1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *   Machine check handler
4  *
5  *    Copyright IBM Corp. 2000, 2009
6  *    Author(s): Ingo Adlung <adlung@de.ibm.com>,
7  *		 Martin Schwidefsky <schwidefsky@de.ibm.com>,
8  *		 Cornelia Huck <cornelia.huck@de.ibm.com>,
9  */
10 
11 #include <linux/kernel_stat.h>
12 #include <linux/cpufeature.h>
13 #include <linux/init.h>
14 #include <linux/errno.h>
15 #include <linux/entry-common.h>
16 #include <linux/hardirq.h>
17 #include <linux/log2.h>
18 #include <linux/kprobes.h>
19 #include <linux/kmemleak.h>
20 #include <linux/time.h>
21 #include <linux/module.h>
22 #include <linux/sched/signal.h>
23 #include <linux/kvm_host.h>
24 #include <linux/export.h>
25 #include <asm/lowcore.h>
26 #include <asm/ctlreg.h>
27 #include <asm/fpu.h>
28 #include <asm/smp.h>
29 #include <asm/stp.h>
30 #include <asm/cputime.h>
31 #include <asm/nmi.h>
32 #include <asm/crw.h>
33 #include <asm/asm-offsets.h>
34 #include <asm/pai.h>
35 #include <asm/vtime.h>
36 
37 struct mcck_struct {
38 	unsigned int kill_task : 1;
39 	unsigned int channel_report : 1;
40 	unsigned int warning : 1;
41 	unsigned int stp_queue : 1;
42 	unsigned long mcck_code;
43 };
44 
45 static DEFINE_PER_CPU(struct mcck_struct, cpu_mcck);
46 
nmi_needs_mcesa(void)47 static inline int nmi_needs_mcesa(void)
48 {
49 	return cpu_has_vx() || cpu_has_gs();
50 }
51 
52 /*
53  * The initial machine check extended save area for the boot CPU.
54  * It will be replaced on the boot CPU reinit with an allocated
55  * structure. The structure is required for machine check happening
56  * early in the boot process.
57  */
58 static struct mcesa boot_mcesa __aligned(MCESA_MAX_SIZE);
59 
nmi_alloc_mcesa_early(u64 * mcesad)60 void __init nmi_alloc_mcesa_early(u64 *mcesad)
61 {
62 	if (!nmi_needs_mcesa())
63 		return;
64 	*mcesad = __pa(&boot_mcesa);
65 	if (cpu_has_gs())
66 		*mcesad |= ilog2(MCESA_MAX_SIZE);
67 }
68 
nmi_alloc_mcesa(u64 * mcesad)69 int nmi_alloc_mcesa(u64 *mcesad)
70 {
71 	unsigned long size;
72 	void *origin;
73 
74 	*mcesad = 0;
75 	if (!nmi_needs_mcesa())
76 		return 0;
77 	size = cpu_has_gs() ? MCESA_MAX_SIZE : MCESA_MIN_SIZE;
78 	origin = kmalloc(size, GFP_KERNEL);
79 	if (!origin)
80 		return -ENOMEM;
81 	/* The pointer is stored with mcesa_bits ORed in */
82 	kmemleak_not_leak(origin);
83 	*mcesad = __pa(origin);
84 	if (cpu_has_gs())
85 		*mcesad |= ilog2(MCESA_MAX_SIZE);
86 	return 0;
87 }
88 
nmi_free_mcesa(u64 * mcesad)89 void nmi_free_mcesa(u64 *mcesad)
90 {
91 	if (!nmi_needs_mcesa())
92 		return;
93 	kfree(__va(*mcesad & MCESA_ORIGIN_MASK));
94 }
95 
nmi_puts(char * dest,const char * src)96 static __always_inline char *nmi_puts(char *dest, const char *src)
97 {
98 	while (*src)
99 		*dest++ = *src++;
100 	*dest = 0;
101 	return dest;
102 }
103 
u64_to_hex(char * dest,u64 val)104 static __always_inline char *u64_to_hex(char *dest, u64 val)
105 {
106 	int i, num;
107 
108 	for (i = 1; i <= 16; i++) {
109 		num = (val >> (64 - 4 * i)) & 0xf;
110 		if (num >= 10)
111 			*dest++ = 'A' + num - 10;
112 		else
113 			*dest++ = '0' + num;
114 	}
115 	*dest = 0;
116 	return dest;
117 }
118 
s390_handle_damage(void)119 static notrace void s390_handle_damage(void)
120 {
121 	struct lowcore *lc = get_lowcore();
122 	union ctlreg0 cr0, cr0_new;
123 	char message[100];
124 	psw_t psw_save;
125 	char *ptr;
126 
127 	smp_emergency_stop();
128 	diag_amode31_ops.diag308_reset();
129 	ptr = nmi_puts(message, "System stopped due to unrecoverable machine check, code: 0x");
130 	u64_to_hex(ptr, lc->mcck_interruption_code);
131 
132 	/*
133 	 * Disable low address protection and make machine check new PSW a
134 	 * disabled wait PSW. Any additional machine check cannot be handled.
