* [PATCH 0/5] Updating Section Deferred Processing
@ 2017-06-14 15:44 Junchang Wang
2017-06-14 15:45 ` [PATCH 1/5] hazptr: Switch from ACCESS_ONCE() to READ_ONCE()/WRITE_ONCE() Junchang Wang
` (5 more replies)
0 siblings, 6 replies; 7+ messages in thread
From: Junchang Wang @ 2017-06-14 15:44 UTC (permalink / raw)
To: perfbook, paulmck; +Cc: Junchang Wang
Hi Paul,
Most of the following patches switch from ACCESS_ONCE to READ_ONCE/WRITE_ONCE in
Section Deferred Processing. Please take a look.
Thanks,
--Jason
Junchang Wang (5):
hazptr: Switch from ACCESS_ONCE() to READ_ONCE()/WRITE_ONCE()
route_rcu: Switch from ACCESS_ONCE() to READ_ONCE()/WRITE_ONCE()
route_rcu: Remove redundant assignment statement to rep->re_freed
seqlocktorture: Switch from ACCESS_ONCE() to READ_ONCE()/WRITE_ONCE()
rcu_rcpls: Switch from ACCESS_ONCE() to READ_ONCE()/WRITE_ONCE()
CodeSamples/defer/hazptr.h | 8 ++++----
CodeSamples/defer/hazptrtorture.h | 6 +++---
CodeSamples/defer/rcu_rcpls.c | 6 +++---
CodeSamples/defer/rcu_rcpls.h | 2 +-
CodeSamples/defer/rcutorture.h | 2 +-
CodeSamples/defer/route_rcu.c | 5 ++---
CodeSamples/defer/seqlocktorture.c | 8 ++++----
appendix/toyrcu/toyrcu.tex | 14 +++++++-------
defer/rcuusage.tex | 4 ++--
9 files changed, 27 insertions(+), 28 deletions(-)
--
2.7.4
^ permalink raw reply [flat|nested] 7+ messages in thread
* [PATCH 1/5] hazptr: Switch from ACCESS_ONCE() to READ_ONCE()/WRITE_ONCE()
2017-06-14 15:44 [PATCH 0/5] Updating Section Deferred Processing Junchang Wang
@ 2017-06-14 15:45 ` Junchang Wang
2017-06-14 15:45 ` [PATCH 2/5] route_rcu: " Junchang Wang
` (4 subsequent siblings)
5 siblings, 0 replies; 7+ messages in thread
From: Junchang Wang @ 2017-06-14 15:45 UTC (permalink / raw)
To: perfbook, paulmck; +Cc: Junchang Wang
Signed-off-by: Junchang Wang <junchangwang@gmail.com>
---
CodeSamples/defer/hazptr.h | 8 ++++----
CodeSamples/defer/hazptrtorture.h | 6 +++---
2 files changed, 7 insertions(+), 7 deletions(-)
diff --git a/CodeSamples/defer/hazptr.h b/CodeSamples/defer/hazptr.h
index 391d281..48bec0e 100644
--- a/CodeSamples/defer/hazptr.h
+++ b/CodeSamples/defer/hazptr.h
@@ -65,17 +65,17 @@ static inline void *hp_record(void **p, hazard_pointer *hp)
hazptr_head_t *tmp;
do {
- tmp = ACCESS_ONCE(*p);
- ACCESS_ONCE(hp->p) = tmp;
+ tmp = READ_ONCE(*p);
+ WRITE_ONCE(hp->p, tmp);
smp_mb();
- } while (tmp != ACCESS_ONCE(*p));
+ } while (tmp != READ_ONCE(*p));
return tmp;
}
static inline void hp_clear(hazard_pointer *hp)
{
smp_mb();
- ACCESS_ONCE(hp->p) = NULL;
+ WRITE_ONCE(hp->p, NULL);
}
#define hazptr_clean_pointer(p) ((typeof(p))((unsigned long)(p) & ~0x1UL))
