util/mmap-alloc.c

   1 /*
   2  * Support for RAM backed by mmaped host memory.
   3  *
   4  * Copyright (c) 2015 Red Hat, Inc.
   5  *
   6  * Authors:
   7  *  Michael S. Tsirkin <mst@redhat.com>
   8  *
   9  * This work is licensed under the terms of the GNU GPL, version 2 or
  10  * later.  See the COPYING file in the top-level directory.
  11  */
  12 #include "qemu/osdep.h"
  13 #include <qemu/mmap-alloc.h>
  14
  15 #define HUGETLBFS_MAGIC       0x958458f6
  16
  17 #ifdef CONFIG_LINUX
  18 #include <sys/vfs.h>
  19 #endif
  20
  21 size_t qemu_fd_getpagesize(int fd)
  22 {
  23 #ifdef CONFIG_LINUX
  24     struct statfs fs;
  25     int ret;
  26
  27     if (fd != -1) {
  28         do {
  29             ret = fstatfs(fd, &fs);
  30         } while (ret != 0 && errno == EINTR);
  31
  32         if (ret == 0 && fs.f_type == HUGETLBFS_MAGIC) {
  33             return fs.f_bsize;
  34         }
  35     }
  36 #endif
  37
  38     return getpagesize();
  39 }
  40
  41 void *qemu_ram_mmap(int fd, size_t size, size_t align, bool shared)
  42 {
  43     /*
  44      * Note: this always allocates at least one extra page of virtual address
  45      * space, even if size is already aligned.
  46      */
  47     size_t total = size + align;
  48 #if defined(__powerpc64__) && defined(__linux__)
  49     /* On ppc64 mappings in the same segment (aka slice) must share the same
  50      * page size. Since we will be re-allocating part of this segment
  51      * from the supplied fd, we should make sure to use the same page size, to
  52      * this end we mmap the supplied fd.  In this case, set MAP_NORESERVE to
  53      * avoid allocating backing store memory.
  54      * We do this unless we are using the system page size, in which case
  55      * anonymous memory is OK.
  56      */
  57     int anonfd = fd == -1 || qemu_fd_getpagesize(fd) == getpagesize() ? -1 : fd;
  58     int flags = anonfd == -1 ? MAP_ANONYMOUS : MAP_NORESERVE;
  59     void *ptr = mmap(0, total, PROT_NONE, flags | MAP_PRIVATE, anonfd, 0);
  60 #else
  61     void *ptr = mmap(0, total, PROT_NONE, MAP_ANONYMOUS | MAP_PRIVATE, -1, 0);
  62 #endif
  63     size_t offset = QEMU_ALIGN_UP((uintptr_t)ptr, align) - (uintptr_t)ptr;
  64     void *ptr1;
  65
  66     if (ptr == MAP_FAILED) {
  67         return MAP_FAILED;
  68     }
  69
  70     /* Make sure align is a power of 2 */
  71     assert(!(align & (align - 1)));
  72     /* Always align to host page size */
  73     assert(align >= getpagesize());
  74
  75     ptr1 = mmap(ptr + offset, size, PROT_READ | PROT_WRITE,
  76                 MAP_FIXED |
  77                 (fd == -1 ? MAP_ANONYMOUS : 0) |
  78                 (shared ? MAP_SHARED : MAP_PRIVATE),
  79                 fd, 0);
  80     if (ptr1 == MAP_FAILED) {
  81         munmap(ptr, total);
  82         return MAP_FAILED;
  83     }
  84
  85     ptr += offset;
  86     total -= offset;
  87
  88     if (offset > 0) {
  89         munmap(ptr - offset, offset);
  90     }
  91
  92     /*
  93      * Leave a single PROT_NONE page allocated after the RAM block, to serve as
  94      * a guard page guarding against potential buffer overflows.
  95      */
  96     if (total > size + getpagesize()) {
  97         munmap(ptr + size + getpagesize(), total - size - getpagesize());
  98     }
  99
 100     return ptr;
 101 }
 102
 103 void qemu_ram_munmap(void *ptr, size_t size)
 104 {
 105     if (ptr) {
 106         /* Unmap both the RAM block and the guard page */
 107         munmap(ptr, size + getpagesize());
 108     }
 109 }