Make RCU lists use the RCU API

This patch moves rcu-protected lists from list.h into a new header file rculist.h.

Quoted 2 lines · On Thu, 17 Jan 2008 21:47:38 +0100 Franck Bui-Huu <[email protected]> wrote:

This patch moves rcu-protected lists from list.h into a new header file rculist.h.

I'm getting way too many compilation errors from this, perhaps because of new rcu-list usages which weren't present in the old tree (ie: current Linus mainline) which you tested it against.

The most practical way to prepare a patch like this is to either make it back-compatible (for a while at least) or to prepare and carefully test it against latest -mm. That way you'll pick up everyone's new code.

Quoted 10 lines · Andrew Morton

Quoted 2 lines · On Thu, 17 Jan 2008 21:47:38 +0100 Franck Bui-Huu <[email protected]> wrote:

This patch moves rcu-protected lists from list.h into a new header file rculist.h.

I'm getting way too many compilation errors from this, perhaps because of new rcu-list usages which weren't present in the old tree (ie: current Linus mainline) which you tested it against.

Not really suprising, sorry for the mess.

Quoted 4 lines

The most practical way to prepare a patch like this is to either make it back-compatible (for a while at least) or to prepare and carefully test it against latest -mm. That way you'll pick up everyone's new code.

I'll try to come up with something better: if rculist helpers are used from list.h then issue a warning otherwise every thing is fine since the helpers are imported from rculist.h.

Do you think it's better ?

In the meanwhile, could you drop the patches related to rculist from mm tree ?

Quoted 1 line · On Sat, 02 Feb 2008 14:32:41 +0100 Franck Bui-Huu <[email protected]> wrote:

Do you think it's better ?

Could. I'd suggest that you redo the header-file split patch around the 2.6.25-rc1 timeframe, test it carefully then let's get it in then.

Quoted 6 lines · Andrew Morton

Quoted 1 line · On Sat, 02 Feb 2008 14:32:41 +0100 Franck Bui-Huu <[email protected]> wrote:

Do you think it's better ?

Could. I'd suggest that you redo the header-file split patch around the 2.6.25-rc1 timeframe, test it carefully then let's get it in then.

Does the mm tree also have a calm down period during release candidates ?

I have modified the patchset so now if rcu helpers are used from rculist.h then fine otherwise gcc warns you that you're using the helpers from list.h like this:

init/foo.c:13: warning: ‘__deprecated_list_add_rcu’ is deprecated (declared at include/linux/rculist.h:76)

But the build process doesn't fail anymore.

For that I added some ugly hacks in list.h and rculist.h but they definitively should be removed for mainline inclusion. I'm sending them in response to this email.

If we include this now, then people can have a chance to notice that rculist.h exists and fix their stuffs until 2.6.25 release candidates but I'll redo the patchset and give it some test around the 2.6.25-rc1 timeframe anyway.

Quoted 7 lines · On Sun, 03 Feb 2008 09:45:25 +0100 Franck Bui-Huu <[email protected]> wrote:

Quoted 5 lines · Andrew Morton

Quoted 1 line · On Sat, 02 Feb 2008 14:32:41 +0100 Franck Bui-Huu <[email protected]> wrote:

Do you think it's better ?

Could. I'd suggest that you redo the header-file split patch around the 2.6.25-rc1 timeframe, test it carefully then let's get it in then.

Does the mm tree also have a calm down period during release candidates ?

Yes, I try to not merge too much material late in the -rcs and in the merge window. Often it's not practical to merge things anyway, because people prepare patches against mainline, which is ancient history...

Quoted 11 lines

I have modified the patchset so now if rcu helpers are used from rculist.h then fine otherwise gcc warns you that you're using the helpers from list.h like this:

init/foo.c:13: warning: ‘__deprecated_list_add_rcu’ is deprecated (declared at include/linux/rculist.h:76)

But the build process doesn't fail anymore.

