Network failures with ACPI enabled (kernel 2.4, 2.6)

Network failures with ACPI enabled (kernel 2.4, 2.6)

[Michael Guntsche]

[-- Attachment #1: Type: text/plain, Size: 1364 bytes --]

Hi,

Im using ACPI on my Acer Travelmate 800 for a very long time.
I never had any big problems with it, until now.

Today I activated my builtin ethernet card (BCM4401) because I wanted to
copy a large amount of data. (normally I only use WLAN).

First I copied some files from my notebook to another computer.
Everything worked fine.

But then I tried to copy an CD-Image file from our fileserver to my
notebook. As soon as the transfer started, the following error messages
appeared
in my syslog and the transfer stalled. 

	b44: eth0: Link is down.
	b44: eth0: Link is up at 100 Mbps, full duplex.
	b44: eth0: Flow control is on for TX and on for RX.
	<repeating>
	.........

Looking at the switch I saw that the card was indeed resetting itself.

-------------
	Server		-> 	Notebook 	OK
	Notebook	->	Server		NOT OK
-------------

Since I was running 2.6.0 I switched back to 2.4.23 to see if it had the
same problem.

To sum it up:

	2.6.0	   +	b44/bcm4400(Broadcom module) + pci=noacpi 	NOT OK
	2.4.2[23]  +	b44/bcm4400(Broadcom module) + pci=noacpi	NOT OK

BUT
	2.6.0      +	b44 + acpi=off OK

With acpi completely off the transfer worked in both directions without
any problems.


Does anybody here have a clue why?
I don't understand how ACPI can interfer with the card, since trying
pci=noacpi didn't work either.
		

Thanks in advance,
Michael










[-- Attachment #2: debug.tar.bz2 --]
[-- Type: application/octet-stream, Size: 37477 bytes --]

Re: Network failures with ACPI enabled (kernel 2.4, 2.6)

[Michael Guntsche]

On Tue, 30 Dec 2003 01:08:22 +0100
Michael Guntsche <mike-Z92qn3yYq0hWk0Htik3J/w@public.gmane.org> wrote:

> But then I tried to copy an CD-Image file from our fileserver to my
> notebook. As soon as the transfer started, the following error messages
> appeared
> in my syslog and the transfer stalled. 
> 
> 	b44: eth0: Link is down.
> 	b44: eth0: Link is up at 100 Mbps, full duplex.
> 	b44: eth0: Flow control is on for TX and on for RX.
> 	<repeating>
> 	.........

Today I tried the same transfer with a XIRCOM PCMCIA network card and ACPI enabled.
It worked without a problem.
I am not sure what this means for ACPI <-> B44 though. ;)


/Michael


-------------------------------------------------------
This SF.net email is sponsored by: IBM Linux Tutorials.
Become an expert in LINUX or just sharpen your skills.  Sign up for IBM's
Free Linux Tutorials.  Learn everything from the bash shell to sys admin.
Click now! http://ads.osdn.com/?ad_id=1278&alloc_id=3371&op=click

Re: Network failures with ACPI enabled (kernel 2.4, 2.6)

[Len Brown]

On Tue, 2003-12-30 at 10:57, Michael Guntsche wrote:
> On Tue, 30 Dec 2003 01:08:22 +0100
> Michael Guntsche <mike-Z92qn3yYq0hWk0Htik3J/w@public.gmane.org> wrote:
> 
> > But then I tried to copy an CD-Image file from our fileserver to my
> > notebook. As soon as the transfer started, the following error messages
> > appeared
> > in my syslog and the transfer stalled. 
> > 
> > 	b44: eth0: Link is down.
> > 	b44: eth0: Link is up at 100 Mbps, full duplex.
> > 	b44: eth0: Flow control is on for TX and on for RX.
> > 	<repeating>
> > 	.........
> 
> Today I tried the same transfer with a XIRCOM PCMCIA network card and ACPI enabled.
> It worked without a problem.
> I am not sure what this means for ACPI <-> B44 though. ;)

Please note which IRQ each kind of ethernet NIC gets,
and if it shares an IRQ with any other device.
If the working and non-working NICs come up on different IRQs
then that may be a clue.

But the fact that B44 works with acpi=off and fails with pci=noacpi
suggests that IRQ assignment isn't the problem.

Looking at the actual ACPI interrupt:

  0:     738650          XT-PIC  timer
  9:       3546          XT-PIC  acpi

Not an interrupt storm, but you're getting a measurable number of acpi
events.


echo 0x4 > /proc/acpi/debug_layer
echo 0x08000000 > /proc/acpi/debug_level

may tell us what they are -- though it is unclear why they might
interfere with the B44 device and not others.

