1 /*
2  * Copyright(c) 2015-2017 Intel Corporation.
3  *
4  * This file is provided under a dual BSD/GPLv2 license.  When using or
5  * redistributing this file, you may do so under either license.
6  *
7  * GPL LICENSE SUMMARY
8  *
9  * This program is free software; you can redistribute it and/or modify
10  * it under the terms of version 2 of the GNU General Public License as
11  * published by the Free Software Foundation.
12  *
13  * This program is distributed in the hope that it will be useful, but
14  * WITHOUT ANY WARRANTY; without even the implied warranty of
15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
16  * General Public License for more details.
17  *
18  * BSD LICENSE
19  *
20  * Redistribution and use in source and binary forms, with or without
21  * modification, are permitted provided that the following conditions
22  * are met:
23  *
24  *  - Redistributions of source code must retain the above copyright
25  *    notice, this list of conditions and the following disclaimer.
26  *  - Redistributions in binary form must reproduce the above copyright
27  *    notice, this list of conditions and the following disclaimer in
28  *    the documentation and/or other materials provided with the
29  *    distribution.
30  *  - Neither the name of Intel Corporation nor the names of its
31  *    contributors may be used to endorse or promote products derived
32  *    from this software without specific prior written permission.
33  *
34  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
35  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
36  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
37  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
38  * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
39  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
40  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
41  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
42  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
43  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
44  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
45  *
46  */
47 
48 #include <linux/spinlock.h>
49 #include <linux/pci.h>
50 #include <linux/io.h>
51 #include <linux/delay.h>
52 #include <linux/netdevice.h>
53 #include <linux/vmalloc.h>
54 #include <linux/module.h>
55 #include <linux/prefetch.h>
56 #include <rdma/ib_verbs.h>
57 
58 #include "hfi.h"
59 #include "trace.h"
60 #include "qp.h"
61 #include "sdma.h"
62 #include "debugfs.h"
63 #include "vnic.h"
64 
65 #undef pr_fmt
66 #define pr_fmt(fmt) DRIVER_NAME ": " fmt
67 
68 /*
69  * The size has to be longer than this string, so we can append
70  * board/chip information to it in the initialization code.
71  */
72 const char ib_hfi1_version[] = HFI1_DRIVER_VERSION "\n";
73 
74 DEFINE_SPINLOCK(hfi1_devs_lock);
75 LIST_HEAD(hfi1_dev_list);
76 DEFINE_MUTEX(hfi1_mutex);	/* general driver use */
77 
78 unsigned int hfi1_max_mtu = HFI1_DEFAULT_MAX_MTU;
79 module_param_named(max_mtu, hfi1_max_mtu, uint, S_IRUGO);
80 MODULE_PARM_DESC(max_mtu, "Set max MTU bytes, default is " __stringify(
81 		 HFI1_DEFAULT_MAX_MTU));
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 *val, const struct kernel_param *kp);
89 static int hfi1_caps_get(char *buffer, const struct kernel_param *kp);
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 /*
106  * MAX_PKT_THREAD_RCV is the max # of packets processed before
107  * the qp_wait_list queue is flushed.
108  */
109 #define MAX_PKT_RECV_THREAD (MAX_PKT_RECV * 4)
110 #define EGR_HEAD_UPDATE_THRESHOLD 16
111 
112 struct hfi1_ib_stats hfi1_stats;
113 
114 static int hfi1_caps_set(const char *val, const struct kernel_param *kp)
115 {
116 	int ret = 0;
117 	unsigned long *cap_mask_ptr = (unsigned long *)kp->arg,
118 		cap_mask = *cap_mask_ptr, value, diff,
119 		write_mask = ((HFI1_CAP_WRITABLE_MASK << HFI1_CAP_USER_SHIFT) |
120 			      HFI1_CAP_WRITABLE_MASK);
121 
122 	ret = kstrtoul(val, 0, &value);
123 	if (ret) {
124 		pr_warn("Invalid module parameter value for 'cap_mask'\n");
125 		goto done;
126 	}
127 	/* Get the changed bits (except the locked bit) */
128 	diff = value ^ (cap_mask & ~HFI1_CAP_LOCKED_SMASK);
129 
130 	/* Remove any bits that are not allowed to change after driver load */
131 	if (HFI1_CAP_LOCKED() && (diff & ~write_mask)) {
132 		pr_warn("Ignoring non-writable capability bits %#lx\n",
133 			diff & ~write_mask);
134 		diff &= write_mask;
135 	}
136 
137 	/* Mask off any reserved bits */
138 	diff &= ~HFI1_CAP_RESERVED_MASK;
139 	/* Clear any previously set and changing bits */
140 	cap_mask &= ~diff;
141 	/* Update the bits with the new capability */
142 	cap_mask |= (value & diff);
143 	/* Check for any kernel/user restrictions */
144 	diff = (cap_mask & (HFI1_CAP_MUST_HAVE_KERN << HFI1_CAP_USER_SHIFT)) ^
145 		((cap_mask & HFI1_CAP_MUST_HAVE_KERN) << HFI1_CAP_USER_SHIFT);
146 	cap_mask &= ~diff;
147 	/* Set the bitmask to the final set */
148 	*cap_mask_ptr = cap_mask;
149 done:
150 	return ret;
151 }
152 
153 static int hfi1_caps_get(char *buffer, const struct kernel_param *kp)
154 {
155 	unsigned long cap_mask = *(unsigned long *)kp->arg;
156 
157 	cap_mask &= ~HFI1_CAP_LOCKED_SMASK;
158 	cap_mask |= ((cap_mask & HFI1_CAP_K2U) << HFI1_CAP_USER_SHIFT);
159 
160 	return scnprintf(buffer, PAGE_SIZE, "0x%lx", cap_mask);
161 }
162 
163 const char *get_unit_name(int unit)
164 {
165 	static char iname[16];
166 
167 	snprintf(iname, sizeof(iname), DRIVER_NAME "_%u", unit);
168 	return iname;
169 }
170 
171 const char *get_card_name(struct rvt_dev_info *rdi)
172 {
173 	struct hfi1_ibdev *ibdev = container_of(rdi, struct hfi1_ibdev, rdi);
174 	struct hfi1_devdata *dd = container_of(ibdev,
175 					       struct hfi1_devdata, verbs_dev);
176 	return get_unit_name(dd->unit);
177 }
178 
179 struct pci_dev *get_pci_dev(struct rvt_dev_info *rdi)
180 {
181 	struct hfi1_ibdev *ibdev = container_of(rdi, struct hfi1_ibdev, rdi);
182 	struct hfi1_devdata *dd = container_of(ibdev,
183 					       struct hfi1_devdata, verbs_dev);
184 	return dd->pcidev;
185 }
186 
187 /*
188  * Return count of units with at least one port ACTIVE.
