On Wed, Aug 02, 2023 at 05:34:42PM +1000, NeilBrown wrote:
> Rather than searching a list of threads to find an idle one, having a
> list of idle threads allows an idle thread to be found immediately.
>
> This adds some spin_lock calls which is not ideal, but as the hold-time
> is tiny it is still faster than searching a list.
Keep in mind that b1691bc03d4e ("sunrpc: convert to lockless lookup
of queued server threads") did the opposite because that very
spin_lock was highly contended. I am skeptical of the above claim
without lock_stat data... but that's sort of moot as this is a
temporary situation, as you point out next.
> A future patch will
> remove them using llist.h. This involves some subtlety and so is left
> to a separate patch.
Since I haven't seen that patch yet, I'm reserving judgement about
whether and how these two changes might be merged.
> This removes the need for the RQ_BUSY flag. The rqst is "busy"
> precisely when it is not on the "idle" list.
I've been having some trouble with this one. The server system
deadlocks hard as soon as the NFS server starts. I tracked it down
this morning: this patch never initialized the sp_idle_threads
list_head.
I will apply this patch (with one-line fix) and the patch that
removes SP_CONGESTED once I hear from the client folks on the
"integrate backchannel" patch.
> Signed-off-by: NeilBrown <neilb@xxxxxxx>
> ---
> include/linux/sunrpc/svc.h | 25 ++++++++++++++++++++++++-
> include/trace/events/sunrpc.h | 1 -
> net/sunrpc/svc.c | 13 ++++++++-----
> net/sunrpc/svc_xprt.c | 15 +++++++++++----
> 4 files changed, 43 insertions(+), 11 deletions(-)
>
> diff --git a/include/linux/sunrpc/svc.h b/include/linux/sunrpc/svc.h
> index 1ac6f74781aa..8b93af92dd53 100644
> --- a/include/linux/sunrpc/svc.h
> +++ b/include/linux/sunrpc/svc.h
> @@ -37,6 +37,7 @@ struct svc_pool {
> struct list_head sp_sockets; /* pending sockets */
> unsigned int sp_nrthreads; /* # of threads in pool */
> struct list_head sp_all_threads; /* all server threads */
> + struct list_head sp_idle_threads; /* idle server threads */
>
> /* statistics on pool operation */
> struct percpu_counter sp_messages_arrived;
> @@ -186,6 +187,7 @@ extern u32 svc_max_payload(const struct svc_rqst *rqstp);
> */
> struct svc_rqst {
> struct list_head rq_all; /* all threads list */
> + struct list_head rq_idle; /* On the idle list */
> struct rcu_head rq_rcu_head; /* for RCU deferred kfree */
> struct svc_xprt * rq_xprt; /* transport ptr */
>
> @@ -262,10 +264,31 @@ enum {
> RQ_SPLICE_OK, /* turned off in gss privacy to prevent
> * encrypting page cache pages */
> RQ_VICTIM, /* Have agreed to shut down */
> - RQ_BUSY, /* request is busy */
> RQ_DATA, /* request has data */
> };
>
> +/**
> + * svc_thread_set_busy - mark a thread as busy
> + * @rqstp: the thread which is now busy
> + *
> + * If rq_idle is "empty", the thread must be busy.
> + */
> +static inline void svc_thread_set_busy(struct svc_rqst *rqstp)
> +{
> + INIT_LIST_HEAD(&rqstp->rq_idle);
> +}
> +
> +/**
> + * svc_thread_busy - check if a thread as busy
> + * @rqstp: the thread which might be busy
> + *
> + * If rq_idle is "empty", the thread must be busy.
