mlxbf-bootctl.c

/*
 * Copyright (c) 2017-2018, Mellanox Technologies Ltd.
 * All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 *
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 *
 * 2. Redistributions in binary form must reproduce the above
 *    copyright notice, this list of conditions and the following
 *    disclaimer in the documentation and/or other materials provided
 *    with the distribution.
 *
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
 * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
 * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
 * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
 * OF THE POSSIBILITY OF SUCH DAMAGE.
 */

#define _GNU_SOURCE  // asprintf
#include <errno.h>
#include <getopt.h>
#include <stdarg.h>
#include <stdbool.h>
#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
#include <sys/fcntl.h>
#include <sys/ioctl.h>
#include <sys/stat.h>
#include <linux/limits.h>
#include <linux/major.h>

/* Boot FIFO constants */
#define BOOT_FIFO_ADDR 0x0408
#define SEGMENT_HEADER_LEN 8
#define MAX_SEG_LEN ((1 << 20) - SEGMENT_HEADER_LEN)
#define SEGMENT_IS_END (1UL << 63)

/* File size limits */
#define INPUT_BFB_MAX_SIZE (20 * 1024 * 1024)
#define MMC_BOOT_PARTITION_MAX_SIZE (4 * 1024 * 1024)

/* DMI constants */
#define DMI_TABLE_PATH "/sys/firmware/dmi/tables/DMI"
#define DMI_PROCESSOR_INFO_TYPE 4
#define DMI_PROCESSOR_VERSION_STR_MAX_SIZE 100

/* PSC APP image ID */
#define PSC_APP_IMG_ID	39

/* Record size of irot signature db. */
#define PSC_APP_IROT_SIG_REC_LEN 104

/* TLV based PSC APP sub-image header. */
typedef struct  __attribute__((__packed__)) {
#define PSC_APP_TYPE_IROT_SIGNATURE_DB 0x1
  uint8_t type;           /* sub-image type */
  uint32_t length;        /* sub-image payload length */
  uint8_t unused[3];
} psc_app_img_hdr_t;

/*
 * Boot watchdog (ARM SBSA) modes of operation.
 *
 * Disabled: watchdog will not be armed for boot (unless used for boot swap).
 * Standard: watchdog will be armed at boot start and refreshed periodically.
 * Time Limit: watchdog will be armed at boot start and never refreshed. In
 *             special cases (like booting in livefish mode or entering UEFI
 *             menu) the watchdog will be disabled.
 */
enum {
  BF_BOOT_WDOG_MODE_DISABLED = 0,
  BF_BOOT_WDOG_MODE_STANDARD = 1,
  BF_BOOT_WDOG_MODE_TIME_LIMIT = 2,
  BF_BOOT_WDOG_MODE_MAX = BF_BOOT_WDOG_MODE_TIME_LIMIT
};

/*
 * Boot watchdog (ARM SBSA) configuration.
 * This struct must match the definition in ATF and UEFI.
 */
typedef union {
	struct __attribute__((__packed__)) {
		uint16_t interval;
		uint8_t mode;
	};
	uint64_t word;
} bf_boot_watchdog_config_t;

/* Boot watchdog mode strings. Can be indexed with enum values above. */
const char * const bf_boot_wdog_mode_str[] = {"disabled", "standard", "time_limit"};

/* Boot watchdog settings */
#define MIN_WDOG_INTERVAL_SEC 45
#define MAX_WDOG_INTERVAL_SEC 4095
#define DEFAULT_WDOG_INTERVAL_SEC 180

/* Other constants */
#define MAX_VERSIONS_COUNT 256
#define BF3_VERSION 2

void die(const char* fmt, ...)
{
  va_list ap;
  va_start(ap, fmt);
  vfprintf(stderr, fmt, ap);
  putc('\n', stderr);
  va_end(ap);
  exit(1);
}

/*
 * BFB image header.
 *
 * This definition is extracted from file
 * atf/plat/mellanox/common/include/drivers/io/bluefield_boot.h
 */
#define BFB_IMGHDR_MAGIC        0x13026642  /* "Bf^B^S" */
typedef struct {
    unsigned long magic:32;
    unsigned long major:4;
    unsigned long minor:4;
    unsigned long reserved:4;
    unsigned long next_img_ver:4;
    unsigned long cur_img_ver:4;
    unsigned long hdr_len:4;
    unsigned long image_id:8;
    unsigned long image_len:32;
    unsigned long image_crc:32;
    unsigned long following_images:64;
} boot_image_header_t;

ssize_t read_or_die(const char* filename, int fd, void* buf, size_t count);

#ifndef OUTPUT_ONLY

#include <linux/mmc/ioctl.h>

/*
 * The Linux MMC driver doesn't export its ioctl command values, so we
 * copy them from include/linux/mmc/mmc.h and include/linux/mmc/core.h.
 */
#define MMC_SWITCH                6   /* ac   [31:0] See below   R1b */
#define MMC_SEND_EXT_CSD          8   /* adtc                    R1  */
#define MMC_RSP_PRESENT	(1 << 0)
#define MMC_RSP_CRC	(1 << 2)		/* expect valid crc */
#define MMC_RSP_BUSY	(1 << 3)		/* card may send busy */
#define MMC_RSP_OPCODE	(1 << 4)		/* response contains opcode */
#define MMC_RSP_SPI_S1	(1 << 7)		/* one status byte */
#define MMC_RSP_SPI_BUSY (1 << 10)		/* card may send busy */
#define MMC_RSP_SPI_R1	(MMC_RSP_SPI_S1)
#define MMC_RSP_R1	(MMC_RSP_PRESENT|MMC_RSP_CRC|MMC_RSP_OPCODE)
#define MMC_RSP_R1B	(MMC_RSP_PRESENT|MMC_RSP_CRC|MMC_RSP_OPCODE|MMC_RSP_BUSY)
#define MMC_RSP_SPI_R1B	(MMC_RSP_SPI_S1|MMC_RSP_SPI_BUSY)
#define MMC_CMD_AC	(0 << 5)
#define MMC_CMD_ADTC	(1 << 5)
#define MMC_SWITCH_MODE_WRITE_BYTE	0x03	/* Set target to value */
#define EXT_CSD_CMD_SET_NORMAL		(1<<0)

/* EXT_CSD register offset. */
#define EXT_CSD_RST_N			162	/* R/W */
#define EXT_CSD_BOOT_BUS_WIDTH		177	/* R/W */
#define EXT_CSD_PART_CONFIG		179	/* R/W */
#define EXT_CSD_BOOT_SIZE_MULT		226	/* R/W */

/* BOOT_BUS_WIDTH register definition. */
#define EXT_CSD_BOOT_BUS_WIDTH_MASK_ALL	0x7
#define EXT_CSD_BOOT_BUS_WIDTH_MASK	0x3
#define EXT_CSD_BOOT_BUS_WIDTH_RESET_MASK	0x4
#define EXT_CSD_BOOT_BUS_WIDTH_X8	0x6

/* EXT_CSD_RST_N register definition. */
#define EXT_CSD_RST_N_MASK		0x3
#define EXT_CSD_RST_N_ENABLE		0x1

/* Program constants */
#define EMMC_MIN_BOOT_SIZE 0x20000
#define EMMC_BLOCK_SIZE 512

#define SYS_PATH1 "/sys/bus/platform/devices/MLNXBF04:00/driver"
#define SYS_PATH2 "/sys/bus/platform/devices/MLNXBF04:00"
#define SECOND_RESET_ACTION_PATH "second_reset_action"
#define POST_RESET_WDOG_PATH "post_reset_wdog"
#define LIFECYCLE_STATE_PATH "lifecycle_state"
#define SECURE_BOOT_FUSE_STATE_PATH "secure_boot_fuse_state"

/* Program variables */
const char *mmc_path = "/dev/mmcblk0";

/* PSC cert update filtering info */
static uint8_t psc_cert_update;    /* whether cert need update. */
static uint64_t psc_cert_key;      /* cert signature lookup key */

/* Run an MMC_IOC_CMD ioctl on mmc_path */
void mmc_command(struct mmc_ioc_cmd *idata)
{
  static int mmc_fd = -1;
  if (mmc_fd < 0)
  {
    mmc_fd = open(mmc_path, O_RDWR);
    if (mmc_fd < 0)
      die("%s: %m", mmc_path);
  }
  if (ioctl(mmc_fd, MMC_IOC_CMD, idata) < 0)
    die("%s: mmc ioctl: %m", mmc_path);
}

uint8_t* get_ext_csd(void)
{
  static uint8_t ext_csd[EMMC_BLOCK_SIZE]
    __attribute__((aligned(EMMC_BLOCK_SIZE))) = { 0 };
  struct mmc_ioc_cmd idata = {
    .write_flag = 0,
    .opcode = MMC_SEND_EXT_CSD,
    .arg = 0,
    .flags = MMC_RSP_SPI_R1 | MMC_RSP_R1 | MMC_CMD_ADTC,
    .blksz = EMMC_BLOCK_SIZE,
    .blocks = 1
  };

  mmc_ioc_cmd_set_data(idata, ext_csd);
  mmc_command(&idata);

  return ext_csd;
}

/* Return the current partition (0 or 1) that we will boot from. */
int get_boot_partition(void)
{
  uint8_t *ext_csd = get_ext_csd();
  int part = ((ext_csd[EXT_CSD_PART_CONFIG] >> 3) & 0x7) - 1;

  /* Set part to 0 by default if it is -1 (boot disabled). */
  if (part < 0 || part > 1)
    part = 0;

  return part;
}

/* Set which partition to boot from. */
void set_boot_partition(int part)
{
  int value = ((part + 1) & 0x7) << 3;  /* Adjust for 1-based numbering */
  struct mmc_ioc_cmd idata = {
    .write_flag = 1,
    .opcode = MMC_SWITCH,
    .arg = ((MMC_SWITCH_MODE_WRITE_BYTE << 24) |
            (EXT_CSD_PART_CONFIG << 16) |
            (value << 8) |
            EXT_CSD_CMD_SET_NORMAL),
    .flags = MMC_RSP_SPI_R1B | MMC_RSP_R1B | MMC_CMD_AC
  };

  mmc_command(&idata);
}

/* Get the boot partition size */
uint64_t get_boot_partition_size(void)
{
  uint64_t part_size;
  uint8_t *ext_csd = get_ext_csd();
  part_size = (ext_csd[EXT_CSD_BOOT_SIZE_MULT]) *
                  EMMC_MIN_BOOT_SIZE;

  return part_size;
}

/* Return the current boot bus width. */
int get_boot_bus_width(void)
{
  uint8_t *ext_csd = get_ext_csd();
  return ext_csd[EXT_CSD_BOOT_BUS_WIDTH] & EXT_CSD_BOOT_BUS_WIDTH_MASK_ALL;
}

/* Set the boot-bus-width to 8-bit mode in ECSD. */
void set_boot_bus_width(void)
{
  uint8_t *ext_csd = get_ext_csd();
  int value = (ext_csd[EXT_CSD_BOOT_BUS_WIDTH] &
    ~EXT_CSD_BOOT_BUS_WIDTH_MASK_ALL) | EXT_CSD_BOOT_BUS_WIDTH_X8;
  struct mmc_ioc_cmd idata = {
    .write_flag = 1,
    .opcode = MMC_SWITCH,
    .arg = ((MMC_SWITCH_MODE_WRITE_BYTE << 24) |
            (EXT_CSD_BOOT_BUS_WIDTH << 16) |
            (value << 8) |
            EXT_CSD_CMD_SET_NORMAL),
    .flags = MMC_RSP_SPI_R1B | MMC_RSP_R1B | MMC_CMD_AC
  };

  mmc_command(&idata);
}

void enable_rst_n(void)
{
  uint8_t *ext_csd = get_ext_csd();
  int value = (ext_csd[EXT_CSD_RST_N] & ~EXT_CSD_RST_N_MASK) |
    EXT_CSD_RST_N_ENABLE;
  struct mmc_ioc_cmd idata = {
    .write_flag = 1,
    .opcode = MMC_SWITCH,
    .arg = ((MMC_SWITCH_MODE_WRITE_BYTE << 24) |
            (EXT_CSD_RST_N << 16) |
            (value << 8) |
            EXT_CSD_CMD_SET_NORMAL),
    .flags = MMC_RSP_SPI_R1B | MMC_RSP_R1B | MMC_CMD_AC
  };

  mmc_command(&idata);
}

FILE *open_sysfs(const char *name, const char *attr)
{
  FILE *f;
  char path[PATH_MAX];

  sprintf(path, "%s/%s", SYS_PATH1, name);
  f = fopen(path, attr);
  if (f == NULL) {
    sprintf(path, "%s/%s", SYS_PATH2, name);
    f = fopen(path, attr);
  }

  if (f == NULL) {
    if (errno == ENOENT) {
      fprintf(stderr, "cannot find required sysfs path %s\n", path);
      die("please load mlxbf_bootctl kernel driver");
    }

    die("%s: %m", name);
  }

  return f;
}

bf_boot_watchdog_config_t get_watchdog_config(void)
{
  bf_boot_watchdog_config_t watchdog_config = {.word = 0};
  FILE *f = open_sysfs(POST_RESET_WDOG_PATH, "r");

  if (fscanf(f, "%d", (int *) &watchdog_config.word) != 1)
    die("%s: failed to read integer", POST_RESET_WDOG_PATH);
  fclose(f);

  return watchdog_config;
}

const char *get_watchdog_mode_str(uint8_t watchdog_mode)
{
  if (watchdog_mode > BF_BOOT_WDOG_MODE_MAX) {
    return "unknown";
  }

