Date: Tue, 09 May 2017 13:59:24 -0700 From: John Baldwin <jhb@freebsd.org> To: Colin Percival <cperciva@tarsnap.com> Cc: Roger Pau =?ISO-8859-1?Q?Monn=E9?= <royger@freebsd.org>, src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-head@freebsd.org Subject: Re: svn commit: r301198 - head/sys/dev/xen/netfront Message-ID: <1571628.UQhQsSBvqn@ralph.baldwin.cx> In-Reply-To: <0100015beeed2f38-6246d7db-7f23-4b8a-ba50-b97ec0953457-000000@email.amazonses.com> References: <201606021116.u52BGajD047287@repo.freebsd.org> <20170509100912.h3ylwugahvfi5cc2@dhcp-3-128.uk.xensource.com> <0100015beeed2f38-6246d7db-7f23-4b8a-ba50-b97ec0953457-000000@email.amazonses.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Tuesday, May 09, 2017 08:36:13 PM Colin Percival wrote: > On 05/09/17 03:09, Roger Pau Monn=EF=BF=BD wrote: > > On Wed, May 03, 2017 at 05:13:40AM +0000, Colin Percival wrote: > >> On 06/02/16 04:16, Roger Pau Monn=EF=BF=BD wrote: > >>> Author: royger > >>> Date: Thu Jun 2 11:16:35 2016 > >>> New Revision: 301198 > >>> URL: https://svnweb.freebsd.org/changeset/base/301198 > >> > >> I think this commit is responsible for panics I'm seeing in EC2 on= T2 family > >> instances. [...] > >> but under high traffic volumes I think a separate thread can alrea= dy be > >> running in xn_rxeof, having dropped the RX lock while it passes a = packet > >> up the stack. This would result in two different threads trying t= o process > >> the same set of responses from the ring, with (unsurprisingly) bad= results. > >=20 > > Hm, right, xn_rxeof drops the lock while pushing the packet up the = stack. > > There's a "XXX" comment on top of that, could you try to remove the= lock > > dripping and see what happens? > >=20 > > I'm not sure there's any reason to drop the lock here, I very much = doubt > > if_input is going to sleep. >=20 > Judging by > $ grep -R -B 1 -A 1 if_input /usr/src/sys/dev > I'm pretty sure that we do indeed need to drop the lock. If it's pos= sible > to enter if_input while holding locks, there are a *lot* of network i= nterface > drivers which are dropping locks unnecessarily... It depends on how the locks are used. If a NIC driver uses a single lo= ck for both TX and RX, then it needs to drop the lock as on the TX side, if_transmit/if_start will be invoked with various network stack locks h= eld. However, if a driver uses separate locks for RX and TX and doesn't acqu= ire the RX lock while holding a TX lock, then it should be safe to hold the= lock across if_input. A lot of the older 100Mbps PCI NIC drivers only use a single lock and t= hus need to drop the lock. Many multiqueue NIC drivers use separate locks however and probably don't need to drop them around if_input. --=20 John Baldwin
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?1571628.UQhQsSBvqn>