[PATCH v2 0/3] Add GPU page fault query interface

[PATCH v2 0/3] Add GPU page fault query interface

[Alex Deucher]

This patch set adds support for an application to query GPU
page faults.  It's useful for debugging and there are
vulkan extensions that could make use of this.  Preliminary
user space code which uses this can be found here:
https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23238
https://gitlab.freedesktop.org/mesa/drm/-/merge_requests/298

Note, that I made a small change to the vmhub definition to
decouple it from how the kernel tracks vmhubs so that we have
a consistent user view even if we decide to add more vmhubs
like we recently did for gfx 9.4.3.

I've also pushed the changed to:
https://gitlab.freedesktop.org/agd5f/linux/-/commits/gpu_fault_info_ioctl

Open question, currently we just expose the raw GPU fault status
register value for each GPU so UMDs need GPU specific knowlege to decode
it, although it's largely the same across generations.  One option would be to
translate to a generic GPU independent fault status.  Opinions?

v2:
- Fix spelling typos noted by Guchun

Alex Deucher (3):
  drm/amdgpu: add cached GPU fault structure to vm struct
  drm/amdgpu: cache gpuvm fault information for gmc7+
  drm/amdgpu: add new INFO ioctl query for the last GPU page fault

 drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c |  3 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c | 16 +++++++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c  | 45 +++++++++++++++++++++++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h  | 31 +++++++++++++++--
 drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c  |  3 ++
 drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c  |  3 ++
 drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c   |  3 ++
 drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c   |  3 ++
 drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c   | 11 +++---
 include/uapi/drm/amdgpu_drm.h           | 16 +++++++++
 10 files changed, 126 insertions(+), 8 deletions(-)

-- 
2.40.1

[PATCH 1/3] drm/amdgpu: add cached GPU fault structure to vm struct

[Alex Deucher]

When we get a GPU page fault, cache the fault for later
analysis.

Cc: samuel.pitoiset@gmail.com
Acked-by: Guchun Chen <guchun.chen@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 31 ++++++++++++++++++++++++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 18 +++++++++++++++
 2 files changed, 49 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index dc80c9c8fd14..5423f66a9ed8 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -2640,3 +2640,34 @@ void amdgpu_debugfs_vm_bo_info(struct amdgpu_vm *vm, struct seq_file *m)
 		   total_done_objs);
 }
 #endif
+
+/**
+ * amdgpu_vm_update_fault_cache - update cached fault into.
+ * @adev: amdgpu device pointer
+ * @pasid: PASID of the VM
+ * @addr: Address of the fault
+ * @status: GPUVM fault status register
+ * @vmhub: which vmhub got the fault
+ *
+ * Cache the fault info for later use by userspace in debugging.
+ */
+void amdgpu_vm_update_fault_cache(struct amdgpu_device *adev,
+				  unsigned int pasid,
+				  uint64_t addr,
+				  uint32_t status,
+				  unsigned int vmhub)
+{
+	struct amdgpu_vm *vm;
+	unsigned long flags;
+
+	xa_lock_irqsave(&adev->vm_manager.pasids, flags);
+
+	vm = xa_load(&adev->vm_manager.pasids, pasid);
+	if (vm) {
+		vm->fault_info.addr = addr;
+		vm->fault_info.status = status;
+		vm->fault_info.vmhub = vmhub;
+	}
+	xa_unlock_irqrestore(&adev->vm_manager.pasids, flags);
+}
+
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
index 14f9a2bf3acb..fb66a413110c 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
@@ -244,6 +244,15 @@ struct amdgpu_vm_update_funcs {
 		      struct dma_fence **fence);
 };
 
+struct amdgpu_vm_fault_info {
+	/* fault address */
+	uint64_t	addr;
+	/* fault status register */
+	uint32_t	status;
+	/* which vmhub? gfxhub, mmhub, etc. */
+	unsigned int	vmhub;
+};
+
 struct amdgpu_vm {
 	/* tree of virtual addresses mapped */
 	struct rb_root_cached	va;
@@ -332,6 +341,9 @@ struct amdgpu_vm {
 
 	/* Memory partition number, -1 means any partition */
 	int8_t			mem_id;
+
+	/* cached fault info */
+	struct amdgpu_vm_fault_info fault_info;
 };
 
 struct amdgpu_vm_manager {
@@ -540,4 +552,10 @@ static inline void amdgpu_vm_eviction_unlock(struct amdgpu_vm *vm)
 	mutex_unlock(&vm->eviction_lock);
 }
 
+void amdgpu_vm_update_fault_cache(struct amdgpu_device *adev,
+				  unsigned int pasid,
+				  uint64_t addr,
+				  uint32_t status,
+				  unsigned int vmhub);
+
 #endif
-- 
2.40.1

[PATCH 2/3] drm/amdgpu: cache gpuvm fault information for gmc7+

[Alex Deucher]

Cache the current fault info in the vm struct.  This can be queried
by userspace later to help debug UMDs.

Cc: samuel.pitoiset@gmail.com
Acked-by: Guchun Chen <guchun.chen@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c |  3 +++
 drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c |  3 +++
 drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c  |  3 +++
 drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c  |  3 +++
 drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c  | 11 +++++++----
 5 files changed, 19 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c
index b2e42f1b0f12..ccb69f5b06fe 100644
--- a/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c
@@ -155,6 +155,9 @@ static int gmc_v10_0_process_interrupt(struct amdgpu_device *adev,
 
 		status = RREG32(hub->vm_l2_pro_fault_status);
 		WREG32_P(hub->vm_l2_pro_fault_cntl, 1, ~1);
+
+		amdgpu_vm_update_fault_cache(adev, entry->pasid, addr, status,
+					     entry->vmid_src ? AMDGPU_MMHUB0(0) : AMDGPU_GFXHUB(0));
 	}
 
 	if (!printk_ratelimit())
diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c
index c571f0d95994..ae35dc6ba502 100644
--- a/drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c
@@ -115,6 +115,9 @@ static int gmc_v11_0_process_interrupt(struct amdgpu_device *adev,
 
 		status = RREG32(hub->vm_l2_pro_fault_status);
 		WREG32_P(hub->vm_l2_pro_fault_cntl, 1, ~1);
+
+		amdgpu_vm_update_fault_cache(adev, entry->pasid, addr, status,
+					     entry->vmid_src ? AMDGPU_MMHUB0(0) : AMDGPU_GFXHUB(0));
 	}
 
 	if (printk_ratelimit()) {
diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c
index acd2b407860f..d51cad788769 100644
--- a/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c
@@ -1273,6 +1273,9 @@ static int gmc_v7_0_process_interrupt(struct amdgpu_device *adev,
 	if (!addr && !status)
 		return 0;
 
+	amdgpu_vm_update_fault_cache(adev, entry->pasid,
+				     ((u64)addr) << AMDGPU_GPU_PAGE_SHIFT, status, AMDGPU_GFXHUB(0));
+
 	if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_FIRST)
 		gmc_v7_0_set_fault_enable_default(adev, false);
 
diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c
index 85dead2a5702..8ce7455ff3f6 100644
--- a/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c
@@ -1448,6 +1448,9 @@ static int gmc_v8_0_process_interrupt(struct amdgpu_device *adev,
 	if (!addr && !status)
 		return 0;
 
+	amdgpu_vm_update_fault_cache(adev, entry->pasid,
+				     ((u64)addr) << AMDGPU_GPU_PAGE_SHIFT, status, AMDGPU_GFXHUB(0));
+
 	if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_FIRST)
 		gmc_v8_0_set_fault_enable_default(adev, false);
 
diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
index 3ed286b72cae..fbd65872050a 100644
--- a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
@@ -555,6 +555,7 @@ static int gmc_v9_0_process_interrupt(struct amdgpu_device *adev,
 	struct amdgpu_vmhub *hub;
 	const char *mmhub_cid;
 	const char *hub_name;
+	unsigned int vmhub;
 	u64 addr;
 	uint32_t cam_index = 0;
 	int ret, xcc_id = 0;
@@ -567,10 +568,10 @@ static int gmc_v9_0_process_interrupt(struct amdgpu_device *adev,
 
 	if (entry->client_id == SOC15_IH_CLIENTID_VMC) {
 		hub_name = "mmhub0";
-		hub = &adev->vmhub[AMDGPU_MMHUB0(node_id / 4)];
+		vmhub = AMDGPU_MMHUB0(node_id / 4);
 	} else if (entry->client_id == SOC15_IH_CLIENTID_VMC1) {
 		hub_name = "mmhub1";
-		hub = &adev->vmhub[AMDGPU_MMHUB1(0)];
+		vmhub = AMDGPU_MMHUB1(0);
 	} else {
 		hub_name = "gfxhub0";
 		if (adev->gfx.funcs->ih_node_to_logical_xcc) {
@@ -579,8 +580,9 @@ static int gmc_v9_0_process_interrupt(struct amdgpu_device *adev,
 			if (xcc_id < 0)
 				xcc_id = 0;
 		}
-		hub = &adev->vmhub[xcc_id];
+		vmhub = xcc_id;
 	}
+	hub = &adev->vmhub[vmhub];
 