  return bf_boot_wdog_mode_str[watchdog_mode];
}

void set_watchdog(uint16_t interval, uint8_t mode)
{
  FILE *f = NULL;
  bf_boot_watchdog_config_t watchdog_config = {.word = 0};

  if (mode > BF_BOOT_WDOG_MODE_MAX) {
    die("Watchdog mode '%d' is unsupported.", mode);
  }

  if (interval == 0) {
    mode = BF_BOOT_WDOG_MODE_DISABLED;
  } else if (interval < MIN_WDOG_INTERVAL_SEC) {
    die("Watchdog interval '%d' is too small (must be >= %ds).", interval, MIN_WDOG_INTERVAL_SEC);
  } else if (interval > MAX_WDOG_INTERVAL_SEC) {
    die("Watchdog interval '%d' is too large (must be < %ds).", interval, MAX_WDOG_INTERVAL_SEC);
  }

  watchdog_config.mode = mode;
  watchdog_config.interval = interval;

  f = open_sysfs(POST_RESET_WDOG_PATH, "w");
  if (fprintf(f, "%d\n", (int) watchdog_config.word) < 0)
    die("%s: failed to set watchdog to '0x%X'", POST_RESET_WDOG_PATH, watchdog_config.word);
  fclose(f);

  watchdog_config = get_watchdog_config();
  if (mode != watchdog_config.mode) {
    die("Failed to set watchdog due to outdated ATF version.");
  }

  if (interval != watchdog_config.interval) {
    die("Failed to set watchdog interval!");
  }
}

uint16_t get_watchdog_interval(const char *watchdog_interval_str)
{
  uint16_t watchdog_interval = DEFAULT_WDOG_INTERVAL_SEC;
  char *end;

  if (!watchdog_interval_str)
    return watchdog_interval;

  watchdog_interval = strtol(watchdog_interval_str, &end, 0);
  if (end == watchdog_interval_str || *end != '\0')
    die("watchdog interval ('%s') must be an integer", watchdog_interval_str);

  if (watchdog_interval != 0 &&
    (watchdog_interval < MIN_WDOG_INTERVAL_SEC || watchdog_interval > MAX_WDOG_INTERVAL_SEC)) {
    die("watchdog interval ('%s') must be between %d-%d",
        watchdog_interval_str, MIN_WDOG_INTERVAL_SEC, MAX_WDOG_INTERVAL_SEC);
  }

  return watchdog_interval;
}

uint8_t get_watchdog_mode(const char *watchdog_mode_str)
{
  uint8_t watchdog_mode = BF_BOOT_WDOG_MODE_DISABLED;

  if (!watchdog_mode_str)
    return watchdog_mode;

  if (strcmp(watchdog_mode_str,
             bf_boot_wdog_mode_str[BF_BOOT_WDOG_MODE_DISABLED]) == 0) {
    watchdog_mode = BF_BOOT_WDOG_MODE_DISABLED;
  } else if (strcmp(watchdog_mode_str,
                    bf_boot_wdog_mode_str[BF_BOOT_WDOG_MODE_STANDARD]) == 0) {
    watchdog_mode = BF_BOOT_WDOG_MODE_STANDARD;
  } else if (strcmp(watchdog_mode_str,
                    bf_boot_wdog_mode_str[BF_BOOT_WDOG_MODE_TIME_LIMIT]) == 0) {
    watchdog_mode = BF_BOOT_WDOG_MODE_TIME_LIMIT;
  } else {
    die("Invalid watchdog mode '%s'", watchdog_mode_str);
  }

  return watchdog_mode;
}

void set_second_reset_action(const char *action)
{
  FILE *f = open_sysfs(SECOND_RESET_ACTION_PATH, "w");
  if (fprintf(f, "%s\n", action) < 0)
    die("%s: failed to set action to '%s'", SECOND_RESET_ACTION_PATH, action);
  fclose(f);
}

void configure_watchdog(bool watchdog_swap, bool nowatchdog_swap,
                        const char *watchdog_mode_str, const char *watchdog_interval_str)
{
  uint16_t watchdog_interval;
  uint8_t watchdog_mode;

  if (!nowatchdog_swap && !watchdog_swap && !watchdog_mode_str && !watchdog_interval_str)
    return;

  watchdog_interval = get_watchdog_interval(watchdog_interval_str);
  watchdog_mode = get_watchdog_mode(watchdog_mode_str);

  if (watchdog_swap) {
    // Swap eMMC on reset after watchdog interval.
    // Ensure watchdog mode 0 is always used for boot swap so
    // this command is still compatible with older ATF versions.
    // The watchdog will be enabled for the next boot, but then
    // disabled for all future boots after that.
    set_watchdog(watchdog_interval, BF_BOOT_WDOG_MODE_DISABLED);
    set_second_reset_action("swap_emmc");
    enable_rst_n();
    return;
  }

  if (nowatchdog_swap) {
    // Disable second reset action (watchdog-swap behavior)
    set_second_reset_action("none");
  }

  if (watchdog_mode == BF_BOOT_WDOG_MODE_DISABLED)
    watchdog_interval = 0;

  set_watchdog(watchdog_interval, watchdog_mode);
}

// Return the (malloced) string describing the lifecycle state (w line return)
char *get_lifecycle_state(void)
{
  FILE *f = open_sysfs(LIFECYCLE_STATE_PATH, "r");
  char *buf = NULL;
  size_t len = 0;

  if (getline(&buf, &len, f) == -1)
    die("%s: failed to get lifecycle state", LIFECYCLE_STATE_PATH);
  fclose(f);

  return buf;
}

// Return number of remaining free secure boot fuse versions
int get_free_sbfuse_slots(void)
{
  int free_slot = 0;
  FILE *f = open_sysfs(SECURE_BOOT_FUSE_STATE_PATH, "r");
  char *buf = NULL, *p;
  size_t len = 0;

  while (getline(&buf, &len, f) != -1)
  {
    p = buf;
    while ((p = strstr(p, "Free")) != NULL) {
      free_slot++;
      p += strlen("Free");
    }
  }
  free(buf);
  fclose(f);

  return free_slot;
}

void show_status(void)
{
  // Display the default boot partition
  int part = get_boot_partition();
  printf("primary: %sboot%d\n", mmc_path, part);
  printf("backup: %sboot%d\n", mmc_path, part ^ 1);

  // Display the boot bus width setting
  int boot_bus_width = get_boot_bus_width();
  printf("boot-bus-width: x%d\n",
    (boot_bus_width & EXT_CSD_BOOT_BUS_WIDTH_MASK) ?
    (boot_bus_width & EXT_CSD_BOOT_BUS_WIDTH_MASK) * 4 : 1);
  printf("reset to x1 after reboot: %s\n",
    (boot_bus_width & EXT_CSD_BOOT_BUS_WIDTH_RESET_MASK)? "FALSE" : "TRUE");

