[okl4-developer] OKL4 2.1: Original version of 'arch/arm/pistachio/cpu/arm926ejs/include/cache.h' is not compilable

[Geoffrey Lee]

On Wed, Jul 09, 2008 at 04:31:14PM +0200, Frank Kaiser wrote:
> Hello
> 
>  

Hi Frank

> 
> When I started my platform development, I drew the okl4_2.1.tar.gz
> tarball from http://wiki.ok-labs.com/FrontPage?action=AttachFile.
> Unzipping it, for some reason, I did not get out the core specific
> include files 'cache.h', 'cpu.h' and 'syscon.h' located at
> 'arch/arm/pistachio/cpu/arm926ejs/include/'. Since there is no example
> platform in the tarball using the ARM926EJ-S core, I thought they were
> left out intentionally, and I created my own ones by combining the
> XSCALE and ARM920T variants with ARM926EJ-S variants from OKL4 1.4.9. It
> just did not occur to me that the unzip went wrong.

Please rest assured that the ARM926EJ-S files are still in the
2.1 distribution.  They can be located in:

okl4_2.1/arch/arm/pistachio/cpu/arm926ejs/
okl4_2.1/arch/arm/pistachio/cpu/arm926ejs/src/
okl4_2.1/arch/arm/pistachio/cpu/arm926ejs/src/rvct/
okl4_2.1/arch/arm/pistachio/cpu/arm926ejs/src/rvct/cache.cc
okl4_2.1/arch/arm/pistachio/cpu/arm926ejs/include/
okl4_2.1/arch/arm/pistachio/cpu/arm926ejs/include/cache.h
okl4_2.1/arch/arm/pistachio/cpu/arm926ejs/include/syscon.h
okl4_2.1/arch/arm/pistachio/cpu/arm926ejs/include/cpu.h

> 
> Meanwhile a colleague also used the mentioned tarball and put it in our
> SVN repository. Recently I began to integrate my new platform code into
> the SVN repository and discovered the missing header file. Comparing
> them with my own creation showed that I have some insignificant
> differences with 'cpu.h' and 'syscon.h', but a significant difference in
> the referenced header files with 'cache.h'.
> 
> Here the include statements of the original:
> 
> #include <debug.h>
> 
> #include <cpu/syscon.h>
> 
> #include <cache.h>
> 
> #include <plat/platform.h> /* Get cache size */
> 
> Here the include statements of my private version:
> 
>              #include <kernel/debug.h>
> 
> #include <kernel/arch/asm.h>
> 
> #include <kernel/cpu/syscon.h>

The header files should read:

#include <kernel/debug.h>
#include <kernel/cpu/syscon.h>
#include <kernel/cache.h>
#include <kernel/arch/asm.h>

In the development cycle we have re-arranged the headers for better
clarify, and the ARM926EJ-S CPU files have not been updated
to reflect this.  While we do not have a platfrom based on this
CPU at this point in time we feel it is still beneficial to release
these files to enable others to jump-start their porting efforts.


> 
> Another difference is that the original version contains the definition 
>              'static const word_t CACHE_LINE_MASK = (CACHE_LINE_SIZE -
> 1)' 
> which does not exist in my private version, and I do not see that
> CACHE_LINE_MASK is anywhere referenced.
> 
> My private version compiles w/o a flaw, but the original one files for
> the first compiled source file (asid.cc) with the following multiple
> error:
> 
> sam/pistachio/include/kernel/cpu/cache.h: In static member function
> `static void arm_cache::cache_flush_i()':
> 
> sam/pistachio/include/kernel/cpu/cache.h:150: error: expected `)' before
> "_"
> 
> sam/pistachio/include/kernel/cpu/cache.h:148: warning: unused variable
> 'zero'
> 
> The relevant code lines are:
> 
>         __asm__ __volatile__ (
> 
>              "  mcr     p15, 0, "_(zero)", c7, c5, 0    ;"      /* Flush
> I cache */
> 
> #if defined(__GNUC__)
> 
>             :: [zero] "r" (zero)
> 
> #endif
> 
> The problem is that '_()' is a preprocessor macro defined in
> 'kernel/arch/asm.h' which the original file misses.
> 
> I made the addition to the original file by following its include path
> scheme:
> 
> #include <debug.h>
> 
> #include <arch/asm.h>
> 
> #include <cpu/syscon.h>
> 
> #include <cache.h>
> 
> #include <plat/platform.h> /* Get cache size */
> 
> Surprisingly this correction does not work. I get most of the files of
> the PISTACHIO kernel compiled until it comes to the platform specific
> ones. With 'platform/at91/pistachio/src/interrupt.cc' the error
> reappears, and additionally the compiler claims that it does not find
> the header files 'debug.h', 'arch/asm.h', 'cpu/syscon.h', 'cache.h' and
> 'plat/platform.h', the 5 ones included by
> '.../arm926ejs/include/cache.h'. Something must we wrong the the include
> paths seen by the compiler, and a view in the log shows the difference.
> 
> When compiling, for instance 'asid.cc', the include paths are ('sam' is
> the build directory):
> 
> -Isam/pistachio/include 
> 
> -Isam/pistachio/object/pistachio/include 
> 
> -Ipistachio/include 
> 
> -Iarch/arm/pistachio/v5/include
> 
> However, for 'interrupt.cc' the include paths are:
> 
> -Isam/pistachio/include 
> 
> -Isam/pistachio/object/pistachio/src 
> 
> -Ipistachio/src
> 
> Both files share only the first include paths of both list. Checking
> this location shows that 'asm.h' is found at
> 'sam/pistachio/include/kernel/arch'. That explains why my private
> version of 'cache.h' finds it. But when compiling 'asid.cc' the original
> version finds it too, and that is thru the second include path
> 'sam/pistachio/object/pistachio/include' under which copy of the arch
> directory exists. It looks as if there is a mix-up of include paths
> versus header file copies in the build directory. The question is:

With the above header changes that I documented above, there
should be no need to modify the build system to change the include
paths.


> 
> *       Why are there different include paths for the pistachio kernel
> source files and for the platform-dependent source files?
> 
> *       Why are there two copies of the architectural-, cpu-, kdb- and
> platform-specific header files, one set in
> '<build_dir>/pistachio/include/kernel' and one set in
> '<build_dir>/pistachio/object/pistachio/include'?


the pistachio/include/kernel headers are the headers which are 
meant to be exported.  The headers in the object directory 
are typically headers which are generated.  However, for kernel
we also populate this with header files.  This is transitional,
in the future exported headers should belong in
pistachio/include/kernel.


> 
> Although the copies are created by SCONS, it is confusing regarding the
> correct set-up of #include directives in the source file. If I had to
> write my platform source files totally from scratch, I'd have had a hard
> time with this. Actually I made a test compilation with an existing
> platform to figure out how the build directory looks like after a
> successful build.
> 

	-gl

>  
> 
> Regards
> 
> Frank
> 

> _______________________________________________
> Developer mailing list
> Developer at okl4.org
> https://lists.okl4.org/mailman/listinfo/developer


--