Date: Mon, 6 Jan 1997 18:44:31 -0700 (MST) From: Terry Lambert <terry@lambert.org> To: bde@zeta.org.au (Bruce Evans) Cc: bde@zeta.org.au, cracauer@wavehh.hanse.de, freebsd-current@FreeBSD.ORG Subject: Re: i586-optimized copyin/out still broken Message-ID: <199701070144.SAA13361@phaeton.artisoft.com> In-Reply-To: <199701030215.NAA13793@godzilla.zeta.org.au> from "Bruce Evans" at Jan 3, 97 01:15:47 pm
next in thread | previous in thread | raw e-mail | index | archive | help
> >I filled a PR (2142) about it, maybe it could be closed then. > > That is much harder to fix. It is caused by the floating point state > not being preserved across signal handlers. There are few, if any, > valid and useful uses for floating point in signal handlers, because > an ANSI signal handler must not make any accesses to a global object > other than assignment to ones of type `volatile sig_atomic_t'. Thus > preserving the state would mainly slow down signal handlers. Is it possible to do a lazy save of the FPU state? If so, it would be a compare/noop until someone did FPU stuff in their signal handler. "The Proper Thing" is probably to set a flag and do the floating point stuff as a result of a signal flag instead of in a signal handler -- that's just good practice anyway, since signals are not events, and it tends to close (a bit) some of the race windows in signals being stupid. Terry Lambert terry@lambert.org --- Any opinions in this posting are my own and not those of my present or previous employers.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199701070144.SAA13361>