Re: [d-kernel] HZ=250?

[Michael Shigorin <mike@osdn.org.ua>]

[-- Attachment #1.1: Type: text/plain, Size: 1747 bytes --]

On Wed, Jun 22, 2005 at 07:16:27PM +0400, Denis Smirnov wrote:
>  MS> Вот и я так думаю.  Вчера с vsu@ обсудили, вышло вот что
>  MS> (также ещё два прилежащих и имеющих отношение к 2.6.12
>  MS> в свете 2.9.9.9 вопроса, насколько понимаю):
> Меня только пните больно, чтобы zaptel проверил.

Держи src.rpm.  Для удобства -- вот и описание:

---
List:       linux-kernel <?l=linux-kernel&r=1&w=2>
Subject:    [RFC][PATCH][2.6] CONFIG_HZ for x86 <?t=106778764700002&r=1&w=2>
From:       Mikael Pettersson <mikpe () csd ! uu ! se> <?a=90367460900004&r=1&w=2>
Date:       2003-11-02 15:36:22 <?l=linux-kernel&r=1&w=2&b=200311>
[Download message RAW <?l=linux-kernel&m=106778757803465&q=raw>]*

This patch adds a CONFIG_HZ option to x86, allowing the kernel-
internal HZ to be reduced from 1000 to 512 or 100. This solves
lost timer interrupt problems on really old machines like my 486.
According to Alan Cox, HZ==1000 is also harmful on some laptops
(presumably due to long SMI windows), so this patch should be
useful for those too.

Selecting a lower HZ may also help thermal efficiency slightly,
by allowing machines to idle longer between timer interrupts.
HZ==100 also reduces the size of the kernel, since it eliminates
a lot of HZ<-->USER_HZ conversions.                                             
HZ==512 may or may not be useful, but it does work. Why 512 and
not 500? It's to avoid having to correct for HZ != 1<<SHIFT_HZ
in kernel/timer.c:second_overflow().
 
The patch also fixes some jiffies_to_clock_t() buglets, which
cause warnings in fs/proc/proc_misc.c when (HZ % USER_HZ) != 0.

/Mikael
---

-- 
 ---- WBR, Michael Shigorin <mike@altlinux.ru>
  ------ Linux.Kiev http://www.linux.kiev.ua/

[-- Attachment #1.2: kernel-fix-CONFIG_HZ-0.1-alt1.src.rpm --]
[-- Type: application/octet-stream, Size: 3339 bytes --]

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]