Commit | Line | Data |
---|---|---|
77241056 MM |
1 | /* |
2 | * | |
3 | * This file is provided under a dual BSD/GPLv2 license. When using or | |
4 | * redistributing this file, you may do so under either license. | |
5 | * | |
6 | * GPL LICENSE SUMMARY | |
7 | * | |
8 | * Copyright(c) 2015 Intel Corporation. | |
9 | * | |
10 | * This program is free software; you can redistribute it and/or modify | |
11 | * it under the terms of version 2 of the GNU General Public License as | |
12 | * published by the Free Software Foundation. | |
13 | * | |
14 | * This program is distributed in the hope that it will be useful, but | |
15 | * WITHOUT ANY WARRANTY; without even the implied warranty of | |
16 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
17 | * General Public License for more details. | |
18 | * | |
19 | * BSD LICENSE | |
20 | * | |
21 | * Copyright(c) 2015 Intel Corporation. | |
22 | * | |
23 | * Redistribution and use in source and binary forms, with or without | |
24 | * modification, are permitted provided that the following conditions | |
25 | * are met: | |
26 | * | |
27 | * - Redistributions of source code must retain the above copyright | |
28 | * notice, this list of conditions and the following disclaimer. | |
29 | * - Redistributions in binary form must reproduce the above copyright | |
30 | * notice, this list of conditions and the following disclaimer in | |
31 | * the documentation and/or other materials provided with the | |
32 | * distribution. | |
33 | * - Neither the name of Intel Corporation nor the names of its | |
34 | * contributors may be used to endorse or promote products derived | |
35 | * from this software without specific prior written permission. | |
36 | * | |
37 | * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS | |
38 | * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT | |
39 | * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR | |
40 | * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT | |
41 | * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, | |
42 | * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT | |
43 | * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, | |
44 | * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY | |
45 | * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT | |
46 | * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE | |
47 | * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. | |
48 | * | |
49 | */ | |
50 | ||
51 | #include <linux/spinlock.h> | |
52 | #include <linux/pci.h> | |
53 | #include <linux/io.h> | |
54 | #include <linux/delay.h> | |
55 | #include <linux/netdevice.h> | |
56 | #include <linux/vmalloc.h> | |
57 | #include <linux/module.h> | |
58 | #include <linux/prefetch.h> | |
8859b4a6 | 59 | #include <rdma/ib_verbs.h> |
77241056 MM |
60 | |
61 | #include "hfi.h" | |
62 | #include "trace.h" | |
63 | #include "qp.h" | |
64 | #include "sdma.h" | |
65 | ||
66 | #undef pr_fmt | |
67 | #define pr_fmt(fmt) DRIVER_NAME ": " fmt | |
68 | ||
69 | /* | |
70 | * The size has to be longer than this string, so we can append | |
71 | * board/chip information to it in the initialization code. | |
72 | */ | |
73 | const char ib_hfi1_version[] = HFI1_DRIVER_VERSION "\n"; | |
74 | ||
75 | DEFINE_SPINLOCK(hfi1_devs_lock); | |
76 | LIST_HEAD(hfi1_dev_list); | |
77 | DEFINE_MUTEX(hfi1_mutex); /* general driver use */ | |
78 | ||
79 | unsigned int hfi1_max_mtu = HFI1_DEFAULT_MAX_MTU; | |
80 | module_param_named(max_mtu, hfi1_max_mtu, uint, S_IRUGO); | |
81 | MODULE_PARM_DESC(max_mtu, "Set max MTU bytes, default is 8192"); | |
82 | ||
83 | unsigned int hfi1_cu = 1; | |
84 | module_param_named(cu, hfi1_cu, uint, S_IRUGO); | |
85 | MODULE_PARM_DESC(cu, "Credit return units"); | |
86 | ||
87 | unsigned long hfi1_cap_mask = HFI1_CAP_MASK_DEFAULT; | |
88 | static int hfi1_caps_set(const char *, const struct kernel_param *); | |
89 | static int hfi1_caps_get(char *, const struct kernel_param *); | |
90 | static const struct kernel_param_ops cap_ops = { | |
91 | .set = hfi1_caps_set, | |
92 | .get = hfi1_caps_get | |
93 | }; | |
94 | module_param_cb(cap_mask, &cap_ops, &hfi1_cap_mask, S_IWUSR | S_IRUGO); | |
95 | MODULE_PARM_DESC(cap_mask, "Bit mask of enabled/disabled HW features"); | |
96 | ||
97 | MODULE_LICENSE("Dual BSD/GPL"); | |
98 | MODULE_DESCRIPTION("Intel Omni-Path Architecture driver"); | |
99 | MODULE_VERSION(HFI1_DRIVER_VERSION); | |
100 | ||
101 | /* | |
102 | * MAX_PKT_RCV is the max # if packets processed per receive interrupt. | |
103 | */ | |
104 | #define MAX_PKT_RECV 64 | |
105 | #define EGR_HEAD_UPDATE_THRESHOLD 16 | |
106 | ||
107 | struct hfi1_ib_stats hfi1_stats; | |
108 | ||
109 | static int hfi1_caps_set(const char *val, const struct kernel_param *kp) | |
110 | { | |
111 | int ret = 0; | |
112 | unsigned long *cap_mask_ptr = (unsigned long *)kp->arg, | |
113 | cap_mask = *cap_mask_ptr, value, diff, | |
114 | write_mask = ((HFI1_CAP_WRITABLE_MASK << HFI1_CAP_USER_SHIFT) | | |
115 | HFI1_CAP_WRITABLE_MASK); | |
116 | ||
117 | ret = kstrtoul(val, 0, &value); | |
118 | if (ret) { | |
119 | pr_warn("Invalid module parameter value for 'cap_mask'\n"); | |
120 | goto done; | |
121 | } | |
122 | /* Get the changed bits (except the locked bit) */ | |
123 | diff = value ^ (cap_mask & ~HFI1_CAP_LOCKED_SMASK); | |
124 | ||
125 | /* Remove any bits that are not allowed to change after driver load */ | |
126 | if (HFI1_CAP_LOCKED() && (diff & ~write_mask)) { | |
127 | pr_warn("Ignoring non-writable capability bits %#lx\n", | |
128 | diff & ~write_mask); | |
129 | diff &= write_mask; | |
130 | } | |
131 | ||
132 | /* Mask off any reserved bits */ | |
133 | diff &= ~HFI1_CAP_RESERVED_MASK; | |
134 | /* Clear any previously set and changing bits */ | |
135 | cap_mask &= ~diff; | |
136 | /* Update the bits with the new capability */ | |
137 | cap_mask |= (value & diff); | |
138 | /* Check for any kernel/user restrictions */ | |
139 | diff = (cap_mask & (HFI1_CAP_MUST_HAVE_KERN << HFI1_CAP_USER_SHIFT)) ^ | |
140 | ((cap_mask & HFI1_CAP_MUST_HAVE_KERN) << HFI1_CAP_USER_SHIFT); | |
141 | cap_mask &= ~diff; | |
142 | /* Set the bitmask to the final set */ | |
143 | *cap_mask_ptr = cap_mask; | |
144 | done: | |
145 | return ret; | |
146 | } | |
147 | ||
148 | static int hfi1_caps_get(char *buffer, const struct kernel_param *kp) | |
149 | { | |
150 | unsigned long cap_mask = *(unsigned long *)kp->arg; | |
