blob: fab88431a06fcb70a92b4fdfe0f475d269be67a6 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Heiko Carstensf5daba12009-03-26 15:24:01 +01002 * Machine check handler
Linus Torvalds1da177e2005-04-16 15:20:36 -07003 *
Heiko Carstensf5daba12009-03-26 15:24:01 +01004 * 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>,
Linus Torvalds1da177e2005-04-16 15:20:36 -07009 */
10
Heiko Carstens052ff462011-01-05 12:47:28 +010011#include <linux/kernel_stat.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070013#include <linux/errno.h>
Heiko Carstens81f64b82009-04-14 15:36:18 +020014#include <linux/hardirq.h>
Heiko Carstens022e4fc2006-05-01 12:16:14 -070015#include <linux/time.h>
Heiko Carstensf5daba12009-03-26 15:24:01 +010016#include <linux/module.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <asm/lowcore.h>
Heiko Carstensf5daba12009-03-26 15:24:01 +010018#include <asm/smp.h>
19#include <asm/etr.h>
Martin Schwidefsky76d4e002009-06-12 10:26:21 +020020#include <asm/cputime.h>
Heiko Carstensf5daba12009-03-26 15:24:01 +010021#include <asm/nmi.h>
22#include <asm/crw.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023
Heiko Carstens77fa2242005-06-25 14:55:30 -070024struct mcck_struct {
25 int kill_task;
26 int channel_report;
27 int warning;
28 unsigned long long mcck_code;
29};
30
31static DEFINE_PER_CPU(struct mcck_struct, cpu_mcck);
32
Heiko Carstensf5daba12009-03-26 15:24:01 +010033static NORET_TYPE void s390_handle_damage(char *msg)
34{
35 smp_send_stop();
36 disabled_wait((unsigned long) __builtin_return_address(0));
37 while (1);
38}
39
Linus Torvalds1da177e2005-04-16 15:20:36 -070040/*
Heiko Carstens77fa2242005-06-25 14:55:30 -070041 * Main machine check handler function. Will be called with interrupts enabled
42 * or disabled and machine checks enabled or disabled.
Linus Torvalds1da177e2005-04-16 15:20:36 -070043 */
Heiko Carstensf5daba12009-03-26 15:24:01 +010044void s390_handle_mcck(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -070045{
Heiko Carstens77fa2242005-06-25 14:55:30 -070046 unsigned long flags;
47 struct mcck_struct mcck;
Linus Torvalds1da177e2005-04-16 15:20:36 -070048
Heiko Carstens77fa2242005-06-25 14:55:30 -070049 /*
50 * Disable machine checks and get the current state of accumulated
51 * machine checks. Afterwards delete the old state and enable machine
52 * checks again.
53 */
54 local_irq_save(flags);
55 local_mcck_disable();
56 mcck = __get_cpu_var(cpu_mcck);
57 memset(&__get_cpu_var(cpu_mcck), 0, sizeof(struct mcck_struct));
58 clear_thread_flag(TIF_MCCK_PENDING);
59 local_mcck_enable();
60 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -070061
Heiko Carstens77fa2242005-06-25 14:55:30 -070062 if (mcck.channel_report)
Heiko Carstensf5daba12009-03-26 15:24:01 +010063 crw_handle_channel_report();
Heiko Carstens7b886412009-03-26 15:24:02 +010064 /*
65 * A warning may remain for a prolonged period on the bare iron.
66 * (actually until the machine is powered off, or the problem is gone)
67 * So we just stop listening for the WARNING MCH and avoid continuously
68 * being interrupted. One caveat is however, that we must do this per
69 * processor and cannot use the smp version of ctl_clear_bit().
70 * On VM we only get one interrupt per virtally presented machinecheck.
71 * Though one suffices, we may get one interrupt per (virtual) cpu.
