Always-on Processor magic: How Find My works while iPhone is powered off

September 30, 2021

iOS 15.0 introduces a new feature: an iPhone can be located with Find My even while the iPhone is turned "off". How does it work? Is it a security concern?

I saw this feature rather early on one of my iPhones with an iOS 15 beta. Here's a screenshot I took in July. The user interface changed a little bit since then.

It took a bit longer until the public realized this feature exists. One needs to update to iOS 15.0, use an iPhone that has location services enabled, a logged in user account, participates in the Find My network, etc. And the weirdest thing nobody does these days: One has to turn the iPhone off. But once Twitter found out, this took off. And so did the rumors how this was implemented.

Apple's Always-on Processor (AOP)

There's only little public documentation about the AOP. All chips and various embedded devices Apple manufactures run a real-time operating system, called RTKitOS. The AOP on the iPhone is no exception. However, the AOP has a special role. It connects to almost every other chip in the iPhone. For some chips, it only does basic tasks like power management, and for other chips, it acts as a transparent proxy that wakes up iOS when needed.

This way, a processor that is always on actually saves energy. iOS can go to sleep while the AOP waits for hardware events. A simple example is the motion sensor. Without touching any button on the iPhone, the display wakes up.

A quick Internet search reveals that even Siri is implemented in the AOP. If you're not too much into technical details, you can skip the remainder of this section and just need to know that the AOP also connects to wireless chips and their power management interfaces :)

Most iOS kernel drivers follow a simple structure. If a chip is running RTKitOS, this is visible in the ioreg output. For example Rose, the U1 Ultra-wideband chip, shows up in this list.

While this is in the iOS kernel, the AOP implements a copy of these drivers. For example, U1 ranging has a duplicate implementation that works without the nearbyd and can run standalone within the AOP.

Even if a chip is not based on RTKitOS, the AOP can connect to it. The Bluetooth chip in the iPhone 11, 12 and 13 is by Broadcom and based on the ThreadX operating system. Nonetheless, the iPhone AOP has some control over Bluetooth. While not following the same RTKitOS-based scheme, it can power the Bluetooth chip using the spmi-bluetooth driver.

Note that this is very different from the HomePod mini and Watch. They both use Apple's own Bluetooth chip, codenamed Marconi. The AOP can control Marconi using the marconi-bluetooth and aop-marconi-bt-control drivers.

Running a Bluetooth application while the iPhone is "off"

All the iPhone needs to have Find My enabled while it is "off" is some basic power control via the AOP as well as a Bluetooth chip that can send Bluetooth LE advertisements. Moreover, these advertisements require secret key material bound to the Apple ID on the iPhone.

Broadcom chips are very similar to Cypress chips. The Cypress SDKs support standalone IoT applications, which only require the Bluetooth chip and no additional host. These applications run in the ThreadX thread called mpaf. Let's check if there's an mpaf thread somewhere in the Bluetooth firmware. For this, download the iOS 15.0 IPSW, mount the largest .dmg file, and browse to /usr/share/firmware/bluetooth. The .bin files are patches loaded via the Bluetooth PCIe driver. Since the iPhone 11, these contain a few debug strings. Running grep mpaf * reveals the following device compatibility list.

iPhone 11 series, BCM4378B1 (Hei, Moana, Tala)
iPhone 12 series, BCM4387C2 (Almond, Cashew, Hazelnut, Pistachio)
iPhone 13 series, BCM4387C2 (Acacia, Camellia, Lilac, Mimosa)
iPad Air 2020 series, BCM4387C2 (Pomegranate)
Some other iPad series, BCM4387C2 (Baobab, Boab, Rambutan)

Interestingly, the iPhone SE 2020, which also has the BCM4378B1 chip and is codenamed Fiti, does not have an mpaf patch. I updated an iPad Air 2020 to iOS 15.0, and despite the patches for Pomegranate, the Find My dialogue is not shown. There's also no location update for the iPad in another iPhone logged into the same account. So there might be a difference in currently supported devices and chips that could in theory support this feature. Apple could add further devices in the future.

The mpaf patch implements a lpm app, which probably stands for low-power mode, that implements a Bluetooth LE gatt service. This can all be found in the strings, and plonk actually saw this before I saw it.

tier2/Olympic/PCIE/Pistachio_CLPC_OS/USI/bld/A_4387C2_ROM/tier2/patch/bcs/mpaf_layer_patch.o.patch2.c

tier2/Olympic/PCIE/Pistachio_CLPC_OS/USI/bld/A_4387C2_ROM/tier2/patch/mpaf/apps/lpm/lpm_app.o.patch2.c

tier2/Olympic/PCIE/Pistachio_CLPC_OS/USI/bld/A_4387C2_ROM/tier2/patch/mpaf/apps/lpm/lpm_app_gatt.o.patch2.c

tier2/Olympic/PCIE/Pistachio_CLPC_OS/USI/bld/A_4387C2_ROM/tier2/patch/mpaf/apps/lpm/lpm_app_fsm.o.patch2.c

These patches were introduced in iOS 15 and not present before. The naming matches the expected functionality. To keep this blog post short and answer your questions early, this blog post doesn't contain instructions on how to analyze these patches, dump the chip's ROM etc. Strings are enough to confirm that this feature is implemented in the Bluetooth chip.

Is the secret key material related to the U1 chip?

On the AirTag, this key material is stored via the U1 chip. Apple calls the nRF on the AirTag "Durian" for a reason. They don't like it, they don't trust it, but it's cheap and low-power. Most time, U1 is on sleep, and from time to time they wake it up.

This is not the design on the iPhone. On the iPhone, keys are stored in the Secure Enclave (SE). Some early U1 build even had debug strings for U1 and the SE exchanging key material, but these are gone in recent builds. So, the little bird is right when it comes to the AirTag, but not the iPhone.

Secret key material transfer

After installing a Bluetooth debug profile to an iPhone 12 on iOS 15.1b2, the idevicesyslog output looks as follows just before shutdown:

Sep 30 22:02:58 BlueTool[126] <Notice>: Completed handling of dictionary-xpc event

Sep 30 22:02:58 bluetoothd[89] <Notice>: BlueTool finished running "hci reset" command - output was "0x0e 0x04 0x01 0x03 0x0c 0x00"

Sep 30 22:02:58 BlueTool[126] <Notice>: Completed handling of dictionary-xpc event

Sep 30 22:02:58 bluetoothd[89] <Notice>: BlueTool finished running "hci cmd 0xFE62 0x06 ..." command - output was "<decode: missing data>"

Sep 30 22:02:59 bluetoothd[89] <Notice>: BlueTool finished running "hci cmd 0xFE62 0x06 ..." command - output was "<decode: missing data>"