summaryrefslogtreecommitdiff
path: root/drivers/infiniband/hw/qib/qib_diag.c
blob: 11da796dd1b7161180473c3b785066b322b37085 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
/*
 * Copyright (c) 2012 Intel Corporation. All rights reserved.
 * Copyright (c) 2006 - 2012 QLogic Corporation. All rights reserved.
 * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
 *
 * This software is available to you under a choice of one of two
 * licenses.  You may choose to be licensed under the terms of the GNU
 * General Public License (GPL) Version 2, available from the file
 * COPYING in the main directory of this source tree, or the
 * OpenIB.org BSD license below:
 *
 *     Redistribution and use in source and binary forms, with or
 *     without modification, are permitted provided that the following
 *     conditions are met:
 *
 *      - Redistributions of source code must retain the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer.
 *
 *      - Redistributions in binary form must reproduce the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer in the documentation and/or other materials
 *        provided with the distribution.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 * SOFTWARE.
 */

/*
 * This file contains support for diagnostic functions.  It is accessed by
 * opening the qib_diag device, normally minor number 129.  Diagnostic use
 * of the QLogic_IB chip may render the chip or board unusable until the
 * driver is unloaded, or in some cases, until the system is rebooted.
 *
 * Accesses to the chip through this interface are not similar to going
 * through the /sys/bus/pci resource mmap interface.
 */

#include <linux/io.h>
#include <linux/pci.h>
#include <linux/poll.h>
#include <linux/vmalloc.h>
#include <linux/export.h>
#include <linux/fs.h>
#include <linux/uaccess.h>

#include "qib.h"
#include "qib_common.h"

#undef pr_fmt
#define pr_fmt(fmt) QIB_DRV_NAME ": " fmt

/*
 * Each client that opens the diag device must read then write
 * offset 0, to prevent lossage from random cat or od. diag_state
 * sequences this "handshake".
 */
enum diag_state { UNUSED = 0, OPENED, INIT, READY };

/* State for an individual client. PID so children cannot abuse handshake */
static struct qib_diag_client {
	struct qib_diag_client *next;
	struct qib_devdata *dd;
	pid_t pid;
	enum diag_state state;
} *client_pool;

/*
 * Get a client struct. Recycled if possible, else kmalloc.
 * Must be called with qib_mutex held
 */
static struct qib_diag_client *get_client(struct qib_devdata *dd)
{
	struct qib_diag_client *dc;

	dc = client_pool;
	if (dc)
		/* got from pool remove it and use */
		client_pool = dc->next;
	else
		/* None in pool, alloc and init */
		dc = kmalloc(sizeof(*dc), GFP_KERNEL);

	if (dc) {
		dc->next = NULL;
		dc->dd = dd;
		dc->pid = current->pid;
		dc->state = OPENED;
	}
	return dc;
}

/*
 * Return to pool. Must be called with qib_mutex held
 */
static void return_client(struct qib_diag_client *dc)
{
	struct qib_devdata *dd = dc->dd;
	struct qib_diag_client *tdc, *rdc;

	rdc = NULL;
	if (dc == dd->diag_client) {
		dd->diag_client = dc->next;
		rdc = dc;
	} else {
		tdc = dc->dd->diag_client;
		while (tdc) {
			if (dc == tdc->next) {
				tdc->next = dc->next;
				rdc = dc;
				break;
			}
			tdc = tdc->next;
		}
	}
	if (rdc) {
		rdc->state = UNUSED;
		rdc->dd = NULL;
		rdc->pid = 0;
		rdc->next = client_pool;
		client_pool = rdc;
	}
}

static int qib_diag_open(struct inode *in, struct file *fp);
static int qib_diag_release(struct inode *in, struct file *fp);
static ssize_t qib_diag_read(struct file *fp, char __user *data,
			     size_t count, loff_t *off);
static ssize_t qib_diag_write(struct file *fp, const char __user *data,
			      size_t count, loff_t *off);

