xref: /linux/drivers/s390/block/dcssblk.c (revision 02680c23d7b3febe45ea3d4f9818c2b2dc89020a)
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_submit_bio(struct bio *bio);
35 static long dcssblk_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
36 		long nr_pages, void **kaddr, pfn_t *pfn);
37 
38 static char dcssblk_segments[DCSSBLK_PARM_LEN] = "\0";
39 
40 static int dcssblk_major;
41 static const struct block_device_operations dcssblk_devops = {
42 	.owner   	= THIS_MODULE,
43 	.submit_bio	= dcssblk_submit_bio,
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 size_t dcssblk_dax_copy_to_iter(struct dax_device *dax_dev,
55 		pgoff_t pgoff, void *addr, size_t bytes, struct iov_iter *i)
56 {
57 	return copy_to_iter(addr, bytes, i);
58 }
59 
60 static int dcssblk_dax_zero_page_range(struct dax_device *dax_dev,
61 				       pgoff_t pgoff, size_t nr_pages)
62 {
63 	long rc;
64 	void *kaddr;
65 
66 	rc = dax_direct_access(dax_dev, pgoff, nr_pages, &kaddr, NULL);
67 	if (rc < 0)
68 		return rc;
69 	memset(kaddr, 0, nr_pages << PAGE_SHIFT);
70 	dax_flush(dax_dev, kaddr, nr_pages << PAGE_SHIFT);
71 	return 0;
72 }
73 
74 static const struct dax_operations dcssblk_dax_ops = {
75 	.direct_access = dcssblk_dax_direct_access,
76 	.dax_supported = generic_fsdax_supported,
77 	.copy_from_iter = dcssblk_dax_copy_from_iter,
78 	.copy_to_iter = dcssblk_dax_copy_to_iter,
79 	.zero_page_range = dcssblk_dax_zero_page_range,
80 };
81 
82 struct dcssblk_dev_info {
83 	struct list_head lh;
84 	struct device dev;
85 	char segment_name[DCSS_BUS_ID_SIZE];
86 	atomic_t use_count;
87 	struct gendisk *gd;
88 	unsigned long start;
89 	unsigned long end;
90 	int segment_type;
91 	unsigned char save_pending;
92 	unsigned char is_shared;
93 	struct request_queue *dcssblk_queue;
94 	int num_of_segments;
95 	struct list_head seg_list;
96 	struct dax_device *dax_dev;
97 };
98 
99 struct segment_info {
100 	struct list_head lh;
101 	char segment_name[DCSS_BUS_ID_SIZE];
102 	unsigned long start;
103 	unsigned long end;
104 	int segment_type;
105 };
106 
107 static ssize_t dcssblk_add_store(struct device * dev, struct device_attribute *attr, const char * buf,
108 				  size_t count);
109 static ssize_t dcssblk_remove_store(struct device * dev, struct device_attribute *attr, const char * buf,
110 				  size_t count);
111 
112 static DEVICE_ATTR(add, S_IWUSR, NULL, dcssblk_add_store);
113 static DEVICE_ATTR(remove, S_IWUSR, NULL, dcssblk_remove_store);
114 
115 static struct device *dcssblk_root_dev;
116 
117 static LIST_HEAD(dcssblk_devices);
118 static struct rw_semaphore dcssblk_devices_sem;
119 
120 /*
121  * release function for segment device.
122  */
123 static void
124 dcssblk_release_segment(struct device *dev)
125 {
126 	struct dcssblk_dev_info *dev_info;
127 	struct segment_info *entry, *temp;
128 
129 	dev_info = container_of(dev, struct dcssblk_dev_info, dev);
130 	list_for_each_entry_safe(entry, temp, &dev_info->seg_list, lh) {
131 		list_del(&entry->lh);
132 		kfree(entry);
133 	}
134 	kfree(dev_info);
135 	module_put(THIS_MODULE);
136 }
137 
138 /*
139  * get a minor number. needs to be called with
140  * down_write(&dcssblk_devices_sem) and the
141  * device needs to be enqueued before the semaphore is
142  * freed.
143  */
144 static int
145 dcssblk_assign_free_minor(struct dcssblk_dev_info *dev_info)
146 {
147 	int minor, found;
148 	struct dcssblk_dev_info *entry;
149 
150 	if (dev_info == NULL)
151 		return -EINVAL;
152 	for (minor = 0; minor < (1<<MINORBITS); minor++) {
153 		found = 0;
154 		// test if minor available
155 		list_for_each_entry(entry, &dcssblk_devices, lh)
156 			if (minor == entry->gd->first_minor)
157 				found++;
158 		if (!found) break; // got unused minor
159 	}
160 	if (found)
161 		return -EBUSY;
162 	dev_info->gd->first_minor = minor;
163 	return 0;
164 }
165 
166 /*
167  * get the struct dcssblk_dev_info from dcssblk_devices
168  * for the given name.
169  * down_read(&dcssblk_devices_sem) must be held.
170  */
171 static struct dcssblk_dev_info *
172 dcssblk_get_device_by_name(char *name)
173 {
174 	struct dcssblk_dev_info *entry;
175 
176 	list_for_each_entry(entry, &dcssblk_devices, lh) {
177 		if (!strcmp(name, entry->segment_name)) {
178 			return entry;
179 		}
180 	}
181 	return NULL;
182 }
183 
184 /*
185  * get the struct segment_info from seg_list
186  * for the given name.
