/* ----------------------------------------------------------------------------

Copyright (c) 2018-2025, Microsoft Research, Daan Leijen

This is free software; you can redistribute it and/or modify it under the

terms of the MIT license. A copy of the license can be found in the file

"LICENSE" at the root of this distribution.

-----------------------------------------------------------------------------*/

// This file is included in `src/prim/prim.c`

#ifndef _DEFAULT_SOURCE

#define _DEFAULT_SOURCE   // ensure mmap flags and syscall are defined

#endif

#if defined(__sun)

// illumos provides new mman.h api when any of these are defined

// otherwise the old api based on caddr_t which predates the void pointers one.

// stock solaris provides only the former, chose to atomically to discard those

// flags only here rather than project wide tough.

#undef _XOPEN_SOURCE

#undef _POSIX_C_SOURCE

#endif

#include "mimalloc.h"

#include "mimalloc/internal.h"

#include "mimalloc/prim.h"

#include <sys/mman.h>  // mmap

#include <unistd.h>    // sysconf

#include <fcntl.h>     // open, close, read, access

#include <stdlib.h>    // getenv, arc4random_buf

#if defined(__linux__)

  #include <features.h>

  #include <sys/prctl.h>    // THP disable, PR_SET_VMA

  #include <sys/sysinfo.h>  // sysinfo

  #if defined(__GLIBC__) && !defined(PR_SET_VMA)

  #include <linux/prctl.h>

  #endif

  #if defined(__GLIBC__)

  #include <linux/mman.h>   // linux mmap flags

  #else

  #include <sys/mman.h>

  #endif

#elif defined(__APPLE__)

  #include <AvailabilityMacros.h>

  #include <TargetConditionals.h>

  #if !defined(TARGET_OS_OSX) || TARGET_OS_OSX   // see issue #879, used to be (!TARGET_IOS_IPHONE && !TARGET_IOS_SIMULATOR)

  #include <mach/vm_statistics.h>    // VM_MAKE_TAG, VM_FLAGS_SUPERPAGE_SIZE_2MB, etc.

  #endif

  #if !defined(MAC_OS_X_VERSION_10_7)

  #define MAC_OS_X_VERSION_10_7   1070

  #endif

  #include <sys/sysctl.h>

#elif defined(__FreeBSD__) || defined(__DragonFly__)

  #include <sys/param.h>

  #if __FreeBSD_version >= 1200000

  #include <sys/cpuset.h>

  #include <sys/domainset.h>

  #endif

  #include <sys/sysctl.h>

#endif

#if (defined(__linux__) && !defined(__ANDROID__)) || defined(__FreeBSD__)

  #define MI_HAS_SYSCALL_H

  #include <sys/syscall.h>

#endif

#if !defined(MADV_DONTNEED) && defined(POSIX_MADV_DONTNEED)  // QNX

#define MADV_DONTNEED  POSIX_MADV_DONTNEED

#endif

#if !defined(MADV_FREE) && defined(POSIX_MADV_FREE)  // QNX

#define MADV_FREE  POSIX_MADV_FREE

#endif

#define MI_UNIX_LARGE_PAGE_SIZE (2*MI_MiB) // TODO: can we query the OS for this?

//------------------------------------------------------------------------------------

// Use syscalls for some primitives to allow for libraries that override open/read/close etc.

// and do allocation themselves; using syscalls prevents recursion when mimalloc is

// still initializing (issue #713)

// Declare inline to avoid unused function warnings.

//------------------------------------------------------------------------------------

#if defined(MI_HAS_SYSCALL_H) && defined(SYS_open) && defined(SYS_close) && defined(SYS_read) && defined(SYS_access)

static inline int mi_prim_open(const char* fpath, int open_flags) {

  return syscall(SYS_open,fpath,open_flags,0);

}

static inline ssize_t mi_prim_read(int fd, void* buf, size_t bufsize) {

  return syscall(SYS_read,fd,buf,bufsize);

}

static inline int mi_prim_close(int fd) {

  return syscall(SYS_close,fd);

}

static inline int mi_prim_access(const char *fpath, int mode) {

  return syscall(SYS_access,fpath,mode);

}

#else

static inline int mi_prim_open(const char* fpath, int open_flags) {

  return open(fpath,open_flags);

}

static inline ssize_t mi_prim_read(int fd, void* buf, size_t bufsize) {

  return read(fd,buf,bufsize);

}

static inline int mi_prim_close(int fd) {

  return close(fd);

}

static inline int mi_prim_access(const char *fpath, int mode) {

  return access(fpath,mode);

}

#endif

//---------------------------------------------

// init

//---------------------------------------------

static bool unix_detect_overcommit(void) {

  bool os_overcommit = true;

  #if defined(__linux__)

    int fd = mi_prim_open("/proc/sys/vm/overcommit_memory", O_RDONLY);

    if (fd >= 0) {

      char buf[32] = {0};  // CPython: zero-initialize to avoid an MSAN report

      ssize_t nread = mi_prim_read(fd, &buf, sizeof(buf));

      mi_prim_close(fd);

