[ath9k-devel] ath9k vs nvidia(proprietary)

[ath9k-devel] ath9k vs nvidia(proprietary)

[Dmitri Seletski]

Hello Guys.

I can pretty much say, unloading nvidia driver and reloading ath9k
module fixes connection.
However, in my case it's pretty bad for me... I am heavy gamer, and SLI
nvidia cards dusting doing nothing ,well, it's pretty bad.
I suspect i may have sort of a trouble with IRQ sharing.

Unfortunately after some goggling i  didn't find any information how i
can force nvidia driver to use some other IRQ channel(which i have
available, by turning off many unused devices that use  IRQ's). Same
question goes here. Is it possible for me to give some load parameters
to ath9k module to force it to use another IRQ?(like some other drivers
do?) Obviously, there is not much point of asking Nvidia to do it, since
those behemoth's won't do anything for sure, since their current driver
(i dont use it for that matter)release is pretty banged up...

Another question, my Access Point and PCI wifi card are from same
manufacturer. AP is set to B,G,N auto mode, but my card only connects to
it using G mode, which is not why i bought it in the first place. Why is
this happening? i am pretty much only user to network myself 99% of
time. i tried to force it to use N mode, it could detect it, but it
couldn't connect.

Thanks in advance for answer!
Dmitri

[ath9k-devel] ath9k vs nvidia(proprietary)

[Pavel Roskin]

On Sun, 2009-04-05 at 10:02 +0100, Dmitri Seletski wrote:
> Hello Guys.
> 
> I can pretty much say, unloading nvidia driver and reloading ath9k
> module fixes connection.
> However, in my case it's pretty bad for me... I am heavy gamer, and SLI
> nvidia cards dusting doing nothing ,well, it's pretty bad.
> I suspect i may have sort of a trouble with IRQ sharing.

http://www.catb.org/~esr/faqs/smart-questions.html#symptoms

> Is it possible for me to give some load parameters
> to ath9k module to force it to use another IRQ?(like some other drivers
> do?)

That's not ath9k specific in any way.

>  Obviously, there is not much point of asking Nvidia to do it, since
> those behemoth's won't do anything for sure, since their current driver
> (i dont use it for that matter)release is pretty banged up...

If you do most of the work finding how to work around the problem, and
it turns out that something could be improved in free software, then
maybe your patch will be applied.

If you expect others to question you about the symptoms, then suggest
patches and push them to the kernel, that's not going to happen.

> Another question, my Access Point and PCI wifi card are from same
> manufacturer. AP is set to B,G,N auto mode, but my card only connects to
> it using G mode, which is not why i bought it in the first place. Why is
> this happening? i am pretty much only user to network myself 99% of
> time. i tried to force it to use N mode, it could detect it, but it
> couldn't connect.

I suggest that you actually show what commands you run and what they
return, not just your interpretation.

In case you are hitting the stuck scan bug, try setting the channel with
iwconfig instead of loading and unloading modules.

-- 
Regards,
Pavel Roskin

[ath9k-devel] ath9k vs nvidia(proprietary)

[Dmitri Seletski]

Pavel Roskin wrote:
> On Sun, 2009-04-05 at 10:02 +0100, Dmitri Seletski wrote:
>   
>> Hello Guys.
>>
>> I can pretty much say, unloading nvidia driver and reloading ath9k
>> module fixes connection.
>> However, in my case it's pretty bad for me... I am heavy gamer, and SLI
>> nvidia cards dusting doing nothing ,well, it's pretty bad.
>> I suspect i may have sort of a trouble with IRQ sharing.
>>     
>  
>   
I was specifically advised to unload nvidia driver previously by one of
developers. I have provided quite some information about my machine,
sorry for not keeping track of those emails, i'll try to be smarter next
time and keep history of previous emails.
>   
>> Is it possible for me to give some load parameters
>> to ath9k module to force it to use another IRQ?(like some other drivers
>> do?)
>>     
>
> That's not ath9k specific in any way.
>
>   
Arm, what do you mean not specific? That phrase makes no sense to me.

[ath9k-devel] ath9k vs nvidia(proprietary)

[Pavel Roskin]

On Mon, 2009-04-06 at 21:14 +0100, Dmitri Seletski wrote:
> Pavel Roskin wrote:
> > On Sun, 2009-04-05 at 10:02 +0100, Dmitri Seletski wrote:
> >   
> >> Hello Guys.
> >>
> >> I can pretty much say, unloading nvidia driver and reloading ath9k
> >> module fixes connection.
> >> However, in my case it's pretty bad for me... I am heavy gamer, and SLI
> >> nvidia cards dusting doing nothing ,well, it's pretty bad.
> >> I suspect i may have sort of a trouble with IRQ sharing.
> >>     
> >  
> >   
> I was specifically advised to unload nvidia driver previously by one of
> developers. I have provided quite some information about my machine,
> sorry for not keeping track of those emails, i'll try to be smarter next
> time and keep history of previous emails.

It's possible that you were asked to do that simply because the nvidia
driver is a "black box", which cannot be audited for code quality.
Developers are happier to help is such black boxes are kept away from
the system under testing.  It doesn't necessarily imply there is a
conflict between the drivers.  It's just an attempt to exclude
unpredictable inputs while testing.

> >> Is it possible for me to give some load parameters
> >> to ath9k module to force it to use another IRQ?(like some other drivers
> >> do?)
> >>     
> >
> > That's not ath9k specific in any way.
> >
> >   
> Arm, what do you mean not specific? That phrase makes no sense to me.
> From what i know not every device driver has ability of changing IRQ as
> required.

Generally, it should not be needed.  Also, IRQ assignment is not done in
individual drivers of PCI devices.  If you need it badly, perhaps the
PCI people can help (linux-pci at vger.kernel.org).  But most likely you
will be told that you are doing something wrong.

>  You can think of my previous phrase as possible feature
> request, or a question of "how to?" type. I know that borh SLI nvidia
> cards are sharing IRQ 16 as well as atheros based dlink wireless card.
> (if it's not difficult, please look into my previous emails, they will
> have dmesg and lspci -vv details, otherwise just ask me for more
> details, i shall provide them)

It should not be a problem.  You can try some workarounds, such as the
"irqpoll" option on the kernel command line.

One fun project would be to enable MSI (message signaled interrupts) in
ath9k.  It's not guaranteed to help in your situation, but it would be a
welcome contribution to the code.  It should not be not very hard.

> >>  Obviously, there is not much point of asking Nvidia to do it, since
> >> those behemoth's won't do anything for sure, since their current driver
> >> (i dont use it for that matter)release is pretty banged up...
> >>     
> >
> > If you do most of the work finding how to work around the problem, and
> > it turns out that something could be improved in free software, then
> > maybe your patch will be applied.
> >
> > If you expect others to question you about the symptoms, then suggest
> > patches and push them to the kernel, that's not going to happen.
> >   
> Symptoms are simple with nvidia module and ath9k loaded at the same time
> connection degrades over short period of time and after several minutes
> disappears all together. I suspect it could be IRQ sharing or other I/O
> or similar hardware conflict, since both of them work pretty well while
> other devices driver is not loaded. no unusual messages in dmesg.
> Machine may hang and even if in terminal(not in X), machine hangs
> completely without providing a bit of information, not even oops message.

ath9k used to hand for me with SLUB debugging enabled.  I debugged it
and submitted a patch recently, but it didn't make it to
wireless-testing yet (it looks like wireless-testing maintainers are
away for more than a week).

If you are lucky, you could find another problem in ath9k.  Then your
contribution will be welcome.  But you may find no obvious problem in
ath9k.

I could dust off an nvidia card and try ath9k with it, but I have other
priorities, such as WDS testing.

