Tag Archives: luton26

Meraki MS220: PoE support

The last several posts in this series have focused primarily on getting a custom firmware running on the Meraki MS220-series switches, without much regard for preserving existing features. Since I am now at a point where my custom firmware is functional as a Layer 2-ish switch, my attention has turned to PoE, since many switches in the series have PoE support and that is feature I think switch owners (especially MS220-8P) are interested in.

From my investigation into libpoecore included in the Meraki firmware, PoE on the MS220-8P appears to be managed by Microsemi’s PD690xx series of Power over Ethernet Management chips (datasheet). The PD690xx series communicates over I2C with the management CPU to manage PoE on the switch ports (enable/disable PoE, set 802.3af/at modes, query power consumed by a PoE device).

We can confirm that the PD690xx communicates via I2C by running poe_server from Meraki’s firmware and enabling I2C tracing in the kernel:

# cat /sys/kernel/debug/tracing/trace
# tracer: nop
#
# entries-in-buffer/entries-written: 1358/1358   #P:1
#
#                              _-----=> irqs-off
#                             / _----=> need-resched
#                            | / _---=> hardirq/softirq
#                            || / _--=> preempt-depth
#                            ||| /     delay
#           TASK-PID   CPU#  ||||    TIMESTAMP  FUNCTION
#              | |       |   ||||       |         |
      poe_server-682   [000] ....   560.356000: i2c_write: i2c-1 #0 a=030 f=0000 l=4 [13-32-0f-ff]
      poe_server-682   [000] ....   560.358000: i2c_result: i2c-1 n=1 ret=1
      poe_server-682   [000] ....   560.358000: i2c_write: i2c-1 #0 a=030 f=0000 l=2 [13-32]
      poe_server-682   [000] ....   560.358000: i2c_read: i2c-1 #1 a=030 f=0001 l=2
      poe_server-682   [000] ....   560.359000: i2c_reply: i2c-1 #1 a=030 f=0001 l=2 [0f-ff]
      poe_server-682   [000] ....   560.359000: i2c_result: i2c-1 n=2 ret=2
      poe_server-682   [000] ....   560.359000: i2c_write: i2c-1 #0 a=030 f=0000 l=4 [13-32-0f-ff]
      poe_server-682   [000] ....   560.360000: i2c_result: i2c-1 n=1 ret=1
      poe_server-682   [000] ....   560.360000: i2c_write: i2c-1 #0 a=030 f=0000 l=4 [13-9e-dc-03]
      poe_server-682   [000] ....   560.362000: i2c_result: i2c-1 n=1 ret=1

I2C tracing is extremely helpful, as running strace against poe_server directly will not yield useful output as to what operations it is performing to configure PoE.

While it is good news that we are able to recover the I2C commands via kernel tracing, it’s bad news in the sense that writing a new daemon to duplicate the features of poe_cli is non-trivial.


Thankfully, with the libpoecore from the Meraki firmware dump and free disassembly tools like Ghidra (sorry Hex-Rays, support MIPS in IDA Free ¯\_(ツ)_/¯), understanding some of the logic behind functionality provided by poe_server and poe_cli becomes much easier.

If you disassemble libpoecore, you can find the function hard_init which contains code to set up GPIO outputs. Interesting to note is that while the GPIO pins change depending on which switch ASIC is present, the sequence of GPIO outputs to configure the PD690xx remains constant.

Disassembler view of the function hard_init from the library libpoecore.so

The same GPIO configuration is executed when switch_brain is started (full strace output):

writev(1, [{iov_base="", iov_len=0}, {iov_base="echo 7 > /sys/class/gpio/export\n", iov_len=32}], 2echo 7 > /sys/class/gpio/export) = 32
writev(1, [{iov_base="", iov_len=0}, {iov_base="echo 12 > /sys/class/gpio/export\n", iov_len=33}], 2echo 12 > /sys/class/gpio/export) = 33
writev(1, [{iov_base="", iov_len=0}, {iov_base="echo out > /sys/class/gpio/gpio7/direction\n", iov_len=43}], 2echo out > /sys/class/gpio/gpio7/direction) = 43
writev(1, [{iov_base="", iov_len=0}, {iov_base="echo out > /sys/class/gpio/gpio12/direction\n", iov_len=44}], 2echo out > /sys/class/gpio/gpio12/direction) = 44
writev(1, [{iov_base="", iov_len=0}, {iov_base="echo 1 > /sys/class/gpio/gpio7/value\n", iov_len=37}], 2echo 1 > /sys/class/gpio/gpio7/value) = 37
writev(1, [{iov_base="", iov_len=0}, {iov_base="echo 0 > /sys/class/gpio/gpio12/value\n", iov_len=38}], 2echo 0 > /sys/class/gpio/gpio12/value) = 38
writev(1, [{iov_base="", iov_len=0}, {iov_base="echo 1 > /sys/class/gpio/gpio12/value\n", iov_len=38}], 2echo 1 > /sys/class/gpio/gpio12/value) = 38
writev(1, [{iov_base="", iov_len=0}, {iov_base="echo 0 > /sys/class/gpio/gpio7/value\n", iov_len=37}], 2echo 0 > /sys/class/gpio/gpio7/value) = 37
writev(1, [{iov_base="", iov_len=0}, {iov_base="echo 1 > /sys/class/gpio/gpio12/value\n", iov_len=38}], 2echo 1 > /sys/class/gpio/gpio12/value) = 38