187  * down_read(&dcssblk_devices_sem) must be held.
188  */
189 static struct segment_info *
190 dcssblk_get_segment_by_name(char *name)
191 {
192 	struct dcssblk_dev_info *dev_info;
193 	struct segment_info *entry;
194 
195 	list_for_each_entry(dev_info, &dcssblk_devices, lh) {
196 		list_for_each_entry(entry, &dev_info->seg_list, lh) {
197 			if (!strcmp(name, entry->segment_name))
198 				return entry;
199 		}
200 	}
201 	return NULL;
202 }
203 
204 /*
205  * get the highest address of the multi-segment block.
206  */
207 static unsigned long
208 dcssblk_find_highest_addr(struct dcssblk_dev_info *dev_info)
209 {
210 	unsigned long highest_addr;
211 	struct segment_info *entry;
212 
213 	highest_addr = 0;
214 	list_for_each_entry(entry, &dev_info->seg_list, lh) {
215 		if (highest_addr < entry->end)
216 			highest_addr = entry->end;
217 	}
218 	return highest_addr;
219 }
220 
221 /*
222  * get the lowest address of the multi-segment block.
223  */
224 static unsigned long
225 dcssblk_find_lowest_addr(struct dcssblk_dev_info *dev_info)
226 {
227 	int set_first;
228 	unsigned long lowest_addr;
229 	struct segment_info *entry;
230 
231 	set_first = 0;
232 	lowest_addr = 0;
233 	list_for_each_entry(entry, &dev_info->seg_list, lh) {
234 		if (set_first == 0) {
235 			lowest_addr = entry->start;
236 			set_first = 1;
237 		} else {
238 			if (lowest_addr > entry->start)
239 				lowest_addr = entry->start;
240 		}
241 	}
242 	return lowest_addr;
243 }
244 
245 /*
246  * Check continuity of segments.
247  */
248 static int
249 dcssblk_is_continuous(struct dcssblk_dev_info *dev_info)
250 {
251 	int i, j, rc;
252 	struct segment_info *sort_list, *entry, temp;
253 
254 	if (dev_info->num_of_segments <= 1)
255 		return 0;
256 
257 	sort_list = kcalloc(dev_info->num_of_segments,
258 			    sizeof(struct segment_info),
259 			    GFP_KERNEL);
260 	if (sort_list == NULL)
261 		return -ENOMEM;
262 	i = 0;
263 	list_for_each_entry(entry, &dev_info->seg_list, lh) {
264 		memcpy(&sort_list[i], entry, sizeof(struct segment_info));
265 		i++;
266 	}
267 
268 	/* sort segments */
269 	for (i = 0; i < dev_info->num_of_segments; i++)
270 		for (j = 0; j < dev_info->num_of_segments; j++)
271 			if (sort_list[j].start > sort_list[i].start) {
272 				memcpy(&temp, &sort_list[i],
273 					sizeof(struct segment_info));
274 				memcpy(&sort_list[i], &sort_list[j],
275 					sizeof(struct segment_info));
276 				memcpy(&sort_list[j], &temp,
277 					sizeof(struct segment_info));
278 			}
279 
280 	/* check continuity */
281 	for (i = 0; i < dev_info->num_of_segments - 1; i++) {
282 		if ((sort_list[i].end + 1) != sort_list[i+1].start) {
283 			pr_err("Adjacent DCSSs %s and %s are not "
284 			       "contiguous\n", sort_list[i].segment_name,
285 			       sort_list[i+1].segment_name);
286 			rc = -EINVAL;
287 			goto out;
288 		}
289 		/* EN and EW are allowed in a block device */
290 		if (sort_list[i].segment_type != sort_list[i+1].segment_type) {
291 			if (!(sort_list[i].segment_type & SEGMENT_EXCLUSIVE) ||
292 				(sort_list[i].segment_type == SEG_TYPE_ER) ||
293 				!(sort_list[i+1].segment_type &
294 				SEGMENT_EXCLUSIVE) ||
295 				(sort_list[i+1].segment_type == SEG_TYPE_ER)) {
296 				pr_err("DCSS %s and DCSS %s have "
297 				       "incompatible types\n",
298 				       sort_list[i].segment_name,
299 				       sort_list[i+1].segment_name);
300 				rc = -EINVAL;
301 				goto out;
302 			}
303 		}
304 	}
305 	rc = 0;
306 out:
307 	kfree(sort_list);
308 	return rc;
309 }
310 
311 /*
312  * Load a segment
313  */
314 static int
315 dcssblk_load_segment(char *name, struct segment_info **seg_info)
316 {
317 	int rc;
318 
319 	/* already loaded? */
320 	down_read(&dcssblk_devices_sem);
321 	*seg_info = dcssblk_get_segment_by_name(name);
322 	up_read(&dcssblk_devices_sem);
323 	if (*seg_info != NULL)
324 		return -EEXIST;
