1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright 2010 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
24 * Copyright 2018 Joyent, Inc.
25 * Copyright 2013 Nexenta Systems, Inc. All rights reserved.
26 * Copyright 2024 Oxide Computer Company
27 */
28
29 /*
30 * MAC data path
31 *
32 * The MAC data path is concerned with the flow of traffic from mac clients --
33 * DLS, IP, etc. -- to various GLDv3 device drivers -- e1000g, vnic, aggr,
34 * ixgbe, etc. -- and from the GLDv3 device drivers back to clients.
35 *
36 * -----------
37 * Terminology
38 * -----------
39 *
40 * MAC uses a lot of different, but related terms that are associated with the
41 * design and structure of the data path. Before we cover other aspects, first
42 * let's review the terminology that MAC uses.
43 *
44 * MAC
45 *
46 * This driver. It interfaces with device drivers and provides abstractions
47 * that the rest of the system consumes. All data links -- things managed
48 * with dladm(8), are accessed through MAC.
49 *
50 * GLDv3 DEVICE DRIVER
51 *
52 * A GLDv3 device driver refers to a driver, both for pseudo-devices and
53 * real devices, which implement the GLDv3 driver API. Common examples of
54 * these are igb and ixgbe, which are drivers for various Intel networking
55 * cards. These devices may or may not have various features, such as
56 * hardware rings and checksum offloading. For MAC, a GLDv3 device is the
57 * final point for the transmission of a packet and the starting point for
58 * the receipt of a packet.
59 *
60 * FLOWS
61 *
62 * At a high level, a flow refers to a series of packets that are related.
63 * Often times the term is used in the context of TCP to indicate a unique
64 * TCP connection and the traffic over it. However, a flow can exist at
65 * other levels of the system as well. MAC has a notion of a default flow
66 * which is used for all unicast traffic addressed to the address of a MAC
67 * device. For example, when a VNIC is created, a default flow is created
68 * for the VNIC's MAC address. In addition, flows are created for broadcast
69 * groups and a user may create a flow with flowadm(8).
70 *
71 * CLASSIFICATION
72 *
73 * Classification refers to the notion of identifying an incoming frame
74 * based on its destination address and optionally its source addresses and
75 * doing different processing based on that information. Classification can
76 * be done in both hardware and software. In general, we usually only
77 * classify based on the layer two destination, eg. for Ethernet, the
78 * destination MAC address.
79 *
80 * The system also will do classification based on layer three and layer
81 * four properties. This is used to support things like flowadm(8), which
82 * allows setting QoS and other properties on a per-flow basis.
83 *
84 * RING
85 *
86 * Conceptually, a ring represents a series of framed messages, often in a
87 * contiguous chunk of memory that acts as a circular buffer. Rings come in
88 * a couple of forms. Generally they are either a hardware construct (hw
89 * ring) or they are a software construct (sw ring) maintained by MAC.
90 *
91 * HW RING
92 *
93 * A hardware ring is a set of resources provided by a GLDv3 device driver
94 * (even if it is a pseudo-device). A hardware ring comes in two different
95 * forms: receive (rx) rings and transmit (tx) rings. An rx hw ring is
96 * something that has a unique DMA (direct memory access) region and
97 * generally supports some form of classification (though it isn't always
98 * used), as well as a means of generating an interrupt specific to that
99 * ring. For example, the device may generate a specific MSI-X for a PCI
100 * express device. A tx ring is similar, except that it is dedicated to
101 * transmission. It may also be a vector for enabling features such as VLAN
102 * tagging and large transmit offloading. It usually has its own dedicated
103 * interrupts for transmit being completed.
104 *
105 * SW RING
106 *
107 * A software ring is a construction of MAC. It represents the same thing
108 * that a hardware ring generally does, a collection of frames. However,
109 * instead of being in a contiguous ring of memory, they're instead linked
110 * by using the mblk_t's b_next pointer. Each frame may itself be multiple
111 * mblk_t's linked together by the b_cont pointer. A software ring always
112 * represents a collection of classified packets; however, it varies as to
113 * whether it uses only layer two information, or a combination of that and
114 * additional layer three and layer four data.
115 *
116 * FANOUT
117 *
118 * Fanout is the idea of spreading out the load of processing frames based
119 * on the source and destination information contained in the layer two,
120 * three, and four headers, such that the data can then be processed in
121 * parallel using multiple hardware threads.
122 *
123 * A fanout algorithm hashes the headers and uses that to place different
124 * flows into a bucket. The most important thing is that packets that are
125 * in the same flow end up in the same bucket. If they do not, performance
126 * can be adversely affected. Consider the case of TCP. TCP severely
127 * penalizes a connection if the data arrives out of order. If a given flow
128 * is processed on different CPUs, then the data will appear out of order,
129 * hence the invariant that fanout always hash a given flow to the same
130 * bucket and thus get processed on the same CPU.
131 *
132 * RECEIVE SIDE SCALING (RSS)
133 *
134 *
135 * Receive side scaling is a term that isn't common in illumos, but is used
136 * by vendors and was popularized by Microsoft. It refers to the idea of
137 * spreading the incoming receive load out across multiple interrupts which
138 * can be directed to different CPUs. This allows a device to leverage
139 * hardware rings even when it doesn't support hardware classification. The
140 * hardware uses an algorithm to perform fanout that ensures the flow
141 * invariant is maintained.
142 *
143 * SOFT RING SET
144 *
145 * A soft ring set, commonly abbreviated SRS, is a collection of rings and
146 * is used for both transmitting and receiving. It is maintained in the
147 * structure mac_soft_ring_set_t. A soft ring set is usually associated
148 * with flows, and coordinates both the use of hardware and software rings.
149 * Because the use of hardware rings can change as devices such as VNICs
150 * come and go, we always ensure that the set has software classification
151 * rules that correspond to the hardware classification rules from rings.
152 *
153 * Soft ring sets are also used for the enforcement of various QoS
154 * properties. For example, if a bandwidth limit has been placed on a
155 * specific flow or device, then that will be enforced by the soft ring
156 * set.
157 *
158 * SERVICE ATTACHMENT POINT (SAP)
159 *
160 * The service attachment point is a DLPI (Data Link Provider Interface)
161 * concept; however, it comes up quite often in MAC. Most MAC devices speak
162 * a protocol that has some notion of different channels or message type
163 * identifiers. For example, Ethernet defines an EtherType which is a part
164 * of the Ethernet header and defines the particular protocol of the data
165 * payload. If the EtherType is set to 0x0800, then it defines that the
166 * contents of that Ethernet frame is IPv4 traffic. For Ethernet, the
167 * EtherType is the SAP.
168 *
169 * In DLPI, a given consumer attaches to a specific SAP. In illumos, the ip
170 * and arp drivers attach to the EtherTypes for IPv4, IPv6, and ARP. Using
171 * libdlpi(3LIB) user software can attach to arbitrary SAPs. With the
172 * exception of 802.1Q VLAN tagged traffic, MAC itself does not directly
173 * consume the SAP; however, it uses that information as part of hashing
174 * and it may be used as part of the construction of flows.
175 *
176 * PRIMARY MAC CLIENT
177 *
178 * The primary mac client refers to a mac client whose unicast address
179 * matches the address of the device itself. For example, if the system has
180 * instance of the e1000g driver such as e1000g0, e1000g1, etc., the
181 * primary mac client is the one named after the device itself. VNICs that
182 * are created on top of such devices are not the primary client.
183 *
184 * TRANSMIT DESCRIPTORS
185 *
186 * Transmit descriptors are a resource that most GLDv3 device drivers have.
187 * Generally, a GLDv3 device driver takes a frame that's meant to be output
188 * and puts a copy of it into a region of memory. Each region of memory
189 * usually has an associated descriptor that the device uses to manage
190 * properties of the frames. Devices have a limited number of such
191 * descriptors. They get reclaimed once the device finishes putting the
192 * frame on the wire.
193 *
194 * If the driver runs out of transmit descriptors, for example, the OS is
195 * generating more frames than it can put on the wire, then it will return
196 * them back to the MAC layer.
197 *
198 * ---------------------------------
199 * Rings, Classification, and Fanout
200 * ---------------------------------
201 *
202 * The heart of MAC is made up of rings, and not those that Elven-kings wear.
203 * When receiving a packet, MAC breaks the work into two different, though
204 * interrelated phases. The first phase is generally classification and then the
205 * second phase is generally fanout. When a frame comes in from a GLDv3 Device,
206 * MAC needs to determine where that frame should be delivered. If it's a
207 * unicast frame (say a normal TCP/IP packet), then it will be delivered to a
208 * single MAC client; however, if it's a broadcast or multicast frame, then MAC
209 * may need to deliver it to multiple MAC clients.
210 *
211 * On transmit, classification isn't quite as important, but may still be used.
212 * Unlike with the receive path, the classification is not used to determine
213 * devices that should transmit something, but rather is used for special
214 * properties of a flow, eg. bandwidth limits for a given IP address, device, or
215 * connection.
216 *
217 * MAC employs a software classifier and leverages hardware classification as
218 * well. The software classifier can leverage the full layer two information,
219 * source, destination, VLAN, and SAP. If the SAP indicates that IP traffic is
220 * being sent, it can classify based on the IP header, and finally, it also
221 * knows how to classify based on the local and remote ports of TCP, UDP, and
222 * SCTP.
223 *
224 * Hardware classifiers vary in capability. Generally all hardware classifiers
225 * provide the capability to classify based on the destination MAC address. Some
226 * hardware has additional filters built in for performing more in-depth
227 * classification; however, it often has much more limited resources for these
228 * activities as compared to the layer two destination address classification.
229 *
230 * The modus operandi in MAC is to always ensure that we have software-based
231 * capabilities and rules in place and then to supplement that with hardware
232 * resources when available. In general, simple layer two classification is
233 * sufficient and nothing else is used, unless a specific flow is created with
234 * tools such as flowadm(8) or bandwidth limits are set on a device with
235 * dladm(8).
236 *
237 * RINGS AND GROUPS
238 *
239 * To get into how rings and classification play together, it's first important
240 * to understand how hardware devices commonly associate rings and allow them to
241 * be programmed. Recall that a hardware ring should be thought of as a DMA
242 * buffer and an interrupt resource. Rings are then collected into groups. A
243 * group itself has a series of classification rules. One or more MAC addresses
244 * are assigned to a group.
245 *
246 * Hardware devices vary in terms of what capabilities they provide. Sometimes
247 * they allow for a dynamic assignment of rings to a group and sometimes they
248 * have a static assignment of rings to a group. For example, the ixgbe driver
249 * has a static assignment of rings to groups such that every group has exactly
250 * one ring and the number of groups is equal to the number of rings.
251 *
252 * Classification and receive side scaling both come into play with how a device
253 * advertises itself to MAC and how MAC uses it. If a device supports layer two
254 * classification of frames, then MAC will assign MAC addresses to a group as a
255 * form of primary classification. If a single MAC address is assigned to a
256 * group, a common case, then MAC will consider packets that come in from rings
257 * on that group to be fully classified and will not need to do any software
258 * classification unless a specific flow has been created.
259 *
260 * If a device supports receive side scaling, then it may advertise or support
261 * groups with multiple rings. In those cases, then receive side scaling will
262 * come into play and MAC will use that as a means of fanning out received
263 * frames across multiple CPUs. This can also be combined with groups that
264 * support layer two classification.
265 *
266 * If a device supports dynamic assignments of rings to groups, then MAC will
267 * change around the way that rings are assigned to various groups as devices
268 * come and go from the system. For example, when a VNIC is created, a new flow
269 * will be created for the VNIC's MAC address. If a hardware ring is available,
270 * MAC may opt to reassign it from one group to another.
271 *
272 * ASSIGNMENT OF HARDWARE RINGS
273 *
274 * This is a bit of a complicated subject that varies depending on the device,
275 * the use of aggregations, the special nature of the primary mac client. This
276 * section deserves being fleshed out.
277 *
278 * FANOUT
279 *
280 * illumos uses fanout to help spread out the incoming processing load of chains
281 * of frames away from a single CPU. If a device supports receive side scaling,
282 * then that provides an initial form of fanout; however, what we're concerned
283 * with all happens after the context of a given set of frames being classified
284 * to a soft ring set.
285 *
286 * After frames reach a soft ring set and account for any potential bandwidth
287 * related accounting, they may be fanned out based on one of the following
288 * three modes:
289 *
290 * o No Fanout
291 * o Protocol level fanout
292 * o Full software ring protocol fanout
293 *
294 * MAC makes the determination as to which of these modes a given soft ring set
295 * obtains based on parameters such as whether or not it's the primary mac
296 * client, whether it's on a 10 GbE or faster device, user controlled dladm(8)
297 * properties, and the nature of the hardware and the resources that it has.
298 *
299 * When there is no fanout, MAC does not create any soft rings for a device and
300 * the device has frames delivered directly to the MAC client.
301 *
302 * Otherwise, all fanout is performed by software. MAC divides incoming frames
303 * into one of three buckets -- IPv4 TCP traffic, IPv4 UDP traffic, and
304 * everything else. Regardless of the type of fanout, these three categories
305 * or buckets are always used.
306 *
307 * The difference between protocol level fanout and full software ring protocol
308 * fanout is the number of software rings that end up getting created. The
309 * system always uses the same number of software rings per protocol bucket. So
310 * in the first case when we're just doing protocol level fanout, we just create
311 * one software ring each for IPv4 TCP traffic, IPv4 UDP traffic, and everything
312 * else.
313 *
314 * In the case where we do full software ring protocol fanout, we generally use
315 * mac_compute_soft_ring_count() to determine the number of rings. There are
316 * other combinations of properties and devices that may send us down other
317 * paths, but this is a common starting point. If it's a non-bandwidth enforced
318 * device and we're on at least a 10 GbE link, then we'll use eight soft rings
319 * per protocol bucket as a starting point. See mac_compute_soft_ring_count()
320 * for more information on the total number.
321 *
322 * For each of these rings, we create a mac_soft_ring_t and an associated worker
323 * thread. Particularly when doing full software ring protocol fanout, we bind
324 * each of the worker threads to individual CPUs.
325 *
326 * The other advantage of these software rings is that it allows upper layers to
327 * optionally poll on them. For example, TCP can leverage an squeue to poll on
328 * the software ring, see squeue.c for more information.
329 *
330 * DLS BYPASS
331 *
332 * DLS is the data link services module. It interfaces with DLPI, which is the
333 * primary way that other parts of the system such as IP interface with the MAC
334 * layer. While DLS is traditionally a STREAMS-based interface, it allows for
335 * certain modules such as IP to negotiate various more modern interfaces to be
336 * used, which are useful for higher performance and allow it to use direct
337 * function calls to DLS instead of using STREAMS.
338 *
339 * When we have IPv4 TCP or UDP software rings, then traffic on those rings is
340 * eligible for what we call the dls bypass. In those cases, rather than going
341 * out mac_rx_deliver() to DLS, DLS instead registers them to go directly via
342 * the direct callback registered with DLS, generally ip_input().
343 *
344 * HARDWARE RING POLLING
345 *
346 * GLDv3 devices with hardware rings generally deliver chains of messages
347 * (mblk_t chain) during the context of a single interrupt. However, interrupts
348 * are not the only way that these devices may be used. As part of implementing
349 * ring support, a GLDv3 device driver must have a way to disable the generation
350 * of that interrupt and allow for the operating system to poll on that ring.
351 *
352 * To implement this, every soft ring set has a worker thread and a polling
353 * thread. If a sufficient packet rate comes into the system, MAC will 'blank'
354 * (disable) interrupts on that specific ring and the polling thread will start
355 * consuming packets from the hardware device and deliver them to the soft ring
356 * set, where the worker thread will take over.
357 *
358 * Once the rate of packet intake drops down below a certain threshold, then
359 * polling on the hardware ring will be quiesced and interrupts will be
360 * re-enabled for the given ring. This effectively allows the system to shift
361 * how it handles a ring based on its load. At high packet rates, polling on the
362 * device as opposed to relying on interrupts can actually reduce overall system
363 * load due to the minimization of interrupt activity.
364 *
365 * Note the importance of each ring having its own interrupt source. The whole
366 * idea here is that we do not disable interrupts on the device as a whole, but
367 * rather each ring can be independently toggled.
368 *
369 * USE OF WORKER THREADS
370 *
371 * Both the soft ring set and individual soft rings have a worker thread
372 * associated with them that may be bound to a specific CPU in the system. Any
373 * such assignment will get reassessed as part of dynamic reconfiguration events
374 * in the system such as the onlining and offlining of CPUs and the creation of
375 * CPU partitions.
376 *
377 * In many cases, while in an interrupt, we try to deliver a frame all the way
378 * through the stack in the context of the interrupt itself. However, if the
379 * amount of queued frames has exceeded a threshold, then we instead defer to
380 * the worker thread to do this work and signal it. This is particularly useful
381 * when you have the soft ring set delivering frames into multiple software
382 * rings. If it was only delivering frames into a single software ring then
383 * there'd be no need to have another thread take over. However, if it's
384 * delivering chains of frames to multiple rings, then it's worthwhile to have
385 * the worker for the software ring take over so that the different software
386 * rings can be processed in parallel.
387 *
388 * In a similar fashion to the hardware polling thread, if we don't have a
389 * backlog or there's nothing to do, then the worker thread will go back to
390 * sleep and frames can be delivered all the way from an interrupt. This
391 * behavior is useful as it's designed to minimize latency and the default
392 * disposition of MAC is to optimize for latency.
393 *
394 * MAINTAINING CHAINS
395 *
396 * Another useful idea that MAC uses is to try and maintain frames in chains for
397 * as long as possible. The idea is that all of MAC can handle chains of frames
398 * structured as a series of mblk_t structures linked with the b_next pointer.
399 * When performing software classification and software fanout, MAC does not
400 * simply determine the destination and send the frame along. Instead, in the
401 * case of classification, it tries to maintain a chain for as long as possible
402 * before passing it along and performing additional processing.
403 *
404 * In the case of fanout, MAC first determines what the target software ring is
405 * for every frame in the original chain and constructs a new chain for each
406 * target. MAC then delivers the new chain to each software ring in succession.
407 *
408 * The whole rationale for doing this is that we want to try and maintain the
409 * pipe as much as possible and deliver as many frames through the stack at once
410 * that we can, rather than just pushing a single frame through. This can often
411 * help bring down latency and allows MAC to get a better sense of the overall
412 * activity in the system and properly engage worker threads.
413 *
414 * --------------------
415 * Bandwidth Management
416 * --------------------
417 *
418 * Bandwidth management is something that's built into the soft ring set itself.
419 * When bandwidth limits are placed on a flow, a corresponding soft ring set is
420 * toggled into bandwidth mode. This changes how we transmit and receive the
421 * frames in question.
422 *
423 * Bandwidth management is done on a per-tick basis. We translate the user's
424 * requested bandwidth from a quantity per-second into a quantity per-tick. MAC
425 * cannot process a frame across more than one tick, thus it sets a lower bound
426 * for the bandwidth cap to be a single MTU. This also means that when
427 * hires ticks are enabled (hz is set to 1000), that the minimum amount of
428 * bandwidth is higher, because the number of ticks has increased and MAC has to
429 * go from accepting 100 packets / sec to 1000 / sec.
430 *
431 * The bandwidth counter is reset by either the soft ring set's worker thread or
432 * a thread that is doing an inline transmit or receive if they discover that
433 * the current tick is in the future from the recorded tick.
434 *
435 * Whenever we're receiving or transmitting data, we end up leaving most of the
436 * work to the soft ring set's worker thread. This forces data inserted into the
437 * soft ring set to be effectively serialized and allows us to exhume bandwidth
438 * at a reasonable rate. If there is nothing in the soft ring set at the moment
439 * and the set has available bandwidth, then it may processed inline.
440 * Otherwise, the worker is responsible for taking care of the soft ring set.
441 *
442 * ---------------------
443 * The Receive Data Path
444 * ---------------------
445 *
446 * The following series of ASCII art images breaks apart the way that a frame
447 * comes in and is processed in MAC.
448 *
449 * Part 1 -- Initial frame receipt, SRS classification
450 *
451 * Here, a frame is received by a GLDv3 driver, generally in the context of an
452 * interrupt, and it ends up in mac_rx_common(). A driver calls either mac_rx or
453 * mac_rx_ring, depending on whether or not it supports rings and can identify
454 * the interrupt as having come from a specific ring. Here we determine whether
455 * or not it's fully classified and perform software classification as
456 * appropriate. From here, everything always ends up going to either entry [A]
457 * or entry [B] based on whether or not they have subflow processing needed. We
458 * leave via fanout or delivery.
459 *
460 * +===========+
461 * v hardware v
462 * v interrupt v
463 * +===========+
464 * |
465 * * . . appropriate
466 * | upcall made
467 * | by GLDv3 driver . . always
468 * | .
469 * +--------+ | +----------+ . +---------------+
470 * | GLDv3 | +---->| mac_rx |-----*--->| mac_rx_common |
471 * | Driver |-->--+ +----------+ +---------------+
472 * +--------+ | ^ |
473 * | | ^ v
474 * ^ | * . . always +----------------------+
475 * | | | | mac_promisc_dispatch |
476 * | | +-------------+ +----------------------+
477 * | +--->| mac_rx_ring | |
478 * | +-------------+ * . . hw classified
479 * | v or single flow?
480 * | |
481 * | +--------++--------------+
482 * | | | * hw class,
483 * | | * hw classified | subflows
484 * | no hw class and . * | or single | exist
485 * | subflows | | flow |
486 * | | v v
487 * | | +-----------+ +-----------+
488 * | | | goto | | goto |
489 * | | | entry [A] | | entry [B] |
490 * | | +-----------+ +-----------+
491 * | v ^
492 * | +-------------+ |
493 * | | mac_rx_flow | * SRS and flow found,
494 * | +-------------+ | call flow cb
495 * | | +------+
496 * | v |
497 * v +==========+ +-----------------+
498 * | v For each v--->| mac_rx_classify |
499 * +----------+ v mblk_t v +-----------------+
500 * | srs | +==========+
501 * | pollling |
502 * | thread |->------------------------------------------+
503 * +----------+ |
504 * v . inline
505 * +--------------------+ +----------+ +---------+ .
506 * [A]---->| mac_rx_srs_process |-->| check bw |-->| enqueue |--*---------+
507 * +--------------------+ | limits | | frames | |
508 * ^ +----------+ | to SRS | |
509 * | +---------+ |
510 * | send chain +--------+ | |
511 * * when clasified | signal | * BW limits, |
512 * | flow changes | srs |<---+ loopback, |
513 * | | worker | stack too |
514 * | +--------+ deep |
515 * +-----------------+ +--------+ |
516 * | mac_flow_lookup | | srs | +---------------------+ |
517 * +-----------------+ | worker |---->| mac_rx_srs_drain |<---+
518 * ^ | thread | | mac_rx_srs_drain_bw |
519 * | +--------+ +---------------------+
520 * | |
521 * +----------------------------+ * software rings
522 * [B]-->| mac_rx_srs_subflow_process | | for fanout?
523 * +----------------------------+ |
524 * +----------+-----------+
525 * | |
526 * v v
527 * +--------+ +--------+
528 * | goto | | goto |
529 * | Part 2 | | Part 3 |
530 * +--------+ +--------+
531 *
532 * Part 2 -- Fanout
533 *
534 * This part is concerned with using software fanout to assign frames to
535 * software rings and then deliver them to MAC clients or allow those rings to
536 * be polled upon. While there are two different primary fanout entry points,
537 * mac_rx_fanout and mac_rx_proto_fanout, they behave in similar ways, and aside
538 * from some of the individual hashing techniques used, most of the general
539 * flow is the same.
540 *
541 * +--------+ +-------------------+
542 * | From |---+--------->| mac_rx_srs_fanout |----+
543 * | Part 1 | | +-------------------+ | +=================+
544 * +--------+ | | v for each mblk_t v
545 * * . . protocol only +--->v assign to new v
546 * | fanout | v chain based on v
547 * | | v hash % nrings v
548 * | +-------------------------+ | +=================+
549 * +--->| mac_rx_srs_proto_fanout |----+ |
550 * +-------------------------+ |
551 * v
552 * +------------+ +--------------------------+ +================+
553 * | enqueue in |<---| mac_rx_soft_ring_process |<------v for each chain v
554 * | soft ring | +--------------------------+ +================+
555 * +------------+
556 * | +-----------+
557 * * soft ring set | soft ring |
558 * | empty and no | worker |
559 * | worker? | thread |
560 * | +-----------+
561 * +------*----------------+ |
562 * | . | v
563 * No . * . Yes | +------------------------+
564 * | +----<--| mac_rx_soft_ring_drain |
565 * | | +------------------------+
566 * v |
567 * +-----------+ v
568 * | signal | +---------------+
569 * | soft ring | | Deliver chain |
570 * | worker | | goto Part 3 |
571 * +-----------+ +---------------+
572 *
573 *
574 * Part 3 -- Packet Delivery
575 *
576 * Here, we go through and deliver the mblk_t chain directly to a given
577 * processing function. In a lot of cases this is mac_rx_deliver(). In the case
578 * of DLS bypass being used, then instead we end up going ahead and deliver it
579 * to the direct callback registered with DLS, generally ip_input.
580 *
581 *
582 * +---------+ +----------------+ +------------------+
583 * | From |---+------->| mac_rx_deliver |--->| Off to DLS, or |
584 * | Parts 1 | | +----------------+ | other MAC client |
585 * | and 2 | * DLS bypass +------------------+
586 * +---------+ | enabled +----------+ +-------------+
587 * +---------->| ip_input |--->| To IP |
588 * +----------+ | and beyond! |
589 * +-------------+
590 *
591 * ----------------------
592 * The Transmit Data Path
593 * ----------------------
594 *
595 * Before we go into the images, it's worth talking about a problem that is a
596 * bit different from the receive data path. GLDv3 device drivers have a finite
597 * amount of transmit descriptors. When they run out, they return unused frames
598 * back to MAC. MAC, at this point has several options about what it will do,
599 * which vary based upon the settings that the client uses.
600 *
601 * When a device runs out of descriptors, the next thing that MAC does is
602 * enqueue them off of the soft ring set or a software ring, depending on the
603 * configuration of the soft ring set. MAC will enqueue up to a high watermark
604 * of mblk_t chains, at which point it will indicate flow control back to the
605 * client. Once this condition is reached, any mblk_t chains that were not
606 * enqueued will be returned to the caller and they will have to decide what to
607 * do with them. There are various flags that control this behavior that a
608 * client may pass, which are discussed below.
609 *
610 * When this condition is hit, MAC also returns a cookie to the client in
611 * addition to unconsumed frames. Clients can poll on that cookie and register a
612 * callback with MAC to be notified when they are no longer subject to flow
613 * control, at which point they may continue to call mac_tx(). This flow control
614 * actually manages to work itself all the way up the stack, back through dls,
615 * to ip, through the various protocols, and to sockfs.
616 *
617 * While the behavior described above is the default, this behavior can be
618 * modified. There are two alternate modes, described below, which are
619 * controlled with flags.
