Re: drivers/block/ub.c

[Matthew Dharm]

On Mon, Jun 28, 2004 at 02:01:57PM -0700, Pete Zaitcev wrote:
> On Mon, 28 Jun 2004 13:50:58 -0700
> Matthew Dharm <mdharm-kernel@xxxxxxxxxxxxxxxxxx> wrote:
> 
> > > 	struct txd {
> > > 		u32 dma_addr;
> > > 		u32 length;
> > > 	};
> > > 
> > > It is just total and utter madness to put a packed or the proposed
> > > __nopadding__ attribute on that structure.  Yet this seems to be
> > > what was suggested now and at the beginning of this thread.
> > 
> > I guess, in the end, what this comes down to is the fact that we're all
> > going to get bitten on the ass when we finally get to a platform where the
> > default alignment is 64-bits, which would then (by default) add padding to
> > the above structure.
> > 
> > How long until that time comes?  Likely within my lifetime, and I'd rather
> > not have to re-write working code into more working code because I couldn't
> > express to the compiler what I needed it to do.
> 
> I, for one, am not engaging into such flights of fancy as a platform
> with larger than natural alignment requirements. Would you even read
> what you're writing? The whole freaking world abandons silly platforms
> and moves to x86 extensions and you're fantasizing about a return
> to Cray-1. It just ain't happening!

Actually, I've been working with a controller for the last several months
(the "latest and greatest", state-of-the-art technology), which can only to
loads/stores in 64-byte units at a minimum.

Yes, I mean "byte".  128-bits wide * 4 DDR beats (2 dobule-edged clocks).
There are no DM pins on the memory interface, so any write less than that
size must be a read-modify-write.

And, I was just having a conversation with a compiler group which was
considering moving to 64-byte alignment as a speed optimization for this
platform, or at least 16-byte (64 bit), which also aligns well with the
native load/store of the CPU involved.

I only wish I was fantasizing about this.

Matt

-- 
Matthew Dharm                              Home: mdharm-usb@xxxxxxxxxxxxxxxxxx 
Maintainer, Linux USB Mass Storage Driver

P:  Nine more messages in admin.policy.
M: I know, I'm typing as fast as I can!
					-- Pitr and Mike
User Friendly, 11/27/97

Attachment: pgp00000.pgp
Description: PGP signature