> >> Another question, my Access Point and PCI wifi card are from same
> >> manufacturer. AP is set to B,G,N auto mode, but my card only connects to
> >> it using G mode, which is not why i bought it in the first place. Why is
> >> this happening? i am pretty much only user to network myself 99% of
> >> time. i tried to force it to use N mode, it could detect it, but it
> >> couldn't connect.
> >>     
> >
> > I suggest that you actually show what commands you run and what they
> > return, not just your interpretation.
> >
> > In case you are hitting the stuck scan bug, try setting the channel with
> > iwconfig instead of loading and unloading modules.
> >
> >   
> Commands are simple as well.
> ip link set wlan0 up
> iwconfig wlan0 essid dlink
> they don't give any errors and don't show any unusual messages in dmesg
> 
> thats pretty much it. (no wireless security used at all)

I mean, it you lose connection, try "iwconfig wlan0 channel 6" (or
whatever channel the AP is on.  It will unblock whatever is stuck in the
kernel.  That's another thing I'd like to have a look at before I even
touch that nvidia card.

> N.B.
> What i really don't want is any kind of aggression towards to me.

Of course.  If you wanted aggression towards you, you would have written
to LKML :-)

> Coding something for a community, even advanced one requires certain
> amount of tolerance, as ANY normal communication.
> I am not perfect tester(i just attempt to become better one), i
> understand that, but i am not coming after difficult day, working with
> difficult people to submit difficult issues(which other people may
> prefer to ignore all together or for those reasons abandon Linux and go
> to MS products, i have seeing enough of those). I am not going to become
> more enthusiastic by being referred to "RTFM noob" like documents, who
> nearly imply my stupidity.
> So far I was very happy in this list, and even my lack of technical
> knowledge did not spoil my attitude towards this list. I want it to stay
> that way. Ignoring problems of people who lack technical knowledge on
> matter is path to nowhere. I seriously, don't want to start flame wars,
> i am not naive and I don't ask for apologies. I just ask one thing,
> please be tolerant, everyone can have a shi*** day, that includes you
> and me!

It's totally unnecessary to vent your frustration here, especially to
those who never called you a noob or something.

-- 
Regards,
Pavel Roskin

[ath9k-devel] ath9k vs nvidia(proprietary)

[Dmitri Seletski]

Pavel Roskin wrote:
> On Mon, 2009-04-06 at 21:14 +0100, Dmitri Seletski wrote:
>   
>> Pavel Roskin wrote:
>>     
>>> On Sun, 2009-04-05 at 10:02 +0100, Dmitri Seletski wrote:
>>>   
>>>       
>>>> Hello Guys.
>>>>
>>>> I can pretty much say, unloading nvidia driver and reloading ath9k
>>>> module fixes connection.
>>>> However, in my case it's pretty bad for me... I am heavy gamer, and SLI
>>>> nvidia cards dusting doing nothing ,well, it's pretty bad.
>>>> I suspect i may have sort of a trouble with IRQ sharing.
>>>>     
>>>>         
>>>  
>>>   
>>>       
>> I was specifically advised to unload nvidia driver previously by one of
>> developers. I have provided quite some information about my machine,
>> sorry for not keeping track of those emails, i'll try to be smarter next
>> time and keep history of previous emails.
>>     
>
> It's possible that you were asked to do that simply because the nvidia
> driver is a "black box", which cannot be audited for code quality.
> Developers are happier to help is such black boxes are kept away from
> the system under testing.  It doesn't necessarily imply there is a
> conflict between the drivers.  It's just an attempt to exclude
> unpredictable inputs while testing.
>   

