Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * SMP boot-related support | |
3 | * | |
82975115 | 4 | * Copyright (C) 1998-2003, 2005 Hewlett-Packard Co |
1da177e4 | 5 | * David Mosberger-Tang <davidm@hpl.hp.com> |
e927ecb0 SS |
6 | * Copyright (C) 2001, 2004-2005 Intel Corp |
7 | * Rohit Seth <rohit.seth@intel.com> | |
8 | * Suresh Siddha <suresh.b.siddha@intel.com> | |
9 | * Gordon Jin <gordon.jin@intel.com> | |
10 | * Ashok Raj <ashok.raj@intel.com> | |
1da177e4 LT |
11 | * |
12 | * 01/05/16 Rohit Seth <rohit.seth@intel.com> Moved SMP booting functions from smp.c to here. | |
13 | * 01/04/27 David Mosberger <davidm@hpl.hp.com> Added ITC synching code. | |
14 | * 02/07/31 David Mosberger <davidm@hpl.hp.com> Switch over to hotplug-CPU boot-sequence. | |
15 | * smp_boot_cpus()/smp_commence() is replaced by | |
16 | * smp_prepare_cpus()/__cpu_up()/smp_cpus_done(). | |
b8d8b883 | 17 | * 04/06/21 Ashok Raj <ashok.raj@intel.com> Added CPU Hotplug Support |
e927ecb0 SS |
18 | * 04/12/26 Jin Gordon <gordon.jin@intel.com> |
19 | * 04/12/26 Rohit Seth <rohit.seth@intel.com> | |
20 | * Add multi-threading and multi-core detection | |
21 | * 05/01/30 Suresh Siddha <suresh.b.siddha@intel.com> | |
22 | * Setup cpu_sibling_map and cpu_core_map | |
1da177e4 | 23 | */ |
1da177e4 LT |
24 | |
25 | #include <linux/module.h> | |
26 | #include <linux/acpi.h> | |
27 | #include <linux/bootmem.h> | |
28 | #include <linux/cpu.h> | |
29 | #include <linux/delay.h> | |
30 | #include <linux/init.h> | |
31 | #include <linux/interrupt.h> | |
32 | #include <linux/irq.h> | |
33 | #include <linux/kernel.h> | |
34 | #include <linux/kernel_stat.h> | |
35 | #include <linux/mm.h> | |
36 | #include <linux/notifier.h> | |
37 | #include <linux/smp.h> | |
1da177e4 LT |
38 | #include <linux/spinlock.h> |
39 | #include <linux/efi.h> | |
40 | #include <linux/percpu.h> | |
41 | #include <linux/bitops.h> | |
42 | ||
60063497 | 43 | #include <linux/atomic.h> |
1da177e4 LT |
44 | #include <asm/cache.h> |
45 | #include <asm/current.h> | |
46 | #include <asm/delay.h> | |
1da177e4 LT |
47 | #include <asm/io.h> |
48 | #include <asm/irq.h> | |
49 | #include <asm/machvec.h> | |
50 | #include <asm/mca.h> | |
51 | #include <asm/page.h> | |
52 | #include <asm/pgalloc.h> | |
53 | #include <asm/pgtable.h> | |
54 | #include <asm/processor.h> | |
55 | #include <asm/ptrace.h> | |
56 | #include <asm/sal.h> | |
1da177e4 LT |
57 | #include <asm/tlbflush.h> |
58 | #include <asm/unistd.h> | |
6e9de181 | 59 | #include <asm/sn/arch.h> |
1da177e4 LT |
60 | |
61 | #define SMP_DEBUG 0 | |
62 | ||
63 | #if SMP_DEBUG | |
64 | #define Dprintk(x...) printk(x) | |
65 | #else | |
66 | #define Dprintk(x...) | |
67 | #endif | |
68 | ||
b8d8b883 | 69 | #ifdef CONFIG_HOTPLUG_CPU |
ff741906 AR |
70 | #ifdef CONFIG_PERMIT_BSP_REMOVE |
71 | #define bsp_remove_ok 1 | |
72 | #else | |
73 | #define bsp_remove_ok 0 | |
74 | #endif | |
75 | ||
b8d8b883 AR |
76 | /* |
77 | * Global array allocated for NR_CPUS at boot time | |
78 | */ | |
79 | struct sal_to_os_boot sal_boot_rendez_state[NR_CPUS]; | |
80 | ||
81 | /* | |
82 | * start_ap in head.S uses this to store current booting cpu | |
83 | * info. | |
84 | */ | |
85 | struct sal_to_os_boot *sal_state_for_booting_cpu = &sal_boot_rendez_state[0]; | |
86 | ||
87 | #define set_brendez_area(x) (sal_state_for_booting_cpu = &sal_boot_rendez_state[(x)]); | |
88 | ||
b8d8b883 | 89 | #else |
b8d8b883 AR |
90 | #define set_brendez_area(x) |
91 | #endif | |
92 | ||
1da177e4 LT |
93 | |
94 | /* | |
95 | * ITC synchronization related stuff: | |
96 | */ | |
ff741906 | 97 | #define MASTER (0) |
1da177e4 LT |
98 | #define SLAVE (SMP_CACHE_BYTES/8) |
99 | ||
100 | #define NUM_ROUNDS 64 /* magic value */ | |
101 | #define NUM_ITERS 5 /* likewise */ | |
102 | ||
103 | static DEFINE_SPINLOCK(itc_sync_lock); | |
104 | static volatile unsigned long go[SLAVE + 1]; | |
105 | ||
106 | #define DEBUG_ITC_SYNC 0 | |
107 | ||
1da177e4 LT |
108 | extern void start_ap (void); |
109 | extern unsigned long ia64_iobase; | |
110 | ||
36c8b586 | 111 | struct task_struct *task_for_booting_cpu; |
1da177e4 LT |
112 | |
113 | /* | |
114 | * State for each CPU | |
115 | */ | |
116 | DEFINE_PER_CPU(int, cpu_state); | |
117 | ||
e927ecb0 | 118 | cpumask_t cpu_core_map[NR_CPUS] __cacheline_aligned; |
42aca483 | 119 | EXPORT_SYMBOL(cpu_core_map); |
d5a7430d MT |
120 | DEFINE_PER_CPU_SHARED_ALIGNED(cpumask_t, cpu_sibling_map); |
121 | EXPORT_PER_CPU_SYMBOL(cpu_sibling_map); | |
122 | ||
e927ecb0 | 123 | int smp_num_siblings = 1; |
e927ecb0 | 124 | |
1da177e4 LT |
125 | /* which logical CPU number maps to which CPU (physical APIC ID) */ |
126 | volatile int ia64_cpu_to_sapicid[NR_CPUS]; | |
127 | EXPORT_SYMBOL(ia64_cpu_to_sapicid); | |
128 | ||
5eda7861 | 129 | static cpumask_t cpu_callin_map; |
1da177e4 LT |
130 | |
131 | struct smp_boot_data smp_boot_data __initdata; | |
132 | ||
133 | unsigned long ap_wakeup_vector = -1; /* External Int use to wakeup APs */ | |
134 | ||
135 | char __initdata no_int_routing; | |
136 | ||
137 | unsigned char smp_int_redirect; /* are INT and IPI redirectable by the chipset? */ | |
138 | ||
ff741906 AR |
139 | #ifdef CONFIG_FORCE_CPEI_RETARGET |
140 | #define CPEI_OVERRIDE_DEFAULT (1) | |
141 | #else | |
142 | #define CPEI_OVERRIDE_DEFAULT (0) | |
143 | #endif | |
144 | ||
145 | unsigned int force_cpei_retarget = CPEI_OVERRIDE_DEFAULT; | |
146 | ||
147 | static int __init | |
148 | cmdl_force_cpei(char *str) | |
149 | { | |
150 | int value=0; | |
151 | ||
152 | get_option (&str, &value); | |
153 | force_cpei_retarget = value; | |
154 | ||
155 | return 1; | |
156 | } | |
157 | ||
158 | __setup("force_cpei=", cmdl_force_cpei); | |
159 | ||
1da177e4 LT |
160 | static int __init |
161 | nointroute (char *str) | |
162 | { | |
163 | no_int_routing = 1; | |
164 | printk ("no_int_routing on\n"); | |
165 | return 1; | |
166 | } | |
167 | ||
168 | __setup("nointroute", nointroute); | |
169 | ||
ff741906 AR |
170 | static void fix_b0_for_bsp(void) |
171 | { | |
172 | #ifdef CONFIG_HOTPLUG_CPU | |
173 | int cpuid; | |
174 | static int fix_bsp_b0 = 1; | |
175 | ||
176 | cpuid = smp_processor_id(); | |
177 | ||
178 | /* | |
179 | * Cache the b0 value on the first AP that comes up | |
180 | */ | |
181 | if (!(fix_bsp_b0 && cpuid)) | |
182 | return; | |
183 | ||
184 | sal_boot_rendez_state[0].br[0] = sal_boot_rendez_state[cpuid].br[0]; | |
185 | printk ("Fixed BSP b0 value from CPU %d\n", cpuid); | |
186 | ||
187 | fix_bsp_b0 = 0; | |
188 | #endif | |
189 | } | |
190 | ||
1da177e4 LT |
191 | void |
192 | sync_master (void *arg) | |
193 | { | |
194 | unsigned long flags, i; | |
195 | ||
196 | go[MASTER] = 0; | |
197 | ||
198 | local_irq_save(flags); | |
199 | { | |
200 | for (i = 0; i < NUM_ROUNDS*NUM_ITERS; ++i) { | |
82975115 DMT |
201 | while (!go[MASTER]) |
202 | cpu_relax(); | |
1da177e4 LT |
203 | go[MASTER] = 0; |
204 | go[SLAVE] = ia64_get_itc(); | |
205 | } | |
206 | } | |
207 | local_irq_restore(flags); | |
208 | } | |
209 | ||
210 | /* | |
211 | * Return the number of cycles by which our itc differs from the itc on the master | |
212 | * (time-keeper) CPU. A positive number indicates our itc is ahead of the master, | |
213 | * negative that it is behind. | |
214 | */ | |
215 | static inline long | |
216 | get_delta (long *rt, long *master) | |
217 | { | |
218 | unsigned long best_t0 = 0, best_t1 = ~0UL, best_tm = 0; | |
219 | unsigned long tcenter, t0, t1, tm; | |
220 | long i; | |
221 | ||
222 | for (i = 0; i < NUM_ITERS; ++i) { | |
223 | t0 = ia64_get_itc(); | |
224 | go[MASTER] = 1; | |
82975115 DMT |
225 | while (!(tm = go[SLAVE])) |
226 | cpu_relax(); | |
1da177e4 LT |
227 | go[SLAVE] = 0; |
228 | t1 = ia64_get_itc(); | |
229 | ||
230 | if (t1 - t0 < best_t1 - best_t0) | |
231 | best_t0 = t0, best_t1 = t1, best_tm = tm; | |
232 | } | |
233 | ||
234 | *rt = best_t1 - best_t0; | |
235 | *master = best_tm - best_t0; | |
236 | ||
237 | /* average best_t0 and best_t1 without overflow: */ | |
238 | tcenter = (best_t0/2 + best_t1/2); | |
239 | if (best_t0 % 2 + best_t1 % 2 == 2) | |
240 | ++tcenter; | |
241 | return tcenter - best_tm; | |
242 | } | |
243 | ||
244 | /* | |
245 | * Synchronize ar.itc of the current (slave) CPU with the ar.itc of the MASTER CPU | |
246 | * (normally the time-keeper CPU). We use a closed loop to eliminate the possibility of | |
247 | * unaccounted-for errors (such as getting a machine check in the middle of a calibration | |
248 | * step). The basic idea is for the slave to ask the master what itc value it has and to | |
249 | * read its own itc before and after the master responds. Each iteration gives us three | |
250 | * timestamps: | |
251 | * | |
252 | * slave master | |
253 | * | |
254 | * t0 ---\ | |
255 | * ---\ | |
256 | * ---> | |
257 | * tm | |
258 | * /--- | |
259 | * /--- | |
260 | * t1 <--- | |
261 | * | |
262 | * | |
263 | * The goal is to adjust the slave's ar.itc such that tm falls exactly half-way between t0 | |
264 | * and t1. If we achieve this, the clocks are synchronized provided the interconnect | |
265 | * between the slave and the master is symmetric. Even if the interconnect were | |
266 | * asymmetric, we would still know that the synchronization error is smaller than the | |
267 | * roundtrip latency (t0 - t1). | |
268 | * | |
269 | * When the interconnect is quiet and symmetric, this lets us synchronize the itc to | |
270 | * within one or two cycles. However, we can only *guarantee* that the synchronization is | |
271 | * accurate to within a round-trip time, which is typically in the range of several | |
272 | * hundred cycles (e.g., ~500 cycles). In practice, this means that the itc's are usually | |
273 | * almost perfectly synchronized, but we shouldn't assume that the accuracy is much better | |
274 | * than half a micro second or so. | |
275 | */ | |
276 | void | |
277 | ia64_sync_itc (unsigned int master) | |
278 | { | |
279 | long i, delta, adj, adjust_latency = 0, done = 0; | |
280 | unsigned long flags, rt, master_time_stamp, bound; | |
281 | #if DEBUG_ITC_SYNC | |
282 | struct { | |
283 | long rt; /* roundtrip time */ | |
284 | long master; /* master's timestamp */ | |
285 | long diff; /* difference between midpoint and master's timestamp */ | |
286 | long lat; /* estimate of itc adjustment latency */ | |
287 | } t[NUM_ROUNDS]; | |
288 | #endif | |
289 | ||
290 | /* | |
291 | * Make sure local timer ticks are disabled while we sync. If | |
292 | * they were enabled, we'd have to worry about nasty issues | |
293 | * like setting the ITC ahead of (or a long time before) the | |
294 | * next scheduled tick. | |
295 | */ | |
296 | BUG_ON((ia64_get_itv() & (1 << 16)) == 0); | |
297 | ||
298 | go[MASTER] = 1; | |
299 | ||
8691e5a8 | 300 | if (smp_call_function_single(master, sync_master, NULL, 0) < 0) { |
1da177e4 LT |
301 | printk(KERN_ERR "sync_itc: failed to get attention of CPU %u!\n", master); |
302 | return; | |
303 | } | |
304 | ||
82975115 DMT |
305 | while (go[MASTER]) |
306 | cpu_relax(); /* wait for master to be ready */ | |
1da177e4 LT |
307 | |
308 | spin_lock_irqsave(&itc_sync_lock, flags); | |
309 | { | |
310 | for (i = 0; i < NUM_ROUNDS; ++i) { | |
311 | delta = get_delta(&rt, &master_time_stamp); | |
312 | if (delta == 0) { | |
313 | done = 1; /* let's lock on to this... */ | |
314 | bound = rt; | |
315 | } | |
316 | ||
317 | if (!done) { | |
318 | if (i > 0) { | |
