nmi.c 9.6 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372
  1. /*
  2. * Machine check handler
  3. *
  4. * Copyright IBM Corp. 2000, 2009
  5. * Author(s): Ingo Adlung <adlung@de.ibm.com>,
  6. * Martin Schwidefsky <schwidefsky@de.ibm.com>,
  7. * Cornelia Huck <cornelia.huck@de.ibm.com>,
  8. * Heiko Carstens <heiko.carstens@de.ibm.com>,
  9. */
  10. #include <linux/kernel_stat.h>
  11. #include <linux/init.h>
  12. #include <linux/errno.h>
  13. #include <linux/hardirq.h>
  14. #include <linux/time.h>
  15. #include <linux/module.h>
  16. #include <asm/lowcore.h>
  17. #include <asm/smp.h>
  18. #include <asm/etr.h>
  19. #include <asm/cputime.h>
  20. #include <asm/nmi.h>
  21. #include <asm/crw.h>
  22. struct mcck_struct {
  23. int kill_task;
  24. int channel_report;
  25. int warning;
  26. unsigned long long mcck_code;
  27. };
  28. static DEFINE_PER_CPU(struct mcck_struct, cpu_mcck);
  29. static void s390_handle_damage(char *msg)
  30. {
  31. smp_send_stop();
  32. disabled_wait((unsigned long) __builtin_return_address(0));
  33. while (1);
  34. }
  35. /*
  36. * Main machine check handler function. Will be called with interrupts enabled
  37. * or disabled and machine checks enabled or disabled.
  38. */
  39. void s390_handle_mcck(void)
  40. {
  41. unsigned long flags;
  42. struct mcck_struct mcck;
  43. /*
  44. * Disable machine checks and get the current state of accumulated
  45. * machine checks. Afterwards delete the old state and enable machine
  46. * checks again.
  47. */
  48. local_irq_save(flags);
  49. local_mcck_disable();
  50. mcck = __get_cpu_var(cpu_mcck);
  51. memset(&__get_cpu_var(cpu_mcck), 0, sizeof(struct mcck_struct));
  52. clear_thread_flag(TIF_MCCK_PENDING);
  53. local_mcck_enable();
  54. local_irq_restore(flags);
  55. if (mcck.channel_report)
  56. crw_handle_channel_report();
  57. /*
  58. * A warning may remain for a prolonged period on the bare iron.
  59. * (actually until the machine is powered off, or the problem is gone)
  60. * So we just stop listening for the WARNING MCH and avoid continuously
  61. * being interrupted. One caveat is however, that we must do this per
  62. * processor and cannot use the smp version of ctl_clear_bit().
  63. * On VM we only get one interrupt per virtally presented machinecheck.
  64. * Though one suffices, we may get one interrupt per (virtual) cpu.
  65. */
  66. if (mcck.warning) { /* WARNING pending ? */
  67. static int mchchk_wng_posted = 0;
  68. /* Use single cpu clear, as we cannot handle smp here. */
  69. __ctl_clear_bit(14, 24); /* Disable WARNING MCH */
  70. if (xchg(&mchchk_wng_posted, 1) == 0)
  71. kill_cad_pid(SIGPWR, 1);
  72. }
  73. if (mcck.kill_task) {
  74. local_irq_enable();
  75. printk(KERN_EMERG "mcck: Terminating task because of machine "
  76. "malfunction (code 0x%016llx).\n", mcck.mcck_code);
  77. printk(KERN_EMERG "mcck: task: %s, pid: %d.\n",
  78. current->comm, current->pid);
  79. do_exit(SIGSEGV);
  80. }
  81. }
  82. EXPORT_SYMBOL_GPL(s390_handle_mcck);
  83. /*
  84. * returns 0 if all registers could be validated
  85. * returns 1 otherwise
  86. */
  87. static int notrace s390_revalidate_registers(struct mci *mci)
  88. {
  89. int kill_task;
  90. u64 zero;
  91. void *fpt_save_area, *fpt_creg_save_area;
  92. kill_task = 0;
  93. zero = 0;
  94. if (!mci->gr) {
  95. /*
  96. * General purpose registers couldn't be restored and have
  97. * unknown contents. Process needs to be terminated.
  98. */
  99. kill_task = 1;
  100. }
  101. if (!mci->fp) {
  102. /*
  103. * Floating point registers can't be restored and
  104. * therefore the process needs to be terminated.
