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 when available. 162 163 1641.3.3 Timestamp Options 165^^^^^^^^^^^^^^^^^^^^^^^ 166 167The interface supports the options 168 169SOF_TIMESTAMPING_OPT_ID: 170 Generate a unique identifier along with each packet. A process can 171 have multiple concurrent timestamping requests outstanding. Packets 172 can be reordered in the transmit path, for instance in the packet 173 scheduler. In that case timestamps will be queued onto the error 174 queue out of order from the original send() calls. It is not always 175 possible to uniquely match timestamps to the original send() calls 176 based on timestamp order or payload inspection alone, then. 177 178 This option associates each packet at send() with a unique 179 identifier and returns that along with the timestamp. The identifier 180 is derived from a per-socket u32 counter (that wraps). For datagram 181 sockets, the counter increments with each sent packet. For stream 182 sockets, it increments with every byte. For stream sockets, also set 183 SOF_TIMESTAMPING_OPT_ID_TCP, see the section below. 184 185 The counter starts at zero. It is initialized the first time that 186 the socket option is enabled. It is reset each time the option is 187 enabled after having been disabled. Resetting the counter does not 188 change the identifiers of existing packets in the system. 189 190 This option is implemented only for transmit timestamps. There, the 191 timestamp is always looped along with a struct sock_extended_err. 192 The option modifies field ee_data to pass an id that is unique 193 among all possibly concurrently outstanding timestamp requests for 194 that socket. 195 196SOF_TIMESTAMPING_OPT_ID_TCP: 197 Pass this modifier along with SOF_TIMESTAMPING_OPT_ID for new TCP 198 timestamping applications. SOF_TIMESTAMPING_OPT_ID defines how the 199 counter increments for stream sockets, but its starting point is 200 not entirely trivial. This option fixes that. 201 202 For stream sockets, if SOF_TIMESTAMPING_OPT_ID is set, this should 203 always be set too. On datagram sockets the option has no effect. 204 205 A reasonable expectation is that the counter is reset to zero with 206 the system call, so that a subsequent write() of N bytes generates 207 a timestamp with counter N-1. SOF_TIMESTAMPING_OPT_ID_TCP 208 implements this behavior under all conditions. 209 210 SOF_TIMESTAMPING_OPT_ID without modifier often reports the same, 211 especially when the socket option is set when no data is in 212 transmission. If data is being transmitted, it may be off by the 213 length of the output queue (SIOCOUTQ). 214 215 The difference is due to being based on snd_una versus write_seq. 216 snd_una is the offset in the stream acknowledged by the peer. This 217 depends on factors outside of process control, such as network RTT. 218 write_seq is the last byte written by the process. This offset is 219 not affected by external inputs. 220 221 The difference is subtle and unlikely to be noticed when configured 222 at initial socket creation, when no data is queued or sent. But 223 SOF_TIMESTAMPING_OPT_ID_TCP behavior is more robust regardless of 224 when the socket option is set. 225 226SOF_TIMESTAMPING_OPT_CMSG: 227 Support recv() cmsg for all timestamped packets. Control messages 228 are already supported unconditionally on all packets with receive 229 timestamps and on IPv6 packets with transmit timestamp. This option 230 extends them to IPv4 packets with transmit timestamp. One use case 231 is to correlate packets with their egress device, by enabling socket 232 option IP_PKTINFO simultaneously. 233 234 235SOF_TIMESTAMPING_OPT_TSONLY: 236 Applies to transmit timestamps only. Makes the kernel return the 237 timestamp as a cmsg alongside an empty packet, as opposed to 238 alongside the original packet. This reduces the amount of memory 239 charged to the socket's receive budget (SO_RCVBUF) and delivers 240 the timestamp even if sysctl net.core.tstamp_allow_data is 0. 241 This option disables SOF_TIMESTAMPING_OPT_CMSG. 242 243SOF_TIMESTAMPING_OPT_STATS: 244 Optional stats that are obtained along with the transmit timestamps. 245 It must be used together with SOF_TIMESTAMPING_OPT_TSONLY. When the 246 transmit timestamp is available, the stats are available in a 247 separate control message of type SCM_TIMESTAMPING_OPT_STATS, as a 248 list of TLVs (struct nlattr) of types. These stats allow the 249 application to associate various transport layer stats with 250 the transmit timestamps, such as how long a certain block of 251 data was limited by peer's receiver window. 252 253SOF_TIMESTAMPING_OPT_PKTINFO: 254 Enable the SCM_TIMESTAMPING_PKTINFO control message for incoming 255 packets with hardware timestamps. The message contains struct 256 scm_ts_pktinfo, which supplies the index of the real interface which 257 received the packet and its length at layer 2. A valid (non-zero) 258 interface index will be returned only if CONFIG_NET_RX_BUSY_POLL is 259 enabled and the driver is using NAPI. The struct contains also two 260 other fields, but they are reserved and undefined. 261 262SOF_TIMESTAMPING_OPT_TX_SWHW: 263 Request both hardware and software timestamps for outgoing packets 264 when SOF_TIMESTAMPING_TX_HARDWARE and SOF_TIMESTAMPING_TX_SOFTWARE 265 are enabled at the same time. If both timestamps are generated, 266 two separate messages will be looped to the socket's error queue, 267 each containing just one timestamp. 268 269New applications are encouraged to pass SOF_TIMESTAMPING_OPT_ID to 270disambiguate timestamps and SOF_TIMESTAMPING_OPT_TSONLY to operate 271regardless of the setting of sysctl net.core.tstamp_allow_data. 272 273An exception is when a process needs additional cmsg data, for 274instance SOL_IP/IP_PKTINFO to detect the egress network interface. 275Then pass option SOF_TIMESTAMPING_OPT_CMSG. This option depends on 276having access to the contents of the original packet, so cannot be 277combined with SOF_TIMESTAMPING_OPT_TSONLY. 278 279 2801.3.4. Enabling timestamps via control messages 281^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 282 283In addition to socket options, timestamp generation can be requested 284per write via cmsg, only for SOF_TIMESTAMPING_TX_* (see Section 1.3.1). 285Using this feature, applications can sample timestamps per sendmsg() 286without paying the overhead of enabling and disabling timestamps via 287setsockopt:: 288 289 struct msghdr *msg; 290 ... 291 cmsg = CMSG_FIRSTHDR(msg); 292 cmsg->cmsg_level = SOL_SOCKET; 293 cmsg->cmsg_type = SO_TIMESTAMPING; 294 cmsg->cmsg_len = CMSG_LEN(sizeof(__u32)); 295 *((__u32 *) CMSG_DATA(cmsg)) = SOF_TIMESTAMPING_TX_SCHED | 296 SOF_TIMESTAMPING_TX_SOFTWARE | 297 SOF_TIMESTAMPING_TX_ACK; 298 err = sendmsg(fd, msg, 0); 299 300The SOF_TIMESTAMPING_TX_* flags set via cmsg will override 301the SOF_TIMESTAMPING_TX_* flags set via setsockopt. 302 303Moreover, applications must still enable timestamp reporting via 304setsockopt to receive timestamps:: 305 306 __u32 val = SOF_TIMESTAMPING_SOFTWARE | 307 SOF_TIMESTAMPING_OPT_ID /* or any other flag */; 308 err = setsockopt(fd, SOL_SOCKET, SO_TIMESTAMPING, &val, sizeof(val)); 309 310 3111.4 Bytestream Timestamps 312------------------------- 313 314The SO_TIMESTAMPING interface supports timestamping of bytes in a 315bytestream. Each request is interpreted as a request for when the 316entire contents of the buffer has passed a timestamping point. That 317is, for streams option SOF_TIMESTAMPING_TX_SOFTWARE will record 318when all bytes have reached the device driver, regardless of how 319many packets the data has been converted into. 320 321In general, bytestreams have no natural delimiters and therefore 322correlating a timestamp with data is non-trivial. A range of bytes 323may be split across segments, any segments may be merged (possibly 324coalescing sections of previously segmented buffers associated with 325independent send() calls). Segments can be reordered and the same 326byte range can coexist in multiple segments for protocols that 327implement retransmissions. 328 329It is essential that all timestamps implement the same semantics, 330regardless of these possible transformations, as otherwise they are 331incomparable. Handling "rare" corner cases differently from the 332simple case (a 1:1 mapping from buffer to skb) is insufficient 333because performance debugging often needs to focus on such outliers. 334 335In practice, timestamps can be correlated with segments of a 336bytestream consistently, if both semantics of the timestamp and the 337timing of measurement are chosen correctly. This challenge is no 338different from deciding on a strategy for IP fragmentation. There, the 339definition is that only the first fragment is timestamped. For 340bytestreams, we chose that a timestamp is generated only when all 341bytes have passed a point. SOF_TIMESTAMPING_TX_ACK as defined is easy to 342implement and reason about. An implementation that has to take into 343account SACK would be more complex due to possible transmission holes 344and out of order arrival. 