Linux Audio

Check our new training course

Loading...
v5.4
  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
 35struct 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
 43static DEFINE_PER_CPU(struct mcck_struct, cpu_mcck);
 44static struct kmem_cache *mcesa_cache;
 45static unsigned long mcesa_origin_lc;
 46
 47static inline int nmi_needs_mcesa(void)
 48{
 49	return MACHINE_HAS_VX || MACHINE_HAS_GS;
 50}
 51
 52static 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 by nmi_init() with an allocated structure.
 62 * The structure is required for machine check happening early in
 63 * the boot process.
 64 */
 65static struct mcesa boot_mcesa __initdata __aligned(MCESA_MAX_SIZE);
 66
 67void __init nmi_alloc_boot_cpu(struct lowcore *lc)
 68{
 69	if (!nmi_needs_mcesa())
 70		return;
 71	lc->mcesad = (unsigned long) &boot_mcesa;
 72	if (MACHINE_HAS_GS)
 73		lc->mcesad |= ilog2(MCESA_MAX_SIZE);
 74}
 75
 76static int __init nmi_init(void)
 77{
 78	unsigned long origin, cr0, size;
 79
 80	if (!nmi_needs_mcesa())
 81		return 0;
 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	origin = (unsigned long) kmem_cache_alloc(mcesa_cache, GFP_KERNEL);
 90	if (!origin)
 91		panic("Couldn't allocate nmi save area");
 92	/* The pointer is stored with mcesa_bits ORed in */
 93	kmemleak_not_leak((void *) origin);
 94	__ctl_store(cr0, 0, 0);
 95	__ctl_clear_bit(0, 28); /* disable lowcore protection */
 96	/* Replace boot_mcesa on the boot CPU */
 97	S390_lowcore.mcesad = origin | mcesa_origin_lc;
 98	__ctl_load(cr0, 0, 0);
 99	return 0;
100}
101early_initcall(nmi_init);
102
103int nmi_alloc_per_cpu(struct lowcore *lc)
104{
105	unsigned long origin;
106
107	if (!nmi_needs_mcesa())
108		return 0;
109	origin = (unsigned long) kmem_cache_alloc(mcesa_cache, GFP_KERNEL);
110	if (!origin)
111		return -ENOMEM;
112	/* The pointer is stored with mcesa_bits ORed in */
113	kmemleak_not_leak((void *) origin);
114	lc->mcesad = origin | mcesa_origin_lc;
115	return 0;
116}
117
118void nmi_free_per_cpu(struct lowcore *lc)
119{
120	if (!nmi_needs_mcesa())
121		return;
122	kmem_cache_free(mcesa_cache, (void *)(lc->mcesad & MCESA_ORIGIN_MASK));
123}
124
125static notrace void s390_handle_damage(void)
126{
127	smp_emergency_stop();
128	disabled_wait();
129	while (1);
130}
131NOKPROBE_SYMBOL(s390_handle_damage);
132
133/*
134 * Main machine check handler function. Will be called with interrupts enabled
135 * or disabled and machine checks enabled or disabled.
136 */
137void s390_handle_mcck(void)
138{
139	unsigned long flags;
140	struct mcck_struct mcck;
141
142	/*
143	 * Disable machine checks and get the current state of accumulated
144	 * machine checks. Afterwards delete the old state and enable machine
145	 * checks again.
146	 */
147	local_irq_save(flags);
148	local_mcck_disable();
149	mcck = *this_cpu_ptr(&cpu_mcck);
150	memset(this_cpu_ptr(&cpu_mcck), 0, sizeof(mcck));
151	clear_cpu_flag(CIF_MCCK_PENDING);
152	local_mcck_enable();
153	local_irq_restore(flags);
154
155	if (mcck.channel_report)
156		crw_handle_channel_report();
157	/*
158	 * A warning may remain for a prolonged period on the bare iron.
159	 * (actually until the machine is powered off, or the problem is gone)
160	 * So we just stop listening for the WARNING MCH and avoid continuously
161	 * being interrupted.  One caveat is however, that we must do this per
162	 * processor and cannot use the smp version of ctl_clear_bit().
163	 * On VM we only get one interrupt per virtally presented machinecheck.
164	 * Though one suffices, we may get one interrupt per (virtual) cpu.
165	 */
166	if (mcck.warning) {	/* WARNING pending ? */
167		static int mchchk_wng_posted = 0;
168
169		/* Use single cpu clear, as we cannot handle smp here. */
170		__ctl_clear_bit(14, 24);	/* Disable WARNING MCH */
171		if (xchg(&mchchk_wng_posted, 1) == 0)
172			kill_cad_pid(SIGPWR, 1);
173	}
174	if (mcck.stp_queue)
175		stp_queue_work();
176	if (mcck.kill_task) {
177		local_irq_enable();
178		printk(KERN_EMERG "mcck: Terminating task because of machine "
179		       "malfunction (code 0x%016lx).\n", mcck.mcck_code);
180		printk(KERN_EMERG "mcck: task: %s, pid: %d.\n",
181		       current->comm, current->pid);
182		do_exit(SIGSEGV);
183	}
184}
185EXPORT_SYMBOL_GPL(s390_handle_mcck);
186
187/*
188 * returns 0 if all required registers are available
189 * returns 1 otherwise
190 */
191static int notrace s390_check_registers(union mci mci, int umode)
192{
193	union ctlreg2 cr2;
194	int kill_task;
 
