Date: Wed, 17 Nov 2010 18:54:47 +0600 From: Eugene Grosbein <eugen@grosbein.pp.ru> To: freebsd-net@freebsd.org Cc: bsdlists@cabstand.com Subject: Re: request for MFC of em/igb drivers Message-ID: <4CE3D097.7030204@grosbein.pp.ru> In-Reply-To: <20101117070422.GA45678@cabstand.com> References: <20101117070422.GA45678@cabstand.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On 17.11.2010 13:04, Chris Peiffer wrote: > > Hi, > > I've been watching the traffic here over the last few months relating > to the em and igb Intel ethernet drivers. It seems like there's a big > consensus that HEAD has some good new fixes. > > Looking at this: > > http://www.freebsd.org/cgi/cvsweb.cgi/src/sys/dev/e1000/if_em.c > > There's a pretty big commit to HEAD in rev 1.58 on Sep 27 that's > marked "MFC: a week" but it doesn't look like anything's been MFC'd > since. (And several other revisions that look more experimental have > since gone into HEAD.) > > We've been seeing some weird issues with em devices under high load > and if these changes in 1.58 are ready for STABLE we'd love to test > them live. > > So this is a beg... if the relevant committers are out here, can that > MFC go through soon? Or if it can't, or I'm reading the logs wrong, > please explain. > > Thanks in advance. Indeed, in RELENG_8 em(4) is not useable for us because of two flaws: 1. It panices system due to NULL pointers dereference. It needs the following patch taken from CURRENT's igb(4): --- if_em.c.orig 2010-11-02 15:45:56.000000000 +0600 +++ if_em.c 2010-11-08 14:24:46.000000000 +0600 @@ -4181,9 +4181,11 @@ ifp->if_ierrors++; /* Reuse loaded DMA map and just update mbuf chain */ mp = rxr->rx_buffers[i].m_head; + if(mp) { mp->m_len = mp->m_pkthdr.len = MCLBYTES; mp->m_data = mp->m_ext.ext_buf; mp->m_next = NULL; + } if (adapter->max_frame_size <= (MCLBYTES - ETHER_ALIGN)) m_adj(mp, ETHER_ALIGN); igb(4) in RELENG_8 has not this fix too and there is a PR: http://www.freebsd.org/cgi/query-pr.cgi?pr=150920 2. It makes physical link down/up for every vlan creation/deletion that leads to 3 seconds of service interruption and down/up events for all other vlans based on same parent interface. It needs the following patch that's already present in CURRENT but not in RELENG_8: @@ -4342,7 +4344,8 @@ em_shadow_vfta[index] |= (1 << bit); ++adapter->num_vlans; /* Re-init to load the changes */ - em_init(adapter); + if (ifp->if_capenable & IFCAP_VLAN_HWFILTER) + em_init(adapter); } /* @@ -4366,7 +4369,8 @@ em_shadow_vfta[index] &= ~(1 << bit); --adapter->num_vlans; /* Re-init to load the changes */ - em_init(adapter); + if (ifp->if_capenable & IFCAP_VLAN_HWFILTER) + em_init(adapter); } static void
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?4CE3D097.7030204>