1.. SPDX-License-Identifier: GPL-2.0 2 3====================== 4 USB4 and Thunderbolt 5====================== 6USB4 is the public specification based on Thunderbolt 3 protocol with 7some differences at the register level among other things. Connection 8manager is an entity running on the host router (host controller) 9responsible for enumerating routers and establishing tunnels. A 10connection manager can be implemented either in firmware or software. 11Typically PCs come with a firmware connection manager for Thunderbolt 3 12and early USB4 capable systems. Apple systems on the other hand use 13software connection manager and the later USB4 compliant devices follow 14the suit. 15 16The Linux Thunderbolt driver supports both and can detect at runtime which 17connection manager implementation is to be used. To be on the safe side the 18software connection manager in Linux also advertises security level 19``user`` which means PCIe tunneling is disabled by default. The 20documentation below applies to both implementations with the exception that 21the software connection manager only supports ``user`` security level and 22is expected to be accompanied with an IOMMU based DMA protection. 23 24Security levels and how to use them 25----------------------------------- 26The interface presented here is not meant for end users. Instead there 27should be a userspace tool that handles all the low-level details, keeps 28a database of the authorized devices and prompts users for new connections. 29 30More details about the sysfs interface for Thunderbolt devices can be 31found in ``Documentation/ABI/testing/sysfs-bus-thunderbolt``. 32 33Those users who just want to connect any device without any sort of 34manual work can add following line to 35``/etc/udev/rules.d/99-local.rules``:: 36 37 ACTION=="add", SUBSYSTEM=="thunderbolt", ATTR{authorized}=="0", ATTR{authorized}="1" 38 39This will authorize all devices automatically when they appear. However, 40keep in mind that this bypasses the security levels and makes the system 41vulnerable to DMA attacks. 42 43Starting with Intel Falcon Ridge Thunderbolt controller there are 4 44security levels available. Intel Titan Ridge added one more security level 45(usbonly). The reason for these is the fact that the connected devices can 46be DMA masters and thus read contents of the host memory without CPU and OS 47knowing about it. There are ways to prevent this by setting up an IOMMU but 48it is not always available for various reasons. 49 50Some USB4 systems have a BIOS setting to disable PCIe tunneling. This is 51treated as another security level (nopcie). 52 53The security levels are as follows: 54 55 none 56 All devices are automatically connected by the firmware. No user 57 approval is needed. In BIOS settings this is typically called 58 *Legacy mode*. 59 60 user 61 User is asked whether the device is allowed to be connected. 62 Based on the device identification information available through 63 ``/sys/bus/thunderbolt/devices``, the user then can make the decision. 64 In BIOS settings this is typically called *Unique ID*. 65 66 secure 67 User is asked whether the device is allowed to be connected. In 68 addition to UUID the device (if it supports secure connect) is sent 69 a challenge that should match the expected one based on a random key 70 written to the ``key`` sysfs attribute. In BIOS settings this is 71 typically called *One time saved key*. 72 73 dponly 74 The firmware automatically creates tunnels for Display Port and 75 USB. No PCIe tunneling is done. In BIOS settings this is 76 typically called *Display Port Only*. 77 78 usbonly 79 The firmware automatically creates tunnels for the USB controller and 80 Display Port in a dock. All PCIe links downstream of the dock are 81 removed. 82 83 nopcie 84 PCIe tunneling is disabled/forbidden from the BIOS. Available in some 85 USB4 systems. 86 87The current security level can be read from 88``/sys/bus/thunderbolt/devices/domainX/security`` where ``domainX`` is 89the Thunderbolt domain the host controller manages. There is typically 90one domain per Thunderbolt host controller. 91 92If the security level reads as ``user`` or ``secure`` the connected 93device must be authorized by the user before PCIe tunnels are created 94(e.g the PCIe device appears). 95 96Each Thunderbolt device plugged in will appear in sysfs under 97``/sys/bus/thunderbolt/devices``. The device directory carries 98information that can be used to identify the particular device, 99including its name and UUID. 100 101Authorizing devices when security level is ``user`` or ``secure`` 102----------------------------------------------------------------- 103When a device is plugged in it will appear in sysfs as follows:: 104 105 /sys/bus/thunderbolt/devices/0-1/authorized - 0 106 /sys/bus/thunderbolt/devices/0-1/device - 0x8004 107 /sys/bus/thunderbolt/devices/0-1/device_name - Thunderbolt to FireWire Adapter 108 /sys/bus/thunderbolt/devices/0-1/vendor - 0x1 109 /sys/bus/thunderbolt/devices/0-1/vendor_name - Apple, Inc. 110 /sys/bus/thunderbolt/devices/0-1/unique_id - e0376f00-0300-0100-ffff-ffffffffffff 111 112The ``authorized`` attribute reads 0 which means no PCIe tunnels are 113created yet. The user can authorize the device by simply entering:: 114 115 # echo 1 > /sys/bus/thunderbolt/devices/0-1/authorized 116 117This will create the PCIe tunnels and the device is now connected. 