319 | adjust_latency += -delta; | |
320 | adj = -delta + adjust_latency/4; | |
321 | } else | |
322 | adj = -delta; | |
323 | ||
324 | ia64_set_itc(ia64_get_itc() + adj); | |
325 | } | |
326 | #if DEBUG_ITC_SYNC | |
327 | t[i].rt = rt; | |
328 | t[i].master = master_time_stamp; | |
329 | t[i].diff = delta; | |
330 | t[i].lat = adjust_latency/4; | |
331 | #endif | |
332 | } | |
333 | } | |
334 | spin_unlock_irqrestore(&itc_sync_lock, flags); | |
335 | ||
336 | #if DEBUG_ITC_SYNC | |
337 | for (i = 0; i < NUM_ROUNDS; ++i) | |
338 | printk("rt=%5ld master=%5ld diff=%5ld adjlat=%5ld\n", | |
339 | t[i].rt, t[i].master, t[i].diff, t[i].lat); | |
340 | #endif | |
341 | ||
342 | printk(KERN_INFO "CPU %d: synchronized ITC with CPU %u (last diff %ld cycles, " | |
343 | "maxerr %lu cycles)\n", smp_processor_id(), master, delta, rt); | |
344 | } | |
345 | ||
346 | /* | |
347 | * Ideally sets up per-cpu profiling hooks. Doesn't do much now... | |
348 | */ | |
5b5e76e9 | 349 | static inline void smp_setup_percpu_timer(void) |
1da177e4 LT |
350 | { |
351 | } | |
352 | ||
ccce9bb8 | 353 | static void |
1da177e4 LT |
354 | smp_callin (void) |
355 | { | |
ff741906 | 356 | int cpuid, phys_id, itc_master; |
ead6caae | 357 | struct cpuinfo_ia64 *last_cpuinfo, *this_cpuinfo; |
1da177e4 | 358 | extern void ia64_init_itm(void); |
ff741906 | 359 | extern volatile int time_keeper_id; |
1da177e4 LT |
360 | |
361 | #ifdef CONFIG_PERFMON | |
362 | extern void pfm_init_percpu(void); | |
363 | #endif | |
364 | ||
365 | cpuid = smp_processor_id(); | |
366 | phys_id = hard_smp_processor_id(); | |
ff741906 | 367 | itc_master = time_keeper_id; |
1da177e4 LT |
368 | |
369 | if (cpu_online(cpuid)) { | |
370 | printk(KERN_ERR "huh, phys CPU#0x%x, CPU#0x%x already present??\n", | |
371 | phys_id, cpuid); | |
372 | BUG(); | |
373 | } | |
374 | ||
ff741906 AR |
375 | fix_b0_for_bsp(); |
376 | ||
3bccd996 LS |
377 | /* |
378 | * numa_node_id() works after this. | |
379 | */ | |
380 | set_numa_node(cpu_to_node_map[cpuid]); | |
fd1197f1 | 381 | set_numa_mem(local_memory_node(cpu_to_node_map[cpuid])); |
3bccd996 | 382 | |
e1b30a39 YI |
383 | spin_lock(&vector_lock); |
384 | /* Setup the per cpu irq handling data structures */ | |
385 | __setup_vector_irq(cpuid); | |
e545a614 | 386 | notify_cpu_starting(cpuid); |
7d7f9848 | 387 | set_cpu_online(cpuid, true); |
a9fa06c2 | 388 | per_cpu(cpu_state, cpuid) = CPU_ONLINE; |
e1b30a39 | 389 | spin_unlock(&vector_lock); |
1da177e4 LT |
390 | |
391 | smp_setup_percpu_timer(); | |
392 | ||
393 | ia64_mca_cmc_vector_setup(); /* Setup vector on AP */ | |
394 | ||
395 | #ifdef CONFIG_PERFMON | |
396 | pfm_init_percpu(); | |
397 | #endif | |
398 | ||
399 | local_irq_enable(); | |
400 | ||
401 | if (!(sal_platform_features & IA64_SAL_PLATFORM_FEATURE_ITC_DRIFT)) { | |
402 | /* | |
403 | * Synchronize the ITC with the BP. Need to do this after irqs are | |
404 | * enabled because ia64_sync_itc() calls smp_call_function_single(), which | |
405 | * calls spin_unlock_bh(), which calls spin_unlock_bh(), which calls | |
406 | * local_bh_enable(), which bugs out if irqs are not enabled... | |
407 | */ | |
ff741906 AR |
408 | Dprintk("Going to syncup ITC with ITC Master.\n"); |
409 | ia64_sync_itc(itc_master); | |
1da177e4 LT |
410 | } |
411 | ||
412 | /* | |
413 | * Get our bogomips. | |
414 | */ | |
415 | ia64_init_itm(); | |
ead6caae JS |
416 | |
417 | /* | |
418 | * Delay calibration can be skipped if new processor is identical to the | |
419 | * previous processor. | |
420 | */ | |
421 | last_cpuinfo = cpu_data(cpuid - 1); | |
422 | this_cpuinfo = local_cpu_data; | |
423 | if (last_cpuinfo->itc_freq != this_cpuinfo->itc_freq || | |
424 | last_cpuinfo->proc_freq != this_cpuinfo->proc_freq || | |
425 | last_cpuinfo->features != this_cpuinfo->features || | |
426 | last_cpuinfo->revision != this_cpuinfo->revision || | |
427 | last_cpuinfo->family != this_cpuinfo->family || | |
428 | last_cpuinfo->archrev != this_cpuinfo->archrev || | |
429 | last_cpuinfo->model != this_cpuinfo->model) | |
430 | calibrate_delay(); | |
1da177e4 LT |
431 | local_cpu_data->loops_per_jiffy = loops_per_jiffy; |
432 | ||
1da177e4 LT |
433 | /* |
434 | * Allow the master to continue. | |
435 | */ | |
5d2068da | 436 | cpumask_set_cpu(cpuid, &cpu_callin_map); |
1da177e4 LT |
437 | Dprintk("Stack on CPU %d at about %p\n",cpuid, &cpuid); |
438 | } | |
439 | ||
440 | ||
441 | /* | |
442 | * Activate a secondary processor. head.S calls this. | |
443 | */ | |
ccce9bb8 | 444 | int |
1da177e4 LT |
445 | start_secondary (void *unused) |
446 | { | |
447 | /* Early console may use I/O ports */ | |
448 | ia64_set_kr(IA64_KR_IO_BASE, __pa(ia64_iobase)); | |
10617bbe | 449 | #ifndef CONFIG_PRINTK_TIME |
1da177e4 | 450 | Dprintk("start_secondary: starting CPU 0x%x\n", hard_smp_processor_id()); |
10617bbe | 451 | #endif |
1da177e4 LT |
452 | efi_map_pal_code(); |
453 | cpu_init(); | |
5bfb5d69 | 454 | preempt_disable(); |
1da177e4 LT |
455 | smp_callin(); |
456 | ||
fc6d73d6 | 457 | cpu_startup_entry(CPUHP_AP_ONLINE_IDLE); |
1da177e4 LT |
458 | return 0; |
459 | } | |
460 | ||
ccce9bb8 | 461 | static int |
13583bf1 | 462 | do_boot_cpu (int sapicid, int cpu, struct task_struct *idle) |
1da177e4 LT |
463 | { |
464 | int timeout; | |
1da177e4 | 465 | |
13583bf1 | 466 | task_for_booting_cpu = idle; |
1da177e4 LT |
467 | Dprintk("Sending wakeup vector %lu to AP 0x%x/0x%x.\n", ap_wakeup_vector, cpu, sapicid); |
468 | ||
b8d8b883 | 469 | set_brendez_area(cpu); |
1da177e4 LT |
470 | platform_send_ipi(cpu, ap_wakeup_vector, IA64_IPI_DM_INT, 0); |
471 | ||
472 | /* | |
473 | * Wait 10s total for the AP to start | |
474 | */ | |
475 | Dprintk("Waiting on callin_map ..."); | |
476 | for (timeout = 0; timeout < 100000; timeout++) { | |
5d2068da | 477 | if (cpumask_test_cpu(cpu, &cpu_callin_map)) |
1da177e4 | 478 | break; /* It has booted */ |
5eda7861 | 479 | barrier(); /* Make sure we re-read cpu_callin_map */ |
1da177e4 LT |
480 | udelay(100); |
481 | } | |
482 | Dprintk("\n"); | |
483 | ||
5d2068da | 484 | if (!cpumask_test_cpu(cpu, &cpu_callin_map)) { |
1da177e4 LT |
485 | printk(KERN_ERR "Processor 0x%x/0x%x is stuck.\n", cpu, sapicid); |
486 | ia64_cpu_to_sapicid[cpu] = -1; | |
7d7f9848 | 487 | set_cpu_online(cpu, false); /* was set in smp_callin() */ |
1da177e4 LT |
488 | return -EINVAL; |
489 | } | |
490 | return 0; | |
491 | } | |
492 | ||
493 | static int __init | |
494 | decay (char *str) | |
495 | { | |
496 | int ticks; | |
497 | get_option (&str, &ticks); | |
498 | return 1; | |
499 | } | |
500 | ||
501 | __setup("decay=", decay); | |
502 | ||
503 | /* | |
504 | * Initialize the logical CPU number to SAPICID mapping | |
505 | */ | |
506 | void __init | |
507 | smp_build_cpu_map (void) | |
508 | { | |
509 | int sapicid, cpu, i; | |
510 | int boot_cpu_id = hard_smp_processor_id(); | |
511 | ||
512 | for (cpu = 0; cpu < NR_CPUS; cpu++) { | |
513 | ia64_cpu_to_sapicid[cpu] = -1; | |
1da177e4 LT |
514 | } |
515 | ||
516 | ia64_cpu_to_sapicid[0] = boot_cpu_id; | |
7d7f9848 | 517 | init_cpu_present(cpumask_of(0)); |
2af51a3f | 518 | set_cpu_possible(0, true); |
1da177e4 LT |
519 | for (cpu = 1, i = 0; i < smp_boot_data.cpu_count; i++) { |
520 | sapicid = smp_boot_data.cpu_phys_id[i]; | |
521 | if (sapicid == boot_cpu_id) | |
522 | continue; | |
2af51a3f RR |
523 | set_cpu_present(cpu, true); |
524 | set_cpu_possible(cpu, true); | |
1da177e4 LT |
525 | ia64_cpu_to_sapicid[cpu] = sapicid; |
526 | cpu++; | |
527 | } | |
528 | } | |
529 | ||
1da177e4 LT |
530 | /* |
531 | * Cycle through the APs sending Wakeup IPIs to boot each. | |
