1.. SPDX-License-Identifier: GPL-2.0+
2
3.. |u8| replace:: :c:type:`u8 <u8>`
4.. |u16| replace:: :c:type:`u16 <u16>`
5.. |TYPE| replace:: ``TYPE``
6.. |LEN| replace:: ``LEN``
7.. |SEQ| replace:: ``SEQ``
8.. |SYN| replace:: ``SYN``
9.. |NAK| replace:: ``NAK``
10.. |ACK| replace:: ``ACK``
11.. |DATA| replace:: ``DATA``
12.. |DATA_SEQ| replace:: ``DATA_SEQ``
13.. |DATA_NSQ| replace:: ``DATA_NSQ``
14.. |TC| replace:: ``TC``
15.. |TID| replace:: ``TID``
16.. |IID| replace:: ``IID``
17.. |RQID| replace:: ``RQID``
18.. |CID| replace:: ``CID``
19
20===========================
21Surface Serial Hub Protocol
22===========================
23
24The Surface Serial Hub (SSH) is the central communication interface for the
25embedded Surface Aggregator Module controller (SAM or EC), found on newer
26Surface generations. We will refer to this protocol and interface as
27SAM-over-SSH, as opposed to SAM-over-HID for the older generations.
28
29On Surface devices with SAM-over-SSH, SAM is connected to the host via UART
30and defined in ACPI as device with ID ``MSHW0084``. On these devices,
31significant functionality is provided via SAM, including access to battery
32and power information and events, thermal read-outs and events, and many
33more. For Surface Laptops, keyboard input is handled via HID directed
34through SAM, on the Surface Laptop 3 and Surface Book 3 this also includes
35touchpad input.
36
37Note that the standard disclaimer for this subsystem also applies to this
38document: All of this has been reverse-engineered and may thus be erroneous
39and/or incomplete.
40
41All CRCs used in the following are two-byte ``crc_ccitt_false(0xffff, ...)``.
42All multi-byte values are little-endian, there is no implicit padding between
43values.
44
45
46SSH Packet Protocol: Definitions
47================================
48
49The fundamental communication unit of the SSH protocol is a frame
50(:c:type:`struct ssh_frame <ssh_frame>`). A frame consists of the following
51fields, packed together and in order:
52
53.. flat-table:: SSH Frame
54   :widths: 1 1 4
55   :header-rows: 1
56
57   * - Field
58     - Type
59     - Description
60
61   * - |TYPE|
62     - |u8|
63     - Type identifier of the frame.
64
65   * - |LEN|
66     - |u16|
67     - Length of the payload associated with the frame.
68
69   * - |SEQ|
70     - |u8|
71     - Sequence ID (see explanation below).
72
73Each frame structure is followed by a CRC over this structure. The CRC over
74the frame structure (|TYPE|, |LEN|, and |SEQ| fields) is placed directly
75after the frame structure and before the payload. The payload is followed by
76its own CRC (over all payload bytes). If the payload is not present (i.e.
77the frame has ``LEN=0``), the CRC of the payload is still present and will
78evaluate to ``0xffff``. The |LEN| field does not include any of the CRCs, it
79equals the number of bytes inbetween the CRC of the frame and the CRC of the
80payload.
81
82Additionally, the following fixed two-byte sequences are used:
83
84.. flat-table:: SSH Byte Sequences
85   :widths: 1 1 4
86   :header-rows: 1
87
88   * - Name
89     - Value
90     - Description
91
92   * - |SYN|
93     - ``[0xAA, 0x55]``
94     - Synchronization bytes.
95
96A message consists of |SYN|, followed by the frame (|TYPE|, |LEN|, |SEQ| and
97CRC) and, if specified in the frame (i.e. ``LEN > 0``), payload bytes,
98followed finally, regardless if the payload is present, the payload CRC. The
99messages corresponding to an exchange are, in part, identified by having the
100same sequence ID (|SEQ|), stored inside the frame (more on this in the next
101section). The sequence ID is a wrapping counter.
102
103A frame can have the following types
104(:c:type:`enum ssh_frame_type <ssh_frame_type>`):
105
106.. flat-table:: SSH Frame Types
107   :widths: 1 1 4
108   :header-rows: 1
109
110   * - Name
111     - Value
112     - Short Description
113
114   * - |NAK|
115     - ``0x04``
116     - Sent on error in previously received message.
117
118   * - |ACK|
119     - ``0x40``
120     - Sent to acknowledge receival of |DATA| frame.
121
122   * - |DATA_SEQ|
123     - ``0x80``
124     - Sent to transfer data. Sequenced.
125
126   * - |DATA_NSQ|
127     - ``0x00``
128     - Same as |DATA_SEQ|, but does not need to be ACKed.
129
130Both |NAK|- and |ACK|-type frames are used to control flow of messages and
131thus do not carry a payload. |DATA_SEQ|- and |DATA_NSQ|-type frames on the
132other hand must carry a payload. The flow sequence and interaction of
133different frame types will be described in more depth in the next section.
