glibc-headers troubles with rawhide on ARM

[Olivier Fourdan]

Hi,

Last Friday, I tried to update the xorg-x11-server package, everything
went well on F30 but the compilation failed on ARM because of a
missing header `sys/io.h`:

https://kojipkgs.fedoraproject.org//work/tasks/2196/35892196/build.log

```
BUILDSTDERR: ../../../hw/xfree86/common/compiler.h:767:10: fatal
error: sys/io.h: No such file or directory
```

This affects only ARM, all other supported architectures built fine.
Also, the same package built fine a couple of weeks ago.

Is that a known issue with glibc in rawhide? A quick search in
bugzilla doesn't reveal anything relevant.

Cheers,
Olivier

[Olivier Fourdan]

Hi,

On Mon, Jul 1, 2019 at 11:12 AM Olivier Fourdan <ofourdan(a)redhat.com&gt; wrote:
>  Hi,
>
>  Last Friday, I tried to update the xorg-x11-server package, everything
>  went well on F30 but the compilation failed on ARM because of a
>  missing header `sys/io.h`:
>
>  https://kojipkgs.fedoraproject.org//work/tasks/2196/35892196/build.log
>
>  ```
>  BUILDSTDERR: ../../../hw/xfree86/common/compiler.h:767:10: fatal
>  error: sys/io.h: No such file or directory
>  ```
>
>  This affects only ARM, all other supported architectures built fine.
>  Also, the same package built fine a couple of weeks ago.
>
>  Is that a known issue with glibc in rawhide? A quick search in
>  bugzilla doesn't reveal anything relevant. 
Quick update, I found this, but it's 5 years old and closed:

https://bugzilla.redhat.com/show_bug.cgi?id=1116162

Cheers
Olivier

[Florian Weimer]

* Olivier Fourdan:

>  Last Friday, I tried to update the xorg-x11-server package,
> everything
>  went well on F30 but the compilation failed on ARM because of a
>  missing header `sys/io.h`:
>
>  https://kojipkgs.fedoraproject.org//work/tasks/2196/35892196/build.log
>
>  ```
>  BUILDSTDERR: ../../../hw/xfree86/common/compiler.h:767:10: fatal
>  error: sys/io.h: No such file or directory
>  ```
>
>  This affects only ARM, all other supported architectures built fine.
>  Also, the same package built fine a couple of weeks ago.
>
>  Is that a known issue with glibc in rawhide? A quick search in
>  bugzilla doesn't reveal anything relevant. 
<sys/io.h> on Arm hasn't worked for a long, long time, so we've removed
it from glibc.  Sorry it broke the build.

Thanks,
Florian

[Peter Robinson]

On Mon, Jul 1, 2019 at 12:02 PM Florian Weimer <fweimer(a)redhat.com&gt; wrote:
>  * Olivier Fourdan:
>
>  > Last Friday, I tried to update the xorg-x11-server package, everything
>  > went well on F30 but the compilation failed on ARM because of a
>  > missing header `sys/io.h`:
>  >
>  > https://kojipkgs.fedoraproject.org//work/tasks/2196/35892196/build.log
>  >
>  > ```
>  > BUILDSTDERR: ../../../hw/xfree86/common/compiler.h:767:10: fatal
>  > error: sys/io.h: No such file or directory
>  > ```
>  >
>  > This affects only ARM, all other supported architectures built fine.
>  > Also, the same package built fine a couple of weeks ago.
>  >
>  > Is that a known issue with glibc in rawhide? A quick search in
>  > bugzilla doesn't reveal anything relevant.
>
>  <sys/io.h> on Arm hasn't worked for a long, long time, so we've removed
>  it from glibc.  Sorry it broke the build. 
It seems xorg-x11-server was also using this in it's builds:

https://koji.fedoraproject.org/koji/taskinfo?taskID=35892190

[Florian Weimer]

* Peter Robinson:

>  On Mon, Jul 1, 2019 at 12:02 PM Florian Weimer
> <fweimer(a)redhat.com&gt; wrote:
> >
> > * Olivier Fourdan:
> >
> > > Last Friday, I tried to update the xorg-x11-server package, everything
> > > went well on F30 but the compilation failed on ARM because of a
> > > missing header `sys/io.h`:
> > >
> > > https://kojipkgs.fedoraproject.org//work/tasks/2196/35892196/build.log
> > >
> > > ```
> > > BUILDSTDERR: ../../../hw/xfree86/common/compiler.h:767:10: fatal
> > > error: sys/io.h: No such file or directory
> > > ```
> > >
> > > This affects only ARM, all other supported architectures built fine.
> > > Also, the same package built fine a couple of weeks ago.
> > >
> > > Is that a known issue with glibc in rawhide? A quick search in
> > > bugzilla doesn't reveal anything relevant.
> >
> > <sys/io.h> on Arm hasn't worked for a long, long time, so we've
> removed
> > it from glibc.  Sorry it broke the build.
>
>  It seems xorg-x11-server was also using this in it's builds:
>
>  https://koji.fedoraproject.org/koji/taskinfo?taskID=35892190 
Peter and Olivier, do you need assistance with this?

Thanks,
Florian

[Peter Robinson]

>  > On Mon, Jul 1, 2019 at 12:02 PM Florian Weimer
> <fweimer(a)redhat.com&gt; wrote:
>  >>
>  >> * Olivier Fourdan:
>  >>
>  >> > Last Friday, I tried to update the xorg-x11-server package, everything
>  >> > went well on F30 but the compilation failed on ARM because of a
>  >> > missing header `sys/io.h`:
>  >> >
>  >> > https://kojipkgs.fedoraproject.org//work/tasks/2196/35892196/build.log
>  >> >
>  >> > ```
>  >> > BUILDSTDERR: ../../../hw/xfree86/common/compiler.h:767:10: fatal
>  >> > error: sys/io.h: No such file or directory
>  >> > ```
>  >> >
>  >> > This affects only ARM, all other supported architectures built fine.
>  >> > Also, the same package built fine a couple of weeks ago.
>  >> >
>  >> > Is that a known issue with glibc in rawhide? A quick search in
>  >> > bugzilla doesn't reveal anything relevant.
>  >>
>  >> <sys/io.h> on Arm hasn't worked for a long, long time, so we've
> removed
>  >> it from glibc.  Sorry it broke the build.
>  >
>  > It seems xorg-x11-server was also using this in it's builds:
>  >
>  > https://koji.fedoraproject.org/koji/taskinfo?taskID=35892190
>
>  Peter and Olivier, do you need assistance with this? 
I'm not the xorg-x11-server maintainer but I came across it when
trying to fix another package that depends on it, so I'm not sure
whether the owner of xorg-x11-server needs assistance.

[Olivier Fourdan]

Hi,

On Tue, Jul 2, 2019 at 2:22 PM Peter Robinson <pbrobinson(a)gmail.com&gt; wrote:
>  On Mon, Jul 1, 2019 at 12:02 PM Florian Weimer
> <fweimer(a)redhat.com&gt; wrote:
>  > [...]
>  >
>  > <sys/io.h> on Arm hasn't worked for a long, long time, so we've
> removed
>  > it from glibc.  Sorry it broke the build.
>
>  It seems xorg-x11-server was also using this in it's builds:
>
>  https://koji.fedoraproject.org/koji/taskinfo?taskID=35892190 
Precisely, that's why I raised the issue here :)

FWIW, I managed to get the Xserver to build by removing the `#include
<sys/io.h>` and references to outb/outw/outl :

http://koji.fedoraproject.org/koji/taskinfo?taskID=35981723

But I don;t think it'll fly with the Xorg drivers, as those are most
likely the consumers of outb/outw/outl.

So I filed https://gitlab.freedesktop.org/xorg/xserver/issues/840
upstream to gauge the water and see what we can do.

Cheers,
Olivier

[David Airlie]

On Tue, Jul 2, 2019 at 10:57 PM Olivier Fourdan <ofourdan(a)redhat.com&gt; wrote:
>  Hi,
>
>  On Tue, Jul 2, 2019 at 2:22 PM Peter Robinson <pbrobinson(a)gmail.com&gt; wrote:
>  > On Mon, Jul 1, 2019 at 12:02 PM Florian Weimer <fweimer(a)redhat.com&gt; wrote:
>  > > [...]
>  > >
>  > > <sys/io.h> on Arm hasn't worked for a long, long time, so we've
> removed
>  > > it from glibc.  Sorry it broke the build.
>  >
>  > It seems xorg-x11-server was also using this in it's builds:
>  >
>  > https://koji.fedoraproject.org/koji/taskinfo?taskID=35892190
>
>  Precisely, that's why I raised the issue here :)
>
>  FWIW, I managed to get the Xserver to build by removing the `#include
>  <sys/io.h>` and references to outb/outw/outl :
>
>  http://koji.fedoraproject.org/koji/taskinfo?taskID=35981723
>
>  But I don;t think it'll fly with the Xorg drivers, as those are most
>  likely the consumers of outb/outw/outl.
>
>  So I filed https://gitlab.freedesktop.org/xorg/xserver/issues/840
>  upstream to gauge the water and see what we can do. 
Any driver that does port io. should probably not be built on ARM,
unless the port io can be disabled anyways.

Dave.

[Olivier Fourdan]

Hi Dave,

On Wed, Jul 3, 2019 at 4:01 AM David Airlie <airlied(a)redhat.com&gt; wrote:
>  On Tue, Jul 2, 2019 at 10:57 PM Olivier Fourdan
> <ofourdan(a)redhat.com&gt; wrote:
>  > [...]
>  >
>  > FWIW, I managed to get the Xserver to build by removing the `#include
>  > <sys/io.h>` and references to outb/outw/outl :
>  >
>  > http://koji.fedoraproject.org/koji/taskinfo?taskID=35981723
>  >
>  > But I don;t think it'll fly with the Xorg drivers, as those are most
>  > likely the consumers of outb/outw/outl.
>  >
>  > So I filed https://gitlab.freedesktop.org/xorg/xserver/issues/840
>  > upstream to gauge the water and see what we can do.
>
>  Any driver that does port io. should probably not be built on ARM,
>  unless the port io can be disabled anyways. 
Right, so I posted my removal patch in a MR upstream linked to the
issue I filed.

Meanwhile, I can push the same downstream and run a new build for
rawhide (similar to the scratch build I ran a couple of days ago which
completed successfully), and then let people see if Xorg drivers fail
to build?

Cheers,
Olivier

[Nicolas Chauvet]

Le mer. 3 juil. 2019 à 04:35, David Airlie <airlied(a)redhat.com&gt; a écrit :
>  On Tue, Jul 2, 2019 at 10:57 PM Olivier Fourdan <ofourdan(a)redhat.com&gt; wrote:
>  >
>  > Hi,
>  >
>  > On Tue, Jul 2, 2019 at 2:22 PM Peter Robinson <pbrobinson(a)gmail.com&gt; wrote:
>  > > On Mon, Jul 1, 2019 at 12:02 PM Florian Weimer <fweimer(a)redhat.com&gt;
> wrote:
>  > > > [...]
>  > > >
>  > > > <sys/io.h> on Arm hasn't worked for a long, long time, so
> we've removed
>  > > > it from glibc.  Sorry it broke the build.
>  > >
>  > > It seems xorg-x11-server was also using this in it's builds:
>  > >
>  > > https://koji.fedoraproject.org/koji/taskinfo?taskID=35892190
>  >
>  > Precisely, that's why I raised the issue here :)
>  >
>  > FWIW, I managed to get the Xserver to build by removing the `#include
>  > <sys/io.h>` and references to outb/outw/outl :
>  >
>  > http://koji.fedoraproject.org/koji/taskinfo?taskID=35981723
>  >
>  > But I don;t think it'll fly with the Xorg drivers, as those are most
>  > likely the consumers of outb/outw/outl.
>  >
>  > So I filed https://gitlab.freedesktop.org/xorg/xserver/issues/840
>  > upstream to gauge the water and see what we can do.
>
>  Any driver that does port io. should probably not be built on ARM,
>  unless the port io can be disabled anyways. 
The opentegra DDX driver uses xorg/compiler.h (that you didn't touched
in the freedesktop PR) but doesn't seems to rely on outb/outw/outl, so
I will see if this driver can be built without it.

For the opentegra case, here an info about why it can still be
considered relevant over the modesetting driver
https://bugzilla.redhat.com/show_bug.cgi?id=1606757#c5
Basically, it still relies on "WIP" libdrm until the kernel tegra
driver abi is reworked...

Thx

Thx

--
-

Nicolas (kwizart)

[Peter Robinson]

On Wed, Jul 3, 2019 at 9:44 AM Nicolas Chauvet <kwizart(a)gmail.com&gt; wrote:
>  Le mer. 3 juil. 2019 à 04:35, David Airlie <airlied(a)redhat.com&gt; a écrit :
>  >
>  > On Tue, Jul 2, 2019 at 10:57 PM Olivier Fourdan <ofourdan(a)redhat.com&gt; wrote:
>  > >
>  > > Hi,
>  > >
>  > > On Tue, Jul 2, 2019 at 2:22 PM Peter Robinson <pbrobinson(a)gmail.com&gt;
> wrote:
>  > > > On Mon, Jul 1, 2019 at 12:02 PM Florian Weimer <fweimer(a)redhat.com&gt;
> wrote:
>  > > > > [...]
>  > > > >
>  > > > > <sys/io.h> on Arm hasn't worked for a long, long time, so
> we've removed
>  > > > > it from glibc.  Sorry it broke the build.
>  > > >
>  > > > It seems xorg-x11-server was also using this in it's builds:
>  > > >
>  > > > https://koji.fedoraproject.org/koji/taskinfo?taskID=35892190
>  > >
>  > > Precisely, that's why I raised the issue here :)
>  > >
>  > > FWIW, I managed to get the Xserver to build by removing the `#include
>  > > <sys/io.h>` and references to outb/outw/outl :
>  > >
>  > > http://koji.fedoraproject.org/koji/taskinfo?taskID=35981723
>  > >
>  > > But I don;t think it'll fly with the Xorg drivers, as those are most
>  > > likely the consumers of outb/outw/outl.
>  > >
>  > > So I filed https://gitlab.freedesktop.org/xorg/xserver/issues/840
>  > > upstream to gauge the water and see what we can do.
>  >
>  > Any driver that does port io. should probably not be built on ARM,
>  > unless the port io can be disabled anyways.
>
>  The opentegra DDX driver uses xorg/compiler.h (that you didn't touched
>  in the freedesktop PR) but doesn't seems to rely on outb/outw/outl, so
>  I will see if this driver can be built without it. 
Similarly xorg-x11-drv-omap fails for the same reason, at least currently.

[Olivier Fourdan]

Hi

On Wed, Jul 3, 2019 at 10:45 AM Nicolas Chauvet <kwizart(a)gmail.com&gt; wrote:
>  The opentegra DDX driver uses xorg/compiler.h (that you didn't
> touched
>  in the freedesktop PR) but doesn't seems to rely on outb/outw/outl, so
>  I will see if this driver can be built without it. 
The installed `xorg/compiler.h` *is* `hw/xfree86/common/compiler.h` so
that's precisely the file my patch chnages.

But I obviously do not remove all of `compiler.h` in my patch, I just
remove the arm + glibc specific code path that refers to an include
(`<sys/io.h>`) that was removed from glibc upstream.

>  For the opentegra case, here an info about why it can still be
>  considered relevant over the modesetting driver
>  https://bugzilla.redhat.com/show_bug.cgi?id=1606757#c5
>  Basically, it still relies on "WIP" libdrm until the kernel tegra
>  driver abi is reworked... 
That's a different issue.

Cheers,
Olivier