lkml.org 
[lkml]   [2012]   [Aug]   [3]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
Date
SubjectRe: [PATCH v2] printk: add option to print cpu id
On Fri, Aug 3, 2012 at 11:56 AM, Pandita, Vikram <vikram.pandita@ti.com> wrote:
> On Fri, Aug 3, 2012 at 2:48 AM, Kay Sievers <kay@vrfy.org> wrote:
>> On Fri, Aug 3, 2012 at 11:36 AM, Pandita, Vikram <vikram.pandita@ti.com> wrote:
>>> On Fri, Aug 3, 2012 at 2:32 AM, Venu Byravarasu <vbyravarasu@nvidia.com> wrote:
>>
>>>> As having Macro locally in the file of interest would serve the purpose,
>>>> Why to change the printk code?
>>>
>>> As stated, the logic followed is exactly similar to well proven and
>>> approved way to handle printk time stamp: CONFIG_PRINTK_TIME
>>> Its an overhead on the system that enables the config option:
>>> CONFIG_PRINTK_CPUID
>>> Otherwise the system remains as is.
>>>
>>> To gain insight on SMP system logging behavior, the price to pay is
>>> the extra 4 chars per printk line,
>>> just like printk-time adds 15 chars to each line. Both options can be
>>> disabled as desired.
>>>
>>> So i am not sure what kind of option you are proposing?
>>> Do u imply PRINTK_TIME is not right then?
>>
>> It's 8 bytes per message for storing the timestamp in the records.
>> There is never 15 bytes storage space needed, the prefix is
>> constructed on-the-fly only while exporting the data.
>
> When i was referring to 15 chars, its coming from here:
> Its the space reserved in each line of output. Corresponding space for
> cpuid is 4 chars: "[x] ":

Just saying, that's just the length of the printed line to the console
or syslog, there is no reservation or space used for that internally.

>> The CPU-ID would need at least two additional bytes (2^16 CPUS) in
>> every record, unless it's a compile-time option.
>
> are u proposing:
> a) to make cpuid a u16?

That would be needed, I guess. We easily have server systems with more
than 255 CPUs. It will only be a matter of time, that the number of
CPUs will increase for everybody, I guess.

> b) to put cpuid in struct cont and struct log - under the #ifdef macro?

As said, I really can't tell how generally useful it is, and if people
think that it should be there unconditionally, should not be there at
all, or as a compile time option. Others might have an opinion on
that.

I personally never missed the CPU-ID in the logs. I personally would
find the PID/task ID more interesting, and even that I never really
missed. :)

Kay


\
 
 \ /
  Last update: 2012-08-03 12:41    [W:0.052 / U:2.376 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site