xref: /openbmc/linux/drivers/s390/block/dcssblk.c (revision 60772e48)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * dcssblk.c -- the S/390 block driver for dcss memory
4  *
5  * Authors: Carsten Otte, Stefan Weinhuber, Gerald Schaefer
6  */
7 
8 #define KMSG_COMPONENT "dcssblk"
9 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
10 
11 #include <linux/module.h>
12 #include <linux/moduleparam.h>
13 #include <linux/ctype.h>
14 #include <linux/errno.h>
15 #include <linux/init.h>
16 #include <linux/slab.h>
17 #include <linux/blkdev.h>
18 #include <linux/completion.h>
19 #include <linux/interrupt.h>
20 #include <linux/platform_device.h>
21 #include <linux/pfn_t.h>
22 #include <linux/uio.h>
23 #include <linux/dax.h>
24 #include <asm/extmem.h>
25 #include <asm/io.h>
26 
27 #define DCSSBLK_NAME "dcssblk"
28 #define DCSSBLK_MINORS_PER_DISK 1
29 #define DCSSBLK_PARM_LEN 400
30 #define DCSS_BUS_ID_SIZE 20
31 
32 static int dcssblk_open(struct block_device *bdev, fmode_t mode);
33 static void dcssblk_release(struct gendisk *disk, fmode_t mode);
34 static blk_qc_t dcssblk_make_request(struct request_queue *q,
35 						struct bio *bio);
36 static long dcssblk_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
37 		long nr_pages, void **kaddr, pfn_t *pfn);
38 
39 static char dcssblk_segments[DCSSBLK_PARM_LEN] = "\0";
40 
41 static int dcssblk_major;
42 static const struct block_device_operations dcssblk_devops = {
43 	.owner   	= THIS_MODULE,
44 	.open    	= dcssblk_open,
45 	.release 	= dcssblk_release,
46 };
47 
48 static size_t dcssblk_dax_copy_from_iter(struct dax_device *dax_dev,
49 		pgoff_t pgoff, void *addr, size_t bytes, struct iov_iter *i)
50 {
51 	return copy_from_iter(addr, bytes, i);
52 }
53 
54 static const struct dax_operations dcssblk_dax_ops = {
55 	.direct_access = dcssblk_dax_direct_access,
56 	.copy_from_iter = dcssblk_dax_copy_from_iter,
57 };
58 
59 struct dcssblk_dev_info {
60 	struct list_head lh;
61 	struct device dev;
62 	char segment_name[DCSS_BUS_ID_SIZE];
63 	atomic_t use_count;
64 	struct gendisk *gd;
65 	unsigned long start;
66 	unsigned long end;
67 	int segment_type;
68 	unsigned char save_pending;
69 	unsigned char is_shared;
70 	struct request_queue *dcssblk_queue;
71 	int num_of_segments;
72 	struct list_head seg_list;
73 	struct dax_device *dax_dev;
74 };
75 
76 struct segment_info {
77 	struct list_head lh;
78 	char segment_name[DCSS_BUS_ID_SIZE];
79 	unsigned long start;
80 	unsigned long end;
81 	int segment_type;
82 };
83 
84 static ssize_t dcssblk_add_store(struct device * dev, struct device_attribute *attr, const char * buf,
85 				  size_t count);
86 static ssize_t dcssblk_remove_store(struct device * dev, struct device_attribute *attr, const char * buf,
87 				  size_t count);
88 
89 static DEVICE_ATTR(add, S_IWUSR, NULL, dcssblk_add_store);
90 static DEVICE_ATTR(remove, S_IWUSR, NULL, dcssblk_remove_store);
91 
92 static struct device *dcssblk_root_dev;
93 
94 static LIST_HEAD(dcssblk_devices);
95 static struct rw_semaphore dcssblk_devices_sem;
96 
97 /*
98  * release function for segment device.
99  */
100 static void
101 dcssblk_release_segment(struct device *dev)
102 {
103 	struct dcssblk_dev_info *dev_info;
104 	struct segment_info *entry, *temp;
105 
106 	dev_info = container_of(dev, struct dcssblk_dev_info, dev);
107 	list_for_each_entry_safe(entry, temp, &dev_info->seg_list, lh) {
108 		list_del(&entry->lh);
109 		kfree(entry);
110 	}
111 	kfree(dev_info);
112 	module_put(THIS_MODULE);
113 }
114 
115 /*
116  * get a minor number. needs to be called with
117  * down_write(&dcssblk_devices_sem) and the
118  * device needs to be enqueued before the semaphore is
119  * freed.
120  */
121 static int
122 dcssblk_assign_free_minor(struct dcssblk_dev_info *dev_info)
123 {
124 	int minor, found;
125 	struct dcssblk_dev_info *entry;
126 
127 	if (dev_info == NULL)
128 		return -EINVAL;
129 	for (minor = 0; minor < (1<<MINORBITS); minor++) {
130 		found = 0;
131 		// test if minor available
132 		list_for_each_entry(entry, &dcssblk_devices, lh)
133 			if (minor == entry->gd->first_minor)
134 				found++;
135 		if (!found) break; // got unused minor
136 	}
137 	if (found)
138 		return -EBUSY;
139 	dev_info->gd->first_minor = minor;
140 	return 0;
141 }
142 
143 /*
144  * get the struct dcssblk_dev_info from dcssblk_devices
145  * for the given name.
