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
  13 #include "qemu/osdep.h"
  14 #include "qemu/mmap-alloc.h"
  15 #include "qemu/host-utils.h"
  16
  17 #define HUGETLBFS_MAGIC       0x958458f6
  18
  19 #ifdef CONFIG_LINUX
  20 #include <sys/vfs.h>
  21 #endif
  22
  23 size_t qemu_fd_getpagesize(int fd)
  24 {
  25 #ifdef CONFIG_LINUX
  26     struct statfs fs;
  27     int ret;
  28
  29     if (fd != -1) {
  30         do {
  31             ret = fstatfs(fd, &fs);
  32         } while (ret != 0 && errno == EINTR);
  33
  34         if (ret == 0 && fs.f_type == HUGETLBFS_MAGIC) {
  35             return fs.f_bsize;
  36         }
  37     }
  38 #endif
  39
  40     return getpagesize();
  41 }
  42
  43 size_t qemu_mempath_getpagesize(const char *mem_path)
  44 {
  45 #ifdef CONFIG_LINUX
  46     struct statfs fs;
  47     int ret;
  48
  49     do {
  50         ret = statfs(mem_path, &fs);
  51     } while (ret != 0 && errno == EINTR);
  52
  53     if (ret != 0) {
  54         fprintf(stderr, "Couldn't statfs() memory path: %s\n",
  55                 strerror(errno));
  56         exit(1);
  57     }
  58
  59     if (fs.f_type == HUGETLBFS_MAGIC) {
  60         /* It's hugepage, return the huge page size */
  61         return fs.f_bsize;
  62     }
  63 #endif
  64
  65     return getpagesize();
  66 }
  67
  68 void *qemu_ram_mmap(int fd, size_t size, size_t align, bool shared)
  69 {
  70     /*
  71      * Note: this always allocates at least one extra page of virtual address
  72      * space, even if size is already aligned.
  73      */
  74     size_t total = size + align;
  75 #if defined(__powerpc64__) && defined(__linux__)
  76     /* On ppc64 mappings in the same segment (aka slice) must share the same
  77      * page size. Since we will be re-allocating part of this segment
  78      * from the supplied fd, we should make sure to use the same page size, to
  79      * this end we mmap the supplied fd.  In this case, set MAP_NORESERVE to
  80      * avoid allocating backing store memory.
  81      * We do this unless we are using the system page size, in which case
  82      * anonymous memory is OK.
  83      */
  84     int anonfd = fd == -1 || qemu_fd_getpagesize(fd) == getpagesize() ? -1 : fd;
  85     int flags = anonfd == -1 ? MAP_ANONYMOUS : MAP_NORESERVE;
  86     void *ptr = mmap(0, total, PROT_NONE, flags | MAP_PRIVATE, anonfd, 0);
  87 #else
  88     void *ptr = mmap(0, total, PROT_NONE, MAP_ANONYMOUS | MAP_PRIVATE, -1, 0);
  89 #endif
  90     size_t offset;
  91     void *ptr1;
  92
  93     if (ptr == MAP_FAILED) {
  94         return MAP_FAILED;
  95     }
  96
  97     assert(is_power_of_2(align));
  98     /* Always align to host page size */
  99     assert(align >= getpagesize());
 100
 101     offset = QEMU_ALIGN_UP((uintptr_t)ptr, align) - (uintptr_t)ptr;
 102     ptr1 = mmap(ptr + offset, size, PROT_READ | PROT_WRITE,
 103                 MAP_FIXED |
 104                 (fd == -1 ? MAP_ANONYMOUS : 0) |
 105                 (shared ? MAP_SHARED : MAP_PRIVATE),
 106                 fd, 0);
 107     if (ptr1 == MAP_FAILED) {
 108         munmap(ptr, total);
 109         return MAP_FAILED;
 110     }
 111
 112     if (offset > 0) {
 113         munmap(ptr, offset);
 114     }
 115
 116     /*
 117      * Leave a single PROT_NONE page allocated after the RAM block, to serve as
 118      * a guard page guarding against potential buffer overflows.
 119      */
 120     total -= offset;
 121     if (total > size + getpagesize()) {
 122         munmap(ptr1 + size + getpagesize(), total - size - getpagesize());
 123     }
 124
 125     return ptr1;
 126 }
 127
 128 void qemu_ram_munmap(void *ptr, size_t size)
 129 {
 130     if (ptr) {
 131         /* Unmap both the RAM block and the guard page */
 132         munmap(ptr, size + getpagesize());
 133     }
 134 }