xref: /linux/arch/powerpc/platforms/pseries/papr_scm.c (revision 7f71507851fc7764b36a3221839607d3a45c2025)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 #define pr_fmt(fmt)	"papr-scm: " fmt
4 
5 #include <linux/of.h>
6 #include <linux/kernel.h>
7 #include <linux/module.h>
8 #include <linux/ioport.h>
9 #include <linux/seq_file.h>
10 #include <linux/slab.h>
11 #include <linux/ndctl.h>
12 #include <linux/sched.h>
13 #include <linux/libnvdimm.h>
14 #include <linux/platform_device.h>
15 #include <linux/delay.h>
16 #include <linux/seq_buf.h>
17 #include <linux/nd.h>
18 
19 #include <asm/plpar_wrappers.h>
20 #include <uapi/linux/papr_pdsm.h>
21 #include <linux/papr_scm.h>
22 #include <asm/mce.h>
23 #include <linux/unaligned.h>
24 #include <linux/perf_event.h>
25 
26 #define BIND_ANY_ADDR (~0ul)
27 
28 #define PAPR_SCM_DIMM_CMD_MASK \
29 	((1ul << ND_CMD_GET_CONFIG_SIZE) | \
30 	 (1ul << ND_CMD_GET_CONFIG_DATA) | \
31 	 (1ul << ND_CMD_SET_CONFIG_DATA) | \
32 	 (1ul << ND_CMD_CALL))
33 
34 /* Struct holding a single performance metric */
35 struct papr_scm_perf_stat {
36 	u8 stat_id[8];
37 	__be64 stat_val;
38 } __packed;
39 
40 /* Struct exchanged between kernel and PHYP for fetching drc perf stats */
41 struct papr_scm_perf_stats {
42 	u8 eye_catcher[8];
43 	/* Should be PAPR_SCM_PERF_STATS_VERSION */
44 	__be32 stats_version;
45 	/* Number of stats following */
46 	__be32 num_statistics;
47 	/* zero or more performance matrics */
48 	struct papr_scm_perf_stat scm_statistic[];
49 } __packed;
50 
51 /* private struct associated with each region */
52 struct papr_scm_priv {
53 	struct platform_device *pdev;
54 	struct device_node *dn;
55 	uint32_t drc_index;
56 	uint64_t blocks;
57 	uint64_t block_size;
58 	int metadata_size;
59 	bool is_volatile;
60 	bool hcall_flush_required;
61 
62 	uint64_t bound_addr;
63 
64 	struct nvdimm_bus_descriptor bus_desc;
65 	struct nvdimm_bus *bus;
66 	struct nvdimm *nvdimm;
67 	struct resource res;
68 	struct nd_region *region;
69 	struct nd_interleave_set nd_set;
70 	struct list_head region_list;
71 
72 	/* Protect dimm health data from concurrent read/writes */
73 	struct mutex health_mutex;
74 
75 	/* Last time the health information of the dimm was updated */
76 	unsigned long lasthealth_jiffies;
77 
78 	/* Health information for the dimm */
79 	u64 health_bitmap;
80 
81 	/* Holds the last known dirty shutdown counter value */
82 	u64 dirty_shutdown_counter;
83 
84 	/* length of the stat buffer as expected by phyp */
85 	size_t stat_buffer_len;
86 
87 	/* The bits which needs to be overridden */
88 	u64 health_bitmap_inject_mask;
89 };
90 
91 static int papr_scm_pmem_flush(struct nd_region *nd_region,
92 			       struct bio *bio __maybe_unused)
93 {
94 	struct papr_scm_priv *p = nd_region_provider_data(nd_region);
95 	unsigned long ret_buf[PLPAR_HCALL_BUFSIZE], token = 0;
96 	long rc;
97 
98 	dev_dbg(&p->pdev->dev, "flush drc 0x%x", p->drc_index);
99 
100 	do {
101 		rc = plpar_hcall(H_SCM_FLUSH, ret_buf, p->drc_index, token);
102 		token = ret_buf[0];
103 
104 		/* Check if we are stalled for some time */
105 		if (H_IS_LONG_BUSY(rc)) {
106 			msleep(get_longbusy_msecs(rc));
107 			rc = H_BUSY;
108 		} else if (rc == H_BUSY) {
109 			cond_resched();
110 		}
111 	} while (rc == H_BUSY);
112 
113 	if (rc) {
114 		dev_err(&p->pdev->dev, "flush error: %ld", rc);
115 		rc = -EIO;
116 	} else {
117 		dev_dbg(&p->pdev->dev, "flush drc 0x%x complete", p->drc_index);
118 	}
119 
120 	return rc;
121 }
122 
123 static LIST_HEAD(papr_nd_regions);
124 static DEFINE_MUTEX(papr_ndr_lock);
125 
126 static int drc_pmem_bind(struct papr_scm_priv *p)
127 {
128 	unsigned long ret[PLPAR_HCALL_BUFSIZE];
129 	uint64_t saved = 0;
130 	uint64_t token;
131 	int64_t rc;
132 
133 	/*
134 	 * When the hypervisor cannot map all the requested memory in a single
135 	 * hcall it returns H_BUSY and we call again with the token until
136 	 * we get H_SUCCESS. Aborting the retry loop before getting H_SUCCESS
137 	 * leave the system in an undefined state, so we wait.
138 	 */
139 	token = 0;
140 
141 	do {
142 		rc = plpar_hcall(H_SCM_BIND_MEM, ret, p->drc_index, 0,
143 				p->blocks, BIND_ANY_ADDR, token);
144 		token = ret[0];
145 		if (!saved)
146 			saved = ret[1];
147 		cond_resched();
148 	} while (rc == H_BUSY);
149 
150 	if (rc)
151 		return rc;
152 
153 	p->bound_addr = saved;
154 	dev_dbg(&p->pdev->dev, "bound drc 0x%x to 0x%lx\n",
155 		p->drc_index, (unsigned long)saved);
156 	return rc;
157 }
158 
159 static void drc_pmem_unbind(struct papr_scm_priv *p)
160 {
161 	unsigned long ret[PLPAR_HCALL_BUFSIZE];
162 	uint64_t token = 0;
163 	int64_t rc;
164 
165 	dev_dbg(&p->pdev->dev, "unbind drc 0x%x\n", p->drc_index);
166 
167 	/* NB: unbind has the same retry requirements as drc_pmem_bind() */
168 	do {
169 
170 		/* Unbind of all SCM resources associated with drcIndex */
171 		rc = plpar_hcall(H_SCM_UNBIND_ALL, ret, H_UNBIND_SCOPE_DRC,
172 				 p->drc_index, token);
173 		token = ret[0];
174 
175 		/* Check if we are stalled for some time */
176 		if (H_IS_LONG_BUSY(rc)) {
177 			msleep(get_longbusy_msecs(rc));
178 			rc = H_BUSY;
179 		} else if (rc == H_BUSY) {
180 			cond_resched();
181 		}
182 
183 	} while (rc == H_BUSY);
184 
185 	if (rc)
186 		dev_err(&p->pdev->dev, "unbind error: %lld\n", rc);
187 	else
188 		dev_dbg(&p->pdev->dev, "unbind drc 0x%x complete\n",
189 			p->drc_index);
190 
191 	return;
192 }
193 
194 static int drc_pmem_query_n_bind(struct papr_scm_priv *p)
195 {
196 	unsigned long start_addr;
197 	unsigned long end_addr;
198 	unsigned long ret[PLPAR_HCALL_BUFSIZE];
199 	int64_t rc;
200 
201 
202 	rc = plpar_hcall(H_SCM_QUERY_BLOCK_MEM_BINDING, ret,
203 			 p->drc_index, 0);
204 	if (rc)
205 		goto err_out;
206 	start_addr = ret[0];
207 
208 	/* Make sure the full region is bound. */
209 	rc = plpar_hcall(H_SCM_QUERY_BLOCK_MEM_BINDING, ret,
210 			 p->drc_index, p->blocks - 1);
211 	if (rc)
212 		goto err_out;
213 	end_addr = ret[0];
214 
215 	if ((end_addr - start_addr) != ((p->blocks - 1) * p->block_size))
216 		goto err_out;
217 
218 	p->bound_addr = start_addr;
219 	dev_dbg(&p->pdev->dev, "bound drc 0x%x to 0x%lx\n", p->drc_index, start_addr);
220 	return rc;
221 
222 err_out:
223 	dev_info(&p->pdev->dev,
224 		 "Failed to query, trying an unbind followed by bind");
225 	drc_pmem_unbind(p);
226 	return drc_pmem_bind(p);
227 }
228 
229 /*
230  * Query the Dimm performance stats from PHYP and copy them (if returned) to
231  * provided struct papr_scm_perf_stats instance 'stats' that can hold atleast
232  * (num_stats + header) bytes.
