Skip site navigation (1)Skip section navigation (2)
Date:      Wed, 22 Jan 2014 16:23:32 -0800
From:      Navdeep Parhar <np@FreeBSD.org>
To:        Garrett Wollman <wollman@freebsd.org>, net@freebsd.org
Subject:   Re: Use of contiguous physical memory in cxgbe driver
Message-ID:  <52E06104.70600@FreeBSD.org>
In-Reply-To: <21216.22944.314697.179039@hergotha.csail.mit.edu>
References:  <21216.22944.314697.179039@hergotha.csail.mit.edu>

next in thread | previous in thread | raw e-mail | index | archive | help
On 01/22/14 15:52, Garrett Wollman wrote:
> At this point everyone is well aware that requiring contiguous
> physical page when the hardware can do scatter-gather is a very bad
> idea.

I wouldn't put it this way.  Using buffers with size > PAGE_SIZE has its
advantages.

> I have a new server under test (running 9.2) which uses Chelsio
> rather than Intel 10G Ethernet controllers.  We fixed the Intel driver
> not to use more-than-page-sized jumbo mbufs.  Can anyone say with
> certainty whether the Chelsio hardware actually requires
> physically-contiguous allocations for jumbo frames?

No it doesn't.  It will do a scatter DMA if the next available rx
buffer's size is less than the size of the incoming frame.

> Has this already
> been fixed in a more recent driver?  (The hardware is identified
> specifically as a T420-CR.)

I'm not sure there's anything to fix here but I can add a knob that
would limit the driver to a maximum of PAGE_SIZE sized rx buffers.

Regards,
Navdeep



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?52E06104.70600>