Re: [ath5k-devel] ACK timeout
Sounds like a plan --- but you should really ask/propose in linux-wireless, as there are way more developers hanging out there than here. I sent it today, in case you're not subscribed there: http://article.gmane.org/gmane.linux.kernel.wireless.general/43648 Ah, and make the measure metric, not fields, yards or inch or whatever :-) Of course, I'm from Europe too :o) Lukas Turek signature.asc Description: This is a digitally signed message part. ___ ath5k-devel mailing list ath5k-devel@lists.ath5k.org https://lists.ath5k.org/mailman/listinfo/ath5k-devel
Re: [ath5k-devel] ACK timeout
On 23.11.2009 08:45 you wrote: I see it differently. Debugfs is for debugging, not for API. If you'd use debugfs for it, then you'd force every embedded developer to compile debugfs into the kernel. Ugly. I agree, but at least it would be some solution for us. We can't afford to maintain patches to ath5k, mac80211 and iw forever. Better IMHO is a module parameter if only one hardware / driver supports this. It couldn't be a module parameter, because it needs to be different for every device. And it would be best if it was tunable at runtime, like it was in Madwifi (via sysctl). Or really go the way via iw, nl80211, cfg80211 etc if several drivers support this feature (now or in future). What about providing a higher abstraction, a distance parameter instead of acktimeout, csttimeout and slottime? For example 'iw phy0 set distance 3000'. Definitely at least ath9k hardware supports this feature. Lukas Turek signature.asc Description: This is a digitally signed message part. ___ ath5k-devel mailing list ath5k-devel@lists.ath5k.org https://lists.ath5k.org/mailman/listinfo/ath5k-devel
Re: [ath5k-devel] ACK timeout
On Sun, Nov 22, 2009 at 11:10:41AM -0800, Felix Fietkau wrote: Luis R. Rodriguez wrote: On Sat, Nov 21, 2009 at 2:14 PM, Lukáš Turek 8...@praha12.net wrote: Hi, it seems Ath5k is missing a Madwifi feature that is essential for our long-distance links: setting ACK timeout value based on the link distance. I found a function ath5k_hw_set_ack_timeout, but it't not called from anywhere. So only the most low-level part is there, the rest and the whole userspace API is missing. What's the best way to add it? I thought about adding it to iw, like iw phy0 set acktimeout, but it would require a change on many layers (iw, nl80211, mac80211 and finally ath5k). Do you think such changes would be accepted? No, not all hardware can change this, we also don't want to clobber iw with hw specific knobs. Actually a lot of hw can set this, just the drivers haven't implemented it yet. Oh OK, thanks for clarifying that. I know that at least Atheros, Broadcom and Ralink hw can set this, so I think we should have an API for nl80211 with iw support. Probably overkill for now still. It could be easily exposed by making it possible to change the Coverage Class in the Country IE, which contains an 'index into air propagation time'. I'm not familiar with coverage class stuff, but since the Country IE is built on hostapd I take it this would be hostapd doing the tweaking back through nl80211 to the driver. Sound right? Luis ___ ath5k-devel mailing list ath5k-devel@lists.ath5k.org https://lists.ath5k.org/mailman/listinfo/ath5k-devel
Re: [ath5k-devel] ACK timeout
On Sun, Nov 22, 2009 at 11:45:27PM -0800, Holger Schurig wrote: On Sunday 22 November 2009 01:36:43 Luis R. Rodriguez wrote: No, not all hardware can change this, we also don't want to clobber iw with hw specific knobs. I see it differently. Debugfs is for debugging, not for API. Well sure. If you'd use debugfs for it, then you'd force every embedded developer to compile debugfs into the kernel. Ugly. Agreed. Better IMHO is a module parameter if only one hardware / driver supports this. Please no, no module parameters! Or really go the way via iw, nl80211, cfg80211 etc if several drivers support this feature (now or in future). nl80211 is really where things should be developed first if we can find a common ground amogst drivers and if there indeed a good use to support a knob. That way we'll also have consistency between feature setup, even if you use different hardware. Sure but debugfs is still a good place to put a knob if you want it accepted seen without much effort and would give some users immediate access to it. Once something gets ironed out through nl80211 we could consider adding a configfs tree for the nl80211 knobs we can expose through a filesystem. I had worked on one way back to add first the country selection but we determined it would be better to first do stuff through netlink (nl80211) than to put effort and support a file system approach. If we can somehow automatically expose the nl80211 command set through configs it would be even nicer but that would require a lot of work. Anyway -- IMHO configs would be the API way to define something properly for through a filesystem for the wireless subsystem. It is questionable though whether a configs solution would provide any size benefit over supporting nl80211 and compiling your own libnl (which I know Felix even had some small port of it). So for now I think nl80211 + debugfs are more than enough, if you want more I'd suggest a configfs interface but don't think that'll buy you much but an API and if you're trying to define something through configfs it likely should be exposed through nl80211 anyway. Luis ___ ath5k-devel mailing list ath5k-devel@lists.ath5k.org https://lists.ath5k.org/mailman/listinfo/ath5k-devel
Re: [ath5k-devel] ACK timeout
Luis R. Rodriguez wrote: On Sat, Nov 21, 2009 at 2:14 PM, Lukáš Turek 8...@praha12.net wrote: Hi, it seems Ath5k is missing a Madwifi feature that is essential for our long-distance links: setting ACK timeout value based on the link distance. I found a function ath5k_hw_set_ack_timeout, but it't not called from anywhere. So only the most low-level part is there, the rest and the whole userspace API is missing. What's the best way to add it? I thought about adding it to iw, like iw phy0 set acktimeout, but it would require a change on many layers (iw, nl80211, mac80211 and finally ath5k). Do you think such changes would be accepted? No, not all hardware can change this, we also don't want to clobber iw with hw specific knobs. Actually a lot of hw can set this, just the drivers haven't implemented it yet. I know that at least Atheros, Broadcom and Ralink hw can set this, so I think we should have an API for nl80211 with iw support. It could be easily exposed by making it possible to change the Coverage Class in the Country IE, which contains an 'index into air propagation time'. - Felix ___ ath5k-devel mailing list ath5k-devel@lists.ath5k.org https://lists.ath5k.org/mailman/listinfo/ath5k-devel