1 /* 2 * Copyright (C) 2001 Dave Engebretsen IBM Corporation 3 * 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. 8 * 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. 13 * 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 17 */ 18 19 /* Change Activity: 20 * 2001/09/21 : engebret : Created with minimal EPOW and HW exception support. 21 * End Change Activity 22 */ 23 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/delay.h> 34 #include <linux/irq.h> 35 #include <linux/random.h> 36 #include <linux/sysrq.h> 37 #include <linux/bitops.h> 38 39 #include <asm/uaccess.h> 40 #include <asm/system.h> 41 #include <asm/io.h> 42 #include <asm/pgtable.h> 43 #include <asm/irq.h> 44 #include <asm/cache.h> 45 #include <asm/prom.h> 46 #include <asm/ptrace.h> 47 #include <asm/machdep.h> 48 #include <asm/rtas.h> 49 #include <asm/udbg.h> 50 #include <asm/firmware.h> 51 52 #include "pseries.h" 53 54 static unsigned char ras_log_buf[RTAS_ERROR_LOG_MAX]; 55 static DEFINE_SPINLOCK(ras_log_buf_lock); 56 57 static char mce_data_buf[RTAS_ERROR_LOG_MAX]; 58 59 static int ras_get_sensor_state_token; 60 static int ras_check_exception_token; 61 62 #define EPOW_SENSOR_TOKEN 9 63 #define EPOW_SENSOR_INDEX 0 64 #define RAS_VECTOR_OFFSET 0x500 65 66 static irqreturn_t ras_epow_interrupt(int irq, void *dev_id); 67 static irqreturn_t ras_error_interrupt(int irq, void *dev_id); 68 69 70 /* 71 * Initialize handlers for the set of interrupts caused by hardware errors 72 * and power system events. 73 */ 74 static int __init init_ras_IRQ(void) 75 { 76 struct device_node *np; 77 78 ras_get_sensor_state_token = rtas_token("get-sensor-state"); 79 ras_check_exception_token = rtas_token("check-exception"); 80 81 /* Internal Errors */ 82 np = of_find_node_by_path("/event-sources/internal-errors"); 83 if (np != NULL) { 84 request_event_sources_irqs(np, ras_error_interrupt, 85 "RAS_ERROR"); 86 of_node_put(np); 87 } 88 89 /* EPOW Events */ 90 np = of_find_node_by_path("/event-sources/epow-events"); 91 if (np != NULL) { 92 request_event_sources_irqs(np, ras_epow_interrupt, "RAS_EPOW"); 93 of_node_put(np); 94 } 95 96 return 0; 97 } 98 __initcall(init_ras_IRQ); 99 100 /* 101 * Handle power subsystem events (EPOW). 102 * 103 * Presently we just log the event has occurred. This should be fixed 104 * to examine the type of power failure and take appropriate action where 105 * the time horizon permits something useful to be done. 106 */ 107 static irqreturn_t ras_epow_interrupt(int irq, void *dev_id) 108 { 109 int status = 0xdeadbeef; 110 int state = 0; 111 int critical; 112 113 status = rtas_call(ras_get_sensor_state_token, 2, 2, &state, 114 EPOW_SENSOR_TOKEN, EPOW_SENSOR_INDEX); 115 116 if (state > 3) 117 critical = 1; /* Time Critical */ 118 else 119 critical = 0; 120 121 spin_lock(&ras_log_buf_lock); 122 123 status = rtas_call(ras_check_exception_token, 6, 1, NULL, 124 RAS_VECTOR_OFFSET, 125 irq_map[irq].hwirq, 126 RTAS_EPOW_WARNING | RTAS_POWERMGM_EVENTS, 127 critical, __pa(&ras_log_buf), 128 rtas_get_error_log_max()); 129 130 udbg_printf("EPOW <0x%lx 0x%x 0x%x>\n", 131 *((unsigned long *)&ras_log_buf), status, state); 132 printk(KERN_WARNING "EPOW <0x%lx 0x%x 0x%x>\n", 133 *((unsigned long *)&ras_log_buf), status, state); 134 135 /* format and print the extended information */ 136 log_error(ras_log_buf, ERR_TYPE_RTAS_LOG, 0); 137 138 spin_unlock(&ras_log_buf_lock); 139 return IRQ_HANDLED; 140 } 141 142 /* 143 * Handle hardware error interrupts. 144 * 145 * RTAS check-exception is called to collect data on the exception. If 146 * the error is deemed recoverable, we log a warning and return. 147 * For nonrecoverable errors, an error is logged and we stop all processing 148 * as quickly as possible in order to prevent propagation of the failure. 149 */ 150 static irqreturn_t ras_error_interrupt(int irq, void *dev_id) 151 { 152 struct rtas_error_log *rtas_elog; 153 int status = 0xdeadbeef; 154 int fatal; 155 156 spin_lock(&ras_log_buf_lock); 157 158 status = rtas_call(ras_check_exception_token, 6, 1, NULL, 159 RAS_VECTOR_OFFSET, 160 irq_map[irq].hwirq, 161 RTAS_INTERNAL_ERROR, 1 /*Time Critical */, 162 __pa(&ras_log_buf), 163 rtas_get_error_log_max()); 164 165 rtas_elog = (struct rtas_error_log *)ras_log_buf; 166 167 if ((status == 0) && (rtas_elog->severity >= RTAS_SEVERITY_ERROR_SYNC)) 168 fatal = 1; 169 else 170 fatal = 0; 171 172 /* format and print the extended information */ 173 log_error(ras_log_buf, ERR_TYPE_RTAS_LOG, fatal); 174 175 if (fatal) { 176 udbg_printf("Fatal HW Error <0x%lx 0x%x>\n", 177 *((unsigned long *)&ras_log_buf), status); 178 printk(KERN_EMERG "Error: Fatal hardware error <0x%lx 0x%x>\n", 179 *((unsigned long *)&ras_log_buf), status); 180 181 #ifndef DEBUG_RTAS_POWER_OFF 182 /* Don't actually power off when debugging so we can test 183 * without actually failing while injecting errors. 