1 /*
2 * Copyright IBM Corp. 2007, 2011
3 * Author(s): Heiko Carstens <heiko.carstens@de.ibm.com>
4 */
5
6 #define KMSG_COMPONENT "cpu"
7 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
8
9 #include <linux/workqueue.h>
10 #include <linux/cpuset.h>
11 #include <linux/device.h>
12 #include <linux/export.h>
13 #include <linux/kernel.h>
14 #include <linux/sched.h>
15 #include <linux/delay.h>
16 #include <linux/init.h>
17 #include <linux/slab.h>
18 #include <linux/cpu.h>
19 #include <linux/smp.h>
20 #include <linux/mm.h>
21 #include <linux/nodemask.h>
22 #include <linux/node.h>
23 #include <asm/sysinfo.h>
24 #include <asm/numa.h>
25
26 #define PTF_HORIZONTAL (0UL)
27 #define PTF_VERTICAL (1UL)
28 #define PTF_CHECK (2UL)
29
30 struct mask_info {
31 struct mask_info *next;
32 unsigned char id;
33 cpumask_t mask;
34 };
35
36 static void set_topology_timer(void);
37 static void topology_work_fn(struct work_struct *work);
38 static struct sysinfo_15_1_x *tl_info;
39
40 static DECLARE_WORK(topology_work, topology_work_fn);
41
42 /*
43 * Socket/Book linked lists and per_cpu(cpu_topology) updates are
44 * protected by "sched_domains_mutex".
45 */
46 static struct mask_info socket_info;
47 static struct mask_info book_info;
48 static struct mask_info drawer_info;
49
50 DEFINE_PER_CPU(struct cpu_topology_s390, cpu_topology);
51 EXPORT_PER_CPU_SYMBOL_GPL(cpu_topology);
52
cpu_group_map(struct mask_info * info,unsigned int cpu)53 static cpumask_t cpu_group_map(struct mask_info *info, unsigned int cpu)
54 {
55 cpumask_t mask;
56
57 cpumask_copy(&mask, cpumask_of(cpu));
58 if (!MACHINE_HAS_TOPOLOGY)
59 return mask;
60 for (; info; info = info->next) {
61 if (cpumask_test_cpu(cpu, &info->mask))
62 return info->mask;
63 }
64 return mask;
65 }
66
cpu_thread_map(unsigned int cpu)67 static cpumask_t cpu_thread_map(unsigned int cpu)
68 {
69 cpumask_t mask;
70 int i;
71
72 cpumask_copy(&mask, cpumask_of(cpu));
73 if (!MACHINE_HAS_TOPOLOGY)
74 return mask;
75 cpu -= cpu % (smp_cpu_mtid + 1);
76 for (i = 0; i <= smp_cpu_mtid; i++)
77 if (cpu_present(cpu + i))
78 cpumask_set_cpu(cpu + i, &mask);
79 return mask;
80 }
81
add_cpus_to_mask(struct topology_core * tl_core,struct mask_info * drawer,struct mask_info * book,struct mask_info * socket)82 static void add_cpus_to_mask(struct topology_core *tl_core,
83 struct mask_info *drawer,
84 struct mask_info *book,
85 struct mask_info *socket)
86 {
87 struct cpu_topology_s390 *topo;
88 unsigned int core;
89
90 for_each_set_bit(core, &tl_core->mask[0], TOPOLOGY_CORE_BITS) {
91 unsigned int rcore;
92 int lcpu, i;
93
94 rcore = TOPOLOGY_CORE_BITS - 1 - core + tl_core->origin;
95 lcpu = smp_find_processor_id(rcore << smp_cpu_mt_shift);
96 if (lcpu < 0)
97 continue;
98 for (i = 0; i <= smp_cpu_mtid; i++) {
99 topo = &per_cpu(cpu_topology, lcpu + i);
100 topo->drawer_id = drawer->id;
101 topo->book_id = book->id;
102 topo->socket_id = socket->id;
103 topo->core_id = rcore;
104 topo->thread_id = lcpu + i;
105 cpumask_set_cpu(lcpu + i, &drawer->mask);
106 cpumask_set_cpu(lcpu + i, &book->mask);
107 cpumask_set_cpu(lcpu + i, &socket->mask);
108 smp_cpu_set_polarization(lcpu + i, tl_core->pp);
109 }
110 }
111 }
112
clear_masks(void)113 static void clear_masks(void)
114 {
115 struct mask_info *info;
116
117 info = &socket_info;
118 while (info) {
119 cpumask_clear(&info->mask);
120 info = info->next;
121 }
122 info = &book_info;
123 while (info) {
124 cpumask_clear(&info->mask);
125 info = info->next;
126 }
127 info = &drawer_info;
128 while (info) {
129 cpumask_clear(&info->mask);
130 info = info->next;
131 }
132 }
133
next_tle(union topology_entry * tle)134 static union topology_entry *next_tle(union topology_entry *tle)
135 {
136 if (!tle->nl)
137 return (union topology_entry *)((struct topology_core *)tle + 1);
138 return (union topology_entry *)((struct topology_container *)tle + 1);
139 }
140
tl_to_masks(struct sysinfo_15_1_x * info)141 static void tl_to_masks(struct sysinfo_15_1_x *info)
142 {
143 struct mask_info *socket = &socket_info;
144 struct mask_info *book = &book_info;
145 struct mask_info *drawer = &drawer_info;
146 union topology_entry *tle, *end;
147
148 clear_masks();
149 tle = info->tle;
150 end = (union topology_entry *)((unsigned long)info + info->length);
151 while (tle < end) {
152 switch (tle->nl) {
153 case 3:
154 drawer = drawer->next;
155 drawer->id = tle->container.id;
156 break;
157 case 2:
158 book = book->next;
159 book->id = tle->container.id;
160 break;
161 case 1:
162 socket = socket->next;
163 socket->id = tle->container.id;
164 break;
165 case 0:
166 add_cpus_to_mask(&tle->cpu, drawer, book, socket);
167 break;
168 default:
169 clear_masks();
170 return;
171 }
172 tle = next_tle(tle);
173 }
174 }
175
topology_update_polarization_simple(void)176 static void topology_update_polarization_simple(void)
177 {
178 int cpu;
179
180 mutex_lock(&smp_cpu_state_mutex);
181 for_each_possible_cpu(cpu)
182 smp_cpu_set_polarization(cpu, POLARIZATION_HRZ);
183 mutex_unlock(&smp_cpu_state_mutex);
184 }
185
ptf(unsigned long fc)186 static int ptf(unsigned long fc)
187 {
188 int rc;
189
190 asm volatile(
191 " .insn rre,0xb9a20000,%1,%1\n"
192 " ipm %0\n"
193 " srl %0,28\n"
194 : "=d" (rc)
195 : "d" (fc) : "cc");
196 return rc;
197 }
198
topology_set_cpu_management(int fc)199 int topology_set_cpu_management(int fc)
200 {
201 int cpu, rc;
202
203 if (!MACHINE_HAS_TOPOLOGY)
204 return -EOPNOTSUPP;
205 if (fc)
206 rc = ptf(PTF_VERTICAL);
207 else
208 rc = ptf(PTF_HORIZONTAL);
209 if (rc)
210 return -EBUSY;
211 for_each_possible_cpu(cpu)
212 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
213 return rc;
214 }
215
update_cpu_masks(void)216 static void update_cpu_masks(void)
217 {
218 struct cpu_topology_s390 *topo;
219 int cpu;
220
221 for_each_possible_cpu(cpu) {
222 topo = &per_cpu(cpu_topology, cpu);
223 topo->thread_mask = cpu_thread_map(cpu);
224 topo->core_mask = cpu_group_map(&socket_info, cpu);
225 topo->book_mask = cpu_group_map(&book_info, cpu);
226 topo->drawer_mask = cpu_group_map(&drawer_info, cpu);
227 if (!MACHINE_HAS_TOPOLOGY) {
228 topo->thread_id = cpu;
229 topo->core_id = cpu;
230 topo->socket_id = cpu;
231 topo->book_id = cpu;
232 topo->drawer_id = cpu;
233 }
234 }
235 numa_update_cpu_topology();
236 }
237
store_topology(struct sysinfo_15_1_x * info)238 void store_topology(struct sysinfo_15_1_x *info)
239 {
240 stsi(info, 15, 1, min(topology_max_mnest, 4));
241 }
242
arch_update_cpu_topology(void)243 int arch_update_cpu_topology(void)
244 {
245 struct sysinfo_15_1_x *info = tl_info;
246 struct device *dev;
247 int cpu, rc = 0;
248
249 if (MACHINE_HAS_TOPOLOGY) {
250 rc = 1;
251 store_topology(info);
252 tl_to_masks(info);
253 }
254 update_cpu_masks();
255 if (!MACHINE_HAS_TOPOLOGY)
256 topology_update_polarization_simple();
257 for_each_online_cpu(cpu) {
258 dev = get_cpu_device(cpu);
259 kobject_uevent(&dev->kobj, KOBJ_CHANGE);
260 }
261 return rc;
262 }
263
topology_work_fn(struct work_struct * work)264 static void topology_work_fn(struct work_struct *work)
265 {
266 rebuild_sched_domains();
267 }
268
topology_schedule_update(void)269 void topology_schedule_update(void)
270 {
271 schedule_work(&topology_work);
272 }
273
topology_timer_fn(unsigned long ignored)274 static void topology_timer_fn(unsigned long ignored)
275 {
276 if (ptf(PTF_CHECK))
277 topology_schedule_update();
278 set_topology_timer();
279 }
280
281 static struct timer_list topology_timer =
282 TIMER_DEFERRED_INITIALIZER(topology_timer_fn, 0, 0);
283
284 static atomic_t topology_poll = ATOMIC_INIT(0);
285
set_topology_timer(void)286 static void set_topology_timer(void)
287 {
288 if (atomic_add_unless(&topology_poll, -1, 0))
289 mod_timer(&topology_timer, jiffies + HZ / 10);
290 else
291 mod_timer(&topology_timer, jiffies + HZ * 60);
292 }
293
topology_expect_change(void)294 void topology_expect_change(void)
295 {
296 if (!MACHINE_HAS_TOPOLOGY)
297 return;
298 /* This is racy, but it doesn't matter since it is just a heuristic.
299 * Worst case is that we poll in a higher frequency for a bit longer.
300 */
301 if (atomic_read(&topology_poll) > 60)
302 return;
303 atomic_add(60, &topology_poll);
304 set_topology_timer();
305 }
306
307 static int cpu_management;
308
dispatching_show(struct device * dev,struct device_attribute * attr,char * buf)309 static ssize_t dispatching_show(struct device *dev,
310 struct device_attribute *attr,
311 char *buf)
312 {
313 ssize_t count;
314
315 mutex_lock(&smp_cpu_state_mutex);
316 count = sprintf(buf, "%d\n", cpu_management);
317 mutex_unlock(&smp_cpu_state_mutex);
318 return count;
319 }
320
dispatching_store(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)321 static ssize_t dispatching_store(struct device *dev,
322 struct device_attribute *attr,
323 const char *buf,
324 size_t count)
325 {
326 int val, rc;
327 char delim;
328
329 if (sscanf(buf, "%d %c", &val, &delim) != 1)
330 return -EINVAL;
331 if (val != 0 && val != 1)
332 return -EINVAL;
333 rc = 0;
334 get_online_cpus();
335 mutex_lock(&smp_cpu_state_mutex);
336 if (cpu_management == val)
337 goto out;
338 rc = topology_set_cpu_management(val);
339 if (rc)
340 goto out;
341 cpu_management = val;
342 topology_expect_change();
343 out:
344 mutex_unlock(&smp_cpu_state_mutex);
345 put_online_cpus();
346 return rc ? rc : count;
347 }
348 static DEVICE_ATTR(dispatching, 0644, dispatching_show,
349 dispatching_store);
350
cpu_polarization_show(struct device * dev,struct device_attribute * attr,char * buf)351 static ssize_t cpu_polarization_show(struct device *dev,
352 struct device_attribute *attr, char *buf)
353 {
354 int cpu = dev->id;
355 ssize_t count;
356
357 mutex_lock(&smp_cpu_state_mutex);
358 switch (smp_cpu_get_polarization(cpu)) {
359 case POLARIZATION_HRZ:
360 count = sprintf(buf, "horizontal\n");
361 break;
362 case POLARIZATION_VL:
363 count = sprintf(buf, "vertical:low\n");
364 break;
365 case POLARIZATION_VM:
366 count = sprintf(buf, "vertical:medium\n");
367 break;
368 case POLARIZATION_VH:
369 count = sprintf(buf, "vertical:high\n");
370 break;
371 default:
372 count = sprintf(buf, "unknown\n");
373 break;
374 }
375 mutex_unlock(&smp_cpu_state_mutex);
376 return count;
377 }
378 static DEVICE_ATTR(polarization, 0444, cpu_polarization_show, NULL);
379
380 static struct attribute *topology_cpu_attrs[] = {
381 &dev_attr_polarization.attr,
382 NULL,
383 };
384
385 static struct attribute_group topology_cpu_attr_group = {
386 .attrs = topology_cpu_attrs,
387 };
388
topology_cpu_init(struct cpu * cpu)389 int topology_cpu_init(struct cpu *cpu)
390 {
391 return sysfs_create_group(&cpu->dev.kobj, &topology_cpu_attr_group);
392 }
393
cpu_thread_mask(int cpu)394 static const struct cpumask *cpu_thread_mask(int cpu)
395 {
396 return &per_cpu(cpu_topology, cpu).thread_mask;
397 }
398
399
cpu_coregroup_mask(int cpu)400 const struct cpumask *cpu_coregroup_mask(int cpu)
401 {
402 return &per_cpu(cpu_topology, cpu).core_mask;
403 }
404
cpu_book_mask(int cpu)405 static const struct cpumask *cpu_book_mask(int cpu)
406 {
407 return &per_cpu(cpu_topology, cpu).book_mask;
408 }
409
cpu_drawer_mask(int cpu)410 static const struct cpumask *cpu_drawer_mask(int cpu)
411 {
412 return &per_cpu(cpu_topology, cpu).drawer_mask;
413 }
414
415 static struct sched_domain_topology_level s390_topology[] = {
416 { cpu_thread_mask, cpu_smt_flags, SD_INIT_NAME(SMT) },
417 { cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) },
418 { cpu_book_mask, SD_INIT_NAME(BOOK) },
419 { cpu_drawer_mask, SD_INIT_NAME(DRAWER) },
420 { cpu_cpu_mask, SD_INIT_NAME(DIE) },
421 { NULL, },
422 };
423
alloc_masks(struct sysinfo_15_1_x * info,struct mask_info * mask,int offset)424 static void __init alloc_masks(struct sysinfo_15_1_x *info,
425 struct mask_info *mask, int offset)
426 {
427 int i, nr_masks;
428
429 nr_masks = info->mag[TOPOLOGY_NR_MAG - offset];
430 for (i = 0; i < info->mnest - offset; i++)
431 nr_masks *= info->mag[TOPOLOGY_NR_MAG - offset - 1 - i];
432 nr_masks = max(nr_masks, 1);
433 for (i = 0; i < nr_masks; i++) {
434 mask->next = kzalloc(sizeof(*mask->next), GFP_KERNEL);
435 mask = mask->next;
436 }
437 }
438
s390_topology_init(void)439 static int __init s390_topology_init(void)
440 {
441 struct sysinfo_15_1_x *info;
442 int i;
443
444 set_sched_topology(s390_topology);
445 if (!MACHINE_HAS_TOPOLOGY)
446 return 0;
447 tl_info = (struct sysinfo_15_1_x *)__get_free_page(GFP_KERNEL);
448 info = tl_info;
449 store_topology(info);
450 pr_info("The CPU configuration topology of the machine is:");
451 for (i = 0; i < TOPOLOGY_NR_MAG; i++)
452 printk(KERN_CONT " %d", info->mag[i]);
453 printk(KERN_CONT " / %d\n", info->mnest);
454 alloc_masks(info, &socket_info, 1);
455 alloc_masks(info, &book_info, 2);
456 alloc_masks(info, &drawer_info, 3);
457 return 0;
458 }
459 early_initcall(s390_topology_init);
460
topology_init(void)461 static int __init topology_init(void)
462 {
463 if (MACHINE_HAS_TOPOLOGY)
464 set_topology_timer();
465 else
466 topology_update_polarization_simple();
467 return device_create_file(cpu_subsys.dev_root, &dev_attr_dispatching);
468 }
469 device_initcall(topology_init);
470