  // Display the watchdog value
  bf_boot_watchdog_config_t watchdog_config = get_watchdog_config();
  const char *watchdog_mode_str = get_watchdog_mode_str(watchdog_config.mode);
  printf("boot watchdog mode: %s\n", watchdog_mode_str);
  printf("boot watchdog interval: %d\n", watchdog_config.interval);
  printf("watchdog-swap: ");
  if (watchdog_config.interval > 0 && watchdog_config.mode == BF_BOOT_WDOG_MODE_DISABLED) {
    // If the interval is non-zero and the watchdog is configured in disabled
    // mode then the watchdog is being used for boot swap protection.
    printf("%d\n", watchdog_config.interval);
  } else {
    printf("disabled\n");
  }

  // Display the secure boot fuse states
  char *lifecycle_state = get_lifecycle_state();
  printf("lifecycle state: %s", lifecycle_state);
  free(lifecycle_state);
  printf("secure boot key free slots: %d\n", get_free_sbfuse_slots());
}

// Get the version of hardware we're currently running on.
// If it can't be found, this will return -1.
int get_hw_version(void) {
  const char *dmi_file = DMI_TABLE_PATH;

  int dmi_fd = open(dmi_file, O_RDONLY);
  if (dmi_fd < 0) {
    fprintf(stderr, "warn: %s: %m, disable version filtering\n", dmi_file);
    return -1;
  }

  struct stat st;
  if (fstat(dmi_fd, &st) < 0)
    die("%s: stat: %m", dmi_file);

  void *dmi_buf = malloc(st.st_size);
  if (dmi_buf == NULL)
    die("out of memory");

  int n_bytes = 0;
  n_bytes = read_or_die(dmi_file, dmi_fd, dmi_buf, st.st_size);
  if (n_bytes != st.st_size)
    die("%s: could not read DMI table", dmi_file);

  if (close(dmi_fd) < 0)
    die("%s: close: %m", dmi_file);

  // Now, skip along table until we reach the processor info
  int bytes_left = st.st_size;
  void *idx = dmi_buf;
  uint8_t table_size = 0;

  while (bytes_left > 0) {
    uint8_t table_type = *((uint8_t*)idx);
    table_size = *((uint8_t*)idx + 1);

    // Break if we hit the processor table.
    if (table_type == DMI_PROCESSOR_INFO_TYPE)
      break;

    // Otherwise, skip to next table.
    idx += table_size;
    bytes_left -= table_size;

    bool first_str = true;

    // We might need to skip over strings, too.
    while (bytes_left > 0) {
      // If idx is over two 0 bytes, we're at the end, so
      // advance to the next table.
      if (*(uint16_t*)idx == 0) {
        idx += sizeof(uint16_t);
        bytes_left -= sizeof(uint16_t);
        break;
      } else {
        // Skip a string. Note the first iteration will
        // place the index at the beginning of the string,
        // whereas all further iterations place the index at
        // the NULL terminator.
        if (!first_str) {
          idx++;
          bytes_left--;
        }
        first_str = false;
        while (*(char*)idx && bytes_left > 0) {
          idx++;
          bytes_left--;
        }
      }
    }
  }

  if (bytes_left <= 0) {
    fprintf(stderr, "warning: could not find BlueField SoC revision\n");
    free(dmi_buf);
    return -1;
  }

  // Now idx is at processor version table - we can read the
  // string now. SMBIOS tables append all strings to the end
  // of the structure, and refer to them by their order.
  // First string is 1, second is 2, etc.
  uint8_t soc_ver_snum = *(uint8_t*)(idx + 0x10);
  if (soc_ver_snum == 0) {
    fprintf(stderr, "warning: found DMI processor ver field, but it's NULL\n");
    free(dmi_buf);
    return -1;
  }

  idx += table_size;
  bytes_left -= table_size;

  // Skip strings until we hit the desired one.
  char version_string[DMI_PROCESSOR_VERSION_STR_MAX_SIZE] = {0};
  bool first_str = true;
  while (soc_ver_snum-- > 1 && bytes_left > 0) {
    if (*(uint16_t*)idx == 0) {
      fprintf(stderr, "warning: DMI processor ver specifies string that does not exist.\n");
      free(dmi_buf);
      return -1;
    } else {
      if (!first_str) {
        idx++;
        bytes_left--;
      }
      first_str = false;
      // Skip a string.
      while (*(char*)idx && bytes_left > 0) {
        idx++;
        bytes_left--;
      }
    }
  }

  idx++;
  bytes_left--;

  // Finally, actually copy the string.
  strncpy(
    version_string,
    idx,
    bytes_left < DMI_PROCESSOR_VERSION_STR_MAX_SIZE - 1 ? bytes_left : DMI_PROCESSOR_VERSION_STR_MAX_SIZE - 1
  );

  // Extract version
  int version = -1;
  if (1 != sscanf(version_string, "Mellanox BlueField-%d", &version))
    fprintf(stderr, "warning: Unknown SoC revision");

  // BlueField 1 is v0, BF2 is v1, etc.
  version--;

  free(dmi_buf);
  return version;
}

/* Find substring within a buffer. */
static char* find_str(char *buf, int buf_len, const char *str)
{
  int len;
  char *p = buf, *find;

  while ((p - buf) < buf_len) {
    len = strlen(p);
    find = strstr(p, str);
    if (find)
      return find;
    p += len + 1;
  }

  return NULL;
}

/*
 * Handle the IROT signature DB and returns the offset of the record which
 * matches the key 'psc_cert_key', or return value < 0 in case of error case
 * or not found.
 */
static int handle_irot_sig_db(uint8_t *buf, uint32_t buf_len)
{
	uint32_t idx0, idx1, idx;
	uint64_t data64;

        /* Ignore sig db if not patch is needed. */
	if (!psc_cert_update)
		return -1;

	/* Sanity check. */
	if (buf_len % PSC_APP_IROT_SIG_REC_LEN)
		return -EINVAL;

	/* Binary search on the sorted records within range [idx0, idx1]. */
	idx0 = 0;
	idx1 = buf_len / PSC_APP_IROT_SIG_REC_LEN - 1;

