linux-kernel - Re: various laptop nagles - any suggestions? (note: 2.6.19-rc2-mm1 but applies to multiple kernels)

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: <20061020203731.GA22407@elte.hu>
Date:	Fri, 20 Oct 2006 22:37:31 +0200
From:	Ingo Molnar <mingo@...e.hu>
To:	Andrew Morton <akpm@...l.org>
Cc:	tglx@...utronix.de, teunis <teunis@...tersgift.com>,
	linux-kernel@...r.kernel.org, Dmitry Torokhov <dtor@...l.ru>
Subject: Re: various laptop nagles - any suggestions?   (note: 2.6.19-rc2-mm1 but applies to multiple kernels)

* Andrew Morton <akpm@...l.org> wrote:

> > > I don't know how many machines will be affected by this, but I'd 
> > > expect it's quite a few - the Vaio has a less-than-one-year-old 
> > > Intel CPU in it.
> > 
> > Is this still the broken lapic issue ?
> 
> yup.  iirc the standard FC5 SMP kernel runs dog-slowly on that machine 
> too.

hm. This is how lapic timer calibration works.

the lapic timer is really simple - it counts down from a value and 
generates an irq if that counter reaches 0. Then it starts counting down 
again.

the 'count down from' value is programmed via __setup_APIC_LVTT().

we first write a 'really large' number into it (1 billion):

        __setup_APIC_LVTT(1000000000);

the unit of counting is '16 system bus cycles'.

i.e. if your system has a system bus of 333 MHz, then a value of 1 
billion takes 48 seconds to count down. (so the calibration ought to be 
pretty robust in this regard.)

then we use the wait_timer_tick() function, which waits until the PIT 
counter reaches 0 (which is attached to the PIT whose frequency we know 
and thus the PIT is already programmed correctly). Hence by calling 
wait_timer_tick() we can generate a delay of one jiffy - and we can read 
out the current lapic timer count and determine the calibration factor.

then we calculate the result as:

        result = (tt1-tt2)*APIC_DIVISOR/LOOPS;

where tt1 is the counter before we start calibration, tt2 is the lapic 
timer counter after we did calibration. (APIC_DIVISOR is 16)

i dont see where the error is - but there must be some calibration 
problem as your system shows a systematic 1:60 difference between 
expected and real lapic timer frequency.

	Ingo
-
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/