2 * Copyright (C) 2001 Dave Engebretsen IBM Corporation
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 * 2001/09/21 : engebret : Created with minimal EPOW and HW exception support.
24 #include <linux/errno.h>
25 #include <linux/threads.h>
26 #include <linux/kernel_stat.h>
27 #include <linux/signal.h>
28 #include <linux/sched.h>
29 #include <linux/ioport.h>
30 #include <linux/interrupt.h>
31 #include <linux/timex.h>
32 #include <linux/init.h>
33 #include <linux/slab.h>
34 #include <linux/pci.h>
35 #include <linux/delay.h>
36 #include <linux/irq.h>
37 #include <linux/random.h>
38 #include <linux/sysrq.h>
39 #include <linux/bitops.h>
41 #include <asm/uaccess.h>
42 #include <asm/system.h>
44 #include <asm/pgtable.h>
46 #include <asm/cache.h>
48 #include <asm/ptrace.h>
49 #include <asm/machdep.h>
52 #include <asm/firmware.h>
56 static unsigned char ras_log_buf[RTAS_ERROR_LOG_MAX];
57 static DEFINE_SPINLOCK(ras_log_buf_lock);
59 char mce_data_buf[RTAS_ERROR_LOG_MAX];
61 static int ras_get_sensor_state_token;
62 static int ras_check_exception_token;
64 #define EPOW_SENSOR_TOKEN 9
65 #define EPOW_SENSOR_INDEX 0
66 #define RAS_VECTOR_OFFSET 0x500
68 static irqreturn_t ras_epow_interrupt(int irq, void *dev_id,
69 struct pt_regs * regs);
70 static irqreturn_t ras_error_interrupt(int irq, void *dev_id,
71 struct pt_regs * regs);
76 static void request_ras_irqs(struct device_node *np,
77 irqreturn_t (*handler)(int, void *, struct pt_regs *),
80 int i, index, count = 0;
83 unsigned int opicplen;
84 unsigned int virqs[16];
86 /* Check for obsolete "open-pic-interrupt" property. If present, then
87 * map those interrupts using the default interrupt host and default
90 opicprop = (u32 *)get_property(np, "open-pic-interrupt", &opicplen);
92 opicplen /= sizeof(u32);
93 for (i = 0; i < opicplen; i++) {
96 virqs[count] = irq_create_mapping(NULL, *(opicprop++),
98 if (virqs[count] == NO_IRQ)
99 printk(KERN_ERR "Unable to allocate interrupt "
100 "number for %s\n", np->full_name);
106 /* Else use normal interrupt tree parsing */
108 /* First try to do a proper OF tree parsing */
109 for (index = 0; of_irq_map_one(np, index, &oirq) == 0;
113 virqs[count] = irq_create_of_mapping(oirq.controller,
116 if (virqs[count] == NO_IRQ)
117 printk(KERN_ERR "Unable to allocate interrupt "
118 "number for %s\n", np->full_name);
124 /* Now request them */
125 for (i = 0; i < count; i++) {
126 if (request_irq(virqs[i], handler, 0, name, NULL)) {
127 printk(KERN_ERR "Unable to request interrupt %d for "
128 "%s\n", virqs[i], np->full_name);
135 * Initialize handlers for the set of interrupts caused by hardware errors
136 * and power system events.
138 static int __init init_ras_IRQ(void)
140 struct device_node *np;
142 ras_get_sensor_state_token = rtas_token("get-sensor-state");
143 ras_check_exception_token = rtas_token("check-exception");
145 /* Internal Errors */
146 np = of_find_node_by_path("/event-sources/internal-errors");
148 request_ras_irqs(np, ras_error_interrupt, "RAS_ERROR");
153 np = of_find_node_by_path("/event-sources/epow-events");
155 request_ras_irqs(np, ras_epow_interrupt, "RAS_EPOW");
161 __initcall(init_ras_IRQ);
164 * Handle power subsystem events (EPOW).
166 * Presently we just log the event has occurred. This should be fixed
167 * to examine the type of power failure and take appropriate action where
168 * the time horizon permits something useful to be done.
171 ras_epow_interrupt(int irq, void *dev_id, struct pt_regs * regs)
173 int status = 0xdeadbeef;
177 status = rtas_call(ras_get_sensor_state_token, 2, 2, &state,
178 EPOW_SENSOR_TOKEN, EPOW_SENSOR_INDEX);
181 critical = 1; /* Time Critical */
185 spin_lock(&ras_log_buf_lock);
187 status = rtas_call(ras_check_exception_token, 6, 1, NULL,
190 RTAS_EPOW_WARNING | RTAS_POWERMGM_EVENTS,
191 critical, __pa(&ras_log_buf),
192 rtas_get_error_log_max());
194 udbg_printf("EPOW <0x%lx 0x%x 0x%x>\n",
195 *((unsigned long *)&ras_log_buf), status, state);
196 printk(KERN_WARNING "EPOW <0x%lx 0x%x 0x%x>\n",
197 *((unsigned long *)&ras_log_buf), status, state);
199 /* format and print the extended information */
200 log_error(ras_log_buf, ERR_TYPE_RTAS_LOG, 0);
202 spin_unlock(&ras_log_buf_lock);
207 * Handle hardware error interrupts.
209 * RTAS check-exception is called to collect data on the exception. If
210 * the error is deemed recoverable, we log a warning and return.
211 * For nonrecoverable errors, an error is logged and we stop all processing
212 * as quickly as possible in order to prevent propagation of the failure.
