Tornado/vxWorks 5.x Information
This page provides a repository for information about using the WRS Tornado
environment and the vxWorks 5.x RTOS with EPICS that doesn't really belong
anywhere else on the EPICS site. If you discover any other information that
ought to go on this page, please let me
know.
Note that there is a separate page provided for
users of vxWorks 6.x and Wind River Workbench.
There is also a reasonably good Tornado 2.0 FAQ available on the web,
mostly comprising answers to questions posted to the
comp.os.vxworks news group.
Tornado 2.2 (vxWorks 5.5)
Installation
According to the Tornado 2.2 release notes, you cannot install multiple
host and/or target architectures in the same directory.
Linux Hosting
See this page for information building vxWorks
target code on Linux.
EPICS Changes
Ron Sluiter has succeeded in building Base R3.13.7 against Tornado 2.2,
but said in an email to tech-talk that in order to do so he had to remove the
switch -nostdinc from the EPICS configuration file
<base>/config/CONFIG.Vx. Base R3.14.x should build on Tornado 2.2
without change.
Tornado 2.0.x (vxWorks 5.4.x)
VxWorks 5.4.x will not be supported by future EPICS Base 3.15 releases. The GNU
C++ compiler provided with 5.4.2 is too old, you will have to upgrade to at
least vxWorks 5.5 or preferably vxWorks 6.x.
Installation
With Tornado 2.0.x it is possible to install multiple host and target
architectures to the same location, and this is probably better than keeping
them separate (Tornado 1.0.1 allowed some combinations, but not all). It is
definitely not a good idea to mix difference versions of Tornado (for
different architectures) in the same location.
WRS Patches
The following patches have been found to fix certain problems with Tornado 2.0
and ought to be installed at all EPICS sites using the indicated architectures
or board types. Patches must be downloaded from WRS's
Customer Support website:
-
sprT2CP4: Cumulative Patch 4 - this converts a Tornado 2.0 installation into
Tornado 2.0.2, and is highly recommended.
-
TDK-13418: Tornado-Comp-Drv Feburary 2002 release v0700 - this upgrades various
BSP drivers for Tornado 2.0.2.
Configuring vxWorks
See the document Configuring Tornado 2.0.x for
EPICS for some information on how to configure your vxWorks image to be
able to load and run EPICS.
Some additional points to note about the Tornado 2 configuration:
-
If you're configuring Tornado 2.2, make sure you don't include RIP unless you
know you need it (very unlikely for EPICS IOCs). This has caused problems at
some sites as it can eat up 100% of the CPU in a fairly high priority
task.
The network memory buffer configuration for Tornado 2.x is significantly
different to that for Tornado 1.x and earlier, because of the new network stack
that is uses. The new configuration process requires that the complete number
and size of these buffers be set at compile time (actually these numbers can be
changed early in the BSP startup, but not afterwards); the old stack could
dynamically grow the number of mbufs it used as necessary, so the initial
configuration was not particularly critical, unlike with the new stack. The
WRS defaults provided in Tornado 2.x are too small for most operational EPICS
IOCs.
The official WRS method of configuring the network stack is described in
section 4.6.3 Network Memory Pool Configuration of the vxWorks Network
Programmer's Guide (that section number is from the vxWorks 5.4 edition, for
the vxWorks 5.5 edition it's 4.3.3). APS and SNS have both developed their own
alternatives to this fixed approach which permit the number of buffers to be
selected at boot time from the vxWorks boot parameters. At APS we provide two
configurations that are selected between by one of the bits in the boot flags
according to the loading on that particular IOC. The actual numbers of buffers
used are as follows:
Parameter | Value (light) | Value (heavy) |
Data Pool clDescTbl [] |
NUM_64 | 125 | 250 |
NUM_128 | 400 | 400 |
NUM_256 | 50 | 50 |
NUM_512 | 25 | 50 |
NUM_1024 | 25 | 25 |
NUM_2048 | 25 | 25 |
NUM_NET_MBLK | 800 | 1200 |
NUM_CL_BLK | 650 | 800 |
System Pool sysClDescTbl [] |
NUM_SYS_64 | 256 | 1024 |
NUM_SYS_128 | 256 | 1024 |
NUM_SYS_256 | 256 | 512 |
NUM_SYS_512 | 256 | 512 |
NUM_SYS_MBLK | 1024 | 3072 |
NUM_SYS_CL_BLKS | 1024 | 3072 |
Known Problems
-
The WRS SPR#31718 "cksum: out of data message logged to console when
target is proxyArp server" may affect sites that are using the Proxy
ARP server to forward packets to a secondary CPU through the backplane (shared
memory) network. The published workaround is to add
proxyPortFwdOff(67) to the startup script of the relevent IOC, and
should be safe providing you are not using DHCP or BOOTP to configure the slave
IOCs on the backplane.
