Re: 2.4.7 softirq incorrectness.

From: Andrea Arcangeli (
Date: Fri Jul 27 2001 - 10:01:07 EST

On Thu, Jul 26, 2001 at 05:47:50PM -0700, Maksim Krasnyanskiy wrote:
> Should we then create generic function (something like netif_rx_from_user) than will call do_softirq
> after calling netif_rx ?

creating such a function is certainly ok (it must first check pending()
before running do_softirq of course). The name shouldn't be "from user"
because we actually call it from normal kernel context.

> I queue it and do tasklet_schedule(tx_task). Everything works just fine but on SMP machine I noticed that sometimes
> data is sent in the wrong order. And the only reason why reordering could happen is if several tx_tasks are runing at the

Do you use tasklet_enable? This patch fixes a bug in tasklet_enable.
(bug found by David Mosemberg) We are thinking at more CPU friendly ways
to handle the tasklet_disable, Linus just had a suggestion, but I don't
have time right now to think much about the alternate approches (i'm at
ols), I will do next week. If you are usng tasklet_enable you may want
to give it a spin.

--- 2.4.7/include/linux/interrupt.h Sun Jul 22 01:16:45 2001
+++ 2.4.7aa1/include/linux/interrupt.h Thu Jul 26 21:08:16 2001
@@ -139,24 +139,26 @@
 static inline void tasklet_disable_nosync(struct tasklet_struct *t)
+ smp_mb__after_atomic_inc();
 static inline void tasklet_disable(struct tasklet_struct *t)
+ smp_mb();
 static inline void tasklet_enable(struct tasklet_struct *t)
- if (atomic_dec_and_test(&t->count))
- tasklet_schedule(t);
+ smp_mb__before_atomic_dec();
+ atomic_dec(&t->count);
 static inline void tasklet_hi_enable(struct tasklet_struct *t)
- if (atomic_dec_and_test(&t->count))
- tasklet_hi_schedule(t);
+ smp_mb__before_atomic_dec();
+ atomic_dec(&t->count);
 extern void tasklet_kill(struct tasklet_struct *t);

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to
More majordomo info at
Please read the FAQ at

This archive was generated by hypermail 2b29 : Tue Jul 31 2001 - 21:00:32 EST