1=========================
2NXP SJA1105 switch driver
3=========================
4
5Overview
6========
7
8The NXP SJA1105 is a family of 10 SPI-managed automotive switches:
9
10- SJA1105E: First generation, no TTEthernet
11- SJA1105T: First generation, TTEthernet
12- SJA1105P: Second generation, no TTEthernet, no SGMII
13- SJA1105Q: Second generation, TTEthernet, no SGMII
14- SJA1105R: Second generation, no TTEthernet, SGMII
15- SJA1105S: Second generation, TTEthernet, SGMII
16- SJA1110A: Third generation, TTEthernet, SGMII, integrated 100base-T1 and
17  100base-TX PHYs
18- SJA1110B: Third generation, TTEthernet, SGMII, 100base-T1, 100base-TX
19- SJA1110C: Third generation, TTEthernet, SGMII, 100base-T1, 100base-TX
20- SJA1110D: Third generation, TTEthernet, SGMII, 100base-T1
21
22Being automotive parts, their configuration interface is geared towards
23set-and-forget use, with minimal dynamic interaction at runtime. They
24require a static configuration to be composed by software and packed
25with CRC and table headers, and sent over SPI.
26
27The static configuration is composed of several configuration tables. Each
28table takes a number of entries. Some configuration tables can be (partially)
29reconfigured at runtime, some not. Some tables are mandatory, some not:
30
31============================= ================== =============================
32Table                          Mandatory          Reconfigurable
33============================= ================== =============================
34Schedule                       no                 no
35Schedule entry points          if Scheduling      no
36VL Lookup                      no                 no
37VL Policing                    if VL Lookup       no
38VL Forwarding                  if VL Lookup       no
39L2 Lookup                      no                 no
40L2 Policing                    yes                no
41VLAN Lookup                    yes                yes
42L2 Forwarding                  yes                partially (fully on P/Q/R/S)
43MAC Config                     yes                partially (fully on P/Q/R/S)
44Schedule Params                if Scheduling      no
45Schedule Entry Points Params   if Scheduling      no
46VL Forwarding Params           if VL Forwarding   no
47L2 Lookup Params               no                 partially (fully on P/Q/R/S)
48L2 Forwarding Params           yes                no
49Clock Sync Params              no                 no
50AVB Params                     no                 no
51General Params                 yes                partially
52Retagging                      no                 yes
53xMII Params                    yes                no
54SGMII                          no                 yes
55============================= ================== =============================
56
57
58Also the configuration is write-only (software cannot read it back from the
59switch except for very few exceptions).
60
61The driver creates a static configuration at probe time, and keeps it at
62all times in memory, as a shadow for the hardware state. When required to
63change a hardware setting, the static configuration is also updated.
64If that changed setting can be transmitted to the switch through the dynamic
65reconfiguration interface, it is; otherwise the switch is reset and
66reprogrammed with the updated static configuration.
67
68Switching features
69==================
70
71The driver supports the configuration of L2 forwarding rules in hardware for
72port bridging. The forwarding, broadcast and flooding domain between ports can
73be restricted through two methods: either at the L2 forwarding level (isolate
74one bridge's ports from another's) or at the VLAN port membership level
75(isolate ports within the same bridge). The final forwarding decision taken by
76the hardware is a logical AND of these two sets of rules.
77
78The hardware tags all traffic internally with a port-based VLAN (pvid), or it
79decodes the VLAN information from the 802.1Q tag. Advanced VLAN classification
80is not possible. Once attributed a VLAN tag, frames are checked against the
81port's membership rules and dropped at ingress if they don't match any VLAN.
82This behavior is available when switch ports are enslaved to a bridge with
83``vlan_filtering 1``.
84
85Normally the hardware is not configurable with respect to VLAN awareness, but
86by changing what TPID the switch searches 802.1Q tags for, the semantics of a
87bridge with ``vlan_filtering 0`` can be kept (accept all traffic, tagged or
88untagged), and therefore this mode is also supported.
89
90Segregating the switch ports in multiple bridges is supported (e.g. 2 + 2), but
91all bridges should have the same level of VLAN awareness (either both have
92``vlan_filtering`` 0, or both 1).
