linux-mips
[Top] [All Lists]

Re: [PATCH v1 4/5] mips specific clock function to get precise timestamp

To: wuzhangjin@gmail.com
Subject: Re: [PATCH v1 4/5] mips specific clock function to get precise timestamp
From: Steven Rostedt <rostedt@goodmis.org>
Date: Thu, 28 May 2009 22:06:42 -0400 (EDT)
Cc: linux-mips@linux-mips.org, linux-kernel@vger.kernel.org, Wu Zhangjin <wuzj@lemote.com>, Ralf Baechle <ralf@linux-mips.org>, Ingo Molnar <mingo@elte.hu>, Andrew Morton <akpm@linux-foundation.org>, Thomas Gleixner <tglx@linutronix.de>, Nicholas Mc Guire <der.herr@hofr.at>
In-reply-to: <d44adaf39284074c93a8dba0209da5b31d0d3c49.1243543471.git.wuzj@lemote.com>
Original-recipient: rfc822;linux-mips@linux-mips.org
References: <cover.1243543471.git.wuzj@lemote.com> <d44adaf39284074c93a8dba0209da5b31d0d3c49.1243543471.git.wuzj@lemote.com>
Sender: linux-mips-bounce@linux-mips.org
User-agent: Alpine 2.00 (DEB 1167 2008-08-23)
On Fri, 29 May 2009, wuzhangjin@gmail.com wrote:

> From: Wu Zhangjin <wuzj@lemote.com>
> 
> by default, ring_buffer_time_stamp calling sched_clock(jiffies-based)
> function to get timestamp, in x86, there is a tsc(64bit) based
> sched_clock, but in mips, the 'tsc'(clock counter) is only 32bit long,
> which will easily rollover, and there is no precise sched_clock in mips,
> we need to get one ourselves.
> 
> to avoid invading the whole linux-mips, i do not want to implement a
> tsc-based native_sched_clock like x86 does. because, there is a need to
> handling rollover of the only 32-bit long 'tsc' of mips, which will need
> extra overhead. in reality, i have tried to implement one(just like the
> ring_buffer_time_stamp here does), but make the kernel hangs when
> booting, i am not sure why it not work.
> 
> herein, i just implement a mips-specific ring_buffer_time_stamp in
> arch/mips/kernel/ftrace.c via adding  __attribute__((weak)) before
> ring_buffer_time_stamp(...) {} in kernel/trace/ring_buffer.c and do
> something in arch/mips/kernel/ftrace.c like this:
> 
> u64  ring_buffer_time_stamp \
>        __attribute__((alias("native_ring_buffer_time_stamp")));
> 
> and, as the same, there is also a need to implement a mips-specific
> trace_clock_local based on the above ring_buffer_timep_stamp, this clock
> function is called in function graph tracer to get calltime & rettime of
> a function.
> 
> and what about the trace_clock and trace_clock_global function, should
> we also implement a mips-secific one? i am not sure.
> 
> Signed-off-by: Wu Zhangjin <wuzj@lemote.com>
> ---
>  arch/mips/kernel/Makefile       |    2 +
>  arch/mips/kernel/csrc-r4k.c     |    2 +-
>  arch/mips/kernel/ftrace_clock.c |   77 
> +++++++++++++++++++++++++++++++++++++++
>  kernel/trace/ring_buffer.c      |    3 +-
>  kernel/trace/trace_clock.c      |    2 +-
>  5 files changed, 83 insertions(+), 3 deletions(-)
>  create mode 100644 arch/mips/kernel/ftrace_clock.c
> 
> diff --git a/arch/mips/kernel/Makefile b/arch/mips/kernel/Makefile
> index 6b1a8a5..5dec76f 100644
> --- a/arch/mips/kernel/Makefile
> +++ b/arch/mips/kernel/Makefile
> @@ -12,6 +12,7 @@ ifdef CONFIG_FUNCTION_TRACER
>  # Do not profile debug and lowlevel utilities
>  CFLAGS_REMOVE_mcount.o = -pg
>  CFLAGS_REMOVE_ftrace.o = -pg
> +CFLAGS_REMOVE_ftrace_clock.o = -pg
>  CFLAGS_REMOVE_early_printk.o = -pg
>  endif
>  
> @@ -33,6 +34,7 @@ obj-$(CONFIG_MODULES)               += mips_ksyms.o module.o
>  
>  obj-$(CONFIG_FUNCTION_TRACER)        += mcount.o
>  obj-$(CONFIG_FUNCTION_TRACER)        += ftrace.o
> +obj-$(CONFIG_NOP_TRACER)     += ftrace_clock.o
>  
>  obj-$(CONFIG_CPU_LOONGSON2)  += r4k_fpu.o r4k_switch.o
>  obj-$(CONFIG_CPU_MIPS32)     += r4k_fpu.o r4k_switch.o
> diff --git a/arch/mips/kernel/csrc-r4k.c b/arch/mips/kernel/csrc-r4k.c
> index e95a3cd..3da1c7a 100644
> --- a/arch/mips/kernel/csrc-r4k.c
> +++ b/arch/mips/kernel/csrc-r4k.c
> @@ -10,7 +10,7 @@
>  
>  #include <asm/time.h>
>  
> -static cycle_t c0_hpt_read(struct clocksource *cs)
> +static cycle_t notrace c0_hpt_read(struct clocksource *cs)
>  {
>       return read_c0_count();
>  }
> diff --git a/arch/mips/kernel/ftrace_clock.c b/arch/mips/kernel/ftrace_clock.c
> new file mode 100644
> index 0000000..2f3b05a
> --- /dev/null
> +++ b/arch/mips/kernel/ftrace_clock.c
> @@ -0,0 +1,77 @@
> +/*
> + * This file is subject to the terms and conditions of the GNU General Public
> + * License.  See the file "COPYING" in the main directory of this archive for
> + * more details.
> + *
> + * Copyright (C) 2009 DSLab, Lanzhou University, China
> + * Author: Wu Zhangjin <wuzj@lemote.com>
> + */
> +
> +#include <linux/types.h>
> +#include <linux/sched.h>
> +#include <linux/jiffies.h>
> +#include <linux/clocksource.h>
> +#include <linux/ring_buffer.h>
> +
> +/* Up this if you want to test the TIME_EXTENTS and normalization */
> +#ifndef DEBUG_SHIFT
> +#define DEBUG_SHIFT 0
> +#endif
> +
> +/* mips-specific ring_buffer_time_stamp implementation,
> + * the original one is defined in kernel/trace/ring_buffer.c
> + */
> +
> +u64 native_ring_buffer_time_stamp(struct ring_buffer *buffer, int cpu)
> +{
> +     u64 current_cycles;
> +     static unsigned long old_jiffies;
> +     static u64 time, old_cycles;
> +
> +     preempt_disable_notrace();
> +    /* update timestamp to avoid missing the timer interrupt */
> +     if (time_before(jiffies, old_jiffies)) {
> +             old_jiffies = jiffies;
> +             time = sched_clock();
> +             old_cycles = clock->cycle_last;
> +     }
> +     current_cycles = clock->read(clock);
> +
> +     time = (time + cyc2ns(clock, (current_cycles - old_cycles)
> +                             & clock->mask)) << DEBUG_SHIFT;
> +
> +     old_cycles = current_cycles;
> +     preempt_enable_no_resched_notrace();
> +
> +     return time;
> +}
> +
> +u64 ring_buffer_time_stamp(struct ring_buffer *buffer, int cpu)
> +             __attribute__((alias("native_ring_buffer_time_stamp")));
> +
> +/*
> + * native_trace_clock_local(): the simplest and least coherent tracing clock.
> + *
> + * Useful for tracing that does not cross to other CPUs nor
> + * does it go through idle events.
> + */
> +u64 native_trace_clock_local(void)
> +{
> +     unsigned long flags;
> +     u64 clock;
> +
> +     /*
> +      * sched_clock() is an architecture implemented, fast, scalable,
> +      * lockless clock. It is not guaranteed to be coherent across
> +      * CPUs, nor across CPU idle events.
> +      */
> +     raw_local_irq_save(flags);
> +     clock = ring_buffer_time_stamp(NULL, raw_smp_processor_id());
> +     raw_local_irq_restore(flags);
> +
> +     return clock;
> +}
> +
> +u64 trace_clock_local(void)
> +             __attribute__((alias("native_trace_clock_local")));
> +
> diff --git a/kernel/trace/ring_buffer.c b/kernel/trace/ring_buffer.c
> index 960cbf4..717bd8e 100644
> --- a/kernel/trace/ring_buffer.c
> +++ b/kernel/trace/ring_buffer.c
> @@ -420,7 +420,8 @@ struct ring_buffer_iter {
>  /* Up this if you want to test the TIME_EXTENTS and normalization */
>  #define DEBUG_SHIFT 0
>  
> -u64 ring_buffer_time_stamp(struct ring_buffer *buffer, int cpu)
> +u64 __attribute__((weak)) ring_buffer_time_stamp(struct ring_buffer *buffer,
> +                             int cpu)
>  {

Note, you want to look at what is in tip:

http://people.redhat.com/~mingo/tip.git/README

The latest ftrace code is there. I changed the ring_buffer_time_stamp to 
use an individule buffer clock, that defaults to trace_clock_local.
Then only the trace_clock_local needs to be weak.

-- Steve


>       u64 time;
>  
> diff --git a/kernel/trace/trace_clock.c b/kernel/trace/trace_clock.c
> index b588fd8..78c98c8 100644
> --- a/kernel/trace/trace_clock.c
> +++ b/kernel/trace/trace_clock.c
> @@ -26,7 +26,7 @@
>   * Useful for tracing that does not cross to other CPUs nor
>   * does it go through idle events.
>   */
> -u64 notrace trace_clock_local(void)
> +u64 __attribute__((weak)) notrace trace_clock_local(void)
>  {
>       unsigned long flags;
>       u64 clock;
> -- 
> 1.6.0.4
> 
> 

<Prev in Thread] Current Thread [Next in Thread>