72 */
Heiko Carstens77fa2242005-06-25 14:55:30 -070073 if (mcck.warning) { /* WARNING pending ? */
Linus Torvalds1da177e2005-04-16 15:20:36 -070074 static int mchchk_wng_posted = 0;
Heiko Carstens7b886412009-03-26 15:24:02 +010075
76 /* Use single cpu clear, as we cannot handle smp here. */
Linus Torvalds1da177e2005-04-16 15:20:36 -070077 __ctl_clear_bit(14, 24); /* Disable WARNING MCH */
78 if (xchg(&mchchk_wng_posted, 1) == 0)
Cedric Le Goater9ec52092006-10-02 02:19:00 -070079 kill_cad_pid(SIGPWR, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 }
Heiko Carstens77fa2242005-06-25 14:55:30 -070081 if (mcck.kill_task) {
82 local_irq_enable();
83 printk(KERN_EMERG "mcck: Terminating task because of machine "
84 "malfunction (code 0x%016llx).\n", mcck.mcck_code);
85 printk(KERN_EMERG "mcck: task: %s, pid: %d.\n",
86 current->comm, current->pid);
87 do_exit(SIGSEGV);
88 }
89}
Christian Borntraeger71cde582008-05-21 13:37:34 +020090EXPORT_SYMBOL_GPL(s390_handle_mcck);
Heiko Carstens77fa2242005-06-25 14:55:30 -070091
92/*
93 * returns 0 if all registers could be validated
94 * returns 1 otherwise
95 */
Heiko Carstensf5daba12009-03-26 15:24:01 +010096static int notrace s390_revalidate_registers(struct mci *mci)
Heiko Carstens77fa2242005-06-25 14:55:30 -070097{
98 int kill_task;
Heiko Carstens77fa2242005-06-25 14:55:30 -070099 u64 zero;
100 void *fpt_save_area, *fpt_creg_save_area;
101
102 kill_task = 0;
103 zero = 0;
Heiko Carstensf5daba12009-03-26 15:24:01 +0100104
105 if (!mci->gr) {
Heiko Carstens77fa2242005-06-25 14:55:30 -0700106 /*
107 * General purpose registers couldn't be restored and have
108 * unknown contents. Process needs to be terminated.
109 */
110 kill_task = 1;
Heiko Carstensf5daba12009-03-26 15:24:01 +0100111 }
112 if (!mci->fp) {
Heiko Carstens77fa2242005-06-25 14:55:30 -0700113 /*
114 * Floating point registers can't be restored and
115 * therefore the process needs to be terminated.
116 */
117 kill_task = 1;
Heiko Carstensf5daba12009-03-26 15:24:01 +0100118 }
Martin Schwidefsky347a8dc2006-01-06 00:19:28 -0800119#ifndef CONFIG_64BIT
Martin Schwidefsky94c12cc2006-09-28 16:56:43 +0200120 asm volatile(
121 " ld 0,0(%0)\n"
122 " ld 2,8(%0)\n"
123 " ld 4,16(%0)\n"
124 " ld 6,24(%0)"
125 : : "a" (&S390_lowcore.floating_pt_save_area));
Heiko Carstens77fa2242005-06-25 14:55:30 -0700126#endif
127
128 if (MACHINE_HAS_IEEE) {
Martin Schwidefsky347a8dc2006-01-06 00:19:28 -0800129#ifdef CONFIG_64BIT
Heiko Carstens77fa2242005-06-25 14:55:30 -0700130 fpt_save_area = &S390_lowcore.floating_pt_save_area;
131 fpt_creg_save_area = &S390_lowcore.fpt_creg_save_area;
132#else
133 fpt_save_area = (void *) S390_lowcore.extended_save_area_addr;
Heiko Carstensf5daba12009-03-26 15:24:01 +0100134 fpt_creg_save_area = fpt_save_area + 128;
Heiko Carstens77fa2242005-06-25 14:55:30 -0700135#endif
Heiko Carstens77fa2242005-06-25 14:55:30 -0700136 if (!mci->fc) {
137 /*
138 * Floating point control register can't be restored.
139 * Task will be terminated.