93
94Topology and loop detection through STP is supported.
95
96Offloads
97========
98
99Time-aware scheduling
100---------------------
101
102The switch supports a variation of the enhancements for scheduled traffic
103specified in IEEE 802.1Q-2018 (formerly 802.1Qbv). This means it can be used to
104ensure deterministic latency for priority traffic that is sent in-band with its
105gate-open event in the network schedule.
106
107This capability can be managed through the tc-taprio offload ('flags 2'). The
108difference compared to the software implementation of taprio is that the latter
109would only be able to shape traffic originated from the CPU, but not
110autonomously forwarded flows.
111
112The device has 8 traffic classes, and maps incoming frames to one of them based
113on the VLAN PCP bits (if no VLAN is present, the port-based default is used).
114As described in the previous sections, depending on the value of
115``vlan_filtering``, the EtherType recognized by the switch as being VLAN can
116either be the typical 0x8100 or a custom value used internally by the driver
117for tagging. Therefore, the switch ignores the VLAN PCP if used in standalone
118or bridge mode with ``vlan_filtering=0``, as it will not recognize the 0x8100
119EtherType. In these modes, injecting into a particular TX queue can only be
120done by the DSA net devices, which populate the PCP field of the tagging header
121on egress. Using ``vlan_filtering=1``, the behavior is the other way around:
122offloaded flows can be steered to TX queues based on the VLAN PCP, but the DSA
123net devices are no longer able to do that. To inject frames into a hardware TX
124queue with VLAN awareness active, it is necessary to create a VLAN
125sub-interface on the DSA master port, and send normal (0x8100) VLAN-tagged
126towards the switch, with the VLAN PCP bits set appropriately.
127
128Management traffic (having DMAC 01-80-C2-xx-xx-xx or 01-19-1B-xx-xx-xx) is the
129notable exception: the switch always treats it with a fixed priority and
130disregards any VLAN PCP bits even if present. The traffic class for management
131traffic has a value of 7 (highest priority) at the moment, which is not
132configurable in the driver.
133
134Below is an example of configuring a 500 us cyclic schedule on egress port
135``swp5``. The traffic class gate for management traffic (7) is open for 100 us,
136and the gates for all other traffic classes are open for 400 us::
137
138  #!/bin/bash
139
140  set -e -u -o pipefail
141
142  NSEC_PER_SEC="1000000000"
143
144  gatemask() {
145          local tc_list="$1"
146          local mask=0
147
148          for tc in ${tc_list}; do
149                  mask=$((${mask} | (1 << ${tc})))
150          done
151
152          printf "%02x" ${mask}
153  }
154
155  if ! systemctl is-active --quiet ptp4l; then
156          echo "Please start the ptp4l service"
157          exit
158  fi
159
160  now=$(phc_ctl /dev/ptp1 get | gawk '/clock time is/ { print $5; }')
161  # Phase-align the base time to the start of the next second.
162  sec=$(echo "${now}" | gawk -F. '{ print $1; }')
163  base_time="$(((${sec} + 1) * ${NSEC_PER_SEC}))"
164
165  tc qdisc add dev swp5 parent root handle 100 taprio \
166          num_tc 8 \
167          map 0 1 2 3 5 6 7 \
168          queues 1@0 1@1 1@2 1@3 1@4 1@5 1@6 1@7 \
169          base-time ${base_time} \
170          sched-entry S $(gatemask 7) 100000 \
171          sched-entry S $(gatemask "0 1 2 3 4 5 6") 400000 \
172          flags 2
173
174It is possible to apply the tc-taprio offload on multiple egress ports. There
175are hardware restrictions related to the fact that no gate event may trigger
176simultaneously on two ports. The driver checks the consistency of the schedules
177against this restriction and errors out when appropriate. Schedule analysis is
178needed to avoid this, which is outside the scope of the document.
179
180Routing actions (redirect, trap, drop)
181--------------------------------------
182
183The switch is able to offload flow-based redirection of packets to a set of
184destination ports specified by the user. Internally, this is implemented by
185making use of Virtual Links, a TTEthernet concept.