345 346On the host, TCP can also break the simple 1:1 mapping from buffer to 347skbuff as a result of Nagle, cork, autocork, segmentation and GSO. The 348implementation ensures correctness in all cases by tracking the 349individual last byte passed to send(), even if it is no longer the 350last byte after an skbuff extend or merge operation. It stores the 351relevant sequence number in skb_shinfo(skb)->tskey. Because an skbuff 352has only one such field, only one timestamp can be generated. 353 354In rare cases, a timestamp request can be missed if two requests are 355collapsed onto the same skb. A process can detect this situation by 356enabling SOF_TIMESTAMPING_OPT_ID and comparing the byte offset at 357send time with the value returned for each timestamp. It can prevent 358the situation by always flushing the TCP stack in between requests, 359for instance by enabling TCP_NODELAY and disabling TCP_CORK and 360autocork. 361 362These precautions ensure that the timestamp is generated only when all 363bytes have passed a timestamp point, assuming that the network stack 364itself does not reorder the segments. The stack indeed tries to avoid 365reordering. The one exception is under administrator control: it is 366possible to construct a packet scheduler configuration that delays 367segments from the same stream differently. Such a setup would be 368unusual. 369 370 3712 Data Interfaces 372================== 373 374Timestamps are read using the ancillary data feature of recvmsg(). 375See `man 3 cmsg` for details of this interface. The socket manual 376page (`man 7 socket`) describes how timestamps generated with 377SO_TIMESTAMP and SO_TIMESTAMPNS records can be retrieved. 378 379 3802.1 SCM_TIMESTAMPING records 381---------------------------- 382 383These timestamps are returned in a control message with cmsg_level 384SOL_SOCKET, cmsg_type SCM_TIMESTAMPING, and payload of type 385 386For SO_TIMESTAMPING_OLD:: 387 388 struct scm_timestamping { 389 struct timespec ts[3]; 390 }; 391 392For SO_TIMESTAMPING_NEW:: 393 394 struct scm_timestamping64 { 395 struct __kernel_timespec ts[3]; 396 397Always use SO_TIMESTAMPING_NEW timestamp to always get timestamp in 398struct scm_timestamping64 format. 399 400SO_TIMESTAMPING_OLD returns incorrect timestamps after the year 2038 401on 32 bit machines. 402 403The structure can return up to three timestamps. This is a legacy 404feature. At least one field is non-zero at any time. Most timestamps 405are passed in ts[0]. Hardware timestamps are passed in ts[2]. 406 407ts[1] used to hold hardware timestamps converted to system time. 408Instead, expose the hardware clock device on the NIC directly as 409a HW PTP clock source, to allow time conversion in userspace and 410optionally synchronize system time with a userspace PTP stack such 411as linuxptp. For the PTP clock API, see Documentation/driver-api/ptp.rst. 412 413Note that if the SO_TIMESTAMP or SO_TIMESTAMPNS option is enabled 414together with SO_TIMESTAMPING using SOF_TIMESTAMPING_SOFTWARE, a false 415software timestamp will be generated in the recvmsg() call and passed 416in ts[0] when a real software timestamp is missing. This happens also 417on hardware transmit timestamps. 418 4192.1.1 Transmit timestamps with MSG_ERRQUEUE 420^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 421 422For transmit timestamps the outgoing packet is looped back to the 423socket's error queue with the send timestamp(s) attached. A process 424receives the timestamps by calling recvmsg() with flag MSG_ERRQUEUE 425set and with a msg_control buffer sufficiently large to receive the 426relevant metadata structures. The recvmsg call returns the original 427outgoing data packet with two ancillary messages attached. 428 429A message of cm_level SOL_IP(V6) and cm_type IP(V6)_RECVERR 430embeds a struct sock_extended_err. This defines the error type. For 431timestamps, the ee_errno field is ENOMSG. The other ancillary message 432will have cm_level SOL_SOCKET and cm_type SCM_TIMESTAMPING. This 433embeds the struct scm_timestamping. 434 435 4362.1.1.2 Timestamp types 437~~~~~~~~~~~~~~~~~~~~~~~ 438 439The semantics of the three struct timespec are defined by field 440ee_info in the extended error structure. It contains a value of 441type SCM_TSTAMP_* to define the actual timestamp passed in 442scm_timestamping. 443 444The SCM_TSTAMP_* types are 1:1 matches to the SOF_TIMESTAMPING_* 445control fields discussed previously, with one exception. For legacy 446reasons, SCM_TSTAMP_SND is equal to zero and can be set for both 447SOF_TIMESTAMPING_TX_HARDWARE and SOF_TIMESTAMPING_TX_SOFTWARE. It 448is the first if ts[2] is non-zero, the second otherwise, in which 449case the timestamp is stored in ts[0]. 