On Wed, 26 Jan 2022, Chuck Lever wrote:
> We have never been able to track down and address the underlying
> cause of the performance issues with workqueue-based service
> support. svo_enqueue_xprt is called multiple times per RPC, so
> it adds instruction path length, but always ends up at the same
> function: svc_xprt_do_enqueue(). We do not anticipate needing
> this flexibility for dynamic nfsd thread management support.
>
> As a micro-optimization, remove .svo_enqueue_xprt because
> Spectre/Meltdown makes virtual function calls more costly.
>
> This change essentially reverts commit b9e13cdfac70 ("nfsd/sunrpc:
> turn enqueueing a svc_xprt into a svc_serv operation").
>
> Signed-off-by: Chuck Lever <chuck.lever@xxxxxxxxxx>
> ---
> fs/lockd/svc.c | 1 -
> fs/nfs/callback.c | 2 --
> fs/nfsd/nfssvc.c | 1 -
> include/linux/sunrpc/svc.h | 3 ---
> include/linux/sunrpc/svc_xprt.h | 1 -
> net/sunrpc/svc_xprt.c | 10 +++++-----
> 6 files changed, 5 insertions(+), 13 deletions(-)
>
> diff --git a/fs/lockd/svc.c b/fs/lockd/svc.c
> index 0475c5a5d061..3a05af873625 100644
> --- a/fs/lockd/svc.c
> +++ b/fs/lockd/svc.c
> @@ -353,7 +353,6 @@ static struct notifier_block lockd_inet6addr_notifier = {
> static const struct svc_serv_ops lockd_sv_ops = {
> .svo_shutdown = svc_rpcb_cleanup,
> .svo_function = lockd,
> - .svo_enqueue_xprt = svc_xprt_do_enqueue,
> .svo_module = THIS_MODULE,
> };
>
> diff --git a/fs/nfs/callback.c b/fs/nfs/callback.c
> index 054cc1255fac..7a810f885063 100644
> --- a/fs/nfs/callback.c
> +++ b/fs/nfs/callback.c
> @@ -234,13 +234,11 @@ static int nfs_callback_up_net(int minorversion, struct svc_serv *serv,
>
> static const struct svc_serv_ops nfs40_cb_sv_ops = {
> .svo_function = nfs4_callback_svc,
> - .svo_enqueue_xprt = svc_xprt_do_enqueue,
> .svo_module = THIS_MODULE,
> };
> #if defined(CONFIG_NFS_V4_1)
> static const struct svc_serv_ops nfs41_cb_sv_ops = {
> .svo_function = nfs41_callback_svc,
> - .svo_enqueue_xprt = svc_xprt_do_enqueue,
> .svo_module = THIS_MODULE,
> };
>
> diff --git a/fs/nfsd/nfssvc.c b/fs/nfsd/nfssvc.c
> index b8c682b62d29..aeeac6de1f0a 100644
> --- a/fs/nfsd/nfssvc.c
> +++ b/fs/nfsd/nfssvc.c
> @@ -615,7 +615,6 @@ static int nfsd_get_default_max_blksize(void)
> static const struct svc_serv_ops nfsd_thread_sv_ops = {
> .svo_shutdown = nfsd_last_thread,
> .svo_function = nfsd,
> - .svo_enqueue_xprt = svc_xprt_do_enqueue,
> .svo_module = THIS_MODULE,
> };
>
> diff --git a/include/linux/sunrpc/svc.h b/include/linux/sunrpc/svc.h
> index f35c22b3355f..6ef9c1cafd0b 100644
> --- a/include/linux/sunrpc/svc.h
> +++ b/include/linux/sunrpc/svc.h
> @@ -61,9 +61,6 @@ struct svc_serv_ops {
> /* function for service threads to run */
> int (*svo_function)(void *);
>
> - /* queue up a transport for servicing */
> - void (*svo_enqueue_xprt)(struct svc_xprt *);
> -
> /* optional module to count when adding threads.
> * Thread function must call module_put_and_kthread_exit() to exit.
