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. 196 197 The process can optionally override the default generated ID, by 198 passing a specific ID with control message SCM_TS_OPT_ID (not 199 supported for TCP sockets):: 200 201 struct msghdr *msg; 202 ... 203 cmsg = CMSG_FIRSTHDR(msg); 204 cmsg->cmsg_level = SOL_SOCKET; 205 cmsg->cmsg_type = SCM_TS_OPT_ID; 206 cmsg->cmsg_len = CMSG_LEN(sizeof(__u32)); 207 *((__u32 *) CMSG_DATA(cmsg)) = opt_id; 208 err = sendmsg(fd, msg, 0); 209 210 211SOF_TIMESTAMPING_OPT_ID_TCP: 212 Pass this modifier along with SOF_TIMESTAMPING_OPT_ID for new TCP 213 timestamping applications. SOF_TIMESTAMPING_OPT_ID defines how the 214 counter increments for stream sockets, but its starting point is 215 not entirely trivial. This option fixes that. 216 217 For stream sockets, if SOF_TIMESTAMPING_OPT_ID is set, this should 218 always be set too. On datagram sockets the option has no effect. 219 220 A reasonable expectation is that the counter is reset to zero with 221 the system call, so that a subsequent write() of N bytes generates 222 a timestamp with counter N-1. SOF_TIMESTAMPING_OPT_ID_TCP 223 implements this behavior under all conditions. 224 225 SOF_TIMESTAMPING_OPT_ID without modifier often reports the same, 226 especially when the socket option is set when no data is in 227 transmission. If data is being transmitted, it may be off by the 228 length of the output queue (SIOCOUTQ). 229 230 The difference is due to being based on snd_una versus write_seq. 231 snd_una is the offset in the stream acknowledged by the peer. This 232 depends on factors outside of process control, such as network RTT. 233 write_seq is the last byte written by the process. This offset is 234 not affected by external inputs. 235 236 The difference is subtle and unlikely to be noticed when configured 237 at initial socket creation, when no data is queued or sent. But 238 SOF_TIMESTAMPING_OPT_ID_TCP behavior is more robust regardless of 239 when the socket option is set. 240 241SOF_TIMESTAMPING_OPT_CMSG: 242 Support recv() cmsg for all timestamped packets. Control messages 243 are already supported unconditionally on all packets with receive 244 timestamps and on IPv6 packets with transmit timestamp. This option 245 extends them to IPv4 packets with transmit timestamp. One use case 246 is to correlate packets with their egress device, by enabling socket 247 option IP_PKTINFO simultaneously. 248 249 250SOF_TIMESTAMPING_OPT_TSONLY: 251 Applies to transmit timestamps only. Makes the kernel return the 252 timestamp as a cmsg alongside an empty packet, as opposed to 253 alongside the original packet. This reduces the amount of memory 254 charged to the socket's receive budget (SO_RCVBUF) and delivers 255 the timestamp even if sysctl net.core.tstamp_allow_data is 0. 256 This option disables SOF_TIMESTAMPING_OPT_CMSG. 257 258SOF_TIMESTAMPING_OPT_STATS: 259 Optional stats that are obtained along with the transmit timestamps. 260 It must be used together with SOF_TIMESTAMPING_OPT_TSONLY. When the 261 transmit timestamp is available, the stats are available in a 262 separate control message of type SCM_TIMESTAMPING_OPT_STATS, as a 263 list of TLVs (struct nlattr) of types. These stats allow the 264 application to associate various transport layer stats with 265 the transmit timestamps, such as how long a certain block of 266 data was limited by peer's receiver window. 267 268SOF_TIMESTAMPING_OPT_PKTINFO: 269 Enable the SCM_TIMESTAMPING_PKTINFO control message for incoming 270 packets with hardware timestamps. The message contains struct 271 scm_ts_pktinfo, which supplies the index of the real interface which 272 received the packet and its length at layer 2. A valid (non-zero) 273 interface index will be returned only if CONFIG_NET_RX_BUSY_POLL is 274 enabled and the driver is using NAPI. The struct contains also two 275 other fields, but they are reserved and undefined. 