habanalabs: add page fault info uapi

Only the first page fault will be saved.
Besides the address which caused the page fault, the driver captures
all of the mmu user mappings.
User can retrieve this data via the new uapi (new opcode in INFO ioctl).

Signed-off-by: Dani Liberman <dliberman@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>

5 files changed, 124 insertions, 1 deletions

diff --git a/drivers/misc/habanalabs/common/device.c b/drivers/misc/habanalabs/common/device.c
index 30ddaaae67e5..5dc6c77b4721 100644
--- a/drivers/misc/habanalabs/common/device.c
+++ b/drivers/misc/habanalabs/common/device.c
@@ -12,6 +12,7 @@
 
 #include <linux/pci.h>
 #include <linux/hwmon.h>
+#include <linux/vmalloc.h>
 
 #include <trace/events/habanalabs.h>
 
@@ -2199,6 +2200,8 @@ void hl_device_fini(struct hl_device *hdev)
 
 	hl_mmu_fini(hdev);
 
+	vfree(hdev->captured_err_info.pgf_info.user_mappings);
+
 	hl_eq_fini(hdev, &hdev->event_queue);
 
 	kfree(hdev->shadow_cs_queue);
@@ -2275,3 +2278,58 @@ void hl_capture_razwi(struct hl_device *hdev, u64 addr, u16 *engine_id, u16 num_
 			num_of_engines * sizeof(u16));
 	hdev->captured_err_info.razwi.flags = flags;
 }
+static void hl_capture_user_mappings(struct hl_device *hdev)
+{
+	struct page_fault_info *pgf_info = &hdev->captured_err_info.pgf_info;
+	struct hl_vm_hash_node *hnode;
+	struct hl_userptr *userptr;
+	struct hl_ctx *ctx;
+	u32 map_idx = 0;
+	int i;
+
+	ctx = hl_get_compute_ctx(hdev);
+	if (!ctx) {
+		dev_err(hdev->dev, "Can't get user context for user mappings\n");
+		return;
+	}
+
+	mutex_lock(&ctx->mem_hash_lock);
+	hash_for_each(ctx->mem_hash, i, hnode, node)
+	pgf_info->num_of_user_mappings++;
+
+	if (!pgf_info->num_of_user_mappings)
+		goto finish;
+
+	/* In case we already allocated in previous session, need to release it before
+	 * allocating new buffer.
+	 */
+	vfree(pgf_info->user_mappings);
+	pgf_info->user_mappings =
+			vmalloc(pgf_info->num_of_user_mappings * sizeof(struct hl_user_mapping));
+	if (!pgf_info->user_mappings) {
+		pgf_info->num_of_user_mappings = 0;
+		goto finish;
+	}
+
+	hash_for_each(ctx->mem_hash, i, hnode, node) {
+		userptr = hnode->ptr;
+		pgf_info->user_mappings[map_idx].dev_va = hnode->vaddr;
+		pgf_info->user_mappings[map_idx].size = userptr->size;
+		map_idx++;
+	}
+finish:
+	mutex_unlock(&ctx->mem_hash_lock);
+	hl_ctx_put(ctx);
+}
+
+void hl_capture_page_fault(struct hl_device *hdev, u64 addr, u16 eng_id, bool is_pmmu)
+{
+	/* Capture only the first page fault */
+	if (atomic_cmpxchg(&hdev->captured_err_info.pgf_info_recorded, 0, 1))
+		return;
+
+	hdev->captured_err_info.pgf_info.pgf.timestamp = ktime_to_ns(ktime_get());
+	hdev->captured_err_info.pgf_info.pgf.addr = addr;
+	hdev->captured_err_info.pgf_info.pgf.engine_id = eng_id;
+	hl_capture_user_mappings(hdev);
+}
diff --git a/drivers/misc/habanalabs/common/habanalabs.h b/drivers/misc/habanalabs/common/habanalabs.h
index f4b3fa4b0976..1489240d5a3a 100644
--- a/drivers/misc/habanalabs/common/habanalabs.h
+++ b/drivers/misc/habanalabs/common/habanalabs.h
@@ -2958,18 +2958,37 @@ struct undefined_opcode_info {
 };
 
 /**
+ * struct page_fault_info - info about page fault
+ * @pgf_info: page fault information.
+ * @user_mappings: buffer containing user mappings.
+ * @num_of_user_mappings: number of user mappings.
+ */
+struct page_fault_info {
+	struct hl_page_fault_info	pgf;
+	struct hl_user_mapping		*user_mappings;
+	u64				num_of_user_mappings;
+};
+
+/**
  * struct hl_error_info - holds information collected during an error.
  * @cs_timeout: CS timeout error information.
  * @razwi: razwi information.
  * @razwi_info_recorded: if set writing to razwi information is enabled.
- *                otherwise - disabled, so the first (root cause) razwi will not be overwritten.
+ *                       otherwise - disabled, so the first (root cause) razwi will not be
+ *                       overwritten.
  * @undef_opcode: undefined opcode information
+ * @pgf_info: page fault information.
+ * @pgf_info_recorded: if set writing to page fault information is enabled.
+ *                     otherwise - disabled, so the first (root cause) page fault will not be
+ *                     overwritten.
  */
 struct hl_error_info {
 	struct cs_timeout_info		cs_timeout;
 	struct hl_info_razwi_event	razwi;
 	atomic_t			razwi_info_recorded;
 	struct undefined_opcode_info	undef_opcode;
+	struct page_fault_info		pgf_info;
+	atomic_t			pgf_info_recorded;
 };
 