        /* Binary search to find the record. */
	for(;;) {
		idx = (idx0 + idx1) / 2;
		data64 = *(uint64_t *)(buf + idx * PSC_APP_IROT_SIG_REC_LEN);
		if (data64 < psc_cert_key) {
			if (idx0 == idx)
				break;
			else
				idx0 = idx;
		} else if (data64 == psc_cert_key) {
			break;
		} else {
			if (idx1 == idx)
				break;
			else
				idx1 = idx;
		}
	}

	return (data64 == psc_cert_key) ? idx : -1;
}

/*
 * Filter the PSC APP image.
 * For the IROT signature DB, only up to one record is needed if matching the
 * lookup key.
 */
size_t filter_psc_app_image(void *image, size_t img_size)
{
  psc_app_img_hdr_t *hdr;
  size_t adjust_len = 0, len = 0;
  int idx;

  while (len < img_size) {
    hdr = (psc_app_img_hdr_t *)(image + len);

    if (len + sizeof(*hdr) + hdr->length > img_size)
      die("corrupted image\n");

    if (len != adjust_len) {
      memmove(image + adjust_len, image + len, sizeof(*hdr) + hdr->length);
      hdr = (psc_app_img_hdr_t *)(image + adjust_len);
    }

    len += sizeof(*hdr) + hdr->length;

    if (hdr->type != PSC_APP_TYPE_IROT_SIGNATURE_DB) {
      adjust_len += sizeof(*hdr) + hdr->length;
      continue;
    }

    idx = handle_irot_sig_db((uint8_t *)&hdr[1], hdr->length);
    if (idx >= 0) {
      memmove((uint8_t *)&hdr[1],
              (uint8_t *)&hdr[1] + idx * PSC_APP_IROT_SIG_REC_LEN,
              PSC_APP_IROT_SIG_REC_LEN);
      hdr->length = PSC_APP_IROT_SIG_REC_LEN;
      adjust_len += sizeof(*hdr) + PSC_APP_IROT_SIG_REC_LEN;
    }
  }

  return adjust_len;
}

static uint32_t crc32_update(uint32_t crc, uint8_t *p, unsigned int len)
{
  int i;

  while (len--) {
    crc ^= *p++;
    for (i = 0; i < 8; i++)
      crc = (crc >> 1) ^ ((crc & 1) ? 0xedb88320 : 0);
  }

  return crc;
}

#else

/* Dummy max partition size for OUTPUT_ONLY */
uint64_t get_boot_partition_size(void)
{
  return INPUT_BFB_MAX_SIZE;
}

#endif  // OUTPUT_ONLY

// Read as much as possible despite EINTR or partial reads, and die on error.
ssize_t read_or_die(const char* filename, int fd, void* buf, size_t count)
{
  ssize_t n = 0;
  while (count > 0)
  {
    ssize_t rc = read(fd, buf, count);
    if (rc < 0)
    {
      if (errno == EINTR)
        continue;
      die("%s: can't read: %m", filename);
    }
    if (rc == 0)
      break;

    n += rc;
    buf += rc;
    count -= rc;
  }
  return n;
}

// Write everything passed in despite EINTR or partial reads, and die on error.
ssize_t write_or_die(const char* filename,
                     int fd, const void* buf, size_t count)
{
  ssize_t n = count;
  while (count > 0)
  {
    ssize_t rc = write(fd, buf, count);
    if (rc < 0)
    {
      if (errno == EINTR)
        continue;
      die("%s: can't write: %m", filename);
    }
    if (rc == 0)
      die("%s: write returned zero", filename);

    buf += rc;
    count -= rc;
  }
  return n;
}

/*
 * Generate the boot stream segment header.
 *
 * is_end: 1 if this segment is the last segment of the boot code, else 0.
 * channel: The channel number to write to.
 * address: The register address to write to.
 * length: The length of this segment in bytes; max is MAX_SEG_LEN.
 *
 * We ignore endian issues here since if the tool is built natively,
 * this is likely correct anyway, and if built cross, we don't have a
 * way to know the endianness of the arm cores anyway.
 */
uint64_t gen_seg_header(int is_end, int channel, int address, size_t length)
{
  return (((is_end & 0x1UL) << 63) |
          ((channel & 0xfUL) << 45) |
          ((address & 0xfff8UL) << 29) |
          (((length + SEGMENT_HEADER_LEN) >> 3) & 0x1ffffUL));
}

uint64_t get_segment_length(uint64_t segheader)
{
  uint64_t length = (segheader & 0x1ffffUL) << 3;
  if (length == 0)
    length = MAX_SEG_LEN;
  else
    length -= SEGMENT_HEADER_LEN;
  return length;
}

void read_bootstream(const char *bootstream, const char *bootfile,
                     uint64_t part_size)
{
  // Copy the contents of the bootfile device to a bootstream
  printf("Copy bootstream from %s to %s\n", bootfile, bootstream);
  int ifd = open(bootfile, O_RDONLY);
  if (ifd < 0)
    die("%s: %m", bootfile);
  int ofd = open(bootstream, O_WRONLY | O_CREAT | O_TRUNC, 0666);
  if (ofd < 0)
    die("%s: %m", bootstream);

  char *buf = malloc(MAX_SEG_LEN);
  if (buf == NULL)
    die("out of memory");

  uint64_t header;
  // Read and discard the header word
  read_or_die(bootfile, ifd, &header, sizeof(header));
  uint64_t segheader = 0, total_size = 0;

  while (!(segheader & SEGMENT_IS_END))
  {
    read_or_die(bootfile, ifd, &segheader, sizeof(segheader));
    uint64_t seg_size = get_segment_length(segheader);
    read_or_die(bootfile, ifd, buf, seg_size);
    write_or_die(bootstream, ofd, buf, seg_size);
    total_size += seg_size;
    if (total_size > part_size)
      die("No valid bfb present");
  }

  if (close(ifd) < 0)
    die("%s: close: %m", bootstream);
  if (close(ofd) < 0)
    die("%s: close: %m", bootfile);
  free(buf);
}

// Read a header from a boot stream. This will consume the
// entire header, verify it, and return the header
// structure.
size_t read_bootstream_header(const char *bootstream, int ifd, boot_image_header_t *hdr) {
    int n;
    uint64_t data;

    // Read and verify the image header.
    n = read_or_die(bootstream, ifd, hdr, sizeof(*hdr));
    if (n != sizeof(*hdr))
      die("Unable to read next header, n=%d", n);
    n = hdr->hdr_len - sizeof(*hdr) / sizeof(uint64_t);
    if (n < 0)
      die("Invalid header length %d, n=%d", hdr->hdr_len, n);
    // Drain the rest of header.
    while (n--) {
      if (read_or_die(bootstream, ifd, &data, sizeof(data)) != sizeof(data))
        die("Not enough data for header");
    }

    return hdr->hdr_len * sizeof(uint64_t);
}

// Correct the following_images and next_img_ver fields in a BFB buffer.
// Note this assumes a well-formed BFB stream in the buffer.
void correct_bootstream_headers(void *buf, int num_images) {
  uint64_t fi_map = 0;
  void *idx = buf;
  boot_image_header_t *hdr;
  int img_size = 0;
  int pad_size = 0;

  if (num_images <= 0)
    die("%s: num_images must be at least 1", __FUNCTION__);

  // Markers to find images next time
  void **images = malloc(num_images * sizeof(void*));
  if (images == NULL)
    die("out of memory");

  // Iterate once to build fi_map and mark image starts
  for (int i = 0; i < num_images; i++) {
    hdr = (boot_image_header_t*) idx;

    // Build image maps
    fi_map |= 1UL << hdr->image_id;
    images[i] = idx;

    // Skip to next image
    img_size = hdr->image_len;
    pad_size = (img_size % 8) ? (8 - img_size % 8) : 0;
    idx += img_size + pad_size + hdr->hdr_len * sizeof(uint64_t);
  }

  boot_image_header_t *prev_hdr;

