xref: /linux/Documentation/networking/timestamping.rst (revision 55d0969c451159cff86949b38c39171cab962069)
1.. SPDX-License-Identifier: GPL-2.0
2
3============
4Timestamping
5============
6
7
81. Control Interfaces
9=====================
10
11The interfaces for receiving network packages timestamps are:
12
13SO_TIMESTAMP
14  Generates a timestamp for each incoming packet in (not necessarily
15  monotonic) system time. Reports the timestamp via recvmsg() in a
16  control message in usec resolution.
17  SO_TIMESTAMP is defined as SO_TIMESTAMP_NEW or SO_TIMESTAMP_OLD
18  based on the architecture type and time_t representation of libc.
19  Control message format is in struct __kernel_old_timeval for
20  SO_TIMESTAMP_OLD and in struct __kernel_sock_timeval for
21  SO_TIMESTAMP_NEW options respectively.
22
23SO_TIMESTAMPNS
24  Same timestamping mechanism as SO_TIMESTAMP, but reports the
25  timestamp as struct timespec in nsec resolution.
26  SO_TIMESTAMPNS is defined as SO_TIMESTAMPNS_NEW or SO_TIMESTAMPNS_OLD
27  based on the architecture type and time_t representation of libc.
28  Control message format is in struct timespec for SO_TIMESTAMPNS_OLD
29  and in struct __kernel_timespec for SO_TIMESTAMPNS_NEW options
30  respectively.
31
32IP_MULTICAST_LOOP + SO_TIMESTAMP[NS]
33  Only for multicast:approximate transmit timestamp obtained by
34  reading the looped packet receive timestamp.
35
36SO_TIMESTAMPING
37  Generates timestamps on reception, transmission or both. Supports
38  multiple timestamp sources, including hardware. Supports generating
39  timestamps for stream sockets.
40
41
421.1 SO_TIMESTAMP (also SO_TIMESTAMP_OLD and SO_TIMESTAMP_NEW)
43-------------------------------------------------------------
44
45This socket option enables timestamping of datagrams on the reception
46path. Because the destination socket, if any, is not known early in
47the network stack, the feature has to be enabled for all packets. The
48same is true for all early receive timestamp options.
49
50For interface details, see `man 7 socket`.
51
52Always use SO_TIMESTAMP_NEW timestamp to always get timestamp in
53struct __kernel_sock_timeval format.
54
55SO_TIMESTAMP_OLD returns incorrect timestamps after the year 2038
56on 32 bit machines.
57
581.2 SO_TIMESTAMPNS (also SO_TIMESTAMPNS_OLD and SO_TIMESTAMPNS_NEW)
59-------------------------------------------------------------------
60
61This option is identical to SO_TIMESTAMP except for the returned data type.
62Its struct timespec allows for higher resolution (ns) timestamps than the
63timeval of SO_TIMESTAMP (ms).
64
65Always use SO_TIMESTAMPNS_NEW timestamp to always get timestamp in
66struct __kernel_timespec format.
67
68SO_TIMESTAMPNS_OLD returns incorrect timestamps after the year 2038
69on 32 bit machines.
70
711.3 SO_TIMESTAMPING (also SO_TIMESTAMPING_OLD and SO_TIMESTAMPING_NEW)
72----------------------------------------------------------------------
73
74Supports multiple types of timestamp requests. As a result, this
75socket option takes a bitmap of flags, not a boolean. In::
76
77  err = setsockopt(fd, SOL_SOCKET, SO_TIMESTAMPING, &val, sizeof(val));
78
79val is an integer with any of the following bits set. Setting other
80bit returns EINVAL and does not change the current state.
81
82The socket option configures timestamp generation for individual
83sk_buffs (1.3.1), timestamp reporting to the socket's error
84queue (1.3.2) and options (1.3.3). Timestamp generation can also
85be enabled for individual sendmsg calls using cmsg (1.3.4).
86
87
881.3.1 Timestamp Generation
89^^^^^^^^^^^^^^^^^^^^^^^^^^
90
91Some bits are requests to the stack to try to generate timestamps. Any
92combination of them is valid. Changes to these bits apply to newly
93created packets, not to packets already in the stack. As a result, it
94is possible to selectively request timestamps for a subset of packets
95(e.g., for sampling) by embedding an send() call within two setsockopt
96calls, one to enable timestamp generation and one to disable it.
97Timestamps may also be generated for reasons other than being
98requested by a particular socket, such as when receive timestamping is
99enabled system wide, as explained earlier.
100
101SOF_TIMESTAMPING_RX_HARDWARE:
102  Request rx timestamps generated by the network adapter.
103
104SOF_TIMESTAMPING_RX_SOFTWARE:
105  Request rx timestamps when data enters the kernel. These timestamps
106  are generated just after a device driver hands a packet to the
107  kernel receive stack.
108
109SOF_TIMESTAMPING_TX_HARDWARE:
110  Request tx timestamps generated by the network adapter. This flag
111  can be enabled via both socket options and control messages.
112
113SOF_TIMESTAMPING_TX_SOFTWARE:
114  Request tx timestamps when data leaves the kernel. These timestamps
115  are generated in the device driver as close as possible, but always
116  prior to, passing the packet to the network interface. Hence, they
117  require driver support and may not be available for all devices.
118  This flag can be enabled via both socket options and control messages.
119
120SOF_TIMESTAMPING_TX_SCHED:
121  Request tx timestamps prior to entering the packet scheduler. Kernel
122  transmit latency is, if long, often dominated by queuing delay. The
123  difference between this timestamp and one taken at
124  SOF_TIMESTAMPING_TX_SOFTWARE will expose this latency independent
125  of protocol processing. The latency incurred in protocol
126  processing, if any, can be computed by subtracting a userspace
127  timestamp taken immediately before send() from this timestamp. On
128  machines with virtual devices where a transmitted packet travels
129  through multiple devices and, hence, multiple packet schedulers,
130  a timestamp is generated at each layer. This allows for fine
131  grained measurement of queuing delay. This flag can be enabled
132  via both socket options and control messages.
133
134SOF_TIMESTAMPING_TX_ACK:
135  Request tx timestamps when all data in the send buffer has been
136  acknowledged. This only makes sense for reliable protocols. It is
137  currently only implemented for TCP. For that protocol, it may
138  over-report measurement, because the timestamp is generated when all
139  data up to and including the buffer at send() was acknowledged: the
140  cumulative acknowledgment. The mechanism ignores SACK and FACK.
141  This flag can be enabled via both socket options and control messages.
142
143
1441.3.2 Timestamp Reporting
145^^^^^^^^^^^^^^^^^^^^^^^^^
146
147The other three bits control which timestamps will be reported in a
148generated control message. Changes to the bits take immediate
149effect at the timestamp reporting locations in the stack. Timestamps
150are only reported for packets that also have the relevant timestamp
151generation request set.
152
153SOF_TIMESTAMPING_SOFTWARE:
154  Report any software timestamps when available.
155
156SOF_TIMESTAMPING_SYS_HARDWARE:
157  This option is deprecated and ignored.
158
159SOF_TIMESTAMPING_RAW_HARDWARE:
160  Report hardware timestamps as generated by
161  SOF_TIMESTAMPING_TX_HARDWARE or SOF_TIMESTAMPING_RX_HARDWARE
162  when available.
163
164
1651.3.3 Timestamp Options
166^^^^^^^^^^^^^^^^^^^^^^^
167
168The interface supports the options
169
170SOF_TIMESTAMPING_OPT_ID:
171  Generate a unique identifier along with each packet. A process can
172  have multiple concurrent timestamping requests outstanding. Packets
173  can be reordered in the transmit path, for instance in the packet
174  scheduler. In that case timestamps will be queued onto the error
175  queue out of order from the original send() calls. It is not always
176  possible to uniquely match timestamps to the original send() calls
177  based on timestamp order or payload inspection alone, then.
178
179  This option associates each packet at send() with a unique
180  identifier and returns that along with the timestamp. The identifier
181  is derived from a per-socket u32 counter (that wraps). For datagram
182  sockets, the counter increments with each sent packet. For stream
183  sockets, it increments with every byte. For stream sockets, also set
184  SOF_TIMESTAMPING_OPT_ID_TCP, see the section below.
185
186  The counter starts at zero. It is initialized the first time that
187  the socket option is enabled. It is reset each time the option is
188  enabled after having been disabled. Resetting the counter does not
189  change the identifiers of existing packets in the system.
190
191  This option is implemented only for transmit timestamps. There, the
192  timestamp is always looped along with a struct sock_extended_err.
193  The option modifies field ee_data to pass an id that is unique
194  among all possibly concurrently outstanding timestamp requests for
195  that socket.
196
197SOF_TIMESTAMPING_OPT_ID_TCP:
198  Pass this modifier along with SOF_TIMESTAMPING_OPT_ID for new TCP
199  timestamping applications. SOF_TIMESTAMPING_OPT_ID defines how the
200  counter increments for stream sockets, but its starting point is
201  not entirely trivial. This option fixes that.
202
203  For stream sockets, if SOF_TIMESTAMPING_OPT_ID is set, this should
204  always be set too. On datagram sockets the option has no effect.
205
206  A reasonable expectation is that the counter is reset to zero with
207  the system call, so that a subsequent write() of N bytes generates
208  a timestamp with counter N-1. SOF_TIMESTAMPING_OPT_ID_TCP
209  implements this behavior under all conditions.
210
211  SOF_TIMESTAMPING_OPT_ID without modifier often reports the same,
212  especially when the socket option is set when no data is in
213  transmission. If data is being transmitted, it may be off by the
214  length of the output queue (SIOCOUTQ).
215
216  The difference is due to being based on snd_una versus write_seq.
217  snd_una is the offset in the stream acknowledged by the peer. This
218  depends on factors outside of process control, such as network RTT.
219  write_seq is the last byte written by the process. This offset is
220  not affected by external inputs.
221
222  The difference is subtle and unlikely to be noticed when configured
223  at initial socket creation, when no data is queued or sent. But
224  SOF_TIMESTAMPING_OPT_ID_TCP behavior is more robust regardless of
225  when the socket option is set.
226
227SOF_TIMESTAMPING_OPT_CMSG:
228  Support recv() cmsg for all timestamped packets. Control messages
229  are already supported unconditionally on all packets with receive
230  timestamps and on IPv6 packets with transmit timestamp. This option
231  extends them to IPv4 packets with transmit timestamp. One use case
232  is to correlate packets with their egress device, by enabling socket
233  option IP_PKTINFO simultaneously.
234
235
236SOF_TIMESTAMPING_OPT_TSONLY:
237  Applies to transmit timestamps only. Makes the kernel return the
238  timestamp as a cmsg alongside an empty packet, as opposed to
239  alongside the original packet. This reduces the amount of memory
240  charged to the socket's receive budget (SO_RCVBUF) and delivers
241  the timestamp even if sysctl net.core.tstamp_allow_data is 0.
242  This option disables SOF_TIMESTAMPING_OPT_CMSG.
243
244SOF_TIMESTAMPING_OPT_STATS:
245  Optional stats that are obtained along with the transmit timestamps.
246  It must be used together with SOF_TIMESTAMPING_OPT_TSONLY. When the
247  transmit timestamp is available, the stats are available in a
248  separate control message of type SCM_TIMESTAMPING_OPT_STATS, as a
249  list of TLVs (struct nlattr) of types. These stats allow the
250  application to associate various transport layer stats with
251  the transmit timestamps, such as how long a certain block of
252  data was limited by peer's receiver window.
253
254SOF_TIMESTAMPING_OPT_PKTINFO:
255  Enable the SCM_TIMESTAMPING_PKTINFO control message for incoming
256  packets with hardware timestamps. The message contains struct
257  scm_ts_pktinfo, which supplies the index of the real interface which
258  received the packet and its length at layer 2. A valid (non-zero)
259  interface index will be returned only if CONFIG_NET_RX_BUSY_POLL is
260  enabled and the driver is using NAPI. The struct contains also two
261  other fields, but they are reserved and undefined.
262
263SOF_TIMESTAMPING_OPT_TX_SWHW:
264  Request both hardware and software timestamps for outgoing packets
265  when SOF_TIMESTAMPING_TX_HARDWARE and SOF_TIMESTAMPING_TX_SOFTWARE
266  are enabled at the same time. If both timestamps are generated,
267  two separate messages will be looped to the socket's error queue,
268  each containing just one timestamp.
269
270SOF_TIMESTAMPING_OPT_RX_FILTER:
271  Filter out spurious receive timestamps: report a receive timestamp
272  only if the matching timestamp generation flag is enabled.
273
274  Receive timestamps are generated early in the ingress path, before a
275  packet's destination socket is known. If any socket enables receive
276  timestamps, packets for all socket will receive timestamped packets.
277  Including those that request timestamp reporting with
278  SOF_TIMESTAMPING_SOFTWARE and/or SOF_TIMESTAMPING_RAW_HARDWARE, but
279  do not request receive timestamp generation. This can happen when
280  requesting transmit timestamps only.
281
282  Receiving spurious timestamps is generally benign. A process can
283  ignore the unexpected non-zero value. But it makes behavior subtly
284  dependent on other sockets. This flag isolates the socket for more
285  deterministic behavior.
286
287New applications are encouraged to pass SOF_TIMESTAMPING_OPT_ID to
288disambiguate timestamps and SOF_TIMESTAMPING_OPT_TSONLY to operate
289regardless of the setting of sysctl net.core.tstamp_allow_data.
290
291An exception is when a process needs additional cmsg data, for
292instance SOL_IP/IP_PKTINFO to detect the egress network interface.
293Then pass option SOF_TIMESTAMPING_OPT_CMSG. This option depends on
294having access to the contents of the original packet, so cannot be
295combined with SOF_TIMESTAMPING_OPT_TSONLY.
296
297
2981.3.4. Enabling timestamps via control messages
299^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
300
301In addition to socket options, timestamp generation can be requested
302per write via cmsg, only for SOF_TIMESTAMPING_TX_* (see Section 1.3.1).
303Using this feature, applications can sample timestamps per sendmsg()
304without paying the overhead of enabling and disabling timestamps via
305setsockopt::
306
307  struct msghdr *msg;
308  ...