151 | ||
152 | cap_mask &= ~HFI1_CAP_LOCKED_SMASK; | |
153 | cap_mask |= ((cap_mask & HFI1_CAP_K2U) << HFI1_CAP_USER_SHIFT); | |
154 | ||
155 | return scnprintf(buffer, PAGE_SIZE, "0x%lx", cap_mask); | |
156 | } | |
157 | ||
158 | const char *get_unit_name(int unit) | |
159 | { | |
160 | static char iname[16]; | |
161 | ||
9805071e | 162 | snprintf(iname, sizeof(iname), DRIVER_NAME "_%u", unit); |
77241056 MM |
163 | return iname; |
164 | } | |
165 | ||
49dbb6cf DD |
166 | const char *get_card_name(struct rvt_dev_info *rdi) |
167 | { | |
168 | struct hfi1_ibdev *ibdev = container_of(rdi, struct hfi1_ibdev, rdi); | |
169 | struct hfi1_devdata *dd = container_of(ibdev, | |
170 | struct hfi1_devdata, verbs_dev); | |
171 | return get_unit_name(dd->unit); | |
172 | } | |
173 | ||
174 | struct pci_dev *get_pci_dev(struct rvt_dev_info *rdi) | |
175 | { | |
176 | struct hfi1_ibdev *ibdev = container_of(rdi, struct hfi1_ibdev, rdi); | |
177 | struct hfi1_devdata *dd = container_of(ibdev, | |
178 | struct hfi1_devdata, verbs_dev); | |
179 | return dd->pcidev; | |
180 | } | |
181 | ||
77241056 MM |
182 | /* |
183 | * Return count of units with at least one port ACTIVE. | |
184 | */ | |
185 | int hfi1_count_active_units(void) | |
186 | { | |
187 | struct hfi1_devdata *dd; | |
188 | struct hfi1_pportdata *ppd; | |
189 | unsigned long flags; | |
190 | int pidx, nunits_active = 0; | |
191 | ||
192 | spin_lock_irqsave(&hfi1_devs_lock, flags); | |
193 | list_for_each_entry(dd, &hfi1_dev_list, list) { | |
194 | if (!(dd->flags & HFI1_PRESENT) || !dd->kregbase) | |
195 | continue; | |
196 | for (pidx = 0; pidx < dd->num_pports; ++pidx) { | |
197 | ppd = dd->pport + pidx; | |
198 | if (ppd->lid && ppd->linkup) { | |
199 | nunits_active++; | |
200 | break; | |
201 | } | |
202 | } | |
203 | } | |
204 | spin_unlock_irqrestore(&hfi1_devs_lock, flags); | |
205 | return nunits_active; | |
206 | } | |
207 | ||
208 | /* | |
209 | * Return count of all units, optionally return in arguments | |
210 | * the number of usable (present) units, and the number of | |
211 | * ports that are up. | |
212 | */ | |
213 | int hfi1_count_units(int *npresentp, int *nupp) | |
214 | { | |
215 | int nunits = 0, npresent = 0, nup = 0; | |
216 | struct hfi1_devdata *dd; | |
217 | unsigned long flags; | |
218 | int pidx; | |
219 | struct hfi1_pportdata *ppd; | |
220 | ||
221 | spin_lock_irqsave(&hfi1_devs_lock, flags); | |
222 | ||
223 | list_for_each_entry(dd, &hfi1_dev_list, list) { | |
224 | nunits++; | |
225 | if ((dd->flags & HFI1_PRESENT) && dd->kregbase) | |
226 | npresent++; | |
227 | for (pidx = 0; pidx < dd->num_pports; ++pidx) { | |
228 | ppd = dd->pport + pidx; | |
229 | if (ppd->lid && ppd->linkup) | |
230 | nup++; | |
231 | } | |
232 | } | |
233 | ||
234 | spin_unlock_irqrestore(&hfi1_devs_lock, flags); | |
235 | ||
236 | if (npresentp) | |
237 | *npresentp = npresent; | |
238 | if (nupp) | |
239 | *nupp = nup; | |
240 | ||
241 | return nunits; | |
242 | } | |
243 | ||
244 | /* | |
245 | * Get address of eager buffer from it's index (allocated in chunks, not | |
246 | * contiguous). | |
247 | */ | |
248 | static inline void *get_egrbuf(const struct hfi1_ctxtdata *rcd, u64 rhf, | |
249 | u8 *update) | |
250 | { | |
251 | u32 idx = rhf_egr_index(rhf), offset = rhf_egr_buf_offset(rhf); | |
252 | ||
253 | *update |= !(idx & (rcd->egrbufs.threshold - 1)) && !offset; | |
254 | return (void *)(((u64)(rcd->egrbufs.rcvtids[idx].addr)) + | |
255 | (offset * RCV_BUF_BLOCK_SIZE)); | |
256 | } | |
257 | ||
258 | /* | |
259 | * Validate and encode the a given RcvArray Buffer size. | |
260 | * The function will check whether the given size falls within | |
261 | * allowed size ranges for the respective type and, optionally, | |
262 | * return the proper encoding. | |
263 | */ | |
264 | inline int hfi1_rcvbuf_validate(u32 size, u8 type, u16 *encoded) | |
265 | { | |
266 | if (unlikely(!IS_ALIGNED(size, PAGE_SIZE))) | |
267 | return 0; | |
268 | if (unlikely(size < MIN_EAGER_BUFFER)) | |
269 | return 0; | |
270 | if (size > | |
271 | (type == PT_EAGER ? MAX_EAGER_BUFFER : MAX_EXPECTED_BUFFER)) | |
272 | return 0; | |
273 | if (encoded) | |
274 | *encoded = ilog2(size / PAGE_SIZE) + 1; | |
275 | return 1; | |
276 | } | |
277 | ||
278 | static void rcv_hdrerr(struct hfi1_ctxtdata *rcd, struct hfi1_pportdata *ppd, | |
279 | struct hfi1_packet *packet) | |
280 | { | |
281 | struct hfi1_message_header *rhdr = packet->hdr; | |
282 | u32 rte = rhf_rcv_type_err(packet->rhf); | |
283 | int lnh = be16_to_cpu(rhdr->lrh[0]) & 3; | |
284 | struct hfi1_ibport *ibp = &ppd->ibport_data; | |
ec4274f1 DD |
285 | struct hfi1_devdata *dd = ppd->dd; |
286 | struct rvt_dev_info *rdi = &dd->verbs_dev.rdi; | |
77241056 MM |
287 | |
288 | if (packet->rhf & (RHF_VCRC_ERR | RHF_ICRC_ERR)) | |
289 | return; | |
290 | ||
291 | if (packet->rhf & RHF_TID_ERR) { | |
292 | /* For TIDERR and RC QPs preemptively schedule a NAK */ | |
293 | struct hfi1_ib_header *hdr = (struct hfi1_ib_header *)rhdr; | |
294 | struct hfi1_other_headers *ohdr = NULL; | |
295 | u32 tlen = rhf_pkt_len(packet->rhf); /* in bytes */ | |
296 | u16 lid = be16_to_cpu(hdr->lrh[1]); | |
297 | u32 qp_num; | |
298 | u32 rcv_flags = 0; | |
299 | ||
300 | /* Sanity check packet */ | |
301 | if (tlen < 24) | |
302 | goto drop; | |
303 | ||
304 | /* Check for GRH */ | |
305 | if (lnh == HFI1_LRH_BTH) | |
306 | ohdr = &hdr->u.oth; | |
307 | else if (lnh == HFI1_LRH_GRH) { | |
308 | u32 vtf; | |
309 | ||
310 | ohdr = &hdr->u.l.oth; | |
311 | if (hdr->u.l.grh.next_hdr != IB_GRH_NEXT_HDR) | |
312 | goto drop; | |
313 | vtf = be32_to_cpu(hdr->u.l.grh.version_tclass_flow); | |
314 | if ((vtf >> IB_GRH_VERSION_SHIFT) != IB_GRH_VERSION) | |
315 | goto drop; | |
316 | rcv_flags |= HFI1_HAS_GRH; | |
317 | } else | |
318 | goto drop; | |
319 | ||
320 | /* Get the destination QP number. */ | |
ec4274f1 | 321 | qp_num = be32_to_cpu(ohdr->bth[1]) & RVT_QPN_MASK; |
8859b4a6 | 322 | if (lid < be16_to_cpu(IB_MULTICAST_LID_BASE)) { |
895420dd | 323 | struct rvt_qp *qp; |
b77d713a | 324 | unsigned long flags; |
77241056 MM |
325 | |
326 | rcu_read_lock(); | |
ec4274f1 | 327 | qp = rvt_lookup_qpn(rdi, &ibp->rvp, qp_num); |
77241056 MM |
328 | if (!qp) { |
329 | rcu_read_unlock(); | |
330 | goto drop; | |
331 | } | |
332 | ||
333 | /* | |
334 | * Handle only RC QPs - for other QP types drop error | |
