Re: [PATCH 1/2] accel: kvm: Fix memory waste under mismatch page size

[Keqian Zhu]

Hi Peter,

On 2020/12/16 1:57, Peter Xu wrote:
> On Tue, Dec 15, 2020 at 03:19:47PM +0800, Keqian Zhu wrote:
>> When handle dirty log, we face qemu_real_host_page_size and
>> TARGET_PAGE_SIZE. The first one is the granule of KVM dirty
>> bitmap, and the second one is the granule of QEMU dirty bitmap.
>>
>> Generally speaking, qemu_real_host_page_size >= TARGET_PAGE_SIZE,
>> so misuse TARGET_PAGE_SIZE to init kvmslot dirty_bmap may waste
>> memory. For example, when qemu_real_host_page_size is 64K and
>> TARGET_PAGE_SIZE is 4K, this bugfix can save 93.75% memory.
>>
>> Signed-off-by: Keqian Zhu <zhukeqian1@huawei.com>
>> ---
>>  accel/kvm/kvm-all.c | 6 +++++-
>>  1 file changed, 5 insertions(+), 1 deletion(-)
>>
>> diff --git a/accel/kvm/kvm-all.c b/accel/kvm/kvm-all.c
>> index baaa54249d..c5e06288eb 100644
>> --- a/accel/kvm/kvm-all.c
>> +++ b/accel/kvm/kvm-all.c
>> @@ -620,8 +620,12 @@ static void kvm_memslot_init_dirty_bitmap(KVMSlot *mem)
>>       * too, in most cases).
>>       * So for now, let's align to 64 instead of HOST_LONG_BITS here, in
>>       * a hope that sizeof(long) won't become >8 any time soon.
>> +     *
>> +     * Note: the granule of kvm dirty log is qemu_real_host_page_size.
>> +     * And mem->memory_size is aligned to it (otherwise this mem can't
>> +     * be registered to KVM).
>>       */
>> -    hwaddr bitmap_size = ALIGN(((mem->memory_size) >> TARGET_PAGE_BITS),
>> +    hwaddr bitmap_size = ALIGN(mem->memory_size / qemu_real_host_page_size,
>>                                          /*HOST_LONG_BITS*/ 64) / 8;
> 
> Yes I think this is correct.  Thanks.
> 
> So one thing I failed to notice is cpu_physical_memory_set_dirty_lebitmap()
> will "amplify" the host dirty pages into guest ones; seems we're all good 
> then.
> 
> Reviewed-by: Peter Xu <peterx@redhat.com>
> 
OK Thanks :-)

One more thing, we should consider whether @start and @size is psize aligned 
(my second
patch). Do you agree to add assert on them directly?


Thanks,
Keqian