For that I added some ugly hacks in list.h and rculist.h but they definitively should be removed for mainline inclusion. I'm sending them in response to this email.

I wouldn't bother, really. Let's just get it as good as we can and slam it in.

Quoted 4 lines

If we include this now, then people can have a chance to notice that rculist.h exists and fix their stuffs until 2.6.25 release candidates but I'll redo the patchset and give it some test around the 2.6.25-rc1 timeframe anyway.

I'm just about finished compilation-testing for 2.6.24-mm1 which is a shame. Please do a best-effort against 2.6.24-mm1 (hopefully I'll get that out tomorrow) and I'll slip it into -rc1 if it gets through cross-build testing without unfixable-with-my-patience-level problems.

From: Franck Bui-Huu <[email protected]>

This patch makes almost all list mutation primitives use rcu_assign_pointer().

The main point of this being readability improvement.

Signed-off-by Franck Bui-Huu <[email protected]>

---
 include/linux/rculist.h |   23 +++++++++--------------
 1 files changed, 9 insertions(+), 14 deletions(-)

include/linux/rculist.h +9 -14

index 8056d38..038b64f 100644
@@ -7,6 +7,7 @@
  * RCU-protected list version
  */
 #include <linux/list.h>
+#include <linux/rcupdate.h>
 
 /*
  * Insert a new entry between two known consecutive entries.

@@ -19,9 +20,8 @@ static inline void __list_add_rcu(struct list_head * new,
 {
 	new->next = next;
 	new->prev = prev;
-	smp_wmb();
+	rcu_assign_pointer(prev->next, new);
 	next->prev = new;
-	prev->next = new;
 }
 
 /**

@@ -110,9 +110,8 @@ static inline void list_replace_rcu(struct list_head *old,
 {
 	new->next = old->next;
 	new->prev = old->prev;
-	smp_wmb();
+	rcu_assign_pointer(new->prev->next, new);
 	new->next->prev = new;
-	new->prev->next = new;
 	old->prev = LIST_POISON2;
 }
 

@@ -166,8 +165,7 @@ static inline void list_splice_init_rcu(struct list_head *list,
 	 */
 
 	last->next = at;
-	smp_wmb();
-	head->next = first;
+	rcu_assign_pointer(head->next, first);
 	first->prev = head;
 	at->prev = last;
 }

@@ -280,10 +278,9 @@ static inline void hlist_replace_rcu(struct hlist_node *old,
 
 	new->next = next;
 	new->pprev = old->pprev;
-	smp_wmb();
+	rcu_assign_pointer(*new->pprev, new);
 	if (next)
 		new->next->pprev = &new->next;
-	*new->pprev = new;
 	old->pprev = LIST_POISON2;
 }
 

@@ -310,12 +307,12 @@ static inline void hlist_add_head_rcu(struct hlist_node *n,
 					struct hlist_head *h)
 {
 	struct hlist_node *first = h->first;
+
 	n->next = first;
 	n->pprev = &h->first;
-	smp_wmb();
+	rcu_assign_pointer(h->first, n);
 	if (first)
 		first->pprev = &n->next;
-	h->first = n;
 }
 
 /**

@@ -341,9 +338,8 @@ static inline void hlist_add_before_rcu(struct hlist_node *n,
 {
 	n->pprev = next->pprev;
 	n->next = next;
-	smp_wmb();
+	rcu_assign_pointer(*(n->pprev), n);
 	next->pprev = &n->next;
-	*(n->pprev) = n;
 }
 
 /**

@@ -369,8 +365,7 @@ static inline void hlist_add_after_rcu(struct hlist_node *prev,
 {
 	n->next = prev->next;
 	n->pprev = &prev->next;
-	smp_wmb();
-	prev->next = n;
+	rcu_assign_pointer(prev->next, n);
 	if (n->next)
 		n->next->pprev = &n->next;
 }

Signature · 1 line

1.5.3.7