189  */
190 int hfi1_count_active_units(void)
191 {
192 	struct hfi1_devdata *dd;
193 	struct hfi1_pportdata *ppd;
194 	unsigned long flags;
195 	int pidx, nunits_active = 0;
196 
197 	spin_lock_irqsave(&hfi1_devs_lock, flags);
198 	list_for_each_entry(dd, &hfi1_dev_list, list) {
199 		if (!(dd->flags & HFI1_PRESENT) || !dd->kregbase)
200 			continue;
201 		for (pidx = 0; pidx < dd->num_pports; ++pidx) {
202 			ppd = dd->pport + pidx;
203 			if (ppd->lid && ppd->linkup) {
204 				nunits_active++;
205 				break;
206 			}
207 		}
208 	}
209 	spin_unlock_irqrestore(&hfi1_devs_lock, flags);
210 	return nunits_active;
211 }
212 
213 /*
214  * Get address of eager buffer from it's index (allocated in chunks, not
215  * contiguous).
216  */
217 static inline void *get_egrbuf(const struct hfi1_ctxtdata *rcd, u64 rhf,
218 			       u8 *update)
219 {
220 	u32 idx = rhf_egr_index(rhf), offset = rhf_egr_buf_offset(rhf);
221 
222 	*update |= !(idx & (rcd->egrbufs.threshold - 1)) && !offset;
223 	return (void *)(((u64)(rcd->egrbufs.rcvtids[idx].addr)) +
224 			(offset * RCV_BUF_BLOCK_SIZE));
225 }
226 
227 /*
228  * Validate and encode the a given RcvArray Buffer size.
229  * The function will check whether the given size falls within
230  * allowed size ranges for the respective type and, optionally,
231  * return the proper encoding.
232  */
233 int hfi1_rcvbuf_validate(u32 size, u8 type, u16 *encoded)
234 {
235 	if (unlikely(!PAGE_ALIGNED(size)))
236 		return 0;
237 	if (unlikely(size < MIN_EAGER_BUFFER))
238 		return 0;
239 	if (size >
240 	    (type == PT_EAGER ? MAX_EAGER_BUFFER : MAX_EXPECTED_BUFFER))
241 		return 0;
242 	if (encoded)
243 		*encoded = ilog2(size / PAGE_SIZE) + 1;
244 	return 1;
245 }
246 
247 static void rcv_hdrerr(struct hfi1_ctxtdata *rcd, struct hfi1_pportdata *ppd,
248 		       struct hfi1_packet *packet)
249 {
250 	struct ib_header *rhdr = packet->hdr;
251 	u32 rte = rhf_rcv_type_err(packet->rhf);
252 	int lnh = ib_get_lnh(rhdr);
253 	struct hfi1_ibport *ibp = rcd_to_iport(rcd);
254 	struct hfi1_devdata *dd = ppd->dd;
255 	struct rvt_dev_info *rdi = &dd->verbs_dev.rdi;
256 
257 	if (packet->rhf & (RHF_VCRC_ERR | RHF_ICRC_ERR))
258 		return;
259 
260 	if (packet->rhf & RHF_TID_ERR) {
261 		/* For TIDERR and RC QPs preemptively schedule a NAK */
262 		struct ib_other_headers *ohdr = NULL;
263 		u32 tlen = rhf_pkt_len(packet->rhf); /* in bytes */
264 		u16 lid  = ib_get_dlid(rhdr);
265 		u32 qp_num;
266 		u32 rcv_flags = 0;
267 
268 		/* Sanity check packet */
269 		if (tlen < 24)
270 			goto drop;
271 
272 		/* Check for GRH */
273 		if (lnh == HFI1_LRH_BTH) {
274 			ohdr = &rhdr->u.oth;
275 		} else if (lnh == HFI1_LRH_GRH) {
276 			u32 vtf;
277 
278 			ohdr = &rhdr->u.l.oth;
279 			if (rhdr->u.l.grh.next_hdr != IB_GRH_NEXT_HDR)
280 				goto drop;
281 			vtf = be32_to_cpu(rhdr->u.l.grh.version_tclass_flow);
282 			if ((vtf >> IB_GRH_VERSION_SHIFT) != IB_GRH_VERSION)
283 				goto drop;
284 			rcv_flags |= HFI1_HAS_GRH;
285 		} else {
286 			goto drop;
287 		}
288 		/* Get the destination QP number. */
289 		qp_num = be32_to_cpu(ohdr->bth[1]) & RVT_QPN_MASK;
290 		if (lid < be16_to_cpu(IB_MULTICAST_LID_BASE)) {
291 			struct rvt_qp *qp;
292 			unsigned long flags;
293 
294 			rcu_read_lock();
295 			qp = rvt_lookup_qpn(rdi, &ibp->rvp, qp_num);
296 			if (!qp) {
297 				rcu_read_unlock();
298 				goto drop;
299 			}
300 
301 			/*
302 			 * Handle only RC QPs - for other QP types drop error
303 			 * packet.
304 			 */
305 			spin_lock_irqsave(&qp->r_lock, flags);
306 
307 			/* Check for valid receive state. */
308 			if (!(ib_rvt_state_ops[qp->state] &
309 			      RVT_PROCESS_RECV_OK)) {
310 				ibp->rvp.n_pkt_drops++;
311 			}
312 
313 			switch (qp->ibqp.qp_type) {
314 			case IB_QPT_RC:
315 				hfi1_rc_hdrerr(
316 					rcd,
317 					rhdr,
318 					rcv_flags,
319 					qp);
320 				break;
321 			default:
322 				/* For now don't handle any other QP types */
323 				break;
324 			}
325 
326 			spin_unlock_irqrestore(&qp->r_lock, flags);
327 			rcu_read_unlock();
328 		} /* Unicast QP */
329 	} /* Valid packet with TIDErr */
330 
331 	/* handle "RcvTypeErr" flags */
332 	switch (rte) {
333 	case RHF_RTE_ERROR_OP_CODE_ERR:
334 	{
335 		u32 opcode;
336 		void *ebuf = NULL;
337 		__be32 *bth = NULL;
338 
339 		if (rhf_use_egr_bfr(packet->rhf))
340 			ebuf = packet->ebuf;
341 
342 		if (!ebuf)
343 			goto drop; /* this should never happen */
344 
345 		if (lnh == HFI1_LRH_BTH)
346 			bth = (__be32 *)ebuf;
347 		else if (lnh == HFI1_LRH_GRH)
348 			bth = (__be32 *)((char *)ebuf + sizeof(struct ib_grh));
349 		else
350 			goto drop;
351 
352 		opcode = be32_to_cpu(bth[0]) >> 24;
353 		opcode &= 0xff;
354 
355 		if (opcode == IB_OPCODE_CNP) {
356 			/*
357 			 * Only in pre-B0 h/w is the CNP_OPCODE handled
358 			 * via this code path.