 
195
196	kill_task = 0;
 
197
198	if (!mci.gr) {
199		/*
200		 * General purpose registers couldn't be restored and have
201		 * unknown contents. Stop system or terminate process.
202		 */
203		if (!umode)
204			s390_handle_damage();
205		kill_task = 1;
206	}
207	/* Check control registers */
208	if (!mci.cr) {
209		/*
210		 * Control registers have unknown contents.
211		 * Can't recover and therefore stopping machine.
212		 */
213		s390_handle_damage();
214	}
215	if (!mci.fp) {
216		/*
217		 * Floating point registers can't be restored. If the
218		 * kernel currently uses floating point registers the
219		 * system is stopped. If the process has its floating
220		 * pointer registers loaded it is terminated.
221		 */
222		if (S390_lowcore.fpu_flags & KERNEL_VXR_V0V7)
223			s390_handle_damage();
224		if (!test_cpu_flag(CIF_FPU))
225			kill_task = 1;
226	}
227	if (!mci.fc) {
228		/*
229		 * Floating point control register can't be restored.
230		 * If the kernel currently uses the floating pointer
231		 * registers and needs the FPC register the system is
232		 * stopped. If the process has its floating pointer
233		 * registers loaded it is terminated.
234		 */
235		if (S390_lowcore.fpu_flags & KERNEL_FPC)
236			s390_handle_damage();
237		if (!test_cpu_flag(CIF_FPU))
238			kill_task = 1;
239	}
240
241	if (MACHINE_HAS_VX) {
242		if (!mci.vr) {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
243			/*
244			 * Vector registers can't be restored. If the kernel
245			 * currently uses vector registers the system is
246			 * stopped. If the process has its vector registers
247			 * loaded it is terminated.
248			 */
249			if (S390_lowcore.fpu_flags & KERNEL_VXR)
250				s390_handle_damage();
251			if (!test_cpu_flag(CIF_FPU))
252				kill_task = 1;
253		}
254	}
255	/* Check if access registers are valid */
256	if (!mci.ar) {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
257		/*
258		 * Access registers have unknown contents.
259		 * Terminating task.
260		 */
261		kill_task = 1;
262	}
263	/* Check guarded storage registers */
264	cr2.val = S390_lowcore.cregs_save_area[2];
265	if (cr2.gse) {
266		if (!mci.gs) {
267			/*
268			 * Guarded storage register can't be restored and
269			 * the current processes uses guarded storage.
270			 * It has to be terminated.
271			 */
272			kill_task = 1;
273		}
 
 
 
 
 
 
274	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
275	/* Check if old PSW is valid */
276	if (!mci.wp) {
277		/*
278		 * Can't tell if we come from user or kernel mode
279		 * -> stopping machine.
280		 */
281		s390_handle_damage();
282	}
283	/* Check for invalid kernel instruction address */
284	if (!mci.ia && !umode) {
285		/*
286		 * The instruction address got lost while running
287		 * in the kernel -> stopping machine.
288		 */
289		s390_handle_damage();
290	}
291
292	if (!mci.ms || !mci.pm || !mci.ia)
293		kill_task = 1;
294
295	return kill_task;
296}
297NOKPROBE_SYMBOL(s390_check_registers);
298
299/*
300 * Backup the guest's machine check info to its description block
301 */
302static void notrace s390_backup_mcck_info(struct pt_regs *regs)
303{
304	struct mcck_volatile_info *mcck_backup;
305	struct sie_page *sie_page;
306
307	/* r14 contains the sie block, which was set in sie64a */
308	struct kvm_s390_sie_block *sie_block =
309			(struct kvm_s390_sie_block *) regs->gprs[14];
310
311	if (sie_block == NULL)
312		/* Something's seriously wrong, stop system. */
313		s390_handle_damage();
314
315	sie_page = container_of(sie_block, struct sie_page, sie_block);
316	mcck_backup = &sie_page->mcck_info;
317	mcck_backup->mcic = S390_lowcore.mcck_interruption_code &
318				~(MCCK_CODE_CP | MCCK_CODE_EXT_DAMAGE);
319	mcck_backup->ext_damage_code = S390_lowcore.external_damage_code;
320	mcck_backup->failing_storage_address
321			= S390_lowcore.failing_storage_address;
322}
323NOKPROBE_SYMBOL(s390_backup_mcck_info);
324
325#define MAX_IPD_COUNT	29
326#define MAX_IPD_TIME	(5 * 60 * USEC_PER_SEC) /* 5 minutes */
327
328#define ED_STP_ISLAND	6	/* External damage STP island check */
329#define ED_STP_SYNC	7	/* External damage STP sync check */
330
331#define MCCK_CODE_NO_GUEST	(MCCK_CODE_CP | MCCK_CODE_EXT_DAMAGE)
332
333/*
334 * machine check handler.
335 */
336void notrace s390_do_machine_check(struct pt_regs *regs)
337{
338	static int ipd_count;
339	static DEFINE_SPINLOCK(ipd_lock);
340	static unsigned long long last_ipd;
341	struct mcck_struct *mcck;
342	unsigned long long tmp;
343	union mci mci;
344	unsigned long mcck_dam_code;
345
346	nmi_enter();
347	inc_irq_stat(NMI_NMI);
348	mci.val = S390_lowcore.mcck_interruption_code;
349	mcck = this_cpu_ptr(&cpu_mcck);
 
350
351	if (mci.sd) {
352		/* System damage -> stopping machine */
353		s390_handle_damage();
354	}
355
356	/*
357	 * Reinject the instruction processing damages' machine checks
358	 * including Delayed Access Exception into the guest
359	 * instead of damaging the host if they happen in the guest.
360	 */
361	if (mci.pd && !test_cpu_flag(CIF_MCCK_GUEST)) {
362		if (mci.b) {
363			/* Processing backup -> verify if we can survive this */
364			u64 z_mcic, o_mcic, t_mcic;
 
365			z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<29);
366			o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
367				  1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
368				  1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 |
369				  1ULL<<16);
370			t_mcic = mci.val;
 
 
 
 
 
 
 
371
372			if (((t_mcic & z_mcic) != 0) ||
373			    ((t_mcic & o_mcic) != o_mcic)) {
374				s390_handle_damage();
 
375			}
376
377			/*
378			 * Nullifying exigent condition, therefore we might
379			 * retry this instruction.
380			 */
381			spin_lock(&ipd_lock);
382			tmp = get_tod_clock();
383			if (((tmp - last_ipd) >> 12) < MAX_IPD_TIME)
384				ipd_count++;
385			else
386				ipd_count = 1;
387			last_ipd = tmp;
388			if (ipd_count == MAX_IPD_COUNT)
389				s390_handle_damage();
390			spin_unlock(&ipd_lock);
391		} else {
392			/* Processing damage -> stopping machine */
393			s390_handle_damage();
 
394		}
395	}
396	if (s390_check_registers(mci, user_mode(regs))) {
397		/*
398		 * Couldn't restore all register contents for the
399		 * user space process -> mark task for termination.
400		 */
401		mcck->kill_task = 1;
402		mcck->mcck_code = mci.val;
403		set_cpu_flag(CIF_MCCK_PENDING);
 
