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/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> 40 41 #include <asm/uaccess.h> 42 #include <asm/system.h> 43 #include <asm/io.h> 44 #include <asm/pgtable.h> 45 #include <asm/irq.h> 46 #include <asm/cache.h> 47 #include <asm/prom.h> 48 #include <asm/ptrace.h> 49 #include <asm/machdep.h> 50 #include <asm/rtas.h> 51 #include <asm/udbg.h> 52 #include <asm/firmware.h> 53 54 static unsigned char ras_log_buf[RTAS_ERROR_LOG_MAX]; 55 static DEFINE_SPINLOCK(ras_log_buf_lock); 56 57 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 struct pt_regs * regs); 68 static irqreturn_t ras_error_interrupt(int irq, void *dev_id, 69 struct pt_regs * regs); 70 71 /* #define DEBUG */ 72 73 static void request_ras_irqs(struct device_node *np, char *propname, 74 irqreturn_t (*handler)(int, void *, struct pt_regs *), 75 const char *name) 76 { 77 unsigned int *ireg, len, i; 78 int virq, n_intr; 79 80 ireg = (unsigned int *)get_property(np, propname, &len); 81 if (ireg == NULL) 82 return; 83 n_intr = prom_n_intr_cells(np); 84 len /= n_intr * sizeof(*ireg); 85 86 for (i = 0; i < len; i++) { 87 virq = virt_irq_create_mapping(*ireg); 88 if (virq == NO_IRQ) { 89 printk(KERN_ERR "Unable to allocate interrupt " 90 "number for %s\n", np->full_name); 91 return; 92 } 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); 96 return; 97 } 98 ireg += n_intr; 99 } 100 } 101 102 /* 103 * Initialize handlers for the set of interrupts caused by hardware errors 104 * and power system events. 105 */ 106 static int __init init_ras_IRQ(void) 107 { 108 struct device_node *np; 109 110 ras_get_sensor_state_token = rtas_token("get-sensor-state"); 111 ras_check_exception_token = rtas_token("check-exception"); 112 113 /* Internal Errors */ 114 np = of_find_node_by_path("/event-sources/internal-errors"); 115 if (np != NULL) { 116 request_ras_irqs(np, "open-pic-interrupt", ras_error_interrupt, 117 "RAS_ERROR"); 118 request_ras_irqs(np, "interrupts", ras_error_interrupt, 119 "RAS_ERROR"); 120 of_node_put(np); 121 } 122 123 /* EPOW Events */ 124 np = of_find_node_by_path("/event-sources/epow-events"); 125 if (np != NULL) { 126 request_ras_irqs(np, "open-pic-interrupt", ras_epow_interrupt, 127 "RAS_EPOW"); 128 request_ras_irqs(np, "interrupts", ras_epow_interrupt, 129 "RAS_EPOW"); 130 of_node_put(np); 131 } 132 133 return 1; 134 } 135 __initcall(init_ras_IRQ); 136 137 /* 138 * Handle power subsystem events (EPOW). 139 * 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. 143 */ 144 static irqreturn_t 145 ras_epow_interrupt(int irq, void *dev_id, struct pt_regs * regs) 146 { 147 int status = 0xdeadbeef; 148 int state = 0; 149 int critical; 150 151 status = rtas_call(ras_get_sensor_state_token, 2, 2, &state, 152 EPOW_SENSOR_TOKEN, EPOW_SENSOR_INDEX); 153 154 if (state > 3) 155 critical = 1; /* Time Critical */ 156 else 157 critical = 0; 158 159 spin_lock(&ras_log_buf_lock); 160 161 status = rtas_call(ras_check_exception_token, 6, 1, NULL, 162 RAS_VECTOR_OFFSET, 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()); 167 168 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); 172 173 /* format and print the extended information */ 174 log_error(ras_log_buf, ERR_TYPE_RTAS_LOG, 0); 175 176 spin_unlock(&ras_log_buf_lock); 177 return IRQ_HANDLED; 178 } 179 180 /* 181 * Handle hardware error interrupts. 182 * 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. 