Date: Tue, 4 Apr 2000 10:09:11 +0930 From: Greg Lehey <grog@lemis.com> To: Alfred Perlstein <bright@wintelcom.net> Cc: Matthew Dillon <dillon@apollo.backplane.com>, current@FreeBSD.ORG Subject: Re: cvs commit: src/sys/i386/i386 support.s src/sys/kern init_sysent.c kern_prot.c kern_sig.c Message-ID: <20000404100911.C53037@freebie.lemis.com> In-Reply-To: <20000402172349.T21029@fw.wintelcom.net> References: <200004021752.KAA13175@freefall.freebsd.org> <20000402163552.P21029@fw.wintelcom.net> <200004022312.QAA51299@apollo.backplane.com> <20000402164700.R21029@fw.wintelcom.net> <200004022338.QAA51565@apollo.backplane.com> <20000402172349.T21029@fw.wintelcom.net>
next in thread | previous in thread | raw e-mail | index | archive | help
On Sunday, 2 April 2000 at 17:23:49 -0700, Alfred Perlstein wrote: > * Matthew Dillon <dillon@apollo.backplane.com> [000402 17:04] wrote: >> >> :I did look at the code, struct proc is allocated from a zone, >> :meaning it won't "go away" once allocated, there's no danger in >> :dereferencing p_pptr, I don't get it. >> : >> :-- >> :-Alfred Perlstein - [bright@wintelcom.net|alfred@freebsd.org] >> :"I have the heart of a child; I keep it in a jar on my desk." >> >> What happens when the parent process exits and the system must >> reassign the parent to process 1? Now think about what happens >> when it occurs on one cpu while another is trying to access the >> ppid. >> >> cpu#1: cpu#2: >> >> read p->p_pptr >> indirect through to get ppid >> (stalls on a cache miss plus, >> due to heavy DMA, stalls on main memory) >> parent process finishes >> exiting, replaces p_pptr >> of children, releases >> struct proc. >> >> struct proc is reused, >> pid is reallocated >> read completes, wrong ppid is returned >> (neither the original ppid nor ppid 1 >> is returned). >> >> In an SMP system you have to assume the worst case, and the worst case >> is that a cpu can stall INDEFINITELY between instructions. > > Good call. > > Ugh, I should think about this more, but i'll just grasp at straws > and suggest that p_pptr is set to volatile variable, and we re-read > it after we snarf the pid from the pointer and make sure it hasn't > changed out from under us. > > Either that or store it in the child's proc struct as well. This seems the obvious solution. Greg -- Finger grog@lemis.com for PGP public key See complete headers for address and phone numbers To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-current" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20000404100911.C53037>