diff --git a/CodeSamples/defer/hazptrtorture.h b/CodeSamples/defer/hazptrtorture.h
index 9341765..2188ec1 100644
--- a/CodeSamples/defer/hazptrtorture.h
+++ b/CodeSamples/defer/hazptrtorture.h
@@ -286,17 +286,17 @@ void *hazptr_update_stress_test(void *arg)
if (i >= HAZPTR_STRESS_PIPE_LEN)
i = 0;
p = &hazptr_stress_array[i];
- if (ACCESS_ONCE(p->inuse)) {
+ if (READ_ONCE(p->inuse)) {
hazptr_stress_idx = i;
continue;
}
p->pipe_count = 0;
- ACCESS_ONCE(p->inuse) = 1;
+ WRITE_ONCE(p->inuse, 1);
p->mbtest = 0;
smp_mb();
p->mbtest = 1;
smp_mb();
- ACCESS_ONCE(hazptr_stress_current) = p;
+ WRITE_ONCE(hazptr_stress_current, p);
hazptr_stress_idx = i;
for (i = 0; i < HAZPTR_STRESS_PIPE_LEN; i++)
if (!hazptr_stress_array[i].inuse) {
--
2.7.4
^ permalink raw reply related [flat|nested] 7+ messages in thread
* [PATCH 2/5] route_rcu: Switch from ACCESS_ONCE() to READ_ONCE()/WRITE_ONCE()
2017-06-14 15:44 [PATCH 0/5] Updating Section Deferred Processing Junchang Wang
2017-06-14 15:45 ` [PATCH 1/5] hazptr: Switch from ACCESS_ONCE() to READ_ONCE()/WRITE_ONCE() Junchang Wang
@ 2017-06-14 15:45 ` Junchang Wang
2017-06-14 15:45 ` [PATCH 3/5] route_rcu: Remove redundant assignment statement to rep->re_freed Junchang Wang
` (3 subsequent siblings)
5 siblings, 0 replies; 7+ messages in thread
From: Junchang Wang @ 2017-06-14 15:45 UTC (permalink / raw)
To: perfbook, paulmck; +Cc: Junchang Wang
Signed-off-by: Junchang Wang <junchangwang@gmail.com>
---
CodeSamples/defer/route_rcu.c | 6 +++---
defer/rcuusage.tex | 4 ++--
2 files changed, 5 insertions(+), 5 deletions(-)
diff --git a/CodeSamples/defer/route_rcu.c b/CodeSamples/defer/route_rcu.c
index 9ac7b25..7716fe8 100644
--- a/CodeSamples/defer/route_rcu.c
+++ b/CodeSamples/defer/route_rcu.c
@@ -44,7 +44,7 @@ DEFINE_SPINLOCK(routelock);
static void re_free(struct route_entry *rep)
{
- ACCESS_ONCE(rep->re_freed) = 1;
+ WRITE_ONCE(rep->re_freed, 1);
free(rep);
}
@@ -60,7 +60,7 @@ unsigned long route_lookup(unsigned long addr)
cds_list_for_each_entry_rcu(rep, &route_list, re_next) {
if (rep->addr == addr) {
ret = rep->iface;
- if (ACCESS_ONCE(rep->re_freed))
+ if (READ_ONCE(rep->re_freed))
abort();
rcu_read_unlock();
return ret;
@@ -93,7 +93,7 @@ static void route_cb(struct rcu_head *rhp)
{
struct route_entry *rep = container_of(rhp, struct route_entry, rh);
- ACCESS_ONCE(rep->re_freed) = 1;
+ WRITE_ONCE(rep->re_freed, 1);
re_free(rep);
}
diff --git a/defer/rcuusage.tex b/defer/rcuusage.tex
index 0a37312..9a90a14 100644
--- a/defer/rcuusage.tex
+++ b/defer/rcuusage.tex
@@ -64,7 +64,7 @@ Section~\ref{sec:defer:RCU Usage Summary} provides a summary.