186
187The driver supports 2 types of keys for Virtual Links:
188
189- VLAN-aware virtual links: these match on destination MAC address, VLAN ID and
190  VLAN PCP.
191- VLAN-unaware virtual links: these match on destination MAC address only.
192
193The VLAN awareness state of the bridge (vlan_filtering) cannot be changed while
194there are virtual link rules installed.
195
196Composing multiple actions inside the same rule is supported. When only routing
197actions are requested, the driver creates a "non-critical" virtual link. When
198the action list also contains tc-gate (more details below), the virtual link
199becomes "time-critical" (draws frame buffers from a reserved memory partition,
200etc).
201
202The 3 routing actions that are supported are "trap", "drop" and "redirect".
203
204Example 1: send frames received on swp2 with a DA of 42:be:24:9b:76:20 to the
205CPU and to swp3. This type of key (DA only) when the port's VLAN awareness
206state is off::
207
208  tc qdisc add dev swp2 clsact
209  tc filter add dev swp2 ingress flower skip_sw dst_mac 42:be:24:9b:76:20 \
210          action mirred egress redirect dev swp3 \
211          action trap
212
213Example 2: drop frames received on swp2 with a DA of 42:be:24:9b:76:20, a VID
214of 100 and a PCP of 0::
215
216  tc filter add dev swp2 ingress protocol 802.1Q flower skip_sw \
217          dst_mac 42:be:24:9b:76:20 vlan_id 100 vlan_prio 0 action drop
218
219Time-based ingress policing
220---------------------------
221
222The TTEthernet hardware abilities of the switch can be constrained to act
223similarly to the Per-Stream Filtering and Policing (PSFP) clause specified in
224IEEE 802.1Q-2018 (formerly 802.1Qci). This means it can be used to perform
225tight timing-based admission control for up to 1024 flows (identified by a
226tuple composed of destination MAC address, VLAN ID and VLAN PCP). Packets which
227are received outside their expected reception window are dropped.
228
229This capability can be managed through the offload of the tc-gate action. As
230routing actions are intrinsic to virtual links in TTEthernet (which performs
231explicit routing of time-critical traffic and does not leave that in the hands
232of the FDB, flooding etc), the tc-gate action may never appear alone when
233asking sja1105 to offload it. One (or more) redirect or trap actions must also
234follow along.
235
236Example: create a tc-taprio schedule that is phase-aligned with a tc-gate
237schedule (the clocks must be synchronized by a 1588 application stack, which is
238outside the scope of this document). No packet delivered by the sender will be
239dropped. Note that the reception window is larger than the transmission window
240(and much more so, in this example) to compensate for the packet propagation
241delay of the link (which can be determined by the 1588 application stack).
242
243Receiver (sja1105)::
244
245  tc qdisc add dev swp2 clsact
246  now=$(phc_ctl /dev/ptp1 get | awk '/clock time is/ {print $5}') && \
247          sec=$(echo $now | awk -F. '{print $1}') && \
248          base_time="$(((sec + 2) * 1000000000))" && \
249          echo "base time ${base_time}"
250  tc filter add dev swp2 ingress flower skip_sw \
251          dst_mac 42:be:24:9b:76:20 \
252          action gate base-time ${base_time} \
253          sched-entry OPEN  60000 -1 -1 \
254          sched-entry CLOSE 40000 -1 -1 \
255          action trap
256
257Sender::
258
259  now=$(phc_ctl /dev/ptp0 get | awk '/clock time is/ {print $5}') && \
260          sec=$(echo $now | awk -F. '{print $1}') && \
261          base_time="$(((sec + 2) * 1000000000))" && \
262          echo "base time ${base_time}"
263  tc qdisc add dev eno0 parent root taprio \
264          num_tc 8 \
265          map 0 1 2 3 4 5 6 7 \
266          queues 1@0 1@1 1@2 1@3 1@4 1@5 1@6 1@7 \
267          base-time ${base_time} \
268          sched-entry S 01  50000 \
269          sched-entry S 00  50000 \
270          flags 2
271
272The engine used to schedule the ingress gate operations is the same that the
273one used for the tc-taprio offload. Therefore, the restrictions regarding the
274fact that no two gate actions (either tc-gate or tc-taprio gates) may fire at
275the same time (during the same 200 ns slot) still apply.
