smpboot_32.c 22 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919
  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. *
  7. * Much of the core SMP work is based on previous work by Thomas Radke, to
  8. * whom a great many thanks are extended.
  9. *
  10. * Thanks to Intel for making available several different Pentium,
  11. * Pentium Pro and Pentium-II/Xeon MP machines.
  12. * Original development of Linux SMP code supported by Caldera.
  13. *
  14. * This code is released under the GNU General Public License version 2 or
  15. * later.
  16. *
  17. * Fixes
  18. * Felix Koop : NR_CPUS used properly
  19. * Jose Renau : Handle single CPU case.
  20. * Alan Cox : By repeated request 8) - Total BogoMIPS 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. * Martin J. Bligh : Added support for multi-quad systems
  33. * Dave Jones : Report invalid combinations of Athlon CPUs.
  34. * Rusty Russell : Hacked into shape for new "hotplug" boot process. */
  35. #include <linux/module.h>
  36. #include <linux/init.h>
  37. #include <linux/kernel.h>
  38. #include <linux/mm.h>
  39. #include <linux/sched.h>
  40. #include <linux/kernel_stat.h>
  41. #include <linux/bootmem.h>
  42. #include <linux/notifier.h>
  43. #include <linux/cpu.h>
  44. #include <linux/percpu.h>
  45. #include <linux/nmi.h>
  46. #include <linux/delay.h>
  47. #include <linux/mc146818rtc.h>
  48. #include <asm/tlbflush.h>
  49. #include <asm/desc.h>
  50. #include <asm/arch_hooks.h>
  51. #include <asm/nmi.h>
  52. #include <mach_apic.h>
  53. #include <mach_wakecpu.h>
  54. #include <smpboot_hooks.h>
  55. #include <asm/vmi.h>
  56. #include <asm/mtrr.h>
  57. /* which logical CPU number maps to which CPU (physical APIC ID) */
  58. u16 x86_cpu_to_apicid_init[NR_CPUS] __initdata =
  59. { [0 ... NR_CPUS-1] = BAD_APICID };
  60. void *x86_cpu_to_apicid_early_ptr;
  61. DEFINE_PER_CPU(u16, x86_cpu_to_apicid) = BAD_APICID;
  62. EXPORT_PER_CPU_SYMBOL(x86_cpu_to_apicid);
  63. u16 x86_bios_cpu_apicid_init[NR_CPUS] __initdata
  64. = { [0 ... NR_CPUS-1] = BAD_APICID };
  65. void *x86_bios_cpu_apicid_early_ptr;
  66. DEFINE_PER_CPU(u16, x86_bios_cpu_apicid) = BAD_APICID;
  67. EXPORT_PER_CPU_SYMBOL(x86_bios_cpu_apicid);
  68. u8 apicid_2_node[MAX_APICID];
  69. static void map_cpu_to_logical_apicid(void);
  70. /* State of each CPU. */
  71. DEFINE_PER_CPU(int, cpu_state) = { 0 };
  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. */
  76. #ifdef CONFIG_HOTPLUG_CPU
  77. /*
  78. * Needed only for CONFIG_HOTPLUG_CPU because __cpuinitdata is
  79. * removed after init for !CONFIG_HOTPLUG_CPU.
  80. */
  81. static 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
  85. struct task_struct *idle_thread_array[NR_CPUS] __cpuinitdata ;
  86. #define get_idle_for_cpu(x) (idle_thread_array[(x)])
  87. #define set_idle_for_cpu(x, p) (idle_thread_array[(x)] = (p))
  88. #endif
  89. static atomic_t init_deasserted;
  90. static void __cpuinit smp_callin(void)
  91. {
  92. int cpuid, phys_id;
  93. unsigned long timeout;
  94. /*
  95. * If waken up by an INIT in an 82489DX configuration
  96. * we may get here before an INIT-deassert IPI reaches
  97. * our local APIC. We have to wait for the IPI or we'll
  98. * lock up on an APIC access.
  99. */
  100. wait_for_init_deassert(&init_deasserted);
  101. /*
  102. * (This works even if the APIC is not enabled.)
