SRCU callback parallelization for 4.12

---
 include/linux/rcutiny.h |  5 ++++
 kernel/rcu/tree.c       | 73 +++++++++++++++++++++++++++++++++++++++----------
 kernel/rcu/tree.h       |  1 +
 3 files changed, 65 insertions(+), 14 deletions(-)

include/linux/rcutiny.h +5 -0

index b452953e21c8..6c9d941e3962 100644
@@ -33,6 +33,11 @@ static inline int rcu_dynticks_snap(struct rcu_dynticks *rdtp)
 	return 0;
 }
 
+static inline bool rcu_eqs_special_set(int cpu)
+{
+	return false;  /* Never flag non-existent other CPUs! */
+}
+
 static inline unsigned long get_state_synchronize_rcu(void)
 {
 	return 0;

kernel/rcu/tree.c +59 ~~-14~~

138 lines not on this page · raw message

kernel/rcu/tree.h +1 -0

index ec62a05bfdb3..7468b4de7e0c 100644
@@ -596,6 +596,7 @@ extern struct rcu_state rcu_preempt_state;
 #endif /* #ifdef CONFIG_PREEMPT_RCU */
 
 int rcu_dynticks_snap(struct rcu_dynticks *rdtp);
+bool rcu_eqs_special_set(int cpu);
 
 #ifdef CONFIG_RCU_BOOST
 DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_status);

Signature · 1 line

2.5.2

Peter Zijlstra

maintainer

2017-04-13 09:21 UTC #2 lore raw

Re: [PATCH tip/core/rcu 01/40] rcu: Maintain special bits at bottom of ->dynticks counter

Quoted 17 lines · Paul E. McKenney

Currently, IPIs are used to force other CPUs to invalidate their TLBs in response to a kernel virtual-memory mapping change. This works, but degrades both battery lifetime (for idle CPUs) and real-time response (for nohz_full CPUs), and in addition results in unnecessary IPIs due to the fact that CPUs executing in usermode are unaffected by stale kernel mappings. It would be better to cause a CPU executing in usermode to wait until it is entering kernel mode to do the flush, first to avoid interrupting usemode tasks and second to handle multiple flush requests with a single flush in the case of a long-running user task.
This commit therefore reserves a bit at the bottom of the ->dynticks counter, which is checked upon exit from extended quiescent states. If it is set, it is cleared and then a new rcu_eqs_special_exit() macro is invoked, which, if not supplied, is an empty single-pass do-while loop. If this bottom bit is set on -entry- to an extended quiescent state, then a WARN_ON_ONCE() triggers.

Quoted 8 lines

This bottom bit may be set using a new rcu_eqs_special_set() function, which returns true if the bit was set, or false if the CPU turned out to not be in an extended quiescent state. Please note that this function refuses to set the bit for a non-nohz_full CPU when that CPU is executing in usermode because usermode execution is tracked by RCU as a dyntick-idle extended quiescent state only for nohz_full CPUs.
Reported-by Andy Lutomirski <[email protected]>

Isn't that more a: Requested-by ?

Paul E. McKenney

author

2017-04-13 16:15 UTC #3 lore raw

Re: [PATCH tip/core/rcu 01/40] rcu: Maintain special bits at bottom of ->dynticks counter

Quoted 25 lines · Peter Zijlstra

Quoted 24 lines · Paul E. McKenney
Currently, IPIs are used to force other CPUs to invalidate their TLBs in response to a kernel virtual-memory mapping change. This works, but degrades both battery lifetime (for idle CPUs) and real-time response (for nohz_full CPUs), and in addition results in unnecessary IPIs due to the fact that CPUs executing in usermode are unaffected by stale kernel mappings. It would be better to cause a CPU executing in usermode to wait until it is entering kernel mode to do the flush, first to avoid interrupting usemode tasks and second to handle multiple flush requests with a single flush in the case of a long-running user task.
This commit therefore reserves a bit at the bottom of the ->dynticks counter, which is checked upon exit from extended quiescent states. If it is set, it is cleared and then a new rcu_eqs_special_exit() macro is invoked, which, if not supplied, is an empty single-pass do-while loop. If this bottom bit is set on -entry- to an extended quiescent state, then a WARN_ON_ONCE() triggers.
This bottom bit may be set using a new rcu_eqs_special_set() function, which returns true if the bit was set, or false if the CPU turned out to not be in an extended quiescent state. Please note that this function refuses to set the bit for a non-nohz_full CPU when that CPU is executing in usermode because usermode execution is tracked by RCU as a dyntick-idle extended quiescent state only for nohz_full CPUs.

Quoted 3 lines

Quoted 1 line
Reported-by Andy Lutomirski <[email protected]>
Isn't that more a: Requested-by ?

I am not too worried about the distinction. Request a feature, report the lack of a needed feature, or report a bug, but either way I had to write the code. ;-)

Thanx, Paul

Paul E. McKenney

author

2017-04-12 17:40 UTC #4 lore raw

PATCH 2/40 rcu: Make arch select smp_mb__after_unlock_lock() strength

The definition of smp_mb__after_unlock_lock() is currently smp_mb() for CONFIG_PPC and a no-op otherwise. It would be better to instead provide an architecture-selectable Kconfig option, and select the strength of smp_mb__after_unlock_lock() based on that option. This commit therefore creates CONFIG_ARCH_WEAK_RELACQ, has PPC select it, and bases the definition of smp_mb__after_unlock_lock() on this new CONFIG_ARCH_WEAK_RELACQ Kconfig option.

Reported-by Ingo Molnar <[email protected]>

Cc Peter Zijlstra <[email protected]>

Cc Will Deacon <[email protected]>

Cc Boqun Feng <[email protected]>

Cc Benjamin Herrenschmidt <[email protected]>

Cc Paul Mackerras <[email protected]>

Acked-by Michael Ellerman <[email protected]>

---
 arch/Kconfig             | 3 +++
 arch/powerpc/Kconfig     | 1 +
 include/linux/rcupdate.h | 6 +++---
 3 files changed, 7 insertions(+), 3 deletions(-)

arch/Kconfig +3 -0

index cd211a14a88f..adefaf344239 100644
@@ -320,6 +320,9 @@ config HAVE_CMPXCHG_LOCAL
 config HAVE_CMPXCHG_DOUBLE
 	bool
 
+config ARCH_WEAK_RELEASE_ACQUIRE
+	bool
+
 config ARCH_WANT_IPC_PARSE_VERSION
 	bool

arch/powerpc/Kconfig +1 -0

index 97a8bc8a095c..7a5c9b764cd2 100644
@@ -99,6 +99,7 @@ config PPC
 	select ARCH_USE_BUILTIN_BSWAP
 	select ARCH_USE_CMPXCHG_LOCKREF		if PPC64
 	select ARCH_WANT_IPC_PARSE_VERSION
+	select ARCH_WEAK_RELEASE_ACQUIRE
 	select BINFMT_ELF
 	select BUILDTIME_EXTABLE_SORT
 	select CLONE_BACKWARDS

include/linux/rcupdate.h +3 -3

index de88b33c0974..e6146d0074f8 100644
@@ -1127,11 +1127,11 @@ do { \
  * if the UNLOCK and LOCK are executed by the same CPU or if the
  * UNLOCK and LOCK operate on the same lock variable.
  */
-#ifdef CONFIG_PPC
+#ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE
 #define smp_mb__after_unlock_lock()	smp_mb()  /* Full ordering for lock. */
-#else /* #ifdef CONFIG_PPC */
+#else /* #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */
 #define smp_mb__after_unlock_lock()	do { } while (0)
-#endif /* #else #ifdef CONFIG_PPC */
+#endif /* #else #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */
 
 
 #endif /* __LINUX_RCUPDATE_H */

Signature · 1 line

2.5.2

Peter Zijlstra

maintainer

2017-04-13 09:22 UTC #5 lore raw

Re: [PATCH tip/core/rcu 02/40] rcu: Make arch select smp_mb__after_unlock_lock() strength

Quoted 1 line · Paul E. McKenney

CONFIG_ARCH_WEAK_RELACQ Kconfig option.

Naming in the changelog and patch don't match.

Quoted 3 lines

arch/Kconfig +3 -0

+config ARCH_WEAK_RELEASE_ACQUIRE
+	bool
+

Paul E. McKenney

author

2017-04-13 16:17 UTC #6 lore raw

Re: [PATCH tip/core/rcu 02/40] rcu: Make arch select smp_mb__after_unlock_lock() strength

Quoted 5 lines · Peter Zijlstra

Quoted 1 line · Paul E. McKenney
CONFIG_ARCH_WEAK_RELACQ Kconfig option.
Naming in the changelog and patch don't match.

Good eyes! Fixed.

Thanx, Paul

Quoted 5 lines

Quoted 3 lines
arch/Kconfig +3 -0
+config ARCH_WEAK_RELEASE_ACQUIRE
+	bool
+

Peter Zijlstra

maintainer

2017-04-13 09:24 UTC #7 lore raw

Re: [PATCH tip/core/rcu 02/40] rcu: Make arch select smp_mb__after_unlock_lock() strength

Quoted 4 lines · Paul E. McKenney

The definition of smp_mb__after_unlock_lock() is currently smp_mb() for CONFIG_PPC and a no-op otherwise. It would be better to instead provide an architecture-selectable Kconfig option, and select the strength of smp_mb__after_unlock_lock() based on that option.

Why is this better? Do we want to have more of this? I thought we still wanted to convince PPC to go RCsc and eradicate all this RCpc 'fun'. But instead now you're making it look like its OK to grow more of this pain.

Paul E. McKenney

author

2017-04-13 16:27 UTC #8 lore raw

Re: [PATCH tip/core/rcu 02/40] rcu: Make arch select smp_mb__after_unlock_lock() strength

Quoted 9 lines · Peter Zijlstra

Quoted 4 lines · Paul E. McKenney
The definition of smp_mb__after_unlock_lock() is currently smp_mb() for CONFIG_PPC and a no-op otherwise. It would be better to instead provide an architecture-selectable Kconfig option, and select the strength of smp_mb__after_unlock_lock() based on that option.
Why is this better? Do we want to have more of this? I thought we still wanted to convince PPC to go RCsc and eradicate all this RCpc 'fun'. But instead now you're making it look like its OK to grow more of this pain.

ARCH_WEAK_RELEASE_ACQUIRE actually works both ways.

To see this, imagine some strange alternate universe in which the Power hardware guys actually did decide to switch PPC to doing RCsc as you suggest. There would still be a lot of Power hardware out there that still does RCpc. Therefore, powerpc builds that needed to run on old Power hardware would select ARCH_WEAK_RELEASE_ACQUIRE, while kernels built to run only on the shiny new (but mythical) alternate-universe Power hardware would avoid selecting this Kconfig option.

But the real reason I queued this patch is that Ingo asked me for it:
https://lkml.org/lkml/2017/1/14/88

Thanx, Paul

Peter Zijlstra

maintainer

2017-04-13 16:38 UTC #9 lore raw

Re: [PATCH tip/core/rcu 02/40] rcu: Make arch select smp_mb__after_unlock_lock() strength

Quoted 9 lines · Paul E. McKenney

ARCH_WEAK_RELEASE_ACQUIRE actually works both ways.
To see this, imagine some strange alternate universe in which the Power hardware guys actually did decide to switch PPC to doing RCsc as you suggest. There would still be a lot of Power hardware out there that still does RCpc. Therefore, powerpc builds that needed to run on old Power hardware would select ARCH_WEAK_RELEASE_ACQUIRE, while kernels built to run only on the shiny new (but mythical) alternate-universe Power hardware would avoid selecting this Kconfig option.

Ah, but Power software guys could do it today by replacing an LWSYNC with a SYNC in say arch_spin_unlock().

And yes, I know this isn't a popular suggestion, but it would do the trick.

Its just that since there's one (PPC) we can sort of pressure them with the pain of being the only ones to hit all the bugs. But the moment more appear (and I'm afraid it'll be MIPS, with the excuse that PPC already does this) it will be ever so much harder to get rid of it.

Then again, maybe I should just give up and accept the Linux kernel has RCpc locks..

Paul E. McKenney

author

2017-04-13 17:04 UTC #10 lore raw

Re: [PATCH tip/core/rcu 02/40] rcu: Make arch select smp_mb__after_unlock_lock() strength

Quoted 12 lines · Peter Zijlstra

Quoted 9 lines · Paul E. McKenney
ARCH_WEAK_RELEASE_ACQUIRE actually works both ways.
To see this, imagine some strange alternate universe in which the Power hardware guys actually did decide to switch PPC to doing RCsc as you suggest. There would still be a lot of Power hardware out there that still does RCpc. Therefore, powerpc builds that needed to run on old Power hardware would select ARCH_WEAK_RELEASE_ACQUIRE, while kernels built to run only on the shiny new (but mythical) alternate-universe Power hardware would avoid selecting this Kconfig option.

Quoted 5 lines

Ah, but Power software guys could do it today by replacing an LWSYNC with a SYNC in say arch_spin_unlock().
And yes, I know this isn't a popular suggestion, but it would do the trick.

Indeed, there is a fine line between motivating people to move to new hardware on the one hand and terminally annoying existing users on the other. ;-)

Quoted 7 lines

Its just that since there's one (PPC) we can sort of pressure them with the pain of being the only ones to hit all the bugs. But the moment more appear (and I'm afraid it'll be MIPS, with the excuse that PPC already does this) it will be ever so much harder to get rid of it.
Then again, maybe I should just give up and accept the Linux kernel has RCpc locks..

As usual, I must defer to the powerpc maintainers on this one.

Thanx, Paul

Michael Ellerman

reviewer

2017-04-19 13:38 UTC #11 lore raw

Re: [PATCH tip/core/rcu 02/40] rcu: Make arch select smp_mb__after_unlock_lock() strength

Quoted 28 lines · Paul E. McKenney

Quoted 12 lines · Peter Zijlstra
Quoted 9 lines · Paul E. McKenney
ARCH_WEAK_RELEASE_ACQUIRE actually works both ways.
To see this, imagine some strange alternate universe in which the Power hardware guys actually did decide to switch PPC to doing RCsc as you suggest. There would still be a lot of Power hardware out there that still does RCpc. Therefore, powerpc builds that needed to run on old Power hardware would select ARCH_WEAK_RELEASE_ACQUIRE, while kernels built to run only on the shiny new (but mythical) alternate-universe Power hardware would avoid selecting this Kconfig option.
Quoted 5 lines
Ah, but Power software guys could do it today by replacing an LWSYNC with a SYNC in say arch_spin_unlock().
And yes, I know this isn't a popular suggestion, but it would do the trick.
Indeed, there is a fine line between motivating people to move to new hardware on the one hand and terminally annoying existing users on the other. ;-)
Quoted 5 lines
Its just that since there's one (PPC) we can sort of pressure them with the pain of being the only ones to hit all the bugs. But the moment more appear (and I'm afraid it'll be MIPS, with the excuse that PPC already does this) it will be ever so much harder to get rid of it.

Quoted 4 lines

Quoted 2 lines
Then again, maybe I should just give up and accept the Linux kernel has RCpc locks..
As usual, I must defer to the powerpc maintainers on this one.

I reworked my locking tests a bit, to run longer, disable ASLR and a few other things, and ran them again. They just bang repeatedly on an uncontended lock, so nothing fancy at all.

Switching the release barrier to sync (from lwsync) slows it down by about 18%.

So I think that pretty much rules it out, at least on current CPUs.

I'll try and get some more time to make sure I didn't do something stupid in the test, and maybe do a version that includes some contention.

cheers

Paul E. McKenney

author

2017-04-19 15:12 UTC #12 lore raw

Re: [PATCH tip/core/rcu 02/40] rcu: Make arch select smp_mb__after_unlock_lock() strength

Quoted 33 lines · Michael Ellerman

Quoted 31 lines · Paul E. McKenney
Quoted 12 lines · Peter Zijlstra
Quoted 9 lines · Paul E. McKenney
ARCH_WEAK_RELEASE_ACQUIRE actually works both ways.
To see this, imagine some strange alternate universe in which the Power hardware guys actually did decide to switch PPC to doing RCsc as you suggest. There would still be a lot of Power hardware out there that still does RCpc. Therefore, powerpc builds that needed to run on old Power hardware would select ARCH_WEAK_RELEASE_ACQUIRE, while kernels built to run only on the shiny new (but mythical) alternate-universe Power hardware would avoid selecting this Kconfig option.
Quoted 5 lines
Ah, but Power software guys could do it today by replacing an LWSYNC with a SYNC in say arch_spin_unlock().
And yes, I know this isn't a popular suggestion, but it would do the trick.
Indeed, there is a fine line between motivating people to move to new hardware on the one hand and terminally annoying existing users on the other. ;-)
Quoted 7 lines
Its just that since there's one (PPC) we can sort of pressure them with the pain of being the only ones to hit all the bugs. But the moment more appear (and I'm afraid it'll be MIPS, with the excuse that PPC already does this) it will be ever so much harder to get rid of it.
Then again, maybe I should just give up and accept the Linux kernel has RCpc locks..

Quoted 8 lines

Quoted 1 line
As usual, I must defer to the powerpc maintainers on this one.
I reworked my locking tests a bit, to run longer, disable ASLR and a few other things, and ran them again. They just bang repeatedly on an uncontended lock, so nothing fancy at all.
Switching the release barrier to sync (from lwsync) slows it down by about 18%.

Ouch!!!

Quoted 5 lines

So I think that pretty much rules it out, at least on current CPUs.
I'll try and get some more time to make sure I didn't do something stupid in the test, and maybe do a version that includes some contention.

Looking forward to seeing what you come up with...

Thanx, Paul

Paul E. McKenney

author

2017-04-12 17:49 UTC #13 lore raw

PATCH 3/40 srcu: Consolidate batch checking into rcu_all_batches_empty()

The srcu_reschedule() function invokes rcu_batch_empty() on each of the four rcu_batch structures in the srcu_struct in question twice. Given that this check will also be needed in cleanup_srcu_struct(), this commit consolidates these four checks into a new rcu_all_batches_empty() function.

---
 kernel/rcu/srcu.c | 21 +++++++++++++--------
 1 file changed, 13 insertions(+), 8 deletions(-)

kernel/rcu/srcu.c +13 -8

index ef3bcfb15b39..ba41a5d04b49 100644
@@ -65,6 +65,17 @@ static inline bool rcu_batch_empty(struct rcu_batch *b)
 }
 
 /*
+ * Are all batches empty for the specified srcu_struct?
+ */
+static inline bool rcu_all_batches_empty(struct srcu_struct *sp)
+{
+	return rcu_batch_empty(&sp->batch_done) &&
+	       rcu_batch_empty(&sp->batch_check1) &&
+	       rcu_batch_empty(&sp->batch_check0) &&
+	       rcu_batch_empty(&sp->batch_queue);
+}
+
+/*
  * Remove the callback at the head of the specified rcu_batch structure
  * and return a pointer to it, or return NULL if the structure is empty.
  */

@@ -619,15 +630,9 @@ static void srcu_reschedule(struct srcu_struct *sp)
 {
 	bool pending = true;
 
-	if (rcu_batch_empty(&sp->batch_done) &&
-	    rcu_batch_empty(&sp->batch_check1) &&
-	    rcu_batch_empty(&sp->batch_check0) &&
-	    rcu_batch_empty(&sp->batch_queue)) {
+	if (rcu_all_batches_empty(sp)) {
 		spin_lock_irq(&sp->queue_lock);
-		if (rcu_batch_empty(&sp->batch_done) &&
-		    rcu_batch_empty(&sp->batch_check1) &&
-		    rcu_batch_empty(&sp->batch_check0) &&
-		    rcu_batch_empty(&sp->batch_queue)) {
+		if (rcu_all_batches_empty(sp)) {
 			sp->running = false;
 			pending = false;
 		}

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-12 17:40 UTC #14 lore raw

PATCH 4/40 srcu: Check for tardy grace-period activity in cleanup_srcu_struct()

Users of SRCU are obliged to complete all grace-period activity before invoking cleanup_srcu_struct(). This means that all calls to either synchronize_srcu() or synchronize_srcu_expedited() must have returned, and all calls to call_srcu() must have returned, and the last call to call_srcu() must have been followed by a call to srcu_barrier(). Furthermore, the caller must have done something to prevent any further calls to synchronize_srcu(), synchronize_srcu_expedited(), and call_srcu().

Therefore, if there has ever been an invocation of call_srcu() on the srcu_struct in question, the sequence of events must be as follows:

1.  Prevent any further calls to call_srcu().
2.  Wait for any pre-existing call_srcu() invocations to return.
3.  Invoke srcu_barrier().
4.  It is now safe to invoke cleanup_srcu_struct().

On the other hand, if there has ever been a call to synchronize_srcu() or synchronize_srcu_expedited(), the sequence of events must be as follows:

1. Prevent any further calls to synchronize_srcu() or synchronize_srcu_expedited().
2. Wait for any pre-existing synchronize_srcu() or synchronize_srcu_expedited() invocations to return.
3. It is now safe to invoke cleanup_srcu_struct().

If there have been calls to all both types of functions (call_srcu() and either of synchronize_srcu() and synchronize_srcu_expedited()), then the caller must do the first three steps of the call_srcu() procedure above and the first two steps of the synchronize_s*() procedure above, and only then invoke cleanup_srcu_struct().

Reported-by Paolo Bonzini <[email protected]>

---
 kernel/rcu/srcu.c | 5 +++++
 1 file changed, 5 insertions(+)

kernel/rcu/srcu.c +5 -0

index ba41a5d04b49..6beeba7b0b67 100644
@@ -261,6 +261,11 @@ void cleanup_srcu_struct(struct srcu_struct *sp)
 {
 	if (WARN_ON(srcu_readers_active(sp)))
 		return; /* Leakage unless caller handles error. */
+	if (WARN_ON(!rcu_all_batches_empty(sp)))
+		return; /* Leakage unless caller handles error. */
+	flush_delayed_work(&sp->work);
+	if (WARN_ON(sp->running))
+		return; /* Caller forgot to stop doing call_srcu()? */
 	free_percpu(sp->per_cpu_ref);
 	sp->per_cpu_ref = NULL;
 }

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-12 17:50 UTC #15 lore raw

PATCH 5/40 rcu: Semicolon inside RCU_TRACE() for rcu.h

The current use of "RCU_TRACE(statement);" can cause odd bugs, especially where "statement" is a local-variable declaration, as it can leave a misplaced ";" in the source code. This commit therefore converts these to "RCU_TRACE(statement;)", which avoids the misplaced ";".

---
 kernel/rcu/rcu.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

kernel/rcu/rcu.h +2 -2

index 0d6ff3e471be..8700a81daf56 100644
@@ -109,12 +109,12 @@ static inline bool __rcu_reclaim(const char *rn, struct rcu_head *head)
 
 	rcu_lock_acquire(&rcu_callback_map);
 	if (__is_kfree_rcu_offset(offset)) {
-		RCU_TRACE(trace_rcu_invoke_kfree_callback(rn, head, offset));
+		RCU_TRACE(trace_rcu_invoke_kfree_callback(rn, head, offset);)
 		kfree((void *)head - offset);
 		rcu_lock_release(&rcu_callback_map);
 		return true;
 	} else {
-		RCU_TRACE(trace_rcu_invoke_callback(rn, head));
+		RCU_TRACE(trace_rcu_invoke_callback(rn, head);)
 		head->func(head);
 		rcu_lock_release(&rcu_callback_map);
 		return false;

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-12 17:40 UTC #16 lore raw

PATCH 6/40 rcu: Semicolon inside RCU_TRACE() for Tiny RCU

---
 kernel/rcu/tiny.c        | 20 ++++++++++----------
 kernel/rcu/tiny_plugin.h |  4 ++--
 2 files changed, 12 insertions(+), 12 deletions(-)

kernel/rcu/tiny.c +10 ~~-10~~

index 6ad330dbbae2..e5385731e391 100644
@@ -79,7 +79,7 @@ EXPORT_SYMBOL(__rcu_is_watching);
  */
 static int rcu_qsctr_help(struct rcu_ctrlblk *rcp)
 {
-	RCU_TRACE(reset_cpu_stall_ticks(rcp));
+	RCU_TRACE(reset_cpu_stall_ticks(rcp);)
 	if (rcp->donetail != rcp->curtail) {
 		rcp->donetail = rcp->curtail;
 		return 1;

@@ -125,7 +125,7 @@ void rcu_bh_qs(void)
  */
 void rcu_check_callbacks(int user)
 {
-	RCU_TRACE(check_cpu_stalls());
+	RCU_TRACE(check_cpu_stalls();)
 	if (user)
 		rcu_sched_qs();
 	else if (!in_softirq())

@@ -143,7 +143,7 @@ static void __rcu_process_callbacks(struct rcu_ctrlblk *rcp)
 	const char *rn = NULL;
 	struct rcu_head *next, *list;
 	unsigned long flags;
-	RCU_TRACE(int cb_count = 0);
+	RCU_TRACE(int cb_count = 0;)
 
 	/* Move the ready-to-invoke callbacks to a local list. */
 	local_irq_save(flags);

@@ -152,7 +152,7 @@ static void __rcu_process_callbacks(struct rcu_ctrlblk *rcp)
 		local_irq_restore(flags);
 		return;
 	}
-	RCU_TRACE(trace_rcu_batch_start(rcp->name, 0, rcp->qlen, -1));
+	RCU_TRACE(trace_rcu_batch_start(rcp->name, 0, rcp->qlen, -1);)
 	list = rcp->rcucblist;
 	rcp->rcucblist = *rcp->donetail;
 	*rcp->donetail = NULL;

@@ -162,7 +162,7 @@ static void __rcu_process_callbacks(struct rcu_ctrlblk *rcp)
 	local_irq_restore(flags);
 
 	/* Invoke the callbacks on the local list. */
-	RCU_TRACE(rn = rcp->name);
+	RCU_TRACE(rn = rcp->name;)
 	while (list) {
 		next = list->next;
 		prefetch(next);

@@ -171,9 +171,9 @@ static void __rcu_process_callbacks(struct rcu_ctrlblk *rcp)
 		__rcu_reclaim(rn, list);
 		local_bh_enable();
 		list = next;
-		RCU_TRACE(cb_count++);
+		RCU_TRACE(cb_count++;)
 	}
-	RCU_TRACE(rcu_trace_sub_qlen(rcp, cb_count));
+	RCU_TRACE(rcu_trace_sub_qlen(rcp, cb_count);)
 	RCU_TRACE(trace_rcu_batch_end(rcp->name,
 				      cb_count, 0, need_resched(),
 				      is_idle_task(current),

@@ -221,7 +221,7 @@ static void __call_rcu(struct rcu_head *head,
 	local_irq_save(flags);
 	*rcp->curtail = head;
 	rcp->curtail = &head->next;
-	RCU_TRACE(rcp->qlen++);
+	RCU_TRACE(rcp->qlen++;)
 	local_irq_restore(flags);
 
 	if (unlikely(is_idle_task(current))) {

@@ -254,8 +254,8 @@ EXPORT_SYMBOL_GPL(call_rcu_bh);
 void __init rcu_init(void)
 {
 	open_softirq(RCU_SOFTIRQ, rcu_process_callbacks);
-	RCU_TRACE(reset_cpu_stall_ticks(&rcu_sched_ctrlblk));
-	RCU_TRACE(reset_cpu_stall_ticks(&rcu_bh_ctrlblk));
+	RCU_TRACE(reset_cpu_stall_ticks(&rcu_sched_ctrlblk);)
+	RCU_TRACE(reset_cpu_stall_ticks(&rcu_bh_ctrlblk);)
 
 	rcu_early_boot_tests();
 }

kernel/rcu/tiny_plugin.h +2 -2

index c64b827ecbca..df3a60e19f07 100644
@@ -162,8 +162,8 @@ static void reset_cpu_stall_ticks(struct rcu_ctrlblk *rcp)
 
 static void check_cpu_stalls(void)
 {
-	RCU_TRACE(check_cpu_stall(&rcu_bh_ctrlblk));
-	RCU_TRACE(check_cpu_stall(&rcu_sched_ctrlblk));
+	RCU_TRACE(check_cpu_stall(&rcu_bh_ctrlblk);)
+	RCU_TRACE(check_cpu_stall(&rcu_sched_ctrlblk);)
 }
 
 #endif /* #ifdef CONFIG_RCU_TRACE */

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-12 17:40 UTC #17 lore raw

PATCH 7/40 rcu: Semicolon inside RCU_TRACE() for tree.c

---
 kernel/rcu/tree.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

kernel/rcu/tree.c +4 -4

index df8f6401f2d5..e8843bf0f5e1 100644
@@ -2793,14 +2793,14 @@ static void rcu_adopt_orphan_cbs(struct rcu_state *rsp, unsigned long flags)
  */
 static void rcu_cleanup_dying_cpu(struct rcu_state *rsp)
 {
-	RCU_TRACE(unsigned long mask);
-	RCU_TRACE(struct rcu_data *rdp = this_cpu_ptr(rsp->rda));
-	RCU_TRACE(struct rcu_node *rnp = rdp->mynode);
+	RCU_TRACE(unsigned long mask;)
+	RCU_TRACE(struct rcu_data *rdp = this_cpu_ptr(rsp->rda);)
+	RCU_TRACE(struct rcu_node *rnp = rdp->mynode;)
 
 	if (!IS_ENABLED(CONFIG_HOTPLUG_CPU))
 		return;
 
-	RCU_TRACE(mask = rdp->grpmask);
+	RCU_TRACE(mask = rdp->grpmask;)
 	trace_rcu_grace_period(rsp->name,
 			       rnp->gpnum + 1 - !!(rnp->qsmask & mask),
 			       TPS("cpuofl"));

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-12 17:40 UTC #18 lore raw

PATCH 8/40 rcu: Make call_rcu() do deferred NOCB wakeups

If a CONFIG_RCU_NOCB_CPUS kernel invokes call_rcu() with interrupts disabled, wakeups must be deferred in order to avoid self-deadlock in the cases where the disabled interrupts are due to scheduler locks being held. In this case, a flag is set and is checked on entry to extended quiescent states (usermode, idle), on exit from the RCU_SOFTIRQ handler, and when the CPU in question goes offline. However, a given CPU could avoid all of those states for a considerable length of time.

This commit therefore allows a subsequent invocation of call_rcu() with interrupts enabled to do the wakeup. It also makes the wakeup function clear the deferred-wakeup flag.

---
 kernel/rcu/tree_plugin.h | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

kernel/rcu/tree_plugin.h +4 -1

index 0a62a8f1caac..5a149091d262 100644
@@ -1768,6 +1768,7 @@ static void wake_nocb_leader(struct rcu_data *rdp, bool force)
 		return;
 	if (READ_ONCE(rdp_leader->nocb_leader_sleep) || force) {
 		/* Prior smp_mb__after_atomic() orders against prior enqueue. */
+		WRITE_ONCE(rdp->nocb_defer_wakeup, RCU_NOGP_WAKE_NOT);
 		WRITE_ONCE(rdp_leader->nocb_leader_sleep, false);
 		swake_up(&rdp_leader->nocb_wq);
 	}

@@ -1853,7 +1854,9 @@ static void __call_rcu_nocb_enqueue(struct rcu_data *rdp,
 		return;
 	}
 	len = atomic_long_read(&rdp->nocb_q_count);
-	if (old_rhpp == &rdp->nocb_head) {
+	if (old_rhpp == &rdp->nocb_head ||
+	    (rcu_nocb_need_deferred_wakeup(rdp) &&
+	     !irqs_disabled_flags(flags))) {
 		if (!irqs_disabled_flags(flags)) {
 			/* ... if queue was empty ... */
 			wake_nocb_leader(rdp, false);

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-12 17:45 UTC #19 lore raw

PATCH 9/40 rcu: Make rcu_all_qs() do deferred NOCB wakeups

If a CONFIG_RCU_NOCB_CPUS kernel invokes call_rcu() with interrupts disabled, wakeups must be deferred in order to avoid self-deadlock in the cases where the disabled interrupts are due to scheduler locks being held. In this case, a flag is set and is checked on entry to extended quiescent states (usermode, idle), on exit from the RCU_SOFTIRQ handler, when the CPU in question goes offline, and on a subsequent invocation of call_rcu(). However, a given CPU could avoid all of those states for a considerable length of time.

This commit therefore allows an invocation of rcu_all_qs() to do the wakeup. It also makes the wakeup function clear the deferred-wakeup flag.

---
 kernel/rcu/tree.c        | 6 ++++++
 kernel/rcu/tree_plugin.h | 3 ++-
 2 files changed, 8 insertions(+), 1 deletion(-)

kernel/rcu/tree.c +6 -0

index e8843bf0f5e1..eaa2b6d7abe2 100644
@@ -522,6 +522,7 @@ EXPORT_SYMBOL_GPL(rcu_note_context_switch);
 void rcu_all_qs(void)
 {
 	unsigned long flags;
+	struct rcu_state *rsp;
 
 	barrier(); /* Avoid RCU read-side critical sections leaking down. */
 	if (unlikely(raw_cpu_read(rcu_sched_qs_mask))) {

@@ -544,6 +545,11 @@ void rcu_all_qs(void)
 		rcu_sched_qs();
 		preempt_enable();
 	}
+	for_each_rcu_flavor(rsp) {
+		preempt_disable();
+		do_nocb_deferred_wakeup(this_cpu_ptr(rsp->rda));
+		preempt_enable();
+	}
 	this_cpu_inc(rcu_qs_ctr);
 	barrier(); /* Avoid RCU read-side critical sections leaking up. */
 }

kernel/rcu/tree_plugin.h +2 -1

index 5a149091d262..0e2114e8076b 100644
@@ -2201,7 +2201,8 @@ static void do_nocb_deferred_wakeup(struct rcu_data *rdp)
 {
 	int ndw;
 
-	if (!rcu_nocb_need_deferred_wakeup(rdp))
+	if (!rcu_nocb_need_deferred_wakeup(rdp) ||
+	    unlikely(rcu_scheduler_active != RCU_SCHEDULER_RUNNING))
 		return;
 	ndw = READ_ONCE(rdp->nocb_defer_wakeup);
 	WRITE_ONCE(rdp->nocb_defer_wakeup, RCU_NOGP_WAKE_NOT);

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-12 17:40 UTC #20 lore raw

PATCH 10/40 rcu: Make rcu_note_context_switch() do deferred NOCB wakeups

If a CONFIG_RCU_NOCB_CPUS kernel invokes call_rcu() with interrupts disabled, wakeups must be deferred in order to avoid self-deadlock in the cases where the disabled interrupts are due to scheduler locks being held. In this case, a flag is set and is checked on entry to extended quiescent states (usermode, idle), on exit from the RCU_SOFTIRQ handler, when the CPU in question goes offline, on a subsequent invocation of call_rcu(), and from rcu_all_qs(). However, a given CPU could avoid all of those states for a considerable length of time.

This commit therefore allows an invocation of rcu_note_context_switch() to do the wakeup. It also makes the wakeup function clear the deferred-wakeup flag.

---
 kernel/rcu/tree.c | 4 ++++
 1 file changed, 4 insertions(+)

kernel/rcu/tree.c +4 -0

index eaa2b6d7abe2..7fa46967021f 100644
@@ -495,12 +495,16 @@ static void rcu_momentary_dyntick_idle(void)
  */
 void rcu_note_context_switch(void)
 {
+	struct rcu_state *rsp;
+
 	barrier(); /* Avoid RCU read-side critical sections leaking down. */
 	trace_rcu_utilization(TPS("Start context switch"));
 	rcu_sched_qs();
 	rcu_preempt_note_context_switch();
 	if (unlikely(raw_cpu_read(rcu_sched_qs_mask)))
 		rcu_momentary_dyntick_idle();
+	for_each_rcu_flavor(rsp)
+		do_nocb_deferred_wakeup(this_cpu_ptr(rsp->rda));
 	trace_rcu_utilization(TPS("End context switch"));
 	barrier(); /* Avoid RCU read-side critical sections leaking up. */
 }

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-12 17:50 UTC #21 lore raw

PATCH 11/40 rcu: Pull rcu_sched_qs_mask into rcu_dynticks structure

The rcu_sched_qs_mask variable is yet another isolated per-CPU variable, so this commit pulls it into the pre-existing rcu_dynticks per-CPU structure.

---
 .../RCU/Design/Data-Structures/Data-Structures.html          |  9 ++++++++-
 kernel/rcu/tree.c                                            | 12 +++++-------
 kernel/rcu/tree.h                                            |  1 +
 3 files changed, 14 insertions(+), 8 deletions(-)

Documentation/RCU/Design/Data-Structures/Data-Structures.html +8 -1

index d583c653a703..bf7f266e8888 100644
@@ -1104,6 +1104,7 @@ Its fields are as follows:
   1   int dynticks_nesting;
   2   int dynticks_nmi_nesting;
   3   atomic_t dynticks;
+  4   int rcu_sched_qs_mask;
 </pre>
 
 <p>The <tt>-&gt;dynticks_nesting</tt> field counts the

@@ -1117,11 +1118,17 @@ NMIs are counted by the <tt>-&gt;dynticks_nmi_nesting</tt>
 field, except that NMIs that interrupt non-dyntick-idle execution
 are not counted.
 
-</p><p>Finally, the <tt>-&gt;dynticks</tt> field counts the corresponding
+</p><p>The <tt>-&gt;dynticks</tt> field counts the corresponding
 CPU's transitions to and from dyntick-idle mode, so that this counter
 has an even value when the CPU is in dyntick-idle mode and an odd
 value otherwise.
 
+</p><p>Finally, the  <tt>-&gt;rcu_sched_qs_mask</tt> field is used
+to record the fact that the RCU core code would really like to
+see a quiescent state from the corresponding CPU.
+This flag is checked by RCU's context-switch and <tt>cond_resched()</tt>
+code, which provide a momentary idle sojourn in response.
+
 <table>
 <tr><th>&nbsp;</th></tr>
 <tr><th align="left">Quick Quiz:</th></tr>

kernel/rcu/tree.c +5 -7

index 7fa46967021f..315647d4e4cd 100644
@@ -272,8 +272,6 @@ void rcu_bh_qs(void)
 	}
 }
 
-static DEFINE_PER_CPU(int, rcu_sched_qs_mask);
-
 /*
  * Steal a bit from the bottom of ->dynticks for idle entry/exit
  * control.  Initially this is for TLB flushing.

@@ -464,8 +462,8 @@ static void rcu_momentary_dyntick_idle(void)
 	 * Yes, we can lose flag-setting operations.  This is OK, because
 	 * the flag will be set again after some delay.
 	 */
-	resched_mask = raw_cpu_read(rcu_sched_qs_mask);
-	raw_cpu_write(rcu_sched_qs_mask, 0);
+	resched_mask = raw_cpu_read(rcu_dynticks.rcu_sched_qs_mask);
+	raw_cpu_write(rcu_dynticks.rcu_sched_qs_mask, 0);
 
 	/* Find the flavor that needs a quiescent state. */
 	for_each_rcu_flavor(rsp) {

@@ -501,7 +499,7 @@ void rcu_note_context_switch(void)
 	trace_rcu_utilization(TPS("Start context switch"));
 	rcu_sched_qs();
 	rcu_preempt_note_context_switch();
-	if (unlikely(raw_cpu_read(rcu_sched_qs_mask)))
+	if (unlikely(raw_cpu_read(rcu_dynticks.rcu_sched_qs_mask)))
 		rcu_momentary_dyntick_idle();
 	for_each_rcu_flavor(rsp)
 		do_nocb_deferred_wakeup(this_cpu_ptr(rsp->rda));

@@ -529,7 +527,7 @@ void rcu_all_qs(void)
 	struct rcu_state *rsp;
 
 	barrier(); /* Avoid RCU read-side critical sections leaking down. */
-	if (unlikely(raw_cpu_read(rcu_sched_qs_mask))) {
+	if (unlikely(raw_cpu_read(rcu_dynticks.rcu_sched_qs_mask))) {
 		local_irq_save(flags);
 		rcu_momentary_dyntick_idle();
 		local_irq_restore(flags);

@@ -1361,7 +1359,7 @@ static int rcu_implicit_dynticks_qs(struct rcu_data *rdp,
 	 * is set too high, we override with half of the RCU CPU stall
 	 * warning delay.
 	 */
-	rcrmp = &per_cpu(rcu_sched_qs_mask, rdp->cpu);
+	rcrmp = &per_cpu(rcu_dynticks.rcu_sched_qs_mask, rdp->cpu);
 	if (time_after(jiffies, rdp->rsp->gp_start + jtsq) ||
 	    time_after(jiffies, rdp->rsp->jiffies_resched)) {
 		if (!(READ_ONCE(*rcrmp) & rdp->rsp->flavor_mask)) {

kernel/rcu/tree.h +1 -0

index 7468b4de7e0c..e298281984dc 100644
@@ -113,6 +113,7 @@ struct rcu_dynticks {
 				    /* Process level is worth LLONG_MAX/2. */
 	int dynticks_nmi_nesting;   /* Track NMI nesting level. */
 	atomic_t dynticks;	    /* Even value for idle, else odd. */
+	int rcu_sched_qs_mask;      /* GP old, need quiescent state. */
 #ifdef CONFIG_NO_HZ_FULL_SYSIDLE
 	long long dynticks_idle_nesting;
 				    /* irq/process nesting level from idle. */

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-12 17:43 UTC #22 lore raw

PATCH 12/40 rcu: Pull rcu_qs_ctr into rcu_dynticks structure

The rcu_qs_ctr variable is yet another isolated per-CPU variable, so this commit pulls it into the pre-existing rcu_dynticks per-CPU structure.

---
 .../RCU/Design/Data-Structures/Data-Structures.html       | 12 ++++++++++--
 kernel/rcu/tree.c                                         | 15 ++++++---------
 kernel/rcu/tree.h                                         |  3 ++-
 kernel/rcu/tree_trace.c                                   |  4 +---
 4 files changed, 19 insertions(+), 15 deletions(-)

Documentation/RCU/Design/Data-Structures/Data-Structures.html +10 -2

index bf7f266e8888..3d0311657533 100644
@@ -1105,6 +1105,7 @@ Its fields are as follows:
   2   int dynticks_nmi_nesting;
   3   atomic_t dynticks;
   4   int rcu_sched_qs_mask;
+  5   unsigned long rcu_qs_ctr;
 </pre>
 
 <p>The <tt>-&gt;dynticks_nesting</tt> field counts the

@@ -1123,12 +1124,19 @@ CPU's transitions to and from dyntick-idle mode, so that this counter
 has an even value when the CPU is in dyntick-idle mode and an odd
 value otherwise.
 
-</p><p>Finally, the  <tt>-&gt;rcu_sched_qs_mask</tt> field is used
+</p><p>The <tt>-&gt;rcu_sched_qs_mask</tt> field is used
 to record the fact that the RCU core code would really like to
-see a quiescent state from the corresponding CPU.
+see a quiescent state from the corresponding CPU, so much so that
+it is willing to call for heavy-weight dyntick-counter operations.
 This flag is checked by RCU's context-switch and <tt>cond_resched()</tt>
 code, which provide a momentary idle sojourn in response.
 
+</p><p>Finally the <tt>-&gt;rcu_qs_ctr</tt> field is used to record
+quiescent states from <tt>cond_resched()</tt>.
+Because <tt>cond_resched()</tt> can execute quite frequently, this
+must be quite lightweight, as in a non-atomic increment of this
+per-CPU field.
+
 <table>
 <tr><th>&nbsp;</th></tr>
 <tr><th align="left">Quick Quiz:</th></tr>

kernel/rcu/tree.c +6 -9

index 315647d4e4cd..be40ebb4f0bf 100644
@@ -436,9 +436,6 @@ bool rcu_eqs_special_set(int cpu)
 	return true;
 }
 
-DEFINE_PER_CPU_SHARED_ALIGNED(unsigned long, rcu_qs_ctr);
-EXPORT_PER_CPU_SYMBOL_GPL(rcu_qs_ctr);
-
 /*
  * Let the RCU core know that this CPU has gone through the scheduler,
  * which is a quiescent state.  This is called when the need for a

@@ -552,7 +549,7 @@ void rcu_all_qs(void)
 		do_nocb_deferred_wakeup(this_cpu_ptr(rsp->rda));
 		preempt_enable();
 	}
-	this_cpu_inc(rcu_qs_ctr);
+	this_cpu_inc(rcu_dynticks.rcu_qs_ctr);
 	barrier(); /* Avoid RCU read-side critical sections leaking up. */
 }
 EXPORT_SYMBOL_GPL(rcu_all_qs);

@@ -1325,7 +1322,7 @@ static int rcu_implicit_dynticks_qs(struct rcu_data *rdp,
 	 */
 	rnp = rdp->mynode;
 	if (time_after(jiffies, rdp->rsp->gp_start + jtsq) &&
-	    READ_ONCE(rdp->rcu_qs_ctr_snap) != per_cpu(rcu_qs_ctr, rdp->cpu) &&
+	    READ_ONCE(rdp->rcu_qs_ctr_snap) != per_cpu(rcu_dynticks.rcu_qs_ctr, rdp->cpu) &&
 	    READ_ONCE(rdp->gpnum) == rnp->gpnum && !rdp->gpwrap) {
 		trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, TPS("rqc"));
 		return 1;

@@ -2034,7 +2031,7 @@ static bool __note_gp_changes(struct rcu_state *rsp, struct rcu_node *rnp,
 		trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("cpustart"));
 		need_gp = !!(rnp->qsmask & rdp->grpmask);
 		rdp->cpu_no_qs.b.norm = need_gp;
-		rdp->rcu_qs_ctr_snap = __this_cpu_read(rcu_qs_ctr);
+		rdp->rcu_qs_ctr_snap = __this_cpu_read(rcu_dynticks.rcu_qs_ctr);
 		rdp->core_needs_qs = need_gp;
 		zero_cpu_stall_ticks(rdp);
 		WRITE_ONCE(rdp->gpwrap, false);

@@ -2632,7 +2629,7 @@ rcu_report_qs_rdp(int cpu, struct rcu_state *rsp, struct rcu_data *rdp)
 		 * within the current grace period.
 		 */
 		rdp->cpu_no_qs.b.norm = true;	/* need qs for new gp. */
-		rdp->rcu_qs_ctr_snap = __this_cpu_read(rcu_qs_ctr);
+		rdp->rcu_qs_ctr_snap = __this_cpu_read(rcu_dynticks.rcu_qs_ctr);
 		raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
 		return;
 	}

@@ -3630,7 +3627,7 @@ static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp)
 	/* Is the RCU core waiting for a quiescent state from this CPU? */
 	if (rcu_scheduler_fully_active &&
 	    rdp->core_needs_qs && rdp->cpu_no_qs.b.norm &&
-	    rdp->rcu_qs_ctr_snap == __this_cpu_read(rcu_qs_ctr)) {
+	    rdp->rcu_qs_ctr_snap == __this_cpu_read(rcu_dynticks.rcu_qs_ctr)) {
 		rdp->n_rp_core_needs_qs++;
 	} else if (rdp->core_needs_qs && !rdp->cpu_no_qs.b.norm) {
 		rdp->n_rp_report_qs++;

@@ -3943,7 +3940,7 @@ rcu_init_percpu_data(int cpu, struct rcu_state *rsp)
 	rdp->gpnum = rnp->completed; /* Make CPU later note any new GP. */
 	rdp->completed = rnp->completed;
 	rdp->cpu_no_qs.b.norm = true;
-	rdp->rcu_qs_ctr_snap = per_cpu(rcu_qs_ctr, cpu);
+	rdp->rcu_qs_ctr_snap = per_cpu(rcu_dynticks.rcu_qs_ctr, cpu);
 	rdp->core_needs_qs = false;
 	trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("cpuonl"));
 	raw_spin_unlock_irqrestore_rcu_node(rnp, flags);

kernel/rcu/tree.h +2 -1

index e298281984dc..76e4467bc765 100644
@@ -113,7 +113,8 @@ struct rcu_dynticks {
 				    /* Process level is worth LLONG_MAX/2. */
 	int dynticks_nmi_nesting;   /* Track NMI nesting level. */
 	atomic_t dynticks;	    /* Even value for idle, else odd. */
-	int rcu_sched_qs_mask;      /* GP old, need quiescent state. */
+	int rcu_sched_qs_mask;      /* GP old, need heavy quiescent state. */
+	unsigned long rcu_qs_ctr;   /* Light universal quiescent state ctr. */
 #ifdef CONFIG_NO_HZ_FULL_SYSIDLE
 	long long dynticks_idle_nesting;
 				    /* irq/process nesting level from idle. */

kernel/rcu/tree_trace.c +1 -3

index 8751a748499a..65b43be38e68 100644
@@ -45,8 +45,6 @@
 #define RCU_TREE_NONCORE
 #include "tree.h"
 
-DECLARE_PER_CPU_SHARED_ALIGNED(unsigned long, rcu_qs_ctr);
-
 static int r_open(struct inode *inode, struct file *file,
 					const struct seq_operations *op)
 {

@@ -121,7 +119,7 @@ static void print_one_rcu_data(struct seq_file *m, struct rcu_data *rdp)
 		   cpu_is_offline(rdp->cpu) ? '!' : ' ',
 		   ulong2long(rdp->completed), ulong2long(rdp->gpnum),
 		   rdp->cpu_no_qs.b.norm,
-		   rdp->rcu_qs_ctr_snap == per_cpu(rcu_qs_ctr, rdp->cpu),
+		   rdp->rcu_qs_ctr_snap == per_cpu(rdp->dynticks->rcu_qs_ctr, rdp->cpu),
 		   rdp->core_needs_qs);
 	seq_printf(m, " dt=%d/%llx/%d df=%lu",
 		   rcu_dynticks_snap(rdp->dynticks),

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-12 17:49 UTC #23 lore raw

PATCH 13/40 rcu: Eliminate flavor scan in rcu_momentary_dyntick_idle()

The rcu_momentary_dyntick_idle() function scans the RCU flavors, checking that one of them still needs a quiescent state before doing an expensive atomic operation on the ->dynticks counter. However, this check reduces overhead only after a rare race condition, and increases complexity. This commit therefore removes the scan and the mechanism enabling the scan.

---
 .../Design/Data-Structures/Data-Structures.html    |  4 +-
 kernel/rcu/tree.c                                  | 62 +++++-----------------
 kernel/rcu/tree.h                                  |  3 +-
 3 files changed, 15 insertions(+), 54 deletions(-)

Documentation/RCU/Design/Data-Structures/Data-Structures.html +2 -2

index 3d0311657533..e4bf20a68fa3 100644
@@ -1104,7 +1104,7 @@ Its fields are as follows:
   1   int dynticks_nesting;
   2   int dynticks_nmi_nesting;
   3   atomic_t dynticks;
-  4   int rcu_sched_qs_mask;
+  4   bool rcu_need_heavy_qs;
   5   unsigned long rcu_qs_ctr;
 </pre>

@@ -1124,7 +1124,7 @@ CPU's transitions to and from dyntick-idle mode, so that this counter
 has an even value when the CPU is in dyntick-idle mode and an odd
 value otherwise.
 
-</p><p>The <tt>-&gt;rcu_sched_qs_mask</tt> field is used
+</p><p>The <tt>-&gt;rcu_need_heavy_qs</tt> field is used
 to record the fact that the RCU core code would really like to
 see a quiescent state from the corresponding CPU, so much so that
 it is willing to call for heavy-weight dyntick-counter operations.

kernel/rcu/tree.c +12 ~~-50~~

115 lines not on this page · raw message

kernel/rcu/tree.h +1 -2

index 76e4467bc765..b212cd0f22c7 100644
@@ -113,7 +113,7 @@ struct rcu_dynticks {
 				    /* Process level is worth LLONG_MAX/2. */
 	int dynticks_nmi_nesting;   /* Track NMI nesting level. */
 	atomic_t dynticks;	    /* Even value for idle, else odd. */
-	int rcu_sched_qs_mask;      /* GP old, need heavy quiescent state. */
+	bool rcu_need_heavy_qs;      /* GP old, need heavy quiescent state. */
 	unsigned long rcu_qs_ctr;   /* Light universal quiescent state ctr. */
 #ifdef CONFIG_NO_HZ_FULL_SYSIDLE
 	long long dynticks_idle_nesting;

@@ -484,7 +484,6 @@ struct rcu_state {
 	struct rcu_node *level[RCU_NUM_LVLS + 1];
 						/* Hierarchy levels (+1 to */
 						/*  shut bogus gcc warning) */
-	u8 flavor_mask;				/* bit in flavor mask. */
 	struct rcu_data __percpu *rda;		/* pointer of percu rcu_data. */
 	call_rcu_func_t call;			/* call_rcu() flavor. */
 	int ncpus;				/* # CPUs seen so far. */

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-12 17:49 UTC #24 lore raw

PATCH 14/40 rcu: Place guard on rcu_all_qs() and rcu_note_context_switch() actions

The rcu_all_qs() and rcu_note_context_switch() do a series of checks, taking various actions to supply RCU with quiescent states, depending on the outcomes of the various checks. This is a bit much for scheduling fastpaths, so this commit creates a separate ->rcu_urgent_qs field in the rcu_dynticks structure that acts as a global guard for these checks. Thus, in the common case, rcu_all_qs() and rcu_note_context_switch() check the ->rcu_urgent_qs field, find it false, and simply return.

Cc Peter Zijlstra <[email protected]>

---
 .../Design/Data-Structures/Data-Structures.html    | 11 +++++-
 kernel/rcu/tree.c                                  | 43 +++++++++++++---------
 kernel/rcu/tree.h                                  |  3 +-
 kernel/rcu/tree_exp.h                              |  2 +
 kernel/rcu/tree_plugin.h                           |  8 +++-
 5 files changed, 45 insertions(+), 22 deletions(-)

Documentation/RCU/Design/Data-Structures/Data-Structures.html +10 -1

index e4bf20a68fa3..4dec89097559 100644
@@ -1106,6 +1106,7 @@ Its fields are as follows:
   3   atomic_t dynticks;
   4   bool rcu_need_heavy_qs;
   5   unsigned long rcu_qs_ctr;
+  6   bool rcu_urgent_qs;
 </pre>
 
 <p>The <tt>-&gt;dynticks_nesting</tt> field counts the

@@ -1131,12 +1132,20 @@ it is willing to call for heavy-weight dyntick-counter operations.
 This flag is checked by RCU's context-switch and <tt>cond_resched()</tt>
 code, which provide a momentary idle sojourn in response.
 
-</p><p>Finally the <tt>-&gt;rcu_qs_ctr</tt> field is used to record
+</p><p>The <tt>-&gt;rcu_qs_ctr</tt> field is used to record
 quiescent states from <tt>cond_resched()</tt>.
 Because <tt>cond_resched()</tt> can execute quite frequently, this
 must be quite lightweight, as in a non-atomic increment of this
 per-CPU field.
 
+</p><p>Finally, the <tt>-&gt;rcu_urgent_qs</tt> field is used to record
+the fact that the RCU core code would really like to see a quiescent
+state from the corresponding CPU, with the various other fields indicating
+just how badly RCU wants this quiescent state.
+This flag is checked by RCU's context-switch and <tt>cond_resched()</tt>
+code, which, if nothing else, non-atomically increment <tt>-&gt;rcu_qs_ctr</tt>
+in response.
+
 <table>
 <tr><th>&nbsp;</th></tr>
 <tr><th align="left">Quick Quiz:</th></tr>

kernel/rcu/tree.c +25 ~~-18~~

92 lines not on this page · raw message

kernel/rcu/tree.h +2 -1

index b212cd0f22c7..d2f276fc2edc 100644
@@ -113,8 +113,9 @@ struct rcu_dynticks {
 				    /* Process level is worth LLONG_MAX/2. */
 	int dynticks_nmi_nesting;   /* Track NMI nesting level. */
 	atomic_t dynticks;	    /* Even value for idle, else odd. */
-	bool rcu_need_heavy_qs;      /* GP old, need heavy quiescent state. */
+	bool rcu_need_heavy_qs;     /* GP old, need heavy quiescent state. */
 	unsigned long rcu_qs_ctr;   /* Light universal quiescent state ctr. */
+	bool rcu_urgent_qs;	    /* GP old need light quiescent state. */
 #ifdef CONFIG_NO_HZ_FULL_SYSIDLE
 	long long dynticks_idle_nesting;
 				    /* irq/process nesting level from idle. */

kernel/rcu/tree_exp.h +2 -0

index a7b639ccd46e..a1f52bbe9db6 100644
@@ -331,6 +331,8 @@ static void sync_sched_exp_handler(void *data)
 		return;
 	}
 	__this_cpu_write(rcu_sched_data.cpu_no_qs.b.exp, true);
+	/* Store .exp before .rcu_urgent_qs. */
+	smp_store_release(this_cpu_ptr(&rcu_dynticks.rcu_urgent_qs), true);
 	resched_cpu(smp_processor_id());
 }

kernel/rcu/tree_plugin.h +6 -2

index 0e2114e8076b..4a9df7ea89a4 100644
@@ -1863,7 +1863,9 @@ static void __call_rcu_nocb_enqueue(struct rcu_data *rdp,
 			trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu,
 					    TPS("WakeEmpty"));
 		} else {
-			rdp->nocb_defer_wakeup = RCU_NOGP_WAKE;
+			WRITE_ONCE(rdp->nocb_defer_wakeup, RCU_NOGP_WAKE);
+			/* Store ->nocb_defer_wakeup before ->rcu_urgent_qs. */
+			smp_store_release(this_cpu_ptr(&rcu_dynticks.rcu_urgent_qs), true);
 			trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu,
 					    TPS("WakeEmptyIsDeferred"));
 		}

@@ -1875,7 +1877,9 @@ static void __call_rcu_nocb_enqueue(struct rcu_data *rdp,
 			trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu,
 					    TPS("WakeOvf"));
 		} else {
-			rdp->nocb_defer_wakeup = RCU_NOGP_WAKE_FORCE;
+			WRITE_ONCE(rdp->nocb_defer_wakeup, RCU_NOGP_WAKE_FORCE);
+			/* Store ->nocb_defer_wakeup before ->rcu_urgent_qs. */
+			smp_store_release(this_cpu_ptr(&rcu_dynticks.rcu_urgent_qs), true);
 			trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu,
 					    TPS("WakeOvfIsDeferred"));
 		}

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-12 17:46 UTC #25 lore raw

PATCH 15/40 rcu: Default RCU_FANOUT_LEAF to 16 unless explicitly changed

If the RCU_EXPERT Kconfig option is not set (the default), then the RCU_FANOUT_LEAF Kconfig option will not be defined, which will cause the leaf-level rcu_node tree fanout to default to 32 on 32-bit systems and 64 on 64-bit systems. This can result in excessive lock contention. This commit therefore changes the computation of the leaf-level rcu_node tree fanout so that the result will be 16 unless an explicit Kconfig or kernel-boot setting says otherwise.

Reported-by Peter Zijlstra <[email protected]>

---
 kernel/rcu/tree.h | 6 +-----
 1 file changed, 1 insertion(+), 5 deletions(-)

kernel/rcu/tree.h +1 -5

index d2f276fc2edc..376c01e539c7 100644
@@ -52,11 +52,7 @@
 #ifdef CONFIG_RCU_FANOUT_LEAF
 #define RCU_FANOUT_LEAF CONFIG_RCU_FANOUT_LEAF
 #else /* #ifdef CONFIG_RCU_FANOUT_LEAF */
-# ifdef CONFIG_64BIT
-# define RCU_FANOUT_LEAF 64
-# else
-# define RCU_FANOUT_LEAF 32
-# endif
+#define RCU_FANOUT_LEAF 16
 #endif /* #else #ifdef CONFIG_RCU_FANOUT_LEAF */
 
 #define RCU_FANOUT_1	      (RCU_FANOUT_LEAF)

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-12 17:43 UTC #26 lore raw

PATCH 16/40 srcu: Abstract multi-tail callback list handling

RCU has only one multi-tail callback list, which is implemented via the nxtlist, nxttail, nxtcompleted, qlen_lazy, and qlen fields in the rcu_data structure, and whose operations are open-code throughout the Tree RCU implementation. This has been more or less OK in the past, but upcoming callback-list optimizations in SRCU could really use a multi-tail callback list there as well.

This commit therefore abstracts the multi-tail callback list handling into a new kernel/rcu/rcu_segcblist.h file, and uses this new API. The simple head-and-tail pointer callback list is also abstracted and applied everywhere except for the NOCB callback-offload lists. (Yes, the plan is to apply them there as well, but this commit is already bigger than would be good.)

