From mboxrd@z Thu Jan  1 00:00:00 1970
From: hch@lst.de (Christoph Hellwig)
Date: Thu, 9 Nov 2017 14:20:46 +0100
Subject: [PATCH 3/5] nvme: track shared namespaces
In-Reply-To: <1a4292d3-b9c6-976c-69aa-a42f84469cb2@grimberg.me>
References: <20171102183034.9320-1-hch@lst.de>
 <20171102183034.9320-4-hch@lst.de>
 <ddc19090-b69b-ffa1-fd85-497b410b2b94@grimberg.me>
 <20171109125214.GA21795@lst.de>
 <1a4292d3-b9c6-976c-69aa-a42f84469cb2@grimberg.me>
Message-ID: <20171109132046.GA22728@lst.de>

On Thu, Nov 09, 2017@02:59:50PM +0200, Sagi Grimberg wrote:
>
>>>> To allow lockless path lookup the list of nvme_ns structures per
>>>> nvme_ns_head is protected by SRCU, which requires freeing the nvme_ns
>>>> structure through call_srcu.
>>>
>>> Can you remind me why isn't rcu sufficient? Can looking up a
>>> path (ns from head->list) block?
>>
>> blk_mq_make_request can block.
>
> Oh I see, so can you explain why srcu should cover direct_make_request?
>
> What if we were to do:
> --
> 	rcu_read_lock();
> 	ns = nvme_find_path(head);
> 	rcu_read_unlock();
> 	if (likely(ns)) {
> 		...
> --

That way we'd have to take and release a namespace reference for every I/O.