1 /* SPDX-License-Identifier: GPL-2.0-only */
2 /*
3 * CPU Topology
4 *
5 * Copyright IBM Corp. 2022, 2023
6 *
7 * Authors:
8 * Pierre Morel <pmorel@linux.ibm.com>
9 */
10
11 #include <libcflat.h>
12 #include <asm/page.h>
13 #include <asm/asm-offsets.h>
14 #include <asm/interrupt.h>
15 #include <asm/facility.h>
16 #include <asm/barrier.h>
17 #include <smp.h>
18 #include <sclp.h>
19 #include <s390x/hardware.h>
20 #include <s390x/stsi.h>
21
22 static uint8_t pagebuf[PAGE_SIZE] __attribute__((aligned(PAGE_SIZE)));
23
24 static int max_nested_lvl;
25 static int number_of_cpus;
26 static int max_cpus;
27
28 /*
29 * Topology level as defined by architecture, all levels exists with
30 * a single container unless overwritten by the QEMU -smp parameter.
31 */
32 static int expected_topo_lvl[CPU_TOPOLOGY_MAX_LEVEL] = { 1, 1, 1, 1, 1, 1 };
33
34 #define PTF_REQ_HORIZONTAL 0
35 #define PTF_REQ_VERTICAL 1
36 #define PTF_CHECK 2
37
38 #define PTF_ERR_NO_REASON 0
39 #define PTF_ERR_ALRDY_POLARIZED 1
40 #define PTF_ERR_IN_PROGRESS 2
41
42 extern int diag308_load_reset(u64);
43
ptf(unsigned long fc,unsigned long * rc)44 static int ptf(unsigned long fc, unsigned long *rc)
45 {
46 int cc;
47
48 asm volatile(
49 " ptf %1 \n"
50 " ipm %0 \n"
51 " srl %0,28 \n"
52 : "=d" (cc), "+d" (fc)
53 :
54 : "cc");
55
56 *rc = fc >> 8;
57 return cc;
58 }
59
check_privilege(int fc)60 static void check_privilege(int fc)
61 {
62 unsigned long rc;
63
64 report_prefix_pushf("Privileged fc %d", fc);
65 enter_pstate();
66 expect_pgm_int();
67 ptf(fc, &rc);
68 check_pgm_int_code(PGM_INT_CODE_PRIVILEGED_OPERATION);
69 report_prefix_pop();
70 }
71
check_specifications(void)72 static void check_specifications(void)
73 {
74 unsigned long error = 0;
75 unsigned long ptf_bits;
76 unsigned long rc;
77 int i;
78
79 report_prefix_push("Specifications");
80
81 /* Function codes above 3 are undefined */
82 for (i = 4; i < 255; i++) {
83 expect_pgm_int();
84 ptf(i, &rc);
85 if (clear_pgm_int() != PGM_INT_CODE_SPECIFICATION) {
86 report_fail("FC %d did not yield specification exception", i);
87 error = 1;
88 }
89 }
90 report(!error, "Undefined function codes");
91
92 /* Reserved bits must be 0 */
93 for (i = 8, error = 0; i < 64; i++) {
94 ptf_bits = 0x01UL << i;
95 expect_pgm_int();
96 ptf(ptf_bits, &rc);
97 if (clear_pgm_int() != PGM_INT_CODE_SPECIFICATION) {
98 report_fail("Reserved bit %d did not yield specification exception", i);
99 error = 1;
100 }
101 }
102
103 report(!error, "Reserved bits");
104
105 report_prefix_pop();
106 }
107
check_polarization_change(void)108 static void check_polarization_change(void)
109 {
110 unsigned long rc;
111 int cc;
112
113 report_prefix_push("Polarization change");
114
115 /* We expect a clean state through reset */
116 assert(diag308_load_reset(1));
117
118 /*
119 * Set vertical polarization to verify that RESET sets
120 * horizontal polarization back.