118 119If the device supports secure connect, and the domain security level is 120set to ``secure``, it has an additional attribute ``key`` which can hold 121a random 32-byte value used for authorization and challenging the device in 122future connects:: 123 124 /sys/bus/thunderbolt/devices/0-3/authorized - 0 125 /sys/bus/thunderbolt/devices/0-3/device - 0x305 126 /sys/bus/thunderbolt/devices/0-3/device_name - AKiTiO Thunder3 PCIe Box 127 /sys/bus/thunderbolt/devices/0-3/key - 128 /sys/bus/thunderbolt/devices/0-3/vendor - 0x41 129 /sys/bus/thunderbolt/devices/0-3/vendor_name - inXtron 130 /sys/bus/thunderbolt/devices/0-3/unique_id - dc010000-0000-8508-a22d-32ca6421cb16 131 132Notice the key is empty by default. 133 134If the user does not want to use secure connect they can just ``echo 1`` 135to the ``authorized`` attribute and the PCIe tunnels will be created in 136the same way as in the ``user`` security level. 137 138If the user wants to use secure connect, the first time the device is 139plugged a key needs to be created and sent to the device:: 140 141 # key=$(openssl rand -hex 32) 142 # echo $key > /sys/bus/thunderbolt/devices/0-3/key 143 # echo 1 > /sys/bus/thunderbolt/devices/0-3/authorized 144 145Now the device is connected (PCIe tunnels are created) and in addition 146the key is stored on the device NVM. 147 148Next time the device is plugged in the user can verify (challenge) the 149device using the same key:: 150 151 # echo $key > /sys/bus/thunderbolt/devices/0-3/key 152 # echo 2 > /sys/bus/thunderbolt/devices/0-3/authorized 153 154If the challenge the device returns back matches the one we expect based 155on the key, the device is connected and the PCIe tunnels are created. 156However, if the challenge fails no tunnels are created and error is 157returned to the user. 158 159If the user still wants to connect the device they can either approve 160the device without a key or write a new key and write 1 to the 161``authorized`` file to get the new key stored on the device NVM. 162 163De-authorizing devices 164---------------------- 165It is possible to de-authorize devices by writing ``0`` to their 166``authorized`` attribute. This requires support from the connection 167manager implementation and can be checked by reading domain 168``deauthorization`` attribute. If it reads ``1`` then the feature is 169supported. 170 171When a device is de-authorized the PCIe tunnel from the parent device 172PCIe downstream (or root) port to the device PCIe upstream port is torn 173down. This is essentially the same thing as PCIe hot-remove and the PCIe 174toplogy in question will not be accessible anymore until the device is 175authorized again. If there is storage such as NVMe or similar involved, 176there is a risk for data loss if the filesystem on that storage is not 177properly shut down. You have been warned! 178 179DMA protection utilizing IOMMU 180------------------------------ 181Recent systems from 2018 and forward with Thunderbolt ports may natively 182support IOMMU. This means that Thunderbolt security is handled by an IOMMU 183so connected devices cannot access memory regions outside of what is 184allocated for them by drivers. When Linux is running on such system it 185automatically enables IOMMU if not enabled by the user already. These 186systems can be identified by reading ``1`` from 187``/sys/bus/thunderbolt/devices/domainX/iommu_dma_protection`` attribute. 188 189The driver does not do anything special in this case but because DMA 190protection is handled by the IOMMU, security levels (if set) are 191redundant. For this reason some systems ship with security level set to 192``none``. Other systems have security level set to ``user`` in order to 193support downgrade to older OS, so users who want to automatically 194authorize devices when IOMMU DMA protection is enabled can use the 195following ``udev`` rule:: 196 197 ACTION=="add", SUBSYSTEM=="thunderbolt", ATTRS{iommu_dma_protection}=="1", ATTR{authorized}=="0", ATTR{authorized}="1" 198 199Upgrading NVM on Thunderbolt device, host or retimer 200---------------------------------------------------- 201Since most of the functionality is handled in firmware running on a 202host controller or a device, it is important that the firmware can be 203upgraded to the latest where possible bugs in it have been fixed. 204Typically OEMs provide this firmware from their support site. 205 206There is also a central site which has links where to download firmware 207for some machines: 208 209 `Thunderbolt Updates <https://thunderbolttechnology.net/updates>`_ 210 211Before you upgrade firmware on a device, host or retimer, please make 212sure it is a suitable upgrade. Failing to do that may render the device 213in a state where it cannot be used properly anymore without special 214tools! 215 216Host NVM upgrade on Apple Macs is not supported. 217 218Once the NVM image has been downloaded, you need to plug in a 219Thunderbolt device so that the host controller appears. It does not 220matter which device is connected (unless you are upgrading NVM on a 221device - then you need to connect that particular device). 