xref: /linux/arch/powerpc/platforms/pseries/papr_scm.c (revision 7ec462100ef9142344ddbf86f2c3008b97acddbe)
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 <uapi/linux/papr_pdsm.h>
20 #include <linux/papr_scm.h>
21 #include <asm/mce.h>
22 #include <linux/unaligned.h>
23 #include <linux/perf_event.h>
24 
25 #define BIND_ANY_ADDR (~0ul)
26 
27 #define PAPR_SCM_DIMM_CMD_MASK \
28 	((1ul << ND_CMD_GET_CONFIG_SIZE) | \
29 	 (1ul << ND_CMD_GET_CONFIG_DATA) | \
30 	 (1ul << ND_CMD_SET_CONFIG_DATA) | \
31 	 (1ul << ND_CMD_CALL))
32 
33 /* Struct holding a single performance metric */
34 struct papr_scm_perf_stat {
35 	u8 stat_id[8];
36 	__be64 stat_val;
37 } __packed;
38 
39 /* Struct exchanged between kernel and PHYP for fetching drc perf stats */
40 struct papr_scm_perf_stats {
41 	u8 eye_catcher[8];
42 	/* Should be PAPR_SCM_PERF_STATS_VERSION */
43 	__be32 stats_version;
44 	/* Number of stats following */
45 	__be32 num_statistics;
46 	/* zero or more performance matrics */
47 	struct papr_scm_perf_stat scm_statistic[];
48 } __packed;
49 
50 /* private struct associated with each region */
51 struct papr_scm_priv {
52 	struct platform_device *pdev;
53 	struct device_node *dn;
54 	uint32_t drc_index;
55 	uint64_t blocks;
56 	uint64_t block_size;
57 	int metadata_size;
58 	bool is_volatile;
59 	bool hcall_flush_required;
60 
61 	uint64_t bound_addr;
62 
63 	struct nvdimm_bus_descriptor bus_desc;
64 	struct nvdimm_bus *bus;
65 	struct nvdimm *nvdimm;
66 	struct resource res;
67 	struct nd_region *region;
68 	struct nd_interleave_set nd_set;
69 	struct list_head region_list;
70 
71 	/* Protect dimm health data from concurrent read/writes */
72 	struct mutex health_mutex;
73 
74 	/* Last time the health information of the dimm was updated */
75 	unsigned long lasthealth_jiffies;
76 
77 	/* Health information for the dimm */
78 	u64 health_bitmap;
79 
80 	/* Holds the last known dirty shutdown counter value */
81 	u64 dirty_shutdown_counter;
82 
83 	/* length of the stat buffer as expected by phyp */
84 	size_t stat_buffer_len;
85 
86 	/* The bits which needs to be overridden */
87 	u64 health_bitmap_inject_mask;
88 };
89 
papr_scm_pmem_flush(struct nd_region * nd_region,struct bio * bio __maybe_unused)90 static int papr_scm_pmem_flush(struct nd_region *nd_region,
91 			       struct bio *bio __maybe_unused)
92 {
93 	struct papr_scm_priv *p = nd_region_provider_data(nd_region);
94 	unsigned long ret_buf[PLPAR_HCALL_BUFSIZE], token = 0;
95 	long rc;
96 
97 	dev_dbg(&p->pdev->dev, "flush drc 0x%x", p->drc_index);
98 
99 	do {
100 		rc = plpar_hcall(H_SCM_FLUSH, ret_buf, p->drc_index, token);
101 		token = ret_buf[0];
102 
103 		/* Check if we are stalled for some time */
104 		if (H_IS_LONG_BUSY(rc)) {
105 			msleep(get_longbusy_msecs(rc));
106 			rc = H_BUSY;
107 		} else if (rc == H_BUSY) {
108 			cond_resched();
109 		}
110 	} while (rc == H_BUSY);
111 
112 	if (rc) {
113 		dev_err(&p->pdev->dev, "flush error: %ld", rc);
114 		rc = -EIO;
115 	} else {
116 		dev_dbg(&p->pdev->dev, "flush drc 0x%x complete", p->drc_index);
117 	}
118 
119 	return rc;
120 }
121 
122 static LIST_HEAD(papr_nd_regions);
123 static DEFINE_MUTEX(papr_ndr_lock);
124 
drc_pmem_bind(struct papr_scm_priv * p)125 static int drc_pmem_bind(struct papr_scm_priv *p)
126 {
127 	unsigned long ret[PLPAR_HCALL_BUFSIZE];
128 	uint64_t saved = 0;
129 	uint64_t token;
130 	int64_t rc;
131 
132 	/*
133 	 * When the hypervisor cannot map all the requested memory in a single
134 	 * hcall it returns H_BUSY and we call again with the token until
135 	 * we get H_SUCCESS. Aborting the retry loop before getting H_SUCCESS
136 	 * leave the system in an undefined state, so we wait.
137 	 */
138 	token = 0;
139 
140 	do {
141 		rc = plpar_hcall(H_SCM_BIND_MEM, ret, p->drc_index, 0,
142 				p->blocks, BIND_ANY_ADDR, token);
143 		token = ret[0];
144 		if (!saved)
145 			saved = ret[1];
146 		cond_resched();
147 	} while (rc == H_BUSY);
148 
149 	if (rc)
150 		return rc;
151 
152 	p->bound_addr = saved;
153 	dev_dbg(&p->pdev->dev, "bound drc 0x%x to 0x%lx\n",
154 		p->drc_index, (unsigned long)saved);
155 	return rc;
156 }
157 
drc_pmem_unbind(struct papr_scm_priv * p)158 static void drc_pmem_unbind(struct papr_scm_priv *p)
159 {
160 	unsigned long ret[PLPAR_HCALL_BUFSIZE];
161 	uint64_t token = 0;
162 	int64_t rc;
163 
164 	dev_dbg(&p->pdev->dev, "unbind drc 0x%x\n", p->drc_index);
165 
166 	/* NB: unbind has the same retry requirements as drc_pmem_bind() */
167 	do {
168 
169 		/* Unbind of all SCM resources associated with drcIndex */
170 		rc = plpar_hcall(H_SCM_UNBIND_ALL, ret, H_UNBIND_SCOPE_DRC,
171 				 p->drc_index, token);
172 		token = ret[0];
173 
174 		/* Check if we are stalled for some time */
175 		if (H_IS_LONG_BUSY(rc)) {
176 			msleep(get_longbusy_msecs(rc));
177 			rc = H_BUSY;
178 		} else if (rc == H_BUSY) {
179 			cond_resched();
180 		}
181 
182 	} while (rc == H_BUSY);
183 
184 	if (rc)
185 		dev_err(&p->pdev->dev, "unbind error: %lld\n", rc);
186 	else
187 		dev_dbg(&p->pdev->dev, "unbind drc 0x%x complete\n",
188 			p->drc_index);
189 
190 	return;
191 }
192 
drc_pmem_query_n_bind(struct papr_scm_priv * p)193 static int drc_pmem_query_n_bind(struct papr_scm_priv *p)
194 {
195 	unsigned long start_addr;
196 	unsigned long end_addr;
197 	unsigned long ret[PLPAR_HCALL_BUFSIZE];
198 	int64_t rc;
199 
200 
201 	rc = plpar_hcall(H_SCM_QUERY_BLOCK_MEM_BINDING, ret,
202 			 p->drc_index, 0);
203 	if (rc)
204 		goto err_out;
205 	start_addr = ret[0];
206 
207 	/* Make sure the full region is bound. */
208 	rc = plpar_hcall(H_SCM_QUERY_BLOCK_MEM_BINDING, ret,
209 			 p->drc_index, p->blocks - 1);
210 	if (rc)
211 		goto err_out;
212 	end_addr = ret[0];
213 
214 	if ((end_addr - start_addr) != ((p->blocks - 1) * p->block_size))
215 		goto err_out;
216 
217 	p->bound_addr = start_addr;
218 	dev_dbg(&p->pdev->dev, "bound drc 0x%x to 0x%lx\n", p->drc_index, start_addr);
219 	return rc;
220 
221 err_out:
222 	dev_info(&p->pdev->dev,
223 		 "Failed to query, trying an unbind followed by bind");
224 	drc_pmem_unbind(p);
225 	return drc_pmem_bind(p);
226 }
227 
228 /*
229  * Query the Dimm performance stats from PHYP and copy them (if returned) to
230  * provided struct papr_scm_perf_stats instance 'stats' that can hold atleast
231  * (num_stats + header) bytes.
