Commit | Line | Data |
---|---|---|
4cedb334 GOC |
1 | /* |
2 | * x86 SMP booting functions | |
3 | * | |
4 | * (c) 1995 Alan Cox, Building #3 <alan@redhat.com> | |
5 | * (c) 1998, 1999, 2000 Ingo Molnar <mingo@redhat.com> | |
6 | * Copyright 2001 Andi Kleen, SuSE Labs. | |
7 | * | |
8 | * Much of the core SMP work is based on previous work by Thomas Radke, to | |
9 | * whom a great many thanks are extended. | |
10 | * | |
11 | * Thanks to Intel for making available several different Pentium, | |
12 | * Pentium Pro and Pentium-II/Xeon MP machines. | |
13 | * Original development of Linux SMP code supported by Caldera. | |
14 | * | |
15 | * This code is released under the GNU General Public License version 2 or | |
16 | * later. | |
17 | * | |
18 | * Fixes | |
19 | * Felix Koop : NR_CPUS used properly | |
20 | * Jose Renau : Handle single CPU case. | |
21 | * Alan Cox : By repeated request 8) - Total BogoMIPS report. | |
22 | * Greg Wright : Fix for kernel stacks panic. | |
23 | * Erich Boleyn : MP v1.4 and additional changes. | |
24 | * Matthias Sattler : Changes for 2.1 kernel map. | |
25 | * Michel Lespinasse : Changes for 2.1 kernel map. | |
26 | * Michael Chastain : Change trampoline.S to gnu as. | |
27 | * Alan Cox : Dumb bug: 'B' step PPro's are fine | |
28 | * Ingo Molnar : Added APIC timers, based on code | |
29 | * from Jose Renau | |
30 | * Ingo Molnar : various cleanups and rewrites | |
31 | * Tigran Aivazian : fixed "0.00 in /proc/uptime on SMP" bug. | |
32 | * Maciej W. Rozycki : Bits for genuine 82489DX APICs | |
33 | * Andi Kleen : Changed for SMP boot into long mode. | |
34 | * Martin J. Bligh : Added support for multi-quad systems | |
35 | * Dave Jones : Report invalid combinations of Athlon CPUs. | |
36 | * Rusty Russell : Hacked into shape for new "hotplug" boot process. | |
37 | * Andi Kleen : Converted to new state machine. | |
38 | * Ashok Raj : CPU hotplug support | |
39 | * Glauber Costa : i386 and x86_64 integration | |
40 | */ | |
41 | ||
68a1c3f8 GC |
42 | #include <linux/init.h> |
43 | #include <linux/smp.h> | |
a355352b | 44 | #include <linux/module.h> |
70708a18 | 45 | #include <linux/sched.h> |
69c18c15 | 46 | #include <linux/percpu.h> |
91718e8d | 47 | #include <linux/bootmem.h> |
cb3c8b90 GOC |
48 | #include <linux/err.h> |
49 | #include <linux/nmi.h> | |
69c18c15 | 50 | |
8aef135c | 51 | #include <asm/acpi.h> |
cb3c8b90 | 52 | #include <asm/desc.h> |
69c18c15 GC |
53 | #include <asm/nmi.h> |
54 | #include <asm/irq.h> | |
07bbc16a | 55 | #include <asm/idle.h> |
69c18c15 | 56 | #include <asm/smp.h> |
e44b7b75 | 57 | #include <asm/trampoline.h> |
69c18c15 GC |
58 | #include <asm/cpu.h> |
59 | #include <asm/numa.h> | |
cb3c8b90 GOC |
60 | #include <asm/pgtable.h> |
61 | #include <asm/tlbflush.h> | |
62 | #include <asm/mtrr.h> | |
bbc2ff6a | 63 | #include <asm/vmi.h> |
34d05591 | 64 | #include <asm/genapic.h> |
cb3c8b90 | 65 | #include <linux/mc146818rtc.h> |
68a1c3f8 | 66 | |
f6bc4029 | 67 | #include <mach_apic.h> |
cb3c8b90 GOC |
68 | #include <mach_wakecpu.h> |
69 | #include <smpboot_hooks.h> | |
70 | ||
16ecf7a4 | 71 | #ifdef CONFIG_X86_32 |
4cedb334 | 72 | u8 apicid_2_node[MAX_APICID]; |
61165d7a | 73 | static int low_mappings; |
acbb6734 GOC |
74 | #endif |
75 | ||
a8db8453 GOC |
76 | /* State of each CPU */ |
77 | DEFINE_PER_CPU(int, cpu_state) = { 0 }; | |
78 | ||
cb3c8b90 GOC |
79 | /* Store all idle threads, this can be reused instead of creating |
80 | * a new thread. Also avoids complicated thread destroy functionality | |
81 | * for idle threads. | |
82 | */ | |
83 | #ifdef CONFIG_HOTPLUG_CPU | |
84 | /* | |
85 | * Needed only for CONFIG_HOTPLUG_CPU because __cpuinitdata is | |
86 | * removed after init for !CONFIG_HOTPLUG_CPU. | |
87 | */ | |
88 | static DEFINE_PER_CPU(struct task_struct *, idle_thread_array); | |
89 | #define get_idle_for_cpu(x) (per_cpu(idle_thread_array, x)) | |
90 | #define set_idle_for_cpu(x, p) (per_cpu(idle_thread_array, x) = (p)) | |
91 | #else | |
f86c9985 | 92 | static struct task_struct *idle_thread_array[NR_CPUS] __cpuinitdata ; |
cb3c8b90 GOC |
93 | #define get_idle_for_cpu(x) (idle_thread_array[(x)]) |
94 | #define set_idle_for_cpu(x, p) (idle_thread_array[(x)] = (p)) | |
95 | #endif | |
f6bc4029 | 96 | |
a355352b GC |
97 | /* Number of siblings per CPU package */ |
98 | int smp_num_siblings = 1; | |
99 | EXPORT_SYMBOL(smp_num_siblings); | |
100 | ||
101 | /* Last level cache ID of each logical CPU */ | |
102 | DEFINE_PER_CPU(u16, cpu_llc_id) = BAD_APICID; | |
103 | ||
104 | /* bitmap of online cpus */ | |
105 | cpumask_t cpu_online_map __read_mostly; | |
106 | EXPORT_SYMBOL(cpu_online_map); | |
107 | ||
108 | cpumask_t cpu_callin_map; | |
109 | cpumask_t cpu_callout_map; | |
110 | cpumask_t cpu_possible_map; | |
111 | EXPORT_SYMBOL(cpu_possible_map); | |
112 | ||
113 | /* representing HT siblings of each logical CPU */ | |
114 | DEFINE_PER_CPU(cpumask_t, cpu_sibling_map); | |
115 | EXPORT_PER_CPU_SYMBOL(cpu_sibling_map); | |
116 | ||
117 | /* representing HT and core siblings of each logical CPU */ | |
118 | DEFINE_PER_CPU(cpumask_t, cpu_core_map); | |
119 | EXPORT_PER_CPU_SYMBOL(cpu_core_map); | |
120 | ||
121 | /* Per CPU bogomips and other parameters */ | |
122 | DEFINE_PER_CPU_SHARED_ALIGNED(struct cpuinfo_x86, cpu_info); | |
123 | EXPORT_PER_CPU_SYMBOL(cpu_info); | |
768d9505 | 124 | |
cb3c8b90 GOC |
125 | static atomic_t init_deasserted; |
126 | ||
8aef135c | 127 | |
768d9505 GC |
128 | /* representing cpus for which sibling maps can be computed */ |
129 | static cpumask_t cpu_sibling_setup_map; | |
130 | ||
1d89a7f0 | 131 | /* Set if we find a B stepping CPU */ |
f86c9985 | 132 | static int __cpuinitdata smp_b_stepping; |
1d89a7f0 | 133 | |
7cc3959e GOC |
134 | #if defined(CONFIG_NUMA) && defined(CONFIG_X86_32) |
135 | ||
136 | /* which logical CPUs are on which nodes */ | |
137 | cpumask_t node_to_cpumask_map[MAX_NUMNODES] __read_mostly = | |
138 | { [0 ... MAX_NUMNODES-1] = CPU_MASK_NONE }; | |
139 | EXPORT_SYMBOL(node_to_cpumask_map); | |
140 | /* which node each logical CPU is on */ | |
141 | int cpu_to_node_map[NR_CPUS] __read_mostly = { [0 ... NR_CPUS-1] = 0 }; | |
142 | EXPORT_SYMBOL(cpu_to_node_map); | |
143 | ||
144 | /* set up a mapping between cpu and node. */ | |
145 | static void map_cpu_to_node(int cpu, int node) | |
146 | { | |
147 | printk(KERN_INFO "Mapping cpu %d to node %d\n", cpu, node); | |
148 | cpu_set(cpu, node_to_cpumask_map[node]); | |
