Re: [RFC PATCH] drm.h: Fix DRM compilation with bare-metal toolchain.

[Arnd Bergmann <arnd@arndb.de>]

On Tuesday 16 April 2013, Nishanth Menon wrote:
> On 12:50-20130416, Arnd Bergmann wrote:
> > On Tuesday 16 April 2013 12:48:28 Paul Sokolovsky wrote:
> > > > diff --git a/include/uapi/drm/drm.h b/include/uapi/drm/drm.h
> > > > index 8d1e2bb..73a99e4 100644
> > > > --- a/include/uapi/drm/drm.h
> > > > +++ b/include/uapi/drm/drm.h
> > > > @@ -36,7 +36,7 @@
> > > >  #ifndef _DRM_H_
> > > >  #define _DRM_H_
> > > >  
> > > > -#if defined(__linux__)
> > > > +#if defined(__KERNEL__) || defined(__linux__)
> > > >  
> > > >  #include <linux/types.h>
> > > >  #include <asm/ioctl.h>
> > 
> > This is still completely bogus, the __KERNEL__ symbol has no significance here.
> > Either make the compiler define __linux__, or remove this #ifdef completely.
> > 
> Searching the v.39-rc7 tag, and greping for _linux_ a few interesting
> list pops up. (pruned):
> arch/arc/Makefile:cflags-y	+= -mA7 -fno-common -pipe -fno-builtin -D__linux__
> arch/h8300/Makefile:KBUILD_CFLAGS += -D__linux__
> arch/hexagon/Makefile:KBUILD_CFLAGS += -ffixed-$(TIR_NAME) -DTHREADINFO_REG=$(TIR_NAME) -D__linux__
> arch/score/Makefile:	-D__linux__ -ffunction-sections -ffreestanding
> arch/xtensa/Makefile:KBUILD_CFLAGS += -ffreestanding -D__linux__
> ^^ these architectures seem to bypass the pain entirely by defining
> __linux__

Right, that seems like a reasonable approach when the compilers are actually known
to be compatible.

> arch/mips/include/uapi/asm/sgidefs.h:#ifndef __linux__

On MIPS, they are not. If you are building a Linux kernel with a gcc that
was targetted at another ABI, the system call interface may change, so they
forbid that here.

> drivers/gpu/drm/radeon/radeon_cp.c:#ifdef __linux__
> {snip}
> drivers/scsi/aic7xxx/aic7770.c:#ifdef __linux__
> drivers/scsi/aic7xxx/aic79xx.h:#ifndef __linux__
> {snip}
> drivers/scsi/aic7xxx/aic79xx_core.c:#ifdef __linux__
> {snip}
> drivers/scsi/aic7xxx/aic79xx_pci.c:#ifdef __linux__
> drivers/scsi/aic7xxx/aic7xxx.h:#ifndef __linux__
> drivers/scsi/aic7xxx/aic7xxx.h:#ifndef __linux__
> drivers/scsi/aic7xxx/aic7xxx_93cx6.c:#ifdef __linux__
> drivers/scsi/aic7xxx/aic7xxx_core.c:#ifdef __linux__
> {snip}
> drivers/scsi/aic7xxx/aic7xxx_pci.c:#ifdef __linux__
> drivers/scsi/aic7xxx/aicasm/aicasm.h:#ifdef __linux__
> drivers/scsi/aic7xxx/aicasm/aicasm_macro_scan.l:#ifdef __linux__
> drivers/scsi/aic7xxx/aicasm/aicasm_scan.l:#ifdef __linux__
> drivers/scsi/aic7xxx/aicasm/aicasm_symbol.c:#ifdef __linux__
> drivers/scsi/aic7xxx/aicasm/aicasm_symbol.h:#ifdef __linux__
> drivers/scsi/dpt/osd_defs.h:#if (defined(__linux__))
> drivers/staging/ced1401/machine.h:#if (defined(__linux__) || defined(_linux) || defined(__linux)) && !defined(LINUX)

These are all drivers that are shared with another OS, or used
to be. In most of them, we can probably just remove the
#else path, since I don't think they are getting synchronized
any more.

> include/acpi/platform/acenv.h:#if defined(_LINUX) || defined(__linux__)

The acpi header files are maintained outside of Linux and are kept
OS-independent.

> include/linux/coda.h:#if defined(__linux__)
> include/uapi/drm/drm.h:#if defined(__linux__)
> include/uapi/linux/coda.h:#if defined(__linux__)
> include/uapi/linux/fuse.h:#ifdef __linux__

In case of coda, we should not need to care any more, that header just
got broken by the uapi-split for other operating systems.

The drm.h and fuse.h header files are in theory still kept
OS-agnostic and are actively maintained.

> And then we have the following as well..
> fs/ext4/ext4.h:#if defined(__KERNEL__) || defined(__linux__)

This seems to have been copied from the ext2 utils. The ext2/ext3
versions of this file don't have support for other operating systems.

> Trying out a few different prebuilt compilers I had around, I see:
> http://pastebin.com/bTVDLTb1
> 
> So, is our approach just to use __linux__ for builds? I am trying to
> understand rationale behind why #include <linux/types.h> #include <asm/ioctl.h>
> would want __linux__ and why __KERNEL__ check is un-wanted.
> Ofcourse, I cant comment about the "One of the BSDs" in else options..
> and why we'd like to keep it around in kernel :)

We might be in a similar situation on ARM that we are in on MIPS. For
instance, there are some compilers that are targetting (old) Android
that have a slightly different ABI, and building a kernel with those
results in a system call ABI that is incompatible with user space built
with a standard compiler. The safest approach is probably to bail
out early if __linux__ is not set, and force anyone that wants to use
a strange compiler to set the macro manually.

	Arnd