  // Now we can iterate over the images directly and fixup
  // the fields that need it
  for (int i = 0; i < num_images; i++) {
    hdr = (boot_image_header_t*) images[i];

    // Update image bitmap. Note that on the last
    // image of a series of versions, this will be wrong,
    // and will be corrected next iteration.
    hdr->following_images = fi_map;

    if (i > 0) {
      prev_hdr = (boot_image_header_t*) images[i-1];

      if (prev_hdr->image_id == hdr->image_id) {
        // Update next_img_ver as long as we haven't gone to
        // a new image ID.
        prev_hdr->next_img_ver = hdr->cur_img_ver;
      } else {
        // If next image ID is different, then update
        // fi_map, and correct the previous image.
        fi_map &= ~(1UL << prev_hdr->image_id);
        prev_hdr->following_images = fi_map;

        // We also set prev_hdr's next_img_ver to 0 since
        // there are no more images of that version.
        prev_hdr->next_img_ver = 0;
      }
    }
  }

  // Edge case: Handle last image.
  hdr = (boot_image_header_t*)images[num_images-1];
  hdr->following_images = 0;
  hdr->next_img_ver = 0;

  free(images);
}

// Tell whether we should write an image to the boot
// partition, based on a version number.
// We do this conservatively, so we don't accidentally
// filter out something common.
bool should_install_image(boot_image_header_t *hdr, int version, uint8_t *max_versions) {
  bool install = false;

  // version < 0 indicates filtering should be off. All
  // images will be installed in this case.
  if (version < 0)
    return true;

  // If version filtering is on, check max_versions is valid
  if (max_versions == NULL)
    die("internal: max_versions is NULL");

#ifndef OUTPUT_ONLY
  if ((hdr->image_id == PSC_APP_IMG_ID) && !psc_cert_update)
    return false;
#endif

  // The logic to decide what to install and what not to install
  // is based on BlueField 1, 2, 3, <X> chips each possibly having
  // their own version of a specific image while at the same time
  // also using "shared" images.  A shared image is defined as an
  // image with a version of 0 AND there are no other higher
  // versions in the bootstream (i.e. images with a version > 0).

  // If the version matches OR the image is a shared image, install it.
  if ((hdr->cur_img_ver == version) || (max_versions[hdr->image_id] == 0)) {
    install = true;
  }

  return install;
}

// Find the maximum version image for each image ID.
// max_versions must be a 256 element array. Each element of max_versions
// corresponds to a maximum version: max_versions[img_id] = max_img_ver
void find_max_versions(const char *bootstream, uint8_t *max_versions) {
  if (max_versions == NULL)
    die("%s: internal: max_versions is NULL", bootstream);

  int ifd = open(bootstream, O_RDONLY);
  if (ifd < 0)
    die("%s: %m", bootstream);

  struct stat st;
  if (fstat(ifd, &st) < 0)
    die("%s: stat: %m", bootstream);

  int bytes_left = st.st_size;

  // Initialize max_versions
  memset (max_versions, 0, MAX_VERSIONS_COUNT * sizeof(uint8_t));

  boot_image_header_t hdr;
  size_t hdr_size;
  size_t pad_size;
  size_t img_size;
  size_t n_bytes;

  while (bytes_left > 0) {
    hdr_size = read_bootstream_header(bootstream, ifd, &hdr);
    bytes_left -= hdr_size;
    img_size = hdr.image_len;

    pad_size = (img_size % 8) ? (8 - img_size % 8) : 0;

    // update max_versions if needed
    if (hdr.cur_img_ver > max_versions[hdr.image_id])
      max_versions[hdr.image_id] = hdr.cur_img_ver;

    // Skip image
    n_bytes = lseek(ifd, img_size + pad_size, SEEK_CUR);
    if (n_bytes < 0)
      die("%s: Could not skip filtered image: %m", bootstream);
    bytes_left -= img_size + pad_size;
  }

  close (ifd);
}

// Read a bootstream to an internal buffer, optionally filtering out images
// of a given version. Supply -1 to version to turn this behavior off.
size_t read_bootstream_to_buffer(const char *bootstream, void *buf, int buf_size, int version)
{
  int ifd = open(bootstream, O_RDONLY);
  if (ifd < 0)
    die("%s: %m", bootstream);

  struct stat st;
  if (fstat(ifd, &st) < 0)
    die("%s: stat: %m", bootstream);

  int bytes_left = st.st_size;

  // If no version filtering requested and file is too large for the
  // buffer, error out.
  if (version < 0 && bytes_left > buf_size) {
    die("%s: Boot stream file is too large", bootstream);
  }

  uint8_t max_versions[MAX_VERSIONS_COUNT];
  find_max_versions(bootstream, max_versions);

  void *idx = buf; // Index along buffer
  size_t n_bytes = 0, n_bytes_adjust = 0;
  boot_image_header_t hdr;
  size_t hdr_size; // Size of the image header, including reserved words
  size_t pad_size;
  size_t img_size;

  int num_images = 0;

  // Otherwise, we'll need to read the whole file and filter it to tell
  // whether stream will fit.
  while (bytes_left > 0) {
    hdr_size = read_bootstream_header(bootstream, ifd, &hdr);
    bytes_left -= hdr_size;
    img_size = hdr.image_len;

    pad_size = (img_size % 8) ? (8 - img_size % 8) : 0;

    // Check whether the version should be included.
    if (should_install_image(&hdr, version, max_versions)) {
      // If so, copy the header and img into the buffer.
      if ((hdr_size + idx) - buf > buf_size)
        die("Boot stream file is too large");

      memcpy(idx, &hdr, sizeof(hdr));
      idx += hdr_size;

      // Copy image into buffer.
      if ((hdr.image_len + idx) - buf > buf_size)
        die("Boot stream file is too large");

      n_bytes = read_or_die(bootstream, ifd, idx, img_size + pad_size);
      if (n_bytes != img_size + pad_size)
        die("Unable to read next image, n=%d", n_bytes);

      n_bytes_adjust = n_bytes;
#ifndef OUTPUT_ONLY
      if (hdr.image_id == PSC_APP_IMG_ID) {
        // Update image len and CRC for trimmed image.
        n_bytes_adjust = filter_psc_app_image(idx, img_size);
        boot_image_header_t *cur_hdr =
            (boot_image_header_t *)(idx - sizeof(hdr));
        cur_hdr->image_len = n_bytes_adjust;
        cur_hdr->image_crc = ~crc32_update(~0, idx, n_bytes_adjust);
      }
#endif

      idx += n_bytes_adjust;
      bytes_left -= n_bytes;

