If the VIRTIO_RING_F_INDIRECT_DESC negotiation succeeds, and the number of sgs used for sending packets is greater than 1. We must constantly call __kmalloc/kfree to allocate/release desc. In the case of extremely fast package delivery, the overhead cannot be ignored: 27.46% [kernel] [k] virtqueue_add 16.66% [kernel] [k] detach_buf_split 16.51% [kernel] [k] virtnet_xsk_xmit 14.04% [kernel] [k] virtqueue_add_outbuf 5.18% [kernel] [k] __kmalloc 4.08% [kernel] [k] kfree 2.80% [kernel] [k] virtqueue_get_buf_ctx 2.22% [kernel] [k] xsk_tx_peek_desc 2.08% [kernel] [k] memset_erms 0.83% [kernel] [k] virtqueue_kick_prepare 0.76% [kernel] [k] virtnet_xsk_run 0.62% [kernel] [k] __free_old_xmit_ptr 0.60% [kernel] [k] vring_map_one_sg 0.53% [kernel] [k] native_apic_mem_write 0.46% [kernel] [k] sg_next 0.43% [kernel] [k] sg_init_table 0.41% [kernel] [k] kmalloc_slab This patch adds a cache function to virtio to cache these allocated indirect desc instead of constantly allocating and releasing desc. Xuan Zhuo (3): virtio: cache indirect desc for split virtio: cache indirect desc for packed virtio-net: enable virtio indirect cache drivers/net/virtio_net.c | 4 ++ drivers/virtio/virtio.c | 6 ++ drivers/virtio/virtio_ring.c | 120 ++++++++++++++++++++++++++++++----- include/linux/virtio.h | 10 +++ 4 files changed, 123 insertions(+), 17 deletions(-) -- 2.31.0
In the case of using indirect, indirect desc must be allocated and
released each time, which increases a lot of cpu overhead.
Here, a cache is added for indirect. If the number of indirect desc to be
applied for is less than VIRT_QUEUE_CACHE_DESC_NUM, the desc array with
the size of VIRT_QUEUE_CACHE_DESC_NUM is fixed and cached for reuse.
Signed-off-by: Xuan Zhuo <xuanzhuo at linux.alibaba.com>
---
drivers/virtio/virtio.c | 6 ++++
drivers/virtio/virtio_ring.c | 63 ++++++++++++++++++++++++++++++------
include/linux/virtio.h | 10 ++++++
3 files changed, 70 insertions(+), 9 deletions(-)
diff --git a/drivers/virtio/virtio.c b/drivers/virtio/virtio.c
index 0a5b54034d4b..04bcb74e5b9a 100644
--- a/drivers/virtio/virtio.c
+++ b/drivers/virtio/virtio.c
@@ -431,6 +431,12 @@ bool is_virtio_device(struct device *dev)
}
EXPORT_SYMBOL_GPL(is_virtio_device);
+void virtio_use_desc_cache(struct virtio_device *dev, bool val)
+{
+ dev->desc_cache = val;
+}
+EXPORT_SYMBOL_GPL(virtio_use_desc_cache);
+
void unregister_virtio_device(struct virtio_device *dev)
{
int index = dev->index; /* save for after device release */
diff --git a/drivers/virtio/virtio_ring.c b/drivers/virtio/virtio_ring.c
index dd95dfd85e98..0b9a8544b0e8 100644
--- a/drivers/virtio/virtio_ring.c
+++ b/drivers/virtio/virtio_ring.c
@@ -117,6 +117,10 @@ struct vring_virtqueue {
/* Hint for event idx: already triggered no need to disable. */
bool event_triggered;
+ /* Is indirect cache used? */
+ bool use_desc_cache;
+ void *desc_cache_chain;
+
union {
/* Available for split ring */
struct {
@@ -423,12 +427,47 @@ static unsigned int vring_unmap_one_split(const struct
vring_virtqueue *vq,
return extra[i].next;
}
-static struct vring_desc *alloc_indirect_split(struct virtqueue *_vq,
+#define VIRT_QUEUE_CACHE_DESC_NUM 4
+
+static void desc_cache_chain_free_split(void *chain)
+{
+ struct vring_desc *desc;
+
+ while (chain) {
+ desc = chain;
+ chain = (void *)desc->addr;
+ kfree(desc);
+ }
+}
+
+static void desc_cache_put_split(struct vring_virtqueue *vq,
+ struct vring_desc *desc, int n)
+{
+ if (vq->use_desc_cache && n <= VIRT_QUEUE_CACHE_DESC_NUM) {
+ desc->addr = (u64)vq->desc_cache_chain;
+ vq->desc_cache_chain = desc;
+ } else {
