2002-09-30 11:02:21

by Dipankar Sarma

Subject: [PATCH] 2.5.39-mm1 fixes 1/3

Hi Andrew,

rcu_ltimer was used mostly for performance measurements and wasn't
completely preemption friendly. With this fix (against 2.5.39-mm1),
it should be now.

Thanks
--
Dipankar Sarma <[email protected]> http://lse.sourceforge.net
Linux Technology Center, IBM Software Lab, Bangalore, India.


--- kernel/rcupdate.c.orig Mon Sep 30 13:55:15 2002
+++ kernel/rcupdate.c Mon Sep 30 13:55:29 2002
@@ -60,12 +60,13 @@
*/
void call_rcu(struct rcu_head *head, void (*func)(void *arg), void *arg)
{
- int cpu = smp_processor_id();
+ int cpu;
unsigned long flags;

head->func = func;
head->arg = arg;
local_irq_save(flags);
+ cpu = smp_processor_id();
list_add_tail(&head->list, &RCU_nxtlist(cpu));
local_irq_restore(flags);
}

2002-09-30 11:04:39

by Dipankar Sarma

Subject: Re: [PATCH] 2.5.39-mm1 fixes 2/3

The read_barrier_depends patch didn't have compilable list_for_each_rcu()
macros. These macros allow a simpler interface to use RCU by taking
care of memory barriers. Fix against 2.5.39-mm1.

Thanks
--
Dipankar Sarma <[email protected]> http://lse.sourceforge.net
Linux Technology Center, IBM Software Lab, Bangalore, India.


--- include/linux/list.h.orig Mon Sep 30 13:59:10 2002
+++ include/linux/list.h Mon Sep 30 13:59:23 2002
@@ -307,7 +307,7 @@
*/
#define list_for_each_rcu(pos, head) \
for (pos = (head)->next, prefetch(pos->next); pos != (head); \
- pos = pos->next, ({ read_barrier_depends(); 0}), prefetch(pos->next))
+ pos = pos->next, ({ read_barrier_depends(); 0;}), prefetch(pos->next))

/**
* list_for_each_safe_rcu - iterate over an rcu-protected list safe
@@ -318,7 +318,7 @@
*/
#define list_for_each_safe_rcu(pos, n, head) \
for (pos = (head)->next, n = pos->next; pos != (head); \
- pos = n, ({ read_barrier_depends(); 0}), n = pos->next)
+ pos = n, ({ read_barrier_depends(); 0;}), n = pos->next)

#endif /* __KERNEL__ || _LVM_H_INCLUDE */

2002-09-30 11:06:33

by Dipankar Sarma

Subject: Re: [PATCH] 2.5.39-mm1 fixes 3/3

The dcache_rcu patch should use the RCU list macros which make sure
of the barrier requirements. Fix against 2.5.39-mm1.

Thanks
--
Dipankar Sarma <[email protected]> http://lse.sourceforge.net
Linux Technology Center, IBM Software Lab, Bangalore, India.


--- fs/dcache.c.orig Mon Sep 30 14:00:40 2002
+++ fs/dcache.c Mon Sep 30 14:05:33 2002
@@ -869,14 +869,8 @@
struct dentry *found = NULL;

rcu_read_lock();
- tmp = head->next;
- for (;;) {
- struct dentry * dentry;
- read_barrier_depends();
- dentry = list_entry(tmp, struct dentry, d_hash);
- if (tmp == head)
- break;
- tmp = tmp->next;
+ list_for_each_rcu(tmp, head) {
+ struct dentry * dentry = list_entry(tmp, struct dentry, d_hash);
if (dentry->d_name.hash != hash)
continue;
if (dentry->d_parent != parent)
@@ -999,7 +993,7 @@
struct list_head *list = d_hash(entry->d_parent, entry->d_name.hash);
spin_lock(&dcache_lock);
if (!list_empty(&entry->d_hash) && !d_unhashed(entry)) BUG();
- list_add(&entry->d_hash, list);
+ list_add_rcu(&entry->d_hash, list);
entry->d_vfs_flags &= ~DCACHE_UNHASHED;
spin_unlock(&dcache_lock);
}