Now that I am releasing a native PCI driver, I am no longer maintaining this project.
This repo currently contains a small tool for flipping an M.2 XMM7360 modem,
like the Fibocom L850-GL, from PCIe into USB mode.
This allows you to talk to it with mbimcli, ModemManager etc.
This only works on some machines: specifically, those that have USB lines routed to the M.2 slot, and enabled. Quite a lot of L850-GL laptops do not have this (eg. Thinkpad T495), so this may not work.
You can build and install it like this:
git clone https://github.com/mkottman/acpi_call.git
cd acpi_call
make && sudo make installThen, you can simply run this script:
sudo ./xmm2usbAfter 5-10 seconds, your new USB device should appear.
My modem came up initially as a 3x ACM + 3x NCM USB device. To switch to MBIM mode I ran:
sudo screen /dev/ttyACM0and issued
AT+GTUSBMODE?
AT+GTUSBMODE=7
AT+CFUN=15This change is permanent; you do not need to do this again after the first time.
Take note of the USB mode reported after the first command in case you want to put it back later...!
(There is an old L8 family AT command doc floating around saying that mode 2 is MBIM. It does not apply to this modem.)
According to the modem docs, the chipset tries talking via PCIe first, then goes to USB if it doesn't succeed. Luckily for us, we have two moving parts here we can use: we can talk to the upstream PCI Express port to disable the link, and then we can use ACPI to hit the modem's reset line so that it starts looking anew. That's all this script does.
The modem seems to happily stay in USB mode across a suspend/resume, but the PCIe link needs to be disabled on each resume. It'd be cute to have a PCI stub driver to do this, I suppose.
My modem, in my Thinkpad T490, initially seems to be stuck in flight mode (AT+CFUN? returns +CFUN: 4,0 and can't be changed).
This turns out to be a mechanism the authors call "FCC Lock".
This prevents the radio from being enabled until it is unlocked;
this is the purpose of the ModemAuthenticator.exe which comes with the driver in Windows.
This has the effect of tying the modem to particular machines:
the unlock key is stored in the system's SMBIOS.
This looks very much like it has been done for regulatory purposes: these days you have to test the software, radio, and antennas together in the final product to meet regulatory requirements. You need a lot of equipment, time, and people who know what they are doing, so this is expensive. Also, porting a PCI driver for this thing would suck, because it's actually quite complicated to talk to. Thus Lenovo only tested the Windows configuration, and possibly implemented this locking mechanism to maintain compliance.
It's also possible, again as a regulatory matter, that these systems actually use dynamic power reduction. This is a feature where, when human body parts are detected near the antennas, the output power is reduced. This allows quite a powerful radio to be used without exceeding the regulatory limits for irradiating people. The modem certainly seems to support such features, though I am yet to see evidence that they are in use here. Nonetheless, this is good reason for caution at this point: bypassing this mechanism may cause body SAR to exceed regulatory limits.
Regulation is a good thing here - this is why we so rarely have problems with interference, and tries to make sure we don't get cooked. Please consider this carefully before you decide to unlock your modem.
I carefully reverse engineered the unlock challenge/response sequence before realising that you can permanently bypass it with just a couple of AT commands.
This should not brick your device, but I make no guarantees. Here be dragons!
Note: as of newer firmware version 18500.5001.00.02.24.09, this appears to cause a reboot loop of the modem. See this issue. I do not recommend performing this at present unless you understand the implications - wait a couple of months for a PCI driver.
You can unlock the modem by issuing
at@nvm:fix_cat_fcclock.fcclock_mode=0(The default mode, at least on my modem, is 2.)
This will work until the modem is next power cycled or reset. To make the change permanent, issue:
at@store_nvm(fix_cat_fcclock)The modem will then power up with radio enabled in future.
I did start doing some reverse engineering and writing a PCI driver, just for fun.
I aim to release something early in 2020 which should be usable for most.
Tested and working on:
Reports of not working on: