[PATCH 1/2] Revert "include/uapi/drm/amdgpu_drm.h: use __u32 and __u64 from <linux/types.h>"

[Marek Olšák]

On Sat, Aug 20, 2016 at 8:28 PM, Marek Olšák <maraeo at gmail.com> wrote:
> On Sat, Aug 20, 2016 at 8:08 PM, Mikko Rapeli <mikko.rapeli at iki.fi> wrote:
>> Cc'ing lkml.
>>
>> On Sat, Aug 20, 2016 at 12:05:54PM +0200, Marek Olšák wrote:
>>> On Sat, Aug 20, 2016 at 12:54 AM, Emil Velikov <emil.l.velikov at gmail.com> wrote:
>>> > On 19 August 2016 at 15:26, Christian König <deathsimple at vodafone.de> wrote:
>>> >> Am 19.08.2016 um 15:50 schrieb Marek Olšák:
>>> >>>
>>> >>> From: Marek Olšák <marek.olsak at amd.com>
>>> >>>
>>> >>> This reverts commit 2ce9dde0d47f2f94ab25c73a30596a7328bcdf1f.
>>> >>>
>>> >>> See the comment in the code. Basically, don't do cleanups in this header.
>>> >>>
>>> >>> Signed-off-by: Marek Olšák <marek.olsak at amd.com>
>>> >>
>>> >>
>>> >> I completely agree with you that this was a bad move, but I fear that we
>>> >> will run into opposition with that.
>>> >>
>>> > Please check the facts before introducing RATHER ANNOYING AND HARD TO
>>> > READ COMMENT IN ALL CAPS.
>>> >
>>> > Story time:
>>> > I was dreaming of a day were we can stop installing these headers,
>>> > thus making deprecation a bit easier process.
>>> > Yet after failing to convince Dave and Daniel on a number of occasions
>>> > I've accepted that those headers _are_ here to stay. And yes they
>>> > _are_ the UAPI, even though no applications are meant to use them but
>>> > the libdrm 'version'.
>>> > Thus any changes to the libdrm ones should be a mirror of the ones
>>> > here and libdrm should _not_ differ.
>>> >
>>> > But let's ignore all that and imagine that those headers are _not_
>>> > UAPI. That gives us even greater reason to _not_ use the uintx_t types
>>> > but the kernel __uX ones. The series that did these changes had a fair
>>> > few references why we want that.
>>> >
>>> > Yes, I can imagine that the situation isn't ideal, and/or not that
>>> > clear. Then again a check with git log should have straightened things
>>> > out.
>>> > If not _please_ help us improve this (documentation and/or others).
>>> >
>>> >
>>> > And last but not least, please share with up what inspired this -
>>> > (build/runtime) regression, attempted sync with libdrm, other ?
>>>
>>> Syncing with libdrm became difficult. I'd like the diff between kernel
>>> and libdrm to be as small as possible.
>>>
>>> We must take into account that the uapi headers can potentially be
>>> implemented by a different OS. That's why they are in libdrm and
>>> that's why nobody should make random changes to them in the kernel
>>> tree. Do not think like a kernel developer isolated in Linux and just
>>> consider the broader use case. If you do, you'll realize that it
>>> simply doesn't make sense to use the __uX types here.
>>
>> When libdrm is combined with Linux kernel, then libdrm should be using
>> the uapi headers from the kernel. For various reasons there are embedded
>> kernel header copies in libdrm, glibc and basically everywhere but there
>> should not be need for that.
>
> You quoted what I had written but you didn't read it.

Here's why libdrm can't use the uapi headers: It would break the very
common use case and that is building the latest userspace driver stack
on top of an old kernel. The userspace needs the latest headers, not
the installed ones. Only then is it fully compatible both ways.

Marek