184 * Error data will not be logged to syslog. 185 */ 186 ppc_md.power_off(); 187 #endif 188 } else { 189 udbg_printf("Recoverable HW Error <0x%lx 0x%x>\n", 190 *((unsigned long *)&ras_log_buf), status); 191 printk(KERN_WARNING 192 "Warning: Recoverable hardware error <0x%lx 0x%x>\n", 193 *((unsigned long *)&ras_log_buf), status); 194 } 195 196 spin_unlock(&ras_log_buf_lock); 197 return IRQ_HANDLED; 198 } 199 200 /* Get the error information for errors coming through the 201 * FWNMI vectors. The pt_regs' r3 will be updated to reflect 202 * the actual r3 if possible, and a ptr to the error log entry 203 * will be returned if found. 204 * 205 * The mce_data_buf does not have any locks or protection around it, 206 * if a second machine check comes in, or a system reset is done 207 * before we have logged the error, then we will get corruption in the 208 * error log. This is preferable over holding off on calling 209 * ibm,nmi-interlock which would result in us checkstopping if a 210 * second machine check did come in. 211 */ 212 static struct rtas_error_log *fwnmi_get_errinfo(struct pt_regs *regs) 213 { 214 unsigned long errdata = regs->gpr[3]; 215 struct rtas_error_log *errhdr = NULL; 216 unsigned long *savep; 217 218 if ((errdata >= 0x7000 && errdata < 0x7fff0) || 219 (errdata >= rtas.base && errdata < rtas.base + rtas.size - 16)) { 220 savep = __va(errdata); 221 regs->gpr[3] = savep[0]; /* restore original r3 */ 222 memset(mce_data_buf, 0, RTAS_ERROR_LOG_MAX); 223 memcpy(mce_data_buf, (char *)(savep + 1), RTAS_ERROR_LOG_MAX); 224 errhdr = (struct rtas_error_log *)mce_data_buf; 225 } else { 226 printk("FWNMI: corrupt r3\n"); 227 } 228 return errhdr; 229 } 230 231 /* Call this when done with the data returned by FWNMI_get_errinfo. 232 * It will release the saved data area for other CPUs in the 233 * partition to receive FWNMI errors. 234 */ 235 static void fwnmi_release_errinfo(void) 236 { 237 int ret = rtas_call(rtas_token("ibm,nmi-interlock"), 0, 1, NULL); 238 if (ret != 0) 239 printk("FWNMI: nmi-interlock failed: %d\n", ret); 240 } 241 242 int pSeries_system_reset_exception(struct pt_regs *regs) 243 { 244 if (fwnmi_active) { 245 struct rtas_error_log *errhdr = fwnmi_get_errinfo(regs); 246 if (errhdr) { 247 /* XXX Should look at FWNMI information */ 248 } 249 fwnmi_release_errinfo(); 250 } 251 return 0; /* need to perform reset */ 252 } 253 254 /* 255 * See if we can recover from a machine check exception. 256 * This is only called on power4 (or above) and only via 257 * the Firmware Non-Maskable Interrupts (fwnmi) handler 258 * which provides the error analysis for us. 259 * 260 * Return 1 if corrected (or delivered a signal). 261 * Return 0 if there is nothing we can do. 262 */ 263 static int recover_mce(struct pt_regs *regs, struct rtas_error_log * err) 264 { 265 int nonfatal = 0; 266 267 if (err->disposition == RTAS_DISP_FULLY_RECOVERED) { 268 /* Platform corrected itself */ 269 nonfatal = 1; 270 } else if ((regs->msr & MSR_RI) && 271 user_mode(regs) && 272 err->severity == RTAS_SEVERITY_ERROR_SYNC && 273 err->disposition == RTAS_DISP_NOT_RECOVERED && 274 err->target == RTAS_TARGET_MEMORY && 275 err->type == RTAS_TYPE_ECC_UNCORR && 276 !(current->pid == 0 || is_global_init(current))) { 277 /* Kill off a user process with an ECC error */ 278 printk(KERN_ERR "MCE: uncorrectable ecc error for pid %d\n", 279 current->pid); 280 /* XXX something better for ECC error? */ 281 _exception(SIGBUS, regs, BUS_ADRERR, regs->nip); 282 nonfatal = 1; 283 } 284 285 log_error((char *)err, ERR_TYPE_RTAS_LOG, !nonfatal); 286 287 return nonfatal; 288 } 289 290 /* 291 * Handle a machine check. 292 * 293 * Note that on Power 4 and beyond Firmware Non-Maskable Interrupts (fwnmi) 294 * should be present. If so the handler which called us tells us if the 295 * error was recovered (never true if RI=0). 296 * 297 * On hardware prior to Power 4 these exceptions were asynchronous which 298 * means we can't tell exactly where it occurred and so we can't recover. 299 */ 300 int pSeries_machine_check_exception(struct pt_regs *regs) 301 { 302 struct rtas_error_log *errp; 303 304 if (fwnmi_active) { 305 errp = fwnmi_get_errinfo(regs); 306 fwnmi_release_errinfo(); 307 if (errp && recover_mce(regs, errp)) 308 return 1; 309 } 310 311 return 0; 312 } 313