Paolo Bonzini <pbonz...@redhat.com> writes:
> Using qemu_memalign only leaves the RAM zero by chance, because libc
> will usually use mmap to satisfy our huge requests. But memory will
> not be zero when using MALLOC_PERTURB_ with a nonzero value. In the
> case of incoming migration, this breaks a recently-introduced
> invariant (commit f1c7279, migration: do not sent zero pages in
> bulk stage, 2013-03-26).
>
> To fix this, use mmap ourselves to get a well-aligned, always zero
> block for the RAM. Mmap-ed memory is easy to "trim" at the sides.
>
> This also removes the need to do something special on valgrind
> (see commit c2a8238a, Support running QEMU on Valgrind, 2011-10-31),
> thus effectively reverts that patch.
>
> Reviewed-by: Juan Quintela <quint...@redhat.com>
> Signed-off-by: Paolo Bonzini <pbonz...@redhat.com>
> ---
> v2->v3: use MAP_FAILED. You learn something every day.
>
> util/oslib-posix.c | 35 ++++++++++++++++++-----------------
> 1 file changed, 18 insertions(+), 17 deletions(-)
>
> diff --git a/util/oslib-posix.c b/util/oslib-posix.c
> index 4e4b819..bda62c0 100644
> --- a/util/oslib-posix.c
> +++ b/util/oslib-posix.c
> @@ -40,7 +40,6 @@ extern int daemon(int, int);
> Valgrind does not support alignments larger than 1 MiB,
> therefore we need special code which handles running on Valgrind. */
> # define QEMU_VMALLOC_ALIGN (512 * 4096)
> -# define CONFIG_VALGRIND
> #elif defined(__linux__) && defined(__s390x__)
> /* Use 1 MiB (segment size) alignment so gmap can be used by KVM. */
> # define QEMU_VMALLOC_ALIGN (256 * 4096)
> @@ -52,12 +51,8 @@ extern int daemon(int, int);
> #include "sysemu/sysemu.h"
> #include "trace.h"
> #include "qemu/sockets.h"
> +#include <sys/mman.h>
>
> -#if defined(CONFIG_VALGRIND)
> -static int running_on_valgrind = -1;
> -#else
> -# define running_on_valgrind 0
> -#endif
> #ifdef CONFIG_LINUX
> #include <sys/syscall.h>
> #endif
> @@ -108,22 +103,28 @@ void *qemu_memalign(size_t alignment, size_t size)
> /* alloc shared memory pages */
> void *qemu_vmalloc(size_t size)
> {
> - void *ptr;
> size_t align = QEMU_VMALLOC_ALIGN;
> + size_t total = size + align - getpagesize();
> + void *ptr = mmap(0, total, PROT_READ | PROT_WRITE,
> + MAP_ANONYMOUS | MAP_PRIVATE, -1, 0);
> + size_t offset = QEMU_ALIGN_UP((uintptr_t)ptr, align) - (uintptr_t)ptr;
>
> -#if defined(CONFIG_VALGRIND)
> - if (running_on_valgrind < 0) {
> - /* First call, test whether we are running on Valgrind.
> - This is a substitute for RUNNING_ON_VALGRIND from valgrind.h. */
> - const char *ld = getenv("LD_PRELOAD");
> - running_on_valgrind = (ld != NULL && strstr(ld, "vgpreload"));
> + if (ptr == MAP_FAILED) {
> + fprintf(stderr, "Failed to allocate %zu B: %s\n",
> + size, strerror(errno));
> + abort();
> }
> -#endif
>
> - if (size < align || running_on_valgrind) {
> - align = getpagesize();
> + ptr += offset;
> + total -= offset;
> +
> + if (offset > 0) {
> + munmap(ptr - offset, offset);
> }
> - ptr = qemu_memalign(align, size);
Hrm, so we switch from qemu_memalign to mmap() but then we don't modify
qemu_vfree() to do a munmap() over free().
qemu_vfree() doesn't know the size so calling munmap() is tricky.
Regards,
Anthony Liguori
> + if (total > size) {
> + munmap(ptr + size, total - size);
> + }
> +
> trace_qemu_vmalloc(size, ptr);
> return ptr;
> }
> --
> 1.8.1.4
