Date: Fri, 01 May 2015 10:47:43 -0400 From: John Baldwin <jhb@freebsd.org> To: Gleb Smirnoff <glebius@freebsd.org> Cc: src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-head@freebsd.org Subject: Re: svn commit: r282280 - in head/sys/dev: e1000 ixgbe ixl Message-ID: <2197979.EUYqekgM4M@ralph.baldwin.cx> In-Reply-To: <20150501135134.GC546@FreeBSD.org> References: <201504301823.t3UINd74073186@svn.freebsd.org> <1998053.shmPH6saZj@ralph.baldwin.cx> <20150501135134.GC546@FreeBSD.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On Friday, May 01, 2015 04:51:34 PM Gleb Smirnoff wrote: > On Fri, May 01, 2015 at 09:11:54AM -0400, John Baldwin wrote: > J> > On Thu, Apr 30, 2015 at 06:23:39PM +0000, John Baldwin wrote: > J> > J> Author: jhb > J> > J> Date: Thu Apr 30 18:23:38 2015 > J> > J> New Revision: 282280 > J> > J> URL: https://svnweb.freebsd.org/changeset/base/282280 > J> > J> > J> > J> Log: > J> > J> Various fixes to the stats in igb(4), ixgbe(4), and ixl(4). > J> > J> - Use hardware counters for ifnet stats in igb(4) when possible. This > J> > J> ensures these stats include packets that bypass the regular stack via > J> > J> netmap. > J> > J> - Don't derefence values off the end of the igb(4) VF stats structure. > J> > J> Instead, add a dedicated if_get_counter method for igb(4) VF interfaces. > J> > J> - Report missed packets on igb(4) as input queue drops rather than an > J> > J> input error. > J> > J> - Report bug_ring drop counts as output queue drops for igb(4) and ixgbe(4). > J> > J> - Export the buf_ring drop stats for individual rings via sysctl on > J> > J> ixgbe(4).^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ > J> > ^^^^^^^^^ > J> > > J> > While you are here, let me remind you about this plan: > J> > > J> > https://lists.freebsd.org/pipermail/svn-src-head/2014-October/063575.html > J> > > J> > We can prototype the API to userland now, write down a utility that uses > J> > it, or add the functionality to an existing utility. And start with Intel > J> > drivers, that seem to be most interested in extra stats. > J> > J> So the importaing thing here is that if_get_counter() is still doing > J> per-ifnet stats. The stat you underlined above is per-queue instead. > J> We well need more explicitly knowledge of queues outside of drivers > J> and in the stack itself to support a generic framework for per-queue > J> stats. > > This depends on how generic we want the API to be. Of course, we can add > an extra argument to if_get_counter(). > > However, if we don't expect the number of queues to exceed a reasonable > number of 255 :), we can fit the functionality into existing API. > We can keep the queue number in the highest 8 bits of the ift_counter > parameter. > > #define IFCOUNTER_MASK 0x00ffffff > #define IFCOUNTER_QUEUE(c) ((c) >> 24) I'd prefer that expose queues more directly and figure out how to handle per-queue stats then (e.g. do we have some sort of driver-independent structure that each ifnet has 1 or more of that maps to a queue and has driver provided methods, etc. If so you could have a driver method for queue stats). Note that I did use if_get_counter to report the per-interface stats instead of adding a new sysctl. -- John Baldwin
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?2197979.EUYqekgM4M>