1 /*
2  * Copyright(c) 2015-2018 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 #include "fault.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 " __stringify(
82 		 HFI1_DEFAULT_MAX_MTU));
83 
84 unsigned int hfi1_cu = 1;
85 module_param_named(cu, hfi1_cu, uint, S_IRUGO);
86 MODULE_PARM_DESC(cu, "Credit return units");
87 
88 unsigned long hfi1_cap_mask = HFI1_CAP_MASK_DEFAULT;
89 static int hfi1_caps_set(const char *val, const struct kernel_param *kp);
90 static int hfi1_caps_get(char *buffer, const struct kernel_param *kp);
91 static const struct kernel_param_ops cap_ops = {
92 	.set = hfi1_caps_set,
93 	.get = hfi1_caps_get
94 };
95 module_param_cb(cap_mask, &cap_ops, &hfi1_cap_mask, S_IWUSR | S_IRUGO);
96 MODULE_PARM_DESC(cap_mask, "Bit mask of enabled/disabled HW features");
97 
98 MODULE_LICENSE("Dual BSD/GPL");
99 MODULE_DESCRIPTION("Intel Omni-Path Architecture driver");
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 struct pci_dev *get_pci_dev(struct rvt_dev_info *rdi)
164 {
165 	struct hfi1_ibdev *ibdev = container_of(rdi, struct hfi1_ibdev, rdi);
166 	struct hfi1_devdata *dd = container_of(ibdev,
167 					       struct hfi1_devdata, verbs_dev);
168 	return dd->pcidev;
169 }
170 
171 /*
172  * Return count of units with at least one port ACTIVE.
173  */
174 int hfi1_count_active_units(void)
175 {
176 	struct hfi1_devdata *dd;
177 	struct hfi1_pportdata *ppd;
178 	unsigned long flags;
179 	int pidx, nunits_active = 0;
180 
181 	spin_lock_irqsave(&hfi1_devs_lock, flags);
182 	list_for_each_entry(dd, &hfi1_dev_list, list) {
183 		if (!(dd->flags & HFI1_PRESENT) || !dd->kregbase1)
184 			continue;
185 		for (pidx = 0; pidx < dd->num_pports; ++pidx) {
186 			ppd = dd->pport + pidx;
187 			if (ppd->lid && ppd->linkup) {
188 				nunits_active++;
189 				break;
190 			}
191 		}
192 	}
193 	spin_unlock_irqrestore(&hfi1_devs_lock, flags);
194 	return nunits_active;
195 }
196 
197 /*
198  * Get address of eager buffer from it's index (allocated in chunks, not
199  * contiguous).
200  */
201 static inline void *get_egrbuf(const struct hfi1_ctxtdata *rcd, u64 rhf,
202 			       u8 *update)
203 {
204 	u32 idx = rhf_egr_index(rhf), offset = rhf_egr_buf_offset(rhf);
205 
206 	*update |= !(idx & (rcd->egrbufs.threshold - 1)) && !offset;
207 	return (void *)(((u64)(rcd->egrbufs.rcvtids[idx].addr)) +
208 			(offset * RCV_BUF_BLOCK_SIZE));
209 }
210 
211 static inline void *hfi1_get_header(struct hfi1_ctxtdata *rcd,
212 				    __le32 *rhf_addr)
213 {
214 	u32 offset = rhf_hdrq_offset(rhf_to_cpu(rhf_addr));
215 
216 	return (void *)(rhf_addr - rcd->rhf_offset + offset);
217 }
218 
219 static inline struct ib_header *hfi1_get_msgheader(struct hfi1_ctxtdata *rcd,
220 						   __le32 *rhf_addr)
221 {
222 	return (struct ib_header *)hfi1_get_header(rcd, rhf_addr);
223 }
224 
225 static inline struct hfi1_16b_header
226 		*hfi1_get_16B_header(struct hfi1_ctxtdata *rcd,
227 				     __le32 *rhf_addr)
228 {
229 	return (struct hfi1_16b_header *)hfi1_get_header(rcd, rhf_addr);
230 }
231 
232 /*
233  * Validate and encode the a given RcvArray Buffer size.
234  * The function will check whether the given size falls within
235  * allowed size ranges for the respective type and, optionally,
236  * return the proper encoding.
237  */
238 int hfi1_rcvbuf_validate(u32 size, u8 type, u16 *encoded)
239 {
240 	if (unlikely(!PAGE_ALIGNED(size)))
241 		return 0;
242 	if (unlikely(size < MIN_EAGER_BUFFER))
243 		return 0;
244 	if (size >
245 	    (type == PT_EAGER ? MAX_EAGER_BUFFER : MAX_EXPECTED_BUFFER))
246 		return 0;
247 	if (encoded)
248 		*encoded = ilog2(size / PAGE_SIZE) + 1;
249 	return 1;
250 }
251 
252 static void rcv_hdrerr(struct hfi1_ctxtdata *rcd, struct hfi1_pportdata *ppd,
253 		       struct hfi1_packet *packet)
254 {
255 	struct ib_header *rhdr = packet->hdr;
256 	u32 rte = rhf_rcv_type_err(packet->rhf);
257 	u32 mlid_base;
258 	struct hfi1_ibport *ibp = rcd_to_iport(rcd);
259 	struct hfi1_devdata *dd = ppd->dd;
260 	struct hfi1_ibdev *verbs_dev = &dd->verbs_dev;
261 	struct rvt_dev_info *rdi = &verbs_dev->rdi;
262 
263 	if ((packet->rhf & RHF_DC_ERR) &&
264 	    hfi1_dbg_fault_suppress_err(verbs_dev))
265 		return;
266 
267 	if (packet->rhf & (RHF_VCRC_ERR | RHF_ICRC_ERR))
268 		return;
269 
270 	if (packet->etype == RHF_RCV_TYPE_BYPASS) {
271 		goto drop;
272 	} else {
273 		u8 lnh = ib_get_lnh(rhdr);
274 
275 		mlid_base = be16_to_cpu(IB_MULTICAST_LID_BASE);
276 		if (lnh == HFI1_LRH_BTH) {
277 			packet->ohdr = &rhdr->u.oth;
278 		} else if (lnh == HFI1_LRH_GRH) {
279 			packet->ohdr = &rhdr->u.l.oth;
280 			packet->grh = &rhdr->u.l.grh;
281 		} else {
282 			goto drop;
283 		}
284 	}
285 
286 	if (packet->rhf & RHF_TID_ERR) {
287 		/* For TIDERR and RC QPs preemptively schedule a NAK */
288 		u32 tlen = rhf_pkt_len(packet->rhf); /* in bytes */
289 		u32 dlid = ib_get_dlid(rhdr);
290 		u32 qp_num;
291 
292 		/* Sanity check packet */
293 		if (tlen < 24)
294 			goto drop;
295 
296 		/* Check for GRH */
297 		if (packet->grh) {
298 			u32 vtf;
299 			struct ib_grh *grh = packet->grh;
300 
301 			if (grh->next_hdr != IB_GRH_NEXT_HDR)
302 				goto drop;
303 			vtf = be32_to_cpu(grh->version_tclass_flow);
304 			if ((vtf >> IB_GRH_VERSION_SHIFT) != IB_GRH_VERSION)
305 				goto drop;
306 		}
307 
308 		/* Get the destination QP number. */
309 		qp_num = ib_bth_get_qpn(packet->ohdr);
310 		if (dlid < mlid_base) {
311 			struct rvt_qp *qp;
312 			unsigned long flags;
313 
314 			rcu_read_lock();
315 			qp = rvt_lookup_qpn(rdi, &ibp->rvp, qp_num);
316 			if (!qp) {
317 				rcu_read_unlock();
318 				goto drop;
319 			}
320 
321 			/*
322 			 * Handle only RC QPs - for other QP types drop error
323 			 * packet.
324 			 */
325 			spin_lock_irqsave(&qp->r_lock, flags);
326 
327 			/* Check for valid receive state. */
328 			if (!(ib_rvt_state_ops[qp->state] &
329 			      RVT_PROCESS_RECV_OK)) {
330 				ibp->rvp.n_pkt_drops++;
331 			}
332 
333 			switch (qp->ibqp.qp_type) {
334 			case IB_QPT_RC:
335 				hfi1_rc_hdrerr(rcd, packet, qp);
336 				break;
337 			default:
338 				/* For now don't handle any other QP types */
339 				break;
340 			}
341 
342 			spin_unlock_irqrestore(&qp->r_lock, flags);
343 			rcu_read_unlock();
344 		} /* Unicast QP */
345 	} /* Valid packet with TIDErr */
346 
347 	/* handle "RcvTypeErr" flags */
348 	switch (rte) {
349 	case RHF_RTE_ERROR_OP_CODE_ERR:
350 	{
351 		void *ebuf = NULL;
352 		u8 opcode;
353 
354 		if (rhf_use_egr_bfr(packet->rhf))
355 			ebuf = packet->ebuf;
356 
357 		if (!ebuf)
358 			goto drop; /* this should never happen */
359 
360 		opcode = ib_bth_get_opcode(packet->ohdr);
361 		if (opcode == IB_OPCODE_CNP) {
362 			/*
363 			 * Only in pre-B0 h/w is the CNP_OPCODE handled
364 			 * via this code path.
