xref: /linux/arch/powerpc/platforms/pseries/papr_scm.c (revision 52990390f91c1c39ca742fc8f390b29891d95127)
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/slab.h>
10 #include <linux/ndctl.h>
11 #include <linux/sched.h>
12 #include <linux/libnvdimm.h>
13 #include <linux/platform_device.h>
14 #include <linux/delay.h>
15 #include <linux/seq_buf.h>
16 #include <linux/nd.h>
17 
18 #include <asm/plpar_wrappers.h>
19 #include <asm/papr_pdsm.h>
20 #include <asm/mce.h>
21 #include <asm/unaligned.h>
22 #include <linux/perf_event.h>
23 
24 #define BIND_ANY_ADDR (~0ul)
25 
26 #define PAPR_SCM_DIMM_CMD_MASK \
27 	((1ul << ND_CMD_GET_CONFIG_SIZE) | \
28 	 (1ul << ND_CMD_GET_CONFIG_DATA) | \
29 	 (1ul << ND_CMD_SET_CONFIG_DATA) | \
30 	 (1ul << ND_CMD_CALL))
31 
32 /* DIMM health bitmap indicators */
33 /* SCM device is unable to persist memory contents */
34 #define PAPR_PMEM_UNARMED                   (1ULL << (63 - 0))
35 /* SCM device failed to persist memory contents */
36 #define PAPR_PMEM_SHUTDOWN_DIRTY            (1ULL << (63 - 1))
37 /* SCM device contents are persisted from previous IPL */
38 #define PAPR_PMEM_SHUTDOWN_CLEAN            (1ULL << (63 - 2))
39 /* SCM device contents are not persisted from previous IPL */
40 #define PAPR_PMEM_EMPTY                     (1ULL << (63 - 3))
41 /* SCM device memory life remaining is critically low */
42 #define PAPR_PMEM_HEALTH_CRITICAL           (1ULL << (63 - 4))
43 /* SCM device will be garded off next IPL due to failure */
44 #define PAPR_PMEM_HEALTH_FATAL              (1ULL << (63 - 5))
45 /* SCM contents cannot persist due to current platform health status */
46 #define PAPR_PMEM_HEALTH_UNHEALTHY          (1ULL << (63 - 6))
47 /* SCM device is unable to persist memory contents in certain conditions */
48 #define PAPR_PMEM_HEALTH_NON_CRITICAL       (1ULL << (63 - 7))
49 /* SCM device is encrypted */
50 #define PAPR_PMEM_ENCRYPTED                 (1ULL << (63 - 8))
51 /* SCM device has been scrubbed and locked */
52 #define PAPR_PMEM_SCRUBBED_AND_LOCKED       (1ULL << (63 - 9))
53 
54 /* Bits status indicators for health bitmap indicating unarmed dimm */
55 #define PAPR_PMEM_UNARMED_MASK (PAPR_PMEM_UNARMED |		\
56 				PAPR_PMEM_HEALTH_UNHEALTHY)
57 
58 /* Bits status indicators for health bitmap indicating unflushed dimm */
59 #define PAPR_PMEM_BAD_SHUTDOWN_MASK (PAPR_PMEM_SHUTDOWN_DIRTY)
60 
61 /* Bits status indicators for health bitmap indicating unrestored dimm */
62 #define PAPR_PMEM_BAD_RESTORE_MASK  (PAPR_PMEM_EMPTY)
63 
64 /* Bit status indicators for smart event notification */
65 #define PAPR_PMEM_SMART_EVENT_MASK (PAPR_PMEM_HEALTH_CRITICAL | \
66 				    PAPR_PMEM_HEALTH_FATAL |	\
67 				    PAPR_PMEM_HEALTH_UNHEALTHY)
68 
69 #define PAPR_SCM_PERF_STATS_EYECATCHER __stringify(SCMSTATS)
70 #define PAPR_SCM_PERF_STATS_VERSION 0x1
71 
72 /* Struct holding a single performance metric */
73 struct papr_scm_perf_stat {
74 	u8 stat_id[8];
75 	__be64 stat_val;
76 } __packed;
77 
78 /* Struct exchanged between kernel and PHYP for fetching drc perf stats */
79 struct papr_scm_perf_stats {
80 	u8 eye_catcher[8];
81 	/* Should be PAPR_SCM_PERF_STATS_VERSION */
82 	__be32 stats_version;
83 	/* Number of stats following */
84 	__be32 num_statistics;
85 	/* zero or more performance matrics */
86 	struct papr_scm_perf_stat scm_statistic[];
87 } __packed;
88 
89 /* private struct associated with each region */
90 struct papr_scm_priv {
91 	struct platform_device *pdev;
92 	struct device_node *dn;
93 	uint32_t drc_index;
94 	uint64_t blocks;
95 	uint64_t block_size;
96 	int metadata_size;
97 	bool is_volatile;
98 	bool hcall_flush_required;
99 
100 	uint64_t bound_addr;
101 
102 	struct nvdimm_bus_descriptor bus_desc;
103 	struct nvdimm_bus *bus;
104 	struct nvdimm *nvdimm;
105 	struct resource res;
106 	struct nd_region *region;
107 	struct nd_interleave_set nd_set;
108 	struct list_head region_list;
109 
110 	/* Protect dimm health data from concurrent read/writes */
111 	struct mutex health_mutex;
112 
113 	/* Last time the health information of the dimm was updated */
114 	unsigned long lasthealth_jiffies;
115 
116 	/* Health information for the dimm */
117 	u64 health_bitmap;
118 
119 	/* Holds the last known dirty shutdown counter value */
120 	u64 dirty_shutdown_counter;
121 
122 	/* length of the stat buffer as expected by phyp */
123 	size_t stat_buffer_len;
124 
125 	/* The bits which needs to be overridden */
126 	u64 health_bitmap_inject_mask;
127 };
128 
129 static int papr_scm_pmem_flush(struct nd_region *nd_region,
130 			       struct bio *bio __maybe_unused)
131 {
132 	struct papr_scm_priv *p = nd_region_provider_data(nd_region);
133 	unsigned long ret_buf[PLPAR_HCALL_BUFSIZE], token = 0;
134 	long rc;
135 
136 	dev_dbg(&p->pdev->dev, "flush drc 0x%x", p->drc_index);
137 
138 	do {
139 		rc = plpar_hcall(H_SCM_FLUSH, ret_buf, p->drc_index, token);
140 		token = ret_buf[0];
141 
142 		/* Check if we are stalled for some time */
143 		if (H_IS_LONG_BUSY(rc)) {
144 			msleep(get_longbusy_msecs(rc));
145 			rc = H_BUSY;
146 		} else if (rc == H_BUSY) {
147 			cond_resched();
148 		}
149 	} while (rc == H_BUSY);
150 
151 	if (rc) {
152 		dev_err(&p->pdev->dev, "flush error: %ld", rc);
153 		rc = -EIO;
154 	} else {
155 		dev_dbg(&p->pdev->dev, "flush drc 0x%x complete", p->drc_index);
156 	}
157 
158 	return rc;
159 }
160 
161 static LIST_HEAD(papr_nd_regions);
162 static DEFINE_MUTEX(papr_ndr_lock);
163 
164 static int drc_pmem_bind(struct papr_scm_priv *p)
165 {
166 	unsigned long ret[PLPAR_HCALL_BUFSIZE];
167 	uint64_t saved = 0;
168 	uint64_t token;
169 	int64_t rc;
170 
171 	/*
172 	 * When the hypervisor cannot map all the requested memory in a single
173 	 * hcall it returns H_BUSY and we call again with the token until
174 	 * we get H_SUCCESS. Aborting the retry loop before getting H_SUCCESS
175 	 * leave the system in an undefined state, so we wait.
