Re: reduce size of pv_pte

[Lars Heidieker <lars%heidieker.de@localhost>]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hi,

On 05/13/11 04:11, YAMAMOTO Takashi wrote:
> hi,
> 
>> -----BEGIN PGP SIGNED MESSAGE-----
>> Hash: SHA1
>>
>> On 04/26/11 00:40, YAMAMOTO Takashi wrote:
>>> hi,
>>>
>>> the following patch reduces the size of pv_pte, thus pv_entry and vm_page.
>>> comments?
>>>
>>> YAMAMOTO Takashi
>>>
>>
>> Hi, that's interesting. It is cutting a pv_entry from 40 bytes down to
>> 32 bytes for 64 bit.
>> I've a concern about the runtime requirements because of the
>> PHYS_TO_VM_PAGE lookup which might be significant.
> 
> if it's significant, it should be improved. :-)
> 

yes obviously ;-) It's just something to keep an eye on. There is
already a alternative physseg_find method utilizing a binary search...

>> Do all amd64 machines have only a few physical memory segments?
> 
> i don't think so.
> linux folks put amazing efforts to speed up their PHYS_TO_VM_PAGE
> equivalent for their big irons.
> 

That's the point were different search strategies might become necessary.

> YAMAMOTO Takashi
> 
>>
>> Lars
> 

As you stated that only P -> V operations need pvkey_decode the
performance hit should be very low if measurable at all...
And that's what I found, at least on my amd64 quad-core with 8gig, the
overhead seems to be totally in the noise.
I'm fine with that and reducing the pv_pte size is definitely a win.

Lars
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.17 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/

iEYEARECAAYFAk3QOLoACgkQcxuYqjT7GRbnrgCgsnZlv1QtxVThvpuQACAoGWaD
92cAniKVszUVB46QcIl7wTpjYlOM8/Xh
=52Uv
-----END PGP SIGNATURE-----