1================================
2I2C muxes and complex topologies
3================================
4
5There are a couple of reasons for building more complex I2C topologies
6than a straight-forward I2C bus with one adapter and one or more devices.
7
8Some example use cases are:
9
101. A mux may be needed on the bus to prevent address collisions.
11
122. The bus may be accessible from some external bus master, and arbitration
13   may be needed to determine if it is ok to access the bus.
14
153. A device (particularly RF tuners) may want to avoid the digital noise
16   from the I2C bus, at least most of the time, and sits behind a gate
17   that has to be operated before the device can be accessed.
18
19Several types of hardware components such as I2C muxes, I2C gates and I2C
20arbitrators allow to handle such needs.
21
22These components are represented as I2C adapter trees by Linux, where
23each adapter has a parent adapter (except the root adapter) and zero or
24more child adapters. The root adapter is the actual adapter that issues
25I2C transfers, and all adapters with a parent are part of an "i2c-mux"
26object (quoted, since it can also be an arbitrator or a gate).
27
28Depending of the particular mux driver, something happens when there is
29an I2C transfer on one of its child adapters. The mux driver can
30obviously operate a mux, but it can also do arbitration with an external
31bus master or open a gate. The mux driver has two operations for this,
32select and deselect. select is called before the transfer and (the
33optional) deselect is called after the transfer.
34
35
36Locking
37=======
38
39There are two variants of locking available to I2C muxes, they can be
40mux-locked or parent-locked muxes.
41
42
43Mux-locked muxes
44----------------
45
46Mux-locked muxes does not lock the entire parent adapter during the
47full select-transfer-deselect transaction, only the muxes on the parent
48adapter are locked. Mux-locked muxes are mostly interesting if the
49select and/or deselect operations must use I2C transfers to complete
50their tasks. Since the parent adapter is not fully locked during the
51full transaction, unrelated I2C transfers may interleave the different
52stages of the transaction. This has the benefit that the mux driver
53may be easier and cleaner to implement, but it has some caveats.
54
55Mux-locked Example
56~~~~~~~~~~~~~~~~~~
57
58::
59
60                   .----------.     .--------.
61    .--------.     |   mux-   |-----| dev D1 |
62    |  root  |--+--|  locked  |     '--------'
63    '--------'  |  |  mux M1  |--.  .--------.
64                |  '----------'  '--| dev D2 |
65                |  .--------.       '--------'
66                '--| dev D3 |
67                   '--------'
68
69When there is an access to D1, this happens:
70
71 1. Someone issues an I2C transfer to D1.
72 2. M1 locks muxes on its parent (the root adapter in this case).
73 3. M1 calls ->select to ready the mux.
74 4. M1 (presumably) does some I2C transfers as part of its select.
75    These transfers are normal I2C transfers that locks the parent
76    adapter.
77 5. M1 feeds the I2C transfer from step 1 to its parent adapter as a
78    normal I2C transfer that locks the parent adapter.
79 6. M1 calls ->deselect, if it has one.
80 7. Same rules as in step 4, but for ->deselect.
81 8. M1 unlocks muxes on its parent.
82
83This means that accesses to D2 are lockout out for the full duration
84of the entire operation. But accesses to D3 are possibly interleaved
85at any point.
86
87Mux-locked caveats
88~~~~~~~~~~~~~~~~~~
89
90When using a mux-locked mux, be aware of the following restrictions:
91
92[ML1]
93  If you build a topology with a mux-locked mux being the parent
94  of a parent-locked mux, this might break the expectation from the
95  parent-locked mux that the root adapter is locked during the
96  transaction.
97
98[ML2]
99  It is not safe to build arbitrary topologies with two (or more)
100  mux-locked muxes that are not siblings, when there are address
101  collisions between the devices on the child adapters of these
102  non-sibling muxes.
