xref: /openbmc/linux/drivers/s390/block/dcssblk.c (revision 0242737d)
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/pfn_t.h>
21 #include <linux/uio.h>
22 #include <linux/dax.h>
23 #include <linux/io.h>
24 #include <asm/extmem.h>
25 
26 #define DCSSBLK_NAME "dcssblk"
27 #define DCSSBLK_MINORS_PER_DISK 1
28 #define DCSSBLK_PARM_LEN 400
29 #define DCSS_BUS_ID_SIZE 20
30 
31 static int dcssblk_open(struct gendisk *disk, blk_mode_t mode);
32 static void dcssblk_release(struct gendisk *disk);
33 static void dcssblk_submit_bio(struct bio *bio);
34 static long dcssblk_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
35 		long nr_pages, enum dax_access_mode mode, void **kaddr,
36 		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 int dcssblk_dax_zero_page_range(struct dax_device *dax_dev,
49 				       pgoff_t pgoff, size_t nr_pages)
50 {
51 	long rc;
52 	void *kaddr;
53 
54 	rc = dax_direct_access(dax_dev, pgoff, nr_pages, DAX_ACCESS,
55 			&kaddr, NULL);
56 	if (rc < 0)
57 		return dax_mem2blk_err(rc);
58 
59 	memset(kaddr, 0, nr_pages << PAGE_SHIFT);
60 	dax_flush(dax_dev, kaddr, nr_pages << PAGE_SHIFT);
61 	return 0;
62 }
63 
64 static const struct dax_operations dcssblk_dax_ops = {
65 	.direct_access = dcssblk_dax_direct_access,
66 	.zero_page_range = dcssblk_dax_zero_page_range,
67 };
68 
69 struct dcssblk_dev_info {
70 	struct list_head lh;
71 	struct device dev;
72 	char segment_name[DCSS_BUS_ID_SIZE];
73 	atomic_t use_count;
74 	struct gendisk *gd;
75 	unsigned long start;
76 	unsigned long end;
77 	int segment_type;
78 	unsigned char save_pending;
79 	unsigned char is_shared;
80 	int num_of_segments;
81 	struct list_head seg_list;
82 	struct dax_device *dax_dev;
83 };
84 
85 struct segment_info {
86 	struct list_head lh;
87 	char segment_name[DCSS_BUS_ID_SIZE];
88 	unsigned long start;
89 	unsigned long end;
90 	int segment_type;
91 };
92 
93 static ssize_t dcssblk_add_store(struct device * dev, struct device_attribute *attr, const char * buf,
94 				  size_t count);
95 static ssize_t dcssblk_remove_store(struct device * dev, struct device_attribute *attr, const char * buf,
96 				  size_t count);
97 
98 static DEVICE_ATTR(add, S_IWUSR, NULL, dcssblk_add_store);
99 static DEVICE_ATTR(remove, S_IWUSR, NULL, dcssblk_remove_store);
100 
101 static struct device *dcssblk_root_dev;
102 
103 static LIST_HEAD(dcssblk_devices);
104 static struct rw_semaphore dcssblk_devices_sem;
105 
106 /*
107  * release function for segment device.
108  */
109 static void
110 dcssblk_release_segment(struct device *dev)
111 {
112 	struct dcssblk_dev_info *dev_info;
113 	struct segment_info *entry, *temp;
114 
115 	dev_info = container_of(dev, struct dcssblk_dev_info, dev);
116 	list_for_each_entry_safe(entry, temp, &dev_info->seg_list, lh) {
117 		list_del(&entry->lh);
118 		kfree(entry);
119 	}
120 	kfree(dev_info);
121 	module_put(THIS_MODULE);
122 }
123 
124 /*
125  * get a minor number. needs to be called with
126  * down_write(&dcssblk_devices_sem) and the
127  * device needs to be enqueued before the semaphore is
128  * freed.
129  */
130 static int
131 dcssblk_assign_free_minor(struct dcssblk_dev_info *dev_info)
132 {
133 	int minor, found;
134 	struct dcssblk_dev_info *entry;
135 
136 	if (dev_info == NULL)
137 		return -EINVAL;
138 	for (minor = 0; minor < (1<<MINORBITS); minor++) {
139 		found = 0;
140 		// test if minor available
141 		list_for_each_entry(entry, &dcssblk_devices, lh)
142 			if (minor == entry->gd->first_minor)
143 				found++;
144 		if (!found) break; // got unused minor
145 	}
146 	if (found)
147 		return -EBUSY;
148 	dev_info->gd->first_minor = minor;
149 	return 0;
150 }
151 
152 /*
153  * get the struct dcssblk_dev_info from dcssblk_devices
154  * for the given name.
155  * down_read(&dcssblk_devices_sem) must be held.