325 
326 	/* get a struct segment_info */
327 	*seg_info = kzalloc(sizeof(struct segment_info), GFP_KERNEL);
328 	if (*seg_info == NULL)
329 		return -ENOMEM;
330 
331 	strcpy((*seg_info)->segment_name, name);
332 
333 	/* load the segment */
334 	rc = segment_load(name, SEGMENT_SHARED,
335 			&(*seg_info)->start, &(*seg_info)->end);
336 	if (rc < 0) {
337 		segment_warning(rc, (*seg_info)->segment_name);
338 		kfree(*seg_info);
339 	} else {
340 		INIT_LIST_HEAD(&(*seg_info)->lh);
341 		(*seg_info)->segment_type = rc;
342 	}
343 	return rc;
344 }
345 
346 /*
347  * device attribute for switching shared/nonshared (exclusive)
348  * operation (show + store)
349  */
350 static ssize_t
351 dcssblk_shared_show(struct device *dev, struct device_attribute *attr, char *buf)
352 {
353 	struct dcssblk_dev_info *dev_info;
354 
355 	dev_info = container_of(dev, struct dcssblk_dev_info, dev);
356 	return sprintf(buf, dev_info->is_shared ? "1\n" : "0\n");
357 }
358 
359 static ssize_t
360 dcssblk_shared_store(struct device *dev, struct device_attribute *attr, const char *inbuf, size_t count)
361 {
362 	struct dcssblk_dev_info *dev_info;
363 	struct segment_info *entry, *temp;
364 	int rc;
365 
366 	if ((count > 1) && (inbuf[1] != '\n') && (inbuf[1] != '\0'))
367 		return -EINVAL;
368 	down_write(&dcssblk_devices_sem);
369 	dev_info = container_of(dev, struct dcssblk_dev_info, dev);
370 	if (atomic_read(&dev_info->use_count)) {
371 		rc = -EBUSY;
372 		goto out;
373 	}
374 	if (inbuf[0] == '1') {
375 		/* reload segments in shared mode */
376 		list_for_each_entry(entry, &dev_info->seg_list, lh) {
377 			rc = segment_modify_shared(entry->segment_name,
378 						SEGMENT_SHARED);
379 			if (rc < 0) {
380 				BUG_ON(rc == -EINVAL);
381 				if (rc != -EAGAIN)
382 					goto removeseg;
383 			}
384 		}
385 		dev_info->is_shared = 1;
386 		switch (dev_info->segment_type) {
387 		case SEG_TYPE_SR:
388 		case SEG_TYPE_ER:
389 		case SEG_TYPE_SC:
390 			set_disk_ro(dev_info->gd, 1);
391 		}
392 	} else if (inbuf[0] == '0') {
393 		/* reload segments in exclusive mode */
394 		if (dev_info->segment_type == SEG_TYPE_SC) {
395 			pr_err("DCSS %s is of type SC and cannot be "
396 			       "loaded as exclusive-writable\n",
397 			       dev_info->segment_name);
398 			rc = -EINVAL;
399 			goto out;
400 		}
401 		list_for_each_entry(entry, &dev_info->seg_list, lh) {
402 			rc = segment_modify_shared(entry->segment_name,
403 						   SEGMENT_EXCLUSIVE);
404 			if (rc < 0) {
405 				BUG_ON(rc == -EINVAL);
406 				if (rc != -EAGAIN)
407 					goto removeseg;
408 			}
409 		}
410 		dev_info->is_shared = 0;
411 		set_disk_ro(dev_info->gd, 0);
412 	} else {
413 		rc = -EINVAL;
414 		goto out;
415 	}
416 	rc = count;
417 	goto out;
418 
419 removeseg:
420 	pr_err("DCSS device %s is removed after a failed access mode "
421 	       "change\n", dev_info->segment_name);
422 	temp = entry;
423 	list_for_each_entry(entry, &dev_info->seg_list, lh) {
424 		if (entry != temp)
425 			segment_unload(entry->segment_name);
426 	}
427 	list_del(&dev_info->lh);
428 
429 	kill_dax(dev_info->dax_dev);
430 	put_dax(dev_info->dax_dev);
431 	del_gendisk(dev_info->gd);
432 	blk_cleanup_queue(dev_info->dcssblk_queue);
433 	dev_info->gd->queue = NULL;
434 	put_disk(dev_info->gd);
435 	up_write(&dcssblk_devices_sem);
436 
437 	if (device_remove_file_self(dev, attr)) {
438 		device_unregister(dev);
439 		put_device(dev);
440 	}
441 	return rc;
442 out:
443 	up_write(&dcssblk_devices_sem);
444 	return rc;
445 }
446 static DEVICE_ATTR(shared, S_IWUSR | S_IRUSR, dcssblk_shared_show,
447 		   dcssblk_shared_store);
448 
449 /*
450  * device attribute for save operation on current copy
451  * of the segment. If the segment is busy, saving will
452  * become pending until it gets released, which can be
453  * undone by storing a non-true value to this entry.
