Skip to contentSkip to conversation

include/linux/coda.h

PATCH In review Me 6 participants 11 messages May 22, 2001 – May 24, 2001 Coda file system

lore
Share & tools
lore

Me

author
#0 lore raw

PATCH 1/1 include/linux/coda.h

Coda.h assumes that __KERNEL__ can only be defined if __linux__ is, which is
painfully false. This allows the kernel compile to get farther on my token FreeBSD box.

-Ryan

include/linux/coda.h +4 -3
@@ -100,6 +100,10 @@
 
 #if defined(__linux__)
 #define cdev_t u_quad_t
+#else
+#define cdev_t dev_t
+#endif
+
 #ifndef __KERNEL__
 #if !defined(_UQUAD_T_) && (!defined(__GLIBC__) || __GLIBC__ < 2)
 #define _UQUAD_T_ 1
@@ -108,9 +112,6 @@
 #else /*__KERNEL__ */
 typedef unsigned long long u_quad_t;
 #endif /* __KERNEL__ */
-#else
-#define cdev_t dev_t
-#endif
 
 #ifdef __CYGWIN32__
 struct timespec {
maintainer
#1 lore raw
Quoted 5 lines · Me

Coda.h assumes that __KERNEL__ can only be defined if __linux__ is, which is
painfully false. This allows the kernel compile to get farther on my token FreeBSD box.

-Ryan

Are you trying to compile a Linux kernel on a FreeBSD machine, or is this a bug in the Coda kernel module in the FreeBSD tree?

Jan

author
#2 lore raw
In reply to Me
Quoted 3 lines · you

Are you trying to compile a Linux kernel on a FreeBSD machine, or is this a bug in the Coda kernel module in the FreeBSD tree?

Sorry, I should've been more specific. I'm trying to compile the Linux kernel (2.4.5pre3) on a FreeBSD machine, which actually works quite well with this patch applied. This is the only place in the core that FreeBSD gets hung up on (init/main.c seems to include it unconditionally). Some drivers may still make incorrect assumptions about __linux__ being defined, though.

-Ryan

#3 lore raw
Quoted 3 lines

Sorry, I should've been more specific. I'm trying to compile the Linux kernel (2.4.5pre3) on a FreeBSD machine, which actually works quite well with this patch applied. This is the only place in the core that FreeBSD gets hung up

Why is your cross compiler outputting different symbols to a linux native compiler ?

Quoted 2 lines

on (init/main.c seems to include it unconditionally). Some drivers may still make incorrect assumptions about __linux__ being defined, though.

If __linux__ is not defined by the cross compiler, then the cross compiler is broken. A cross compiler has the same environment as the native compiler for the target. The only stuff that should break (well should as in might) is tools native built

Or am I misunderstanding the report ?

#4 lore raw
Quoted 9 lines · [email protected] said:

Why is your cross compiler outputting different symbols to a linux
native compiler ?

If __linux__ is not defined by the cross compiler, then the cross compiler is broken. A cross compiler has the same environment as the native compiler for the target. The only stuff that should break (well should as in might) is tools native built

Or am I misunderstanding the report ?

Why use a cross compiler? With the obvious exception of UML, the Linux kernel is not a Linux executable, so why should it need to be compiled with a compiler which targets such?

The kernel compiles quite happily with compilers which aren't targetted specifically at Linux -- the CODA compatibility cruft being the one exception. I often just comment out the CODA includes from <linux/fs.h> to get round the same problem.

--
dwmw2

maintainer
#5 lore raw
In reply to Alan Cox
Quoted 4 lines · David Woodhouse

The kernel compiles quite happily with compilers which aren't targetted specifically at Linux -- the CODA compatibility cruft being the one exception. I often just comment out the CODA includes from <linux/fs.h> to get round the same problem.

Well, the original idea for having all that compatibility cruft in coda.h was that the identical header could be used both by userspace, and as part of the kernel code of all Coda ports.

But with problems like these popping up and the silly thing is just turning into an unmaintainable mess of #ifdef bloat.

#include <linux/coda.h> can be taken out of linux/fs.h when I change a few lines of code. In fact with those changes all Coda related headers can be moved from include/linux/ to fs/coda/, although coda.h contains the 'exported interface' and should probably remain in include/linux.

At least this will solve your problem as long as Coda isn't compiled into the kernel, but not the reported problem of compiling a Linux kernel on a different (Coda supported) OS without using a properly configured cross-compiler.

I will try cleaning things up by having a common 'coda.h' for all platforms (and userspace) and platform specific 'coda_types.h' files which define missing types and such. It would definitely get rid of a lot of the bloat, and probably make things somewhat easier to maintain.