156  */
157 static struct dcssblk_dev_info *
158 dcssblk_get_device_by_name(char *name)
159 {
160 	struct dcssblk_dev_info *entry;
161 
162 	list_for_each_entry(entry, &dcssblk_devices, lh) {
163 		if (!strcmp(name, entry->segment_name)) {
164 			return entry;
165 		}
166 	}
167 	return NULL;
168 }
169 
170 /*
171  * get the struct segment_info from seg_list
172  * for the given name.
173  * down_read(&dcssblk_devices_sem) must be held.
174  */
175 static struct segment_info *
176 dcssblk_get_segment_by_name(char *name)
177 {
178 	struct dcssblk_dev_info *dev_info;
179 	struct segment_info *entry;
180 
181 	list_for_each_entry(dev_info, &dcssblk_devices, lh) {
182 		list_for_each_entry(entry, &dev_info->seg_list, lh) {
183 			if (!strcmp(name, entry->segment_name))
184 				return entry;
185 		}
186 	}
187 	return NULL;
188 }
189 
190 /*
191  * get the highest address of the multi-segment block.
192  */
193 static unsigned long
194 dcssblk_find_highest_addr(struct dcssblk_dev_info *dev_info)
195 {
196 	unsigned long highest_addr;
197 	struct segment_info *entry;
198 
199 	highest_addr = 0;
200 	list_for_each_entry(entry, &dev_info->seg_list, lh) {
201 		if (highest_addr < entry->end)
202 			highest_addr = entry->end;
203 	}
204 	return highest_addr;
205 }
206 
207 /*
208  * get the lowest address of the multi-segment block.
209  */
210 static unsigned long
211 dcssblk_find_lowest_addr(struct dcssblk_dev_info *dev_info)
212 {
213 	int set_first;
214 	unsigned long lowest_addr;
215 	struct segment_info *entry;
216 
217 	set_first = 0;
218 	lowest_addr = 0;
219 	list_for_each_entry(entry, &dev_info->seg_list, lh) {
220 		if (set_first == 0) {
221 			lowest_addr = entry->start;
222 			set_first = 1;
223 		} else {
224 			if (lowest_addr > entry->start)
225 				lowest_addr = entry->start;
226 		}
227 	}
228 	return lowest_addr;
229 }
230 
231 /*
232  * Check continuity of segments.
233  */
234 static int
235 dcssblk_is_continuous(struct dcssblk_dev_info *dev_info)
236 {
237 	int i, j, rc;
238 	struct segment_info *sort_list, *entry, temp;
239 
240 	if (dev_info->num_of_segments <= 1)
241 		return 0;
242 
243 	sort_list = kcalloc(dev_info->num_of_segments,
244 			    sizeof(struct segment_info),
245 			    GFP_KERNEL);
246 	if (sort_list == NULL)
247 		return -ENOMEM;
248 	i = 0;
249 	list_for_each_entry(entry, &dev_info->seg_list, lh) {
250 		memcpy(&sort_list[i], entry, sizeof(struct segment_info));
251 		i++;
252 	}
253 
254 	/* sort segments */
255 	for (i = 0; i < dev_info->num_of_segments; i++)
256 		for (j = 0; j < dev_info->num_of_segments; j++)
257 			if (sort_list[j].start > sort_list[i].start) {
258 				memcpy(&temp, &sort_list[i],
259 					sizeof(struct segment_info));
260 				memcpy(&sort_list[i], &sort_list[j],
261 					sizeof(struct segment_info));
262 				memcpy(&sort_list[j], &temp,
263 					sizeof(struct segment_info));
264 			}
265 
266 	/* check continuity */
267 	for (i = 0; i < dev_info->num_of_segments - 1; i++) {
268 		if ((sort_list[i].end + 1) != sort_list[i+1].start) {
269 			pr_err("Adjacent DCSSs %s and %s are not "
270 			       "contiguous\n", sort_list[i].segment_name,
271 			       sort_list[i+1].segment_name);
272 			rc = -EINVAL;
273 			goto out;
274 		}
275 		/* EN and EW are allowed in a block device */
276 		if (sort_list[i].segment_type != sort_list[i+1].segment_type) {
277 			if (!(sort_list[i].segment_type & SEGMENT_EXCLUSIVE) ||
278 				(sort_list[i].segment_type == SEG_TYPE_ER) ||
279 				!(sort_list[i+1].segment_type &
280 				SEGMENT_EXCLUSIVE) ||
281 				(sort_list[i+1].segment_type == SEG_TYPE_ER)) {
282 				pr_err("DCSS %s and DCSS %s have "
283 				       "incompatible types\n",
284 				       sort_list[i].segment_name,
285 				       sort_list[i+1].segment_name);
286 				rc = -EINVAL;
287 				goto out;
288 			}
289 		}
290 	}
291 	rc = 0;
292 out:
293 	kfree(sort_list);
294 	return rc;
295 }
296 
297 /*
298  * Load a segment
299  */
300 static int
301 dcssblk_load_segment(char *name, struct segment_info **seg_info)
