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/io.h> 41 #include <asm/pgtable.h> 42 #include <asm/irq.h> 43 #include <asm/cache.h> 44 #include <asm/prom.h> 45 #include <asm/ptrace.h> 46 #include <asm/machdep.h> 47 #include <asm/rtas.h> 48 #include <asm/udbg.h> 49 #include <asm/firmware.h> 50 51 #include "pseries.h" 52 53 static unsigned char ras_log_buf[RTAS_ERROR_LOG_MAX]; 54 static DEFINE_SPINLOCK(ras_log_buf_lock); 55 56 static char global_mce_data_buf[RTAS_ERROR_LOG_MAX]; 57 static DEFINE_PER_CPU(__u64, mce_data_buf); 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 65 static irqreturn_t ras_epow_interrupt(int irq, void *dev_id); 66 static irqreturn_t ras_error_interrupt(int irq, void *dev_id); 67 68 69 /* 70 * Initialize handlers for the set of interrupts caused by hardware errors 71 * and power system events. 72 */ 73 static int __init init_ras_IRQ(void) 74 { 75 struct device_node *np; 76 77 ras_get_sensor_state_token = rtas_token("get-sensor-state"); 78 ras_check_exception_token = rtas_token("check-exception"); 79 80 /* Internal Errors */ 81 np = of_find_node_by_path("/event-sources/internal-errors"); 82 if (np != NULL) { 83 request_event_sources_irqs(np, ras_error_interrupt, 84 "RAS_ERROR"); 85 of_node_put(np); 86 } 87 88 /* EPOW Events */ 89 np = of_find_node_by_path("/event-sources/epow-events"); 90 if (np != NULL) { 91 request_event_sources_irqs(np, ras_epow_interrupt, "RAS_EPOW"); 92 of_node_put(np); 93 } 94 95 return 0; 96 } 97 __initcall(init_ras_IRQ); 98 99 /* 100 * Handle power subsystem events (EPOW). 101 * 102 * Presently we just log the event has occurred. This should be fixed 103 * to examine the type of power failure and take appropriate action where 104 * the time horizon permits something useful to be done. 105 */ 106 static irqreturn_t ras_epow_interrupt(int irq, void *dev_id) 107 { 108 int status = 0xdeadbeef; 109 int state = 0; 110 int critical; 111 112 status = rtas_call(ras_get_sensor_state_token, 2, 2, &state, 113 EPOW_SENSOR_TOKEN, EPOW_SENSOR_INDEX); 114 115 if (state > 3) 116 critical = 1; /* Time Critical */ 117 else 118 critical = 0; 119 120 spin_lock(&ras_log_buf_lock); 121 122 status = rtas_call(ras_check_exception_token, 6, 1, NULL, 123 RTAS_VECTOR_EXTERNAL_INTERRUPT, 124 virq_to_hw(irq), 125 RTAS_EPOW_WARNING | RTAS_POWERMGM_EVENTS, 126 critical, __pa(&ras_log_buf), 127 rtas_get_error_log_max()); 128 129 udbg_printf("EPOW <0x%lx 0x%x 0x%x>\n", 130 *((unsigned long *)&ras_log_buf), status, state); 131 printk(KERN_WARNING "EPOW <0x%lx 0x%x 0x%x>\n", 132 *((unsigned long *)&ras_log_buf), status, state); 133 134 /* format and print the extended information */ 135 log_error(ras_log_buf, ERR_TYPE_RTAS_LOG, 0); 136 137 spin_unlock(&ras_log_buf_lock); 138 return IRQ_HANDLED; 139 } 140 141 /* 142 * Handle hardware error interrupts. 143 * 144 * RTAS check-exception is called to collect data on the exception. If 145 * the error is deemed recoverable, we log a warning and return. 146 * For nonrecoverable errors, an error is logged and we stop all processing 147 * as quickly as possible in order to prevent propagation of the failure. 148 */ 149 static irqreturn_t ras_error_interrupt(int irq, void *dev_id) 150 { 151 struct rtas_error_log *rtas_elog; 152 int status = 0xdeadbeef; 153 int fatal; 154 155 spin_lock(&ras_log_buf_lock); 156 157 status = rtas_call(ras_check_exception_token, 6, 1, NULL, 158 RTAS_VECTOR_EXTERNAL_INTERRUPT, 159 virq_to_hw(irq), 160 RTAS_INTERNAL_ERROR, 1 /*Time Critical */, 161 __pa(&ras_log_buf), 162 rtas_get_error_log_max()); 163 164 rtas_elog = (struct rtas_error_log *)ras_log_buf; 165 166 if ((status == 0) && (rtas_elog->severity >= RTAS_SEVERITY_ERROR_SYNC)) 167 fatal = 1; 168 else 169 fatal = 0; 170 171 /* format and print the extended information */ 172 log_error(ras_log_buf, ERR_TYPE_RTAS_LOG, fatal); 173 174 if (fatal) { 175 udbg_printf("Fatal HW Error <0x%lx 0x%x>\n", 176 *((unsigned long *)&ras_log_buf), status); 177 printk(KERN_EMERG "Error: Fatal hardware error <0x%lx 0x%x>\n", 178 *((unsigned long *)&ras_log_buf), status); 179 180 #ifndef DEBUG_RTAS_POWER_OFF 181 /* Don't actually power off when debugging so we can test 182 * without actually failing while injecting errors. 183 * Error data will not be logged to syslog. 184 */ 185 ppc_md.power_off(); 186 #endif 187 } else { 188 udbg_printf("Recoverable HW Error <0x%lx 0x%x>\n", 189 *((unsigned long *)&ras_log_buf), status); 190 printk(KERN_WARNING 191 "Warning: Recoverable hardware error <0x%lx 0x%x>\n", 192 *((unsigned long *)&ras_log_buf), status); 193 } 194 195 spin_unlock(&ras_log_buf_lock); 196 return IRQ_HANDLED; 197 } 198 199 /* 200 * Some versions of FWNMI place the buffer inside the 4kB page starting at 201 * 0x7000. Other versions place it inside the rtas buffer. We check both. 202 */ 203 #define VALID_FWNMI_BUFFER(A) \ 204 ((((A) >= 0x7000) && ((A) < 0x7ff0)) || \ 205 (((A) >= rtas.base) && ((A) < (rtas.base + rtas.size - 16)))) 206 207 /* 208 * Get the error information for errors coming through the 209 * FWNMI vectors. The pt_regs' r3 will be updated to reflect 210 * the actual r3 if possible, and a ptr to the error log entry 211 * will be returned if found. 212 * 213 * If the RTAS error is not of the extended type, then we put it in a per 214 * cpu 64bit buffer. If it is the extended type we use global_mce_data_buf. 215 * 216 * The global_mce_data_buf does not have any locks or protection around it, 217 * if a second machine check comes in, or a system reset is done 218 * before we have logged the error, then we will get corruption in the 219 * error log. This is preferable over holding off on calling 220 * ibm,nmi-interlock which would result in us checkstopping if a 221 * second machine check did come in. 222 */ 223 static struct rtas_error_log *fwnmi_get_errinfo(struct pt_regs *regs) 224 { 225 unsigned long *savep; 226 struct rtas_error_log *h, *errhdr = NULL; 227 228 if (!VALID_FWNMI_BUFFER(regs->gpr[3])) { 229 printk(KERN_ERR "FWNMI: corrupt r3 0x%016lx\n", regs->gpr[3]); 230 return NULL; 231 } 232 233 savep = __va(regs->gpr[3]); 234 regs->gpr[3] = savep[0]; /* restore original r3 */ 235 236 /* If it isn't an extended log we can use the per cpu 64bit buffer */ 237 h = (struct rtas_error_log *)&savep[1]; 238 if (!h->extended) { 239 memcpy(&__get_cpu_var(mce_data_buf), h, sizeof(__u64)); 240 errhdr = (struct rtas_error_log *)&__get_cpu_var(mce_data_buf); 241 } else { 242 int len; 243 244 len = max_t(int, 8+h->extended_log_length, RTAS_ERROR_LOG_MAX); 245 memset(global_mce_data_buf, 0, RTAS_ERROR_LOG_MAX); 246 memcpy(global_mce_data_buf, h, len); 247 errhdr = (struct rtas_error_log *)global_mce_data_buf; 248 } 249 250 return errhdr; 251 } 252 253 /* Call this when done with the data returned by FWNMI_get_errinfo. 