[PATCH net-next v3 0/3] locklessly protect left members in struct rps_dev_flow

[PATCH net-next v3 0/3] locklessly protect left members in struct rps_dev_flow

[Jason Xing]

From: Jason Xing <kernelxing@tencent.com>

Since Eric did a more complicated locklessly change to last_qtail
member[1] in struct rps_dev_flow, the left members are easier to change
as the same.

One thing important I would like to share by qooting Eric:
"rflow is located in rxqueue->rps_flow_table, it is thus private to current
thread. Only one cpu can service an RX queue at a time."
So we only pay attention to the reader in the rps_may_expire_flow() and
writer in the set_rps_cpu(). They are in the two different contexts.

[1]:
https://git.kernel.org/pub/scm/linux/kernel/git/netdev/net-next.git/commit/?id=3b4cf29bdab

v3
Link: https://lore.kernel.org/all/20240417062721.45652-1-kerneljasonxing@gmail.com/
1. adjust the protection in a right way (Eric)

v2
1. fix passing wrong type qtail.

Jason Xing (3):
  net: rps: protect last_qtail with rps_input_queue_tail_save() helper
  net: rps: protect filter locklessly
  net: rps: locklessly access rflow->cpu

 net/core/dev.c | 18 +++++++++---------
 1 file changed, 9 insertions(+), 9 deletions(-)

-- 
2.37.3

[PATCH net-next v3 1/3] net: rps: protect last_qtail with rps_input_queue_tail_save() helper

[Jason Xing]

From: Jason Xing <kernelxing@tencent.com>

Removing one unnecessary reader protection and add another writer
protection to finish the locklessly proctection job.

Note: the removed READ_ONCE() is not needed because we only have to protect
the locklessly reader in the different context (rps_may_expire_flow()).

Signed-off-by: Jason Xing <kernelxing@tencent.com>
---
 net/core/dev.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/net/core/dev.c b/net/core/dev.c
index 854a3a28a8d8..58e0da91bfef 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -4501,7 +4501,7 @@ set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
 		struct netdev_rx_queue *rxqueue;
 		struct rps_dev_flow_table *flow_table;
 		struct rps_dev_flow *old_rflow;
-		u32 flow_id;
+		u32 flow_id, head;
 		u16 rxq_index;
 		int rc;
 
@@ -4529,8 +4529,8 @@ set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
 			old_rflow->filter = RPS_NO_FILTER;
 	out:
 #endif
-		rflow->last_qtail =
-			READ_ONCE(per_cpu(softnet_data, next_cpu).input_queue_head);
+		head = READ_ONCE(per_cpu(softnet_data, next_cpu).input_queue_head);
+		rps_input_queue_tail_save(&rflow->last_qtail, head);
 	}
 
 	rflow->cpu = next_cpu;
@@ -4613,7 +4613,7 @@ static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
 		if (unlikely(tcpu != next_cpu) &&
 		    (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
 		     ((int)(READ_ONCE(per_cpu(softnet_data, tcpu).input_queue_head) -
-		      READ_ONCE(rflow->last_qtail))) >= 0)) {
+		      rflow->last_qtail)) >= 0)) {
 			tcpu = next_cpu;
 			rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
 		}
-- 
2.37.3

Re: [PATCH net-next v3 1/3] net: rps: protect last_qtail with rps_input_queue_tail_save() helper

[Eric Dumazet]

On Thu, Apr 18, 2024 at 9:36 AM Jason Xing <kerneljasonxing@gmail.com> wrote:
>
> From: Jason Xing <kernelxing@tencent.com>
>
> Removing one unnecessary reader protection and add another writer
> protection to finish the locklessly proctection job.
>
> Note: the removed READ_ONCE() is not needed because we only have to protect
> the locklessly reader in the different context (rps_may_expire_flow()).
>
> Signed-off-by: Jason Xing <kernelxing@tencent.com>

Reviewed-by: Eric Dumazet <edumazet@google.com>

compile error in set_rps_cpu() without CONFIG_RFS_ACCEL?

[John Sperbeck]

If CONFIG_RFS_ACCEL is off, then I think there will be a compile error with this change, since 'head' is used, but no longer defined.

Re: compile error in set_rps_cpu() without CONFIG_RFS_ACCEL?

[Eric Dumazet]

On Wed, May 29, 2024 at 10:34 PM John Sperbeck <jsperbeck@google.com> wrote:
>
> If CONFIG_RFS_ACCEL is off, then I think there will be a compile error with this change, since 'head' is used, but no longer defined.

I assume you are speaking of this commit ?

commit 84b6823cd96b38c40b3b30beabbfa48d92990e1a
Author: Jason Xing <kernelxing@tencent.com>
Date:   Thu Apr 18 15:36:01 2024 +0800

    net: rps: protect last_qtail with rps_input_queue_tail_save() helper

Re: compile error in set_rps_cpu() without CONFIG_RFS_ACCEL?

[Jason Xing]

On Thu, May 30, 2024 at 4:45 AM Eric Dumazet <edumazet@google.com> wrote:
>
> On Wed, May 29, 2024 at 10:34 PM John Sperbeck <jsperbeck@google.com> wrote:
> >
> > If CONFIG_RFS_ACCEL is off, then I think there will be a compile error with this change, since 'head' is used, but no longer defined.
>
> I assume you are speaking of this commit ?
>
> commit 84b6823cd96b38c40b3b30beabbfa48d92990e1a
> Author: Jason Xing <kernelxing@tencent.com>
> Date:   Thu Apr 18 15:36:01 2024 +0800
>
>     net: rps: protect last_qtail with rps_input_queue_tail_save() helper

Yes, I will fix it soon.

Thanks,
Jason

Re: compile error in set_rps_cpu() without CONFIG_RFS_ACCEL?

[Jason Xing]

Hi John,

On Thu, May 30, 2024 at 4:34 AM John Sperbeck <jsperbeck@google.com> wrote:
>
> If CONFIG_RFS_ACCEL is off, then I think there will be a compile error with this change, since 'head' is used, but no longer defined.

Thanks for your report. I'll fix it today.

[PATCH net-next v3 2/3] net: rps: protect filter locklessly

[Jason Xing]

From: Jason Xing <kernelxing@tencent.com>

As we can see, rflow->filter can be written/read concurrently, so
lockless access is needed.

Signed-off-by: Jason Xing <kernelxing@tencent.com>
---
 net/core/dev.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/net/core/dev.c b/net/core/dev.c
index 58e0da91bfef..ed6efef01582 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -4524,9 +4524,9 @@ set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
 			goto out;
 		old_rflow = rflow;
 		rflow = &flow_table->flows[flow_id];
-		rflow->filter = rc;
-		if (old_rflow->filter == rflow->filter)
-			old_rflow->filter = RPS_NO_FILTER;
+		WRITE_ONCE(rflow->filter, rc);
+		if (old_rflow->filter == rc)
+			WRITE_ONCE(old_rflow->filter, RPS_NO_FILTER);
 	out:
 #endif
 		head = READ_ONCE(per_cpu(softnet_data, next_cpu).input_queue_head);
@@ -4666,7 +4666,7 @@ bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
 	if (flow_table && flow_id <= flow_table->mask) {
 		rflow = &flow_table->flows[flow_id];
 		cpu = READ_ONCE(rflow->cpu);
-		if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
+		if (READ_ONCE(rflow->filter) == filter_id && cpu < nr_cpu_ids &&
 		    ((int)(READ_ONCE(per_cpu(softnet_data, cpu).input_queue_head) -
 			   READ_ONCE(rflow->last_qtail)) <
 		     (int)(10 * flow_table->mask)))
-- 
2.37.3

Re: [PATCH net-next v3 2/3] net: rps: protect filter locklessly

[Eric Dumazet]

On Thu, Apr 18, 2024 at 9:36 AM Jason Xing <kerneljasonxing@gmail.com> wrote:
>
> From: Jason Xing <kernelxing@tencent.com>
>
> As we can see, rflow->filter can be written/read concurrently, so
> lockless access is needed.
>
> Signed-off-by: Jason Xing <kernelxing@tencent.com>

Reviewed-by: Eric Dumazet <edumazet@google.com>

[PATCH net-next v3 3/3] net: rps: locklessly access rflow->cpu

[Jason Xing]

From: Jason Xing <kernelxing@tencent.com>

This is the last member in struct rps_dev_flow which should be
protected locklessly. So finish it.

Signed-off-by: Jason Xing <kernelxing@tencent.com>
---
 net/core/dev.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/net/core/dev.c b/net/core/dev.c
index ed6efef01582..8010036c07b6 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -4533,7 +4533,7 @@ set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
 		rps_input_queue_tail_save(&rflow->last_qtail, head);
 	}
 
-	rflow->cpu = next_cpu;
+	WRITE_ONCE(rflow->cpu, next_cpu);
 	return rflow;
 }
 
-- 
2.37.3

Re: [PATCH net-next v3 3/3] net: rps: locklessly access rflow->cpu

[Eric Dumazet]

On Thu, Apr 18, 2024 at 9:36 AM Jason Xing <kerneljasonxing@gmail.com> wrote:
>
> From: Jason Xing <kernelxing@tencent.com>
>
> This is the last member in struct rps_dev_flow which should be
> protected locklessly. So finish it.
>
> Signed-off-by: Jason Xing <kernelxing@tencent.com>

Reviewed-by: Eric Dumazet <edumazet@google.com>

Re: [PATCH net-next v3 0/3] locklessly protect left members in struct rps_dev_flow

[patchwork-bot+netdevbpf]

Hello:

This series was applied to netdev/net-next.git (main)
by David S. Miller <davem@davemloft.net>:

On Thu, 18 Apr 2024 15:36:00 +0800 you wrote:
> From: Jason Xing <kernelxing@tencent.com>
> 
> Since Eric did a more complicated locklessly change to last_qtail
> member[1] in struct rps_dev_flow, the left members are easier to change
> as the same.
> 
> One thing important I would like to share by qooting Eric:
> "rflow is located in rxqueue->rps_flow_table, it is thus private to current
> thread. Only one cpu can service an RX queue at a time."
> So we only pay attention to the reader in the rps_may_expire_flow() and
> writer in the set_rps_cpu(). They are in the two different contexts.
> 
> [...]

Here is the summary with links:
  - [net-next,v3,1/3] net: rps: protect last_qtail with rps_input_queue_tail_save() helper
    https://git.kernel.org/netdev/net-next/c/84b6823cd96b
  - [net-next,v3,2/3] net: rps: protect filter locklessly
    https://git.kernel.org/netdev/net-next/c/f00bf5dc8320
  - [net-next,v3,3/3] net: rps: locklessly access rflow->cpu
    https://git.kernel.org/netdev/net-next/c/f7b60cce8470

You are awesome, thank you!
-- 
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html