620 *
621 * DROP MODE
622 *
623 * This mode is controlled by having the client pass the MAC_DROP_ON_NO_DESC
624 * flag. When this is passed, if a device driver runs out of transmit
625 * descriptors, then the MAC layer will drop any unsent traffic. The client in
626 * this case will never have any frames returned to it.
627 *
628 * DON'T ENQUEUE
629 *
630 * This mode is controlled by having the client pass the MAC_TX_NO_ENQUEUE flag.
631 * If the MAC_DROP_ON_NO_DESC flag is also passed, it takes precedence. In this
632 * mode, when we hit a case where a driver runs out of transmit descriptors,
633 * then instead of enqueuing packets in a soft ring set or software ring, we
634 * instead return the mblk_t chain back to the caller and immediately put the
635 * soft ring set into flow control mode.
636 *
637 * The following series of ASCII art images describe the transmit data path that
638 * MAC clients enter into based on calling into mac_tx(). A soft ring set has a
639 * transmission function associated with it. There are seven possible
640 * transmission modes, some of which share function entry points. The one that a
641 * soft ring set gets depends on properties such as whether there are
642 * transmission rings for fanout, whether the device involves aggregations,
643 * whether any bandwidth limits exist, etc.
644 *
645 *
646 * Part 1 -- Initial checks
647 *
648 * * . called by
649 * | MAC clients
650 * v . . No
651 * +--------+ +-----------+ . +-------------------+ +====================+
652 * | mac_tx |->| device |-*-->| mac_protect_check |->v Is this the simple v
653 * +--------+ | quiesced? | +-------------------+ v case? See [1] v
654 * +-----------+ | +====================+
655 * * . Yes * failed |
656 * v | frames |
657 * +--------------+ | +-------+---------+
658 * | freemsgchain |<---------+ Yes . * No . *
659 * +--------------+ v v
660 * +-----------+ +--------+
661 * | goto | | goto |
662 * | Part 2 | | SRS TX |
663 * | Entry [A] | | func |
664 * +-----------+ +--------+
665 * | |
666 * | v
667 * | +--------+
668 * +---------->| return |
669 * | cookie |
670 * +--------+
671 *
672 * [1] The simple case refers to the SRS being configured with the
673 * SRS_TX_DEFAULT transmission mode, having a single mblk_t (not a chain), their
674 * being only a single active client, and not having a backlog in the srs.
675 *
676 *
677 * Part 2 -- The SRS transmission functions
678 *
679 * This part is a bit more complicated. The different transmission paths often
680 * leverage one another. In this case, we'll draw out the more common ones
681 * before the parts that depend upon them. Here, we're going to start with the
682 * workings of mac_tx_send() a common function that most of the others end up
683 * calling.
684 *
685 * +-------------+
686 * | mac_tx_send |
687 * +-------------+
688 * |
689 * v
690 * +=============+ +==============+
691 * v more than v--->v check v
692 * v one client? v v VLAN and add v
693 * +=============+ v VLAN tags v
694 * | +==============+
695 * | |
696 * +------------------+
697 * |
698 * | [A]
699 * v |
700 * +============+ . No v
701 * v more than v . +==========+ +--------------------------+
702 * v one active v-*---->v for each v---->| mac_promisc_dispatch_one |---+
703 * v client? v v mblk_t v +--------------------------+ |
704 * +============+ +==========+ ^ |
705 * | | +==========+ |
706 * * . Yes | v hardware v<-------+
707 * v +------------+ v rings? v
708 * +==========+ | +==========+
709 * v for each v No . . . * |
710 * v mblk_t v specific | |
711 * +==========+ flow | +-----+-----+
712 * | | | |
713 * v | v v
714 * +-----------------+ | +-------+ +---------+
715 * | mac_tx_classify |------------+ | GLDv3 | | GLDv3 |
716 * +-----------------+ |TX func| | ring tx |
717 * | +-------+ | func |
718 * * Specific flow, generally | +---------+
719 * | bcast, mcast, loopback | |
720 * v +-----+-----+
721 * +==========+ +---------+ |
722 * v valid L2 v--*--->| freemsg | v
723 * v header v . No +---------+ +-------------------+
724 * +==========+ | return unconsumed |
725 * * . Yes | frames to the |
726 * v | caller |
727 * +===========+ +-------------------+
728 * v braodcast v +----------------+ ^
729 * v flow? v--*-->| mac_bcast_send |------------------+
730 * +===========+ . +----------------+ |
731 * | . . Yes |
732 * No . * v
733 * | +---------------------+ +---------------+ +----------+
734 * +->|mac_promisc_dispatch |->| mac_fix_cksum |->| flow |
735 * +---------------------+ +---------------+ | callback |
736 * +----------+
737 *
738 *
739 * In addition, many but not all of the routines, all rely on
740 * mac_tx_softring_process as an entry point.
741 *
742 *
743 * . No . No
744 * +--------------------------+ +========+ . +===========+ . +-------------+
745 * | mac_tx_soft_ring_process |-->v worker v-*->v out of tx v-*->| goto |
746 * +--------------------------+ v only? v v descr.? v | mac_tx_send |
747 * +========+ +===========+ +-------------+
748 * Yes . * * . Yes |
749 * . No v | v
750 * v=========+ . +===========+ . Yes | Yes . +==========+
751 * v apppend v<--*----------v out of tx v-*-------+---------*--v returned v
752 * v mblk_t v v descr.? v | v frames? v
753 * v chain v +===========+ | +==========+
754 * +=========+ | *. No
755 * | | v
756 * v v +------------+
757 * +===================+ +----------------------+ | done |
758 * v worker scheduled? v | mac_tx_sring_enqueue | | processing |
759 * v Out of tx descr? v +----------------------+ +------------+
760 * +===================+ |
761 * | | . Yes v
762 * * Yes * No . +============+
763 * | v +-*---------v drop on no v
764 * | +========+ v v TX desc? v
765 * | v wake v +----------+ +============+
766 * | v worker v | mac_pkt_ | * . No
767 * | +========+ | drop | | . Yes . No
768 * | | +----------+ v . .
769 * | | v ^ +===============+ . +========+ .
770 * +--+--------+---------+ | v Don't enqueue v-*->v ring v-*----+
771 * | | v Set? v v empty? v |
772 * | +---------------+ +===============+ +========+ |
773 * | | | | |
774 * | | +-------------------+ | |
775 * | *. Yes | +---------+ |
776 * | | v v v
777 * | | +===========+ +========+ +--------------+
778 * | +<-v At hiwat? v v append v | return |
779 * | +===========+ v mblk_t v | mblk_t chain |
780 * | * No v chain v | and flow |
781 * | v +========+ | control |
782 * | +=========+ | | cookie |
783 * | v append v v +--------------+
784 * | v mblk_t v +========+
785 * | v chain v v wake v +------------+
786 * | +=========+ v worker v-->| done |
787 * | | +========+ | processing |
788 * | v .. Yes +------------+
789 * | +=========+ . +========+
790 * | v first v--*-->v wake v
791 * | v append? v v worker v
792 * | +=========+ +========+
793 * | | |
794 * | No . * |
795 * | v |
796 * | +--------------+ |
797 * +------>| Return | |
798 * | flow control |<------------+
799 * | cookie |
800 * +--------------+
801 *
802 *
803 * The remaining images are all specific to each of the different transmission
804 * modes.
805 *
806 * SRS TX DEFAULT
807 *
808 * [ From Part 1 ]
809 * |
810 * v
811 * +-------------------------+
812 * | mac_tx_single_ring_mode |
813 * +-------------------------+
814 * |
815 * | . Yes
816 * v .
817 * +==========+ . +============+
818 * v SRS v-*->v Try to v---->---------------------+
819 * v backlog? v v enqueue in v |
820 * +==========+ v SRS v-->------+ * . . Queue too
821 * | +============+ * don't enqueue | deep or
822 * * . No ^ | | flag or at | drop flag
823 * | | v | hiwat, |
824 * v | | | return +---------+
825 * +-------------+ | | | cookie | freemsg |
826 * | goto |-*-----+ | | +---------+
827 * | mac_tx_send | . returned | | |
828 * +-------------+ mblk_t | | |
829 * | | | |
830 * | | | |
831 * * . . all mblk_t * queued, | |
832 * v consumed | may return | |
833 * +-------------+ | tx cookie | |
834 * | SRS TX func |<------------+------------+----------------+
835 * | completed |
836 * +-------------+
837 *
838 * SRS_TX_SERIALIZE
839 *
840 * +------------------------+
841 * | mac_tx_serializer_mode |
842 * +------------------------+
843 * |
844 * | . No
845 * v .
846 * +============+ . +============+ +-------------+ +============+
847 * v srs being v-*->v set SRS v--->| goto |-->v remove SRS v
848 * v processed? v v proc flags v | mac_tx_send | v proc flag v
849 * +============+ +============+ +-------------+ +============+
850 * | |
851 * * Yes |
852 * v . No v
853 * +--------------------+ . +==========+
854 * | mac_tx_srs_enqueue | +------------------------*-----<--v returned v
855 * +--------------------+ | v frames? v
856 * | | . Yes +==========+
857 * | | . |
858 * | | . +=========+ v
859 * v +-<-*-v queued v +--------------------+
860 * +-------------+ | v frames? v<----| mac_tx_srs_enqueue |
861 * | SRS TX func | | +=========+ +--------------------+
862 * | completed, |<------+ * . Yes
863 * | may return | | v
864 * | cookie | | +========+
865 * +-------------+ +-<---v wake v
866 * v worker v
867 * +========+
868 *
869 *
870 * SRS_TX_FANOUT
871 *
872 * . Yes
873 * +--------------------+ +=============+ . +--------------------------+
874 * | mac_tx_fanout_mode |--->v Have fanout v-*-->| goto |
875 * +--------------------+ v hint? v | mac_rx_soft_ring_process |
876 * +=============+ +--------------------------+
877 * * . No |
878 * v ^
879 * +===========+ |
880 * +--->v for each v +===============+
881 * | v mblk_t v v pick softring v
882 * same * +===========+ v from hash v
883 * hash | | +===============+
884 * | v |
885 * | +--------------+ |
886 * +---| mac_pkt_hash |--->*------------+
887 * +--------------+ . different
888 * hash or
889 * done proc.
890 * SRS_TX_AGGR chain
891 *
892 * +------------------+ +================================+
893 * | mac_tx_aggr_mode |--->v Use aggr capab function to v
894 * +------------------+ v find appropriate tx ring. v
895 * v Applies hash based on aggr v
896 * v policy, see mac_tx_aggr_mode() v
897 * +================================+
898 * |
899 * v
900 * +-------------------------------+
901 * | goto |
902 * | mac_rx_srs_soft_ring_process |
903 * +-------------------------------+
904 *
905 *
906 * SRS_TX_BW, SRS_TX_BW_FANOUT, SRS_TX_BW_AGGR
907 *
908 * Note, all three of these tx functions start from the same place --
909 * mac_tx_bw_mode().
910 *
911 * +----------------+
912 * | mac_tx_bw_mode |
913 * +----------------+
914 * |
915 * v . No . No . Yes
916 * +==============+ . +============+ . +=============+ . +=========+
917 * v Out of BW? v--*->v SRS empty? v--*->v reset BW v-*->v Bump BW v
918 * +==============+ +============+ v tick count? v v Usage v
919 * | | +=============+ +=========+
920 * | +---------+ | |
921 * | | +--------------------+ |
922 * | | | +----------------------+
923 * v | v v
924 * +===============+ | +==========+ +==========+ +------------------+
925 * v Don't enqueue v | v set bw v v Is aggr? v--*-->| goto |
926 * v flag set? v | v enforced v +==========+ . | mac_tx_aggr_mode |-+
927 * +===============+ | +==========+ | . +------------------+ |
928 * | Yes .* | | No . * . |
929 * | | | | | . Yes |
930 * * . No | | v | |
931 * | +---------+ | +========+ v +======+ |
932 * | | freemsg | | v append v +============+ . Yes v pick v |
933 * | +---------+ | v mblk_t v v Is fanout? v--*---->v ring v |
934 * | | | v chain v +============+ +======+ |
935 * +------+ | +========+ | | |
936 * v | | v v |
937 * +---------+ | v +-------------+ +--------------------+ |
938 * | return | | +========+ | goto | | goto | |
939 * | flow | | v wakeup v | mac_tx_send | | mac_tx_fanout_mode | |
940 * | control | | v worker v +-------------+ +--------------------+ |
941 * | cookie | | +========+ | | |
942 * +---------+ | | | +------+------+
943 * | v | |
944 * | +---------+ | v
945 * | | return | +============+ +------------+
946 * | | flow | v unconsumed v-------+ | done |
947 * | | control | v frames? v | | processing |
948 * | | cookie | +============+ | +------------+
949 * | +---------+ | |
950 * | Yes * |
951 * | | |
952 * | +===========+ |
953 * | v subtract v |
954 * | v unused bw v |
955 * | +===========+ |
956 * | | |
957 * | v |
958 * | +--------------------+ |
959 * +------------->| mac_tx_srs_enqueue | |
960 * +--------------------+ |
961 * | |
962 * | |
963 * +------------+ |
964 * | return fc | |
965 * | cookie and |<------+
966 * | mblk_t |
967 * +------------+
968 */
969
970 #include <sys/types.h>
971 #include <sys/callb.h>
972 #include <sys/pattr.h>
973 #include <sys/sdt.h>
974 #include <sys/strsubr.h>
975 #include <sys/strsun.h>
976 #include <sys/vlan.h>
977 #include <sys/stack.h>
978 #include <sys/archsystm.h>
979 #include <inet/ipsec_impl.h>
980 #include <inet/ip_impl.h>
981 #include <inet/sadb.h>
982 #include <inet/ipsecesp.h>
983 #include <inet/ipsecah.h>
984 #include <inet/ip6.h>
985
986 #include <sys/mac_impl.h>
987 #include <sys/mac_client_impl.h>
988 #include <sys/mac_client_priv.h>
989 #include <sys/mac_soft_ring.h>
990 #include <sys/mac_flow_impl.h>
991
992 static mac_tx_cookie_t mac_tx_single_ring_mode(mac_soft_ring_set_t *, mblk_t *,
993 uintptr_t, uint16_t, mblk_t **);
994 static mac_tx_cookie_t mac_tx_serializer_mode(mac_soft_ring_set_t *, mblk_t *,
995 uintptr_t, uint16_t, mblk_t **);
996 static mac_tx_cookie_t mac_tx_fanout_mode(mac_soft_ring_set_t *, mblk_t *,
997 uintptr_t, uint16_t, mblk_t **);
998 static mac_tx_cookie_t mac_tx_bw_mode(mac_soft_ring_set_t *, mblk_t *,
999 uintptr_t, uint16_t, mblk_t **);
1000 static mac_tx_cookie_t mac_tx_aggr_mode(mac_soft_ring_set_t *, mblk_t *,
1001 uintptr_t, uint16_t, mblk_t **);
1002
1003 typedef struct mac_tx_mode_s {
1004 mac_tx_srs_mode_t mac_tx_mode;
1005 mac_tx_func_t mac_tx_func;
1006 } mac_tx_mode_t;
1007
1008 /*
1009 * There are seven modes of operation on the Tx side. These modes get set
1010 * in mac_tx_srs_setup(). Except for the experimental TX_SERIALIZE mode,
1011 * none of the other modes are user configurable. They get selected by
1012 * the system depending upon whether the link (or flow) has multiple Tx
1013 * rings or a bandwidth configured, or if the link is an aggr, etc.
1014 *
1015 * When the Tx SRS is operating in aggr mode (st_mode) or if there are
1016 * multiple Tx rings owned by Tx SRS, then each Tx ring (pseudo or
1017 * otherwise) will have a soft ring associated with it. These soft rings
1018 * are stored in srs_tx_soft_rings[] array.
1019 *
1020 * Additionally in the case of aggr, there is the st_soft_rings[] array
1021 * in the mac_srs_tx_t structure. This array is used to store the same
1022 * set of soft rings that are present in srs_tx_soft_rings[] array but
1023 * in a different manner. The soft ring associated with the pseudo Tx
1024 * ring is saved at mr_index (of the pseudo ring) in st_soft_rings[]
1025 * array. This helps in quickly getting the soft ring associated with the
1026 * Tx ring when aggr_find_tx_ring() returns the pseudo Tx ring that is to
1027 * be used for transmit.
1028 */
1029 mac_tx_mode_t mac_tx_mode_list[] = {
1030 {SRS_TX_DEFAULT, mac_tx_single_ring_mode},
1031 {SRS_TX_SERIALIZE, mac_tx_serializer_mode},
1032 {SRS_TX_FANOUT, mac_tx_fanout_mode},
1033 {SRS_TX_BW, mac_tx_bw_mode},
1034 {SRS_TX_BW_FANOUT, mac_tx_bw_mode},
1035 {SRS_TX_AGGR, mac_tx_aggr_mode},
1036 {SRS_TX_BW_AGGR, mac_tx_bw_mode}
1037 };
1038
1039 /*
1040 * Soft Ring Set (SRS) - The Run time code that deals with
1041 * dynamic polling from the hardware, bandwidth enforcement,
1042 * fanout etc.
1043 *
1044 * We try to use H/W classification on NIC and assign traffic for
1045 * a MAC address to a particular Rx ring or ring group. There is a
1046 * 1-1 mapping between a SRS and a Rx ring. The SRS dynamically
1047 * switches the underlying Rx ring between interrupt and
1048 * polling mode and enforces any specified B/W control.
1049 *
1050 * There is always a SRS created and tied to each H/W and S/W rule.
1051 * Whenever we create a H/W rule, we always add the the same rule to
1052 * S/W classifier and tie a SRS to it.
1053 *
1054 * In case a B/W control is specified, it is broken into bytes
1055 * per ticks and as soon as the quota for a tick is exhausted,
1056 * the underlying Rx ring is forced into poll mode for remainder of
1057 * the tick. The SRS poll thread only polls for bytes that are
1058 * allowed to come in the SRS. We typically let 4x the configured
1059 * B/W worth of packets to come in the SRS (to prevent unnecessary
1060 * drops due to bursts) but only process the specified amount.
1061 *
1062 * A MAC client (e.g. a VNIC or aggr) can have 1 or more
1063 * Rx rings (and corresponding SRSs) assigned to it. The SRS
1064 * in turn can have softrings to do protocol level fanout or
1065 * softrings to do S/W based fanout or both. In case the NIC
1066 * has no Rx rings, we do S/W classification to respective SRS.
1067 * The S/W classification rule is always setup and ready. This
1068 * allows the MAC layer to reassign Rx rings whenever needed
1069 * but packets still continue to flow via the default path and
1070 * getting S/W classified to correct SRS.
1071 *
1072 * The SRS's are used on both Tx and Rx side. They use the same
1073 * data structure but the processing routines have slightly different
1074 * semantics due to the fact that Rx side needs to do dynamic
1075 * polling etc.
1076 *
1077 * Dynamic Polling Notes
1078 * =====================
1079 *
1080 * Each Soft ring set is capable of switching its Rx ring between
1081 * interrupt and poll mode and actively 'polls' for packets in
1082 * poll mode. If the SRS is implementing a B/W limit, it makes
1083 * sure that only Max allowed packets are pulled in poll mode
1084 * and goes to poll mode as soon as B/W limit is exceeded. As
1085 * such, there are no overheads to implement B/W limits.
1086 *
1087 * In poll mode, its better to keep the pipeline going where the
1088 * SRS worker thread keeps processing packets and poll thread
1089 * keeps bringing more packets (specially if they get to run
1090 * on different CPUs). This also prevents the overheads associated
1091 * by excessive signalling (on NUMA machines, this can be
1092 * pretty devastating). The exception is latency optimized case
1093 * where worker thread does no work and interrupt and poll thread
1094 * are allowed to do their own drain.
1095 *
1096 * We use the following policy to control Dynamic Polling:
1097 * 1) We switch to poll mode anytime the processing
1098 * thread causes a backlog to build up in SRS and
1099 * its associated Soft Rings (sr_poll_pkt_cnt > 0).
1100 * 2) As long as the backlog stays under the low water
1101 * mark (sr_lowat), we poll the H/W for more packets.
1102 * 3) If the backlog (sr_poll_pkt_cnt) exceeds low
1103 * water mark, we stay in poll mode but don't poll
1104 * the H/W for more packets.
1105 * 4) Anytime in polling mode, if we poll the H/W for
1106 * packets and find nothing plus we have an existing
1107 * backlog (sr_poll_pkt_cnt > 0), we stay in polling
1108 * mode but don't poll the H/W for packets anymore
1109 * (let the polling thread go to sleep).
1110 * 5) Once the backlog is relived (packets are processed)
1111 * we reenable polling (by signalling the poll thread)
1112 * only when the backlog dips below sr_poll_thres.
1113 * 6) sr_hiwat is used exclusively when we are not
1114 * polling capable and is used to decide when to
1115 * drop packets so the SRS queue length doesn't grow
1116 * infinitely.
1117 *
1118 * NOTE: Also see the block level comment on top of mac_soft_ring.c
1119 */
1120
1121 /*
1122 * mac_latency_optimize
1123 *
1124 * Controls whether the poll thread can process the packets inline
1125 * or let the SRS worker thread do the processing. This applies if
1126 * the SRS was not being processed. For latency sensitive traffic,
1127 * this needs to be true to allow inline processing. For throughput
1128 * under load, this should be false.
1129 *
1130 * This (and other similar) tunable should be rolled into a link
1131 * or flow specific workload hint that can be set using dladm
1132 * linkprop (instead of multiple such tunables).
1133 */
1134 boolean_t mac_latency_optimize = B_TRUE;
1135
1136 /*
1137 * MAC_RX_SRS_ENQUEUE_CHAIN and MAC_TX_SRS_ENQUEUE_CHAIN
1138 *
1139 * queue a mp or chain in soft ring set and increment the
1140 * local count (srs_count) for the SRS and the shared counter
1141 * (srs_poll_pkt_cnt - shared between SRS and its soft rings
1142 * to track the total unprocessed packets for polling to work
1143 * correctly).
1144 *
1145 * The size (total bytes queued) counters are incremented only
1146 * if we are doing B/W control.
1147 */
1148 #define MAC_SRS_ENQUEUE_CHAIN(mac_srs, head, tail, count, sz) { \
1149 ASSERT(MUTEX_HELD(&(mac_srs)->srs_lock)); \
1150 if ((mac_srs)->srs_last != NULL) \
1151 (mac_srs)->srs_last->b_next = (head); \
1152 else \
1153 (mac_srs)->srs_first = (head); \
1154 (mac_srs)->srs_last = (tail); \
1155 (mac_srs)->srs_count += count; \
1156 }
1157
1158 #define MAC_RX_SRS_ENQUEUE_CHAIN(mac_srs, head, tail, count, sz) { \
1159 mac_srs_rx_t *srs_rx = &(mac_srs)->srs_rx; \
1160 \
1161 MAC_SRS_ENQUEUE_CHAIN(mac_srs, head, tail, count, sz); \
1162 srs_rx->sr_poll_pkt_cnt += count; \
1163 ASSERT(srs_rx->sr_poll_pkt_cnt > 0); \
1164 if ((mac_srs)->srs_type & SRST_BW_CONTROL) { \
1165 (mac_srs)->srs_size += (sz); \
1166 mutex_enter(&(mac_srs)->srs_bw->mac_bw_lock); \
1167 (mac_srs)->srs_bw->mac_bw_sz += (sz); \
1168 mutex_exit(&(mac_srs)->srs_bw->mac_bw_lock); \
1169 } \
1170 }
1171
1172 #define MAC_TX_SRS_ENQUEUE_CHAIN(mac_srs, head, tail, count, sz) { \
1173 mac_srs->srs_state |= SRS_ENQUEUED; \
1174 MAC_SRS_ENQUEUE_CHAIN(mac_srs, head, tail, count, sz); \
1175 if ((mac_srs)->srs_type & SRST_BW_CONTROL) { \
1176 (mac_srs)->srs_size += (sz); \
1177 (mac_srs)->srs_bw->mac_bw_sz += (sz); \
1178 } \
1179 }
1180
1181 /*
1182 * Turn polling on routines
1183 */
1184 #define MAC_SRS_POLLING_ON(mac_srs) { \
1185 ASSERT(MUTEX_HELD(&(mac_srs)->srs_lock)); \
1186 if (((mac_srs)->srs_state & \
1187 (SRS_POLLING_CAPAB|SRS_POLLING)) == SRS_POLLING_CAPAB) { \
1188 (mac_srs)->srs_state |= SRS_POLLING; \
1189 (void) mac_hwring_disable_intr((mac_ring_handle_t) \
1190 (mac_srs)->srs_ring); \
1191 (mac_srs)->srs_rx.sr_poll_on++; \
1192 } \
1193 }
1194
1195 #define MAC_SRS_WORKER_POLLING_ON(mac_srs) { \
1196 ASSERT(MUTEX_HELD(&(mac_srs)->srs_lock)); \
1197 if (((mac_srs)->srs_state & \
1198 (SRS_POLLING_CAPAB|SRS_WORKER|SRS_POLLING)) == \
1199 (SRS_POLLING_CAPAB|SRS_WORKER)) { \
1200 (mac_srs)->srs_state |= SRS_POLLING; \
1201 (void) mac_hwring_disable_intr((mac_ring_handle_t) \
1202 (mac_srs)->srs_ring); \
1203 (mac_srs)->srs_rx.sr_worker_poll_on++; \
1204 } \
1205 }
1206
1207 /*
1208 * MAC_SRS_POLL_RING
1209 *
1210 * Signal the SRS poll thread to poll the underlying H/W ring
1211 * provided it wasn't already polling (SRS_GET_PKTS was set).
1212 *
1213 * Poll thread gets to run only from mac_rx_srs_drain() and only
1214 * if the drain was being done by the worker thread.
1215 */
1216 #define MAC_SRS_POLL_RING(mac_srs) { \
1217 mac_srs_rx_t *srs_rx = &(mac_srs)->srs_rx; \
1218 \
1219 ASSERT(MUTEX_HELD(&(mac_srs)->srs_lock)); \
1220 srs_rx->sr_poll_thr_sig++; \
1221 if (((mac_srs)->srs_state & \
1222 (SRS_POLLING_CAPAB|SRS_WORKER|SRS_GET_PKTS)) == \
1223 (SRS_WORKER|SRS_POLLING_CAPAB)) { \
1224 (mac_srs)->srs_state |= SRS_GET_PKTS; \
1225 cv_signal(&(mac_srs)->srs_cv); \
1226 } else { \
1227 srs_rx->sr_poll_thr_busy++; \
1228 } \
1229 }
1230
1231 /*
1232 * MAC_SRS_CHECK_BW_CONTROL
1233 *
1234 * Check to see if next tick has started so we can reset the
1235 * SRS_BW_ENFORCED flag and allow more packets to come in the
1236 * system.