 	if (retry_fault) {
 		if (adev->irq.retry_cam_enabled) {
@@ -626,7 +628,6 @@ static int gmc_v9_0_process_interrupt(struct amdgpu_device *adev,
 	if (!printk_ratelimit())
 		return 0;
 
-
 	memset(&task_info, 0, sizeof(struct amdgpu_task_info));
 	amdgpu_vm_get_task_info(adev, entry->pasid, &task_info);
 
@@ -663,6 +664,8 @@ static int gmc_v9_0_process_interrupt(struct amdgpu_device *adev,
 	rw = REG_GET_FIELD(status, VM_L2_PROTECTION_FAULT_STATUS, RW);
 	WREG32_P(hub->vm_l2_pro_fault_cntl, 1, ~1);
 
+	amdgpu_vm_update_fault_cache(adev, entry->pasid, addr, status, vmhub);
+
 	dev_err(adev->dev,
 		"VM_L2_PROTECTION_FAULT_STATUS:0x%08X\n",
 		status);
-- 
2.40.1

[PATCH 3/3] drm/amdgpu: add new INFO ioctl query for the last GPU page fault

[Alex Deucher]

Add a interface to query the last GPU page fault for the process.
Useful for debugging context lost errors.

v2: split vmhub representation between kernel and userspace

Mesa MR: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23238
libdrm MR: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23238

Cc: samuel.pitoiset@gmail.com
Acked-by: Guchun Chen <guchun.chen@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c |  3 ++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c | 16 ++++++++++++++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c  | 16 +++++++++++++++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h  | 13 ++++++++++---
 include/uapi/drm/amdgpu_drm.h           | 16 ++++++++++++++++
 5 files changed, 59 insertions(+), 5 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
index 999d008b6b48..6b053bab799c 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
@@ -113,9 +113,10 @@
  *            gl1c_cache_size, gl2c_cache_size, mall_size, enabled_rb_pipes_mask_hi
  *   3.53.0 - Support for GFX11 CP GFX shadowing
  *   3.54.0 - Add AMDGPU_CTX_QUERY2_FLAGS_RESET_IN_PROGRESS support
+ * - 3.55.0 - Add AMDGPU_INFO_GPUVM_FAULT query
  */
 #define KMS_DRIVER_MAJOR	3
-#define KMS_DRIVER_MINOR	54
+#define KMS_DRIVER_MINOR	55
 #define KMS_DRIVER_PATCHLEVEL	0
 