      // <https://www.kernel.org/doc/Documentation/vm/overcommit-accounting>

      // 0: heuristic overcommit, 1: always overcommit, 2: never overcommit (ignore NORESERVE)

      if (nread >= 1) {

        os_overcommit = (buf[0] == '0' || buf[0] == '1');

      }

    }

  #elif defined(__FreeBSD__)

    int val = 0;

    size_t olen = sizeof(val);

    if (sysctlbyname("vm.overcommit", &val, &olen, NULL, 0) == 0) {

      os_overcommit = (val != 0);

    }

  #else

    // default: overcommit is true

  #endif

  return os_overcommit;

}

// try to detect the physical memory dynamically (if possible)

static void unix_detect_physical_memory( size_t page_size, size_t* physical_memory_in_kib ) {

  #if defined(CTL_HW) && (defined(HW_PHYSMEM64) || defined(HW_MEMSIZE))  // freeBSD, macOS

    MI_UNUSED(page_size);

    int64_t physical_memory = 0;

    size_t length = sizeof(int64_t);

    #if defined(HW_PHYSMEM64)

    int mib[2] = { CTL_HW, HW_PHYSMEM64 };

    #else

    int mib[2] = { CTL_HW, HW_MEMSIZE };

    #endif

    const int err = sysctl(mib, 2, &physical_memory, &length, NULL, 0);

    if (err==0 && physical_memory > 0) {

      const int64_t phys_in_kib = physical_memory / MI_KiB;

      if (phys_in_kib > 0 && (uint64_t)phys_in_kib <= SIZE_MAX) {

        *physical_memory_in_kib = (size_t)phys_in_kib;

      }

    }

  #elif defined(__linux__)

    MI_UNUSED(page_size);

    struct sysinfo info; _mi_memzero_var(info);

    const int err = sysinfo(&info);

    if (err==0 && info.mem_unit > 0 && info.totalram <= SIZE_MAX) {

      if (info.mem_unit==MI_KiB) {

        *physical_memory_in_kib = (size_t)info.totalram;

      }

      else {

        size_t total = 0;

        if (!mi_mul_overflow((size_t)info.totalram, (size_t)info.mem_unit, &total)) {

          *physical_memory_in_kib = (total / MI_KiB);

        }

      }

    }

  #elif defined(_SC_PHYS_PAGES)  // do not use by default as it might cause allocation (by using `fopen` to parse /proc/meminfo) (issue #1100)

    const long pphys = sysconf(_SC_PHYS_PAGES);

    const size_t psize_in_kib = page_size / MI_KiB;

    if (psize_in_kib > 0 && pphys > 0 && (unsigned long)pphys <= SIZE_MAX && (size_t)pphys <= (SIZE_MAX/psize_in_kib)) {

      *physical_memory_in_kib = (size_t)pphys * psize_in_kib;

    }

  #endif

}

void _mi_prim_mem_init( mi_os_mem_config_t* config )

{

  long psize = sysconf(_SC_PAGESIZE);

  if (psize > 0 && (unsigned long)psize < SIZE_MAX) {

    config->page_size = (size_t)psize;

    config->alloc_granularity = (size_t)psize;

    unix_detect_physical_memory(config->page_size, &config->physical_memory_in_kib);

  }

  config->large_page_size = MI_UNIX_LARGE_PAGE_SIZE;

  config->has_overcommit = unix_detect_overcommit();

  config->has_partial_free = true;    // mmap can free in parts

  config->has_virtual_reserve = true; // todo: check if this true for NetBSD?  (for anonymous mmap with PROT_NONE)

  // disable transparent huge pages for this process?

  #if (defined(__linux__) || defined(__ANDROID__)) && defined(PR_GET_THP_DISABLE)

  if (!mi_option_is_enabled(mi_option_allow_thp)) // disable THP if requested through an option

  {

    if (prctl(PR_GET_THP_DISABLE, 0, 0, 0, 0) == 0) {   // -1 on error, 1 if already disabled

      // Most likely since distros often come with always/madvise settings.

      // Disabling only for mimalloc process rather than touching system wide settings

      (void)prctl(PR_SET_THP_DISABLE, 1, 0, 0, 0);

    }

  }

  #endif

}

//---------------------------------------------

// free

//---------------------------------------------

int _mi_prim_free(void* addr, size_t size ) {

  if (size==0) return 0;

  bool err = (munmap(addr, size) == -1);

  return (err ? errno : 0);

}

//---------------------------------------------

// mmap

//---------------------------------------------

// return errno on failure

static int unix_madvise(void* addr, size_t size, int advice) {

  #if defined(__sun) || defined(_AIX)

  const int res = madvise((caddr_t)addr, size, advice);  // Solaris needs cast (issue #520)

  return (res==0 ? 0 : errno);

  #elif defined(__QNX__)

  return posix_madvise(addr, size, advice);              // posix returns errno

  #else

  const int res = madvise(addr, size, advice);           // linux returns -1 on failure and sets errno

  return (res==0 ? 0 : errno);

  #endif

}

static void* unix_mmap_prim(void* addr, size_t size, int protect_flags, int flags, int fd) {

  void* p = mmap(addr, size, protect_flags, flags, fd, 0 /* offset */);

