Category Archives: Embedded

Meraki MS350 hardware overview

The Meraki MS350 (MS350-24 and MS350-48) series switches offer 24 or 48 ports of Gigabit Ethernet. The MS350-24X offers 16 ports of Gigabit Ethernet, and 8 ports of multi-Gigabit (1/2.5/5/10G) Ethernet. All models have four SFP/SFP+ uplink ports, a dedicated remote management port, and stacking capabilities via QSFP. Today we will be looking at the MS350-48 and MS350-24X models specifically.

MS350-48LP from Meraki’s datasheet

Here is a quick summary of the MS350 specs:

  • Intel Atom C2358 CPU (2C/2T, 1.74GHz)
  • 2GB DDR3 ECC RAM (SODIMM)
  • 16MB SPI flash, 2GB NAND flash (TSOP48 NAND on motherboard, USB via Phison)
  • MS350-24X: 30 Network interfaces (16 Gigabit Ethernet, 8 mGig Ethernet, 4 SFP+, 2 QSFP stacking)
  • MS350-48: 54 Network interfaces (48 Gigabit Ethernet, 4 SFP+, 2 QSFP stacking)

The MS350-48 uses the Broadcom BCM56547 (A0) ASIC, with BCM84740 PHYs. The MS350-24X uses the Broadcom BCM56546 (B0) ASIC, with BCM82578 and Aquantia AQR405 PHYs. PoE versions of the switch use the Broadcom BCM59121 PSE controller.

MS350-48:

00:00.0 Host bridge: Intel Corporation Atom processor C2000 SoC Transaction Router (rev 02)
00:01.0 PCI bridge: Intel Corporation Atom processor C2000 PCIe Root Port 1 (rev 02)
00:03.0 PCI bridge: Intel Corporation Atom processor C2000 PCIe Root Port 3 (rev 02)
00:0b.0 Co-processor: Intel Corporation Atom processor C2000 QAT (rev 02)
00:0e.0 Host bridge: Intel Corporation Atom processor C2000 RAS (rev 02)
00:0f.0 IOMMU: Intel Corporation Atom processor C2000 RCEC (rev 02)
00:13.0 System peripheral: Intel Corporation Atom processor C2000 SMBus 2.0 (rev 02)
00:14.0 Ethernet controller: Intel Corporation Ethernet Connection I354 (rev 03)
00:14.1 Ethernet controller: Intel Corporation Ethernet Connection I354 1.0 GbE Backplane (rev 03)
00:14.2 Ethernet controller: Intel Corporation Ethernet Connection I354 (rev 03)
00:14.3 Ethernet controller: Intel Corporation Ethernet Connection I354 (rev 03)
00:1f.0 ISA bridge: Intel Corporation Atom processor C2000 PCU (rev 02)
00:1f.3 SMBus: Intel Corporation Atom processor C2000 PCU SMBus (rev 02)
01:00.0 Ethernet controller: Broadcom Inc. and subsidiaries Device b547 (rev 01)
01:00.1 Ethernet controller: Broadcom Inc. and subsidiaries Device b547 (rev 01)

MS350-24X:

00:00.0 Host bridge: Intel Corporation Atom processor C2000 SoC Transaction Router (rev 02)
00:01.0 PCI bridge: Intel Corporation Atom processor C2000 PCIe Root Port 1 (rev 02)
00:03.0 PCI bridge: Intel Corporation Atom processor C2000 PCIe Root Port 3 (rev 02)
00:0b.0 Co-processor: Intel Corporation Atom processor C2000 QAT (rev 02)
00:0e.0 Host bridge: Intel Corporation Atom processor C2000 RAS (rev 02)
00:0f.0 IOMMU: Intel Corporation Atom processor C2000 RCEC (rev 02)
00:13.0 System peripheral: Intel Corporation Atom processor C2000 SMBus 2.0 (rev 02)
00:14.0 Ethernet controller: Intel Corporation Ethernet Connection I354 (rev 03)
00:14.1 Ethernet controller: Intel Corporation Ethernet Connection I354 1.0 GbE Backplane (rev 03)
00:14.2 Ethernet controller: Intel Corporation Ethernet Connection I354 1.0 GbE Backplane (rev 03)
00:14.3 Ethernet controller: Intel Corporation Ethernet Connection I354 1.0 GbE Backplane (rev 03)
00:1f.0 ISA bridge: Intel Corporation Atom processor C2000 PCU (rev 02)
00:1f.3 SMBus: Intel Corporation Atom processor C2000 PCU SMBus (rev 02)
01:00.0 Ethernet controller: Broadcom Inc. and subsidiaries Device b546 (rev 11)

