drivers/watchdog/hpwdt.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *      HPE WatchDog Driver
   4  *      based on
   5  *
   6  *      SoftDog 0.05:   A Software Watchdog Device
   7  *
   8  *      (c) Copyright 2018 Hewlett Packard Enterprise Development LP
   9  *      Thomas Mingarelli <thomas.mingarelli@hpe.com>
  10  */
  11
  12 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  13
  14 #include <linux/device.h>
  15 #include <linux/io.h>
  16 #include <linux/kernel.h>
  17 #include <linux/module.h>
  18 #include <linux/moduleparam.h>
  19 #include <linux/pci.h>
  20 #include <linux/pci_ids.h>
  21 #include <linux/types.h>
  22 #include <linux/watchdog.h>
  23 #include <asm/nmi.h>
  24 #include <linux/crash_dump.h>
  25
  26 #define HPWDT_VERSION                   "2.0.4"
  27 #define SECS_TO_TICKS(secs)             ((secs) * 1000 / 128)
  28 #define TICKS_TO_SECS(ticks)            ((ticks) * 128 / 1000)
  29 #define HPWDT_MAX_TICKS                 65535
  30 #define HPWDT_MAX_TIMER                 TICKS_TO_SECS(HPWDT_MAX_TICKS)
  31 #define DEFAULT_MARGIN                  30
  32 #define PRETIMEOUT_SEC                  9
  33
  34 static bool ilo5;
  35 static unsigned int soft_margin = DEFAULT_MARGIN;       /* in seconds */
  36 static bool nowayout = WATCHDOG_NOWAYOUT;
  37 static bool pretimeout = IS_ENABLED(CONFIG_HPWDT_NMI_DECODING);
  38 static int kdumptimeout = -1;
  39
  40 static void __iomem *pci_mem_addr;              /* the PCI-memory address */
  41 static unsigned long __iomem *hpwdt_nmistat;
  42 static unsigned long __iomem *hpwdt_timer_reg;
  43 static unsigned long __iomem *hpwdt_timer_con;
  44
  45 static const struct pci_device_id hpwdt_devices[] = {
  46         { PCI_DEVICE(PCI_VENDOR_ID_COMPAQ, 0xB203) },   /* iLO2 */
  47         { PCI_DEVICE(PCI_VENDOR_ID_HP, 0x3306) },       /* iLO3 */
  48         { PCI_DEVICE(PCI_VENDOR_ID_HP_3PAR, 0x0389) },  /* PCtrl */
  49         {0},                    /* terminate list */
  50 };
  51 MODULE_DEVICE_TABLE(pci, hpwdt_devices);
  52
  53 static const struct pci_device_id hpwdt_blacklist[] = {
  54         { PCI_DEVICE_SUB(PCI_VENDOR_ID_HP, 0x3306, PCI_VENDOR_ID_HP, 0x1979) }, /* auxilary iLO */
  55         { PCI_DEVICE_SUB(PCI_VENDOR_ID_HP, 0x3306, PCI_VENDOR_ID_HP_3PAR, 0x0289) },  /* CL */
  56         {0},                    /* terminate list */
  57 };
  58
  59 static struct watchdog_device hpwdt_dev;
  60 /*
  61  *      Watchdog operations
  62  */
  63 static int hpwdt_hw_is_running(void)
  64 {
  65         return ioread8(hpwdt_timer_con) & 0x01;
  66 }
  67
  68 static int hpwdt_start(struct watchdog_device *wdd)
  69 {
  70         int control = 0x81 | (pretimeout ? 0x4 : 0);
  71         int reload = SECS_TO_TICKS(min(wdd->timeout, wdd->max_hw_heartbeat_ms/1000));
  72
  73         dev_dbg(wdd->parent, "start watchdog 0x%08x:0x%08x:0x%02x\n", wdd->timeout, reload, control);
  74         iowrite16(reload, hpwdt_timer_reg);
  75         iowrite8(control, hpwdt_timer_con);
  76
  77         return 0;
  78 }
  79
  80 static void hpwdt_stop(void)
  81 {
  82         unsigned long data;
  83
  84         pr_debug("stop  watchdog\n");
  85
  86         data = ioread8(hpwdt_timer_con);
  87         data &= 0xFE;
