[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <alpine.LRH.2.03.1408140920520.14603@AMR>
Date: Thu, 14 Aug 2014 09:33:38 -0600 (MDT)
From: Keith Busch <keith.busch@...el.com>
To: Jens Axboe <axboe@...com>
cc: Matias Bjørling <m@...rling.me>,
Keith Busch <keith.busch@...el.com>,
Matthew Wilcox <willy@...ux.intel.com>,
"Sam Bradshaw (sbradshaw)" <sbradshaw@...ron.com>,
LKML <linux-kernel@...r.kernel.org>,
linux-nvme <linux-nvme@...ts.infradead.org>,
Christoph Hellwig <hch@...radead.org>,
Rob Nelson <rlnelson@...gle.com>,
Ming Lei <tom.leiming@...il.com>
Subject: Re: [PATCH v11] NVMe: Convert to blk-mq
On Thu, 14 Aug 2014, Jens Axboe wrote:
> On 08/14/2014 02:25 AM, Matias Bjørling wrote:
>> The result is set to BLK_MQ_RQ_QUEUE_ERROR, or am I mistaken?
>
> Looks OK to me, looking at the code, 'result' is initialized to
> BLK_MQ_RQ_QUEUE_BUSY though. Which looks correct, we don't want to error
> on a suspended queue.
My mistake missing how the result was initialized.
> nr_tags must be uninitialized or screwed up somehow, otherwise I don't
> see how that kmalloc() could warn on being too large. Keith, are you
> running with slab debugging? Matias, might be worth trying.
I'm not running with slab debugging. If it's any clue at all, blk-mq is
using 16 of the 31 allocated h/w queues (which is okay as we discussed
earlier), and the oops happens when clearing the first unused queue.
I'll have time to mess with this more today, so I can either help find
the problem or apply a patch if one becomes available.
Powered by blists - more mailing lists