134
135
136SSH Packet Protocol: Flow Sequence
137==================================
138
139Each exchange begins with |SYN|, followed by a |DATA_SEQ|- or
140|DATA_NSQ|-type frame, followed by its CRC, payload, and payload CRC. In
141case of a |DATA_NSQ|-type frame, the exchange is then finished. In case of a
142|DATA_SEQ|-type frame, the receiving party has to acknowledge receival of
143the frame by responding with a message containing an |ACK|-type frame with
144the same sequence ID of the |DATA| frame. In other words, the sequence ID of
145the |ACK| frame specifies the |DATA| frame to be acknowledged. In case of an
146error, e.g. an invalid CRC, the receiving party responds with a message
147containing an |NAK|-type frame. As the sequence ID of the previous data
148frame, for which an error is indicated via the |NAK| frame, cannot be relied
149upon, the sequence ID of the |NAK| frame should not be used and is set to
150zero. After receival of an |NAK| frame, the sending party should re-send all
151outstanding (non-ACKed) messages.
152
153Sequence IDs are not synchronized between the two parties, meaning that they
154are managed independently for each party. Identifying the messages
155corresponding to a single exchange thus relies on the sequence ID as well as
156the type of the message, and the context. Specifically, the sequence ID is
157used to associate an ``ACK`` with its ``DATA_SEQ``-type frame, but not
158``DATA_SEQ``- or ``DATA_NSQ``-type frames with other ``DATA``- type frames.
159
160An example exchange might look like this:
161
162::
163
164    tx: -- SYN FRAME(D) CRC(F) PAYLOAD CRC(P) -----------------------------
165    rx: ------------------------------------- SYN FRAME(A) CRC(F) CRC(P) --
166
167where both frames have the same sequence ID (``SEQ``). Here, ``FRAME(D)``
168indicates a |DATA_SEQ|-type frame, ``FRAME(A)`` an ``ACK``-type frame,
169``CRC(F)`` the CRC over the previous frame, ``CRC(P)`` the CRC over the
170previous payload. In case of an error, the exchange would look like this:
171
172::
173
174    tx: -- SYN FRAME(D) CRC(F) PAYLOAD CRC(P) -----------------------------
175    rx: ------------------------------------- SYN FRAME(N) CRC(F) CRC(P) --
176
177upon which the sender should re-send the message. ``FRAME(N)`` indicates an
178|NAK|-type frame. Note that the sequence ID of the |NAK|-type frame is fixed
179to zero. For |DATA_NSQ|-type frames, both exchanges are the same:
180
181::
182
183    tx: -- SYN FRAME(DATA_NSQ) CRC(F) PAYLOAD CRC(P) ----------------------
184    rx: -------------------------------------------------------------------
185
186Here, an error can be detected, but not corrected or indicated to the
187sending party. These exchanges are symmetric, i.e. switching ``rx`` and
188``tx`` results again in a valid exchange. Currently, no longer exchanges are
189known.
190
191
192Commands: Requests, Responses, and Events
193=========================================
194
195Commands are sent as payload inside a data frame. Currently, this is the
196only known payload type of |DATA| frames, with a payload-type value of
197``0x80`` (:c:type:`SSH_PLD_TYPE_CMD <ssh_payload_type>`).
198
199The command-type payload (:c:type:`struct ssh_command <ssh_command>`)
200consists of an eight-byte command structure, followed by optional and
201variable length command data. The length of this optional data is derived
202from the frame payload length given in the corresponding frame, i.e. it is
203``frame.len - sizeof(struct ssh_command)``. The command struct contains the
204following fields, packed together and in order:
205
206.. flat-table:: SSH Command
207   :widths: 1 1 4
208   :header-rows: 1
209
210   * - Field
211     - Type
212     - Description
213
214   * - |TYPE|
215     - |u8|
216     - Type of the payload. For commands always ``0x80``.
217
218   * - |TC|
219     - |u8|
220     - Target category.
221
222   * - |TID| (out)
223     - |u8|
224     - Target ID for outgoing (host to EC) commands.
225
226   * - |TID| (in)
227     - |u8|
228     - Target ID for incoming (EC to host) commands.
229
230   * - |IID|
231     - |u8|
232     - Instance ID.
233
234   * - |RQID|
235     - |u16|
236     - Request ID.
237
238   * - |CID|
239     - |u8|
240     - Command ID.
241
242The command struct and data, in general, does not contain any failure
243detection mechanism (e.g. CRCs), this is solely done on the frame level.
244
245Command-type payloads are used by the host to send commands and requests to
246the EC as well as by the EC to send responses and events back to the host.
247We differentiate between requests (sent by the host), responses (sent by the
248EC in response to a request), and events (sent by the EC without a preceding
249request).
