Re: VFS 64-bit clean? Not yet (was Re: large-file-system)

Linus Torvalds (torvalds@transmeta.com)
16 Feb 1998 20:25:51 GMT


In article <m0y4OTP-0005FsC@lightning.swansea.linux.org.uk>,
Alan Cox <alan@lxorguk.ukuu.org.uk> wrote:
>> > Make it 64-bit. Things like NFS-V3 require it, and there's no point
>> > in reworking a lot of the VFS code if we're still going to impose
>> > artificial limits in it.
>>
>> NFS V3 can't require anything of our page structure. But your point
>> is well-made.
>
>It can. NFS V3 is 64bit and needs to put 64bit values in the page
>cache to get full results. It also has a write/commit based operation
>which needs the page cache too

No it does not.

32 bits is plenty, if you use a scaling factor. And gcc is known for
generating very bad code for multi-register "long long" (ie any
non-64-bit architecture) - especially on the x86 where the register
pressure means that you get truly bad code.

Alan, I don't see why you are hung up about getting the full 64 bits
when there are _known_ problems with it, and the 32+scale approach is
for all intents equivalent and avoids all the problems (and allows
74-bit filesystems on 64-bit architectures - you should like that).

Sure, the 64-bit version may _look_ simpler to do, but if you actually
try to optimize it for performance you'll wish you hadn't gone to "long
long". And the page cache is timing critical.

Linus

Linus

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu