[PATCH net-next] dql: annotate data-races around dql->last_obj_cnt

[PATCH net-next] dql: annotate data-races around dql->last_obj_cnt

[Eric Dumazet]

dql->last_obj_cnt is read/written from different contexts,
without any lock synchronization.

Use READ_ONCE()/WRITE_ONCE() to avoid load/store tearing.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 include/linux/dynamic_queue_limits.h | 2 +-
 lib/dynamic_queue_limits.c           | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/include/linux/dynamic_queue_limits.h b/include/linux/dynamic_queue_limits.h
index 281298e77a1579cba1f92a3b3f03b8be089fd38f..808b1a5102e7c0bbbcd9676b0dacadad2f0ee49a 100644
--- a/include/linux/dynamic_queue_limits.h
+++ b/include/linux/dynamic_queue_limits.h
@@ -127,7 +127,7 @@ static inline void dql_queued(struct dql *dql, unsigned int count)
 	if (WARN_ON_ONCE(count > DQL_MAX_OBJECT))
 		return;
 
-	dql->last_obj_cnt = count;
+	WRITE_ONCE(dql->last_obj_cnt, count);
 
 	/* We want to force a write first, so that cpu do not attempt
 	 * to get cache line containing last_obj_cnt, num_queued, adj_limit
diff --git a/lib/dynamic_queue_limits.c b/lib/dynamic_queue_limits.c
index e49deddd3de9fe9e98d6712559cf48d12a0a2537..c1b7638a594ac43f947e00decabbd3468dcb53de 100644
--- a/lib/dynamic_queue_limits.c
+++ b/lib/dynamic_queue_limits.c
@@ -179,7 +179,7 @@ void dql_completed(struct dql *dql, unsigned int count)
 
 	dql->adj_limit = limit + completed;
 	dql->prev_ovlimit = ovlimit;
-	dql->prev_last_obj_cnt = dql->last_obj_cnt;
+	dql->prev_last_obj_cnt = READ_ONCE(dql->last_obj_cnt);
 	dql->num_completed = completed;
 	dql->prev_num_queued = num_queued;
 
-- 
2.47.0.163.g1226f6d8fa-goog

Re: [PATCH net-next] dql: annotate data-races around dql->last_obj_cnt

[Joe Damato]

On Tue, Oct 29, 2024 at 07:14:25PM +0000, Eric Dumazet wrote:
> dql->last_obj_cnt is read/written from different contexts,
> without any lock synchronization.
> 
> Use READ_ONCE()/WRITE_ONCE() to avoid load/store tearing.
> 
> Signed-off-by: Eric Dumazet <edumazet@google.com>
> ---
>  include/linux/dynamic_queue_limits.h | 2 +-
>  lib/dynamic_queue_limits.c           | 2 +-
>  2 files changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/include/linux/dynamic_queue_limits.h b/include/linux/dynamic_queue_limits.h
> index 281298e77a1579cba1f92a3b3f03b8be089fd38f..808b1a5102e7c0bbbcd9676b0dacadad2f0ee49a 100644
> --- a/include/linux/dynamic_queue_limits.h
> +++ b/include/linux/dynamic_queue_limits.h
> @@ -127,7 +127,7 @@ static inline void dql_queued(struct dql *dql, unsigned int count)
>  	if (WARN_ON_ONCE(count > DQL_MAX_OBJECT))
>  		return;
>  
> -	dql->last_obj_cnt = count;
> +	WRITE_ONCE(dql->last_obj_cnt, count);
>  
>  	/* We want to force a write first, so that cpu do not attempt
>  	 * to get cache line containing last_obj_cnt, num_queued, adj_limit
> diff --git a/lib/dynamic_queue_limits.c b/lib/dynamic_queue_limits.c
> index e49deddd3de9fe9e98d6712559cf48d12a0a2537..c1b7638a594ac43f947e00decabbd3468dcb53de 100644
> --- a/lib/dynamic_queue_limits.c
> +++ b/lib/dynamic_queue_limits.c
> @@ -179,7 +179,7 @@ void dql_completed(struct dql *dql, unsigned int count)
>  
>  	dql->adj_limit = limit + completed;
>  	dql->prev_ovlimit = ovlimit;
> -	dql->prev_last_obj_cnt = dql->last_obj_cnt;
> +	dql->prev_last_obj_cnt = READ_ONCE(dql->last_obj_cnt);
>  	dql->num_completed = completed;
>  	dql->prev_num_queued = num_queued;
>  

This looks fine to me. I noted that dql_reset writes last_obj_cnt,
but AFAIU that write is not a problem (from the 1 driver I looked
at).