The datasheet for the luton26 ASIC used in the MS220-8P, MS220-24P, and MS22 (VDMS-10393) doesn’t list anything connected to GPIO 7, and GPIO 12 is used for either SFP17_SD or PHY7_LED1 depending on the overlay function chosen. The functionality of these GPIO pins is undefined in the ASIC datasheet, however libpoecore is setting them and manipulating their outputs.

We can implement the logic of hard_init in an init script to set up the GPIO pins in the same way, and the result is that the PD690xx is configured for auto mode. I am not sure how, there is nothing in the PD690xx datasheet which suggests GPIO pins can be used to configure the operating mode, but the switch will automatically negotiate and power a PoE device.

Writing a new daemon to communicate with the PD690xx will ultimately be necessary if fine control over PoE functionality is to be achieved. Without I2C communication to the PD690xx, it is not possible to query the power budget, or limit port power delivery. In the mean time, for those who do not mind unmanaged “plug-and-play” style, PoE can be considered functional.

MS220-8P: Custom firmware from scratch

To follow up on my previous posts about modifying the firmware for the Cisco Meraki MS220-8P, I have more progress to report.

Since I could not figure out how to fix serial output from userspace when booting from u-boot, I went back to booting with RedBoot and tried to focus on improving userspace from my first sloppy attempt to coerce the Meraki firmware into booting from NOR.

Here is the current situation:

  • RedBoot without CRC or kernel size checks
  • Kernel 3.18.123 with xz compression
  • Busybox userspace based on buildroot 2020.02.1
  • Meraki kernel modules: vtss_core, proclikefs, merakiclick, elts_meraki, and vc_click are successfully loaded during boot
  • The click router is successfully initialized, creating the interfaces arping, linux_mgmt, sw0_pcap, and wired0

However, networking is non-functional. The switch broadcasts DHCP requests, but no packets are ever passed from the switching ASIC to the management CPU:

arping    Link encap:Ethernet  HWaddr 88:15:44:73:22:31
          inet addr:169.254.55.143  Bcast:169.254.255.255  Mask:255.255.0.0
          inet6 addr: fe80::8a15:44ff:fe73:2231/64 Scope:Link
          UP BROADCAST RUNNING MULTICAST  MTU:1500  Metric:1
          RX packets:0 errors:0 dropped:0 overruns:0 frame:0
          TX packets:33 errors:0 dropped:0 overruns:0 carrier:0
          collisions:0 txqueuelen:0
          RX bytes:0 (0.0 B)  TX bytes:5546 (5.4 KiB)

linux_mgmt Link encap:UNSPEC  HWaddr 00-00-00-00-00-00-00-00-00-00-00-00-00-00-00-00
          UP POINTOPOINT RUNNING NOARP MULTICAST  MTU:1500  Metric:1
          RX packets:0 errors:0 dropped:0 overruns:0 frame:0
          TX packets:0 errors:0 dropped:0 overruns:0 carrier:0
          collisions:0 txqueuelen:0
          RX bytes:0 (0.0 B)  TX bytes:0 (0.0 B)

lo        Link encap:Local Loopback
          inet addr:127.0.0.1  Mask:255.255.255.255
          inet6 addr: ::1/128 Scope:Host
          UP LOOPBACK RUNNING  MTU:65536  Metric:1
          RX packets:0 errors:0 dropped:0 overruns:0 frame:0
          TX packets:0 errors:0 dropped:0 overruns:0 carrier:0
          collisions:0 txqueuelen:0
          RX bytes:0 (0.0 B)  TX bytes:0 (0.0 B)