149 | cpu_to_node_map[cpu] = node; | |
150 | } | |
151 | ||
152 | /* undo a mapping between cpu and node. */ | |
153 | static void unmap_cpu_to_node(int cpu) | |
154 | { | |
155 | int node; | |
156 | ||
157 | printk(KERN_INFO "Unmapping cpu %d from all nodes\n", cpu); | |
158 | for (node = 0; node < MAX_NUMNODES; node++) | |
159 | cpu_clear(cpu, node_to_cpumask_map[node]); | |
160 | cpu_to_node_map[cpu] = 0; | |
161 | } | |
162 | #else /* !(CONFIG_NUMA && CONFIG_X86_32) */ | |
163 | #define map_cpu_to_node(cpu, node) ({}) | |
164 | #define unmap_cpu_to_node(cpu) ({}) | |
165 | #endif | |
166 | ||
167 | #ifdef CONFIG_X86_32 | |
1b374e4d SS |
168 | static int boot_cpu_logical_apicid; |
169 | ||
7cc3959e GOC |
170 | u8 cpu_2_logical_apicid[NR_CPUS] __read_mostly = |
171 | { [0 ... NR_CPUS-1] = BAD_APICID }; | |
172 | ||
a4928cff | 173 | static void map_cpu_to_logical_apicid(void) |
7cc3959e GOC |
174 | { |
175 | int cpu = smp_processor_id(); | |
176 | int apicid = logical_smp_processor_id(); | |
177 | int node = apicid_to_node(apicid); | |
178 | ||
179 | if (!node_online(node)) | |
180 | node = first_online_node; | |
181 | ||
182 | cpu_2_logical_apicid[cpu] = apicid; | |
183 | map_cpu_to_node(cpu, node); | |
184 | } | |
185 | ||
1481a3dd | 186 | void numa_remove_cpu(int cpu) |
7cc3959e GOC |
187 | { |
188 | cpu_2_logical_apicid[cpu] = BAD_APICID; | |
189 | unmap_cpu_to_node(cpu); | |
190 | } | |
191 | #else | |
7cc3959e GOC |
192 | #define map_cpu_to_logical_apicid() do {} while (0) |
193 | #endif | |
194 | ||
cb3c8b90 GOC |
195 | /* |
196 | * Report back to the Boot Processor. | |
197 | * Running on AP. | |
198 | */ | |
a4928cff | 199 | static void __cpuinit smp_callin(void) |
cb3c8b90 GOC |
200 | { |
201 | int cpuid, phys_id; | |
202 | unsigned long timeout; | |
203 | ||
204 | /* | |
205 | * If waken up by an INIT in an 82489DX configuration | |
206 | * we may get here before an INIT-deassert IPI reaches | |
207 | * our local APIC. We have to wait for the IPI or we'll | |
208 | * lock up on an APIC access. | |
209 | */ | |
210 | wait_for_init_deassert(&init_deasserted); | |
211 | ||
212 | /* | |
213 | * (This works even if the APIC is not enabled.) | |
214 | */ | |
4c9961d5 | 215 | phys_id = read_apic_id(); |
cb3c8b90 GOC |
216 | cpuid = smp_processor_id(); |
217 | if (cpu_isset(cpuid, cpu_callin_map)) { | |
218 | panic("%s: phys CPU#%d, CPU#%d already present??\n", __func__, | |
219 | phys_id, cpuid); | |
220 | } | |
cfc1b9a6 | 221 | pr_debug("CPU#%d (phys ID: %d) waiting for CALLOUT\n", cpuid, phys_id); |
cb3c8b90 GOC |
222 | |
223 | /* | |
224 | * STARTUP IPIs are fragile beasts as they might sometimes | |
225 | * trigger some glue motherboard logic. Complete APIC bus | |
226 | * silence for 1 second, this overestimates the time the | |
227 | * boot CPU is spending to send the up to 2 STARTUP IPIs | |
228 | * by a factor of two. This should be enough. | |
229 | */ | |
230 | ||
231 | /* | |
232 | * Waiting 2s total for startup (udelay is not yet working) | |
233 | */ | |
234 | timeout = jiffies + 2*HZ; | |
235 | while (time_before(jiffies, timeout)) { | |
236 | /* | |
237 | * Has the boot CPU finished it's STARTUP sequence? | |
238 | */ | |
239 | if (cpu_isset(cpuid, cpu_callout_map)) | |
240 | break; | |
241 | cpu_relax(); | |
242 | } | |
243 | ||
244 | if (!time_before(jiffies, timeout)) { | |
245 | panic("%s: CPU%d started up but did not get a callout!\n", | |
246 | __func__, cpuid); | |
247 | } | |
248 | ||
249 | /* | |
250 | * the boot CPU has finished the init stage and is spinning | |
251 | * on callin_map until we finish. We are free to set up this | |
252 | * CPU, first the APIC. (this is probably redundant on most | |
253 | * boards) | |
254 | */ | |
255 | ||
cfc1b9a6 | 256 | pr_debug("CALLIN, before setup_local_APIC().\n"); |
cb3c8b90 GOC |
257 | smp_callin_clear_local_apic(); |
258 | setup_local_APIC(); | |
259 | end_local_APIC_setup(); | |
260 | map_cpu_to_logical_apicid(); | |
261 | ||
e545a614 | 262 | notify_cpu_starting(cpuid); |
cb3c8b90 GOC |
263 | /* |
264 | * Get our bogomips. | |
265 | * | |
266 | * Need to enable IRQs because it can take longer and then | |
267 | * the NMI watchdog might kill us. | |
268 | */ | |
269 | local_irq_enable(); | |
270 | calibrate_delay(); | |
271 | local_irq_disable(); | |
cfc1b9a6 | 272 | pr_debug("Stack at about %p\n", &cpuid); |
cb3c8b90 GOC |
273 | |
274 | /* | |
275 | * Save our processor parameters | |
276 | */ | |
277 | smp_store_cpu_info(cpuid); | |
278 | ||
279 | /* | |
280 | * Allow the master to continue. | |
281 | */ | |
282 | cpu_set(cpuid, cpu_callin_map); | |
283 | } | |
284 | ||
bbc2ff6a GOC |
285 | /* |
286 | * Activate a secondary processor. | |
287 | */ | |
dbe55f47 | 288 | static void __cpuinit start_secondary(void *unused) |
bbc2ff6a GOC |
289 | { |
290 | /* | |
291 | * Don't put *anything* before cpu_init(), SMP booting is too | |
292 | * fragile that we want to limit the things done here to the | |
293 | * most necessary things. | |
294 | */ | |
295 | #ifdef CONFIG_VMI | |
296 | vmi_bringup(); | |
297 | #endif | |
298 | cpu_init(); | |
299 | preempt_disable(); | |
300 | smp_callin(); | |
301 | ||
302 | /* otherwise gcc will move up smp_processor_id before the cpu_init */ | |
303 | barrier(); | |
304 | /* | |
305 | * Check TSC synchronization with the BP: | |
306 | */ | |
307 | check_tsc_sync_target(); | |
308 | ||
309 | if (nmi_watchdog == NMI_IO_APIC) { | |
310 | disable_8259A_irq(0); | |
311 | enable_NMI_through_LVT0(); | |
312 | enable_8259A_irq(0); | |
313 | } | |
314 | ||
61165d7a HD |
315 | #ifdef CONFIG_X86_32 |
316 | while (low_mappings) | |
317 | cpu_relax(); | |
318 | __flush_tlb_all(); | |
319 | #endif | |
320 | ||
bbc2ff6a GOC |
321 | /* This must be done before setting cpu_online_map */ |
322 | set_cpu_sibling_map(raw_smp_processor_id()); | |
323 | wmb(); | |
324 | ||
325 | /* | |
326 | * We need to hold call_lock, so there is no inconsistency | |
327 | * between the time smp_call_function() determines number of | |
328 | * IPI recipients, and the time when the determination is made | |
329 | * for which cpus receive the IPI. Holding this | |
330 | * lock helps us to not include this cpu in a currently in progress | |
331 | * smp_call_function(). | |
d388e5fd EB |
332 | * |
333 | * We need to hold vector_lock so there the set of online cpus | |
334 | * does not change while we are assigning vectors to cpus. Holding | |
335 | * this lock ensures we don't half assign or remove an irq from a cpu. | |
