提交 76cc8b13 编写于 作者: T Tom Herbert 提交者: David S. Miller

net: fix problem in dequeuing from input_pkt_queue

Fix some issues introduced in batch skb dequeuing for input_pkt_queue.
The primary issue it that the queue head must be incremented only
after a packet has been processed, that is only after
__netif_receive_skb has been called.  This is needed for the mechanism
to prevent OOO packet in RFS.  Also when flushing the input_pkt_queue
and process_queue, the process queue should be done first to prevent
OOO packets.

Because the input_pkt_queue has been effectively split into two queues,
the calculation of the tail ptr is no longer correct.  The correct value
would be head+input_pkt_queue->len+process_queue->len.  To avoid
this calculation we added an explict input_queue_tail in softnet_data.
The tail value is simply incremented when queuing to input_pkt_queue.
Signed-off-by: NTom Herbert <therbert@google.com>
Acked-by: NEric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>
上级 1f01bfd2
...@@ -1407,17 +1407,25 @@ struct softnet_data { ...@@ -1407,17 +1407,25 @@ struct softnet_data {
struct softnet_data *rps_ipi_next; struct softnet_data *rps_ipi_next;
unsigned int cpu; unsigned int cpu;
unsigned int input_queue_head; unsigned int input_queue_head;
unsigned int input_queue_tail;
#endif #endif
unsigned dropped; unsigned dropped;
struct sk_buff_head input_pkt_queue; struct sk_buff_head input_pkt_queue;
struct napi_struct backlog; struct napi_struct backlog;
}; };
static inline void input_queue_head_add(struct softnet_data *sd, static inline void input_queue_head_incr(struct softnet_data *sd)
unsigned int len)
{ {
#ifdef CONFIG_RPS #ifdef CONFIG_RPS
sd->input_queue_head += len; sd->input_queue_head++;
#endif
}
static inline void input_queue_tail_incr_save(struct softnet_data *sd,
unsigned int *qtail)
{
#ifdef CONFIG_RPS
*qtail = ++sd->input_queue_tail;
#endif #endif
} }
......
...@@ -2426,10 +2426,7 @@ static int enqueue_to_backlog(struct sk_buff *skb, int cpu, ...@@ -2426,10 +2426,7 @@ static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
if (skb_queue_len(&sd->input_pkt_queue)) { if (skb_queue_len(&sd->input_pkt_queue)) {
enqueue: enqueue:
__skb_queue_tail(&sd->input_pkt_queue, skb); __skb_queue_tail(&sd->input_pkt_queue, skb);
#ifdef CONFIG_RPS input_queue_tail_incr_save(sd, qtail);
*qtail = sd->input_queue_head +
skb_queue_len(&sd->input_pkt_queue);
#endif
rps_unlock(sd); rps_unlock(sd);
local_irq_restore(flags); local_irq_restore(flags);
return NET_RX_SUCCESS; return NET_RX_SUCCESS;
...@@ -2964,7 +2961,7 @@ static void flush_backlog(void *arg) ...@@ -2964,7 +2961,7 @@ static void flush_backlog(void *arg)
if (skb->dev == dev) { if (skb->dev == dev) {
__skb_unlink(skb, &sd->input_pkt_queue); __skb_unlink(skb, &sd->input_pkt_queue);
kfree_skb(skb); kfree_skb(skb);
input_queue_head_add(sd, 1); input_queue_head_incr(sd);
} }
} }
rps_unlock(sd); rps_unlock(sd);
...@@ -2973,6 +2970,7 @@ static void flush_backlog(void *arg) ...@@ -2973,6 +2970,7 @@ static void flush_backlog(void *arg)
if (skb->dev == dev) { if (skb->dev == dev) {
__skb_unlink(skb, &sd->process_queue); __skb_unlink(skb, &sd->process_queue);
kfree_skb(skb); kfree_skb(skb);
input_queue_head_incr(sd);
} }
} }
} }
...@@ -3328,18 +3326,20 @@ static int process_backlog(struct napi_struct *napi, int quota) ...@@ -3328,18 +3326,20 @@ static int process_backlog(struct napi_struct *napi, int quota)
while ((skb = __skb_dequeue(&sd->process_queue))) { while ((skb = __skb_dequeue(&sd->process_queue))) {
local_irq_enable(); local_irq_enable();
__netif_receive_skb(skb); __netif_receive_skb(skb);
if (++work >= quota)
return work;
local_irq_disable(); local_irq_disable();
input_queue_head_incr(sd);
if (++work >= quota) {
local_irq_enable();
return work;
}
} }
rps_lock(sd); rps_lock(sd);
qlen = skb_queue_len(&sd->input_pkt_queue); qlen = skb_queue_len(&sd->input_pkt_queue);
if (qlen) { if (qlen)
input_queue_head_add(sd, qlen);
skb_queue_splice_tail_init(&sd->input_pkt_queue, skb_queue_splice_tail_init(&sd->input_pkt_queue,
&sd->process_queue); &sd->process_queue);
}
if (qlen < quota - work) { if (qlen < quota - work) {
/* /*
* Inline a custom version of __napi_complete(). * Inline a custom version of __napi_complete().
...@@ -5679,12 +5679,14 @@ static int dev_cpu_callback(struct notifier_block *nfb, ...@@ -5679,12 +5679,14 @@ static int dev_cpu_callback(struct notifier_block *nfb,
local_irq_enable(); local_irq_enable();
/* Process offline CPU's input_pkt_queue */ /* Process offline CPU's input_pkt_queue */
while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) { while ((skb = __skb_dequeue(&oldsd->process_queue))) {
netif_rx(skb); netif_rx(skb);
input_queue_head_add(oldsd, 1); input_queue_head_incr(oldsd);
} }
while ((skb = __skb_dequeue(&oldsd->process_queue))) while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
netif_rx(skb); netif_rx(skb);
input_queue_head_incr(oldsd);
}
return NOTIFY_OK; return NOTIFY_OK;
} }
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册