cvs commit: src/share/man/man4 coretemp.4 src/sys/dev/coretemp coretemp.c

[Rui Paulo]

On 14 Oct 2007, at 19:48, Nate Lawson wrote:

> Alexander Leidinger wrote:
>> netchild    2007-10-14 10:59:44 UTC
>>
>>   FreeBSD src repository
>>
>>   Modified files:
>>     share/man/man4       coretemp.4
>>     sys/dev/coretemp     coretemp.c
>>   Log:
>>   Convert coretemp(4) to the hardware sensors framework and
>>   make sure to never call sched_bind() for uninitialised CPUs.
>>
>>   Submitted by:   Constantine A. Murenin <cnst at FreeBSD.org>
>>   Sponsored by:   Google Summer of Code 2007 (GSoC2007/cnst-sensors)
>>   Mentored by:    syrinx
>>   Tested by:      many
>>   OKed by:        kensmith
>>
>>   Revision  Changes    Path
>>   1.2       +14 -6     src/share/man/man4/coretemp.4
>>   1.3       +46 -35    src/sys/dev/coretemp/coretemp.c
>
>> -	thread_lock(curthread);
>> -	sched_unbind(curthread);
>> -	thread_unlock(curthread);
>> +	 * Bind to specific CPU to read the correct temperature.
>> +	 * If not all CPUs are initialised, then only read from
>> +	 * cpu0, returning -1 on all other CPUs.
>> +	 */
>> +	if (smp_cpus > 1) {
>> +		thread_lock(curthread);
>> +		sched_bind(curthread, cpu);
>> +		msr = rdmsr(MSR_THERM_STATUS);
>> +		sched_unbind(curthread);
>> +		thread_unlock(curthread);
>> +	} else if (cpu != 0)
>> +		return (-1);
>> +	else
>> +		msr = rdmsr(MSR_THERM_STATUS);
>
> I reviewed this part of the patch for Rui Paulo and told him to use
> smp_active instead of smp_cpus > 1.

Sorry, for not getting back to you sooner, but there's a better way  
than that. I briefly chatted with jhb and the proper way is to use  
SYSINIT(9). This way the module works on UP and SMP machines.
I have a patch, but my devel machine has panic()ed.

Regards.
--
Rui Paulo