diff mbox series

[net-next,v2,2/3] net: rps: protect filter locklessly

Message ID 20240417062721.45652-3-kerneljasonxing@gmail.com (mailing list archive)
State Superseded
Delegated to: Netdev Maintainers
Headers show
Series locklessly protect left members in struct rps_dev_flow | expand

Checks

Context Check Description
netdev/series_format success Posting correctly formatted
netdev/tree_selection success Clearly marked for net-next
netdev/ynl success Generated files up to date; no warnings/errors; no diff in generated;
netdev/fixes_present success Fixes tag not required for -next series
netdev/header_inline success No static functions without inline keyword in header files
netdev/build_32bit success Errors and warnings before: 933 this patch: 933
netdev/build_tools success No tools touched, skip
netdev/cc_maintainers success CCed 4 of 4 maintainers
netdev/build_clang success Errors and warnings before: 938 this patch: 938
netdev/verify_signedoff success Signed-off-by tag matches author and committer
netdev/deprecated_api success None detected
netdev/check_selftest success No net selftest shell script
netdev/verify_fixes success No Fixes tag
netdev/build_allmodconfig_warn success Errors and warnings before: 944 this patch: 944
netdev/checkpatch success total: 0 errors, 0 warnings, 0 checks, 18 lines checked
netdev/build_clang_rust success No Rust files in patch. Skipping build
netdev/kdoc success Errors and warnings before: 79 this patch: 79
netdev/source_inline success Was 0 now: 0
netdev/contest success net-next-2024-04-18--00-00 (tests: 961)

Commit Message

Jason Xing April 17, 2024, 6:27 a.m. UTC
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>
---
I'm not very sure if the READ_ONCE in set_rps_cpu() is useful. I
scaned/checked the codes and found no lock can prevent multiple
threads from calling set_rps_cpu() and handling the same flow
simultaneously. The same question still exists in patch [3/3].
---
 net/core/dev.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

Comments

Eric Dumazet April 17, 2024, 10:04 a.m. UTC | #1
On Wed, Apr 17, 2024 at 8:27 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>
> ---
> I'm not very sure if the READ_ONCE in set_rps_cpu() is useful. I
> scaned/checked the codes and found no lock can prevent multiple
> threads from calling set_rps_cpu() and handling the same flow
> simultaneously. The same question still exists in patch [3/3].
> ---
>  net/core/dev.c | 6 +++---
>  1 file changed, 3 insertions(+), 3 deletions(-)
>
> diff --git a/net/core/dev.c b/net/core/dev.c
> index 2003b9a61e40..40a535158e45 100644
> --- a/net/core/dev.c
> +++ b/net/core/dev.c
> @@ -4524,8 +4524,8 @@ 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)
> +               WRITE_ONCE(rflow->filter, rc);
> +               if (old_rflow->filter == READ_ONCE(rflow->filter))

You missed the obvious opportunity to use

               if (old_rflow->filter ==  rc)

Here your code is going to force the compiler to read the memory right
after a prior write, adding a stall on some arches.

>                         old_rflow->filter = RPS_NO_FILTER;
>         out:
>  #endif
> @@ -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
>
Jason Xing April 17, 2024, 11:03 a.m. UTC | #2
On Wed, Apr 17, 2024 at 6:04 PM Eric Dumazet <edumazet@google.com> wrote:
>
> On Wed, Apr 17, 2024 at 8:27 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>
> > ---
> > I'm not very sure if the READ_ONCE in set_rps_cpu() is useful. I
> > scaned/checked the codes and found no lock can prevent multiple
> > threads from calling set_rps_cpu() and handling the same flow
> > simultaneously. The same question still exists in patch [3/3].
> > ---
> >  net/core/dev.c | 6 +++---
> >  1 file changed, 3 insertions(+), 3 deletions(-)
> >
> > diff --git a/net/core/dev.c b/net/core/dev.c
> > index 2003b9a61e40..40a535158e45 100644
> > --- a/net/core/dev.c
> > +++ b/net/core/dev.c
> > @@ -4524,8 +4524,8 @@ 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)
> > +               WRITE_ONCE(rflow->filter, rc);
> > +               if (old_rflow->filter == READ_ONCE(rflow->filter))
>
> You missed the obvious opportunity to use
>
>                if (old_rflow->filter ==  rc)
>
> Here your code is going to force the compiler to read the memory right
> after a prior write, adding a stall on some arches.

