2019-10-29 14:23:44

by Paul Menzel

[permalink] [raw]
Subject: [PATCH] smpboot: reuse timer calibration

From: Arjan van de Ven <[email protected]>
Date: Wed, 11 Feb 2015 17:28:14 -0600

NO point recalibrating for known-constant tsc ...
saves 200ms+ of boot time.

Signed-off-by: Paul Menzel <[email protected]>
---

Arjan, can your Signed-off-by line be added? On what device, did you
test this?

This upstreams the patch from Clear Linux.

https://github.com/clearlinux-pkgs/linux/blob/master/0108-smpboot-reuse-timer-calibration.patch

arch/x86/kernel/tsc.c | 3 +++
1 file changed, 3 insertions(+)

diff --git a/arch/x86/kernel/tsc.c b/arch/x86/kernel/tsc.c
index c59454c382fd..42b07ed467d2 100644
--- a/arch/x86/kernel/tsc.c
+++ b/arch/x86/kernel/tsc.c
@@ -1525,6 +1525,9 @@ unsigned long calibrate_delay_is_known(void)
if (!constant_tsc || !mask)
return 0;

+ if (cpu != 0)
+ return cpu_data(0).loops_per_jiffy;
+
sibling = cpumask_any_but(mask, cpu);
if (sibling < nr_cpu_ids)
return cpu_data(sibling).loops_per_jiffy;
--
2.24.0.rc1


2019-10-29 19:42:38

by Peter Zijlstra

[permalink] [raw]
Subject: Re: [PATCH] smpboot: reuse timer calibration

On Tue, Oct 29, 2019 at 03:19:57PM +0100, Paul Menzel wrote:
> From: Arjan van de Ven <[email protected]>
> Date: Wed, 11 Feb 2015 17:28:14 -0600
>
> NO point recalibrating for known-constant tsc ...
> saves 200ms+ of boot time.
>
> Signed-off-by: Paul Menzel <[email protected]>
> ---
>
> Arjan, can your Signed-off-by line be added? On what device, did you test
> this?

This patch makes no sense what so ever given the current function, also
it hard codes 0 as being the BSP.

All of this for platforms (!constant_tsc) that barely exist anymore.

2019-11-05 10:06:57

by Thomas Gleixner

[permalink] [raw]
Subject: Re: [PATCH] smpboot: reuse timer calibration

On Tue, 29 Oct 2019, Peter Zijlstra wrote:
> On Tue, Oct 29, 2019 at 03:19:57PM +0100, Paul Menzel wrote:
> > From: Arjan van de Ven <[email protected]>
> > Date: Wed, 11 Feb 2015 17:28:14 -0600
> >
> > NO point recalibrating for known-constant tsc ...
> > saves 200ms+ of boot time.
> >
> > Signed-off-by: Paul Menzel <[email protected]>
> > ---
> >
> > Arjan, can your Signed-off-by line be added? On what device, did you test
> > this?
>
> This patch makes no sense what so ever given the current function, also
> it hard codes 0 as being the BSP.
>
> All of this for platforms (!constant_tsc) that barely exist anymore.

And it's only an issue on multi-socket systems with !constant_tsc, which
are even more rare.

Thanks,

tglx