 
 
 
 
 
 
 
404	}
405
406	/*
407	 * Backup the machine check's info if it happens when the guest
408	 * is running.
409	 */
410	if (test_cpu_flag(CIF_MCCK_GUEST))
411		s390_backup_mcck_info(regs);
412
413	if (mci.cd) {
414		/* Timing facility damage */
415		s390_handle_damage();
416	}
417	if (mci.ed && mci.ec) {
418		/* External damage */
 
 
 
 
419		if (S390_lowcore.external_damage_code & (1U << ED_STP_SYNC))
420			mcck->stp_queue |= stp_sync_check();
421		if (S390_lowcore.external_damage_code & (1U << ED_STP_ISLAND))
422			mcck->stp_queue |= stp_island_check();
423		if (mcck->stp_queue)
424			set_cpu_flag(CIF_MCCK_PENDING);
425	}
426
427	/*
428	 * Reinject storage related machine checks into the guest if they
429	 * happen when the guest is running.
430	 */
431	if (!test_cpu_flag(CIF_MCCK_GUEST)) {
432		if (mci.se)
433			/* Storage error uncorrected */
434			s390_handle_damage();
435		if (mci.ke)
436			/* Storage key-error uncorrected */
437			s390_handle_damage();
438		if (mci.ds && mci.fa)
439			/* Storage degradation */
440			s390_handle_damage();
441	}
442	if (mci.cp) {
443		/* Channel report word pending */
444		mcck->channel_report = 1;
445		set_cpu_flag(CIF_MCCK_PENDING);
446	}
447	if (mci.w) {
448		/* Warning pending */
449		mcck->warning = 1;
450		set_cpu_flag(CIF_MCCK_PENDING);
451	}
452
453	/*
454	 * If there are only Channel Report Pending and External Damage
455	 * machine checks, they will not be reinjected into the guest
456	 * because they refer to host conditions only.
457	 */
458	mcck_dam_code = (mci.val & MCIC_SUBCLASS_MASK);
459	if (test_cpu_flag(CIF_MCCK_GUEST) &&
460	(mcck_dam_code & MCCK_CODE_NO_GUEST) != mcck_dam_code) {
461		/* Set exit reason code for host's later handling */
462		*((long *)(regs->gprs[15] + __SF_SIE_REASON)) = -EINTR;
463	}
464	clear_cpu_flag(CIF_MCCK_GUEST);
465	nmi_exit();
466}
467NOKPROBE_SYMBOL(s390_do_machine_check);
468
469static int __init machine_check_init(void)
470{
471	ctl_set_bit(14, 25);	/* enable external damage MCH */
472	ctl_set_bit(14, 27);	/* enable system recovery MCH */
473	ctl_set_bit(14, 24);	/* enable warning MCH */
474	return 0;
475}
476early_initcall(machine_check_init);
v3.5.6
 
  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
 11#include <linux/kernel_stat.h>
 12#include <linux/init.h>
 13#include <linux/errno.h>
 14#include <linux/hardirq.h>
 
 
 
 15#include <linux/time.h>
 16#include <linux/module.h>
 
 
 
 17#include <asm/lowcore.h>
 18#include <asm/smp.h>
 19#include <asm/etr.h>
 20#include <asm/cputime.h>
 21#include <asm/nmi.h>
 22#include <asm/crw.h>
 
 
 
 
 23
 24struct 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
 33static 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
 40/*
 41 * Main machine check handler function. Will be called with interrupts enabled
 42 * or disabled and machine checks enabled or disabled.
 43 */
 44void s390_handle_mcck(void)
 45{
 46	unsigned long flags;
 47	struct mcck_struct mcck;
 48
 49	/*
 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);
 61
 62	if (mcck.channel_report)
 63		crw_handle_channel_report();
 64	/*
 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	 */
 73	if (mcck.warning) {	/* WARNING pending ? */
 74		static int mchchk_wng_posted = 0;
 75
 76		/* Use single cpu clear, as we cannot handle smp here. */
 77		__ctl_clear_bit(14, 24);	/* Disable WARNING MCH */
 78		if (xchg(&mchchk_wng_posted, 1) == 0)
 79			kill_cad_pid(SIGPWR, 1);
 80	}
 
 
 81	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}
 90EXPORT_SYMBOL_GPL(s390_handle_mcck);
 91
 92/*
 93 * returns 0 if all registers could be validated
 94 * returns 1 otherwise
 95 */
 96static int notrace s390_revalidate_registers(struct mci *mci)
 97{
 
 98	int kill_task;
 99	u64 zero;
100	void *fpt_save_area, *fpt_creg_save_area;
101
102	kill_task = 0;
103	zero = 0;
104
105	if (!mci->gr) {
106		/*
107		 * General purpose registers couldn't be restored and have
108		 * unknown contents. Process needs to be terminated.
109		 */
 
 
110		kill_task = 1;
111	}
112	if (!mci->fp) {
 
 
 
 
 
 
 
 
113		/*
114		 * Floating point registers can't be restored and
115		 * therefore the process needs to be terminated.
 
 
116		 */
117		kill_task = 1;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
118	}
119#ifndef CONFIG_64BIT
120	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));
126#endif
127
128	if (MACHINE_HAS_IEEE) {
129#ifdef CONFIG_64BIT
130		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;
134		fpt_creg_save_area = fpt_save_area + 128;
135#endif
136		if (!mci->fc) {
137			/*
138			 * Floating point control register can't be restored.
139			 * Task will be terminated.
 
 
140			 */
141			asm volatile("lfpc 0(%0)" : : "a" (&zero), "m" (zero));
142			kill_task = 1;
143
144		} else
145			asm volatile("lfpc 0(%0)" : : "a" (fpt_creg_save_area));
146
147		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));
165	}
166	/* Revalidate access registers */
167	asm volatile(
168		"	lam	0,15,0(%0)"
169		: : "a" (&S390_lowcore.access_regs_save_area));
170	if (!mci->ar) {
171		/*
172		 * Access registers have unknown contents.
173		 * Terminating task.
174		 */
175		kill_task = 1;
176	}
177	/* Revalidate control registers */
178	if (!mci->cr) {
179		/*
180		 * Control registers have unknown contents.
181		 * Can't recover and therefore stopping machine.
182		 */
183		s390_handle_damage("invalid control registers.");
184	} else {
185#ifdef CONFIG_64BIT
186		asm volatile(
187			"	lctlg	0,15,0(%0)"
188			: : "a" (&S390_lowcore.cregs_save_area));
189#else
190		asm volatile(
191			"	lctl	0,15,0(%0)"
192			: : "a" (&S390_lowcore.cregs_save_area));
193#endif
194	}
195	/*
196	 * We don't even try to revalidate the TOD register, since we simply
197	 * can't write something sensible into that register.
198	 */
199#ifdef CONFIG_64BIT
200	/*
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)
205		asm volatile(
206			"	sr	0,0\n"
207			"	sckpf"
208			: : : "0", "cc");
209	else
210		asm volatile(
211			"	l	0,0(%0)\n"
212			"	sckpf"
213			: : "a" (&S390_lowcore.tod_progreg_save_area)
214			: "0", "cc");
215#endif
216	/* Revalidate clock comparator register */
217	if (S390_lowcore.clock_comparator == -1)
218		set_clock_comparator(S390_lowcore.mcck_clock);
219	else
220		set_clock_comparator(S390_lowcore.clock_comparator);
221	/* 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
235#define MAX_IPD_COUNT	29
236#define MAX_IPD_TIME	(5 * 60 * USEC_PER_SEC) /* 5 minutes */
237
238#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
243/*
244 * machine check handler.
245 */
246void notrace s390_do_machine_check(struct pt_regs *regs)
247{
248	static int ipd_count;
249	static DEFINE_SPINLOCK(ipd_lock);
250	static unsigned long long last_ipd;
251	struct mcck_struct *mcck;
252	unsigned long long tmp;
253	struct mci *mci;
254	int umode;
255
256	nmi_enter();
257	kstat_cpu(smp_processor_id()).irqs[NMI_NMI]++;
258	mci = (struct mci *) &S390_lowcore.mcck_interruption_code;
259	mcck = &__get_cpu_var(cpu_mcck);
260	umode = user_mode(regs);
261
262	if (mci->sd) {
263		/* System damage -> stopping machine */
264		s390_handle_damage("received system damage machine check.");
265	}
266	if (mci->pd) {
267		if (mci->b) {
 
