Date: Wed, 02 Mar 2005 22:09:29 -0800 From: Nate Lawson <nate@root.org> To: David Xu <davidxu@freebsd.org> Cc: John Baldwin <jhb@freebsd.org> Subject: Re: cvs commit: src/sys/kern kern_sig.c Message-ID: <4226AA19.2080809@root.org> In-Reply-To: <4226A46B.2090704@freebsd.org> References: <200503021343.j22DhpQ3075008@repoman.freebsd.org> <200503020915.28512.jhb@FreeBSD.org> <4226446B.7020406@freebsd.org> <20050303033115.GA13174@VARK.MIT.EDU> <42269DB0.6070107@freebsd.org> <20050303052902.GA14011@VARK.MIT.EDU> <4226A46B.2090704@freebsd.org>
next in thread | previous in thread | raw e-mail | index | archive | help
David Xu wrote: > David Schultz wrote: > >> You have to worry about that anyway, though. A and B need to know >> that they're not allowed to hold locks across the calls if C calls >> msleep(), for instance. Anyway, your proposal if having a flag >> for msleep() is basically the same as my proposal of having a >> separate function. (The only difference is that adding a separate >> function doesn't break the ABI.) So it sounds like we're more or >> less in agreement here. >> >> >> > This is not a lock problem, this is the problem why a stack variable can > not > be used when thread is going to sleep, this is a rather odd behavior to me. > For example, thread A stack variable address p is put on a known place, > e.g, a queue, thread A unlocks the lock of the queue and sleeps, > sometimes later, a producer thread B writes the data into memory pointed > by p, > and wake up A, that's a very simple code, here malloc is not needed at all. > At the time, kernel shoudn't swap out the thread stack, any code trying > to swap > it out is totally broken. I've always treated local variables as valid only within the current stack frame, from the current context. If you had different per-context protection domains, for instance, there would be the same problem. >>> First find all code written in such way, but it is not that easy. >> >> True. If we changed msleep() to disable swapping by default, then >> we wouldn't have to worry about correctness problems related to >> missing some. Or better, you could add a debugging option to _always_ emulate a swap on msleep by marking the page not present until msleep returns. Then you wouldn't have to worry about correctness problems related to missing some. -- Nate
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?4226AA19.2080809>