arch/mips/dec/ecc-berr.c

   1 /*
   2  *      linux/arch/mips/dec/ecc-berr.c
   3  *
   4  *      Bus error event handling code for systems equipped with ECC
   5  *      handling logic, i.e. DECstation/DECsystem 5000/200 (KN02),
   6  *      5000/240 (KN03), 5000/260 (KN05) and DECsystem 5900 (KN03),
   7  *      5900/260 (KN05) systems.
   8  *
   9  *      Copyright (c) 2003, 2005  Maciej W. Rozycki
  10  *
  11  *      This program is free software; you can redistribute it and/or
  12  *      modify it under the terms of the GNU General Public License
  13  *      as published by the Free Software Foundation; either version
  14  *      2 of the License, or (at your option) any later version.
  15  */
  16
  17 #include <linux/init.h>
  18 #include <linux/interrupt.h>
  19 #include <linux/kernel.h>
  20 #include <linux/sched.h>
  21 #include <linux/spinlock.h>
  22 #include <linux/types.h>
  23
  24 #include <asm/addrspace.h>
  25 #include <asm/bootinfo.h>
  26 #include <asm/cpu.h>
  27 #include <asm/irq_regs.h>
  28 #include <asm/processor.h>
  29 #include <asm/system.h>
  30 #include <asm/traps.h>
  31
  32 #include <asm/dec/ecc.h>
  33 #include <asm/dec/kn02.h>
  34 #include <asm/dec/kn03.h>
  35 #include <asm/dec/kn05.h>
  36
  37 static volatile u32 *kn0x_erraddr;
  38 static volatile u32 *kn0x_chksyn;
  39
  40 static inline void dec_ecc_be_ack(void)
  41 {
  42         *kn0x_erraddr = 0;                      /* any write clears the IRQ */
  43         iob();
  44 }
  45
  46 static int dec_ecc_be_backend(struct pt_regs *regs, int is_fixup, int invoker)
  47 {
  48         static const char excstr[] = "exception";
  49         static const char intstr[] = "interrupt";
  50         static const char cpustr[] = "CPU";
  51         static const char dmastr[] = "DMA";
  52         static const char readstr[] = "read";
  53         static const char mreadstr[] = "memory read";
  54         static const char writestr[] = "write";
  55         static const char mwritstr[] = "partial memory write";
  56         static const char timestr[] = "timeout";
  57         static const char overstr[] = "overrun";
  58         static const char eccstr[] = "ECC error";
  59
  60         const char *kind, *agent, *cycle, *event;
  61         const char *status = "", *xbit = "", *fmt = "";
  62         unsigned long address;
  63         u16 syn = 0, sngl;
  64
  65         int i = 0;
  66
  67         u32 erraddr = *kn0x_erraddr;
  68         u32 chksyn = *kn0x_chksyn;
  69         int action = MIPS_BE_FATAL;
  70
  71         /* For non-ECC ack ASAP, so that any subsequent errors get caught. */
  72         if ((erraddr & (KN0X_EAR_VALID | KN0X_EAR_ECCERR)) == KN0X_EAR_VALID)
  73                 dec_ecc_be_ack();
  74
  75         kind = invoker ? intstr : excstr;
  76
  77         if (!(erraddr & KN0X_EAR_VALID)) {
  78                 /* No idea what happened. */
  79                 printk(KERN_ALERT "Unidentified bus error %s\n", kind);
  80                 return action;
  81         }
  82
  83         agent = (erraddr & KN0X_EAR_CPU) ? cpustr : dmastr;
  84
  85         if (erraddr & KN0X_EAR_ECCERR) {
  86                 /* An ECC error on a CPU or DMA transaction. */
  87                 cycle = (erraddr & KN0X_EAR_WRITE) ? mwritstr : mreadstr;
  88                 event = eccstr;
  89         } else {
  90                 /* A CPU timeout or a DMA overrun. */
  91                 cycle = (erraddr & KN0X_EAR_WRITE) ? writestr : readstr;
  92                 event = (erraddr & KN0X_EAR_CPU) ? timestr : overstr;
  93         }
  94
  95         address = erraddr & KN0X_EAR_ADDRESS;
  96         /* For ECC errors on reads adjust for MT pipelining. */
  97         if ((erraddr & (KN0X_EAR_WRITE | KN0X_EAR_ECCERR)) == KN0X_EAR_ECCERR)
  98                 address = (address & ~0xfffLL) | ((address - 5) & 0xfffLL);
  99         address <<= 2;
 100
 101         /* Only CPU errors are fixable. */
 102         if (erraddr & KN0X_EAR_CPU && is_fixup)
 103                 action = MIPS_BE_FIXUP;
 104
 105         if (erraddr & KN0X_EAR_ECCERR) {
 106                 static const u8 data_sbit[32] = {
 107                         0x4f, 0x4a, 0x52, 0x54, 0x57, 0x58, 0x5b, 0x5d,
 108                         0x23, 0x25, 0x26, 0x29, 0x2a, 0x2c, 0x31, 0x34,
 109                         0x0e, 0x0b, 0x13, 0x15, 0x16, 0x19, 0x1a, 0x1c,
 110                         0x62, 0x64, 0x67, 0x68, 0x6b, 0x6d, 0x70, 0x75,
 111                 };
 112                 static const u8 data_mbit[25] = {
 113                         0x07, 0x0d, 0x1f,