335 | * packet. | |
336 | */ | |
b77d713a | 337 | spin_lock_irqsave(&qp->r_lock, flags); |
77241056 MM |
338 | |
339 | /* Check for valid receive state. */ | |
83693bd1 DD |
340 | if (!(ib_rvt_state_ops[qp->state] & |
341 | RVT_PROCESS_RECV_OK)) { | |
4eb06882 | 342 | ibp->rvp.n_pkt_drops++; |
77241056 MM |
343 | } |
344 | ||
345 | switch (qp->ibqp.qp_type) { | |
346 | case IB_QPT_RC: | |
347 | hfi1_rc_hdrerr( | |
348 | rcd, | |
349 | hdr, | |
350 | rcv_flags, | |
351 | qp); | |
352 | break; | |
353 | default: | |
354 | /* For now don't handle any other QP types */ | |
355 | break; | |
356 | } | |
357 | ||
b77d713a | 358 | spin_unlock_irqrestore(&qp->r_lock, flags); |
77241056 MM |
359 | rcu_read_unlock(); |
360 | } /* Unicast QP */ | |
361 | } /* Valid packet with TIDErr */ | |
362 | ||
363 | /* handle "RcvTypeErr" flags */ | |
364 | switch (rte) { | |
365 | case RHF_RTE_ERROR_OP_CODE_ERR: | |
366 | { | |
367 | u32 opcode; | |
368 | void *ebuf = NULL; | |
369 | __be32 *bth = NULL; | |
370 | ||
371 | if (rhf_use_egr_bfr(packet->rhf)) | |
372 | ebuf = packet->ebuf; | |
373 | ||
374 | if (ebuf == NULL) | |
375 | goto drop; /* this should never happen */ | |
376 | ||
377 | if (lnh == HFI1_LRH_BTH) | |
378 | bth = (__be32 *)ebuf; | |
379 | else if (lnh == HFI1_LRH_GRH) | |
380 | bth = (__be32 *)((char *)ebuf + sizeof(struct ib_grh)); | |
381 | else | |
382 | goto drop; | |
383 | ||
384 | opcode = be32_to_cpu(bth[0]) >> 24; | |
385 | opcode &= 0xff; | |
386 | ||
387 | if (opcode == IB_OPCODE_CNP) { | |
388 | /* | |
389 | * Only in pre-B0 h/w is the CNP_OPCODE handled | |
624be1db | 390 | * via this code path. |
77241056 | 391 | */ |
895420dd | 392 | struct rvt_qp *qp = NULL; |
77241056 MM |
393 | u32 lqpn, rqpn; |
394 | u16 rlid; | |
395 | u8 svc_type, sl, sc5; | |
396 | ||
397 | sc5 = (be16_to_cpu(rhdr->lrh[0]) >> 12) & 0xf; | |
398 | if (rhf_dc_info(packet->rhf)) | |
399 | sc5 |= 0x10; | |
400 | sl = ibp->sc_to_sl[sc5]; | |
401 | ||
ec4274f1 | 402 | lqpn = be32_to_cpu(bth[1]) & RVT_QPN_MASK; |
77241056 | 403 | rcu_read_lock(); |
ec4274f1 | 404 | qp = rvt_lookup_qpn(rdi, &ibp->rvp, lqpn); |
77241056 MM |
405 | if (qp == NULL) { |
406 | rcu_read_unlock(); | |
407 | goto drop; | |
408 | } | |
409 | ||
410 | switch (qp->ibqp.qp_type) { | |
411 | case IB_QPT_UD: | |
412 | rlid = 0; | |
413 | rqpn = 0; | |
414 | svc_type = IB_CC_SVCTYPE_UD; | |
415 | break; | |
416 | case IB_QPT_UC: | |
417 | rlid = be16_to_cpu(rhdr->lrh[3]); | |
418 | rqpn = qp->remote_qpn; | |
419 | svc_type = IB_CC_SVCTYPE_UC; | |
420 | break; | |
421 | default: | |
422 | goto drop; | |
423 | } | |
424 | ||
425 | process_becn(ppd, sl, rlid, lqpn, rqpn, svc_type); | |
426 | rcu_read_unlock(); | |
427 | } | |
428 | ||
429 | packet->rhf &= ~RHF_RCV_TYPE_ERR_SMASK; | |
430 | break; | |
431 | } | |
432 | default: | |
433 | break; | |
434 | } | |
435 | ||
436 | drop: | |
437 | return; | |
438 | } | |
439 | ||
440 | static inline void init_packet(struct hfi1_ctxtdata *rcd, | |
441 | struct hfi1_packet *packet) | |
442 | { | |
443 | ||
444 | packet->rsize = rcd->rcvhdrqentsize; /* words */ | |
445 | packet->maxcnt = rcd->rcvhdrq_cnt * packet->rsize; /* words */ | |
446 | packet->rcd = rcd; | |
447 | packet->updegr = 0; | |
448 | packet->etail = -1; | |
f4f30031 | 449 | packet->rhf_addr = get_rhf_addr(rcd); |
77241056 MM |
450 | packet->rhf = rhf_to_cpu(packet->rhf_addr); |
451 | packet->rhqoff = rcd->head; | |
452 | packet->numpkt = 0; | |
453 | packet->rcv_flags = 0; | |
454 | } | |
455 | ||
895420dd | 456 | static void process_ecn(struct rvt_qp *qp, struct hfi1_ib_header *hdr, |
77241056 | 457 | struct hfi1_other_headers *ohdr, |
977940b8 | 458 | u64 rhf, u32 bth1, struct ib_grh *grh) |
77241056 MM |
459 | { |
460 | struct hfi1_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num); | |
977940b8 AK |
461 | u32 rqpn = 0; |
462 | u16 rlid; | |
77241056 | 463 | u8 sc5, svc_type; |
77241056 MM |
464 | |
465 | switch (qp->ibqp.qp_type) { | |
977940b8 AK |
466 | case IB_QPT_SMI: |
467 | case IB_QPT_GSI: | |
77241056 | 468 | case IB_QPT_UD: |
977940b8 | 469 | rlid = be16_to_cpu(hdr->lrh[3]); |
ec4274f1 | 470 | rqpn = be32_to_cpu(ohdr->u.ud.deth[1]) & RVT_QPN_MASK; |
77241056 MM |
471 | svc_type = IB_CC_SVCTYPE_UD; |
472 | break; | |
977940b8 AK |
473 | case IB_QPT_UC: |
474 | rlid = qp->remote_ah_attr.dlid; | |
475 | rqpn = qp->remote_qpn; | |
476 | svc_type = IB_CC_SVCTYPE_UC; | |
477 | break; | |
478 | case IB_QPT_RC: | |
479 | rlid = qp->remote_ah_attr.dlid; | |
480 | rqpn = qp->remote_qpn; | |
481 | svc_type = IB_CC_SVCTYPE_RC; | |
482 | break; | |
77241056 MM |
483 | default: |
484 | return; | |
485 | } | |
486 | ||
77241056 MM |
487 | sc5 = (be16_to_cpu(hdr->lrh[0]) >> 12) & 0xf; |
488 | if (rhf_dc_info(rhf)) | |
489 | sc5 |= 0x10; | |
490 | ||
977940b8 | 491 | if (bth1 & HFI1_FECN_SMASK) { |
77241056 MM |
492 | u16 pkey = (u16)be32_to_cpu(ohdr->bth[0]); |
493 | u16 dlid = be16_to_cpu(hdr->lrh[1]); | |
77241056 | 494 | |
977940b8 | 495 | return_cnp(ibp, qp, rqpn, pkey, dlid, rlid, sc5, grh); |
77241056 MM |
496 | } |
497 | ||
977940b8 | 498 | if (bth1 & HFI1_BECN_SMASK) { |
77241056 | 499 | struct hfi1_pportdata *ppd = ppd_from_ibp(ibp); |
ec4274f1 | 500 | u32 lqpn = bth1 & RVT_QPN_MASK; |
77241056 MM |
501 | u8 sl = ibp->sc_to_sl[sc5]; |
502 | ||
977940b8 | 503 | process_becn(ppd, sl, rlid, lqpn, rqpn, svc_type); |
77241056 | 504 | } |
77241056 MM |
505 | } |
506 | ||
507 | struct ps_mdata { | |
508 | struct hfi1_ctxtdata *rcd; | |
509 | u32 rsize; | |
510 | u32 maxcnt; | |
511 | u32 ps_head; | |
512 | u32 ps_tail; | |
513 | u32 ps_seq; | |
514 | }; | |
515 | ||
516 | static inline void init_ps_mdata(struct ps_mdata *mdata, | |
517 | struct hfi1_packet *packet) | |
518 | { | |
519 | struct hfi1_ctxtdata *rcd = packet->rcd; | |
520 | ||
521 | mdata->rcd = rcd; | |
522 | mdata->rsize = packet->rsize; | |
523 | mdata->maxcnt = packet->maxcnt; | |
3e7ccca0 | 524 | mdata->ps_head = packet->rhqoff; |
77241056 | 525 | |
82c2611d | 526 | if (HFI1_CAP_KGET_MASK(rcd->flags, DMA_RTAIL)) { |
3e7ccca0 | 527 | mdata->ps_tail = get_rcvhdrtail(rcd); |
82c2611d NV |
528 | if (rcd->ctxt == HFI1_CTRL_CTXT) |
529 | mdata->ps_seq = rcd->seq_cnt; | |
530 | else | |
531 | mdata->ps_seq = 0; /* not used with DMA_RTAIL */ | |
77241056 MM |
532 | } else { |
533 | mdata->ps_tail = 0; /* used only with DMA_RTAIL*/ | |
534 | mdata->ps_seq = rcd->seq_cnt; | |