222 223Note an OEM-specific method to power the controller up ("force power") may 224be available for your system in which case there is no need to plug in a 225Thunderbolt device. 226 227After that we can write the firmware to the non-active parts of the NVM 228of the host or device. As an example here is how Intel NUC6i7KYK (Skull 229Canyon) Thunderbolt controller NVM is upgraded:: 230 231 # dd if=KYK_TBT_FW_0018.bin of=/sys/bus/thunderbolt/devices/0-0/nvm_non_active0/nvmem 232 233Once the operation completes we can trigger NVM authentication and 234upgrade process as follows:: 235 236 # echo 1 > /sys/bus/thunderbolt/devices/0-0/nvm_authenticate 237 238If no errors are returned, the host controller shortly disappears. Once 239it comes back the driver notices it and initiates a full power cycle. 240After a while the host controller appears again and this time it should 241be fully functional. 242 243We can verify that the new NVM firmware is active by running the following 244commands:: 245 246 # cat /sys/bus/thunderbolt/devices/0-0/nvm_authenticate 247 0x0 248 # cat /sys/bus/thunderbolt/devices/0-0/nvm_version 249 18.0 250 251If ``nvm_authenticate`` contains anything other than 0x0 it is the error 252code from the last authentication cycle, which means the authentication 253of the NVM image failed. 254 255Note names of the NVMem devices ``nvm_activeN`` and ``nvm_non_activeN`` 256depend on the order they are registered in the NVMem subsystem. N in 257the name is the identifier added by the NVMem subsystem. 258 259Upgrading on-board retimer NVM when there is no cable connected 260--------------------------------------------------------------- 261If the platform supports, it may be possible to upgrade the retimer NVM 262firmware even when there is nothing connected to the USB4 263ports. When this is the case the ``usb4_portX`` devices have two special 264attributes: ``offline`` and ``rescan``. The way to upgrade the firmware 265is to first put the USB4 port into offline mode:: 266 267 # echo 1 > /sys/bus/thunderbolt/devices/0-0/usb4_port1/offline 268 269This step makes sure the port does not respond to any hotplug events, 270and also ensures the retimers are powered on. The next step is to scan 271for the retimers:: 272 273 # echo 1 > /sys/bus/thunderbolt/devices/0-0/usb4_port1/rescan 274 275This enumerates and adds the on-board retimers. Now retimer NVM can be 276upgraded in the same way than with cable connected (see previous 277section). However, the retimer is not disconnected as we are offline 278mode) so after writing ``1`` to ``nvm_authenticate`` one should wait for 2795 or more seconds before running rescan again:: 280 281 # echo 1 > /sys/bus/thunderbolt/devices/0-0/usb4_port1/rescan 282 283This point if everything went fine, the port can be put back to 284functional state again:: 285 286 # echo 0 > /sys/bus/thunderbolt/devices/0-0/usb4_port1/offline 287 288Upgrading NVM when host controller is in safe mode 289-------------------------------------------------- 290If the existing NVM is not properly authenticated (or is missing) the 291host controller goes into safe mode which means that the only available 292functionality is flashing a new NVM image. When in this mode, reading 293``nvm_version`` fails with ``ENODATA`` and the device identification 294information is missing. 295 296To recover from this mode, one needs to flash a valid NVM image to the 297host controller in the same way it is done in the previous chapter. 298 299Networking over Thunderbolt cable 300--------------------------------- 301Thunderbolt technology allows software communication between two hosts 302connected by a Thunderbolt cable. 303 304It is possible to tunnel any kind of traffic over a Thunderbolt link but 305currently we only support Apple ThunderboltIP protocol. 306 307If the other host is running Windows or macOS, the only thing you need to 308do is to connect a Thunderbolt cable between the two hosts; the 309``thunderbolt-net`` driver is loaded automatically. If the other host is 310also Linux you should load ``thunderbolt-net`` manually on one host (it 311does not matter which one):: 312 313 # modprobe thunderbolt-net 314 315This triggers module load on the other host automatically. If the driver 316is built-in to the kernel image, there is no need to do anything. 317 318The driver will create one virtual ethernet interface per Thunderbolt 319port which are named like ``thunderbolt0`` and so on. From this point 320you can either use standard userspace tools like ``ifconfig`` to 321configure the interface or let your GUI handle it automatically. 322 323Forcing power 324------------- 325Many OEMs include a method that can be used to force the power of a 326Thunderbolt controller to an "On" state even if nothing is connected. 327If supported by your machine this will be exposed by the WMI bus with 328a sysfs attribute called "force_power". 329 330For example the intel-wmi-thunderbolt driver exposes this attribute in: 331 /sys/bus/wmi/devices/86CCFD48-205E-4A77-9C48-2021CBEDE341/force_power 332 333 To force the power to on, write 1 to this attribute file. 334 To disable force power, write 0 to this attribute file. 335 336Note: it's currently not possible to query the force power state of a platform. 337