On Wed December 8 2010 18:33:23 Neo Tida wrote: > Bruno, that was great. It worked like a charm. > Thank you very much. You really helped me. My speed is a little less than > madwifi, but that's great.
Glad i could help. Please ask the openwrt guys to include this patch. I did already 2 times and got ignored. We have that really well tested here, and the company i work for uses it in their product. > About madwifi and 40Mbps, yes that was weird for me too. But my receive > speed is 40Mbps ( I can see it in my task manager too ) and send speed is > exactly 33Mbps. How can you receive faster than you can send something? ;) > Now with ath5k and your patch, receive speed is about > 27Mbps and send speed is about 30Mbps. > Is there any other trick to make it better? None that i know of. bruno > > > ________________________________ > From: Bruno Randolf <b...@einfach.org> > To: ath5k-devel@lists.ath5k.org > Cc: Neo Tida <neo.t...@yahoo.com>; n...@openwrt.org > Sent: Wed, December 8, 2010 4:34:55 AM > Subject: Re: [ath5k-devel] ath5k low speed with ar5414 based cards > > On Tue December 7 2010 22:18:20 Neo Tida wrote: > > I have a problem with ath5k. I can't get more than 14 Mbps with ath5k, > > but with madwifi and exactly that platform, I can get about 40 Mbps. > > For more info, I use openwrt(trunk) on two routerstation pro platform > > with ar5414 based mini pci card in 11a mode. > > > > As I can get a great performance with madwifi, it's not a hw problem. > > 40Mbps??? Sounds incredible. I've never seen more than 33Mbps with either > madwifi or ath5k. But I can consistently get 33Mbps with ath5k on the same > chipset and platform. > > Oh - just remembering: there is a patch in OpenWRT, which limits DMA bust > sizes to work around a HW problem, but it's too conservative. I have sent a > patch to them but they didn't apply it. That's the change you have to make: > > commit dacc09a722cc0064b299b59deca1806d93d661db > Author: Bruno Randolf <b...@einfach.org> > Date: Tue Oct 19 12:22:57 2010 +0900 > > mac80211: improve ar71xx PCI bug WAR > > Since this PCI bug affects RX only, we can keep the 128B DMA size for > TX. > > We have been using this setting since month without problems and a > significant > performance improvement. > > diff --git > a/package/mac80211/patches/201-ath5k-WAR-for-AR71xx-PCI-bug.patch > b/package/mac80211/pa > index 38f56ed..a0d757c 100644 > --- a/package/mac80211/patches/201-ath5k-WAR-for-AR71xx-PCI-bug.patch > +++ b/package/mac80211/patches/201-ath5k-WAR-for-AR71xx-PCI-bug.patch > @@ -12,7 +12,7 @@ > +#else > + /* WAR for AR71xx PCI bug */ > + AR5K_REG_WRITE_BITS(ah, AR5K_TXCFG, > -+ AR5K_TXCFG_SDMAMR, AR5K_DMASIZE_4B); > ++ AR5K_TXCFG_SDMAMR, AR5K_DMASIZE_128B); > + AR5K_REG_WRITE_BITS(ah, AR5K_RXCFG, > + AR5K_RXCFG_SDMAMW, AR5K_DMASIZE_4B); > > > +#endif > > > @@ -30,7 +30,7 @@ > > > { AR5K_RXCFG, AR5K_DMASIZE_128B }, > > > +#else > > > + /* WAR for AR71xx PCI bug */ > > > -+ { AR5K_TXCFG, AR5K_DMASIZE_4B }, > > > ++ { AR5K_TXCFG, AR5K_DMASIZE_128B }, > > > + { AR5K_RXCFG, AR5K_DMASIZE_4B }, > > > +#endif > > > { AR5K_CFG, AR5K_INIT_CFG }, > > Hope that helps, > bruno _______________________________________________ ath5k-devel mailing list ath5k-devel@lists.ath5k.org https://lists.ath5k.org/mailman/listinfo/ath5k-devel