Bug 5065

Summary: SIS900 NIC fails -- 2.6.10 and later, unless acpi=off
Product: ACPI Reporter: Mathieu Jobin (somekool)
Component: Config-InterruptsAssignee: Len Brown (lenb)
Status: CLOSED PATCH_ALREADY_AVAILABLE    
Severity: high CC: acpi-bugzilla, bunk, fedefreire
Priority: P2    
Hardware: i386   
OS: Linux   
Kernel Version: >= 2.6.10 Subsystem:
Regression: --- Bisected commit-id:
Attachments: dmesg -s 64000 from working 2.6.9
/proc/interrupts with 2.6.9
all 7 files.

Description Mathieu Jobin 2005-08-14 12:50:36 UTC
Distribution: Gentoo 
Hardware Environment: Laptop, P4 2.4, 512mhz 
Software Environment: Linux 
Problem Description: I am stuck in 2.6.9, and I have enough reasons so I would 
like to upgrade..... but I never could, there is a problem since 2.6.10, which 
include current 2.6.12 
 
sorry I could not pick up the correct component, was not even sure If maybe I 
should have choose network category instead of ACPI category. 
 
in 2.6.9 is all fine, in 2.6.12, I have an eth0, I can assign an IP manually, 
but no network activity is possible. dhcpcd times out. If I disable ACPI, 
either in the kernel or using boot options, network works just fine. but  
then there is tons of other stuff on my laptop that requires ACPI. 
 
I have SIS900 network card, I know there is another obscure network card that 
experience the same problem, I found a post on through google once, but I can 
not find it anymore... I can't remember the name of the network card, but it 
something I never heard before. these are probably 2 of the rare network card 
that have this problem with the kernel. 
 
 
 
Steps to reproduce: try to use a sis900 with acpi under linux 2.6.12 
 
I'll try to post some more informations as I get some.
Comment 1 Len Brown 2005-08-14 21:21:38 UTC
if you booted with acpi=off and it caused it to start working,
then you got the right category:-)

Please confirm that (working) 2.6.9 was running an ACPI-enabled kernel,
so we can tell if this is a regression in ACPI support, or if
perhaps ACPI on this system never worked properly.
dmesg -s64000 and /proc/interrupts from 2.6.9 would be helpful.

Likely this is an interrupt or PCI configuration issue.
Please boot a recent kernel, say 2.6.12,
with "acpi=noirq" and then
"pci=noacpi" and report the results.

If either of those fix it, then please attach the output from
dmesg -s64000 for the working and failing case.  Also
attach the /proc/interrupts and output from lspci -vv in the
working case and the output from acpidump, available in pmtools here:
http://ftp.kernel.org/pub/linux/kernel/people/lenb/acpi/utils/


Comment 2 Mathieu Jobin 2005-08-15 08:55:49 UTC
Created attachment 5644 [details]
dmesg -s 64000 from working 2.6.9

I confirm 2.6.9 works fine with ACPI enabled. my laptop does not enjoy his 
life without ACPI. 
 
I'm actually not totally sure about 2.6.10, but I am positive, 2.6.11 and 
2.6.12 both have the bug. and it seems like a regression, or at least 
something that changed.
Comment 3 Mathieu Jobin 2005-08-15 09:04:44 UTC
Created attachment 5645 [details]
/proc/interrupts with 2.6.9
Comment 4 Mathieu Jobin 2005-08-15 09:34:53 UTC
Created attachment 5646 [details]
all 7 files.

bad news.... 
 
they both work 
 
acpi=noirq works 
pci=noacpi works 
but I tried again, with acpi fully enabled, and dhcp does not give me an IP.

that makes 7 new files to attach,... I think a .tar would be best, no ?

also, whenever I boot on 2.6.12, I get a strange noise, I don't know if its
related. it seems to start always at the same momment in a last step of the
boot process. its a repetive noise, hard to descrive, like the harddrive would
start and stop. it does not do it in 2.6.9. The noise could be related with
acpi not being fully enabled.

I'll try to get more information between the different states.
Comment 5 Mathieu Jobin 2005-08-15 09:59:09 UTC
ok, I confirm, with acpi enabled, I can't get an IP. and even though I set one  
manually, I can't ping other host in my LAN.  
  
NETDEV WATCHDOG: eth0: transmit timed out  
eth0: Transmit timeout, status 00000004 00000240   
NETDEV WATCHDOG: eth0: transmit timed out  
eth0: Transmit timeout, status 00000000 00000279   
  
those seems to be the offensive lines.  
  
about the noise, it does it with acpi=on, acpi=noirq and pci=noacpi as well as  
acpi=off, however, the noise is MUCH louder using acpi=off, its like the FAN  
constantly start and stop every second. in 2.6.9 its quite just like on a  
lake.  
  
please ask me, if there is any other info I could bring. 
 
 
Comment 6 Federico Freire 2006-01-31 09:05:53 UTC
I have the same problem. If I don't use acpi=off or acpi=noirq sis900 doesn't work.
This is my dmesg:
irq 3: nobody cared (try booting with the "irqpoll" option)
 [<c012aa7c>] __report_bad_irq+0x31/0x77
 [<c012ab4f>] note_interrupt+0x75/0x9a
 [<c012a4e6>] __do_IRQ+0x9f/0xe9
 [<c0103df8>] do_IRQ+0x1b/0x28
 [<c010290a>] common_interrupt+0x1a/0x20
handlers:
[<df21446b>] (sis900_interrupt+0x0/0xdb [sis900])
Disabling IRQ #3
eth0: Media Link On 10mbps half-duplex
NETDEV WATCHDOG: eth0: transmit timed out
eth0: Transmit timeout, status 00000004 00000241

I'm using 2.6.15

PS: Sorry for my english
Comment 7 Mathieu Jobin 2006-01-31 10:44:54 UTC
mine is fixed since I reinstalled from 2.6.14
Comment 8 Federico Freire 2006-02-01 07:18:19 UTC
Maybe this is a different bug. Did you change anything in your .config ? Maybe
you should close this bug.