blob: c6c993f4c4158a9b0bf397d7c80435e40af361d1 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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 *
Andi Kleena8ab26f2005-04-16 15:25:19 -070015 * This code is released under the GNU General Public License version 2
Linus Torvalds1da177e2005-04-16 15:20:36 -070016 *
17 * Fixes
18 * Felix Koop : NR_CPUS used properly
19 * Jose Renau : Handle single CPU case.
20 * Alan Cox : By repeated request 8) - Total BogoMIP report.
21 * Greg Wright : Fix for kernel stacks panic.
22 * Erich Boleyn : MP v1.4 and additional changes.
23 * Matthias Sattler : Changes for 2.1 kernel map.
24 * Michel Lespinasse : Changes for 2.1 kernel map.
25 * Michael Chastain : Change trampoline.S to gnu as.
26 * Alan Cox : Dumb bug: 'B' step PPro's are fine
27 * Ingo Molnar : Added APIC timers, based on code
28 * from Jose Renau
29 * Ingo Molnar : various cleanups and rewrites
30 * Tigran Aivazian : fixed "0.00 in /proc/uptime on SMP" bug.
31 * Maciej W. Rozycki : Bits for genuine 82489DX APICs
32 * Andi Kleen : Changed for SMP boot into long mode.
Andi Kleena8ab26f2005-04-16 15:25:19 -070033 * Rusty Russell : Hacked into shape for new "hotplug" boot process.
34 * Andi Kleen : Converted to new state machine.
35 * Various cleanups.
36 * Probably mostly hotplug CPU ready now.
Ashok Raj76e4f662005-06-25 14:55:00 -070037 * Ashok Raj : CPU hotplug support
Linus Torvalds1da177e2005-04-16 15:20:36 -070038 */
39
Andi Kleena8ab26f2005-04-16 15:25:19 -070040
Linus Torvalds1da177e2005-04-16 15:20:36 -070041#include <linux/init.h>
42
43#include <linux/mm.h>
44#include <linux/kernel_stat.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070045#include <linux/bootmem.h>
46#include <linux/thread_info.h>
47#include <linux/module.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070048#include <linux/delay.h>
49#include <linux/mc146818rtc.h>
Andrew Mortona3bc0db2006-09-25 23:32:33 -070050#include <linux/smp.h>
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -070051#include <linux/kdebug.h>
Andrew Mortona3bc0db2006-09-25 23:32:33 -070052
Linus Torvalds1da177e2005-04-16 15:20:36 -070053#include <asm/mtrr.h>
54#include <asm/pgalloc.h>
55#include <asm/desc.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070056#include <asm/tlbflush.h>
57#include <asm/proto.h>
Andi Kleen75152112005-05-16 21:53:34 -070058#include <asm/nmi.h>
Al Viro9cdd3042005-09-12 18:49:25 +020059#include <asm/irq.h>
60#include <asm/hw_irq.h>
Ravikiran G Thirumalai488fc082006-02-07 12:58:23 -080061#include <asm/numa.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070062
Glauber de Oliveira Costa8d770102008-03-19 14:25:31 -030063#include <mach_wakecpu.h>
64
Andi Kleena8ab26f2005-04-16 15:25:19 -070065/* Set when the idlers are all forked */
66int smp_threads_ready;
67
Ashok Raj76e4f662005-06-25 14:55:00 -070068/* State of each CPU */
69DEFINE_PER_CPU(int, cpu_state) = { 0 };
70
71/*
72 * Store all idle threads, this can be reused instead of creating
73 * a new thread. Also avoids complicated thread destroy functionality
74 * for idle threads.
75 */
travis@sgi.com24b0d222008-01-30 13:33:11 +010076#ifdef CONFIG_HOTPLUG_CPU
77/*
78 * Needed only for CONFIG_HOTPLUG_CPU because __cpuinitdata is
79 * removed after init for !CONFIG_HOTPLUG_CPU.
80 */
81static DEFINE_PER_CPU(struct task_struct *, idle_thread_array);
82#define get_idle_for_cpu(x) (per_cpu(idle_thread_array, x))
83#define set_idle_for_cpu(x,p) (per_cpu(idle_thread_array, x) = (p))
84#else
Ashok Raj76e4f662005-06-25 14:55:00 -070085struct task_struct *idle_thread_array[NR_CPUS] __cpuinitdata ;
Ashok Raj76e4f662005-06-25 14:55:00 -070086#define get_idle_for_cpu(x) (idle_thread_array[(x)])
87#define set_idle_for_cpu(x,p) (idle_thread_array[(x)] = (p))
travis@sgi.com24b0d222008-01-30 13:33:11 +010088#endif
89
Andi Kleena8ab26f2005-04-16 15:25:19 -070090static atomic_t init_deasserted __cpuinitdata;
91
92/*
93 * Report back to the Boot Processor.
94 * Running on AP.
95 */
96void __cpuinit smp_callin(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -070097{
98 int cpuid, phys_id;
99 unsigned long timeout;
100
101 /*
102 * If waken up by an INIT in an 82489DX configuration
103 * we may get here before an INIT-deassert IPI reaches
104 * our local APIC. We have to wait for the IPI or we'll
105 * lock up on an APIC access.
