Skip to contentSkip to conversation

IO-APIC on nforce2 [PATCH] + [PATCH] for nmi_debug=1 + [PATCH]

DISCUSSION Ross Dickson 5 participants 7 messages May 10, 2004 – May 24, 2004

lore
Share & tools
lore
#0 lore raw

Craig Bradney wrote

Quoted 7 lines

Well.. 2.6.6 is released.. and THANK YOU Linus and all the patch writers.. we have nforce2 fixes in the released kernel now. I'm just waiting for a gentoo-dev-sources release now..

Craig

MOMENT PLEASE.
ALMOST complete nforce2 support. Job not done yet.

Unfortunately 2.6.6 still has the old check_timer code which inhibits nmi_watchdog=1 on all nforce2 from working by having timer_ack=1 when checking io-apic pit routing.

It is a hardware issue - NOT A BUGGY BIOS ISSUE inside the integrated nforce2 interrupt routing.

To my understanding IT WILL NEVER BE FIXED BY A BIOS REVISION and after reading the 8259 datasheets I think it is a mistake within the existing code to have the timer_ack on there in the first place.

I would still like to see Maciej's check_timer patch in the kernel. It was pulled after only a single user mobo complaint was posted yet it helps both nforce2 and ibm bios pc's. To my knowledge little effort was made by that user to accomodate the patch - it was just outright pulled in spite of its benefit to others?

Who do we ask to revisit this? Linus? the io-apic.c maintainer? or the one user with a complaint?

That patch that was dropped by Linus? after appearing in 2.6.3-mm3. For those nforce2 users with problems of clock skew with the timer into pin0 routing, that patch gave a virtual wire timer routing which worked well.

It also works around serious problems for ibm users who also want it in.
http://linux.derkeiler.com/Mailing-Lists/Kernel/2004-04/4421.html

Regards
Ross.

#1 lore raw
Quoted 8 lines · Ross Dickson

Craig Bradney wrote

Quoted 6 lines

Well.. 2.6.6 is released.. and THANK YOU Linus and all the patch writers.. we have nforce2 fixes in the released kernel now. I'm just waiting for a gentoo-dev-sources release now..

Quoted 4 lines
Quoted 1 line

Craig

MOMENT PLEASE.
ALMOST complete nforce2 support. Job not done yet.

I was waiting for that...

Quoted 17 lines

Unfortunately 2.6.6 still has the old check_timer code which inhibits nmi_watchdog=1 on all nforce2 from working by having timer_ack=1 when checking io-apic pit routing.

It is a hardware issue - NOT A BUGGY BIOS ISSUE inside the integrated nforce2 interrupt routing.

To my understanding IT WILL NEVER BE FIXED BY A BIOS REVISION and after reading the 8259 datasheets I think it is a mistake within the existing code to have the timer_ack on there in the first place.

I would still like to see Maciej's check_timer patch in the kernel. It was pulled after only a single user mobo complaint was posted yet it helps both nforce2 and ibm bios pc's. To my knowledge little effort was made by that user to accomodate the patch - it was just outright pulled in spite of its benefit to others?

Quoted 6 lines

Who do we ask to revisit this? Linus? the io-apic.c maintainer? or the one user with a complaint?

That patch that was dropped by Linus? after appearing in 2.6.3-mm3. For those nforce2 users with problems of clock skew with the timer into pin0 routing, that patch gave a virtual wire timer routing which worked well.

Why was it dropped?

Quoted 2 lines

It also works around serious problems for ibm users who also want it in.
http://linux.derkeiler.com/Mailing-Lists/Kernel/2004-04/4421.html

It looks like theres IDE caching problems with the new flushing mechanism on Maxtor 8mb cache drives.. of which I have 2, so, looks like I'm awaiting 2.6.7 now anyway or a patch to remove it.

Craig

#2 lore raw
In reply to Ross Dickson
Quoted 10 lines · Ross Dickson

Craig Bradney wrote

Quoted 5 lines

Well.. 2.6.6 is released.. and THANK YOU Linus and all the patch writers.. we have nforce2 fixes in the released kernel now. I'm just waiting for a gentoo-dev-sources release now..

Craig

MOMENT PLEASE.
ALMOST complete nforce2 support. Job not done yet.

Damn =)

Quoted 6 lines

Unfortunately 2.6.6 still has the old check_timer code which inhibits nmi_watchdog=1 on all nforce2 from working by having timer_ack=1 when checking io-apic pit routing.

