1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * store hypervisor information instruction emulation functions.
4 *
5 * Copyright IBM Corp. 2016
6 * Author(s): Janosch Frank <frankja@linux.vnet.ibm.com>
7 */
8
9 #include <linux/export.h>
10 #include <linux/errno.h>
11 #include <linux/pagemap.h>
12 #include <linux/vmalloc.h>
13 #include <linux/syscalls.h>
14 #include <linux/mutex.h>
15 #include <asm/asm-offsets.h>
16 #include <asm/sclp.h>
17 #include <asm/diag.h>
18 #include <asm/sysinfo.h>
19 #include <asm/ebcdic.h>
20 #include <asm/facility.h>
21 #include <asm/sthyi.h>
22 #include <asm/asm.h>
23 #include "entry.h"
24
25 #define DED_WEIGHT 0xffff
26 /*
27 * CP and IFL as EBCDIC strings, SP/0x40 determines the end of string
28 * as they are justified with spaces.
29 */
30 #define CP 0xc3d7404040404040UL
31 #define IFL 0xc9c6d34040404040UL
32
33 enum hdr_flags {
34 HDR_NOT_LPAR = 0x10,
35 HDR_STACK_INCM = 0x20,
36 HDR_STSI_UNAV = 0x40,
37 HDR_PERF_UNAV = 0x80,
38 };
39
40 enum mac_validity {
41 MAC_NAME_VLD = 0x20,
42 MAC_ID_VLD = 0x40,
43 MAC_CNT_VLD = 0x80,
44 };
45
46 enum par_flag {
47 PAR_MT_EN = 0x80,
48 };
49
50 enum par_validity {
51 PAR_GRP_VLD = 0x08,
52 PAR_ID_VLD = 0x10,
53 PAR_ABS_VLD = 0x20,
54 PAR_WGHT_VLD = 0x40,
55 PAR_PCNT_VLD = 0x80,
56 };
57
58 struct hdr_sctn {
59 u8 infhflg1;
60 u8 infhflg2; /* reserved */
61 u8 infhval1; /* reserved */
62 u8 infhval2; /* reserved */
63 u8 reserved[3];
64 u8 infhygct;
65 u16 infhtotl;
66 u16 infhdln;
67 u16 infmoff;
68 u16 infmlen;
69 u16 infpoff;
70 u16 infplen;
71 u16 infhoff1;
72 u16 infhlen1;
73 u16 infgoff1;
74 u16 infglen1;
75 u16 infhoff2;
76 u16 infhlen2;
77 u16 infgoff2;
78 u16 infglen2;
79 u16 infhoff3;
80 u16 infhlen3;
81 u16 infgoff3;
82 u16 infglen3;
83 u8 reserved2[4];
84 } __packed;
85
86 struct mac_sctn {
87 u8 infmflg1; /* reserved */
88 u8 infmflg2; /* reserved */
89 u8 infmval1;
90 u8 infmval2; /* reserved */
91 u16 infmscps;
92 u16 infmdcps;
93 u16 infmsifl;
94 u16 infmdifl;
95 char infmname[8];
96 char infmtype[4];
97 char infmmanu[16];
98 char infmseq[16];
99 char infmpman[4];
100 u8 reserved[4];
101 } __packed;
102
103 struct par_sctn {
104 u8 infpflg1;
105 u8 infpflg2; /* reserved */
106 u8 infpval1;
107 u8 infpval2; /* reserved */
108 u16 infppnum;
109 u16 infpscps;
110 u16 infpdcps;
111 u16 infpsifl;
112 u16 infpdifl;
113 u16 reserved;
114 char infppnam[8];
115 u32 infpwbcp;
116 u32 infpabcp;
117 u32 infpwbif;
118 u32 infpabif;
119 char infplgnm[8];
120 u32 infplgcp;
121 u32 infplgif;
122 } __packed;
123
124 struct sthyi_sctns {
125 struct hdr_sctn hdr;
126 struct mac_sctn mac;
127 struct par_sctn par;
128 } __packed;
129
130 struct cpu_inf {
131 u64 lpar_cap;
132 u64 lpar_grp_cap;
133 u64 lpar_weight;
134 u64 all_weight;
135 int cpu_num_ded;
136 int cpu_num_shd;
137 };
138
139 struct lpar_cpu_inf {
140 struct cpu_inf cp;
141 struct cpu_inf ifl;
142 };
143
144 /*
145 * STHYI requires extensive locking in the higher hypervisors
146 * and is very computational/memory expensive. Therefore we
147 * cache the retrieved data whose valid period is 1s.
