3 * Copyright (C) 2001 Dave Engebretsen IBM Corporation
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
21 * 2001/09/21 : engebret : Created with minimal EPOW and HW exception support.
25 #include <linux/errno.h>
26 #include <linux/threads.h>
27 #include <linux/kernel_stat.h>
28 #include <linux/signal.h>
29 #include <linux/sched.h>
30 #include <linux/ioport.h>
31 #include <linux/interrupt.h>
32 #include <linux/timex.h>
33 #include <linux/init.h>
34 #include <linux/slab.h>
35 #include <linux/pci.h>
36 #include <linux/delay.h>
37 #include <linux/irq.h>
38 #include <linux/random.h>
39 #include <linux/sysrq.h>
40 #include <linux/bitops.h>
42 #include <asm/uaccess.h>
43 #include <asm/system.h>
45 #include <asm/pgtable.h>
47 #include <asm/cache.h>
49 #include <asm/ptrace.h>
50 #include <asm/machdep.h>
52 #include <asm/ppcdebug.h>
54 static unsigned char ras_log_buf[RTAS_ERROR_LOG_MAX];
55 static DEFINE_SPINLOCK(ras_log_buf_lock);
57 char mce_data_buf[RTAS_ERROR_LOG_MAX]
59 /* This is true if we are using the firmware NMI handler (typically LPAR) */
60 extern int fwnmi_active;
62 extern void _exception(int signr, struct pt_regs *regs, int code, unsigned long addr);
64 static int ras_get_sensor_state_token;
65 static int ras_check_exception_token;
67 #define EPOW_SENSOR_TOKEN 9
68 #define EPOW_SENSOR_INDEX 0
69 #define RAS_VECTOR_OFFSET 0x500
71 static irqreturn_t ras_epow_interrupt(int irq, void *dev_id,
72 struct pt_regs * regs);
73 static irqreturn_t ras_error_interrupt(int irq, void *dev_id,
74 struct pt_regs * regs);
78 static void request_ras_irqs(struct device_node *np, char *propname,
79 irqreturn_t (*handler)(int, void *, struct pt_regs *),
82 unsigned int *ireg, len, i;
85 ireg = (unsigned int *)get_property(np, propname, &len);
88 n_intr = prom_n_intr_cells(np);
89 len /= n_intr * sizeof(*ireg);
91 for (i = 0; i < len; i++) {
92 virq = virt_irq_create_mapping(*ireg);
94 printk(KERN_ERR "Unable to allocate interrupt "
95 "number for %s\n", np->full_name);
98 if (request_irq(irq_offset_up(virq), handler, 0, name, NULL)) {
99 printk(KERN_ERR "Unable to request interrupt %d for "
100 "%s\n", irq_offset_up(virq), np->full_name);
108 * Initialize handlers for the set of interrupts caused by hardware errors
109 * and power system events.
111 static int __init init_ras_IRQ(void)
113 struct device_node *np;
115 ras_get_sensor_state_token = rtas_token("get-sensor-state");
116 ras_check_exception_token = rtas_token("check-exception");
118 /* Internal Errors */
119 np = of_find_node_by_path("/event-sources/internal-errors");
121 request_ras_irqs(np, "open-pic-interrupt", ras_error_interrupt,
123 request_ras_irqs(np, "interrupts", ras_error_interrupt,
129 np = of_find_node_by_path("/event-sources/epow-events");
131 request_ras_irqs(np, "open-pic-interrupt", ras_epow_interrupt,
133 request_ras_irqs(np, "interrupts", ras_epow_interrupt,
140 __initcall(init_ras_IRQ);
143 * Handle power subsystem events (EPOW).
145 * Presently we just log the event has occurred. This should be fixed
146 * to examine the type of power failure and take appropriate action where
147 * the time horizon permits something useful to be done.
150 ras_epow_interrupt(int irq, void *dev_id, struct pt_regs * regs)
152 int status = 0xdeadbeef;
156 status = rtas_call(ras_get_sensor_state_token, 2, 2, &state,
157 EPOW_SENSOR_TOKEN, EPOW_SENSOR_INDEX);
160 critical = 1; /* Time Critical */
164 spin_lock(&ras_log_buf_lock);
166 status = rtas_call(ras_check_exception_token, 6, 1, NULL,
168 virt_irq_to_real(irq_offset_down(irq)),
169 RTAS_EPOW_WARNING | RTAS_POWERMGM_EVENTS,
170 critical, __pa(&ras_log_buf),
171 rtas_get_error_log_max());
173 udbg_printf("EPOW <0x%lx 0x%x 0x%x>\n",
174 *((unsigned long *)&ras_log_buf), status, state);
175 printk(KERN_WARNING "EPOW <0x%lx 0x%x 0x%x>\n",
176 *((unsigned long *)&ras_log_buf), status, state);
178 /* format and print the extended information */
179 log_error(ras_log_buf, ERR_TYPE_RTAS_LOG, 0);
181 spin_unlock(&ras_log_buf_lock);
186 * Handle hardware error interrupts.
188 * RTAS check-exception is called to collect data on the exception. If
189 * the error is deemed recoverable, we log a warning and return.
190 * For nonrecoverable errors, an error is logged and we stop all processing
191 * as quickly as possible in order to prevent propagation of the failure.