140 */
Martin Schwidefsky94c12cc2006-09-28 16:56:43 +0200141 asm volatile("lfpc 0(%0)" : : "a" (&zero), "m" (zero));
Heiko Carstens77fa2242005-06-25 14:55:30 -0700142 kill_task = 1;
143
Martin Schwidefsky94c12cc2006-09-28 16:56:43 +0200144 } else
145 asm volatile("lfpc 0(%0)" : : "a" (fpt_creg_save_area));
Heiko Carstens77fa2242005-06-25 14:55:30 -0700146
Martin Schwidefsky94c12cc2006-09-28 16:56:43 +0200147 asm volatile(
148 " ld 0,0(%0)\n"
149 " ld 1,8(%0)\n"
150 " ld 2,16(%0)\n"
151 " ld 3,24(%0)\n"
152 " ld 4,32(%0)\n"
153 " ld 5,40(%0)\n"
154 " ld 6,48(%0)\n"
155 " ld 7,56(%0)\n"
156 " ld 8,64(%0)\n"
157 " ld 9,72(%0)\n"
158 " ld 10,80(%0)\n"
159 " ld 11,88(%0)\n"
160 " ld 12,96(%0)\n"
161 " ld 13,104(%0)\n"
162 " ld 14,112(%0)\n"
163 " ld 15,120(%0)\n"
164 : : "a" (fpt_save_area));
Heiko Carstens77fa2242005-06-25 14:55:30 -0700165 }
Heiko Carstens77fa2242005-06-25 14:55:30 -0700166 /* Revalidate access registers */
Martin Schwidefsky94c12cc2006-09-28 16:56:43 +0200167 asm volatile(
168 " lam 0,15,0(%0)"
169 : : "a" (&S390_lowcore.access_regs_save_area));
Heiko Carstensf5daba12009-03-26 15:24:01 +0100170 if (!mci->ar) {
Heiko Carstens77fa2242005-06-25 14:55:30 -0700171 /*
172 * Access registers have unknown contents.
173 * Terminating task.
174 */
175 kill_task = 1;
Heiko Carstensf5daba12009-03-26 15:24:01 +0100176 }
Heiko Carstens77fa2242005-06-25 14:55:30 -0700177 /* Revalidate control registers */
Heiko Carstensf5daba12009-03-26 15:24:01 +0100178 if (!mci->cr) {
Heiko Carstens77fa2242005-06-25 14:55:30 -0700179 /*
180 * Control registers have unknown contents.
181 * Can't recover and therefore stopping machine.
182 */
183 s390_handle_damage("invalid control registers.");
Heiko Carstensf5daba12009-03-26 15:24:01 +0100184 } else {
Martin Schwidefsky347a8dc2006-01-06 00:19:28 -0800185#ifdef CONFIG_64BIT
Martin Schwidefsky94c12cc2006-09-28 16:56:43 +0200186 asm volatile(
187 " lctlg 0,15,0(%0)"
188 : : "a" (&S390_lowcore.cregs_save_area));
Heiko Carstens77fa2242005-06-25 14:55:30 -0700189#else
Martin Schwidefsky94c12cc2006-09-28 16:56:43 +0200190 asm volatile(
191 " lctl 0,15,0(%0)"
192 : : "a" (&S390_lowcore.cregs_save_area));
Heiko Carstens77fa2242005-06-25 14:55:30 -0700193#endif
Heiko Carstensf5daba12009-03-26 15:24:01 +0100194 }
Heiko Carstens77fa2242005-06-25 14:55:30 -0700195 /*
196 * We don't even try to revalidate the TOD register, since we simply
197 * can't write something sensible into that register.
198 */
Martin Schwidefsky347a8dc2006-01-06 00:19:28 -0800199#ifdef CONFIG_64BIT
Heiko Carstens77fa2242005-06-25 14:55:30 -0700200 /*
201 * See if we can revalidate the TOD programmable register with its
202 * old contents (should be zero) otherwise set it to zero.
203 */
204 if (!mci->pr)
Martin Schwidefsky94c12cc2006-09-28 16:56:43 +0200205 asm volatile(
206 " sr 0,0\n"
207 " sckpf"
208 : : : "0", "cc");
Heiko Carstens77fa2242005-06-25 14:55:30 -0700209 else
210 asm volatile(
Martin Schwidefsky94c12cc2006-09-28 16:56:43 +0200211 " l 0,0(%0)\n"
212 " sckpf"
213 : : "a" (&S390_lowcore.tod_progreg_save_area)
214 : "0", "cc");
Heiko Carstens77fa2242005-06-25 14:55:30 -0700215#endif
Heiko Carstens77fa2242005-06-25 14:55:30 -0700216 /* Revalidate clock comparator register */
Heiko Carstense8129c62010-11-25 09:52:45 +0100217 if (S390_lowcore.clock_comparator == -1)
218 set_clock_comparator(S390_lowcore.mcck_clock);
219 else
220 set_clock_comparator(S390_lowcore.clock_comparator);
Heiko Carstens77fa2242005-06-25 14:55:30 -0700221 /* Check if old PSW is valid */
222 if (!mci->wp)
223 /*
224 * Can't tell if we come from user or kernel mode
225 * -> stopping machine.
