xref: /openbmc/linux/drivers/s390/block/dcssblk.c (revision fe4549b1)
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 	up_write(&dcssblk_devices_sem);
415 
416 	dax_remove_host(dev_info->gd);
417 	kill_dax(dev_info->dax_dev);
418 	put_dax(dev_info->dax_dev);
419 	del_gendisk(dev_info->gd);
420 	put_disk(dev_info->gd);
421 
422 	if (device_remove_file_self(dev, attr)) {
423 		device_unregister(dev);
424 		put_device(dev);
425 	}
426 	return rc;
427 out:
428 	up_write(&dcssblk_devices_sem);
429 	return rc;
430 }
431 static DEVICE_ATTR(shared, S_IWUSR | S_IRUSR, dcssblk_shared_show,
432 		   dcssblk_shared_store);
433 
434 /*
435  * device attribute for save operation on current copy
436  * of the segment. If the segment is busy, saving will
437  * become pending until it gets released, which can be
438  * undone by storing a non-true value to this entry.
439  * (show + store)
440  */
441 static ssize_t
442 dcssblk_save_show(struct device *dev, struct device_attribute *attr, char *buf)
443 {
444 	struct dcssblk_dev_info *dev_info;
445 
446 	dev_info = container_of(dev, struct dcssblk_dev_info, dev);
447 	return sprintf(buf, dev_info->save_pending ? "1\n" : "0\n");
448 }
449 
450 static ssize_t
451 dcssblk_save_store(struct device *dev, struct device_attribute *attr, const char *inbuf, size_t count)
452 {
453 	struct dcssblk_dev_info *dev_info;
454 	struct segment_info *entry;
455 
456 	if ((count > 1) && (inbuf[1] != '\n') && (inbuf[1] != '\0'))
457 		return -EINVAL;
458 	dev_info = container_of(dev, struct dcssblk_dev_info, dev);
459 
460 	down_write(&dcssblk_devices_sem);
461 	if (inbuf[0] == '1') {
462 		if (atomic_read(&dev_info->use_count) == 0) {
463 			// device is idle => we save immediately
464 			pr_info("All DCSSs that map to device %s are "
465 				"saved\n", dev_info->segment_name);
466 			list_for_each_entry(entry, &dev_info->seg_list, lh) {
467 				if (entry->segment_type == SEG_TYPE_EN ||
468 				    entry->segment_type == SEG_TYPE_SN)
469 					pr_warn("DCSS %s is of type SN or EN"
470 						" and cannot be saved\n",
471 						entry->segment_name);
472 				else
473 					segment_save(entry->segment_name);
474 			}
475 		}  else {
476 			// device is busy => we save it when it becomes
477 			// idle in dcssblk_release
478 			pr_info("Device %s is in use, its DCSSs will be "
479 				"saved when it becomes idle\n",
480 				dev_info->segment_name);
481 			dev_info->save_pending = 1;
482 		}
483 	} else if (inbuf[0] == '0') {
484 		if (dev_info->save_pending) {
485 			// device is busy & the user wants to undo his save
486 			// request
487 			dev_info->save_pending = 0;
488 			pr_info("A pending save request for device %s "
489 				"has been canceled\n",
490 				dev_info->segment_name);
491 		}
492 	} else {
493 		up_write(&dcssblk_devices_sem);
494 		return -EINVAL;
495 	}
496 	up_write(&dcssblk_devices_sem);
497 	return count;
498 }
499 static DEVICE_ATTR(save, S_IWUSR | S_IRUSR, dcssblk_save_show,
500 		   dcssblk_save_store);
501 
502 /*