static const struct file_operations diag_file_ops = {
	.owner = THIS_MODULE,
	.write = qib_diag_write,
	.read = qib_diag_read,
	.open = qib_diag_open,
	.release = qib_diag_release,
	.llseek = default_llseek,
};

static atomic_t diagpkt_count = ATOMIC_INIT(0);
static struct cdev *diagpkt_cdev;
static struct device *diagpkt_device;

static ssize_t qib_diagpkt_write(struct file *fp, const char __user *data,
				 size_t count, loff_t *off);

static const struct file_operations diagpkt_file_ops = {
	.owner = THIS_MODULE,
	.write = qib_diagpkt_write,
	.llseek = noop_llseek,
};

int qib_diag_add(struct qib_devdata *dd)
{
	char name[16];
	int ret = 0;

	if (atomic_inc_return(&diagpkt_count) == 1) {
		ret = qib_cdev_init(QIB_DIAGPKT_MINOR, "ipath_diagpkt",
				    &diagpkt_file_ops, &diagpkt_cdev,
				    &diagpkt_device);
		if (ret)
			goto done;
	}

	snprintf(name, sizeof(name), "ipath_diag%d", dd->unit);
	ret = qib_cdev_init(QIB_DIAG_MINOR_BASE + dd->unit, name,
			    &diag_file_ops, &dd->diag_cdev,
			    &dd->diag_device);
done:
	return ret;
}

static void qib_unregister_observers(struct qib_devdata *dd);

void qib_diag_remove(struct qib_devdata *dd)
{
	struct qib_diag_client *dc;

	if (atomic_dec_and_test(&diagpkt_count))
		qib_cdev_cleanup(&diagpkt_cdev, &diagpkt_device);

	qib_cdev_cleanup(&dd->diag_cdev, &dd->diag_device);

	/*
	 * Return all diag_clients of this device. There should be none,
	 * as we are "guaranteed" that no clients are still open
	 */
	while (dd->diag_client)
		return_client(dd->diag_client);

	/* Now clean up all unused client structs */
	while (client_pool) {
		dc = client_pool;
		client_pool = dc->next;
		kfree(dc);
	}
	/* Clean up observer list */
	qib_unregister_observers(dd);
}

/* qib_remap_ioaddr32 - remap an offset into chip address space to __iomem *
 *
 * @dd: the qlogic_ib device
 * @offs: the offset in chip-space
 * @cntp: Pointer to max (byte) count for transfer starting at offset
 * This returns a u32 __iomem * so it can be used for both 64 and 32-bit
 * mapping. It is needed because with the use of PAT for control of
 * write-combining, the logically contiguous address-space of the chip
 * may be split into virtually non-contiguous spaces, with different
 * attributes, which are them mapped to contiguous physical space
 * based from the first BAR.
 *
 * The code below makes the same assumptions as were made in
 * init_chip_wc_pat() (qib_init.c), copied here:
 * Assumes chip address space looks like:
 *		- kregs + sregs + cregs + uregs (in any order)
 *		- piobufs (2K and 4K bufs in either order)
 *	or:
 *		- kregs + sregs + cregs (in any order)
 *		- piobufs (2K and 4K bufs in either order)
 *		- uregs
 *
 * If cntp is non-NULL, returns how many bytes from offset can be accessed
 * Returns 0 if the offset is not mapped.
 */
static u32 __iomem *qib_remap_ioaddr32(struct qib_devdata *dd, u32 offset,
				       u32 *cntp)
{
	u32 kreglen;
	u32 snd_bottom, snd_lim = 0;
	u32 __iomem *krb32 = (u32 __iomem *)dd->kregbase;
	u32 __iomem *map = NULL;
	u32 cnt = 0;
	u32 tot4k, offs4k;

	/* First, simplest case, offset is within the first map. */
	kreglen = (dd->kregend - dd->kregbase) * sizeof(u64);
	if (offset < kreglen) {
		map = krb32 + (offset / sizeof(u32));
		cnt = kreglen - offset;
		goto mapped;
	}

