Re: [Qemu-devel] vga optmization

[Stefano Stabellini <stefano.stabellini@eu.citrix.com>]

andrzej zaborowski wrote:

> Hi,
> 
> 2008/11/3 Glauber Costa <glommer@redhat.com>:
> [...]
>> diff --git a/cpu-all.h b/cpu-all.h
>> index cdd79bc..9118f4d 100644
>> --- a/cpu-all.h
>> +++ b/cpu-all.h
>> @@ -46,6 +46,8 @@
>>
>>  #ifdef BSWAP_NEEDED
>>
>> +#include "kvm.h"
>> +
>>  static inline uint16_t tswap16(uint16_t s)
>>  {
>>     return bswap16(s);
>> @@ -909,17 +911,10 @@ int cpu_memory_rw_debug(CPUState *env, target_ulong addr,
>>  #define KQEMU_DIRTY_FLAG     0x04
>>  #define MIGRATION_DIRTY_FLAG 0x08
>>
>> -/* read dirty bit (return 0 or 1) */
>> -static inline int cpu_physical_memory_is_dirty(ram_addr_t addr)
>> -{
>> -    return phys_ram_dirty[addr >> TARGET_PAGE_BITS] == 0xff;
>> -}
>> +int cpu_physical_memory_get_dirty(ram_addr_t addr, int dirty_flags);
>> +int cpu_physical_memory_is_dirty(ram_addr_t addr);
>>
>> -static inline int cpu_physical_memory_get_dirty(ram_addr_t addr,
>> -                                                int dirty_flags)
>> -{
>> -    return phys_ram_dirty[addr >> TARGET_PAGE_BITS] & dirty_flags;
>> -}
>> +void qemu_physical_sync_dirty_bitmap(ram_addr_t start_addr);
> 
> This will prevent the functions from being inlined even if KVM is
> disabled (e.g. on other archs) and I think it could be easily
> retained.
> 


I agree on this.

> void qemu_physical_sync_dirty_bitmap(ram_addr_t start_addr)
> {
>     if (kvm_enabled())
>         kvm_physical_sync_dirty_bitmap(start_addr);
> }
> 


Why don't you make qemu_physical_sync_dirty_bitmap take also and end
address, and you merge the two bitmaps in this address range in this
function, so you don't have to change cpu_physical_memory_get_dirty at
all?

I am saying to do something like:

void qemu_physical_sync_dirty_bitmap(ram_addr_t start_addr, ram_addr_t end_addr)
{
	/* sync and merge the two bitmaps between start_addr and end_addr */
}

then leave cpu_physical_memory_get_dirty untouched.
I would prefer this approch, it also leaves more space to other
optimizations.