18 re_next) {
19 if (rep->addr == addr) {
20 ret = rep->iface;
-21 if (ACCESS_ONCE(rep->re_freed))
+21 if (READ_ONCE(rep->re_freed))
22 abort();
23 rcu_read_unlock();
24 return ret;
@@ -106,7 +106,7 @@ Section~\ref{sec:defer:RCU Usage Summary} provides a summary.
20 struct route_entry *rep;
21
22 rep = container_of(rhp, struct route_entry, rh);
-23 ACCESS_ONCE(rep->re_freed) = 1;
+23 WRITE_ONCE(rep->re_freed, 1);
24 free(rep);
25 }
26
--
2.7.4
^ permalink raw reply related [flat|nested] 7+ messages in thread
* [PATCH 3/5] route_rcu: Remove redundant assignment statement to rep->re_freed
2017-06-14 15:44 [PATCH 0/5] Updating Section Deferred Processing Junchang Wang
2017-06-14 15:45 ` [PATCH 1/5] hazptr: Switch from ACCESS_ONCE() to READ_ONCE()/WRITE_ONCE() Junchang Wang
2017-06-14 15:45 ` [PATCH 2/5] route_rcu: " Junchang Wang
@ 2017-06-14 15:45 ` Junchang Wang
2017-06-14 15:45 ` [PATCH 4/5] seqlocktorture: Switch from ACCESS_ONCE() to READ_ONCE()/WRITE_ONCE() Junchang Wang
` (2 subsequent siblings)
5 siblings, 0 replies; 7+ messages in thread
From: Junchang Wang @ 2017-06-14 15:45 UTC (permalink / raw)
To: perfbook, paulmck; +Cc: Junchang Wang
Signed-off-by: Junchang Wang <junchangwang@gmail.com>
---
CodeSamples/defer/route_rcu.c | 1 -
1 file changed, 1 deletion(-)
diff --git a/CodeSamples/defer/route_rcu.c b/CodeSamples/defer/route_rcu.c
index 7716fe8..1fd69ea 100644
--- a/CodeSamples/defer/route_rcu.c
+++ b/CodeSamples/defer/route_rcu.c
@@ -93,7 +93,6 @@ static void route_cb(struct rcu_head *rhp)
{
struct route_entry *rep = container_of(rhp, struct route_entry, rh);
- WRITE_ONCE(rep->re_freed, 1);
re_free(rep);
}
--
2.7.4
^ permalink raw reply related [flat|nested] 7+ messages in thread
* [PATCH 4/5] seqlocktorture: Switch from ACCESS_ONCE() to READ_ONCE()/WRITE_ONCE()
2017-06-14 15:44 [PATCH 0/5] Updating Section Deferred Processing Junchang Wang
` (2 preceding siblings ...)
2017-06-14 15:45 ` [PATCH 3/5] route_rcu: Remove redundant assignment statement to rep->re_freed Junchang Wang
@ 2017-06-14 15:45 ` Junchang Wang
2017-06-14 15:45 ` [PATCH 5/5] rcu_rcpls: " Junchang Wang
2017-06-14 17:52 ` [PATCH 0/5] Updating Section Deferred Processing Paul E. McKenney
5 siblings, 0 replies; 7+ messages in thread
From: Junchang Wang @ 2017-06-14 15:45 UTC (permalink / raw)
To: perfbook, paulmck; +Cc: Junchang Wang
Signed-off-by: Junchang Wang <junchangwang@gmail.com>
---
CodeSamples/defer/seqlocktorture.c | 8 ++++----
1 file changed, 4 insertions(+), 4 deletions(-)
diff --git a/CodeSamples/defer/seqlocktorture.c b/CodeSamples/defer/seqlocktorture.c