176 	 */
177 	token = 0;
178 
179 	do {
180 		rc = plpar_hcall(H_SCM_BIND_MEM, ret, p->drc_index, 0,
181 				p->blocks, BIND_ANY_ADDR, token);
182 		token = ret[0];
183 		if (!saved)
184 			saved = ret[1];
185 		cond_resched();
186 	} while (rc == H_BUSY);
187 
188 	if (rc)
189 		return rc;
190 
191 	p->bound_addr = saved;
192 	dev_dbg(&p->pdev->dev, "bound drc 0x%x to 0x%lx\n",
193 		p->drc_index, (unsigned long)saved);
194 	return rc;
195 }
196 
197 static void drc_pmem_unbind(struct papr_scm_priv *p)
198 {
199 	unsigned long ret[PLPAR_HCALL_BUFSIZE];
200 	uint64_t token = 0;
201 	int64_t rc;
202 
203 	dev_dbg(&p->pdev->dev, "unbind drc 0x%x\n", p->drc_index);
204 
205 	/* NB: unbind has the same retry requirements as drc_pmem_bind() */
206 	do {
207 
208 		/* Unbind of all SCM resources associated with drcIndex */
209 		rc = plpar_hcall(H_SCM_UNBIND_ALL, ret, H_UNBIND_SCOPE_DRC,
210 				 p->drc_index, token);
211 		token = ret[0];
212 
213 		/* Check if we are stalled for some time */
214 		if (H_IS_LONG_BUSY(rc)) {
215 			msleep(get_longbusy_msecs(rc));
216 			rc = H_BUSY;
217 		} else if (rc == H_BUSY) {
218 			cond_resched();
219 		}
220 
221 	} while (rc == H_BUSY);
222 
223 	if (rc)
224 		dev_err(&p->pdev->dev, "unbind error: %lld\n", rc);
225 	else
226 		dev_dbg(&p->pdev->dev, "unbind drc 0x%x complete\n",
227 			p->drc_index);
228 
229 	return;
230 }
231 
232 static int drc_pmem_query_n_bind(struct papr_scm_priv *p)
233 {
234 	unsigned long start_addr;
235 	unsigned long end_addr;
236 	unsigned long ret[PLPAR_HCALL_BUFSIZE];
237 	int64_t rc;
238 
239 
240 	rc = plpar_hcall(H_SCM_QUERY_BLOCK_MEM_BINDING, ret,
241 			 p->drc_index, 0);
242 	if (rc)
243 		goto err_out;
244 	start_addr = ret[0];
245 
246 	/* Make sure the full region is bound. */
247 	rc = plpar_hcall(H_SCM_QUERY_BLOCK_MEM_BINDING, ret,
248 			 p->drc_index, p->blocks - 1);
249 	if (rc)
250 		goto err_out;
251 	end_addr = ret[0];
252 
253 	if ((end_addr - start_addr) != ((p->blocks - 1) * p->block_size))
254 		goto err_out;
255 
256 	p->bound_addr = start_addr;
257 	dev_dbg(&p->pdev->dev, "bound drc 0x%x to 0x%lx\n", p->drc_index, start_addr);
258 	return rc;
259 
260 err_out:
261 	dev_info(&p->pdev->dev,
262 		 "Failed to query, trying an unbind followed by bind");
263 	drc_pmem_unbind(p);
264 	return drc_pmem_bind(p);
265 }
266 
267 /*
268  * Query the Dimm performance stats from PHYP and copy them (if returned) to
269  * provided struct papr_scm_perf_stats instance 'stats' that can hold atleast
270  * (num_stats + header) bytes.
271  * - If buff_stats == NULL the return value is the size in bytes of the buffer
272  * needed to hold all supported performance-statistics.
273  * - If buff_stats != NULL and num_stats == 0 then we copy all known
274  * performance-statistics to 'buff_stat' and expect to be large enough to
275  * hold them.
276  * - if buff_stats != NULL and num_stats > 0 then copy the requested
277  * performance-statistics to buff_stats.
278  */
279 static ssize_t drc_pmem_query_stats(struct papr_scm_priv *p,
280 				    struct papr_scm_perf_stats *buff_stats,
281 				    unsigned int num_stats)
282 {
283 	unsigned long ret[PLPAR_HCALL_BUFSIZE];
284 	size_t size;
285 	s64 rc;
286 
287 	/* Setup the out buffer */
288 	if (buff_stats) {
289 		memcpy(buff_stats->eye_catcher,
290 		       PAPR_SCM_PERF_STATS_EYECATCHER, 8);
291 		buff_stats->stats_version =
292 			cpu_to_be32(PAPR_SCM_PERF_STATS_VERSION);
293 		buff_stats->num_statistics =
294 			cpu_to_be32(num_stats);
295 
296 		/*
297 		 * Calculate the buffer size based on num-stats provided
298 		 * or use the prefetched max buffer length
299 		 */
300 		if (num_stats)
301 			/* Calculate size from the num_stats */
302 			size = sizeof(struct papr_scm_perf_stats) +
303 				num_stats * sizeof(struct papr_scm_perf_stat);
304 		else
305 			size = p->stat_buffer_len;
306 	} else {
307 		/* In case of no out buffer ignore the size */
308 		size = 0;
309 	}
310 
311 	/* Do the HCALL asking PHYP for info */
312 	rc = plpar_hcall(H_SCM_PERFORMANCE_STATS, ret, p->drc_index,
313 			 buff_stats ? virt_to_phys(buff_stats) : 0,
314 			 size);
315 
316 	/* Check if the error was due to an unknown stat-id */
317 	if (rc == H_PARTIAL) {
318 		dev_err(&p->pdev->dev,
319 			"Unknown performance stats, Err:0x%016lX\n", ret[0]);
320 		return -ENOENT;
321 	} else if (rc == H_AUTHORITY) {
322 		dev_info(&p->pdev->dev,
323 			 "Permission denied while accessing performance stats");
324 		return -EPERM;
325 	} else if (rc == H_UNSUPPORTED) {
326 		dev_dbg(&p->pdev->dev, "Performance stats unsupported\n");
327 		return -EOPNOTSUPP;
328 	} else if (rc != H_SUCCESS) {
329 		dev_err(&p->pdev->dev,
330 			"Failed to query performance stats, Err:%lld\n", rc);
331 		return -EIO;
332 
333 	} else if (!size) {
334 		/* Handle case where stat buffer size was requested */
335 		dev_dbg(&p->pdev->dev,
336 			"Performance stats size %ld\n", ret[0]);
337 		return ret[0];
338 	}
339 
340 	/* Successfully fetched the requested stats from phyp */
341 	dev_dbg(&p->pdev->dev,
342 		"Performance stats returned %d stats\n",
343 		be32_to_cpu(buff_stats->num_statistics));
344 	return 0;
345 }
346 
347 #ifdef CONFIG_PERF_EVENTS
348 #define to_nvdimm_pmu(_pmu)	container_of(_pmu, struct nvdimm_pmu, pmu)
349 
350 static const char * const nvdimm_events_map[] = {
351 	[1] = "CtlResCt",
352 	[2] = "CtlResTm",
353 	[3] = "PonSecs ",
354 	[4] = "MemLife ",
355 	[5] = "CritRscU",
356 	[6] = "HostLCnt",
357 	[7] = "HostSCnt",
358 	[8] = "HostSDur",
359 	[9] = "HostLDur",
360 	[10] = "MedRCnt ",
361 	[11] = "MedWCnt ",
362 	[12] = "MedRDur ",
363 	[13] = "MedWDur ",
364 	[14] = "CchRHCnt",
365 	[15] = "CchWHCnt",
366 	[16] = "FastWCnt",
367 };
368 
369 static int papr_scm_pmu_get_value(struct perf_event *event, struct device *dev, u64 *count)
370 {
371 	struct papr_scm_perf_stat *stat;
372 	struct papr_scm_perf_stats *stats;
373 	struct papr_scm_priv *p = dev_get_drvdata(dev);
374 	int rc, size;
375 
376 	/* Invalid eventcode */
