From owner-freebsd-current  Tue Sep  7 10:42:31 1999
Delivered-To: freebsd-current@freebsd.org
Received: by hub.freebsd.org (Postfix, from userid 583)
	id A22F814DC4; Tue,  7 Sep 1999 10:42:28 -0700 (PDT)
To: nordwick@erdos.askjeeves.com, wollman@khavrinen.lcs.mit.edu
Subject: Re: (P)review: sigset_t for more than 32 signals
Cc: current@FreeBSD.ORG
In-Reply-To: <199909061833.OAA03865@khavrinen.lcs.mit.edu>
Message-Id: <19990907174228.A22F814DC4@hub.freebsd.org>
Date: Tue,  7 Sep 1999 10:42:28 -0700 (PDT)
From: bde@FreeBSD.ORG (Bruce Evans)
Sender: owner-freebsd-current@FreeBSD.ORG
Precedence: bulk
X-Loop: FreeBSD.ORG

> It's true that this would (currently) slow down uthreads.  However, I
> suspect that uthreads may want to do its own signal-mask handling
> anyway (even though it currently doesn't).  On the other hand, if we
> made jmp_buf and struct sigcontext compatible, it could clean up the
> thread scheduler's inner workings.

Note that C9x requires longjmp() to leave the floating point
environment alone, so the full sigcontext must not be restored.
This doesn't prevent using sigreturn() to implement longjmp(),
but it guarantees that setjmp()/longjmp() can't be (ab)used to
implement full context switching.

Making jmp_buf the same as struct sigcontext would be a bit wasteful.

Bruce


To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-current" in the body of the message