Bug 7595

Summary: irq 3: nobody cared - PIC-mode - sis900 - Acer Aspire 3003LCi
Product: ACPI Reporter: Piter PUNK (piterpk)
Component: Config-InterruptsAssignee: ykzhao (yakui.zhao)
Status: REJECTED INVALID    
Severity: high CC: acpi-bugzilla, bunk, venza
Priority: P2    
Hardware: i386   
OS: Linux   
Kernel Version: 2.6.19-rc6 Subsystem:
Regression: --- Bisected commit-id:
Attachments: dmesg with acpi=off
dmesg with acpi and without irqpoll
dmesg with acpi and with irqpoll
/proc/interrupts without acpi
/proc/interrupts with acpi activated
Include Acer Aspire 3000 in acpi blacklist

Description Piter PUNK 2006-11-29 05:26:13 UTC
Most recent kernel where this bug did *NOT* occur: 2.6.17.13 (not tested in
newer ones before 2.6.19-rc6).
Distribution: Slackware 11.0
Hardware Environment: Acer Aspire 3003LCi with an embedded sis900 chipset and
PCI1410 CardBus controller
Software Environment: kernel 2.6.19-rc6
Problem Description: In 2.6.17.13, when i use the eth0 (sis900) for the first
time, it hangs the machine for a while, give one error message about "nobody
cares" and them works (with irqpoll option on). In 2.6.19-rc6 i got the machine
hang, the error message but the card doesn't communicate with or without
irqpoll. The only way it works is with acpi=off. My CardBus controller only got
one IRQ in acpi=off. (and sis900 and yenta shares the same IRQ 3).

Steps to reproduce:
1) Turn on the machine with acpi activated (with or without irqpoll)
2) ifconfig eth0 some.ip or dhcpcd eth0
3) Not works

To make sis900 work:
1) Turn on the machine with acpi=off
2) ifconfig eth0 some.ip or dhcpcd eth0
3) Use the network.
Comment 1 Piter PUNK 2006-11-29 05:28:11 UTC
Created attachment 9656 [details]
dmesg with acpi=off

This is the dmesg when i use 2.6.19-rc6 with acpi=off
Comment 2 Piter PUNK 2006-11-29 05:30:18 UTC
Created attachment 9657 [details]
dmesg with acpi and without irqpoll

This is the dmesg when i use 2.6.19-rc6 with acpi activated. Look the error
message in the end of dmesg telling about "disableing" IRQ 3 and suggesting to
use irqpoll option at boot.
Comment 3 Piter PUNK 2006-11-29 05:32:43 UTC
Created attachment 9658 [details]
dmesg with acpi and with irqpoll

Now with irqpoll. We can see nothing different. The same lines telling to try
"irqpoll" option and disalibling IRQ#3
Comment 4 Piter PUNK 2006-11-29 05:35:36 UTC
Created attachment 9659 [details]
/proc/interrupts without acpi

Look at irq 3, shared by yenta and eth0 (sis900)
Comment 5 Piter PUNK 2006-11-29 05:39:29 UTC
Created attachment 9660 [details]
/proc/interrupts with acpi activated

Now, the /proc/interrupts with acpi activated. Nobody in irq 3.
Comment 6 Adrian Bunk 2006-11-29 06:54:50 UTC
Can you test whether it works with kernel 2.6.18?
Comment 7 Piter PUNK 2006-11-29 08:08:53 UTC
Yes, i test with 2.6.18 now and the behavior is the same of 2.6.17.13.

1) kernel 2.6.18 + acpi (without irqpoll): Not work
2) kernel 2.6.18 + acpi (with irqpoll): Work
3) kernel 2.6.18 (without acpi): Work

In 1, irq 3 isn't in /proc/interrupts
In 2, the eth0 is in irq 3 in /proc/interrupts.
In 3, the eth0 and yenta is sharing irq 3 in /proc/interrupts

Do you want the dmesg and /proc/interrupts of all cases?
Comment 8 Piter PUNK 2006-12-08 10:55:37 UTC
Created attachment 9764 [details]
Include Acer Aspire 3000 in acpi blacklist

I saw using acpi=noirq solve my problem in 2.6.19-rc6 (and probably does the
same in 2.6.19). Now i got IRQ shared like in acpi=off, no error messages and a
system working.

I made this small patch to blacklist Acer Aspire 3000. I saw many other systems
blacklisted and i don't think is a bad idea made a system that works -;)
Comment 9 Len Brown 2007-03-30 19:29:21 UTC
The message above seem to all be in PIC-mode.
But the dmesg shows this system has an MADT.

Please build a kernel with IOAPIC support and re-test.
Attach the resulting dmesg and paste the resulting /proc/interrupts.

Yes, it is true that PIC mode should work properly,
but that becomes somewhat less interesting if IOAPIC mode works.
If we are to persue the PIC-mode issue, we'll also need the
output from acpidump and lspci -vv.


Comment 10 Piter PUNK 2007-04-03 22:29:02 UTC
It works with IO-APIC enabled.
This is the solution?
(test with 2.6.18.8)
Even the irqpoll option isn't needed anymore.
Comment 11 ykzhao 2007-10-08 07:33:17 UTC
Hi, Peter
Please retest this problem using the latest stable kernel. It is noted that local APIC and IO APIC should be enabled in the kernel configuration. 
Will you please upload the following files after the system is booted successfully ?
1.acpidump
2. .config
3. dmesg
Thanks.
Comment 12 Piter PUNK 2007-10-11 09:47:44 UTC
(In reply to comment #11)
> Please retest this problem using the latest stable kernel. It is noted that
> local APIC and IO APIC should be enabled in the kernel configuration. 
> Will you please upload the following files after the system is booted
> successfully ?

I will do the tests ASAP and post here what happens.
Do you want tests against 2.6.23 or 2.6.22.10?
Comment 13 Fu Michael 2007-10-22 17:35:40 UTC
Piter, plesae test the latest upstream kernel you can get. thanks. And also the acpidump .config and dmesg requested in comment# 11.
Comment 14 Fu Michael 2007-11-28 00:14:06 UTC
Piter, any response?
Comment 15 Piter PUNK 2007-12-16 22:00:45 UTC
Machine working fine. Using 2.6.23.9 with APIC and IO-APIC on.
Sorry the delay to answer. I only got my hands in problematic
machine last week.

Thank you!

dmesg and .config will be uploaded tomorrow.
Comment 16 ykzhao 2007-12-18 23:18:17 UTC
As this is not a linux/ACPI bug, and it can work well in IOAPIC mode.
Reject this bug and mark it as INVALID.
Pleas reopen it if you still have any questions.