Make rt_cache_stat use kmalloc_percpu

Hi,
Here's another patch to use kmalloc_percpu. As usual, this removes NR_CPUS bloat, can work when modulized and helps node local allocation.

Thanks,
Kiran

D: Name: rtstats-2.5.52-1.patch
D: Author: Ravikiran Thirumalai
D: Description: Make rt_cache_stat use kmalloc_percpu

 include/net/route.h |    6 +++-
 net/ipv4/route.c    |   71 +++++++++++++++++++++++++++++-----------------------
 2 files changed, 45 insertions(+), 32 deletions(-)

diff -ruN -X dontdiff linux-2.5.52/include/net/route.h rtstats-2.5.52/include/net/route.h
@@ -102,7 +102,11 @@
         unsigned int gc_ignored;
         unsigned int gc_goal_miss;
         unsigned int gc_dst_overflow;
-} ____cacheline_aligned_in_smp;
+};
+
+extern struct rt_cache_stat *rt_cache_stat;
+#define RT_CACHE_STAT_INC(field)					  \
+		(per_cpu_ptr(rt_cache_stat, smp_processor_id())->field++)
 
 extern struct ip_rt_acct *ip_rt_acct;
 

From: Ravikiran G Thirumalai <[email protected]>
Date: Mon, 16 Dec 2002 19:22:12 +0530

Here's another patch to use kmalloc_percpu. As usual, this removes NR_CPUS bloat, can work when modulized and helps node local allocation.

I can't consider this seriously until the kmalloc_percpu stuff actually makes it into Linus's tree. Last I checked, Andrew had a lot of legitimate gripes with the ideas.

Kiran's latest (vastly simpler) version looks fine to my eye.

From: Andrew Morton <[email protected]>
Date: Fri, 20 Dec 2002 23:31:56 -0800

Kiran's latest (vastly simpler) version looks fine to my eye.

Ok, so once that is in, he can feel free to resubmit the rt_cache_state patch.

Ok, I'll do that

Thanks,
Kiran