Re: Why two stacks per process?

[Patrick McManus <mcmanus@xxxxxxxxxxxx>]

On Thu, 2008-04-24 at 09:35 -0400, Robert P. J. Day wrote:
> On Thu, 24 Apr 2008, sahlot arvind wrote:
> 
> > Yes. I agree with Robert. There are two stacks per process. But why
> > cannot we just use one stack as I said earlier?
> 
> while there may be other reasons, there's one obvious one -- as i
> mentioned in my earlier email, the kernel likes to store a small,
> pre-process "thread_info" structure at the bottom of the kernel stack.

the other prominent reason is the kernel needs somewhere to keep the
state for when you get scheduled off while you're running kernel code on
behalf of the userspace process.. this might be in the preemptible
kernel, or maybe just when you "sleep" on something like a semaphore,
page fault, wait_event*, or any of the other things you can't do in an
atomic context ;)

I've read speculation about just using one kernel stack per CPU for
normal system call usage, another for irqs, and then tearing them up and
down dynamically for these interrupted kernel code situations (and
perhaps other situs?).. I guess that would work, but it certainly has
some overhead.

-Patrick
--
www.ducksong.com



--
To unsubscribe from this list: send an email with
"unsubscribe kernelnewbies" to ecartis@xxxxxxxxxxxx
Please read the FAQ at http://kernelnewbies.org/FAQ