276
277To come in handy, it is possible to share time-triggered virtual links across
278more than 1 ingress port, via flow blocks. In this case, the restriction of
279firing at the same time does not apply because there is a single schedule in
280the system, that of the shared virtual link::
281
282  tc qdisc add dev swp2 ingress_block 1 clsact
283  tc qdisc add dev swp3 ingress_block 1 clsact
284  tc filter add block 1 flower skip_sw dst_mac 42:be:24:9b:76:20 \
285          action gate index 2 \
286          base-time 0 \
287          sched-entry OPEN 50000000 -1 -1 \
288          sched-entry CLOSE 50000000 -1 -1 \
289          action trap
290
291Hardware statistics for each flow are also available ("pkts" counts the number
292of dropped frames, which is a sum of frames dropped due to timing violations,
293lack of destination ports and MTU enforcement checks). Byte-level counters are
294not available.
295
296Limitations
297===========
298
299The SJA1105 switch family always performs VLAN processing. When configured as
300VLAN-unaware, frames carry a different VLAN tag internally, depending on
301whether the port is standalone or under a VLAN-unaware bridge.
302
303The virtual link keys are always fixed at {MAC DA, VLAN ID, VLAN PCP}, but the
304driver asks for the VLAN ID and VLAN PCP when the port is under a VLAN-aware
305bridge. Otherwise, it fills in the VLAN ID and PCP automatically, based on
306whether the port is standalone or in a VLAN-unaware bridge, and accepts only
307"VLAN-unaware" tc-flower keys (MAC DA).
308
309The existing tc-flower keys that are offloaded using virtual links are no
310longer operational after one of the following happens:
311
312- port was standalone and joins a bridge (VLAN-aware or VLAN-unaware)
313- port is part of a bridge whose VLAN awareness state changes
314- port was part of a bridge and becomes standalone
315- port was standalone, but another port joins a VLAN-aware bridge and this
316  changes the global VLAN awareness state of the bridge
317
318The driver cannot veto all these operations, and it cannot update/remove the
319existing tc-flower filters either. So for proper operation, the tc-flower
320filters should be installed only after the forwarding configuration of the port
321has been made, and removed by user space before making any changes to it.
322
323Device Tree bindings and board design
324=====================================
325
326This section references ``Documentation/devicetree/bindings/net/dsa/nxp,sja1105.yaml``
327and aims to showcase some potential switch caveats.
328
329RMII PHY role and out-of-band signaling
330---------------------------------------
331
332In the RMII spec, the 50 MHz clock signals are either driven by the MAC or by
333an external oscillator (but not by the PHY).
334But the spec is rather loose and devices go outside it in several ways.
335Some PHYs go against the spec and may provide an output pin where they source
336the 50 MHz clock themselves, in an attempt to be helpful.
337On the other hand, the SJA1105 is only binary configurable - when in the RMII
338MAC role it will also attempt to drive the clock signal. To prevent this from
339happening it must be put in RMII PHY role.
340But doing so has some unintended consequences.
341In the RMII spec, the PHY can transmit extra out-of-band signals via RXD[1:0].
342These are practically some extra code words (/J/ and /K/) sent prior to the
343preamble of each frame. The MAC does not have this out-of-band signaling
344mechanism defined by the RMII spec.
345So when the SJA1105 port is put in PHY role to avoid having 2 drivers on the
346clock signal, inevitably an RMII PHY-to-PHY connection is created. The SJA1105
347emulates a PHY interface fully and generates the /J/ and /K/ symbols prior to
348frame preambles, which the real PHY is not expected to understand. So the PHY
349simply encodes the extra symbols received from the SJA1105-as-PHY onto the
350100Base-Tx wire.
351On the other side of the wire, some link partners might discard these extra
352symbols, while others might choke on them and discard the entire Ethernet
353frames that follow along. This looks like packet loss with some link partners
354but not with others.
355The take-away is that in RMII mode, the SJA1105 must be let to drive the
356reference clock if connected to a PHY.
357
358RGMII fixed-link and internal delays
359------------------------------------
360
361As mentioned in the bindings document, the second generation of devices has
362tunable delay lines as part of the MAC, which can be used to establish the
363correct RGMII timing budget.