Both models have the same USB devices present:

Bus 001 Device 002: ID 8087:07db
Bus 001 Device 001: ID 1d6b:0002
Bus 001 Device 003: ID 13fe:5200

MS350-24X and MS350-48 both use coreboot as the bootloader, although the MS350-24X model has a different build. In both cases, the ROM has the following layout:

00000000:00010000 reserved
00010000:0070ffff bk1
00710000:00dfffff bk2
00e00000:00ffffff coreboot

The cbfs contains the following:

FMAP REGION: COREBOOT
ms350-24x_w25q128.bin: 16384 kB, bootblocksize 1024, romsize 16777216, offset 0xe10000
alignment: 64 bytes, architecture: x86

Name                           Offset     Type           Size   Comp
cmos_layout.bin                0xe10000   cmos_layout      1396 none
fallback/romstage              0xe105c0   (unknown)       21624 none
fallback/ramstage              0xe15a80   (unknown)       49421 none
fallback/payload               0xe21c00   simple elf      23042 none
config                         0xe27640   raw              4676 none
revision                       0xe288c0   raw               566 none
(empty)                        0xe28b40   null          1209432 none
mrc.cache                      0xf4ffc0   mrc_cache       65536 none
cpu_microcode_blob.bin         0xf60000   microcode       84992 none
(empty)                        0xf74c40   null            45912 none
fsp.bin                        0xf7ffc0   fsp            389120 none
(empty)                        0xfdf000   null           134040 none

coreboot was built with an ELF payload (miles) which by default loads and jumps into the bootkernel FIT image located at 0x10000. A secondary bootkernel exists on flash at offset 0x710000.

This is very similar to the MX84 as they are both based on the same Rangeley platform.


The entire MS350 series is based on the Intel Atom C2000 series CPU, which Meraki also used in the MX84. Sadly, the MS350 also suffers from the AVR54 errata, as the C2358 in both the MS350-48 and MS350-24X is the B0 revision.

LPC_CLK is exposed on pin 1 of J35, with R3635 carrying 3.3V (MS350-48 and MS350-24X). Therefore, you can add a 100 Ohm resistor between R3635 and pin 1 to pull up the LPC clock. Just be sure to use an “extended-life” resistor for the modification, you wouldn’t want to compromise the MTBF of your Meraki product with anything sub-par 😉

100 Ohm resistor to pull up LPC clock (MS350-24X)


If you wish to flash your MS350, you will need to remove or socket the SOIC8 SPI flash (SK_U1).

This is because there are other devices powered by the +3.3V voltage rail used by SPI flash, which interferes with your ability to read/write the contents of flash. I prefer the Wieson G6179-10 SOIC8 socket (available from Adafruit). People outside the US will probably find it easier to desolder the flash and use a SOIC8 socket with prototype wires, as the G6179-10 is difficult to obtain for a reasonable price.

The UART header is J31 on both the MS350-48 and MS350-24X and follows the standard Meraki UART pinout (1: VCC, 2: Tx, 3: Rx, 4: GND)

Similar to the MS210/225 series, the Broadcom SDK implements the packet engine in userspace, using the linux_kernel_bde and linux_user_bde kernel modules to interface with the ASIC. In the Meraki firmware, the packet engine is a component of the userspace click daemon, which loads the bcm_click shared object during click router initialisation.


Idle power consumption:

  • MS35-48: 54W
  • MS350-24X: 96W

GPL source code for the MS350 was requested from Meraki in July 2023. At the time of writing, they have not provided any. I will update this post with links to the source code when it is provided.

Meraki MX84 overview

I became aware of the Meraki MX84 from Lathe Abusaid’s blog post about tearing down the hardware. After setting up an eBay alert and waiting, I finally won a job lot which included an MX84.


Here is a quick summary of the MX84 specs:

  • Intel Atom C2358 CPU (2C/2T, 1.74GHz)
  • 4GB DDR3 ECC RAM (H5TC4G83CFR-PBA)
  • Internal SATA port (1TB Western Digital Green)
  • External USB2.0 port
  • 13 Network interfaces (Vitesse VSC7425: 11 Gigabit Ethernet, 2 SFP)
  • 16MB SPI flash, 1GB NAND flash (Phison PS2251, USB on motherboard)
  • Fanless
  • Open frame 12V 2.5A power supply

The device runs Linux 3.18.131.

00:00.0 Host bridge: Intel Corporation Atom processor C2000 SoC Transaction Router (rev 02)
00:01.0 PCI bridge: Intel Corporation Atom processor C2000 PCIe Root Port 1 (rev 02)
00:03.0 PCI bridge: Intel Corporation Atom processor C2000 PCIe Root Port 3 (rev 02)
00:0b.0 Co-processor: Intel Corporation Atom processor C2000 QAT (rev 02)
00:0e.0 Host bridge: Intel Corporation Atom processor C2000 RAS (rev 02)
00:0f.0 IOMMU: Intel Corporation Atom processor C2000 RCEC (rev 02)
00:13.0 System peripheral: Intel Corporation Atom processor C2000 SMBus 2.0 (rev 02)
00:14.0 Ethernet controller: Intel Corporation Ethernet Connection I354 1.0 GbE Backplane (rev 03)
00:14.1 Ethernet controller: Intel Corporation Ethernet Connection I354 1.0 GbE Backplane (rev 03)
00:14.2 Ethernet controller: Intel Corporation Ethernet Connection I354 1.0 GbE Backplane (rev 03)
00:14.3 Ethernet controller: Intel Corporation Ethernet Connection I354 1.0 GbE Backplane (rev 03)
00:16.0 USB controller: Intel Corporation Atom processor C2000 USB Enhanced Host Controller (rev 02)
00:17.0 SATA controller: Intel Corporation Atom processor C2000 AHCI SATA2 Controller (rev 02)
00:18.0 SATA controller: Intel Corporation Atom processor C2000 AHCI SATA3 Controller (rev 02)
00:1f.0 ISA bridge: Intel Corporation Atom processor C2000 PCU (rev 02)
00:1f.3 SMBus: Intel Corporation Atom processor C2000 PCU SMBus (rev 02)

The MX84 uses coreboot as the bootloader (coreboot-af6fa06-dirty-Liteon_GRM1001_MFG_v4.0.0; bootlog) and the ROM has the following layout:

00000000:00010000 reserved
00010000:0070ffff bk1
00710000:00dfffff bk2
00e00000:00ffffff coreboot

The cbfs contains the following:

FMAP REGION: COREBOOT
mx84.bin: 16384 kB, bootblocksize 1024, romsize 16777216, offset 0xe00000
alignment: 64 bytes, architecture: x86

Name                           Offset     Type           Size   Comp
cmos_layout.bin                0xe00000   cmos_layout      1352 none
fallback/romstage              0xe00580   (unknown)       25820 none
fallback/ramstage              0xe06ac0   (unknown)       61965 none
fallback/payload               0xe15d40   simple elf      20349 none
config                         0xe1ad00   raw              4310 none
revision                       0xe1be00   raw               712 none
(empty)                        0xe1c100   null          1261208 none
mrc.cache                      0xf4ffc0   mrc_cache       65536 none
cpu_microcode_blob.bin         0xf60000   microcode       84992 none
(empty)                        0xf74c40   null            45912 none
fsp.bin                        0xf7ffc0   spd            389120 none
(empty)                        0xfdf000   null           134040 none

coreboot was built with an ELF payload (miles) which by default loads and jumps into the bootkernel FIT image (dts here) located at 0x10000. A secondary bootkernel exists on flash at offset 0x710000.


Let us revisit those Intel I354 interfaces. As a networking appliance, the MX84 has a lot of network interfaces.

There are 13 network interfaces on the front (Management, Internet 1 & 2, Ethernet 3-10, and 2 SFP cages) so there should be a switch inside the MX84 or we would expect to see more than four interfaces in lspci.

In this case, the switch is the VSC7425, and even if you use the 3.18.131 kernel from Meraki, you won’t have any connectivity because all four of the I354 interfaces connect directly to the VSC7425

The stock Meraki firmware uses a binary called vtss_poca_d to initialise and configure the VSC7425, which does so using a proprietary Vitesse framework (PDF).

vtss_poca_d is a static binary, so could we use it with a newer kernel such as 5.10.146 found in OpenWrt 22.03?

