On 12/08/14 17:49, Jens Axboe wrote:
> On 12/08/2014 07:55 AM, Bart Van Assche wrote:
>> diff --git a/block/blk-mq-tag.c b/block/blk-mq-tag.c
>> index 67ab88b..e88af88 100644
>> --- a/block/blk-mq-tag.c
>> +++ b/block/blk-mq-tag.c
>> @@ -256,6 +256,8 @@ static int bt_get(struct blk_mq_alloc_data *data,
>> break;
>> }
>>
>> + blk_mq_run_hw_queue(hctx, false);
>> +
>> blk_mq_put_ctx(data->ctx);
>>
>> io_schedule();
>
> Ah yes, that could be an issue for some cases, we do need to run the
> queue there. For a tag map shared across hardware queues, we might need
> to run more than just the current queue, however. For now we can safely
> assume that we allocate fairly, so it should not be an issue.
>
> It might be worth experimenting with doing a __bt_get() after the queue
> run before going to sleep, in case the queue run found completions as well.
Unless anyone objects I will start testing the following patch:
[PATCH] blk-mq: Fix bt_get() hang
Avoid that if there are fewer hardware queues than CPU threads that
bt_get() can hang. The symptoms of the hang were as follows:
* All tags allocated for a particular hardware queue.
* (nr_tags) pending commands for that hardware queue.
* No pending commands for the software queues associated with that
hardware queue.
The call stack that corresponds to the hang is as follows:
io_schedule+0x9c/0x130
bt_get+0xef/0x180
blk_mq_get_tag+0x9f/0xd0
__blk_mq_alloc_request+0x16/0x1f0
blk_mq_map_request+0x123/0x130
blk_mq_make_request+0x69/0x280
generic_make_request+0xc0/0x110
submit_bio+0x64/0x130
do_blockdev_direct_IO+0x1dc8/0x2da0
__blockdev_direct_IO+0x47/0x50
blkdev_direct_IO+0x49/0x50
generic_file_read_iter+0x546/0x610
blkdev_read_iter+0x32/0x40
aio_run_iocb+0x1f8/0x400
do_io_submit+0x121/0x490
SyS_io_submit+0xb/0x10
system_call_fastpath+0x12/0x17
---
block/blk-mq-tag.c | 6 ++++++
1 file changed, 6 insertions(+)
diff --git a/block/blk-mq-tag.c b/block/blk-mq-tag.c
index c22491e..14ab120 100644
--- a/block/blk-mq-tag.c
+++ b/block/blk-mq-tag.c
@@ -256,6 +256,12 @@ static int bt_get(struct blk_mq_alloc_data *data,
if (tag != -1)
break;
+ blk_mq_run_hw_queue(hctx, false);
+
+ tag = __bt_get(hctx, bt, last_tag);
+ if (tag != -1)
+ break;
+
blk_mq_put_ctx(data->ctx);
io_schedule();
--
2.1.2
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