309  cmsg			       = CMSG_FIRSTHDR(msg);
310  cmsg->cmsg_level	       = SOL_SOCKET;
311  cmsg->cmsg_type	       = SO_TIMESTAMPING;
312  cmsg->cmsg_len	       = CMSG_LEN(sizeof(__u32));
313  *((__u32 *) CMSG_DATA(cmsg)) = SOF_TIMESTAMPING_TX_SCHED |
314				 SOF_TIMESTAMPING_TX_SOFTWARE |
315				 SOF_TIMESTAMPING_TX_ACK;
316  err = sendmsg(fd, msg, 0);
317
318The SOF_TIMESTAMPING_TX_* flags set via cmsg will override
319the SOF_TIMESTAMPING_TX_* flags set via setsockopt.
320
321Moreover, applications must still enable timestamp reporting via
322setsockopt to receive timestamps::
323
324  __u32 val = SOF_TIMESTAMPING_SOFTWARE |
325	      SOF_TIMESTAMPING_OPT_ID /* or any other flag */;
326  err = setsockopt(fd, SOL_SOCKET, SO_TIMESTAMPING, &val, sizeof(val));
327
328
3291.4 Bytestream Timestamps
330-------------------------
331
332The SO_TIMESTAMPING interface supports timestamping of bytes in a
333bytestream. Each request is interpreted as a request for when the
334entire contents of the buffer has passed a timestamping point. That
335is, for streams option SOF_TIMESTAMPING_TX_SOFTWARE will record
336when all bytes have reached the device driver, regardless of how
337many packets the data has been converted into.
338
339In general, bytestreams have no natural delimiters and therefore
340correlating a timestamp with data is non-trivial. A range of bytes
341may be split across segments, any segments may be merged (possibly
342coalescing sections of previously segmented buffers associated with
343independent send() calls). Segments can be reordered and the same
344byte range can coexist in multiple segments for protocols that
345implement retransmissions.
346
347It is essential that all timestamps implement the same semantics,
348regardless of these possible transformations, as otherwise they are
349incomparable. Handling "rare" corner cases differently from the
350simple case (a 1:1 mapping from buffer to skb) is insufficient
351because performance debugging often needs to focus on such outliers.
352
353In practice, timestamps can be correlated with segments of a
354bytestream consistently, if both semantics of the timestamp and the
355timing of measurement are chosen correctly. This challenge is no
356different from deciding on a strategy for IP fragmentation. There, the
357definition is that only the first fragment is timestamped. For
358bytestreams, we chose that a timestamp is generated only when all
359bytes have passed a point. SOF_TIMESTAMPING_TX_ACK as defined is easy to
360implement and reason about. An implementation that has to take into
361account SACK would be more complex due to possible transmission holes
362and out of order arrival.
363
364On the host, TCP can also break the simple 1:1 mapping from buffer to
365skbuff as a result of Nagle, cork, autocork, segmentation and GSO. The
366implementation ensures correctness in all cases by tracking the
367individual last byte passed to send(), even if it is no longer the
368last byte after an skbuff extend or merge operation. It stores the
369relevant sequence number in skb_shinfo(skb)->tskey. Because an skbuff
370has only one such field, only one timestamp can be generated.
371
372In rare cases, a timestamp request can be missed if two requests are
373collapsed onto the same skb. A process can detect this situation by
374enabling SOF_TIMESTAMPING_OPT_ID and comparing the byte offset at
375send time with the value returned for each timestamp. It can prevent
376the situation by always flushing the TCP stack in between requests,
377for instance by enabling TCP_NODELAY and disabling TCP_CORK and
378autocork. After linux-4.7, a better way to prevent coalescing is
379to use MSG_EOR flag at sendmsg() time.
380
381These precautions ensure that the timestamp is generated only when all
382bytes have passed a timestamp point, assuming that the network stack
383itself does not reorder the segments. The stack indeed tries to avoid
384reordering. The one exception is under administrator control: it is
385possible to construct a packet scheduler configuration that delays
386segments from the same stream differently. Such a setup would be
387unusual.
388
389
3902 Data Interfaces
391==================
392
393Timestamps are read using the ancillary data feature of recvmsg().
394See `man 3 cmsg` for details of this interface. The socket manual
395page (`man 7 socket`) describes how timestamps generated with
396SO_TIMESTAMP and SO_TIMESTAMPNS records can be retrieved.
397
398
3992.1 SCM_TIMESTAMPING records
400----------------------------
401
402These timestamps are returned in a control message with cmsg_level
403SOL_SOCKET, cmsg_type SCM_TIMESTAMPING, and payload of type
404
405For SO_TIMESTAMPING_OLD::
406
407	struct scm_timestamping {
408		struct timespec ts[3];
409	};
410
411For SO_TIMESTAMPING_NEW::
412
413	struct scm_timestamping64 {
414		struct __kernel_timespec ts[3];
415
416Always use SO_TIMESTAMPING_NEW timestamp to always get timestamp in
417struct scm_timestamping64 format.
418
419SO_TIMESTAMPING_OLD returns incorrect timestamps after the year 2038
420on 32 bit machines.
421
422The structure can return up to three timestamps. This is a legacy
423feature. At least one field is non-zero at any time. Most timestamps
424are passed in ts[0]. Hardware timestamps are passed in ts[2].
425
426ts[1] used to hold hardware timestamps converted to system time.
427Instead, expose the hardware clock device on the NIC directly as
428a HW PTP clock source, to allow time conversion in userspace and
429optionally synchronize system time with a userspace PTP stack such
430as linuxptp. For the PTP clock API, see Documentation/driver-api/ptp.rst.
431
432Note that if the SO_TIMESTAMP or SO_TIMESTAMPNS option is enabled
433together with SO_TIMESTAMPING using SOF_TIMESTAMPING_SOFTWARE, a false
434software timestamp will be generated in the recvmsg() call and passed
435in ts[0] when a real software timestamp is missing. This happens also
436on hardware transmit timestamps.
437
4382.1.1 Transmit timestamps with MSG_ERRQUEUE
439^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
440
441For transmit timestamps the outgoing packet is looped back to the
442socket's error queue with the send timestamp(s) attached. A process
443receives the timestamps by calling recvmsg() with flag MSG_ERRQUEUE
444set and with a msg_control buffer sufficiently large to receive the
445relevant metadata structures. The recvmsg call returns the original
446outgoing data packet with two ancillary messages attached.
447
448A message of cm_level SOL_IP(V6) and cm_type IP(V6)_RECVERR
449embeds a struct sock_extended_err. This defines the error type. For
450timestamps, the ee_errno field is ENOMSG. The other ancillary message
451will have cm_level SOL_SOCKET and cm_type SCM_TIMESTAMPING. This
452embeds the struct scm_timestamping.
453
454
4552.1.1.2 Timestamp types
456~~~~~~~~~~~~~~~~~~~~~~~
457
458The semantics of the three struct timespec are defined by field
459ee_info in the extended error structure. It contains a value of
460type SCM_TSTAMP_* to define the actual timestamp passed in
461scm_timestamping.
462
463The SCM_TSTAMP_* types are 1:1 matches to the SOF_TIMESTAMPING_*
464control fields discussed previously, with one exception. For legacy
465reasons, SCM_TSTAMP_SND is equal to zero and can be set for both
466SOF_TIMESTAMPING_TX_HARDWARE and SOF_TIMESTAMPING_TX_SOFTWARE. It
467is the first if ts[2] is non-zero, the second otherwise, in which
468case the timestamp is stored in ts[0].
469
470
4712.1.1.3 Fragmentation
472~~~~~~~~~~~~~~~~~~~~~
473
474Fragmentation of outgoing datagrams is rare, but is possible, e.g., by
475explicitly disabling PMTU discovery. If an outgoing packet is fragmented,
476then only the first fragment is timestamped and returned to the sending
477socket.
478
479
4802.1.1.4 Packet Payload
481~~~~~~~~~~~~~~~~~~~~~~
482
483The calling application is often not interested in receiving the whole
484packet payload that it passed to the stack originally: the socket
485error queue mechanism is just a method to piggyback the timestamp on.
486In this case, the application can choose to read datagrams with a
487smaller buffer, possibly even of length 0. The payload is truncated
488accordingly. Until the process calls recvmsg() on the error queue,
489however, the full packet is queued, taking up budget from SO_RCVBUF.
490
491
4922.1.1.5 Blocking Read
493~~~~~~~~~~~~~~~~~~~~~
494
495Reading from the error queue is always a non-blocking operation. To
496block waiting on a timestamp, use poll or select. poll() will return
497POLLERR in pollfd.revents if any data is ready on the error queue.
498There is no need to pass this flag in pollfd.events. This flag is
499ignored on request. See also `man 2 poll`.
500
501
5022.1.2 Receive timestamps
503^^^^^^^^^^^^^^^^^^^^^^^^
504
505On reception, there is no reason to read from the socket error queue.
506The SCM_TIMESTAMPING ancillary data is sent along with the packet data
507on a normal recvmsg(). Since this is not a socket error, it is not
508accompanied by a message SOL_IP(V6)/IP(V6)_RECVERROR. In this case,
509the meaning of the three fields in struct scm_timestamping is
510implicitly defined. ts[0] holds a software timestamp if set, ts[1]
511is again deprecated and ts[2] holds a hardware timestamp if set.
512
513
5143. Hardware Timestamping configuration: SIOCSHWTSTAMP and SIOCGHWTSTAMP
515=======================================================================
516
517Hardware time stamping must also be initialized for each device driver
518that is expected to do hardware time stamping. The parameter is defined in
519include/uapi/linux/net_tstamp.h as::
520
521	struct hwtstamp_config {
522		int flags;	/* no flags defined right now, must be zero */
523		int tx_type;	/* HWTSTAMP_TX_* */
524		int rx_filter;	/* HWTSTAMP_FILTER_* */
525	};
526
527Desired behavior is passed into the kernel and to a specific device by
528calling ioctl(SIOCSHWTSTAMP) with a pointer to a struct ifreq whose
529ifr_data points to a struct hwtstamp_config. The tx_type and
530rx_filter are hints to the driver what it is expected to do. If
531the requested fine-grained filtering for incoming packets is not
532supported, the driver may time stamp more than just the requested types
533of packets.
