David Daney | 4c076fb | 2010-07-24 10:16:05 -0700 | [diff] [blame] | 1 | /* |
| 2 | * Octeon Watchdog driver |
| 3 | * |
| 4 | * Copyright (C) 2007, 2008, 2009, 2010 Cavium Networks |
| 5 | * |
| 6 | * Some parts derived from wdt.c |
| 7 | * |
| 8 | * (c) Copyright 1996-1997 Alan Cox <alan@lxorguk.ukuu.org.uk>, |
| 9 | * All Rights Reserved. |
| 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 | * Neither Alan Cox nor CymruNet Ltd. admit liability nor provide |
| 17 | * warranty for any of this software. This material is provided |
| 18 | * "AS-IS" and at no charge. |
| 19 | * |
| 20 | * (c) Copyright 1995 Alan Cox <alan@lxorguk.ukuu.org.uk> |
| 21 | * |
| 22 | * This file is subject to the terms and conditions of the GNU General Public |
| 23 | * License. See the file "COPYING" in the main directory of this archive |
| 24 | * for more details. |
| 25 | * |
| 26 | * |
| 27 | * The OCTEON watchdog has a maximum timeout of 2^32 * io_clock. |
| 28 | * For most systems this is less than 10 seconds, so to allow for |
| 29 | * software to request longer watchdog heartbeats, we maintain software |
| 30 | * counters to count multiples of the base rate. If the system locks |
| 31 | * up in such a manner that we can not run the software counters, the |
| 32 | * only result is a watchdog reset sooner than was requested. But |
| 33 | * that is OK, because in this case userspace would likely not be able |
| 34 | * to do anything anyhow. |
| 35 | * |
| 36 | * The hardware watchdog interval we call the period. The OCTEON |
| 37 | * watchdog goes through several stages, after the first period an |
| 38 | * irq is asserted, then if it is not reset, after the next period NMI |
| 39 | * is asserted, then after an additional period a chip wide soft reset. |
| 40 | * So for the software counters, we reset watchdog after each period |
| 41 | * and decrement the counter. But for the last two periods we need to |
| 42 | * let the watchdog progress to the NMI stage so we disable the irq |
| 43 | * and let it proceed. Once in the NMI, we print the register state |
| 44 | * to the serial port and then wait for the reset. |
| 45 | * |
| 46 | * A watchdog is maintained for each CPU in the system, that way if |
| 47 | * one CPU suffers a lockup, we also get a register dump and reset. |
| 48 | * The userspace ping resets the watchdog on all CPUs. |
| 49 | * |
| 50 | * Before userspace opens the watchdog device, we still run the |
| 51 | * watchdogs to catch any lockups that may be kernel related. |
| 52 | * |
| 53 | */ |
| 54 | |
Joe Perches | 27c766a | 2012-02-15 15:06:19 -0800 | [diff] [blame] | 55 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
| 56 | |
David Daney | 4c076fb | 2010-07-24 10:16:05 -0700 | [diff] [blame] | 57 | #include <linux/miscdevice.h> |
| 58 | #include <linux/interrupt.h> |
| 59 | #include <linux/watchdog.h> |
| 60 | #include <linux/cpumask.h> |
| 61 | #include <linux/bitops.h> |
| 62 | #include <linux/kernel.h> |
| 63 | #include <linux/module.h> |
| 64 | #include <linux/string.h> |
| 65 | #include <linux/delay.h> |
| 66 | #include <linux/cpu.h> |
| 67 | #include <linux/smp.h> |
| 68 | #include <linux/fs.h> |
David Howells | ca4d3e67 | 2010-10-07 14:08:54 +0100 | [diff] [blame] | 69 | #include <linux/irq.h> |
David Daney | 4c076fb | 2010-07-24 10:16:05 -0700 | [diff] [blame] | 70 | |
| 71 | #include <asm/mipsregs.h> |
| 72 | #include <asm/uasm.h> |
| 73 | |
| 74 | #include <asm/octeon/octeon.h> |
| 75 | |
| 76 | /* The count needed to achieve timeout_sec. */ |
| 77 | static unsigned int timeout_cnt; |
| 78 | |
| 79 | /* The maximum period supported. */ |
| 80 | static unsigned int max_timeout_sec; |
| 81 | |
| 82 | /* The current period. */ |
| 83 | static unsigned int timeout_sec; |
| 84 | |
| 85 | /* Set to non-zero when userspace countdown mode active */ |
| 86 | static int do_coundown; |
| 87 | static unsigned int countdown_reset; |
| 88 | static unsigned int per_cpu_countdown[NR_CPUS]; |
| 89 | |
| 90 | static cpumask_t irq_enabled_cpus; |
| 91 | |
| 92 | #define WD_TIMO 60 /* Default heartbeat = 60 seconds */ |
| 93 | |
| 94 | static int heartbeat = WD_TIMO; |
| 95 | module_param(heartbeat, int, S_IRUGO); |
| 96 | MODULE_PARM_DESC(heartbeat, |
| 97 | "Watchdog heartbeat in seconds. (0 < heartbeat, default=" |
| 98 | __MODULE_STRING(WD_TIMO) ")"); |
| 99 | |
Wim Van Sebroeck | 86a1e18 | 2012-03-05 16:51:11 +0100 | [diff] [blame] | 100 | static bool nowayout = WATCHDOG_NOWAYOUT; |
| 101 | module_param(nowayout, bool, S_IRUGO); |
David Daney | 4c076fb | 2010-07-24 10:16:05 -0700 | [diff] [blame] | 102 | MODULE_PARM_DESC(nowayout, |
| 103 | "Watchdog cannot be stopped once started (default=" |
| 104 | __MODULE_STRING(WATCHDOG_NOWAYOUT) ")"); |
| 105 | |
| 106 | static unsigned long octeon_wdt_is_open; |
| 107 | static char expect_close; |
| 108 | |
| 109 | static u32 __initdata nmi_stage1_insns[64]; |
| 110 | /* We need one branch and therefore one relocation per target label. */ |
| 111 | static struct uasm_label __initdata labels[5]; |
| 112 | static struct uasm_reloc __initdata relocs[5]; |
| 113 | |
| 114 | enum lable_id { |
| 115 | label_enter_bootloader = 1 |
| 116 | }; |
| 117 | |
| 118 | /* Some CP0 registers */ |
| 119 | #define K0 26 |
| 120 | #define C0_CVMMEMCTL 11, 7 |
| 121 | #define C0_STATUS 12, 0 |
| 122 | #define C0_EBASE 15, 1 |
| 123 | #define C0_DESAVE 31, 0 |
| 124 | |
| 125 | void octeon_wdt_nmi_stage2(void); |
| 126 | |
| 127 | static void __init octeon_wdt_build_stage1(void) |
| 128 | { |
| 129 | int i; |
| 130 | int len; |
| 131 | u32 *p = nmi_stage1_insns; |
| 132 | #ifdef CONFIG_HOTPLUG_CPU |
| 133 | struct uasm_label *l = labels; |
| 134 | struct uasm_reloc *r = relocs; |
| 135 | #endif |
| 136 | |
| 137 | /* |
| 138 | * For the next few instructions running the debugger may |
| 139 | * cause corruption of k0 in the saved registers. Since we're |
| 140 | * about to crash, nobody probably cares. |
| 141 | * |
| 142 | * Save K0 into the debug scratch register |
| 143 | */ |
| 144 | uasm_i_dmtc0(&p, K0, C0_DESAVE); |
| 145 | |
| 146 | uasm_i_mfc0(&p, K0, C0_STATUS); |
| 147 | #ifdef CONFIG_HOTPLUG_CPU |
| 148 | uasm_il_bbit0(&p, &r, K0, ilog2(ST0_NMI), label_enter_bootloader); |
| 149 | #endif |
| 150 | /* Force 64-bit addressing enabled */ |
| 151 | uasm_i_ori(&p, K0, K0, ST0_UX | ST0_SX | ST0_KX); |
| 152 | uasm_i_mtc0(&p, K0, C0_STATUS); |
| 153 | |
| 154 | #ifdef CONFIG_HOTPLUG_CPU |
| 155 | uasm_i_mfc0(&p, K0, C0_EBASE); |
| 156 | /* Coreid number in K0 */ |
| 157 | uasm_i_andi(&p, K0, K0, 0xf); |
| 158 | /* 8 * coreid in bits 16-31 */ |
| 159 | uasm_i_dsll_safe(&p, K0, K0, 3 + 16); |
| 160 | uasm_i_ori(&p, K0, K0, 0x8001); |
| 161 | uasm_i_dsll_safe(&p, K0, K0, 16); |
| 162 | uasm_i_ori(&p, K0, K0, 0x0700); |
| 163 | uasm_i_drotr_safe(&p, K0, K0, 32); |
| 164 | /* |
| 165 | * Should result in: 0x8001,0700,0000,8*coreid which is |
| 166 | * CVMX_CIU_WDOGX(coreid) - 0x0500 |
| 167 | * |