index fa0b69a..a345a5f 100644
--- a/CodeSamples/defer/seqlocktorture.c
+++ b/CodeSamples/defer/seqlocktorture.c
@@ -78,9 +78,9 @@ void *seqlock_read_test(void *arg)
run_on(me);
atomic_inc(&nthreadsrunning);
- while (ACCESS_ONCE(goflag) == GOFLAG_INIT)
+ while (READ_ONCE(goflag) == GOFLAG_INIT)
poll(NULL, 0, 1);
- while (ACCESS_ONCE(goflag) == GOFLAG_RUN) {
+ while (READ_ONCE(goflag) == GOFLAG_RUN) {
for (i = COUNT_READ_RUN; i > 0; i--) {
n_retries_local_cur = -1;
do {
@@ -115,9 +115,9 @@ void *seqlock_write_test(void *arg)
run_on(me);
atomic_inc(&nthreadsrunning);
- while (ACCESS_ONCE(goflag) == GOFLAG_INIT)
+ while (READ_ONCE(goflag) == GOFLAG_INIT)
poll(NULL, 0, 1);
- while (ACCESS_ONCE(goflag) == GOFLAG_RUN) {
+ while (READ_ONCE(goflag) == GOFLAG_RUN) {
for (i = COUNT_UPDATE_RUN; i > 0; i--) {
write_seqlock(&test_seqlock);
for (j = 0; j < n_elems; j++)
--
2.7.4
^ permalink raw reply related [flat|nested] 7+ messages in thread
* [PATCH 5/5] rcu_rcpls: Switch from ACCESS_ONCE() to READ_ONCE()/WRITE_ONCE()
2017-06-14 15:44 [PATCH 0/5] Updating Section Deferred Processing Junchang Wang
` (3 preceding siblings ...)
2017-06-14 15:45 ` [PATCH 4/5] seqlocktorture: Switch from ACCESS_ONCE() to READ_ONCE()/WRITE_ONCE() Junchang Wang
@ 2017-06-14 15:45 ` Junchang Wang
2017-06-14 17:52 ` [PATCH 0/5] Updating Section Deferred Processing Paul E. McKenney
5 siblings, 0 replies; 7+ messages in thread
From: Junchang Wang @ 2017-06-14 15:45 UTC (permalink / raw)
To: perfbook, paulmck; +Cc: Junchang Wang
Signed-off-by: Junchang Wang <junchangwang@gmail.com>
---
CodeSamples/defer/rcu_rcpls.c | 6 +++---
CodeSamples/defer/rcu_rcpls.h | 2 +-
CodeSamples/defer/rcutorture.h | 2 +-
appendix/toyrcu/toyrcu.tex | 14 +++++++-------
4 files changed, 12 insertions(+), 12 deletions(-)
diff --git a/CodeSamples/defer/rcu_rcpls.c b/CodeSamples/defer/rcu_rcpls.c
index 8dbf624..3018ae9 100644
--- a/CodeSamples/defer/rcu_rcpls.c
+++ b/CodeSamples/defer/rcu_rcpls.c
@@ -28,7 +28,7 @@ static void flip_counter_and_wait(int ctr)
int i;
int t;
- ACCESS_ONCE(rcu_idx) = ctr + 1;
+ WRITE_ONCE(rcu_idx, ctr + 1);
i = ctr & 0x1;
smp_mb();
for_each_thread(t) {
@@ -46,10 +46,10 @@ void synchronize_rcu(void)
int oldctr;
smp_mb();
- oldctr = ACCESS_ONCE(rcu_idx);
+ oldctr = READ_ONCE(rcu_idx);
smp_mb();
spin_lock(&rcu_gp_lock);
- ctr = ACCESS_ONCE(rcu_idx);
+ ctr = READ_ONCE(rcu_idx);
if (ctr - oldctr >= 3) {
/*
diff --git a/CodeSamples/defer/rcu_rcpls.h b/CodeSamples/defer/rcu_rcpls.h
index a9c7723..6825270 100644
--- a/CodeSamples/defer/rcu_rcpls.h
+++ b/CodeSamples/defer/rcu_rcpls.h