534
535Drivers are free to use a more permissive configuration than the requested
536configuration. It is expected that drivers should only implement directly the
537most generic mode that can be supported. For example if the hardware can
538support HWTSTAMP_FILTER_PTP_V2_EVENT, then it should generally always upscale
539HWTSTAMP_FILTER_PTP_V2_L2_SYNC, and so forth, as HWTSTAMP_FILTER_PTP_V2_EVENT
540is more generic (and more useful to applications).
541
542A driver which supports hardware time stamping shall update the struct
543with the actual, possibly more permissive configuration. If the
544requested packets cannot be time stamped, then nothing should be
545changed and ERANGE shall be returned (in contrast to EINVAL, which
546indicates that SIOCSHWTSTAMP is not supported at all).
547
548Only a processes with admin rights may change the configuration. User
549space is responsible to ensure that multiple processes don't interfere
550with each other and that the settings are reset.
551
552Any process can read the actual configuration by passing this
553structure to ioctl(SIOCGHWTSTAMP) in the same way.  However, this has
554not been implemented in all drivers.
555
556::
557
558    /* possible values for hwtstamp_config->tx_type */
559    enum {
560	    /*
561	    * no outgoing packet will need hardware time stamping;
562	    * should a packet arrive which asks for it, no hardware
563	    * time stamping will be done
564	    */
565	    HWTSTAMP_TX_OFF,
566
567	    /*
568	    * enables hardware time stamping for outgoing packets;
569	    * the sender of the packet decides which are to be
570	    * time stamped by setting SOF_TIMESTAMPING_TX_SOFTWARE
571	    * before sending the packet
572	    */
573	    HWTSTAMP_TX_ON,
574    };
575
576    /* possible values for hwtstamp_config->rx_filter */
577    enum {
578	    /* time stamp no incoming packet at all */
579	    HWTSTAMP_FILTER_NONE,
580
581	    /* time stamp any incoming packet */
582	    HWTSTAMP_FILTER_ALL,
583
584	    /* return value: time stamp all packets requested plus some others */
585	    HWTSTAMP_FILTER_SOME,
586
587	    /* PTP v1, UDP, any kind of event packet */
588	    HWTSTAMP_FILTER_PTP_V1_L4_EVENT,
589
590	    /* for the complete list of values, please check
591	    * the include file include/uapi/linux/net_tstamp.h
592	    */
593    };
594
5953.1 Hardware Timestamping Implementation: Device Drivers
596--------------------------------------------------------
597
598A driver which supports hardware time stamping must support the
599SIOCSHWTSTAMP ioctl and update the supplied struct hwtstamp_config with
600the actual values as described in the section on SIOCSHWTSTAMP.  It
601should also support SIOCGHWTSTAMP.
602
603Time stamps for received packets must be stored in the skb. To get a pointer
604to the shared time stamp structure of the skb call skb_hwtstamps(). Then
605set the time stamps in the structure::
606
607    struct skb_shared_hwtstamps {
608	    /* hardware time stamp transformed into duration
609	    * since arbitrary point in time
610	    */
611	    ktime_t	hwtstamp;
612    };
613
614Time stamps for outgoing packets are to be generated as follows:
615
616- In hard_start_xmit(), check if (skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)
617  is set no-zero. If yes, then the driver is expected to do hardware time
618  stamping.
619- If this is possible for the skb and requested, then declare
620  that the driver is doing the time stamping by setting the flag
621  SKBTX_IN_PROGRESS in skb_shinfo(skb)->tx_flags , e.g. with::
622
623      skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
624
625  You might want to keep a pointer to the associated skb for the next step
626  and not free the skb. A driver not supporting hardware time stamping doesn't
627  do that. A driver must never touch sk_buff::tstamp! It is used to store
628  software generated time stamps by the network subsystem.
629- Driver should call skb_tx_timestamp() as close to passing sk_buff to hardware
630  as possible. skb_tx_timestamp() provides a software time stamp if requested
631  and hardware timestamping is not possible (SKBTX_IN_PROGRESS not set).
632- As soon as the driver has sent the packet and/or obtained a
633  hardware time stamp for it, it passes the time stamp back by
634  calling skb_tstamp_tx() with the original skb, the raw
635  hardware time stamp. skb_tstamp_tx() clones the original skb and
636  adds the timestamps, therefore the original skb has to be freed now.
637  If obtaining the hardware time stamp somehow fails, then the driver
638  should not fall back to software time stamping. The rationale is that
639  this would occur at a later time in the processing pipeline than other
640  software time stamping and therefore could lead to unexpected deltas
641  between time stamps.
