[Gta04-owner] Host AP mode
neilb at suse.de
Thu May 31 03:00:44 CEST 2012
On Wed, 30 May 2012 08:23:49 +0200 Andreas Kemnade <andreas at kemnade.info>
> On Tue, 29 May 2012 23:11:01 +0200
> Christoph Mair <christoph.mair at gmail.com> wrote:
> > > But my earlier mentioned wifi performance problem with 3.x, which does
> > > not exist in 2.6.32, is not better with the libertas_tf driver.
> > > Could the SD/MMC host driver be the bottleneck?
> > I tried to debug the MMC interface and found out that the chip
> > interface runs at 25MHz. I've read somewhere that the module should
> > support up to 50MHz but I could not convince the module to use this
> > speed. Anyways, 25MHz with a 4 bits wide interface should be enough
> > for everyone [TM].
> Hmm, that transfers the 640KByte, which are enough for everyone in
> approx 50ms. Maybe people can live with that.
> Does the data get collected from there in time? Are interrupts
> correctly handled?
> > Unfortunately I could not find out why the chip does not use higher
> > data rates than 5MBit/s. I've seen very short periods where iwconfig
> > displayed higher data rates (up to 48MBits/s) but the value returned
> > immediately to 5MBits/s when I called iwconfig again (a few times).
> The libertastf driver adjusts the bit rates to higher values but
> that does not lead to higher data rates.
> But why does the standard wifi driver in kernel 2.6.32 does not
> have the problem? Does the improved power management in 3.x play a
> role here?
> The most interesting question is probably when the problem did start.
> We know:
> 2.6.32: no problem
> 3.2, 3.3, 3.4: problem
> Is that problem only with GTA04A4+Kernel3.x or also with GTA04A3+Kernel3.x?
> What about the first 3.x kernels?
> How to bisect that problem? Do some minimal gta04 porting for the kernels,
> so that only the most important devices work?
> Andreas Kemnade
I can confirm that I see the same slowdown.
With 2.6.32, it reports a Bit Rate of 24Mb/s and a link quality of 39/100,
and I get about 1.1 to 1.4 MB/s throughput.
With 3.4, it reports 5.5Mb/sec (and won't let me change it) with a link
quality of 63/70 (!) and I get about 150 kB/s throughput.
So it does look like something is broken in 3.4.
The difference between the hw kernel and 3.4 is
34 files changed, 5633 insertions(+), 3319 deletions(-)
which is quite a lot:
./drivers/net/wireless/libertas/cfg.c | 2142 ++++++++++++++++++++-
Maybe scan and assoc were merged into cfg ??
The difference between vanilla 2.6.32 and the hw kernel are about as big:
35 files changed, 3888 insertions(+), 4293 deletions(-)
Between vanilla 2.6.32 and 3.4 is
38 files changed, 7449 insertions(+), 11722 deletions(-)
So lots of changes all over the place.
I don't think there is any real chance of usefully bisecting this.
One possible approach would be to revert all the libertas changes
from hw and see if that compiles, runs, and is slow. If it was,
you could then add all the changes back in a new branch and run
git-bisect on that to see where it gets fixes.
I don't have a lot of hope for this though.
You would probably be better off posting a question to some relevant mailing list.
Daniel Drake <dsd at laptop.org> seems to be the main contributor, so 'cc' him.
Dan Williams <dcbw at redhat.com> is listed as the maintainer.
libertas-dev at lists.infradead.org and linux-wireless at vger.kernel.org
seem to be the relevant email lists.
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 828 bytes
Desc: not available
More information about the Gta04-owner