302 {
303 	int rc;
304 
305 	/* already loaded? */
306 	down_read(&dcssblk_devices_sem);
307 	*seg_info = dcssblk_get_segment_by_name(name);
308 	up_read(&dcssblk_devices_sem);
309 	if (*seg_info != NULL)
310 		return -EEXIST;
311 
312 	/* get a struct segment_info */
313 	*seg_info = kzalloc(sizeof(struct segment_info), GFP_KERNEL);
314 	if (*seg_info == NULL)
315 		return -ENOMEM;
316 
317 	strcpy((*seg_info)->segment_name, name);
318 
319 	/* load the segment */
320 	rc = segment_load(name, SEGMENT_SHARED,
321 			&(*seg_info)->start, &(*seg_info)->end);
322 	if (rc < 0) {
323 		segment_warning(rc, (*seg_info)->segment_name);
324 		kfree(*seg_info);
325 	} else {
326 		INIT_LIST_HEAD(&(*seg_info)->lh);
327 		(*seg_info)->segment_type = rc;
328 	}
329 	return rc;
330 }
331 
332 /*
333  * device attribute for switching shared/nonshared (exclusive)
334  * operation (show + store)
335  */
336 static ssize_t
337 dcssblk_shared_show(struct device *dev, struct device_attribute *attr, char *buf)
338 {
339 	struct dcssblk_dev_info *dev_info;
340 
341 	dev_info = container_of(dev, struct dcssblk_dev_info, dev);
342 	return sprintf(buf, dev_info->is_shared ? "1\n" : "0\n");
343 }
344 
345 static ssize_t
346 dcssblk_shared_store(struct device *dev, struct device_attribute *attr, const char *inbuf, size_t count)
347 {
348 	struct dcssblk_dev_info *dev_info;
349 	struct segment_info *entry, *temp;
350 	int rc;
351 
352 	if ((count > 1) && (inbuf[1] != '\n') && (inbuf[1] != '\0'))
353 		return -EINVAL;
354 	down_write(&dcssblk_devices_sem);
355 	dev_info = container_of(dev, struct dcssblk_dev_info, dev);
356 	if (atomic_read(&dev_info->use_count)) {
357 		rc = -EBUSY;
358 		goto out;
359 	}
360 	if (inbuf[0] == '1') {
361 		/* reload segments in shared mode */
362 		list_for_each_entry(entry, &dev_info->seg_list, lh) {
363 			rc = segment_modify_shared(entry->segment_name,
364 						SEGMENT_SHARED);
365 			if (rc < 0) {
366 				BUG_ON(rc == -EINVAL);
367 				if (rc != -EAGAIN)
368 					goto removeseg;
369 			}
370 		}
371 		dev_info->is_shared = 1;
372 		switch (dev_info->segment_type) {
373 		case SEG_TYPE_SR:
374 		case SEG_TYPE_ER:
375 		case SEG_TYPE_SC:
376 			set_disk_ro(dev_info->gd, 1);
377 		}
378 	} else if (inbuf[0] == '0') {
379 		/* reload segments in exclusive mode */
380 		if (dev_info->segment_type == SEG_TYPE_SC) {
381 			pr_err("DCSS %s is of type SC and cannot be "
382 			       "loaded as exclusive-writable\n",
383 			       dev_info->segment_name);
384 			rc = -EINVAL;
385 			goto out;
386 		}
387 		list_for_each_entry(entry, &dev_info->seg_list, lh) {
388 			rc = segment_modify_shared(entry->segment_name,
389 						   SEGMENT_EXCLUSIVE);
390 			if (rc < 0) {
391 				BUG_ON(rc == -EINVAL);
392 				if (rc != -EAGAIN)
393 					goto removeseg;
394 			}
395 		}
396 		dev_info->is_shared = 0;
397 		set_disk_ro(dev_info->gd, 0);
398 	} else {
399 		rc = -EINVAL;
400 		goto out;
401 	}
402 	rc = count;
403 	goto out;
404 
405 removeseg:
406 	pr_err("DCSS device %s is removed after a failed access mode "
407 	       "change\n", dev_info->segment_name);
408 	temp = entry;
409 	list_for_each_entry(entry, &dev_info->seg_list, lh) {
410 		if (entry != temp)
411 			segment_unload(entry->segment_name);
412 	}
413 	list_del(&dev_info->lh);
414 
415 	kill_dax(dev_info->dax_dev);
416 	put_dax(dev_info->dax_dev);
417 	del_gendisk(dev_info->gd);
418 	put_disk(dev_info->gd);
419 	up_write(&dcssblk_devices_sem);
420 
421 	if (device_remove_file_self(dev, attr)) {
422 		device_unregister(dev);
423 		put_device(dev);
424 	}
425 	return rc;
426 out:
427 	up_write(&dcssblk_devices_sem);
428 	return rc;
429 }
430 static DEVICE_ATTR(shared, S_IWUSR | S_IRUSR, dcssblk_shared_show,
431 		   dcssblk_shared_store);
432 
433 /*
434  * device attribute for save operation on current copy
435  * of the segment. If the segment is busy, saving will
436  * become pending until it gets released, which can be
437  * undone by storing a non-true value to this entry.
