xref: /openbmc/linux/arch/s390/kernel/nmi.c (revision 3994a52b)
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/export.h>
17 #include <asm/lowcore.h>
18 #include <asm/smp.h>
19 #include <asm/stp.h>
20 #include <asm/cputime.h>
21 #include <asm/nmi.h>
22 #include <asm/crw.h>
23 #include <asm/switch_to.h>
24 #include <asm/ctl_reg.h>
25 
26 struct mcck_struct {
27 	unsigned int kill_task : 1;
28 	unsigned int channel_report : 1;
29 	unsigned int warning : 1;
30 	unsigned int stp_queue : 1;
31 	unsigned long mcck_code;
32 };
33 
34 static DEFINE_PER_CPU(struct mcck_struct, cpu_mcck);
35 
36 static void s390_handle_damage(void)
37 {
38 	smp_send_stop();
39 	disabled_wait((unsigned long) __builtin_return_address(0));
40 	while (1);
41 }
42 
43 /*
44  * Main machine check handler function. Will be called with interrupts enabled
45  * or disabled and machine checks enabled or disabled.
46  */
47 void s390_handle_mcck(void)
48 {
49 	unsigned long flags;
50 	struct mcck_struct mcck;
51 
52 	/*
53 	 * Disable machine checks and get the current state of accumulated
54 	 * machine checks. Afterwards delete the old state and enable machine
55 	 * checks again.
56 	 */
57 	local_irq_save(flags);
58 	local_mcck_disable();
59 	mcck = *this_cpu_ptr(&cpu_mcck);
60 	memset(this_cpu_ptr(&cpu_mcck), 0, sizeof(mcck));
61 	clear_cpu_flag(CIF_MCCK_PENDING);
62 	local_mcck_enable();
63 	local_irq_restore(flags);
64 
65 	if (mcck.channel_report)
66 		crw_handle_channel_report();
67 	/*
68 	 * A warning may remain for a prolonged period on the bare iron.
69 	 * (actually until the machine is powered off, or the problem is gone)
70 	 * So we just stop listening for the WARNING MCH and avoid continuously
71 	 * being interrupted.  One caveat is however, that we must do this per
72 	 * processor and cannot use the smp version of ctl_clear_bit().
73 	 * On VM we only get one interrupt per virtally presented machinecheck.
74 	 * Though one suffices, we may get one interrupt per (virtual) cpu.
75 	 */
76 	if (mcck.warning) {	/* WARNING pending ? */
77 		static int mchchk_wng_posted = 0;
78 
79 		/* Use single cpu clear, as we cannot handle smp here. */
80 		__ctl_clear_bit(14, 24);	/* Disable WARNING MCH */
81 		if (xchg(&mchchk_wng_posted, 1) == 0)
82 			kill_cad_pid(SIGPWR, 1);
83 	}
84 	if (mcck.stp_queue)
85 		stp_queue_work();
86 	if (mcck.kill_task) {
87 		local_irq_enable();
88 		printk(KERN_EMERG "mcck: Terminating task because of machine "
89 		       "malfunction (code 0x%016lx).\n", mcck.mcck_code);
90 		printk(KERN_EMERG "mcck: task: %s, pid: %d.\n",
91 		       current->comm, current->pid);
92 		do_exit(SIGSEGV);
93 	}
94 }
95 EXPORT_SYMBOL_GPL(s390_handle_mcck);
96 
97 /*
98  * returns 0 if all registers could be validated
99  * returns 1 otherwise
100  */
101 static int notrace s390_validate_registers(union mci mci, int umode)
102 {
103 	int kill_task;
104 	u64 zero;
105 	void *fpt_save_area;
106 
107 	kill_task = 0;
108 	zero = 0;
109 
110 	if (!mci.gr) {
111 		/*
112 		 * General purpose registers couldn't be restored and have
113 		 * unknown contents. Stop system or terminate process.
114 		 */
115 		if (!umode)
116 			s390_handle_damage();
117 		kill_task = 1;
118 	}
119 	if (!mci.fp) {
120 		/*
121 		 * Floating point registers can't be restored. If the
122 		 * kernel currently uses floating point registers the
123 		 * system is stopped. If the process has its floating
124 		 * pointer registers loaded it is terminated.
125 		 * Otherwise just revalidate the registers.
126 		 */
127 		if (S390_lowcore.fpu_flags & KERNEL_VXR_V0V7)
128 			s390_handle_damage();
129 		if (!test_cpu_flag(CIF_FPU))
130 			kill_task = 1;
131 	}
132 	fpt_save_area = &S390_lowcore.floating_pt_save_area;
133 	if (!mci.fc) {
134 		/*
135 		 * Floating point control register can't be restored.
136 		 * If the kernel currently uses the floating pointer
137 		 * registers and needs the FPC register the system is
138 		 * stopped. If the process has its floating pointer
139 		 * registers loaded it is terminated. Otherwiese the
140 		 * FPC is just revalidated.
141 		 */
142 		if (S390_lowcore.fpu_flags & KERNEL_FPC)
143 			s390_handle_damage();
144 		asm volatile("lfpc %0" : : "Q" (zero));
145 		if (!test_cpu_flag(CIF_FPU))
146 			kill_task = 1;
147 	} else {
148 		asm volatile("lfpc %0"
149 			     : : "Q" (S390_lowcore.fpt_creg_save_area));
150 	}
151 
152 	if (!MACHINE_HAS_VX) {
153 		/* Validate floating point registers */
154 		asm volatile(
155 			"	ld	0,0(%0)\n"
156 			"	ld	1,8(%0)\n"
157 			"	ld	2,16(%0)\n"
158 			"	ld	3,24(%0)\n"
159 			"	ld	4,32(%0)\n"
160 			"	ld	5,40(%0)\n"
161 			"	ld	6,48(%0)\n"
162 			"	ld	7,56(%0)\n"
163 			"	ld	8,64(%0)\n"
164 			"	ld	9,72(%0)\n"
165 			"	ld	10,80(%0)\n"
166 			"	ld	11,88(%0)\n"
167 			"	ld	12,96(%0)\n"
168 			"	ld	13,104(%0)\n"
169 			"	ld	14,112(%0)\n"
170 			"	ld	15,120(%0)\n"
171 			: : "a" (fpt_save_area) : "memory");
172 	} else {
173 		/* Validate vector registers */
174 		union ctlreg0 cr0;
175 
176 		if (!mci.vr) {
177 			/*
178 			 * Vector registers can't be restored. If the kernel
179 			 * currently uses vector registers the system is
180 			 * stopped. If the process has its vector registers
181 			 * loaded it is terminated. Otherwise just revalidate
182 			 * the registers.
183 			 */
184 			if (S390_lowcore.fpu_flags & KERNEL_VXR)
185 				s390_handle_damage();
186 			if (!test_cpu_flag(CIF_FPU))
187 				kill_task = 1;
188 		}
189 		cr0.val = S390_lowcore.cregs_save_area[0];
190 		cr0.afp = cr0.vx = 1;
191 		__ctl_load(cr0.val, 0, 0);
192 		asm volatile(
193 			"	la	1,%0\n"
194 			"	.word	0xe70f,0x1000,0x0036\n"	/* vlm 0,15,0(1) */
195 			"	.word	0xe70f,0x1100,0x0c36\n"	/* vlm 16,31,256(1) */
196 			: : "Q" (*(struct vx_array *)
197 				 &S390_lowcore.vector_save_area) : "1");
198 		__ctl_load(S390_lowcore.cregs_save_area[0], 0, 0);
199 	}
200 	/* Validate access registers */
201 	asm volatile(
202 		"	lam	0,15,0(%0)"
203 		: : "a" (&S390_lowcore.access_regs_save_area));
204 	if (!mci.ar) {
205 		/*
206 		 * Access registers have unknown contents.
207 		 * Terminating task.
208 		 */
209 		kill_task = 1;
210 	}
211 	/* Validate control registers */
212 	if (!mci.cr) {
213 		/*
214 		 * Control registers have unknown contents.
215 		 * Can't recover and therefore stopping machine.
216 		 */
217 		s390_handle_damage();
218 	} else {
219 		asm volatile(
220 			"	lctlg	0,15,0(%0)"
221 			: : "a" (&S390_lowcore.cregs_save_area) : "memory");
222 	}
223 	/*
224 	 * We don't even try to validate the TOD register, since we simply
225 	 * can't write something sensible into that register.
226 	 */
227 	/*
228 	 * See if we can validate the TOD programmable register with its
229 	 * old contents (should be zero) otherwise set it to zero.
230 	 */
231 	if (!mci.pr)
232 		asm volatile(
233 			"	sr	0,0\n"
234 			"	sckpf"
235 			: : : "0", "cc");
236 	else
237 		asm volatile(
238 			"	l	0,%0\n"
239 			"	sckpf"
240 			: : "Q" (S390_lowcore.tod_progreg_save_area)
241 			: "0", "cc");
242 	/* Validate clock comparator register */
243 	set_clock_comparator(S390_lowcore.clock_comparator);
244 	/* Check if old PSW is valid */
245 	if (!mci.wp)
246 		/*
247 		 * Can't tell if we come from user or kernel mode
248 		 * -> stopping machine.