276 277SOF_TIMESTAMPING_OPT_TX_SWHW: 278 Request both hardware and software timestamps for outgoing packets 279 when SOF_TIMESTAMPING_TX_HARDWARE and SOF_TIMESTAMPING_TX_SOFTWARE 280 are enabled at the same time. If both timestamps are generated, 281 two separate messages will be looped to the socket's error queue, 282 each containing just one timestamp. 283 284SOF_TIMESTAMPING_OPT_RX_FILTER: 285 Filter out spurious receive timestamps: report a receive timestamp 286 only if the matching timestamp generation flag is enabled. 287 288 Receive timestamps are generated early in the ingress path, before a 289 packet's destination socket is known. If any socket enables receive 290 timestamps, packets for all socket will receive timestamped packets. 291 Including those that request timestamp reporting with 292 SOF_TIMESTAMPING_SOFTWARE and/or SOF_TIMESTAMPING_RAW_HARDWARE, but 293 do not request receive timestamp generation. This can happen when 294 requesting transmit timestamps only. 295 296 Receiving spurious timestamps is generally benign. A process can 297 ignore the unexpected non-zero value. But it makes behavior subtly 298 dependent on other sockets. This flag isolates the socket for more 299 deterministic behavior. 300 301New applications are encouraged to pass SOF_TIMESTAMPING_OPT_ID to 302disambiguate timestamps and SOF_TIMESTAMPING_OPT_TSONLY to operate 303regardless of the setting of sysctl net.core.tstamp_allow_data. 304 305An exception is when a process needs additional cmsg data, for 306instance SOL_IP/IP_PKTINFO to detect the egress network interface. 307Then pass option SOF_TIMESTAMPING_OPT_CMSG. This option depends on 308having access to the contents of the original packet, so cannot be 309combined with SOF_TIMESTAMPING_OPT_TSONLY. 310 311 3121.3.4. Enabling timestamps via control messages 313^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 314 315In addition to socket options, timestamp generation can be requested 316per write via cmsg, only for SOF_TIMESTAMPING_TX_* (see Section 1.3.1). 317Using this feature, applications can sample timestamps per sendmsg() 318without paying the overhead of enabling and disabling timestamps via 319setsockopt:: 320 321 struct msghdr *msg; 322 ... 323 cmsg = CMSG_FIRSTHDR(msg); 324 cmsg->cmsg_level = SOL_SOCKET; 325 cmsg->cmsg_type = SO_TIMESTAMPING; 326 cmsg->cmsg_len = CMSG_LEN(sizeof(__u32)); 327 *((__u32 *) CMSG_DATA(cmsg)) = SOF_TIMESTAMPING_TX_SCHED | 328 SOF_TIMESTAMPING_TX_SOFTWARE | 329 SOF_TIMESTAMPING_TX_ACK; 330 err = sendmsg(fd, msg, 0); 331 332The SOF_TIMESTAMPING_TX_* flags set via cmsg will override 333the SOF_TIMESTAMPING_TX_* flags set via setsockopt. 334 335Moreover, applications must still enable timestamp reporting via 336setsockopt to receive timestamps:: 337 338 __u32 val = SOF_TIMESTAMPING_SOFTWARE | 339 SOF_TIMESTAMPING_OPT_ID /* or any other flag */; 340 err = setsockopt(fd, SOL_SOCKET, SO_TIMESTAMPING, &val, sizeof(val)); 341 342 3431.4 Bytestream Timestamps 344------------------------- 345 346The SO_TIMESTAMPING interface supports timestamping of bytes in a 347bytestream. Each request is interpreted as a request for when the 348entire contents of the buffer has passed a timestamping point. That 349is, for streams option SOF_TIMESTAMPING_TX_SOFTWARE will record 350when all bytes have reached the device driver, regardless of how 351many packets the data has been converted into. 352 353In general, bytestreams have no natural delimiters and therefore 354correlating a timestamp with data is non-trivial. A range of bytes 355may be split across segments, any segments may be merged (possibly 356coalescing sections of previously segmented buffers associated with 357independent send() calls). Segments can be reordered and the same 358byte range can coexist in multiple segments for protocols that 359implement retransmissions. 360 361It is essential that all timestamps implement the same semantics, 362regardless of these possible transformations, as otherwise they are 363incomparable. Handling "rare" corner cases differently from the 364simple case (a 1:1 mapping from buffer to skb) is insufficient 365because performance debugging often needs to focus on such outliers. 