Skip to contentSkip to conversation

timers: Clear must_forward_clk inside base lock

PATCH In review Gaurav Kohli 2 participants 4 messages Jul 26, 2018 – Jul 31, 2018

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 timers: Clear must_forward_clk inside base lock

While migrating timer to new base, there is a need to update base clk by calling forward_timer_base to avoid stale clock , but at the same time if run_timer is exectuing in new core it may set must_forward_clk to false and due to this forward base logic may fail as per below check:

if (likely(!base->must_forward_clk))

                return;

So preventing the same by putting must_forward_clk inside base lock.

kernel/time/timer.c +13 -13
index cc2d23e..675241d 100644
@@ -1657,6 +1657,19 @@ static inline void __run_timers(struct timer_base *base)
 
 	raw_spin_lock_irq(&base->lock);
 
+	/*
+	 * must_forward_clk must be cleared before running timers so that any
+	 * timer functions that call mod_timer will not try to forward the
+	 * base. idle trcking / clock forwarding logic is only used with
+	 * BASE_STD timers.
+	 *
+	 * The deferrable base does not do idle tracking at all, so we do
+	 * not forward it. This can result in very large variations in
+	 * granularity for deferrable timers, but they can be deferred for
+	 * long periods due to idle.
+	 */
+	base->must_forward_clk = false;
+
 	while (time_after_eq(jiffies, base->clk)) {
 
 		levels = collect_expired_timers(base, heads);
@@ -1676,19 +1689,6 @@ static __latent_entropy void run_timer_softirq(struct softirq_action *h)
 {
 	struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
 
-	/*
-	 * must_forward_clk must be cleared before running timers so that any
-	 * timer functions that call mod_timer will not try to forward the
-	 * base. idle trcking / clock forwarding logic is only used with
-	 * BASE_STD timers.
-	 *
-	 * The deferrable base does not do idle tracking at all, so we do
-	 * not forward it. This can result in very large variations in
-	 * granularity for deferrable timers, but they can be deferred for
-	 * long periods due to idle.
-	 */
-	base->must_forward_clk = false;
-
 	__run_timers(base);
 	if (IS_ENABLED(CONFIG_NO_HZ_COMMON))
 		__run_timers(this_cpu_ptr(&timer_bases[BASE_DEF]));
Signature · 1 line
1.9.1
#1 lore raw

Hi John, Thomas,

Can you please review below patch and update your comments:

Regards
Gaurav

Quoted 60 lines · Gaurav Kohli

While migrating timer to new base, there is a need to update base clk by calling forward_timer_base to avoid stale clock , but at the same time if run_timer is exectuing in new core it may set must_forward_clk to false and due to this forward base logic may fail as per below check:

if (likely(!base->must_forward_clk))

                 return;

So preventing the same by putting must_forward_clk inside base lock.

kernel/time/timer.c +13 -13
index cc2d23e..675241d 100644
@@ -1657,6 +1657,19 @@ static inline void __run_timers(struct timer_base *base)
 
  	raw_spin_lock_irq(&base->lock);
 
+	/*
+	 * must_forward_clk must be cleared before running timers so that any
+	 * timer functions that call mod_timer will not try to forward the
+	 * base. idle trcking / clock forwarding logic is only used with
+	 * BASE_STD timers.
+	 *
+	 * The deferrable base does not do idle tracking at all, so we do
+	 * not forward it. This can result in very large variations in
+	 * granularity for deferrable timers, but they can be deferred for
+	 * long periods due to idle.
+	 */
+	base->must_forward_clk = false;
+
  	while (time_after_eq(jiffies, base->clk)) {
 
  		levels = collect_expired_timers(base, heads);
@@ -1676,19 +1689,6 @@ static __latent_entropy void run_timer_softirq(struct softirq_action *h)
  {
  	struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
 