  88         iowrite8(data, hpwdt_timer_con);
  89 }
  90
  91 static int hpwdt_stop_core(struct watchdog_device *wdd)
  92 {
  93         hpwdt_stop();
  94
  95         return 0;
  96 }
  97
  98 static void hpwdt_ping_ticks(int val)
  99 {
 100         val = min(val, HPWDT_MAX_TICKS);
 101         iowrite16(val, hpwdt_timer_reg);
 102 }
 103
 104 static int hpwdt_ping(struct watchdog_device *wdd)
 105 {
 106         int reload = SECS_TO_TICKS(min(wdd->timeout, wdd->max_hw_heartbeat_ms/1000));
 107
 108         dev_dbg(wdd->parent, "ping  watchdog 0x%08x:0x%08x\n", wdd->timeout, reload);
 109         hpwdt_ping_ticks(reload);
 110
 111         return 0;
 112 }
 113
 114 static unsigned int hpwdt_gettimeleft(struct watchdog_device *wdd)
 115 {
 116         return TICKS_TO_SECS(ioread16(hpwdt_timer_reg));
 117 }
 118
 119 static int hpwdt_settimeout(struct watchdog_device *wdd, unsigned int val)
 120 {
 121         dev_dbg(wdd->parent, "set_timeout = %d\n", val);
 122
 123         wdd->timeout = val;
 124         if (val <= wdd->pretimeout) {
 125                 dev_dbg(wdd->parent, "pretimeout < timeout. Setting to zero\n");
 126                 wdd->pretimeout = 0;
 127                 pretimeout = false;
 128                 if (watchdog_active(wdd))
 129                         hpwdt_start(wdd);
 130         }
 131         hpwdt_ping(wdd);
 132
 133         return 0;
 134 }
 135
 136 #ifdef CONFIG_HPWDT_NMI_DECODING
 137 static int hpwdt_set_pretimeout(struct watchdog_device *wdd, unsigned int req)
 138 {
 139         unsigned int val = 0;
 140
 141         dev_dbg(wdd->parent, "set_pretimeout = %d\n", req);
 142         if (req) {
 143                 val = PRETIMEOUT_SEC;
 144                 if (val >= wdd->timeout)
 145                         return -EINVAL;
 146         }
 147
 148         if (val != req)
 149                 dev_dbg(wdd->parent, "Rounding pretimeout to: %d\n", val);
 150
 151         wdd->pretimeout = val;
 152         pretimeout = !!val;
 153
 154         if (watchdog_active(wdd))
 155                 hpwdt_start(wdd);
 156
 157         return 0;
 158 }
 159
 160 static int hpwdt_my_nmi(void)
 161 {
 162         return ioread8(hpwdt_nmistat) & 0x6;
 163 }
 164
 165 /*
 166  *      NMI Handler
 167  */
 168 static int hpwdt_pretimeout(unsigned int ulReason, struct pt_regs *regs)
 169 {
 170         unsigned int mynmi = hpwdt_my_nmi();
 171         static char panic_msg[] =
 172                 "00: An NMI occurred. Depending on your system the reason "
 173                 "for the NMI is logged in any one of the following resources:\n"
 174                 "1. Integrated Management Log (IML)\n"
 175                 "2. OA Syslog\n"
 176                 "3. OA Forward Progress Log\n"
 177                 "4. iLO Event Log";
 178
 179         if (ilo5 && ulReason == NMI_UNKNOWN && !mynmi)
 180                 return NMI_DONE;
 181
 182         if (ilo5 && !pretimeout && !mynmi)
 183                 return NMI_DONE;
 184
 185         if (kdumptimeout < 0)
 186                 hpwdt_stop();
 187         else if (kdumptimeout == 0)
 188                 ;
 189         else {
 190                 unsigned int val = max((unsigned int)kdumptimeout, hpwdt_dev.timeout);
 191                 hpwdt_ping_ticks(SECS_TO_TICKS(val));
