lkml.org 
[lkml]   [2021]   [Oct]   [13]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
From
Subject[PATCH] habanalabs: refactor fence handling in hl_cs_poll_fences
Date
From: Dani Liberman <dliberman@habana.ai>

To avoid checking if fence exists multipled times, changed fence
handling to depend only on the fence status field:

Busy, which means CS still did not completed :
Add its QID so multi CS wait on its completion.
Finished, which means CS completed and fence exists:
Raise its completion bit if it finished mcs handling and
update if necessary the earliest timestamp.
Gone, which means CS already completed and fence deleted:
Update multi CS data to ignore timestamp and raise its
completion bit.

Signed-off-by: Dani Liberman <dliberman@habana.ai>
Reported-by: kernel test robot <lkp@intel.com>
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
---
.../habanalabs/common/command_submission.c | 71 ++++++++++---------
1 file changed, 36 insertions(+), 35 deletions(-)

diff --git a/drivers/misc/habanalabs/common/command_submission.c b/drivers/misc/habanalabs/common/command_submission.c
index f94ac2350a9b..4c8000fd246c 100644
--- a/drivers/misc/habanalabs/common/command_submission.c
+++ b/drivers/misc/habanalabs/common/command_submission.c
@@ -2382,47 +2382,48 @@ static int hl_cs_poll_fences(struct multi_cs_data *mcs_data)
break;
}

- /*
- * It is possible to get an old sequence numbers from user
- * which related to already completed CSs and their fences
- * already gone. In this case, no need to consider its QID for
- * mcs completion.
- */
- if (fence)
+ switch (status) {
+ case CS_WAIT_STATUS_BUSY:
+ /* CS did not finished, keep waiting on its QID*/
mcs_data->stream_master_qid_map |=
fence->stream_master_qid_map;
+ break;
+ case CS_WAIT_STATUS_COMPLETED:
+ /*
+ * Using mcs_handling_done to avoid possibility of mcs_data
+ * returns to user indicating CS completed before it finished
+ * all of its mcs handling, to avoid race the next time the
+ * user waits for mcs.
+ */
+ if (!fence->mcs_handling_done)
+ break;

- /*
- * Using mcs_handling_done to avoid possibility of mcs_data
- * returns to user indicating CS completed before it finished
- * all of its mcs handling, to avoid race the next time the
- * user waits for mcs.
- */
- if (status == CS_WAIT_STATUS_BUSY ||
- (fence && !fence->mcs_handling_done))
- continue;
-
- mcs_data->completion_bitmap |= BIT(i);
-
- /*
- * best effort to extract timestamp. few notes:
- * - if even single fence is gone we cannot extract timestamp
- * (as fence not exist anymore)
- * - for all completed CSs we take the earliest timestamp.
- * for this we have to validate that:
- * 1. given timestamp was indeed set
- * 2. the timestamp is earliest of all timestamps so far
- */
-
- if (status == CS_WAIT_STATUS_GONE) {
+ mcs_data->completion_bitmap |= BIT(i);
+ /*
+ * For all completed CSs we take the earliest timestamp.
+ * For this we have to validate that the timestamp is
+ * earliest of all timestamps so far.
+ */
+ if (mcs_data->update_ts &&
+ (ktime_compare(fence->timestamp, first_cs_time) < 0))
+ first_cs_time = fence->timestamp;
+ break;
+ case CS_WAIT_STATUS_GONE:
mcs_data->update_ts = false;
mcs_data->gone_cs = true;
- } else if (mcs_data->update_ts &&
- (ktime_compare(fence->timestamp,
- ktime_set(0, 0)) > 0) &&
- (ktime_compare(fence->timestamp, first_cs_time) < 0)) {
- first_cs_time = fence->timestamp;
+ /*
+ * It is possible to get an old sequence numbers from user
+ * which related to already completed CSs and their fences
+ * already gone. In this case, CS set as completed but
+ * no need to consider its QID for mcs completion.
+ */
+ mcs_data->completion_bitmap |= BIT(i);
+ break;
+ default:
+ dev_err(hdev->dev, "Invalid fence status\n");
+ return -EINVAL;
}
+
}

hl_fences_put(mcs_data->fence_arr, arr_len);
--
2.17.1
\
 
 \ /
  Last update: 2021-10-13 13:09    [W:0.022 / U:0.172 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site