146  * down_read(&dcssblk_devices_sem) must be held.
147  */
148 static struct dcssblk_dev_info *
149 dcssblk_get_device_by_name(char *name)
150 {
151 	struct dcssblk_dev_info *entry;
152 
153 	list_for_each_entry(entry, &dcssblk_devices, lh) {
154 		if (!strcmp(name, entry->segment_name)) {
155 			return entry;
156 		}
157 	}
158 	return NULL;
159 }
160 
161 /*
162  * get the struct segment_info from seg_list
163  * for the given name.
164  * down_read(&dcssblk_devices_sem) must be held.
165  */
166 static struct segment_info *
167 dcssblk_get_segment_by_name(char *name)
168 {
169 	struct dcssblk_dev_info *dev_info;
170 	struct segment_info *entry;
171 
172 	list_for_each_entry(dev_info, &dcssblk_devices, lh) {
173 		list_for_each_entry(entry, &dev_info->seg_list, lh) {
174 			if (!strcmp(name, entry->segment_name))
175 				return entry;
176 		}
177 	}
178 	return NULL;
179 }
180 
181 /*
182  * get the highest address of the multi-segment block.
183  */
184 static unsigned long
185 dcssblk_find_highest_addr(struct dcssblk_dev_info *dev_info)
186 {
187 	unsigned long highest_addr;
188 	struct segment_info *entry;
189 
190 	highest_addr = 0;
191 	list_for_each_entry(entry, &dev_info->seg_list, lh) {
192 		if (highest_addr < entry->end)
193 			highest_addr = entry->end;
194 	}
195 	return highest_addr;
196 }
197 
198 /*
199  * get the lowest address of the multi-segment block.
200  */
201 static unsigned long
202 dcssblk_find_lowest_addr(struct dcssblk_dev_info *dev_info)
203 {
204 	int set_first;
205 	unsigned long lowest_addr;
206 	struct segment_info *entry;
207 
208 	set_first = 0;
209 	lowest_addr = 0;
210 	list_for_each_entry(entry, &dev_info->seg_list, lh) {
211 		if (set_first == 0) {
212 			lowest_addr = entry->start;
213 			set_first = 1;
214 		} else {
215 			if (lowest_addr > entry->start)
216 				lowest_addr = entry->start;
217 		}
218 	}
219 	return lowest_addr;
220 }
221 
222 /*
223  * Check continuity of segments.
224  */
225 static int
226 dcssblk_is_continuous(struct dcssblk_dev_info *dev_info)
227 {
228 	int i, j, rc;
229 	struct segment_info *sort_list, *entry, temp;
230 
231 	if (dev_info->num_of_segments <= 1)
232 		return 0;
233 
234 	sort_list = kzalloc(
235 			sizeof(struct segment_info) * dev_info->num_of_segments,
236 			GFP_KERNEL);
237 	if (sort_list == NULL)
238 		return -ENOMEM;
239 	i = 0;
240 	list_for_each_entry(entry, &dev_info->seg_list, lh) {
241 		memcpy(&sort_list[i], entry, sizeof(struct segment_info));
242 		i++;
243 	}
244 
245 	/* sort segments */
246 	for (i = 0; i < dev_info->num_of_segments; i++)
247 		for (j = 0; j < dev_info->num_of_segments; j++)
248 			if (sort_list[j].start > sort_list[i].start) {
249 				memcpy(&temp, &sort_list[i],
250 					sizeof(struct segment_info));
251 				memcpy(&sort_list[i], &sort_list[j],
252 					sizeof(struct segment_info));
253 				memcpy(&sort_list[j], &temp,
254 					sizeof(struct segment_info));
255 			}
256 
257 	/* check continuity */
258 	for (i = 0; i < dev_info->num_of_segments - 1; i++) {
259 		if ((sort_list[i].end + 1) != sort_list[i+1].start) {
260 			pr_err("Adjacent DCSSs %s and %s are not "
261 			       "contiguous\n", sort_list[i].segment_name,
262 			       sort_list[i+1].segment_name);
263 			rc = -EINVAL;
264 			goto out;
265 		}
266 		/* EN and EW are allowed in a block device */
267 		if (sort_list[i].segment_type != sort_list[i+1].segment_type) {
268 			if (!(sort_list[i].segment_type & SEGMENT_EXCLUSIVE) ||
269 				(sort_list[i].segment_type == SEG_TYPE_ER) ||
270 				!(sort_list[i+1].segment_type &
271 				SEGMENT_EXCLUSIVE) ||
272 				(sort_list[i+1].segment_type == SEG_TYPE_ER)) {
273 				pr_err("DCSS %s and DCSS %s have "
274 				       "incompatible types\n",
275 				       sort_list[i].segment_name,
276 				       sort_list[i+1].segment_name);
277 				rc = -EINVAL;
278 				goto out;
279 			}
280 		}
281 	}
282 	rc = 0;
283 out:
284 	kfree(sort_list);
285 	return rc;
286 }
287 
288 /*
289  * Load a segment
290  */
291 static int
292 dcssblk_load_segment(char *name, struct segment_info **seg_info)
293 {
