VME Master/Slave Window Setup.
DJ
I have a few questions regarding VME master/slave window setup for the
Tundra PCI Universe II with the VxWorks OS. I am using a PPC604 processor
for info.
Can somebody confirm with me whether my understanding of master/slave
windows is correct?
- Master windows are mappings of local physical memory to VME memory.
- Slave windows are mappings of VME memory to local physical memory.
If I had four processors, two per board - One of which is the "master"
processor that is in charge of transferring 40MB of data at a time to the
other processors (known as the slave processors), would the following setup
work? Each slave processor needs a copy of the 40MB of data and needs to
send a few acknowledgements back to the master.
- On the "master" processor, setup one master window that is 40MB is size.
- On the rest of the processors, set up one slave window that is also 40MB
is size which ensures the 40MB of memory the master window addresses is seen
on the slave processors.
When the master processor places 40MB of data in its own physical memory
that is mapped to the VME, am I right in thinking using this setup, the data
would then appear on each of the slave processors local memory?
I would then need to set up the reserve on for the acknowlegements from the
slaves to the master processors.
I have read tech note 56 from WindRiver on VME and the VxWorks Programmers
Guide and this is what I can deduce from it!
If you can respond, could you do so via the newsgroup, instead of via email.
Thank you.
Regards,
Dan
mitch
>Hi All,
>
>I have a few questions regarding VME master/slave window setup for the
>Tundra PCI Universe II with the VxWorks OS. I am using a PPC604 processor
>for info.
>
>Can somebody confirm with me whether my understanding of master/slave
>windows is correct?
>
>- Master windows are mappings of local physical memory to VME memory.
Essentially correct. Master windows map UNOCCUPIED local physical
memory spaces to one or more VME address spaces via the Universe
mapping registers.
>- Slave windows are mappings of VME memory to local physical memory.
The local physical memory space must be assigned to real RAM or some
other memory-mapped device.
>If I had four processors, two per board - One of which is the "master"
>processor that is in charge of transferring 40MB of data at a time to the
>other processors (known as the slave processors), would the following setup
>work? Each slave processor needs a copy of the 40MB of data and needs to
>send a few acknowledgements back to the master.
>
>- On the "master" processor, setup one master window that is 40MB is size.
>- On the rest of the processors, set up one slave window that is also 40MB
>is size which ensures the 40MB of memory the master window addresses is seen
>on the slave processors.
>
>When the master processor places 40MB of data in its own physical memory
>that is mapped to the VME, am I right in thinking using this setup, the data
>would then appear on each of the slave processors local memory?
Correct, but you would have to notify the slaves, preferably by mail box
interrupt.
>I would then need to set up the reserve on for the acknowlegements from the
>slaves to the master processors.
>
>I have read tech note 56 from WindRiver on VME and the VxWorks Programmers
>Guide and this is what I can deduce from it!
Also read WTN 44 and 47 (if they are still available), section 3.5
Shared-Memory Network on the Backplane in the VxWorks 5.4 Network
Programmer's Guide, and the BSP header file for more details.
HTH,
--mitch