bbc2ff6a | 336 | */ |
3b16cf87 | 337 | ipi_call_lock_irq(); |
d388e5fd EB |
338 | lock_vector_lock(); |
339 | __setup_vector_irq(smp_processor_id()); | |
bbc2ff6a | 340 | cpu_set(smp_processor_id(), cpu_online_map); |
d388e5fd | 341 | unlock_vector_lock(); |
3b16cf87 | 342 | ipi_call_unlock_irq(); |
bbc2ff6a GOC |
343 | per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE; |
344 | ||
345 | setup_secondary_clock(); | |
346 | ||
347 | wmb(); | |
348 | cpu_idle(); | |
349 | } | |
350 | ||
1d89a7f0 GOC |
351 | static void __cpuinit smp_apply_quirks(struct cpuinfo_x86 *c) |
352 | { | |
1d89a7f0 GOC |
353 | /* |
354 | * Mask B, Pentium, but not Pentium MMX | |
355 | */ | |
356 | if (c->x86_vendor == X86_VENDOR_INTEL && | |
357 | c->x86 == 5 && | |
358 | c->x86_mask >= 1 && c->x86_mask <= 4 && | |
359 | c->x86_model <= 3) | |
360 | /* | |
361 | * Remember we have B step Pentia with bugs | |
362 | */ | |
363 | smp_b_stepping = 1; | |
364 | ||
365 | /* | |
366 | * Certain Athlons might work (for various values of 'work') in SMP | |
367 | * but they are not certified as MP capable. | |
368 | */ | |
369 | if ((c->x86_vendor == X86_VENDOR_AMD) && (c->x86 == 6)) { | |
370 | ||
371 | if (num_possible_cpus() == 1) | |
372 | goto valid_k7; | |
373 | ||
374 | /* Athlon 660/661 is valid. */ | |
375 | if ((c->x86_model == 6) && ((c->x86_mask == 0) || | |
376 | (c->x86_mask == 1))) | |
377 | goto valid_k7; | |
378 | ||
379 | /* Duron 670 is valid */ | |
380 | if ((c->x86_model == 7) && (c->x86_mask == 0)) | |
381 | goto valid_k7; | |
382 | ||
383 | /* | |
384 | * Athlon 662, Duron 671, and Athlon >model 7 have capability | |
385 | * bit. It's worth noting that the A5 stepping (662) of some | |
386 | * Athlon XP's have the MP bit set. | |
387 | * See http://www.heise.de/newsticker/data/jow-18.10.01-000 for | |
388 | * more. | |
389 | */ | |
390 | if (((c->x86_model == 6) && (c->x86_mask >= 2)) || | |
391 | ((c->x86_model == 7) && (c->x86_mask >= 1)) || | |
392 | (c->x86_model > 7)) | |
393 | if (cpu_has_mp) | |
394 | goto valid_k7; | |
395 | ||
396 | /* If we get here, not a certified SMP capable AMD system. */ | |
397 | add_taint(TAINT_UNSAFE_SMP); | |
398 | } | |
399 | ||
400 | valid_k7: | |
401 | ; | |
1d89a7f0 GOC |
402 | } |
403 | ||
a4928cff | 404 | static void __cpuinit smp_checks(void) |
693d4b8a GOC |
405 | { |
406 | if (smp_b_stepping) | |
407 | printk(KERN_WARNING "WARNING: SMP operation may be unreliable" | |
408 | "with B stepping processors.\n"); | |
409 | ||
410 | /* | |
411 | * Don't taint if we are running SMP kernel on a single non-MP | |
412 | * approved Athlon | |
413 | */ | |
414 | if (tainted & TAINT_UNSAFE_SMP) { | |
f68e00a3 | 415 | if (num_online_cpus()) |
693d4b8a GOC |
416 | printk(KERN_INFO "WARNING: This combination of AMD" |
417 | "processors is not suitable for SMP.\n"); | |
418 | else | |
419 | tainted &= ~TAINT_UNSAFE_SMP; | |
420 | } | |
421 | } | |
422 | ||
1d89a7f0 GOC |
423 | /* |
424 | * The bootstrap kernel entry code has set these up. Save them for | |
425 | * a given CPU | |
426 | */ | |
427 | ||
428 | void __cpuinit smp_store_cpu_info(int id) | |
429 | { | |
430 | struct cpuinfo_x86 *c = &cpu_data(id); | |
431 | ||
432 | *c = boot_cpu_data; | |
433 | c->cpu_index = id; | |
434 | if (id != 0) | |
435 | identify_secondary_cpu(c); | |
436 | smp_apply_quirks(c); | |
437 | } | |
438 | ||
439 | ||
768d9505 GC |
440 | void __cpuinit set_cpu_sibling_map(int cpu) |
441 | { | |
442 | int i; | |
443 | struct cpuinfo_x86 *c = &cpu_data(cpu); | |
444 | ||
445 | cpu_set(cpu, cpu_sibling_setup_map); | |
446 | ||
447 | if (smp_num_siblings > 1) { | |
334ef7a7 | 448 | for_each_cpu_mask_nr(i, cpu_sibling_setup_map) { |
768d9505 GC |
449 | if (c->phys_proc_id == cpu_data(i).phys_proc_id && |
450 | c->cpu_core_id == cpu_data(i).cpu_core_id) { | |
451 | cpu_set(i, per_cpu(cpu_sibling_map, cpu)); | |
452 | cpu_set(cpu, per_cpu(cpu_sibling_map, i)); | |
453 | cpu_set(i, per_cpu(cpu_core_map, cpu)); | |
454 | cpu_set(cpu, per_cpu(cpu_core_map, i)); | |
455 | cpu_set(i, c->llc_shared_map); | |
456 | cpu_set(cpu, cpu_data(i).llc_shared_map); | |
457 | } | |
458 | } | |
459 | } else { | |
460 | cpu_set(cpu, per_cpu(cpu_sibling_map, cpu)); | |
461 | } | |
462 | ||
463 | cpu_set(cpu, c->llc_shared_map); | |
464 | ||
465 | if (current_cpu_data.x86_max_cores == 1) { | |
466 | per_cpu(cpu_core_map, cpu) = per_cpu(cpu_sibling_map, cpu); | |
467 | c->booted_cores = 1; | |
468 | return; | |
469 | } | |
470 | ||
334ef7a7 | 471 | for_each_cpu_mask_nr(i, cpu_sibling_setup_map) { |
768d9505 GC |
472 | if (per_cpu(cpu_llc_id, cpu) != BAD_APICID && |
473 | per_cpu(cpu_llc_id, cpu) == per_cpu(cpu_llc_id, i)) { | |
474 | cpu_set(i, c->llc_shared_map); | |
475 | cpu_set(cpu, cpu_data(i).llc_shared_map); | |
476 | } | |
477 | if (c->phys_proc_id == cpu_data(i).phys_proc_id) { | |
478 | cpu_set(i, per_cpu(cpu_core_map, cpu)); | |
479 | cpu_set(cpu, per_cpu(cpu_core_map, i)); | |
480 | /* | |
481 | * Does this new cpu bringup a new core? | |
482 | */ | |
483 | if (cpus_weight(per_cpu(cpu_sibling_map, cpu)) == 1) { | |
484 | /* | |
485 | * for each core in package, increment | |
486 | * the booted_cores for this new cpu | |
487 | */ | |
488 | if (first_cpu(per_cpu(cpu_sibling_map, i)) == i) | |
489 | c->booted_cores++; | |
490 | /* | |
491 | * increment the core count for all | |
492 | * the other cpus in this package | |
493 | */ | |
494 | if (i != cpu) | |
495 | cpu_data(i).booted_cores++; | |
496 | } else if (i != cpu && !c->booted_cores) | |
497 | c->booted_cores = cpu_data(i).booted_cores; | |
498 | } | |
499 | } | |
500 | } | |
501 | ||
70708a18 GC |
502 | /* maps the cpu to the sched domain representing multi-core */ |
503 | cpumask_t cpu_coregroup_map(int cpu) | |
504 | { | |
505 | struct cpuinfo_x86 *c = &cpu_data(cpu); | |
506 | /* | |
507 | * For perf, we return last level cache shared map. | |
508 | * And for power savings, we return cpu_core_map | |
509 | */ | |
510 | if (sched_mc_power_savings || sched_smt_power_savings) | |
511 | return per_cpu(cpu_core_map, cpu); | |
512 | else | |
513 | return c->llc_shared_map; | |
514 | } | |
515 | ||
a4928cff | 516 | static void impress_friends(void) |
904541e2 GOC |
517 | { |
518 | int cpu; | |
519 | unsigned long bogosum = 0; | |
520 | /* | |
521 | * Allow the user to impress friends. | |
522 | */ | |
cfc1b9a6 | 523 | pr_debug("Before bogomips.\n"); |
904541e2 GOC |