503  * device attribute for showing all segments in a device
504  */
505 static ssize_t
506 dcssblk_seglist_show(struct device *dev, struct device_attribute *attr,
507 		char *buf)
508 {
509 	int i;
510 
511 	struct dcssblk_dev_info *dev_info;
512 	struct segment_info *entry;
513 
514 	down_read(&dcssblk_devices_sem);
515 	dev_info = container_of(dev, struct dcssblk_dev_info, dev);
516 	i = 0;
517 	buf[0] = '\0';
518 	list_for_each_entry(entry, &dev_info->seg_list, lh) {
519 		strcpy(&buf[i], entry->segment_name);
520 		i += strlen(entry->segment_name);
521 		buf[i] = '\n';
522 		i++;
523 	}
524 	up_read(&dcssblk_devices_sem);
525 	return i;
526 }
527 static DEVICE_ATTR(seglist, S_IRUSR, dcssblk_seglist_show, NULL);
528 
529 static struct attribute *dcssblk_dev_attrs[] = {
530 	&dev_attr_shared.attr,
531 	&dev_attr_save.attr,
532 	&dev_attr_seglist.attr,
533 	NULL,
534 };
535 static struct attribute_group dcssblk_dev_attr_group = {
536 	.attrs = dcssblk_dev_attrs,
537 };
538 static const struct attribute_group *dcssblk_dev_attr_groups[] = {
539 	&dcssblk_dev_attr_group,
540 	NULL,
541 };
542 
543 /*
544  * device attribute for adding devices
545  */
546 static ssize_t
547 dcssblk_add_store(struct device *dev, struct device_attribute *attr, const char *buf, size_t count)
548 {
549 	int rc, i, j, num_of_segments;
550 	struct dcssblk_dev_info *dev_info;
551 	struct segment_info *seg_info, *temp;
552 	char *local_buf;
553 	unsigned long seg_byte_size;
554 
555 	dev_info = NULL;
556 	seg_info = NULL;
557 	if (dev != dcssblk_root_dev) {
558 		rc = -EINVAL;
559 		goto out_nobuf;
560 	}
561 	if ((count < 1) || (buf[0] == '\0') || (buf[0] == '\n')) {
562 		rc = -ENAMETOOLONG;
563 		goto out_nobuf;
564 	}
565 
566 	local_buf = kmalloc(count + 1, GFP_KERNEL);
567 	if (local_buf == NULL) {
568 		rc = -ENOMEM;
569 		goto out_nobuf;
570 	}
571 
572 	/*
573 	 * parse input
574 	 */
575 	num_of_segments = 0;
576 	for (i = 0; (i < count && (buf[i] != '\0') && (buf[i] != '\n')); i++) {
577 		for (j = i; j < count &&
578 			(buf[j] != ':') &&
579 			(buf[j] != '\0') &&
580 			(buf[j] != '\n'); j++) {
581 			local_buf[j-i] = toupper(buf[j]);
582 		}
583 		local_buf[j-i] = '\0';
584 		if (((j - i) == 0) || ((j - i) > 8)) {
585 			rc = -ENAMETOOLONG;
586 			goto seg_list_del;
587 		}
588 
589 		rc = dcssblk_load_segment(local_buf, &seg_info);
590 		if (rc < 0)
591 			goto seg_list_del;
592 		/*
593 		 * get a struct dcssblk_dev_info
594 		 */
595 		if (num_of_segments == 0) {
596 			dev_info = kzalloc(sizeof(struct dcssblk_dev_info),
597 					GFP_KERNEL);
598 			if (dev_info == NULL) {
599 				rc = -ENOMEM;
600 				goto out;
601 			}
602 			strcpy(dev_info->segment_name, local_buf);
603 			dev_info->segment_type = seg_info->segment_type;
604 			INIT_LIST_HEAD(&dev_info->seg_list);
605 		}
606 		list_add_tail(&seg_info->lh, &dev_info->seg_list);
607 		num_of_segments++;
608 		i = j;
609 
610 		if ((buf[j] == '\0') || (buf[j] == '\n'))
611 			break;
612 	}
613 
614 	/* no trailing colon at the end of the input */