 114                         0x2f, 0x32, 0x37, 0x38, 0x3b, 0x3d, 0x3e,
 115                         0x43, 0x45, 0x46, 0x49, 0x4c, 0x51, 0x5e,
 116                         0x61, 0x6e, 0x73, 0x76, 0x79, 0x7a, 0x7c, 0x7f,
 117                 };
 118                 static const char sbestr[] = "corrected single";
 119                 static const char dbestr[] = "uncorrectable double";
 120                 static const char mbestr[] = "uncorrectable multiple";
 121
 122                 if (!(address & 0x4))
 123                         syn = chksyn;                   /* Low bank. */
 124                 else
 125                         syn = chksyn >> 16;             /* High bank. */
 126
 127                 if (!(syn & KN0X_ESR_VLDLO)) {
 128                         /* Ack now, no rewrite will happen. */
 129                         dec_ecc_be_ack();
 130
 131                         fmt = KERN_ALERT "%s" "invalid\n";
 132                 } else {
 133                         sngl = syn & KN0X_ESR_SNGLO;
 134                         syn &= KN0X_ESR_SYNLO;
 135
 136                         /*
 137                          * Multibit errors may be tagged incorrectly;
 138                          * check the syndrome explicitly.
 139                          */
 140                         for (i = 0; i < 25; i++)
 141                                 if (syn == data_mbit[i])
 142                                         break;
 143
 144                         if (i < 25) {
 145                                 status = mbestr;
 146                         } else if (!sngl) {
 147                                 status = dbestr;
 148                         } else {
 149                                 volatile u32 *ptr =
 150                                         (void *)CKSEG1ADDR(address);
 151
 152                                 *ptr = *ptr;            /* Rewrite. */
 153                                 iob();
 154
 155                                 status = sbestr;
 156                                 action = MIPS_BE_DISCARD;
 157                         }
 158
 159                         /* Ack now, now we've rewritten (or not). */
 160                         dec_ecc_be_ack();
 161
 162                         if (syn && syn == (syn & -syn)) {
 163                                 if (syn == 0x01) {
 164                                         fmt = KERN_ALERT "%s"
 165                                               "%#04x -- %s bit error "
 166                                               "at check bit C%s\n";
 167                                         xbit = "X";
 168                                 } else {
 169                                         fmt = KERN_ALERT "%s"
 170                                               "%#04x -- %s bit error "
 171                                               "at check bit C%s%u\n";
 172                                 }
 173                                 i = syn >> 2;
 174                         } else {
 175                                 for (i = 0; i < 32; i++)
 176                                         if (syn == data_sbit[i])
 177                                                 break;
 178                                 if (i < 32)
 179                                         fmt = KERN_ALERT "%s"
 180                                               "%#04x -- %s bit error "
 181                                               "at data bit D%s%u\n";
 182                                 else
 183                                         fmt = KERN_ALERT "%s"
 184                                               "%#04x -- %s bit error\n";
 185                         }
 186                 }
 187         }
 188
 189         if (action != MIPS_BE_FIXUP)
 190                 printk(KERN_ALERT "Bus error %s: %s %s %s at %#010lx\n",
 191                         kind, agent, cycle, event, address);
 192
 193         if (action != MIPS_BE_FIXUP && erraddr & KN0X_EAR_ECCERR)
 194                 printk(fmt, "  ECC syndrome ", syn, status, xbit, i);
 195
 196         return action;
 197 }
 198
 199 int dec_ecc_be_handler(struct pt_regs *regs, int is_fixup)
 200 {
 201         return dec_ecc_be_backend(regs, is_fixup, 0);
 202 }
 203
 204 irqreturn_t dec_ecc_be_interrupt(int irq, void *dev_id)
 205 {
 206         struct pt_regs *regs = get_irq_regs();
 207
 208         int action = dec_ecc_be_backend(regs, 0, 1);
 209
 210         if (action == MIPS_BE_DISCARD)
 211                 return IRQ_HANDLED;
 212
 213         /*
 214          * FIXME: Find the affected processes and kill them, otherwise
 215          * we must die.
