| Message ID | 20200710004317.3017137-1-sjg@chromium.org |
|---|---|
| State | Superseded |
| Headers | show |
| Series | [v3,1/4] timer: Allow delays with a 32-bit microsecond timer | expand |
On Fri, Jul 10, 2020 at 8:49 AM Simon Glass <sjg at chromium.org> wrote: > > The current get_timer_us() uses 64-bit arithmetic on 32-bit machines. > When implementing microsecond-level timeouts, 32-bits is plenty. Add a > new function that uses an unsigned long. On 64-bit machines this is > still 64-bit, but this doesn't introduce a penalty. On 32-bit machines > it is more efficient. > > Signed-off-by: Simon Glass <sjg at chromium.org> > --- > > Changes in v3: > - Expand the commit message and function comment > > include/time.h | 11 +++++++++++ > lib/time.c | 5 +++++ > 2 files changed, 16 insertions(+) > > diff --git a/include/time.h b/include/time.h > index e99f9c8012..3f00e68713 100644 > --- a/include/time.h > +++ b/include/time.h > @@ -17,6 +17,17 @@ unsigned long get_timer(unsigned long base); > unsigned long timer_get_us(void); > uint64_t get_timer_us(uint64_t base); > > +/** > + * get_timer_us_long() - Get the number of elapsed microseconds > + * > + * This uses 32-bit arithmetic on 32-bit machines, which is enough to handle > + * delays of over an hour. For 64-bit machines it uses a 64-bit value. > + * > + *@base: Base time to consider > + *@return elapsed time since @base > + */ > +unsigned long get_timer_us_long(unsigned long base); > + > /* > * timer_test_add_offset() > * > diff --git a/lib/time.c b/lib/time.c > index 65db0f6cda..47f8c84327 100644 > --- a/lib/time.c > +++ b/lib/time.c > @@ -152,6 +152,11 @@ uint64_t __weak get_timer_us(uint64_t base) > return tick_to_time_us(get_ticks()) - base; > } > > +unsigned long __weak get_timer_us_long(unsigned long base) > +{ > + return timer_get_us() - base; > +} > + > unsigned long __weak notrace timer_get_us(void) > { > return tick_to_time(get_ticks() * 1000); > -- Reviewed-by: Bin Meng <bmeng.cn at gmail.com> Some day we should clean up the timer APIs. We have get_timer_us(), and now get_timer_us_long(), but we also have timer_get_us() that returns long value! Regards, Bin
On Mon, Jul 13, 2020 at 10:13 AM Bin Meng <bmeng.cn at gmail.com> wrote: > > On Fri, Jul 10, 2020 at 8:49 AM Simon Glass <sjg at chromium.org> wrote: > > > > The current get_timer_us() uses 64-bit arithmetic on 32-bit machines. > > When implementing microsecond-level timeouts, 32-bits is plenty. Add a > > new function that uses an unsigned long. On 64-bit machines this is > > still 64-bit, but this doesn't introduce a penalty. On 32-bit machines > > it is more efficient. > > > > Signed-off-by: Simon Glass <sjg at chromium.org> > > --- > > > > Changes in v3: > > - Expand the commit message and function comment > > > > include/time.h | 11 +++++++++++ > > lib/time.c | 5 +++++ > > 2 files changed, 16 insertions(+) > > > > diff --git a/include/time.h b/include/time.h > > index e99f9c8012..3f00e68713 100644 > > --- a/include/time.h > > +++ b/include/time.h > > @@ -17,6 +17,17 @@ unsigned long get_timer(unsigned long base); > > unsigned long timer_get_us(void); > > uint64_t get_timer_us(uint64_t base); > > > > +/** > > + * get_timer_us_long() - Get the number of elapsed microseconds > > + * > > + * This uses 32-bit arithmetic on 32-bit machines, which is enough to handle > > + * delays of over an hour. For 64-bit machines it uses a 64-bit value. > > + * > > + *@base: Base time to consider > > + *@return elapsed time since @base > > + */ > > +unsigned long get_timer_us_long(unsigned long base); > > + > > /* > > * timer_test_add_offset() > > * > > diff --git a/lib/time.c b/lib/time.c > > index 65db0f6cda..47f8c84327 100644 > > --- a/lib/time.c > > +++ b/lib/time.c > > @@ -152,6 +152,11 @@ uint64_t __weak get_timer_us(uint64_t base) > > return tick_to_time_us(get_ticks()) - base; > > } > > > > +unsigned long __weak get_timer_us_long(unsigned long base) > > +{ > > + return timer_get_us() - base; > > +} > > + > > unsigned long __weak notrace timer_get_us(void) > > { > > return tick_to_time(get_ticks() * 1000); > > -- > > Reviewed-by: Bin Meng <bmeng.cn at gmail.com> > > Some day we should clean up the timer APIs. We have get_timer_us(), > and now get_timer_us_long(), but we also have timer_get_us() that > returns long value! applied to u-boot-x86, thanks!
