quoted 1 line

if UMD find the same gem object for 3 times, it also need to explicitly free(put) that object for 3 times?

Correct yes. Thinking more about this the real problem is to translate the handle into a structure in libdrm.

Here we are back to the problem Marek and Michel has been working on for a while that we always need to be able to translate a handle into a bo structure.....

So that needs to be solved before we can upstream the changes.

Regards,
Christian.

Quoted 128 lines
Quoted 4 lines

I think we could as well try to use the DMA-buf handle tree for that.

If the bo is not import/export, DMA-buf will be NULL always.

Regards,
Jerry

Quoted 110 lines

Christian.

Quoted 107 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +49 -0
+ spin_unlock(&filp->table_lock);
+
+    return 0;
+}
+
+int amdgpu_gem_find_bo_by_cpu_mapping_ioctl(struct drm_device *dev, 
 void *data,
+                        struct drm_file *filp)
+{
+    union drm_amdgpu_gem_find_bo *args = data;
+    struct drm_gem_object *gobj;
+    struct amdgpu_bo *bo;
+    struct ttm_buffer_object *tbo;
+    struct vm_area_struct *vma;
+    uint32_t handle;
+    int r;
+
+    if (offset_in_page(args->in.addr | args->in.size))
+        return -EINVAL;
+
+    down_read(&current->mm->mmap_sem);
+    vma = find_vma(current->mm, args->in.addr);
+    if (!vma || vma->vm_file != filp->filp ||
+        (args->in.size > (vma->vm_end - args->in.addr))) {
+        args->out.handle = 0;
+        up_read(&current->mm->mmap_sem);
+        return -EINVAL;
+    }
+    args->out.offset = args->in.addr - vma->vm_start;
+
+    tbo = vma->vm_private_data;
+    bo = container_of(tbo, struct amdgpu_bo, tbo);
+    amdgpu_bo_ref(bo);
+    gobj = &bo->gem_base;
+
+    handle = amdgpu_gem_get_handle_from_object(filp, gobj);
+    if (!handle) {
+        r = drm_gem_handle_create(filp, gobj, &handle);
+        if (r) {
+            DRM_ERROR("create gem handle failed\n");
+            up_read(&current->mm->mmap_sem);
+            return r;
+        }
+    }
+    args->out.handle = handle;
+    up_read(&current->mm->mmap_sem);
+
+    return 0;
+}
+
   int amdgpu_gem_userptr_ioctl(struct drm_device *dev, void *data,
                    struct drm_file *filp)

  {

drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c +2 -1
index 7956848..1bd2cc1 100644
@@ -1100,7 +1100,8 @@ void amdgpu_disable_vblank_kms(struct 
 drm_device *dev, unsigned int pipe)
       DRM_IOCTL_DEF_DRV(AMDGPU_GEM_VA, amdgpu_gem_va_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
       DRM_IOCTL_DEF_DRV(AMDGPU_GEM_OP, amdgpu_gem_op_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
       DRM_IOCTL_DEF_DRV(AMDGPU_GEM_USERPTR, amdgpu_gem_userptr_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
-    DRM_IOCTL_DEF_DRV(AMDGPU_FREESYNC, 
 amdgpu_display_freesync_ioctl, DRM_MASTER)
+    DRM_IOCTL_DEF_DRV(AMDGPU_FREESYNC, 
 amdgpu_display_freesync_ioctl, DRM_MASTER),
+    DRM_IOCTL_DEF_DRV(AMDGPU_GEM_FIND_BO, 

amdgpu_gem_find_bo_by_cpu_mapping_ioctl, DRM_AUTH|DRM_RENDER_ALLOW) };
  const int amdgpu_max_kms_ioctl = ARRAY_SIZE(amdgpu_ioctls_kms);

include/uapi/drm/amdgpu_drm.h +3 -0
index 94444ee..000c415 100644
@@ -54,6 +54,7 @@
   #define DRM_AMDGPU_VM            0x13
   #define DRM_AMDGPU_FENCE_TO_HANDLE    0x14
   #define DRM_AMDGPU_SCHED        0x15
+#define DRM_AMDGPU_GEM_FIND_BO        0x16
   /* not upstream */
   #define DRM_AMDGPU_FREESYNC            0x5d
@@ -74,6 +75,7 @@
   #define DRM_IOCTL_AMDGPU_FENCE_TO_HANDLE DRM_IOWR(DRM_COMMAND_BASE 
+ DRM_AMDGPU_FENCE_TO_HANDLE, union drm_amdgpu_fence_to_handle)
   #define DRM_IOCTL_AMDGPU_SCHED DRM_IOW(DRM_COMMAND_BASE + DRM_AMDGPU_SCHED, union drm_amdgpu_sched)
   #define DRM_IOCTL_AMDGPU_FREESYNC DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_FREESYNC, struct drm_amdgpu_freesync)
+#define DRM_IOCTL_AMDGPU_GEM_FIND_BO DRM_IOWR(DRM_COMMAND_BASE + 
 DRM_AMDGPU_GEM_FIND_BO, union drm_amdgpu_gem_find_bo)
   /**