  #if defined(__linux__) && defined(PR_SET_VMA)

  if (p!=MAP_FAILED && p!=NULL) {

    prctl(PR_SET_VMA, PR_SET_VMA_ANON_NAME, p, size, "mimalloc");

  }

  #endif

  return p;

}

static void* unix_mmap_prim_aligned(void* addr, size_t size, size_t try_alignment, int protect_flags, int flags, int fd) {

  MI_UNUSED(try_alignment);

  void* p = NULL;

  #if defined(MAP_ALIGNED)  // BSD

  if (addr == NULL && try_alignment > 1 && (try_alignment % _mi_os_page_size()) == 0) {

    size_t n = mi_bsr(try_alignment);

    if (((size_t)1 << n) == try_alignment && n >= 12 && n <= 30) {  // alignment is a power of 2 and 4096 <= alignment <= 1GiB

      p = unix_mmap_prim(addr, size, protect_flags, flags | MAP_ALIGNED(n), fd);

      if (p==MAP_FAILED || !_mi_is_aligned(p,try_alignment)) {

        int err = errno;

        _mi_trace_message("unable to directly request aligned OS memory (error: %d (0x%x), size: 0x%zx bytes, alignment: 0x%zx, hint address: %p)\n", err, err, size, try_alignment, addr);

      }

      if (p!=MAP_FAILED) return p;

      // fall back to regular mmap

    }

  }

  #elif defined(MAP_ALIGN)  // Solaris

  if (addr == NULL && try_alignment > 1 && (try_alignment % _mi_os_page_size()) == 0) {

    p = unix_mmap_prim((void*)try_alignment, size, protect_flags, flags | MAP_ALIGN, fd);  // addr parameter is the required alignment

    if (p!=MAP_FAILED) return p;

    // fall back to regular mmap

  }

  #endif

  #if (MI_INTPTR_SIZE >= 8) && !defined(MAP_ALIGNED)

  // on 64-bit systems, use the virtual address area after 2TiB for 4MiB aligned allocations

  if (addr == NULL) {

    void* hint = _mi_os_get_aligned_hint(try_alignment, size);

    if (hint != NULL) {

      p = unix_mmap_prim(hint, size, protect_flags, flags, fd);

      if (p==MAP_FAILED || !_mi_is_aligned(p,try_alignment)) {

        #if MI_TRACK_ENABLED  // asan sometimes does not instrument errno correctly?

        int err = 0;

        #else

        int err = errno;

        #endif

        _mi_trace_message("unable to directly request hinted aligned OS memory (error: %d (0x%x), size: 0x%zx bytes, alignment: 0x%zx, hint address: %p)\n", err, err, size, try_alignment, hint);

      }

      if (p!=MAP_FAILED) return p;

      // fall back to regular mmap

    }

  }

  #endif

  // regular mmap

  p = unix_mmap_prim(addr, size, protect_flags, flags, fd);

  if (p!=MAP_FAILED) return p;

  // failed to allocate

  return NULL;

}

static int unix_mmap_fd(void) {

  #if defined(VM_MAKE_TAG)

  // macOS: tracking anonymous page with a specific ID. (All up to 98 are taken officially but LLVM sanitizers had taken 99)

  int os_tag = (int)mi_option_get(mi_option_os_tag);

  if (os_tag < 100 || os_tag > 255) { os_tag = 254; }

  return VM_MAKE_TAG(os_tag);

  #else

  return -1;

  #endif

}

#if defined(MAP_ALIGNED_SUPER) || defined(MAP_HUGETLB) || defined(MAP_HUGE_1GB) || defined(MAP_HUGE_2MB) || defined(VM_FLAGS_SUPERPAGE_SIZE_2MB)

#define MI_OS_HAS_HUGE_PAGES  1

#endif

static void* unix_mmap(void* addr, size_t size, size_t try_alignment, int protect_flags, bool large_only, bool allow_large, bool* is_large) {

  #if !defined(MAP_ANONYMOUS)

  #define MAP_ANONYMOUS  MAP_ANON

  #endif

  #if !defined(MAP_NORESERVE)

  #define MAP_NORESERVE  0

  #endif

  void* p = NULL;

  const int fd = unix_mmap_fd();

  int flags = MAP_PRIVATE | MAP_ANONYMOUS;

  if (_mi_os_has_overcommit()) {

    flags |= MAP_NORESERVE;

  }

  #if defined(PROT_MAX)

  protect_flags |= PROT_MAX(PROT_READ | PROT_WRITE); // BSD

  #endif

  // huge page allocation

  #if MI_OS_HAS_HUGE_PAGES

  if (allow_large && (large_only || (_mi_os_canuse_large_page(size, try_alignment) && mi_option_is_enabled(mi_option_allow_large_os_pages)))) {

    static _Atomic(size_t) large_page_try_ok; // = 0;

    size_t try_ok = mi_atomic_load_acquire(&large_page_try_ok);

    if (!large_only && try_ok > 0) {

      // If the OS is not configured for large OS pages, or the user does not have

      // enough permission, the `mmap` will always fail (but it might also fail for other reasons).

