* Mathieu Desnoyers ([email protected]) wrote:
> * Paul E. McKenney ([email protected]) wrote:
> > On Wed, Sep 28, 2011 at 04:34:33PM +0800, Lai Jiangshan wrote:
> > > Signed-off-by: Lai Jiangshan <[email protected]>
> > 
> > This one I don't understand.  It looks to me that all dereferences
> > of the per_cpu_call_rcu_data pointer are done while holding the
> > call_rcu_mutex, so there should be no need for the memory barrier.
> > 
> > Yes, there is an access to the pointer without lock protection
> > in create_all_cpu_call_rcu_data(), but that access does not
> > do a dereference, so still no need for the memory barrier.
> > 
> > So, what am I missing?
> 
> A concurrent get_cpu_call_rcu_data(), called by get_call_rcu_data(),
> could dereference this pointer without holding any mutex. So this
> situation would happen if we have a concurrent call_rcu() executing
> while we do the create_all_cpu_call_rcu_data().
> 
> I think we would need to put a rcu_dereference() around
> per_cpu_call_rcu_data read within get_cpu_call_rcu_data() too.
> per_cpu_call_rcu_data should be done with rcu_set_pointer.
> 
> Also, a rcu read-side critical section would be required around any
> usage of per_cpu_call_rcu_data, and the action of tearing down the
> per-cpu data would require to wait for a quiescent state. So we would
> basically require that the call_rcu users need to be registered as
> RCU reader threads.
> 
> Or we clearly state that no call_rcu should happen while
> create_all_cpu_call_rcu_data is executed. But given we allow call_rcu to
> be called within call_rcu callbacks, this seems to be hard to guarantee.
> 
> Thoughts ?

Implemented in this commit:

commit 618b25958fec4d76310f0d9c59e42128e73a8719
Author: Mathieu Desnoyers <[email protected]>
Date:   Thu Sep 29 17:13:48 2011 -0400

    urcu call_rcu: Use RCU read-side protection for per-cpu call_rcu data
    
    A concurrent get_cpu_call_rcu_data(), called by get_call_rcu_data(),
    could dereference this pointer without holding any mutex. So this
    situation would happen if we have a concurrent call_rcu() executing
    while we do the create_all_cpu_call_rcu_data().
    
    I think we would need to put a rcu_dereference() around
    per_cpu_call_rcu_data read within get_cpu_call_rcu_data() too.
    per_cpu_call_rcu_data should be done with rcu_set_pointer.
    
    Also, a rcu read-side critical section would be required around any
    usage of per_cpu_call_rcu_data, and the action of tearing down the
    per-cpu data would require to wait for a quiescent state. So we would
    basically require that the call_rcu users need to be registered as
    RCU reader threads.
    
    Signed-off-by: Mathieu Desnoyers <[email protected]>

diff --git a/urcu-bp.c b/urcu-bp.c
index 2ae3408..4c0ab54 100644
--- a/urcu-bp.c
+++ b/urcu-bp.c
@@ -39,6 +39,7 @@
 #include "urcu/wfqueue.h"
 #include "urcu/map/urcu-bp.h"
 #include "urcu/static/urcu-bp.h"
+#include "urcu-pointer.h"
 
 /* Do not #define _LGPL_SOURCE to ensure we can emit the wrapper symbols */
 #undef _LGPL_SOURCE
diff --git a/urcu-call-rcu-impl.h b/urcu-call-rcu-impl.h
index 0a47d96..d09adb1 100644
--- a/urcu-call-rcu-impl.h
+++ b/urcu-call-rcu-impl.h
@@ -82,7 +82,10 @@ static struct call_rcu_data *default_call_rcu_data;
 
 /*
  * Pointer to array of pointers to per-CPU call_rcu_data structures
- * and # CPUs.
+ * and # CPUs. per_cpu_call_rcu_data is a RCU-protected pointer to an
+ * array of RCU-protected pointers to call_rcu_data. call_rcu acts as a
+ * RCU read-side and reads per_cpu_call_rcu_data and the per-cpu pointer
+ * without mutex. The call_rcu_mutex protects updates.
  */
 
 static struct call_rcu_data **per_cpu_call_rcu_data;
@@ -109,7 +112,7 @@ static void alloc_cpu_call_rcu_data(void)
        p = malloc(maxcpus * sizeof(*per_cpu_call_rcu_data));
        if (p != NULL) {
                memset(p, '\0', maxcpus * sizeof(*per_cpu_call_rcu_data));
-               per_cpu_call_rcu_data = p;
+               rcu_set_pointer(&per_cpu_call_rcu_data, p);
        } else {
                if (!warned) {
                        fprintf(stderr, "[error] liburcu: unable to allocate 
per-CPU pointer array\n");
@@ -330,13 +333,18 @@ static void call_rcu_data_init(struct call_rcu_data 
**crdpp,
  * CPU, returning NULL if there is none.  We cannot automatically
  * created it because the platform we are running on might not define
  * sched_getcpu().
+ *
+ * The call to this function and use of the returned call_rcu_data
+ * should be protected by RCU read-side lock.
  */
 
 struct call_rcu_data *get_cpu_call_rcu_data(int cpu)
 {
        static int warned = 0;
+       struct call_rcu_data **pcpu_crdp;
 