233  * - If buff_stats == NULL the return value is the size in bytes of the buffer
234  * needed to hold all supported performance-statistics.
235  * - If buff_stats != NULL and num_stats == 0 then we copy all known
236  * performance-statistics to 'buff_stat' and expect to be large enough to
237  * hold them.
238  * - if buff_stats != NULL and num_stats > 0 then copy the requested
239  * performance-statistics to buff_stats.
240  */
241 static ssize_t drc_pmem_query_stats(struct papr_scm_priv *p,
242 				    struct papr_scm_perf_stats *buff_stats,
243 				    unsigned int num_stats)
244 {
245 	unsigned long ret[PLPAR_HCALL_BUFSIZE];
246 	size_t size;
247 	s64 rc;
248 
249 	/* Setup the out buffer */
250 	if (buff_stats) {
251 		memcpy(buff_stats->eye_catcher,
252 		       PAPR_SCM_PERF_STATS_EYECATCHER, 8);
253 		buff_stats->stats_version =
254 			cpu_to_be32(PAPR_SCM_PERF_STATS_VERSION);
255 		buff_stats->num_statistics =
256 			cpu_to_be32(num_stats);
257 
258 		/*
259 		 * Calculate the buffer size based on num-stats provided
260 		 * or use the prefetched max buffer length
261 		 */
262 		if (num_stats)
263 			/* Calculate size from the num_stats */
264 			size = sizeof(struct papr_scm_perf_stats) +
265 				num_stats * sizeof(struct papr_scm_perf_stat);
266 		else
267 			size = p->stat_buffer_len;
268 	} else {
269 		/* In case of no out buffer ignore the size */
270 		size = 0;
271 	}
272 
273 	/* Do the HCALL asking PHYP for info */
274 	rc = plpar_hcall(H_SCM_PERFORMANCE_STATS, ret, p->drc_index,
275 			 buff_stats ? virt_to_phys(buff_stats) : 0,
276 			 size);
277 
278 	/* Check if the error was due to an unknown stat-id */
279 	if (rc == H_PARTIAL) {
280 		dev_err(&p->pdev->dev,
281 			"Unknown performance stats, Err:0x%016lX\n", ret[0]);
282 		return -ENOENT;
283 	} else if (rc == H_AUTHORITY) {
284 		dev_info(&p->pdev->dev,
285 			 "Permission denied while accessing performance stats");
286 		return -EPERM;
287 	} else if (rc == H_UNSUPPORTED) {
288 		dev_dbg(&p->pdev->dev, "Performance stats unsupported\n");
289 		return -EOPNOTSUPP;
290 	} else if (rc != H_SUCCESS) {
291 		dev_err(&p->pdev->dev,
292 			"Failed to query performance stats, Err:%lld\n", rc);
293 		return -EIO;
294 
295 	} else if (!size) {
296 		/* Handle case where stat buffer size was requested */
297 		dev_dbg(&p->pdev->dev,
298 			"Performance stats size %ld\n", ret[0]);
299 		return ret[0];
300 	}
301 
302 	/* Successfully fetched the requested stats from phyp */
303 	dev_dbg(&p->pdev->dev,
304 		"Performance stats returned %d stats\n",
305 		be32_to_cpu(buff_stats->num_statistics));
306 	return 0;
307 }
308 
309 #ifdef CONFIG_PERF_EVENTS
310 #define to_nvdimm_pmu(_pmu)	container_of(_pmu, struct nvdimm_pmu, pmu)
311 
312 static const char * const nvdimm_events_map[] = {
313 	[1] = "CtlResCt",
314 	[2] = "CtlResTm",
315 	[3] = "PonSecs ",
316 	[4] = "MemLife ",
317 	[5] = "CritRscU",
318 	[6] = "HostLCnt",
319 	[7] = "HostSCnt",
320 	[8] = "HostSDur",
321 	[9] = "HostLDur",
322 	[10] = "MedRCnt ",
323 	[11] = "MedWCnt ",
324 	[12] = "MedRDur ",
325 	[13] = "MedWDur ",
326 	[14] = "CchRHCnt",
327 	[15] = "CchWHCnt",
328 	[16] = "FastWCnt",
329 };
330 
331 static int papr_scm_pmu_get_value(struct perf_event *event, struct device *dev, u64 *count)
332 {
333 	struct papr_scm_perf_stat *stat;
334 	struct papr_scm_perf_stats *stats;
335 	struct papr_scm_priv *p = dev_get_drvdata(dev);
336 	int rc, size;
337 
338 	/* Invalid eventcode */
339 	if (event->attr.config == 0 || event->attr.config >= ARRAY_SIZE(nvdimm_events_map))
340 		return -EINVAL;
341 
342 	/* Allocate request buffer enough to hold single performance stat */
343 	size = sizeof(struct papr_scm_perf_stats) +
344 		sizeof(struct papr_scm_perf_stat);
345 
346 	if (!p)
347 		return -EINVAL;
348 
349 	stats = kzalloc(size, GFP_KERNEL);
350 	if (!stats)
351 		return -ENOMEM;
352 
353 	stat = &stats->scm_statistic[0];
354 	memcpy(&stat->stat_id,
355 	       nvdimm_events_map[event->attr.config],
356 		sizeof(stat->stat_id));
357 	stat->stat_val = 0;
358 
359 	rc = drc_pmem_query_stats(p, stats, 1);
360 	if (rc < 0) {
361 		kfree(stats);
362 		return rc;
363 	}
364 
365 	*count = be64_to_cpu(stat->stat_val);
366 	kfree(stats);
367 	return 0;
368 }
369 
370 static int papr_scm_pmu_event_init(struct perf_event *event)
371 {
372 	struct nvdimm_pmu *nd_pmu = to_nvdimm_pmu(event->pmu);
373 	struct papr_scm_priv *p;
374 
375 	if (!nd_pmu)
376 		return -EINVAL;
377 
378 	/* test the event attr type for PMU enumeration */
379 	if (event->attr.type != event->pmu->type)
380 		return -ENOENT;
381 
382 	/* it does not support event sampling mode */
383 	if (is_sampling_event(event))
384 		return -EOPNOTSUPP;
385 
386 	/* no branch sampling */
387 	if (has_branch_stack(event))
388 		return -EOPNOTSUPP;
389 
390 	p = (struct papr_scm_priv *)nd_pmu->dev->driver_data;
391 	if (!p)
392 		return -EINVAL;
393 
394 	/* Invalid eventcode */
395 	if (event->attr.config == 0 || event->attr.config > 16)
396 		return -EINVAL;
397 
398 	return 0;
399 }
400 
401 static int papr_scm_pmu_add(struct perf_event *event, int flags)
402 {
403 	u64 count;
404 	int rc;
405 	struct nvdimm_pmu *nd_pmu = to_nvdimm_pmu(event->pmu);
406 
407 	if (!nd_pmu)
408 		return -EINVAL;
409 
410 	if (flags & PERF_EF_START) {
411 		rc = papr_scm_pmu_get_value(event, nd_pmu->dev, &count);
412 		if (rc)
413 			return rc;
414 
415 		local64_set(&event->hw.prev_count, count);
416 	}
417 
418 	return 0;
419 }
420 
421 static void papr_scm_pmu_read(struct perf_event *event)
422 {
423 	u64 prev, now;
424 	int rc;
425 	struct nvdimm_pmu *nd_pmu = to_nvdimm_pmu(event->pmu);
426 
427 	if (!nd_pmu)
428 		return;
429 
430 	rc = papr_scm_pmu_get_value(event, nd_pmu->dev, &now);
431 	if (rc)
432 		return;
433 
434 	prev = local64_xchg(&event->hw.prev_count, now);
435 	local64_add(now - prev, &event->count);
436 }
437 
438 static void papr_scm_pmu_del(struct perf_event *event, int flags)
439 {
440 	papr_scm_pmu_read(event);
441 }
442 
443 static void papr_scm_pmu_register(struct papr_scm_priv *p)
444 {
445 	struct nvdimm_pmu *nd_pmu;
446 	int rc, nodeid;
447 
448 	nd_pmu = kzalloc(sizeof(*nd_pmu), GFP_KERNEL);
449 	if (!nd_pmu) {