438  * (show + store)
439  */
440 static ssize_t
441 dcssblk_save_show(struct device *dev, struct device_attribute *attr, char *buf)
442 {
443 	struct dcssblk_dev_info *dev_info;
444 
445 	dev_info = container_of(dev, struct dcssblk_dev_info, dev);
446 	return sprintf(buf, dev_info->save_pending ? "1\n" : "0\n");
447 }
448 
449 static ssize_t
450 dcssblk_save_store(struct device *dev, struct device_attribute *attr, const char *inbuf, size_t count)
451 {
452 	struct dcssblk_dev_info *dev_info;
453 	struct segment_info *entry;
454 
455 	if ((count > 1) && (inbuf[1] != '\n') && (inbuf[1] != '\0'))
456 		return -EINVAL;
457 	dev_info = container_of(dev, struct dcssblk_dev_info, dev);
458 
459 	down_write(&dcssblk_devices_sem);
460 	if (inbuf[0] == '1') {
461 		if (atomic_read(&dev_info->use_count) == 0) {
462 			// device is idle => we save immediately
463 			pr_info("All DCSSs that map to device %s are "
464 				"saved\n", dev_info->segment_name);
465 			list_for_each_entry(entry, &dev_info->seg_list, lh) {
466 				if (entry->segment_type == SEG_TYPE_EN ||
467 				    entry->segment_type == SEG_TYPE_SN)
468 					pr_warn("DCSS %s is of type SN or EN"
469 						" and cannot be saved\n",
470 						entry->segment_name);
471 				else
472 					segment_save(entry->segment_name);
473 			}
474 		}  else {
475 			// device is busy => we save it when it becomes
476 			// idle in dcssblk_release
477 			pr_info("Device %s is in use, its DCSSs will be "
478 				"saved when it becomes idle\n",
479 				dev_info->segment_name);
480 			dev_info->save_pending = 1;
481 		}
482 	} else if (inbuf[0] == '0') {
483 		if (dev_info->save_pending) {
484 			// device is busy & the user wants to undo his save
485 			// request
486 			dev_info->save_pending = 0;
487 			pr_info("A pending save request for device %s "
488 				"has been canceled\n",
489 				dev_info->segment_name);
490 		}
491 	} else {
492 		up_write(&dcssblk_devices_sem);
493 		return -EINVAL;
494 	}
495 	up_write(&dcssblk_devices_sem);
496 	return count;
497 }
498 static DEVICE_ATTR(save, S_IWUSR | S_IRUSR, dcssblk_save_show,
499 		   dcssblk_save_store);
500 
501 /*
502  * device attribute for showing all segments in a device
503  */
504 static ssize_t
505 dcssblk_seglist_show(struct device *dev, struct device_attribute *attr,
506 		char *buf)
507 {
508 	int i;
509 
510 	struct dcssblk_dev_info *dev_info;
511 	struct segment_info *entry;
512 
513 	down_read(&dcssblk_devices_sem);
514 	dev_info = container_of(dev, struct dcssblk_dev_info, dev);
515 	i = 0;
516 	buf[0] = '\0';
517 	list_for_each_entry(entry, &dev_info->seg_list, lh) {
518 		strcpy(&buf[i], entry->segment_name);
519 		i += strlen(entry->segment_name);
520 		buf[i] = '\n';
521 		i++;
522 	}
523 	up_read(&dcssblk_devices_sem);
524 	return i;
525 }
526 static DEVICE_ATTR(seglist, S_IRUSR, dcssblk_seglist_show, NULL);
527 
528 static struct attribute *dcssblk_dev_attrs[] = {
529 	&dev_attr_shared.attr,
530 	&dev_attr_save.attr,
531 	&dev_attr_seglist.attr,
532 	NULL,
533 };
534 static struct attribute_group dcssblk_dev_attr_group = {
535 	.attrs = dcssblk_dev_attrs,
536 };
537 static const struct attribute_group *dcssblk_dev_attr_groups[] = {
538 	&dcssblk_dev_attr_group,
539 	NULL,
540 };
541 
542 /*
543  * device attribute for adding devices
544  */