  105. */
  106. kill_task = 1;
  107. }
  108. #ifndef CONFIG_64BIT
  109. asm volatile(
  110. " ld 0,0(%0)\n"
  111. " ld 2,8(%0)\n"
  112. " ld 4,16(%0)\n"
  113. " ld 6,24(%0)"
  114. : : "a" (&S390_lowcore.floating_pt_save_area));
  115. #endif
  116. if (MACHINE_HAS_IEEE) {
  117. #ifdef CONFIG_64BIT
  118. fpt_save_area = &S390_lowcore.floating_pt_save_area;
  119. fpt_creg_save_area = &S390_lowcore.fpt_creg_save_area;
  120. #else
  121. fpt_save_area = (void *) S390_lowcore.extended_save_area_addr;
  122. fpt_creg_save_area = fpt_save_area + 128;
  123. #endif
  124. if (!mci->fc) {
  125. /*
  126. * Floating point control register can't be restored.
  127. * Task will be terminated.
  128. */
  129. asm volatile("lfpc 0(%0)" : : "a" (&zero), "m" (zero));
  130. kill_task = 1;
  131. } else
  132. asm volatile("lfpc 0(%0)" : : "a" (fpt_creg_save_area));
  133. asm volatile(
  134. " ld 0,0(%0)\n"
  135. " ld 1,8(%0)\n"
  136. " ld 2,16(%0)\n"
  137. " ld 3,24(%0)\n"
  138. " ld 4,32(%0)\n"
  139. " ld 5,40(%0)\n"
  140. " ld 6,48(%0)\n"
  141. " ld 7,56(%0)\n"
  142. " ld 8,64(%0)\n"
  143. " ld 9,72(%0)\n"
  144. " ld 10,80(%0)\n"
  145. " ld 11,88(%0)\n"
  146. " ld 12,96(%0)\n"
  147. " ld 13,104(%0)\n"
  148. " ld 14,112(%0)\n"
  149. " ld 15,120(%0)\n"
  150. : : "a" (fpt_save_area));
  151. }
  152. /* Revalidate access registers */
  153. asm volatile(
  154. " lam 0,15,0(%0)"
  155. : : "a" (&S390_lowcore.access_regs_save_area));
  156. if (!mci->ar) {
  157. /*
  158. * Access registers have unknown contents.
  159. * Terminating task.
  160. */
  161. kill_task = 1;
  162. }
  163. /* Revalidate control registers */
  164. if (!mci->cr) {
  165. /*
  166. * Control registers have unknown contents.
  167. * Can't recover and therefore stopping machine.
  168. */
  169. s390_handle_damage("invalid control registers.");
  170. } else {
  171. #ifdef CONFIG_64BIT
  172. asm volatile(
  173. " lctlg 0,15,0(%0)"
  174. : : "a" (&S390_lowcore.cregs_save_area));
  175. #else
  176. asm volatile(
  177. " lctl 0,15,0(%0)"
  178. : : "a" (&S390_lowcore.cregs_save_area));
  179. #endif
  180. }
  181. /*
  182. * We don't even try to revalidate the TOD register, since we simply
  183. * can't write something sensible into that register.
  184. */
  185. #ifdef CONFIG_64BIT
  186. /*
  187. * See if we can revalidate the TOD programmable register with its
  188. * old contents (should be zero) otherwise set it to zero.
  189. */
  190. if (!mci->pr)
  191. asm volatile(
  192. " sr 0,0\n"
  193. " sckpf"
  194. : : : "0", "cc");
  195. else
  196. asm volatile(
  197. " l 0,0(%0)\n"
  198. " sckpf"
  199. : : "a" (&S390_lowcore.tod_progreg_save_area)
  200. : "0", "cc");
  201. #endif
  202. /* Revalidate clock comparator register */
  203. set_clock_comparator(S390_lowcore.clock_comparator);
  204. /* Check if old PSW is valid */
  205. if (!mci->wp)
  206. /*
  207. * Can't tell if we come from user or kernel mode
  208. * -> stopping machine.
  209. */
  210. s390_handle_damage("old psw invalid.");
  211. if (!mci->ms || !mci->pm || !mci->ia)
  212. kill_task = 1;
  213. return kill_task;
  214. }
  215. #define MAX_IPD_COUNT 29
  216. #define MAX_IPD_TIME (5 * 60 * USEC_PER_SEC) /* 5 minutes */
  217. #define ED_STP_ISLAND 6 /* External damage STP island check */
  218. #define ED_STP_SYNC 7 /* External damage STP sync check */
  219. #define ED_ETR_SYNC 12 /* External damage ETR sync check */
  220. #define ED_ETR_SWITCH 13 /* External damage ETR switch to local */
  221. /*
  222. * machine check handler.