366 367In practice, timestamps can be correlated with segments of a 368bytestream consistently, if both semantics of the timestamp and the 369timing of measurement are chosen correctly. This challenge is no 370different from deciding on a strategy for IP fragmentation. There, the 371definition is that only the first fragment is timestamped. For 372bytestreams, we chose that a timestamp is generated only when all 373bytes have passed a point. SOF_TIMESTAMPING_TX_ACK as defined is easy to 374implement and reason about. An implementation that has to take into 375account SACK would be more complex due to possible transmission holes 376and out of order arrival. 377 378On the host, TCP can also break the simple 1:1 mapping from buffer to 379skbuff as a result of Nagle, cork, autocork, segmentation and GSO. The 380implementation ensures correctness in all cases by tracking the 381individual last byte passed to send(), even if it is no longer the 382last byte after an skbuff extend or merge operation. It stores the 383relevant sequence number in skb_shinfo(skb)->tskey. Because an skbuff 384has only one such field, only one timestamp can be generated. 385 386In rare cases, a timestamp request can be missed if two requests are 387collapsed onto the same skb. A process can detect this situation by 388enabling SOF_TIMESTAMPING_OPT_ID and comparing the byte offset at 389send time with the value returned for each timestamp. It can prevent 390the situation by always flushing the TCP stack in between requests, 391for instance by enabling TCP_NODELAY and disabling TCP_CORK and 392autocork. After linux-4.7, a better way to prevent coalescing is 393to use MSG_EOR flag at sendmsg() time. 394 395These precautions ensure that the timestamp is generated only when all 396bytes have passed a timestamp point, assuming that the network stack 397itself does not reorder the segments. The stack indeed tries to avoid 398reordering. The one exception is under administrator control: it is 399possible to construct a packet scheduler configuration that delays 400segments from the same stream differently. Such a setup would be 401unusual. 402 403 4042 Data Interfaces 405================== 406 407Timestamps are read using the ancillary data feature of recvmsg(). 408See `man 3 cmsg` for details of this interface. The socket manual 409page (`man 7 socket`) describes how timestamps generated with 410SO_TIMESTAMP and SO_TIMESTAMPNS records can be retrieved. 411 412 4132.1 SCM_TIMESTAMPING records 414---------------------------- 415 416These timestamps are returned in a control message with cmsg_level 417SOL_SOCKET, cmsg_type SCM_TIMESTAMPING, and payload of type 418 419For SO_TIMESTAMPING_OLD:: 420 421 struct scm_timestamping { 422 struct timespec ts[3]; 423 }; 424 425For SO_TIMESTAMPING_NEW:: 426 427 struct scm_timestamping64 { 428 struct __kernel_timespec ts[3]; 429 430Always use SO_TIMESTAMPING_NEW timestamp to always get timestamp in 431struct scm_timestamping64 format. 432 433SO_TIMESTAMPING_OLD returns incorrect timestamps after the year 2038 434on 32 bit machines. 435 436The structure can return up to three timestamps. This is a legacy 437feature. At least one field is non-zero at any time. Most timestamps 438are passed in ts[0]. Hardware timestamps are passed in ts[2]. 439 440ts[1] used to hold hardware timestamps converted to system time. 441Instead, expose the hardware clock device on the NIC directly as 442a HW PTP clock source, to allow time conversion in userspace and 443optionally synchronize system time with a userspace PTP stack such 444as linuxptp. For the PTP clock API, see Documentation/driver-api/ptp.rst. 445 446Note that if the SO_TIMESTAMP or SO_TIMESTAMPNS option is enabled 447together with SO_TIMESTAMPING using SOF_TIMESTAMPING_SOFTWARE, a false 448software timestamp will be generated in the recvmsg() call and passed 449in ts[0] when a real software timestamp is missing. This happens also 450on hardware transmit timestamps. 451 4522.1.1 Transmit timestamps with MSG_ERRQUEUE 453^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 454 455For transmit timestamps the outgoing packet is looped back to the 456socket's error queue with the send timestamp(s) attached. A process 457receives the timestamps by calling recvmsg() with flag MSG_ERRQUEUE 458set and with a msg_control buffer sufficiently large to receive the 459relevant metadata structures. The recvmsg call returns the original 460outgoing data packet with two ancillary messages attached. 461 462A message of cm_level SOL_IP(V6) and cm_type IP(V6)_RECVERR 463embeds a struct sock_extended_err. This defines the error type. For 464timestamps, the ee_errno field is ENOMSG. The other ancillary message 465will have cm_level SOL_SOCKET and cm_type SCM_TIMESTAMPING. This 466embeds the struct scm_timestamping. 467 468 4692.1.1.2 Timestamp types 470~~~~~~~~~~~~~~~~~~~~~~~ 471 472The semantics of the three struct timespec are defined by field 473ee_info in the extended error structure. It contains a value of 474type SCM_TSTAMP_* to define the actual timestamp passed in 475scm_timestamping. 476 477The SCM_TSTAMP_* types are 1:1 matches to the SOF_TIMESTAMPING_* 478control fields discussed previously, with one exception. For legacy 479reasons, SCM_TSTAMP_SND is equal to zero and can be set for both 480SOF_TIMESTAMPING_TX_HARDWARE and SOF_TIMESTAMPING_TX_SOFTWARE. It 481is the first if ts[2] is non-zero, the second otherwise, in which 482case the timestamp is stored in ts[0]. 483 484 4852.1.1.3 Fragmentation 486~~~~~~~~~~~~~~~~~~~~~ 487 488Fragmentation of outgoing datagrams is rare, but is possible, e.g., by 489explicitly disabling PMTU discovery. If an outgoing packet is fragmented, 490then only the first fragment is timestamped and returned to the sending 491socket. 492 493 4942.1.1.4 Packet Payload 495~~~~~~~~~~~~~~~~~~~~~~ 496 497The calling application is often not interested in receiving the whole 498packet payload that it passed to the stack originally: the socket 499error queue mechanism is just a method to piggyback the timestamp on. 500In this case, the application can choose to read datagrams with a 501smaller buffer, possibly even of length 0. The payload is truncated 502accordingly. Until the process calls recvmsg() on the error queue, 503however, the full packet is queued, taking up budget from SO_RCVBUF. 504 505 5062.1.1.5 Blocking Read 507~~~~~~~~~~~~~~~~~~~~~ 508 509Reading from the error queue is always a non-blocking operation. To 510block waiting on a timestamp, use poll or select. poll() will return 511POLLERR in pollfd.revents if any data is ready on the error queue. 512There is no need to pass this flag in pollfd.events. This flag is 513ignored on request. See also `man 2 poll`. 514 515 5162.1.2 Receive timestamps 517^^^^^^^^^^^^^^^^^^^^^^^^ 518 519On reception, there is no reason to read from the socket error queue. 520The SCM_TIMESTAMPING ancillary data is sent along with the packet data 521on a normal recvmsg(). Since this is not a socket error, it is not 522accompanied by a message SOL_IP(V6)/IP(V6)_RECVERROR. In this case, 523the meaning of the three fields in struct scm_timestamping is 524implicitly defined. ts[0] holds a software timestamp if set, ts[1] 525is again deprecated and ts[2] holds a hardware timestamp if set. 526 527 5283. Hardware Timestamping configuration: SIOCSHWTSTAMP and SIOCGHWTSTAMP 529======================================================================= 530 531Hardware time stamping must also be initialized for each device driver 532that is expected to do hardware time stamping. The parameter is defined in 533include/uapi/linux/net_tstamp.h as:: 534 535 struct hwtstamp_config { 536 int flags; /* no flags defined right now, must be zero */ 537 int tx_type; /* HWTSTAMP_TX_* */ 538 int rx_filter; /* HWTSTAMP_FILTER_* */ 539 }; 540 541Desired behavior is passed into the kernel and to a specific device by 542calling ioctl(SIOCSHWTSTAMP) with a pointer to a struct ifreq whose 543ifr_data points to a struct hwtstamp_config. The tx_type and 544rx_filter are hints to the driver what it is expected to do. If 545the requested fine-grained filtering for incoming packets is not 546supported, the driver may time stamp more than just the requested types 547of packets. 