---
 kernel/rcu/rcu_segcblist.h | 626 +++++++++++++++++++++++++++++++++++++++++++++
 kernel/rcu/tree.c          | 348 ++++++++-----------------
 kernel/rcu/tree.h          |  41 +--
 kernel/rcu/tree_plugin.h   |  54 ++--
 kernel/rcu/tree_trace.c    |  21 +-
 5 files changed, 781 insertions(+), 309 deletions(-)
 create mode 100644 kernel/rcu/rcu_segcblist.h

kernel/rcu/rcu_segcblist.h new file +626 -0

new file mode 100644
index 000000000000..55b20ab5f940
@@ -0,0 +1,626 @@
+/*
+ * RCU segmented callback lists
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, you can access it online at
+ * http://www.gnu.org/licenses/gpl-2.0.html.
+ *
+ * Copyright IBM Corporation, 2017
+ *
+ * Authors: Paul E. McKenney <[email protected]>
+ */
+
+#ifndef __KERNEL_RCU_SEGCBLIST_H
+#define __KERNEL_RCU_SEGCBLIST_H
+
+/* Simple unsegmented callback lists. */
+struct rcu_cblist {
+	struct rcu_head *head;
+	struct rcu_head **tail;
+	long len;
+	long len_lazy;
+};
+
+#define RCU_CBLIST_INITIALIZER(n) { .head = NULL, .tail = &n.head }
+
+/* Initialize simple callback list. */
+static inline void rcu_cblist_init(struct rcu_cblist *rclp)
+{
+	rclp->head = NULL;
+	rclp->tail = &rclp->head;
+	rclp->len = 0;
+	rclp->len_lazy = 0;
+}
+
+/* Is simple callback list empty? */
+static inline bool rcu_cblist_empty(struct rcu_cblist *rclp)
+{
+	return !rclp->head;
+}
+
+/* Return number of callbacks in simple callback list. */
+static inline long rcu_cblist_n_cbs(struct rcu_cblist *rclp)
+{
+	return rclp->len;
+}
+
+/* Return number of lazy callbacks in simple callback list. */
+static inline long rcu_cblist_n_lazy_cbs(struct rcu_cblist *rclp)
+{
+	return rclp->len_lazy;
+}
+
+/*
+ * Debug function to actually count the number of callbacks.
+ * If the number exceeds the limit specified, return -1.
+ */
+static inline long rcu_cblist_count_cbs(struct rcu_cblist *rclp, long lim)
+{
+	int cnt = 0;
+	struct rcu_head **rhpp = &rclp->head;
+
+	for (;;) {
+		if (!*rhpp)
+			return cnt;
+		if (++cnt > lim)
+			return -1;
+		rhpp = &(*rhpp)->next;
+	}
+}
+
+/*
+ * Dequeue the oldest rcu_head structure from the specified callback
+ * list.  This function assumes that the callback is non-lazy, but
+ * the caller can later invoke rcu_cblist_dequeued_lazy() if it
+ * finds otherwise (and if it cares about laziness).  This allows
+ * different users to have different ways of determining laziness.
+ */
+static inline struct rcu_head *rcu_cblist_dequeue(struct rcu_cblist *rclp)
+{
+	struct rcu_head *rhp;
+
+	rhp = rclp->head;
+	if (!rhp)
+		return NULL;
+	prefetch(rhp);
+	rclp->len--;
+	rclp->head = rhp->next;
+	if (!rclp->head)
+		rclp->tail = &rclp->head;
+	return rhp;
+}
+
+/*
+ * Account for the fact that a previously dequeued callback turned out
+ * to be marked as lazy.
+ */
+static inline void rcu_cblist_dequeued_lazy(struct rcu_cblist *rclp)
+{
+	rclp->len_lazy--;
+}
+
+/*
+ * Interim function to return rcu_cblist head pointer.  Longer term, the
+ * rcu_cblist will be used more pervasively, removing the need for this
+ * function.
+ */
+static inline struct rcu_head *rcu_cblist_head(struct rcu_cblist *rclp)
+{
+	return rclp->head;
+}
+
+/*
+ * Interim function to return rcu_cblist head pointer.  Longer term, the
+ * rcu_cblist will be used more pervasively, removing the need for this
+ * function.
+ */
+static inline struct rcu_head **rcu_cblist_tail(struct rcu_cblist *rclp)
+{
+	WARN_ON_ONCE(rcu_cblist_empty(rclp));
+	return rclp->tail;
+}
+
+/* Complicated segmented callback lists.  ;-) */
+
+/*
+ * Index values for segments in rcu_segcblist structure.
+ *
+ * The segments are as follows:
+ *
+ * [head, *tails[RCU_DONE_TAIL]):
+ *	Callbacks whose grace period has elapsed, and thus can be invoked.
+ * [*tails[RCU_DONE_TAIL], *tails[RCU_WAIT_TAIL]):
+ *	Callbacks waiting for the current GP from the current CPU's viewpoint.
+ * [*tails[RCU_WAIT_TAIL], *tails[RCU_NEXT_READY_TAIL]):
+ *	Callbacks that arrived before the next GP started, again from
+ *	the current CPU's viewpoint.  These can be handled by the next GP.
+ * [*tails[RCU_NEXT_READY_TAIL], *tails[RCU_NEXT_TAIL]):
+ *	Callbacks that might have arrived after the next GP started.
+ *	There is some uncertainty as to when a given GP starts and
+ *	ends, but a CPU knows the exact times if it is the one starting
+ *	or ending the GP.  Other CPUs know that the previous GP ends
+ *	before the next one starts.
+ *
+ * Note that RCU_WAIT_TAIL cannot be empty unless RCU_NEXT_READY_TAIL is also
+ * empty.
+ *
+ * The ->gp_seq[] array contains the grace-period number at which the
+ * corresponding segment of callbacks will be ready to invoke.  A given
+ * element of this array is meaningful only when the corresponding segment
+ * is non-empty, and it is never valid for RCU_DONE_TAIL (whose callbacks
+ * are already ready to invoke) or for RCU_NEXT_TAIL (whose callbacks have
+ * not yet been assigned a grace-period number).
+ */
+#define RCU_DONE_TAIL		0	/* Also RCU_WAIT head. */
+#define RCU_WAIT_TAIL		1	/* Also RCU_NEXT_READY head. */
+#define RCU_NEXT_READY_TAIL	2	/* Also RCU_NEXT head. */
+#define RCU_NEXT_TAIL		3
+#define RCU_CBLIST_NSEGS	4
+
+struct rcu_segcblist {
+	struct rcu_head *head;
+	struct rcu_head **tails[RCU_CBLIST_NSEGS];
+	unsigned long gp_seq[RCU_CBLIST_NSEGS];
+	long len;
+	long len_lazy;
+};
+
+/*
+ * Initialize an rcu_segcblist structure.
+ */
+static inline void rcu_segcblist_init(struct rcu_segcblist *rsclp)
+{
+	int i;
+
+	BUILD_BUG_ON(RCU_NEXT_TAIL + 1 != ARRAY_SIZE(rsclp->gp_seq));
+	BUILD_BUG_ON(ARRAY_SIZE(rsclp->tails) != ARRAY_SIZE(rsclp->gp_seq));
+	rsclp->head = NULL;
+	for (i = 0; i < RCU_CBLIST_NSEGS; i++)
+		rsclp->tails[i] = &rsclp->head;
+	rsclp->len = 0;
+	rsclp->len_lazy = 0;
+}
+
+/*
+ * Is the specified rcu_segcblist structure empty?
+ *
+ * But careful!  The fact that the ->head field is NULL does not
+ * necessarily imply that there are no callbacks associated with
+ * this structure.  When callbacks are being invoked, they are
+ * removed as a group.  If callback invocation must be preempted,
+ * the remaining callbacks will be added back to the list.  Either
+ * way, the counts are updated later.
+ *
+ * So it is often the case that rcu_segcblist_n_cbs() should be used
+ * instead.
+ */
+static inline bool rcu_segcblist_empty(struct rcu_segcblist *rsclp)
+{
+	return !rsclp->head;
+}
+
+/* Return number of callbacks in segmented callback list. */
+static inline long rcu_segcblist_n_cbs(struct rcu_segcblist *rsclp)
+{
+	return READ_ONCE(rsclp->len);
+}
+
+/* Return number of lazy callbacks in segmented callback list. */
+static inline long rcu_segcblist_n_lazy_cbs(struct rcu_segcblist *rsclp)
+{
+	return rsclp->len_lazy;
+}
+
+/* Return number of lazy callbacks in segmented callback list. */
+static inline long rcu_segcblist_n_nonlazy_cbs(struct rcu_segcblist *rsclp)
+{
+	return rsclp->len - rsclp->len_lazy;
+}
+
+/*
+ * Is the specified rcu_segcblist enabled, for example, not corresponding
+ * to an offline or callback-offloaded CPU?
+ */
+static inline bool rcu_segcblist_is_enabled(struct rcu_segcblist *rsclp)
+{
+	return !!rsclp->tails[RCU_NEXT_TAIL];
+}
+
+/*
+ * Disable the specified rcu_segcblist structure, so that callbacks can
+ * no longer be posted to it.  This structure must be empty.
+ */
+static inline void rcu_segcblist_disable(struct rcu_segcblist *rsclp)
+{
+	WARN_ON_ONCE(!rcu_segcblist_empty(rsclp));
+	WARN_ON_ONCE(rcu_segcblist_n_cbs(rsclp));
+	WARN_ON_ONCE(rcu_segcblist_n_lazy_cbs(rsclp));
+	rsclp->tails[RCU_NEXT_TAIL] = NULL;
+}
+
+/*
+ * Is the specified segment of the specified rcu_segcblist structure
+ * empty of callbacks?
+ */
+static inline bool rcu_segcblist_segempty(struct rcu_segcblist *rsclp, int seg)
+{
+	if (seg == RCU_DONE_TAIL)
+		return &rsclp->head == rsclp->tails[RCU_DONE_TAIL];
+	return rsclp->tails[seg - 1] == rsclp->tails[seg];
+}
+
+/*
+ * Are all segments following the specified segment of the specified
+ * rcu_segcblist structure empty of callbacks?  (The specified
+ * segment might well contain callbacks.)
+ */
+static inline bool rcu_segcblist_restempty(struct rcu_segcblist *rsclp, int seg)
+{
+	return !*rsclp->tails[seg];
+}
+
+/*
+ * Does the specified rcu_segcblist structure contain callbacks that
+ * are ready to be invoked?
+ */
+static inline bool rcu_segcblist_ready_cbs(struct rcu_segcblist *rsclp)
+{
+	return rcu_segcblist_is_enabled(rsclp) &&
+	       &rsclp->head != rsclp->tails[RCU_DONE_TAIL];
+}
+
+/*
+ * Does the specified rcu_segcblist structure contain callbacks that
+ * are still pending, that is, not yet ready to be invoked?
+ */
+static inline bool rcu_segcblist_pend_cbs(struct rcu_segcblist *rsclp)
+{
+	return rcu_segcblist_is_enabled(rsclp) &&
+	       !rcu_segcblist_restempty(rsclp, RCU_DONE_TAIL);
+}
+
+/*
+ * Return a pointer to the first callback in the specified rcu_segcblist
+ * structure.  This is useful for diagnostics.
+ */
+static inline struct rcu_head *
+rcu_segcblist_first_cb(struct rcu_segcblist *rsclp)
+{
+	if (rcu_segcblist_is_enabled(rsclp))
+	       return rsclp->head;
+	return NULL;
+}
+
+/*
+ * Return a pointer to the first pending callback in the specified
+ * rcu_segcblist structure.  This is useful just after posting a given
+ * callback -- if that callback is the first pending callback, then
+ * you cannot rely on someone else having already started up the required
+ * grace period.
+ */
+static inline struct rcu_head *
+rcu_segcblist_first_pend_cb(struct rcu_segcblist *rsclp)
+{
+	if (rcu_segcblist_is_enabled(rsclp))
+	       return *rsclp->tails[RCU_DONE_TAIL];
+	return NULL;
+}
+
+/*
+ * Does the specified rcu_segcblist structure contain callbacks that
+ * have not yet been processed beyond having been posted, that is,
+ * does it contain callbacks in its last segment?
+ */
+static inline bool rcu_segcblist_new_cbs(struct rcu_segcblist *rsclp)
+{
+	return rcu_segcblist_is_enabled(rsclp) &&
+	       !rcu_segcblist_restempty(rsclp, RCU_NEXT_READY_TAIL);
+}
+
+/*
+ * Enqueue the specified callback onto the specified rcu_segcblist
+ * structure, updating accounting as needed.  Note that the ->len
+ * field may be accessed locklessly, hence the WRITE_ONCE().
+ * The ->len field is used by rcu_barrier() and friends to determine
+ * if it must post a callback on this structure, and it is OK
+ * for rcu_barrier() to sometimes post callbacks needlessly, but
+ * absolutely not OK for it to ever miss posting a callback.
+ */
+static inline void rcu_segcblist_enqueue(struct rcu_segcblist *rsclp,
+					 struct rcu_head *rhp, bool lazy)
+{
+	WRITE_ONCE(rsclp->len, rsclp->len + 1); /* ->len sampled locklessly. */
+	if (lazy)
+		rsclp->len_lazy++;
+	smp_mb(); /* Ensure counts are updated before callback is enqueued. */
+	rhp->next = NULL;
+	*rsclp->tails[RCU_NEXT_TAIL] = rhp;
+	rsclp->tails[RCU_NEXT_TAIL] = &rhp->next;
+}
+
+/*
+ * Extract only the counts from the specified rcu_segcblist structure,
+ * and place them in the specified rcu_cblist structure.  This function
+ * supports both callback orphaning and invocation, hence the separation
+ * of counts and callbacks.  (Callbacks ready for invocation must be
+ * orphaned and adopted separately from pending callbacks, but counts
+ * apply to all callbacks.  Locking must be used to make sure that
+ * both orphaned-callbacks lists are consistent.)
+ */
+static inline void rcu_segcblist_extract_count(struct rcu_segcblist *rsclp,
+					       struct rcu_cblist *rclp)
+{
+	rclp->len_lazy += rsclp->len_lazy;
+	rclp->len += rsclp->len;
+	rsclp->len_lazy = 0;
+	WRITE_ONCE(rsclp->len, 0); /* ->len sampled locklessly. */
+}
+
+/*
+ * Extract only those callbacks ready to be invoked from the specified
+ * rcu_segcblist structure and place them in the specified rcu_cblist
+ * structure.
+ */
+static inline void rcu_segcblist_extract_done_cbs(struct rcu_segcblist *rsclp,
+						  struct rcu_cblist *rclp)
+{
+	int i;
+
+	if (!rcu_segcblist_ready_cbs(rsclp))
+		return; /* Nothing to do. */
+	*rclp->tail = rsclp->head;
+	rsclp->head = *rsclp->tails[RCU_DONE_TAIL];
+	*rsclp->tails[RCU_DONE_TAIL] = NULL;
+	rclp->tail = rsclp->tails[RCU_DONE_TAIL];
+	for (i = RCU_CBLIST_NSEGS - 1; i >= RCU_DONE_TAIL; i--)
+		if (rsclp->tails[i] == rsclp->tails[RCU_DONE_TAIL])
+			rsclp->tails[i] = &rsclp->head;
+}
+
+/*
+ * Extract only those callbacks still pending (not yet ready to be
+ * invoked) from the specified rcu_segcblist structure and place them in
+ * the specified rcu_cblist structure.  Note that this loses information
+ * about any callbacks that might have been partway done waiting for
+ * their grace period.  Too bad!  They will have to start over.
+ */
+static inline void
+rcu_segcblist_extract_pend_cbs(struct rcu_segcblist *rsclp,
+			       struct rcu_cblist *rclp)
+{
+	int i;
+
+	if (!rcu_segcblist_pend_cbs(rsclp))
+		return; /* Nothing to do. */
+	*rclp->tail = *rsclp->tails[RCU_DONE_TAIL];
+	rclp->tail = rsclp->tails[RCU_NEXT_TAIL];
+	*rsclp->tails[RCU_DONE_TAIL] = NULL;
+	for (i = RCU_DONE_TAIL + 1; i < RCU_CBLIST_NSEGS; i++)
+		rsclp->tails[i] = rsclp->tails[RCU_DONE_TAIL];
+}
+
+/*
+ * Move the entire contents of the specified rcu_segcblist structure,
+ * counts, callbacks, and all, to the specified rcu_cblist structure.
+ * @@@ Why do we need this???  Moving early-boot CBs to NOCB lists?
+ * @@@ Memory barrier needed?  (Not if only used at boot time...)
+ */
+static inline void rcu_segcblist_extract_all(struct rcu_segcblist *rsclp,
+					     struct rcu_cblist *rclp)
+{
+	rcu_segcblist_extract_done_cbs(rsclp, rclp);
+	rcu_segcblist_extract_pend_cbs(rsclp, rclp);
+	rcu_segcblist_extract_count(rsclp, rclp);
+}
+
+/*
+ * Insert counts from the specified rcu_cblist structure in the
+ * specified rcu_segcblist structure.
+ */
+static inline void rcu_segcblist_insert_count(struct rcu_segcblist *rsclp,
+					      struct rcu_cblist *rclp)
+{
+	
+	rsclp->len_lazy += rclp->len_lazy;
+	/* ->len sampled locklessly. */
+	WRITE_ONCE(rsclp->len, rsclp->len + rclp->len);
+	rclp->len_lazy = 0;
+	rclp->len = 0;
+}
+
+/*
+ * Move callbacks from the specified rcu_cblist to the beginning of the
+ * done-callbacks segment of the specified rcu_segcblist.
+ */
+static inline void rcu_segcblist_insert_done_cbs(struct rcu_segcblist *rsclp,
+						 struct rcu_cblist *rclp)
+{
+	int i;
+
+	if (!rclp->head)
+		return; /* No callbacks to move. */
+	*rclp->tail = rsclp->head;
+	rsclp->head = rclp->head;
+	for (i = RCU_DONE_TAIL; i < RCU_CBLIST_NSEGS; i++)
+		if (&rsclp->head == rsclp->tails[i])
+			rsclp->tails[i] = rclp->tail;
+		else
+			break;
+	rclp->head = NULL;
+	rclp->tail = &rclp->head;
+}
+
+/*
+ * Move callbacks from the specified rcu_cblist to the end of the
+ * new-callbacks segment of the specified rcu_segcblist.
+ */
+static inline void rcu_segcblist_insert_pend_cbs(struct rcu_segcblist *rsclp,
+						 struct rcu_cblist *rclp)
+{
+	if (!rclp->head)
+		return; /* Nothing to do. */
+	*rsclp->tails[RCU_NEXT_TAIL] = rclp->head;
+	rsclp->tails[RCU_NEXT_TAIL] = rclp->tail;
+	rclp->head = NULL;
+	rclp->tail = &rclp->head;
+}
+
+/*
+ * Advance the callbacks in the specified rcu_segcblist structure based
+ * on the current value passed in for the grace-period counter.
+ */
+static inline void rcu_segcblist_advance(struct rcu_segcblist *rsclp,
+					 unsigned long seq)
+{
+	int i, j;
+
+	WARN_ON_ONCE(!rcu_segcblist_is_enabled(rsclp));
+	WARN_ON_ONCE(rcu_segcblist_restempty(rsclp, RCU_DONE_TAIL));
+
+	/*
+	 * Find all callbacks whose ->gp_seq numbers indicate that they
+	 * are ready to invoke, and put them into the RCU_DONE_TAIL segment.
+	 */
+	for (i = RCU_WAIT_TAIL; i < RCU_NEXT_TAIL; i++) {
+		if (ULONG_CMP_LT(seq, rsclp->gp_seq[i]))
+			break;
+		rsclp->tails[RCU_DONE_TAIL] = rsclp->tails[i];
+	}
+
+	/* If no callbacks moved, nothing more need be done. */
+	if (i == RCU_WAIT_TAIL)
+		return;
+
+	/* Clean up tail pointers that might have been misordered above. */
+	for (j = RCU_WAIT_TAIL; j < i; j++)
+		rsclp->tails[j] = rsclp->tails[RCU_DONE_TAIL];
+
+	/*
+	 * Callbacks moved, so clean up the misordered ->tails[] pointers
+	 * that now point into the middle of the list of ready-to-invoke
+	 * callbacks.  The overall effect is to copy down the later pointers
+	 * into the gap that was created by the now-ready segments.
+	 */
+	for (j = RCU_WAIT_TAIL; i < RCU_NEXT_TAIL; i++, j++) {
+		if (rsclp->tails[j] == rsclp->tails[RCU_NEXT_TAIL])
+			break;  /* No more callbacks. */
+		rsclp->tails[j] = rsclp->tails[i];
+		rsclp->gp_seq[j] = rsclp->gp_seq[i];
+	}
+}
+
+/*
+ * "Accelerate" callbacks based on more-accurate grace-period information.
+ * The reason for this is that RCU does not synchronize the beginnings and
+ * ends of grace periods, and that callbacks are posted locally.  This in
+ * turn means that the callbacks must be labelled conservatively early
+ * on, as getting exact information would degrade both performance and
+ * scalability.  When more accurate grace-period information becomes
+ * available, previously posted callbacks can be "accelerated", marking
+ * them to complete at the end of the earlier grace period.
+ *
+ * This function operates on an rcu_segcblist structure, and also the
+ * grace-period sequence number at which new callbacks would become
+ * ready to invoke.
+ */
+static inline bool rcu_segcblist_accelerate(struct rcu_segcblist *rsclp,
+					    unsigned long seq)
+{
+	int i;
+
+	WARN_ON_ONCE(!rcu_segcblist_is_enabled(rsclp));
+	WARN_ON_ONCE(rcu_segcblist_restempty(rsclp, RCU_DONE_TAIL));
+
+	/*
+	 * Find the segment preceding the oldest segment of callbacks
+	 * whose ->gp_seq[] completion is at or after that passed in via
+	 * "seq", skipping any empty segments.  This oldest segment, along
+	 * with any later segments, can be merged in with any newly arrived
+	 * callbacks in the RCU_NEXT_TAIL segment, and assigned "seq"
+	 * as their ->gp_seq[] grace-period completion sequence number.
+	 */
+	for (i = RCU_NEXT_READY_TAIL; i > RCU_DONE_TAIL; i--)
+		if (rsclp->tails[i] != rsclp->tails[i - 1] &&
+		    ULONG_CMP_LT(rsclp->gp_seq[i], seq))
+			break;
+
+	/*
+	 * If all the segments contain callbacks that correspond to
+	 * earlier grace-period sequence numbers than "seq", leave.
+	 * Assuming that the rcu_segcblist structure has enough
+	 * segments in its arrays, this can only happen if some of
+	 * the non-done segments contain callbacks that really are
+	 * ready to invoke.  This situation will get straightened
+	 * out by the next call to rcu_segcblist_advance().
+	 *
+	 * Also advance to the oldest segment of callbacks whose
+	 * ->gp_seq[] completion is at or after that passed in via "seq",
+	 * skipping any empty segments.
+	 */
+	if (++i >= RCU_NEXT_TAIL)
+		return false;
+
+	/*
+	 * Merge all later callbacks, including newly arrived callbacks,
+	 * into the segment located by the for-loop above.  Assign "seq"
+	 * as the ->gp_seq[] value in order to correctly handle the case
+	 * where there were no pending callbacks in the rcu_segcblist
+	 * structure other than in the RCU_NEXT_TAIL segment.
+	 */
+	for (; i < RCU_NEXT_TAIL; i++) {
+		rsclp->tails[i] = rsclp->tails[RCU_NEXT_TAIL];
+		rsclp->gp_seq[i] = seq;
+	}
+	return true;
+}
+
+/*
+ * Scan the specified rcu_segcblist structure for callbacks that need
+ * a grace period later than the one specified by "seq".  We don't look
+ * at the RCU_DONE_TAIL or RCU_NEXT_TAIL segments because they don't
+ * have a grace-period sequence number.
+ */
+static inline bool rcu_segcblist_future_gp_needed(struct rcu_segcblist *rsclp,
+						  unsigned long seq)
+{
+	int i;
+
+	for (i = RCU_WAIT_TAIL; i < RCU_NEXT_TAIL; i++)
+		if (rsclp->tails[i - 1] != rsclp->tails[i] &&
+		    ULONG_CMP_LT(seq, rsclp->gp_seq[i]))
+		    	return true;
+	return false;
+}
+
+/*
+ * Interim function to return rcu_segcblist head pointer.  Longer term, the
+ * rcu_segcblist will be used more pervasively, removing the need for this
+ * function.
+ */
+static inline struct rcu_head *rcu_segcblist_head(struct rcu_segcblist *rsclp)
+{
+	return rsclp->head;
+}
+
+/*
+ * Interim function to return rcu_segcblist head pointer.  Longer term, the
+ * rcu_segcblist will be used more pervasively, removing the need for this
+ * function.
+ */
+static inline struct rcu_head **rcu_segcblist_tail(struct rcu_segcblist *rsclp)
+{
+	WARN_ON_ONCE(rcu_segcblist_empty(rsclp));
+	return rsclp->tails[RCU_NEXT_TAIL];
+}
+
+#endif /* __KERNEL_RCU_SEGCBLIST_H */

kernel/rcu/tree.c +112 ~~-236~~

580 lines not on this page · raw message

kernel/rcu/tree.h +7 ~~-34~~

61 lines not on this page · raw message

kernel/rcu/tree_plugin.h +25 ~~-29~~

95 lines not on this page · raw message

kernel/rcu/tree_trace.c +11 ~~-10~~

index 65b43be38e68..066c64071a7b 100644
@@ -41,6 +41,7 @@
 #include <linux/mutex.h>
 #include <linux/debugfs.h>
 #include <linux/seq_file.h>
+#include <linux/prefetch.h>
 
 #define RCU_TREE_NONCORE
 #include "tree.h"

@@ -128,17 +129,15 @@ static void print_one_rcu_data(struct seq_file *m, struct rcu_data *rdp)
 		   rdp->dynticks_fqs);
 	seq_printf(m, " of=%lu", rdp->offline_fqs);
 	rcu_nocb_q_lengths(rdp, &ql, &qll);
-	qll += rdp->qlen_lazy;
-	ql += rdp->qlen;
+	qll += rcu_segcblist_n_lazy_cbs(&rdp->cblist);
+	ql += rcu_segcblist_n_cbs(&rdp->cblist);
 	seq_printf(m, " ql=%ld/%ld qs=%c%c%c%c",
 		   qll, ql,
-		   ".N"[rdp->nxttail[RCU_NEXT_READY_TAIL] !=
-			rdp->nxttail[RCU_NEXT_TAIL]],
-		   ".R"[rdp->nxttail[RCU_WAIT_TAIL] !=
-			rdp->nxttail[RCU_NEXT_READY_TAIL]],
-		   ".W"[rdp->nxttail[RCU_DONE_TAIL] !=
-			rdp->nxttail[RCU_WAIT_TAIL]],
-		   ".D"[&rdp->nxtlist != rdp->nxttail[RCU_DONE_TAIL]]);
+		   ".N"[!rcu_segcblist_segempty(&rdp->cblist, RCU_NEXT_TAIL)],
+		   ".R"[!rcu_segcblist_segempty(&rdp->cblist,
+						RCU_NEXT_READY_TAIL)],
+		   ".W"[!rcu_segcblist_segempty(&rdp->cblist, RCU_WAIT_TAIL)],
+		   ".D"[!rcu_segcblist_segempty(&rdp->cblist, RCU_DONE_TAIL)]);
 #ifdef CONFIG_RCU_BOOST
 	seq_printf(m, " kt=%d/%c ktl=%x",
 		   per_cpu(rcu_cpu_has_work, rdp->cpu),

@@ -276,7 +275,9 @@ static void print_one_rcu_state(struct seq_file *m, struct rcu_state *rsp)
 	seq_printf(m, "nfqs=%lu/nfqsng=%lu(%lu) fqlh=%lu oqlen=%ld/%ld\n",
 		   rsp->n_force_qs, rsp->n_force_qs_ngp,
 		   rsp->n_force_qs - rsp->n_force_qs_ngp,
-		   READ_ONCE(rsp->n_force_qs_lh), rsp->qlen_lazy, rsp->qlen);
+		   READ_ONCE(rsp->n_force_qs_lh),
+		   rcu_cblist_n_lazy_cbs(&rsp->orphan_done),
+		   rcu_cblist_n_cbs(&rsp->orphan_done));
 	for (rnp = &rsp->node[0]; rnp - &rsp->node[0] < rcu_num_nodes; rnp++) {
 		if (rnp->level != level) {
 			seq_puts(m, "\n");

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-12 17:45 UTC #27 lore raw

PATCH 17/40 srcu: Allow SRCU to access rcu_scheduler_active

This is primarily a code-movement commit in preparation for allowing SRCU to handle early-boot SRCU grace periods.

---
 include/linux/rcutiny.h  |  6 +++---
 kernel/rcu/tiny_plugin.h |  9 +++++----
 kernel/rcu/tree.c        |  2 +-
 kernel/rcu/tree_exp.h    | 12 -----------
 kernel/rcu/update.c      | 52 +++++++++++++++++++++++++++++++-----------------
 5 files changed, 43 insertions(+), 38 deletions(-)

include/linux/rcutiny.h +3 -3

index 6c9d941e3962..5219be250f00 100644
@@ -217,14 +217,14 @@ static inline void exit_rcu(void)
 {
 }
 
-#ifdef CONFIG_DEBUG_LOCK_ALLOC
+#if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_SRCU)
 extern int rcu_scheduler_active __read_mostly;
 void rcu_scheduler_starting(void);
-#else /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
+#else /* #if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_SRCU) */
 static inline void rcu_scheduler_starting(void)
 {
 }
-#endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */
+#endif /* #else #if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_SRCU) */
 
 #if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_RCU_TRACE)

kernel/rcu/tiny_plugin.h +5 -4

index df3a60e19f07..371034e77f87 100644
@@ -52,7 +52,7 @@ static struct rcu_ctrlblk rcu_bh_ctrlblk = {
 	RCU_TRACE(.name = "rcu_bh")
 };
 
-#ifdef CONFIG_DEBUG_LOCK_ALLOC
+#if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_SRCU)
 #include <linux/kernel_stat.h>
 
 int rcu_scheduler_active __read_mostly;

@@ -65,15 +65,16 @@ EXPORT_SYMBOL_GPL(rcu_scheduler_active);
  * to RCU_SCHEDULER_RUNNING, skipping the RCU_SCHEDULER_INIT stage.
  * The reason for this is that Tiny RCU does not need kthreads, so does
  * not have to care about the fact that the scheduler is half-initialized
- * at a certain phase of the boot process.
+ * at a certain phase of the boot process.  Unless SRCU is in the mix.
  */
 void __init rcu_scheduler_starting(void)
 {
 	WARN_ON(nr_context_switches() > 0);
-	rcu_scheduler_active = RCU_SCHEDULER_RUNNING;
+	rcu_scheduler_active = IS_ENABLED(CONFIG_SRCU)
+		? RCU_SCHEDULER_INIT : RCU_SCHEDULER_RUNNING;
 }
 
-#endif /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
+#endif /* #if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_SRCU) */
 
 #ifdef CONFIG_RCU_TRACE

kernel/rcu/tree.c +1 -1

index 00390965c843..5e827d636e0b 100644
@@ -3981,7 +3981,7 @@ early_initcall(rcu_spawn_gp_kthread);
  * task is booting the system, and such primitives are no-ops).  After this
  * function is called, any synchronous grace-period primitives are run as
  * expedited, with the requesting task driving the grace period forward.
- * A later core_initcall() rcu_exp_runtime_mode() will switch to full
+ * A later core_initcall() rcu_set_runtime_mode() will switch to full
  * runtime RCU functionality.
  */
 void rcu_scheduler_starting(void)

kernel/rcu/tree_exp.h +0 ~~-12~~

index a1f52bbe9db6..51ca287828a2 100644
@@ -737,15 +737,3 @@ void synchronize_rcu_expedited(void)
 EXPORT_SYMBOL_GPL(synchronize_rcu_expedited);
 
 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */
-
-/*
- * Switch to run-time mode once Tree RCU has fully initialized.
- */
-static int __init rcu_exp_runtime_mode(void)
-{
-	rcu_test_sync_prims();
-	rcu_scheduler_active = RCU_SCHEDULER_RUNNING;
-	rcu_test_sync_prims();
-	return 0;
-}
-core_initcall(rcu_exp_runtime_mode);

kernel/rcu/update.c +34 ~~-18~~

70 lines not on this page · raw message

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-12 17:46 UTC #28 lore raw

PATCH 18/40 srcu: Allow early boot use of synchronize_srcu()

This commit checks for pre-scheduler state, and if that early in the boot process, synchronize_srcu() and friends are no-ops.

---
 kernel/rcu/srcu.c | 2 ++
 1 file changed, 2 insertions(+)

kernel/rcu/srcu.c +2 -0

index 6beeba7b0b67..1026ce24922f 100644
@@ -411,6 +411,8 @@ static void __synchronize_srcu(struct srcu_struct *sp, int trycount)
 			 lock_is_held(&rcu_sched_lock_map),
 			 "Illegal synchronize_srcu() in same-type SRCU (or in RCU) read-side critical section");
 
+	if (rcu_scheduler_active == RCU_SCHEDULER_INACTIVE)
+		return;
 	might_sleep();
 	init_completion(&rcu.completion);

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-12 17:46 UTC #29 lore raw

PATCH 19/40 rcu: Add single-element dequeue functions to rcu_segcblist

This commit adds single-element dequeue functions to rcu_segcblist. These are less efficient than using the extract and insert functions, but allow more precise debugging code. These functions are thus expected to be used only in debug builds, for example, CONFIG_PROVE_RCU.

---
 kernel/rcu/rcu_segcblist.h | 45 +++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 45 insertions(+)

kernel/rcu/rcu_segcblist.h +45 -0

index 55b20ab5f940..c32d13b368ac 100644
@@ -290,6 +290,51 @@ static inline bool rcu_segcblist_pend_cbs(struct rcu_segcblist *rsclp)
 }
 
 /*
+ * Dequeue and return the first ready-to-invoke callback.  If there
+ * are no ready-to-invoke callbacks, return NULL.  Disables interrupts
+ * to avoid interference.  Does not protect from interference from other
+ * CPUs or tasks.
+ */
+static inline struct rcu_head *
+rcu_segcblist_dequeue(struct rcu_segcblist *rsclp)
+{
+	unsigned long flags;
+	int i;
+	struct rcu_head *rhp;
+
+	local_irq_save(flags);
+	if (!rcu_segcblist_ready_cbs(rsclp)) {
+		local_irq_restore(flags);
+		return NULL;
+	}
+	rhp = rsclp->head;
+	BUG_ON(!rhp);
+	rsclp->head = rhp->next;
+	for (i = RCU_DONE_TAIL; i < RCU_CBLIST_NSEGS; i++) {
+		if (rsclp->tails[i] != &rhp->next)
+			break;
+		rsclp->tails[i] = &rsclp->head;
+	}
+	smp_mb(); /* Dequeue before decrement for rcu_barrier(). */
+	WRITE_ONCE(rsclp->len, rsclp->len - 1);
+	local_irq_restore(flags);
+	return rhp;
+}
+
+/*
+ * Account for the fact that a previously dequeued callback turned out
+ * to be marked as lazy.
+ */
+static inline void rcu_segcblist_dequeued_lazy(struct rcu_segcblist *rsclp)
+{
+	unsigned long flags;
+
+	local_irq_save(flags);
+	rsclp->len_lazy--;
+	local_irq_restore(flags);
+}
+
+/*
  * Return a pointer to the first callback in the specified rcu_segcblist
  * structure.  This is useful for diagnostics.
  */

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-12 17:46 UTC #30 lore raw

PATCH 20/40 srcu: Move rcu_seq_start() and friends to rcu.h

This commit moves rcu_seq_start(), rcu_seq_end(), rcu_seq_snap(), and rcu_seq_done() from kernel/rcu/tree.c to kernel/rcu/rcu.h. This will allow SRCU to use these functions, which in turn will allow SRCU to move from a single global callback queue to a per-CPU callback queue.

---
 kernel/rcu/rcu.h  | 40 ++++++++++++++++++++++++++++++++++++++++
 kernel/rcu/tree.c | 35 -----------------------------------
 2 files changed, 40 insertions(+), 35 deletions(-)

kernel/rcu/rcu.h +40 -0

index 8700a81daf56..91e0bf31f6ce 100644
@@ -56,6 +56,46 @@
 #define DYNTICK_TASK_EXIT_IDLE	   (DYNTICK_TASK_NEST_VALUE + \
 				    DYNTICK_TASK_FLAG)
 
+
+/*
+ * Grace-period counter management.
+ */
+
+/* Adjust sequence number for start of update-side operation. */
+static inline void rcu_seq_start(unsigned long *sp)
+{
+	WRITE_ONCE(*sp, *sp + 1);
+	smp_mb(); /* Ensure update-side operation after counter increment. */
+	WARN_ON_ONCE(!(*sp & 0x1));
+}
+
+/* Adjust sequence number for end of update-side operation. */
+static inline void rcu_seq_end(unsigned long *sp)
+{
+	smp_mb(); /* Ensure update-side operation before counter increment. */
+	WRITE_ONCE(*sp, *sp + 1);
+	WARN_ON_ONCE(*sp & 0x1);
+}
+
+/* Take a snapshot of the update side's sequence number. */
+static inline unsigned long rcu_seq_snap(unsigned long *sp)
+{
+	unsigned long s;
+
+	s = (READ_ONCE(*sp) + 3) & ~0x1;
+	smp_mb(); /* Above access must not bleed into critical section. */
+	return s;
+}
+
+/*
+ * Given a snapshot from rcu_seq_snap(), determine whether or not a
+ * full update-side operation has occurred.
+ */
+static inline bool rcu_seq_done(unsigned long *sp, unsigned long s)
+{
+	return ULONG_CMP_GE(READ_ONCE(*sp), s);
+}
+
 /*
  * debug_rcu_head_queue()/debug_rcu_head_unqueue() are used internally
  * by call_rcu() and rcu callback execution, and are therefore not part of the

kernel/rcu/tree.c +0 ~~-35~~

index 5e827d636e0b..f8b100902e21 100644
@@ -3416,41 +3416,6 @@ void cond_synchronize_sched(unsigned long oldstate)
 }
 EXPORT_SYMBOL_GPL(cond_synchronize_sched);
 
-/* Adjust sequence number for start of update-side operation. */
-static void rcu_seq_start(unsigned long *sp)
-{
-	WRITE_ONCE(*sp, *sp + 1);
-	smp_mb(); /* Ensure update-side operation after counter increment. */
-	WARN_ON_ONCE(!(*sp & 0x1));
-}
-
-/* Adjust sequence number for end of update-side operation. */
-static void rcu_seq_end(unsigned long *sp)
-{
-	smp_mb(); /* Ensure update-side operation before counter increment. */
-	WRITE_ONCE(*sp, *sp + 1);
-	WARN_ON_ONCE(*sp & 0x1);
-}
-
-/* Take a snapshot of the update side's sequence number. */
-static unsigned long rcu_seq_snap(unsigned long *sp)
-{
-	unsigned long s;
-
-	s = (READ_ONCE(*sp) + 3) & ~0x1;
-	smp_mb(); /* Above access must not bleed into critical section. */
-	return s;
-}
-
-/*
- * Given a snapshot from rcu_seq_snap(), determine whether or not a
- * full update-side operation has occurred.
- */
-static bool rcu_seq_done(unsigned long *sp, unsigned long s)
-{
-	return ULONG_CMP_GE(READ_ONCE(*sp), s);
-}
-
 /*
  * Check to see if there is any immediate RCU-related work to be done
  * by the current CPU, for the specified type of RCU, returning 1 if so.

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-12 17:46 UTC #31 lore raw

PATCH 21/40 rcu: Expedited wakeups need to be fully ordered

Expedited grace periods use workqueue handlers that wake up the requesters, but there is no lock mediating this wakeup. Therefore, memory barriers are required to ensure that the handler's memory references are seen by all to occur before synchronize_*_expedited() returns to its caller. Possibly detected by syzkaller.

Reported-by Dmitry Vyukov <[email protected]>

---
 kernel/rcu/tree_exp.h | 2 ++
 1 file changed, 2 insertions(+)

kernel/rcu/tree_exp.h +2 -0

index 51ca287828a2..027e123d93c7 100644
@@ -533,6 +533,7 @@ static void rcu_exp_wait_wake(struct rcu_state *rsp, unsigned long s)
 				rnp->exp_seq_rq = s;
 			spin_unlock(&rnp->exp_lock);
 		}
+		smp_mb(); /* All above changes before wakeup. */
 		wake_up_all(&rnp->exp_wq[(rsp->expedited_sequence >> 1) & 0x3]);
 	}
 	trace_rcu_exp_grace_period(rsp->name, s, TPS("endwake"));

@@ -614,6 +615,7 @@ static void _synchronize_rcu_expedited(struct rcu_state *rsp,
 	wait_event(rnp->exp_wq[(s >> 1) & 0x3],
 		   sync_exp_work_done(rsp,
 				      &rdp->exp_workdone0, s));
+	smp_mb(); /* Workqueue actions happen before return. */
 
 	/* Let the next expedited grace period start. */
 	mutex_unlock(&rsp->exp_mutex);

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-12 17:42 UTC #32 lore raw

PATCH 22/40 rcu: Fix warning in rcu_seq_end()

From: Dmitry Vyukov <[email protected]>

The rcu_seq_end() function increments seq signifying completion of a grace period, after that checks that the seq is even and wakes _synchronize_rcu_expedited(). The _synchronize_rcu_expedited() function uses wait_event() to wait for even seq. The problem is that wait_event() can return as soon as seq becomes even without waiting for the wakeup. In such case the warning in rcu_seq_end() can falsely fire if the next expedited grace period starts before the check.

Check that seq has good value before incrementing it.

Signed-off-by Dmitry Vyukov <[email protected]>

Cc Paul E. McKenney <[email protected]>

Cc Steven Rostedt <[email protected]>

Cc Mathieu Desnoyers <[email protected]>

---

syzkaller-triggered warning:

WARNING: CPU: 0 PID: 4832 at kernel/rcu/tree.c:3533
rcu_seq_end+0x110/0x140 kernel/rcu/tree.c:3533
CPU: 0 PID: 4832 Comm: kworker/0:3 Not tainted 4.10.0+ #276
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Bochs 01/01/2011
Workqueue: events wait_rcu_exp_gp
Call Trace:
 __dump_stack lib/dump_stack.c:15 [inline]
 dump_stack+0x2ee/0x3ef lib/dump_stack.c:51
 panic+0x1fb/0x412 kernel/panic.c:179
 __warn+0x1c4/0x1e0 kernel/panic.c:540
 warn_slowpath_null+0x2c/0x40 kernel/panic.c:583
 rcu_seq_end+0x110/0x140 kernel/rcu/tree.c:3533
 rcu_exp_gp_seq_end kernel/rcu/tree_exp.h:36 [inline]
 rcu_exp_wait_wake+0x8a9/0x1330 kernel/rcu/tree_exp.h:517
 rcu_exp_sel_wait_wake kernel/rcu/tree_exp.h:559 [inline]
 wait_rcu_exp_gp+0x83/0xc0 kernel/rcu/tree_exp.h:570
 process_one_work+0xc06/0x1c20 kernel/workqueue.c:2096
 worker_thread+0x223/0x19c0 kernel/workqueue.c:2230
 kthread+0x326/0x3f0 kernel/kthread.c:227
 ret_from_fork+0x31/0x40 arch/x86/entry/entry_64.S:430
---
---
 kernel/rcu/rcu.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

kernel/rcu/rcu.h +1 -1

index 91e0bf31f6ce..0bc1313c49e2 100644
@@ -73,8 +73,8 @@ static inline void rcu_seq_start(unsigned long *sp)
 static inline void rcu_seq_end(unsigned long *sp)
 {
 	smp_mb(); /* Ensure update-side operation before counter increment. */
+	WARN_ON_ONCE(!(*sp & 0x1));
 	WRITE_ONCE(*sp, *sp + 1);
-	WARN_ON_ONCE(*sp & 0x1);
 }
 
 /* Take a snapshot of the update side's sequence number. */

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-12 17:43 UTC #33 lore raw

PATCH 23/40 srcu: Push srcu_advance_batches() fastpath into common case

This commit simplifies the SRCU state machine by pushing the srcu_advance_batches() idle-SRCU fastpath into the common case. This is done by giving srcu_reschedule() a delay parameter, which is zero in the call from srcu_advance_batches().

This commit is a step towards numbering callbacks in order to efficiently handle per-CPU callback lists.

---
 kernel/rcu/srcu.c | 27 +++++++--------------------
 1 file changed, 7 insertions(+), 20 deletions(-)

kernel/rcu/srcu.c +7 ~~-20~~

index 1026ce24922f..febf87deaccb 100644
@@ -393,8 +393,7 @@ void call_srcu(struct srcu_struct *sp, struct rcu_head *head,
 }
 EXPORT_SYMBOL_GPL(call_srcu);
 
-static void srcu_advance_batches(struct srcu_struct *sp, int trycount);
-static void srcu_reschedule(struct srcu_struct *sp);
+static void srcu_reschedule(struct srcu_struct *sp, unsigned long delay);
 
 /*
  * Helper function for synchronize_srcu() and synchronize_srcu_expedited().

@@ -403,7 +402,6 @@ static void __synchronize_srcu(struct srcu_struct *sp, int trycount)
 {
 	struct rcu_synchronize rcu;
 	struct rcu_head *head = &rcu.head;
-	bool done = false;
 
 	RCU_LOCKDEP_WARN(lock_is_held(&sp->dep_map) ||
 			 lock_is_held(&rcu_bh_lock_map) ||

@@ -425,25 +423,15 @@ static void __synchronize_srcu(struct srcu_struct *sp, int trycount)
 		sp->running = true;
 		rcu_batch_queue(&sp->batch_check0, head);
 		spin_unlock_irq(&sp->queue_lock);
-
-		srcu_advance_batches(sp, trycount);
-		if (!rcu_batch_empty(&sp->batch_done)) {
-			BUG_ON(sp->batch_done.head != head);
-			rcu_batch_dequeue(&sp->batch_done);
-			done = true;
-		}
 		/* give the processing owner to work_struct */
-		srcu_reschedule(sp);
+		srcu_reschedule(sp, 0);
 	} else {
 		rcu_batch_queue(&sp->batch_queue, head);
 		spin_unlock_irq(&sp->queue_lock);
 	}
 
-	if (!done) {
-		wait_for_completion(&rcu.completion);
-		smp_mb(); /* Caller's later accesses after GP. */
-	}
-
+	wait_for_completion(&rcu.completion);
+	smp_mb(); /* Caller's later accesses after GP. */
 }
 
 /**

@@ -633,7 +621,7 @@ static void srcu_invoke_callbacks(struct srcu_struct *sp)
  * Finished one round of SRCU grace period.  Start another if there are
  * more SRCU callbacks queued, otherwise put SRCU into not-running state.
  */
-static void srcu_reschedule(struct srcu_struct *sp)
+static void srcu_reschedule(struct srcu_struct *sp, unsigned long delay)
 {
 	bool pending = true;

@@ -647,8 +635,7 @@ static void srcu_reschedule(struct srcu_struct *sp)
 	}
 
 	if (pending)
-		queue_delayed_work(system_power_efficient_wq,
-				   &sp->work, SRCU_INTERVAL);
+		queue_delayed_work(system_power_efficient_wq, &sp->work, delay);
 }
 
 /*

@@ -663,6 +650,6 @@ void process_srcu(struct work_struct *work)
 	srcu_collect_new(sp);
 	srcu_advance_batches(sp, 1);
 	srcu_invoke_callbacks(sp);
-	srcu_reschedule(sp);
+	srcu_reschedule(sp, SRCU_INTERVAL);
 }
 EXPORT_SYMBOL_GPL(process_srcu);

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-12 17:42 UTC #34 lore raw

PATCH 24/40 srcu: Move to state-based grace-period sequencing

The current SRCU grace-period processing might never reach the last portion of srcu_advance_batches(). This is OK given the current implementation, as the first portion, up to the try_check_zero() following the srcu_flip() is sufficient to drive grace periods forward. However, it has the unfortunate side-effect of making it impossible to determine when a given grace period has ended, and it will be necessary to efficiently trace ends of grace periods in order to efficiently handle per-CPU SRCU callback lists.

This commit therefore adds states to the SRCU grace-period processing, so that the end of a given SRCU grace period is marked by the transition to the SRCU_STATE_DONE state.

---
 include/linux/srcu.h |  10 ++++-
 kernel/rcu/srcu.c    | 111 ++++++++++++++++++++++++++++-----------------------
 2 files changed, 69 insertions(+), 52 deletions(-)

include/linux/srcu.h +8 -2

index a598cf3ac70c..f149a685896c 100644
@@ -48,7 +48,7 @@ struct srcu_struct {
 	unsigned long completed;
 	struct srcu_array __percpu *per_cpu_ref;
 	spinlock_t queue_lock; /* protect ->batch_queue, ->running */
-	bool running;
+	int srcu_state;
 	/* callbacks just queued */
 	struct rcu_batch batch_queue;
 	/* callbacks try to do the first check_zero */

@@ -62,6 +62,12 @@ struct srcu_struct {
 #endif /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
 };
 
+/* Values for -> state variable. */
+#define SRCU_STATE_IDLE		0
+#define SRCU_STATE_SCAN1	1
+#define SRCU_STATE_SCAN2	2
+#define SRCU_STATE_DONE		3
+
 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 
 int __init_srcu_struct(struct srcu_struct *sp, const char *name,

@@ -89,7 +95,7 @@ void process_srcu(struct work_struct *work);
 		.completed = -300,					\
 		.per_cpu_ref = &name##_srcu_array,			\
 		.queue_lock = __SPIN_LOCK_UNLOCKED(name.queue_lock),	\
-		.running = false,					\
+		.srcu_state = SRCU_STATE_IDLE,				\
 		.batch_queue = RCU_BATCH_INIT(name.batch_queue),	\
 		.batch_check0 = RCU_BATCH_INIT(name.batch_check0),	\
 		.batch_check1 = RCU_BATCH_INIT(name.batch_check1),	\

kernel/rcu/srcu.c +61 ~~-50~~

155 lines not on this page · raw message

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-12 17:46 UTC #35 lore raw

PATCH 25/40 srcu: Add grace-period sequence numbers

This commit adds grace-period sequence numbers, which will be used to handle mid-boot grace periods and per-CPU callback lists.

---
 include/linux/srcu.h |  1 +
 kernel/rcu/srcu.c    | 27 +++++++++++++++++++++++----
 2 files changed, 24 insertions(+), 4 deletions(-)

include/linux/srcu.h +1 -0

index f149a685896c..047ac8c28a4e 100644
@@ -46,6 +46,7 @@ struct rcu_batch {
 
 struct srcu_struct {
 	unsigned long completed;
+	unsigned long srcu_gp_seq;
 	struct srcu_array __percpu *per_cpu_ref;
 	spinlock_t queue_lock; /* protect ->batch_queue, ->running */
 	int srcu_state;

kernel/rcu/srcu.c +23 -4

index 84da09c02971..ed7c3d082b9f 100644
@@ -110,6 +110,7 @@ static inline void rcu_batch_move(struct rcu_batch *to, struct rcu_batch *from)
 static int init_srcu_struct_fields(struct srcu_struct *sp)
 {
 	sp->completed = 0;
+	sp->srcu_gp_seq = 0;
 	spin_lock_init(&sp->queue_lock);
 	sp->srcu_state = SRCU_STATE_IDLE;
 	rcu_batch_init(&sp->batch_queue);

@@ -313,6 +314,15 @@ EXPORT_SYMBOL_GPL(__srcu_read_unlock);
 #define SYNCHRONIZE_SRCU_EXP_TRYCOUNT	12
 
 /*
+ * Start an SRCU grace period.
+ */
+static void srcu_gp_start(struct srcu_struct *sp)
+{
+	WRITE_ONCE(sp->srcu_state, SRCU_STATE_SCAN1);
+	rcu_seq_start(&sp->srcu_gp_seq);
+}
+
+/*
  * @@@ Wait until all pre-existing readers complete.  Such readers
  * will have used the index specified by "idx".
  * the caller should ensures the ->completed is not changed while checking

@@ -349,6 +359,15 @@ static void srcu_flip(struct srcu_struct *sp)
 }
 
 /*
+ * End an SRCU grace period.
+ */
+static void srcu_gp_end(struct srcu_struct *sp)
+{
+	rcu_seq_end(&sp->srcu_gp_seq);
+	WRITE_ONCE(sp->srcu_state, SRCU_STATE_DONE);
+}
+
+/*
  * Enqueue an SRCU callback on the specified srcu_struct structure,
  * initiating grace-period processing if it is not already running.
  *

@@ -386,7 +405,7 @@ void call_srcu(struct srcu_struct *sp, struct rcu_head *head,
 	smp_mb__after_unlock_lock(); /* Caller's prior accesses before GP. */
 	rcu_batch_queue(&sp->batch_queue, head);
 	if (READ_ONCE(sp->srcu_state) == SRCU_STATE_IDLE) {
-		WRITE_ONCE(sp->srcu_state, SRCU_STATE_SCAN1);
+		srcu_gp_start(sp);
 		queue_delayed_work(system_power_efficient_wq, &sp->work, 0);
 	}
 	spin_unlock_irqrestore(&sp->queue_lock, flags);

@@ -420,7 +439,7 @@ static void __synchronize_srcu(struct srcu_struct *sp, int trycount)
 	smp_mb__after_unlock_lock(); /* Caller's prior accesses before GP. */
 	if (READ_ONCE(sp->srcu_state) == SRCU_STATE_IDLE) {
 		/* steal the processing owner */
-		WRITE_ONCE(sp->srcu_state, SRCU_STATE_SCAN1);
+		srcu_gp_start(sp);
 		rcu_batch_queue(&sp->batch_check0, head);
 		spin_unlock_irq(&sp->queue_lock);
 		/* give the processing owner to work_struct */

@@ -555,7 +574,7 @@ static void srcu_advance_batches(struct srcu_struct *sp, int trycount)
 	 */
 
 	if (sp->srcu_state == SRCU_STATE_DONE)
-		WRITE_ONCE(sp->srcu_state, SRCU_STATE_SCAN1);
+		srcu_gp_start(sp);
 
 	if (sp->srcu_state == SRCU_STATE_SCAN1) {
 		idx = 1 ^ (sp->completed & 1);

@@ -602,7 +621,7 @@ static void srcu_advance_batches(struct srcu_struct *sp, int trycount)
 		 */
 		rcu_batch_move(&sp->batch_done, &sp->batch_check1);
 
-		WRITE_ONCE(sp->srcu_state, SRCU_STATE_DONE);
+		srcu_gp_end(sp);
 	}
 }

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-12 17:41 UTC #36 lore raw

PATCH 26/40 srcu: Use rcu_segcblist to track SRCU callbacks

This commit switches SRCU from custom-built callback queues to the new rcu_segcblist structure. This change associates grace-period sequence numbers with groups of callbacks, which will be needed for efficient processing of per-CPU callbacks.

---
 include/linux/rcu_segcblist.h | 678 ++++++++++++++++++++++++++++++++++++++++++
 include/linux/srcu.h          |  24 +-
 kernel/rcu/rcu.h              |   6 +
 kernel/rcu/rcu_segcblist.h    | 671 -----------------------------------------
 kernel/rcu/srcu.c             | 159 ++--------
 kernel/rcu/tree.h             |   2 +-
 6 files changed, 721 insertions(+), 819 deletions(-)
 create mode 100644 include/linux/rcu_segcblist.h
 delete mode 100644 kernel/rcu/rcu_segcblist.h

include/linux/rcu_segcblist.h new file +678 -0

new file mode 100644
index 000000000000..74b1e7243955
@@ -0,0 +1,678 @@
+/*
+ * RCU segmented callback lists
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, you can access it online at
+ * http://www.gnu.org/licenses/gpl-2.0.html.
+ *
+ * Copyright IBM Corporation, 2017
+ *
+ * Authors: Paul E. McKenney <[email protected]>
+ */
+
+#ifndef __KERNEL_RCU_SEGCBLIST_H
+#define __KERNEL_RCU_SEGCBLIST_H
+
+/* Simple unsegmented callback lists. */
+struct rcu_cblist {
+	struct rcu_head *head;
+	struct rcu_head **tail;
+	long len;
+	long len_lazy;
+};
+
+#define RCU_CBLIST_INITIALIZER(n) { .head = NULL, .tail = &n.head }
+
+/* Initialize simple callback list. */
+static inline void rcu_cblist_init(struct rcu_cblist *rclp)
+{
+	rclp->head = NULL;
+	rclp->tail = &rclp->head;
+	rclp->len = 0;
+	rclp->len_lazy = 0;
+}
+
+/* Is simple callback list empty? */
+static inline bool rcu_cblist_empty(struct rcu_cblist *rclp)
+{
+	return !rclp->head;
+}
+
+/* Return number of callbacks in simple callback list. */
+static inline long rcu_cblist_n_cbs(struct rcu_cblist *rclp)
+{
+	return rclp->len;
+}
+
+/* Return number of lazy callbacks in simple callback list. */
+static inline long rcu_cblist_n_lazy_cbs(struct rcu_cblist *rclp)
+{
+	return rclp->len_lazy;
+}
+
+/*
+ * Debug function to actually count the number of callbacks.
+ * If the number exceeds the limit specified, return -1.
+ */
+static inline long rcu_cblist_count_cbs(struct rcu_cblist *rclp, long lim)
+{
+	int cnt = 0;
+	struct rcu_head **rhpp = &rclp->head;
+
+	for (;;) {
+		if (!*rhpp)
+			return cnt;
+		if (++cnt > lim)
+			return -1;
+		rhpp = &(*rhpp)->next;
+	}
+}
+
+/*
+ * Dequeue the oldest rcu_head structure from the specified callback
+ * list.  This function assumes that the callback is non-lazy, but
+ * the caller can later invoke rcu_cblist_dequeued_lazy() if it
+ * finds otherwise (and if it cares about laziness).  This allows
+ * different users to have different ways of determining laziness.
+ */
+static inline struct rcu_head *rcu_cblist_dequeue(struct rcu_cblist *rclp)
+{
+	struct rcu_head *rhp;
+
+	rhp = rclp->head;
+	if (!rhp)
+		return NULL;
+	rclp->len--;
+	rclp->head = rhp->next;
+	if (!rclp->head)
+		rclp->tail = &rclp->head;
+	return rhp;
+}
+
+/*
+ * Account for the fact that a previously dequeued callback turned out
+ * to be marked as lazy.
+ */
+static inline void rcu_cblist_dequeued_lazy(struct rcu_cblist *rclp)
+{
+	rclp->len_lazy--;
+}
+
+/*
+ * Interim function to return rcu_cblist head pointer.  Longer term, the
+ * rcu_cblist will be used more pervasively, removing the need for this
+ * function.
+ */
+static inline struct rcu_head *rcu_cblist_head(struct rcu_cblist *rclp)
+{
+	return rclp->head;
+}
+
+/*
+ * Interim function to return rcu_cblist head pointer.  Longer term, the
+ * rcu_cblist will be used more pervasively, removing the need for this
+ * function.
+ */
+static inline struct rcu_head **rcu_cblist_tail(struct rcu_cblist *rclp)
+{
+	WARN_ON_ONCE(rcu_cblist_empty(rclp));
+	return rclp->tail;
+}
+
+/* Complicated segmented callback lists.  ;-) */
+
+/*
+ * Index values for segments in rcu_segcblist structure.
+ *
+ * The segments are as follows:
+ *
+ * [head, *tails[RCU_DONE_TAIL]):
+ *	Callbacks whose grace period has elapsed, and thus can be invoked.
+ * [*tails[RCU_DONE_TAIL], *tails[RCU_WAIT_TAIL]):
+ *	Callbacks waiting for the current GP from the current CPU's viewpoint.
+ * [*tails[RCU_WAIT_TAIL], *tails[RCU_NEXT_READY_TAIL]):
+ *	Callbacks that arrived before the next GP started, again from
+ *	the current CPU's viewpoint.  These can be handled by the next GP.
+ * [*tails[RCU_NEXT_READY_TAIL], *tails[RCU_NEXT_TAIL]):
+ *	Callbacks that might have arrived after the next GP started.
+ *	There is some uncertainty as to when a given GP starts and
+ *	ends, but a CPU knows the exact times if it is the one starting
+ *	or ending the GP.  Other CPUs know that the previous GP ends
+ *	before the next one starts.
+ *
+ * Note that RCU_WAIT_TAIL cannot be empty unless RCU_NEXT_READY_TAIL is also
+ * empty.
+ *
+ * The ->gp_seq[] array contains the grace-period number at which the
+ * corresponding segment of callbacks will be ready to invoke.  A given
+ * element of this array is meaningful only when the corresponding segment
+ * is non-empty, and it is never valid for RCU_DONE_TAIL (whose callbacks
+ * are already ready to invoke) or for RCU_NEXT_TAIL (whose callbacks have
+ * not yet been assigned a grace-period number).
+ */
+#define RCU_DONE_TAIL		0	/* Also RCU_WAIT head. */
+#define RCU_WAIT_TAIL		1	/* Also RCU_NEXT_READY head. */
+#define RCU_NEXT_READY_TAIL	2	/* Also RCU_NEXT head. */
+#define RCU_NEXT_TAIL		3
+#define RCU_CBLIST_NSEGS	4
+
+struct rcu_segcblist {
+	struct rcu_head *head;
+	struct rcu_head **tails[RCU_CBLIST_NSEGS];
+	unsigned long gp_seq[RCU_CBLIST_NSEGS];
+	long len;
+	long len_lazy;
+};
+
+#define RCU_SEGCBLIST_INITIALIZER(n) \
+{ \
+	.head = NULL, \
+	.tails[RCU_DONE_TAIL] = &n.head, \
+	.tails[RCU_WAIT_TAIL] = &n.head, \
+	.tails[RCU_NEXT_READY_TAIL] = &n.head, \
+	.tails[RCU_NEXT_TAIL] = &n.head, \
+}
+
+/*
+ * Initialize an rcu_segcblist structure.
+ */
+static inline void rcu_segcblist_init(struct rcu_segcblist *rsclp)
+{
+	int i;
+
+	BUILD_BUG_ON(RCU_NEXT_TAIL + 1 != ARRAY_SIZE(rsclp->gp_seq));
+	BUILD_BUG_ON(ARRAY_SIZE(rsclp->tails) != ARRAY_SIZE(rsclp->gp_seq));
+	rsclp->head = NULL;
+	for (i = 0; i < RCU_CBLIST_NSEGS; i++)
+		rsclp->tails[i] = &rsclp->head;
+	rsclp->len = 0;
+	rsclp->len_lazy = 0;
+}
+
+/*
+ * Is the specified rcu_segcblist structure empty?
+ *
+ * But careful!  The fact that the ->head field is NULL does not
+ * necessarily imply that there are no callbacks associated with
+ * this structure.  When callbacks are being invoked, they are
+ * removed as a group.  If callback invocation must be preempted,
+ * the remaining callbacks will be added back to the list.  Either
+ * way, the counts are updated later.
+ *
+ * So it is often the case that rcu_segcblist_n_cbs() should be used
+ * instead.
+ */
+static inline bool rcu_segcblist_empty(struct rcu_segcblist *rsclp)
+{
+	return !rsclp->head;
+}
+
+/* Return number of callbacks in segmented callback list. */
+static inline long rcu_segcblist_n_cbs(struct rcu_segcblist *rsclp)
+{
+	return READ_ONCE(rsclp->len);
+}
+
+/* Return number of lazy callbacks in segmented callback list. */
+static inline long rcu_segcblist_n_lazy_cbs(struct rcu_segcblist *rsclp)
+{
+	return rsclp->len_lazy;
+}
+
+/* Return number of lazy callbacks in segmented callback list. */
+static inline long rcu_segcblist_n_nonlazy_cbs(struct rcu_segcblist *rsclp)
+{
+	return rsclp->len - rsclp->len_lazy;
+}
+
+/*
+ * Is the specified rcu_segcblist enabled, for example, not corresponding
+ * to an offline or callback-offloaded CPU?
+ */
+static inline bool rcu_segcblist_is_enabled(struct rcu_segcblist *rsclp)
+{
+	return !!rsclp->tails[RCU_NEXT_TAIL];
+}
+
+/*
+ * Disable the specified rcu_segcblist structure, so that callbacks can
+ * no longer be posted to it.  This structure must be empty.
+ */
+static inline void rcu_segcblist_disable(struct rcu_segcblist *rsclp)
+{
+	WARN_ON_ONCE(!rcu_segcblist_empty(rsclp));
+	WARN_ON_ONCE(rcu_segcblist_n_cbs(rsclp));
+	WARN_ON_ONCE(rcu_segcblist_n_lazy_cbs(rsclp));
+	rsclp->tails[RCU_NEXT_TAIL] = NULL;
+}
+
+/*
+ * Is the specified segment of the specified rcu_segcblist structure
+ * empty of callbacks?
+ */
+static inline bool rcu_segcblist_segempty(struct rcu_segcblist *rsclp, int seg)
+{
+	if (seg == RCU_DONE_TAIL)
+		return &rsclp->head == rsclp->tails[RCU_DONE_TAIL];
+	return rsclp->tails[seg - 1] == rsclp->tails[seg];
+}
+
+/*
+ * Are all segments following the specified segment of the specified
+ * rcu_segcblist structure empty of callbacks?  (The specified
+ * segment might well contain callbacks.)
+ */
+static inline bool rcu_segcblist_restempty(struct rcu_segcblist *rsclp, int seg)
+{
+	return !*rsclp->tails[seg];
+}
+
+/*
+ * Does the specified rcu_segcblist structure contain callbacks that
+ * are ready to be invoked?
+ */
+static inline bool rcu_segcblist_ready_cbs(struct rcu_segcblist *rsclp)
+{
+	return rcu_segcblist_is_enabled(rsclp) &&
+	       &rsclp->head != rsclp->tails[RCU_DONE_TAIL];
+}
+
+/*
+ * Does the specified rcu_segcblist structure contain callbacks that
+ * are still pending, that is, not yet ready to be invoked?
+ */
+static inline bool rcu_segcblist_pend_cbs(struct rcu_segcblist *rsclp)
+{
+	return rcu_segcblist_is_enabled(rsclp) &&
+	       !rcu_segcblist_restempty(rsclp, RCU_DONE_TAIL);
+}
+
+/*
+ * Dequeue and return the first ready-to-invoke callback.  If there
+ * are no ready-to-invoke callbacks, return NULL.  Disables interrupts
+ * to avoid interference.  Does not protect from interference from other
+ * CPUs or tasks.
+ */
+static inline struct rcu_head *
+rcu_segcblist_dequeue(struct rcu_segcblist *rsclp)
+{
+	unsigned long flags;
+	int i;
+	struct rcu_head *rhp;
+
+	local_irq_save(flags);
+	if (!rcu_segcblist_ready_cbs(rsclp)) {
+		local_irq_restore(flags);
+		return NULL;
+	}
+	rhp = rsclp->head;
+	BUG_ON(!rhp);
+	rsclp->head = rhp->next;
+	for (i = RCU_DONE_TAIL; i < RCU_CBLIST_NSEGS; i++) {
+		if (rsclp->tails[i] != &rhp->next)
+			break;
+		rsclp->tails[i] = &rsclp->head;
+	}
+	smp_mb(); /* Dequeue before decrement for rcu_barrier(). */
+	WRITE_ONCE(rsclp->len, rsclp->len - 1);
+	local_irq_restore(flags);
+	return rhp;
+}
+
+/*
+ * Account for the fact that a previously dequeued callback turned out
+ * to be marked as lazy.
+ */
+static inline void rcu_segcblist_dequeued_lazy(struct rcu_segcblist *rsclp)
+{
+	unsigned long flags;
+
+	local_irq_save(flags);
+	rsclp->len_lazy--;
+	local_irq_restore(flags);
+}
+
+/*
+ * Return a pointer to the first callback in the specified rcu_segcblist
+ * structure.  This is useful for diagnostics.
+ */
+static inline struct rcu_head *
+rcu_segcblist_first_cb(struct rcu_segcblist *rsclp)
+{
+	if (rcu_segcblist_is_enabled(rsclp))
+		return rsclp->head;
+	return NULL;
+}
+
+/*
+ * Return a pointer to the first pending callback in the specified
+ * rcu_segcblist structure.  This is useful just after posting a given
+ * callback -- if that callback is the first pending callback, then
+ * you cannot rely on someone else having already started up the required
+ * grace period.
+ */
+static inline struct rcu_head *
+rcu_segcblist_first_pend_cb(struct rcu_segcblist *rsclp)
+{
+	if (rcu_segcblist_is_enabled(rsclp))
+		return *rsclp->tails[RCU_DONE_TAIL];
+	return NULL;
+}
+
+/*
+ * Does the specified rcu_segcblist structure contain callbacks that
+ * have not yet been processed beyond having been posted, that is,
+ * does it contain callbacks in its last segment?
+ */
+static inline bool rcu_segcblist_new_cbs(struct rcu_segcblist *rsclp)
+{
+	return rcu_segcblist_is_enabled(rsclp) &&
+	       !rcu_segcblist_restempty(rsclp, RCU_NEXT_READY_TAIL);
+}
+
+/*
+ * Enqueue the specified callback onto the specified rcu_segcblist
+ * structure, updating accounting as needed.  Note that the ->len
+ * field may be accessed locklessly, hence the WRITE_ONCE().
+ * The ->len field is used by rcu_barrier() and friends to determine
+ * if it must post a callback on this structure, and it is OK
+ * for rcu_barrier() to sometimes post callbacks needlessly, but
+ * absolutely not OK for it to ever miss posting a callback.
+ */
+static inline void rcu_segcblist_enqueue(struct rcu_segcblist *rsclp,
+					 struct rcu_head *rhp, bool lazy)
+{
+	WRITE_ONCE(rsclp->len, rsclp->len + 1); /* ->len sampled locklessly. */
+	if (lazy)
+		rsclp->len_lazy++;
+	smp_mb(); /* Ensure counts are updated before callback is enqueued. */
+	rhp->next = NULL;
+	*rsclp->tails[RCU_NEXT_TAIL] = rhp;
+	rsclp->tails[RCU_NEXT_TAIL] = &rhp->next;
+}
+
+/*
+ * Extract only the counts from the specified rcu_segcblist structure,
+ * and place them in the specified rcu_cblist structure.  This function
+ * supports both callback orphaning and invocation, hence the separation
+ * of counts and callbacks.  (Callbacks ready for invocation must be
+ * orphaned and adopted separately from pending callbacks, but counts
+ * apply to all callbacks.  Locking must be used to make sure that
+ * both orphaned-callbacks lists are consistent.)
+ */
+static inline void rcu_segcblist_extract_count(struct rcu_segcblist *rsclp,
+					       struct rcu_cblist *rclp)
+{
+	rclp->len_lazy += rsclp->len_lazy;
+	rclp->len += rsclp->len;
+	rsclp->len_lazy = 0;
+	WRITE_ONCE(rsclp->len, 0); /* ->len sampled locklessly. */
+}
+
+/*
+ * Extract only those callbacks ready to be invoked from the specified
+ * rcu_segcblist structure and place them in the specified rcu_cblist
+ * structure.
+ */
+static inline void rcu_segcblist_extract_done_cbs(struct rcu_segcblist *rsclp,
+						  struct rcu_cblist *rclp)
+{
+	int i;
+
+	if (!rcu_segcblist_ready_cbs(rsclp))
+		return; /* Nothing to do. */
+	*rclp->tail = rsclp->head;
+	rsclp->head = *rsclp->tails[RCU_DONE_TAIL];
+	*rsclp->tails[RCU_DONE_TAIL] = NULL;
+	rclp->tail = rsclp->tails[RCU_DONE_TAIL];
+	for (i = RCU_CBLIST_NSEGS - 1; i >= RCU_DONE_TAIL; i--)
+		if (rsclp->tails[i] == rsclp->tails[RCU_DONE_TAIL])
+			rsclp->tails[i] = &rsclp->head;
+}
+
+/*
+ * Extract only those callbacks still pending (not yet ready to be
+ * invoked) from the specified rcu_segcblist structure and place them in
+ * the specified rcu_cblist structure.  Note that this loses information
+ * about any callbacks that might have been partway done waiting for
+ * their grace period.  Too bad!  They will have to start over.
+ */
+static inline void
+rcu_segcblist_extract_pend_cbs(struct rcu_segcblist *rsclp,
+			       struct rcu_cblist *rclp)
+{
+	int i;
+
+	if (!rcu_segcblist_pend_cbs(rsclp))
+		return; /* Nothing to do. */
+	*rclp->tail = *rsclp->tails[RCU_DONE_TAIL];
+	rclp->tail = rsclp->tails[RCU_NEXT_TAIL];
+	*rsclp->tails[RCU_DONE_TAIL] = NULL;
+	for (i = RCU_DONE_TAIL + 1; i < RCU_CBLIST_NSEGS; i++)
+		rsclp->tails[i] = rsclp->tails[RCU_DONE_TAIL];
+}
+
+/*
+ * Move the entire contents of the specified rcu_segcblist structure,
+ * counts, callbacks, and all, to the specified rcu_cblist structure.
+ * @@@ Why do we need this???  Moving early-boot CBs to NOCB lists?
+ * @@@ Memory barrier needed?  (Not if only used at boot time...)
+ */
+static inline void rcu_segcblist_extract_all(struct rcu_segcblist *rsclp,
+					     struct rcu_cblist *rclp)
+{
+	rcu_segcblist_extract_done_cbs(rsclp, rclp);
+	rcu_segcblist_extract_pend_cbs(rsclp, rclp);
+	rcu_segcblist_extract_count(rsclp, rclp);
+}
+
+/*
+ * Insert counts from the specified rcu_cblist structure in the
+ * specified rcu_segcblist structure.
+ */
+static inline void rcu_segcblist_insert_count(struct rcu_segcblist *rsclp,
+					      struct rcu_cblist *rclp)
+{
+	rsclp->len_lazy += rclp->len_lazy;
+	/* ->len sampled locklessly. */
+	WRITE_ONCE(rsclp->len, rsclp->len + rclp->len);
+	rclp->len_lazy = 0;
+	rclp->len = 0;
+}
+
+/*
+ * Move callbacks from the specified rcu_cblist to the beginning of the
+ * done-callbacks segment of the specified rcu_segcblist.
+ */
+static inline void rcu_segcblist_insert_done_cbs(struct rcu_segcblist *rsclp,
+						 struct rcu_cblist *rclp)
+{
+	int i;
+
+	if (!rclp->head)
+		return; /* No callbacks to move. */
+	*rclp->tail = rsclp->head;
+	rsclp->head = rclp->head;
+	for (i = RCU_DONE_TAIL; i < RCU_CBLIST_NSEGS; i++)
+		if (&rsclp->head == rsclp->tails[i])
+			rsclp->tails[i] = rclp->tail;
+		else
+			break;
+	rclp->head = NULL;
+	rclp->tail = &rclp->head;
+}
+
+/*
+ * Move callbacks from the specified rcu_cblist to the end of the
+ * new-callbacks segment of the specified rcu_segcblist.
+ */
+static inline void rcu_segcblist_insert_pend_cbs(struct rcu_segcblist *rsclp,
+						 struct rcu_cblist *rclp)
+{
+	if (!rclp->head)
+		return; /* Nothing to do. */
+	*rsclp->tails[RCU_NEXT_TAIL] = rclp->head;
+	rsclp->tails[RCU_NEXT_TAIL] = rclp->tail;
+	rclp->head = NULL;
+	rclp->tail = &rclp->head;
+}
+
+/*
+ * Advance the callbacks in the specified rcu_segcblist structure based
+ * on the current value passed in for the grace-period counter.
+ */
+static inline void rcu_segcblist_advance(struct rcu_segcblist *rsclp,
+					 unsigned long seq)
+{
+	int i, j;
+
+	WARN_ON_ONCE(!rcu_segcblist_is_enabled(rsclp));
+	WARN_ON_ONCE(rcu_segcblist_restempty(rsclp, RCU_DONE_TAIL));
+
+	/*
+	 * Find all callbacks whose ->gp_seq numbers indicate that they
+	 * are ready to invoke, and put them into the RCU_DONE_TAIL segment.
+	 */
+	for (i = RCU_WAIT_TAIL; i < RCU_NEXT_TAIL; i++) {
+		if (ULONG_CMP_LT(seq, rsclp->gp_seq[i]))
+			break;
+		rsclp->tails[RCU_DONE_TAIL] = rsclp->tails[i];
+	}
+
+	/* If no callbacks moved, nothing more need be done. */
+	if (i == RCU_WAIT_TAIL)
+		return;
+
+	/* Clean up tail pointers that might have been misordered above. */
+	for (j = RCU_WAIT_TAIL; j < i; j++)
+		rsclp->tails[j] = rsclp->tails[RCU_DONE_TAIL];
+
+	/*
+	 * Callbacks moved, so clean up the misordered ->tails[] pointers
+	 * that now point into the middle of the list of ready-to-invoke
+	 * callbacks.  The overall effect is to copy down the later pointers
+	 * into the gap that was created by the now-ready segments.
+	 */
+	for (j = RCU_WAIT_TAIL; i < RCU_NEXT_TAIL; i++, j++) {
+		if (rsclp->tails[j] == rsclp->tails[RCU_NEXT_TAIL])
+			break;  /* No more callbacks. */
+		rsclp->tails[j] = rsclp->tails[i];
+		rsclp->gp_seq[j] = rsclp->gp_seq[i];
+	}
+}
+
+/*
+ * "Accelerate" callbacks based on more-accurate grace-period information.
+ * The reason for this is that RCU does not synchronize the beginnings and
+ * ends of grace periods, and that callbacks are posted locally.  This in
+ * turn means that the callbacks must be labelled conservatively early
+ * on, as getting exact information would degrade both performance and
+ * scalability.  When more accurate grace-period information becomes
+ * available, previously posted callbacks can be "accelerated", marking
+ * them to complete at the end of the earlier grace period.
+ *
+ * This function operates on an rcu_segcblist structure, and also the
+ * grace-period sequence number at which new callbacks would become
+ * ready to invoke.
+ */
+static inline bool rcu_segcblist_accelerate(struct rcu_segcblist *rsclp,
+					    unsigned long seq)
+{
+	int i;
+
+	WARN_ON_ONCE(!rcu_segcblist_is_enabled(rsclp));
+	WARN_ON_ONCE(rcu_segcblist_restempty(rsclp, RCU_DONE_TAIL));
+
+	/*
+	 * Find the segment preceding the oldest segment of callbacks
+	 * whose ->gp_seq[] completion is at or after that passed in via
+	 * "seq", skipping any empty segments.  This oldest segment, along
+	 * with any later segments, can be merged in with any newly arrived
+	 * callbacks in the RCU_NEXT_TAIL segment, and assigned "seq"
+	 * as their ->gp_seq[] grace-period completion sequence number.
+	 */
+	for (i = RCU_NEXT_READY_TAIL; i > RCU_DONE_TAIL; i--)
+		if (rsclp->tails[i] != rsclp->tails[i - 1] &&
+		    ULONG_CMP_LT(rsclp->gp_seq[i], seq))
+			break;
+
+	/*
+	 * If all the segments contain callbacks that correspond to
+	 * earlier grace-period sequence numbers than "seq", leave.
+	 * Assuming that the rcu_segcblist structure has enough
+	 * segments in its arrays, this can only happen if some of
+	 * the non-done segments contain callbacks that really are
+	 * ready to invoke.  This situation will get straightened
+	 * out by the next call to rcu_segcblist_advance().
+	 *
+	 * Also advance to the oldest segment of callbacks whose
+	 * ->gp_seq[] completion is at or after that passed in via "seq",
+	 * skipping any empty segments.
+	 */
+	if (++i >= RCU_NEXT_TAIL)
+		return false;
+
+	/*
+	 * Merge all later callbacks, including newly arrived callbacks,
+	 * into the segment located by the for-loop above.  Assign "seq"
+	 * as the ->gp_seq[] value in order to correctly handle the case
+	 * where there were no pending callbacks in the rcu_segcblist
+	 * structure other than in the RCU_NEXT_TAIL segment.
+	 */
+	for (; i < RCU_NEXT_TAIL; i++) {
+		rsclp->tails[i] = rsclp->tails[RCU_NEXT_TAIL];
+		rsclp->gp_seq[i] = seq;
+	}
+	return true;
+}
+
+/*
+ * Scan the specified rcu_segcblist structure for callbacks that need
+ * a grace period later than the one specified by "seq".  We don't look
+ * at the RCU_DONE_TAIL or RCU_NEXT_TAIL segments because they don't
+ * have a grace-period sequence number.
+ */
+static inline bool rcu_segcblist_future_gp_needed(struct rcu_segcblist *rsclp,
+						  unsigned long seq)
+{
+	int i;
+
+	for (i = RCU_WAIT_TAIL; i < RCU_NEXT_TAIL; i++)
+		if (rsclp->tails[i - 1] != rsclp->tails[i] &&
+		    ULONG_CMP_LT(seq, rsclp->gp_seq[i]))
+			return true;
+	return false;
+}
+
+/*
+ * Interim function to return rcu_segcblist head pointer.  Longer term, the
+ * rcu_segcblist will be used more pervasively, removing the need for this
+ * function.
+ */
+static inline struct rcu_head *rcu_segcblist_head(struct rcu_segcblist *rsclp)
+{
+	return rsclp->head;
+}
+
+/*
+ * Interim function to return rcu_segcblist head pointer.  Longer term, the
+ * rcu_segcblist will be used more pervasively, removing the need for this
+ * function.
+ */
+static inline struct rcu_head **rcu_segcblist_tail(struct rcu_segcblist *rsclp)
+{
+	WARN_ON_ONCE(rcu_segcblist_empty(rsclp));
+	return rsclp->tails[RCU_NEXT_TAIL];
+}
+
+#endif /* __KERNEL_RCU_SEGCBLIST_H */

include/linux/srcu.h +5 ~~-19~~

index 047ac8c28a4e..ad154a7bc114 100644
@@ -22,7 +22,7 @@
  *	   Lai Jiangshan <[email protected]>
  *
  * For detailed explanation of Read-Copy Update mechanism see -
- * 		Documentation/RCU/ *.txt
+ *		Documentation/RCU/ *.txt
  *
  */

@@ -32,31 +32,20 @@
 #include <linux/mutex.h>
 #include <linux/rcupdate.h>
 #include <linux/workqueue.h>
+#include <linux/rcu_segcblist.h>
 
 struct srcu_array {
 	unsigned long lock_count[2];
 	unsigned long unlock_count[2];
 };
 
-struct rcu_batch {
-	struct rcu_head *head, **tail;
-};
-
-#define RCU_BATCH_INIT(name) { NULL, &(name.head) }
-
 struct srcu_struct {
 	unsigned long completed;
 	unsigned long srcu_gp_seq;
 	struct srcu_array __percpu *per_cpu_ref;
-	spinlock_t queue_lock; /* protect ->batch_queue, ->running */
+	spinlock_t queue_lock; /* protect ->srcu_cblist, ->srcu_state */
 	int srcu_state;
-	/* callbacks just queued */
-	struct rcu_batch batch_queue;
-	/* callbacks try to do the first check_zero */
-	struct rcu_batch batch_check0;
-	/* callbacks done with the first check_zero and the flip */
-	struct rcu_batch batch_check1;
-	struct rcu_batch batch_done;
+	struct rcu_segcblist srcu_cblist;
 	struct delayed_work work;
 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 	struct lockdep_map dep_map;

@@ -97,10 +86,7 @@ void process_srcu(struct work_struct *work);
 		.per_cpu_ref = &name##_srcu_array,			\
 		.queue_lock = __SPIN_LOCK_UNLOCKED(name.queue_lock),	\
 		.srcu_state = SRCU_STATE_IDLE,				\
-		.batch_queue = RCU_BATCH_INIT(name.batch_queue),	\
-		.batch_check0 = RCU_BATCH_INIT(name.batch_check0),	\
-		.batch_check1 = RCU_BATCH_INIT(name.batch_check1),	\
-		.batch_done = RCU_BATCH_INIT(name.batch_done),		\
+		.srcu_cblist = RCU_SEGCBLIST_INITIALIZER(name.srcu_cblist),\
 		.work = __DELAYED_WORK_INITIALIZER(name.work, process_srcu, 0),\
 		__SRCU_DEP_MAP_INIT(name)				\
 	}

kernel/rcu/rcu.h +6 -0

index 0bc1313c49e2..a943b42a9cf7 100644
@@ -87,6 +87,12 @@ static inline unsigned long rcu_seq_snap(unsigned long *sp)
 	return s;
 }
 
+/* Return the current value the update side's sequence number, no ordering. */
+static inline unsigned long rcu_seq_current(unsigned long *sp)
+{
+	return READ_ONCE(*sp);
+}
+
 /*
  * Given a snapshot from rcu_seq_snap(), determine whether or not a
  * full update-side operation has occurred.

kernel/rcu/rcu_segcblist.h deleted +0 ~~-671~~

671 lines not on this page · raw message

kernel/rcu/srcu.c +31 ~~-128~~

254 lines not on this page · raw message

kernel/rcu/tree.h +1 -1

index 93889ff21dbb..4f62651588ea 100644
@@ -30,7 +30,7 @@
 #include <linux/seqlock.h>
 #include <linux/swait.h>
 #include <linux/stop_machine.h>
-#include "rcu_segcblist.h"
+#include <linux/rcu_segcblist.h>
 
 /*
  * Define shape of hierarchy based on NR_CPUS, CONFIG_RCU_FANOUT, and

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-12 17:43 UTC #37 lore raw

PATCH 27/40 srcu: Move combining-tree definitions for SRCU's benefit

This commit moves the C preprocessor code that defines the default shape of the rcu_node combining tree to a new include/linux/rcu_node_tree.h file as a first step towards enabling SRCU to create its own combining tree, which in turn enables SRCU to implement per-CPU callback handling, thus avoiding contention on the lock currently guarding the single list of callbacks. Note that users of SRCU still need to know the size of the srcu_struct structure, hence include/linux rather than kernel/rcu.

This commit is code-movement only.

---
 include/linux/rcu_node_tree.h | 102 ++++++++++++++++++++++++++++++++++++++++++
 kernel/rcu/tree.h             |  71 +----------------------------
 2 files changed, 103 insertions(+), 70 deletions(-)
 create mode 100644 include/linux/rcu_node_tree.h

include/linux/rcu_node_tree.h new file +102 -0

new file mode 100644
index 000000000000..b7eb97096b1c
@@ -0,0 +1,102 @@
+/*
+ * RCU node combining tree definitions.  These are used to compute
+ * global attributes while avoiding common-case global contention.  A key
+ * property that these computations rely on is a tournament-style approach
+ * where only one of the tasks contending a lower level in the tree need
+ * advance to the next higher level.  If properly configured, this allows
+ * unlimited scalability while maintaining a constant level of contention
+ * on the root node.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, you can access it online at
+ * http://www.gnu.org/licenses/gpl-2.0.html.
+ *
+ * Copyright IBM Corporation, 2017
+ *
+ * Author: Paul E. McKenney <[email protected]>
+ */
+
+#ifndef __LINUX_RCU_NODE_TREE_H
+#define __LINUX_RCU_NODE_TREE_H
+
+/*
+ * Define shape of hierarchy based on NR_CPUS, CONFIG_RCU_FANOUT, and
+ * CONFIG_RCU_FANOUT_LEAF.
+ * In theory, it should be possible to add more levels straightforwardly.
+ * In practice, this did work well going from three levels to four.
+ * Of course, your mileage may vary.
+ */
+
+#ifdef CONFIG_RCU_FANOUT
+#define RCU_FANOUT CONFIG_RCU_FANOUT
+#else /* #ifdef CONFIG_RCU_FANOUT */
+# ifdef CONFIG_64BIT
+# define RCU_FANOUT 64
+# else
+# define RCU_FANOUT 32
+# endif
+#endif /* #else #ifdef CONFIG_RCU_FANOUT */
+
+#ifdef CONFIG_RCU_FANOUT_LEAF
+#define RCU_FANOUT_LEAF CONFIG_RCU_FANOUT_LEAF
+#else /* #ifdef CONFIG_RCU_FANOUT_LEAF */
+#define RCU_FANOUT_LEAF 16
+#endif /* #else #ifdef CONFIG_RCU_FANOUT_LEAF */
+
+#define RCU_FANOUT_1	      (RCU_FANOUT_LEAF)
+#define RCU_FANOUT_2	      (RCU_FANOUT_1 * RCU_FANOUT)
+#define RCU_FANOUT_3	      (RCU_FANOUT_2 * RCU_FANOUT)
+#define RCU_FANOUT_4	      (RCU_FANOUT_3 * RCU_FANOUT)
+
+#if NR_CPUS <= RCU_FANOUT_1
+#  define RCU_NUM_LVLS	      1
+#  define NUM_RCU_LVL_0	      1
+#  define NUM_RCU_NODES	      NUM_RCU_LVL_0
+#  define NUM_RCU_LVL_INIT    { NUM_RCU_LVL_0 }
+#  define RCU_NODE_NAME_INIT  { "rcu_node_0" }
+#  define RCU_FQS_NAME_INIT   { "rcu_node_fqs_0" }
+#elif NR_CPUS <= RCU_FANOUT_2
+#  define RCU_NUM_LVLS	      2
+#  define NUM_RCU_LVL_0	      1
+#  define NUM_RCU_LVL_1	      DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
+#  define NUM_RCU_NODES	      (NUM_RCU_LVL_0 + NUM_RCU_LVL_1)
+#  define NUM_RCU_LVL_INIT    { NUM_RCU_LVL_0, NUM_RCU_LVL_1 }
+#  define RCU_NODE_NAME_INIT  { "rcu_node_0", "rcu_node_1" }
+#  define RCU_FQS_NAME_INIT   { "rcu_node_fqs_0", "rcu_node_fqs_1" }
+#elif NR_CPUS <= RCU_FANOUT_3
+#  define RCU_NUM_LVLS	      3
+#  define NUM_RCU_LVL_0	      1
+#  define NUM_RCU_LVL_1	      DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_2)
+#  define NUM_RCU_LVL_2	      DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
+#  define NUM_RCU_NODES	      (NUM_RCU_LVL_0 + NUM_RCU_LVL_1 + NUM_RCU_LVL_2)
+#  define NUM_RCU_LVL_INIT    { NUM_RCU_LVL_0, NUM_RCU_LVL_1, NUM_RCU_LVL_2 }
+#  define RCU_NODE_NAME_INIT  { "rcu_node_0", "rcu_node_1", "rcu_node_2" }
+#  define RCU_FQS_NAME_INIT   { "rcu_node_fqs_0", "rcu_node_fqs_1", "rcu_node_fqs_2" }
+#elif NR_CPUS <= RCU_FANOUT_4
+#  define RCU_NUM_LVLS	      4
+#  define NUM_RCU_LVL_0	      1
+#  define NUM_RCU_LVL_1	      DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_3)
+#  define NUM_RCU_LVL_2	      DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_2)
+#  define NUM_RCU_LVL_3	      DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
+#  define NUM_RCU_NODES	      (NUM_RCU_LVL_0 + NUM_RCU_LVL_1 + NUM_RCU_LVL_2 + NUM_RCU_LVL_3)
+#  define NUM_RCU_LVL_INIT    { NUM_RCU_LVL_0, NUM_RCU_LVL_1, NUM_RCU_LVL_2, NUM_RCU_LVL_3 }
+#  define RCU_NODE_NAME_INIT  { "rcu_node_0", "rcu_node_1", "rcu_node_2", "rcu_node_3" }
+#  define RCU_FQS_NAME_INIT   { "rcu_node_fqs_0", "rcu_node_fqs_1", "rcu_node_fqs_2", "rcu_node_fqs_3" }
+#else
+# error "CONFIG_RCU_FANOUT insufficient for NR_CPUS"
+#endif /* #if (NR_CPUS) <= RCU_FANOUT_1 */
+
+extern int rcu_num_lvls;
+extern int rcu_num_nodes;
+
+#endif /* __LINUX_RCU_NODE_TREE_H */

kernel/rcu/tree.h +1 ~~-70~~

77 lines not on this page · raw message

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-12 17:43 UTC #38 lore raw

PATCH 28/40 srcu: Move rcu_init_levelspread() to rcu_tree_node.h

This commit moves the rcu_init_levelspread() function from kernel/rcu/tree.c to kernel/rcu/rcu.h so that SRCU can access it. This is another step towards enabling SRCU to create its own combining tree. This commit is code-movement only, give or take knock-on adjustments.

---
 include/linux/rcu_node_tree.h |  3 ---
 kernel/rcu/rcu.h              | 36 ++++++++++++++++++++++++++++++++++++
 kernel/rcu/srcu.c             |  1 +
 kernel/rcu/tree.c             | 25 -------------------------
 kernel/rcu/tree_trace.c       |  1 +
 5 files changed, 38 insertions(+), 28 deletions(-)

include/linux/rcu_node_tree.h +0 -3

index b7eb97096b1c..4b766b61e1a0 100644
@@ -96,7 +96,4 @@
 # error "CONFIG_RCU_FANOUT insufficient for NR_CPUS"
 #endif /* #if (NR_CPUS) <= RCU_FANOUT_1 */
 
-extern int rcu_num_lvls;
-extern int rcu_num_nodes;
-
 #endif /* __LINUX_RCU_NODE_TREE_H */

kernel/rcu/rcu.h +36 -0

index a943b42a9cf7..87326479b39a 100644
@@ -190,4 +190,40 @@ void rcu_test_sync_prims(void);
  */
 extern void resched_cpu(int cpu);
 
+#if defined(SRCU) || !defined(TINY_RCU)
+
+#include <linux/rcu_node_tree.h>
+
+extern int rcu_num_lvls;
+extern int rcu_num_nodes;
+static bool rcu_fanout_exact;
+static int rcu_fanout_leaf;
+
+/*
+ * Compute the per-level fanout, either using the exact fanout specified
+ * or balancing the tree, depending on the rcu_fanout_exact boot parameter.
+ */
+static inline void rcu_init_levelspread(int *levelspread, const int *levelcnt)
+{
+	int i;
+
+	if (rcu_fanout_exact) {
+		levelspread[rcu_num_lvls - 1] = rcu_fanout_leaf;
+		for (i = rcu_num_lvls - 2; i >= 0; i--)
+			levelspread[i] = RCU_FANOUT;
+	} else {
+		int ccur;
+		int cprv;
+
+		cprv = nr_cpu_ids;
+		for (i = rcu_num_lvls - 1; i >= 0; i--) {
+			ccur = levelcnt[i];
+			levelspread[i] = (cprv + ccur - 1) / ccur;
+			cprv = ccur;
+		}
+	}
+}
+
+#endif /* #if defined(SRCU) || !defined(TINY_RCU) */
+
 #endif /* __LINUX_RCU_H */

kernel/rcu/srcu.c +1 -0

index 132cd57c07f1..d51ab050f777 100644
@@ -36,6 +36,7 @@
 #include <linux/delay.h>
 #include <linux/srcu.h>
 
+#include <linux/rcu_node_tree.h>
 #include "rcu.h"
 
 static int init_srcu_struct_fields(struct srcu_struct *sp)

kernel/rcu/tree.c +0 ~~-25~~

index f8b100902e21..c5a9674824ef 100644
@@ -3959,31 +3959,6 @@ void rcu_scheduler_starting(void)
 }
 
 /*
- * Compute the per-level fanout, either using the exact fanout specified
- * or balancing the tree, depending on the rcu_fanout_exact boot parameter.
- */
-static void __init rcu_init_levelspread(int *levelspread, const int *levelcnt)
-{
-	int i;
-
-	if (rcu_fanout_exact) {
-		levelspread[rcu_num_lvls - 1] = rcu_fanout_leaf;
-		for (i = rcu_num_lvls - 2; i >= 0; i--)
-			levelspread[i] = RCU_FANOUT;
-	} else {
-		int ccur;
-		int cprv;
-
-		cprv = nr_cpu_ids;
-		for (i = rcu_num_lvls - 1; i >= 0; i--) {
-			ccur = levelcnt[i];
-			levelspread[i] = (cprv + ccur - 1) / ccur;
-			cprv = ccur;
-		}
-	}
-}
-
-/*
  * Helper function for rcu_init() that initializes one rcu_state structure.
  */
 static void __init rcu_init_one(struct rcu_state *rsp)

kernel/rcu/tree_trace.c +1 -0

index 066c64071a7b..30c5bf89ee58 100644
@@ -45,6 +45,7 @@
 
 #define RCU_TREE_NONCORE
 #include "tree.h"
+#include "rcu.h"
 
 static int r_open(struct inode *inode, struct file *file,
 					const struct seq_operations *op)

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-12 17:45 UTC #39 lore raw

PATCH 29/40 rcu: Remove redundant levelcnt[] array from rcu_init_one()

The levelcnt[] array is identical to num_rcu_lvl[], so this commit removes levelcnt[].

---
 kernel/rcu/tree.c | 9 +++------
 1 file changed, 3 insertions(+), 6 deletions(-)

kernel/rcu/tree.c +3 -6

index c5a9674824ef..101473f92ff5 100644
@@ -3968,7 +3968,6 @@ static void __init rcu_init_one(struct rcu_state *rsp)
 	static struct lock_class_key rcu_node_class[RCU_NUM_LVLS];
 	static struct lock_class_key rcu_fqs_class[RCU_NUM_LVLS];
 
-	int levelcnt[RCU_NUM_LVLS];		/* # nodes in each level. */
 	int levelspread[RCU_NUM_LVLS];		/* kids/node in each level. */
 	int cpustride = 1;
 	int i;

@@ -3983,18 +3982,16 @@ static void __init rcu_init_one(struct rcu_state *rsp)
 
 	/* Initialize the level-tracking arrays. */
 
-	for (i = 0; i < rcu_num_lvls; i++)
-		levelcnt[i] = num_rcu_lvl[i];
 	for (i = 1; i < rcu_num_lvls; i++)
-		rsp->level[i] = rsp->level[i - 1] + levelcnt[i - 1];
-	rcu_init_levelspread(levelspread, levelcnt);
+		rsp->level[i] = rsp->level[i - 1] + num_rcu_lvl[i - 1];
+	rcu_init_levelspread(levelspread, num_rcu_lvl);
 
 	/* Initialize the elements themselves, starting from the leaves. */
 
 	for (i = rcu_num_lvls - 1; i >= 0; i--) {
 		cpustride *= levelspread[i];
 		rnp = rsp->level[i];
-		for (j = 0; j < levelcnt[i]; j++, rnp++) {
+		for (j = 0; j < num_rcu_lvl[i]; j++, rnp++) {
 			raw_spin_lock_init(&ACCESS_PRIVATE(rnp, lock));
 			lockdep_set_class_and_name(&ACCESS_PRIVATE(rnp, lock),
 						   &rcu_node_class[i], buf[i]);

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-12 17:42 UTC #40 lore raw

PATCH 30/40 srcu: Move rcu_node traversal macros to rcu.h

This commit moves rcu_for_each_node_breadth_first(), rcu_for_each_nonleaf_node_breadth_first(), and rcu_for_each_leaf_node() from kernel/rcu/tree.h to kernel/rcu/rcu.h so that SRCU can access them. This commit is code-movement only.

---
 kernel/rcu/rcu.h  | 35 +++++++++++++++++++++++++++++++++++
 kernel/rcu/tree.h | 35 -----------------------------------
 2 files changed, 35 insertions(+), 35 deletions(-)

kernel/rcu/rcu.h +35 -0

index 87326479b39a..5c1798ec32f7 100644
@@ -224,6 +224,41 @@ static inline void rcu_init_levelspread(int *levelspread, const int *levelcnt)
 	}
 }
 
+/*
+ * Do a full breadth-first scan of the rcu_node structures for the
+ * specified rcu_state structure.
+ */
+#define rcu_for_each_node_breadth_first(rsp, rnp) \
+	for ((rnp) = &(rsp)->node[0]; \
+	     (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++)
+
+/*
+ * Do a breadth-first scan of the non-leaf rcu_node structures for the
+ * specified rcu_state structure.  Note that if there is a singleton
+ * rcu_node tree with but one rcu_node structure, this loop is a no-op.
+ */
+#define rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) \
+	for ((rnp) = &(rsp)->node[0]; \
+	     (rnp) < (rsp)->level[rcu_num_lvls - 1]; (rnp)++)
+
+/*
+ * Scan the leaves of the rcu_node hierarchy for the specified rcu_state
+ * structure.  Note that if there is a singleton rcu_node tree with but
+ * one rcu_node structure, this loop -will- visit the rcu_node structure.
+ * It is still a leaf node, even if it is also the root node.
+ */
+#define rcu_for_each_leaf_node(rsp, rnp) \
+	for ((rnp) = (rsp)->level[rcu_num_lvls - 1]; \
+	     (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++)
+
+/*
+ * Iterate over all possible CPUs in a leaf RCU node.
+ */
+#define for_each_leaf_node_possible_cpu(rnp, cpu) \
+	for ((cpu) = cpumask_next(rnp->grplo - 1, cpu_possible_mask); \
+	     cpu <= rnp->grphi; \
+	     cpu = cpumask_next((cpu), cpu_possible_mask))
+
 #endif /* #if defined(SRCU) || !defined(TINY_RCU) */
 
 #endif /* __LINUX_RCU_H */

kernel/rcu/tree.h +0 ~~-35~~

index 1bec3958d44f..a2a45cb629d6 100644
@@ -193,41 +193,6 @@ struct rcu_node {
 #define leaf_node_cpu_bit(rnp, cpu) (1UL << ((cpu) - (rnp)->grplo))
 
 /*
- * Do a full breadth-first scan of the rcu_node structures for the
- * specified rcu_state structure.
- */
-#define rcu_for_each_node_breadth_first(rsp, rnp) \
-	for ((rnp) = &(rsp)->node[0]; \
-	     (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++)
-
-/*
- * Do a breadth-first scan of the non-leaf rcu_node structures for the
- * specified rcu_state structure.  Note that if there is a singleton
- * rcu_node tree with but one rcu_node structure, this loop is a no-op.
- */
-#define rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) \
-	for ((rnp) = &(rsp)->node[0]; \
-	     (rnp) < (rsp)->level[rcu_num_lvls - 1]; (rnp)++)
-
-/*
- * Scan the leaves of the rcu_node hierarchy for the specified rcu_state
- * structure.  Note that if there is a singleton rcu_node tree with but
- * one rcu_node structure, this loop -will- visit the rcu_node structure.
- * It is still a leaf node, even if it is also the root node.
- */
-#define rcu_for_each_leaf_node(rsp, rnp) \
-	for ((rnp) = (rsp)->level[rcu_num_lvls - 1]; \
-	     (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++)
-
-/*
- * Iterate over all possible CPUs in a leaf RCU node.
- */
-#define for_each_leaf_node_possible_cpu(rnp, cpu) \
-	for ((cpu) = cpumask_next(rnp->grplo - 1, cpu_possible_mask); \
-	     cpu <= rnp->grphi; \
-	     cpu = cpumask_next((cpu), cpu_possible_mask))
-
-/*
  * Union to allow "aggregate OR" operation on the need for a quiescent
  * state by the normal and expedited grace periods.
  */

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-12 17:46 UTC #41 lore raw

PATCH 31/40 srcu: Make num_rcu_lvl[] array be external

This commit makes the num_rcu_lvl[] array external so that SRCU can make use of it for initializing its upcoming srcu_node tree.

---
 kernel/rcu/rcu.h  | 1 +
 kernel/rcu/tree.c | 2 +-
 2 files changed, 2 insertions(+), 1 deletion(-)

kernel/rcu/rcu.h +1 -0

index 5c1798ec32f7..4303b880ac99 100644
@@ -195,6 +195,7 @@ extern void resched_cpu(int cpu);
 #include <linux/rcu_node_tree.h>
 
 extern int rcu_num_lvls;
+extern int num_rcu_lvl[];
 extern int rcu_num_nodes;
 static bool rcu_fanout_exact;
 static int rcu_fanout_leaf;

kernel/rcu/tree.c +1 -1

index 101473f92ff5..e0eb798181bc 100644
@@ -123,7 +123,7 @@ static int rcu_fanout_leaf = RCU_FANOUT_LEAF;
 module_param(rcu_fanout_leaf, int, 0444);
 int rcu_num_lvls __read_mostly = RCU_NUM_LVLS;
 /* Number of rcu_nodes at specified level. */
-static int num_rcu_lvl[] = NUM_RCU_LVL_INIT;
+int num_rcu_lvl[] = NUM_RCU_LVL_INIT;
 int rcu_num_nodes __read_mostly = NUM_RCU_NODES; /* Total # rcu_nodes in use. */
 /* panic() on RCU Stall sysctl. */
 int sysctl_panic_on_rcu_stall __read_mostly;

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-12 17:46 UTC #42 lore raw

PATCH 32/40 srcu: Fix bogus try_check_zero() comment

---
 kernel/rcu/srcu.c | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)

kernel/rcu/srcu.c +3 -4

index d51ab050f777..5aeeaecfb673 100644
@@ -254,10 +254,9 @@ static void srcu_gp_start(struct srcu_struct *sp)
 }
 
 /*
- * @@@ Wait until all pre-existing readers complete.  Such readers
- * will have used the index specified by "idx".
- * the caller should ensures the ->completed is not changed while checking
- * and idx = (->completed & 1) ^ 1
+ * Wait until all readers counted by array index idx complete, but loop
+ * a maximum of trycount times.  The caller must ensure that ->completed
+ * is not changed while checking.
  */
 static bool try_check_zero(struct srcu_struct *sp, int idx, int trycount)
 {

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-12 17:43 UTC #43 lore raw

PATCH 33/40 srcu: Improve rcu_seq grace-period-counter abstraction

The expedited grace-period code contains several open-coded shifts know the format of an rcu_seq grace-period counter, which is not particularly good style. This commit therefore creates a new rcu_seq_ctr() function that extracts the counter portion of the counter, and an rcu_seq_state() function that extracts the low-order state bit. This commit prepares for SRCU callback parallelization, which will require two state bits.

---
 kernel/rcu/rcu.h      | 29 +++++++++++++++++++++++++----
 kernel/rcu/tree_exp.h |  9 ++++-----
 2 files changed, 29 insertions(+), 9 deletions(-)

kernel/rcu/rcu.h +25 -4

index 4303b880ac99..c62df93bfc1b 100644
@@ -61,20 +61,41 @@
  * Grace-period counter management.
  */
 
+#define RCU_SEQ_CTR_SHIFT	1
+#define RCU_SEQ_STATE_MASK	((1 << RCU_SEQ_CTR_SHIFT) - 1)
+
+/*
+ * Return the counter portion of a sequence number previously returned
+ * by rcu_seq_snap() or rcu_seq_current().
+ */
+static inline unsigned long rcu_seq_ctr(unsigned long s)
+{
+	return s >> RCU_SEQ_CTR_SHIFT;
+}
+
+/*
+ * Return the state portion of a sequence number previously returned
+ * by rcu_seq_snap() or rcu_seq_current().
+ */
+static inline int rcu_seq_state(unsigned long s)
+{
+	return s & RCU_SEQ_STATE_MASK;
+}
+
 /* Adjust sequence number for start of update-side operation. */
 static inline void rcu_seq_start(unsigned long *sp)
 {
 	WRITE_ONCE(*sp, *sp + 1);
 	smp_mb(); /* Ensure update-side operation after counter increment. */
-	WARN_ON_ONCE(!(*sp & 0x1));
+	WARN_ON_ONCE(rcu_seq_state(*sp) != 1);
 }
 
 /* Adjust sequence number for end of update-side operation. */
 static inline void rcu_seq_end(unsigned long *sp)
 {
 	smp_mb(); /* Ensure update-side operation before counter increment. */
-	WARN_ON_ONCE(!(*sp & 0x1));
-	WRITE_ONCE(*sp, *sp + 1);
+	WARN_ON_ONCE(!rcu_seq_state(*sp));
+	WRITE_ONCE(*sp, (*sp | RCU_SEQ_STATE_MASK) + 1);
 }
 
 /* Take a snapshot of the update side's sequence number. */

@@ -82,7 +103,7 @@ static inline unsigned long rcu_seq_snap(unsigned long *sp)
 {
 	unsigned long s;
 
-	s = (READ_ONCE(*sp) + 3) & ~0x1;
+	s = (READ_ONCE(*sp) + 2 * RCU_SEQ_STATE_MASK + 1) & ~RCU_SEQ_STATE_MASK;
 	smp_mb(); /* Above access must not bleed into critical section. */
 	return s;
 }

kernel/rcu/tree_exp.h +4 -5

index 027e123d93c7..e513b4ab1197 100644
@@ -292,7 +292,7 @@ static bool exp_funnel_lock(struct rcu_state *rsp, unsigned long s)
 			trace_rcu_exp_funnel_lock(rsp->name, rnp->level,
 						  rnp->grplo, rnp->grphi,
 						  TPS("wait"));
-			wait_event(rnp->exp_wq[(s >> 1) & 0x3],
+			wait_event(rnp->exp_wq[rcu_seq_ctr(s) & 0x3],
 				   sync_exp_work_done(rsp,
 						      &rdp->exp_workdone2, s));
 			return true;

@@ -534,7 +534,7 @@ static void rcu_exp_wait_wake(struct rcu_state *rsp, unsigned long s)
 			spin_unlock(&rnp->exp_lock);
 		}
 		smp_mb(); /* All above changes before wakeup. */
-		wake_up_all(&rnp->exp_wq[(rsp->expedited_sequence >> 1) & 0x3]);
+		wake_up_all(&rnp->exp_wq[rcu_seq_ctr(rsp->expedited_sequence) & 0x3]);
 	}
 	trace_rcu_exp_grace_period(rsp->name, s, TPS("endwake"));
 	mutex_unlock(&rsp->exp_wake_mutex);

@@ -612,9 +612,8 @@ static void _synchronize_rcu_expedited(struct rcu_state *rsp,
 	/* Wait for expedited grace period to complete. */
 	rdp = per_cpu_ptr(rsp->rda, raw_smp_processor_id());
 	rnp = rcu_get_root(rsp);
-	wait_event(rnp->exp_wq[(s >> 1) & 0x3],
-		   sync_exp_work_done(rsp,
-				      &rdp->exp_workdone0, s));
+	wait_event(rnp->exp_wq[rcu_seq_ctr(s) & 0x3],
+		   sync_exp_work_done(rsp, &rdp->exp_workdone0, s));
 	smp_mb(); /* Workqueue actions happen before return. */
 
 	/* Let the next expedited grace period start. */

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-12 17:41 UTC #44 lore raw

PATCH 34/40 srcu: Allow a second bit in rcu_seq for SRCU state

This commit increases the number of reserved bits at the bottom of an rcu_seq grace-period counter from one to two, as will be needed to accommodate SRCU's three-state grace periods.

---
 kernel/rcu/rcu.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

kernel/rcu/rcu.h +1 -1

index c62df93bfc1b..87a0ac95b551 100644
@@ -61,7 +61,7 @@
  * Grace-period counter management.
  */
 
-#define RCU_SEQ_CTR_SHIFT	1
+#define RCU_SEQ_CTR_SHIFT	2
 #define RCU_SEQ_STATE_MASK	((1 << RCU_SEQ_CTR_SHIFT) - 1)
 
 /*

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-12 17:46 UTC #45 lore raw

PATCH 35/40 srcu: Merge ->srcu_state into ->srcu_gp_seq

Updating ->srcu_state and ->srcu_gp_seq will lead to extremely complex race conditions given multiple callback queues, so this commit takes advantage of the two-bit state now available in rcu_seq counters to store the state in the bottom two bits of ->srcu_gp_seq.

---
 include/linux/srcu.h |  5 +----
 kernel/rcu/rcu.h     | 10 ++++++++++
 kernel/rcu/srcu.c    | 55 +++++++++++++++++++++++++++++++++-------------------
 3 files changed, 46 insertions(+), 24 deletions(-)

include/linux/srcu.h +1 -4

index ad154a7bc114..e7dbc01b61a1 100644
@@ -43,8 +43,7 @@ struct srcu_struct {
 	unsigned long completed;
 	unsigned long srcu_gp_seq;
 	struct srcu_array __percpu *per_cpu_ref;
-	spinlock_t queue_lock; /* protect ->srcu_cblist, ->srcu_state */
-	int srcu_state;
+	spinlock_t queue_lock; /* protect ->srcu_cblist */
 	struct rcu_segcblist srcu_cblist;
 	struct delayed_work work;
 #ifdef CONFIG_DEBUG_LOCK_ALLOC

@@ -56,7 +55,6 @@ struct srcu_struct {
 #define SRCU_STATE_IDLE		0
 #define SRCU_STATE_SCAN1	1
 #define SRCU_STATE_SCAN2	2
-#define SRCU_STATE_DONE		3
 
 #ifdef CONFIG_DEBUG_LOCK_ALLOC

@@ -85,7 +83,6 @@ void process_srcu(struct work_struct *work);
 		.completed = -300,					\
 		.per_cpu_ref = &name##_srcu_array,			\
 		.queue_lock = __SPIN_LOCK_UNLOCKED(name.queue_lock),	\
-		.srcu_state = SRCU_STATE_IDLE,				\
 		.srcu_cblist = RCU_SEGCBLIST_INITIALIZER(name.srcu_cblist),\
 		.work = __DELAYED_WORK_INITIALIZER(name.work, process_srcu, 0),\
 		__SRCU_DEP_MAP_INIT(name)				\

kernel/rcu/rcu.h +10 -0

index 87a0ac95b551..73e16ec4054b 100644
@@ -82,6 +82,16 @@ static inline int rcu_seq_state(unsigned long s)
 	return s & RCU_SEQ_STATE_MASK;
 }
 
+/*
+ * Set the state portion of the pointed-to sequence number.
+ * The caller is responsible for preventing conflicting updates.
+ */
+static inline void rcu_seq_set_state(unsigned long *sp, int newstate)
+{
+	WARN_ON_ONCE(newstate & ~RCU_SEQ_STATE_MASK);
+	WRITE_ONCE(*sp, (*sp & ~RCU_SEQ_STATE_MASK) + newstate);
+}
+
 /* Adjust sequence number for start of update-side operation. */
 static inline void rcu_seq_start(unsigned long *sp)
 {

kernel/rcu/srcu.c +35 ~~-20~~

138 lines not on this page · raw message

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-12 17:45 UTC #46 lore raw

PATCH 36/40 srcu: Crude control of expedited grace periods

SRCU's implementation of expedited grace periods has always assumed that the SRCU instance is idle when the expedited request arrives. This commit improves this a bit by maintaining a count of the number of outstanding expedited requests, thus allowing prior non-expedited grace periods accommodate these requests by shifting to expedited mode. However, any non-expedited wait already in progress will still wait for the full duration.

Improved control of expedited grace periods is planned, but one step at a time.

---
 include/linux/srcu.h |  1 +
 kernel/rcu/srcu.c    | 84 ++++++++++++++++++++++++++++------------------------
 2 files changed, 47 insertions(+), 38 deletions(-)

include/linux/srcu.h +1 -0

index e7dbc01b61a1..73a1b6296224 100644
@@ -42,6 +42,7 @@ struct srcu_array {
 struct srcu_struct {
 	unsigned long completed;
 	unsigned long srcu_gp_seq;
+	atomic_t srcu_exp_cnt;
 	struct srcu_array __percpu *per_cpu_ref;
 	spinlock_t queue_lock; /* protect ->srcu_cblist */
 	struct rcu_segcblist srcu_cblist;

kernel/rcu/srcu.c +46 ~~-38~~

168 lines not on this page · raw message

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-12 17:45 UTC #47 lore raw

PATCH 37/40 srcu: Create a tiny SRCU

In response to automated complaints about modifications to SRCU increasing its size, this commit creates a tiny SRCU that is used in SMP=y && PREEMPT=n builds.

---
 include/linux/srcu.h     |  69 ++-------------
 include/linux/srcutiny.h |  81 ++++++++++++++++++
 include/linux/srcutree.h |  91 ++++++++++++++++++++
 init/Kconfig             |  12 +++
 kernel/rcu/Makefile      |   3 +-
 kernel/rcu/rcutorture.c  |   2 +
 kernel/rcu/srcutiny.c    | 215 +++++++++++++++++++++++++++++++++++++++++++++++
 7 files changed, 411 insertions(+), 62 deletions(-)
 create mode 100644 include/linux/srcutiny.h
 create mode 100644 include/linux/srcutree.h
 create mode 100644 kernel/rcu/srcutiny.c

include/linux/srcu.h +8 ~~-61~~

index 73a1b6296224..907f09b14eda 100644
@@ -34,28 +34,7 @@
 #include <linux/workqueue.h>
 #include <linux/rcu_segcblist.h>
 
-struct srcu_array {
-	unsigned long lock_count[2];
-	unsigned long unlock_count[2];
-};
-
-struct srcu_struct {
-	unsigned long completed;
-	unsigned long srcu_gp_seq;
-	atomic_t srcu_exp_cnt;
-	struct srcu_array __percpu *per_cpu_ref;
-	spinlock_t queue_lock; /* protect ->srcu_cblist */
-	struct rcu_segcblist srcu_cblist;
-	struct delayed_work work;
-#ifdef CONFIG_DEBUG_LOCK_ALLOC
-	struct lockdep_map dep_map;
-#endif /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
-};
-
-/* Values for -> state variable. */
-#define SRCU_STATE_IDLE		0
-#define SRCU_STATE_SCAN1	1
-#define SRCU_STATE_SCAN2	2
+struct srcu_struct;
 
 #ifdef CONFIG_DEBUG_LOCK_ALLOC

@@ -77,42 +56,13 @@ int init_srcu_struct(struct srcu_struct *sp);
 #define __SRCU_DEP_MAP_INIT(srcu_name)
 #endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */
 
-void process_srcu(struct work_struct *work);
-
-#define __SRCU_STRUCT_INIT(name)					\
-	{								\
-		.completed = -300,					\
-		.per_cpu_ref = &name##_srcu_array,			\
-		.queue_lock = __SPIN_LOCK_UNLOCKED(name.queue_lock),	\
-		.srcu_cblist = RCU_SEGCBLIST_INITIALIZER(name.srcu_cblist),\
-		.work = __DELAYED_WORK_INITIALIZER(name.work, process_srcu, 0),\
-		__SRCU_DEP_MAP_INIT(name)				\
-	}
-
-/*
- * Define and initialize a srcu struct at build time.
- * Do -not- call init_srcu_struct() nor cleanup_srcu_struct() on it.
- *
- * Note that although DEFINE_STATIC_SRCU() hides the name from other
- * files, the per-CPU variable rules nevertheless require that the
- * chosen name be globally unique.  These rules also prohibit use of
- * DEFINE_STATIC_SRCU() within a function.  If these rules are too
- * restrictive, declare the srcu_struct manually.  For example, in
- * each file:
- *
- *	static struct srcu_struct my_srcu;
- *
- * Then, before the first use of each my_srcu, manually initialize it:
- *
- *	init_srcu_struct(&my_srcu);
- *
- * See include/linux/percpu-defs.h for the rules on per-CPU variables.
- */
-#define __DEFINE_SRCU(name, is_static)					\
-	static DEFINE_PER_CPU(struct srcu_array, name##_srcu_array);\
-	is_static struct srcu_struct name = __SRCU_STRUCT_INIT(name)
-#define DEFINE_SRCU(name)		__DEFINE_SRCU(name, /* not static */)
-#define DEFINE_STATIC_SRCU(name)	__DEFINE_SRCU(name, static)
+#ifdef CONFIG_TINY_SRCU
+#include <linux/srcutiny.h>
+#elif defined(CONFIG_TREE_SRCU)
+#include <linux/srcutree.h>
+#else
+#error "Unknown SRCU implementation specified to kernel configuration"
+#endif
 
 /**
  * call_srcu() - Queue a callback for invocation after an SRCU grace period

@@ -138,9 +88,6 @@ void cleanup_srcu_struct(struct srcu_struct *sp);
 int __srcu_read_lock(struct srcu_struct *sp) __acquires(sp);
 void __srcu_read_unlock(struct srcu_struct *sp, int idx) __releases(sp);
 void synchronize_srcu(struct srcu_struct *sp);
-void synchronize_srcu_expedited(struct srcu_struct *sp);
-unsigned long srcu_batches_completed(struct srcu_struct *sp);
-void srcu_barrier(struct srcu_struct *sp);
 
 #ifdef CONFIG_DEBUG_LOCK_ALLOC

include/linux/srcutiny.h new file +81 -0

81 lines not on this page · raw message

include/linux/srcutree.h new file +91 -0

91 lines not on this page · raw message

init/Kconfig +12 -0

index a92f27da4a27..d269f2ca17b8 100644
@@ -526,6 +526,18 @@ config SRCU
 	  permits arbitrary sleeping or blocking within RCU read-side critical
 	  sections.
 
+config TINY_SRCU
+	bool
+	default y if TINY_RCU
+	help
+	  This option selects the single-CPU non-preemptible version of SRCU.
+
+config TREE_SRCU
+	bool
+	default y if !TINY_RCU
+	help
+	  This option selects the full-fledged version of SRCU.
+
 config TASKS_RCU
 	bool
 	default n

kernel/rcu/Makefile +2 -1

index 18dfc485225c..b853214a2b99 100644
@@ -3,7 +3,8 @@
 KCOV_INSTRUMENT := n
 
 obj-y += update.o sync.o
-obj-$(CONFIG_SRCU) += srcu.o
+obj-$(CONFIG_TREE_SRCU) += srcu.o
+obj-$(CONFIG_TINY_SRCU) += srcutiny.o
 obj-$(CONFIG_RCU_TORTURE_TEST) += rcutorture.o
 obj-$(CONFIG_RCU_PERF_TEST) += rcuperf.o
 obj-$(CONFIG_TREE_RCU) += tree.o

kernel/rcu/rcutorture.c +2 -0

index cccc417a8135..98591e16db1a 100644
@@ -559,6 +559,7 @@ static void srcu_torture_barrier(void)
 
 static void srcu_torture_stats(void)
 {
+#ifdef CONFIG_TREE_SRCU
 	int cpu;
 	int idx = srcu_ctlp->completed & 0x1;

@@ -587,6 +588,7 @@ static void srcu_torture_stats(void)
 		pr_cont(" %d(%ld,%ld)", cpu, c0, c1);
 	}
 	pr_cont("\n");
+#endif
 }
 
 static void srcu_torture_synchronize_expedited(void)

kernel/rcu/srcutiny.c new file +215 -0

215 lines not on this page · raw message

Signature · 1 line

2.5.2

Peter Zijlstra

maintainer

2017-04-13 09:48 UTC #48 lore raw

Re: [PATCH tip/core/rcu 37/40] srcu: Create a tiny SRCU

Quoted 3 lines · Paul E. McKenney

In response to automated complaints about modifications to SRCU increasing its size, this commit creates a tiny SRCU that is used in SMP=y && PREEMPT=n builds.

You mean SMP=n && PREEMPT=n, right?

config TINY_RCU

        bool
        default y if !PREEMPT && !SMP

Quoted 5 lines

init/Kconfig +5 -0

+config TINY_SRCU
+	bool
+	default y if TINY_RCU
+	help
+	  This option selects the single-CPU non-preemptible version of SRCU.

Paul E. McKenney

author

2017-04-13 16:30 UTC #49 lore raw

Re: [PATCH tip/core/rcu 37/40] srcu: Create a tiny SRCU

Quoted 6 lines · Peter Zijlstra

Quoted 3 lines · Paul E. McKenney
In response to automated complaints about modifications to SRCU increasing its size, this commit creates a tiny SRCU that is used in SMP=y && PREEMPT=n builds.
You mean SMP=n && PREEMPT=n, right?

I do, thank you for spotting this! Fixed.

Thanx, Paul

Quoted 10 lines

config TINY_RCU

        bool
        default y if !PREEMPT && !SMP

Quoted 5 lines

init/Kconfig +5 -0

+config TINY_SRCU
+	bool
+	default y if TINY_RCU
+	help
+	  This option selects the single-CPU non-preemptible version of SRCU.

Paul E. McKenney

author

2017-04-12 17:42 UTC #50 lore raw

PATCH 38/40 srcutorture: Print Tiny SRCU reader statistics

The srcu_torture_stats() function is adapted to the specific srcu_struct layout traditionally used by SRCU. This commit therefore adds support for Tiny SRCU.

---
 kernel/rcu/rcutorture.c | 15 +++++++++++----
 1 file changed, 11 insertions(+), 4 deletions(-)

kernel/rcu/rcutorture.c +11 -4

index 98591e16db1a..9cbb8a7b909d 100644
@@ -559,11 +559,12 @@ static void srcu_torture_barrier(void)
 
 static void srcu_torture_stats(void)
 {
-#ifdef CONFIG_TREE_SRCU
-	int cpu;
-	int idx = srcu_ctlp->completed & 0x1;
+	int __maybe_unused cpu;
+	int idx;
 
-	pr_alert("%s%s per-CPU(idx=%d):",
+#ifdef CONFIG_TREE_SRCU
+	idx = srcu_ctlp->completed & 0x1;
+	pr_alert("%s%s Tree SRCU per-CPU(idx=%d):",
 		 torture_type, TORTURE_FLAG, idx);
 	for_each_possible_cpu(cpu) {
 		unsigned long l0, l1;

@@ -588,6 +589,12 @@ static void srcu_torture_stats(void)
 		pr_cont(" %d(%ld,%ld)", cpu, c0, c1);
 	}
 	pr_cont("\n");
+#elif defined(CONFIG_TINY_SRCU)
+	idx = READ_ONCE(srcu_ctlp->srcu_idx) & 0x1;
+	pr_alert("%s%s Tiny SRCU per-CPU(idx=%d): (%d,%d)\n",
+		 torture_type, TORTURE_FLAG, idx,
+		 READ_ONCE(srcu_ctlp->srcu_lock_nesting[!idx]),
+		 READ_ONCE(srcu_ctlp->srcu_lock_nesting[idx]));
 #endif
 }

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-12 17:42 UTC #51 lore raw

PATCH 39/40 srcu: Introduce CLASSIC_SRCU Kconfig option

The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm.

---
 include/linux/srcu.h        |   2 +
 include/linux/srcuclassic.h | 101 ++++++++
 init/Kconfig                |  21 +-
 kernel/rcu/Makefile         |   3 +-
 kernel/rcu/rcutorture.c     |   2 +-
 kernel/rcu/srcu.c           | 347 ++++++++++++++-----------
 kernel/rcu/srcutree.c       | 613 ++++++++++++++++++++++++++++++++++++++++++++
 7 files changed, 934 insertions(+), 155 deletions(-)
 create mode 100644 include/linux/srcuclassic.h
 create mode 100644 kernel/rcu/srcutree.c

include/linux/srcu.h +2 -0

index 907f09b14eda..167ad8831aaf 100644
@@ -60,6 +60,8 @@ int init_srcu_struct(struct srcu_struct *sp);
 #include <linux/srcutiny.h>
 #elif defined(CONFIG_TREE_SRCU)
 #include <linux/srcutree.h>
+#elif defined(CONFIG_CLASSIC_SRCU)
+#include <linux/srcuclassic.h>
 #else
 #error "Unknown SRCU implementation specified to kernel configuration"
 #endif

include/linux/srcuclassic.h new file +101 -0

101 lines not on this page · raw message

init/Kconfig +19 -2

index d269f2ca17b8..558cc3638ab9 100644
@@ -526,15 +526,32 @@ config SRCU
 	  permits arbitrary sleeping or blocking within RCU read-side critical
 	  sections.
 
+config CLASSIC_SRCU
+	bool "Use v4.11 classic SRCU implementation"
+	default n
+	depends on RCU_EXPERT && SRCU
+	help
+	  This option selects the traditional well-tested classic SRCU
+	  implementation from v4.11, as might be desired for enterprise
+	  Linux distributions.  Without this option, the shiny new
+	  Tiny SRCU and Tree SRCU implementations are used instead.
+	  At some point, it is hoped that Tiny SRCU and Tree SRCU
+	  will accumulate enough test time and confidence to allow
+	  Classic SRCU to be dropped entirely.
+
+	  Say Y if you need a rock-solid SRCU.
+
+	  Say N if you would like help test Tree SRCU.
+
 config TINY_SRCU
 	bool
-	default y if TINY_RCU
+	default y if TINY_RCU && !CLASSIC_SRCU
 	help
 	  This option selects the single-CPU non-preemptible version of SRCU.
 
 config TREE_SRCU
 	bool
-	default y if !TINY_RCU
+	default y if !TINY_RCU && !CLASSIC_SRCU
 	help
 	  This option selects the full-fledged version of SRCU.

kernel/rcu/Makefile +2 -1

index b853214a2b99..158e6593d58c 100644
@@ -3,7 +3,8 @@
 KCOV_INSTRUMENT := n
 
 obj-y += update.o sync.o
-obj-$(CONFIG_TREE_SRCU) += srcu.o
+obj-$(CONFIG_CLASSIC_SRCU) += srcu.o
+obj-$(CONFIG_TREE_SRCU) += srcutree.o
 obj-$(CONFIG_TINY_SRCU) += srcutiny.o
 obj-$(CONFIG_RCU_TORTURE_TEST) += rcutorture.o
 obj-$(CONFIG_RCU_PERF_TEST) += rcuperf.o

kernel/rcu/rcutorture.c +1 -1

index 9cbb8a7b909d..6f344b6748a8 100644
@@ -562,7 +562,7 @@ static void srcu_torture_stats(void)
 	int __maybe_unused cpu;
 	int idx;
 
-#ifdef CONFIG_TREE_SRCU
+#if defined(CONFIG_TREE_SRCU) || defined(CONFIG_CLASSIC_SRCU)
 	idx = srcu_ctlp->completed & 0x1;
 	pr_alert("%s%s Tree SRCU per-CPU(idx=%d):",
 		 torture_type, TORTURE_FLAG, idx);

kernel/rcu/srcu.c +196 ~~-151~~

497 lines not on this page · raw message

kernel/rcu/srcutree.c new file +613 -0

613 lines not on this page · raw message

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-12 17:43 UTC #52 lore raw

PATCH 40/40 srcu: Parallelize callback handling

Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention.

Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks each require a bit more work. These are handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree).

In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked.

The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but this gets complex and racy. And interestingly enough, it is never necessary to start a grace period in this case because srcu_barrier() only enqueues a callback when a callback is already present. And a grace period has to have already been started for this pre-existing callback. And it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the pre-existing callback. The special case where all the pre-existing callbacks are on a different list being invoked is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked.

Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times.

This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU.

[1] https://lwn.net/Articles/309030/

---
 include/linux/rcu_segcblist.h |  42 ++-
 include/linux/srcutree.h      |  80 ++++--
 kernel/rcu/rcutorture.c       |  20 +-
 kernel/rcu/srcutree.c         | 639 +++++++++++++++++++++++++++++++++---------
 kernel/rcu/tree.c             |   6 +
 kernel/rcu/tree.h             |   8 +
 6 files changed, 645 insertions(+), 150 deletions(-)

include/linux/rcu_segcblist.h +38 -4

index 74b1e7243955..ced8f313fd05 100644
@@ -402,6 +402,37 @@ static inline void rcu_segcblist_enqueue(struct rcu_segcblist *rsclp,
 }
 
 /*
+ * Entrain the specified callback onto the specified rcu_segcblist at
+ * the end of the last non-empty segment.  If the entire rcu_segcblist
+ * is empty, make no change, but return false.
+ *
+ * This is intended for use by rcu_barrier()-like primitives, -not-
+ * for normal grace-period use.  IMPORTANT:  The callback you enqueue
+ * will wait for all prior callbacks, NOT necessarily for a grace
+ * period.  You have been warned.
+ */
+static inline bool rcu_segcblist_entrain(struct rcu_segcblist *rsclp,
+					 struct rcu_head *rhp, bool lazy)
+{
+	int i;
+
+	if (rcu_segcblist_n_cbs(rsclp) == 0)
+		return false;
+	WRITE_ONCE(rsclp->len, rsclp->len + 1);
+	if (lazy)
+		rsclp->len_lazy++;
+	smp_mb(); /* Ensure counts are updated before callback is entrained. */
+	rhp->next = NULL;
+	for (i = RCU_NEXT_TAIL; i > RCU_DONE_TAIL; i--)
+		if (rsclp->tails[i] != rsclp->tails[i - 1])
+			break;
+	*rsclp->tails[i] = rhp;
+	for (; i <= RCU_NEXT_TAIL; i++)
+		rsclp->tails[i] = &rhp->next;
+	return true;
+}
+
+/*
  * Extract only the counts from the specified rcu_segcblist structure,
  * and place them in the specified rcu_cblist structure.  This function
  * supports both callback orphaning and invocation, hence the separation

@@ -537,7 +568,8 @@ static inline void rcu_segcblist_advance(struct rcu_segcblist *rsclp,
 	int i, j;
 
 	WARN_ON_ONCE(!rcu_segcblist_is_enabled(rsclp));
-	WARN_ON_ONCE(rcu_segcblist_restempty(rsclp, RCU_DONE_TAIL));
+	if (rcu_segcblist_restempty(rsclp, RCU_DONE_TAIL))
+		return;
 
 	/*
 	 * Find all callbacks whose ->gp_seq numbers indicate that they

@@ -582,8 +614,9 @@ static inline void rcu_segcblist_advance(struct rcu_segcblist *rsclp,
  * them to complete at the end of the earlier grace period.
  *
  * This function operates on an rcu_segcblist structure, and also the
- * grace-period sequence number at which new callbacks would become
- * ready to invoke.
+ * grace-period sequence number seq at which new callbacks would become
+ * ready to invoke.  Returns true if there are callbacks that won't be
+ * ready to invoke until seq, false otherwise.
  */
 static inline bool rcu_segcblist_accelerate(struct rcu_segcblist *rsclp,
 					    unsigned long seq)

@@ -591,7 +624,8 @@ static inline bool rcu_segcblist_accelerate(struct rcu_segcblist *rsclp,
 	int i;
 
 	WARN_ON_ONCE(!rcu_segcblist_is_enabled(rsclp));
-	WARN_ON_ONCE(rcu_segcblist_restempty(rsclp, RCU_DONE_TAIL));
+	if (rcu_segcblist_restempty(rsclp, RCU_DONE_TAIL))
+		return false;
 
 	/*
 	 * Find the segment preceding the oldest segment of callbacks

include/linux/srcutree.h +64 ~~-16~~

107 lines not on this page · raw message

kernel/rcu/rcutorture.c +19 -1

index 6f344b6748a8..e9d4527cdd43 100644
@@ -563,17 +563,30 @@ static void srcu_torture_stats(void)
 	int idx;
 
 #if defined(CONFIG_TREE_SRCU) || defined(CONFIG_CLASSIC_SRCU)
+#ifdef CONFIG_TREE_SRCU
+	idx = srcu_ctlp->srcu_idx & 0x1;
+#else /* #ifdef CONFIG_TREE_SRCU */
 	idx = srcu_ctlp->completed & 0x1;
+#endif /* #else #ifdef CONFIG_TREE_SRCU */
 	pr_alert("%s%s Tree SRCU per-CPU(idx=%d):",
 		 torture_type, TORTURE_FLAG, idx);
 	for_each_possible_cpu(cpu) {
 		unsigned long l0, l1;
 		unsigned long u0, u1;
 		long c0, c1;
-		struct srcu_array *counts = per_cpu_ptr(srcu_ctlp->per_cpu_ref, cpu);
+#ifdef CONFIG_TREE_SRCU
+		struct srcu_data *counts;
 
+		counts = per_cpu_ptr(srcu_ctlp->sda, cpu);
+		u0 = counts->srcu_unlock_count[!idx];
+		u1 = counts->srcu_unlock_count[idx];
+#else /* #ifdef CONFIG_TREE_SRCU */
+		struct srcu_array *counts;
+
+		counts = per_cpu_ptr(srcu_ctlp->per_cpu_ref, cpu);
 		u0 = counts->unlock_count[!idx];
 		u1 = counts->unlock_count[idx];
+#endif /* #else #ifdef CONFIG_TREE_SRCU */
 
 		/*
 		 * Make sure that a lock is always counted if the corresponding

@@ -581,8 +594,13 @@ static void srcu_torture_stats(void)
 		 */
 		smp_rmb();
 
+#ifdef CONFIG_TREE_SRCU
+		l0 = counts->srcu_lock_count[!idx];
+		l1 = counts->srcu_lock_count[idx];
+#else /* #ifdef CONFIG_TREE_SRCU */
 		l0 = counts->lock_count[!idx];
 		l1 = counts->lock_count[idx];
+#endif /* #else #ifdef CONFIG_TREE_SRCU */
 
 		c0 = l0 - u0;
 		c1 = l1 - u1;

kernel/rcu/srcutree.c +510 ~~-129~~

877 lines not on this page · raw message

kernel/rcu/tree.c +6 -0

index e0eb798181bc..75ca946734e6 100644
@@ -3783,12 +3783,16 @@ int rcutree_online_cpu(unsigned int cpu)
 {
 	sync_sched_exp_online_cleanup(cpu);
 	rcutree_affinity_setting(cpu, -1);
+	if (IS_ENABLED(CONFIG_TREE_SRCU))
+		srcu_online_cpu(cpu);
 	return 0;
 }
 
 int rcutree_offline_cpu(unsigned int cpu)
 {
 	rcutree_affinity_setting(cpu, cpu);
+	if (IS_ENABLED(CONFIG_TREE_SRCU))
+		srcu_offline_cpu(cpu);
 	return 0;
 }

@@ -4164,6 +4168,8 @@ void __init rcu_init(void)
 	for_each_online_cpu(cpu) {
 		rcutree_prepare_cpu(cpu);
 		rcu_cpu_starting(cpu);
+		if (IS_ENABLED(CONFIG_TREE_SRCU))
+			srcu_online_cpu(cpu);
 	}
 }

kernel/rcu/tree.h +8 -0

index a2a45cb629d6..0e598ab08fea 100644
@@ -541,6 +541,14 @@ static bool rcu_nohz_full_cpu(struct rcu_state *rsp);
 static void rcu_dynticks_task_enter(void);
 static void rcu_dynticks_task_exit(void);
 
+#ifdef CONFIG_SRCU
+void srcu_online_cpu(unsigned int cpu);
+void srcu_offline_cpu(unsigned int cpu);
+#else /* #ifdef CONFIG_SRCU */
+void srcu_online_cpu(unsigned int cpu) { }
+void srcu_offline_cpu(unsigned int cpu) { }
+#endif /* #else #ifdef CONFIG_SRCU */
+
 #endif /* #ifndef RCU_TREE_NONCORE */
 
 #ifdef CONFIG_RCU_TRACE

Signature · 1 line

2.5.2

Peter Zijlstra

maintainer

2017-04-13 09:50 UTC #53 lore raw

Re: [PATCH tip/core/rcu 40/40] srcu: Parallelize callback handling

Quoted 1 line · Paul E. McKenney

Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1],

Bugger, now you're making me feel bad for not having updated those patches in ages.. I'll try and bump it on the todo list.

Paul E. McKenney

author

2017-04-13 16:37 UTC #54 lore raw

Re: [PATCH tip/core/rcu 40/40] srcu: Parallelize callback handling

Quoted 5 lines · Peter Zijlstra

Quoted 1 line · Paul E. McKenney
Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1],
Bugger, now you're making me feel bad for not having updated those patches in ages.. I'll try and bump it on the todo list.

Apologies, I wasn't trying to make you feel bad. But I did get an old version of the patch. A much more recent one is here, which I have added to the commit log:

	https://patchwork.kernel.org/patch/5108281/

For my part, I feel bad that I didn't realize much earlier that parallel SRCU callbacks were needed. As it was, someone had to tell me late last year.

But yes, it would be really cool to have lockless VMA lookup!!!

Thanx, Paul

Paul E. McKenney

author

2017-04-13 17:43 UTC #55 lore raw

Re: [PATCH tip/core/rcu 40/40] srcu: Parallelize callback handling

Quoted 11 lines · Paul E. McKenney

Quoted 5 lines · Peter Zijlstra
Quoted 1 line · Paul E. McKenney
Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1],
Bugger, now you're making me feel bad for not having updated those patches in ages.. I'll try and bump it on the todo list.
Apologies, I wasn't trying to make you feel bad. But I did get an old version of the patch. A much more recent one is here, which I have added to the commit log:

Quoted 5 lines

	https://patchwork.kernel.org/patch/5108281/
For my part, I feel bad that I didn't realize much earlier that parallel SRCU callbacks were needed. As it was, someone had to tell me late last year.

And it turns out that Laurent Dufour (CCed) forward-ported your patch from the above patchworks URL to 4.11. He is chasing down some mmseq bugs on a best-effort basis.

Thanx, Paul

Quoted 3 lines

But yes, it would be really cool to have lockless VMA lookup!!!
Thanx, Paul

Peter Zijlstra

maintainer

2017-04-13 09:54 UTC #56 lore raw

Re: [PATCH tip/core/rcu 40/40] srcu: Parallelize callback handling

Quoted 41 lines · Paul E. McKenney

Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention.
Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks each require a bit more work. These are handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree).
In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked.
The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but this gets complex and racy. And interestingly enough, it is never necessary to start a grace period in this case because srcu_barrier() only enqueues a callback when a callback is already present. And a grace period has to have already been started for this pre-existing callback. And it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the pre-existing callback. The special case where all the pre-existing callbacks are on a different list being invoked is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked.

Quoted 4 lines

Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times.

So one thing I've asked before I think, would it not be possible to abstract PREEMPT_RCU and use the exact same code for PREEMPT_RCU and SRCU ?

Paul E. McKenney

author

2017-04-13 16:50 UTC #57 lore raw

Re: [PATCH tip/core/rcu 40/40] srcu: Parallelize callback handling

Quoted 42 lines · Peter Zijlstra

Quoted 41 lines · Paul E. McKenney
Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention.
Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks each require a bit more work. These are handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree).
In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked.
The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but this gets complex and racy. And interestingly enough, it is never necessary to start a grace period in this case because srcu_barrier() only enqueues a callback when a callback is already present. And a grace period has to have already been started for this pre-existing callback. And it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the pre-existing callback. The special case where all the pre-existing callbacks are on a different list being invoked is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked.

Quoted 8 lines

Quoted 4 lines
Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times.
So one thing I've asked before I think, would it not be possible to abstract PREEMPT_RCU and use the exact same code for PREEMPT_RCU and SRCU ?

I took a hard look at that some time ago, and it gets pretty ugly pretty quickly. Much of the PREEMPT_RCU code has the idea that there is only one global PREEMPT_RCU implementation baked deeply into it. For but one example, the handling of an arbitrarily large number of ->blkd_tasks lists at context-switch time would not be pretty, especially if the task in question blocked while in both a PREEMPT_RCU and in an SRCU read-side critical section. Or, worse yet, if it blocked while in several different SRCU read-side critical sections.

It might be easier to go the other way and implement PREEMPT_RCU in terms of SRCU, but I don't believe that the read-side smp_mb() calls would make people happy. Plus there are use cases that would not be well-served by idle no longer being an extended quiescent state. And SRCU currently has inconvenient restrictions about use in interrupt and NMI handlers.

It might well be that there is a global solution for all this, but in the meantime I am instead sharing common code and doing a bit of consolidation.

Thanx, Paul

Michael Ellerman

2017-04-20 03:40 UTC #58 lore raw

powerpc KVM build break in linux-next (was Re: [PATCH tip/core/rcu 40/40] srcu: Parallelize callback handling)

In reply to Michael Ellerman

Quoted 5 lines · Paul E. McKenney

include/linux/srcutree.h +0 -0
index f2b3bd6c6bc2..0400e211aa44 100644
@@ -24,25 +24,75 @@

...

Quoted 18 lines

include/linux/srcutree.h +11 -6

+/*
+ * Per-SRCU-domain structure, similar in function to rcu_state.
+ */
 struct srcu_struct {
-	unsigned long completed;
-	unsigned long srcu_gp_seq;
-	atomic_t srcu_exp_cnt;
-	struct srcu_array __percpu *per_cpu_ref;
-	spinlock_t queue_lock; /* protect ->srcu_cblist */
-	struct rcu_segcblist srcu_cblist;
+	struct srcu_node node[NUM_RCU_NODES];	/* Combining tree. */
+	struct srcu_node *level[RCU_NUM_LVLS + 1];
+						/* First node at each level. */
+	struct mutex srcu_cb_mutex;		/* Serialize CB preparation. */
+	spinlock_t gp_lock;			/* protect ->srcu_cblist */
+	struct mutex srcu_gp_mutex;		/* Serialize GP work. */
+	unsigned int srcu_idx;			/* Current rdr array element. */
+	unsigned long srcu_gp_seq;		/* Grace-period seq #. */

include/linux/srcutree.h line 88

Quoted 10 lines

include/linux/srcutree.h +10 -0

+	unsigned long srcu_gp_seq_needed;	/* Latest gp_seq needed. */
+	atomic_t srcu_exp_cnt;			/* # ongoing expedited GPs. */
+	struct srcu_data __percpu *sda;		/* Per-CPU srcu_data array. */
+	unsigned long srcu_barrier_seq;		/* srcu_barrier seq #. */
+	struct mutex srcu_barrier_mutex;	/* Serialize barrier ops. */
+	struct completion srcu_barrier_completion;
+						/* Awaken barrier rq at end. */
+	atomic_t srcu_barrier_cpu_cnt;		/* # CPUs not yet posting a */
+						/*  callback for the barrier */
+						/*  operation. */

This change seems to have had the non-obvious effect of breaking the powerpc KVM build.

Because struct kvm contains two srcu_structs which are located before the kvm_arch struct, the increase in size of srcu_struct has caused the offset from the start of struct kvm to kvm_arch to be too big for some of our asm.

struct kvm {
	spinlock_t mmu_lock;
	struct mutex slots_lock;
	struct mm_struct *mm; /* userspace tied to this vm */
	struct kvm_memslots *memslots[KVM_ADDRESS_SPACE_NUM];
	struct srcu_struct srcu;
	struct srcu_struct irq_srcu;
        ...
	struct kvm_arch arch;

Example error:
arch/powerpc/kvm/book3s_hv_rmhandlers.S:617: Error: operand out of range (0x000000000000b328 is not between 0xffffffffffff8000 and 0x0000000000007fff)

Where line 617 is:
lwz r7,KVM_LPID(r9)

And the KVM_LPID constant comes from asm-offsets.s. The diff of old vs new is:

-->KVM_LPID 17752 offsetof(struct kvm, arch.lpid) #
+->KVM_LPID 45864 offsetof(struct kvm, arch.lpid) #

We can probably just fix it by changing the asm to keep the address of kvm_arch in a register, and then offset from that.

cheers

Paul E. McKenney

author

2017-04-20 14:19 UTC #59 lore raw

Re: powerpc KVM build break in linux-next (was Re: [PATCH tip/core/rcu 40/40] srcu: Parallelize callback handling)

Quoted 65 lines · Michael Ellerman

Quoted 5 lines · Paul E. McKenney

include/linux/srcutree.h +0 -0
index f2b3bd6c6bc2..0400e211aa44 100644
@@ -24,25 +24,75 @@

...

Quoted 18 lines

include/linux/srcutree.h +11 -6

+/*
+ * Per-SRCU-domain structure, similar in function to rcu_state.
+ */
 struct srcu_struct {
-	unsigned long completed;
-	unsigned long srcu_gp_seq;
-	atomic_t srcu_exp_cnt;
-	struct srcu_array __percpu *per_cpu_ref;
-	spinlock_t queue_lock; /* protect ->srcu_cblist */
-	struct rcu_segcblist srcu_cblist;
+	struct srcu_node node[NUM_RCU_NODES];	/* Combining tree. */
+	struct srcu_node *level[RCU_NUM_LVLS + 1];
+						/* First node at each level. */
+	struct mutex srcu_cb_mutex;		/* Serialize CB preparation. */
+	spinlock_t gp_lock;			/* protect ->srcu_cblist */
+	struct mutex srcu_gp_mutex;		/* Serialize GP work. */
+	unsigned int srcu_idx;			/* Current rdr array element. */
+	unsigned long srcu_gp_seq;		/* Grace-period seq #. */

include/linux/srcutree.h line 88

Quoted 10 lines

include/linux/srcutree.h +10 -0

+	unsigned long srcu_gp_seq_needed;	/* Latest gp_seq needed. */
+	atomic_t srcu_exp_cnt;			/* # ongoing expedited GPs. */
+	struct srcu_data __percpu *sda;		/* Per-CPU srcu_data array. */
+	unsigned long srcu_barrier_seq;		/* srcu_barrier seq #. */
+	struct mutex srcu_barrier_mutex;	/* Serialize barrier ops. */
+	struct completion srcu_barrier_completion;
+						/* Awaken barrier rq at end. */
+	atomic_t srcu_barrier_cpu_cnt;		/* # CPUs not yet posting a */
+						/*  callback for the barrier */
+						/*  operation. */

This change seems to have had the non-obvious effect of breaking the powerpc KVM build.

struct kvm {
	spinlock_t mmu_lock;
	struct mutex slots_lock;
	struct mm_struct *mm; /* userspace tied to this vm */
	struct kvm_memslots *memslots[KVM_ADDRESS_SPACE_NUM];
	struct srcu_struct srcu;
	struct srcu_struct irq_srcu;
        ...
	struct kvm_arch arch;

Example error:
arch/powerpc/kvm/book3s_hv_rmhandlers.S:617: Error: operand out of range (0x000000000000b328 is not between 0xffffffffffff8000 and 0x0000000000007fff)

Where line 617 is:
lwz r7,KVM_LPID(r9)

And the KVM_LPID constant comes from asm-offsets.s. The diff of old vs new is:

Quoted 6 lines

-->KVM_LPID 17752 offsetof(struct kvm, arch.lpid) #
+->KVM_LPID 45864 offsetof(struct kvm, arch.lpid) #
We can probably just fix it by changing the asm to keep the address of kvm_arch in a register, and then offset from that.

Ouch!

What do you need from me?

Thanx, Paul

Paolo Bonzini

maintainer

2017-04-20 15:28 UTC #60 lore raw

Re: powerpc KVM build break in linux-next (was Re: [PATCH tip/core/rcu 40/40] srcu: Parallelize callback handling)

Quoted 65 lines · Michael Ellerman

Quoted 5 lines · Paul E. McKenney

include/linux/srcutree.h +0 -0
index f2b3bd6c6bc2..0400e211aa44 100644
@@ -24,25 +24,75 @@

...

Quoted 18 lines

include/linux/srcutree.h +11 -6

+/*
+ * Per-SRCU-domain structure, similar in function to rcu_state.
+ */
 struct srcu_struct {
-	unsigned long completed;
-	unsigned long srcu_gp_seq;
-	atomic_t srcu_exp_cnt;
-	struct srcu_array __percpu *per_cpu_ref;
-	spinlock_t queue_lock; /* protect ->srcu_cblist */
-	struct rcu_segcblist srcu_cblist;
+	struct srcu_node node[NUM_RCU_NODES];	/* Combining tree. */
+	struct srcu_node *level[RCU_NUM_LVLS + 1];
+						/* First node at each level. */
+	struct mutex srcu_cb_mutex;		/* Serialize CB preparation. */
+	spinlock_t gp_lock;			/* protect ->srcu_cblist */
+	struct mutex srcu_gp_mutex;		/* Serialize GP work. */
+	unsigned int srcu_idx;			/* Current rdr array element. */
+	unsigned long srcu_gp_seq;		/* Grace-period seq #. */

include/linux/srcutree.h line 88

Quoted 10 lines

include/linux/srcutree.h +10 -0

+	unsigned long srcu_gp_seq_needed;	/* Latest gp_seq needed. */
+	atomic_t srcu_exp_cnt;			/* # ongoing expedited GPs. */
+	struct srcu_data __percpu *sda;		/* Per-CPU srcu_data array. */
+	unsigned long srcu_barrier_seq;		/* srcu_barrier seq #. */
+	struct mutex srcu_barrier_mutex;	/* Serialize barrier ops. */
+	struct completion srcu_barrier_completion;
+						/* Awaken barrier rq at end. */
+	atomic_t srcu_barrier_cpu_cnt;		/* # CPUs not yet posting a */
+						/*  callback for the barrier */
+						/*  operation. */

This change seems to have had the non-obvious effect of breaking the powerpc KVM build.

struct kvm {
	spinlock_t mmu_lock;
	struct mutex slots_lock;
	struct mm_struct *mm; /* userspace tied to this vm */
	struct kvm_memslots *memslots[KVM_ADDRESS_SPACE_NUM];
	struct srcu_struct srcu;
	struct srcu_struct irq_srcu;
        ...
	struct kvm_arch arch;

Example error:
arch/powerpc/kvm/book3s_hv_rmhandlers.S:617: Error: operand out of range (0x000000000000b328 is not between 0xffffffffffff8000 and 0x0000000000007fff)

Where line 617 is:
lwz r7,KVM_LPID(r9)

And the KVM_LPID constant comes from asm-offsets.s. The diff of old vs new is:

Quoted 6 lines

-->KVM_LPID 17752 offsetof(struct kvm, arch.lpid) #
+->KVM_LPID 45864 offsetof(struct kvm, arch.lpid) #
We can probably just fix it by changing the asm to keep the address of kvm_arch in a register, and then offset from that.

Michael, if you want to move the two srcu structs at the end of struct kvm, that would be fine by me. Please send a patch yourself so you can test it on PPC. Thanks,

Paolo

Paul E. McKenney

author

2017-04-21 00:38 UTC #61 lore raw

Re: powerpc KVM build break in linux-next (was Re: [PATCH tip/core/rcu 40/40] srcu: Parallelize callback handling)

Quoted 72 lines · Paolo Bonzini

Quoted 69 lines · Michael Ellerman
Quoted 5 lines · Paul E. McKenney
include/linux/srcutree.h +0 -0
index f2b3bd6c6bc2..0400e211aa44 100644
@@ -24,25 +24,75 @@
...
Quoted 18 lines
include/linux/srcutree.h +11 -6
+/*
+ * Per-SRCU-domain structure, similar in function to rcu_state.
+ */
 struct srcu_struct {
-	unsigned long completed;
-	unsigned long srcu_gp_seq;
-	atomic_t srcu_exp_cnt;
-	struct srcu_array __percpu *per_cpu_ref;
-	spinlock_t queue_lock; /* protect ->srcu_cblist */
-	struct rcu_segcblist srcu_cblist;
+	struct srcu_node node[NUM_RCU_NODES];	/* Combining tree. */
+	struct srcu_node *level[RCU_NUM_LVLS + 1];
+						/* First node at each level. */
+	struct mutex srcu_cb_mutex;		/* Serialize CB preparation. */
+	spinlock_t gp_lock;			/* protect ->srcu_cblist */
+	struct mutex srcu_gp_mutex;		/* Serialize GP work. */
+	unsigned int srcu_idx;			/* Current rdr array element. */
+	unsigned long srcu_gp_seq;		/* Grace-period seq #. */
include/linux/srcutree.h line 88
Quoted 10 lines
include/linux/srcutree.h +10 -0
+	unsigned long srcu_gp_seq_needed;	/* Latest gp_seq needed. */
+	atomic_t srcu_exp_cnt;			/* # ongoing expedited GPs. */
+	struct srcu_data __percpu *sda;		/* Per-CPU srcu_data array. */
+	unsigned long srcu_barrier_seq;		/* srcu_barrier seq #. */
+	struct mutex srcu_barrier_mutex;	/* Serialize barrier ops. */
+	struct completion srcu_barrier_completion;
+						/* Awaken barrier rq at end. */
+	atomic_t srcu_barrier_cpu_cnt;		/* # CPUs not yet posting a */
+						/*  callback for the barrier */
+						/*  operation. */
This change seems to have had the non-obvious effect of breaking the powerpc KVM build.
Because struct kvm contains two srcu_structs which are located before the kvm_arch struct, the increase in size of srcu_struct has caused the offset from the start of struct kvm to kvm_arch to be too big for some of our asm.
struct kvm {
	spinlock_t mmu_lock;
	struct mutex slots_lock;
	struct mm_struct *mm; /* userspace tied to this vm */
	struct kvm_memslots *memslots[KVM_ADDRESS_SPACE_NUM];
	struct srcu_struct srcu;
	struct srcu_struct irq_srcu;
        ...
	struct kvm_arch arch;
Example error:
arch/powerpc/kvm/book3s_hv_rmhandlers.S:617: Error: operand out of range (0x000000000000b328 is not between 0xffffffffffff8000 and 0x0000000000007fff)
Where line 617 is:
lwz r7,KVM_LPID(r9)
And the KVM_LPID constant comes from asm-offsets.s. The diff of old vs new is:
-->KVM_LPID 17752 offsetof(struct kvm, arch.lpid) #
+->KVM_LPID 45864 offsetof(struct kvm, arch.lpid) #

Quoted 6 lines

Quoted 2 lines
We can probably just fix it by changing the asm to keep the address of kvm_arch in a register, and then offset from that.
Michael, if you want to move the two srcu structs at the end of struct kvm, that would be fine by me. Please send a patch yourself so you can test it on PPC. Thanks,

On the off-chance that it is at all helpful, I have added the commit shown below to -rcu. If it helps, I am happy to push this, but am just as happy to drop it in favor of some other fix. If at all possible, I would like to get this into the upcoming merge window.

Thanx, Paul

Michael Ellerman

Tested-by

2017-04-21 01:42 UTC #62 lore raw

Re: powerpc KVM build break in linux-next (was Re: [PATCH tip/core/rcu 40/40] srcu: Parallelize callback handling)

Quoted 6 lines · Paul E. McKenney

Quoted 5 lines · Paolo Bonzini
Quoted 4 lines · Michael Ellerman
"Paul E. McKenney" <[email protected]> writes:
This change seems to have had the non-obvious effect of breaking the powerpc KVM build.

...

Quoted 7 lines

Quoted 4 lines
Michael, if you want to move the two srcu structs at the end of struct kvm, that would be fine by me. Please send a patch yourself so you can test it on PPC. Thanks,
On the off-chance that it is at all helpful, I have added the commit shown below to -rcu.

As shown below :)

Quoted 3 lines

If it helps, I am happy to push this, but am just as happy to drop it in favor of some other fix. If at all possible, I would like to get this into the upcoming merge window.

Thanks, this looks perfect to me, and if you're happy to put it on top of your tree that would limit the breakage to a smaller history window, so that would be ideal.

Tested-by Michael Ellerman <[email protected]>

cheers

Quoted 1 line

From a19a6617e2817e485ccc2f7cc5a97bd7ff769b87 Mon Sep 17 00:00:00 2001

From: "Paul E. McKenney" <[email protected]>
Date: Thu, 20 Apr 2017 17:30:06 -0700
Subject: kvm: Move srcu_struct fields to end of struct kvm

Parallelizing SRCU callback handling increased the size of srcu_struct, which moved the kvm_arch field within the kvm struct out of reach of powerpc's current assembly code, resulting in the following sort of build error:

arch/powerpc/kvm/book3s_hv_rmhandlers.S:617: Error: operand out of range (0x000000000000b328 is not between 0xffffffffffff8000 and 0x0000000000007fff)

This commit moves the srcu_struct fields in the kvm structure to follow the kvm_arch field, which again allows powerpc's assembly code to reach it.

Reported-by Stephen Rothwell <[email protected]>

Reported-by Michael Ellerman <[email protected]>

Reported-by kbuild test robot <[email protected]>

In reply to Paolo Bonzini

---
 include/linux/kvm_host.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

include/linux/kvm_host.h +2 -2

index 2c14ad9..96c8e29 100644
@@ -375,8 +375,6 @@ struct kvm {
 	struct mutex slots_lock;
 	struct mm_struct *mm; /* userspace tied to this vm */
 	struct kvm_memslots *memslots[KVM_ADDRESS_SPACE_NUM];
-	struct srcu_struct srcu;
-	struct srcu_struct irq_srcu;
 	struct kvm_vcpu *vcpus[KVM_MAX_VCPUS];
 
 	/*

@@ -429,6 +427,8 @@ struct kvm {
 	struct list_head devices;
 	struct dentry *debugfs_dentry;
 	struct kvm_stat_data **debugfs_stat_data;
+	struct srcu_struct srcu;
+	struct srcu_struct irq_srcu;
 };
 
 #define kvm_err(fmt, ...) \

Signature · 1 line

cgit v1.1

Paul E. McKenney

author

2017-04-21 04:17 UTC #63 lore raw

Re: powerpc KVM build break in linux-next (was Re: [PATCH tip/core/rcu 40/40] srcu: Parallelize callback handling)

Quoted 22 lines · Michael Ellerman

Quoted 6 lines · Paul E. McKenney
Quoted 5 lines · Paolo Bonzini
Quoted 4 lines · Michael Ellerman
"Paul E. McKenney" <[email protected]> writes:
This change seems to have had the non-obvious effect of breaking the powerpc KVM build.
...
Quoted 7 lines
Quoted 4 lines
Michael, if you want to move the two srcu structs at the end of struct kvm, that would be fine by me. Please send a patch yourself so you can test it on PPC. Thanks,
On the off-chance that it is at all helpful, I have added the commit shown below to -rcu.
As shown below :)
Quoted 3 lines
If it helps, I am happy to push this, but am just as happy to drop it in favor of some other fix. If at all possible, I would like to get this into the upcoming merge window.

Quoted 5 lines

Thanks, this looks perfect to me, and if you're happy to put it on top of your tree that would limit the breakage to a smaller history window, so that would be ideal.
Tested-by Michael Ellerman <[email protected]>

Thank you, Michael!

Paolo, does this look good to you? My upstream maintainer will be much more likely to take this with your ack. ;-)

Thanx, Paul

Quoted 52 lines

cheers
Quoted 1 line
From a19a6617e2817e485ccc2f7cc5a97bd7ff769b87 Mon Sep 17 00:00:00 2001
From: "Paul E. McKenney" <[email protected]>
Date: Thu, 20 Apr 2017 17:30:06 -0700
Subject: kvm: Move srcu_struct fields to end of struct kvm
Parallelizing SRCU callback handling increased the size of srcu_struct, which moved the kvm_arch field within the kvm struct out of reach of powerpc's current assembly code, resulting in the following sort of build error:
arch/powerpc/kvm/book3s_hv_rmhandlers.S:617: Error: operand out of range (0x000000000000b328 is not between 0xffffffffffff8000 and 0x0000000000007fff)
This commit moves the srcu_struct fields in the kvm structure to follow the kvm_arch field, which again allows powerpc's assembly code to reach it.
Reported-by Stephen Rothwell <[email protected]>
Reported-by Michael Ellerman <[email protected]>
Reported-by kbuild test robot <[email protected]>
Suggested-by Paolo Bonzini <[email protected]>
Signed-off-by Paul E. McKenney <[email protected]>
---
 include/linux/kvm_host.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
include/linux/kvm_host.h +2 -2
index 2c14ad9..96c8e29 100644
@@ -375,8 +375,6 @@ struct kvm {
 	struct mutex slots_lock;
 	struct mm_struct *mm; /* userspace tied to this vm */
 	struct kvm_memslots *memslots[KVM_ADDRESS_SPACE_NUM];
-	struct srcu_struct srcu;
-	struct srcu_struct irq_srcu;
 	struct kvm_vcpu *vcpus[KVM_MAX_VCPUS];
 
 	/*
@@ -429,6 +427,8 @@ struct kvm {
 	struct list_head devices;
 	struct dentry *debugfs_dentry;
 	struct kvm_stat_data **debugfs_stat_data;
+	struct srcu_struct srcu;
+	struct srcu_struct irq_srcu;
 };
 
 #define kvm_err(fmt, ...) \
Signature · 1 line
cgit v1.1

Paolo Bonzini

maintainer Acked-by

2017-04-21 07:27 UTC #64 lore raw

Re: powerpc KVM build break in linux-next (was Re: [PATCH tip/core/rcu 40/40] srcu: Parallelize callback handling)

Quoted 9 lines · Paul E. McKenney

Quoted 5 lines
Thanks, this looks perfect to me, and if you're happy to put it on top of your tree that would limit the breakage to a smaller history window, so that would be ideal.
Tested-by Michael Ellerman <[email protected]>
Thank you, Michael!
Paolo, does this look good to you? My upstream maintainer will be much more likely to take this with your ack. ;-)

Sure,

Acked-by Paolo Bonzini <[email protected]>

since I don't expect conflicts between whoever applies this patch and the KVM tree.

Paolo

Paul E. McKenney

author Applied

2017-04-21 12:51 UTC #65 lore raw

Re: powerpc KVM build break in linux-next (was Re: [PATCH tip/core/rcu 40/40] srcu: Parallelize callback handling)

Quoted 13 lines · Paolo Bonzini

Quoted 9 lines · Paul E. McKenney
Quoted 5 lines
Thanks, this looks perfect to me, and if you're happy to put it on top of your tree that would limit the breakage to a smaller history window, so that would be ideal.
Tested-by Michael Ellerman <[email protected]>
Thank you, Michael!
Paolo, does this look good to you? My upstream maintainer will be much more likely to take this with your ack. ;-)

Quoted 6 lines

Sure,
Acked-by Paolo Bonzini <[email protected]>
since I don't expect conflicts between whoever applies this patch and the KVM tree.

Applied, thank you!

Thanx, Paul

Michael Ellerman

2017-04-22 06:09 UTC #66 lore raw

Re: powerpc KVM build break in linux-next (was Re: [PATCH tip/core/rcu 40/40] srcu: Parallelize callback handling)

Quoted 12 lines · Paul E. McKenney

Quoted 11 lines · Paolo Bonzini
Quoted 9 lines · Paul E. McKenney
Quoted 5 lines
Thanks, this looks perfect to me, and if you're happy to put it on top of your tree that would limit the breakage to a smaller history window, so that would be ideal.
Tested-by Michael Ellerman <[email protected]>
Thank you, Michael!
Paolo, does this look good to you? My upstream maintainer will be much more likely to take this with your ack. ;-)

Quoted 8 lines

Quoted 6 lines
Sure,
Acked-by Paolo Bonzini <[email protected]>
since I don't expect conflicts between whoever applies this patch and the KVM tree.
Applied, thank you!

Thanks all.

cheers

Michael Ellerman

2017-04-21 02:13 UTC #67 lore raw

Re: powerpc KVM build break in linux-next (was Re: [PATCH tip/core/rcu 40/40] srcu: Parallelize callback handling)

Quoted 3 lines · Paolo Bonzini

Michael, if you want to move the two srcu structs at the end of struct kvm, that would be fine by me. Please send a patch yourself so you can test it on PPC. Thanks,

Paul beat me to it :)

I've confirmed it fixes the build break and the resulting kernel can boot a KVM guest happily.

cheers

Paul E. McKenney

author

2017-04-17 23:45 UTC #68 lore raw

v2 cover SRCU callback parallelization for 4.12

Hello!

This v2 series moves SRCU from its traditional single per-srcu_struct callback queue to per-srcu_struct/per-CPU callback queues. This involves abstracting functionality from Tree RCU, which results in a large conflict footprint, which in turn results in some otherwise unrelated patches coming along for the ride.

2. Make arch select smp_mb__after_unlock_lock() strength, which
gets rid of an arch-specific #ifdef.

3. Consolidate SRCU batch checking into rcu_all_batches_empty().

4. Check for tardy grace-period activity in cleanup_srcu_struct().

5-7. Semicolon inside RCU_TRACE() for various parts of RCU.

8. Pull rcu_sched_qs_mask into rcu_dynticks structure in order to
eliminate an isolated per-CPU variable.

9. Pull rcu_qs_ctr into rcu_dynticks structure.

10. Eliminate flavor scan in rcu_momentary_dyntick_idle() to
reduce semi-common-case context-switch overhead.

11. Place guard on rcu_all_qs() and rcu_note_context_switch()
actions to reduce common-case scheduler-fastpath overhead.

12. Default RCU_FANOUT_LEAF to 16 unless explicitly changed.

13. Abstract multi-tail callback list handling for SRCU.

14. Allow SRCU to access rcu_scheduler_active.

15. Allow early boot use of synchronize_srcu(), though not yet
mid-boot use.

16. Add single-element dequeue functions to rcu_segcblist for
debug use.

17. Move rcu_seq_start() and friends to rcu.h for SRCU's benefit.

18. Expedited wakeups need to be fully ordered.

19. Fix warning in rcu_seq_end().

20. Push srcu_advance_batches() fastpath into common case as a
step towards callback parallelization.

21. Move to state-based grace-period sequencing, also as a step
towards callback parallelization.

22. Add grace-period sequence numbers to SRCU.

23. Use rcu_segcblist to track SRCU callbacks.

24. Move combining-tree definitions for SRCU's benefit.

25. Move rcu_init_levelspread() to rcu_tree_node.h for SRCU's benefit.

26. Remove redundant levelcnt[] array from rcu_init_one().

27. Move rcu_node traversal macros to rcu.h for SRCU's benefit.

28. Make num_rcu_lvl[] array be external for SRCU's benefit.

29. Fix bogus try_check_zero() comment.

30. Improve rcu_seq grace-period-counter abstraction for SRCU's
benefit.

31. Allow a second bit in rcu_seq for SRCU state.

32. Merge ->srcu_state into ->srcu_gp_seq to allow atomic updates.

33. Provide crude control of expedited SRCU grace periods.

34. Use static initialization for "srcu" in mm/mmu_notifier.c.

35. Create a tiny SRCU for bloatwatch/tinification.

36. Print Tiny SRCU reader statistics in rcutorture.

37. Introduce CLASSIC_SRCU Kconfig option for those who do not
wish to help debug Tree SRCU.

38. Parallelize SRCU callback handling.

39. Make non-preemptive schedule be Tasks RCU quiescent state.

Updates since v1:

o Incorporate feedback from Peter Zijlstra.

o Dropped v1 patches 8-10 ("Make various parts of RCU do deferred
NOCB wakeups in order to prevent callback blockages, and thus hangs"). These patches turned out to be papering over a no-CBs CPU design flaw. There will be patches in v4.13 to fix the design flaw directly.

o Added v2 patch #34 ("Use static initialization for "srcu" in
mm/mmu_notifier.c"), moving it from its v1 location in the fixes series.

o Added v2 patch #39 ("Make non-preemptive schedule be Tasks RCU
quiescent state") for the benefit of upcoming ftrace work at Steve Rostedt's request.

Thanx, Paul

------------------------------------------------------------------------

 /kernel/rcu/rcu_segcblist.h                                     |  670 -----
 b/Documentation/RCU/Design/Data-Structures/Data-Structures.html |   36 
 b/arch/Kconfig                                                  |    3 
 b/arch/powerpc/Kconfig                                          |    1 
 b/include/linux/rcu_node_tree.h                                 |  105 
 b/include/linux/rcu_segcblist.h                                 |  720 +++++
 b/include/linux/rcupdate.h                                      |   17 
 b/include/linux/rcutiny.h                                       |   24 
 b/include/linux/rcutree.h                                       |    5 
 b/include/linux/srcu.h                                          |  112 
 b/include/linux/srcuclassic.h                                   |  101 
 b/include/linux/srcutiny.h                                      |   81 
 b/include/linux/srcutree.h                                      |  171 +
 b/init/Kconfig                                                  |   33 
 b/kernel/rcu/Makefile                                           |    6 
 b/kernel/rcu/rcu.h                                              |  165 +
 b/kernel/rcu/rcu_segcblist.h                                    |  670 +++++
 b/kernel/rcu/rcutorture.c                                       |   39 
 b/kernel/rcu/srcu.c                                             |  846 +++---
 b/kernel/rcu/srcutiny.c                                         |  215 +
 b/kernel/rcu/srcutree.c                                         | 1252 ++++++++--
 b/kernel/rcu/tiny.c                                             |   20 
 b/kernel/rcu/tiny_plugin.h                                      |   13 
 b/kernel/rcu/tree.c                                             |  657 ++---
 b/kernel/rcu/tree.h                                             |  174 -
 b/kernel/rcu/tree_exp.h                                         |   25 
 b/kernel/rcu/tree_plugin.h                                      |   62 
 b/kernel/rcu/tree_trace.c                                       |   26 
 b/kernel/rcu/update.c                                           |   53 
 b/kernel/sched/core.c                                           |    2 
 b/mm/mmu_notifier.c                                             |   14 
 31 files changed, 4287 insertions(+), 2031 deletions(-)

Paul E. McKenney

author

2017-04-17 23:53 UTC #69 lore raw

PATCH v2 1/39 rcu: Maintain special bits at bottom of ->dynticks counter

Reported-by Andy Lutomirski <[email protected]>

---
 include/linux/rcutiny.h |  5 ++++
 kernel/rcu/tree.c       | 73 +++++++++++++++++++++++++++++++++++++++----------
 kernel/rcu/tree.h       |  1 +
 3 files changed, 65 insertions(+), 14 deletions(-)

include/linux/rcutiny.h +5 -0

index b452953e21c8..6c9d941e3962 100644
@@ -33,6 +33,11 @@ static inline int rcu_dynticks_snap(struct rcu_dynticks *rdtp)
 	return 0;
 }
 
+static inline bool rcu_eqs_special_set(int cpu)
+{
+	return false;  /* Never flag non-existent other CPUs! */
+}
+
 static inline unsigned long get_state_synchronize_rcu(void)
 {
 	return 0;

kernel/rcu/tree.c +59 ~~-14~~

138 lines not on this page · raw message

kernel/rcu/tree.h +1 -0

index ec62a05bfdb3..7468b4de7e0c 100644
@@ -596,6 +596,7 @@ extern struct rcu_state rcu_preempt_state;
 #endif /* #ifdef CONFIG_PREEMPT_RCU */
 
 int rcu_dynticks_snap(struct rcu_dynticks *rdtp);
+bool rcu_eqs_special_set(int cpu);
 
 #ifdef CONFIG_RCU_BOOST
 DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_status);

Signature · 1 line

2.5.2

Josh Triplett

maintainer Reviewed-by on v2 1/39

2017-04-18 00:08 UTC #70 lore raw

Re: [PATCH v2 tip/core/rcu 01/39] rcu: Maintain special bits at bottom of ->dynticks counter

Quoted 24 lines · Paul E. McKenney

Currently, IPIs are used to force other CPUs to invalidate their TLBs in response to a kernel virtual-memory mapping change. This works, but degrades both battery lifetime (for idle CPUs) and real-time response (for nohz_full CPUs), and in addition results in unnecessary IPIs due to the fact that CPUs executing in usermode are unaffected by stale kernel mappings. It would be better to cause a CPU executing in usermode to wait until it is entering kernel mode to do the flush, first to avoid interrupting usemode tasks and second to handle multiple flush requests with a single flush in the case of a long-running user task.
This commit therefore reserves a bit at the bottom of the ->dynticks counter, which is checked upon exit from extended quiescent states. If it is set, it is cleared and then a new rcu_eqs_special_exit() macro is invoked, which, if not supplied, is an empty single-pass do-while loop. If this bottom bit is set on -entry- to an extended quiescent state, then a WARN_ON_ONCE() triggers.
This bottom bit may be set using a new rcu_eqs_special_set() function, which returns true if the bit was set, or false if the CPU turned out to not be in an extended quiescent state. Please note that this function refuses to set the bit for a non-nohz_full CPU when that CPU is executing in usermode because usermode execution is tracked by RCU as a dyntick-idle extended quiescent state only for nohz_full CPUs.

Quoted 2 lines

Reported-by Andy Lutomirski <[email protected]>
Signed-off-by Paul E. McKenney <[email protected]>

Comments below. With those fixed:

Quoted 7 lines

kernel/rcu/tree.c +0 -1

@@ -290,15 +300,20 @@ static DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = {
 static void rcu_dynticks_eqs_enter(void)
 {
 	struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks);
-	int special;

kernel/rcu/tree.c static DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = { · line 310

Quoted 10 lines

kernel/rcu/tree.c +2 -2

+	int seq;
 
 	/*
 	 * CPUs seeing atomic_inc_return() must see prior RCU read-side
 	 * critical sections, and we also must force ordering with the
 	 * next idle sojourn.
 	 */
-	special = atomic_inc_return(&rdtp->dynticks);
-	WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && special & 0x1);
+	seq = atomic_add_return(RCU_DYNTICK_CTRL_CTR, &rdtp->dynticks);

You changed atomic_inc_return to atomic_add_return here, but the comment above still says atomic_inc_return.

Quoted 5 lines

kernel/rcu/tree.c +0 -1

@@ -308,15 +323,22 @@ static void rcu_dynticks_eqs_enter(void)
 static void rcu_dynticks_eqs_exit(void)
 {
 	struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks);
-	int special;

kernel/rcu/tree.c static void rcu_dynticks_eqs_enter(void) · line 333

Quoted 10 lines

kernel/rcu/tree.c +2 -2

+	int seq;
 
 	/*
 	 * CPUs seeing atomic_inc_return() must see prior idle sojourns,
 	 * and we also must force ordering with the next RCU read-side
 	 * critical section.
 	 */
-	special = atomic_inc_return(&rdtp->dynticks);
-	WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && !(special & 0x1));
+	seq = atomic_add_return(RCU_DYNTICK_CTRL_CTR, &rdtp->dynticks);

Likewise.

- Josh Triplett

Paul E. McKenney

author

2017-04-18 18:19 UTC #71 lore raw

Re: [PATCH v2 tip/core/rcu 01/39] rcu: Maintain special bits at bottom of ->dynticks counter

Quoted 25 lines · Josh Triplett

Quoted 24 lines · Paul E. McKenney
Currently, IPIs are used to force other CPUs to invalidate their TLBs in response to a kernel virtual-memory mapping change. This works, but degrades both battery lifetime (for idle CPUs) and real-time response (for nohz_full CPUs), and in addition results in unnecessary IPIs due to the fact that CPUs executing in usermode are unaffected by stale kernel mappings. It would be better to cause a CPU executing in usermode to wait until it is entering kernel mode to do the flush, first to avoid interrupting usemode tasks and second to handle multiple flush requests with a single flush in the case of a long-running user task.
This commit therefore reserves a bit at the bottom of the ->dynticks counter, which is checked upon exit from extended quiescent states. If it is set, it is cleared and then a new rcu_eqs_special_exit() macro is invoked, which, if not supplied, is an empty single-pass do-while loop. If this bottom bit is set on -entry- to an extended quiescent state, then a WARN_ON_ONCE() triggers.
This bottom bit may be set using a new rcu_eqs_special_set() function, which returns true if the bit was set, or false if the CPU turned out to not be in an extended quiescent state. Please note that this function refuses to set the bit for a non-nohz_full CPU when that CPU is executing in usermode because usermode execution is tracked by RCU as a dyntick-idle extended quiescent state only for nohz_full CPUs.

Quoted 5 lines

Quoted 2 lines
Reported-by Andy Lutomirski <[email protected]>
Signed-off-by Paul E. McKenney <[email protected]>
Comments below. With those fixed:
Reviewed-by Josh Triplett <[email protected]>

Good eyes, fixed!

Thanx, Paul

Quoted 41 lines

Quoted 7 lines

kernel/rcu/tree.c +0 -1

@@ -290,15 +300,20 @@ static DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = {
 static void rcu_dynticks_eqs_enter(void)
 {
 	struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks);
-	int special;

kernel/rcu/tree.c static DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = { · line 310

Quoted 10 lines

kernel/rcu/tree.c +2 -2

+	int seq;
 
 	/*
 	 * CPUs seeing atomic_inc_return() must see prior RCU read-side
 	 * critical sections, and we also must force ordering with the
 	 * next idle sojourn.
 	 */
-	special = atomic_inc_return(&rdtp->dynticks);
-	WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && special & 0x1);
+	seq = atomic_add_return(RCU_DYNTICK_CTRL_CTR, &rdtp->dynticks);

You changed atomic_inc_return to atomic_add_return here, but the comment above still says atomic_inc_return.

Quoted 5 lines

kernel/rcu/tree.c +0 -1

@@ -308,15 +323,22 @@ static void rcu_dynticks_eqs_enter(void)
 static void rcu_dynticks_eqs_exit(void)
 {
 	struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks);
-	int special;

kernel/rcu/tree.c static void rcu_dynticks_eqs_enter(void) · line 333

Quoted 10 lines

kernel/rcu/tree.c +2 -2

+	int seq;
 
 	/*
 	 * CPUs seeing atomic_inc_return() must see prior idle sojourns,
 	 * and we also must force ordering with the next RCU read-side
 	 * critical section.
 	 */
-	special = atomic_inc_return(&rdtp->dynticks);
-	WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && !(special & 0x1));
+	seq = atomic_add_return(RCU_DYNTICK_CTRL_CTR, &rdtp->dynticks);

Likewise.

- Josh Triplett

Paul E. McKenney

author

2017-04-17 23:45 UTC #72 lore raw

PATCH v2 2/39 rcu: Make arch select smp_mb__after_unlock_lock() strength

The definition of smp_mb__after_unlock_lock() is currently smp_mb() for CONFIG_PPC and a no-op otherwise. It would be better to instead provide an architecture-selectable Kconfig option, and select the strength of smp_mb__after_unlock_lock() based on that option. This commit therefore creates ARCH_WEAK_RELEASE_ACQUIRE, has PPC select it, and bases the definition of smp_mb__after_unlock_lock() on this new ARCH_WEAK_RELEASE_ACQUIRE Kconfig option.

