xref: /openbmc/linux/arch/s390/kernel/nmi.c (revision 904818e2)
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 #include <asm/switch_to.h>
24 #include <asm/fpu-internal.h>
25 #include <asm/ctl_reg.h>
26 
27 struct mcck_struct {
28 	int kill_task;
29 	int channel_report;
30 	int warning;
31 	unsigned long long mcck_code;
32 };
33 
34 static DEFINE_PER_CPU(struct mcck_struct, cpu_mcck);
35 
36 static void s390_handle_damage(char *msg)
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.kill_task) {
85 		local_irq_enable();
86 		printk(KERN_EMERG "mcck: Terminating task because of machine "
87 		       "malfunction (code 0x%016llx).\n", mcck.mcck_code);
88 		printk(KERN_EMERG "mcck: task: %s, pid: %d.\n",
89 		       current->comm, current->pid);
90 		do_exit(SIGSEGV);
91 	}
92 }
93 EXPORT_SYMBOL_GPL(s390_handle_mcck);
94 
95 /*
96  * returns 0 if all registers could be validated
97  * returns 1 otherwise
98  */
99 static int notrace s390_revalidate_registers(struct mci *mci)
100 {
101 	int kill_task;
102 	u64 zero;
103 	void *fpt_save_area, *fpt_creg_save_area;
104 
105 	kill_task = 0;
106 	zero = 0;
107 
108 	if (!mci->gr) {
109 		/*
110 		 * General purpose registers couldn't be restored and have
111 		 * unknown contents. Process needs to be terminated.
112 		 */
113 		kill_task = 1;
114 	}
115 	if (!mci->fp) {
116 		/*
117 		 * Floating point registers can't be restored and
118 		 * therefore the process needs to be terminated.
119 		 */
120 		kill_task = 1;
121 	}
122 	fpt_save_area = &S390_lowcore.floating_pt_save_area;
123 	fpt_creg_save_area = &S390_lowcore.fpt_creg_save_area;
124 	if (!mci->fc) {
125 		/*
126 		 * Floating point control register can't be restored.
127 		 * Task will be terminated.
128 		 */
129 		asm volatile("lfpc 0(%0)" : : "a" (&zero), "m" (zero));
130 		kill_task = 1;
131 	} else
132 		asm volatile("lfpc 0(%0)" : : "a" (fpt_creg_save_area));
133 
134 	if (!MACHINE_HAS_VX) {
135 		/* Revalidate floating point registers */
136 		asm volatile(
137 			"	ld	0,0(%0)\n"
138 			"	ld	1,8(%0)\n"
139 			"	ld	2,16(%0)\n"
140 			"	ld	3,24(%0)\n"
141 			"	ld	4,32(%0)\n"
142 			"	ld	5,40(%0)\n"
143 			"	ld	6,48(%0)\n"
144 			"	ld	7,56(%0)\n"
145 			"	ld	8,64(%0)\n"
146 			"	ld	9,72(%0)\n"
147 			"	ld	10,80(%0)\n"
148 			"	ld	11,88(%0)\n"
149 			"	ld	12,96(%0)\n"
150 			"	ld	13,104(%0)\n"
151 			"	ld	14,112(%0)\n"
152 			"	ld	15,120(%0)\n"
153 			: : "a" (fpt_save_area));
154 	} else {
155 		/* Revalidate vector registers */
156 		union ctlreg0 cr0;
157 
158 		if (!mci->vr) {
159 			/*
160 			 * Vector registers can't be restored and therefore
161 			 * the process needs to be terminated.
162 			 */
163 			kill_task = 1;
164 		}
165 		cr0.val = S390_lowcore.cregs_save_area[0];
166 		cr0.afp = cr0.vx = 1;
167 		__ctl_load(cr0.val, 0, 0);
168 		restore_vx_regs((__vector128 *)
169 				&S390_lowcore.vector_save_area);
170 		__ctl_load(S390_lowcore.cregs_save_area[0], 0, 0);
171 	}
172 	/* Revalidate access registers */
173 	asm volatile(
174 		"	lam	0,15,0(%0)"
175 		: : "a" (&S390_lowcore.access_regs_save_area));
176 	if (!mci->ar) {
177 		/*
178 		 * Access registers have unknown contents.
179 		 * Terminating task.
180 		 */
181 		kill_task = 1;
182 	}
183 	/* Revalidate control registers */
184 	if (!mci->cr) {
185 		/*
186 		 * Control registers have unknown contents.
187 		 * Can't recover and therefore stopping machine.
188 		 */
189 		s390_handle_damage("invalid control registers.");
190 	} else {
191 		asm volatile(
192 			"	lctlg	0,15,0(%0)"
193 			: : "a" (&S390_lowcore.cregs_save_area));
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 	/*
200 	 * See if we can revalidate the TOD programmable register with its
201 	 * old contents (should be zero) otherwise set it to zero.
202 	 */
203 	if (!mci->pr)
204 		asm volatile(
205 			"	sr	0,0\n"
206 			"	sckpf"
207 			: : : "0", "cc");
208 	else
209 		asm volatile(
210 			"	l	0,0(%0)\n"
211 			"	sckpf"
212 			: : "a" (&S390_lowcore.tod_progreg_save_area)
213 			: "0", "cc");
214 	/* Revalidate clock comparator register */
215 	set_clock_comparator(S390_lowcore.clock_comparator);
216 	/* Check if old PSW is valid */
217 	if (!mci->wp)
218 		/*
219 		 * Can't tell if we come from user or kernel mode
220 		 * -> stopping machine.