548 549Drivers are free to use a more permissive configuration than the requested 550configuration. It is expected that drivers should only implement directly the 551most generic mode that can be supported. For example if the hardware can 552support HWTSTAMP_FILTER_PTP_V2_EVENT, then it should generally always upscale 553HWTSTAMP_FILTER_PTP_V2_L2_SYNC, and so forth, as HWTSTAMP_FILTER_PTP_V2_EVENT 554is more generic (and more useful to applications). 555 556A driver which supports hardware time stamping shall update the struct 557with the actual, possibly more permissive configuration. If the 558requested packets cannot be time stamped, then nothing should be 559changed and ERANGE shall be returned (in contrast to EINVAL, which 560indicates that SIOCSHWTSTAMP is not supported at all). 561 562Only a processes with admin rights may change the configuration. User 563space is responsible to ensure that multiple processes don't interfere 564with each other and that the settings are reset. 565 566Any process can read the actual configuration by passing this 567structure to ioctl(SIOCGHWTSTAMP) in the same way. However, this has 568not been implemented in all drivers. 569 570:: 571 572 /* possible values for hwtstamp_config->tx_type */ 573 enum { 574 /* 575 * no outgoing packet will need hardware time stamping; 576 * should a packet arrive which asks for it, no hardware 577 * time stamping will be done 578 */ 579 HWTSTAMP_TX_OFF, 580 581 /* 582 * enables hardware time stamping for outgoing packets; 583 * the sender of the packet decides which are to be 584 * time stamped by setting SOF_TIMESTAMPING_TX_SOFTWARE 585 * before sending the packet 586 */ 587 HWTSTAMP_TX_ON, 588 }; 589 590 /* possible values for hwtstamp_config->rx_filter */ 591 enum { 592 /* time stamp no incoming packet at all */ 593 HWTSTAMP_FILTER_NONE, 594 595 /* time stamp any incoming packet */ 596 HWTSTAMP_FILTER_ALL, 597 598 /* return value: time stamp all packets requested plus some others */ 599 HWTSTAMP_FILTER_SOME, 600 601 /* PTP v1, UDP, any kind of event packet */ 602 HWTSTAMP_FILTER_PTP_V1_L4_EVENT, 603 604 /* for the complete list of values, please check 605 * the include file include/uapi/linux/net_tstamp.h 606 */ 607 }; 608 6093.1 Hardware Timestamping Implementation: Device Drivers 610-------------------------------------------------------- 611 612A driver which supports hardware time stamping must support the 613SIOCSHWTSTAMP ioctl and update the supplied struct hwtstamp_config with 614the actual values as described in the section on SIOCSHWTSTAMP. It 615should also support SIOCGHWTSTAMP. 616 617Time stamps for received packets must be stored in the skb. To get a pointer 618to the shared time stamp structure of the skb call skb_hwtstamps(). Then 619set the time stamps in the structure:: 620 621 struct skb_shared_hwtstamps { 622 /* hardware time stamp transformed into duration 623 * since arbitrary point in time 624 */ 625 ktime_t hwtstamp; 626 }; 627 628Time stamps for outgoing packets are to be generated as follows: 629 630- In hard_start_xmit(), check if (skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) 631 is set no-zero. If yes, then the driver is expected to do hardware time 632 stamping. 633- If this is possible for the skb and requested, then declare 634 that the driver is doing the time stamping by setting the flag 635 SKBTX_IN_PROGRESS in skb_shinfo(skb)->tx_flags , e.g. with:: 636 637 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS; 638 639 You might want to keep a pointer to the associated skb for the next step 640 and not free the skb. A driver not supporting hardware time stamping doesn't 641 do that. A driver must never touch sk_buff::tstamp! It is used to store 642 software generated time stamps by the network subsystem. 