187 */ 188 static irqreturn_t 189 ras_error_interrupt(int irq, void *dev_id, struct pt_regs * regs) 190 { 191 struct rtas_error_log *rtas_elog; 192 int status = 0xdeadbeef; 193 int fatal; 194 195 spin_lock(&ras_log_buf_lock); 196 197 status = rtas_call(ras_check_exception_token, 6, 1, NULL, 198 RAS_VECTOR_OFFSET, 199 virt_irq_to_real(irq_offset_down(irq)), 200 RTAS_INTERNAL_ERROR, 1 /*Time Critical */, 201 __pa(&ras_log_buf), 202 rtas_get_error_log_max()); 203 204 rtas_elog = (struct rtas_error_log *)ras_log_buf; 205 206 if ((status == 0) && (rtas_elog->severity >= RTAS_SEVERITY_ERROR_SYNC)) 207 fatal = 1; 208 else 209 fatal = 0; 210 211 /* format and print the extended information */ 212 log_error(ras_log_buf, ERR_TYPE_RTAS_LOG, fatal); 213 214 if (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); 219 220 #ifndef DEBUG 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. 224 */ 225 ppc_md.power_off(); 226 #endif 227 } else { 228 udbg_printf("Recoverable HW Error <0x%lx 0x%x>\n", 229 *((unsigned long *)&ras_log_buf), status); 230 printk(KERN_WARNING 231 "Warning: Recoverable hardware error <0x%lx 0x%x>\n", 232 *((unsigned long *)&ras_log_buf), status); 233 } 234 235 spin_unlock(&ras_log_buf_lock); 236 return IRQ_HANDLED; 237 } 238 239 /* 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. 243 * 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. 250 */ 251 static struct rtas_error_log *fwnmi_get_errinfo(struct pt_regs *regs) 252 { 253 unsigned long errdata = regs->gpr[3]; 254 struct rtas_error_log *errhdr = NULL; 255 unsigned long *savep; 256 257 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; 264 } else { 265 printk("FWNMI: corrupt r3\n"); 266 } 267 return errhdr; 268 } 269 270 /* 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. 273 */ 274 static void fwnmi_release_errinfo(void) 275 { 276 int ret = rtas_call(rtas_token("ibm,nmi-interlock"), 0, 1, NULL); 277 if (ret != 0) 278 printk("FWNMI: nmi-interlock failed: %d\n", ret); 279 } 280 281 void pSeries_system_reset_exception(struct pt_regs *regs) 282 { 283 if (fwnmi_active) { 284 struct rtas_error_log *errhdr = fwnmi_get_errinfo(regs); 285 if (errhdr) { 286 /* XXX Should look at FWNMI information */ 287 } 288 fwnmi_release_errinfo(); 289 } 290 } 291 292 /* 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. 297 * 298 * Return 1 if corrected (or delivered a signal). 299 * Return 0 if there is nothing we can do. 300 */ 301 static int recover_mce(struct pt_regs *regs, struct rtas_error_log * err) 302 { 303 int nonfatal = 0; 304 305 if (err->disposition == RTAS_DISP_FULLY_RECOVERED) { 306 /* Platform corrected itself */ 307 nonfatal = 1; 308 } else if ((regs->msr & MSR_RI) && 309 user_mode(regs) && 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", 317 current->pid); 318 /* XXX something better for ECC error? */ 319 _exception(SIGBUS, regs, BUS_ADRERR, regs->nip); 320 nonfatal = 1; 321 } 322 323 log_error((char *)err, ERR_TYPE_RTAS_LOG, !nonfatal); 324 325 return nonfatal; 326 } 327 328 /* 329 * Handle a machine check. 330 * 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). 334 * 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. 337 */ 338 int pSeries_machine_check_exception(struct pt_regs *regs) 339 { 340 struct rtas_error_log *errp; 341 342 if (fwnmi_active) { 343 errp = fwnmi_get_errinfo(regs); 344 fwnmi_release_errinfo(); 345 if (errp && recover_mce(regs, errp)) 346 return 1; 347 } 348 349 return 0; 350 } 351