454  * (show + store)
455  */
456 static ssize_t
457 dcssblk_save_show(struct device *dev, struct device_attribute *attr, char *buf)
458 {
459 	struct dcssblk_dev_info *dev_info;
460 
461 	dev_info = container_of(dev, struct dcssblk_dev_info, dev);
462 	return sprintf(buf, dev_info->save_pending ? "1\n" : "0\n");
463 }
464 
465 static ssize_t
466 dcssblk_save_store(struct device *dev, struct device_attribute *attr, const char *inbuf, size_t count)
467 {
468 	struct dcssblk_dev_info *dev_info;
469 	struct segment_info *entry;
470 
471 	if ((count > 1) && (inbuf[1] != '\n') && (inbuf[1] != '\0'))
472 		return -EINVAL;
473 	dev_info = container_of(dev, struct dcssblk_dev_info, dev);
474 
475 	down_write(&dcssblk_devices_sem);
476 	if (inbuf[0] == '1') {
477 		if (atomic_read(&dev_info->use_count) == 0) {
478 			// device is idle => we save immediately
479 			pr_info("All DCSSs that map to device %s are "
480 				"saved\n", dev_info->segment_name);
481 			list_for_each_entry(entry, &dev_info->seg_list, lh) {
482 				if (entry->segment_type == SEG_TYPE_EN ||
483 				    entry->segment_type == SEG_TYPE_SN)
484 					pr_warn("DCSS %s is of type SN or EN"
485 						" and cannot be saved\n",
486 						entry->segment_name);
487 				else
488 					segment_save(entry->segment_name);
489 			}
490 		}  else {
491 			// device is busy => we save it when it becomes
492 			// idle in dcssblk_release
493 			pr_info("Device %s is in use, its DCSSs will be "
494 				"saved when it becomes idle\n",
495 				dev_info->segment_name);
496 			dev_info->save_pending = 1;
497 		}
498 	} else if (inbuf[0] == '0') {
499 		if (dev_info->save_pending) {
500 			// device is busy & the user wants to undo his save
501 			// request
502 			dev_info->save_pending = 0;
503 			pr_info("A pending save request for device %s "
504 				"has been canceled\n",
505 				dev_info->segment_name);
506 		}
507 	} else {
508 		up_write(&dcssblk_devices_sem);
509 		return -EINVAL;
510 	}
511 	up_write(&dcssblk_devices_sem);
512 	return count;
513 }
514 static DEVICE_ATTR(save, S_IWUSR | S_IRUSR, dcssblk_save_show,
515 		   dcssblk_save_store);
516 
517 /*
518  * device attribute for showing all segments in a device
519  */
520 static ssize_t
521 dcssblk_seglist_show(struct device *dev, struct device_attribute *attr,
522 		char *buf)
523 {
524 	int i;
525 
526 	struct dcssblk_dev_info *dev_info;
527 	struct segment_info *entry;
528 
529 	down_read(&dcssblk_devices_sem);
530 	dev_info = container_of(dev, struct dcssblk_dev_info, dev);
531 	i = 0;
532 	buf[0] = '\0';
533 	list_for_each_entry(entry, &dev_info->seg_list, lh) {
534 		strcpy(&buf[i], entry->segment_name);
535 		i += strlen(entry->segment_name);
536 		buf[i] = '\n';
537 		i++;
538 	}
539 	up_read(&dcssblk_devices_sem);
540 	return i;
541 }
542 static DEVICE_ATTR(seglist, S_IRUSR, dcssblk_seglist_show, NULL);
543 
544 static struct attribute *dcssblk_dev_attrs[] = {
545 	&dev_attr_shared.attr,
546 	&dev_attr_save.attr,
547 	&dev_attr_seglist.attr,
548 	NULL,
549 };
550 static struct attribute_group dcssblk_dev_attr_group = {
551 	.attrs = dcssblk_dev_attrs,
552 };
553 static const struct attribute_group *dcssblk_dev_attr_groups[] = {
554 	&dcssblk_dev_attr_group,
555 	NULL,
556 };
557 
558 /*
559  * device attribute for adding devices
560  */
561 static ssize_t
562 dcssblk_add_store(struct device *dev, struct device_attribute *attr, const char *buf, size_t count)
563 {
564 	int rc, i, j, num_of_segments;
565 	struct dcssblk_dev_info *dev_info;
566 	struct segment_info *seg_info, *temp;
567 	char *local_buf;
568 	unsigned long seg_byte_size;
569 
570 	dev_info = NULL;
571 	seg_info = NULL;
572 	if (dev != dcssblk_root_dev) {
573 		rc = -EINVAL;
574 		goto out_nobuf;
575 	}
576 	if ((count < 1) || (buf[0] == '\0') || (buf[0] == '\n')) {
577 		rc = -ENAMETOOLONG;
578 		goto out_nobuf;
579 	}
580 
581 	local_buf = kmalloc(count + 1, GFP_KERNEL);
582 	if (local_buf == NULL) {
583 		rc = -ENOMEM;
584 		goto out_nobuf;
585 	}
586 
587 	/*
588 	 * parse input
589 	 */
590 	num_of_segments = 0;
591 	for (i = 0; (i < count && (buf[i] != '\0') && (buf[i] != '\n')); i++) {
592 		for (j = i; j < count &&
593 			(buf[j] != ':') &&
594 			(buf[j] != '\0') &&
595 			(buf[j] != '\n'); j++) {
596 			local_buf[j-i] = toupper(buf[j]);
597 		}
598 		local_buf[j-i] = '\0';
599 		if (((j - i) == 0) || ((j - i) > 8)) {
600 			rc = -ENAMETOOLONG;
601 			goto seg_list_del;
602 		}
603 
