Re: [Qemu-devel] [PATCH v2] exec: factor out duplicate mmap code

["Michael S. Tsirkin" <mst@redhat.com>]

On Fri, Oct 02, 2015 at 10:48:13AM +1000, Richard Henderson wrote:
> On 10/01/2015 10:58 PM, Michael S. Tsirkin wrote:
> >Anonymous and file-backed RAM allocation are now almost exactly the same.
> >
> >Reduce code duplication by moving RAM mmap code out of oslib-posix.c and
> >exec.c.
> >
> >Reported-by: Marc-André Lureau <mlureau@redhat.com>
> >Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
> >Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
> >Acked-by: Paolo Bonzini <pbonzini@redhat.com>
> >---
> >
> >Changes from v1: add shared flag to get MAP_SHARED mappings
> >(for vhost-user), only set MAP_ANONYMOUS for anonymous RAM.
> >
> >  include/qemu/mmap-alloc.h | 10 +++++++
> >  exec.c                    | 47 +++++++------------------------
> >  util/mmap-alloc.c         | 71 +++++++++++++++++++++++++++++++++++++++++++++++
> >  util/oslib-posix.c        | 28 +++----------------
> >  util/Makefile.objs        |  2 +-
> >  5 files changed, 96 insertions(+), 62 deletions(-)
> >  create mode 100644 include/qemu/mmap-alloc.h
> >  create mode 100644 util/mmap-alloc.c
> >
> >diff --git a/include/qemu/mmap-alloc.h b/include/qemu/mmap-alloc.h
> >new file mode 100644
> >index 0000000..56388e6
> >--- /dev/null
> >+++ b/include/qemu/mmap-alloc.h
> >@@ -0,0 +1,10 @@
> >+#ifndef QEMU_MMAP_ALLOC
> >+#define QEMU_MMAP_ALLOC
> >+
> >+#include "qemu-common.h"
> >+
> >+void *qemu_ram_mmap(int fd, size_t size, size_t align, bool shared);
> >+
> >+void qemu_ram_munmap(void *ptr, size_t size);
> >+
> >+#endif
> >diff --git a/exec.c b/exec.c
> >index 7d90a52..4505dc7 100644
> >--- a/exec.c
> >+++ b/exec.c
> >@@ -55,6 +55,9 @@
> >  #include "exec/ram_addr.h"
> >
> >  #include "qemu/range.h"
> >+#ifndef _WIN32
> >+#include "qemu/mmap-alloc.h"
> >+#endif
> >
> >  //#define DEBUG_SUBPAGE
> >
> >@@ -84,9 +87,9 @@ static MemoryRegion io_mem_unassigned;
> >   */
> >  #define RAM_RESIZEABLE (1 << 2)
> >
> >-/* An extra page is mapped on top of this RAM.
> >+/* RAM is backed by an mmapped file.
> >   */
> >-#define RAM_EXTRA (1 << 3)
> >+#define RAM_FILE (1 << 3)
> >  #endif
> >
> >  struct CPUTailQ cpus = QTAILQ_HEAD_INITIALIZER(cpus);
> >@@ -1188,13 +1191,10 @@ static void *file_ram_alloc(RAMBlock *block,
> >      char *filename;
> >      char *sanitized_name;
> >      char *c;
> >-    void *ptr;
> >-    void *area = NULL;
> >+    void *area;
> >      int fd;
> >      uint64_t hpagesize;
> >-    uint64_t total;
> >      Error *local_err = NULL;
> >-    size_t offset;
> >
> >      hpagesize = gethugepagesize(path, &local_err);
> >      if (local_err) {
> >@@ -1238,7 +1238,6 @@ static void *file_ram_alloc(RAMBlock *block,
> >      g_free(filename);
> >
> >      memory = ROUND_UP(memory, hpagesize);
> >-    total = memory + hpagesize;
> >
> >      /*
> >       * ftruncate is not supported by hugetlbfs in older
> >@@ -1250,40 +1249,14 @@ static void *file_ram_alloc(RAMBlock *block,
> >          perror("ftruncate");
> >      }
> >
> >-    ptr = mmap(0, total, PROT_NONE, MAP_PRIVATE | MAP_ANONYMOUS,
> >-                -1, 0);
> >-    if (ptr == MAP_FAILED) {
> >-        error_setg_errno(errp, errno,
> >-                         "unable to allocate memory range for hugepages");
> >-        close(fd);
> >-        goto error;
> >-    }
> >-
> >-    offset = QEMU_ALIGN_UP((uintptr_t)ptr, hpagesize) - (uintptr_t)ptr;
> >-
> >-    area = mmap(ptr + offset, memory, PROT_READ | PROT_WRITE,
> >-                (block->flags & RAM_SHARED ? MAP_SHARED : MAP_PRIVATE) |
> >-                MAP_FIXED,
> >-                fd, 0);
> >+    area = qemu_ram_mmap(fd, memory, hpagesize, block->flags & RAM_SHARED);
> >      if (area == MAP_FAILED) {
> >          error_setg_errno(errp, errno,
> >                           "unable to map backing store for hugepages");
> >-        munmap(ptr, total);
> >          close(fd);
> >          goto error;
> >      }
> >
> >-    if (offset > 0) {
> >-        munmap(ptr, offset);
> >-    }
> >-    ptr += offset;
> >-    total -= offset;
> >-
> >-    if (total > memory + getpagesize()) {
> >-        munmap(ptr + memory + getpagesize(),
> >-               total - memory - getpagesize());
> >-    }
> >-
> >      if (mem_prealloc) {
> >          os_mem_prealloc(fd, area, memory);
> >      }
> >@@ -1601,7 +1574,7 @@ ram_addr_t qemu_ram_alloc_from_file(ram_addr_t size, MemoryRegion *mr,
> >      new_block->used_length = size;
> >      new_block->max_length = size;
> >      new_block->flags = share ? RAM_SHARED : 0;
> >-    new_block->flags |= RAM_EXTRA;
> >+    new_block->flags |= RAM_FILE;
> >      new_block->host = file_ram_alloc(new_block, size,
> >                                       mem_path, errp);
> >      if (!new_block->host) {
> >@@ -1703,8 +1676,8 @@ static void reclaim_ramblock(RAMBlock *block)
> >          xen_invalidate_map_cache_entry(block->host);
> >  #ifndef _WIN32
> >      } else if (block->fd >= 0) {
> >-        if (block->flags & RAM_EXTRA) {
> >-            munmap(block->host, block->max_length + getpagesize());
> >+        if (block->flags & RAM_FILE) {
> >+            qemu_ram_munmap(block->host, block->max_length);
> >          } else {
> >              munmap(block->host, block->max_length);
> >          }
> >diff --git a/util/mmap-alloc.c b/util/mmap-alloc.c
> >new file mode 100644
> >index 0000000..e82cc94
> >--- /dev/null
> >+++ b/util/mmap-alloc.c
> >@@ -0,0 +1,71 @@
> >+/*
> >+ * Support for RAM backed by mmaped host memory.
> >+ *
> >+ * Copyright (c) 2015 Red Hat, Inc.
> >+ *
> >+ * Authors:
> >+ *  Michael S. Tsirkin <mst@redhat.com>
> >+ *
> >+ * This work is licensed under the terms of the GNU GPL, version 2 or
> >+ * later.  See the COPYING file in the top-level directory.
> >+ */
> >+#include <qemu/mmap-alloc.h>
> >+#include <sys/types.h>
> >+#include <sys/mman.h>
> >+#include <assert.h>
> >+
> >+void *qemu_ram_mmap(int fd, size_t size, size_t align, bool shared)
> 
> Do we need generic "align", or would a bool hugepage suffice?  Or I suppose
> this is also used for non-anonymous hugepages via /dev/hugepage where we
> have multiple huge page sizes?