121 */
122 cc = ptf(PTF_REQ_VERTICAL, &rc);
123 report(cc == 0, "Set vertical polarization.");
124
125 assert(diag308_load_reset(1));
126
127 cc = ptf(PTF_CHECK, &rc);
128 report(cc == 0, "Reset should clear topology report");
129
130 cc = ptf(PTF_REQ_HORIZONTAL, &rc);
131 report(cc == 2 && rc == PTF_ERR_ALRDY_POLARIZED,
132 "After RESET polarization is horizontal");
133
134 /* Flip between vertical and horizontal polarization */
135 cc = ptf(PTF_REQ_VERTICAL, &rc);
136 report(cc == 0, "Change to vertical");
137
138 cc = ptf(PTF_CHECK, &rc);
139 report(cc == 1, "Should report change after horizontal -> vertical");
140
141 cc = ptf(PTF_REQ_VERTICAL, &rc);
142 report(cc == 2 && rc == PTF_ERR_ALRDY_POLARIZED, "Double change to vertical");
143
144 cc = ptf(PTF_CHECK, &rc);
145 report(cc == 0, "Should not report change after vertical -> vertical");
146
147 cc = ptf(PTF_REQ_HORIZONTAL, &rc);
148 report(cc == 0, "Change to horizontal");
149
150 cc = ptf(PTF_CHECK, &rc);
151 report(cc == 1, "Should report change after vertical -> horizontal");
152
153 cc = ptf(PTF_REQ_HORIZONTAL, &rc);
154 report(cc == 2 && rc == PTF_ERR_ALRDY_POLARIZED, "Double change to horizontal");
155
156 cc = ptf(PTF_CHECK, &rc);
157 report(cc == 0, "Should not report change after horizontal -> horizontal");
158
159 report_prefix_pop();
160 }
161
test_ptf(void)162 static void test_ptf(void)
163 {
164 check_privilege(PTF_REQ_HORIZONTAL);
165 check_privilege(PTF_REQ_VERTICAL);
166 check_privilege(PTF_CHECK);
167 check_specifications();
168 check_polarization_change();
169 }
170
171 /*
172 * stsi_check_maxcpus
173 * @info: Pointer to the stsi information
174 *
175 * The product of the numbers of containers per level
176 * is the maximum number of CPU allowed by the machine.
177 */
stsi_check_maxcpus(struct sysinfo_15_1_x * info)178 static void stsi_check_maxcpus(struct sysinfo_15_1_x *info)
179 {
180 int n, i;
181
182 for (i = 0, n = 1; i < CPU_TOPOLOGY_MAX_LEVEL; i++)
183 n *= info->mag[i] ?: 1;
184
185 report(n == max_cpus, "Calculated max CPUs: %d", n);
186 }
187
188 /*
189 * stsi_check_header
190 * @info: Pointer to the stsi information
191 * @sel2: stsi selector 2 value
192 *
193 * MAG field should match the architecture defined containers
194 * when MNEST as returned by SCLP matches MNEST of the SYSIB.
195 */
stsi_check_header(struct sysinfo_15_1_x * info,int sel2)196 static void stsi_check_header(struct sysinfo_15_1_x *info, int sel2)
197 {
198 int i;
199
200 report_prefix_push("Header");
201
202 /* Header is 16 bytes, each TLE 8 or 16, therefore alignment must be 8 at least */
203 report(IS_ALIGNED(info->length, 8), "Length %d multiple of 8", info->length);
204 report(info->length < PAGE_SIZE, "Length %d in bounds", info->length);
205 report(sel2 == info->mnest, "Valid mnest");
206 stsi_check_maxcpus(info);
207
208 /*
209 * It is not clear how the MAG fields are calculated when mnest
210 * in the SYSIB 15.x is different from the maximum nested level
211 * in the SCLP info, so we skip here for now.
212 */
213 if (max_nested_lvl != info->mnest) {
214 report_skip("No specification on layer aggregation");
215 goto done;
216 }
217
218 /*
219 * MAG up to max_nested_lvl must match the architecture
220 * defined containers.
221 */
222 for (i = 0; i < max_nested_lvl; i++)
223 report(info->mag[CPU_TOPOLOGY_MAX_LEVEL - i - 1] == expected_topo_lvl[i],
224 "MAG %d field match %d == %d",
225 i + 1,
226 info->mag[CPU_TOPOLOGY_MAX_LEVEL - i - 1],
227 expected_topo_lvl[i]);
228
229 /* Above max_nested_lvl the MAG field must be null */
230 for (; i < CPU_TOPOLOGY_MAX_LEVEL; i++)
231 report(info->mag[CPU_TOPOLOGY_MAX_LEVEL - i - 1] == 0,
232 "MAG %d field match %d == %d", i + 1,
233 info->mag[CPU_TOPOLOGY_MAX_LEVEL - i - 1], 0);
234
235 done:
236 report_prefix_pop();
237 }
238
239 /**
240 * stsi_get_sysib:
241 * @info: pointer to the STSI info structure
242 * @sel2: the selector giving the topology level to check
243 *
244 * Fill the sysinfo_15_1_x info structure and check the
245 * SYSIB header.
246 *
247 * Returns instruction validity.
248 */
stsi_get_sysib(struct sysinfo_15_1_x * info,int sel2)249 static int stsi_get_sysib(struct sysinfo_15_1_x *info, int sel2)
250 {
251 int ret;
252
253 report_prefix_pushf("SYSIB");
254
255 ret = stsi(info, 15, 1, sel2);
256
257 if (max_nested_lvl >= sel2) {
258 report(!ret, "Valid instruction");
259 } else {
260 report(ret, "Invalid instruction");
261 }
262
263 report_prefix_pop();
264
265 return ret;
266 }
267
check_cpu(union topology_cpu * cpu,union topology_container * parent)268 static int check_cpu(union topology_cpu *cpu,
269 union topology_container *parent)
270 {
271 report_prefix_pushf("%d:%d:%d:%d", cpu->d, cpu->pp, cpu->type, cpu->origin);
272
273 report(!(cpu->raw[0] & CPUS_TLE_RES_BITS), "reserved bits %016lx",
274 cpu->raw[0] & CPUS_TLE_RES_BITS);
275
276 report(cpu->type == CPU_TYPE_IFL, "type IFL");
277
278 if (cpu->d)
279 report(cpu->pp == POLARIZATION_VERTICAL_HIGH ||
280 cpu->pp == POLARIZATION_HORIZONTAL,
281 "Dedicated CPUs are either horizontally polarized or have high entitlement");
282 else
283 report_skip("Not dedicated");
284
285 report_prefix_pop();
286
287 return __builtin_popcountl(cpu->mask);
288 }
289
check_child_cpus(struct sysinfo_15_1_x * info,union topology_container * cont,union topology_cpu * child,unsigned int * cpus_in_masks)290 static union topology_container *check_child_cpus(struct sysinfo_15_1_x *info,
291 union topology_container *cont,
292 union topology_cpu *child,
293 unsigned int *cpus_in_masks)
294 {
295 void *last = ((void *)info) + info->length;
296 union topology_cpu *prev_cpu = NULL;
297 bool correct_ordering = true;
298 unsigned int cpus = 0;
299 int i;
300
301 for (i = 0; (void *)&child[i] < last && child[i].nl == 0; prev_cpu = &child[i++]) {
302 cpus += check_cpu(&child[i], cont);
303 if (prev_cpu) {
304 if (prev_cpu->type > child[i].type) {
305 report_info("Incorrect ordering wrt type for child %d", i);
306 correct_ordering = false;
307 }
308 if (prev_cpu->type < child[i].type)
309 continue;
310 if (prev_cpu->pp < child[i].pp) {
311 report_info("Incorrect ordering wrt polarization for child %d", i);
312 correct_ordering = false;
313 }
314 if (prev_cpu->pp > child[i].pp)
315 continue;
316 if (!prev_cpu->d && child[i].d) {
317 report_info("Incorrect ordering wrt dedication for child %d", i);
318 correct_ordering = false;
319 }
320 if (prev_cpu->d && !child[i].d)
321 continue;
322 if (prev_cpu->origin > child[i].origin) {
323 report_info("Incorrect ordering wrt origin for child %d", i);
324 correct_ordering = false;
325 }
326 }
327 }
328 report(correct_ordering, "children correctly ordered");
329 report(cpus <= expected_topo_lvl[0], "%d children <= max of %d",
330 cpus, expected_topo_lvl[0]);
331 *cpus_in_masks += cpus;
332
333 return (union topology_container *)&child[i];
334 }
335
336 static union topology_container *check_container(struct sysinfo_15_1_x *info,
337 union topology_container *cont,
338 union topology_entry *child,
339 unsigned int *cpus_in_masks);
340
check_child_containers(struct sysinfo_15_1_x * info,union topology_container * cont,union topology_container * child,unsigned int * cpus_in_masks)341 static union topology_container *check_child_containers(struct sysinfo_15_1_x *info,
342 union topology_container *cont,
343 union topology_container *child,
344 unsigned int *cpus_in_masks)
345 {
346 void *last = ((void *)info) + info->length;
347 union topology_container *entry;
348 int i;
349
350 for (i = 0, entry = child; (void *)entry < last && entry->nl == cont->nl - 1; i++) {
351 entry = check_container(info, entry, (union topology_entry *)(entry + 1),
352 cpus_in_masks);
353 }
354 if (max_nested_lvl == info->mnest)
355 report(i <= expected_topo_lvl[cont->nl - 1], "%d children <= max of %d",
356 i, expected_topo_lvl[cont->nl - 1]);
357
358 return entry;
359 }
360
check_container(struct sysinfo_15_1_x * info,union topology_container * cont,union topology_entry * child,unsigned int * cpus_in_masks)361 static union topology_container *check_container(struct sysinfo_15_1_x *info,
362 union topology_container *cont,
363 union topology_entry *child,
364 unsigned int *cpus_in_masks)
365 {
366 union topology_container *entry;
367
368 report_prefix_pushf("%d", cont->id);
369
370 report(cont->nl - 1 == child->nl, "Level %d one above child level %d",
371 cont->nl, child->nl);
372 report(!(cont->raw & CONTAINER_TLE_RES_BITS), "reserved bits %016lx",
373 cont->raw & CONTAINER_TLE_RES_BITS);
374
375 if (cont->nl > 1)
376 entry = check_child_containers(info, cont, &child->container, cpus_in_masks);
377 else
378 entry = check_child_cpus(info, cont, &child->cpu, cpus_in_masks);
379
380 report_prefix_pop();
381 return entry;
382 }
383
check_topology_list(struct sysinfo_15_1_x * info,int sel2)384 static void check_topology_list(struct sysinfo_15_1_x *info, int sel2)
385 {
386 union topology_container dummy = { .nl = sel2, .id = 0 };
387 unsigned int cpus_in_masks = 0;
388
389 report_prefix_push("TLE");
390
391 check_container(info, &dummy, info->tle, &cpus_in_masks);
392 report(cpus_in_masks == number_of_cpus,
393 "Number of CPUs %d equals %d CPUs in masks",
394 number_of_cpus, cpus_in_masks);
395
396 report_prefix_pop();
397 }
398
399 /**
400 * check_sysinfo_15_1_x:
401 * @info: pointer to the STSI info structure
402 * @sel2: the selector giving the topology level to check
403 *
404 * Check if the validity of the STSI instruction and then
405 * calls specific checks on the information buffer.
406 */
check_sysinfo_15_1_x(struct sysinfo_15_1_x * info,int sel2)407 static void check_sysinfo_15_1_x(struct sysinfo_15_1_x *info, int sel2)
408 {
409 int ret;
410 int cc;
411 unsigned long rc;
412
413 report_prefix_pushf("15_1_%d", sel2);
414
415 ret = stsi_get_sysib(info, sel2);
416 if (ret) {
417 report_skip("Selector 2 not supported by architecture");
418 goto end;
419 }
420
421 report_prefix_pushf("H");
422 cc = ptf(PTF_REQ_HORIZONTAL, &rc);
423 if (cc != 0 && rc != PTF_ERR_ALRDY_POLARIZED) {
424 report_fail("Unable to set horizontal polarization");
425 goto vertical;
426 }
427
428 stsi_check_header(info, sel2);
429 check_topology_list(info, sel2);
430
431 vertical:
432 report_prefix_pop();
433 report_prefix_pushf("V");
434
435 cc = ptf(PTF_REQ_VERTICAL, &rc);
436 if (cc != 0 && rc != PTF_ERR_ALRDY_POLARIZED) {
437 report_fail("Unable to set vertical polarization");
438 goto end;
439 }
440
441 stsi_check_header(info, sel2);
442 check_topology_list(info, sel2);
443 report_prefix_pop();
444
445 end:
446 report_prefix_pop();
447 }
448
449 /*
450 * The Maximum Nested level is given by SCLP READ_SCP_INFO if the MNEST facility
451 * is available.