604 		rc = dcssblk_load_segment(local_buf, &seg_info);
605 		if (rc < 0)
606 			goto seg_list_del;
607 		/*
608 		 * get a struct dcssblk_dev_info
609 		 */
610 		if (num_of_segments == 0) {
611 			dev_info = kzalloc(sizeof(struct dcssblk_dev_info),
612 					GFP_KERNEL);
613 			if (dev_info == NULL) {
614 				rc = -ENOMEM;
615 				goto out;
616 			}
617 			strcpy(dev_info->segment_name, local_buf);
618 			dev_info->segment_type = seg_info->segment_type;
619 			INIT_LIST_HEAD(&dev_info->seg_list);
620 		}
621 		list_add_tail(&seg_info->lh, &dev_info->seg_list);
622 		num_of_segments++;
623 		i = j;
624 
625 		if ((buf[j] == '\0') || (buf[j] == '\n'))
626 			break;
627 	}
628 
629 	/* no trailing colon at the end of the input */
630 	if ((i > 0) && (buf[i-1] == ':')) {
631 		rc = -ENAMETOOLONG;
632 		goto seg_list_del;
633 	}
634 	strlcpy(local_buf, buf, i + 1);
635 	dev_info->num_of_segments = num_of_segments;
636 	rc = dcssblk_is_continuous(dev_info);
637 	if (rc < 0)
638 		goto seg_list_del;
639 
640 	dev_info->start = dcssblk_find_lowest_addr(dev_info);
641 	dev_info->end = dcssblk_find_highest_addr(dev_info);
642 
643 	dev_set_name(&dev_info->dev, "%s", dev_info->segment_name);
644 	dev_info->dev.release = dcssblk_release_segment;
645 	dev_info->dev.groups = dcssblk_dev_attr_groups;
646 	INIT_LIST_HEAD(&dev_info->lh);
647 	dev_info->gd = alloc_disk(DCSSBLK_MINORS_PER_DISK);
648 	if (dev_info->gd == NULL) {
649 		rc = -ENOMEM;
650 		goto seg_list_del;
651 	}
652 	dev_info->gd->major = dcssblk_major;
653 	dev_info->gd->fops = &dcssblk_devops;
654 	dev_info->dcssblk_queue = blk_alloc_queue(NUMA_NO_NODE);
655 	dev_info->gd->queue = dev_info->dcssblk_queue;
656 	dev_info->gd->private_data = dev_info;
657 	blk_queue_logical_block_size(dev_info->dcssblk_queue, 4096);
658 	blk_queue_flag_set(QUEUE_FLAG_DAX, dev_info->dcssblk_queue);
659 
660 	seg_byte_size = (dev_info->end - dev_info->start + 1);
661 	set_capacity(dev_info->gd, seg_byte_size >> 9); // size in sectors
662 	pr_info("Loaded %s with total size %lu bytes and capacity %lu "
663 		"sectors\n", local_buf, seg_byte_size, seg_byte_size >> 9);
664 
665 	dev_info->save_pending = 0;
666 	dev_info->is_shared = 1;
667 	dev_info->dev.parent = dcssblk_root_dev;
668 
669 	/*
670 	 *get minor, add to list
671 	 */
672 	down_write(&dcssblk_devices_sem);
673 	if (dcssblk_get_segment_by_name(local_buf)) {
674 		rc = -EEXIST;
675 		goto release_gd;
676 	}
677 	rc = dcssblk_assign_free_minor(dev_info);
678 	if (rc)
679 		goto release_gd;
680 	sprintf(dev_info->gd->disk_name, "dcssblk%d",
681 		dev_info->gd->first_minor);
682 	list_add_tail(&dev_info->lh, &dcssblk_devices);
683 
684 	if (!try_module_get(THIS_MODULE)) {
685 		rc = -ENODEV;
686 		goto dev_list_del;
687 	}
688 	/*
689 	 * register the device
690 	 */
691 	rc = device_register(&dev_info->dev);
692 	if (rc)
693 		goto put_dev;
694 
695 	dev_info->dax_dev = alloc_dax(dev_info, dev_info->gd->disk_name,
696 			&dcssblk_dax_ops, DAXDEV_F_SYNC);
697 	if (IS_ERR(dev_info->dax_dev)) {
698 		rc = PTR_ERR(dev_info->dax_dev);
699 		dev_info->dax_dev = NULL;
700 		goto put_dev;
701 	}
702 
703 	get_device(&dev_info->dev);
704 	device_add_disk(&dev_info->dev, dev_info->gd, NULL);
705 
706 	switch (dev_info->segment_type) {
707 		case SEG_TYPE_SR:
708 		case SEG_TYPE_ER:
709 		case SEG_TYPE_SC:
710 			set_disk_ro(dev_info->gd,1);
711 			break;
712 		default:
713 			set_disk_ro(dev_info->gd,0);
714 			break;
715 	}
716 	up_write(&dcssblk_devices_sem);
717 	rc = count;
718 	goto out;
719 
720 put_dev:
721 	list_del(&dev_info->lh);
722 	blk_cleanup_queue(dev_info->dcssblk_queue);
723 	dev_info->gd->queue = NULL;
724 	put_disk(dev_info->gd);
725 	list_for_each_entry(seg_info, &dev_info->seg_list, lh) {
726 		segment_unload(seg_info->segment_name);
727 	}
728 	put_device(&dev_info->dev);
729 	up_write(&dcssblk_devices_sem);
730 	goto out;
731 dev_list_del:
732 	list_del(&dev_info->lh);
733 release_gd:
734 	blk_cleanup_queue(dev_info->dcssblk_queue);
735 	dev_info->gd->queue = NULL;
736 	put_disk(dev_info->gd);
737 	up_write(&dcssblk_devices_sem);
738 seg_list_del:
739 	if (dev_info == NULL)
740 		goto out;
741 	list_for_each_entry_safe(seg_info, temp, &dev_info->seg_list, lh) {
742 		list_del(&seg_info->lh);
743 		segment_unload(seg_info->segment_name);
744 		kfree(seg_info);
745 	}
746 	kfree(dev_info);
747 out:
748 	kfree(local_buf);
749 out_nobuf:
750 	return rc;
751 }
752 
753 /*
754  * device attribute for removing devices
755  */
756 static ssize_t
757 dcssblk_remove_store(struct device *dev, struct device_attribute *attr, const char *buf, size_t count)
758 {
759 	struct dcssblk_dev_info *dev_info;
760 	struct segment_info *entry;
761 	int rc, i;
762 	char *local_buf;
763 
764 	if (dev != dcssblk_root_dev) {
765 		return -EINVAL;
766 	}
767 	local_buf = kmalloc(count + 1, GFP_KERNEL);
768 	if (local_buf == NULL) {
769 		return -ENOMEM;
770 	}
771 	/*
772 	 * parse input
773 	 */
774 	for (i = 0; (i < count && (*(buf+i)!='\0') && (*(buf+i)!='\n')); i++) {
775 		local_buf[i] = toupper(buf[i]);
776 	}
777 	local_buf[i] = '\0';
778 	if ((i == 0) || (i > 8)) {
779 		rc = -ENAMETOOLONG;
780 		goto out_buf;
781 	}
782 
783 	down_write(&dcssblk_devices_sem);
784 	dev_info = dcssblk_get_device_by_name(local_buf);
785 	if (dev_info == NULL) {
786 		up_write(&dcssblk_devices_sem);
787 		pr_warn("Device %s cannot be removed because it is not a known device\n",
788 			local_buf);
789 		rc = -ENODEV;
790 		goto out_buf;
791 	}
792 	if (atomic_read(&dev_info->use_count) != 0) {
793 		up_write(&dcssblk_devices_sem);
794 		pr_warn("Device %s cannot be removed while it is in use\n",
795 			local_buf);
796 		rc = -EBUSY;
797 		goto out_buf;
798 	}
799 
800 	list_del(&dev_info->lh);
801 	kill_dax(dev_info->dax_dev);
802 	put_dax(dev_info->dax_dev);
803 	del_gendisk(dev_info->gd);
804 	blk_cleanup_queue(dev_info->dcssblk_queue);
805 	dev_info->gd->queue = NULL;
806 	put_disk(dev_info->gd);
807 
808 	/* unload all related segments */
809 	list_for_each_entry(entry, &dev_info->seg_list, lh)
810 		segment_unload(entry->segment_name);
811 
812 	up_write(&dcssblk_devices_sem);
813 
814 	device_unregister(&dev_info->dev);
815 	put_device(&dev_info->dev);
816 
817 	rc = count;
818 out_buf:
819 	kfree(local_buf);
820 	return rc;
821 }
822 
823 static int
824 dcssblk_open(struct block_device *bdev, fmode_t mode)
825 {
826 	struct dcssblk_dev_info *dev_info;
827 	int rc;
828 
829 	dev_info = bdev->bd_disk->private_data;
830 	if (NULL == dev_info) {
831 		rc = -ENODEV;
832 		goto out;
833 	}
834 	atomic_inc(&dev_info->use_count);
835 	rc = 0;
836 out:
837 	return rc;
838 }
839 
840 static void
841 dcssblk_release(struct gendisk *disk, fmode_t mode)
842 {
843 	struct dcssblk_dev_info *dev_info = disk->private_data;
844 	struct segment_info *entry;
845 
846 	if (!dev_info) {
847 		WARN_ON(1);
848 		return;
849 	}
850 	down_write(&dcssblk_devices_sem);
851 	if (atomic_dec_and_test(&dev_info->use_count)
852 	    && (dev_info->save_pending)) {
853 		pr_info("Device %s has become idle and is being saved "
854 			"now\n", dev_info->segment_name);
855 		list_for_each_entry(entry, &dev_info->seg_list, lh) {
856 			if (entry->segment_type == SEG_TYPE_EN ||
857 			    entry->segment_type == SEG_TYPE_SN)
858 				pr_warn("DCSS %s is of type SN or EN and cannot"
859 					" be saved\n", entry->segment_name);
860 			else
861 				segment_save(entry->segment_name);
862 		}
863 		dev_info->save_pending = 0;
864 	}
865 	up_write(&dcssblk_devices_sem);
866 }
867 
868 static blk_qc_t
869 dcssblk_submit_bio(struct bio *bio)
870 {
871 	struct dcssblk_dev_info *dev_info;
872 	struct bio_vec bvec;
873 	struct bvec_iter iter;
874 	unsigned long index;
875 	unsigned long page_addr;
876 	unsigned long source_addr;
877 	unsigned long bytes_done;
878 
879 	blk_queue_split(&bio);
880 
881 	bytes_done = 0;
882 	dev_info = bio->bi_bdev->bd_disk->private_data;
883 	if (dev_info == NULL)
884 		goto fail;
885 	if ((bio->bi_iter.bi_sector & 7) != 0 ||
886 	    (bio->bi_iter.bi_size & 4095) != 0)
887 		/* Request is not page-aligned. */
888 		goto fail;
889 	/* verify data transfer direction */
890 	if (dev_info->is_shared) {
891 		switch (dev_info->segment_type) {
892 		case SEG_TYPE_SR:
893 		case SEG_TYPE_ER:
894 		case SEG_TYPE_SC:
895 			/* cannot write to these segments */
896 			if (bio_data_dir(bio) == WRITE) {
897 				pr_warn("Writing to %s failed because it is a read-only device\n",
898 					dev_name(&dev_info->dev));
899 				goto fail;
900 			}
901 		}
902 	}
903 
904 	index = (bio->bi_iter.bi_sector >> 3);
905 	bio_for_each_segment(bvec, bio, iter) {
906 		page_addr = (unsigned long)
907 			page_address(bvec.bv_page) + bvec.bv_offset;
908 		source_addr = dev_info->start + (index<<12) + bytes_done;
909 		if (unlikely((page_addr & 4095) != 0) || (bvec.bv_len & 4095) != 0)
910 			// More paranoia.
911 			goto fail;
912 		if (bio_data_dir(bio) == READ) {
913 			memcpy((void*)page_addr, (void*)source_addr,
914 				bvec.bv_len);
915 		} else {
916 			memcpy((void*)source_addr, (void*)page_addr,
917 				bvec.bv_len);
918 		}
919 		bytes_done += bvec.bv_len;
920 	}
921 	bio_endio(bio);
922 	return BLK_QC_T_NONE;
923 fail:
924 	bio_io_error(bio);
925 	return BLK_QC_T_NONE;
926 }
927 
928 static long
929 __dcssblk_direct_access(struct dcssblk_dev_info *dev_info, pgoff_t pgoff,
930 		long nr_pages, void **kaddr, pfn_t *pfn)
931 {
932 	resource_size_t offset = pgoff * PAGE_SIZE;
933 	unsigned long dev_sz;
934 
935 	dev_sz = dev_info->end - dev_info->start + 1;
936 	if (kaddr)
937 		*kaddr = (void *) dev_info->start + offset;
938 	if (pfn)
939 		*pfn = __pfn_to_pfn_t(PFN_DOWN(dev_info->start + offset),
940 				PFN_DEV|PFN_SPECIAL);
941 
942 	return (dev_sz - offset) / PAGE_SIZE;
943 }
944 
945 static long
946 dcssblk_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
947 		long nr_pages, void **kaddr, pfn_t *pfn)
948 {
949 	struct dcssblk_dev_info *dev_info = dax_get_private(dax_dev);
950 
951 	return __dcssblk_direct_access(dev_info, pgoff, nr_pages, kaddr, pfn);
952 }
953 
954 static void
955 dcssblk_check_params(void)
956 {
957 	int rc, i, j, k;
958 	char buf[DCSSBLK_PARM_LEN + 1];
959 	struct dcssblk_dev_info *dev_info;
960 
961 	for (i = 0; (i < DCSSBLK_PARM_LEN) && (dcssblk_segments[i] != '\0');
962 	     i++) {
963 		for (j = i; (j < DCSSBLK_PARM_LEN) &&
964 			    (dcssblk_segments[j] != ',')  &&
965 			    (dcssblk_segments[j] != '\0') &&
966 			    (dcssblk_segments[j] != '('); j++)
967 		{
968 			buf[j-i] = dcssblk_segments[j];
969 		}
970 		buf[j-i] = '\0';
971 		rc = dcssblk_add_store(dcssblk_root_dev, NULL, buf, j-i);
972 		if ((rc >= 0) && (dcssblk_segments[j] == '(')) {
973 			for (k = 0; (buf[k] != ':') && (buf[k] != '\0'); k++)
974 				buf[k] = toupper(buf[k]);
975 			buf[k] = '\0';
976 			if (!strncmp(&dcssblk_segments[j], "(local)", 7)) {
977 				down_read(&dcssblk_devices_sem);
978 				dev_info = dcssblk_get_device_by_name(buf);
979 				up_read(&dcssblk_devices_sem);
980 				if (dev_info)
981 					dcssblk_shared_store(&dev_info->dev,
982 							     NULL, "0\n", 2);
983 			}
984 		}
985 		while ((dcssblk_segments[j] != ',') &&
986 		       (dcssblk_segments[j] != '\0'))
987 		{
988 			j++;
989 		}
990 		if (dcssblk_segments[j] == '\0')
991 			break;
992 		i = j;
993 	}
994 }
995 
996 /*
997  * Suspend / Resume
998  */
999 static int dcssblk_freeze(struct device *dev)
1000 {
1001 	struct dcssblk_dev_info *dev_info;
1002 	int rc = 0;
1003 
1004 	list_for_each_entry(dev_info, &dcssblk_devices, lh) {
1005 		switch (dev_info->segment_type) {
1006 			case SEG_TYPE_SR:
1007 			case SEG_TYPE_ER:
1008 			case SEG_TYPE_SC:
1009 				if (!dev_info->is_shared)
1010 					rc = -EINVAL;
1011 				break;
1012 			default:
1013 				rc = -EINVAL;
1014 				break;
1015 		}
1016 		if (rc)
1017 			break;
1018 	}
1019 	if (rc)
1020 		pr_err("Suspending the system failed because DCSS device %s "
1021 		       "is writable\n",
1022 		       dev_info->segment_name);
1023 	return rc;
1024 }
1025 
1026 static int dcssblk_restore(struct device *dev)
1027 {
1028 	struct dcssblk_dev_info *dev_info;
1029 	struct segment_info *entry;
1030 	unsigned long start, end;
1031 	int rc = 0;
1032 
1033 	list_for_each_entry(dev_info, &dcssblk_devices, lh) {
1034 		list_for_each_entry(entry, &dev_info->seg_list, lh) {
1035 			segment_unload(entry->segment_name);
1036 			rc = segment_load(entry->segment_name, SEGMENT_SHARED,
1037 					  &start, &end);
1038 			if (rc < 0) {
1039 // TODO in_use check ?
1040 				segment_warning(rc, entry->segment_name);
1041 				goto out_panic;
1042 			}
1043 			if (start != entry->start || end != entry->end) {
1044 				pr_err("The address range of DCSS %s changed "
1045 				       "while the system was suspended\n",
1046 				       entry->segment_name);
1047 				goto out_panic;
1048 			}
1049 		}
1050 	}
1051 	return 0;
1052 out_panic:
1053 	panic("fatal dcssblk resume error\n");
1054 }
1055 
1056 static int dcssblk_thaw(struct device *dev)
1057 {
1058 	return 0;
1059 }
1060 
1061 static const struct dev_pm_ops dcssblk_pm_ops = {
1062 	.freeze		= dcssblk_freeze,
1063 	.thaw		= dcssblk_thaw,
1064 	.restore	= dcssblk_restore,
1065 };
1066 
1067 static struct platform_driver dcssblk_pdrv = {
1068 	.driver = {
1069 		.name	= "dcssblk",
1070 		.pm	= &dcssblk_pm_ops,
1071 	},
1072 };
1073 
1074 static struct platform_device *dcssblk_pdev;
1075 
1076 
1077 /*
1078  * The init/exit functions.
1079  */
1080 static void __exit
1081 dcssblk_exit(void)
1082 {
1083 	platform_device_unregister(dcssblk_pdev);
1084 	platform_driver_unregister(&dcssblk_pdrv);
1085 	root_device_unregister(dcssblk_root_dev);
1086 	unregister_blkdev(dcssblk_major, DCSSBLK_NAME);
1087 }
1088 
1089 static int __init
1090 dcssblk_init(void)
1091 {
1092 	int rc;
1093 
1094 	rc = platform_driver_register(&dcssblk_pdrv);
1095 	if (rc)
1096 		return rc;
1097 
1098 	dcssblk_pdev = platform_device_register_simple("dcssblk", -1, NULL,
1099 							0);
1100 	if (IS_ERR(dcssblk_pdev)) {
1101 		rc = PTR_ERR(dcssblk_pdev);
1102 		goto out_pdrv;
1103 	}
1104 
1105 	dcssblk_root_dev = root_device_register("dcssblk");
1106 	if (IS_ERR(dcssblk_root_dev)) {
1107 		rc = PTR_ERR(dcssblk_root_dev);
1108 		goto out_pdev;
1109 	}
1110 	rc = device_create_file(dcssblk_root_dev, &dev_attr_add);
1111 	if (rc)
1112 		goto out_root;
1113 	rc = device_create_file(dcssblk_root_dev, &dev_attr_remove);
1114 	if (rc)
1115 		goto out_root;
1116 	rc = register_blkdev(0, DCSSBLK_NAME);
1117 	if (rc < 0)
1118 		goto out_root;
1119 	dcssblk_major = rc;
1120 	init_rwsem(&dcssblk_devices_sem);
1121 
1122 	dcssblk_check_params();
1123 	return 0;
1124 
1125 out_root:
1126 	root_device_unregister(dcssblk_root_dev);
1127 out_pdev:
1128 	platform_device_unregister(dcssblk_pdev);
1129 out_pdrv:
1130 	platform_driver_unregister(&dcssblk_pdrv);
1131 	return rc;
1132 }
1133 
1134 module_init(dcssblk_init);
1135 module_exit(dcssblk_exit);
1136 
1137 module_param_string(segments, dcssblk_segments, DCSSBLK_PARM_LEN, 0444);
1138 MODULE_PARM_DESC(segments, "Name of DCSS segment(s) to be loaded, "
1139 		 "comma-separated list, names in each set separated "
1140 		 "by commas are separated by colons, each set contains "
1141 		 "names of contiguous segments and each name max. 8 chars.\n"
1142 		 "Adding \"(local)\" to the end of each set equals echoing 0 "
1143 		 "to /sys/devices/dcssblk/<device name>/shared after loading "
1144 		 "the contiguous segments - \n"
1145 		 "e.g. segments=\"mydcss1,mydcss2:mydcss3,mydcss4(local)\"");
1146 
1147 MODULE_LICENSE("GPL");
1148