106 */
Glauber Costae90009b2008-03-03 14:13:13 -0300107 wait_for_init_deassert(&init_deasserted);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108
109 /*
110 * (This works even if the APIC is not enabled.)
111 */
112 phys_id = GET_APIC_ID(apic_read(APIC_ID));
113 cpuid = smp_processor_id();
114 if (cpu_isset(cpuid, cpu_callin_map)) {
115 panic("smp_callin: phys CPU#%d, CPU#%d already present??\n",
116 phys_id, cpuid);
117 }
118 Dprintk("CPU#%d (phys ID: %d) waiting for CALLOUT\n", cpuid, phys_id);
119
120 /*
121 * STARTUP IPIs are fragile beasts as they might sometimes
122 * trigger some glue motherboard logic. Complete APIC bus
123 * silence for 1 second, this overestimates the time the
124 * boot CPU is spending to send the up to 2 STARTUP IPIs
125 * by a factor of two. This should be enough.
126 */
127
128 /*
129 * Waiting 2s total for startup (udelay is not yet working)
130 */
131 timeout = jiffies + 2*HZ;
132 while (time_before(jiffies, timeout)) {
133 /*
134 * Has the boot CPU finished it's STARTUP sequence?
135 */
136 if (cpu_isset(cpuid, cpu_callout_map))
137 break;
Andi Kleena8ab26f2005-04-16 15:25:19 -0700138 cpu_relax();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 }
140
141 if (!time_before(jiffies, timeout)) {
142 panic("smp_callin: CPU%d started up but did not get a callout!\n",
143 cpuid);
144 }
145
146 /*
147 * the boot CPU has finished the init stage and is spinning
148 * on callin_map until we finish. We are free to set up this
149 * CPU, first the APIC. (this is probably redundant on most
150 * boards)
151 */
152
153 Dprintk("CALLIN, before setup_local_APIC().\n");
154 setup_local_APIC();
Andi Kleen739f33b2008-01-30 13:30:40 +0100155 end_local_APIC_setup();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 /*
158 * Get our bogomips.
Andi Kleenb4452212005-09-12 18:49:24 +0200159 *
160 * Need to enable IRQs because it can take longer and then
161 * the NMI watchdog might kill us.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 */
Andi Kleenb4452212005-09-12 18:49:24 +0200163 local_irq_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 calibrate_delay();
Andi Kleenb4452212005-09-12 18:49:24 +0200165 local_irq_disable();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166 Dprintk("Stack at about %p\n",&cpuid);
167
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168 /*
169 * Save our processor parameters
170 */
171 smp_store_cpu_info(cpuid);
172
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 /*
174 * Allow the master to continue.
175 */
176 cpu_set(cpuid, cpu_callin_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177}
178
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179/*
Andi Kleena8ab26f2005-04-16 15:25:19 -0700180 * Setup code on secondary processor (after comming out of the trampoline)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 */
Andi Kleena8ab26f2005-04-16 15:25:19 -0700182void __cpuinit start_secondary(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183{
184 /*
185 * Dont put anything before smp_callin(), SMP
186 * booting is too fragile that we want to limit the
187 * things done here to the most necessary things.
188 */
189 cpu_init();
Nick Piggin5bfb5d62005-11-08 21:39:01 -0800190 preempt_disable();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 smp_callin();
192
193 /* otherwise gcc will move up the smp_processor_id before the cpu_init */
194 barrier();
195
Ingo Molnar95492e42007-02-16 01:27:34 -0800196 /*
197 * Check TSC sync first:
198 */
199 check_tsc_sync_target();
200
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 if (nmi_watchdog == NMI_IO_APIC) {
202 disable_8259A_irq(0);
Jan Beuliche9427102008-01-30 13:31:24 +0100203 enable_NMI_through_LVT0();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 enable_8259A_irq(0);
205 }
206
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 /*
Ashok Rajcb0cd8d2005-06-25 14:55:01 -0700208 * The sibling maps must be set before turing the online map on for
209 * this cpu
210 */
211 set_cpu_sibling_map(smp_processor_id());
212
213 /*
Ashok Raj884d9e42005-06-25 14:55:02 -0700214 * We need to hold call_lock, so there is no inconsistency
215 * between the time smp_call_function() determines number of
Simon Arlott676b1852007-10-20 01:25:36 +0200216 * IPI recipients, and the time when the determination is made
Ashok Raj884d9e42005-06-25 14:55:02 -0700217 * for which cpus receive the IPI in genapic_flat.c. Holding this
218 * lock helps us to not include this cpu in a currently in progress
219 * smp_call_function().
