On Fri, 2020-11-13 at 14:14 -0600, Tom Zanussi wrote:
>
> This patch seems to fix it for me:
If there was any discussion about this patch, I missed it.
> From 4855377d0cb34b1b67a5c6d84cc8609c9da0bc3e Mon Sep 17 00:00:00 2001
> Message-Id: <4855377d0cb34b1b67a5c6d84cc8609c9da0bc3e.1605297603.git.zanussi@xxxxxxxxxx>
> From: Tom Zanussi <zanussi@xxxxxxxxxx>
> Date: Fri, 13 Nov 2020 13:04:15 -0600
> Subject: [PATCH] net: Add missing __raw_write_seqcount_end() and
> seqcount_release()
>
> The patch ('net: Properly annotate the try-lock for the seqlock") adds
> __raw_write_seqcount_begin() in qdisc_run_begin() but omits the
> corresponding __raw_write_seqcount_end() and seqcount_release() in
> qdisc_run_end().
>
> Add it unconditionally, since qdisc_run_end() is never called unless
> qdisc_run_begin() succeeds, and if it succeeds,
> __raw_write_seqcount_begin() seqcount_acquire() will have been called.
>
> Signed-off-by: Tom Zanussi <zanussi@xxxxxxxxxx>
> ---
> include/net/sch_generic.h | 4 ++++
> 1 file changed, 4 insertions(+)
>
> diff --git a/include/net/sch_generic.h b/include/net/sch_generic.h
> index 112d2dca8b08..c5ccce4f8f62 100644
> --- a/include/net/sch_generic.h
> +++ b/include/net/sch_generic.h
> @@ -192,7 +192,11 @@ static inline bool qdisc_run_begin(struct Qdisc *qdisc)
> static inline void qdisc_run_end(struct Qdisc *qdisc)
> {
> #ifdef CONFIG_PREEMPT_RT
> + seqcount_t *s = &qdisc->running.seqcount;
> +
> write_sequnlock(&qdisc->running);
> + __raw_write_seqcount_end(s);
> + seqcount_release(&s->dep_map, 1, _RET_IP_);
> #else
> write_seqcount_end(&qdisc->running);
> #endif
__raw_write_seqcount_end() is an integral part of write_sequnlock(),
but we do seem to be missing a seqcount_release() in 5.4-rt.
-Mike
