From: Bernard Metzler <bmt@zurich.ibm.com>
Date: Fri, 28 Jul 2023 13:44:18 +0200
Subject: RDMA/siw: Fix tx thread initialization.
Patch-mainline: v6.6-rc1
Git-commit: 91f36237b4b9bdce7610c7450a906d46704a566a
References: jsc#PED-6864

Immediately removing the siw module after insertion may
crash in siw_stop_tx_thread(), if the according thread did
not yet had a chance to initialize its wait queue and
siw_stop_tx_thread() tries to wakeup that thread. Initializing
the threads state before spwaning it fixes it.

Reported-by: Guoqing Jiang <guoqing.jiang@linux.dev>
Signed-off-by: Bernard Metzler <bmt@zurich.ibm.com>
Link: https://lore.kernel.org/r/20230728114418.124328-1-bmt@zurich.ibm.com
Tested-by: Guoqing Jiang <guoqing.jiang@linux.dev>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Acked-by: Thomas Bogendoerfer <tbogendoerfer@suse.de>
---
 drivers/infiniband/sw/siw/siw.h       |    3 +-
 drivers/infiniband/sw/siw/siw_main.c  |   40 ++----------------------------
 drivers/infiniband/sw/siw/siw_qp_tx.c |   44 +++++++++++++++++++++++++++++-----
 3 files changed, 43 insertions(+), 44 deletions(-)

--- a/drivers/infiniband/sw/siw/siw.h
+++ b/drivers/infiniband/sw/siw/siw.h
@@ -531,11 +531,12 @@ void siw_qp_llp_data_ready(struct sock *
 void siw_qp_llp_write_space(struct sock *sk);
 
 /* QP TX path functions */
+int siw_create_tx_threads(void);
+void siw_stop_tx_threads(void);
 int siw_run_sq(void *arg);
 int siw_qp_sq_process(struct siw_qp *qp);
 int siw_sq_start(struct siw_qp *qp);
 int siw_activate_tx(struct siw_qp *qp);
-void siw_stop_tx_thread(int nr_cpu);
 int siw_get_tx_cpu(struct siw_device *sdev);
 void siw_put_tx_cpu(int cpu);
 
--- a/drivers/infiniband/sw/siw/siw_main.c
+++ b/drivers/infiniband/sw/siw/siw_main.c
@@ -87,29 +87,6 @@ static void siw_device_cleanup(struct ib
 	xa_destroy(&sdev->mem_xa);
 }
 
-static int siw_create_tx_threads(void)
-{
-	int cpu, assigned = 0;
-
-	for_each_online_cpu(cpu) {
-		/* Skip HT cores */
-		if (cpu % cpumask_weight(topology_sibling_cpumask(cpu)))
-			continue;
-
-		siw_tx_thread[cpu] =
-			kthread_run_on_cpu(siw_run_sq,
-					   (unsigned long *)(long)cpu,
-					   cpu, "siw_tx/%u");
-		if (IS_ERR(siw_tx_thread[cpu])) {
-			siw_tx_thread[cpu] = NULL;
-			continue;
-		}
-
-		assigned++;
-	}
-	return assigned;
-}
-
 static int siw_dev_qualified(struct net_device *netdev)
 {
 	/*
@@ -529,7 +506,6 @@ static struct rdma_link_ops siw_link_ops
 static __init int siw_init_module(void)
 {
 	int rv;
-	int nr_cpu;
 
 	if (SENDPAGE_THRESH < SIW_MAX_INLINE) {
 		pr_info("siw: sendpage threshold too small: %u\n",
@@ -574,12 +550,8 @@ static __init int siw_init_module(void)
 	return 0;
 
 out_error:
-	for (nr_cpu = 0; nr_cpu < nr_cpu_ids; nr_cpu++) {
-		if (siw_tx_thread[nr_cpu]) {
-			siw_stop_tx_thread(nr_cpu);
-			siw_tx_thread[nr_cpu] = NULL;
-		}
-	}
+	siw_stop_tx_threads();
+
 	if (siw_crypto_shash)
 		crypto_free_shash(siw_crypto_shash);
 
@@ -593,14 +565,8 @@ out_error:
 
 static void __exit siw_exit_module(void)
 {
-	int cpu;
+	siw_stop_tx_threads();
 
-	for_each_possible_cpu(cpu) {
-		if (siw_tx_thread[cpu]) {
-			siw_stop_tx_thread(cpu);
-			siw_tx_thread[cpu] = NULL;
-		}
-	}
 	unregister_netdevice_notifier(&siw_netdev_nb);
 	rdma_link_unregister(&siw_link_ops);
 	ib_unregister_driver(RDMA_DRIVER_SIW);
--- a/drivers/infiniband/sw/siw/siw_qp_tx.c
+++ b/drivers/infiniband/sw/siw/siw_qp_tx.c
@@ -1202,10 +1202,45 @@ struct tx_task_t {
 
 static DEFINE_PER_CPU(struct tx_task_t, siw_tx_task_g);
 
-void siw_stop_tx_thread(int nr_cpu)
+int siw_create_tx_threads(void)
 {
-	kthread_stop(siw_tx_thread[nr_cpu]);
-	wake_up(&per_cpu(siw_tx_task_g, nr_cpu).waiting);
+	int cpu, assigned = 0;
+
+	for_each_online_cpu(cpu) {
+		struct tx_task_t *tx_task;
+
+		/* Skip HT cores */
+		if (cpu % cpumask_weight(topology_sibling_cpumask(cpu)))
+			continue;
+
+		tx_task = &per_cpu(siw_tx_task_g, cpu);
+		init_llist_head(&tx_task->active);
+		init_waitqueue_head(&tx_task->waiting);
+
+		siw_tx_thread[cpu] =
+			kthread_run_on_cpu(siw_run_sq,
+					   (unsigned long *)(long)cpu,
+					   cpu, "siw_tx/%u");
+		if (IS_ERR(siw_tx_thread[cpu])) {
+			siw_tx_thread[cpu] = NULL;
+			continue;
+		}
+		assigned++;
+	}
+	return assigned;
+}
+
+void siw_stop_tx_threads(void)
+{
+	int cpu;
+
+	for_each_possible_cpu(cpu) {
+		if (siw_tx_thread[cpu]) {
+			kthread_stop(siw_tx_thread[cpu]);
+			wake_up(&per_cpu(siw_tx_task_g, cpu).waiting);
+			siw_tx_thread[cpu] = NULL;
+		}
+	}
 }
 
 int siw_run_sq(void *data)
@@ -1215,9 +1250,6 @@ int siw_run_sq(void *data)
 	struct siw_qp *qp;
 	struct tx_task_t *tx_task = &per_cpu(siw_tx_task_g, nr_cpu);
 
-	init_llist_head(&tx_task->active);
-	init_waitqueue_head(&tx_task->waiting);
-
 	while (1) {
 		struct llist_node *fifo_list = NULL;