Perfectly understandable, now i am in my X from uner VESA compiled in...
>   
>>>> Is it possible for me to give some load parameters
>>>> to ath9k module to force it to use another IRQ?(like some other drivers
>>>> do?)
>>>>     
>>>>         
>>> That's not ath9k specific in any way.
>>>
>>>   
>>>       
>> Arm, what do you mean not specific? That phrase makes no sense to me.
>> From what i know not every device driver has ability of changing IRQ as
>> required.
>>     
>
> Generally, it should not be needed.  Also, IRQ assignment is not done in
> individual drivers of PCI devices.  If you need it badly, perhaps the
> PCI people can help (linux-pci at vger.kernel.org).  But most likely you
> will be told that you are doing something wrong.
>   
Not that I am doing something wrong. i am NOT doing anything.
>   
>>  You can think of my previous phrase as possible feature
>> request, or a question of "how to?" type. I know that borh SLI nvidia
>> cards are sharing IRQ 16 as well as atheros based dlink wireless card.
>> (if it's not difficult, please look into my previous emails, they will
>> have dmesg and lspci -vv details, otherwise just ask me for more
>> details, i shall provide them)
>>     
>
> It should not be a problem.  You can try some workarounds, such as the
> "irqpoll" option on the kernel command line.
>
> One fun project would be to enable MSI (message signaled interrupts) in
> ath9k.  It's not guaranteed to help in your situation, but it would be a
> welcome contribution to the code.  It should not be not very hard.
>   
Well, how can i help? I am not coder... I dont speak C ;)
>   
>>>>  Obviously, there is not much point of asking Nvidia to do it, since
>>>> those behemoth's won't do anything for sure, since their current driver
>>>> (i dont use it for that matter)release is pretty banged up...
>>>>     
>>>>         
>>> If you do most of the work finding how to work around the problem, and
>>> it turns out that something could be improved in free software, then
>>> maybe your patch will be applied.
>>>
>>> If you expect others to question you about the symptoms, then suggest
>>> patches and push them to the kernel, that's not going to happen.
>>>   
>>>       
>> Symptoms are simple with nvidia module and ath9k loaded at the same time
>> connection degrades over short period of time and after several minutes
>> disappears all together. I suspect it could be IRQ sharing or other I/O
>> or similar hardware conflict, since both of them work pretty well while
>> other devices driver is not loaded. no unusual messages in dmesg.
>> Machine may hang and even if in terminal(not in X), machine hangs
>> completely without providing a bit of information, not even oops message.
>>     
>
> ath9k used to hand for me with SLUB debugging enabled.  I debugged it
> and submitted a patch recently, but it didn't make it to
> wireless-testing yet (it looks like wireless-testing maintainers are
> away for more than a week).
>
> If you are lucky, you could find another problem in ath9k.  Then your
> contribution will be welcome.  But you may find no obvious problem in
> ath9k.
>
> I could dust off an nvidia card and try ath9k with it, but I have other
> priorities, such as WDS testing.
>
>   
>>>> Another question, my Access Point and PCI wifi card are from same
>>>> manufacturer. AP is set to B,G,N auto mode, but my card only connects to
>>>> it using G mode, which is not why i bought it in the first place. Why is
>>>> this happening? i am pretty much only user to network myself 99% of
>>>> time. i tried to force it to use N mode, it could detect it, but it
>>>> couldn't connect.
>>>>     
>>>>         
>>> I suggest that you actually show what commands you run and what they
>>> return, not just your interpretation.
>>>
>>> In case you are hitting the stuck scan bug, try setting the channel with
>>> iwconfig instead of loading and unloading modules.
>>>
>>>   
>>>       
>> Commands are simple as well.
>> ip link set wlan0 up
>> iwconfig wlan0 essid dlink
>> they don't give any errors and don't show any unusual messages in dmesg
>>
>> thats pretty much it. (no wireless security used at all)
>>     
>
> I mean, it you lose connection, try "iwconfig wlan0 channel 6" (or
> whatever channel the AP is on.  It will unblock whatever is stuck in the
> kernel.  That's another thing I'd like to have a look at before I even
> touch that nvidia card.
>
>   
>> N.B.
>> What i really don't want is any kind of aggression towards to me.
>>     
>
> Of course.  If you wanted aggression towards you, you would have written
> to LKML :-)
>
>   
>> Coding something for a community, even advanced one requires certain
>> amount of tolerance, as ANY normal communication.
>> I am not perfect tester(i just attempt to become better one), i
>> understand that, but i am not coming after difficult day, working with
>> difficult people to submit difficult issues(which other people may
>> prefer to ignore all together or for those reasons abandon Linux and go
>> to MS products, i have seeing enough of those). I am not going to become
>> more enthusiastic by being referred to "RTFM noob" like documents, who
>> nearly imply my stupidity.
>> So far I was very happy in this list, and even my lack of technical
>> knowledge did not spoil my attitude towards this list. I want it to stay
>> that way. Ignoring problems of people who lack technical knowledge on
>> matter is path to nowhere. I seriously, don't want to start flame wars,
>> i am not naive and I don't ask for apologies. I just ask one thing,
>> please be tolerant, everyone can have a shi*** day, that includes you
>> and me!
>>     
>
> It's totally unnecessary to vent your frustration here, especially to
> those who never called you a noob or something.
>   
Like I said, i am not going to start flame wars... An no, you did not
call me noob. Its not my point, i am here to solve my problems, and in
some way to help other people to solve theirs. If I suck at it - tell me
what to do to improve my skills of "debugging" drivers. Links to
documentation that call me stupid are not doing any help.