377 	if (event->attr.config == 0 || event->attr.config >= ARRAY_SIZE(nvdimm_events_map))
378 		return -EINVAL;
379 
380 	/* Allocate request buffer enough to hold single performance stat */
381 	size = sizeof(struct papr_scm_perf_stats) +
382 		sizeof(struct papr_scm_perf_stat);
383 
384 	if (!p)
385 		return -EINVAL;
386 
387 	stats = kzalloc(size, GFP_KERNEL);
388 	if (!stats)
389 		return -ENOMEM;
390 
391 	stat = &stats->scm_statistic[0];
392 	memcpy(&stat->stat_id,
393 	       nvdimm_events_map[event->attr.config],
394 		sizeof(stat->stat_id));
395 	stat->stat_val = 0;
396 
397 	rc = drc_pmem_query_stats(p, stats, 1);
398 	if (rc < 0) {
399 		kfree(stats);
400 		return rc;
401 	}
402 
403 	*count = be64_to_cpu(stat->stat_val);
404 	kfree(stats);
405 	return 0;
406 }
407 
408 static int papr_scm_pmu_event_init(struct perf_event *event)
409 {
410 	struct nvdimm_pmu *nd_pmu = to_nvdimm_pmu(event->pmu);
411 	struct papr_scm_priv *p;
412 
413 	if (!nd_pmu)
414 		return -EINVAL;
415 
416 	/* test the event attr type for PMU enumeration */
417 	if (event->attr.type != event->pmu->type)
418 		return -ENOENT;
419 
420 	/* it does not support event sampling mode */
421 	if (is_sampling_event(event))
422 		return -EOPNOTSUPP;
423 
424 	/* no branch sampling */
425 	if (has_branch_stack(event))
426 		return -EOPNOTSUPP;
427 
428 	p = (struct papr_scm_priv *)nd_pmu->dev->driver_data;
429 	if (!p)
430 		return -EINVAL;
431 
432 	/* Invalid eventcode */
433 	if (event->attr.config == 0 || event->attr.config > 16)
434 		return -EINVAL;
435 
436 	return 0;
437 }
438 
439 static int papr_scm_pmu_add(struct perf_event *event, int flags)
440 {
441 	u64 count;
442 	int rc;
443 	struct nvdimm_pmu *nd_pmu = to_nvdimm_pmu(event->pmu);
444 
445 	if (!nd_pmu)
446 		return -EINVAL;
447 
448 	if (flags & PERF_EF_START) {
449 		rc = papr_scm_pmu_get_value(event, nd_pmu->dev, &count);
450 		if (rc)
451 			return rc;
452 
453 		local64_set(&event->hw.prev_count, count);
454 	}
455 
456 	return 0;
457 }
458 
459 static void papr_scm_pmu_read(struct perf_event *event)
460 {
461 	u64 prev, now;
462 	int rc;
463 	struct nvdimm_pmu *nd_pmu = to_nvdimm_pmu(event->pmu);
464 
465 	if (!nd_pmu)
466 		return;
467 
468 	rc = papr_scm_pmu_get_value(event, nd_pmu->dev, &now);
469 	if (rc)
470 		return;
471 
472 	prev = local64_xchg(&event->hw.prev_count, now);
473 	local64_add(now - prev, &event->count);
474 }
475 
476 static void papr_scm_pmu_del(struct perf_event *event, int flags)
477 {
478 	papr_scm_pmu_read(event);
479 }
480 
481 static void papr_scm_pmu_register(struct papr_scm_priv *p)
482 {
483 	struct nvdimm_pmu *nd_pmu;
484 	int rc, nodeid;
485 
486 	nd_pmu = kzalloc(sizeof(*nd_pmu), GFP_KERNEL);
487 	if (!nd_pmu) {
488 		rc = -ENOMEM;
489 		goto pmu_err_print;
490 	}
491 
492 	if (!p->stat_buffer_len) {
493 		rc = -ENOENT;
494 		goto pmu_check_events_err;
495 	}
496 
497 	nd_pmu->pmu.task_ctx_nr = perf_invalid_context;
498 	nd_pmu->pmu.name = nvdimm_name(p->nvdimm);
499 	nd_pmu->pmu.event_init = papr_scm_pmu_event_init;
500 	nd_pmu->pmu.read = papr_scm_pmu_read;
501 	nd_pmu->pmu.add = papr_scm_pmu_add;
502 	nd_pmu->pmu.del = papr_scm_pmu_del;
503 
504 	nd_pmu->pmu.capabilities = PERF_PMU_CAP_NO_INTERRUPT |
505 				PERF_PMU_CAP_NO_EXCLUDE;
506 
507 	/*updating the cpumask variable */
508 	nodeid = numa_map_to_online_node(dev_to_node(&p->pdev->dev));
509 	nd_pmu->arch_cpumask = *cpumask_of_node(nodeid);
510 
511 	rc = register_nvdimm_pmu(nd_pmu, p->pdev);
512 	if (rc)
513 		goto pmu_check_events_err;
514 
515 	/*
516 	 * Set archdata.priv value to nvdimm_pmu structure, to handle the
517 	 * unregistering of pmu device.
518 	 */
519 	p->pdev->archdata.priv = nd_pmu;
520 	return;
521 
522 pmu_check_events_err:
523 	kfree(nd_pmu);
524 pmu_err_print:
525 	dev_info(&p->pdev->dev, "nvdimm pmu didn't register rc=%d\n", rc);
526 }
527 
528 #else
529 static void papr_scm_pmu_register(struct papr_scm_priv *p) { }
530 #endif
531 
532 /*
533  * Issue hcall to retrieve dimm health info and populate papr_scm_priv with the
534  * health information.
535  */
536 static int __drc_pmem_query_health(struct papr_scm_priv *p)
537 {
538 	unsigned long ret[PLPAR_HCALL_BUFSIZE];
539 	u64 bitmap = 0;
540 	long rc;
541 
542 	/* issue the hcall */
543 	rc = plpar_hcall(H_SCM_HEALTH, ret, p->drc_index);
544 	if (rc == H_SUCCESS)
545 		bitmap = ret[0] & ret[1];
546 	else if (rc == H_FUNCTION)
547 		dev_info_once(&p->pdev->dev,
548 			      "Hcall H_SCM_HEALTH not implemented, assuming empty health bitmap");
549 	else {
550 
551 		dev_err(&p->pdev->dev,
552 			"Failed to query health information, Err:%ld\n", rc);
553 		return -ENXIO;
554 	}
555 
556 	p->lasthealth_jiffies = jiffies;
557 	/* Allow injecting specific health bits via inject mask. */
558 	if (p->health_bitmap_inject_mask)
559 		bitmap = (bitmap & ~p->health_bitmap_inject_mask) |
560 			p->health_bitmap_inject_mask;
561 	WRITE_ONCE(p->health_bitmap, bitmap);
562 	dev_dbg(&p->pdev->dev,
563 		"Queried dimm health info. Bitmap:0x%016lx Mask:0x%016lx\n",
564 		ret[0], ret[1]);
565 
566 	return 0;
567 }
568 
569 /* Min interval in seconds for assuming stable dimm health */
570 #define MIN_HEALTH_QUERY_INTERVAL 60
571 
572 /* Query cached health info and if needed call drc_pmem_query_health */
573 static int drc_pmem_query_health(struct papr_scm_priv *p)
574 {
575 	unsigned long cache_timeout;
576 	int rc;
577 
578 	/* Protect concurrent modifications to papr_scm_priv */
579 	rc = mutex_lock_interruptible(&p->health_mutex);
580 	if (rc)
581 		return rc;
582 
583 	/* Jiffies offset for which the health data is assumed to be same */
584 	cache_timeout = p->lasthealth_jiffies +
585 		msecs_to_jiffies(MIN_HEALTH_QUERY_INTERVAL * 1000);
586 
587 	/* Fetch new health info is its older than MIN_HEALTH_QUERY_INTERVAL */
588 	if (time_after(jiffies, cache_timeout))
589 		rc = __drc_pmem_query_health(p);
590 	else
591 		/* Assume cached health data is valid */
592 		rc = 0;
593 
594 	mutex_unlock(&p->health_mutex);
595 	return rc;
596 }
597 
598 static int papr_scm_meta_get(struct papr_scm_priv *p,
599 			     struct nd_cmd_get_config_data_hdr *hdr)
600 {
601 	unsigned long data[PLPAR_HCALL_BUFSIZE];
602 	unsigned long offset, data_offset;
603 	int len, read;
604 	int64_t ret;
605 
606 	if ((hdr->in_offset + hdr->in_length) > p->metadata_size)
607 		return -EINVAL;
608 
609 	for (len = hdr->in_length; len; len -= read) {
610 
611 		data_offset = hdr->in_length - len;
612 		offset = hdr->in_offset + data_offset;
613 
614 		if (len >= 8)
615 			read = 8;
616 		else if (len >= 4)
617 			read = 4;
618 		else if (len >= 2)
619 			read = 2;
620 		else
621 			read = 1;
622 
623 		ret = plpar_hcall(H_SCM_READ_METADATA, data, p->drc_index,
624 				  offset, read);
625 
626 		if (ret == H_PARAMETER) /* bad DRC index */
627 			return -ENODEV;
628 		if (ret)
629 			return -EINVAL; /* other invalid parameter */
630 
631 		switch (read) {
632 		case 8:
633 			*(uint64_t *)(hdr->out_buf + data_offset) = be64_to_cpu(data[0]);
634 			break;
635 		case 4:
636 			*(uint32_t *)(hdr->out_buf + data_offset) = be32_to_cpu(data[0] & 0xffffffff);
637 			break;
638 
639 		case 2:
640 			*(uint16_t *)(hdr->out_buf + data_offset) = be16_to_cpu(data[0] & 0xffff);
641 			break;
642 
643 		case 1:
644 			*(uint8_t *)(hdr->out_buf + data_offset) = (data[0] & 0xff);
645 			break;
646 		}
647 	}
648 	return 0;
649 }
650 
651 static int papr_scm_meta_set(struct papr_scm_priv *p,
652 			     struct nd_cmd_set_config_hdr *hdr)
653 {
654 	unsigned long offset, data_offset;
655 	int len, wrote;
656 	unsigned long data;
657 	__be64 data_be;
658 	int64_t ret;
659 
660 	if ((hdr->in_offset + hdr->in_length) > p->metadata_size)
661 		return -EINVAL;
662 
663 	for (len = hdr->in_length; len; len -= wrote) {
664 
665 		data_offset = hdr->in_length - len;
666 		offset = hdr->in_offset + data_offset;
667 
668 		if (len >= 8) {
669 			data = *(uint64_t *)(hdr->in_buf + data_offset);
670 			data_be = cpu_to_be64(data);
671 			wrote = 8;
672 		} else if (len >= 4) {
673 			data = *(uint32_t *)(hdr->in_buf + data_offset);
674 			data &= 0xffffffff;
675 			data_be = cpu_to_be32(data);
676 			wrote = 4;
677 		} else if (len >= 2) {
678 			data = *(uint16_t *)(hdr->in_buf + data_offset);
679 			data &= 0xffff;
680 			data_be = cpu_to_be16(data);
681 			wrote = 2;
682 		} else {
683 			data_be = *(uint8_t *)(hdr->in_buf + data_offset);
684 			data_be &= 0xff;
685 			wrote = 1;
686 		}
687 
688 		ret = plpar_hcall_norets(H_SCM_WRITE_METADATA, p->drc_index,
689 					 offset, data_be, wrote);
690 		if (ret == H_PARAMETER) /* bad DRC index */
691 			return -ENODEV;
692 		if (ret)
693 			return -EINVAL; /* other invalid parameter */
694 	}
695 
696 	return 0;
697 }
698 
699 /*
700  * Do a sanity checks on the inputs args to dimm-control function and return
701  * '0' if valid. Validation of PDSM payloads happens later in
702  * papr_scm_service_pdsm.
703  */
704 static int is_cmd_valid(struct nvdimm *nvdimm, unsigned int cmd, void *buf,
705 			unsigned int buf_len)
706 {
707 	unsigned long cmd_mask = PAPR_SCM_DIMM_CMD_MASK;
708 	struct nd_cmd_pkg *nd_cmd;
709 	struct papr_scm_priv *p;
710 	enum papr_pdsm pdsm;
711 
712 	/* Only dimm-specific calls are supported atm */
713 	if (!nvdimm)
714 		return -EINVAL;
715 
716 	/* get the provider data from struct nvdimm */
717 	p = nvdimm_provider_data(nvdimm);
718 
719 	if (!test_bit(cmd, &cmd_mask)) {
720 		dev_dbg(&p->pdev->dev, "Unsupported cmd=%u\n", cmd);
721 		return -EINVAL;
722 	}
723 
724 	/* For CMD_CALL verify pdsm request */
725 	if (cmd == ND_CMD_CALL) {
726 		/* Verify the envelope and envelop size */
727 		if (!buf ||
728 		    buf_len < (sizeof(struct nd_cmd_pkg) + ND_PDSM_HDR_SIZE)) {
729 			dev_dbg(&p->pdev->dev, "Invalid pkg size=%u\n",
730 				buf_len);
731 			return -EINVAL;
732 		}
733 
734 		/* Verify that the nd_cmd_pkg.nd_family is correct */
735 		nd_cmd = (struct nd_cmd_pkg *)buf;
736 
737 		if (nd_cmd->nd_family != NVDIMM_FAMILY_PAPR) {
738 			dev_dbg(&p->pdev->dev, "Invalid pkg family=0x%llx\n",
739 				nd_cmd->nd_family);
740 			return -EINVAL;
741 		}
742 
743 		pdsm = (enum papr_pdsm)nd_cmd->nd_command;
744 
745 		/* Verify if the pdsm command is valid */
746 		if (pdsm <= PAPR_PDSM_MIN || pdsm >= PAPR_PDSM_MAX) {
747 			dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Invalid PDSM\n",
748 				pdsm);
749 			return -EINVAL;
750 		}
751 
752 		/* Have enough space to hold returned 'nd_pkg_pdsm' header */
753 		if (nd_cmd->nd_size_out < ND_PDSM_HDR_SIZE) {
754 			dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Invalid payload\n",
755 				pdsm);
756 			return -EINVAL;
757 		}
758 	}
759 
760 	/* Let the command be further processed */
761 	return 0;
762 }
763 
764 static int papr_pdsm_fuel_gauge(struct papr_scm_priv *p,
765 				union nd_pdsm_payload *payload)
766 {
767 	int rc, size;
768 	u64 statval;
769 	struct papr_scm_perf_stat *stat;
770 	struct papr_scm_perf_stats *stats;
771 
772 	/* Silently fail if fetching performance metrics isn't  supported */
773 	if (!p->stat_buffer_len)
774 		return 0;
775 
776 	/* Allocate request buffer enough to hold single performance stat */
777 	size = sizeof(struct papr_scm_perf_stats) +
778 		sizeof(struct papr_scm_perf_stat);
779 
780 	stats = kzalloc(size, GFP_KERNEL);
781 	if (!stats)
782 		return -ENOMEM;
783 
784 	stat = &stats->scm_statistic[0];
785 	memcpy(&stat->stat_id, "MemLife ", sizeof(stat->stat_id));
786 	stat->stat_val = 0;
787 
788 	/* Fetch the fuel gauge and populate it in payload */
789 	rc = drc_pmem_query_stats(p, stats, 1);
790 	if (rc < 0) {
791 		dev_dbg(&p->pdev->dev, "Err(%d) fetching fuel gauge\n", rc);
792 		goto free_stats;
793 	}
794 
795 	statval = be64_to_cpu(stat->stat_val);
796 	dev_dbg(&p->pdev->dev,
797 		"Fetched fuel-gauge %llu", statval);
798 	payload->health.extension_flags |=
799 		PDSM_DIMM_HEALTH_RUN_GAUGE_VALID;
800 	payload->health.dimm_fuel_gauge = statval;
801 
802 	rc = sizeof(struct nd_papr_pdsm_health);
803 
804 free_stats:
805 	kfree(stats);
806 	return rc;
807 }
808 
809 /* Add the dirty-shutdown-counter value to the pdsm */
810 static int papr_pdsm_dsc(struct papr_scm_priv *p,
811 			 union nd_pdsm_payload *payload)
812 {
813 	payload->health.extension_flags |= PDSM_DIMM_DSC_VALID;
814 	payload->health.dimm_dsc = p->dirty_shutdown_counter;
815 
816 	return sizeof(struct nd_papr_pdsm_health);
817 }
818 
819 /* Fetch the DIMM health info and populate it in provided package. */
820 static int papr_pdsm_health(struct papr_scm_priv *p,
821 			    union nd_pdsm_payload *payload)
822 {
823 	int rc;
824 
825 	/* Ensure dimm health mutex is taken preventing concurrent access */
826 	rc = mutex_lock_interruptible(&p->health_mutex);
827 	if (rc)
828 		goto out;
829 
830 	/* Always fetch upto date dimm health data ignoring cached values */
831 	rc = __drc_pmem_query_health(p);
832 	if (rc) {
833 		mutex_unlock(&p->health_mutex);
834 		goto out;
835 	}
836 
837 	/* update health struct with various flags derived from health bitmap */
838 	payload->health = (struct nd_papr_pdsm_health) {
839 		.extension_flags = 0,
840 		.dimm_unarmed = !!(p->health_bitmap & PAPR_PMEM_UNARMED_MASK),
841 		.dimm_bad_shutdown = !!(p->health_bitmap & PAPR_PMEM_BAD_SHUTDOWN_MASK),
842 		.dimm_bad_restore = !!(p->health_bitmap & PAPR_PMEM_BAD_RESTORE_MASK),
843 		.dimm_scrubbed = !!(p->health_bitmap & PAPR_PMEM_SCRUBBED_AND_LOCKED),
844 		.dimm_locked = !!(p->health_bitmap & PAPR_PMEM_SCRUBBED_AND_LOCKED),
845 		.dimm_encrypted = !!(p->health_bitmap & PAPR_PMEM_ENCRYPTED),
846 		.dimm_health = PAPR_PDSM_DIMM_HEALTHY,
847 	};
848 
849 	/* Update field dimm_health based on health_bitmap flags */
850 	if (p->health_bitmap & PAPR_PMEM_HEALTH_FATAL)
851 		payload->health.dimm_health = PAPR_PDSM_DIMM_FATAL;
852 	else if (p->health_bitmap & PAPR_PMEM_HEALTH_CRITICAL)
853 		payload->health.dimm_health = PAPR_PDSM_DIMM_CRITICAL;
854 	else if (p->health_bitmap & PAPR_PMEM_HEALTH_UNHEALTHY)
855 		payload->health.dimm_health = PAPR_PDSM_DIMM_UNHEALTHY;
856 
857 	/* struct populated hence can release the mutex now */
858 	mutex_unlock(&p->health_mutex);
859 
860 	/* Populate the fuel gauge meter in the payload */
861 	papr_pdsm_fuel_gauge(p, payload);
862 	/* Populate the dirty-shutdown-counter field */
863 	papr_pdsm_dsc(p, payload);
864 
865 	rc = sizeof(struct nd_papr_pdsm_health);
866 
867 out:
868 	return rc;
869 }
870 
871 /* Inject a smart error Add the dirty-shutdown-counter value to the pdsm */
872 static int papr_pdsm_smart_inject(struct papr_scm_priv *p,
873 				  union nd_pdsm_payload *payload)
874 {
875 	int rc;
876 	u32 supported_flags = 0;
877 	u64 inject_mask = 0, clear_mask = 0;
878 	u64 mask;
879 
880 	/* Check for individual smart error flags and update inject/clear masks */
881 	if (payload->smart_inject.flags & PDSM_SMART_INJECT_HEALTH_FATAL) {
882 		supported_flags |= PDSM_SMART_INJECT_HEALTH_FATAL;
883 		if (payload->smart_inject.fatal_enable)
884 			inject_mask |= PAPR_PMEM_HEALTH_FATAL;
885 		else
886 			clear_mask |= PAPR_PMEM_HEALTH_FATAL;
887 	}
888 
889 	if (payload->smart_inject.flags & PDSM_SMART_INJECT_BAD_SHUTDOWN) {
890 		supported_flags |= PDSM_SMART_INJECT_BAD_SHUTDOWN;
891 		if (payload->smart_inject.unsafe_shutdown_enable)
892 			inject_mask |= PAPR_PMEM_SHUTDOWN_DIRTY;
893 		else
894 			clear_mask |= PAPR_PMEM_SHUTDOWN_DIRTY;
895 	}
896 
897 	dev_dbg(&p->pdev->dev, "[Smart-inject] inject_mask=%#llx clear_mask=%#llx\n",
898 		inject_mask, clear_mask);
899 
900 	/* Prevent concurrent access to dimm health bitmap related members */
901 	rc = mutex_lock_interruptible(&p->health_mutex);
902 	if (rc)
903 		return rc;
904 
905 	/* Use inject/clear masks to set health_bitmap_inject_mask */
906 	mask = READ_ONCE(p->health_bitmap_inject_mask);
907 	mask = (mask & ~clear_mask) | inject_mask;
908 	WRITE_ONCE(p->health_bitmap_inject_mask, mask);
909 
910 	/* Invalidate cached health bitmap */
911 	p->lasthealth_jiffies = 0;
912 
913 	mutex_unlock(&p->health_mutex);
914 
915 	/* Return the supported flags back to userspace */
916 	payload->smart_inject.flags = supported_flags;
917 
918 	return sizeof(struct nd_papr_pdsm_health);
919 }
920 
921 /*
922  * 'struct pdsm_cmd_desc'
923  * Identifies supported PDSMs' expected length of in/out payloads
924  * and pdsm service function.
925  *
926  * size_in	: Size of input payload if any in the PDSM request.
927  * size_out	: Size of output payload if any in the PDSM request.
928  * service	: Service function for the PDSM request. Return semantics:
929  *		  rc < 0 : Error servicing PDSM and rc indicates the error.
930  *		  rc >=0 : Serviced successfully and 'rc' indicate number of
931  *			bytes written to payload.
932  */
933 struct pdsm_cmd_desc {
934 	u32 size_in;
935 	u32 size_out;
936 	int (*service)(struct papr_scm_priv *dimm,
937 		       union nd_pdsm_payload *payload);
938 };
939 
940 /* Holds all supported PDSMs' command descriptors */
941 static const struct pdsm_cmd_desc __pdsm_cmd_descriptors[] = {
942 	[PAPR_PDSM_MIN] = {
943 		.size_in = 0,
944 		.size_out = 0,
945 		.service = NULL,
946 	},
947 	/* New PDSM command descriptors to be added below */
948 
949 	[PAPR_PDSM_HEALTH] = {
950 		.size_in = 0,
951 		.size_out = sizeof(struct nd_papr_pdsm_health),
952 		.service = papr_pdsm_health,
953 	},
954 
955 	[PAPR_PDSM_SMART_INJECT] = {
956 		.size_in = sizeof(struct nd_papr_pdsm_smart_inject),
957 		.size_out = sizeof(struct nd_papr_pdsm_smart_inject),
958 		.service = papr_pdsm_smart_inject,
959 	},
960 	/* Empty */
961 	[PAPR_PDSM_MAX] = {
962 		.size_in = 0,
963 		.size_out = 0,
964 		.service = NULL,
965 	},
966 };
967 
968 /* Given a valid pdsm cmd return its command descriptor else return NULL */
969 static inline const struct pdsm_cmd_desc *pdsm_cmd_desc(enum papr_pdsm cmd)
970 {
971 	if (cmd >= 0 || cmd < ARRAY_SIZE(__pdsm_cmd_descriptors))
972 		return &__pdsm_cmd_descriptors[cmd];
973 
974 	return NULL;
975 }
976 
977 /*
978  * For a given pdsm request call an appropriate service function.
979  * Returns errors if any while handling the pdsm command package.
980  */
981 static int papr_scm_service_pdsm(struct papr_scm_priv *p,
982 				 struct nd_cmd_pkg *pkg)
983 {
984 	/* Get the PDSM header and PDSM command */
985 	struct nd_pkg_pdsm *pdsm_pkg = (struct nd_pkg_pdsm *)pkg->nd_payload;
986 	enum papr_pdsm pdsm = (enum papr_pdsm)pkg->nd_command;
987 	const struct pdsm_cmd_desc *pdsc;
988 	int rc;
989 
990 	/* Fetch corresponding pdsm descriptor for validation and servicing */
991 	pdsc = pdsm_cmd_desc(pdsm);
992 
993 	/* Validate pdsm descriptor */
994 	/* Ensure that reserved fields are 0 */
995 	if (pdsm_pkg->reserved[0] || pdsm_pkg->reserved[1]) {
996 		dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Invalid reserved field\n",
997 			pdsm);
998 		return -EINVAL;
999 	}
1000 
1001 	/* If pdsm expects some input, then ensure that the size_in matches */
1002 	if (pdsc->size_in &&
1003 	    pkg->nd_size_in != (pdsc->size_in + ND_PDSM_HDR_SIZE)) {
1004 		dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Mismatched size_in=%d\n",
1005 			pdsm, pkg->nd_size_in);
1006 		return -EINVAL;
1007 	}
1008 
1009 	/* If pdsm wants to return data, then ensure that  size_out matches */
1010 	if (pdsc->size_out &&
1011 	    pkg->nd_size_out != (pdsc->size_out + ND_PDSM_HDR_SIZE)) {
1012 		dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Mismatched size_out=%d\n",
1013 			pdsm, pkg->nd_size_out);
1014 		return -EINVAL;
1015 	}
1016 
1017 	/* Service the pdsm */
1018 	if (pdsc->service) {
1019 		dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Servicing..\n", pdsm);
1020 
1021 		rc = pdsc->service(p, &pdsm_pkg->payload);
1022 
1023 		if (rc < 0) {
1024 			/* error encountered while servicing pdsm */
1025 			pdsm_pkg->cmd_status = rc;
1026 			pkg->nd_fw_size = ND_PDSM_HDR_SIZE;
1027 		} else {
1028 			/* pdsm serviced and 'rc' bytes written to payload */
1029 			pdsm_pkg->cmd_status = 0;
1030 			pkg->nd_fw_size = ND_PDSM_HDR_SIZE + rc;
1031 		}
1032 	} else {
1033 		dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Unsupported PDSM request\n",
1034 			pdsm);
1035 		pdsm_pkg->cmd_status = -ENOENT;
1036 		pkg->nd_fw_size = ND_PDSM_HDR_SIZE;
1037 	}
1038 
1039 	return pdsm_pkg->cmd_status;
1040 }
1041 
1042 static int papr_scm_ndctl(struct nvdimm_bus_descriptor *nd_desc,
1043 			  struct nvdimm *nvdimm, unsigned int cmd, void *buf,
1044 			  unsigned int buf_len, int *cmd_rc)
1045 {
1046 	struct nd_cmd_get_config_size *get_size_hdr;
1047 	struct nd_cmd_pkg *call_pkg = NULL;
1048 	struct papr_scm_priv *p;
1049 	int rc;
1050 
1051 	rc = is_cmd_valid(nvdimm, cmd, buf, buf_len);
1052 	if (rc) {
1053 		pr_debug("Invalid cmd=0x%x. Err=%d\n", cmd, rc);
1054 		return rc;
1055 	}
1056 
1057 	/* Use a local variable in case cmd_rc pointer is NULL */
1058 	if (!cmd_rc)
1059 		cmd_rc = &rc;
1060 
1061 	p = nvdimm_provider_data(nvdimm);
1062 
1063 	switch (cmd) {
1064 	case ND_CMD_GET_CONFIG_SIZE:
1065 		get_size_hdr = buf;
1066 
1067 		get_size_hdr->status = 0;
1068 		get_size_hdr->max_xfer = 8;
1069 		get_size_hdr->config_size = p->metadata_size;
1070 		*cmd_rc = 0;
1071 		break;
1072 
1073 	case ND_CMD_GET_CONFIG_DATA:
1074 		*cmd_rc = papr_scm_meta_get(p, buf);
1075 		break;
1076 
1077 	case ND_CMD_SET_CONFIG_DATA:
1078 		*cmd_rc = papr_scm_meta_set(p, buf);
1079 		break;
1080 
1081 	case ND_CMD_CALL:
1082 		call_pkg = (struct nd_cmd_pkg *)buf;
1083 		*cmd_rc = papr_scm_service_pdsm(p, call_pkg);
1084 		break;
1085 
1086 	default:
1087 		dev_dbg(&p->pdev->dev, "Unknown command = %d\n", cmd);
1088 		return -EINVAL;
1089 	}
1090 
1091 	dev_dbg(&p->pdev->dev, "returned with cmd_rc = %d\n", *cmd_rc);
1092 
1093 	return 0;
1094 }
1095 
1096 static ssize_t health_bitmap_inject_show(struct device *dev,
1097 					 struct device_attribute *attr,
1098 					 char *buf)
1099 {
1100 	struct nvdimm *dimm = to_nvdimm(dev);
1101 	struct papr_scm_priv *p = nvdimm_provider_data(dimm);
1102 
1103 	return sprintf(buf, "%#llx\n",
1104 		       READ_ONCE(p->health_bitmap_inject_mask));
1105 }
1106 
1107 static DEVICE_ATTR_ADMIN_RO(health_bitmap_inject);
1108 
1109 static ssize_t perf_stats_show(struct device *dev,
1110 			       struct device_attribute *attr, char *buf)
1111 {
1112 	int index;
1113 	ssize_t rc;
1114 	struct seq_buf s;
1115 	struct papr_scm_perf_stat *stat;
1116 	struct papr_scm_perf_stats *stats;
1117 	struct nvdimm *dimm = to_nvdimm(dev);
1118 	struct papr_scm_priv *p = nvdimm_provider_data(dimm);
1119 
1120 	if (!p->stat_buffer_len)
1121 		return -ENOENT;
1122 
1123 	/* Allocate the buffer for phyp where stats are written */
1124 	stats = kzalloc(p->stat_buffer_len, GFP_KERNEL);
1125 	if (!stats)
1126 		return -ENOMEM;
1127 
1128 	/* Ask phyp to return all dimm perf stats */
1129 	rc = drc_pmem_query_stats(p, stats, 0);
1130 	if (rc)
1131 		goto free_stats;
1132 	/*
1133 	 * Go through the returned output buffer and print stats and
1134 	 * values. Since stat_id is essentially a char string of
1135 	 * 8 bytes, simply use the string format specifier to print it.
1136 	 */
1137 	seq_buf_init(&s, buf, PAGE_SIZE);
1138 	for (index = 0, stat = stats->scm_statistic;
1139 	     index < be32_to_cpu(stats->num_statistics);
1140 	     ++index, ++stat) {
1141 		seq_buf_printf(&s, "%.8s = 0x%016llX\n",
1142 			       stat->stat_id,
1143 			       be64_to_cpu(stat->stat_val));
1144 	}
1145 
1146 free_stats:
1147 	kfree(stats);
1148 	return rc ? rc : (ssize_t)seq_buf_used(&s);
1149 }
1150 static DEVICE_ATTR_ADMIN_RO(perf_stats);
1151 
1152 static ssize_t flags_show(struct device *dev,
1153 			  struct device_attribute *attr, char *buf)
1154 {
1155 	struct nvdimm *dimm = to_nvdimm(dev);
1156 	struct papr_scm_priv *p = nvdimm_provider_data(dimm);
1157 	struct seq_buf s;
1158 	u64 health;
1159 	int rc;
1160 
1161 	rc = drc_pmem_query_health(p);
1162 	if (rc)
1163 		return rc;
1164 
1165 	/* Copy health_bitmap locally, check masks & update out buffer */
1166 	health = READ_ONCE(p->health_bitmap);
1167 
1168 	seq_buf_init(&s, buf, PAGE_SIZE);
1169 	if (health & PAPR_PMEM_UNARMED_MASK)
1170 		seq_buf_printf(&s, "not_armed ");
1171 
1172 	if (health & PAPR_PMEM_BAD_SHUTDOWN_MASK)
1173 		seq_buf_printf(&s, "flush_fail ");
1174 
1175 	if (health & PAPR_PMEM_BAD_RESTORE_MASK)
1176 		seq_buf_printf(&s, "restore_fail ");
1177 
1178 	if (health & PAPR_PMEM_ENCRYPTED)
1179 		seq_buf_printf(&s, "encrypted ");
1180 
1181 	if (health & PAPR_PMEM_SMART_EVENT_MASK)
1182 		seq_buf_printf(&s, "smart_notify ");
1183 
1184 	if (health & PAPR_PMEM_SCRUBBED_AND_LOCKED)
1185 		seq_buf_printf(&s, "scrubbed locked ");
1186 
1187 	if (seq_buf_used(&s))
1188 		seq_buf_printf(&s, "\n");
1189 
1190 	return seq_buf_used(&s);
1191 }
1192 DEVICE_ATTR_RO(flags);
1193 
1194 static ssize_t dirty_shutdown_show(struct device *dev,
1195 			  struct device_attribute *attr, char *buf)
1196 {
1197 	struct nvdimm *dimm = to_nvdimm(dev);
1198 	struct papr_scm_priv *p = nvdimm_provider_data(dimm);
1199 
1200 	return sysfs_emit(buf, "%llu\n", p->dirty_shutdown_counter);
1201 }
1202 DEVICE_ATTR_RO(dirty_shutdown);
1203 
1204 static umode_t papr_nd_attribute_visible(struct kobject *kobj,
1205 					 struct attribute *attr, int n)
1206 {
1207 	struct device *dev = kobj_to_dev(kobj);
1208 	struct nvdimm *nvdimm = to_nvdimm(dev);
1209 	struct papr_scm_priv *p = nvdimm_provider_data(nvdimm);
1210 
1211 	/* For if perf-stats not available remove perf_stats sysfs */
1212 	if (attr == &dev_attr_perf_stats.attr && p->stat_buffer_len == 0)
1213 		return 0;
1214 
1215 	return attr->mode;
1216 }
1217 
1218 /* papr_scm specific dimm attributes */
1219 static struct attribute *papr_nd_attributes[] = {
1220 	&dev_attr_flags.attr,
1221 	&dev_attr_perf_stats.attr,
1222 	&dev_attr_dirty_shutdown.attr,
1223 	&dev_attr_health_bitmap_inject.attr,
1224 	NULL,
1225 };
1226 
1227 static const struct attribute_group papr_nd_attribute_group = {
1228 	.name = "papr",
1229 	.is_visible = papr_nd_attribute_visible,
1230 	.attrs = papr_nd_attributes,
1231 };
1232 
1233 static const struct attribute_group *papr_nd_attr_groups[] = {
1234 	&papr_nd_attribute_group,
1235 	NULL,
1236 };
1237 
1238 static int papr_scm_nvdimm_init(struct papr_scm_priv *p)
1239 {
1240 	struct device *dev = &p->pdev->dev;
1241 	struct nd_mapping_desc mapping;
1242 	struct nd_region_desc ndr_desc;
1243 	unsigned long dimm_flags;
1244 	int target_nid, online_nid;
1245 
1246 	p->bus_desc.ndctl = papr_scm_ndctl;
1247 	p->bus_desc.module = THIS_MODULE;
1248 	p->bus_desc.of_node = p->pdev->dev.of_node;
1249 	p->bus_desc.provider_name = kstrdup(p->pdev->name, GFP_KERNEL);
1250 
1251 	/* Set the dimm command family mask to accept PDSMs */
1252 	set_bit(NVDIMM_FAMILY_PAPR, &p->bus_desc.dimm_family_mask);
1253 
1254 	if (!p->bus_desc.provider_name)
1255 		return -ENOMEM;
1256 
1257 	p->bus = nvdimm_bus_register(NULL, &p->bus_desc);
1258 	if (!p->bus) {
1259 		dev_err(dev, "Error creating nvdimm bus %pOF\n", p->dn);
1260 		kfree(p->bus_desc.provider_name);
1261 		return -ENXIO;
1262 	}
1263 
1264 	dimm_flags = 0;
1265 	set_bit(NDD_LABELING, &dimm_flags);
1266 
1267 	/*
1268 	 * Check if the nvdimm is unarmed. No locking needed as we are still
1269 	 * initializing. Ignore error encountered if any.
1270 	 */
1271 	__drc_pmem_query_health(p);
1272 
1273 	if (p->health_bitmap & PAPR_PMEM_UNARMED_MASK)
1274 		set_bit(NDD_UNARMED, &dimm_flags);
1275 
1276 	p->nvdimm = nvdimm_create(p->bus, p, papr_nd_attr_groups,
1277 				  dimm_flags, PAPR_SCM_DIMM_CMD_MASK, 0, NULL);
1278 	if (!p->nvdimm) {
1279 		dev_err(dev, "Error creating DIMM object for %pOF\n", p->dn);
1280 		goto err;
1281 	}
1282 
1283 	if (nvdimm_bus_check_dimm_count(p->bus, 1))
1284 		goto err;
1285 
1286 	/* now add the region */
1287 
1288 	memset(&mapping, 0, sizeof(mapping));
1289 	mapping.nvdimm = p->nvdimm;
1290 	mapping.start = 0;
1291 	mapping.size = p->blocks * p->block_size; // XXX: potential overflow?
1292 
1293 	memset(&ndr_desc, 0, sizeof(ndr_desc));
1294 	target_nid = dev_to_node(&p->pdev->dev);
1295 	online_nid = numa_map_to_online_node(target_nid);
1296 	ndr_desc.numa_node = online_nid;
1297 	ndr_desc.target_node = target_nid;
1298 	ndr_desc.res = &p->res;
1299 	ndr_desc.of_node = p->dn;
1300 	ndr_desc.provider_data = p;
1301 	ndr_desc.mapping = &mapping;
1302 	ndr_desc.num_mappings = 1;
1303 	ndr_desc.nd_set = &p->nd_set;
1304 
1305 	if (p->hcall_flush_required) {
1306 		set_bit(ND_REGION_ASYNC, &ndr_desc.flags);
1307 		ndr_desc.flush = papr_scm_pmem_flush;
1308 	}
1309 
1310 	if (p->is_volatile)
1311 		p->region = nvdimm_volatile_region_create(p->bus, &ndr_desc);
1312 	else {
1313 		set_bit(ND_REGION_PERSIST_MEMCTRL, &ndr_desc.flags);
1314 		p->region = nvdimm_pmem_region_create(p->bus, &ndr_desc);
1315 	}
1316 	if (!p->region) {
1317 		dev_err(dev, "Error registering region %pR from %pOF\n",
1318 				ndr_desc.res, p->dn);
1319 		goto err;
1320 	}
1321 	if (target_nid != online_nid)
1322 		dev_info(dev, "Region registered with target node %d and online node %d",
1323 			 target_nid, online_nid);
1324 
1325 	mutex_lock(&papr_ndr_lock);
1326 	list_add_tail(&p->region_list, &papr_nd_regions);
1327 	mutex_unlock(&papr_ndr_lock);
1328 
1329 	return 0;
1330 
1331 err:	nvdimm_bus_unregister(p->bus);
1332 	kfree(p->bus_desc.provider_name);
1333 	return -ENXIO;
1334 }
1335 
1336 static void papr_scm_add_badblock(struct nd_region *region,
1337 				  struct nvdimm_bus *bus, u64 phys_addr)
1338 {
1339 	u64 aligned_addr = ALIGN_DOWN(phys_addr, L1_CACHE_BYTES);
1340 
1341 	if (nvdimm_bus_add_badrange(bus, aligned_addr, L1_CACHE_BYTES)) {
1342 		pr_err("Bad block registration for 0x%llx failed\n", phys_addr);
1343 		return;
1344 	}
1345 
1346 	pr_debug("Add memory range (0x%llx - 0x%llx) as bad range\n",
1347 		 aligned_addr, aligned_addr + L1_CACHE_BYTES);
1348 
1349 	nvdimm_region_notify(region, NVDIMM_REVALIDATE_POISON);
1350 }
1351 
1352 static int handle_mce_ue(struct notifier_block *nb, unsigned long val,
1353 			 void *data)
1354 {
1355 	struct machine_check_event *evt = data;
1356 	struct papr_scm_priv *p;
1357 	u64 phys_addr;
1358 	bool found = false;
1359 
1360 	if (evt->error_type != MCE_ERROR_TYPE_UE)
1361 		return NOTIFY_DONE;
1362 
1363 	if (list_empty(&papr_nd_regions))
1364 		return NOTIFY_DONE;
1365 
1366 	/*
1367 	 * The physical address obtained here is PAGE_SIZE aligned, so get the
1368 	 * exact address from the effective address
1369 	 */
1370 	phys_addr = evt->u.ue_error.physical_address +
1371 			(evt->u.ue_error.effective_address & ~PAGE_MASK);
1372 
1373 	if (!evt->u.ue_error.physical_address_provided ||
1374 	    !is_zone_device_page(pfn_to_page(phys_addr >> PAGE_SHIFT)))
1375 		return NOTIFY_DONE;
1376 
1377 	/* mce notifier is called from a process context, so mutex is safe */
1378 	mutex_lock(&papr_ndr_lock);
1379 	list_for_each_entry(p, &papr_nd_regions, region_list) {
1380 		if (phys_addr >= p->res.start && phys_addr <= p->res.end) {
1381 			found = true;
1382 			break;
1383 		}
1384 	}
1385 
1386 	if (found)
1387 		papr_scm_add_badblock(p->region, p->bus, phys_addr);
1388 
1389 	mutex_unlock(&papr_ndr_lock);
1390 
1391 	return found ? NOTIFY_OK : NOTIFY_DONE;
1392 }
1393 
1394 static struct notifier_block mce_ue_nb = {
1395 	.notifier_call = handle_mce_ue
1396 };
1397 
1398 static int papr_scm_probe(struct platform_device *pdev)
1399 {
1400 	struct device_node *dn = pdev->dev.of_node;
1401 	u32 drc_index, metadata_size;
1402 	u64 blocks, block_size;
1403 	struct papr_scm_priv *p;
1404 	u8 uuid_raw[UUID_SIZE];
1405 	const char *uuid_str;
1406 	ssize_t stat_size;
1407 	uuid_t uuid;
1408 	int rc;
1409 
1410 	/* check we have all the required DT properties */
1411 	if (of_property_read_u32(dn, "ibm,my-drc-index", &drc_index)) {
1412 		dev_err(&pdev->dev, "%pOF: missing drc-index!\n", dn);
1413 		return -ENODEV;
1414 	}
1415 
1416 	if (of_property_read_u64(dn, "ibm,block-size", &block_size)) {
1417 		dev_err(&pdev->dev, "%pOF: missing block-size!\n", dn);
1418 		return -ENODEV;
1419 	}
1420 
1421 	if (of_property_read_u64(dn, "ibm,number-of-blocks", &blocks)) {
1422 		dev_err(&pdev->dev, "%pOF: missing number-of-blocks!\n", dn);
1423 		return -ENODEV;
1424 	}
1425 
1426 	if (of_property_read_string(dn, "ibm,unit-guid", &uuid_str)) {
1427 		dev_err(&pdev->dev, "%pOF: missing unit-guid!\n", dn);
1428 		return -ENODEV;
1429 	}
1430 
1431 	/*
1432 	 * open firmware platform device create won't update the NUMA
1433 	 * distance table. For PAPR SCM devices we use numa_map_to_online_node()
1434 	 * to find the nearest online NUMA node and that requires correct
1435 	 * distance table information.
1436 	 */
1437 	update_numa_distance(dn);
1438 
1439 	p = kzalloc(sizeof(*p), GFP_KERNEL);
1440 	if (!p)
1441 		return -ENOMEM;
1442 
1443 	/* Initialize the dimm mutex */
1444 	mutex_init(&p->health_mutex);
1445 
1446 	/* optional DT properties */
1447 	of_property_read_u32(dn, "ibm,metadata-size", &metadata_size);
1448 
1449 	p->dn = dn;
1450 	p->drc_index = drc_index;
1451 	p->block_size = block_size;
1452 	p->blocks = blocks;
1453 	p->is_volatile = !of_property_read_bool(dn, "ibm,cache-flush-required");
1454 	p->hcall_flush_required = of_property_read_bool(dn, "ibm,hcall-flush-required");
1455 
1456 	if (of_property_read_u64(dn, "ibm,persistence-failed-count",
1457 				 &p->dirty_shutdown_counter))
1458 		p->dirty_shutdown_counter = 0;
1459 
1460 	/* We just need to ensure that set cookies are unique across */
1461 	uuid_parse(uuid_str, &uuid);
1462 
1463 	/*
1464 	 * The cookie1 and cookie2 are not really little endian.
1465 	 * We store a raw buffer representation of the
1466 	 * uuid string so that we can compare this with the label
1467 	 * area cookie irrespective of the endian configuration
1468 	 * with which the kernel is built.
1469 	 *
1470 	 * Historically we stored the cookie in the below format.
1471 	 * for a uuid string 72511b67-0b3b-42fd-8d1d-5be3cae8bcaa
1472 	 *	cookie1 was 0xfd423b0b671b5172
1473 	 *	cookie2 was 0xaabce8cae35b1d8d
1474 	 */
1475 	export_uuid(uuid_raw, &uuid);
1476 	p->nd_set.cookie1 = get_unaligned_le64(&uuid_raw[0]);
1477 	p->nd_set.cookie2 = get_unaligned_le64(&uuid_raw[8]);
1478 
1479 	/* might be zero */
1480 	p->metadata_size = metadata_size;
1481 	p->pdev = pdev;
1482 
1483 	/* request the hypervisor to bind this region to somewhere in memory */
1484 	rc = drc_pmem_bind(p);
1485 
1486 	/* If phyp says drc memory still bound then force unbound and retry */
1487 	if (rc == H_OVERLAP)
1488 		rc = drc_pmem_query_n_bind(p);
1489 
1490 	if (rc != H_SUCCESS) {
1491 		dev_err(&p->pdev->dev, "bind err: %d\n", rc);
1492 		rc = -ENXIO;
1493 		goto err;
1494 	}
1495 
1496 	/* setup the resource for the newly bound range */
1497 	p->res.start = p->bound_addr;
1498 	p->res.end   = p->bound_addr + p->blocks * p->block_size - 1;
1499 	p->res.name  = pdev->name;
1500 	p->res.flags = IORESOURCE_MEM;
1501 
1502 	/* Try retrieving the stat buffer and see if its supported */
1503 	stat_size = drc_pmem_query_stats(p, NULL, 0);
1504 	if (stat_size > 0) {
1505 		p->stat_buffer_len = stat_size;
1506 		dev_dbg(&p->pdev->dev, "Max perf-stat size %lu-bytes\n",
1507 			p->stat_buffer_len);
1508 	}
1509 
1510 	rc = papr_scm_nvdimm_init(p);
1511 	if (rc)
1512 		goto err2;
1513 
1514 	platform_set_drvdata(pdev, p);
1515 	papr_scm_pmu_register(p);
1516 
1517 	return 0;
1518 
1519 err2:	drc_pmem_unbind(p);
1520 err:	kfree(p);
1521 	return rc;
1522 }
1523 
1524 static int papr_scm_remove(struct platform_device *pdev)
1525 {
1526 	struct papr_scm_priv *p = platform_get_drvdata(pdev);
1527 
1528 	mutex_lock(&papr_ndr_lock);
1529 	list_del(&p->region_list);
1530 	mutex_unlock(&papr_ndr_lock);
1531 
1532 	nvdimm_bus_unregister(p->bus);
1533 	drc_pmem_unbind(p);
1534 
1535 	if (pdev->archdata.priv)
1536 		unregister_nvdimm_pmu(pdev->archdata.priv);
1537 
1538 	pdev->archdata.priv = NULL;
1539 	kfree(p->bus_desc.provider_name);
1540 	kfree(p);
1541 
1542 	return 0;
1543 }
1544 
1545 static const struct of_device_id papr_scm_match[] = {
1546 	{ .compatible = "ibm,pmemory" },
1547 	{ .compatible = "ibm,pmemory-v2" },
1548 	{ },
1549 };
1550 
1551 static struct platform_driver papr_scm_driver = {
1552 	.probe = papr_scm_probe,
1553 	.remove = papr_scm_remove,
1554 	.driver = {
1555 		.name = "papr_scm",
1556 		.of_match_table = papr_scm_match,
1557 	},
1558 };
1559 
1560 static int __init papr_scm_init(void)
1561 {
1562 	int ret;
1563 
1564 	ret = platform_driver_register(&papr_scm_driver);
1565 	if (!ret)
1566 		mce_register_notifier(&mce_ue_nb);
1567 
1568 	return ret;
1569 }
1570 module_init(papr_scm_init);
1571 
1572 static void __exit papr_scm_exit(void)
1573 {
1574 	mce_unregister_notifier(&mce_ue_nb);
1575 	platform_driver_unregister(&papr_scm_driver);
1576 }
1577 module_exit(papr_scm_exit);
1578 
1579 MODULE_DEVICE_TABLE(of, papr_scm_match);
1580 MODULE_LICENSE("GPL");
1581 MODULE_AUTHOR("IBM Corporation");
1582