148 */
149 #define CACHE_VALID_JIFFIES HZ
150
151 struct sthyi_info {
152 void *info;
153 unsigned long end;
154 };
155
156 static DEFINE_MUTEX(sthyi_mutex);
157 static struct sthyi_info sthyi_cache;
158
cpu_id(u8 ctidx,void * diag224_buf)159 static inline u64 cpu_id(u8 ctidx, void *diag224_buf)
160 {
161 return *((u64 *)(diag224_buf + (ctidx + 1) * DIAG204_CPU_NAME_LEN));
162 }
163
164 /*
165 * Scales the cpu capping from the lpar range to the one expected in
166 * sthyi data.
167 *
168 * diag204 reports a cap in hundredths of processor units.
169 * z/VM's range for one core is 0 - 0x10000.
170 */
scale_cap(u32 in)171 static u32 scale_cap(u32 in)
172 {
173 return (0x10000 * in) / 100;
174 }
175
fill_hdr(struct sthyi_sctns * sctns)176 static void fill_hdr(struct sthyi_sctns *sctns)
177 {
178 sctns->hdr.infhdln = sizeof(sctns->hdr);
179 sctns->hdr.infmoff = sizeof(sctns->hdr);
180 sctns->hdr.infmlen = sizeof(sctns->mac);
181 sctns->hdr.infplen = sizeof(sctns->par);
182 sctns->hdr.infpoff = sctns->hdr.infhdln + sctns->hdr.infmlen;
183 sctns->hdr.infhtotl = sctns->hdr.infpoff + sctns->hdr.infplen;
184 }
185
fill_stsi_mac(struct sthyi_sctns * sctns,struct sysinfo_1_1_1 * sysinfo)186 static void fill_stsi_mac(struct sthyi_sctns *sctns,
187 struct sysinfo_1_1_1 *sysinfo)
188 {
189 sclp_ocf_cpc_name_copy(sctns->mac.infmname);
190 if (*(u64 *)sctns->mac.infmname != 0)
191 sctns->mac.infmval1 |= MAC_NAME_VLD;
192
193 if (stsi(sysinfo, 1, 1, 1))
194 return;
195
196 memcpy(sctns->mac.infmtype, sysinfo->type, sizeof(sctns->mac.infmtype));
197 memcpy(sctns->mac.infmmanu, sysinfo->manufacturer, sizeof(sctns->mac.infmmanu));
198 memcpy(sctns->mac.infmpman, sysinfo->plant, sizeof(sctns->mac.infmpman));
199 memcpy(sctns->mac.infmseq, sysinfo->sequence, sizeof(sctns->mac.infmseq));
200
201 sctns->mac.infmval1 |= MAC_ID_VLD;
202 }
203
fill_stsi_par(struct sthyi_sctns * sctns,struct sysinfo_2_2_2 * sysinfo)204 static void fill_stsi_par(struct sthyi_sctns *sctns,
205 struct sysinfo_2_2_2 *sysinfo)
206 {
207 if (stsi(sysinfo, 2, 2, 2))
208 return;
209
210 sctns->par.infppnum = sysinfo->lpar_number;
211 memcpy(sctns->par.infppnam, sysinfo->name, sizeof(sctns->par.infppnam));
212
213 sctns->par.infpval1 |= PAR_ID_VLD;
214 }
215
fill_stsi(struct sthyi_sctns * sctns)216 static void fill_stsi(struct sthyi_sctns *sctns)
217 {
218 void *sysinfo;
219
220 /* Errors are handled through the validity bits in the response. */
221 sysinfo = (void *)__get_free_page(GFP_KERNEL);
222 if (!sysinfo)
223 return;
224
225 fill_stsi_mac(sctns, sysinfo);
226 fill_stsi_par(sctns, sysinfo);
227
228 free_pages((unsigned long)sysinfo, 0);
229 }
230
fill_diag_mac(struct sthyi_sctns * sctns,struct diag204_x_phys_block * block,void * diag224_buf)231 static void fill_diag_mac(struct sthyi_sctns *sctns,
232 struct diag204_x_phys_block *block,
233 void *diag224_buf)
234 {
235 int i;
236
237 for (i = 0; i < block->hdr.cpus; i++) {
238 switch (cpu_id(block->cpus[i].ctidx, diag224_buf)) {
239 case CP:
240 if (block->cpus[i].weight == DED_WEIGHT)
241 sctns->mac.infmdcps++;
242 else
243 sctns->mac.infmscps++;
244 break;
245 case IFL:
246 if (block->cpus[i].weight == DED_WEIGHT)
247 sctns->mac.infmdifl++;
248 else
249 sctns->mac.infmsifl++;
250 break;
251 }
252 }
253 sctns->mac.infmval1 |= MAC_CNT_VLD;
254 }
255
256 /* Returns a pointer to the the next partition block. */
lpar_cpu_inf(struct lpar_cpu_inf * part_inf,bool this_lpar,void * diag224_buf,struct diag204_x_part_block * block)257 static struct diag204_x_part_block *lpar_cpu_inf(struct lpar_cpu_inf *part_inf,
258 bool this_lpar,
259 void *diag224_buf,
260 struct diag204_x_part_block *block)
261 {
262 int i, capped = 0, weight_cp = 0, weight_ifl = 0;
263 struct cpu_inf *cpu_inf;
264
265 for (i = 0; i < block->hdr.rcpus; i++) {
266 if (!(block->cpus[i].cflag & DIAG204_CPU_ONLINE))
267 continue;
268
269 switch (cpu_id(block->cpus[i].ctidx, diag224_buf)) {
270 case CP:
271 cpu_inf = &part_inf->cp;
272 if (block->cpus[i].cur_weight < DED_WEIGHT)
273 weight_cp |= block->cpus[i].cur_weight;
274 break;
275 case IFL:
276 cpu_inf = &part_inf->ifl;
277 if (block->cpus[i].cur_weight < DED_WEIGHT)
278 weight_ifl |= block->cpus[i].cur_weight;
279 break;
280 default:
281 continue;
282 }
283
284 if (!this_lpar)
285 continue;
286
287 capped |= block->cpus[i].cflag & DIAG204_CPU_CAPPED;
288 cpu_inf->lpar_cap |= block->cpus[i].cpu_type_cap;
289 cpu_inf->lpar_grp_cap |= block->cpus[i].group_cpu_type_cap;
290
291 if (block->cpus[i].weight == DED_WEIGHT)
292 cpu_inf->cpu_num_ded += 1;
293 else
294 cpu_inf->cpu_num_shd += 1;
295 }
296
297 if (this_lpar && capped) {
298 part_inf->cp.lpar_weight = weight_cp;
299 part_inf->ifl.lpar_weight = weight_ifl;
300 }
301 part_inf->cp.all_weight += weight_cp;
302 part_inf->ifl.all_weight += weight_ifl;
303 return (struct diag204_x_part_block *)&block->cpus[i];
304 }
305
diag204_get_data(bool diag204_allow_busy)306 static void *diag204_get_data(bool diag204_allow_busy)
307 {
308 unsigned long subcode;
309 void *diag204_buf;
310 int pages, rc;
311
312 subcode = DIAG204_SUBC_RSI;
313 subcode |= DIAG204_INFO_EXT;
314 pages = diag204(subcode, 0, NULL);
315 if (pages < 0)
316 return ERR_PTR(pages);
317 if (pages == 0)
318 return ERR_PTR(-ENODATA);
319 diag204_buf = __vmalloc_node(array_size(pages, PAGE_SIZE),
320 PAGE_SIZE, GFP_KERNEL, NUMA_NO_NODE,
321 __builtin_return_address(0));
322 if (!diag204_buf)
323 return ERR_PTR(-ENOMEM);
324 subcode = DIAG204_SUBC_STIB7;
325 subcode |= DIAG204_INFO_EXT;
326 if (diag204_has_bif() && diag204_allow_busy)
327 subcode |= DIAG204_BIF_BIT;
328 rc = diag204(subcode, pages, diag204_buf);
329 if (rc < 0) {
330 vfree(diag204_buf);
331 return ERR_PTR(rc);
332 }
333 return diag204_buf;
334 }
335
is_diag204_cached(struct sthyi_sctns * sctns)336 static bool is_diag204_cached(struct sthyi_sctns *sctns)
337 {
338 /*
339 * Check if validity bits are set when diag204 data
340 * is gathered.
341 */
342 if (sctns->par.infpval1)
343 return true;
344 return false;
345 }
346
fill_diag(struct sthyi_sctns * sctns,void * diag204_buf)347 static void fill_diag(struct sthyi_sctns *sctns, void *diag204_buf)
348 {
349 int i;
350 bool this_lpar;
351 void *diag224_buf = NULL;
352 struct diag204_x_info_blk_hdr *ti_hdr;
353 struct diag204_x_part_block *part_block;
354 struct diag204_x_phys_block *phys_block;
355 struct lpar_cpu_inf lpar_inf = {};
356
357 diag224_buf = (void *)__get_free_page(GFP_KERNEL | GFP_DMA);
358 if (!diag224_buf || diag224(diag224_buf))
359 goto out;
360
361 ti_hdr = diag204_buf;
362 part_block = diag204_buf + sizeof(*ti_hdr);
363
364 for (i = 0; i < ti_hdr->npar; i++) {
365 /*
366 * For the calling lpar we also need to get the cpu
367 * caps and weights. The time information block header
368 * specifies the offset to the partition block of the
369 * caller lpar, so we know when we process its data.
370 */
371 this_lpar = (void *)part_block - diag204_buf == ti_hdr->this_part;
372 part_block = lpar_cpu_inf(&lpar_inf, this_lpar, diag224_buf,
373 part_block);
374 }
375
376 phys_block = (struct diag204_x_phys_block *)part_block;
377 part_block = diag204_buf + ti_hdr->this_part;
378 if (part_block->hdr.mtid)
379 sctns->par.infpflg1 = PAR_MT_EN;
380
381 sctns->par.infpval1 |= PAR_GRP_VLD;
382 sctns->par.infplgcp = scale_cap(lpar_inf.cp.lpar_grp_cap);
383 sctns->par.infplgif = scale_cap(lpar_inf.ifl.lpar_grp_cap);
384 memcpy(sctns->par.infplgnm, part_block->hdr.hardware_group_name,
385 sizeof(sctns->par.infplgnm));
386
387 sctns->par.infpscps = lpar_inf.cp.cpu_num_shd;
388 sctns->par.infpdcps = lpar_inf.cp.cpu_num_ded;
389 sctns->par.infpsifl = lpar_inf.ifl.cpu_num_shd;
390 sctns->par.infpdifl = lpar_inf.ifl.cpu_num_ded;
391 sctns->par.infpval1 |= PAR_PCNT_VLD;
392
393 sctns->par.infpabcp = scale_cap(lpar_inf.cp.lpar_cap);
394 sctns->par.infpabif = scale_cap(lpar_inf.ifl.lpar_cap);
395 sctns->par.infpval1 |= PAR_ABS_VLD;
396
397 /*
398 * Everything below needs global performance data to be
399 * meaningful.
400 */
401 if (!(ti_hdr->flags & DIAG204_LPAR_PHYS_FLG)) {
402 sctns->hdr.infhflg1 |= HDR_PERF_UNAV;
403 goto out;
404 }
405
406 fill_diag_mac(sctns, phys_block, diag224_buf);
407
408 if (lpar_inf.cp.lpar_weight) {
409 sctns->par.infpwbcp = sctns->mac.infmscps * 0x10000 *
410 lpar_inf.cp.lpar_weight / lpar_inf.cp.all_weight;
411 }
412
413 if (lpar_inf.ifl.lpar_weight) {
414 sctns->par.infpwbif = sctns->mac.infmsifl * 0x10000 *
415 lpar_inf.ifl.lpar_weight / lpar_inf.ifl.all_weight;
416 }
417 sctns->par.infpval1 |= PAR_WGHT_VLD;
418
419 out:
420 free_page((unsigned long)diag224_buf);
421 }
422
sthyi(u64 vaddr,u64 * rc)423 static int sthyi(u64 vaddr, u64 *rc)
424 {
425 union register_pair r1 = { .even = 0, }; /* subcode */
426 union register_pair r2 = { .even = vaddr, };
427 int cc;
428
429 asm volatile(
430 ".insn rre,0xB2560000,%[r1],%[r2]\n"
431 CC_IPM(cc)
432 : CC_OUT(cc, cc), [r2] "+&d" (r2.pair)
433 : [r1] "d" (r1.pair)
434 : CC_CLOBBER_LIST("memory"));
435 *rc = r2.odd;
436 return CC_TRANSFORM(cc);
437 }
438
fill_dst(void * dst,u64 * rc)439 static int fill_dst(void *dst, u64 *rc)
440 {
441 void *diag204_buf;
442
443 struct sthyi_sctns *sctns = (struct sthyi_sctns *)dst;
444
445 /*
446 * If the facility is on, we don't want to emulate the instruction.
447 * We ask the hypervisor to provide the data.
448 */
449 if (test_facility(74)) {
450 memset(dst, 0, PAGE_SIZE);
451 return sthyi((u64)dst, rc);
452 }
453 /*
454 * When emulating, if diag204 returns BUSY don't reset dst buffer
455 * and use cached data.
456 */
457 *rc = 0;
458 diag204_buf = diag204_get_data(is_diag204_cached(sctns));
459 if (IS_ERR(diag204_buf))
460 return PTR_ERR(diag204_buf);
461 memset(dst, 0, PAGE_SIZE);
462 fill_hdr(sctns);
463 fill_stsi(sctns);
464 fill_diag(sctns, diag204_buf);
465 vfree(diag204_buf);
466 return 0;
467 }
468
sthyi_init_cache(void)469 static int sthyi_init_cache(void)
470 {
471 if (sthyi_cache.info)
472 return 0;
473 sthyi_cache.info = (void *)get_zeroed_page(GFP_KERNEL);
474 if (!sthyi_cache.info)
475 return -ENOMEM;
476 sthyi_cache.end = jiffies - 1; /* expired */
477 return 0;
478 }
479
sthyi_update_cache(u64 * rc)480 static int sthyi_update_cache(u64 *rc)
481 {
482 int r;
483
484 r = fill_dst(sthyi_cache.info, rc);
485 if (r == 0) {
486 sthyi_cache.end = jiffies + CACHE_VALID_JIFFIES;
487 } else if (r == -EBUSY) {
488 /* mark as expired and return 0 to keep using cached data */
489 sthyi_cache.end = jiffies - 1;
490 r = 0;
491 }
492 return r;
493 }
494
495 /*
496 * sthyi_fill - Fill page with data returned by the STHYI instruction
497 *
498 * @dst: Pointer to zeroed page
499 * @rc: Pointer for storing the return code of the instruction
500 *
501 * Fills the destination with system information returned by the STHYI
502 * instruction. The data is generated by emulation or execution of STHYI,
503 * if available. The return value is either a negative error value or
504 * the condition code that would be returned, the rc parameter is the
505 * return code which is passed in register R2 + 1.
506 */
sthyi_fill(void * dst,u64 * rc)507 int sthyi_fill(void *dst, u64 *rc)
508 {
509 int r;
510
511 mutex_lock(&sthyi_mutex);
512 r = sthyi_init_cache();
513 if (r)
514 goto out;
515
516 if (time_is_before_jiffies(sthyi_cache.end)) {
517 /* cache expired */
518 r = sthyi_update_cache(rc);
519 if (r)
520 goto out;
521 }
522 *rc = 0;
523 memcpy(dst, sthyi_cache.info, PAGE_SIZE);
524 out:
525 mutex_unlock(&sthyi_mutex);
526 return r;
527 }
528 EXPORT_SYMBOL_GPL(sthyi_fill);
529
SYSCALL_DEFINE4(s390_sthyi,unsigned long,function_code,void __user *,buffer,u64 __user *,return_code,unsigned long,flags)530 SYSCALL_DEFINE4(s390_sthyi, unsigned long, function_code, void __user *, buffer,
531 u64 __user *, return_code, unsigned long, flags)
532 {
533 u64 sthyi_rc;
534 void *info;
535 int r;
536
537 if (flags)
538 return -EINVAL;
539 if (function_code != STHYI_FC_CP_IFL_CAP)
540 return -EOPNOTSUPP;
541 info = (void *)get_zeroed_page(GFP_KERNEL);
542 if (!info)
543 return -ENOMEM;
544 r = sthyi_fill(info, &sthyi_rc);
545 if (r < 0)
546 goto out;
547 if (return_code && put_user(sthyi_rc, return_code)) {
548 r = -EFAULT;
549 goto out;
550 }
551 if (copy_to_user(buffer, info, PAGE_SIZE))
552 r = -EFAULT;
553 out:
554 free_page((unsigned long)info);
555 return r;
556 }
557