sw0_pcap  Link encap:Ethernet  HWaddr 00:01:02:03:04:05
          inet addr:169.254.83.119  Bcast:169.254.255.255  Mask:255.255.0.0
          inet6 addr: fe80::201:2ff:fe03:405/64 Scope:Link
          UP BROADCAST RUNNING MULTICAST  MTU:1500  Metric:1
          RX packets:0 errors:0 dropped:0 overruns:0 frame:0
          TX packets:32 errors:0 dropped:0 overruns:0 carrier:0
          collisions:0 txqueuelen:0
          RX bytes:0 (0.0 B)  TX bytes:5476 (5.3 KiB)

wired0    Link encap:UNSPEC  HWaddr 00-00-00-00-00-00-00-00-00-00-00-00-00-00-00-00
          UP POINTOPOINT RUNNING NOARP MULTICAST  MTU:1500  Metric:1
          RX packets:0 errors:0 dropped:0 overruns:0 frame:0
          TX packets:0 errors:0 dropped:0 overruns:0 carrier:0
          collisions:0 txqueuelen:0
          RX bytes:0 (0.0 B)  TX bytes:0 (0.0 B)

On the other side:

tcpdump: listening on eth1, link-type EN10MB (Ethernet), capture size 262144 bytes
06:02:32.423798 IP6 (hlim 1, next-header Options (0) payload length: 36) fe80::201:c0ff:fe1f:9fb2 > ff02::16: HBH (rtalert: 0x0000) (padn) [icmp6 sum ok] ICMP6, multicast listener report v2, 1 group record(s) [gaddr ff02::1:ff1f:9fb2 to_ex { }]
06:02:33.097129 IP6 (hlim 1, next-header Options (0) payload length: 36) fe80::201:c0ff:fe1f:9fb2 > ff02::16: HBH (rtalert: 0x0000) (padn) [icmp6 sum ok] ICMP6, multicast listener report v2, 1 group record(s) [gaddr ff02::1:ff1f:9fb2 to_ex { }]
06:03:20.969517 IP (tos 0x0, ttl 64, id 20247, offset 0, flags [none], proto UDP (17), length 332, bad cksum 2a8b (->1677)!)
    0.0.10.10.68 > 10.10.255.255.67: [bad udp cksum 0x17ca -> 0x03b6!] BOOTP/DHCP, Request from 88:15:44:21:65:10 (oui Unknown), length 304, xid 0xd3650d76, secs 123, Flags [none] (0x0000)
          Client-Ethernet-Address 88:15:44:21:65:10 (oui Unknown)
          Vendor-rfc1048 Extensions
            Magic Cookie 0x63825363
            DHCP-Message Option 53, length 1: Discover
            Client-ID Option 61, length 15: hardware-type 255, 44:21:65:10:00:03:00:01:88:15:44:21:65:10
            SLP-NA Option 80, length 0""
            NOAUTO Option 116, length 1: Y
            MSZ Option 57, length 2: 1472
            Hostname Option 12, length 13: "m881544216510"
            T145 Option 145, length 1: 1
            Parameter-Request Option 55, length 14:
              Subnet-Mask, Classless-Static-Route, Static-Route, Default-Gateway
              Domain-Name-Server, Hostname, Domain-Name, MTU
              BR, Lease-Time, Server-ID, RN
              RB, Option 119
            END Option 255, length 0
06:04:25.620540 IP (tos 0x0, ttl 64, id 13784, offset 0, flags [none], proto UDP (17), length 332, bad cksum 43ca (->2fb6)!)
    0.0.10.10.68 > 10.10.255.255.67: [bad udp cksum 0x178a -> 0x0376!] BOOTP/DHCP, Request from 88:15:44:21:65:10 (oui Unknown), length 304, xid 0xd3650d76, secs 187, Flags [none] (0x0000)
          Client-Ethernet-Address 88:15:44:21:65:10 (oui Unknown)
          Vendor-rfc1048 Extensions
            Magic Cookie 0x63825363
            DHCP-Message Option 53, length 1: Discover
            Client-ID Option 61, length 15: hardware-type 255, 44:21:65:10:00:03:00:01:88:15:44:21:65:10
            SLP-NA Option 80, length 0""
            NOAUTO Option 116, length 1: Y
            MSZ Option 57, length 2: 1472
            Hostname Option 12, length 13: "m881544216510"
            T145 Option 145, length 1: 1
            Parameter-Request Option 55, length 14:
              Subnet-Mask, Classless-Static-Route, Static-Route, Default-Gateway
              Domain-Name-Server, Hostname, Domain-Name, MTU
              BR, Lease-Time, Server-ID, RN
              RB, Option 119
            END Option 255, length 0

The primary reason for this appears to be because Meraki is using the click modular router. Click seems to be an attempt to lobotomize the Linux kernel networking stack for the sake of writing academic research papers. The kernel side of click was never upstreamed to mainline and is no longer maintained. Let us pour one out for the poor soul at Cisco who has to maintain this.