450 451 4522.1.1.3 Fragmentation 453~~~~~~~~~~~~~~~~~~~~~ 454 455Fragmentation of outgoing datagrams is rare, but is possible, e.g., by 456explicitly disabling PMTU discovery. If an outgoing packet is fragmented, 457then only the first fragment is timestamped and returned to the sending 458socket. 459 460 4612.1.1.4 Packet Payload 462~~~~~~~~~~~~~~~~~~~~~~ 463 464The calling application is often not interested in receiving the whole 465packet payload that it passed to the stack originally: the socket 466error queue mechanism is just a method to piggyback the timestamp on. 467In this case, the application can choose to read datagrams with a 468smaller buffer, possibly even of length 0. The payload is truncated 469accordingly. Until the process calls recvmsg() on the error queue, 470however, the full packet is queued, taking up budget from SO_RCVBUF. 471 472 4732.1.1.5 Blocking Read 474~~~~~~~~~~~~~~~~~~~~~ 475 476Reading from the error queue is always a non-blocking operation. To 477block waiting on a timestamp, use poll or select. poll() will return 478POLLERR in pollfd.revents if any data is ready on the error queue. 479There is no need to pass this flag in pollfd.events. This flag is 480ignored on request. See also `man 2 poll`. 481 482 4832.1.2 Receive timestamps 484^^^^^^^^^^^^^^^^^^^^^^^^ 485 486On reception, there is no reason to read from the socket error queue. 487The SCM_TIMESTAMPING ancillary data is sent along with the packet data 488on a normal recvmsg(). Since this is not a socket error, it is not 489accompanied by a message SOL_IP(V6)/IP(V6)_RECVERROR. In this case, 490the meaning of the three fields in struct scm_timestamping is 491implicitly defined. ts[0] holds a software timestamp if set, ts[1] 492is again deprecated and ts[2] holds a hardware timestamp if set. 493 494 4953. Hardware Timestamping configuration: SIOCSHWTSTAMP and SIOCGHWTSTAMP 496======================================================================= 497 498Hardware time stamping must also be initialized for each device driver 499that is expected to do hardware time stamping. The parameter is defined in 500include/uapi/linux/net_tstamp.h as:: 501 502 struct hwtstamp_config { 503 int flags; /* no flags defined right now, must be zero */ 504 int tx_type; /* HWTSTAMP_TX_* */ 505 int rx_filter; /* HWTSTAMP_FILTER_* */ 506 }; 507 508Desired behavior is passed into the kernel and to a specific device by 509calling ioctl(SIOCSHWTSTAMP) with a pointer to a struct ifreq whose 510ifr_data points to a struct hwtstamp_config. The tx_type and 511rx_filter are hints to the driver what it is expected to do. If 512the requested fine-grained filtering for incoming packets is not 513supported, the driver may time stamp more than just the requested types 514of packets. 515 516Drivers are free to use a more permissive configuration than the requested 517configuration. It is expected that drivers should only implement directly the 518most generic mode that can be supported. For example if the hardware can 519support HWTSTAMP_FILTER_PTP_V2_EVENT, then it should generally always upscale 520HWTSTAMP_FILTER_PTP_V2_L2_SYNC, and so forth, as HWTSTAMP_FILTER_PTP_V2_EVENT 521is more generic (and more useful to applications). 522 523A driver which supports hardware time stamping shall update the struct 524with the actual, possibly more permissive configuration. If the 525requested packets cannot be time stamped, then nothing should be 526changed and ERANGE shall be returned (in contrast to EINVAL, which 527indicates that SIOCSHWTSTAMP is not supported at all). 528 529Only a processes with admin rights may change the configuration. User 530space is responsible to ensure that multiple processes don't interfere 531with each other and that the settings are reset. 532 533Any process can read the actual configuration by passing this 534structure to ioctl(SIOCGHWTSTAMP) in the same way. However, this has 535not been implemented in all drivers. 