# Application Processor to EC communication [TOC] ## Overview The Application Processor (sometimes called the host) communicates with the EC by issuing *host commands*, which are identified by a command ID and version number, and then reading a response. When a host command is issued through `ectool`, two or three software components are involved: * `ectool`, the user-space binary, * normally the `cros-ec` Kernel driver, and * the code on the EC itself. This can be thought of as two parts: * a chip-specific driver for the appropriate transport, and * the generic host command handling code (mostly in the [host command task]). We'll go into detail of each of these, as well as the traffic on the wire, in the following sections. ### `ectool` `ectool` contains wrapper functions for the host commands exposed by the EC, providing a CLI. They call one of the transport-specific `ec_command` implementations in the `util/comm-*.c` files to send and receive from the EC. #### Stress test The `ectool stress` command sends a large amount of host commands continuously to the EC. This verifies that the EC can respond to all commands received without timeouts or communication errors. It will log the time elapsed for every iteration of sending 10000 host commands with the time that it took to send them. Example output of stress test is: `Update: attempt 10000 round 1 | took 205 seconds` Killing the process will result in displaying the total runtime and failures that happened during it. ### EC kernel driver In most cases, `ectool` communicates via the [`cros-ec` Kernel driver], rather than directly from userspace. It sends raw commands to the Kernel driver, which sends them on to the EC, bypassing a lot of the other Kernel driver functionality. There are other CrOS EC-related Kernel drivers, which use host commands to act as adapters to existing Linux APIs. For example, sensors from the EC are mapped to the Linux [Industrial I/O] system. ### On the wire Now we come to the protocol itself. All transactions take this general form: * Host writes the request packet, consisting of: * a transport-specific header; * a `struct ec_host_request` containing the command ID, data length, and a checksum; and * zero or more bytes of parameters for the command, the format of which depends on the command. * Host reads the response to its request, consisting of: * a transport-specific header; * a `struct ec_host_response` containing the result code, data length, and a checksum; and * zero or more bytes of response from the command, again with a command-specific format. ### On the EC The host packet is received on the EC by some chip-specific code which checks its transport-specific header, then passes it on to the common host command code, starting at `host_packet_receive`. The common code validates the packet and then sends it on to the handler function (annotated with the `DECLARE_HOST_COMMAND` macro), which runs in the `HOSTCMD` task. The handler can set a response by modifying its arguments struct, which is sent back to the host via the chip-specific code. While this is happening, the EC needs to indicate to the host that it is busy processing and not yet ready to give a response. How it does this depends on the transport method used (see [Transport-specific details] below). ## Versions There are two different concepts of "version" involved in host commands: version of the overarching protocol, and versions of individual commands. ### Protocol versions There have been three protocol versions so far, and this document describes version 3. Version 1 was superseded by 2 before it shipped, so no devices use it anymore. Version 2 is generally deprecated, but you might still encounter it occasionally. Which version is in use can be determined using the `EC_CMD_GET_PROTOCOL_INFO` command. This was only introduced in version 3, however, so if errors, `EC_CMD_HELLO` should be sent in version 2. If the hello command succeeds, the EC speaks version 2. ### Command versions Individual commands also have versions, independent of the protocol version they're being called with. Different versions of a command may have different parameter or response formats. `EC_CMD_GET_CMD_VERSIONS` returns the versions of the given command supported by the EC. These version numbers start at 0. ## Tracing Command tracing can be done on the EC console (`hc debug` command) and/or in the kernel: ``` echo 1 > /sys/kernel/debug/tracing/events/cros_ec/enable cat /sys/kernel/debug/tracing/trace 369.416372: cros_ec_request_start: version: 0, command: EC_CMD_USB_PD_POWER_INFO 369.420528: cros_ec_request_done: version: 0, command: EC_CMD_USB_PD_POWER_INFO, ec result: EC_RES_SUCCESS, retval: 16 369.420529: cros_ec_request_start: version: 0, command: EC_CMD_USB_PD_DISCOVERY 369.421383: cros_ec_request_done: version: 0, command: EC_CMD_USB_PD_DISCOVERY, ec result: EC_RES_SUCCESS, retval: 5 ``` ## API synchronization EC, coreboot, user space (`platform2)` and kernel needs to access the same information contained in `ec_commands.h`. * Coreboot regularly update [ec_commands.h]. * User space/depthcharge use `ec_commands.h` installed by `chromeos-base/chromeos-ec-headers` package. * Kernel `ec_commands.h` - named `./include/linux/platform_data/cros_ec_commands.h` - is not a direct copy. New commands and options are added on a need-to-know basis alongside the kernel code that uses these commands. A copy of `ec_commands.h`, matching kernel standard is auto-generated at `/build/kernel/include/linux/mfd/cros_ec_commands.h` when `make build_cros_ec_commands` or `make buildall` is run. (It is far from perfect, since the kernel include file has not been updated with `clang-format` as `ec_commands.h` has been). Copy the necessary changes from that file to `include/linux/platform_data/cros_ec_commands.h` and push the changes upstream first following the [upstream instructions]. ## Transport-specific details Although the command and response formats are the same across all transports, some details of how they are transmitted differ, which may be of interest when implementing the EC side of the protocol on a new chip. ### I2C I2C is very flexible with its timing, so when the EC receives a packet from the host, it should stretch the clock, holding it low until it is ready for the host to read the response. If the host tries to read more bytes than were in the response, the EC should respond with an obvious filler byte (such as 0xEC). For example, if a command that normally returns 50 bytes errors, its response will only be 8 bytes (the size of the response struct). The host will probably try to read 50 bytes anyway, so the EC should send the 8 bytes of the struct followed by 42 copies of the filler byte. ### SPI The SPI bus is similar to I2C, but with two major exceptions. First, there's a minimum speed on the SPI bus. If slave devices don't respond quickly enough, the master will assume they're broken and give up. Second, every transaction is bidirectional. When bits are being clocked from master to slave on the MOSI line, the master will simultaneously read bits in the other direction on the MISO line. Hardware devices can usually handle this, and often some hardware-based flow control used to "stretch" the transaction by a bit or byte if the slave device needs a little extra time to respond to the master's demands. When exchanging messages with the EC on the SPI bus, the EC's host commands are communicated using our own software flow-control scheme, because most of the embedded controllers either aren't fast enough or don't have any support for hardware flow-control. It works like this: When the AP sends a byte to the EC, if the EC doesn't have a response queued up in advance, a default byte is returned. The EC preconfigures that default response byte to indicate its status (ready, busy, waiting for more input, etc.). Once the AP has sent a complete command message, it continues clocking bytes to the EC (which the EC ignores) and just looks at the response byte that comes back. Once the EC has parsed the AP's command and is ready to reply, it sends a "start of frame" byte, followed by the actual response. The AP continues to read and ignore bytes from the EC until it sees the start of frame byte, and then it knows that the EC's response is starting with the next byte. Once the response packet has been read, any additional reads should return `EC_SPI_PAST_END`. ### LPC or eSPI The EC should set `EC_LPC_STATUS_PROCESSING` in its command status register after receiving a host packet and before it has a response ready. [`cros-ec` Kernel driver]: https://chromium.googlesource.com/chromiumos/third_party/kernel/+/refs/heads/chromeos-4.19/drivers/mfd/cros_ec_dev.c [Industrial I/O]: https://www.kernel.org/doc/html/v4.14/driver-api/iio/index.html [host command task]: https://chromium.googlesource.com/chromiumos/platform/ec/+/HEAD/common/host_command.c [Transport-specific details]: #Transport_specific-details [ec_commands.h]: https://source.chromium.org/chromiumos/chromiumos/codesearch/+/main:src/third_party/coreboot/src/ec/google/chromeec/ec_commands.h [upstream instructions]: https://chromium.googlesource.com/chromiumos/docs/+/HEAD/kernel_development.md#Upstream-development.