nmi.c 9.6 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376
  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/init.h>
  11. #include <linux/errno.h>
  12. #include <linux/time.h>
  13. #include <linux/module.h>
  14. #include <asm/lowcore.h>
  15. #include <asm/smp.h>
  16. #include <asm/etr.h>
  17. #include <asm/cpu.h>
  18. #include <asm/nmi.h>
  19. #include <asm/crw.h>
  20. struct mcck_struct {
  21. int kill_task;
  22. int channel_report;
  23. int warning;
  24. unsigned long long mcck_code;
  25. };
  26. static DEFINE_PER_CPU(struct mcck_struct, cpu_mcck);
  27. static NORET_TYPE void s390_handle_damage(char *msg)
  28. {
  29. smp_send_stop();
  30. disabled_wait((unsigned long) __builtin_return_address(0));
  31. while (1);
  32. }
  33. /*
  34. * Main machine check handler function. Will be called with interrupts enabled
  35. * or disabled and machine checks enabled or disabled.
  36. */
  37. void s390_handle_mcck(void)
  38. {
  39. unsigned long flags;
  40. struct mcck_struct mcck;
  41. /*
  42. * Disable machine checks and get the current state of accumulated
  43. * machine checks. Afterwards delete the old state and enable machine
  44. * checks again.
  45. */
  46. local_irq_save(flags);
  47. local_mcck_disable();
  48. mcck = __get_cpu_var(cpu_mcck);
  49. memset(&__get_cpu_var(cpu_mcck), 0, sizeof(struct mcck_struct));
  50. clear_thread_flag(TIF_MCCK_PENDING);
  51. local_mcck_enable();
  52. local_irq_restore(flags);
  53. if (mcck.channel_report)
  54. crw_handle_channel_report();
  55. /*
  56. * A warning may remain for a prolonged period on the bare iron.
  57. * (actually until the machine is powered off, or the problem is gone)
  58. * So we just stop listening for the WARNING MCH and avoid continuously
  59. * being interrupted. One caveat is however, that we must do this per
  60. * processor and cannot use the smp version of ctl_clear_bit().
  61. * On VM we only get one interrupt per virtally presented machinecheck.
  62. * Though one suffices, we may get one interrupt per (virtual) cpu.
  63. */
  64. if (mcck.warning) { /* WARNING pending ? */
  65. static int mchchk_wng_posted = 0;
  66. /* Use single cpu clear, as we cannot handle smp here. */
  67. __ctl_clear_bit(14, 24); /* Disable WARNING MCH */
  68. if (xchg(&mchchk_wng_posted, 1) == 0)
  69. kill_cad_pid(SIGPWR, 1);
  70. }
  71. if (mcck.kill_task) {
  72. local_irq_enable();
  73. printk(KERN_EMERG "mcck: Terminating task because of machine "
  74. "malfunction (code 0x%016llx).\n", mcck.mcck_code);
  75. printk(KERN_EMERG "mcck: task: %s, pid: %d.\n",
  76. current->comm, current->pid);
  77. do_exit(SIGSEGV);
  78. }
  79. }
  80. EXPORT_SYMBOL_GPL(s390_handle_mcck);
  81. /*
  82. * returns 0 if all registers could be validated
  83. * returns 1 otherwise
  84. */
  85. static int notrace s390_revalidate_registers(struct mci *mci)
  86. {
  87. int kill_task;
  88. u64 tmpclock;
  89. u64 zero;
  90. void *fpt_save_area, *fpt_creg_save_area;
  91. kill_task = 0;
  92. zero = 0;
  93. if (!mci->gr) {
  94. /*
  95. * General purpose registers couldn't be restored and have
  96. * unknown contents. Process needs to be terminated.
  97. */
  98. kill_task = 1;
  99. }
  100. if (!mci->fp) {
  101. /*
  102. * Floating point registers can't be restored and
  103. * therefore the process needs to be terminated.
  104. */
  105. kill_task = 1;
  106. }
  107. #ifndef CONFIG_64BIT
  108. asm volatile(
  109. " ld 0,0(%0)\n"
  110. " ld 2,8(%0)\n"
  111. " ld 4,16(%0)\n"
  112. " ld 6,24(%0)"
  113. : : "a" (&S390_lowcore.floating_pt_save_area));
  114. #endif
  115. if (MACHINE_HAS_IEEE) {
  116. #ifdef CONFIG_64BIT
  117. fpt_save_area = &S390_lowcore.floating_pt_save_area;
  118. fpt_creg_save_area = &S390_lowcore.fpt_creg_save_area;
  119. #else
  120. fpt_save_area = (void *) S390_lowcore.extended_save_area_addr;
  121. fpt_creg_save_area = fpt_save_area + 128;
  122. #endif
  123. if (!mci->fc) {
  124. /*
  125. * Floating point control register can't be restored.