+ kfree(desc);
+ }
+}
+
+static struct vring_desc *alloc_indirect_split(struct vring_virtqueue *vq,
unsigned int total_sg,
gfp_t gfp)
{
struct vring_desc *desc;
- unsigned int i;
+ unsigned int i, n;
+
+ if (vq->use_desc_cache && total_sg <= VIRT_QUEUE_CACHE_DESC_NUM)
{
+ if (vq->desc_cache_chain) {
+ desc = vq->desc_cache_chain;
+ vq->desc_cache_chain = (void *)desc->addr;
+ goto got;
+ }
+ n = VIRT_QUEUE_CACHE_DESC_NUM;
+ } else {
+ n = total_sg;
+ }
/*
* We require lowmem mappings for the descriptors because
@@ -437,12 +476,13 @@ static struct vring_desc *alloc_indirect_split(struct
virtqueue *_vq,
*/
gfp &= ~__GFP_HIGHMEM;
- desc = kmalloc_array(total_sg, sizeof(struct vring_desc), gfp);
+ desc = kmalloc_array(n, sizeof(struct vring_desc), gfp);
if (!desc)
return NULL;
+got:
for (i = 0; i < total_sg; i++)
- desc[i].next = cpu_to_virtio16(_vq->vdev, i + 1);
+ desc[i].next = cpu_to_virtio16(vq->vq.vdev, i + 1);
return desc;
}
@@ -508,7 +548,7 @@ static inline int virtqueue_add_split(struct virtqueue *_vq,
head = vq->free_head;
if (virtqueue_use_indirect(_vq, total_sg))
- desc = alloc_indirect_split(_vq, total_sg, gfp);
+ desc = alloc_indirect_split(vq, total_sg, gfp);
else {
desc = NULL;
WARN_ON_ONCE(total_sg > vq->split.vring.num &&
!vq->indirect);
@@ -652,7 +692,7 @@ static inline int virtqueue_add_split(struct virtqueue *_vq,
}
if (indirect)
- kfree(desc);
+ desc_cache_put_split(vq, desc, total_sg);
END_USE(vq);
return -ENOMEM;
@@ -717,7 +757,7 @@ static void detach_buf_split(struct vring_virtqueue *vq,
unsigned int head,
if (vq->indirect) {
struct vring_desc *indir_desc vq->split.desc_state[head].indir_desc;
- u32 len;
+ u32 len, n;
/* Free the indirect table, if any, now that it's unmapped. */
if (!indir_desc)
@@ -729,10 +769,12 @@ static void detach_buf_split(struct vring_virtqueue *vq,
unsigned int head,
VRING_DESC_F_INDIRECT));
BUG_ON(len == 0 || len % sizeof(struct vring_desc));
- for (j = 0; j < len / sizeof(struct vring_desc); j++)
+ n = len / sizeof(struct vring_desc);
+
+ for (j = 0; j < n; j++)
vring_unmap_one_split_indirect(vq, &indir_desc[j]);
- kfree(indir_desc);
+ desc_cache_put_split(vq, indir_desc, n);
vq->split.desc_state[head].indir_desc = NULL;
} else if (ctx) {
*ctx = vq->split.desc_state[head].indir_desc;
@@ -2199,6 +2241,8 @@ struct virtqueue *__vring_new_virtqueue(unsigned int
index,
vq->indirect = virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC)
&&
!context;
vq->event = virtio_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX);
+ vq->desc_cache_chain = NULL;
+ vq->use_desc_cache = vdev->desc_cache;
if (virtio_has_feature(vdev, VIRTIO_F_ORDER_PLATFORM))
vq->weak_barriers = false;
@@ -2329,6 +2373,7 @@ void vring_del_virtqueue(struct virtqueue *_vq)
if (!vq->packed_ring) {
kfree(vq->split.desc_state);
kfree(vq->split.desc_extra);
+ desc_cache_chain_free_split(vq->desc_cache_chain);
}
kfree(vq);
}
diff --git a/include/linux/virtio.h b/include/linux/virtio.h
index 41edbc01ffa4..d84b7b8f4070 100644
--- a/include/linux/virtio.h
+++ b/include/linux/virtio.h
@@ -109,6 +109,7 @@ struct virtio_device {
bool failed;
bool config_enabled;
bool config_change_pending;
+ bool desc_cache;
spinlock_t config_lock;
spinlock_t vqs_list_lock; /* Protects VQs list access */
struct device dev;
@@ -130,6 +131,15 @@ int register_virtio_device(struct virtio_device *dev);
void unregister_virtio_device(struct virtio_device *dev);
bool is_virtio_device(struct device *dev);
+/**
+ * virtio_use_desc_cache - virtio ring use desc cache
+ *
+ * virtio will cache the allocated indirect desc.
