Date: Wed, 10 Apr 1996 20:30:50 +1000 From: Bruce Evans <bde@zeta.org.au> To: freebsd-bugs@freefall.freebsd.org, hsu@clinet.fi, scrappy@freefall.freebsd.org Subject: Re: kern/528 Message-ID: <199604101030.UAA15447@godzilla.zeta.org.au>
next in thread | raw e-mail | index | archive | help
>Synopsis: slow 386 reports excessive interrupt-level buffer overlows >State-Changed-From-To: open-closed >State-Changed-By: scrappy >State-Changed-When: Tue Apr 9 21:19:42 PDT 1996 >State-Changed-Why: >fixed setting 'options RS_IBUFSIZE=1024', with appropriate >modifications made to sio.c to allow for this Erm. This isn't an acceptable fix. It breaks things unless you also increase TTYHOG and perhaps some other magic numbers (TTYHOG must be at least few hundred bytes larger than 2*RS_IBUFSIZE...). I left the PR open to remind me to fix the problem properly someday. Part of the problem was that softclock() is often delayed for several clock ticks but sio assumed that it was only long delayed on extremely overloaded systems. This problem was moved sometime between 2.0.5R and 2.1R when sio switched from timeouts to softtty interrupts. Now the problems are: - softtty interrupts are masked by splhigh() and there is a bug somewhere that results in even clock interrupts on my p133 being masked for up to 5 clock ticks. - softtty interrupt handlers are preemptible by almost all h/w interrupt handlers, so the same hogs that prevent softclock() from running often enough (IDE disk interrupt handlers?) may prevent the softtty handlers from completing soon enough. Bruce
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199604101030.UAA15447>