Thanks. I see. I will remove READ_ONCE() and then reuse 'rc'.

I would like to ask one relational question: could multiple threads
access the same rflow in set_rps_cpu() concurrently? Because I was
thinking a lot about whether I should use the READ_ONCE() here to
prevent another thread accessing/modifying this value concurrently.
The answer is probably yes?

Thanks,
Jason

>
> >                         old_rflow->filter = RPS_NO_FILTER;
> >         out:
> >  #endif
> > @@ -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
> >
Eric Dumazet April 17, 2024, 11:52 a.m. UTC | #3
On Wed, Apr 17, 2024 at 1:03 PM Jason Xing <kerneljasonxing@gmail.com> wrote:
>
> On Wed, Apr 17, 2024 at 6:04 PM Eric Dumazet <edumazet@google.com> wrote:
> >
> > On Wed, Apr 17, 2024 at 8:27 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>
> > > ---
> > > I'm not very sure if the READ_ONCE in set_rps_cpu() is useful. I
> > > scaned/checked the codes and found no lock can prevent multiple
> > > threads from calling set_rps_cpu() and handling the same flow
> > > simultaneously. The same question still exists in patch [3/3].
> > > ---
> > >  net/core/dev.c | 6 +++---
> > >  1 file changed, 3 insertions(+), 3 deletions(-)
> > >
> > > diff --git a/net/core/dev.c b/net/core/dev.c
> > > index 2003b9a61e40..40a535158e45 100644
> > > --- a/net/core/dev.c
> > > +++ b/net/core/dev.c
> > > @@ -4524,8 +4524,8 @@ 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)
> > > +               WRITE_ONCE(rflow->filter, rc);
> > > +               if (old_rflow->filter == READ_ONCE(rflow->filter))
> >
> > You missed the obvious opportunity to use
> >
> >                if (old_rflow->filter ==  rc)
> >
> > Here your code is going to force the compiler to read the memory right
> > after a prior write, adding a stall on some arches.
>
> Thanks. I see. I will remove READ_ONCE() and then reuse 'rc'.
>
> I would like to ask one relational question: could multiple threads
> access the same rflow in set_rps_cpu() concurrently? Because I was
> thinking a lot about whether I should use the READ_ONCE() here to
> prevent another thread accessing/modifying this value concurrently.

READ_ONCE() would not prevent this.

> The answer is probably yes?

I think the answer is no.  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.

I think you can scrap the patch series.

I will instead remove the not needed annotations :

diff --git a/include/net/rps.h b/include/net/rps.h
index a93401d23d66e45210acc73f0326087813b69d59..3f913464a2b321efe38a05dd107bf134fae6ad17
100644
--- a/include/net/rps.h
+++ b/include/net/rps.h
@@ -134,7 +134,7 @@ static inline u32 rps_input_queue_tail_incr(struct
softnet_data *sd)
 static inline void rps_input_queue_tail_save(u32 *dest, u32 tail)
 {
 #ifdef CONFIG_RPS
-       WRITE_ONCE(*dest, tail);
+       *dest = tail;
 #endif
 }

diff --git a/net/core/dev.c b/net/core/dev.c
index 854a3a28a8d85b335a9158378ae0cca6dfbf8b36..d774e4009790c9af30d3c8f9a5eab83e9cf01bd8
100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -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);
                }
@@ -4668,7 +4668,7 @@ bool rps_may_expire_flow(struct net_device *dev,
u16 rxq_index,
                cpu = READ_ONCE(rflow->cpu);
                if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
                    ((int)(READ_ONCE(per_cpu(softnet_data,
cpu).input_queue_head) -
-                          READ_ONCE(rflow->last_qtail)) <
+                          rflow->last_qtail) <
                     (int)(10 * flow_table->mask)))
                        expire = false;
        }
Eric Dumazet April 17, 2024, 11:58 a.m. UTC | #4
On Wed, Apr 17, 2024 at 1:52 PM Eric Dumazet <edumazet@google.com> wrote:
>
> @@ -4668,7 +4668,7 @@ bool rps_may_expire_flow(struct net_device *dev,
> u16 rxq_index,
>                 cpu = READ_ONCE(rflow->cpu);
>                 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
>                     ((int)(READ_ONCE(per_cpu(softnet_data,
> cpu).input_queue_head) -
> -                          READ_ONCE(rflow->last_qtail)) <
> +                          rflow->last_qtail) <
>                      (int)(10 * flow_table->mask)))
>                         expire = false;
>         }