@@ -47,7 +47,7 @@ static void rcu_read_lock(void)
n = __get_thread_var(rcu_nesting);
if (n == 0) {
- i = ACCESS_ONCE(rcu_idx) & 0x1;
+ i = READ_ONCE(rcu_idx) & 0x1;
__get_thread_var(rcu_read_idx) = i;
__get_thread_var(rcu_refcnt)[i]++;
}
diff --git a/CodeSamples/defer/rcutorture.h b/CodeSamples/defer/rcutorture.h
index c5bc87e..ca03df9 100644
--- a/CodeSamples/defer/rcutorture.h
+++ b/CodeSamples/defer/rcutorture.h
@@ -271,7 +271,7 @@ void *rcu_read_stress_test(void *arg)
n_mberror++;
rcu_read_lock_nest();
for (i = 0; i < 100; i++)
- ACCESS_ONCE(garbage)++;
+ WRITE_ONCE(garbage, READ_ONCE(garbage) + 1);
rcu_read_unlock_nest();
pc = p->pipe_count;
rcu_read_unlock();
diff --git a/appendix/toyrcu/toyrcu.tex b/appendix/toyrcu/toyrcu.tex
index b4882c3..5e74afa 100644
--- a/appendix/toyrcu/toyrcu.tex
+++ b/appendix/toyrcu/toyrcu.tex
@@ -1003,7 +1003,7 @@ concurrent RCU updates.
5
6 n = __get_thread_var(rcu_nesting);
7 if (n == 0) {
- 8 i = ACCESS_ONCE(rcu_idx) & 0x1;
+ 8 i = READ_ONCE(rcu_idx) & 0x1;
9 __get_thread_var(rcu_read_idx) = i;
10 __get_thread_var(rcu_refcnt)[i]++;
11 }
@@ -1044,7 +1044,7 @@ so that line~8 of
Figure~\ref{fig:app:toyrcu:RCU Read-Side Using Per-Thread Reference-Count Pair and Shared Update}
must mask off the low-order bit.
We also switched from using \co{atomic_read()} and \co{atomic_set()}
-to using \co{ACCESS_ONCE()}.
+to using \co{READ_ONCE()}.
The data is also quite similar, as shown in
Figure~\ref{fig:app:toyrcu:RCU Read-Side Using Per-Thread Reference-Count Pair and Shared Update Data},
with \co{rcu_idx} now being a \co{long} instead of an
@@ -1058,7 +1058,7 @@ with \co{rcu_idx} now being a \co{long} instead of an
3 int i;
4 int t;
5
- 6 ACCESS_ONCE(rcu_idx) = ctr + 1;
+ 6 WRITE_ONCE(rcu_idx, ctr + 1);
7 i = ctr & 0x1;
8 smp_mb();
9 for_each_thread(t) {
@@ -1075,10 +1075,10 @@ with \co{rcu_idx} now being a \co{long} instead of an
20 int oldctr;
21
22 smp_mb();
- 23 oldctr = ACCESS_ONCE(rcu_idx);
+ 23 oldctr = READ_ONCE(rcu_idx);
24 smp_mb();
25 spin_lock(&rcu_gp_lock);
- 26 ctr = ACCESS_ONCE(rcu_idx);
+ 26 ctr = READ_ONCE(rcu_idx);
27 if (ctr - oldctr >= 3) {
28 spin_unlock(&rcu_gp_lock);
29 smp_mb();
@@ -1106,7 +1106,7 @@ These are similar to those in
Figure~\ref{fig:app:toyrcu:RCU Update Using Per-Thread Reference-Count Pair}.
The differences in \co{flip_counter_and_wait()} include:
\begin{enumerate}
-\item Line~6 uses \co{ACCESS_ONCE()} instead of \co{atomic_set()},
+\item Line~6 uses \co{WRITE_ONCE()} instead of \co{atomic_set()},
and increments rather than complementing.
\item A new line~7 masks the counter down to its bottom bit.