 192         }
 193
 194         hex_byte_pack(panic_msg, mynmi);
 195         nmi_panic(regs, panic_msg);
 196
 197         return NMI_HANDLED;
 198 }
 199 #endif /* CONFIG_HPWDT_NMI_DECODING */
 200
 201
 202 static const struct watchdog_info ident = {
 203         .options = WDIOF_PRETIMEOUT    |
 204                    WDIOF_SETTIMEOUT    |
 205                    WDIOF_KEEPALIVEPING |
 206                    WDIOF_MAGICCLOSE,
 207         .identity = "HPE iLO2+ HW Watchdog Timer",
 208 };
 209
 210 /*
 211  *      Kernel interfaces
 212  */
 213
 214 static const struct watchdog_ops hpwdt_ops = {
 215         .owner          = THIS_MODULE,
 216         .start          = hpwdt_start,
 217         .stop           = hpwdt_stop_core,
 218         .ping           = hpwdt_ping,
 219         .set_timeout    = hpwdt_settimeout,
 220         .get_timeleft   = hpwdt_gettimeleft,
 221 #ifdef CONFIG_HPWDT_NMI_DECODING
 222         .set_pretimeout = hpwdt_set_pretimeout,
 223 #endif
 224 };
 225
 226 static struct watchdog_device hpwdt_dev = {
 227         .info           = &ident,
 228         .ops            = &hpwdt_ops,
 229         .min_timeout    = 1,
 230         .timeout        = DEFAULT_MARGIN,
 231         .pretimeout     = PRETIMEOUT_SEC,
 232         .max_hw_heartbeat_ms    = HPWDT_MAX_TIMER * 1000,
 233 };
 234
 235
 236 /*
 237  *      Init & Exit
 238  */
 239
 240 static int hpwdt_init_nmi_decoding(struct pci_dev *dev)
 241 {
 242 #ifdef CONFIG_HPWDT_NMI_DECODING
 243         int retval;
 244         /*
 245          * Only one function can register for NMI_UNKNOWN
 246          */
 247         retval = register_nmi_handler(NMI_UNKNOWN, hpwdt_pretimeout, 0, "hpwdt");
 248         if (retval)
 249                 goto error;
 250         retval = register_nmi_handler(NMI_SERR, hpwdt_pretimeout, 0, "hpwdt");
 251         if (retval)
 252                 goto error1;
 253         retval = register_nmi_handler(NMI_IO_CHECK, hpwdt_pretimeout, 0, "hpwdt");
 254         if (retval)
 255                 goto error2;
 256
 257         dev_info(&dev->dev,
 258                 "HPE Watchdog Timer Driver: NMI decoding initialized\n");
 259
 260         return 0;
 261
 262 error2:
 263         unregister_nmi_handler(NMI_SERR, "hpwdt");
 264 error1:
 265         unregister_nmi_handler(NMI_UNKNOWN, "hpwdt");
 266 error:
 267         dev_warn(&dev->dev,
 268                 "Unable to register a die notifier (err=%d).\n",
 269                 retval);
 270         return retval;
 271 #endif  /* CONFIG_HPWDT_NMI_DECODING */
 272         return 0;
 273 }
 274
 275 static void hpwdt_exit_nmi_decoding(void)
 276 {
 277 #ifdef CONFIG_HPWDT_NMI_DECODING
 278         unregister_nmi_handler(NMI_UNKNOWN, "hpwdt");
 279         unregister_nmi_handler(NMI_SERR, "hpwdt");
 280         unregister_nmi_handler(NMI_IO_CHECK, "hpwdt");
 281 #endif
 282 }
 283
 284 static int hpwdt_init_one(struct pci_dev *dev,
 285                                         const struct pci_device_id *ent)
 286 {
 287         int retval;
 288
 289         /*
 290          * First let's find out if we are on an iLO2+ server. We will
 291          * not run on a legacy ASM box.
 292          * So we only support the G5 ProLiant servers and higher.
