| Message ID | 1482169657-15773-1-git-send-email-olof@lixom.net (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Mon, Dec 19, 2016 at 9:47 AM, Olof Johansson <olof@lixom.net> wrote: > This is useful to get an indication of how much time we spent in firmware. > > It's not guaranteed that the timer started at 0 on reset, so it's just > an approximation, and might very well be invalid on some systems. But > it's still a useful metric to have access to. > > Signed-off-by: Olof Johansson <olof@lixom.net> Friendly ping. Would be nice to see this land in 4.11. Thanks, -Olof
On Mon, Dec 19, 2016 at 09:47:37AM -0800, Olof Johansson wrote: > This is useful to get an indication of how much time we spent in firmware. > > It's not guaranteed that the timer started at 0 on reset, so it's just > an approximation, and might very well be invalid on some systems. But > it's still a useful metric to have access to. Hi Olof, [ ... ] > --- a/drivers/clocksource/arm_arch_timer.c > +++ b/drivers/clocksource/arm_arch_timer.c > @@ -521,6 +521,8 @@ arch_timer_detect_rate(void __iomem *cntbase, struct device_node *np) > > static void arch_timer_banner(unsigned type) > { > + unsigned long cnt = arch_timer_read_counter(); > + arch_timer_banner() is called before arch_counter_register() where the arch_timer_read_counter() function pointer is set. Perhaps the arch_timer_banner() and arch_counter_register() should be swapped in arch_timer_common_init(). > pr_info("Architected %s%s%s timer(s) running at %lu.%02luMHz (%s%s%s).\n", > type & ARCH_CP15_TIMER ? "cp15" : "", > type == (ARCH_CP15_TIMER | ARCH_MEM_TIMER) ? " and " : "", > @@ -534,6 +536,8 @@ static void arch_timer_banner(unsigned type) > type & ARCH_MEM_TIMER ? > arch_timer_mem_use_virtual ? "virt" : "phys" : > ""); > + pr_info("Initial timer value: 0x%lx: %ld.%02lds\n", > + cnt, cnt/arch_timer_rate, (cnt/(arch_timer_rate/100)) % 100); > } >
On Sat, Feb 04, 2017 at 09:41:18AM +0100, Daniel Lezcano wrote: > On Mon, Dec 19, 2016 at 09:47:37AM -0800, Olof Johansson wrote: > > This is useful to get an indication of how much time we spent in firmware. > > > > It's not guaranteed that the timer started at 0 on reset, so it's just > > an approximation, and might very well be invalid on some systems. But > > it's still a useful metric to have access to. > > Hi Olof, > > [ ... ] > > > --- a/drivers/clocksource/arm_arch_timer.c > > +++ b/drivers/clocksource/arm_arch_timer.c > > @@ -521,6 +521,8 @@ arch_timer_detect_rate(void __iomem *cntbase, struct device_node *np) > > > > static void arch_timer_banner(unsigned type) > > { > > + unsigned long cnt = arch_timer_read_counter(); > > + > > arch_timer_banner() is called before arch_counter_register() where the > arch_timer_read_counter() function pointer is set. > > Perhaps the arch_timer_banner() and arch_counter_register() should be swapped in > arch_timer_common_init(). That would make sense to me. > > pr_info("Architected %s%s%s timer(s) running at %lu.%02luMHz (%s%s%s).\n", > > type & ARCH_CP15_TIMER ? "cp15" : "", > > type == (ARCH_CP15_TIMER | ARCH_MEM_TIMER) ? " and " : "", > > @@ -534,6 +536,8 @@ static void arch_timer_banner(unsigned type) > > type & ARCH_MEM_TIMER ? > > arch_timer_mem_use_virtual ? "virt" : "phys" : > > ""); > > + pr_info("Initial timer value: 0x%lx: %ld.%02lds\n", > > + cnt, cnt/arch_timer_rate, (cnt/(arch_timer_rate/100)) % 100); Our tiemrs should be precise enough to give us a few more digits here (e.g. down to ns, like printk). Are there any helpers we can use to do that? It would also be nice to log which counter we're reading from. Thanks, Mark.
diff --git a/drivers/clocksource/arm_arch_timer.c b/drivers/clocksource/arm_arch_timer.c index 02fef68..c26078b 100644 --- a/drivers/clocksource/arm_arch_timer.c +++ b/drivers/clocksource/arm_arch_timer.c @@ -521,6 +521,8 @@ arch_timer_detect_rate(void __iomem *cntbase, struct device_node *np) static void arch_timer_banner(unsigned type) { + unsigned long cnt = arch_timer_read_counter(); + pr_info("Architected %s%s%s timer(s) running at %lu.%02luMHz (%s%s%s).\n", type & ARCH_CP15_TIMER ? "cp15" : "", type == (ARCH_CP15_TIMER | ARCH_MEM_TIMER) ? " and " : "", @@ -534,6 +536,8 @@ static void arch_timer_banner(unsigned type) type & ARCH_MEM_TIMER ? arch_timer_mem_use_virtual ? "virt" : "phys" : ""); + pr_info("Initial timer value: 0x%lx: %ld.%02lds\n", + cnt, cnt/arch_timer_rate, (cnt/(arch_timer_rate/100)) % 100); } u32 arch_timer_get_rate(void)
This is useful to get an indication of how much time we spent in firmware. It's not guaranteed that the timer started at 0 on reset, so it's just an approximation, and might very well be invalid on some systems. But it's still a useful metric to have access to. Signed-off-by: Olof Johansson <olof@lixom.net> --- drivers/clocksource/arm_arch_timer.c | 4 ++++ 1 file changed, 4 insertions(+)