 unsigned int amdgpu_vram_limit = UINT_MAX;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
index e3531aa3c8bd..2289f8eb3d1d 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
@@ -1163,6 +1163,22 @@ int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
 		return copy_to_user(out, max_ibs,
 				    min((size_t)size, sizeof(max_ibs))) ? -EFAULT : 0;
 	}
+	case AMDGPU_INFO_GPUVM_FAULT: {
+		struct amdgpu_fpriv *fpriv = filp->driver_priv;
+		struct amdgpu_vm *vm = &fpriv->vm;
+		struct drm_amdgpu_info_gpuvm_fault gpuvm_fault;
+
+		if (!vm)
+			return -EINVAL;
+
+		memset(&gpuvm_fault, 0, sizeof(gpuvm_fault));
+		gpuvm_fault.addr = vm->fault_info.addr;
+		gpuvm_fault.status = vm->fault_info.status;
+		gpuvm_fault.vmhub = vm->fault_info.vmhub;
+
+		return copy_to_user(out, &gpuvm_fault,
+				    min((size_t)size, sizeof(gpuvm_fault))) ? -EFAULT : 0;
+	}
 	default:
 		DRM_DEBUG_KMS("Invalid request %d\n", info->query);
 		return -EINVAL;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 5423f66a9ed8..2c1106855492 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -2666,7 +2666,21 @@ void amdgpu_vm_update_fault_cache(struct amdgpu_device *adev,
 	if (vm) {
 		vm->fault_info.addr = addr;
 		vm->fault_info.status = status;
-		vm->fault_info.vmhub = vmhub;
+		if (AMDGPU_IS_GFXHUB(vmhub)) {
+			vm->fault_info.vmhub = AMDGPU_VMHUB_TYPE_GFX;
+			vm->fault_info.vmhub |=
+				(vmhub - AMDGPU_GFXHUB_START) << AMDGPU_VMHUB_IDX_SHIFT;
+		} else if (AMDGPU_IS_MMHUB0(vmhub)) {
+			vm->fault_info.vmhub = AMDGPU_VMHUB_TYPE_MM0;
+			vm->fault_info.vmhub |=
+				(vmhub - AMDGPU_MMHUB0_START) << AMDGPU_VMHUB_IDX_SHIFT;
+		} else if (AMDGPU_IS_MMHUB1(vmhub)) {
+			vm->fault_info.vmhub = AMDGPU_VMHUB_TYPE_MM1;
+			vm->fault_info.vmhub |=
+				(vmhub - AMDGPU_MMHUB1_START) << AMDGPU_VMHUB_IDX_SHIFT;
+		} else {
+			WARN_ONCE(1, "Invalid vmhub %u\n", vmhub);
+		}
 	}
 	xa_unlock_irqrestore(&adev->vm_manager.pasids, flags);
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
index fb66a413110c..1a34fea9acb9 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
@@ -116,9 +116,16 @@ struct amdgpu_mem_stats;
  * layout: max 8 GFXHUB + 4 MMHUB0 + 1 MMHUB1
  */
 #define AMDGPU_MAX_VMHUBS			13
-#define AMDGPU_GFXHUB(x)			(x)
-#define AMDGPU_MMHUB0(x)			(8 + x)
-#define AMDGPU_MMHUB1(x)			(8 + 4 + x)
+#define AMDGPU_GFXHUB_START			0
+#define AMDGPU_MMHUB0_START			8
+#define AMDGPU_MMHUB1_START			12
+#define AMDGPU_GFXHUB(x)			(AMDGPU_GFXHUB_START + (x))
+#define AMDGPU_MMHUB0(x)			(AMDGPU_MMHUB0_START + (x))
+#define AMDGPU_MMHUB1(x)			(AMDGPU_MMHUB1_START + (x))
+
+#define AMDGPU_IS_GFXHUB(x) ((x) >= AMDGPU_GFXHUB_START && (x) < AMDGPU_MMHUB0_START)
+#define AMDGPU_IS_MMHUB0(x) ((x) >= AMDGPU_MMHUB0_START && (x) < AMDGPU_MMHUB1_START)
+#define AMDGPU_IS_MMHUB1(x) ((x) >= AMDGPU_MMHUB1_START && (x) < AMDGPU_MAX_VMHUBS)
 
 /* Reserve 2MB at top/bottom of address space for kernel use */
 #define AMDGPU_VA_RESERVED_SIZE			(2ULL << 20)
diff --git a/include/uapi/drm/amdgpu_drm.h b/include/uapi/drm/amdgpu_drm.h
index 79b14828d542..bbe3ae6dc546 100644
--- a/include/uapi/drm/amdgpu_drm.h
+++ b/include/uapi/drm/amdgpu_drm.h
@@ -890,6 +890,8 @@ struct drm_amdgpu_cs_chunk_cp_gfx_shadow {
 	#define AMDGPU_INFO_VIDEO_CAPS_ENCODE		1
 /* Query the max number of IBs per gang per submission */
 #define AMDGPU_INFO_MAX_IBS			0x22
+/* query last page fault info */
+#define AMDGPU_INFO_GPUVM_FAULT			0x23
 
 #define AMDGPU_INFO_MMR_SE_INDEX_SHIFT	0
 #define AMDGPU_INFO_MMR_SE_INDEX_MASK	0xff
@@ -1215,6 +1217,20 @@ struct drm_amdgpu_info_video_caps {
 	struct drm_amdgpu_info_video_codec_info codec_info[AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_COUNT];
 };
 
+#define AMDGPU_VMHUB_TYPE_MASK			0xff
+#define AMDGPU_VMHUB_TYPE_SHIFT			0
+#define AMDGPU_VMHUB_TYPE_GFX			0
+#define AMDGPU_VMHUB_TYPE_MM0			1
+#define AMDGPU_VMHUB_TYPE_MM1			2
+#define AMDGPU_VMHUB_IDX_MASK			0xff00
+#define AMDGPU_VMHUB_IDX_SHIFT			8
+
+struct drm_amdgpu_info_gpuvm_fault {
+	__u64 addr;
+	__u32 status;
+	__u32 vmhub;
+};
+
 /*
  * Supported GPU families
  */
-- 
2.40.1

