hi all:
I read the function of process_backlog, and I found it don't delete the NIC form poll list. the codes is:
4321 static int process_backlog(struct napi_struct *napi, int quota)
4322 {
4323 int work = 0;
4324 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4325
4326 /* Check if we have pending ipi, its better to send them now,
4327 * not waiting net_rx_action() end.
4328 */
4329 if (sd_has_rps_ipi_waiting(sd)) {
4330 local_irq_disable();
4331 net_rps_action_and_irq_enable(sd);
4332 }
4333
4334 napi->weight = weight_p;
4335 local_irq_disable();
4336 while (1) {
4337 struct sk_buff *skb;
4338
4339 while ((skb = __skb_dequeue(&sd->process_queue))) {
4340 local_irq_enable();
4341 __netif_receive_skb(skb);
4342 local_irq_disable();
4343 input_queue_head_incr(sd);
4344 if (++work >= quota) {
4345 local_irq_enable();
4346 return work;
4347 }
4348 }
4349
4350 rps_lock(sd);
4351 if (skb_queue_empty(&sd->input_pkt_queue)) {
4352 /*
4353 * Inline a custom version of __napi_complete().
4354 * only current cpu owns and manipulates this napi,
4355 * and NAPI_STATE_SCHED is the only possible flag set
4356 * on backlog.
4357 * We can use a plain write instead of clear_bit(),
4358 * and we dont need an smp_mb() memory barrier.
4359 */
4360 napi->state = 0;
4361 rps_unlock(sd);
4362
4363 break;
4364 }
4365
4366 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4322 {
4323 int work = 0;
4324 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4325
4326 /* Check if we have pending ipi, its better to send them now,
4327 * not waiting net_rx_action() end.
4328 */
4329 if (sd_has_rps_ipi_waiting(sd)) {
4330 local_irq_disable();
4331 net_rps_action_and_irq_enable(sd);
4332 }
4333
4334 napi->weight = weight_p;
4335 local_irq_disable();
4336 while (1) {
4337 struct sk_buff *skb;
4338
4339 while ((skb = __skb_dequeue(&sd->process_queue))) {
4340 local_irq_enable();
4341 __netif_receive_skb(skb);
4342 local_irq_disable();
4343 input_queue_head_incr(sd);
4344 if (++work >= quota) {
4345 local_irq_enable();
4346 return work;
4347 }
4348 }
4349
4350 rps_lock(sd);
4351 if (skb_queue_empty(&sd->input_pkt_queue)) {
4352 /*
4353 * Inline a custom version of __napi_complete().
4354 * only current cpu owns and manipulates this napi,
4355 * and NAPI_STATE_SCHED is the only possible flag set
4356 * on backlog.
4357 * We can use a plain write instead of clear_bit(),
4358 * and we dont need an smp_mb() memory barrier.
4359 */
4360 napi->state = 0;
4361 rps_unlock(sd);
4362
4363 break;
4364 }
4365
4366 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4367 &sd->process_queue);
4368 rps_unlock(sd);
4369 }
4370 local_irq_enable();
4371
4372 return work;
4373 }
4368 rps_unlock(sd);
4369 }
4370 local_irq_enable();
4371
4372 return work;
4373 }
I think we should delete the NIC from the poll list by:
list_del(&napi->poll_list);
and my git repository infromation is:
[root@localhost linux]# git show
commit aa39477b5692611b91ac9455ae588738852b3f60
Merge: 48ec833 5164bec
Author: Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx>
Date: Mon Dec 22 14:47:17 2014 -0800
Merge tag 'dm-3.19-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm
Pull device mapper fixes from Mike Snitzer:
"Thre stable fixes and one fix for a regression introduced during 3.19
merge:
- Fix inability to discard used space when the thin-pool target is in
out-of-data-space mode and also transition the thin-pool back to
write mode once free space is made available.
- Fix DM core bio-based end_io bug that prevented proper
post-processing of the error code returned from the block layer.
- Fix crash in DM thin-pool due to thin device being added to the
pool's active_thins list before properly initializing the thin
device's refcount"
* tag 'dm-3.19-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm:
dm: fix missed error code if .end_io isn't implemented by target_type
dm thin: fix crash by initializing thin device's refcount and completion earlier
dm thin: fix missing out-of-data-space to write mode transition if blocks are released
dm thin: fix inability to discard blocks when in out-of-data-space mode
Don't handle the list_del is right, or this is a bug?
Thank you.
_______________________________________________ Kernelnewbies mailing list Kernelnewbies@xxxxxxxxxxxxxxxxx http://lists.kernelnewbies.org/mailman/listinfo/kernelnewbies
