[openib-general] RFC: detecting duplicate MAD requests

[Hal Rosenstock]

On Tue, 2006-06-13 at 00:40, Michael S. Tsirkin wrote:
> Quoting r. Hal Rosenstock <halr at voltaire.com>:
> > > Without some sort of restriction, a userspace app that's slow to pull receive 
> > > MADs from the kernel would result in consuming a large amount of kernel memory.
> > 
> > Understood but dropping a MAD after acknowledging also seems like a bad
> > thing to me.
> 
> True. Maybe we can find a way to avoid acknowledging the MAD?

There are architected ways to do that. There's busy for MADs which could
be used for some MADs. For RMPP, would the transfer be ABORTed ? I don't
think you can switch to BUSY in the middle (but I'm not 100% sure). I
don't know how this limit is being used exactly, but it might be best if
the RMPP receive were treated as 1 MAD regardless of of how many
segments it was.

-- Hal

> > Couldn't this be controlled on the request side (assuming
> > the request has a response as opposed to unsolicited sends/receives) ?
> 
> Sounds like the wrong thing to do.