Skip to contentSkip to conversation

io_apic & timer_ack fix

DISCUSSION Ross Dickson 3 participants 4 messages Apr 9, 2004 – Apr 14, 2004

lore
Share & tools
lore
#0 lore raw
Quoted 5 lines · Jon Grimm

Hmmm....

I see that the following patch got pulled in by Andrew:
http://linux.bkbits.net:8080/linux-2.5/diffs/arch/i386/kernel/[email protected]?nav=index.html|src/|src/arch|src/arch/i386|src/arch/i386/kernel|hist/arch/i386/kernel/io_apic.c

Quoted 7 lines

The patch had a couple bugs:
http://seclists.org/lists/linux-kernel/2004/Mar/4152.html

But the patch was pulled out entirely by Linus:
http://linux.bkbits.net:8080/linux-2.5/diffs/arch/i386/kernel/[email protected]?nav=index.html|src/|src/arch|src/arch/i386|src/arch/i386/kernel|hist/arch/i386/kernel/io_apic.c

Was it determined that the fix was bogus? damaging? fixable?

I thought the patch was OK with typos fixed.

Quoted 7 lines

I ask as I see behavior identical for which this patch seems to have been originally carved up for (buggy SMM BIOS at fault, but this was a workaround in the OS).

http://marc.theaimsgroup.com/?l=linux-kernel&m=101604672921823&w=2
http://www.ussg.iu.edu/hypermail/linux/kernel/0203.2/0698.html

Quoted 6 lines

Its a fair answer to force the BIOS vendor to fix, but in the meantime, I'm trying to figure out how safe/unsafe the workaround patch is ? I've ran on it overnight (with the semi-colon's fixed) and it hasn't exhibited the troubling behavior (where timer interrupts seem stuck or in some cases just extremely slow.... and the 8259 IMR is mucked up when Linux isn't even touching anymore).

I read the thread you mention about the IMR muckup along the way to creating my nforce2 patches - it was most enlightening as to how bad consumer computers can be.

Prakash tracked his overheat to a buggy binary nvidia driver
http://marc.theaimsgroup.com/?l=linux-kernel&m=108059111721363&w=2
and not Maciej's patch.

Thomas was tracking down C1 C2 etc states but I do not know the results of his search?
http://marc.theaimsgroup.com/?l=linux-kernel&m=107972277920929&w=2
Was it a problem only with one machine?

I do not recollect any other threads indicating problems with the patch.

I remember rediffing my nforce2 io-apic patch using the 2.6.3-mm3 kernel with Maciej's patch and having no heat trouble. I am surprised it got pulled out but then I only tested it on one type of chipset.

BTW I just rebooted to my modified 2.6.3-mm3 and got my normal 38C cpu. I have to have timer_ack=0 in my io-apic timer routing patch for nforce2 to get nmi_debug=1 to work. This was all along the way to trying to stop lockups. In fact I have been running no timer_ack kernel mods since December on 4 machines and all have been cool and hard lockup free.

Regards
Ross Dickson

#1 lore raw
Quoted 3 lines · Ross Dickson
Quoted 1 line

Was it determined that the fix was bogus? damaging? fixable?

I thought the patch was OK with typos fixed.

I consider it final.

Quoted 4 lines

Thomas was tracking down C1 C2 etc states but I do not know the results of his search?
http://marc.theaimsgroup.com/?l=linux-kernel&m=107972277920929&w=2
Was it a problem only with one machine?

The effect seems impossible, but I can't discuss with facts. I need to
get at AMD processor documents to find out what are the conditions to switch between the power-save states. Perhaps this is yet another SMM bug. E.g. the pair of PIC poll I/O accesses interacts with the SMM somehow.

Unfortunately I lack time to dig into it right now -- perhaps someone
else can do the research?

Signature · 3 lines
+  Maciej W. Rozycki, Technical University of Gdansk, Poland   +
+--------------------------------------------------------------+
+        e-mail: [email protected], PGP key available        +
#2 lore raw
Quoted 11 lines · Maciej W. Rozycki
Quoted 5 lines · Ross Dickson
Quoted 1 line

Was it determined that the fix was bogus? damaging? fixable?

I thought the patch was OK with typos fixed.

I consider it final.

I 've been running the patch on a 2.4.21 code base. I note that /proc/interrupts shows NMI interrupts coming in fast and furious, where without it there were none. I'm not sure what to think of this.

Any idea whether this could/should be expected?

Best Regards,
Jon

#3 lore raw
Quoted 3 lines · Jon Grimm

I 've been running the patch on a 2.4.21 code base. I note that /proc/interrupts shows NMI interrupts coming in fast and furious, where
without it there were none. I'm not sure what to think of this.

This is the NMI watchdog -- an aid to debug lock-ups. You can control it
with the "nmi_watchdog=" kernel option -- see documentation. I suppose without the timer_ack change the watchdog wouldn't work for some reason -- perhaps due to problems with your 8259A core or with the SMM.

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