643- Driver should call skb_tx_timestamp() as close to passing sk_buff to hardware 644 as possible. skb_tx_timestamp() provides a software time stamp if requested 645 and hardware timestamping is not possible (SKBTX_IN_PROGRESS not set). 646- As soon as the driver has sent the packet and/or obtained a 647 hardware time stamp for it, it passes the time stamp back by 648 calling skb_tstamp_tx() with the original skb, the raw 649 hardware time stamp. skb_tstamp_tx() clones the original skb and 650 adds the timestamps, therefore the original skb has to be freed now. 651 If obtaining the hardware time stamp somehow fails, then the driver 652 should not fall back to software time stamping. The rationale is that 653 this would occur at a later time in the processing pipeline than other 654 software time stamping and therefore could lead to unexpected deltas 655 between time stamps. 656 6573.2 Special considerations for stacked PTP Hardware Clocks 658---------------------------------------------------------- 659 660There are situations when there may be more than one PHC (PTP Hardware Clock) 661in the data path of a packet. The kernel has no explicit mechanism to allow the 662user to select which PHC to use for timestamping Ethernet frames. Instead, the 663assumption is that the outermost PHC is always the most preferable, and that 664kernel drivers collaborate towards achieving that goal. Currently there are 3 665cases of stacked PHCs, detailed below: 666 6673.2.1 DSA (Distributed Switch Architecture) switches 668^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 669 670These are Ethernet switches which have one of their ports connected to an 671(otherwise completely unaware) host Ethernet interface, and perform the role of 672a port multiplier with optional forwarding acceleration features. Each DSA 673switch port is visible to the user as a standalone (virtual) network interface, 674and its network I/O is performed, under the hood, indirectly through the host 675interface (redirecting to the host port on TX, and intercepting frames on RX). 676 677When a DSA switch is attached to a host port, PTP synchronization has to 678suffer, since the switch's variable queuing delay introduces a path delay 679jitter between the host port and its PTP partner. For this reason, some DSA 680switches include a timestamping clock of their own, and have the ability to 681perform network timestamping on their own MAC, such that path delays only 682measure wire and PHY propagation latencies. Timestamping DSA switches are 683supported in Linux and expose the same ABI as any other network interface (save 684for the fact that the DSA interfaces are in fact virtual in terms of network 685I/O, they do have their own PHC). It is typical, but not mandatory, for all 686interfaces of a DSA switch to share the same PHC. 687 688By design, PTP timestamping with a DSA switch does not need any special 689handling in the driver for the host port it is attached to. However, when the 690host port also supports PTP timestamping, DSA will take care of intercepting 691the ``.ndo_eth_ioctl`` calls towards the host port, and block attempts to enable 692hardware timestamping on it. This is because the SO_TIMESTAMPING API does not 693allow the delivery of multiple hardware timestamps for the same packet, so 694anybody else except for the DSA switch port must be prevented from doing so. 695 696In the generic layer, DSA provides the following infrastructure for PTP 697timestamping: 698 699- ``.port_txtstamp()``: a hook called prior to the transmission of 700 packets with a hardware TX timestamping request from user space. 