couple fixes for block and md helpers #77
Conversation
Support new requet op type which was introduced by this commit:
commit ff07a02e9e8e("treewide: Rename enum req_opf into enum req_op")
Also just dump cmd_flags in binary mode if no enum type was used
with it.
Signed-off-by: Junxiao Bi <[email protected]>
rq.part is set only when doing IO statistics, if RQF_IO_STAT is not set for the request, we got the target disk by scanning all disks to match request_queue. Signed-off-by: Junxiao Bi <[email protected]>
Since commit 568f27006577("blk-mq: centralise related handling
into blk_mq_get_driver_tag") was merged in v5.10, for requests
that are not yet dispatched, like those in the plug list, we
should get from "tags.static_rqs", otherwise we may get a wrong
request. To resolve this, first check request from "tags.rqs"
is valid or not, if not get it from "tags.static_rqs".
Signed-off-by: Junxiao Bi <[email protected]>
commit 72adae23a72c("md: Change active_io to percpu") which was
introduced by v6.3 changed mddev.active_io from type atomic_t
to percpu_ref.
Signed-off-by: Junxiao Bi <[email protected]>
There were bugs which caused some inflight io requests was not drained when offline cpu. Add cpu filed in the output to show that. Signed-off-by: Junxiao Bi <[email protected]>
oracle-contributor-agreement
bot
added
the
OCA Verified
I think our CI test did great job. The 1st and 4th commits are for kernel v6.x, so only if CI test supports LUCI, that can be detected. The 2nd commit is for a special upstream kernel build which didn't have RQF_IO_STAT configured, this will not happen for uek. 3rd patch is a corner case which is only for io request in the plug list, it's other kernel bug which caused io pending in that list, it's very hard to get a vmcore with io in that list if no that bug. The 5th commit is adding a new "cpu" field to show cpu status which is triggered by a customer issue. |
Couple fixes was added to make block and md helpers working better with upstream which will benefit future uek. Beside that one fix was adding "cpu" to inflight io output to help tell whether the cpu that request was issued from is in offline status.