Anyway, I am loading nvidia drivers and see how all that works... will
try that trick with with channel number if connection fades away.  What
i also noticed, "iwlist wlan0 scan"(interface is up) doesn't bring any
results there either...
Anyway, let me try it and i shall report back.

Dmitri

[ath9k-devel] ath9k vs nvidia(proprietary)

[Jason Tackaberry]

On Mon, 2009-04-06 at 22:17 +0100, Dmitri Seletski wrote:
> > It should not be a problem.  You can try some workarounds, such as the
> > "irqpoll" option on the kernel command line.
> [...]
>    
> Well, how can i help? I am not coder... I dont speak C ;)

Dmitri, I'd be very curious to hear if booting with the irqpoll option
helps your problem.

The other day I was playing with a D-Link DWA-652 and whenever I started
doing bulk transfers dmesg would show "received PCI FATAL interrupt" and
the wireless connection would go dead.

I noticed in /proc/interrupts that both ath9k and nvidia were using IRQ
16.  However, as the nvidia binary driver is more important to me than
the wireless card, and I didn't have very long to decide if I wanted to
keep it, I decided to return the DWA-652.

Thanks,
Jason.

[ath9k-devel] ath9k vs nvidia(proprietary)

[Dmitri Seletski]

Hello Jason.
I will probably get a slap from developers, which i do and do not deserve...
Anyway, i was using gnome on linux for as long as i remember.
And every time i have loaded nvidia drivers, i was loading gnome too.
(hence, my assumption of hardware issue)
Now for several days i am running xfce, pinging local computers over
Acess Point  in G mode for hours without one packet being dropped.
Nvidia drivers are loaded, SLI is enabled, all 3 devices are actually
IRQ 16...
I could never guess, it looks like it was not hardware problem after all.

So irqpoll option was unnecessary for me...

I still have issue with connecting using N mode...
It silently fails, even by manually assigning channel.

Jason Tackaberry wrote:
> On Mon, 2009-04-06 at 22:17 +0100, Dmitri Seletski wrote:
>   
>>> It should not be a problem.  You can try some workarounds, such as the
>>> "irqpoll" option on the kernel command line.
>>>       
>> [...]
>>    
>> Well, how can i help? I am not coder... I dont speak C ;)
>>     
>
> Dmitri, I'd be very curious to hear if booting with the irqpoll option
> helps your problem.
>
> The other day I was playing with a D-Link DWA-652 and whenever I started
> doing bulk transfers dmesg would show "received PCI FATAL interrupt" and
> the wireless connection would go dead.
>
> I noticed in /proc/interrupts that both ath9k and nvidia were using IRQ
> 16.  However, as the nvidia binary driver is more important to me than
> the wireless card, and I didn't have very long to decide if I wanted to
> keep it, I decided to return the DWA-652.
>
> Thanks,
> Jason.
>
>