Date: Fri, 31 Oct 2008 09:59:22 +0900 From: PYUN Yong-Hyeon <pyunyh@gmail.com> To: Jeremy Chadwick <koitsu@FreeBSD.org> Cc: freebsd-questions@FreeBSD.org, Graham Bentley <admin@cpcnw.co.uk> Subject: Re: Marvell 88E8052 PCI-E LAN on FreeBSD 7.0 Message-ID: <20081031005921.GA82781@cdnetworks.co.kr> In-Reply-To: <20081030123815.GA4568@icarus.home.lan> References: <op.ujt418h2t4j7n6@3bsd.cpcnw.co.uk> <20081030123815.GA4568@icarus.home.lan>
next in thread | previous in thread | raw e-mail | index | archive | help
On Thu, Oct 30, 2008 at 05:38:15AM -0700, Jeremy Chadwick wrote: > On Thu, Oct 30, 2008 at 11:33:34AM -0000, Graham Bentley wrote: > > Hi ALl, > > > > I have a DFI LanParty Mobo that includes Marvells 88E8052 > > and 88E8053 LAN IF. > > > > Using the module with 7.0 [msk] the network preformance > > is terrible, Opera / Links stall, or wont page load at > > all although pings to the router are fine? > > > > I then tried using Marvells own driver the website [myk] > > and the results where about the same. > > > > Just before I was about to give up and put in my trusty > > old 3Com 3C509 [xl] I noticed that in the Marvell doco > > there where several tunable params so decided to try > > out a few. > > > > I discovered that adding -txcsum and -rxcsum (i.e. > > disabling hardware checksuming) to the ifconfig > > statement, the performance was as quick as it is > > on that other OS! I'm not sure you suffers from the same problem but there was a Tx checksum offload related bug in msk(4) driver and it was fixed in HEAD. How about applying the diff in CVS rev 1.33 of if_msk? http://www.freebsd.org/cgi/cvsweb.cgi/src/sys/dev/msk/if_msk.c.diff?r1=1.32;r2=1.33;f=h > > Yong-Hyeon Pyun can probably explain what's going on here. I've CC'd > him on this mail; he usually hangs out on -stable though. > > You need to keep something in mind here: Marvell does not give out > documentation for their cards publicly, so Yong-Hyeon has to > reverse-engineer and "tinker" with what he already knows. Some hardware > feature do not work, others are buggy, others work fine on specific > revisions of the chip while later ones break. > > And if you tell me "Well Linux has support for this chip!", I will throw > the following evidence back in your face: Marvell and other companies > are giving Linux developers development PCI cards to develop drivers > with (sometimes even before the card is out in the market), so Linux has > the upper hand here. They are not doing this with the BSDs. > > Purely as an example: in my Wiki, section "Network devices", see the > entry for the 88E8040 NIC. I'm still working with Yong-Hyeon to try to > get him access to a laptop that has this chip so he can write the > driver. > > http://wiki.freebsd.org/JeremyChadwick/Commonly_reported_issues > > This should give you some idea of the complexities. As I said, Linux > has the upper hand, because they're getting support from Marvell. > > > It seems that hw cs is on by default so I added the > > above to my ifoncfig line in /etc/rc.conf and now > > all is snappy! > > > > I was wondering how could this be the case and also > > if anyone else had this issue with Marvell chips? > > I would urge you to go out and purchase an Intel Pro/1000 PT card, which > runs for around 30-40 USD. It's good to have a spare NIC on hand > anyways -- your 3C509 xl(4) based NIC probably won't cut it, especially > if you're complaining about performance. :-) No one uses those cards > any more except individuals running on hardware from 1997, which you are > obviously not. > -- Regards, Pyun YongHyeon
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20081031005921.GA82781>