$ vtss_poca_d
mdio_write16: SIOCSMIIREG on eth0 phy:0 reg:0 failed: Not supported
mdio_write16: SIOCSMIIREG on eth0 phy:0 reg:1 failed: Not supported
mdio_write16: SIOCSMIIREG on eth0 phy:0 reg:2 failed: Not supported
mdio_write16: SIOCSMIIREG on eth0 phy:0 reg:3 failed: Not supported
mdio_write16: SIOCSMIIREG on eth0 phy:0 reg:0 failed: Not supported
mdio_write16: SIOCSMIIREG on eth0 phy:0 reg:1 failed: Not supported
mdio_write16: SIOCSMIIREG on eth0 phy:0 reg:2 failed: Not supported
mdio_write16: SIOCSMIIREG on eth0 phy:0 reg:3 failed: Not supported
mdio_write16: SIOCSMIIREG on eth0 phy:0 reg:0 failed: Not supported
mdio_write16: SIOCSMIIREG on eth0 phy:0 reg:1 failed: Not supported
mdio_read16: SIOCGMIIREG on eth0 phy:0 reg:2 failed: Not supported
mdio_read16: SIOCGMIIREG on eth0 phy:0 reg:3 failed: Not supported
mdio_read16: SIOCGMIIREG on eth0 phy:0 reg:2 failed: Not supported
mdio_read16: SIOCGMIIREG on eth0 phy:0 reg:3 failed: Not supported

Nope! As such, it is very unlikely this device will ever be supported by OpenWrt.


You may have noticed that the MX84 is based on the Atom C2000, a CPU which suffers from the AVR54 errata. When I first received my MX84, there was no output on UART and the power consumption was a suspiciously consistent 6W, all the hallmarks of a device dead from AVR54. There are numerous instructions for how to revive a Synology NAS with a dead Atom, but no such instructions exist for the MX84.

Fortunately for me, there was a photo from Lathe Abusaid’s blog post which provided a crucial hint. It appears that the MX84 unit photographed in their teardown includes a 100 Ohm resistor between pins 1 (LPC clock) and 9 (3.3V) of header J7.

This solution appears to have been chosen because it was the most convenient for Cisco, however note that pin 9 of J7 appears to be a GPIO output which, depending on the coreboot payload, may not always be active high. I would suggest instead soldering to pin 8 of the unpopulated SOIC8 nearby (U47), which will provide 3.3V regardless of the payload GPIO configuration.

That being said, here is a photo of the resistor fix to J7 that I just advised against doing, taken before realizing pin 9 was a GPIO

After soldering the pull-up resistor, the LPC clock is back on pin 1 of J7:

Oscilloscope output of the LPC clock on pin 1 of J7 after adding a 100Ω resistor


There is an unpopulated footprint for a Micro-USB port on the left side of the motherboard. By default the D-/D+ are not connected, as the 0 Ohm resistors are unpopulated (they are instead populated on R467/R468 connecting the Phison ps2303q to the Atom CPU). I believe this port was used during development to easily swap the USB drive connected to the SoC.


One question I had about the MX84 was: why coreboot? It seems that this design is based on Intel’s “Mohon Peak” reference platform. From the Intel customer reference board (CRB) documentation (PDF):

The embedded firmware ecosystem has developed an example boot loader solution
for the CRB that uses the FSP kit. This solution is based on the open source Coreboot
project at coreboot.org. While Intel does not endorse or support boot loader solutions
based on the Coreboot project, the example Coreboot-based boot loader provides a
good teaching model for how to integrate the Intel FSP into a complete boot loader
solution.

Now it is clear why Meraki chose to use coreboot, that is simply the bootloader reference provided by Intel for Mohon Peak. Other manufacturers who made Atom C2000 products also used coreboot (such as the VeloCloud 520-AC).

Meraki provided the coreboot source code in December 2022, after a delay of more than 12 months. The coreboot source code for the MX84 is available on GitHub.


Meraki hardware commanding the premium that it does, if you are considering buying an MX84: don’t. The VeloCloud 520-AC (C2358) and 540-AC (C2558) are available for ~$30 on eBay and have the C0 revision which doesn’t suffer from AVR54.