524 | for_each_possible_cpu(cpu) |
525 | if (cpu_isset(cpu, cpu_callout_map)) | |
526 | bogosum += cpu_data(cpu).loops_per_jiffy; | |
527 | printk(KERN_INFO | |
528 | "Total of %d processors activated (%lu.%02lu BogoMIPS).\n", | |
f68e00a3 | 529 | num_online_cpus(), |
904541e2 GOC |
530 | bogosum/(500000/HZ), |
531 | (bogosum/(5000/HZ))%100); | |
532 | ||
cfc1b9a6 | 533 | pr_debug("Before bogocount - setting activated=1.\n"); |
904541e2 GOC |
534 | } |
535 | ||
cb3c8b90 GOC |
536 | static inline void __inquire_remote_apic(int apicid) |
537 | { | |
538 | unsigned i, regs[] = { APIC_ID >> 4, APIC_LVR >> 4, APIC_SPIV >> 4 }; | |
539 | char *names[] = { "ID", "VERSION", "SPIV" }; | |
540 | int timeout; | |
541 | u32 status; | |
542 | ||
543 | printk(KERN_INFO "Inquiring remote APIC #%d...\n", apicid); | |
544 | ||
545 | for (i = 0; i < ARRAY_SIZE(regs); i++) { | |
546 | printk(KERN_INFO "... APIC #%d %s: ", apicid, names[i]); | |
547 | ||
548 | /* | |
549 | * Wait for idle. | |
550 | */ | |
551 | status = safe_apic_wait_icr_idle(); | |
552 | if (status) | |
553 | printk(KERN_CONT | |
554 | "a previous APIC delivery may have failed\n"); | |
555 | ||
1b374e4d | 556 | apic_icr_write(APIC_DM_REMRD | regs[i], apicid); |
cb3c8b90 GOC |
557 | |
558 | timeout = 0; | |
559 | do { | |
560 | udelay(100); | |
561 | status = apic_read(APIC_ICR) & APIC_ICR_RR_MASK; | |
562 | } while (status == APIC_ICR_RR_INPROG && timeout++ < 1000); | |
563 | ||
564 | switch (status) { | |
565 | case APIC_ICR_RR_VALID: | |
566 | status = apic_read(APIC_RRR); | |
567 | printk(KERN_CONT "%08x\n", status); | |
568 | break; | |
569 | default: | |
570 | printk(KERN_CONT "failed\n"); | |
571 | } | |
572 | } | |
573 | } | |
574 | ||
575 | #ifdef WAKE_SECONDARY_VIA_NMI | |
576 | /* | |
577 | * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal | |
578 | * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this | |
579 | * won't ... remember to clear down the APIC, etc later. | |
580 | */ | |
581 | static int __devinit | |
582 | wakeup_secondary_cpu(int logical_apicid, unsigned long start_eip) | |
583 | { | |
584 | unsigned long send_status, accept_status = 0; | |
585 | int maxlvt; | |
586 | ||
587 | /* Target chip */ | |
cb3c8b90 GOC |
588 | /* Boot on the stack */ |
589 | /* Kick the second */ | |
1b374e4d | 590 | apic_icr_write(APIC_DM_NMI | APIC_DEST_LOGICAL, logical_apicid); |
cb3c8b90 | 591 | |
cfc1b9a6 | 592 | pr_debug("Waiting for send to finish...\n"); |
cb3c8b90 GOC |
593 | send_status = safe_apic_wait_icr_idle(); |
594 | ||
595 | /* | |
596 | * Give the other CPU some time to accept the IPI. | |
597 | */ | |
598 | udelay(200); | |
cb3c8b90 | 599 | maxlvt = lapic_get_maxlvt(); |
593f4a78 | 600 | if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */ |
cb3c8b90 | 601 | apic_write(APIC_ESR, 0); |
cb3c8b90 | 602 | accept_status = (apic_read(APIC_ESR) & 0xEF); |
cfc1b9a6 | 603 | pr_debug("NMI sent.\n"); |
cb3c8b90 GOC |
604 | |
605 | if (send_status) | |
606 | printk(KERN_ERR "APIC never delivered???\n"); | |
607 | if (accept_status) | |
608 | printk(KERN_ERR "APIC delivery error (%lx).\n", accept_status); | |
609 | ||
610 | return (send_status | accept_status); | |
611 | } | |
612 | #endif /* WAKE_SECONDARY_VIA_NMI */ | |
613 | ||
cb3c8b90 GOC |
614 | #ifdef WAKE_SECONDARY_VIA_INIT |
615 | static int __devinit | |
616 | wakeup_secondary_cpu(int phys_apicid, unsigned long start_eip) | |
617 | { | |
618 | unsigned long send_status, accept_status = 0; | |
619 | int maxlvt, num_starts, j; | |
620 | ||
34d05591 JS |
621 | if (get_uv_system_type() == UV_NON_UNIQUE_APIC) { |
622 | send_status = uv_wakeup_secondary(phys_apicid, start_eip); | |
623 | atomic_set(&init_deasserted, 1); | |
624 | return send_status; | |
625 | } | |
626 | ||
593f4a78 MR |
627 | maxlvt = lapic_get_maxlvt(); |
628 | ||
cb3c8b90 GOC |
629 | /* |
630 | * Be paranoid about clearing APIC errors. | |
631 | */ | |
632 | if (APIC_INTEGRATED(apic_version[phys_apicid])) { | |
593f4a78 MR |
633 | if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */ |
634 | apic_write(APIC_ESR, 0); | |
cb3c8b90 GOC |
635 | apic_read(APIC_ESR); |
636 | } | |
637 | ||
cfc1b9a6 | 638 | pr_debug("Asserting INIT.\n"); |
cb3c8b90 GOC |
639 | |
640 | /* | |
641 | * Turn INIT on target chip | |
642 | */ | |
cb3c8b90 GOC |
643 | /* |
644 | * Send IPI | |
645 | */ | |
1b374e4d SS |
646 | apic_icr_write(APIC_INT_LEVELTRIG | APIC_INT_ASSERT | APIC_DM_INIT, |
647 | phys_apicid); | |
cb3c8b90 | 648 | |
cfc1b9a6 | 649 | pr_debug("Waiting for send to finish...\n"); |
cb3c8b90 GOC |
650 | send_status = safe_apic_wait_icr_idle(); |
651 | ||
652 | mdelay(10); | |
653 | ||
cfc1b9a6 | 654 | pr_debug("Deasserting INIT.\n"); |
cb3c8b90 GOC |
655 | |
656 | /* Target chip */ | |
cb3c8b90 | 657 | /* Send IPI */ |
1b374e4d | 658 | apic_icr_write(APIC_INT_LEVELTRIG | APIC_DM_INIT, phys_apicid); |
cb3c8b90 | 659 | |
cfc1b9a6 | 660 | pr_debug("Waiting for send to finish...\n"); |
cb3c8b90 GOC |
661 | send_status = safe_apic_wait_icr_idle(); |
662 | ||
663 | mb(); | |
664 | atomic_set(&init_deasserted, 1); | |
665 | ||
666 | /* | |
667 | * Should we send STARTUP IPIs ? | |
668 | * | |
669 | * Determine this based on the APIC version. | |
670 | * If we don't have an integrated APIC, don't send the STARTUP IPIs. | |
671 | */ | |
672 | if (APIC_INTEGRATED(apic_version[phys_apicid])) | |
673 | num_starts = 2; | |
674 | else | |
675 | num_starts = 0; | |
676 | ||
677 | /* | |
678 | * Paravirt / VMI wants a startup IPI hook here to set up the | |
679 | * target processor state. | |
680 | */ | |
681 | startup_ipi_hook(phys_apicid, (unsigned long) start_secondary, | |
cb3c8b90 | 682 | (unsigned long)stack_start.sp); |
cb3c8b90 GOC |
683 | |
684 | /* | |
685 | * Run STARTUP IPI loop. | |
686 | */ | |
cfc1b9a6 | 687 | pr_debug("#startup loops: %d.\n", num_starts); |
cb3c8b90 | 688 | |
cb3c8b90 | 689 | for (j = 1; j <= num_starts; j++) { |
cfc1b9a6 | 690 | pr_debug("Sending STARTUP #%d.\n", j); |
593f4a78 MR |
691 | if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */ |
692 | apic_write(APIC_ESR, 0); | |
cb3c8b90 | 693 | apic_read(APIC_ESR); |
cfc1b9a6 | 694 | pr_debug("After apic_write.\n"); |
cb3c8b90 GOC |
695 | |
696 | /* | |
697 | * STARTUP IPI | |
698 | */ | |
699 | ||
700 | /* Target chip */ | |
cb3c8b90 GOC |
701 | /* Boot on the stack */ |
702 | /* Kick the second */ | |
1b374e4d SS |
703 | apic_icr_write(APIC_DM_STARTUP | (start_eip >> 12), |
704 | phys_apicid); | |
cb3c8b90 GOC |
705 | |
706 | /* | |
707 | * Give the other CPU some time to accept the IPI. | |
708 | */ | |
709 | udelay(300); | |
710 | ||
cfc1b9a6 | 711 | pr_debug("Startup point 1.\n"); |
cb3c8b90 | 712 | |
cfc1b9a6 | 713 | pr_debug("Waiting for send to finish...\n"); |
cb3c8b90 GOC |
714 | send_status = safe_apic_wait_icr_idle(); |
715 | ||
716 | /* | |
717 | * Give the other CPU some time to accept the IPI. | |
718 | */ | |
719 | udelay(200); | |
593f4a78 | 720 | if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */ |
cb3c8b90 | 721 | apic_write(APIC_ESR, 0); |
cb3c8b90 GOC |
722 | accept_status = (apic_read(APIC_ESR) & 0xEF); |
723 | if (send_status || accept_status) | |
724 | break; | |
725 | } | |
cfc1b9a6 | 726 | pr_debug("After Startup.\n"); |
cb3c8b90 GOC |
727 | |
728 | if (send_status) | |
729 | printk(KERN_ERR "APIC never delivered???\n"); | |
730 | if (accept_status) | |
731 | printk(KERN_ERR "APIC delivery error (%lx).\n", accept_status); | |
732 | ||
733 | return (send_status | accept_status); | |
734 | } | |
735 | #endif /* WAKE_SECONDARY_VIA_INIT */ | |
736 | ||
737 | struct create_idle { | |
738 | struct work_struct work; | |
739 | struct task_struct *idle; | |
740 | struct completion done; | |
741 | int cpu; | |
742 | }; | |
743 | ||
744 | static void __cpuinit do_fork_idle(struct work_struct *work) | |
745 | { | |
746 | struct create_idle *c_idle = | |
747 | container_of(work, struct create_idle, work); | |
748 | ||
749 | c_idle->idle = fork_idle(c_idle->cpu); | |
750 | complete(&c_idle->done); | |
751 | } | |
752 | ||
f307d25e | 753 | #ifdef CONFIG_X86_64 |
d19fbfdf MS |
754 | |
755 | /* __ref because it's safe to call free_bootmem when after_bootmem == 0. */ | |
756 | static void __ref free_bootmem_pda(struct x8664_pda *oldpda) | |
757 | { | |
758 | if (!after_bootmem) | |
759 | free_bootmem((unsigned long)oldpda, sizeof(*oldpda)); | |
760 | } | |
761 | ||
3461b0af MT |
762 | /* |
763 | * Allocate node local memory for the AP pda. | |
764 | * | |
765 | * Must be called after the _cpu_pda pointer table is initialized. | |
766 | */ | |
7c33b1e6 | 767 | int __cpuinit get_local_pda(int cpu) |
3461b0af MT |
768 | { |
769 | struct x8664_pda *oldpda, *newpda; | |
770 | unsigned long size = sizeof(struct x8664_pda); | |
771 | int node = cpu_to_node(cpu); | |
772 | ||
773 | if (cpu_pda(cpu) && !cpu_pda(cpu)->in_bootmem) | |
774 | return 0; | |
775 | ||
776 | oldpda = cpu_pda(cpu); | |
777 | newpda = kmalloc_node(size, GFP_ATOMIC, node); | |
778 | if (!newpda) { | |
779 | printk(KERN_ERR "Could not allocate node local PDA " | |
780 | "for CPU %d on node %d\n", cpu, node); | |
781 | ||
782 | if (oldpda) | |
783 | return 0; /* have a usable pda */ | |
784 | else | |
785 | return -1; | |
786 | } | |
787 | ||
788 | if (oldpda) { | |
789 | memcpy(newpda, oldpda, size); | |
d19fbfdf | 790 | free_bootmem_pda(oldpda); |
3461b0af MT |
791 | } |
792 | ||
793 | newpda->in_bootmem = 0; | |
794 | cpu_pda(cpu) = newpda; | |
795 | return 0; | |
796 | } | |
f307d25e | 797 | #endif /* CONFIG_X86_64 */ |
3461b0af | 798 | |
cb3c8b90 GOC |
799 | static int __cpuinit do_boot_cpu(int apicid, int cpu) |
800 | /* | |
801 | * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad | |
802 | * (ie clustered apic addressing mode), this is a LOGICAL apic ID. | |
803 | * Returns zero if CPU booted OK, else error code from wakeup_secondary_cpu. | |
804 | */ | |
805 | { | |
806 | unsigned long boot_error = 0; | |
807 | int timeout; | |
808 | unsigned long start_ip; | |
809 | unsigned short nmi_high = 0, nmi_low = 0; | |
810 | struct create_idle c_idle = { | |
811 | .cpu = cpu, | |
812 | .done = COMPLETION_INITIALIZER_ONSTACK(c_idle.done), | |
813 | }; | |
814 | INIT_WORK(&c_idle.work, do_fork_idle); | |
cb3c8b90 | 815 | |
a939098a | 816 | #ifdef CONFIG_X86_64 |
cb3c8b90 | 817 | /* Allocate node local memory for AP pdas */ |
3461b0af MT |
818 | if (cpu > 0) { |
819 | boot_error = get_local_pda(cpu); | |
820 | if (boot_error) | |
821 | goto restore_state; | |
822 | /* if can't get pda memory, can't start cpu */ | |
cb3c8b90 GOC |
823 | } |
824 | #endif | |
825 | ||
826 | alternatives_smp_switch(1); | |
827 | ||
828 | c_idle.idle = get_idle_for_cpu(cpu); | |
829 | ||
830 | /* | |
831 | * We can't use kernel_thread since we must avoid to | |
832 | * reschedule the child. | |
833 | */ | |
834 | if (c_idle.idle) { | |
835 | c_idle.idle->thread.sp = (unsigned long) (((struct pt_regs *) | |
836 | (THREAD_SIZE + task_stack_page(c_idle.idle))) - 1); | |
837 | init_idle(c_idle.idle, cpu); | |
838 | goto do_rest; | |
839 | } | |
840 | ||
841 | if (!keventd_up() || current_is_keventd()) | |
842 | c_idle.work.func(&c_idle.work); | |
843 | else { | |
844 | schedule_work(&c_idle.work); | |
845 | wait_for_completion(&c_idle.done); | |
846 | } | |
847 | ||
848 | if (IS_ERR(c_idle.idle)) { | |
849 | printk("failed fork for CPU %d\n", cpu); | |
850 | return PTR_ERR(c_idle.idle); | |
851 | } | |
852 | ||
853 | set_idle_for_cpu(cpu, c_idle.idle); | |
854 | do_rest: | |
855 | #ifdef CONFIG_X86_32 | |
856 | per_cpu(current_task, cpu) = c_idle.idle; | |
857 | init_gdt(cpu); | |
cb3c8b90 | 858 | /* Stack for startup_32 can be just as for start_secondary onwards */ |
cb3c8b90 GOC |
859 | irq_ctx_init(cpu); |
860 | #else | |
861 | cpu_pda(cpu)->pcurrent = c_idle.idle; | |
cb3c8b90 GOC |
862 | clear_tsk_thread_flag(c_idle.idle, TIF_FORK); |
863 | #endif | |
a939098a | 864 | early_gdt_descr.address = (unsigned long)get_cpu_gdt_table(cpu); |
3e970473 | 865 | initial_code = (unsigned long)start_secondary; |
9cf4f298 | 866 | stack_start.sp = (void *) c_idle.idle->thread.sp; |
cb3c8b90 GOC |
867 | |
868 | /* start_ip had better be page-aligned! */ | |
869 | start_ip = setup_trampoline(); | |
870 | ||
871 | /* So we see what's up */ | |
872 | printk(KERN_INFO "Booting processor %d/%d ip %lx\n", | |
873 | cpu, apicid, start_ip); | |
874 | ||
875 | /* | |
876 | * This grunge runs the startup process for | |
877 | * the targeted processor. | |
878 | */ | |
879 | ||
880 | atomic_set(&init_deasserted, 0); | |
881 | ||
34d05591 | 882 | if (get_uv_system_type() != UV_NON_UNIQUE_APIC) { |
cb3c8b90 | 883 | |
cfc1b9a6 | 884 | pr_debug("Setting warm reset code and vector.\n"); |
cb3c8b90 | 885 | |
34d05591 JS |
886 | store_NMI_vector(&nmi_high, &nmi_low); |
887 | ||
888 | smpboot_setup_warm_reset_vector(start_ip); | |
889 | /* | |
890 | * Be paranoid about clearing APIC errors. | |
891 | */ | |