  126. * Task will be terminated.
  127. */
  128. asm volatile("lfpc 0(%0)" : : "a" (&zero), "m" (zero));
  129. kill_task = 1;
  130. } else
  131. asm volatile("lfpc 0(%0)" : : "a" (fpt_creg_save_area));
  132. asm volatile(
  133. " ld 0,0(%0)\n"
  134. " ld 1,8(%0)\n"
  135. " ld 2,16(%0)\n"
  136. " ld 3,24(%0)\n"
  137. " ld 4,32(%0)\n"
  138. " ld 5,40(%0)\n"
  139. " ld 6,48(%0)\n"
  140. " ld 7,56(%0)\n"
  141. " ld 8,64(%0)\n"
  142. " ld 9,72(%0)\n"
  143. " ld 10,80(%0)\n"
  144. " ld 11,88(%0)\n"
  145. " ld 12,96(%0)\n"
  146. " ld 13,104(%0)\n"
  147. " ld 14,112(%0)\n"
  148. " ld 15,120(%0)\n"
  149. : : "a" (fpt_save_area));
  150. }
  151. /* Revalidate access registers */
  152. asm volatile(
  153. " lam 0,15,0(%0)"
  154. : : "a" (&S390_lowcore.access_regs_save_area));
  155. if (!mci->ar) {
  156. /*
  157. * Access registers have unknown contents.
  158. * Terminating task.
  159. */
  160. kill_task = 1;
  161. }
  162. /* Revalidate control registers */
  163. if (!mci->cr) {
  164. /*
  165. * Control registers have unknown contents.
  166. * Can't recover and therefore stopping machine.
  167. */
  168. s390_handle_damage("invalid control registers.");
  169. } else {
  170. #ifdef CONFIG_64BIT
  171. asm volatile(
  172. " lctlg 0,15,0(%0)"
  173. : : "a" (&S390_lowcore.cregs_save_area));
  174. #else
  175. asm volatile(
  176. " lctl 0,15,0(%0)"
  177. : : "a" (&S390_lowcore.cregs_save_area));
  178. #endif
  179. }
  180. /*
  181. * We don't even try to revalidate the TOD register, since we simply
  182. * can't write something sensible into that register.
  183. */
  184. #ifdef CONFIG_64BIT
  185. /*
  186. * See if we can revalidate the TOD programmable register with its
  187. * old contents (should be zero) otherwise set it to zero.
  188. */
  189. if (!mci->pr)
  190. asm volatile(
  191. " sr 0,0\n"
  192. " sckpf"
  193. : : : "0", "cc");
  194. else
  195. asm volatile(
  196. " l 0,0(%0)\n"
  197. " sckpf"
  198. : : "a" (&S390_lowcore.tod_progreg_save_area)
  199. : "0", "cc");
  200. #endif
  201. /* Revalidate clock comparator register */
  202. asm volatile(
  203. " stck 0(%1)\n"
  204. " sckc 0(%1)"
  205. : "=m" (tmpclock) : "a" (&(tmpclock)) : "cc", "memory");
  206. /* Check if old PSW is valid */
  207. if (!mci->wp)
  208. /*
  209. * Can't tell if we come from user or kernel mode
  210. * -> stopping machine.
  211. */
  212. s390_handle_damage("old psw invalid.");
  213. if (!mci->ms || !mci->pm || !mci->ia)
  214. kill_task = 1;
  215. return kill_task;
  216. }
  217. #define MAX_IPD_COUNT 29
  218. #define MAX_IPD_TIME (5 * 60 * USEC_PER_SEC) /* 5 minutes */
  219. #define ED_STP_ISLAND 6 /* External damage STP island check */
  220. #define ED_STP_SYNC 7 /* External damage STP sync check */
  221. #define ED_ETR_SYNC 12 /* External damage ETR sync check */
  222. #define ED_ETR_SWITCH 13 /* External damage ETR switch to local */
  223. /*
  224. * machine check handler.