249 		 */
250 		s390_handle_damage();
251 
252 	if (!mci.ms || !mci.pm || !mci.ia)
253 		kill_task = 1;
254 
255 	return kill_task;
256 }
257 
258 #define MAX_IPD_COUNT	29
259 #define MAX_IPD_TIME	(5 * 60 * USEC_PER_SEC) /* 5 minutes */
260 
261 #define ED_STP_ISLAND	6	/* External damage STP island check */
262 #define ED_STP_SYNC	7	/* External damage STP sync check */
263 
264 /*
265  * machine check handler.
266  */
267 void notrace s390_do_machine_check(struct pt_regs *regs)
268 {
269 	static int ipd_count;
270 	static DEFINE_SPINLOCK(ipd_lock);
271 	static unsigned long long last_ipd;
272 	struct mcck_struct *mcck;
273 	unsigned long long tmp;
274 	union mci mci;
275 
276 	nmi_enter();
277 	inc_irq_stat(NMI_NMI);
278 	mci.val = S390_lowcore.mcck_interruption_code;
279 	mcck = this_cpu_ptr(&cpu_mcck);
280 
281 	if (mci.sd) {
282 		/* System damage -> stopping machine */
283 		s390_handle_damage();
284 	}
285 	if (mci.pd) {
286 		if (mci.b) {
287 			/* Processing backup -> verify if we can survive this */
288 			u64 z_mcic, o_mcic, t_mcic;
289 			z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<29);
290 			o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
291 				  1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
292 				  1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 |
293 				  1ULL<<16);
294 			t_mcic = mci.val;
295 
296 			if (((t_mcic & z_mcic) != 0) ||
297 			    ((t_mcic & o_mcic) != o_mcic)) {
298 				s390_handle_damage();
299 			}
300 
301 			/*
302 			 * Nullifying exigent condition, therefore we might
303 			 * retry this instruction.
304 			 */
305 			spin_lock(&ipd_lock);
306 			tmp = get_tod_clock();
307 			if (((tmp - last_ipd) >> 12) < MAX_IPD_TIME)
308 				ipd_count++;
309 			else
310 				ipd_count = 1;
311 			last_ipd = tmp;
312 			if (ipd_count == MAX_IPD_COUNT)
313 				s390_handle_damage();
314 			spin_unlock(&ipd_lock);
315 		} else {
316 			/* Processing damage -> stopping machine */
317 			s390_handle_damage();
318 		}
319 	}
320 	if (s390_validate_registers(mci, user_mode(regs))) {
321 		/*
322 		 * Couldn't restore all register contents for the
323 		 * user space process -> mark task for termination.
324 		 */
325 		mcck->kill_task = 1;
326 		mcck->mcck_code = mci.val;
327 		set_cpu_flag(CIF_MCCK_PENDING);
328 	}
329 	if (mci.cd) {
330 		/* Timing facility damage */
331 		s390_handle_damage();
332 	}
333 	if (mci.ed && mci.ec) {
334 		/* External damage */
335 		if (S390_lowcore.external_damage_code & (1U << ED_STP_SYNC))
336 			mcck->stp_queue |= stp_sync_check();
337 		if (S390_lowcore.external_damage_code & (1U << ED_STP_ISLAND))
338 			mcck->stp_queue |= stp_island_check();
339 		if (mcck->stp_queue)
340 			set_cpu_flag(CIF_MCCK_PENDING);
341 	}
342 	if (mci.se)
343 		/* Storage error uncorrected */
344 		s390_handle_damage();
345 	if (mci.ke)
346 		/* Storage key-error uncorrected */
347 		s390_handle_damage();
348 	if (mci.ds && mci.fa)
349 		/* Storage degradation */
350 		s390_handle_damage();
351 	if (mci.cp) {
352 		/* Channel report word pending */
353 		mcck->channel_report = 1;
354 		set_cpu_flag(CIF_MCCK_PENDING);
355 	}
356 	if (mci.w) {
357 		/* Warning pending */
358 		mcck->warning = 1;
359 		set_cpu_flag(CIF_MCCK_PENDING);
360 	}
361 	nmi_exit();
362 }
363 
364 static int __init machine_check_init(void)
365 {
366 	ctl_set_bit(14, 25);	/* enable external damage MCH */
367 	ctl_set_bit(14, 27);	/* enable system recovery MCH */
368 	ctl_set_bit(14, 24);	/* enable warning MCH */
369 	return 0;
370 }
371 early_initcall(machine_check_init);
372