As the original USB-C Power Delivery (PD) solution for the ChromeOS Embedded Controller has aged, it has grown to the point where it is difficult to add new features and address bugs. A new PD stack (generally referred to as TCPMv2) has been introduced to the codebase for use moving forward. It implements a layered, state-based design which tracks more closely with the USB Type-C and USB PD specifications.
[TOC]
Boards may enable TCPMv2 by adding the following defines:
CONFIG_USB_PD_TCPMV2Configures the board to use the new stack.CONFIG_USB_DRP_ACC_TRYSRC: Configures the type of state machine to run (in this case, a DRP which performs Try.SRC behavior). Currently available are DRP and charge-through Vconn-powered device optionsCONFIG_USB_PD_DECODE_SOP: Sets up messaging for SOP’ and SOP’’, which is strongly encouraged in the TCPMv2 code moving forwardCONFIG_USB_PID 0x1234: Sets the USB Product Identifier. This will be shared for all boards within one reference design, and new PIDs may be requested by sending an email to the ChromeOS FW Team.CONFIG_USB_PD_REV30: The TCPMv2 stack defaults to PD2.0 operation but defining this macro enables PD3.0 functionality.
Other configurations to specify behaviors within the task still apply (ex.
CONFIG_USB_PD_ALT_MODE_DFP and CONFIG_USB_PD_TCPC_LOW_POWER).
The basis of the TCPMv2 state machines is a generic state machine framework. This framework can be found in common/usbc/usb_sm.c. For each state, there may be defined:
- Entry: Called upon entering a state
- Run: Called while steady in a state
- Exit: Called upon leaving a state
- Parent: Superstate. Enters, exits, and runs alongside the child state. Only enters and exits when transitioning between states which do not share the parent.
All fields are optional and may be set to NULL. A new state is transitioned to with a call into set_state(), which exits the old state and parents and enters the new parents and state. States may be changed with set_state() in any entry or run function, but any call from an exit function is ignored since exit is only called when a change is already in progress. While in a state, run_state() executes the run function for the current state and parents. If set_state() is called from either an entry function or a run function, the remaining run or entry functions are stopped.
Below is a graphical example of what some states may look like. States 1 and 2 share Parent State 1, while State 3 has Parent State of 2.
Translated into code, this would be something like below (note it is not necessary that the states be a part of an array, but the TCPMv2 code generally organizes the states in this way):
static const struct usb_state test_states[] = {
[PARENT_1] = {
.entry = parent_1_entry,
.run = parent_1_run,
.exit = parent_1_exit,
},
[PARENT_2] = {
.entry = parent_2_entry,
.run = parent_2_run,
.exit = parent_2_exit,
},
[STATE_1] = {
.entry = state_1_entry,
.run = state_1_run,
.exit = state_1_exit,
.parent = &test_states[PARENT_1],
},
[STATE_2] = {
.entry = state_2_entry,
.run = state_2_run,
.exit = state_2_exit,
.parent = &test_states[PARENT_1],
},
[STATE_3] = {
.entry = state_3_entry,
.run = state_3_run,
.exit = state_3_exit,
.parent = &test_states[PARENT_2],
},
};
For this example, each test state is written simply to produce a print of its function name. The two exceptions are:
- parent_1_run() calls set_state() into STATE_2 when called a second time
- state_2_entry() calls set_state() into STATE_3
Graphically, this is represented below:
And the following code output is produced:
Calling run_state()
state_1_run
parent_1_run
Calling run_state()
state_1_run
state_1_run calling set_state() to state 2
state_1_exit
state_2_entry
state_2_entry calling set_state() to state 3
state_2_exit
parent_1_exit
parent_2_entry
state_3_entry
Calling run_state()
state_3_run
parent_2_run
The TCPMv2 PD task is built upon state machines using the above framework and is located in common/usbc/usbc_task.c. It is separated into three layers which track with the USB Type-C and USB PD specification states and run in a loop with 5 ms periods between executions. A graphical representation of these layers is below.
The task is designed such that the Type-C (TC) layer could run independently for any application which doesn’t wish to enable PD messaging. Boards define their appropriate Policy Engine (PE) and TC state machines through their selection of a CONFIG_USB_* define, with current options supporting both Dual-Role Ports (DRPs) and Charge-Through Vconn-Powered Device (CTVPD). All boards use the same Protocol Layer (PRL) code.
The three state machines mentioned above interact with each other and the EC drivers in order to orchestrate all Type-C connection behavior. Graphically, they are represented below.
Layers communicate with each other using atomic operations on flags and shared buffers. Functions calling into each layer are clearly named to indicate the layer they are acting on, and anything calling into the PD task should be doing so through pd_* named functions.
Some specific examples of how this communication works between layers is below. If a port partner sends in a Vconn_swap request, then:
- PRL will communicate that a message was received to the PE layer through pe_message_received(), which sets PE_FLAGS_MSG_RECEIVED and indicates the receive buffer has a message
- PE asks with the TC whether the board is sourcing Vconn with tc_is_vconn_src() which checks TC_FLAGS_VCONN_ON
- PE tells the PRL to send an ACCEPT message to the port partner through prl_send_ctrl_msg() which fills in shared message information and sets PRL_FLAGS_MSG_XMIT
- PRL lets the PE know that the message transmit was successful through pe_message_sent() which sets PE_FLAGS_TX_COMPLETE
- TC tells the PE layer that the Vconn swap completed with pe_vconn_swap_complete() which sets PE_FLAGS_VCONN_SWAP_COMPLETE
Defined in the USB Type-C specification, this layer is responsible for basic connection. It takes care of setting the CC lines, detecting and debouncing the partner CC lines, and performs most of the interactions needed with the PPC and USB mux. Once the TC layer has gotten the connection to the point of being Attached.SNK or Attached.SRC, it will enable the PRL and PE layers accordingly.