1237 */
1238 #define MAC_SRS_CHECK_BW_CONTROL(mac_srs) { \
1239 ASSERT(MUTEX_HELD(&(mac_srs)->srs_lock)); \
1240 ASSERT(((mac_srs)->srs_type & SRST_TX) || \
1241 MUTEX_HELD(&(mac_srs)->srs_bw->mac_bw_lock)); \
1242 clock_t now = ddi_get_lbolt(); \
1243 if ((mac_srs)->srs_bw->mac_bw_curr_time != now) { \
1244 (mac_srs)->srs_bw->mac_bw_curr_time = now; \
1245 (mac_srs)->srs_bw->mac_bw_used = 0; \
1246 if ((mac_srs)->srs_bw->mac_bw_state & SRS_BW_ENFORCED) \
1247 (mac_srs)->srs_bw->mac_bw_state &= ~SRS_BW_ENFORCED; \
1248 } \
1249 }
1250
1251 /*
1252 * MAC_SRS_WORKER_WAKEUP
1253 *
1254 * Wake up the SRS worker thread to process the queue as long as
1255 * no one else is processing the queue. If we are optimizing for
1256 * latency, we wake up the worker thread immediately or else we
1257 * wait mac_srs_worker_wakeup_ticks before worker thread gets
1258 * woken up.
1259 */
1260 int mac_srs_worker_wakeup_ticks = 0;
1261 #define MAC_SRS_WORKER_WAKEUP(mac_srs) { \
1262 ASSERT(MUTEX_HELD(&(mac_srs)->srs_lock)); \
1263 if (!((mac_srs)->srs_state & SRS_PROC) && \
1264 (mac_srs)->srs_tid == NULL) { \
1265 if (((mac_srs)->srs_state & SRS_LATENCY_OPT) || \
1266 (mac_srs_worker_wakeup_ticks == 0)) \
1267 cv_signal(&(mac_srs)->srs_async); \
1268 else \
1269 (mac_srs)->srs_tid = \
1270 timeout(mac_srs_fire, (mac_srs), \
1271 mac_srs_worker_wakeup_ticks); \
1272 } \
1273 }
1274
1275 #define TX_BANDWIDTH_MODE(mac_srs) \
1276 ((mac_srs)->srs_tx.st_mode == SRS_TX_BW || \
1277 (mac_srs)->srs_tx.st_mode == SRS_TX_BW_FANOUT || \
1278 (mac_srs)->srs_tx.st_mode == SRS_TX_BW_AGGR)
1279
1280 #define TX_SRS_TO_SOFT_RING(mac_srs, head, hint) { \
1281 if (tx_mode == SRS_TX_BW_FANOUT) \
1282 (void) mac_tx_fanout_mode(mac_srs, head, hint, 0, NULL);\
1283 else \
1284 (void) mac_tx_aggr_mode(mac_srs, head, hint, 0, NULL); \
1285 }
1286
1287 /*
1288 * MAC_TX_SRS_BLOCK
1289 *
1290 * Always called from mac_tx_srs_drain() function. SRS_TX_BLOCKED
1291 * will be set only if srs_tx_woken_up is FALSE. If
1292 * srs_tx_woken_up is TRUE, it indicates that the wakeup arrived
1293 * before we grabbed srs_lock to set SRS_TX_BLOCKED. We need to
1294 * attempt to transmit again and not setting SRS_TX_BLOCKED does
1295 * that.
1296 */
1297 #define MAC_TX_SRS_BLOCK(srs, mp) { \
1298 ASSERT(MUTEX_HELD(&(srs)->srs_lock)); \
1299 if ((srs)->srs_tx.st_woken_up) { \
1300 (srs)->srs_tx.st_woken_up = B_FALSE; \
1301 } else { \
1302 ASSERT(!((srs)->srs_state & SRS_TX_BLOCKED)); \
1303 (srs)->srs_state |= SRS_TX_BLOCKED; \
1304 (srs)->srs_tx.st_stat.mts_blockcnt++; \
1305 } \
1306 }
1307
1308 /*
1309 * MAC_TX_SRS_TEST_HIWAT
1310 *
1311 * Called before queueing a packet onto Tx SRS to test and set
1312 * SRS_TX_HIWAT if srs_count exceeds srs_tx_hiwat.
1313 */
1314 #define MAC_TX_SRS_TEST_HIWAT(srs, mp, tail, cnt, sz, cookie) { \
1315 boolean_t enqueue = 1; \
1316 \
1317 if ((srs)->srs_count > (srs)->srs_tx.st_hiwat) { \
1318 /* \
1319 * flow-controlled. Store srs in cookie so that it \
1320 * can be returned as mac_tx_cookie_t to client \
1321 */ \
1322 (srs)->srs_state |= SRS_TX_HIWAT; \
1323 cookie = (mac_tx_cookie_t)srs; \
1324 (srs)->srs_tx.st_hiwat_cnt++; \
1325 if ((srs)->srs_count > (srs)->srs_tx.st_max_q_cnt) { \
1326 /* increment freed stats */ \
1327 (srs)->srs_tx.st_stat.mts_sdrops += cnt; \
1328 /* \
1329 * b_prev may be set to the fanout hint \
1330 * hence can't use freemsg directly \
1331 */ \
1332 mac_drop_chain(mp_chain, "SRS Tx max queue"); \
1333 DTRACE_PROBE1(tx_queued_hiwat, \
1334 mac_soft_ring_set_t *, srs); \
1335 enqueue = 0; \
1336 } \
1337 } \
1338 if (enqueue) \
1339 MAC_TX_SRS_ENQUEUE_CHAIN(srs, mp, tail, cnt, sz); \
1340 }
1341
1342 /* Some utility macros */
1343 #define MAC_SRS_BW_LOCK(srs) \
1344 if (!(srs->srs_type & SRST_TX)) \
1345 mutex_enter(&srs->srs_bw->mac_bw_lock);
1346
1347 #define MAC_SRS_BW_UNLOCK(srs) \
1348 if (!(srs->srs_type & SRST_TX)) \
1349 mutex_exit(&srs->srs_bw->mac_bw_lock);
1350
1351 #define MAC_TX_SRS_DROP_MESSAGE(srs, chain, cookie, s) { \
1352 mac_drop_chain((chain), (s)); \
1353 /* increment freed stats */ \
1354 (srs)->srs_tx.st_stat.mts_sdrops++; \
1355 (cookie) = (mac_tx_cookie_t)(srs); \
1356 }
1357
1358 #define MAC_TX_SET_NO_ENQUEUE(srs, mp_chain, ret_mp, cookie) { \
1359 mac_srs->srs_state |= SRS_TX_WAKEUP_CLIENT; \
1360 cookie = (mac_tx_cookie_t)srs; \
1361 *ret_mp = mp_chain; \
1362 }
1363
1364 /*
1365 * Threshold used in receive-side processing to determine if handling
1366 * can occur in situ (in the interrupt thread) or if it should be left to a
1367 * worker thread. Note that the constant used to make this determination is
1368 * not entirely made-up, and is a result of some emprical validation. That
1369 * said, the constant is left as a global variable to allow it to be
1370 * dynamically tuned in the field if and as needed.
1371 */
1372 uintptr_t mac_rx_srs_stack_needed = 14336;
1373 uint_t mac_rx_srs_stack_toodeep;
1374
1375 #ifndef STACK_GROWTH_DOWN
1376 #error Downward stack growth assumed.
1377 #endif
1378
1379 /*
1380 * Drop the rx packet and advance to the next one in the chain.
1381 */
1382 static void
mac_rx_drop_pkt(mac_soft_ring_set_t * srs,mblk_t * mp)1383 mac_rx_drop_pkt(mac_soft_ring_set_t *srs, mblk_t *mp)
1384 {
1385 mac_srs_rx_t *srs_rx = &srs->srs_rx;
1386
1387 ASSERT(mp->b_next == NULL);
1388 mutex_enter(&srs->srs_lock);
1389 MAC_UPDATE_SRS_COUNT_LOCKED(srs, 1);
1390 MAC_UPDATE_SRS_SIZE_LOCKED(srs, msgdsize(mp));
1391 mutex_exit(&srs->srs_lock);
1392
1393 srs_rx->sr_stat.mrs_sdrops++;
1394 freemsg(mp);
1395 }
1396
1397 /* DATAPATH RUNTIME ROUTINES */
1398
1399 /*
1400 * mac_srs_fire
1401 *
1402 * Timer callback routine for waking up the SRS worker thread.
1403 */
1404 static void
mac_srs_fire(void * arg)1405 mac_srs_fire(void *arg)
1406 {
1407 mac_soft_ring_set_t *mac_srs = (mac_soft_ring_set_t *)arg;
1408
1409 mutex_enter(&mac_srs->srs_lock);
1410 if (mac_srs->srs_tid == NULL) {
1411 mutex_exit(&mac_srs->srs_lock);
1412 return;
1413 }
1414
1415 mac_srs->srs_tid = NULL;
1416 if (!(mac_srs->srs_state & SRS_PROC))
1417 cv_signal(&mac_srs->srs_async);
1418
1419 mutex_exit(&mac_srs->srs_lock);
1420 }
1421
1422 /*
1423 * 'hint' is fanout_hint (type of uint64_t) which is given by the TCP/IP stack,
1424 * and it is used on the TX path.
1425 */
1426 #define HASH_HINT(hint) \
1427 ((hint) ^ ((hint) >> 24) ^ ((hint) >> 16) ^ ((hint) >> 8))
1428
1429
1430 /*
1431 * hash based on the src address, dst address and the port information.
1432 */
1433 #define HASH_ADDR(src, dst, ports) \
1434 (ntohl((src) + (dst)) ^ ((ports) >> 24) ^ ((ports) >> 16) ^ \
1435 ((ports) >> 8) ^ (ports))
1436
1437 #define COMPUTE_INDEX(key, sz) (key % sz)
1438
1439 #define FANOUT_ENQUEUE_MP(head, tail, cnt, bw_ctl, sz, sz0, mp) { \
1440 if ((tail) != NULL) { \
1441 ASSERT((tail)->b_next == NULL); \
1442 (tail)->b_next = (mp); \
1443 } else { \
1444 ASSERT((head) == NULL); \
1445 (head) = (mp); \
1446 } \
1447 (tail) = (mp); \
1448 (cnt)++; \
1449 if ((bw_ctl)) \
1450 (sz) += (sz0); \
1451 }
1452
1453 #define MAC_FANOUT_DEFAULT 0
1454 #define MAC_FANOUT_RND_ROBIN 1
1455 int mac_fanout_type = MAC_FANOUT_DEFAULT;
1456
1457 #define MAX_SR_TYPES 3
1458 /* fanout types for port based hashing */
1459 enum pkt_type {
1460 V4_TCP = 0,
1461 V4_UDP,
1462 OTH,
1463 UNDEF
1464 };
1465
1466 /*
1467 * Pair of local and remote ports in the transport header
1468 */
1469 #define PORTS_SIZE 4
1470
1471 /*
1472 * This routine delivers packets destined for an SRS into one of the
1473 * protocol soft rings.
1474 *
1475 * Given a chain of packets we need to split it up into multiple sub
1476 * chains: TCP, UDP or OTH soft ring. Instead of entering the soft
1477 * ring one packet at a time, we want to enter it in the form of a
1478 * chain otherwise we get this start/stop behaviour where the worker
1479 * thread goes to sleep and then next packet comes in forcing it to
1480 * wake up.
1481 */
1482 static void
mac_rx_srs_proto_fanout(mac_soft_ring_set_t * mac_srs,mblk_t * head)1483 mac_rx_srs_proto_fanout(mac_soft_ring_set_t *mac_srs, mblk_t *head)
1484 {
1485 struct ether_header *ehp;
1486 struct ether_vlan_header *evhp;
1487 uint32_t sap;
1488 ipha_t *ipha;
1489 uint8_t *dstaddr;
1490 size_t hdrsize;
1491 mblk_t *mp;
1492 mblk_t *headmp[MAX_SR_TYPES];
1493 mblk_t *tailmp[MAX_SR_TYPES];
1494 int cnt[MAX_SR_TYPES];
1495 size_t sz[MAX_SR_TYPES];
1496 size_t sz1;
1497 boolean_t bw_ctl;
1498 boolean_t hw_classified;
1499 boolean_t dls_bypass;
1500 boolean_t is_ether;
1501 boolean_t is_unicast;
1502 enum pkt_type type;
1503 mac_client_impl_t *mcip = mac_srs->srs_mcip;
1504
1505 is_ether = (mcip->mci_mip->mi_info.mi_nativemedia == DL_ETHER);
1506 bw_ctl = ((mac_srs->srs_type & SRST_BW_CONTROL) != 0);
1507
1508 /*
1509 * If we don't have a Rx ring, S/W classification would have done
1510 * its job and its a packet meant for us. If we were polling on
1511 * the default ring (i.e. there was a ring assigned to this SRS),
1512 * then we need to make sure that the mac address really belongs
1513 * to us.
1514 */
1515 hw_classified = mac_srs->srs_ring != NULL &&
1516 mac_srs->srs_ring->mr_classify_type == MAC_HW_CLASSIFIER;
1517
1518 /*
1519 * Some clients, such as non-ethernet, need DLS processing in
1520 * the Rx path. Such clients clear the SRST_DLS_BYPASS flag.
1521 * DLS bypass may also be disabled via the
1522 * MCIS_RX_BYPASS_DISABLE flag.
1523 */
1524 dls_bypass = ((mac_srs->srs_type & SRST_DLS_BYPASS) != 0) &&
1525 ((mcip->mci_state_flags & MCIS_RX_BYPASS_DISABLE) == 0);
1526
1527 bzero(headmp, MAX_SR_TYPES * sizeof (mblk_t *));
1528 bzero(tailmp, MAX_SR_TYPES * sizeof (mblk_t *));
1529 bzero(cnt, MAX_SR_TYPES * sizeof (int));
1530 bzero(sz, MAX_SR_TYPES * sizeof (size_t));
1531
1532 /*
1533 * We have a chain from SRS that we need to split across the
1534 * soft rings. The squeues for the TCP and IPv4 SAPs use their
1535 * own soft rings to allow polling from the squeue. The rest of
1536 * the packets are delivered on the OTH soft ring which cannot
1537 * be polled.
1538 */
1539 while (head != NULL) {
1540 mp = head;
1541 head = head->b_next;
1542 mp->b_next = NULL;
1543
1544 type = OTH;
1545 sz1 = (mp->b_cont == NULL) ? MBLKL(mp) : msgdsize(mp);
1546
1547 if (is_ether) {
1548 /*
1549 * At this point we can be sure the packet at least
1550 * has an ether header.
1551 */
1552 if (sz1 < sizeof (struct ether_header)) {
1553 mac_rx_drop_pkt(mac_srs, mp);
1554 continue;
1555 }
1556 ehp = (struct ether_header *)mp->b_rptr;
1557
1558 /*
1559 * Determine if this is a VLAN or non-VLAN packet.
1560 */
1561 if ((sap = ntohs(ehp->ether_type)) == VLAN_TPID) {
1562 evhp = (struct ether_vlan_header *)mp->b_rptr;
1563 sap = ntohs(evhp->ether_type);
1564 hdrsize = sizeof (struct ether_vlan_header);
1565
1566 /*
1567 * Check if the VID of the packet, if
1568 * any, belongs to this client.
1569 * Technically, if this packet came up
1570 * via a HW classified ring then we
1571 * don't need to perform this check.
1572 * Perhaps a future optimization.
1573 */
1574 if (!mac_client_check_flow_vid(mcip,
1575 VLAN_ID(ntohs(evhp->ether_tci)))) {
1576 mac_rx_drop_pkt(mac_srs, mp);
1577 continue;
1578 }
1579 } else {
1580 hdrsize = sizeof (struct ether_header);
1581 }
1582 is_unicast =
1583 ((((uint8_t *)&ehp->ether_dhost)[0] & 0x01) == 0);
1584 dstaddr = (uint8_t *)&ehp->ether_dhost;
1585 } else {
1586 mac_header_info_t mhi;
1587
1588 if (mac_header_info((mac_handle_t)mcip->mci_mip,
1589 mp, &mhi) != 0) {
1590 mac_rx_drop_pkt(mac_srs, mp);
1591 continue;
1592 }
1593 hdrsize = mhi.mhi_hdrsize;
1594 sap = mhi.mhi_bindsap;
1595 is_unicast = (mhi.mhi_dsttype == MAC_ADDRTYPE_UNICAST);
1596 dstaddr = (uint8_t *)mhi.mhi_daddr;
1597 }
1598
1599 if (!dls_bypass) {
1600 FANOUT_ENQUEUE_MP(headmp[type], tailmp[type],
1601 cnt[type], bw_ctl, sz[type], sz1, mp);
1602 continue;
1603 }
1604
1605 if (sap == ETHERTYPE_IP) {
1606 /*
1607 * If we are H/W classified, but we have promisc
1608 * on, then we need to check for the unicast address.
1609 */
1610 if (hw_classified && mcip->mci_promisc_list != NULL) {
1611 mac_address_t *map;
1612
1613 rw_enter(&mcip->mci_rw_lock, RW_READER);
1614 map = mcip->mci_unicast;
1615 if (bcmp(dstaddr, map->ma_addr,
1616 map->ma_len) == 0)
1617 type = UNDEF;
1618 rw_exit(&mcip->mci_rw_lock);
1619 } else if (is_unicast) {
1620 type = UNDEF;
1621 }
1622 }
1623
1624 /*
1625 * This needs to become a contract with the driver for
1626 * the fast path.
1627 *
1628 * In the normal case the packet will have at least the L2
1629 * header and the IP + Transport header in the same mblk.
1630 * This is usually the case when the NIC driver sends up
1631 * the packet. This is also true when the stack generates
1632 * a packet that is looped back and when the stack uses the
1633 * fastpath mechanism. The normal case is optimized for
1634 * performance and may bypass DLS. All other cases go through
1635 * the 'OTH' type path without DLS bypass.
1636 */
1637 ipha = (ipha_t *)(mp->b_rptr + hdrsize);
1638 if ((type != OTH) && MBLK_RX_FANOUT_SLOWPATH(mp, ipha))
1639 type = OTH;
1640
1641 if (type == OTH) {
1642 FANOUT_ENQUEUE_MP(headmp[type], tailmp[type],
1643 cnt[type], bw_ctl, sz[type], sz1, mp);
1644 continue;
1645 }
1646
1647 ASSERT(type == UNDEF);
1648
1649 /*
1650 * Determine the type from the IP protocol value. If
1651 * classified as TCP or UDP, then update the read
1652 * pointer to the beginning of the IP header.
1653 * Otherwise leave the message as is for further
1654 * processing by DLS.
1655 */
1656 switch (ipha->ipha_protocol) {
1657 case IPPROTO_TCP:
1658 type = V4_TCP;
1659 mp->b_rptr += hdrsize;
1660 break;
1661 case IPPROTO_UDP:
1662 type = V4_UDP;
1663 mp->b_rptr += hdrsize;
1664 break;
1665 default:
1666 type = OTH;
1667 break;
1668 }
1669
1670 FANOUT_ENQUEUE_MP(headmp[type], tailmp[type], cnt[type],
1671 bw_ctl, sz[type], sz1, mp);
1672 }
1673
1674 for (type = V4_TCP; type < UNDEF; type++) {
1675 if (headmp[type] != NULL) {
1676 mac_soft_ring_t *softring;
1677
1678 ASSERT(tailmp[type]->b_next == NULL);
1679 switch (type) {
1680 case V4_TCP:
1681 softring = mac_srs->srs_tcp_soft_rings[0];
1682 break;
1683 case V4_UDP:
1684 softring = mac_srs->srs_udp_soft_rings[0];
1685 break;
1686 case OTH:
1687 softring = mac_srs->srs_oth_soft_rings[0];
1688 }
1689 mac_rx_soft_ring_process(mcip, softring,
1690 headmp[type], tailmp[type], cnt[type], sz[type]);
1691 }
1692 }
1693 }
1694
1695 int fanout_unaligned = 0;
1696
1697 /*
1698 * The fanout routine for any clients with DLS bypass disabled or for
1699 * traffic classified as "other". Returns -1 on an error (drop the
1700 * packet due to a malformed packet), 0 on success, with values
1701 * written in *indx and *type.
1702 */
1703 static int
mac_rx_srs_long_fanout(mac_soft_ring_set_t * mac_srs,mblk_t * mp,uint32_t sap,size_t hdrsize,enum pkt_type * type,uint_t * indx)1704 mac_rx_srs_long_fanout(mac_soft_ring_set_t *mac_srs, mblk_t *mp,
1705 uint32_t sap, size_t hdrsize, enum pkt_type *type, uint_t *indx)
1706 {
1707 ip6_t *ip6h;
1708 ipha_t *ipha;
1709 uint8_t *whereptr;
1710 uint_t hash;
1711 uint16_t remlen;
1712 uint8_t nexthdr;
1713 uint16_t hdr_len;
1714 uint32_t src_val, dst_val;
1715 boolean_t modifiable = B_TRUE;
1716 boolean_t v6;
1717
1718 ASSERT(MBLKL(mp) >= hdrsize);
1719
1720 if (sap == ETHERTYPE_IPV6) {
1721 v6 = B_TRUE;
1722 hdr_len = IPV6_HDR_LEN;
1723 } else if (sap == ETHERTYPE_IP) {
1724 v6 = B_FALSE;
1725 hdr_len = IP_SIMPLE_HDR_LENGTH;
1726 } else {
1727 *indx = 0;
1728 *type = OTH;
1729 return (0);
1730 }
1731
1732 ip6h = (ip6_t *)(mp->b_rptr + hdrsize);
1733 ipha = (ipha_t *)ip6h;
1734
1735 if ((uint8_t *)ip6h == mp->b_wptr) {
1736 /*
1737 * The first mblk_t only includes the mac header.
1738 * Note that it is safe to change the mp pointer here,
1739 * as the subsequent operation does not assume mp
1740 * points to the start of the mac header.
1741 */
1742 mp = mp->b_cont;
1743
1744 /*
1745 * Make sure the IP header points to an entire one.
1746 */
1747 if (mp == NULL)
1748 return (-1);
1749
1750 if (MBLKL(mp) < hdr_len) {
1751 modifiable = (DB_REF(mp) == 1);
1752
1753 if (modifiable && !pullupmsg(mp, hdr_len))
1754 return (-1);
1755 }
1756
1757 ip6h = (ip6_t *)mp->b_rptr;
1758 ipha = (ipha_t *)ip6h;
1759 }
1760
1761 if (!modifiable || !(OK_32PTR((char *)ip6h)) ||
1762 ((uint8_t *)ip6h + hdr_len > mp->b_wptr)) {
1763 /*
1764 * If either the IP header is not aligned, or it does not hold
1765 * the complete simple structure (a pullupmsg() is not an
1766 * option since it would result in an unaligned IP header),
1767 * fanout to the default ring.
1768 *
1769 * Note that this may cause packet reordering.
1770 */
1771 *indx = 0;
1772 *type = OTH;
1773 fanout_unaligned++;
1774 return (0);
1775 }
1776
1777 /*
1778 * Extract next-header, full header length, and source-hash value
1779 * using v4/v6 specific fields.
1780 */
1781 if (v6) {
1782 remlen = ntohs(ip6h->ip6_plen);
1783 nexthdr = ip6h->ip6_nxt;
1784 src_val = V4_PART_OF_V6(ip6h->ip6_src);
1785 dst_val = V4_PART_OF_V6(ip6h->ip6_dst);
1786 /*
1787 * Do src based fanout if below tunable is set to B_TRUE or
1788 * when mac_ip_hdr_length_v6() fails because of malformed
1789 * packets or because mblks need to be concatenated using
1790 * pullupmsg().
1791 *
1792 * Perform a version check to prevent parsing weirdness...
1793 */
1794 if (IPH_HDR_VERSION(ip6h) != IPV6_VERSION ||
1795 !mac_ip_hdr_length_v6(ip6h, mp->b_wptr, &hdr_len, &nexthdr,
1796 NULL)) {
1797 goto src_dst_based_fanout;
1798 }
1799 } else {
1800 hdr_len = IPH_HDR_LENGTH(ipha);
1801 remlen = ntohs(ipha->ipha_length) - hdr_len;
1802 nexthdr = ipha->ipha_protocol;
1803 src_val = (uint32_t)ipha->ipha_src;
1804 dst_val = (uint32_t)ipha->ipha_dst;
1805 /*
1806 * Catch IPv4 fragment case here. IPv6 has nexthdr == FRAG
1807 * for its equivalent case.
1808 */
1809 if ((ntohs(ipha->ipha_fragment_offset_and_flags) &
1810 (IPH_MF | IPH_OFFSET)) != 0) {
1811 goto src_dst_based_fanout;
1812 }
1813 }
1814 if (remlen < MIN_EHDR_LEN)
1815 return (-1);
1816 whereptr = (uint8_t *)ip6h + hdr_len;
1817
1818 /* If the transport is one of below, we do port/SPI based fanout */
1819 switch (nexthdr) {
1820 case IPPROTO_TCP:
1821 case IPPROTO_UDP:
1822 case IPPROTO_SCTP:
1823 case IPPROTO_ESP:
1824 /*
1825 * If the ports or SPI in the transport header is not part of
1826 * the mblk, do src_based_fanout, instead of calling
1827 * pullupmsg().
1828 */
1829 if (mp->b_cont == NULL || whereptr + PORTS_SIZE <= mp->b_wptr)
1830 break; /* out of switch... */
1831 /* FALLTHRU */
1832 default:
1833 goto src_dst_based_fanout;
1834 }
1835
1836 switch (nexthdr) {
1837 case IPPROTO_TCP:
1838 hash = HASH_ADDR(src_val, dst_val, *(uint32_t *)whereptr);
1839 *indx = COMPUTE_INDEX(hash, mac_srs->srs_tcp_ring_count);
1840 *type = OTH;
1841 break;
1842 case IPPROTO_UDP:
1843 case IPPROTO_SCTP:
1844 case IPPROTO_ESP:
1845 if (mac_fanout_type == MAC_FANOUT_DEFAULT) {
1846 hash = HASH_ADDR(src_val, dst_val,
1847 *(uint32_t *)whereptr);
1848 *indx = COMPUTE_INDEX(hash,
1849 mac_srs->srs_udp_ring_count);
1850 } else {
1851 *indx = mac_srs->srs_ind % mac_srs->srs_udp_ring_count;
1852 mac_srs->srs_ind++;
1853 }
1854 *type = OTH;
1855 break;
1856 }
1857 return (0);
1858
1859 src_dst_based_fanout:
1860 hash = HASH_ADDR(src_val, dst_val, (uint32_t)0);
1861 *indx = COMPUTE_INDEX(hash, mac_srs->srs_oth_ring_count);
1862 *type = OTH;
1863 return (0);
1864 }
1865
1866 /*
1867 * This routine delivers packets destined for an SRS into a soft ring member
1868 * of the set.
1869 *
1870 * Given a chain of packets we need to split it up into multiple sub
1871 * chains: TCP, UDP or OTH soft ring. Instead of entering the soft
1872 * ring one packet at a time, we want to enter it in the form of a
1873 * chain otherwise we get this start/stop behaviour where the worker
1874 * thread goes to sleep and then next packet comes in forcing it to
1875 * wake up.
