Wu Zhangjin <wuzhangjin@xxxxxxxxx> writes:
Hi,
> Hi,
>
> On Wed, 2009-05-27 at 11:46 +0200, Arnaud Patard wrote:
>> > Loongson2F add a new capability to dynamic scaling cpu frequency. However the
>> > cpu count timer depends on cpu frequency. So an alternative clock must be used
>> > if this driver is enabled. Besides, the CPU enter wait state when the frequency
>> > is setting zero. All these features help power save.
>> >
>> > In fuloong(2f) and yeeloong(2f), if you want to use this feature, you
>> > should enable the cs5536 mfgpt timer.
> [...]
>> > diff --git a/arch/mips/Kconfig b/arch/mips/Kconfig
>> > index 74efb43..aa8cd64 100644
>> > --- a/arch/mips/Kconfig
>> > +++ b/arch/mips/Kconfig
>> > @@ -2136,6 +2136,23 @@ source "kernel/power/Kconfig"
>> >
>> > endmenu
>> >
>> > +menu "CPU Frequency scaling"
>> > +
>> > +source "drivers/cpufreq/Kconfig"
>> > +
>> > +config LOONGSON2F_CPU_FREQ
>> > + bool "Loongson-2F CPU Frequency driver"
>> > + depends on CPU_LOONGSON2F && CPU_FREQ && CS5536_MFGPT
>>
>> If I have a clock from (for instance) a i8253 compatible source, one will
>> have to add something here. I'm not sure it's a good idea. Did you try
>> with something like "select LOONGSON2F_CPU_FREQ" in the machine Kconfig
>> entry ?
>>
>
> something like "select LOONGSON2F_CPU_FREQ" not work, because users may
> do not need LOONGSON2F_CPU_FREQ.
>
> what about this solution?
>
> +config LOONGSON2F_CPU_FREQ + bool "Loongson-2F CPU Frequency driver"
> + depends on CPU_LOONGSON2F && CPU_FREQ && (CS5536_MFGPT || I8253)
well, I was thinking about something like :
config LOONGSON2F_EXTERNAL_CLOCK
bool
config LOONGSON2F_CPU_FREQ
bool "Loongson-2F CPU Frequency driver"
depends on CPU_LOONGSON2F && CPU_FREQ && LOONGSON2F_EXTERNAL_CLOCK
Maybe it's overkill. Don't know.
>
>
>> > diff --git a/arch/mips/kernel/loongson2f_freq.c b/arch/mips/kernel/loongson2f_freq.c
>> > new file mode 100644
>> > index 0000000..183f36b
> [...]
>> > +static int __init loongson2f_cpufreq_module_init(void)
>> > +{
>> > + struct cpuinfo_mips *c = &cpu_data[0];
>> > + int result;
>> > +
>> > + if (c->processor_id != PRID_IMP_LOONGSON2F)
>> > + return -ENODEV;
>>
>> How can this happen ? the Kconfig entry depends on CPU_LOONGSON2F so I
>> would expect this is useless.
>>
>
> i just removed it, thanks!
>
>> > diff --git a/arch/mips/loongson/common/clock.c b/arch/mips/loongson/common/clock.c
>> > new file mode 100644
>> > index 0000000..a8c648d
>> > --- /dev/null
>> > +++ b/arch/mips/loongson/common/clock.c
> [...]
>> > +/*
>> > + * This is the simple version of Loongson-2F wait
>> > + * Maybe we need do this in interrupt disabled content
>> > + */
>> > +DEFINE_SPINLOCK(loongson2f_wait_lock);
>> > +void loongson2f_cpu_wait(void)
>> > +{
>> > + u32 cpu_freq;
>> > + unsigned long flags;
>> > +
>> > + spin_lock_irqsave(&loongson2f_wait_lock, flags);
>> > + cpu_freq = LOONGSON_CHIPCFG0;
>> > + LOONGSON_CHIPCFG0 &= ~0x7; /* Put CPU into wait mode */
>>
>> by doing this, if you want to "wake up" the 2f, you need an external
>> interrupt source otherwise your only solution is to power down your
>> machine. Are you sure that it's really safe to enable it by default ?
>>
>
> sorry, I am not clear why Yanhua use a loongson-specific cpu_wait
> here(seems used in loongson2f_freq.c)? perhaps he can explain it. and I
> also have a question about it: why not use the "wait" instruction
> directly here like r4k_wait() does? or just use the r4k_wait() function
> via adding some code in arch/mips/kernel/cpu-probe.c:
>From what I remember, the loongson does not have the wait
instruction. That's why r4k_wait() can't be used.
>
> void __init check_wait(void)
> {
> ...
> switch (c->cputype) {
> ...
> case CPU_PR4450:
> case CPU_BCM3302:
> case CPU_CAVIUM_OCTEON:
> + case CPU_LOONGSON2:
> cpu_wait = r4k_wait;
> break;
> ...
> }
>
> and even if we not use cpufreq, is better to enable a loongson-specific
> cpu_wait()? cpu_wait() is called in cpu_idle() (defined in
> arch/mips/kernel/process.c), does we need cpu_wait() to save power when
> cpu is idle?
with this specific cpu_wait(), the loongson is nearly stopped and the
power usage is iirc around 0 to 1 W (this is rather a theorical value,
I've never measured it). The problem is that if you don't have an
external interrupt source, the loongson will stay in sleep state
forever. For instance, if you have only the r4k clocksource, when it'll
reach cpu_wait(), the machine will hang. iirc, depending on the external
clock, in some cases, this may have some side effects. I've played with
that monthes ago so take it with a grain of salt :)
Arnaud
