Re: [PATCH 3/4] exec, memory: Call to xen_modified_memory.

[Paolo Bonzini <pbonzini@redhat.com>]

Il 17/07/2012 20:06, Stefano Stabellini ha scritto:
> On Tue, 17 Jul 2012, Anthony PERARD wrote:
>> This patch add some calls to xen_modified_memory to notify Xen about dirtybits
>> during migration.
>>
>> Signed-off-by: Anthony PERARD <anthony.perard@citrix.com>
>> ---
>>  exec.c   |    4 ++++
>>  memory.c |    2 ++
>>  2 files changed, 6 insertions(+), 0 deletions(-)
>>
>> diff --git a/exec.c b/exec.c
>> index c9fa17d..9f7a4f7 100644
>> --- a/exec.c
>> +++ b/exec.c
>> @@ -3438,6 +3438,7 @@ void cpu_physical_memory_rw(target_phys_addr_t addr, uint8_t *buf,
>>                      cpu_physical_memory_set_dirty_flags(
>>                          addr1, (0xff & ~CODE_DIRTY_FLAG));
>>                  }
>> +                xen_modified_memory(addr1, TARGET_PAGE_SIZE);
>>                  qemu_put_ram_ptr(ptr);
>>              }
>>          } else {
>> @@ -3623,6 +3624,7 @@ void cpu_physical_memory_unmap(void *buffer, target_phys_addr_t len,
>>      if (buffer != bounce.buffer) {
>>          if (is_write) {
>>              ram_addr_t addr1 = qemu_ram_addr_from_host_nofail(buffer);
>> +            xen_modified_memory(addr1, access_len);
>>              while (access_len) {
>>                  unsigned l;
>>                  l = TARGET_PAGE_SIZE;
> 
> You need to add xen_modified_memory in cpu_physical_memory_map, rather
> than cpu_physical_memory_unmap.

No, adding it to map is wrong, because the RAM save routine can migrate
(and mark as non-dirty) the read buffers _before_ the device models have
written to them.

Paolo