Commit | Line | Data |
---|---|---|
8fe2f761 WB |
1 | |
2 | 1. Control Interfaces | |
3 | ||
4 | The interfaces for receiving network packages timestamps are: | |
cb9eff09 PO |
5 | |
6 | * SO_TIMESTAMP | |
8fe2f761 WB |
7 | Generates a timestamp for each incoming packet in (not necessarily |
8 | monotonic) system time. Reports the timestamp via recvmsg() in a | |
9 | control message as struct timeval (usec resolution). | |
cb9eff09 PO |
10 | |
11 | * SO_TIMESTAMPNS | |
8fe2f761 WB |
12 | Same timestamping mechanism as SO_TIMESTAMP, but reports the |
13 | timestamp as struct timespec (nsec resolution). | |
cb9eff09 PO |
14 | |
15 | * IP_MULTICAST_LOOP + SO_TIMESTAMP[NS] | |
8fe2f761 WB |
16 | Only for multicast:approximate transmit timestamp obtained by |
17 | reading the looped packet receive timestamp. | |
cb9eff09 | 18 | |
8fe2f761 WB |
19 | * SO_TIMESTAMPING |
20 | Generates timestamps on reception, transmission or both. Supports | |
21 | multiple timestamp sources, including hardware. Supports generating | |
22 | timestamps for stream sockets. | |
cb9eff09 | 23 | |
cb9eff09 | 24 | |
8fe2f761 | 25 | 1.1 SO_TIMESTAMP: |
adca4767 | 26 | |
8fe2f761 WB |
27 | This socket option enables timestamping of datagrams on the reception |
28 | path. Because the destination socket, if any, is not known early in | |
29 | the network stack, the feature has to be enabled for all packets. The | |
30 | same is true for all early receive timestamp options. | |
adca4767 | 31 | |
8fe2f761 WB |
32 | For interface details, see `man 7 socket`. |
33 | ||
34 | ||
35 | 1.2 SO_TIMESTAMPNS: | |
36 | ||
37 | This option is identical to SO_TIMESTAMP except for the returned data type. | |
38 | Its struct timespec allows for higher resolution (ns) timestamps than the | |
39 | timeval of SO_TIMESTAMP (ms). | |
40 | ||
41 | ||
42 | 1.3 SO_TIMESTAMPING: | |
43 | ||
44 | Supports multiple types of timestamp requests. As a result, this | |
45 | socket option takes a bitmap of flags, not a boolean. In | |
46 | ||
47 | err = setsockopt(fd, SOL_SOCKET, SO_TIMESTAMPING, (void *) val, &val); | |
48 | ||
49 | val is an integer with any of the following bits set. Setting other | |
50 | bit returns EINVAL and does not change the current state. | |
adca4767 | 51 | |
adca4767 | 52 | |
8fe2f761 | 53 | 1.3.1 Timestamp Generation |
adca4767 | 54 | |
8fe2f761 WB |
55 | Some bits are requests to the stack to try to generate timestamps. Any |
56 | combination of them is valid. Changes to these bits apply to newly | |
57 | created packets, not to packets already in the stack. As a result, it | |
58 | is possible to selectively request timestamps for a subset of packets | |
59 | (e.g., for sampling) by embedding an send() call within two setsockopt | |
60 | calls, one to enable timestamp generation and one to disable it. | |
61 | Timestamps may also be generated for reasons other than being | |
62 | requested by a particular socket, such as when receive timestamping is | |
63 | enabled system wide, as explained earlier. | |
adca4767 | 64 | |
8fe2f761 WB |
65 | SOF_TIMESTAMPING_RX_HARDWARE: |
66 | Request rx timestamps generated by the network adapter. | |
67 | ||
68 | SOF_TIMESTAMPING_RX_SOFTWARE: | |
69 | Request rx timestamps when data enters the kernel. These timestamps | |
70 | are generated just after a device driver hands a packet to the | |
71 | kernel receive stack. | |
72 | ||
73 | SOF_TIMESTAMPING_TX_HARDWARE: | |
74 | Request tx timestamps generated by the network adapter. | |
75 | ||
76 | SOF_TIMESTAMPING_TX_SOFTWARE: | |
77 | Request tx timestamps when data leaves the kernel. These timestamps | |
78 | are generated in the device driver as close as possible, but always | |
79 | prior to, passing the packet to the network interface. Hence, they | |
80 | require driver support and may not be available for all devices. | |
81 | ||
82 | SOF_TIMESTAMPING_TX_SCHED: | |
83 | Request tx timestamps prior to entering the packet scheduler. Kernel | |
84 | transmit latency is, if long, often dominated by queuing delay. The | |
85 | difference between this timestamp and one taken at | |
86 | SOF_TIMESTAMPING_TX_SOFTWARE will expose this latency independent | |
87 | of protocol processing. The latency incurred in protocol | |
88 | processing, if any, can be computed by subtracting a userspace | |
