Re: [PATCH] scsi: scsi_debug: introduce module parameter of 'use_blk_mq'

[Jens Axboe <axboe@kernel.dk>]

On 7/2/18 5:43 PM, Ming Lei wrote:
> On Tue, Jul 3, 2018 at 4:37 AM, Jens Axboe <axboe@kernel.dk> wrote:
>> On 7/2/18 2:28 PM, Douglas Gilbert wrote:
>>> On 2018-07-02 07:06 AM, Ming Lei wrote:
>>>> With the introduced module parameter of 'use_blk_mq', it is easy
>>>> to switch between 'blk_mq' and 'non_blk_mq' by reloading scsi_debug
>>>> module, so that we can test scsi_mq/blk_mq related regressions easily.
>>>>
>>>> Cc: Douglas Gilbert <dgilbert@interlog.com>
>>>> Cc: Bart Van Assche <bart.vanassche@wdc.com>
>>>> Cc: Jens Axboe <axboe@kernel.dk>
>>>> Cc: Omar Sandoval <osandov@fb.com>
>>>> Signed-off-by: Ming Lei <ming.lei@redhat.com>
>>>> ---
>>>>   drivers/scsi/scsi_debug.c | 5 +++++
>>>>   1 file changed, 5 insertions(+)
>>>>
>>>> diff --git a/drivers/scsi/scsi_debug.c b/drivers/scsi/scsi_debug.c
>>>> index 24d7496cd9e2..236cfb669df3 100644
>>>> --- a/drivers/scsi/scsi_debug.c
>>>> +++ b/drivers/scsi/scsi_debug.c
>>>> @@ -144,6 +144,7 @@ static const char *sdebug_version_date = "20180128";
>>>>   #define DEF_SUBMIT_QUEUES 1
>>>>   #define DEF_UUID_CTL 0
>>>>   #define JDELAY_OVERRIDDEN -9999
>>>> +#define DEF_USE_BLK_MQ  0
>>>>
>>>>   #define SDEBUG_LUN_0_VAL 0
>>>>
>>>> @@ -671,6 +672,7 @@ static bool sdebug_verbose;
>>>>   static bool have_dif_prot;
>>>>   static bool write_since_sync;
>>>>   static bool sdebug_statistics = DEF_STATISTICS;
>>>> +static bool sdebug_use_blk_mq = DEF_USE_BLK_MQ;
>>>>
>>>>   static unsigned int sdebug_store_sectors;
>>>>   static sector_t sdebug_capacity;   /* in sectors */
>>>> @@ -4537,6 +4539,7 @@ module_param_named(vpd_use_hostno, sdebug_vpd_use_hostno, int,
>>>>                 S_IRUGO | S_IWUSR);
>>>>   module_param_named(write_same_length, sdebug_write_same_length, int,
>>>>                 S_IRUGO | S_IWUSR);
>>>> +module_param_named(use_blk_mq, sdebug_use_blk_mq, bool, S_IRUGO | S_IWUSR);
>>>>
>>>>   MODULE_AUTHOR("Eric Youngdale + Douglas Gilbert");
>>>>   MODULE_DESCRIPTION("SCSI debug adapter driver");
>>>> @@ -5849,6 +5852,8 @@ static int sdebug_driver_probe(struct device *dev)
>>>>      sdebug_driver_template.can_queue = sdebug_max_queue;
>>>>      if (sdebug_clustering)
>>>>              sdebug_driver_template.use_clustering = ENABLE_CLUSTERING;
>>>> +    if (sdebug_use_blk_mq)
>>>> +            sdebug_driver_template.force_blk_mq = 1;
>>>>      hpnt = scsi_host_alloc(&sdebug_driver_template, sizeof(sdbg_host));
>>>>      if (NULL == hpnt) {
>>>>              pr_err("scsi_host_alloc failed\n");
>>>>
>>>
>>> Hi,
>>> It is up to others whether this patch goes through. It seems the default
>>> associated with blk_mq may soon change. So two things:
>>>    - I'd like to see a MODULE_PARM_DESC(use_blk_mq, ...) entry in patch
>>>    - perhaps a 3 valued use_blk_mq: 1 --> force blk_mq; 0 (def) --> accept
>>>      default mq setting; -1 --> turn off mq (if possible, if not it's a
>>>      NOP)
>>
>> The point is that the parameter is going to be short lived, since the
>> non-mq path will be going away. We don't want drivers exposing this
>> sort of thing, just to remove the option shortly again. Once we're
>> happy with scsi-mq for a release or two, the old code should be
>> deleted. Once that happens, then the option will have no purpose.
> 
> Not like other drivers, scsi_debug is a bit special, since it is for
> test purpose. We may switch to scsi_mq at default soon, but the
> whole MQ(block, scsi_mq) code may take ages to be removed,
> maybe never.

It'll definitely get removed, the motivation to do that is strong. I'm
not in violent disagreement with you, my main point is just that it's
going to be a temporary option. Once we flip the switch again and remove
the non-mq path for SCSI a few releases later, then the option is dead.

> We may use the per-driver parameter to test both mq and non-mq
> path, especially for comparing both, then speed up to make MQ code
> mature & stable. That is why I think this patch is very useful.

Also possible with just a reboot, or fiddle with the scsi_mod.use_blk_mq
option...

-- 
Jens Axboe