Date: Wed, 18 Mar 1998 01:08:59 -0600 (CST) From: Lars Fredriksen <fredriks@Mcs.Net> To: shimon@simon-shapiro.org Cc: thyerm@camtech.net.au, mike@smith.net.au, current@FreeBSD.ORG, nate@mt.sri.com Subject: Re: silo overflows (Was Re: 3.0-RELEASE?) Message-ID: <199803180708.BAA10857@Mercury.mcs.net> In-Reply-To: <XFMail.980310093045.shimon@simon-shapiro.org> from "Simon Shapiro" at Mar 10, 98 09:30:45 am
next in thread | previous in thread | raw e-mail | index | archive | help
Simon Shapiro writes: > > [stuff deleted] > > Because I get the silo overflows consistently, without any X11 on the > machine, I doubt X11 is key to this. From Mike's excellent overview of > interrupts, it is pretty clear that an interrupt is leaking someplace. > I ALWAYS see this silo overflow associated wit ha DPT lost interrupt; > > Some time ago (several months) I noticed that the DPT driver is not getting > certain interrupts. I simply did not get them. It was not a case where I > receive them and not proces them. At first I suspected the hardware. It > happens, under SMP, on two drastically different motherboards. > I added a simple timer to the DPT driver that occasionally wakes up and > checks the hardware for posted interrupt. If I find one, I process it and > complain. > > I always get a silo overflow when using PPP. It always happens with a DPT > lost interrupt recovery. Not every DPT lost interrupt recovery has an > sio.c silo overflow, but every silo overflow has a DPT lost interrupt > associated with it. > > Probably a race condition in interrupt handling. If X11 causes the video > card to generate interrupts, this is our clue; A video card can generate > interrupts very quickly. So can a DPT; Less than 2us between interrupts > is not unusual. > > For all it's worth; Fast Interrupts remind me of Linux interrupts; You > shutdown all interrupts, etc. Linux chronically loses interrupts. If you > block all interrupts, there simply is no way to guarantee you will not lose > one. This is how I understand it. I am probably wrong. > > Simon > Hi, I also see lots of : sio3: 151 more interrupt-level buffer overflows (total 151) Usually serveral thousands per day. All mine are from ppp using the serial link. When using netscape it makes the problem worse but linx hardly ever causes problem. Just as Simon did attribute the problem related to DPT, mine seems related to disk access as well. I am using the ahc driver (cam at the moment), and I wonder if my problems doesn't stem from the funny MP table that SuperMicro puts in. I haven't tried it in UP mode to see if there is a problem there. I hope to try the MP table patches that Tor put together in a couple of days. Just my 2 cents worth. Lars -- ------------------------------------------------------------------- Lars Fredriksen fredriks@mcs.com (home) lars@odin-corporation.com (home-home) 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?199803180708.BAA10857>