535 | } | |
536 | } | |
537 | ||
82c2611d NV |
538 | static inline int ps_done(struct ps_mdata *mdata, u64 rhf, |
539 | struct hfi1_ctxtdata *rcd) | |
77241056 | 540 | { |
82c2611d | 541 | if (HFI1_CAP_KGET_MASK(rcd->flags, DMA_RTAIL)) |
77241056 MM |
542 | return mdata->ps_head == mdata->ps_tail; |
543 | return mdata->ps_seq != rhf_rcv_seq(rhf); | |
544 | } | |
545 | ||
82c2611d NV |
546 | static inline int ps_skip(struct ps_mdata *mdata, u64 rhf, |
547 | struct hfi1_ctxtdata *rcd) | |
548 | { | |
549 | /* | |
550 | * Control context can potentially receive an invalid rhf. | |
551 | * Drop such packets. | |
552 | */ | |
553 | if ((rcd->ctxt == HFI1_CTRL_CTXT) && (mdata->ps_head != mdata->ps_tail)) | |
554 | return mdata->ps_seq != rhf_rcv_seq(rhf); | |
555 | ||
556 | return 0; | |
557 | } | |
558 | ||
559 | static inline void update_ps_mdata(struct ps_mdata *mdata, | |
560 | struct hfi1_ctxtdata *rcd) | |
77241056 | 561 | { |
77241056 | 562 | mdata->ps_head += mdata->rsize; |
3e7ccca0 | 563 | if (mdata->ps_head >= mdata->maxcnt) |
77241056 | 564 | mdata->ps_head = 0; |
82c2611d NV |
565 | |
566 | /* Control context must do seq counting */ | |
567 | if (!HFI1_CAP_KGET_MASK(rcd->flags, DMA_RTAIL) || | |
568 | (rcd->ctxt == HFI1_CTRL_CTXT)) { | |
77241056 MM |
569 | if (++mdata->ps_seq > 13) |
570 | mdata->ps_seq = 1; | |
571 | } | |
572 | } | |
573 | ||
574 | /* | |
575 | * prescan_rxq - search through the receive queue looking for packets | |
576 | * containing Excplicit Congestion Notifications (FECNs, or BECNs). | |
577 | * When an ECN is found, process the Congestion Notification, and toggle | |
578 | * it off. | |
6c9e50f8 VM |
579 | * This is declared as a macro to allow quick checking of the port to avoid |
580 | * the overhead of a function call if not enabled. | |
77241056 | 581 | */ |
6c9e50f8 VM |
582 | #define prescan_rxq(rcd, packet) \ |
583 | do { \ | |
584 | if (rcd->ppd->cc_prescan) \ | |
585 | __prescan_rxq(packet); \ | |
586 | } while (0) | |
587 | static void __prescan_rxq(struct hfi1_packet *packet) | |
77241056 MM |
588 | { |
589 | struct hfi1_ctxtdata *rcd = packet->rcd; | |
590 | struct ps_mdata mdata; | |
591 | ||
77241056 MM |
592 | init_ps_mdata(&mdata, packet); |
593 | ||
594 | while (1) { | |
595 | struct hfi1_devdata *dd = rcd->dd; | |
596 | struct hfi1_ibport *ibp = &rcd->ppd->ibport_data; | |
597 | __le32 *rhf_addr = (__le32 *) rcd->rcvhdrq + mdata.ps_head + | |
598 | dd->rhf_offset; | |
895420dd | 599 | struct rvt_qp *qp; |
77241056 MM |
600 | struct hfi1_ib_header *hdr; |
601 | struct hfi1_other_headers *ohdr; | |
602 | struct ib_grh *grh = NULL; | |
ec4274f1 | 603 | struct rvt_dev_info *rdi = &dd->verbs_dev.rdi; |
77241056 | 604 | u64 rhf = rhf_to_cpu(rhf_addr); |
977940b8 | 605 | u32 etype = rhf_rcv_type(rhf), qpn, bth1; |
77241056 MM |
606 | int is_ecn = 0; |
607 | u8 lnh; | |
608 | ||
82c2611d | 609 | if (ps_done(&mdata, rhf, rcd)) |
77241056 MM |
610 | break; |
611 | ||
82c2611d NV |
612 | if (ps_skip(&mdata, rhf, rcd)) |
613 | goto next; | |
614 | ||
77241056 MM |
615 | if (etype != RHF_RCV_TYPE_IB) |
616 | goto next; | |
617 | ||
618 | hdr = (struct hfi1_ib_header *) | |
619 | hfi1_get_msgheader(dd, rhf_addr); | |
620 | lnh = be16_to_cpu(hdr->lrh[0]) & 3; | |
621 | ||
622 | if (lnh == HFI1_LRH_BTH) | |
623 | ohdr = &hdr->u.oth; | |
624 | else if (lnh == HFI1_LRH_GRH) { | |
625 | ohdr = &hdr->u.l.oth; | |
626 | grh = &hdr->u.l.grh; | |
627 | } else | |
628 | goto next; /* just in case */ | |
629 | ||
977940b8 AK |
630 | bth1 = be32_to_cpu(ohdr->bth[1]); |
631 | is_ecn = !!(bth1 & (HFI1_FECN_SMASK | HFI1_BECN_SMASK)); | |
77241056 MM |
632 | |
633 | if (!is_ecn) | |
634 | goto next; | |
635 | ||
ec4274f1 | 636 | qpn = bth1 & RVT_QPN_MASK; |
77241056 | 637 | rcu_read_lock(); |
ec4274f1 | 638 | qp = rvt_lookup_qpn(rdi, &ibp->rvp, qpn); |
77241056 MM |
639 | |
640 | if (qp == NULL) { | |
641 | rcu_read_unlock(); | |
642 | goto next; | |
643 | } | |
644 | ||
977940b8 | 645 | process_ecn(qp, hdr, ohdr, rhf, bth1, grh); |
77241056 | 646 | rcu_read_unlock(); |
977940b8 AK |
647 | |
648 | /* turn off BECN, FECN */ | |
649 | bth1 &= ~(HFI1_FECN_SMASK | HFI1_BECN_SMASK); | |
650 | ohdr->bth[1] = cpu_to_be32(bth1); | |
77241056 | 651 | next: |
82c2611d | 652 | update_ps_mdata(&mdata, rcd); |
77241056 MM |
653 | } |
654 | } | |
82c2611d NV |
655 | |
656 | static inline int skip_rcv_packet(struct hfi1_packet *packet, int thread) | |
657 | { | |
658 | int ret = RCV_PKT_OK; | |
659 | ||
660 | /* Set up for the next packet */ | |
661 | packet->rhqoff += packet->rsize; | |
662 | if (packet->rhqoff >= packet->maxcnt) | |
663 | packet->rhqoff = 0; | |
664 | ||
665 | packet->numpkt++; | |
666 | if (unlikely((packet->numpkt & (MAX_PKT_RECV - 1)) == 0)) { | |
667 | if (thread) { | |
668 | cond_resched(); | |
669 | } else { | |
670 | ret = RCV_PKT_LIMIT; | |
671 | this_cpu_inc(*packet->rcd->dd->rcv_limit); | |
672 | } | |
673 | } | |
674 | ||
675 | packet->rhf_addr = (__le32 *)packet->rcd->rcvhdrq + packet->rhqoff + | |
676 | packet->rcd->dd->rhf_offset; | |
677 | packet->rhf = rhf_to_cpu(packet->rhf_addr); | |
678 | ||
679 | return ret; | |
680 | } | |
77241056 | 681 | |
f4f30031 | 682 | static inline int process_rcv_packet(struct hfi1_packet *packet, int thread) |
77241056 MM |
683 | { |
684 | int ret = RCV_PKT_OK; | |
685 | ||
686 | packet->hdr = hfi1_get_msgheader(packet->rcd->dd, | |
687 | packet->rhf_addr); | |
688 | packet->hlen = (u8 *)packet->rhf_addr - (u8 *)packet->hdr; | |
689 | packet->etype = rhf_rcv_type(packet->rhf); | |
690 | /* total length */ | |
691 | packet->tlen = rhf_pkt_len(packet->rhf); /* in bytes */ | |
692 | /* retrieve eager buffer details */ | |
693 | packet->ebuf = NULL; | |
694 | if (rhf_use_egr_bfr(packet->rhf)) { | |
695 | packet->etail = rhf_egr_index(packet->rhf); | |
696 | packet->ebuf = get_egrbuf(packet->rcd, packet->rhf, | |
697 | &packet->updegr); | |
698 | /* | |
699 | * Prefetch the contents of the eager buffer. It is | |
700 | * OK to send a negative length to prefetch_range(). | |
701 | * The +2 is the size of the RHF. | |
702 | */ | |
703 | prefetch_range(packet->ebuf, | |
704 | packet->tlen - ((packet->rcd->rcvhdrqentsize - | |
705 | (rhf_hdrq_offset(packet->rhf)+2)) * 4)); | |
706 | } | |
707 | ||
708 | /* | |
709 | * Call a type specific handler for the packet. We | |
710 | * should be able to trust that etype won't be beyond | |
711 | * the range of valid indexes. If so something is really | |
712 | * wrong and we can probably just let things come | |
713 | * crashing down. There is no need to eat another | |
714 | * comparison in this performance critical code. | |
715 | */ | |
716 | packet->rcd->dd->rhf_rcv_function_map[packet->etype](packet); | |
717 | packet->numpkt++; | |
718 | ||
719 | /* Set up for the next packet */ | |
720 | packet->rhqoff += packet->rsize; | |
721 | if (packet->rhqoff >= packet->maxcnt) | |
722 | packet->rhqoff = 0; | |
723 | ||
f4f30031 DL |
724 | if (unlikely((packet->numpkt & (MAX_PKT_RECV - 1)) == 0)) { |
725 | if (thread) { | |
726 | cond_resched(); | |
727 | } else { | |
728 | ret = RCV_PKT_LIMIT; | |
729 | this_cpu_inc(*packet->rcd->dd->rcv_limit); | |
730 | } | |
77241056 MM |
731 | } |
732 | ||
733 | packet->rhf_addr = (__le32 *) packet->rcd->rcvhdrq + packet->rhqoff + | |
734 | packet->rcd->dd->rhf_offset; | |
735 | packet->rhf = rhf_to_cpu(packet->rhf_addr); | |
736 | ||
737 | return ret; | |
738 | } | |
739 | ||
740 | static inline void process_rcv_update(int last, struct hfi1_packet *packet) | |
741 | { | |
742 | /* | |
743 | * Update head regs etc., every 16 packets, if not last pkt, | |
744 | * to help prevent rcvhdrq overflows, when many packets | |
745 | * are processed and queue is nearly full. | |
746 | * Don't request an interrupt for intermediate updates. | |
747 | */ | |
748 | if (!last && !(packet->numpkt & 0xf)) { | |
749 | update_usrhead(packet->rcd, packet->rhqoff, packet->updegr, | |
750 | packet->etail, 0, 0); | |
751 | packet->updegr = 0; | |
752 | } | |
753 | packet->rcv_flags = 0; | |
754 | } | |
755 | ||
756 | static inline void finish_packet(struct hfi1_packet *packet) | |
757 | { | |
758 | ||
759 | /* | |
760 | * Nothing we need to free for the packet. | |
761 | * | |
762 | * The only thing we need to do is a final update and call for an | |
763 | * interrupt | |
764 | */ | |
765 | update_usrhead(packet->rcd, packet->rcd->head, packet->updegr, | |
766 | packet->etail, rcv_intr_dynamic, packet->numpkt); | |
767 | ||
768 | } | |
769 | ||
770 | static inline void process_rcv_qp_work(struct hfi1_packet *packet) | |
771 | { | |
772 | ||
773 | struct hfi1_ctxtdata *rcd; | |
895420dd | 774 | struct rvt_qp *qp, *nqp; |
77241056 MM |
775 | |
776 | rcd = packet->rcd; | |
777 | rcd->head = packet->rhqoff; | |
778 | ||
779 | /* | |
780 | * Iterate over all QPs waiting to respond. | |
781 | * The list won't change since the IRQ is only run on one CPU. | |
782 | */ | |
783 | list_for_each_entry_safe(qp, nqp, &rcd->qp_wait_list, rspwait) { | |
784 | list_del_init(&qp->rspwait); | |
54d10c1e DD |
785 | if (qp->r_flags & RVT_R_RSP_NAK) { |
786 | qp->r_flags &= ~RVT_R_RSP_NAK; | |
77241056 MM |
787 | hfi1_send_rc_ack(rcd, qp, 0); |
788 | } | |
54d10c1e | 789 | if (qp->r_flags & RVT_R_RSP_SEND) { |
77241056 MM |
790 | unsigned long flags; |
791 | ||
54d10c1e | 792 | qp->r_flags &= ~RVT_R_RSP_SEND; |
77241056 | 793 | spin_lock_irqsave(&qp->s_lock, flags); |
83693bd1 DD |
794 | if (ib_rvt_state_ops[qp->state] & |
795 | RVT_PROCESS_OR_FLUSH_SEND) | |
77241056 MM |
796 | hfi1_schedule_send(qp); |
797 | spin_unlock_irqrestore(&qp->s_lock, flags); | |
798 | } | |
799 | if (atomic_dec_and_test(&qp->refcount)) | |
800 | wake_up(&qp->wait); | |
801 | } | |
802 | } | |
803 | ||
804 | /* | |
805 | * Handle receive interrupts when using the no dma rtail option. | |
806 | */ | |
f4f30031 | 807 | int handle_receive_interrupt_nodma_rtail(struct hfi1_ctxtdata *rcd, int thread) |
77241056 MM |
808 | { |
809 | u32 seq; | |
f4f30031 | 810 | int last = RCV_PKT_OK; |
77241056 MM |
811 | struct hfi1_packet packet; |
812 | ||
813 | init_packet(rcd, &packet); | |
814 | seq = rhf_rcv_seq(packet.rhf); | |
f4f30031 DL |
815 | if (seq != rcd->seq_cnt) { |
816 | last = RCV_PKT_DONE; | |
77241056 | 817 | goto bail; |
f4f30031 | 818 | } |
77241056 | 819 | |
6c9e50f8 | 820 | prescan_rxq(rcd, &packet); |
77241056 | 821 | |
f4f30031 DL |
822 | while (last == RCV_PKT_OK) { |
823 | last = process_rcv_packet(&packet, thread); | |
77241056 MM |
824 | seq = rhf_rcv_seq(packet.rhf); |
825 | if (++rcd->seq_cnt > 13) | |
826 | rcd->seq_cnt = 1; | |
827 | if (seq != rcd->seq_cnt) | |
f4f30031 | 828 | last = RCV_PKT_DONE; |
77241056 MM |
829 | process_rcv_update(last, &packet); |
830 | } | |
831 | process_rcv_qp_work(&packet); | |
832 | bail: | |
833 | finish_packet(&packet); | |
f4f30031 | 834 | return last; |
77241056 MM |
835 | } |
836 | ||
f4f30031 | 837 | int handle_receive_interrupt_dma_rtail(struct hfi1_ctxtdata *rcd, int thread) |
77241056 MM |
838 | { |
839 | u32 hdrqtail; | |
f4f30031 | 840 | int last = RCV_PKT_OK; |
77241056 MM |
841 | struct hfi1_packet packet; |
842 | ||
843 | init_packet(rcd, &packet); | |
844 | hdrqtail = get_rcvhdrtail(rcd); | |
f4f30031 DL |
845 | if (packet.rhqoff == hdrqtail) { |
846 | last = RCV_PKT_DONE; | |
77241056 | 847 | goto bail; |
f4f30031 | 848 | } |
77241056 MM |
849 | smp_rmb(); /* prevent speculative reads of dma'ed hdrq */ |
850 | ||
6c9e50f8 | 851 | prescan_rxq(rcd, &packet); |
77241056 | 852 | |
f4f30031 DL |
853 | while (last == RCV_PKT_OK) { |
854 | last = process_rcv_packet(&packet, thread); | |
77241056 | 855 | if (packet.rhqoff == hdrqtail) |
f4f30031 | 856 | last = RCV_PKT_DONE; |
77241056 MM |
857 | process_rcv_update(last, &packet); |
858 | } | |
859 | process_rcv_qp_work(&packet); | |
860 | bail: | |
861 | finish_packet(&packet); | |
f4f30031 | 862 | return last; |
77241056 MM |
863 | } |
864 | ||
865 | static inline void set_all_nodma_rtail(struct hfi1_devdata *dd) | |
866 | { | |
867 | int i; | |
868 | ||
82c2611d | 869 | for (i = HFI1_CTRL_CTXT + 1; i < dd->first_user_ctxt; i++) |
77241056 MM |
870 | dd->rcd[i]->do_interrupt = |
871 | &handle_receive_interrupt_nodma_rtail; | |
872 | } | |
873 | ||
874 | static inline void set_all_dma_rtail(struct hfi1_devdata *dd) | |
875 | { | |
876 | int i; | |
877 | ||
82c2611d | 878 | for (i = HFI1_CTRL_CTXT + 1; i < dd->first_user_ctxt; i++) |
77241056 MM |
879 | dd->rcd[i]->do_interrupt = |
880 | &handle_receive_interrupt_dma_rtail; | |
881 | } | |
882 | ||
fb9036dd JS |
883 | void set_all_slowpath(struct hfi1_devdata *dd) |
884 | { | |
885 | int i; | |
886 | ||
887 | /* HFI1_CTRL_CTXT must always use the slow path interrupt handler */ | |
888 | for (i = HFI1_CTRL_CTXT + 1; i < dd->first_user_ctxt; i++) | |
889 | dd->rcd[i]->do_interrupt = &handle_receive_interrupt; | |
890 | } | |
891 | ||
892 | static inline int set_armed_to_active(struct hfi1_ctxtdata *rcd, | |
893 | struct hfi1_packet packet, | |
894 | struct hfi1_devdata *dd) | |
895 | { | |
896 | struct work_struct *lsaw = &rcd->ppd->linkstate_active_work; | |
897 | struct hfi1_message_header *hdr = hfi1_get_msgheader(packet.rcd->dd, | |
898 | packet.rhf_addr); | |
899 | ||
900 | if (hdr2sc(hdr, packet.rhf) != 0xf) { | |
901 | int hwstate = read_logical_state(dd); | |
902 | ||
903 | if (hwstate != LSTATE_ACTIVE) { | |
904 | dd_dev_info(dd, "Unexpected link state %d\n", hwstate); | |
905 | return 0; | |
906 | } | |
907 | ||
908 | queue_work(rcd->ppd->hfi1_wq, lsaw); | |
909 | return 1; | |
910 | } | |
911 | return 0; | |
912 | } | |
913 | ||
77241056 MM |
914 | /* |
915 | * handle_receive_interrupt - receive a packet | |
916 | * @rcd: the context | |
917 | * | |
918 | * Called from interrupt handler for errors or receive interrupt. | |
919 | * This is the slow path interrupt handler. | |
920 | */ | |
f4f30031 | 921 | int handle_receive_interrupt(struct hfi1_ctxtdata *rcd, int thread) |
77241056 | 922 | { |
77241056 MM |
923 | struct hfi1_devdata *dd = rcd->dd; |
924 | u32 hdrqtail; | |
82c2611d | 925 | int needset, last = RCV_PKT_OK; |
77241056 | 926 | struct hfi1_packet packet; |
82c2611d NV |
927 | int skip_pkt = 0; |
928 | ||
929 | /* Control context will always use the slow path interrupt handler */ | |
930 | needset = (rcd->ctxt == HFI1_CTRL_CTXT) ? 0 : 1; | |
77241056 MM |
931 | |
932 | init_packet(rcd, &packet); | |
933 | ||
82c2611d | 934 | if (!HFI1_CAP_KGET_MASK(rcd->flags, DMA_RTAIL)) { |
77241056 MM |
935 | u32 seq = rhf_rcv_seq(packet.rhf); |
936 | ||
f4f30031 DL |
937 | if (seq != rcd->seq_cnt) { |
938 | last = RCV_PKT_DONE; | |
77241056 | 939 | goto bail; |
f4f30031 | 940 | } |
77241056 MM |
941 | hdrqtail = 0; |
942 | } else { | |
943 | hdrqtail = get_rcvhdrtail(rcd); | |
f4f30031 DL |
944 | if (packet.rhqoff == hdrqtail) { |
945 | last = RCV_PKT_DONE; | |
77241056 | 946 | goto bail; |
f4f30031 | 947 | } |
77241056 | 948 | smp_rmb(); /* prevent speculative reads of dma'ed hdrq */ |
82c2611d NV |
949 | |
950 | /* | |
951 | * Control context can potentially receive an invalid | |
952 | * rhf. Drop such packets. | |
953 | */ | |
954 | if (rcd->ctxt == HFI1_CTRL_CTXT) { | |
955 | u32 seq = rhf_rcv_seq(packet.rhf); | |
956 | ||
957 | if (seq != rcd->seq_cnt) | |
958 | skip_pkt = 1; | |
959 | } | |
77241056 MM |
960 | } |
961 | ||
6c9e50f8 | 962 | prescan_rxq(rcd, &packet); |
77241056 | 963 | |
f4f30031 | 964 | while (last == RCV_PKT_OK) { |
77241056 MM |
965 | |
966 | if (unlikely(dd->do_drop && atomic_xchg(&dd->drop_packet, | |
967 | DROP_PACKET_OFF) == DROP_PACKET_ON)) { | |
968 | dd->do_drop = 0; | |
969 | ||
970 | /* On to the next packet */ | |
971 | packet.rhqoff += packet.rsize; | |
972 | packet.rhf_addr = (__le32 *) rcd->rcvhdrq + | |
973 | packet.rhqoff + | |
974 | dd->rhf_offset; | |
975 | packet.rhf = rhf_to_cpu(packet.rhf_addr); | |
976 | ||
82c2611d NV |
977 | } else if (skip_pkt) { |
978 | last = skip_rcv_packet(&packet, thread); | |
979 | skip_pkt = 0; | |
77241056 | 980 | } else { |
fb9036dd JS |
981 | /* Auto activate link on non-SC15 packet receive */ |
982 | if (unlikely(rcd->ppd->host_link_state == | |
983 | HLS_UP_ARMED) && | |
984 | set_armed_to_active(rcd, packet, dd)) | |
985 | goto bail; | |
f4f30031 | 986 | last = process_rcv_packet(&packet, thread); |
77241056 MM |
987 | } |
988 | ||
82c2611d | 989 | if (!HFI1_CAP_KGET_MASK(rcd->flags, DMA_RTAIL)) { |
77241056 MM |
990 | u32 seq = rhf_rcv_seq(packet.rhf); |
991 | ||
992 | if (++rcd->seq_cnt > 13) | |
993 | rcd->seq_cnt = 1; | |
994 | if (seq != rcd->seq_cnt) | |
f4f30031 | 995 | last = RCV_PKT_DONE; |
77241056 MM |
996 | if (needset) { |
997 | dd_dev_info(dd, | |
998 | "Switching to NO_DMA_RTAIL\n"); | |
999 | set_all_nodma_rtail(dd); | |
1000 | needset = 0; | |
1001 | } | |
1002 | } else { | |
1003 | if (packet.rhqoff == hdrqtail) | |
f4f30031 | 1004 | last = RCV_PKT_DONE; |
82c2611d NV |
1005 | /* |
1006 | * Control context can potentially receive an invalid | |
1007 | * rhf. Drop such packets. | |
1008 | */ | |
1009 | if (rcd->ctxt == HFI1_CTRL_CTXT) { | |
1010 | u32 seq = rhf_rcv_seq(packet.rhf); | |
1011 | ||
1012 | if (++rcd->seq_cnt > 13) | |
1013 | rcd->seq_cnt = 1; | |
1014 | if (!last && (seq != rcd->seq_cnt)) | |
1015 | skip_pkt = 1; | |
1016 | } | |
1017 | ||
77241056 MM |
1018 | if (needset) { |
1019 | dd_dev_info(dd, | |
1020 | "Switching to DMA_RTAIL\n"); | |
1021 | set_all_dma_rtail(dd); | |
1022 | needset = 0; | |
1023 | } | |
1024 | } | |
1025 | ||
1026 | process_rcv_update(last, &packet); | |
1027 | } | |
1028 | ||
1029 | process_rcv_qp_work(&packet); | |
1030 | ||
1031 | bail: | |
1032 | /* | |
1033 | * Always write head at end, and setup rcv interrupt, even | |
1034 | * if no packets were processed. | |
1035 | */ | |
1036 | finish_packet(&packet); | |
f4f30031 | 1037 | return last; |
77241056 MM |
1038 | } |
1039 | ||
fb9036dd JS |
1040 | /* |
1041 | * We may discover in the interrupt that the hardware link state has | |
1042 | * changed from ARMED to ACTIVE (due to the arrival of a non-SC15 packet), | |
1043 | * and we need to update the driver's notion of the link state. We cannot | |
1044 | * run set_link_state from interrupt context, so we queue this function on | |
1045 | * a workqueue. | |
1046 | * | |
1047 | * We delay the regular interrupt processing until after the state changes | |
1048 | * so that the link will be in the correct state by the time any application | |
1049 | * we wake up attempts to send a reply to any message it received. | |
1050 | * (Subsequent receive interrupts may possibly force the wakeup before we | |
1051 | * update the link state.) | |
1052 | * | |
1053 | * The rcd is freed in hfi1_free_ctxtdata after hfi1_postinit_cleanup invokes | |
1054 | * dd->f_cleanup(dd) to disable the interrupt handler and flush workqueues, | |
1055 | * so we're safe from use-after-free of the rcd. | |
1056 | */ | |
1057 | void receive_interrupt_work(struct work_struct *work) | |
1058 | { | |
1059 | struct hfi1_pportdata *ppd = container_of(work, struct hfi1_pportdata, | |
1060 | linkstate_active_work); | |
1061 | struct hfi1_devdata *dd = ppd->dd; | |
1062 | int i; | |
1063 | ||
1064 | /* Received non-SC15 packet implies neighbor_normal */ | |
1065 | ppd->neighbor_normal = 1; | |
1066 | set_link_state(ppd, HLS_UP_ACTIVE); | |
1067 | ||
1068 | /* | |
1069 | * Interrupt all kernel contexts that could have had an | |
1070 | * interrupt during auto activation. | |
1071 | */ | |
1072 | for (i = HFI1_CTRL_CTXT; i < dd->first_user_ctxt; i++) | |
1073 | force_recv_intr(dd->rcd[i]); | |
1074 | } | |
1075 | ||
77241056 MM |
1076 | /* |
1077 | * Convert a given MTU size to the on-wire MAD packet enumeration. | |
1078 | * Return -1 if the size is invalid. | |
1079 | */ | |
1080 | int mtu_to_enum(u32 mtu, int default_if_bad) | |
1081 | { | |
1082 | switch (mtu) { | |
1083 | case 0: return OPA_MTU_0; | |
1084 | case 256: return OPA_MTU_256; | |
1085 | case 512: return OPA_MTU_512; | |
1086 | case 1024: return OPA_MTU_1024; | |
1087 | case 2048: return OPA_MTU_2048; | |
1088 | case 4096: return OPA_MTU_4096; | |
1089 | case 8192: return OPA_MTU_8192; | |
1090 | case 10240: return OPA_MTU_10240; | |
1091 | } | |
1092 | return default_if_bad; | |
1093 | } | |
1094 | ||
1095 | u16 enum_to_mtu(int mtu) | |
1096 | { | |
1097 | switch (mtu) { | |
1098 | case OPA_MTU_0: return 0; | |
1099 | case OPA_MTU_256: return 256; | |
1100 | case OPA_MTU_512: return 512; | |
1101 | case OPA_MTU_1024: return 1024; | |
1102 | case OPA_MTU_2048: return 2048; | |
1103 | case OPA_MTU_4096: return 4096; | |
1104 | case OPA_MTU_8192: return 8192; | |
1105 | case OPA_MTU_10240: return 10240; | |
1106 | default: return 0xffff; | |
1107 | } | |
1108 | } | |
1109 | ||
1110 | /* | |
1111 | * set_mtu - set the MTU | |
1112 | * @ppd: the per port data | |
1113 | * | |
1114 | * We can handle "any" incoming size, the issue here is whether we | |
1115 | * need to restrict our outgoing size. We do not deal with what happens | |
1116 | * to programs that are already running when the size changes. | |
1117 | */ | |
1118 | int set_mtu(struct hfi1_pportdata *ppd) | |
1119 | { | |
1120 | struct hfi1_devdata *dd = ppd->dd; | |
1121 | int i, drain, ret = 0, is_up = 0; | |
1122 | ||
1123 | ppd->ibmtu = 0; | |
1124 | for (i = 0; i < ppd->vls_supported; i++) | |
1125 | if (ppd->ibmtu < dd->vld[i].mtu) | |
1126 | ppd->ibmtu = dd->vld[i].mtu; | |
1127 | ppd->ibmaxlen = ppd->ibmtu + lrh_max_header_bytes(ppd->dd); | |
1128 | ||
1129 | mutex_lock(&ppd->hls_lock); | |
1130 | if (ppd->host_link_state == HLS_UP_INIT | |
1131 | || ppd->host_link_state == HLS_UP_ARMED | |
1132 | || ppd->host_link_state == HLS_UP_ACTIVE) | |
1133 | is_up = 1; | |
1134 | ||
1135 | drain = !is_ax(dd) && is_up; | |
1136 | ||
1137 | if (drain) | |
1138 | /* | |
1139 | * MTU is specified per-VL. To ensure that no packet gets | |
1140 | * stuck (due, e.g., to the MTU for the packet's VL being | |
1141 | * reduced), empty the per-VL FIFOs before adjusting MTU. | |
1142 | */ | |
1143 | ret = stop_drain_data_vls(dd); | |
1144 | ||
1145 | if (ret) { | |
1146 | dd_dev_err(dd, "%s: cannot stop/drain VLs - refusing to change per-VL MTUs\n", | |
1147 | __func__); | |
1148 | goto err; | |
1149 | } | |
1150 | ||
1151 | hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_MTU, 0); | |
1152 | ||
1153 | if (drain) | |
1154 | open_fill_data_vls(dd); /* reopen all VLs */ | |
1155 | ||
1156 | err: | |
1157 | mutex_unlock(&ppd->hls_lock); | |
1158 | ||
1159 | return ret; | |
1160 | } | |
1161 | ||
1162 | int hfi1_set_lid(struct hfi1_pportdata *ppd, u32 lid, u8 lmc) | |
1163 | { | |
1164 | struct hfi1_devdata *dd = ppd->dd; | |
1165 | ||
1166 | ppd->lid = lid; | |
1167 | ppd->lmc = lmc; | |
1168 | hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_LIDLMC, 0); | |
1169 | ||
1170 | dd_dev_info(dd, "IB%u:%u got a lid: 0x%x\n", dd->unit, ppd->port, lid); | |
1171 | ||
1172 | return 0; | |
1173 | } | |
1174 | ||
1175 | /* | |
1176 | * Following deal with the "obviously simple" task of overriding the state | |
1177 | * of the LEDs, which normally indicate link physical and logical status. | |
1178 | * The complications arise in dealing with different hardware mappings | |
1179 | * and the board-dependent routine being called from interrupts. | |
1180 | * and then there's the requirement to _flash_ them. | |
1181 | */ | |
1182 | #define LED_OVER_FREQ_SHIFT 8 | |
1183 | #define LED_OVER_FREQ_MASK (0xFF<<LED_OVER_FREQ_SHIFT) | |
1184 | /* Below is "non-zero" to force override, but both actual LEDs are off */ | |
1185 | #define LED_OVER_BOTH_OFF (8) | |
1186 | ||
1187 | static void run_led_override(unsigned long opaque) | |
1188 | { | |
1189 | struct hfi1_pportdata *ppd = (struct hfi1_pportdata *)opaque; | |
1190 | struct hfi1_devdata *dd = ppd->dd; | |
1191 | int timeoff; | |
1192 | int ph_idx; | |
1193 | ||
1194 | if (!(dd->flags & HFI1_INITTED)) | |
1195 | return; | |
1196 | ||
1197 | ph_idx = ppd->led_override_phase++ & 1; | |
1198 | ppd->led_override = ppd->led_override_vals[ph_idx]; | |
1199 | timeoff = ppd->led_override_timeoff; | |
1200 | ||
1201 | /* | |
1202 | * don't re-fire the timer if user asked for it to be off; we let | |
1203 | * it fire one more time after they turn it off to simplify | |
1204 | */ | |
1205 | if (ppd->led_override_vals[0] || ppd->led_override_vals[1]) | |
1206 | mod_timer(&ppd->led_override_timer, jiffies + timeoff); | |
1207 | } | |
1208 | ||
1209 | void hfi1_set_led_override(struct hfi1_pportdata *ppd, unsigned int val) | |
1210 | { | |
1211 | struct hfi1_devdata *dd = ppd->dd; | |
1212 | int timeoff, freq; | |
1213 | ||
1214 | if (!(dd->flags & HFI1_INITTED)) | |
1215 | return; | |
1216 | ||
1217 | /* First check if we are blinking. If not, use 1HZ polling */ | |
1218 | timeoff = HZ; | |
1219 | freq = (val & LED_OVER_FREQ_MASK) >> LED_OVER_FREQ_SHIFT; | |
1220 | ||
1221 | if (freq) { | |
1222 | /* For blink, set each phase from one nybble of val */ | |
1223 | ppd->led_override_vals[0] = val & 0xF; | |
1224 | ppd->led_override_vals[1] = (val >> 4) & 0xF; | |
1225 | timeoff = (HZ << 4)/freq; | |
1226 | } else { | |
1227 | /* Non-blink set both phases the same. */ | |
1228 | ppd->led_override_vals[0] = val & 0xF; | |
1229 | ppd->led_override_vals[1] = val & 0xF; | |
1230 | } | |
1231 | ppd->led_override_timeoff = timeoff; | |
1232 | ||
1233 | /* | |
1234 | * If the timer has not already been started, do so. Use a "quick" | |
1235 | * timeout so the function will be called soon, to look at our request. | |
1236 | */ | |