\end{enumerate}
@@ -1116,7 +1116,7 @@ The changes to \co{synchronize_rcu()} are more pervasive:
\item There is a new \co{oldctr} local variable that captures
the pre-lock-acquisition value of \co{rcu_idx} on
line~23.
-\item Line~26 uses \co{ACCESS_ONCE()} instead of \co{atomic_read()}.
+\item Line~26 uses \co{READ_ONCE()} instead of \co{atomic_read()}.
\item Lines~27-30 check to see if at least three counter flips were
performed by other threads while the lock was being acquired,
and, if so, releases the lock, does a memory barrier, and returns.
--
2.7.4
^ permalink raw reply related [flat|nested] 7+ messages in thread
* Re: [PATCH 0/5] Updating Section Deferred Processing
2017-06-14 15:44 [PATCH 0/5] Updating Section Deferred Processing Junchang Wang
` (4 preceding siblings ...)
2017-06-14 15:45 ` [PATCH 5/5] rcu_rcpls: " Junchang Wang
@ 2017-06-14 17:52 ` Paul E. McKenney
5 siblings, 0 replies; 7+ messages in thread
From: Paul E. McKenney @ 2017-06-14 17:52 UTC (permalink / raw)
To: Junchang Wang; +Cc: perfbook
On Wed, Jun 14, 2017 at 11:44:59PM +0800, Junchang Wang wrote:
> Hi Paul,
>
> Most of the following patches switch from ACCESS_ONCE to READ_ONCE/WRITE_ONCE in
> Section Deferred Processing. Please take a look.
Applied and pushed, thank you!
And good catch on the redundant assignment! ;-)
Thanx, Paul
> Thanks,
> --Jason
>
> Junchang Wang (5):
> hazptr: Switch from ACCESS_ONCE() to READ_ONCE()/WRITE_ONCE()
> route_rcu: Switch from ACCESS_ONCE() to READ_ONCE()/WRITE_ONCE()
> route_rcu: Remove redundant assignment statement to rep->re_freed
> seqlocktorture: Switch from ACCESS_ONCE() to READ_ONCE()/WRITE_ONCE()
> rcu_rcpls: Switch from ACCESS_ONCE() to READ_ONCE()/WRITE_ONCE()
>
> CodeSamples/defer/hazptr.h | 8 ++++----
> CodeSamples/defer/hazptrtorture.h | 6 +++---
> CodeSamples/defer/rcu_rcpls.c | 6 +++---
> CodeSamples/defer/rcu_rcpls.h | 2 +-
> CodeSamples/defer/rcutorture.h | 2 +-
> CodeSamples/defer/route_rcu.c | 5 ++---
> CodeSamples/defer/seqlocktorture.c | 8 ++++----
> appendix/toyrcu/toyrcu.tex | 14 +++++++-------
> defer/rcuusage.tex | 4 ++--
> 9 files changed, 27 insertions(+), 28 deletions(-)
>
> --
> 2.7.4
>
^ permalink raw reply [flat|nested] 7+ messages in thread
end of thread, other threads:[~2017-06-14 17:52 UTC | newest]
Thread overview: 7+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2017-06-14 15:44 [PATCH 0/5] Updating Section Deferred Processing Junchang Wang
2017-06-14 15:45 ` [PATCH 1/5] hazptr: Switch from ACCESS_ONCE() to READ_ONCE()/WRITE_ONCE() Junchang Wang
2017-06-14 15:45 ` [PATCH 2/5] route_rcu: " Junchang Wang
2017-06-14 15:45 ` [PATCH 3/5] route_rcu: Remove redundant assignment statement to rep->re_freed Junchang Wang
2017-06-14 15:45 ` [PATCH 4/5] seqlocktorture: Switch from ACCESS_ONCE() to READ_ONCE()/WRITE_ONCE() Junchang Wang
2017-06-14 15:45 ` [PATCH 5/5] rcu_rcpls: " Junchang Wang
2017-06-14 17:52 ` [PATCH 0/5] Updating Section Deferred Processing Paul E. McKenney
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.