   * DOC: memory domains

include/uapi/drm/amdgpu_drm.h +14 -0
@@ -392,6 +394,25 @@ struct drm_amdgpu_gem_wait_idle_out {
       struct drm_amdgpu_gem_wait_idle_out out;
   };
+struct drm_amdgpu_gem_find_bo_in {
+    /* CPU address */
+    __u64            addr;
+    /* memory size from CPU address */
+    __u64            size;
+};
+
+struct drm_amdgpu_gem_find_bo_out {
+    /* offset in bo */
+    __u64            offset;
+    /* resulting GEM handle */
+    __u32            handle;
+};
+
Quoted 9 lines

include/uapi/drm/amdgpu_drm.h struct drm_amdgpu_gem_wait_idle_out { · line 418

Quoted 8 lines
include/uapi/drm/amdgpu_drm.h +5 -0
+union drm_amdgpu_gem_find_bo {
+    struct drm_amdgpu_gem_find_bo_in in;
+    struct drm_amdgpu_gem_find_bo_out out;
+};
+
   struct drm_amdgpu_wait_cs_in {
       /* Command submission handle
            * handle equals 0 means none to wait for
Quoted 4 lines

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

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

#15 lore raw
In reply to a message not in the archive
Quoted 66 lines · Christian König

Am 31.07.2018 um 08:58 schrieb Zhang, Jerry (Junwei):

Quoted 59 lines · Christian König
Quoted 49 lines

Am 30.07.2018 um 12:02 schrieb Junwei Zhang:

Quoted 5 lines

From: Chunming Zhou <[email protected]>

v2: get original gem handle from gobj
v3: update find bo data structure as union(in, out)
simply some code logic

Do we now have an open source user for this, so that we can upstream it? One more point below.

Quoted 40 lines
---
  drivers/gpu/drm/amd/amdgpu/amdgpu.h     |  2 ++
  drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 63 +++++++++++++++++++++++++++++++++
  drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c |  3 +-
  include/uapi/drm/amdgpu_drm.h           | 21 +++++++++++
  4 files changed, 88 insertions(+), 1 deletion(-)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +2 -0
index 4cd20e7..46c370b 100644
@@ -1213,6 +1213,8 @@ int amdgpu_gem_info_ioctl(struct drm_device *dev, void *data,
                struct drm_file *filp);
  int amdgpu_gem_userptr_ioctl(struct drm_device *dev, void *data,
              struct drm_file *filp);
+int amdgpu_gem_find_bo_by_cpu_mapping_ioctl(struct drm_device *dev, void *data,
+                        struct drm_file *filp);
  int amdgpu_gem_mmap_ioctl(struct drm_device *dev, void *data,
                struct drm_file *filp);
  int amdgpu_gem_wait_idle_ioctl(struct drm_device *dev, void *data,
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +8 -0
index 71792d8..bae8417 100644
@@ -288,6 +288,69 @@ int amdgpu_gem_create_ioctl(struct drm_device *dev, void *data,
      return 0;
  }
+static int amdgpu_gem_get_handle_from_object(struct drm_file *filp,
+                         struct drm_gem_object *obj)
+{
+    int i;
+    struct drm_gem_object *tmp;
+
+    spin_lock(&filp->table_lock);
+    idr_for_each_entry(&filp->object_idr, tmp, i) {
Quoted 8 lines

drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c int amdgpu_gem_create_ioctl(struct drm_device *dev, void *data, · line 303

Quoted 6 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +6 -0
+        if (obj == tmp) {
+            drm_gem_object_reference(obj);
+            spin_unlock(&filp->table_lock);
+            return i;
+        }
+    }

Please double check if that is still up to date.

Quoted 3 lines

We may have to replace drm_gem_object_reference() with drm_gem_object_get().

On 2nd thought, do we really need to do reference every time?

Yes, that's a must have. Otherwise the handle could be freed and reused already when we return.

Quoted 7 lines
Quoted 1 line

if UMD find the same gem object for 3 times, it also need to explicitly free(put) that object for 3 times?

Correct yes. Thinking more about this the real problem is to translate the handle into a structure in libdrm.

Here we are back to the problem Marek and Michel has been working on for a while that we always need to be able to translate a handle into a bo structure.....

So that needs to be solved before we can upstream the changes.

Thanks for your info.
It's better to fix that before upstream.

Regards,
Jerry

Quoted 130 lines

Regards,
Christian.

Quoted 125 lines
Quoted 4 lines

I think we could as well try to use the DMA-buf handle tree for that.

If the bo is not import/export, DMA-buf will be NULL always.

Regards,
Jerry

Quoted 103 lines

Christian.