89 | timestamp taken immediately before send() from this timestamp. On | |
90 | machines with virtual devices where a transmitted packet travels | |
91 | through multiple devices and, hence, multiple packet schedulers, | |
92 | a timestamp is generated at each layer. This allows for fine | |
93 | grained measurement of queuing delay. | |
94 | ||
95 | SOF_TIMESTAMPING_TX_ACK: | |
96 | Request tx timestamps when all data in the send buffer has been | |
97 | acknowledged. This only makes sense for reliable protocols. It is | |
98 | currently only implemented for TCP. For that protocol, it may | |
99 | over-report measurement, because the timestamp is generated when all | |
100 | data up to and including the buffer at send() was acknowledged: the | |
101 | cumulative acknowledgment. The mechanism ignores SACK and FACK. | |
102 | ||
103 | ||
104 | 1.3.2 Timestamp Reporting | |
105 | ||
106 | The other three bits control which timestamps will be reported in a | |
107 | generated control message. Changes to the bits take immediate | |
108 | effect at the timestamp reporting locations in the stack. Timestamps | |
109 | are only reported for packets that also have the relevant timestamp | |
110 | generation request set. | |
111 | ||
112 | SOF_TIMESTAMPING_SOFTWARE: | |
113 | Report any software timestamps when available. | |
114 | ||
115 | SOF_TIMESTAMPING_SYS_HARDWARE: | |
116 | This option is deprecated and ignored. | |
117 | ||
118 | SOF_TIMESTAMPING_RAW_HARDWARE: | |
119 | Report hardware timestamps as generated by | |
120 | SOF_TIMESTAMPING_TX_HARDWARE when available. | |
121 | ||
122 | ||
123 | 1.3.3 Timestamp Options | |
124 | ||
125 | The interface supports one option | |
126 | ||
127 | SOF_TIMESTAMPING_OPT_ID: | |
128 | ||
129 | Generate a unique identifier along with each packet. A process can | |
130 | have multiple concurrent timestamping requests outstanding. Packets | |
131 | can be reordered in the transmit path, for instance in the packet | |
132 | scheduler. In that case timestamps will be queued onto the error | |
133 | queue out of order from the original send() calls. This option | |
134 | embeds a counter that is incremented at send() time, to order | |
135 | timestamps within a flow. | |
136 | ||
137 | This option is implemented only for transmit timestamps. There, the | |
138 | timestamp is always looped along with a struct sock_extended_err. | |
139 | The option modifies field ee_info to pass an id that is unique | |
140 | among all possibly concurrently outstanding timestamp requests for | |
141 | that socket. In practice, it is a monotonically increasing u32 | |
142 | (that wraps). | |
143 | ||
144 | In datagram sockets, the counter increments on each send call. In | |
145 | stream sockets, it increments with every byte. | |
146 | ||
147 | ||
148 | 1.4 Bytestream Timestamps | |
149 | ||
150 | The SO_TIMESTAMPING interface supports timestamping of bytes in a | |
151 | bytestream. Each request is interpreted as a request for when the | |
152 | entire contents of the buffer has passed a timestamping point. That | |
153 | is, for streams option SOF_TIMESTAMPING_TX_SOFTWARE will record | |
154 | when all bytes have reached the device driver, regardless of how | |
155 | many packets the data has been converted into. | |
156 | ||
157 | In general, bytestreams have no natural delimiters and therefore | |
158 | correlating a timestamp with data is non-trivial. A range of bytes | |
159 | may be split across segments, any segments may be merged (possibly | |
160 | coalescing sections of previously segmented buffers associated with | |
161 | independent send() calls). Segments can be reordered and the same | |
162 | byte range can coexist in multiple segments for protocols that | |
163 | implement retransmissions. | |
164 | ||
165 | It is essential that all timestamps implement the same semantics, | |
166 | regardless of these possible transformations, as otherwise they are | |
167 | incomparable. Handling "rare" corner cases differently from the | |
168 | simple case (a 1:1 mapping from buffer to skb) is insufficient | |
169 | because performance debugging often needs to focus on such outliers. | |
170 | ||
171 | In practice, timestamps can be correlated with segments of a | |
172 | bytestream consistently, if both semantics of the timestamp and the | |
