Re: [PATCH -next 2/3] md/raid10: convert resync_lock to use seqlock

[Logan Gunthorpe <logang@deltatee.com>]

On 2022-09-01 18:49, Guoqing Jiang wrote:
> 
> 
> On 9/2/22 2:41 AM, Logan Gunthorpe wrote:
>> Hi,
>>
>> On 2022-08-29 07:15, Yu Kuai wrote:
>>> From: Yu Kuai <yukuai3@huawei.com>
>>>
>>> Currently, wait_barrier() will hold 'resync_lock' to read
>>> 'conf->barrier',
>>> and io can't be dispatched until 'barrier' is dropped.
>>>
>>> Since holding the 'barrier' is not common, convert 'resync_lock' to use
>>> seqlock so that holding lock can be avoided in fast path.
>>>
>>> Signed-off-by: Yu Kuai <yukuai3@huawei.com>
>> I've found some lockdep issues starting with this patch in md-next while
>> running mdadm tests (specifically 00raid10 when run about 10 times in a
>> row).
>>
>> I've seen a couple different lock dep errors. The first seems to be
>> reproducible on this patch, then it possibly changes to the second on
>> subsequent patches. Not sure exactly.
> 
> That's why I said "try mdadm test suites too to avoid regression." ...

You may have to run it multiple times, a single run tends not to catch
all errors. I had to loop the noted test 10 times to be sure I hit this
every time when I did the simple bisect.

And ensure that all the debug options are on when you run it (take a
look at the Kernel Hacking section in menuconfig). You won't hit this
bug without at least CONFIG_PROVE_LOCKING=y.

Logan