892 | apic_write(APIC_ESR, 0); | |
893 | apic_read(APIC_ESR); | |
894 | } | |
cb3c8b90 | 895 | |
cb3c8b90 GOC |
896 | /* |
897 | * Starting actual IPI sequence... | |
898 | */ | |
899 | boot_error = wakeup_secondary_cpu(apicid, start_ip); | |
900 | ||
901 | if (!boot_error) { | |
902 | /* | |
903 | * allow APs to start initializing. | |
904 | */ | |
cfc1b9a6 | 905 | pr_debug("Before Callout %d.\n", cpu); |
cb3c8b90 | 906 | cpu_set(cpu, cpu_callout_map); |
cfc1b9a6 | 907 | pr_debug("After Callout %d.\n", cpu); |
cb3c8b90 GOC |
908 | |
909 | /* | |
910 | * Wait 5s total for a response | |
911 | */ | |
912 | for (timeout = 0; timeout < 50000; timeout++) { | |
913 | if (cpu_isset(cpu, cpu_callin_map)) | |
914 | break; /* It has booted */ | |
915 | udelay(100); | |
916 | } | |
917 | ||
918 | if (cpu_isset(cpu, cpu_callin_map)) { | |
919 | /* number CPUs logically, starting from 1 (BSP is 0) */ | |
cfc1b9a6 | 920 | pr_debug("OK.\n"); |
cb3c8b90 GOC |
921 | printk(KERN_INFO "CPU%d: ", cpu); |
922 | print_cpu_info(&cpu_data(cpu)); | |
cfc1b9a6 | 923 | pr_debug("CPU has booted.\n"); |
cb3c8b90 GOC |
924 | } else { |
925 | boot_error = 1; | |
926 | if (*((volatile unsigned char *)trampoline_base) | |
927 | == 0xA5) | |
928 | /* trampoline started but...? */ | |
929 | printk(KERN_ERR "Stuck ??\n"); | |
930 | else | |
931 | /* trampoline code not run */ | |
932 | printk(KERN_ERR "Not responding.\n"); | |
34d05591 JS |
933 | if (get_uv_system_type() != UV_NON_UNIQUE_APIC) |
934 | inquire_remote_apic(apicid); | |
cb3c8b90 GOC |
935 | } |
936 | } | |
6f585e01 | 937 | #ifdef CONFIG_X86_64 |
3461b0af | 938 | restore_state: |
6f585e01 | 939 | #endif |
cb3c8b90 GOC |
940 | if (boot_error) { |
941 | /* Try to put things back the way they were before ... */ | |
23ca4bba | 942 | numa_remove_cpu(cpu); /* was set by numa_add_cpu */ |
cb3c8b90 GOC |
943 | cpu_clear(cpu, cpu_callout_map); /* was set by do_boot_cpu() */ |
944 | cpu_clear(cpu, cpu_initialized); /* was set by cpu_init() */ | |
cb3c8b90 GOC |
945 | cpu_clear(cpu, cpu_present_map); |
946 | per_cpu(x86_cpu_to_apicid, cpu) = BAD_APICID; | |
947 | } | |
948 | ||
949 | /* mark "stuck" area as not stuck */ | |
950 | *((volatile unsigned long *)trampoline_base) = 0; | |
951 | ||
63d38198 AK |
952 | /* |
953 | * Cleanup possible dangling ends... | |
954 | */ | |
955 | smpboot_restore_warm_reset_vector(); | |
956 | ||
cb3c8b90 GOC |
957 | return boot_error; |
958 | } | |
959 | ||
960 | int __cpuinit native_cpu_up(unsigned int cpu) | |
961 | { | |
962 | int apicid = cpu_present_to_apicid(cpu); | |
963 | unsigned long flags; | |
964 | int err; | |
965 | ||
966 | WARN_ON(irqs_disabled()); | |
967 | ||
cfc1b9a6 | 968 | pr_debug("++++++++++++++++++++=_---CPU UP %u\n", cpu); |
cb3c8b90 GOC |
969 | |
970 | if (apicid == BAD_APICID || apicid == boot_cpu_physical_apicid || | |
971 | !physid_isset(apicid, phys_cpu_present_map)) { | |
972 | printk(KERN_ERR "%s: bad cpu %d\n", __func__, cpu); | |
973 | return -EINVAL; | |
974 | } | |
975 | ||
976 | /* | |
977 | * Already booted CPU? | |
978 | */ | |
979 | if (cpu_isset(cpu, cpu_callin_map)) { | |
cfc1b9a6 | 980 | pr_debug("do_boot_cpu %d Already started\n", cpu); |
cb3c8b90 GOC |
981 | return -ENOSYS; |
982 | } | |
983 | ||
984 | /* | |
985 | * Save current MTRR state in case it was changed since early boot | |
986 | * (e.g. by the ACPI SMI) to initialize new CPUs with MTRRs in sync: | |
987 | */ | |
988 | mtrr_save_state(); | |
989 | ||
990 | per_cpu(cpu_state, cpu) = CPU_UP_PREPARE; | |
991 | ||
992 | #ifdef CONFIG_X86_32 | |
993 | /* init low mem mapping */ | |
68db065c | 994 | clone_pgd_range(swapper_pg_dir, swapper_pg_dir + KERNEL_PGD_BOUNDARY, |
61165d7a | 995 | min_t(unsigned long, KERNEL_PGD_PTRS, KERNEL_PGD_BOUNDARY)); |
cb3c8b90 | 996 | flush_tlb_all(); |
61165d7a | 997 | low_mappings = 1; |
cb3c8b90 GOC |
998 | |
999 | err = do_boot_cpu(apicid, cpu); | |
61165d7a HD |
1000 | |
1001 | zap_low_mappings(); | |
1002 | low_mappings = 0; | |
1003 | #else | |
1004 | err = do_boot_cpu(apicid, cpu); | |
1005 | #endif | |
1006 | if (err) { | |
cfc1b9a6 | 1007 | pr_debug("do_boot_cpu failed %d\n", err); |
61165d7a | 1008 | return -EIO; |
cb3c8b90 GOC |
1009 | } |
1010 | ||
1011 | /* | |
1012 | * Check TSC synchronization with the AP (keep irqs disabled | |
1013 | * while doing so): | |
1014 | */ | |
1015 | local_irq_save(flags); | |
1016 | check_tsc_sync_source(cpu); | |
1017 | local_irq_restore(flags); | |
1018 | ||
7c04e64a | 1019 | while (!cpu_online(cpu)) { |
cb3c8b90 GOC |
1020 | cpu_relax(); |
1021 | touch_nmi_watchdog(); | |
1022 | } | |
1023 | ||
1024 | return 0; | |
1025 | } | |
1026 | ||
8aef135c GOC |
1027 | /* |
1028 | * Fall back to non SMP mode after errors. | |
1029 | * | |
1030 | * RED-PEN audit/test this more. I bet there is more state messed up here. | |
1031 | */ | |
1032 | static __init void disable_smp(void) | |
1033 | { | |
1034 | cpu_present_map = cpumask_of_cpu(0); | |
1035 | cpu_possible_map = cpumask_of_cpu(0); | |
8aef135c | 1036 | smpboot_clear_io_apic_irqs(); |
0f385d1d | 1037 | |
8aef135c | 1038 | if (smp_found_config) |
b6df1b8b | 1039 | physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map); |
8aef135c | 1040 | else |
b6df1b8b | 1041 | physid_set_mask_of_physid(0, &phys_cpu_present_map); |
8aef135c GOC |
1042 | map_cpu_to_logical_apicid(); |
1043 | cpu_set(0, per_cpu(cpu_sibling_map, 0)); | |
1044 | cpu_set(0, per_cpu(cpu_core_map, 0)); | |
1045 | } | |
1046 | ||
1047 | /* | |
1048 | * Various sanity checks. | |
1049 | */ | |
1050 | static int __init smp_sanity_check(unsigned max_cpus) | |
1051 | { | |
ac23d4ee | 1052 | preempt_disable(); |
a58f03b0 YL |
1053 | |
1054 | #if defined(CONFIG_X86_PC) && defined(CONFIG_X86_32) | |
1055 | if (def_to_bigsmp && nr_cpu_ids > 8) { | |
1056 | unsigned int cpu; | |
1057 | unsigned nr; | |
1058 | ||
1059 | printk(KERN_WARNING | |
1060 | "More than 8 CPUs detected - skipping them.\n" | |
1061 | "Use CONFIG_X86_GENERICARCH and CONFIG_X86_BIGSMP.\n"); | |
1062 | ||
1063 | nr = 0; | |
1064 | for_each_present_cpu(cpu) { | |
1065 | if (nr >= 8) | |
1066 | cpu_clear(cpu, cpu_present_map); | |
1067 | nr++; | |
1068 | } | |
1069 | ||
1070 | nr = 0; | |
1071 | for_each_possible_cpu(cpu) { | |
1072 | if (nr >= 8) | |
1073 | cpu_clear(cpu, cpu_possible_map); | |
1074 | nr++; | |
1075 | } | |
1076 | ||
1077 | nr_cpu_ids = 8; | |
1078 | } | |
1079 | #endif | |
1080 | ||
8aef135c GOC |
1081 | if (!physid_isset(hard_smp_processor_id(), phys_cpu_present_map)) { |
