Messages in this thread | | | From | (Alan Cox) | Subject | Re: tq_immediate | Date | Mon, 22 Feb 1999 12:01:53 +0000 (GMT) |
| |
> I'm working on a network device driver which has a global data (very > complex dynamic data structure actually) which is accessed for read
Thats generally a bad sign 8)
> and write from dev->hard_start_xmit and dev->do_ioctl and a third > function, let's say foo(), which is not called directly but frequently > queued as task on tq_immediate. Should the global data be protected by > sema/spinlock or something?
You can be executing open and interrupt handlers together the moment you don't have interrupts on.
You can be executing get_stats, do_ioctl, hard_start_xmit and the receive irq in parallel if you have enough cpus. I think close as well, but I'd have to check.
> The ioctl can not interrupt other functions, that's clear. I hope that > hard_start_xmit can not either. But what about the queued task foo?
Its not just about interrupting. They can be running in parallel on multiple processors
> BTW is that true that dev->hard_start_xmit call is always initiated > by a user program (send/write call). Or is there any case when it is > called by the kernel itself (for example from a queued task)?
It can be run from a bottom half (normally timer_bh or net_bh). The transmit function itself won't be recursively re-entered and when it is called bh handling is blocked (please dont rely on that for 2.3, just rely on the drivers transmit function not being re-entered)
Alan
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
| |