220 */
221 lock_ipi_call_lock();
Eric W. Biederman70a0a532006-10-25 01:00:23 +0200222 spin_lock(&vector_lock);
Ashok Raj884d9e42005-06-25 14:55:02 -0700223
Eric W. Biederman70a0a532006-10-25 01:00:23 +0200224 /* Setup the per cpu irq handling data structures */
225 __setup_vector_irq(smp_processor_id());
Ashok Raj884d9e42005-06-25 14:55:02 -0700226 /*
Andi Kleena8ab26f2005-04-16 15:25:19 -0700227 * Allow the master to continue.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 */
Glauber Costafc25da92008-03-03 14:13:04 -0300229 spin_unlock(&vector_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 cpu_set(smp_processor_id(), cpu_online_map);
Ashok Raj884d9e42005-06-25 14:55:02 -0700231 unlock_ipi_call_lock();
232
Glauber de Oliveira Costa5733f622008-03-19 14:25:09 -0300233 per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE;
234
Glauber de Oliveira Costa746ef0c2008-01-30 13:31:11 +0100235 setup_secondary_clock();
Thomas Gleixner3ac508b2007-10-14 22:57:45 +0200236
Glauber de Oliveira Costafa8004d2008-03-19 14:25:12 -0300237 wmb();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 cpu_idle();
239}
240
Andi Kleena8ab26f2005-04-16 15:25:19 -0700241extern volatile unsigned long init_rsp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242extern void (*initial_code)(void);
243
Olaf Hering44456d32005-07-27 11:45:17 -0700244#ifdef APIC_DEBUG
Glauber de Oliveira Costa8d770102008-03-19 14:25:31 -0300245static void __inquire_remote_apic(int apicid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246{
247 unsigned i, regs[] = { APIC_ID >> 4, APIC_LVR >> 4, APIC_SPIV >> 4 };
248 char *names[] = { "ID", "VERSION", "SPIV" };
Fernando Luis VazquezCao3144c332007-05-02 19:27:17 +0200249 int timeout;
Thomas Gleixner3c6bb072008-01-30 13:30:15 +0100250 u32 status;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251
252 printk(KERN_INFO "Inquiring remote APIC #%d...\n", apicid);
253
Alejandro Martinez Ruiz4d022ad2007-10-17 14:38:58 +0200254 for (i = 0; i < ARRAY_SIZE(regs); i++) {
Thomas Gleixner3c6bb072008-01-30 13:30:15 +0100255 printk(KERN_INFO "... APIC #%d %s: ", apicid, names[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256
257 /*
258 * Wait for idle.
259 */
Fernando Luis VazquezCao3144c332007-05-02 19:27:17 +0200260 status = safe_apic_wait_icr_idle();
261 if (status)
Thomas Gleixner3c6bb072008-01-30 13:30:15 +0100262 printk(KERN_CONT
263 "a previous APIC delivery may have failed\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264
Glauber de Oliveira Costa1af8a0c2008-03-19 14:24:58 -0300265 apic_write_around(APIC_ICR2, SET_APIC_DEST_FIELD(apicid));
266 apic_write_around(APIC_ICR, APIC_DM_REMRD | regs[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267
268 timeout = 0;
269 do {
270 udelay(100);
271 status = apic_read(APIC_ICR) & APIC_ICR_RR_MASK;
272 } while (status == APIC_ICR_RR_INPROG && timeout++ < 1000);
273
274 switch (status) {
275 case APIC_ICR_RR_VALID:
276 status = apic_read(APIC_RRR);
Thomas Gleixner3c6bb072008-01-30 13:30:15 +0100277 printk(KERN_CONT "%08x\n", status);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278 break;
279 default:
Thomas Gleixner3c6bb072008-01-30 13:30:15 +0100280 printk(KERN_CONT "failed\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 }
282 }
283}
284#endif
285
Andi Kleena8ab26f2005-04-16 15:25:19 -0700286/*
287 * Kick the secondary to wake up.
288 */
289static int __cpuinit wakeup_secondary_via_INIT(int phys_apicid, unsigned int start_rip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290{
Fernando Luis VazquezCaoea8c7332007-05-02 19:27:17 +0200291 unsigned long send_status, accept_status = 0;
292 int maxlvt, num_starts, j;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293
Glauber de Oliveira Costa148a30f2008-03-19 14:25:11 -0300294 /*
295 * Be paranoid about clearing APIC errors.
