powerpc/papr_scm: Fetch nvdimm performance stats from PHYP
[linux-2.6-microblaze.git] / arch / powerpc / platforms / pseries / papr_scm.c
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
22 #define BIND_ANY_ADDR (~0ul)
23
24 #define PAPR_SCM_DIMM_CMD_MASK \
25         ((1ul << ND_CMD_GET_CONFIG_SIZE) | \
26          (1ul << ND_CMD_GET_CONFIG_DATA) | \
27          (1ul << ND_CMD_SET_CONFIG_DATA) | \
28          (1ul << ND_CMD_CALL))
29
30 /* DIMM health bitmap bitmap indicators */
31 /* SCM device is unable to persist memory contents */
32 #define PAPR_PMEM_UNARMED                   (1ULL << (63 - 0))
33 /* SCM device failed to persist memory contents */
34 #define PAPR_PMEM_SHUTDOWN_DIRTY            (1ULL << (63 - 1))
35 /* SCM device contents are persisted from previous IPL */
36 #define PAPR_PMEM_SHUTDOWN_CLEAN            (1ULL << (63 - 2))
37 /* SCM device contents are not persisted from previous IPL */
38 #define PAPR_PMEM_EMPTY                     (1ULL << (63 - 3))
39 /* SCM device memory life remaining is critically low */
40 #define PAPR_PMEM_HEALTH_CRITICAL           (1ULL << (63 - 4))
41 /* SCM device will be garded off next IPL due to failure */
42 #define PAPR_PMEM_HEALTH_FATAL              (1ULL << (63 - 5))
43 /* SCM contents cannot persist due to current platform health status */
44 #define PAPR_PMEM_HEALTH_UNHEALTHY          (1ULL << (63 - 6))
45 /* SCM device is unable to persist memory contents in certain conditions */
46 #define PAPR_PMEM_HEALTH_NON_CRITICAL       (1ULL << (63 - 7))
47 /* SCM device is encrypted */
48 #define PAPR_PMEM_ENCRYPTED                 (1ULL << (63 - 8))
49 /* SCM device has been scrubbed and locked */
50 #define PAPR_PMEM_SCRUBBED_AND_LOCKED       (1ULL << (63 - 9))
51
52 /* Bits status indicators for health bitmap indicating unarmed dimm */
53 #define PAPR_PMEM_UNARMED_MASK (PAPR_PMEM_UNARMED |             \
54                                 PAPR_PMEM_HEALTH_UNHEALTHY)
55
56 /* Bits status indicators for health bitmap indicating unflushed dimm */
57 #define PAPR_PMEM_BAD_SHUTDOWN_MASK (PAPR_PMEM_SHUTDOWN_DIRTY)
58
59 /* Bits status indicators for health bitmap indicating unrestored dimm */
60 #define PAPR_PMEM_BAD_RESTORE_MASK  (PAPR_PMEM_EMPTY)
61
62 /* Bit status indicators for smart event notification */
63 #define PAPR_PMEM_SMART_EVENT_MASK (PAPR_PMEM_HEALTH_CRITICAL | \
64                                     PAPR_PMEM_HEALTH_FATAL |    \
65                                     PAPR_PMEM_HEALTH_UNHEALTHY)
66
67 #define PAPR_SCM_PERF_STATS_EYECATCHER __stringify(SCMSTATS)
68 #define PAPR_SCM_PERF_STATS_VERSION 0x1
69
70 /* Struct holding a single performance metric */
71 struct papr_scm_perf_stat {
72         u8 stat_id[8];
73         __be64 stat_val;
74 } __packed;
75
76 /* Struct exchanged between kernel and PHYP for fetching drc perf stats */
77 struct papr_scm_perf_stats {
78         u8 eye_catcher[8];
79         /* Should be PAPR_SCM_PERF_STATS_VERSION */
80         __be32 stats_version;
81         /* Number of stats following */
82         __be32 num_statistics;
83         /* zero or more performance matrics */
84         struct papr_scm_perf_stat scm_statistic[];
85 } __packed;
86
87 /* private struct associated with each region */
88 struct papr_scm_priv {
89         struct platform_device *pdev;
90         struct device_node *dn;
91         uint32_t drc_index;
92         uint64_t blocks;
93         uint64_t block_size;
94         int metadata_size;
95         bool is_volatile;
96
97         uint64_t bound_addr;
98
99         struct nvdimm_bus_descriptor bus_desc;
100         struct nvdimm_bus *bus;
101         struct nvdimm *nvdimm;
102         struct resource res;
103         struct nd_region *region;
104         struct nd_interleave_set nd_set;
105         struct list_head region_list;
106
107         /* Protect dimm health data from concurrent read/writes */
108         struct mutex health_mutex;
109
110         /* Last time the health information of the dimm was updated */
111         unsigned long lasthealth_jiffies;
112
113         /* Health information for the dimm */
114         u64 health_bitmap;
115
116         /* length of the stat buffer as expected by phyp */
117         size_t stat_buffer_len;
118 };
119
120 static LIST_HEAD(papr_nd_regions);
121 static DEFINE_MUTEX(papr_ndr_lock);
122
123 static int drc_pmem_bind(struct papr_scm_priv *p)
124 {
125         unsigned long ret[PLPAR_HCALL_BUFSIZE];
126         uint64_t saved = 0;
127         uint64_t token;
128         int64_t rc;
129
130         /*
131          * When the hypervisor cannot map all the requested memory in a single
132          * hcall it returns H_BUSY and we call again with the token until
133          * we get H_SUCCESS. Aborting the retry loop before getting H_SUCCESS
134          * leave the system in an undefined state, so we wait.