  225. */
  226. void notrace s390_do_machine_check(struct pt_regs *regs)
  227. {
  228. static int ipd_count;
  229. static DEFINE_SPINLOCK(ipd_lock);
  230. static unsigned long long last_ipd;
  231. struct mcck_struct *mcck;
  232. unsigned long long tmp;
  233. struct mci *mci;
  234. int umode;
  235. lockdep_off();
  236. s390_idle_check();
  237. mci = (struct mci *) &S390_lowcore.mcck_interruption_code;
  238. mcck = &__get_cpu_var(cpu_mcck);
  239. umode = user_mode(regs);
  240. if (mci->sd) {
  241. /* System damage -> stopping machine */
  242. s390_handle_damage("received system damage machine check.");
  243. }
  244. if (mci->pd) {
  245. if (mci->b) {
  246. /* Processing backup -> verify if we can survive this */
  247. u64 z_mcic, o_mcic, t_mcic;
  248. #ifdef CONFIG_64BIT
  249. z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<29);
  250. o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
  251. 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
  252. 1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 |
  253. 1ULL<<16);
  254. #else
  255. z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<57 | 1ULL<<50 |
  256. 1ULL<<29);
  257. o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
  258. 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
  259. 1ULL<<30 | 1ULL<<20 | 1ULL<<17 | 1ULL<<16);
  260. #endif
  261. t_mcic = *(u64 *)mci;
  262. if (((t_mcic & z_mcic) != 0) ||
  263. ((t_mcic & o_mcic) != o_mcic)) {
  264. s390_handle_damage("processing backup machine "
  265. "check with damage.");
  266. }
  267. /*
  268. * Nullifying exigent condition, therefore we might
  269. * retry this instruction.
  270. */
  271. spin_lock(&ipd_lock);
  272. tmp = get_clock();
  273. if (((tmp - last_ipd) >> 12) < MAX_IPD_TIME)
  274. ipd_count++;
  275. else
  276. ipd_count = 1;
  277. last_ipd = tmp;
  278. if (ipd_count == MAX_IPD_COUNT)
  279. s390_handle_damage("too many ipd retries.");
  280. spin_unlock(&ipd_lock);
  281. } else {
  282. /* Processing damage -> stopping machine */
  283. s390_handle_damage("received instruction processing "
  284. "damage machine check.");
  285. }
  286. }
  287. if (s390_revalidate_registers(mci)) {
  288. if (umode) {
  289. /*
  290. * Couldn't restore all register contents while in
  291. * user mode -> mark task for termination.
  292. */
  293. mcck->kill_task = 1;
  294. mcck->mcck_code = *(unsigned long long *) mci;
  295. set_thread_flag(TIF_MCCK_PENDING);
  296. } else {
  297. /*
  298. * Couldn't restore all register contents while in
  299. * kernel mode -> stopping machine.
  300. */
  301. s390_handle_damage("unable to revalidate registers.");
  302. }
  303. }
  304. if (mci->cd) {
  305. /* Timing facility damage */
  306. s390_handle_damage("TOD clock damaged");
  307. }
  308. if (mci->ed && mci->ec) {
  309. /* External damage */
  310. if (S390_lowcore.external_damage_code & (1U << ED_ETR_SYNC))
  311. etr_sync_check();
  312. if (S390_lowcore.external_damage_code & (1U << ED_ETR_SWITCH))
  313. etr_switch_to_local();
  314. if (S390_lowcore.external_damage_code & (1U << ED_STP_SYNC))
  315. stp_sync_check();
  316. if (S390_lowcore.external_damage_code & (1U << ED_STP_ISLAND))
  317. stp_island_check();
  318. }
  319. if (mci->se)
  320. /* Storage error uncorrected */
  321. s390_handle_damage("received storage error uncorrected "
  322. "machine check.");
  323. if (mci->ke)
  324. /* Storage key-error uncorrected */
  325. s390_handle_damage("received storage key-error uncorrected "
  326. "machine check.");
  327. if (mci->ds && mci->fa)
  328. /* Storage degradation */
  329. s390_handle_damage("received storage degradation machine "
  330. "check.");
  331. if (mci->cp) {
  332. /* Channel report word pending */
  333. mcck->channel_report = 1;
  334. set_thread_flag(TIF_MCCK_PENDING);
  335. }
  336. if (mci->w) {
  337. /* Warning pending */
  338. mcck->warning = 1;
  339. set_thread_flag(TIF_MCCK_PENDING);
  340. }
  341. lockdep_on();
  342. }
  343. static int __init machine_check_init(void)
  344. {
  345. ctl_set_bit(14, 25); /* enable external damage MCH */
  346. ctl_set_bit(14, 27); /* enable system recovery MCH */
  347. ctl_set_bit(14, 24); /* enable warning MCH */
  348. return 0;
  349. }
  350. arch_initcall(machine_check_init);