From: keith.busch@intel.com (Keith Busch)
Date: Fri, 22 May 2015 18:01:14 +0000 (UTC)
Subject: [PATCH 9/9] nvme: submit internal commands through the block layer
In-Reply-To: <alpine.LNX.2.00.1505221731500.15930@localhost.lm.intel.com>
References: <1432285966-31770-1-git-send-email-hch@lst.de>
 <1432285966-31770-10-git-send-email-hch@lst.de>
 <alpine.LNX.2.00.1505221731500.15930@localhost.lm.intel.com>
Message-ID: <alpine.LNX.2.00.1505221759250.15930@localhost.lm.intel.com>

On Fri, 22 May 2015, Keith Busch wrote:
> On Fri, 22 May 2015, Christoph Hellwig wrote:
>> -		req->errors = nvme_error_status(status);
>> +		if (req->cmd_type == REQ_TYPE_DRV_PRIV) {
>> +			req->sense_len = le32_to_cpup(&cqe->result);
>> +			req->errors = status;
>> +		} else {
>> +			req->errors = nvme_error_status(status);
>> +		}
>
> The "sense_len" is now the command result, but it's getting set only in
> the error case, so the initial set features for number of queues gets
> the wrong result; the driver creates only one IO queue when I wanted a
> lot more.

... and blk_execute_rq overwrites rq->sense_len to 0 before returning
if we don't set sense data, so we can never get command results anymore.