296 */
297 if (APIC_INTEGRATED(apic_version[phys_apicid])) {
298 apic_read_around(APIC_SPIV);
299 apic_write(APIC_ESR, 0);
300 apic_read(APIC_ESR);
301 }
302
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303 Dprintk("Asserting INIT.\n");
304
305 /*
306 * Turn INIT on target chip
307 */
Glauber de Oliveira Costa1af8a0c2008-03-19 14:24:58 -0300308 apic_write_around(APIC_ICR2, SET_APIC_DEST_FIELD(phys_apicid));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309
310 /*
311 * Send IPI
312 */
Glauber de Oliveira Costa1af8a0c2008-03-19 14:24:58 -0300313 apic_write_around(APIC_ICR, APIC_INT_LEVELTRIG | APIC_INT_ASSERT
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 | APIC_DM_INIT);
315
316 Dprintk("Waiting for send to finish...\n");
Fernando Luis VazquezCaoea8c7332007-05-02 19:27:17 +0200317 send_status = safe_apic_wait_icr_idle();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318
319 mdelay(10);
320
321 Dprintk("Deasserting INIT.\n");
322
323 /* Target chip */
Glauber de Oliveira Costa1af8a0c2008-03-19 14:24:58 -0300324 apic_write_around(APIC_ICR2, SET_APIC_DEST_FIELD(phys_apicid));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325
326 /* Send IPI */
Glauber de Oliveira Costa1af8a0c2008-03-19 14:24:58 -0300327 apic_write_around(APIC_ICR, APIC_INT_LEVELTRIG | APIC_DM_INIT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328
329 Dprintk("Waiting for send to finish...\n");
Fernando Luis VazquezCaoea8c7332007-05-02 19:27:17 +0200330 send_status = safe_apic_wait_icr_idle();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331
Benjamin LaHaisef2ecfab2006-01-11 22:43:03 +0100332 mb();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 atomic_set(&init_deasserted, 1);
334
Glauber de Oliveira Costa148a30f2008-03-19 14:25:11 -0300335 if (APIC_INTEGRATED(apic_version[phys_apicid]))
336 num_starts = 2;
337 else
338 num_starts = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339
340 /*
Glauber de Oliveira Costad0173ae2008-03-19 14:24:59 -0300341 * Paravirt / VMI wants a startup IPI hook here to set up the
342 * target processor state.
343 */
344 startup_ipi_hook(phys_apicid, (unsigned long) start_secondary,
345 (unsigned long) init_rsp);
346
347
348 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 * Run STARTUP IPI loop.
350 */
351 Dprintk("#startup loops: %d.\n", num_starts);
352
Thomas Gleixner37e650c2008-01-30 13:30:14 +0100353 maxlvt = lapic_get_maxlvt();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354
355 for (j = 1; j <= num_starts; j++) {
356 Dprintk("Sending STARTUP #%d.\n",j);
Glauber de Oliveira Costa1af8a0c2008-03-19 14:24:58 -0300357 apic_read_around(APIC_SPIV);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 apic_write(APIC_ESR, 0);
359 apic_read(APIC_ESR);
360 Dprintk("After apic_write.\n");
361
362 /*
363 * STARTUP IPI
364 */
365
366 /* Target chip */
Glauber de Oliveira Costa1af8a0c2008-03-19 14:24:58 -0300367 apic_write_around(APIC_ICR2, SET_APIC_DEST_FIELD(phys_apicid));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368
369 /* Boot on the stack */
370 /* Kick the second */
Glauber de Oliveira Costa1af8a0c2008-03-19 14:24:58 -0300371 apic_write_around(APIC_ICR, APIC_DM_STARTUP | (start_rip>>12));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372
373 /*
374 * Give the other CPU some time to accept the IPI.
375 */
376 udelay(300);
377
378 Dprintk("Startup point 1.\n");
379
380 Dprintk("Waiting for send to finish...\n");
Fernando Luis VazquezCaoea8c7332007-05-02 19:27:17 +0200381 send_status = safe_apic_wait_icr_idle();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382
383 /*
384 * Give the other CPU some time to accept the IPI.
385 */
386 udelay(200);
387 /*
388 * Due to the Pentium erratum 3AP.
389 */
390 if (maxlvt > 3) {
Glauber de Oliveira Costa1af8a0c2008-03-19 14:24:58 -0300391 apic_read_around(APIC_SPIV);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 apic_write(APIC_ESR, 0);
393 }
394 accept_status = (apic_read(APIC_ESR) & 0xEF);
395 if (send_status || accept_status)
396 break;
397 }
398 Dprintk("After Startup.\n");
399
400 if (send_status)
401 printk(KERN_ERR "APIC never delivered???\n");
402 if (accept_status)
403 printk(KERN_ERR "APIC delivery error (%lx).\n", accept_status);
404
405 return (send_status | accept_status);
406}
407
Ashok Raj76e4f662005-06-25 14:55:00 -0700408struct create_idle {
David Howells65f27f32006-11-22 14:55:48 +0000409 struct work_struct work;
Ashok Raj76e4f662005-06-25 14:55:00 -0700410 struct task_struct *idle;
411 struct completion done;
412 int cpu;
413};
414
Thomas Gleixnera2b484a2008-01-09 00:18:28 +0100415static void __cpuinit do_fork_idle(struct work_struct *work)
Ashok Raj76e4f662005-06-25 14:55:00 -0700416{
David Howells65f27f32006-11-22 14:55:48 +0000417 struct create_idle *c_idle =
418 container_of(work, struct create_idle, work);
Ashok Raj76e4f662005-06-25 14:55:00 -0700419
420 c_idle->idle = fork_idle(c_idle->cpu);
421 complete(&c_idle->done);
422}
423
Andi Kleena8ab26f2005-04-16 15:25:19 -0700424/*
425 * Boot one CPU.