 
 
 
 
 
268			/* Processing backup -> verify if we can survive this */
269			u64 z_mcic, o_mcic, t_mcic;
270#ifdef CONFIG_64BIT
271			z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<29);
272			o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
273				  1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
274				  1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 |
275				  1ULL<<16);
276#else
277			z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<57 | 1ULL<<50 |
278				  1ULL<<29);
279			o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
280				  1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
281				  1ULL<<30 | 1ULL<<20 | 1ULL<<17 | 1ULL<<16);
282#endif
283			t_mcic = *(u64 *)mci;
284
285			if (((t_mcic & z_mcic) != 0) ||
286			    ((t_mcic & o_mcic) != o_mcic)) {
287				s390_handle_damage("processing backup machine "
288						   "check with damage.");
289			}
290
291			/*
292			 * Nullifying exigent condition, therefore we might
293			 * retry this instruction.
294			 */
295			spin_lock(&ipd_lock);
296			tmp = get_clock();
297			if (((tmp - last_ipd) >> 12) < MAX_IPD_TIME)
298				ipd_count++;
299			else
300				ipd_count = 1;
301			last_ipd = tmp;
302			if (ipd_count == MAX_IPD_COUNT)
303				s390_handle_damage("too many ipd retries.");
304			spin_unlock(&ipd_lock);
305		} else {
306			/* Processing damage -> stopping machine */
307			s390_handle_damage("received instruction processing "
308					   "damage machine check.");
309		}
310	}
311	if (s390_revalidate_registers(mci)) {
312		if (umode) {
313			/*
314			 * Couldn't restore all register contents while in
315			 * user mode -> mark task for termination.
316			 */
317			mcck->kill_task = 1;
318			mcck->mcck_code = *(unsigned long long *) mci;
319			set_thread_flag(TIF_MCCK_PENDING);
320		} else {
321			/*
322			 * Couldn't restore all register contents while in
323			 * kernel mode -> stopping machine.
324			 */
325			s390_handle_damage("unable to revalidate registers.");
326		}
327	}
328	if (mci->cd) {
 
 
 
 
 
 
 
 
329		/* Timing facility damage */
330		s390_handle_damage("TOD clock damaged");
331	}
332	if (mci->ed && mci->ec) {
333		/* External damage */
334		if (S390_lowcore.external_damage_code & (1U << ED_ETR_SYNC))
335			etr_sync_check();
336		if (S390_lowcore.external_damage_code & (1U << ED_ETR_SWITCH))
337			etr_switch_to_local();
338		if (S390_lowcore.external_damage_code & (1U << ED_STP_SYNC))
339			stp_sync_check();
340		if (S390_lowcore.external_damage_code & (1U << ED_STP_ISLAND))
341			stp_island_check();
 
 
342	}
343	if (mci->se)
344		/* Storage error uncorrected */
345		s390_handle_damage("received storage error uncorrected "
346				   "machine check.");
347	if (mci->ke)
348		/* Storage key-error uncorrected */
349		s390_handle_damage("received storage key-error uncorrected "
350				   "machine check.");
351	if (mci->ds && mci->fa)
352		/* Storage degradation */
353		s390_handle_damage("received storage degradation machine "
354				   "check.");
355	if (mci->cp) {
 
 
 
 
356		/* Channel report word pending */
357		mcck->channel_report = 1;
358		set_thread_flag(TIF_MCCK_PENDING);
359	}
360	if (mci->w) {
361		/* Warning pending */
362		mcck->warning = 1;
363		set_thread_flag(TIF_MCCK_PENDING);
 
 
 
 
 
 
 
 
 
 
 
 
364	}
 
365	nmi_exit();
366}
 
367
368static int __init machine_check_init(void)
369{
370	ctl_set_bit(14, 25);	/* enable external damage MCH */
371	ctl_set_bit(14, 27);	/* enable system recovery MCH */
372	ctl_set_bit(14, 24);	/* enable warning MCH */
373	return 0;
374}
375arch_initcall(machine_check_init);