Message-ID: <493960F2.6070102@oracle.com>
Date:	Fri, 05 Dec 2008 09:12:18 -0800
From:	Randy Dunlap <randy.dunlap@...cle.com>
To:	Joe Korty <joe.korty@...r.com>
CC:	"mtk.manpages@...il.com" <mtk.manpages@...il.com>,
	Greg KH <greg@...ah.com>, Thomas Gleixner <tglx@...utronix.de>,
	Ingo Molnar <mingo@...e.hu>,
	Alexey Dobriyan <adobriyan@...il.com>,
	Linux API <linux-api@...r.kernel.org>,
	LKML <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH] ABI Documentation for /proc/timer_list, v2

Joe Korty wrote:
> Document /proc/timer_list ABI, version 2.
> 
> This partially documents /timer_list, including the
> proposed 'Version 0.5' extensions that add a jiffie timer
> display.
> 
> v2 exists to address some of the concerns Michael Kerrisk
> brought up.  What was left out: I did not document old
> versions of /timer_list, I did not document the meaning
> of the x.y version numbering system (which only Ingo
> can answer anyway), and I did not document fields of
> secondary importance that already had adequate 'DocBook'
> documentation in the kernel sources.
> 
> Signed-off-by: Joe Korty <joe.korty@...r.com>
> 
> Index: 2.6.28-rc6/Documentation/ABI/stable/procfs-timer_list
> ===================================================================
> --- /dev/null	1970-01-01 00:00:00.000000000 +0000
> +++ 2.6.28-rc6/Documentation/ABI/stable/procfs-timer_list	2008-12-01 13:07:15.000000000 -0500
> @@ -0,0 +1,129 @@
> +What:		/proc/timer_list
> +Date:		November 2008
> +Contact:	Ingo Molnar <mingo@...e.hu>
> +		Thomas Gleixner <tglx@...utronix.de>
> +		Joe Korty <joe.korty@...r.com>
> +Revision-Rate:	Moderate
> +At-Revision:	0.5
> +Description:
> +		/proc/timer_list displays most everything about every kind
> +		of timer, and some things about time too.
> +
> +		The contents of this file should be expected to change,
> +		as the data displayed corresponds directly to various
> +		kernel-internal data structures.  For this reason, the first
> +		line contains the file revision.  It is the responsibility
> +		of this file's maintainers to bump the revision each time a
> +		kernel is released having incompatible changes in this file.
> +
> +		This document covers only the version of /proc/timer_list
> +		located in the kernel sources to which it is attached.
> +		Documentation for previous (and later) versions of
> +		/proc/timer_list is to be found (if they exist) in the
> +		kernel sources of those earlier (or later) kernels.
> +
> +		Section Overview
> +		----------------
> +		The file contains several somewhat independent sections.
> +
> +		The first section contains a few lines of global info:
> +		   1 - Timer List Version: File revision.
> +		   2 - HRTIMER_MAX_CLOCK_BASES: number of clock types that
> +		       support high resolution timers.
> +		   3 - now at x nsecs: number of nsecs since boot.
> +
> +		The second section is organized per-CPU.  Each CPU subsection
> +		in turn contains several sub-subsections which are, in order
> +		of appearance:
> +
> +		   The contents of the data structures associated with each
> +		   clock on this CPU:
> +		    1 - clock ID: 0 == CLOCK_REALTIME, 1 == CLOCK_MONOTONIC
> +		    2 - base: kernel address of this clock's
> +		        hrtimer_clock_base structure.
> +		    3 - resolution: resolution of this clock.
> +		    4 - get_time: name of kernel function used to fetch
> +		        time from this clock.
> +		    5 - offset: difference, in nsecs, between this clock
> +		        and the reference clock for this clock.
> +		   Under each of these clocks is, in turn, a display of all
> +		   the active high resolution timers queued to that clock.
> +		   These are the lines beginning with '#' and are described
> +		   in detail later in this document.

Are we supposed to be able to see lines beginning with '#' in this text file,
or only in /proc/timer_list ?

> +
> +		   The contents of per-CPU hrtimer data fields not
> +		   associated with a particular cpu clock (ie, shared by

Please use "CPU" consistenly (instead of "cpu").

> +		   both clocks or not associated with any clock).  These
> +		   are: expires_next, hres_active, nr_event, nohz_mode, all
> +		   things idle_*, tick_stopped, last_jiffies, next_jiffies.
> +		   The above are field names from 'struct tick_sched' and
> +		   'struct hrtimer_cpu_base', documentation for these may

		   'struct hrtimer_cpu_base'; documentation for these may

> +		   be found in the kernel DocBook.
> +
> +		   A display of low resolution (ie, jiffie) timer wheel
> +		   data.  These are prefixed by the lines:
> +		    1 - base: kernel virtual address of the timer wheel
> +		        data structure (struct tvec_base) for this cpu.
> +		    2 - running timer: kernel virtual address of the
> +		        expired timer being processed, NULL if none.
> +		    3 - timer_jiffies: what this wheel considers to
> +		        be the current time, will be == jiffies or
> +			will lag it by a tick or two if it has not
> +			caught up with the current time.
> +		   Also under this section is a display, one per line, of
> +		   each active jiffie timer queued to this CPU.  These are
> +		   the lines under an 'active jiffie timers' section that
> +		   begin with a number.
> +
> +		The third and final section describes each 'tick device'
> +		known to the kernel.  A tick device is a piece of hardware
> +		capable of generating periodic and/or one-shot interrupts
> +		under software control, and thus is capable of generating
> +		the interrupts needed to expire the various active timers
> +		at their given expiration times.  Examples of tick devices:
> +		hpet, pit, lapic.  All but the first two lines display
> +		fields corresponding to structure elements from 'struct
> +		clock_event_device', documentation for which can be found
> +		in the kernel Docbook. The first two lines are:
> +		  1 - mode: 0 == periodic timer, 1 == one-shot timer
> +		  2 - is 'Per CPU device' or is 'Broadcast device'
> +
> +		Hires Timer Layout
> +		------------------
> +		High-resolution timers are displayed on lines that begin
> +		with a '#' and always appear under one of the many sections
> +		labeled 'active timers'.  There is an 'active timers'
> +		section for every CPU and every clock.
> +
> +		The fields of a hrtimer, spread out over two lines, are:

		              an hrtimer,
> +
> +		line 1 fields:
> +		  1 - unique hrtimer index (#0, #1, #2, etc)
> +		  2 - kernel address of the hrtimer data structure
> +		      in question
> +		  3 - function to be called when timer expires
> +		  4 - timer state (eg, S:01), avail states, OR-able:
> +		      0 - inactive
> +		      1 - enqueued
> +		      2 - callback
> +		      4 - pending
> +		      8 - migrate
> +		  5 - function which created the timer
> +		  6 - process name & pid which created the timer
> +
> +		line 2 fields:
> +		  1 - absolute expiration time, range format (start - end)
> +		  2 - relative expiration time, range format (start - end)
> +
> +		Lowres Timer Layout
> +		-------------------
> +		Low-resolution timers are displayed one-per-line under
> +		sections labeled 'active jiffie timers'.  There is one such
> +		section per CPU.  A lowres timer has the following fields:
> +
> +		  1 - number of jiffies remaining until timer expires
> +		  2 - function to be called on expiration
> +		  3 - data value to be given to the above function on
> +		      expiration
> +		  4 - function which created this timer
> +		  5 - name & pid of the process that created this timer


Thanks,
~Randy
--
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/

Powered by blists - more mailing lists