[PATCH 0/3] Add GPU page fault query interface

[Alex Deucher]

This patch set adds support for an application to query GPU
page faults.  It's useful for debugging and there are
vulkan extensions that could make use of this.  Preliminary
user space code which uses this can be found here:
https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23238
https://gitlab.freedesktop.org/mesa/drm/-/merge_requests/298

Note, that I made a small change to the vmhub definition to
decouple it from how the kernel tracks vmhubs so that we have
a consistent user view even if we decide to add more vmhubs
like we recently did for gfx 9.4.3.

I've also pushed the changed to:
https://gitlab.freedesktop.org/agd5f/linux/-/commits/gpu_fault_info_ioctl


Alex Deucher (3):
  drm/amdgpu: add cached GPU fault structure to vm struct
  drm/amdgpu: cache gpuvm fault information for gmc7+
  drm/amdgpu: add new INFO ioctl query for the last GPU page fault

 drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c |  3 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c | 16 +++++++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c  | 45 +++++++++++++++++++++++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h  | 31 +++++++++++++++--
 drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c  |  3 ++
 drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c  |  3 ++
 drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c   |  3 ++
 drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c   |  3 ++
 drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c   | 11 +++---
 include/uapi/drm/amdgpu_drm.h           | 16 +++++++++
 10 files changed, 126 insertions(+), 8 deletions(-)

-- 
2.40.1

[PATCH 3/3] drm/amdgpu: add new INFO ioctl query for the last GPU page fault

[Alex Deucher]

Add a interface to query the last GPU page fault for the process.
Useful for debugging context lost errors.

v2: split vmhub representation between kernel and userspace

Mesa MR: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23238
libdrm MR: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23238

Cc: samuel.pitoiset@gmail.com
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c |  3 ++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c | 16 ++++++++++++++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c  | 16 +++++++++++++++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h  | 13 ++++++++++---
 include/uapi/drm/amdgpu_drm.h           | 16 ++++++++++++++++
 5 files changed, 59 insertions(+), 5 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
index 7300df2a342c..7e17b285decc 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
@@ -112,9 +112,10 @@
  *            gl1c_cache_size, gl2c_cache_size, mall_size, enabled_rb_pipes_mask_hi
  *   3.53.0 - Support for GFX11 CP GFX shadowing
  *   3.54.0 - Add AMDGPU_CTX_QUERY2_FLAGS_RESET_IN_PROGRESS support
+ * - 3.55.0 - Add AMDGPU_INFO_GPUVM_FAULT query
  */
 #define KMS_DRIVER_MAJOR	3
-#define KMS_DRIVER_MINOR	54
+#define KMS_DRIVER_MINOR	55
 #define KMS_DRIVER_PATCHLEVEL	0
 