> */
> diff --git a/include/linux/sunrpc/svc_xprt.h b/include/linux/sunrpc/svc_xprt.h
> index 571f605bc91e..a3ba027fb4ba 100644
> --- a/include/linux/sunrpc/svc_xprt.h
> +++ b/include/linux/sunrpc/svc_xprt.h
> @@ -131,7 +131,6 @@ int svc_create_xprt(struct svc_serv *, const char *, struct net *,
> const int, const unsigned short, int,
> const struct cred *);
> void svc_xprt_received(struct svc_xprt *xprt);
> -void svc_xprt_do_enqueue(struct svc_xprt *xprt);
> void svc_xprt_enqueue(struct svc_xprt *xprt);
> void svc_xprt_put(struct svc_xprt *xprt);
> void svc_xprt_copy_addrs(struct svc_rqst *rqstp, struct svc_xprt *xprt);
> diff --git a/net/sunrpc/svc_xprt.c b/net/sunrpc/svc_xprt.c
> index b21ad7994147..9fce4f7774bb 100644
> --- a/net/sunrpc/svc_xprt.c
> +++ b/net/sunrpc/svc_xprt.c
> @@ -32,6 +32,7 @@ static int svc_deferred_recv(struct svc_rqst *rqstp);
> static struct cache_deferred_req *svc_defer(struct cache_req *req);
> static void svc_age_temp_xprts(struct timer_list *t);
> static void svc_delete_xprt(struct svc_xprt *xprt);
> +static void svc_xprt_do_enqueue(struct svc_xprt *xprt);
>
> /* apparently the "standard" is that clients close
> * idle connections after 5 minutes, servers after
> @@ -266,12 +267,12 @@ void svc_xprt_received(struct svc_xprt *xprt)
> }
>
> /* As soon as we clear busy, the xprt could be closed and
> - * 'put', so we need a reference to call svc_enqueue_xprt with:
> + * 'put', so we need a reference to call svc_xprt_do_enqueue with:
> */
> svc_xprt_get(xprt);
> smp_mb__before_atomic();
> clear_bit(XPT_BUSY, &xprt->xpt_flags);
> - xprt->xpt_server->sv_ops->svo_enqueue_xprt(xprt);
> + svc_xprt_do_enqueue(xprt);
> svc_xprt_put(xprt);
> }
> EXPORT_SYMBOL_GPL(svc_xprt_received);
> @@ -423,7 +424,7 @@ static bool svc_xprt_ready(struct svc_xprt *xprt)
> return false;
> }
>
> -void svc_xprt_do_enqueue(struct svc_xprt *xprt)
> +static void svc_xprt_do_enqueue(struct svc_xprt *xprt)
> {
> struct svc_pool *pool;
> struct svc_rqst *rqstp = NULL;
> @@ -467,7 +468,6 @@ void svc_xprt_do_enqueue(struct svc_xprt *xprt)
> put_cpu();
> trace_svc_xprt_enqueue(xprt, rqstp);
> }
> -EXPORT_SYMBOL_GPL(svc_xprt_do_enqueue);
>
> /*
> * Queue up a transport with data pending. If there are idle nfsd
> @@ -478,7 +478,7 @@ void svc_xprt_enqueue(struct svc_xprt *xprt)
> {
> if (test_bit(XPT_BUSY, &xprt->xpt_flags))
> return;
> - xprt->xpt_server->sv_ops->svo_enqueue_xprt(xprt);
> + svc_xprt_do_enqueue(xprt);
> }
> EXPORT_SYMBOL_GPL(svc_xprt_enqueue);
>
I obviously like this, but I think we can do better - merge
svc_xprt_enqueue and svc_xprt_do_enqueue, and drop the extra test on
XPT_BUSY.
They were separated in Commit 0971374e2818 ("SUNRPC: Reduce contention
in svc_xprt_enqueue()") so that the XPT_BUSY check happened before
taking any spinlocks.
We have since moved or removed the spinlocks so the extra test is fairly
pointless.
Possibly we could add
if (xpt_flags & BIT(XPR_BUSY))
return false;
at the start of svc_xprt_ready(). It would make sense there and be
close to free.
Thanks,
NeilBrown