426 */
427static int __cpuinit do_boot_cpu(int cpu, int apicid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 unsigned long boot_error;
Andi Kleena8ab26f2005-04-16 15:25:19 -0700430 int timeout;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 unsigned long start_rip;
Ashok Raj76e4f662005-06-25 14:55:00 -0700432 struct create_idle c_idle = {
433 .cpu = cpu,
Ingo Molnarf86bf9b2006-07-10 04:44:05 -0700434 .done = COMPLETION_INITIALIZER_ONSTACK(c_idle.done),
Ashok Raj76e4f662005-06-25 14:55:00 -0700435 };
Glauber Costa2b775a22008-02-22 12:09:29 -0300436 INIT_WORK(&c_idle.work, do_fork_idle);
Ashok Raj76e4f662005-06-25 14:55:00 -0700437
Ravikiran G Thirumalaic11efdf2006-01-11 22:43:57 +0100438 /* allocate memory for gdts of secondary cpus. Hotplug is considered */
439 if (!cpu_gdt_descr[cpu].address &&
440 !(cpu_gdt_descr[cpu].address = get_zeroed_page(GFP_KERNEL))) {
441 printk(KERN_ERR "Failed to allocate GDT for CPU %d\n", cpu);
442 return -1;
443 }
444
Ravikiran G Thirumalai365ba912006-01-11 22:45:42 +0100445 /* Allocate node local memory for AP pdas */
446 if (cpu_pda(cpu) == &boot_cpu_pda[cpu]) {
447 struct x8664_pda *newpda, *pda;
448 int node = cpu_to_node(cpu);
449 pda = cpu_pda(cpu);
450 newpda = kmalloc_node(sizeof (struct x8664_pda), GFP_ATOMIC,
451 node);
452 if (newpda) {
453 memcpy(newpda, pda, sizeof (struct x8664_pda));
454 cpu_pda(cpu) = newpda;
455 } else
456 printk(KERN_ERR
457 "Could not allocate node local PDA for CPU %d on node %d\n",
458 cpu, node);
459 }
460
Gerd Hoffmannd167a512006-06-26 13:56:16 +0200461 alternatives_smp_switch(1);
462
Ashok Raj76e4f662005-06-25 14:55:00 -0700463 c_idle.idle = get_idle_for_cpu(cpu);
464
465 if (c_idle.idle) {
H. Peter Anvinfaca6222008-01-30 13:31:02 +0100466 c_idle.idle->thread.sp = (unsigned long) (((struct pt_regs *)
Al Viro57eafdc2006-01-12 01:05:39 -0800467 (THREAD_SIZE + task_stack_page(c_idle.idle))) - 1);
Ashok Raj76e4f662005-06-25 14:55:00 -0700468 init_idle(c_idle.idle, cpu);
469 goto do_rest;
Andi Kleena8ab26f2005-04-16 15:25:19 -0700470 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471
Ashok Raj76e4f662005-06-25 14:55:00 -0700472 /*
473 * During cold boot process, keventd thread is not spun up yet.
474 * When we do cpu hot-add, we create idle threads on the fly, we should
475 * not acquire any attributes from the calling context. Hence the clean
476 * way to create kernel_threads() is to do that from keventd().
477 * We do the current_is_keventd() due to the fact that ACPI notifier
478 * was also queuing to keventd() and when the caller is already running
479 * in context of keventd(), we would end up with locking up the keventd
480 * thread.
481 */
482 if (!keventd_up() || current_is_keventd())
David Howells65f27f32006-11-22 14:55:48 +0000483 c_idle.work.func(&c_idle.work);
Ashok Raj76e4f662005-06-25 14:55:00 -0700484 else {
David Howells65f27f32006-11-22 14:55:48 +0000485 schedule_work(&c_idle.work);
Ashok Raj76e4f662005-06-25 14:55:00 -0700486 wait_for_completion(&c_idle.done);
487 }
488
489 if (IS_ERR(c_idle.idle)) {
490 printk("failed fork for CPU %d\n", cpu);
491 return PTR_ERR(c_idle.idle);
492 }
493
494 set_idle_for_cpu(cpu, c_idle.idle);
495
496do_rest:
497
Ravikiran G Thirumalaidf79efd2006-01-11 22:45:39 +0100498 cpu_pda(cpu)->pcurrent = c_idle.idle;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499
500 start_rip = setup_trampoline();
501
H. Peter Anvinfaca6222008-01-30 13:31:02 +0100502 init_rsp = c_idle.idle->thread.sp;
Glauber de Oliveira Costa7818a1e2008-01-30 13:31:31 +0100503 load_sp0(&per_cpu(init_tss, cpu), &c_idle.idle->thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 initial_code = start_secondary;
Al Viroe4f17c42006-01-12 01:05:38 -0800505 clear_tsk_thread_flag(c_idle.idle, TIF_FORK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506
Andi Kleende04f322005-07-28 21:15:29 -0700507 printk(KERN_INFO "Booting processor %d/%d APIC 0x%x\n", cpu,
508 cpus_weight(cpu_present_map),
509 apicid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510
511 /*
512 * This grunge runs the startup process for
513 * the targeted processor.
