Building gdbserver on m68k Coldfire

Waldemar,

As you know, gdbserver currently fails to build on m68k Coldfire due to
the use of fork(). I've investigated this, and it's due to the fact
that gdbserver looks at uClibc __UCLIBC_HAS_MMU__ define (see
https://sourceware.org/git/gitweb.cgi?p=binutils-gdb.git;a=blob;f=gdb/nat/linux-ptrace.h;h=59549452c0990204ef48451dc2c902c85ee48215;hb=HEAD#l25)
to decide whether to define HAS_NOMMU. If HAS_NOMMU is defined, it uses
vfork().

So, I've fixed this with:

With this fix, the build goes a little bit further, but then fails with:

../nat/linux-ptrace.c: In function 'linux_fork_to_function':
../nat/linux-ptrace.c:282:19: warning: implicit declaration of function 'clone' [-Wimplicit-function-declaration]
       child_pid = clone (function, child_stack + STACK_SIZE,
                   ^
../nat/linux-ptrace.c:283:5: error: 'CLONE_VM' undeclared (first use in this function)
     CLONE_VM | SIGCHLD, child_stack + STACK_SIZE * 2);
     ^
../nat/linux-ptrace.c:283:5: note: each undeclared identifier is reported only once for each function it appears in

It's weird, because this piece of code in linux-ptrace is specifically
written for uClibc/noMMU systems:

https://sourceware.org/git/gitweb.cgi?p=binutils-gdb.git;a=blob;f=gdb/nat/linux-ptrace.c;h=3265b160749b3b635676b423ec1298a26627d78f;hb=HEAD#l271

Is uClibc/noMMU supposed to support clone() and CLONE_VM ?

Thanks!

Thomas

Hi Thomas,
Thomas Petazzoni wrote,

> Waldemar,
> 
> As you know, gdbserver currently fails to build on m68k Coldfire due to
> the use of fork(). I've investigated this, and it's due to the fact
> that gdbserver looks at uClibc __UCLIBC_HAS_MMU__ define (see
> https://sourceware.org/git/gitweb.cgi?p=binutils-gdb.git;a=blob;f=gdb/nat/linux-ptrace.h;h=59549452c0990204ef48451dc2c902c85ee48215;hb=HEAD#l25)
> to decide whether to define HAS_NOMMU. If HAS_NOMMU is defined, it uses
> vfork().
> 
> So, I've fixed this with:
> 
> diff --git a/package/uclibc/uclibc.mk b/package/uclibc/uclibc.mk
> index bbf389e..7bbbe3e 100644
> --- a/package/uclibc/uclibc.mk
> +++ b/package/uclibc/uclibc.mk
> @@ -260,10 +260,12 @@ endif
>  ifeq ($(BR2_USE_MMU),y)
>  define UCLIBC_MMU_CONFIG
>         $(call KCONFIG_ENABLE_OPT,ARCH_USE_MMU,$(@D)/.config)
> +       $(call KCONFIG_ENABLE_OPT,ARCH_HAS_MMU,$(@D)/.config)
>  endef
>  else
>  define UCLIBC_MMU_CONFIG
>         $(call KCONFIG_DISABLE_OPT,ARCH_USE_MMU,$(@D)/.config)
> +       $(call KCONFIG_DISABLE_OPT,ARCH_HAS_MMU,$(@D)/.config)
>  endef
>  endif
>  
> With this fix, the build goes a little bit further, but then fails with:
> 
> ../nat/linux-ptrace.c: In function 'linux_fork_to_function':
> ../nat/linux-ptrace.c:282:19: warning: implicit declaration of function 'clone' [-Wimplicit-function-declaration]
>        child_pid = clone (function, child_stack + STACK_SIZE,
>                    ^
> ../nat/linux-ptrace.c:283:5: error: 'CLONE_VM' undeclared (first use in this function)
>      CLONE_VM | SIGCHLD, child_stack + STACK_SIZE * 2);
>      ^
> ../nat/linux-ptrace.c:283:5: note: each undeclared identifier is reported only once for each function it appears in
> 
> It's weird, because this piece of code in linux-ptrace is specifically
> written for uClibc/noMMU systems:
> 
> https://sourceware.org/git/gitweb.cgi?p=binutils-gdb.git;a=blob;f=gdb/nat/linux-ptrace.c;h=3265b160749b3b635676b423ec1298a26627d78f;hb=HEAD#l271
> 
> Is uClibc/noMMU supposed to support clone() and CLONE_VM ?

Yes, just add this patch:
https://cgit.openadk.org/cgi/cgit/openadk.git/tree/package/gdbserver/patches/nommu.patch

best regards
 Waldemar