xref: /openbmc/linux/arch/s390/kernel/nmi.c (revision bf3608f3)
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  *		 Heiko Carstens <heiko.carstens@de.ibm.com>,
10  */
11 
12 #include <linux/kernel_stat.h>
13 #include <linux/init.h>
14 #include <linux/errno.h>
15 #include <linux/hardirq.h>
16 #include <linux/log2.h>
17 #include <linux/kprobes.h>
18 #include <linux/kmemleak.h>
19 #include <linux/time.h>
20 #include <linux/module.h>
21 #include <linux/sched/signal.h>
22 
23 #include <linux/export.h>
24 #include <asm/lowcore.h>
25 #include <asm/smp.h>
26 #include <asm/stp.h>
27 #include <asm/cputime.h>
28 #include <asm/nmi.h>
29 #include <asm/crw.h>
30 #include <asm/switch_to.h>
31 #include <asm/ctl_reg.h>
32 #include <asm/asm-offsets.h>
33 #include <linux/kvm_host.h>
34 
35 struct mcck_struct {
36 	unsigned int kill_task : 1;
37 	unsigned int channel_report : 1;
38 	unsigned int warning : 1;
39 	unsigned int stp_queue : 1;
40 	unsigned long mcck_code;
41 };
42 
43 static DEFINE_PER_CPU(struct mcck_struct, cpu_mcck);
44 static struct kmem_cache *mcesa_cache;
45 static unsigned long mcesa_origin_lc;
46 
47 static inline int nmi_needs_mcesa(void)
48 {
49 	return MACHINE_HAS_VX || MACHINE_HAS_GS;
50 }
51 
52 static inline unsigned long nmi_get_mcesa_size(void)
53 {
54 	if (MACHINE_HAS_GS)
55 		return MCESA_MAX_SIZE;
56 	return MCESA_MIN_SIZE;
57 }
58 
59 /*
60  * The initial machine check extended save area for the boot CPU.
61  * It will be replaced on the boot CPU reinit with an allocated
62  * structure. The structure is required for machine check happening
63  * early in the boot process.
64  */
65 static struct mcesa boot_mcesa __initdata __aligned(MCESA_MAX_SIZE);
66 
67 void __init nmi_alloc_mcesa_early(u64 *mcesad)
68 {
69 	if (!nmi_needs_mcesa())
70 		return;
71 	*mcesad = __pa(&boot_mcesa);
72 	if (MACHINE_HAS_GS)
73 		*mcesad |= ilog2(MCESA_MAX_SIZE);
74 }
75 
76 static void __init nmi_alloc_cache(void)
77 {
78 	unsigned long size;
79 
80 	if (!nmi_needs_mcesa())
81 		return;
82 	size = nmi_get_mcesa_size();
83 	if (size > MCESA_MIN_SIZE)
84 		mcesa_origin_lc = ilog2(size);
85 	/* create slab cache for the machine-check-extended-save-areas */
86 	mcesa_cache = kmem_cache_create("nmi_save_areas", size, size, 0, NULL);
87 	if (!mcesa_cache)
88 		panic("Couldn't create nmi save area cache");
89 }
90 
91 int __ref nmi_alloc_mcesa(u64 *mcesad)
92 {
93 	unsigned long origin;
94 
95 	*mcesad = 0;
96 	if (!nmi_needs_mcesa())
97 		return 0;
98 	if (!mcesa_cache)
99 		nmi_alloc_cache();
100 	origin = (unsigned long) kmem_cache_alloc(mcesa_cache, GFP_KERNEL);
101 	if (!origin)
102 		return -ENOMEM;
103 	/* The pointer is stored with mcesa_bits ORed in */
104 	kmemleak_not_leak((void *) origin);
105 	*mcesad = __pa(origin) | mcesa_origin_lc;
106 	return 0;
107 }
108 
109 void nmi_free_mcesa(u64 *mcesad)
110 {
111 	if (!nmi_needs_mcesa())
112 		return;
113 	kmem_cache_free(mcesa_cache, __va(*mcesad & MCESA_ORIGIN_MASK));
114 }
115 
116 static notrace void s390_handle_damage(void)
117 {
118 	smp_emergency_stop();
119 	disabled_wait();
120 	while (1);
121 }
122 NOKPROBE_SYMBOL(s390_handle_damage);
123 
124 /*
125  * Main machine check handler function. Will be called with interrupts disabled
126  * and machine checks enabled.