Right. Further, passing in alignment makes it possible to
keep page size probing in exec.c which is the only user.


> Should we call this function something else, so that we can use it to
> allocate hugepage aligned storage for code_gen_buffer and/or the tbs array?


I don't much care what do we call it - with two callers it would be easy
to rename down the road.
The reason I put in "RAM" is because it allocates a guard page
which only seems to make sense for RAM memory that can be mapped.

But if you have a better name in mind, please let me know.


> >+{
> >+    /*
> >+     * Note: this always allocates at least one extra page of virtual address
> >+     * space, even if size is already aligned.
> >+     */
> >+    size_t total = size + align;
> >+    void *ptr = mmap(0, total, PROT_NONE, MAP_ANONYMOUS | MAP_PRIVATE, -1, 0);
> >+    size_t offset = QEMU_ALIGN_UP((uintptr_t)ptr, align) - (uintptr_t)ptr;
> >+    void *ptr1;
> >+
> >+    if (ptr == MAP_FAILED) {
> >+        return NULL;
> >+    }
> >+
> >+    /* Make sure align is a power of 2 */
> >+    assert(!(align & (align - 1)));
> >+    /* Always align to host page size */
> >+    assert(align >= getpagesize());
> 
> Is qemu_real_host_page_size initialized yet?
> If not, should we move the initialization forward?

I think it is.
Though qemu_real_host_page_size seems to only be needed
for portability, and this file is posix-specific.

Anyway, we are moving code from util/oslib-posix.c and exec.c that
currently both call getpagesize in this function, it seems better to be
consistent.

We can refactor afterwards, correct?


> >+
> >+    ptr1 = mmap(ptr + offset, size, PROT_READ | PROT_WRITE,
> >+                MAP_FIXED |
> >+                (fd == -1 ? MAP_ANONYMOUS : 0) |
> >+                (shared ? MAP_SHARED : MAP_PRIVATE),
> >+                fd, 0);
> 
> If fd == -1, we really only need mprotect on the region, correct?

Yes. But that's already like this in util/oslib-posix.c,
this is just re-factoring without logic changes.

Any reason to prefer mprotect? That would be more LOC.

> >+    if (ptr1 == MAP_FAILED) {
> >+        munmap(ptr, total);
> >+        return NULL;
> >+    }
> >+
> >+    ptr += offset;
> >+    total -= offset;
> >+
> >+    if (offset > 0) {
> >+        munmap(ptr - offset, offset);
> >+    }
> >+
> >+    /*
> >+     * Leave a single PROT_NONE page allocated after the RAM block, to serve as
> >+     * a guard page guarding against potential buffer overflows.
> >+     */
> >+    if (total > size + getpagesize()) {
> >+        munmap(ptr + size + getpagesize(), total - size - getpagesize());
> >+    }
> >+
> >+    return ptr;
> >+}