365 			 */
366 			struct rvt_qp *qp = NULL;
367 			u32 lqpn, rqpn;
368 			u16 rlid;
369 			u8 svc_type, sl, sc5;
370 
371 			sc5 = hfi1_9B_get_sc5(rhdr, packet->rhf);
372 			sl = ibp->sc_to_sl[sc5];
373 
374 			lqpn = ib_bth_get_qpn(packet->ohdr);
375 			rcu_read_lock();
376 			qp = rvt_lookup_qpn(rdi, &ibp->rvp, lqpn);
377 			if (!qp) {
378 				rcu_read_unlock();
379 				goto drop;
380 			}
381 
382 			switch (qp->ibqp.qp_type) {
383 			case IB_QPT_UD:
384 				rlid = 0;
385 				rqpn = 0;
386 				svc_type = IB_CC_SVCTYPE_UD;
387 				break;
388 			case IB_QPT_UC:
389 				rlid = ib_get_slid(rhdr);
390 				rqpn = qp->remote_qpn;
391 				svc_type = IB_CC_SVCTYPE_UC;
392 				break;
393 			default:
394 				rcu_read_unlock();
395 				goto drop;
396 			}
397 
398 			process_becn(ppd, sl, rlid, lqpn, rqpn, svc_type);
399 			rcu_read_unlock();
400 		}
401 
402 		packet->rhf &= ~RHF_RCV_TYPE_ERR_SMASK;
403 		break;
404 	}
405 	default:
406 		break;
407 	}
408 
409 drop:
410 	return;
411 }
412 
413 static inline void init_packet(struct hfi1_ctxtdata *rcd,
414 			       struct hfi1_packet *packet)
415 {
416 	packet->rsize = rcd->rcvhdrqentsize; /* words */
417 	packet->maxcnt = rcd->rcvhdrq_cnt * packet->rsize; /* words */
418 	packet->rcd = rcd;
419 	packet->updegr = 0;
420 	packet->etail = -1;
421 	packet->rhf_addr = get_rhf_addr(rcd);
422 	packet->rhf = rhf_to_cpu(packet->rhf_addr);
423 	packet->rhqoff = rcd->head;
424 	packet->numpkt = 0;
425 }
426 
427 /* We support only two types - 9B and 16B for now */
428 static const hfi1_handle_cnp hfi1_handle_cnp_tbl[2] = {
429 	[HFI1_PKT_TYPE_9B] = &return_cnp,
430 	[HFI1_PKT_TYPE_16B] = &return_cnp_16B
431 };
432 
433 /**
434  * hfi1_process_ecn_slowpath - Process FECN or BECN bits
435  * @qp: The packet's destination QP
436  * @pkt: The packet itself.
437  * @prescan: Is the caller the RXQ prescan
438  *
439  * Process the packet's FECN or BECN bits. By now, the packet
440  * has already been evaluated whether processing of those bit should
441  * be done.
442  * The significance of the @prescan argument is that if the caller
443  * is the RXQ prescan, a CNP will be send out instead of waiting for the
444  * normal packet processing to send an ACK with BECN set (or a CNP).
445  */
446 bool hfi1_process_ecn_slowpath(struct rvt_qp *qp, struct hfi1_packet *pkt,
447 			       bool prescan)
448 {
449 	struct hfi1_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num);
450 	struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
451 	struct ib_other_headers *ohdr = pkt->ohdr;
452 	struct ib_grh *grh = pkt->grh;
453 	u32 rqpn = 0;
454 	u16 pkey;
455 	u32 rlid, slid, dlid = 0;
456 	u8 hdr_type, sc, svc_type, opcode;
457 	bool is_mcast = false, ignore_fecn = false, do_cnp = false,
458 		fecn, becn;
459 
460 	/* can be called from prescan */
461 	if (pkt->etype == RHF_RCV_TYPE_BYPASS) {
462 		pkey = hfi1_16B_get_pkey(pkt->hdr);
463 		sc = hfi1_16B_get_sc(pkt->hdr);
464 		dlid = hfi1_16B_get_dlid(pkt->hdr);
465 		slid = hfi1_16B_get_slid(pkt->hdr);
466 		is_mcast = hfi1_is_16B_mcast(dlid);
467 		opcode = ib_bth_get_opcode(ohdr);
468 		hdr_type = HFI1_PKT_TYPE_16B;
469 		fecn = hfi1_16B_get_fecn(pkt->hdr);
470 		becn = hfi1_16B_get_becn(pkt->hdr);
471 	} else {
472 		pkey = ib_bth_get_pkey(ohdr);
473 		sc = hfi1_9B_get_sc5(pkt->hdr, pkt->rhf);
474 		dlid = qp->ibqp.qp_type != IB_QPT_UD ? ib_get_dlid(pkt->hdr) :
475 			ppd->lid;
476 		slid = ib_get_slid(pkt->hdr);
477 		is_mcast = (dlid > be16_to_cpu(IB_MULTICAST_LID_BASE)) &&
478 			   (dlid != be16_to_cpu(IB_LID_PERMISSIVE));
479 		opcode = ib_bth_get_opcode(ohdr);
480 		hdr_type = HFI1_PKT_TYPE_9B;
481 		fecn = ib_bth_get_fecn(ohdr);
482 		becn = ib_bth_get_becn(ohdr);
483 	}
484 
485 	switch (qp->ibqp.qp_type) {
486 	case IB_QPT_UD:
487 		rlid = slid;
488 		rqpn = ib_get_sqpn(pkt->ohdr);
489 		svc_type = IB_CC_SVCTYPE_UD;
490 		break;
491 	case IB_QPT_SMI:
492 	case IB_QPT_GSI:
493 		rlid = slid;
494 		rqpn = ib_get_sqpn(pkt->ohdr);
495 		svc_type = IB_CC_SVCTYPE_UD;
496 		break;
497 	case IB_QPT_UC:
498 		rlid = rdma_ah_get_dlid(&qp->remote_ah_attr);
499 		rqpn = qp->remote_qpn;
500 		svc_type = IB_CC_SVCTYPE_UC;
501 		break;
502 	case IB_QPT_RC:
503 		rlid = rdma_ah_get_dlid(&qp->remote_ah_attr);
504 		rqpn = qp->remote_qpn;
505 		svc_type = IB_CC_SVCTYPE_RC;
506 		break;
507 	default:
508 		return false;
509 	}
510 
511 	ignore_fecn = is_mcast || (opcode == IB_OPCODE_CNP) ||
512 		(opcode == IB_OPCODE_RC_ACKNOWLEDGE);
513 	/*
514 	 * ACKNOWLEDGE packets do not get a CNP but this will be
515 	 * guarded by ignore_fecn above.
516 	 */
517 	do_cnp = prescan ||
518 		(opcode >= IB_OPCODE_RC_RDMA_READ_RESPONSE_FIRST &&
519 		 opcode <= IB_OPCODE_RC_ATOMIC_ACKNOWLEDGE);
520 
521 	/* Call appropriate CNP handler */
522 	if (!ignore_fecn && do_cnp && fecn)
523 		hfi1_handle_cnp_tbl[hdr_type](ibp, qp, rqpn, pkey,
524 					      dlid, rlid, sc, grh);
525 
526 	if (becn) {
527 		u32 lqpn = be32_to_cpu(ohdr->bth[1]) & RVT_QPN_MASK;
528 		u8 sl = ibp->sc_to_sl[sc];
529 
530 		process_becn(ppd, sl, rlid, lqpn, rqpn, svc_type);
531 	}
532 	return !ignore_fecn && fecn;
533 }
534 
535 struct ps_mdata {
536 	struct hfi1_ctxtdata *rcd;
537 	u32 rsize;
538 	u32 maxcnt;
539 	u32 ps_head;
540 	u32 ps_tail;
541 	u32 ps_seq;
542 };
543 
544 static inline void init_ps_mdata(struct ps_mdata *mdata,
545 				 struct hfi1_packet *packet)
546 {
547 	struct hfi1_ctxtdata *rcd = packet->rcd;
548 
549 	mdata->rcd = rcd;
550 	mdata->rsize = packet->rsize;
551 	mdata->maxcnt = packet->maxcnt;
552 	mdata->ps_head = packet->rhqoff;
553 
554 	if (HFI1_CAP_KGET_MASK(rcd->flags, DMA_RTAIL)) {
555 		mdata->ps_tail = get_rcvhdrtail(rcd);
556 		if (rcd->ctxt == HFI1_CTRL_CTXT)
557 			mdata->ps_seq = rcd->seq_cnt;
558 		else
559 			mdata->ps_seq = 0; /* not used with DMA_RTAIL */
560 	} else {
561 		mdata->ps_tail = 0; /* used only with DMA_RTAIL*/
562 		mdata->ps_seq = rcd->seq_cnt;
563 	}
564 }
565 
566 static inline int ps_done(struct ps_mdata *mdata, u64 rhf,
567 			  struct hfi1_ctxtdata *rcd)
568 {
569 	if (HFI1_CAP_KGET_MASK(rcd->flags, DMA_RTAIL))
570 		return mdata->ps_head == mdata->ps_tail;
571 	return mdata->ps_seq != rhf_rcv_seq(rhf);
572 }
573 
574 static inline int ps_skip(struct ps_mdata *mdata, u64 rhf,
575 			  struct hfi1_ctxtdata *rcd)
576 {
577 	/*
578 	 * Control context can potentially receive an invalid rhf.