127  */
128 void __s390_handle_mcck(void)
129 {
130 	struct mcck_struct mcck;
131 
132 	/*
133 	 * Disable machine checks and get the current state of accumulated
134 	 * machine checks. Afterwards delete the old state and enable machine
135 	 * checks again.
136 	 */
137 	local_mcck_disable();
138 	mcck = *this_cpu_ptr(&cpu_mcck);
139 	memset(this_cpu_ptr(&cpu_mcck), 0, sizeof(mcck));
140 	local_mcck_enable();
141 
142 	if (mcck.channel_report)
143 		crw_handle_channel_report();
144 	/*
145 	 * A warning may remain for a prolonged period on the bare iron.
146 	 * (actually until the machine is powered off, or the problem is gone)
147 	 * So we just stop listening for the WARNING MCH and avoid continuously
148 	 * being interrupted.  One caveat is however, that we must do this per
149 	 * processor and cannot use the smp version of ctl_clear_bit().
150 	 * On VM we only get one interrupt per virtally presented machinecheck.
151 	 * Though one suffices, we may get one interrupt per (virtual) cpu.
152 	 */
153 	if (mcck.warning) {	/* WARNING pending ? */
154 		static int mchchk_wng_posted = 0;
155 
156 		/* Use single cpu clear, as we cannot handle smp here. */
157 		__ctl_clear_bit(14, 24);	/* Disable WARNING MCH */
158 		if (xchg(&mchchk_wng_posted, 1) == 0)
159 			kill_cad_pid(SIGPWR, 1);
160 	}
161 	if (mcck.stp_queue)
162 		stp_queue_work();
163 	if (mcck.kill_task) {
164 		local_irq_enable();
165 		printk(KERN_EMERG "mcck: Terminating task because of machine "
166 		       "malfunction (code 0x%016lx).\n", mcck.mcck_code);
167 		printk(KERN_EMERG "mcck: task: %s, pid: %d.\n",
168 		       current->comm, current->pid);
169 		make_task_dead(SIGSEGV);
170 	}
171 }
172 
173 void noinstr s390_handle_mcck(void)
174 {
175 	trace_hardirqs_off();
176 	__s390_handle_mcck();
177 	trace_hardirqs_on();
178 }
179 /*
180  * returns 0 if all required registers are available
181  * returns 1 otherwise
182  */
183 static int notrace s390_validate_registers(union mci mci, int umode)
184 {
185 	struct mcesa *mcesa;
186 	void *fpt_save_area;
187 	union ctlreg2 cr2;
188 	int kill_task;
189 	u64 zero;
190 
191 	kill_task = 0;
192 	zero = 0;
193 
194 	if (!mci.gr) {
195 		/*
196 		 * General purpose registers couldn't be restored and have
197 		 * unknown contents. Stop system or terminate process.
198 		 */
199 		if (!umode)
200 			s390_handle_damage();
201 		kill_task = 1;
202 	}
203 	if (!mci.fp) {
204 		/*
205 		 * Floating point registers can't be restored. If the
206 		 * kernel currently uses floating point registers the
207 		 * system is stopped. If the process has its floating
208 		 * pointer registers loaded it is terminated.
209 		 */
210 		if (S390_lowcore.fpu_flags & KERNEL_VXR_V0V7)
211 			s390_handle_damage();
212 		if (!test_cpu_flag(CIF_FPU))
213 			kill_task = 1;
214 	}
215 	fpt_save_area = &S390_lowcore.floating_pt_save_area;
216 	if (!mci.fc) {
217 		/*
218 		 * Floating point control register can't be restored.