194 ras_error_interrupt(int irq, void *dev_id, struct pt_regs * regs)
196 struct rtas_error_log *rtas_elog;
197 int status = 0xdeadbeef;
200 spin_lock(&ras_log_buf_lock);
202 status = rtas_call(ras_check_exception_token, 6, 1, NULL,
204 virt_irq_to_real(irq_offset_down(irq)),
205 RTAS_INTERNAL_ERROR, 1 /*Time Critical */,
207 rtas_get_error_log_max());
209 rtas_elog = (struct rtas_error_log *)ras_log_buf;
211 if ((status == 0) && (rtas_elog->severity >= RTAS_SEVERITY_ERROR_SYNC))
216 /* format and print the extended information */
217 log_error(ras_log_buf, ERR_TYPE_RTAS_LOG, fatal);
220 udbg_printf("Fatal HW Error <0x%lx 0x%x>\n",
221 *((unsigned long *)&ras_log_buf), status);
222 printk(KERN_EMERG "Error: Fatal hardware error <0x%lx 0x%x>\n",
223 *((unsigned long *)&ras_log_buf), status);
226 /* Don't actually power off when debugging so we can test
227 * without actually failing while injecting errors.
228 * Error data will not be logged to syslog.
233 udbg_printf("Recoverable HW Error <0x%lx 0x%x>\n",
234 *((unsigned long *)&ras_log_buf), status);
236 "Warning: Recoverable hardware error <0x%lx 0x%x>\n",
237 *((unsigned long *)&ras_log_buf), status);
240 spin_unlock(&ras_log_buf_lock);
244 /* Get the error information for errors coming through the
245 * FWNMI vectors. The pt_regs' r3 will be updated to reflect
246 * the actual r3 if possible, and a ptr to the error log entry
247 * will be returned if found.
249 * The mce_data_buf does not have any locks or protection around it,
250 * if a second machine check comes in, or a system reset is done
251 * before we have logged the error, then we will get corruption in the
252 * error log. This is preferable over holding off on calling
253 * ibm,nmi-interlock which would result in us checkstopping if a
254 * second machine check did come in.
256 static struct rtas_error_log *fwnmi_get_errinfo(struct pt_regs *regs)
258 unsigned long errdata = regs->gpr[3];
259 struct rtas_error_log *errhdr = NULL;
260 unsigned long *savep;
262 if ((errdata >= 0x7000 && errdata < 0x7fff0) ||
263 (errdata >= rtas.base && errdata < rtas.base + rtas.size - 16)) {
264 savep = __va(errdata);
265 regs->gpr[3] = savep[0]; /* restore original r3 */
266 memset(mce_data_buf, 0, RTAS_ERROR_LOG_MAX);
267 memcpy(mce_data_buf, (char *)(savep + 1), RTAS_ERROR_LOG_MAX);
268 errhdr = (struct rtas_error_log *)mce_data_buf;
270 printk("FWNMI: corrupt r3\n");
275 /* Call this when done with the data returned by FWNMI_get_errinfo.
276 * It will release the saved data area for other CPUs in the
277 * partition to receive FWNMI errors.
279 static void fwnmi_release_errinfo(void)
281 int ret = rtas_call(rtas_token("ibm,nmi-interlock"), 0, 1, NULL);
283 printk("FWNMI: nmi-interlock failed: %d\n", ret);
286 void pSeries_system_reset_exception(struct pt_regs *regs)
289 struct rtas_error_log *errhdr = fwnmi_get_errinfo(regs);
291 /* XXX Should look at FWNMI information */
293 fwnmi_release_errinfo();
298 * See if we can recover from a machine check exception.
299 * This is only called on power4 (or above) and only via
300 * the Firmware Non-Maskable Interrupts (fwnmi) handler
301 * which provides the error analysis for us.
303 * Return 1 if corrected (or delivered a signal).
304 * Return 0 if there is nothing we can do.
306 static int recover_mce(struct pt_regs *regs, struct rtas_error_log * err)
310 if (err->disposition == RTAS_DISP_FULLY_RECOVERED) {
311 /* Platform corrected itself */
313 } else if ((regs->msr & MSR_RI) &&
315 err->severity == RTAS_SEVERITY_ERROR_SYNC &&
316 err->disposition == RTAS_DISP_NOT_RECOVERED &&
317 err->target == RTAS_TARGET_MEMORY &&
318 err->type == RTAS_TYPE_ECC_UNCORR &&
319 !(current->pid == 0 || current->pid == 1)) {
320 /* Kill off a user process with an ECC error */
321 printk(KERN_ERR "MCE: uncorrectable ecc error for pid %d\n",
323 /* XXX something better for ECC error? */
324 _exception(SIGBUS, regs, BUS_ADRERR, regs->nip);
328 log_error((char *)err, ERR_TYPE_RTAS_LOG, !nonfatal);
334 * Handle a machine check.
336 * Note that on Power 4 and beyond Firmware Non-Maskable Interrupts (fwnmi)
337 * should be present. If so the handler which called us tells us if the
338 * error was recovered (never true if RI=0).
340 * On hardware prior to Power 4 these exceptions were asynchronous which
341 * means we can't tell exactly where it occurred and so we can't recover.
343 int pSeries_machine_check_exception(struct pt_regs *regs)
345 struct rtas_error_log *errp;
348 errp = fwnmi_get_errinfo(regs);
349 fwnmi_release_errinfo();
350 if (errp && recover_mce(regs, errp))