359 			 */
360 			struct rvt_qp *qp = NULL;
361 			u32 lqpn, rqpn;
362 			u16 rlid;
363 			u8 svc_type, sl, sc5;
364 
365 			sc5 = hfi1_9B_get_sc5(rhdr, packet->rhf);
366 			sl = ibp->sc_to_sl[sc5];
367 
368 			lqpn = be32_to_cpu(bth[1]) & RVT_QPN_MASK;
369 			rcu_read_lock();
370 			qp = rvt_lookup_qpn(rdi, &ibp->rvp, lqpn);
371 			if (!qp) {
372 				rcu_read_unlock();
373 				goto drop;
374 			}
375 
376 			switch (qp->ibqp.qp_type) {
377 			case IB_QPT_UD:
378 				rlid = 0;
379 				rqpn = 0;
380 				svc_type = IB_CC_SVCTYPE_UD;
381 				break;
382 			case IB_QPT_UC:
383 				rlid = ib_get_slid(rhdr);
384 				rqpn = qp->remote_qpn;
385 				svc_type = IB_CC_SVCTYPE_UC;
386 				break;
387 			default:
388 				goto drop;
389 			}
390 
391 			process_becn(ppd, sl, rlid, lqpn, rqpn, svc_type);
392 			rcu_read_unlock();
393 		}
394 
395 		packet->rhf &= ~RHF_RCV_TYPE_ERR_SMASK;
396 		break;
397 	}
398 	default:
399 		break;
400 	}
401 
402 drop:
403 	return;
404 }
405 
406 static inline void init_packet(struct hfi1_ctxtdata *rcd,
407 			       struct hfi1_packet *packet)
408 {
409 	packet->rsize = rcd->rcvhdrqentsize; /* words */
410 	packet->maxcnt = rcd->rcvhdrq_cnt * packet->rsize; /* words */
411 	packet->rcd = rcd;
412 	packet->updegr = 0;
413 	packet->etail = -1;
414 	packet->rhf_addr = get_rhf_addr(rcd);
415 	packet->rhf = rhf_to_cpu(packet->rhf_addr);
416 	packet->rhqoff = rcd->head;
417 	packet->numpkt = 0;
418 	packet->rcv_flags = 0;
419 }
420 
421 void hfi1_process_ecn_slowpath(struct rvt_qp *qp, struct hfi1_packet *pkt,
422 			       bool do_cnp)
423 {
424 	struct hfi1_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num);
425 	struct ib_header *hdr = pkt->hdr;
426 	struct ib_other_headers *ohdr = pkt->ohdr;
427 	struct ib_grh *grh = NULL;
428 	u32 rqpn = 0, bth1;
429 	u16 rlid, dlid = ib_get_dlid(hdr);
430 	u8 sc, svc_type;
431 	bool is_mcast = false;
432 
433 	if (pkt->rcv_flags & HFI1_HAS_GRH)
434 		grh = &hdr->u.l.grh;
435 
436 	switch (qp->ibqp.qp_type) {
437 	case IB_QPT_SMI:
438 	case IB_QPT_GSI:
439 	case IB_QPT_UD:
440 		rlid = ib_get_slid(hdr);
441 		rqpn = be32_to_cpu(ohdr->u.ud.deth[1]) & RVT_QPN_MASK;
442 		svc_type = IB_CC_SVCTYPE_UD;
443 		is_mcast = (dlid > be16_to_cpu(IB_MULTICAST_LID_BASE)) &&
444 			(dlid != be16_to_cpu(IB_LID_PERMISSIVE));
445 		break;
446 	case IB_QPT_UC:
447 		rlid = rdma_ah_get_dlid(&qp->remote_ah_attr);
448 		rqpn = qp->remote_qpn;
449 		svc_type = IB_CC_SVCTYPE_UC;
450 		break;
451 	case IB_QPT_RC:
452 		rlid = rdma_ah_get_dlid(&qp->remote_ah_attr);
453 		rqpn = qp->remote_qpn;
454 		svc_type = IB_CC_SVCTYPE_RC;
455 		break;
456 	default:
457 		return;
458 	}
459 
460 	sc = hfi1_9B_get_sc5(hdr, pkt->rhf);
461 
462 	bth1 = be32_to_cpu(ohdr->bth[1]);
463 	if (do_cnp && (bth1 & IB_FECN_SMASK)) {
464 		u16 pkey = (u16)be32_to_cpu(ohdr->bth[0]);
465 
466 		return_cnp(ibp, qp, rqpn, pkey, dlid, rlid, sc, grh);
467 	}
468 
469 	if (!is_mcast && (bth1 & IB_BECN_SMASK)) {
470 		struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
471 		u32 lqpn = bth1 & RVT_QPN_MASK;
472 		u8 sl = ibp->sc_to_sl[sc];
473 
474 		process_becn(ppd, sl, rlid, lqpn, rqpn, svc_type);
475 	}
476 
477 }
478 
479 struct ps_mdata {
480 	struct hfi1_ctxtdata *rcd;
481 	u32 rsize;
482 	u32 maxcnt;
483 	u32 ps_head;
484 	u32 ps_tail;
485 	u32 ps_seq;
486 };
487 
488 static inline void init_ps_mdata(struct ps_mdata *mdata,
489 				 struct hfi1_packet *packet)
490 {
491 	struct hfi1_ctxtdata *rcd = packet->rcd;
492 
493 	mdata->rcd = rcd;
494 	mdata->rsize = packet->rsize;
495 	mdata->maxcnt = packet->maxcnt;
496 	mdata->ps_head = packet->rhqoff;
497 
498 	if (HFI1_CAP_KGET_MASK(rcd->flags, DMA_RTAIL)) {
499 		mdata->ps_tail = get_rcvhdrtail(rcd);
500 		if (rcd->ctxt == HFI1_CTRL_CTXT)
501 			mdata->ps_seq = rcd->seq_cnt;
502 		else
503 			mdata->ps_seq = 0; /* not used with DMA_RTAIL */
504 	} else {
505 		mdata->ps_tail = 0; /* used only with DMA_RTAIL*/
506 		mdata->ps_seq = rcd->seq_cnt;
507 	}
508 }
509 
510 static inline int ps_done(struct ps_mdata *mdata, u64 rhf,
511 			  struct hfi1_ctxtdata *rcd)
512 {
513 	if (HFI1_CAP_KGET_MASK(rcd->flags, DMA_RTAIL))
514 		return mdata->ps_head == mdata->ps_tail;
515 	return mdata->ps_seq != rhf_rcv_seq(rhf);
516 }
517 
518 static inline int ps_skip(struct ps_mdata *mdata, u64 rhf,
519 			  struct hfi1_ctxtdata *rcd)
520 {
521 	/*
522 	 * Control context can potentially receive an invalid rhf.
523 	 * Drop such packets.
524 	 */
525 	if ((rcd->ctxt == HFI1_CTRL_CTXT) && (mdata->ps_head != mdata->ps_tail))
526 		return mdata->ps_seq != rhf_rcv_seq(rhf);
527 
528 	return 0;
529 }
530 
531 static inline void update_ps_mdata(struct ps_mdata *mdata,
532 				   struct hfi1_ctxtdata *rcd)
533 {
534 	mdata->ps_head += mdata->rsize;
535 	if (mdata->ps_head >= mdata->maxcnt)
536 		mdata->ps_head = 0;
537 
538 	/* Control context must do seq counting */
539 	if (!HFI1_CAP_KGET_MASK(rcd->flags, DMA_RTAIL) ||
540 	    (rcd->ctxt == HFI1_CTRL_CTXT)) {
541 		if (++mdata->ps_seq > 13)
542 			mdata->ps_seq = 1;
543 	}
544 }
545 
546 /*
547  * prescan_rxq - search through the receive queue looking for packets
548  * containing Excplicit Congestion Notifications (FECNs, or BECNs).
549  * When an ECN is found, process the Congestion Notification, and toggle
550  * it off.
551  * This is declared as a macro to allow quick checking of the port to avoid
552  * the overhead of a function call if not enabled.
553  */
554 #define prescan_rxq(rcd, packet) \
555 	do { \
556 		if (rcd->ppd->cc_prescan) \
557 			__prescan_rxq(packet); \
558 	} while (0)
559 static void __prescan_rxq(struct hfi1_packet *packet)
560 {
561 	struct hfi1_ctxtdata *rcd = packet->rcd;
562 	struct ps_mdata mdata;
563 
564 	init_ps_mdata(&mdata, packet);
565 
566 	while (1) {
567 		struct hfi1_devdata *dd = rcd->dd;
568 		struct hfi1_ibport *ibp = rcd_to_iport(rcd);
569 		__le32 *rhf_addr = (__le32 *)rcd->rcvhdrq + mdata.ps_head +
570 					 dd->rhf_offset;
571 		struct rvt_qp *qp;
572 		struct ib_header *hdr;
573 		struct rvt_dev_info *rdi = &dd->verbs_dev.rdi;
574 		u64 rhf = rhf_to_cpu(rhf_addr);
575 		u32 etype = rhf_rcv_type(rhf), qpn, bth1;
576 		int is_ecn = 0;
577 		u8 lnh;
578 
579 		if (ps_done(&mdata, rhf, rcd))
580 			break;
581 
582 		if (ps_skip(&mdata, rhf, rcd))
583 			goto next;
584 
585 		if (etype != RHF_RCV_TYPE_IB)
586 			goto next;
587 
588 		packet->hdr = hfi1_get_msgheader(dd, rhf_addr);
589 		hdr = packet->hdr;
590 		lnh = ib_get_lnh(hdr);
591 
592 		if (lnh == HFI1_LRH_BTH) {
593 			packet->ohdr = &hdr->u.oth;
594 		} else if (lnh == HFI1_LRH_GRH) {
595 			packet->ohdr = &hdr->u.l.oth;
596 			packet->rcv_flags |= HFI1_HAS_GRH;
597 		} else {
598 			goto next; /* just in case */
599 		}
600 
601 		bth1 = be32_to_cpu(packet->ohdr->bth[1]);
602 		is_ecn = !!(bth1 & (IB_FECN_SMASK | IB_BECN_SMASK));
603 
604 		if (!is_ecn)
605 			goto next;
606 
607 		qpn = bth1 & RVT_QPN_MASK;
608 		rcu_read_lock();
609 		qp = rvt_lookup_qpn(rdi, &ibp->rvp, qpn);
610 
611 		if (!qp) {
612 			rcu_read_unlock();
613 			goto next;
614 		}
615 
616 		process_ecn(qp, packet, true);
617 		rcu_read_unlock();
618 
619 		/* turn off BECN, FECN */
620 		bth1 &= ~(IB_FECN_SMASK | IB_BECN_SMASK);
621 		packet->ohdr->bth[1] = cpu_to_be32(bth1);
622 next:
623 		update_ps_mdata(&mdata, rcd);
624 	}
625 }
626 
627 static void process_rcv_qp_work(struct hfi1_ctxtdata *rcd)
628 {
629 	struct rvt_qp *qp, *nqp;
630 
631 	/*
632 	 * Iterate over all QPs waiting to respond.
633 	 * The list won't change since the IRQ is only run on one CPU.
634 	 */
635 	list_for_each_entry_safe(qp, nqp, &rcd->qp_wait_list, rspwait) {
636 		list_del_init(&qp->rspwait);
637 		if (qp->r_flags & RVT_R_RSP_NAK) {
638 			qp->r_flags &= ~RVT_R_RSP_NAK;
639 			hfi1_send_rc_ack(rcd, qp, 0);
640 		}
641 		if (qp->r_flags & RVT_R_RSP_SEND) {
642 			unsigned long flags;
643 
644 			qp->r_flags &= ~RVT_R_RSP_SEND;
645 			spin_lock_irqsave(&qp->s_lock, flags);
646 			if (ib_rvt_state_ops[qp->state] &
647 					RVT_PROCESS_OR_FLUSH_SEND)
648 				hfi1_schedule_send(qp);
649 			spin_unlock_irqrestore(&qp->s_lock, flags);
650 		}
651 		rvt_put_qp(qp);
652 	}
653 }
654 
655 static noinline int max_packet_exceeded(struct hfi1_packet *packet, int thread)
656 {
657 	if (thread) {
658 		if ((packet->numpkt & (MAX_PKT_RECV_THREAD - 1)) == 0)
659 			/* allow defered processing */
660 			process_rcv_qp_work(packet->rcd);
661 		cond_resched();
662 		return RCV_PKT_OK;
663 	} else {
664 		this_cpu_inc(*packet->rcd->dd->rcv_limit);
665 		return RCV_PKT_LIMIT;
666 	}
667 }
668 
669 static inline int check_max_packet(struct hfi1_packet *packet, int thread)
670 {
671 	int ret = RCV_PKT_OK;
672 
673 	if (unlikely((packet->numpkt & (MAX_PKT_RECV - 1)) == 0))
674 		ret = max_packet_exceeded(packet, thread);
675 	return ret;
676 }
677 
678 static noinline int skip_rcv_packet(struct hfi1_packet *packet, int thread)
679 {
680 	int ret;
681 
682 	/* Set up for the next packet */
683 	packet->rhqoff += packet->rsize;
684 	if (packet->rhqoff >= packet->maxcnt)
685 		packet->rhqoff = 0;
686 
687 	packet->numpkt++;
688 	ret = check_max_packet(packet, thread);
689 
690 	packet->rhf_addr = (__le32 *)packet->rcd->rcvhdrq + packet->rhqoff +
691 				     packet->rcd->dd->rhf_offset;
692 	packet->rhf = rhf_to_cpu(packet->rhf_addr);
693 
694 	return ret;
695 }
696 
697 static inline int process_rcv_packet(struct hfi1_packet *packet, int thread)
698 {
699 	int ret;
700 
701 	packet->hdr = hfi1_get_msgheader(packet->rcd->dd,
702 					 packet->rhf_addr);
703 	packet->hlen = (u8 *)packet->rhf_addr - (u8 *)packet->hdr;
704 	packet->etype = rhf_rcv_type(packet->rhf);
705 	/* total length */
706 	packet->tlen = rhf_pkt_len(packet->rhf); /* in bytes */
707 	/* retrieve eager buffer details */
708 	packet->ebuf = NULL;
709 	if (rhf_use_egr_bfr(packet->rhf)) {
710 		packet->etail = rhf_egr_index(packet->rhf);
711 		packet->ebuf = get_egrbuf(packet->rcd, packet->rhf,
712 				 &packet->updegr);
713 		/*
714 		 * Prefetch the contents of the eager buffer.  It is
715 		 * OK to send a negative length to prefetch_range().
716 		 * The +2 is the size of the RHF.
717 		 */
718 		prefetch_range(packet->ebuf,
719 			       packet->tlen - ((packet->rcd->rcvhdrqentsize -
720 					       (rhf_hdrq_offset(packet->rhf)
721 						+ 2)) * 4));
722 	}
723 
724 	/*
725 	 * Call a type specific handler for the packet. We
726 	 * should be able to trust that etype won't be beyond
727 	 * the range of valid indexes. If so something is really
728 	 * wrong and we can probably just let things come
729 	 * crashing down. There is no need to eat another
730 	 * comparison in this performance critical code.
731 	 */
732 	packet->rcd->dd->rhf_rcv_function_map[packet->etype](packet);
733 	packet->numpkt++;
734 
735 	/* Set up for the next packet */
736 	packet->rhqoff += packet->rsize;
737 	if (packet->rhqoff >= packet->maxcnt)
738 		packet->rhqoff = 0;
739 
740 	ret = check_max_packet(packet, thread);
741 
742 	packet->rhf_addr = (__le32 *)packet->rcd->rcvhdrq + packet->rhqoff +
743 				      packet->rcd->dd->rhf_offset;
744 	packet->rhf = rhf_to_cpu(packet->rhf_addr);
745 
746 	return ret;
747 }
748 
749 static inline void process_rcv_update(int last, struct hfi1_packet *packet)
750 {
751 	/*
752 	 * Update head regs etc., every 16 packets, if not last pkt,
753 	 * to help prevent rcvhdrq overflows, when many packets
754 	 * are processed and queue is nearly full.
755 	 * Don't request an interrupt for intermediate updates.
756 	 */
757 	if (!last && !(packet->numpkt & 0xf)) {
758 		update_usrhead(packet->rcd, packet->rhqoff, packet->updegr,
759 			       packet->etail, 0, 0);
760 		packet->updegr = 0;
761 	}
762 	packet->rcv_flags = 0;
763 }
764 
765 static inline void finish_packet(struct hfi1_packet *packet)
766 {
767 	/*
768 	 * Nothing we need to free for the packet.
769 	 *
770 	 * The only thing we need to do is a final update and call for an
771 	 * interrupt
772 	 */
773 	update_usrhead(packet->rcd, packet->rcd->head, packet->updegr,
774 		       packet->etail, rcv_intr_dynamic, packet->numpkt);
775 }
776 
777 /*
778  * Handle receive interrupts when using the no dma rtail option.
779  */
780 int handle_receive_interrupt_nodma_rtail(struct hfi1_ctxtdata *rcd, int thread)
781 {
782 	u32 seq;
783 	int last = RCV_PKT_OK;
784 	struct hfi1_packet packet;
785 
786 	init_packet(rcd, &packet);
787 	seq = rhf_rcv_seq(packet.rhf);
788 	if (seq != rcd->seq_cnt) {
789 		last = RCV_PKT_DONE;
790 		goto bail;
791 	}
792 
793 	prescan_rxq(rcd, &packet);
794 
795 	while (last == RCV_PKT_OK) {
796 		last = process_rcv_packet(&packet, thread);
797 		seq = rhf_rcv_seq(packet.rhf);
798 		if (++rcd->seq_cnt > 13)
799 			rcd->seq_cnt = 1;
800 		if (seq != rcd->seq_cnt)
801 			last = RCV_PKT_DONE;
802 		process_rcv_update(last, &packet);
803 	}
804 	process_rcv_qp_work(rcd);
805 	rcd->head = packet.rhqoff;
806 bail:
807 	finish_packet(&packet);
808 	return last;
809 }
810 
811 int handle_receive_interrupt_dma_rtail(struct hfi1_ctxtdata *rcd, int thread)
812 {
813 	u32 hdrqtail;
814 	int last = RCV_PKT_OK;
815 	struct hfi1_packet packet;
816 
817 	init_packet(rcd, &packet);
818 	hdrqtail = get_rcvhdrtail(rcd);
819 	if (packet.rhqoff == hdrqtail) {
820 		last = RCV_PKT_DONE;
821 		goto bail;
822 	}
823 	smp_rmb();  /* prevent speculative reads of dma'ed hdrq */
824 
825 	prescan_rxq(rcd, &packet);
826 
827 	while (last == RCV_PKT_OK) {
828 		last = process_rcv_packet(&packet, thread);
829 		if (packet.rhqoff == hdrqtail)
830 			last = RCV_PKT_DONE;
831 		process_rcv_update(last, &packet);
832 	}
833 	process_rcv_qp_work(rcd);
834 	rcd->head = packet.rhqoff;
835 bail:
836 	finish_packet(&packet);
837 	return last;
838 }
839 
840 static inline void set_nodma_rtail(struct hfi1_devdata *dd, u8 ctxt)
841 {
842 	int i;
843 
844 	/*
845 	 * For dynamically allocated kernel contexts (like vnic) switch
846 	 * interrupt handler only for that context. Otherwise, switch
847 	 * interrupt handler for all statically allocated kernel contexts.
848 	 */
849 	if (ctxt >= dd->first_dyn_alloc_ctxt) {
850 		dd->rcd[ctxt]->do_interrupt =
851 			&handle_receive_interrupt_nodma_rtail;
852 		return;
853 	}
854 
855 	for (i = HFI1_CTRL_CTXT + 1; i < dd->first_dyn_alloc_ctxt; i++)
856 		dd->rcd[i]->do_interrupt =
857 			&handle_receive_interrupt_nodma_rtail;
858 }
859 
860 static inline void set_dma_rtail(struct hfi1_devdata *dd, u8 ctxt)
861 {
862 	int i;
863 
864 	/*
865 	 * For dynamically allocated kernel contexts (like vnic) switch
866 	 * interrupt handler only for that context. Otherwise, switch
867 	 * interrupt handler for all statically allocated kernel contexts.
868 	 */
869 	if (ctxt >= dd->first_dyn_alloc_ctxt) {
870 		dd->rcd[ctxt]->do_interrupt =
871 			&handle_receive_interrupt_dma_rtail;
872 		return;
873 	}
874 
875 	for (i = HFI1_CTRL_CTXT + 1; i < dd->first_dyn_alloc_ctxt; i++)
876 		dd->rcd[i]->do_interrupt =
877 			&handle_receive_interrupt_dma_rtail;
878 }
879 
880 void set_all_slowpath(struct hfi1_devdata *dd)
881 {
882 	int i;
883 
884 	/* HFI1_CTRL_CTXT must always use the slow path interrupt handler */
885 	for (i = HFI1_CTRL_CTXT + 1; i < dd->num_rcv_contexts; i++) {
886 		struct hfi1_ctxtdata *rcd = dd->rcd[i];
887 
888 		if ((i < dd->first_dyn_alloc_ctxt) ||
889 		    (rcd && rcd->sc && (rcd->sc->type == SC_KERNEL)))
890 			rcd->do_interrupt = &handle_receive_interrupt;
891 	}
892 }
893 
894 static inline int set_armed_to_active(struct hfi1_ctxtdata *rcd,
895 				      struct hfi1_packet *packet,
896 				      struct hfi1_devdata *dd)
897 {
898 	struct work_struct *lsaw = &rcd->ppd->linkstate_active_work;
899 	struct ib_header *hdr = hfi1_get_msgheader(packet->rcd->dd,
900 						   packet->rhf_addr);
901 	u8 etype = rhf_rcv_type(packet->rhf);
902 
903 	if (etype == RHF_RCV_TYPE_IB &&
904 	    hfi1_9B_get_sc5(hdr, packet->rhf) != 0xf) {
905 		int hwstate = read_logical_state(dd);
906 
907 		if (hwstate != LSTATE_ACTIVE) {
908 			dd_dev_info(dd, "Unexpected link state %d\n", hwstate);
909 			return 0;
910 		}
911 
912 		queue_work(rcd->ppd->hfi1_wq, lsaw);
913 		return 1;
914 	}
915 	return 0;
916 }
917 
918 /*
919  * handle_receive_interrupt - receive a packet
920  * @rcd: the context
921  *
922  * Called from interrupt handler for errors or receive interrupt.
923  * This is the slow path interrupt handler.
924  */
925 int handle_receive_interrupt(struct hfi1_ctxtdata *rcd, int thread)
926 {
927 	struct hfi1_devdata *dd = rcd->dd;
928 	u32 hdrqtail;
929 	int needset, last = RCV_PKT_OK;
930 	struct hfi1_packet packet;
931 	int skip_pkt = 0;
932 
933 	/* Control context will always use the slow path interrupt handler */
934 	needset = (rcd->ctxt == HFI1_CTRL_CTXT) ? 0 : 1;
935 
936 	init_packet(rcd, &packet);
937 
938 	if (!HFI1_CAP_KGET_MASK(rcd->flags, DMA_RTAIL)) {
939 		u32 seq = rhf_rcv_seq(packet.rhf);
940 
941 		if (seq != rcd->seq_cnt) {
942 			last = RCV_PKT_DONE;
943 			goto bail;
944 		}
945 		hdrqtail = 0;
946 	} else {
947 		hdrqtail = get_rcvhdrtail(rcd);
948 		if (packet.rhqoff == hdrqtail) {
949 			last = RCV_PKT_DONE;
950 			goto bail;
951 		}
952 		smp_rmb();  /* prevent speculative reads of dma'ed hdrq */
953 
954 		/*
955 		 * Control context can potentially receive an invalid
956 		 * rhf. Drop such packets.
957 		 */
958 		if (rcd->ctxt == HFI1_CTRL_CTXT) {
959 			u32 seq = rhf_rcv_seq(packet.rhf);
960 
961 			if (seq != rcd->seq_cnt)
962 				skip_pkt = 1;
963 		}
964 	}
965 
966 	prescan_rxq(rcd, &packet);
967 
968 	while (last == RCV_PKT_OK) {
969 		if (unlikely(dd->do_drop &&
970 			     atomic_xchg(&dd->drop_packet, DROP_PACKET_OFF) ==
971 			     DROP_PACKET_ON)) {
972 			dd->do_drop = 0;
973 
974 			/* On to the next packet */
975 			packet.rhqoff += packet.rsize;
976 			packet.rhf_addr = (__le32 *)rcd->rcvhdrq +
977 					  packet.rhqoff +
978 					  dd->rhf_offset;
979 			packet.rhf = rhf_to_cpu(packet.rhf_addr);
980 
981 		} else if (skip_pkt) {
982 			last = skip_rcv_packet(&packet, thread);
983 			skip_pkt = 0;
984 		} else {
985 			/* Auto activate link on non-SC15 packet receive */
986 			if (unlikely(rcd->ppd->host_link_state ==
987 				     HLS_UP_ARMED) &&
988 			    set_armed_to_active(rcd, &packet, dd))
989 				goto bail;
990 			last = process_rcv_packet(&packet, thread);
991 		}
992 
993 		if (!HFI1_CAP_KGET_MASK(rcd->flags, DMA_RTAIL)) {
994 			u32 seq = rhf_rcv_seq(packet.rhf);
995 
996 			if (++rcd->seq_cnt > 13)
997 				rcd->seq_cnt = 1;
998 			if (seq != rcd->seq_cnt)
999 				last = RCV_PKT_DONE;
1000 			if (needset) {
1001 				dd_dev_info(dd, "Switching to NO_DMA_RTAIL\n");
1002 				set_nodma_rtail(dd, rcd->ctxt);
1003 				needset = 0;
1004 			}
1005 		} else {
1006 			if (packet.rhqoff == hdrqtail)
1007 				last = RCV_PKT_DONE;
1008 			/*
1009 			 * Control context can potentially receive an invalid
1010 			 * rhf. Drop such packets.
1011 			 */
1012 			if (rcd->ctxt == HFI1_CTRL_CTXT) {
1013 				u32 seq = rhf_rcv_seq(packet.rhf);
1014 
1015 				if (++rcd->seq_cnt > 13)
1016 					rcd->seq_cnt = 1;
1017 				if (!last && (seq != rcd->seq_cnt))
1018 					skip_pkt = 1;
1019 			}
1020 
1021 			if (needset) {
1022 				dd_dev_info(dd,
1023 					    "Switching to DMA_RTAIL\n");
1024 				set_dma_rtail(dd, rcd->ctxt);
1025 				needset = 0;
1026 			}
1027 		}
1028 
1029 		process_rcv_update(last, &packet);
1030 	}
1031 
1032 	process_rcv_qp_work(rcd);
1033 	rcd->head = packet.rhqoff;
1034 
1035 bail:
1036 	/*
1037 	 * Always write head at end, and setup rcv interrupt, even
1038 	 * if no packets were processed.
1039 	 */
1040 	finish_packet(&packet);
1041 	return last;
1042 }
1043 
1044 /*
1045  * We may discover in the interrupt that the hardware link state has
1046  * changed from ARMED to ACTIVE (due to the arrival of a non-SC15 packet),
1047  * and we need to update the driver's notion of the link state.  We cannot
1048  * run set_link_state from interrupt context, so we queue this function on
1049  * a workqueue.
1050  *
1051  * We delay the regular interrupt processing until after the state changes
1052  * so that the link will be in the correct state by the time any application
1053  * we wake up attempts to send a reply to any message it received.
1054  * (Subsequent receive interrupts may possibly force the wakeup before we
1055  * update the link state.)
1056  *
1057  * The rcd is freed in hfi1_free_ctxtdata after hfi1_postinit_cleanup invokes
1058  * dd->f_cleanup(dd) to disable the interrupt handler and flush workqueues,
1059  * so we're safe from use-after-free of the rcd.
1060  */
1061 void receive_interrupt_work(struct work_struct *work)
1062 {
1063 	struct hfi1_pportdata *ppd = container_of(work, struct hfi1_pportdata,
1064 						  linkstate_active_work);
1065 	struct hfi1_devdata *dd = ppd->dd;
1066 	int i;
1067 
1068 	/* Received non-SC15 packet implies neighbor_normal */
1069 	ppd->neighbor_normal = 1;
1070 	set_link_state(ppd, HLS_UP_ACTIVE);
1071 
1072 	/*
1073 	 * Interrupt all statically allocated kernel contexts that could
1074 	 * have had an interrupt during auto activation.
1075 	 */
1076 	for (i = HFI1_CTRL_CTXT; i < dd->first_dyn_alloc_ctxt; i++)
1077 		force_recv_intr(dd->rcd[i]);
1078 }
1079 
1080 /*
1081  * Convert a given MTU size to the on-wire MAD packet enumeration.
1082  * Return -1 if the size is invalid.
1083  */
1084 int mtu_to_enum(u32 mtu, int default_if_bad)
1085 {
1086 	switch (mtu) {
1087 	case     0: return OPA_MTU_0;
1088 	case   256: return OPA_MTU_256;
1089 	case   512: return OPA_MTU_512;
1090 	case  1024: return OPA_MTU_1024;
1091 	case  2048: return OPA_MTU_2048;
1092 	case  4096: return OPA_MTU_4096;
1093 	case  8192: return OPA_MTU_8192;
1094 	case 10240: return OPA_MTU_10240;
1095 	}
1096 	return default_if_bad;
1097 }
1098 
1099 u16 enum_to_mtu(int mtu)
1100 {
1101 	switch (mtu) {
1102 	case OPA_MTU_0:     return 0;
1103 	case OPA_MTU_256:   return 256;
1104 	case OPA_MTU_512:   return 512;
1105 	case OPA_MTU_1024:  return 1024;
1106 	case OPA_MTU_2048:  return 2048;
1107 	case OPA_MTU_4096:  return 4096;
1108 	case OPA_MTU_8192:  return 8192;
1109 	case OPA_MTU_10240: return 10240;
1110 	default: return 0xffff;
1111 	}
1112 }
1113 
1114 /*
1115  * set_mtu - set the MTU
1116  * @ppd: the per port data
1117  *
1118  * We can handle "any" incoming size, the issue here is whether we
1119  * need to restrict our outgoing size.  We do not deal with what happens
1120  * to programs that are already running when the size changes.
1121  */
1122 int set_mtu(struct hfi1_pportdata *ppd)
1123 {
1124 	struct hfi1_devdata *dd = ppd->dd;
1125 	int i, drain, ret = 0, is_up = 0;
1126 
1127 	ppd->ibmtu = 0;
1128 	for (i = 0; i < ppd->vls_supported; i++)
1129 		if (ppd->ibmtu < dd->vld[i].mtu)
1130 			ppd->ibmtu = dd->vld[i].mtu;
1131 	ppd->ibmaxlen = ppd->ibmtu + lrh_max_header_bytes(ppd->dd);
1132 
1133 	mutex_lock(&ppd->hls_lock);
1134 	if (ppd->host_link_state == HLS_UP_INIT ||
1135 	    ppd->host_link_state == HLS_UP_ARMED ||
1136 	    ppd->host_link_state == HLS_UP_ACTIVE)
1137 		is_up = 1;
1138 
1139 	drain = !is_ax(dd) && is_up;
1140 
1141 	if (drain)
1142 		/*
1143 		 * MTU is specified per-VL. To ensure that no packet gets
1144 		 * stuck (due, e.g., to the MTU for the packet's VL being
1145 		 * reduced), empty the per-VL FIFOs before adjusting MTU.
1146 		 */
1147 		ret = stop_drain_data_vls(dd);
1148 
1149 	if (ret) {
1150 		dd_dev_err(dd, "%s: cannot stop/drain VLs - refusing to change per-VL MTUs\n",
1151 			   __func__);
1152 		goto err;
1153 	}
1154 
1155 	hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_MTU, 0);
1156 
1157 	if (drain)
1158 		open_fill_data_vls(dd); /* reopen all VLs */
1159 
1160 err:
1161 	mutex_unlock(&ppd->hls_lock);
1162 
1163 	return ret;
1164 }
1165 
1166 int hfi1_set_lid(struct hfi1_pportdata *ppd, u32 lid, u8 lmc)
1167 {
1168 	struct hfi1_devdata *dd = ppd->dd;
1169 
1170 	ppd->lid = lid;
1171 	ppd->lmc = lmc;
1172 	hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_LIDLMC, 0);
1173 
1174 	dd_dev_info(dd, "port %u: got a lid: 0x%x\n", ppd->port, lid);
1175 
1176 	return 0;
1177 }
1178 
1179 void shutdown_led_override(struct hfi1_pportdata *ppd)
1180 {
1181 	struct hfi1_devdata *dd = ppd->dd;
1182 
1183 	/*
1184 	 * This pairs with the memory barrier in hfi1_start_led_override to
1185 	 * ensure that we read the correct state of LED beaconing represented
1186 	 * by led_override_timer_active
1187 	 */
1188 	smp_rmb();
1189 	if (atomic_read(&ppd->led_override_timer_active)) {
1190 		del_timer_sync(&ppd->led_override_timer);
1191 		atomic_set(&ppd->led_override_timer_active, 0);
1192 		/* Ensure the atomic_set is visible to all CPUs */
1193 		smp_wmb();
1194 	}
1195 
1196 	/* Hand control of the LED to the DC for normal operation */
1197 	write_csr(dd, DCC_CFG_LED_CNTRL, 0);
1198 }
1199 
1200 static void run_led_override(unsigned long opaque)
1201 {
1202 	struct hfi1_pportdata *ppd = (struct hfi1_pportdata *)opaque;
1203 	struct hfi1_devdata *dd = ppd->dd;
1204 	unsigned long timeout;
1205 	int phase_idx;
1206 
1207 	if (!(dd->flags & HFI1_INITTED))
1208 		return;
1209 
1210 	phase_idx = ppd->led_override_phase & 1;
1211 
1212 	setextled(dd, phase_idx);
1213 
1214 	timeout = ppd->led_override_vals[phase_idx];
1215 
1216 	/* Set up for next phase */
1217 	ppd->led_override_phase = !ppd->led_override_phase;
1218 
1219 	mod_timer(&ppd->led_override_timer, jiffies + timeout);
1220 }
1221 
1222 /*
1223  * To have the LED blink in a particular pattern, provide timeon and timeoff
1224  * in milliseconds.
1225  * To turn off custom blinking and return to normal operation, use
1226  * shutdown_led_override()
1227  */
1228 void hfi1_start_led_override(struct hfi1_pportdata *ppd, unsigned int timeon,
1229 			     unsigned int timeoff)
1230 {
1231 	if (!(ppd->dd->flags & HFI1_INITTED))
1232 		return;
1233 
1234 	/* Convert to jiffies for direct use in timer */
1235 	ppd->led_override_vals[0] = msecs_to_jiffies(timeoff);
1236 	ppd->led_override_vals[1] = msecs_to_jiffies(timeon);
1237 
1238 	/* Arbitrarily start from LED on phase */
1239 	ppd->led_override_phase = 1;
1240 
1241 	/*
1242 	 * If the timer has not already been started, do so. Use a "quick"
1243 	 * timeout so the handler will be called soon to look at our request.
