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>
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 static int ras_get_sensor_state_token;
60 static int ras_check_exception_token;
62 #define EPOW_SENSOR_TOKEN 9
63 #define EPOW_SENSOR_INDEX 0
64 #define RAS_VECTOR_OFFSET 0x500
66 static irqreturn_t ras_epow_interrupt(int irq, void *dev_id,
67 struct pt_regs * regs);
68 static irqreturn_t ras_error_interrupt(int irq, void *dev_id,
69 struct pt_regs * regs);
73 static void request_ras_irqs(struct device_node *np, char *propname,
74 irqreturn_t (*handler)(int, void *, struct pt_regs *),
77 unsigned int *ireg, len, i;
80 ireg = (unsigned int *)get_property(np, propname, &len);
83 n_intr = prom_n_intr_cells(np);
84 len /= n_intr * sizeof(*ireg);
86 for (i = 0; i < len; i++) {
87 virq = virt_irq_create_mapping(*ireg);
89 printk(KERN_ERR "Unable to allocate interrupt "
90 "number for %s\n", np->full_name);
93 if (request_irq(irq_offset_up(virq), handler, 0, name, NULL)) {
94 printk(KERN_ERR "Unable to request interrupt %d for "
95 "%s\n", irq_offset_up(virq), np->full_name);
103 * Initialize handlers for the set of interrupts caused by hardware errors
104 * and power system events.
106 static int __init init_ras_IRQ(void)
108 struct device_node *np;
110 ras_get_sensor_state_token = rtas_token("get-sensor-state");
111 ras_check_exception_token = rtas_token("check-exception");
113 /* Internal Errors */
114 np = of_find_node_by_path("/event-sources/internal-errors");
116 request_ras_irqs(np, "open-pic-interrupt", ras_error_interrupt,
118 request_ras_irqs(np, "interrupts", ras_error_interrupt,
124 np = of_find_node_by_path("/event-sources/epow-events");
126 request_ras_irqs(np, "open-pic-interrupt", ras_epow_interrupt,
128 request_ras_irqs(np, "interrupts", ras_epow_interrupt,
135 __initcall(init_ras_IRQ);
138 * Handle power subsystem events (EPOW).
140 * Presently we just log the event has occurred. This should be fixed
141 * to examine the type of power failure and take appropriate action where
142 * the time horizon permits something useful to be done.
145 ras_epow_interrupt(int irq, void *dev_id, struct pt_regs * regs)
147 int status = 0xdeadbeef;
151 status = rtas_call(ras_get_sensor_state_token, 2, 2, &state,
152 EPOW_SENSOR_TOKEN, EPOW_SENSOR_INDEX);
155 critical = 1; /* Time Critical */
159 spin_lock(&ras_log_buf_lock);
161 status = rtas_call(ras_check_exception_token, 6, 1, NULL,
163 virt_irq_to_real(irq_offset_down(irq)),
164 RTAS_EPOW_WARNING | RTAS_POWERMGM_EVENTS,
165 critical, __pa(&ras_log_buf),
166 rtas_get_error_log_max());
168 udbg_printf("EPOW <0x%lx 0x%x 0x%x>\n",
169 *((unsigned long *)&ras_log_buf), status, state);
170 printk(KERN_WARNING "EPOW <0x%lx 0x%x 0x%x>\n",
171 *((unsigned long *)&ras_log_buf), status, state);
173 /* format and print the extended information */
174 log_error(ras_log_buf, ERR_TYPE_RTAS_LOG, 0);
176 spin_unlock(&ras_log_buf_lock);
181 * Handle hardware error interrupts.
183 * RTAS check-exception is called to collect data on the exception. If
184 * the error is deemed recoverable, we log a warning and return.
185 * For nonrecoverable errors, an error is logged and we stop all processing
186 * as quickly as possible in order to prevent propagation of the failure.
189 ras_error_interrupt(int irq, void *dev_id, struct pt_regs * regs)
191 struct rtas_error_log *rtas_elog;
192 int status = 0xdeadbeef;
195 spin_lock(&ras_log_buf_lock);
197 status = rtas_call(ras_check_exception_token, 6, 1, NULL,
199 virt_irq_to_real(irq_offset_down(irq)),
200 RTAS_INTERNAL_ERROR, 1 /*Time Critical */,
202 rtas_get_error_log_max());
204 rtas_elog = (struct rtas_error_log *)ras_log_buf;
206 if ((status == 0) && (rtas_elog->severity >= RTAS_SEVERITY_ERROR_SYNC))
211 /* format and print the extended information */
212 log_error(ras_log_buf, ERR_TYPE_RTAS_LOG, fatal);
215 udbg_printf("Fatal HW Error <0x%lx 0x%x>\n",
216 *((unsigned long *)&ras_log_buf), status);
217 printk(KERN_EMERG "Error: Fatal hardware error <0x%lx 0x%x>\n",
218 *((unsigned long *)&ras_log_buf), status);
221 /* Don't actually power off when debugging so we can test
222 * without actually failing while injecting errors.
