[ofa-general] [PATCH 14/16] ib/ipoib: Support modifying IPOIB CQ moderation params

[Eli Cohen]

On Sun, 2008-01-20 at 12:13 +0200, Or Gerlitz wrote:
> Eli Cohen wrote:
> > On Thu, 2008-01-17 at 15:22 +0200, Or Gerlitz wrote:
> >> Eli Cohen wrote:
> 
> >> IPoIB has one CQ. As I see it, this means that you should either let the 
> >> user specify only one of rx or tx coalescing params, or make sure that 
> >> the user did not provide something that the driver can not deploy, eg 
> >> rx_usecs 12 tx_usecs 1
> 
> > I think making sure you can't provide two different values for rx and tx
> > will do.
> 
> ok.
> 
> Anyway, note that with your planned re-separation of the rx and tx CQs, 
> with no interrupts used for the tx CQ, you would need to remove all the 
> control given to the user on tx interrupt moderation... so if you will 
> not let them control tx from day one your life would be easier here, no?
> 
Thinking it over , if I want to make the code correct as it is now (e.g.
a single cq for both tx and rx), I must require that the user supply
both tx and rx values in each invocation of ethtool and that these
values be identical. Alternatively we can say that rx dictates
configuration for both rx and tx.
When we get separate tx and rx CQs we I will modify this patch to
exclude tx configuration.