Re: [Update PATCH V3] md: don't unregister sync_thread with reconfig_mutex held

[Guoqing Jiang <guoqing.jiang@linux.dev>]

>>>> I noticed a clear regression with mdadm tests with this patch in 
>>>> md-next
>>>> (7e6ba434cc6080).
>>>>
>>>> Before the patch, tests 07reshape5intr and 07revert-grow would fail
>>>> fairly infrequently (about 1 in 4 runs for the former and 1 in 30 runs
>>>> for the latter).
>>>>
>>>> After this patch, both tests always fail.
>>>>
>>>> I don't have time to dig into why this is, but it would be nice if
>>>> someone can at least fix the regression. It is hard to make any 
>>>> progress
>>>> on these tests if we are continuing to further break them.

I have tried with both ubuntu 22.04 kernel which is 5.15 and vanilla 
5.12, none of them
can pass your mentioned tests.

[root@localhost mdadm]# lsblk|grep vd
vda          252:0    0    1G  0 disk
vdb          252:16   0    1G  0 disk
vdc          252:32   0    1G  0 disk
vdd          252:48   0    1G  0 disk
[root@localhost mdadm]# ./test --dev=disk --disks=/dev/vd{a..d} 
--tests=05r1-add-internalbitmap
Testing on linux-5.12.0-default kernel
/root/mdadm/tests/05r1-add-internalbitmap... succeeded
[root@localhost mdadm]# ./test --dev=disk --disks=/dev/vd{a..d} 
--tests=07reshape5intr
Testing on linux-5.12.0-default kernel
/root/mdadm/tests/07reshape5intr... FAILED - see 
/var/tmp/07reshape5intr.log and /var/tmp/fail07reshape5intr.log for details
[root@localhost mdadm]# ./test --dev=disk --disks=/dev/vd{a..d} 
--tests=07revert-grow
Testing on linux-5.12.0-default kernel
/root/mdadm/tests/07revert-grow... FAILED - see 
/var/tmp/07revert-grow.log and /var/tmp/fail07revert-grow.log for details
[root@localhost mdadm]# head -10  /var/tmp/07revert-grow.log | grep mdadm
+ . /root/mdadm/tests/07revert-grow
*++ mdadm -CR --assume-clean /dev/md0 -l5 -n4 -x1 /dev/vda /dev/vdb 
/dev/vdc /dev/vdd /dev/vda /dev/vdb /dev/vdc /dev/vdd --metadata=0.9**
*
The above line is clearly wrong from my understanding.

And let's check ubuntu 22.04.

root@vm:/home/gjiang/mdadm# lsblk|grep vd
vda    252:0    0     1G  0 disk
vdb    252:16   0     1G  0 disk
vdc    252:32   0     1G  0 disk
root@vm:/home/gjiang/mdadm# ./test --dev=disk --disks=/dev/vd{a..d} 
--tests=05r1-failfast
Testing on linux-5.15.0-30-generic kernel
/home/gjiang/mdadm/tests/05r1-failfast... succeeded
root@vm:/home/gjiang/mdadm# ./test --dev=disk --disks=/dev/vd{a..c}   
--tests=07reshape5intr
Testing on linux-5.15.0-30-generic kernel
/home/gjiang/mdadm/tests/07reshape5intr... FAILED - see 
/var/tmp/07reshape5intr.log and /var/tmp/fail07reshape5intr.log for details
root@vm:/home/gjiang/mdadm# ./test --dev=disk --disks=/dev/vd{a..c} 
--tests=07revert-grow
Testing on linux-5.15.0-30-generic kernel
/home/gjiang/mdadm/tests/07revert-grow... FAILED - see 
/var/tmp/07revert-grow.log and /var/tmp/fail07revert-grow.log for details

So I would not consider it is regression.

[ ... ]

> FYI: I've used loop devices on a virtio disk.
>
> I later discovered Logans patches [1], which I were not aware of, as 
> I'm not subscribed to the lists.
>
> [1]: 
> https://lore.kernel.org/linux-raid/20220519191311.17119-6-logang@deltatee.com/T/#u
>
> The series seems to acknowledge that there are open problems and tries 
> to fix them.
> So I've used his md-bug branch from 
> https://github.com/sbates130272/linux-p2pmem but it didn't look better.

Thanks for your effort.

> So I understand, the mdadm tests *are* supposed to work and every bug 
> I see here is worth analyzing? Or is Logan hunting down everything 
> anyway?

Yes, it was supposed to be. But unfortunately, it was kind of broken, 
good news is people are aware
of it and try to make it works/better, pls see other links.

[1] 
https://lore.kernel.org/linux-raid/EA6887B4-2A44-49D0-ACF9-C04CC92AFD87@oracle.com/T/#t
[2] 
https://lore.kernel.org/linux-raid/CALTww2-mbfZRcHu_95Q+WANXZ9ayRwjXvyvqP5Gseeb86dEy=w@mail.gmail.com/T/#t

Thanks,
Guoqing