From: Mathieu Desnoyers <compudj@krystal.dyndns.org>
To: Andi Kleen <ak@suse.de>
Cc: patches@x86-64.org, linux-kernel@vger.kernel.org,
Daniel Walker <dwalker@mvista.com>
Subject: Re: [PATCH] [15/58] i386: Rewrite sched_clock
Date: Fri, 20 Jul 2007 10:39:07 -0400 [thread overview]
Message-ID: <20070720143907.GC29979@Krystal> (raw)
In-Reply-To: <20070720141210.GA29979@Krystal>
* Mathieu Desnoyers (mathieu.desnoyers@polymtl.ca) wrote:
> * Andi Kleen (ak@suse.de) wrote:
> >
> > > I noticed the same thing about interrupts off when going through the
> > > code.
> >
> > That's only on a slow path during cpu frequency changing while the TSC is instable.
> > Shouldn't be that common.
> >
> > -Andi
>
> Hrm, I don't see why you can get away without disabling interrupts in
> the fast path:
>
> +unsigned long long tsc_sched_clock(void)
> +{
> + unsigned long long r;
> + struct sc_data *sc = &get_cpu_var(sc_data);
> +
> + if (unlikely(sc->unstable)) {
> + r = (jiffies_64 - sc->sync_base) * (1000000000 / HZ);
> + r += sc->ns_base;
> + /*
> + * last_val is used to avoid non monotonity on a
> + * stable->unstable transition. Make sure the time
> + * never goes to before the last value returned by the
> + * TSC clock.
> + */
> + while (r <= sc->last_val) {
> + rmb();
> + r = sc->last_val + 1;
> + rmb();
> + }
> + sc->last_val = r;
>
> Here, slow path, we update last_val (64 bits value). Must be protected.
>
> + } else {
> + rdtscll(r);
> + r = __cycles_2_ns(sc, r);
> + sc->last_val = r;
>
> Here, fast path, we update last_val too so it is ready to be read when
> the tsc will become unstable.
>
> If we don't disable interrupts around its update, we could have: (LSB vs
> MSB update order is arbitrary)
>
> update sc->last_val 32MSB
> interrupt comes
> update sc->last_val 32MSB
> update sc->last_val 32LSB
> iret
> update sc->last_val 32LSB
>
> So if, after this, we run tsc_sched_clock() with an unstable TSC, we
> read a last_val containing the interrupt's MSB and the last_val LSB. It
> can particularity hurt if we are around a 32 bits overflow, because time
> could "jump" forward of about 1.43 seconds on a 3 GHz system.
>
> So I guess we need synchronization on the fast path, and therefore using
> cmpxchg_local on x86_64 and cmpxchg64_local on i386 makes sense.
>
The case above explained the issue for i386. For x86_64, the race goes
like this:
read tsc
interrupt
read tsc
update sc->last_val
iret
update sc->last_val
Here, last_val is not at its highest value anymore. This is why a
cmpxchg is useful on x86_64.
Mathieu
> Mathieu
>
> + }
> +
> + put_cpu_var(sc_data);
> +
> + return r;
> +}
>
>
>
> --
> Mathieu Desnoyers
> Computer Engineering Ph.D. Student, Ecole Polytechnique de Montreal
> OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F BA06 3F25 A8FE 3BAE 9A68
--
Mathieu Desnoyers
Computer Engineering Ph.D. Student, Ecole Polytechnique de Montreal
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F BA06 3F25 A8FE 3BAE 9A68
next prev parent reply other threads:[~2007-07-20 14:39 UTC|newest]
Thread overview: 119+ messages / expand[flat|nested] mbox.gz Atom feed top
2007-07-19 9:54 [PATCH] [0/58] First batch of x86 patches for .23 Andi Kleen
2007-07-19 9:54 ` [PATCH] [1/58] x86: Always flush pages in change_page_attr Andi Kleen
2007-08-06 10:15 ` [patches] " Jan Beulich
2007-08-06 10:36 ` Andi Kleen
2007-08-06 10:49 ` Jan Beulich
2007-07-19 9:54 ` [PATCH] [2/58] x86_64: Tell gcc to only align stack to 8 bytes Andi Kleen
2007-07-19 11:50 ` Serge Belyshev
2007-07-19 12:06 ` Andi Kleen
2007-07-19 14:42 ` Chuck Ebbert