1237 | if (atomic_inc_return(&ppd->led_override_timer_active) == 1) { | |
1238 | /* Need to start timer */ | |
a3faf606 MFW |
1239 | setup_timer(&ppd->led_override_timer, run_led_override, |
1240 | (unsigned long)ppd); | |
1241 | ||
77241056 MM |
1242 | ppd->led_override_timer.expires = jiffies + 1; |
1243 | add_timer(&ppd->led_override_timer); | |
1244 | } else { | |
1245 | if (ppd->led_override_vals[0] || ppd->led_override_vals[1]) | |
1246 | mod_timer(&ppd->led_override_timer, jiffies + 1); | |
1247 | atomic_dec(&ppd->led_override_timer_active); | |
1248 | } | |
1249 | } | |
1250 | ||
1251 | /** | |
1252 | * hfi1_reset_device - reset the chip if possible | |
1253 | * @unit: the device to reset | |
1254 | * | |
1255 | * Whether or not reset is successful, we attempt to re-initialize the chip | |
1256 | * (that is, much like a driver unload/reload). We clear the INITTED flag | |
1257 | * so that the various entry points will fail until we reinitialize. For | |
1258 | * now, we only allow this if no user contexts are open that use chip resources | |
1259 | */ | |
1260 | int hfi1_reset_device(int unit) | |
1261 | { | |
1262 | int ret, i; | |
1263 | struct hfi1_devdata *dd = hfi1_lookup(unit); | |
1264 | struct hfi1_pportdata *ppd; | |
1265 | unsigned long flags; | |
1266 | int pidx; | |
1267 | ||
1268 | if (!dd) { | |
1269 | ret = -ENODEV; | |
1270 | goto bail; | |
1271 | } | |
1272 | ||
1273 | dd_dev_info(dd, "Reset on unit %u requested\n", unit); | |
1274 | ||
1275 | if (!dd->kregbase || !(dd->flags & HFI1_PRESENT)) { | |
1276 | dd_dev_info(dd, | |
1277 | "Invalid unit number %u or not initialized or not present\n", | |
1278 | unit); | |
1279 | ret = -ENXIO; | |
1280 | goto bail; | |
1281 | } | |
1282 | ||
1283 | spin_lock_irqsave(&dd->uctxt_lock, flags); | |
1284 | if (dd->rcd) | |
1285 | for (i = dd->first_user_ctxt; i < dd->num_rcv_contexts; i++) { | |
1286 | if (!dd->rcd[i] || !dd->rcd[i]->cnt) | |
1287 | continue; | |
1288 | spin_unlock_irqrestore(&dd->uctxt_lock, flags); | |
1289 | ret = -EBUSY; | |
1290 | goto bail; | |
1291 | } | |
1292 | spin_unlock_irqrestore(&dd->uctxt_lock, flags); | |
1293 | ||
1294 | for (pidx = 0; pidx < dd->num_pports; ++pidx) { | |
1295 | ppd = dd->pport + pidx; | |
1296 | if (atomic_read(&ppd->led_override_timer_active)) { | |
1297 | /* Need to stop LED timer, _then_ shut off LEDs */ | |
1298 | del_timer_sync(&ppd->led_override_timer); | |
1299 | atomic_set(&ppd->led_override_timer_active, 0); | |
1300 | } | |
1301 | ||
1302 | /* Shut off LEDs after we are sure timer is not running */ | |
1303 | ppd->led_override = LED_OVER_BOTH_OFF; | |
1304 | } | |
1305 | if (dd->flags & HFI1_HAS_SEND_DMA) | |
1306 | sdma_exit(dd); | |
1307 | ||
1308 | hfi1_reset_cpu_counters(dd); | |
1309 | ||
1310 | ret = hfi1_init(dd, 1); | |
1311 | ||
1312 | if (ret) | |
1313 | dd_dev_err(dd, | |
1314 | "Reinitialize unit %u after reset failed with %d\n", | |
1315 | unit, ret); | |
1316 | else | |
1317 | dd_dev_info(dd, "Reinitialized unit %u after resetting\n", | |
1318 | unit); | |
1319 | ||
1320 | bail: | |
1321 | return ret; | |
1322 | } | |
1323 | ||
1324 | void handle_eflags(struct hfi1_packet *packet) | |
1325 | { | |
1326 | struct hfi1_ctxtdata *rcd = packet->rcd; | |
1327 | u32 rte = rhf_rcv_type_err(packet->rhf); | |
1328 | ||
77241056 | 1329 | rcv_hdrerr(rcd, rcd->ppd, packet); |
a03a03e9 IH |
1330 | if (rhf_err_flags(packet->rhf)) |
1331 | dd_dev_err(rcd->dd, | |
1332 | "receive context %d: rhf 0x%016llx, errs [ %s%s%s%s%s%s%s%s] rte 0x%x\n", | |
1333 | rcd->ctxt, packet->rhf, | |
1334 | packet->rhf & RHF_K_HDR_LEN_ERR ? "k_hdr_len " : "", | |
1335 | packet->rhf & RHF_DC_UNC_ERR ? "dc_unc " : "", | |
1336 | packet->rhf & RHF_DC_ERR ? "dc " : "", | |
1337 | packet->rhf & RHF_TID_ERR ? "tid " : "", | |
1338 | packet->rhf & RHF_LEN_ERR ? "len " : "", | |
1339 | packet->rhf & RHF_ECC_ERR ? "ecc " : "", | |
1340 | packet->rhf & RHF_VCRC_ERR ? "vcrc " : "", | |
1341 | packet->rhf & RHF_ICRC_ERR ? "icrc " : "", | |
1342 | rte); | |
77241056 MM |
1343 | } |
1344 | ||
1345 | /* | |
1346 | * The following functions are called by the interrupt handler. They are type | |
1347 | * specific handlers for each packet type. | |
1348 | */ | |
1349 | int process_receive_ib(struct hfi1_packet *packet) | |
1350 | { | |
1351 | trace_hfi1_rcvhdr(packet->rcd->ppd->dd, | |
1352 | packet->rcd->ctxt, | |
1353 | rhf_err_flags(packet->rhf), | |
1354 | RHF_RCV_TYPE_IB, | |
1355 | packet->hlen, | |
1356 | packet->tlen, | |
1357 | packet->updegr, | |
1358 | rhf_egr_index(packet->rhf)); | |
1359 | ||
1360 | if (unlikely(rhf_err_flags(packet->rhf))) { | |
1361 | handle_eflags(packet); | |
1362 | return RHF_RCV_CONTINUE; | |
1363 | } | |
1364 | ||
1365 | hfi1_ib_rcv(packet); | |
1366 | return RHF_RCV_CONTINUE; | |
1367 | } | |
1368 | ||
1369 | int process_receive_bypass(struct hfi1_packet *packet) | |
1370 | { | |
1371 | if (unlikely(rhf_err_flags(packet->rhf))) | |
1372 | handle_eflags(packet); | |
1373 | ||
1374 | dd_dev_err(packet->rcd->dd, | |
1375 | "Bypass packets are not supported in normal operation. Dropping\n"); | |
1376 | return RHF_RCV_CONTINUE; | |
1377 | } | |
1378 | ||
1379 | int process_receive_error(struct hfi1_packet *packet) | |
1380 | { | |
1381 | handle_eflags(packet); | |
1382 | ||
1383 | if (unlikely(rhf_err_flags(packet->rhf))) | |
1384 | dd_dev_err(packet->rcd->dd, | |
1385 | "Unhandled error packet received. Dropping.\n"); | |
1386 | ||
1387 | return RHF_RCV_CONTINUE; | |
1388 | } | |
1389 | ||
1390 | int kdeth_process_expected(struct hfi1_packet *packet) | |
1391 | { | |
1392 | if (unlikely(rhf_err_flags(packet->rhf))) | |
1393 | handle_eflags(packet); | |
1394 | ||
1395 | dd_dev_err(packet->rcd->dd, | |
1396 | "Unhandled expected packet received. Dropping.\n"); | |
1397 | return RHF_RCV_CONTINUE; | |
1398 | } | |
1399 | ||
1400 | int kdeth_process_eager(struct hfi1_packet *packet) | |
1401 | { | |
1402 | if (unlikely(rhf_err_flags(packet->rhf))) | |
1403 | handle_eflags(packet); | |
1404 | ||
1405 | dd_dev_err(packet->rcd->dd, | |
1406 | "Unhandled eager packet received. Dropping.\n"); | |
1407 | return RHF_RCV_CONTINUE; | |
1408 | } | |
1409 | ||
1410 | int process_receive_invalid(struct hfi1_packet *packet) | |
1411 | { | |
1412 | dd_dev_err(packet->rcd->dd, "Invalid packet type %d. Dropping\n", | |
1413 | rhf_rcv_type(packet->rhf)); | |
1414 | return RHF_RCV_CONTINUE; | |
1415 | } |