223 * Error data will not be logged to syslog.
228 udbg_printf("Recoverable HW Error <0x%lx 0x%x>\n",
229 *((unsigned long *)&ras_log_buf), status);
231 "Warning: Recoverable hardware error <0x%lx 0x%x>\n",
232 *((unsigned long *)&ras_log_buf), status);
235 spin_unlock(&ras_log_buf_lock);
239 /* Get the error information for errors coming through the
240 * FWNMI vectors. The pt_regs' r3 will be updated to reflect
241 * the actual r3 if possible, and a ptr to the error log entry
242 * will be returned if found.
244 * The mce_data_buf does not have any locks or protection around it,
245 * if a second machine check comes in, or a system reset is done
246 * before we have logged the error, then we will get corruption in the
247 * error log. This is preferable over holding off on calling
248 * ibm,nmi-interlock which would result in us checkstopping if a
249 * second machine check did come in.
251 static struct rtas_error_log *fwnmi_get_errinfo(struct pt_regs *regs)
253 unsigned long errdata = regs->gpr[3];
254 struct rtas_error_log *errhdr = NULL;
255 unsigned long *savep;
257 if ((errdata >= 0x7000 && errdata < 0x7fff0) ||
258 (errdata >= rtas.base && errdata < rtas.base + rtas.size - 16)) {
259 savep = __va(errdata);
260 regs->gpr[3] = savep[0]; /* restore original r3 */
261 memset(mce_data_buf, 0, RTAS_ERROR_LOG_MAX);
262 memcpy(mce_data_buf, (char *)(savep + 1), RTAS_ERROR_LOG_MAX);
263 errhdr = (struct rtas_error_log *)mce_data_buf;
265 printk("FWNMI: corrupt r3\n");
270 /* Call this when done with the data returned by FWNMI_get_errinfo.
271 * It will release the saved data area for other CPUs in the
272 * partition to receive FWNMI errors.
274 static void fwnmi_release_errinfo(void)
276 int ret = rtas_call(rtas_token("ibm,nmi-interlock"), 0, 1, NULL);
278 printk("FWNMI: nmi-interlock failed: %d\n", ret);
281 void pSeries_system_reset_exception(struct pt_regs *regs)
284 struct rtas_error_log *errhdr = fwnmi_get_errinfo(regs);
286 /* XXX Should look at FWNMI information */
288 fwnmi_release_errinfo();
293 * See if we can recover from a machine check exception.
294 * This is only called on power4 (or above) and only via
295 * the Firmware Non-Maskable Interrupts (fwnmi) handler
296 * which provides the error analysis for us.
298 * Return 1 if corrected (or delivered a signal).
299 * Return 0 if there is nothing we can do.
301 static int recover_mce(struct pt_regs *regs, struct rtas_error_log * err)
305 if (err->disposition == RTAS_DISP_FULLY_RECOVERED) {
306 /* Platform corrected itself */
308 } else if ((regs->msr & MSR_RI) &&
310 err->severity == RTAS_SEVERITY_ERROR_SYNC &&
311 err->disposition == RTAS_DISP_NOT_RECOVERED &&
312 err->target == RTAS_TARGET_MEMORY &&
313 err->type == RTAS_TYPE_ECC_UNCORR &&
314 !(current->pid == 0 || current->pid == 1)) {
315 /* Kill off a user process with an ECC error */
316 printk(KERN_ERR "MCE: uncorrectable ecc error for pid %d\n",
318 /* XXX something better for ECC error? */
319 _exception(SIGBUS, regs, BUS_ADRERR, regs->nip);
323 log_error((char *)err, ERR_TYPE_RTAS_LOG, !nonfatal);
329 * Handle a machine check.
331 * Note that on Power 4 and beyond Firmware Non-Maskable Interrupts (fwnmi)
332 * should be present. If so the handler which called us tells us if the
333 * error was recovered (never true if RI=0).
335 * On hardware prior to Power 4 these exceptions were asynchronous which
336 * means we can't tell exactly where it occurred and so we can't recover.
338 int pSeries_machine_check_exception(struct pt_regs *regs)
340 struct rtas_error_log *errp;
343 errp = fwnmi_get_errinfo(regs);
344 fwnmi_release_errinfo();
345 if (errp && recover_mce(regs, errp))