452 * If the MNEST facility is not available, sclp_get_stsi_mnest returns 0 and the
453 * Maximum Nested level is 2
454 */
455 #define S390_DEFAULT_MNEST 2
sclp_get_mnest(void)456 static int sclp_get_mnest(void)
457 {
458 return sclp_get_stsi_mnest() ?: S390_DEFAULT_MNEST;
459 }
460
expected_num_cpus(void)461 static int expected_num_cpus(void)
462 {
463 int i;
464 int ncpus = 1;
465
466 for (i = 0; i < CPU_TOPOLOGY_MAX_LEVEL; i++)
467 ncpus *= expected_topo_lvl[i] ?: 1;
468
469 return ncpus;
470 }
471
472 /**
473 * test_stsi:
474 *
475 * Retrieves the maximum nested topology level supported by the architecture
476 * and the number of CPUs.
477 * Calls the checking for the STSI instruction in sel2 reverse level order
478 * from 6 (CPU_TOPOLOGY_MAX_LEVEL) to 2 to have the most interesting level,
479 * the one triggering a topology-change-report-pending condition, level 2,
480 * at the end of the report.
481 *
482 */
test_stsi(void)483 static void test_stsi(void)
484 {
485 int sel2;
486
487 max_cpus = expected_num_cpus();
488 report_info("Architecture max CPUs: %d", max_cpus);
489
490 max_nested_lvl = sclp_get_mnest();
491 report_info("SCLP maximum nested level : %d", max_nested_lvl);
492
493 number_of_cpus = sclp_get_cpu_num();
494 report_info("SCLP number of CPU: %d", number_of_cpus);
495
496 /* STSI selector 2 can takes values between 2 and 6 */
497 for (sel2 = 6; sel2 >= 2; sel2--)
498 check_sysinfo_15_1_x((struct sysinfo_15_1_x *)pagebuf, sel2);
499 }
500
501 /**
502 * parse_topology_args:
503 * @argc: number of arguments
504 * @argv: argument array
505 *
506 * This function initialize the architecture topology levels
507 * which should be the same as the one provided by the hypervisor.
508 *
509 * We use the current names found in IBM/Z literature, Linux and QEMU:
510 * cores, sockets/packages, books, drawers and nodes to facilitate the
511 * human machine interface but store the result in a machine abstract
512 * array of architecture topology levels.
513 * Note that when QEMU uses socket as a name for the topology level 1
514 * Linux uses package or physical_package.
515 */
parse_topology_args(int argc,char ** argv)516 static void parse_topology_args(int argc, char **argv)
517 {
518 int i;
519 static const char * const levels[] = { "cores", "sockets",
520 "books", "drawers" };
521
522 for (i = 1; i < argc; i++) {
523 char *flag = argv[i];
524 int level;
525
526 if (flag[0] != '-')
527 report_abort("Argument is expected to begin with '-'");
528 flag++;
529 for (level = 0; level < ARRAY_SIZE(levels); level++) {
530 if (!strcmp(levels[level], flag))
531 break;
532 }
533 if (level == ARRAY_SIZE(levels))
534 report_abort("Unknown parameter %s", flag);
535
536 expected_topo_lvl[level] = atol(argv[++i]);
537 report_info("%s: %d", levels[level], expected_topo_lvl[level]);
538 }
539 }
540
541 static struct {
542 const char *name;
543 void (*func)(void);
544 } tests[] = {
545 { "PTF", test_ptf },
546 { "STSI", test_stsi },
547 { NULL, NULL }
548 };
549
main(int argc,char * argv[])550 int main(int argc, char *argv[])
551 {
552 int i;
553
554 report_prefix_push("CPU Topology");
555
556 parse_topology_args(argc, argv);
557
558 if (!test_facility(11)) {
559 report_skip("Topology facility not present");
560 goto end;
561 }
562
563 report_info("Virtual machine level %ld", stsi_get_fc());
564
565 for (i = 0; tests[i].name; i++) {
566 report_prefix_push(tests[i].name);
567 tests[i].func();
568 report_prefix_pop();
569 }
570
571 end:
572 report_prefix_pop();
573 return report_summary();
574 }
575