250
251Commands and events are uniquely identified by their target category
252(``TC``) and command ID (``CID``). The target category specifies a general
253category for the command (e.g. system in general, vs. battery and AC, vs.
254temperature, and so on), while the command ID specifies the command inside
255that category. Only the combination of |TC| + |CID| is unique. Additionally,
256commands have an instance ID (``IID``), which is used to differentiate
257between different sub-devices. For example ``TC=3`` ``CID=1`` is a
258request to get the temperature on a thermal sensor, where |IID| specifies
259the respective sensor. If the instance ID is not used, it should be set to
260zero. If instance IDs are used, they, in general, start with a value of one,
261whereas zero may be used for instance independent queries, if applicable. A
262response to a request should have the same target category, command ID, and
263instance ID as the corresponding request.
264
265Responses are matched to their corresponding request via the request ID
266(``RQID``) field. This is a 16 bit wrapping counter similar to the sequence
267ID on the frames. Note that the sequence ID of the frames for a
268request-response pair does not match. Only the request ID has to match.
269Frame-protocol wise these are two separate exchanges, and may even be
270separated, e.g. by an event being sent after the request but before the
271response. Not all commands produce a response, and this is not detectable by
272|TC| + |CID|. It is the responsibility of the issuing party to wait for a
273response (or signal this to the communication framework, as is done in
274SAN/ACPI via the ``SNC`` flag).
275
276Events are identified by unique and reserved request IDs. These IDs should
277not be used by the host when sending a new request. They are used on the
278host to, first, detect events and, second, match them with a registered
279event handler. Request IDs for events are chosen by the host and directed to
280the EC when setting up and enabling an event source (via the
281enable-event-source request). The EC then uses the specified request ID for
282events sent from the respective source. Note that an event should still be
283identified by its target category, command ID, and, if applicable, instance
284ID, as a single event source can send multiple different event types. In
285general, however, a single target category should map to a single reserved
286event request ID.
287
288Furthermore, requests, responses, and events have an associated target ID
289(``TID``). This target ID is split into output (host to EC) and input (EC to
290host) fields, with the respecting other field (e.g. output field on incoming
291messages) set to zero. Two ``TID`` values are known: Primary (``0x01``) and
292secondary (``0x02``). In general, the response to a request should have the
293same ``TID`` value, however, the field (output vs. input) should be used in
294accordance to the direction in which the response is sent (i.e. on the input
295field, as responses are generally sent from the EC to the host).
296
297Note that, even though requests and events should be uniquely identifiable
298by target category and command ID alone, the EC may require specific
299target ID and instance ID values to accept a command. A command that is
300accepted for ``TID=1``, for example, may not be accepted for ``TID=2``
301and vice versa.
302
303
304Limitations and Observations
305============================
306
307The protocol can, in theory, handle up to ``U8_MAX`` frames in parallel,
308with up to ``U16_MAX`` pending requests (neglecting request IDs reserved for
309events). In practice, however, this is more limited. From our testing
310(although via a python and thus a user-space program), it seems that the EC
311can handle up to four requests (mostly) reliably in parallel at a certain
312time. With five or more requests in parallel, consistent discarding of
313commands (ACKed frame but no command response) has been observed. For five
314simultaneous commands, this reproducibly resulted in one command being
315dropped and four commands being handled.
316
317However, it has also been noted that, even with three requests in parallel,
318occasional frame drops happen. Apart from this, with a limit of three
319pending requests, no dropped commands (i.e. command being dropped but frame
320carrying command being ACKed) have been observed. In any case, frames (and
321possibly also commands) should be re-sent by the host if a certain timeout
322is exceeded. This is done by the EC for frames with a timeout of one second,
323up to two re-tries (i.e. three transmissions in total). The limit of
324re-tries also applies to received NAKs, and, in a worst case scenario, can
325lead to entire messages being dropped.
326
327While this also seems to work fine for pending data frames as long as no
328transmission failures occur, implementation and handling of these seems to
329depend on the assumption that there is only one non-acknowledged data frame.
330In particular, the detection of repeated frames relies on the last sequence
331number. This means that, if a frame that has been successfully received by
332the EC is sent again, e.g. due to the host not receiving an |ACK|, the EC
333will only detect this if it has the sequence ID of the last frame received
334by the EC. As an example: Sending two frames with ``SEQ=0`` and ``SEQ=1``
335followed by a repetition of ``SEQ=0`` will not detect the second ``SEQ=0``
336frame as such, and thus execute the command in this frame each time it has
337been received, i.e. twice in this example. Sending ``SEQ=0``, ``SEQ=1`` and
338then repeating ``SEQ=1`` will detect the second ``SEQ=1`` as repetition of
339the first one and ignore it, thus executing the contained command only once.
340
341In conclusion, this suggests a limit of at most one pending un-ACKed frame
342(per party, effectively leading to synchronous communication regarding
343frames) and at most three pending commands. The limit to synchronous frame
344transfers seems to be consistent with behavior observed on Windows.
345