 unsigned int amdgpu_vram_limit = UINT_MAX;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
index 41d047e5de69..bca2a56046ae 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
@@ -1163,6 +1163,22 @@ int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
 		return copy_to_user(out, max_ibs,
 				    min((size_t)size, sizeof(max_ibs))) ? -EFAULT : 0;
 	}
+	case AMDGPU_INFO_GPUVM_FAULT: {
+		struct amdgpu_fpriv *fpriv = filp->driver_priv;
+		struct amdgpu_vm *vm = &fpriv->vm;
+		struct drm_amdgpu_info_gpuvm_fault gpuvm_fault;
+
+		if (!vm)
+			return -EINVAL;
+
+		memset(&gpuvm_fault, 0, sizeof(gpuvm_fault));
+		gpuvm_fault.addr = vm->fault_info.addr;
+		gpuvm_fault.status = vm->fault_info.status;
+		gpuvm_fault.vmhub = vm->fault_info.vmhub;
+
+		return copy_to_user(out, &gpuvm_fault,
+				    min((size_t)size, sizeof(gpuvm_fault))) ? -EFAULT : 0;
+	}
 	default:
 		DRM_DEBUG_KMS("Invalid request %d\n", info->query);
 		return -EINVAL;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 73e022f3daa4..c1b0c5f3c1f8 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -2657,7 +2657,21 @@ void amdgpu_vm_update_fault_cache(struct amdgpu_device *adev,
 	if (vm) {
 		vm->fault_info.addr = addr;
 		vm->fault_info.status = status;
-		vm->fault_info.vmhub = vmhub;
+		if (AMDGPU_IS_GFXHUB(vmhub)) {
+			vm->fault_info.vmhub = AMDGPU_VMHUB_TYPE_GFX;
+			vm->fault_info.vmhub |=
+				(vmhub - AMDGPU_GFXHUB_START) << AMDGPU_VMHUB_IDX_SHIFT;
+		} else if (AMDGPU_IS_MMHUB0(vmhub)) {
+			vm->fault_info.vmhub = AMDGPU_VMHUB_TYPE_MM0;
+			vm->fault_info.vmhub |=
+				(vmhub - AMDGPU_MMHUB0_START) << AMDGPU_VMHUB_IDX_SHIFT;
+		} else if (AMDGPU_IS_MMHUB1(vmhub)) {
+			vm->fault_info.vmhub = AMDGPU_VMHUB_TYPE_MM1;
+			vm->fault_info.vmhub |=
+				(vmhub - AMDGPU_MMHUB1_START) << AMDGPU_VMHUB_IDX_SHIFT;
+		} else {
+			WARN_ONCE(1, "Invalid vmhub %u\n", vmhub);
+		}
 	}
 	xa_unlock_irqrestore(&adev->vm_manager.pasids, flags);
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
index fb66a413110c..1a34fea9acb9 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
@@ -116,9 +116,16 @@ struct amdgpu_mem_stats;
  * layout: max 8 GFXHUB + 4 MMHUB0 + 1 MMHUB1
  */
 #define AMDGPU_MAX_VMHUBS			13
-#define AMDGPU_GFXHUB(x)			(x)
-#define AMDGPU_MMHUB0(x)			(8 + x)
-#define AMDGPU_MMHUB1(x)			(8 + 4 + x)
+#define AMDGPU_GFXHUB_START			0
+#define AMDGPU_MMHUB0_START			8
+#define AMDGPU_MMHUB1_START			12
+#define AMDGPU_GFXHUB(x)			(AMDGPU_GFXHUB_START + (x))
+#define AMDGPU_MMHUB0(x)			(AMDGPU_MMHUB0_START + (x))
+#define AMDGPU_MMHUB1(x)			(AMDGPU_MMHUB1_START + (x))
+
+#define AMDGPU_IS_GFXHUB(x) ((x) >= AMDGPU_GFXHUB_START && (x) < AMDGPU_MMHUB0_START)
+#define AMDGPU_IS_MMHUB0(x) ((x) >= AMDGPU_MMHUB0_START && (x) < AMDGPU_MMHUB1_START)
+#define AMDGPU_IS_MMHUB1(x) ((x) >= AMDGPU_MMHUB1_START && (x) < AMDGPU_MAX_VMHUBS)
 
 /* Reserve 2MB at top/bottom of address space for kernel use */
 #define AMDGPU_VA_RESERVED_SIZE			(2ULL << 20)
diff --git a/include/uapi/drm/amdgpu_drm.h b/include/uapi/drm/amdgpu_drm.h
index 79b14828d542..bbe3ae6dc546 100644
--- a/include/uapi/drm/amdgpu_drm.h
+++ b/include/uapi/drm/amdgpu_drm.h
@@ -890,6 +890,8 @@ struct drm_amdgpu_cs_chunk_cp_gfx_shadow {
 	#define AMDGPU_INFO_VIDEO_CAPS_ENCODE		1
 /* Query the max number of IBs per gang per submission */
 #define AMDGPU_INFO_MAX_IBS			0x22
+/* query last page fault info */
+#define AMDGPU_INFO_GPUVM_FAULT			0x23
 
 #define AMDGPU_INFO_MMR_SE_INDEX_SHIFT	0
 #define AMDGPU_INFO_MMR_SE_INDEX_MASK	0xff
@@ -1215,6 +1217,20 @@ struct drm_amdgpu_info_video_caps {
 	struct drm_amdgpu_info_video_codec_info codec_info[AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_COUNT];
 };
 
+#define AMDGPU_VMHUB_TYPE_MASK			0xff
+#define AMDGPU_VMHUB_TYPE_SHIFT			0
+#define AMDGPU_VMHUB_TYPE_GFX			0
+#define AMDGPU_VMHUB_TYPE_MM0			1
+#define AMDGPU_VMHUB_TYPE_MM1			2
+#define AMDGPU_VMHUB_IDX_MASK			0xff00
+#define AMDGPU_VMHUB_IDX_SHIFT			8
+
+struct drm_amdgpu_info_gpuvm_fault {
+	__u64 addr;
+	__u32 status;
+	__u32 vmhub;
+};
+
 /*
  * Supported GPU families
  */
-- 
2.40.1

Re: [PATCH 3/3] drm/amdgpu: add new INFO ioctl query for the last GPU page fault

[Christian König]

Am 25.05.23 um 18:52 schrieb Alex Deucher:
> Add a interface to query the last GPU page fault for the process.
> Useful for debugging context lost errors.
>
> v2: split vmhub representation between kernel and userspace
>
> Mesa MR: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23238
> libdrm MR: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23238
>
> Cc: samuel.pitoiset@gmail.com
> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c |  3 ++-
>   drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c | 16 ++++++++++++++++
>   drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c  | 16 +++++++++++++++-
>   drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h  | 13 ++++++++++---
>   include/uapi/drm/amdgpu_drm.h           | 16 ++++++++++++++++
>   5 files changed, 59 insertions(+), 5 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
> index 7300df2a342c..7e17b285decc 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
> @@ -112,9 +112,10 @@
>    *            gl1c_cache_size, gl2c_cache_size, mall_size, enabled_rb_pipes_mask_hi
>    *   3.53.0 - Support for GFX11 CP GFX shadowing
>    *   3.54.0 - Add AMDGPU_CTX_QUERY2_FLAGS_RESET_IN_PROGRESS support
> + * - 3.55.0 - Add AMDGPU_INFO_GPUVM_FAULT query
>    */
>   #define KMS_DRIVER_MAJOR	3
> -#define KMS_DRIVER_MINOR	54
> +#define KMS_DRIVER_MINOR	55
>   #define KMS_DRIVER_PATCHLEVEL	0
>   
>   unsigned int amdgpu_vram_limit = UINT_MAX;
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
> index 41d047e5de69..bca2a56046ae 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
> @@ -1163,6 +1163,22 @@ int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
>   		return copy_to_user(out, max_ibs,
>   				    min((size_t)size, sizeof(max_ibs))) ? -EFAULT : 0;
>   	}
> +	case AMDGPU_INFO_GPUVM_FAULT: {
> +		struct amdgpu_fpriv *fpriv = filp->driver_priv;
> +		struct amdgpu_vm *vm = &fpriv->vm;
> +		struct drm_amdgpu_info_gpuvm_fault gpuvm_fault;
> +
> +		if (!vm)
> +			return -EINVAL;
> +
> +		memset(&gpuvm_fault, 0, sizeof(gpuvm_fault));
> +		gpuvm_fault.addr = vm->fault_info.addr;
> +		gpuvm_fault.status = vm->fault_info.status;
> +		gpuvm_fault.vmhub = vm->fault_info.vmhub;

You need something to provide locking and barrier here. I suggest to 
just grab the xa lock.

