Received-SPF: pass (google.com: domain of linux-kernel+bounces-214247-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.80.249 as permitted sender) client-ip=147.75.80.249;
Subject: Re: [PATCH] KVM: Remove duplicated zero clear with dirty_bitmap
 buffer
To: Christophe JAILLET <christophe.jaillet@wanadoo.fr>,
 Paolo Bonzini <pbonzini@redhat.com>
Cc: kvm@vger.kernel.org, linux-kernel@vger.kernel.org
References: <20240613125407.1126587-1-maobibo@loongson.cn>
 <115973a9-caa6-4d53-a477-dea2d2291598@wanadoo.fr>
From: maobibo <maobibo@loongson.cn>
Message-ID: <fb2da53e-791d-aef7-4dbb-dcf054675f9b@loongson.cn>
Date: Fri, 14 Jun 2024 10:45:56 +0800
User-Agent: Mozilla/5.0 (X11; Linux loongarch64; rv:68.0) Gecko/20100101
 Thunderbird/68.7.0
Precedence: bulk
MIME-Version: 1.0
In-Reply-To: <115973a9-caa6-4d53-a477-dea2d2291598@wanadoo.fr>
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Language: en-US
Content-Transfer-Encoding: 8bit


On 2024/6/14 上午3:25, Christophe JAILLET wrote:
> Le 13/06/2024 à 14:54, Bibo Mao a écrit :
>> Since dirty_bitmap pointer is allocated with function __vcalloc(),
>> there is __GFP_ZERO flag set in the implementation about this function
>> __vcalloc_noprof(). It is not necessary to clear dirty_bitmap buffer
>> with zero again.
>>
>> Signed-off-by: Bibo Mao <maobibo@loongson.cn>
>> ---
>>   virt/kvm/kvm_main.c | 3 ---
>>   1 file changed, 3 deletions(-)
>>
> 
> Hi,
> 
>> diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
>> index 14841acb8b95..c7d4a041dcfa 100644
>> --- a/virt/kvm/kvm_main.c
>> +++ b/virt/kvm/kvm_main.c
>> @@ -1669,9 +1669,6 @@ static int kvm_prepare_memory_region(struct kvm 
>> *kvm,
>>               r = kvm_alloc_dirty_bitmap(new);
>>               if (r)
>>                   return r;
>> -
>> -            if (kvm_dirty_log_manual_protect_and_init_set(kvm))
>> -                bitmap_set(new->dirty_bitmap, 0, new->npages);
> 
> unless I miss something obvious, this does not clear anything, but set 
> all bits to 1.
> 
> 0 is not for "write 0" (i.e. clear), but for "start at offset 0".
> So all bits are set to 1 in this case.
you are right, I had thought it is to write 0 :(

I do not know whether KVM_DIRTY_LOG_INITIALLY_SET should be enabled on 
LoongArch. If it is set, write protection for second MMU will start one 
by one in function kvm_arch_mmu_enable_log_dirty_pt_masked() when dirty 
log is cleared if it is set, else write protection will start in 
function kvm_arch_commit_memory_region() when flag of memslot is changed.

I do not see the obvious benefits between these two write protect 
stages. Can anyone give me any hints?

Regards
Bibo Mao

> 
> CJ
> 
>>           }
>>       }
>>
>> base-commit: 83a7eefedc9b56fe7bfeff13b6c7356688ffa670