532 | */ | |
533 | void __init | |
534 | smp_prepare_cpus (unsigned int max_cpus) | |
535 | { | |
536 | int boot_cpu_id = hard_smp_processor_id(); | |
537 | ||
538 | /* | |
539 | * Initialize the per-CPU profiling counter/multiplier | |
540 | */ | |
541 | ||
542 | smp_setup_percpu_timer(); | |
543 | ||
5d2068da | 544 | cpumask_set_cpu(0, &cpu_callin_map); |
1da177e4 LT |
545 | |
546 | local_cpu_data->loops_per_jiffy = loops_per_jiffy; | |
547 | ia64_cpu_to_sapicid[0] = boot_cpu_id; | |
548 | ||
549 | printk(KERN_INFO "Boot processor id 0x%x/0x%x\n", 0, boot_cpu_id); | |
550 | ||
551 | current_thread_info()->cpu = 0; | |
552 | ||
553 | /* | |
554 | * If SMP should be disabled, then really disable it! | |
555 | */ | |
556 | if (!max_cpus) { | |
557 | printk(KERN_INFO "SMP mode deactivated.\n"); | |
2af51a3f RR |
558 | init_cpu_online(cpumask_of(0)); |
559 | init_cpu_present(cpumask_of(0)); | |
560 | init_cpu_possible(cpumask_of(0)); | |
1da177e4 LT |
561 | return; |
562 | } | |
563 | } | |
564 | ||
5b5e76e9 | 565 | void smp_prepare_boot_cpu(void) |
1da177e4 | 566 | { |
7d7f9848 | 567 | set_cpu_online(smp_processor_id(), true); |
5d2068da | 568 | cpumask_set_cpu(smp_processor_id(), &cpu_callin_map); |
3bccd996 | 569 | set_numa_node(cpu_to_node_map[smp_processor_id()]); |
a9fa06c2 | 570 | per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE; |
1da177e4 LT |
571 | } |
572 | ||
573 | #ifdef CONFIG_HOTPLUG_CPU | |
e927ecb0 SS |
574 | static inline void |
575 | clear_cpu_sibling_map(int cpu) | |
576 | { | |
577 | int i; | |
578 | ||
5d2068da RR |
579 | for_each_cpu(i, &per_cpu(cpu_sibling_map, cpu)) |
580 | cpumask_clear_cpu(cpu, &per_cpu(cpu_sibling_map, i)); | |
581 | for_each_cpu(i, &cpu_core_map[cpu]) | |
582 | cpumask_clear_cpu(cpu, &cpu_core_map[i]); | |
e927ecb0 | 583 | |
d5a7430d | 584 | per_cpu(cpu_sibling_map, cpu) = cpu_core_map[cpu] = CPU_MASK_NONE; |
e927ecb0 SS |
585 | } |
586 | ||
587 | static void | |
588 | remove_siblinginfo(int cpu) | |
589 | { | |
590 | int last = 0; | |
591 | ||
592 | if (cpu_data(cpu)->threads_per_core == 1 && | |
593 | cpu_data(cpu)->cores_per_socket == 1) { | |
5d2068da RR |
594 | cpumask_clear_cpu(cpu, &cpu_core_map[cpu]); |
595 | cpumask_clear_cpu(cpu, &per_cpu(cpu_sibling_map, cpu)); | |
e927ecb0 SS |
596 | return; |
597 | } | |
598 | ||
5d2068da | 599 | last = (cpumask_weight(&cpu_core_map[cpu]) == 1 ? 1 : 0); |
e927ecb0 SS |
600 | |
601 | /* remove it from all sibling map's */ | |
602 | clear_cpu_sibling_map(cpu); | |
e927ecb0 SS |
603 | } |
604 | ||
1da177e4 | 605 | extern void fixup_irqs(void); |
ff741906 AR |
606 | |
607 | int migrate_platform_irqs(unsigned int cpu) | |
608 | { | |
609 | int new_cpei_cpu; | |
097e98b4 | 610 | struct irq_data *data = NULL; |
0de26520 | 611 | const struct cpumask *mask; |
ff741906 AR |
612 | int retval = 0; |
613 | ||
614 | /* | |
615 | * dont permit CPEI target to removed. | |
616 | */ | |
617 | if (cpe_vector > 0 && is_cpu_cpei_target(cpu)) { | |
618 | printk ("CPU (%d) is CPEI Target\n", cpu); | |
619 | if (can_cpei_retarget()) { | |
620 | /* | |
621 | * Now re-target the CPEI to a different processor | |
622 | */ | |
7d7f9848 | 623 | new_cpei_cpu = cpumask_any(cpu_online_mask); |
0de26520 | 624 | mask = cpumask_of(new_cpei_cpu); |
ff741906 | 625 | set_cpei_target_cpu(new_cpei_cpu); |
097e98b4 | 626 | data = irq_get_irq_data(ia64_cpe_irq); |
ff741906 | 627 | /* |
72fdbdce | 628 | * Switch for now, immediately, we need to do fake intr |
ff741906 AR |
629 | * as other interrupts, but need to study CPEI behaviour with |
630 | * polling before making changes. | |
631 | */ | |
097e98b4 TG |
632 | if (data && data->chip) { |
633 | data->chip->irq_disable(data); | |
634 | data->chip->irq_set_affinity(data, mask, false); | |
635 | data->chip->irq_enable(data); | |
25985edc | 636 | printk ("Re-targeting CPEI to cpu %d\n", new_cpei_cpu); |
ff741906 AR |
637 | } |
638 | } | |
097e98b4 | 639 | if (!data) { |
ff741906 AR |
640 | printk ("Unable to retarget CPEI, offline cpu [%d] failed\n", cpu); |
641 | retval = -EBUSY; | |
642 | } | |
643 | } | |
644 | return retval; | |
645 | } | |
646 | ||
1da177e4 | 647 | /* must be called with cpucontrol mutex held */ |
1da177e4 LT |
648 | int __cpu_disable(void) |
649 | { | |
650 | int cpu = smp_processor_id(); | |
651 | ||
652 | /* | |
653 | * dont permit boot processor for now | |
654 | */ | |
ff741906 AR |
655 | if (cpu == 0 && !bsp_remove_ok) { |
656 | printk ("Your platform does not support removal of BSP\n"); | |
657 | return (-EBUSY); | |
658 | } | |
659 | ||
6e9de181 JK |
660 | if (ia64_platform_is("sn2")) { |
661 | if (!sn_cpu_disable_allowed(cpu)) | |
662 | return -EBUSY; | |
663 | } | |
664 | ||
7d7f9848 | 665 | set_cpu_online(cpu, false); |
66db2e63 | 666 | |
ff741906 | 667 | if (migrate_platform_irqs(cpu)) { |
7d7f9848 | 668 | set_cpu_online(cpu, true); |
c0acdea2 | 669 | return -EBUSY; |
ff741906 | 670 | } |
1da177e4 | 671 | |
e927ecb0 | 672 | remove_siblinginfo(cpu); |
66db2e63 | 673 | fixup_irqs(); |
1da177e4 | 674 | local_flush_tlb_all(); |
5d2068da | 675 | cpumask_clear_cpu(cpu, &cpu_callin_map); |
1da177e4 LT |
676 | return 0; |
677 | } | |
678 | ||
679 | void __cpu_die(unsigned int cpu) | |
680 | { | |
681 | unsigned int i; | |
682 | ||
683 | for (i = 0; i < 100; i++) { | |
684 | /* They ack this in play_dead by setting CPU_DEAD */ | |
685 | if (per_cpu(cpu_state, cpu) == CPU_DEAD) | |
686 | { | |
b8d8b883 | 687 | printk ("CPU %d is now offline\n", cpu); |
1da177e4 LT |
688 | return; |
689 | } | |
690 | msleep(100); | |
691 | } | |
692 | printk(KERN_ERR "CPU %u didn't die...\n", cpu); | |
693 | } | |
1da177e4 LT |
694 | #endif /* CONFIG_HOTPLUG_CPU */ |
695 | ||
696 | void | |
697 | smp_cpus_done (unsigned int dummy) | |
698 | { | |
699 | int cpu; | |
700 | unsigned long bogosum = 0; | |
701 | ||
702 | /* | |
703 | * Allow the user to impress friends. | |
704 | */ | |
705 | ||
dc565b52 | 706 | for_each_online_cpu(cpu) { |
707 | bogosum += cpu_data(cpu)->loops_per_jiffy; | |
708 | } | |
1da177e4 LT |
709 | |
710 | printk(KERN_INFO "Total of %d processors activated (%lu.%02lu BogoMIPS).\n", | |
711 | (int)num_online_cpus(), bogosum/(500000/HZ), (bogosum/(5000/HZ))%100); | |
712 | } | |
713 | ||
5b5e76e9 | 714 | static inline void set_cpu_sibling_map(int cpu) |
e927ecb0 SS |
715 | { |
716 | int i; | |
717 | ||
718 | for_each_online_cpu(i) { | |
719 | if ((cpu_data(cpu)->socket_id == cpu_data(i)->socket_id)) { | |
5d2068da RR |
720 | cpumask_set_cpu(i, &cpu_core_map[cpu]); |
721 | cpumask_set_cpu(cpu, &cpu_core_map[i]); | |
e927ecb0 | 722 | if (cpu_data(cpu)->core_id == cpu_data(i)->core_id) { |
5d2068da RR |
723 | cpumask_set_cpu(i, |
724 | &per_cpu(cpu_sibling_map, cpu)); | |
725 | cpumask_set_cpu(cpu, | |
726 | &per_cpu(cpu_sibling_map, i)); | |
e927ecb0 SS |
727 | } |
728 | } | |
729 | } | |
730 | } | |
731 | ||
ccce9bb8 | 732 | int |
8239c25f | 733 | __cpu_up(unsigned int cpu, struct task_struct *tidle) |
1da177e4 LT |
734 | { |
735 | int ret; | |
736 | int sapicid; | |
737 | ||
738 | sapicid = ia64_cpu_to_sapicid[cpu]; | |
739 | if (sapicid == -1) | |
740 | return -EINVAL; | |
741 | ||
742 | /* | |
b8d8b883 AR |
743 | * Already booted cpu? not valid anymore since we dont |
744 | * do idle loop tightspin anymore. | |
1da177e4 | 745 | */ |
5d2068da | 746 | if (cpumask_test_cpu(cpu, &cpu_callin_map)) |
b8d8b883 AR |
747 | return -EINVAL; |
748 | ||
a9fa06c2 | 749 | per_cpu(cpu_state, cpu) = CPU_UP_PREPARE; |
1da177e4 | 750 | /* Processor goes to start_secondary(), sets online flag */ |
13583bf1 | 751 | ret = do_boot_cpu(sapicid, cpu, tidle); |
