Patch "crypto: mcryptd - protect the per-CPU queue with a lock" has been added to the 3.18-stable tree

This is a note to let you know that I've just added the patch titled

crypto: mcryptd - protect the per-CPU queue with a lock

to the 3.18-stable tree which can be found at:
http://www.kernel.org/git/?p=linux/kernel/git/stable/stable-queue.git;a=summary

The filename of the patch is:
crypto-mcryptd-protect-the-per-cpu-queue-with-a-lock.patch
and it can be found in the queue-3.18 subdirectory.

If you, or anyone else, feels it should not be added to the stable tree, please let <[email protected]> know about it.

From: Sebastian Andrzej Siewior <[email protected]>
Date: Thu, 30 Nov 2017 13:39:27 +0100
Subject: crypto: mcryptd - protect the per-CPU queue with a lock

From: Sebastian Andrzej Siewior <[email protected]>

commit 9abffc6f2efe46c3564c04312e52e07622d40e51 upstream.

mcryptd_enqueue_request() grabs the per-CPU queue struct and protects access to it with disabled preemption. Then it schedules a worker on the same CPU. The worker in mcryptd_queue_worker() guards access to the same per-CPU variable with disabled preemption.

If we take CPU-hotplug into account then it is possible that between queue_work_on() and the actual invocation of the worker the CPU goes down and the worker will be scheduled on _another_ CPU. And here the preempt_disable() protection does not work anymore. The easiest thing is to add a spin_lock() to guard access to the list.

Another detail: mcryptd_queue_worker() is not processing more than MCRYPTD_BATCH invocation in a row. If there are still items left, then it will invoke queue_work() to proceed with more later. *I* would suggest to simply drop that check because it does not use a system workqueue and the workqueue is already marked as "CPU_INTENSIVE". And if preemption is required then the scheduler should do it. However if queue_work() is used then the work item is marked as CPU unbound. That means it will try to run on the local CPU but it may run on another CPU as well. Especially with CONFIG_DEBUG_WQ_FORCE_RR_CPU=y. Again, the preempt_disable() won't work here but lock which was introduced will help.
In order to keep work-item on the local CPU (and avoid RR) I changed it to queue_work_on().

---
 crypto/mcryptd.c         |   23 ++++++++++-------------
 include/crypto/mcryptd.h |    1 +
 2 files changed, 11 insertions(+), 13 deletions(-)

@@ -80,6 +80,7 @@ static int mcryptd_init_queue(struct mcr
 		pr_debug("cpu_queue #%d %p\n", cpu, queue->cpu_queue);
 		crypto_init_queue(&cpu_queue->queue, max_cpu_qlen);
 		INIT_WORK(&cpu_queue->work, mcryptd_queue_worker);
+		spin_lock_init(&cpu_queue->q_lock);
 	}
 	return 0;
 }

@@ -103,15 +104,16 @@ static int mcryptd_enqueue_request(struc
 	int cpu, err;
 	struct mcryptd_cpu_queue *cpu_queue;
 
-	cpu = get_cpu();
-	cpu_queue = this_cpu_ptr(queue->cpu_queue);
-	rctx->tag.cpu = cpu;
+	cpu_queue = raw_cpu_ptr(queue->cpu_queue);
+	spin_lock(&cpu_queue->q_lock);
+	cpu = smp_processor_id();
+	rctx->tag.cpu = smp_processor_id();
 
 	err = crypto_enqueue_request(&cpu_queue->queue, request);
 	pr_debug("enqueue request: cpu %d cpu_queue %p request %p\n",
 		 cpu, cpu_queue, request);
+	spin_unlock(&cpu_queue->q_lock);
 	queue_work_on(cpu, kcrypto_wq, &cpu_queue->work);
-	put_cpu();
 
 	return err;
 }

@@ -164,16 +166,11 @@ static void mcryptd_queue_worker(struct
 	cpu_queue = container_of(work, struct mcryptd_cpu_queue, work);
 	i = 0;
 	while (i < MCRYPTD_BATCH || single_task_running()) {
-		/*
-		 * preempt_disable/enable is used to prevent
-		 * being preempted by mcryptd_enqueue_request()
-		 */
-		local_bh_disable();
-		preempt_disable();
+
+		spin_lock_bh(&cpu_queue->q_lock);
 		backlog = crypto_get_backlog(&cpu_queue->queue);
 		req = crypto_dequeue_request(&cpu_queue->queue);
-		preempt_enable();
-		local_bh_enable();
+		spin_unlock_bh(&cpu_queue->q_lock);
 
 		if (!req) {
 			mcryptd_opportunistic_flush();

@@ -188,7 +185,7 @@ static void mcryptd_queue_worker(struct
 		++i;
 	}
 	if (cpu_queue->queue.qlen)
-		queue_work(kcrypto_wq, &cpu_queue->work);
+		queue_work_on(smp_processor_id(), kcrypto_wq, &cpu_queue->work);
 }
 
 void mcryptd_flusher(struct work_struct *__work)

@@ -26,6 +26,7 @@ static inline struct mcryptd_ahash *__mc
 
 struct mcryptd_cpu_queue {
 	struct crypto_queue queue;
+	spinlock_t q_lock;
 	struct work_struct work;
 };
 

Patches currently in stable-queue which might be from [email protected] are

queue-3.18/crypto-mcryptd-protect-the-per-cpu-queue-with-a-lock.patch