ipc/msg: mitigate the lock contention with percpu counter

The msg_bytes and msg_hdrs atomic counters are frequently updated when IPC
msg queue is in heavy use, causing heavy cache bounce and overhead.
Change them to percpu_counter greatly improve the performance.  Since
there is one percpu struct per namespace, additional memory cost is
minimal.  Reading of the count done in msgctl call, which is infrequent.
So the need to sum up the counts in each CPU is infrequent.

Apply the patch and test the pts/stress-ng-1.4.0
-- system v message passing (160 threads).

Score gain: 3.99x

CPU: ICX 8380 x 2 sockets
Core number: 40 x 2 physical cores
Benchmark: pts/stress-ng-1.4.0
-- system v message passing (160 threads)

[akpm@linux-foundation.org: coding-style cleanups]
[jiebin.sun@intel.com: avoid negative value by overflow in msginfo]
Link: https://lkml.kernel.org/r/20220920150809.4014944-1-jiebin.sun@intel.com
[akpm@linux-foundation.org: fix min() warnings]
Link: https://lkml.kernel.org/r/20220913192538.3023708-3-jiebin.sun@intel.com
Signed-off-by: Jiebin Sun <jiebin.sun@intel.com>
Reviewed-by: Tim Chen <tim.c.chen@linux.intel.com>
Cc: Alexander Mikhalitsyn <alexander.mikhalitsyn@virtuozzo.com>
Cc: Alexey Gladkov <legion@kernel.org>
Cc: Christoph Lameter <cl@linux.com>
Cc: Davidlohr Bueso <dave@stgolabs.net>
Cc: Dennis Zhou <dennis@kernel.org>
Cc: "Eric W . Biederman" <ebiederm@xmission.com>
Cc: Manfred Spraul <manfred@colorfullife.com>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Vasily Averin <vasily.averin@linux.dev>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

diff --git a/include/linux/ipc_namespace.h b/include/linux/ipc_namespace.h
index e3e8c86..e8240cf 100644 (file)
--- a/include/linux/ipc_namespace.h
+++ b/include/linux/ipc_namespace.h
@@ -11,6 +11,7 @@
 #include <linux/refcount.h>
 #include <linux/rhashtable-types.h>
 #include <linux/sysctl.h>
+#include <linux/percpu_counter.h>
 
 struct user_namespace;
 
@@ -36,8 +37,8 @@ struct ipc_namespace {
        unsigned int    msg_ctlmax;
        unsigned int    msg_ctlmnb;
        unsigned int    msg_ctlmni;
-       atomic_t        msg_bytes;
-       atomic_t        msg_hdrs;
+       struct percpu_counter percpu_msg_bytes;
+       struct percpu_counter percpu_msg_hdrs;
 
        size_t          shm_ctlmax;
        size_t          shm_ctlall;

diff --git a/ipc/msg.c b/ipc/msg.c
index a0d0577..e4e0990 100644 (file)
--- a/ipc/msg.c
+++ b/ipc/msg.c
@@ -39,6 +39,7 @@
 #include <linux/nsproxy.h>
 #include <linux/ipc_namespace.h>
 #include <linux/rhashtable.h>
+#include <linux/percpu_counter.h>
 
 #include <asm/current.h>
 #include <linux/uaccess.h>
@@ -285,10 +286,10 @@ static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
        rcu_read_unlock();
 
        list_for_each_entry_safe(msg, t, &msq->q_messages, m_list) {
-               atomic_dec(&ns->msg_hdrs);
+               percpu_counter_sub_local(&ns->percpu_msg_hdrs, 1);
                free_msg(msg);
        }
-       atomic_sub(msq->q_cbytes, &ns->msg_bytes);
+       percpu_counter_sub_local(&ns->percpu_msg_bytes, msq->q_cbytes);
        ipc_update_pid(&msq->q_lspid, NULL);
        ipc_update_pid(&msq->q_lrpid, NULL);
        ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
@@ -495,17 +496,22 @@ static int msgctl_info(struct ipc_namespace *ns, int msqid,
        msginfo->msgssz = MSGSSZ;
        msginfo->msgseg = MSGSEG;
        down_read(&msg_ids(ns).rwsem);
-       if (cmd == MSG_INFO) {
+       if (cmd == MSG_INFO)
                msginfo->msgpool = msg_ids(ns).in_use;
-               msginfo->msgmap = atomic_read(&ns->msg_hdrs);
-               msginfo->msgtql = atomic_read(&ns->msg_bytes);
+       max_idx = ipc_get_maxidx(&msg_ids(ns));
+       up_read(&msg_ids(ns).rwsem);
+       if (cmd == MSG_INFO) {
+               msginfo->msgmap = min_t(int,
+                                    percpu_counter_sum(&ns->percpu_msg_hdrs),
+                                    INT_MAX);
+               msginfo->msgtql = min_t(int,
+                                    percpu_counter_sum(&ns->percpu_msg_bytes),
+                                    INT_MAX);
        } else {
                msginfo->msgmap = MSGMAP;
                msginfo->msgpool = MSGPOOL;
                msginfo->msgtql = MSGTQL;
        }
-       max_idx = ipc_get_maxidx(&msg_ids(ns));
-       up_read(&msg_ids(ns).rwsem);
        return (max_idx < 0) ? 0 : max_idx;
 }
 
