[dpdk-dev] [PATCH v1] net/i40e: support request any number of queues

[Kevin Traynor]

On 17/01/2019 07:26, Yan, Zhirun wrote:
> 
> 
>> -----Original Message-----
>> From: Kevin Traynor [mailto:ktraynor at redhat.com]
>> Sent: Wednesday, January 16, 2019 1:34 AM
>> To: Yan, Zhirun <zhirun.yan at intel.com>; dev at dpdk.org; Zhang, Qi Z
>> <qi.z.zhang at intel.com>
>> Subject: Re: [dpdk-dev] [PATCH v1] net/i40e: support request any number of
>> queues
>>
>> On 01/15/2019 02:57 PM, Zhirun Yan wrote:
>>> Before this patch, VF must request a specific queues(1/2/4/8/16) with
>>> DPDK PF. This patch align the number of requested queues to next power
>>> of 2. So VF can request any number queues from 1 to 16.
>>>
>>> Signed-off-by: Zhirun Yan <zhirun.yan at intel.com>
>>> ---
>>>  drivers/net/i40e/i40e_pf.c | 5 ++++-
>>>  1 file changed, 4 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/net/i40e/i40e_pf.c b/drivers/net/i40e/i40e_pf.c
>>> index 092e0d3e9..d110dffb7 100644
>>> --- a/drivers/net/i40e/i40e_pf.c
>>> +++ b/drivers/net/i40e/i40e_pf.c
>>> @@ -1271,7 +1271,10 @@
>> i40e_pf_host_process_cmd_request_queues(struct i40e_pf_vf *vf, uint8_t
>> *msg)
>>>  	} else {
>>>  		i40e_vc_notify_vf_reset(vf);
>>>  		vf->vsi->nb_qps = req_pairs;
>>> -		pf->vf_nb_qps = req_pairs;
>>> +		if (rte_is_power_of_2(req_pairs))
>>> +			pf->vf_nb_qps = req_pairs;
>>> +		else
>>> +			pf->vf_nb_qps = i40e_align_floor(req_pairs) << 1;
>>
>> Shouldn't you do this before you validate against the number of free queue
>> pairs? or it is somehow handled
> 
> Before PF alloc queue to VF, PF will check the queue number in i40e_res_pool_alloc().
> Suppose there are 15 queues left in PF, this part will check 12 as a valid value if request 12 queues,
>  But PF  will refuse to alloc 16 queue.
> 

Hello. In that scenario the vsi is reset and there is a failed setup,
whereas it could be caught here and max rounded value sent back to vf. I
will send a patch to show what I mean (compile tested only). Please take
a look.

>>
>>>  		i40e_pf_host_process_cmd_reset_vf(vf);
>>>
>>>  		return 0;
>>>
>