294 	int rc;
295 
296 	/* already loaded? */
297 	down_read(&dcssblk_devices_sem);
298 	*seg_info = dcssblk_get_segment_by_name(name);
299 	up_read(&dcssblk_devices_sem);
300 	if (*seg_info != NULL)
301 		return -EEXIST;
302 
303 	/* get a struct segment_info */
304 	*seg_info = kzalloc(sizeof(struct segment_info), GFP_KERNEL);
305 	if (*seg_info == NULL)
306 		return -ENOMEM;
307 
308 	strcpy((*seg_info)->segment_name, name);
309 
310 	/* load the segment */
311 	rc = segment_load(name, SEGMENT_SHARED,
312 			&(*seg_info)->start, &(*seg_info)->end);
313 	if (rc < 0) {
314 		segment_warning(rc, (*seg_info)->segment_name);
315 		kfree(*seg_info);
316 	} else {
317 		INIT_LIST_HEAD(&(*seg_info)->lh);
318 		(*seg_info)->segment_type = rc;
319 	}
320 	return rc;
321 }
322 
323 /*
324  * device attribute for switching shared/nonshared (exclusive)
325  * operation (show + store)
326  */
327 static ssize_t
328 dcssblk_shared_show(struct device *dev, struct device_attribute *attr, char *buf)
329 {
330 	struct dcssblk_dev_info *dev_info;
331 
332 	dev_info = container_of(dev, struct dcssblk_dev_info, dev);
333 	return sprintf(buf, dev_info->is_shared ? "1\n" : "0\n");
334 }
335 
336 static ssize_t
337 dcssblk_shared_store(struct device *dev, struct device_attribute *attr, const char *inbuf, size_t count)
338 {
339 	struct dcssblk_dev_info *dev_info;
340 	struct segment_info *entry, *temp;
341 	int rc;
342 
343 	if ((count > 1) && (inbuf[1] != '\n') && (inbuf[1] != '\0'))
344 		return -EINVAL;
345 	down_write(&dcssblk_devices_sem);
346 	dev_info = container_of(dev, struct dcssblk_dev_info, dev);
347 	if (atomic_read(&dev_info->use_count)) {
348 		rc = -EBUSY;
349 		goto out;
350 	}
351 	if (inbuf[0] == '1') {
352 		/* reload segments in shared mode */
353 		list_for_each_entry(entry, &dev_info->seg_list, lh) {
354 			rc = segment_modify_shared(entry->segment_name,
355 						SEGMENT_SHARED);
356 			if (rc < 0) {
357 				BUG_ON(rc == -EINVAL);
358 				if (rc != -EAGAIN)
359 					goto removeseg;
360 			}
361 		}
362 		dev_info->is_shared = 1;
363 		switch (dev_info->segment_type) {
364 		case SEG_TYPE_SR:
365 		case SEG_TYPE_ER:
366 		case SEG_TYPE_SC:
367 			set_disk_ro(dev_info->gd, 1);
368 		}
369 	} else if (inbuf[0] == '0') {
370 		/* reload segments in exclusive mode */
371 		if (dev_info->segment_type == SEG_TYPE_SC) {
372 			pr_err("DCSS %s is of type SC and cannot be "
373 			       "loaded as exclusive-writable\n",
374 			       dev_info->segment_name);
375 			rc = -EINVAL;
376 			goto out;
377 		}
378 		list_for_each_entry(entry, &dev_info->seg_list, lh) {
379 			rc = segment_modify_shared(entry->segment_name,
380 						   SEGMENT_EXCLUSIVE);
381 			if (rc < 0) {
382 				BUG_ON(rc == -EINVAL);
383 				if (rc != -EAGAIN)
384 					goto removeseg;
385 			}
386 		}
387 		dev_info->is_shared = 0;
388 		set_disk_ro(dev_info->gd, 0);
389 	} else {
390 		rc = -EINVAL;
391 		goto out;
392 	}
393 	rc = count;
394 	goto out;
395 
396 removeseg:
397 	pr_err("DCSS device %s is removed after a failed access mode "
398 	       "change\n", dev_info->segment_name);
399 	temp = entry;
400 	list_for_each_entry(entry, &dev_info->seg_list, lh) {
401 		if (entry != temp)
402 			segment_unload(entry->segment_name);
403 	}
404 	list_del(&dev_info->lh);
405 
406 	kill_dax(dev_info->dax_dev);
407 	put_dax(dev_info->dax_dev);
408 	del_gendisk(dev_info->gd);
409 	blk_cleanup_queue(dev_info->dcssblk_queue);
410 	dev_info->gd->queue = NULL;
411 	put_disk(dev_info->gd);
412 	up_write(&dcssblk_devices_sem);
413 
414 	if (device_remove_file_self(dev, attr)) {
415 		device_unregister(dev);
416 		put_device(dev);
417 	}
418 	return rc;
419 out:
420 	up_write(&dcssblk_devices_sem);
421 	return rc;
422 }
423 static DEVICE_ATTR(shared, S_IWUSR | S_IRUSR, dcssblk_shared_show,
424 		   dcssblk_shared_store);
425 
426 /*
427  * device attribute for save operation on current copy
428  * of the segment. If the segment is busy, saving will
429  * become pending until it gets released, which can be
430  * undone by storing a non-true value to this entry.
431  * (show + store)
432  */
433 static ssize_t
434 dcssblk_save_show(struct device *dev, struct device_attribute *attr, char *buf)
435 {
436 	struct dcssblk_dev_info *dev_info;
437 
438 	dev_info = container_of(dev, struct dcssblk_dev_info, dev);
439 	return sprintf(buf, dev_info->save_pending ? "1\n" : "0\n");
440 }
441 
442 static ssize_t
443 dcssblk_save_store(struct device *dev, struct device_attribute *attr, const char *inbuf, size_t count)
444 {
445 	struct dcssblk_dev_info *dev_info;
446 	struct segment_info *entry;
447 
448 	if ((count > 1) && (inbuf[1] != '\n') && (inbuf[1] != '\0'))
449 		return -EINVAL;
450 	dev_info = container_of(dev, struct dcssblk_dev_info, dev);
451 
452 	down_write(&dcssblk_devices_sem);
453 	if (inbuf[0] == '1') {
454 		if (atomic_read(&dev_info->use_count) == 0) {
455 			// device is idle => we save immediately
456 			pr_info("All DCSSs that map to device %s are "
457 				"saved\n", dev_info->segment_name);
458 			list_for_each_entry(entry, &dev_info->seg_list, lh) {
459 				if (entry->segment_type == SEG_TYPE_EN ||
460 				    entry->segment_type == SEG_TYPE_SN)
461 					pr_warn("DCSS %s is of type SN or EN"
462 						" and cannot be saved\n",
463 						entry->segment_name);
464 				else
465 					segment_save(entry->segment_name);
466 			}
467 		}  else {
468 			// device is busy => we save it when it becomes
469 			// idle in dcssblk_release
470 			pr_info("Device %s is in use, its DCSSs will be "
471 				"saved when it becomes idle\n",
472 				dev_info->segment_name);
473 			dev_info->save_pending = 1;
474 		}
475 	} else if (inbuf[0] == '0') {
476 		if (dev_info->save_pending) {
477 			// device is busy & the user wants to undo his save
478 			// request
479 			dev_info->save_pending = 0;
480 			pr_info("A pending save request for device %s "
481 				"has been canceled\n",
482 				dev_info->segment_name);
483 		}
484 	} else {
485 		up_write(&dcssblk_devices_sem);
486 		return -EINVAL;
487 	}
488 	up_write(&dcssblk_devices_sem);
489 	return count;
490 }
491 static DEVICE_ATTR(save, S_IWUSR | S_IRUSR, dcssblk_save_show,
492 		   dcssblk_save_store);
493 
494 /*
495  * device attribute for showing all segments in a device
496  */
497 static ssize_t
498 dcssblk_seglist_show(struct device *dev, struct device_attribute *attr,
499 		char *buf)
500 {
501 	int i;
502 
503 	struct dcssblk_dev_info *dev_info;
504 	struct segment_info *entry;
505 
506 	down_read(&dcssblk_devices_sem);
507 	dev_info = container_of(dev, struct dcssblk_dev_info, dev);
508 	i = 0;
509 	buf[0] = '\0';
510 	list_for_each_entry(entry, &dev_info->seg_list, lh) {
511 		strcpy(&buf[i], entry->segment_name);
512 		i += strlen(entry->segment_name);
513 		buf[i] = '\n';
514 		i++;
515 	}
516 	up_read(&dcssblk_devices_sem);
517 	return i;
518 }
519 static DEVICE_ATTR(seglist, S_IRUSR, dcssblk_seglist_show, NULL);
520 
521 static struct attribute *dcssblk_dev_attrs[] = {
522 	&dev_attr_shared.attr,
523 	&dev_attr_save.attr,
524 	&dev_attr_seglist.attr,
525 	NULL,
526 };
527 static struct attribute_group dcssblk_dev_attr_group = {
528 	.attrs = dcssblk_dev_attrs,
529 };
530 static const struct attribute_group *dcssblk_dev_attr_groups[] = {
531 	&dcssblk_dev_attr_group,
532 	NULL,
533 };
534 
535 /*
536  * device attribute for adding devices
537  */
538 static ssize_t
539 dcssblk_add_store(struct device *dev, struct device_attribute *attr, const char *buf, size_t count)
540 {
541 	int rc, i, j, num_of_segments;
542 	struct dcssblk_dev_info *dev_info;
543 	struct segment_info *seg_info, *temp;
544 	char *local_buf;
545 	unsigned long seg_byte_size;
546 
547 	dev_info = NULL;
548 	seg_info = NULL;
549 	if (dev != dcssblk_root_dev) {
550 		rc = -EINVAL;
551 		goto out_nobuf;
552 	}
553 	if ((count < 1) || (buf[0] == '\0') || (buf[0] == '\n')) {
554 		rc = -ENAMETOOLONG;
555 		goto out_nobuf;
556 	}
557 
558 	local_buf = kmalloc(count + 1, GFP_KERNEL);
559 	if (local_buf == NULL) {
560 		rc = -ENOMEM;
561 		goto out_nobuf;
562 	}
563 
564 	/*
565 	 * parse input
566 	 */
567 	num_of_segments = 0;
568 	for (i = 0; (i < count && (buf[i] != '\0') && (buf[i] != '\n')); i++) {
569 		for (j = i; j < count &&
570 			(buf[j] != ':') &&
571 			(buf[j] != '\0') &&
572 			(buf[j] != '\n'); j++) {
573 			local_buf[j-i] = toupper(buf[j]);
574 		}
575 		local_buf[j-i] = '\0';
576 		if (((j - i) == 0) || ((j - i) > 8)) {
577 			rc = -ENAMETOOLONG;
578 			goto seg_list_del;
579 		}
580 
581 		rc = dcssblk_load_segment(local_buf, &seg_info);
582 		if (rc < 0)
583 			goto seg_list_del;
584 		/*
585 		 * get a struct dcssblk_dev_info
586 		 */
587 		if (num_of_segments == 0) {
588 			dev_info = kzalloc(sizeof(struct dcssblk_dev_info),
589 					GFP_KERNEL);
590 			if (dev_info == NULL) {
591 				rc = -ENOMEM;
592 				goto out;
593 			}
594 			strcpy(dev_info->segment_name, local_buf);
595 			dev_info->segment_type = seg_info->segment_type;
596 			INIT_LIST_HEAD(&dev_info->seg_list);
597 		}
598 		list_add_tail(&seg_info->lh, &dev_info->seg_list);
599 		num_of_segments++;
600 		i = j;
601 
602 		if ((buf[j] == '\0') || (buf[j] == '\n'))
603 			break;
604 	}
605 
606 	/* no trailing colon at the end of the input */
607 	if ((i > 0) && (buf[i-1] == ':')) {
608 		rc = -ENAMETOOLONG;
609 		goto seg_list_del;
610 	}
611 	strlcpy(local_buf, buf, i + 1);
612 	dev_info->num_of_segments = num_of_segments;
613 	rc = dcssblk_is_continuous(dev_info);
614 	if (rc < 0)
615 		goto seg_list_del;
616 
617 	dev_info->start = dcssblk_find_lowest_addr(dev_info);
618 	dev_info->end = dcssblk_find_highest_addr(dev_info);
619 
620 	dev_set_name(&dev_info->dev, "%s", dev_info->segment_name);
621 	dev_info->dev.release = dcssblk_release_segment;
622 	dev_info->dev.groups = dcssblk_dev_attr_groups;
623 	INIT_LIST_HEAD(&dev_info->lh);
624 	dev_info->gd = alloc_disk(DCSSBLK_MINORS_PER_DISK);
625 	if (dev_info->gd == NULL) {
626 		rc = -ENOMEM;
627 		goto seg_list_del;
628 	}
629 	dev_info->gd->major = dcssblk_major;
630 	dev_info->gd->fops = &dcssblk_devops;
631 	dev_info->dcssblk_queue = blk_alloc_queue(GFP_KERNEL);
632 	dev_info->gd->queue = dev_info->dcssblk_queue;
633 	dev_info->gd->private_data = dev_info;
634 	blk_queue_make_request(dev_info->dcssblk_queue, dcssblk_make_request);
635 	blk_queue_logical_block_size(dev_info->dcssblk_queue, 4096);
636 	queue_flag_set_unlocked(QUEUE_FLAG_DAX, dev_info->dcssblk_queue);
637 
638 	seg_byte_size = (dev_info->end - dev_info->start + 1);
639 	set_capacity(dev_info->gd, seg_byte_size >> 9); // size in sectors
640 	pr_info("Loaded %s with total size %lu bytes and capacity %lu "
641 		"sectors\n", local_buf, seg_byte_size, seg_byte_size >> 9);
642 
643 	dev_info->save_pending = 0;
644 	dev_info->is_shared = 1;
645 	dev_info->dev.parent = dcssblk_root_dev;
646 
647 	/*
648 	 *get minor, add to list
649 	 */
650 	down_write(&dcssblk_devices_sem);
651 	if (dcssblk_get_segment_by_name(local_buf)) {
652 		rc = -EEXIST;
653 		goto release_gd;
654 	}
655 	rc = dcssblk_assign_free_minor(dev_info);
656 	if (rc)
657 		goto release_gd;
658 	sprintf(dev_info->gd->disk_name, "dcssblk%d",
659 		dev_info->gd->first_minor);
660 	list_add_tail(&dev_info->lh, &dcssblk_devices);
661 
662 	if (!try_module_get(THIS_MODULE)) {
663 		rc = -ENODEV;
664 		goto dev_list_del;
665 	}
666 	/*
667 	 * register the device
668 	 */
669 	rc = device_register(&dev_info->dev);
670 	if (rc)
671 		goto put_dev;
672 
673 	dev_info->dax_dev = alloc_dax(dev_info, dev_info->gd->disk_name,
674 			&dcssblk_dax_ops);
675 	if (!dev_info->dax_dev) {
676 		rc = -ENOMEM;
677 		goto put_dev;
678 	}
679 
680 	get_device(&dev_info->dev);
681 	device_add_disk(&dev_info->dev, dev_info->gd);
682 
683 	switch (dev_info->segment_type) {
684 		case SEG_TYPE_SR:
685 		case SEG_TYPE_ER:
686 		case SEG_TYPE_SC:
687 			set_disk_ro(dev_info->gd,1);
688 			break;
689 		default:
690 			set_disk_ro(dev_info->gd,0);
691 			break;
692 	}
693 	up_write(&dcssblk_devices_sem);
694 	rc = count;
695 	goto out;
696 
697 put_dev:
698 	list_del(&dev_info->lh);
699 	blk_cleanup_queue(dev_info->dcssblk_queue);
700 	dev_info->gd->queue = NULL;
701 	put_disk(dev_info->gd);
702 	list_for_each_entry(seg_info, &dev_info->seg_list, lh) {
703 		segment_unload(seg_info->segment_name);
704 	}
705 	put_device(&dev_info->dev);
706 	up_write(&dcssblk_devices_sem);
707 	goto out;
708 dev_list_del:
709 	list_del(&dev_info->lh);
710 release_gd:
711 	blk_cleanup_queue(dev_info->dcssblk_queue);
712 	dev_info->gd->queue = NULL;
713 	put_disk(dev_info->gd);
714 	up_write(&dcssblk_devices_sem);
715 seg_list_del:
716 	if (dev_info == NULL)
717 		goto out;
718 	list_for_each_entry_safe(seg_info, temp, &dev_info->seg_list, lh) {
719 		list_del(&seg_info->lh);
720 		segment_unload(seg_info->segment_name);
721 		kfree(seg_info);
722 	}
723 	kfree(dev_info);
724 out:
725 	kfree(local_buf);
726 out_nobuf:
727 	return rc;
728 }
729 
730 /*
731  * device attribute for removing devices
732  */
733 static ssize_t
734 dcssblk_remove_store(struct device *dev, struct device_attribute *attr, const char *buf, size_t count)
735 {
736 	struct dcssblk_dev_info *dev_info;
737 	struct segment_info *entry;
738 	int rc, i;
739 	char *local_buf;
740 
741 	if (dev != dcssblk_root_dev) {
742 		return -EINVAL;
743 	}
744 	local_buf = kmalloc(count + 1, GFP_KERNEL);
745 	if (local_buf == NULL) {
746 		return -ENOMEM;
747 	}
748 	/*
749 	 * parse input
750 	 */
751 	for (i = 0; (i < count && (*(buf+i)!='\0') && (*(buf+i)!='\n')); i++) {
752 		local_buf[i] = toupper(buf[i]);
753 	}
754 	local_buf[i] = '\0';
755 	if ((i == 0) || (i > 8)) {
756 		rc = -ENAMETOOLONG;
757 		goto out_buf;
758 	}
759 
760 	down_write(&dcssblk_devices_sem);
761 	dev_info = dcssblk_get_device_by_name(local_buf);
762 	if (dev_info == NULL) {
763 		up_write(&dcssblk_devices_sem);
764 		pr_warn("Device %s cannot be removed because it is not a known device\n",
765 			local_buf);
766 		rc = -ENODEV;
767 		goto out_buf;
768 	}
769 	if (atomic_read(&dev_info->use_count) != 0) {
770 		up_write(&dcssblk_devices_sem);
771 		pr_warn("Device %s cannot be removed while it is in use\n",
772 			local_buf);
773 		rc = -EBUSY;
774 		goto out_buf;
775 	}
776 
777 	list_del(&dev_info->lh);
778 	kill_dax(dev_info->dax_dev);
779 	put_dax(dev_info->dax_dev);
780 	del_gendisk(dev_info->gd);
781 	blk_cleanup_queue(dev_info->dcssblk_queue);
782 	dev_info->gd->queue = NULL;
783 	put_disk(dev_info->gd);
784 
785 	/* unload all related segments */
786 	list_for_each_entry(entry, &dev_info->seg_list, lh)
787 		segment_unload(entry->segment_name);
788 
789 	up_write(&dcssblk_devices_sem);
790 
791 	device_unregister(&dev_info->dev);
792 	put_device(&dev_info->dev);
793 
794 	rc = count;
795 out_buf:
796 	kfree(local_buf);
797 	return rc;
798 }
799 
800 static int
801 dcssblk_open(struct block_device *bdev, fmode_t mode)
802 {
803 	struct dcssblk_dev_info *dev_info;
804 	int rc;
805 
806 	dev_info = bdev->bd_disk->private_data;
807 	if (NULL == dev_info) {
808 		rc = -ENODEV;
809 		goto out;
810 	}
811 	atomic_inc(&dev_info->use_count);
812 	bdev->bd_block_size = 4096;
813 	rc = 0;
814 out:
815 	return rc;
816 }
817 
818 static void
819 dcssblk_release(struct gendisk *disk, fmode_t mode)
820 {
821 	struct dcssblk_dev_info *dev_info = disk->private_data;
822 	struct segment_info *entry;
823 
824 	if (!dev_info) {
825 		WARN_ON(1);
826 		return;
827 	}
828 	down_write(&dcssblk_devices_sem);
829 	if (atomic_dec_and_test(&dev_info->use_count)
830 	    && (dev_info->save_pending)) {
831 		pr_info("Device %s has become idle and is being saved "
832 			"now\n", dev_info->segment_name);
833 		list_for_each_entry(entry, &dev_info->seg_list, lh) {
834 			if (entry->segment_type == SEG_TYPE_EN ||
835 			    entry->segment_type == SEG_TYPE_SN)
836 				pr_warn("DCSS %s is of type SN or EN and cannot"
837 					" be saved\n", entry->segment_name);
838 			else
839 				segment_save(entry->segment_name);
840 		}
841 		dev_info->save_pending = 0;
842 	}
843 	up_write(&dcssblk_devices_sem);
844 }
845 
846 static blk_qc_t
847 dcssblk_make_request(struct request_queue *q, struct bio *bio)
848 {
849 	struct dcssblk_dev_info *dev_info;
850 	struct bio_vec bvec;
851 	struct bvec_iter iter;
852 	unsigned long index;
853 	unsigned long page_addr;
854 	unsigned long source_addr;
855 	unsigned long bytes_done;
856 
857 	blk_queue_split(q, &bio);
858 
859 	bytes_done = 0;
860 	dev_info = bio->bi_disk->private_data;
861 	if (dev_info == NULL)
862 		goto fail;
863 	if ((bio->bi_iter.bi_sector & 7) != 0 ||
864 	    (bio->bi_iter.bi_size & 4095) != 0)
865 		/* Request is not page-aligned. */
866 		goto fail;
867 	if (bio_end_sector(bio) > get_capacity(bio->bi_disk)) {
868 		/* Request beyond end of DCSS segment. */
869 		goto fail;
870 	}
871 	/* verify data transfer direction */
872 	if (dev_info->is_shared) {
873 		switch (dev_info->segment_type) {
874 		case SEG_TYPE_SR:
875 		case SEG_TYPE_ER:
876 		case SEG_TYPE_SC:
877 			/* cannot write to these segments */
878 			if (bio_data_dir(bio) == WRITE) {
879 				pr_warn("Writing to %s failed because it is a read-only device\n",
880 					dev_name(&dev_info->dev));
881 				goto fail;
882 			}
883 		}
884 	}
885 
886 	index = (bio->bi_iter.bi_sector >> 3);
887 	bio_for_each_segment(bvec, bio, iter) {
888 		page_addr = (unsigned long)
889 			page_address(bvec.bv_page) + bvec.bv_offset;
890 		source_addr = dev_info->start + (index<<12) + bytes_done;
891 		if (unlikely((page_addr & 4095) != 0) || (bvec.bv_len & 4095) != 0)
892 			// More paranoia.
893 			goto fail;
894 		if (bio_data_dir(bio) == READ) {
895 			memcpy((void*)page_addr, (void*)source_addr,
896 				bvec.bv_len);
897 		} else {
898 			memcpy((void*)source_addr, (void*)page_addr,
899 				bvec.bv_len);
900 		}
901 		bytes_done += bvec.bv_len;
902 	}
903 	bio_endio(bio);
904 	return BLK_QC_T_NONE;
905 fail:
906 	bio_io_error(bio);
907 	return BLK_QC_T_NONE;
908 }
909 
910 static long
911 __dcssblk_direct_access(struct dcssblk_dev_info *dev_info, pgoff_t pgoff,
912 		long nr_pages, void **kaddr, pfn_t *pfn)
913 {
914 	resource_size_t offset = pgoff * PAGE_SIZE;
915 	unsigned long dev_sz;
916 
917 	dev_sz = dev_info->end - dev_info->start + 1;
918 	*kaddr = (void *) dev_info->start + offset;
919 	*pfn = __pfn_to_pfn_t(PFN_DOWN(dev_info->start + offset),
920 			PFN_DEV|PFN_SPECIAL);
921 
922 	return (dev_sz - offset) / PAGE_SIZE;
923 }
924 
925 static long
926 dcssblk_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
927 		long nr_pages, void **kaddr, pfn_t *pfn)
928 {
929 	struct dcssblk_dev_info *dev_info = dax_get_private(dax_dev);
930 
931 	return __dcssblk_direct_access(dev_info, pgoff, nr_pages, kaddr, pfn);
932 }
933 
934 static void
935 dcssblk_check_params(void)
936 {
937 	int rc, i, j, k;
938 	char buf[DCSSBLK_PARM_LEN + 1];
939 	struct dcssblk_dev_info *dev_info;
940 
941 	for (i = 0; (i < DCSSBLK_PARM_LEN) && (dcssblk_segments[i] != '\0');
942 	     i++) {
943 		for (j = i; (j < DCSSBLK_PARM_LEN) &&
944 			    (dcssblk_segments[j] != ',')  &&
945 			    (dcssblk_segments[j] != '\0') &&
946 			    (dcssblk_segments[j] != '('); j++)
947 		{
948 			buf[j-i] = dcssblk_segments[j];
949 		}
950 		buf[j-i] = '\0';
951 		rc = dcssblk_add_store(dcssblk_root_dev, NULL, buf, j-i);
952 		if ((rc >= 0) && (dcssblk_segments[j] == '(')) {
953 			for (k = 0; (buf[k] != ':') && (buf[k] != '\0'); k++)
954 				buf[k] = toupper(buf[k]);
955 			buf[k] = '\0';
956 			if (!strncmp(&dcssblk_segments[j], "(local)", 7)) {
957 				down_read(&dcssblk_devices_sem);
958 				dev_info = dcssblk_get_device_by_name(buf);
959 				up_read(&dcssblk_devices_sem);
960 				if (dev_info)
961 					dcssblk_shared_store(&dev_info->dev,
962 							     NULL, "0\n", 2);
963 			}
964 		}
965 		while ((dcssblk_segments[j] != ',') &&
966 		       (dcssblk_segments[j] != '\0'))
967 		{
968 			j++;
969 		}
970 		if (dcssblk_segments[j] == '\0')
971 			break;
972 		i = j;
973 	}
974 }
975 
976 /*
977  * Suspend / Resume
978  */
979 static int dcssblk_freeze(struct device *dev)
980 {
981 	struct dcssblk_dev_info *dev_info;
982 	int rc = 0;
983 
984 	list_for_each_entry(dev_info, &dcssblk_devices, lh) {
985 		switch (dev_info->segment_type) {
986 			case SEG_TYPE_SR:
987 			case SEG_TYPE_ER:
988 			case SEG_TYPE_SC:
989 				if (!dev_info->is_shared)
990 					rc = -EINVAL;
991 				break;
992 			default:
993 				rc = -EINVAL;
994 				break;
995 		}
996 		if (rc)
997 			break;
998 	}
999 	if (rc)
1000 		pr_err("Suspending the system failed because DCSS device %s "
1001 		       "is writable\n",
1002 		       dev_info->segment_name);
1003 	return rc;
1004 }
1005 
1006 static int dcssblk_restore(struct device *dev)
1007 {
1008 	struct dcssblk_dev_info *dev_info;
1009 	struct segment_info *entry;
1010 	unsigned long start, end;
1011 	int rc = 0;
1012 
1013 	list_for_each_entry(dev_info, &dcssblk_devices, lh) {
1014 		list_for_each_entry(entry, &dev_info->seg_list, lh) {
1015 			segment_unload(entry->segment_name);
1016 			rc = segment_load(entry->segment_name, SEGMENT_SHARED,
1017 					  &start, &end);
1018 			if (rc < 0) {
1019 // TODO in_use check ?
1020 				segment_warning(rc, entry->segment_name);
1021 				goto out_panic;
1022 			}
1023 			if (start != entry->start || end != entry->end) {
1024 				pr_err("The address range of DCSS %s changed "
1025 				       "while the system was suspended\n",
1026 				       entry->segment_name);
1027 				goto out_panic;
1028 			}
1029 		}
1030 	}
1031 	return 0;
1032 out_panic:
1033 	panic("fatal dcssblk resume error\n");
1034 }
1035 
1036 static int dcssblk_thaw(struct device *dev)
1037 {
1038 	return 0;
1039 }
1040 
1041 static const struct dev_pm_ops dcssblk_pm_ops = {
1042 	.freeze		= dcssblk_freeze,
1043 	.thaw		= dcssblk_thaw,
1044 	.restore	= dcssblk_restore,
1045 };
1046 
1047 static struct platform_driver dcssblk_pdrv = {
1048 	.driver = {
1049 		.name	= "dcssblk",
1050 		.pm	= &dcssblk_pm_ops,
1051 	},
1052 };
1053 
1054 static struct platform_device *dcssblk_pdev;
1055 
1056 
1057 /*
1058  * The init/exit functions.
1059  */
1060 static void __exit
1061 dcssblk_exit(void)
1062 {
1063 	platform_device_unregister(dcssblk_pdev);
1064 	platform_driver_unregister(&dcssblk_pdrv);
1065 	root_device_unregister(dcssblk_root_dev);
1066 	unregister_blkdev(dcssblk_major, DCSSBLK_NAME);
1067 }
1068 
1069 static int __init
1070 dcssblk_init(void)
1071 {
1072 	int rc;
1073 
1074 	rc = platform_driver_register(&dcssblk_pdrv);
1075 	if (rc)
1076 		return rc;
1077 
1078 	dcssblk_pdev = platform_device_register_simple("dcssblk", -1, NULL,
1079 							0);
1080 	if (IS_ERR(dcssblk_pdev)) {
1081 		rc = PTR_ERR(dcssblk_pdev);
1082 		goto out_pdrv;
1083 	}
1084 
1085 	dcssblk_root_dev = root_device_register("dcssblk");
1086 	if (IS_ERR(dcssblk_root_dev)) {
1087 		rc = PTR_ERR(dcssblk_root_dev);
1088 		goto out_pdev;
1089 	}
1090 	rc = device_create_file(dcssblk_root_dev, &dev_attr_add);
1091 	if (rc)
1092 		goto out_root;
1093 	rc = device_create_file(dcssblk_root_dev, &dev_attr_remove);
1094 	if (rc)
1095 		goto out_root;
1096 	rc = register_blkdev(0, DCSSBLK_NAME);
1097 	if (rc < 0)
1098 		goto out_root;
1099 	dcssblk_major = rc;
1100 	init_rwsem(&dcssblk_devices_sem);
1101 
1102 	dcssblk_check_params();
1103 	return 0;
1104 
1105 out_root:
1106 	root_device_unregister(dcssblk_root_dev);
1107 out_pdev:
1108 	platform_device_unregister(dcssblk_pdev);
1109 out_pdrv:
1110 	platform_driver_unregister(&dcssblk_pdrv);
1111 	return rc;
1112 }
1113 
1114 module_init(dcssblk_init);
1115 module_exit(dcssblk_exit);
1116 
1117 module_param_string(segments, dcssblk_segments, DCSSBLK_PARM_LEN, 0444);
1118 MODULE_PARM_DESC(segments, "Name of DCSS segment(s) to be loaded, "
1119 		 "comma-separated list, names in each set separated "
1120 		 "by commas are separated by colons, each set contains "
1121 		 "names of contiguous segments and each name max. 8 chars.\n"
1122 		 "Adding \"(local)\" to the end of each set equals echoing 0 "
1123 		 "to /sys/devices/dcssblk/<device name>/shared after loading "
1124 		 "the contiguous segments - \n"
1125 		 "e.g. segments=\"mydcss1,mydcss2:mydcss3,mydcss4(local)\"");
1126 
1127 MODULE_LICENSE("GPL");
1128