      // Therefore, once a large page allocation failed, we don't try again for `large_page_try_ok` times

      // to avoid too many failing calls to mmap.

      mi_atomic_cas_strong_acq_rel(&large_page_try_ok, &try_ok, try_ok - 1);

    }

    else {

      int lflags = flags & ~MAP_NORESERVE;  // using NORESERVE on huge pages seems to fail on Linux

      int lfd = fd;

      #ifdef MAP_ALIGNED_SUPER

      lflags |= MAP_ALIGNED_SUPER;

      #endif

      #ifdef MAP_HUGETLB

      lflags |= MAP_HUGETLB;

      #endif

      #ifdef MAP_HUGE_1GB

      static _Atomic(size_t) mi_huge_1gib_pages_unavailable;

      if (large_only && (size % MI_GiB) == 0 && (mi_atomic_load_relaxed(&mi_huge_1gib_pages_unavailable)==0)) {

        lflags |= MAP_HUGE_1GB;

      }

      else

      #endif

      {

        #ifdef MAP_HUGE_2MB

        lflags |= MAP_HUGE_2MB;

        #endif

      }

      #ifdef VM_FLAGS_SUPERPAGE_SIZE_2MB

      lfd |= VM_FLAGS_SUPERPAGE_SIZE_2MB;

      #endif

      if (large_only || lflags != flags) {

        // try large OS page allocation

        *is_large = true;

        p = unix_mmap_prim_aligned(addr, size, try_alignment, protect_flags, lflags, lfd);

        #ifdef MAP_HUGE_1GB

        if (p == NULL && (lflags & MAP_HUGE_1GB) == MAP_HUGE_1GB) {

          mi_atomic_store_relaxed(&mi_huge_1gib_pages_unavailable,1); // don't try huge 1GiB pages again

          if (large_only) {

            _mi_warning_message("unable to allocate huge (1GiB) page, trying large (2MiB) pages instead (errno: %i)\n", errno);

          }

          lflags = ((lflags & ~MAP_HUGE_1GB) | MAP_HUGE_2MB);

          p = unix_mmap_prim_aligned(addr, size, try_alignment, protect_flags, lflags, lfd);

        }

        #endif

        if (large_only) return p;

        if (p == NULL) {

          mi_atomic_store_release(&large_page_try_ok, (size_t)8);  // on error, don't try again for the next N allocations

        }

      }

    }

  } // huge pages

  #endif

  // regular allocation

  if (p == NULL) {

    *is_large = false;

    p = unix_mmap_prim_aligned(addr, size, try_alignment, protect_flags, flags, fd);

    #if !defined(MI_NO_THP)

    if (p != NULL && mi_option_is_enabled(mi_option_allow_thp) && _mi_os_canuse_large_page(size, try_alignment)) {

      #if defined(MADV_HUGEPAGE)

      // Many Linux systems don't allow MAP_HUGETLB but they support instead

      // transparent huge pages (THP). Generally, it is not required to call `madvise` with MADV_HUGE

      // though since properly aligned allocations will already use large pages if available

      // in that case -- in particular for our large regions (in `memory.c`).

      // However, some systems only allow THP if called with explicit `madvise`, so

      // when large OS pages are enabled for mimalloc, we call `madvise` anyways.

      if (unix_madvise(p, size, MADV_HUGEPAGE) == 0) {

        // *is_large = true; // possibly

      };

      #elif defined(__sun)

      struct memcntl_mha cmd = {0};

      cmd.mha_pagesize = _mi_os_large_page_size();

      cmd.mha_cmd = MHA_MAPSIZE_VA;

      if (memcntl((caddr_t)p, size, MC_HAT_ADVISE, (caddr_t)&cmd, 0, 0) == 0) {

        // *is_large = true; // possibly

      }

      #endif

    }

    #endif

  }

  return p;

}

// Note: the `try_alignment` is just a hint and the returned pointer is not guaranteed to be aligned.

int _mi_prim_alloc(void* hint_addr, size_t size, size_t try_alignment, bool commit, bool allow_large, bool* is_large, bool* is_zero, void** addr) {

  mi_assert_internal(size > 0 && (size % _mi_os_page_size()) == 0);

  mi_assert_internal(commit || !allow_large);

  mi_assert_internal(try_alignment > 0);

  *is_zero = true;

  int protect_flags = (commit ? (PROT_WRITE | PROT_READ) : PROT_NONE);

  *addr = unix_mmap(hint_addr, size, try_alignment, protect_flags, false, allow_large, is_large);

  return (*addr != NULL ? 0 : errno);

}

//---------------------------------------------

// Commit/Reset

//---------------------------------------------

static void unix_mprotect_hint(int err) {

  #if defined(__linux__) && (MI_SECURE>=5) // guard page around every mimalloc page

  if (err == ENOMEM) {

    _mi_warning_message("The next warning may be caused by a low memory map limit.\n"

                        "  On Linux this is controlled by the vm.max_map_count -- maybe increase it?\n"

                        "  For example: sudo sysctl -w vm.max_map_count=262144\n");

  }

  #else

  MI_UNUSED(err);

  #endif

}

int _mi_prim_commit(void* start, size_t size, bool* is_zero) {

  // commit: ensure we can access the area

  // note: we may think that *is_zero can be true since the memory

  // was either from mmap PROT_NONE, or from decommit MADV_DONTNEED, but

  // we sometimes call commit on a range with still partially committed

  // memory and `mprotect` does not zero the range.

  *is_zero = false;

  int err = mprotect(start, size, (PROT_READ | PROT_WRITE));

  if (err != 0) {

    err = errno;

    unix_mprotect_hint(err);

  }

  return err;

}

int _mi_prim_reuse(void* start, size_t size) {

  MI_UNUSED(start); MI_UNUSED(size);

  #if defined(__APPLE__) && defined(MADV_FREE_REUSE)

  return unix_madvise(start, size, MADV_FREE_REUSE);

  #endif

  return 0;

}

int _mi_prim_decommit(void* start, size_t size, bool* needs_recommit) {

  int err = 0;

  #if defined(__APPLE__) && defined(MADV_FREE_REUSABLE)

    // decommit on macOS: use MADV_FREE_REUSABLE as it does immediate rss accounting (issue #1097)

    err = unix_madvise(start, size, MADV_FREE_REUSABLE);

    if (err) { err = unix_madvise(start, size, MADV_DONTNEED); }

  #else

    // decommit: use MADV_DONTNEED as it decreases rss immediately (unlike MADV_FREE)

    err = unix_madvise(start, size, MADV_DONTNEED);

  #endif

  #if !MI_DEBUG && MI_SECURE<=2

    *needs_recommit = false;

  #else

    *needs_recommit = true;

    mprotect(start, size, PROT_NONE);

  #endif

  /*

  // decommit: use mmap with MAP_FIXED and PROT_NONE to discard the existing memory (and reduce rss)

  *needs_recommit = true;

  const int fd = unix_mmap_fd();

  void* p = mmap(start, size, PROT_NONE, (MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS | MAP_NORESERVE), fd, 0);

  if (p != start) { err = errno; }

  */

  return err;

}

int _mi_prim_reset(void* start, size_t size) {

  int err = 0;

  // on macOS can use MADV_FREE_REUSABLE (but we disable this for now as it seems slower)

  #if 0 && defined(__APPLE__) && defined(MADV_FREE_REUSABLE)

  err = unix_madvise(start, size, MADV_FREE_REUSABLE);

  if (err==0) return 0;

  // fall through

  #endif

  #if defined(MADV_FREE)

  // Otherwise, we try to use `MADV_FREE` as that is the fastest. A drawback though is that it

  // will not reduce the `rss` stats in tools like `top` even though the memory is available

  // to other processes. With the default `MIMALLOC_PURGE_DECOMMITS=1` we ensure that by

  // default `MADV_DONTNEED` is used though.

  static _Atomic(size_t) advice = MI_ATOMIC_VAR_INIT(MADV_FREE);

  int oadvice = (int)mi_atomic_load_relaxed(&advice);

  while ((err = unix_madvise(start, size, oadvice)) != 0 && err == EAGAIN) { /* try again */ };

  if (err == EINVAL && oadvice == MADV_FREE) {

    // if MADV_FREE is not supported, fall back to MADV_DONTNEED from now on

    mi_atomic_store_release(&advice, (size_t)MADV_DONTNEED);

    err = unix_madvise(start, size, MADV_DONTNEED);

  }

  #else

  err = unix_madvise(start, size, MADV_DONTNEED);

  #endif

  return err;

}

int _mi_prim_protect(void* start, size_t size, bool protect) {

  int err = mprotect(start, size, protect ? PROT_NONE : (PROT_READ | PROT_WRITE));

  if (err != 0) { err = errno; }

  unix_mprotect_hint(err);

  return err;

}

//---------------------------------------------

// Huge page allocation

//---------------------------------------------

#if (MI_INTPTR_SIZE >= 8) && !defined(__HAIKU__) && !defined(__CYGWIN__)

#ifndef MPOL_PREFERRED

#define MPOL_PREFERRED 1

#endif

#if defined(MI_HAS_SYSCALL_H) && defined(SYS_mbind)

static long mi_prim_mbind(void* start, unsigned long len, unsigned long mode, const unsigned long* nmask, unsigned long maxnode, unsigned flags) {

  return syscall(SYS_mbind, start, len, mode, nmask, maxnode, flags);

}

#else

static long mi_prim_mbind(void* start, unsigned long len, unsigned long mode, const unsigned long* nmask, unsigned long maxnode, unsigned flags) {

  MI_UNUSED(start); MI_UNUSED(len); MI_UNUSED(mode); MI_UNUSED(nmask); MI_UNUSED(maxnode); MI_UNUSED(flags);

  return 0;

}

#endif

int _mi_prim_alloc_huge_os_pages(void* hint_addr, size_t size, int numa_node, bool* is_zero, void** addr) {

  bool is_large = true;

  *is_zero = true;

  *addr = unix_mmap(hint_addr, size, MI_SEGMENT_SIZE, PROT_READ | PROT_WRITE, true, true, &is_large);

  if (*addr != NULL && numa_node >= 0 && numa_node < (8*MI_INTPTR_SIZE - 1)) { // at most 63 nodes

    unsigned long numa_mask = (1UL << numa_node);

    // TODO: does `mbind` work correctly for huge OS pages? should we

    // use `set_mempolicy` before calling mmap instead?