@@ -935,8 +941,8 @@ static long do_msgsnd(int msqid, long mtype, void __user *mtext,
                list_add_tail(&msg->m_list, &msq->q_messages);
                msq->q_cbytes += msgsz;
                msq->q_qnum++;
-               atomic_add(msgsz, &ns->msg_bytes);
-               atomic_inc(&ns->msg_hdrs);
+               percpu_counter_add_local(&ns->percpu_msg_bytes, msgsz);
+               percpu_counter_add_local(&ns->percpu_msg_hdrs, 1);
        }
 
        err = 0;
@@ -1159,8 +1165,8 @@ static long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp, in
                        msq->q_rtime = ktime_get_real_seconds();
                        ipc_update_pid(&msq->q_lrpid, task_tgid(current));
                        msq->q_cbytes -= msg->m_ts;
-                       atomic_sub(msg->m_ts, &ns->msg_bytes);
-                       atomic_dec(&ns->msg_hdrs);
+                       percpu_counter_sub_local(&ns->percpu_msg_bytes, msg->m_ts);
+                       percpu_counter_sub_local(&ns->percpu_msg_hdrs, 1);
                        ss_wakeup(msq, &wake_q, false);
 
                        goto out_unlock0;
@@ -1297,20 +1303,34 @@ COMPAT_SYSCALL_DEFINE5(msgrcv, int, msqid, compat_uptr_t, msgp,
 }
 #endif
 
-void msg_init_ns(struct ipc_namespace *ns)
+int msg_init_ns(struct ipc_namespace *ns)
 {
+       int ret;
+
        ns->msg_ctlmax = MSGMAX;
        ns->msg_ctlmnb = MSGMNB;
        ns->msg_ctlmni = MSGMNI;
 
-       atomic_set(&ns->msg_bytes, 0);
-       atomic_set(&ns->msg_hdrs, 0);
+       ret = percpu_counter_init(&ns->percpu_msg_bytes, 0, GFP_KERNEL);
+       if (ret)
+               goto fail_msg_bytes;
+       ret = percpu_counter_init(&ns->percpu_msg_hdrs, 0, GFP_KERNEL);
+       if (ret)
+               goto fail_msg_hdrs;
        ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
+       return 0;
+
+fail_msg_hdrs:
+       percpu_counter_destroy(&ns->percpu_msg_bytes);
+fail_msg_bytes:
+       return ret;
 }
 
 #ifdef CONFIG_IPC_NS
 void msg_exit_ns(struct ipc_namespace *ns)
 {
+       percpu_counter_destroy(&ns->percpu_msg_bytes);
+       percpu_counter_destroy(&ns->percpu_msg_hdrs);
        free_ipcs(ns, &msg_ids(ns), freeque);
        idr_destroy(&ns->ids[IPC_MSG_IDS].ipcs_idr);
        rhashtable_destroy(&ns->ids[IPC_MSG_IDS].key_ht);

diff --git a/ipc/namespace.c b/ipc/namespace.c
index e1fcaed..8316ea5 100644 (file)
--- a/ipc/namespace.c
+++ b/ipc/namespace.c
@@ -66,8 +66,11 @@ static struct ipc_namespace *create_ipc_ns(struct user_namespace *user_ns,
        if (!setup_ipc_sysctls(ns))
                goto fail_mq;
 
+       err = msg_init_ns(ns);
+       if (err)
+               goto fail_put;
+
        sem_init_ns(ns);
-       msg_init_ns(ns);
        shm_init_ns(ns);
 
        return ns;

diff --git a/ipc/util.h b/ipc/util.h
index 2dd7ce0..b2906e3 100644 (file)
--- a/ipc/util.h
+++ b/ipc/util.h
@@ -64,7 +64,7 @@ static inline void mq_put_mnt(struct ipc_namespace *ns) { }
 
 #ifdef CONFIG_SYSVIPC
 void sem_init_ns(struct ipc_namespace *ns);
-void msg_init_ns(struct ipc_namespace *ns);
+int msg_init_ns(struct ipc_namespace *ns);
 void shm_init_ns(struct ipc_namespace *ns);
 
 void sem_exit_ns(struct ipc_namespace *ns);
@@ -72,7 +72,7 @@ void msg_exit_ns(struct ipc_namespace *ns);
 void shm_exit_ns(struct ipc_namespace *ns);
 #else
 static inline void sem_init_ns(struct ipc_namespace *ns) { }
-static inline void msg_init_ns(struct ipc_namespace *ns) { }
+static inline int msg_init_ns(struct ipc_namespace *ns) { return 0; }
 static inline void shm_init_ns(struct ipc_namespace *ns) { }
 
 static inline void sem_exit_ns(struct ipc_namespace *ns) { }