[ath9k-devel] Performance problems with the AR9287

[ath9k-devel] Performance problems with the AR9287

[Mitch Davis]

Hello,

I am writing to you all on behalf of the many Fedora users with AR9287
chips.  In Fedora 15, wireless performance with the AR9287 is
terrible.  Here's a bug about it:

  https://bugzilla.redhat.com/show_bug.cgi?id=702722

It's now on Fedora's "common bugs" list:

  http://fedoraproject.org/wiki/Common_F15_bugs#Poor_transfer_speeds_with_Atheros-based_wireless_chipsets

I have tried compat-wireless and compat-wireless-next (see comment 15)
and neither gives any improvement.

This chip works well with the driver that was shipped in Fedora 14
(kernel 2.6.35).

Can someone help me find the problem please?  I am happy to do
debugging and information collection, but I'll need some hand-holding.

Thank you,

Mitch.

[ath9k-devel] Performance problems with the AR9287

[Adrian Chadd]

Hi,

I'd start debugging it by looking at the various debugfs entries in
ath9k and try to find some error counters that are "high".

I ported the AR9287 code from ath9k to FreeBSD (minus some code that's
never ever called) and performance for me is just dandy.
(Or at least was when I last checked.)

If disabling encryption helps, there's likely some bugs lying around
in encryption/decryption handling which I haven't encountered in
FreeBSD.

I'd suggest looking at the debugfs entries when there's significant,
reproducible packet loss (and not just "it's slow", as that isn't
nearly as easily debuggable) to see if you can find counters that
increment whenever a packet is dropped. Once you've established that,
we can hopefully do some further digging and find what in particular
is messed up.

Thanks,


Adrian


On 20 October 2011 06:26, Mitch Davis <mjd+ath9k.org@afork.com> wrote:
> Hello,
>
> I am writing to you all on behalf of the many Fedora users with AR9287
> chips. ?In Fedora 15, wireless performance with the AR9287 is
> terrible. ?Here's a bug about it:
>
> ?https://bugzilla.redhat.com/show_bug.cgi?id=702722
>
> It's now on Fedora's "common bugs" list:
>
> ?http://fedoraproject.org/wiki/Common_F15_bugs#Poor_transfer_speeds_with_Atheros-based_wireless_chipsets
>
> I have tried compat-wireless and compat-wireless-next (see comment 15)
> and neither gives any improvement.
>
> This chip works well with the driver that was shipped in Fedora 14
> (kernel 2.6.35).
>
> Can someone help me find the problem please? ?I am happy to do
> debugging and information collection, but I'll need some hand-holding.
>
> Thank you,
>
> Mitch.
> _______________________________________________
> ath9k-devel mailing list
> ath9k-devel at lists.ath9k.org
> https://lists.ath9k.org/mailman/listinfo/ath9k-devel
>

[ath9k-devel] Performance problems with the AR9287

[Adrian Chadd]

.. and since the driver in 2.6.35 works fine, you have a starting
point for bisecting the driver code. :-)
That will _definitely_ get you a fix.


Adrian

[ath9k-devel] Performance problems with the AR9287

[Peter Stuge]

Mitch Davis wrote:
> Can someone help me find the problem please?  I am happy to do
> debugging and information collection, but I'll need some
> hand-holding.

Sorry to be the bearer of bad news, but I'm afraid the answer is no.
You basically have to find the problem on your own. Of course patches
that fix problems which you find are welcome, but you can expect zero
hand-holding.


Adrian Chadd wrote:
> a starting point for bisecting the driver code. :-)
> That will _definitely_ get you a fix.

I disagree strongly. I would not at all expect that simply finding a
commit which exposes broken behavior will result in a fix. I find it
equally likely, or more likely, that a bad commit (if one can be
found at all!) will not be simple enough for even the skilled ath9k
developers to immediately see what the problem is.

To be clear, so that noone misunderstands - I am not claiming, and
have never claimed, that driver debugging is easy.


//Peter

[ath9k-devel] Performance problems with the AR9287

[Mitch Davis]

On Thu, Oct 20, 2011 at 1:02 PM, Peter Stuge <peter@stuge.se> wrote:
>
> I disagree strongly. I would not at all expect that simply finding a
> commit which exposes broken behavior will result in a fix. I find it
> equally likely, or more likely, that a bad commit (if one can be
> found at all!) will not be simple enough for even the skilled ath9k
> developers to immediately see what the problem is.

I thought so, and I agree.  This particular problem could have been
broken and fixed many times, and the odds of me finding the relevant
one isn't good.

> You basically have to find the problem on your own.