 216          *
 217          * The interrupt is asynchronously delivered thus EPC and RA
 218          * may be irrelevant, but are printed for a reference.
 219          */
 220         printk(KERN_ALERT "Fatal bus interrupt, epc == %08lx, ra == %08lx\n",
 221                regs->cp0_epc, regs->regs[31]);
 222         die("Unrecoverable bus error", regs);
 223 }
 224
 225
 226 /*
 227  * Initialization differs a bit between KN02 and KN03/KN05, so we
 228  * need two variants.  Once set up, all systems can be handled the
 229  * same way.
 230  */
 231 static inline void dec_kn02_be_init(void)
 232 {
 233         volatile u32 *csr = (void *)CKSEG1ADDR(KN02_SLOT_BASE + KN02_CSR);
 234         unsigned long flags;
 235
 236         kn0x_erraddr = (void *)CKSEG1ADDR(KN02_SLOT_BASE + KN02_ERRADDR);
 237         kn0x_chksyn = (void *)CKSEG1ADDR(KN02_SLOT_BASE + KN02_CHKSYN);
 238
 239         spin_lock_irqsave(&kn02_lock, flags);
 240
 241         /* Preset write-only bits of the Control Register cache. */
 242         cached_kn02_csr = *csr | KN02_CSR_LEDS;
 243
 244         /* Set normal ECC detection and generation. */
 245         cached_kn02_csr &= ~(KN02_CSR_DIAGCHK | KN02_CSR_DIAGGEN);
 246         /* Enable ECC correction. */
 247         cached_kn02_csr |= KN02_CSR_CORRECT;
 248         *csr = cached_kn02_csr;
 249         iob();
 250
 251         spin_unlock_irqrestore(&kn02_lock, flags);
 252 }
 253
 254 static inline void dec_kn03_be_init(void)
 255 {
 256         volatile u32 *mcr = (void *)CKSEG1ADDR(KN03_SLOT_BASE + IOASIC_MCR);
 257         volatile u32 *mbcs = (void *)CKSEG1ADDR(KN4K_SLOT_BASE + KN4K_MB_CSR);
 258
 259         kn0x_erraddr = (void *)CKSEG1ADDR(KN03_SLOT_BASE + IOASIC_ERRADDR);
 260         kn0x_chksyn = (void *)CKSEG1ADDR(KN03_SLOT_BASE + IOASIC_CHKSYN);
 261
 262         /*
 263          * Set normal ECC detection and generation, enable ECC correction.
 264          * For KN05 we also need to make sure EE (?) is enabled in the MB.
 265          * Otherwise DBE/IBE exceptions would be masked but bus error
 266          * interrupts would still arrive, resulting in an inevitable crash
 267          * if get_dbe() triggers one.
 268          */
 269         *mcr = (*mcr & ~(KN03_MCR_DIAGCHK | KN03_MCR_DIAGGEN)) |
 270                KN03_MCR_CORRECT;
 271         if (current_cpu_data.cputype == CPU_R4400SC)
 272                 *mbcs |= KN4K_MB_CSR_EE;
 273         fast_iob();
 274 }
 275
 276 void __init dec_ecc_be_init(void)
 277 {
 278         if (mips_machtype == MACH_DS5000_200)
 279                 dec_kn02_be_init();
 280         else
 281                 dec_kn03_be_init();
 282
 283         /* Clear any leftover errors from the firmware. */
 284         dec_ecc_be_ack();
 285 }