Hi Bin, On Sun, 12 Jul 2020 at 20:13, Bin Meng <bmeng.cn at gmail.com> wrote: > > On Fri, Jul 10, 2020 at 8:49 AM Simon Glass <sjg at chromium.org> wrote: > > > > The current get_timer_us() uses 64-bit arithmetic on 32-bit machines. > > When implementing microsecond-level timeouts, 32-bits is plenty. Add a > > new function that uses an unsigned long. On 64-bit machines this is > > still 64-bit, but this doesn't introduce a penalty. On 32-bit machines > > it is more efficient. > > > > Signed-off-by: Simon Glass <sjg at chromium.org> > > --- > > > > Changes in v3: > > - Expand the commit message and function comment > > > > include/time.h | 11 +++++++++++ > > lib/time.c | 5 +++++ > > 2 files changed, 16 insertions(+) > > > > diff --git a/include/time.h b/include/time.h > > index e99f9c8012..3f00e68713 100644 > > --- a/include/time.h > > +++ b/include/time.h > > @@ -17,6 +17,17 @@ unsigned long get_timer(unsigned long base); > > unsigned long timer_get_us(void); > > uint64_t get_timer_us(uint64_t base); > > > > +/** > > + * get_timer_us_long() - Get the number of elapsed microseconds > > + * > > + * This uses 32-bit arithmetic on 32-bit machines, which is enough to handle > > + * delays of over an hour. For 64-bit machines it uses a 64-bit value. > > + * > > + *@base: Base time to consider > > + *@return elapsed time since @base > > + */ > > +unsigned long get_timer_us_long(unsigned long base); > > + > > /* > > * timer_test_add_offset() > > * > > diff --git a/lib/time.c b/lib/time.c > > index 65db0f6cda..47f8c84327 100644 > > --- a/lib/time.c > > +++ b/lib/time.c > > @@ -152,6 +152,11 @@ uint64_t __weak get_timer_us(uint64_t base) > > return tick_to_time_us(get_ticks()) - base; > > } > > > > +unsigned long __weak get_timer_us_long(unsigned long base) > > +{ > > + return timer_get_us() - base; > > +} > > + > > unsigned long __weak notrace timer_get_us(void) > > { > > return tick_to_time(get_ticks() * 1000); > > -- > > Reviewed-by: Bin Meng <bmeng.cn at gmail.com> > > Some day we should clean up the timer APIs. We have get_timer_us(), > and now get_timer_us_long(), but we also have timer_get_us() that > returns long value! Yes, but that one returns the monotonic time, so isn't so easy for timeouts. Perhaps we should remove it since get_timer_us_long(0) does the same thing? Regards, Simon
Hi Simon, On Mon, Jul 13, 2020 at 10:25 AM Simon Glass <sjg at chromium.org> wrote: > > Hi Bin, > > On Sun, 12 Jul 2020 at 20:13, Bin Meng <bmeng.cn at gmail.com> wrote: > > > > On Fri, Jul 10, 2020 at 8:49 AM Simon Glass <sjg at chromium.org> wrote: > > > > > > The current get_timer_us() uses 64-bit arithmetic on 32-bit machines. > > > When implementing microsecond-level timeouts, 32-bits is plenty. Add a > > > new function that uses an unsigned long. On 64-bit machines this is > > > still 64-bit, but this doesn't introduce a penalty. On 32-bit machines > > > it is more efficient. > > > > > > Signed-off-by: Simon Glass <sjg at chromium.org> > > > --- > > > > > > Changes in v3: > > > - Expand the commit message and function comment > > > > > > include/time.h | 11 +++++++++++ > > > lib/time.c | 5 +++++ > > > 2 files changed, 16 insertions(+) > > > > > > diff --git a/include/time.h b/include/time.h > > > index e99f9c8012..3f00e68713 100644 > > > --- a/include/time.h > > > +++ b/include/time.h > > > @@ -17,6 +17,17 @@ unsigned long get_timer(unsigned long base); > > > unsigned long timer_get_us(void); > > > uint64_t get_timer_us(uint64_t base); > > > > > > +/** > > > + * get_timer_us_long() - Get the number of elapsed microseconds > > > + * > > > + * This uses 32-bit arithmetic on 32-bit machines, which is enough to handle > > > + * delays of over an hour. For 64-bit machines it uses a 64-bit value. > > > + * > > > + *@base: Base time to consider > > > + *@return elapsed time since @base > > > + */ > > > +unsigned long get_timer_us_long(unsigned long base); > > > + > > > /* > > > * timer_test_add_offset() > > > * > > > diff --git a/lib/time.c b/lib/time.c > > > index 65db0f6cda..47f8c84327 100644 > > > --- a/lib/time.c > > > +++ b/lib/time.c > > > @@ -152,6 +152,11 @@ uint64_t __weak get_timer_us(uint64_t base) > > > return tick_to_time_us(get_ticks()) - base; > > > } > > > > > > +unsigned long __weak get_timer_us_long(unsigned long base) > > > +{ > > > + return timer_get_us() - base; > > > +} > > > + > > > unsigned long __weak notrace timer_get_us(void) > > > { > > > return tick_to_time(get_ticks() * 1000); > > > -- > > > > Reviewed-by: Bin Meng <bmeng.cn at gmail.com> > > > > Some day we should clean up the timer APIs. We have get_timer_us(), > > and now get_timer_us_long(), but we also have timer_get_us() that > > returns long value! > > Yes, but that one returns the monotonic time, so isn't so easy for > timeouts. Perhaps we should remove it since get_timer_us_long(0) does > the same thing? Yes, I think so. Regards, Bin
diff --git a/include/time.h b/include/time.h index e99f9c8012..3f00e68713 100644 --- a/include/time.h +++ b/include/time.h @@ -17,6 +17,17 @@ unsigned long get_timer(unsigned long base); unsigned long timer_get_us(void); uint64_t get_timer_us(uint64_t base); +/** + * get_timer_us_long() - Get the number of elapsed microseconds + * + * This uses 32-bit arithmetic on 32-bit machines, which is enough to handle + * delays of over an hour. For 64-bit machines it uses a 64-bit value. + * + *@base: Base time to consider + *@return elapsed time since @base + */ +unsigned long get_timer_us_long(unsigned long base); + /* * timer_test_add_offset() * diff --git a/lib/time.c b/lib/time.c index 65db0f6cda..47f8c84327 100644 --- a/lib/time.c +++ b/lib/time.c @@ -152,6 +152,11 @@ uint64_t __weak get_timer_us(uint64_t base) return tick_to_time_us(get_ticks()) - base; } +unsigned long __weak get_timer_us_long(unsigned long base) +{ + return timer_get_us() - base; +} + unsigned long __weak notrace timer_get_us(void) { return tick_to_time(get_ticks() * 1000);
The current get_timer_us() uses 64-bit arithmetic on 32-bit machines. When implementing microsecond-level timeouts, 32-bits is plenty. Add a new function that uses an unsigned long. On 64-bit machines this is still 64-bit, but this doesn't introduce a penalty. On 32-bit machines it is more efficient. Signed-off-by: Simon Glass <sjg at chromium.org> --- Changes in v3: - Expand the commit message and function comment include/time.h | 11 +++++++++++ lib/time.c | 5 +++++ 2 files changed, 16 insertions(+)