	/*
	 * Next check for user regs, the next most common case,
	 * and a cheap check because if they are not in the first map
	 * they are last in chip.
	 */
	if (dd->userbase) {
		/* If user regs mapped, they are after send, so set limit. */
		u32 ulim = (dd->cfgctxts * dd->ureg_align) + dd->uregbase;

		if (!dd->piovl15base)
			snd_lim = dd->uregbase;
		krb32 = (u32 __iomem *)dd->userbase;
		if (offset >= dd->uregbase && offset < ulim) {
			map = krb32 + (offset - dd->uregbase) / sizeof(u32);
			cnt = ulim - offset;
			goto mapped;
		}
	}

	/*
	 * Lastly, check for offset within Send Buffers.
	 * This is gnarly because struct devdata is deliberately vague
	 * about things like 7322 VL15 buffers, and we are not in
	 * chip-specific code here, so should not make many assumptions.
	 * The one we _do_ make is that the only chip that has more sndbufs
	 * than we admit is the 7322, and it has userregs above that, so
	 * we know the snd_lim.
	 */
	/* Assume 2K buffers are first. */
	snd_bottom = dd->pio2k_bufbase;
	if (snd_lim == 0) {
		u32 tot2k = dd->piobcnt2k * ALIGN(dd->piosize2k, dd->palign);

		snd_lim = snd_bottom + tot2k;
	}
	/* If 4k buffers exist, account for them by bumping
	 * appropriate limit.
	 */
	tot4k = dd->piobcnt4k * dd->align4k;
	offs4k = dd->piobufbase >> 32;
	if (dd->piobcnt4k) {
		if (snd_bottom > offs4k)
			snd_bottom = offs4k;
		else {
			/* 4k above 2k. Bump snd_lim, if needed*/
			if (!dd->userbase || dd->piovl15base)
				snd_lim = offs4k + tot4k;
		}
	}
	/*
	 * Judgement call: can we ignore the space between SendBuffs and
	 * UserRegs, where we would like to see vl15 buffs, but not more?
	 */
	if (offset >= snd_bottom && offset < snd_lim) {
		offset -= snd_bottom;
		map = (u32 __iomem *)dd->piobase + (offset / sizeof(u32));
		cnt = snd_lim - offset;
	}

	if (!map && offs4k && dd->piovl15base) {
		snd_lim = offs4k + tot4k + 2 * dd->align4k;
		if (offset >= (offs4k + tot4k) && offset < snd_lim) {
			map = (u32 __iomem *)dd->piovl15base +
				((offset - (offs4k + tot4k)) / sizeof(u32));
			cnt = snd_lim - offset;
		}
	}

mapped:
	if (cntp)
		*cntp = cnt;
	return map;
}

/*
 * qib_read_umem64 - read a 64-bit quantity from the chip into user space
 * @dd: the qlogic_ib device
 * @uaddr: the location to store the data in user memory
 * @regoffs: the offset from BAR0 (_NOT_ full pointer, anymore)
 * @count: number of bytes to copy (multiple of 32 bits)
 *
 * This function also localizes all chip memory accesses.
 * The copy should be written such that we read full cacheline packets
 * from the chip.  This is usually used for a single qword
 *
 * NOTE:  This assumes the chip address is 64-bit aligned.
 */
static int qib_read_umem64(struct qib_devdata *dd, void __user *uaddr,
			   u32 regoffs, size_t count)
{
	const u64 __iomem *reg_addr;
	const u64 __iomem *reg_end;
	u32 limit;
	int ret;

	reg_addr = (const u64 __iomem *)qib_remap_ioaddr32(dd, regoffs, &limit);
	if (reg_addr == NULL || limit == 0 || !(dd->flags & QIB_PRESENT)) {
		ret = -EINVAL;
		goto bail;
	}
	if (count >= limit)
		count = limit;
	reg_end = reg_addr + (count / sizeof(u64));

	/* not very efficient, but it works for now */
	while (reg_addr < reg_end) {
		u64 data = readq(reg_addr);

		if (copy_to_user(uaddr, &data, sizeof(u64))) {
			ret = -EFAULT;
			goto bail;
		}
		reg_addr++;
		uaddr += sizeof(u64);
	}
	ret = 0;
bail:
	return ret;
}

/*
 * qib_write_umem64 - write a 64-bit quantity to the chip from user space
 * @dd: the qlogic_ib device
 * @regoffs: the offset from BAR0 (_NOT_ full pointer, anymore)
 * @uaddr: the source of the data in user memory
 * @count: the number of bytes to copy (multiple of 32 bits)
 *
 * This is usually used for a single qword
 * NOTE:  This assumes the chip address is 64-bit aligned.
 */

static int qib_write_umem64(struct qib_devdata *dd, u32 regoffs,
			    const void __user *uaddr, size_t count)
{
	u64 __iomem *reg_addr;
	const u64 __iomem *reg_end;
	u32 limit;
	int ret;

	reg_addr = (u64 __iomem *)qib_remap_ioaddr32(dd, regoffs, &limit);
	if (reg_addr == NULL || limit == 0 || !(dd->flags & QIB_PRESENT)) {
		ret = -EINVAL;
		goto bail;
	}
	if (count >= limit)
		count = limit;
	reg_end = reg_addr + (count / sizeof(u64));

	/* not very efficient, but it works for now */
	while (reg_addr < reg_end) {
		u64 data;

		if (copy_from_user(&data, uaddr, sizeof(data))) {
			ret = -EFAULT;
			goto bail;
		}
		writeq(data, reg_addr);

		reg_addr++;
		uaddr += sizeof(u64);
	}
	ret = 0;
bail:
	return ret;
}

/*
 * qib_read_umem32 - read a 32-bit quantity from the chip into user space
 * @dd: the qlogic_ib device
 * @uaddr: the location to store the data in user memory
 * @regoffs: the offset from BAR0 (_NOT_ full pointer, anymore)
 * @count: number of bytes to copy
 *
 * read 32 bit values, not 64 bit; for memories that only
 * support 32 bit reads; usually a single dword.
 */
static int qib_read_umem32(struct qib_devdata *dd, void __user *uaddr,
			   u32 regoffs, size_t count)
{
	const u32 __iomem *reg_addr;
	const u32 __iomem *reg_end;
	u32 limit;
	int ret;

	reg_addr = qib_remap_ioaddr32(dd, regoffs, &limit);
	if (reg_addr == NULL || limit == 0 || !(dd->flags & QIB_PRESENT)) {
		ret = -EINVAL;
		goto bail;
	}
	if (count >= limit)
		count = limit;
	reg_end = reg_addr + (count / sizeof(u32));

	/* not very efficient, but it works for now */
	while (reg_addr < reg_end) {
		u32 data = readl(reg_addr);

		if (copy_to_user(uaddr, &data, sizeof(data))) {
			ret = -EFAULT;
			goto bail;
		}

		reg_addr++;
		uaddr += sizeof(u32);

	}
	ret = 0;
bail:
	return ret;
}

/*
 * qib_write_umem32 - write a 32-bit quantity to the chip from user space
 * @dd: the qlogic_ib device
 * @regoffs: the offset from BAR0 (_NOT_ full pointer, anymore)
 * @uaddr: the source of the data in user memory
 * @count: number of bytes to copy
 *
 * write 32 bit values, not 64 bit; for memories that only
 * support 32 bit write; usually a single dword.
 */

static int qib_write_umem32(struct qib_devdata *dd, u32 regoffs,
			    const void __user *uaddr, size_t count)
{
	u32 __iomem *reg_addr;
	const u32 __iomem *reg_end;
	u32 limit;
	int ret;

	reg_addr = qib_remap_ioaddr32(dd, regoffs, &limit);
	if (reg_addr == NULL || limit == 0 || !(dd->flags & QIB_PRESENT)) {
		ret = -EINVAL;
		goto bail;
	}
	if (count >= limit)
		count = limit;
	reg_end = reg_addr + (count / sizeof(u32));

	while (reg_addr < reg_end) {
		u32 data;

		if (copy_from_user(&data, uaddr, sizeof(data))) {
			ret = -EFAULT;
			goto bail;
		}
		writel(data, reg_addr);