1082 | printk(KERN_WARNING "weird, boot CPU (#%d) not listed" | |
1083 | "by the BIOS.\n", hard_smp_processor_id()); | |
1084 | physid_set(hard_smp_processor_id(), phys_cpu_present_map); | |
1085 | } | |
1086 | ||
1087 | /* | |
1088 | * If we couldn't find an SMP configuration at boot time, | |
1089 | * get out of here now! | |
1090 | */ | |
1091 | if (!smp_found_config && !acpi_lapic) { | |
ac23d4ee | 1092 | preempt_enable(); |
8aef135c GOC |
1093 | printk(KERN_NOTICE "SMP motherboard not detected.\n"); |
1094 | disable_smp(); | |
1095 | if (APIC_init_uniprocessor()) | |
1096 | printk(KERN_NOTICE "Local APIC not detected." | |
1097 | " Using dummy APIC emulation.\n"); | |
1098 | return -1; | |
1099 | } | |
1100 | ||
1101 | /* | |
1102 | * Should not be necessary because the MP table should list the boot | |
1103 | * CPU too, but we do it for the sake of robustness anyway. | |
1104 | */ | |
1105 | if (!check_phys_apicid_present(boot_cpu_physical_apicid)) { | |
1106 | printk(KERN_NOTICE | |
1107 | "weird, boot CPU (#%d) not listed by the BIOS.\n", | |
1108 | boot_cpu_physical_apicid); | |
1109 | physid_set(hard_smp_processor_id(), phys_cpu_present_map); | |
1110 | } | |
ac23d4ee | 1111 | preempt_enable(); |
8aef135c GOC |
1112 | |
1113 | /* | |
1114 | * If we couldn't find a local APIC, then get out of here now! | |
1115 | */ | |
1116 | if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid]) && | |
1117 | !cpu_has_apic) { | |
1118 | printk(KERN_ERR "BIOS bug, local APIC #%d not detected!...\n", | |
1119 | boot_cpu_physical_apicid); | |
1120 | printk(KERN_ERR "... forcing use of dummy APIC emulation." | |
1121 | "(tell your hw vendor)\n"); | |
1122 | smpboot_clear_io_apic(); | |
1123 | return -1; | |
1124 | } | |
1125 | ||
1126 | verify_local_APIC(); | |
1127 | ||
1128 | /* | |
1129 | * If SMP should be disabled, then really disable it! | |
1130 | */ | |
1131 | if (!max_cpus) { | |
73d08e63 | 1132 | printk(KERN_INFO "SMP mode deactivated.\n"); |
8aef135c | 1133 | smpboot_clear_io_apic(); |
d54db1ac MR |
1134 | |
1135 | localise_nmi_watchdog(); | |
1136 | ||
e90955c2 | 1137 | connect_bsp_APIC(); |
e90955c2 JB |
1138 | setup_local_APIC(); |
1139 | end_local_APIC_setup(); | |
8aef135c GOC |
1140 | return -1; |
1141 | } | |
1142 | ||
1143 | return 0; | |
1144 | } | |
1145 | ||
1146 | static void __init smp_cpu_index_default(void) | |
1147 | { | |
1148 | int i; | |
1149 | struct cpuinfo_x86 *c; | |
1150 | ||
7c04e64a | 1151 | for_each_possible_cpu(i) { |
8aef135c GOC |
1152 | c = &cpu_data(i); |
1153 | /* mark all to hotplug */ | |
1154 | c->cpu_index = NR_CPUS; | |
1155 | } | |
1156 | } | |
1157 | ||
1158 | /* | |
1159 | * Prepare for SMP bootup. The MP table or ACPI has been read | |
1160 | * earlier. Just do some sanity checking here and enable APIC mode. | |
1161 | */ | |
1162 | void __init native_smp_prepare_cpus(unsigned int max_cpus) | |
1163 | { | |
deef3250 | 1164 | preempt_disable(); |
8aef135c GOC |
1165 | smp_cpu_index_default(); |
1166 | current_cpu_data = boot_cpu_data; | |
1167 | cpu_callin_map = cpumask_of_cpu(0); | |
1168 | mb(); | |
1169 | /* | |
1170 | * Setup boot CPU information | |
1171 | */ | |
1172 | smp_store_cpu_info(0); /* Final full version of the data */ | |
1b374e4d | 1173 | #ifdef CONFIG_X86_32 |
8aef135c | 1174 | boot_cpu_logical_apicid = logical_smp_processor_id(); |
1b374e4d | 1175 | #endif |
8aef135c GOC |
1176 | current_thread_info()->cpu = 0; /* needed? */ |
1177 | set_cpu_sibling_map(0); | |
1178 | ||
6e1cb38a SS |
1179 | #ifdef CONFIG_X86_64 |
1180 | enable_IR_x2apic(); | |
1181 | setup_apic_routing(); | |
1182 | #endif | |
1183 | ||
8aef135c GOC |
1184 | if (smp_sanity_check(max_cpus) < 0) { |
1185 | printk(KERN_INFO "SMP disabled\n"); | |
1186 | disable_smp(); | |
deef3250 | 1187 | goto out; |
8aef135c GOC |
1188 | } |
1189 | ||
ac23d4ee | 1190 | preempt_disable(); |
4c9961d5 | 1191 | if (read_apic_id() != boot_cpu_physical_apicid) { |
8aef135c | 1192 | panic("Boot APIC ID in local APIC unexpected (%d vs %d)", |
4c9961d5 | 1193 | read_apic_id(), boot_cpu_physical_apicid); |
8aef135c GOC |
1194 | /* Or can we switch back to PIC here? */ |
1195 | } | |
ac23d4ee | 1196 | preempt_enable(); |
8aef135c | 1197 | |
8aef135c | 1198 | connect_bsp_APIC(); |
b5841765 | 1199 | |
8aef135c GOC |
1200 | /* |
1201 | * Switch from PIC to APIC mode. | |
1202 | */ | |
1203 | setup_local_APIC(); | |
1204 | ||
1205 | #ifdef CONFIG_X86_64 | |
1206 | /* | |
1207 | * Enable IO APIC before setting up error vector | |
1208 | */ | |
1209 | if (!skip_ioapic_setup && nr_ioapics) | |
1210 | enable_IO_APIC(); | |
1211 | #endif | |
1212 | end_local_APIC_setup(); | |
1213 | ||
1214 | map_cpu_to_logical_apicid(); | |
1215 | ||
1216 | setup_portio_remap(); | |
1217 | ||
1218 | smpboot_setup_io_apic(); | |
1219 | /* | |
1220 | * Set up local APIC timer on boot CPU. | |
1221 | */ | |
1222 | ||
1223 | printk(KERN_INFO "CPU%d: ", 0); | |
1224 | print_cpu_info(&cpu_data(0)); | |
1225 | setup_boot_clock(); | |
c4bd1fda MS |
1226 | |
1227 | if (is_uv_system()) | |
1228 | uv_system_init(); | |
deef3250 IM |
1229 | out: |
1230 | preempt_enable(); | |
8aef135c | 1231 | } |
a8db8453 GOC |
1232 | /* |
1233 | * Early setup to make printk work. | |
1234 | */ | |
1235 | void __init native_smp_prepare_boot_cpu(void) | |
1236 | { | |
1237 | int me = smp_processor_id(); | |
1238 | #ifdef CONFIG_X86_32 | |
1239 | init_gdt(me); | |
a8db8453 | 1240 | #endif |
a939098a | 1241 | switch_to_new_gdt(); |
a8db8453 GOC |
1242 | /* already set me in cpu_online_map in boot_cpu_init() */ |
1243 | cpu_set(me, cpu_callout_map); | |
1244 | per_cpu(cpu_state, me) = CPU_ONLINE; | |
1245 | } | |
1246 | ||
83f7eb9c GOC |
1247 | void __init native_smp_cpus_done(unsigned int max_cpus) |
1248 | { | |
cfc1b9a6 | 1249 | pr_debug("Boot done.\n"); |
83f7eb9c GOC |
1250 | |
1251 | impress_friends(); | |
1252 | smp_checks(); | |
1253 | #ifdef CONFIG_X86_IO_APIC | |
1254 | setup_ioapic_dest(); | |
1255 | #endif | |
1256 | check_nmi_watchdog(); | |
83f7eb9c GOC |
1257 | } |
1258 | ||
68a1c3f8 | 1259 | #ifdef CONFIG_HOTPLUG_CPU |
2cd9fb71 | 1260 | |
a4928cff | 1261 | static void remove_siblinginfo(int cpu) |
768d9505 GC |
1262 | { |
1263 | int sibling; | |
1264 | struct cpuinfo_x86 *c = &cpu_data(cpu); | |
1265 | ||
334ef7a7 | 1266 | for_each_cpu_mask_nr(sibling, per_cpu(cpu_core_map, cpu)) { |
768d9505 GC |
1267 | cpu_clear(cpu, per_cpu(cpu_core_map, sibling)); |