 /**
@@ -3781,6 +3800,7 @@ hl_mmap_mem_buf_alloc(struct hl_mem_mgr *mmg,
 __printf(2, 3) void hl_engine_data_sprintf(struct engines_data *e, const char *fmt, ...);
 void hl_capture_razwi(struct hl_device *hdev, u64 addr, u16 *engine_id, u16 num_of_engines,
 			u8 flags);
+void hl_capture_page_fault(struct hl_device *hdev, u64 addr, u16 eng_id, bool is_pmmu);
 
 #ifdef CONFIG_DEBUG_FS
 
diff --git a/drivers/misc/habanalabs/common/habanalabs_drv.c b/drivers/misc/habanalabs/common/habanalabs_drv.c
index d87434b9bc16..714994725224 100644
--- a/drivers/misc/habanalabs/common/habanalabs_drv.c
+++ b/drivers/misc/habanalabs/common/habanalabs_drv.c
@@ -213,6 +213,7 @@ int hl_device_open(struct inode *inode, struct file *filp)
 
 	atomic_set(&hdev->captured_err_info.cs_timeout.write_enable, 1);
 	atomic_set(&hdev->captured_err_info.razwi_info_recorded, 0);
+	atomic_set(&hdev->captured_err_info.pgf_info_recorded, 0);
 	hdev->captured_err_info.undef_opcode.write_enable = true;
 
 	hdev->open_counter++;
diff --git a/drivers/misc/habanalabs/common/habanalabs_ioctl.c b/drivers/misc/habanalabs/common/habanalabs_ioctl.c
index 6aef4e24d122..cac2c7fb14f1 100644
--- a/drivers/misc/habanalabs/common/habanalabs_ioctl.c
+++ b/drivers/misc/habanalabs/common/habanalabs_ioctl.c
@@ -778,6 +778,42 @@ static int engine_status_info(struct hl_fpriv *hpriv, struct hl_info_args *args)
 	return rc;
 }
 
+static int page_fault_info(struct hl_fpriv *hpriv, struct hl_info_args *args)
+{
+	struct hl_device *hdev = hpriv->hdev;
+	u32 max_size = args->return_size;
+	struct hl_page_fault_info *info = &hdev->captured_err_info.pgf_info.pgf;
+	void __user *out = (void __user *) (uintptr_t) args->return_pointer;
+
+	if ((!max_size) || (!out))
+		return -EINVAL;
+
+	return copy_to_user(out, info, min_t(size_t, max_size, sizeof(struct hl_page_fault_info)))
+				? -EFAULT : 0;
+}
+
+static int user_mappings_info(struct hl_fpriv *hpriv, struct hl_info_args *args)
+{
+	void __user *out = (void __user *) (uintptr_t) args->return_pointer;
+	u32 user_buf_size = args->return_size;
+	struct hl_device *hdev = hpriv->hdev;
+	struct page_fault_info *pgf_info;
+	u64 actual_size;
+
+	pgf_info = &hdev->captured_err_info.pgf_info;
+	args->array_size = pgf_info->num_of_user_mappings;
+
+	if (!out)
+		return -EINVAL;
+
+	actual_size = pgf_info->num_of_user_mappings * sizeof(struct hl_user_mapping);
+	if (user_buf_size < actual_size)
+		return -ENOMEM;
+
+	return copy_to_user(out, pgf_info->user_mappings, min_t(size_t, user_buf_size, actual_size))
+				? -EFAULT : 0;
+}
+
 static int _hl_info_ioctl(struct hl_fpriv *hpriv, void *data,
 				struct device *dev)
 {
@@ -837,6 +873,12 @@ static int _hl_info_ioctl(struct hl_fpriv *hpriv, void *data,
 	case HL_INFO_GET_EVENTS:
 		return events_info(hpriv, args);
 
+	case HL_INFO_PAGE_FAULT_EVENT:
+		return page_fault_info(hpriv, args);
+
+	case HL_INFO_USER_MAPPINGS:
+		return user_mappings_info(hpriv, args);
+
 	default:
 		break;
 	}
diff --git a/drivers/misc/habanalabs/gaudi/gaudi.c b/drivers/misc/habanalabs/gaudi/gaudi.c
index f856ac51fde1..1a99f7be8b60 100644
--- a/drivers/misc/habanalabs/gaudi/gaudi.c
+++ b/drivers/misc/habanalabs/gaudi/gaudi.c
@@ -6755,6 +6755,8 @@ static void gaudi_print_and_get_mmu_error_info(struct hl_device *hdev, u64 *addr
 		*addr |= RREG32(mmMMU_UP_PAGE_ERROR_CAPTURE_VA);
 
 		dev_err_ratelimited(hdev->dev, "MMU page fault on va 0x%llx\n", *addr);
+		hl_capture_page_fault(hdev, *addr, 0, true);
+
 		WREG32(mmMMU_UP_PAGE_ERROR_CAPTURE, 0);
 	}