Re: [net-next-2.6 PATCH 3/3] ixgbe: Do not allocate too many netdev txqueues

[John Fastabend <john.r.fastabend@intel.com>]

Waskiewicz Jr, Peter P wrote:
> On Sat, 2010-02-27 at 20:57 -0700, Eric Dumazet wrote:
>   
>> Le samedi 27 février 2010 à 17:02 -0800, Peter P Waskiewicz Jr a écrit :
>>     
>>> On Fri, 2010-02-26 at 06:04 -0800, Eric Dumazet wrote:
>>>       
>>>> Le vendredi 26 février 2010 à 01:15 -0800, Jeff Kirsher a écrit :
>>>>         
>>>>> +	if (ii->mac == ixgbe_mac_82598EB)
>>>>> +		indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
>>>>> +	else
>>>>> +		indices = min_t(unsigned int, indices, IXGBE_MAX_FDIR_INDICES);
>>>>> +
>>>>> +	indices = max_t(unsigned int, indices, IXGBE_MAX_DCB_INDICES);
>>>>> +#ifdef IXGBE_FCOE
>>>>> +	indices += min_t(unsigned int, num_possible_cpus(),
>>>>> +			 IXGBE_MAX_FCOE_INDICES);
>>>>> +#endif
>>>>> +	indices = min_t(unsigned int, indices, MAX_TX_QUEUES);
>>>>> +	netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
>>>>>  	if (!netdev) {
>>>>>  		err = -ENOMEM;
>>>>>  		goto err_alloc_etherdev;
>>>>>
>>>>>           
>>>> Thanks Jeff, but what is the reason for limiting to MAX_TX_QUEUES ?
>>>> Is it a hardware issue ?
>>>>
>>>>         
>>> MAX_TX_QUEUES is 128, which is the maximum the 82599 device supports in
>>> hardware (82598 supports 32 Tx queues).  I'm not sure why you'd ever
>>> want to have more Tx queues than what you have in the network device.
>>>       
>> I was not sure MAX_TX_QUEUES capping was still necessary after the
>> block :
>>
>> if (ii->mac == ixgbe_mac_82598EB)
>>           indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
>> else
>>           indices = min_t(unsigned int, indices,
>> IXGBE_MAX_FDIR_INDICES);
>>
>> indices = max_t(unsigned int, indices, IXGBE_MAX_DCB_INDICES);
>> #ifdef IXGBE_FCOE
>> 	indices += min_t(unsigned int, num_possible_cpus(),
>>                     IXGBE_MAX_FCOE_INDICES);
>> #endif
>>
>> So I asked to be sure that MAX_TX_QUEUES was not a leftover from the
>> previous default allocation.
>>
>> Thanks
>>     
>
> I see what you're getting at now.  The most we could have from this
> codepath is 72 indices for 82599, and 24 for 82598, so yeah, this is
> probably unneeded.
>
> We can get the patch cleaned up.
>
> Cheers,
> -PJ
>
>   
Thanks for catching this.  I'll submit another patch to clean this up.  
Additionally, we can probably remove MAX_TX_QUEUES all together and use 
the above values instead.  No reason to have a tx_ring array larger then 
we are ever going to use.

thanks
john.