Jan

btw. Coda is not an acronym, no need to capitalize it.

#6 lore raw
In reply to Alan Cox
Quoted 6 lines · Alan Cox

If __linux__ is not defined by the cross compiler, then the cross compiler is broken. A cross compiler has the same environment as the native compiler for the target. The only stuff that should break (well should as in might) is tools native built

Or am I misunderstanding the report ?

It is not a cross compiler, it is the FreeBSD native gcc. As I understood it, Linux is a self contained project and should be targetting the native platform, not Linux on that platform. A cross compiler should not be needed unless one is building for another CPU.

The coda.h file was doing something along the lines of:

#ifdef __linux__
#ifdef __KERNEL__
/* Define things */
#endif
#endif

When compiling the kernel under FreeBSD, __KERNEL__ is defined, but __linux__ is not. I think this is an error on the part of the header file, because on non-Linux build environments, which would otherwise compile the Linux kernel correctly, do not have __linux__ defined.

However, not many people will probably find much use in compiling the kernel on other platforms, so if you think this isn't worth inclusion, I totally understand. I'm in the process of porting UML to FreeBSD, and having this patch in the tree would make my job slightly easier.

-Ryan

#7 lore raw
Quoted 4 lines

When compiling the kernel under FreeBSD, __KERNEL__ is defined, but __linux__ is not. I think this is an error on the part of the header file, because on non-Linux build environments, which would otherwise compile the Linux kernel correctly, do not have __linux__ defined.

Thats a problem with your compiler setup.

Quoted 4 lines

However, not many people will probably find much use in compiling the kernel on other platforms, so if you think this isn't worth inclusion, I totally understand. I'm in the process of porting UML to FreeBSD, and having this patch in the tree would make my job slightly easier.

Teaching your gcc to do -D__linux__ would be cleaner and also catch the zillion other cases that simply came out wrong instead of erroring

One shell script for your CC and off you go

#8 lore raw
In reply to Ryan Cumming
Quoted 7 lines · Ryan Cumming

When compiling the kernel under FreeBSD, __KERNEL__ is defined, but __linux__ is not. I think this is an error on the part of the header file, because on non-Linux build environments, which would otherwise compile the Linux kernel correctly, do not have __linux__ defined.

However, not many people will probably find much use in compiling the kernel on other platforms, so if you think this isn't worth inclusion, I

hmmm...

building for:
SPARC under Solaris
PPC under BeOS
PA-RISC under HP-UX
M68k under HP-UX

I'd really like the last one.
I have a HP, M68k box I'd like to run linux on and I've not seen a M68K distro yet. But I haven't had time to try it yet either.

-Thomas

maintainer
#9 lore raw

UML cross-platform build problems (was Re: [PATCH] include/linux/coda.h)

In reply to Alan Cox
Quoted 7 lines · Ryan Cumming
Quoted 5 lines · Alan Cox

If __linux__ is not defined by the cross compiler, then the cross compiler is broken. A cross compiler has the same environment as the native compiler for the target. The only stuff that should break (well should as in might) is tools native built

Quoted 6 lines
Quoted 1 line

Or am I misunderstanding the report ?

It is not a cross compiler, it is the FreeBSD native gcc. As I understood it, Linux is a self contained project and should be targetting the native platform, not Linux on that platform. A cross compiler should not be needed unless one is building for another CPU.

The problem is that the same coda.h header is also used as part of the FreeBSD kernel to build a Coda kernel module. Both Linux and FreeBSD have different include and typedef requirements for their kernel code.

There is no way for an identical coda.h header to recognize whether it is built as part of the FreeBSD kernel or as part of a (UML) Linux kernel, except if the parts that are OS dependent are not part of the same header.

I agree that a UML kernel on FreeBSD should be a native binary and not cross-compiled. However, this could be an UML specific problem and -D__linux__ should be added to CFLAGS in arch/uml/Makefile as all drivers and filesystems that are developed for different platforms and share parts of their code are possibly affected.

Jan

#10 lore raw

Re: UML cross-platform build problems (was Re: [PATCH] include/linux/coda.h)

Quoted 3 lines · [email protected] said:

I agree that a UML kernel on FreeBSD should be a native binary and not cross-compiled. However, this could be an UML specific problem and -D__linux__ should be added to CFLAGS in arch/uml/Makefile

Exactly. Don't be shy about fiddling CFLAGS in UML-specific Makefiles. This is already done quite a bit (-D__i386__ and the construction of userspace CFLAGS are examples).

This is very UML-specific, so there's no reason to bother the rest of the kernel (unless, of course, you find a bug that could affect other ports).

Jeff