701 This is required for two-step timestamping, since the hardware 702 timestamp becomes available after the actual MAC transmission, so the 703 driver must be prepared to correlate the timestamp with the original 704 packet so that it can re-enqueue the packet back into the socket's 705 error queue. To save the packet for when the timestamp becomes 706 available, the driver can call ``skb_clone_sk`` , save the clone pointer 707 in skb->cb and enqueue a tx skb queue. Typically, a switch will have a 708 PTP TX timestamp register (or sometimes a FIFO) where the timestamp 709 becomes available. In case of a FIFO, the hardware might store 710 key-value pairs of PTP sequence ID/message type/domain number and the 711 actual timestamp. To perform the correlation correctly between the 712 packets in a queue waiting for timestamping and the actual timestamps, 713 drivers can use a BPF classifier (``ptp_classify_raw``) to identify 714 the PTP transport type, and ``ptp_parse_header`` to interpret the PTP 715 header fields. There may be an IRQ that is raised upon this 716 timestamp's availability, or the driver might have to poll after 717 invoking ``dev_queue_xmit()`` towards the host interface. 718 One-step TX timestamping do not require packet cloning, since there is 719 no follow-up message required by the PTP protocol (because the 720 TX timestamp is embedded into the packet by the MAC), and therefore 721 user space does not expect the packet annotated with the TX timestamp 722 to be re-enqueued into its socket's error queue. 723 724- ``.port_rxtstamp()``: On RX, the BPF classifier is run by DSA to 725 identify PTP event messages (any other packets, including PTP general 726 messages, are not timestamped). The original (and only) timestampable 727 skb is provided to the driver, for it to annotate it with a timestamp, 728 if that is immediately available, or defer to later. On reception, 729 timestamps might either be available in-band (through metadata in the 730 DSA header, or attached in other ways to the packet), or out-of-band 731 (through another RX timestamping FIFO). Deferral on RX is typically 732 necessary when retrieving the timestamp needs a sleepable context. In 733 that case, it is the responsibility of the DSA driver to call 734 ``netif_rx()`` on the freshly timestamped skb. 735 7363.2.2 Ethernet PHYs 737^^^^^^^^^^^^^^^^^^^ 738 739These are devices that typically fulfill a Layer 1 role in the network stack, 740hence they do not have a representation in terms of a network interface as DSA 741switches do. However, PHYs may be able to detect and timestamp PTP packets, for 742performance reasons: timestamps taken as close as possible to the wire have the 743potential to yield a more stable and precise synchronization. 744 745A PHY driver that supports PTP timestamping must create a ``struct 746mii_timestamper`` and add a pointer to it in ``phydev->mii_ts``. The presence 747of this pointer will be checked by the networking stack. 748 749Since PHYs do not have network interface representations, the timestamping and 750ethtool ioctl operations for them need to be mediated by their respective MAC 751driver. Therefore, as opposed to DSA switches, modifications need to be done 752to each individual MAC driver for PHY timestamping support. This entails: 753 754- Checking, in ``.ndo_eth_ioctl``, whether ``phy_has_hwtstamp(netdev->phydev)`` 755 is true or not. If it is, then the MAC driver should not process this request 756 but instead pass it on to the PHY using ``phy_mii_ioctl()``. 757 758- On RX, special intervention may or may not be needed, depending on the 759 function used to deliver skb's up the network stack. In the case of plain 760 ``netif_rx()`` and similar, MAC drivers must check whether 761 ``skb_defer_rx_timestamp(skb)`` is necessary or not - and if it is, don't 762 call ``netif_rx()`` at all. If ``CONFIG_NETWORK_PHY_TIMESTAMPING`` is 763 enabled, and ``skb->dev->phydev->mii_ts`` exists, its ``.rxtstamp()`` hook 764 will be called now, to determine, using logic very similar to DSA, whether 765 deferral for RX timestamping is necessary. Again like DSA, it becomes the 766 responsibility of the PHY driver to send the packet up the stack when the 767 timestamp is available. 