514 */
515
516 atomic_set(&init_deasserted, 0);
517
518 Dprintk("Setting warm reset code and vector.\n");
519
520 CMOS_WRITE(0xa, 0xf);
521 local_flush_tlb();
522 Dprintk("1.\n");
523 *((volatile unsigned short *) phys_to_virt(0x469)) = start_rip >> 4;
524 Dprintk("2.\n");
525 *((volatile unsigned short *) phys_to_virt(0x467)) = start_rip & 0xf;
526 Dprintk("3.\n");
527
528 /*
529 * Be paranoid about clearing APIC errors.
530 */
Andi Kleen11a8e772006-01-11 22:46:51 +0100531 apic_write(APIC_ESR, 0);
532 apic_read(APIC_ESR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533
534 /*
535 * Status is now clean
536 */
537 boot_error = 0;
538
539 /*
540 * Starting actual IPI sequence...
541 */
Andi Kleena8ab26f2005-04-16 15:25:19 -0700542 boot_error = wakeup_secondary_via_INIT(apicid, start_rip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543
544 if (!boot_error) {
545 /*
546 * allow APs to start initializing.
547 */
548 Dprintk("Before Callout %d.\n", cpu);
549 cpu_set(cpu, cpu_callout_map);
550 Dprintk("After Callout %d.\n", cpu);
551
552 /*
553 * Wait 5s total for a response
554 */
555 for (timeout = 0; timeout < 50000; timeout++) {
556 if (cpu_isset(cpu, cpu_callin_map))
557 break; /* It has booted */
558 udelay(100);
559 }
560
561 if (cpu_isset(cpu, cpu_callin_map)) {
562 /* number CPUs logically, starting from 1 (BSP is 0) */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 Dprintk("CPU has booted.\n");
Glauber de Oliveira Costa4f3ab192008-03-19 14:25:01 -0300564 printk(KERN_INFO "CPU%d: ", cpu);
565 print_cpu_info(&cpu_data(cpu));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 } else {
567 boot_error = 1;
568 if (*((volatile unsigned char *)phys_to_virt(SMP_TRAMPOLINE_BASE))
569 == 0xA5)
570 /* trampoline started but...? */
571 printk("Stuck ??\n");
572 else
573 /* trampoline code not run */
574 printk("Not responding.\n");
Olaf Hering44456d32005-07-27 11:45:17 -0700575#ifdef APIC_DEBUG
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 inquire_remote_apic(apicid);
577#endif
578 }
579 }
580 if (boot_error) {
581 cpu_clear(cpu, cpu_callout_map); /* was set here (do_boot_cpu()) */
Jeremy Fitzhardinge5548fec2008-01-30 13:30:55 +0100582 clear_bit(cpu, (unsigned long *)&cpu_initialized); /* was set by cpu_init() */
Ravikiran G Thirumalai488fc082006-02-07 12:58:23 -0800583 clear_node_cpumask(cpu); /* was set by numa_add_cpu */
Andi Kleena8ab26f2005-04-16 15:25:19 -0700584 cpu_clear(cpu, cpu_present_map);
585 cpu_clear(cpu, cpu_possible_map);
Mike Travis71fff5e2007-10-19 20:35:03 +0200586 per_cpu(x86_cpu_to_apicid, cpu) = BAD_APICID;
Andi Kleena8ab26f2005-04-16 15:25:19 -0700587 return -EIO;
588 }
589
590 return 0;
591}
592
593cycles_t cacheflush_time;
594unsigned long cache_decay_ticks;
595
596/*
Andi Kleena8ab26f2005-04-16 15:25:19 -0700597 * Cleanup possible dangling ends...
598 */
599static __cpuinit void smp_cleanup_boot(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 /*
Andi Kleena8ab26f2005-04-16 15:25:19 -0700602 * Paranoid: Set warm reset code and vector here back
603 * to default values.
604 */
605 CMOS_WRITE(0, 0xf);
606
607 /*
608 * Reset trampoline flag
609 */
610 *((volatile int *) phys_to_virt(0x467)) = 0;
Andi Kleena8ab26f2005-04-16 15:25:19 -0700611}
612
613/*
614 * Fall back to non SMP mode after errors.
615 *
616 * RED-PEN audit/test this more. I bet there is more state messed up here.
617 */
Ashok Raje6982c62005-06-25 14:54:58 -0700618static __init void disable_smp(void)
Andi Kleena8ab26f2005-04-16 15:25:19 -0700619{
620 cpu_present_map = cpumask_of_cpu(0);
621 cpu_possible_map = cpumask_of_cpu(0);
622 if (smp_found_config)
623 phys_cpu_present_map = physid_mask_of_physid(boot_cpu_id);
624 else
625 phys_cpu_present_map = physid_mask_of_physid(0);
Mike Travisd5a74302007-10-16 01:24:05 -0700626 cpu_set(0, per_cpu(cpu_sibling_map, 0));
Mike Travis08357612007-10-16 01:24:04 -0700627 cpu_set(0, per_cpu(cpu_core_map, 0));
Andi Kleena8ab26f2005-04-16 15:25:19 -0700628}
629
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630/*
Andi Kleena8ab26f2005-04-16 15:25:19 -0700631 * Various sanity checks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 */
Ashok Raje6982c62005-06-25 14:54:58 -0700633static int __init smp_sanity_check(unsigned max_cpus)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 if (!physid_isset(hard_smp_processor_id(), phys_cpu_present_map)) {
636 printk("weird, boot CPU (#%d) not listed by the BIOS.\n",
637 hard_smp_processor_id());
638 physid_set(hard_smp_processor_id(), phys_cpu_present_map);
639 }
640
641 /*
642 * If we couldn't find an SMP configuration at boot time,
643 * get out of here now!
