Re: [PATCH v2 2/5] ufs: mcq: Add support for clean up mcq resources

[Bart Van Assche <bvanassche@acm.org>]

On 5/3/23 21:01, Bao D. Nguyen wrote:
> On 4/25/2023 5:08 PM, Bart Van Assche wrote:
>> On 4/17/23 14:05, Bao D. Nguyen wrote:
>>> @@ -3110,7 +3128,7 @@ static int ufshcd_wait_for_dev_cmd(struct ufs_hba *hba,
>>>           err = -ETIMEDOUT;
>>>           dev_dbg(hba->dev, "%s: dev_cmd request timedout, tag %d\n",
>>>               __func__, lrbp->task_tag);
>>> -        if (ufshcd_clear_cmds(hba, 1U << lrbp->task_tag) == 0) {
>>> +        if (ufshcd_clear_cmds(hba, 1UL << lrbp->task_tag) == 0) {
>>>               /* successfully cleared the command, retry if needed */
>>>               err = -EAGAIN;
>>>               /*
>>
>> Is this change necessary?
> My intention was to support tag greater than 31 and less than 64.
> The 1U << only works up to 32 bits.

The UFSHCI 4.0 standard and also the Linux kernel UFS driver support more than
64 outstanding commands so the above change is not sufficient.

>> Please do not use lrbp->cmd to check whether or not a command has completed. See also my patch "scsi: ufs: Fix handling of lrbp->cmd".
> I have been thinking how to replace lrbp->cmd, but could not find a good solution. Throughout this patch series, I am using lrbp->cmd as a way to find the pending command that is being aborted and clean up the resources associated with it. Any suggestions to achieve it? 

Using lrbp->cmd is fine but checking whether or not it is NULL is not OK. How
about using blk_mq_request_started() to check whether or not a request is still
pending?

Thanks,

Bart.