450 		rc = -ENOMEM;
451 		goto pmu_err_print;
452 	}
453 
454 	if (!p->stat_buffer_len) {
455 		rc = -ENOENT;
456 		goto pmu_check_events_err;
457 	}
458 
459 	nd_pmu->pmu.task_ctx_nr = perf_invalid_context;
460 	nd_pmu->pmu.name = nvdimm_name(p->nvdimm);
461 	nd_pmu->pmu.event_init = papr_scm_pmu_event_init;
462 	nd_pmu->pmu.read = papr_scm_pmu_read;
463 	nd_pmu->pmu.add = papr_scm_pmu_add;
464 	nd_pmu->pmu.del = papr_scm_pmu_del;
465 
466 	nd_pmu->pmu.capabilities = PERF_PMU_CAP_NO_INTERRUPT |
467 				PERF_PMU_CAP_NO_EXCLUDE;
468 
469 	/*updating the cpumask variable */
470 	nodeid = numa_map_to_online_node(dev_to_node(&p->pdev->dev));
471 	nd_pmu->arch_cpumask = *cpumask_of_node(nodeid);
472 
473 	rc = register_nvdimm_pmu(nd_pmu, p->pdev);
474 	if (rc)
475 		goto pmu_check_events_err;
476 
477 	/*
478 	 * Set archdata.priv value to nvdimm_pmu structure, to handle the
479 	 * unregistering of pmu device.
480 	 */
481 	p->pdev->archdata.priv = nd_pmu;
482 	return;
483 
484 pmu_check_events_err:
485 	kfree(nd_pmu);
486 pmu_err_print:
487 	dev_info(&p->pdev->dev, "nvdimm pmu didn't register rc=%d\n", rc);
488 }
489 
490 #else
491 static void papr_scm_pmu_register(struct papr_scm_priv *p) { }
492 #endif
493 
494 /*
495  * Issue hcall to retrieve dimm health info and populate papr_scm_priv with the
496  * health information.
497  */
498 static int __drc_pmem_query_health(struct papr_scm_priv *p)
499 {
500 	unsigned long ret[PLPAR_HCALL_BUFSIZE];
501 	u64 bitmap = 0;
502 	long rc;
503 
504 	/* issue the hcall */
505 	rc = plpar_hcall(H_SCM_HEALTH, ret, p->drc_index);
506 	if (rc == H_SUCCESS)
507 		bitmap = ret[0] & ret[1];
508 	else if (rc == H_FUNCTION)
509 		dev_info_once(&p->pdev->dev,
510 			      "Hcall H_SCM_HEALTH not implemented, assuming empty health bitmap");
511 	else {
512 
513 		dev_err(&p->pdev->dev,
514 			"Failed to query health information, Err:%ld\n", rc);
515 		return -ENXIO;
516 	}
517 
518 	p->lasthealth_jiffies = jiffies;
519 	/* Allow injecting specific health bits via inject mask. */
520 	if (p->health_bitmap_inject_mask)
521 		bitmap = (bitmap & ~p->health_bitmap_inject_mask) |
522 			p->health_bitmap_inject_mask;
523 	WRITE_ONCE(p->health_bitmap, bitmap);
524 	dev_dbg(&p->pdev->dev,
525 		"Queried dimm health info. Bitmap:0x%016lx Mask:0x%016lx\n",
526 		ret[0], ret[1]);
527 
528 	return 0;
529 }
530 
531 /* Min interval in seconds for assuming stable dimm health */
532 #define MIN_HEALTH_QUERY_INTERVAL 60
533 
534 /* Query cached health info and if needed call drc_pmem_query_health */
535 static int drc_pmem_query_health(struct papr_scm_priv *p)
536 {
537 	unsigned long cache_timeout;
538 	int rc;
539 
540 	/* Protect concurrent modifications to papr_scm_priv */
541 	rc = mutex_lock_interruptible(&p->health_mutex);
542 	if (rc)
543 		return rc;
544 
545 	/* Jiffies offset for which the health data is assumed to be same */
546 	cache_timeout = p->lasthealth_jiffies +
547 		msecs_to_jiffies(MIN_HEALTH_QUERY_INTERVAL * 1000);
548 
549 	/* Fetch new health info is its older than MIN_HEALTH_QUERY_INTERVAL */
550 	if (time_after(jiffies, cache_timeout))
551 		rc = __drc_pmem_query_health(p);
552 	else
553 		/* Assume cached health data is valid */
554 		rc = 0;
555 
556 	mutex_unlock(&p->health_mutex);
557 	return rc;
558 }
559 
560 static int papr_scm_meta_get(struct papr_scm_priv *p,
561 			     struct nd_cmd_get_config_data_hdr *hdr)
562 {
563 	unsigned long data[PLPAR_HCALL_BUFSIZE];
564 	unsigned long offset, data_offset;
565 	int len, read;
566 	int64_t ret;
567 
568 	if ((hdr->in_offset + hdr->in_length) > p->metadata_size)
569 		return -EINVAL;
570 
571 	for (len = hdr->in_length; len; len -= read) {
572 
573 		data_offset = hdr->in_length - len;
574 		offset = hdr->in_offset + data_offset;
575 
576 		if (len >= 8)
577 			read = 8;
578 		else if (len >= 4)
579 			read = 4;
580 		else if (len >= 2)
581 			read = 2;
582 		else
583 			read = 1;
584 
585 		ret = plpar_hcall(H_SCM_READ_METADATA, data, p->drc_index,
586 				  offset, read);
587 
588 		if (ret == H_PARAMETER) /* bad DRC index */
589 			return -ENODEV;
590 		if (ret)
591 			return -EINVAL; /* other invalid parameter */
592 
593 		switch (read) {
594 		case 8:
595 			*(uint64_t *)(hdr->out_buf + data_offset) = be64_to_cpu(data[0]);
596 			break;
597 		case 4:
598 			*(uint32_t *)(hdr->out_buf + data_offset) = be32_to_cpu(data[0] & 0xffffffff);
599 			break;
600 
601 		case 2:
602 			*(uint16_t *)(hdr->out_buf + data_offset) = be16_to_cpu(data[0] & 0xffff);
603 			break;
604 
605 		case 1:
606 			*(uint8_t *)(hdr->out_buf + data_offset) = (data[0] & 0xff);
607 			break;
608 		}
609 	}
610 	return 0;
611 }
612 
613 static int papr_scm_meta_set(struct papr_scm_priv *p,
614 			     struct nd_cmd_set_config_hdr *hdr)
615 {
616 	unsigned long offset, data_offset;
617 	int len, wrote;
618 	unsigned long data;
619 	__be64 data_be;
620 	int64_t ret;
621 
622 	if ((hdr->in_offset + hdr->in_length) > p->metadata_size)
623 		return -EINVAL;
624 
625 	for (len = hdr->in_length; len; len -= wrote) {
626 
627 		data_offset = hdr->in_length - len;
628 		offset = hdr->in_offset + data_offset;
629 
630 		if (len >= 8) {
631 			data = *(uint64_t *)(hdr->in_buf + data_offset);
632 			data_be = cpu_to_be64(data);
633 			wrote = 8;
634 		} else if (len >= 4) {
635 			data = *(uint32_t *)(hdr->in_buf + data_offset);
636 			data &= 0xffffffff;
637 			data_be = cpu_to_be32(data);
638 			wrote = 4;
639 		} else if (len >= 2) {
640 			data = *(uint16_t *)(hdr->in_buf + data_offset);
641 			data &= 0xffff;
642 			data_be = cpu_to_be16(data);
643 			wrote = 2;
644 		} else {
645 			data_be = *(uint8_t *)(hdr->in_buf + data_offset);
646 			data_be &= 0xff;
647 			wrote = 1;
648 		}
649 
650 		ret = plpar_hcall_norets(H_SCM_WRITE_METADATA, p->drc_index,
651 					 offset, data_be, wrote);
652 		if (ret == H_PARAMETER) /* bad DRC index */
653 			return -ENODEV;
654 		if (ret)
655 			return -EINVAL; /* other invalid parameter */
656 	}
657 
658 	return 0;
659 }
660 
661 /*
662  * Do a sanity checks on the inputs args to dimm-control function and return
663  * '0' if valid. Validation of PDSM payloads happens later in
664  * papr_scm_service_pdsm.
665  */
666 static int is_cmd_valid(struct nvdimm *nvdimm, unsigned int cmd, void *buf,
667 			unsigned int buf_len)
668 {
669 	unsigned long cmd_mask = PAPR_SCM_DIMM_CMD_MASK;
670 	struct nd_cmd_pkg *nd_cmd;
671 	struct papr_scm_priv *p;
672 	enum papr_pdsm pdsm;
673 
674 	/* Only dimm-specific calls are supported atm */
675 	if (!nvdimm)
676 		return -EINVAL;
677 
678 	/* get the provider data from struct nvdimm */
679 	p = nvdimm_provider_data(nvdimm);
680 
681 	if (!test_bit(cmd, &cmd_mask)) {
682 		dev_dbg(&p->pdev->dev, "Unsupported cmd=%u\n", cmd);
683 		return -EINVAL;
684 	}
685 
686 	/* For CMD_CALL verify pdsm request */
687 	if (cmd == ND_CMD_CALL) {
688 		/* Verify the envelope and envelop size */
689 		if (!buf ||
690 		    buf_len < (sizeof(struct nd_cmd_pkg) + ND_PDSM_HDR_SIZE)) {
691 			dev_dbg(&p->pdev->dev, "Invalid pkg size=%u\n",
692 				buf_len);
693 			return -EINVAL;
694 		}
695 
696 		/* Verify that the nd_cmd_pkg.nd_family is correct */
697 		nd_cmd = (struct nd_cmd_pkg *)buf;
698 
699 		if (nd_cmd->nd_family != NVDIMM_FAMILY_PAPR) {
700 			dev_dbg(&p->pdev->dev, "Invalid pkg family=0x%llx\n",
701 				nd_cmd->nd_family);
702 			return -EINVAL;
703 		}
704 
705 		pdsm = (enum papr_pdsm)nd_cmd->nd_command;
706 
707 		/* Verify if the pdsm command is valid */
708 		if (pdsm <= PAPR_PDSM_MIN || pdsm >= PAPR_PDSM_MAX) {
709 			dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Invalid PDSM\n",
710 				pdsm);
711 			return -EINVAL;
712 		}
713 
714 		/* Have enough space to hold returned 'nd_pkg_pdsm' header */
715 		if (nd_cmd->nd_size_out < ND_PDSM_HDR_SIZE) {
716 			dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Invalid payload\n",
717 				pdsm);
718 			return -EINVAL;
719 		}
720 	}
721 
722 	/* Let the command be further processed */
723 	return 0;
724 }
725 
726 static int papr_pdsm_fuel_gauge(struct papr_scm_priv *p,
727 				union nd_pdsm_payload *payload)
728 {
729 	int rc, size;
730 	u64 statval;
731 	struct papr_scm_perf_stat *stat;
732 	struct papr_scm_perf_stats *stats;
733 
734 	/* Silently fail if fetching performance metrics isn't  supported */
735 	if (!p->stat_buffer_len)
736 		return 0;
737 
738 	/* Allocate request buffer enough to hold single performance stat */
739 	size = sizeof(struct papr_scm_perf_stats) +
740 		sizeof(struct papr_scm_perf_stat);
741 
742 	stats = kzalloc(size, GFP_KERNEL);
743 	if (!stats)
744 		return -ENOMEM;
745 
746 	stat = &stats->scm_statistic[0];
747 	memcpy(&stat->stat_id, "MemLife ", sizeof(stat->stat_id));
748 	stat->stat_val = 0;
749 
750 	/* Fetch the fuel gauge and populate it in payload */
751 	rc = drc_pmem_query_stats(p, stats, 1);
752 	if (rc < 0) {
753 		dev_dbg(&p->pdev->dev, "Err(%d) fetching fuel gauge\n", rc);
754 		goto free_stats;
755 	}
756 
757 	statval = be64_to_cpu(stat->stat_val);
758 	dev_dbg(&p->pdev->dev,
759 		"Fetched fuel-gauge %llu", statval);
760 	payload->health.extension_flags |=
761 		PDSM_DIMM_HEALTH_RUN_GAUGE_VALID;
762 	payload->health.dimm_fuel_gauge = statval;
763 
764 	rc = sizeof(struct nd_papr_pdsm_health);
765 
766 free_stats:
767 	kfree(stats);
768 	return rc;
769 }
770 
771 /* Add the dirty-shutdown-counter value to the pdsm */
772 static int papr_pdsm_dsc(struct papr_scm_priv *p,
773 			 union nd_pdsm_payload *payload)
774 {
775 	payload->health.extension_flags |= PDSM_DIMM_DSC_VALID;
776 	payload->health.dimm_dsc = p->dirty_shutdown_counter;
777 
778 	return sizeof(struct nd_papr_pdsm_health);
779 }
780 
781 /* Fetch the DIMM health info and populate it in provided package. */
782 static int papr_pdsm_health(struct papr_scm_priv *p,
783 			    union nd_pdsm_payload *payload)
784 {
785 	int rc;
786 
787 	/* Ensure dimm health mutex is taken preventing concurrent access */
788 	rc = mutex_lock_interruptible(&p->health_mutex);
789 	if (rc)
790 		goto out;
791 
792 	/* Always fetch upto date dimm health data ignoring cached values */
793 	rc = __drc_pmem_query_health(p);
794 	if (rc) {
795 		mutex_unlock(&p->health_mutex);
796 		goto out;
797 	}
798 
799 	/* update health struct with various flags derived from health bitmap */
800 	payload->health = (struct nd_papr_pdsm_health) {
801 		.extension_flags = 0,
802 		.dimm_unarmed = !!(p->health_bitmap & PAPR_PMEM_UNARMED_MASK),
803 		.dimm_bad_shutdown = !!(p->health_bitmap & PAPR_PMEM_BAD_SHUTDOWN_MASK),
804 		.dimm_bad_restore = !!(p->health_bitmap & PAPR_PMEM_BAD_RESTORE_MASK),
805 		.dimm_scrubbed = !!(p->health_bitmap & PAPR_PMEM_SCRUBBED_AND_LOCKED),
806 		.dimm_locked = !!(p->health_bitmap & PAPR_PMEM_SCRUBBED_AND_LOCKED),
807 		.dimm_encrypted = !!(p->health_bitmap & PAPR_PMEM_ENCRYPTED),
808 		.dimm_health = PAPR_PDSM_DIMM_HEALTHY,
809 	};
810 
811 	/* Update field dimm_health based on health_bitmap flags */
812 	if (p->health_bitmap & PAPR_PMEM_HEALTH_FATAL)
813 		payload->health.dimm_health = PAPR_PDSM_DIMM_FATAL;
814 	else if (p->health_bitmap & PAPR_PMEM_HEALTH_CRITICAL)
815 		payload->health.dimm_health = PAPR_PDSM_DIMM_CRITICAL;
816 	else if (p->health_bitmap & PAPR_PMEM_HEALTH_UNHEALTHY)
817 		payload->health.dimm_health = PAPR_PDSM_DIMM_UNHEALTHY;
818 
819 	/* struct populated hence can release the mutex now */
820 	mutex_unlock(&p->health_mutex);
821 
822 	/* Populate the fuel gauge meter in the payload */
823 	papr_pdsm_fuel_gauge(p, payload);
824 	/* Populate the dirty-shutdown-counter field */
825 	papr_pdsm_dsc(p, payload);
826 
827 	rc = sizeof(struct nd_papr_pdsm_health);
828 
829 out:
830 	return rc;
831 }
832 
833 /* Inject a smart error Add the dirty-shutdown-counter value to the pdsm */
834 static int papr_pdsm_smart_inject(struct papr_scm_priv *p,
835 				  union nd_pdsm_payload *payload)
836 {
837 	int rc;
838 	u32 supported_flags = 0;
839 	u64 inject_mask = 0, clear_mask = 0;
840 	u64 mask;
841 
842 	/* Check for individual smart error flags and update inject/clear masks */
843 	if (payload->smart_inject.flags & PDSM_SMART_INJECT_HEALTH_FATAL) {
844 		supported_flags |= PDSM_SMART_INJECT_HEALTH_FATAL;
845 		if (payload->smart_inject.fatal_enable)
846 			inject_mask |= PAPR_PMEM_HEALTH_FATAL;
847 		else
848 			clear_mask |= PAPR_PMEM_HEALTH_FATAL;
849 	}
850 
851 	if (payload->smart_inject.flags & PDSM_SMART_INJECT_BAD_SHUTDOWN) {
852 		supported_flags |= PDSM_SMART_INJECT_BAD_SHUTDOWN;
853 		if (payload->smart_inject.unsafe_shutdown_enable)
854 			inject_mask |= PAPR_PMEM_SHUTDOWN_DIRTY;
855 		else
856 			clear_mask |= PAPR_PMEM_SHUTDOWN_DIRTY;
857 	}
858 
859 	dev_dbg(&p->pdev->dev, "[Smart-inject] inject_mask=%#llx clear_mask=%#llx\n",
860 		inject_mask, clear_mask);
861 
862 	/* Prevent concurrent access to dimm health bitmap related members */
863 	rc = mutex_lock_interruptible(&p->health_mutex);
864 	if (rc)
865 		return rc;
866 
867 	/* Use inject/clear masks to set health_bitmap_inject_mask */
868 	mask = READ_ONCE(p->health_bitmap_inject_mask);
869 	mask = (mask & ~clear_mask) | inject_mask;
870 	WRITE_ONCE(p->health_bitmap_inject_mask, mask);
871 
872 	/* Invalidate cached health bitmap */
873 	p->lasthealth_jiffies = 0;
874 
875 	mutex_unlock(&p->health_mutex);
876 
877 	/* Return the supported flags back to userspace */
878 	payload->smart_inject.flags = supported_flags;
879 
880 	return sizeof(struct nd_papr_pdsm_health);
881 }
882 
883 /*
884  * 'struct pdsm_cmd_desc'
885  * Identifies supported PDSMs' expected length of in/out payloads
886  * and pdsm service function.
887  *
888  * size_in	: Size of input payload if any in the PDSM request.
889  * size_out	: Size of output payload if any in the PDSM request.
890  * service	: Service function for the PDSM request. Return semantics:
891  *		  rc < 0 : Error servicing PDSM and rc indicates the error.
892  *		  rc >=0 : Serviced successfully and 'rc' indicate number of
893  *			bytes written to payload.
894  */
895 struct pdsm_cmd_desc {
896 	u32 size_in;
897 	u32 size_out;
898 	int (*service)(struct papr_scm_priv *dimm,
899 		       union nd_pdsm_payload *payload);
900 };
901 
902 /* Holds all supported PDSMs' command descriptors */
903 static const struct pdsm_cmd_desc __pdsm_cmd_descriptors[] = {
904 	[PAPR_PDSM_MIN] = {
905 		.size_in = 0,
906 		.size_out = 0,
907 		.service = NULL,
908 	},
909 	/* New PDSM command descriptors to be added below */
910 
911 	[PAPR_PDSM_HEALTH] = {
912 		.size_in = 0,
913 		.size_out = sizeof(struct nd_papr_pdsm_health),
914 		.service = papr_pdsm_health,
915 	},
916 
917 	[PAPR_PDSM_SMART_INJECT] = {
918 		.size_in = sizeof(struct nd_papr_pdsm_smart_inject),
919 		.size_out = sizeof(struct nd_papr_pdsm_smart_inject),
920 		.service = papr_pdsm_smart_inject,
921 	},
922 	/* Empty */
923 	[PAPR_PDSM_MAX] = {
924 		.size_in = 0,
925 		.size_out = 0,
926 		.service = NULL,
927 	},
928 };
929 
930 /* Given a valid pdsm cmd return its command descriptor else return NULL */
931 static inline const struct pdsm_cmd_desc *pdsm_cmd_desc(enum papr_pdsm cmd)
932 {
933 	if (cmd >= 0 || cmd < ARRAY_SIZE(__pdsm_cmd_descriptors))
934 		return &__pdsm_cmd_descriptors[cmd];
935 
936 	return NULL;
937 }
938 
939 /*
940  * For a given pdsm request call an appropriate service function.
941  * Returns errors if any while handling the pdsm command package.
942  */
943 static int papr_scm_service_pdsm(struct papr_scm_priv *p,
944 				 struct nd_cmd_pkg *pkg)
945 {
946 	/* Get the PDSM header and PDSM command */
947 	struct nd_pkg_pdsm *pdsm_pkg = (struct nd_pkg_pdsm *)pkg->nd_payload;
948 	enum papr_pdsm pdsm = (enum papr_pdsm)pkg->nd_command;
949 	const struct pdsm_cmd_desc *pdsc;
950 	int rc;
951 
952 	/* Fetch corresponding pdsm descriptor for validation and servicing */
953 	pdsc = pdsm_cmd_desc(pdsm);
954 
955 	/* Validate pdsm descriptor */
956 	/* Ensure that reserved fields are 0 */
957 	if (pdsm_pkg->reserved[0] || pdsm_pkg->reserved[1]) {
958 		dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Invalid reserved field\n",
959 			pdsm);
960 		return -EINVAL;
961 	}
962 
963 	/* If pdsm expects some input, then ensure that the size_in matches */
964 	if (pdsc->size_in &&
965 	    pkg->nd_size_in != (pdsc->size_in + ND_PDSM_HDR_SIZE)) {
966 		dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Mismatched size_in=%d\n",
967 			pdsm, pkg->nd_size_in);
968 		return -EINVAL;
969 	}
970 
971 	/* If pdsm wants to return data, then ensure that  size_out matches */
972 	if (pdsc->size_out &&
973 	    pkg->nd_size_out != (pdsc->size_out + ND_PDSM_HDR_SIZE)) {
974 		dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Mismatched size_out=%d\n",
975 			pdsm, pkg->nd_size_out);
976 		return -EINVAL;
977 	}
978 
979 	/* Service the pdsm */
980 	if (pdsc->service) {
981 		dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Servicing..\n", pdsm);
982 
983 		rc = pdsc->service(p, &pdsm_pkg->payload);
984 
985 		if (rc < 0) {
986 			/* error encountered while servicing pdsm */
987 			pdsm_pkg->cmd_status = rc;
988 			pkg->nd_fw_size = ND_PDSM_HDR_SIZE;
989 		} else {
990 			/* pdsm serviced and 'rc' bytes written to payload */
991 			pdsm_pkg->cmd_status = 0;
992 			pkg->nd_fw_size = ND_PDSM_HDR_SIZE + rc;
993 		}
994 	} else {
995 		dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Unsupported PDSM request\n",
996 			pdsm);
997 		pdsm_pkg->cmd_status = -ENOENT;
998 		pkg->nd_fw_size = ND_PDSM_HDR_SIZE;
999 	}
1000 
1001 	return pdsm_pkg->cmd_status;
1002 }
1003 
1004 static int papr_scm_ndctl(struct nvdimm_bus_descriptor *nd_desc,
1005 			  struct nvdimm *nvdimm, unsigned int cmd, void *buf,
1006 			  unsigned int buf_len, int *cmd_rc)
1007 {
1008 	struct nd_cmd_get_config_size *get_size_hdr;
1009 	struct nd_cmd_pkg *call_pkg = NULL;
1010 	struct papr_scm_priv *p;
1011 	int rc;
1012 
1013 	rc = is_cmd_valid(nvdimm, cmd, buf, buf_len);
1014 	if (rc) {
1015 		pr_debug("Invalid cmd=0x%x. Err=%d\n", cmd, rc);
1016 		return rc;
1017 	}
1018 
1019 	/* Use a local variable in case cmd_rc pointer is NULL */
1020 	if (!cmd_rc)
1021 		cmd_rc = &rc;
1022 
1023 	p = nvdimm_provider_data(nvdimm);
1024 
1025 	switch (cmd) {
1026 	case ND_CMD_GET_CONFIG_SIZE:
1027 		get_size_hdr = buf;
1028 
1029 		get_size_hdr->status = 0;
1030 		get_size_hdr->max_xfer = 8;
1031 		get_size_hdr->config_size = p->metadata_size;
1032 		*cmd_rc = 0;
1033 		break;
1034 
1035 	case ND_CMD_GET_CONFIG_DATA:
1036 		*cmd_rc = papr_scm_meta_get(p, buf);
1037 		break;
1038 
1039 	case ND_CMD_SET_CONFIG_DATA:
1040 		*cmd_rc = papr_scm_meta_set(p, buf);
1041 		break;
1042 
1043 	case ND_CMD_CALL:
1044 		call_pkg = (struct nd_cmd_pkg *)buf;
1045 		*cmd_rc = papr_scm_service_pdsm(p, call_pkg);
1046 		break;
1047 
1048 	default:
1049 		dev_dbg(&p->pdev->dev, "Unknown command = %d\n", cmd);
1050 		return -EINVAL;
1051 	}
1052 
1053 	dev_dbg(&p->pdev->dev, "returned with cmd_rc = %d\n", *cmd_rc);
1054 
1055 	return 0;
1056 }
1057 
1058 static ssize_t health_bitmap_inject_show(struct device *dev,
1059 					 struct device_attribute *attr,
1060 					 char *buf)
1061 {
1062 	struct nvdimm *dimm = to_nvdimm(dev);
1063 	struct papr_scm_priv *p = nvdimm_provider_data(dimm);
1064 
1065 	return sprintf(buf, "%#llx\n",
1066 		       READ_ONCE(p->health_bitmap_inject_mask));
1067 }
1068 
1069 static DEVICE_ATTR_ADMIN_RO(health_bitmap_inject);
1070 
1071 static ssize_t perf_stats_show(struct device *dev,
1072 			       struct device_attribute *attr, char *buf)
1073 {
1074 	int index;
1075 	ssize_t rc;
1076 	struct seq_buf s;
1077 	struct papr_scm_perf_stat *stat;
1078 	struct papr_scm_perf_stats *stats;
1079 	struct nvdimm *dimm = to_nvdimm(dev);
1080 	struct papr_scm_priv *p = nvdimm_provider_data(dimm);
1081 
1082 	if (!p->stat_buffer_len)
1083 		return -ENOENT;
1084 
1085 	/* Allocate the buffer for phyp where stats are written */
1086 	stats = kzalloc(p->stat_buffer_len, GFP_KERNEL);
1087 	if (!stats)
1088 		return -ENOMEM;
1089 
1090 	/* Ask phyp to return all dimm perf stats */
1091 	rc = drc_pmem_query_stats(p, stats, 0);
1092 	if (rc)
1093 		goto free_stats;
1094 	/*
1095 	 * Go through the returned output buffer and print stats and
1096 	 * values. Since stat_id is essentially a char string of
1097 	 * 8 bytes, simply use the string format specifier to print it.
1098 	 */
1099 	seq_buf_init(&s, buf, PAGE_SIZE);
1100 	for (index = 0, stat = stats->scm_statistic;
1101 	     index < be32_to_cpu(stats->num_statistics);
1102 	     ++index, ++stat) {
1103 		seq_buf_printf(&s, "%.8s = 0x%016llX\n",
1104 			       stat->stat_id,
1105 			       be64_to_cpu(stat->stat_val));
1106 	}
1107 
1108 free_stats:
1109 	kfree(stats);
1110 	return rc ? rc : (ssize_t)seq_buf_used(&s);
1111 }
1112 static DEVICE_ATTR_ADMIN_RO(perf_stats);
1113 
1114 static ssize_t flags_show(struct device *dev,
1115 			  struct device_attribute *attr, char *buf)
1116 {
1117 	struct nvdimm *dimm = to_nvdimm(dev);
1118 	struct papr_scm_priv *p = nvdimm_provider_data(dimm);
1119 	struct seq_buf s;
1120 	u64 health;
1121 	int rc;
1122 
1123 	rc = drc_pmem_query_health(p);
1124 	if (rc)
1125 		return rc;
1126 
1127 	/* Copy health_bitmap locally, check masks & update out buffer */
1128 	health = READ_ONCE(p->health_bitmap);
1129 
1130 	seq_buf_init(&s, buf, PAGE_SIZE);
1131 	if (health & PAPR_PMEM_UNARMED_MASK)
1132 		seq_buf_printf(&s, "not_armed ");
1133 
1134 	if (health & PAPR_PMEM_BAD_SHUTDOWN_MASK)
1135 		seq_buf_printf(&s, "flush_fail ");
1136 
1137 	if (health & PAPR_PMEM_BAD_RESTORE_MASK)
1138 		seq_buf_printf(&s, "restore_fail ");
1139 
1140 	if (health & PAPR_PMEM_ENCRYPTED)
1141 		seq_buf_printf(&s, "encrypted ");
1142 
1143 	if (health & PAPR_PMEM_SMART_EVENT_MASK)
1144 		seq_buf_printf(&s, "smart_notify ");
1145 
1146 	if (health & PAPR_PMEM_SCRUBBED_AND_LOCKED)
1147 		seq_buf_printf(&s, "scrubbed locked ");
1148 
1149 	if (seq_buf_used(&s))
1150 		seq_buf_printf(&s, "\n");
1151 
1152 	return seq_buf_used(&s);
1153 }
1154 DEVICE_ATTR_RO(flags);
1155 
1156 static ssize_t dirty_shutdown_show(struct device *dev,
1157 			  struct device_attribute *attr, char *buf)
1158 {
1159 	struct nvdimm *dimm = to_nvdimm(dev);
1160 	struct papr_scm_priv *p = nvdimm_provider_data(dimm);
1161 
1162 	return sysfs_emit(buf, "%llu\n", p->dirty_shutdown_counter);
1163 }
1164 DEVICE_ATTR_RO(dirty_shutdown);
1165 
1166 static umode_t papr_nd_attribute_visible(struct kobject *kobj,
1167 					 struct attribute *attr, int n)
1168 {
1169 	struct device *dev = kobj_to_dev(kobj);
1170 	struct nvdimm *nvdimm = to_nvdimm(dev);
1171 	struct papr_scm_priv *p = nvdimm_provider_data(nvdimm);
1172 
1173 	/* For if perf-stats not available remove perf_stats sysfs */
1174 	if (attr == &dev_attr_perf_stats.attr && p->stat_buffer_len == 0)
1175 		return 0;
1176 
1177 	return attr->mode;
1178 }
1179 
1180 /* papr_scm specific dimm attributes */
1181 static struct attribute *papr_nd_attributes[] = {
1182 	&dev_attr_flags.attr,
1183 	&dev_attr_perf_stats.attr,
1184 	&dev_attr_dirty_shutdown.attr,
1185 	&dev_attr_health_bitmap_inject.attr,
1186 	NULL,
1187 };
1188 
1189 static const struct attribute_group papr_nd_attribute_group = {
1190 	.name = "papr",
1191 	.is_visible = papr_nd_attribute_visible,
1192 	.attrs = papr_nd_attributes,
1193 };
1194 
1195 static const struct attribute_group *papr_nd_attr_groups[] = {
1196 	&papr_nd_attribute_group,
1197 	NULL,
1198 };
1199 
1200 static int papr_scm_nvdimm_init(struct papr_scm_priv *p)
1201 {
1202 	struct device *dev = &p->pdev->dev;
1203 	struct nd_mapping_desc mapping;
1204 	struct nd_region_desc ndr_desc;
1205 	unsigned long dimm_flags;
1206 	int target_nid, online_nid;
1207 
1208 	p->bus_desc.ndctl = papr_scm_ndctl;
1209 	p->bus_desc.module = THIS_MODULE;
1210 	p->bus_desc.of_node = p->pdev->dev.of_node;
1211 	p->bus_desc.provider_name = kstrdup(p->pdev->name, GFP_KERNEL);
1212 
1213 	/* Set the dimm command family mask to accept PDSMs */
1214 	set_bit(NVDIMM_FAMILY_PAPR, &p->bus_desc.dimm_family_mask);
1215 
1216 	if (!p->bus_desc.provider_name)
1217 		return -ENOMEM;
1218 
1219 	p->bus = nvdimm_bus_register(NULL, &p->bus_desc);
1220 	if (!p->bus) {
1221 		dev_err(dev, "Error creating nvdimm bus %pOF\n", p->dn);
1222 		kfree(p->bus_desc.provider_name);
1223 		return -ENXIO;
1224 	}
1225 
1226 	dimm_flags = 0;
1227 	set_bit(NDD_LABELING, &dimm_flags);
1228 
1229 	/*
1230 	 * Check if the nvdimm is unarmed. No locking needed as we are still
1231 	 * initializing. Ignore error encountered if any.
1232 	 */
1233 	__drc_pmem_query_health(p);
1234 
1235 	if (p->health_bitmap & PAPR_PMEM_UNARMED_MASK)
1236 		set_bit(NDD_UNARMED, &dimm_flags);
1237 
1238 	p->nvdimm = nvdimm_create(p->bus, p, papr_nd_attr_groups,
1239 				  dimm_flags, PAPR_SCM_DIMM_CMD_MASK, 0, NULL);
1240 	if (!p->nvdimm) {
1241 		dev_err(dev, "Error creating DIMM object for %pOF\n", p->dn);
1242 		goto err;
1243 	}
1244 
1245 	if (nvdimm_bus_check_dimm_count(p->bus, 1))
1246 		goto err;
1247 
1248 	/* now add the region */
1249 
1250 	memset(&mapping, 0, sizeof(mapping));
1251 	mapping.nvdimm = p->nvdimm;
1252 	mapping.start = 0;
1253 	mapping.size = p->blocks * p->block_size; // XXX: potential overflow?
1254 
1255 	memset(&ndr_desc, 0, sizeof(ndr_desc));
1256 	target_nid = dev_to_node(&p->pdev->dev);
1257 	online_nid = numa_map_to_online_node(target_nid);
1258 	ndr_desc.numa_node = online_nid;
1259 	ndr_desc.target_node = target_nid;
1260 	ndr_desc.res = &p->res;
1261 	ndr_desc.of_node = p->dn;
1262 	ndr_desc.provider_data = p;
1263 	ndr_desc.mapping = &mapping;
1264 	ndr_desc.num_mappings = 1;
1265 	ndr_desc.nd_set = &p->nd_set;
1266 
1267 	if (p->hcall_flush_required) {
1268 		set_bit(ND_REGION_ASYNC, &ndr_desc.flags);
1269 		ndr_desc.flush = papr_scm_pmem_flush;
1270 	}
1271 
1272 	if (p->is_volatile)
1273 		p->region = nvdimm_volatile_region_create(p->bus, &ndr_desc);
1274 	else {
1275 		set_bit(ND_REGION_PERSIST_MEMCTRL, &ndr_desc.flags);
1276 		p->region = nvdimm_pmem_region_create(p->bus, &ndr_desc);
1277 	}
1278 	if (!p->region) {
1279 		dev_err(dev, "Error registering region %pR from %pOF\n",
1280 				ndr_desc.res, p->dn);
1281 		goto err;
1282 	}
1283 	if (target_nid != online_nid)
1284 		dev_info(dev, "Region registered with target node %d and online node %d",
1285 			 target_nid, online_nid);
1286 
1287 	mutex_lock(&papr_ndr_lock);
1288 	list_add_tail(&p->region_list, &papr_nd_regions);
1289 	mutex_unlock(&papr_ndr_lock);
1290 
1291 	return 0;
1292 
1293 err:	nvdimm_bus_unregister(p->bus);
1294 	kfree(p->bus_desc.provider_name);
1295 	return -ENXIO;
1296 }
1297 
1298 static void papr_scm_add_badblock(struct nd_region *region,
1299 				  struct nvdimm_bus *bus, u64 phys_addr)
1300 {
1301 	u64 aligned_addr = ALIGN_DOWN(phys_addr, L1_CACHE_BYTES);
1302 
1303 	if (nvdimm_bus_add_badrange(bus, aligned_addr, L1_CACHE_BYTES)) {
1304 		pr_err("Bad block registration for 0x%llx failed\n", phys_addr);
1305 		return;
1306 	}
1307 
1308 	pr_debug("Add memory range (0x%llx - 0x%llx) as bad range\n",
1309 		 aligned_addr, aligned_addr + L1_CACHE_BYTES);
1310 
1311 	nvdimm_region_notify(region, NVDIMM_REVALIDATE_POISON);
1312 }
1313 
1314 static int handle_mce_ue(struct notifier_block *nb, unsigned long val,
1315 			 void *data)
1316 {
1317 	struct machine_check_event *evt = data;
1318 	struct papr_scm_priv *p;
1319 	u64 phys_addr;
1320 	bool found = false;
1321 
1322 	if (evt->error_type != MCE_ERROR_TYPE_UE)
1323 		return NOTIFY_DONE;
1324 
1325 	if (list_empty(&papr_nd_regions))
1326 		return NOTIFY_DONE;
1327 
1328 	/*
1329 	 * The physical address obtained here is PAGE_SIZE aligned, so get the
1330 	 * exact address from the effective address
1331 	 */
1332 	phys_addr = evt->u.ue_error.physical_address +
1333 			(evt->u.ue_error.effective_address & ~PAGE_MASK);
1334 
1335 	if (!evt->u.ue_error.physical_address_provided ||
1336 	    !is_zone_device_page(pfn_to_page(phys_addr >> PAGE_SHIFT)))
1337 		return NOTIFY_DONE;
1338 
1339 	/* mce notifier is called from a process context, so mutex is safe */
1340 	mutex_lock(&papr_ndr_lock);
1341 	list_for_each_entry(p, &papr_nd_regions, region_list) {
1342 		if (phys_addr >= p->res.start && phys_addr <= p->res.end) {
1343 			found = true;
1344 			break;
1345 		}
1346 	}
1347 
1348 	if (found)
1349 		papr_scm_add_badblock(p->region, p->bus, phys_addr);
1350 
1351 	mutex_unlock(&papr_ndr_lock);
1352 
1353 	return found ? NOTIFY_OK : NOTIFY_DONE;
1354 }
1355 
1356 static struct notifier_block mce_ue_nb = {
1357 	.notifier_call = handle_mce_ue
1358 };
1359 
1360 static int papr_scm_probe(struct platform_device *pdev)
1361 {
1362 	struct device_node *dn = pdev->dev.of_node;
1363 	u32 drc_index, metadata_size;
1364 	u64 blocks, block_size;
1365 	struct papr_scm_priv *p;
1366 	u8 uuid_raw[UUID_SIZE];
1367 	const char *uuid_str;
1368 	ssize_t stat_size;
1369 	uuid_t uuid;
1370 	int rc;
1371 
1372 	/* check we have all the required DT properties */
1373 	if (of_property_read_u32(dn, "ibm,my-drc-index", &drc_index)) {
1374 		dev_err(&pdev->dev, "%pOF: missing drc-index!\n", dn);
1375 		return -ENODEV;
1376 	}
1377 
1378 	if (of_property_read_u64(dn, "ibm,block-size", &block_size)) {
1379 		dev_err(&pdev->dev, "%pOF: missing block-size!\n", dn);
1380 		return -ENODEV;
1381 	}
1382 
1383 	if (of_property_read_u64(dn, "ibm,number-of-blocks", &blocks)) {
1384 		dev_err(&pdev->dev, "%pOF: missing number-of-blocks!\n", dn);
1385 		return -ENODEV;
1386 	}
1387 
1388 	if (of_property_read_string(dn, "ibm,unit-guid", &uuid_str)) {
1389 		dev_err(&pdev->dev, "%pOF: missing unit-guid!\n", dn);
1390 		return -ENODEV;
1391 	}
1392 
1393 	/*
1394 	 * open firmware platform device create won't update the NUMA
1395 	 * distance table. For PAPR SCM devices we use numa_map_to_online_node()
1396 	 * to find the nearest online NUMA node and that requires correct
1397 	 * distance table information.
1398 	 */
1399 	update_numa_distance(dn);
1400 
1401 	p = kzalloc(sizeof(*p), GFP_KERNEL);
1402 	if (!p)
1403 		return -ENOMEM;
1404 
1405 	/* Initialize the dimm mutex */
1406 	mutex_init(&p->health_mutex);
1407 
1408 	/* optional DT properties */
1409 	of_property_read_u32(dn, "ibm,metadata-size", &metadata_size);
1410 
1411 	p->dn = dn;
1412 	p->drc_index = drc_index;
1413 	p->block_size = block_size;
1414 	p->blocks = blocks;
1415 	p->is_volatile = !of_property_read_bool(dn, "ibm,cache-flush-required");
1416 	p->hcall_flush_required = of_property_read_bool(dn, "ibm,hcall-flush-required");
1417 
1418 	if (of_property_read_u64(dn, "ibm,persistence-failed-count",
1419 				 &p->dirty_shutdown_counter))
1420 		p->dirty_shutdown_counter = 0;
1421 
1422 	/* We just need to ensure that set cookies are unique across */
1423 	uuid_parse(uuid_str, &uuid);
1424 
1425 	/*
1426 	 * The cookie1 and cookie2 are not really little endian.
1427 	 * We store a raw buffer representation of the
1428 	 * uuid string so that we can compare this with the label
1429 	 * area cookie irrespective of the endian configuration
1430 	 * with which the kernel is built.
1431 	 *
1432 	 * Historically we stored the cookie in the below format.
1433 	 * for a uuid string 72511b67-0b3b-42fd-8d1d-5be3cae8bcaa
1434 	 *	cookie1 was 0xfd423b0b671b5172
1435 	 *	cookie2 was 0xaabce8cae35b1d8d
1436 	 */
1437 	export_uuid(uuid_raw, &uuid);
1438 	p->nd_set.cookie1 = get_unaligned_le64(&uuid_raw[0]);
1439 	p->nd_set.cookie2 = get_unaligned_le64(&uuid_raw[8]);
1440 
1441 	/* might be zero */
1442 	p->metadata_size = metadata_size;
1443 	p->pdev = pdev;
1444 
1445 	/* request the hypervisor to bind this region to somewhere in memory */
1446 	rc = drc_pmem_bind(p);
1447 
1448 	/* If phyp says drc memory still bound then force unbound and retry */
1449 	if (rc == H_OVERLAP)
1450 		rc = drc_pmem_query_n_bind(p);
1451 
1452 	if (rc != H_SUCCESS) {
1453 		dev_err(&p->pdev->dev, "bind err: %d\n", rc);
1454 		rc = -ENXIO;
1455 		goto err;
1456 	}
1457 
1458 	/* setup the resource for the newly bound range */
1459 	p->res.start = p->bound_addr;
1460 	p->res.end   = p->bound_addr + p->blocks * p->block_size - 1;
1461 	p->res.name  = pdev->name;
1462 	p->res.flags = IORESOURCE_MEM;
1463 
1464 	/* Try retrieving the stat buffer and see if its supported */
1465 	stat_size = drc_pmem_query_stats(p, NULL, 0);
1466 	if (stat_size > 0) {
1467 		p->stat_buffer_len = stat_size;
1468 		dev_dbg(&p->pdev->dev, "Max perf-stat size %lu-bytes\n",
1469 			p->stat_buffer_len);
1470 	}
1471 
1472 	rc = papr_scm_nvdimm_init(p);
1473 	if (rc)
1474 		goto err2;
1475 
1476 	platform_set_drvdata(pdev, p);
1477 	papr_scm_pmu_register(p);
1478 
1479 	return 0;
1480 
1481 err2:	drc_pmem_unbind(p);
1482 err:	kfree(p);
1483 	return rc;
1484 }
1485 
1486 static void papr_scm_remove(struct platform_device *pdev)
1487 {
1488 	struct papr_scm_priv *p = platform_get_drvdata(pdev);
1489 
1490 	mutex_lock(&papr_ndr_lock);
1491 	list_del(&p->region_list);
1492 	mutex_unlock(&papr_ndr_lock);
1493 
1494 	nvdimm_bus_unregister(p->bus);
1495 	drc_pmem_unbind(p);
1496 
1497 	if (pdev->archdata.priv)
1498 		unregister_nvdimm_pmu(pdev->archdata.priv);
1499 
1500 	pdev->archdata.priv = NULL;
1501 	kfree(p->bus_desc.provider_name);
1502 	kfree(p);
1503 }
1504 
1505 static const struct of_device_id papr_scm_match[] = {
1506 	{ .compatible = "ibm,pmemory" },
1507 	{ .compatible = "ibm,pmemory-v2" },
1508 	{ },
1509 };
1510 
1511 static struct platform_driver papr_scm_driver = {
1512 	.probe = papr_scm_probe,
1513 	.remove = papr_scm_remove,
1514 	.driver = {
1515 		.name = "papr_scm",
1516 		.of_match_table = papr_scm_match,
1517 	},
1518 };
1519 
1520 static int __init papr_scm_init(void)
1521 {
1522 	int ret;
1523 
1524 	ret = platform_driver_register(&papr_scm_driver);
1525 	if (!ret)
1526 		mce_register_notifier(&mce_ue_nb);
1527 
1528 	return ret;
1529 }
1530 module_init(papr_scm_init);
1531 
1532 static void __exit papr_scm_exit(void)
1533 {
1534 	mce_unregister_notifier(&mce_ue_nb);
1535 	platform_driver_unregister(&papr_scm_driver);
1536 }
1537 module_exit(papr_scm_exit);
1538 
1539 MODULE_DEVICE_TABLE(of, papr_scm_match);
1540 MODULE_DESCRIPTION("PAPR Storage Class Memory interface driver");
1541 MODULE_LICENSE("GPL");
1542 MODULE_AUTHOR("IBM Corporation");
1543