173 | timing of measurement are chosen correctly. This challenge is no | |
174 | different from deciding on a strategy for IP fragmentation. There, the | |
175 | definition is that only the first fragment is timestamped. For | |
176 | bytestreams, we chose that a timestamp is generated only when all | |
177 | bytes have passed a point. SOF_TIMESTAMPING_TX_ACK as defined is easy to | |
178 | implement and reason about. An implementation that has to take into | |
179 | account SACK would be more complex due to possible transmission holes | |
180 | and out of order arrival. | |
181 | ||
182 | On the host, TCP can also break the simple 1:1 mapping from buffer to | |
183 | skbuff as a result of Nagle, cork, autocork, segmentation and GSO. The | |
184 | implementation ensures correctness in all cases by tracking the | |
185 | individual last byte passed to send(), even if it is no longer the | |
186 | last byte after an skbuff extend or merge operation. It stores the | |
187 | relevant sequence number in skb_shinfo(skb)->tskey. Because an skbuff | |
188 | has only one such field, only one timestamp can be generated. | |
189 | ||
190 | In rare cases, a timestamp request can be missed if two requests are | |
191 | collapsed onto the same skb. A process can detect this situation by | |
192 | enabling SOF_TIMESTAMPING_OPT_ID and comparing the byte offset at | |
193 | send time with the value returned for each timestamp. It can prevent | |
194 | the situation by always flushing the TCP stack in between requests, | |
195 | for instance by enabling TCP_NODELAY and disabling TCP_CORK and | |
196 | autocork. | |
197 | ||
198 | These precautions ensure that the timestamp is generated only when all | |
199 | bytes have passed a timestamp point, assuming that the network stack | |
200 | itself does not reorder the segments. The stack indeed tries to avoid | |
201 | reordering. The one exception is under administrator control: it is | |
202 | possible to construct a packet scheduler configuration that delays | |
203 | segments from the same stream differently. Such a setup would be | |
204 | unusual. | |
205 | ||
206 | ||
207 | 2 Data Interfaces | |
208 | ||
209 | Timestamps are read using the ancillary data feature of recvmsg(). | |
210 | See `man 3 cmsg` for details of this interface. The socket manual | |
211 | page (`man 7 socket`) describes how timestamps generated with | |
212 | SO_TIMESTAMP and SO_TIMESTAMPNS records can be retrieved. | |
213 | ||
214 | ||
215 | 2.1 SCM_TIMESTAMPING records | |
216 | ||
217 | These timestamps are returned in a control message with cmsg_level | |
218 | SOL_SOCKET, cmsg_type SCM_TIMESTAMPING, and payload of type | |
69298698 PL |
219 | |
220 | struct scm_timestamping { | |
8fe2f761 | 221 | struct timespec ts[3]; |
69298698 | 222 | }; |
cb9eff09 | 223 | |
8fe2f761 WB |
224 | The structure can return up to three timestamps. This is a legacy |
225 | feature. Only one field is non-zero at any time. Most timestamps | |
226 | are passed in ts[0]. Hardware timestamps are passed in ts[2]. | |
227 | ||
228 | ts[1] used to hold hardware timestamps converted to system time. | |
229 | Instead, expose the hardware clock device on the NIC directly as | |
230 | a HW PTP clock source, to allow time conversion in userspace and | |
231 | optionally synchronize system time with a userspace PTP stack such | |
232 | as linuxptp. For the PTP clock API, see Documentation/ptp/ptp.txt. | |
233 | ||
234 | 2.1.1 Transmit timestamps with MSG_ERRQUEUE | |
235 | ||
236 | For transmit timestamps the outgoing packet is looped back to the | |
237 | socket's error queue with the send timestamp(s) attached. A process | |
238 | receives the timestamps by calling recvmsg() with flag MSG_ERRQUEUE | |
239 | set and with a msg_control buffer sufficiently large to receive the | |
240 | relevant metadata structures. The recvmsg call returns the original | |
241 | outgoing data packet with two ancillary messages attached. | |
242 | ||
243 | A message of cm_level SOL_IP(V6) and cm_type IP(V6)_RECVERR | |
244 | embeds a struct sock_extended_err. This defines the error type. For | |
245 | timestamps, the ee_errno field is ENOMSG. The other ancillary message | |
246 | will have cm_level SOL_SOCKET and cm_type SCM_TIMESTAMPING. This | |
247 | embeds the struct scm_timestamping. | |
248 | ||
249 | ||
250 | 2.1.1.2 Timestamp types | |
