| Message ID | OS3P286MB22955AB6FF67B67778343FEDF5D79@OS3P286MB2295.JPNP286.PROD.OUTLOOK.COM |
|---|---|
| State | Superseded |
| Headers | show |
| Series | [ovs-dev,net-next,v5,1/1] net:openvswitch:reduce cpu_used_mask memory | expand |
Fri, Feb 03, 2023 at 09:52:56AM CET, taoyuan_eddy@hotmail.com wrote: >Use actual CPU number instead of hardcoded value to decide the size >of 'cpu_used_mask' in 'struct sw_flow'. Below is the reason. > >'struct cpumask cpu_used_mask' is embedded in struct sw_flow. >Its size is hardcoded to CONFIG_NR_CPUS bits, which can be >8192 by default, it costs memory and slows down ovs_flow_alloc > >To address this, redefine cpu_used_mask to pointer >append cpumask_size() bytes after 'stat' to hold cpumask > >cpumask APIs like cpumask_next and cpumask_set_cpu never access >bits beyond cpu count, cpumask_size() bytes of memory is enough > >Signed-off-by: Eddy Tao <taoyuan_eddy@hotmail.com> Eddy, could you please slow down a bit? Why did you send v5 right after v4? Could you please always put a changelog to the patch submitted to contain info about changes in between the submitted version? Thanks!
sorry, Jiri, my bad, will hold on before the next submission/hope it is the final one eddy
Fri, Feb 03, 2023 at 10:03:28AM CET, taoyuan_eddy@hotmail.com wrote: >sorry, Jiri, my bad, will hold on before the next submission/hope it is the final one Don't top-post please. So what is the change comparing to v4? > >eddy >________________________________ >发件人: Jiri Pirko <jiri@resnulli.us> >发送时间: 2023年2月3日 9:00 >收件人: Eddy Tao <taoyuan_eddy@hotmail.com> >抄送: netdev@vger.kernel.org <netdev@vger.kernel.org>; Pravin B Shelar <pshelar@ovn.org>; David S. Miller <davem@davemloft.net>; Eric Dumazet <edumazet@google.com>; Jakub Kicinski <kuba@kernel.org>; Paolo Abeni <pabeni@redhat.com>; dev@openvswitch.org <dev@openvswitch.org>; linux-kernel@vger.kernel.org <linux-kernel@vger.kernel.org> >主题: Re: [PATCH net-next v5 1/1] net:openvswitch:reduce cpu_used_mask memory > >Fri, Feb 03, 2023 at 09:52:56AM CET, taoyuan_eddy@hotmail.com wrote: >>Use actual CPU number instead of hardcoded value to decide the size >>of 'cpu_used_mask' in 'struct sw_flow'. Below is the reason. >> >>'struct cpumask cpu_used_mask' is embedded in struct sw_flow. >>Its size is hardcoded to CONFIG_NR_CPUS bits, which can be >>8192 by default, it costs memory and slows down ovs_flow_alloc >> >>To address this, redefine cpu_used_mask to pointer >>append cpumask_size() bytes after 'stat' to hold cpumask >> >>cpumask APIs like cpumask_next and cpumask_set_cpu never access >>bits beyond cpu count, cpumask_size() bytes of memory is enough >> >>Signed-off-by: Eddy Tao <taoyuan_eddy@hotmail.com> > >Eddy, could you please slow down a bit? Why did you send v5 right >after v4? Could you please always put a changelog to the patch >submitted to contain info about changes in between the >submitted version? > >Thanks! >
Hi, Jiri:
Difference between v5 and v4 are a wrap to fix max-line-length warning
- cpumask_set_cpu(cpu, &flow->cpu_used_mask);
+ cpumask_set_cpu(cpu,
+ flow->cpu_used_mask);
I will make the newer submission contain the diff from previous one in the next submission.
sorry for the inconvenience
eddy
diff --git a/net/openvswitch/flow.c b/net/openvswitch/flow.c index e20d1a973417..a56483eda015 100644 --- a/net/openvswitch/flow.c +++ b/net/openvswitch/flow.c @@ -107,7 +107,8 @@ 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 +136,8 @@ 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 +161,8 @@ 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..dc6a174c3194 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]); @@ -1196,7 +1197,8 @@ int ovs_flow_init(void) 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;
Use actual CPU number instead of hardcoded value to decide the size of 'cpu_used_mask' in 'struct sw_flow'. Below is the reason. 'struct cpumask cpu_used_mask' is embedded in struct sw_flow. Its size is hardcoded to CONFIG_NR_CPUS bits, which can be 8192 by default, it costs memory and slows down ovs_flow_alloc To address this, redefine cpu_used_mask to pointer append cpumask_size() bytes after 'stat' to hold cpumask cpumask APIs like cpumask_next and cpumask_set_cpu never access bits beyond cpu count, cpumask_size() bytes of memory is enough Signed-off-by: Eddy Tao <taoyuan_eddy@hotmail.com> --- net/openvswitch/flow.c | 9 ++++++--- net/openvswitch/flow.h | 2 +- net/openvswitch/flow_table.c | 8 +++++--- 3 files changed, 12 insertions(+), 7 deletions(-)