1268 | /*/ | |
1269 | * last thread sibling in this cpu core going down | |
1270 | */ | |
1271 | if (cpus_weight(per_cpu(cpu_sibling_map, cpu)) == 1) | |
1272 | cpu_data(sibling).booted_cores--; | |
1273 | } | |
1274 | ||
334ef7a7 | 1275 | for_each_cpu_mask_nr(sibling, per_cpu(cpu_sibling_map, cpu)) |
768d9505 GC |
1276 | cpu_clear(cpu, per_cpu(cpu_sibling_map, sibling)); |
1277 | cpus_clear(per_cpu(cpu_sibling_map, cpu)); | |
1278 | cpus_clear(per_cpu(cpu_core_map, cpu)); | |
1279 | c->phys_proc_id = 0; | |
1280 | c->cpu_core_id = 0; | |
1281 | cpu_clear(cpu, cpu_sibling_setup_map); | |
1282 | } | |
68a1c3f8 | 1283 | |
c5562fae | 1284 | static int additional_cpus __initdata = -1; |
68a1c3f8 GC |
1285 | |
1286 | static __init int setup_additional_cpus(char *s) | |
1287 | { | |
1288 | return s && get_option(&s, &additional_cpus) ? 0 : -EINVAL; | |
1289 | } | |
1290 | early_param("additional_cpus", setup_additional_cpus); | |
1291 | ||
1292 | /* | |
1293 | * cpu_possible_map should be static, it cannot change as cpu's | |
1294 | * are onlined, or offlined. The reason is per-cpu data-structures | |
1295 | * are allocated by some modules at init time, and dont expect to | |
1296 | * do this dynamically on cpu arrival/departure. | |
1297 | * cpu_present_map on the other hand can change dynamically. | |
1298 | * In case when cpu_hotplug is not compiled, then we resort to current | |
1299 | * behaviour, which is cpu_possible == cpu_present. | |
1300 | * - Ashok Raj | |
1301 | * | |
1302 | * Three ways to find out the number of additional hotplug CPUs: | |
1303 | * - If the BIOS specified disabled CPUs in ACPI/mptables use that. | |
1304 | * - The user can overwrite it with additional_cpus=NUM | |
1305 | * - Otherwise don't reserve additional CPUs. | |
1306 | * We do this because additional CPUs waste a lot of memory. | |
1307 | * -AK | |
1308 | */ | |
1309 | __init void prefill_possible_map(void) | |
1310 | { | |
1311 | int i; | |
1312 | int possible; | |
1313 | ||
329513a3 YL |
1314 | /* no processor from mptable or madt */ |
1315 | if (!num_processors) | |
1316 | num_processors = 1; | |
1317 | ||
68a1c3f8 GC |
1318 | if (additional_cpus == -1) { |
1319 | if (disabled_cpus > 0) | |
1320 | additional_cpus = disabled_cpus; | |
1321 | else | |
1322 | additional_cpus = 0; | |
1323 | } | |
2bd455db | 1324 | |
68a1c3f8 GC |
1325 | possible = num_processors + additional_cpus; |
1326 | if (possible > NR_CPUS) | |
1327 | possible = NR_CPUS; | |
1328 | ||
1329 | printk(KERN_INFO "SMP: Allowing %d CPUs, %d hotplug CPUs\n", | |
1330 | possible, max_t(int, possible - num_processors, 0)); | |
1331 | ||
1332 | for (i = 0; i < possible; i++) | |
1333 | cpu_set(i, cpu_possible_map); | |
3461b0af MT |
1334 | |
1335 | nr_cpu_ids = possible; | |
68a1c3f8 | 1336 | } |
69c18c15 GC |
1337 | |
1338 | static void __ref remove_cpu_from_maps(int cpu) | |
1339 | { | |
1340 | cpu_clear(cpu, cpu_online_map); | |
69c18c15 GC |
1341 | cpu_clear(cpu, cpu_callout_map); |
1342 | cpu_clear(cpu, cpu_callin_map); | |
1343 | /* was set by cpu_init() */ | |
29cbeb0e | 1344 | cpu_clear(cpu, cpu_initialized); |
23ca4bba | 1345 | numa_remove_cpu(cpu); |
69c18c15 GC |
1346 | } |
1347 | ||
8227dce7 | 1348 | void cpu_disable_common(void) |
69c18c15 GC |
1349 | { |
1350 | int cpu = smp_processor_id(); | |
69c18c15 GC |
1351 | /* |
1352 | * HACK: | |
1353 | * Allow any queued timer interrupts to get serviced | |
1354 | * This is only a temporary solution until we cleanup | |
1355 | * fixup_irqs as we do for IA64. | |
1356 | */ | |
1357 | local_irq_enable(); | |
1358 | mdelay(1); | |
1359 | ||
1360 | local_irq_disable(); | |
1361 | remove_siblinginfo(cpu); | |
1362 | ||
1363 | /* It's now safe to remove this processor from the online map */ | |
d388e5fd | 1364 | lock_vector_lock(); |
69c18c15 | 1365 | remove_cpu_from_maps(cpu); |
d388e5fd | 1366 | unlock_vector_lock(); |
69c18c15 | 1367 | fixup_irqs(cpu_online_map); |
8227dce7 AN |
1368 | } |
1369 | ||
1370 | int native_cpu_disable(void) | |
1371 | { | |
1372 | int cpu = smp_processor_id(); | |
1373 | ||
1374 | /* | |
1375 | * Perhaps use cpufreq to drop frequency, but that could go | |
1376 | * into generic code. | |
1377 | * | |
1378 | * We won't take down the boot processor on i386 due to some | |
1379 | * interrupts only being able to be serviced by the BSP. | |
1380 | * Especially so if we're not using an IOAPIC -zwane | |
1381 | */ | |
1382 | if (cpu == 0) | |
1383 | return -EBUSY; | |
1384 | ||
1385 | if (nmi_watchdog == NMI_LOCAL_APIC) | |
1386 | stop_apic_nmi_watchdog(NULL); | |
1387 | clear_local_APIC(); | |
1388 | ||
1389 | cpu_disable_common(); | |
69c18c15 GC |
1390 | return 0; |
1391 | } | |
1392 | ||
93be71b6 | 1393 | void native_cpu_die(unsigned int cpu) |
69c18c15 GC |
1394 | { |
1395 | /* We don't do anything here: idle task is faking death itself. */ | |
1396 | unsigned int i; | |
1397 | ||
1398 | for (i = 0; i < 10; i++) { | |
1399 | /* They ack this in play_dead by setting CPU_DEAD */ | |
1400 | if (per_cpu(cpu_state, cpu) == CPU_DEAD) { | |
1401 | printk(KERN_INFO "CPU %d is now offline\n", cpu); | |
1402 | if (1 == num_online_cpus()) | |
1403 | alternatives_smp_switch(0); | |
1404 | return; | |
1405 | } | |
1406 | msleep(100); | |
1407 | } | |
1408 | printk(KERN_ERR "CPU %u didn't die...\n", cpu); | |
1409 | } | |
a21f5d88 AN |
1410 | |
1411 | void play_dead_common(void) | |
1412 | { | |
1413 | idle_task_exit(); | |
1414 | reset_lazy_tlbstate(); | |
1415 | irq_ctx_exit(raw_smp_processor_id()); | |
07bbc16a | 1416 | c1e_remove_cpu(raw_smp_processor_id()); |
a21f5d88 AN |
1417 | |
1418 | mb(); | |
1419 | /* Ack it */ | |
1420 | __get_cpu_var(cpu_state) = CPU_DEAD; | |
1421 | ||
1422 | /* | |
1423 | * With physical CPU hotplug, we should halt the cpu | |
1424 | */ | |
1425 | local_irq_disable(); | |
1426 | } | |
1427 | ||
1428 | void native_play_dead(void) | |
1429 | { | |
1430 | play_dead_common(); | |
1431 | wbinvd_halt(); | |
1432 | } | |
1433 | ||
69c18c15 | 1434 | #else /* ... !CONFIG_HOTPLUG_CPU */ |
93be71b6 | 1435 | int native_cpu_disable(void) |
69c18c15 GC |
1436 | { |
1437 | return -ENOSYS; | |
1438 | } | |
1439 | ||
93be71b6 | 1440 | void native_cpu_die(unsigned int cpu) |
69c18c15 GC |
1441 | { |
1442 | /* We said "no" in __cpu_disable */ | |
1443 | BUG(); | |
1444 | } | |
a21f5d88 AN |
1445 | |
1446 | void native_play_dead(void) | |
1447 | { | |
1448 | BUG(); | |
1449 | } | |
1450 | ||
68a1c3f8 | 1451 | #endif |