Re: [PATCH] arm: remove unused code in delay.S

[Felipe Contreras]

On Mon, Sep 14, 2009 at 6:21 PM, Russell King - ARM Linux
<linux@xxxxxxxxxxxxxxxx> wrote:
> On Mon, Sep 14, 2009 at 06:14:08PM +0300, Felipe Contreras wrote:
>> On Mon, Sep 14, 2009 at 5:40 PM, Russell King - ARM Linux
>> <linux@xxxxxxxxxxxxxxxx> wrote:
>> > On Mon, Sep 14, 2009 at 05:38:32PM +0300, Felipe Contreras wrote:
>> >> On Mon, Sep 14, 2009 at 5:00 PM, Russell King - ARM Linux
>> >> <linux@xxxxxxxxxxxxxxxx> wrote:
>> >> > On Mon, Sep 14, 2009 at 03:58:24PM +0300, Felipe Contreras wrote:
>> >> >> On Mon, Sep 14, 2009 at 11:10 AM, Russell King - ARM Linux
>> >> >> <linux@xxxxxxxxxxxxxxxx> wrote:
>> >> >> > On Mon, Sep 14, 2009 at 01:21:00AM +0100, Jamie Lokier wrote:
>> >> >> >> Russell King - ARM Linux wrote:
>> >> >> >> > On Sun, Sep 13, 2009 at 11:28:47PM +0200, Marek Vasut wrote:
>> >> >> >> > > >         bhi   __delay
>> >> >> >> > > >         mov   pc, lr
>> >> >> >> > > > ÂENDPROC(__udelay)
>> >> >> >> > > >
>> >> >> >> > > Hi
>> >> >> >> > >
>> >> >> >> > > why was this code there in the first place ?
>> >> >> >> >
>> >> >> >> > To make the delay loop more stable and predictable on older CPUs.
>> >> >> >>
>> >> >> >> So why has it been commented out, if it's needed for that?
>> >> >> >
>> >> >> > We moved on and it penalises later CPUs, leading to udelay providing
>> >> >> > shorter delays than requested.
>> >> >> >
>> >> >> > So the choice was either stable and predictable on older CPUs but
>> >> >> > buggy on newer CPUs, or correct on all CPUs but gives unnecessarily
>> >> >> > longer delays on older CPUs.
>> >> >>
>> >> >> Why not add an #ifdef CPU_V4 or whatever?
>> >> >
>> >> > Because then you get it whenever you configure for V4 as the lowest
>> >> > denominator CPU, which leads to the buggy behaviour on better CPUs.
>> >> > It's far better to leave it as is and just accept that the old CPUs
>> >> > will have longer than necessary delays. ÂIf people really really
>> >> > care (and there's likely to only be a small minority of them now)
>> >> > changing the '0' to a '1' is a very simple change for them to carry
>> >> > in their local tree. ÂUnlike getting the right unrolling etc.
>> >>
>> >> Well, they can also 'git revert' this patch. If somebody really cares
>> >> I think they should shout now and provide a better patch, otherwise
>> >> this one should be merged.
>> >
>> > On the other hand, having the code there as it currently stands is not
>> > harmful in any way, so leaving it there is just as easy.
>>
>> It makes the code less understandable. I'm not sure about linux's
>> practices, but an #if 0 generally means somebody is being lazy.
>
> I would agree with you if it was a complicated bit of code, but it
> isn't. ÂIt is a simple count to zero (or overflow) and terminate
> loop. ÂAnd it's certainly not about me being lazy.

Maybe it's not complicated to you, but not everyone is so literate
about ARM assembly code (e.g. me). When I first looked at the code I
didn't even realize there was an #if 0 there, which yes, I grant is a
problem of my editor, but the issue wouldn't have happened if the code
wasn't there in the first place.

And I'm not saying your are being lazy, if anything it's probably the
people using this code. They should figure a way to avoid patching the
code. However, these users are hypothetical at this point.

> Unless there is a strong argument for removing it, the code stays as
> is.
>
> So far, the argument is basically "it's a #if 0, we must get rid of
> it" which is a religous argument, not a technical one. ÂThe fact is
> that (as I said above) keeping it there provides the code for when
> people want to enable it. ÂThat's a technical reason for keeping it.
>
> Please can we now move to something more productive instead of this
> religous argument?

I don't think a good argument has been stated, I'll try to do that on
Wolfram's reply.

-- 
Felipe Contreras
--
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/