[PATCH 05/12] blk-mq: Introduce blk_mq_quiesce_queue()

[Bart Van Assche]

On 10/26/16 18:30, Ming Lei wrote:
> On Thu, Oct 27, 2016 at 6:53 AM, Bart Van Assche
> <bart.vanassche at sandisk.com> wrote:
>> blk_mq_quiesce_queue() waits until ongoing .queue_rq() invocations
>> have finished. This function does *not* wait until all outstanding
>> requests have finished (this means invocation of request.end_io()).
>> The algorithm used by blk_mq_quiesce_queue() is as follows:
>> * Hold either an RCU read lock or an SRCU read lock around
>>   .queue_rq() calls. The former is used if .queue_rq() does not
>>   block and the latter if .queue_rq() may block.
>> * blk_mq_quiesce_queue() calls synchronize_srcu() or
>>   synchronize_rcu() to wait for .queue_rq() invocations that
>>   started before blk_mq_quiesce_queue() was called.
>> * The blk_mq_hctx_stopped() calls that control whether or not
>>   .queue_rq() will be called are called with the (S)RCU read lock
>>   held. This is necessary to avoid race conditions against
>>   the "blk_mq_stop_hw_queues(q); blk_mq_quiesce_queue(q);"
>>   sequence from another thread.
>>
>> Signed-off-by: Bart Van Assche <bart.vanassche at sandisk.com>
>> Cc: Christoph Hellwig <hch at lst.de>
>> Cc: Ming Lei <tom.leiming at gmail.com>
>> Cc: Hannes Reinecke <hare at suse.com>
>> Cc: Johannes Thumshirn <jthumshirn at suse.de>
>> ---
>>  block/Kconfig          |  1 +
>>  block/blk-mq.c         | 69 +++++++++++++++++++++++++++++++++++++++++++++-----
>>  include/linux/blk-mq.h |  3 +++
>>  include/linux/blkdev.h |  1 +
>>  4 files changed, 67 insertions(+), 7 deletions(-)
>>
>> diff --git a/block/Kconfig b/block/Kconfig
>> index 1d4d624..0562ef9 100644
>> --- a/block/Kconfig
>> +++ b/block/Kconfig
>> @@ -5,6 +5,7 @@ menuconfig BLOCK
>>         bool "Enable the block layer" if EXPERT
>>         default y
>>         select SBITMAP
>> +       select SRCU
>>         help
>>          Provide block layer support for the kernel.
>>
>> diff --git a/block/blk-mq.c b/block/blk-mq.c
>> index 0cf21c2..4945437 100644
>> --- a/block/blk-mq.c
>> +++ b/block/blk-mq.c
>> @@ -115,6 +115,31 @@ void blk_mq_unfreeze_queue(struct request_queue *q)
>>  }
>>  EXPORT_SYMBOL_GPL(blk_mq_unfreeze_queue);
>>
>> +/**
>> + * blk_mq_quiesce_queue() - wait until all ongoing queue_rq calls have finished
>> + * @q: request queue.
>> + *
>> + * Note: this function does not prevent that the struct request end_io()
>> + * callback function is invoked. Additionally, it is not prevented that
>> + * new queue_rq() calls occur unless the queue has been stopped first.
>> + */
>> +void blk_mq_quiesce_queue(struct request_queue *q)
>> +{
>> +       struct blk_mq_hw_ctx *hctx;
>> +       unsigned int i;
>> +       bool rcu = false;
>
> Before synchronizing SRCU/RCU, we have to set a per-hctx flag
> or per-queue flag to block comming .queue_rq(), seems I mentioned
> that before:
>
>    https://www.spinics.net/lists/linux-rdma/msg41389.html

Hello Ming,

Thanks for having included an URL to an archived version of that 
discussion. What I remember about that discussion is that I proposed to 
use the existing flag BLK_MQ_S_STOPPED instead of introducing a
new QUEUE_FLAG_QUIESCING flag and that you agreed with that proposal. 
See also https://www.spinics.net/lists/linux-rdma/msg41430.html.

Bart.