579 	 * Drop such packets.
580 	 */
581 	if ((rcd->ctxt == HFI1_CTRL_CTXT) && (mdata->ps_head != mdata->ps_tail))
582 		return mdata->ps_seq != rhf_rcv_seq(rhf);
583 
584 	return 0;
585 }
586 
587 static inline void update_ps_mdata(struct ps_mdata *mdata,
588 				   struct hfi1_ctxtdata *rcd)
589 {
590 	mdata->ps_head += mdata->rsize;
591 	if (mdata->ps_head >= mdata->maxcnt)
592 		mdata->ps_head = 0;
593 
594 	/* Control context must do seq counting */
595 	if (!HFI1_CAP_KGET_MASK(rcd->flags, DMA_RTAIL) ||
596 	    (rcd->ctxt == HFI1_CTRL_CTXT)) {
597 		if (++mdata->ps_seq > 13)
598 			mdata->ps_seq = 1;
599 	}
600 }
601 
602 /*
603  * prescan_rxq - search through the receive queue looking for packets
604  * containing Excplicit Congestion Notifications (FECNs, or BECNs).
605  * When an ECN is found, process the Congestion Notification, and toggle
606  * it off.
607  * This is declared as a macro to allow quick checking of the port to avoid
608  * the overhead of a function call if not enabled.
609  */
610 #define prescan_rxq(rcd, packet) \
611 	do { \
612 		if (rcd->ppd->cc_prescan) \
613 			__prescan_rxq(packet); \
614 	} while (0)
615 static void __prescan_rxq(struct hfi1_packet *packet)
616 {
617 	struct hfi1_ctxtdata *rcd = packet->rcd;
618 	struct ps_mdata mdata;
619 
620 	init_ps_mdata(&mdata, packet);
621 
622 	while (1) {
623 		struct hfi1_ibport *ibp = rcd_to_iport(rcd);
624 		__le32 *rhf_addr = (__le32 *)rcd->rcvhdrq + mdata.ps_head +
625 					 packet->rcd->rhf_offset;
626 		struct rvt_qp *qp;
627 		struct ib_header *hdr;
628 		struct rvt_dev_info *rdi = &rcd->dd->verbs_dev.rdi;
629 		u64 rhf = rhf_to_cpu(rhf_addr);
630 		u32 etype = rhf_rcv_type(rhf), qpn, bth1;
631 		u8 lnh;
632 
633 		if (ps_done(&mdata, rhf, rcd))
634 			break;
635 
636 		if (ps_skip(&mdata, rhf, rcd))
637 			goto next;
638 
639 		if (etype != RHF_RCV_TYPE_IB)
640 			goto next;
641 
642 		packet->hdr = hfi1_get_msgheader(packet->rcd, rhf_addr);
643 		hdr = packet->hdr;
644 		lnh = ib_get_lnh(hdr);
645 
646 		if (lnh == HFI1_LRH_BTH) {
647 			packet->ohdr = &hdr->u.oth;
648 			packet->grh = NULL;
649 		} else if (lnh == HFI1_LRH_GRH) {
650 			packet->ohdr = &hdr->u.l.oth;
651 			packet->grh = &hdr->u.l.grh;
652 		} else {
653 			goto next; /* just in case */
654 		}
655 
656 		if (!hfi1_may_ecn(packet))
657 			goto next;
658 
659 		bth1 = be32_to_cpu(packet->ohdr->bth[1]);
660 		qpn = bth1 & RVT_QPN_MASK;
661 		rcu_read_lock();
662 		qp = rvt_lookup_qpn(rdi, &ibp->rvp, qpn);
663 
664 		if (!qp) {
665 			rcu_read_unlock();
666 			goto next;
667 		}
668 
669 		hfi1_process_ecn_slowpath(qp, packet, true);
670 		rcu_read_unlock();
671 
672 		/* turn off BECN, FECN */
673 		bth1 &= ~(IB_FECN_SMASK | IB_BECN_SMASK);
674 		packet->ohdr->bth[1] = cpu_to_be32(bth1);
675 next:
676 		update_ps_mdata(&mdata, rcd);
677 	}
678 }
679 
680 static void process_rcv_qp_work(struct hfi1_packet *packet)
681 {
682 	struct rvt_qp *qp, *nqp;
683 	struct hfi1_ctxtdata *rcd = packet->rcd;
684 
685 	/*
686 	 * Iterate over all QPs waiting to respond.
687 	 * The list won't change since the IRQ is only run on one CPU.
688 	 */
689 	list_for_each_entry_safe(qp, nqp, &rcd->qp_wait_list, rspwait) {
690 		list_del_init(&qp->rspwait);
691 		if (qp->r_flags & RVT_R_RSP_NAK) {
692 			qp->r_flags &= ~RVT_R_RSP_NAK;
693 			packet->qp = qp;
694 			hfi1_send_rc_ack(packet, 0);
695 		}
696 		if (qp->r_flags & RVT_R_RSP_SEND) {
697 			unsigned long flags;
698 
699 			qp->r_flags &= ~RVT_R_RSP_SEND;
700 			spin_lock_irqsave(&qp->s_lock, flags);
701 			if (ib_rvt_state_ops[qp->state] &
702 					RVT_PROCESS_OR_FLUSH_SEND)
703 				hfi1_schedule_send(qp);
704 			spin_unlock_irqrestore(&qp->s_lock, flags);
705 		}
706 		rvt_put_qp(qp);
707 	}
708 }
709 
710 static noinline int max_packet_exceeded(struct hfi1_packet *packet, int thread)
711 {
712 	if (thread) {
713 		if ((packet->numpkt & (MAX_PKT_RECV_THREAD - 1)) == 0)
714 			/* allow defered processing */
715 			process_rcv_qp_work(packet);
716 		cond_resched();
717 		return RCV_PKT_OK;
718 	} else {
719 		this_cpu_inc(*packet->rcd->dd->rcv_limit);
720 		return RCV_PKT_LIMIT;
721 	}
722 }
723 
724 static inline int check_max_packet(struct hfi1_packet *packet, int thread)
725 {
726 	int ret = RCV_PKT_OK;
727 
728 	if (unlikely((packet->numpkt & (MAX_PKT_RECV - 1)) == 0))
729 		ret = max_packet_exceeded(packet, thread);
730 	return ret;
731 }
732 
733 static noinline int skip_rcv_packet(struct hfi1_packet *packet, int thread)
734 {
735 	int ret;
736 
737 	/* Set up for the next packet */
738 	packet->rhqoff += packet->rsize;
739 	if (packet->rhqoff >= packet->maxcnt)
740 		packet->rhqoff = 0;
741 
742 	packet->numpkt++;
743 	ret = check_max_packet(packet, thread);
744 
745 	packet->rhf_addr = (__le32 *)packet->rcd->rcvhdrq + packet->rhqoff +
746 				     packet->rcd->rhf_offset;
747 	packet->rhf = rhf_to_cpu(packet->rhf_addr);
748 
749 	return ret;
750 }
751 
752 static inline int process_rcv_packet(struct hfi1_packet *packet, int thread)
753 {
754 	int ret;
755 
756 	packet->etype = rhf_rcv_type(packet->rhf);
757 
758 	/* total length */
759 	packet->tlen = rhf_pkt_len(packet->rhf); /* in bytes */
760 	/* retrieve eager buffer details */
761 	packet->ebuf = NULL;
762 	if (rhf_use_egr_bfr(packet->rhf)) {
763 		packet->etail = rhf_egr_index(packet->rhf);
764 		packet->ebuf = get_egrbuf(packet->rcd, packet->rhf,
765 				 &packet->updegr);
766 		/*
767 		 * Prefetch the contents of the eager buffer.  It is
768 		 * OK to send a negative length to prefetch_range().
769 		 * The +2 is the size of the RHF.
770 		 */
771 		prefetch_range(packet->ebuf,
772 			       packet->tlen - ((packet->rcd->rcvhdrqentsize -
773 					       (rhf_hdrq_offset(packet->rhf)
774 						+ 2)) * 4));
775 	}
776 
777 	/*
778 	 * Call a type specific handler for the packet. We
779 	 * should be able to trust that etype won't be beyond
780 	 * the range of valid indexes. If so something is really
781 	 * wrong and we can probably just let things come
782 	 * crashing down. There is no need to eat another
783 	 * comparison in this performance critical code.