615 	if ((i > 0) && (buf[i-1] == ':')) {
616 		rc = -ENAMETOOLONG;
617 		goto seg_list_del;
618 	}
619 	strscpy(local_buf, buf, i + 1);
620 	dev_info->num_of_segments = num_of_segments;
621 	rc = dcssblk_is_continuous(dev_info);
622 	if (rc < 0)
623 		goto seg_list_del;
624 
625 	dev_info->start = dcssblk_find_lowest_addr(dev_info);
626 	dev_info->end = dcssblk_find_highest_addr(dev_info);
627 
628 	dev_set_name(&dev_info->dev, "%s", dev_info->segment_name);
629 	dev_info->dev.release = dcssblk_release_segment;
630 	dev_info->dev.groups = dcssblk_dev_attr_groups;
631 	INIT_LIST_HEAD(&dev_info->lh);
632 	dev_info->gd = blk_alloc_disk(NUMA_NO_NODE);
633 	if (dev_info->gd == NULL) {
634 		rc = -ENOMEM;
635 		goto seg_list_del;
636 	}
637 	dev_info->gd->major = dcssblk_major;
638 	dev_info->gd->minors = DCSSBLK_MINORS_PER_DISK;
639 	dev_info->gd->fops = &dcssblk_devops;
640 	dev_info->gd->private_data = dev_info;
641 	dev_info->gd->flags |= GENHD_FL_NO_PART;
642 	blk_queue_logical_block_size(dev_info->gd->queue, 4096);
643 	blk_queue_flag_set(QUEUE_FLAG_DAX, dev_info->gd->queue);
644 
645 	seg_byte_size = (dev_info->end - dev_info->start + 1);
646 	set_capacity(dev_info->gd, seg_byte_size >> 9); // size in sectors
647 	pr_info("Loaded %s with total size %lu bytes and capacity %lu "
648 		"sectors\n", local_buf, seg_byte_size, seg_byte_size >> 9);
649 
650 	dev_info->save_pending = 0;
651 	dev_info->is_shared = 1;
652 	dev_info->dev.parent = dcssblk_root_dev;
653 
654 	/*
655 	 *get minor, add to list
656 	 */
657 	down_write(&dcssblk_devices_sem);
658 	if (dcssblk_get_segment_by_name(local_buf)) {
659 		rc = -EEXIST;
660 		goto release_gd;
661 	}
662 	rc = dcssblk_assign_free_minor(dev_info);
663 	if (rc)
664 		goto release_gd;
665 	sprintf(dev_info->gd->disk_name, "dcssblk%d",
666 		dev_info->gd->first_minor);
667 	list_add_tail(&dev_info->lh, &dcssblk_devices);
668 
669 	if (!try_module_get(THIS_MODULE)) {
670 		rc = -ENODEV;
671 		goto dev_list_del;
672 	}
673 	/*
674 	 * register the device
675 	 */
676 	rc = device_register(&dev_info->dev);
677 	if (rc)
678 		goto put_dev;
679 
680 	dev_info->dax_dev = alloc_dax(dev_info, &dcssblk_dax_ops);
681 	if (IS_ERR(dev_info->dax_dev)) {
682 		rc = PTR_ERR(dev_info->dax_dev);
683 		dev_info->dax_dev = NULL;
684 		goto put_dev;
685 	}
686 	set_dax_synchronous(dev_info->dax_dev);
687 	rc = dax_add_host(dev_info->dax_dev, dev_info->gd);
688 	if (rc)
689 		goto out_dax;
690 
691 	get_device(&dev_info->dev);
692 	rc = device_add_disk(&dev_info->dev, dev_info->gd, NULL);
693 	if (rc)
694 		goto out_dax_host;
695 
696 	switch (dev_info->segment_type) {
697 		case SEG_TYPE_SR:
698 		case SEG_TYPE_ER:
699 		case SEG_TYPE_SC:
700 			set_disk_ro(dev_info->gd,1);
701 			break;
702 		default:
703 			set_disk_ro(dev_info->gd,0);
704 			break;
705 	}
706 	up_write(&dcssblk_devices_sem);
707 	rc = count;
708 	goto out;
709 
710 out_dax_host:
711 	put_device(&dev_info->dev);
712 	dax_remove_host(dev_info->gd);
713 out_dax:
714 	kill_dax(dev_info->dax_dev);
