Re: New Patch series for the UBD Driver

[Jens Axboe <axboe@kernel.dk>]

On 11/12/18 12:11 PM, Jens Axboe wrote:
> On 11/12/18 11:41 AM, Anton Ivanov wrote:
>>>> And hang. Let me try a few more times to see if I can get the spinlock
>>>> recursion and/or abort instead of hang case (they are more difficult to
>>>> reproduce).
>>> This shows exactly what I was talking about, you are trying to
>>> issue IO from irq/non-blocking context. If you look up the warning,
>>> that's exactly what it is telling you.
>>
>>
>> So why does this happen only in the case of an error? Why nothing is 
>> triggered in the OK path?
>>
>> After adding this flag I no longer get the  block/blk.h:439 The spinlock 
>> recursion and the other barfs are alive and well.
>>
>> Back to "what is that" - can I execute blk_update_request, 
>> __blk_mq_end_request, blk_mq_end_request out of an interrupt context?
>>
>> It seems that I can when error is BLK_STS_OK, but not when it is not.
> 
> Post a new stack trace with BLOCKING set. Of course you can call the end
> request functions from irq context, most drivers end up doing that. For
> OK and error.
> 
> There's something fundamentally wrong with your case, it might be reuse
> or something like that.

Took a quick look at your code, and a quick guess would be that your
discard completions are bad. For discard, initialize io_req->length
with -bio->bi_size >> 9. I think this is preventing your discard
completion from ended properly.

-- 
Jens Axboe


_______________________________________________
linux-um mailing list
linux-um@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-um