-       if (per_cpu_call_rcu_data == NULL)
+       pcpu_crdp = rcu_dereference(per_cpu_call_rcu_data);
+       if (pcpu_crdp == NULL)
                return NULL;
        if (!warned && maxcpus > 0 && (cpu < 0 || maxcpus <= cpu)) {
                fprintf(stderr, "[error] liburcu: get CPU # out of range\n");
@@ -344,7 +352,7 @@ struct call_rcu_data *get_cpu_call_rcu_data(int cpu)
        }
        if (cpu < 0 || maxcpus <= cpu)
                return NULL;
-       return per_cpu_call_rcu_data[cpu];
+       return rcu_dereference(pcpu_crdp[cpu]);
 }
 
 /*
@@ -418,7 +426,7 @@ int set_cpu_call_rcu_data(int cpu, struct call_rcu_data 
*crdp)
                return -EEXIST;
        }
 
-       per_cpu_call_rcu_data[cpu] = crdp;
+       rcu_set_pointer(&per_cpu_call_rcu_data[cpu], crdp);
        call_rcu_unlock(&call_rcu_mutex);
        return 0;
 }
@@ -450,6 +458,9 @@ struct call_rcu_data *get_default_call_rcu_data(void)
  * structure assigned to the CPU on which the thread is running,
  * followed by the default call_rcu_data structure.  If there is not
  * yet a default call_rcu_data structure, one will be created.
+ *
+ * Calls to this function and use of the returned call_rcu_data should
+ * be protected by RCU read-side lock.
  */
 struct call_rcu_data *get_call_rcu_data(void)
 {
@@ -564,6 +575,8 @@ static void wake_call_rcu_thread(struct call_rcu_data *crdp)
  * need the first invocation of call_rcu() to be fast, make sure
  * to create a call_rcu thread first.  One way to accomplish this is
  * "get_call_rcu_data();", and another is create_all_cpu_call_rcu_data().
+ *
+ * call_rcu must be called by registered RCU read-side threads.
  */
 
 void call_rcu(struct rcu_head *head,
@@ -573,10 +586,13 @@ void call_rcu(struct rcu_head *head,
 
        cds_wfq_node_init(&head->next);
        head->func = func;
+       /* Holding rcu read-side lock across use of per-cpu crdp */
+       rcu_read_lock();
        crdp = get_call_rcu_data();
        cds_wfq_enqueue(&crdp->cbs, &head->next);
        uatomic_inc(&crdp->qlen);
        wake_call_rcu_thread(crdp);
+       rcu_read_unlock();
 }
 
 /*
@@ -641,17 +657,37 @@ void call_rcu_data_free(struct call_rcu_data *crdp)
 void free_all_cpu_call_rcu_data(void)
 {
        int cpu;
-       struct call_rcu_data *crdp;
+       struct call_rcu_data **crdp;
+       static int warned = 0;
 
        if (maxcpus <= 0)
                return;
+
+       crdp = malloc(sizeof(*crdp) * maxcpus);
+       if (!crdp) {
+               if (!warned) {
+                       fprintf(stderr, "[error] liburcu: unable to allocate 
per-CPU pointer array\n");
+               }
+               warned = 1;
+       }
+
        for (cpu = 0; cpu < maxcpus; cpu++) {
-               crdp = get_cpu_call_rcu_data(cpu);
-               if (crdp == NULL)
+               crdp[cpu] = get_cpu_call_rcu_data(cpu);
+               if (crdp[cpu] == NULL)
                        continue;
                set_cpu_call_rcu_data(cpu, NULL);
-               call_rcu_data_free(crdp);
        }
+       /*
+        * Wait for call_rcu sites acting as RCU readers of the
+        * call_rcu_data to become quiescent.
+        */
+       synchronize_rcu();
+       for (cpu = 0; cpu < maxcpus; cpu++) {
+               if (crdp[cpu] == NULL)
+                       continue;
+               call_rcu_data_free(crdp[cpu]);
+       }
+       free(crdp);
 }
 