232  * - If buff_stats == NULL the return value is the size in bytes of the buffer
233  * needed to hold all supported performance-statistics.
234  * - If buff_stats != NULL and num_stats == 0 then we copy all known
235  * performance-statistics to 'buff_stat' and expect to be large enough to
236  * hold them.
237  * - if buff_stats != NULL and num_stats > 0 then copy the requested
238  * performance-statistics to buff_stats.
239  */
drc_pmem_query_stats(struct papr_scm_priv * p,struct papr_scm_perf_stats * buff_stats,unsigned int num_stats)240 static ssize_t drc_pmem_query_stats(struct papr_scm_priv *p,
241 				    struct papr_scm_perf_stats *buff_stats,
242 				    unsigned int num_stats)
243 {
244 	unsigned long ret[PLPAR_HCALL_BUFSIZE];
245 	size_t size;
246 	s64 rc;
247 
248 	/* Setup the out buffer */
249 	if (buff_stats) {
250 		memcpy(buff_stats->eye_catcher,
251 		       PAPR_SCM_PERF_STATS_EYECATCHER, 8);
252 		buff_stats->stats_version =
253 			cpu_to_be32(PAPR_SCM_PERF_STATS_VERSION);
254 		buff_stats->num_statistics =
255 			cpu_to_be32(num_stats);
256 
257 		/*
258 		 * Calculate the buffer size based on num-stats provided
259 		 * or use the prefetched max buffer length
260 		 */
261 		if (num_stats)
262 			/* Calculate size from the num_stats */
263 			size = sizeof(struct papr_scm_perf_stats) +
264 				num_stats * sizeof(struct papr_scm_perf_stat);
265 		else
266 			size = p->stat_buffer_len;
267 	} else {
268 		/* In case of no out buffer ignore the size */
269 		size = 0;
270 	}
271 
272 	/* Do the HCALL asking PHYP for info */
273 	rc = plpar_hcall(H_SCM_PERFORMANCE_STATS, ret, p->drc_index,
274 			 buff_stats ? virt_to_phys(buff_stats) : 0,
275 			 size);
276 
277 	/* Check if the error was due to an unknown stat-id */
278 	if (rc == H_PARTIAL) {
279 		dev_err(&p->pdev->dev,
280 			"Unknown performance stats, Err:0x%016lX\n", ret[0]);
281 		return -ENOENT;
282 	} else if (rc == H_AUTHORITY) {
283 		dev_info(&p->pdev->dev,
284 			 "Permission denied while accessing performance stats");
285 		return -EPERM;
286 	} else if (rc == H_UNSUPPORTED) {
287 		dev_dbg(&p->pdev->dev, "Performance stats unsupported\n");
288 		return -EOPNOTSUPP;
289 	} else if (rc != H_SUCCESS) {
290 		dev_err(&p->pdev->dev,
291 			"Failed to query performance stats, Err:%lld\n", rc);
292 		return -EIO;
293 
294 	} else if (!size) {
295 		/* Handle case where stat buffer size was requested */
296 		dev_dbg(&p->pdev->dev,
297 			"Performance stats size %ld\n", ret[0]);
298 		return ret[0];
299 	}
300 
301 	/* Successfully fetched the requested stats from phyp */
302 	dev_dbg(&p->pdev->dev,
303 		"Performance stats returned %d stats\n",
304 		be32_to_cpu(buff_stats->num_statistics));
305 	return 0;
306 }
307 
308 #ifdef CONFIG_PERF_EVENTS
309 #define to_nvdimm_pmu(_pmu)	container_of(_pmu, struct nvdimm_pmu, pmu)
310 
311 static const char * const nvdimm_events_map[] = {
312 	[1] = "CtlResCt",
313 	[2] = "CtlResTm",
314 	[3] = "PonSecs ",
315 	[4] = "MemLife ",
316 	[5] = "CritRscU",
317 	[6] = "HostLCnt",
318 	[7] = "HostSCnt",
319 	[8] = "HostSDur",
320 	[9] = "HostLDur",
321 	[10] = "MedRCnt ",
322 	[11] = "MedWCnt ",
323 	[12] = "MedRDur ",
324 	[13] = "MedWDur ",
325 	[14] = "CchRHCnt",
326 	[15] = "CchWHCnt",
327 	[16] = "FastWCnt",
328 };
329 
papr_scm_pmu_get_value(struct perf_event * event,struct device * dev,u64 * count)330 static int papr_scm_pmu_get_value(struct perf_event *event, struct device *dev, u64 *count)
331 {
332 	struct papr_scm_perf_stat *stat;
333 	struct papr_scm_perf_stats *stats;
334 	struct papr_scm_priv *p = dev_get_drvdata(dev);
335 	int rc, size;
336 
337 	/* Invalid eventcode */
338 	if (event->attr.config == 0 || event->attr.config >= ARRAY_SIZE(nvdimm_events_map))
339 		return -EINVAL;
340 
341 	/* Allocate request buffer enough to hold single performance stat */
342 	size = sizeof(struct papr_scm_perf_stats) +
343 		sizeof(struct papr_scm_perf_stat);
344 
345 	if (!p)
346 		return -EINVAL;
347 
348 	stats = kzalloc(size, GFP_KERNEL);
349 	if (!stats)
350 		return -ENOMEM;
351 
352 	stat = &stats->scm_statistic[0];
353 	memcpy(&stat->stat_id,
354 	       nvdimm_events_map[event->attr.config],
355 		sizeof(stat->stat_id));
356 	stat->stat_val = 0;
357 
358 	rc = drc_pmem_query_stats(p, stats, 1);
359 	if (rc < 0) {
360 		kfree(stats);
361 		return rc;
362 	}
363 
364 	*count = be64_to_cpu(stat->stat_val);
365 	kfree(stats);
366 	return 0;
367 }
368 
papr_scm_pmu_event_init(struct perf_event * event)369 static int papr_scm_pmu_event_init(struct perf_event *event)
370 {
371 	struct nvdimm_pmu *nd_pmu = to_nvdimm_pmu(event->pmu);
372 	struct papr_scm_priv *p;
373 
374 	if (!nd_pmu)
375 		return -EINVAL;
376 
377 	/* test the event attr type for PMU enumeration */
378 	if (event->attr.type != event->pmu->type)
379 		return -ENOENT;
380 
381 	/* it does not support event sampling mode */
382 	if (is_sampling_event(event))
383 		return -EOPNOTSUPP;
384 
385 	/* no branch sampling */
386 	if (has_branch_stack(event))
387 		return -EOPNOTSUPP;
388 
389 	p = (struct papr_scm_priv *)nd_pmu->dev->driver_data;
390 	if (!p)
391 		return -EINVAL;
392 
393 	/* Invalid eventcode */
394 	if (event->attr.config == 0 || event->attr.config > 16)
395 		return -EINVAL;
396 
397 	return 0;
398 }
399 
papr_scm_pmu_add(struct perf_event * event,int flags)400 static int papr_scm_pmu_add(struct perf_event *event, int flags)
401 {
402 	u64 count;
403 	int rc;
404 	struct nvdimm_pmu *nd_pmu = to_nvdimm_pmu(event->pmu);
405 
406 	if (!nd_pmu)
407 		return -EINVAL;
408 
409 	if (flags & PERF_EF_START) {
410 		rc = papr_scm_pmu_get_value(event, nd_pmu->dev, &count);
411 		if (rc)
412 			return rc;
413 
414 		local64_set(&event->hw.prev_count, count);
415 	}
416 
417 	return 0;
418 }
419 
papr_scm_pmu_read(struct perf_event * event)420 static void papr_scm_pmu_read(struct perf_event *event)
421 {
422 	u64 prev, now;
423 	int rc;
424 	struct nvdimm_pmu *nd_pmu = to_nvdimm_pmu(event->pmu);
425 
426 	if (!nd_pmu)
427 		return;
428 
429 	rc = papr_scm_pmu_get_value(event, nd_pmu->dev, &now);
430 	if (rc)
431 		return;
432 
433 	prev = local64_xchg(&event->hw.prev_count, now);
434 	local64_add(now - prev, &event->count);
435 }
436 
papr_scm_pmu_del(struct perf_event * event,int flags)437 static void papr_scm_pmu_del(struct perf_event *event, int flags)
438 {
439 	papr_scm_pmu_read(event);
440 }
441 
papr_scm_pmu_register(struct papr_scm_priv * p)442 static void papr_scm_pmu_register(struct papr_scm_priv *p)
443 {
444 	struct nvdimm_pmu *nd_pmu;
445 	int rc, nodeid;
446 
447 	nd_pmu = kzalloc(sizeof(*nd_pmu), GFP_KERNEL);
448 	if (!nd_pmu) {
449 		rc = -ENOMEM;
450 		goto pmu_err_print;
451 	}
452 
453 	if (!p->stat_buffer_len) {
454 		rc = -ENOENT;
455 		goto pmu_check_events_err;
456 	}
457 
458 	nd_pmu->pmu.task_ctx_nr = perf_invalid_context;
459 	nd_pmu->pmu.name = nvdimm_name(p->nvdimm);
460 	nd_pmu->pmu.event_init = papr_scm_pmu_event_init;
461 	nd_pmu->pmu.read = papr_scm_pmu_read;
462 	nd_pmu->pmu.add = papr_scm_pmu_add;
463 	nd_pmu->pmu.del = papr_scm_pmu_del;
464 
465 	nd_pmu->pmu.capabilities = PERF_PMU_CAP_NO_INTERRUPT |
466 				PERF_PMU_CAP_NO_EXCLUDE;
467 
468 	/*updating the cpumask variable */
469 	nodeid = numa_map_to_online_node(dev_to_node(&p->pdev->dev));
470 	nd_pmu->arch_cpumask = *cpumask_of_node(nodeid);
471 
472 	rc = register_nvdimm_pmu(nd_pmu, p->pdev);
473 	if (rc)
474 		goto pmu_check_events_err;
475 
476 	/*
477 	 * Set archdata.priv value to nvdimm_pmu structure, to handle the
478 	 * unregistering of pmu device.