    // see: <https://lkml.org/lkml/2017/2/9/875>

    long err = mi_prim_mbind(*addr, size, MPOL_PREFERRED, &numa_mask, 8*MI_INTPTR_SIZE, 0);

    if (err != 0) {

      err = errno;

      _mi_warning_message("failed to bind huge (1GiB) pages to numa node %d (error: %ld (0x%lx))\n", numa_node, err, err);

    }

  }

  return (*addr != NULL ? 0 : errno);

}

#else

int _mi_prim_alloc_huge_os_pages(void* hint_addr, size_t size, int numa_node, bool* is_zero, void** addr) {

  MI_UNUSED(hint_addr); MI_UNUSED(size); MI_UNUSED(numa_node);

  *is_zero = false;

  *addr = NULL;

  return ENOMEM;

}

#endif

//---------------------------------------------

// NUMA nodes

//---------------------------------------------

#if defined(__linux__)

size_t _mi_prim_numa_node(void) {

  #if defined(MI_HAS_SYSCALL_H) && defined(SYS_getcpu)

    unsigned int node = 0;

    unsigned int ncpu = 0;

    int err = syscall(SYS_getcpu, &ncpu, &node, NULL);

    if (err != 0) return 0;

    return node;

  #else

    return 0;

  #endif

}

size_t _mi_prim_numa_node_count(void) {

  char buf[128];

  unsigned last_found = 0;

  for(unsigned node = 1; node < 256; node++) {

    // enumerate node entries -- todo: it there a more efficient way to do this? (but ensure there is no allocation)

    _mi_snprintf(buf, 127, "/sys/devices/system/node/node%u", node);

    if (mi_prim_access(buf,R_OK) != 0) {

      if (node - last_found > 4) break; // allow some sparseness of nodes but not more than 4

    }

    else { last_found = node; }         // highest found node

  }

  return last_found + 1;

}

#elif defined(__FreeBSD__) && __FreeBSD_version >= 1200000

size_t _mi_prim_numa_node(void) {

  domainset_t dom;

  size_t node;

  int policy;

  if (cpuset_getdomain(CPU_LEVEL_CPUSET, CPU_WHICH_PID, -1, sizeof(dom), &dom, &policy) == -1) return 0ul;

  for (node = 0; node < MAXMEMDOM; node++) {

    if (DOMAINSET_ISSET(node, &dom)) return node;

  }

  return 0ul;

}

size_t _mi_prim_numa_node_count(void) {

  size_t ndomains = 0;

  size_t len = sizeof(ndomains);

  if (sysctlbyname("vm.ndomains", &ndomains, &len, NULL, 0) == -1) return 0ul;

  return ndomains;

}

#elif defined(__DragonFly__)

size_t _mi_prim_numa_node(void) {

  // TODO: DragonFly does not seem to provide any userland means to get this information.

  return 0ul;

}

size_t _mi_prim_numa_node_count(void) {

  size_t ncpus = 0, nvirtcoresperphys = 0;

  size_t len = sizeof(size_t);

  if (sysctlbyname("hw.ncpu", &ncpus, &len, NULL, 0) == -1) return 0ul;

  if (sysctlbyname("hw.cpu_topology_ht_ids", &nvirtcoresperphys, &len, NULL, 0) == -1) return 0ul;

  return nvirtcoresperphys * ncpus;

}

#else

size_t _mi_prim_numa_node(void) {

  return 0;

}

size_t _mi_prim_numa_node_count(void) {

  return 1;

}

#endif

// ----------------------------------------------------------------

// Clock

// ----------------------------------------------------------------

#include <time.h>

// low resolution timer

static mi_msecs_t mi_prim_clock_now_lowres(void) {

  const int64_t ticks = (int64_t)clock();

  #if !defined(CLOCKS_PER_SEC)

    return ticks;

  #else

    if (CLOCKS_PER_SEC <= 0 || CLOCKS_PER_SEC == 1000) {

      return ticks;

    }

    else if (CLOCKS_PER_SEC > 0 && CLOCKS_PER_SEC < 1000) {

      return ticks * (1000 / (mi_msecs_t)CLOCKS_PER_SEC);

    }

    else {

      return ticks / ((mi_msecs_t)CLOCKS_PER_SEC / 1000);

    }

  #endif

}

mi_msecs_t _mi_prim_clock_now(void) {

  #if defined(CLOCK_REALTIME) || defined(CLOCK_MONOTONIC)

    #ifdef CLOCK_MONOTONIC

    const clockid_t clockid = CLOCK_MONOTONIC;

    #else

    const clockid_t clockid = CLOCK_REALTIME;

    #endif

    struct timespec t;

    if (clock_gettime(clockid,&t) == 0) {

      return ((mi_msecs_t)t.tv_sec * 1000) + ((mi_msecs_t)t.tv_nsec / 1000000L);

