Skip to content

Commit

Permalink
net: Convert NAPI gro list into a small hash table.
Browse files Browse the repository at this point in the history 
Improve the performance of GRO receive by splitting flows into
multiple hash chains.

Suggested-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
  • Loading branch information
David Miller committed Jun 26, 2018
1 parent d4546c2 commit 07d7836
Show file tree
Hide file tree
Showing 2 changed files with 81 additions and 27 deletions.
3 changes: 2 additions & 1 deletion include/linux/netdevice.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -305,6 +305,7 @@ int __init netdev_boot_setup(char *str);
/*
* Structure for NAPI scheduling similar to tasklet but with weighting
*/
#define GRO_HASH_BUCKETS 8
struct napi_struct {
/* The poll_list must only be managed by the entity which
* changes the state of the NAPI_STATE_SCHED bit. This means
Expand All @@ -322,7 +323,7 @@ struct napi_struct {
int poll_owner;
#endif
struct net_device *dev;
struct list_head gro_list;
struct list_head gro_hash[GRO_HASH_BUCKETS];
struct sk_buff *skb;
struct hrtimer timer;
struct list_head dev_list;
Expand Down
105 changes: 79 additions & 26 deletions net/core/dev.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -4875,31 +4875,46 @@ static int napi_gro_complete(struct sk_buff *skb)
return netif_receive_skb_internal(skb);
}

/* napi->gro_list contains packets ordered by age.
* youngest packets at the head of it.
* Complete skbs in reverse order to reduce latencies.
*/
void napi_gro_flush(struct napi_struct *napi, bool flush_old)
static void __napi_gro_flush_chain(struct napi_struct *napi, struct list_head *head,
bool flush_old)
{
struct sk_buff *skb, *p;

list_for_each_entry_safe_reverse(skb, p, &napi->gro_list, list) {
list_for_each_entry_safe_reverse(skb, p, head, list) {
if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
return;
list_del_init(&skb->list);
napi_gro_complete(skb);
napi->gro_count--;
}
}

/* napi->gro_hash contains packets ordered by age.
* youngest packets at the head of it.
* Complete skbs in reverse order to reduce latencies.
*/
void napi_gro_flush(struct napi_struct *napi, bool flush_old)
{
int i;

for (i = 0; i < GRO_HASH_BUCKETS; i++) {
struct list_head *head = &napi->gro_hash[i];

__napi_gro_flush_chain(napi, head, flush_old);
}
}
EXPORT_SYMBOL(napi_gro_flush);

static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
static struct list_head *gro_list_prepare(struct napi_struct *napi,
struct sk_buff *skb)
{
unsigned int maclen = skb->dev->hard_header_len;
u32 hash = skb_get_hash_raw(skb);
struct list_head *head;
struct sk_buff *p;

list_for_each_entry(p, &napi->gro_list, list) {
head = &napi->gro_hash[hash & (GRO_HASH_BUCKETS - 1)];
list_for_each_entry(p, head, list) {
unsigned long diffs;

NAPI_GRO_CB(p)->flush = 0;
Expand All @@ -4922,6 +4937,8 @@ static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
maclen);
NAPI_GRO_CB(p)->same_flow = !diffs;
}

return head;
}

static void skb_gro_reset_offset(struct sk_buff *skb)
Expand Down Expand Up @@ -4964,11 +4981,45 @@ static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
}
}

static void gro_flush_oldest(struct napi_struct *napi)
{
struct sk_buff *oldest = NULL;
unsigned long age = jiffies;
int i;

for (i = 0; i < GRO_HASH_BUCKETS; i++) {
struct list_head *head = &napi->gro_hash[i];
struct sk_buff *skb;

if (list_empty(head))
continue;

skb = list_last_entry(head, struct sk_buff, list);
if (!oldest || time_before(NAPI_GRO_CB(skb)->age, age)) {
oldest = skb;
age = NAPI_GRO_CB(skb)->age;
}
}

/* We are called with napi->gro_count >= MAX_GRO_SKBS, so this is
* impossible.
*/
if (WARN_ON_ONCE(!oldest))
return;

