Great news.
The value of getting a driver is very clear and obvious, but it has not
yet risen in the priority list.
We had a preliminary driver started a year or so ago, and we also have
a NIC/network style driver that the NetFPGA project
at Stanford (with Xilinx/Cisco support) is using when they use the
OpenCPI infrastructure for FPGAs acting as networking front-ends.
But neither is ready for prime time. The driver-level requirements are
actually pretty modest, but no linux driver wizard has
done it yet. Assuming we still want to configure the DMA memory pool
statically (the grub config), its pretty basic:
- allow allocation within, and mapping to, the dedicated DMA memory pool
(we mmap to /dev/mem today)
- allow PIO mapping to each worker's config space. (we mmap to /dev/mem
today)
- connect MSIX interrupts to the most efficient async wakeup/signal
mechanism to user space. (we don't do this today)
Finally, for security, we need to separate the location of the DMA
address/counter registers to a page that is only accessed
in kernel space to ensure that the device's DMA addresses are only
programmed to talk to the DMA memory pool.
Of course anyone reloading a bitstream can violate security...
Happily, the DMA design would still not involve any kernel activity
during data movement. Only for setting up the stream.
Any linux driver wizards in your shop? This shouldn't take more than a
month or so...
The core team is on a sprint to finish a major release this month, but
after that it might rise to the surface...
Cheers,
Jim
On 1/15/12 8:48 PM, Ziersch, Troy (Contractor) wrote:
UNCLASSIFIED
Thanks, I've got testRpl working now which means my SW and HW install is
complete.
Are there any plans for a Linux driver so everything doesn't need to be
run as root?
Cheers,
Troy
________________________________
From: James Kulp [mailto:[email protected]]
Sent: Friday, 13 January 2012 11:19 PM
To: Ziersch, Troy (Contractor)
Cc: [email protected]
Subject: Re: [opencpi_dev] testRpl Usage [SEC=UNCLASSIFIED]
Hi Troy,
The -x flag should point directly to the zipped bitstream file.
(bit.gz).
The default value for the -x flag, "file", isn't very useful...
All opencpi "binary artifact" files (files that hold one or more worker
binaries), have embedded XML that describes what is inside.
The documentation/usage message for testRpl is not as god as it should.
The excuse is mostly that we are going to rewrite the testRpl program
using the new xml-based application features, which will make it about
1/4 the size.
(See tools/cdk/examples/*_xml).
Jim
On 1/13/12 12:39 AM, Ziersch, Troy (Contractor) wrote:
UNCLASSIFIED
Hi,
I've managed to follow the instructions in Appendix VI of "Open
Reference Platform 2 Specification" up to the tests at the end.
The ocfrp_check at step 5a seems OK:
opencpi/ocpi/bin/linux-x86_64$ sudo ./ocfrp_check
OpenCPI FPGA at PCI 0000:03:00.0: BSV date Wed Nov 30 03:26:32
2011, bitstream date Tue Jan 10 16:49:23 2012, platform "ml605", device
"xc6vlx240t", UUID 303268bb-2d6c-45be-b441-56eee86ea508
But the testRpl at step 5b complains about a "file":
opencpi/ocpi/bin/linux-x86_64$ sudo ./testRpl -r1if -r3of -N0o 1
-N1i 1 -N3o 1 -N0i 1 -I 0 -i 1 0000:03:00.0
In PCIPIOXferFactory::PCIPIOXferFactory()
In PIOXferFactory::PIOXferFactory()
Starting: sending 1 messages of 0 (buffer 9000)
Buffer counts: 0o 1 1i 1 3o 1 6i 2 8o 2 0i 1 9i 2 9o 2
Active indicators: 0o (null) 1i flowcontrol 3o flowcontrol 6i
(null) 8o (null) 0i (null)
++++++++ In void XferFactoryManager::startup(), ref count = 0
OCFRP: 0000:03:00.0, with bitstream birthday: Wed Nov 30
03:26:32 2011
API Error: Cannot open file: "file"
I tried using the -x arg to point to various xml files in
opencpi/hdl/applications/testbias wihout any luck.
Any ideas what I'm doing wrong?
Cheers,
Troy
IMPORTANT: This email remains the property of the Department of
Defence and is subject to the jurisdiction of section 70 of the Crimes
Act 1914. If you have received this email in error, you are requested to
contact the sender and delete the email.
_______________________________________________
opencpi_dev mailing list
[email protected]
http://lists.opencpi.org/listinfo.cgi/opencpi_dev-opencpi.org
IMPORTANT: This email remains the property of the Department of Defence
and is subject to the jurisdiction of section 70 of the Crimes Act 1914.
If you have received this email in error, you are requested to contact
the sender and delete the email.
_______________________________________________
opencpi_dev mailing list
[email protected]
http://lists.opencpi.org/listinfo.cgi/opencpi_dev-opencpi.org