Reported-by Ingo Molnar <[email protected]>

Cc Peter Zijlstra <[email protected]>

Cc Will Deacon <[email protected]>

Cc Boqun Feng <[email protected]>

Cc Benjamin Herrenschmidt <[email protected]>

Cc Paul Mackerras <[email protected]>

Acked-by Michael Ellerman <[email protected]>

---
 arch/Kconfig             | 3 +++
 arch/powerpc/Kconfig     | 1 +
 include/linux/rcupdate.h | 6 +++---
 3 files changed, 7 insertions(+), 3 deletions(-)

arch/Kconfig +3 -0

index cd211a14a88f..adefaf344239 100644
@@ -320,6 +320,9 @@ config HAVE_CMPXCHG_LOCAL
 config HAVE_CMPXCHG_DOUBLE
 	bool
 
+config ARCH_WEAK_RELEASE_ACQUIRE
+	bool
+
 config ARCH_WANT_IPC_PARSE_VERSION
 	bool

arch/powerpc/Kconfig +1 -0

index 97a8bc8a095c..7a5c9b764cd2 100644
@@ -99,6 +99,7 @@ config PPC
 	select ARCH_USE_BUILTIN_BSWAP
 	select ARCH_USE_CMPXCHG_LOCKREF		if PPC64
 	select ARCH_WANT_IPC_PARSE_VERSION
+	select ARCH_WEAK_RELEASE_ACQUIRE
 	select BINFMT_ELF
 	select BUILDTIME_EXTABLE_SORT
 	select CLONE_BACKWARDS

include/linux/rcupdate.h +3 -3

index de88b33c0974..e6146d0074f8 100644
@@ -1127,11 +1127,11 @@ do { \
  * if the UNLOCK and LOCK are executed by the same CPU or if the
  * UNLOCK and LOCK operate on the same lock variable.
  */
-#ifdef CONFIG_PPC
+#ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE
 #define smp_mb__after_unlock_lock()	smp_mb()  /* Full ordering for lock. */
-#else /* #ifdef CONFIG_PPC */
+#else /* #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */
 #define smp_mb__after_unlock_lock()	do { } while (0)
-#endif /* #else #ifdef CONFIG_PPC */
+#endif /* #else #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */
 
 
 #endif /* __LINUX_RCUPDATE_H */

Signature · 1 line

2.5.2

Josh Triplett

maintainer Reviewed-by on v2 2/39

2017-04-18 00:19 UTC #73 lore raw

Re: [PATCH v2 tip/core/rcu 02/39] rcu: Make arch select smp_mb__after_unlock_lock() strength

Quoted 8 lines · Paul E. McKenney

The definition of smp_mb__after_unlock_lock() is currently smp_mb() for CONFIG_PPC and a no-op otherwise. It would be better to instead provide an architecture-selectable Kconfig option, and select the strength of smp_mb__after_unlock_lock() based on that option. This commit therefore creates ARCH_WEAK_RELEASE_ACQUIRE, has PPC select it, and bases the definition of smp_mb__after_unlock_lock() on this new ARCH_WEAK_RELEASE_ACQUIRE Kconfig option.

Quoted 9 lines

Reported-by Ingo Molnar <[email protected]>
Signed-off-by Paul E. McKenney <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Will Deacon <[email protected]>
Cc Boqun Feng <[email protected]>
Cc Benjamin Herrenschmidt <[email protected]>
Cc Paul Mackerras <[email protected]>
Acked-by Michael Ellerman <[email protected]>
Cc <[email protected]>

Seems sensible.

Quoted 53 lines

 arch/Kconfig             | 3 +++
 arch/powerpc/Kconfig     | 1 +
 include/linux/rcupdate.h | 6 +++---
 3 files changed, 7 insertions(+), 3 deletions(-)

arch/Kconfig +3 -0

index cd211a14a88f..adefaf344239 100644
@@ -320,6 +320,9 @@ config HAVE_CMPXCHG_LOCAL
 config HAVE_CMPXCHG_DOUBLE
 	bool
 
+config ARCH_WEAK_RELEASE_ACQUIRE
+	bool
+
 config ARCH_WANT_IPC_PARSE_VERSION
 	bool

arch/powerpc/Kconfig +1 -0

index 97a8bc8a095c..7a5c9b764cd2 100644
@@ -99,6 +99,7 @@ config PPC
 	select ARCH_USE_BUILTIN_BSWAP
 	select ARCH_USE_CMPXCHG_LOCKREF		if PPC64
 	select ARCH_WANT_IPC_PARSE_VERSION
+	select ARCH_WEAK_RELEASE_ACQUIRE
 	select BINFMT_ELF
 	select BUILDTIME_EXTABLE_SORT
 	select CLONE_BACKWARDS

include/linux/rcupdate.h +3 -3

index de88b33c0974..e6146d0074f8 100644
@@ -1127,11 +1127,11 @@ do { \
  * if the UNLOCK and LOCK are executed by the same CPU or if the
  * UNLOCK and LOCK operate on the same lock variable.
  */
-#ifdef CONFIG_PPC
+#ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE
 #define smp_mb__after_unlock_lock()	smp_mb()  /* Full ordering for lock. */
-#else /* #ifdef CONFIG_PPC */
+#else /* #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */
 #define smp_mb__after_unlock_lock()	do { } while (0)
-#endif /* #else #ifdef CONFIG_PPC */
+#endif /* #else #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */
 
 
 #endif /* __LINUX_RCUPDATE_H */

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-17 23:52 UTC #74 lore raw

PATCH v2 3/39 srcu: Consolidate batch checking into rcu_all_batches_empty()

---
 kernel/rcu/srcu.c | 21 +++++++++++++--------
 1 file changed, 13 insertions(+), 8 deletions(-)

kernel/rcu/srcu.c +13 -8

index ef3bcfb15b39..ba41a5d04b49 100644
@@ -65,6 +65,17 @@ static inline bool rcu_batch_empty(struct rcu_batch *b)
 }
 
 /*
+ * Are all batches empty for the specified srcu_struct?
+ */
+static inline bool rcu_all_batches_empty(struct srcu_struct *sp)
+{
+	return rcu_batch_empty(&sp->batch_done) &&
+	       rcu_batch_empty(&sp->batch_check1) &&
+	       rcu_batch_empty(&sp->batch_check0) &&
+	       rcu_batch_empty(&sp->batch_queue);
+}
+
+/*
  * Remove the callback at the head of the specified rcu_batch structure
  * and return a pointer to it, or return NULL if the structure is empty.
  */

@@ -619,15 +630,9 @@ static void srcu_reschedule(struct srcu_struct *sp)
 {
 	bool pending = true;
 
-	if (rcu_batch_empty(&sp->batch_done) &&
-	    rcu_batch_empty(&sp->batch_check1) &&
-	    rcu_batch_empty(&sp->batch_check0) &&
-	    rcu_batch_empty(&sp->batch_queue)) {
+	if (rcu_all_batches_empty(sp)) {
 		spin_lock_irq(&sp->queue_lock);
-		if (rcu_batch_empty(&sp->batch_done) &&
-		    rcu_batch_empty(&sp->batch_check1) &&
-		    rcu_batch_empty(&sp->batch_check0) &&
-		    rcu_batch_empty(&sp->batch_queue)) {
+		if (rcu_all_batches_empty(sp)) {
 			sp->running = false;
 			pending = false;
 		}

Signature · 1 line

2.5.2

Josh Triplett

maintainer Reviewed-by on v2 3/39

2017-04-18 00:30 UTC #75 lore raw

Re: [PATCH v2 tip/core/rcu 03/39] srcu: Consolidate batch checking into rcu_all_batches_empty()

Quoted 7 lines · Paul E. McKenney

The srcu_reschedule() function invokes rcu_batch_empty() on each of the four rcu_batch structures in the srcu_struct in question twice. Given that this check will also be needed in cleanup_srcu_struct(), this commit consolidates these four checks into a new rcu_all_batches_empty() function.
Signed-off-by Paul E. McKenney <[email protected]>

Quoted 46 lines

 kernel/rcu/srcu.c | 21 +++++++++++++--------
 1 file changed, 13 insertions(+), 8 deletions(-)

kernel/rcu/srcu.c +13 -8

index ef3bcfb15b39..ba41a5d04b49 100644
@@ -65,6 +65,17 @@ static inline bool rcu_batch_empty(struct rcu_batch *b)
 }
 
 /*
+ * Are all batches empty for the specified srcu_struct?
+ */
+static inline bool rcu_all_batches_empty(struct srcu_struct *sp)
+{
+	return rcu_batch_empty(&sp->batch_done) &&
+	       rcu_batch_empty(&sp->batch_check1) &&
+	       rcu_batch_empty(&sp->batch_check0) &&
+	       rcu_batch_empty(&sp->batch_queue);
+}
+
+/*
  * Remove the callback at the head of the specified rcu_batch structure
  * and return a pointer to it, or return NULL if the structure is empty.
  */

@@ -619,15 +630,9 @@ static void srcu_reschedule(struct srcu_struct *sp)
 {
 	bool pending = true;
 
-	if (rcu_batch_empty(&sp->batch_done) &&
-	    rcu_batch_empty(&sp->batch_check1) &&
-	    rcu_batch_empty(&sp->batch_check0) &&
-	    rcu_batch_empty(&sp->batch_queue)) {
+	if (rcu_all_batches_empty(sp)) {
 		spin_lock_irq(&sp->queue_lock);
-		if (rcu_batch_empty(&sp->batch_done) &&
-		    rcu_batch_empty(&sp->batch_check1) &&
-		    rcu_batch_empty(&sp->batch_check0) &&
-		    rcu_batch_empty(&sp->batch_queue)) {
+		if (rcu_all_batches_empty(sp)) {
 			sp->running = false;
 			pending = false;
 		}

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-17 23:54 UTC #76 lore raw

PATCH v2 4/39 srcu: Check for tardy grace-period activity in cleanup_srcu_struct()

Therefore, if there has ever been an invocation of call_srcu() on the srcu_struct in question, the sequence of events must be as follows:

1.  Prevent any further calls to call_srcu().
2.  Wait for any pre-existing call_srcu() invocations to return.
3.  Invoke srcu_barrier().
4.  It is now safe to invoke cleanup_srcu_struct().

On the other hand, if there has ever been a call to synchronize_srcu() or synchronize_srcu_expedited(), the sequence of events must be as follows:

Reported-by Paolo Bonzini <[email protected]>

---
 kernel/rcu/srcu.c | 5 +++++
 1 file changed, 5 insertions(+)

kernel/rcu/srcu.c +5 -0

index ba41a5d04b49..6beeba7b0b67 100644
@@ -261,6 +261,11 @@ void cleanup_srcu_struct(struct srcu_struct *sp)
 {
 	if (WARN_ON(srcu_readers_active(sp)))
 		return; /* Leakage unless caller handles error. */
+	if (WARN_ON(!rcu_all_batches_empty(sp)))
+		return; /* Leakage unless caller handles error. */
+	flush_delayed_work(&sp->work);
+	if (WARN_ON(sp->running))
+		return; /* Caller forgot to stop doing call_srcu()? */
 	free_percpu(sp->per_cpu_ref);
 	sp->per_cpu_ref = NULL;
 }

Signature · 1 line

2.5.2

Josh Triplett

maintainer Reviewed-by on v2 4/39

2017-04-18 00:33 UTC #77 lore raw

Re: [PATCH v2 tip/core/rcu 04/39] srcu: Check for tardy grace-period activity in cleanup_srcu_struct()

Quoted 28 lines · Paul E. McKenney

Users of SRCU are obliged to complete all grace-period activity before invoking cleanup_srcu_struct(). This means that all calls to either synchronize_srcu() or synchronize_srcu_expedited() must have returned, and all calls to call_srcu() must have returned, and the last call to call_srcu() must have been followed by a call to srcu_barrier(). Furthermore, the caller must have done something to prevent any further calls to synchronize_srcu(), synchronize_srcu_expedited(), and call_srcu().
Therefore, if there has ever been an invocation of call_srcu() on the srcu_struct in question, the sequence of events must be as follows:
1.  Prevent any further calls to call_srcu().
2.  Wait for any pre-existing call_srcu() invocations to return.
3.  Invoke srcu_barrier().
4.  It is now safe to invoke cleanup_srcu_struct().
On the other hand, if there has ever been a call to synchronize_srcu() or synchronize_srcu_expedited(), the sequence of events must be as follows:
1. Prevent any further calls to synchronize_srcu() or synchronize_srcu_expedited().
2. Wait for any pre-existing synchronize_srcu() or synchronize_srcu_expedited() invocations to return.
3. It is now safe to invoke cleanup_srcu_struct().

Quoted 5 lines

If there have been calls to all both types of functions (call_srcu() and either of synchronize_srcu() and synchronize_srcu_expedited()), then the caller must do the first three steps of the call_srcu() procedure above and the first two steps of the synchronize_s*() procedure above, and only then invoke cleanup_srcu_struct().

This commit message clearly explains the correct sequence for the client, but not which aspects of this the change now enforces. Some of the steps above remain the responsibility of the caller, while the callee now checks more of them. Could you add something at the end explaining the change and what it enforces?

Quoted 2 lines

Reported-by Paolo Bonzini <[email protected]>
Signed-off-by Paul E. McKenney <[email protected]>

With the above change:

Quoted 22 lines

 kernel/rcu/srcu.c | 5 +++++
 1 file changed, 5 insertions(+)

kernel/rcu/srcu.c +5 -0

index ba41a5d04b49..6beeba7b0b67 100644
@@ -261,6 +261,11 @@ void cleanup_srcu_struct(struct srcu_struct *sp)
 {
 	if (WARN_ON(srcu_readers_active(sp)))
 		return; /* Leakage unless caller handles error. */
+	if (WARN_ON(!rcu_all_batches_empty(sp)))
+		return; /* Leakage unless caller handles error. */
+	flush_delayed_work(&sp->work);
+	if (WARN_ON(sp->running))
+		return; /* Caller forgot to stop doing call_srcu()? */
 	free_percpu(sp->per_cpu_ref);
 	sp->per_cpu_ref = NULL;
 }

Signature · 1 line

2.5.2

Josh Triplett

maintainer

2017-04-18 00:34 UTC #78 lore raw

Re: [PATCH v2 tip/core/rcu 04/39] srcu: Check for tardy grace-period activity in cleanup_srcu_struct()

Quoted 35 lines · Josh Triplett

Quoted 28 lines · Paul E. McKenney
Users of SRCU are obliged to complete all grace-period activity before invoking cleanup_srcu_struct(). This means that all calls to either synchronize_srcu() or synchronize_srcu_expedited() must have returned, and all calls to call_srcu() must have returned, and the last call to call_srcu() must have been followed by a call to srcu_barrier(). Furthermore, the caller must have done something to prevent any further calls to synchronize_srcu(), synchronize_srcu_expedited(), and call_srcu().
Therefore, if there has ever been an invocation of call_srcu() on the srcu_struct in question, the sequence of events must be as follows:
1.  Prevent any further calls to call_srcu().
2.  Wait for any pre-existing call_srcu() invocations to return.
3.  Invoke srcu_barrier().
4.  It is now safe to invoke cleanup_srcu_struct().
On the other hand, if there has ever been a call to synchronize_srcu() or synchronize_srcu_expedited(), the sequence of events must be as follows:
1. Prevent any further calls to synchronize_srcu() or synchronize_srcu_expedited().
2. Wait for any pre-existing synchronize_srcu() or synchronize_srcu_expedited() invocations to return.
3. It is now safe to invoke cleanup_srcu_struct().
Quoted 5 lines
If there have been calls to all both types of functions (call_srcu() and either of synchronize_srcu() and synchronize_srcu_expedited()), then the caller must do the first three steps of the call_srcu() procedure above and the first two steps of the synchronize_s*() procedure above, and only then invoke cleanup_srcu_struct().

Quoted 5 lines

This commit message clearly explains the correct sequence for the client, but not which aspects of this the change now enforces. Some of the steps above remain the responsibility of the caller, while the callee now checks more of them. Could you add something at the end explaining the change and what it enforces?

More importantly, perhaps this explanation could find its way into the documentation of cleanup_srcu_struct?

Quoted 28 lines

Quoted 2 lines

Reported-by Paolo Bonzini <[email protected]>
Signed-off-by Paul E. McKenney <[email protected]>

With the above change:

Quoted 22 lines

 kernel/rcu/srcu.c | 5 +++++
 1 file changed, 5 insertions(+)

kernel/rcu/srcu.c +5 -0

index ba41a5d04b49..6beeba7b0b67 100644
@@ -261,6 +261,11 @@ void cleanup_srcu_struct(struct srcu_struct *sp)
 {
 	if (WARN_ON(srcu_readers_active(sp)))
 		return; /* Leakage unless caller handles error. */
+	if (WARN_ON(!rcu_all_batches_empty(sp)))
+		return; /* Leakage unless caller handles error. */
+	flush_delayed_work(&sp->work);
+	if (WARN_ON(sp->running))
+		return; /* Caller forgot to stop doing call_srcu()? */
 	free_percpu(sp->per_cpu_ref);
 	sp->per_cpu_ref = NULL;
 }

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-18 18:35 UTC #79 lore raw

Re: [PATCH v2 tip/core/rcu 04/39] srcu: Check for tardy grace-period activity in cleanup_srcu_struct()

Quoted 36 lines · Josh Triplett

Quoted 35 lines · Josh Triplett
Quoted 33 lines · Paul E. McKenney
Users of SRCU are obliged to complete all grace-period activity before invoking cleanup_srcu_struct(). This means that all calls to either synchronize_srcu() or synchronize_srcu_expedited() must have returned, and all calls to call_srcu() must have returned, and the last call to call_srcu() must have been followed by a call to srcu_barrier(). Furthermore, the caller must have done something to prevent any further calls to synchronize_srcu(), synchronize_srcu_expedited(), and call_srcu().
Therefore, if there has ever been an invocation of call_srcu() on the srcu_struct in question, the sequence of events must be as follows:
1.  Prevent any further calls to call_srcu().
2.  Wait for any pre-existing call_srcu() invocations to return.
3.  Invoke srcu_barrier().
4.  It is now safe to invoke cleanup_srcu_struct().
On the other hand, if there has ever been a call to synchronize_srcu() or synchronize_srcu_expedited(), the sequence of events must be as follows:
1. Prevent any further calls to synchronize_srcu() or synchronize_srcu_expedited().
2. Wait for any pre-existing synchronize_srcu() or synchronize_srcu_expedited() invocations to return.
3. It is now safe to invoke cleanup_srcu_struct().
If there have been calls to all both types of functions (call_srcu() and either of synchronize_srcu() and synchronize_srcu_expedited()), then the caller must do the first three steps of the call_srcu() procedure above and the first two steps of the synchronize_s*() procedure above, and only then invoke cleanup_srcu_struct().

Quoted 8 lines

Quoted 5 lines
This commit message clearly explains the correct sequence for the client, but not which aspects of this the change now enforces. Some of the steps above remain the responsibility of the caller, while the callee now checks more of them. Could you add something at the end explaining the change and what it enforces?
More importantly, perhaps this explanation could find its way into the documentation of cleanup_srcu_struct?

Like this?

/**
* cleanup_srcu_struct - deconstruct a sleep-RCU structure
* @sp: structure to clean up.
*
* Must invoke this only after you are finished using a given srcu_struct
* that was initialized via init_srcu_struct(). This code does some
* probabalistic checking, spotting late uses of srcu_read_lock(),
* synchronize_srcu(), synchronize_srcu_expedited(), and call_srcu().
* If any such late uses are detected, the per-CPU memory associated with
* the srcu_struct is simply leaked and WARN_ON() is invoked. If the
* caller frees the srcu_struct itself, a use-after-free crash will likely
* ensue, but at least there will be a warning printed.
*/

I added the following paragraph to the commit log:

Note that cleanup_srcu_struct() does some probabilistic checks for the caller failing to follow these procedures, in which case cleanup_srcu_struct() does WARN_ON() and avoids freeing the per-CPU structures associated with the specified srcu_struct structure.

And added your Reviewed-by, but please let me if more is needed.

Thanx, Paul

Quoted 29 lines

Quoted 28 lines

Quoted 2 lines

Reported-by Paolo Bonzini <[email protected]>
Signed-off-by Paul E. McKenney <[email protected]>

With the above change:

Quoted 22 lines

 kernel/rcu/srcu.c | 5 +++++
 1 file changed, 5 insertions(+)

kernel/rcu/srcu.c +5 -0

index ba41a5d04b49..6beeba7b0b67 100644
@@ -261,6 +261,11 @@ void cleanup_srcu_struct(struct srcu_struct *sp)
 {
 	if (WARN_ON(srcu_readers_active(sp)))
 		return; /* Leakage unless caller handles error. */
+	if (WARN_ON(!rcu_all_batches_empty(sp)))
+		return; /* Leakage unless caller handles error. */
+	flush_delayed_work(&sp->work);
+	if (WARN_ON(sp->running))
+		return; /* Caller forgot to stop doing call_srcu()? */
 	free_percpu(sp->per_cpu_ref);
 	sp->per_cpu_ref = NULL;
 }

Signature · 1 line

2.5.2

Josh Triplett

maintainer

2017-04-18 19:45 UTC #80 lore raw

Re: [PATCH v2 tip/core/rcu 04/39] srcu: Check for tardy grace-period activity in cleanup_srcu_struct()

Quoted 64 lines · Paul E. McKenney

Quoted 36 lines · Josh Triplett
Quoted 35 lines · Josh Triplett
Quoted 33 lines · Paul E. McKenney
Users of SRCU are obliged to complete all grace-period activity before invoking cleanup_srcu_struct(). This means that all calls to either synchronize_srcu() or synchronize_srcu_expedited() must have returned, and all calls to call_srcu() must have returned, and the last call to call_srcu() must have been followed by a call to srcu_barrier(). Furthermore, the caller must have done something to prevent any further calls to synchronize_srcu(), synchronize_srcu_expedited(), and call_srcu().
Therefore, if there has ever been an invocation of call_srcu() on the srcu_struct in question, the sequence of events must be as follows:
1.  Prevent any further calls to call_srcu().
2.  Wait for any pre-existing call_srcu() invocations to return.
3.  Invoke srcu_barrier().
4.  It is now safe to invoke cleanup_srcu_struct().
On the other hand, if there has ever been a call to synchronize_srcu() or synchronize_srcu_expedited(), the sequence of events must be as follows:
1. Prevent any further calls to synchronize_srcu() or synchronize_srcu_expedited().
2. Wait for any pre-existing synchronize_srcu() or synchronize_srcu_expedited() invocations to return.
3. It is now safe to invoke cleanup_srcu_struct().
If there have been calls to all both types of functions (call_srcu() and either of synchronize_srcu() and synchronize_srcu_expedited()), then the caller must do the first three steps of the call_srcu() procedure above and the first two steps of the synchronize_s*() procedure above, and only then invoke cleanup_srcu_struct().
Quoted 8 lines
Quoted 5 lines
This commit message clearly explains the correct sequence for the client, but not which aspects of this the change now enforces. Some of the steps above remain the responsibility of the caller, while the callee now checks more of them. Could you add something at the end explaining the change and what it enforces?
More importantly, perhaps this explanation could find its way into the documentation of cleanup_srcu_struct?
Like this?
/**
* cleanup_srcu_struct - deconstruct a sleep-RCU structure
* @sp: structure to clean up.
*
* Must invoke this only after you are finished using a given srcu_struct
* that was initialized via init_srcu_struct(). This code does some
* probabalistic checking, spotting late uses of srcu_read_lock(),
* synchronize_srcu(), synchronize_srcu_expedited(), and call_srcu().
* If any such late uses are detected, the per-CPU memory associated with
* the srcu_struct is simply leaked and WARN_ON() is invoked. If the
* caller frees the srcu_struct itself, a use-after-free crash will likely
* ensue, but at least there will be a warning printed.
*/
I added the following paragraph to the commit log:

Quoted 7 lines

Note that cleanup_srcu_struct() does some probabilistic checks for the caller failing to follow these procedures, in which case cleanup_srcu_struct() does WARN_ON() and avoids freeing the per-CPU structures associated with the specified srcu_struct structure.
And added your Reviewed-by, but please let me if more is needed.

Looks good to me, thanks for improving the documentation.

Paul E. McKenney

author

2017-04-17 23:45 UTC #81 lore raw

PATCH v2 5/39 rcu: Semicolon inside RCU_TRACE() for rcu.h

---
 kernel/rcu/rcu.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

kernel/rcu/rcu.h +2 -2

index 0d6ff3e471be..8700a81daf56 100644
@@ -109,12 +109,12 @@ static inline bool __rcu_reclaim(const char *rn, struct rcu_head *head)
 
 	rcu_lock_acquire(&rcu_callback_map);
 	if (__is_kfree_rcu_offset(offset)) {
-		RCU_TRACE(trace_rcu_invoke_kfree_callback(rn, head, offset));
+		RCU_TRACE(trace_rcu_invoke_kfree_callback(rn, head, offset);)
 		kfree((void *)head - offset);
 		rcu_lock_release(&rcu_callback_map);
 		return true;
 	} else {
-		RCU_TRACE(trace_rcu_invoke_callback(rn, head));
+		RCU_TRACE(trace_rcu_invoke_callback(rn, head);)
 		head->func(head);
 		rcu_lock_release(&rcu_callback_map);
 		return false;

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-17 23:46 UTC #82 lore raw

PATCH v2 6/39 rcu: Semicolon inside RCU_TRACE() for Tiny RCU

---
 kernel/rcu/tiny.c        | 20 ++++++++++----------
 kernel/rcu/tiny_plugin.h |  4 ++--
 2 files changed, 12 insertions(+), 12 deletions(-)

kernel/rcu/tiny.c +10 ~~-10~~

index 6ad330dbbae2..e5385731e391 100644
@@ -79,7 +79,7 @@ EXPORT_SYMBOL(__rcu_is_watching);
  */
 static int rcu_qsctr_help(struct rcu_ctrlblk *rcp)
 {
-	RCU_TRACE(reset_cpu_stall_ticks(rcp));
+	RCU_TRACE(reset_cpu_stall_ticks(rcp);)
 	if (rcp->donetail != rcp->curtail) {
 		rcp->donetail = rcp->curtail;
 		return 1;

@@ -125,7 +125,7 @@ void rcu_bh_qs(void)
  */
 void rcu_check_callbacks(int user)
 {
-	RCU_TRACE(check_cpu_stalls());
+	RCU_TRACE(check_cpu_stalls();)
 	if (user)
 		rcu_sched_qs();
 	else if (!in_softirq())

@@ -143,7 +143,7 @@ static void __rcu_process_callbacks(struct rcu_ctrlblk *rcp)
 	const char *rn = NULL;
 	struct rcu_head *next, *list;
 	unsigned long flags;
-	RCU_TRACE(int cb_count = 0);
+	RCU_TRACE(int cb_count = 0;)
 
 	/* Move the ready-to-invoke callbacks to a local list. */
 	local_irq_save(flags);

@@ -152,7 +152,7 @@ static void __rcu_process_callbacks(struct rcu_ctrlblk *rcp)
 		local_irq_restore(flags);
 		return;
 	}
-	RCU_TRACE(trace_rcu_batch_start(rcp->name, 0, rcp->qlen, -1));
+	RCU_TRACE(trace_rcu_batch_start(rcp->name, 0, rcp->qlen, -1);)
 	list = rcp->rcucblist;
 	rcp->rcucblist = *rcp->donetail;
 	*rcp->donetail = NULL;

@@ -162,7 +162,7 @@ static void __rcu_process_callbacks(struct rcu_ctrlblk *rcp)
 	local_irq_restore(flags);
 
 	/* Invoke the callbacks on the local list. */
-	RCU_TRACE(rn = rcp->name);
+	RCU_TRACE(rn = rcp->name;)
 	while (list) {
 		next = list->next;
 		prefetch(next);

@@ -171,9 +171,9 @@ static void __rcu_process_callbacks(struct rcu_ctrlblk *rcp)
 		__rcu_reclaim(rn, list);
 		local_bh_enable();
 		list = next;
-		RCU_TRACE(cb_count++);
+		RCU_TRACE(cb_count++;)
 	}
-	RCU_TRACE(rcu_trace_sub_qlen(rcp, cb_count));
+	RCU_TRACE(rcu_trace_sub_qlen(rcp, cb_count);)
 	RCU_TRACE(trace_rcu_batch_end(rcp->name,
 				      cb_count, 0, need_resched(),
 				      is_idle_task(current),

@@ -221,7 +221,7 @@ static void __call_rcu(struct rcu_head *head,
 	local_irq_save(flags);
 	*rcp->curtail = head;
 	rcp->curtail = &head->next;
-	RCU_TRACE(rcp->qlen++);
+	RCU_TRACE(rcp->qlen++;)
 	local_irq_restore(flags);
 
 	if (unlikely(is_idle_task(current))) {

@@ -254,8 +254,8 @@ EXPORT_SYMBOL_GPL(call_rcu_bh);
 void __init rcu_init(void)
 {
 	open_softirq(RCU_SOFTIRQ, rcu_process_callbacks);
-	RCU_TRACE(reset_cpu_stall_ticks(&rcu_sched_ctrlblk));
-	RCU_TRACE(reset_cpu_stall_ticks(&rcu_bh_ctrlblk));
+	RCU_TRACE(reset_cpu_stall_ticks(&rcu_sched_ctrlblk);)
+	RCU_TRACE(reset_cpu_stall_ticks(&rcu_bh_ctrlblk);)
 
 	rcu_early_boot_tests();
 }

kernel/rcu/tiny_plugin.h +2 -2

index c64b827ecbca..df3a60e19f07 100644
@@ -162,8 +162,8 @@ static void reset_cpu_stall_ticks(struct rcu_ctrlblk *rcp)
 
 static void check_cpu_stalls(void)
 {
-	RCU_TRACE(check_cpu_stall(&rcu_bh_ctrlblk));
-	RCU_TRACE(check_cpu_stall(&rcu_sched_ctrlblk));
+	RCU_TRACE(check_cpu_stall(&rcu_bh_ctrlblk);)
+	RCU_TRACE(check_cpu_stall(&rcu_sched_ctrlblk);)
 }
 
 #endif /* #ifdef CONFIG_RCU_TRACE */

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-17 23:48 UTC #83 lore raw

PATCH v2 7/39 rcu: Semicolon inside RCU_TRACE() for tree.c

---
 kernel/rcu/tree.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

kernel/rcu/tree.c +4 -4

index df8f6401f2d5..e8843bf0f5e1 100644
@@ -2793,14 +2793,14 @@ static void rcu_adopt_orphan_cbs(struct rcu_state *rsp, unsigned long flags)
  */
 static void rcu_cleanup_dying_cpu(struct rcu_state *rsp)
 {
-	RCU_TRACE(unsigned long mask);
-	RCU_TRACE(struct rcu_data *rdp = this_cpu_ptr(rsp->rda));
-	RCU_TRACE(struct rcu_node *rnp = rdp->mynode);
+	RCU_TRACE(unsigned long mask;)
+	RCU_TRACE(struct rcu_data *rdp = this_cpu_ptr(rsp->rda);)
+	RCU_TRACE(struct rcu_node *rnp = rdp->mynode;)
 
 	if (!IS_ENABLED(CONFIG_HOTPLUG_CPU))
 		return;
 
-	RCU_TRACE(mask = rdp->grpmask);
+	RCU_TRACE(mask = rdp->grpmask;)
 	trace_rcu_grace_period(rsp->name,
 			       rnp->gpnum + 1 - !!(rnp->qsmask & mask),
 			       TPS("cpuofl"));

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-17 23:51 UTC #84 lore raw

PATCH v2 8/39 rcu: Pull rcu_sched_qs_mask into rcu_dynticks structure

The rcu_sched_qs_mask variable is yet another isolated per-CPU variable, so this commit pulls it into the pre-existing rcu_dynticks per-CPU structure.

---
 .../RCU/Design/Data-Structures/Data-Structures.html          |  9 ++++++++-
 kernel/rcu/tree.c                                            | 12 +++++-------
 kernel/rcu/tree.h                                            |  1 +
 3 files changed, 14 insertions(+), 8 deletions(-)

Documentation/RCU/Design/Data-Structures/Data-Structures.html +8 -1

index d583c653a703..bf7f266e8888 100644
@@ -1104,6 +1104,7 @@ Its fields are as follows:
   1   int dynticks_nesting;
   2   int dynticks_nmi_nesting;
   3   atomic_t dynticks;
+  4   int rcu_sched_qs_mask;
 </pre>
 
 <p>The <tt>-&gt;dynticks_nesting</tt> field counts the

@@ -1117,11 +1118,17 @@ NMIs are counted by the <tt>-&gt;dynticks_nmi_nesting</tt>
 field, except that NMIs that interrupt non-dyntick-idle execution
 are not counted.
 
-</p><p>Finally, the <tt>-&gt;dynticks</tt> field counts the corresponding
+</p><p>The <tt>-&gt;dynticks</tt> field counts the corresponding
 CPU's transitions to and from dyntick-idle mode, so that this counter
 has an even value when the CPU is in dyntick-idle mode and an odd
 value otherwise.
 
+</p><p>Finally, the  <tt>-&gt;rcu_sched_qs_mask</tt> field is used
+to record the fact that the RCU core code would really like to
+see a quiescent state from the corresponding CPU.
+This flag is checked by RCU's context-switch and <tt>cond_resched()</tt>
+code, which provide a momentary idle sojourn in response.
+
 <table>
 <tr><th>&nbsp;</th></tr>
 <tr><th align="left">Quick Quiz:</th></tr>

kernel/rcu/tree.c +5 -7

index e8843bf0f5e1..0e8d91f36bb6 100644
@@ -272,8 +272,6 @@ void rcu_bh_qs(void)
 	}
 }
 
-static DEFINE_PER_CPU(int, rcu_sched_qs_mask);
-
 /*
  * Steal a bit from the bottom of ->dynticks for idle entry/exit
  * control.  Initially this is for TLB flushing.

@@ -464,8 +462,8 @@ static void rcu_momentary_dyntick_idle(void)
 	 * Yes, we can lose flag-setting operations.  This is OK, because
 	 * the flag will be set again after some delay.
 	 */
-	resched_mask = raw_cpu_read(rcu_sched_qs_mask);
-	raw_cpu_write(rcu_sched_qs_mask, 0);
+	resched_mask = raw_cpu_read(rcu_dynticks.rcu_sched_qs_mask);
+	raw_cpu_write(rcu_dynticks.rcu_sched_qs_mask, 0);
 
 	/* Find the flavor that needs a quiescent state. */
 	for_each_rcu_flavor(rsp) {

@@ -499,7 +497,7 @@ void rcu_note_context_switch(void)
 	trace_rcu_utilization(TPS("Start context switch"));
 	rcu_sched_qs();
 	rcu_preempt_note_context_switch();
-	if (unlikely(raw_cpu_read(rcu_sched_qs_mask)))
+	if (unlikely(raw_cpu_read(rcu_dynticks.rcu_sched_qs_mask)))
 		rcu_momentary_dyntick_idle();
 	trace_rcu_utilization(TPS("End context switch"));
 	barrier(); /* Avoid RCU read-side critical sections leaking up. */

@@ -524,7 +522,7 @@ void rcu_all_qs(void)
 	unsigned long flags;
 
 	barrier(); /* Avoid RCU read-side critical sections leaking down. */
-	if (unlikely(raw_cpu_read(rcu_sched_qs_mask))) {
+	if (unlikely(raw_cpu_read(rcu_dynticks.rcu_sched_qs_mask))) {
 		local_irq_save(flags);
 		rcu_momentary_dyntick_idle();
 		local_irq_restore(flags);

@@ -1351,7 +1349,7 @@ static int rcu_implicit_dynticks_qs(struct rcu_data *rdp,
 	 * is set too high, we override with half of the RCU CPU stall
 	 * warning delay.
 	 */
-	rcrmp = &per_cpu(rcu_sched_qs_mask, rdp->cpu);
+	rcrmp = &per_cpu(rcu_dynticks.rcu_sched_qs_mask, rdp->cpu);
 	if (time_after(jiffies, rdp->rsp->gp_start + jtsq) ||
 	    time_after(jiffies, rdp->rsp->jiffies_resched)) {
 		if (!(READ_ONCE(*rcrmp) & rdp->rsp->flavor_mask)) {

kernel/rcu/tree.h +1 -0

index 7468b4de7e0c..e298281984dc 100644
@@ -113,6 +113,7 @@ struct rcu_dynticks {
 				    /* Process level is worth LLONG_MAX/2. */
 	int dynticks_nmi_nesting;   /* Track NMI nesting level. */
 	atomic_t dynticks;	    /* Even value for idle, else odd. */
+	int rcu_sched_qs_mask;      /* GP old, need quiescent state. */
 #ifdef CONFIG_NO_HZ_FULL_SYSIDLE
 	long long dynticks_idle_nesting;
 				    /* irq/process nesting level from idle. */

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-17 23:54 UTC #85 lore raw

PATCH v2 9/39 rcu: Pull rcu_qs_ctr into rcu_dynticks structure

The rcu_qs_ctr variable is yet another isolated per-CPU variable, so this commit pulls it into the pre-existing rcu_dynticks per-CPU structure.

---
 .../RCU/Design/Data-Structures/Data-Structures.html       | 12 ++++++++++--
 kernel/rcu/tree.c                                         | 15 ++++++---------
 kernel/rcu/tree.h                                         |  3 ++-
 kernel/rcu/tree_trace.c                                   |  4 +---
 4 files changed, 19 insertions(+), 15 deletions(-)

Documentation/RCU/Design/Data-Structures/Data-Structures.html +10 -2

index bf7f266e8888..3d0311657533 100644
@@ -1105,6 +1105,7 @@ Its fields are as follows:
   2   int dynticks_nmi_nesting;
   3   atomic_t dynticks;
   4   int rcu_sched_qs_mask;
+  5   unsigned long rcu_qs_ctr;
 </pre>
 
 <p>The <tt>-&gt;dynticks_nesting</tt> field counts the

@@ -1123,12 +1124,19 @@ CPU's transitions to and from dyntick-idle mode, so that this counter
 has an even value when the CPU is in dyntick-idle mode and an odd
 value otherwise.
 
-</p><p>Finally, the  <tt>-&gt;rcu_sched_qs_mask</tt> field is used
+</p><p>The <tt>-&gt;rcu_sched_qs_mask</tt> field is used
 to record the fact that the RCU core code would really like to
-see a quiescent state from the corresponding CPU.
+see a quiescent state from the corresponding CPU, so much so that
+it is willing to call for heavy-weight dyntick-counter operations.
 This flag is checked by RCU's context-switch and <tt>cond_resched()</tt>
 code, which provide a momentary idle sojourn in response.
 
+</p><p>Finally the <tt>-&gt;rcu_qs_ctr</tt> field is used to record
+quiescent states from <tt>cond_resched()</tt>.
+Because <tt>cond_resched()</tt> can execute quite frequently, this
+must be quite lightweight, as in a non-atomic increment of this
+per-CPU field.
+
 <table>
 <tr><th>&nbsp;</th></tr>
 <tr><th align="left">Quick Quiz:</th></tr>

kernel/rcu/tree.c +6 -9

index 0e8d91f36bb6..fbee1d729c4b 100644
@@ -436,9 +436,6 @@ bool rcu_eqs_special_set(int cpu)
 	return true;
 }
 
-DEFINE_PER_CPU_SHARED_ALIGNED(unsigned long, rcu_qs_ctr);
-EXPORT_PER_CPU_SYMBOL_GPL(rcu_qs_ctr);
-
 /*
  * Let the RCU core know that this CPU has gone through the scheduler,
  * which is a quiescent state.  This is called when the need for a

@@ -542,7 +539,7 @@ void rcu_all_qs(void)
 		rcu_sched_qs();
 		preempt_enable();
 	}
-	this_cpu_inc(rcu_qs_ctr);
+	this_cpu_inc(rcu_dynticks.rcu_qs_ctr);
 	barrier(); /* Avoid RCU read-side critical sections leaking up. */
 }
 EXPORT_SYMBOL_GPL(rcu_all_qs);

@@ -1315,7 +1312,7 @@ static int rcu_implicit_dynticks_qs(struct rcu_data *rdp,
 	 */
 	rnp = rdp->mynode;
 	if (time_after(jiffies, rdp->rsp->gp_start + jtsq) &&
-	    READ_ONCE(rdp->rcu_qs_ctr_snap) != per_cpu(rcu_qs_ctr, rdp->cpu) &&
+	    READ_ONCE(rdp->rcu_qs_ctr_snap) != per_cpu(rcu_dynticks.rcu_qs_ctr, rdp->cpu) &&
 	    READ_ONCE(rdp->gpnum) == rnp->gpnum && !rdp->gpwrap) {
 		trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, TPS("rqc"));
 		return 1;

@@ -2024,7 +2021,7 @@ static bool __note_gp_changes(struct rcu_state *rsp, struct rcu_node *rnp,
 		trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("cpustart"));
 		need_gp = !!(rnp->qsmask & rdp->grpmask);
 		rdp->cpu_no_qs.b.norm = need_gp;
-		rdp->rcu_qs_ctr_snap = __this_cpu_read(rcu_qs_ctr);
+		rdp->rcu_qs_ctr_snap = __this_cpu_read(rcu_dynticks.rcu_qs_ctr);
 		rdp->core_needs_qs = need_gp;
 		zero_cpu_stall_ticks(rdp);
 		WRITE_ONCE(rdp->gpwrap, false);

@@ -2622,7 +2619,7 @@ rcu_report_qs_rdp(int cpu, struct rcu_state *rsp, struct rcu_data *rdp)
 		 * within the current grace period.
 		 */
 		rdp->cpu_no_qs.b.norm = true;	/* need qs for new gp. */
-		rdp->rcu_qs_ctr_snap = __this_cpu_read(rcu_qs_ctr);
+		rdp->rcu_qs_ctr_snap = __this_cpu_read(rcu_dynticks.rcu_qs_ctr);
 		raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
 		return;
 	}

@@ -3620,7 +3617,7 @@ static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp)
 	/* Is the RCU core waiting for a quiescent state from this CPU? */
 	if (rcu_scheduler_fully_active &&
 	    rdp->core_needs_qs && rdp->cpu_no_qs.b.norm &&
-	    rdp->rcu_qs_ctr_snap == __this_cpu_read(rcu_qs_ctr)) {
+	    rdp->rcu_qs_ctr_snap == __this_cpu_read(rcu_dynticks.rcu_qs_ctr)) {
 		rdp->n_rp_core_needs_qs++;
 	} else if (rdp->core_needs_qs && !rdp->cpu_no_qs.b.norm) {
 		rdp->n_rp_report_qs++;

@@ -3933,7 +3930,7 @@ rcu_init_percpu_data(int cpu, struct rcu_state *rsp)
 	rdp->gpnum = rnp->completed; /* Make CPU later note any new GP. */
 	rdp->completed = rnp->completed;
 	rdp->cpu_no_qs.b.norm = true;
-	rdp->rcu_qs_ctr_snap = per_cpu(rcu_qs_ctr, cpu);
+	rdp->rcu_qs_ctr_snap = per_cpu(rcu_dynticks.rcu_qs_ctr, cpu);
 	rdp->core_needs_qs = false;
 	trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("cpuonl"));
 	raw_spin_unlock_irqrestore_rcu_node(rnp, flags);

kernel/rcu/tree.h +2 -1

index e298281984dc..76e4467bc765 100644
@@ -113,7 +113,8 @@ struct rcu_dynticks {
 				    /* Process level is worth LLONG_MAX/2. */
 	int dynticks_nmi_nesting;   /* Track NMI nesting level. */
 	atomic_t dynticks;	    /* Even value for idle, else odd. */
-	int rcu_sched_qs_mask;      /* GP old, need quiescent state. */
+	int rcu_sched_qs_mask;      /* GP old, need heavy quiescent state. */
+	unsigned long rcu_qs_ctr;   /* Light universal quiescent state ctr. */
 #ifdef CONFIG_NO_HZ_FULL_SYSIDLE
 	long long dynticks_idle_nesting;
 				    /* irq/process nesting level from idle. */

kernel/rcu/tree_trace.c +1 -3

index 8751a748499a..65b43be38e68 100644
@@ -45,8 +45,6 @@
 #define RCU_TREE_NONCORE
 #include "tree.h"
 
-DECLARE_PER_CPU_SHARED_ALIGNED(unsigned long, rcu_qs_ctr);
-
 static int r_open(struct inode *inode, struct file *file,
 					const struct seq_operations *op)
 {

@@ -121,7 +119,7 @@ static void print_one_rcu_data(struct seq_file *m, struct rcu_data *rdp)
 		   cpu_is_offline(rdp->cpu) ? '!' : ' ',
 		   ulong2long(rdp->completed), ulong2long(rdp->gpnum),
 		   rdp->cpu_no_qs.b.norm,
-		   rdp->rcu_qs_ctr_snap == per_cpu(rcu_qs_ctr, rdp->cpu),
+		   rdp->rcu_qs_ctr_snap == per_cpu(rdp->dynticks->rcu_qs_ctr, rdp->cpu),
 		   rdp->core_needs_qs);
 	seq_printf(m, " dt=%d/%llx/%d df=%lu",
 		   rcu_dynticks_snap(rdp->dynticks),

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-17 23:48 UTC #86 lore raw

PATCH v2 10/39 rcu: Eliminate flavor scan in rcu_momentary_dyntick_idle()

---
 .../Design/Data-Structures/Data-Structures.html    |  4 +-
 kernel/rcu/tree.c                                  | 62 +++++-----------------
 kernel/rcu/tree.h                                  |  3 +-
 3 files changed, 15 insertions(+), 54 deletions(-)

Documentation/RCU/Design/Data-Structures/Data-Structures.html +2 -2

index 3d0311657533..e4bf20a68fa3 100644
@@ -1104,7 +1104,7 @@ Its fields are as follows:
   1   int dynticks_nesting;
   2   int dynticks_nmi_nesting;
   3   atomic_t dynticks;
-  4   int rcu_sched_qs_mask;
+  4   bool rcu_need_heavy_qs;
   5   unsigned long rcu_qs_ctr;
 </pre>

@@ -1124,7 +1124,7 @@ CPU's transitions to and from dyntick-idle mode, so that this counter
 has an even value when the CPU is in dyntick-idle mode and an odd
 value otherwise.
 
-</p><p>The <tt>-&gt;rcu_sched_qs_mask</tt> field is used
+</p><p>The <tt>-&gt;rcu_need_heavy_qs</tt> field is used
 to record the fact that the RCU core code would really like to
 see a quiescent state from the corresponding CPU, so much so that
 it is willing to call for heavy-weight dyntick-counter operations.

kernel/rcu/tree.c +12 ~~-50~~

115 lines not on this page · raw message

kernel/rcu/tree.h +1 -2

index 76e4467bc765..b212cd0f22c7 100644
@@ -113,7 +113,7 @@ struct rcu_dynticks {
 				    /* Process level is worth LLONG_MAX/2. */
 	int dynticks_nmi_nesting;   /* Track NMI nesting level. */
 	atomic_t dynticks;	    /* Even value for idle, else odd. */
-	int rcu_sched_qs_mask;      /* GP old, need heavy quiescent state. */
+	bool rcu_need_heavy_qs;      /* GP old, need heavy quiescent state. */
 	unsigned long rcu_qs_ctr;   /* Light universal quiescent state ctr. */
 #ifdef CONFIG_NO_HZ_FULL_SYSIDLE
 	long long dynticks_idle_nesting;

@@ -484,7 +484,6 @@ struct rcu_state {
 	struct rcu_node *level[RCU_NUM_LVLS + 1];
 						/* Hierarchy levels (+1 to */
 						/*  shut bogus gcc warning) */
-	u8 flavor_mask;				/* bit in flavor mask. */
 	struct rcu_data __percpu *rda;		/* pointer of percu rcu_data. */
 	call_rcu_func_t call;			/* call_rcu() flavor. */
 	int ncpus;				/* # CPUs seen so far. */

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-17 23:53 UTC #87 lore raw

PATCH v2 11/39 rcu: Place guard on rcu_all_qs() and rcu_note_context_switch() actions

Cc Peter Zijlstra <[email protected]>

---
 .../Design/Data-Structures/Data-Structures.html    | 11 ++++++-
 kernel/rcu/tree.c                                  | 38 ++++++++++++++--------
 kernel/rcu/tree.h                                  |  3 +-
 kernel/rcu/tree_exp.h                              |  2 ++
 kernel/rcu/tree_plugin.h                           |  8 +++--
 5 files changed, 44 insertions(+), 18 deletions(-)

Documentation/RCU/Design/Data-Structures/Data-Structures.html +10 -1

index e4bf20a68fa3..4dec89097559 100644
@@ -1106,6 +1106,7 @@ Its fields are as follows:
   3   atomic_t dynticks;
   4   bool rcu_need_heavy_qs;
   5   unsigned long rcu_qs_ctr;
+  6   bool rcu_urgent_qs;
 </pre>
 
 <p>The <tt>-&gt;dynticks_nesting</tt> field counts the

@@ -1131,12 +1132,20 @@ it is willing to call for heavy-weight dyntick-counter operations.
 This flag is checked by RCU's context-switch and <tt>cond_resched()</tt>
 code, which provide a momentary idle sojourn in response.
 
-</p><p>Finally the <tt>-&gt;rcu_qs_ctr</tt> field is used to record
+</p><p>The <tt>-&gt;rcu_qs_ctr</tt> field is used to record
 quiescent states from <tt>cond_resched()</tt>.
 Because <tt>cond_resched()</tt> can execute quite frequently, this
 must be quite lightweight, as in a non-atomic increment of this
 per-CPU field.
 
+</p><p>Finally, the <tt>-&gt;rcu_urgent_qs</tt> field is used to record
+the fact that the RCU core code would really like to see a quiescent
+state from the corresponding CPU, with the various other fields indicating
+just how badly RCU wants this quiescent state.
+This flag is checked by RCU's context-switch and <tt>cond_resched()</tt>
+code, which, if nothing else, non-atomically increment <tt>-&gt;rcu_qs_ctr</tt>
+in response.
+
 <table>
 <tr><th>&nbsp;</th></tr>
 <tr><th align="left">Quick Quiz:</th></tr>

kernel/rcu/tree.c +24 ~~-14~~

index 3c62ea06edb3..aa81cc6c365e 100644
@@ -464,8 +464,14 @@ void rcu_note_context_switch(void)
 	trace_rcu_utilization(TPS("Start context switch"));
 	rcu_sched_qs();
 	rcu_preempt_note_context_switch();
+	/* Load rcu_urgent_qs before other flags. */
+	if (!smp_load_acquire(this_cpu_ptr(&rcu_dynticks.rcu_urgent_qs)))
+		goto out;
+	this_cpu_write(rcu_dynticks.rcu_urgent_qs, false);
 	if (unlikely(raw_cpu_read(rcu_dynticks.rcu_need_heavy_qs)))
 		rcu_momentary_dyntick_idle();
+	this_cpu_inc(rcu_dynticks.rcu_qs_ctr);
+out:
 	trace_rcu_utilization(TPS("End context switch"));
 	barrier(); /* Avoid RCU read-side critical sections leaking up. */
 }

@@ -488,29 +494,26 @@ void rcu_all_qs(void)
 {
 	unsigned long flags;
 
+	if (!raw_cpu_read(rcu_dynticks.rcu_urgent_qs))
+		return;
+	preempt_disable();
+	/* Load rcu_urgent_qs before other flags. */
+	if (!smp_load_acquire(this_cpu_ptr(&rcu_dynticks.rcu_urgent_qs))) {
+		preempt_enable();
+		return;
+	}
+	this_cpu_write(rcu_dynticks.rcu_urgent_qs, false);
 	barrier(); /* Avoid RCU read-side critical sections leaking down. */
 	if (unlikely(raw_cpu_read(rcu_dynticks.rcu_need_heavy_qs))) {
 		local_irq_save(flags);
 		rcu_momentary_dyntick_idle();
 		local_irq_restore(flags);
 	}
-	if (unlikely(raw_cpu_read(rcu_sched_data.cpu_no_qs.b.exp))) {
-		/*
-		 * Yes, we just checked a per-CPU variable with preemption
-		 * enabled, so we might be migrated to some other CPU at
-		 * this point.  That is OK because in that case, the
-		 * migration will supply the needed quiescent state.
-		 * We might end up needlessly disabling preemption and
-		 * invoking rcu_sched_qs() on the destination CPU, but
-		 * the probability and cost are both quite low, so this
-		 * should not be a problem in practice.
-		 */
-		preempt_disable();
+	if (unlikely(raw_cpu_read(rcu_sched_data.cpu_no_qs.b.exp)))
 		rcu_sched_qs();
-		preempt_enable();
-	}
 	this_cpu_inc(rcu_dynticks.rcu_qs_ctr);
 	barrier(); /* Avoid RCU read-side critical sections leaking up. */
+	preempt_enable();
 }
 EXPORT_SYMBOL_GPL(rcu_all_qs);

@@ -1246,6 +1249,7 @@ static int rcu_implicit_dynticks_qs(struct rcu_data *rdp,
 {
 	unsigned long jtsq;
 	bool *rnhqp;
+	bool *ruqp;
 	unsigned long rjtsc;
 	struct rcu_node *rnp;

@@ -1281,11 +1285,15 @@ static int rcu_implicit_dynticks_qs(struct rcu_data *rdp,
 	 * might not be the case for nohz_full CPUs looping in the kernel.
 	 */
 	rnp = rdp->mynode;
+	ruqp = per_cpu_ptr(&rcu_dynticks.rcu_urgent_qs, rdp->cpu);
 	if (time_after(jiffies, rdp->rsp->gp_start + jtsq) &&
 	    READ_ONCE(rdp->rcu_qs_ctr_snap) != per_cpu(rcu_dynticks.rcu_qs_ctr, rdp->cpu) &&
 	    READ_ONCE(rdp->gpnum) == rnp->gpnum && !rdp->gpwrap) {
 		trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, TPS("rqc"));
 		return 1;
+	} else {
+		/* Load rcu_qs_ctr before store to rcu_urgent_qs. */
+		smp_store_release(ruqp, true);
 	}
 
 	/* Check for the CPU being offline. */

@@ -1321,6 +1329,8 @@ static int rcu_implicit_dynticks_qs(struct rcu_data *rdp,
 	    (time_after(jiffies, rdp->rsp->gp_start + jtsq) ||
 	     time_after(jiffies, rdp->rsp->jiffies_resched))) {
 		WRITE_ONCE(*rnhqp, true);
+		/* Store rcu_need_heavy_qs before rcu_urgent_qs. */
+		smp_store_release(ruqp, true);
 		rdp->rsp->jiffies_resched += 5; /* Re-enable beating. */
 	}

kernel/rcu/tree.h +2 -1

index b212cd0f22c7..d2f276fc2edc 100644
@@ -113,8 +113,9 @@ struct rcu_dynticks {
 				    /* Process level is worth LLONG_MAX/2. */
 	int dynticks_nmi_nesting;   /* Track NMI nesting level. */
 	atomic_t dynticks;	    /* Even value for idle, else odd. */
-	bool rcu_need_heavy_qs;      /* GP old, need heavy quiescent state. */
+	bool rcu_need_heavy_qs;     /* GP old, need heavy quiescent state. */
 	unsigned long rcu_qs_ctr;   /* Light universal quiescent state ctr. */
+	bool rcu_urgent_qs;	    /* GP old need light quiescent state. */
 #ifdef CONFIG_NO_HZ_FULL_SYSIDLE
 	long long dynticks_idle_nesting;
 				    /* irq/process nesting level from idle. */

kernel/rcu/tree_exp.h +2 -0

index a7b639ccd46e..a1f52bbe9db6 100644
@@ -331,6 +331,8 @@ static void sync_sched_exp_handler(void *data)
 		return;
 	}
 	__this_cpu_write(rcu_sched_data.cpu_no_qs.b.exp, true);
+	/* Store .exp before .rcu_urgent_qs. */
+	smp_store_release(this_cpu_ptr(&rcu_dynticks.rcu_urgent_qs), true);
 	resched_cpu(smp_processor_id());
 }

kernel/rcu/tree_plugin.h +6 -2

index 0a62a8f1caac..621296a6694b 100644
@@ -1860,7 +1860,9 @@ static void __call_rcu_nocb_enqueue(struct rcu_data *rdp,
 			trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu,
 					    TPS("WakeEmpty"));
 		} else {
-			rdp->nocb_defer_wakeup = RCU_NOGP_WAKE;
+			WRITE_ONCE(rdp->nocb_defer_wakeup, RCU_NOGP_WAKE);
+			/* Store ->nocb_defer_wakeup before ->rcu_urgent_qs. */
+			smp_store_release(this_cpu_ptr(&rcu_dynticks.rcu_urgent_qs), true);
 			trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu,
 					    TPS("WakeEmptyIsDeferred"));
 		}

@@ -1872,7 +1874,9 @@ static void __call_rcu_nocb_enqueue(struct rcu_data *rdp,
 			trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu,
 					    TPS("WakeOvf"));
 		} else {
-			rdp->nocb_defer_wakeup = RCU_NOGP_WAKE_FORCE;
+			WRITE_ONCE(rdp->nocb_defer_wakeup, RCU_NOGP_WAKE_FORCE);
+			/* Store ->nocb_defer_wakeup before ->rcu_urgent_qs. */
+			smp_store_release(this_cpu_ptr(&rcu_dynticks.rcu_urgent_qs), true);
 			trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu,
 					    TPS("WakeOvfIsDeferred"));
 		}

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-17 23:47 UTC #88 lore raw

PATCH v2 12/39 rcu: Default RCU_FANOUT_LEAF to 16 unless explicitly changed

Reported-by Peter Zijlstra <[email protected]>

---
 kernel/rcu/tree.h | 6 +-----
 1 file changed, 1 insertion(+), 5 deletions(-)

kernel/rcu/tree.h +1 -5

index d2f276fc2edc..376c01e539c7 100644
@@ -52,11 +52,7 @@
 #ifdef CONFIG_RCU_FANOUT_LEAF
 #define RCU_FANOUT_LEAF CONFIG_RCU_FANOUT_LEAF
 #else /* #ifdef CONFIG_RCU_FANOUT_LEAF */
-# ifdef CONFIG_64BIT
-# define RCU_FANOUT_LEAF 64
-# else
-# define RCU_FANOUT_LEAF 32
-# endif
+#define RCU_FANOUT_LEAF 16
 #endif /* #else #ifdef CONFIG_RCU_FANOUT_LEAF */
 
 #define RCU_FANOUT_1	      (RCU_FANOUT_LEAF)