 /*
@@ -700,7 +736,7 @@ void call_rcu_after_fork_child(void)
        /* Cleanup call_rcu_data pointers before use */
        maxcpus_reset();
        free(per_cpu_call_rcu_data);
-       per_cpu_call_rcu_data = NULL;
+       rcu_set_pointer(&per_cpu_call_rcu_data, NULL);
        thread_call_rcu_data = NULL;
 
        /* Dispose of all of the rest of the call_rcu_data structures. */
diff --git a/urcu-call-rcu.h b/urcu-call-rcu.h
index e76a844..5ea0c23 100644
--- a/urcu-call-rcu.h
+++ b/urcu-call-rcu.h
@@ -62,16 +62,28 @@ struct rcu_head {
 /*
  * Exported functions
  */
+
+/*
+ * get_cpu_call_rcu_data should be called with RCU read-side lock held.
+ * Callers should be registered RCU read-side threads.
+ */
 struct call_rcu_data *get_cpu_call_rcu_data(int cpu);
 pthread_t get_call_rcu_thread(struct call_rcu_data *crdp);
 struct call_rcu_data *create_call_rcu_data(unsigned long flags,
                                           int cpu_affinity);
 int set_cpu_call_rcu_data(int cpu, struct call_rcu_data *crdp);
 struct call_rcu_data *get_default_call_rcu_data(void);
+/*
+ * get_call_rcu_data should be called from registered RCU read-side
+ * threads.
+ */
 struct call_rcu_data *get_call_rcu_data(void);
 struct call_rcu_data *get_thread_call_rcu_data(void);
 void set_thread_call_rcu_data(struct call_rcu_data *crdp);
 int create_all_cpu_call_rcu_data(unsigned long flags);
+/*
+ * call_rcu should be called from registered RCU read-side threads.
+ */
 void call_rcu(struct rcu_head *head,
              void (*func)(struct rcu_head *head));
 void call_rcu_data_free(struct call_rcu_data *crdp);
diff --git a/urcu-qsbr.c b/urcu-qsbr.c
index 1adaa94..a59a87a 100644
--- a/urcu-qsbr.c
+++ b/urcu-qsbr.c
@@ -39,6 +39,7 @@
 #include "urcu/map/urcu-qsbr.h"
 #define BUILD_QSBR_LIB
 #include "urcu/static/urcu-qsbr.h"
+#include "urcu-pointer.h"
 
 /* Do not #define _LGPL_SOURCE to ensure we can emit the wrapper symbols */
 #undef _LGPL_SOURCE
diff --git a/urcu.c b/urcu.c
index 20bbf36..77f6888 100644
--- a/urcu.c
+++ b/urcu.c
@@ -39,6 +39,7 @@
 #include "urcu/wfqueue.h"
 #include "urcu/map/urcu.h"
 #include "urcu/static/urcu.h"
+#include "urcu-pointer.h"
 
 /* Do not #define _LGPL_SOURCE to ensure we can emit the wrapper symbols */
 #undef _LGPL_SOURCE

> 
> Mathieu
> 
> > 
> >                                                     Thanx, Paul
> > 
> > > ---
> > >  urcu-call-rcu-impl.h |    4 ++++
> > >  1 files changed, 4 insertions(+), 0 deletions(-)
> > > 
> > > diff --git a/urcu-call-rcu-impl.h b/urcu-call-rcu-impl.h
> > > index 3c68ae7..462139a 100644
> > > --- a/urcu-call-rcu-impl.h
> > > +++ b/urcu-call-rcu-impl.h
> > > @@ -104,6 +104,10 @@ static void alloc_cpu_call_rcu_data(void)
> > >   p = malloc(maxcpus * sizeof(*per_cpu_call_rcu_data));
> > >   if (p != NULL) {
> > >           memset(p, '\0', maxcpus * sizeof(*per_cpu_call_rcu_data));
> > > +
> > > +         /* Array initialized before pointer is planted. */
> > > +         cmm_smp_mb();
> > > +
> > >           per_cpu_call_rcu_data = p;
> > >   } else {
> > >           if (!warned) {
> > > -- 
> > > 1.7.4.4
> > > 
> > 
> 
> -- 
> Mathieu Desnoyers
> Operating System Efficiency R&D Consultant
> EfficiOS Inc.
> http://www.efficios.com

-- 
Mathieu Desnoyers
Operating System Efficiency R&D Consultant
EfficiOS Inc.
http://www.efficios.com

_______________________________________________
ltt-dev mailing list
[email protected]
http://lists.casi.polymtl.ca/cgi-bin/mailman/listinfo/ltt-dev

Reply via email to