364When powered up, these can shift the Rx and Tx clocks with a phase difference
365between 73.8 and 101.7 degrees.
366The catch is that the delay lines need to lock onto a clock signal with a
367stable frequency. This means that there must be at least 2 microseconds of
368silence between the clock at the old vs at the new frequency. Otherwise the
369lock is lost and the delay lines must be reset (powered down and back up).
370In RGMII the clock frequency changes with link speed (125 MHz at 1000 Mbps, 25
371MHz at 100 Mbps and 2.5 MHz at 10 Mbps), and link speed might change during the
372AN process.
373In the situation where the switch port is connected through an RGMII fixed-link
374to a link partner whose link state life cycle is outside the control of Linux
375(such as a different SoC), then the delay lines would remain unlocked (and
376inactive) until there is manual intervention (ifdown/ifup on the switch port).
377The take-away is that in RGMII mode, the switch's internal delays are only
378reliable if the link partner never changes link speeds, or if it does, it does
379so in a way that is coordinated with the switch port (practically, both ends of
380the fixed-link are under control of the same Linux system).
381As to why would a fixed-link interface ever change link speeds: there are
382Ethernet controllers out there which come out of reset in 100 Mbps mode, and
383their driver inevitably needs to change the speed and clock frequency if it's
384required to work at gigabit.
385
386MDIO bus and PHY management
387---------------------------
388
389The SJA1105 does not have an MDIO bus and does not perform in-band AN either.
390Therefore there is no link state notification coming from the switch device.
391A board would need to hook up the PHYs connected to the switch to any other
392MDIO bus available to Linux within the system (e.g. to the DSA master's MDIO
393bus). Link state management then works by the driver manually keeping in sync
394(over SPI commands) the MAC link speed with the settings negotiated by the PHY.
395
396By comparison, the SJA1110 supports an MDIO slave access point over which its
397internal 100base-T1 PHYs can be accessed from the host. This is, however, not
398used by the driver, instead the internal 100base-T1 and 100base-TX PHYs are
399accessed through SPI commands, modeled in Linux as virtual MDIO buses.
400
401The microcontroller attached to the SJA1110 port 0 also has an MDIO controller
402operating in master mode, however the driver does not support this either,
403since the microcontroller gets disabled when the Linux driver operates.
404Discrete PHYs connected to the switch ports should have their MDIO interface
405attached to an MDIO controller from the host system and not to the switch,
406similar to SJA1105.
407
408Port compatibility matrix
409-------------------------
410
411The SJA1105 port compatibility matrix is:
412
413===== ============== ============== ==============
414Port   SJA1105E/T     SJA1105P/Q     SJA1105R/S
415===== ============== ============== ==============
4160      xMII           xMII           xMII
4171      xMII           xMII           xMII
4182      xMII           xMII           xMII
4193      xMII           xMII           xMII
4204      xMII           xMII           SGMII
421===== ============== ============== ==============
422
423
424The SJA1110 port compatibility matrix is:
425
426===== ============== ============== ============== ==============
427Port   SJA1110A       SJA1110B       SJA1110C       SJA1110D
428===== ============== ============== ============== ==============
4290      RevMII (uC)    RevMII (uC)    RevMII (uC)    RevMII (uC)
4301      100base-TX     100base-TX     100base-TX
431       or SGMII                                     SGMII
4322      xMII           xMII           xMII           xMII
433       or SGMII                                     or SGMII
4343      xMII           xMII           xMII
435       or SGMII       or SGMII                      SGMII
436       or 2500base-X  or 2500base-X                 or 2500base-X
4374      SGMII          SGMII          SGMII          SGMII
438       or 2500base-X  or 2500base-X  or 2500base-X  or 2500base-X
4395      100base-T1     100base-T1     100base-T1     100base-T1
4406      100base-T1     100base-T1     100base-T1     100base-T1
4417      100base-T1     100base-T1     100base-T1     100base-T1
4428      100base-T1     100base-T1     n/a            n/a
4439      100base-T1     100base-T1     n/a            n/a
44410     100base-T1     n/a            n/a            n/a
445===== ============== ============== ============== ==============
446