-
If the shell on an MVME1xx IOC (68k family) ever runs continuously without ever
relinquishing control to lower priority tasks for more than 1.5 seconds, you
might get the error message
interrupt:
ei0: reset
This comes from the ei ethernet driver whenever tNetTask
takes longer than 1.5 seconds to process some incoming packets, and may or may
not be serious - at iocInit it appears to be benign (it often occurs in
dbLoadRecords with large databases), but if it occurs later there are
reports that it can cause a complete hang. If you're not using an NFS
filesystem I would suggest switching to this if possible as it might solve it,
otherwise you can temporarily reduce the priority of the tShell task
at the beginning of the startup script are restore it again afterwards as
follows:
taskPrioritySet(0,60)
...
taskPrioritySet(0,1)
-
It is unwise to configure Tornado 2.x to include the DNS resolver if you are
running EPICS R3.13.x. If you do and your DNS server dies or doesn't contain
some data that it should, Channel Access connections can take a very long time
to be made and the IOC will appear to be very flaky, although nothing obviously
appears to be wrong other than bad connections. This is caused by a CA task
being held up waiting for a DNS reply and preventing further connections until
this times out. EPICS R3.14 does not suffer from this problem however.
-
When upgrading a shared-memory backplane networked system from Tornado 1.x to
Tornado 2.x, you should be careful about any subnet mask given in the
"inet on backplane" boot parameter setting. In Tornado 1.x the
subnet mask appears to have been less critical, whereas in Tornado 2.x values
that worked fine before can cause the main network to stop working (processor 0
might decide to route all packets destined for your boot host through the
backplane network, which isn't very helpful when it's looking for the symbol
table file or startup script). This is a standard TCP/IP network configuration
issue associated with subnets, and you should talk to someone knowledgable in
that area if you can't work out what the setting should be yourself.
-
The error message "arpresolve: unable to allocate llinfo" may
be an indication that you have something wrong with your routing tables,
although there may be other causes for this message too. Tornado 2 is less
forgiving of network configuration problems than earlier versions, so if you
get these you should check your boot parameters (gateway inet) and any
routeAdd() calls in your startup to make sure they're correct for your network
configuration.
-
A message from tNetTask "arptnew failed on xxxxxxxx"
where xxxxxxxx is the hexadecimal representation of an IP address is
similar to the previous error, indicating some kind of routing configuration
problem. I received this explanation from a former WRS developer:
An IP packet is being transmitted, and based upon how it matches network
interfaces' IP addresses and mask, an outgoing interface is selected, at
which point the interface driver calls arpresolve to get a MAC address
for that IP address, either by sending an ARP request, or by using
a cached entry.
At that point, if the ARP code, looking for an ARP cache entry, finds
that this particular IP address has a specific routing entry which is
not an ARP entry, it will return that particular error.
Since ARP uses has a common mechanism with the IP routing table,
there could not be a routing entry and an ARP entry for the same
destination IP address at the same time.
One possible reason for this is if an ICMP Redirect is received for
a particuar IP address, which would be considered "local" if mathcing
its address to the network interface's IP address and mask.
mRouteShow() at the time of this error should pretty much paint the picture
in bright colour.
Tornado 1.0.1 (vxWorks 5.3.1)
Ethernet on mv2700 (dc driver)
There can be a network reliability problem with Tornado 1.0.1 on the mv2700
through certain types of 10/100baseT media. A quick way to resolve this
problem (rather than getting a patch for Tornado 1.0.1 from WRS) is to change
the setting of DC_MODE in the config/mv2700/config.h file
from 0x08 to 0x18. This puts the driver into Full Duplex
mode. Note that this setting must be correct for the network port you're
using, so you might need two boot files if you have both full- and half-duplex
ports.
PowerPC Issues
There is a separate page discussing the specific problems
associated with using PowerPC CPUs under vxWorks/Tornado.
|