On 01/13/2017 05:41 PM, Omar Sandoval wrote:
> On Fri, Jan 13, 2017 at 12:15:17PM +0100, Hannes Reinecke wrote:
>> On 01/11/2017 10:40 PM, Jens Axboe wrote:
>>> This adds a set of hooks that intercepts the blk-mq path of
>>> allocating/inserting/issuing/completing requests, allowing
>>> us to develop a scheduler within that framework.
>>>
>>> We reuse the existing elevator scheduler API on the registration
>>> side, but augment that with the scheduler flagging support for
>>> the blk-mq interfce, and with a separate set of ops hooks for MQ
>>> devices.
>>>
>>> We split driver and scheduler tags, so we can run the scheduling
>>> independent of device queue depth.
>>>
>>> Signed-off-by: Jens Axboe <ax...@fb.com>
>> [ .. ]
>>> @@ -823,6 +847,35 @@ static inline unsigned int queued_to_index(unsigned
>>> int queued)
>>> return min(BLK_MQ_MAX_DISPATCH_ORDER - 1, ilog2(queued) + 1);
>>> }
>>>
>>> +static bool blk_mq_get_driver_tag(struct request *rq,
>>> + struct blk_mq_hw_ctx **hctx, bool wait)
>>> +{
>>> + struct blk_mq_alloc_data data = {
>>> + .q = rq->q,
>>> + .ctx = rq->mq_ctx,
>>> + .hctx = blk_mq_map_queue(rq->q, rq->mq_ctx->cpu),
>>> + .flags = wait ? 0 : BLK_MQ_REQ_NOWAIT,
>>> + };
>>> +
>>> + if (blk_mq_hctx_stopped(data.hctx))
>>> + return false;
>>> +
>>> + if (rq->tag != -1) {
>>> +done:
>>> + if (hctx)
>>> + *hctx = data.hctx;
>>> + return true;
>>> + }
>>> +
>>> + rq->tag = blk_mq_get_tag(&data);
>>> + if (rq->tag >= 0) {
>>> + data.hctx->tags->rqs[rq->tag] = rq;
>>> + goto done;
>>> + }
>>> +
>>> + return false;
>>> +}
>>> +
>> What happens with the existing request at 'rqs[rq->tag]' ?
>> Surely there is one already, right?
>> Things like '->init_request' assume a fully populated array, so moving
>> one entry to another location is ... interesting.
>>
>> I would have thought we need to do a request cloning here,
>> otherwise this would introduce a memory leak, right?
>> (Not to mention a potential double completion, as the request is now at
>> two positions in the array)
>>
>> Cheers,
>>
>> Hannes
>
> The entries in tags->rqs aren't slab objects, they're pointers into
> pages allocated separately and tracked on tags->page_list. See
> blk_mq_alloc_rqs(). In blk_mq_free_rqs(), we free all of the pages on
> tags->page_list, so there shouldn't be a memory leak.
>
> As for hctx->tags->rqs, entries are only overwritten when a scheduler is
> enabled. In that case, the rqs array is storing pointers to requests
> actually from hctx->sched_tags, so overwriting/leaking isn't an issue.
Ah. Thanks.
That explains it.
Cheers,
Hannes
--
Dr. Hannes Reinecke zSeries & Storage
h...@suse.de +49 911 74053 688
SUSE LINUX Products GmbH, Maxfeldstr. 5, 90409 Nürnberg
GF: J. Hawn, J. Guild, F. Imendörffer, HRB 16746 (AG Nürnberg)