715 	put_dax(dev_info->dax_dev);
716 put_dev:
717 	list_del(&dev_info->lh);
718 	put_disk(dev_info->gd);
719 	list_for_each_entry(seg_info, &dev_info->seg_list, lh) {
720 		segment_unload(seg_info->segment_name);
721 	}
722 	put_device(&dev_info->dev);
723 	up_write(&dcssblk_devices_sem);
724 	goto out;
725 dev_list_del:
726 	list_del(&dev_info->lh);
727 release_gd:
728 	put_disk(dev_info->gd);
729 	up_write(&dcssblk_devices_sem);
730 seg_list_del:
731 	if (dev_info == NULL)
732 		goto out;
733 	list_for_each_entry_safe(seg_info, temp, &dev_info->seg_list, lh) {
734 		list_del(&seg_info->lh);
735 		segment_unload(seg_info->segment_name);
736 		kfree(seg_info);
737 	}
738 	kfree(dev_info);
739 out:
740 	kfree(local_buf);
741 out_nobuf:
742 	return rc;
743 }
744 
745 /*
746  * device attribute for removing devices
747  */
748 static ssize_t
749 dcssblk_remove_store(struct device *dev, struct device_attribute *attr, const char *buf, size_t count)
750 {
751 	struct dcssblk_dev_info *dev_info;
752 	struct segment_info *entry;
753 	int rc, i;
754 	char *local_buf;
755 
756 	if (dev != dcssblk_root_dev) {
757 		return -EINVAL;
758 	}
759 	local_buf = kmalloc(count + 1, GFP_KERNEL);
760 	if (local_buf == NULL) {
761 		return -ENOMEM;
762 	}
763 	/*
764 	 * parse input
765 	 */
766 	for (i = 0; (i < count && (*(buf+i)!='\0') && (*(buf+i)!='\n')); i++) {
767 		local_buf[i] = toupper(buf[i]);
768 	}
769 	local_buf[i] = '\0';
770 	if ((i == 0) || (i > 8)) {
771 		rc = -ENAMETOOLONG;
772 		goto out_buf;
773 	}
774 
775 	down_write(&dcssblk_devices_sem);
776 	dev_info = dcssblk_get_device_by_name(local_buf);
777 	if (dev_info == NULL) {
778 		up_write(&dcssblk_devices_sem);
779 		pr_warn("Device %s cannot be removed because it is not a known device\n",
780 			local_buf);
781 		rc = -ENODEV;
782 		goto out_buf;
783 	}
784 	if (atomic_read(&dev_info->use_count) != 0) {
785 		up_write(&dcssblk_devices_sem);
786 		pr_warn("Device %s cannot be removed while it is in use\n",
787 			local_buf);
788 		rc = -EBUSY;
789 		goto out_buf;
790 	}
791 
792 	list_del(&dev_info->lh);
793 	/* unload all related segments */
794 	list_for_each_entry(entry, &dev_info->seg_list, lh)
795 		segment_unload(entry->segment_name);
796 	up_write(&dcssblk_devices_sem);
797 
798 	dax_remove_host(dev_info->gd);
799 	kill_dax(dev_info->dax_dev);
800 	put_dax(dev_info->dax_dev);
801 	del_gendisk(dev_info->gd);
802 	put_disk(dev_info->gd);
803 
804 	device_unregister(&dev_info->dev);
805 	put_device(&dev_info->dev);
806 
807 	rc = count;
808 out_buf:
809 	kfree(local_buf);
810 	return rc;
811 }
812 
813 static int
814 dcssblk_open(struct gendisk *disk, blk_mode_t mode)
815 {
816 	struct dcssblk_dev_info *dev_info = disk->private_data;
817 	int rc;
818 
819 	if (NULL == dev_info) {
820 		rc = -ENODEV;
821 		goto out;
822 	}
823 	atomic_inc(&dev_info->use_count);
824 	rc = 0;
825 out:
826 	return rc;
827 }
828 
829 static void
830 dcssblk_release(struct gendisk *disk)
831 {
832 	struct dcssblk_dev_info *dev_info = disk->private_data;
833 	struct segment_info *entry;
834 
835 	if (!dev_info) {
836 		WARN_ON(1);
837 		return;
838 	}
839 	down_write(&dcssblk_devices_sem);
840 	if (atomic_dec_and_test(&dev_info->use_count)
841 	    && (dev_info->save_pending)) {
842 		pr_info("Device %s has become idle and is being saved "
843 			"now\n", dev_info->segment_name);
844 		list_for_each_entry(entry, &dev_info->seg_list, lh) {
845 			if (entry->segment_type == SEG_TYPE_EN ||
846 			    entry->segment_type == SEG_TYPE_SN)
847 				pr_warn("DCSS %s is of type SN or EN and cannot"
848 					" be saved\n", entry->segment_name);
849 			else
850 				segment_save(entry->segment_name);
851 		}
852 		dev_info->save_pending = 0;
853 	}
854 	up_write(&dcssblk_devices_sem);
855 }
856 
857 static void
858 dcssblk_submit_bio(struct bio *bio)
859 {
860 	struct dcssblk_dev_info *dev_info;
861 	struct bio_vec bvec;
862 	struct bvec_iter iter;
863 	unsigned long index;
864 	void *page_addr;
865 	unsigned long source_addr;
866 	unsigned long bytes_done;
867 
868 	bytes_done = 0;
869 	dev_info = bio->bi_bdev->bd_disk->private_data;
870 	if (dev_info == NULL)
871 		goto fail;
872 	if (!IS_ALIGNED(bio->bi_iter.bi_sector, 8) ||
873 	    !IS_ALIGNED(bio->bi_iter.bi_size, PAGE_SIZE))
874 		/* Request is not page-aligned. */
875 		goto fail;
876 	/* verify data transfer direction */
877 	if (dev_info->is_shared) {
878 		switch (dev_info->segment_type) {
879 		case SEG_TYPE_SR:
880 		case SEG_TYPE_ER:
881 		case SEG_TYPE_SC:
882 			/* cannot write to these segments */
883 			if (bio_data_dir(bio) == WRITE) {
884 				pr_warn("Writing to %s failed because it is a read-only device\n",
885 					dev_name(&dev_info->dev));
886 				goto fail;
887 			}
888 		}
889 	}
890 
891 	index = (bio->bi_iter.bi_sector >> 3);
892 	bio_for_each_segment(bvec, bio, iter) {
893 		page_addr = bvec_virt(&bvec);
894 		source_addr = dev_info->start + (index<<12) + bytes_done;
895 		if (unlikely(!IS_ALIGNED((unsigned long)page_addr, PAGE_SIZE) ||
896 			     !IS_ALIGNED(bvec.bv_len, PAGE_SIZE)))
897 			// More paranoia.
898 			goto fail;
899 		if (bio_data_dir(bio) == READ)
900 			memcpy(page_addr, __va(source_addr), bvec.bv_len);
901 		else
902 			memcpy(__va(source_addr), page_addr, bvec.bv_len);
903 		bytes_done += bvec.bv_len;
904 	}
905 	bio_endio(bio);
906 	return;
907 fail:
908 	bio_io_error(bio);
909 }
910 
911 static long
912 __dcssblk_direct_access(struct dcssblk_dev_info *dev_info, pgoff_t pgoff,
913 		long nr_pages, void **kaddr, pfn_t *pfn)
914 {
915 	resource_size_t offset = pgoff * PAGE_SIZE;
916 	unsigned long dev_sz;
917 
918 	dev_sz = dev_info->end - dev_info->start + 1;
919 	if (kaddr)
920 		*kaddr = (void *) dev_info->start + offset;
921 	if (pfn)
922 		*pfn = __pfn_to_pfn_t(PFN_DOWN(dev_info->start + offset),
923 				PFN_DEV|PFN_SPECIAL);
924 
925 	return (dev_sz - offset) / PAGE_SIZE;
926 }
927 
928 static long
929 dcssblk_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
930 		long nr_pages, enum dax_access_mode mode, void **kaddr,
931 		pfn_t *pfn)
