[Mesh] Meraki routers and watchdog problems

[Charles N Wyble]

I came across instructions to disable the watchdog. I will send it to the list when I'm back at my computer.

Look through the links I sent you on Friday night, I believe it's in one of those. Anyway, I will try to run it down soon.

Marc Juul <juul at labitat.dk> wrote:
>As some of you may already know, Noisebridge received a donation of
>somewhere in the neighborhood of 250 "Meraki Outdoor" routers. I moved
>most
>of them to sudo mesh so they wouldn't disappear from Noisebridge. I
>expect
>we'll have around 150 of them once we've given some out to everyone who
>has
>declared interest.
>
>They're nearly the same specs and chipset as the picostations. The
>chipset
>is an AR2317 instead of an AR2315 and it has an extra ethernet adapter
>and
>only 200 mW of output power.
>
>Here's a datasheet:
>
>  https://meraki.cisco.com/lib/pdf/meraki_datasheet_outdoor.pdf
>
>To use them, we need to solve four problems:
>
>* Antennas. They use the same connectors as mini-pci cards. I took
>apart
>a broken laptop at sudo room and extracted two antennas, but that's
>probably not a viable solution for all of them. We need a source for
>these
>antennas, but I don't know the official name of that type of connector.
>
>* PoE injectors. They take 5 to 22 volts.
>
>* Cases: For indoor use we can use dollar store food storage cases. For
>outdoor we can use cases design for outdoor electrical installations
>(Mitar's suggestion). If someone's handy with 3D modeling we could also
>3D
>print cases. PLA is fine for indoor, but for outdoor we need ABS
>plastic
>which requires a heated print bed.
>
>* Software: I've been trying to get OpenWRT running on these, but it
>has
>proven difficult.
>
>The watchdog reboots the routers after 5 minutes and nothing can be
>done to
>stop that from within the built-in RedBoot. Unfortunately it takes
>longer
>than 5 minutes to flash them with OpenWRT. I've written a script that
>flashes the routers in four stages, using a serial console and tftp
>over
>ethernet and it works:
>
>  https://github.com/sudomesh/merakiflasher
>
>There are two remaining problems. One is that the sudomesh firmware
>boots
>up and then immediately issues a shutdown command. This should be
>fairly
>simple to solve but I haven't delved into it.
>
>The second issue is the watchdog. The official meraki firmware that
>comes
>with the routers correctly talks to the watchdog. Normal OpenWRT does
>not.
>The kernel driver for the AR2315 watchdog is supposed to create the
>/dev/watchdog device and the watchdog can be controlled from userland
>by
>the watchdog daemon (part of busybox utilities) by writing to
>/dev/watchdog
>or issuing ioctls. I've spent some time debugging the issue.
>
>The relevant files are:
>
>  drivers/watchdog/ar2315-wtd.c
>  arch/mips/ar231x/ar2315.c
>  arch/mips/ar231x/devices.c
>  arch/mips/include/asm/mach-ar231x.h
>
>The watchdog is being controlled by writing to memory-mapped registers
>defined in mach-ar231x.h and by compiling a kernel with lots of printk
>statements inserted into these files I have verified that the register
>write functions are being called at the correct times.
>
>I can imagine three things that may be going wrong:
>
> 1. The memory mapped register addresses in mach-ar231x.h are different
>for the AR2317 chipset (this seems unlikely based on the info I've been
>able to find).
>
>2. The registers are not being mapped into memory correctly. I don't
>know
>where to find the code that deals with the MMU and mapping registers.
>
>  3. There is some additional initialization required for the AR2317
>watchdog that isn't being handled correctly.
>
>4. The register read/write functions are not implemented correctly.
>These
>seem to be implemented in arch/mips/include/asm/mipsregs.h as assembly,
>but
>the actual ar231x_write_reg function is defined as:
>
>  static inline void
>  ar231x_write_reg(u32 reg, u32 val)
>  {
>      __raw_writel(val, (u32 *) KSEG1ADDR(reg));
>  }
>
>and __raw_write is defined in include/asm-generic/io.h as:
>
>static inline void __raw_writel(u32 b, volatile void __iomem *addr)
>{
>        *(volatile u32 __force *) addr = b;
>}
>
>So it's just an assignment. Maybe if the registers are not getting
>memory
>mapped (no idea how to check if they are) we should instead be using
>something like:
>
>#define __write_32bit_c0_register(register, sel, value)            \
>do {                                    \
>    if (sel == 0)                            \
>        __asm__ __volatile__(                    \
>            "mtc0\t%z0, " #register "\n\t"            \
>            : : "Jr" ((unsigned int)(value)));        \
>    else                                \
>        __asm__ __volatile__(                    \
>            ".set\tmips32\n\t"                \
>            "mtc0\t%z0, " #register ", " #sel "\n\t"    \
>            ".set\tmips0"                    \
>            : : "Jr" ((unsigned int)(value)));        \
>} while (0)
>
>from mipsregs.h
>
>I'm not sure. I'm speculating at this point.
>
>Adrian: I could really use some help on this. Do you have time to join
>us
>for a hack night soon? or maybe we can just meet on irc?

-- 
Charles Wyble charles at thefnf.org 
818 280 7059
CTO / co founder thefnf.org and guifi.us