251 | ||
252 | The semantics of the three struct timespec are defined by field | |
253 | ee_info in the extended error structure. It contains a value of | |
254 | type SCM_TSTAMP_* to define the actual timestamp passed in | |
255 | scm_timestamping. | |
256 | ||
257 | The SCM_TSTAMP_* types are 1:1 matches to the SOF_TIMESTAMPING_* | |
258 | control fields discussed previously, with one exception. For legacy | |
259 | reasons, SCM_TSTAMP_SND is equal to zero and can be set for both | |
260 | SOF_TIMESTAMPING_TX_HARDWARE and SOF_TIMESTAMPING_TX_SOFTWARE. It | |
261 | is the first if ts[2] is non-zero, the second otherwise, in which | |
262 | case the timestamp is stored in ts[0]. | |
263 | ||
264 | ||
265 | 2.1.1.3 Fragmentation | |
266 | ||
267 | Fragmentation of outgoing datagrams is rare, but is possible, e.g., by | |
268 | explicitly disabling PMTU discovery. If an outgoing packet is fragmented, | |
269 | then only the first fragment is timestamped and returned to the sending | |
270 | socket. | |
271 | ||
272 | ||
273 | 2.1.1.4 Packet Payload | |
274 | ||
275 | The calling application is often not interested in receiving the whole | |
276 | packet payload that it passed to the stack originally: the socket | |
277 | error queue mechanism is just a method to piggyback the timestamp on. | |
278 | In this case, the application can choose to read datagrams with a | |
279 | smaller buffer, possibly even of length 0. The payload is truncated | |
280 | accordingly. Until the process calls recvmsg() on the error queue, | |
281 | however, the full packet is queued, taking up budget from SO_RCVBUF. | |
282 | ||
283 | ||
284 | 2.1.1.5 Blocking Read | |
285 | ||
286 | Reading from the error queue is always a non-blocking operation. To | |
287 | block waiting on a timestamp, use poll or select. poll() will return | |
288 | POLLERR in pollfd.revents if any data is ready on the error queue. | |
289 | There is no need to pass this flag in pollfd.events. This flag is | |
290 | ignored on request. See also `man 2 poll`. | |
291 | ||
292 | ||
293 | 2.1.2 Receive timestamps | |
294 | ||
295 | On reception, there is no reason to read from the socket error queue. | |
296 | The SCM_TIMESTAMPING ancillary data is sent along with the packet data | |
297 | on a normal recvmsg(). Since this is not a socket error, it is not | |
298 | accompanied by a message SOL_IP(V6)/IP(V6)_RECVERROR. In this case, | |
299 | the meaning of the three fields in struct scm_timestamping is | |
300 | implicitly defined. ts[0] holds a software timestamp if set, ts[1] | |
301 | is again deprecated and ts[2] holds a hardware timestamp if set. | |
302 | ||
303 | ||
304 | 3. Hardware Timestamping configuration: SIOCSHWTSTAMP and SIOCGHWTSTAMP | |
cb9eff09 PO |
305 | |
306 | Hardware time stamping must also be initialized for each device driver | |
69298698 PL |
307 | that is expected to do hardware time stamping. The parameter is defined in |
308 | /include/linux/net_tstamp.h as: | |
cb9eff09 PO |
309 | |
310 | struct hwtstamp_config { | |
69298698 PL |
311 | int flags; /* no flags defined right now, must be zero */ |
312 | int tx_type; /* HWTSTAMP_TX_* */ | |
313 | int rx_filter; /* HWTSTAMP_FILTER_* */ | |
cb9eff09 PO |
314 | }; |
315 | ||
316 | Desired behavior is passed into the kernel and to a specific device by | |
317 | calling ioctl(SIOCSHWTSTAMP) with a pointer to a struct ifreq whose | |
318 | ifr_data points to a struct hwtstamp_config. The tx_type and | |
319 | rx_filter are hints to the driver what it is expected to do. If | |
320 | the requested fine-grained filtering for incoming packets is not | |
321 | supported, the driver may time stamp more than just the requested types | |
322 | of packets. | |
323 | ||
324 | A driver which supports hardware time stamping shall update the struct | |
325 | with the actual, possibly more permissive configuration. If the | |
326 | requested packets cannot be time stamped, then nothing should be | |
327 | changed and ERANGE shall be returned (in contrast to EINVAL, which | |
328 | indicates that SIOCSHWTSTAMP is not supported at all). | |
329 | ||
330 | Only a processes with admin rights may change the configuration. User | |
331 | space is responsible to ensure that multiple processes don't interfere | |
332 | with each other and that the settings are reset. | |
