Short:        improve reliability of serial transfers
Author:       Ralph Babel, Robert Miranda
Uploader:     Matt_H amiga org
Type:         driver/other
Version:      1.7
Architecture: m68k-amigaos

***********************************
* GvpPatch 1.7 README, 1993-01-17 *
***********************************

The executable "GvpPatch", which this text file is supposed
to be accompanied by, is a patch utility that serves to make
serial-port transmissions more reliable when using a GVP
Series-II host adapter.  The problem this patch addresses is
neither a deficiency in the GVP hard- or software nor in the
serial.device's code, but rather a limitation of the Amiga's
on-board serial port, which can only hold a total of one
byte of incoming data (plus one byte currently being
shifted).  The port requires CPU intervention to handle flow
control, so any delay (caused by system load, disk activity,
etc.) to the CPU's ability to respond to interrupts may - at
higher baud rates - cause loss of data while receiving, e.g.
during downloads.  The GVP Series-II is not the only DMA
host adapter affected by this problem, for instance when
using fast SCSI drives employing on-board cache memory; one
may also encounter this problem on the Microbotics
HardFrame, Commodore's A2090 and A2091, and even
programmed-I/O cards.

In order not to tie up the Zorro-II bus for long periods of
time, GvpPatch will therefore transparently split large data
blocks into smaller ones while the serial.device is in use.
Its syntax is:

    Run >NIL: GvpPatch -s

Sending a Ctrl-C break signal to the process will terminate
it. If a device driver other than the serial.device is used,
"-s" may be followed by the name of that driver (e.g.
"-sgvpser.device") to override the default. Multiple "-s"
options can be specified to monitor several device drivers.

In addition, using the "-o" option, it is possible to make
certain types of Direct SCSI applications work, the authors
of which did not interpret Commodore's ambiguous HD_SCSICMD
"documentation" the same way GVP did.  This option is not
required by gvpscsi.device 4.4 and later versions.

A standard 68000-based Amiga using a Series-II hard card
should be able to handle 19200 baud, even without Fast RAM,
provided that system load is not too high.  Having several
floppy-disk or hard-disk filesystem processes active, or
running high-resolution multicolor screens, for example, can
degrade system performance enough that even downloads to the
RAM disk may cause problems.  Using a GVP combo accelerator
with 32-bit-wide RAM, transfers at 38400 baud are possible.
If you have a 68020 or a 68030, the use of "SetCPU FASTROM
CACHE" is strongly recommended with Kickstart 1.3.  Note
that when using the "CPU" command included with Amiga OS
2.04 instead of "SetCPU", the table of exception vectors is
moved into Fast memory as well when using the "FASTROM"
option.  GVP's original Series-I SCSI host adapters and the
GVP AT interface do not need GvpPatch.  By the way: the baud
rates mentioned above refer to full-speed null-modem
transmissions, e.g. ~1800 cps at 19200 baud and ~3600 cps at
38400, not just the usual HST/V.32/V.32bis modem transfer
rates, which are usually lower.

The following FaaastROM SCSI driver versions are supported:

- gvpscsi.device 3. 7
- gvpscsi.device 3.12 (= Omni 3.0 to 3.4)
- gvpscsi.device 3.14 (= Omni 3.5)
- gvpscsi.device 3.15 (= FaaastROM 4.0)
- gvpscsi.device 4. 3
- gvpscsi.device 4. 4
- gvpscsi.device 4. 5
- gvpscsi.device 4.12
- gvpscsi.device 4.13

If you are uncertain as to which driver version is installed
in your system, use the CLI command "Version gvpscsi.device"
to find out.

In the authors' opinion, GvpPatch is just a kludge that
attempts to hide a major deficiency in the built-in
serial-port hardware and is therefore not officially
supported or endorsed by GVP.  It is nevertheless released
in the hope that it will help certain system configurations.
It is provided "as is" without warranty of any kind, either
expressed or implied, including, but not limited to,
correctness, reliability, and the fitness for any particular
purpose.  The entire risk as to its quality and accuracy is
assumed solely by the user.

Ralph Babel

Robert Miranda