* [PATCH v2 1/1] net: openvswitch: reduce cpu_used_mask memory
@ 2023-02-01 13:24 taoyuan_eddy
2023-02-01 13:44 ` Jiri Pirko
0 siblings, 1 reply; 4+ messages in thread
From: taoyuan_eddy @ 2023-02-01 13:24 UTC (permalink / raw)
To: netdev
Cc: Eddy Tao, Pravin B Shelar, David S. Miller, Eric Dumazet,
Jakub Kicinski, Paolo Abeni, dev, linux-kernel
From: Eddy Tao <taoyuan_eddy@hotmail.com>
'struct cpumask cpu_used_mask' is embedded in struct sw_flow.
However, its size is hardcoded to CONFIG_NR_CPUS bits, which can be
8192 by default, it costs memory and slows down ovs_flow_alloc.
This fix uses actual CPU number instead
Signed-off-by: Eddy Tao <taoyuan_eddy@hotmail.com>
---
net/openvswitch/flow.c | 6 +++---
net/openvswitch/flow.h | 2 +-
net/openvswitch/flow_table.c | 24 +++++++++++++++++++++---
3 files changed, 25 insertions(+), 7 deletions(-)
diff --git a/net/openvswitch/flow.c b/net/openvswitch/flow.c
index e20d1a973417..06345cd8c777 100644
--- a/net/openvswitch/flow.c
+++ b/net/openvswitch/flow.c
@@ -107,7 +107,7 @@ void ovs_flow_stats_update(struct sw_flow *flow, __be16 tcp_flags,
rcu_assign_pointer(flow->stats[cpu],
new_stats);
- cpumask_set_cpu(cpu, &flow->cpu_used_mask);
+ cpumask_set_cpu(cpu, flow->cpu_used_mask);
goto unlock;
}
}
@@ -135,7 +135,7 @@ void ovs_flow_stats_get(const struct sw_flow *flow,
memset(ovs_stats, 0, sizeof(*ovs_stats));
/* We open code this to make sure cpu 0 is always considered */
- for (cpu = 0; cpu < nr_cpu_ids; cpu = cpumask_next(cpu, &flow->cpu_used_mask)) {
+ for (cpu = 0; cpu < nr_cpu_ids; cpu = cpumask_next(cpu, flow->cpu_used_mask)) {
struct sw_flow_stats *stats = rcu_dereference_ovsl(flow->stats[cpu]);
if (stats) {
@@ -159,7 +159,7 @@ void ovs_flow_stats_clear(struct sw_flow *flow)
int cpu;
/* We open code this to make sure cpu 0 is always considered */
- for (cpu = 0; cpu < nr_cpu_ids; cpu = cpumask_next(cpu, &flow->cpu_used_mask)) {
+ for (cpu = 0; cpu < nr_cpu_ids; cpu = cpumask_next(cpu, flow->cpu_used_mask)) {
struct sw_flow_stats *stats = ovsl_dereference(flow->stats[cpu]);
if (stats) {
diff --git a/net/openvswitch/flow.h b/net/openvswitch/flow.h
index 073ab73ffeaa..b5711aff6e76 100644
--- a/net/openvswitch/flow.h
+++ b/net/openvswitch/flow.h
@@ -229,7 +229,7 @@ struct sw_flow {
*/
struct sw_flow_key key;
struct sw_flow_id id;
- struct cpumask cpu_used_mask;
+ struct cpumask *cpu_used_mask;
struct sw_flow_mask *mask;
struct sw_flow_actions __rcu *sf_acts;
struct sw_flow_stats __rcu *stats[]; /* One for each CPU. First one
diff --git a/net/openvswitch/flow_table.c b/net/openvswitch/flow_table.c
index 0a0e4c283f02..63c95c9a814d 100644
--- a/net/openvswitch/flow_table.c
+++ b/net/openvswitch/flow_table.c
@@ -87,11 +87,12 @@ struct sw_flow *ovs_flow_alloc(void)
if (!stats)
goto err;
+ flow->cpu_used_mask = (struct cpumask *)&flow->stats[nr_cpu_ids];
spin_lock_init(&stats->lock);
RCU_INIT_POINTER(flow->stats[0], stats);
- cpumask_set_cpu(0, &flow->cpu_used_mask);
+ cpumask_set_cpu(0, flow->cpu_used_mask);
return flow;
err:
@@ -115,7 +116,7 @@ static void flow_free(struct sw_flow *flow)
flow->sf_acts);
/* We open code this to make sure cpu 0 is always considered */
for (cpu = 0; cpu < nr_cpu_ids;
- cpu = cpumask_next(cpu, &flow->cpu_used_mask)) {
+ cpu = cpumask_next(cpu, flow->cpu_used_mask)) {
if (flow->stats[cpu])
kmem_cache_free(flow_stats_cache,
(struct sw_flow_stats __force *)flow->stats[cpu]);
@@ -1194,9 +1195,26 @@ int ovs_flow_init(void)
BUILD_BUG_ON(__alignof__(struct sw_flow_key) % __alignof__(long));
BUILD_BUG_ON(sizeof(struct sw_flow_key) % sizeof(long));
+ /* Simply embedding 'struct cpumask' in 'struct sw_flow'
+ * consumes memory unnecessarily large. Cpumask is an bitmap
+ * of CONFIG_NR_CPUS bits, which is hardcoded in .config
+ * and default value can be 8192, in this case is 1024 bytes.
+ * It drops ovs_flow_alloc performance and cost memory.
+ * We should use actual CPU count instead of hardcoded value.