  103. */
  104. phys_id = GET_APIC_ID(apic_read(APIC_ID));
  105. cpuid = smp_processor_id();
  106. if (cpu_isset(cpuid, cpu_callin_map)) {
  107. printk("huh, phys CPU#%d, CPU#%d already present??\n",
  108. phys_id, cpuid);
  109. BUG();
  110. }
  111. Dprintk("CPU#%d (phys ID: %d) waiting for CALLOUT\n", cpuid, phys_id);
  112. /*
  113. * STARTUP IPIs are fragile beasts as they might sometimes
  114. * trigger some glue motherboard logic. Complete APIC bus
  115. * silence for 1 second, this overestimates the time the
  116. * boot CPU is spending to send the up to 2 STARTUP IPIs
  117. * by a factor of two. This should be enough.
  118. */
  119. /*
  120. * Waiting 2s total for startup (udelay is not yet working)
  121. */
  122. timeout = jiffies + 2*HZ;
  123. while (time_before(jiffies, timeout)) {
  124. /*
  125. * Has the boot CPU finished it's STARTUP sequence?
  126. */
  127. if (cpu_isset(cpuid, cpu_callout_map))
  128. break;
  129. cpu_relax();
  130. }
  131. if (!time_before(jiffies, timeout)) {
  132. printk("BUG: CPU%d started up but did not get a callout!\n",
  133. cpuid);
  134. BUG();
  135. }
  136. /*
  137. * the boot CPU has finished the init stage and is spinning
  138. * on callin_map until we finish. We are free to set up this
  139. * CPU, first the APIC. (this is probably redundant on most
  140. * boards)
  141. */
  142. Dprintk("CALLIN, before setup_local_APIC().\n");
  143. smp_callin_clear_local_apic();
  144. setup_local_APIC();
  145. end_local_APIC_setup();
  146. map_cpu_to_logical_apicid();
  147. /*
  148. * Get our bogomips.
  149. */
  150. local_irq_enable();
  151. calibrate_delay();
  152. local_irq_disable();
  153. Dprintk("Stack at about %p\n",&cpuid);
  154. /*
  155. * Save our processor parameters
  156. */
  157. smp_store_cpu_info(cpuid);
  158. /*
  159. * Allow the master to continue.
  160. */
  161. cpu_set(cpuid, cpu_callin_map);
  162. }
  163. /*
  164. * Activate a secondary processor.
  165. */
  166. static void __cpuinit start_secondary(void *unused)
  167. {
  168. /*
  169. * Don't put *anything* before cpu_init(), SMP booting is too
  170. * fragile that we want to limit the things done here to the
  171. * most necessary things.
  172. */
  173. #ifdef CONFIG_VMI
  174. vmi_bringup();
  175. #endif
  176. cpu_init();
  177. preempt_disable();
  178. smp_callin();
  179. /* otherwise gcc will move up smp_processor_id before the cpu_init */
  180. barrier();
  181. /*
  182. * Check TSC synchronization with the BP:
  183. */
  184. check_tsc_sync_target();
  185. if (nmi_watchdog == NMI_IO_APIC) {
  186. disable_8259A_irq(0);
  187. enable_NMI_through_LVT0();
  188. enable_8259A_irq(0);
  189. }
  190. /* This must be done before setting cpu_online_map */
  191. set_cpu_sibling_map(raw_smp_processor_id());
  192. wmb();
  193. /*
  194. * We need to hold call_lock, so there is no inconsistency
  195. * between the time smp_call_function() determines number of
  196. * IPI recipients, and the time when the determination is made
  197. * for which cpus receive the IPI. Holding this
  198. * lock helps us to not include this cpu in a currently in progress
  199. * smp_call_function().
  200. */
  201. lock_ipi_call_lock();
  202. cpu_set(smp_processor_id(), cpu_online_map);
  203. unlock_ipi_call_lock();
  204. per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE;
  205. setup_secondary_clock();
  206. wmb();
  207. cpu_idle();
  208. }
  209. /*
  210. * Everything has been set up for the secondary
  211. * CPUs - they just need to reload everything
  212. * from the task structure
  213. * This function must not return.
  214. */
  215. void __devinit initialize_secondary(void)
  216. {
  217. /*
  218. * We don't actually need to load the full TSS,
  219. * basically just the stack pointer and the ip.
  220. */
  221. asm volatile(
  222. "movl %0,%%esp\n\t"
  223. "jmp *%1"
  224. :
  225. :"m" (current->thread.sp),"m" (current->thread.ip));
  226. }
  227. /* Static state in head.S used to set up a CPU */
  228. extern struct {
  229. void * sp;
  230. unsigned short ss;
  231. } stack_start;
  232. #ifdef CONFIG_NUMA
  233. /* which logical CPUs are on which nodes */
  234. cpumask_t node_to_cpumask_map[MAX_NUMNODES] __read_mostly =
  235. { [0 ... MAX_NUMNODES-1] = CPU_MASK_NONE };
  236. EXPORT_SYMBOL(node_to_cpumask_map);
  237. /* which node each logical CPU is on */
  238. int cpu_to_node_map[NR_CPUS] __read_mostly = { [0 ... NR_CPUS-1] = 0 };
  239. EXPORT_SYMBOL(cpu_to_node_map);
  240. /* set up a mapping between cpu and node. */
  241. static inline void map_cpu_to_node(int cpu, int node)
  242. {
  243. printk("Mapping cpu %d to node %d\n", cpu, node);
  244. cpu_set(cpu, node_to_cpumask_map[node]);
  245. cpu_to_node_map[cpu] = node;
  246. }
  247. /* undo a mapping between cpu and node. */
  248. static inline void unmap_cpu_to_node(int cpu)
  249. {
  250. int node;
  251. printk("Unmapping cpu %d from all nodes\n", cpu);
  252. for (node = 0; node < MAX_NUMNODES; node ++)
  253. cpu_clear(cpu, node_to_cpumask_map[node]);
  254. cpu_to_node_map[cpu] = 0;
  255. }
  256. #else /* !CONFIG_NUMA */
  257. #define map_cpu_to_node(cpu, node) ({})
  258. #define unmap_cpu_to_node(cpu) ({})
  259. #endif /* CONFIG_NUMA */
  260. u8 cpu_2_logical_apicid[NR_CPUS] __read_mostly = { [0 ... NR_CPUS-1] = BAD_APICID };
  261. static void map_cpu_to_logical_apicid(void)
  262. {
  263. int cpu = smp_processor_id();
  264. int apicid = logical_smp_processor_id();
  265. int node = apicid_to_node(apicid);
  266. if (!node_online(node))
  267. node = first_online_node;
  268. cpu_2_logical_apicid[cpu] = apicid;
  269. map_cpu_to_node(cpu, node);
  270. }
  271. static void unmap_cpu_to_logical_apicid(int cpu)
  272. {
  273. cpu_2_logical_apicid[cpu] = BAD_APICID;
  274. unmap_cpu_to_node(cpu);
  275. }
  276. static inline void __inquire_remote_apic(int apicid)
  277. {
  278. unsigned i, regs[] = { APIC_ID >> 4, APIC_LVR >> 4, APIC_SPIV >> 4 };
  279. char *names[] = { "ID", "VERSION", "SPIV" };
  280. int timeout;
  281. u32 status;
  282. printk(KERN_INFO "Inquiring remote APIC #%d...\n", apicid);
  283. for (i = 0; i < ARRAY_SIZE(regs); i++) {
  284. printk(KERN_INFO "... APIC #%d %s: ", apicid, names[i]);
  285. /*
  286. * Wait for idle.
  287. */
  288. status = safe_apic_wait_icr_idle();
  289. if (status)
  290. printk(KERN_CONT
  291. "a previous APIC delivery may have failed\n");
  292. apic_write_around(APIC_ICR2, SET_APIC_DEST_FIELD(apicid));
  293. apic_write_around(APIC_ICR, APIC_DM_REMRD | regs[i]);
  294. timeout = 0;
  295. do {
  296. udelay(100);
  297. status = apic_read(APIC_ICR) & APIC_ICR_RR_MASK;
  298. } while (status == APIC_ICR_RR_INPROG && timeout++ < 1000);
  299. switch (status) {
  300. case APIC_ICR_RR_VALID:
  301. status = apic_read(APIC_RRR);
  302. printk(KERN_CONT "%08x\n", status);
  303. break;
  304. default:
  305. printk(KERN_CONT "failed\n");
  306. }
  307. }
  308. }
  309. #ifdef WAKE_SECONDARY_VIA_NMI
  310. /*
  311. * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal
  312. * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this
  313. * won't ... remember to clear down the APIC, etc later.
  314. */
  315. static int __devinit
  316. wakeup_secondary_cpu(int logical_apicid, unsigned long start_eip)
  317. {
  318. unsigned long send_status, accept_status = 0;
  319. int maxlvt;
  320. /* Target chip */
  321. apic_write_around(APIC_ICR2, SET_APIC_DEST_FIELD(logical_apicid));
  322. /* Boot on the stack */
  323. /* Kick the second */
  324. apic_write_around(APIC_ICR, APIC_DM_NMI | APIC_DEST_LOGICAL);
  325. Dprintk("Waiting for send to finish...\n");
  326. send_status = safe_apic_wait_icr_idle();
  327. /*
  328. * Give the other CPU some time to accept the IPI.
  329. */
  330. udelay(200);
  331. /*
  332. * Due to the Pentium erratum 3AP.
  333. */
  334. maxlvt = lapic_get_maxlvt();
  335. if (maxlvt > 3) {
  336. apic_read_around(APIC_SPIV);
  337. apic_write(APIC_ESR, 0);
  338. }
  339. accept_status = (apic_read(APIC_ESR) & 0xEF);
  340. Dprintk("NMI sent.\n");
  341. if (send_status)
  342. printk("APIC never delivered???\n");
  343. if (accept_status)
  344. printk("APIC delivery error (%lx).\n", accept_status);
  345. return (send_status | accept_status);
  346. }
  347. #endif /* WAKE_SECONDARY_VIA_NMI */
  348. #ifdef WAKE_SECONDARY_VIA_INIT
  349. static int __devinit
  350. wakeup_secondary_cpu(int phys_apicid, unsigned long start_eip)
  351. {
  352. unsigned long send_status, accept_status = 0;
  353. int maxlvt, num_starts, j;
  354. /*
  355. * Be paranoid about clearing APIC errors.
  356. */
  357. if (APIC_INTEGRATED(apic_version[phys_apicid])) {
  358. apic_read_around(APIC_SPIV);
  359. apic_write(APIC_ESR, 0);
  360. apic_read(APIC_ESR);
  361. }
  362. Dprintk("Asserting INIT.\n");
  363. /*
  364. * Turn INIT on target chip
  365. */
  366. apic_write_around(APIC_ICR2, SET_APIC_DEST_FIELD(phys_apicid));
  367. /*
  368. * Send IPI
  369. */
  370. apic_write_around(APIC_ICR, APIC_INT_LEVELTRIG | APIC_INT_ASSERT
  371. | APIC_DM_INIT);
  372. Dprintk("Waiting for send to finish...\n");
  373. send_status = safe_apic_wait_icr_idle();
  374. mdelay(10);
  375. Dprintk("Deasserting INIT.\n");
  376. /* Target chip */
  377. apic_write_around(APIC_ICR2, SET_APIC_DEST_FIELD(phys_apicid));
  378. /* Send IPI */
  379. apic_write_around(APIC_ICR, APIC_INT_LEVELTRIG | APIC_DM_INIT);
  380. Dprintk("Waiting for send to finish...\n");
  381. send_status = safe_apic_wait_icr_idle();
  382. mb();
  383. atomic_set(&init_deasserted, 1);
  384. /*
  385. * Should we send STARTUP IPIs ?
  386. *
  387. * Determine this based on the APIC version.
  388. * If we don't have an integrated APIC, don't send the STARTUP IPIs.
  389. */
  390. if (APIC_INTEGRATED(apic_version[phys_apicid]))
  391. num_starts = 2;
  392. else
  393. num_starts = 0;
  394. /*
  395. * Paravirt / VMI wants a startup IPI hook here to set up the
  396. * target processor state.
  397. */
  398. startup_ipi_hook(phys_apicid, (unsigned long) start_secondary,
  399. (unsigned long) stack_start.sp);
  400. /*
  401. * Run STARTUP IPI loop.
  402. */
  403. Dprintk("#startup loops: %d.\n", num_starts);
  404. maxlvt = lapic_get_maxlvt();
  405. for (j = 1; j <= num_starts; j++) {
  406. Dprintk("Sending STARTUP #%d.\n",j);
  407. apic_read_around(APIC_SPIV);
  408. apic_write(APIC_ESR, 0);
  409. apic_read(APIC_ESR);
  410. Dprintk("After apic_write.\n");
  411. /*
  412. * STARTUP IPI
  413. */
  414. /* Target chip */
  415. apic_write_around(APIC_ICR2, SET_APIC_DEST_FIELD(phys_apicid));
  416. /* Boot on the stack */
  417. /* Kick the second */
  418. apic_write_around(APIC_ICR, APIC_DM_STARTUP
  419. | (start_eip >> 12));
  420. /*
  421. * Give the other CPU some time to accept the IPI.
  422. */
  423. udelay(300);
  424. Dprintk("Startup point 1.\n");
  425. Dprintk("Waiting for send to finish...\n");
  426. send_status = safe_apic_wait_icr_idle();
  427. /*
  428. * Give the other CPU some time to accept the IPI.
  429. */
  430. udelay(200);
  431. /*
  432. * Due to the Pentium erratum 3AP.
  433. */
  434. if (maxlvt > 3) {
  435. apic_read_around(APIC_SPIV);
  436. apic_write(APIC_ESR, 0);
  437. }
  438. accept_status = (apic_read(APIC_ESR) & 0xEF);
  439. if (send_status || accept_status)
  440. break;
  441. }
  442. Dprintk("After Startup.\n");
  443. if (send_status)
  444. printk("APIC never delivered???\n");
  445. if (accept_status)
  446. printk("APIC delivery error (%lx).\n", accept_status);
  447. return (send_status | accept_status);
  448. }
  449. #endif /* WAKE_SECONDARY_VIA_INIT */
  450. extern cpumask_t cpu_initialized;
  451. struct create_idle {
  452. struct work_struct work;
  453. struct task_struct *idle;
  454. struct completion done;
  455. int cpu;
  456. };
  457. static void __cpuinit do_fork_idle(struct work_struct *work)
  458. {
  459. struct create_idle *c_idle =
  460. container_of(work, struct create_idle, work);
  461. c_idle->idle = fork_idle(c_idle->cpu);
  462. complete(&c_idle->done);
  463. }
  464. static int __cpuinit do_boot_cpu(int apicid, int cpu)
  465. /*
  466. * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad
  467. * (ie clustered apic addressing mode), this is a LOGICAL apic ID.
  468. * Returns zero if CPU booted OK, else error code from wakeup_secondary_cpu.
  469. */
  470. {
  471. unsigned long boot_error = 0;
  472. int timeout;
  473. unsigned long start_eip;
  474. unsigned short nmi_high = 0, nmi_low = 0;
  475. struct create_idle c_idle = {
  476. .cpu = cpu,
  477. .done = COMPLETION_INITIALIZER_ONSTACK(c_idle.done),
  478. };
  479. INIT_WORK(&c_idle.work, do_fork_idle);
  480. alternatives_smp_switch(1);
  481. c_idle.idle = get_idle_for_cpu(cpu);
  482. /*
  483. * We can't use kernel_thread since we must avoid to
  484. * reschedule the child.
  485. */
  486. if (c_idle.idle) {
  487. c_idle.idle->thread.sp = (unsigned long) (((struct pt_regs *)
  488. (THREAD_SIZE + task_stack_page(c_idle.idle))) - 1);
  489. init_idle(c_idle.idle, cpu);
  490. goto do_rest;
  491. }
  492. if (!keventd_up() || current_is_keventd())
  493. c_idle.work.func(&c_idle.work);
  494. else {
  495. schedule_work(&c_idle.work);
  496. wait_for_completion(&c_idle.done);
  497. }
  498. if (IS_ERR(c_idle.idle)) {
  499. printk(KERN_ERR "failed fork for CPU %d\n", cpu);
  500. return PTR_ERR(c_idle.idle);
  501. }
  502. set_idle_for_cpu(cpu, c_idle.idle);
  503. do_rest:
  504. per_cpu(current_task, cpu) = c_idle.idle;
  505. init_gdt(cpu);
  506. early_gdt_descr.address = (unsigned long)get_cpu_gdt_table(cpu);
  507. c_idle.idle->thread.ip = (unsigned long) start_secondary;
  508. /* start_eip had better be page-aligned! */
  509. start_eip = setup_trampoline();
  510. /* So we see what's up */
  511. printk("Booting processor %d/%d ip %lx\n", cpu, apicid, start_eip);
  512. /* Stack for startup_32 can be just as for start_secondary onwards */
  513. stack_start.sp = (void *) c_idle.idle->thread.sp;
  514. irq_ctx_init(cpu);
  515. /*
  516. * This grunge runs the startup process for
  517. * the targeted processor.
  518. */
  519. atomic_set(&init_deasserted, 0);
  520. Dprintk("Setting warm reset code and vector.\n");
  521. store_NMI_vector(&nmi_high, &nmi_low);
  522. smpboot_setup_warm_reset_vector(start_eip);
  523. /*
  524. * Be paranoid about clearing APIC errors.
  525. */
  526. apic_write(APIC_ESR, 0);
  527. apic_read(APIC_ESR);
  528. /*
  529. * Starting actual IPI sequence...
  530. */
  531. boot_error = wakeup_secondary_cpu(apicid, start_eip);
  532. if (!boot_error) {
  533. /*
  534. * allow APs to start initializing.
  535. */
  536. Dprintk("Before Callout %d.\n", cpu);
  537. cpu_set(cpu, cpu_callout_map);
  538. Dprintk("After Callout %d.\n", cpu);
  539. /*
  540. * Wait 5s total for a response
  541. */
  542. for (timeout = 0; timeout < 50000; timeout++) {
  543. if (cpu_isset(cpu, cpu_callin_map))
  544. break; /* It has booted */
  545. udelay(100);
  546. }
  547. if (cpu_isset(cpu, cpu_callin_map)) {
  548. /* number CPUs logically, starting from 1 (BSP is 0) */
  549. Dprintk("OK.\n");
  550. printk("CPU%d: ", cpu);
  551. print_cpu_info(&cpu_data(cpu));
  552. Dprintk("CPU has booted.\n");
  553. } else {
  554. boot_error= 1;
  555. if (*((volatile unsigned char *)trampoline_base)
  556. == 0xA5)
  557. /* trampoline started but...? */
  558. printk("Stuck ??\n");
  559. else
  560. /* trampoline code not run */
  561. printk("Not responding.\n");
  562. inquire_remote_apic(apicid);
  563. }
  564. }
  565. if (boot_error) {
  566. /* Try to put things back the way they were before ... */
  567. unmap_cpu_to_logical_apicid(cpu);
  568. cpu_clear(cpu, cpu_callout_map); /* was set here (do_boot_cpu()) */
  569. cpu_clear(cpu, cpu_initialized); /* was set by cpu_init() */
  570. cpu_clear(cpu, cpu_possible_map);
  571. per_cpu(x86_cpu_to_apicid, cpu) = BAD_APICID;
  572. }
  573. /* mark "stuck" area as not stuck */
  574. *((volatile unsigned long *)trampoline_base) = 0;
  575. return boot_error;
  576. }
  577. #ifdef CONFIG_HOTPLUG_CPU
  578. void cpu_exit_clear(void)
  579. {
  580. int cpu = raw_smp_processor_id();
  581. idle_task_exit();
  582. cpu_uninit();
  583. irq_ctx_exit(cpu);
  584. cpu_clear(cpu, cpu_callout_map);
  585. cpu_clear(cpu, cpu_callin_map);
  586. unmap_cpu_to_logical_apicid(cpu);
  587. }
  588. #endif
  589. static void __cpuinit __smp_prepare_cpu(int cpu)
  590. {
  591. int apicid;
  592. apicid = per_cpu(x86_cpu_to_apicid, cpu);
  593. /* init low mem mapping */
  594. clone_pgd_range(swapper_pg_dir, swapper_pg_dir + USER_PGD_PTRS,
  595. min_t(unsigned long, KERNEL_PGD_PTRS, USER_PGD_PTRS));
  596. flush_tlb_all();
  597. do_boot_cpu(apicid, cpu);
  598. }
  599. static int boot_cpu_logical_apicid;
  600. /* Where the IO area was mapped on multiquad, always 0 otherwise */
  601. void *xquad_portio;
  602. #ifdef CONFIG_X86_NUMAQ
  603. EXPORT_SYMBOL(xquad_portio);
  604. #endif
  605. static void __init disable_smp(void)
  606. {
  607. cpu_possible_map = cpumask_of_cpu(0);
  608. cpu_present_map = cpumask_of_cpu(0);
  609. smpboot_clear_io_apic_irqs();
  610. phys_cpu_present_map = physid_mask_of_physid(0);
  611. map_cpu_to_logical_apicid();
  612. cpu_set(0, per_cpu(cpu_sibling_map, 0));
  613. cpu_set(0, per_cpu(cpu_core_map, 0));
  614. }
  615. static int __init smp_sanity_check(unsigned max_cpus)
  616. {
  617. /*
  618. * If we couldn't find an SMP configuration at boot time,
  619. * get out of here now!
  620. */
  621. if (!smp_found_config && !acpi_lapic) {
  622. printk(KERN_NOTICE "SMP motherboard not detected.\n");
  623. disable_smp();
  624. if (APIC_init_uniprocessor())
  625. printk(KERN_NOTICE "Local APIC not detected."
  626. " Using dummy APIC emulation.\n");
  627. return -1;
  628. }
  629. /*
  630. * Should not be necessary because the MP table should list the boot
  631. * CPU too, but we do it for the sake of robustness anyway.
  632. * Makes no sense to do this check in clustered apic mode, so skip it
  633. */
  634. if (!check_phys_apicid_present(boot_cpu_physical_apicid)) {
  635. printk("weird, boot CPU (#%d) not listed by the BIOS.\n",
  636. boot_cpu_physical_apicid);
  637. physid_set(hard_smp_processor_id(), phys_cpu_present_map);
  638. }
  639. /*
  640. * If we couldn't find a local APIC, then get out of here now!
  641. */
  642. if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid]) && !cpu_has_apic) {
  643. printk(KERN_ERR "BIOS bug, local APIC #%d not detected!...\n",
  644. boot_cpu_physical_apicid);
  645. printk(KERN_ERR "... forcing use of dummy APIC emulation. (tell your hw vendor)\n");
  646. return -1;
  647. }
  648. verify_local_APIC();
  649. /*
  650. * If SMP should be disabled, then really disable it!
  651. */
  652. if (!max_cpus) {
  653. smp_found_config = 0;
  654. printk(KERN_INFO "SMP mode deactivated, forcing use of dummy APIC emulation.\n");
  655. if (nmi_watchdog == NMI_LOCAL_APIC) {
  656. printk(KERN_INFO "activating minimal APIC for NMI watchdog use.\n");
  657. connect_bsp_APIC();
  658. setup_local_APIC();
  659. end_local_APIC_setup();
  660. }
  661. return -1;
  662. }
  663. return 0;
  664. }
  665. /*
  666. * Cycle through the processors sending APIC IPIs to boot each.
