Re: [PATCH] nvme: Add ctrl attributes for queue_count and sqsize

[James Smart <james.smart@broadcom.com>]

On 9/25/2019 2:36 PM, Keith Busch wrote:
> On Tue, Sep 24, 2019 at 02:22:08PM -0700, James Smart wrote:
>> Current controller interrogation requires a lot of guesswork
>> on how many io queues were created and what the io sq size is.
>> The numbers are dependent upon core/fabric defaults, connect
>> arguments, and target responses.
>>
>> Add sysfs attributes for queue_count and sqsize.
>>
>> Signed-off-by: James Smart <jsmart2021@gmail.com>
>> ---
>>   drivers/nvme/host/core.c | 4 ++++
>>   1 file changed, 4 insertions(+)
>>
>> diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
>> index e0e2dcbfd05e..b1c24af284ca 100644
>> --- a/drivers/nvme/host/core.c
>> +++ b/drivers/nvme/host/core.c
>> @@ -3040,6 +3040,8 @@ static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
>>   
>>   nvme_show_int_function(cntlid);
>>   nvme_show_int_function(numa_node);
>> +nvme_show_int_function(queue_count);
>> +nvme_show_int_function(sqsize);
> The sqsize is only valid for fabrics, so I think you need to update the
> 'is_visible' to suppress this attribute for pcie.

I would think knowing the SQ size or an I/O queue is still interesting 
although perhaps not as useful.

Looking in pci.c:
- dev->q_depth is the baseline, getting stamped into each of the queue 
structures
- nvme_pci_enable() is setting dev->ctrl.sqsize to dev->q_depth - 1;

Or are you asking not to display it as the implied model for how many 
outstanding ios is different ?
note: I wasn't trying to go there - only to say what the SQ size is for 
an I/O queue

-- james


_______________________________________________
Linux-nvme mailing list
Linux-nvme@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-nvme