      // Finally, count the image. We reuse this later for
      // fixing up the headers.
      num_images++;
    } else {
      // Otherwise, skip over the entire image.
      n_bytes = lseek(ifd, img_size + pad_size, SEEK_CUR);
      if (n_bytes < 0)
        die("%s: Could not skip filtered image: %m", bootstream);
      bytes_left -= img_size + pad_size;
    }
  }

  if (close(ifd) < 0)
    die("%s: close: %m", bootstream);

  correct_bootstream_headers(buf, num_images);

  return idx - buf;
}

/*
 * Get the filtering information from ACPI, which has the
 * irot-cert-update: L4 IROT cert needs to be updated.
 * irot-cert-key: L4 IROT cert signature lookup key.
 */
static void get_psc_app_filter_info(void)
{
#ifndef OUTPUT_ONLY
  char path[] = "/sys/firmware/acpi/tables/SSDT1";
  struct stat st;
  int i, fd;
  char *buf = NULL, *p;

  /* Support SSDT[1~9] */
  for (i = 1; i <= 9; i++) {
    path[strlen(path) - 1] = '0' + i;
    fd = open(path, O_RDONLY);
    if (fd == -1)
      break;

    // Get size
    if (fstat(fd, &st) < 0)
      die("%s: stat: %m", path);

    // Allocate memory and read the file.
    if (buf)
      free(buf);
    buf = malloc(st.st_size + 1);
    if (!buf)
      die("fail to alloc buf: %m");
    buf[st.st_size] = 0;
    if (read_or_die(path, fd, buf, st.st_size) != st.st_size)
      die("%s: unexpected read", path);

    close(fd);

    p = find_str(buf, st.st_size, "irot-cert-key");
    if (!p)
      continue;
    memcpy(&psc_cert_key, p + strlen("irot-cert-key") + 2, sizeof(uint64_t));

    p = find_str(buf, st.st_size, "irot-cert-update");
    if (!p)
      break;
    memcpy(&psc_cert_update, p + strlen("irot-cert-update") + 2, sizeof(uint8_t));

    break;
  }

  if (buf)
    free(buf);
#endif
}

void write_bootstream(const char *bootstream, const char *bootfile, int flags, int version)
{
  int sysfd = -1;
  char *sysname;
  size_t ibuf_maxsize = INPUT_BFB_MAX_SIZE;

  get_psc_app_filter_info();

  // Reset the force_ro setting if need be
  if (strncmp(bootfile, "/dev/", 5) == 0)
  {
    if (asprintf(&sysname, "/sys/block/%s/force_ro", &bootfile[5]) <= 0)
      die("unexpected failure in asprintf (%s/%d)", __FILE__, __LINE__);

    sysfd = open(sysname, O_RDWR | O_SYNC);
    if (sysfd >= 0)
    {
      char status;
      if (read_or_die(sysname, sysfd, &status, 1) != 1)
        die("%s: unexpected EOF on read", sysname);
      if (status == '1')
      {
        char disabled = '0';
        if (lseek(sysfd, 0, SEEK_SET) != 0)
          die("%s: can't seek back to start: %m", sysname);
        write_or_die(sysname, sysfd, &disabled, 1);
      }
      else
      {
        close(sysfd);
        sysfd = -1;
        free(sysname);
        sysname = NULL;
      }
    }
    else
    {
      if (errno != ENOENT)
        die("%s: open: %m", sysname);
      printf("WARNING: No matching %s for %s\n", sysname, bootfile);
      free(sysname);
      sysname = NULL;
    }

    // If writing to /dev/... assume it's an MMC partition
    ibuf_maxsize = get_boot_partition_size();
  }

  // Copy the bootstream to the bootfile device
  void *ibuf = malloc(ibuf_maxsize);
  void *inidx = ibuf;
  uint8_t padbuf[8] = {0}; // There are at most 8 pad bytes.
  if (ibuf == NULL)
    die("out of memory");
  size_t bytes_left = read_bootstream_to_buffer(bootstream, ibuf, ibuf_maxsize, version);
  int ofd = open(bootfile, O_WRONLY | flags, 0666);
  if (ofd < 0)
    die("%s: %m", bootfile);

  // Write the bootstream header word first.  This has the byte to
  // be displayed in the rev_id register as the low 8 bits (zero for now).
  uint64_t header = 0;
  write_or_die(bootfile, ofd, &header, sizeof(header));

  while (bytes_left > 0)
  {
    size_t seg_size = (bytes_left <= MAX_SEG_LEN) ? bytes_left : MAX_SEG_LEN;
    bytes_left -= seg_size;

    // Generate the segment header.
    size_t pad_size = seg_size % 8 ? (8 - seg_size % 8) : 0;
    uint64_t segheader = gen_seg_header(bytes_left == 0, 1, BOOT_FIFO_ADDR,
                                        seg_size + pad_size);
    write_or_die(bootfile, ofd, &segheader, sizeof(segheader));

    // Copy the segment plus any padding.
    write_or_die(bootfile, ofd, inidx, seg_size);
    inidx += seg_size;
    write_or_die(bootfile, ofd, padbuf, pad_size);
  }

  if (close(ofd) < 0)
    die("%s: close: %m", bootfile);
  free(ibuf);

  // Put back the force_ro setting if need be
  if (sysfd >= 0)
  {
    if (lseek(sysfd, 0, SEEK_SET) != 0)
      die("%s: can't seek back to start: %m", sysname);
    char enabled = '1';
    write_or_die(sysname, sysfd, &enabled, 1);
    close(sysfd);
    free(sysname);
  }
}

#ifdef OUTPUT_ONLY

int main(int argc, char **argv)
{
  static struct option long_options[] = {
    { "bootstream", required_argument, NULL, 'b' },
    { "output", required_argument, NULL, 'o' },
    { "help", no_argument, NULL, 'h' },
    { NULL, 0, NULL, 0 }
  };
  static const char short_options[] = "b:o:h";
  static const char help_text[] =
   "syntax: mlx-bootctl [--help|-h] --bootstream|-b BFBFILE --output|-o OUTPUT";

  const char *bootstream = NULL;
  const char *output_file = NULL;
  int opt;

  while ((opt = getopt_long(argc, argv, short_options, long_options, NULL))
         != -1)
  {
    switch (opt)
    {
    case 'b':
      bootstream = optarg;
      break;

    case 'o':
      output_file = optarg;
      break;

    case 'h':
    default:
      die(help_text);
      break;
    }
  }

  if (bootstream == NULL || output_file == NULL)
    die("mlx-bootctl: Must specify --output and --bootstream");

  write_bootstream(bootstream, output_file, O_CREAT | O_TRUNC, -1);
  return 0;
}

#else

static void verify_bootstream(const char *bootfile)
{
  int ifd, bytes_left, img_size;
  boot_image_header_t hdr;
  struct stat st;
  uint64_t data;
  uint32_t crc;

  ifd = open(bootfile, O_RDONLY);
  if (ifd < 0)
    die("%s: %m", bootfile);

  if (fstat(ifd, &st) < 0)
    die("%s: stat: %m", bootfile);
  bytes_left = st.st_size;
  if (bytes_left % sizeof(uint64_t))
    die("Invalid size %d", bytes_left);
  if (bytes_left > INPUT_BFB_MAX_SIZE)
    die("Input boot file size too big");

  while (bytes_left > 0) {
    bytes_left -= read_bootstream_header(bootfile, ifd, &hdr);
    if (bytes_left < 0)
      die("Invalid header length");