1876 *
1877 * Note:
1878 * Since we know what is the maximum fanout possible, we create a 2D array
1879 * of 'softring types * MAX_SR_FANOUT' for the head, tail, cnt and sz
1880 * variables so that we can enter the softrings with chain. We need the
1881 * MAX_SR_FANOUT so we can allocate the arrays on the stack (a kmem_alloc
1882 * for each packet would be expensive). If we ever want to have the
1883 * ability to have unlimited fanout, we should probably declare a head,
1884 * tail, cnt, sz with each soft ring (a data struct which contains a softring
1885 * along with these members) and create an array of this uber struct so we
1886 * don't have to do kmem_alloc.
1887 */
1888 int fanout_oth1 = 0;
1889 int fanout_oth2 = 0;
1890 int fanout_oth3 = 0;
1891 int fanout_oth4 = 0;
1892 int fanout_oth5 = 0;
1893
1894 static void
mac_rx_srs_fanout(mac_soft_ring_set_t * mac_srs,mblk_t * head)1895 mac_rx_srs_fanout(mac_soft_ring_set_t *mac_srs, mblk_t *head)
1896 {
1897 struct ether_header *ehp;
1898 struct ether_vlan_header *evhp;
1899 uint32_t sap;
1900 ipha_t *ipha;
1901 uint8_t *dstaddr;
1902 uint_t indx;
1903 size_t ports_offset;
1904 size_t ipha_len;
1905 size_t hdrsize;
1906 uint_t hash;
1907 mblk_t *mp;
1908 mblk_t *headmp[MAX_SR_TYPES][MAX_SR_FANOUT];
1909 mblk_t *tailmp[MAX_SR_TYPES][MAX_SR_FANOUT];
1910 int cnt[MAX_SR_TYPES][MAX_SR_FANOUT];
1911 size_t sz[MAX_SR_TYPES][MAX_SR_FANOUT];
1912 size_t sz1;
1913 boolean_t bw_ctl;
1914 boolean_t hw_classified;
1915 boolean_t dls_bypass;
1916 boolean_t is_ether;
1917 boolean_t is_unicast;
1918 int fanout_cnt;
1919 enum pkt_type type;
1920 mac_client_impl_t *mcip = mac_srs->srs_mcip;
1921
1922 is_ether = (mcip->mci_mip->mi_info.mi_nativemedia == DL_ETHER);
1923 bw_ctl = ((mac_srs->srs_type & SRST_BW_CONTROL) != 0);
1924
1925 /*
1926 * If we don't have a Rx ring, S/W classification would have done
1927 * its job and its a packet meant for us. If we were polling on
1928 * the default ring (i.e. there was a ring assigned to this SRS),
1929 * then we need to make sure that the mac address really belongs
1930 * to us.
1931 */
1932 hw_classified = mac_srs->srs_ring != NULL &&
1933 mac_srs->srs_ring->mr_classify_type == MAC_HW_CLASSIFIER;
1934
1935 /*
1936 * Some clients, such as non Ethernet, need DLS processing in
1937 * the Rx path. Such clients clear the SRST_DLS_BYPASS flag.
1938 * DLS bypass may also be disabled via the
1939 * MCIS_RX_BYPASS_DISABLE flag, but this is only consumed by
1940 * sun4v vsw currently.
1941 */
1942 dls_bypass = ((mac_srs->srs_type & SRST_DLS_BYPASS) != 0) &&
1943 ((mcip->mci_state_flags & MCIS_RX_BYPASS_DISABLE) == 0);
1944
1945 /*
1946 * Since the softrings are never destroyed and we always
1947 * create equal number of softrings for TCP, UDP and rest,
1948 * its OK to check one of them for count and use it without
1949 * any lock. In future, if soft rings get destroyed because
1950 * of reduction in fanout, we will need to ensure that happens
1951 * behind the SRS_PROC.
1952 */
1953 fanout_cnt = mac_srs->srs_tcp_ring_count;
1954
1955 bzero(headmp, MAX_SR_TYPES * MAX_SR_FANOUT * sizeof (mblk_t *));
1956 bzero(tailmp, MAX_SR_TYPES * MAX_SR_FANOUT * sizeof (mblk_t *));
1957 bzero(cnt, MAX_SR_TYPES * MAX_SR_FANOUT * sizeof (int));
1958 bzero(sz, MAX_SR_TYPES * MAX_SR_FANOUT * sizeof (size_t));
1959
1960 /*
1961 * We got a chain from SRS that we need to send to the soft rings.
1962 * Since squeues for TCP & IPv4 SAP poll their soft rings (for
1963 * performance reasons), we need to separate out v4_tcp, v4_udp
1964 * and the rest goes in other.
1965 */
1966 while (head != NULL) {
1967 mp = head;
1968 head = head->b_next;
1969 mp->b_next = NULL;
1970
1971 type = OTH;
1972 sz1 = (mp->b_cont == NULL) ? MBLKL(mp) : msgdsize(mp);
1973
1974 if (is_ether) {
1975 /*
1976 * At this point we can be sure the packet at least
1977 * has an ether header.
1978 */
1979 if (sz1 < sizeof (struct ether_header)) {
1980 mac_rx_drop_pkt(mac_srs, mp);
1981 continue;
1982 }
1983 ehp = (struct ether_header *)mp->b_rptr;
1984
1985 /*
1986 * Determine if this is a VLAN or non-VLAN packet.
1987 */
1988 if ((sap = ntohs(ehp->ether_type)) == VLAN_TPID) {
1989 evhp = (struct ether_vlan_header *)mp->b_rptr;
1990 sap = ntohs(evhp->ether_type);
1991 hdrsize = sizeof (struct ether_vlan_header);
1992
1993 /*
1994 * Check if the VID of the packet, if
1995 * any, belongs to this client.
1996 * Technically, if this packet came up
1997 * via a HW classified ring then we
1998 * don't need to perform this check.
1999 * Perhaps a future optimization.
2000 */
2001 if (!mac_client_check_flow_vid(mcip,
2002 VLAN_ID(ntohs(evhp->ether_tci)))) {
2003 mac_rx_drop_pkt(mac_srs, mp);
2004 continue;
2005 }
2006 } else {
2007 hdrsize = sizeof (struct ether_header);
2008 }
2009 is_unicast =
2010 ((((uint8_t *)&ehp->ether_dhost)[0] & 0x01) == 0);
2011 dstaddr = (uint8_t *)&ehp->ether_dhost;
2012 } else {
2013 mac_header_info_t mhi;
2014
2015 if (mac_header_info((mac_handle_t)mcip->mci_mip,
2016 mp, &mhi) != 0) {
2017 mac_rx_drop_pkt(mac_srs, mp);
2018 continue;
2019 }
2020 hdrsize = mhi.mhi_hdrsize;
2021 sap = mhi.mhi_bindsap;
2022 is_unicast = (mhi.mhi_dsttype == MAC_ADDRTYPE_UNICAST);
2023 dstaddr = (uint8_t *)mhi.mhi_daddr;
2024 }
2025
2026 if (!dls_bypass) {
2027 if (mac_rx_srs_long_fanout(mac_srs, mp, sap,
2028 hdrsize, &type, &indx) == -1) {
2029 mac_rx_drop_pkt(mac_srs, mp);
2030 continue;
2031 }
2032
2033 FANOUT_ENQUEUE_MP(headmp[type][indx],
2034 tailmp[type][indx], cnt[type][indx], bw_ctl,
2035 sz[type][indx], sz1, mp);
2036 continue;
2037 }
2038
2039 /*
2040 * If we are using the default Rx ring where H/W or S/W
2041 * classification has not happened, we need to verify if
2042 * this unicast packet really belongs to us.
2043 */
2044 if (sap == ETHERTYPE_IP) {
2045 /*
2046 * If we are H/W classified, but we have promisc
2047 * on, then we need to check for the unicast address.
2048 */
2049 if (hw_classified && mcip->mci_promisc_list != NULL) {
2050 mac_address_t *map;
2051
2052 rw_enter(&mcip->mci_rw_lock, RW_READER);
2053 map = mcip->mci_unicast;
2054 if (bcmp(dstaddr, map->ma_addr,
2055 map->ma_len) == 0)
2056 type = UNDEF;
2057 rw_exit(&mcip->mci_rw_lock);
2058 } else if (is_unicast) {
2059 type = UNDEF;
2060 }
2061 }
2062
2063 /*
2064 * This needs to become a contract with the driver for
2065 * the fast path.
2066 */
2067
2068 ipha = (ipha_t *)(mp->b_rptr + hdrsize);
2069 if ((type != OTH) && MBLK_RX_FANOUT_SLOWPATH(mp, ipha)) {
2070 type = OTH;
2071 fanout_oth1++;
2072 }
2073
2074 if (type != OTH) {
2075 uint16_t frag_offset_flags;
2076
2077 switch (ipha->ipha_protocol) {
2078 case IPPROTO_TCP:
2079 case IPPROTO_UDP:
2080 case IPPROTO_SCTP:
2081 case IPPROTO_ESP:
2082 ipha_len = IPH_HDR_LENGTH(ipha);
2083 if ((uchar_t *)ipha + ipha_len + PORTS_SIZE >
2084 mp->b_wptr) {
2085 type = OTH;
2086 break;
2087 }
2088 frag_offset_flags =
2089 ntohs(ipha->ipha_fragment_offset_and_flags);
2090 if ((frag_offset_flags &
2091 (IPH_MF | IPH_OFFSET)) != 0) {
2092 type = OTH;
2093 fanout_oth3++;
2094 break;
2095 }
2096 ports_offset = hdrsize + ipha_len;
2097 break;
2098 default:
2099 type = OTH;
2100 fanout_oth4++;
2101 break;
2102 }
2103 }
2104
2105 if (type == OTH) {
2106 if (mac_rx_srs_long_fanout(mac_srs, mp, sap,
2107 hdrsize, &type, &indx) == -1) {
2108 mac_rx_drop_pkt(mac_srs, mp);
2109 continue;
2110 }
2111
2112 FANOUT_ENQUEUE_MP(headmp[type][indx],
2113 tailmp[type][indx], cnt[type][indx], bw_ctl,
2114 sz[type][indx], sz1, mp);
2115 continue;
2116 }
2117
2118 ASSERT(type == UNDEF);
2119
2120 /*
2121 * XXX-Sunay: We should hold srs_lock since ring_count
2122 * below can change. But if we are always called from
2123 * mac_rx_srs_drain and SRS_PROC is set, then we can
2124 * enforce that ring_count can't be changed i.e.
2125 * to change fanout type or ring count, the calling
2126 * thread needs to be behind SRS_PROC.
2127 */
2128 switch (ipha->ipha_protocol) {
2129 case IPPROTO_TCP:
2130 /*
2131 * Note that for ESP, we fanout on SPI and it is at the
2132 * same offset as the 2x16-bit ports. So it is clumped
2133 * along with TCP, UDP and SCTP.
2134 */
2135 hash = HASH_ADDR(ipha->ipha_src, ipha->ipha_dst,
2136 *(uint32_t *)(mp->b_rptr + ports_offset));
2137 indx = COMPUTE_INDEX(hash, mac_srs->srs_tcp_ring_count);
2138 type = V4_TCP;
2139 mp->b_rptr += hdrsize;
2140 break;
2141 case IPPROTO_UDP:
2142 case IPPROTO_SCTP:
2143 case IPPROTO_ESP:
2144 if (mac_fanout_type == MAC_FANOUT_DEFAULT) {
2145 hash = HASH_ADDR(ipha->ipha_src, ipha->ipha_dst,
2146 *(uint32_t *)(mp->b_rptr + ports_offset));
2147 indx = COMPUTE_INDEX(hash,
2148 mac_srs->srs_udp_ring_count);
2149 } else {
2150 indx = mac_srs->srs_ind %
2151 mac_srs->srs_udp_ring_count;
2152 mac_srs->srs_ind++;
2153 }
2154 type = V4_UDP;
2155 mp->b_rptr += hdrsize;
2156 break;
2157 default:
2158 indx = 0;
2159 type = OTH;
2160 }
2161
2162 FANOUT_ENQUEUE_MP(headmp[type][indx], tailmp[type][indx],
2163 cnt[type][indx], bw_ctl, sz[type][indx], sz1, mp);
2164 }
2165
2166 for (type = V4_TCP; type < UNDEF; type++) {
2167 int i;
2168
2169 for (i = 0; i < fanout_cnt; i++) {
2170 if (headmp[type][i] != NULL) {
2171 mac_soft_ring_t *softring;
2172
2173 ASSERT(tailmp[type][i]->b_next == NULL);
2174 switch (type) {
2175 case V4_TCP:
2176 softring =
2177 mac_srs->srs_tcp_soft_rings[i];
2178 break;
2179 case V4_UDP:
2180 softring =
2181 mac_srs->srs_udp_soft_rings[i];
2182 break;
2183 case OTH:
2184 softring =
2185 mac_srs->srs_oth_soft_rings[i];
2186 break;
2187 }
2188 mac_rx_soft_ring_process(mcip,
2189 softring, headmp[type][i], tailmp[type][i],
2190 cnt[type][i], sz[type][i]);
2191 }
2192 }
2193 }
2194 }
2195
2196 #define SRS_BYTES_TO_PICKUP 150000
2197 ssize_t max_bytes_to_pickup = SRS_BYTES_TO_PICKUP;
2198
2199 /*
2200 * mac_rx_srs_poll_ring
2201 *
2202 * This SRS Poll thread uses this routine to poll the underlying hardware
2203 * Rx ring to get a chain of packets. It can inline process that chain
2204 * if mac_latency_optimize is set (default) or signal the SRS worker thread
2205 * to do the remaining processing.
2206 *
2207 * Since packets come in the system via interrupt or poll path, we also
2208 * update the stats and deal with promiscous clients here.
2209 */
2210 void
mac_rx_srs_poll_ring(mac_soft_ring_set_t * mac_srs)2211 mac_rx_srs_poll_ring(mac_soft_ring_set_t *mac_srs)
2212 {
2213 kmutex_t *lock = &mac_srs->srs_lock;
2214 kcondvar_t *async = &mac_srs->srs_cv;
2215 mac_srs_rx_t *srs_rx = &mac_srs->srs_rx;
2216 mblk_t *head, *tail, *mp;
2217 callb_cpr_t cprinfo;
2218 ssize_t bytes_to_pickup;
2219 size_t sz;
2220 int count;
2221 mac_client_impl_t *smcip;
2222
2223 CALLB_CPR_INIT(&cprinfo, lock, callb_generic_cpr, "mac_srs_poll");
2224 mutex_enter(lock);
2225
2226 start:
2227 for (;;) {
2228 if (mac_srs->srs_state & SRS_PAUSE)
2229 goto done;
2230
2231 CALLB_CPR_SAFE_BEGIN(&cprinfo);
2232 cv_wait(async, lock);
2233 CALLB_CPR_SAFE_END(&cprinfo, lock);
2234
2235 if (mac_srs->srs_state & SRS_PAUSE)
2236 goto done;
2237
2238 check_again:
2239 if (mac_srs->srs_type & SRST_BW_CONTROL) {
2240 /*
2241 * We pick as many bytes as we are allowed to queue.
2242 * Its possible that we will exceed the total
2243 * packets queued in case this SRS is part of the
2244 * Rx ring group since > 1 poll thread can be pulling
2245 * upto the max allowed packets at the same time
2246 * but that should be OK.
2247 */
2248 mutex_enter(&mac_srs->srs_bw->mac_bw_lock);
2249 bytes_to_pickup =
2250 mac_srs->srs_bw->mac_bw_drop_threshold -
2251 mac_srs->srs_bw->mac_bw_sz;
2252 /*
2253 * We shouldn't have been signalled if we
2254 * have 0 or less bytes to pick but since
2255 * some of the bytes accounting is driver
2256 * dependant, we do the safety check.
2257 */
2258 if (bytes_to_pickup < 0)
2259 bytes_to_pickup = 0;
2260 mutex_exit(&mac_srs->srs_bw->mac_bw_lock);
2261 } else {
2262 /*
2263 * ToDO: Need to change the polling API
2264 * to add a packet count and a flag which
2265 * tells the driver whether we want packets
2266 * based on a count, or bytes, or all the
2267 * packets queued in the driver/HW. This
2268 * way, we never have to check the limits
2269 * on poll path. We truly let only as many
2270 * packets enter the system as we are willing
2271 * to process or queue.
2272 *
2273 * Something along the lines of
2274 * pkts_to_pickup = mac_soft_ring_max_q_cnt -
2275 * mac_srs->srs_poll_pkt_cnt
2276 */
2277
2278 /*
2279 * Since we are not doing B/W control, pick
2280 * as many packets as allowed.
2281 */
2282 bytes_to_pickup = max_bytes_to_pickup;
2283 }
2284
2285 /* Poll the underlying Hardware */
2286 mutex_exit(lock);
2287 head = MAC_HWRING_POLL(mac_srs->srs_ring, (int)bytes_to_pickup);
2288 mutex_enter(lock);
2289
2290 ASSERT((mac_srs->srs_state & SRS_POLL_THR_OWNER) ==
2291 SRS_POLL_THR_OWNER);
2292
2293 mp = tail = head;
2294 count = 0;
2295 sz = 0;
2296 while (mp != NULL) {
2297 tail = mp;
2298 sz += msgdsize(mp);
2299 mp = mp->b_next;
2300 count++;
2301 }
2302
2303 if (head != NULL) {
2304 tail->b_next = NULL;
2305 smcip = mac_srs->srs_mcip;
2306
2307 SRS_RX_STAT_UPDATE(mac_srs, pollbytes, sz);
2308 SRS_RX_STAT_UPDATE(mac_srs, pollcnt, count);
2309
2310 /*
2311 * If there are any promiscuous mode callbacks
2312 * defined for this MAC client, pass them a copy
2313 * if appropriate and also update the counters.
2314 */
2315 if (smcip != NULL) {
2316 if (smcip->mci_mip->mi_promisc_list != NULL) {
2317 mutex_exit(lock);
2318 mac_promisc_dispatch(smcip->mci_mip,
2319 head, NULL, B_FALSE);
2320 mutex_enter(lock);
2321 }
2322 }
2323 if (mac_srs->srs_type & SRST_BW_CONTROL) {
2324 mutex_enter(&mac_srs->srs_bw->mac_bw_lock);
2325 mac_srs->srs_bw->mac_bw_polled += sz;
2326 mutex_exit(&mac_srs->srs_bw->mac_bw_lock);
2327 }
2328 MAC_RX_SRS_ENQUEUE_CHAIN(mac_srs, head, tail,
2329 count, sz);
2330 if (count <= 10)
2331 srs_rx->sr_stat.mrs_chaincntundr10++;
2332 else if (count > 10 && count <= 50)
2333 srs_rx->sr_stat.mrs_chaincnt10to50++;
2334 else
2335 srs_rx->sr_stat.mrs_chaincntover50++;
2336 }
2337
2338 /*
2339 * We are guaranteed that SRS_PROC will be set if we
2340 * are here. Also, poll thread gets to run only if
2341 * the drain was being done by a worker thread although
2342 * its possible that worker thread is still running
2343 * and poll thread was sent down to keep the pipeline
2344 * going instead of doing a complete drain and then
2345 * trying to poll the NIC.
2346 *
2347 * So we need to check SRS_WORKER flag to make sure
2348 * that the worker thread is not processing the queue
2349 * in parallel to us. The flags and conditions are
2350 * protected by the srs_lock to prevent any race. We
2351 * ensure that we don't drop the srs_lock from now
2352 * till the end and similarly we don't drop the srs_lock
2353 * in mac_rx_srs_drain() till similar condition check
2354 * are complete. The mac_rx_srs_drain() needs to ensure
2355 * that SRS_WORKER flag remains set as long as its
2356 * processing the queue.
2357 */
2358 if (!(mac_srs->srs_state & SRS_WORKER) &&
2359 (mac_srs->srs_first != NULL)) {
2360 /*
2361 * We have packets to process and worker thread
2362 * is not running. Check to see if poll thread is
2363 * allowed to process.
2364 */
2365 if (mac_srs->srs_state & SRS_LATENCY_OPT) {
2366 mac_srs->srs_drain_func(mac_srs, SRS_POLL_PROC);
2367 if (!(mac_srs->srs_state & SRS_PAUSE) &&
2368 srs_rx->sr_poll_pkt_cnt <=
2369 srs_rx->sr_lowat) {
2370 srs_rx->sr_poll_again++;
2371 goto check_again;
2372 }
2373 /*
2374 * We are already above low water mark
2375 * so stay in the polling mode but no
2376 * need to poll. Once we dip below
2377 * the polling threshold, the processing
2378 * thread (soft ring) will signal us
2379 * to poll again (MAC_UPDATE_SRS_COUNT)
2380 */
2381 srs_rx->sr_poll_drain_no_poll++;
2382 mac_srs->srs_state &= ~(SRS_PROC|SRS_GET_PKTS);
2383 /*
2384 * In B/W control case, its possible
2385 * that the backlog built up due to
2386 * B/W limit being reached and packets
2387 * are queued only in SRS. In this case,
2388 * we should schedule worker thread
2389 * since no one else will wake us up.
2390 */
2391 if ((mac_srs->srs_type & SRST_BW_CONTROL) &&
2392 (mac_srs->srs_tid == NULL)) {
2393 mac_srs->srs_tid =
2394 timeout(mac_srs_fire, mac_srs, 1);
2395 srs_rx->sr_poll_worker_wakeup++;
2396 }
2397 } else {
2398 /*
2399 * Wakeup the worker thread for more processing.
2400 * We optimize for throughput in this case.
2401 */
2402 mac_srs->srs_state &= ~(SRS_PROC|SRS_GET_PKTS);
2403 MAC_SRS_WORKER_WAKEUP(mac_srs);
2404 srs_rx->sr_poll_sig_worker++;
2405 }
2406 } else if ((mac_srs->srs_first == NULL) &&
2407 !(mac_srs->srs_state & SRS_WORKER)) {
2408 /*
2409 * There is nothing queued in SRS and
2410 * no worker thread running. Plus we
2411 * didn't get anything from the H/W
2412 * as well (head == NULL);
2413 */
2414 ASSERT(head == NULL);
2415 mac_srs->srs_state &=
2416 ~(SRS_PROC|SRS_GET_PKTS);
2417
2418 /*
2419 * If we have a packets in soft ring, don't allow
2420 * more packets to come into this SRS by keeping the
2421 * interrupts off but not polling the H/W. The
2422 * poll thread will get signaled as soon as
2423 * srs_poll_pkt_cnt dips below poll threshold.
2424 */
2425 if (srs_rx->sr_poll_pkt_cnt == 0) {
2426 srs_rx->sr_poll_intr_enable++;
2427 MAC_SRS_POLLING_OFF(mac_srs);
2428 } else {
2429 /*
2430 * We know nothing is queued in SRS
2431 * since we are here after checking
2432 * srs_first is NULL. The backlog
2433 * is entirely due to packets queued
2434 * in Soft ring which will wake us up
2435 * and get the interface out of polling
2436 * mode once the backlog dips below
2437 * sr_poll_thres.
2438 */
2439 srs_rx->sr_poll_no_poll++;
2440 }
2441 } else {
2442 /*
2443 * Worker thread is already running.
2444 * Nothing much to do. If the polling
2445 * was enabled, worker thread will deal
2446 * with that.
2447 */
2448 mac_srs->srs_state &= ~SRS_GET_PKTS;
2449 srs_rx->sr_poll_goto_sleep++;
2450 }
2451 }
2452 done:
2453 mac_srs->srs_state |= SRS_POLL_THR_QUIESCED;
2454 cv_signal(&mac_srs->srs_async);
2455 /*
2456 * If this is a temporary quiesce then wait for the restart signal
2457 * from the srs worker. Then clear the flags and signal the srs worker
2458 * to ensure a positive handshake and go back to start.
2459 */
2460 while (!(mac_srs->srs_state & (SRS_CONDEMNED | SRS_POLL_THR_RESTART)))
2461 cv_wait(async, lock);
2462 if (mac_srs->srs_state & SRS_POLL_THR_RESTART) {
2463 ASSERT(!(mac_srs->srs_state & SRS_CONDEMNED));
2464 mac_srs->srs_state &=
2465 ~(SRS_POLL_THR_QUIESCED | SRS_POLL_THR_RESTART);
2466 cv_signal(&mac_srs->srs_async);
2467 goto start;
2468 } else {
2469 mac_srs->srs_state |= SRS_POLL_THR_EXITED;
2470 cv_signal(&mac_srs->srs_async);
2471 CALLB_CPR_EXIT(&cprinfo);
2472 thread_exit();
2473 }
2474 }
2475
2476 /*
2477 * mac_srs_pick_chain
2478 *
2479 * In Bandwidth control case, checks how many packets can be processed
2480 * and return them in a sub chain.
2481 */
2482 static mblk_t *
mac_srs_pick_chain(mac_soft_ring_set_t * mac_srs,mblk_t ** chain_tail,size_t * chain_sz,int * chain_cnt)2483 mac_srs_pick_chain(mac_soft_ring_set_t *mac_srs, mblk_t **chain_tail,
2484 size_t *chain_sz, int *chain_cnt)
2485 {
2486 mblk_t *head = NULL;
2487 mblk_t *tail = NULL;
2488 size_t sz;
2489 size_t tsz = 0;
2490 int cnt = 0;
2491 mblk_t *mp;
2492
2493 ASSERT(MUTEX_HELD(&mac_srs->srs_lock));
2494 mutex_enter(&mac_srs->srs_bw->mac_bw_lock);
2495 if (((mac_srs->srs_bw->mac_bw_used + mac_srs->srs_size) <=
2496 mac_srs->srs_bw->mac_bw_limit) ||
2497 (mac_srs->srs_bw->mac_bw_limit == 0)) {
2498 mutex_exit(&mac_srs->srs_bw->mac_bw_lock);
2499 head = mac_srs->srs_first;
2500 mac_srs->srs_first = NULL;
2501 *chain_tail = mac_srs->srs_last;
2502 mac_srs->srs_last = NULL;
2503 *chain_sz = mac_srs->srs_size;
2504 *chain_cnt = mac_srs->srs_count;
2505 mac_srs->srs_count = 0;
2506 mac_srs->srs_size = 0;
2507 return (head);
2508 }
2509
2510 /*
2511 * Can't clear the entire backlog.
2512 * Need to find how many packets to pick
2513 */
2514 ASSERT(MUTEX_HELD(&mac_srs->srs_bw->mac_bw_lock));
2515 while ((mp = mac_srs->srs_first) != NULL) {
2516 sz = msgdsize(mp);
2517 if ((tsz + sz + mac_srs->srs_bw->mac_bw_used) >
2518 mac_srs->srs_bw->mac_bw_limit) {
2519 if (!(mac_srs->srs_bw->mac_bw_state & SRS_BW_ENFORCED))
2520 mac_srs->srs_bw->mac_bw_state |=
2521 SRS_BW_ENFORCED;
2522 break;
2523 }
2524
2525 /*
2526 * The _size & cnt is decremented from the softrings
2527 * when they send up the packet for polling to work
2528 * properly.
