Re: [PATCH v4 16/17] scsi: ufs: Optimize the command queueing code

["Asutosh Das (asd)" <asutoshd@codeaurora.org>]

On 12/6/2021 2:41 PM, Asutosh Das (asd) wrote:
> On 12/3/2021 3:19 PM, Bart Van Assche wrote:
>> Remove the clock scaling lock from ufshcd_queuecommand() since it is a
>> performance bottleneck. Instead check the SCSI device budget bitmaps in
>> the code that waits for ongoing ufshcd_queuecommand() calls. A bit is
>> set in sdev->budget_map just before scsi_queue_rq() is called and a bit
>> is cleared from that bitmap if scsi_queue_rq() does not submit the
>> request or after the request has finished. See also the
>> blk_mq_{get,put}_dispatch_budget() calls in the block layer.
>>
>> There is no risk for a livelock since the block layer delays queue
>> reruns if queueing a request fails because the SCSI host has been
>> blocked.
>>
>> Cc: Asutosh Das (asd) <asutoshd@codeaurora.org>
>> Signed-off-by: Bart Van Assche <bvanassche@acm.org>
>> ---
> 
> Reviewed-by: Asutosh Das <asutoshd@codeaurora.org>
> 

Replying to my own mail.

Hi Bart,
>>   drivers/scsi/ufs/ufshcd.c | 33 +++++++++++++++++++++++----------
>>   drivers/scsi/ufs/ufshcd.h |  1 +
>>   2 files changed, 24 insertions(+), 10 deletions(-)
>>
>> diff --git a/drivers/scsi/ufs/ufshcd.c b/drivers/scsi/ufs/ufshcd.c
>> index 9f0a1f637030..650dddf960c2 100644
>> --- a/drivers/scsi/ufs/ufshcd.c
>> +++ b/drivers/scsi/ufs/ufshcd.c
>> @@ -1070,13 +1070,31 @@ static bool 
>> ufshcd_is_devfreq_scaling_required(struct ufs_hba *hba,
>>       return false;
>>   }
>> +/*
>> + * Determine the number of pending commands by counting the bits in 
>> the SCSI
>> + * device budget maps. This approach has been selected because a bit 
>> is set in
>> + * the budget map before scsi_host_queue_ready() checks the 
>> host_self_blocked
>> + * flag. The host_self_blocked flag can be modified by calling
>> + * scsi_block_requests() or scsi_unblock_requests().
>> + */
>> +static u32 ufshcd_pending_cmds(struct ufs_hba *hba)
>> +{
>> +    struct scsi_device *sdev;
>> +    u32 pending = 0;
>> +
>> +    shost_for_each_device(sdev, hba->host)
>> +        pending += sbitmap_weight(&sdev->budget_map);
>> +
I was porting this change to my downstream code and it occurred to me 
that in a high IO rate scenario it's possible that bits in the 
budget_map may be set even when that particular IO may not be issued to 
driver. So there would unnecessary waiting for that to be cleared.
Do you think it's possible?
I think we should wait only for requests which are already started.
e.g. blk_mq_tagset_busy_iter() ?

PLMK your thoughts on this.

>> +    return pending;
>> +}
>> +
>>   static int ufshcd_wait_for_doorbell_clr(struct ufs_hba *hba,
>>                       u64 wait_timeout_us)
>>   {
>>       unsigned long flags;
>>       int ret = 0;
>>       u32 tm_doorbell;
>> -    u32 tr_doorbell;
>> +    u32 tr_pending;
>>       bool timeout = false, do_last_check = false;
>>       ktime_t start;
>> @@ -1094,8 +1112,8 @@ static int ufshcd_wait_for_doorbell_clr(struct 
>> ufs_hba *hba,
>>           }
>>           tm_doorbell = ufshcd_readl(hba, REG_UTP_TASK_REQ_DOOR_BELL);
>> -        tr_doorbell = ufshcd_readl(hba, REG_UTP_TRANSFER_REQ_DOOR_BELL);
>> -        if (!tm_doorbell && !tr_doorbell) {
>> +        tr_pending = ufshcd_pending_cmds(hba);
>> +        if (!tm_doorbell && !tr_pending) {
>>               timeout = false;
>>               break;
>>           } else if (do_last_check) {
>> @@ -1115,12 +1133,12 @@ static int ufshcd_wait_for_doorbell_clr(struct 
>> ufs_hba *hba,
>>               do_last_check = true;
>>           }
>>           spin_lock_irqsave(hba->host->host_lock, flags);
>> -    } while (tm_doorbell || tr_doorbell);
>> +    } while (tm_doorbell || tr_pending);
>>       if (timeout) {
>>           dev_err(hba->dev,
>>               "%s: timedout waiting for doorbell to clear (tm=0x%x, 
>> tr=0x%x)\n",
>> -            __func__, tm_doorbell, tr_doorbell);
>> +            __func__, tm_doorbell, tr_pending);
>>           ret = -EBUSY;
>>       }
>>   out:
>> @@ -2681,9 +2699,6 @@ static int ufshcd_queuecommand(struct Scsi_Host 
>> *host, struct scsi_cmnd *cmd)
>>       WARN_ONCE(tag < 0, "Invalid tag %d\n", tag);
>> -    if (!down_read_trylock(&hba->clk_scaling_lock))
>> -        return SCSI_MLQUEUE_HOST_BUSY;
>> -
>>       /*
>>        * Allows the UFS error handler to wait for prior 
>> ufshcd_queuecommand()
>>        * calls.
>> @@ -2772,8 +2787,6 @@ static int ufshcd_queuecommand(struct Scsi_Host 
>> *host, struct scsi_cmnd *cmd)
>>   out:
>>       rcu_read_unlock();
>> -    up_read(&hba->clk_scaling_lock);
>> -
>>       if (ufs_trigger_eh()) {
>>           unsigned long flags;
>> diff --git a/drivers/scsi/ufs/ufshcd.h b/drivers/scsi/ufs/ufshcd.h
>> index 8e942762e668..88c20f3608c2 100644
>> --- a/drivers/scsi/ufs/ufshcd.h
>> +++ b/drivers/scsi/ufs/ufshcd.h
>> @@ -778,6 +778,7 @@ struct ufs_hba_monitor {
>>    * @clk_list_head: UFS host controller clocks list node head
>>    * @pwr_info: holds current power mode
>>    * @max_pwr_info: keeps the device max valid pwm
>> + * @clk_scaling_lock: used to serialize device commands and clock 
>> scaling
>>    * @desc_size: descriptor sizes reported by device
>>    * @urgent_bkops_lvl: keeps track of urgent bkops level for device
>>    * @is_urgent_bkops_lvl_checked: keeps track if the urgent bkops 
>> level for
>>
> 
> 


-- 
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
Linux Foundation Collaborative Project