Since Meraki was spun out of MIT and the researchers who wrote Click were also at MIT, the relationship between Meraki and Click is clear. Finally Click found a practical use, in a now-obsolete product line from Cisco.

More reverse engineering is necessary to make Click functional enough to have basic connectivity to the management CPU.


The strace output of switch_brain is available in this gist. By first running switch_brain, and then overwriting the default traffic rules with “allow all” I was able to SSH to the management CPU:

/ # echo "allow all" > /click/nat/common_switch_nat/from_smc_filter/config
/ # echo "allow all" > /click/nat/common_switch_nat/from_mgmt_filter/config
/ # tail /tmp/messages | grep dropbear
Jan  1 00:15:49 buildroot authpriv.info dropbear[680]: Child connection from 10.10.10.1:39248
Jan  1 00:15:50 buildroot authpriv.notice dropbear[680]: Password auth succeeded for 'root' from 10.10.10.1:39248
/ # w
USER            TTY             IDLE    TIME             HOST
root            pts/0           00:00   Jan  1 00:15:50  10.10.10.1

I think it is now very clear that the “only” thing blocking full access to the management CPU is the Click configuration. My hope is to build the configuration from the switch_brain strace output and package that into an initscript.

I have updated the meraki-builder repository with the buildroot config file and overlay directory I use to inject init scripts (among other things).

If you would like to contribute, I have written some installation instructions for the current development firmware. If you find the correct voodoo of Click commands to access the management CPU, please open a PR 🥰 The voodoo has been found.

u-boot, boot args, and compressed kernels on the Meraki MS220-8P

The last time we looked at the Cisco Meraki MS220, it was possible to boot kernel 3.18.123 compiled from the Cisco GPL archive. The ability to compile the kernel, with our own command line and modules opened up new possibilities for firmware modification.

That was a very good start, but there is more improvement possible. Support for the Microsemi VCore III SoCs was added to u-boot in late 2018. The MS220-8P contains the Luton26 SoC, but with only 10 ports. The MS220-8P appears to follow the Microsemi PCB090 reference design, and it is possible to build and boot u-boot (v2019.10):

U-Boot 2019.10 (Apr 04 2020 - 09:26:23 +0000)

MSCC VCore-III MIPS 24Kec
Model: Luton26 PCB090 Reference Board
DRAM:  128 MiB
Loading Environment from SPI Flash... SF: Detected mx25l12805d with page size 256 Bytes, erase size 64 KiB, total 16 MiB
OK
In:    serial@10100000
Out:   serial@10100000
Err:   serial@10100000
Net:   Could not get PHY for miim-bus1: addr 0
Could not get PHY for miim-bus1: addr 1
Could not get PHY for miim-bus1: addr 2
Could not get PHY for miim-bus1: addr 3
Could not get PHY for miim-bus1: addr 4
Could not get PHY for miim-bus1: addr 5
Could not get PHY for miim-bus1: addr 6
Could not get PHY for miim-bus1: addr 7
Could not get PHY for miim-bus1: addr 8
Could not get PHY for miim-bus1: addr 9
Could not get PHY for miim-bus1: addr 10
Could not get PHY for miim-bus1: addr 11

Warning: switch@1010000 (eth0) using random MAC address - 62:bf:42:f4:e4:5f
eth0: switch@1010000
Hit any key to stop autoboot:  0 
luton #

u-boot for the Luton26 specifies the environment to be saved at 1MB. This does not match my desired flash layout, so I redefined the ENV_OFFSET and associated variables in board/mscc/luton/Kconfig:

# 64KiB
config ENV_SECT_SIZE
        default 0x10000
# 128KiB at 512KiB
config ENV_SIZE
        default 0x20000
# at 512KiB
config ENV_OFFSET
        default 0x80000 if ENV_IS_IN_SPI_FLASH

Though for some reason this is not working as anticipated, and the environment is still corrupted after saveenv (╯°□°)╯︵ ┻━┻


With u-boot, I was finally able to test booting a compressed kernel. I wonder why Meraki is not compressing the kernel, since compression does not appear to slow down boot at all, and the compressed kernel is much smaller than an uncompressed kernel. Using xz compression, the 3.18.123 kernel is ~2MB, instead of ~5MB. Some fairly trivial modifications were necessary to achieve this.


With u-boot support, testing is much easier. u-boot has network support on the Luton26, making it possible to tftpboot a kernel. This vastly increases the speed of testing as you don’t have to write the kernel to SPI flash before testing changes.