333 | ||
fd468c74 BH |
334 | Any process can read the actual configuration by passing this |
335 | structure to ioctl(SIOCGHWTSTAMP) in the same way. However, this has | |
336 | not been implemented in all drivers. | |
337 | ||
cb9eff09 PO |
338 | /* possible values for hwtstamp_config->tx_type */ |
339 | enum { | |
340 | /* | |
341 | * no outgoing packet will need hardware time stamping; | |
342 | * should a packet arrive which asks for it, no hardware | |
343 | * time stamping will be done | |
344 | */ | |
345 | HWTSTAMP_TX_OFF, | |
346 | ||
347 | /* | |
348 | * enables hardware time stamping for outgoing packets; | |
349 | * the sender of the packet decides which are to be | |
350 | * time stamped by setting SOF_TIMESTAMPING_TX_SOFTWARE | |
351 | * before sending the packet | |
352 | */ | |
353 | HWTSTAMP_TX_ON, | |
354 | }; | |
355 | ||
356 | /* possible values for hwtstamp_config->rx_filter */ | |
357 | enum { | |
358 | /* time stamp no incoming packet at all */ | |
359 | HWTSTAMP_FILTER_NONE, | |
360 | ||
361 | /* time stamp any incoming packet */ | |
362 | HWTSTAMP_FILTER_ALL, | |
363 | ||
69298698 PL |
364 | /* return value: time stamp all packets requested plus some others */ |
365 | HWTSTAMP_FILTER_SOME, | |
cb9eff09 PO |
366 | |
367 | /* PTP v1, UDP, any kind of event packet */ | |
368 | HWTSTAMP_FILTER_PTP_V1_L4_EVENT, | |
369 | ||
69298698 PL |
370 | /* for the complete list of values, please check |
371 | * the include file /include/linux/net_tstamp.h | |
372 | */ | |
cb9eff09 PO |
373 | }; |
374 | ||
8fe2f761 | 375 | 3.1 Hardware Timestamping Implementation: Device Drivers |
cb9eff09 PO |
376 | |
377 | A driver which supports hardware time stamping must support the | |
69298698 | 378 | SIOCSHWTSTAMP ioctl and update the supplied struct hwtstamp_config with |
fd468c74 BH |
379 | the actual values as described in the section on SIOCSHWTSTAMP. It |
380 | should also support SIOCGHWTSTAMP. | |
69298698 PL |
381 | |
382 | Time stamps for received packets must be stored in the skb. To get a pointer | |
383 | to the shared time stamp structure of the skb call skb_hwtstamps(). Then | |
384 | set the time stamps in the structure: | |
385 | ||
386 | struct skb_shared_hwtstamps { | |
387 | /* hardware time stamp transformed into duration | |
388 | * since arbitrary point in time | |
389 | */ | |
390 | ktime_t hwtstamp; | |
69298698 | 391 | }; |
cb9eff09 PO |
392 | |
393 | Time stamps for outgoing packets are to be generated as follows: | |
2244d07b OH |
394 | - In hard_start_xmit(), check if (skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) |
395 | is set no-zero. If yes, then the driver is expected to do hardware time | |
396 | stamping. | |
cb9eff09 | 397 | - If this is possible for the skb and requested, then declare |
2244d07b OH |
398 | that the driver is doing the time stamping by setting the flag |
399 | SKBTX_IN_PROGRESS in skb_shinfo(skb)->tx_flags , e.g. with | |
400 | ||
401 | skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS; | |
402 | ||
403 | You might want to keep a pointer to the associated skb for the next step | |
404 | and not free the skb. A driver not supporting hardware time stamping doesn't | |
405 | do that. A driver must never touch sk_buff::tstamp! It is used to store | |
406 | software generated time stamps by the network subsystem. | |
59cb89e6 JK |
407 | - Driver should call skb_tx_timestamp() as close to passing sk_buff to hardware |
408 | as possible. skb_tx_timestamp() provides a software time stamp if requested | |
409 | and hardware timestamping is not possible (SKBTX_IN_PROGRESS not set). | |
cb9eff09 PO |
410 | - As soon as the driver has sent the packet and/or obtained a |
411 | hardware time stamp for it, it passes the time stamp back by | |
412 | calling skb_hwtstamp_tx() with the original skb, the raw | |
69298698 PL |
413 | hardware time stamp. skb_hwtstamp_tx() clones the original skb and |
414 | adds the timestamps, therefore the original skb has to be freed now. | |
415 | If obtaining the hardware time stamp somehow fails, then the driver | |
416 | should not fall back to software time stamping. The rationale is that | |
417 | this would occur at a later time in the processing pipeline than other | |
418 | software time stamping and therefore could lead to unexpected deltas | |
419 | between time stamps. |