135 	 */
136 	local_ctl_store(0, &cr0.reg);
137 	cr0_new = cr0;
138 	cr0_new.lap = 0;
139 	local_ctl_load(0, &cr0_new.reg);
140 	psw_save = lc->mcck_new_psw;
141 	psw_bits(lc->mcck_new_psw).io = 0;
142 	psw_bits(lc->mcck_new_psw).ext = 0;
143 	psw_bits(lc->mcck_new_psw).wait = 1;
144 	sclp_emergency_printk(message);
145 
146 	/*
147 	 * Restore machine check new PSW and control register 0 to original
148 	 * values. This makes possible system dump analysis easier.
149 	 */
150 	lc->mcck_new_psw = psw_save;
151 	local_ctl_load(0, &cr0.reg);
152 	disabled_wait();
153 	while (1);
154 }
155 NOKPROBE_SYMBOL(s390_handle_damage);
156 
157 /*
158  * Main machine check handler function. Will be called with interrupts disabled
159  * and machine checks enabled.
160  */
s390_handle_mcck(void)161 void s390_handle_mcck(void)
162 {
163 	struct mcck_struct mcck;
164 	unsigned long mflags;
165 
166 	/*
167 	 * Disable machine checks and get the current state of accumulated
168 	 * machine checks. Afterwards delete the old state and enable machine
169 	 * checks again.
170 	 */
171 	local_mcck_save(mflags);
172 	mcck = *this_cpu_ptr(&cpu_mcck);
173 	memset(this_cpu_ptr(&cpu_mcck), 0, sizeof(mcck));
174 	local_mcck_restore(mflags);
175 
176 	if (mcck.channel_report)
177 		crw_handle_channel_report();
178 	/*
179 	 * A warning may remain for a prolonged period on the bare iron.
180 	 * (actually until the machine is powered off, or the problem is gone)
181 	 * So we just stop listening for the WARNING MCH and avoid continuously
182 	 * being interrupted.  One caveat is however, that we must do this per
183 	 * processor and cannot use the smp version of ctl_clear_bit().
184 	 * On VM we only get one interrupt per virtally presented machinecheck.
185 	 * Though one suffices, we may get one interrupt per (virtual) cpu.
186 	 */
187 	if (mcck.warning) {	/* WARNING pending ? */
188 		static int mchchk_wng_posted = 0;
189 
190 		/* Use single cpu clear, as we cannot handle smp here. */
191 		local_ctl_clear_bit(14, CR14_WARNING_SUBMASK_BIT);
192 		if (xchg(&mchchk_wng_posted, 1) == 0)
193 			kill_cad_pid(SIGPWR, 1);
194 	}
195 	if (mcck.stp_queue)
196 		stp_queue_work();
197 	if (mcck.kill_task) {
198 		printk(KERN_EMERG "mcck: Terminating task because of machine "
199 		       "malfunction (code 0x%016lx).\n", mcck.mcck_code);
200 		printk(KERN_EMERG "mcck: task: %s, pid: %d.\n",
201 		       current->comm, current->pid);
202 		if (is_global_init(current))
203 			panic("mcck: Attempting to kill init!\n");
204 		do_send_sig_info(SIGKILL, SEND_SIG_PRIV, current, PIDTYPE_PID);
205 	}
206 }
207 
208 /**
209  * nmi_registers_valid - verify if registers are valid
210  * @mci: machine check interruption code
211  *
212  * Inspect a machine check interruption code and verify if all required
213  * registers are valid. For some registers the corresponding validity bit is
214  * ignored and the registers are set to the expected value.
215  * Returns true if all registers are valid, otherwise false.
216  */
nmi_registers_valid(union mci mci)217 static bool notrace nmi_registers_valid(union mci mci)
218 {
219 	union ctlreg2 cr2;
220 
221 	/*
222 	 * The getcpu vdso syscall reads the CPU number from the programmable
223 	 * field of the TOD clock. Disregard the TOD programmable register
224 	 * validity bit and load the CPU number into the TOD programmable field
225 	 * unconditionally.