784 	 */
785 	packet->rcd->rhf_rcv_function_map[packet->etype](packet);
786 	packet->numpkt++;
787 
788 	/* Set up for the next packet */
789 	packet->rhqoff += packet->rsize;
790 	if (packet->rhqoff >= packet->maxcnt)
791 		packet->rhqoff = 0;
792 
793 	ret = check_max_packet(packet, thread);
794 
795 	packet->rhf_addr = (__le32 *)packet->rcd->rcvhdrq + packet->rhqoff +
796 				      packet->rcd->rhf_offset;
797 	packet->rhf = rhf_to_cpu(packet->rhf_addr);
798 
799 	return ret;
800 }
801 
802 static inline void process_rcv_update(int last, struct hfi1_packet *packet)
803 {
804 	/*
805 	 * Update head regs etc., every 16 packets, if not last pkt,
806 	 * to help prevent rcvhdrq overflows, when many packets
807 	 * are processed and queue is nearly full.
808 	 * Don't request an interrupt for intermediate updates.
809 	 */
810 	if (!last && !(packet->numpkt & 0xf)) {
811 		update_usrhead(packet->rcd, packet->rhqoff, packet->updegr,
812 			       packet->etail, 0, 0);
813 		packet->updegr = 0;
814 	}
815 	packet->grh = NULL;
816 }
817 
818 static inline void finish_packet(struct hfi1_packet *packet)
819 {
820 	/*
821 	 * Nothing we need to free for the packet.
822 	 *
823 	 * The only thing we need to do is a final update and call for an
824 	 * interrupt
825 	 */
826 	update_usrhead(packet->rcd, packet->rcd->head, packet->updegr,
827 		       packet->etail, rcv_intr_dynamic, packet->numpkt);
828 }
829 
830 /*
831  * Handle receive interrupts when using the no dma rtail option.
832  */
833 int handle_receive_interrupt_nodma_rtail(struct hfi1_ctxtdata *rcd, int thread)
834 {
835 	u32 seq;
836 	int last = RCV_PKT_OK;
837 	struct hfi1_packet packet;
838 
839 	init_packet(rcd, &packet);
840 	seq = rhf_rcv_seq(packet.rhf);
841 	if (seq != rcd->seq_cnt) {
842 		last = RCV_PKT_DONE;
843 		goto bail;
844 	}
845 
846 	prescan_rxq(rcd, &packet);
847 
848 	while (last == RCV_PKT_OK) {
849 		last = process_rcv_packet(&packet, thread);
850 		seq = rhf_rcv_seq(packet.rhf);
851 		if (++rcd->seq_cnt > 13)
852 			rcd->seq_cnt = 1;
853 		if (seq != rcd->seq_cnt)
854 			last = RCV_PKT_DONE;
855 		process_rcv_update(last, &packet);
856 	}
857 	process_rcv_qp_work(&packet);
858 	rcd->head = packet.rhqoff;
859 bail:
860 	finish_packet(&packet);
861 	return last;
862 }
863 
864 int handle_receive_interrupt_dma_rtail(struct hfi1_ctxtdata *rcd, int thread)
865 {
866 	u32 hdrqtail;
867 	int last = RCV_PKT_OK;
868 	struct hfi1_packet packet;
869 
870 	init_packet(rcd, &packet);
871 	hdrqtail = get_rcvhdrtail(rcd);
872 	if (packet.rhqoff == hdrqtail) {
873 		last = RCV_PKT_DONE;
874 		goto bail;
875 	}
876 	smp_rmb();  /* prevent speculative reads of dma'ed hdrq */
877 
878 	prescan_rxq(rcd, &packet);
879 
880 	while (last == RCV_PKT_OK) {
881 		last = process_rcv_packet(&packet, thread);
882 		if (packet.rhqoff == hdrqtail)
883 			last = RCV_PKT_DONE;
884 		process_rcv_update(last, &packet);
885 	}
886 	process_rcv_qp_work(&packet);
887 	rcd->head = packet.rhqoff;
888 bail:
889 	finish_packet(&packet);
890 	return last;
891 }
892 
893 static inline void set_nodma_rtail(struct hfi1_devdata *dd, u16 ctxt)
894 {
895 	struct hfi1_ctxtdata *rcd;
896 	u16 i;
897 
898 	/*
899 	 * For dynamically allocated kernel contexts (like vnic) switch
900 	 * interrupt handler only for that context. Otherwise, switch
901 	 * interrupt handler for all statically allocated kernel contexts.
902 	 */
903 	if (ctxt >= dd->first_dyn_alloc_ctxt) {
904 		rcd = hfi1_rcd_get_by_index_safe(dd, ctxt);
905 		if (rcd) {
906 			rcd->do_interrupt =
907 				&handle_receive_interrupt_nodma_rtail;
908 			hfi1_rcd_put(rcd);
909 		}
910 		return;
911 	}
912 
913 	for (i = HFI1_CTRL_CTXT + 1; i < dd->first_dyn_alloc_ctxt; i++) {
914 		rcd = hfi1_rcd_get_by_index(dd, i);
915 		if (rcd)
916 			rcd->do_interrupt =
917 				&handle_receive_interrupt_nodma_rtail;
918 		hfi1_rcd_put(rcd);
919 	}
920 }
921 
922 static inline void set_dma_rtail(struct hfi1_devdata *dd, u16 ctxt)
923 {
924 	struct hfi1_ctxtdata *rcd;
925 	u16 i;
926 
927 	/*
928 	 * For dynamically allocated kernel contexts (like vnic) switch
929 	 * interrupt handler only for that context. Otherwise, switch
930 	 * interrupt handler for all statically allocated kernel contexts.
931 	 */
932 	if (ctxt >= dd->first_dyn_alloc_ctxt) {
933 		rcd = hfi1_rcd_get_by_index_safe(dd, ctxt);
934 		if (rcd) {
935 			rcd->do_interrupt =
936 				&handle_receive_interrupt_dma_rtail;
937 			hfi1_rcd_put(rcd);
938 		}
939 		return;
940 	}
941 
942 	for (i = HFI1_CTRL_CTXT + 1; i < dd->first_dyn_alloc_ctxt; i++) {
943 		rcd = hfi1_rcd_get_by_index(dd, i);
944 		if (rcd)
945 			rcd->do_interrupt =
946 				&handle_receive_interrupt_dma_rtail;
947 		hfi1_rcd_put(rcd);
948 	}
949 }
950 
951 void set_all_slowpath(struct hfi1_devdata *dd)
952 {
953 	struct hfi1_ctxtdata *rcd;
954 	u16 i;
955 
956 	/* HFI1_CTRL_CTXT must always use the slow path interrupt handler */
957 	for (i = HFI1_CTRL_CTXT + 1; i < dd->num_rcv_contexts; i++) {
958 		rcd = hfi1_rcd_get_by_index(dd, i);
959 		if (!rcd)
960 			continue;
961 		if (i < dd->first_dyn_alloc_ctxt || rcd->is_vnic)
962 			rcd->do_interrupt = &handle_receive_interrupt;
963 
964 		hfi1_rcd_put(rcd);
965 	}
966 }
967 
968 static inline int set_armed_to_active(struct hfi1_ctxtdata *rcd,
969 				      struct hfi1_packet *packet,
970 				      struct hfi1_devdata *dd)
971 {
972 	struct work_struct *lsaw = &rcd->ppd->linkstate_active_work;
973 	u8 etype = rhf_rcv_type(packet->rhf);
974 	u8 sc = SC15_PACKET;
975 
976 	if (etype == RHF_RCV_TYPE_IB) {
977 		struct ib_header *hdr = hfi1_get_msgheader(packet->rcd,
978 							   packet->rhf_addr);
979 		sc = hfi1_9B_get_sc5(hdr, packet->rhf);
980 	} else if (etype == RHF_RCV_TYPE_BYPASS) {
981 		struct hfi1_16b_header *hdr = hfi1_get_16B_header(
982 						packet->rcd,
983 						packet->rhf_addr);
984 		sc = hfi1_16B_get_sc(hdr);
985 	}
986 	if (sc != SC15_PACKET) {
987 		int hwstate = driver_lstate(rcd->ppd);
988 
989 		if (hwstate != IB_PORT_ACTIVE) {
990 			dd_dev_info(dd,
991 				    "Unexpected link state %s\n",
992 				    opa_lstate_name(hwstate));
993 			return 0;
994 		}
995 
996 		queue_work(rcd->ppd->link_wq, lsaw);
997 		return 1;
998 	}
999 	return 0;
1000 }
1001 
1002 /*
1003  * handle_receive_interrupt - receive a packet
1004  * @rcd: the context
1005  *
1006  * Called from interrupt handler for errors or receive interrupt.
1007  * This is the slow path interrupt handler.
