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
--- linux-2.5.52/include/net/route.h Mon Dec 16 07:37:42 2002
+++ rtstats-2.5.52/include/net/route.h Mon Dec 16 14:56:01 2002
@@ -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;
diff -ruN -X dontdiff linux-2.5.52/net/ipv4/route.c rtstats-2.5.52/net/ipv4/route.c
--- linux-2.5.52/net/ipv4/route.c Mon Dec 16 07:38:23 2002
+++ rtstats-2.5.52/net/ipv4/route.c Mon Dec 16 16:14:09 2002
@@ -195,7 +195,7 @@
static unsigned rt_hash_mask;
static int rt_hash_log;
-struct rt_cache_stat rt_cache_stat[NR_CPUS];
+struct rt_cache_stat *rt_cache_stat;
static int rt_intern_hash(unsigned hash, struct rtable *rth,
struct rtable **res);
@@ -319,24 +319,26 @@
int len = 0;
for (i = 0; i < NR_CPUS; i++) {
+ if (!cpu_possible(i))
+ continue;
len += sprintf(buffer+len, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
dst_entries,
- rt_cache_stat[i].in_hit,
- rt_cache_stat[i].in_slow_tot,
- rt_cache_stat[i].in_slow_mc,
- rt_cache_stat[i].in_no_route,
- rt_cache_stat[i].in_brd,
- rt_cache_stat[i].in_martian_dst,
- rt_cache_stat[i].in_martian_src,
-
- rt_cache_stat[i].out_hit,
- rt_cache_stat[i].out_slow_tot,
- rt_cache_stat[i].out_slow_mc,
-
- rt_cache_stat[i].gc_total,
- rt_cache_stat[i].gc_ignored,
- rt_cache_stat[i].gc_goal_miss,
- rt_cache_stat[i].gc_dst_overflow
+ per_cpu_ptr(rt_cache_stat, i)->in_hit,
+ per_cpu_ptr(rt_cache_stat, i)->in_slow_tot,
+ per_cpu_ptr(rt_cache_stat, i)->in_slow_mc,
+ per_cpu_ptr(rt_cache_stat, i)->in_no_route,
+ per_cpu_ptr(rt_cache_stat, i)->in_brd,
+ per_cpu_ptr(rt_cache_stat, i)->in_martian_dst,
+ per_cpu_ptr(rt_cache_stat, i)->in_martian_src,
+
+ per_cpu_ptr(rt_cache_stat, i)->out_hit,
+ per_cpu_ptr(rt_cache_stat, i)->out_slow_tot,
+ per_cpu_ptr(rt_cache_stat, i)->out_slow_mc,
+
+ per_cpu_ptr(rt_cache_stat, i)->gc_total,
+ per_cpu_ptr(rt_cache_stat, i)->gc_ignored,
+ per_cpu_ptr(rt_cache_stat, i)->gc_goal_miss,
+ per_cpu_ptr(rt_cache_stat, i)->gc_dst_overflow
);
}
@@ -592,11 +594,11 @@
* do not make it too frequently.
*/
- rt_cache_stat[smp_processor_id()].gc_total++;
+ RT_CACHE_STAT_INC(gc_total);
if (now - last_gc < ip_rt_gc_min_interval &&
atomic_read(&ipv4_dst_ops.entries) < ip_rt_max_size) {
- rt_cache_stat[smp_processor_id()].gc_ignored++;
+ RT_CACHE_STAT_INC(gc_ignored);
goto out;
}
@@ -664,7 +666,7 @@
We will not spin here for long time in any case.