226 	 */
227 	set_tod_programmable_field(raw_smp_processor_id());
228 	/*
229 	 * Set the clock comparator register to the next expected value.
230 	 */
231 	set_clock_comparator(get_lowcore()->clock_comparator);
232 	if (!mci.gr || !mci.fp || !mci.fc)
233 		return false;
234 	/*
235 	 * The vector validity must only be checked if not running a
236 	 * KVM guest. For KVM guests the machine check is forwarded by
237 	 * KVM and it is the responsibility of the guest to take
238 	 * appropriate actions. The host vector or FPU values have been
239 	 * saved by KVM and will be restored by KVM.
240 	 */
241 	if (!mci.vr && !test_cpu_flag(CIF_MCCK_GUEST))
242 		return false;
243 	if (!mci.ar)
244 		return false;
245 	/*
246 	 * Two cases for guarded storage registers:
247 	 * - machine check in kernel or userspace
248 	 * - machine check while running SIE (KVM guest)
249 	 * For kernel or userspace the userspace values of guarded storage
250 	 * control can not be recreated, the process must be terminated.
251 	 * For SIE the guest values of guarded storage can not be recreated.
252 	 * This is either due to a bug or due to GS being disabled in the
253 	 * guest. The guest will be notified by KVM code and the guests machine
254 	 * check handling must take care of this. The host values are saved by
255 	 * KVM and are not affected.
256 	 */
257 	cr2.reg = get_lowcore()->cregs_save_area[2];
258 	if (cr2.gse && !mci.gs && !test_cpu_flag(CIF_MCCK_GUEST))
259 		return false;
260 	if (!mci.ms || !mci.pm || !mci.ia)
261 		return false;
262 	return true;
263 }
264 NOKPROBE_SYMBOL(nmi_registers_valid);
265 
266 /*
267  * Backup the guest's machine check info to its description block
268  */
s390_backup_mcck_info(struct pt_regs * regs)269 static void notrace s390_backup_mcck_info(struct pt_regs *regs)
270 {
271 	struct mcck_volatile_info *mcck_backup;
272 	struct sie_page *sie_page;
273 
274 	/* r14 contains the sie block, which was set in sie64a */
275 	struct kvm_s390_sie_block *sie_block = phys_to_virt(regs->gprs[14]);
276 
277 	if (sie_block == NULL)
278 		/* Something's seriously wrong, stop system. */
279 		s390_handle_damage();
280 
281 	sie_page = container_of(sie_block, struct sie_page, sie_block);
282 	mcck_backup = &sie_page->mcck_info;
283 	mcck_backup->mcic = get_lowcore()->mcck_interruption_code &
284 				~(MCCK_CODE_CP | MCCK_CODE_EXT_DAMAGE);
285 	mcck_backup->ext_damage_code = get_lowcore()->external_damage_code;
286 	mcck_backup->failing_storage_address = get_lowcore()->failing_storage_address;
287 }
288 NOKPROBE_SYMBOL(s390_backup_mcck_info);
289 
290 #define MAX_IPD_COUNT	29
291 #define MAX_IPD_TIME	(5 * 60 * USEC_PER_SEC) /* 5 minutes */
292 
293 #define ED_STP_ISLAND	6	/* External damage STP island check */
294 #define ED_STP_SYNC	7	/* External damage STP sync check */
295 
296 #define MCCK_CODE_NO_GUEST	(MCCK_CODE_CP | MCCK_CODE_EXT_DAMAGE)
297 
298 /*
299  * machine check handler.
300  */
s390_do_machine_check(struct pt_regs * regs)301 void notrace s390_do_machine_check(struct pt_regs *regs)
302 {
303 	static int ipd_count;
304 	static DEFINE_SPINLOCK(ipd_lock);
305 	static unsigned long long last_ipd;
306 	struct lowcore *lc = get_lowcore();
307 	struct mcck_struct *mcck;
308 	unsigned long long tmp;
309 	irqentry_state_t irq_state;
310 	union mci mci;
311 	unsigned long mcck_dam_code;
312 	int mcck_pending = 0;
313 
314 	irq_state = irqentry_nmi_enter(regs);
315 
316 	if (user_mode(regs))
317 		update_timer_mcck();
318 	inc_irq_stat(NMI_NMI);
319 	mci.val = lc->mcck_interruption_code;
320 	mcck = this_cpu_ptr(&cpu_mcck);
321 
322 	/*
323 	 * Reinject the instruction processing damages' machine checks
324 	 * including Delayed Access Exception into the guest
325 	 * instead of damaging the host if they happen in the guest.