1008  */
1009 int handle_receive_interrupt(struct hfi1_ctxtdata *rcd, int thread)
1010 {
1011 	struct hfi1_devdata *dd = rcd->dd;
1012 	u32 hdrqtail;
1013 	int needset, last = RCV_PKT_OK;
1014 	struct hfi1_packet packet;
1015 	int skip_pkt = 0;
1016 
1017 	/* Control context will always use the slow path interrupt handler */
1018 	needset = (rcd->ctxt == HFI1_CTRL_CTXT) ? 0 : 1;
1019 
1020 	init_packet(rcd, &packet);
1021 
1022 	if (!HFI1_CAP_KGET_MASK(rcd->flags, DMA_RTAIL)) {
1023 		u32 seq = rhf_rcv_seq(packet.rhf);
1024 
1025 		if (seq != rcd->seq_cnt) {
1026 			last = RCV_PKT_DONE;
1027 			goto bail;
1028 		}
1029 		hdrqtail = 0;
1030 	} else {
1031 		hdrqtail = get_rcvhdrtail(rcd);
1032 		if (packet.rhqoff == hdrqtail) {
1033 			last = RCV_PKT_DONE;
1034 			goto bail;
1035 		}
1036 		smp_rmb();  /* prevent speculative reads of dma'ed hdrq */
1037 
1038 		/*
1039 		 * Control context can potentially receive an invalid
1040 		 * rhf. Drop such packets.
1041 		 */
1042 		if (rcd->ctxt == HFI1_CTRL_CTXT) {
1043 			u32 seq = rhf_rcv_seq(packet.rhf);
1044 
1045 			if (seq != rcd->seq_cnt)
1046 				skip_pkt = 1;
1047 		}
1048 	}
1049 
1050 	prescan_rxq(rcd, &packet);
1051 
1052 	while (last == RCV_PKT_OK) {
1053 		if (unlikely(dd->do_drop &&
1054 			     atomic_xchg(&dd->drop_packet, DROP_PACKET_OFF) ==
1055 			     DROP_PACKET_ON)) {
1056 			dd->do_drop = 0;
1057 
1058 			/* On to the next packet */
1059 			packet.rhqoff += packet.rsize;
1060 			packet.rhf_addr = (__le32 *)rcd->rcvhdrq +
1061 					  packet.rhqoff +
1062 					  rcd->rhf_offset;
1063 			packet.rhf = rhf_to_cpu(packet.rhf_addr);
1064 
1065 		} else if (skip_pkt) {
1066 			last = skip_rcv_packet(&packet, thread);
1067 			skip_pkt = 0;
1068 		} else {
1069 			/* Auto activate link on non-SC15 packet receive */
1070 			if (unlikely(rcd->ppd->host_link_state ==
1071 				     HLS_UP_ARMED) &&
1072 			    set_armed_to_active(rcd, &packet, dd))
1073 				goto bail;
1074 			last = process_rcv_packet(&packet, thread);
1075 		}
1076 
1077 		if (!HFI1_CAP_KGET_MASK(rcd->flags, DMA_RTAIL)) {
1078 			u32 seq = rhf_rcv_seq(packet.rhf);
1079 
1080 			if (++rcd->seq_cnt > 13)
1081 				rcd->seq_cnt = 1;
1082 			if (seq != rcd->seq_cnt)
1083 				last = RCV_PKT_DONE;
1084 			if (needset) {
1085 				dd_dev_info(dd, "Switching to NO_DMA_RTAIL\n");
1086 				set_nodma_rtail(dd, rcd->ctxt);
1087 				needset = 0;
1088 			}
1089 		} else {
1090 			if (packet.rhqoff == hdrqtail)
1091 				last = RCV_PKT_DONE;
1092 			/*
1093 			 * Control context can potentially receive an invalid
1094 			 * rhf. Drop such packets.
1095 			 */
1096 			if (rcd->ctxt == HFI1_CTRL_CTXT) {
1097 				u32 seq = rhf_rcv_seq(packet.rhf);
1098 
1099 				if (++rcd->seq_cnt > 13)
1100 					rcd->seq_cnt = 1;
1101 				if (!last && (seq != rcd->seq_cnt))
1102 					skip_pkt = 1;
1103 			}
1104 
1105 			if (needset) {
1106 				dd_dev_info(dd,
1107 					    "Switching to DMA_RTAIL\n");
1108 				set_dma_rtail(dd, rcd->ctxt);
1109 				needset = 0;
1110 			}
1111 		}
1112 
1113 		process_rcv_update(last, &packet);
1114 	}
1115 
1116 	process_rcv_qp_work(&packet);
1117 	rcd->head = packet.rhqoff;
1118 
1119 bail:
1120 	/*
1121 	 * Always write head at end, and setup rcv interrupt, even
1122 	 * if no packets were processed.
1123 	 */
1124 	finish_packet(&packet);
1125 	return last;
1126 }
1127 
1128 /*
1129  * We may discover in the interrupt that the hardware link state has
1130  * changed from ARMED to ACTIVE (due to the arrival of a non-SC15 packet),
1131  * and we need to update the driver's notion of the link state.  We cannot
1132  * run set_link_state from interrupt context, so we queue this function on
1133  * a workqueue.
1134  *
1135  * We delay the regular interrupt processing until after the state changes
1136  * so that the link will be in the correct state by the time any application
1137  * we wake up attempts to send a reply to any message it received.
1138  * (Subsequent receive interrupts may possibly force the wakeup before we
1139  * update the link state.)
1140  *
1141  * The rcd is freed in hfi1_free_ctxtdata after hfi1_postinit_cleanup invokes
1142  * dd->f_cleanup(dd) to disable the interrupt handler and flush workqueues,
1143  * so we're safe from use-after-free of the rcd.
1144  */
1145 void receive_interrupt_work(struct work_struct *work)
1146 {
1147 	struct hfi1_pportdata *ppd = container_of(work, struct hfi1_pportdata,
1148 						  linkstate_active_work);
1149 	struct hfi1_devdata *dd = ppd->dd;
1150 	struct hfi1_ctxtdata *rcd;
1151 	u16 i;
1152 
1153 	/* Received non-SC15 packet implies neighbor_normal */
1154 	ppd->neighbor_normal = 1;
1155 	set_link_state(ppd, HLS_UP_ACTIVE);
1156 
1157 	/*
1158 	 * Interrupt all statically allocated kernel contexts that could
1159 	 * have had an interrupt during auto activation.
1160 	 */
1161 	for (i = HFI1_CTRL_CTXT; i < dd->first_dyn_alloc_ctxt; i++) {
1162 		rcd = hfi1_rcd_get_by_index(dd, i);
1163 		if (rcd)
1164 			force_recv_intr(rcd);
1165 		hfi1_rcd_put(rcd);
1166 	}
1167 }
1168 
1169 /*
1170  * Convert a given MTU size to the on-wire MAD packet enumeration.
1171  * Return -1 if the size is invalid.
1172  */
1173 int mtu_to_enum(u32 mtu, int default_if_bad)
1174 {
1175 	switch (mtu) {
1176 	case     0: return OPA_MTU_0;
1177 	case   256: return OPA_MTU_256;
1178 	case   512: return OPA_MTU_512;
1179 	case  1024: return OPA_MTU_1024;
1180 	case  2048: return OPA_MTU_2048;
1181 	case  4096: return OPA_MTU_4096;
1182 	case  8192: return OPA_MTU_8192;
1183 	case 10240: return OPA_MTU_10240;
1184 	}
1185 	return default_if_bad;
1186 }
1187 
1188 u16 enum_to_mtu(int mtu)
1189 {
1190 	switch (mtu) {
1191 	case OPA_MTU_0:     return 0;
1192 	case OPA_MTU_256:   return 256;
1193 	case OPA_MTU_512:   return 512;
1194 	case OPA_MTU_1024:  return 1024;
1195 	case OPA_MTU_2048:  return 2048;
1196 	case OPA_MTU_4096:  return 4096;
1197 	case OPA_MTU_8192:  return 8192;
1198 	case OPA_MTU_10240: return 10240;
1199 	default: return 0xffff;
1200 	}
1201 }
1202 
1203 /*
1204  * set_mtu - set the MTU
1205  * @ppd: the per port data
1206  *
1207  * We can handle "any" incoming size, the issue here is whether we
1208  * need to restrict our outgoing size.  We do not deal with what happens
1209  * to programs that are already running when the size changes.
1210  */
1211 int set_mtu(struct hfi1_pportdata *ppd)
1212 {
1213 	struct hfi1_devdata *dd = ppd->dd;
1214 	int i, drain, ret = 0, is_up = 0;
1215 
1216 	ppd->ibmtu = 0;
1217 	for (i = 0; i < ppd->vls_supported; i++)
1218 		if (ppd->ibmtu < dd->vld[i].mtu)
1219 			ppd->ibmtu = dd->vld[i].mtu;
1220 	ppd->ibmaxlen = ppd->ibmtu + lrh_max_header_bytes(ppd->dd);
1221 
1222 	mutex_lock(&ppd->hls_lock);
1223 	if (ppd->host_link_state == HLS_UP_INIT ||
1224 	    ppd->host_link_state == HLS_UP_ARMED ||
1225 	    ppd->host_link_state == HLS_UP_ACTIVE)
1226 		is_up = 1;
1227 
1228 	drain = !is_ax(dd) && is_up;
1229 
1230 	if (drain)
1231 		/*
1232 		 * MTU is specified per-VL. To ensure that no packet gets
1233 		 * stuck (due, e.g., to the MTU for the packet's VL being
1234 		 * reduced), empty the per-VL FIFOs before adjusting MTU.
