Quoting is (19-Apr-01 06:45:08):

> On Thu, Apr 19, 2001 at 07:07:23AM +0100, Georges Heinesch wrote:

>> What does the error message in <subject> mean

> The 2 byte minus one bit buffer provided by the Amiga serial
> hardware did overflow.

What a huge buffer.

>> and how can I avoid it.

>> It happens during ftp download on my A3000, especially when moving
>> the mouse with screenblanker or when moving windows around quickly
>> while download is in progress.

> What is the disk controller you are using?

- internal A3000 scsi
  (netbsd partitions are installed on that one)

- csppc scsi
  (none of this disks were mounted while the silo error occurred)

>> It's an A3000, ser0 is the internal serial prot (modem is
>> connected),
>> /netbsd is 1.5.
>> 
>> Any ideas?

> Without knowing your full configuration:

What do specifically need?

> - make sure both modem and PPP and cable in between know about
> hardware 
>  handshake (crtsdts).

modem: &H1 / &R2 - RTSCTS is selected
ppp  : /etc/ppp/options.tty00 contains "crtscts"
cable: lines are provides

> Read "man pppd" and your modem manual about
>  this. Look at the modem control lights to verify that the computer
>  really does switch rts off, and that the rx data really DO stop to
>  come in in this case.

Hard to see with my US Robotics V.EVERYTHING. The lights RS and CS are
steadily on, which, according docs, means that RTSCTS is used.

Hence the lights behave identical as when used from AmigaOS (MiamiDx).

> - if this doesn't help, use options LEV6_DEFER in your kernel
> configuration.

Ok, gonna try this when the 38400 also fails!

> - if this still doesn't help, increasing SERIBUF_SIZE might slightly
> decrease
>   the overflow rate a bit.

Same here. Gonna do that if 38400 and LEV6_DEFER fails.

> - overall, you should operate the serial-modem connection at the
> slowest
>  speed possible. That is, do NOT use V42bis or MNP5 with your modem;
>  use PPP compression instead,

AFAIK, my modem doesn't support PPP compression (if it has to be
supported by the modem). However I don't see your point here. V.42bis
and MNP5 is modem-modem stuff. My problem is however computer-modem
related. Where's the relation?

> and use the slowest serial speed that
>  is slightly above your modem bitrate (e.g., for 28800 and 33600 use
>  38400). Commodore did specify the port for 57600 bps maximum, less
>  DOES help a lot.

I'm gonna try it out with 38400, even if this solution is highly
inacceptable for me since the modem works fine with Miami (internal
serial device on 57600). When using 8n1, I could even use it with
115200 without any problem. Other serial devices (like artser.device)
also work fine with 57600 on the same computer. I can remember that
the standard Commodore serial.device created problems with 57600,
while 38400 worked fine again.

> - if you have a disk controller that does LARGE dma transfers beyond
> what
>  Commodore specified as maximum allowable without being interuptible
>  by the CPU, tough luck. We did our best to limit DMA transfers and
>  corresponding
>  68040/68060 cache flushes in the NetBSD drivers if ser0 is used,
>  but you might be forced to buy a serial port with a deeper fifo.

Blk=1 DriveName=NetBSD_root DevFlags=0 HostID=7 Next=2 Mount Boot
Surfaces=1 BlocksPerTrack=2864 Reserved=00000000 PreAlloc=00000000
Interleave=0 LowCyl=2 HighCyl=145 NumBuffers=30 BufMemType=0
MaxTransfer=00ffffff Mask=7ffffffe BootPri=-15 DosType=4E425207
'NBR\7'

0xffffff is standard, isn't it?

I don't think it's a disk related problem. Remember that it also
happens when I move windows around quickly. No disk activity is
involved while doing this.

Thanks.

-- 
Cu  Georges Heinesch, Luxembourg
    geohei@geohei.lu
    http://www.geohei.lu
    PGP RSA & DH/DSS public key on request and on public servers

... what goes up, must come down ...