2529 */
2530 tsz += sz;
2531 cnt++;
2532 mac_srs->srs_count--;
2533 mac_srs->srs_size -= sz;
2534 if (tail != NULL)
2535 tail->b_next = mp;
2536 else
2537 head = mp;
2538 tail = mp;
2539 mac_srs->srs_first = mac_srs->srs_first->b_next;
2540 }
2541 mutex_exit(&mac_srs->srs_bw->mac_bw_lock);
2542 if (mac_srs->srs_first == NULL)
2543 mac_srs->srs_last = NULL;
2544
2545 if (tail != NULL)
2546 tail->b_next = NULL;
2547 *chain_tail = tail;
2548 *chain_cnt = cnt;
2549 *chain_sz = tsz;
2550
2551 return (head);
2552 }
2553
2554 /*
2555 * mac_rx_srs_drain
2556 *
2557 * The SRS drain routine. Gets to run to clear the queue. Any thread
2558 * (worker, interrupt, poll) can call this based on processing model.
2559 * The first thing we do is disable interrupts if possible and then
2560 * drain the queue. we also try to poll the underlying hardware if
2561 * there is a dedicated hardware Rx ring assigned to this SRS.
2562 *
2563 * There is a equivalent drain routine in bandwidth control mode
2564 * mac_rx_srs_drain_bw. There is some code duplication between the two
2565 * routines but they are highly performance sensitive and are easier
2566 * to read/debug if they stay separate. Any code changes here might
2567 * also apply to mac_rx_srs_drain_bw as well.
2568 */
2569 void
mac_rx_srs_drain(mac_soft_ring_set_t * mac_srs,uint_t proc_type)2570 mac_rx_srs_drain(mac_soft_ring_set_t *mac_srs, uint_t proc_type)
2571 {
2572 mblk_t *head;
2573 mblk_t *tail;
2574 timeout_id_t tid;
2575 int cnt = 0;
2576 mac_client_impl_t *mcip = mac_srs->srs_mcip;
2577 mac_srs_rx_t *srs_rx = &mac_srs->srs_rx;
2578
2579 ASSERT(MUTEX_HELD(&mac_srs->srs_lock));
2580 ASSERT(!(mac_srs->srs_type & SRST_BW_CONTROL));
2581
2582 /* If we are blanked i.e. can't do upcalls, then we are done */
2583 if (mac_srs->srs_state & (SRS_BLANK | SRS_PAUSE)) {
2584 ASSERT((mac_srs->srs_type & SRST_NO_SOFT_RINGS) ||
2585 (mac_srs->srs_state & SRS_PAUSE));
2586 goto out;
2587 }
2588
2589 if (mac_srs->srs_first == NULL)
2590 goto out;
2591
2592 if (!(mac_srs->srs_state & SRS_LATENCY_OPT) &&
2593 (srs_rx->sr_poll_pkt_cnt <= srs_rx->sr_lowat)) {
2594 /*
2595 * In the normal case, the SRS worker thread does no
2596 * work and we wait for a backlog to build up before
2597 * we switch into polling mode. In case we are
2598 * optimizing for throughput, we use the worker thread
2599 * as well. The goal is to let worker thread process
2600 * the queue and poll thread to feed packets into
2601 * the queue. As such, we should signal the poll
2602 * thread to try and get more packets.
2603 *
2604 * We could have pulled this check in the POLL_RING
2605 * macro itself but keeping it explicit here makes
2606 * the architecture more human understandable.
2607 */
2608 MAC_SRS_POLL_RING(mac_srs);
2609 }
2610
2611 again:
2612 head = mac_srs->srs_first;
2613 mac_srs->srs_first = NULL;
2614 tail = mac_srs->srs_last;
2615 mac_srs->srs_last = NULL;
2616 cnt = mac_srs->srs_count;
2617 mac_srs->srs_count = 0;
2618
2619 ASSERT(head != NULL);
2620 ASSERT(tail != NULL);
2621
2622 if ((tid = mac_srs->srs_tid) != NULL)
2623 mac_srs->srs_tid = NULL;
2624
2625 mac_srs->srs_state |= (SRS_PROC|proc_type);
2626
2627 /*
2628 * mcip is NULL for broadcast and multicast flows. The promisc
2629 * callbacks for broadcast and multicast packets are delivered from
2630 * mac_rx() and we don't need to worry about that case in this path
2631 */
2632 if (mcip != NULL) {
2633 if (mcip->mci_promisc_list != NULL) {
2634 mutex_exit(&mac_srs->srs_lock);
2635 mac_promisc_client_dispatch(mcip, head);
2636 mutex_enter(&mac_srs->srs_lock);
2637 }
2638 if (MAC_PROTECT_ENABLED(mcip, MPT_IPNOSPOOF)) {
2639 mutex_exit(&mac_srs->srs_lock);
2640 mac_protect_intercept_dynamic(mcip, head);
2641 mutex_enter(&mac_srs->srs_lock);
2642 }
2643 }
2644
2645 /*
2646 * Check if SRS itself is doing the processing. This direct
2647 * path applies only when subflows are present.
2648 */
2649 if (mac_srs->srs_type & SRST_NO_SOFT_RINGS) {
2650 mac_direct_rx_t proc;
2651 void *arg1;
2652 mac_resource_handle_t arg2;
2653
2654 /*
2655 * This is the case when a Rx is directly
2656 * assigned and we have a fully classified
2657 * protocol chain. We can deal with it in
2658 * one shot.
2659 */
2660 proc = srs_rx->sr_func;
2661 arg1 = srs_rx->sr_arg1;
2662 arg2 = srs_rx->sr_arg2;
2663
2664 mac_srs->srs_state |= SRS_CLIENT_PROC;
2665 mutex_exit(&mac_srs->srs_lock);
2666 if (tid != NULL) {
2667 (void) untimeout(tid);
2668 tid = NULL;
2669 }
2670
2671 proc(arg1, arg2, head, NULL);
2672 /*
2673 * Decrement the size and count here itelf
2674 * since the packet has been processed.
2675 */
2676 mutex_enter(&mac_srs->srs_lock);
2677 MAC_UPDATE_SRS_COUNT_LOCKED(mac_srs, cnt);
2678 if (mac_srs->srs_state & SRS_CLIENT_WAIT)
2679 cv_signal(&mac_srs->srs_client_cv);
2680 mac_srs->srs_state &= ~SRS_CLIENT_PROC;
2681 } else {
2682 /* Some kind of softrings based fanout is required */
2683 mutex_exit(&mac_srs->srs_lock);
2684 if (tid != NULL) {
2685 (void) untimeout(tid);
2686 tid = NULL;
2687 }
2688
2689 /*
2690 * Since the fanout routines can deal with chains,
2691 * shoot the entire chain up.
2692 */
2693 if (mac_srs->srs_type & SRST_FANOUT_SRC_IP)
2694 mac_rx_srs_fanout(mac_srs, head);
2695 else
2696 mac_rx_srs_proto_fanout(mac_srs, head);
2697 mutex_enter(&mac_srs->srs_lock);
2698 }
2699
2700 if (!(mac_srs->srs_state & (SRS_BLANK|SRS_PAUSE)) &&
2701 (mac_srs->srs_first != NULL)) {
2702 /*
2703 * More packets arrived while we were clearing the
2704 * SRS. This can be possible because of one of
2705 * three conditions below:
2706 * 1) The driver is using multiple worker threads
2707 * to send the packets to us.
2708 * 2) The driver has a race in switching
2709 * between interrupt and polling mode or
2710 * 3) Packets are arriving in this SRS via the
2711 * S/W classification as well.
2712 *
2713 * We should switch to polling mode and see if we
2714 * need to send the poll thread down. Also, signal
2715 * the worker thread to process whats just arrived.
2716 */
2717 MAC_SRS_POLLING_ON(mac_srs);
2718 if (srs_rx->sr_poll_pkt_cnt <= srs_rx->sr_lowat) {
2719 srs_rx->sr_drain_poll_sig++;
2720 MAC_SRS_POLL_RING(mac_srs);
2721 }
2722
2723 /*
2724 * If we didn't signal the poll thread, we need
2725 * to deal with the pending packets ourselves.
2726 */
2727 if (proc_type == SRS_WORKER) {
2728 srs_rx->sr_drain_again++;
2729 goto again;
2730 } else {
2731 srs_rx->sr_drain_worker_sig++;
2732 cv_signal(&mac_srs->srs_async);
2733 }
2734 }
2735
2736 out:
2737 if (mac_srs->srs_state & SRS_GET_PKTS) {
2738 /*
2739 * Poll thread is already running. Leave the
2740 * SRS_RPOC set and hand over the control to
2741 * poll thread.
2742 */
2743 mac_srs->srs_state &= ~proc_type;
2744 srs_rx->sr_drain_poll_running++;
2745 return;
2746 }
2747
2748 /*
2749 * Even if there are no packets queued in SRS, we
2750 * need to make sure that the shared counter is
2751 * clear and any associated softrings have cleared
2752 * all the backlog. Otherwise, leave the interface
2753 * in polling mode and the poll thread will get
2754 * signalled once the count goes down to zero.
2755 *
2756 * If someone is already draining the queue (SRS_PROC is
2757 * set) when the srs_poll_pkt_cnt goes down to zero,
2758 * then it means that drain is already running and we
2759 * will turn off polling at that time if there is
2760 * no backlog.
2761 *
2762 * As long as there are packets queued either
2763 * in soft ring set or its soft rings, we will leave
2764 * the interface in polling mode (even if the drain
2765 * was done being the interrupt thread). We signal
2766 * the poll thread as well if we have dipped below
2767 * low water mark.
2768 *
2769 * NOTE: We can't use the MAC_SRS_POLLING_ON macro
2770 * since that turn polling on only for worker thread.
2771 * Its not worth turning polling on for interrupt
2772 * thread (since NIC will not issue another interrupt)
2773 * unless a backlog builds up.
2774 */
2775 if ((srs_rx->sr_poll_pkt_cnt > 0) &&
2776 (mac_srs->srs_state & SRS_POLLING_CAPAB)) {
2777 mac_srs->srs_state &= ~(SRS_PROC|proc_type);
2778 srs_rx->sr_drain_keep_polling++;
2779 MAC_SRS_POLLING_ON(mac_srs);
2780 if (srs_rx->sr_poll_pkt_cnt <= srs_rx->sr_lowat)
2781 MAC_SRS_POLL_RING(mac_srs);
2782 return;
2783 }
2784
2785 /* Nothing else to do. Get out of poll mode */
2786 MAC_SRS_POLLING_OFF(mac_srs);
2787 mac_srs->srs_state &= ~(SRS_PROC|proc_type);
2788 srs_rx->sr_drain_finish_intr++;
2789 }
2790
2791 /*
2792 * mac_rx_srs_drain_bw
2793 *
2794 * The SRS BW drain routine. Gets to run to clear the queue. Any thread
2795 * (worker, interrupt, poll) can call this based on processing model.
2796 * The first thing we do is disable interrupts if possible and then
2797 * drain the queue. we also try to poll the underlying hardware if
2798 * there is a dedicated hardware Rx ring assigned to this SRS.
2799 *
2800 * There is a equivalent drain routine in non bandwidth control mode
2801 * mac_rx_srs_drain. There is some code duplication between the two
2802 * routines but they are highly performance sensitive and are easier
2803 * to read/debug if they stay separate. Any code changes here might
2804 * also apply to mac_rx_srs_drain as well.
2805 */
2806 void
mac_rx_srs_drain_bw(mac_soft_ring_set_t * mac_srs,uint_t proc_type)2807 mac_rx_srs_drain_bw(mac_soft_ring_set_t *mac_srs, uint_t proc_type)
2808 {
2809 mblk_t *head;
2810 mblk_t *tail;
2811 timeout_id_t tid;
2812 size_t sz = 0;
2813 int cnt = 0;
2814 mac_client_impl_t *mcip = mac_srs->srs_mcip;
2815 mac_srs_rx_t *srs_rx = &mac_srs->srs_rx;
2816 clock_t now;
2817
2818 ASSERT(MUTEX_HELD(&mac_srs->srs_lock));
2819 ASSERT(mac_srs->srs_type & SRST_BW_CONTROL);
2820 again:
2821 /* Check if we are doing B/W control */
2822 mutex_enter(&mac_srs->srs_bw->mac_bw_lock);
2823 now = ddi_get_lbolt();
2824 if (mac_srs->srs_bw->mac_bw_curr_time != now) {
2825 mac_srs->srs_bw->mac_bw_curr_time = now;
2826 mac_srs->srs_bw->mac_bw_used = 0;
2827 if (mac_srs->srs_bw->mac_bw_state & SRS_BW_ENFORCED)
2828 mac_srs->srs_bw->mac_bw_state &= ~SRS_BW_ENFORCED;
2829 } else if (mac_srs->srs_bw->mac_bw_state & SRS_BW_ENFORCED) {
2830 mutex_exit(&mac_srs->srs_bw->mac_bw_lock);
2831 goto done;
2832 } else if (mac_srs->srs_bw->mac_bw_used >
2833 mac_srs->srs_bw->mac_bw_limit) {
2834 mac_srs->srs_bw->mac_bw_state |= SRS_BW_ENFORCED;
2835 mutex_exit(&mac_srs->srs_bw->mac_bw_lock);
2836 goto done;
2837 }
2838 mutex_exit(&mac_srs->srs_bw->mac_bw_lock);
2839
2840 /* If we are blanked i.e. can't do upcalls, then we are done */
2841 if (mac_srs->srs_state & (SRS_BLANK | SRS_PAUSE)) {
2842 ASSERT((mac_srs->srs_type & SRST_NO_SOFT_RINGS) ||
2843 (mac_srs->srs_state & SRS_PAUSE));
2844 goto done;
2845 }
2846
2847 sz = 0;
2848 cnt = 0;
2849 if ((head = mac_srs_pick_chain(mac_srs, &tail, &sz, &cnt)) == NULL) {
2850 /*
2851 * We couldn't pick up a single packet.
2852 */
2853 mutex_enter(&mac_srs->srs_bw->mac_bw_lock);
2854 if ((mac_srs->srs_bw->mac_bw_used == 0) &&
2855 (mac_srs->srs_size != 0) &&
2856 !(mac_srs->srs_bw->mac_bw_state & SRS_BW_ENFORCED)) {
2857 /*
2858 * Seems like configured B/W doesn't
2859 * even allow processing of 1 packet
2860 * per tick.
2861 *
2862 * XXX: raise the limit to processing
2863 * at least 1 packet per tick.
2864 */
2865 mac_srs->srs_bw->mac_bw_limit +=
2866 mac_srs->srs_bw->mac_bw_limit;
2867 mac_srs->srs_bw->mac_bw_drop_threshold +=
2868 mac_srs->srs_bw->mac_bw_drop_threshold;
2869 cmn_err(CE_NOTE, "mac_rx_srs_drain: srs(%p) "
2870 "raised B/W limit to %d since not even a "
2871 "single packet can be processed per "
2872 "tick %d\n", (void *)mac_srs,
2873 (int)mac_srs->srs_bw->mac_bw_limit,
2874 (int)msgdsize(mac_srs->srs_first));
2875 }
2876 mutex_exit(&mac_srs->srs_bw->mac_bw_lock);
2877 goto done;
2878 }
2879
2880 ASSERT(head != NULL);
2881 ASSERT(tail != NULL);
2882
2883 /* zero bandwidth: drop all and return to interrupt mode */
2884 mutex_enter(&mac_srs->srs_bw->mac_bw_lock);
2885 if (mac_srs->srs_bw->mac_bw_limit == 0) {
2886 srs_rx->sr_stat.mrs_sdrops += cnt;
2887 ASSERT(mac_srs->srs_bw->mac_bw_sz >= sz);
2888 mac_srs->srs_bw->mac_bw_sz -= sz;
2889 mac_srs->srs_bw->mac_bw_drop_bytes += sz;
2890 mutex_exit(&mac_srs->srs_bw->mac_bw_lock);
2891 mac_drop_chain(head, "Rx no bandwidth");
2892 goto leave_poll;
2893 } else {
2894 mutex_exit(&mac_srs->srs_bw->mac_bw_lock);
2895 }
2896
2897 if ((tid = mac_srs->srs_tid) != NULL)
2898 mac_srs->srs_tid = NULL;
2899
2900 mac_srs->srs_state |= (SRS_PROC|proc_type);
2901 MAC_SRS_WORKER_POLLING_ON(mac_srs);
2902
2903 /*
2904 * mcip is NULL for broadcast and multicast flows. The promisc
2905 * callbacks for broadcast and multicast packets are delivered from
2906 * mac_rx() and we don't need to worry about that case in this path
2907 */
2908 if (mcip != NULL) {
2909 if (mcip->mci_promisc_list != NULL) {
2910 mutex_exit(&mac_srs->srs_lock);
2911 mac_promisc_client_dispatch(mcip, head);
2912 mutex_enter(&mac_srs->srs_lock);
2913 }
2914 if (MAC_PROTECT_ENABLED(mcip, MPT_IPNOSPOOF)) {
2915 mutex_exit(&mac_srs->srs_lock);
2916 mac_protect_intercept_dynamic(mcip, head);
2917 mutex_enter(&mac_srs->srs_lock);
2918 }
2919 }
2920
2921 /*
2922 * Check if SRS itself is doing the processing
2923 * This direct path does not apply when subflows are present. In this
2924 * case, packets need to be dispatched to a soft ring according to the
2925 * flow's bandwidth and other resources contraints.
2926 */
2927 if (mac_srs->srs_type & SRST_NO_SOFT_RINGS) {
2928 mac_direct_rx_t proc;
2929 void *arg1;
2930 mac_resource_handle_t arg2;
2931
2932 /*
2933 * This is the case when a Rx is directly
2934 * assigned and we have a fully classified
2935 * protocol chain. We can deal with it in
2936 * one shot.
2937 */
2938 proc = srs_rx->sr_func;
2939 arg1 = srs_rx->sr_arg1;
2940 arg2 = srs_rx->sr_arg2;
2941
2942 mac_srs->srs_state |= SRS_CLIENT_PROC;
2943 mutex_exit(&mac_srs->srs_lock);
2944 if (tid != NULL) {
2945 (void) untimeout(tid);
2946 tid = NULL;
2947 }
2948
2949 proc(arg1, arg2, head, NULL);
2950 /*
2951 * Decrement the size and count here itelf
2952 * since the packet has been processed.
2953 */
2954 mutex_enter(&mac_srs->srs_lock);
2955 MAC_UPDATE_SRS_COUNT_LOCKED(mac_srs, cnt);
2956 MAC_UPDATE_SRS_SIZE_LOCKED(mac_srs, sz);
2957
2958 if (mac_srs->srs_state & SRS_CLIENT_WAIT)
2959 cv_signal(&mac_srs->srs_client_cv);
2960 mac_srs->srs_state &= ~SRS_CLIENT_PROC;
2961 } else {
2962 /* Some kind of softrings based fanout is required */
2963 mutex_exit(&mac_srs->srs_lock);
2964 if (tid != NULL) {
2965 (void) untimeout(tid);
2966 tid = NULL;
2967 }
2968
2969 /*
2970 * Since the fanout routines can deal with chains,
2971 * shoot the entire chain up.
2972 */
2973 if (mac_srs->srs_type & SRST_FANOUT_SRC_IP)
2974 mac_rx_srs_fanout(mac_srs, head);
2975 else
2976 mac_rx_srs_proto_fanout(mac_srs, head);
2977 mutex_enter(&mac_srs->srs_lock);
2978 }
2979
2980 /*
2981 * Send the poll thread to pick up any packets arrived
2982 * so far. This also serves as the last check in case
2983 * nothing else is queued in the SRS. The poll thread
2984 * is signalled only in the case the drain was done
2985 * by the worker thread and SRS_WORKER is set. The
2986 * worker thread can run in parallel as long as the
2987 * SRS_WORKER flag is set. We we have nothing else to
2988 * process, we can exit while leaving SRS_PROC set
2989 * which gives the poll thread control to process and
2990 * cleanup once it returns from the NIC.
2991 *
2992 * If we have nothing else to process, we need to
2993 * ensure that we keep holding the srs_lock till
2994 * all the checks below are done and control is
2995 * handed to the poll thread if it was running.
2996 */
2997 mutex_enter(&mac_srs->srs_bw->mac_bw_lock);
2998 if (!(mac_srs->srs_bw->mac_bw_state & SRS_BW_ENFORCED)) {
2999 if (mac_srs->srs_first != NULL) {
3000 if (proc_type == SRS_WORKER) {
3001 mutex_exit(&mac_srs->srs_bw->mac_bw_lock);
3002 if (srs_rx->sr_poll_pkt_cnt <=
3003 srs_rx->sr_lowat)
3004 MAC_SRS_POLL_RING(mac_srs);
3005 goto again;
3006 } else {
3007 cv_signal(&mac_srs->srs_async);
3008 }
3009 }
3010 }
3011 mutex_exit(&mac_srs->srs_bw->mac_bw_lock);
3012
3013 done:
3014
3015 if (mac_srs->srs_state & SRS_GET_PKTS) {
3016 /*
3017 * Poll thread is already running. Leave the
3018 * SRS_RPOC set and hand over the control to
3019 * poll thread.
3020 */
3021 mac_srs->srs_state &= ~proc_type;
3022 return;
3023 }
3024
3025 /*
3026 * If we can't process packets because we have exceeded
3027 * B/W limit for this tick, just set the timeout
3028 * and leave.
3029 *
3030 * Even if there are no packets queued in SRS, we
3031 * need to make sure that the shared counter is
3032 * clear and any associated softrings have cleared
3033 * all the backlog. Otherwise, leave the interface
3034 * in polling mode and the poll thread will get
3035 * signalled once the count goes down to zero.
3036 *
3037 * If someone is already draining the queue (SRS_PROC is
3038 * set) when the srs_poll_pkt_cnt goes down to zero,
3039 * then it means that drain is already running and we
3040 * will turn off polling at that time if there is
3041 * no backlog. As long as there are packets queued either
3042 * is soft ring set or its soft rings, we will leave
3043 * the interface in polling mode.
3044 */
3045 mutex_enter(&mac_srs->srs_bw->mac_bw_lock);
3046 if ((mac_srs->srs_state & SRS_POLLING_CAPAB) &&
3047 ((mac_srs->srs_bw->mac_bw_state & SRS_BW_ENFORCED) ||
3048 (srs_rx->sr_poll_pkt_cnt > 0))) {
3049 MAC_SRS_POLLING_ON(mac_srs);
3050 mac_srs->srs_state &= ~(SRS_PROC|proc_type);
3051 if ((mac_srs->srs_first != NULL) &&
3052 (mac_srs->srs_tid == NULL))
3053 mac_srs->srs_tid = timeout(mac_srs_fire,
3054 mac_srs, 1);
3055 mutex_exit(&mac_srs->srs_bw->mac_bw_lock);
3056 return;
3057 }
3058 mutex_exit(&mac_srs->srs_bw->mac_bw_lock);
3059
3060 leave_poll:
3061
3062 /* Nothing else to do. Get out of poll mode */
3063 MAC_SRS_POLLING_OFF(mac_srs);
3064 mac_srs->srs_state &= ~(SRS_PROC|proc_type);
3065 }
3066
3067 /*
3068 * mac_srs_worker
3069 *
3070 * The SRS worker routine. Drains the queue when no one else is
3071 * processing it.
3072 */
3073 void
mac_srs_worker(mac_soft_ring_set_t * mac_srs)3074 mac_srs_worker(mac_soft_ring_set_t *mac_srs)
3075 {
3076 kmutex_t *lock = &mac_srs->srs_lock;
3077 kcondvar_t *async = &mac_srs->srs_async;
3078 callb_cpr_t cprinfo;
3079 boolean_t bw_ctl_flag;
3080
3081 CALLB_CPR_INIT(&cprinfo, lock, callb_generic_cpr, "srs_worker");
3082 mutex_enter(lock);
3083
3084 start:
3085 for (;;) {
3086 bw_ctl_flag = B_FALSE;
3087 if (mac_srs->srs_type & SRST_BW_CONTROL) {
3088 MAC_SRS_BW_LOCK(mac_srs);
3089 MAC_SRS_CHECK_BW_CONTROL(mac_srs);
3090 if (mac_srs->srs_bw->mac_bw_state & SRS_BW_ENFORCED)
3091 bw_ctl_flag = B_TRUE;
3092 MAC_SRS_BW_UNLOCK(mac_srs);
3093 }
3094 /*
3095 * The SRS_BW_ENFORCED flag may change since we have dropped
3096 * the mac_bw_lock. However the drain function can handle both
3097 * a drainable SRS or a bandwidth controlled SRS, and the
3098 * effect of scheduling a timeout is to wakeup the worker
3099 * thread which in turn will call the drain function. Since
3100 * we release the srs_lock atomically only in the cv_wait there
3101 * isn't a fear of waiting for ever.
3102 */
3103 while (((mac_srs->srs_state & SRS_PROC) ||
3104 (mac_srs->srs_first == NULL) || bw_ctl_flag ||
3105 (mac_srs->srs_state & SRS_TX_BLOCKED)) &&
3106 !(mac_srs->srs_state & SRS_PAUSE)) {
3107 /*
3108 * If we have packets queued and we are here
3109 * because B/W control is in place, we better
3110 * schedule the worker wakeup after 1 tick
3111 * to see if bandwidth control can be relaxed.
3112 */
3113 if (bw_ctl_flag && mac_srs->srs_tid == NULL) {
3114 /*
3115 * We need to ensure that a timer is already
3116 * scheduled or we force schedule one for
3117 * later so that we can continue processing
3118 * after this quanta is over.
3119 */
3120 mac_srs->srs_tid = timeout(mac_srs_fire,
3121 mac_srs, 1);
3122 }
3123 wait:
3124 CALLB_CPR_SAFE_BEGIN(&cprinfo);
3125 cv_wait(async, lock);
3126 CALLB_CPR_SAFE_END(&cprinfo, lock);
3127
3128 if (mac_srs->srs_state & SRS_PAUSE)
3129 goto done;
3130 if (mac_srs->srs_state & SRS_PROC)
3131 goto wait;
3132
3133 if (mac_srs->srs_first != NULL &&
3134 mac_srs->srs_type & SRST_BW_CONTROL) {
3135 MAC_SRS_BW_LOCK(mac_srs);
3136 if (mac_srs->srs_bw->mac_bw_state &
3137 SRS_BW_ENFORCED) {
3138 MAC_SRS_CHECK_BW_CONTROL(mac_srs);
3139 }
3140 bw_ctl_flag = mac_srs->srs_bw->mac_bw_state &
3141 SRS_BW_ENFORCED;
3142 MAC_SRS_BW_UNLOCK(mac_srs);
3143 }
3144 }
3145
3146 if (mac_srs->srs_state & SRS_PAUSE)
3147 goto done;
3148 mac_srs->srs_drain_func(mac_srs, SRS_WORKER);
3149 }
3150 done:
3151 /*
3152 * The Rx SRS quiesce logic first cuts off packet supply to the SRS
3153 * from both hard and soft classifications and waits for such threads
3154 * to finish before signaling the worker. So at this point the only
3155 * thread left that could be competing with the worker is the poll
3156 * thread. In the case of Tx, there shouldn't be any thread holding
3157 * SRS_PROC at this point.
3158 */
3159 if (!(mac_srs->srs_state & SRS_PROC)) {
3160 mac_srs->srs_state |= SRS_PROC;
3161 } else {
3162 ASSERT((mac_srs->srs_type & SRST_TX) == 0);
3163 /*
3164 * Poll thread still owns the SRS and is still running
3165 */
3166 ASSERT((mac_srs->srs_poll_thr == NULL) ||
3167 ((mac_srs->srs_state & SRS_POLL_THR_OWNER) ==
3168 SRS_POLL_THR_OWNER));
3169 }
3170 mac_srs_worker_quiesce(mac_srs);
3171 /*
3172 * Wait for the SRS_RESTART or SRS_CONDEMNED signal from the initiator
3173 * of the quiesce operation
3174 */
3175 while (!(mac_srs->srs_state & (SRS_CONDEMNED | SRS_RESTART)))
3176 cv_wait(&mac_srs->srs_async, &mac_srs->srs_lock);
3177
3178 if (mac_srs->srs_state & SRS_RESTART) {
3179 ASSERT(!(mac_srs->srs_state & SRS_CONDEMNED));
3180 mac_srs_worker_restart(mac_srs);
3181 mac_srs->srs_state &= ~SRS_PROC;
3182 goto start;
3183 }
3184
3185 if (!(mac_srs->srs_state & SRS_CONDEMNED_DONE))
3186 mac_srs_worker_quiesce(mac_srs);
3187
3188 mac_srs->srs_state &= ~SRS_PROC;
3189 /* The macro drops the srs_lock */
3190 CALLB_CPR_EXIT(&cprinfo);
3191 thread_exit();
3192 }
3193
3194 /*
3195 * mac_rx_srs_subflow_process
3196 *
3197 * Receive side routine called from interrupt path when there are
3198 * sub flows present on this SRS.
3199 */
3200 /* ARGSUSED */
3201 void
mac_rx_srs_subflow_process(void * arg,mac_resource_handle_t srs,mblk_t * mp_chain,boolean_t loopback)3202 mac_rx_srs_subflow_process(void *arg, mac_resource_handle_t srs,
3203 mblk_t *mp_chain, boolean_t loopback)
3204 {
3205 flow_entry_t *flent = NULL;
3206 flow_entry_t *prev_flent = NULL;
3207 mblk_t *mp = NULL;
3208 mblk_t *tail = NULL;
3209 mac_soft_ring_set_t *mac_srs = (mac_soft_ring_set_t *)srs;
3210 mac_client_impl_t *mcip;
3211
3212 mcip = mac_srs->srs_mcip;
3213 ASSERT(mcip != NULL);
3214
3215 /*
3216 * We need to determine the SRS for every packet
3217 * by walking the flow table, if we don't get any,
3218 * then we proceed using the SRS we came with.
3219 */
3220 mp = tail = mp_chain;
3221 while (mp != NULL) {
3222
3223 /*
3224 * We will increment the stats for the matching subflow.
3225 * when we get the bytes/pkt count for the classified packets
3226 * later in mac_rx_srs_process.
3227 */
3228 (void) mac_flow_lookup(mcip->mci_subflow_tab, mp,
3229 FLOW_INBOUND, &flent);
3230
3231 if (mp == mp_chain || flent == prev_flent) {
3232 if (prev_flent != NULL)
3233 FLOW_REFRELE(prev_flent);
3234 prev_flent = flent;
3235 flent = NULL;
3236 tail = mp;
3237 mp = mp->b_next;
3238 continue;
3239 }
3240 tail->b_next = NULL;
3241 /*
3242 * A null indicates, this is for the mac_srs itself.
3243 * XXX-venu : probably assert for fe_rx_srs_cnt == 0.
3244 */
3245 if (prev_flent == NULL || prev_flent->fe_rx_srs_cnt == 0) {
3246 mac_rx_srs_process(arg,
3247 (mac_resource_handle_t)mac_srs, mp_chain,
3248 loopback);
3249 } else {
3250 (prev_flent->fe_cb_fn)(prev_flent->fe_cb_arg1,
3251 prev_flent->fe_cb_arg2, mp_chain, loopback);
3252 FLOW_REFRELE(prev_flent);
3253 }
3254 prev_flent = flent;
3255 flent = NULL;
3256 mp_chain = mp;
3257 tail = mp;
3258 mp = mp->b_next;
3259 }
3260 /* Last chain */
3261 ASSERT(mp_chain != NULL);
3262 if (prev_flent == NULL || prev_flent->fe_rx_srs_cnt == 0) {
3263 mac_rx_srs_process(arg,
3264 (mac_resource_handle_t)mac_srs, mp_chain, loopback);
3265 } else {
3266 (prev_flent->fe_cb_fn)(prev_flent->fe_cb_arg1,
3267 prev_flent->fe_cb_arg2, mp_chain, loopback);
3268 FLOW_REFRELE(prev_flent);
3269 }
3270 }
3271
3272 /*
3273 * MAC SRS receive side routine. If the data is coming from the
3274 * network (i.e. from a NIC) then this is called in interrupt context.
3275 * If the data is coming from a local sender (e.g. mac_tx_send() or
3276 * bridge_forward()) then this is not called in interrupt context.
3277 *
3278 * loopback is set to force a context switch on the loopback
3279 * path between MAC clients.
3280 */
3281 /* ARGSUSED */
3282 void
mac_rx_srs_process(void * arg,mac_resource_handle_t srs,mblk_t * mp_chain,boolean_t loopback)3283 mac_rx_srs_process(void *arg, mac_resource_handle_t srs, mblk_t *mp_chain,
3284 boolean_t loopback)
3285 {
3286 mac_soft_ring_set_t *mac_srs = (mac_soft_ring_set_t *)srs;
3287 mblk_t *mp, *tail, *head;
3288 int count = 0;
3289 int count1;
3290 size_t sz = 0;
3291 size_t chain_sz, sz1;
3292 mac_bw_ctl_t *mac_bw;
3293 mac_srs_rx_t *srs_rx = &mac_srs->srs_rx;
3294
3295 /*
3296 * Set the tail, count and sz. We set the sz irrespective
3297 * of whether we are doing B/W control or not for the
3298 * purpose of updating the stats.
3299 */
3300 mp = tail = mp_chain;
3301 while (mp != NULL) {
3302 tail = mp;
3303 count++;
3304 sz += msgdsize(mp);
3305 mp = mp->b_next;
3306 }
3307
3308 mutex_enter(&mac_srs->srs_lock);
3309
3310 if (loopback) {
3311 SRS_RX_STAT_UPDATE(mac_srs, lclbytes, sz);
3312 SRS_RX_STAT_UPDATE(mac_srs, lclcnt, count);
3313
3314 } else {
3315 SRS_RX_STAT_UPDATE(mac_srs, intrbytes, sz);
3316 SRS_RX_STAT_UPDATE(mac_srs, intrcnt, count);
3317 }
3318
3319 /*
3320 * If the SRS in already being processed; has been blanked;
3321 * can be processed by worker thread only; or the B/W limit
3322 * has been reached, then queue the chain and check if
3323 * worker thread needs to be awakend.
3324 */
3325 if (mac_srs->srs_type & SRST_BW_CONTROL) {
3326 mac_bw = mac_srs->srs_bw;
3327 ASSERT(mac_bw != NULL);
3328 mutex_enter(&mac_bw->mac_bw_lock);
3329 mac_bw->mac_bw_intr += sz;
3330 if (mac_bw->mac_bw_limit == 0) {
3331 /* zero bandwidth: drop all */
3332 srs_rx->sr_stat.mrs_sdrops += count;
3333 mac_bw->mac_bw_drop_bytes += sz;
3334 mutex_exit(&mac_bw->mac_bw_lock);
3335 mutex_exit(&mac_srs->srs_lock);
3336 mac_drop_chain(mp_chain, "Rx no bandwidth");
3337 return;
3338 } else {
3339 if ((mac_bw->mac_bw_sz + sz) <=
3340 mac_bw->mac_bw_drop_threshold) {
3341 mutex_exit(&mac_bw->mac_bw_lock);
3342 MAC_RX_SRS_ENQUEUE_CHAIN(mac_srs, mp_chain,
3343 tail, count, sz);
3344 } else {
3345 mp = mp_chain;
3346 chain_sz = 0;
3347 count1 = 0;
3348 tail = NULL;
3349 head = NULL;
3350 while (mp != NULL) {
3351 sz1 = msgdsize(mp);
3352 if (mac_bw->mac_bw_sz + chain_sz + sz1 >
3353 mac_bw->mac_bw_drop_threshold)
3354 break;
3355 chain_sz += sz1;
3356 count1++;
3357 tail = mp;
3358 mp = mp->b_next;
3359 }
3360 mutex_exit(&mac_bw->mac_bw_lock);
3361 if (tail != NULL) {
3362 head = tail->b_next;
3363 tail->b_next = NULL;
3364 MAC_RX_SRS_ENQUEUE_CHAIN(mac_srs,
3365 mp_chain, tail, count1, chain_sz);
3366 sz -= chain_sz;
3367 count -= count1;
3368 } else {
3369 /* Can't pick up any */
3370 head = mp_chain;
3371 }
3372 if (head != NULL) {
3373 /* Drop any packet over the threshold */
3374 srs_rx->sr_stat.mrs_sdrops += count;
3375 mutex_enter(&mac_bw->mac_bw_lock);
3376 mac_bw->mac_bw_drop_bytes += sz;
3377 mutex_exit(&mac_bw->mac_bw_lock);
3378 freemsgchain(head);
3379 }
3380 }
3381 MAC_SRS_WORKER_WAKEUP(mac_srs);
3382 mutex_exit(&mac_srs->srs_lock);
3383 return;
3384 }
3385 }
3386
3387 /*
3388 * If the total number of packets queued in the SRS and
3389 * its associated soft rings exceeds the max allowed,
3390 * then drop the chain. If we are polling capable, this
3391 * shouldn't be happening.
3392 */
3393 if (!(mac_srs->srs_type & SRST_BW_CONTROL) &&
3394 (srs_rx->sr_poll_pkt_cnt > srs_rx->sr_hiwat)) {
3395 mac_bw = mac_srs->srs_bw;
3396 srs_rx->sr_stat.mrs_sdrops += count;
3397 mutex_enter(&mac_bw->mac_bw_lock);
3398 mac_bw->mac_bw_drop_bytes += sz;
3399 mutex_exit(&mac_bw->mac_bw_lock);
3400 freemsgchain(mp_chain);
3401 mutex_exit(&mac_srs->srs_lock);
3402 return;
3403 }
3404
3405 MAC_RX_SRS_ENQUEUE_CHAIN(mac_srs, mp_chain, tail, count, sz);
3406
3407 if (!(mac_srs->srs_state & SRS_PROC)) {
3408 /*
3409 * If we are coming via loopback, if we are not optimizing for
3410 * latency, or if our stack is running deep, we should signal
3411 * the worker thread.
3412 */
3413 if (loopback || !(mac_srs->srs_state & SRS_LATENCY_OPT)) {
3414 /*
3415 * For loopback, We need to let the worker take
3416 * over as we don't want to continue in the same
3417 * thread even if we can. This could lead to stack
3418 * overflows and may also end up using
3419 * resources (cpu) incorrectly.
3420 */
3421 cv_signal(&mac_srs->srs_async);
3422 } else if (STACK_BIAS + (uintptr_t)getfp() -
3423 (uintptr_t)curthread->t_stkbase < mac_rx_srs_stack_needed) {
3424 if (++mac_rx_srs_stack_toodeep == 0)
3425 mac_rx_srs_stack_toodeep = 1;
3426 cv_signal(&mac_srs->srs_async);
3427 } else {
3428 /*
3429 * Seems like no one is processing the SRS and
3430 * there is no backlog. We also inline process
3431 * our packet if its a single packet in non
3432 * latency optimized case (in latency optimized
3433 * case, we inline process chains of any size).
3434 */
3435 mac_srs->srs_drain_func(mac_srs, SRS_PROC_FAST);
3436 }
3437 }
3438 mutex_exit(&mac_srs->srs_lock);
3439 }
3440
3441 /* TX SIDE ROUTINES (RUNTIME) */
3442
3443 /*
3444 * mac_tx_srs_no_desc
3445 *
3446 * This routine is called by Tx single ring default mode
3447 * when Tx ring runs out of descs.
3448 */
3449 mac_tx_cookie_t
mac_tx_srs_no_desc(mac_soft_ring_set_t * mac_srs,mblk_t * mp_chain,uint16_t flag,mblk_t ** ret_mp)3450 mac_tx_srs_no_desc(mac_soft_ring_set_t *mac_srs, mblk_t *mp_chain,
3451 uint16_t flag, mblk_t **ret_mp)
3452 {
3453 mac_tx_cookie_t cookie = 0;
3454 mac_srs_tx_t *srs_tx = &mac_srs->srs_tx;
3455 boolean_t wakeup_worker = B_TRUE;
3456 uint32_t tx_mode = srs_tx->st_mode;
3457 int cnt, sz;
3458 mblk_t *tail;
3459
3460 ASSERT(tx_mode == SRS_TX_DEFAULT || tx_mode == SRS_TX_BW);
3461 if (flag & MAC_DROP_ON_NO_DESC) {
3462 MAC_TX_SRS_DROP_MESSAGE(mac_srs, mp_chain, cookie,
3463 "Tx no desc");
3464 } else {
3465 if (mac_srs->srs_first != NULL)
3466 wakeup_worker = B_FALSE;
3467 MAC_COUNT_CHAIN(mac_srs, mp_chain, tail, cnt, sz);
3468 if (flag & MAC_TX_NO_ENQUEUE) {
3469 /*
3470 * If TX_QUEUED is not set, queue the
3471 * packet and let mac_tx_srs_drain()
3472 * set the TX_BLOCKED bit for the
3473 * reasons explained above. Otherwise,
3474 * return the mblks.
3475 */
3476 if (wakeup_worker) {
3477 MAC_TX_SRS_ENQUEUE_CHAIN(mac_srs,
3478 mp_chain, tail, cnt, sz);
3479 } else {
3480 MAC_TX_SET_NO_ENQUEUE(mac_srs,
3481 mp_chain, ret_mp, cookie);
3482 }
3483 } else {
3484 MAC_TX_SRS_TEST_HIWAT(mac_srs, mp_chain,
3485 tail, cnt, sz, cookie);
3486 }
3487 if (wakeup_worker)
3488 cv_signal(&mac_srs->srs_async);
3489 }
3490 return (cookie);
3491 }
3492
3493 /*
3494 * mac_tx_srs_enqueue
3495 *
3496 * This routine is called when Tx SRS is operating in either serializer
3497 * or bandwidth mode. In serializer mode, a packet will get enqueued
3498 * when a thread cannot enter SRS exclusively. In bandwidth mode,
3499 * packets gets queued if allowed byte-count limit for a tick is
3500 * exceeded. The action that gets taken when MAC_DROP_ON_NO_DESC and
3501 * MAC_TX_NO_ENQUEUE is set is different than when operaing in either
3502 * the default mode or fanout mode. Here packets get dropped or
3503 * returned back to the caller only after hi-watermark worth of data
3504 * is queued.
3505 */
3506 static mac_tx_cookie_t
mac_tx_srs_enqueue(mac_soft_ring_set_t * mac_srs,mblk_t * mp_chain,uint16_t flag,uintptr_t fanout_hint,mblk_t ** ret_mp)3507 mac_tx_srs_enqueue(mac_soft_ring_set_t *mac_srs, mblk_t *mp_chain,
3508 uint16_t flag, uintptr_t fanout_hint, mblk_t **ret_mp)
3509 {
3510 mac_tx_cookie_t cookie = 0;
3511 int cnt, sz;
3512 mblk_t *tail;
3513 boolean_t wakeup_worker = B_TRUE;
3514
3515 /*
3516 * Ignore fanout hint if we don't have multiple tx rings.
3517 */
3518 if (!MAC_TX_SOFT_RINGS(mac_srs))
3519 fanout_hint = 0;
3520
3521 if (mac_srs->srs_first != NULL)
3522 wakeup_worker = B_FALSE;
3523 MAC_COUNT_CHAIN(mac_srs, mp_chain, tail, cnt, sz);
3524 if (flag & MAC_DROP_ON_NO_DESC) {
3525 if (mac_srs->srs_count > mac_srs->srs_tx.st_hiwat) {
3526 MAC_TX_SRS_DROP_MESSAGE(mac_srs, mp_chain, cookie,
3527 "Tx SRS hiwat");
3528 } else {
3529 MAC_TX_SRS_ENQUEUE_CHAIN(mac_srs,
3530 mp_chain, tail, cnt, sz);
3531 }
3532 } else if (flag & MAC_TX_NO_ENQUEUE) {
3533 if ((mac_srs->srs_count > mac_srs->srs_tx.st_hiwat) ||
3534 (mac_srs->srs_state & SRS_TX_WAKEUP_CLIENT)) {
3535 MAC_TX_SET_NO_ENQUEUE(mac_srs, mp_chain,
3536 ret_mp, cookie);
3537 } else {
3538 mp_chain->b_prev = (mblk_t *)fanout_hint;
3539 MAC_TX_SRS_ENQUEUE_CHAIN(mac_srs,
3540 mp_chain, tail, cnt, sz);
3541 }
3542 } else {
3543 /*
3544 * If you are BW_ENFORCED, just enqueue the
3545 * packet. srs_worker will drain it at the
3546 * prescribed rate. Before enqueueing, save
3547 * the fanout hint.
3548 */
3549 mp_chain->b_prev = (mblk_t *)fanout_hint;
3550 MAC_TX_SRS_TEST_HIWAT(mac_srs, mp_chain,
3551 tail, cnt, sz, cookie);
3552 }
3553 if (wakeup_worker)
3554 cv_signal(&mac_srs->srs_async);
3555 return (cookie);
3556 }
3557
3558 /*
3559 * There are seven tx modes:
3560 *
3561 * 1) Default mode (SRS_TX_DEFAULT)
3562 * 2) Serialization mode (SRS_TX_SERIALIZE)
3563 * 3) Fanout mode (SRS_TX_FANOUT)
3564 * 4) Bandwdith mode (SRS_TX_BW)
3565 * 5) Fanout and Bandwidth mode (SRS_TX_BW_FANOUT)
3566 * 6) aggr Tx mode (SRS_TX_AGGR)
3567 * 7) aggr Tx bw mode (SRS_TX_BW_AGGR)
3568 *
3569 * The tx mode in which an SRS operates is decided in mac_tx_srs_setup()
3570 * based on the number of Tx rings requested for an SRS and whether
3571 * bandwidth control is requested or not.
3572 *
3573 * The default mode (i.e., no fanout/no bandwidth) is used when the
3574 * underlying NIC does not have Tx rings or just one Tx ring. In this mode,
3575 * the SRS acts as a pass-thru. Packets will go directly to mac_tx_send().
3576 * When the underlying Tx ring runs out of Tx descs, it starts queueing up
3577 * packets in SRS. When flow-control is relieved, the srs_worker drains
3578 * the queued packets and informs blocked clients to restart sending
3579 * packets.
3580 *
3581 * In the SRS_TX_SERIALIZE mode, all calls to mac_tx() are serialized. This
3582 * mode is used when the link has no Tx rings or only one Tx ring.
3583 *
3584 * In the SRS_TX_FANOUT mode, packets will be fanned out to multiple
3585 * Tx rings. Each Tx ring will have a soft ring associated with it.
3586 * These soft rings will be hung off the Tx SRS. Queueing if it happens
3587 * due to lack of Tx desc will be in individual soft ring (and not srs)
3588 * associated with Tx ring.
3589 *
3590 * In the TX_BW mode, tx srs will allow packets to go down to Tx ring
3591 * only if bw is available. Otherwise the packets will be queued in
3592 * SRS. If fanout to multiple Tx rings is configured, the packets will
3593 * be fanned out among the soft rings associated with the Tx rings.
3594 *
3595 * In SRS_TX_AGGR mode, mac_tx_aggr_mode() routine is called. This routine
3596 * invokes an aggr function, aggr_find_tx_ring(), to find a pseudo Tx ring
3597 * belonging to a port on which the packet has to be sent. Aggr will
3598 * always have a pseudo Tx ring associated with it even when it is an
3599 * aggregation over a single NIC that has no Tx rings. Even in such a
3600 * case, the single pseudo Tx ring will have a soft ring associated with
3601 * it and the soft ring will hang off the SRS.
3602 *
3603 * If a bandwidth is specified for an aggr, SRS_TX_BW_AGGR mode is used.
3604 * In this mode, the bandwidth is first applied on the outgoing packets
3605 * and later mac_tx_addr_mode() function is called to send the packet out
3606 * of one of the pseudo Tx rings.
3607 *
3608 * Four flags are used in srs_state for indicating flow control
3609 * conditions : SRS_TX_BLOCKED, SRS_TX_HIWAT, SRS_TX_WAKEUP_CLIENT.
3610 * SRS_TX_BLOCKED indicates out of Tx descs. SRS expects a wakeup from the
3611 * driver below.
3612 * SRS_TX_HIWAT indicates packet count enqueued in Tx SRS exceeded Tx hiwat
3613 * and flow-control pressure is applied back to clients. The clients expect
3614 * wakeup when flow-control is relieved.
3615 * SRS_TX_WAKEUP_CLIENT get set when (flag == MAC_TX_NO_ENQUEUE) and mblk
3616 * got returned back to client either due to lack of Tx descs or due to bw
3617 * control reasons. The clients expect a wakeup when condition is relieved.
3618 *
3619 * The fourth argument to mac_tx() is the flag. Normally it will be 0 but
3620 * some clients set the following values too: MAC_DROP_ON_NO_DESC,
3621 * MAC_TX_NO_ENQUEUE
3622 * Mac clients that do not want packets to be enqueued in the mac layer set
3623 * MAC_DROP_ON_NO_DESC value. The packets won't be queued in the Tx SRS or
3624 * Tx soft rings but instead get dropped when the NIC runs out of desc. The
3625 * behaviour of this flag is different when the Tx is running in serializer
3626 * or bandwidth mode. Under these (Serializer, bandwidth) modes, the packet
3627 * get dropped when Tx high watermark is reached.
3628 * There are some mac clients like vsw, aggr that want the mblks to be
3629 * returned back to clients instead of being queued in Tx SRS (or Tx soft
3630 * rings) under flow-control (i.e., out of desc or exceeding bw limits)
3631 * conditions. These clients call mac_tx() with MAC_TX_NO_ENQUEUE flag set.
3632 * In the default and Tx fanout mode, the un-transmitted mblks will be
3633 * returned back to the clients when the driver runs out of Tx descs.
3634 * SRS_TX_WAKEUP_CLIENT (or S_RING_WAKEUP_CLIENT) will be set in SRS (or
3635 * soft ring) so that the clients can be woken up when Tx desc become
3636 * available. When running in serializer or bandwidth mode mode,
3637 * SRS_TX_WAKEUP_CLIENT will be set when tx hi-watermark is reached.
3638 */
3639
3640 mac_tx_func_t
mac_tx_get_func(uint32_t mode)3641 mac_tx_get_func(uint32_t mode)
3642 {
3643 return (mac_tx_mode_list[mode].mac_tx_func);
3644 }
3645
3646 /* ARGSUSED */
3647 static mac_tx_cookie_t
mac_tx_single_ring_mode(mac_soft_ring_set_t * mac_srs,mblk_t * mp_chain,uintptr_t fanout_hint,uint16_t flag,mblk_t ** ret_mp)3648 mac_tx_single_ring_mode(mac_soft_ring_set_t *mac_srs, mblk_t *mp_chain,
3649 uintptr_t fanout_hint, uint16_t flag, mblk_t **ret_mp)
3650 {
3651 mac_srs_tx_t *srs_tx = &mac_srs->srs_tx;
3652 mac_tx_stats_t stats;
3653 mac_tx_cookie_t cookie = 0;
3654
3655 ASSERT(srs_tx->st_mode == SRS_TX_DEFAULT);
3656
3657 /* Regular case with a single Tx ring */
3658 /*
3659 * SRS_TX_BLOCKED is set when underlying NIC runs
3660 * out of Tx descs and messages start getting
3661 * queued. It won't get reset until
3662 * tx_srs_drain() completely drains out the
3663 * messages.
3664 */
3665 if ((mac_srs->srs_state & SRS_ENQUEUED) != 0) {
3666 /* Tx descs/resources not available */
3667 mutex_enter(&mac_srs->srs_lock);
3668 if ((mac_srs->srs_state & SRS_ENQUEUED) != 0) {
3669 cookie = mac_tx_srs_no_desc(mac_srs, mp_chain,
3670 flag, ret_mp);
3671 mutex_exit(&mac_srs->srs_lock);
3672 return (cookie);
3673 }
3674 /*
3675 * While we were computing mblk count, the
3676 * flow control condition got relieved.
3677 * Continue with the transmission.
3678 */
3679 mutex_exit(&mac_srs->srs_lock);
3680 }
3681
3682 mp_chain = mac_tx_send(srs_tx->st_arg1, srs_tx->st_arg2,
3683 mp_chain, &stats);
3684
3685 /*
3686 * Multiple threads could be here sending packets.
3687 * Under such conditions, it is not possible to
3688 * automically set SRS_TX_BLOCKED bit to indicate
3689 * out of tx desc condition. To atomically set
3690 * this, we queue the returned packet and do
3691 * the setting of SRS_TX_BLOCKED in
3692 * mac_tx_srs_drain().
3693 */
3694 if (mp_chain != NULL) {
3695 mutex_enter(&mac_srs->srs_lock);
3696 cookie = mac_tx_srs_no_desc(mac_srs, mp_chain, flag, ret_mp);
3697 mutex_exit(&mac_srs->srs_lock);
3698 return (cookie);
3699 }
3700 SRS_TX_STATS_UPDATE(mac_srs, &stats);
3701
3702 return (0);
3703 }
3704
3705 /*
3706 * mac_tx_serialize_mode
3707 *
3708 * This is an experimental mode implemented as per the request of PAE.
3709 * In this mode, all callers attempting to send a packet to the NIC
3710 * will get serialized. Only one thread at any time will access the
3711 * NIC to send the packet out.
3712 */
3713 /* ARGSUSED */
3714 static mac_tx_cookie_t
mac_tx_serializer_mode(mac_soft_ring_set_t * mac_srs,mblk_t * mp_chain,uintptr_t fanout_hint,uint16_t flag,mblk_t ** ret_mp)3715 mac_tx_serializer_mode(mac_soft_ring_set_t *mac_srs, mblk_t *mp_chain,
3716 uintptr_t fanout_hint, uint16_t flag, mblk_t **ret_mp)
3717 {
3718 mac_tx_stats_t stats;
3719 mac_tx_cookie_t cookie = 0;
3720 mac_srs_tx_t *srs_tx = &mac_srs->srs_tx;
3721
3722 /* Single ring, serialize below */
3723 ASSERT(srs_tx->st_mode == SRS_TX_SERIALIZE);
3724 mutex_enter(&mac_srs->srs_lock);
3725 if ((mac_srs->srs_first != NULL) ||
3726 (mac_srs->srs_state & SRS_PROC)) {
3727 /*
3728 * In serialization mode, queue all packets until
3729 * TX_HIWAT is set.
3730 * If drop bit is set, drop if TX_HIWAT is set.
3731 * If no_enqueue is set, still enqueue until hiwat
3732 * is set and return mblks after TX_HIWAT is set.
3733 */
3734 cookie = mac_tx_srs_enqueue(mac_srs, mp_chain,
3735 flag, 0, ret_mp);
3736 mutex_exit(&mac_srs->srs_lock);
3737 return (cookie);
3738 }
3739 /*
3740 * No packets queued, nothing on proc and no flow
3741 * control condition. Fast-path, ok. Do inline
3742 * processing.
3743 */
3744 mac_srs->srs_state |= SRS_PROC;
3745 mutex_exit(&mac_srs->srs_lock);
3746
3747 mp_chain = mac_tx_send(srs_tx->st_arg1, srs_tx->st_arg2,
3748 mp_chain, &stats);
3749
3750 mutex_enter(&mac_srs->srs_lock);
3751 mac_srs->srs_state &= ~SRS_PROC;
3752 if (mp_chain != NULL) {
3753 cookie = mac_tx_srs_enqueue(mac_srs,
3754 mp_chain, flag, 0, ret_mp);
3755 }
3756 if (mac_srs->srs_first != NULL) {
3757 /*
3758 * We processed inline our packet and a new
3759 * packet/s got queued while we were
3760 * processing. Wakeup srs worker
3761 */
3762 cv_signal(&mac_srs->srs_async);
3763 }
3764 mutex_exit(&mac_srs->srs_lock);
3765
3766 if (cookie == 0)
3767 SRS_TX_STATS_UPDATE(mac_srs, &stats);
3768
3769 return (cookie);
3770 }
3771
3772 /*
3773 * mac_tx_fanout_mode
3774 *
3775 * In this mode, the SRS will have access to multiple Tx rings to send
3776 * the packet out. The fanout hint that is passed as an argument is
3777 * used to find an appropriate ring to fanout the traffic. Each Tx
3778 * ring, in turn, will have a soft ring associated with it. If a Tx
3779 * ring runs out of Tx desc's the returned packet will be queued in
3780 * the soft ring associated with that Tx ring. The srs itself will not
3781 * queue any packets.
3782 */
3783
3784 #define MAC_TX_SOFT_RING_PROCESS(chain) { \
3785 index = COMPUTE_INDEX(hash, mac_srs->srs_tx_ring_count), \
3786 softring = mac_srs->srs_tx_soft_rings[index]; \
3787 cookie = mac_tx_soft_ring_process(softring, chain, flag, ret_mp); \
3788 DTRACE_PROBE2(tx__fanout, uint64_t, hash, uint_t, index); \
3789 }
3790
3791 static mac_tx_cookie_t
mac_tx_fanout_mode(mac_soft_ring_set_t * mac_srs,mblk_t * mp_chain,uintptr_t fanout_hint,uint16_t flag,mblk_t ** ret_mp)3792 mac_tx_fanout_mode(mac_soft_ring_set_t *mac_srs, mblk_t *mp_chain,
3793 uintptr_t fanout_hint, uint16_t flag, mblk_t **ret_mp)
3794 {
3795 mac_soft_ring_t *softring;
3796 uint64_t hash;
3797 uint_t index;
3798 mac_tx_cookie_t cookie = 0;
3799
3800 ASSERT(mac_srs->srs_tx.st_mode == SRS_TX_FANOUT ||
3801 mac_srs->srs_tx.st_mode == SRS_TX_BW_FANOUT);
3802 if (fanout_hint != 0) {
3803 /*
3804 * The hint is specified by the caller, simply pass the
3805 * whole chain to the soft ring.
3806 */
3807 hash = HASH_HINT(fanout_hint);
3808 MAC_TX_SOFT_RING_PROCESS(mp_chain);
3809 } else {
3810 mblk_t *last_mp, *cur_mp, *sub_chain;
3811 uint64_t last_hash = 0;
3812 uint_t media = mac_srs->srs_mcip->mci_mip->mi_info.mi_media;
3813
3814 /*
3815 * Compute the hash from the contents (headers) of the
3816 * packets of the mblk chain. Split the chains into
3817 * subchains of the same conversation.
3818 *
3819 * Since there may be more than one ring used for
3820 * sub-chains of the same call, and since the caller
3821 * does not maintain per conversation state since it
3822 * passed a zero hint, unsent subchains will be
3823 * dropped.
3824 */
3825
3826 flag |= MAC_DROP_ON_NO_DESC;
3827 ret_mp = NULL;
3828
3829 ASSERT(ret_mp == NULL);
3830
3831 sub_chain = NULL;
3832 last_mp = NULL;
3833
3834 for (cur_mp = mp_chain; cur_mp != NULL;
3835 cur_mp = cur_mp->b_next) {
3836 hash = mac_pkt_hash(media, cur_mp, MAC_PKT_HASH_L4,
3837 B_TRUE);
3838 if (last_hash != 0 && hash != last_hash) {
3839 /*
3840 * Starting a different subchain, send current
3841 * chain out.
3842 */
3843 ASSERT(last_mp != NULL);
3844 last_mp->b_next = NULL;
3845 MAC_TX_SOFT_RING_PROCESS(sub_chain);
3846 sub_chain = NULL;
3847 }
3848
3849 /* add packet to subchain */
3850 if (sub_chain == NULL)
3851 sub_chain = cur_mp;
3852 last_mp = cur_mp;
3853 last_hash = hash;
3854 }
3855
3856 if (sub_chain != NULL) {
3857 /* send last subchain */
3858 ASSERT(last_mp != NULL);
3859 last_mp->b_next = NULL;
3860 MAC_TX_SOFT_RING_PROCESS(sub_chain);
3861 }
3862
3863 cookie = 0;
3864 }
3865
3866 return (cookie);
3867 }
3868
3869 /*
3870 * mac_tx_bw_mode
3871 *
3872 * In the bandwidth mode, Tx srs will allow packets to go down to Tx ring
3873 * only if bw is available. Otherwise the packets will be queued in
3874 * SRS. If the SRS has multiple Tx rings, then packets will get fanned
3875 * out to a Tx rings.
3876 */
3877 static mac_tx_cookie_t
mac_tx_bw_mode(mac_soft_ring_set_t * mac_srs,mblk_t * mp_chain,uintptr_t fanout_hint,uint16_t flag,mblk_t ** ret_mp)3878 mac_tx_bw_mode(mac_soft_ring_set_t *mac_srs, mblk_t *mp_chain,
3879 uintptr_t fanout_hint, uint16_t flag, mblk_t **ret_mp)
3880 {
3881 int cnt, sz;
3882 mblk_t *tail;
3883 mac_tx_cookie_t cookie = 0;
3884 mac_srs_tx_t *srs_tx = &mac_srs->srs_tx;
3885 clock_t now;
3886
3887 ASSERT(TX_BANDWIDTH_MODE(mac_srs));
3888 ASSERT(mac_srs->srs_type & SRST_BW_CONTROL);
3889 mutex_enter(&mac_srs->srs_lock);
3890 if (mac_srs->srs_bw->mac_bw_limit == 0) {
3891 /*
3892 * zero bandwidth, no traffic is sent: drop the packets,
3893 * or return the whole chain if the caller requests all
3894 * unsent packets back.
3895 */
3896 if (flag & MAC_TX_NO_ENQUEUE) {
3897 cookie = (mac_tx_cookie_t)mac_srs;
3898 *ret_mp = mp_chain;
3899 } else {
3900 MAC_TX_SRS_DROP_MESSAGE(mac_srs, mp_chain, cookie,
3901 "Tx no bandwidth");
3902 }
3903 mutex_exit(&mac_srs->srs_lock);
3904 return (cookie);
3905 } else if ((mac_srs->srs_first != NULL) ||
3906 (mac_srs->srs_bw->mac_bw_state & SRS_BW_ENFORCED)) {
3907 cookie = mac_tx_srs_enqueue(mac_srs, mp_chain, flag,
3908 fanout_hint, ret_mp);
3909 mutex_exit(&mac_srs->srs_lock);
3910 return (cookie);
3911 }
3912 MAC_COUNT_CHAIN(mac_srs, mp_chain, tail, cnt, sz);
3913 now = ddi_get_lbolt();
3914 if (mac_srs->srs_bw->mac_bw_curr_time != now) {
3915 mac_srs->srs_bw->mac_bw_curr_time = now;
3916 mac_srs->srs_bw->mac_bw_used = 0;
3917 } else if (mac_srs->srs_bw->mac_bw_used >
3918 mac_srs->srs_bw->mac_bw_limit) {
3919 mac_srs->srs_bw->mac_bw_state |= SRS_BW_ENFORCED;
3920 MAC_TX_SRS_ENQUEUE_CHAIN(mac_srs,
3921 mp_chain, tail, cnt, sz);
3922 /*
3923 * Wakeup worker thread. Note that worker
3924 * thread has to be woken up so that it
3925 * can fire up the timer to be woken up
3926 * on the next tick. Also once
3927 * BW_ENFORCED is set, it can only be
3928 * reset by srs_worker thread. Until then
3929 * all packets will get queued up in SRS
3930 * and hence this this code path won't be
3931 * entered until BW_ENFORCED is reset.
3932 */
3933 cv_signal(&mac_srs->srs_async);
3934 mutex_exit(&mac_srs->srs_lock);
3935 return (cookie);
3936 }
3937
3938 mac_srs->srs_bw->mac_bw_used += sz;
3939 mutex_exit(&mac_srs->srs_lock);
3940
3941 if (srs_tx->st_mode == SRS_TX_BW_FANOUT) {
3942 mac_soft_ring_t *softring;
3943 uint_t indx, hash;
3944
3945 hash = HASH_HINT(fanout_hint);
3946 indx = COMPUTE_INDEX(hash,
3947 mac_srs->srs_tx_ring_count);
3948 softring = mac_srs->srs_tx_soft_rings[indx];
3949 return (mac_tx_soft_ring_process(softring, mp_chain, flag,
3950 ret_mp));
3951 } else if (srs_tx->st_mode == SRS_TX_BW_AGGR) {
3952 return (mac_tx_aggr_mode(mac_srs, mp_chain,
3953 fanout_hint, flag, ret_mp));
3954 } else {
3955 mac_tx_stats_t stats;
3956
3957 mp_chain = mac_tx_send(srs_tx->st_arg1, srs_tx->st_arg2,
3958 mp_chain, &stats);
3959
3960 if (mp_chain != NULL) {
3961 mutex_enter(&mac_srs->srs_lock);
3962 MAC_COUNT_CHAIN(mac_srs, mp_chain, tail, cnt, sz);
3963 if (mac_srs->srs_bw->mac_bw_used > sz)
3964 mac_srs->srs_bw->mac_bw_used -= sz;
3965 else
3966 mac_srs->srs_bw->mac_bw_used = 0;
3967 cookie = mac_tx_srs_enqueue(mac_srs, mp_chain, flag,
3968 fanout_hint, ret_mp);
3969 mutex_exit(&mac_srs->srs_lock);
3970 return (cookie);
3971 }
3972 SRS_TX_STATS_UPDATE(mac_srs, &stats);
3973
3974 return (0);
3975 }
3976 }
3977
3978 /*
3979 * mac_tx_aggr_mode
3980 *
3981 * This routine invokes an aggr function, aggr_find_tx_ring(), to find
3982 * a (pseudo) Tx ring belonging to a port on which the packet has to
3983 * be sent. aggr_find_tx_ring() first finds the outgoing port based on
3984 * L2/L3/L4 policy and then uses the fanout_hint passed to it to pick
3985 * a Tx ring from the selected port.
3986 *
3987 * Note that a port can be deleted from the aggregation. In such a case,
3988 * the aggregation layer first separates the port from the rest of the
3989 * ports making sure that port (and thus any Tx rings associated with
3990 * it) won't get selected in the call to aggr_find_tx_ring() function.
3991 * Later calls are made to mac_group_rem_ring() passing pseudo Tx ring
3992 * handles one by one which in turn will quiesce the Tx SRS and remove
3993 * the soft ring associated with the pseudo Tx ring. Unlike Rx side
3994 * where a cookie is used to protect against mac_rx_ring() calls on
3995 * rings that have been removed, no such cookie is needed on the Tx
3996 * side as the pseudo Tx ring won't be available anymore to
3997 * aggr_find_tx_ring() once the port has been removed.
3998 */
3999 static mac_tx_cookie_t
mac_tx_aggr_mode(mac_soft_ring_set_t * mac_srs,mblk_t * mp_chain,uintptr_t fanout_hint,uint16_t flag,mblk_t ** ret_mp)4000 mac_tx_aggr_mode(mac_soft_ring_set_t *mac_srs, mblk_t *mp_chain,
4001 uintptr_t fanout_hint, uint16_t flag, mblk_t **ret_mp)
4002 {
4003 mac_srs_tx_t *srs_tx = &mac_srs->srs_tx;
4004 mac_tx_ring_fn_t find_tx_ring_fn;
4005 mac_ring_handle_t ring = NULL;
4006 void *arg;
4007 mac_soft_ring_t *sringp;
4008
4009 find_tx_ring_fn = srs_tx->st_capab_aggr.mca_find_tx_ring_fn;
4010 arg = srs_tx->st_capab_aggr.mca_arg;
4011 if (find_tx_ring_fn(arg, mp_chain, fanout_hint, &ring) == NULL)
4012 return (0);
4013 sringp = srs_tx->st_soft_rings[((mac_ring_t *)ring)->mr_index];
4014 return (mac_tx_soft_ring_process(sringp, mp_chain, flag, ret_mp));
4015 }
4016
4017 void
mac_tx_invoke_callbacks(mac_client_impl_t * mcip,mac_tx_cookie_t cookie)4018 mac_tx_invoke_callbacks(mac_client_impl_t *mcip, mac_tx_cookie_t cookie)
4019 {
4020 mac_cb_t *mcb;
4021 mac_tx_notify_cb_t *mtnfp;
4022
4023 /* Wakeup callback registered clients */
4024 MAC_CALLBACK_WALKER_INC(&mcip->mci_tx_notify_cb_info);
4025 for (mcb = mcip->mci_tx_notify_cb_list; mcb != NULL;
4026 mcb = mcb->mcb_nextp) {
4027 mtnfp = (mac_tx_notify_cb_t *)mcb->mcb_objp;
4028 mtnfp->mtnf_fn(mtnfp->mtnf_arg, cookie);
4029 }
4030 MAC_CALLBACK_WALKER_DCR(&mcip->mci_tx_notify_cb_info,
4031 &mcip->mci_tx_notify_cb_list);
4032 }
4033
4034 /* ARGSUSED */
4035 void
mac_tx_srs_drain(mac_soft_ring_set_t * mac_srs,uint_t proc_type)4036 mac_tx_srs_drain(mac_soft_ring_set_t *mac_srs, uint_t proc_type)
4037 {
4038 mblk_t *head, *tail;
4039 size_t sz;
4040 uint32_t tx_mode;
4041 uint_t saved_pkt_count;
4042 mac_tx_stats_t stats;
4043 mac_srs_tx_t *srs_tx = &mac_srs->srs_tx;
4044 clock_t now;
4045
4046 saved_pkt_count = 0;
4047 ASSERT(mutex_owned(&mac_srs->srs_lock));
4048 ASSERT(!(mac_srs->srs_state & SRS_PROC));
4049
4050 mac_srs->srs_state |= SRS_PROC;
4051
4052 tx_mode = srs_tx->st_mode;
4053 if (tx_mode == SRS_TX_DEFAULT || tx_mode == SRS_TX_SERIALIZE) {
4054 if (mac_srs->srs_first != NULL) {
4055 head = mac_srs->srs_first;
4056 tail = mac_srs->srs_last;
4057 saved_pkt_count = mac_srs->srs_count;
4058 mac_srs->srs_first = NULL;
4059 mac_srs->srs_last = NULL;
4060 mac_srs->srs_count = 0;
4061 mutex_exit(&mac_srs->srs_lock);
4062
4063 head = mac_tx_send(srs_tx->st_arg1, srs_tx->st_arg2,
4064 head, &stats);
4065
4066 mutex_enter(&mac_srs->srs_lock);
4067 if (head != NULL) {
4068 /* Device out of tx desc, set block */
4069 if (head->b_next == NULL)
4070 VERIFY(head == tail);
4071 tail->b_next = mac_srs->srs_first;
4072 mac_srs->srs_first = head;
4073 mac_srs->srs_count +=
4074 (saved_pkt_count - stats.mts_opackets);
4075 if (mac_srs->srs_last == NULL)
4076 mac_srs->srs_last = tail;
4077 MAC_TX_SRS_BLOCK(mac_srs, head);
4078 } else {
4079 srs_tx->st_woken_up = B_FALSE;
4080 SRS_TX_STATS_UPDATE(mac_srs, &stats);
4081 }
4082 }
4083 } else if (tx_mode == SRS_TX_BW) {
4084 /*
4085 * We are here because the timer fired and we have some data
4086 * to tranmit. Also mac_tx_srs_worker should have reset
4087 * SRS_BW_ENFORCED flag
4088 */
4089 ASSERT(!(mac_srs->srs_bw->mac_bw_state & SRS_BW_ENFORCED));
4090 head = tail = mac_srs->srs_first;
4091 while (mac_srs->srs_first != NULL) {
4092 tail = mac_srs->srs_first;
4093 tail->b_prev = NULL;
4094 mac_srs->srs_first = tail->b_next;
4095 if (mac_srs->srs_first == NULL)
4096 mac_srs->srs_last = NULL;
4097 mac_srs->srs_count--;
4098 sz = msgdsize(tail);
4099 mac_srs->srs_size -= sz;
4100 saved_pkt_count++;
4101 MAC_TX_UPDATE_BW_INFO(mac_srs, sz);
4102
4103 if (mac_srs->srs_bw->mac_bw_used <
4104 mac_srs->srs_bw->mac_bw_limit)
4105 continue;
4106
4107 now = ddi_get_lbolt();
4108 if (mac_srs->srs_bw->mac_bw_curr_time != now) {
4109 mac_srs->srs_bw->mac_bw_curr_time = now;
4110 mac_srs->srs_bw->mac_bw_used = sz;
4111 continue;
4112 }
4113 mac_srs->srs_bw->mac_bw_state |= SRS_BW_ENFORCED;
4114 break;
4115 }
4116
4117 ASSERT((head == NULL && tail == NULL) ||
4118 (head != NULL && tail != NULL));
4119 if (tail != NULL) {
4120 tail->b_next = NULL;
4121 mutex_exit(&mac_srs->srs_lock);
4122
4123 head = mac_tx_send(srs_tx->st_arg1, srs_tx->st_arg2,
4124 head, &stats);
4125
4126 mutex_enter(&mac_srs->srs_lock);
4127 if (head != NULL) {
4128 uint_t size_sent;
4129
4130 /* Device out of tx desc, set block */
4131 if (head->b_next == NULL)
4132 VERIFY(head == tail);
4133 tail->b_next = mac_srs->srs_first;
4134 mac_srs->srs_first = head;
4135 mac_srs->srs_count +=
4136 (saved_pkt_count - stats.mts_opackets);
4137 if (mac_srs->srs_last == NULL)
4138 mac_srs->srs_last = tail;
4139 size_sent = sz - stats.mts_obytes;
4140 mac_srs->srs_size += size_sent;
4141 mac_srs->srs_bw->mac_bw_sz += size_sent;
4142 if (mac_srs->srs_bw->mac_bw_used > size_sent) {
4143 mac_srs->srs_bw->mac_bw_used -=
4144 size_sent;
4145 } else {
4146 mac_srs->srs_bw->mac_bw_used = 0;
4147 }
4148 MAC_TX_SRS_BLOCK(mac_srs, head);
4149 } else {
4150 srs_tx->st_woken_up = B_FALSE;
4151 SRS_TX_STATS_UPDATE(mac_srs, &stats);
4152 }
4153 }
4154 } else if (tx_mode == SRS_TX_BW_FANOUT || tx_mode == SRS_TX_BW_AGGR) {
4155 mblk_t *prev;
4156 uint64_t hint;
4157
4158 /*
4159 * We are here because the timer fired and we
4160 * have some quota to tranmit.
4161 */
4162 prev = NULL;
4163 head = tail = mac_srs->srs_first;
4164 while (mac_srs->srs_first != NULL) {
4165 tail = mac_srs->srs_first;
4166 mac_srs->srs_first = tail->b_next;
4167 if (mac_srs->srs_first == NULL)
4168 mac_srs->srs_last = NULL;
4169 mac_srs->srs_count--;
4170 sz = msgdsize(tail);
4171 mac_srs->srs_size -= sz;
4172 mac_srs->srs_bw->mac_bw_used += sz;
4173 if (prev == NULL)
4174 hint = (ulong_t)tail->b_prev;
4175 if (hint != (ulong_t)tail->b_prev) {
4176 prev->b_next = NULL;
4177 mutex_exit(&mac_srs->srs_lock);
4178 TX_SRS_TO_SOFT_RING(mac_srs, head, hint);
4179 head = tail;
4180 hint = (ulong_t)tail->b_prev;
4181 mutex_enter(&mac_srs->srs_lock);
4182 }
4183
4184 prev = tail;
4185 tail->b_prev = NULL;
4186 if (mac_srs->srs_bw->mac_bw_used <
4187 mac_srs->srs_bw->mac_bw_limit)
4188 continue;
4189
4190 now = ddi_get_lbolt();
4191 if (mac_srs->srs_bw->mac_bw_curr_time != now) {
4192 mac_srs->srs_bw->mac_bw_curr_time = now;
4193 mac_srs->srs_bw->mac_bw_used = 0;
4194 continue;
4195 }
4196 mac_srs->srs_bw->mac_bw_state |= SRS_BW_ENFORCED;
4197 break;
4198 }
4199 ASSERT((head == NULL && tail == NULL) ||
4200 (head != NULL && tail != NULL));
4201 if (tail != NULL) {
4202 tail->b_next = NULL;
4203 mutex_exit(&mac_srs->srs_lock);
4204 TX_SRS_TO_SOFT_RING(mac_srs, head, hint);
4205 mutex_enter(&mac_srs->srs_lock);
4206 }
4207 }
4208 /*
4209 * SRS_TX_FANOUT case not considered here because packets
4210 * won't be queued in the SRS for this case. Packets will
4211 * be sent directly to soft rings underneath and if there
4212 * is any queueing at all, it would be in Tx side soft
4213 * rings.
4214 */
4215
4216 /*
4217 * When srs_count becomes 0, reset SRS_TX_HIWAT and
4218 * SRS_TX_WAKEUP_CLIENT and wakeup registered clients.
4219 */
4220 if (mac_srs->srs_count == 0 && (mac_srs->srs_state &
4221 (SRS_TX_HIWAT | SRS_TX_WAKEUP_CLIENT | SRS_ENQUEUED))) {
4222 mac_client_impl_t *mcip = mac_srs->srs_mcip;
4223 boolean_t wakeup_required = B_FALSE;
4224
4225 if (mac_srs->srs_state &
4226 (SRS_TX_HIWAT|SRS_TX_WAKEUP_CLIENT)) {
4227 wakeup_required = B_TRUE;
4228 }
4229 mac_srs->srs_state &= ~(SRS_TX_HIWAT |
4230 SRS_TX_WAKEUP_CLIENT | SRS_ENQUEUED);
4231 mutex_exit(&mac_srs->srs_lock);
4232 if (wakeup_required) {
4233 mac_tx_invoke_callbacks(mcip, (mac_tx_cookie_t)mac_srs);
4234 /*
4235 * If the client is not the primary MAC client, then we
4236 * need to send the notification to the clients upper
4237 * MAC, i.e. mci_upper_mip.
4238 */
4239 mac_tx_notify(mcip->mci_upper_mip != NULL ?
4240 mcip->mci_upper_mip : mcip->mci_mip);
4241 }
4242 mutex_enter(&mac_srs->srs_lock);
4243 }
4244 mac_srs->srs_state &= ~SRS_PROC;
4245 }
4246
4247 /*
4248 * Given a packet, get the flow_entry that identifies the flow
4249 * to which that packet belongs. The flow_entry will contain
4250 * the transmit function to be used to send the packet. If the
4251 * function returns NULL, the packet should be sent using the
4252 * underlying NIC.
4253 */
4254 static flow_entry_t *
mac_tx_classify(mac_impl_t * mip,mblk_t * mp)4255 mac_tx_classify(mac_impl_t *mip, mblk_t *mp)
4256 {
4257 flow_entry_t *flent = NULL;
4258 mac_client_impl_t *mcip;
4259 int err;
4260
4261 /*
4262 * Do classification on the packet.
4263 */
4264 err = mac_flow_lookup(mip->mi_flow_tab, mp, FLOW_OUTBOUND, &flent);
4265 if (err != 0)
4266 return (NULL);
4267
4268 /*
4269 * This flent might just be an additional one on the MAC client,
4270 * i.e. for classification purposes (different fdesc), however
4271 * the resources, SRS et. al., are in the mci_flent, so if
4272 * this isn't the mci_flent, we need to get it.
4273 */
4274 if ((mcip = flent->fe_mcip) != NULL && mcip->mci_flent != flent) {
4275 FLOW_REFRELE(flent);
4276 flent = mcip->mci_flent;
4277 FLOW_TRY_REFHOLD(flent, err);
4278 if (err != 0)
4279 return (NULL);
4280 }
4281
4282 return (flent);
4283 }
4284
4285 /*
4286 * This macro is only meant to be used by mac_tx_send().
4287 */
4288 #define CHECK_VID_AND_ADD_TAG(mp) { \
4289 if (vid_check) { \
4290 int err = 0; \
4291 \
4292 MAC_VID_CHECK(src_mcip, (mp), err); \
4293 if (err != 0) { \
4294 freemsg((mp)); \
4295 (mp) = next; \
4296 oerrors++; \
4297 continue; \
4298 } \
4299 } \
4300 if (add_tag) { \
4301 (mp) = mac_add_vlan_tag((mp), 0, vid); \
4302 if ((mp) == NULL) { \
4303 (mp) = next; \
4304 oerrors++; \
4305 continue; \
4306 } \
4307 } \
4308 }
4309
4310 mblk_t *
mac_tx_send(mac_client_handle_t mch,mac_ring_handle_t ring,mblk_t * mp_chain,mac_tx_stats_t * stats)4311 mac_tx_send(mac_client_handle_t mch, mac_ring_handle_t ring, mblk_t *mp_chain,
4312 mac_tx_stats_t *stats)
4313 {
4314 mac_client_impl_t *src_mcip = (mac_client_impl_t *)mch;
4315 mac_impl_t *mip = src_mcip->mci_mip;
4316 uint_t obytes = 0, opackets = 0, oerrors = 0;
4317 mblk_t *mp = NULL, *next;
4318 boolean_t vid_check, add_tag;
4319 uint16_t vid = 0;
4320
4321 if (mip->mi_nclients > 1) {
4322 vid_check = MAC_VID_CHECK_NEEDED(src_mcip);
4323 add_tag = MAC_TAG_NEEDED(src_mcip);
4324 if (add_tag)
4325 vid = mac_client_vid(mch);
4326 } else {
4327 ASSERT(mip->mi_nclients == 1);
4328 vid_check = add_tag = B_FALSE;
4329 }
4330
4331 /*
4332 * Fastpath: if there's only one client, we simply send
4333 * the packet down to the underlying NIC.
4334 */
4335 if (mip->mi_nactiveclients == 1) {
4336 DTRACE_PROBE2(fastpath,
4337 mac_client_impl_t *, src_mcip, mblk_t *, mp_chain);
4338
4339 mp = mp_chain;
4340 while (mp != NULL) {
4341 next = mp->b_next;
4342 mp->b_next = NULL;
4343 opackets++;
4344 obytes += (mp->b_cont == NULL ? MBLKL(mp) :
4345 msgdsize(mp));
4346
4347 CHECK_VID_AND_ADD_TAG(mp);
4348 mp = mac_provider_tx(mip, ring, mp, src_mcip);
4349
4350 /*
4351 * If the driver is out of descriptors and does a
4352 * partial send it will return a chain of unsent
4353 * mblks. Adjust the accounting stats.
4354 */
4355 if (mp != NULL) {
4356 opackets--;
4357 obytes -= msgdsize(mp);
4358 mp->b_next = next;
4359 break;
4360 }
4361 mp = next;
4362 }
4363 goto done;
4364 }
4365
4366 /*
4367 * No fastpath, we either have more than one MAC client
4368 * defined on top of the same MAC, or one or more MAC
4369 * client promiscuous callbacks.
4370 */
4371 DTRACE_PROBE3(slowpath, mac_client_impl_t *,
4372 src_mcip, int, mip->mi_nclients, mblk_t *, mp_chain);
4373
4374 mp = mp_chain;
4375 while (mp != NULL) {
4376 flow_entry_t *dst_flow_ent;
4377 void *flow_cookie;
4378 size_t pkt_size;
4379
4380 next = mp->b_next;
4381 mp->b_next = NULL;
4382 opackets++;
4383 pkt_size = (mp->b_cont == NULL ? MBLKL(mp) : msgdsize(mp));
4384 obytes += pkt_size;
4385 CHECK_VID_AND_ADD_TAG(mp);
4386
4387 /*
4388 * Find the destination.
4389 */
4390 dst_flow_ent = mac_tx_classify(mip, mp);
4391
4392 if (dst_flow_ent != NULL) {
4393 /*
4394 * Got a matching flow. It's either another
4395 * MAC client, or a broadcast/multicast flow.
4396 */
4397 flow_cookie = mac_flow_get_client_cookie(dst_flow_ent);
4398
4399 if (flow_cookie != NULL) {
4400 /*
4401 * The vnic_bcast_send function expects
4402 * to receive the sender MAC client
4403 * as value for arg2.
4404 */
4405 mac_bcast_send(flow_cookie, src_mcip, mp,
4406 B_TRUE);
4407 } else {
4408 /*
4409 * loopback the packet to a local MAC
4410 * client. We force a context switch
4411 * if both source and destination MAC
4412 * clients are used by IP, i.e.
4413 * bypass is set.
4414 */
4415 boolean_t do_switch;
4416
4417 mac_client_impl_t *dst_mcip =
4418 dst_flow_ent->fe_mcip;
4419
4420 /*
4421 * Check if there are promiscuous mode
4422 * callbacks defined. This check is
4423 * done here in the 'else' case and
4424 * not in other cases because this
4425 * path is for local loopback
4426 * communication which does not go
4427 * through MAC_TX(). For paths that go
4428 * through MAC_TX(), the promisc_list
4429 * check is done inside the MAC_TX()
4430 * macro.
4431 */
4432 if (mip->mi_promisc_list != NULL) {
4433 mac_promisc_dispatch(mip, mp, src_mcip,
4434 B_TRUE);
4435 }
4436
4437 do_switch = ((src_mcip->mci_state_flags &
4438 dst_mcip->mci_state_flags &
4439 MCIS_CLIENT_POLL_CAPABLE) != 0);
4440
4441 mac_hw_emul(&mp, NULL, NULL, MAC_ALL_EMULS);
4442 if (mp != NULL) {
4443 (dst_flow_ent->fe_cb_fn)(
4444 dst_flow_ent->fe_cb_arg1,
4445 dst_flow_ent->fe_cb_arg2,
4446 mp, do_switch);
4447 }
4448
4449 }
4450 FLOW_REFRELE(dst_flow_ent);
4451 } else {
4452 /*
4453 * Unknown destination, send via the underlying
4454 * NIC.
4455 */
4456 mp = mac_provider_tx(mip, ring, mp, src_mcip);
4457 if (mp != NULL) {
4458 /*
4459 * Adjust for the last packet that
4460 * could not be transmitted
4461 */
4462 opackets--;
4463 obytes -= pkt_size;
4464 mp->b_next = next;
4465 break;
4466 }
4467 }
4468 mp = next;
4469 }
4470
4471 done:
4472 stats->mts_obytes = obytes;
4473 stats->mts_opackets = opackets;
4474 stats->mts_oerrors = oerrors;
4475 return (mp);
4476 }
4477
4478 /*
4479 * mac_tx_srs_ring_present
4480 *
4481 * Returns whether the specified ring is part of the specified SRS.
4482 */
4483 boolean_t
mac_tx_srs_ring_present(mac_soft_ring_set_t * srs,mac_ring_t * tx_ring)4484 mac_tx_srs_ring_present(mac_soft_ring_set_t *srs, mac_ring_t *tx_ring)
4485 {
4486 int i;
4487 mac_soft_ring_t *soft_ring;
4488
4489 if (srs->srs_tx.st_arg2 == tx_ring)
4490 return (B_TRUE);
4491
4492 for (i = 0; i < srs->srs_tx_ring_count; i++) {
4493 soft_ring = srs->srs_tx_soft_rings[i];
4494 if (soft_ring->s_ring_tx_arg2 == tx_ring)
4495 return (B_TRUE);
4496 }
4497
4498 return (B_FALSE);
4499 }
4500
4501 /*
4502 * mac_tx_srs_get_soft_ring
4503 *
4504 * Returns the TX soft ring associated with the given ring, if present.
4505 */
4506 mac_soft_ring_t *
mac_tx_srs_get_soft_ring(mac_soft_ring_set_t * srs,mac_ring_t * tx_ring)4507 mac_tx_srs_get_soft_ring(mac_soft_ring_set_t *srs, mac_ring_t *tx_ring)
4508 {
4509 int i;
4510 mac_soft_ring_t *soft_ring;
4511
4512 if (srs->srs_tx.st_arg2 == tx_ring)
4513 return (NULL);
4514
4515 for (i = 0; i < srs->srs_tx_ring_count; i++) {
4516 soft_ring = srs->srs_tx_soft_rings[i];
4517 if (soft_ring->s_ring_tx_arg2 == tx_ring)
4518 return (soft_ring);
4519 }
4520
4521 return (NULL);
4522 }
4523
4524 /*
4525 * mac_tx_srs_wakeup
4526 *
4527 * Called when Tx desc become available. Wakeup the appropriate worker
4528 * thread after resetting the SRS_TX_BLOCKED/S_RING_BLOCK bit in the
4529 * state field.
4530 */
4531 void
mac_tx_srs_wakeup(mac_soft_ring_set_t * mac_srs,mac_ring_handle_t ring)4532 mac_tx_srs_wakeup(mac_soft_ring_set_t *mac_srs, mac_ring_handle_t ring)
4533 {
4534 int i;
4535 mac_soft_ring_t *sringp;
4536 mac_srs_tx_t *srs_tx = &mac_srs->srs_tx;
4537
4538 mutex_enter(&mac_srs->srs_lock);
4539 /*
4540 * srs_tx_ring_count == 0 is the single ring mode case. In
4541 * this mode, there will not be Tx soft rings associated
4542 * with the SRS.
4543 */
4544 if (!MAC_TX_SOFT_RINGS(mac_srs)) {
4545 if (srs_tx->st_arg2 == ring &&
4546 mac_srs->srs_state & SRS_TX_BLOCKED) {
4547 mac_srs->srs_state &= ~SRS_TX_BLOCKED;
4548 srs_tx->st_stat.mts_unblockcnt++;
4549 cv_signal(&mac_srs->srs_async);
4550 }
4551 /*
4552 * A wakeup can come before tx_srs_drain() could
4553 * grab srs lock and set SRS_TX_BLOCKED. So
4554 * always set woken_up flag when we come here.
4555 */
4556 srs_tx->st_woken_up = B_TRUE;
4557 mutex_exit(&mac_srs->srs_lock);
4558 return;
4559 }
4560
4561 /*
4562 * If you are here, it is for FANOUT, BW_FANOUT,
4563 * AGGR_MODE or AGGR_BW_MODE case
4564 */
4565 for (i = 0; i < mac_srs->srs_tx_ring_count; i++) {
4566 sringp = mac_srs->srs_tx_soft_rings[i];
4567 mutex_enter(&sringp->s_ring_lock);
4568 if (sringp->s_ring_tx_arg2 == ring) {
4569 if (sringp->s_ring_state & S_RING_BLOCK) {
4570 sringp->s_ring_state &= ~S_RING_BLOCK;
4571 sringp->s_st_stat.mts_unblockcnt++;
4572 cv_signal(&sringp->s_ring_async);
4573 }
4574 sringp->s_ring_tx_woken_up = B_TRUE;
4575 }
4576 mutex_exit(&sringp->s_ring_lock);
4577 }
4578 mutex_exit(&mac_srs->srs_lock);
4579 }
4580
4581 /*
4582 * Once the driver is done draining, send a MAC_NOTE_TX notification to unleash
4583 * the blocked clients again.
4584 */
4585 void
mac_tx_notify(mac_impl_t * mip)4586 mac_tx_notify(mac_impl_t *mip)
4587 {
4588 i_mac_notify(mip, MAC_NOTE_TX);
4589 }
4590
4591 /*
4592 * RX SOFTRING RELATED FUNCTIONS
4593 *
4594 * These functions really belong in mac_soft_ring.c and here for
4595 * a short period.
4596 */
4597
4598 #define SOFT_RING_ENQUEUE_CHAIN(ringp, mp, tail, cnt, sz) { \
4599 /* \
4600 * Enqueue our mblk chain. \
4601 */ \
4602 ASSERT(MUTEX_HELD(&(ringp)->s_ring_lock)); \
4603 \
4604 if ((ringp)->s_ring_last != NULL) \
4605 (ringp)->s_ring_last->b_next = (mp); \
4606 else \
4607 (ringp)->s_ring_first = (mp); \
4608 (ringp)->s_ring_last = (tail); \
4609 (ringp)->s_ring_count += (cnt); \
4610 ASSERT((ringp)->s_ring_count > 0); \
4611 if ((ringp)->s_ring_type & ST_RING_BW_CTL) { \
4612 (ringp)->s_ring_size += sz; \
4613 } \
4614 }
4615
4616 /*
4617 * Default entry point to deliver a packet chain to a MAC client.
4618 * If the MAC client has flows, do the classification with these
4619 * flows as well.
4620 */
4621 /* ARGSUSED */
4622 void
mac_rx_deliver(void * arg1,mac_resource_handle_t mrh,mblk_t * mp_chain,mac_header_info_t * arg3)4623 mac_rx_deliver(void *arg1, mac_resource_handle_t mrh, mblk_t *mp_chain,
4624 mac_header_info_t *arg3)
4625 {
4626 mac_client_impl_t *mcip = arg1;
4627
4628 if (mcip->mci_nvids == 1 &&
4629 !(mcip->mci_state_flags & MCIS_STRIP_DISABLE)) {
4630 /*
4631 * If the client has exactly one VID associated with it
4632 * and striping of VLAN header is not disabled,
4633 * remove the VLAN tag from the packet before
4634 * passing it on to the client's receive callback.
4635 * Note that this needs to be done after we dispatch
4636 * the packet to the promiscuous listeners of the
4637 * client, since they expect to see the whole
4638 * frame including the VLAN headers.
4639 *
4640 * The MCIS_STRIP_DISABLE is only issued when sun4v
4641 * vsw is in play.
4642 */
4643 mp_chain = mac_strip_vlan_tag_chain(mp_chain);
4644 }
4645
4646 mcip->mci_rx_fn(mcip->mci_rx_arg, mrh, mp_chain, B_FALSE);
4647 }
4648
4649 /*
4650 * Process a chain for a given soft ring. If the number of packets
4651 * queued in the SRS and its associated soft rings (including this
4652 * one) is very small (tracked by srs_poll_pkt_cnt) then allow the
4653 * entering thread (interrupt or poll thread) to process the chain
4654 * inline. This is meant to reduce latency under low load.
4655 *
4656 * The proc and arg for each mblk is already stored in the mblk in
4657 * appropriate places.
4658 */
4659 /* ARGSUSED */
4660 void
mac_rx_soft_ring_process(mac_client_impl_t * mcip,mac_soft_ring_t * ringp,mblk_t * mp_chain,mblk_t * tail,int cnt,size_t sz)4661 mac_rx_soft_ring_process(mac_client_impl_t *mcip, mac_soft_ring_t *ringp,
4662 mblk_t *mp_chain, mblk_t *tail, int cnt, size_t sz)
4663 {
4664 mac_direct_rx_t proc;
4665 void *arg1;
4666 mac_resource_handle_t arg2;
4667 mac_soft_ring_set_t *mac_srs = ringp->s_ring_set;
4668
4669 ASSERT(ringp != NULL);
4670 ASSERT(mp_chain != NULL);
4671 ASSERT(tail != NULL);
4672 ASSERT(MUTEX_NOT_HELD(&ringp->s_ring_lock));
4673
4674 mutex_enter(&ringp->s_ring_lock);
4675 ringp->s_ring_total_inpkt += cnt;
4676 ringp->s_ring_total_rbytes += sz;
4677 if ((mac_srs->srs_rx.sr_poll_pkt_cnt <= 1) &&
4678 !(ringp->s_ring_type & ST_RING_WORKER_ONLY)) {
4679 /* If on processor or blanking on, then enqueue and return */
4680 if (ringp->s_ring_state & S_RING_BLANK ||
4681 ringp->s_ring_state & S_RING_PROC) {
4682 SOFT_RING_ENQUEUE_CHAIN(ringp, mp_chain, tail, cnt, sz);
4683 mutex_exit(&ringp->s_ring_lock);
4684 return;
4685 }
4686 proc = ringp->s_ring_rx_func;
4687 arg1 = ringp->s_ring_rx_arg1;
4688 arg2 = ringp->s_ring_rx_arg2;
4689 /*
4690 * See if anything is already queued. If we are the
4691 * first packet, do inline processing else queue the
4692 * packet and do the drain.
4693 */
4694 if (ringp->s_ring_first == NULL) {
4695 /*
4696 * Fast-path, ok to process and nothing queued.
4697 */
4698 ringp->s_ring_run = curthread;
4699 ringp->s_ring_state |= (S_RING_PROC);
4700
4701 mutex_exit(&ringp->s_ring_lock);
4702
4703 /*
4704 * We are the chain of 1 packet so
4705 * go through this fast path.
4706 */
4707 ASSERT(mp_chain->b_next == NULL);
4708
4709 (*proc)(arg1, arg2, mp_chain, NULL);
4710
4711 ASSERT(MUTEX_NOT_HELD(&ringp->s_ring_lock));
4712 /*
4713 * If we have an SRS performing bandwidth
4714 * control then we need to decrement the size
4715 * and count so the SRS has an accurate count
4716 * of the data queued between the SRS and its
4717 * soft rings. We decrement the counters only
4718 * when the packet is processed by both the
4719 * SRS and the soft ring.
4720 */
4721 mutex_enter(&mac_srs->srs_lock);
4722 MAC_UPDATE_SRS_COUNT_LOCKED(mac_srs, cnt);
4723 MAC_UPDATE_SRS_SIZE_LOCKED(mac_srs, sz);
4724 mutex_exit(&mac_srs->srs_lock);
4725
4726 mutex_enter(&ringp->s_ring_lock);
4727 ringp->s_ring_run = NULL;
4728 ringp->s_ring_state &= ~S_RING_PROC;
4729 if (ringp->s_ring_state & S_RING_CLIENT_WAIT)
4730 cv_signal(&ringp->s_ring_client_cv);
4731
4732 if ((ringp->s_ring_first == NULL) ||
4733 (ringp->s_ring_state & S_RING_BLANK)) {
4734 /*
4735 * We processed a single packet inline
4736 * and nothing new has arrived or our
4737 * receiver doesn't want to receive
4738 * any packets. We are done.
4739 */
4740 mutex_exit(&ringp->s_ring_lock);
4741 return;
4742 }
4743 } else {
4744 SOFT_RING_ENQUEUE_CHAIN(ringp,
4745 mp_chain, tail, cnt, sz);
4746 }
4747
4748 /*
4749 * We are here because either we couldn't do inline
4750 * processing (because something was already
4751 * queued), or we had a chain of more than one
4752 * packet, or something else arrived after we were
4753 * done with inline processing.
4754 */
4755 ASSERT(MUTEX_HELD(&ringp->s_ring_lock));
4756 ASSERT(ringp->s_ring_first != NULL);
4757
4758 ringp->s_ring_drain_func(ringp);
4759 mutex_exit(&ringp->s_ring_lock);
4760 return;
4761 } else {
4762 /* ST_RING_WORKER_ONLY case */
4763 SOFT_RING_ENQUEUE_CHAIN(ringp, mp_chain, tail, cnt, sz);
4764 mac_soft_ring_worker_wakeup(ringp);
4765 mutex_exit(&ringp->s_ring_lock);
4766 }
4767 }
4768
4769 /*
4770 * TX SOFTRING RELATED FUNCTIONS
4771 *
4772 * These functions really belong in mac_soft_ring.c and here for
4773 * a short period.
4774 */
4775
4776 #define TX_SOFT_RING_ENQUEUE_CHAIN(ringp, mp, tail, cnt, sz) { \
4777 ASSERT(MUTEX_HELD(&ringp->s_ring_lock)); \
4778 ringp->s_ring_state |= S_RING_ENQUEUED; \
4779 SOFT_RING_ENQUEUE_CHAIN(ringp, mp_chain, tail, cnt, sz); \
4780 }
4781
4782 /*
4783 * mac_tx_sring_queued
4784 *
4785 * When we are out of transmit descriptors and we already have a
4786 * queue that exceeds hiwat (or the client called us with
4787 * MAC_TX_NO_ENQUEUE or MAC_DROP_ON_NO_DESC flag), return the
4788 * soft ring pointer as the opaque cookie for the client enable
4789 * flow control.
4790 */
4791 static mac_tx_cookie_t
mac_tx_sring_enqueue(mac_soft_ring_t * ringp,mblk_t * mp_chain,uint16_t flag,mblk_t ** ret_mp)4792 mac_tx_sring_enqueue(mac_soft_ring_t *ringp, mblk_t *mp_chain, uint16_t flag,
4793 mblk_t **ret_mp)
4794 {
4795 int cnt;
4796 size_t sz;
4797 mblk_t *tail;
4798 mac_soft_ring_set_t *mac_srs = ringp->s_ring_set;
4799 mac_tx_cookie_t cookie = 0;
4800 boolean_t wakeup_worker = B_TRUE;
4801
4802 ASSERT(MUTEX_HELD(&ringp->s_ring_lock));
4803 MAC_COUNT_CHAIN(mac_srs, mp_chain, tail, cnt, sz);
4804 if (flag & MAC_DROP_ON_NO_DESC) {
4805 mac_drop_chain(mp_chain, "Tx softring no desc");
4806 /* increment freed stats */
4807 ringp->s_ring_drops += cnt;
4808 cookie = (mac_tx_cookie_t)ringp;
4809 } else {
4810 if (ringp->s_ring_first != NULL)
4811 wakeup_worker = B_FALSE;
4812
4813 if (flag & MAC_TX_NO_ENQUEUE) {
4814 /*
4815 * If QUEUED is not set, queue the packet
4816 * and let mac_tx_soft_ring_drain() set
4817 * the TX_BLOCKED bit for the reasons
4818 * explained above. Otherwise, return the
4819 * mblks.
4820 */
4821 if (wakeup_worker) {
4822 TX_SOFT_RING_ENQUEUE_CHAIN(ringp,
4823 mp_chain, tail, cnt, sz);
4824 } else {
4825 ringp->s_ring_state |= S_RING_WAKEUP_CLIENT;
4826 cookie = (mac_tx_cookie_t)ringp;
4827 *ret_mp = mp_chain;
4828 }
4829 } else {
4830 boolean_t enqueue = B_TRUE;
4831
4832 if (ringp->s_ring_count > ringp->s_ring_tx_hiwat) {
4833 /*
4834 * flow-controlled. Store ringp in cookie
4835 * so that it can be returned as
4836 * mac_tx_cookie_t to client
4837 */
4838 ringp->s_ring_state |= S_RING_TX_HIWAT;
4839 cookie = (mac_tx_cookie_t)ringp;
4840 ringp->s_ring_hiwat_cnt++;
4841 if (ringp->s_ring_count >
4842 ringp->s_ring_tx_max_q_cnt) {
4843 /* increment freed stats */
4844 ringp->s_ring_drops += cnt;
4845 /*
4846 * b_prev may be set to the fanout hint
4847 * hence can't use freemsg directly
4848 */
4849 mac_drop_chain(mp_chain,
4850 "Tx softring max queue");
4851 DTRACE_PROBE1(tx_queued_hiwat,
4852 mac_soft_ring_t *, ringp);
4853 enqueue = B_FALSE;
4854 }
4855 }
4856 if (enqueue) {
4857 TX_SOFT_RING_ENQUEUE_CHAIN(ringp, mp_chain,
4858 tail, cnt, sz);
4859 }
4860 }
4861 if (wakeup_worker)
4862 cv_signal(&ringp->s_ring_async);
4863 }
4864 return (cookie);
4865 }
4866
4867
4868 /*
4869 * mac_tx_soft_ring_process
4870 *
4871 * This routine is called when fanning out outgoing traffic among
4872 * multipe Tx rings.
4873 * Note that a soft ring is associated with a h/w Tx ring.
4874 */
4875 mac_tx_cookie_t
mac_tx_soft_ring_process(mac_soft_ring_t * ringp,mblk_t * mp_chain,uint16_t flag,mblk_t ** ret_mp)4876 mac_tx_soft_ring_process(mac_soft_ring_t *ringp, mblk_t *mp_chain,
4877 uint16_t flag, mblk_t **ret_mp)
4878 {
4879 mac_soft_ring_set_t *mac_srs = ringp->s_ring_set;
4880 int cnt;
4881 size_t sz;
4882 mblk_t *tail;
4883 mac_tx_cookie_t cookie = 0;
4884
4885 ASSERT(ringp != NULL);
4886 ASSERT(mp_chain != NULL);
4887 ASSERT(MUTEX_NOT_HELD(&ringp->s_ring_lock));
4888 /*
4889 * The following modes can come here: SRS_TX_BW_FANOUT,
4890 * SRS_TX_FANOUT, SRS_TX_AGGR, SRS_TX_BW_AGGR.
4891 */
4892 ASSERT(MAC_TX_SOFT_RINGS(mac_srs));
4893 ASSERT(mac_srs->srs_tx.st_mode == SRS_TX_FANOUT ||
4894 mac_srs->srs_tx.st_mode == SRS_TX_BW_FANOUT ||
4895 mac_srs->srs_tx.st_mode == SRS_TX_AGGR ||
4896 mac_srs->srs_tx.st_mode == SRS_TX_BW_AGGR);
4897
4898 if (ringp->s_ring_type & ST_RING_WORKER_ONLY) {
4899 /* Serialization mode */
4900
4901 mutex_enter(&ringp->s_ring_lock);
4902 if (ringp->s_ring_count > ringp->s_ring_tx_hiwat) {
4903 cookie = mac_tx_sring_enqueue(ringp, mp_chain,
4904 flag, ret_mp);
4905 mutex_exit(&ringp->s_ring_lock);
4906 return (cookie);
4907 }
4908 MAC_COUNT_CHAIN(mac_srs, mp_chain, tail, cnt, sz);
4909 TX_SOFT_RING_ENQUEUE_CHAIN(ringp, mp_chain, tail, cnt, sz);
4910 if (ringp->s_ring_state & (S_RING_BLOCK | S_RING_PROC)) {
4911 /*
4912 * If ring is blocked due to lack of Tx
4913 * descs, just return. Worker thread
4914 * will get scheduled when Tx desc's
4915 * become available.
4916 */
4917 mutex_exit(&ringp->s_ring_lock);
4918 return (cookie);
4919 }
4920 mac_soft_ring_worker_wakeup(ringp);
4921 mutex_exit(&ringp->s_ring_lock);
4922 return (cookie);
4923 } else {
4924 /* Default fanout mode */
4925 /*
4926 * S_RING_BLOCKED is set when underlying NIC runs
4927 * out of Tx descs and messages start getting
4928 * queued. It won't get reset until
4929 * tx_srs_drain() completely drains out the
4930 * messages.
4931 */
4932 mac_tx_stats_t stats;
4933
4934 if (ringp->s_ring_state & S_RING_ENQUEUED) {
4935 /* Tx descs/resources not available */
4936 mutex_enter(&ringp->s_ring_lock);
4937 if (ringp->s_ring_state & S_RING_ENQUEUED) {
4938 cookie = mac_tx_sring_enqueue(ringp, mp_chain,
4939 flag, ret_mp);
4940 mutex_exit(&ringp->s_ring_lock);
4941 return (cookie);
4942 }
4943 /*
4944 * While we were computing mblk count, the
4945 * flow control condition got relieved.
4946 * Continue with the transmission.
4947 */
4948 mutex_exit(&ringp->s_ring_lock);
4949 }
4950
4951 mp_chain = mac_tx_send(ringp->s_ring_tx_arg1,
4952 ringp->s_ring_tx_arg2, mp_chain, &stats);
4953
4954 /*
4955 * Multiple threads could be here sending packets.
4956 * Under such conditions, it is not possible to
4957 * automically set S_RING_BLOCKED bit to indicate
4958 * out of tx desc condition. To atomically set
4959 * this, we queue the returned packet and do
4960 * the setting of S_RING_BLOCKED in
4961 * mac_tx_soft_ring_drain().
4962 */
4963 if (mp_chain != NULL) {
4964 mutex_enter(&ringp->s_ring_lock);
4965 cookie =
4966 mac_tx_sring_enqueue(ringp, mp_chain, flag, ret_mp);
4967 mutex_exit(&ringp->s_ring_lock);
4968 return (cookie);
4969 }
4970 SRS_TX_STATS_UPDATE(mac_srs, &stats);
4971 SOFTRING_TX_STATS_UPDATE(ringp, &stats);
4972
4973 return (0);
4974 }
4975 }
4976