It is a hardware issue - NOT A BUGGY BIOS ISSUE inside the integrated nforce2 interrupt routing.

So it should be in.

Quoted 3 lines

To my understanding IT WILL NEVER BE FIXED BY A BIOS REVISION and after reading the 8259 datasheets I think it is a mistake within the existing code to have the timer_ack on there in the first place.

The "breaking" machine could be blacklisted instead.

Quoted 5 lines

I would still like to see Maciej's check_timer patch in the kernel. It was pulled after only a single user mobo complaint was posted yet it helps both nforce2 and ibm bios pc's. To my knowledge little effort was made by that user to accomodate the patch - it was just outright pulled in spite of its benefit to others?

Hummm, and now ppl seem to be forcing 4k stacks that breaks my laptop to a level of non usability.. =P

Quoted 2 lines

Who do we ask to revisit this? Linus? the io-apic.c maintainer? or the one user with a complaint?

Perhaps there should be a workaround option, ie like acpi=force etc.

Quoted 6 lines

That patch that was dropped by Linus? after appearing in 2.6.3-mm3. For those nforce2 users with problems of clock skew with the timer into pin0 routing, that patch gave a virtual wire timer routing which worked well.

It also works around serious problems for ibm users who also want it in.
http://linux.derkeiler.com/Mailing-Lists/Kernel/2004-04/4421.html

Rediffed it against 2.6.6
But i don't see how the virtual wire mode was done, i just rediffed the patch that i found.

Signature · 1 line
Ian Kumlien <pomac () vapor ! com> -- http://pomac.netswarm.net
arch/i386/kernel/io_apic.c renamed from arch/i386/kernel/io_apic.c.orig +14 -4
@@ -2159,6 +2159,10 @@
 {
 	int pin1, pin2;
 	int vector;
+	unsigned int ver;
+
+	ver = apic_read(APIC_LVR);
+	ver = GET_APIC_VERSION(ver);
 
 	/*
 	 * get/set the timer IRQ vector:
@@ -2172,11 +2176,15 @@
 	 * mode for the 8259A whenever interrupts are routed
 	 * through I/O APICs.  Also IRQ0 has to be enabled in
 	 * the 8259A which implies the virtual wire has to be
-	 * disabled in the local APIC.
+	 * disabled in the local APIC. Finally timer interrupts
+	 * need to be acknowledged manually in the 8259A for
+	 * do_slow_timeoffset() and for the i82489DX when using
+	 * the NMI watchdog.
 	 */
 	apic_write_around(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
 	init_8259A(1);
-	timer_ack = 1;
+	timer_ack = !cpu_has_tsc;
+	timer_ack |= nmi_watchdog == NMI_IO_APIC && !APIC_INTEGRATED(ver);
 	enable_8259A_irq(0);
 
 	pin1 = find_isa_irq_pin(0, mp_INT);
@@ -2194,7 +2202,8 @@
 				disable_8259A_irq(0);
 				setup_nmi();
 				enable_8259A_irq(0);
-				check_nmi_watchdog();
+				if (check_nmi_watchdog() < 0)
+					timer_ack = !cpu_has_tsc;
 			}
 			return;
 		}
@@ -2217,7 +2226,8 @@
 				add_pin_to_irq(0, 0, pin2);
 			if (nmi_watchdog == NMI_IO_APIC) {
 				setup_nmi();
-				check_nmi_watchdog();
+				if (check_nmi_watchdog() < 0)
+					timer_ack = !cpu_has_tsc;
 			}
 			return;
 		}
#3 lore raw
In reply to Ross Dickson

Does the 2.6.6 Changelog section about nforce2 imply that Ross' Patches of Stability are no longer required? I don't read lkml regularely, so this would be new for me.

Anyway, if this works (tm) I am very happy. Will attempt to test it later on, even though I have Maxtor 8MB discs :)

Regards,

Arjen

<[email protected]>
[PATCH] fixup for C1 Halt Disconnect problem on nForce2 chipsets

Based on information provided by "Allen Martin" <[email protected]>:

A hang is caused when the CPU generates a very fast CONNECT/HALT cycle sequence. Workaround is to set the SYSTEM_IDLE_TIMEOUT to 80 ns. This allows the state-machine and timer to return to a proper state within 80 ns of the CONNECT and probe appearing together. Since the CPU will not issue another HALT within 80 ns of the initial HALT, the failure condition is avoided.