If you already own an MX84 and want to poke around, here is a buildroot based firmware that you flash to SPI. The firmware will boot, initialize the switch, DHCP, and start an SSH server (the root password is the device serial without hyphens). Note that it is initramfs based, so no changes are persisted.

Caveat emptor: VeloCloud devices have an issue with the igb/I354 compatibility, meaning that only the two SFP cages are functional. However, that is two more interfaces than you will get from the MX84 (zero) with any other kernel.

Fujitsu iRMC S4 License

A few years ago we looked at iRMC S4 on the Fujitsu TX140 S2. iRMC S4 provides typical remote management features that you would expect to find in a BMC: remote power control, sensor monitoring and alerting, hardware inventory, and boot order over-ride/selection. Some additional features like the remote KVM and remote media require a license key.

Licensed IPMI features are not new and other vendors, such as Supermicro, have had their IPMI license reverse engineered.

Fujitsu are a somewhat niche vendor when it comes to servers, and to date I am not aware that anyone has publicly reverse engineered the iRMC S4 license.


They say a picture is worth a thousand words, so we will start with a diagram

iRMC S4 license contents

An iRMC S4 license has four distinct fields

  1. Header/magic: 4 bytes (iRMC)
  2. Features to be enabled by the license (bitmask): 4 bytes
  3. Type of license (temporary or permanent): 4 bytes
  4. CRC32 of the system serial number: 4 bytes

The above data is encrypted using AES-128, and the output is base32 encoded with hyphens every 4 characters.

For example, here is an iRMC S4 license (enabling KVM and remote media) for an RX chassis with the serial number YLNS012345:

ZKAF-Z5EG-PL5G-6GFR-YEG6-CKGM-KQ

And the actual license contents:

69524d43 0300000 0ffffff05 2e4dbb51

Licensed features in iRMC S4 include:

  • Remote KVM
  • Remote media
  • eLCM

Feature bit 1 is for KVM, bit 2 remote media, and bit 3 seems to be for eLCM (eLCM appears to only be available on some models).

Installing an iRMC S4 license on a TX chassis


Back in 2014, Fujitsu changed the iRMC S4 licensing to be “node-locked”, which means that a license is tied to a specific server and cannot be transferred. The installation of a volume license is not possible after 2015-01-01 00:00:00.

iRMC S4 tracks the “Power on Hours (PoH)” of the chassis, and it appears that there is the capability to generate a temporary license which will expire after a certain number of Power on Hours is reached, probably to provide customers with time to evaluate the value proposition of purchasing iRMC licenses.

iRMC S4 time limited license

If you are reading this, then you are probably not interested in generating temporary licenses. Setting the field to 0xffffff00 for a TX chassis and 0xffffff05 for an RX chassis will result in a permanent license.


Now that we have covered the fields in an unencrypted iRMC S4 license, it will be obvious that the example license ZKAF-Z5EG-PL5G-6GFR-YEG6-CKGM-KQ is not simply the base32 encoded binary license data.

Unlike Supermicro, Fujitsu use a static HMAC message and key to create an HMAC-SHA1 hash, the first 16 bytes of which are used as the key for AES-128. The AES encrypted data is then base32 encoded and the output is the iRMC license you install via the web interface.

I will not be disclosing Fujitsu’s HMAC key and AES IV here, but suffice to say you can download and unpack the iRMC firmware from Fujitsu and find the values in /usr/local/lib/libfts_license.so.1.12.1. Thanks Fujitsu!


For anyone interested in reverse engineering the iRMC S4 license validation themselves:

  • the HMAC key and message are used in lkeyInitCipherKey in libfts_license
  • the AES IV is used in decrypt_with_license in libfts_license

libfts_license in Ghidra, showing decompiled function and hexdump

Anyone looking for a simpler solution, a proof-of-concept for python is here. Note that you need to provide the correct HMAC/AES values obtained from libfts_license.


To anyone wondering, the license logic from iRMC S4 is not applicable to older iRMC platforms such as iRMC S2 or iRMC S3.

However, the license logic appears to be unchanged between iRMC S4 and S5. Hardware with iRMC S5 is too expensive to justify purchasing to verify this, but maybe someone will leave a comment as to whether the license logic described here is still applicable to iRMC S5.

Edit: An anonymous reader has written to say that the logic is unchanged for iRMC S5 ✨