[PATCH 3/5] habanalabs: fix bug in timestamps registration code

[Oded Gabbay]

From: farah kassabri <fkassabri@xxxxxxxxx>

Protect re-using the same timestamp buffer record before actually
adding it to the to interrupt wait list.
Mark ts buff offset as in use in the spinlock protection area of the
interrupt wait list to avoid getting in the re-use section in
ts_buff_get_kernel_ts_record before adding the node to the list.
this scenario might happen when multiple threads are racing on
same offset and one thread could set data in the ts buff in
ts_buff_get_kernel_ts_record then the other thread takes over
and get to ts_buff_get_kernel_ts_record and we will try
to re-use the same ts buff offset then we will try to
delete a non existing node from the list.

Signed-off-by: farah kassabri <fkassabri@xxxxxxxxx>
Reviewed-by: Oded Gabbay <ogabbay@xxxxxxxxxx>
Signed-off-by: Oded Gabbay <ogabbay@xxxxxxxxxx>
---
 .../habanalabs/common/command_submission.c    | 33 ++++++++++++-------
 1 file changed, 22 insertions(+), 11 deletions(-)

diff --git a/drivers/accel/habanalabs/common/command_submission.c b/drivers/accel/habanalabs/common/command_submission.c
index c54f504383ac..00fedf2d8654 100644
--- a/drivers/accel/habanalabs/common/command_submission.c
+++ b/drivers/accel/habanalabs/common/command_submission.c
@@ -3180,19 +3180,18 @@ static int ts_buff_get_kernel_ts_record(struct hl_mmap_mem_buf *buf,
 			goto start_over;
 		}
 	} else {
+		/* Fill up the new registration node info */
+		requested_offset_record->ts_reg_info.buf = buf;
+		requested_offset_record->ts_reg_info.cq_cb = cq_cb;
+		requested_offset_record->ts_reg_info.timestamp_kernel_addr =
+				(u64 *) ts_buff->user_buff_address + ts_offset;
+		requested_offset_record->cq_kernel_addr =
+				(u64 *) cq_cb->kernel_address + cq_offset;
+		requested_offset_record->cq_target_value = target_value;
+
 		spin_unlock_irqrestore(wait_list_lock, flags);
 	}
 
-	/* Fill up the new registration node info */
-	requested_offset_record->ts_reg_info.in_use = 1;
-	requested_offset_record->ts_reg_info.buf = buf;
-	requested_offset_record->ts_reg_info.cq_cb = cq_cb;
-	requested_offset_record->ts_reg_info.timestamp_kernel_addr =
-			(u64 *) ts_buff->user_buff_address + ts_offset;
-	requested_offset_record->cq_kernel_addr =
-			(u64 *) cq_cb->kernel_address + cq_offset;
-	requested_offset_record->cq_target_value = target_value;
-
 	*pend = requested_offset_record;
 
 	dev_dbg(buf->mmg->dev, "Found available node in TS kernel CB %p\n",
@@ -3240,7 +3239,7 @@ static int _hl_interrupt_wait_ioctl(struct hl_device *hdev, struct hl_ctx *ctx,
 			goto put_cq_cb;
 		}
 
-		/* Find first available record */
+		/* get ts buffer record */
 		rc = ts_buff_get_kernel_ts_record(buf, cq_cb, ts_offset,
 						cq_counters_offset, target_value,
 						&interrupt->wait_list_lock, &pend);
@@ -3288,7 +3287,19 @@ static int _hl_interrupt_wait_ioctl(struct hl_device *hdev, struct hl_ctx *ctx,
 	 * Note that we cannot have sorted list by target value,
 	 * in order to shorten the list pass loop, since
 	 * same list could have nodes for different cq counter handle.
+	 * Note:
+	 * Mark ts buff offset as in use here in the spinlock protection area
+	 * to avoid getting in the re-use section in ts_buff_get_kernel_ts_record
+	 * before adding the node to the list. this scenario might happen when
+	 * multiple threads are racing on same offset and one thread could
+	 * set the ts buff in ts_buff_get_kernel_ts_record then the other thread
+	 * takes over and get to ts_buff_get_kernel_ts_record and then we will try
+	 * to re-use the same ts buff offset, and will try to delete a non existing
+	 * node from the list.
 	 */
+	if (register_ts_record)
+		pend->ts_reg_info.in_use = 1;
+
 	list_add_tail(&pend->wait_list_node, &interrupt->wait_list_head);
 	spin_unlock_irqrestore(&interrupt->wait_list_lock, flags);
 
-- 
2.25.1