>Local APIC disabled by BIOS -- reenabling.
> Found and enabled local APIC!

This can be risky -- please verify that "nolapic" has no effect.

thanks,
-Len




-------------------------------------------------------
This SF.net email is sponsored by: IBM Linux Tutorials.
Become an expert in LINUX or just sharpen your skills.  Sign up for IBM's
Free Linux Tutorials.  Learn everything from the bash shell to sys admin.
Click now! http://ads.osdn.com/?ad_id=1278&alloc_id=3371&op=click

Re: Network failures with ACPI enabled (kernel 2.4, 2.6)

[Michael Guntsche]

On 30 Dec 2003 12:48:02 -0500
Len Brown <len.brown-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org> wrote:

> > Today I tried the same transfer with a XIRCOM PCMCIA network card
> > and ACPI enabled. It worked without a problem.
> > I am not sure what this means for ACPI <-> B44 though. ;)
> 
> Please note which IRQ each kind of ethernet NIC gets,
> and if it shares an IRQ with any other device.
> If the working and non-working NICs come up on different IRQs
> then that may be a clue.

The broadcom card gets IRQ 5 and the Xircom card gets IRQ3.
Both IRQ are non-shared. Well IRQ 5 was shared with USB, but for testing
purposes I removed the USB modules.


> 
> But the fact that B44 works with acpi=off and fails with pci=noacpi
> suggests that IRQ assignment isn't the problem.
> 
> Looking at the actual ACPI interrupt:
> 
>   0:     738650          XT-PIC  timer
>   9:       3546          XT-PIC  acpi
> 
> Not an interrupt storm, but you're getting a measurable number of acpi
> events.
> 
> 
> echo 0x4 > /proc/acpi/debug_layer
> echo 0x08000000 > /proc/acpi/debug_level
> 
> may tell us what they are -- though it is unclear why they might
> interfere with the B44 device and not others.

I know where the IRQ's come from. I have an applet running that shows me
the battery state and temperature via ACPI. Every time it polls the stats
the interrupts increase.
For testing purposes I disabled it, verified that the interrupts didn't increase
and tried copying a file, same result.

> 
> >Local APIC disabled by BIOS -- reenabling.
> > Found and enabled local APIC!
> 
> This can be risky -- please verify that "nolapic" has no effect.
I tried "nolapic" and even compiled a kernel completely without IOAPIC support.
No change.

Is ACPI doing something nasty with IRQ 5?


Cheers,
Michael


-------------------------------------------------------
This SF.net email is sponsored by: IBM Linux Tutorials.
Become an expert in LINUX or just sharpen your skills.  Sign up for IBM's
Free Linux Tutorials.  Learn everything from the bash shell to sys admin.
Click now! http://ads.osdn.com/?ad_id=1278&alloc_id=3371&op=click

Re: Network failures with ACPI enabled (kernel 2.4, 2.6) PROBLEM FOUND!!!!!!

[Michael Guntsche]

On Tue, 30 Dec 2003 19:23:01 +0100
Michael Guntsche <mike-Z92qn3yYq0hWk0Htik3J/w@public.gmane.org> wrote:

> On 30 Dec 2003 12:48:02 -0500
> Len Brown <len.brown-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org> wrote:
> 
> > > Today I tried the same transfer with a XIRCOM PCMCIA network card
> > > and ACPI enabled. It worked without a problem.
> > > I am not sure what this means for ACPI <-> B44 though. ;)
> > 
> > Please note which IRQ each kind of ethernet NIC gets,
> > and if it shares an IRQ with any other device.
> > If the working and non-working NICs come up on different IRQs
> > then that may be a clue.
> 
> The broadcom card gets IRQ 5 and the Xircom card gets IRQ3.
> Both IRQ are non-shared. Well IRQ 5 was shared with USB, but for
> testing purposes I removed the USB modules.

Hia Len,

You won't believe me but I found out what was causing the problem.
Since I never use the parallel port on this notebook I turned it OFF in
the BIOS.
Setting it to AUTOMATIC mode solved my issues.
The Linux-ACPI still has a problem somewhere since I don't see the
network timeouts in WinXP, if I turn off the parport.
For me it looks like that the parport and the internal card share
something, either ACPI or hardware wise and by turning off the parport
the network card somehow changes too.

What do you think about that?

Michael


-------------------------------------------------------
This SF.net email is sponsored by: IBM Linux Tutorials.
Become an expert in LINUX or just sharpen your skills.  Sign up for IBM's
Free Linux Tutorials.  Learn everything from the bash shell to sys admin.
Click now! http://ads.osdn.com/?ad_id=1278&alloc_id=3371&op=click

RE: Network failures with ACPI enabled (kernel 2.4, 2.6) PROBLEM FOUND!!!!!!

[Brown, Len]

> > The broadcom card gets IRQ 5 and the Xircom card gets IRQ3.
> > Both IRQ are non-shared. Well IRQ 5 was shared with USB, but for
> > testing purposes I removed the USB modules.

Can you put the xircom in the slot where the broadcom lived and
See if it too gets IRQ5 and fails with the EPP set to OFF?

Doing so would suggest that this failure is device independent
and IRQ related.

> You won't believe me but I found out what was causing the problem.
> Since I never use the parallel port on this notebook I turned 
> it OFF in the BIOS.
> Setting it to AUTOMATIC mode solved my issues.

> The Linux-ACPI still has a problem somewhere since I don't see the
> network timeouts in WinXP, if I turn off the parport.
> For me it looks like that the parport and the internal card share
> something, either ACPI or hardware wise and by turning off the parport
> the network card somehow changes too.

It may be I/O port ranges rather than IRQs at the bottom of this one.

-Len


-------------------------------------------------------
This SF.net email is sponsored by: IBM Linux Tutorials.
Become an expert in LINUX or just sharpen your skills.  Sign up for IBM's
Free Linux Tutorials.  Learn everything from the bash shell to sys admin.
Click now! http://ads.osdn.com/?ad_id\x1278&alloc_id371&op=click

Re: Network failures with ACPI enabled (kernel 2.4, 2.6) PROBLEM FOUND!!!!!!

[Michael Guntsche]

On Tue, 30 Dec 2003 16:17:47 -0500
"Brown, Len" <len.brown-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org> wrote:

> Can you put the xircom in the slot where the broadcom lived and
> See if it too gets IRQ5 and fails with the EPP set to OFF?
> 
> Doing so would suggest that this failure is device independent
> and IRQ related.

Sorry, I can't do that since the broadcom network card is build-in.
I tried to get the pcmcia subsystem to assign IRQ 5 to the Xircom 
which didn't work.


> It may be I/O port ranges rather than IRQs at the bottom of this one.

But why is it working when the parport is turned on?
Shouldn't it be the other way around, if there is a resource
conflict?

Michael


-------------------------------------------------------
This SF.net email is sponsored by: IBM Linux Tutorials.
Become an expert in LINUX or just sharpen your skills.  Sign up for IBM's
Free Linux Tutorials.  Learn everything from the bash shell to sys admin.
Click now! http://ads.osdn.com/?ad_id=1278&alloc_id=3371&op=click

RE: Network failures with ACPI enabled (kernel 2.4, 2.6) PROBLEM FOUND!!!!!!

[Brown, Len]

> But why is it working when the parport is turned on?
> Shouldn't it be the other way around, if there is a resource
> conflict?

The BIOS' AML may not be well coordinated with the CMOS setting
for enabling/disabling the device.

It is interesting that Windows XP works either way.
Could be that Linux has an ACPI bug that XP does not have,
or that the BIOS happens to work with XP's implementation
and not Linux's implementation.

-Len


-------------------------------------------------------
This SF.net email is sponsored by: IBM Linux Tutorials.
Become an expert in LINUX or just sharpen your skills.  Sign up for IBM's
Free Linux Tutorials.  Learn everything from the bash shell to sys admin.
Click now! http://ads.osdn.com/?ad_id\x1278&alloc_id371&op=click

Re: Network failures with ACPI enabled (kernel 2.4, 2.6) PROBLEM FOUND!!!!!!

[Michael Guntsche]

On Tue, 30 Dec 2003 17:36:14 -0500
"Brown, Len" <len.brown-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org> wrote:
> The BIOS' AML may not be well coordinated with the CMOS setting
> for enabling/disabling the device.
> 
> It is interesting that Windows XP works either way.
> Could be that Linux has an ACPI bug that XP does not have,
> or that the BIOS happens to work with XP's implementation
> and not Linux's implementation.
> 

Is there a way to see what the change in the CMOS does to ACPI?
What puzzles me is that with the parport disabled outgoing traffic
still works. It would make more sense if the network didn't work at all.

Michael



-------------------------------------------------------
This SF.net email is sponsored by: IBM Linux Tutorials.
Become an expert in LINUX or just sharpen your skills.  Sign up for IBM's
Free Linux Tutorials.  Learn everything from the bash shell to sys admin.
Click now! http://ads.osdn.com/?ad_id=1278&alloc_id=3371&op=click

Re: Network failures with ACPI enabled (kernel 2.4, 2.6) PROBLEM FOUND!!!!!!

