linux-kernel - Re: CONFIG_NO_HZ: missed ticks, stall (keyb IRQ required) [2.6.18-rc4-mm1]

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-Id: <1162892758.4715.362.camel@localhost.localdomain>
Date:	Tue, 07 Nov 2006 10:45:58 +0100
From:	Thomas Gleixner <tglx@...utronix.de>
To:	Andreas Mohr <andi@...x01.fht-esslingen.de>
Cc:	Ingo Molnar <mingo@...e.hu>, Len Brown <lenb@...nel.org>,
	linux-kernel@...r.kernel.org
Subject: Re: CONFIG_NO_HZ: missed ticks, stall (keyb IRQ required)
	[2.6.18-rc4-mm1]

On Tue, 2006-11-07 at 10:28 +0100, Thomas Gleixner wrote:
> > That's what I didn't understand all that time:
> > I do get the "C2 unusable, kills APIC timer" message, so I expected the code
> > to not use C2, but it seems it did use it (causing hangs) and I didn't
> > fully analyze the code whether it truly tried to prevent C2 here
> > (handling was a bit opaque to me, should have analyzed it
> > more thoroughly to get to know exactly what happens).
> > 
> > And like I said, brutally hard-wiring max_cstate to C1 already fixed
> > dynticks things for me, so it seems as if it still touched C2 before.
> 
> Yes, it leaves the C states untouched, it uses (should use) PIT instead
> of the local APIC timer. I'm a bit confused, why this does not work on
> your box.

Can you try the patch below please ? I just noticed, that the local APIC
gets reprogrammed before we switch back from PIT, which is perfectly
fine, but maybe your box does not like that treatment. If this does not
help, I'm going to do the never switch back from PIT hackery which I
wanted to avoid for performance reasons.

	tglx

Index: linux-2.6.19-rc4-mm1/arch/i386/kernel/apic.c
===================================================================
--- linux-2.6.19-rc4-mm1.orig/arch/i386/kernel/apic.c	2006-11-07 10:33:56.000000000 +0100
+++ linux-2.6.19-rc4-mm1/arch/i386/kernel/apic.c	2006-11-07 10:33:18.000000000 +0100
@@ -100,6 +100,7 @@ static struct clock_event_device lapic_c
  */
 struct lapic_event_device {
 	struct clock_event_device	evdev;
+	enum clock_event_mode		mode;
 	unsigned long			last_delta;
 	unsigned long			counter;
 };
@@ -224,9 +225,11 @@ static void lapic_next_event(unsigned lo
 	struct lapic_event_device *ldev;
 
 	ldev = container_of(evt, struct lapic_event_device, evdev);
-	ldev->last_delta = delta;
 
-	apic_write_around(APIC_TMICT, delta);
+	if (ldev->mode == CLOCK_EVT_PERIODIC) {
+		ldev->last_delta = delta;
+		apic_write_around(APIC_TMICT, delta);
+	}
 }
 
 /*
@@ -257,7 +260,7 @@ static void lapic_timer_setup(enum clock
 		apic_write_around(APIC_LVTT, v);
 		break;
 	}
-
+	ldev->mode = mode;
 	local_irq_restore(flags);
 }
 


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/