1235 		 */
1236 		ret = stop_drain_data_vls(dd);
1237 
1238 	if (ret) {
1239 		dd_dev_err(dd, "%s: cannot stop/drain VLs - refusing to change per-VL MTUs\n",
1240 			   __func__);
1241 		goto err;
1242 	}
1243 
1244 	hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_MTU, 0);
1245 
1246 	if (drain)
1247 		open_fill_data_vls(dd); /* reopen all VLs */
1248 
1249 err:
1250 	mutex_unlock(&ppd->hls_lock);
1251 
1252 	return ret;
1253 }
1254 
1255 int hfi1_set_lid(struct hfi1_pportdata *ppd, u32 lid, u8 lmc)
1256 {
1257 	struct hfi1_devdata *dd = ppd->dd;
1258 
1259 	ppd->lid = lid;
1260 	ppd->lmc = lmc;
1261 	hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_LIDLMC, 0);
1262 
1263 	dd_dev_info(dd, "port %u: got a lid: 0x%x\n", ppd->port, lid);
1264 
1265 	return 0;
1266 }
1267 
1268 void shutdown_led_override(struct hfi1_pportdata *ppd)
1269 {
1270 	struct hfi1_devdata *dd = ppd->dd;
1271 
1272 	/*
1273 	 * This pairs with the memory barrier in hfi1_start_led_override to
1274 	 * ensure that we read the correct state of LED beaconing represented
1275 	 * by led_override_timer_active
1276 	 */
1277 	smp_rmb();
1278 	if (atomic_read(&ppd->led_override_timer_active)) {
1279 		del_timer_sync(&ppd->led_override_timer);
1280 		atomic_set(&ppd->led_override_timer_active, 0);
1281 		/* Ensure the atomic_set is visible to all CPUs */
1282 		smp_wmb();
1283 	}
1284 
1285 	/* Hand control of the LED to the DC for normal operation */
1286 	write_csr(dd, DCC_CFG_LED_CNTRL, 0);
1287 }
1288 
1289 static void run_led_override(struct timer_list *t)
1290 {
1291 	struct hfi1_pportdata *ppd = from_timer(ppd, t, led_override_timer);
1292 	struct hfi1_devdata *dd = ppd->dd;
1293 	unsigned long timeout;
1294 	int phase_idx;
1295 
1296 	if (!(dd->flags & HFI1_INITTED))
1297 		return;
1298 
1299 	phase_idx = ppd->led_override_phase & 1;
1300 
1301 	setextled(dd, phase_idx);
1302 
1303 	timeout = ppd->led_override_vals[phase_idx];
1304 
1305 	/* Set up for next phase */
1306 	ppd->led_override_phase = !ppd->led_override_phase;
1307 
1308 	mod_timer(&ppd->led_override_timer, jiffies + timeout);
1309 }
1310 
1311 /*
1312  * To have the LED blink in a particular pattern, provide timeon and timeoff
1313  * in milliseconds.
1314  * To turn off custom blinking and return to normal operation, use
1315  * shutdown_led_override()
1316  */
1317 void hfi1_start_led_override(struct hfi1_pportdata *ppd, unsigned int timeon,
1318 			     unsigned int timeoff)
1319 {
1320 	if (!(ppd->dd->flags & HFI1_INITTED))
1321 		return;
1322 
1323 	/* Convert to jiffies for direct use in timer */
1324 	ppd->led_override_vals[0] = msecs_to_jiffies(timeoff);
1325 	ppd->led_override_vals[1] = msecs_to_jiffies(timeon);
1326 
1327 	/* Arbitrarily start from LED on phase */
1328 	ppd->led_override_phase = 1;
1329 
1330 	/*
1331 	 * If the timer has not already been started, do so. Use a "quick"
1332 	 * timeout so the handler will be called soon to look at our request.
1333 	 */
1334 	if (!timer_pending(&ppd->led_override_timer)) {
1335 		timer_setup(&ppd->led_override_timer, run_led_override, 0);
1336 		ppd->led_override_timer.expires = jiffies + 1;
1337 		add_timer(&ppd->led_override_timer);
1338 		atomic_set(&ppd->led_override_timer_active, 1);
1339 		/* Ensure the atomic_set is visible to all CPUs */
1340 		smp_wmb();
1341 	}
1342 }
1343 
1344 /**
1345  * hfi1_reset_device - reset the chip if possible
1346  * @unit: the device to reset
1347  *
1348  * Whether or not reset is successful, we attempt to re-initialize the chip
1349  * (that is, much like a driver unload/reload).  We clear the INITTED flag
1350  * so that the various entry points will fail until we reinitialize.  For
1351  * now, we only allow this if no user contexts are open that use chip resources
1352  */
1353 int hfi1_reset_device(int unit)
1354 {
1355 	int ret;
1356 	struct hfi1_devdata *dd = hfi1_lookup(unit);
1357 	struct hfi1_pportdata *ppd;
1358 	int pidx;
1359 
1360 	if (!dd) {
1361 		ret = -ENODEV;
1362 		goto bail;
1363 	}
1364 
1365 	dd_dev_info(dd, "Reset on unit %u requested\n", unit);
1366 
1367 	if (!dd->kregbase1 || !(dd->flags & HFI1_PRESENT)) {
1368 		dd_dev_info(dd,
1369 			    "Invalid unit number %u or not initialized or not present\n",
1370 			    unit);
1371 		ret = -ENXIO;
1372 		goto bail;
1373 	}
1374 
1375 	/* If there are any user/vnic contexts, we cannot reset */
1376 	mutex_lock(&hfi1_mutex);
1377 	if (dd->rcd)
1378 		if (hfi1_stats.sps_ctxts) {
1379 			mutex_unlock(&hfi1_mutex);
1380 			ret = -EBUSY;
1381 			goto bail;
1382 		}
1383 	mutex_unlock(&hfi1_mutex);
1384 
1385 	for (pidx = 0; pidx < dd->num_pports; ++pidx) {
1386 		ppd = dd->pport + pidx;
1387 
1388 		shutdown_led_override(ppd);
1389 	}
1390 	if (dd->flags & HFI1_HAS_SEND_DMA)
1391 		sdma_exit(dd);
1392 
1393 	hfi1_reset_cpu_counters(dd);
1394 
1395 	ret = hfi1_init(dd, 1);
1396 
1397 	if (ret)
1398 		dd_dev_err(dd,
1399 			   "Reinitialize unit %u after reset failed with %d\n",
1400 			   unit, ret);
1401 	else
1402 		dd_dev_info(dd, "Reinitialized unit %u after resetting\n",
1403 			    unit);
1404 
1405 bail:
1406 	return ret;
1407 }
1408 
1409 static inline void hfi1_setup_ib_header(struct hfi1_packet *packet)
1410 {
1411 	packet->hdr = (struct hfi1_ib_message_header *)
1412 			hfi1_get_msgheader(packet->rcd,
1413 					   packet->rhf_addr);
1414 	packet->hlen = (u8 *)packet->rhf_addr - (u8 *)packet->hdr;
1415 }
1416 
1417 static int hfi1_bypass_ingress_pkt_check(struct hfi1_packet *packet)
1418 {
1419 	struct hfi1_pportdata *ppd = packet->rcd->ppd;
1420 
1421 	/* slid and dlid cannot be 0 */
1422 	if ((!packet->slid) || (!packet->dlid))
1423 		return -EINVAL;
1424 
1425 	/* Compare port lid with incoming packet dlid */
1426 	if ((!(hfi1_is_16B_mcast(packet->dlid))) &&
1427 	    (packet->dlid !=
1428 		opa_get_lid(be32_to_cpu(OPA_LID_PERMISSIVE), 16B))) {
1429 		if ((packet->dlid & ~((1 << ppd->lmc) - 1)) != ppd->lid)
1430 			return -EINVAL;
1431 	}
1432 
1433 	/* No multicast packets with SC15 */
1434 	if ((hfi1_is_16B_mcast(packet->dlid)) && (packet->sc == 0xF))
1435 		return -EINVAL;
1436 
1437 	/* Packets with permissive DLID always on SC15 */
1438 	if ((packet->dlid == opa_get_lid(be32_to_cpu(OPA_LID_PERMISSIVE),
1439 					 16B)) &&
1440 	    (packet->sc != 0xF))
1441 		return -EINVAL;
1442 
1443 	return 0;
1444 }
1445 
1446 static int hfi1_setup_9B_packet(struct hfi1_packet *packet)
1447 {
1448 	struct hfi1_ibport *ibp = rcd_to_iport(packet->rcd);
1449 	struct ib_header *hdr;
1450 	u8 lnh;
1451 
1452 	hfi1_setup_ib_header(packet);
1453 	hdr = packet->hdr;
1454 
1455 	lnh = ib_get_lnh(hdr);
1456 	if (lnh == HFI1_LRH_BTH) {
1457 		packet->ohdr = &hdr->u.oth;
1458 		packet->grh = NULL;
1459 	} else if (lnh == HFI1_LRH_GRH) {
1460 		u32 vtf;
1461 
1462 		packet->ohdr = &hdr->u.l.oth;
1463 		packet->grh = &hdr->u.l.grh;
1464 		if (packet->grh->next_hdr != IB_GRH_NEXT_HDR)
1465 			goto drop;
1466 		vtf = be32_to_cpu(packet->grh->version_tclass_flow);
1467 		if ((vtf >> IB_GRH_VERSION_SHIFT) != IB_GRH_VERSION)
1468 			goto drop;
1469 	} else {
1470 		goto drop;
1471 	}
1472 
1473 	/* Query commonly used fields from packet header */
1474 	packet->payload = packet->ebuf;
1475 	packet->opcode = ib_bth_get_opcode(packet->ohdr);
1476 	packet->slid = ib_get_slid(hdr);
1477 	packet->dlid = ib_get_dlid(hdr);
1478 	if (unlikely((packet->dlid >= be16_to_cpu(IB_MULTICAST_LID_BASE)) &&
1479 		     (packet->dlid != be16_to_cpu(IB_LID_PERMISSIVE))))
1480 		packet->dlid += opa_get_mcast_base(OPA_MCAST_NR) -
1481 				be16_to_cpu(IB_MULTICAST_LID_BASE);
1482 	packet->sl = ib_get_sl(hdr);
1483 	packet->sc = hfi1_9B_get_sc5(hdr, packet->rhf);
1484 	packet->pad = ib_bth_get_pad(packet->ohdr);
1485 	packet->extra_byte = 0;
1486 	packet->pkey = ib_bth_get_pkey(packet->ohdr);
1487 	packet->migrated = ib_bth_is_migration(packet->ohdr);
1488 
1489 	return 0;
1490 drop:
1491 	ibp->rvp.n_pkt_drops++;
1492 	return -EINVAL;
1493 }
1494 
1495 static int hfi1_setup_bypass_packet(struct hfi1_packet *packet)
1496 {
1497 	/*
1498 	 * Bypass packets have a different header/payload split
1499 	 * compared to an IB packet.
