Re: NAPI documentation needed

[Ben Hutchings <bhutchings@solarflare.com>]

On Thu, 2012-12-20 at 12:04 -0800, Eric Dumazet wrote:
> On Thu, 2012-12-20 at 20:39 +0100, Rafał Miłecki wrote:
> > I wanted to report some problem I've encouraged during bgmac driver development.
> > 
> > At the very beginning I've implemented IRQ using threaded IRQ
> > (request_threaded_irq). I didn't know about NAPI until someone pointed
> > me that mistake. So I decided to rewrite IRQs handling to use NAPI.
> > I've found following documents:
> > http://www.linuxfoundation.org/collaborate/workgroups/networking/napi
> > ftp://robur.slu.se/pub/Linux/net-development/NAPI/README
> > ftp://robur.slu.se/pub/Linux/net-development/NAPI/NAPI_HOWTO.txt
> > ftp://robur.slu.se/pub/Linux/net-development/NAPI/converting-to-NAPI.txt~
> > but nothing really official sitting in kernel's Documentation dir.
> > 
> > So I started to using found documents, but then noticed they are quite outdated.
> > 
> > 1) We don't have netif_rx_schedule and netif_rx_complete anymore.
> > 2) We don't set poll and weight manually anymore but use netif_napi_add
> > 3) Return type and arguments has changed in poll. None of the
> > following is up-to-date:
> > static void my_poll (struct net_device *dev, int *budget)
> > int (*poll)(struct net_device *dev, int *budget);
> > 
> > It would be great if someone with NAPI knowledge could document it in
> > a kernel. Would be really helpful for new network drivers developers.
> 
> I think you might be the one to update/create the official NAPI
> documentation, now ideas are clear for you.
> 
> That would be really great indeed.

It would.  Last time someone asked, my answer was:

> The initial change to napi_struct is explained in
> <http://lwn.net/Articles/244640/>.
> 
> Since then there have been further changes:
> 
> - netif_napi_del() has been added.  You must call it to clean up NAPI
> contexts before freeing the associated net device(s).
> 
> - Instead of netif_rx_schedule(), netif_rx_complete(), etc. you must use
> napi_schedule(), napi_complete() etc. which just take a napi_struct
> pointer.

Ben.

-- 
Ben Hutchings, Staff Engineer, Solarflare
Not speaking for my employer; that's the marketing department's job.
They asked us to note that Solarflare product names are trademarked.