> + */
> +static inline bool svc_thread_busy(struct svc_rqst *rqstp)
> +{
> + return list_empty(&rqstp->rq_idle);
> +}
> +
> #define SVC_NET(rqst) (rqst->rq_xprt ? rqst->rq_xprt->xpt_net : rqst->rq_bc_net)
>
> /*
> diff --git a/include/trace/events/sunrpc.h b/include/trace/events/sunrpc.h
> index 6beb38c1dcb5..337c90787fb1 100644
> --- a/include/trace/events/sunrpc.h
> +++ b/include/trace/events/sunrpc.h
> @@ -1677,7 +1677,6 @@ DEFINE_SVCXDRBUF_EVENT(sendto);
> svc_rqst_flag(DROPME) \
> svc_rqst_flag(SPLICE_OK) \
> svc_rqst_flag(VICTIM) \
> - svc_rqst_flag(BUSY) \
> svc_rqst_flag_end(DATA)
>
> #undef svc_rqst_flag
> diff --git a/net/sunrpc/svc.c b/net/sunrpc/svc.c
> index 1233d72714b9..dce433dea1bd 100644
> --- a/net/sunrpc/svc.c
> +++ b/net/sunrpc/svc.c
> @@ -641,7 +641,7 @@ svc_rqst_alloc(struct svc_serv *serv, struct svc_pool *pool, int node)
>
> folio_batch_init(&rqstp->rq_fbatch);
>
> - __set_bit(RQ_BUSY, &rqstp->rq_flags);
> + svc_thread_set_busy(rqstp);
> rqstp->rq_server = serv;
> rqstp->rq_pool = pool;
>
> @@ -702,10 +702,13 @@ void svc_pool_wake_idle_thread(struct svc_pool *pool)
> struct svc_rqst *rqstp;
>
> rcu_read_lock();
> - list_for_each_entry_rcu(rqstp, &pool->sp_all_threads, rq_all) {
> - if (test_and_set_bit(RQ_BUSY, &rqstp->rq_flags))
> - continue;
> -
> + spin_lock_bh(&pool->sp_lock);
> + rqstp = list_first_entry_or_null(&pool->sp_idle_threads,
> + struct svc_rqst, rq_idle);
> + if (rqstp)
> + list_del_init(&rqstp->rq_idle);
> + spin_unlock_bh(&pool->sp_lock);
> + if (rqstp) {
> WRITE_ONCE(rqstp->rq_qtime, ktime_get());
> wake_up_process(rqstp->rq_task);
> rcu_read_unlock();
> diff --git a/net/sunrpc/svc_xprt.c b/net/sunrpc/svc_xprt.c
> index 0a300ae6a7ed..e44efcc21b63 100644
> --- a/net/sunrpc/svc_xprt.c
> +++ b/net/sunrpc/svc_xprt.c
> @@ -737,8 +737,9 @@ static void svc_rqst_wait_for_work(struct svc_rqst *rqstp)
> set_current_state(TASK_IDLE);
> smp_mb__before_atomic();
> clear_bit(SP_CONGESTED, &pool->sp_flags);
> - clear_bit(RQ_BUSY, &rqstp->rq_flags);
> - smp_mb__after_atomic();
> + spin_lock_bh(&pool->sp_lock);
> + list_add(&rqstp->rq_idle, &pool->sp_idle_threads);
> + spin_unlock_bh(&pool->sp_lock);
>
> /* Need to check should_sleep() again after
> * setting task state in case a wakeup happened
> @@ -751,8 +752,14 @@ static void svc_rqst_wait_for_work(struct svc_rqst *rqstp)
> cond_resched();
> }
>
> - set_bit(RQ_BUSY, &rqstp->rq_flags);
> - smp_mb__after_atomic();
> + /* We *must* be removed from the list before we can continue.
> + * If we were woken, this is already done
> + */
> + if (!svc_thread_busy(rqstp)) {
> + spin_lock_bh(&pool->sp_lock);
> + list_del_init(&rqstp->rq_idle);
> + spin_unlock_bh(&pool->sp_lock);
> + }
> } else
> cond_resched();
> try_to_freeze();
> --
> 2.40.1
>
--
Chuck Lever