Quoted 45 lines · Ross Dickson

Craig Bradney wrote

Quoted 7 lines

Well.. 2.6.6 is released.. and THANK YOU Linus and all the patch writers.. we have nforce2 fixes in the released kernel now. I'm just waiting for a gentoo-dev-sources release now..

Craig

MOMENT PLEASE.
ALMOST complete nforce2 support. Job not done yet.

Unfortunately 2.6.6 still has the old check_timer code which inhibits nmi_watchdog=1 on all nforce2 from working by having timer_ack=1 when checking io-apic pit routing.

It is a hardware issue - NOT A BUGGY BIOS ISSUE inside the integrated nforce2 interrupt routing.

To my understanding IT WILL NEVER BE FIXED BY A BIOS REVISION and after reading the 8259 datasheets I think it is a mistake within the existing code to have the timer_ack on there in the first place.

I would still like to see Maciej's check_timer patch in the kernel. It was pulled after only a single user mobo complaint was posted yet it helps both nforce2 and ibm bios pc's. To my knowledge little effort was made by that user to accomodate the patch - it was just outright pulled in spite of its benefit to others?

Who do we ask to revisit this? Linus? the io-apic.c maintainer? or the one user with a complaint?

That patch that was dropped by Linus? after appearing in 2.6.3-mm3. For those nforce2 users with problems of clock skew with the timer into pin0 routing, that patch gave a virtual wire timer routing which worked well.

It also works around serious problems for ibm users who also want it in.
http://linux.derkeiler.com/Mailing-Lists/Kernel/2004-04/4421.html

Regards
Ross.

#4 lore raw
Quoted 3 lines · Arjen Verweij

Does the 2.6.6 Changelog section about nforce2 imply that Ross' Patches of Stability are no longer required? I don't read lkml regularely, so this would be new for me.

Do a grep for nforce and you'll find:
<[email protected]>
[PATCH] fixup for C1 Halt Disconnect problem on nForce2 chipsets

Based on information provided by "Allen Martin" <[email protected]>:

A hang is caused when the CPU generates a very fast CONNECT/HALT cycle sequence. Workaround is to set the SYSTEM_IDLE_TIMEOUT to 80 ns. This allows the state-machine and timer to return to a proper state within 80 ns of the CONNECT and probe appearing together. Since the CPU will not issue another HALT within 80 ns of the initial HALT, the failure condition is avoided.

and:
<[email protected]>
[ACPI] workaround for nForce2 BIOS bug: XT-PIC timer in IOAPIC mode "acpi_skip_timer_override" boot parameter
dmi_scan for common platforms, may be replaced with PCI-ID in future.
http://bugzilla.kernel.org/show_bug.cgi?id=1203

So it's only missing the patches that ross mentioned, it will still work without em. (except for clock drift afair)

Quoted 2 lines

Anyway, if this works (tm) I am very happy. Will attempt to test it later on, even though I have Maxtor 8MB discs :)

Signature · 1 line
Ian Kumlien <pomac () vapor ! com> -- http://pomac.netswarm.net
#5 lore raw
In reply to Arjen Verweij
Quoted 6 lines · Arjen Verweij

Does the 2.6.6 Changelog section about nforce2 imply that Ross' Patches of Stability are no longer required? I don't read lkml regularely, so this would be new for me.

Anyway, if this works (tm) I am very happy. Will attempt to test it later on, even though I have Maxtor 8MB discs :)

Sorry, i must have needed food or so, i missed the post you had there... =P

Signature · 1 line
Ian Kumlien <pomac () vapor ! com> -- http://pomac.netswarm.net
#6 lore raw
In reply to Ross Dickson
Quoted 3 lines · Ross Dickson

To my understanding IT WILL NEVER BE FIXED BY A BIOS REVISION and after reading the 8259 datasheets I think it is a mistake within the existing code to have the timer_ack on there in the first place.

The timer_ack variation exploits the polling mode of the i8259A with the
AEOI enabled. It's a valid configuration. It's needed for a correct operation of the timer interrupt handler and the NMI watchdog in certain configurations. What's there in the datasheets you that makes you think the code is a mistake? A reference would be appreciated.

Maciej

Signature · 3 lines
+  Maciej W. Rozycki, Technical University of Gdansk, Poland   +
+--------------------------------------------------------------+
+        e-mail: [email protected], PGP key available        +