Date: Mon, 24 Apr 2000 14:42:09 -0500 From: "Jim King" <king@sstar.com> To: <e-masson@kisoft-services.com>, <freebsd-stable@FreeBSD.ORG> Subject: Re: sio interrupt handler problem Message-ID: <003801bfae25$39a6d100$a44b8486@jking> References: <5564.000424@pd.chel.ru><20000424114057.A692@pir.net> <14596.37283.563515.214828@localhost.nantes.kisoft-services.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Monday, April 24, 2000 1:25 PM, Eric Masson wrote: > Peter Radcliffe writes: > >Get some real serial ports. > > > >16450 ports, in my experience, really suck. From the era when they > >were common you can find dual fast serial cards with two 16550A ports > >on them, which are considerably better. > Hello > > Thinkpad 390 2626-700 (PII-266 64Mo) > > FreeBSD notbsdems.nantes.kisoft-services.com 4.0-STABLE > FreeBSD 4.0-STABLE #0: Mon Apr 24 13:02:58 CEST 2000 > emss@notbsdems.nantes.kisoft-services.com: > /usr/src/sys/compile/THINKPAD i386 > > sio0 at port 0x3f8-0x3ff irq 4 flags 0x10 on isa0 > sio0: type 16550A > > So i assume my box has real serial ports :). > > I encounter lots of silo overflows while using userland ppp under Xfree. > - Under XFree86 4.0, approximately 250 Silo Overflows (SO) while fetching > mail & news (10 minutes). > - Under XFree86 3.3.6, approximately 220 SO for the same task. > - Shell mode, xemacs -nw & fetchnews generate 1 SO from time to time. > > >From the previous, i could think that XFree86 is responsible for the SO, > but one month ago this box was running 3.4-STABLE & XFree86 3.3.6 and I was > wondering what were these SO messages in freebsd-stable mailing list ;). > > I don't really understand the point. Maybe i misconfigured my kernel. Well > all comments welcome. One thing I've done in the past to overcome serial port problems is to patch sys/isa/sio.c to set the FIFO receive trigger level to 8 bytes instead of 14 (FIFO_RX_MEDH instead of FIFO_RX_HIGH). If you have a piece of hardware or a misbehaved driver that's causing high interrupt latency this change be really beneficial, and greatly outweighs the higher interrupt overhead when using an 8 byte trigger. <soapbox> An 8 byte trigger is a safer option that causes better operation on weird hardware without signifcantly changing anything on good hardware. I really wish sio.c would be changed to set the trigger to 8 bytes by default, or at least have a flags option to change the trigger level. </soapbox> Jim To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-stable" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?003801bfae25$39a6d100$a44b8486>