"Time police" - best way to submit?

"Time police" - best way to submit?

[Henry Hallam]

Hi, my name is Henry and I'm a kernel newbie.

Recently plagued by a perplexing issue where *something* would step
the clock on an important server by exactly 5 minutes at infrequent,
unpredictable intervals, I put together my first kernel module - a
trivial hook into do_settimeofday via a jprobe, that prints the PID,
process name and time adjustment to the kernel message log.

Here's what I have so far:
https://github.com/henryhallam/timepolice

Questions:

1. Is a module using kprobes/jprobes the right way to do it, vs adding
printks to linux/time/timekeeping.c?

2. If a module is the way to go, would it be better to create a
read/write /proc entry to allow enabling and disabling, rather than
inserting the module to enable and removing to disable?

3. Is this sort of thing worth submitting to mainline?  What are the
next steps for that, after choosing a better name, changing the silly
load/unload messages and ensuring it passes checkpatch.pl?

Thanks!

"Time police" - best way to submit?

[Valdis.Kletnieks at vt.edu]

On Wed, 16 Apr 2014 14:27:50 -0700, Henry Hallam said:

> 3. Is this sort of thing worth submitting to mainline?

Don't bother doing it unless you have first verified that it
plays nice with xntpd.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 848 bytes
Desc: not available
Url : http://lists.kernelnewbies.org/pipermail/kernelnewbies/attachments/20140416/9aea555f/attachment.bin 

"Time police" - best way to submit?

[Henry Hallam]

On Wed, Apr 16, 2014 at 4:27 PM,  <Valdis.Kletnieks@vt.edu> wrote:
> On Wed, 16 Apr 2014 14:27:50 -0700, Henry Hallam said:
>
>> 3. Is this sort of thing worth submitting to mainline?
>
> Don't bother doing it unless you have first verified that it
> plays nice with xntpd.

It does; it's strictly instrumentation, and ntpd uses adjtime rather
than settimeofday anyway.  NTP behaves as normal with the module
installed.

Thanks, Henry

"Time police" - best way to submit?

[Josh Cartwright]

On Wed, Apr 16, 2014 at 02:27:50PM -0700, Henry Hallam wrote:
> Hi, my name is Henry and I'm a kernel newbie.

Hello, Henry.

> Recently plagued by a perplexing issue where *something* would step
> the clock on an important server by exactly 5 minutes at infrequent,
> unpredictable intervals, I put together my first kernel module - a
> trivial hook into do_settimeofday via a jprobe, that prints the PID,
> process name and time adjustment to the kernel message log.
> 
> Here's what I have so far:
> https://github.com/henryhallam/timepolice
> 
> Questions:
> 
> 1. Is a module using kprobes/jprobes the right way to do it, vs adding
> printks to linux/time/timekeeping.c?

Have you taken a look at the various kernel tracing features?  I suspect
you could do exactly what you want without modifying the kernel at all
using the function tracer.

  Josh

"Time police" - best way to submit?

[Henry Hallam]

Thanks Josh.  I had not, and I appreciate the pointer.
Henry

On Fri, Apr 18, 2014 at 1:51 PM, Josh Cartwright <joshc@eso.teric.us> wrote:
> On Wed, Apr 16, 2014 at 02:27:50PM -0700, Henry Hallam wrote:
>> Hi, my name is Henry and I'm a kernel newbie.
>
> Hello, Henry.
>
>> Recently plagued by a perplexing issue where *something* would step
>> the clock on an important server by exactly 5 minutes at infrequent,
>> unpredictable intervals, I put together my first kernel module - a
>> trivial hook into do_settimeofday via a jprobe, that prints the PID,
>> process name and time adjustment to the kernel message log.
>>
>> Here's what I have so far:
>> https://github.com/henryhallam/timepolice
>>
>> Questions:
>>
>> 1. Is a module using kprobes/jprobes the right way to do it, vs adding
>> printks to linux/time/timekeeping.c?
>
> Have you taken a look at the various kernel tracing features?  I suspect
> you could do exactly what you want without modifying the kernel at all
> using the function tracer.
>
>   Josh