644 */
645 if (!smp_found_config) {
646 printk(KERN_NOTICE "SMP motherboard not detected.\n");
Andi Kleena8ab26f2005-04-16 15:25:19 -0700647 disable_smp();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 if (APIC_init_uniprocessor())
649 printk(KERN_NOTICE "Local APIC not detected."
650 " Using dummy APIC emulation.\n");
Andi Kleena8ab26f2005-04-16 15:25:19 -0700651 return -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 }
653
654 /*
655 * Should not be necessary because the MP table should list the boot
656 * CPU too, but we do it for the sake of robustness anyway.
657 */
658 if (!physid_isset(boot_cpu_id, phys_cpu_present_map)) {
659 printk(KERN_NOTICE "weird, boot CPU (#%d) not listed by the BIOS.\n",
660 boot_cpu_id);
661 physid_set(hard_smp_processor_id(), phys_cpu_present_map);
662 }
663
664 /*
665 * If we couldn't find a local APIC, then get out of here now!
666 */
Andi Kleen11a8e772006-01-11 22:46:51 +0100667 if (!cpu_has_apic) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 printk(KERN_ERR "BIOS bug, local APIC #%d not detected!...\n",
669 boot_cpu_id);
670 printk(KERN_ERR "... forcing use of dummy APIC emulation. (tell your hw vendor)\n");
Andi Kleena8ab26f2005-04-16 15:25:19 -0700671 nr_ioapics = 0;
672 return -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 }
674
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 /*
676 * If SMP should be disabled, then really disable it!
677 */
678 if (!max_cpus) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 printk(KERN_INFO "SMP mode deactivated, forcing use of dummy APIC emulation.\n");
Andi Kleena8ab26f2005-04-16 15:25:19 -0700680 nr_ioapics = 0;
681 return -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 }
683
Andi Kleena8ab26f2005-04-16 15:25:19 -0700684 return 0;
685}
686
Yinghai Lu949ec322008-01-30 13:30:46 +0100687static void __init smp_cpu_index_default(void)
688{
689 int i;
690 struct cpuinfo_x86 *c;
691
692 for_each_cpu_mask(i, cpu_possible_map) {
693 c = &cpu_data(i);
694 /* mark all to hotplug */
695 c->cpu_index = NR_CPUS;
696 }
697}
698
Mike Travis71fff5e2007-10-19 20:35:03 +0200699/*
Andi Kleena8ab26f2005-04-16 15:25:19 -0700700 * Prepare for SMP bootup. The MP table or ACPI has been read
701 * earlier. Just do some sanity checking here and enable APIC mode.
702 */
Glauber Costa7557da62008-03-03 14:12:38 -0300703void __init native_smp_prepare_cpus(unsigned int max_cpus)
Andi Kleena8ab26f2005-04-16 15:25:19 -0700704{
Andi Kleena8ab26f2005-04-16 15:25:19 -0700705 nmi_watchdog_default();
Yinghai Lu949ec322008-01-30 13:30:46 +0100706 smp_cpu_index_default();
Andi Kleena8ab26f2005-04-16 15:25:19 -0700707 current_cpu_data = boot_cpu_data;
708 current_thread_info()->cpu = 0; /* needed? */
Siddha, Suresh B94605ef2005-11-05 17:25:54 +0100709 set_cpu_sibling_map(0);
Andi Kleena8ab26f2005-04-16 15:25:19 -0700710
Andi Kleena8ab26f2005-04-16 15:25:19 -0700711 if (smp_sanity_check(max_cpus) < 0) {
712 printk(KERN_INFO "SMP disabled\n");
713 disable_smp();
714 return;
715 }
716
717
718 /*
719 * Switch from PIC to APIC mode.
720 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 setup_local_APIC();
722
Andi Kleen739f33b2008-01-30 13:30:40 +0100723 /*
724 * Enable IO APIC before setting up error vector
725 */
726 if (!skip_ioapic_setup && nr_ioapics)
727 enable_IO_APIC();
728 end_local_APIC_setup();
729
Andi Kleena8ab26f2005-04-16 15:25:19 -0700730 if (GET_APIC_ID(apic_read(APIC_ID)) != boot_cpu_id) {
731 panic("Boot APIC ID in local APIC unexpected (%d vs %d)",
732 GET_APIC_ID(apic_read(APIC_ID)), boot_cpu_id);
733 /* Or can we switch back to PIC here? */
734 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735
736 /*
Andi Kleena8ab26f2005-04-16 15:25:19 -0700737 * Now start the IO-APICs
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 */
739 if (!skip_ioapic_setup && nr_ioapics)
740 setup_IO_APIC();
741 else
742 nr_ioapics = 0;
743
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 /*
Andi Kleena8ab26f2005-04-16 15:25:19 -0700745 * Set up local APIC timer on boot CPU.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747
Glauber de Oliveira Costa746ef0c2008-01-30 13:31:11 +0100748 setup_boot_clock();
Glauber de Oliveira Costa4f3ab192008-03-19 14:25:01 -0300749 printk(KERN_INFO "CPU%d: ", 0);
750 print_cpu_info(&cpu_data(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751}
752
Andi Kleena8ab26f2005-04-16 15:25:19 -0700753/*
754 * Early setup to make printk work.