    }

  #endif

  return mi_prim_clock_now_lowres();

}

//----------------------------------------------------------------

// Process info

//----------------------------------------------------------------

#if defined(__unix__) || defined(__unix) || defined(unix) || defined(__APPLE__) || defined(__HAIKU__)

#include <stdio.h>

#include <unistd.h>

#include <sys/resource.h>

#if defined(__APPLE__)

#include <mach/mach.h>

#endif

#if defined(__HAIKU__)

#include <kernel/OS.h>

#endif

static mi_msecs_t timeval_secs(const struct timeval* tv) {

  return ((mi_msecs_t)tv->tv_sec * 1000L) + ((mi_msecs_t)tv->tv_usec / 1000L);

}

void _mi_prim_process_info(mi_process_info_t* pinfo)

{

  struct rusage rusage;

  if (getrusage(RUSAGE_SELF, &rusage) == 0) {

    pinfo->utime = timeval_secs(&rusage.ru_utime);

    pinfo->stime = timeval_secs(&rusage.ru_stime);

    #if !defined(__HAIKU__)

      pinfo->page_faults = rusage.ru_majflt;

    #endif

    #if defined(__APPLE__)

      pinfo->peak_rss = rusage.ru_maxrss;         // macos reports in bytes

    #else

      pinfo->peak_rss = rusage.ru_maxrss * 1024;  // Linux/BSD report in KiB

    #endif

  }

  #if defined(__HAIKU__)

    // Haiku does not have (yet?) a way to

    // get these stats per process

    thread_info tid;

    if (get_thread_info(find_thread(0), &tid) == B_OK) {

      area_info mem;

      ssize_t c;

      while (get_next_area_info(tid.team, &c, &mem) == B_OK) {

        pinfo->peak_rss += mem.ram_size;

      }

    }

    pinfo->page_faults = 0;

  #elif defined(__APPLE__)

    #ifdef MACH_TASK_BASIC_INFO

    struct mach_task_basic_info info;

    mach_msg_type_number_t infoCount = MACH_TASK_BASIC_INFO_COUNT;

    if (task_info(mach_task_self(), MACH_TASK_BASIC_INFO, (task_info_t)&info, &infoCount) == KERN_SUCCESS) {

      pinfo->current_rss = (size_t)info.resident_size;

    }

    #else

    struct task_basic_info info;

    mach_msg_type_number_t infoCount = TASK_BASIC_INFO_COUNT;

    if (task_info(mach_task_self(), TASK_BASIC_INFO, (task_info_t)&info, &infoCount) == KERN_SUCCESS) {

      pinfo->current_rss = (size_t)info.resident_size;

    }

    #endif

  #endif

  // use defaults for commit

}

#else

#ifndef __wasi__

// WebAssembly instances are not processes

#pragma message("define a way to get process info")

#endif

void _mi_prim_process_info(mi_process_info_t* pinfo)

{

  // use defaults

  MI_UNUSED(pinfo);

}

#endif

//----------------------------------------------------------------

// Output

//----------------------------------------------------------------

void _mi_prim_out_stderr( const char* msg ) {

  fputs(msg,stderr);

}

//----------------------------------------------------------------

// Environment

//----------------------------------------------------------------

#if !defined(MI_USE_ENVIRON) || (MI_USE_ENVIRON!=0)

// On Posix systemsr use `environ` to access environment variables

// even before the C runtime is initialized.

#if defined(__APPLE__) && defined(__has_include) && __has_include(<crt_externs.h>)

#include <crt_externs.h>

static char** mi_get_environ(void) {

  return (*_NSGetEnviron());

}

#else

extern char** environ;

static char** mi_get_environ(void) {

  return environ;

}

#endif

int _mi_prim_getenv(const char* name, char* result, size_t result_size) {

  if (name==NULL) return -1;

  const size_t len = _mi_strlen(name);

  if (len == 0) return -1;

  char** env = mi_get_environ();

  if (env == NULL) return -1;

  // compare up to 10000 entries

  for (int i = 0; i < 10000 && env[i] != NULL; i++) {

    const char* s = env[i];

    if (_mi_strnicmp(name, s, len) == 0 && s[len] == '=') { // case insensitive

      // found it

      if (!_mi_strlcpy(result, s + len + 1, result_size)) return -1;

      return 1;   // success

    }

  }

  return 0; // not found

}

#else

// fallback: use standard C `getenv` but this cannot be used while initializing the C runtime

int _mi_prim_getenv(const char* name, char* result, size_t result_size) {

  // cannot call getenv() when still initializing the C runtime.

  if (_mi_preloading()) return -1;  // error, try again later

  const char* s = getenv(name);

  if (s == NULL) {

    // we check the upper case name too.

    char buf[64+1];

    size_t len = _mi_strnlen(name,sizeof(buf)-1);

    for (size_t i = 0; i < len; i++) {

      buf[i] = _mi_toupper(name[i]);

    }

    buf[len] = 0;

    s = getenv(buf);

  }

  if (s == NULL || _mi_strnlen(s,result_size) >= result_size) return 0; // not found

  if (!_mi_strlcpy(result, s, result_size)) return -1;

  return 1;  // success

}

#endif  // !MI_USE_ENVIRON

//----------------------------------------------------------------

// Random

//----------------------------------------------------------------

#if defined(__APPLE__) && defined(MAC_OS_X_VERSION_10_15) && (MAC_OS_X_VERSION_MIN_REQUIRED >= MAC_OS_X_VERSION_10_15)

#include <CommonCrypto/CommonCryptoError.h>

#include <CommonCrypto/CommonRandom.h>

bool _mi_prim_random_buf(void* buf, size_t buf_len) {

  // We prefer CCRandomGenerateBytes as it returns an error code while arc4random_buf

  // may fail silently on macOS. See PR #390, and <https://opensource.apple.com/source/Libc/Libc-1439.40.11/gen/FreeBSD/arc4random.c.auto.html>

  return (CCRandomGenerateBytes(buf, buf_len) == kCCSuccess);

}

#elif defined(__ANDROID__) || defined(__DragonFly__) || \

      defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__) || \

      defined(__sun) || \

      defined(__CYGWIN__) /* CPython: gh-148461 */ || \

      (defined(__APPLE__) && (MAC_OS_X_VERSION_MIN_REQUIRED >= MAC_OS_X_VERSION_10_7))

bool _mi_prim_random_buf(void* buf, size_t buf_len) {

  arc4random_buf(buf, buf_len);

  return true;

}

#elif defined(__APPLE__) || defined(__linux__) || defined(__HAIKU__)   // also for old apple versions < 10.7 (issue #829)

#include <sys/types.h>

#include <sys/stat.h>

#include <errno.h>

bool _mi_prim_random_buf(void* buf, size_t buf_len) {

  // Modern Linux provides `getrandom` but different distributions either use `sys/random.h` or `linux/random.h`

  // and for the latter the actual `getrandom` call is not always defined.

  // (see <https://stackoverflow.com/questions/45237324/why-doesnt-getrandom-compile>)

  // We therefore use a syscall directly and fall back dynamically to /dev/urandom when needed.

  #if defined(MI_HAS_SYSCALL_H) && defined(SYS_getrandom)

    #ifndef GRND_NONBLOCK

    #define GRND_NONBLOCK (1)

    #endif

    static _Atomic(uintptr_t) no_getrandom; // = 0

    if (mi_atomic_load_acquire(&no_getrandom)==0) {

      ssize_t ret = syscall(SYS_getrandom, buf, buf_len, GRND_NONBLOCK);

      if (ret >= 0) return (buf_len == (size_t)ret);

      if (errno != ENOSYS) return false;

      mi_atomic_store_release(&no_getrandom, (uintptr_t)1); // don't call again, and fall back to /dev/urandom

    }

  #endif

  int flags = O_RDONLY;

  #if defined(O_CLOEXEC)

  flags |= O_CLOEXEC;

  #endif

  int fd = mi_prim_open("/dev/urandom", flags);

  if (fd < 0) return false;

  size_t count = 0;

  while(count < buf_len) {

    ssize_t ret = mi_prim_read(fd, (char*)buf + count, buf_len - count);

    if (ret==0) {

      break;

    }

    else if (ret<0) {

      if (errno!=EAGAIN && errno!=EINTR) break;

    }

    else {

      count += ret;

    }

  }

  mi_prim_close(fd);

  return (count==buf_len);

}

#else

bool _mi_prim_random_buf(void* buf, size_t buf_len) {

  return false;

}

#endif

//----------------------------------------------------------------

// Thread init/done

//----------------------------------------------------------------

#if defined(MI_USE_PTHREADS)

// use pthread local storage keys to detect thread ending

// (and used with MI_TLS_PTHREADS for the default heap)

pthread_key_t _mi_heap_default_key = (pthread_key_t)(-1);

static void mi_pthread_done(void* value) {

  if (value!=NULL) {

    _mi_thread_done((mi_heap_t*)value);

  }

}

void _mi_prim_thread_init_auto_done(void) {

  mi_assert_internal(_mi_heap_default_key == (pthread_key_t)(-1));

  const int err = pthread_key_create(&_mi_heap_default_key, &mi_pthread_done);

  if (err!=0) {

    _mi_error_message(err,"unable to create a pthread thread local key (error %d (0x%x))", err, err);

    _mi_heap_default_key = (pthread_key_t)(-1);

  };

}

void _mi_prim_thread_done_auto_done(void) {

  if (_mi_heap_default_key != (pthread_key_t)(-1)) {  // do not leak the key, see issue #809

    pthread_key_delete(_mi_heap_default_key);

  }

}

void _mi_prim_thread_associate_default_heap(mi_heap_t* heap) {

  if (_mi_heap_default_key != (pthread_key_t)(-1)) {  // can happen during recursive invocation on freeBSD

    pthread_setspecific(_mi_heap_default_key, heap);

  }

}

#else

void _mi_prim_thread_init_auto_done(void) {

  // nothing

}

void _mi_prim_thread_done_auto_done(void) {

  // nothing

}

void _mi_prim_thread_associate_default_heap(mi_heap_t* heap) {

  MI_UNUSED(heap);

}

#endif