Quoted 100 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +49 -0
+ spin_unlock(&filp->table_lock);
+
+    return 0;
+}
+
+int amdgpu_gem_find_bo_by_cpu_mapping_ioctl(struct drm_device *dev, void *data,
+                        struct drm_file *filp)
+{
+    union drm_amdgpu_gem_find_bo *args = data;
+    struct drm_gem_object *gobj;
+    struct amdgpu_bo *bo;
+    struct ttm_buffer_object *tbo;
+    struct vm_area_struct *vma;
+    uint32_t handle;
+    int r;
+
+    if (offset_in_page(args->in.addr | args->in.size))
+        return -EINVAL;
+
+    down_read(&current->mm->mmap_sem);
+    vma = find_vma(current->mm, args->in.addr);
+    if (!vma || vma->vm_file != filp->filp ||
+        (args->in.size > (vma->vm_end - args->in.addr))) {
+        args->out.handle = 0;
+        up_read(&current->mm->mmap_sem);
+        return -EINVAL;
+    }
+    args->out.offset = args->in.addr - vma->vm_start;
+
+    tbo = vma->vm_private_data;
+    bo = container_of(tbo, struct amdgpu_bo, tbo);
+    amdgpu_bo_ref(bo);
+    gobj = &bo->gem_base;
+
+    handle = amdgpu_gem_get_handle_from_object(filp, gobj);
+    if (!handle) {
+        r = drm_gem_handle_create(filp, gobj, &handle);
+        if (r) {
+            DRM_ERROR("create gem handle failed\n");
+            up_read(&current->mm->mmap_sem);
+            return r;
+        }
+    }
+    args->out.handle = handle;
+    up_read(&current->mm->mmap_sem);
+
+    return 0;
+}
+
  int amdgpu_gem_userptr_ioctl(struct drm_device *dev, void *data,
                   struct drm_file *filp)
  {
drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c +2 -1
index 7956848..1bd2cc1 100644
@@ -1100,7 +1100,8 @@ void amdgpu_disable_vblank_kms(struct drm_device *dev, unsigned int pipe)
      DRM_IOCTL_DEF_DRV(AMDGPU_GEM_VA, amdgpu_gem_va_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
      DRM_IOCTL_DEF_DRV(AMDGPU_GEM_OP, amdgpu_gem_op_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
      DRM_IOCTL_DEF_DRV(AMDGPU_GEM_USERPTR, amdgpu_gem_userptr_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
-    DRM_IOCTL_DEF_DRV(AMDGPU_FREESYNC, amdgpu_display_freesync_ioctl, DRM_MASTER)
+    DRM_IOCTL_DEF_DRV(AMDGPU_FREESYNC, amdgpu_display_freesync_ioctl, DRM_MASTER),
+    DRM_IOCTL_DEF_DRV(AMDGPU_GEM_FIND_BO, amdgpu_gem_find_bo_by_cpu_mapping_ioctl, DRM_AUTH|DRM_RENDER_ALLOW)
  };
  const int amdgpu_max_kms_ioctl = ARRAY_SIZE(amdgpu_ioctls_kms);
include/uapi/drm/amdgpu_drm.h +16 -0
index 94444ee..000c415 100644
@@ -54,6 +54,7 @@
  #define DRM_AMDGPU_VM            0x13
  #define DRM_AMDGPU_FENCE_TO_HANDLE    0x14
  #define DRM_AMDGPU_SCHED        0x15
+#define DRM_AMDGPU_GEM_FIND_BO        0x16
  /* not upstream */
  #define DRM_AMDGPU_FREESYNC            0x5d
@@ -74,6 +75,7 @@
  #define DRM_IOCTL_AMDGPU_FENCE_TO_HANDLE DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_FENCE_TO_HANDLE, union drm_amdgpu_fence_to_handle)
  #define DRM_IOCTL_AMDGPU_SCHED DRM_IOW(DRM_COMMAND_BASE + DRM_AMDGPU_SCHED, union drm_amdgpu_sched)
  #define DRM_IOCTL_AMDGPU_FREESYNC DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_FREESYNC, struct drm_amdgpu_freesync)
+#define DRM_IOCTL_AMDGPU_GEM_FIND_BO DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_GEM_FIND_BO, union drm_amdgpu_gem_find_bo)
  /**
   * DOC: memory domains
@@ -392,6 +394,25 @@ struct drm_amdgpu_gem_wait_idle_out {
      struct drm_amdgpu_gem_wait_idle_out out;
  };
+struct drm_amdgpu_gem_find_bo_in {
+    /* CPU address */
+    __u64            addr;
+    /* memory size from CPU address */
+    __u64            size;
+};
+
+struct drm_amdgpu_gem_find_bo_out {
+    /* offset in bo */
+    __u64            offset;
+    /* resulting GEM handle */
+    __u32            handle;
+};
+
Quoted 9 lines

include/uapi/drm/amdgpu_drm.h struct drm_amdgpu_gem_wait_idle_out { · line 417

Quoted 8 lines
include/uapi/drm/amdgpu_drm.h +5 -0
+union drm_amdgpu_gem_find_bo {
+    struct drm_amdgpu_gem_find_bo_in in;
+    struct drm_amdgpu_gem_find_bo_out out;
+};
+
  struct drm_amdgpu_wait_cs_in {
      /* Command submission handle
           * handle equals 0 means none to wait for

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

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

#17 lore raw
In reply to a message not in the archive

Am 31.07.2018 um 10:05 schrieb Zhang, Jerry (Junwei):

Quoted 73 lines · Christian König
Quoted 72 lines

Am 31.07.2018 um 08:58 schrieb Zhang, Jerry (Junwei):

Quoted 63 lines · Christian König
Quoted 53 lines

Am 30.07.2018 um 12:02 schrieb Junwei Zhang:

Quoted 5 lines

From: Chunming Zhou <[email protected]>

v2: get original gem handle from gobj
v3: update find bo data structure as union(in, out) simply some code logic

Do we now have an open source user for this, so that we can upstream it? One more point below.

Quoted 44 lines

---
  drivers/gpu/drm/amd/amdgpu/amdgpu.h     |  2 ++ drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 63
+++++++++++++++++++++++++++++++++
  drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c |  3 +- include/uapi/drm/amdgpu_drm.h           | 21 +++++++++++

  4 files changed, 88 insertions(+), 1 deletion(-)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +2 -0
index 4cd20e7..46c370b 100644
@@ -1213,6 +1213,8 @@ int amdgpu_gem_info_ioctl(struct drm_device 
 *dev, void *data,
                 struct drm_file *filp);
   int amdgpu_gem_userptr_ioctl(struct drm_device *dev, void *data,
               struct drm_file *filp);
+int amdgpu_gem_find_bo_by_cpu_mapping_ioctl(struct drm_device 
 *dev, void *data,
+                        struct drm_file *filp);
   int amdgpu_gem_mmap_ioctl(struct drm_device *dev, void *data,

                struct drm_file *filp);
  int amdgpu_gem_wait_idle_ioctl(struct drm_device *dev, void *data,

drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +8 -0
index 71792d8..bae8417 100644
@@ -288,6 +288,69 @@ int amdgpu_gem_create_ioctl(struct drm_device 
 *dev, void *data,
       return 0;
   }
+static int amdgpu_gem_get_handle_from_object(struct drm_file *filp,
+                         struct drm_gem_object *obj)
+{
+    int i;
+    struct drm_gem_object *tmp;
+
+    spin_lock(&filp->table_lock);
+    idr_for_each_entry(&filp->object_idr, tmp, i) {
Quoted 8 lines

drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c int amdgpu_gem_create_ioctl(struct drm_device · line 304