2007-07-19 9:54 ` [PATCH] [3/58] x86_64: asm/ptrace.h needs linux/compiler.h Andi Kleen
2007-07-19 9:54 ` [PATCH] [4/58] x86_64: Don't rely on a unique IO-APIC ID Andi Kleen
2007-07-19 9:54 ` [PATCH] [5/58] x86_64: Report the pending irq if available in smp_affinity Andi Kleen
2007-07-19 10:23 ` Ingo Molnar
2007-07-19 9:54 ` [PATCH] [6/58] x86_64: Use LOCAL_DISTANCE and REMOTE_DISTANCE in x86_64 ACPI code Andi Kleen
2007-07-19 9:54 ` [PATCH] [7/58] x86_64: various cleanups in NUMA scan node Andi Kleen
2007-07-19 17:15 ` Yinghai Lu
2007-07-19 17:21 ` Andi Kleen
2007-07-19 17:38 ` Yinghai Lu
2007-07-19 20:00 ` Andi Kleen
2007-07-19 21:01 ` David Rientjes
2007-07-19 9:54 ` [PATCH] [8/58] x86_64: Use string instruction memcpy/memset on AMD Fam10 Andi Kleen
2007-07-19 16:43 ` Jan Engelhardt
2007-07-19 17:00 ` Yinghai Lu
2007-07-19 9:54 ` [PATCH] [9/58] x86_64: Always use builtin memcpy on gcc 4.3 Andi Kleen
2007-07-21 23:16 ` Oleg Verych
2007-07-21 23:27 ` Andi Kleen
2007-07-22 0:29 ` Denis Vlasenko
2007-07-19 9:54 ` [PATCH] [10/58] i386: Move all simple string operations out of line Andi Kleen
2007-07-19 9:54 ` [PATCH] [11/58] x86: Support __attribute__((__cold__)) in gcc 4.3 Andi Kleen
2007-07-19 9:54 ` [PATCH] [12/58] x86_64: Add vDSO for x86-64 with gettimeofday/clock_gettime/getcpu Andi Kleen
2007-08-21 16:25 ` Daniel Walker
2007-08-21 18:45 ` Andi Kleen
2007-08-21 18:40 ` Andrew Morton
2007-07-19 9:54 ` [PATCH] [13/58] x86: Separate checking of unsynchronized and unstable TSC Andi Kleen
2007-07-19 9:54 ` [PATCH] [14/58] x86_64: Add on_cpu_single Andi Kleen
2007-07-19 11:09 ` Satyam Sharma
2007-07-19 12:07 ` Andi Kleen
2007-07-19 9:54 ` [PATCH] [15/58] i386: Rewrite sched_clock Andi Kleen
2007-07-19 16:51 ` Daniel Walker
2007-07-19 17:13 ` Andi Kleen
2007-07-19 17:15 ` Daniel Walker
2007-07-19 17:22 ` Andi Kleen
2007-07-19 17:31 ` Daniel Walker
2007-07-19 17:38 ` Andi Kleen
2007-07-19 17:43 ` Daniel Walker
2007-07-19 18:00 ` Andi Kleen
2007-07-19 18:00 ` Daniel Walker
2007-07-20 3:11 ` Mathieu Desnoyers
2007-07-20 3:47 ` Mathieu Desnoyers
2007-07-20 4:18 ` [PATCH] [15/58] i386: Rewrite sched_clock (cmpxchg8b) Mathieu Desnoyers
2007-07-20 5:07 ` Nick Piggin
2007-07-20 5:47 ` Mathieu Desnoyers
2007-07-20 8:27 ` [PATCH] [15/58] i386: Rewrite sched_clock Andi Kleen
2007-07-20 14:12 ` Mathieu Desnoyers
2007-07-20 14:39 ` Mathieu Desnoyers [this message]
2007-07-20 15:14 ` Andi Kleen
2007-07-20 15:22 ` Mathieu Desnoyers
2007-07-20 16:49 ` [PATCH] 80386 and 80486 cmpxchg64 and cmpxchg64_local fallback Mathieu Desnoyers
2007-07-19 9:55 ` [PATCH] [16/58] x86_64: Use new shared sched_clock in x86-64 too Andi Kleen
2007-07-19 9:55 ` [PATCH] [17/58] i386: Add L3 cache support to AMD CPUID4 emulation Andi Kleen
2007-07-20 17:00 ` [patches] " Andreas Herrmann
2007-07-20 17:15 ` Andreas Herrmann
2007-07-19 9:55 ` [PATCH] [18/58] x86_64: remove extra extern declaring about dmi_ioremap Andi Kleen
2007-07-19 9:55 ` [PATCH] [19/58] x86_64: Don't use softirq save locks in smp_call_function Andi Kleen
2007-07-19 12:16 ` Satyam Sharma
2007-07-19 12:19 ` Andi Kleen
2007-07-19 9:55 ` [PATCH] [20/58] x86: Always probe the NMI watchdog Andi Kleen
2007-07-19 10:24 ` Björn Steinbrink
2007-07-19 10:42 ` Andi Kleen
2007-07-19 9:55 ` [PATCH] [21/58] i386: Reserve the right performance counter for the Intel PerfMon " Andi Kleen
2007-07-19 10:21 ` Björn Steinbrink
2007-07-19 10:45 ` Andi Kleen
2007-07-19 9:55 ` [PATCH] [22/58] x86_64: hpet tsc calibration fix broken smi detection logic Andi Kleen
2007-07-19 9:55 ` [PATCH] [23/58] i386: remove pit_interrupt_hook Andi Kleen