| 168 | * Now ld K0, CVMX_CIU_WDOGX(coreid) |
| 169 | */ |
| 170 | uasm_i_ld(&p, K0, 0x500, K0); |
| 171 | /* |
| 172 | * If bit one set handle the NMI as a watchdog event. |
| 173 | * otherwise transfer control to bootloader. |
| 174 | */ |
| 175 | uasm_il_bbit0(&p, &r, K0, 1, label_enter_bootloader); |
| 176 | uasm_i_nop(&p); |
| 177 | #endif |
| 178 | |
| 179 | /* Clear Dcache so cvmseg works right. */ |
| 180 | uasm_i_cache(&p, 1, 0, 0); |
| 181 | |
| 182 | /* Use K0 to do a read/modify/write of CVMMEMCTL */ |
| 183 | uasm_i_dmfc0(&p, K0, C0_CVMMEMCTL); |
| 184 | /* Clear out the size of CVMSEG */ |
| 185 | uasm_i_dins(&p, K0, 0, 0, 6); |
| 186 | /* Set CVMSEG to its largest value */ |
| 187 | uasm_i_ori(&p, K0, K0, 0x1c0 | 54); |
| 188 | /* Store the CVMMEMCTL value */ |
| 189 | uasm_i_dmtc0(&p, K0, C0_CVMMEMCTL); |
| 190 | |
| 191 | /* Load the address of the second stage handler */ |
| 192 | UASM_i_LA(&p, K0, (long)octeon_wdt_nmi_stage2); |
| 193 | uasm_i_jr(&p, K0); |
| 194 | uasm_i_dmfc0(&p, K0, C0_DESAVE); |
| 195 | |
| 196 | #ifdef CONFIG_HOTPLUG_CPU |
| 197 | uasm_build_label(&l, p, label_enter_bootloader); |
| 198 | /* Jump to the bootloader and restore K0 */ |
| 199 | UASM_i_LA(&p, K0, (long)octeon_bootloader_entry_addr); |
| 200 | uasm_i_jr(&p, K0); |
| 201 | uasm_i_dmfc0(&p, K0, C0_DESAVE); |
| 202 | #endif |
| 203 | uasm_resolve_relocs(relocs, labels); |
| 204 | |
| 205 | len = (int)(p - nmi_stage1_insns); |
Joe Perches | 27c766a | 2012-02-15 15:06:19 -0800 | [diff] [blame] | 206 | pr_debug("Synthesized NMI stage 1 handler (%d instructions)\n", len); |
David Daney | 4c076fb | 2010-07-24 10:16:05 -0700 | [diff] [blame] | 207 | |
| 208 | pr_debug("\t.set push\n"); |
| 209 | pr_debug("\t.set noreorder\n"); |
| 210 | for (i = 0; i < len; i++) |
| 211 | pr_debug("\t.word 0x%08x\n", nmi_stage1_insns[i]); |
| 212 | pr_debug("\t.set pop\n"); |
| 213 | |
| 214 | if (len > 32) |
| 215 | panic("NMI stage 1 handler exceeds 32 instructions, was %d\n", len); |
| 216 | } |
| 217 | |
| 218 | static int cpu2core(int cpu) |
| 219 | { |
| 220 | #ifdef CONFIG_SMP |
| 221 | return cpu_logical_map(cpu); |
| 222 | #else |
| 223 | return cvmx_get_core_num(); |
| 224 | #endif |
| 225 | } |
| 226 | |
| 227 | static int core2cpu(int coreid) |
| 228 | { |
| 229 | #ifdef CONFIG_SMP |
| 230 | return cpu_number_map(coreid); |
| 231 | #else |
| 232 | return 0; |
| 233 | #endif |
| 234 | } |
| 235 | |
| 236 | /** |
| 237 | * Poke the watchdog when an interrupt is received |
| 238 | * |
| 239 | * @cpl: |
| 240 | * @dev_id: |
| 241 | * |
| 242 | * Returns |
| 243 | */ |
| 244 | static irqreturn_t octeon_wdt_poke_irq(int cpl, void *dev_id) |
| 245 | { |
| 246 | unsigned int core = cvmx_get_core_num(); |
| 247 | int cpu = core2cpu(core); |
| 248 | |
| 249 | if (do_coundown) { |
| 250 | if (per_cpu_countdown[cpu] > 0) { |
| 251 | /* We're alive, poke the watchdog */ |
| 252 | cvmx_write_csr(CVMX_CIU_PP_POKEX(core), 1); |
| 253 | per_cpu_countdown[cpu]--; |
| 254 | } else { |
| 255 | /* Bad news, you are about to reboot. */ |
| 256 | disable_irq_nosync(cpl); |
| 257 | cpumask_clear_cpu(cpu, &irq_enabled_cpus); |
| 258 | } |
| 259 | } else { |
| 260 | /* Not open, just ping away... */ |
| 261 | cvmx_write_csr(CVMX_CIU_PP_POKEX(core), 1); |
| 262 | } |
| 263 | return IRQ_HANDLED; |
| 264 | } |
| 265 | |
| 266 | /* From setup.c */ |
| 267 | extern int prom_putchar(char c); |
| 268 | |
| 269 | /** |
| 270 | * Write a string to the uart |
| 271 | * |
| 272 | * @str: String to write |
| 273 | */ |
| 274 | static void octeon_wdt_write_string(const char *str) |
| 275 | { |
| 276 | /* Just loop writing one byte at a time */ |
| 277 | while (*str) |
| 278 | prom_putchar(*str++); |
| 279 | } |
| 280 | |
| 281 | /** |
| 282 | * Write a hex number out of the uart |
| 283 | * |
| 284 | * @value: Number to display |
| 285 | * @digits: Number of digits to print (1 to 16) |
| 286 | */ |
| 287 | static void octeon_wdt_write_hex(u64 value, int digits) |
| 288 | { |
| 289 | int d; |
| 290 | int v; |
| 291 | for (d = 0; d < digits; d++) { |
| 292 | v = (value >> ((digits - d - 1) * 4)) & 0xf; |
| 293 | if (v >= 10) |
| 294 | prom_putchar('a' + v - 10); |
| 295 | else |
| 296 | prom_putchar('0' + v); |
| 297 | } |
| 298 | } |
| 299 | |
| 300 | const char *reg_name[] = { |
| 301 | "$0", "at", "v0", "v1", "a0", "a1", "a2", "a3", |
| 302 | "a4", "a5", "a6", "a7", "t0", "t1", "t2", "t3", |
| 303 | "s0", "s1", "s2", "s3", "s4", "s5", "s6", "s7", |
| 304 | "t8", "t9", "k0", "k1", "gp", "sp", "s8", "ra" |
| 305 | }; |
| 306 | |
| 307 | /** |
| 308 | * NMI stage 3 handler. NMIs are handled in the following manner: |
| 309 | * 1) The first NMI handler enables CVMSEG and transfers from |
| 310 | * the bootbus region into normal memory. It is careful to not |
| 311 | * destroy any registers. |
| 312 | * 2) The second stage handler uses CVMSEG to save the registers |
| 313 | * and create a stack for C code. It then calls the third level |
| 314 | * handler with one argument, a pointer to the register values. |
| 315 | * 3) The third, and final, level handler is the following C |
| 316 | * function that prints out some useful infomration. |
| 317 | * |
| 318 | * @reg: Pointer to register state before the NMI |
| 319 | */ |
| 320 | void octeon_wdt_nmi_stage3(u64 reg[32]) |
| 321 | { |
| 322 | u64 i; |
| 323 | |
| 324 | unsigned int coreid = cvmx_get_core_num(); |
| 325 | /* |
| 326 | * Save status and cause early to get them before any changes |
| 327 | * might happen. |
| 328 | */ |
| 329 | u64 cp0_cause = read_c0_cause(); |
| 330 | u64 cp0_status = read_c0_status(); |
| 331 | u64 cp0_error_epc = read_c0_errorepc(); |
| 332 | u64 cp0_epc = read_c0_epc(); |
| 333 | |
| 334 | /* Delay so output from all cores output is not jumbled together. */ |
| 335 | __delay(100000000ull * coreid); |
| 336 | |
| 337 | octeon_wdt_write_string("\r\n*** NMI Watchdog interrupt on Core 0x"); |
| 338 | octeon_wdt_write_hex(coreid, 1); |
| 339 | octeon_wdt_write_string(" ***\r\n"); |
| 340 | for (i = 0; i < 32; i++) { |
| 341 | octeon_wdt_write_string("\t"); |
| 342 | octeon_wdt_write_string(reg_name[i]); |
| 343 | octeon_wdt_write_string("\t0x"); |
| 344 | octeon_wdt_write_hex(reg[i], 16); |
| 345 | if (i & 1) |
| 346 | octeon_wdt_write_string("\r\n"); |
| 347 | } |
| 348 | octeon_wdt_write_string("\terr_epc\t0x"); |
| 349 | octeon_wdt_write_hex(cp0_error_epc, 16); |
| 350 | |
| 351 | octeon_wdt_write_string("\tepc\t0x"); |
| 352 | octeon_wdt_write_hex(cp0_epc, 16); |
| 353 | octeon_wdt_write_string("\r\n"); |
| 354 | |
| 355 | octeon_wdt_write_string("\tstatus\t0x"); |
| 356 | octeon_wdt_write_hex(cp0_status, 16); |
| 357 | octeon_wdt_write_string("\tcause\t0x"); |
| 358 | octeon_wdt_write_hex(cp0_cause, 16); |
| 359 | octeon_wdt_write_string("\r\n"); |
| 360 | |
| 361 | octeon_wdt_write_string("\tsum0\t0x"); |
| 362 | octeon_wdt_write_hex(cvmx_read_csr(CVMX_CIU_INTX_SUM0(coreid * 2)), 16); |
| 363 | octeon_wdt_write_string("\ten0\t0x"); |
| 364 | octeon_wdt_write_hex(cvmx_read_csr(CVMX_CIU_INTX_EN0(coreid * 2)), 16); |
| 365 | octeon_wdt_write_string("\r\n"); |
| 366 | |
| 367 | octeon_wdt_write_string("*** Chip soft reset soon ***\r\n"); |
| 368 | } |
| 369 | |
| 370 | static void octeon_wdt_disable_interrupt(int cpu) |
| 371 | { |
| 372 | unsigned int core; |
| 373 | unsigned int irq; |
| 374 | union cvmx_ciu_wdogx ciu_wdog; |
| 375 | |
| 376 | core = cpu2core(cpu); |
| 377 | |
| 378 | irq = OCTEON_IRQ_WDOG0 + core; |
| 379 | |
| 380 | /* Poke the watchdog to clear out its state */ |
| 381 | cvmx_write_csr(CVMX_CIU_PP_POKEX(core), 1); |
| 382 | |
| 383 | /* Disable the hardware. */ |
| 384 | ciu_wdog.u64 = 0; |
| 385 | cvmx_write_csr(CVMX_CIU_WDOGX(core), ciu_wdog.u64); |
| 386 | |
| 387 | free_irq(irq, octeon_wdt_poke_irq); |
| 388 | } |
| 389 | |
| 390 | static void octeon_wdt_setup_interrupt(int cpu) |
| 391 | { |
| 392 | unsigned int core; |
| 393 | unsigned int irq; |
| 394 | union cvmx_ciu_wdogx ciu_wdog; |
| 395 | |
| 396 | core = cpu2core(cpu); |
| 397 | |
| 398 | /* Disable it before doing anything with the interrupts. */ |
| 399 | ciu_wdog.u64 = 0; |
| 400 | cvmx_write_csr(CVMX_CIU_WDOGX(core), ciu_wdog.u64); |
| 401 | |
| 402 | per_cpu_countdown[cpu] = countdown_reset; |
| 403 | |
| 404 | irq = OCTEON_IRQ_WDOG0 + core; |
| 405 | |
| 406 | if (request_irq(irq, octeon_wdt_poke_irq, |
Venkat Subbiah | 47bfd05 | 2011-10-03 17:22:04 -0700 | [diff] [blame] | 407 | IRQF_NO_THREAD, "octeon_wdt", octeon_wdt_poke_irq)) |
David Daney | 4c076fb | 2010-07-24 10:16:05 -0700 | [diff] [blame] | 408 | panic("octeon_wdt: Couldn't obtain irq %d", irq); |
| 409 | |
| 410 | cpumask_set_cpu(cpu, &irq_enabled_cpus); |
| 411 | |
| 412 | /* Poke the watchdog to clear out its state */ |
| 413 | cvmx_write_csr(CVMX_CIU_PP_POKEX(core), 1); |
| 414 | |
| 415 | /* Finally enable the watchdog now that all handlers are installed */ |
| 416 | ciu_wdog.u64 = 0; |
| 417 | ciu_wdog.s.len = timeout_cnt; |
| 418 | ciu_wdog.s.mode = 3; /* 3 = Interrupt + NMI + Soft-Reset */ |
| 419 | cvmx_write_csr(CVMX_CIU_WDOGX(core), ciu_wdog.u64); |
| 420 | } |
| 421 | |
| 422 | static int octeon_wdt_cpu_callback(struct notifier_block *nfb, |
| 423 | unsigned long action, void *hcpu) |
| 424 | { |
| 425 | unsigned int cpu = (unsigned long)hcpu; |
| 426 | |
| 427 | switch (action) { |
| 428 | case CPU_DOWN_PREPARE: |
| 429 | octeon_wdt_disable_interrupt(cpu); |
| 430 | break; |
| 431 | case CPU_ONLINE: |
| 432 | case CPU_DOWN_FAILED: |
| 433 | octeon_wdt_setup_interrupt(cpu); |
| 434 | break; |
| 435 | default: |
| 436 | break; |
| 437 | } |
| 438 | return NOTIFY_OK; |
| 439 | } |
| 440 | |
| 441 | static void octeon_wdt_ping(void) |
| 442 | { |
| 443 | int cpu; |
| 444 | int coreid; |
| 445 | |
| 446 | for_each_online_cpu(cpu) { |
| 447 | coreid = cpu2core(cpu); |
| 448 | cvmx_write_csr(CVMX_CIU_PP_POKEX(coreid), 1); |
| 449 | per_cpu_countdown[cpu] = countdown_reset; |
| 450 | if ((countdown_reset || !do_coundown) && |
| 451 | !cpumask_test_cpu(cpu, &irq_enabled_cpus)) { |
| 452 | /* We have to enable the irq */ |
| 453 | int irq = OCTEON_IRQ_WDOG0 + coreid; |
| 454 | enable_irq(irq); |
| 455 | cpumask_set_cpu(cpu, &irq_enabled_cpus); |
| 456 | } |
| 457 | } |
| 458 | } |
| 459 | |
| 460 | static void octeon_wdt_calc_parameters(int t) |
| 461 | { |
| 462 | unsigned int periods; |
| 463 | |
| 464 | timeout_sec = max_timeout_sec; |
| 465 | |
| 466 | |
| 467 | /* |
| 468 | * Find the largest interrupt period, that can evenly divide |
| 469 | * the requested heartbeat time. |
| 470 | */ |
| 471 | while ((t % timeout_sec) != 0) |
| 472 | timeout_sec--; |
| 473 | |
| 474 | periods = t / timeout_sec; |
| 475 | |
| 476 | /* |
| 477 | * The last two periods are after the irq is disabled, and |
| 478 | * then to the nmi, so we subtract them off. |
| 479 | */ |
| 480 | |
| 481 | countdown_reset = periods > 2 ? periods - 2 : 0; |
| 482 | heartbeat = t; |
David Daney | 468ffde | 2010-10-07 16:03:52 -0700 | [diff] [blame] | 483 | timeout_cnt = ((octeon_get_io_clock_rate() >> 8) * timeout_sec) >> 8; |
David Daney | 4c076fb | 2010-07-24 10:16:05 -0700 | [diff] [blame] | 484 | } |
| 485 | |
| 486 | static int octeon_wdt_set_heartbeat(int t) |
| 487 | { |
| 488 | int cpu; |
| 489 | int coreid; |
| 490 | union cvmx_ciu_wdogx ciu_wdog; |
| 491 | |
| 492 | if (t <= 0) |
| 493 | return -1; |
| 494 | |
| 495 | octeon_wdt_calc_parameters(t); |
| 496 | |
| 497 | for_each_online_cpu(cpu) { |
| 498 | coreid = cpu2core(cpu); |
| 499 | cvmx_write_csr(CVMX_CIU_PP_POKEX(coreid), 1); |
| 500 | ciu_wdog.u64 = 0; |
| 501 | ciu_wdog.s.len = timeout_cnt; |
| 502 | ciu_wdog.s.mode = 3; /* 3 = Interrupt + NMI + Soft-Reset */ |
| 503 | cvmx_write_csr(CVMX_CIU_WDOGX(coreid), ciu_wdog.u64); |
| 504 | cvmx_write_csr(CVMX_CIU_PP_POKEX(coreid), 1); |
| 505 | } |
| 506 | octeon_wdt_ping(); /* Get the irqs back on. */ |
| 507 | return 0; |
| 508 | } |
| 509 | |
| 510 | /** |
| 511 | * octeon_wdt_write: |
| 512 | * @file: file handle to the watchdog |
| 513 | * @buf: buffer to write (unused as data does not matter here |
| 514 | * @count: count of bytes |
| 515 | * @ppos: pointer to the position to write. No seeks allowed |
| 516 | * |
| 517 | * A write to a watchdog device is defined as a keepalive signal. Any |
| 518 | * write of data will do, as we we don't define content meaning. |
| 519 | */ |
| 520 | |
| 521 | static ssize_t octeon_wdt_write(struct file *file, const char __user *buf, |
| 522 | size_t count, loff_t *ppos) |
| 523 | { |
| 524 | if (count) { |
| 525 | if (!nowayout) { |
| 526 | size_t i; |
| 527 | |
| 528 | /* In case it was set long ago */ |
| 529 | expect_close = 0; |
| 530 | |
| 531 | for (i = 0; i != count; i++) { |
| 532 | char c; |
| 533 | if (get_user(c, buf + i)) |
| 534 | return -EFAULT; |
| 535 | if (c == 'V') |
| 536 | expect_close = 1; |
| 537 | } |
| 538 | } |
| 539 | octeon_wdt_ping(); |
| 540 | } |
| 541 | return count; |
| 542 | } |
| 543 | |
| 544 | /** |
| 545 | * octeon_wdt_ioctl: |
| 546 | * @file: file handle to the device |
| 547 | * @cmd: watchdog command |
| 548 | * @arg: argument pointer |
| 549 | * |
| 550 | * The watchdog API defines a common set of functions for all |
| 551 | * watchdogs according to their available features. We only |
| 552 | * actually usefully support querying capabilities and setting |
| 553 | * the timeout. |
| 554 | */ |
| 555 | |
| 556 | static long octeon_wdt_ioctl(struct file *file, unsigned int cmd, |
| 557 | unsigned long arg) |
| 558 | { |
| 559 | void __user *argp = (void __user *)arg; |
| 560 | int __user *p = argp; |
| 561 | int new_heartbeat; |
| 562 | |
| 563 | static struct watchdog_info ident = { |
| 564 | .options = WDIOF_SETTIMEOUT| |
| 565 | WDIOF_MAGICCLOSE| |
| 566 | WDIOF_KEEPALIVEPING, |
| 567 | .firmware_version = 1, |
| 568 | .identity = "OCTEON", |
| 569 | }; |
| 570 | |
| 571 | switch (cmd) { |
| 572 | case WDIOC_GETSUPPORT: |
| 573 | return copy_to_user(argp, &ident, sizeof(ident)) ? -EFAULT : 0; |
| 574 | case WDIOC_GETSTATUS: |
| 575 | case WDIOC_GETBOOTSTATUS: |
| 576 | return put_user(0, p); |
| 577 | case WDIOC_KEEPALIVE: |
| 578 | octeon_wdt_ping(); |
| 579 | return 0; |
| 580 | case WDIOC_SETTIMEOUT: |
| 581 | if (get_user(new_heartbeat, p)) |
| 582 | return -EFAULT; |
| 583 | if (octeon_wdt_set_heartbeat(new_heartbeat)) |
| 584 | return -EINVAL; |
| 585 | /* Fall through. */ |
| 586 | case WDIOC_GETTIMEOUT: |
| 587 | return put_user(heartbeat, p); |
| 588 | default: |
| 589 | return -ENOTTY; |
| 590 | } |
| 591 | } |
| 592 | |
| 593 | /** |
| 594 | * octeon_wdt_open: |
| 595 | * @inode: inode of device |
| 596 | * @file: file handle to device |
| 597 | * |
| 598 | * The watchdog device has been opened. The watchdog device is single |
| 599 | * open and on opening we do a ping to reset the counters. |
| 600 | */ |
| 601 | |
| 602 | static int octeon_wdt_open(struct inode *inode, struct file *file) |
| 603 | { |
| 604 | if (test_and_set_bit(0, &octeon_wdt_is_open)) |
| 605 | return -EBUSY; |
| 606 | /* |
| 607 | * Activate |
| 608 | */ |
| 609 | octeon_wdt_ping(); |
| 610 | do_coundown = 1; |
| 611 | return nonseekable_open(inode, file); |
| 612 | } |
| 613 | |
| 614 | /** |
| 615 | * octeon_wdt_release: |
| 616 | * @inode: inode to board |
| 617 | * @file: file handle to board |
| 618 | * |
| 619 | * The watchdog has a configurable API. There is a religious dispute |
| 620 | * between people who want their watchdog to be able to shut down and |
| 621 | * those who want to be sure if the watchdog manager dies the machine |
| 622 | * reboots. In the former case we disable the counters, in the latter |
| 623 | * case you have to open it again very soon. |
| 624 | */ |
| 625 | |
| 626 | static int octeon_wdt_release(struct inode *inode, struct file *file) |
| 627 | { |
| 628 | if (expect_close) { |
| 629 | do_coundown = 0; |
| 630 | octeon_wdt_ping(); |
| 631 | } else { |
Joe Perches | 27c766a | 2012-02-15 15:06:19 -0800 | [diff] [blame] | 632 | pr_crit("WDT device closed unexpectedly. WDT will not stop!\n"); |
David Daney | 4c076fb | 2010-07-24 10:16:05 -0700 | [diff] [blame] | 633 | } |
| 634 | clear_bit(0, &octeon_wdt_is_open); |
| 635 | expect_close = 0; |
| 636 | return 0; |
| 637 | } |
| 638 | |
| 639 | static const struct file_operations octeon_wdt_fops = { |
| 640 | .owner = THIS_MODULE, |
| 641 | .llseek = no_llseek, |
| 642 | .write = octeon_wdt_write, |
| 643 | .unlocked_ioctl = octeon_wdt_ioctl, |
| 644 | .open = octeon_wdt_open, |
| 645 | .release = octeon_wdt_release, |
| 646 | }; |
| 647 | |
| 648 | static struct miscdevice octeon_wdt_miscdev = { |
| 649 | .minor = WATCHDOG_MINOR, |
| 650 | .name = "watchdog", |
| 651 | .fops = &octeon_wdt_fops, |
| 652 | }; |
| 653 | |
| 654 | static struct notifier_block octeon_wdt_cpu_notifier = { |
| 655 | .notifier_call = octeon_wdt_cpu_callback, |
| 656 | }; |
| 657 | |
| 658 | |
| 659 | /** |
| 660 | * Module/ driver initialization. |
| 661 | * |
| 662 | * Returns Zero on success |
| 663 | */ |
| 664 | static int __init octeon_wdt_init(void) |
| 665 | { |
| 666 | int i; |
| 667 | int ret; |
| 668 | int cpu; |
| 669 | u64 *ptr; |
| 670 | |
| 671 | /* |
| 672 | * Watchdog time expiration length = The 16 bits of LEN |
| 673 | * represent the most significant bits of a 24 bit decrementer |
| 674 | * that decrements every 256 cycles. |
| 675 | * |
| 676 | * Try for a timeout of 5 sec, if that fails a smaller number |
| 677 | * of even seconds, |
| 678 | */ |
| 679 | max_timeout_sec = 6; |
| 680 | do { |
| 681 | max_timeout_sec--; |
David Daney | 468ffde | 2010-10-07 16:03:52 -0700 | [diff] [blame] | 682 | timeout_cnt = ((octeon_get_io_clock_rate() >> 8) * max_timeout_sec) >> 8; |
David Daney | 4c076fb | 2010-07-24 10:16:05 -0700 | [diff] [blame] | 683 | } while (timeout_cnt > 65535); |
| 684 | |
| 685 | BUG_ON(timeout_cnt == 0); |
| 686 | |
| 687 | octeon_wdt_calc_parameters(heartbeat); |
| 688 | |
Joe Perches | 27c766a | 2012-02-15 15:06:19 -0800 | [diff] [blame] | 689 | pr_info("Initial granularity %d Sec\n", timeout_sec); |
David Daney | 4c076fb | 2010-07-24 10:16:05 -0700 | [diff] [blame] | 690 | |
| 691 | ret = misc_register(&octeon_wdt_miscdev); |
| 692 | if (ret) { |
Joe Perches | 27c766a | 2012-02-15 15:06:19 -0800 | [diff] [blame] | 693 | pr_err("cannot register miscdev on minor=%d (err=%d)\n", |
| 694 | WATCHDOG_MINOR, ret); |
David Daney | 4c076fb | 2010-07-24 10:16:05 -0700 | [diff] [blame] | 695 | goto out; |
| 696 | } |
| 697 | |
| 698 | /* Build the NMI handler ... */ |
| 699 | octeon_wdt_build_stage1(); |
| 700 | |
| 701 | /* ... and install it. */ |
| 702 | ptr = (u64 *) nmi_stage1_insns; |
| 703 | for (i = 0; i < 16; i++) { |
| 704 | cvmx_write_csr(CVMX_MIO_BOOT_LOC_ADR, i * 8); |
| 705 | cvmx_write_csr(CVMX_MIO_BOOT_LOC_DAT, ptr[i]); |
| 706 | } |
| 707 | cvmx_write_csr(CVMX_MIO_BOOT_LOC_CFGX(0), 0x81fc0000); |
| 708 | |
| 709 | cpumask_clear(&irq_enabled_cpus); |
| 710 | |
Srivatsa S. Bhat | 99c3bf3 | 2014-03-11 02:10:43 +0530 | [diff] [blame] | 711 | cpu_notifier_register_begin(); |
David Daney | 4c076fb | 2010-07-24 10:16:05 -0700 | [diff] [blame] | 712 | for_each_online_cpu(cpu) |
| 713 | octeon_wdt_setup_interrupt(cpu); |
| 714 | |
Srivatsa S. Bhat | 99c3bf3 | 2014-03-11 02:10:43 +0530 | [diff] [blame] | 715 | __register_hotcpu_notifier(&octeon_wdt_cpu_notifier); |
| 716 | cpu_notifier_register_done(); |
| 717 | |
David Daney | 4c076fb | 2010-07-24 10:16:05 -0700 | [diff] [blame] | 718 | out: |
| 719 | return ret; |
| 720 | } |
| 721 | |
| 722 | /** |
| 723 | * Module / driver shutdown |
| 724 | */ |
| 725 | static void __exit octeon_wdt_cleanup(void) |
| 726 | { |
| 727 | int cpu; |
| 728 | |
| 729 | misc_deregister(&octeon_wdt_miscdev); |
| 730 | |
Srivatsa S. Bhat | 99c3bf3 | 2014-03-11 02:10:43 +0530 | [diff] [blame] | 731 | cpu_notifier_register_begin(); |
| 732 | __unregister_hotcpu_notifier(&octeon_wdt_cpu_notifier); |
David Daney | 4c076fb | 2010-07-24 10:16:05 -0700 | [diff] [blame] | 733 | |
| 734 | for_each_online_cpu(cpu) { |
| 735 | int core = cpu2core(cpu); |
| 736 | /* Disable the watchdog */ |
| 737 | cvmx_write_csr(CVMX_CIU_WDOGX(core), 0); |
| 738 | /* Free the interrupt handler */ |
| 739 | free_irq(OCTEON_IRQ_WDOG0 + core, octeon_wdt_poke_irq); |
| 740 | } |
Srivatsa S. Bhat | 99c3bf3 | 2014-03-11 02:10:43 +0530 | [diff] [blame] | 741 | |
| 742 | cpu_notifier_register_done(); |
| 743 | |
David Daney | 4c076fb | 2010-07-24 10:16:05 -0700 | [diff] [blame] | 744 | /* |
| 745 | * Disable the boot-bus memory, the code it points to is soon |
| 746 | * to go missing. |
| 747 | */ |
| 748 | cvmx_write_csr(CVMX_MIO_BOOT_LOC_CFGX(0), 0); |
| 749 | } |
| 750 | |
| 751 | MODULE_LICENSE("GPL"); |
| 752 | MODULE_AUTHOR("Cavium Networks <support@caviumnetworks.com>"); |
| 753 | MODULE_DESCRIPTION("Cavium Networks Octeon Watchdog driver."); |
| 754 | module_init(octeon_wdt_init); |
| 755 | module_exit(octeon_wdt_cleanup); |