1500 	 * Current split is set such that 16 bytes of the actual
1501 	 * header is in the header buffer and the remining is in
1502 	 * the eager buffer. We chose 16 since hfi1 driver only
1503 	 * supports 16B bypass packets and we will be able to
1504 	 * receive the entire LRH with such a split.
1505 	 */
1506 
1507 	struct hfi1_ctxtdata *rcd = packet->rcd;
1508 	struct hfi1_pportdata *ppd = rcd->ppd;
1509 	struct hfi1_ibport *ibp = &ppd->ibport_data;
1510 	u8 l4;
1511 
1512 	packet->hdr = (struct hfi1_16b_header *)
1513 			hfi1_get_16B_header(packet->rcd,
1514 					    packet->rhf_addr);
1515 	l4 = hfi1_16B_get_l4(packet->hdr);
1516 	if (l4 == OPA_16B_L4_IB_LOCAL) {
1517 		packet->ohdr = packet->ebuf;
1518 		packet->grh = NULL;
1519 		packet->opcode = ib_bth_get_opcode(packet->ohdr);
1520 		packet->pad = hfi1_16B_bth_get_pad(packet->ohdr);
1521 		/* hdr_len_by_opcode already has an IB LRH factored in */
1522 		packet->hlen = hdr_len_by_opcode[packet->opcode] +
1523 			(LRH_16B_BYTES - LRH_9B_BYTES);
1524 		packet->migrated = opa_bth_is_migration(packet->ohdr);
1525 	} else if (l4 == OPA_16B_L4_IB_GLOBAL) {
1526 		u32 vtf;
1527 		u8 grh_len = sizeof(struct ib_grh);
1528 
1529 		packet->ohdr = packet->ebuf + grh_len;
1530 		packet->grh = packet->ebuf;
1531 		packet->opcode = ib_bth_get_opcode(packet->ohdr);
1532 		packet->pad = hfi1_16B_bth_get_pad(packet->ohdr);
1533 		/* hdr_len_by_opcode already has an IB LRH factored in */
1534 		packet->hlen = hdr_len_by_opcode[packet->opcode] +
1535 			(LRH_16B_BYTES - LRH_9B_BYTES) + grh_len;
1536 		packet->migrated = opa_bth_is_migration(packet->ohdr);
1537 
1538 		if (packet->grh->next_hdr != IB_GRH_NEXT_HDR)
1539 			goto drop;
1540 		vtf = be32_to_cpu(packet->grh->version_tclass_flow);
1541 		if ((vtf >> IB_GRH_VERSION_SHIFT) != IB_GRH_VERSION)
1542 			goto drop;
1543 	} else if (l4 == OPA_16B_L4_FM) {
1544 		packet->mgmt = packet->ebuf;
1545 		packet->ohdr = NULL;
1546 		packet->grh = NULL;
1547 		packet->opcode = IB_OPCODE_UD_SEND_ONLY;
1548 		packet->pad = OPA_16B_L4_FM_PAD;
1549 		packet->hlen = OPA_16B_L4_FM_HLEN;
1550 		packet->migrated = false;
1551 	} else {
1552 		goto drop;
1553 	}
1554 
1555 	/* Query commonly used fields from packet header */
1556 	packet->payload = packet->ebuf + packet->hlen - LRH_16B_BYTES;
1557 	packet->slid = hfi1_16B_get_slid(packet->hdr);
1558 	packet->dlid = hfi1_16B_get_dlid(packet->hdr);
1559 	if (unlikely(hfi1_is_16B_mcast(packet->dlid)))
1560 		packet->dlid += opa_get_mcast_base(OPA_MCAST_NR) -
1561 				opa_get_lid(opa_get_mcast_base(OPA_MCAST_NR),
1562 					    16B);
1563 	packet->sc = hfi1_16B_get_sc(packet->hdr);
1564 	packet->sl = ibp->sc_to_sl[packet->sc];
1565 	packet->extra_byte = SIZE_OF_LT;
1566 	packet->pkey = hfi1_16B_get_pkey(packet->hdr);
1567 
1568 	if (hfi1_bypass_ingress_pkt_check(packet))
1569 		goto drop;
1570 
1571 	return 0;
1572 drop:
1573 	hfi1_cdbg(PKT, "%s: packet dropped\n", __func__);
1574 	ibp->rvp.n_pkt_drops++;
1575 	return -EINVAL;
1576 }
1577 
1578 static void show_eflags_errs(struct hfi1_packet *packet)
1579 {
1580 	struct hfi1_ctxtdata *rcd = packet->rcd;
1581 	u32 rte = rhf_rcv_type_err(packet->rhf);
1582 
1583 	dd_dev_err(rcd->dd,
1584 		   "receive context %d: rhf 0x%016llx, errs [ %s%s%s%s%s%s%s%s] rte 0x%x\n",
1585 		   rcd->ctxt, packet->rhf,
1586 		   packet->rhf & RHF_K_HDR_LEN_ERR ? "k_hdr_len " : "",
1587 		   packet->rhf & RHF_DC_UNC_ERR ? "dc_unc " : "",
1588 		   packet->rhf & RHF_DC_ERR ? "dc " : "",
1589 		   packet->rhf & RHF_TID_ERR ? "tid " : "",
1590 		   packet->rhf & RHF_LEN_ERR ? "len " : "",
1591 		   packet->rhf & RHF_ECC_ERR ? "ecc " : "",
1592 		   packet->rhf & RHF_VCRC_ERR ? "vcrc " : "",
1593 		   packet->rhf & RHF_ICRC_ERR ? "icrc " : "",
1594 		   rte);
1595 }
1596 
1597 void handle_eflags(struct hfi1_packet *packet)
1598 {
1599 	struct hfi1_ctxtdata *rcd = packet->rcd;
1600 
1601 	rcv_hdrerr(rcd, rcd->ppd, packet);
1602 	if (rhf_err_flags(packet->rhf))
1603 		show_eflags_errs(packet);
1604 }
1605 
1606 /*
1607  * The following functions are called by the interrupt handler. They are type
1608  * specific handlers for each packet type.