545 static ssize_t
546 dcssblk_add_store(struct device *dev, struct device_attribute *attr, const char *buf, size_t count)
547 {
548 	int rc, i, j, num_of_segments;
549 	struct dcssblk_dev_info *dev_info;
550 	struct segment_info *seg_info, *temp;
551 	char *local_buf;
552 	unsigned long seg_byte_size;
553 
554 	dev_info = NULL;
555 	seg_info = NULL;
556 	if (dev != dcssblk_root_dev) {
557 		rc = -EINVAL;
558 		goto out_nobuf;
559 	}
560 	if ((count < 1) || (buf[0] == '\0') || (buf[0] == '\n')) {
561 		rc = -ENAMETOOLONG;
562 		goto out_nobuf;
563 	}
564 
565 	local_buf = kmalloc(count + 1, GFP_KERNEL);
566 	if (local_buf == NULL) {
567 		rc = -ENOMEM;
568 		goto out_nobuf;
569 	}
570 
571 	/*
572 	 * parse input
573 	 */
574 	num_of_segments = 0;
575 	for (i = 0; (i < count && (buf[i] != '\0') && (buf[i] != '\n')); i++) {
576 		for (j = i; j < count &&
577 			(buf[j] != ':') &&
578 			(buf[j] != '\0') &&
579 			(buf[j] != '\n'); j++) {
580 			local_buf[j-i] = toupper(buf[j]);
581 		}
582 		local_buf[j-i] = '\0';
583 		if (((j - i) == 0) || ((j - i) > 8)) {
584 			rc = -ENAMETOOLONG;
585 			goto seg_list_del;
586 		}
587 
588 		rc = dcssblk_load_segment(local_buf, &seg_info);
589 		if (rc < 0)
590 			goto seg_list_del;
591 		/*
592 		 * get a struct dcssblk_dev_info
593 		 */
594 		if (num_of_segments == 0) {
595 			dev_info = kzalloc(sizeof(struct dcssblk_dev_info),
596 					GFP_KERNEL);
597 			if (dev_info == NULL) {
598 				rc = -ENOMEM;
599 				goto out;
600 			}
601 			strcpy(dev_info->segment_name, local_buf);
602 			dev_info->segment_type = seg_info->segment_type;
603 			INIT_LIST_HEAD(&dev_info->seg_list);
604 		}
605 		list_add_tail(&seg_info->lh, &dev_info->seg_list);
606 		num_of_segments++;
607 		i = j;
608 
609 		if ((buf[j] == '\0') || (buf[j] == '\n'))
610 			break;
611 	}
612 
613 	/* no trailing colon at the end of the input */
614 	if ((i > 0) && (buf[i-1] == ':')) {
615 		rc = -ENAMETOOLONG;
616 		goto seg_list_del;
617 	}
618 	strscpy(local_buf, buf, i + 1);
619 	dev_info->num_of_segments = num_of_segments;
620 	rc = dcssblk_is_continuous(dev_info);
621 	if (rc < 0)
622 		goto seg_list_del;
623 
624 	dev_info->start = dcssblk_find_lowest_addr(dev_info);
625 	dev_info->end = dcssblk_find_highest_addr(dev_info);
626 
627 	dev_set_name(&dev_info->dev, "%s", dev_info->segment_name);
628 	dev_info->dev.release = dcssblk_release_segment;
629 	dev_info->dev.groups = dcssblk_dev_attr_groups;
630 	INIT_LIST_HEAD(&dev_info->lh);
631 	dev_info->gd = blk_alloc_disk(NUMA_NO_NODE);
632 	if (dev_info->gd == NULL) {
633 		rc = -ENOMEM;
634 		goto seg_list_del;
635 	}
636 	dev_info->gd->major = dcssblk_major;
637 	dev_info->gd->minors = DCSSBLK_MINORS_PER_DISK;
638 	dev_info->gd->fops = &dcssblk_devops;
639 	dev_info->gd->private_data = dev_info;
640 	dev_info->gd->flags |= GENHD_FL_NO_PART;
641 	blk_queue_logical_block_size(dev_info->gd->queue, 4096);
642 	blk_queue_flag_set(QUEUE_FLAG_DAX, dev_info->gd->queue);
643 
644 	seg_byte_size = (dev_info->end - dev_info->start + 1);
645 	set_capacity(dev_info->gd, seg_byte_size >> 9); // size in sectors
646 	pr_info("Loaded %s with total size %lu bytes and capacity %lu "
647 		"sectors\n", local_buf, seg_byte_size, seg_byte_size >> 9);
648 
649 	dev_info->save_pending = 0;
650 	dev_info->is_shared = 1;