+ *
+ * This function must be called before find_vqs.
+ */
+void virtio_use_desc_cache(struct virtio_device *dev, bool val);
+
void virtio_break_device(struct virtio_device *dev);
void virtio_config_changed(struct virtio_device *dev);
--
2.31.0
In the case of using indirect, indirect desc must be allocated and
released each time, which increases a lot of cpu overhead.
Here, a cache is added for indirect. If the number of indirect desc to be
applied for is less than VIRT_QUEUE_CACHE_DESC_NUM, the desc array with
the size of VIRT_QUEUE_CACHE_DESC_NUM is fixed and cached for reuse.
Signed-off-by: Xuan Zhuo <xuanzhuo at linux.alibaba.com>
---
drivers/virtio/virtio_ring.c | 57 +++++++++++++++++++++++++++++++-----
1 file changed, 49 insertions(+), 8 deletions(-)
diff --git a/drivers/virtio/virtio_ring.c b/drivers/virtio/virtio_ring.c
index 0b9a8544b0e8..4fd7bd5bcd70 100644
--- a/drivers/virtio/virtio_ring.c
+++ b/drivers/virtio/virtio_ring.c
@@ -1074,10 +1074,45 @@ static void vring_unmap_desc_packed(const struct
vring_virtqueue *vq,
}
}
-static struct vring_packed_desc *alloc_indirect_packed(unsigned int total_sg,
+static void desc_cache_chain_free_packed(void *chain)
+{
+ struct vring_packed_desc *desc;
+
+ while (chain) {
+ desc = chain;
+ chain = (void *)desc->addr;
+ kfree(desc);
+ }
+}
+
+static void desc_cache_put_packed(struct vring_virtqueue *vq,
+ struct vring_packed_desc *desc, int n)
+{
+ if (vq->use_desc_cache && n <= VIRT_QUEUE_CACHE_DESC_NUM) {
+ desc->addr = (u64)vq->desc_cache_chain;
+ vq->desc_cache_chain = desc;
+ } else {
+ kfree(desc);
+ }
+}
+
+static struct vring_packed_desc *alloc_indirect_packed(struct vring_virtqueue
*vq,
+ unsigned int total_sg,
gfp_t gfp)
{
struct vring_packed_desc *desc;
+ unsigned int n;
+
+ if (vq->use_desc_cache && total_sg <= VIRT_QUEUE_CACHE_DESC_NUM)
{
+ if (vq->desc_cache_chain) {
+ desc = vq->desc_cache_chain;
+ vq->desc_cache_chain = (void *)desc->addr;
+ return desc;
+ }
+ n = VIRT_QUEUE_CACHE_DESC_NUM;
+ } else {
+ n = total_sg;
+ }
/*
* We require lowmem mappings for the descriptors because
@@ -1086,7 +1121,7 @@ static struct vring_packed_desc
*alloc_indirect_packed(unsigned int total_sg,
*/
gfp &= ~__GFP_HIGHMEM;
- desc = kmalloc_array(total_sg, sizeof(struct vring_packed_desc), gfp);
+ desc = kmalloc_array(n, sizeof(struct vring_packed_desc), gfp);
return desc;
}
@@ -1106,7 +1141,7 @@ static int virtqueue_add_indirect_packed(struct
vring_virtqueue *vq,
dma_addr_t addr;
head = vq->packed.next_avail_idx;
- desc = alloc_indirect_packed(total_sg, gfp);
+ desc = alloc_indirect_packed(vq, total_sg, gfp);
if (unlikely(vq->vq.num_free < 1)) {
pr_debug("Can't add buf len 1 - avail = 0\n");
@@ -1197,7 +1232,7 @@ static int virtqueue_add_indirect_packed(struct
vring_virtqueue *vq,
for (i = 0; i < err_idx; i++)
vring_unmap_desc_packed(vq, &desc[i]);
- kfree(desc);
+ desc_cache_put_packed(vq, desc, total_sg);
END_USE(vq);
return -ENOMEM;
@@ -1422,20 +1457,22 @@ static void detach_buf_packed(struct vring_virtqueue
*vq,
}
if (vq->indirect) {
- u32 len;
+ u32 len, n;
/* Free the indirect table, if any, now that it's unmapped. */
desc = state->indir_desc;
if (!desc)
return;
+ n = len / sizeof(struct vring_packed_desc);
+
if (vq->use_dma_api) {
len = vq->packed.desc_extra[id].len;
- for (i = 0; i < len / sizeof(struct vring_packed_desc);
- i++)
