Those following the source-changes mailing list closely may have noticed several evbarm kernels getting "options MULTIPROCESSOR" in the last few days. This is due to those configurations now running properly in SMP mode, thanks to work mostly done by Matt Thomas and Nick Hudson.

The list of supported multiprocessor boards currently is:

  • Banana Pi (BPI)
  • Cubieboard 2 (CUBIEBOARD)
  • Cubietruck (CUBIETRUCK)
  • Merrii Hummingbird A31 (HUMMINGBIRD_A31)
  • CUBOX-I
  • NITROGEN6X

Details how to create bootable media and various other information for the Allwinner boards can be found on the NetBSD/evbarm on Allwinner Technology SoCs wiki page.

The release engineering team is discussing how to bring all those changes into the netbsd-7 branch as well, so that we can call NetBSD 7.0 "the ARM SoC release".

While multicore ARM chips are mostly known for being used in cell phones and tablet devices, there are also some nice "tiny PC" variants out there, like the CubieTruck, which originally comes with a small transparent case that allows piggybacking it onto a 2.5" hard disk:


Image from cubieboard.org under creative commons license.

This is nice to put next to your display, but a bit too tiny and fragile for my test lab - so I reused an old (originally mac68k cartridge) SCSI enclosure for mine:


Image by myself under creative commons license.

This machine is used to run regular tests for big endian (!) arm, the results are gathered here. Running it big-endian is just a way to trigger more bugs.

The last test run logged on the page is already done with an SMP kernel. No regressions were found so far, and the other bugs (sligtly more than 30 failures in the test run is way too much) will be addressed one by one.

Now happy multi-ARM-ing everyone, and I am looking forward to a great NetBSD 7.0 release!

Posted at lunch time on Thursday, November 6th, 2014 Tags:

The NetBSD Project is pleased to announce NetBSD 5.1.5, the fifth security/bugfix update of the NetBSD 5.1 release branch, and NetBSD 5.2.3, the third security/bugfix update of the NetBSD 5.2 release branch. They represent a selected subset of fixes deemed important for security or stability reasons, and if you are running a prior release of either branch, we strongly suggest that you update to one of these releases.

For more details, please see the NetBSD 5.1.5 release notes or NetBSD 5.2.3 release notes.

Complete source and binaries for NetBSD are available for download at many sites around the world. A list of download sites providing FTP, AnonCVS, SUP, and other services may be found at http://www.NetBSD.org/mirrors/.

Posted Friday evening, November 21st, 2014 Tags:
When I found out that Ingenic is giving away some of their MIPS Creator CI20 boards I applied, and to my surprise they sent me one. Of course, the point was to make NetBSD work on it. I just finished the first step.

That is, make it load a kernel, identify / setup the CPU, attach a serial console. This is what it looks like:

U-Boot SPL 2013.10-rc3-g9329ab16a204 (Jun 26 2014 - 09:43:22)
SDRAM H5TQ2G83CFR initialization... done


U-Boot 2013.10-rc3-g9329ab16a204 (Jun 26 2014 - 09:43:22)

Board: ci20 (Ingenic XBurst JZ4780 SoC)
DRAM: 1 GiB
NAND: 8192 MiB
MMC: jz_mmc msc1: 0
In: eserial3
Out: eserial3
Err: eserial3
Net: dm9000
ci20# dhcp
ERROR: resetting DM9000 -> not responding
dm9000 i/o: 0xb6000000, id: 0x90000a46
DM9000: running in 8 bit mode
MAC: d0:31:10:ff:7e:89
operating at 100M full duplex mode
BOOTP broadcast 1
DHCP client bound to address 192.168.0.47
*** Warning: no boot file name; using 'C0A8002F.img'
Using dm9000 device
TFTP from server 192.168.0.44; our IP address is 192.168.0.47
Filename 'C0A8002F.img'.
Load address: 0x88000000
Loading: #################################################################
##############
284.2 KiB/s
done
Bytes transferred = 1146945 (118041 hex)
ci20# bootm
## Booting kernel from Legacy Image at 88000000 ...
Image Name: evbmips 7.99.1 (CI20)
Image Type: MIPS NetBSD Kernel Image (gzip compressed)
Data Size: 1146881 Bytes = 1.1 MiB
Load Address: 80020000
Entry Point: 80020000
Verifying Checksum ... OK
Uncompressing Kernel Image ... OK
subcommand not supported
ci20# g 80020000
## Starting applicatpmap_steal_memory: seg 0: 0x30c 0x30c 0xffff 0xffff
Loaded initial symtab at 0x802502d4, strtab at 0x80270cb4, # entries 8323
Copyright (c) 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003, 2004, 2005,
2006, 2007, 2008, 2009, 2010, 2011, 2012, 2013, 2014
The NetBSD Foundation, Inc. All rights reserved.
Copyright (c) 1982, 1986, 1989, 1991, 1993
The Regents of the University of California. All rights reserved.

NetBSD 7.99.1 (CI20) #113: Sat Nov 22 09:58:39 EST 2014
ml@blackbush:/home/build/obj_evbmips32/sys/arch/evbmips/compile/CI20
Ingenic XBurst
total memory = 1024 MB
avail memory = 1001 MB
kern.module.path=/stand/evbmips/7.99.1/modules
mainbus0 (root)
cpu0 at mainbus0: 1200.00MHz (hz cycles = 120000, delay divisor = 12)
cpu0: Ingenic XBurst (0x3ee1024f) Rev. 79 with unknown FPC type (0x330000) Rev. 0
cpu0: 32 TLB entries, 16MB max page size
cpu0: 32KB/32B 8-way set-associative L1 instruction cache
cpu0: 32KB/32B 8-way set-associative write-back L1 data cache
com0 at mainbus0: Ingenic UART, working fifo
com0: console

root device:

What works:

  • CPU identification and setup
  • serial console via UART0
  • reset ( by provoking a watchdog timeout )
  • basic timers - enough for delay(), since the CPUs don't have MIPS cycle counters
  • dropping into ddb and poking around

What doesn't work (yet):

  • interrupts
  • everything else

Biggest obstacle - believe it or not, the serial port. The on-chip UARTs are mostly 16550 compatible. Mostly. The difference is one bit in the FIFO control register which, if not set, powers down the UART. So throwing data at the UART by hand worked but as soon as the com driver took over the line went dead. It took me a while to find that one.
Posted Sunday afternoon, November 23rd, 2014 Tags:
Add a comment