  223. */
  224. void notrace s390_do_machine_check(struct pt_regs *regs)
  225. {
  226. static int ipd_count;
  227. static DEFINE_SPINLOCK(ipd_lock);
  228. static unsigned long long last_ipd;
  229. struct mcck_struct *mcck;
  230. unsigned long long tmp;
  231. struct mci *mci;
  232. int umode;
  233. nmi_enter();
  234. inc_irq_stat(NMI_NMI);
  235. mci = (struct mci *) &S390_lowcore.mcck_interruption_code;
  236. mcck = &__get_cpu_var(cpu_mcck);
  237. umode = user_mode(regs);
  238. if (mci->sd) {
  239. /* System damage -> stopping machine */
  240. s390_handle_damage("received system damage machine check.");
  241. }
  242. if (mci->pd) {
  243. if (mci->b) {
  244. /* Processing backup -> verify if we can survive this */
  245. u64 z_mcic, o_mcic, t_mcic;
  246. #ifdef CONFIG_64BIT
  247. z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<29);
  248. o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
  249. 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
  250. 1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 |
  251. 1ULL<<16);
  252. #else
  253. z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<57 | 1ULL<<50 |
  254. 1ULL<<29);
  255. o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
  256. 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
  257. 1ULL<<30 | 1ULL<<20 | 1ULL<<17 | 1ULL<<16);
  258. #endif
  259. t_mcic = *(u64 *)mci;
  260. if (((t_mcic & z_mcic) != 0) ||
  261. ((t_mcic & o_mcic) != o_mcic)) {
  262. s390_handle_damage("processing backup machine "
  263. "check with damage.");
  264. }
  265. /*
  266. * Nullifying exigent condition, therefore we might
  267. * retry this instruction.
  268. */
  269. spin_lock(&ipd_lock);
  270. tmp = get_tod_clock();
  271. if (((tmp - last_ipd) >> 12) < MAX_IPD_TIME)
  272. ipd_count++;
  273. else
  274. ipd_count = 1;
  275. last_ipd = tmp;
  276. if (ipd_count == MAX_IPD_COUNT)
  277. s390_handle_damage("too many ipd retries.");
  278. spin_unlock(&ipd_lock);
  279. } else {
  280. /* Processing damage -> stopping machine */
  281. s390_handle_damage("received instruction processing "
  282. "damage machine check.");
  283. }
  284. }
  285. if (s390_revalidate_registers(mci)) {
  286. if (umode) {
  287. /*
  288. * Couldn't restore all register contents while in
  289. * user mode -> mark task for termination.
  290. */
  291. mcck->kill_task = 1;
  292. mcck->mcck_code = *(unsigned long long *) mci;
  293. set_thread_flag(TIF_MCCK_PENDING);
  294. } else {
  295. /*
  296. * Couldn't restore all register contents while in
  297. * kernel mode -> stopping machine.
  298. */
  299. s390_handle_damage("unable to revalidate registers.");
  300. }
  301. }
  302. if (mci->cd) {
  303. /* Timing facility damage */
  304. s390_handle_damage("TOD clock damaged");
  305. }
  306. if (mci->ed && mci->ec) {
  307. /* External damage */
  308. if (S390_lowcore.external_damage_code & (1U << ED_ETR_SYNC))
  309. etr_sync_check();
  310. if (S390_lowcore.external_damage_code & (1U << ED_ETR_SWITCH))
  311. etr_switch_to_local();
  312. if (S390_lowcore.external_damage_code & (1U << ED_STP_SYNC))
  313. stp_sync_check();
  314. if (S390_lowcore.external_damage_code & (1U << ED_STP_ISLAND))
  315. stp_island_check();
  316. }
  317. if (mci->se)
  318. /* Storage error uncorrected */
  319. s390_handle_damage("received storage error uncorrected "
  320. "machine check.");
  321. if (mci->ke)
  322. /* Storage key-error uncorrected */
  323. s390_handle_damage("received storage key-error uncorrected "
  324. "machine check.");
  325. if (mci->ds && mci->fa)
  326. /* Storage degradation */
  327. s390_handle_damage("received storage degradation machine "
  328. "check.");
  329. if (mci->cp) {
  330. /* Channel report word pending */
  331. mcck->channel_report = 1;
  332. set_thread_flag(TIF_MCCK_PENDING);
  333. }
  334. if (mci->w) {
  335. /* Warning pending */
  336. mcck->warning = 1;
  337. set_thread_flag(TIF_MCCK_PENDING);
  338. }
  339. nmi_exit();
  340. }
  341. static int __init machine_check_init(void)
  342. {
  343. ctl_set_bit(14, 25); /* enable external damage MCH */
  344. ctl_set_bit(14, 27); /* enable system recovery MCH */
  345. ctl_set_bit(14, 24); /* enable warning MCH */
  346. return 0;
  347. }
  348. arch_initcall(machine_check_init);