On Fri, 2018-05-11 at 20:06 +0800, jianchao.wang wrote:
> Hi bart
>
> I add debug log in blk_mq_add_timer as following
>
> void blk_mq_add_timer(struct request *req, enum mq_rq_state old,
> enum mq_rq_state new)
> {
> struct request_queue *q = req->q;
>
> if (!req->timeout)
> req->timeout = q->rq_timeout;
> if (!blk_mq_rq_set_deadline(req, jiffies + req->timeout, old, new))
> WARN_ON_ONCE(true);
>
> trace_printk("jiffies %lx to %x ldl %lx gen %u dl %x\n",
> jiffies,
> req->timeout,
> blk_rq_deadline(req),
> req->das.generation,
> req->das.deadline);
>
> return __blk_add_timer(req);
> }
>
> And get log below:
>
> jbd2/sda2-8-320 [000] ...1 95.030824: blk_mq_add_timer: jiffies ffff37c0 to 1d4c ldl ffff550c40000000 gen 0 dl ffff550c
> kworker/0:1H-136 [000] ...1 95.031822: blk_mq_add_timer: jiffies ffff37c0 to 1d4c ldl ffff550c40000000 gen 0 dl ffff550c
> kworker/6:1H-244 [006] ...1 95.041695: blk_mq_add_timer: jiffies ffff37c3 to 1d4c ldl ffff550f40000000 gen 0 dl ffff550f
> kworker/6:1H-244 [006] ...1 95.041954: blk_mq_add_timer: jiffies ffff37c3 to 1d4c ldl ffff550f40000000 gen 0 dl ffff550f
>
> The blk_rq_deadline return ffff550c40000000 which looks really crazy.
The bug is in the above trace_printk() call: blk_rq_deadline() must only be used
for the legacy block layer and not for blk-mq code. If you have a look at the value
of the das.deadline field then one can see that the value of that field is correct:
0xffff550c - 0xffff37c0 = 7500 = 30 * 250. Does that mean that HZ = 250 on the system
on which you ran this test?
> And generation never change.
That's a good catch. The code for incrementing the generation number occurs in
blk_mq_change_rq_state() but is missing from blk_mq_rq_set_deadline(). I will fix this.
Bart.
