Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
From:   Jiachen Zhang <zhangjiachen.jaycee@bytedance.com>
To:     Vivek Goyal <vgoyal@redhat.com>,
        Stefan Hajnoczi <stefanha@redhat.com>,
        Miklos Szeredi <miklos@szeredi.hu>
Cc:     virtualization@lists.linux-foundation.org,
        linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org,
        Jiachen Zhang <zhangjiachen.jaycee@bytedance.com>,
        Connor Kuehl <ckuehl@redhat.com>
Subject: [PATCH] virtiofs: enable multiple request queues
Date:   Sun, 11 Dec 2022 18:38:57 +0800
Message-Id: <20221211103857.25805-1-zhangjiachen.jaycee@bytedance.com>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Precedence: bulk

Support virtio-fs multiple virtqueues and distribute requests across the
multiqueue complex automatically based on the IRQ affinity.

This commit is based on Connor's patch in the virtio-fs mailing-list,
and additionally intergates cpu-to-vq map into struct virtio_fs so that
this virtio-fs multi-queue feature can fit into multiple virtio-fs mounts.

Link: https://www.mail-archive.com/virtio-fs@redhat.com/msg03320.html
Suggested-by: Stefan Hajnoczi <stefanha@redhat.com>
Cc: Connor Kuehl <ckuehl@redhat.com>
Signed-off-by: Jiachen Zhang <zhangjiachen.jaycee@bytedance.com>
---
 fs/fuse/virtio_fs.c | 37 +++++++++++++++++++++++++++++--------
 1 file changed, 29 insertions(+), 8 deletions(-)

diff --git a/fs/fuse/virtio_fs.c b/fs/fuse/virtio_fs.c
index 4d8d4f16c727..410968dede0c 100644
--- a/fs/fuse/virtio_fs.c
+++ b/fs/fuse/virtio_fs.c
@@ -32,8 +32,9 @@ static DEFINE_MUTEX(virtio_fs_mutex);
 static LIST_HEAD(virtio_fs_instances);
 
 enum {
-	VQ_HIPRIO,
-	VQ_REQUEST
+	VQ_HIPRIO = 0,
+	/* TODO add VQ_NOTIFICATION according to the virtio 1.2 spec. */
+	VQ_REQUEST = 1,
 };
 
 #define VQ_NAME_LEN	24
@@ -59,6 +60,7 @@ struct virtio_fs {
 	struct list_head list;    /* on virtio_fs_instances */
 	char *tag;
 	struct virtio_fs_vq *vqs;
+	struct virtio_fs_vq * __percpu *vq_proxy;
 	unsigned int nvqs;               /* number of virtqueues */
 	unsigned int num_request_queues; /* number of request queues */
 	struct dax_device *dax_dev;
@@ -686,6 +688,7 @@ static int virtio_fs_setup_vqs(struct virtio_device *vdev,
 	struct virtqueue **vqs;
 	vq_callback_t **callbacks;
 	const char **names;
+	struct irq_affinity desc = { .pre_vectors = 1, .nr_sets = 1, };
 	unsigned int i;
 	int ret = 0;
 
@@ -694,11 +697,16 @@ static int virtio_fs_setup_vqs(struct virtio_device *vdev,
 	if (fs->num_request_queues == 0)
 		return -EINVAL;
 
+	fs->num_request_queues = min_t(unsigned int, nr_cpu_ids,
+				       fs->num_request_queues);
+
 	fs->nvqs = VQ_REQUEST + fs->num_request_queues;
 	fs->vqs = kcalloc(fs->nvqs, sizeof(fs->vqs[VQ_HIPRIO]), GFP_KERNEL);
 	if (!fs->vqs)
 		return -ENOMEM;
 
+	pr_debug("virtio-fs: number of vqs: %d\n", fs->nvqs);
+
 	vqs = kmalloc_array(fs->nvqs, sizeof(vqs[VQ_HIPRIO]), GFP_KERNEL);
 	callbacks = kmalloc_array(fs->nvqs, sizeof(callbacks[VQ_HIPRIO]),
 					GFP_KERNEL);
@@ -723,12 +731,26 @@ static int virtio_fs_setup_vqs(struct virtio_device *vdev,
 		names[i] = fs->vqs[i].name;
 	}
 
-	ret = virtio_find_vqs(vdev, fs->nvqs, vqs, callbacks, names, NULL);
+	ret = virtio_find_vqs(vdev, fs->nvqs, vqs, callbacks, names, &desc);
 	if (ret < 0)
 		goto out;
 
-	for (i = 0; i < fs->nvqs; i++)
+	fs->vq_proxy = alloc_percpu(struct virtio_fs_vq *);
+	for (i = 0; i < fs->nvqs; i++) {
+		const struct cpumask *mask;
+		unsigned int cpu;
+
 		fs->vqs[i].vq = vqs[i];
+		if (i == VQ_HIPRIO)
+			continue;
+
+		mask = vdev->config->get_vq_affinity(vdev, i);
+		for_each_cpu(cpu, mask) {
+			struct virtio_fs_vq **cpu_vq = per_cpu_ptr(fs->vq_proxy, cpu);
+			*cpu_vq = &fs->vqs[i];
+			pr_debug("virtio-fs: map cpu %d to vq%d\n", cpu, i);
+		}
+	}
 
 	virtio_fs_start_all_queues(fs);
 out:
@@ -875,8 +897,6 @@ static int virtio_fs_probe(struct virtio_device *vdev)
 	if (ret < 0)
 		goto out;
 
-	/* TODO vq affinity */
-
 	ret = virtio_fs_setup_dax(vdev, fs);
 	if (ret < 0)
 		goto out_vqs;
@@ -926,6 +946,7 @@ static void virtio_fs_remove(struct virtio_device *vdev)
 	virtio_fs_stop_all_queues(fs);
 	virtio_fs_drain_all_queues_locked(fs);
 	virtio_reset_device(vdev);
+	free_percpu(fs->vq_proxy);
 	virtio_fs_cleanup_vqs(vdev);
 
 	vdev->priv = NULL;
@@ -1223,7 +1244,6 @@ static int virtio_fs_enqueue_req(struct virtio_fs_vq *fsvq,
 static void virtio_fs_wake_pending_and_unlock(struct fuse_iqueue *fiq)
 __releases(fiq->lock)
 {
-	unsigned int queue_id = VQ_REQUEST; /* TODO multiqueue */
 	struct virtio_fs *fs;
 	struct fuse_req *req;
 	struct virtio_fs_vq *fsvq;
@@ -1243,7 +1263,8 @@ __releases(fiq->lock)
 		 req->in.h.nodeid, req->in.h.len,
 		 fuse_len_args(req->args->out_numargs, req->args->out_args));
 
-	fsvq = &fs->vqs[queue_id];
+	fsvq = this_cpu_read(*fs->vq_proxy);
+
 	ret = virtio_fs_enqueue_req(fsvq, req, false);
 	if (ret < 0) {
 		if (ret == -ENOMEM || ret == -ENOSPC) {
-- 
2.20.1