Date: Tue, 19 May 2020 11:02:27 -0400 From: Andrew Gallatin <gallatin@cs.duke.edu> To: Kristof Provost <kp@FreeBSD.org>, Eric Joyner <erj@FreeBSD.org>, Jacob Keller <jacob.e.keller@intel.com>, shurd@FreeBSD.org Cc: src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-head@freebsd.org Subject: Re: svn commit: r347418 - head/sys/net Message-ID: <c954afea-b96d-a2c3-7037-4a419d23be86@cs.duke.edu> In-Reply-To: <3BBFB371-EA44-4EE9-8A55-542CDE273CC4@FreeBSD.org> References: <201905100041.x4A0fhNT083122@repo.freebsd.org> <3BBFB371-EA44-4EE9-8A55-542CDE273CC4@FreeBSD.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On 2020-05-19 04:21, Kristof Provost wrote:
> The if_bnxt driver initialises |.isc_nrxd_max = {INT32_MAX, INT32_MAX,
> INT32_MAX},|, so presumably that’s the cause.
> I don’t know what a sane value would be though. I’ve defaulted to 4096
> (because that’s what some other iflib users seems to do) for now, and
> that seems to work. It doesn’t panic and I can get traffic through it at
> least:
You seem to be setting the max, not the default, and 4K max descriptors
on a 100g device is going to basically cripple it.
How about setting to the next power of 2 below max int so as to keep
with the authors intent?
If we don't already have a macro, something like (INT32_MAX >> 1) + 1
Drew
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?c954afea-b96d-a2c3-7037-4a419d23be86>