mirrors/linux
1
0
Fork 1
mirror of https://github.com/torvalds/linux.git synced 2025-11-04 02:30:34 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

timers, sched/clock: Avoid deadlock during read from NMI

Browse source 
Currently it is possible for an NMI (or FIQ on ARM) to come in
and read sched_clock() whilst update_sched_clock() has locked
the seqcount for writing. This results in the NMI handler
locking up when it calls raw_read_seqcount_begin().

This patch fixes the NMI safety issues by providing banked clock
data. This is a similar approach to the one used in Thomas
Gleixner's 4396e058c52e("timekeeping: Provide fast and NMI safe
access to CLOCK_MONOTONIC").

Suggested-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Daniel Thompson <daniel.thompson@linaro.org>
Signed-off-by: John Stultz <john.stultz@linaro.org>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Russell King <linux@arm.linux.org.uk>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will.deacon@arm.com>
Link: http://lkml.kernel.org/r/1427397806-20889-6-git-send-email-john.stultz@linaro.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
This commit is contained in:
Daniel Thompson 2015-03-26 12:23:26 -07:00 committed by Ingo Molnar
parent 9fee69a8c8
commit 1809bfa44e
1 changed files with 67 additions and 34 deletions
repo.diff.show_diff_stats Download patch file Download diff file Expand all files Collapse all files