Quoted 6 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +6 -0
+        if (obj == tmp) {
+            drm_gem_object_reference(obj);
+            spin_unlock(&filp->table_lock);
+            return i;
+        }
+    }

Please double check if that is still up to date.

Quoted 3 lines

We may have to replace drm_gem_object_reference() with drm_gem_object_get().

On 2nd thought, do we really need to do reference every time?

Yes, that's a must have. Otherwise the handle could be freed and reused already when we return.

Quoted 1 line

if UMD find the same gem object for 3 times, it also need to explicitly free(put) that object for 3 times?

Quoted 8 lines
Quoted 5 lines

Correct yes. Thinking more about this the real problem is to translate the handle into a structure in libdrm.

Here we are back to the problem Marek and Michel has been working on for a while that we always need to be able to translate a handle into a bo structure.....

So that needs to be solved before we can upstream the changes.

Thanks for your info.
It's better to fix that before upstream.

Thinking more about this the hash currently used in libdrm is not adequate any more.

E.g. we now need to be able to find all BOs based on their handle. Since the handles are dense either an r/b tree or a radix tree now sounds like the best approach to me.

Christian.

Quoted 3 lines

Regards,
Jerry

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

#19 lore raw
In reply to a message not in the archive
Quoted 74 lines · Christian König

Am 31.07.2018 um 10:05 schrieb Zhang, Jerry (Junwei):

Quoted 73 lines · Christian König
Quoted 72 lines

Am 31.07.2018 um 08:58 schrieb Zhang, Jerry (Junwei):

Quoted 59 lines · Christian König
Quoted 49 lines

Am 30.07.2018 um 12:02 schrieb Junwei Zhang:

Quoted 5 lines

From: Chunming Zhou <[email protected]>

v2: get original gem handle from gobj
v3: update find bo data structure as union(in, out)
simply some code logic

Do we now have an open source user for this, so that we can upstream it? One more point below.

Quoted 40 lines
---
  drivers/gpu/drm/amd/amdgpu/amdgpu.h     |  2 ++
  drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 63 +++++++++++++++++++++++++++++++++
  drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c |  3 +-
  include/uapi/drm/amdgpu_drm.h           | 21 +++++++++++
  4 files changed, 88 insertions(+), 1 deletion(-)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +2 -0
index 4cd20e7..46c370b 100644
@@ -1213,6 +1213,8 @@ int amdgpu_gem_info_ioctl(struct drm_device *dev, void *data,
                struct drm_file *filp);
  int amdgpu_gem_userptr_ioctl(struct drm_device *dev, void *data,
              struct drm_file *filp);
+int amdgpu_gem_find_bo_by_cpu_mapping_ioctl(struct drm_device *dev, void *data,
+                        struct drm_file *filp);
  int amdgpu_gem_mmap_ioctl(struct drm_device *dev, void *data,
                struct drm_file *filp);
  int amdgpu_gem_wait_idle_ioctl(struct drm_device *dev, void *data,
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +8 -0
index 71792d8..bae8417 100644
@@ -288,6 +288,69 @@ int amdgpu_gem_create_ioctl(struct drm_device *dev, void *data,
      return 0;
  }
+static int amdgpu_gem_get_handle_from_object(struct drm_file *filp,
+                         struct drm_gem_object *obj)
+{
+    int i;
+    struct drm_gem_object *tmp;
+
+    spin_lock(&filp->table_lock);
+    idr_for_each_entry(&filp->object_idr, tmp, i) {
Quoted 8 lines

drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c int amdgpu_gem_create_ioctl(struct drm_device *dev, void *data, · line 303

Quoted 6 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +6 -0
+        if (obj == tmp) {
+            drm_gem_object_reference(obj);
+            spin_unlock(&filp->table_lock);
+            return i;
+        }
+    }

Please double check if that is still up to date.

Quoted 3 lines

We may have to replace drm_gem_object_reference() with drm_gem_object_get().

On 2nd thought, do we really need to do reference every time?

Yes, that's a must have. Otherwise the handle could be freed and reused already when we return.

Quoted 1 line

if UMD find the same gem object for 3 times, it also need to explicitly free(put) that object for 3 times?

Correct yes. Thinking more about this the real problem is to translate the handle into a structure in libdrm.

Here we are back to the problem Marek and Michel has been working on for a while that we always need to be able to translate a handle into a bo structure.....

Quoted 8 lines
Quoted 4 lines
Quoted 1 line

So that needs to be solved before we can upstream the changes.

Thanks for your info.
It's better to fix that before upstream.

Thinking more about this the hash currently used in libdrm is not adequate any more.

E.g. we now need to be able to find all BOs based on their handle. Since the handles are dense either an r/b tree or a radix tree now sounds like the best approach to me.

Not sure the exact reason that we added hash table in libdrm.
But it really costs much less time than calling IOCTL to find BO by their handles.

In this case, UMD seems not to be able to get BO handle and try to verify it by cpu address then. In another word, UMD would like to find if the memory is created as BO or system memory, I suppose.