Oh well, rps_may_expire_flow() might be called from other contexts, so
only the  READ_ONCE()
from get_rps_cpu() is not really necessary.
Jason Xing April 17, 2024, 12:14 p.m. UTC | #5
On Wed, Apr 17, 2024 at 7:58 PM Eric Dumazet <edumazet@google.com> wrote:
>
> On Wed, Apr 17, 2024 at 1:52 PM Eric Dumazet <edumazet@google.com> wrote:
> >
> > @@ -4668,7 +4668,7 @@ bool rps_may_expire_flow(struct net_device *dev,
> > u16 rxq_index,
> >                 cpu = READ_ONCE(rflow->cpu);
> >                 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
> >                     ((int)(READ_ONCE(per_cpu(softnet_data,
> > cpu).input_queue_head) -
> > -                          READ_ONCE(rflow->last_qtail)) <
> > +                          rflow->last_qtail) <
> >                      (int)(10 * flow_table->mask)))
> >                         expire = false;
> >         }
>
> Oh well, rps_may_expire_flow() might be called from other contexts, so
> only the  READ_ONCE()
> from get_rps_cpu() is not really necessary.

Thanks for telling me the access logic about qtail in the previous email.

Yes, I'm writing exactly what you're saying now :) I can keep
protecting rflow->cpu and rflow->filter locklessly.

I can remove the unneeded annotations around qtail as you suggested
with those two patches if I can, or you can submit it first. It's up
to you :)

Thanks,
Jason
Jason Xing April 17, 2024, 12:20 p.m. UTC | #6
On Wed, Apr 17, 2024 at 8:14 PM Jason Xing <kerneljasonxing@gmail.com> wrote:
>
> On Wed, Apr 17, 2024 at 7:58 PM Eric Dumazet <edumazet@google.com> wrote:
> >
> > On Wed, Apr 17, 2024 at 1:52 PM Eric Dumazet <edumazet@google.com> wrote:
> > >
> > > @@ -4668,7 +4668,7 @@ bool rps_may_expire_flow(struct net_device *dev,
> > > u16 rxq_index,
> > >                 cpu = READ_ONCE(rflow->cpu);
> > >                 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
> > >                     ((int)(READ_ONCE(per_cpu(softnet_data,
> > > cpu).input_queue_head) -
> > > -                          READ_ONCE(rflow->last_qtail)) <
> > > +                          rflow->last_qtail) <
> > >                      (int)(10 * flow_table->mask)))
> > >                         expire = false;
> > >         }
> >
> > Oh well, rps_may_expire_flow() might be called from other contexts, so
> > only the  READ_ONCE()
> > from get_rps_cpu() is not really necessary.
>
> Thanks for telling me the access logic about qtail in the previous email.
>
> Yes, I'm writing exactly what you're saying now :) I can keep
> protecting rflow->cpu and rflow->filter locklessly.
>

I can keep these three patches just like now only without that
READ_ONCE(), I have to update my statement.

[...]
> I can remove the unneeded annotations around qtail as you suggested
> with those two patches if I can, or you can submit it first. It's up
> to you :)

The 'qtail' also needs protection. What I was saying is not true.

>
> Thanks,
> Jason
diff mbox series

Patch

diff --git a/net/core/dev.c b/net/core/dev.c
index 2003b9a61e40..40a535158e45 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -4524,8 +4524,8 @@  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)
+		WRITE_ONCE(rflow->filter, rc);
+		if (old_rflow->filter == READ_ONCE(rflow->filter))
 			old_rflow->filter = RPS_NO_FILTER;
 	out:
 #endif
@@ -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)))