2007-07-19 9:55 ` [PATCH] [24/58] x86_64: Untangle asm/hpet.h from asm/timex.h Andi Kleen
2007-07-19 9:55 ` [PATCH] [25/58] x86_64: use generic cmos update Andi Kleen
2007-07-19 9:55 ` [PATCH] [26/58] x86_64: Use generic xtime init Andi Kleen
2007-07-19 9:55 ` [PATCH] [27/58] x86_64: Remove dead code and other janitor work in tsc.c Andi Kleen
2007-07-19 9:55 ` [PATCH] [28/58] x86_64: Fix APIC typo Andi Kleen
2007-07-19 9:55 ` [PATCH] [29/58] x86_64: fiuxp pt_reqs leftovers Andi Kleen
2007-07-19 9:55 ` [PATCH] [30/58] x86: share hpet.h with i386 Andi Kleen
2007-07-19 9:55 ` [PATCH] [31/58] x86_64: apic.c coding style janitor work Andi Kleen
2007-07-19 9:55 ` [PATCH] [32/58] x86_64: time.c white space wreckage cleanup Andi Kleen
2007-07-19 9:55 ` [PATCH] [33/58] x86_64: Avoid too many remote cpu references due to /proc/stat Andi Kleen
2007-07-19 10:21 ` Christoph Hellwig
2007-07-19 10:41 ` Andi Kleen
2007-07-19 10:55 ` Adrian Bunk
2007-07-19 9:55 ` [PATCH] [34/58] x86_64: ia32entry adjustments Andi Kleen
2007-07-19 14:46 ` Jeff Garzik
2007-08-06 10:43 ` Jan Beulich
2007-07-19 9:55 ` [PATCH] [35/58] i386: allow debuggers to access the vsyscall page with compat vDSO Andi Kleen
2007-07-19 9:55 ` [PATCH] [36/58] x86_64: minor exception trace variables cleanup Andi Kleen
2007-07-19 9:55 ` [PATCH] [37/58] x86_64: remove unused variable maxcpus Andi Kleen
2007-07-19 9:55 ` [PATCH] [38/58] i386: smp-alt-once option is only useful with HOTPLUG_CPU Andi Kleen
2007-07-19 9:55 ` [PATCH] [39/58] i386: minor nx handling adjustment Andi Kleen
2007-07-19 9:55 ` [PATCH] [40/58] i386: remapped_pgdat_init() static Andi Kleen
2007-07-19 9:55 ` [PATCH] [41/58] i386: arch/i386/kernel/i8253.c should #include <asm/timer.h> Andi Kleen
2007-07-19 9:55 ` [PATCH] [42/58] i386: timer_irq_works() static again Andi Kleen
2007-07-19 9:55 ` [PATCH] [43/58] x86_64: Quicklist support for x86_64 Andi Kleen
2007-07-19 9:55 ` [PATCH] [44/58] x86_64: extract helper function from e820_register_active_regions Andi Kleen
2007-07-19 9:55 ` [PATCH] [45/58] x86_64: fake pxm-to-node mapping for fake numa Andi Kleen
2007-07-19 9:55 ` [PATCH] [46/58] x86_64: fake apicid_to_node " Andi Kleen
2007-07-19 9:55 ` [PATCH] [47/58] i386: insert unclaimed MMCONFIG resources Andi Kleen
2007-07-19 9:55 ` [PATCH] [48/58] x86_64: O_EXCL on /dev/mcelog Andi Kleen
2007-07-19 9:55 ` [PATCH] [49/58] x86_64: support poll() " Andi Kleen
2007-07-19 9:55 ` [PATCH] [50/58] x86_64: mcelog tolerant level cleanup Andi Kleen
2007-07-19 9:55 ` [PATCH] [51/58] i386: fix machine rebooting Andi Kleen
2007-07-19 9:55 ` [PATCH] [52/58] i386: fix section mismatch warnings in mtrr Andi Kleen
2007-07-19 9:55 ` [PATCH] [53/58] x86: PM_TRACE support Andi Kleen
2007-07-19 9:55 ` [PATCH] [54/58] x86: Make Alt-SysRq-p display the debug register contents Andi Kleen
2007-07-19 9:55 ` [PATCH] [55/58] i386: add reference to the arguments Andi Kleen
2007-07-19 9:55 ` [PATCH] [56/58] x86: round_jiffies() for i386 and x86-64 non-critical/corrected MCE polling Andi Kleen
2007-07-19 9:55 ` [PATCH] [57/58] x86_64: check remote IRR bit before migrating level triggered irq Andi Kleen
2007-07-19 9:55 ` [PATCH] [58/58] x86: remove support for the Rise CPU Andi Kleen
2007-07-19 10:45 ` Alan Cox
2007-07-19 10:48 ` Adrian Bunk
2007-07-19 11:13 ` Alan Cox
2007-07-19 12:03 ` Andi Kleen
2007-07-19 14:56 ` Jeff Garzik
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20070720143907.GC29979@Krystal \
--to=compudj@krystal.dyndns.org \
--cc=ak@suse.de \
--cc=dwalker@mvista.com \
--cc=linux-kernel@vger.kernel.org \
--cc=patches@x86-64.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox