smp.c 9.9 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362
  1. /*
  2. * Intel SMP support routines.
  3. *
  4. * (c) 1995 Alan Cox, Building #3 <alan@lxorguk.ukuu.org.uk>
  5. * (c) 1998-99, 2000, 2009 Ingo Molnar <mingo@redhat.com>
  6. * (c) 2002,2003 Andi Kleen, SuSE Labs.
  7. *
  8. * i386 and x86_64 integration by Glauber Costa <gcosta@redhat.com>
  9. *
  10. * This code is released under the GNU General Public License version 2 or
  11. * later.
  12. */
  13. #include <linux/init.h>
  14. #include <linux/mm.h>
  15. #include <linux/delay.h>
  16. #include <linux/spinlock.h>
  17. #include <linux/export.h>
  18. #include <linux/kernel_stat.h>
  19. #include <linux/mc146818rtc.h>
  20. #include <linux/cache.h>
  21. #include <linux/interrupt.h>
  22. #include <linux/cpu.h>
  23. #include <linux/gfp.h>
  24. #include <asm/mtrr.h>
  25. #include <asm/tlbflush.h>
  26. #include <asm/mmu_context.h>
  27. #include <asm/proto.h>
  28. #include <asm/apic.h>
  29. #include <asm/nmi.h>
  30. #include <asm/mce.h>
  31. #include <asm/trace/irq_vectors.h>
  32. #include <asm/kexec.h>
  33. /*
  34. * Some notes on x86 processor bugs affecting SMP operation:
  35. *
  36. * Pentium, Pentium Pro, II, III (and all CPUs) have bugs.
  37. * The Linux implications for SMP are handled as follows:
  38. *
  39. * Pentium III / [Xeon]
  40. * None of the E1AP-E3AP errata are visible to the user.
  41. *
  42. * E1AP. see PII A1AP
  43. * E2AP. see PII A2AP
  44. * E3AP. see PII A3AP
  45. *
  46. * Pentium II / [Xeon]
  47. * None of the A1AP-A3AP errata are visible to the user.
  48. *
  49. * A1AP. see PPro 1AP
  50. * A2AP. see PPro 2AP
  51. * A3AP. see PPro 7AP
  52. *
  53. * Pentium Pro
  54. * None of 1AP-9AP errata are visible to the normal user,
  55. * except occasional delivery of 'spurious interrupt' as trap #15.
  56. * This is very rare and a non-problem.
  57. *
  58. * 1AP. Linux maps APIC as non-cacheable
  59. * 2AP. worked around in hardware
  60. * 3AP. fixed in C0 and above steppings microcode update.
  61. * Linux does not use excessive STARTUP_IPIs.
  62. * 4AP. worked around in hardware
  63. * 5AP. symmetric IO mode (normal Linux operation) not affected.
  64. * 'noapic' mode has vector 0xf filled out properly.
  65. * 6AP. 'noapic' mode might be affected - fixed in later steppings
  66. * 7AP. We do not assume writes to the LVT deassering IRQs
  67. * 8AP. We do not enable low power mode (deep sleep) during MP bootup
  68. * 9AP. We do not use mixed mode
  69. *
  70. * Pentium
  71. * There is a marginal case where REP MOVS on 100MHz SMP
  72. * machines with B stepping processors can fail. XXX should provide
  73. * an L1cache=Writethrough or L1cache=off option.
  74. *
  75. * B stepping CPUs may hang. There are hardware work arounds
  76. * for this. We warn about it in case your board doesn't have the work
  77. * arounds. Basically that's so I can tell anyone with a B stepping
  78. * CPU and SMP problems "tough".
  79. *
  80. * Specific items [From Pentium Processor Specification Update]
  81. *
  82. * 1AP. Linux doesn't use remote read
  83. * 2AP. Linux doesn't trust APIC errors
  84. * 3AP. We work around this
  85. * 4AP. Linux never generated 3 interrupts of the same priority
  86. * to cause a lost local interrupt.
  87. * 5AP. Remote read is never used
  88. * 6AP. not affected - worked around in hardware
  89. * 7AP. not affected - worked around in hardware
  90. * 8AP. worked around in hardware - we get explicit CS errors if not
  91. * 9AP. only 'noapic' mode affected. Might generate spurious
  92. * interrupts, we log only the first one and count the
  93. * rest silently.
  94. * 10AP. not affected - worked around in hardware
  95. * 11AP. Linux reads the APIC between writes to avoid this, as per
  96. * the documentation. Make sure you preserve this as it affects
  97. * the C stepping chips too.
  98. * 12AP. not affected - worked around in hardware
  99. * 13AP. not affected - worked around in hardware
  100. * 14AP. we always deassert INIT during bootup
  101. * 15AP. not affected - worked around in hardware
  102. * 16AP. not affected - worked around in hardware
  103. * 17AP. not affected - worked around in hardware
  104. * 18AP. not affected - worked around in hardware
  105. * 19AP. not affected - worked around in BIOS
  106. *
  107. * If this sounds worrying believe me these bugs are either ___RARE___,
  108. * or are signal timing bugs worked around in hardware and there's
  109. * about nothing of note with C stepping upwards.
  110. */
  111. static atomic_t stopping_cpu = ATOMIC_INIT(-1);
  112. static bool smp_no_nmi_ipi = false;
  113. /*
  114. * this function sends a 'reschedule' IPI to another CPU.
  115. * it goes straight through and wastes no time serializing
  116. * anything. Worst case is that we lose a reschedule ...
  117. */
  118. static void native_smp_send_reschedule(int cpu)
  119. {
  120. if (unlikely(cpu_is_offline(cpu))) {
  121. WARN_ON(1);
  122. return;
  123. }
  124. apic->send_IPI(cpu, RESCHEDULE_VECTOR);
  125. }
  126. void native_send_call_func_single_ipi(int cpu)
  127. {
  128. apic->send_IPI(cpu, CALL_FUNCTION_SINGLE_VECTOR);
  129. }
  130. void native_send_call_func_ipi(const struct cpumask *mask)
  131. {
  132. cpumask_var_t allbutself;
  133. if (!alloc_cpumask_var(&allbutself, GFP_ATOMIC)) {
  134. apic->send_IPI_mask(mask, CALL_FUNCTION_VECTOR);
  135. return;
  136. }
  137. cpumask_copy(allbutself, cpu_online_mask);
  138. cpumask_clear_cpu(smp_processor_id(), allbutself);
  139. if (cpumask_equal(mask, allbutself) &&
  140. cpumask_equal(cpu_online_mask, cpu_callout_mask))
  141. apic->send_IPI_allbutself(CALL_FUNCTION_VECTOR);
  142. else
  143. apic->send_IPI_mask(mask, CALL_FUNCTION_VECTOR);
  144. free_cpumask_var(allbutself);
  145. }
  146. static int smp_stop_nmi_callback(unsigned int val, struct pt_regs *regs)
  147. {
  148. /* We are registered on stopping cpu too, avoid spurious NMI */
  149. if (raw_smp_processor_id() == atomic_read(&stopping_cpu))
  150. return NMI_HANDLED;
  151. stop_this_cpu(NULL);
  152. return NMI_HANDLED;
  153. }
  154. /*
  155. * this function calls the 'stop' function on all other CPUs in the system.
  156. */
  157. asmlinkage __visible void smp_reboot_interrupt(void)
  158. {
  159. ipi_entering_ack_irq();
  160. stop_this_cpu(NULL);
  161. irq_exit();
  162. }
  163. static void native_stop_other_cpus(int wait)
  164. {
  165. unsigned long flags;
  166. unsigned long timeout;
  167. if (reboot_force)
  168. return;
  169. /*
  170. * Use an own vector here because smp_call_function
  171. * does lots of things not suitable in a panic situation.
  172. */
  173. /*
  174. * We start by using the REBOOT_VECTOR irq.
  175. * The irq is treated as a sync point to allow critical
  176. * regions of code on other cpus to release their spin locks
  177. * and re-enable irqs. Jumping straight to an NMI might
  178. * accidentally cause deadlocks with further shutdown/panic
  179. * code. By syncing, we give the cpus up to one second to
  180. * finish their work before we force them off with the NMI.
  181. */
  182. if (num_online_cpus() > 1) {
  183. /* did someone beat us here? */
  184. if (atomic_cmpxchg(&stopping_cpu, -1, safe_smp_processor_id()) != -1)
  185. return;
  186. /* sync above data before sending IRQ */
  187. wmb();
  188. apic->send_IPI_allbutself(REBOOT_VECTOR);
  189. /*
  190. * Don't wait longer than a second if the caller
  191. * didn't ask us to wait.
  192. */
  193. timeout = USEC_PER_SEC;
  194. while (num_online_cpus() > 1 && (wait || timeout--))
  195. udelay(1);
  196. }
  197. /* if the REBOOT_VECTOR didn't work, try with the NMI */
  198. if ((num_online_cpus() > 1) && (!smp_no_nmi_ipi)) {
  199. if (register_nmi_handler(NMI_LOCAL, smp_stop_nmi_callback,
  200. NMI_FLAG_FIRST, "smp_stop"))
  201. /* Note: we ignore failures here */
  202. /* Hope the REBOOT_IRQ is good enough */
  203. goto finish;
  204. /* sync above data before sending IRQ */
  205. wmb();
  206. pr_emerg("Shutting down cpus with NMI\n");
  207. apic->send_IPI_allbutself(NMI_VECTOR);
  208. /*
  209. * Don't wait longer than a 10 ms if the caller
  210. * didn't ask us to wait.
  211. */
  212. timeout = USEC_PER_MSEC * 10;
  213. while (num_online_cpus() > 1 && (wait || timeout--))
  214. udelay(1);
  215. }
  216. finish:
  217. local_irq_save(flags);
  218. disable_local_APIC();
  219. mcheck_cpu_clear(this_cpu_ptr(&cpu_info));
  220. local_irq_restore(flags);
  221. }
  222. /*
  223. * Reschedule call back.
  224. */
  225. static inline void __smp_reschedule_interrupt(void)
  226. {
  227. inc_irq_stat(irq_resched_count);
  228. scheduler_ipi();
  229. }
  230. __visible void smp_reschedule_interrupt(struct pt_regs *regs)
  231. {
  232. irq_enter();
  233. ack_APIC_irq();
  234. __smp_reschedule_interrupt();
  235. irq_exit();
  236. /*
  237. * KVM uses this interrupt to force a cpu out of guest mode
  238. */
  239. }
  240. __visible void smp_trace_reschedule_interrupt(struct pt_regs *regs)
  241. {
  242. /*
  243. * Need to call irq_enter() before calling the trace point.
  244. * __smp_reschedule_interrupt() calls irq_enter/exit() too (in
  245. * scheduler_ipi(). This is OK, since those functions are allowed
  246. * to nest.
  247. */
  248. ipi_entering_ack_irq();
  249. trace_reschedule_entry(RESCHEDULE_VECTOR);
  250. __smp_reschedule_interrupt();
  251. trace_reschedule_exit(RESCHEDULE_VECTOR);
  252. exiting_irq();
  253. /*
  254. * KVM uses this interrupt to force a cpu out of guest mode
  255. */
  256. }
  257. static inline void __smp_call_function_interrupt(void)
  258. {
  259. generic_smp_call_function_interrupt();
  260. inc_irq_stat(irq_call_count);
  261. }
  262. __visible void smp_call_function_interrupt(struct pt_regs *regs)
  263. {
  264. ipi_entering_ack_irq();
  265. __smp_call_function_interrupt();
  266. exiting_irq();
  267. }
  268. __visible void smp_trace_call_function_interrupt(struct pt_regs *regs)
  269. {
  270. ipi_entering_ack_irq();
  271. trace_call_function_entry(CALL_FUNCTION_VECTOR);
  272. __smp_call_function_interrupt();
  273. trace_call_function_exit(CALL_FUNCTION_VECTOR);
  274. exiting_irq();
  275. }
  276. static inline void __smp_call_function_single_interrupt(void)
  277. {
  278. generic_smp_call_function_single_interrupt();
  279. inc_irq_stat(irq_call_count);
  280. }
  281. __visible void smp_call_function_single_interrupt(struct pt_regs *regs)
  282. {
  283. ipi_entering_ack_irq();
  284. __smp_call_function_single_interrupt();
  285. exiting_irq();
  286. }
  287. __visible void smp_trace_call_function_single_interrupt(struct pt_regs *regs)
  288. {
  289. ipi_entering_ack_irq();
  290. trace_call_function_single_entry(CALL_FUNCTION_SINGLE_VECTOR);
  291. __smp_call_function_single_interrupt();
  292. trace_call_function_single_exit(CALL_FUNCTION_SINGLE_VECTOR);
  293. exiting_irq();
  294. }
  295. static int __init nonmi_ipi_setup(char *str)
  296. {
  297. smp_no_nmi_ipi = true;
  298. return 1;
  299. }
  300. __setup("nonmi_ipi", nonmi_ipi_setup);
  301. struct smp_ops smp_ops = {
  302. .smp_prepare_boot_cpu = native_smp_prepare_boot_cpu,
  303. .smp_prepare_cpus = native_smp_prepare_cpus,
  304. .smp_cpus_done = native_smp_cpus_done,
  305. .stop_other_cpus = native_stop_other_cpus,
  306. #if defined(CONFIG_KEXEC_CORE)
  307. .crash_stop_other_cpus = kdump_nmi_shootdown_cpus,
  308. #endif
  309. .smp_send_reschedule = native_smp_send_reschedule,
  310. .cpu_up = native_cpu_up,
  311. .cpu_die = native_cpu_die,
  312. .cpu_disable = native_cpu_disable,
  313. .play_dead = native_play_dead,
  314. .send_call_func_ipi = native_send_call_func_ipi,
  315. .send_call_func_single_ipi = native_send_call_func_single_ipi,
  316. };
  317. EXPORT_SYMBOL_GPL(smp_ops);