Re: [PATCH RFC 5/5] KVM: This is the main part of the "moving dirty bitmaps to user space"

[Avi Kivity <avi@redhat.com>]

On 04/12/2010 05:29 AM, Takuya Yoshikawa wrote:
>
>>
>>> TODO:
>>> 1. We want to use copy_in_user() for 32bit case too.
>>
>> Definitely. Why doesn't it work now?
>
> Sadly we don't have that for 32bit. We have to implement by ourselves.
>
> I tested two temporary implementations for 32bit:
>   1. This version using copy_from_user() and copy_to_user() with
>      not nice vmalloc().
>   2. Loop with __get_user() and __put_user().
>
> The result was 1 is much faster than 2.

What about copy_from_user()/copy_to_user() through a 512 byte buffer on 
the kernel stack?  That should give the speed of 1 without the vmalloc() 
(which will fail on 32 bit if you copy large blocks).

>
>>
>>> Note that this is only for the compatibility issue: in the future,
>>> we hope, qemu will not need to use this ioctl.
>>> 2. We have to implement test_bit_user() to avoid extra set_bit.
>>
>> This was important in the days of shadow paging. I'm not so sure about
>> it with nested paging, since we'll typically only fault a page once per
>> iteration. Since we're very likely to actually write these days, the
>> extra access is wasteful.
>
> Nice news for me! So all we need to ask x86(asm-generic) people to 
> merge are:
>
> set bit user and copy_in_user 32bit version.

They might still want test_bit_user and clear_bit_user for completeness :)

-- 
I have a truly marvellous patch that fixes the bug which this
signature is too narrow to contain.