From mboxrd@z Thu Jan  1 00:00:00 1970
From: =?UTF-8?B?TWF0aWFzIEJqw7hybGluZw==?= <m@bjorling.me>
Subject: Re: [PATCH v7 1/5] lightnvm: Support for Open-Channel SSDs
Date: Wed, 02 Sep 2015 12:48:58 +0200
Message-ID: <55E6D41A.5060100@bjorling.me>
References: <1438957791-24632-1-git-send-email-mb@lightnvm.io> <1438957791-24632-2-git-send-email-mb@lightnvm.io> <55E67218.9010701@cn.fujitsu.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Transfer-Encoding: 7bit
Cc: jg@lightnvm.io, Stephen.Bates@pmcs.com, keith.busch@intel.com,
	=?UTF-8?B?TWF0aWFzIEJqw7hybGluZw==?= <mb@lightnvm.io>
To: Dongsheng Yang <yangds.fnst@cn.fujitsu.com>, hch@infradead.org,
	axboe@fb.com, linux-fsdevel@vger.kernel.org,
	linux-kernel@vger.kernel.org, linux-nvme@lists.infradead.org
Return-path: <linux-kernel-owner@vger.kernel.org>
In-Reply-To: <55E67218.9010701@cn.fujitsu.com>
Sender: linux-kernel-owner@vger.kernel.org
List-Id: linux-fsdevel.vger.kernel.org

>> +
>> +    /* register with device with a supported BM */
>> +    list_for_each_entry(bt, &nvm_bms, list) {
>> +        ret = bt->register_bm(dev);
>> +        if (ret < 0)
>> +            goto err; /* initialization failed */
>> +        if (ret > 0) {
>> +            dev->bm = bt;
>> +            break; /* successfully initialized */
>> +        }
>> +    }
>
> Why just search it from head to tail? Can user specific it
> in nvm_create_target()?

Hi Yang,

Currently only the rrpc and a couple of out of tree block managers are 
built. The register_bm only tries to find a block manager that supports 
the device, when it finds it, that  one is initialized. It is an open 
question on how we choose the right block manager, e.g. a proprietary 
and a open-source block manager is in place. Priorities might be a way 
to go? or mark certain block managers as a catch all?

Hopefully we will get away with only a single or two block managers in 
the future, so we won't have one for each type of device.

>> +
>> +    if (!ret) {
>> +        pr_info("nvm: no compatible bm was found.\n");
>> +        return 0;
>> +    }
>
> If we allow nvm_device registered with no bm, we would get
> a NULL pointer reference problem in later using.
>

Yes, definitely. In the care that happens, I envision it should be 
possible to register a block manager after a device is loaded, and then 
any outstanding devices (which does not have a registered block 
manager), will be probed again.

> As mentioned above, why we have to choose bm for nvm in nvm_register?

Without a block manager, we don't know the structure of the device and 
how to interact with it. I want to initialize that as soon as possible. 
So that layers on top can start interacting.

>
> Thanx
> Yang