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-17 23:53 UTC #89 lore raw

PATCH v2 13/39 srcu: Abstract multi-tail callback list handling

---
 kernel/rcu/rcu_segcblist.h | 625 +++++++++++++++++++++++++++++++++++++++++++++
 kernel/rcu/tree.c          | 348 ++++++++-----------------
 kernel/rcu/tree.h          |  41 +--
 kernel/rcu/tree_plugin.h   |  54 ++--
 kernel/rcu/tree_trace.c    |  21 +-
 5 files changed, 780 insertions(+), 309 deletions(-)
 create mode 100644 kernel/rcu/rcu_segcblist.h

kernel/rcu/rcu_segcblist.h new file +625 -0

new file mode 100644
index 000000000000..24078f3c0218
@@ -0,0 +1,625 @@
+/*
+ * RCU segmented callback lists
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, you can access it online at
+ * http://www.gnu.org/licenses/gpl-2.0.html.
+ *
+ * Copyright IBM Corporation, 2017
+ *
+ * Authors: Paul E. McKenney <[email protected]>
+ */
+
+#ifndef __KERNEL_RCU_SEGCBLIST_H
+#define __KERNEL_RCU_SEGCBLIST_H
+
+/* Simple unsegmented callback lists. */
+struct rcu_cblist {
+	struct rcu_head *head;
+	struct rcu_head **tail;
+	long len;
+	long len_lazy;
+};
+
+#define RCU_CBLIST_INITIALIZER(n) { .head = NULL, .tail = &n.head }
+
+/* Initialize simple callback list. */
+static inline void rcu_cblist_init(struct rcu_cblist *rclp)
+{
+	rclp->head = NULL;
+	rclp->tail = &rclp->head;
+	rclp->len = 0;
+	rclp->len_lazy = 0;
+}
+
+/* Is simple callback list empty? */
+static inline bool rcu_cblist_empty(struct rcu_cblist *rclp)
+{
+	return !rclp->head;
+}
+
+/* Return number of callbacks in simple callback list. */
+static inline long rcu_cblist_n_cbs(struct rcu_cblist *rclp)
+{
+	return rclp->len;
+}
+
+/* Return number of lazy callbacks in simple callback list. */
+static inline long rcu_cblist_n_lazy_cbs(struct rcu_cblist *rclp)
+{
+	return rclp->len_lazy;
+}
+
+/*
+ * Debug function to actually count the number of callbacks.
+ * If the number exceeds the limit specified, return -1.
+ */
+static inline long rcu_cblist_count_cbs(struct rcu_cblist *rclp, long lim)
+{
+	int cnt = 0;
+	struct rcu_head **rhpp = &rclp->head;
+
+	for (;;) {
+		if (!*rhpp)
+			return cnt;
+		if (++cnt > lim)
+			return -1;
+		rhpp = &(*rhpp)->next;
+	}
+}
+
+/*
+ * Dequeue the oldest rcu_head structure from the specified callback
+ * list.  This function assumes that the callback is non-lazy, but
+ * the caller can later invoke rcu_cblist_dequeued_lazy() if it
+ * finds otherwise (and if it cares about laziness).  This allows
+ * different users to have different ways of determining laziness.
+ */
+static inline struct rcu_head *rcu_cblist_dequeue(struct rcu_cblist *rclp)
+{
+	struct rcu_head *rhp;
+
+	rhp = rclp->head;
+	if (!rhp)
+		return NULL;
+	prefetch(rhp);
+	rclp->len--;
+	rclp->head = rhp->next;
+	if (!rclp->head)
+		rclp->tail = &rclp->head;
+	return rhp;
+}
+
+/*
+ * Account for the fact that a previously dequeued callback turned out
+ * to be marked as lazy.
+ */
+static inline void rcu_cblist_dequeued_lazy(struct rcu_cblist *rclp)
+{
+	rclp->len_lazy--;
+}
+
+/*
+ * Interim function to return rcu_cblist head pointer.  Longer term, the
+ * rcu_cblist will be used more pervasively, removing the need for this
+ * function.
+ */
+static inline struct rcu_head *rcu_cblist_head(struct rcu_cblist *rclp)
+{
+	return rclp->head;
+}
+
+/*
+ * Interim function to return rcu_cblist head pointer.  Longer term, the
+ * rcu_cblist will be used more pervasively, removing the need for this
+ * function.
+ */
+static inline struct rcu_head **rcu_cblist_tail(struct rcu_cblist *rclp)
+{
+	WARN_ON_ONCE(rcu_cblist_empty(rclp));
+	return rclp->tail;
+}
+
+/* Complicated segmented callback lists.  ;-) */
+
+/*
+ * Index values for segments in rcu_segcblist structure.
+ *
+ * The segments are as follows:
+ *
+ * [head, *tails[RCU_DONE_TAIL]):
+ *	Callbacks whose grace period has elapsed, and thus can be invoked.
+ * [*tails[RCU_DONE_TAIL], *tails[RCU_WAIT_TAIL]):
+ *	Callbacks waiting for the current GP from the current CPU's viewpoint.
+ * [*tails[RCU_WAIT_TAIL], *tails[RCU_NEXT_READY_TAIL]):
+ *	Callbacks that arrived before the next GP started, again from
+ *	the current CPU's viewpoint.  These can be handled by the next GP.
+ * [*tails[RCU_NEXT_READY_TAIL], *tails[RCU_NEXT_TAIL]):
+ *	Callbacks that might have arrived after the next GP started.
+ *	There is some uncertainty as to when a given GP starts and
+ *	ends, but a CPU knows the exact times if it is the one starting
+ *	or ending the GP.  Other CPUs know that the previous GP ends
+ *	before the next one starts.
+ *
+ * Note that RCU_WAIT_TAIL cannot be empty unless RCU_NEXT_READY_TAIL is also
+ * empty.
+ *
+ * The ->gp_seq[] array contains the grace-period number at which the
+ * corresponding segment of callbacks will be ready to invoke.  A given
+ * element of this array is meaningful only when the corresponding segment
+ * is non-empty, and it is never valid for RCU_DONE_TAIL (whose callbacks
+ * are already ready to invoke) or for RCU_NEXT_TAIL (whose callbacks have
+ * not yet been assigned a grace-period number).
+ */
+#define RCU_DONE_TAIL		0	/* Also RCU_WAIT head. */
+#define RCU_WAIT_TAIL		1	/* Also RCU_NEXT_READY head. */
+#define RCU_NEXT_READY_TAIL	2	/* Also RCU_NEXT head. */
+#define RCU_NEXT_TAIL		3
+#define RCU_CBLIST_NSEGS	4
+
+struct rcu_segcblist {
+	struct rcu_head *head;
+	struct rcu_head **tails[RCU_CBLIST_NSEGS];
+	unsigned long gp_seq[RCU_CBLIST_NSEGS];
+	long len;
+	long len_lazy;
+};
+
+/*
+ * Initialize an rcu_segcblist structure.
+ */
+static inline void rcu_segcblist_init(struct rcu_segcblist *rsclp)
+{
+	int i;
+
+	BUILD_BUG_ON(RCU_NEXT_TAIL + 1 != ARRAY_SIZE(rsclp->gp_seq));
+	BUILD_BUG_ON(ARRAY_SIZE(rsclp->tails) != ARRAY_SIZE(rsclp->gp_seq));
+	rsclp->head = NULL;
+	for (i = 0; i < RCU_CBLIST_NSEGS; i++)
+		rsclp->tails[i] = &rsclp->head;
+	rsclp->len = 0;
+	rsclp->len_lazy = 0;
+}
+
+/*
+ * Is the specified rcu_segcblist structure empty?
+ *
+ * But careful!  The fact that the ->head field is NULL does not
+ * necessarily imply that there are no callbacks associated with
+ * this structure.  When callbacks are being invoked, they are
+ * removed as a group.  If callback invocation must be preempted,
+ * the remaining callbacks will be added back to the list.  Either
+ * way, the counts are updated later.
+ *
+ * So it is often the case that rcu_segcblist_n_cbs() should be used
+ * instead.
+ */
+static inline bool rcu_segcblist_empty(struct rcu_segcblist *rsclp)
+{
+	return !rsclp->head;
+}
+
+/* Return number of callbacks in segmented callback list. */
+static inline long rcu_segcblist_n_cbs(struct rcu_segcblist *rsclp)
+{
+	return READ_ONCE(rsclp->len);
+}
+
+/* Return number of lazy callbacks in segmented callback list. */
+static inline long rcu_segcblist_n_lazy_cbs(struct rcu_segcblist *rsclp)
+{
+	return rsclp->len_lazy;
+}
+
+/* Return number of lazy callbacks in segmented callback list. */
+static inline long rcu_segcblist_n_nonlazy_cbs(struct rcu_segcblist *rsclp)
+{
+	return rsclp->len - rsclp->len_lazy;
+}
+
+/*
+ * Is the specified rcu_segcblist enabled, for example, not corresponding
+ * to an offline or callback-offloaded CPU?
+ */
+static inline bool rcu_segcblist_is_enabled(struct rcu_segcblist *rsclp)
+{
+	return !!rsclp->tails[RCU_NEXT_TAIL];
+}
+
+/*
+ * Disable the specified rcu_segcblist structure, so that callbacks can
+ * no longer be posted to it.  This structure must be empty.
+ */
+static inline void rcu_segcblist_disable(struct rcu_segcblist *rsclp)
+{
+	WARN_ON_ONCE(!rcu_segcblist_empty(rsclp));
+	WARN_ON_ONCE(rcu_segcblist_n_cbs(rsclp));
+	WARN_ON_ONCE(rcu_segcblist_n_lazy_cbs(rsclp));
+	rsclp->tails[RCU_NEXT_TAIL] = NULL;
+}
+
+/*
+ * Is the specified segment of the specified rcu_segcblist structure
+ * empty of callbacks?
+ */
+static inline bool rcu_segcblist_segempty(struct rcu_segcblist *rsclp, int seg)
+{
+	if (seg == RCU_DONE_TAIL)
+		return &rsclp->head == rsclp->tails[RCU_DONE_TAIL];
+	return rsclp->tails[seg - 1] == rsclp->tails[seg];
+}
+
+/*
+ * Are all segments following the specified segment of the specified
+ * rcu_segcblist structure empty of callbacks?  (The specified
+ * segment might well contain callbacks.)
+ */
+static inline bool rcu_segcblist_restempty(struct rcu_segcblist *rsclp, int seg)
+{
+	return !*rsclp->tails[seg];
+}
+
+/*
+ * Does the specified rcu_segcblist structure contain callbacks that
+ * are ready to be invoked?
+ */
+static inline bool rcu_segcblist_ready_cbs(struct rcu_segcblist *rsclp)
+{
+	return rcu_segcblist_is_enabled(rsclp) &&
+	       &rsclp->head != rsclp->tails[RCU_DONE_TAIL];
+}
+
+/*
+ * Does the specified rcu_segcblist structure contain callbacks that
+ * are still pending, that is, not yet ready to be invoked?
+ */
+static inline bool rcu_segcblist_pend_cbs(struct rcu_segcblist *rsclp)
+{
+	return rcu_segcblist_is_enabled(rsclp) &&
+	       !rcu_segcblist_restempty(rsclp, RCU_DONE_TAIL);
+}
+
+/*
+ * Return a pointer to the first callback in the specified rcu_segcblist
+ * structure.  This is useful for diagnostics.
+ */
+static inline struct rcu_head *
+rcu_segcblist_first_cb(struct rcu_segcblist *rsclp)
+{
+	if (rcu_segcblist_is_enabled(rsclp))
+		return rsclp->head;
+	return NULL;
+}
+
+/*
+ * Return a pointer to the first pending callback in the specified
+ * rcu_segcblist structure.  This is useful just after posting a given
+ * callback -- if that callback is the first pending callback, then
+ * you cannot rely on someone else having already started up the required
+ * grace period.
+ */
+static inline struct rcu_head *
+rcu_segcblist_first_pend_cb(struct rcu_segcblist *rsclp)
+{
+	if (rcu_segcblist_is_enabled(rsclp))
+		return *rsclp->tails[RCU_DONE_TAIL];
+	return NULL;
+}
+
+/*
+ * Does the specified rcu_segcblist structure contain callbacks that
+ * have not yet been processed beyond having been posted, that is,
+ * does it contain callbacks in its last segment?
+ */
+static inline bool rcu_segcblist_new_cbs(struct rcu_segcblist *rsclp)
+{
+	return rcu_segcblist_is_enabled(rsclp) &&
+	       !rcu_segcblist_restempty(rsclp, RCU_NEXT_READY_TAIL);
+}
+
+/*
+ * Enqueue the specified callback onto the specified rcu_segcblist
+ * structure, updating accounting as needed.  Note that the ->len
+ * field may be accessed locklessly, hence the WRITE_ONCE().
+ * The ->len field is used by rcu_barrier() and friends to determine
+ * if it must post a callback on this structure, and it is OK
+ * for rcu_barrier() to sometimes post callbacks needlessly, but
+ * absolutely not OK for it to ever miss posting a callback.
+ */
+static inline void rcu_segcblist_enqueue(struct rcu_segcblist *rsclp,
+					 struct rcu_head *rhp, bool lazy)
+{
+	WRITE_ONCE(rsclp->len, rsclp->len + 1); /* ->len sampled locklessly. */
+	if (lazy)
+		rsclp->len_lazy++;
+	smp_mb(); /* Ensure counts are updated before callback is enqueued. */
+	rhp->next = NULL;
+	*rsclp->tails[RCU_NEXT_TAIL] = rhp;
+	rsclp->tails[RCU_NEXT_TAIL] = &rhp->next;
+}
+
+/*
+ * Extract only the counts from the specified rcu_segcblist structure,
+ * and place them in the specified rcu_cblist structure.  This function
+ * supports both callback orphaning and invocation, hence the separation
+ * of counts and callbacks.  (Callbacks ready for invocation must be
+ * orphaned and adopted separately from pending callbacks, but counts
+ * apply to all callbacks.  Locking must be used to make sure that
+ * both orphaned-callbacks lists are consistent.)
+ */
+static inline void rcu_segcblist_extract_count(struct rcu_segcblist *rsclp,
+					       struct rcu_cblist *rclp)
+{
+	rclp->len_lazy += rsclp->len_lazy;
+	rclp->len += rsclp->len;
+	rsclp->len_lazy = 0;
+	WRITE_ONCE(rsclp->len, 0); /* ->len sampled locklessly. */
+}
+
+/*
+ * Extract only those callbacks ready to be invoked from the specified
+ * rcu_segcblist structure and place them in the specified rcu_cblist
+ * structure.
+ */
+static inline void rcu_segcblist_extract_done_cbs(struct rcu_segcblist *rsclp,
+						  struct rcu_cblist *rclp)
+{
+	int i;
+
+	if (!rcu_segcblist_ready_cbs(rsclp))
+		return; /* Nothing to do. */
+	*rclp->tail = rsclp->head;
+	rsclp->head = *rsclp->tails[RCU_DONE_TAIL];
+	*rsclp->tails[RCU_DONE_TAIL] = NULL;
+	rclp->tail = rsclp->tails[RCU_DONE_TAIL];
+	for (i = RCU_CBLIST_NSEGS - 1; i >= RCU_DONE_TAIL; i--)
+		if (rsclp->tails[i] == rsclp->tails[RCU_DONE_TAIL])
+			rsclp->tails[i] = &rsclp->head;
+}
+
+/*
+ * Extract only those callbacks still pending (not yet ready to be
+ * invoked) from the specified rcu_segcblist structure and place them in
+ * the specified rcu_cblist structure.  Note that this loses information
+ * about any callbacks that might have been partway done waiting for
+ * their grace period.  Too bad!  They will have to start over.
+ */
+static inline void
+rcu_segcblist_extract_pend_cbs(struct rcu_segcblist *rsclp,
+			       struct rcu_cblist *rclp)
+{
+	int i;
+
+	if (!rcu_segcblist_pend_cbs(rsclp))
+		return; /* Nothing to do. */
+	*rclp->tail = *rsclp->tails[RCU_DONE_TAIL];
+	rclp->tail = rsclp->tails[RCU_NEXT_TAIL];
+	*rsclp->tails[RCU_DONE_TAIL] = NULL;
+	for (i = RCU_DONE_TAIL + 1; i < RCU_CBLIST_NSEGS; i++)
+		rsclp->tails[i] = rsclp->tails[RCU_DONE_TAIL];
+}
+
+/*
+ * Move the entire contents of the specified rcu_segcblist structure,
+ * counts, callbacks, and all, to the specified rcu_cblist structure.
+ * @@@ Why do we need this???  Moving early-boot CBs to NOCB lists?
+ * @@@ Memory barrier needed?  (Not if only used at boot time...)
+ */
+static inline void rcu_segcblist_extract_all(struct rcu_segcblist *rsclp,
+					     struct rcu_cblist *rclp)
+{
+	rcu_segcblist_extract_done_cbs(rsclp, rclp);
+	rcu_segcblist_extract_pend_cbs(rsclp, rclp);
+	rcu_segcblist_extract_count(rsclp, rclp);
+}
+
+/*
+ * Insert counts from the specified rcu_cblist structure in the
+ * specified rcu_segcblist structure.
+ */
+static inline void rcu_segcblist_insert_count(struct rcu_segcblist *rsclp,
+					      struct rcu_cblist *rclp)
+{
+	rsclp->len_lazy += rclp->len_lazy;
+	/* ->len sampled locklessly. */
+	WRITE_ONCE(rsclp->len, rsclp->len + rclp->len);
+	rclp->len_lazy = 0;
+	rclp->len = 0;
+}
+
+/*
+ * Move callbacks from the specified rcu_cblist to the beginning of the
+ * done-callbacks segment of the specified rcu_segcblist.
+ */
+static inline void rcu_segcblist_insert_done_cbs(struct rcu_segcblist *rsclp,
+						 struct rcu_cblist *rclp)
+{
+	int i;
+
+	if (!rclp->head)
+		return; /* No callbacks to move. */
+	*rclp->tail = rsclp->head;
+	rsclp->head = rclp->head;
+	for (i = RCU_DONE_TAIL; i < RCU_CBLIST_NSEGS; i++)
+		if (&rsclp->head == rsclp->tails[i])
+			rsclp->tails[i] = rclp->tail;
+		else
+			break;
+	rclp->head = NULL;
+	rclp->tail = &rclp->head;
+}
+
+/*
+ * Move callbacks from the specified rcu_cblist to the end of the
+ * new-callbacks segment of the specified rcu_segcblist.
+ */
+static inline void rcu_segcblist_insert_pend_cbs(struct rcu_segcblist *rsclp,
+						 struct rcu_cblist *rclp)
+{
+	if (!rclp->head)
+		return; /* Nothing to do. */
+	*rsclp->tails[RCU_NEXT_TAIL] = rclp->head;
+	rsclp->tails[RCU_NEXT_TAIL] = rclp->tail;
+	rclp->head = NULL;
+	rclp->tail = &rclp->head;
+}
+
+/*
+ * Advance the callbacks in the specified rcu_segcblist structure based
+ * on the current value passed in for the grace-period counter.
+ */
+static inline void rcu_segcblist_advance(struct rcu_segcblist *rsclp,
+					 unsigned long seq)
+{
+	int i, j;
+
+	WARN_ON_ONCE(!rcu_segcblist_is_enabled(rsclp));
+	WARN_ON_ONCE(rcu_segcblist_restempty(rsclp, RCU_DONE_TAIL));
+
+	/*
+	 * Find all callbacks whose ->gp_seq numbers indicate that they
+	 * are ready to invoke, and put them into the RCU_DONE_TAIL segment.
+	 */
+	for (i = RCU_WAIT_TAIL; i < RCU_NEXT_TAIL; i++) {
+		if (ULONG_CMP_LT(seq, rsclp->gp_seq[i]))
+			break;
+		rsclp->tails[RCU_DONE_TAIL] = rsclp->tails[i];
+	}
+
+	/* If no callbacks moved, nothing more need be done. */
+	if (i == RCU_WAIT_TAIL)
+		return;
+
+	/* Clean up tail pointers that might have been misordered above. */
+	for (j = RCU_WAIT_TAIL; j < i; j++)
+		rsclp->tails[j] = rsclp->tails[RCU_DONE_TAIL];
+
+	/*
+	 * Callbacks moved, so clean up the misordered ->tails[] pointers
+	 * that now point into the middle of the list of ready-to-invoke
+	 * callbacks.  The overall effect is to copy down the later pointers
+	 * into the gap that was created by the now-ready segments.
+	 */
+	for (j = RCU_WAIT_TAIL; i < RCU_NEXT_TAIL; i++, j++) {
+		if (rsclp->tails[j] == rsclp->tails[RCU_NEXT_TAIL])
+			break;  /* No more callbacks. */
+		rsclp->tails[j] = rsclp->tails[i];
+		rsclp->gp_seq[j] = rsclp->gp_seq[i];
+	}
+}
+
+/*
+ * "Accelerate" callbacks based on more-accurate grace-period information.
+ * The reason for this is that RCU does not synchronize the beginnings and
+ * ends of grace periods, and that callbacks are posted locally.  This in
+ * turn means that the callbacks must be labelled conservatively early
+ * on, as getting exact information would degrade both performance and
+ * scalability.  When more accurate grace-period information becomes
+ * available, previously posted callbacks can be "accelerated", marking
+ * them to complete at the end of the earlier grace period.
+ *
+ * This function operates on an rcu_segcblist structure, and also the
+ * grace-period sequence number at which new callbacks would become
+ * ready to invoke.
+ */
+static inline bool rcu_segcblist_accelerate(struct rcu_segcblist *rsclp,
+					    unsigned long seq)
+{
+	int i;
+
+	WARN_ON_ONCE(!rcu_segcblist_is_enabled(rsclp));
+	WARN_ON_ONCE(rcu_segcblist_restempty(rsclp, RCU_DONE_TAIL));
+
+	/*
+	 * Find the segment preceding the oldest segment of callbacks
+	 * whose ->gp_seq[] completion is at or after that passed in via
+	 * "seq", skipping any empty segments.  This oldest segment, along
+	 * with any later segments, can be merged in with any newly arrived
+	 * callbacks in the RCU_NEXT_TAIL segment, and assigned "seq"
+	 * as their ->gp_seq[] grace-period completion sequence number.
+	 */
+	for (i = RCU_NEXT_READY_TAIL; i > RCU_DONE_TAIL; i--)
+		if (rsclp->tails[i] != rsclp->tails[i - 1] &&
+		    ULONG_CMP_LT(rsclp->gp_seq[i], seq))
+			break;
+
+	/*
+	 * If all the segments contain callbacks that correspond to
+	 * earlier grace-period sequence numbers than "seq", leave.
+	 * Assuming that the rcu_segcblist structure has enough
+	 * segments in its arrays, this can only happen if some of
+	 * the non-done segments contain callbacks that really are
+	 * ready to invoke.  This situation will get straightened
+	 * out by the next call to rcu_segcblist_advance().
+	 *
+	 * Also advance to the oldest segment of callbacks whose
+	 * ->gp_seq[] completion is at or after that passed in via "seq",
+	 * skipping any empty segments.
+	 */
+	if (++i >= RCU_NEXT_TAIL)
+		return false;
+
+	/*
+	 * Merge all later callbacks, including newly arrived callbacks,
+	 * into the segment located by the for-loop above.  Assign "seq"
+	 * as the ->gp_seq[] value in order to correctly handle the case
+	 * where there were no pending callbacks in the rcu_segcblist
+	 * structure other than in the RCU_NEXT_TAIL segment.
+	 */
+	for (; i < RCU_NEXT_TAIL; i++) {
+		rsclp->tails[i] = rsclp->tails[RCU_NEXT_TAIL];
+		rsclp->gp_seq[i] = seq;
+	}
+	return true;
+}
+
+/*
+ * Scan the specified rcu_segcblist structure for callbacks that need
+ * a grace period later than the one specified by "seq".  We don't look
+ * at the RCU_DONE_TAIL or RCU_NEXT_TAIL segments because they don't
+ * have a grace-period sequence number.
+ */
+static inline bool rcu_segcblist_future_gp_needed(struct rcu_segcblist *rsclp,
+						  unsigned long seq)
+{
+	int i;
+
+	for (i = RCU_WAIT_TAIL; i < RCU_NEXT_TAIL; i++)
+		if (rsclp->tails[i - 1] != rsclp->tails[i] &&
+		    ULONG_CMP_LT(seq, rsclp->gp_seq[i]))
+			return true;
+	return false;
+}
+
+/*
+ * Interim function to return rcu_segcblist head pointer.  Longer term, the
+ * rcu_segcblist will be used more pervasively, removing the need for this
+ * function.
+ */
+static inline struct rcu_head *rcu_segcblist_head(struct rcu_segcblist *rsclp)
+{
+	return rsclp->head;
+}
+
+/*
+ * Interim function to return rcu_segcblist head pointer.  Longer term, the
+ * rcu_segcblist will be used more pervasively, removing the need for this
+ * function.
+ */
+static inline struct rcu_head **rcu_segcblist_tail(struct rcu_segcblist *rsclp)
+{
+	WARN_ON_ONCE(rcu_segcblist_empty(rsclp));
+	return rsclp->tails[RCU_NEXT_TAIL];
+}
+
+#endif /* __KERNEL_RCU_SEGCBLIST_H */

kernel/rcu/tree.c +112 ~~-236~~

580 lines not on this page · raw message

kernel/rcu/tree.h +7 ~~-34~~

61 lines not on this page · raw message

kernel/rcu/tree_plugin.h +25 ~~-29~~

95 lines not on this page · raw message

kernel/rcu/tree_trace.c +11 ~~-10~~

index 65b43be38e68..066c64071a7b 100644
@@ -41,6 +41,7 @@
 #include <linux/mutex.h>
 #include <linux/debugfs.h>
 #include <linux/seq_file.h>
+#include <linux/prefetch.h>
 
 #define RCU_TREE_NONCORE
 #include "tree.h"

@@ -128,17 +129,15 @@ static void print_one_rcu_data(struct seq_file *m, struct rcu_data *rdp)
 		   rdp->dynticks_fqs);
 	seq_printf(m, " of=%lu", rdp->offline_fqs);
 	rcu_nocb_q_lengths(rdp, &ql, &qll);
-	qll += rdp->qlen_lazy;
-	ql += rdp->qlen;
+	qll += rcu_segcblist_n_lazy_cbs(&rdp->cblist);
+	ql += rcu_segcblist_n_cbs(&rdp->cblist);
 	seq_printf(m, " ql=%ld/%ld qs=%c%c%c%c",
 		   qll, ql,
-		   ".N"[rdp->nxttail[RCU_NEXT_READY_TAIL] !=
-			rdp->nxttail[RCU_NEXT_TAIL]],
-		   ".R"[rdp->nxttail[RCU_WAIT_TAIL] !=
-			rdp->nxttail[RCU_NEXT_READY_TAIL]],
-		   ".W"[rdp->nxttail[RCU_DONE_TAIL] !=
-			rdp->nxttail[RCU_WAIT_TAIL]],
-		   ".D"[&rdp->nxtlist != rdp->nxttail[RCU_DONE_TAIL]]);
+		   ".N"[!rcu_segcblist_segempty(&rdp->cblist, RCU_NEXT_TAIL)],
+		   ".R"[!rcu_segcblist_segempty(&rdp->cblist,
+						RCU_NEXT_READY_TAIL)],
+		   ".W"[!rcu_segcblist_segempty(&rdp->cblist, RCU_WAIT_TAIL)],
+		   ".D"[!rcu_segcblist_segempty(&rdp->cblist, RCU_DONE_TAIL)]);
 #ifdef CONFIG_RCU_BOOST
 	seq_printf(m, " kt=%d/%c ktl=%x",
 		   per_cpu(rcu_cpu_has_work, rdp->cpu),

@@ -276,7 +275,9 @@ static void print_one_rcu_state(struct seq_file *m, struct rcu_state *rsp)
 	seq_printf(m, "nfqs=%lu/nfqsng=%lu(%lu) fqlh=%lu oqlen=%ld/%ld\n",
 		   rsp->n_force_qs, rsp->n_force_qs_ngp,
 		   rsp->n_force_qs - rsp->n_force_qs_ngp,
-		   READ_ONCE(rsp->n_force_qs_lh), rsp->qlen_lazy, rsp->qlen);
+		   READ_ONCE(rsp->n_force_qs_lh),
+		   rcu_cblist_n_lazy_cbs(&rsp->orphan_done),
+		   rcu_cblist_n_cbs(&rsp->orphan_done));
 	for (rnp = &rsp->node[0]; rnp - &rsp->node[0] < rcu_num_nodes; rnp++) {
 		if (rnp->level != level) {
 			seq_puts(m, "\n");

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-17 23:45 UTC #90 lore raw

PATCH v2 14/39 srcu: Allow SRCU to access rcu_scheduler_active

This is primarily a code-movement commit in preparation for allowing SRCU to handle early-boot SRCU grace periods.

---
 include/linux/rcutiny.h  |  6 +++---
 kernel/rcu/tiny_plugin.h |  9 +++++----
 kernel/rcu/tree.c        |  2 +-
 kernel/rcu/tree_exp.h    | 12 -----------
 kernel/rcu/update.c      | 52 +++++++++++++++++++++++++++++++-----------------
 5 files changed, 43 insertions(+), 38 deletions(-)

include/linux/rcutiny.h +3 -3

index 6c9d941e3962..5219be250f00 100644
@@ -217,14 +217,14 @@ static inline void exit_rcu(void)
 {
 }
 
-#ifdef CONFIG_DEBUG_LOCK_ALLOC
+#if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_SRCU)
 extern int rcu_scheduler_active __read_mostly;
 void rcu_scheduler_starting(void);
-#else /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
+#else /* #if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_SRCU) */
 static inline void rcu_scheduler_starting(void)
 {
 }
-#endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */
+#endif /* #else #if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_SRCU) */
 
 #if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_RCU_TRACE)

kernel/rcu/tiny_plugin.h +5 -4

index df3a60e19f07..371034e77f87 100644
@@ -52,7 +52,7 @@ static struct rcu_ctrlblk rcu_bh_ctrlblk = {
 	RCU_TRACE(.name = "rcu_bh")
 };
 
-#ifdef CONFIG_DEBUG_LOCK_ALLOC
+#if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_SRCU)
 #include <linux/kernel_stat.h>
 
 int rcu_scheduler_active __read_mostly;

@@ -65,15 +65,16 @@ EXPORT_SYMBOL_GPL(rcu_scheduler_active);
  * to RCU_SCHEDULER_RUNNING, skipping the RCU_SCHEDULER_INIT stage.
  * The reason for this is that Tiny RCU does not need kthreads, so does
  * not have to care about the fact that the scheduler is half-initialized
- * at a certain phase of the boot process.
+ * at a certain phase of the boot process.  Unless SRCU is in the mix.
  */
 void __init rcu_scheduler_starting(void)
 {
 	WARN_ON(nr_context_switches() > 0);
-	rcu_scheduler_active = RCU_SCHEDULER_RUNNING;
+	rcu_scheduler_active = IS_ENABLED(CONFIG_SRCU)
+		? RCU_SCHEDULER_INIT : RCU_SCHEDULER_RUNNING;
 }
 
-#endif /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
+#endif /* #if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_SRCU) */
 
 #ifdef CONFIG_RCU_TRACE

kernel/rcu/tree.c +1 -1

index ea14e6410cb8..03a1e3e09e82 100644
@@ -3974,7 +3974,7 @@ early_initcall(rcu_spawn_gp_kthread);
  * task is booting the system, and such primitives are no-ops).  After this
  * function is called, any synchronous grace-period primitives are run as
  * expedited, with the requesting task driving the grace period forward.
- * A later core_initcall() rcu_exp_runtime_mode() will switch to full
+ * A later core_initcall() rcu_set_runtime_mode() will switch to full
  * runtime RCU functionality.
  */
 void rcu_scheduler_starting(void)

kernel/rcu/tree_exp.h +0 ~~-12~~

index a1f52bbe9db6..51ca287828a2 100644
@@ -737,15 +737,3 @@ void synchronize_rcu_expedited(void)
 EXPORT_SYMBOL_GPL(synchronize_rcu_expedited);
 
 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */
-
-/*
- * Switch to run-time mode once Tree RCU has fully initialized.
- */
-static int __init rcu_exp_runtime_mode(void)
-{
-	rcu_test_sync_prims();
-	rcu_scheduler_active = RCU_SCHEDULER_RUNNING;
-	rcu_test_sync_prims();
-	return 0;
-}
-core_initcall(rcu_exp_runtime_mode);

kernel/rcu/update.c +34 ~~-18~~

70 lines not on this page · raw message

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-17 23:45 UTC #91 lore raw

PATCH v2 15/39 srcu: Allow early boot use of synchronize_srcu()

This commit checks for pre-scheduler state, and if that early in the boot process, synchronize_srcu() and friends are no-ops.

---
 kernel/rcu/srcu.c | 2 ++
 1 file changed, 2 insertions(+)

kernel/rcu/srcu.c +2 -0

index 6beeba7b0b67..1026ce24922f 100644
@@ -411,6 +411,8 @@ static void __synchronize_srcu(struct srcu_struct *sp, int trycount)
 			 lock_is_held(&rcu_sched_lock_map),
 			 "Illegal synchronize_srcu() in same-type SRCU (or in RCU) read-side critical section");
 
+	if (rcu_scheduler_active == RCU_SCHEDULER_INACTIVE)
+		return;
 	might_sleep();
 	init_completion(&rcu.completion);

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-17 23:47 UTC #92 lore raw

PATCH v2 16/39 rcu: Add single-element dequeue functions to rcu_segcblist

---
 kernel/rcu/rcu_segcblist.h | 45 +++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 45 insertions(+)

kernel/rcu/rcu_segcblist.h +45 -0

index 24078f3c0218..982e3e05b22a 100644
@@ -290,6 +290,51 @@ static inline bool rcu_segcblist_pend_cbs(struct rcu_segcblist *rsclp)
 }
 
 /*
+ * Dequeue and return the first ready-to-invoke callback.  If there
+ * are no ready-to-invoke callbacks, return NULL.  Disables interrupts
+ * to avoid interference.  Does not protect from interference from other
+ * CPUs or tasks.
+ */
+static inline struct rcu_head *
+rcu_segcblist_dequeue(struct rcu_segcblist *rsclp)
+{
+	unsigned long flags;
+	int i;
+	struct rcu_head *rhp;
+
+	local_irq_save(flags);
+	if (!rcu_segcblist_ready_cbs(rsclp)) {
+		local_irq_restore(flags);
+		return NULL;
+	}
+	rhp = rsclp->head;
+	BUG_ON(!rhp);
+	rsclp->head = rhp->next;
+	for (i = RCU_DONE_TAIL; i < RCU_CBLIST_NSEGS; i++) {
+		if (rsclp->tails[i] != &rhp->next)
+			break;
+		rsclp->tails[i] = &rsclp->head;
+	}
+	smp_mb(); /* Dequeue before decrement for rcu_barrier(). */
+	WRITE_ONCE(rsclp->len, rsclp->len - 1);
+	local_irq_restore(flags);
+	return rhp;
+}
+
+/*
+ * Account for the fact that a previously dequeued callback turned out
+ * to be marked as lazy.
+ */
+static inline void rcu_segcblist_dequeued_lazy(struct rcu_segcblist *rsclp)
+{
+	unsigned long flags;
+
+	local_irq_save(flags);
+	rsclp->len_lazy--;
+	local_irq_restore(flags);
+}
+
+/*
  * Return a pointer to the first callback in the specified rcu_segcblist
  * structure.  This is useful for diagnostics.
  */

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-17 23:52 UTC #93 lore raw

PATCH v2 17/39 srcu: Move rcu_seq_start() and friends to rcu.h

---
 kernel/rcu/rcu.h  | 40 ++++++++++++++++++++++++++++++++++++++++
 kernel/rcu/tree.c | 35 -----------------------------------
 2 files changed, 40 insertions(+), 35 deletions(-)

kernel/rcu/rcu.h +40 -0

index 8700a81daf56..91e0bf31f6ce 100644
@@ -56,6 +56,46 @@
 #define DYNTICK_TASK_EXIT_IDLE	   (DYNTICK_TASK_NEST_VALUE + \
 				    DYNTICK_TASK_FLAG)
 
+
+/*
+ * Grace-period counter management.
+ */
+
+/* Adjust sequence number for start of update-side operation. */
+static inline void rcu_seq_start(unsigned long *sp)
+{
+	WRITE_ONCE(*sp, *sp + 1);
+	smp_mb(); /* Ensure update-side operation after counter increment. */
+	WARN_ON_ONCE(!(*sp & 0x1));
+}
+
+/* Adjust sequence number for end of update-side operation. */
+static inline void rcu_seq_end(unsigned long *sp)
+{
+	smp_mb(); /* Ensure update-side operation before counter increment. */
+	WRITE_ONCE(*sp, *sp + 1);
+	WARN_ON_ONCE(*sp & 0x1);
+}
+
+/* Take a snapshot of the update side's sequence number. */
+static inline unsigned long rcu_seq_snap(unsigned long *sp)
+{
+	unsigned long s;
+
+	s = (READ_ONCE(*sp) + 3) & ~0x1;
+	smp_mb(); /* Above access must not bleed into critical section. */
+	return s;
+}
+
+/*
+ * Given a snapshot from rcu_seq_snap(), determine whether or not a
+ * full update-side operation has occurred.
+ */
+static inline bool rcu_seq_done(unsigned long *sp, unsigned long s)
+{
+	return ULONG_CMP_GE(READ_ONCE(*sp), s);
+}
+
 /*
  * debug_rcu_head_queue()/debug_rcu_head_unqueue() are used internally
  * by call_rcu() and rcu callback execution, and are therefore not part of the

kernel/rcu/tree.c +0 ~~-35~~

index 03a1e3e09e82..a3eff40a6c17 100644
@@ -3409,41 +3409,6 @@ void cond_synchronize_sched(unsigned long oldstate)
 }
 EXPORT_SYMBOL_GPL(cond_synchronize_sched);
 
-/* Adjust sequence number for start of update-side operation. */
-static void rcu_seq_start(unsigned long *sp)
-{
-	WRITE_ONCE(*sp, *sp + 1);
-	smp_mb(); /* Ensure update-side operation after counter increment. */
-	WARN_ON_ONCE(!(*sp & 0x1));
-}
-
-/* Adjust sequence number for end of update-side operation. */
-static void rcu_seq_end(unsigned long *sp)
-{
-	smp_mb(); /* Ensure update-side operation before counter increment. */
-	WRITE_ONCE(*sp, *sp + 1);
-	WARN_ON_ONCE(*sp & 0x1);
-}
-
-/* Take a snapshot of the update side's sequence number. */
-static unsigned long rcu_seq_snap(unsigned long *sp)
-{
-	unsigned long s;
-
-	s = (READ_ONCE(*sp) + 3) & ~0x1;
-	smp_mb(); /* Above access must not bleed into critical section. */
-	return s;
-}
-
-/*
- * Given a snapshot from rcu_seq_snap(), determine whether or not a
- * full update-side operation has occurred.
- */
-static bool rcu_seq_done(unsigned long *sp, unsigned long s)
-{
-	return ULONG_CMP_GE(READ_ONCE(*sp), s);
-}
-
 /*
  * Check to see if there is any immediate RCU-related work to be done
  * by the current CPU, for the specified type of RCU, returning 1 if so.

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-17 23:47 UTC #94 lore raw

PATCH v2 18/39 rcu: Expedited wakeups need to be fully ordered

Reported-by Dmitry Vyukov <[email protected]>

---
 kernel/rcu/tree_exp.h | 2 ++
 1 file changed, 2 insertions(+)

kernel/rcu/tree_exp.h +2 -0

index 51ca287828a2..027e123d93c7 100644
@@ -533,6 +533,7 @@ static void rcu_exp_wait_wake(struct rcu_state *rsp, unsigned long s)
 				rnp->exp_seq_rq = s;
 			spin_unlock(&rnp->exp_lock);
 		}
+		smp_mb(); /* All above changes before wakeup. */
 		wake_up_all(&rnp->exp_wq[(rsp->expedited_sequence >> 1) & 0x3]);
 	}
 	trace_rcu_exp_grace_period(rsp->name, s, TPS("endwake"));

@@ -614,6 +615,7 @@ static void _synchronize_rcu_expedited(struct rcu_state *rsp,
 	wait_event(rnp->exp_wq[(s >> 1) & 0x3],
 		   sync_exp_work_done(rsp,
 				      &rdp->exp_workdone0, s));
+	smp_mb(); /* Workqueue actions happen before return. */
 
 	/* Let the next expedited grace period start. */
 	mutex_unlock(&rsp->exp_mutex);

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-17 23:51 UTC #95 lore raw

PATCH v2 19/39 rcu: Fix warning in rcu_seq_end()

From: Dmitry Vyukov <[email protected]>

Check that seq has good value before incrementing it.

Signed-off-by Dmitry Vyukov <[email protected]>

Cc Paul E. McKenney <[email protected]>

Cc Steven Rostedt <[email protected]>

Cc Mathieu Desnoyers <[email protected]>

---

syzkaller-triggered warning:

WARNING: CPU: 0 PID: 4832 at kernel/rcu/tree.c:3533
rcu_seq_end+0x110/0x140 kernel/rcu/tree.c:3533
CPU: 0 PID: 4832 Comm: kworker/0:3 Not tainted 4.10.0+ #276
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Bochs 01/01/2011
Workqueue: events wait_rcu_exp_gp
Call Trace:
 __dump_stack lib/dump_stack.c:15 [inline]
 dump_stack+0x2ee/0x3ef lib/dump_stack.c:51
 panic+0x1fb/0x412 kernel/panic.c:179
 __warn+0x1c4/0x1e0 kernel/panic.c:540
 warn_slowpath_null+0x2c/0x40 kernel/panic.c:583
 rcu_seq_end+0x110/0x140 kernel/rcu/tree.c:3533
 rcu_exp_gp_seq_end kernel/rcu/tree_exp.h:36 [inline]
 rcu_exp_wait_wake+0x8a9/0x1330 kernel/rcu/tree_exp.h:517
 rcu_exp_sel_wait_wake kernel/rcu/tree_exp.h:559 [inline]
 wait_rcu_exp_gp+0x83/0xc0 kernel/rcu/tree_exp.h:570
 process_one_work+0xc06/0x1c20 kernel/workqueue.c:2096
 worker_thread+0x223/0x19c0 kernel/workqueue.c:2230
 kthread+0x326/0x3f0 kernel/kthread.c:227
 ret_from_fork+0x31/0x40 arch/x86/entry/entry_64.S:430
---
---
 kernel/rcu/rcu.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

kernel/rcu/rcu.h +1 -1

index 91e0bf31f6ce..0bc1313c49e2 100644
@@ -73,8 +73,8 @@ static inline void rcu_seq_start(unsigned long *sp)
 static inline void rcu_seq_end(unsigned long *sp)
 {
 	smp_mb(); /* Ensure update-side operation before counter increment. */
+	WARN_ON_ONCE(!(*sp & 0x1));
 	WRITE_ONCE(*sp, *sp + 1);
-	WARN_ON_ONCE(*sp & 0x1);
 }
 
 /* Take a snapshot of the update side's sequence number. */

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-17 23:47 UTC #96 lore raw

PATCH v2 20/39 srcu: Push srcu_advance_batches() fastpath into common case

This commit is a step towards numbering callbacks in order to efficiently handle per-CPU callback lists.

---
 kernel/rcu/srcu.c | 27 +++++++--------------------
 1 file changed, 7 insertions(+), 20 deletions(-)

kernel/rcu/srcu.c +7 ~~-20~~

index 1026ce24922f..febf87deaccb 100644
@@ -393,8 +393,7 @@ void call_srcu(struct srcu_struct *sp, struct rcu_head *head,
 }
 EXPORT_SYMBOL_GPL(call_srcu);
 
-static void srcu_advance_batches(struct srcu_struct *sp, int trycount);
-static void srcu_reschedule(struct srcu_struct *sp);
+static void srcu_reschedule(struct srcu_struct *sp, unsigned long delay);
 
 /*
  * Helper function for synchronize_srcu() and synchronize_srcu_expedited().

@@ -403,7 +402,6 @@ static void __synchronize_srcu(struct srcu_struct *sp, int trycount)
 {
 	struct rcu_synchronize rcu;
 	struct rcu_head *head = &rcu.head;
-	bool done = false;
 
 	RCU_LOCKDEP_WARN(lock_is_held(&sp->dep_map) ||
 			 lock_is_held(&rcu_bh_lock_map) ||

@@ -425,25 +423,15 @@ static void __synchronize_srcu(struct srcu_struct *sp, int trycount)
 		sp->running = true;
 		rcu_batch_queue(&sp->batch_check0, head);
 		spin_unlock_irq(&sp->queue_lock);
-
-		srcu_advance_batches(sp, trycount);
-		if (!rcu_batch_empty(&sp->batch_done)) {
-			BUG_ON(sp->batch_done.head != head);
-			rcu_batch_dequeue(&sp->batch_done);
-			done = true;
-		}
 		/* give the processing owner to work_struct */
-		srcu_reschedule(sp);
+		srcu_reschedule(sp, 0);
 	} else {
 		rcu_batch_queue(&sp->batch_queue, head);
 		spin_unlock_irq(&sp->queue_lock);
 	}
 
-	if (!done) {
-		wait_for_completion(&rcu.completion);
-		smp_mb(); /* Caller's later accesses after GP. */
-	}
-
+	wait_for_completion(&rcu.completion);
+	smp_mb(); /* Caller's later accesses after GP. */
 }
 
 /**

@@ -633,7 +621,7 @@ static void srcu_invoke_callbacks(struct srcu_struct *sp)
  * Finished one round of SRCU grace period.  Start another if there are
  * more SRCU callbacks queued, otherwise put SRCU into not-running state.
  */
-static void srcu_reschedule(struct srcu_struct *sp)
+static void srcu_reschedule(struct srcu_struct *sp, unsigned long delay)
 {
 	bool pending = true;

@@ -647,8 +635,7 @@ static void srcu_reschedule(struct srcu_struct *sp)
 	}
 
 	if (pending)
-		queue_delayed_work(system_power_efficient_wq,
-				   &sp->work, SRCU_INTERVAL);
+		queue_delayed_work(system_power_efficient_wq, &sp->work, delay);
 }
 
 /*

@@ -663,6 +650,6 @@ void process_srcu(struct work_struct *work)
 	srcu_collect_new(sp);
 	srcu_advance_batches(sp, 1);
 	srcu_invoke_callbacks(sp);
-	srcu_reschedule(sp);
+	srcu_reschedule(sp, SRCU_INTERVAL);
 }
 EXPORT_SYMBOL_GPL(process_srcu);

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-17 23:48 UTC #97 lore raw

PATCH v2 21/39 srcu: Move to state-based grace-period sequencing

This commit therefore adds states to the SRCU grace-period processing, so that the end of a given SRCU grace period is marked by the transition to the SRCU_STATE_DONE state.

---
 include/linux/srcu.h |  10 ++++-
 kernel/rcu/srcu.c    | 111 ++++++++++++++++++++++++++++-----------------------
 2 files changed, 69 insertions(+), 52 deletions(-)

include/linux/srcu.h +8 -2

index a598cf3ac70c..f149a685896c 100644
@@ -48,7 +48,7 @@ struct srcu_struct {
 	unsigned long completed;
 	struct srcu_array __percpu *per_cpu_ref;
 	spinlock_t queue_lock; /* protect ->batch_queue, ->running */
-	bool running;
+	int srcu_state;
 	/* callbacks just queued */
 	struct rcu_batch batch_queue;
 	/* callbacks try to do the first check_zero */

@@ -62,6 +62,12 @@ struct srcu_struct {
 #endif /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
 };
 
+/* Values for -> state variable. */
+#define SRCU_STATE_IDLE		0
+#define SRCU_STATE_SCAN1	1
+#define SRCU_STATE_SCAN2	2
+#define SRCU_STATE_DONE		3
+
 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 
 int __init_srcu_struct(struct srcu_struct *sp, const char *name,

@@ -89,7 +95,7 @@ void process_srcu(struct work_struct *work);
 		.completed = -300,					\
 		.per_cpu_ref = &name##_srcu_array,			\
 		.queue_lock = __SPIN_LOCK_UNLOCKED(name.queue_lock),	\
-		.running = false,					\
+		.srcu_state = SRCU_STATE_IDLE,				\
 		.batch_queue = RCU_BATCH_INIT(name.batch_queue),	\
 		.batch_check0 = RCU_BATCH_INIT(name.batch_check0),	\
 		.batch_check1 = RCU_BATCH_INIT(name.batch_check1),	\

kernel/rcu/srcu.c +61 ~~-50~~

155 lines not on this page · raw message

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-17 23:48 UTC #98 lore raw

PATCH v2 22/39 srcu: Add grace-period sequence numbers

This commit adds grace-period sequence numbers, which will be used to handle mid-boot grace periods and per-CPU callback lists.

---
 include/linux/srcu.h |  1 +
 kernel/rcu/srcu.c    | 27 +++++++++++++++++++++++----
 2 files changed, 24 insertions(+), 4 deletions(-)

include/linux/srcu.h +1 -0

index f149a685896c..047ac8c28a4e 100644
@@ -46,6 +46,7 @@ struct rcu_batch {
 
 struct srcu_struct {
 	unsigned long completed;
+	unsigned long srcu_gp_seq;
 	struct srcu_array __percpu *per_cpu_ref;
 	spinlock_t queue_lock; /* protect ->batch_queue, ->running */
 	int srcu_state;

kernel/rcu/srcu.c +23 -4

index 84da09c02971..ed7c3d082b9f 100644
@@ -110,6 +110,7 @@ static inline void rcu_batch_move(struct rcu_batch *to, struct rcu_batch *from)
 static int init_srcu_struct_fields(struct srcu_struct *sp)
 {
 	sp->completed = 0;
+	sp->srcu_gp_seq = 0;
 	spin_lock_init(&sp->queue_lock);
 	sp->srcu_state = SRCU_STATE_IDLE;
 	rcu_batch_init(&sp->batch_queue);

@@ -313,6 +314,15 @@ EXPORT_SYMBOL_GPL(__srcu_read_unlock);
 #define SYNCHRONIZE_SRCU_EXP_TRYCOUNT	12
 
 /*
+ * Start an SRCU grace period.
+ */
+static void srcu_gp_start(struct srcu_struct *sp)
+{
+	WRITE_ONCE(sp->srcu_state, SRCU_STATE_SCAN1);
+	rcu_seq_start(&sp->srcu_gp_seq);
+}
+
+/*
  * @@@ Wait until all pre-existing readers complete.  Such readers
  * will have used the index specified by "idx".
  * the caller should ensures the ->completed is not changed while checking

@@ -349,6 +359,15 @@ static void srcu_flip(struct srcu_struct *sp)
 }
 
 /*
+ * End an SRCU grace period.
+ */
+static void srcu_gp_end(struct srcu_struct *sp)
+{
+	rcu_seq_end(&sp->srcu_gp_seq);
+	WRITE_ONCE(sp->srcu_state, SRCU_STATE_DONE);
+}
+
+/*
  * Enqueue an SRCU callback on the specified srcu_struct structure,
  * initiating grace-period processing if it is not already running.
  *

@@ -386,7 +405,7 @@ void call_srcu(struct srcu_struct *sp, struct rcu_head *head,
 	smp_mb__after_unlock_lock(); /* Caller's prior accesses before GP. */
 	rcu_batch_queue(&sp->batch_queue, head);
 	if (READ_ONCE(sp->srcu_state) == SRCU_STATE_IDLE) {
-		WRITE_ONCE(sp->srcu_state, SRCU_STATE_SCAN1);
+		srcu_gp_start(sp);
 		queue_delayed_work(system_power_efficient_wq, &sp->work, 0);
 	}
 	spin_unlock_irqrestore(&sp->queue_lock, flags);

@@ -420,7 +439,7 @@ static void __synchronize_srcu(struct srcu_struct *sp, int trycount)
 	smp_mb__after_unlock_lock(); /* Caller's prior accesses before GP. */
 	if (READ_ONCE(sp->srcu_state) == SRCU_STATE_IDLE) {
 		/* steal the processing owner */
-		WRITE_ONCE(sp->srcu_state, SRCU_STATE_SCAN1);
+		srcu_gp_start(sp);
 		rcu_batch_queue(&sp->batch_check0, head);
 		spin_unlock_irq(&sp->queue_lock);
 		/* give the processing owner to work_struct */

@@ -555,7 +574,7 @@ static void srcu_advance_batches(struct srcu_struct *sp, int trycount)
 	 */
 
 	if (sp->srcu_state == SRCU_STATE_DONE)
-		WRITE_ONCE(sp->srcu_state, SRCU_STATE_SCAN1);
+		srcu_gp_start(sp);
 
 	if (sp->srcu_state == SRCU_STATE_SCAN1) {
 		idx = 1 ^ (sp->completed & 1);

@@ -602,7 +621,7 @@ static void srcu_advance_batches(struct srcu_struct *sp, int trycount)
 		 */
 		rcu_batch_move(&sp->batch_done, &sp->batch_check1);
 
-		WRITE_ONCE(sp->srcu_state, SRCU_STATE_DONE);
+		srcu_gp_end(sp);
 	}
 }

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-17 23:46 UTC #99 lore raw

PATCH v2 23/39 srcu: Use rcu_segcblist to track SRCU callbacks

---
 include/linux/rcu_segcblist.h | 678 ++++++++++++++++++++++++++++++++++++++++++
 include/linux/srcu.h          |  24 +-
 kernel/rcu/rcu.h              |   6 +
 kernel/rcu/rcu_segcblist.h    | 670 -----------------------------------------
 kernel/rcu/srcu.c             | 159 ++--------
 kernel/rcu/tree.h             |   2 +-
 6 files changed, 721 insertions(+), 818 deletions(-)
 create mode 100644 include/linux/rcu_segcblist.h
 delete mode 100644 kernel/rcu/rcu_segcblist.h

