debugfs: add skip_reset_on_timeout option

To be able to debug long-running CS better, without changing the
userspace code, we are adding a new option through debugfs interface
to skip the reset of the device in case of CS timeout.
Signed-off-by: Yuri Nudelman <ynudelman@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>

Documentation/ABI/testing/debugfs-driver-habanalabs

@@ -207,6 +207,14 @@ Contact:        ogabbay@kernel.org
 Description:    Sets the PCI power state. Valid values are "1" for D0 and "2"
                 for D3Hot
 
+What:           /sys/kernel/debug/habanalabs/hl<n>/skip_reset_on_timeout
+Date:           Jun 2021
+KernelVersion:  5.13
+Contact:        ynudelman@habana.ai
+Description:    Sets the skip reset on timeout option for the device. Value of
+                "0" means device will be reset in case some CS has timed out,
+                otherwise it will not be reset.
+
 What:           /sys/kernel/debug/habanalabs/hl<n>/stop_on_err
 Date:           Mar 2020
 KernelVersion:  5.6

drivers/misc/habanalabs/common/command_submission.c

@@ -663,6 +663,7 @@ static int allocate_cs(struct hl_device *hdev, struct hl_ctx *ctx,
 	cs->timestamp = !!(flags & HL_CS_FLAGS_TIMESTAMP);
 	cs->timeout_jiffies = timeout;
 	cs->skip_reset_on_timeout =
+		hdev->skip_reset_on_timeout ||
 		!!(flags & HL_CS_FLAGS_SKIP_RESET_ON_TIMEOUT);
 	cs->submission_time_jiffies = jiffies;
 	INIT_LIST_HEAD(&cs->job_list);

drivers/misc/habanalabs/common/debugfs.c

@@ -1278,6 +1278,11 @@ void hl_debugfs_add_device(struct hl_device *hdev)
 				dev_entry->root,
 				&dev_entry->blob_desc);
 
+	debugfs_create_x8("skip_reset_on_timeout",
+				0644,
+				dev_entry->root,
+				&hdev->skip_reset_on_timeout);
+
 	for (i = 0, entry = dev_entry->entry_arr ; i < count ; i++, entry++) {
 		debugfs_create_file(hl_debugfs_list[i].name,
 					0444,

drivers/misc/habanalabs/common/habanalabs.h

@@ -2191,6 +2191,8 @@ struct hl_mmu_funcs {
  * @supports_staged_submission: true if staged submissions are supported
  * @curr_reset_cause: saves an enumerated reset cause when a hard reset is
  *                    triggered, and cleared after it is shared with preboot.
+ * @skip_reset_on_timeout: Skip device reset if CS has timed out, wait for it to
+ *                         complete instead.
  */
 struct hl_device {
 	struct pci_dev			*pdev;
@@ -2305,6 +2307,7 @@ struct hl_device {
 	u8				device_fini_pending;
 	u8				supports_staged_submission;
 	u8				curr_reset_cause;
+	u8				skip_reset_on_timeout;
 
 	/* Parameters for bring-up */
 	u64				nic_ports_mask;