Can you recommend a basic approach?  Are Adrian's comments about the
entries in debugfs useful?  What should I be looking for?

Mitch.

[ath9k-devel] Performance problems with the AR9287

[Peter Stuge]

Mitch Davis wrote:
> > I disagree strongly. I would not at all expect that simply finding a
> > commit which exposes broken behavior will result in a fix.
..
> I thought so, and I agree.  This particular problem could have been
> broken and fixed many times, and the odds of me finding the
> relevant one isn't good.

On the other hand, maybe the bisect is easy enough to do, and it
*might* be enough to identify the problem.


> > You basically have to find the problem on your own.
> 
> Can you recommend a basic approach?

I'm afraid not. I'm also not an ath9k developer. If I would have the
same issues I expect I would spend a week or so on studying the
driver code and trying to reverse engineer the hardware.


> Are Adrian's comments about the entries in debugfs useful?

Yes!


> What should I be looking for?

I really don't know. Apparently the driver code quality is now good,
and there are no issues. I'm really happy that I haven't had to use
it for a while; issues that I might have run into together with sad
state of affairs on this mailing list would just make me too mad.


//Peter

[ath9k-devel] Performance problems with the AR9287

[Adrian Chadd]

On 20 October 2011 10:20, Mitch Davis <mjd+ath9k.org@afork.com> wrote:

> I thought so, and I agree. ?This particular problem could have been
> broken and fixed many times, and the odds of me finding the relevant
> one isn't good.

You don't have to find the relevant one. You just have to find one
example of it breaking.



Adrian

[ath9k-devel] Performance problems with the AR9287

[Ben Greear]

On 10/19/2011 03:26 PM, Mitch Davis wrote:
> Hello,
>
> I am writing to you all on behalf of the many Fedora users with AR9287
> chips.  In Fedora 15, wireless performance with the AR9287 is
> terrible.  Here's a bug about it:
>
>    https://bugzilla.redhat.com/show_bug.cgi?id=702722
>
> It's now on Fedora's "common bugs" list:
>
>    http://fedoraproject.org/wiki/Common_F15_bugs#Poor_transfer_speeds_with_Atheros-based_wireless_chipsets
>
> I have tried compat-wireless and compat-wireless-next (see comment 15)
> and neither gives any improvement.
>
> This chip works well with the driver that was shipped in Fedora 14
> (kernel 2.6.35).
>
> Can someone help me find the problem please?  I am happy to do
> debugging and information collection, but I'll need some hand-holding.

 From reading that report, it seems users of ath5k had issues as well,
and that upgrading to the latest wireless code fixed issues.

It's not clear at all that it's always a certain chipset.

Maybe you could get lspci info from anyone reporting these problems,
as well as confirm that the problem is fixed or not by using the
latest wireless-compat code?

For what it's worth, we see fine performance on ath9k chipsets
with the 3.0 kernel, though we are using 9380 chips mostly.

Thanks,
Ben

-- 
Ben Greear <greearb@candelatech.com>
Candela Technologies Inc  http://www.candelatech.com

[ath9k-devel] Performance problems with the AR9287

[Mohammed Shafi]

On Thu, Oct 20, 2011 at 3:56 AM, Mitch Davis <mjd+ath9k.org@afork.com> wrote:
> Hello,
>
> I am writing to you all on behalf of the many Fedora users with AR9287
> chips. ?In Fedora 15, wireless performance with the AR9287 is
> terrible. ?Here's a bug about it:
>
> ?https://bugzilla.redhat.com/show_bug.cgi?id=702722
>
> It's now on Fedora's "common bugs" list:
>
> ?http://fedoraproject.org/wiki/Common_F15_bugs#Poor_transfer_speeds_with_Atheros-based_wireless_chipsets
>
> I have tried compat-wireless and compat-wireless-next (see comment 15)
> and neither gives any improvement.
>
> This chip works well with the driver that was shipped in Fedora 14
> (kernel 2.6.35).
>
> Can someone help me find the problem please? ?I am happy to do
> debugging and information collection, but I'll need some hand-holding.

as Ben, kindly verify this issue can be recreate in the latest compat wireless
or wireless-testing(it would take time to compile the kernel).
http://linuxwireless.org/en/developers/Documentation/git-guide
i will try my level tbest o badly get some time to recreate the issue
in a clean environment.

>
> Thank you,
>
> Mitch.
> _______________________________________________
> ath9k-devel mailing list
> ath9k-devel at lists.ath9k.org
> https://lists.ath9k.org/mailman/listinfo/ath9k-devel
>



-- 
shafi