215 ras_error_interrupt(int irq, void *dev_id, struct pt_regs * regs)
217 struct rtas_error_log *rtas_elog;
218 int status = 0xdeadbeef;
221 spin_lock(&ras_log_buf_lock);
223 status = rtas_call(ras_check_exception_token, 6, 1, NULL,
226 RTAS_INTERNAL_ERROR, 1 /*Time Critical */,
228 rtas_get_error_log_max());
230 rtas_elog = (struct rtas_error_log *)ras_log_buf;
232 if ((status == 0) && (rtas_elog->severity >= RTAS_SEVERITY_ERROR_SYNC))
237 /* format and print the extended information */
238 log_error(ras_log_buf, ERR_TYPE_RTAS_LOG, fatal);
241 udbg_printf("Fatal HW Error <0x%lx 0x%x>\n",
242 *((unsigned long *)&ras_log_buf), status);
243 printk(KERN_EMERG "Error: Fatal hardware error <0x%lx 0x%x>\n",
244 *((unsigned long *)&ras_log_buf), status);
247 /* Don't actually power off when debugging so we can test
248 * without actually failing while injecting errors.
249 * Error data will not be logged to syslog.
254 udbg_printf("Recoverable HW Error <0x%lx 0x%x>\n",
255 *((unsigned long *)&ras_log_buf), status);
257 "Warning: Recoverable hardware error <0x%lx 0x%x>\n",
258 *((unsigned long *)&ras_log_buf), status);
261 spin_unlock(&ras_log_buf_lock);
265 /* Get the error information for errors coming through the
266 * FWNMI vectors. The pt_regs' r3 will be updated to reflect
267 * the actual r3 if possible, and a ptr to the error log entry
268 * will be returned if found.
270 * The mce_data_buf does not have any locks or protection around it,
271 * if a second machine check comes in, or a system reset is done
272 * before we have logged the error, then we will get corruption in the
273 * error log. This is preferable over holding off on calling
274 * ibm,nmi-interlock which would result in us checkstopping if a
275 * second machine check did come in.
277 static struct rtas_error_log *fwnmi_get_errinfo(struct pt_regs *regs)
279 unsigned long errdata = regs->gpr[3];
280 struct rtas_error_log *errhdr = NULL;
281 unsigned long *savep;
283 if ((errdata >= 0x7000 && errdata < 0x7fff0) ||
284 (errdata >= rtas.base && errdata < rtas.base + rtas.size - 16)) {
285 savep = __va(errdata);
286 regs->gpr[3] = savep[0]; /* restore original r3 */
287 memset(mce_data_buf, 0, RTAS_ERROR_LOG_MAX);
288 memcpy(mce_data_buf, (char *)(savep + 1), RTAS_ERROR_LOG_MAX);
289 errhdr = (struct rtas_error_log *)mce_data_buf;
291 printk("FWNMI: corrupt r3\n");
296 /* Call this when done with the data returned by FWNMI_get_errinfo.
297 * It will release the saved data area for other CPUs in the
298 * partition to receive FWNMI errors.
300 static void fwnmi_release_errinfo(void)
302 int ret = rtas_call(rtas_token("ibm,nmi-interlock"), 0, 1, NULL);
304 printk("FWNMI: nmi-interlock failed: %d\n", ret);
307 int pSeries_system_reset_exception(struct pt_regs *regs)
310 struct rtas_error_log *errhdr = fwnmi_get_errinfo(regs);
312 /* XXX Should look at FWNMI information */
314 fwnmi_release_errinfo();
316 return 0; /* need to perform reset */
320 * See if we can recover from a machine check exception.
321 * This is only called on power4 (or above) and only via
322 * the Firmware Non-Maskable Interrupts (fwnmi) handler
323 * which provides the error analysis for us.
325 * Return 1 if corrected (or delivered a signal).
326 * Return 0 if there is nothing we can do.
328 static int recover_mce(struct pt_regs *regs, struct rtas_error_log * err)
332 if (err->disposition == RTAS_DISP_FULLY_RECOVERED) {
333 /* Platform corrected itself */
335 } else if ((regs->msr & MSR_RI) &&
337 err->severity == RTAS_SEVERITY_ERROR_SYNC &&
338 err->disposition == RTAS_DISP_NOT_RECOVERED &&
339 err->target == RTAS_TARGET_MEMORY &&
340 err->type == RTAS_TYPE_ECC_UNCORR &&
341 !(current->pid == 0 || current->pid == 1)) {
342 /* Kill off a user process with an ECC error */
343 printk(KERN_ERR "MCE: uncorrectable ecc error for pid %d\n",
345 /* XXX something better for ECC error? */
346 _exception(SIGBUS, regs, BUS_ADRERR, regs->nip);
350 log_error((char *)err, ERR_TYPE_RTAS_LOG, !nonfatal);
356 * Handle a machine check.
358 * Note that on Power 4 and beyond Firmware Non-Maskable Interrupts (fwnmi)
359 * should be present. If so the handler which called us tells us if the
360 * error was recovered (never true if RI=0).
362 * On hardware prior to Power 4 these exceptions were asynchronous which
363 * means we can't tell exactly where it occurred and so we can't recover.
365 int pSeries_machine_check_exception(struct pt_regs *regs)
367 struct rtas_error_log *errp;
370 errp = fwnmi_get_errinfo(regs);
371 fwnmi_release_errinfo();
372 if (errp && recover_mce(regs, errp))