219 		 * If the kernel currently uses the floating pointer
220 		 * registers and needs the FPC register the system is
221 		 * stopped. If the process has its floating pointer
222 		 * registers loaded it is terminated. Otherwise the
223 		 * FPC is just validated.
224 		 */
225 		if (S390_lowcore.fpu_flags & KERNEL_FPC)
226 			s390_handle_damage();
227 		asm volatile(
228 			"	lfpc	%0\n"
229 			:
230 			: "Q" (zero));
231 		if (!test_cpu_flag(CIF_FPU))
232 			kill_task = 1;
233 	} else {
234 		asm volatile(
235 			"	lfpc	%0\n"
236 			:
237 			: "Q" (S390_lowcore.fpt_creg_save_area));
238 	}
239 
240 	mcesa = __va(S390_lowcore.mcesad & MCESA_ORIGIN_MASK);
241 	if (!MACHINE_HAS_VX) {
242 		/* Validate floating point registers */
243 		asm volatile(
244 			"	ld	0,0(%0)\n"
245 			"	ld	1,8(%0)\n"
246 			"	ld	2,16(%0)\n"
247 			"	ld	3,24(%0)\n"
248 			"	ld	4,32(%0)\n"
249 			"	ld	5,40(%0)\n"
250 			"	ld	6,48(%0)\n"
251 			"	ld	7,56(%0)\n"
252 			"	ld	8,64(%0)\n"
253 			"	ld	9,72(%0)\n"
254 			"	ld	10,80(%0)\n"
255 			"	ld	11,88(%0)\n"
256 			"	ld	12,96(%0)\n"
257 			"	ld	13,104(%0)\n"
258 			"	ld	14,112(%0)\n"
259 			"	ld	15,120(%0)\n"
260 			:
261 			: "a" (fpt_save_area)
262 			: "memory");
263 	} else {
264 		/* Validate vector registers */
265 		union ctlreg0 cr0;
266 
267 		/*
268 		 * The vector validity must only be checked if not running a
269 		 * KVM guest. For KVM guests the machine check is forwarded by
270 		 * KVM and it is the responsibility of the guest to take
271 		 * appropriate actions. The host vector or FPU values have been
272 		 * saved by KVM and will be restored by KVM.
273 		 */
274 		if (!mci.vr && !test_cpu_flag(CIF_MCCK_GUEST)) {
275 			/*
276 			 * Vector registers can't be restored. If the kernel
277 			 * currently uses vector registers the system is
278 			 * stopped. If the process has its vector registers
279 			 * loaded it is terminated. Otherwise just validate
280 			 * the registers.
281 			 */
282 			if (S390_lowcore.fpu_flags & KERNEL_VXR)
283 				s390_handle_damage();
284 			if (!test_cpu_flag(CIF_FPU))
285 				kill_task = 1;
286 		}
287 		cr0.val = S390_lowcore.cregs_save_area[0];
288 		cr0.afp = cr0.vx = 1;
289 		__ctl_load(cr0.val, 0, 0);
290 		asm volatile(
291 			"	la	1,%0\n"
292 			"	.word	0xe70f,0x1000,0x0036\n" /* vlm 0,15,0(1) */
293 			"	.word	0xe70f,0x1100,0x0c36\n" /* vlm 16,31,256(1) */
294 			:
295 			: "Q" (*(struct vx_array *)mcesa->vector_save_area)
296 			: "1");
297 		__ctl_load(S390_lowcore.cregs_save_area[0], 0, 0);
298 	}
299 	/* Validate access registers */
300 	asm volatile(
301 		"	lam	0,15,0(%0)\n"
302 		:
303 		: "a" (&S390_lowcore.access_regs_save_area)
304 		: "memory");
305 	if (!mci.ar) {
306 		/*
307 		 * Access registers have unknown contents.
308 		 * Terminating task.
309 		 */
310 		kill_task = 1;
311 	}
312 	/* Validate guarded storage registers */
313 	cr2.val = S390_lowcore.cregs_save_area[2];
314 	if (cr2.gse) {
315 		if (!mci.gs) {
316 			/*
317 			 * 2 cases:
318 			 * - machine check in kernel or userspace
319 			 * - machine check while running SIE (KVM guest)
320 			 * For kernel or userspace the userspace values of
321 			 * guarded storage control can not be recreated, the
322 			 * process must be terminated.
