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 #ifdef __sparc__
  39     /* SPARC Linux needs greater alignment than the pagesize */
  40     return QEMU_VMALLOC_ALIGN;
  41 #endif
  42 #endif
  43
  44     return getpagesize();
  45 }
  46
  47 size_t qemu_mempath_getpagesize(const char *mem_path)
  48 {
  49 #ifdef CONFIG_LINUX
  50     struct statfs fs;
  51     int ret;
  52
  53     do {
  54         ret = statfs(mem_path, &fs);
  55     } while (ret != 0 && errno == EINTR);
  56
  57     if (ret != 0) {
  58         fprintf(stderr, "Couldn't statfs() memory path: %s\n",
  59                 strerror(errno));
  60         exit(1);
  61     }
  62
  63     if (fs.f_type == HUGETLBFS_MAGIC) {
  64         /* It's hugepage, return the huge page size */
  65         return fs.f_bsize;
  66     }
  67 #ifdef __sparc__
  68     /* SPARC Linux needs greater alignment than the pagesize */
  69     return QEMU_VMALLOC_ALIGN;
  70 #endif
  71 #endif
  72
  73     return getpagesize();
  74 }
  75
  76 void *qemu_ram_mmap(int fd, size_t size, size_t align, bool shared)
  77 {
  78     /*
  79      * Note: this always allocates at least one extra page of virtual address
  80      * space, even if size is already aligned.
  81      */
  82     size_t total = size + align;
  83 #if defined(__powerpc64__) && defined(__linux__)
  84     /* On ppc64 mappings in the same segment (aka slice) must share the same
  85      * page size. Since we will be re-allocating part of this segment
  86      * from the supplied fd, we should make sure to use the same page size, to
  87      * this end we mmap the supplied fd.  In this case, set MAP_NORESERVE to
  88      * avoid allocating backing store memory.
  89      * We do this unless we are using the system page size, in which case
  90      * anonymous memory is OK.
  91      */
  92     int anonfd = fd == -1 || qemu_fd_getpagesize(fd) == getpagesize() ? -1 : fd;
  93     int flags = anonfd == -1 ? MAP_ANONYMOUS : MAP_NORESERVE;
  94     void *ptr = mmap(0, total, PROT_NONE, flags | MAP_PRIVATE, anonfd, 0);
  95 #else
  96     void *ptr = mmap(0, total, PROT_NONE, MAP_ANONYMOUS | MAP_PRIVATE, -1, 0);
  97 #endif
  98     size_t offset;
  99     void *ptr1;
 100
 101     if (ptr == MAP_FAILED) {
 102         return MAP_FAILED;
 103     }
 104
 105     assert(is_power_of_2(align));
 106     /* Always align to host page size */
 107     assert(align >= getpagesize());
 108
 109     offset = QEMU_ALIGN_UP((uintptr_t)ptr, align) - (uintptr_t)ptr;
 110     ptr1 = mmap(ptr + offset, size, PROT_READ | PROT_WRITE,
 111                 MAP_FIXED |
 112                 (fd == -1 ? MAP_ANONYMOUS : 0) |
 113                 (shared ? MAP_SHARED : MAP_PRIVATE),
 114                 fd, 0);
 115     if (ptr1 == MAP_FAILED) {
 116         munmap(ptr, total);
 117         return MAP_FAILED;
 118     }
 119
 120     if (offset > 0) {
 121         munmap(ptr, offset);
 122     }
 123
 124     /*
 125      * Leave a single PROT_NONE page allocated after the RAM block, to serve as
 126      * a guard page guarding against potential buffer overflows.
 127      */
 128     total -= offset;
 129     if (total > size + getpagesize()) {
 130         munmap(ptr1 + size + getpagesize(), total - size - getpagesize());
 131     }
 132
 133     return ptr1;
 134 }
 135
 136 void qemu_ram_munmap(void *ptr, size_t size)
 137 {
 138     if (ptr) {
 139         /* Unmap both the RAM block and the guard page */
 140         munmap(ptr, size + getpagesize());
 141     }
 142 }