From owner-freebsd-current Wed Jun 5 11:46:38 1996 Return-Path: owner-current Received: (from root@localhost) by freefall.freebsd.org (8.7.5/8.7.3) id LAA23190 for current-outgoing; Wed, 5 Jun 1996 11:46:38 -0700 (PDT) Received: from phaeton.artisoft.com (phaeton.Artisoft.COM [198.17.250.211]) by freefall.freebsd.org (8.7.5/8.7.3) with SMTP id LAA23185 for ; Wed, 5 Jun 1996 11:46:36 -0700 (PDT) Received: (from terry@localhost) by phaeton.artisoft.com (8.6.11/8.6.9) id LAA29393; Wed, 5 Jun 1996 11:41:00 -0700 From: Terry Lambert Message-Id: <199606051841.LAA29393@phaeton.artisoft.com> Subject: Re: Vm fixes NG To: toor@dyson.iquest.net (John S. Dyson) Date: Wed, 5 Jun 1996 11:40:59 -0700 (MST) Cc: terry@lambert.org, james@miller.cs.uwm.edu, FreeBSD-current@FreeBSD.org, fcurrent@jraynard.demon.co.uk, grog@lemis.de In-Reply-To: <199606050242.VAA02177@dyson.iquest.net> from "John S. Dyson" at Jun 4, 96 09:42:41 pm X-Mailer: ELM [version 2.4 PL24] MIME-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Sender: owner-current@FreeBSD.org X-Loop: FreeBSD.org Precedence: bulk > This discussion of trying to solve the problem of a GUI leaving the > console in a state that the kernel doesn't know how to change during > a system crash is interesting. There is a potential solution (hack) > that I don't remember being mentioned. HACK ALERT!!!: Specify a callback > into user space -- wired specially, with a special system call, that the > kernel (syscons, pcvt, etc) can call in the case of a crash to fix the problem. > To make it easiest on the kernel VM people, it would be nice if it would be > a self-contained module and PIC. We don't support user space ISRs (yet), but > this isn't very different from an user space ISR. > > Just a thought. It's a good thought. The segment containing such a routine and its data would need to be locked in core. If it took a fault, it would need to fail the process, not the kernel, s a result of the fault. There is also the problem of associating a credential with the callback registration. This gets back into the issue of whether credentials should be seperate from the process as a container object. In the case of an AST implementation (what you are talking about here is the kernel running user-provided code, preferrably in the context of a credential and a page map), this gets especially hairy. 8-). It probably wants support for multiple code and data segments (COFF or ELF, basically) to be able to pull it of without locking the whole thing in core following a registration. 8-(. Terry Lambert terry@lambert.org --- Any opinions in this posting are my own and not those of my present or previous employers.