A part of the USB PD specification, the protocol layer is responsible for the actual sending and receiving of PD messages with the TCPCs. The layer is actually composed of 4 separate state machines running one after the other. These state machines are:
- Chunked receive (RCH): passes messages up to the PE and requests chunks when chunking
- Chunked transmit (TCH): receives messages from the PE and waits for chunk requests when chunking
- Protocol transmit (PRL_TX): passes messages to the TCPCs and handles PD 3.0 collision avoidance
- Protocol hard reset (PRL_HR): responds to or transmits hard resets, resets PRL layer variables, notifies PE of hard reset receipt or sent
The PE layer states are defined as a part of the USB PD specification. State names are meant to track very closely with the specification so they can be easily searchable and understood. The PE’s primary responsibility is to send and process PD messages in order to implement the port’s policy.
The USB PD specification defines some responsibilities of the DPM, e.g. selecting Source Capabilities to offer and alternate modes to enter. Broadly, the DPM is supposed to make discretionary decisions about the behavior of the PE, i.e. policy decisions. However, the spec is relatively silent on the structure and scope of the DPM.
In TCPMv2, the DPM consists of a central code module in usb_pd_dpm.c and
several code modules for supported alternate or USB modes. The mode entry and
exit logic in the DPM takes the form of a state machine, and each mode has its
own state machine to control the specific mode entry and exit steps. The DPM
also contains logic to choose which Source PDO to offer in Source
Capabilities.
The DPM state machines do not currently use the USB state machine framework. Additionally, there are policy decisions encoded in the lower layers and in board code that ought to be made by the DPM. An effort to centralize policy decisions and improve DPM code organization is ongoing.
TCPMv2 supports several modes of operation besides USB 3.2. To determine which
modes the port partner and cable support, the TCPM sends SOP and SOP'
Vendor-Defined Messages, specifically Discover Identity, Discover SVIDs,
and Discover Modes. The responses to these messages list supported modes. The
AP may query this information via EC_CMD_TYPEC_DISCOVERY.
Based on the discovery responses, the DPM decides which mode to enter (if any)
and drives mode entry. If CONFIG_USB_PD_REQUIRE_AP_MODE_ENTRY is enabled, the
AP directs the DPM to enter and exit modes via EC_CMD_TYPEC_CONTROL.
DisplayPort alternate mode repurposes the SuperSpeed lanes as video data lanes, all flowing from the host to the monitor. DP mode can use 2 lanes, allowing simultaneous USB traffic on the cable, or it can use all 4 lanes, supporting greater display bandwidth at the expense of USB.
The sequence to enter DP alternate mode (v. 1.4) is
- The TCPM sends an SOP VDM Enter Mode REQ with the DP VID.
- The TCPM sends an SOP DP Status VDM REQ.
- The TCPM sends an SOP DP Configure VDM REQ.
- Display sends an SOP Attention VDM REQ.
To exit DP alternate mode, the TCPM sends an SOP VDM Exit Mode REQ.
dp_alt_mode.c and usb_pd_alt_mode_dfp.c contain the majority of the DP entry
and exit logic.
TBT3 alternate mode repurposes all 4 SuperSpeed lanes for Thunderbolt 3, which is itself a bidirectional tunneling protocol supporting various types of hardware, most commonly docks.
The sequence to enter TBT3 alternate mode depends on the cable used:
- If the cable supports TBT3, the TCPM sends an SOP' VDM Enter Mode REQ with the Intel VID.
- If the cable is also an active cable with SOP'' support, the TCPM sends an SOP'' VDM Enter Mode REQ. (The TBT3 mode VDO may identify the cable as active for this purpose when the ID Header VDO does not.)
- The TCPM sends an SOP VDM Enter Mode REQ.
To exit TBT3 alternate mode, the TCPM sends VDM Exit Mode REQs to SOP, SOP'', and SOP', as appropriate.
tbt_alt_mode.c and usb_pd_alt_mode_dfp.c contain the TBT entry and exit
logic.
USB4 is not an alternate mode, but it does repurpose the SuperSpeed lines (using different signaling and routing than USB 3.2), so the TCPM must explicitly enter USB4.
The sequence to enter USB4 is conceptually similar to that for TBT3:
- If the cable supports USB4, the TCPM sends an SOP' Enter USB message.
- If the cable is also an active cable, the TCPM sends an SOP'' Enter USB.
- The TCPM sends an SOP Enter USB. Complicating matters, if the cable does not support USB4 but does support TBT3, the TCPM may enter TBT3 for the cable as above, then enter USB4 for the port partner.
To exit USB4 (returning to a USB 3.2 mux configuration), the TCPM performs a Data Reset sequence.
usb_mode.c and usb_pd_alt_mode_dfp.c contain the USB4 entry and exit logic.
- Always call return after set_state(). Once the state has been changed, nothing further should be done in the current state.
- Never call set_state() from an exit function. The call will be ignored as there is already a state transition taking place.
- Never call set_state() from outside the PD task. The task may be waiting in any number of locations and the context should not change around it while it does so.
- Always use flags to communicate between layers, and to communicate with the PD task from other tasks. Flags should be accessed through atomic operations.
- Always use pd_* functions to access the PD task from other tasks.
- Always write unit tests as code is added, to verify new code works and protect against regressions as development continues.