		reg_addr++;
		uaddr += sizeof(u32);
	}
	ret = 0;
bail:
	return ret;
}

static int qib_diag_open(struct inode *in, struct file *fp)
{
	int unit = iminor(in) - QIB_DIAG_MINOR_BASE;
	struct qib_devdata *dd;
	struct qib_diag_client *dc;
	int ret;

	mutex_lock(&qib_mutex);

	dd = qib_lookup(unit);

	if (dd == NULL || !(dd->flags & QIB_PRESENT) ||
	    !dd->kregbase) {
		ret = -ENODEV;
		goto bail;
	}

	dc = get_client(dd);
	if (!dc) {
		ret = -ENOMEM;
		goto bail;
	}
	dc->next = dd->diag_client;
	dd->diag_client = dc;
	fp->private_data = dc;
	ret = 0;
bail:
	mutex_unlock(&qib_mutex);

	return ret;
}

/**
 * qib_diagpkt_write - write an IB packet
 * @fp: the diag data device file pointer
 * @data: qib_diag_pkt structure saying where to get the packet
 * @count: size of data to write
 * @off: unused by this code
 */
static ssize_t qib_diagpkt_write(struct file *fp,
				 const char __user *data,
				 size_t count, loff_t *off)
{
	u32 __iomem *piobuf;
	u32 plen, pbufn, maxlen_reserve;
	struct qib_diag_xpkt dp;
	u32 *tmpbuf = NULL;
	struct qib_devdata *dd;
	struct qib_pportdata *ppd;
	ssize_t ret = 0;

	if (count != sizeof(dp)) {
		ret = -EINVAL;
		goto bail;
	}
	if (copy_from_user(&dp, data, sizeof(dp))) {
		ret = -EFAULT;
		goto bail;
	}

	dd = qib_lookup(dp.unit);
	if (!dd || !(dd->flags & QIB_PRESENT) || !dd->kregbase) {
		ret = -ENODEV;
		goto bail;
	}
	if (!(dd->flags & QIB_INITTED)) {
		/* no hardware, freeze, etc. */
		ret = -ENODEV;
		goto bail;
	}

	if (dp.version != _DIAG_XPKT_VERS) {
		qib_dev_err(dd, "Invalid version %u for diagpkt_write\n",
			    dp.version);
		ret = -EINVAL;
		goto bail;
	}
	/* send count must be an exact number of dwords */
	if (dp.len & 3) {
		ret = -EINVAL;
		goto bail;
	}
	if (!dp.port || dp.port > dd->num_pports) {
		ret = -EINVAL;
		goto bail;
	}
	ppd = &dd->pport[dp.port - 1];

	/*
	 * need total length before first word written, plus 2 Dwords. One Dword
	 * is for padding so we get the full user data when not aligned on
	 * a word boundary. The other Dword is to make sure we have room for the
	 * ICRC which gets tacked on later.
	 */
	maxlen_reserve = 2 * sizeof(u32);
	if (dp.len > ppd->ibmaxlen - maxlen_reserve) {
		ret = -EINVAL;
		goto bail;
	}

	plen = sizeof(u32) + dp.len;

	tmpbuf = vmalloc(plen);
	if (!tmpbuf) {
		ret = -ENOMEM;
		goto bail;
	}

	if (copy_from_user(tmpbuf,
			   u64_to_user_ptr(dp.data),
			   dp.len)) {
		ret = -EFAULT;
		goto bail;
	}

	plen >>= 2;             /* in dwords */

	if (dp.pbc_wd == 0)
		dp.pbc_wd = plen;

	piobuf = dd->f_getsendbuf(ppd, dp.pbc_wd, &pbufn);
	if (!piobuf) {
		ret = -EBUSY;
		goto bail;
	}
	/* disarm it just to be extra sure */
	dd->f_sendctrl(dd->pport, QIB_SENDCTRL_DISARM_BUF(pbufn));