-	/*
-	 * must_forward_clk must be cleared before running timers so that any
-	 * timer functions that call mod_timer will not try to forward the
-	 * base. idle trcking / clock forwarding logic is only used with
-	 * BASE_STD timers.
-	 *
-	 * The deferrable base does not do idle tracking at all, so we do
-	 * not forward it. This can result in very large variations in
-	 * granularity for deferrable timers, but they can be deferred for
-	 * long periods due to idle.
-	 */
-	base->must_forward_clk = false;
-
  	__run_timers(base);
  	if (IS_ENABLED(CONFIG_NO_HZ_COMMON))
  		__run_timers(this_cpu_ptr(&timer_bases[BASE_DEF]));
Signature · 2 lines
Qualcomm India Private Limited, on behalf of Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project.
#2 lore raw

Gaurav,

Quoted 1 line · Gaurav

Can you please review below patch and update your comments:

It's on the todo list already. We're not machines and a week time before sending a reminder is really appropriate.

Thanks,

tglx

#3 lore raw
In reply to Gaurav Kohli

Gaurav,

Quoted 9 lines · Gaurav Kohli

While migrating timer to new base, there is a need to update base clk by calling forward_timer_base to avoid stale clock , but at the same time if run_timer is exectuing in new core it may set must_forward_clk to false and due to this forward base logic may fail as per below check:

if (likely(!base->must_forward_clk))

                return;

After twisting my brain for a while I can understand what you are trying to say, but please look at your own sentence once again. One sentence spawning 6 lines with a really convoluted structure and then you spend 3 lines to copy a code snippet which is really not helpful.

Please try to structure the description and use a simple table to show the race, e.g.:

base->must_forward_clock is indicating that the base clock might be stale due to a long idle sleep. The forwarding takes either place in the timer softirq or when a timer is enqueued while the base is idle. If the enqueue to an idle base happens from a remote CPU then the following race can happen:

CPU0 CPU1

run_timer_softirq() mod_timer(timer)

    base->must_forward_clk = false;	  base = lock_base(timer);
    __run_timers(base)	     		  if (base->must_forward_clk)
    					     forward(base);
      lock(base->lock);
					  queue_timer(base, timer);
      					  ^^^ Based on stale base->clk
					  
      					  unlock(base);
      forward(base);					  

The root cause is that base->must_forward_clk is cleared outside the base->lock held region, so the remote queueing CPU observes it as cleared, but the base clock is still stale. This can cause large granularity values for timers, i.e. the accuracy of the expiry time suffers.

Prevent this by clearing the flag with base->lock held, so that the forwarding takes place before the cleared flag is observable by a remote CPU.

Can you see the difference?

kernel/time/timer.c static inline void __run_timers(struct timer_base *base) · line 1663

Quoted 6 lines
kernel/time/timer.c +4 -0
 	raw_spin_lock_irq(&base->lock);
 
+	/*
+	 * must_forward_clk must be cleared before running timers so that any
+	 * timer functions that call mod_timer will not try to forward the
+	 * base. idle trcking / clock forwarding logic is only used with

I know that the 'trcking' typo was in the original comment, but it does not make anything better if you just blindly move it.

kernel/time/timer.c static inline void __run_timers(struct timer_base *base) · line 1669

Quoted 6 lines
kernel/time/timer.c +6 -0
+	 * BASE_STD timers.
+	 *
+	 * The deferrable base does not do idle tracking at all, so we do
+	 * not forward it. This can result in very large variations in
+	 * granularity for deferrable timers, but they can be deferred for
+	 * long periods due to idle.

This part of the comment is still correct, but now it's also confusing because the flag is cleared for _ALL_ bases and not only for BASE_STD. So at least you want to change that to something like this:

* The must_forward_clk flag is cleared unconditionally also for
* the deferrable base. The deferrable base is not affected by idle
* tracking and never forwarded, so clearing the flag is a NOOP. *
* The fact that the deferrable base is never forwarded can cause
* large varations in granularity for deferrable timers, but they
* can be deferred for long periods due to idle anyway.

See? If you move a comment you really have to think about whether it is still correct. If not, then you have to adjust it so it makes sense and not just move it blindly around and be done with it. Think about yourself looking at that code in a year from now when you forgot all the gory details already.

Thanks,

tglx