Soft IRQs

[Jamie Iles <lkml@jamieiles.com>]

Hi,

I am trying to understand whether it is correct behaviour for soft IRQs 
to be executed when interrupts with disabled. In particular, if I have 
some code that does:

    spin_lock_t mylock = SPIN_LOCK_UNLOCKED;
    unsigned long flags;
    spin_lock_irqsave( &mylock, flags );
   
    ...
   
    spin_lock_irqrestore( &mylock, flags );

Can soft IRQs run in the critical section above? I have a problem where 
'local_bh_enable_ip()' is being called as a result of 'dev_kfree_skb()' 
and a NET_RX_SOFTIRQ is being raised when I expect interrupts to be 
disabled.

'local_bh_enable()' only uses the 'in_irq()' macro to check whether we 
should do 'do_softirq()' and it also only checks we do not have the 
softirq field of 'preempt_count' non-zero before enabling soft IRQs.

I can see that if I was to replace 'spin_lock_irqsave()' with 
'spin_lock_bh()' then the softirq field of 'preempt_count' would be 
incremented and prevent soft IRQs until the lock was released. Should 
'spin_lock_irqsave()' also disable soft interrupts?

Thanks,

Jamie