 293          */
 294         if (dev->subsystem_vendor != PCI_VENDOR_ID_HP &&
 295             dev->subsystem_vendor != PCI_VENDOR_ID_HP_3PAR) {
 296                 dev_warn(&dev->dev,
 297                         "This server does not have an iLO2+ ASIC.\n");
 298                 return -ENODEV;
 299         }
 300
 301         if (pci_match_id(hpwdt_blacklist, dev)) {
 302                 dev_dbg(&dev->dev, "Not supported on this device\n");
 303                 return -ENODEV;
 304         }
 305
 306         if (pci_enable_device(dev)) {
 307                 dev_warn(&dev->dev,
 308                         "Not possible to enable PCI Device: 0x%x:0x%x.\n",
 309                         ent->vendor, ent->device);
 310                 return -ENODEV;
 311         }
 312
 313         pci_mem_addr = pci_iomap(dev, 1, 0x80);
 314         if (!pci_mem_addr) {
 315                 dev_warn(&dev->dev,
 316                         "Unable to detect the iLO2+ server memory.\n");
 317                 retval = -ENOMEM;
 318                 goto error_pci_iomap;
 319         }
 320         hpwdt_nmistat   = pci_mem_addr + 0x6e;
 321         hpwdt_timer_reg = pci_mem_addr + 0x70;
 322         hpwdt_timer_con = pci_mem_addr + 0x72;
 323
 324         /* Have the core update running timer until user space is ready */
 325         if (hpwdt_hw_is_running()) {
 326                 dev_info(&dev->dev, "timer is running\n");
 327                 set_bit(WDOG_HW_RUNNING, &hpwdt_dev.status);
 328         }
 329
 330         /* Initialize NMI Decoding functionality */
 331         retval = hpwdt_init_nmi_decoding(dev);
 332         if (retval != 0)
 333                 goto error_init_nmi_decoding;
 334
 335         watchdog_stop_on_unregister(&hpwdt_dev);
 336         watchdog_set_nowayout(&hpwdt_dev, nowayout);
 337         watchdog_init_timeout(&hpwdt_dev, soft_margin, NULL);
 338
 339         if (is_kdump_kernel()) {
 340                 pretimeout = false;
 341                 kdumptimeout = 0;
 342         }
 343
 344         if (pretimeout && hpwdt_dev.timeout <= PRETIMEOUT_SEC) {
 345                 dev_warn(&dev->dev, "timeout <= pretimeout. Setting pretimeout to zero\n");
 346                 pretimeout = false;
 347         }
 348         hpwdt_dev.pretimeout = pretimeout ? PRETIMEOUT_SEC : 0;
 349         kdumptimeout = min(kdumptimeout, HPWDT_MAX_TIMER);
 350
 351         hpwdt_dev.parent = &dev->dev;
 352         retval = watchdog_register_device(&hpwdt_dev);
 353         if (retval < 0)
 354                 goto error_wd_register;
 355
 356         dev_info(&dev->dev, "HPE Watchdog Timer Driver: Version: %s\n",
 357                                 HPWDT_VERSION);
 358         dev_info(&dev->dev, "timeout: %d seconds (nowayout=%d)\n",
 359                                 hpwdt_dev.timeout, nowayout);
 360         dev_info(&dev->dev, "pretimeout: %s.\n",
 361                                 pretimeout ? "on" : "off");
 362         dev_info(&dev->dev, "kdumptimeout: %d.\n", kdumptimeout);
 363
 364         if (dev->subsystem_vendor == PCI_VENDOR_ID_HP_3PAR)
 365                 ilo5 = true;
 366
 367         return 0;
 368
 369 error_wd_register:
 370         hpwdt_exit_nmi_decoding();
 371 error_init_nmi_decoding:
 372         pci_iounmap(dev, pci_mem_addr);
 373 error_pci_iomap:
 374         pci_disable_device(dev);
 375         return retval;
 376 }
 377
 378 static void hpwdt_exit(struct pci_dev *dev)
 379 {
 380         watchdog_unregister_device(&hpwdt_dev);
 381         hpwdt_exit_nmi_decoding();
 382         pci_iounmap(dev, pci_mem_addr);
 383         pci_disable_device(dev);
 384 }
 385
 386 static struct pci_driver hpwdt_driver = {
 387         .name = "hpwdt",
 388         .id_table = hpwdt_devices,
 389         .probe = hpwdt_init_one,
 390         .remove = hpwdt_exit,
 391 };
 392
 393 MODULE_AUTHOR("Tom Mingarelli");
 394 MODULE_DESCRIPTION("hpe watchdog driver");
 395 MODULE_LICENSE("GPL");
 396 MODULE_VERSION(HPWDT_VERSION);
 397
 398 module_param(soft_margin, int, 0);
 399 MODULE_PARM_DESC(soft_margin, "Watchdog timeout in seconds");
 400
 401 module_param_named(timeout, soft_margin, int, 0);
 402 MODULE_PARM_DESC(timeout, "Alias of soft_margin");
 403
 404 module_param(nowayout, bool, 0);
 405 MODULE_PARM_DESC(nowayout, "Watchdog cannot be stopped once started (default="
 406                 __MODULE_STRING(WATCHDOG_NOWAYOUT) ")");
 407
 408 module_param(kdumptimeout, int, 0444);
 409 MODULE_PARM_DESC(kdumptimeout, "Timeout applied for crash kernel transition in seconds");
 410
 411 #ifdef CONFIG_HPWDT_NMI_DECODING
 412 module_param(pretimeout, bool, 0);
 413 MODULE_PARM_DESC(pretimeout, "Watchdog pretimeout enabled");
 414 #endif
 415
 416 module_pci_driver(hpwdt_driver);