Regards,
Jerry

Quoted 7 lines

Christian.

Quoted 3 lines

Regards,
Jerry

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

#21 lore raw
In reply to a message not in the archive

Am 31.07.2018 um 10:58 schrieb Zhang, Jerry (Junwei):

Quoted 23 lines · Christian König
Quoted 22 lines

Am 31.07.2018 um 10:05 schrieb Zhang, Jerry (Junwei):

Quoted 21 lines · Christian König
Quoted 19 lines

Am 31.07.2018 um 08:58 schrieb Zhang, Jerry (Junwei):

Quoted 8 lines · Christian König
Quoted 3 lines

Am 30.07.2018 um 12:02 schrieb Junwei Zhang:
[SNIP]
Please double check if that is still up to date.

We may have to replace drm_gem_object_reference() with drm_gem_object_get().

On 2nd thought, do we really need to do reference every time?

Yes, that's a must have. Otherwise the handle could be freed and reused already when we return.

Quoted 1 line

if UMD find the same gem object for 3 times, it also need to explicitly free(put) that object for 3 times?

Correct yes. Thinking more about this the real problem is to translate the handle into a structure in libdrm.

Here we are back to the problem Marek and Michel has been working on for a while that we always need to be able to translate a handle into a bo structure.....

So that needs to be solved before we can upstream the changes.

Quoted 8 lines
Quoted 6 lines
Quoted 2 lines

Thanks for your info.
It's better to fix that before upstream.

Thinking more about this the hash currently used in libdrm is not adequate any more.

E.g. we now need to be able to find all BOs based on their handle. Since the handles are dense either an r/b tree or a radix tree now sounds like the best approach to me.

Not sure the exact reason that we added hash table in libdrm.

The reason for that was that when a kernel function returns a handle we need to make sure that we always use the same struct amdgpu_bo for it.

Otherwise you run into quite some problems with syncing etc...

Quoted 1 line

But it really costs much less time than calling IOCTL to find BO by their handles.

Well we could just completely drop the kernel implementation and use an userspace implementation.

And yes I agree when we need a tree anyway it would probably be faster than calling the IOCTL to find the BO.

Christian.

Quoted 12 lines

In this case, UMD seems not to be able to get BO handle and try to verify it by cpu address then. In another word, UMD would like to find if the memory is created as BO or system memory, I suppose.

Regards,
Jerry

Quoted 4 lines

Christian.

Quoted 1 line
Quoted 7 lines
Quoted 3 lines
Quoted 2 lines

Regards,
Jerry

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

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

#23 lore raw
In reply to a message not in the archive
Quoted 29 lines · Christian König

Am 31.07.2018 um 10:58 schrieb Zhang, Jerry (Junwei):

Quoted 28 lines · Christian König
Quoted 27 lines

Am 31.07.2018 um 10:05 schrieb Zhang, Jerry (Junwei):

Quoted 15 lines · Christian König
Quoted 14 lines

Am 31.07.2018 um 08:58 schrieb Zhang, Jerry (Junwei):

Quoted 8 lines · Christian König
Quoted 3 lines

Am 30.07.2018 um 12:02 schrieb Junwei Zhang:
[SNIP]
Please double check if that is still up to date.

We may have to replace drm_gem_object_reference() with drm_gem_object_get().

On 2nd thought, do we really need to do reference every time?

Yes, that's a must have. Otherwise the handle could be freed and reused already when we return.

Quoted 1 line

if UMD find the same gem object for 3 times, it also need to explicitly free(put) that object for 3 times?

Quoted 8 lines
Quoted 5 lines

Correct yes. Thinking more about this the real problem is to translate the handle into a structure in libdrm.

Here we are back to the problem Marek and Michel has been working on for a while that we always need to be able to translate a handle into a bo structure.....

So that needs to be solved before we can upstream the changes.

Thanks for your info.
It's better to fix that before upstream.

Thinking more about this the hash currently used in libdrm is not adequate any more.

Quoted 7 lines
Quoted 3 lines
Quoted 1 line

E.g. we now need to be able to find all BOs based on their handle. Since the handles are dense either an r/b tree or a radix tree now sounds like the best approach to me.

Not sure the exact reason that we added hash table in libdrm.

The reason for that was that when a kernel function returns a handle we need to make sure that we always use the same struct amdgpu_bo for it.

Otherwise you run into quite some problems with syncing etc...

Thanks for your explanation.

Quoted 4 lines
Quoted 2 lines

But it really costs much less time than calling IOCTL to find BO by their handles.

Well we could just completely drop the kernel implementation and use an userspace implementation.

Do you mean to implement finding bo by cpu address in libdrm completely? e.g. to create a tree to manage bo handle in libdrm?

Jerry

Quoted 25 lines

And yes I agree when we need a tree anyway it would probably be faster than calling the IOCTL to find the BO.

Christian.

Quoted 19 lines

In this case, UMD seems not to be able to get BO handle and try to verify it by cpu address then. In another word, UMD would like to find if the memory is created as BO or system memory, I suppose.

Regards,
Jerry

Quoted 7 lines

Christian.

Quoted 3 lines

Regards,
Jerry

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

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

#25 lore raw
In reply to a message not in the archive

Am 31.07.2018 um 11:54 schrieb Zhang, Jerry (Junwei):

Quoted 38 lines · Christian König
Quoted 29 lines

Am 31.07.2018 um 10:58 schrieb Zhang, Jerry (Junwei):

Quoted 28 lines · Christian König
Quoted 27 lines

Am 31.07.2018 um 10:05 schrieb Zhang, Jerry (Junwei):

Quoted 15 lines · Christian König
Quoted 14 lines

Am 31.07.2018 um 08:58 schrieb Zhang, Jerry (Junwei):

Quoted 8 lines · Christian König
Quoted 3 lines

Am 30.07.2018 um 12:02 schrieb Junwei Zhang:
[SNIP]
Please double check if that is still up to date.

We may have to replace drm_gem_object_reference() with drm_gem_object_get().

On 2nd thought, do we really need to do reference every time?

Yes, that's a must have. Otherwise the handle could be freed and reused already when we return.

Quoted 1 line

if UMD find the same gem object for 3 times, it also need to explicitly free(put) that object for 3 times?

Quoted 8 lines
Quoted 5 lines

Correct yes. Thinking more about this the real problem is to translate the handle into a structure in libdrm.

Here we are back to the problem Marek and Michel has been working on for a while that we always need to be able to translate a handle into a bo structure.....

So that needs to be solved before we can upstream the changes.

Thanks for your info.
It's better to fix that before upstream.

Thinking more about this the hash currently used in libdrm is not adequate any more.

Quoted 7 lines
Quoted 3 lines
Quoted 1 line

E.g. we now need to be able to find all BOs based on their handle. Since the handles are dense either an r/b tree or a radix tree now sounds like the best approach to me.

Not sure the exact reason that we added hash table in libdrm.

The reason for that was that when a kernel function returns a handle we need to make sure that we always use the same struct amdgpu_bo for it.

Otherwise you run into quite some problems with syncing etc...

Quoted 8 lines

Thanks for your explanation.

Quoted 4 lines
Quoted 2 lines

But it really costs much less time than calling IOCTL to find BO by their handles.

Well we could just completely drop the kernel implementation and use an userspace implementation.

Do you mean to implement finding bo by cpu address in libdrm completely?

Yes, exactly.

Quoted 1 line

e.g. to create a tree to manage bo handle in libdrm?

I mean when we need to create a tree to map the handle to a BO you could also create a tree to map the CPU pointer to the BO directly and avoid the IOCTL overhead completely.

Christian.

Quoted 28 lines

Jerry

Quoted 25 lines

And yes I agree when we need a tree anyway it would probably be faster than calling the IOCTL to find the BO.

Christian.

Quoted 19 lines

In this case, UMD seems not to be able to get BO handle and try to verify it by cpu address then. In another word, UMD would like to find if the memory is created as BO or system memory, I suppose.

Regards,
Jerry

Quoted 7 lines

Christian.

Quoted 3 lines

Regards,
Jerry

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

Quoted 4 lines

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

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

#27 lore raw
In reply to a message not in the archive

Can this be implemented as a wrapper on top of libdrm? So that the tree (or hash table) isn't created for UMDs that don't need it.

Marek

Quoted 130 lines · Christian König

Am 31.07.2018 um 11:54 schrieb Zhang, Jerry (Junwei):

Quoted 71 lines · Christian König
Quoted 53 lines

Am 31.07.2018 um 10:58 schrieb Zhang, Jerry (Junwei):

Quoted 51 lines · Christian König
Quoted 47 lines

Am 31.07.2018 um 10:05 schrieb Zhang, Jerry (Junwei):

Quoted 32 lines · Christian König
Quoted 30 lines

Am 31.07.2018 um 08:58 schrieb Zhang, Jerry (Junwei):

Quoted 8 lines · Christian König
Quoted 4 lines

Am 30.07.2018 um 12:02 schrieb Junwei Zhang:
[SNIP]
Please double check if that is still up to date.

Quoted 4 lines

We may have to replace drm_gem_object_reference() with drm_gem_object_get().

On 2nd thought, do we really need to do reference every time?

Yes, that's a must have. Otherwise the handle could be freed and reused already when we return.

Quoted 2 lines

if UMD find the same gem object for 3 times, it also need to explicitly free(put) that object for 3 times?

Correct yes. Thinking more about this the real problem is to translate the handle into a structure in libdrm.

Here we are back to the problem Marek and Michel has been working on for a while that we always need to be able to translate a handle into a bo structure.....

Quoted 5 lines
Quoted 1 line

So that needs to be solved before we can upstream the changes.

Thanks for your info.
It's better to fix that before upstream.

Thinking more about this the hash currently used in libdrm is not adequate any more.

E.g. we now need to be able to find all BOs based on their handle. Since the handles are dense either an r/b tree or a radix tree now sounds like the best approach to me.

Quoted 7 lines
Quoted 1 line

Not sure the exact reason that we added hash table in libdrm.

The reason for that was that when a kernel function returns a handle we need to make sure that we always use the same struct amdgpu_bo for it.

Otherwise you run into quite some problems with syncing etc...

Thanks for your explanation.

Quoted 5 lines
Quoted 3 lines

But it really costs much less time than calling IOCTL to find BO by their handles.

Quoted 5 lines
Quoted 2 lines

Well we could just completely drop the kernel implementation and use an userspace implementation.

Do you mean to implement finding bo by cpu address in libdrm completely?

Yes, exactly.

Quoted 1 line

e.g. to create a tree to manage bo handle in libdrm?

I mean when we need to create a tree to map the handle to a BO you could also create a tree to map the CPU pointer to the BO directly and avoid the IOCTL overhead completely.

Christian.

Quoted 33 lines

Jerry

Quoted 30 lines

And yes I agree when we need a tree anyway it would probably be faster than calling the IOCTL to find the BO.

Christian.

Quoted 22 lines

In this case, UMD seems not to be able to get BO handle and try to verify it by cpu address then.
In another word, UMD would like to find if the memory is created as BO or system memory, I suppose.

