Re: [PATCH 1/7] x86: Limit the number of processor bootup messages

[Yinghai Lu <yhlu.kernel@gmail.com>]

On Fri, Nov 13, 2009 at 5:46 AM, Mike Travis <travis@sgi.com> wrote:
>
>
> Yinghai Lu wrote:
>>
>> On Thu, Nov 12, 2009 at 9:19 AM, Mike Travis <travis@sgi.com> wrote:
>>>
>>> With a large number of processors in a system there is an excessive
>>> amount
>>> of messages sent to the system console.  It's estimated that with 4096
>>> processors in a system, and the console baudrate set to 56K, the startup
>>> messages will take about 84 minutes to clear the serial port.
>>>
>>> This set of patches limits the number of repetitious messages which
>>> contain
>>> no additional information.  Much of this information is obtainable from
>>> the
>>> /proc and /sysfs.   Most of the messages are also sent to the kernel log
>>> buffer as KERN_DEBUG messages so it can be used to examine more closely
>>> any
>>> details specific to a processor.
>>>
>>> The list of message transformations....
>>>
>>> For system_state == SYSTEM_BOOTING:
>>>
>>> [   18.669304] Booting Node   0, Processors  1 2 3 4 5 6 7 Ok.
>>> [   19.321065] Booting Node   1, Processors  8 9 10 11 12 13 14 15 Ok.
>>> [   20.065325] Booting Node   2, Processors  16 17 18 19 20 21 22 23 Ok.
>>> ..
>>> [  117.153053] Booting Node  63, Processors  1016 1017 1018 1019 1020
>>> 1021 1022 1023 Ok.
>>> [  117.952235] Brought up 1024 CPUs
>>>
>>> Timing shows that with NO bootup messages, it takes only slightly less
>>> time
>>> so printing adds very little overhead:
>>>
>>> [   18.670219] Booting all processors
>>> [  117.180248] Brought up 1024 CPUs
>>>
>>> For Processor Information printout, the specifics of CPU0 are printed and
>>> then at the end of the bootup sequence, a summary is printed:
>>>
>>> [  117.957682] Processor Information for CPUS: 0-191,208-223,256-271,...
>>> [  117.968034] Genuine Intel(R) CPU             0000 @ 2.13GHz stepping
>>> 04
>>> [  117.977406] BogoMIPS: MIN 3989.01 (7978031) AVG 4266.62 (8533249) MAX
>>> 4537.51 (9075028)
>>> [  117.984496] Processor Information for CPUS:
>>> 192-207,240-255,272-287,...
>>> [  117.996032] Genuine Intel(R) CPU             0000 @ 2.13GHz stepping
>>> 03
>>> [  118.001404] BogoMIPS: MIN 4021.49 (8042995) AVG 4265.91 (8531833) MAX
>>> 4479.79 (8959584)
>>> [  118.012373] Processor Information for CPUS: 224-239,736-751
>>> [  118.020032] Genuine Intel(R) CPU             0000 @ 1.87GHz stepping
>>> 03
>>> [  118.028033] BogoMIPS: MIN 3733.92 (7467855) AVG 3746.96 (7493933) MAX
>>> 3939.52 (7879056)
>>> [  118.036360] Processor Information for CPUS:
>>> 320-335,384-415,432-447,...
>>> [  118.044032] Intel(R) Xeon(R) CPU           X7560  @ 2.27GHz stepping
>>> 05
>>> [  118.053404] BogoMIPS: MIN 4244.65 (8489318) AVG 4532.45 (9064917) MAX
>>> 4666.80 (9333604)
>>> [  118.060644] Total of 1024 processors activated (4386353.46 BogoMIPS).
>>>
>>>
>>> The following lines have been removed:
>>>
>>>       CPU: Physical Processor ID:
>>>       CPU: Processor Core ID:
>>>       CPU %d/0x%x -> Node %d
>>
>> why?
>
> Someone asked that they be removed?
>
> The intent of the patch was to remove repetitious per cpu printouts where
> the
> information is easily available via some other means, which is the case of
> the
> above 3 lines.  Would you prefer they stay as KERN_DEBUG or pr_debug lines?
> (I had it this way in the first version but then was told that the messages
> were
> not needed.)

at least on one distribution kernel ( SLES 11?), when BSP is from
socket1, the kernel is not happy
(thought that cpu is from socket 0).

YH