Re: 2.6.29-rc libata sff 32bit PIO regression

[Sergei Shtylyov]

Hello.

Hugh Dickins wrote:

> > > (though much more verbose: please simplify if you see a better way).
> > >       
> >   How about the following?
> >
> > 		unsigned char *tail = buf + buflen - slop;
> > 		unsigned char pad[4];
> >
> > 		if (rw == READ) {
> > 			if (slop <= 2)
> > 				ioread16_rep(data_addr, pad, 1);
> > 			else
> > 				ioread32_rep(data_addr, pad, 1);
> > 					memcpy(tail, pad, slop);
> >     
>
> Too many tabs on the memcpy.
>   

  Hey, this is not a patch, and I was using Thunderbird's msg editor -- 
which isn;t really good to tabs. :-)

> > 		} else {
> > 			memcpy(pad, tail, slop);
> > 			memset(pad + slop, 0, 4 - slop);
> >     
>
> And we could make that line even more complicated!
>   

  We could use memzero() but memset() should boil down to it anyway.

> But I think unsigned char pad[4] = {0, 0, 0, 0} would be better.
>   

  Not really, we don't need to waste time initiazlizing it on reads -- 
I hope you understand that it will require real code to write all those 
zeros?). Besides, only {0} should be enough as other entries should be 
implitly zeroed).

> Though Alan didn't have it initialized at all: I don't know if
> that was oversight or superior knowledge.  In Alan's case, one
> should usually assume the latter.

  These bytes can be anything actually as a device should just ignore them.

> > 		if (slop <= 2)
> > 				iowrite16_rep(data_addr, pad, 1);
> > 			else
> > 				iowrite32_rep(data_addr, pad, 1);
> > 		}
> >     
>
> Well, I don't know.

  I do. :-)

> I felt really pleased with using ioread16_rep
> and the char array in my original patch, where slop might be 1 or 2
> or 3; but once it comes down to always one single PIO op, I felt
> it too lazy to be using the _rep form.
>   

  It should do the Right Thing WRT the byte reordering (which is a lack 
thereof ;-) while your code had to muck with it explicitly. And of 
course it's shorter -- because of that.

> I really don't care, whatever works and best satisfies Alan.
>   

 I thought we should care about general user satisfaction, not just 
Alan's... :-)

> > > -	return words << 2;
> > > +
> > > +	return buflen + (buflen & 1);
> > >   
> > >       
> >    return (buflen + 1) & ~1;
> >
> >   Well, I guess I could just have posted my own patch... :-)
> >     
>
> Yes, do go ahead, I'm not desperate to get my name in there!
>   

  I'm not actually very enthusiastic in getting blamed for the 
breakage, given the Alan's example. ;-)

> Hugh
>   

MBR, Sergei


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/