536 537:: 538 539 /* possible values for hwtstamp_config->tx_type */ 540 enum { 541 /* 542 * no outgoing packet will need hardware time stamping; 543 * should a packet arrive which asks for it, no hardware 544 * time stamping will be done 545 */ 546 HWTSTAMP_TX_OFF, 547 548 /* 549 * enables hardware time stamping for outgoing packets; 550 * the sender of the packet decides which are to be 551 * time stamped by setting SOF_TIMESTAMPING_TX_SOFTWARE 552 * before sending the packet 553 */ 554 HWTSTAMP_TX_ON, 555 }; 556 557 /* possible values for hwtstamp_config->rx_filter */ 558 enum { 559 /* time stamp no incoming packet at all */ 560 HWTSTAMP_FILTER_NONE, 561 562 /* time stamp any incoming packet */ 563 HWTSTAMP_FILTER_ALL, 564 565 /* return value: time stamp all packets requested plus some others */ 566 HWTSTAMP_FILTER_SOME, 567 568 /* PTP v1, UDP, any kind of event packet */ 569 HWTSTAMP_FILTER_PTP_V1_L4_EVENT, 570 571 /* for the complete list of values, please check 572 * the include file include/uapi/linux/net_tstamp.h 573 */ 574 }; 575 5763.1 Hardware Timestamping Implementation: Device Drivers 577-------------------------------------------------------- 578 579A driver which supports hardware time stamping must support the 580SIOCSHWTSTAMP ioctl and update the supplied struct hwtstamp_config with 581the actual values as described in the section on SIOCSHWTSTAMP. It 582should also support SIOCGHWTSTAMP. 583 584Time stamps for received packets must be stored in the skb. To get a pointer 585to the shared time stamp structure of the skb call skb_hwtstamps(). Then 586set the time stamps in the structure:: 587 588 struct skb_shared_hwtstamps { 589 /* hardware time stamp transformed into duration 590 * since arbitrary point in time 591 */ 592 ktime_t hwtstamp; 593 }; 594 595Time stamps for outgoing packets are to be generated as follows: 596 597- In hard_start_xmit(), check if (skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) 598 is set no-zero. If yes, then the driver is expected to do hardware time 599 stamping. 600- If this is possible for the skb and requested, then declare 601 that the driver is doing the time stamping by setting the flag 602 SKBTX_IN_PROGRESS in skb_shinfo(skb)->tx_flags , e.g. with:: 603 604 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS; 605 606 You might want to keep a pointer to the associated skb for the next step 607 and not free the skb. A driver not supporting hardware time stamping doesn't 608 do that. A driver must never touch sk_buff::tstamp! It is used to store 609 software generated time stamps by the network subsystem. 610- Driver should call skb_tx_timestamp() as close to passing sk_buff to hardware 611 as possible. skb_tx_timestamp() provides a software time stamp if requested 612 and hardware timestamping is not possible (SKBTX_IN_PROGRESS not set). 613- As soon as the driver has sent the packet and/or obtained a 614 hardware time stamp for it, it passes the time stamp back by 615 calling skb_tstamp_tx() with the original skb, the raw 616 hardware time stamp. skb_tstamp_tx() clones the original skb and 617 adds the timestamps, therefore the original skb has to be freed now. 618 If obtaining the hardware time stamp somehow fails, then the driver 619 should not fall back to software time stamping. The rationale is that 620 this would occur at a later time in the processing pipeline than other 621 software time stamping and therefore could lead to unexpected deltas 622 between time stamps. 623 6243.2 Special considerations for stacked PTP Hardware Clocks 625---------------------------------------------------------- 626 627There are situations when there may be more than one PHC (PTP Hardware Clock) 628in the data path of a packet. The kernel has no explicit mechanism to allow the 629user to select which PHC to use for timestamping Ethernet frames. Instead, the 630assumption is that the outermost PHC is always the most preferable, and that 631kernel drivers collaborate towards achieving that goal. Currently there are 3 632cases of stacked PHCs, detailed below: 633 6343.2.1 DSA (Distributed Switch Architecture) switches 635^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 636 637These are Ethernet switches which have one of their ports connected to an 638(otherwise completely unaware) host Ethernet interface, and perform the role of 639a port multiplier with optional forwarding acceleration features. Each DSA 640switch port is visible to the user as a standalone (virtual) network interface, 641and its network I/O is performed, under the hood, indirectly through the host 642interface (redirecting to the host port on TX, and intercepting frames on RX). 