[Michael Guntsche]

On Wed, 31 Dec 2003 00:17:16 +0100
Michael Guntsche <mike-Z92qn3yYq0hWk0Htik3J/w@public.gmane.org> wrote:

> Is there a way to see what the change in the CMOS does to ACPI?
> What puzzles me is that with the parport disabled outgoing traffic
> still works. It would make more sense if the network didn't work at all.
> 
One more piece of information.
The parport must be set to ECP, else you still get the same errors.


Michael


-------------------------------------------------------
This SF.net email is sponsored by: IBM Linux Tutorials.
Become an expert in LINUX or just sharpen your skills.  Sign up for IBM's
Free Linux Tutorials.  Learn everything from the bash shell to sys admin.
Click now! http://ads.osdn.com/?ad_id=1278&alloc_id=3371&op=click

Re: Network failures with ACPI enabled (kernel 2.4, 2.6) PROBLEM FOUND!!!!!!

[Len Brown]

On Tue, 2003-12-30 at 18:46, Michael Guntsche wrote:
> On Wed, 31 Dec 2003 00:17:16 +0100
> Michael Guntsche <mike-Z92qn3yYq0hWk0Htik3J/w@public.gmane.org> wrote:
> 
> > Is there a way to see what the change in the CMOS does to ACPI?
> > What puzzles me is that with the parport disabled outgoing traffic
> > still works. It would make more sense if the network didn't work at all.
> > 
> One more piece of information.
> The parport must be set to ECP, else you still get the same errors.

The fact that transmit works and receive fails is a clue.
Probably the broadcom uses interrupts for receive, but doesn't
use interrupts for transmit.  So if it were not getting interrupts,
transmit could still send frames.

So if you enable the parallel port in ECP mode, broadcom works --
but if the port is disabled, or enabled but not in ECP mode, it fails?

This is curious.  Indeed, the AML for this box shows three devices,
ECP, EPP, and LPTB.  Unfortunately, the _PRS for them disassembles
as a Buffer that one has to hand-disassemble to see what resources
are actually possible...

To answer your question about observing what the ECP enable/disable
does to the AML, the DSDT could be extracted, modified
and used to over-ride the DSDT in the BIOS.  This isn't rocket science,
but isn't trivial either.

As curious as the failure is, the system works both ways when
running XP, so I would encourage you to put the info into a
bug report.  The observation about TX working and RX failing
raises the prospect that this really is an IRQ issue, and
with the /proc/interrupts for the different cases at hand
the solution to this puzzle may become more clear.

thanks,
-Len





-------------------------------------------------------
This SF.net email is sponsored by: IBM Linux Tutorials.
Become an expert in LINUX or just sharpen your skills.  Sign up for IBM's
Free Linux Tutorials.  Learn everything from the bash shell to sys admin.
Click now! http://ads.osdn.com/?ad_id=1278&alloc_id=3371&op=click

Re: Network failures with ACPI enabled (kernel 2.4, 2.6) PROBLEM FOUND!!!!!!

[Michael Guntsche]

On 31 Dec 2003 01:21:01 -0500
Len Brown <len.brown-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org> wrote:

> The fact that transmit works and receive fails is a clue.
> Probably the broadcom uses interrupts for receive, but doesn't
> use interrupts for transmit.  So if it were not getting interrupts,
> transmit could still send frames.
> 
> So if you enable the parallel port in ECP mode, broadcom works --
> but if the port is disabled, or enabled but not in ECP mode, it fails?

Hmm I played around with it some more. If I lower the speed on my
notebook
to 10baseT-HD the transfer works at 800KB/s.
But errors and frames on the device (eth0) are increasing at an alarming
rate.
For me it looks like that the problem occurs if the network card gets a
lot of inbound traffic.


> 
> This is curious.  Indeed, the AML for this box shows three devices,
> ECP, EPP, and LPTB.  Unfortunately, the _PRS for them disassembles
> as a Buffer that one has to hand-disassemble to see what resources
> are actually possible...
So this means that ACPI IS doing something different depending on the
setting in the BIOS, right?


> 
> To answer your question about observing what the ECP enable/disable
> does to the AML, the DSDT could be extracted, modified
> and used to over-ride the DSDT in the BIOS.  This isn't rocket
> science, but isn't trivial either.

> 
> As curious as the failure is, the system works both ways when
> running XP, so I would encourage you to put the info into a
> bug report.  The observation about TX working and RX failing
> raises the prospect that this really is an IRQ issue, and
> with the /proc/interrupts for the different cases at hand
> the solution to this puzzle may become more clear.