*/
- rt_cache_stat[smp_processor_id()].gc_goal_miss++;
+ RT_CACHE_STAT_INC(gc_goal_miss);
if (expire == 0)
break;
@@ -683,7 +685,7 @@
goto out;
if (net_ratelimit())
printk(KERN_WARNING "dst cache overflow\n");
- rt_cache_stat[smp_processor_id()].gc_dst_overflow++;
+ RT_CACHE_STAT_INC(gc_dst_overflow);
return 1;
work_done:
@@ -1387,7 +1389,7 @@
if (!LOCAL_MCAST(daddr) && IN_DEV_MFORWARD(in_dev))
rth->u.dst.input = ip_mr_input;
#endif
- rt_cache_stat[smp_processor_id()].in_slow_mc++;
+ RT_CACHE_STAT_INC(in_slow_mc);
in_dev_put(in_dev);
hash = rt_hash_code(daddr, saddr ^ (dev->ifindex << 5), tos);
@@ -1472,7 +1474,7 @@
}
free_res = 1;
- rt_cache_stat[smp_processor_id()].in_slow_tot++;
+ RT_CACHE_STAT_INC(in_slow_tot);
#ifdef CONFIG_IP_ROUTE_NAT
/* Policy is applied before mapping destination,
@@ -1629,7 +1631,7 @@
}
flags |= RTCF_BROADCAST;
res.type = RTN_BROADCAST;
- rt_cache_stat[smp_processor_id()].in_brd++;
+ RT_CACHE_STAT_INC(in_brd);
local_input:
rth = dst_alloc(&ipv4_dst_ops);
@@ -1674,7 +1676,7 @@
goto intern;
no_route:
- rt_cache_stat[smp_processor_id()].in_no_route++;
+ RT_CACHE_STAT_INC(in_no_route);
spec_dst = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
res.type = RTN_UNREACHABLE;
goto local_input;
@@ -1683,7 +1685,7 @@
* Do not cache martian addresses: they should be logged (RFC1812)
*/
martian_destination:
- rt_cache_stat[smp_processor_id()].in_martian_dst++;
+ RT_CACHE_STAT_INC(in_martian_dst);
#ifdef CONFIG_IP_ROUTE_VERBOSE
if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
printk(KERN_WARNING "martian destination %u.%u.%u.%u from "
@@ -1700,7 +1702,7 @@
martian_source:
- rt_cache_stat[smp_processor_id()].in_martian_src++;
+ RT_CACHE_STAT_INC(in_martian_src);
#ifdef CONFIG_IP_ROUTE_VERBOSE
if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
/*
@@ -1749,7 +1751,7 @@
rth->u.dst.lastuse = jiffies;
dst_hold(&rth->u.dst);
rth->u.dst.__use++;
- rt_cache_stat[smp_processor_id()].in_hit++;
+ RT_CACHE_STAT_INC(in_hit);
read_unlock(&rt_hash_table[hash].lock);
skb->dst = (struct dst_entry*)rth;
return 0;
@@ -2046,7 +2048,7 @@
rth->u.dst.output=ip_output;
- rt_cache_stat[smp_processor_id()].out_slow_tot++;
+ RT_CACHE_STAT_INC(out_slow_tot);
if (flags & RTCF_LOCAL) {
rth->u.dst.input = ip_local_deliver;
@@ -2056,7 +2058,7 @@
rth->rt_spec_dst = fl.fl4_src;
if (flags & RTCF_LOCAL && !(dev_out->flags & IFF_LOOPBACK)) {
rth->u.dst.output = ip_mc_output;
- rt_cache_stat[smp_processor_id()].out_slow_mc++;
+ RT_CACHE_STAT_INC(out_slow_mc);
}
#ifdef CONFIG_IP_MROUTE
if (res.type == RTN_MULTICAST) {
@@ -2114,7 +2116,7 @@
rth->u.dst.lastuse = jiffies;
dst_hold(&rth->u.dst);
rth->u.dst.__use++;
- rt_cache_stat[smp_processor_id()].out_hit++;
+ RT_CACHE_STAT_INC(out_hit);
read_unlock_bh(&rt_hash_table[hash].lock);
*rp = rth;
return 0;
@@ -2634,6 +2636,11 @@
ipv4_dst_ops.gc_thresh = (rt_hash_mask + 1);
ip_rt_max_size = (rt_hash_mask + 1) * 16;
+ rt_cache_stat = kmalloc_percpu(sizeof (struct rt_cache_stat),
+ GFP_KERNEL);
+ if (!rt_cache_stat)
+ goto out_enomem1;
+
devinet_init();
ip_fib_init();
@@ -2659,6 +2666,8 @@
out:
return rc;
out_enomem:
+ kfree_percpu(rt_cache_stat);
+out_enomem1:
rc = -ENOMEM;
goto out;
}
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.
"David S. Miller" wrote:
>
> 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
"David S. Miller" wrote:
> 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.
Ok, so once that is in, he can feel free to resubmit the
rt_cache_state patch.
On Fri, Dec 20, 2002 at 11:29:09PM -0800, David S. Miller wrote:
>...
> 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