Re: [PATCH] ARM: tegra: fix erroneous address in dts

From: Thierry Reding
Date: Mon Jul 18 2016 - 02:01:39 EST

On Fri, Jul 15, 2016 at 06:18:03PM +0200, Ralf Ramsauer wrote:
> On 07/15/2016 06:01 PM, Stephen Warren wrote:
> > On 07/15/2016 03:37 AM, Ralf Ramsauer wrote:
> >> On 07/15/2016 12:02 AM, Thierry Reding wrote:
> >>> On Thu, Jul 14, 2016 at 06:48:57PM +0200, Ralf Ramsauer wrote:
> >>>> c90bb7b enabled the high speed UARTs of the Jetson TK1. The address
> >>>> specification inside the dts is wrong. Fix it and use the correct
> >>>> address.
> >>>>
> >>>> Fixes: c90bb7b9b9 ("ARM: tegra: Add high speed UARTs to Jetson TK1
> >>>> device tree")
> >>>> Signed-off-by: Ralf Ramsauer <ralf@xxxxxxxxxxxxxxxxxxxxxx>
> >>>> ---
> >>>> arch/arm/boot/dts/tegra124-jetson-tk1.dts | 4 ++--
> >>>> 1 file changed, 2 insertions(+), 2 deletions(-)
> >>>
> >>> These addresses are correct. The 0, prefix was dropped from the unit
> >>> address in commit b5896f67ab3c ("ARM: tegra: Remove commas from unit
> >>> addresses on Tegra124").
> >>>
> >>> What's the problem that you're seeing? What's not working for you?
> >>
> >> I cannot find b5896f67ab3c neither in swarren's tree nor in linux
> >> upstream. But there's d0bc5aaf890 in swarren's linux-tegra tree that
> >> matches your described changes and was committed on 1st of July. But
> >> this patch is not upstream yet, while the other patch is.
> >
> > The fix is in linux-next, and will be in mainline soon I expect.
> Ah okay, but I still wonder how my original patch got changed on its
> way... The original version on the mailinglist was not buggy.

It's most likely my fault. My recollection is that we used to have an
earlier patch that removed all 0, prefixes from unit addresses and what
happens is probably that I applied your patch on top, adjusted it to
omit the 0, prefix and then we noticed that removing the 0, prefix does
not work for the GPU (because U-Boot looks it up by node name to enable
it), so the patch that removes the 0, prefix was dropped and I forgot
to adjust your patch to compensate.

> > My github linux-tegra.git isn't relevant since it's just my own personal
> > dev branch, but for reference, the commit is there since it's based on
> > linux-next.
> >
> >> Have a look at mainline tegra124-jetson-tk1.dts, there the addresses are
> >> erroneous as they still use the 0, annotation. And I just realised, that
> >> somehow, upstream patch c90bb7b slightly differs from my initial patch
> >> [1] on the mailing list.
> >
> > Your patch should probably be CC: stable so that existing kernel
> > versions get fixed. That said, I'd argue that since the original patch
> > never actually had any effect since it applied to the wrong node, the
> > best fix for stable kernels is simply to revert it rather than fixing
> > it, to avoid the potential for behaviour changes and regressions.
> > Starting with kernel 4.8 (I think), that patch will begin to have actual
> > effect.
> There is no current existing stable kernel that is affected, as it went
> in during the last merge window in 4.6-rc1. So no need for fixing
> stable. Maybe it's still possible to fix it as the stabilisation window
> is still open and 4.7 is not released yet?

Unfortunately a patch to remove the 0, prefix from all nodes except the
GPU was now merged into the arm-soc tree, which means that if we try to
get this into v4.7, then we'll have to make sure to patch up the unit
address again for v4.8.

I have no objections to doing that, so feel free to add my Acked-by and
resend this to arm@xxxxxxxxxxx When you do, please keep linux-tegra in
Cc so that we can track it, and maybe mention that you want this to go
into v4.7 and how this came about to be, so that the arm-soc maintainers
won't be surprised when the patch to revert this shows up in a couple of
days for v4.8.

Can you take care of sending out the second patch as well? Probably best
to send it the same way as this, with a mention that it's targetted at
v4.8 and a revert of this one.


Attachment: signature.asc
Description: PGP signature