651 	dev_info->dev.parent = dcssblk_root_dev;
652 
653 	/*
654 	 *get minor, add to list
655 	 */
656 	down_write(&dcssblk_devices_sem);
657 	if (dcssblk_get_segment_by_name(local_buf)) {
658 		rc = -EEXIST;
659 		goto release_gd;
660 	}
661 	rc = dcssblk_assign_free_minor(dev_info);
662 	if (rc)
663 		goto release_gd;
664 	sprintf(dev_info->gd->disk_name, "dcssblk%d",
665 		dev_info->gd->first_minor);
666 	list_add_tail(&dev_info->lh, &dcssblk_devices);
667 
668 	if (!try_module_get(THIS_MODULE)) {
669 		rc = -ENODEV;
670 		goto dev_list_del;
671 	}
672 	/*
673 	 * register the device
674 	 */
675 	rc = device_register(&dev_info->dev);
676 	if (rc)
677 		goto put_dev;
678 
679 	dev_info->dax_dev = alloc_dax(dev_info, &dcssblk_dax_ops);
680 	if (IS_ERR(dev_info->dax_dev)) {
681 		rc = PTR_ERR(dev_info->dax_dev);
682 		dev_info->dax_dev = NULL;
683 		goto put_dev;
684 	}
685 	set_dax_synchronous(dev_info->dax_dev);
686 	rc = dax_add_host(dev_info->dax_dev, dev_info->gd);
687 	if (rc)
688 		goto out_dax;
689 
690 	get_device(&dev_info->dev);
691 	rc = device_add_disk(&dev_info->dev, dev_info->gd, NULL);
692 	if (rc)
693 		goto out_dax_host;
694 
695 	switch (dev_info->segment_type) {
696 		case SEG_TYPE_SR:
697 		case SEG_TYPE_ER:
698 		case SEG_TYPE_SC:
699 			set_disk_ro(dev_info->gd,1);
700 			break;
701 		default:
702 			set_disk_ro(dev_info->gd,0);
703 			break;
704 	}
705 	up_write(&dcssblk_devices_sem);
706 	rc = count;
707 	goto out;
708 
709 out_dax_host:
710 	dax_remove_host(dev_info->gd);
711 out_dax:
712 	put_device(&dev_info->dev);
713 	kill_dax(dev_info->dax_dev);
714 	put_dax(dev_info->dax_dev);
715 put_dev:
716 	list_del(&dev_info->lh);
717 	put_disk(dev_info->gd);
718 	list_for_each_entry(seg_info, &dev_info->seg_list, lh) {
719 		segment_unload(seg_info->segment_name);
720 	}
721 	put_device(&dev_info->dev);
722 	up_write(&dcssblk_devices_sem);
723 	goto out;
724 dev_list_del:
725 	list_del(&dev_info->lh);
726 release_gd:
727 	put_disk(dev_info->gd);
728 	up_write(&dcssblk_devices_sem);
729 seg_list_del:
730 	if (dev_info == NULL)
731 		goto out;
732 	list_for_each_entry_safe(seg_info, temp, &dev_info->seg_list, lh) {
733 		list_del(&seg_info->lh);
734 		segment_unload(seg_info->segment_name);
735 		kfree(seg_info);
736 	}
737 	kfree(dev_info);
738 out:
739 	kfree(local_buf);
740 out_nobuf:
741 	return rc;
742 }
743 
744 /*
745  * device attribute for removing devices
746  */
747 static ssize_t
748 dcssblk_remove_store(struct device *dev, struct device_attribute *attr, const char *buf, size_t count)
749 {
750 	struct dcssblk_dev_info *dev_info;
751 	struct segment_info *entry;
752 	int rc, i;
753 	char *local_buf;
754 
755 	if (dev != dcssblk_root_dev) {
756 		return -EINVAL;
757 	}
758 	local_buf = kmalloc(count + 1, GFP_KERNEL);
759 	if (local_buf == NULL) {
760 		return -ENOMEM;
761 	}
762 	/*
763 	 * parse input
764 	 */
765 	for (i = 0; (i < count && (*(buf+i)!='\0') && (*(buf+i)!='\n')); i++) {
766 		local_buf[i] = toupper(buf[i]);
767 	}
768 	local_buf[i] = '\0';
769 	if ((i == 0) || (i > 8)) {
770 		rc = -ENAMETOOLONG;
771 		goto out_buf;
772 	}
773 
774 	down_write(&dcssblk_devices_sem);
775 	dev_info = dcssblk_get_device_by_name(local_buf);
776 	if (dev_info == NULL) {
777 		up_write(&dcssblk_devices_sem);