In both cases (with or without ECP enabled) interrupts are increasing,
although it stops in the error case since the card resets itself.


/michael


-------------------------------------------------------
This SF.net email is sponsored by: IBM Linux Tutorials.
Become an expert in LINUX or just sharpen your skills.  Sign up for IBM's
Free Linux Tutorials.  Learn everything from the bash shell to sys admin.
Click now! http://ads.osdn.com/?ad_id=1278&alloc_id=3371&op=click

Re: Network failures with ACPI enabled (kernel 2.4, 2.6) PROBLEM FOUND!!!!!!

[Nate Lawson]

On Wed, 31 Dec 2003, Michael Guntsche wrote:
> On Tue, 30 Dec 2003 17:36:14 -0500
> "Brown, Len" <len.brown-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org> wrote:
> > The BIOS' AML may not be well coordinated with the CMOS setting
> > for enabling/disabling the device.
> >
> > It is interesting that Windows XP works either way.
> > Could be that Linux has an ACPI bug that XP does not have,
> > or that the BIOS happens to work with XP's implementation
> > and not Linux's implementation.
> >
>
> Is there a way to see what the change in the CMOS does to ACPI?
> What puzzles me is that with the parport disabled outgoing traffic
> still works. It would make more sense if the network didn't work at all.

Only incoming packets generate an interrupt.

-Nate


-------------------------------------------------------
This SF.net email is sponsored by: IBM Linux Tutorials.
Become an expert in LINUX or just sharpen your skills.  Sign up for IBM's
Free Linux Tutorials.  Learn everything from the bash shell to sys admin.
Click now! http://ads.osdn.com/?ad_id=1278&alloc_id=3371&op=click

RE: Network failures with ACPI enabled (kernel 2.4, 2.6) PROBLEM FOUND!!!!!!

[Nate Lawson]

On Tue, 30 Dec 2003, Brown, Len wrote:
> > You won't believe me but I found out what was causing the problem.
> > Since I never use the parallel port on this notebook I turned
> > it OFF in the BIOS.
> > Setting it to AUTOMATIC mode solved my issues.
>
> > The Linux-ACPI still has a problem somewhere since I don't see the
> > network timeouts in WinXP, if I turn off the parport.
> > For me it looks like that the parport and the internal card share
> > something, either ACPI or hardware wise and by turning off the parport
> > the network card somehow changes too.
>
> It may be I/O port ranges rather than IRQs at the bottom of this one.

You can use the -r flag for acpidump to decode Buffers as
ResourceTemplate()s:

     -r   Additionally outputs commented ResourceTemplate() macros for
	  Buffer objects that contain valid resource streams.  These macros
	  are defined in the ACPI 2.0 specification section 16.2.4.

Since moving to iasl as the acpidump backend in FreeBSD, we've lost this
capability.  I'd love it if iasl would grow a flag to have this behavior.
You could restrict it to only outputing the macros if the Buffer starts
with a valid resource.

-Nate


-------------------------------------------------------
This SF.net email is sponsored by: IBM Linux Tutorials.
Become an expert in LINUX or just sharpen your skills.  Sign up for IBM's
Free Linux Tutorials.  Learn everything from the bash shell to sys admin.
Click now! http://ads.osdn.com/?ad_id=1278&alloc_id=3371&op=click

Re: Network failures with ACPI enabled (kernel 2.4, 2.6) PROBLEM FOUND!!!!!!

[Pavel Machek]

Hi!

> > The fact that transmit works and receive fails is a clue.
> > Probably the broadcom uses interrupts for receive, but doesn't
> > use interrupts for transmit.  So if it were not getting interrupts,
> > transmit could still send frames.
> > 
> > So if you enable the parallel port in ECP mode, broadcom works --
> > but if the port is disabled, or enabled but not in ECP mode, it fails?
> 
> Hmm I played around with it some more. If I lower the speed on my
> notebook
> to 10baseT-HD the transfer works at 800KB/s.
> But errors and frames on the device (eth0) are increasing at an alarming
> rate.
> For me it looks like that the problem occurs if the network card gets a
> lot of inbound traffic.

Is not this plain old "ACPI stays to log with interrupts disabled"?
									Pavel
-- 
When do you have a heart between your knees?
[Johanka's followup: and *two* hearts?]


-------------------------------------------------------
This SF.net email is sponsored by: IBM Linux Tutorials.
Become an expert in LINUX or just sharpen your skills.  Sign up for IBM's
Free Linux Tutorials.  Learn everything from the bash shell to sys admin.
Click now! http://ads.osdn.com/?ad_id=1278&alloc_id=3371&op=click