From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from [140.186.70.92] (port=36655 helo=eggs.gnu.org) by lists.gnu.org with esmtp (Exim 4.43) id 1Q0SZZ-0003rT-Oa for qemu-devel@nongnu.org; Fri, 18 Mar 2011 01:50:07 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1Q0SZY-0004hC-Cm for qemu-devel@nongnu.org; Fri, 18 Mar 2011 01:50:05 -0400 Received: from mx1.redhat.com ([209.132.183.28]:26049) by eggs.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1Q0SZY-0004h6-49 for qemu-devel@nongnu.org; Fri, 18 Mar 2011 01:50:04 -0400 Message-ID: <4D82F29F.8050902@redhat.com> Date: Fri, 18 Mar 2011 13:50:23 +0800 From: Jason Wang MIME-Version: 1.0 Subject: Re: [Qemu-devel] [PATCH v6] rtl8139: add vlan support References: <1299803757-26812-1-git-send-email-benjamin.poirier@gmail.com> In-Reply-To: <1299803757-26812-1-git-send-email-benjamin.poirier@gmail.com> Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit List-Id: qemu-devel.nongnu.org List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Benjamin Poirier Cc: qemu-devel@nongnu.org On 03/11/2011 08:35 AM, Benjamin Poirier wrote: > Here is version 6 of my patchset to add vlan support to the emulated rtl8139 > nic. > > Changes since v5: > * moved all receive changes to "add vlan tag extraction" > * fixed checkpatch.pl style issues > * fixed bugs in receive case related to small buffers and loopback > mode. Moved "too small buffer" code back where it used to be, though > it is changed in content. > > Changes since v4: > * removed alloca(), for real. Thanks to the reviewers for their > patience. This patchset now has more versions than the vlan header > has bytes! > * corrected the unlikely, debug printf and long lines, as per comments > * cleaned out ifdef's pertaining to ethernet checksum calculation. > According to a comment since removed they were related to an > "optimization": > > RTL8139 provides frame CRC with received packet, this feature > > seems to be ignored by most drivers, disabled by default > see commit ccf1d14 > > I've tested v5 using x86_64 host/guest with the usual procedure. I've also ran > the clang analyzer on the qemu code base, just for fun. > > Changes since v3: > * removed alloca() and #include as per comments > * reordered patches to put extraction before insertion. Extraction > touches only the receive path but insertion touches both. The two > patches are now needed to have vlan functionnality. > > I've tested v4 with x86_64 host/guest. I used the same testing procedure as > before. I've tested a plain configuration as well as one with tso + vlan > offload, successfully. > > I had to hack around the Linux 8139cp driver to be able to enable tso on vlan > which leads me to wonder, can someone with access to the C+ spec or a real > card confirm that it can do tso and vlan offload at the same time? The patch > I used for the kernel is at https://gist.github.com/851895. 8139cp does have the support for VLAN with TSO. See IFCAP_VLAN_HWTSO capabilities in FreeBSD re(4) driver: http://svn.freebsd.org/viewvc/base/release/8.2.0/sys/dev/re/if_re.c?revision=218742&view=markup. You can use it in your testing for both vlan tso and loopback TX. Additionally, you can submit your 8139 modifications to linux upstream I think. > Changes since v2: > insertion: > * moved insertion later in the process, to handle tso > * use qemu_sendv_packet() to insert the tag for us > * added dot1q_buf parameter to rtl8139_do_receive() to avoid some > memcpy() in loopback mode. Note that the code path through that > function is unchanged when dot1q_buf is NULL. > > extraction: > * reduced the amount of copying by moving the "frame too short" logic > after the removal of the vlan tag (as is done in e1000.c for > example). Unfortunately, that logic can no longer be shared betwen > C+ and C mode. > > I've posted v2 of these patches back in November > http://article.gmane.org/gmane.comp.emulators.qemu/84252 > > I've tested v3 on the following combinations of guest and hosts: > host: x86_64, guest: x86_64 > host: x86_64, guest: ppc32 > host: ppc32, guest: ppc32 > > Testing on the x86_64 host used '-net tap' and consisted of: > * making an http transfert on the untagged interface. > * ping -s 0-1472 to another host on a vlan. > * making an scp upload to another host on a vlan. > > Testing on the ppc32 host used '-net socket' connected to an x86_64 qemu-kvm > running the virtio nic and consisted of: > * establishing an ssh connection between the two using an untagged interface. > * ping -s 0-1472 between the two using a vlan. > * making an scp transfer in both directions using a vlan. > > All that was successful. Nevertheless, it doesn't exercise all code paths so > care is in order. > > Please note that the lack of vlan support in rtl8139 has taken a few people > aback: > https://bugzilla.redhat.com/show_bug.cgi?id=516587 > http://article.gmane.org/gmane.linux.network.general/14266 > > Thanks, > -Ben >