1/* 2 * Bus error event handling code for systems equipped with ECC 3 * handling logic, i.e. DECstation/DECsystem 5000/200 (KN02), 4 * 5000/240 (KN03), 5000/260 (KN05) and DECsystem 5900 (KN03), 5 * 5900/260 (KN05) systems. 6 * 7 * Copyright (c) 2003, 2005 Maciej W. Rozycki 8 * 9 * This program is free software; you can redistribute it and/or 10 * modify it under the terms of the GNU General Public License 11 * as published by the Free Software Foundation; either version 12 * 2 of the License, or (at your option) any later version. 13 */ 14 15#include <linux/init.h> 16#include <linux/interrupt.h> 17#include <linux/kernel.h> 18#include <linux/sched.h> 19#include <linux/types.h> 20 21#include <asm/addrspace.h> 22#include <asm/bootinfo.h> 23#include <asm/cpu.h> 24#include <asm/cpu-type.h> 25#include <asm/irq_regs.h> 26#include <asm/processor.h> 27#include <asm/ptrace.h> 28#include <asm/traps.h> 29 30#include <asm/dec/ecc.h> 31#include <asm/dec/kn02.h> 32#include <asm/dec/kn03.h> 33#include <asm/dec/kn05.h> 34 35static volatile u32 *kn0x_erraddr; 36static volatile u32 *kn0x_chksyn; 37 38static inline void dec_ecc_be_ack(void) 39{ 40 *kn0x_erraddr = 0; /* any write clears the IRQ */ 41 iob(); 42} 43 44static int dec_ecc_be_backend(struct pt_regs *regs, int is_fixup, int invoker) 45{ 46 static const char excstr[] = "exception"; 47 static const char intstr[] = "interrupt"; 48 static const char cpustr[] = "CPU"; 49 static const char dmastr[] = "DMA"; 50 static const char readstr[] = "read"; 51 static const char mreadstr[] = "memory read"; 52 static const char writestr[] = "write"; 53 static const char mwritstr[] = "partial memory write"; 54 static const char timestr[] = "timeout"; 55 static const char overstr[] = "overrun"; 56 static const char eccstr[] = "ECC error"; 57 58 const char *kind, *agent, *cycle, *event; 59 const char *status = "", *xbit = "", *fmt = ""; 60 unsigned long address; 61 u16 syn = 0, sngl; 62 63 int i = 0; 64 65 u32 erraddr = *kn0x_erraddr; 66 u32 chksyn = *kn0x_chksyn; 67 int action = MIPS_BE_FATAL; 68 69 /* For non-ECC ack ASAP, so that any subsequent errors get caught. */ 70 if ((erraddr & (KN0X_EAR_VALID | KN0X_EAR_ECCERR)) == KN0X_EAR_VALID) 71 dec_ecc_be_ack(); 72 73 kind = invoker ? intstr : excstr; 74 75 if (!(erraddr & KN0X_EAR_VALID)) { 76 /* No idea what happened. */ 77 printk(KERN_ALERT "Unidentified bus error %s\n", kind); 78 return action; 79 } 80 81 agent = (erraddr & KN0X_EAR_CPU) ? cpustr : dmastr; 82 83 if (erraddr & KN0X_EAR_ECCERR) { 84 /* An ECC error on a CPU or DMA transaction. */ 85 cycle = (erraddr & KN0X_EAR_WRITE) ? mwritstr : mreadstr; 86 event = eccstr; 87 } else { 88 /* A CPU timeout or a DMA overrun. */ 89 cycle = (erraddr & KN0X_EAR_WRITE) ? writestr : readstr; 90 event = (erraddr & KN0X_EAR_CPU) ? timestr : overstr; 91 } 92 93 address = erraddr & KN0X_EAR_ADDRESS; 94 /* For ECC errors on reads adjust for MT pipelining. */ 95 if ((erraddr & (KN0X_EAR_WRITE | KN0X_EAR_ECCERR)) == KN0X_EAR_ECCERR) 96 address = (address & ~0xfffLL) | ((address - 5) & 0xfffLL); 97 address <<= 2; 98 99 /* Only CPU errors are fixable. */ 100 if (erraddr & KN0X_EAR_CPU && is_fixup) 101 action = MIPS_BE_FIXUP; 102 103 if (erraddr & KN0X_EAR_ECCERR) { 104 static const u8 data_sbit[32] = { 105 0x4f, 0x4a, 0x52, 0x54, 0x57, 0x58, 0x5b, 0x5d, 106 0x23, 0x25, 0x26, 0x29, 0x2a, 0x2c, 0x31, 0x34, 107 0x0e, 0x0b, 0x13, 0x15, 0x16, 0x19, 0x1a, 0x1c, 108 0x62, 0x64, 0x67, 0x68, 0x6b, 0x6d, 0x70, 0x75, 109 }; 110 static const u8 data_mbit[25] = { 111 0x07, 0x0d, 0x1f, 112 0x2f, 0x32, 0x37, 0x38, 0x3b, 0x3d, 0x3e, 113 0x43, 0x45, 0x46, 0x49, 0x4c, 0x51, 0x5e, 114 0x61, 0x6e, 0x73, 0x76, 0x79, 0x7a, 0x7c, 0x7f, 115 }; 116 static const char sbestr[] = "corrected single"; 117 static const char dbestr[] = "uncorrectable double"; 118 static const char mbestr[] = "uncorrectable multiple"; 119 120 if (!(address & 0x4)) 121 syn = chksyn; /* Low bank. */ 122 else 123 syn = chksyn >> 16; /* High bank. */ 124 125 if (!(syn & KN0X_ESR_VLDLO)) { 126 /* Ack now, no rewrite will happen. */ 127 dec_ecc_be_ack(); 128 129 fmt = KERN_ALERT "%s" "invalid\n"; 130 } else { 131 sngl = syn & KN0X_ESR_SNGLO; 132 syn &= KN0X_ESR_SYNLO; 133 134 /* 135 * Multibit errors may be tagged incorrectly; 136 * check the syndrome explicitly. 137 */ 138 for (i = 0; i < 25; i++) 139 if (syn == data_mbit[i]) 140 break; 141 142 if (i < 25) { 143 status = mbestr; 144 } else if (!sngl) { 145 status = dbestr; 146 } else { 147 volatile u32 *ptr = 148 (void *)CKSEG1ADDR(address); 149 150 *ptr = *ptr; /* Rewrite. */ 151 iob(); 152 153 status = sbestr; 154 action = MIPS_BE_DISCARD; 155 } 156 157 /* Ack now, now we've rewritten (or not). */ 158 dec_ecc_be_ack(); 159 160 if (syn && syn == (syn & -syn)) { 161 if (syn == 0x01) { 162 fmt = KERN_ALERT "%s" 163 "%#04x -- %s bit error " 164 "at check bit C%s\n"; 165 xbit = "X"; 166 } else { 167 fmt = KERN_ALERT "%s" 168 "%#04x -- %s bit error " 169 "at check bit C%s%u\n"; 170 } 171 i = syn >> 2; 172 } else { 173 for (i = 0; i < 32; i++) 174 if (syn == data_sbit[i]) 175 break; 176 if (i < 32) 177 fmt = KERN_ALERT "%s" 178 "%#04x -- %s bit error " 179 "at data bit D%s%u\n"; 180 else 181 fmt = KERN_ALERT "%s" 182 "%#04x -- %s bit error\n"; 183 } 184 } 185 } 186 187 if (action != MIPS_BE_FIXUP) 188 printk(KERN_ALERT "Bus error %s: %s %s %s at %#010lx\n", 189 kind, agent, cycle, event, address); 190 191 if (action != MIPS_BE_FIXUP && erraddr & KN0X_EAR_ECCERR) 192 printk(fmt, " ECC syndrome ", syn, status, xbit, i); 193 194 return action; 195} 196 197int dec_ecc_be_handler(struct pt_regs *regs, int is_fixup) 198{ 199 return dec_ecc_be_backend(regs, is_fixup, 0); 200} 201 202irqreturn_t dec_ecc_be_interrupt(int irq, void *dev_id) 203{ 204 struct pt_regs *regs = get_irq_regs(); 205 206 int action = dec_ecc_be_backend(regs, 0, 1); 207 208 if (action == MIPS_BE_DISCARD) 209 return IRQ_HANDLED; 210 211 /* 212 * FIXME: Find the affected processes and kill them, otherwise 213 * we must die. 214 * 215 * The interrupt is asynchronously delivered thus EPC and RA 216 * may be irrelevant, but are printed for a reference. 217 */ 218 printk(KERN_ALERT "Fatal bus interrupt, epc == %08lx, ra == %08lx\n", 219 regs->cp0_epc, regs->regs[31]); 220 die("Unrecoverable bus error", regs); 221} 222 223 224/* 225 * Initialization differs a bit between KN02 and KN03/KN05, so we 226 * need two variants. Once set up, all systems can be handled the 227 * same way. 228 */ 229static inline void dec_kn02_be_init(void) 230{ 231 volatile u32 *csr = (void *)CKSEG1ADDR(KN02_SLOT_BASE + KN02_CSR); 232 233 kn0x_erraddr = (void *)CKSEG1ADDR(KN02_SLOT_BASE + KN02_ERRADDR); 234 kn0x_chksyn = (void *)CKSEG1ADDR(KN02_SLOT_BASE + KN02_CHKSYN); 235 236 /* Preset write-only bits of the Control Register cache. */ 237 cached_kn02_csr = *csr | KN02_CSR_LEDS; 238 239 /* Set normal ECC detection and generation. */ 240 cached_kn02_csr &= ~(KN02_CSR_DIAGCHK | KN02_CSR_DIAGGEN); 241 /* Enable ECC correction. */ 242 cached_kn02_csr |= KN02_CSR_CORRECT; 243 *csr = cached_kn02_csr; 244 iob(); 245} 246 247static inline void dec_kn03_be_init(void) 248{ 249 volatile u32 *mcr = (void *)CKSEG1ADDR(KN03_SLOT_BASE + IOASIC_MCR); 250 volatile u32 *mbcs = (void *)CKSEG1ADDR(KN4K_SLOT_BASE + KN4K_MB_CSR); 251 252 kn0x_erraddr = (void *)CKSEG1ADDR(KN03_SLOT_BASE + IOASIC_ERRADDR); 253 kn0x_chksyn = (void *)CKSEG1ADDR(KN03_SLOT_BASE + IOASIC_CHKSYN); 254 255 /* 256 * Set normal ECC detection and generation, enable ECC correction. 257 * For KN05 we also need to make sure EE (?) is enabled in the MB. 258 * Otherwise DBE/IBE exceptions would be masked but bus error 259 * interrupts would still arrive, resulting in an inevitable crash 260 * if get_dbe() triggers one. 261 */ 262 *mcr = (*mcr & ~(KN03_MCR_DIAGCHK | KN03_MCR_DIAGGEN)) | 263 KN03_MCR_CORRECT; 264 if (current_cpu_type() == CPU_R4400SC) 265 *mbcs |= KN4K_MB_CSR_EE; 266 fast_iob(); 267} 268 269void __init dec_ecc_be_init(void) 270{ 271 if (mips_machtype == MACH_DS5000_200) 272 dec_kn02_be_init(); 273 else 274 dec_kn03_be_init(); 275 276 /* Clear any leftover errors from the firmware. */ 277 dec_ecc_be_ack(); 278} 279