Getting the handoff between u-boot and the kernel for the command line was difficult to figure out. The original bootloader, Redboot, doesn’t have the kernel command line built-in the command line is always compiled into the kernel. In that context, it’s logical why Vitesse didn’t include support for reading the bootloader command line in the kernel (and why Cisco didn’t modify the kernel to include support, since they’re shipping Redboot in these switches).

u-boot was placing the variables somewhere in RAM, but for some reason the kernel wasn’t getting them in its boot command line. u-boot has a humorously titled page on exactly this issue: Linux Kernel Ignores my bootargs. After digging through the u-boot source, it was clear that argv are being passed at the start of DRAM at 0x80000000:

in include/configs/vcoreiii.h: 
#define CONFIG_SYS_SDRAM_BASE           0x80000000
in board/mscc/luton/luton.c:
gd->bd->bi_boot_params = CONFIG_SYS_SDRAM_BASE;

After much searching through the 3.18.123 kernel source code, it became clear that prom is typically where the boot args receiving code is typically placed, so I took inspiration from the lantiq prom code to copy the bootloader argc and argv.

Now, it’s possible to tftpboot compressed kernels with boot args!


Except… there is no serial output from userspace. printk messages are displayed on serial, but no userspace programs (e.g. init) seem able to print out /dev/ttyS0 when using u-boot.

Compiling the kernel with /dev/ttyprintk support and adding custom scripts in /etc/init.d shows that userspace is indeed alive:

[    4.779000] [U] S05printk starting; userspace is alive
[    5.689000] random: dropbear urandom read with 85 bits of entropy available
[    7.356000] [U] Hello from S51printk; userspace is alive but serial is broken

/dev/ttyS0 seems normal, and the memory map is the same as RedBoot (which is logical, since it’s the same kernel version):

[    7.706000] [U] S99 starting; we print stuff
[    7.712000] [U] Now we print /dev
...
[    8.150000] [U] crw-------    1 root     root        4,  64 Jan  1 00:00 ttyS0
[    8.159000] [U] crw-------    1 root     root        5,   3 Jan  1 00:00 ttyprintk
[    8.207000] [U] Now we print /proc/iomem
[    8.222000] [U] 00000000-07feffff : System RAM
[    8.227000] [U]   00000000-00000000 : Crash kernel
[    8.232000] [U]   00100000-0048cc83 : Kernel code
[    8.237000] [U]   0048cc84-0057610f : Kernel data
[    8.241000] [U] 40000000-4fffffff : Serial interface
[    8.247000] [U] 50000000-5fffffff : Parallel interface
[    8.252000] [U]   50000000-50000010 : gen_nand.0
[    8.256000] [U]     50000000-50000010 : gen_nand.0
[    8.261000] [U] 60000000-60ffffff : Switch registers
[    8.266000] [U] 70000000-701fffff : CPU Registers
[    8.271000] [U]   70100000-7010001f : serial
[    8.279000] [U] Now we print /proc/vmallocinfo
[    8.294000] [U] 0xc0000000-0xc1001000 16781312 __ioremap+0x128/0x4ac ioremap
[    8.301000] [U] 0xc1004000-0xc1009000   20480 jffs2_lzo_init+0x18/0x84 pages=4 vmalloc
[    8.309000] [U] 0xc1009000-0xc100c000   12288 jffs2_lzo_init+0x24/0x84 pages=2 vmalloc
[    8.317000] [U] 0xc100c000-0xc100e000    8192 __ioremap+0x128/0x4ac ioremap
[    8.324000] [U] 0xc100e000-0xc1010000    8192 __ioremap+0x128/0x4ac ioremap
[    8.331000] [U] 0xc1010000-0xc1035000  151552 deflate_init+0x44/0xf8 pages=36 vmalloc
[    8.339000] [U] 0xc1035000-0xc1041000   49152 deflate_init+0x98/0xf8 pages=11 vmalloc
[    8.347000] [U] 0xc1041000-0xc1045000   16384 n_tty_open+0x2c/0x14c pages=3 vmalloc
[    8.355000] [U] 0xc1045000-0xc1066000  135168 xz_dec_lzma2_create+0x54/0x9c pages=32 vmalloc
[    8.363000] [U] 0xc1080000-0xc1281000 2101248 __ioremap+0x128/0x4ac ioremap
[    8.370000] [U] 0xc1285000-0xc1289000   16384 n_tty_open+0x2c/0x14c pages=3 vmalloc
[   97.943000] random: nonblocking pool is initialized

I haven’t solved the issue yet of why serial output is broken in userspace 😔
If anyone has experienced broken serial in userspace but not for kernel printk, I’d love to hear how you resolved the problem!