Re: [PATCH] time/sched_clock: Use raw_read_seqcount_latch()
Hi Peter, Ahemd, On Wed, Jul 15, 2020 at 05:58:50PM +0200, Peter Zijlstra wrote: [...] > > > diff --git a/kernel/time/sched_clock.c b/kernel/time/sched_clock.c > > > index fa3f800d7d76..ea007928d681 100644 > > > --- a/kernel/time/sched_clock.c > > > +++ b/kernel/time/sched_clock.c > > > @@ -100,7 +100,7 @@ unsigned long long notrace sched_clock(void) > > > struct clock_read_data *rd; > > > > > > do { > > > - seq = raw_read_seqcount(); > > > + seq = raw_read_seqcount_latch(); > > > > Understand this is doing the same thing with __ktime_get_fast_ns() and > > I saw Peter acked to make change for this. > > > > Just want to confirm, since this patch introduces conflict with the > > patch set "arm64: perf: Proper cap_user_time* support" [1], I should > > rebase the patch set on top of this patch, right? > > Or rebase this patch on top of yours and include it, either way. Have rebased this patch and included it in the patch set v3 for "arm64: perf: Proper cap_user_time* support" [1]. Thanks! Leo [1] https://lore.kernel.org/linux-arm-kernel/20200716051130.4359-3-leo@linaro.org/T/#u
Re: [PATCH] time/sched_clock: Use raw_read_seqcount_latch()
On Wed, Jul 15, 2020 at 11:29:26PM +0800, Leo Yan wrote: > Hi Peter, Ahmed, > > On Wed, Jul 15, 2020 at 01:59:01PM +0200, Ahmed S. Darwish wrote: > > sched_clock uses seqcount_t latching to switch between two storage > > places protected by the sequence counter. This allows it to have > > interruptible, NMI-safe, seqcount_t write side critical sections. > > > > Since 7fc26327b756 ("seqlock: Introduce raw_read_seqcount_latch()"), > > raw_read_seqcount_latch() became the standardized way for seqcount_t > > latch read paths. Due to the dependent load, it also has one read > > memory barrier less than the currently used raw_read_seqcount() API. > > > > Use raw_read_seqcount_latch() for the seqcount_t latch read path. > > > > Link: > > https://lkml.kernel.org/r/20200625085745.gd117...@hirez.programming.kicks-ass.net > > Link: > > https://lkml.kernel.org/r/20200715092345.ga231...@debian-buster-darwi.lab.linutronix.de > > References: 1809bfa44e10 ("timers, sched/clock: Avoid deadlock during read > > from NMI") > > Signed-off-by: Ahmed S. Darwish > > --- > > kernel/time/sched_clock.c | 2 +- > > 1 file changed, 1 insertion(+), 1 deletion(-) > > > > diff --git a/kernel/time/sched_clock.c b/kernel/time/sched_clock.c > > index fa3f800d7d76..ea007928d681 100644 > > --- a/kernel/time/sched_clock.c > > +++ b/kernel/time/sched_clock.c > > @@ -100,7 +100,7 @@ unsigned long long notrace sched_clock(void) > > struct clock_read_data *rd; > > > > do { > > - seq = raw_read_seqcount(); > > + seq = raw_read_seqcount_latch(); > > Understand this is doing the same thing with __ktime_get_fast_ns() and > I saw Peter acked to make change for this. > > Just want to confirm, since this patch introduces conflict with the > patch set "arm64: perf: Proper cap_user_time* support" [1], I should > rebase the patch set on top of this patch, right? Or rebase this patch on top of yours and include it, either way.
Re: [PATCH] time/sched_clock: Use raw_read_seqcount_latch()
Hi Peter, Ahmed, On Wed, Jul 15, 2020 at 01:59:01PM +0200, Ahmed S. Darwish wrote: > sched_clock uses seqcount_t latching to switch between two storage > places protected by the sequence counter. This allows it to have > interruptible, NMI-safe, seqcount_t write side critical sections. > > Since 7fc26327b756 ("seqlock: Introduce raw_read_seqcount_latch()"), > raw_read_seqcount_latch() became the standardized way for seqcount_t > latch read paths. Due to the dependent load, it also has one read > memory barrier less than the currently used raw_read_seqcount() API. > > Use raw_read_seqcount_latch() for the seqcount_t latch read path. > > Link: > https://lkml.kernel.org/r/20200625085745.gd117...@hirez.programming.kicks-ass.net > Link: > https://lkml.kernel.org/r/20200715092345.ga231...@debian-buster-darwi.lab.linutronix.de > References: 1809bfa44e10 ("timers, sched/clock: Avoid deadlock during read > from NMI") > Signed-off-by: Ahmed S. Darwish > --- > kernel/time/sched_clock.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/kernel/time/sched_clock.c b/kernel/time/sched_clock.c > index fa3f800d7d76..ea007928d681 100644 > --- a/kernel/time/sched_clock.c > +++ b/kernel/time/sched_clock.c > @@ -100,7 +100,7 @@ unsigned long long notrace sched_clock(void) > struct clock_read_data *rd; > > do { > - seq = raw_read_seqcount(); > + seq = raw_read_seqcount_latch(); Understand this is doing the same thing with __ktime_get_fast_ns() and I saw Peter acked to make change for this. Just want to confirm, since this patch introduces conflict with the patch set "arm64: perf: Proper cap_user_time* support" [1], I should rebase the patch set on top of this patch, right? Thanks, Leo [1] https://patchwork.kernel.org/cover/11664031/ > rd = cd.read_data + (seq & 1); > > cyc = (rd->read_sched_clock() - rd->epoch_cyc) & > -- > 2.20.1
[PATCH] time/sched_clock: Use raw_read_seqcount_latch()
sched_clock uses seqcount_t latching to switch between two storage places protected by the sequence counter. This allows it to have interruptible, NMI-safe, seqcount_t write side critical sections. Since 7fc26327b756 ("seqlock: Introduce raw_read_seqcount_latch()"), raw_read_seqcount_latch() became the standardized way for seqcount_t latch read paths. Due to the dependent load, it also has one read memory barrier less than the currently used raw_read_seqcount() API. Use raw_read_seqcount_latch() for the seqcount_t latch read path. Link: https://lkml.kernel.org/r/20200625085745.gd117...@hirez.programming.kicks-ass.net Link: https://lkml.kernel.org/r/20200715092345.ga231...@debian-buster-darwi.lab.linutronix.de References: 1809bfa44e10 ("timers, sched/clock: Avoid deadlock during read from NMI") Signed-off-by: Ahmed S. Darwish --- kernel/time/sched_clock.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/kernel/time/sched_clock.c b/kernel/time/sched_clock.c index fa3f800d7d76..ea007928d681 100644 --- a/kernel/time/sched_clock.c +++ b/kernel/time/sched_clock.c @@ -100,7 +100,7 @@ unsigned long long notrace sched_clock(void) struct clock_read_data *rd; do { - seq = raw_read_seqcount(); + seq = raw_read_seqcount_latch(); rd = cd.read_data + (seq & 1); cyc = (rd->read_sched_clock() - rd->epoch_cyc) & -- 2.20.1