135          */
136         token = 0;
137
138         do {
139                 rc = plpar_hcall(H_SCM_BIND_MEM, ret, p->drc_index, 0,
140                                 p->blocks, BIND_ANY_ADDR, token);
141                 token = ret[0];
142                 if (!saved)
143                         saved = ret[1];
144                 cond_resched();
145         } while (rc == H_BUSY);
146
147         if (rc)
148                 return rc;
149
150         p->bound_addr = saved;
151         dev_dbg(&p->pdev->dev, "bound drc 0x%x to 0x%lx\n",
152                 p->drc_index, (unsigned long)saved);
153         return rc;
154 }
155
156 static void drc_pmem_unbind(struct papr_scm_priv *p)
157 {
158         unsigned long ret[PLPAR_HCALL_BUFSIZE];
159         uint64_t token = 0;
160         int64_t rc;
161
162         dev_dbg(&p->pdev->dev, "unbind drc 0x%x\n", p->drc_index);
163
164         /* NB: unbind has the same retry requirements as drc_pmem_bind() */
165         do {
166
167                 /* Unbind of all SCM resources associated with drcIndex */
168                 rc = plpar_hcall(H_SCM_UNBIND_ALL, ret, H_UNBIND_SCOPE_DRC,
169                                  p->drc_index, token);
170                 token = ret[0];
171
172                 /* Check if we are stalled for some time */
173                 if (H_IS_LONG_BUSY(rc)) {
174                         msleep(get_longbusy_msecs(rc));
175                         rc = H_BUSY;
176                 } else if (rc == H_BUSY) {
177                         cond_resched();
178                 }
179
180         } while (rc == H_BUSY);
181
182         if (rc)
183                 dev_err(&p->pdev->dev, "unbind error: %lld\n", rc);
184         else
185                 dev_dbg(&p->pdev->dev, "unbind drc 0x%x complete\n",
186                         p->drc_index);
187
188         return;
189 }
190
191 static int drc_pmem_query_n_bind(struct papr_scm_priv *p)
192 {
193         unsigned long start_addr;
194         unsigned long end_addr;
195         unsigned long ret[PLPAR_HCALL_BUFSIZE];
196         int64_t rc;
197
198
199         rc = plpar_hcall(H_SCM_QUERY_BLOCK_MEM_BINDING, ret,
200                          p->drc_index, 0);
201         if (rc)
202                 goto err_out;
203         start_addr = ret[0];
204
205         /* Make sure the full region is bound. */
206         rc = plpar_hcall(H_SCM_QUERY_BLOCK_MEM_BINDING, ret,
207                          p->drc_index, p->blocks - 1);
208         if (rc)
209                 goto err_out;
210         end_addr = ret[0];
211
212         if ((end_addr - start_addr) != ((p->blocks - 1) * p->block_size))
213                 goto err_out;
214
215         p->bound_addr = start_addr;
216         dev_dbg(&p->pdev->dev, "bound drc 0x%x to 0x%lx\n", p->drc_index, start_addr);
217         return rc;
218
219 err_out:
220         dev_info(&p->pdev->dev,
221                  "Failed to query, trying an unbind followed by bind");
222         drc_pmem_unbind(p);
223         return drc_pmem_bind(p);
224 }
225
226 /*
227  * Query the Dimm performance stats from PHYP and copy them (if returned) to
228  * provided struct papr_scm_perf_stats instance 'stats' that can hold atleast
229  * (num_stats + header) bytes.
230  * - If buff_stats == NULL the return value is the size in byes of the buffer
231  * needed to hold all supported performance-statistics.
232  * - If buff_stats != NULL and num_stats == 0 then we copy all known
233  * performance-statistics to 'buff_stat' and expect to be large enough to
234  * hold them.
235  * - if buff_stats != NULL and num_stats > 0 then copy the requested
236  * performance-statistics to buff_stats.
237  */
238 static ssize_t drc_pmem_query_stats(struct papr_scm_priv *p,
239                                     struct papr_scm_perf_stats *buff_stats,
240                                     unsigned int num_stats)
241 {
242         unsigned long ret[PLPAR_HCALL_BUFSIZE];
243         size_t size;
244         s64 rc;
245
246         /* Setup the out buffer */
247         if (buff_stats) {
248                 memcpy(buff_stats->eye_catcher,
249                        PAPR_SCM_PERF_STATS_EYECATCHER, 8);
250                 buff_stats->stats_version =
251                         cpu_to_be32(PAPR_SCM_PERF_STATS_VERSION);
252                 buff_stats->num_statistics =
253                         cpu_to_be32(num_stats);
254
255                 /*
256                  * Calculate the buffer size based on num-stats provided
257                  * or use the prefetched max buffer length
258                  */
259                 if (num_stats)
260                         /* Calculate size from the num_stats */
261                         size = sizeof(struct papr_scm_perf_stats) +
262                                 num_stats * sizeof(struct papr_scm_perf_stat);
263                 else
264                         size = p->stat_buffer_len;
265         } else {
266                 /* In case of no out buffer ignore the size */
267                 size = 0;
268         }
269
270         /* Do the HCALL asking PHYP for info */
271         rc = plpar_hcall(H_SCM_PERFORMANCE_STATS, ret, p->drc_index,
272                          buff_stats ? virt_to_phys(buff_stats) : 0,
273                          size);
274
275         /* Check if the error was due to an unknown stat-id */
276         if (rc == H_PARTIAL) {
277                 dev_err(&p->pdev->dev,
278                         "Unknown performance stats, Err:0x%016lX\n", ret[0]);
279                 return -ENOENT;
280         } else if (rc != H_SUCCESS) {
281                 dev_err(&p->pdev->dev,
282                         "Failed to query performance stats, Err:%lld\n", rc);
283                 return -EIO;
284
285         } else if (!size) {
286                 /* Handle case where stat buffer size was requested */
287                 dev_dbg(&p->pdev->dev,
288                         "Performance stats size %ld\n", ret[0]);
289                 return ret[0];
290         }
291
292         /* Successfully fetched the requested stats from phyp */
293         dev_dbg(&p->pdev->dev,
294                 "Performance stats returned %d stats\n",
295                 be32_to_cpu(buff_stats->num_statistics));
296         return 0;
297 }
298
299 /*
300  * Issue hcall to retrieve dimm health info and populate papr_scm_priv with the
301  * health information.
302  */
303 static int __drc_pmem_query_health(struct papr_scm_priv *p)
304 {
305         unsigned long ret[PLPAR_HCALL_BUFSIZE];
306         long rc;
307
308         /* issue the hcall */
309         rc = plpar_hcall(H_SCM_HEALTH, ret, p->drc_index);
310         if (rc != H_SUCCESS) {
311                 dev_err(&p->pdev->dev,
312                         "Failed to query health information, Err:%ld\n", rc);
313                 return -ENXIO;
314         }
315
316         p->lasthealth_jiffies = jiffies;
317         p->health_bitmap = ret[0] & ret[1];
318
319         dev_dbg(&p->pdev->dev,
320                 "Queried dimm health info. Bitmap:0x%016lx Mask:0x%016lx\n",
321                 ret[0], ret[1]);
322
323         return 0;
324 }
325
326 /* Min interval in seconds for assuming stable dimm health */
327 #define MIN_HEALTH_QUERY_INTERVAL 60
328
329 /* Query cached health info and if needed call drc_pmem_query_health */
330 static int drc_pmem_query_health(struct papr_scm_priv *p)
331 {
332         unsigned long cache_timeout;
333         int rc;
334
335         /* Protect concurrent modifications to papr_scm_priv */
336         rc = mutex_lock_interruptible(&p->health_mutex);
337         if (rc)
338                 return rc;
339
340         /* Jiffies offset for which the health data is assumed to be same */
341         cache_timeout = p->lasthealth_jiffies +
342                 msecs_to_jiffies(MIN_HEALTH_QUERY_INTERVAL * 1000);
343
344         /* Fetch new health info is its older than MIN_HEALTH_QUERY_INTERVAL */
345         if (time_after(jiffies, cache_timeout))
346                 rc = __drc_pmem_query_health(p);
347         else
348                 /* Assume cached health data is valid */
349                 rc = 0;
350
351         mutex_unlock(&p->health_mutex);
352         return rc;
353 }
354
355 static int papr_scm_meta_get(struct papr_scm_priv *p,
356                              struct nd_cmd_get_config_data_hdr *hdr)
357 {
358         unsigned long data[PLPAR_HCALL_BUFSIZE];
359         unsigned long offset, data_offset;
360         int len, read;
361         int64_t ret;
362
363         if ((hdr->in_offset + hdr->in_length) > p->metadata_size)
364                 return -EINVAL;
365
366         for (len = hdr->in_length; len; len -= read) {
367
368                 data_offset = hdr->in_length - len;
369                 offset = hdr->in_offset + data_offset;
370
371                 if (len >= 8)
372                         read = 8;
373                 else if (len >= 4)
374                         read = 4;
375                 else if (len >= 2)
376                         read = 2;
377                 else
378                         read = 1;
379
380                 ret = plpar_hcall(H_SCM_READ_METADATA, data, p->drc_index,
381                                   offset, read);
382
383                 if (ret == H_PARAMETER) /* bad DRC index */
384                         return -ENODEV;
385                 if (ret)
386                         return -EINVAL; /* other invalid parameter */
387
388                 switch (read) {
389                 case 8:
390                         *(uint64_t *)(hdr->out_buf + data_offset) = be64_to_cpu(data[0]);
391                         break;
392                 case 4:
393                         *(uint32_t *)(hdr->out_buf + data_offset) = be32_to_cpu(data[0] & 0xffffffff);
394                         break;
395
396                 case 2:
397                         *(uint16_t *)(hdr->out_buf + data_offset) = be16_to_cpu(data[0] & 0xffff);
398                         break;
399
400                 case 1:
401                         *(uint8_t *)(hdr->out_buf + data_offset) = (data[0] & 0xff);
402                         break;
403                 }
404         }
405         return 0;
406 }
407
408 static int papr_scm_meta_set(struct papr_scm_priv *p,
409                              struct nd_cmd_set_config_hdr *hdr)
410 {
411         unsigned long offset, data_offset;
412         int len, wrote;
413         unsigned long data;
414         __be64 data_be;
415         int64_t ret;
416
417         if ((hdr->in_offset + hdr->in_length) > p->metadata_size)
418                 return -EINVAL;
419
420         for (len = hdr->in_length; len; len -= wrote) {
421
422                 data_offset = hdr->in_length - len;
423                 offset = hdr->in_offset + data_offset;
424
425                 if (len >= 8) {
426                         data = *(uint64_t *)(hdr->in_buf + data_offset);
427                         data_be = cpu_to_be64(data);
428                         wrote = 8;
429                 } else if (len >= 4) {
430                         data = *(uint32_t *)(hdr->in_buf + data_offset);
431                         data &= 0xffffffff;
432                         data_be = cpu_to_be32(data);
433                         wrote = 4;
434                 } else if (len >= 2) {
435                         data = *(uint16_t *)(hdr->in_buf + data_offset);
436                         data &= 0xffff;
437                         data_be = cpu_to_be16(data);
438                         wrote = 2;
439                 } else {
440                         data_be = *(uint8_t *)(hdr->in_buf + data_offset);
441                         data_be &= 0xff;
442                         wrote = 1;
443                 }
444
445                 ret = plpar_hcall_norets(H_SCM_WRITE_METADATA, p->drc_index,
446                                          offset, data_be, wrote);
447                 if (ret == H_PARAMETER) /* bad DRC index */
448                         return -ENODEV;
449                 if (ret)
450                         return -EINVAL; /* other invalid parameter */
451         }
452
453         return 0;
454 }
455
456 /*
457  * Do a sanity checks on the inputs args to dimm-control function and return
458  * '0' if valid. Validation of PDSM payloads happens later in
459  * papr_scm_service_pdsm.
460  */
461 static int is_cmd_valid(struct nvdimm *nvdimm, unsigned int cmd, void *buf,
462                         unsigned int buf_len)
463 {
464         unsigned long cmd_mask = PAPR_SCM_DIMM_CMD_MASK;
465         struct nd_cmd_pkg *nd_cmd;
466         struct papr_scm_priv *p;
467         enum papr_pdsm pdsm;
468
469         /* Only dimm-specific calls are supported atm */
470         if (!nvdimm)
471                 return -EINVAL;
472
473         /* get the provider data from struct nvdimm */
474         p = nvdimm_provider_data(nvdimm);
475
476         if (!test_bit(cmd, &cmd_mask)) {
477                 dev_dbg(&p->pdev->dev, "Unsupported cmd=%u\n", cmd);
478                 return -EINVAL;
479         }
480
481         /* For CMD_CALL verify pdsm request */
482         if (cmd == ND_CMD_CALL) {
483                 /* Verify the envelope and envelop size */
484                 if (!buf ||
485                     buf_len < (sizeof(struct nd_cmd_pkg) + ND_PDSM_HDR_SIZE)) {
486                         dev_dbg(&p->pdev->dev, "Invalid pkg size=%u\n",
487                                 buf_len);
488                         return -EINVAL;
489                 }
490
491                 /* Verify that the nd_cmd_pkg.nd_family is correct */
492                 nd_cmd = (struct nd_cmd_pkg *)buf;
493
494                 if (nd_cmd->nd_family != NVDIMM_FAMILY_PAPR) {
495                         dev_dbg(&p->pdev->dev, "Invalid pkg family=0x%llx\n",
496                                 nd_cmd->nd_family);
497                         return -EINVAL;
498                 }
499
500                 pdsm = (enum papr_pdsm)nd_cmd->nd_command;
501
502                 /* Verify if the pdsm command is valid */
503                 if (pdsm <= PAPR_PDSM_MIN || pdsm >= PAPR_PDSM_MAX) {
504                         dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Invalid PDSM\n",
505                                 pdsm);
506                         return -EINVAL;
507                 }
508
509                 /* Have enough space to hold returned 'nd_pkg_pdsm' header */
510                 if (nd_cmd->nd_size_out < ND_PDSM_HDR_SIZE) {
511                         dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Invalid payload\n",
512                                 pdsm);
513                         return -EINVAL;
514                 }
515         }
516
517         /* Let the command be further processed */
518         return 0;
519 }
520
521 /* Fetch the DIMM health info and populate it in provided package. */
522 static int papr_pdsm_health(struct papr_scm_priv *p,
523                             union nd_pdsm_payload *payload)
524 {
525         int rc;
526
527         /* Ensure dimm health mutex is taken preventing concurrent access */
528         rc = mutex_lock_interruptible(&p->health_mutex);
529         if (rc)
530                 goto out;
531
532         /* Always fetch upto date dimm health data ignoring cached values */
533         rc = __drc_pmem_query_health(p);
534         if (rc) {
535                 mutex_unlock(&p->health_mutex);
536                 goto out;
537         }
538
539         /* update health struct with various flags derived from health bitmap */
540         payload->health = (struct nd_papr_pdsm_health) {
541                 .extension_flags = 0,
542                 .dimm_unarmed = !!(p->health_bitmap & PAPR_PMEM_UNARMED_MASK),
543                 .dimm_bad_shutdown = !!(p->health_bitmap & PAPR_PMEM_BAD_SHUTDOWN_MASK),
544                 .dimm_bad_restore = !!(p->health_bitmap & PAPR_PMEM_BAD_RESTORE_MASK),
545                 .dimm_scrubbed = !!(p->health_bitmap & PAPR_PMEM_SCRUBBED_AND_LOCKED),
546                 .dimm_locked = !!(p->health_bitmap & PAPR_PMEM_SCRUBBED_AND_LOCKED),
547                 .dimm_encrypted = !!(p->health_bitmap & PAPR_PMEM_ENCRYPTED),
548                 .dimm_health = PAPR_PDSM_DIMM_HEALTHY,
549         };
550
551         /* Update field dimm_health based on health_bitmap flags */
552         if (p->health_bitmap & PAPR_PMEM_HEALTH_FATAL)
553                 payload->health.dimm_health = PAPR_PDSM_DIMM_FATAL;
554         else if (p->health_bitmap & PAPR_PMEM_HEALTH_CRITICAL)
555                 payload->health.dimm_health = PAPR_PDSM_DIMM_CRITICAL;
556         else if (p->health_bitmap & PAPR_PMEM_HEALTH_UNHEALTHY)
557                 payload->health.dimm_health = PAPR_PDSM_DIMM_UNHEALTHY;
558
559         /* struct populated hence can release the mutex now */
560         mutex_unlock(&p->health_mutex);
561         rc = sizeof(struct nd_papr_pdsm_health);
562
563 out:
564         return rc;
565 }
566
567 /*
568  * 'struct pdsm_cmd_desc'
569  * Identifies supported PDSMs' expected length of in/out payloads
570  * and pdsm service function.
571  *
572  * size_in      : Size of input payload if any in the PDSM request.
573  * size_out     : Size of output payload if any in the PDSM request.
574  * service      : Service function for the PDSM request. Return semantics:
575  *                rc < 0 : Error servicing PDSM and rc indicates the error.
576  *                rc >=0 : Serviced successfully and 'rc' indicate number of
577  *                      bytes written to payload.
578  */
579 struct pdsm_cmd_desc {
580         u32 size_in;
581         u32 size_out;
582         int (*service)(struct papr_scm_priv *dimm,
583                        union nd_pdsm_payload *payload);
584 };
585
586 /* Holds all supported PDSMs' command descriptors */
587 static const struct pdsm_cmd_desc __pdsm_cmd_descriptors[] = {
588         [PAPR_PDSM_MIN] = {
589                 .size_in = 0,
590                 .size_out = 0,
591                 .service = NULL,
592         },
593         /* New PDSM command descriptors to be added below */
594
595         [PAPR_PDSM_HEALTH] = {
596                 .size_in = 0,
597                 .size_out = sizeof(struct nd_papr_pdsm_health),
598                 .service = papr_pdsm_health,
599         },
600         /* Empty */
601         [PAPR_PDSM_MAX] = {
602                 .size_in = 0,
603                 .size_out = 0,
604                 .service = NULL,
605         },
606 };
607
608 /* Given a valid pdsm cmd return its command descriptor else return NULL */
609 static inline const struct pdsm_cmd_desc *pdsm_cmd_desc(enum papr_pdsm cmd)
610 {
611         if (cmd >= 0 || cmd < ARRAY_SIZE(__pdsm_cmd_descriptors))
612                 return &__pdsm_cmd_descriptors[cmd];
613
614         return NULL;
615 }
616
617 /*
618  * For a given pdsm request call an appropriate service function.
619  * Returns errors if any while handling the pdsm command package.
620  */
621 static int papr_scm_service_pdsm(struct papr_scm_priv *p,
622                                  struct nd_cmd_pkg *pkg)
623 {
624         /* Get the PDSM header and PDSM command */
625         struct nd_pkg_pdsm *pdsm_pkg = (struct nd_pkg_pdsm *)pkg->nd_payload;
626         enum papr_pdsm pdsm = (enum papr_pdsm)pkg->nd_command;
627         const struct pdsm_cmd_desc *pdsc;
628         int rc;
629
630         /* Fetch corresponding pdsm descriptor for validation and servicing */
631         pdsc = pdsm_cmd_desc(pdsm);
632
633         /* Validate pdsm descriptor */
634         /* Ensure that reserved fields are 0 */
635         if (pdsm_pkg->reserved[0] || pdsm_pkg->reserved[1]) {
636                 dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Invalid reserved field\n",
637                         pdsm);
638                 return -EINVAL;
639         }
640
641         /* If pdsm expects some input, then ensure that the size_in matches */
642         if (pdsc->size_in &&
643             pkg->nd_size_in != (pdsc->size_in + ND_PDSM_HDR_SIZE)) {
644                 dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Mismatched size_in=%d\n",
645                         pdsm, pkg->nd_size_in);
646                 return -EINVAL;
647         }
648
649         /* If pdsm wants to return data, then ensure that  size_out matches */
650         if (pdsc->size_out &&
651             pkg->nd_size_out != (pdsc->size_out + ND_PDSM_HDR_SIZE)) {
652                 dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Mismatched size_out=%d\n",
653                         pdsm, pkg->nd_size_out);
654                 return -EINVAL;
655         }
656
657         /* Service the pdsm */
658         if (pdsc->service) {
659                 dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Servicing..\n", pdsm);
660
661                 rc = pdsc->service(p, &pdsm_pkg->payload);
662
663                 if (rc < 0) {
664                         /* error encountered while servicing pdsm */
665                         pdsm_pkg->cmd_status = rc;
666                         pkg->nd_fw_size = ND_PDSM_HDR_SIZE;
667                 } else {
668                         /* pdsm serviced and 'rc' bytes written to payload */
669                         pdsm_pkg->cmd_status = 0;
670                         pkg->nd_fw_size = ND_PDSM_HDR_SIZE + rc;
671                 }
672         } else {
673                 dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Unsupported PDSM request\n",
674                         pdsm);
675                 pdsm_pkg->cmd_status = -ENOENT;
676                 pkg->nd_fw_size = ND_PDSM_HDR_SIZE;
677         }
678
679         return pdsm_pkg->cmd_status;
680 }
681
682 static int papr_scm_ndctl(struct nvdimm_bus_descriptor *nd_desc,
683                           struct nvdimm *nvdimm, unsigned int cmd, void *buf,
684                           unsigned int buf_len, int *cmd_rc)
685 {
686         struct nd_cmd_get_config_size *get_size_hdr;
687         struct nd_cmd_pkg *call_pkg = NULL;
688         struct papr_scm_priv *p;
689         int rc;
690
691         rc = is_cmd_valid(nvdimm, cmd, buf, buf_len);
692         if (rc) {
693                 pr_debug("Invalid cmd=0x%x. Err=%d\n", cmd, rc);
694                 return rc;
695         }
696
697         /* Use a local variable in case cmd_rc pointer is NULL */
698         if (!cmd_rc)
699                 cmd_rc = &rc;
700
701         p = nvdimm_provider_data(nvdimm);
702
703         switch (cmd) {
704         case ND_CMD_GET_CONFIG_SIZE:
705                 get_size_hdr = buf;
706
707                 get_size_hdr->status = 0;
708                 get_size_hdr->max_xfer = 8;
709                 get_size_hdr->config_size = p->metadata_size;
710                 *cmd_rc = 0;
711                 break;
712
713         case ND_CMD_GET_CONFIG_DATA:
714                 *cmd_rc = papr_scm_meta_get(p, buf);
715                 break;
716
717         case ND_CMD_SET_CONFIG_DATA:
718                 *cmd_rc = papr_scm_meta_set(p, buf);
719                 break;
720
721         case ND_CMD_CALL:
722                 call_pkg = (struct nd_cmd_pkg *)buf;
723                 *cmd_rc = papr_scm_service_pdsm(p, call_pkg);
724                 break;
725
726         default:
727                 dev_dbg(&p->pdev->dev, "Unknown command = %d\n", cmd);
728                 return -EINVAL;
729         }
730
731         dev_dbg(&p->pdev->dev, "returned with cmd_rc = %d\n", *cmd_rc);
732
733         return 0;
734 }
735
736 static ssize_t perf_stats_show(struct device *dev,
737                                struct device_attribute *attr, char *buf)
738 {
739         int index, rc;
740         struct seq_buf s;
741         struct papr_scm_perf_stat *stat;
742         struct papr_scm_perf_stats *stats;
743         struct nvdimm *dimm = to_nvdimm(dev);
744         struct papr_scm_priv *p = nvdimm_provider_data(dimm);
745
746         if (!p->stat_buffer_len)
747                 return -ENOENT;
748
749         /* Allocate the buffer for phyp where stats are written */
750         stats = kzalloc(p->stat_buffer_len, GFP_KERNEL);
751         if (!stats)
752                 return -ENOMEM;
753
754         /* Ask phyp to return all dimm perf stats */
755         rc = drc_pmem_query_stats(p, stats, 0);
756         if (rc)
757                 goto free_stats;
758         /*
759          * Go through the returned output buffer and print stats and
760          * values. Since stat_id is essentially a char string of
761          * 8 bytes, simply use the string format specifier to print it.
762          */
763         seq_buf_init(&s, buf, PAGE_SIZE);
764         for (index = 0, stat = stats->scm_statistic;
765              index < be32_to_cpu(stats->num_statistics);
766              ++index, ++stat) {
767                 seq_buf_printf(&s, "%.8s = 0x%016llX\n",
768                                stat->stat_id,
769                                be64_to_cpu(stat->stat_val));
770         }
771
772 free_stats:
773         kfree(stats);
774         return rc ? rc : seq_buf_used(&s);
775 }
776 DEVICE_ATTR_RO(perf_stats);
777
778 static ssize_t flags_show(struct device *dev,
779                           struct device_attribute *attr, char *buf)
780 {
781         struct nvdimm *dimm = to_nvdimm(dev);
782         struct papr_scm_priv *p = nvdimm_provider_data(dimm);
783         struct seq_buf s;
784         u64 health;
785         int rc;
786
787         rc = drc_pmem_query_health(p);
788         if (rc)
789                 return rc;
790
791         /* Copy health_bitmap locally, check masks & update out buffer */
792         health = READ_ONCE(p->health_bitmap);
793
794         seq_buf_init(&s, buf, PAGE_SIZE);
795         if (health & PAPR_PMEM_UNARMED_MASK)
796                 seq_buf_printf(&s, "not_armed ");
797
798         if (health & PAPR_PMEM_BAD_SHUTDOWN_MASK)
799                 seq_buf_printf(&s, "flush_fail ");
800
801         if (health & PAPR_PMEM_BAD_RESTORE_MASK)
802                 seq_buf_printf(&s, "restore_fail ");
803
804         if (health & PAPR_PMEM_ENCRYPTED)
805                 seq_buf_printf(&s, "encrypted ");
806
807         if (health & PAPR_PMEM_SMART_EVENT_MASK)
808                 seq_buf_printf(&s, "smart_notify ");
809
810         if (health & PAPR_PMEM_SCRUBBED_AND_LOCKED)
811                 seq_buf_printf(&s, "scrubbed locked ");
812
813         if (seq_buf_used(&s))
814                 seq_buf_printf(&s, "\n");
815
816         return seq_buf_used(&s);
817 }
818 DEVICE_ATTR_RO(flags);
819
820 /* papr_scm specific dimm attributes */
821 static struct attribute *papr_nd_attributes[] = {
822         &dev_attr_flags.attr,
823         &dev_attr_perf_stats.attr,
824         NULL,
825 };
826
827 static struct attribute_group papr_nd_attribute_group = {
828         .name = "papr",
829         .attrs = papr_nd_attributes,
830 };
831
832 static const struct attribute_group *papr_nd_attr_groups[] = {
833         &papr_nd_attribute_group,
834         NULL,
835 };
836
837 static int papr_scm_nvdimm_init(struct papr_scm_priv *p)
838 {
839         struct device *dev = &p->pdev->dev;
840         struct nd_mapping_desc mapping;
841         struct nd_region_desc ndr_desc;
842         unsigned long dimm_flags;
843         int target_nid, online_nid;
844         ssize_t stat_size;
845
846         p->bus_desc.ndctl = papr_scm_ndctl;
847         p->bus_desc.module = THIS_MODULE;
848         p->bus_desc.of_node = p->pdev->dev.of_node;
849         p->bus_desc.provider_name = kstrdup(p->pdev->name, GFP_KERNEL);
850
851         if (!p->bus_desc.provider_name)
852                 return -ENOMEM;
853
854         p->bus = nvdimm_bus_register(NULL, &p->bus_desc);
855         if (!p->bus) {
856                 dev_err(dev, "Error creating nvdimm bus %pOF\n", p->dn);
857                 kfree(p->bus_desc.provider_name);
858                 return -ENXIO;
859         }
860
861         dimm_flags = 0;
862         set_bit(NDD_LABELING, &dimm_flags);
863
864         p->nvdimm = nvdimm_create(p->bus, p, papr_nd_attr_groups,
865                                   dimm_flags, PAPR_SCM_DIMM_CMD_MASK, 0, NULL);
866         if (!p->nvdimm) {
867                 dev_err(dev, "Error creating DIMM object for %pOF\n", p->dn);
868                 goto err;
869         }
870
871         if (nvdimm_bus_check_dimm_count(p->bus, 1))
872                 goto err;
873
874         /* now add the region */
875
876         memset(&mapping, 0, sizeof(mapping));
877         mapping.nvdimm = p->nvdimm;
878         mapping.start = 0;
879         mapping.size = p->blocks * p->block_size; // XXX: potential overflow?
880
881         memset(&ndr_desc, 0, sizeof(ndr_desc));
882         target_nid = dev_to_node(&p->pdev->dev);
883         online_nid = numa_map_to_online_node(target_nid);
884         ndr_desc.numa_node = online_nid;
885         ndr_desc.target_node = target_nid;
886         ndr_desc.res = &p->res;
887         ndr_desc.of_node = p->dn;
888         ndr_desc.provider_data = p;
889         ndr_desc.mapping = &mapping;
890         ndr_desc.num_mappings = 1;
891         ndr_desc.nd_set = &p->nd_set;
892
893         if (p->is_volatile)
894                 p->region = nvdimm_volatile_region_create(p->bus, &ndr_desc);
895         else {
896                 set_bit(ND_REGION_PERSIST_MEMCTRL, &ndr_desc.flags);
897                 p->region = nvdimm_pmem_region_create(p->bus, &ndr_desc);
898         }
899         if (!p->region) {
900                 dev_err(dev, "Error registering region %pR from %pOF\n",
901                                 ndr_desc.res, p->dn);
902                 goto err;
903         }
904         if (target_nid != online_nid)
905                 dev_info(dev, "Region registered with target node %d and online node %d",
906                          target_nid, online_nid);
907
908         mutex_lock(&papr_ndr_lock);
909         list_add_tail(&p->region_list, &papr_nd_regions);
910         mutex_unlock(&papr_ndr_lock);
911
912         /* Try retriving the stat buffer and see if its supported */
913         stat_size = drc_pmem_query_stats(p, NULL, 0);
914         if (stat_size > 0) {
915                 p->stat_buffer_len = stat_size;
916                 dev_dbg(&p->pdev->dev, "Max perf-stat size %lu-bytes\n",
917                         p->stat_buffer_len);
918         } else {
919                 dev_info(&p->pdev->dev, "Dimm performance stats unavailable\n");
920         }
921
922         return 0;
923
924 err:    nvdimm_bus_unregister(p->bus);
925         kfree(p->bus_desc.provider_name);
926         return -ENXIO;
927 }
928
929 static void papr_scm_add_badblock(struct nd_region *region,
930                                   struct nvdimm_bus *bus, u64 phys_addr)
931 {
932         u64 aligned_addr = ALIGN_DOWN(phys_addr, L1_CACHE_BYTES);
933
934         if (nvdimm_bus_add_badrange(bus, aligned_addr, L1_CACHE_BYTES)) {
935                 pr_err("Bad block registration for 0x%llx failed\n", phys_addr);
936                 return;
937         }
938
939         pr_debug("Add memory range (0x%llx - 0x%llx) as bad range\n",
940                  aligned_addr, aligned_addr + L1_CACHE_BYTES);
941
942         nvdimm_region_notify(region, NVDIMM_REVALIDATE_POISON);
943 }
944
945 static int handle_mce_ue(struct notifier_block *nb, unsigned long val,
946                          void *data)
947 {
948         struct machine_check_event *evt = data;
949         struct papr_scm_priv *p;
950         u64 phys_addr;
951         bool found = false;
952
953         if (evt->error_type != MCE_ERROR_TYPE_UE)
954                 return NOTIFY_DONE;
955
956         if (list_empty(&papr_nd_regions))
957                 return NOTIFY_DONE;
958
959         /*
960          * The physical address obtained here is PAGE_SIZE aligned, so get the
961          * exact address from the effective address
962          */
963         phys_addr = evt->u.ue_error.physical_address +
964                         (evt->u.ue_error.effective_address & ~PAGE_MASK);
965
966         if (!evt->u.ue_error.physical_address_provided ||
967             !is_zone_device_page(pfn_to_page(phys_addr >> PAGE_SHIFT)))
968                 return NOTIFY_DONE;
969
970         /* mce notifier is called from a process context, so mutex is safe */
971         mutex_lock(&papr_ndr_lock);
972         list_for_each_entry(p, &papr_nd_regions, region_list) {
973                 if (phys_addr >= p->res.start && phys_addr <= p->res.end) {
974                         found = true;
975                         break;
976                 }
977         }
978
979         if (found)
980                 papr_scm_add_badblock(p->region, p->bus, phys_addr);
981
982         mutex_unlock(&papr_ndr_lock);
983
984         return found ? NOTIFY_OK : NOTIFY_DONE;
985 }
986
987 static struct notifier_block mce_ue_nb = {
988         .notifier_call = handle_mce_ue
989 };
990
991 static int papr_scm_probe(struct platform_device *pdev)
992 {
993         struct device_node *dn = pdev->dev.of_node;
994         u32 drc_index, metadata_size;
995         u64 blocks, block_size;
996         struct papr_scm_priv *p;
997         const char *uuid_str;
998         u64 uuid[2];
999         int rc;
1000
1001         /* check we have all the required DT properties */