101
kernel/time/sched_clock.c
View file

@ -47,19 +47,20 @@ struct clock_read_data {
* struct clock_data - all data needed for sched_clock (including * struct clock_data - all data needed for sched_clock (including
* registration of a new clock source) * registration of a new clock source)
* *
* @seq: Sequence counter for protecting updates. * @seq: Sequence counter for protecting updates. The lowest
* bit is the index for @read_data.
* @read_data: Data required to read from sched_clock. * @read_data: Data required to read from sched_clock.
* @wrap_kt: Duration for which clock can run before wrapping * @wrap_kt: Duration for which clock can run before wrapping
* @rate: Tick rate of the registered clock * @rate: Tick rate of the registered clock
* @actual_read_sched_clock: Registered clock read function * @actual_read_sched_clock: Registered clock read function
* *
* The ordering of this structure has been chosen to optimize cache * The ordering of this structure has been chosen to optimize cache
* performance. In particular seq and read_data (combined) should fit * performance. In particular seq and read_data[0] (combined) should fit
* into a single 64 byte cache line. * into a single 64 byte cache line.
*/ */
struct clock_data { struct clock_data {
seqcount_t seq; seqcount_t seq;
struct clock_read_data read_data; struct clock_read_data read_data[2];
ktime_t wrap_kt; ktime_t wrap_kt;
unsigned long rate; unsigned long rate;
u64 (*actual_read_sched_clock)(void); u64 (*actual_read_sched_clock)(void);
@ -80,10 +81,9 @@ static u64 notrace jiffy_sched_clock_read(void)
} }
static struct clock_data cd ____cacheline_aligned = { static struct clock_data cd ____cacheline_aligned = {
.read_data = { .mult = NSEC_PER_SEC / HZ, .read_data[0] = { .mult = NSEC_PER_SEC / HZ,
.read_sched_clock = jiffy_sched_clock_read, }, .read_sched_clock = jiffy_sched_clock_read, },
.actual_read_sched_clock = jiffy_sched_clock_read, .actual_read_sched_clock = jiffy_sched_clock_read,
}; };
static inline u64 notrace cyc_to_ns(u64 cyc, u32 mult, u32 shift) static inline u64 notrace cyc_to_ns(u64 cyc, u32 mult, u32 shift)
@ -95,10 +95,11 @@ unsigned long long notrace sched_clock(void)
{ {
u64 cyc, res; u64 cyc, res;
unsigned long seq; unsigned long seq;
struct clock_read_data *rd = &cd.read_data; struct clock_read_data *rd;
do { do {
seq = raw_read_seqcount_begin(&cd.seq); seq = raw_read_seqcount(&cd.seq);
rd = cd.read_data + (seq & 1);
cyc = (rd->read_sched_clock() - rd->epoch_cyc) & cyc = (rd->read_sched_clock() - rd->epoch_cyc) &
rd->sched_clock_mask; rd->sched_clock_mask;
@ -108,27 +109,51 @@ unsigned long long notrace sched_clock(void)
return res; return res;
} }
/*
* Updating the data required to read the clock.
*
* sched_clock will never observe mis-matched data even if called from
* an NMI. We do this by maintaining an odd/even copy of the data and
* steering sched_clock to one or the other using a sequence counter.
* In order to preserve the data cache profile of sched_clock as much
* as possible the system reverts back to the even copy when the update
* completes; the odd copy is used *only* during an update.
*/
static void update_clock_read_data(struct clock_read_data *rd)
{
/* update the backup (odd) copy with the new data */
cd.read_data[1] = *rd;
/* steer readers towards the odd copy */
raw_write_seqcount_latch(&cd.seq);
/* now its safe for us to update the normal (even) copy */
cd.read_data[0] = *rd;
/* switch readers back to the even copy */
raw_write_seqcount_latch(&cd.seq);
}
/* /*
* Atomically update the sched_clock epoch. * Atomically update the sched_clock epoch.
*/ */
static void update_sched_clock(void) static void update_sched_clock(void)
{ {
unsigned long flags;
u64 cyc; u64 cyc;
u64 ns; u64 ns;
struct clock_read_data *rd = &cd.read_data; struct clock_read_data rd;
rd = cd.read_data[0];
cyc = cd.actual_read_sched_clock(); cyc = cd.actual_read_sched_clock();
ns = rd->epoch_ns + ns = rd.epoch_ns +
cyc_to_ns((cyc - rd->epoch_cyc) & rd->sched_clock_mask, cyc_to_ns((cyc - rd.epoch_cyc) & rd.sched_clock_mask,
rd->mult, rd->shift); rd.mult, rd.shift);
raw_local_irq_save(flags); rd.epoch_ns = ns;
raw_write_seqcount_begin(&cd.seq); rd.epoch_cyc = cyc;
rd->epoch_ns = ns;
rd->epoch_cyc = cyc; update_clock_read_data(&rd);
raw_write_seqcount_end(&cd.seq);
raw_local_irq_restore(flags);
} }
static enum hrtimer_restart sched_clock_poll(struct hrtimer *hrt) static enum hrtimer_restart sched_clock_poll(struct hrtimer *hrt)
@ -145,7 +170,7 @@ void __init sched_clock_register(u64 (*read)(void), int bits,
u32 new_mult, new_shift; u32 new_mult, new_shift;
unsigned long r; unsigned long r;
char r_unit; char r_unit;
struct clock_read_data *rd = &cd.read_data; struct clock_read_data rd;
if (cd.rate > rate) if (cd.rate > rate)
return; return;
@ -162,22 +187,23 @@ void __init sched_clock_register(u64 (*read)(void), int bits,
wrap = clocks_calc_max_nsecs(new_mult, new_shift, 0, new_mask, NULL); wrap = clocks_calc_max_nsecs(new_mult, new_shift, 0, new_mask, NULL);
cd.wrap_kt = ns_to_ktime(wrap); cd.wrap_kt = ns_to_ktime(wrap);
rd = cd.read_data[0];
/* update epoch for new counter and update epoch_ns from old counter*/ /* update epoch for new counter and update epoch_ns from old counter*/
new_epoch = read(); new_epoch = read();
cyc = cd.actual_read_sched_clock(); cyc = cd.actual_read_sched_clock();
ns = rd->epoch_ns + ns = rd.epoch_ns +
cyc_to_ns((cyc - rd->epoch_cyc) & rd->sched_clock_mask, cyc_to_ns((cyc - rd.epoch_cyc) & rd.sched_clock_mask,
rd->mult, rd->shift); rd.mult, rd.shift);
cd.actual_read_sched_clock = read; cd.actual_read_sched_clock = read;
raw_write_seqcount_begin(&cd.seq); rd.read_sched_clock = read;
rd->read_sched_clock = read; rd.sched_clock_mask = new_mask;
rd->sched_clock_mask = new_mask; rd.mult = new_mult;
rd->mult = new_mult; rd.shift = new_shift;
rd->shift = new_shift; rd.epoch_cyc = new_epoch;
rd->epoch_cyc = new_epoch; rd.epoch_ns = ns;
rd->epoch_ns = ns; update_clock_read_data(&rd);
raw_write_seqcount_end(&cd.seq);
r = rate; r = rate;
if (r >= 4000000) { if (r >= 4000000) {
@ -227,15 +253,22 @@ void __init sched_clock_postinit(void)
* *
* This function makes it appear to sched_clock() as if the clock * This function makes it appear to sched_clock() as if the clock
* stopped counting at its last update. * stopped counting at its last update.
*
* This function must only be called from the critical
* section in sched_clock(). It relies on the read_seqcount_retry()
* at the end of the critical section to be sure we observe the
* correct copy of epoch_cyc.
*/ */
static u64 notrace suspended_sched_clock_read(void) static u64 notrace suspended_sched_clock_read(void)
{ {
return cd.read_data.epoch_cyc; unsigned long seq = raw_read_seqcount(&cd.seq);
return cd.read_data[seq & 1].epoch_cyc;
} }
static int sched_clock_suspend(void) static int sched_clock_suspend(void)
{ {
struct clock_read_data *rd = &cd.read_data; struct clock_read_data *rd = &cd.read_data[0];
update_sched_clock(); update_sched_clock();
hrtimer_cancel(&sched_clock_timer); hrtimer_cancel(&sched_clock_timer);
@ -245,7 +278,7 @@ static int sched_clock_suspend(void)
static void sched_clock_resume(void) static void sched_clock_resume(void)
{ {
struct clock_read_data *rd = &cd.read_data; struct clock_read_data *rd = &cd.read_data[0];
rd->epoch_cyc = cd.actual_read_sched_clock(); rd->epoch_cyc = cd.actual_read_sched_clock();
hrtimer_start(&sched_clock_timer, cd.wrap_kt, HRTIMER_MODE_REL); hrtimer_start(&sched_clock_timer, cd.wrap_kt, HRTIMER_MODE_REL);