Regards,
Jerry

Quoted 8 lines

Christian.

Quoted 3 lines

Regards,
Jerry

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

Quoted 4 lines

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

Quoted 4 lines

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

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

#29 lore raw
In reply to a message not in the archive

Am 01.08.2018 um 00:07 schrieb Marek Olšák:

Quoted 2 lines

Can this be implemented as a wrapper on top of libdrm? So that the tree (or hash table) isn't created for UMDs that don't need it.

No, the problem is that an application gets a CPU pointer from one API and tries to import that pointer into another one.

In other words we need to implement this independent of the UMD who mapped the BO.

Christian.

Quoted 3 lines

Marek

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

#31 lore raw
In reply to a message not in the archive
Quoted 11 lines · Christian König

Am 01.08.2018 um 00:07 schrieb Marek Olšák:

Quoted 3 lines

Can this be implemented as a wrapper on top of libdrm? So that the tree (or hash table) isn't created for UMDs that don't need it.

No, the problem is that an application gets a CPU pointer from one API and tries to import that pointer into another one.

In other words we need to implement this independent of the UMD who mapped the BO.

Yeah, it could be an optional feature of libdrm, and other components should be able to disable it to remove the overhead.

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

#33 lore raw
In reply to a message not in the archive

Am 01.08.2018 um 19:39 schrieb Marek Olšák:

Quoted 6 lines
Quoted 4 lines · Christian König

Am 01.08.2018 um 00:07 schrieb Marek Olšák:

Quoted 2 lines

Can this be implemented as a wrapper on top of libdrm? So that the tree (or hash table) isn't created for UMDs that don't need it.

Quoted 7 lines
Quoted 5 lines

No, the problem is that an application gets a CPU pointer from one API and tries to import that pointer into another one.

In other words we need to implement this independent of the UMD who mapped the BO.

Yeah, it could be an optional feature of libdrm, and other components should be able to disable it to remove the overhead.

The overhead is negligible, the real problem is the memory footprint.

A brief look at the hash implementation in libdrm showed that this is actually really inefficient.

I think we have the choice of implementing a r/b tree to map the CPU pointer addresses or implement a quadratic tree to map the handles.

The later is easy to do and would also allow to get rid of the hash table as well.

Christian.

Quoted 2 lines

Marek

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

#35 lore raw
In reply to a message not in the archive
Quoted 25 lines · Christian König

Am 01.08.2018 um 19:39 schrieb Marek Olšák:

Quoted 14 lines
Quoted 11 lines · Christian König

Am 01.08.2018 um 00:07 schrieb Marek Olšák:

Quoted 3 lines

Can this be implemented as a wrapper on top of libdrm? So that the tree (or hash table) isn't created for UMDs that don't need it.

No, the problem is that an application gets a CPU pointer from one API and
tries to import that pointer into another one.

Quoted 6 lines
Quoted 3 lines

In other words we need to implement this independent of the UMD who mapped
the BO.

Yeah, it could be an optional feature of libdrm, and other components should be able to disable it to remove the overhead.

The overhead is negligible, the real problem is the memory footprint.

Quoted 8 lines

A brief look at the hash implementation in libdrm showed that this is actually really inefficient.

I think we have the choice of implementing a r/b tree to map the CPU pointer addresses or implement a quadratic tree to map the handles.

The later is easy to do and would also allow to get rid of the hash table as well.

We can also use the hash table from mesa/src/util.

I don't think the overhead would be negligible. It would be a log(n) insertion in bo_map and a log(n) deletion in bo_unmap. If you did bo_map+bo_unmap 10000 times, would it be negligible?

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

#37 lore raw
In reply to a message not in the archive

Am 01.08.2018 um 19:59 schrieb Marek Olšák:

Quoted 27 lines
Quoted 25 lines · Christian König

Am 01.08.2018 um 19:39 schrieb Marek Olšák:

Quoted 10 lines
Quoted 8 lines · Christian König

Am 01.08.2018 um 00:07 schrieb Marek Olšák:

Quoted 2 lines

Can this be implemented as a wrapper on top of libdrm? So that the tree (or hash table) isn't created for UMDs that don't need it.

No, the problem is that an application gets a CPU pointer from one API and
tries to import that pointer into another one.

Quoted 5 lines
Quoted 3 lines

In other words we need to implement this independent of the UMD who mapped
the BO.

Yeah, it could be an optional feature of libdrm, and other components should be able to disable it to remove the overhead.

The overhead is negligible, the real problem is the memory footprint.

A brief look at the hash implementation in libdrm showed that this is actually really inefficient.

I think we have the choice of implementing a r/b tree to map the CPU pointer addresses or implement a quadratic tree to map the handles.

Quoted 7 lines
Quoted 2 lines

The later is easy to do and would also allow to get rid of the hash table as well.

We can also use the hash table from mesa/src/util.

I don't think the overhead would be negligible. It would be a log(n) insertion in bo_map and a log(n) deletion in bo_unmap. If you did bo_map+bo_unmap 10000 times, would it be negligible?

Compared to what the kernel needs to do for updating the page tables it is less than 1% of the total work.

The real question is if it wouldn't be simpler to use a tree for the handles. Since the handles are dense you can just use an unbalanced tree which is really easy.

For a tree of the CPU mappings we would need an r/b interval tree, which is hard to implement and quite some overkill.

Do you have any numbers how many BOs really get a CPU mapping in a real world application?

Christian.