	/* disable header check on pbufn for this packet */
	dd->f_txchk_change(dd, pbufn, 1, TXCHK_CHG_TYPE_DIS1, NULL);

	writeq(dp.pbc_wd, piobuf);
	/*
	 * Copy all but the trigger word, then flush, so it's written
	 * to chip before trigger word, then write trigger word, then
	 * flush again, so packet is sent.
	 */
	if (dd->flags & QIB_PIO_FLUSH_WC) {
		qib_flush_wc();
		qib_pio_copy(piobuf + 2, tmpbuf, plen - 1);
		qib_flush_wc();
		__raw_writel(tmpbuf[plen - 1], piobuf + plen + 1);
	} else
		qib_pio_copy(piobuf + 2, tmpbuf, plen);

	if (dd->flags & QIB_USE_SPCL_TRIG) {
		u32 spcl_off = (pbufn >= dd->piobcnt2k) ? 2047 : 1023;

		qib_flush_wc();
		__raw_writel(0xaebecede, piobuf + spcl_off);
	}

	/*
	 * Ensure buffer is written to the chip, then re-enable
	 * header checks (if supported by chip).  The txchk
	 * code will ensure seen by chip before returning.
	 */
	qib_flush_wc();
	qib_sendbuf_done(dd, pbufn);
	dd->f_txchk_change(dd, pbufn, 1, TXCHK_CHG_TYPE_ENAB1, NULL);

	ret = sizeof(dp);

bail:
	vfree(tmpbuf);
	return ret;
}

static int qib_diag_release(struct inode *in, struct file *fp)
{
	mutex_lock(&qib_mutex);
	return_client(fp->private_data);
	fp->private_data = NULL;
	mutex_unlock(&qib_mutex);
	return 0;
}

/*
 * Chip-specific code calls to register its interest in
 * a specific range.
 */
struct diag_observer_list_elt {
	struct diag_observer_list_elt *next;
	const struct diag_observer *op;
};

int qib_register_observer(struct qib_devdata *dd,
			  const struct diag_observer *op)
{
	struct diag_observer_list_elt *olp;
	unsigned long flags;

	if (!dd || !op)
		return -EINVAL;
	olp = vmalloc(sizeof(*olp));
	if (!olp)
		return -ENOMEM;

	spin_lock_irqsave(&dd->qib_diag_trans_lock, flags);
	olp->op = op;
	olp->next = dd->diag_observer_list;
	dd->diag_observer_list = olp;
	spin_unlock_irqrestore(&dd->qib_diag_trans_lock, flags);

	return 0;
}

/* Remove all registered observers when device is closed */
static void qib_unregister_observers(struct qib_devdata *dd)
{
	struct diag_observer_list_elt *olp;
	unsigned long flags;

	spin_lock_irqsave(&dd->qib_diag_trans_lock, flags);
	olp = dd->diag_observer_list;
	while (olp) {
		/* Pop one observer, let go of lock */
		dd->diag_observer_list = olp->next;
		spin_unlock_irqrestore(&dd->qib_diag_trans_lock, flags);
		vfree(olp);
		/* try again. */
		spin_lock_irqsave(&dd->qib_diag_trans_lock, flags);
		olp = dd->diag_observer_list;
	}
	spin_unlock_irqrestore(&dd->qib_diag_trans_lock, flags);
}

/*
 * Find the observer, if any, for the specified address. Initial implementation
 * is simple stack of observers. This must be called with diag transaction
 * lock held.
 */
static const struct diag_observer *diag_get_observer(struct qib_devdata *dd,
						     u32 addr)
{
	struct diag_observer_list_elt *olp;
	const struct diag_observer *op = NULL;

	olp = dd->diag_observer_list;
	while (olp) {
		op = olp->op;
		if (addr >= op->bottom && addr <= op->top)
			break;
		olp = olp->next;
	}
	if (!olp)
		op = NULL;

	return op;
}

static ssize_t qib_diag_read(struct file *fp, char __user *data,
			     size_t count, loff_t *off)
{
	struct qib_diag_client *dc = fp->private_data;
	struct qib_devdata *dd = dc->dd;
	ssize_t ret;