103
104  I.e. the select-transfer-deselect transaction targeting e.g. device
105  address 0x42 behind mux-one may be interleaved with a similar
106  operation targeting device address 0x42 behind mux-two. The
107  intent with such a topology would in this hypothetical example
108  be that mux-one and mux-two should not be selected simultaneously,
109  but mux-locked muxes do not guarantee that in all topologies.
110
111[ML3]
112  A mux-locked mux cannot be used by a driver for auto-closing
113  gates/muxes, i.e. something that closes automatically after a given
114  number (one, in most cases) of I2C transfers. Unrelated I2C transfers
115  may creep in and close prematurely.
116
117[ML4]
118  If any non-I2C operation in the mux driver changes the I2C mux state,
119  the driver has to lock the root adapter during that operation.
120  Otherwise garbage may appear on the bus as seen from devices
121  behind the mux, when an unrelated I2C transfer is in flight during
122  the non-I2C mux-changing operation.
123
124
125Parent-locked muxes
126-------------------
127
128Parent-locked muxes lock the parent adapter during the full select-
129transfer-deselect transaction. The implication is that the mux driver
130has to ensure that any and all I2C transfers through that parent
131adapter during the transaction are unlocked I2C transfers (using e.g.
132__i2c_transfer), or a deadlock will follow.
133
134Parent-locked Example
135~~~~~~~~~~~~~~~~~~~~~
136
137::
138
139                   .----------.     .--------.
140    .--------.     |  parent- |-----| dev D1 |
141    |  root  |--+--|  locked  |     '--------'
142    '--------'  |  |  mux M1  |--.  .--------.
143                |  '----------'  '--| dev D2 |
144                |  .--------.       '--------'
145                '--| dev D3 |
146                   '--------'
147
148When there is an access to D1, this happens:
149
150 1.  Someone issues an I2C transfer to D1.
151 2.  M1 locks muxes on its parent (the root adapter in this case).
152 3.  M1 locks its parent adapter.
153 4.  M1 calls ->select to ready the mux.
154 5.  If M1 does any I2C transfers (on this root adapter) as part of
155     its select, those transfers must be unlocked I2C transfers so
156     that they do not deadlock the root adapter.
157 6.  M1 feeds the I2C transfer from step 1 to the root adapter as an
158     unlocked I2C transfer, so that it does not deadlock the parent
159     adapter.
160 7.  M1 calls ->deselect, if it has one.
161 8.  Same rules as in step 5, but for ->deselect.
162 9.  M1 unlocks its parent adapter.
163 10. M1 unlocks muxes on its parent.
164
165This means that accesses to both D2 and D3 are locked out for the full
166duration of the entire operation.
167
168Parent-locked Caveats
169~~~~~~~~~~~~~~~~~~~~~
170
171When using a parent-locked mux, be aware of the following restrictions:
172
173[PL1]
174  If you build a topology with a parent-locked mux being the child
175  of another mux, this might break a possible assumption from the
176  child mux that the root adapter is unused between its select op
177  and the actual transfer (e.g. if the child mux is auto-closing
178  and the parent mux issues I2C transfers as part of its select).
179  This is especially the case if the parent mux is mux-locked, but
180  it may also happen if the parent mux is parent-locked.
181
182[PL2]
183  If select/deselect calls out to other subsystems such as gpio,
184  pinctrl, regmap or iio, it is essential that any I2C transfers
185  caused by these subsystems are unlocked. This can be convoluted to
186  accomplish, maybe even impossible if an acceptably clean solution
187  is sought.
188
189
190Complex Examples
191================
192
193Parent-locked mux as parent of parent-locked mux
194------------------------------------------------
195
196This is a useful topology, but it can be bad::
197
198                   .----------.     .----------.     .--------.
199    .--------.     |  parent- |-----|  parent- |-----| dev D1 |
200    |  root  |--+--|  locked  |     |  locked  |     '--------'
201    '--------'  |  |  mux M1  |--.  |  mux M2  |--.  .--------.