221 		 */
222 		s390_handle_damage("old psw invalid.");
223 
224 	if (!mci->ms || !mci->pm || !mci->ia)
225 		kill_task = 1;
226 
227 	return kill_task;
228 }
229 
230 #define MAX_IPD_COUNT	29
231 #define MAX_IPD_TIME	(5 * 60 * USEC_PER_SEC) /* 5 minutes */
232 
233 #define ED_STP_ISLAND	6	/* External damage STP island check */
234 #define ED_STP_SYNC	7	/* External damage STP sync check */
235 #define ED_ETR_SYNC	12	/* External damage ETR sync check */
236 #define ED_ETR_SWITCH	13	/* External damage ETR switch to local */
237 
238 /*
239  * machine check handler.
240  */
241 void notrace s390_do_machine_check(struct pt_regs *regs)
242 {
243 	static int ipd_count;
244 	static DEFINE_SPINLOCK(ipd_lock);
245 	static unsigned long long last_ipd;
246 	struct mcck_struct *mcck;
247 	unsigned long long tmp;
248 	struct mci *mci;
249 	int umode;
250 
251 	nmi_enter();
252 	inc_irq_stat(NMI_NMI);
253 	mci = (struct mci *) &S390_lowcore.mcck_interruption_code;
254 	mcck = this_cpu_ptr(&cpu_mcck);
255 	umode = user_mode(regs);
256 
257 	if (mci->sd) {
258 		/* System damage -> stopping machine */
259 		s390_handle_damage("received system damage machine check.");
260 	}
261 	if (mci->pd) {
262 		if (mci->b) {
263 			/* Processing backup -> verify if we can survive this */
264 			u64 z_mcic, o_mcic, t_mcic;
265 			z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<29);
266 			o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
267 				  1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
268 				  1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 |
269 				  1ULL<<16);
270 			t_mcic = *(u64 *)mci;
271 
272 			if (((t_mcic & z_mcic) != 0) ||
273 			    ((t_mcic & o_mcic) != o_mcic)) {
274 				s390_handle_damage("processing backup machine "
275 						   "check with damage.");
276 			}
277 
278 			/*
279 			 * Nullifying exigent condition, therefore we might
280 			 * retry this instruction.
281 			 */
282 			spin_lock(&ipd_lock);
283 			tmp = get_tod_clock();
284 			if (((tmp - last_ipd) >> 12) < MAX_IPD_TIME)
285 				ipd_count++;
286 			else
287 				ipd_count = 1;
288 			last_ipd = tmp;
289 			if (ipd_count == MAX_IPD_COUNT)
290 				s390_handle_damage("too many ipd retries.");
291 			spin_unlock(&ipd_lock);
292 		} else {
293 			/* Processing damage -> stopping machine */
294 			s390_handle_damage("received instruction processing "
295 					   "damage machine check.");
296 		}
297 	}
298 	if (s390_revalidate_registers(mci)) {
299 		if (umode) {
300 			/*
301 			 * Couldn't restore all register contents while in
302 			 * user mode -> mark task for termination.
303 			 */
304 			mcck->kill_task = 1;
305 			mcck->mcck_code = *(unsigned long long *) mci;
306 			set_cpu_flag(CIF_MCCK_PENDING);
307 		} else {
308 			/*
309 			 * Couldn't restore all register contents while in
310 			 * kernel mode -> stopping machine.
311 			 */
312 			s390_handle_damage("unable to revalidate registers.");
313 		}
314 	}
315 	if (mci->cd) {
316 		/* Timing facility damage */
317 		s390_handle_damage("TOD clock damaged");
318 	}
319 	if (mci->ed && mci->ec) {
320 		/* External damage */
321 		if (S390_lowcore.external_damage_code & (1U << ED_ETR_SYNC))
322 			etr_sync_check();
323 		if (S390_lowcore.external_damage_code & (1U << ED_ETR_SWITCH))
324 			etr_switch_to_local();
325 		if (S390_lowcore.external_damage_code & (1U << ED_STP_SYNC))
326 			stp_sync_check();
327 		if (S390_lowcore.external_damage_code & (1U << ED_STP_ISLAND))
328 			stp_island_check();
329 	}
330 	if (mci->se)
331 		/* Storage error uncorrected */
332 		s390_handle_damage("received storage error uncorrected "
333 				   "machine check.");
334 	if (mci->ke)
335 		/* Storage key-error uncorrected */
336 		s390_handle_damage("received storage key-error uncorrected "
337 				   "machine check.");
338 	if (mci->ds && mci->fa)
339 		/* Storage degradation */
340 		s390_handle_damage("received storage degradation machine "
341 				   "check.");
342 	if (mci->cp) {
343 		/* Channel report word pending */
344 		mcck->channel_report = 1;
345 		set_cpu_flag(CIF_MCCK_PENDING);
346 	}
347 	if (mci->w) {
348 		/* Warning pending */
349 		mcck->warning = 1;
350 		set_cpu_flag(CIF_MCCK_PENDING);
351 	}
352 	nmi_exit();
353 }
354 
355 static int __init machine_check_init(void)
356 {
357 	ctl_set_bit(14, 25);	/* enable external damage MCH */
358 	ctl_set_bit(14, 27);	/* enable system recovery MCH */
359 	ctl_set_bit(14, 24);	/* enable warning MCH */
360 	return 0;
361 }
362 arch_initcall(machine_check_init);
363