/* Do not adjust napi->gro_count, caller is adding a new SKB to
* the chain.
*/
list_del(&oldest->list);
napi_gro_complete(oldest);
}

static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
{
struct list_head *head = &offload_base;
struct packet_offload *ptype;
__be16 type = skb->protocol;
struct list_head *gro_head;
struct sk_buff *pp = NULL;
enum gro_result ret;
int same_flow;
Expand All @@ -4977,7 +5028,7 @@ static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff
if (netif_elide_gro(skb->dev))
goto normal;

gro_list_prepare(napi, skb);
gro_head = gro_list_prepare(napi, skb);

rcu_read_lock();
list_for_each_entry_rcu(ptype, head, list) {
Expand Down Expand Up @@ -5011,7 +5062,7 @@ static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff
NAPI_GRO_CB(skb)->csum_valid = 0;
}

pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
pp = ptype->callbacks.gro_receive(gro_head, skb);
break;
}
rcu_read_unlock();
Expand Down Expand Up @@ -5040,19 +5091,15 @@ static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff
goto normal;

if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
struct sk_buff *nskb;

nskb = list_last_entry(&napi->gro_list, struct sk_buff, list);
list_del(&nskb->list);
napi_gro_complete(nskb);
gro_flush_oldest(napi);
} else {
napi->gro_count++;
}
NAPI_GRO_CB(skb)->count = 1;
NAPI_GRO_CB(skb)->age = jiffies;
NAPI_GRO_CB(skb)->last = skb;
skb_shinfo(skb)->gso_size = skb_gro_len(skb);
list_add(&skb->list, &napi->gro_list);
list_add(&skb->list, gro_head);
ret = GRO_HELD;

pull:
Expand Down Expand Up @@ -5458,7 +5505,7 @@ bool napi_complete_done(struct napi_struct *n, int work_done)
NAPIF_STATE_IN_BUSY_POLL)))
return false;

if (!list_empty(&n->gro_list)) {
if (n->gro_count) {
unsigned long timeout = 0;

if (work_done)
Expand Down Expand Up @@ -5667,7 +5714,7 @@ static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
/* Note : we use a relaxed variant of napi_schedule_prep() not setting
* NAPI_STATE_MISSED, since we do not react to a device IRQ.
*/
if (!list_empty(&napi->gro_list) && !napi_disable_pending(napi) &&
if (napi->gro_count && !napi_disable_pending(napi) &&
!test_and_set_bit(NAPI_STATE_SCHED, &napi->state))
__napi_schedule_irqoff(napi);

Expand All @@ -5677,11 +5724,14 @@ static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
int (*poll)(struct napi_struct *, int), int weight)
{
int i;

INIT_LIST_HEAD(&napi->poll_list);
hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
napi->timer.function = napi_watchdog;
napi->gro_count = 0;
INIT_LIST_HEAD(&napi->gro_list);
for (i = 0; i < GRO_HASH_BUCKETS; i++)
INIT_LIST_HEAD(&napi->gro_hash[i]);
napi->skb = NULL;
napi->poll = poll;
if (weight > NAPI_POLL_WEIGHT)
Expand Down Expand Up @@ -5714,12 +5764,16 @@ void napi_disable(struct napi_struct *n)
}
EXPORT_SYMBOL(napi_disable);

static void gro_list_free(struct list_head *head)
static void flush_gro_hash(struct napi_struct *napi)
{
struct sk_buff *skb, *p;
int i;

list_for_each_entry_safe(skb, p, head, list)
kfree_skb(skb);
for (i = 0; i < GRO_HASH_BUCKETS; i++) {
struct sk_buff *skb, *n;

list_for_each_entry_safe(skb, n, &napi->gro_hash[i], list)
kfree_skb(skb);
}
}

/* Must be called in process context */
Expand All @@ -5731,8 +5785,7 @@ void netif_napi_del(struct napi_struct *napi)
list_del_init(&napi->dev_list);
napi_free_frags(napi);

gro_list_free(&napi->gro_list);
INIT_LIST_HEAD(&napi->gro_list);
flush_gro_hash(napi);
napi->gro_count = 0;
}
EXPORT_SYMBOL(netif_napi_del);
Expand Down Expand Up @@ -5775,7 +5828,7 @@ static int napi_poll(struct napi_struct *n, struct list_head *repoll)
goto out_unlock;
}

if (!list_empty(&n->gro_list)) {
if (n->gro_count) {
/* flush too old packets
* If HZ < 1000, flush all packets.
*/
Expand Down

0 comments on commit 07d7836

Please sign in to comment.