+ *
+ * To address this, 'cpu_used_mask' is redefined to pointer
+ * and append a cpumask_size() after 'stat' to hold the memory
+ * for struct cpumask.
+ *
+ * cpumask APIs like cpumask_next and cpumask_set_cpu are defined
+ * to never access bits beyond cpu count, as such above change is
+ * safe even though the actual memory provided is smaller than
+ * sizeof(struct cpumask)
+ */
flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow)
+ (nr_cpu_ids
- * sizeof(struct sw_flow_stats *)),
+ * sizeof(struct sw_flow_stats *))
+ + cpumask_size(),
0, 0, NULL);
if (flow_cache == NULL)
return -ENOMEM;
--
2.27.0
^ permalink raw reply related [flat|nested] 4+ messages in thread
* Re: [PATCH v2 1/1] net: openvswitch: reduce cpu_used_mask memory
2023-02-01 13:24 [PATCH v2 1/1] net: openvswitch: reduce cpu_used_mask memory taoyuan_eddy
@ 2023-02-01 13:44 ` Jiri Pirko
[not found] ` <OS3P286MB2295325EEB3F45E1F974001EF5D69@OS3P286MB2295.JPNP286.PROD.OUTLOOK.COM>
0 siblings, 1 reply; 4+ messages in thread
From: Jiri Pirko @ 2023-02-01 13:44 UTC (permalink / raw)
To: taoyuan_eddy
Cc: netdev, Pravin B Shelar, David S. Miller, Eric Dumazet,
Jakub Kicinski, Paolo Abeni, dev, linux-kernel
Wed, Feb 01, 2023 at 02:24:39PM CET, taoyuan_eddy@hotmail.com wrote:
>From: Eddy Tao <taoyuan_eddy@hotmail.com>
>
>'struct cpumask cpu_used_mask' is embedded in struct sw_flow.
>However, its size is hardcoded to CONFIG_NR_CPUS bits, which can be
>8192 by default, it costs memory and slows down ovs_flow_alloc.
>This fix uses actual CPU number instead
>
>Signed-off-by: Eddy Tao <taoyuan_eddy@hotmail.com>
Eddy, looks like you missed my second comment to v1. Could you please
check again?
Thanks!
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: 回复: [PATCH v2 1/1] net: openvswitch: reduce cpu_used_mask memory
[not found] ` <OS3P286MB2295325EEB3F45E1F974001EF5D69@OS3P286MB2295.JPNP286.PROD.OUTLOOK.COM>
@ 2023-02-02 5:35 ` Jakub Kicinski
2023-02-02 5:36 ` Jakub Kicinski
0 siblings, 1 reply; 4+ messages in thread
From: Jakub Kicinski @ 2023-02-02 5:35 UTC (permalink / raw)
To: 陶 缘
Cc: Jiri Pirko, netdev, Pravin B Shelar, David S. Miller,
Eric Dumazet, Paolo Abeni, dev, linux-kernel
On Thu, 2 Feb 2023 05:09:51 +0000 陶 缘 wrote:
> I guest you are pointing to the field "From: taoyuan_eddy@hotmail.com" in the patch header linked from "Headers show" section in the patch page
>
>
> https://patchwork.kernel.org/project/netdevbpf/patch/OS3P286MB22954422E3DD09FF5FD6B091F5D19@OS3P286MB2295.JPNP286.PROD.OUTLOOK.COM/
>
> I will fix that accordingly.
The From is correct, please look at the entire email Jiri also
commented about the code.
Two more notes:
- don't top post on the list
- reply in plain text not HTML
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: 回复: [PATCH v2 1/1] net: openvswitch: reduce cpu_used_mask memory
2023-02-02 5:35 ` 回复: " Jakub Kicinski
@ 2023-02-02 5:36 ` Jakub Kicinski
0 siblings, 0 replies; 4+ messages in thread
From: Jakub Kicinski @ 2023-02-02 5:36 UTC (permalink / raw)
To: 陶 缘
Cc: Jiri Pirko, netdev, Pravin B Shelar, David S. Miller,
Eric Dumazet, Paolo Abeni, dev, linux-kernel
On Wed, 1 Feb 2023 21:35:15 -0800 Jakub Kicinski wrote:
> On Thu, 2 Feb 2023 05:09:51 +0000 陶 缘 wrote:
> > I guest you are pointing to the field "From: taoyuan_eddy@hotmail.com" in the patch header linked from "Headers show" section in the patch page
> >
> >
> > https://patchwork.kernel.org/project/netdevbpf/patch/OS3P286MB22954422E3DD09FF5FD6B091F5D19@OS3P286MB2295.JPNP286.PROD.OUTLOOK.COM/
> >
> > I will fix that accordingly.
>
> The From is correct, please look at the entire email Jiri also
> commented about the code.
>
> Two more notes:
> - don't top post on the list
> - reply in plain text not HTML
One more thing, please make sure you read the process information
for Linux networking:
https://www.kernel.org/doc/html/next/process/maintainer-netdev.html
^ permalink raw reply [flat|nested] 4+ messages in thread
end of thread, other threads:[~2023-02-02 5:39 UTC | newest]
Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2023-02-01 13:24 [PATCH v2 1/1] net: openvswitch: reduce cpu_used_mask memory taoyuan_eddy
2023-02-01 13:44 ` Jiri Pirko
[not found] ` <OS3P286MB2295325EEB3F45E1F974001EF5D69@OS3P286MB2295.JPNP286.PROD.OUTLOOK.COM>
2023-02-02 5:35 ` 回复: " Jakub Kicinski
2023-02-02 5:36 ` Jakub Kicinski
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).