768 769 For other skb receive functions, such as ``napi_gro_receive`` and 770 ``netif_receive_skb``, the stack automatically checks whether 771 ``skb_defer_rx_timestamp()`` is necessary, so this check is not needed inside 772 the driver. 773 774- On TX, again, special intervention might or might not be needed. The 775 function that calls the ``mii_ts->txtstamp()`` hook is named 776 ``skb_clone_tx_timestamp()``. This function can either be called directly 777 (case in which explicit MAC driver support is indeed needed), but the 778 function also piggybacks from the ``skb_tx_timestamp()`` call, which many MAC 779 drivers already perform for software timestamping purposes. Therefore, if a 780 MAC supports software timestamping, it does not need to do anything further 781 at this stage. 782 7833.2.3 MII bus snooping devices 784^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 785 786These perform the same role as timestamping Ethernet PHYs, save for the fact 787that they are discrete devices and can therefore be used in conjunction with 788any PHY even if it doesn't support timestamping. In Linux, they are 789discoverable and attachable to a ``struct phy_device`` through Device Tree, and 790for the rest, they use the same mii_ts infrastructure as those. See 791Documentation/devicetree/bindings/ptp/timestamper.txt for more details. 792 7933.2.4 Other caveats for MAC drivers 794^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 795 796Stacked PHCs, especially DSA (but not only) - since that doesn't require any 797modification to MAC drivers, so it is more difficult to ensure correctness of 798all possible code paths - is that they uncover bugs which were impossible to 799trigger before the existence of stacked PTP clocks. One example has to do with 800this line of code, already presented earlier:: 801 802 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS; 803 804Any TX timestamping logic, be it a plain MAC driver, a DSA switch driver, a PHY 805driver or a MII bus snooping device driver, should set this flag. 806But a MAC driver that is unaware of PHC stacking might get tripped up by 807somebody other than itself setting this flag, and deliver a duplicate 808timestamp. 809For example, a typical driver design for TX timestamping might be to split the 810transmission part into 2 portions: 811 8121. "TX": checks whether PTP timestamping has been previously enabled through 813 the ``.ndo_eth_ioctl`` ("``priv->hwtstamp_tx_enabled == true``") and the 814 current skb requires a TX timestamp ("``skb_shinfo(skb)->tx_flags & 815 SKBTX_HW_TSTAMP``"). If this is true, it sets the 816 "``skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS``" flag. Note: as 817 described above, in the case of a stacked PHC system, this condition should 818 never trigger, as this MAC is certainly not the outermost PHC. But this is 819 not where the typical issue is. Transmission proceeds with this packet. 820 8212. "TX confirmation": Transmission has finished. The driver checks whether it 822 is necessary to collect any TX timestamp for it. Here is where the typical 823 issues are: the MAC driver takes a shortcut and only checks whether 824 "``skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS``" was set. With a stacked 825 PHC system, this is incorrect because this MAC driver is not the only entity 826 in the TX data path who could have enabled SKBTX_IN_PROGRESS in the first 827 place. 828 829The correct solution for this problem is for MAC drivers to have a compound 830check in their "TX confirmation" portion, not only for 831"``skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS``", but also for 832"``priv->hwtstamp_tx_enabled == true``". Because the rest of the system ensures 833that PTP timestamping is not enabled for anything other than the outermost PHC, 834this enhanced check will avoid delivering a duplicated TX timestamp to user 835space. 836