326 	 */
327 	if (mci.pd && !test_cpu_flag(CIF_MCCK_GUEST)) {
328 		if (mci.b) {
329 			/* Processing backup -> verify if we can survive this */
330 			u64 z_mcic, o_mcic, t_mcic;
331 			z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<29);
332 			o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
333 				  1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
334 				  1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 |
335 				  1ULL<<16);
336 			t_mcic = mci.val;
337 
338 			if (((t_mcic & z_mcic) != 0) ||
339 			    ((t_mcic & o_mcic) != o_mcic)) {
340 				s390_handle_damage();
341 			}
342 
343 			/*
344 			 * Nullifying exigent condition, therefore we might
345 			 * retry this instruction.
346 			 */
347 			spin_lock(&ipd_lock);
348 			tmp = get_tod_clock();
349 			if (((tmp - last_ipd) >> 12) < MAX_IPD_TIME)
350 				ipd_count++;
351 			else
352 				ipd_count = 1;
353 			last_ipd = tmp;
354 			if (ipd_count == MAX_IPD_COUNT)
355 				s390_handle_damage();
356 			spin_unlock(&ipd_lock);
357 		} else {
358 			/* Processing damage -> stopping machine */
359 			s390_handle_damage();
360 		}
361 	}
362 	if (!nmi_registers_valid(mci)) {
363 		if (!user_mode(regs))
364 			s390_handle_damage();
365 		/*
366 		 * Couldn't restore all register contents for the
367 		 * user space process -> mark task for termination.
368 		 */
369 		mcck->kill_task = 1;
370 		mcck->mcck_code = mci.val;
371 		mcck_pending = 1;
372 	}
373 
374 	/*
375 	 * Backup the machine check's info if it happens when the guest
376 	 * is running.
377 	 */
378 	if (test_cpu_flag(CIF_MCCK_GUEST))
379 		s390_backup_mcck_info(regs);
380 
381 	if (mci.cd) {
382 		/* Timing facility damage */
383 		s390_handle_damage();
384 	}
385 	if (mci.ed && mci.ec) {
386 		/* External damage */
387 		if (lc->external_damage_code & (1U << ED_STP_SYNC))
388 			mcck->stp_queue |= stp_sync_check();
389 		if (lc->external_damage_code & (1U << ED_STP_ISLAND))
390 			mcck->stp_queue |= stp_island_check();
391 		mcck_pending = 1;
392 	}
393 	/*
394 	 * Reinject storage related machine checks into the guest if they
395 	 * happen when the guest is running.
396 	 */
397 	if (!test_cpu_flag(CIF_MCCK_GUEST)) {
398 		/* Storage error uncorrected */
399 		if (mci.se)
400 			s390_handle_damage();
401 		/* Storage key-error uncorrected */
402 		if (mci.ke)
403 			s390_handle_damage();
404 		/* Storage degradation */
405 		if (mci.ds && mci.fa)
406 			s390_handle_damage();
407 	}
408 	if (mci.cp) {
409 		/* Channel report word pending */
410 		mcck->channel_report = 1;
411 		mcck_pending = 1;
412 	}
413 	if (mci.w) {
414 		/* Warning pending */
415 		mcck->warning = 1;
416 		mcck_pending = 1;
417 	}
418 
419 	/*
420 	 * If there are only Channel Report Pending and External Damage
421 	 * machine checks, they will not be reinjected into the guest
422 	 * because they refer to host conditions only.
423 	 */
424 	mcck_dam_code = (mci.val & MCIC_SUBCLASS_MASK);
425 	if (test_cpu_flag(CIF_MCCK_GUEST) &&
426 	(mcck_dam_code & MCCK_CODE_NO_GUEST) != mcck_dam_code) {
427 		/* Set exit reason code for host's later handling */
428 		*((long *)(regs->gprs[15] + __SF_SIE_REASON)) = -EINTR;
429 	}
430 	clear_cpu_flag(CIF_MCCK_GUEST);
431 
432 	if (mcck_pending)
433 		schedule_mcck_handler();
434 
435 	irqentry_nmi_exit(regs, irq_state);
436 }
437 NOKPROBE_SYMBOL(s390_do_machine_check);
438 
machine_check_init(void)439 static int __init machine_check_init(void)
440 {
441 	system_ctl_set_bit(14, CR14_EXTERNAL_DAMAGE_SUBMASK_BIT);
442 	system_ctl_set_bit(14, CR14_RECOVERY_SUBMASK_BIT);
443 	system_ctl_set_bit(14, CR14_WARNING_SUBMASK_BIT);
444 	return 0;
445 }
446 early_initcall(machine_check_init);
447