    // Verify the image crc.
    crc = ~0;
    img_size = hdr.image_len;
    while (img_size > 0) {
      if (read_or_die(bootfile, ifd, &data, sizeof(data)) != sizeof(data))
        die("Not enough data for image id %d, missing size %d", hdr.image_id, img_size);
      crc = crc32_update(crc, (uint8_t *)&data, sizeof(data));
      img_size -= sizeof(data);
      bytes_left -= sizeof(data);
    }
    if (hdr.image_crc != ~crc)
      die("Invalid CRC for image_id %d", hdr.image_id);
  }

  close(ifd);
}

int main(int argc, char **argv)
{
  static struct option long_options[] = {
    { "swap", no_argument, NULL, 's' },
    { "watchdog-swap", required_argument, NULL, 'w' },
    { "nowatchdog-swap", no_argument, NULL, 'n' },
    { "watchdog-boot-mode", required_argument, NULL, 'm' },
    { "watchdog-boot-interval", required_argument, NULL, 'i' },
    { "bootstream", required_argument, NULL, 'b' },
    { "overwrite-current", no_argument, NULL, 'c' },
    { "device", required_argument, NULL, 'd' },
    { "output", required_argument, NULL, 'o' },
    { "read", required_argument, NULL, 'r' },
    { "version", required_argument, NULL, 'v' },
    { "help", no_argument, NULL, 'h' },
    { NULL, 0, NULL, 0 }
  };
  static const char short_options[] = "sb:d:o:r:hev:";
  static const char help_text[] =
    "syntax: mlxbf-bootctl [--help|-h] [--swap|-s] [--device|-d MMCFILE]\n"
    "                      [--output|-o OUTPUT] [--read|-r INPUT]\n"
    "                      [--bootstream|-b BFBFILE] [--overwrite-current]\n"
    "                      [--watchdog-swap interval | --nowatchdog-swap]\n"
    "                      [--watchdog-boot-mode disabled|standard|time_limit]\n"
    "                      [--watchdog-boot-interval %d-%d]\n"
    "                      [--version|-v VERSION]";

  const char *watchdog_mode_str = NULL;
  const char *watchdog_interval_str = NULL;
  const char *bootstream = NULL;
  const char *output_file = NULL;
  const char *input_file = NULL;
  bool watchdog_swap = false;
  bool nowatchdog_swap = false;
  bool watchdog_interval_param = false;
  bool swap = false;
  bool auto_version = true;
  int version_arg = -1;
  int which_boot = 1;   // alternate boot partition by default
  int opt;

  while ((opt = getopt_long(argc, argv, short_options, long_options, NULL))
         != -1)
  {
    switch (opt)
    {
    case 's':
      swap = true;
      break;

    case 'w':
      watchdog_interval_str = optarg;
      watchdog_swap = true;
      break;

    case 'n':
      nowatchdog_swap = true;
      break;

    case 'm':
      watchdog_mode_str = optarg;
      nowatchdog_swap = true;
      if (get_hw_version() != BF3_VERSION)
        die("'--watchdog-boot-mode' is only supported on Bluefield 3");
      break;

    case 'i':
      watchdog_interval_str = optarg;
      watchdog_interval_param = true;
      if (get_hw_version() != BF3_VERSION)
        die("'--watchdog-boot-interval' is only supported on Bluefield 3");
      break;

    case 'b':
      bootstream = optarg;
      break;

    case 'c':
      which_boot = 0;    // overwrite current boot partition (dangerous)
      break;

    case 'd':
      mmc_path = optarg;
      break;

    case 'o':
      output_file = optarg;
      break;

    case 'r':
      input_file = optarg;
      break;

    case 'e':
      enable_rst_n();
      break;

    case 'v':
      auto_version = false;
      char *end;
      version_arg = strtol(optarg, &end, 0);
      if (end == optarg || *end != '\0')
        die("version argument ('%s') must be an integer", optarg);
      break;

    case 'h':
    default:
      die(help_text, MIN_WDOG_INTERVAL_SEC, MAX_WDOG_INTERVAL_SEC);
      break;
    }
  }

  if (argc <= 1)
  {
    show_status();
    return 0;
  }

  if (watchdog_swap && watchdog_mode_str)
    die("--watchdog-swap cannot be used with --watchdog-boot-mode.");

  if (watchdog_swap && nowatchdog_swap)
    die("--watchdog-swap cannot be used with --nowatchdog-swap.");

  if (watchdog_interval_param && !watchdog_mode_str)
    die("--watchdog-boot-interval cannot be used without --watchdog-boot-mode.");

  if (bootstream)
  {
    if (input_file)
    {
      uint64_t boot_part_size = get_boot_partition_size();
      read_bootstream(bootstream, input_file, boot_part_size);
    }
    else if (output_file)
    {
      // Write the bootstream to the given file, creating it if needed
      // Don't filter anything here, since we're not writing to EMMC.
      write_bootstream(bootstream, output_file, O_CREAT | O_TRUNC, -1);
    }
    else
    {
      verify_bootstream(bootstream);

      struct stat st;
      int error;
      error = stat(bootstream, &st);

      if (error < 0)
        die("%s: %m", bootstream);

      // Get the active partition and write to the appropriate *bootN file
      // Must save/restore boot partition, which I/O otherwise resets to zero.
      int boot_part = get_boot_partition();
      char *bootfile;
      if (asprintf(&bootfile, "%sboot%d", mmc_path, boot_part ^ which_boot) <= 0)
        die("unexpected failure in asprintf (%s/%d)", __FILE__, __LINE__);

      int version;
      if (auto_version) {
        version = get_hw_version();
      } else {
        version = version_arg;
      }

      // Note: There is no check to see if the image will fit in the
      // boot partition in main(). This is because write_bootstream()
      // may filter the contents of the stream, meaning the size check
      // must be done in write_bootstream().
      write_bootstream(bootstream, bootfile, O_SYNC, version);

      // The eMMC driver works in an asynchronous way, thus any commands sent
      // should occur after the write to the bootstream has retired. Otherwise
      // a blk_update_request I/O error would be raised and the success of the
      // command cannot be guaranteed. The check for the status of the eMMC
      // at user level is non-trival, thus a delay is added here instead.
      // @TODO: Update the delay to check for the eMMC status.
      sleep(1);
      set_boot_partition(boot_part);
      set_boot_bus_width();
      enable_rst_n();
    }
  }

  if (swap)
    set_boot_partition(get_boot_partition() ^ 1);

  configure_watchdog(watchdog_swap, nowatchdog_swap,
                     watchdog_mode_str, watchdog_interval_str);
  return 0;
}

#endif // OUTPUT_ONLY