1244 	 */
1245 	if (!timer_pending(&ppd->led_override_timer)) {
1246 		setup_timer(&ppd->led_override_timer, run_led_override,
1247 			    (unsigned long)ppd);
1248 		ppd->led_override_timer.expires = jiffies + 1;
1249 		add_timer(&ppd->led_override_timer);
1250 		atomic_set(&ppd->led_override_timer_active, 1);
1251 		/* Ensure the atomic_set is visible to all CPUs */
1252 		smp_wmb();
1253 	}
1254 }
1255 
1256 /**
1257  * hfi1_reset_device - reset the chip if possible
1258  * @unit: the device to reset
1259  *
1260  * Whether or not reset is successful, we attempt to re-initialize the chip
1261  * (that is, much like a driver unload/reload).  We clear the INITTED flag
1262  * so that the various entry points will fail until we reinitialize.  For
1263  * now, we only allow this if no user contexts are open that use chip resources
1264  */
1265 int hfi1_reset_device(int unit)
1266 {
1267 	int ret, i;
1268 	struct hfi1_devdata *dd = hfi1_lookup(unit);
1269 	struct hfi1_pportdata *ppd;
1270 	unsigned long flags;
1271 	int pidx;
1272 
1273 	if (!dd) {
1274 		ret = -ENODEV;
1275 		goto bail;
1276 	}
1277 
1278 	dd_dev_info(dd, "Reset on unit %u requested\n", unit);
1279 
1280 	if (!dd->kregbase || !(dd->flags & HFI1_PRESENT)) {
1281 		dd_dev_info(dd,
1282 			    "Invalid unit number %u or not initialized or not present\n",
1283 			    unit);
1284 		ret = -ENXIO;
1285 		goto bail;
1286 	}
1287 
1288 	spin_lock_irqsave(&dd->uctxt_lock, flags);
1289 	if (dd->rcd)
1290 		for (i = dd->first_dyn_alloc_ctxt;
1291 		     i < dd->num_rcv_contexts; i++) {
1292 			if (!dd->rcd[i])
1293 				continue;
1294 			spin_unlock_irqrestore(&dd->uctxt_lock, flags);
1295 			ret = -EBUSY;
1296 			goto bail;
1297 		}
1298 	spin_unlock_irqrestore(&dd->uctxt_lock, flags);
1299 
1300 	for (pidx = 0; pidx < dd->num_pports; ++pidx) {
1301 		ppd = dd->pport + pidx;
1302 
1303 		shutdown_led_override(ppd);
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 
1329 	rcv_hdrerr(rcd, rcd->ppd, packet);
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);
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 	if (unlikely(hfi1_dbg_fault_packet(packet)))
1352 		return RHF_RCV_CONTINUE;
1353 
1354 	trace_hfi1_rcvhdr(packet->rcd->ppd->dd,
1355 			  packet->rcd->ctxt,
1356 			  rhf_err_flags(packet->rhf),
1357 			  RHF_RCV_TYPE_IB,
1358 			  packet->hlen,
1359 			  packet->tlen,
1360 			  packet->updegr,
1361 			  rhf_egr_index(packet->rhf));
1362 
1363 	if (unlikely(
1364 		 (hfi1_dbg_fault_suppress_err(&packet->rcd->dd->verbs_dev) &&
1365 		 (packet->rhf & RHF_DC_ERR))))
1366 		return RHF_RCV_CONTINUE;
1367 
1368 	if (unlikely(rhf_err_flags(packet->rhf))) {
1369 		handle_eflags(packet);
1370 		return RHF_RCV_CONTINUE;
1371 	}
1372 
1373 	hfi1_ib_rcv(packet);
1374 	return RHF_RCV_CONTINUE;
1375 }
1376 
1377 static inline bool hfi1_is_vnic_packet(struct hfi1_packet *packet)
1378 {
1379 	/* Packet received in VNIC context via RSM */
1380 	if (packet->rcd->is_vnic)
1381 		return true;
1382 
1383 	if ((HFI1_GET_L2_TYPE(packet->ebuf) == OPA_VNIC_L2_TYPE) &&
1384 	    (HFI1_GET_L4_TYPE(packet->ebuf) == OPA_VNIC_L4_ETHR))
1385 		return true;
1386 
1387 	return false;
1388 }
1389 
1390 int process_receive_bypass(struct hfi1_packet *packet)
1391 {
1392 	struct hfi1_devdata *dd = packet->rcd->dd;
1393 
1394 	if (unlikely(rhf_err_flags(packet->rhf))) {
1395 		handle_eflags(packet);
1396 	} else if (hfi1_is_vnic_packet(packet)) {
1397 		hfi1_vnic_bypass_rcv(packet);
1398 		return RHF_RCV_CONTINUE;
1399 	}
1400 
1401 	dd_dev_err(dd, "Unsupported bypass packet. Dropping\n");
1402 	incr_cntr64(&dd->sw_rcv_bypass_packet_errors);
1403 	if (!(dd->err_info_rcvport.status_and_code & OPA_EI_STATUS_SMASK)) {
1404 		u64 *flits = packet->ebuf;
1405 
1406 		if (flits && !(packet->rhf & RHF_LEN_ERR)) {
1407 			dd->err_info_rcvport.packet_flit1 = flits[0];
1408 			dd->err_info_rcvport.packet_flit2 =
1409 				packet->tlen > sizeof(flits[0]) ? flits[1] : 0;
1410 		}
1411 		dd->err_info_rcvport.status_and_code |=
1412 			(OPA_EI_STATUS_SMASK | BAD_L2_ERR);
1413 	}
1414 	return RHF_RCV_CONTINUE;
1415 }
1416 
1417 int process_receive_error(struct hfi1_packet *packet)
1418 {
1419 	/* KHdrHCRCErr -- KDETH packet with a bad HCRC */
1420 	if (unlikely(
1421 		 hfi1_dbg_fault_suppress_err(&packet->rcd->dd->verbs_dev) &&
1422 		 rhf_rcv_type_err(packet->rhf) == 3))
1423 		return RHF_RCV_CONTINUE;
1424 
1425 	handle_eflags(packet);
1426 
1427 	if (unlikely(rhf_err_flags(packet->rhf)))
1428 		dd_dev_err(packet->rcd->dd,
1429 			   "Unhandled error packet received. Dropping.\n");
1430 
1431 	return RHF_RCV_CONTINUE;
1432 }
1433 
1434 int kdeth_process_expected(struct hfi1_packet *packet)
1435 {
1436 	if (unlikely(hfi1_dbg_fault_packet(packet)))
1437 		return RHF_RCV_CONTINUE;
1438 	if (unlikely(rhf_err_flags(packet->rhf)))
1439 		handle_eflags(packet);
1440 
1441 	dd_dev_err(packet->rcd->dd,
1442 		   "Unhandled expected packet received. Dropping.\n");
1443 	return RHF_RCV_CONTINUE;
1444 }
1445 
1446 int kdeth_process_eager(struct hfi1_packet *packet)
1447 {
1448 	if (unlikely(rhf_err_flags(packet->rhf)))
1449 		handle_eflags(packet);
1450 	if (unlikely(hfi1_dbg_fault_packet(packet)))
1451 		return RHF_RCV_CONTINUE;
1452 
1453 	dd_dev_err(packet->rcd->dd,
1454 		   "Unhandled eager packet received. Dropping.\n");
1455 	return RHF_RCV_CONTINUE;
1456 }
1457 
1458 int process_receive_invalid(struct hfi1_packet *packet)
1459 {
1460 	dd_dev_err(packet->rcd->dd, "Invalid packet type %d. Dropping\n",
1461 		   rhf_rcv_type(packet->rhf));
1462 	return RHF_RCV_CONTINUE;
1463 }
1464