Bug 71

Summary: RTL8100BL (8139) do not work on acpi UP without local apic
Product: ACPI Reporter: Nicolas Mailhot (Nicolas.Mailhot)
Component: Config-InterruptsAssignee: Len Brown (lenb)
Status: CLOSED CODE_FIX    
Severity: low CC: acpi-bugzilla, Marko.Macek, snxiao
Priority: P2    
Hardware: IA-32   
OS: Linux   
Kernel Version: 2.5.46-bk2 to 2.6.0-test3 Subsystem:
Regression: --- Bisected commit-id:
Bug Depends on:    
Bug Blocks: 845    
Attachments: GA-7VAX kernel config ( 2.6.0-test4-bk2)
dmesg for GA-7VAX (2.6.0-tes4-bk2, acpi + local apic - ioapic)
GA-7VAX interrupts without acpi or apic
GA-7VAX interrupts with acpi and io-apic
GA-7VAX interrupts with acpi, local apic and no io-apic

Description Nicolas Mailhot 2002-11-16 04:24:44 UTC
[ This is just a rehash of the networking part of bug #10 so people not
interested in the usb part can follow this bug. While bug #10 form may have been
good for linux-kernel, compound bugs are really not a good bugzilla practice ]

Exact Kernel version: 2.5.46-bk2 + about a dozen 2.5/2.5-ac kernels I've tried
in the 2.5.44-2.5.47 range

Distribution: Red Hat Raw Hide

Hardware Environment: 
Gigabyte GA -7VAX http://www.giga-byte.com/products/7vax.htm (Via KT 400 + Via
VT 8235 + Realtek RTL8100BL), F6 bios

Problem Description: When I build a 2.5 acpi kernel without local apic, network
initialization fails with a timeout. Enabling local apic kills usb.

nov  5 20:50:30 rousalka ifup:
nov  5 20:50:30 rousalka ifup: D
Comment 1 Stacy Woods 2002-11-21 08:54:04 UTC
Andy,  I think bug belongs under the ACPI component.
Comment 2 Nicolas Mailhot 2002-12-08 07:31:40 UTC
Seems via rhine has the exact invert problem btw, with it its enablig io-apic
that kills network
Comment 3 Andy Grover 2003-03-06 16:51:29 UTC
marking normal - workaround available (acpi=off)
Comment 4 Nicolas Mailhot 2003-03-07 13:26:45 UTC
Well, back to high since there is *no* workaround.

One can build a kernel without acpi, and it will work (therefore not blocker).

However *all* acpi kernels, be it without local apic, with local apic and no
io-apic, with local apic, io-apic and black magic options (pci=noacpi or
acpi=off) will fail either on usb or on network (I even found a combo that
disabled both : acpi+local apic+io-apci+pci=noacpi).

So on my hardware acpi on linux is a big no-go now. And getting acpi to work on
2.5 like on 2.4 was the whole point of this bug.

I won't qualify a bug that requires removal of a user-visible functionnality 
Comment 5 Nicolas Mailhot 2003-03-07 13:30:28 UTC
Just to be clear : I re-did the whole test series on 2.5.64-bk2 (all APIC/ACPI
compile option combos, all acpi startup options on acpi+local apic+io-apic kernel)
Comment 6 Nicolas Mailhot 2003-03-07 13:35:41 UTC
This one is normal. If one bug can be fixed, I'd rather have bug 10.
*If* this one were fixed it'd be a workaround for 10
Comment 7 Andy Grover 2003-05-21 14:13:42 UTC
*** Bug 399 has been marked as a duplicate of this bug. ***
Comment 8 Andy Grover 2003-05-21 14:16:38 UTC
still present in .69 or .69 + acpi-patch?
Comment 9 Nicolas Mailhot 2003-05-24 15:53:45 UTC
Well, seems I can no longuer test this.
Without ioapic or local apic I get a hda: lost interrupt on the new disks. Seems
the new SiI680 ATA card has the same problem as the embeded RTL8100BL controller.

Do you want to to consider it as a separate bug or should we say it's the same
as the original one ? I can fill a new bug if you like - with the usb one
that'll only be the third acpi-related bug on this system.
Comment 10 Nicolas Mailhot 2003-06-17 12:58:36 UTC
(of course without acpi local/ioapic are not needed)
Comment 11 Luming Yu 2003-08-24 19:24:17 UTC
Please test 2.6.0-test4 kernel. 
Comment 12 Nicolas Mailhot 2003-08-25 13:25:30 UTC
With 2.6.0-test4-bk2 the problem is semi-fixed so this might be considered a
workaround for bug 10.

acpi + local apic - io-apic = eth + usb + raid working

*However* it is still not really usable since usb is anemic (for the mouse at
least). The cursor moves like a turtle at empty load (ie likesome of
theinteractivity tests peope heve been doing, but without anything serious
running). This might be due to the way interupts areassigned - instead of being
properly spread like with io-apic all heavy users are lumped together.
Comment 13 Nicolas Mailhot 2003-08-25 13:26:55 UTC
Created attachment 729 [details]
GA-7VAX kernel config ( 2.6.0-test4-bk2)
Comment 14 Nicolas Mailhot 2003-08-25 13:28:52 UTC
Created attachment 730 [details]
dmesg for GA-7VAX (2.6.0-tes4-bk2, acpi + local apic - ioapic)
Comment 15 Nicolas Mailhot 2003-08-25 13:29:43 UTC
Created attachment 731 [details]
GA-7VAX interrupts without acpi or apic
Comment 16 Nicolas Mailhot 2003-08-25 13:30:50 UTC
Created attachment 732 [details]
GA-7VAX interrupts with acpi and io-apic
Comment 17 Nicolas Mailhot 2003-08-25 13:32:18 UTC
Created attachment 733 [details]
GA-7VAX interrupts with acpi, local apic and no io-apic
Comment 18 Nicolas Mailhot 2003-09-06 11:16:04 UTC
Severity -> low since bug #10 is fixed and this was a workarounf scenario
Comment 19 Nicolas Mailhot 2003-10-02 15:29:24 UTC
Since bug #10 is fixed and patched was merged in mainline, I really don't care
about this one now.

-> LATER
Comment 20 Len Brown 2003-11-28 21:50:05 UTC
booted 2.6.0-test11 on a GA-7VAX with no local APIC suport built into the (ACPI) kernel, 
and it the ethernet now runs fine.   
 
Closing -- let me know if you have any further problems with this box. 
thanks, 
-Len