643 644When a DSA switch is attached to a host port, PTP synchronization has to 645suffer, since the switch's variable queuing delay introduces a path delay 646jitter between the host port and its PTP partner. For this reason, some DSA 647switches include a timestamping clock of their own, and have the ability to 648perform network timestamping on their own MAC, such that path delays only 649measure wire and PHY propagation latencies. Timestamping DSA switches are 650supported in Linux and expose the same ABI as any other network interface (save 651for the fact that the DSA interfaces are in fact virtual in terms of network 652I/O, they do have their own PHC). It is typical, but not mandatory, for all 653interfaces of a DSA switch to share the same PHC. 654 655By design, PTP timestamping with a DSA switch does not need any special 656handling in the driver for the host port it is attached to. However, when the 657host port also supports PTP timestamping, DSA will take care of intercepting 658the ``.ndo_eth_ioctl`` calls towards the host port, and block attempts to enable 659hardware timestamping on it. This is because the SO_TIMESTAMPING API does not 660allow the delivery of multiple hardware timestamps for the same packet, so 661anybody else except for the DSA switch port must be prevented from doing so. 662 663In the generic layer, DSA provides the following infrastructure for PTP 664timestamping: 665 666- ``.port_txtstamp()``: a hook called prior to the transmission of 667 packets with a hardware TX timestamping request from user space. 668 This is required for two-step timestamping, since the hardware 669 timestamp becomes available after the actual MAC transmission, so the 670 driver must be prepared to correlate the timestamp with the original 671 packet so that it can re-enqueue the packet back into the socket's 672 error queue. To save the packet for when the timestamp becomes 673 available, the driver can call ``skb_clone_sk`` , save the clone pointer 674 in skb->cb and enqueue a tx skb queue. Typically, a switch will have a 675 PTP TX timestamp register (or sometimes a FIFO) where the timestamp 676 becomes available. In case of a FIFO, the hardware might store 677 key-value pairs of PTP sequence ID/message type/domain number and the 678 actual timestamp. To perform the correlation correctly between the 679 packets in a queue waiting for timestamping and the actual timestamps, 680 drivers can use a BPF classifier (``ptp_classify_raw``) to identify 681 the PTP transport type, and ``ptp_parse_header`` to interpret the PTP 682 header fields. There may be an IRQ that is raised upon this 683 timestamp's availability, or the driver might have to poll after 684 invoking ``dev_queue_xmit()`` towards the host interface. 685 One-step TX timestamping do not require packet cloning, since there is 686 no follow-up message required by the PTP protocol (because the 687 TX timestamp is embedded into the packet by the MAC), and therefore 688 user space does not expect the packet annotated with the TX timestamp 689 to be re-enqueued into its socket's error queue. 690 691- ``.port_rxtstamp()``: On RX, the BPF classifier is run by DSA to 692 identify PTP event messages (any other packets, including PTP general 693 messages, are not timestamped). The original (and only) timestampable 694 skb is provided to the driver, for it to annotate it with a timestamp, 695 if that is immediately available, or defer to later. On reception, 696 timestamps might either be available in-band (through metadata in the 697 DSA header, or attached in other ways to the packet), or out-of-band 698 (through another RX timestamping FIFO). Deferral on RX is typically 699 necessary when retrieving the timestamp needs a sleepable context. In 700 that case, it is the responsibility of the DSA driver to call 701 ``netif_rx()`` on the freshly timestamped skb. 702 7033.2.2 Ethernet PHYs 704^^^^^^^^^^^^^^^^^^^ 705 706These are devices that typically fulfill a Layer 1 role in the network stack, 707hence they do not have a representation in terms of a network interface as DSA 708switches do. However, PHYs may be able to detect and timestamp PTP packets, for 709performance reasons: timestamps taken as close as possible to the wire have the 710potential to yield a more stable and precise synchronization. 711 712A PHY driver that supports PTP timestamping must create a ``struct 713mii_timestamper`` and add a pointer to it in ``phydev->mii_ts``. The presence 714of this pointer will be checked by the networking stack. 715 716Since PHYs do not have network interface representations, the timestamping and 717ethtool ioctl operations for them need to be mediated by their respective MAC 718driver. Therefore, as opposed to DSA switches, modifications need to be done 719to each individual MAC driver for PHY timestamping support. This entails: 720 721- Checking, in ``.ndo_eth_ioctl``, whether ``phy_has_hwtstamp(netdev->phydev)`` 722 is true or not. If it is, then the MAC driver should not process this request 723 but instead pass it on to the PHY using ``phy_mii_ioctl()``. 