226 */
227 s390_handle_damage("old psw invalid.");
228
229 if (!mci->ms || !mci->pm || !mci->ia)
230 kill_task = 1;
231
232 return kill_task;
233}
234
Heiko Carstensb73d40c2006-04-27 18:40:23 -0700235#define MAX_IPD_COUNT 29
Heiko Carstens022e4fc2006-05-01 12:16:14 -0700236#define MAX_IPD_TIME (5 * 60 * USEC_PER_SEC) /* 5 minutes */
Heiko Carstensb73d40c2006-04-27 18:40:23 -0700237
Heiko Carstensf5daba12009-03-26 15:24:01 +0100238#define ED_STP_ISLAND 6 /* External damage STP island check */
239#define ED_STP_SYNC 7 /* External damage STP sync check */
240#define ED_ETR_SYNC 12 /* External damage ETR sync check */
241#define ED_ETR_SWITCH 13 /* External damage ETR switch to local */
242
Heiko Carstens77fa2242005-06-25 14:55:30 -0700243/*
244 * machine check handler.
245 */
Heiko Carstenscc54c1e2009-03-26 15:23:59 +0100246void notrace s390_do_machine_check(struct pt_regs *regs)
Heiko Carstens77fa2242005-06-25 14:55:30 -0700247{
Heiko Carstensf5daba12009-03-26 15:24:01 +0100248 static int ipd_count;
Heiko Carstensb73d40c2006-04-27 18:40:23 -0700249 static DEFINE_SPINLOCK(ipd_lock);
250 static unsigned long long last_ipd;
Heiko Carstensf5daba12009-03-26 15:24:01 +0100251 struct mcck_struct *mcck;
Heiko Carstensb73d40c2006-04-27 18:40:23 -0700252 unsigned long long tmp;
Heiko Carstens77fa2242005-06-25 14:55:30 -0700253 struct mci *mci;
Heiko Carstens77fa2242005-06-25 14:55:30 -0700254 int umode;
255
Heiko Carstens81f64b82009-04-14 15:36:18 +0200256 nmi_enter();
Martin Schwidefsky63779812010-05-17 10:00:03 +0200257 s390_idle_check(regs, S390_lowcore.mcck_clock,
258 S390_lowcore.mcck_enter_timer);
Heiko Carstens052ff462011-01-05 12:47:28 +0100259 kstat_cpu(smp_processor_id()).irqs[NMI_NMI]++;
Heiko Carstens77fa2242005-06-25 14:55:30 -0700260 mci = (struct mci *) &S390_lowcore.mcck_interruption_code;
261 mcck = &__get_cpu_var(cpu_mcck);
262 umode = user_mode(regs);
263
Heiko Carstensf5daba12009-03-26 15:24:01 +0100264 if (mci->sd) {
Heiko Carstens77fa2242005-06-25 14:55:30 -0700265 /* System damage -> stopping machine */
266 s390_handle_damage("received system damage machine check.");
Heiko Carstensf5daba12009-03-26 15:24:01 +0100267 }
Heiko Carstens77fa2242005-06-25 14:55:30 -0700268 if (mci->pd) {
269 if (mci->b) {
270 /* Processing backup -> verify if we can survive this */
271 u64 z_mcic, o_mcic, t_mcic;
Martin Schwidefsky347a8dc2006-01-06 00:19:28 -0800272#ifdef CONFIG_64BIT
Heiko Carstens77fa2242005-06-25 14:55:30 -0700273 z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<29);
274 o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
275 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
276 1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 |
277 1ULL<<16);
278#else
279 z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<57 | 1ULL<<50 |
280 1ULL<<29);
281 o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
282 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
283 1ULL<<30 | 1ULL<<20 | 1ULL<<17 | 1ULL<<16);
284#endif
285 t_mcic = *(u64 *)mci;
286
287 if (((t_mcic & z_mcic) != 0) ||
288 ((t_mcic & o_mcic) != o_mcic)) {
289 s390_handle_damage("processing backup machine "
290 "check with damage.");
291 }
Heiko Carstensb73d40c2006-04-27 18:40:23 -0700292
293 /*
294 * Nullifying exigent condition, therefore we might
295 * retry this instruction.