> +
> +		return copy_to_user(out, &gpuvm_fault,
> +				    min((size_t)size, sizeof(gpuvm_fault))) ? -EFAULT : 0;
> +	}
>   	default:
>   		DRM_DEBUG_KMS("Invalid request %d\n", info->query);
>   		return -EINVAL;
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
> index 73e022f3daa4..c1b0c5f3c1f8 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
> @@ -2657,7 +2657,21 @@ void amdgpu_vm_update_fault_cache(struct amdgpu_device *adev,
>   	if (vm) {
>   		vm->fault_info.addr = addr;
>   		vm->fault_info.status = status;
> -		vm->fault_info.vmhub = vmhub;
> +		if (AMDGPU_IS_GFXHUB(vmhub)) {
> +			vm->fault_info.vmhub = AMDGPU_VMHUB_TYPE_GFX;
> +			vm->fault_info.vmhub |=
> +				(vmhub - AMDGPU_GFXHUB_START) << AMDGPU_VMHUB_IDX_SHIFT;
> +		} else if (AMDGPU_IS_MMHUB0(vmhub)) {
> +			vm->fault_info.vmhub = AMDGPU_VMHUB_TYPE_MM0;
> +			vm->fault_info.vmhub |=
> +				(vmhub - AMDGPU_MMHUB0_START) << AMDGPU_VMHUB_IDX_SHIFT;
> +		} else if (AMDGPU_IS_MMHUB1(vmhub)) {
> +			vm->fault_info.vmhub = AMDGPU_VMHUB_TYPE_MM1;
> +			vm->fault_info.vmhub |=
> +				(vmhub - AMDGPU_MMHUB1_START) << AMDGPU_VMHUB_IDX_SHIFT;
> +		} else {
> +			WARN_ONCE(1, "Invalid vmhub %u\n", vmhub);
> +		}
>   	}
>   	xa_unlock_irqrestore(&adev->vm_manager.pasids, flags);
>   }
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
> index fb66a413110c..1a34fea9acb9 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
> @@ -116,9 +116,16 @@ struct amdgpu_mem_stats;
>    * layout: max 8 GFXHUB + 4 MMHUB0 + 1 MMHUB1
>    */
>   #define AMDGPU_MAX_VMHUBS			13
> -#define AMDGPU_GFXHUB(x)			(x)
> -#define AMDGPU_MMHUB0(x)			(8 + x)
> -#define AMDGPU_MMHUB1(x)			(8 + 4 + x)
> +#define AMDGPU_GFXHUB_START			0
> +#define AMDGPU_MMHUB0_START			8
> +#define AMDGPU_MMHUB1_START			12
> +#define AMDGPU_GFXHUB(x)			(AMDGPU_GFXHUB_START + (x))
> +#define AMDGPU_MMHUB0(x)			(AMDGPU_MMHUB0_START + (x))
> +#define AMDGPU_MMHUB1(x)			(AMDGPU_MMHUB1_START + (x))
> +
> +#define AMDGPU_IS_GFXHUB(x) ((x) >= AMDGPU_GFXHUB_START && (x) < AMDGPU_MMHUB0_START)
> +#define AMDGPU_IS_MMHUB0(x) ((x) >= AMDGPU_MMHUB0_START && (x) < AMDGPU_MMHUB1_START)
> +#define AMDGPU_IS_MMHUB1(x) ((x) >= AMDGPU_MMHUB1_START && (x) < AMDGPU_MAX_VMHUBS)
>   
>   /* Reserve 2MB at top/bottom of address space for kernel use */
>   #define AMDGPU_VA_RESERVED_SIZE			(2ULL << 20)
> diff --git a/include/uapi/drm/amdgpu_drm.h b/include/uapi/drm/amdgpu_drm.h
> index 79b14828d542..bbe3ae6dc546 100644
> --- a/include/uapi/drm/amdgpu_drm.h
> +++ b/include/uapi/drm/amdgpu_drm.h
> @@ -890,6 +890,8 @@ struct drm_amdgpu_cs_chunk_cp_gfx_shadow {
>   	#define AMDGPU_INFO_VIDEO_CAPS_ENCODE		1
>   /* Query the max number of IBs per gang per submission */
>   #define AMDGPU_INFO_MAX_IBS			0x22
> +/* query last page fault info */
> +#define AMDGPU_INFO_GPUVM_FAULT			0x23
>   
>   #define AMDGPU_INFO_MMR_SE_INDEX_SHIFT	0
>   #define AMDGPU_INFO_MMR_SE_INDEX_MASK	0xff
> @@ -1215,6 +1217,20 @@ struct drm_amdgpu_info_video_caps {
>   	struct drm_amdgpu_info_video_codec_info codec_info[AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_COUNT];
>   };
>   
> +#define AMDGPU_VMHUB_TYPE_MASK			0xff
> +#define AMDGPU_VMHUB_TYPE_SHIFT			0
> +#define AMDGPU_VMHUB_TYPE_GFX			0
> +#define AMDGPU_VMHUB_TYPE_MM0			1
> +#define AMDGPU_VMHUB_TYPE_MM1			2
> +#define AMDGPU_VMHUB_IDX_MASK			0xff00
> +#define AMDGPU_VMHUB_IDX_SHIFT			8

IIRC UMR already established the standard to put the hub type in the 
higher bits and the hub number/VMID in the lower ones.

Probably best to sync up with Tom so that both interfaces use the same 
numbering.

Apart from that looks good to me.

Christian.

> +
> +struct drm_amdgpu_info_gpuvm_fault {
> +	__u64 addr;
> +	__u32 status;
> +	__u32 vmhub;
> +};
> +
>   /*
>    * Supported GPU families
>    */