642
6433.2 Special considerations for stacked PTP Hardware Clocks
644----------------------------------------------------------
645
646There are situations when there may be more than one PHC (PTP Hardware Clock)
647in the data path of a packet. The kernel has no explicit mechanism to allow the
648user to select which PHC to use for timestamping Ethernet frames. Instead, the
649assumption is that the outermost PHC is always the most preferable, and that
650kernel drivers collaborate towards achieving that goal. Currently there are 3
651cases of stacked PHCs, detailed below:
652
6533.2.1 DSA (Distributed Switch Architecture) switches
654^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
655
656These are Ethernet switches which have one of their ports connected to an
657(otherwise completely unaware) host Ethernet interface, and perform the role of
658a port multiplier with optional forwarding acceleration features.  Each DSA
659switch port is visible to the user as a standalone (virtual) network interface,
660and its network I/O is performed, under the hood, indirectly through the host
661interface (redirecting to the host port on TX, and intercepting frames on RX).
662
663When a DSA switch is attached to a host port, PTP synchronization has to
664suffer, since the switch's variable queuing delay introduces a path delay
665jitter between the host port and its PTP partner. For this reason, some DSA
666switches include a timestamping clock of their own, and have the ability to
667perform network timestamping on their own MAC, such that path delays only
668measure wire and PHY propagation latencies. Timestamping DSA switches are
669supported in Linux and expose the same ABI as any other network interface (save
670for the fact that the DSA interfaces are in fact virtual in terms of network
671I/O, they do have their own PHC).  It is typical, but not mandatory, for all
672interfaces of a DSA switch to share the same PHC.
673
674By design, PTP timestamping with a DSA switch does not need any special
675handling in the driver for the host port it is attached to.  However, when the
676host port also supports PTP timestamping, DSA will take care of intercepting
677the ``.ndo_eth_ioctl`` calls towards the host port, and block attempts to enable
678hardware timestamping on it. This is because the SO_TIMESTAMPING API does not
679allow the delivery of multiple hardware timestamps for the same packet, so
680anybody else except for the DSA switch port must be prevented from doing so.
681
682In the generic layer, DSA provides the following infrastructure for PTP
683timestamping:
684
685- ``.port_txtstamp()``: a hook called prior to the transmission of
686  packets with a hardware TX timestamping request from user space.
687  This is required for two-step timestamping, since the hardware
688  timestamp becomes available after the actual MAC transmission, so the
689  driver must be prepared to correlate the timestamp with the original
690  packet so that it can re-enqueue the packet back into the socket's
691  error queue. To save the packet for when the timestamp becomes
692  available, the driver can call ``skb_clone_sk`` , save the clone pointer
693  in skb->cb and enqueue a tx skb queue. Typically, a switch will have a
694  PTP TX timestamp register (or sometimes a FIFO) where the timestamp
695  becomes available. In case of a FIFO, the hardware might store
696  key-value pairs of PTP sequence ID/message type/domain number and the
697  actual timestamp. To perform the correlation correctly between the
698  packets in a queue waiting for timestamping and the actual timestamps,
699  drivers can use a BPF classifier (``ptp_classify_raw``) to identify
700  the PTP transport type, and ``ptp_parse_header`` to interpret the PTP
701  header fields. There may be an IRQ that is raised upon this
702  timestamp's availability, or the driver might have to poll after
703  invoking ``dev_queue_xmit()`` towards the host interface.
704  One-step TX timestamping do not require packet cloning, since there is
705  no follow-up message required by the PTP protocol (because the
706  TX timestamp is embedded into the packet by the MAC), and therefore
707  user space does not expect the packet annotated with the TX timestamp
708  to be re-enqueued into its socket's error queue.
709
710- ``.port_rxtstamp()``: On RX, the BPF classifier is run by DSA to
711  identify PTP event messages (any other packets, including PTP general
712  messages, are not timestamped). The original (and only) timestampable
713  skb is provided to the driver, for it to annotate it with a timestamp,
714  if that is immediately available, or defer to later. On reception,
715  timestamps might either be available in-band (through metadata in the
716  DSA header, or attached in other ways to the packet), or out-of-band
717  (through another RX timestamping FIFO). Deferral on RX is typically
718  necessary when retrieving the timestamp needs a sleepable context. In
719  that case, it is the responsibility of the DSA driver to call
720  ``netif_rx()`` on the freshly timestamped skb.
721
7223.2.2 Ethernet PHYs
723^^^^^^^^^^^^^^^^^^^
724
725These are devices that typically fulfill a Layer 1 role in the network stack,
726hence they do not have a representation in terms of a network interface as DSA
727switches do. However, PHYs may be able to detect and timestamp PTP packets, for
728performance reasons: timestamps taken as close as possible to the wire have the
729potential to yield a more stable and precise synchronization.