932 {
933 	struct dcssblk_dev_info *dev_info = dax_get_private(dax_dev);
934 
935 	return __dcssblk_direct_access(dev_info, pgoff, nr_pages, kaddr, pfn);
936 }
937 
938 static void
939 dcssblk_check_params(void)
940 {
941 	int rc, i, j, k;
942 	char buf[DCSSBLK_PARM_LEN + 1];
943 	struct dcssblk_dev_info *dev_info;
944 
945 	for (i = 0; (i < DCSSBLK_PARM_LEN) && (dcssblk_segments[i] != '\0');
946 	     i++) {
947 		for (j = i; (j < DCSSBLK_PARM_LEN) &&
948 			    (dcssblk_segments[j] != ',')  &&
949 			    (dcssblk_segments[j] != '\0') &&
950 			    (dcssblk_segments[j] != '('); j++)
951 		{
952 			buf[j-i] = dcssblk_segments[j];
953 		}
954 		buf[j-i] = '\0';
955 		rc = dcssblk_add_store(dcssblk_root_dev, NULL, buf, j-i);
956 		if ((rc >= 0) && (dcssblk_segments[j] == '(')) {
957 			for (k = 0; (buf[k] != ':') && (buf[k] != '\0'); k++)
958 				buf[k] = toupper(buf[k]);
959 			buf[k] = '\0';
960 			if (!strncmp(&dcssblk_segments[j], "(local)", 7)) {
961 				down_read(&dcssblk_devices_sem);
962 				dev_info = dcssblk_get_device_by_name(buf);
963 				up_read(&dcssblk_devices_sem);
964 				if (dev_info)
965 					dcssblk_shared_store(&dev_info->dev,
966 							     NULL, "0\n", 2);
967 			}
968 		}
969 		while ((dcssblk_segments[j] != ',') &&
970 		       (dcssblk_segments[j] != '\0'))
971 		{
972 			j++;
973 		}
974 		if (dcssblk_segments[j] == '\0')
975 			break;
976 		i = j;
977 	}
978 }
979 
980 /*
981  * The init/exit functions.
982  */
983 static void __exit
984 dcssblk_exit(void)
985 {
986 	root_device_unregister(dcssblk_root_dev);
987 	unregister_blkdev(dcssblk_major, DCSSBLK_NAME);
988 }
989 
990 static int __init
991 dcssblk_init(void)
992 {
993 	int rc;
994 
995 	dcssblk_root_dev = root_device_register("dcssblk");
996 	if (IS_ERR(dcssblk_root_dev))
997 		return PTR_ERR(dcssblk_root_dev);
998 	rc = device_create_file(dcssblk_root_dev, &dev_attr_add);
999 	if (rc)
1000 		goto out_root;
1001 	rc = device_create_file(dcssblk_root_dev, &dev_attr_remove);
1002 	if (rc)
1003 		goto out_root;
1004 	rc = register_blkdev(0, DCSSBLK_NAME);
1005 	if (rc < 0)
1006 		goto out_root;
1007 	dcssblk_major = rc;
1008 	init_rwsem(&dcssblk_devices_sem);
1009 
1010 	dcssblk_check_params();
1011 	return 0;
1012 
1013 out_root:
1014 	root_device_unregister(dcssblk_root_dev);
1015 
1016 	return rc;
1017 }
1018 
1019 module_init(dcssblk_init);
1020 module_exit(dcssblk_exit);
1021 
1022 module_param_string(segments, dcssblk_segments, DCSSBLK_PARM_LEN, 0444);
1023 MODULE_PARM_DESC(segments, "Name of DCSS segment(s) to be loaded, "
1024 		 "comma-separated list, names in each set separated "
1025 		 "by commas are separated by colons, each set contains "
1026 		 "names of contiguous segments and each name max. 8 chars.\n"
1027 		 "Adding \"(local)\" to the end of each set equals echoing 0 "
1028 		 "to /sys/devices/dcssblk/<device name>/shared after loading "
1029 		 "the contiguous segments - \n"
1030 		 "e.g. segments=\"mydcss1,mydcss2:mydcss3,mydcss4(local)\"");
1031 
1032 MODULE_LICENSE("GPL");
1033