755 */
Glauber Costa1e3fac82008-03-03 14:12:37 -0300756void __init native_smp_prepare_boot_cpu(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757{
Andi Kleena8ab26f2005-04-16 15:25:19 -0700758 int me = smp_processor_id();
Yinghai Lu23916d42008-01-30 13:33:39 +0100759 /* already set me in cpu_online_map in boot_cpu_init() */
Andi Kleena8ab26f2005-04-16 15:25:19 -0700760 cpu_set(me, cpu_callout_map);
Ashok Raj884d9e42005-06-25 14:55:02 -0700761 per_cpu(cpu_state, me) = CPU_ONLINE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762}
763
Andi Kleena8ab26f2005-04-16 15:25:19 -0700764/*
765 * Entry point to boot a CPU.
Andi Kleena8ab26f2005-04-16 15:25:19 -0700766 */
Glauber Costa71d19542008-03-03 14:12:36 -0300767int __cpuinit native_cpu_up(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768{
Andi Kleena8ab26f2005-04-16 15:25:19 -0700769 int apicid = cpu_present_to_apicid(cpu);
Ingo Molnard04f41e2007-03-07 18:12:31 +0100770 unsigned long flags;
771 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772
Andi Kleena8ab26f2005-04-16 15:25:19 -0700773 WARN_ON(irqs_disabled());
774
775 Dprintk("++++++++++++++++++++=_---CPU UP %u\n", cpu);
776
777 if (apicid == BAD_APICID || apicid == boot_cpu_id ||
778 !physid_isset(apicid, phys_cpu_present_map)) {
779 printk("__cpu_up: bad cpu %d\n", cpu);
780 return -EINVAL;
781 }
Andi Kleena8ab26f2005-04-16 15:25:19 -0700782
Ashok Raj76e4f662005-06-25 14:55:00 -0700783 /*
784 * Already booted CPU?
785 */
786 if (cpu_isset(cpu, cpu_callin_map)) {
787 Dprintk("do_boot_cpu %d Already started\n", cpu);
788 return -ENOSYS;
789 }
790
Bernhard Kaindl2b1f6272007-05-02 19:27:17 +0200791 /*
792 * Save current MTRR state in case it was changed since early boot
793 * (e.g. by the ACPI SMI) to initialize new CPUs with MTRRs in sync:
794 */
795 mtrr_save_state();
796
Ashok Raj884d9e42005-06-25 14:55:02 -0700797 per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
Andi Kleena8ab26f2005-04-16 15:25:19 -0700798 /* Boot it! */
799 err = do_boot_cpu(cpu, apicid);
800 if (err < 0) {
Andi Kleena8ab26f2005-04-16 15:25:19 -0700801 Dprintk("do_boot_cpu failed %d\n", err);
802 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 }
804
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 /* Unleash the CPU! */
806 Dprintk("waiting for cpu %d\n", cpu);
807
Ingo Molnar95492e42007-02-16 01:27:34 -0800808 /*
809 * Make sure and check TSC sync:
810 */
Ingo Molnard04f41e2007-03-07 18:12:31 +0100811 local_irq_save(flags);
Ingo Molnar95492e42007-02-16 01:27:34 -0800812 check_tsc_sync_source(cpu);
Ingo Molnard04f41e2007-03-07 18:12:31 +0100813 local_irq_restore(flags);
Ingo Molnar95492e42007-02-16 01:27:34 -0800814
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 while (!cpu_isset(cpu, cpu_online_map))
Andi Kleena8ab26f2005-04-16 15:25:19 -0700816 cpu_relax();
Ashok Raj76e4f662005-06-25 14:55:00 -0700817 err = 0;
818
819 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820}
821
Glauber de Oliveira Costaf68e00a2008-03-19 14:25:29 -0300822extern void impress_friends(void);
823extern void smp_checks(void);
824
Andi Kleena8ab26f2005-04-16 15:25:19 -0700825/*
826 * Finish the SMP boot.
827 */
Glauber Costac5597642008-03-03 14:12:39 -0300828void __init native_smp_cpus_done(unsigned int max_cpus)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829{
Andi Kleena8ab26f2005-04-16 15:25:19 -0700830 smp_cleanup_boot();
Glauber de Oliveira Costaf68e00a2008-03-19 14:25:29 -0300831
832 Dprintk("Boot done.\n");
833
834 impress_friends();
835 smp_checks();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 setup_ioapic_dest();
Andi Kleen75152112005-05-16 21:53:34 -0700837 check_nmi_watchdog();
Andi Kleena8ab26f2005-04-16 15:25:19 -0700838}