* Wu Zhangjin <wuzhangjin@xxxxxxxxx> wrote:
> From: Wu Zhangjin <wuzhangjin@xxxxxxxxx>
>
> (This v4 revision incorporates with the feedbacks from Sergei.)
>
> This patch adds a cnt32_to_63() and MIPS c0 count based sched_clock(),
> which provides high resolution. and also, one new kernel option
> (HR_SCHED_CLOCK) is added to enable/disable this sched_clock().
>
> Without it, the Ftrace for MIPS will give useless timestamp information.
>
> Because cnt32_to_63() needs to be called at least once per half period
> to work properly, Differ from the old version, this v2 revision set up a
> kernel timer to ensure the requirement of some MIPSs which have short c0
> count period.
>
> Signed-off-by: Wu Zhangjin <wuzhangjin@xxxxxxxxx>
> ---
> arch/mips/Kconfig | 18 +++++++++++++
> arch/mips/kernel/csrc-r4k.c | 57 +++++++++++++++++++++++++++++++++++++++++++
> 2 files changed, 75 insertions(+), 0 deletions(-)
Looks good in general, just a few small nits:
> diff --git a/arch/mips/Kconfig b/arch/mips/Kconfig
> index b342197..5ea55f5 100644
> --- a/arch/mips/Kconfig
> +++ b/arch/mips/Kconfig
> @@ -1952,6 +1952,24 @@ config NR_CPUS
> source "kernel/time/Kconfig"
>
> #
> +# High Resolution sched_clock() Configuration
> +#
> +
> +config HR_SCHED_CLOCK
> + bool "High Resolution sched_clock()"
> + depends on CSRC_R4K
> + default n
> + help
> + This option enables the MIPS c0 count based high resolution
> + sched_clock().
> +
> + If you need a ns precision timestamp, you are recommended to enable
> + this option. For example, If you are using the Ftrace subsystem to do
s/If/if
> + real time tracing, this option is needed.
> +
> + If unsure, disable it.
> +
> +#
> # Timer Interrupt Frequency Configuration
> #
>
> diff --git a/arch/mips/kernel/csrc-r4k.c b/arch/mips/kernel/csrc-r4k.c
> index e95a3cd..7762c95 100644
> --- a/arch/mips/kernel/csrc-r4k.c
> +++ b/arch/mips/kernel/csrc-r4k.c
> @@ -10,6 +10,61 @@
>
> #include <asm/time.h>
>
> +#ifdef CONFIG_HR_SCHED_CLOCK
> +#include <linux/cnt32_to_63.h>
> +#include <linux/timer.h>
> +
> +/*
> + * MIPS sched_clock implementation.
> + *
> + * because cnt32_to_63() needs to be called at least once per half period to
> + * work properly, and some of the MIPS frequency is high, perhaps a kernel
> + * timer is needed to be set up to ensure this requirement is always met.
> + * Please refer to arch/arm/plat-orion/time.c and include/linux/cnt32_to_63.h
> + */
> +static unsigned long __read_mostly tclk2ns_scale;
> +static unsigned long __read_mostly tclk2ns_scale_factor;
> +
> +unsigned long long notrace sched_clock(void)
> +{
> + unsigned long long v = cnt32_to_63(read_c0_count());
> + return (v * tclk2ns_scale) >> tclk2ns_scale_factor;
> +}
> +
> +static struct timer_list cnt32_to_63_keepwarm_timer;
> +
> +static void cnt32_to_63_keepwarm(unsigned long data)
> +{
> + mod_timer(&cnt32_to_63_keepwarm_timer, round_jiffies(jiffies + data));
> + (void) sched_clock();
sched_clock() isnt __must_check so the cast of the return value seems
unnecessary.
> +}
> +
> +static void __init setup_sched_clock(struct clocksource *cs, unsigned long tclk)
> +{
> + unsigned long long v;
> + unsigned long data;
> +
> + v = cs->mult;
> + /*
> + * We want an even value to automatically clear the top bit
> + * returned by cnt32_to_63() without an additional run time
> + * instruction. So if the LSB is 1 then round it up.
> + */
> + if (v & 1)
> + v++;
> + tclk2ns_scale = v;
> + tclk2ns_scale_factor = cs->shift;
> +
> + data = 0x80000000 / tclk * HZ;
> + setup_timer(&cnt32_to_63_keepwarm_timer, cnt32_to_63_keepwarm, data);
> + mod_timer(&cnt32_to_63_keepwarm_timer, round_jiffies(jiffies + data));
> +}
> +#else /* !CONFIG_HR_SCHED_CLOCK */
> +static void __init setup_sched_clock(struct clocksource *cs, unsigned long tclk)
> +{
> +}
> +#endif
Could this be put into a new, separate file,
arch/mips/kernel/csrc-r4k-hres.c, to avoid the ugly #ifdef block?
> +
> static cycle_t c0_hpt_read(struct clocksource *cs)
> {
> return read_c0_count();
> @@ -32,6 +87,8 @@ int __init init_r4k_clocksource(void)
>
> clocksource_set_clock(&clocksource_mips, mips_hpt_frequency);
>
> + setup_sched_clock(&clocksource_mips, mips_hpt_frequency);
And i'd suggest to rename it to setup_hres_sched_clock() - as i presume
this system has a sched_clock() even if CONFIG_HR_SCHED_CLOCK is
disabled.
Thanks,
Ingo