include/linux/rcu_segcblist.h new file +678 -0

new file mode 100644
index 000000000000..74b1e7243955
@@ -0,0 +1,678 @@
+/*
+ * RCU segmented callback lists
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, you can access it online at
+ * http://www.gnu.org/licenses/gpl-2.0.html.
+ *
+ * Copyright IBM Corporation, 2017
+ *
+ * Authors: Paul E. McKenney <[email protected]>
+ */
+
+#ifndef __KERNEL_RCU_SEGCBLIST_H
+#define __KERNEL_RCU_SEGCBLIST_H
+
+/* Simple unsegmented callback lists. */
+struct rcu_cblist {
+	struct rcu_head *head;
+	struct rcu_head **tail;
+	long len;
+	long len_lazy;
+};
+
+#define RCU_CBLIST_INITIALIZER(n) { .head = NULL, .tail = &n.head }
+
+/* Initialize simple callback list. */
+static inline void rcu_cblist_init(struct rcu_cblist *rclp)
+{
+	rclp->head = NULL;
+	rclp->tail = &rclp->head;
+	rclp->len = 0;
+	rclp->len_lazy = 0;
+}
+
+/* Is simple callback list empty? */
+static inline bool rcu_cblist_empty(struct rcu_cblist *rclp)
+{
+	return !rclp->head;
+}
+
+/* Return number of callbacks in simple callback list. */
+static inline long rcu_cblist_n_cbs(struct rcu_cblist *rclp)
+{
+	return rclp->len;
+}
+
+/* Return number of lazy callbacks in simple callback list. */
+static inline long rcu_cblist_n_lazy_cbs(struct rcu_cblist *rclp)
+{
+	return rclp->len_lazy;
+}
+
+/*
+ * Debug function to actually count the number of callbacks.
+ * If the number exceeds the limit specified, return -1.
+ */
+static inline long rcu_cblist_count_cbs(struct rcu_cblist *rclp, long lim)
+{
+	int cnt = 0;
+	struct rcu_head **rhpp = &rclp->head;
+
+	for (;;) {
+		if (!*rhpp)
+			return cnt;
+		if (++cnt > lim)
+			return -1;
+		rhpp = &(*rhpp)->next;
+	}
+}
+
+/*
+ * Dequeue the oldest rcu_head structure from the specified callback
+ * list.  This function assumes that the callback is non-lazy, but
+ * the caller can later invoke rcu_cblist_dequeued_lazy() if it
+ * finds otherwise (and if it cares about laziness).  This allows
+ * different users to have different ways of determining laziness.
+ */
+static inline struct rcu_head *rcu_cblist_dequeue(struct rcu_cblist *rclp)
+{
+	struct rcu_head *rhp;
+
+	rhp = rclp->head;
+	if (!rhp)
+		return NULL;
+	rclp->len--;
+	rclp->head = rhp->next;
+	if (!rclp->head)
+		rclp->tail = &rclp->head;
+	return rhp;
+}
+
+/*
+ * Account for the fact that a previously dequeued callback turned out
+ * to be marked as lazy.
+ */
+static inline void rcu_cblist_dequeued_lazy(struct rcu_cblist *rclp)
+{
+	rclp->len_lazy--;
+}
+
+/*
+ * Interim function to return rcu_cblist head pointer.  Longer term, the
+ * rcu_cblist will be used more pervasively, removing the need for this
+ * function.
+ */
+static inline struct rcu_head *rcu_cblist_head(struct rcu_cblist *rclp)
+{
+	return rclp->head;
+}
+
+/*
+ * Interim function to return rcu_cblist head pointer.  Longer term, the
+ * rcu_cblist will be used more pervasively, removing the need for this
+ * function.
+ */
+static inline struct rcu_head **rcu_cblist_tail(struct rcu_cblist *rclp)
+{
+	WARN_ON_ONCE(rcu_cblist_empty(rclp));
+	return rclp->tail;
+}
+
+/* Complicated segmented callback lists.  ;-) */
+
+/*
+ * Index values for segments in rcu_segcblist structure.
+ *
+ * The segments are as follows:
+ *
+ * [head, *tails[RCU_DONE_TAIL]):
+ *	Callbacks whose grace period has elapsed, and thus can be invoked.
+ * [*tails[RCU_DONE_TAIL], *tails[RCU_WAIT_TAIL]):
+ *	Callbacks waiting for the current GP from the current CPU's viewpoint.
+ * [*tails[RCU_WAIT_TAIL], *tails[RCU_NEXT_READY_TAIL]):
+ *	Callbacks that arrived before the next GP started, again from
+ *	the current CPU's viewpoint.  These can be handled by the next GP.
+ * [*tails[RCU_NEXT_READY_TAIL], *tails[RCU_NEXT_TAIL]):
+ *	Callbacks that might have arrived after the next GP started.
+ *	There is some uncertainty as to when a given GP starts and
+ *	ends, but a CPU knows the exact times if it is the one starting
+ *	or ending the GP.  Other CPUs know that the previous GP ends
+ *	before the next one starts.
+ *
+ * Note that RCU_WAIT_TAIL cannot be empty unless RCU_NEXT_READY_TAIL is also
+ * empty.
+ *
+ * The ->gp_seq[] array contains the grace-period number at which the
+ * corresponding segment of callbacks will be ready to invoke.  A given
+ * element of this array is meaningful only when the corresponding segment
+ * is non-empty, and it is never valid for RCU_DONE_TAIL (whose callbacks
+ * are already ready to invoke) or for RCU_NEXT_TAIL (whose callbacks have
+ * not yet been assigned a grace-period number).
+ */
+#define RCU_DONE_TAIL		0	/* Also RCU_WAIT head. */
+#define RCU_WAIT_TAIL		1	/* Also RCU_NEXT_READY head. */
+#define RCU_NEXT_READY_TAIL	2	/* Also RCU_NEXT head. */
+#define RCU_NEXT_TAIL		3
+#define RCU_CBLIST_NSEGS	4
+
+struct rcu_segcblist {
+	struct rcu_head *head;
+	struct rcu_head **tails[RCU_CBLIST_NSEGS];
+	unsigned long gp_seq[RCU_CBLIST_NSEGS];
+	long len;
+	long len_lazy;
+};
+
+#define RCU_SEGCBLIST_INITIALIZER(n) \
+{ \
+	.head = NULL, \
+	.tails[RCU_DONE_TAIL] = &n.head, \
+	.tails[RCU_WAIT_TAIL] = &n.head, \
+	.tails[RCU_NEXT_READY_TAIL] = &n.head, \
+	.tails[RCU_NEXT_TAIL] = &n.head, \
+}
+
+/*
+ * Initialize an rcu_segcblist structure.
+ */
+static inline void rcu_segcblist_init(struct rcu_segcblist *rsclp)
+{
+	int i;
+
+	BUILD_BUG_ON(RCU_NEXT_TAIL + 1 != ARRAY_SIZE(rsclp->gp_seq));
+	BUILD_BUG_ON(ARRAY_SIZE(rsclp->tails) != ARRAY_SIZE(rsclp->gp_seq));
+	rsclp->head = NULL;
+	for (i = 0; i < RCU_CBLIST_NSEGS; i++)
+		rsclp->tails[i] = &rsclp->head;
+	rsclp->len = 0;
+	rsclp->len_lazy = 0;
+}
+
+/*
+ * Is the specified rcu_segcblist structure empty?
+ *
+ * But careful!  The fact that the ->head field is NULL does not
+ * necessarily imply that there are no callbacks associated with
+ * this structure.  When callbacks are being invoked, they are
+ * removed as a group.  If callback invocation must be preempted,
+ * the remaining callbacks will be added back to the list.  Either
+ * way, the counts are updated later.
+ *
+ * So it is often the case that rcu_segcblist_n_cbs() should be used
+ * instead.
+ */
+static inline bool rcu_segcblist_empty(struct rcu_segcblist *rsclp)
+{
+	return !rsclp->head;
+}
+
+/* Return number of callbacks in segmented callback list. */
+static inline long rcu_segcblist_n_cbs(struct rcu_segcblist *rsclp)
+{
+	return READ_ONCE(rsclp->len);
+}
+
+/* Return number of lazy callbacks in segmented callback list. */
+static inline long rcu_segcblist_n_lazy_cbs(struct rcu_segcblist *rsclp)
+{
+	return rsclp->len_lazy;
+}
+
+/* Return number of lazy callbacks in segmented callback list. */
+static inline long rcu_segcblist_n_nonlazy_cbs(struct rcu_segcblist *rsclp)
+{
+	return rsclp->len - rsclp->len_lazy;
+}
+
+/*
+ * Is the specified rcu_segcblist enabled, for example, not corresponding
+ * to an offline or callback-offloaded CPU?
+ */
+static inline bool rcu_segcblist_is_enabled(struct rcu_segcblist *rsclp)
+{
+	return !!rsclp->tails[RCU_NEXT_TAIL];
+}
+
+/*
+ * Disable the specified rcu_segcblist structure, so that callbacks can
+ * no longer be posted to it.  This structure must be empty.
+ */
+static inline void rcu_segcblist_disable(struct rcu_segcblist *rsclp)
+{
+	WARN_ON_ONCE(!rcu_segcblist_empty(rsclp));
+	WARN_ON_ONCE(rcu_segcblist_n_cbs(rsclp));
+	WARN_ON_ONCE(rcu_segcblist_n_lazy_cbs(rsclp));
+	rsclp->tails[RCU_NEXT_TAIL] = NULL;
+}
+
+/*
+ * Is the specified segment of the specified rcu_segcblist structure
+ * empty of callbacks?
+ */
+static inline bool rcu_segcblist_segempty(struct rcu_segcblist *rsclp, int seg)
+{
+	if (seg == RCU_DONE_TAIL)
+		return &rsclp->head == rsclp->tails[RCU_DONE_TAIL];
+	return rsclp->tails[seg - 1] == rsclp->tails[seg];
+}
+
+/*
+ * Are all segments following the specified segment of the specified
+ * rcu_segcblist structure empty of callbacks?  (The specified
+ * segment might well contain callbacks.)
+ */
+static inline bool rcu_segcblist_restempty(struct rcu_segcblist *rsclp, int seg)
+{
+	return !*rsclp->tails[seg];
+}
+
+/*
+ * Does the specified rcu_segcblist structure contain callbacks that
+ * are ready to be invoked?
+ */
+static inline bool rcu_segcblist_ready_cbs(struct rcu_segcblist *rsclp)
+{
+	return rcu_segcblist_is_enabled(rsclp) &&
+	       &rsclp->head != rsclp->tails[RCU_DONE_TAIL];
+}
+
+/*
+ * Does the specified rcu_segcblist structure contain callbacks that
+ * are still pending, that is, not yet ready to be invoked?
+ */
+static inline bool rcu_segcblist_pend_cbs(struct rcu_segcblist *rsclp)
+{
+	return rcu_segcblist_is_enabled(rsclp) &&
+	       !rcu_segcblist_restempty(rsclp, RCU_DONE_TAIL);
+}
+
+/*
+ * Dequeue and return the first ready-to-invoke callback.  If there
+ * are no ready-to-invoke callbacks, return NULL.  Disables interrupts
+ * to avoid interference.  Does not protect from interference from other
+ * CPUs or tasks.
+ */
+static inline struct rcu_head *
+rcu_segcblist_dequeue(struct rcu_segcblist *rsclp)
+{
+	unsigned long flags;
+	int i;
+	struct rcu_head *rhp;
+
+	local_irq_save(flags);
+	if (!rcu_segcblist_ready_cbs(rsclp)) {
+		local_irq_restore(flags);
+		return NULL;
+	}
+	rhp = rsclp->head;
+	BUG_ON(!rhp);
+	rsclp->head = rhp->next;
+	for (i = RCU_DONE_TAIL; i < RCU_CBLIST_NSEGS; i++) {
+		if (rsclp->tails[i] != &rhp->next)
+			break;
+		rsclp->tails[i] = &rsclp->head;
+	}
+	smp_mb(); /* Dequeue before decrement for rcu_barrier(). */
+	WRITE_ONCE(rsclp->len, rsclp->len - 1);
+	local_irq_restore(flags);
+	return rhp;
+}
+
+/*
+ * Account for the fact that a previously dequeued callback turned out
+ * to be marked as lazy.
+ */
+static inline void rcu_segcblist_dequeued_lazy(struct rcu_segcblist *rsclp)
+{
+	unsigned long flags;
+
+	local_irq_save(flags);
+	rsclp->len_lazy--;
+	local_irq_restore(flags);
+}
+
+/*
+ * Return a pointer to the first callback in the specified rcu_segcblist
+ * structure.  This is useful for diagnostics.
+ */
+static inline struct rcu_head *
+rcu_segcblist_first_cb(struct rcu_segcblist *rsclp)
+{
+	if (rcu_segcblist_is_enabled(rsclp))
+		return rsclp->head;
+	return NULL;
+}
+
+/*
+ * Return a pointer to the first pending callback in the specified
+ * rcu_segcblist structure.  This is useful just after posting a given
+ * callback -- if that callback is the first pending callback, then
+ * you cannot rely on someone else having already started up the required
+ * grace period.
+ */
+static inline struct rcu_head *
+rcu_segcblist_first_pend_cb(struct rcu_segcblist *rsclp)
+{
+	if (rcu_segcblist_is_enabled(rsclp))
+		return *rsclp->tails[RCU_DONE_TAIL];
+	return NULL;
+}
+
+/*
+ * Does the specified rcu_segcblist structure contain callbacks that
+ * have not yet been processed beyond having been posted, that is,
+ * does it contain callbacks in its last segment?
+ */
+static inline bool rcu_segcblist_new_cbs(struct rcu_segcblist *rsclp)
+{
+	return rcu_segcblist_is_enabled(rsclp) &&
+	       !rcu_segcblist_restempty(rsclp, RCU_NEXT_READY_TAIL);
+}
+
+/*
+ * Enqueue the specified callback onto the specified rcu_segcblist
+ * structure, updating accounting as needed.  Note that the ->len
+ * field may be accessed locklessly, hence the WRITE_ONCE().
+ * The ->len field is used by rcu_barrier() and friends to determine
+ * if it must post a callback on this structure, and it is OK
+ * for rcu_barrier() to sometimes post callbacks needlessly, but
+ * absolutely not OK for it to ever miss posting a callback.
+ */
+static inline void rcu_segcblist_enqueue(struct rcu_segcblist *rsclp,
+					 struct rcu_head *rhp, bool lazy)
+{
+	WRITE_ONCE(rsclp->len, rsclp->len + 1); /* ->len sampled locklessly. */
+	if (lazy)
+		rsclp->len_lazy++;
+	smp_mb(); /* Ensure counts are updated before callback is enqueued. */
+	rhp->next = NULL;
+	*rsclp->tails[RCU_NEXT_TAIL] = rhp;
+	rsclp->tails[RCU_NEXT_TAIL] = &rhp->next;
+}
+
+/*
+ * Extract only the counts from the specified rcu_segcblist structure,
+ * and place them in the specified rcu_cblist structure.  This function
+ * supports both callback orphaning and invocation, hence the separation
+ * of counts and callbacks.  (Callbacks ready for invocation must be
+ * orphaned and adopted separately from pending callbacks, but counts
+ * apply to all callbacks.  Locking must be used to make sure that
+ * both orphaned-callbacks lists are consistent.)
+ */
+static inline void rcu_segcblist_extract_count(struct rcu_segcblist *rsclp,
+					       struct rcu_cblist *rclp)
+{
+	rclp->len_lazy += rsclp->len_lazy;
+	rclp->len += rsclp->len;
+	rsclp->len_lazy = 0;
+	WRITE_ONCE(rsclp->len, 0); /* ->len sampled locklessly. */
+}
+
+/*
+ * Extract only those callbacks ready to be invoked from the specified
+ * rcu_segcblist structure and place them in the specified rcu_cblist
+ * structure.
+ */
+static inline void rcu_segcblist_extract_done_cbs(struct rcu_segcblist *rsclp,
+						  struct rcu_cblist *rclp)
+{
+	int i;
+
+	if (!rcu_segcblist_ready_cbs(rsclp))
+		return; /* Nothing to do. */
+	*rclp->tail = rsclp->head;
+	rsclp->head = *rsclp->tails[RCU_DONE_TAIL];
+	*rsclp->tails[RCU_DONE_TAIL] = NULL;
+	rclp->tail = rsclp->tails[RCU_DONE_TAIL];
+	for (i = RCU_CBLIST_NSEGS - 1; i >= RCU_DONE_TAIL; i--)
+		if (rsclp->tails[i] == rsclp->tails[RCU_DONE_TAIL])
+			rsclp->tails[i] = &rsclp->head;
+}
+
+/*
+ * Extract only those callbacks still pending (not yet ready to be
+ * invoked) from the specified rcu_segcblist structure and place them in
+ * the specified rcu_cblist structure.  Note that this loses information
+ * about any callbacks that might have been partway done waiting for
+ * their grace period.  Too bad!  They will have to start over.
+ */
+static inline void
+rcu_segcblist_extract_pend_cbs(struct rcu_segcblist *rsclp,
+			       struct rcu_cblist *rclp)
+{
+	int i;
+
+	if (!rcu_segcblist_pend_cbs(rsclp))
+		return; /* Nothing to do. */
+	*rclp->tail = *rsclp->tails[RCU_DONE_TAIL];
+	rclp->tail = rsclp->tails[RCU_NEXT_TAIL];
+	*rsclp->tails[RCU_DONE_TAIL] = NULL;
+	for (i = RCU_DONE_TAIL + 1; i < RCU_CBLIST_NSEGS; i++)
+		rsclp->tails[i] = rsclp->tails[RCU_DONE_TAIL];
+}
+
+/*
+ * Move the entire contents of the specified rcu_segcblist structure,
+ * counts, callbacks, and all, to the specified rcu_cblist structure.
+ * @@@ Why do we need this???  Moving early-boot CBs to NOCB lists?
+ * @@@ Memory barrier needed?  (Not if only used at boot time...)
+ */
+static inline void rcu_segcblist_extract_all(struct rcu_segcblist *rsclp,
+					     struct rcu_cblist *rclp)
+{
+	rcu_segcblist_extract_done_cbs(rsclp, rclp);
+	rcu_segcblist_extract_pend_cbs(rsclp, rclp);
+	rcu_segcblist_extract_count(rsclp, rclp);
+}
+
+/*
+ * Insert counts from the specified rcu_cblist structure in the
+ * specified rcu_segcblist structure.
+ */
+static inline void rcu_segcblist_insert_count(struct rcu_segcblist *rsclp,
+					      struct rcu_cblist *rclp)
+{
+	rsclp->len_lazy += rclp->len_lazy;
+	/* ->len sampled locklessly. */
+	WRITE_ONCE(rsclp->len, rsclp->len + rclp->len);
+	rclp->len_lazy = 0;
+	rclp->len = 0;
+}
+
+/*
+ * Move callbacks from the specified rcu_cblist to the beginning of the
+ * done-callbacks segment of the specified rcu_segcblist.
+ */
+static inline void rcu_segcblist_insert_done_cbs(struct rcu_segcblist *rsclp,
+						 struct rcu_cblist *rclp)
+{
+	int i;
+
+	if (!rclp->head)
+		return; /* No callbacks to move. */
+	*rclp->tail = rsclp->head;
+	rsclp->head = rclp->head;
+	for (i = RCU_DONE_TAIL; i < RCU_CBLIST_NSEGS; i++)
+		if (&rsclp->head == rsclp->tails[i])
+			rsclp->tails[i] = rclp->tail;
+		else
+			break;
+	rclp->head = NULL;
+	rclp->tail = &rclp->head;
+}
+
+/*
+ * Move callbacks from the specified rcu_cblist to the end of the
+ * new-callbacks segment of the specified rcu_segcblist.
+ */
+static inline void rcu_segcblist_insert_pend_cbs(struct rcu_segcblist *rsclp,
+						 struct rcu_cblist *rclp)
+{
+	if (!rclp->head)
+		return; /* Nothing to do. */
+	*rsclp->tails[RCU_NEXT_TAIL] = rclp->head;
+	rsclp->tails[RCU_NEXT_TAIL] = rclp->tail;
+	rclp->head = NULL;
+	rclp->tail = &rclp->head;
+}
+
+/*
+ * Advance the callbacks in the specified rcu_segcblist structure based
+ * on the current value passed in for the grace-period counter.
+ */
+static inline void rcu_segcblist_advance(struct rcu_segcblist *rsclp,
+					 unsigned long seq)
+{
+	int i, j;
+
+	WARN_ON_ONCE(!rcu_segcblist_is_enabled(rsclp));
+	WARN_ON_ONCE(rcu_segcblist_restempty(rsclp, RCU_DONE_TAIL));
+
+	/*
+	 * Find all callbacks whose ->gp_seq numbers indicate that they
+	 * are ready to invoke, and put them into the RCU_DONE_TAIL segment.
+	 */
+	for (i = RCU_WAIT_TAIL; i < RCU_NEXT_TAIL; i++) {
+		if (ULONG_CMP_LT(seq, rsclp->gp_seq[i]))
+			break;
+		rsclp->tails[RCU_DONE_TAIL] = rsclp->tails[i];
+	}
+
+	/* If no callbacks moved, nothing more need be done. */
+	if (i == RCU_WAIT_TAIL)
+		return;
+
+	/* Clean up tail pointers that might have been misordered above. */
+	for (j = RCU_WAIT_TAIL; j < i; j++)
+		rsclp->tails[j] = rsclp->tails[RCU_DONE_TAIL];
+
+	/*
+	 * Callbacks moved, so clean up the misordered ->tails[] pointers
+	 * that now point into the middle of the list of ready-to-invoke
+	 * callbacks.  The overall effect is to copy down the later pointers
+	 * into the gap that was created by the now-ready segments.
+	 */
+	for (j = RCU_WAIT_TAIL; i < RCU_NEXT_TAIL; i++, j++) {
+		if (rsclp->tails[j] == rsclp->tails[RCU_NEXT_TAIL])
+			break;  /* No more callbacks. */
+		rsclp->tails[j] = rsclp->tails[i];
+		rsclp->gp_seq[j] = rsclp->gp_seq[i];
+	}
+}
+
+/*
+ * "Accelerate" callbacks based on more-accurate grace-period information.
+ * The reason for this is that RCU does not synchronize the beginnings and
+ * ends of grace periods, and that callbacks are posted locally.  This in
+ * turn means that the callbacks must be labelled conservatively early
+ * on, as getting exact information would degrade both performance and
+ * scalability.  When more accurate grace-period information becomes
+ * available, previously posted callbacks can be "accelerated", marking
+ * them to complete at the end of the earlier grace period.
+ *
+ * This function operates on an rcu_segcblist structure, and also the
+ * grace-period sequence number at which new callbacks would become
+ * ready to invoke.
+ */
+static inline bool rcu_segcblist_accelerate(struct rcu_segcblist *rsclp,
+					    unsigned long seq)
+{
+	int i;
+
+	WARN_ON_ONCE(!rcu_segcblist_is_enabled(rsclp));
+	WARN_ON_ONCE(rcu_segcblist_restempty(rsclp, RCU_DONE_TAIL));
+
+	/*
+	 * Find the segment preceding the oldest segment of callbacks
+	 * whose ->gp_seq[] completion is at or after that passed in via
+	 * "seq", skipping any empty segments.  This oldest segment, along
+	 * with any later segments, can be merged in with any newly arrived
+	 * callbacks in the RCU_NEXT_TAIL segment, and assigned "seq"
+	 * as their ->gp_seq[] grace-period completion sequence number.
+	 */
+	for (i = RCU_NEXT_READY_TAIL; i > RCU_DONE_TAIL; i--)
+		if (rsclp->tails[i] != rsclp->tails[i - 1] &&
+		    ULONG_CMP_LT(rsclp->gp_seq[i], seq))
+			break;
+
+	/*
+	 * If all the segments contain callbacks that correspond to
+	 * earlier grace-period sequence numbers than "seq", leave.
+	 * Assuming that the rcu_segcblist structure has enough
+	 * segments in its arrays, this can only happen if some of
+	 * the non-done segments contain callbacks that really are
+	 * ready to invoke.  This situation will get straightened
+	 * out by the next call to rcu_segcblist_advance().
+	 *
+	 * Also advance to the oldest segment of callbacks whose
+	 * ->gp_seq[] completion is at or after that passed in via "seq",
+	 * skipping any empty segments.
+	 */
+	if (++i >= RCU_NEXT_TAIL)
+		return false;
+
+	/*
+	 * Merge all later callbacks, including newly arrived callbacks,
+	 * into the segment located by the for-loop above.  Assign "seq"
+	 * as the ->gp_seq[] value in order to correctly handle the case
+	 * where there were no pending callbacks in the rcu_segcblist
+	 * structure other than in the RCU_NEXT_TAIL segment.
+	 */
+	for (; i < RCU_NEXT_TAIL; i++) {
+		rsclp->tails[i] = rsclp->tails[RCU_NEXT_TAIL];
+		rsclp->gp_seq[i] = seq;
+	}
+	return true;
+}
+
+/*
+ * Scan the specified rcu_segcblist structure for callbacks that need
+ * a grace period later than the one specified by "seq".  We don't look
+ * at the RCU_DONE_TAIL or RCU_NEXT_TAIL segments because they don't
+ * have a grace-period sequence number.
+ */
+static inline bool rcu_segcblist_future_gp_needed(struct rcu_segcblist *rsclp,
+						  unsigned long seq)
+{
+	int i;
+
+	for (i = RCU_WAIT_TAIL; i < RCU_NEXT_TAIL; i++)
+		if (rsclp->tails[i - 1] != rsclp->tails[i] &&
+		    ULONG_CMP_LT(seq, rsclp->gp_seq[i]))
+			return true;
+	return false;
+}
+
+/*
+ * Interim function to return rcu_segcblist head pointer.  Longer term, the
+ * rcu_segcblist will be used more pervasively, removing the need for this
+ * function.
+ */
+static inline struct rcu_head *rcu_segcblist_head(struct rcu_segcblist *rsclp)
+{
+	return rsclp->head;
+}
+
+/*
+ * Interim function to return rcu_segcblist head pointer.  Longer term, the
+ * rcu_segcblist will be used more pervasively, removing the need for this
+ * function.
+ */
+static inline struct rcu_head **rcu_segcblist_tail(struct rcu_segcblist *rsclp)
+{
+	WARN_ON_ONCE(rcu_segcblist_empty(rsclp));
+	return rsclp->tails[RCU_NEXT_TAIL];
+}
+
+#endif /* __KERNEL_RCU_SEGCBLIST_H */

include/linux/srcu.h +5 ~~-19~~

index 047ac8c28a4e..ad154a7bc114 100644
@@ -22,7 +22,7 @@
  *	   Lai Jiangshan <[email protected]>
  *
  * For detailed explanation of Read-Copy Update mechanism see -
- * 		Documentation/RCU/ *.txt
+ *		Documentation/RCU/ *.txt
  *
  */

@@ -32,31 +32,20 @@
 #include <linux/mutex.h>
 #include <linux/rcupdate.h>
 #include <linux/workqueue.h>
+#include <linux/rcu_segcblist.h>
 
 struct srcu_array {
 	unsigned long lock_count[2];
 	unsigned long unlock_count[2];
 };
 
-struct rcu_batch {
-	struct rcu_head *head, **tail;
-};
-
-#define RCU_BATCH_INIT(name) { NULL, &(name.head) }
-
 struct srcu_struct {
 	unsigned long completed;
 	unsigned long srcu_gp_seq;
 	struct srcu_array __percpu *per_cpu_ref;
-	spinlock_t queue_lock; /* protect ->batch_queue, ->running */
+	spinlock_t queue_lock; /* protect ->srcu_cblist, ->srcu_state */
 	int srcu_state;
-	/* callbacks just queued */
-	struct rcu_batch batch_queue;
-	/* callbacks try to do the first check_zero */
-	struct rcu_batch batch_check0;
-	/* callbacks done with the first check_zero and the flip */
-	struct rcu_batch batch_check1;
-	struct rcu_batch batch_done;
+	struct rcu_segcblist srcu_cblist;
 	struct delayed_work work;
 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 	struct lockdep_map dep_map;

@@ -97,10 +86,7 @@ void process_srcu(struct work_struct *work);
 		.per_cpu_ref = &name##_srcu_array,			\
 		.queue_lock = __SPIN_LOCK_UNLOCKED(name.queue_lock),	\
 		.srcu_state = SRCU_STATE_IDLE,				\
-		.batch_queue = RCU_BATCH_INIT(name.batch_queue),	\
-		.batch_check0 = RCU_BATCH_INIT(name.batch_check0),	\
-		.batch_check1 = RCU_BATCH_INIT(name.batch_check1),	\
-		.batch_done = RCU_BATCH_INIT(name.batch_done),		\
+		.srcu_cblist = RCU_SEGCBLIST_INITIALIZER(name.srcu_cblist),\
 		.work = __DELAYED_WORK_INITIALIZER(name.work, process_srcu, 0),\
 		__SRCU_DEP_MAP_INIT(name)				\
 	}

kernel/rcu/rcu.h +6 -0

index 0bc1313c49e2..a943b42a9cf7 100644
@@ -87,6 +87,12 @@ static inline unsigned long rcu_seq_snap(unsigned long *sp)
 	return s;
 }
 
+/* Return the current value the update side's sequence number, no ordering. */
+static inline unsigned long rcu_seq_current(unsigned long *sp)
+{
+	return READ_ONCE(*sp);
+}
+
 /*
  * Given a snapshot from rcu_seq_snap(), determine whether or not a
  * full update-side operation has occurred.

kernel/rcu/rcu_segcblist.h deleted +0 ~~-670~~

670 lines not on this page · raw message

kernel/rcu/srcu.c +31 ~~-128~~

254 lines not on this page · raw message

kernel/rcu/tree.h +1 -1

index 93889ff21dbb..4f62651588ea 100644
@@ -30,7 +30,7 @@
 #include <linux/seqlock.h>
 #include <linux/swait.h>
 #include <linux/stop_machine.h>
-#include "rcu_segcblist.h"
+#include <linux/rcu_segcblist.h>
 
 /*
  * Define shape of hierarchy based on NR_CPUS, CONFIG_RCU_FANOUT, and

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-17 23:47 UTC #100 lore raw

PATCH v2 24/39 srcu: Move combining-tree definitions for SRCU's benefit

This commit is code-movement only.

---
 include/linux/rcu_node_tree.h | 102 ++++++++++++++++++++++++++++++++++++++++++
 kernel/rcu/tree.h             |  71 +----------------------------
 2 files changed, 103 insertions(+), 70 deletions(-)
 create mode 100644 include/linux/rcu_node_tree.h

include/linux/rcu_node_tree.h new file +102 -0

new file mode 100644
index 000000000000..b7eb97096b1c
@@ -0,0 +1,102 @@
+/*
+ * RCU node combining tree definitions.  These are used to compute
+ * global attributes while avoiding common-case global contention.  A key
+ * property that these computations rely on is a tournament-style approach
+ * where only one of the tasks contending a lower level in the tree need
+ * advance to the next higher level.  If properly configured, this allows
+ * unlimited scalability while maintaining a constant level of contention
+ * on the root node.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, you can access it online at
+ * http://www.gnu.org/licenses/gpl-2.0.html.
+ *
+ * Copyright IBM Corporation, 2017
+ *
+ * Author: Paul E. McKenney <[email protected]>
+ */
+
+#ifndef __LINUX_RCU_NODE_TREE_H
+#define __LINUX_RCU_NODE_TREE_H
+
+/*
+ * Define shape of hierarchy based on NR_CPUS, CONFIG_RCU_FANOUT, and
+ * CONFIG_RCU_FANOUT_LEAF.
+ * In theory, it should be possible to add more levels straightforwardly.
+ * In practice, this did work well going from three levels to four.
+ * Of course, your mileage may vary.
+ */
+
+#ifdef CONFIG_RCU_FANOUT
+#define RCU_FANOUT CONFIG_RCU_FANOUT
+#else /* #ifdef CONFIG_RCU_FANOUT */
+# ifdef CONFIG_64BIT
+# define RCU_FANOUT 64
+# else
+# define RCU_FANOUT 32
+# endif
+#endif /* #else #ifdef CONFIG_RCU_FANOUT */
+
+#ifdef CONFIG_RCU_FANOUT_LEAF
+#define RCU_FANOUT_LEAF CONFIG_RCU_FANOUT_LEAF
+#else /* #ifdef CONFIG_RCU_FANOUT_LEAF */
+#define RCU_FANOUT_LEAF 16
+#endif /* #else #ifdef CONFIG_RCU_FANOUT_LEAF */
+
+#define RCU_FANOUT_1	      (RCU_FANOUT_LEAF)
+#define RCU_FANOUT_2	      (RCU_FANOUT_1 * RCU_FANOUT)
+#define RCU_FANOUT_3	      (RCU_FANOUT_2 * RCU_FANOUT)
+#define RCU_FANOUT_4	      (RCU_FANOUT_3 * RCU_FANOUT)
+
+#if NR_CPUS <= RCU_FANOUT_1
+#  define RCU_NUM_LVLS	      1
+#  define NUM_RCU_LVL_0	      1
+#  define NUM_RCU_NODES	      NUM_RCU_LVL_0
+#  define NUM_RCU_LVL_INIT    { NUM_RCU_LVL_0 }
+#  define RCU_NODE_NAME_INIT  { "rcu_node_0" }
+#  define RCU_FQS_NAME_INIT   { "rcu_node_fqs_0" }
+#elif NR_CPUS <= RCU_FANOUT_2
+#  define RCU_NUM_LVLS	      2
+#  define NUM_RCU_LVL_0	      1
+#  define NUM_RCU_LVL_1	      DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
+#  define NUM_RCU_NODES	      (NUM_RCU_LVL_0 + NUM_RCU_LVL_1)
+#  define NUM_RCU_LVL_INIT    { NUM_RCU_LVL_0, NUM_RCU_LVL_1 }
+#  define RCU_NODE_NAME_INIT  { "rcu_node_0", "rcu_node_1" }
+#  define RCU_FQS_NAME_INIT   { "rcu_node_fqs_0", "rcu_node_fqs_1" }
+#elif NR_CPUS <= RCU_FANOUT_3
+#  define RCU_NUM_LVLS	      3
+#  define NUM_RCU_LVL_0	      1
+#  define NUM_RCU_LVL_1	      DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_2)
+#  define NUM_RCU_LVL_2	      DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
+#  define NUM_RCU_NODES	      (NUM_RCU_LVL_0 + NUM_RCU_LVL_1 + NUM_RCU_LVL_2)
+#  define NUM_RCU_LVL_INIT    { NUM_RCU_LVL_0, NUM_RCU_LVL_1, NUM_RCU_LVL_2 }
+#  define RCU_NODE_NAME_INIT  { "rcu_node_0", "rcu_node_1", "rcu_node_2" }
+#  define RCU_FQS_NAME_INIT   { "rcu_node_fqs_0", "rcu_node_fqs_1", "rcu_node_fqs_2" }
+#elif NR_CPUS <= RCU_FANOUT_4
+#  define RCU_NUM_LVLS	      4
+#  define NUM_RCU_LVL_0	      1
+#  define NUM_RCU_LVL_1	      DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_3)
+#  define NUM_RCU_LVL_2	      DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_2)
+#  define NUM_RCU_LVL_3	      DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
+#  define NUM_RCU_NODES	      (NUM_RCU_LVL_0 + NUM_RCU_LVL_1 + NUM_RCU_LVL_2 + NUM_RCU_LVL_3)
+#  define NUM_RCU_LVL_INIT    { NUM_RCU_LVL_0, NUM_RCU_LVL_1, NUM_RCU_LVL_2, NUM_RCU_LVL_3 }
+#  define RCU_NODE_NAME_INIT  { "rcu_node_0", "rcu_node_1", "rcu_node_2", "rcu_node_3" }
+#  define RCU_FQS_NAME_INIT   { "rcu_node_fqs_0", "rcu_node_fqs_1", "rcu_node_fqs_2", "rcu_node_fqs_3" }
+#else
+# error "CONFIG_RCU_FANOUT insufficient for NR_CPUS"
+#endif /* #if (NR_CPUS) <= RCU_FANOUT_1 */
+
+extern int rcu_num_lvls;
+extern int rcu_num_nodes;
+
+#endif /* __LINUX_RCU_NODE_TREE_H */

kernel/rcu/tree.h +1 ~~-70~~

77 lines not on this page · raw message

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-17 23:46 UTC #101 lore raw

PATCH v2 25/39 srcu: Move rcu_init_levelspread() to rcu_tree_node.h

---
 include/linux/rcu_node_tree.h |  3 ---
 kernel/rcu/rcu.h              | 36 ++++++++++++++++++++++++++++++++++++
 kernel/rcu/srcu.c             |  1 +
 kernel/rcu/tree.c             | 25 -------------------------
 kernel/rcu/tree_trace.c       |  1 +
 5 files changed, 38 insertions(+), 28 deletions(-)

include/linux/rcu_node_tree.h +0 -3

index b7eb97096b1c..4b766b61e1a0 100644
@@ -96,7 +96,4 @@
 # error "CONFIG_RCU_FANOUT insufficient for NR_CPUS"
 #endif /* #if (NR_CPUS) <= RCU_FANOUT_1 */
 
-extern int rcu_num_lvls;
-extern int rcu_num_nodes;
-
 #endif /* __LINUX_RCU_NODE_TREE_H */

kernel/rcu/rcu.h +36 -0

index a943b42a9cf7..87326479b39a 100644
@@ -190,4 +190,40 @@ void rcu_test_sync_prims(void);
  */
 extern void resched_cpu(int cpu);
 
+#if defined(SRCU) || !defined(TINY_RCU)
+
+#include <linux/rcu_node_tree.h>
+
+extern int rcu_num_lvls;
+extern int rcu_num_nodes;
+static bool rcu_fanout_exact;
+static int rcu_fanout_leaf;
+
+/*
+ * Compute the per-level fanout, either using the exact fanout specified
+ * or balancing the tree, depending on the rcu_fanout_exact boot parameter.
+ */
+static inline void rcu_init_levelspread(int *levelspread, const int *levelcnt)
+{
+	int i;
+
+	if (rcu_fanout_exact) {
+		levelspread[rcu_num_lvls - 1] = rcu_fanout_leaf;
+		for (i = rcu_num_lvls - 2; i >= 0; i--)
+			levelspread[i] = RCU_FANOUT;
+	} else {
+		int ccur;
+		int cprv;
+
+		cprv = nr_cpu_ids;
+		for (i = rcu_num_lvls - 1; i >= 0; i--) {
+			ccur = levelcnt[i];
+			levelspread[i] = (cprv + ccur - 1) / ccur;
+			cprv = ccur;
+		}
+	}
+}
+
+#endif /* #if defined(SRCU) || !defined(TINY_RCU) */
+
 #endif /* __LINUX_RCU_H */

kernel/rcu/srcu.c +1 -0

index 132cd57c07f1..d51ab050f777 100644
@@ -36,6 +36,7 @@
 #include <linux/delay.h>
 #include <linux/srcu.h>
 
+#include <linux/rcu_node_tree.h>
 #include "rcu.h"
 
 static int init_srcu_struct_fields(struct srcu_struct *sp)

kernel/rcu/tree.c +0 ~~-25~~

index a3eff40a6c17..436eaa3047e6 100644
@@ -3952,31 +3952,6 @@ void rcu_scheduler_starting(void)
 }
 
 /*
- * Compute the per-level fanout, either using the exact fanout specified
- * or balancing the tree, depending on the rcu_fanout_exact boot parameter.
- */
-static void __init rcu_init_levelspread(int *levelspread, const int *levelcnt)
-{
-	int i;
-
-	if (rcu_fanout_exact) {
-		levelspread[rcu_num_lvls - 1] = rcu_fanout_leaf;
-		for (i = rcu_num_lvls - 2; i >= 0; i--)
-			levelspread[i] = RCU_FANOUT;
-	} else {
-		int ccur;
-		int cprv;
-
-		cprv = nr_cpu_ids;
-		for (i = rcu_num_lvls - 1; i >= 0; i--) {
-			ccur = levelcnt[i];
-			levelspread[i] = (cprv + ccur - 1) / ccur;
-			cprv = ccur;
-		}
-	}
-}
-
-/*
  * Helper function for rcu_init() that initializes one rcu_state structure.
  */
 static void __init rcu_init_one(struct rcu_state *rsp)

kernel/rcu/tree_trace.c +1 -0

index 066c64071a7b..30c5bf89ee58 100644
@@ -45,6 +45,7 @@
 
 #define RCU_TREE_NONCORE
 #include "tree.h"
+#include "rcu.h"
 
 static int r_open(struct inode *inode, struct file *file,
 					const struct seq_operations *op)

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-17 23:51 UTC #102 lore raw

PATCH v2 26/39 rcu: Remove redundant levelcnt[] array from rcu_init_one()

The levelcnt[] array is identical to num_rcu_lvl[], so this commit removes levelcnt[].

---
 kernel/rcu/tree.c | 9 +++------
 1 file changed, 3 insertions(+), 6 deletions(-)

kernel/rcu/tree.c +3 -6

index 436eaa3047e6..21140192fd76 100644
@@ -3961,7 +3961,6 @@ static void __init rcu_init_one(struct rcu_state *rsp)
 	static struct lock_class_key rcu_node_class[RCU_NUM_LVLS];
 	static struct lock_class_key rcu_fqs_class[RCU_NUM_LVLS];
 
-	int levelcnt[RCU_NUM_LVLS];		/* # nodes in each level. */
 	int levelspread[RCU_NUM_LVLS];		/* kids/node in each level. */
 	int cpustride = 1;
 	int i;

@@ -3976,18 +3975,16 @@ static void __init rcu_init_one(struct rcu_state *rsp)
 
 	/* Initialize the level-tracking arrays. */
 
-	for (i = 0; i < rcu_num_lvls; i++)
-		levelcnt[i] = num_rcu_lvl[i];
 	for (i = 1; i < rcu_num_lvls; i++)
-		rsp->level[i] = rsp->level[i - 1] + levelcnt[i - 1];
-	rcu_init_levelspread(levelspread, levelcnt);
+		rsp->level[i] = rsp->level[i - 1] + num_rcu_lvl[i - 1];
+	rcu_init_levelspread(levelspread, num_rcu_lvl);
 
 	/* Initialize the elements themselves, starting from the leaves. */
 
 	for (i = rcu_num_lvls - 1; i >= 0; i--) {
 		cpustride *= levelspread[i];
 		rnp = rsp->level[i];
-		for (j = 0; j < levelcnt[i]; j++, rnp++) {
+		for (j = 0; j < num_rcu_lvl[i]; j++, rnp++) {
 			raw_spin_lock_init(&ACCESS_PRIVATE(rnp, lock));
 			lockdep_set_class_and_name(&ACCESS_PRIVATE(rnp, lock),
 						   &rcu_node_class[i], buf[i]);

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-17 23:47 UTC #103 lore raw

PATCH v2 27/39 srcu: Move rcu_node traversal macros to rcu.h

---
 kernel/rcu/rcu.h  | 35 +++++++++++++++++++++++++++++++++++
 kernel/rcu/tree.h | 35 -----------------------------------
 2 files changed, 35 insertions(+), 35 deletions(-)

kernel/rcu/rcu.h +35 -0

index 87326479b39a..5c1798ec32f7 100644
@@ -224,6 +224,41 @@ static inline void rcu_init_levelspread(int *levelspread, const int *levelcnt)
 	}
 }
 
+/*
+ * Do a full breadth-first scan of the rcu_node structures for the
+ * specified rcu_state structure.
+ */
+#define rcu_for_each_node_breadth_first(rsp, rnp) \
+	for ((rnp) = &(rsp)->node[0]; \
+	     (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++)
+
+/*
+ * Do a breadth-first scan of the non-leaf rcu_node structures for the
+ * specified rcu_state structure.  Note that if there is a singleton
+ * rcu_node tree with but one rcu_node structure, this loop is a no-op.
+ */
+#define rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) \
+	for ((rnp) = &(rsp)->node[0]; \
+	     (rnp) < (rsp)->level[rcu_num_lvls - 1]; (rnp)++)
+
+/*
+ * Scan the leaves of the rcu_node hierarchy for the specified rcu_state
+ * structure.  Note that if there is a singleton rcu_node tree with but
+ * one rcu_node structure, this loop -will- visit the rcu_node structure.
+ * It is still a leaf node, even if it is also the root node.
+ */
+#define rcu_for_each_leaf_node(rsp, rnp) \
+	for ((rnp) = (rsp)->level[rcu_num_lvls - 1]; \
+	     (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++)
+
+/*
+ * Iterate over all possible CPUs in a leaf RCU node.
+ */
+#define for_each_leaf_node_possible_cpu(rnp, cpu) \
+	for ((cpu) = cpumask_next(rnp->grplo - 1, cpu_possible_mask); \
+	     cpu <= rnp->grphi; \
+	     cpu = cpumask_next((cpu), cpu_possible_mask))
+
 #endif /* #if defined(SRCU) || !defined(TINY_RCU) */
 
 #endif /* __LINUX_RCU_H */

kernel/rcu/tree.h +0 ~~-35~~

index 1bec3958d44f..a2a45cb629d6 100644
@@ -193,41 +193,6 @@ struct rcu_node {
 #define leaf_node_cpu_bit(rnp, cpu) (1UL << ((cpu) - (rnp)->grplo))
 
 /*
- * Do a full breadth-first scan of the rcu_node structures for the
- * specified rcu_state structure.
- */
-#define rcu_for_each_node_breadth_first(rsp, rnp) \
-	for ((rnp) = &(rsp)->node[0]; \
-	     (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++)
-
-/*
- * Do a breadth-first scan of the non-leaf rcu_node structures for the
- * specified rcu_state structure.  Note that if there is a singleton
- * rcu_node tree with but one rcu_node structure, this loop is a no-op.
- */
-#define rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) \
-	for ((rnp) = &(rsp)->node[0]; \
-	     (rnp) < (rsp)->level[rcu_num_lvls - 1]; (rnp)++)
-
-/*
- * Scan the leaves of the rcu_node hierarchy for the specified rcu_state
- * structure.  Note that if there is a singleton rcu_node tree with but
- * one rcu_node structure, this loop -will- visit the rcu_node structure.
- * It is still a leaf node, even if it is also the root node.
- */
-#define rcu_for_each_leaf_node(rsp, rnp) \
-	for ((rnp) = (rsp)->level[rcu_num_lvls - 1]; \
-	     (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++)
-
-/*
- * Iterate over all possible CPUs in a leaf RCU node.
- */
-#define for_each_leaf_node_possible_cpu(rnp, cpu) \
-	for ((cpu) = cpumask_next(rnp->grplo - 1, cpu_possible_mask); \
-	     cpu <= rnp->grphi; \
-	     cpu = cpumask_next((cpu), cpu_possible_mask))
-
-/*
  * Union to allow "aggregate OR" operation on the need for a quiescent
  * state by the normal and expedited grace periods.
  */

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-17 23:53 UTC #104 lore raw

PATCH v2 28/39 srcu: Make num_rcu_lvl[] array be external

This commit makes the num_rcu_lvl[] array external so that SRCU can make use of it for initializing its upcoming srcu_node tree.

---
 kernel/rcu/rcu.h  | 1 +
 kernel/rcu/tree.c | 2 +-
 2 files changed, 2 insertions(+), 1 deletion(-)

kernel/rcu/rcu.h +1 -0

index 5c1798ec32f7..4303b880ac99 100644
@@ -195,6 +195,7 @@ extern void resched_cpu(int cpu);
 #include <linux/rcu_node_tree.h>
 
 extern int rcu_num_lvls;
+extern int num_rcu_lvl[];
 extern int rcu_num_nodes;
 static bool rcu_fanout_exact;
 static int rcu_fanout_leaf;

kernel/rcu/tree.c +1 -1

index 21140192fd76..bef5765c3f68 100644
@@ -123,7 +123,7 @@ static int rcu_fanout_leaf = RCU_FANOUT_LEAF;
 module_param(rcu_fanout_leaf, int, 0444);
 int rcu_num_lvls __read_mostly = RCU_NUM_LVLS;
 /* Number of rcu_nodes at specified level. */
-static int num_rcu_lvl[] = NUM_RCU_LVL_INIT;
+int num_rcu_lvl[] = NUM_RCU_LVL_INIT;
 int rcu_num_nodes __read_mostly = NUM_RCU_NODES; /* Total # rcu_nodes in use. */
 /* panic() on RCU Stall sysctl. */
 int sysctl_panic_on_rcu_stall __read_mostly;

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-17 23:48 UTC #105 lore raw

PATCH v2 29/39 srcu: Fix bogus try_check_zero() comment

---
 kernel/rcu/srcu.c | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)

kernel/rcu/srcu.c +3 -4

index d51ab050f777..5aeeaecfb673 100644
@@ -254,10 +254,9 @@ static void srcu_gp_start(struct srcu_struct *sp)
 }
 
 /*
- * @@@ Wait until all pre-existing readers complete.  Such readers
- * will have used the index specified by "idx".
- * the caller should ensures the ->completed is not changed while checking
- * and idx = (->completed & 1) ^ 1
+ * Wait until all readers counted by array index idx complete, but loop
+ * a maximum of trycount times.  The caller must ensure that ->completed
+ * is not changed while checking.
  */
 static bool try_check_zero(struct srcu_struct *sp, int idx, int trycount)
 {

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-17 23:52 UTC #106 lore raw

PATCH v2 30/39 srcu: Improve rcu_seq grace-period-counter abstraction

---
 kernel/rcu/rcu.h      | 29 +++++++++++++++++++++++++----
 kernel/rcu/tree_exp.h |  9 ++++-----
 2 files changed, 29 insertions(+), 9 deletions(-)

kernel/rcu/rcu.h +25 -4

index 4303b880ac99..c62df93bfc1b 100644
@@ -61,20 +61,41 @@
  * Grace-period counter management.
  */
 
+#define RCU_SEQ_CTR_SHIFT	1
+#define RCU_SEQ_STATE_MASK	((1 << RCU_SEQ_CTR_SHIFT) - 1)
+
+/*
+ * Return the counter portion of a sequence number previously returned
+ * by rcu_seq_snap() or rcu_seq_current().
+ */
+static inline unsigned long rcu_seq_ctr(unsigned long s)
+{
+	return s >> RCU_SEQ_CTR_SHIFT;
+}
+
+/*
+ * Return the state portion of a sequence number previously returned
+ * by rcu_seq_snap() or rcu_seq_current().
+ */
+static inline int rcu_seq_state(unsigned long s)
+{
+	return s & RCU_SEQ_STATE_MASK;
+}
+
 /* Adjust sequence number for start of update-side operation. */
 static inline void rcu_seq_start(unsigned long *sp)
 {
 	WRITE_ONCE(*sp, *sp + 1);
 	smp_mb(); /* Ensure update-side operation after counter increment. */
-	WARN_ON_ONCE(!(*sp & 0x1));
+	WARN_ON_ONCE(rcu_seq_state(*sp) != 1);
 }
 
 /* Adjust sequence number for end of update-side operation. */
 static inline void rcu_seq_end(unsigned long *sp)
 {
 	smp_mb(); /* Ensure update-side operation before counter increment. */
-	WARN_ON_ONCE(!(*sp & 0x1));
-	WRITE_ONCE(*sp, *sp + 1);
+	WARN_ON_ONCE(!rcu_seq_state(*sp));
+	WRITE_ONCE(*sp, (*sp | RCU_SEQ_STATE_MASK) + 1);
 }
 
 /* Take a snapshot of the update side's sequence number. */

@@ -82,7 +103,7 @@ static inline unsigned long rcu_seq_snap(unsigned long *sp)
 {
 	unsigned long s;
 
-	s = (READ_ONCE(*sp) + 3) & ~0x1;
+	s = (READ_ONCE(*sp) + 2 * RCU_SEQ_STATE_MASK + 1) & ~RCU_SEQ_STATE_MASK;
 	smp_mb(); /* Above access must not bleed into critical section. */
 	return s;
 }

kernel/rcu/tree_exp.h +4 -5

index 027e123d93c7..e513b4ab1197 100644
@@ -292,7 +292,7 @@ static bool exp_funnel_lock(struct rcu_state *rsp, unsigned long s)
 			trace_rcu_exp_funnel_lock(rsp->name, rnp->level,
 						  rnp->grplo, rnp->grphi,
 						  TPS("wait"));
-			wait_event(rnp->exp_wq[(s >> 1) & 0x3],
+			wait_event(rnp->exp_wq[rcu_seq_ctr(s) & 0x3],
 				   sync_exp_work_done(rsp,
 						      &rdp->exp_workdone2, s));
 			return true;

@@ -534,7 +534,7 @@ static void rcu_exp_wait_wake(struct rcu_state *rsp, unsigned long s)
 			spin_unlock(&rnp->exp_lock);
 		}
 		smp_mb(); /* All above changes before wakeup. */
-		wake_up_all(&rnp->exp_wq[(rsp->expedited_sequence >> 1) & 0x3]);
+		wake_up_all(&rnp->exp_wq[rcu_seq_ctr(rsp->expedited_sequence) & 0x3]);
 	}
 	trace_rcu_exp_grace_period(rsp->name, s, TPS("endwake"));
 	mutex_unlock(&rsp->exp_wake_mutex);

@@ -612,9 +612,8 @@ static void _synchronize_rcu_expedited(struct rcu_state *rsp,
 	/* Wait for expedited grace period to complete. */
 	rdp = per_cpu_ptr(rsp->rda, raw_smp_processor_id());
 	rnp = rcu_get_root(rsp);
-	wait_event(rnp->exp_wq[(s >> 1) & 0x3],
-		   sync_exp_work_done(rsp,
-				      &rdp->exp_workdone0, s));
+	wait_event(rnp->exp_wq[rcu_seq_ctr(s) & 0x3],
+		   sync_exp_work_done(rsp, &rdp->exp_workdone0, s));
 	smp_mb(); /* Workqueue actions happen before return. */
 
 	/* Let the next expedited grace period start. */

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-17 23:45 UTC #107 lore raw

PATCH v2 31/39 srcu: Allow a second bit in rcu_seq for SRCU state

This commit increases the number of reserved bits at the bottom of an rcu_seq grace-period counter from one to two, as will be needed to accommodate SRCU's three-state grace periods.

---
 kernel/rcu/rcu.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

kernel/rcu/rcu.h +1 -1

index c62df93bfc1b..87a0ac95b551 100644
@@ -61,7 +61,7 @@
  * Grace-period counter management.
  */
 
-#define RCU_SEQ_CTR_SHIFT	1
+#define RCU_SEQ_CTR_SHIFT	2
 #define RCU_SEQ_STATE_MASK	((1 << RCU_SEQ_CTR_SHIFT) - 1)
 
 /*

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-17 23:51 UTC #108 lore raw

PATCH v2 32/39 srcu: Merge ->srcu_state into ->srcu_gp_seq

---
 include/linux/srcu.h |  5 +----
 kernel/rcu/rcu.h     | 10 ++++++++++
 kernel/rcu/srcu.c    | 55 +++++++++++++++++++++++++++++++++-------------------
 3 files changed, 46 insertions(+), 24 deletions(-)

include/linux/srcu.h +1 -4

index ad154a7bc114..e7dbc01b61a1 100644
@@ -43,8 +43,7 @@ struct srcu_struct {
 	unsigned long completed;
 	unsigned long srcu_gp_seq;
 	struct srcu_array __percpu *per_cpu_ref;
-	spinlock_t queue_lock; /* protect ->srcu_cblist, ->srcu_state */
-	int srcu_state;
+	spinlock_t queue_lock; /* protect ->srcu_cblist */
 	struct rcu_segcblist srcu_cblist;
 	struct delayed_work work;
 #ifdef CONFIG_DEBUG_LOCK_ALLOC

@@ -56,7 +55,6 @@ struct srcu_struct {
 #define SRCU_STATE_IDLE		0
 #define SRCU_STATE_SCAN1	1
 #define SRCU_STATE_SCAN2	2
-#define SRCU_STATE_DONE		3
 
 #ifdef CONFIG_DEBUG_LOCK_ALLOC

@@ -85,7 +83,6 @@ void process_srcu(struct work_struct *work);
 		.completed = -300,					\
 		.per_cpu_ref = &name##_srcu_array,			\
 		.queue_lock = __SPIN_LOCK_UNLOCKED(name.queue_lock),	\
-		.srcu_state = SRCU_STATE_IDLE,				\
 		.srcu_cblist = RCU_SEGCBLIST_INITIALIZER(name.srcu_cblist),\
 		.work = __DELAYED_WORK_INITIALIZER(name.work, process_srcu, 0),\
 		__SRCU_DEP_MAP_INIT(name)				\

kernel/rcu/rcu.h +10 -0

index 87a0ac95b551..73e16ec4054b 100644
@@ -82,6 +82,16 @@ static inline int rcu_seq_state(unsigned long s)
 	return s & RCU_SEQ_STATE_MASK;
 }
 
+/*
+ * Set the state portion of the pointed-to sequence number.
+ * The caller is responsible for preventing conflicting updates.
+ */
+static inline void rcu_seq_set_state(unsigned long *sp, int newstate)
+{
+	WARN_ON_ONCE(newstate & ~RCU_SEQ_STATE_MASK);
+	WRITE_ONCE(*sp, (*sp & ~RCU_SEQ_STATE_MASK) + newstate);
+}
+
 /* Adjust sequence number for start of update-side operation. */
 static inline void rcu_seq_start(unsigned long *sp)
 {

kernel/rcu/srcu.c +35 ~~-20~~

138 lines not on this page · raw message

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-17 23:46 UTC #109 lore raw

PATCH v2 33/39 srcu: Crude control of expedited grace periods

Improved control of expedited grace periods is planned, but one step at a time.

---
 include/linux/srcu.h |  1 +
 kernel/rcu/srcu.c    | 84 ++++++++++++++++++++++++++++------------------------
 2 files changed, 47 insertions(+), 38 deletions(-)

include/linux/srcu.h +1 -0

index e7dbc01b61a1..73a1b6296224 100644
@@ -42,6 +42,7 @@ struct srcu_array {
 struct srcu_struct {
 	unsigned long completed;
 	unsigned long srcu_gp_seq;
+	atomic_t srcu_exp_cnt;
 	struct srcu_array __percpu *per_cpu_ref;
 	spinlock_t queue_lock; /* protect ->srcu_cblist */
 	struct rcu_segcblist srcu_cblist;

kernel/rcu/srcu.c +46 ~~-38~~

168 lines not on this page · raw message

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-17 23:45 UTC #110 lore raw

PATCH v2 34/39 mm: Use static initialization for "srcu"

The MM-notifier code currently dynamically initializes the srcu_struct named "srcu" at subsys_initcall() time, and includes a BUG_ON() to check this initialization in do_mmu_notifier_register(). Unfortunately, there is no foolproof way to verify that an srcu_struct has been initialized, given the possibility of an srcu_struct being allocated on the stack or on the heap. This means that creating an srcu_struct_is_initialized() function is not a reasonable course of action. Nor is peppering do_mmu_notifier_register() with SRCU-specific #ifdefs an attractive alternative.

This commit therefore uses DEFINE_STATIC_SRCU() to initialize this srcu_struct at compile time, thus eliminating both the subsys_initcall()-time initialization and the runtime BUG_ON().

Cc Andrew Morton <[email protected]>

Cc Ingo Molnar <[email protected]>

Cc Michal Hocko <[email protected]>

Cc "Peter Zijlstra (Intel)" <[email protected]>

Cc Vegard Nossum <[email protected]>

---
 mm/mmu_notifier.c | 14 +-------------
 1 file changed, 1 insertion(+), 13 deletions(-)

mm/mmu_notifier.c +1 ~~-13~~

index a7652acd2ab9..54ca54562928 100644
@@ -21,7 +21,7 @@
 #include <linux/slab.h>
 
 /* global SRCU for all MMs */
-static struct srcu_struct srcu;
+DEFINE_STATIC_SRCU(srcu);
 
 /*
  * This function allows mmu_notifier::release callback to delay a call to

@@ -252,12 +252,6 @@ static int do_mmu_notifier_register(struct mmu_notifier *mn,
 
 	BUG_ON(atomic_read(&mm->mm_users) <= 0);
 
-	/*
-	 * Verify that mmu_notifier_init() already run and the global srcu is
-	 * initialized.
-	 */
-	BUG_ON(!srcu.per_cpu_ref);
-
 	ret = -ENOMEM;
 	mmu_notifier_mm = kmalloc(sizeof(struct mmu_notifier_mm), GFP_KERNEL);
 	if (unlikely(!mmu_notifier_mm))

@@ -406,9 +400,3 @@ void mmu_notifier_unregister_no_release(struct mmu_notifier *mn,
 	mmdrop(mm);
 }
 EXPORT_SYMBOL_GPL(mmu_notifier_unregister_no_release);
-
-static int __init mmu_notifier_init(void)
-{
-	return init_srcu_struct(&srcu);
-}
-subsys_initcall(mmu_notifier_init);

Signature · 7 lines

2.5.2

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to [email protected].  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

Paul E. McKenney

author

2017-04-17 23:47 UTC #111 lore raw

PATCH v2 35/39 srcu: Create a tiny SRCU

In response to automated complaints about modifications to SRCU increasing its size, this commit creates a tiny SRCU that is used in SMP=n && PREEMPT=n builds.

---
 include/linux/srcu.h     |  69 ++-------------
 include/linux/srcutiny.h |  81 ++++++++++++++++++
 include/linux/srcutree.h |  91 ++++++++++++++++++++
 init/Kconfig             |  12 +++
 kernel/rcu/Makefile      |   3 +-
 kernel/rcu/rcutorture.c  |   2 +
 kernel/rcu/srcutiny.c    | 215 +++++++++++++++++++++++++++++++++++++++++++++++
 7 files changed, 411 insertions(+), 62 deletions(-)
 create mode 100644 include/linux/srcutiny.h
 create mode 100644 include/linux/srcutree.h
 create mode 100644 kernel/rcu/srcutiny.c

include/linux/srcu.h +8 ~~-61~~

index 73a1b6296224..907f09b14eda 100644
@@ -34,28 +34,7 @@
 #include <linux/workqueue.h>
 #include <linux/rcu_segcblist.h>
 
-struct srcu_array {
-	unsigned long lock_count[2];
-	unsigned long unlock_count[2];
-};
-
-struct srcu_struct {
-	unsigned long completed;
-	unsigned long srcu_gp_seq;
-	atomic_t srcu_exp_cnt;
-	struct srcu_array __percpu *per_cpu_ref;
-	spinlock_t queue_lock; /* protect ->srcu_cblist */
-	struct rcu_segcblist srcu_cblist;
-	struct delayed_work work;
-#ifdef CONFIG_DEBUG_LOCK_ALLOC
-	struct lockdep_map dep_map;
-#endif /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
-};
-
-/* Values for -> state variable. */
-#define SRCU_STATE_IDLE		0
-#define SRCU_STATE_SCAN1	1
-#define SRCU_STATE_SCAN2	2
+struct srcu_struct;
 
 #ifdef CONFIG_DEBUG_LOCK_ALLOC

@@ -77,42 +56,13 @@ int init_srcu_struct(struct srcu_struct *sp);
 #define __SRCU_DEP_MAP_INIT(srcu_name)
 #endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */
 
-void process_srcu(struct work_struct *work);
-
-#define __SRCU_STRUCT_INIT(name)					\
-	{								\
-		.completed = -300,					\
-		.per_cpu_ref = &name##_srcu_array,			\
-		.queue_lock = __SPIN_LOCK_UNLOCKED(name.queue_lock),	\
-		.srcu_cblist = RCU_SEGCBLIST_INITIALIZER(name.srcu_cblist),\
-		.work = __DELAYED_WORK_INITIALIZER(name.work, process_srcu, 0),\
-		__SRCU_DEP_MAP_INIT(name)				\
-	}
-
-/*
- * Define and initialize a srcu struct at build time.
- * Do -not- call init_srcu_struct() nor cleanup_srcu_struct() on it.
- *
- * Note that although DEFINE_STATIC_SRCU() hides the name from other
- * files, the per-CPU variable rules nevertheless require that the
- * chosen name be globally unique.  These rules also prohibit use of
- * DEFINE_STATIC_SRCU() within a function.  If these rules are too
- * restrictive, declare the srcu_struct manually.  For example, in
- * each file:
- *
- *	static struct srcu_struct my_srcu;
- *
- * Then, before the first use of each my_srcu, manually initialize it:
- *
- *	init_srcu_struct(&my_srcu);
- *
- * See include/linux/percpu-defs.h for the rules on per-CPU variables.
- */
-#define __DEFINE_SRCU(name, is_static)					\
-	static DEFINE_PER_CPU(struct srcu_array, name##_srcu_array);\
-	is_static struct srcu_struct name = __SRCU_STRUCT_INIT(name)
-#define DEFINE_SRCU(name)		__DEFINE_SRCU(name, /* not static */)
-#define DEFINE_STATIC_SRCU(name)	__DEFINE_SRCU(name, static)
+#ifdef CONFIG_TINY_SRCU
+#include <linux/srcutiny.h>
+#elif defined(CONFIG_TREE_SRCU)
+#include <linux/srcutree.h>
+#else
+#error "Unknown SRCU implementation specified to kernel configuration"
+#endif
 
 /**
  * call_srcu() - Queue a callback for invocation after an SRCU grace period

@@ -138,9 +88,6 @@ void cleanup_srcu_struct(struct srcu_struct *sp);
 int __srcu_read_lock(struct srcu_struct *sp) __acquires(sp);
 void __srcu_read_unlock(struct srcu_struct *sp, int idx) __releases(sp);
 void synchronize_srcu(struct srcu_struct *sp);
-void synchronize_srcu_expedited(struct srcu_struct *sp);
-unsigned long srcu_batches_completed(struct srcu_struct *sp);
-void srcu_barrier(struct srcu_struct *sp);
 
 #ifdef CONFIG_DEBUG_LOCK_ALLOC

include/linux/srcutiny.h new file +81 -0

81 lines not on this page · raw message

include/linux/srcutree.h new file +91 -0

91 lines not on this page · raw message

init/Kconfig +12 -0

index a92f27da4a27..d269f2ca17b8 100644
@@ -526,6 +526,18 @@ config SRCU
 	  permits arbitrary sleeping or blocking within RCU read-side critical
 	  sections.
 
+config TINY_SRCU
+	bool
+	default y if TINY_RCU
+	help
+	  This option selects the single-CPU non-preemptible version of SRCU.
+
+config TREE_SRCU
+	bool
+	default y if !TINY_RCU
+	help
+	  This option selects the full-fledged version of SRCU.
+
 config TASKS_RCU
 	bool
 	default n

kernel/rcu/Makefile +2 -1

index 18dfc485225c..b853214a2b99 100644
@@ -3,7 +3,8 @@
 KCOV_INSTRUMENT := n
 
 obj-y += update.o sync.o
-obj-$(CONFIG_SRCU) += srcu.o
+obj-$(CONFIG_TREE_SRCU) += srcu.o
+obj-$(CONFIG_TINY_SRCU) += srcutiny.o
 obj-$(CONFIG_RCU_TORTURE_TEST) += rcutorture.o
 obj-$(CONFIG_RCU_PERF_TEST) += rcuperf.o
 obj-$(CONFIG_TREE_RCU) += tree.o

kernel/rcu/rcutorture.c +2 -0

index cccc417a8135..98591e16db1a 100644
@@ -559,6 +559,7 @@ static void srcu_torture_barrier(void)
 
 static void srcu_torture_stats(void)
 {
+#ifdef CONFIG_TREE_SRCU
 	int cpu;
 	int idx = srcu_ctlp->completed & 0x1;

@@ -587,6 +588,7 @@ static void srcu_torture_stats(void)
 		pr_cont(" %d(%ld,%ld)", cpu, c0, c1);
 	}
 	pr_cont("\n");
+#endif
 }
 
 static void srcu_torture_synchronize_expedited(void)

kernel/rcu/srcutiny.c new file +215 -0

215 lines not on this page · raw message

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-17 23:47 UTC #112 lore raw

PATCH v2 36/39 srcutorture: Print Tiny SRCU reader statistics

The srcu_torture_stats() function is adapted to the specific srcu_struct layout traditionally used by SRCU. This commit therefore adds support for Tiny SRCU.

---
 kernel/rcu/rcutorture.c | 15 +++++++++++----
 1 file changed, 11 insertions(+), 4 deletions(-)

kernel/rcu/rcutorture.c +11 -4

index 98591e16db1a..9cbb8a7b909d 100644
@@ -559,11 +559,12 @@ static void srcu_torture_barrier(void)
 
 static void srcu_torture_stats(void)
 {
-#ifdef CONFIG_TREE_SRCU
-	int cpu;
-	int idx = srcu_ctlp->completed & 0x1;
+	int __maybe_unused cpu;
+	int idx;
 
-	pr_alert("%s%s per-CPU(idx=%d):",
+#ifdef CONFIG_TREE_SRCU
+	idx = srcu_ctlp->completed & 0x1;
+	pr_alert("%s%s Tree SRCU per-CPU(idx=%d):",
 		 torture_type, TORTURE_FLAG, idx);
 	for_each_possible_cpu(cpu) {
 		unsigned long l0, l1;

@@ -588,6 +589,12 @@ static void srcu_torture_stats(void)
 		pr_cont(" %d(%ld,%ld)", cpu, c0, c1);
 	}
 	pr_cont("\n");
+#elif defined(CONFIG_TINY_SRCU)
+	idx = READ_ONCE(srcu_ctlp->srcu_idx) & 0x1;
+	pr_alert("%s%s Tiny SRCU per-CPU(idx=%d): (%d,%d)\n",
+		 torture_type, TORTURE_FLAG, idx,
+		 READ_ONCE(srcu_ctlp->srcu_lock_nesting[!idx]),
+		 READ_ONCE(srcu_ctlp->srcu_lock_nesting[idx]));
 #endif
 }

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-17 23:47 UTC #113 lore raw

PATCH v2 37/39 srcu: Introduce CLASSIC_SRCU Kconfig option

---
 include/linux/srcu.h        |   2 +
 include/linux/srcuclassic.h | 101 ++++++++
 init/Kconfig                |  21 +-
 kernel/rcu/Makefile         |   3 +-
 kernel/rcu/rcutorture.c     |   2 +-
 kernel/rcu/srcu.c           | 347 ++++++++++++++-----------
 kernel/rcu/srcutree.c       | 613 ++++++++++++++++++++++++++++++++++++++++++++
 7 files changed, 934 insertions(+), 155 deletions(-)
 create mode 100644 include/linux/srcuclassic.h
 create mode 100644 kernel/rcu/srcutree.c

include/linux/srcu.h +2 -0

index 907f09b14eda..167ad8831aaf 100644
@@ -60,6 +60,8 @@ int init_srcu_struct(struct srcu_struct *sp);
 #include <linux/srcutiny.h>
 #elif defined(CONFIG_TREE_SRCU)
 #include <linux/srcutree.h>
+#elif defined(CONFIG_CLASSIC_SRCU)
+#include <linux/srcuclassic.h>
 #else
 #error "Unknown SRCU implementation specified to kernel configuration"
 #endif

include/linux/srcuclassic.h new file +101 -0

101 lines not on this page · raw message

init/Kconfig +19 -2

index d269f2ca17b8..558cc3638ab9 100644
@@ -526,15 +526,32 @@ config SRCU
 	  permits arbitrary sleeping or blocking within RCU read-side critical
 	  sections.
 