296 */
Heiko Carstensb73d40c2006-04-27 18:40:23 -0700297 spin_lock(&ipd_lock);
Heiko Carstensb73d40c2006-04-27 18:40:23 -0700298 tmp = get_clock();
Heiko Carstensb73d40c2006-04-27 18:40:23 -0700299 if (((tmp - last_ipd) >> 12) < MAX_IPD_TIME)
300 ipd_count++;
301 else
302 ipd_count = 1;
Heiko Carstensb73d40c2006-04-27 18:40:23 -0700303 last_ipd = tmp;
Heiko Carstensb73d40c2006-04-27 18:40:23 -0700304 if (ipd_count == MAX_IPD_COUNT)
305 s390_handle_damage("too many ipd retries.");
Heiko Carstensb73d40c2006-04-27 18:40:23 -0700306 spin_unlock(&ipd_lock);
Heiko Carstensf5daba12009-03-26 15:24:01 +0100307 } else {
Heiko Carstens77fa2242005-06-25 14:55:30 -0700308 /* Processing damage -> stopping machine */
309 s390_handle_damage("received instruction processing "
310 "damage machine check.");
311 }
312 }
313 if (s390_revalidate_registers(mci)) {
314 if (umode) {
315 /*
316 * Couldn't restore all register contents while in
317 * user mode -> mark task for termination.
318 */
319 mcck->kill_task = 1;
320 mcck->mcck_code = *(unsigned long long *) mci;
321 set_thread_flag(TIF_MCCK_PENDING);
Heiko Carstensf5daba12009-03-26 15:24:01 +0100322 } else {
Heiko Carstens77fa2242005-06-25 14:55:30 -0700323 /*
324 * Couldn't restore all register contents while in
325 * kernel mode -> stopping machine.
326 */
327 s390_handle_damage("unable to revalidate registers.");
Heiko Carstensf5daba12009-03-26 15:24:01 +0100328 }
Heiko Carstens77fa2242005-06-25 14:55:30 -0700329 }
Martin Schwidefskyd54853e2007-02-05 21:18:19 +0100330 if (mci->cd) {
331 /* Timing facility damage */
332 s390_handle_damage("TOD clock damaged");
333 }
Martin Schwidefskyd54853e2007-02-05 21:18:19 +0100334 if (mci->ed && mci->ec) {
335 /* External damage */
336 if (S390_lowcore.external_damage_code & (1U << ED_ETR_SYNC))
337 etr_sync_check();
338 if (S390_lowcore.external_damage_code & (1U << ED_ETR_SWITCH))
339 etr_switch_to_local();
Martin Schwidefskyd2fec592008-07-14 09:58:56 +0200340 if (S390_lowcore.external_damage_code & (1U << ED_STP_SYNC))
341 stp_sync_check();
342 if (S390_lowcore.external_damage_code & (1U << ED_STP_ISLAND))
343 stp_island_check();
Martin Schwidefskyd54853e2007-02-05 21:18:19 +0100344 }
Heiko Carstens77fa2242005-06-25 14:55:30 -0700345 if (mci->se)
346 /* Storage error uncorrected */
347 s390_handle_damage("received storage error uncorrected "
348 "machine check.");
Heiko Carstens77fa2242005-06-25 14:55:30 -0700349 if (mci->ke)
350 /* Storage key-error uncorrected */
351 s390_handle_damage("received storage key-error uncorrected "
352 "machine check.");
Heiko Carstens77fa2242005-06-25 14:55:30 -0700353 if (mci->ds && mci->fa)
354 /* Storage degradation */
355 s390_handle_damage("received storage degradation machine "
356 "check.");
Heiko Carstens77fa2242005-06-25 14:55:30 -0700357 if (mci->cp) {
358 /* Channel report word pending */
359 mcck->channel_report = 1;
360 set_thread_flag(TIF_MCCK_PENDING);
361 }
Heiko Carstens77fa2242005-06-25 14:55:30 -0700362 if (mci->w) {
363 /* Warning pending */
364 mcck->warning = 1;
365 set_thread_flag(TIF_MCCK_PENDING);
366 }
Heiko Carstens81f64b82009-04-14 15:36:18 +0200367 nmi_exit();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368}
369
Heiko Carstensf5daba12009-03-26 15:24:01 +0100370static int __init machine_check_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371{
Martin Schwidefskyd54853e2007-02-05 21:18:19 +0100372 ctl_set_bit(14, 25); /* enable external damage MCH */
Heiko Carstensf5daba12009-03-26 15:24:01 +0100373 ctl_set_bit(14, 27); /* enable system recovery MCH */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 ctl_set_bit(14, 24); /* enable warning MCH */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 return 0;
376}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377arch_initcall(machine_check_init);