202                |  '----------'  |  '----------'  '--| dev D2 |
203                |  .--------.    |  .--------.       '--------'
204                '--| dev D4 |    '--| dev D3 |
205                   '--------'       '--------'
206
207When any device is accessed, all other devices are locked out for
208the full duration of the operation (both muxes lock their parent,
209and specifically when M2 requests its parent to lock, M1 passes
210the buck to the root adapter).
211
212This topology is bad if M2 is an auto-closing mux and M1->select
213issues any unlocked I2C transfers on the root adapter that may leak
214through and be seen by the M2 adapter, thus closing M2 prematurely.
215
216
217Mux-locked mux as parent of mux-locked mux
218------------------------------------------
219
220This is a good topology::
221
222                   .----------.     .----------.     .--------.
223    .--------.     |   mux-   |-----|   mux-   |-----| dev D1 |
224    |  root  |--+--|  locked  |     |  locked  |     '--------'
225    '--------'  |  |  mux M1  |--.  |  mux M2  |--.  .--------.
226                |  '----------'  |  '----------'  '--| dev D2 |
227                |  .--------.    |  .--------.       '--------'
228                '--| dev D4 |    '--| dev D3 |
229                   '--------'       '--------'
230
231When device D1 is accessed, accesses to D2 are locked out for the
232full duration of the operation (muxes on the top child adapter of M1
233are locked). But accesses to D3 and D4 are possibly interleaved at
234any point.
235
236Accesses to D3 locks out D1 and D2, but accesses to D4 are still possibly
237interleaved.
238
239
240Mux-locked mux as parent of parent-locked mux
241---------------------------------------------
242
243This is probably a bad topology::
244
245                   .----------.     .----------.     .--------.
246    .--------.     |   mux-   |-----|  parent- |-----| dev D1 |
247    |  root  |--+--|  locked  |     |  locked  |     '--------'
248    '--------'  |  |  mux M1  |--.  |  mux M2  |--.  .--------.
249                |  '----------'  |  '----------'  '--| dev D2 |
250                |  .--------.    |  .--------.       '--------'
251                '--| dev D4 |    '--| dev D3 |
252                   '--------'       '--------'
253
254When device D1 is accessed, accesses to D2 and D3 are locked out
255for the full duration of the operation (M1 locks child muxes on the
256root adapter). But accesses to D4 are possibly interleaved at any
257point.
258
259This kind of topology is generally not suitable and should probably
260be avoided. The reason is that M2 probably assumes that there will
261be no I2C transfers during its calls to ->select and ->deselect, and
262if there are, any such transfers might appear on the slave side of M2
263as partial I2C transfers, i.e. garbage or worse. This might cause
264device lockups and/or other problems.
265
266The topology is especially troublesome if M2 is an auto-closing
267mux. In that case, any interleaved accesses to D4 might close M2
268prematurely, as might any I2C transfers part of M1->select.
269
270But if M2 is not making the above stated assumption, and if M2 is not
271auto-closing, the topology is fine.
272
273
274Parent-locked mux as parent of mux-locked mux
275---------------------------------------------
276
277This is a good topology::
278
279                   .----------.     .----------.     .--------.
280    .--------.     |  parent- |-----|   mux-   |-----| dev D1 |
281    |  root  |--+--|  locked  |     |  locked  |     '--------'
282    '--------'  |  |  mux M1  |--.  |  mux M2  |--.  .--------.
283                |  '----------'  |  '----------'  '--| dev D2 |
284                |  .--------.    |  .--------.       '--------'
285                '--| dev D4 |    '--| dev D3 |
286                   '--------'       '--------'
287
288When D1 is accessed, accesses to D2 are locked out for the full
289duration of the operation (muxes on the top child adapter of M1
290are locked). Accesses to D3 and D4 are possibly interleaved at
291any point, just as is expected for mux-locked muxes.
292
293When D3 or D4 are accessed, everything else is locked out. For D3
294accesses, M1 locks the root adapter. For D4 accesses, the root
295adapter is locked directly.
296
297
298Two mux-locked sibling muxes
299----------------------------
300
301This is a good topology::
302
303                                    .--------.