+config CLASSIC_SRCU
+	bool "Use v4.11 classic SRCU implementation"
+	default n
+	depends on RCU_EXPERT && SRCU
+	help
+	  This option selects the traditional well-tested classic SRCU
+	  implementation from v4.11, as might be desired for enterprise
+	  Linux distributions.  Without this option, the shiny new
+	  Tiny SRCU and Tree SRCU implementations are used instead.
+	  At some point, it is hoped that Tiny SRCU and Tree SRCU
+	  will accumulate enough test time and confidence to allow
+	  Classic SRCU to be dropped entirely.
+
+	  Say Y if you need a rock-solid SRCU.
+
+	  Say N if you would like help test Tree SRCU.
+
 config TINY_SRCU
 	bool
-	default y if TINY_RCU
+	default y if TINY_RCU && !CLASSIC_SRCU
 	help
 	  This option selects the single-CPU non-preemptible version of SRCU.
 
 config TREE_SRCU
 	bool
-	default y if !TINY_RCU
+	default y if !TINY_RCU && !CLASSIC_SRCU
 	help
 	  This option selects the full-fledged version of SRCU.

kernel/rcu/Makefile +2 -1

index b853214a2b99..158e6593d58c 100644
@@ -3,7 +3,8 @@
 KCOV_INSTRUMENT := n
 
 obj-y += update.o sync.o
-obj-$(CONFIG_TREE_SRCU) += srcu.o
+obj-$(CONFIG_CLASSIC_SRCU) += srcu.o
+obj-$(CONFIG_TREE_SRCU) += srcutree.o
 obj-$(CONFIG_TINY_SRCU) += srcutiny.o
 obj-$(CONFIG_RCU_TORTURE_TEST) += rcutorture.o
 obj-$(CONFIG_RCU_PERF_TEST) += rcuperf.o

kernel/rcu/rcutorture.c +1 -1

index 9cbb8a7b909d..6f344b6748a8 100644
@@ -562,7 +562,7 @@ static void srcu_torture_stats(void)
 	int __maybe_unused cpu;
 	int idx;
 
-#ifdef CONFIG_TREE_SRCU
+#if defined(CONFIG_TREE_SRCU) || defined(CONFIG_CLASSIC_SRCU)
 	idx = srcu_ctlp->completed & 0x1;
 	pr_alert("%s%s Tree SRCU per-CPU(idx=%d):",
 		 torture_type, TORTURE_FLAG, idx);

kernel/rcu/srcu.c +196 ~~-151~~

497 lines not on this page · raw message

kernel/rcu/srcutree.c new file +613 -0

613 lines not on this page · raw message

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-17 23:46 UTC #114 lore raw

PATCH v2 38/39 srcu: Parallelize callback handling

Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention.

Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree).

The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked.

This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU.

Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability:

			Callback Queuing Overhead
			-------------------------
	# CPUS		Classic SRCU	Tree SRCU
	------          ------------    ---------
	     2              0.349 us     0.342 us
	    16             31.66  us     0.4   us
	    41             ---------     0.417 us

The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages.

[1] https://lwn.net/Articles/309030/
[2] https://patchwork.kernel.org/patch/5108281/

---
 include/linux/rcu_segcblist.h |  42 ++-
 include/linux/srcutree.h      |  80 ++++--
 kernel/rcu/rcutorture.c       |  20 +-
 kernel/rcu/srcutree.c         | 639 +++++++++++++++++++++++++++++++++---------
 kernel/rcu/tree.c             |   6 +
 kernel/rcu/tree.h             |   8 +
 6 files changed, 645 insertions(+), 150 deletions(-)

include/linux/rcu_segcblist.h +38 -4

index 74b1e7243955..ced8f313fd05 100644
@@ -402,6 +402,37 @@ static inline void rcu_segcblist_enqueue(struct rcu_segcblist *rsclp,
 }
 
 /*
+ * Entrain the specified callback onto the specified rcu_segcblist at
+ * the end of the last non-empty segment.  If the entire rcu_segcblist
+ * is empty, make no change, but return false.
+ *
+ * This is intended for use by rcu_barrier()-like primitives, -not-
+ * for normal grace-period use.  IMPORTANT:  The callback you enqueue
+ * will wait for all prior callbacks, NOT necessarily for a grace
+ * period.  You have been warned.
+ */
+static inline bool rcu_segcblist_entrain(struct rcu_segcblist *rsclp,
+					 struct rcu_head *rhp, bool lazy)
+{
+	int i;
+
+	if (rcu_segcblist_n_cbs(rsclp) == 0)
+		return false;
+	WRITE_ONCE(rsclp->len, rsclp->len + 1);
+	if (lazy)
+		rsclp->len_lazy++;
+	smp_mb(); /* Ensure counts are updated before callback is entrained. */
+	rhp->next = NULL;
+	for (i = RCU_NEXT_TAIL; i > RCU_DONE_TAIL; i--)
+		if (rsclp->tails[i] != rsclp->tails[i - 1])
+			break;
+	*rsclp->tails[i] = rhp;
+	for (; i <= RCU_NEXT_TAIL; i++)
+		rsclp->tails[i] = &rhp->next;
+	return true;
+}
+
+/*
  * Extract only the counts from the specified rcu_segcblist structure,
  * and place them in the specified rcu_cblist structure.  This function
  * supports both callback orphaning and invocation, hence the separation

@@ -537,7 +568,8 @@ static inline void rcu_segcblist_advance(struct rcu_segcblist *rsclp,
 	int i, j;
 
 	WARN_ON_ONCE(!rcu_segcblist_is_enabled(rsclp));
-	WARN_ON_ONCE(rcu_segcblist_restempty(rsclp, RCU_DONE_TAIL));
+	if (rcu_segcblist_restempty(rsclp, RCU_DONE_TAIL))
+		return;
 
 	/*
 	 * Find all callbacks whose ->gp_seq numbers indicate that they

@@ -582,8 +614,9 @@ static inline void rcu_segcblist_advance(struct rcu_segcblist *rsclp,
  * them to complete at the end of the earlier grace period.
  *
  * This function operates on an rcu_segcblist structure, and also the
- * grace-period sequence number at which new callbacks would become
- * ready to invoke.
+ * grace-period sequence number seq at which new callbacks would become
+ * ready to invoke.  Returns true if there are callbacks that won't be
+ * ready to invoke until seq, false otherwise.
  */
 static inline bool rcu_segcblist_accelerate(struct rcu_segcblist *rsclp,
 					    unsigned long seq)

@@ -591,7 +624,8 @@ static inline bool rcu_segcblist_accelerate(struct rcu_segcblist *rsclp,
 	int i;
 
 	WARN_ON_ONCE(!rcu_segcblist_is_enabled(rsclp));
-	WARN_ON_ONCE(rcu_segcblist_restempty(rsclp, RCU_DONE_TAIL));
+	if (rcu_segcblist_restempty(rsclp, RCU_DONE_TAIL))
+		return false;
 
 	/*
 	 * Find the segment preceding the oldest segment of callbacks

include/linux/srcutree.h +64 ~~-16~~

107 lines not on this page · raw message

kernel/rcu/rcutorture.c +19 -1

index 6f344b6748a8..e9d4527cdd43 100644
@@ -563,17 +563,30 @@ static void srcu_torture_stats(void)
 	int idx;
 
 #if defined(CONFIG_TREE_SRCU) || defined(CONFIG_CLASSIC_SRCU)
+#ifdef CONFIG_TREE_SRCU
+	idx = srcu_ctlp->srcu_idx & 0x1;
+#else /* #ifdef CONFIG_TREE_SRCU */
 	idx = srcu_ctlp->completed & 0x1;
+#endif /* #else #ifdef CONFIG_TREE_SRCU */
 	pr_alert("%s%s Tree SRCU per-CPU(idx=%d):",
 		 torture_type, TORTURE_FLAG, idx);
 	for_each_possible_cpu(cpu) {
 		unsigned long l0, l1;
 		unsigned long u0, u1;
 		long c0, c1;
-		struct srcu_array *counts = per_cpu_ptr(srcu_ctlp->per_cpu_ref, cpu);
+#ifdef CONFIG_TREE_SRCU
+		struct srcu_data *counts;
 
+		counts = per_cpu_ptr(srcu_ctlp->sda, cpu);
+		u0 = counts->srcu_unlock_count[!idx];
+		u1 = counts->srcu_unlock_count[idx];
+#else /* #ifdef CONFIG_TREE_SRCU */
+		struct srcu_array *counts;
+
+		counts = per_cpu_ptr(srcu_ctlp->per_cpu_ref, cpu);
 		u0 = counts->unlock_count[!idx];
 		u1 = counts->unlock_count[idx];
+#endif /* #else #ifdef CONFIG_TREE_SRCU */
 
 		/*
 		 * Make sure that a lock is always counted if the corresponding

@@ -581,8 +594,13 @@ static void srcu_torture_stats(void)
 		 */
 		smp_rmb();
 
+#ifdef CONFIG_TREE_SRCU
+		l0 = counts->srcu_lock_count[!idx];
+		l1 = counts->srcu_lock_count[idx];
+#else /* #ifdef CONFIG_TREE_SRCU */
 		l0 = counts->lock_count[!idx];
 		l1 = counts->lock_count[idx];
+#endif /* #else #ifdef CONFIG_TREE_SRCU */
 
 		c0 = l0 - u0;
 		c1 = l1 - u1;

kernel/rcu/srcutree.c +510 ~~-129~~

877 lines not on this page · raw message

kernel/rcu/tree.c +6 -0

index bef5765c3f68..8bf6dfab8055 100644
@@ -3776,12 +3776,16 @@ int rcutree_online_cpu(unsigned int cpu)
 {
 	sync_sched_exp_online_cleanup(cpu);
 	rcutree_affinity_setting(cpu, -1);
+	if (IS_ENABLED(CONFIG_TREE_SRCU))
+		srcu_online_cpu(cpu);
 	return 0;
 }
 
 int rcutree_offline_cpu(unsigned int cpu)
 {
 	rcutree_affinity_setting(cpu, cpu);
+	if (IS_ENABLED(CONFIG_TREE_SRCU))
+		srcu_offline_cpu(cpu);
 	return 0;
 }

@@ -4157,6 +4161,8 @@ void __init rcu_init(void)
 	for_each_online_cpu(cpu) {
 		rcutree_prepare_cpu(cpu);
 		rcu_cpu_starting(cpu);
+		if (IS_ENABLED(CONFIG_TREE_SRCU))
+			srcu_online_cpu(cpu);
 	}
 }

kernel/rcu/tree.h +8 -0

index a2a45cb629d6..0e598ab08fea 100644
@@ -541,6 +541,14 @@ static bool rcu_nohz_full_cpu(struct rcu_state *rsp);
 static void rcu_dynticks_task_enter(void);
 static void rcu_dynticks_task_exit(void);
 
+#ifdef CONFIG_SRCU
+void srcu_online_cpu(unsigned int cpu);
+void srcu_offline_cpu(unsigned int cpu);
+#else /* #ifdef CONFIG_SRCU */
+void srcu_online_cpu(unsigned int cpu) { }
+void srcu_offline_cpu(unsigned int cpu) { }
+#endif /* #else #ifdef CONFIG_SRCU */
+
 #endif /* #ifndef RCU_TREE_NONCORE */
 
 #ifdef CONFIG_RCU_TRACE

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-17 23:47 UTC #115 lore raw

PATCH v2 39/39 rcu: Make non-preemptive schedule be Tasks RCU quiescent state

Currently, a call to schedule() acts as a Tasks RCU quiescent state only if a context switch actually takes place. However, just the call to schedule() guarantees that the calling task has moved off of whatever tracing trampoline that it might have been one previously. This commit therefore plumbs schedule()'s "preempt" parameter into rcu_note_context_switch(), which then records the Tasks RCU quiescent state, but only if this call to schedule() was -not- due to a preemption.

To avoid adding overhead to the common-case context-switch path, this commit hides the rcu_note_context_switch() check under an existing non-common-case check.

---
 include/linux/rcupdate.h | 11 ++++++++---
 include/linux/rcutiny.h  | 13 +++++++++----
 include/linux/rcutree.h  |  5 +++--
 kernel/rcu/tree.c        | 22 +++++++++++++++++++++-
 kernel/rcu/update.c      |  1 +
 kernel/sched/core.c      |  2 +-
 6 files changed, 43 insertions(+), 11 deletions(-)

include/linux/rcupdate.h +8 -3

index e6146d0074f8..f531b29207da 100644
@@ -363,15 +363,20 @@ static inline void rcu_init_nohz(void)
 #ifdef CONFIG_TASKS_RCU
 #define TASKS_RCU(x) x
 extern struct srcu_struct tasks_rcu_exit_srcu;
-#define rcu_note_voluntary_context_switch(t) \
+#define rcu_note_voluntary_context_switch_lite(t) \
 	do { \
-		rcu_all_qs(); \
 		if (READ_ONCE((t)->rcu_tasks_holdout)) \
 			WRITE_ONCE((t)->rcu_tasks_holdout, false); \
 	} while (0)
+#define rcu_note_voluntary_context_switch(t) \
+	do { \
+		rcu_all_qs(); \
+		rcu_note_voluntary_context_switch_lite(t); \
+	} while (0)
 #else /* #ifdef CONFIG_TASKS_RCU */
 #define TASKS_RCU(x) do { } while (0)
-#define rcu_note_voluntary_context_switch(t)	rcu_all_qs()
+#define rcu_note_voluntary_context_switch_lite(t)	do { } while (0)
+#define rcu_note_voluntary_context_switch(t)		rcu_all_qs()
 #endif /* #else #ifdef CONFIG_TASKS_RCU */
 
 /**

include/linux/rcutiny.h +9 -4

index 5219be250f00..74d9c3a1feee 100644
@@ -92,10 +92,11 @@ static inline void kfree_call_rcu(struct rcu_head *head,
 	call_rcu(head, func);
 }
 
-static inline void rcu_note_context_switch(void)
-{
-	rcu_sched_qs();
-}
+#define rcu_note_context_switch(preempt) \
+	do { \
+		rcu_sched_qs(); \
+		rcu_note_voluntary_context_switch_lite(current); \
+	} while (0)
 
 /*
  * Take advantage of the fact that there is only one CPU, which

@@ -242,6 +243,10 @@ static inline bool rcu_is_watching(void)
 
 #endif /* #else defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_RCU_TRACE) */
 
+static inline void rcu_request_urgent_qs_task(struct task_struct *t)
+{
+}
+
 static inline void rcu_all_qs(void)
 {
 	barrier(); /* Avoid RCU read-side critical sections leaking across. */

include/linux/rcutree.h +3 -2

index 63a4e4cf40a5..0bacb6b2af69 100644
@@ -30,7 +30,7 @@
 #ifndef __LINUX_RCUTREE_H
 #define __LINUX_RCUTREE_H
 
-void rcu_note_context_switch(void);
+void rcu_note_context_switch(bool preempt);
 int rcu_needs_cpu(u64 basem, u64 *nextevt);
 void rcu_cpu_stall_reset(void);

@@ -41,7 +41,7 @@ void rcu_cpu_stall_reset(void);
  */
 static inline void rcu_virt_note_context_switch(int cpu)
 {
-	rcu_note_context_switch();
+	rcu_note_context_switch(false);
 }
 
 void synchronize_rcu_bh(void);

@@ -108,6 +108,7 @@ void rcu_scheduler_starting(void);
 extern int rcu_scheduler_active __read_mostly;
 
 bool rcu_is_watching(void);
+void rcu_request_urgent_qs_task(struct task_struct *t);
 
 void rcu_all_qs(void);

kernel/rcu/tree.c +21 -1

index 8bf6dfab8055..21b707ce4dfa 100644
@@ -458,7 +458,7 @@ static void rcu_momentary_dyntick_idle(void)
  * and requires special handling for preemptible RCU.
  * The caller must have disabled interrupts.
  */
-void rcu_note_context_switch(void)
+void rcu_note_context_switch(bool preempt)
 {
 	barrier(); /* Avoid RCU read-side critical sections leaking down. */
 	trace_rcu_utilization(TPS("Start context switch"));

@@ -471,6 +471,8 @@ void rcu_note_context_switch(void)
 	if (unlikely(raw_cpu_read(rcu_dynticks.rcu_need_heavy_qs)))
 		rcu_momentary_dyntick_idle();
 	this_cpu_inc(rcu_dynticks.rcu_qs_ctr);
+	if (!preempt)
+		rcu_note_voluntary_context_switch_lite(current);
 out:
 	trace_rcu_utilization(TPS("End context switch"));
 	barrier(); /* Avoid RCU read-side critical sections leaking up. */

@@ -1149,6 +1151,24 @@ bool notrace rcu_is_watching(void)
 }
 EXPORT_SYMBOL_GPL(rcu_is_watching);
 
+/*
+ * If a holdout task is actually running, request an urgent quiescent
+ * state from its CPU.  This is unsynchronized, so migrations can cause
+ * the request to go to the wrong CPU.  Which is OK, all that will happen
+ * is that the CPU's next context switch will be a bit slower and next
+ * time around this task will generate another request.
+ */
+void rcu_request_urgent_qs_task(struct task_struct *t)
+{
+	int cpu;
+
+	barrier();
+	cpu = task_cpu(t);
+	if (!task_curr(t))
+		return; /* This task is not running on that CPU. */
+	smp_store_release(per_cpu_ptr(&rcu_dynticks.rcu_urgent_qs, cpu), true);
+}
+
 #if defined(CONFIG_PROVE_RCU) && defined(CONFIG_HOTPLUG_CPU)
 
 /*

kernel/rcu/update.c +1 -0

index c5df0d756900..273e869ca21d 100644
@@ -665,6 +665,7 @@ static void check_holdout_task(struct task_struct *t,
 		put_task_struct(t);
 		return;
 	}
+	rcu_request_urgent_qs_task(t);
 	if (!needreport)
 		return;
 	if (*firstreport) {

kernel/sched/core.c +1 -1

index 3b31fc05a0f1..2adf7b6c04e7 100644
@@ -3378,7 +3378,7 @@ static void __sched notrace __schedule(bool preempt)
 		hrtick_clear(rq);
 
 	local_irq_disable();
-	rcu_note_context_switch();
+	rcu_note_context_switch(preempt);
 
 	/*
 	 * Make sure that signal_pending_state()->signal_pending() below

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-19 16:58 UTC #116 lore raw

v3 cover SRCU callback parallelization for 4.12

Hello!

This v3 series moves SRCU from its traditional single per-srcu_struct callback queue to per-srcu_struct/per-CPU callback queues. This involves abstracting functionality from Tree RCU, which results in a large conflict footprint, which in turn results in some otherwise unrelated patches coming along for the ride.

2. Make arch select smp_mb__after_unlock_lock() strength, which
gets rid of an arch-specific #ifdef.

3. Consolidate SRCU batch checking into rcu_all_batches_empty().

4. Check for tardy grace-period activity in cleanup_srcu_struct().

5-7. Semicolon inside RCU_TRACE() for various parts of RCU.

8. Pull rcu_sched_qs_mask into rcu_dynticks structure in order to
eliminate an isolated per-CPU variable.

9. Pull rcu_qs_ctr into rcu_dynticks structure.

10. Eliminate flavor scan in rcu_momentary_dyntick_idle() to
reduce semi-common-case context-switch overhead.

11. Place guard on rcu_all_qs() and rcu_note_context_switch()
actions to reduce common-case scheduler-fastpath overhead.

12. Default RCU_FANOUT_LEAF to 16 unless explicitly changed.

13. Abstract multi-tail callback list handling for SRCU.

14. Allow SRCU to access rcu_scheduler_active.

15. Allow early boot use of synchronize_srcu(), though not yet
mid-boot use.

16. Add single-element dequeue functions to rcu_segcblist for
debug use.

17. Move rcu_seq_start() and friends to rcu.h for SRCU's benefit.

18. Expedited wakeups need to be fully ordered.

19. Fix warning in rcu_seq_end().

20. Push srcu_advance_batches() fastpath into common case as a
step towards callback parallelization.

21. Move to state-based grace-period sequencing, also as a step
towards callback parallelization.

22. Add grace-period sequence numbers to SRCU.

23. Use rcu_segcblist to track SRCU callbacks.

24. Move combining-tree definitions for SRCU's benefit.

25. Move rcu_init_levelspread() to rcu_tree_node.h for SRCU's benefit.

26. Remove redundant levelcnt[] array from rcu_init_one().

27. Move rcu_node traversal macros to rcu.h for SRCU's benefit.

28. Make num_rcu_lvl[] array be external for SRCU's benefit.

29. Fix bogus try_check_zero() comment.

30. Improve rcu_seq grace-period-counter abstraction for SRCU's
benefit.

31. Allow a second bit in rcu_seq for SRCU state.

32. Merge ->srcu_state into ->srcu_gp_seq to allow atomic updates.

33. Provide crude control of expedited SRCU grace periods.

34. Use static initialization for "srcu" in mm/mmu_notifier.c.

35. Create a tiny SRCU for bloatwatch/tinification.

36. Print Tiny SRCU reader statistics in rcutorture.

37. Introduce CLASSIC_SRCU Kconfig option for those who do not
wish to help debug Tree SRCU.

38. Parallelize SRCU callback handling.

39. Make expedited parallel SRCU callback handling really be fully
expedited.

40. Make non-preemptive schedule be Tasks RCU quiescent state.

Updates since v2:

o Apply Josh Triplett feedback.

o Fix a performance regression found by Marc Zyngier.

Updates since v1:

o Incorporate feedback from Peter Zijlstra.

o Added v2 patch #34 ("Use static initialization for "srcu" in
mm/mmu_notifier.c"), moving it from its v1 location in the fixes series.

o Added v2 patch #39 ("Make non-preemptive schedule be Tasks RCU
quiescent state") for the benefit of upcoming ftrace work at Steve Rostedt's request.

Thanx, Paul

------------------------------------------------------------------------

 /kernel/rcu/rcu_segcblist.h                                     |  670 -----
 b/Documentation/RCU/Design/Data-Structures/Data-Structures.html |   36 
 b/arch/Kconfig                                                  |    3 
 b/arch/powerpc/Kconfig                                          |    1 
 b/include/linux/rcu_node_tree.h                                 |  105 
 b/include/linux/rcu_segcblist.h                                 |  720 +++++
 b/include/linux/rcupdate.h                                      |   17 
 b/include/linux/rcutiny.h                                       |   24 
 b/include/linux/rcutree.h                                       |    5 
 b/include/linux/srcu.h                                          |  112 
 b/include/linux/srcuclassic.h                                   |  101 
 b/include/linux/srcutiny.h                                      |   81 
 b/include/linux/srcutree.h                                      |  171 +
 b/init/Kconfig                                                  |   33 
 b/kernel/rcu/Makefile                                           |    6 
 b/kernel/rcu/rcu.h                                              |  165 +
 b/kernel/rcu/rcu_segcblist.h                                    |  670 +++++
 b/kernel/rcu/rcutorture.c                                       |   39 
 b/kernel/rcu/srcu.c                                             |  856 +++---
 b/kernel/rcu/srcutiny.c                                         |  215 +
 b/kernel/rcu/srcutree.c                                         | 1255 ++++++++--
 b/kernel/rcu/tiny.c                                             |   20 
 b/kernel/rcu/tiny_plugin.h                                      |   13 
 b/kernel/rcu/tree.c                                             |  661 ++---
 b/kernel/rcu/tree.h                                             |  174 -
 b/kernel/rcu/tree_exp.h                                         |   25 
 b/kernel/rcu/tree_plugin.h                                      |   62 
 b/kernel/rcu/tree_trace.c                                       |   26 
 b/kernel/rcu/update.c                                           |   53 
 b/kernel/sched/core.c                                           |    2 
 b/mm/mmu_notifier.c                                             |   14 
 31 files changed, 4299 insertions(+), 2036 deletions(-)

Paul E. McKenney

author

2017-04-19 17:02 UTC #117 lore raw

PATCH v3 1/40 rcu: Maintain special bits at bottom of ->dynticks counter

Reported-by Andy Lutomirski <[email protected]>

---
 include/linux/rcutiny.h |  5 ++++
 kernel/rcu/tree.c       | 77 +++++++++++++++++++++++++++++++++++++++----------
 kernel/rcu/tree.h       |  1 +
 3 files changed, 67 insertions(+), 16 deletions(-)

include/linux/rcutiny.h +5 -0

index b452953e21c8..6c9d941e3962 100644
@@ -33,6 +33,11 @@ static inline int rcu_dynticks_snap(struct rcu_dynticks *rdtp)
 	return 0;
 }
 
+static inline bool rcu_eqs_special_set(int cpu)
+{
+	return false;  /* Never flag non-existent other CPUs! */
+}
+
 static inline unsigned long get_state_synchronize_rcu(void)
 {
 	return 0;

kernel/rcu/tree.c +61 ~~-16~~

140 lines not on this page · raw message

kernel/rcu/tree.h +1 -0

index ec62a05bfdb3..7468b4de7e0c 100644
@@ -596,6 +596,7 @@ extern struct rcu_state rcu_preempt_state;
 #endif /* #ifdef CONFIG_PREEMPT_RCU */
 
 int rcu_dynticks_snap(struct rcu_dynticks *rdtp);
+bool rcu_eqs_special_set(int cpu);
 
 #ifdef CONFIG_RCU_BOOST
 DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_status);

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-19 16:58 UTC #118 lore raw

PATCH v3 2/40 rcu: Make arch select smp_mb__after_unlock_lock() strength

The definition of smp_mb__after_unlock_lock() is currently smp_mb() for CONFIG_PPC and a no-op otherwise. It would be better to instead provide an architecture-selectable Kconfig option, and select the strength of smp_mb__after_unlock_lock() based on that option. This commit therefore creates ARCH_WEAK_RELEASE_ACQUIRE, has PPC select it, and bases the definition of smp_mb__after_unlock_lock() on this new ARCH_WEAK_RELEASE_ACQUIRE Kconfig option.

Reported-by Ingo Molnar <[email protected]>

Cc Peter Zijlstra <[email protected]>

Cc Will Deacon <[email protected]>

Cc Boqun Feng <[email protected]>

Cc Benjamin Herrenschmidt <[email protected]>

Cc Paul Mackerras <[email protected]>

Acked-by Michael Ellerman <[email protected]>

---
 arch/Kconfig             | 3 +++
 arch/powerpc/Kconfig     | 1 +
 include/linux/rcupdate.h | 6 +++---
 3 files changed, 7 insertions(+), 3 deletions(-)

arch/Kconfig +3 -0

index cd211a14a88f..adefaf344239 100644
@@ -320,6 +320,9 @@ config HAVE_CMPXCHG_LOCAL
 config HAVE_CMPXCHG_DOUBLE
 	bool
 
+config ARCH_WEAK_RELEASE_ACQUIRE
+	bool
+
 config ARCH_WANT_IPC_PARSE_VERSION
 	bool

arch/powerpc/Kconfig +1 -0

index 97a8bc8a095c..7a5c9b764cd2 100644
@@ -99,6 +99,7 @@ config PPC
 	select ARCH_USE_BUILTIN_BSWAP
 	select ARCH_USE_CMPXCHG_LOCKREF		if PPC64
 	select ARCH_WANT_IPC_PARSE_VERSION
+	select ARCH_WEAK_RELEASE_ACQUIRE
 	select BINFMT_ELF
 	select BUILDTIME_EXTABLE_SORT
 	select CLONE_BACKWARDS

include/linux/rcupdate.h +3 -3

index de88b33c0974..e6146d0074f8 100644
@@ -1127,11 +1127,11 @@ do { \
  * if the UNLOCK and LOCK are executed by the same CPU or if the
  * UNLOCK and LOCK operate on the same lock variable.
  */
-#ifdef CONFIG_PPC
+#ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE
 #define smp_mb__after_unlock_lock()	smp_mb()  /* Full ordering for lock. */
-#else /* #ifdef CONFIG_PPC */
+#else /* #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */
 #define smp_mb__after_unlock_lock()	do { } while (0)
-#endif /* #else #ifdef CONFIG_PPC */
+#endif /* #else #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */
 
 
 #endif /* __LINUX_RCUPDATE_H */

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-19 17:07 UTC #119 lore raw

PATCH v3 3/40 srcu: Consolidate batch checking into rcu_all_batches_empty()

---
 kernel/rcu/srcu.c | 21 +++++++++++++--------
 1 file changed, 13 insertions(+), 8 deletions(-)

kernel/rcu/srcu.c +13 -8

index ef3bcfb15b39..ba41a5d04b49 100644
@@ -65,6 +65,17 @@ static inline bool rcu_batch_empty(struct rcu_batch *b)
 }
 
 /*
+ * Are all batches empty for the specified srcu_struct?
+ */
+static inline bool rcu_all_batches_empty(struct srcu_struct *sp)
+{
+	return rcu_batch_empty(&sp->batch_done) &&
+	       rcu_batch_empty(&sp->batch_check1) &&
+	       rcu_batch_empty(&sp->batch_check0) &&
+	       rcu_batch_empty(&sp->batch_queue);
+}
+
+/*
  * Remove the callback at the head of the specified rcu_batch structure
  * and return a pointer to it, or return NULL if the structure is empty.
  */

@@ -619,15 +630,9 @@ static void srcu_reschedule(struct srcu_struct *sp)
 {
 	bool pending = true;
 
-	if (rcu_batch_empty(&sp->batch_done) &&
-	    rcu_batch_empty(&sp->batch_check1) &&
-	    rcu_batch_empty(&sp->batch_check0) &&
-	    rcu_batch_empty(&sp->batch_queue)) {
+	if (rcu_all_batches_empty(sp)) {
 		spin_lock_irq(&sp->queue_lock);
-		if (rcu_batch_empty(&sp->batch_done) &&
-		    rcu_batch_empty(&sp->batch_check1) &&
-		    rcu_batch_empty(&sp->batch_check0) &&
-		    rcu_batch_empty(&sp->batch_queue)) {
+		if (rcu_all_batches_empty(sp)) {
 			sp->running = false;
 			pending = false;
 		}

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-19 17:08 UTC #120 lore raw

PATCH v3 4/40 srcu: Check for tardy grace-period activity in cleanup_srcu_struct()

Therefore, if there has ever been an invocation of call_srcu() on the srcu_struct in question, the sequence of events must be as follows:

1.  Prevent any further calls to call_srcu().
2.  Wait for any pre-existing call_srcu() invocations to return.
3.  Invoke srcu_barrier().
4.  It is now safe to invoke cleanup_srcu_struct().

On the other hand, if there has ever been a call to synchronize_srcu() or synchronize_srcu_expedited(), the sequence of events must be as follows:

Reported-by Paolo Bonzini <[email protected]>

---
 kernel/rcu/srcu.c | 15 +++++++++++++--
 1 file changed, 13 insertions(+), 2 deletions(-)

kernel/rcu/srcu.c +13 -2

index ba41a5d04b49..e6da9fc1f0e9 100644
@@ -254,13 +254,24 @@ static bool srcu_readers_active(struct srcu_struct *sp)
  * cleanup_srcu_struct - deconstruct a sleep-RCU structure
  * @sp: structure to clean up.
  *
- * Must invoke this after you are finished using a given srcu_struct that
- * was initialized via init_srcu_struct(), else you leak memory.
+ * Must invoke this only after you are finished using a given srcu_struct
+ * that was initialized via init_srcu_struct().  This code does some
+ * probabalistic checking, spotting late uses of srcu_read_lock(),
+ * synchronize_srcu(), synchronize_srcu_expedited(), and call_srcu().
+ * If any such late uses are detected, the per-CPU memory associated with
+ * the srcu_struct is simply leaked and WARN_ON() is invoked.  If the
+ * caller frees the srcu_struct itself, a use-after-free crash will likely
+ * ensue, but at least there will be a warning printed.
  */
 void cleanup_srcu_struct(struct srcu_struct *sp)
 {
 	if (WARN_ON(srcu_readers_active(sp)))
 		return; /* Leakage unless caller handles error. */
+	if (WARN_ON(!rcu_all_batches_empty(sp)))
+		return; /* Leakage unless caller handles error. */
+	flush_delayed_work(&sp->work);
+	if (WARN_ON(sp->running))
+		return; /* Caller forgot to stop doing call_srcu()? */
 	free_percpu(sp->per_cpu_ref);
 	sp->per_cpu_ref = NULL;
 }

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-19 17:04 UTC #121 lore raw

PATCH v3 5/40 rcu: Semicolon inside RCU_TRACE() for rcu.h

---
 kernel/rcu/rcu.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

kernel/rcu/rcu.h +2 -2

index 0d6ff3e471be..8700a81daf56 100644
@@ -109,12 +109,12 @@ static inline bool __rcu_reclaim(const char *rn, struct rcu_head *head)
 
 	rcu_lock_acquire(&rcu_callback_map);
 	if (__is_kfree_rcu_offset(offset)) {
-		RCU_TRACE(trace_rcu_invoke_kfree_callback(rn, head, offset));
+		RCU_TRACE(trace_rcu_invoke_kfree_callback(rn, head, offset);)
 		kfree((void *)head - offset);
 		rcu_lock_release(&rcu_callback_map);
 		return true;
 	} else {
-		RCU_TRACE(trace_rcu_invoke_callback(rn, head));
+		RCU_TRACE(trace_rcu_invoke_callback(rn, head);)
 		head->func(head);
 		rcu_lock_release(&rcu_callback_map);
 		return false;

Signature · 1 line

2.5.2

Joe Perches

2017-04-19 17:41 UTC #122 lore raw

Re: [PATCH v3 tip/core/rcu 05/40] rcu: Semicolon inside RCU_TRACE() for rcu.h

Quoted 4 lines · Paul E. McKenney

The current use of "RCU_TRACE(statement);" can cause odd bugs, especially where "statement" is a local-variable declaration, as it can leave a misplaced ";" in the source code. This commit therefore converts these to "RCU_TRACE(statement;)", which avoids the misplaced ";".

Perhaps instead the single RCU_TRACE macro should be split into RCU_TRACE_DECLARE and RCU_TRACE, and then RCU_TRACE should use do {} while (0)

Paul E. McKenney

author

2017-04-25 21:00 UTC #123 lore raw

Re: [PATCH v3 tip/core/rcu 05/40] rcu: Semicolon inside RCU_TRACE() for rcu.h

Quoted 9 lines · Joe Perches

Quoted 4 lines · Paul E. McKenney
The current use of "RCU_TRACE(statement);" can cause odd bugs, especially where "statement" is a local-variable declaration, as it can leave a misplaced ";" in the source code. This commit therefore converts these to "RCU_TRACE(statement;)", which avoids the misplaced ";".
Perhaps instead the single RCU_TRACE macro should be split into RCU_TRACE_DECLARE and RCU_TRACE, and then RCU_TRACE should use do {} while (0)

Cute trick, but I am having a hard time getting very excited either way. Thoughts from others?

Thanx, Paul

Steven Rostedt

reviewer

2017-04-25 21:14 UTC #124 lore raw

Re: [PATCH v3 tip/core/rcu 05/40] rcu: Semicolon inside RCU_TRACE() for rcu.h

Quoted 16 lines · On Wed, 19 Apr 2017 09:58:02 -0700 "Paul E. McKenney"…

The current use of "RCU_TRACE(statement);" can cause odd bugs, especially where "statement" is a local-variable declaration, as it can leave a misplaced ";" in the source code. This commit therefore converts these to "RCU_TRACE(statement;)", which avoids the misplaced ";".
Reported-by Josh Triplett <[email protected]>
Signed-off-by Paul E. McKenney <[email protected]>
---
 kernel/rcu/rcu.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
kernel/rcu/rcu.h +0 -0
index 0d6ff3e471be..8700a81daf56 100644

kernel/rcu/rcu.h static inline bool __rcu_reclaim(const char *rn, struct rcu_head *head) · line 112

Quoted 5 lines

kernel/rcu/rcu.h +1 -1

@@ -109,12 +109,12 @@ static inline bool __rcu_reclaim(const char *rn, struct rcu_head *head)
 
 	rcu_lock_acquire(&rcu_callback_map);
 	if (__is_kfree_rcu_offset(offset)) {
-		RCU_TRACE(trace_rcu_invoke_kfree_callback(rn, head, offset));
+		RCU_TRACE(trace_rcu_invoke_kfree_callback(rn, head, offset);)

I have to ask. Why is this a RCU_TRACE() anyway? tracepoints use jump labels, and this is basically a nop here when tracing is off.

-- Steve

Quoted 9 lines

kernel/rcu/rcu.h +1 -1

 		kfree((void *)head - offset);
 		rcu_lock_release(&rcu_callback_map);
 		return true;
 	} else {
-		RCU_TRACE(trace_rcu_invoke_callback(rn, head));
+		RCU_TRACE(trace_rcu_invoke_callback(rn, head);)
 		head->func(head);
 		rcu_lock_release(&rcu_callback_map);
 		return false;

Paul E. McKenney

author

2017-04-25 21:27 UTC #125 lore raw

Re: [PATCH v3 tip/core/rcu 05/40] rcu: Semicolon inside RCU_TRACE() for rcu.h

Quoted 20 lines · Steven Rostedt

Quoted 17 lines · On Wed, 19 Apr 2017 09:58:02 -0700 "Paul E. McKenney"…
The current use of "RCU_TRACE(statement);" can cause odd bugs, especially where "statement" is a local-variable declaration, as it can leave a misplaced ";" in the source code. This commit therefore converts these to "RCU_TRACE(statement;)", which avoids the misplaced ";".
Reported-by Josh Triplett <[email protected]>
Signed-off-by Paul E. McKenney <[email protected]>
---
 kernel/rcu/rcu.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
kernel/rcu/rcu.h +0 -0
index 0d6ff3e471be..8700a81daf56 100644
@@ -109,12 +109,12 @@ static inline bool __rcu_reclaim(const char *rn, struct rcu_head *head)
 

Quoted 7 lines

kernel/rcu/rcu.h static inline bool __rcu_reclaim(const char *rn, struct rcu_head *head) · line 112
Quoted 4 lines
kernel/rcu/rcu.h +1 -1
 	rcu_lock_acquire(&rcu_callback_map);
 	if (__is_kfree_rcu_offset(offset)) {
-		RCU_TRACE(trace_rcu_invoke_kfree_callback(rn, head, offset));
+		RCU_TRACE(trace_rcu_invoke_kfree_callback(rn, head, offset);)
I have to ask. Why is this a RCU_TRACE() anyway? tracepoints use jump labels, and this is basically a nop here when tracing is off.

Because otherwise the build breaks if CONFIG_RCU_TRACE=n. So it is not the tracepoints themselves, but their helper functions. Worse, the variables declared for them. Even worse, in Tiny RCU, the fields in structures that are used only for tracing.

I freely admit that there might be a better way to structure this, give or take the Tiny RCU structure fields.

Thanx, Paul

Quoted 13 lines

-- Steve

Quoted 9 lines

kernel/rcu/rcu.h +1 -1

 		kfree((void *)head - offset);
 		rcu_lock_release(&rcu_callback_map);
 		return true;
 	} else {
-		RCU_TRACE(trace_rcu_invoke_callback(rn, head));
+		RCU_TRACE(trace_rcu_invoke_callback(rn, head);)
 		head->func(head);
 		rcu_lock_release(&rcu_callback_map);
 		return false;

Paul E. McKenney

author

2017-04-19 17:07 UTC #126 lore raw

PATCH v3 6/40 rcu: Semicolon inside RCU_TRACE() for Tiny RCU

---
 kernel/rcu/tiny.c        | 20 ++++++++++----------
 kernel/rcu/tiny_plugin.h |  4 ++--
 2 files changed, 12 insertions(+), 12 deletions(-)

kernel/rcu/tiny.c +10 ~~-10~~

index 6ad330dbbae2..e5385731e391 100644
@@ -79,7 +79,7 @@ EXPORT_SYMBOL(__rcu_is_watching);
  */
 static int rcu_qsctr_help(struct rcu_ctrlblk *rcp)
 {
-	RCU_TRACE(reset_cpu_stall_ticks(rcp));
+	RCU_TRACE(reset_cpu_stall_ticks(rcp);)
 	if (rcp->donetail != rcp->curtail) {
 		rcp->donetail = rcp->curtail;
 		return 1;

@@ -125,7 +125,7 @@ void rcu_bh_qs(void)
  */
 void rcu_check_callbacks(int user)
 {
-	RCU_TRACE(check_cpu_stalls());
+	RCU_TRACE(check_cpu_stalls();)
 	if (user)
 		rcu_sched_qs();
 	else if (!in_softirq())

@@ -143,7 +143,7 @@ static void __rcu_process_callbacks(struct rcu_ctrlblk *rcp)
 	const char *rn = NULL;
 	struct rcu_head *next, *list;
 	unsigned long flags;
-	RCU_TRACE(int cb_count = 0);
+	RCU_TRACE(int cb_count = 0;)
 
 	/* Move the ready-to-invoke callbacks to a local list. */
 	local_irq_save(flags);

@@ -152,7 +152,7 @@ static void __rcu_process_callbacks(struct rcu_ctrlblk *rcp)
 		local_irq_restore(flags);
 		return;
 	}
-	RCU_TRACE(trace_rcu_batch_start(rcp->name, 0, rcp->qlen, -1));
+	RCU_TRACE(trace_rcu_batch_start(rcp->name, 0, rcp->qlen, -1);)
 	list = rcp->rcucblist;
 	rcp->rcucblist = *rcp->donetail;
 	*rcp->donetail = NULL;

@@ -162,7 +162,7 @@ static void __rcu_process_callbacks(struct rcu_ctrlblk *rcp)
 	local_irq_restore(flags);
 
 	/* Invoke the callbacks on the local list. */
-	RCU_TRACE(rn = rcp->name);
+	RCU_TRACE(rn = rcp->name;)
 	while (list) {
 		next = list->next;
 		prefetch(next);

@@ -171,9 +171,9 @@ static void __rcu_process_callbacks(struct rcu_ctrlblk *rcp)
 		__rcu_reclaim(rn, list);
 		local_bh_enable();
 		list = next;
-		RCU_TRACE(cb_count++);
+		RCU_TRACE(cb_count++;)
 	}
-	RCU_TRACE(rcu_trace_sub_qlen(rcp, cb_count));
+	RCU_TRACE(rcu_trace_sub_qlen(rcp, cb_count);)
 	RCU_TRACE(trace_rcu_batch_end(rcp->name,
 				      cb_count, 0, need_resched(),
 				      is_idle_task(current),

@@ -221,7 +221,7 @@ static void __call_rcu(struct rcu_head *head,
 	local_irq_save(flags);
 	*rcp->curtail = head;
 	rcp->curtail = &head->next;
-	RCU_TRACE(rcp->qlen++);
+	RCU_TRACE(rcp->qlen++;)
 	local_irq_restore(flags);
 
 	if (unlikely(is_idle_task(current))) {

@@ -254,8 +254,8 @@ EXPORT_SYMBOL_GPL(call_rcu_bh);
 void __init rcu_init(void)
 {
 	open_softirq(RCU_SOFTIRQ, rcu_process_callbacks);
-	RCU_TRACE(reset_cpu_stall_ticks(&rcu_sched_ctrlblk));
-	RCU_TRACE(reset_cpu_stall_ticks(&rcu_bh_ctrlblk));
+	RCU_TRACE(reset_cpu_stall_ticks(&rcu_sched_ctrlblk);)
+	RCU_TRACE(reset_cpu_stall_ticks(&rcu_bh_ctrlblk);)
 
 	rcu_early_boot_tests();
 }

kernel/rcu/tiny_plugin.h +2 -2

index c64b827ecbca..df3a60e19f07 100644
@@ -162,8 +162,8 @@ static void reset_cpu_stall_ticks(struct rcu_ctrlblk *rcp)
 
 static void check_cpu_stalls(void)
 {
-	RCU_TRACE(check_cpu_stall(&rcu_bh_ctrlblk));
-	RCU_TRACE(check_cpu_stall(&rcu_sched_ctrlblk));
+	RCU_TRACE(check_cpu_stall(&rcu_bh_ctrlblk);)
+	RCU_TRACE(check_cpu_stall(&rcu_sched_ctrlblk);)
 }
 
 #endif /* #ifdef CONFIG_RCU_TRACE */

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-19 17:04 UTC #127 lore raw

PATCH v3 7/40 rcu: Semicolon inside RCU_TRACE() for tree.c

---
 kernel/rcu/tree.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

kernel/rcu/tree.c +4 -4

index 0efad311ded4..3747277aae67 100644
@@ -2793,14 +2793,14 @@ static void rcu_adopt_orphan_cbs(struct rcu_state *rsp, unsigned long flags)
  */
 static void rcu_cleanup_dying_cpu(struct rcu_state *rsp)
 {
-	RCU_TRACE(unsigned long mask);
-	RCU_TRACE(struct rcu_data *rdp = this_cpu_ptr(rsp->rda));
-	RCU_TRACE(struct rcu_node *rnp = rdp->mynode);
+	RCU_TRACE(unsigned long mask;)
+	RCU_TRACE(struct rcu_data *rdp = this_cpu_ptr(rsp->rda);)
+	RCU_TRACE(struct rcu_node *rnp = rdp->mynode;)
 
 	if (!IS_ENABLED(CONFIG_HOTPLUG_CPU))
 		return;
 
-	RCU_TRACE(mask = rdp->grpmask);
+	RCU_TRACE(mask = rdp->grpmask;)
 	trace_rcu_grace_period(rsp->name,
 			       rnp->gpnum + 1 - !!(rnp->qsmask & mask),
 			       TPS("cpuofl"));

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-19 16:59 UTC #128 lore raw

PATCH v3 8/40 rcu: Pull rcu_sched_qs_mask into rcu_dynticks structure

The rcu_sched_qs_mask variable is yet another isolated per-CPU variable, so this commit pulls it into the pre-existing rcu_dynticks per-CPU structure.

---
 .../RCU/Design/Data-Structures/Data-Structures.html          |  9 ++++++++-
 kernel/rcu/tree.c                                            | 12 +++++-------
 kernel/rcu/tree.h                                            |  1 +
 3 files changed, 14 insertions(+), 8 deletions(-)

Documentation/RCU/Design/Data-Structures/Data-Structures.html +8 -1

index d583c653a703..bf7f266e8888 100644
@@ -1104,6 +1104,7 @@ Its fields are as follows:
   1   int dynticks_nesting;
   2   int dynticks_nmi_nesting;
   3   atomic_t dynticks;
+  4   int rcu_sched_qs_mask;
 </pre>
 
 <p>The <tt>-&gt;dynticks_nesting</tt> field counts the

@@ -1117,11 +1118,17 @@ NMIs are counted by the <tt>-&gt;dynticks_nmi_nesting</tt>
 field, except that NMIs that interrupt non-dyntick-idle execution
 are not counted.
 
-</p><p>Finally, the <tt>-&gt;dynticks</tt> field counts the corresponding
+</p><p>The <tt>-&gt;dynticks</tt> field counts the corresponding
 CPU's transitions to and from dyntick-idle mode, so that this counter
 has an even value when the CPU is in dyntick-idle mode and an odd
 value otherwise.
 
+</p><p>Finally, the  <tt>-&gt;rcu_sched_qs_mask</tt> field is used
+to record the fact that the RCU core code would really like to
+see a quiescent state from the corresponding CPU.
+This flag is checked by RCU's context-switch and <tt>cond_resched()</tt>
+code, which provide a momentary idle sojourn in response.
+
 <table>
 <tr><th>&nbsp;</th></tr>
 <tr><th align="left">Quick Quiz:</th></tr>

kernel/rcu/tree.c +5 -7

index 3747277aae67..3a0703035874 100644
@@ -272,8 +272,6 @@ void rcu_bh_qs(void)
 	}
 }
 
-static DEFINE_PER_CPU(int, rcu_sched_qs_mask);
-
 /*
  * Steal a bit from the bottom of ->dynticks for idle entry/exit
  * control.  Initially this is for TLB flushing.

@@ -464,8 +462,8 @@ static void rcu_momentary_dyntick_idle(void)
 	 * Yes, we can lose flag-setting operations.  This is OK, because
 	 * the flag will be set again after some delay.
 	 */
-	resched_mask = raw_cpu_read(rcu_sched_qs_mask);
-	raw_cpu_write(rcu_sched_qs_mask, 0);
+	resched_mask = raw_cpu_read(rcu_dynticks.rcu_sched_qs_mask);
+	raw_cpu_write(rcu_dynticks.rcu_sched_qs_mask, 0);
 
 	/* Find the flavor that needs a quiescent state. */
 	for_each_rcu_flavor(rsp) {

@@ -499,7 +497,7 @@ void rcu_note_context_switch(void)
 	trace_rcu_utilization(TPS("Start context switch"));
 	rcu_sched_qs();
 	rcu_preempt_note_context_switch();
-	if (unlikely(raw_cpu_read(rcu_sched_qs_mask)))
+	if (unlikely(raw_cpu_read(rcu_dynticks.rcu_sched_qs_mask)))
 		rcu_momentary_dyntick_idle();
 	trace_rcu_utilization(TPS("End context switch"));
 	barrier(); /* Avoid RCU read-side critical sections leaking up. */

@@ -524,7 +522,7 @@ void rcu_all_qs(void)
 	unsigned long flags;
 
 	barrier(); /* Avoid RCU read-side critical sections leaking down. */
-	if (unlikely(raw_cpu_read(rcu_sched_qs_mask))) {
+	if (unlikely(raw_cpu_read(rcu_dynticks.rcu_sched_qs_mask))) {
 		local_irq_save(flags);
 		rcu_momentary_dyntick_idle();
 		local_irq_restore(flags);

@@ -1351,7 +1349,7 @@ static int rcu_implicit_dynticks_qs(struct rcu_data *rdp,
 	 * is set too high, we override with half of the RCU CPU stall
 	 * warning delay.
 	 */
-	rcrmp = &per_cpu(rcu_sched_qs_mask, rdp->cpu);
+	rcrmp = &per_cpu(rcu_dynticks.rcu_sched_qs_mask, rdp->cpu);
 	if (time_after(jiffies, rdp->rsp->gp_start + jtsq) ||
 	    time_after(jiffies, rdp->rsp->jiffies_resched)) {
 		if (!(READ_ONCE(*rcrmp) & rdp->rsp->flavor_mask)) {

kernel/rcu/tree.h +1 -0

index 7468b4de7e0c..e298281984dc 100644
@@ -113,6 +113,7 @@ struct rcu_dynticks {
 				    /* Process level is worth LLONG_MAX/2. */
 	int dynticks_nmi_nesting;   /* Track NMI nesting level. */
 	atomic_t dynticks;	    /* Even value for idle, else odd. */
+	int rcu_sched_qs_mask;      /* GP old, need quiescent state. */
 #ifdef CONFIG_NO_HZ_FULL_SYSIDLE
 	long long dynticks_idle_nesting;
 				    /* irq/process nesting level from idle. */

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-19 17:08 UTC #129 lore raw

PATCH v3 9/40 rcu: Pull rcu_qs_ctr into rcu_dynticks structure

The rcu_qs_ctr variable is yet another isolated per-CPU variable, so this commit pulls it into the pre-existing rcu_dynticks per-CPU structure.

---
 .../RCU/Design/Data-Structures/Data-Structures.html       | 12 ++++++++++--
 kernel/rcu/tree.c                                         | 15 ++++++---------
 kernel/rcu/tree.h                                         |  3 ++-
 kernel/rcu/tree_trace.c                                   |  4 +---
 4 files changed, 19 insertions(+), 15 deletions(-)

Documentation/RCU/Design/Data-Structures/Data-Structures.html +10 -2

index bf7f266e8888..3d0311657533 100644
@@ -1105,6 +1105,7 @@ Its fields are as follows:
   2   int dynticks_nmi_nesting;
   3   atomic_t dynticks;
   4   int rcu_sched_qs_mask;
+  5   unsigned long rcu_qs_ctr;
 </pre>
 
 <p>The <tt>-&gt;dynticks_nesting</tt> field counts the

@@ -1123,12 +1124,19 @@ CPU's transitions to and from dyntick-idle mode, so that this counter
 has an even value when the CPU is in dyntick-idle mode and an odd
 value otherwise.
 
-</p><p>Finally, the  <tt>-&gt;rcu_sched_qs_mask</tt> field is used
+</p><p>The <tt>-&gt;rcu_sched_qs_mask</tt> field is used
 to record the fact that the RCU core code would really like to
-see a quiescent state from the corresponding CPU.
+see a quiescent state from the corresponding CPU, so much so that
+it is willing to call for heavy-weight dyntick-counter operations.
 This flag is checked by RCU's context-switch and <tt>cond_resched()</tt>
 code, which provide a momentary idle sojourn in response.
 
+</p><p>Finally the <tt>-&gt;rcu_qs_ctr</tt> field is used to record
+quiescent states from <tt>cond_resched()</tt>.
+Because <tt>cond_resched()</tt> can execute quite frequently, this
+must be quite lightweight, as in a non-atomic increment of this
+per-CPU field.
+
 <table>
 <tr><th>&nbsp;</th></tr>
 <tr><th align="left">Quick Quiz:</th></tr>

kernel/rcu/tree.c +6 -9

index 3a0703035874..82a86a67c92a 100644
@@ -436,9 +436,6 @@ bool rcu_eqs_special_set(int cpu)
 	return true;
 }
 
-DEFINE_PER_CPU_SHARED_ALIGNED(unsigned long, rcu_qs_ctr);
-EXPORT_PER_CPU_SYMBOL_GPL(rcu_qs_ctr);
-
 /*
  * Let the RCU core know that this CPU has gone through the scheduler,
  * which is a quiescent state.  This is called when the need for a

@@ -542,7 +539,7 @@ void rcu_all_qs(void)
 		rcu_sched_qs();
 		preempt_enable();
 	}
-	this_cpu_inc(rcu_qs_ctr);
+	this_cpu_inc(rcu_dynticks.rcu_qs_ctr);
 	barrier(); /* Avoid RCU read-side critical sections leaking up. */
 }
 EXPORT_SYMBOL_GPL(rcu_all_qs);

@@ -1315,7 +1312,7 @@ static int rcu_implicit_dynticks_qs(struct rcu_data *rdp,
 	 */
 	rnp = rdp->mynode;
 	if (time_after(jiffies, rdp->rsp->gp_start + jtsq) &&
-	    READ_ONCE(rdp->rcu_qs_ctr_snap) != per_cpu(rcu_qs_ctr, rdp->cpu) &&
+	    READ_ONCE(rdp->rcu_qs_ctr_snap) != per_cpu(rcu_dynticks.rcu_qs_ctr, rdp->cpu) &&
 	    READ_ONCE(rdp->gpnum) == rnp->gpnum && !rdp->gpwrap) {
 		trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, TPS("rqc"));
 		return 1;

@@ -2024,7 +2021,7 @@ static bool __note_gp_changes(struct rcu_state *rsp, struct rcu_node *rnp,
 		trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("cpustart"));
 		need_gp = !!(rnp->qsmask & rdp->grpmask);
 		rdp->cpu_no_qs.b.norm = need_gp;
-		rdp->rcu_qs_ctr_snap = __this_cpu_read(rcu_qs_ctr);
+		rdp->rcu_qs_ctr_snap = __this_cpu_read(rcu_dynticks.rcu_qs_ctr);
 		rdp->core_needs_qs = need_gp;
 		zero_cpu_stall_ticks(rdp);
 		WRITE_ONCE(rdp->gpwrap, false);

@@ -2622,7 +2619,7 @@ rcu_report_qs_rdp(int cpu, struct rcu_state *rsp, struct rcu_data *rdp)
 		 * within the current grace period.
 		 */
 		rdp->cpu_no_qs.b.norm = true;	/* need qs for new gp. */
-		rdp->rcu_qs_ctr_snap = __this_cpu_read(rcu_qs_ctr);
+		rdp->rcu_qs_ctr_snap = __this_cpu_read(rcu_dynticks.rcu_qs_ctr);
 		raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
 		return;
 	}

@@ -3620,7 +3617,7 @@ static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp)
 	/* Is the RCU core waiting for a quiescent state from this CPU? */
 	if (rcu_scheduler_fully_active &&
 	    rdp->core_needs_qs && rdp->cpu_no_qs.b.norm &&
-	    rdp->rcu_qs_ctr_snap == __this_cpu_read(rcu_qs_ctr)) {
+	    rdp->rcu_qs_ctr_snap == __this_cpu_read(rcu_dynticks.rcu_qs_ctr)) {
 		rdp->n_rp_core_needs_qs++;
 	} else if (rdp->core_needs_qs && !rdp->cpu_no_qs.b.norm) {
 		rdp->n_rp_report_qs++;

@@ -3933,7 +3930,7 @@ rcu_init_percpu_data(int cpu, struct rcu_state *rsp)
 	rdp->gpnum = rnp->completed; /* Make CPU later note any new GP. */
 	rdp->completed = rnp->completed;
 	rdp->cpu_no_qs.b.norm = true;
-	rdp->rcu_qs_ctr_snap = per_cpu(rcu_qs_ctr, cpu);
+	rdp->rcu_qs_ctr_snap = per_cpu(rcu_dynticks.rcu_qs_ctr, cpu);
 	rdp->core_needs_qs = false;
 	trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("cpuonl"));
 	raw_spin_unlock_irqrestore_rcu_node(rnp, flags);

kernel/rcu/tree.h +2 -1

index e298281984dc..76e4467bc765 100644
@@ -113,7 +113,8 @@ struct rcu_dynticks {
 				    /* Process level is worth LLONG_MAX/2. */
 	int dynticks_nmi_nesting;   /* Track NMI nesting level. */
 	atomic_t dynticks;	    /* Even value for idle, else odd. */
-	int rcu_sched_qs_mask;      /* GP old, need quiescent state. */
+	int rcu_sched_qs_mask;      /* GP old, need heavy quiescent state. */
+	unsigned long rcu_qs_ctr;   /* Light universal quiescent state ctr. */
 #ifdef CONFIG_NO_HZ_FULL_SYSIDLE
 	long long dynticks_idle_nesting;
 				    /* irq/process nesting level from idle. */

kernel/rcu/tree_trace.c +1 -3

index 8751a748499a..65b43be38e68 100644
@@ -45,8 +45,6 @@
 #define RCU_TREE_NONCORE
 #include "tree.h"
 
-DECLARE_PER_CPU_SHARED_ALIGNED(unsigned long, rcu_qs_ctr);
-
 static int r_open(struct inode *inode, struct file *file,
 					const struct seq_operations *op)
 {

@@ -121,7 +119,7 @@ static void print_one_rcu_data(struct seq_file *m, struct rcu_data *rdp)
 		   cpu_is_offline(rdp->cpu) ? '!' : ' ',
 		   ulong2long(rdp->completed), ulong2long(rdp->gpnum),
 		   rdp->cpu_no_qs.b.norm,
-		   rdp->rcu_qs_ctr_snap == per_cpu(rcu_qs_ctr, rdp->cpu),
+		   rdp->rcu_qs_ctr_snap == per_cpu(rdp->dynticks->rcu_qs_ctr, rdp->cpu),
 		   rdp->core_needs_qs);
 	seq_printf(m, " dt=%d/%llx/%d df=%lu",
 		   rcu_dynticks_snap(rdp->dynticks),

Signature · 1 line

2.5.2

Paul E. McKenney

author

2017-04-19 17:08 UTC #130 lore raw

PATCH v3 10/40 rcu: Eliminate flavor scan in rcu_momentary_dyntick_idle()