	if (dc->pid != current->pid) {
		ret = -EPERM;
		goto bail;
	}

	if (count == 0)
		ret = 0;
	else if ((count % 4) || (*off % 4))
		/* address or length is not 32-bit aligned, hence invalid */
		ret = -EINVAL;
	else if (dc->state < READY && (*off || count != 8))
		ret = -EINVAL;  /* prevent cat /dev/qib_diag* */
	else {
		unsigned long flags;
		u64 data64 = 0;
		int use_32;
		const struct diag_observer *op;

		use_32 = (count % 8) || (*off % 8);
		ret = -1;
		spin_lock_irqsave(&dd->qib_diag_trans_lock, flags);
		/*
		 * Check for observer on this address range.
		 * we only support a single 32 or 64-bit read
		 * via observer, currently.
		 */
		op = diag_get_observer(dd, *off);
		if (op) {
			u32 offset = *off;

			ret = op->hook(dd, op, offset, &data64, 0, use_32);
		}
		/*
		 * We need to release lock before any copy_to_user(),
		 * whether implicit in qib_read_umem* or explicit below.
		 */
		spin_unlock_irqrestore(&dd->qib_diag_trans_lock, flags);
		if (!op) {
			if (use_32)
				/*
				 * Address or length is not 64-bit aligned;
				 * do 32-bit rd
				 */
				ret = qib_read_umem32(dd, data, (u32) *off,
						      count);
			else
				ret = qib_read_umem64(dd, data, (u32) *off,
						      count);
		} else if (ret == count) {
			/* Below finishes case where observer existed */
			ret = copy_to_user(data, &data64, use_32 ?
					   sizeof(u32) : sizeof(u64));
			if (ret)
				ret = -EFAULT;
		}
	}

	if (ret >= 0) {
		*off += count;
		ret = count;
		if (dc->state == OPENED)
			dc->state = INIT;
	}
bail:
	return ret;
}

static ssize_t qib_diag_write(struct file *fp, const char __user *data,
			      size_t count, loff_t *off)
{
	struct qib_diag_client *dc = fp->private_data;
	struct qib_devdata *dd = dc->dd;
	ssize_t ret;

	if (dc->pid != current->pid) {
		ret = -EPERM;
		goto bail;
	}

	if (count == 0)
		ret = 0;
	else if ((count % 4) || (*off % 4))
		/* address or length is not 32-bit aligned, hence invalid */
		ret = -EINVAL;
	else if (dc->state < READY &&
		((*off || count != 8) || dc->state != INIT))
		/* No writes except second-step of init seq */
		ret = -EINVAL;  /* before any other write allowed */
	else {
		unsigned long flags;
		const struct diag_observer *op = NULL;
		int use_32 =  (count % 8) || (*off % 8);

		/*
		 * Check for observer on this address range.
		 * We only support a single 32 or 64-bit write
		 * via observer, currently. This helps, because
		 * we would otherwise have to jump through hoops
		 * to make "diag transaction" meaningful when we
		 * cannot do a copy_from_user while holding the lock.
		 */
		if (count == 4 || count == 8) {
			u64 data64;
			u32 offset = *off;

			ret = copy_from_user(&data64, data, count);
			if (ret) {
				ret = -EFAULT;
				goto bail;
			}
			spin_lock_irqsave(&dd->qib_diag_trans_lock, flags);
			op = diag_get_observer(dd, *off);
			if (op)
				ret = op->hook(dd, op, offset, &data64, ~0Ull,
					       use_32);
			spin_unlock_irqrestore(&dd->qib_diag_trans_lock, flags);
		}

		if (!op) {
			if (use_32)
				/*
				 * Address or length is not 64-bit aligned;
				 * do 32-bit write
				 */
				ret = qib_write_umem32(dd, (u32) *off, data,
						       count);
			else
				ret = qib_write_umem64(dd, (u32) *off, data,
						       count);
		}
	}

	if (ret >= 0) {
		*off += count;
		ret = count;
		if (dc->state == INIT)
			dc->state = READY; /* all read/write OK now */
	}
bail:
	return ret;
}