730
731A PHY driver that supports PTP timestamping must create a ``struct
732mii_timestamper`` and add a pointer to it in ``phydev->mii_ts``. The presence
733of this pointer will be checked by the networking stack.
734
735Since PHYs do not have network interface representations, the timestamping and
736ethtool ioctl operations for them need to be mediated by their respective MAC
737driver.  Therefore, as opposed to DSA switches, modifications need to be done
738to each individual MAC driver for PHY timestamping support. This entails:
739
740- Checking, in ``.ndo_eth_ioctl``, whether ``phy_has_hwtstamp(netdev->phydev)``
741  is true or not. If it is, then the MAC driver should not process this request
742  but instead pass it on to the PHY using ``phy_mii_ioctl()``.
743
744- On RX, special intervention may or may not be needed, depending on the
745  function used to deliver skb's up the network stack. In the case of plain
746  ``netif_rx()`` and similar, MAC drivers must check whether
747  ``skb_defer_rx_timestamp(skb)`` is necessary or not - and if it is, don't
748  call ``netif_rx()`` at all.  If ``CONFIG_NETWORK_PHY_TIMESTAMPING`` is
749  enabled, and ``skb->dev->phydev->mii_ts`` exists, its ``.rxtstamp()`` hook
750  will be called now, to determine, using logic very similar to DSA, whether
751  deferral for RX timestamping is necessary.  Again like DSA, it becomes the
752  responsibility of the PHY driver to send the packet up the stack when the
753  timestamp is available.
754
755  For other skb receive functions, such as ``napi_gro_receive`` and
756  ``netif_receive_skb``, the stack automatically checks whether
757  ``skb_defer_rx_timestamp()`` is necessary, so this check is not needed inside
758  the driver.
759
760- On TX, again, special intervention might or might not be needed.  The
761  function that calls the ``mii_ts->txtstamp()`` hook is named
762  ``skb_clone_tx_timestamp()``. This function can either be called directly
763  (case in which explicit MAC driver support is indeed needed), but the
764  function also piggybacks from the ``skb_tx_timestamp()`` call, which many MAC
765  drivers already perform for software timestamping purposes. Therefore, if a
766  MAC supports software timestamping, it does not need to do anything further
767  at this stage.
768
7693.2.3 MII bus snooping devices
770^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
771
772These perform the same role as timestamping Ethernet PHYs, save for the fact
773that they are discrete devices and can therefore be used in conjunction with
774any PHY even if it doesn't support timestamping. In Linux, they are
775discoverable and attachable to a ``struct phy_device`` through Device Tree, and
776for the rest, they use the same mii_ts infrastructure as those. See
777Documentation/devicetree/bindings/ptp/timestamper.txt for more details.
778
7793.2.4 Other caveats for MAC drivers
780^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
781
782Stacked PHCs, especially DSA (but not only) - since that doesn't require any
783modification to MAC drivers, so it is more difficult to ensure correctness of
784all possible code paths - is that they uncover bugs which were impossible to
785trigger before the existence of stacked PTP clocks.  One example has to do with
786this line of code, already presented earlier::
787
788      skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
789
790Any TX timestamping logic, be it a plain MAC driver, a DSA switch driver, a PHY
791driver or a MII bus snooping device driver, should set this flag.
792But a MAC driver that is unaware of PHC stacking might get tripped up by
793somebody other than itself setting this flag, and deliver a duplicate
794timestamp.
795For example, a typical driver design for TX timestamping might be to split the
796transmission part into 2 portions:
797
7981. "TX": checks whether PTP timestamping has been previously enabled through
799   the ``.ndo_eth_ioctl`` ("``priv->hwtstamp_tx_enabled == true``") and the
800   current skb requires a TX timestamp ("``skb_shinfo(skb)->tx_flags &
801   SKBTX_HW_TSTAMP``"). If this is true, it sets the
802   "``skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS``" flag. Note: as
803   described above, in the case of a stacked PHC system, this condition should
804   never trigger, as this MAC is certainly not the outermost PHC. But this is
805   not where the typical issue is.  Transmission proceeds with this packet.
806
8072. "TX confirmation": Transmission has finished. The driver checks whether it
808   is necessary to collect any TX timestamp for it. Here is where the typical
809   issues are: the MAC driver takes a shortcut and only checks whether
810   "``skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS``" was set. With a stacked
811   PHC system, this is incorrect because this MAC driver is not the only entity
812   in the TX data path who could have enabled SKBTX_IN_PROGRESS in the first
813   place.
814
815The correct solution for this problem is for MAC drivers to have a compound
816check in their "TX confirmation" portion, not only for
817"``skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS``", but also for
818"``priv->hwtstamp_tx_enabled == true``". Because the rest of the system ensures
819that PTP timestamping is not enabled for anything other than the outermost PHC,
820this enhanced check will avoid delivering a duplicated TX timestamp to user
821space.
822