When qemu fork's (ssh migration, qemu-nbd, slirp), the guest memory
mapping becomes shared and write-protected by parent and child, until
execve switches to a new mm.
Since kernel commit 7885a4fbebcbae1f8594445a46aefb9d97353594,
get_user_pages with force=1 parameter will break COW during this window,
leaving stale shadow mappings that point to the previously shared page.
Fix this by madvising the range as MADV_DONTFORK, if mmu notifiers are
disabled.
Signed-off-by: Marcelo Tosatti <[EMAIL PROTECTED]>
Index: kvm-userspace.tip/qemu/qemu-kvm.c
===================================================================
--- kvm-userspace.tip.orig/qemu/qemu-kvm.c
+++ kvm-userspace.tip/qemu/qemu-kvm.c
@@ -25,6 +25,7 @@ int kvm_pit = 1;
#include <pthread.h>
#include <sys/utsname.h>
#include <sys/syscall.h>
+#include <sys/mman.h>
#define bool _Bool
#define false 0
@@ -812,6 +813,19 @@ void kvm_cpu_register_physical_memory(ta
}
}
+int kvm_setup_guest_memory(void *area, unsigned long size)
+{
+ int ret = 0;
+
+ if (kvm_enabled() && !kvm_has_sync_mmu(kvm_context))
+ ret = madvise(area, size, MADV_DONTFORK);
+
+ if (ret)
+ perror ("madvise");
+
+ return ret;
+}
+
int kvm_qemu_check_extension(int ext)
{
return kvm_check_extension(kvm_context, ext);
Index: kvm-userspace.tip/qemu/qemu-kvm.h
===================================================================
--- kvm-userspace.tip.orig/qemu/qemu-kvm.h
+++ kvm-userspace.tip/qemu/qemu-kvm.h
@@ -45,6 +45,7 @@ void *kvm_cpu_create_phys_mem(target_phy
void kvm_cpu_destroy_phys_mem(target_phys_addr_t start_addr,
unsigned long size);
+int kvm_setup_guest_memory(void *area, unsigned long size);
int kvm_arch_qemu_create_context(void);
Index: kvm-userspace.tip/qemu/vl.c
===================================================================
--- kvm-userspace.tip.orig/qemu/vl.c
+++ kvm-userspace.tip/qemu/vl.c
@@ -8890,7 +8890,7 @@ static int gethugepagesize(void)
return hugepagesize;
}
-void *alloc_mem_area(unsigned long memory, const char *path)
+void *alloc_mem_area(size_t memory, unsigned long *len, const char *path)
{
char *filename;
void *area;
@@ -8929,18 +8929,23 @@ void *alloc_mem_area(unsigned long memor
return NULL;
}
+ *len = memory;
return area;
}
void *qemu_alloc_physram(unsigned long memory)
{
void *area = NULL;
+ unsigned long map_len = memory;
if (mem_path)
- area = alloc_mem_area(memory, mem_path);
+ area = alloc_mem_area(memory, &map_len, mem_path);
if (!area)
area = qemu_vmalloc(memory);
-
+#ifdef USE_KVM
+ if (kvm_setup_guest_memory(area, map_len))
+ area = NULL;
+#endif
return area;
}
--
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to [EMAIL PROTECTED]
More majordomo info at http://vger.kernel.org/majordomo-info.html