Quoted 2 lines

Marek

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

#39 lore raw
In reply to a message not in the archive
Quoted 55 lines · Christian König

Am 01.08.2018 um 19:59 schrieb Marek Olšák:

Quoted 40 lines
Quoted 29 lines · Christian König

Am 01.08.2018 um 19:39 schrieb Marek Olšák:

Quoted 14 lines
Quoted 11 lines · Christian König

Am 01.08.2018 um 00:07 schrieb Marek Olšák:

Quoted 3 lines

Can this be implemented as a wrapper on top of libdrm? So that the tree (or hash table) isn't created for UMDs that don't need it.

No, the problem is that an application gets a CPU pointer from one API and
tries to import that pointer into another one.

Quoted 6 lines
Quoted 3 lines

In other words we need to implement this independent of the UMD who mapped
the BO.

Yeah, it could be an optional feature of libdrm, and other components should be able to disable it to remove the overhead.

The overhead is negligible, the real problem is the memory footprint.

A brief look at the hash implementation in libdrm showed that this is actually really inefficient.

Quoted 7 lines

I think we have the choice of implementing a r/b tree to map the CPU pointer
addresses or implement a quadratic tree to map the handles.

The later is easy to do and would also allow to get rid of the hash table as
well.

Quoted 5 lines

We can also use the hash table from mesa/src/util.

I don't think the overhead would be negligible. It would be a log(n) insertion in bo_map and a log(n) deletion in bo_unmap. If you did bo_map+bo_unmap 10000 times, would it be negligible?

Compared to what the kernel needs to do for updating the page tables it is less than 1% of the total work.

The real question is if it wouldn't be simpler to use a tree for the handles. Since the handles are dense you can just use an unbalanced tree which is really easy.

Quoted 5 lines

For a tree of the CPU mappings we would need an r/b interval tree, which is hard to implement and quite some overkill.

Do you have any numbers how many BOs really get a CPU mapping in a real world application?

Without our suballocator, we sometimes exceeded the max. mmap limit (~64K). It should be much less with the suballocator with 128KB slabs, probably a few thousands.

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

#41 lore raw
In reply to a message not in the archive
Quoted 60 lines · Marek Olšák
Quoted 58 lines · Christian König

Am 01.08.2018 um 19:59 schrieb Marek Olšák:

Quoted 40 lines
Quoted 29 lines · Christian König

Am 01.08.2018 um 19:39 schrieb Marek Olšák:

Quoted 14 lines
Quoted 11 lines · Christian König

Am 01.08.2018 um 00:07 schrieb Marek Olšák:

Quoted 3 lines

Can this be implemented as a wrapper on top of libdrm? So that the tree (or hash table) isn't created for UMDs that don't need it.

No, the problem is that an application gets a CPU pointer from one API and
tries to import that pointer into another one.

Quoted 6 lines
Quoted 3 lines

In other words we need to implement this independent of the UMD who mapped
the BO.

Yeah, it could be an optional feature of libdrm, and other components should be able to disable it to remove the overhead.

The overhead is negligible, the real problem is the memory footprint.

A brief look at the hash implementation in libdrm showed that this is actually really inefficient.

Quoted 7 lines

I think we have the choice of implementing a r/b tree to map the CPU pointer
addresses or implement a quadratic tree to map the handles.

The later is easy to do and would also allow to get rid of the hash table as
well.

Quoted 5 lines

We can also use the hash table from mesa/src/util.

I don't think the overhead would be negligible. It would be a log(n) insertion in bo_map and a log(n) deletion in bo_unmap. If you did bo_map+bo_unmap 10000 times, would it be negligible?

Compared to what the kernel needs to do for updating the page tables it is less than 1% of the total work.

The real question is if it wouldn't be simpler to use a tree for the handles. Since the handles are dense you can just use an unbalanced tree which is really easy.

For a tree of the CPU mappings we would need an r/b interval tree, which is hard to implement and quite some overkill.

Quoted 6 lines
Quoted 2 lines

Do you have any numbers how many BOs really get a CPU mapping in a real world application?

Without our suballocator, we sometimes exceeded the max. mmap limit (~64K). It should be much less with the suballocator with 128KB slabs, probably a few thousands.

Is there a way to verify that it has performance issue if moving the cpu mapping in libdrm hash table from kernel side?

AFAIW, only one game will use that with close OGL.

Regards,
Jerry

Quoted 3 lines

Marek

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

#43 lore raw
In reply to a message not in the archive

Am 02.08.2018 um 08:20 schrieb Zhang, Jerry (Junwei):

Quoted 19 lines · Marek Olšák
Quoted 18 lines
Quoted 15 lines · Christian König

Am 01.08.2018 um 19:59 schrieb Marek Olšák:

Quoted 1 line

[SNIP]

Compared to what the kernel needs to do for updating the page tables it is less than 1% of the total work.

The real question is if it wouldn't be simpler to use a tree for the handles. Since the handles are dense you can just use an unbalanced tree which is really easy.

For a tree of the CPU mappings we would need an r/b interval tree, which is hard to implement and quite some overkill.

Do you have any numbers how many BOs really get a CPU mapping in a real world application?

Quoted 6 lines
Quoted 3 lines

Without our suballocator, we sometimes exceeded the max. mmap limit (~64K). It should be much less with the suballocator with 128KB slabs, probably a few thousands.

Is there a way to verify that it has performance issue if moving the cpu mapping in libdrm hash table from kernel side?

Going to take a look at this, but I have the feeling that the CPU overhead we need to add in userspace isn't that much.

Quoted 2 lines

AFAIW, only one game will use that with close OGL.

Well, I would also say that this is a bug in the game and should be fixed there.

Christian.

Quoted 7 lines

Regards,
Jerry

Quoted 3 lines

Marek

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