323 			 * For SIE the guest values of guarded storage can not
324 			 * be recreated. This is either due to a bug or due to
325 			 * GS being disabled in the guest. The guest will be
326 			 * notified by KVM code and the guests machine check
327 			 * handling must take care of this.  The host values
328 			 * are saved by KVM and are not affected.
329 			 */
330 			if (!test_cpu_flag(CIF_MCCK_GUEST))
331 				kill_task = 1;
332 		} else {
333 			load_gs_cb((struct gs_cb *)mcesa->guarded_storage_save_area);
334 		}
335 	}
336 	/*
337 	 * The getcpu vdso syscall reads CPU number from the programmable
338 	 * field of the TOD clock. Disregard the TOD programmable register
339 	 * validity bit and load the CPU number into the TOD programmable
340 	 * field unconditionally.
341 	 */
342 	set_tod_programmable_field(raw_smp_processor_id());
343 	/* Validate clock comparator register */
344 	set_clock_comparator(S390_lowcore.clock_comparator);
345 
346 	if (!mci.ms || !mci.pm || !mci.ia)
347 		kill_task = 1;
348 
349 	return kill_task;
350 }
351 NOKPROBE_SYMBOL(s390_validate_registers);
352 
353 /*
354  * Backup the guest's machine check info to its description block
355  */
356 static void notrace s390_backup_mcck_info(struct pt_regs *regs)
357 {
358 	struct mcck_volatile_info *mcck_backup;
359 	struct sie_page *sie_page;
360 
361 	/* r14 contains the sie block, which was set in sie64a */
362 	struct kvm_s390_sie_block *sie_block =
363 			(struct kvm_s390_sie_block *) regs->gprs[14];
364 
365 	if (sie_block == NULL)
366 		/* Something's seriously wrong, stop system. */
367 		s390_handle_damage();
368 
369 	sie_page = container_of(sie_block, struct sie_page, sie_block);
370 	mcck_backup = &sie_page->mcck_info;
371 	mcck_backup->mcic = S390_lowcore.mcck_interruption_code &
372 				~(MCCK_CODE_CP | MCCK_CODE_EXT_DAMAGE);
373 	mcck_backup->ext_damage_code = S390_lowcore.external_damage_code;
374 	mcck_backup->failing_storage_address
375 			= S390_lowcore.failing_storage_address;
376 }
377 NOKPROBE_SYMBOL(s390_backup_mcck_info);
378 
379 #define MAX_IPD_COUNT	29
380 #define MAX_IPD_TIME	(5 * 60 * USEC_PER_SEC) /* 5 minutes */
381 
382 #define ED_STP_ISLAND	6	/* External damage STP island check */
383 #define ED_STP_SYNC	7	/* External damage STP sync check */
384 
385 #define MCCK_CODE_NO_GUEST	(MCCK_CODE_CP | MCCK_CODE_EXT_DAMAGE)
386 
387 /*
388  * machine check handler.
389  */
390 int notrace s390_do_machine_check(struct pt_regs *regs)
391 {
392 	static int ipd_count;
393 	static DEFINE_SPINLOCK(ipd_lock);
394 	static unsigned long long last_ipd;
395 	struct mcck_struct *mcck;
396 	unsigned long long tmp;
397 	union mci mci;
398 	unsigned long mcck_dam_code;
399 	int mcck_pending = 0;
400 
401 	nmi_enter();
402 
403 	if (user_mode(regs))
404 		update_timer_mcck();
405 	inc_irq_stat(NMI_NMI);
406 	mci.val = S390_lowcore.mcck_interruption_code;
407 	mcck = this_cpu_ptr(&cpu_mcck);
408 
409 	/*
410 	 * Reinject the instruction processing damages' machine checks
411 	 * including Delayed Access Exception into the guest
412 	 * instead of damaging the host if they happen in the guest.