479 	 */
480 	p->pdev->archdata.priv = nd_pmu;
481 	return;
482 
483 pmu_check_events_err:
484 	kfree(nd_pmu);
485 pmu_err_print:
486 	dev_info(&p->pdev->dev, "nvdimm pmu didn't register rc=%d\n", rc);
487 }
488 
489 #else
papr_scm_pmu_register(struct papr_scm_priv * p)490 static void papr_scm_pmu_register(struct papr_scm_priv *p) { }
491 #endif
492 
493 /*
494  * Issue hcall to retrieve dimm health info and populate papr_scm_priv with the
495  * health information.
496  */
__drc_pmem_query_health(struct papr_scm_priv * p)497 static int __drc_pmem_query_health(struct papr_scm_priv *p)
498 {
499 	unsigned long ret[PLPAR_HCALL_BUFSIZE];
500 	u64 bitmap = 0;
501 	long rc;
502 
503 	/* issue the hcall */
504 	rc = plpar_hcall(H_SCM_HEALTH, ret, p->drc_index);
505 	if (rc == H_SUCCESS)
506 		bitmap = ret[0] & ret[1];
507 	else if (rc == H_FUNCTION)
508 		dev_info_once(&p->pdev->dev,
509 			      "Hcall H_SCM_HEALTH not implemented, assuming empty health bitmap");
510 	else {
511 
512 		dev_err(&p->pdev->dev,
513 			"Failed to query health information, Err:%ld\n", rc);
514 		return -ENXIO;
515 	}
516 
517 	p->lasthealth_jiffies = jiffies;
518 	/* Allow injecting specific health bits via inject mask. */
519 	if (p->health_bitmap_inject_mask)
520 		bitmap = (bitmap & ~p->health_bitmap_inject_mask) |
521 			p->health_bitmap_inject_mask;
522 	WRITE_ONCE(p->health_bitmap, bitmap);
523 	dev_dbg(&p->pdev->dev,
524 		"Queried dimm health info. Bitmap:0x%016lx Mask:0x%016lx\n",
525 		ret[0], ret[1]);
526 
527 	return 0;
528 }
529 
530 /* Min interval in seconds for assuming stable dimm health */
531 #define MIN_HEALTH_QUERY_INTERVAL 60
532 
533 /* Query cached health info and if needed call drc_pmem_query_health */
drc_pmem_query_health(struct papr_scm_priv * p)534 static int drc_pmem_query_health(struct papr_scm_priv *p)
535 {
536 	unsigned long cache_timeout;
537 	int rc;
538 
539 	/* Protect concurrent modifications to papr_scm_priv */
540 	rc = mutex_lock_interruptible(&p->health_mutex);
541 	if (rc)
542 		return rc;
543 
544 	/* Jiffies offset for which the health data is assumed to be same */
545 	cache_timeout = p->lasthealth_jiffies +
546 		msecs_to_jiffies(MIN_HEALTH_QUERY_INTERVAL * 1000);
547 
548 	/* Fetch new health info is its older than MIN_HEALTH_QUERY_INTERVAL */
549 	if (time_after(jiffies, cache_timeout))
550 		rc = __drc_pmem_query_health(p);
551 	else
552 		/* Assume cached health data is valid */
553 		rc = 0;
554 
555 	mutex_unlock(&p->health_mutex);
556 	return rc;
557 }
558 
papr_scm_meta_get(struct papr_scm_priv * p,struct nd_cmd_get_config_data_hdr * hdr)559 static int papr_scm_meta_get(struct papr_scm_priv *p,
560 			     struct nd_cmd_get_config_data_hdr *hdr)
561 {
562 	unsigned long data[PLPAR_HCALL_BUFSIZE];
563 	unsigned long offset, data_offset;
564 	int len, read;
565 	int64_t ret;
566 
567 	if ((hdr->in_offset + hdr->in_length) > p->metadata_size)
568 		return -EINVAL;
569 
570 	for (len = hdr->in_length; len; len -= read) {
571 
572 		data_offset = hdr->in_length - len;
573 		offset = hdr->in_offset + data_offset;
574 
575 		if (len >= 8)
576 			read = 8;
577 		else if (len >= 4)
578 			read = 4;
579 		else if (len >= 2)
580 			read = 2;
581 		else
582 			read = 1;
583 
584 		ret = plpar_hcall(H_SCM_READ_METADATA, data, p->drc_index,
585 				  offset, read);
586 
587 		if (ret == H_PARAMETER) /* bad DRC index */
588 			return -ENODEV;
589 		if (ret)
590 			return -EINVAL; /* other invalid parameter */
591 
592 		switch (read) {
593 		case 8:
594 			*(uint64_t *)(hdr->out_buf + data_offset) = be64_to_cpu(data[0]);
595 			break;
596 		case 4:
597 			*(uint32_t *)(hdr->out_buf + data_offset) = be32_to_cpu(data[0] & 0xffffffff);
598 			break;
599 
600 		case 2:
601 			*(uint16_t *)(hdr->out_buf + data_offset) = be16_to_cpu(data[0] & 0xffff);
602 			break;
603 
604 		case 1:
605 			*(uint8_t *)(hdr->out_buf + data_offset) = (data[0] & 0xff);
606 			break;
607 		}
608 	}
609 	return 0;
610 }
611 
papr_scm_meta_set(struct papr_scm_priv * p,struct nd_cmd_set_config_hdr * hdr)612 static int papr_scm_meta_set(struct papr_scm_priv *p,
613 			     struct nd_cmd_set_config_hdr *hdr)
614 {
615 	unsigned long offset, data_offset;
616 	int len, wrote;
617 	unsigned long data;
618 	__be64 data_be;
619 	int64_t ret;
620 
621 	if ((hdr->in_offset + hdr->in_length) > p->metadata_size)
622 		return -EINVAL;
623 
624 	for (len = hdr->in_length; len; len -= wrote) {
625 
626 		data_offset = hdr->in_length - len;
627 		offset = hdr->in_offset + data_offset;
628 
629 		if (len >= 8) {
630 			data = *(uint64_t *)(hdr->in_buf + data_offset);
631 			data_be = cpu_to_be64(data);
632 			wrote = 8;
633 		} else if (len >= 4) {
634 			data = *(uint32_t *)(hdr->in_buf + data_offset);
635 			data &= 0xffffffff;
636 			data_be = cpu_to_be32(data);
637 			wrote = 4;
638 		} else if (len >= 2) {
639 			data = *(uint16_t *)(hdr->in_buf + data_offset);
640 			data &= 0xffff;
641 			data_be = cpu_to_be16(data);
642 			wrote = 2;
643 		} else {
644 			data_be = *(uint8_t *)(hdr->in_buf + data_offset);
645 			data_be &= 0xff;
646 			wrote = 1;
647 		}
648 
649 		ret = plpar_hcall_norets(H_SCM_WRITE_METADATA, p->drc_index,
650 					 offset, data_be, wrote);
651 		if (ret == H_PARAMETER) /* bad DRC index */
652 			return -ENODEV;
653 		if (ret)
654 			return -EINVAL; /* other invalid parameter */
655 	}
656 
657 	return 0;
658 }
659 
660 /*
661  * Do a sanity checks on the inputs args to dimm-control function and return
662  * '0' if valid. Validation of PDSM payloads happens later in
663  * papr_scm_service_pdsm.
664  */
is_cmd_valid(struct nvdimm * nvdimm,unsigned int cmd,void * buf,unsigned int buf_len)665 static int is_cmd_valid(struct nvdimm *nvdimm, unsigned int cmd, void *buf,
666 			unsigned int buf_len)
667 {
668 	unsigned long cmd_mask = PAPR_SCM_DIMM_CMD_MASK;
669 	struct nd_cmd_pkg *nd_cmd;
670 	struct papr_scm_priv *p;
671 	enum papr_pdsm pdsm;
672 
673 	/* Only dimm-specific calls are supported atm */
674 	if (!nvdimm)
675 		return -EINVAL;
676 
677 	/* get the provider data from struct nvdimm */
678 	p = nvdimm_provider_data(nvdimm);
679 
680 	if (!test_bit(cmd, &cmd_mask)) {
681 		dev_dbg(&p->pdev->dev, "Unsupported cmd=%u\n", cmd);
682 		return -EINVAL;
683 	}
684 
685 	/* For CMD_CALL verify pdsm request */
686 	if (cmd == ND_CMD_CALL) {
687 		/* Verify the envelope and envelop size */
688 		if (!buf ||
689 		    buf_len < (sizeof(struct nd_cmd_pkg) + ND_PDSM_HDR_SIZE)) {
690 			dev_dbg(&p->pdev->dev, "Invalid pkg size=%u\n",
691 				buf_len);
692 			return -EINVAL;
693 		}
694 
695 		/* Verify that the nd_cmd_pkg.nd_family is correct */
696 		nd_cmd = (struct nd_cmd_pkg *)buf;
697 
698 		if (nd_cmd->nd_family != NVDIMM_FAMILY_PAPR) {
699 			dev_dbg(&p->pdev->dev, "Invalid pkg family=0x%llx\n",
700 				nd_cmd->nd_family);
701 			return -EINVAL;
702 		}
703 
704 		pdsm = (enum papr_pdsm)nd_cmd->nd_command;
705 
706 		/* Verify if the pdsm command is valid */
707 		if (pdsm <= PAPR_PDSM_MIN || pdsm >= PAPR_PDSM_MAX) {
708 			dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Invalid PDSM\n",
709 				pdsm);
710 			return -EINVAL;
711 		}
712 
713 		/* Have enough space to hold returned 'nd_pkg_pdsm' header */
714 		if (nd_cmd->nd_size_out < ND_PDSM_HDR_SIZE) {
715 			dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Invalid payload\n",
716 				pdsm);
717 			return -EINVAL;
718 		}
719 	}
720 
721 	/* Let the command be further processed */
722 	return 0;
723 }
724 
papr_pdsm_fuel_gauge(struct papr_scm_priv * p,union nd_pdsm_payload * payload)725 static int papr_pdsm_fuel_gauge(struct papr_scm_priv *p,
726 				union nd_pdsm_payload *payload)
727 {
728 	int rc, size;
729 	u64 statval;
730 	struct papr_scm_perf_stat *stat;
731 	struct papr_scm_perf_stats *stats;
732 
733 	/* Silently fail if fetching performance metrics isn't  supported */
734 	if (!p->stat_buffer_len)
735 		return 0;
736 
737 	/* Allocate request buffer enough to hold single performance stat */
738 	size = sizeof(struct papr_scm_perf_stats) +
739 		sizeof(struct papr_scm_perf_stat);
740 
741 	stats = kzalloc(size, GFP_KERNEL);
742 	if (!stats)
743 		return -ENOMEM;
744 
745 	stat = &stats->scm_statistic[0];
746 	memcpy(&stat->stat_id, "MemLife ", sizeof(stat->stat_id));
747 	stat->stat_val = 0;
748 
749 	/* Fetch the fuel gauge and populate it in payload */
750 	rc = drc_pmem_query_stats(p, stats, 1);
751 	if (rc < 0) {
752 		dev_dbg(&p->pdev->dev, "Err(%d) fetching fuel gauge\n", rc);
753 		goto free_stats;
754 	}
755 
756 	statval = be64_to_cpu(stat->stat_val);
757 	dev_dbg(&p->pdev->dev,
758 		"Fetched fuel-gauge %llu", statval);
759 	payload->health.extension_flags |=
760 		PDSM_DIMM_HEALTH_RUN_GAUGE_VALID;
761 	payload->health.dimm_fuel_gauge = statval;
762 
763 	rc = sizeof(struct nd_papr_pdsm_health);
764 
765 free_stats:
766 	kfree(stats);
767 	return rc;
768 }
769 
770 /* Add the dirty-shutdown-counter value to the pdsm */
papr_pdsm_dsc(struct papr_scm_priv * p,union nd_pdsm_payload * payload)771 static int papr_pdsm_dsc(struct papr_scm_priv *p,
772 			 union nd_pdsm_payload *payload)
773 {
774 	payload->health.extension_flags |= PDSM_DIMM_DSC_VALID;
775 	payload->health.dimm_dsc = p->dirty_shutdown_counter;
776 
777 	return sizeof(struct nd_papr_pdsm_health);
778 }
779 
780 /* Fetch the DIMM health info and populate it in provided package. */
papr_pdsm_health(struct papr_scm_priv * p,union nd_pdsm_payload * payload)781 static int papr_pdsm_health(struct papr_scm_priv *p,
782 			    union nd_pdsm_payload *payload)
783 {
784 	int rc;
785 
786 	/* Ensure dimm health mutex is taken preventing concurrent access */
787 	rc = mutex_lock_interruptible(&p->health_mutex);
788 	if (rc)
789 		goto out;
790 
791 	/* Always fetch upto date dimm health data ignoring cached values */
792 	rc = __drc_pmem_query_health(p);
793 	if (rc) {
794 		mutex_unlock(&p->health_mutex);
795 		goto out;
796 	}
797 
798 	/* update health struct with various flags derived from health bitmap */
799 	payload->health = (struct nd_papr_pdsm_health) {
800 		.extension_flags = 0,
801 		.dimm_unarmed = !!(p->health_bitmap & PAPR_PMEM_UNARMED_MASK),
802 		.dimm_bad_shutdown = !!(p->health_bitmap & PAPR_PMEM_BAD_SHUTDOWN_MASK),
803 		.dimm_bad_restore = !!(p->health_bitmap & PAPR_PMEM_BAD_RESTORE_MASK),
804 		.dimm_scrubbed = !!(p->health_bitmap & PAPR_PMEM_SCRUBBED_AND_LOCKED),
805 		.dimm_locked = !!(p->health_bitmap & PAPR_PMEM_SCRUBBED_AND_LOCKED),
806 		.dimm_encrypted = !!(p->health_bitmap & PAPR_PMEM_ENCRYPTED),
807 		.dimm_health = PAPR_PDSM_DIMM_HEALTHY,
808 	};
809 
810 	/* Update field dimm_health based on health_bitmap flags */
811 	if (p->health_bitmap & PAPR_PMEM_HEALTH_FATAL)
812 		payload->health.dimm_health = PAPR_PDSM_DIMM_FATAL;
813 	else if (p->health_bitmap & PAPR_PMEM_HEALTH_CRITICAL)
814 		payload->health.dimm_health = PAPR_PDSM_DIMM_CRITICAL;
815 	else if (p->health_bitmap & PAPR_PMEM_HEALTH_UNHEALTHY)
816 		payload->health.dimm_health = PAPR_PDSM_DIMM_UNHEALTHY;
817 
818 	/* struct populated hence can release the mutex now */
819 	mutex_unlock(&p->health_mutex);
820 
821 	/* Populate the fuel gauge meter in the payload */
822 	papr_pdsm_fuel_gauge(p, payload);
823 	/* Populate the dirty-shutdown-counter field */
824 	papr_pdsm_dsc(p, payload);
825 
826 	rc = sizeof(struct nd_papr_pdsm_health);
827 
828 out:
829 	return rc;
830 }
831 
832 /* Inject a smart error Add the dirty-shutdown-counter value to the pdsm */
papr_pdsm_smart_inject(struct papr_scm_priv * p,union nd_pdsm_payload * payload)833 static int papr_pdsm_smart_inject(struct papr_scm_priv *p,
834 				  union nd_pdsm_payload *payload)
835 {
836 	int rc;
837 	u32 supported_flags = 0;
838 	u64 inject_mask = 0, clear_mask = 0;
839 	u64 mask;
840 
841 	/* Check for individual smart error flags and update inject/clear masks */
842 	if (payload->smart_inject.flags & PDSM_SMART_INJECT_HEALTH_FATAL) {
843 		supported_flags |= PDSM_SMART_INJECT_HEALTH_FATAL;
844 		if (payload->smart_inject.fatal_enable)
845 			inject_mask |= PAPR_PMEM_HEALTH_FATAL;
846 		else
847 			clear_mask |= PAPR_PMEM_HEALTH_FATAL;
848 	}
849 
850 	if (payload->smart_inject.flags & PDSM_SMART_INJECT_BAD_SHUTDOWN) {
851 		supported_flags |= PDSM_SMART_INJECT_BAD_SHUTDOWN;
852 		if (payload->smart_inject.unsafe_shutdown_enable)
853 			inject_mask |= PAPR_PMEM_SHUTDOWN_DIRTY;
854 		else
855 			clear_mask |= PAPR_PMEM_SHUTDOWN_DIRTY;
856 	}
857 
858 	dev_dbg(&p->pdev->dev, "[Smart-inject] inject_mask=%#llx clear_mask=%#llx\n",
859 		inject_mask, clear_mask);
860 
861 	/* Prevent concurrent access to dimm health bitmap related members */
862 	rc = mutex_lock_interruptible(&p->health_mutex);
863 	if (rc)
864 		return rc;
865 
866 	/* Use inject/clear masks to set health_bitmap_inject_mask */
867 	mask = READ_ONCE(p->health_bitmap_inject_mask);
868 	mask = (mask & ~clear_mask) | inject_mask;
869 	WRITE_ONCE(p->health_bitmap_inject_mask, mask);
870 
871 	/* Invalidate cached health bitmap */
872 	p->lasthealth_jiffies = 0;
873 
874 	mutex_unlock(&p->health_mutex);
875 
876 	/* Return the supported flags back to userspace */
877 	payload->smart_inject.flags = supported_flags;
878 
879 	return sizeof(struct nd_papr_pdsm_health);
880 }
881 
882 /*
883  * 'struct pdsm_cmd_desc'
884  * Identifies supported PDSMs' expected length of in/out payloads
885  * and pdsm service function.
886  *
887  * size_in	: Size of input payload if any in the PDSM request.
888  * size_out	: Size of output payload if any in the PDSM request.
889  * service	: Service function for the PDSM request. Return semantics:
890  *		  rc < 0 : Error servicing PDSM and rc indicates the error.
891  *		  rc >=0 : Serviced successfully and 'rc' indicate number of
892  *			bytes written to payload.
893  */
894 struct pdsm_cmd_desc {
895 	u32 size_in;
896 	u32 size_out;
897 	int (*service)(struct papr_scm_priv *dimm,
898 		       union nd_pdsm_payload *payload);
899 };
900 
901 /* Holds all supported PDSMs' command descriptors */
902 static const struct pdsm_cmd_desc __pdsm_cmd_descriptors[] = {
903 	[PAPR_PDSM_MIN] = {
904 		.size_in = 0,
905 		.size_out = 0,
906 		.service = NULL,
907 	},
908 	/* New PDSM command descriptors to be added below */
909 
910 	[PAPR_PDSM_HEALTH] = {
911 		.size_in = 0,
912 		.size_out = sizeof(struct nd_papr_pdsm_health),
913 		.service = papr_pdsm_health,
914 	},
915 
916 	[PAPR_PDSM_SMART_INJECT] = {
917 		.size_in = sizeof(struct nd_papr_pdsm_smart_inject),
918 		.size_out = sizeof(struct nd_papr_pdsm_smart_inject),
919 		.service = papr_pdsm_smart_inject,
920 	},
921 	/* Empty */
922 	[PAPR_PDSM_MAX] = {
923 		.size_in = 0,
924 		.size_out = 0,
925 		.service = NULL,
926 	},
927 };
928 
929 /* Given a valid pdsm cmd return its command descriptor else return NULL */
pdsm_cmd_desc(enum papr_pdsm cmd)930 static inline const struct pdsm_cmd_desc *pdsm_cmd_desc(enum papr_pdsm cmd)
931 {
932 	if (cmd >= 0 || cmd < ARRAY_SIZE(__pdsm_cmd_descriptors))
933 		return &__pdsm_cmd_descriptors[cmd];
934 
935 	return NULL;
936 }
937 
938 /*
939  * For a given pdsm request call an appropriate service function.
940  * Returns errors if any while handling the pdsm command package.
941  */
papr_scm_service_pdsm(struct papr_scm_priv * p,struct nd_cmd_pkg * pkg)942 static int papr_scm_service_pdsm(struct papr_scm_priv *p,
943 				 struct nd_cmd_pkg *pkg)
944 {
945 	/* Get the PDSM header and PDSM command */
946 	struct nd_pkg_pdsm *pdsm_pkg = (struct nd_pkg_pdsm *)pkg->nd_payload;
947 	enum papr_pdsm pdsm = (enum papr_pdsm)pkg->nd_command;
948 	const struct pdsm_cmd_desc *pdsc;
949 	int rc;
950 
951 	/* Fetch corresponding pdsm descriptor for validation and servicing */
952 	pdsc = pdsm_cmd_desc(pdsm);
953 
954 	/* Validate pdsm descriptor */
955 	/* Ensure that reserved fields are 0 */
956 	if (pdsm_pkg->reserved[0] || pdsm_pkg->reserved[1]) {
957 		dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Invalid reserved field\n",
958 			pdsm);
959 		return -EINVAL;
960 	}
961 
962 	/* If pdsm expects some input, then ensure that the size_in matches */
963 	if (pdsc->size_in &&
964 	    pkg->nd_size_in != (pdsc->size_in + ND_PDSM_HDR_SIZE)) {
965 		dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Mismatched size_in=%d\n",
966 			pdsm, pkg->nd_size_in);
967 		return -EINVAL;
968 	}
969 
970 	/* If pdsm wants to return data, then ensure that  size_out matches */
971 	if (pdsc->size_out &&
972 	    pkg->nd_size_out != (pdsc->size_out + ND_PDSM_HDR_SIZE)) {
973 		dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Mismatched size_out=%d\n",
974 			pdsm, pkg->nd_size_out);
975 		return -EINVAL;
976 	}
977 
978 	/* Service the pdsm */
979 	if (pdsc->service) {
980 		dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Servicing..\n", pdsm);
981 
982 		rc = pdsc->service(p, &pdsm_pkg->payload);
983 
984 		if (rc < 0) {
985 			/* error encountered while servicing pdsm */
986 			pdsm_pkg->cmd_status = rc;
987 			pkg->nd_fw_size = ND_PDSM_HDR_SIZE;
988 		} else {
989 			/* pdsm serviced and 'rc' bytes written to payload */
990 			pdsm_pkg->cmd_status = 0;
991 			pkg->nd_fw_size = ND_PDSM_HDR_SIZE + rc;
992 		}
993 	} else {
994 		dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Unsupported PDSM request\n",
995 			pdsm);
996 		pdsm_pkg->cmd_status = -ENOENT;
997 		pkg->nd_fw_size = ND_PDSM_HDR_SIZE;
998 	}
999 
1000 	return pdsm_pkg->cmd_status;
1001 }
1002 
papr_scm_ndctl(struct nvdimm_bus_descriptor * nd_desc,struct nvdimm * nvdimm,unsigned int cmd,void * buf,unsigned int buf_len,int * cmd_rc)1003 static int papr_scm_ndctl(struct nvdimm_bus_descriptor *nd_desc,
1004 			  struct nvdimm *nvdimm, unsigned int cmd, void *buf,
1005 			  unsigned int buf_len, int *cmd_rc)
1006 {
1007 	struct nd_cmd_get_config_size *get_size_hdr;
1008 	struct nd_cmd_pkg *call_pkg = NULL;
1009 	struct papr_scm_priv *p;
1010 	int rc;
1011 
1012 	rc = is_cmd_valid(nvdimm, cmd, buf, buf_len);
1013 	if (rc) {
1014 		pr_debug("Invalid cmd=0x%x. Err=%d\n", cmd, rc);
1015 		return rc;
1016 	}
1017 
1018 	/* Use a local variable in case cmd_rc pointer is NULL */
1019 	if (!cmd_rc)
1020 		cmd_rc = &rc;
1021 
1022 	p = nvdimm_provider_data(nvdimm);
1023 
1024 	switch (cmd) {
1025 	case ND_CMD_GET_CONFIG_SIZE:
1026 		get_size_hdr = buf;
1027 
1028 		get_size_hdr->status = 0;
1029 		get_size_hdr->max_xfer = 8;
1030 		get_size_hdr->config_size = p->metadata_size;
1031 		*cmd_rc = 0;
1032 		break;
1033 
1034 	case ND_CMD_GET_CONFIG_DATA:
1035 		*cmd_rc = papr_scm_meta_get(p, buf);
1036 		break;
1037 
1038 	case ND_CMD_SET_CONFIG_DATA:
1039 		*cmd_rc = papr_scm_meta_set(p, buf);
1040 		break;
1041 
1042 	case ND_CMD_CALL:
1043 		call_pkg = (struct nd_cmd_pkg *)buf;
1044 		*cmd_rc = papr_scm_service_pdsm(p, call_pkg);
1045 		break;
1046 
1047 	default:
1048 		dev_dbg(&p->pdev->dev, "Unknown command = %d\n", cmd);
1049 		return -EINVAL;
1050 	}
1051 
1052 	dev_dbg(&p->pdev->dev, "returned with cmd_rc = %d\n", *cmd_rc);
1053 
1054 	return 0;
1055 }
1056 
health_bitmap_inject_show(struct device * dev,struct device_attribute * attr,char * buf)1057 static ssize_t health_bitmap_inject_show(struct device *dev,
1058 					 struct device_attribute *attr,
1059 					 char *buf)
1060 {
1061 	struct nvdimm *dimm = to_nvdimm(dev);
1062 	struct papr_scm_priv *p = nvdimm_provider_data(dimm);
1063 
1064 	return sprintf(buf, "%#llx\n",
1065 		       READ_ONCE(p->health_bitmap_inject_mask));
1066 }
1067 
1068 static DEVICE_ATTR_ADMIN_RO(health_bitmap_inject);
1069 
perf_stats_show(struct device * dev,struct device_attribute * attr,char * buf)1070 static ssize_t perf_stats_show(struct device *dev,
1071 			       struct device_attribute *attr, char *buf)
1072 {
1073 	int index;
1074 	ssize_t rc;
1075 	struct seq_buf s;
1076 	struct papr_scm_perf_stat *stat;
1077 	struct papr_scm_perf_stats *stats;
1078 	struct nvdimm *dimm = to_nvdimm(dev);
1079 	struct papr_scm_priv *p = nvdimm_provider_data(dimm);
1080 
1081 	if (!p->stat_buffer_len)
1082 		return -ENOENT;
1083 
1084 	/* Allocate the buffer for phyp where stats are written */
1085 	stats = kzalloc(p->stat_buffer_len, GFP_KERNEL);
1086 	if (!stats)
1087 		return -ENOMEM;
1088 
1089 	/* Ask phyp to return all dimm perf stats */
1090 	rc = drc_pmem_query_stats(p, stats, 0);
1091 	if (rc)
1092 		goto free_stats;
1093 	/*
1094 	 * Go through the returned output buffer and print stats and
1095 	 * values. Since stat_id is essentially a char string of
1096 	 * 8 bytes, simply use the string format specifier to print it.
1097 	 */
1098 	seq_buf_init(&s, buf, PAGE_SIZE);
1099 	for (index = 0, stat = stats->scm_statistic;
1100 	     index < be32_to_cpu(stats->num_statistics);
1101 	     ++index, ++stat) {
1102 		seq_buf_printf(&s, "%.8s = 0x%016llX\n",
1103 			       stat->stat_id,
1104 			       be64_to_cpu(stat->stat_val));
1105 	}
1106 
1107 free_stats:
1108 	kfree(stats);
1109 	return rc ? rc : (ssize_t)seq_buf_used(&s);
1110 }
1111 static DEVICE_ATTR_ADMIN_RO(perf_stats);
1112 
flags_show(struct device * dev,struct device_attribute * attr,char * buf)1113 static ssize_t flags_show(struct device *dev,
1114 			  struct device_attribute *attr, char *buf)
1115 {
1116 	struct nvdimm *dimm = to_nvdimm(dev);
1117 	struct papr_scm_priv *p = nvdimm_provider_data(dimm);
1118 	struct seq_buf s;
1119 	u64 health;
1120 	int rc;
1121 
1122 	rc = drc_pmem_query_health(p);
1123 	if (rc)
1124 		return rc;
1125 
1126 	/* Copy health_bitmap locally, check masks & update out buffer */
1127 	health = READ_ONCE(p->health_bitmap);
1128 
1129 	seq_buf_init(&s, buf, PAGE_SIZE);
1130 	if (health & PAPR_PMEM_UNARMED_MASK)
1131 		seq_buf_printf(&s, "not_armed ");
1132 
1133 	if (health & PAPR_PMEM_BAD_SHUTDOWN_MASK)
1134 		seq_buf_printf(&s, "flush_fail ");
1135 
1136 	if (health & PAPR_PMEM_BAD_RESTORE_MASK)
1137 		seq_buf_printf(&s, "restore_fail ");
1138 
1139 	if (health & PAPR_PMEM_ENCRYPTED)
1140 		seq_buf_printf(&s, "encrypted ");
1141 
1142 	if (health & PAPR_PMEM_SMART_EVENT_MASK)
1143 		seq_buf_printf(&s, "smart_notify ");
1144 
1145 	if (health & PAPR_PMEM_SCRUBBED_AND_LOCKED)
1146 		seq_buf_printf(&s, "scrubbed locked ");
1147 
1148 	if (seq_buf_used(&s))
1149 		seq_buf_printf(&s, "\n");
1150 
1151 	return seq_buf_used(&s);
1152 }
1153 DEVICE_ATTR_RO(flags);
1154 
dirty_shutdown_show(struct device * dev,struct device_attribute * attr,char * buf)1155 static ssize_t dirty_shutdown_show(struct device *dev,
1156 			  struct device_attribute *attr, char *buf)
1157 {
1158 	struct nvdimm *dimm = to_nvdimm(dev);
1159 	struct papr_scm_priv *p = nvdimm_provider_data(dimm);
1160 
1161 	return sysfs_emit(buf, "%llu\n", p->dirty_shutdown_counter);
1162 }
1163 DEVICE_ATTR_RO(dirty_shutdown);
1164 
papr_nd_attribute_visible(struct kobject * kobj,struct attribute * attr,int n)1165 static umode_t papr_nd_attribute_visible(struct kobject *kobj,
1166 					 struct attribute *attr, int n)
1167 {
1168 	struct device *dev = kobj_to_dev(kobj);
1169 	struct nvdimm *nvdimm = to_nvdimm(dev);
1170 	struct papr_scm_priv *p = nvdimm_provider_data(nvdimm);
1171 
1172 	/* For if perf-stats not available remove perf_stats sysfs */
1173 	if (attr == &dev_attr_perf_stats.attr && p->stat_buffer_len == 0)
1174 		return 0;
1175 
1176 	return attr->mode;
1177 }
1178 
1179 /* papr_scm specific dimm attributes */
1180 static struct attribute *papr_nd_attributes[] = {
1181 	&dev_attr_flags.attr,
1182 	&dev_attr_perf_stats.attr,
1183 	&dev_attr_dirty_shutdown.attr,
1184 	&dev_attr_health_bitmap_inject.attr,
1185 	NULL,
1186 };
1187 
1188 static const struct attribute_group papr_nd_attribute_group = {
1189 	.name = "papr",
1190 	.is_visible = papr_nd_attribute_visible,
1191 	.attrs = papr_nd_attributes,
1192 };
1193 
1194 static const struct attribute_group *papr_nd_attr_groups[] = {
1195 	&papr_nd_attribute_group,
1196 	NULL,
1197 };
1198 
papr_scm_nvdimm_init(struct papr_scm_priv * p)1199 static int papr_scm_nvdimm_init(struct papr_scm_priv *p)
1200 {
1201 	struct device *dev = &p->pdev->dev;
1202 	struct nd_mapping_desc mapping;
1203 	struct nd_region_desc ndr_desc;
1204 	unsigned long dimm_flags;
1205 	int target_nid, online_nid;
1206 
1207 	p->bus_desc.ndctl = papr_scm_ndctl;
1208 	p->bus_desc.module = THIS_MODULE;
1209 	p->bus_desc.of_node = p->pdev->dev.of_node;
1210 	p->bus_desc.provider_name = kstrdup(p->pdev->name, GFP_KERNEL);
1211 
1212 	/* Set the dimm command family mask to accept PDSMs */
1213 	set_bit(NVDIMM_FAMILY_PAPR, &p->bus_desc.dimm_family_mask);
1214 
1215 	if (!p->bus_desc.provider_name)
1216 		return -ENOMEM;
1217 
1218 	p->bus = nvdimm_bus_register(NULL, &p->bus_desc);
1219 	if (!p->bus) {
1220 		dev_err(dev, "Error creating nvdimm bus %pOF\n", p->dn);
1221 		kfree(p->bus_desc.provider_name);
1222 		return -ENXIO;
1223 	}
1224 
1225 	dimm_flags = 0;
1226 	set_bit(NDD_LABELING, &dimm_flags);
1227 
1228 	/*
1229 	 * Check if the nvdimm is unarmed. No locking needed as we are still
1230 	 * initializing. Ignore error encountered if any.
1231 	 */
1232 	__drc_pmem_query_health(p);
1233 
1234 	if (p->health_bitmap & PAPR_PMEM_UNARMED_MASK)
1235 		set_bit(NDD_UNARMED, &dimm_flags);
1236 
1237 	p->nvdimm = nvdimm_create(p->bus, p, papr_nd_attr_groups,
1238 				  dimm_flags, PAPR_SCM_DIMM_CMD_MASK, 0, NULL);
1239 	if (!p->nvdimm) {
1240 		dev_err(dev, "Error creating DIMM object for %pOF\n", p->dn);
1241 		goto err;
1242 	}
1243 
1244 	if (nvdimm_bus_check_dimm_count(p->bus, 1))
1245 		goto err;
1246 
1247 	/* now add the region */
1248 
1249 	memset(&mapping, 0, sizeof(mapping));
1250 	mapping.nvdimm = p->nvdimm;
1251 	mapping.start = 0;
1252 	mapping.size = p->blocks * p->block_size; // XXX: potential overflow?
1253 
1254 	memset(&ndr_desc, 0, sizeof(ndr_desc));
1255 	target_nid = dev_to_node(&p->pdev->dev);
1256 	online_nid = numa_map_to_online_node(target_nid);
1257 	ndr_desc.numa_node = online_nid;
1258 	ndr_desc.target_node = target_nid;
1259 	ndr_desc.res = &p->res;
1260 	ndr_desc.of_node = p->dn;
1261 	ndr_desc.provider_data = p;
1262 	ndr_desc.mapping = &mapping;
1263 	ndr_desc.num_mappings = 1;
1264 	ndr_desc.nd_set = &p->nd_set;
1265 
1266 	if (p->hcall_flush_required) {
1267 		set_bit(ND_REGION_ASYNC, &ndr_desc.flags);
1268 		ndr_desc.flush = papr_scm_pmem_flush;
1269 	}
1270 
1271 	if (p->is_volatile)
1272 		p->region = nvdimm_volatile_region_create(p->bus, &ndr_desc);
1273 	else {
1274 		set_bit(ND_REGION_PERSIST_MEMCTRL, &ndr_desc.flags);
1275 		p->region = nvdimm_pmem_region_create(p->bus, &ndr_desc);
1276 	}
1277 	if (!p->region) {
1278 		dev_err(dev, "Error registering region %pR from %pOF\n",
1279 				ndr_desc.res, p->dn);
1280 		goto err;
1281 	}
1282 	if (target_nid != online_nid)
1283 		dev_info(dev, "Region registered with target node %d and online node %d",
1284 			 target_nid, online_nid);
1285 
1286 	mutex_lock(&papr_ndr_lock);
1287 	list_add_tail(&p->region_list, &papr_nd_regions);
1288 	mutex_unlock(&papr_ndr_lock);
1289 
1290 	return 0;
1291 
1292 err:	nvdimm_bus_unregister(p->bus);
1293 	kfree(p->bus_desc.provider_name);
1294 	return -ENXIO;
1295 }
1296 
papr_scm_add_badblock(struct nd_region * region,struct nvdimm_bus * bus,u64 phys_addr)1297 static void papr_scm_add_badblock(struct nd_region *region,
1298 				  struct nvdimm_bus *bus, u64 phys_addr)
1299 {
1300 	u64 aligned_addr = ALIGN_DOWN(phys_addr, L1_CACHE_BYTES);
1301 
1302 	if (nvdimm_bus_add_badrange(bus, aligned_addr, L1_CACHE_BYTES)) {
1303 		pr_err("Bad block registration for 0x%llx failed\n", phys_addr);
1304 		return;
1305 	}
1306 
1307 	pr_debug("Add memory range (0x%llx - 0x%llx) as bad range\n",
1308 		 aligned_addr, aligned_addr + L1_CACHE_BYTES);
1309 
1310 	nvdimm_region_notify(region, NVDIMM_REVALIDATE_POISON);
1311 }
1312 
handle_mce_ue(struct notifier_block * nb,unsigned long val,void * data)1313 static int handle_mce_ue(struct notifier_block *nb, unsigned long val,
1314 			 void *data)
1315 {
1316 	struct machine_check_event *evt = data;
1317 	struct papr_scm_priv *p;
1318 	u64 phys_addr;
1319 	bool found = false;
1320 
1321 	if (evt->error_type != MCE_ERROR_TYPE_UE)
1322 		return NOTIFY_DONE;
1323 
1324 	if (list_empty(&papr_nd_regions))
1325 		return NOTIFY_DONE;
1326 
1327 	/*
1328 	 * The physical address obtained here is PAGE_SIZE aligned, so get the
1329 	 * exact address from the effective address
1330 	 */
1331 	phys_addr = evt->u.ue_error.physical_address +
1332 			(evt->u.ue_error.effective_address & ~PAGE_MASK);
1333 
1334 	if (!evt->u.ue_error.physical_address_provided ||
1335 	    !is_zone_device_page(pfn_to_page(phys_addr >> PAGE_SHIFT)))
1336 		return NOTIFY_DONE;
1337 
1338 	/* mce notifier is called from a process context, so mutex is safe */
1339 	mutex_lock(&papr_ndr_lock);
1340 	list_for_each_entry(p, &papr_nd_regions, region_list) {
1341 		if (phys_addr >= p->res.start && phys_addr <= p->res.end) {
1342 			found = true;
1343 			break;
1344 		}
1345 	}
1346 
1347 	if (found)
1348 		papr_scm_add_badblock(p->region, p->bus, phys_addr);
1349 
1350 	mutex_unlock(&papr_ndr_lock);
1351 
1352 	return found ? NOTIFY_OK : NOTIFY_DONE;
1353 }
1354 
1355 static struct notifier_block mce_ue_nb = {
1356 	.notifier_call = handle_mce_ue
1357 };
1358 
papr_scm_probe(struct platform_device * pdev)1359 static int papr_scm_probe(struct platform_device *pdev)
1360 {
1361 	struct device_node *dn = pdev->dev.of_node;
1362 	u32 drc_index, metadata_size;
1363 	u64 blocks, block_size;
1364 	struct papr_scm_priv *p;
1365 	u8 uuid_raw[UUID_SIZE];
1366 	const char *uuid_str;
1367 	ssize_t stat_size;
1368 	uuid_t uuid;
1369 	int rc;
1370 
1371 	/* check we have all the required DT properties */
1372 	if (of_property_read_u32(dn, "ibm,my-drc-index", &drc_index)) {
1373 		dev_err(&pdev->dev, "%pOF: missing drc-index!\n", dn);
1374 		return -ENODEV;
1375 	}
1376 
1377 	if (of_property_read_u64(dn, "ibm,block-size", &block_size)) {
1378 		dev_err(&pdev->dev, "%pOF: missing block-size!\n", dn);
1379 		return -ENODEV;
1380 	}
1381 
1382 	if (of_property_read_u64(dn, "ibm,number-of-blocks", &blocks)) {
1383 		dev_err(&pdev->dev, "%pOF: missing number-of-blocks!\n", dn);
1384 		return -ENODEV;
1385 	}
1386 
1387 	if (of_property_read_string(dn, "ibm,unit-guid", &uuid_str)) {
1388 		dev_err(&pdev->dev, "%pOF: missing unit-guid!\n", dn);
1389 		return -ENODEV;
1390 	}
1391 
1392 	/*
1393 	 * open firmware platform device create won't update the NUMA
1394 	 * distance table. For PAPR SCM devices we use numa_map_to_online_node()
1395 	 * to find the nearest online NUMA node and that requires correct
1396 	 * distance table information.
1397 	 */
1398 	update_numa_distance(dn);
1399 
1400 	p = kzalloc(sizeof(*p), GFP_KERNEL);
1401 	if (!p)
1402 		return -ENOMEM;
1403 
1404 	/* Initialize the dimm mutex */
1405 	mutex_init(&p->health_mutex);
1406 
1407 	/* optional DT properties */
1408 	of_property_read_u32(dn, "ibm,metadata-size", &metadata_size);
1409 
1410 	p->dn = dn;
1411 	p->drc_index = drc_index;
1412 	p->block_size = block_size;
1413 	p->blocks = blocks;
1414 	p->is_volatile = !of_property_read_bool(dn, "ibm,cache-flush-required");
1415 	p->hcall_flush_required = of_property_read_bool(dn, "ibm,hcall-flush-required");
1416 
1417 	if (of_property_read_u64(dn, "ibm,persistence-failed-count",
1418 				 &p->dirty_shutdown_counter))
1419 		p->dirty_shutdown_counter = 0;
1420 
1421 	/* We just need to ensure that set cookies are unique across */
1422 	uuid_parse(uuid_str, &uuid);
1423 
1424 	/*
1425 	 * The cookie1 and cookie2 are not really little endian.
1426 	 * We store a raw buffer representation of the
1427 	 * uuid string so that we can compare this with the label
1428 	 * area cookie irrespective of the endian configuration
1429 	 * with which the kernel is built.
1430 	 *
1431 	 * Historically we stored the cookie in the below format.
1432 	 * for a uuid string 72511b67-0b3b-42fd-8d1d-5be3cae8bcaa
1433 	 *	cookie1 was 0xfd423b0b671b5172
1434 	 *	cookie2 was 0xaabce8cae35b1d8d
1435 	 */
1436 	export_uuid(uuid_raw, &uuid);
1437 	p->nd_set.cookie1 = get_unaligned_le64(&uuid_raw[0]);
1438 	p->nd_set.cookie2 = get_unaligned_le64(&uuid_raw[8]);
1439 
1440 	/* might be zero */
1441 	p->metadata_size = metadata_size;
1442 	p->pdev = pdev;
1443 
1444 	/* request the hypervisor to bind this region to somewhere in memory */
1445 	rc = drc_pmem_bind(p);
1446 
1447 	/* If phyp says drc memory still bound then force unbound and retry */
1448 	if (rc == H_OVERLAP)
1449 		rc = drc_pmem_query_n_bind(p);
1450 
1451 	if (rc != H_SUCCESS) {
1452 		dev_err(&p->pdev->dev, "bind err: %d\n", rc);
1453 		rc = -ENXIO;
1454 		goto err;
1455 	}
1456 
1457 	/* setup the resource for the newly bound range */
1458 	p->res.start = p->bound_addr;
1459 	p->res.end   = p->bound_addr + p->blocks * p->block_size - 1;
1460 	p->res.name  = pdev->name;
1461 	p->res.flags = IORESOURCE_MEM;
1462 
1463 	/* Try retrieving the stat buffer and see if its supported */
1464 	stat_size = drc_pmem_query_stats(p, NULL, 0);
1465 	if (stat_size > 0) {
1466 		p->stat_buffer_len = stat_size;
1467 		dev_dbg(&p->pdev->dev, "Max perf-stat size %lu-bytes\n",
1468 			p->stat_buffer_len);
1469 	}
1470 
1471 	rc = papr_scm_nvdimm_init(p);
1472 	if (rc)
1473 		goto err2;
1474 
1475 	platform_set_drvdata(pdev, p);
1476 	papr_scm_pmu_register(p);
1477 
1478 	return 0;
1479 
1480 err2:	drc_pmem_unbind(p);
1481 err:	kfree(p);
1482 	return rc;
1483 }
1484 
papr_scm_remove(struct platform_device * pdev)1485 static void papr_scm_remove(struct platform_device *pdev)
1486 {
1487 	struct papr_scm_priv *p = platform_get_drvdata(pdev);
1488 
1489 	mutex_lock(&papr_ndr_lock);
1490 	list_del(&p->region_list);
1491 	mutex_unlock(&papr_ndr_lock);
1492 
1493 	nvdimm_bus_unregister(p->bus);
1494 	drc_pmem_unbind(p);
1495 
1496 	if (pdev->archdata.priv)
1497 		unregister_nvdimm_pmu(pdev->archdata.priv);
1498 
1499 	pdev->archdata.priv = NULL;
1500 	kfree(p->bus_desc.provider_name);
1501 	kfree(p);
1502 }
1503 
1504 static const struct of_device_id papr_scm_match[] = {
1505 	{ .compatible = "ibm,pmemory" },
1506 	{ .compatible = "ibm,pmemory-v2" },
1507 	{ },
1508 };
1509 
1510 static struct platform_driver papr_scm_driver = {
1511 	.probe = papr_scm_probe,
1512 	.remove = papr_scm_remove,
1513 	.driver = {
1514 		.name = "papr_scm",
1515 		.of_match_table = papr_scm_match,
1516 	},
1517 };
1518 
papr_scm_init(void)1519 static int __init papr_scm_init(void)
1520 {
1521 	int ret;
1522 
1523 	ret = platform_driver_register(&papr_scm_driver);
1524 	if (!ret)
1525 		mce_register_notifier(&mce_ue_nb);
1526 
1527 	return ret;
1528 }
1529 module_init(papr_scm_init);
1530 
papr_scm_exit(void)1531 static void __exit papr_scm_exit(void)
1532 {
1533 	mce_unregister_notifier(&mce_ue_nb);
1534 	platform_driver_unregister(&papr_scm_driver);
1535 }
1536 module_exit(papr_scm_exit);
1537 
1538 MODULE_DEVICE_TABLE(of, papr_scm_match);
1539 MODULE_DESCRIPTION("PAPR Storage Class Memory interface driver");
1540 MODULE_LICENSE("GPL");
1541 MODULE_AUTHOR("IBM Corporation");
1542