304                   .----------.  .--| dev D1 |
305                   |   mux-   |--'  '--------'
306                .--|  locked  |     .--------.
307                |  |  mux M1  |-----| dev D2 |
308                |  '----------'     '--------'
309                |  .----------.     .--------.
310    .--------.  |  |   mux-   |-----| dev D3 |
311    |  root  |--+--|  locked  |     '--------'
312    '--------'  |  |  mux M2  |--.  .--------.
313                |  '----------'  '--| dev D4 |
314                |  .--------.       '--------'
315                '--| dev D5 |
316                   '--------'
317
318When D1 is accessed, accesses to D2, D3 and D4 are locked out. But
319accesses to D5 may be interleaved at any time.
320
321
322Two parent-locked sibling muxes
323-------------------------------
324
325This is a good topology::
326
327                                    .--------.
328                   .----------.  .--| dev D1 |
329                   |  parent- |--'  '--------'
330                .--|  locked  |     .--------.
331                |  |  mux M1  |-----| dev D2 |
332                |  '----------'     '--------'
333                |  .----------.     .--------.
334    .--------.  |  |  parent- |-----| dev D3 |
335    |  root  |--+--|  locked  |     '--------'
336    '--------'  |  |  mux M2  |--.  .--------.
337                |  '----------'  '--| dev D4 |
338                |  .--------.       '--------'
339                '--| dev D5 |
340                   '--------'
341
342When any device is accessed, accesses to all other devices are locked
343out.
344
345
346Mux-locked and parent-locked sibling muxes
347------------------------------------------
348
349This is a good topology::
350
351                                    .--------.
352                   .----------.  .--| dev D1 |
353                   |   mux-   |--'  '--------'
354                .--|  locked  |     .--------.
355                |  |  mux M1  |-----| dev D2 |
356                |  '----------'     '--------'
357                |  .----------.     .--------.
358    .--------.  |  |  parent- |-----| dev D3 |
359    |  root  |--+--|  locked  |     '--------'
360    '--------'  |  |  mux M2  |--.  .--------.
361                |  '----------'  '--| dev D4 |
362                |  .--------.       '--------'
363                '--| dev D5 |
364                   '--------'
365
366When D1 or D2 are accessed, accesses to D3 and D4 are locked out while
367accesses to D5 may interleave. When D3 or D4 are accessed, accesses to
368all other devices are locked out.
369
370
371Mux type of existing device drivers
372===================================
373
374Whether a device is mux-locked or parent-locked depends on its
375implementation. The following list was correct at the time of writing:
376
377In drivers/i2c/muxes/:
378
379======================    =============================================
380i2c-arb-gpio-challenge    Parent-locked
381i2c-mux-gpio              Normally parent-locked, mux-locked iff
382                          all involved gpio pins are controlled by the
383                          same I2C root adapter that they mux.
384i2c-mux-gpmux             Normally parent-locked, mux-locked iff
385                          specified in device-tree.
386i2c-mux-ltc4306           Mux-locked
387i2c-mux-mlxcpld           Parent-locked
388i2c-mux-pca9541           Parent-locked
389i2c-mux-pca954x           Parent-locked
390i2c-mux-pinctrl           Normally parent-locked, mux-locked iff
391                          all involved pinctrl devices are controlled
392                          by the same I2C root adapter that they mux.
393i2c-mux-reg               Parent-locked
394======================    =============================================
395
396In drivers/iio/:
397
398======================    =============================================
399gyro/mpu3050              Mux-locked
400imu/inv_mpu6050/          Mux-locked
401======================    =============================================
402
403In drivers/media/:
404
405=======================   =============================================
406dvb-frontends/lgdt3306a   Mux-locked
407dvb-frontends/m88ds3103   Parent-locked
408dvb-frontends/rtl2830     Parent-locked
409dvb-frontends/rtl2832     Mux-locked
410dvb-frontends/si2168      Mux-locked
411usb/cx231xx/              Parent-locked
412=======================   =============================================
413