1da177e4 LT |
752 | if (ret < 0) |
753 | return ret; | |
754 | ||
e927ecb0 SS |
755 | if (cpu_data(cpu)->threads_per_core == 1 && |
756 | cpu_data(cpu)->cores_per_socket == 1) { | |
5d2068da RR |
757 | cpumask_set_cpu(cpu, &per_cpu(cpu_sibling_map, cpu)); |
758 | cpumask_set_cpu(cpu, &cpu_core_map[cpu]); | |
e927ecb0 SS |
759 | return 0; |
760 | } | |
761 | ||
762 | set_cpu_sibling_map(cpu); | |
763 | ||
1da177e4 LT |
764 | return 0; |
765 | } | |
766 | ||
767 | /* | |
72fdbdce | 768 | * Assume that CPUs have been discovered by some platform-dependent interface. For |
1da177e4 LT |
769 | * SoftSDV/Lion, that would be ACPI. |
770 | * | |
771 | * Setup of the IPI irq handler is done in irq.c:init_IRQ_SMP(). | |
772 | */ | |
773 | void __init | |
774 | init_smp_config(void) | |
775 | { | |
776 | struct fptr { | |
777 | unsigned long fp; | |
778 | unsigned long gp; | |
779 | } *ap_startup; | |
780 | long sal_ret; | |
781 | ||
72fdbdce | 782 | /* Tell SAL where to drop the APs. */ |
1da177e4 LT |
783 | ap_startup = (struct fptr *) start_ap; |
784 | sal_ret = ia64_sal_set_vectors(SAL_VECTOR_OS_BOOT_RENDEZ, | |
785 | ia64_tpa(ap_startup->fp), ia64_tpa(ap_startup->gp), 0, 0, 0, 0); | |
786 | if (sal_ret < 0) | |
787 | printk(KERN_ERR "SMP: Can't set SAL AP Boot Rendezvous: %s\n", | |
788 | ia64_sal_strerror(sal_ret)); | |
789 | } | |
790 | ||
e927ecb0 SS |
791 | /* |
792 | * identify_siblings(cpu) gets called from identify_cpu. This populates the | |
793 | * information related to logical execution units in per_cpu_data structure. | |
794 | */ | |
5b5e76e9 | 795 | void identify_siblings(struct cpuinfo_ia64 *c) |
e927ecb0 | 796 | { |
e088a4ad | 797 | long status; |
e927ecb0 | 798 | u16 pltid; |
e927ecb0 SS |
799 | pal_logical_to_physical_t info; |
800 | ||
6ff0bc94 AC |
801 | status = ia64_pal_logical_to_phys(-1, &info); |
802 | if (status != PAL_STATUS_SUCCESS) { | |
113134fc AC |
803 | if (status != PAL_STATUS_UNIMPLEMENTED) { |
804 | printk(KERN_ERR | |
805 | "ia64_pal_logical_to_phys failed with %ld\n", | |
806 | status); | |
807 | return; | |
808 | } | |
809 | ||
810 | info.overview_ppid = 0; | |
811 | info.overview_cpp = 1; | |
812 | info.overview_tpc = 1; | |
e927ecb0 | 813 | } |
6ff0bc94 AC |
814 | |
815 | status = ia64_sal_physical_id_info(&pltid); | |
816 | if (status != PAL_STATUS_SUCCESS) { | |
817 | if (status != PAL_STATUS_UNIMPLEMENTED) | |
818 | printk(KERN_ERR | |
819 | "ia64_sal_pltid failed with %ld\n", | |
820 | status); | |
e927ecb0 SS |
821 | return; |
822 | } | |
e927ecb0 SS |
823 | |
824 | c->socket_id = (pltid << 8) | info.overview_ppid; | |
113134fc AC |
825 | |
826 | if (info.overview_cpp == 1 && info.overview_tpc == 1) | |
827 | return; | |
828 | ||
e927ecb0 SS |
829 | c->cores_per_socket = info.overview_cpp; |
830 | c->threads_per_core = info.overview_tpc; | |
4129a953 | 831 | c->num_log = info.overview_num_log; |
e927ecb0 | 832 | |
4129a953 FY |
833 | c->core_id = info.log1_cid; |
834 | c->thread_id = info.log1_tid; | |
e927ecb0 | 835 | } |
dd562c05 SE |
836 | |
837 | /* | |
838 | * returns non zero, if multi-threading is enabled | |
839 | * on at least one physical package. Due to hotplug cpu | |
840 | * and (maxcpus=), all threads may not necessarily be enabled | |
841 | * even though the processor supports multi-threading. | |
842 | */ | |
843 | int is_multithreading_enabled(void) | |
844 | { | |
845 | int i, j; | |
846 | ||
847 | for_each_present_cpu(i) { | |
848 | for_each_present_cpu(j) { | |
849 | if (j == i) | |
850 | continue; | |
851 | if ((cpu_data(j)->socket_id == cpu_data(i)->socket_id)) { | |
852 | if (cpu_data(j)->core_id == cpu_data(i)->core_id) | |
853 | return 1; | |
854 | } | |
855 | } | |
856 | } | |
857 | return 0; | |
858 | } | |
859 | EXPORT_SYMBOL_GPL(is_multithreading_enabled); |