  667. */
  668. static void __init smp_boot_cpus(unsigned int max_cpus)
  669. {
  670. /*
  671. * Setup boot CPU information
  672. */
  673. smp_store_cpu_info(0); /* Final full version of the data */
  674. printk(KERN_INFO "CPU%d: ", 0);
  675. print_cpu_info(&cpu_data(0));
  676. boot_cpu_physical_apicid = GET_APIC_ID(apic_read(APIC_ID));
  677. boot_cpu_logical_apicid = logical_smp_processor_id();
  678. current_thread_info()->cpu = 0;
  679. set_cpu_sibling_map(0);
  680. if (smp_sanity_check(max_cpus) < 0) {
  681. printk(KERN_INFO "SMP disabled\n");
  682. disable_smp();
  683. return;
  684. }
  685. connect_bsp_APIC();
  686. setup_local_APIC();
  687. end_local_APIC_setup();
  688. map_cpu_to_logical_apicid();
  689. setup_portio_remap();
  690. smpboot_setup_io_apic();
  691. setup_boot_clock();
  692. }
  693. /* These are wrappers to interface to the new boot process. Someone
  694. who understands all this stuff should rewrite it properly. --RR 15/Jul/02 */
  695. void __init native_smp_prepare_cpus(unsigned int max_cpus)
  696. {
  697. nmi_watchdog_default();
  698. cpu_callin_map = cpumask_of_cpu(0);
  699. mb();
  700. smp_boot_cpus(max_cpus);
  701. }
  702. void __init native_smp_prepare_boot_cpu(void)
  703. {
  704. unsigned int cpu = smp_processor_id();
  705. init_gdt(cpu);
  706. switch_to_new_gdt();
  707. cpu_set(cpu, cpu_callout_map);
  708. __get_cpu_var(cpu_state) = CPU_ONLINE;
  709. }
  710. int __cpuinit native_cpu_up(unsigned int cpu)
  711. {
  712. int apicid = cpu_present_to_apicid(cpu);
  713. unsigned long flags;
  714. WARN_ON(irqs_disabled());
  715. Dprintk("++++++++++++++++++++=_---CPU UP %u\n", cpu);
  716. if (apicid == BAD_APICID || apicid == boot_cpu_physical_apicid ||
  717. !physid_isset(apicid, phys_cpu_present_map)) {
  718. printk(KERN_ERR "%s: bad cpu %d\n", __func__, cpu);
  719. return -EINVAL;
  720. }
  721. /*
  722. * Save current MTRR state in case it was changed since early boot
  723. * (e.g. by the ACPI SMI) to initialize new CPUs with MTRRs in sync:
  724. */
  725. mtrr_save_state();
  726. per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
  727. __smp_prepare_cpu(cpu);
  728. /* In case one didn't come up */
  729. if (!cpu_isset(cpu, cpu_callin_map)) {
  730. printk(KERN_DEBUG "skipping cpu%d, didn't come online\n", cpu);
  731. return -EIO;
  732. }
  733. /*
  734. * Check TSC synchronization with the AP (keep irqs disabled
  735. * while doing so):
  736. */
  737. local_irq_save(flags);
  738. check_tsc_sync_source(cpu);
  739. local_irq_restore(flags);
  740. while (!cpu_isset(cpu, cpu_online_map)) {
  741. cpu_relax();
  742. touch_nmi_watchdog();
  743. }
  744. return 0;
  745. }
  746. extern void impress_friends(void);
  747. extern void smp_checks(void);
  748. void __init native_smp_cpus_done(unsigned int max_cpus)
  749. {
  750. /*
  751. * Cleanup possible dangling ends...
  752. */
  753. smpboot_restore_warm_reset_vector();
  754. Dprintk("Boot done.\n");
  755. impress_friends();
  756. smp_checks();
  757. #ifdef CONFIG_X86_IO_APIC
  758. setup_ioapic_dest();
  759. #endif
  760. check_nmi_watchdog();
  761. zap_low_mappings();
  762. }