1002         if (of_property_read_u32(dn, "ibm,my-drc-index", &drc_index)) {
1003                 dev_err(&pdev->dev, "%pOF: missing drc-index!\n", dn);
1004                 return -ENODEV;
1005         }
1006
1007         if (of_property_read_u64(dn, "ibm,block-size", &block_size)) {
1008                 dev_err(&pdev->dev, "%pOF: missing block-size!\n", dn);
1009                 return -ENODEV;
1010         }
1011
1012         if (of_property_read_u64(dn, "ibm,number-of-blocks", &blocks)) {
1013                 dev_err(&pdev->dev, "%pOF: missing number-of-blocks!\n", dn);
1014                 return -ENODEV;
1015         }
1016
1017         if (of_property_read_string(dn, "ibm,unit-guid", &uuid_str)) {
1018                 dev_err(&pdev->dev, "%pOF: missing unit-guid!\n", dn);
1019                 return -ENODEV;
1020         }
1021
1022
1023         p = kzalloc(sizeof(*p), GFP_KERNEL);
1024         if (!p)
1025                 return -ENOMEM;
1026
1027         /* Initialize the dimm mutex */
1028         mutex_init(&p->health_mutex);
1029
1030         /* optional DT properties */
1031         of_property_read_u32(dn, "ibm,metadata-size", &metadata_size);
1032
1033         p->dn = dn;
1034         p->drc_index = drc_index;
1035         p->block_size = block_size;
1036         p->blocks = blocks;
1037         p->is_volatile = !of_property_read_bool(dn, "ibm,cache-flush-required");
1038
1039         /* We just need to ensure that set cookies are unique across */
1040         uuid_parse(uuid_str, (uuid_t *) uuid);
1041         /*
1042          * cookie1 and cookie2 are not really little endian
1043          * we store a little endian representation of the
1044          * uuid str so that we can compare this with the label
1045          * area cookie irrespective of the endian config with which
1046          * the kernel is built.
1047          */
1048         p->nd_set.cookie1 = cpu_to_le64(uuid[0]);
1049         p->nd_set.cookie2 = cpu_to_le64(uuid[1]);
1050
1051         /* might be zero */
1052         p->metadata_size = metadata_size;
1053         p->pdev = pdev;
1054
1055         /* request the hypervisor to bind this region to somewhere in memory */
1056         rc = drc_pmem_bind(p);
1057
1058         /* If phyp says drc memory still bound then force unbound and retry */
1059         if (rc == H_OVERLAP)
1060                 rc = drc_pmem_query_n_bind(p);
1061
1062         if (rc != H_SUCCESS) {
1063                 dev_err(&p->pdev->dev, "bind err: %d\n", rc);
1064                 rc = -ENXIO;
1065                 goto err;
1066         }
1067
1068         /* setup the resource for the newly bound range */
1069         p->res.start = p->bound_addr;
1070         p->res.end   = p->bound_addr + p->blocks * p->block_size - 1;
1071         p->res.name  = pdev->name;
1072         p->res.flags = IORESOURCE_MEM;
1073
1074         rc = papr_scm_nvdimm_init(p);
1075         if (rc)
1076                 goto err2;
1077
1078         platform_set_drvdata(pdev, p);
1079
1080         return 0;
1081
1082 err2:   drc_pmem_unbind(p);
1083 err:    kfree(p);
1084         return rc;
1085 }
1086
1087 static int papr_scm_remove(struct platform_device *pdev)
1088 {
1089         struct papr_scm_priv *p = platform_get_drvdata(pdev);
1090
1091         mutex_lock(&papr_ndr_lock);
1092         list_del(&p->region_list);
1093         mutex_unlock(&papr_ndr_lock);
1094
1095         nvdimm_bus_unregister(p->bus);
1096         drc_pmem_unbind(p);
1097         kfree(p->bus_desc.provider_name);
1098         kfree(p);
1099
1100         return 0;
1101 }
1102
1103 static const struct of_device_id papr_scm_match[] = {
1104         { .compatible = "ibm,pmemory" },
1105         { .compatible = "ibm,pmemory-v2" },
1106         { },
1107 };
1108
1109 static struct platform_driver papr_scm_driver = {
1110         .probe = papr_scm_probe,
1111         .remove = papr_scm_remove,
1112         .driver = {
1113                 .name = "papr_scm",
1114                 .of_match_table = papr_scm_match,
1115         },
1116 };
1117
1118 static int __init papr_scm_init(void)
1119 {
1120         int ret;
1121
1122         ret = platform_driver_register(&papr_scm_driver);
1123         if (!ret)
1124                 mce_register_notifier(&mce_ue_nb);
1125
1126         return ret;
1127 }
1128 module_init(papr_scm_init);
1129
1130 static void __exit papr_scm_exit(void)
1131 {
1132         mce_unregister_notifier(&mce_ue_nb);
1133         platform_driver_unregister(&papr_scm_driver);
1134 }
1135 module_exit(papr_scm_exit);
1136
1137 MODULE_DEVICE_TABLE(of, papr_scm_match);
1138 MODULE_LICENSE("GPL");
1139 MODULE_AUTHOR("IBM Corporation");