Reviewed-by: Joe Damato <jdamato@fastly.com>

Re: [PATCH net-next] dql: annotate data-races around dql->last_obj_cnt

[Eric Dumazet]

On Tue, Oct 29, 2024 at 8:34 PM Joe Damato <jdamato@fastly.com> wrote:
>
> On Tue, Oct 29, 2024 at 07:14:25PM +0000, Eric Dumazet wrote:
> > dql->last_obj_cnt is read/written from different contexts,
> > without any lock synchronization.
> >
> > Use READ_ONCE()/WRITE_ONCE() to avoid load/store tearing.
> >
> > Signed-off-by: Eric Dumazet <edumazet@google.com>
> > ---
> >  include/linux/dynamic_queue_limits.h | 2 +-
> >  lib/dynamic_queue_limits.c           | 2 +-
> >  2 files changed, 2 insertions(+), 2 deletions(-)
> >
> > diff --git a/include/linux/dynamic_queue_limits.h b/include/linux/dynamic_queue_limits.h
> > index 281298e77a1579cba1f92a3b3f03b8be089fd38f..808b1a5102e7c0bbbcd9676b0dacadad2f0ee49a 100644
> > --- a/include/linux/dynamic_queue_limits.h
> > +++ b/include/linux/dynamic_queue_limits.h
> > @@ -127,7 +127,7 @@ static inline void dql_queued(struct dql *dql, unsigned int count)
> >       if (WARN_ON_ONCE(count > DQL_MAX_OBJECT))
> >               return;
> >
> > -     dql->last_obj_cnt = count;
> > +     WRITE_ONCE(dql->last_obj_cnt, count);
> >
> >       /* We want to force a write first, so that cpu do not attempt
> >        * to get cache line containing last_obj_cnt, num_queued, adj_limit
> > diff --git a/lib/dynamic_queue_limits.c b/lib/dynamic_queue_limits.c
> > index e49deddd3de9fe9e98d6712559cf48d12a0a2537..c1b7638a594ac43f947e00decabbd3468dcb53de 100644
> > --- a/lib/dynamic_queue_limits.c
> > +++ b/lib/dynamic_queue_limits.c
> > @@ -179,7 +179,7 @@ void dql_completed(struct dql *dql, unsigned int count)
> >
> >       dql->adj_limit = limit + completed;
> >       dql->prev_ovlimit = ovlimit;
> > -     dql->prev_last_obj_cnt = dql->last_obj_cnt;
> > +     dql->prev_last_obj_cnt = READ_ONCE(dql->last_obj_cnt);
> >       dql->num_completed = completed;
> >       dql->prev_num_queued = num_queued;
> >
>
> This looks fine to me. I noted that dql_reset writes last_obj_cnt,
> but AFAIU that write is not a problem (from the 1 driver I looked
> at).

Yeah, I think that dql_reset() should not exist in the first place.

When all skbs are properly tx completed, BQL state should be known and clean.

Re: [PATCH net-next] dql: annotate data-races around dql->last_obj_cnt

[patchwork-bot+netdevbpf]

Hello:

This patch was applied to netdev/net-next.git (main)
by Jakub Kicinski <kuba@kernel.org>:

On Tue, 29 Oct 2024 19:14:25 +0000 you wrote:
> dql->last_obj_cnt is read/written from different contexts,
> without any lock synchronization.
> 
> Use READ_ONCE()/WRITE_ONCE() to avoid load/store tearing.
> 
> Signed-off-by: Eric Dumazet <edumazet@google.com>
> 
> [...]

Here is the summary with links:
  - [net-next] dql: annotate data-races around dql->last_obj_cnt
    https://git.kernel.org/netdev/net-next/c/a911bad094b0

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