FreeBSD on the Ubiquiti Routerstation Pro
- AR7161 - 680MHz
- 128MB SDRAM
- 16MB serial flash
- 1 x USB 2.0 socket
- 1 x SD card slot
- 7 x GPIO
- Atheros AR8316 Gigabit switch PHY
- 3 + 1 1000baseTX ports, via AR8316 PHY
- 3 x MiniPCI slots
Stuff that's weird/broken
RedBoot caches the flags passed to exec -c. I'm not sure where it does this currently.
The RTC support is there but not yet functioning. It is on the SPI bus; the device datasheet is here - http://www.nxp.com/acrobat_download/datasheets/PCF2123_1.pdf The problem is that it shares the CS line with flash and thus requires some hackery in order to correctly function.
The USB code in -head needs some alignment fixes to work properly. A symptom is "AutoSense failed" when trying to probe the on-board SD slot which is a USB mass device. Kan has written a patch - http://people.freebsd.org/~kan/usb_rspro.diff .
- The SD slot can't easily be used as a root device - it doesn't probe successfully until -after- single user mode startup begins.
- Reading/writing data from /dev/redboot/XXX character devices must be done using an alignment and blocksize of 64k. It is slow and ties up the kernel!
Playing with MDROOT
- I'm using FFS version 1 and "makefs" to build an MDROOT. Note that the endian-ness needs to be forced to match the MIPS endian-ness (I'm using big endian) and the kernel vfs mount output when this is incorrect is very unhelpful.
- OpenWRT has some code from Ubquity which builds the images:
- The above tool needs to be modified to generate different load and entry points - the Linux images use the same load/entry points but FreeBSD's elf image doesn't
- The version string must start with RSPRO or the Redboot loader will refuse to run it!
- The redboot loader understands GZIP - so the kernel image can simply be gzip -9'ed to fit in the available space. The redboot loader will ungzip it into memory for you before executing.
RedBoot will refuse to write a kernel image that isn't a multiple of 64k (flash sector size?) Thus the kernel image must be padded before being fed to the firmware image program.
- The image generated by the above program will overwrite the non-system FIS partitions - so it's in effect a "fresh" start. It also avoids having to manually fiddle with fis commands at the redboot prompt and potentially bricking the unit.
- I'm building binary packages on the units themselves via an NFS root. That works fine - the ports system however needs a bit of hand holding to tell it to use the right endian-ness. Add TARGET_BIG_ENDIAN=true to /etc/make.conf before building any ports.
Stuff that's been fixed
- PCI didn't work unless you boot verbose. You can do this by passing a -v flag to the kernel by using the -c flag to exec. Ie, exec -c "-v". gonzo fixed this in -HEAD by adding a longer delay into the PCI bus probing logic