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: <200611070141.16593.len.brown@intel.com>
Date:	Tue, 7 Nov 2006 01:41:16 -0500
From:	Len Brown <len.brown@...el.com>
To:	Andreas Mohr <andi@...x01.fht-esslingen.de>
Cc:	Thomas Gleixner <tglx@...utronix.de>, Ingo Molnar <mingo@...e.hu>,
	linux-kernel@...r.kernel.org
Subject: Re: CONFIG_NO_HZ: missed ticks, stall (keyb IRQ required) [2.6.18-rc4-mm1]

On Monday 06 November 2006 15:58, Andreas Mohr wrote:

> > > How useful would it be to simply disable C2 operation (but not C1)
> > > in CONFIG_NO_HZ mode after's been determined to kill APIC timer?:

If the goal is saving power, then disabling dynticks will likely
be more attractive than disabling C2.  Perhaps you can measure it?
eg. simply run "bltk -I" to measure idle battery life (http://sourceforge.net/projects/bltk)

But this is even more true when talking about C3 -- it certainly saves more
power than dynticks does.  This is true for the example system here:
http://ftp.kernel.org/pub/linux/kernel/people/lenb/acpi/doc/OLS2006-bltk-paper.pdf

So given that C3 on every known system that has shipped to date
breaks the LAPIC timer (and apparently this applies to C2 on these AMD boxes),
dynticks needs a solid story for co-existing with C3.

If  (HPET re-programming latency is not prohibititive)
	If  (there is an HPET/thread)
		then using the HPET/thread instead of the LAPIC timer is the way to go
	else If ( there is an HPET/core, or /package)
		then using the HPET for the timer interrupt to a group of threads
		maybe the way to go.  (interrupt needs to be re-sent to the actual
		destination via IPI)
else if (PIT re-programming latency is not prohibitive)
		then using PIT to cpu0 and IPI to the cpu of interest would work
else
	The HW isn't well suited to dynticks,
	fall back to a static HZ mode at boot-time.
	Though I suppose that "static HZ" could actually be a variable
	set at boot-time rather than the compile-time scheme we use today.
	Eg. if a box has high-latency C-states, it certainly wants hz<=64
	Boxes with just C1 will not care and would go for hz= faster

-Len

-
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/