778 		pr_warn("Device %s cannot be removed because it is not a known device\n",
779 			local_buf);
780 		rc = -ENODEV;
781 		goto out_buf;
782 	}
783 	if (atomic_read(&dev_info->use_count) != 0) {
784 		up_write(&dcssblk_devices_sem);
785 		pr_warn("Device %s cannot be removed while it is in use\n",
786 			local_buf);
787 		rc = -EBUSY;
788 		goto out_buf;
789 	}
790 
791 	list_del(&dev_info->lh);
792 	kill_dax(dev_info->dax_dev);
793 	put_dax(dev_info->dax_dev);
794 	del_gendisk(dev_info->gd);
795 	put_disk(dev_info->gd);
796 
797 	/* unload all related segments */
798 	list_for_each_entry(entry, &dev_info->seg_list, lh)
799 		segment_unload(entry->segment_name);
800 
801 	up_write(&dcssblk_devices_sem);
802 
803 	device_unregister(&dev_info->dev);
804 	put_device(&dev_info->dev);
805 
806 	rc = count;
807 out_buf:
808 	kfree(local_buf);
809 	return rc;
810 }
811 
812 static int
813 dcssblk_open(struct gendisk *disk, blk_mode_t mode)
814 {
815 	struct dcssblk_dev_info *dev_info = disk->private_data;
816 	int rc;
817 
818 	if (NULL == dev_info) {
819 		rc = -ENODEV;
820 		goto out;
821 	}
822 	atomic_inc(&dev_info->use_count);
823 	rc = 0;
824 out:
825 	return rc;
826 }
827 
828 static void
829 dcssblk_release(struct gendisk *disk)
830 {
831 	struct dcssblk_dev_info *dev_info = disk->private_data;
832 	struct segment_info *entry;
833 
834 	if (!dev_info) {
835 		WARN_ON(1);
836 		return;
837 	}
838 	down_write(&dcssblk_devices_sem);
839 	if (atomic_dec_and_test(&dev_info->use_count)
840 	    && (dev_info->save_pending)) {
841 		pr_info("Device %s has become idle and is being saved "
842 			"now\n", dev_info->segment_name);
843 		list_for_each_entry(entry, &dev_info->seg_list, lh) {
844 			if (entry->segment_type == SEG_TYPE_EN ||
845 			    entry->segment_type == SEG_TYPE_SN)
846 				pr_warn("DCSS %s is of type SN or EN and cannot"
847 					" be saved\n", entry->segment_name);
848 			else
849 				segment_save(entry->segment_name);
850 		}
851 		dev_info->save_pending = 0;
852 	}
853 	up_write(&dcssblk_devices_sem);
854 }
855 
856 static void
857 dcssblk_submit_bio(struct bio *bio)
858 {
859 	struct dcssblk_dev_info *dev_info;
860 	struct bio_vec bvec;
861 	struct bvec_iter iter;
862 	unsigned long index;
863 	unsigned long page_addr;
864 	unsigned long source_addr;
865 	unsigned long bytes_done;
866 
867 	bytes_done = 0;
868 	dev_info = bio->bi_bdev->bd_disk->private_data;
869 	if (dev_info == NULL)
870 		goto fail;
871 	if ((bio->bi_iter.bi_sector & 7) != 0 ||
872 	    (bio->bi_iter.bi_size & 4095) != 0)
873 		/* Request is not page-aligned. */
874 		goto fail;
875 	/* verify data transfer direction */
876 	if (dev_info->is_shared) {
877 		switch (dev_info->segment_type) {
878 		case SEG_TYPE_SR:
879 		case SEG_TYPE_ER:
880 		case SEG_TYPE_SC:
881 			/* cannot write to these segments */
882 			if (bio_data_dir(bio) == WRITE) {
883 				pr_warn("Writing to %s failed because it is a read-only device\n",
884 					dev_name(&dev_info->dev));
885 				goto fail;
886 			}
887 		}
888 	}
889 
890 	index = (bio->bi_iter.bi_sector >> 3);
891 	bio_for_each_segment(bvec, bio, iter) {
892 		page_addr = (unsigned long)bvec_virt(&bvec);
893 		source_addr = dev_info->start + (index<<12) + bytes_done;
894 		if (unlikely((page_addr & 4095) != 0) || (bvec.bv_len & 4095) != 0)
895 			// More paranoia.
896 			goto fail;
897 		if (bio_data_dir(bio) == READ) {
898 			memcpy((void*)page_addr, (void*)source_addr,
899 				bvec.bv_len);
900 		} else {
901 			memcpy((void*)source_addr, (void*)page_addr,
902 				bvec.bv_len);
903 		}
904 		bytes_done += bvec.bv_len;
905 	}
906 	bio_endio(bio);
907 	return;
908 fail:
909 	bio_io_error(bio);
910 }
911 
912 static long
913 __dcssblk_direct_access(struct dcssblk_dev_info *dev_info, pgoff_t pgoff,
914 		long nr_pages, void **kaddr, pfn_t *pfn)
915 {
916 	resource_size_t offset = pgoff * PAGE_SIZE;
917 	unsigned long dev_sz;
918 
919 	dev_sz = dev_info->end - dev_info->start + 1;
920 	if (kaddr)
921 		*kaddr = (void *) dev_info->start + offset;
922 	if (pfn)
923 		*pfn = __pfn_to_pfn_t(PFN_DOWN(dev_info->start + offset),
924 				PFN_DEV|PFN_SPECIAL);
925 
926 	return (dev_sz - offset) / PAGE_SIZE;
927 }
928 
929 static long
930 dcssblk_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
931 		long nr_pages, enum dax_access_mode mode, void **kaddr,
932 		pfn_t *pfn)
933 {
934 	struct dcssblk_dev_info *dev_info = dax_get_private(dax_dev);
935 
936 	return __dcssblk_direct_access(dev_info, pgoff, nr_pages, kaddr, pfn);
937 }
938 
939 static void
940 dcssblk_check_params(void)
941 {
942 	int rc, i, j, k;
943 	char buf[DCSSBLK_PARM_LEN + 1];
944 	struct dcssblk_dev_info *dev_info;
945 
946 	for (i = 0; (i < DCSSBLK_PARM_LEN) && (dcssblk_segments[i] != '\0');
947 	     i++) {
948 		for (j = i; (j < DCSSBLK_PARM_LEN) &&
949 			    (dcssblk_segments[j] != ',')  &&
950 			    (dcssblk_segments[j] != '\0') &&
951 			    (dcssblk_segments[j] != '('); j++)
952 		{
953 			buf[j-i] = dcssblk_segments[j];
954 		}
955 		buf[j-i] = '\0';
956 		rc = dcssblk_add_store(dcssblk_root_dev, NULL, buf, j-i);
957 		if ((rc >= 0) && (dcssblk_segments[j] == '(')) {
958 			for (k = 0; (buf[k] != ':') && (buf[k] != '\0'); k++)
959 				buf[k] = toupper(buf[k]);
960 			buf[k] = '\0';
961 			if (!strncmp(&dcssblk_segments[j], "(local)", 7)) {
962 				down_read(&dcssblk_devices_sem);
963 				dev_info = dcssblk_get_device_by_name(buf);
964 				up_read(&dcssblk_devices_sem);
965 				if (dev_info)
966 					dcssblk_shared_store(&dev_info->dev,
967 							     NULL, "0\n", 2);
968 			}
969 		}
970 		while ((dcssblk_segments[j] != ',') &&
971 		       (dcssblk_segments[j] != '\0'))
972 		{
973 			j++;
974 		}
975 		if (dcssblk_segments[j] == '\0')
976 			break;
977 		i = j;
978 	}
979 }
980 
981 /*
982  * The init/exit functions.
983  */
984 static void __exit
985 dcssblk_exit(void)
986 {
987 	root_device_unregister(dcssblk_root_dev);
988 	unregister_blkdev(dcssblk_major, DCSSBLK_NAME);
989 }
990 
991 static int __init
992 dcssblk_init(void)
993 {
994 	int rc;
995 
996 	dcssblk_root_dev = root_device_register("dcssblk");
997 	if (IS_ERR(dcssblk_root_dev))
998 		return PTR_ERR(dcssblk_root_dev);
999 	rc = device_create_file(dcssblk_root_dev, &dev_attr_add);
1000 	if (rc)
1001 		goto out_root;
1002 	rc = device_create_file(dcssblk_root_dev, &dev_attr_remove);
1003 	if (rc)
1004 		goto out_root;
1005 	rc = register_blkdev(0, DCSSBLK_NAME);
1006 	if (rc < 0)
1007 		goto out_root;
1008 	dcssblk_major = rc;
1009 	init_rwsem(&dcssblk_devices_sem);
1010 
1011 	dcssblk_check_params();
1012 	return 0;
1013 
1014 out_root:
1015 	root_device_unregister(dcssblk_root_dev);
1016 
1017 	return rc;
1018 }
1019 
1020 module_init(dcssblk_init);
1021 module_exit(dcssblk_exit);
1022 
1023 module_param_string(segments, dcssblk_segments, DCSSBLK_PARM_LEN, 0444);
1024 MODULE_PARM_DESC(segments, "Name of DCSS segment(s) to be loaded, "
1025 		 "comma-separated list, names in each set separated "
1026 		 "by commas are separated by colons, each set contains "
1027 		 "names of contiguous segments and each name max. 8 chars.\n"
1028 		 "Adding \"(local)\" to the end of each set equals echoing 0 "
1029 		 "to /sys/devices/dcssblk/<device name>/shared after loading "
1030 		 "the contiguous segments - \n"
1031 		 "e.g. segments=\"mydcss1,mydcss2:mydcss3,mydcss4(local)\"");
1032 
1033 MODULE_LICENSE("GPL");
1034