Loongson2 cpu_wait function

Loongson2 cpu_wait function

[Ralf Baechle]

During the 3.10 merge cycle many MIPS platforms were broken by the
generic idle loop patches.  A patch series to fix this has already been
merged but I'm wondering if fb40bc3e94933007d3e42e96daf1ec8044821cb8
[MIPS: Idle: Re-enable irqs at the end of r3081, au1k and loongson2
cpu_wait.] is sufficient and correct for Loongson 2.

In particular:

 o drivers/cpufreq/loongson2_cpufreq.c protects accesses to LOONGSON_CHIPCFG0
   in loongson2_cpu_wait with a spinlock.  This spinlock is not used anywhere
   else in the kernel so it would appear there is still a race with other
   accesses to LOONGSON_CHIPCFG0.
 o It's not SMPly correct - even if cpufreq_exit restores the old value of
   cpu_wait on a SMP system another processor might still be executing
   loongson2_cpu_wait().
 o I'd appreciate if at least some basic power saving would be used even if
   CONFIG_LOONGSON2_CPUFREQ was disabled, that is loongson2_cpu_wait should
   go back to arch/mips/kernel/idle.c.
 o Could somebody test if Loongson 2 is working?  Thanks!

  Ralf

Re: Loongson2 cpu_wait function

[Aaro Koskinen]

Hi,

On Thu, May 23, 2013 at 02:27:37PM +0200, Ralf Baechle wrote:
> During the 3.10 merge cycle many MIPS platforms were broken by the
> generic idle loop patches.  A patch series to fix this has already been
> merged but I'm wondering if fb40bc3e94933007d3e42e96daf1ec8044821cb8
> [MIPS: Idle: Re-enable irqs at the end of r3081, au1k and loongson2
> cpu_wait.] is sufficient and correct for Loongson 2.
> 
> In particular:
> 
>  o drivers/cpufreq/loongson2_cpufreq.c protects accesses to LOONGSON_CHIPCFG0
>    in loongson2_cpu_wait with a spinlock.  This spinlock is not used anywhere
>    else in the kernel so it would appear there is still a race with other
>    accesses to LOONGSON_CHIPCFG0.
>  o It's not SMPly correct - even if cpufreq_exit restores the old value of
>    cpu_wait on a SMP system another processor might still be executing
>    loongson2_cpu_wait().

I think Loongson2 is UP-only?

>  o I'd appreciate if at least some basic power saving would be used even if
>    CONFIG_LOONGSON2_CPUFREQ was disabled, that is loongson2_cpu_wait should
>    go back to arch/mips/kernel/idle.c.
>  o Could somebody test if Loongson 2 is working?  Thanks!

It works (tested d97955625710b57f24427e403f150126078273c2), but cpufreq
seems to be broken for some reason:

[    6.136000] calling  cpufreq_init+0x0/0x8c @ 1
[    6.140000] cpufreq: Loongson-2F CPU frequency driver.
[    6.144000] initcall cpufreq_init+0x0/0x8c returned -19 after 4000 usecs

I would also like to remind that the boot is still unreliable without
this patch: http://patchwork.linux-mips.org/patch/4958/

A.