+ for (i = 0; i < n; i++)
vring_unmap_desc_packed(vq, &desc[i]);
}
- kfree(desc);
+
+ desc_cache_put_packed(vq, desc, n);
state->indir_desc = NULL;
} else if (ctx) {
*ctx = state->indir_desc;
@@ -1753,6 +1790,8 @@ static struct virtqueue *vring_create_virtqueue_packed(
vq->indirect = virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC)
&&
!context;
vq->event = virtio_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX);
+ vq->desc_cache_chain = NULL;
+ vq->use_desc_cache = vdev->desc_cache;
if (virtio_has_feature(vdev, VIRTIO_F_ORDER_PLATFORM))
vq->weak_barriers = false;
@@ -2374,6 +2413,8 @@ void vring_del_virtqueue(struct virtqueue *_vq)
kfree(vq->split.desc_state);
kfree(vq->split.desc_extra);
desc_cache_chain_free_split(vq->desc_cache_chain);
+ } else {
+ desc_cache_chain_free_packed(vq->desc_cache_chain);
}
kfree(vq);
}
--
2.31.0
If the VIRTIO_RING_F_INDIRECT_DESC negotiation succeeds, and the number
of sgs used for sending packets is greater than 1. We must constantly
call __kmalloc/kfree to allocate/release desc.
In the case of extremely fast package delivery, the overhead cannot be
ignored:
27.46% [kernel] [k] virtqueue_add
16.66% [kernel] [k] detach_buf_split
16.51% [kernel] [k] virtnet_xsk_xmit
14.04% [kernel] [k] virtqueue_add_outbuf
5.18% [kernel] [k] __kmalloc
4.08% [kernel] [k] kfree
2.80% [kernel] [k] virtqueue_get_buf_ctx
2.22% [kernel] [k] xsk_tx_peek_desc
2.08% [kernel] [k] memset_erms
0.83% [kernel] [k] virtqueue_kick_prepare
0.76% [kernel] [k] virtnet_xsk_run
0.62% [kernel] [k] __free_old_xmit_ptr
0.60% [kernel] [k] vring_map_one_sg
0.53% [kernel] [k] native_apic_mem_write
0.46% [kernel] [k] sg_next
0.43% [kernel] [k] sg_init_table
0.41% [kernel] [k] kmalloc_slab
Compared to not using virtio indirect cache, virtio-net can get a 16%
performance improvement when using indirect desc cache.
In the test case, the CPU where the package is sent has reached 100%.
The following are the PPS in two cases:
indirect desc cache | no cache
3074658 | 2685132
3111866 | 2666118
3152527 | 2653632
3125867 | 2669820
3027147 | 2644464
3069211 | 2669777
3038522 | 2675645
3034507 | 2671302
3102257 | 2685504
3083712 | 2692800
3051771 | 2676928
3080684 | 2695040
3147816 | 2720876
3123887 | 2705492
3180963 | 2699520
3191579 | 2676480
3161670 | 2686272
3189768 | 2692588
3174272 | 2686692
3143434 | 2682416
Signed-off-by: Xuan Zhuo <xuanzhuo at linux.alibaba.com>
---
drivers/net/virtio_net.c | 4 ++++
1 file changed, 4 insertions(+)
diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index 4ad25a8b0870..0ec29cf90d0a 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c
@@ -27,9 +27,11 @@ static int napi_weight = NAPI_POLL_WEIGHT;
module_param(napi_weight, int, 0444);
static bool csum = true, gso = true, napi_tx = true;
+static bool virtio_desc_cache = true;
module_param(csum, bool, 0444);
module_param(gso, bool, 0444);
module_param(napi_tx, bool, 0644);
+module_param(virtio_desc_cache, bool, 0644);
/* FIXME: MTU in config. */
#define GOOD_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN)
@@ -3214,6 +3216,8 @@ static int virtnet_probe(struct virtio_device *vdev)
vi->curr_queue_pairs = num_online_cpus();
vi->max_queue_pairs = max_queue_pairs;
+ virtio_use_desc_cache(vdev, virtio_desc_cache);
+
/* Allocate/initialize the rx/tx queues, and invoke find_vqs */
err = init_vqs(vi);
if (err)
--
2.31.0