254 * It will release the saved data area for other CPUs in the 255 * partition to receive FWNMI errors. 256 */ 257 static void fwnmi_release_errinfo(void) 258 { 259 int ret = rtas_call(rtas_token("ibm,nmi-interlock"), 0, 1, NULL); 260 if (ret != 0) 261 printk(KERN_ERR "FWNMI: nmi-interlock failed: %d\n", ret); 262 } 263 264 int pSeries_system_reset_exception(struct pt_regs *regs) 265 { 266 if (fwnmi_active) { 267 struct rtas_error_log *errhdr = fwnmi_get_errinfo(regs); 268 if (errhdr) { 269 /* XXX Should look at FWNMI information */ 270 } 271 fwnmi_release_errinfo(); 272 } 273 return 0; /* need to perform reset */ 274 } 275 276 /* 277 * See if we can recover from a machine check exception. 278 * This is only called on power4 (or above) and only via 279 * the Firmware Non-Maskable Interrupts (fwnmi) handler 280 * which provides the error analysis for us. 281 * 282 * Return 1 if corrected (or delivered a signal). 283 * Return 0 if there is nothing we can do. 284 */ 285 static int recover_mce(struct pt_regs *regs, struct rtas_error_log *err) 286 { 287 int recovered = 0; 288 289 if (!(regs->msr & MSR_RI)) { 290 /* If MSR_RI isn't set, we cannot recover */ 291 recovered = 0; 292 293 } else if (err->disposition == RTAS_DISP_FULLY_RECOVERED) { 294 /* Platform corrected itself */ 295 recovered = 1; 296 297 } else if (err->disposition == RTAS_DISP_LIMITED_RECOVERY) { 298 /* Platform corrected itself but could be degraded */ 299 printk(KERN_ERR "MCE: limited recovery, system may " 300 "be degraded\n"); 301 recovered = 1; 302 303 } else if (user_mode(regs) && !is_global_init(current) && 304 err->severity == RTAS_SEVERITY_ERROR_SYNC) { 305 306 /* 307 * If we received a synchronous error when in userspace 308 * kill the task. Firmware may report details of the fail 309 * asynchronously, so we can't rely on the target and type 310 * fields being valid here. 311 */ 312 printk(KERN_ERR "MCE: uncorrectable error, killing task " 313 "%s:%d\n", current->comm, current->pid); 314 315 _exception(SIGBUS, regs, BUS_MCEERR_AR, regs->nip); 316 recovered = 1; 317 } 318 319 log_error((char *)err, ERR_TYPE_RTAS_LOG, 0); 320 321 return recovered; 322 } 323 324 /* 325 * Handle a machine check. 326 * 327 * Note that on Power 4 and beyond Firmware Non-Maskable Interrupts (fwnmi) 328 * should be present. If so the handler which called us tells us if the 329 * error was recovered (never true if RI=0). 330 * 331 * On hardware prior to Power 4 these exceptions were asynchronous which 332 * means we can't tell exactly where it occurred and so we can't recover. 333 */ 334 int pSeries_machine_check_exception(struct pt_regs *regs) 335 { 336 struct rtas_error_log *errp; 337 338 if (fwnmi_active) { 339 errp = fwnmi_get_errinfo(regs); 340 fwnmi_release_errinfo(); 341 if (errp && recover_mce(regs, errp)) 342 return 1; 343 } 344 345 return 0; 346 } 347