Subject: Re: com "silo overflow" messages
To: Ken Raeburn <email@example.com>
From: Tom Yu <tlyu@MIT.EDU>
Date: 01/21/1997 16:19:22
>>>>> "raeburn" == Ken Raeburn <firstname.lastname@example.org> writes:
raeburn> However... In my current configuration, I don't have a wd
raeburn> controller any more. All disks are connected to an Adaptec
raeburn> 2940U SCSI controller. And yet I still get "silo overflow"
raeburn> reports during heavy disk activity.
raeburn> My kernel dates from mid-November. I tried updating, but it
raeburn> looks like I'll need to update some other tools first so I
raeburn> can build the new kernel.
I'm running a 11-Jan or so snapshot of -current on my 486/66 with a
82450 (fifoless 8250 clone) serial port, and that still had silo
overflow problems at 38400 bps until I dropped in Onno van der
Linden's com driver patches (updated by Bill Sommerfeld). I suspect
there are still some com driver problems related to dropped
interrupts, since I still experience serial port lockups.
raeburn> Any chance the ahc driver might have problems like Michael
raeburn> had suggested for the wd driver? Any way I can tell?
I'm using the bha driver with a BT-445S and I was still seeing a
similar problem. I suspect that a broader problem relating to
disabled interrupts is present here, considering that the problem
appears with wd, aha, and bha drivers. Onno's patches do appear to
help somewhat, given that I've had at least an order of magnitude drop
in silo overflows since applying them. I suspect this is due to the
new two-level interrupt handling structure implemented by the
I would really like to see NetBSD have a more robust com driver, but I
have no idea if this is actively being worked on. Certainly, Onno's
patches are a good place to start, given the performance change I've
seen as a result of using them. Bill? Any thoughts? :-)