413 	 */
414 	if (mci.pd && !test_cpu_flag(CIF_MCCK_GUEST)) {
415 		if (mci.b) {
416 			/* Processing backup -> verify if we can survive this */
417 			u64 z_mcic, o_mcic, t_mcic;
418 			z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<29);
419 			o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
420 				  1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
421 				  1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 |
422 				  1ULL<<16);
423 			t_mcic = mci.val;
424 
425 			if (((t_mcic & z_mcic) != 0) ||
426 			    ((t_mcic & o_mcic) != o_mcic)) {
427 				s390_handle_damage();
428 			}
429 
430 			/*
431 			 * Nullifying exigent condition, therefore we might
432 			 * retry this instruction.
433 			 */
434 			spin_lock(&ipd_lock);
435 			tmp = get_tod_clock();
436 			if (((tmp - last_ipd) >> 12) < MAX_IPD_TIME)
437 				ipd_count++;
438 			else
439 				ipd_count = 1;
440 			last_ipd = tmp;
441 			if (ipd_count == MAX_IPD_COUNT)
442 				s390_handle_damage();
443 			spin_unlock(&ipd_lock);
444 		} else {
445 			/* Processing damage -> stopping machine */
446 			s390_handle_damage();
447 		}
448 	}
449 	if (s390_validate_registers(mci, user_mode(regs))) {
450 		/*
451 		 * Couldn't restore all register contents for the
452 		 * user space process -> mark task for termination.
453 		 */
454 		mcck->kill_task = 1;
455 		mcck->mcck_code = mci.val;
456 		mcck_pending = 1;
457 	}
458 
459 	/*
460 	 * Backup the machine check's info if it happens when the guest
461 	 * is running.
462 	 */
463 	if (test_cpu_flag(CIF_MCCK_GUEST))
464 		s390_backup_mcck_info(regs);
465 
466 	if (mci.cd) {
467 		/* Timing facility damage */
468 		s390_handle_damage();
469 	}
470 	if (mci.ed && mci.ec) {
471 		/* External damage */
472 		if (S390_lowcore.external_damage_code & (1U << ED_STP_SYNC))
473 			mcck->stp_queue |= stp_sync_check();
474 		if (S390_lowcore.external_damage_code & (1U << ED_STP_ISLAND))
475 			mcck->stp_queue |= stp_island_check();
476 		mcck_pending = 1;
477 	}
478 
479 	if (mci.cp) {
480 		/* Channel report word pending */
481 		mcck->channel_report = 1;
482 		mcck_pending = 1;
483 	}
484 	if (mci.w) {
485 		/* Warning pending */
486 		mcck->warning = 1;
487 		mcck_pending = 1;
488 	}
489 
490 	/*
491 	 * If there are only Channel Report Pending and External Damage
492 	 * machine checks, they will not be reinjected into the guest
493 	 * because they refer to host conditions only.
494 	 */
495 	mcck_dam_code = (mci.val & MCIC_SUBCLASS_MASK);
496 	if (test_cpu_flag(CIF_MCCK_GUEST) &&
497 	(mcck_dam_code & MCCK_CODE_NO_GUEST) != mcck_dam_code) {
498 		/* Set exit reason code for host's later handling */
499 		*((long *)(regs->gprs[15] + __SF_SIE_REASON)) = -EINTR;
500 	}
501 	clear_cpu_flag(CIF_MCCK_GUEST);
502 
503 	if (user_mode(regs) && mcck_pending) {
504 		nmi_exit();
505 		return 1;
506 	}
507 
508 	if (mcck_pending)
509 		schedule_mcck_handler();
510 
511 	nmi_exit();
512 	return 0;
513 }
514 NOKPROBE_SYMBOL(s390_do_machine_check);
515 
516 static int __init machine_check_init(void)
517 {
518 	ctl_set_bit(14, 25);	/* enable external damage MCH */
519 	ctl_set_bit(14, 27);	/* enable system recovery MCH */
520 	ctl_set_bit(14, 24);	/* enable warning MCH */
521 	return 0;
522 }
523 early_initcall(machine_check_init);
524