724 725- On RX, special intervention may or may not be needed, depending on the 726 function used to deliver skb's up the network stack. In the case of plain 727 ``netif_rx()`` and similar, MAC drivers must check whether 728 ``skb_defer_rx_timestamp(skb)`` is necessary or not - and if it is, don't 729 call ``netif_rx()`` at all. If ``CONFIG_NETWORK_PHY_TIMESTAMPING`` is 730 enabled, and ``skb->dev->phydev->mii_ts`` exists, its ``.rxtstamp()`` hook 731 will be called now, to determine, using logic very similar to DSA, whether 732 deferral for RX timestamping is necessary. Again like DSA, it becomes the 733 responsibility of the PHY driver to send the packet up the stack when the 734 timestamp is available. 735 736 For other skb receive functions, such as ``napi_gro_receive`` and 737 ``netif_receive_skb``, the stack automatically checks whether 738 ``skb_defer_rx_timestamp()`` is necessary, so this check is not needed inside 739 the driver. 740 741- On TX, again, special intervention might or might not be needed. The 742 function that calls the ``mii_ts->txtstamp()`` hook is named 743 ``skb_clone_tx_timestamp()``. This function can either be called directly 744 (case in which explicit MAC driver support is indeed needed), but the 745 function also piggybacks from the ``skb_tx_timestamp()`` call, which many MAC 746 drivers already perform for software timestamping purposes. Therefore, if a 747 MAC supports software timestamping, it does not need to do anything further 748 at this stage. 749 7503.2.3 MII bus snooping devices 751^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 752 753These perform the same role as timestamping Ethernet PHYs, save for the fact 754that they are discrete devices and can therefore be used in conjunction with 755any PHY even if it doesn't support timestamping. In Linux, they are 756discoverable and attachable to a ``struct phy_device`` through Device Tree, and 757for the rest, they use the same mii_ts infrastructure as those. See 758Documentation/devicetree/bindings/ptp/timestamper.txt for more details. 759 7603.2.4 Other caveats for MAC drivers 761^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 762 763Stacked PHCs, especially DSA (but not only) - since that doesn't require any 764modification to MAC drivers, so it is more difficult to ensure correctness of 765all possible code paths - is that they uncover bugs which were impossible to 766trigger before the existence of stacked PTP clocks. One example has to do with 767this line of code, already presented earlier:: 768 769 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS; 770 771Any TX timestamping logic, be it a plain MAC driver, a DSA switch driver, a PHY 772driver or a MII bus snooping device driver, should set this flag. 773But a MAC driver that is unaware of PHC stacking might get tripped up by 774somebody other than itself setting this flag, and deliver a duplicate 775timestamp. 776For example, a typical driver design for TX timestamping might be to split the 777transmission part into 2 portions: 778 7791. "TX": checks whether PTP timestamping has been previously enabled through 780 the ``.ndo_eth_ioctl`` ("``priv->hwtstamp_tx_enabled == true``") and the 781 current skb requires a TX timestamp ("``skb_shinfo(skb)->tx_flags & 782 SKBTX_HW_TSTAMP``"). If this is true, it sets the 783 "``skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS``" flag. Note: as 784 described above, in the case of a stacked PHC system, this condition should 785 never trigger, as this MAC is certainly not the outermost PHC. But this is 786 not where the typical issue is. Transmission proceeds with this packet. 787 7882. "TX confirmation": Transmission has finished. The driver checks whether it 789 is necessary to collect any TX timestamp for it. Here is where the typical 790 issues are: the MAC driver takes a shortcut and only checks whether 791 "``skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS``" was set. With a stacked 792 PHC system, this is incorrect because this MAC driver is not the only entity 793 in the TX data path who could have enabled SKBTX_IN_PROGRESS in the first 794 place. 795 796The correct solution for this problem is for MAC drivers to have a compound 797check in their "TX confirmation" portion, not only for 798"``skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS``", but also for 799"``priv->hwtstamp_tx_enabled == true``". Because the rest of the system ensures 800that PTP timestamping is not enabled for anything other than the outermost PHC, 801this enhanced check will avoid delivering a duplicated TX timestamp to user 802space. 803