1609  */
1610 static int process_receive_ib(struct hfi1_packet *packet)
1611 {
1612 	if (hfi1_setup_9B_packet(packet))
1613 		return RHF_RCV_CONTINUE;
1614 
1615 	if (unlikely(hfi1_dbg_should_fault_rx(packet)))
1616 		return RHF_RCV_CONTINUE;
1617 
1618 	trace_hfi1_rcvhdr(packet);
1619 
1620 	if (unlikely(rhf_err_flags(packet->rhf))) {
1621 		handle_eflags(packet);
1622 		return RHF_RCV_CONTINUE;
1623 	}
1624 
1625 	hfi1_ib_rcv(packet);
1626 	return RHF_RCV_CONTINUE;
1627 }
1628 
1629 static inline bool hfi1_is_vnic_packet(struct hfi1_packet *packet)
1630 {
1631 	/* Packet received in VNIC context via RSM */
1632 	if (packet->rcd->is_vnic)
1633 		return true;
1634 
1635 	if ((hfi1_16B_get_l2(packet->ebuf) == OPA_16B_L2_TYPE) &&
1636 	    (hfi1_16B_get_l4(packet->ebuf) == OPA_16B_L4_ETHR))
1637 		return true;
1638 
1639 	return false;
1640 }
1641 
1642 static int process_receive_bypass(struct hfi1_packet *packet)
1643 {
1644 	struct hfi1_devdata *dd = packet->rcd->dd;
1645 
1646 	if (hfi1_is_vnic_packet(packet)) {
1647 		hfi1_vnic_bypass_rcv(packet);
1648 		return RHF_RCV_CONTINUE;
1649 	}
1650 
1651 	if (hfi1_setup_bypass_packet(packet))
1652 		return RHF_RCV_CONTINUE;
1653 
1654 	trace_hfi1_rcvhdr(packet);
1655 
1656 	if (unlikely(rhf_err_flags(packet->rhf))) {
1657 		handle_eflags(packet);
1658 		return RHF_RCV_CONTINUE;
1659 	}
1660 
1661 	if (hfi1_16B_get_l2(packet->hdr) == 0x2) {
1662 		hfi1_16B_rcv(packet);
1663 	} else {
1664 		dd_dev_err(dd,
1665 			   "Bypass packets other than 16B are not supported in normal operation. Dropping\n");
1666 		incr_cntr64(&dd->sw_rcv_bypass_packet_errors);
1667 		if (!(dd->err_info_rcvport.status_and_code &
1668 		      OPA_EI_STATUS_SMASK)) {
1669 			u64 *flits = packet->ebuf;
1670 
1671 			if (flits && !(packet->rhf & RHF_LEN_ERR)) {
1672 				dd->err_info_rcvport.packet_flit1 = flits[0];
1673 				dd->err_info_rcvport.packet_flit2 =
1674 					packet->tlen > sizeof(flits[0]) ?
1675 					flits[1] : 0;
1676 			}
1677 			dd->err_info_rcvport.status_and_code |=
1678 				(OPA_EI_STATUS_SMASK | BAD_L2_ERR);
1679 		}
1680 	}
1681 	return RHF_RCV_CONTINUE;
1682 }
1683 
1684 static int process_receive_error(struct hfi1_packet *packet)
1685 {
1686 	/* KHdrHCRCErr -- KDETH packet with a bad HCRC */
1687 	if (unlikely(
1688 		 hfi1_dbg_fault_suppress_err(&packet->rcd->dd->verbs_dev) &&
1689 		 (rhf_rcv_type_err(packet->rhf) == RHF_RCV_TYPE_ERROR ||
1690 		  packet->rhf & RHF_DC_ERR)))
1691 		return RHF_RCV_CONTINUE;
1692 
1693 	hfi1_setup_ib_header(packet);
1694 	handle_eflags(packet);
1695 
1696 	if (unlikely(rhf_err_flags(packet->rhf)))
1697 		dd_dev_err(packet->rcd->dd,
1698 			   "Unhandled error packet received. Dropping.\n");
1699 
1700 	return RHF_RCV_CONTINUE;
1701 }
1702 
1703 static int kdeth_process_expected(struct hfi1_packet *packet)
1704 {
1705 	hfi1_setup_9B_packet(packet);
1706 	if (unlikely(hfi1_dbg_should_fault_rx(packet)))
1707 		return RHF_RCV_CONTINUE;
1708 
1709 	if (unlikely(rhf_err_flags(packet->rhf))) {
1710 		struct hfi1_ctxtdata *rcd = packet->rcd;
1711 
1712 		if (hfi1_handle_kdeth_eflags(rcd, rcd->ppd, packet))
1713 			return RHF_RCV_CONTINUE;
1714 	}
1715 
1716 	hfi1_kdeth_expected_rcv(packet);
1717 	return RHF_RCV_CONTINUE;
1718 }
1719 
1720 static int kdeth_process_eager(struct hfi1_packet *packet)
1721 {
1722 	hfi1_setup_9B_packet(packet);
1723 	if (unlikely(hfi1_dbg_should_fault_rx(packet)))
1724 		return RHF_RCV_CONTINUE;
1725 
1726 	trace_hfi1_rcvhdr(packet);
1727 	if (unlikely(rhf_err_flags(packet->rhf))) {
1728 		struct hfi1_ctxtdata *rcd = packet->rcd;
1729 
1730 		show_eflags_errs(packet);
1731 		if (hfi1_handle_kdeth_eflags(rcd, rcd->ppd, packet))
1732 			return RHF_RCV_CONTINUE;
1733 	}
1734 
1735 	hfi1_kdeth_eager_rcv(packet);
1736 	return RHF_RCV_CONTINUE;
1737 }
1738 
1739 static int process_receive_invalid(struct hfi1_packet *packet)
1740 {
1741 	dd_dev_err(packet->rcd->dd, "Invalid packet type %d. Dropping\n",
1742 		   rhf_rcv_type(packet->rhf));
1743 	return RHF_RCV_CONTINUE;
1744 }
1745 
1746 void seqfile_dump_rcd(struct seq_file *s, struct hfi1_ctxtdata *rcd)
1747 {
1748 	struct hfi1_packet packet;
1749 	struct ps_mdata mdata;
1750 
1751 	seq_printf(s, "Rcd %u: RcvHdr cnt %u entsize %u %s head %llu tail %llu\n",
1752 		   rcd->ctxt, rcd->rcvhdrq_cnt, rcd->rcvhdrqentsize,
1753 		   HFI1_CAP_KGET_MASK(rcd->flags, DMA_RTAIL) ?
1754 		   "dma_rtail" : "nodma_rtail",
1755 		   read_uctxt_csr(rcd->dd, rcd->ctxt, RCV_HDR_HEAD) &
1756 		   RCV_HDR_HEAD_HEAD_MASK,
1757 		   read_uctxt_csr(rcd->dd, rcd->ctxt, RCV_HDR_TAIL));
1758 
1759 	init_packet(rcd, &packet);
1760 	init_ps_mdata(&mdata, &packet);
1761 
1762 	while (1) {
1763 		__le32 *rhf_addr = (__le32 *)rcd->rcvhdrq + mdata.ps_head +
1764 					 rcd->rhf_offset;
1765 		struct ib_header *hdr;
1766 		u64 rhf = rhf_to_cpu(rhf_addr);
1767 		u32 etype = rhf_rcv_type(rhf), qpn;
1768 		u8 opcode;
1769 		u32 psn;
1770 		u8 lnh;
1771 
1772 		if (ps_done(&mdata, rhf, rcd))
1773 			break;
1774 
1775 		if (ps_skip(&mdata, rhf, rcd))
1776 			goto next;
1777 
1778 		if (etype > RHF_RCV_TYPE_IB)
1779 			goto next;
1780 
1781 		packet.hdr = hfi1_get_msgheader(rcd, rhf_addr);
1782 		hdr = packet.hdr;
1783 
1784 		lnh = be16_to_cpu(hdr->lrh[0]) & 3;
1785 
1786 		if (lnh == HFI1_LRH_BTH)
1787 			packet.ohdr = &hdr->u.oth;
1788 		else if (lnh == HFI1_LRH_GRH)
1789 			packet.ohdr = &hdr->u.l.oth;
1790 		else
1791 			goto next; /* just in case */
1792 
1793 		opcode = (be32_to_cpu(packet.ohdr->bth[0]) >> 24);
1794 		qpn = be32_to_cpu(packet.ohdr->bth[1]) & RVT_QPN_MASK;
1795 		psn = mask_psn(be32_to_cpu(packet.ohdr->bth[2]));
1796 
1797 		seq_printf(s, "\tEnt %u: opcode 0x%x, qpn 0x%x, psn 0x%x\n",
1798 			   mdata.ps_head, opcode, qpn, psn);
1799 next:
1800 		update_ps_mdata(&mdata, rcd);
1801 	}
1802 }
1803 
1804 const rhf_rcv_function_ptr normal_rhf_rcv_functions[] = {
1805 	[RHF_RCV_TYPE_EXPECTED] = kdeth_process_expected,
1806 	[RHF_RCV_TYPE_EAGER] = kdeth_process_eager,
1807 	[RHF_RCV_TYPE_IB] = process_receive_ib,
1808 	[RHF_RCV_TYPE_ERROR] = process_receive_error,
1809 	[RHF_RCV_TYPE_BYPASS] = process_receive_bypass,
1810 	[RHF_RCV_TYPE_INVALID5] = process_receive_invalid,
1811 	[RHF_RCV_TYPE_INVALID6] = process_receive_invalid,
1812 	[RHF_RCV_TYPE_INVALID7] = process_receive_invalid,
1813 };
1814