Date: Wed, 22 Nov 2017 17:19:54 +0100 From: =?UTF-8?Q?Micha=C5=82_Krawczyk?= <mk@semihalf.com> To: Gleb Smirnoff <glebius@freebsd.org> Cc: Marcin Wojtas <mw@freebsd.org>, src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-head@freebsd.org Subject: Re: svn commit: r325239 - head/sys/dev/ena Message-ID: <CAJMMOfP6tpAKv2EBNiSkwNNKCPNr2s19Ek8QkVZsANYBQiTfCg@mail.gmail.com> In-Reply-To: <20171122000816.GH1063@FreeBSD.org> References: <201710311631.v9VGVNOs058255@repo.freebsd.org> <20171122000816.GH1063@FreeBSD.org>
next in thread | previous in thread | raw e-mail | index | archive | help
Hi Gleb, We are counting the statistics in the software because of the alignment with the newest ena-com API (HAL) which is delivered to us from NIC's vendor. Regarding fetching them periodically - there are 2 issues with that. First of all, there may be problem with the counter overflow. Secondly, reading statistics on demand requires sending admin message to the NIC and then the cv_timedwait function is called to wait for the response. However there are a lot of witness warnings when the cv_timedwait is called from the if_get_counter() context, because of the non sleepable lock which is hold there (I don't remember exactly which lock was causing this issue). Best regards, Michal 2017-11-22 1:08 GMT+01:00 Gleb Smirnoff <glebius@freebsd.org>: > Hi Martin and Michal, > > On Tue, Oct 31, 2017 at 04:31:23PM +0000, Marcin Wojtas wrote: > M> Author: mw > M> Date: Tue Oct 31 16:31:23 2017 > M> New Revision: 325239 > M> URL: https://svnweb.freebsd.org/changeset/base/325239 > M> > M> Log: > M> Rework counting of hardware statistics in ENA driver > M> > M> Do not read all statistics from the device, instead count them in the > M> driver except from RX drops - they are received directly from the NIC > M> in the AENQ descriptor. > M> > M> Submitted by: Michal Krawczyk <mk@semihalf.com> > M> Reviewed by: imp > M> Obtained from: Semihalf > M> Sponsored by: Amazon.com, Inc. > M> Differential Revision: https://reviews.freebsd.org/D12852 > > Is it possible not to count them in software, as well as not fetch > them from hardware periodically, but instead just fetch them on > demand, when either if_get_counter() or a sysctl is called? > > That would be more efficient. > > -- > Gleb Smirnoff >
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?CAJMMOfP6tpAKv2EBNiSkwNNKCPNr2s19Ek8QkVZsANYBQiTfCg>