Re: [PATCH 2/4] Add NR_syscalls for x86_64

[Jason Baron <jbaron@redhat.com>]

On Wed, Aug 26, 2009 at 10:39:26AM -0400, Steven Rostedt wrote:
> On Wed, 26 Aug 2009, Jason Baron wrote:
> > > > diff --git a/Kbuild b/Kbuild
> > > > index f056b4f..5f43d4d 100644
> > > > --- a/Kbuild
> > > > +++ b/Kbuild
> > > > @@ -78,7 +78,7 @@ endef
> > > >  arch/$(SRCARCH)/kernel/asm-offsets.s: arch/$(SRCARCH)/kernel/asm-offsets.c \
> > > >                                        $(obj)/$(bounds-file) FORCE
> > > >  	$(Q)mkdir -p $(dir $@)
> > > > -	$(call if_changed_dep,cc_s_c)
> > > > +	$(call if_changed_dep,cc_s_c_define_offset)
> > > >  
> > > >  $(obj)/$(offsets-file): arch/$(SRCARCH)/kernel/asm-offsets.s Kbuild
> > > >  	$(call cmd,offsets)
> > > > diff --git a/arch/x86/include/asm/unistd_64.h b/arch/x86/include/asm/unistd_64.h
> > > > index 900e161..70c0c3d 100644
> > > > --- a/arch/x86/include/asm/unistd_64.h
> > > > +++ b/arch/x86/include/asm/unistd_64.h
> > > > @@ -688,6 +688,12 @@ __SYSCALL(__NR_perf_counter_open, sys_perf_counter_open)
> > > >  #endif	/* __NO_STUBS */
> > > >  
> > > >  #ifdef __KERNEL__
> > > > +
> > > > +#ifndef CREATE_OFFSETS
> > > > +#include <asm/asm-offsets.h>
> > > > +#define NR_syscalls (__NR_syscall_max + 1)
> > > > +#endif
> > > > +
> > > >  /*
> > > >   * "Conditional" syscalls
> > > >   *
> > > > diff --git a/scripts/Makefile.build b/scripts/Makefile.build
> > > > index 5c4b7a4..4177858 100644
> > > > --- a/scripts/Makefile.build
> > > > +++ b/scripts/Makefile.build
> > > > @@ -145,6 +145,10 @@ $(multi-objs-y:.o=.lst) : modname = $(modname-multi)
> > > >  quiet_cmd_cc_s_c = CC $(quiet_modtag)  $@
> > > >  cmd_cc_s_c       = $(CC) $(c_flags) -fverbose-asm -S -o $@ $<
> > > >  
> > > > +quiet_cmd_cc_s_c_define_offset = CC $(quiet_modtag)  $@
> > > > +cmd_cc_s_c_define_offset       = $(CC) $(c_flags) -DCREATE_OFFSETS \
> > > > +					-fverbose-asm -S -o $@ $<
> > > > +
> > > >  $(obj)/%.s: $(src)/%.c FORCE
> > > >  	$(call if_changed_dep,cc_s_c)
> > > 
> > > 
> > > Ug, yuck yuck yuck. Sometimes "it works" is not good enough ;-)
> > > 
> > > How about just adding in asm/unistd.h
> > > 
> > >  #ifdef __KERNEL__
> > >  # ifdef CONFIG_X86_32
> > >  #  include "unistd_32.h"
> > >  # else
> > >  #  include "unistd_64.h"
> > > +#  define NR_syscalls (__NR_syscall_max + 1)
> > >  # endif
> > >  #else
> > >  # ifdef __i386__
> > >  #  include "unistd_32.h"
> > >  # else
> > >  #  include "unistd_64.h"
> > >  # endif
> > >  #endif
> > > 
> > > And if this is a problem because asm-offset_64.c includes asm/unistd.h, 
> > > then make it just include unistd_64.h.
> > > 
> > > Or am I missing something?
> > > 
> > > -- Steve
> > > 
> > > 
> > 
> > ok, so using the above patch gives me:
> > 
> >   CC      arch/x86/kernel/ftrace.o
> > arch/x86/kernel/ftrace.c: In function ?syscall_nr_to_meta?:
> > arch/x86/kernel/ftrace.c:497:35: error: ?__NR_syscall_max? undeclared
> > (first use in this function)
> > arch/x86/kernel/ftrace.c:497:35: error: (Each undeclared identifier is
> > reported only once
> > arch/x86/kernel/ftrace.c:497:35: error: for each function it appears
> > in.)
> > arch/x86/kernel/ftrace.c: In function ?syscall_name_to_nr?:
> > arch/x86/kernel/ftrace.c:510:19: error: ?__NR_syscall_max? undeclared
> > (first use in this function)
> > arch/x86/kernel/ftrace.c: In function ?arch_init_ftrace_syscalls?:
> > arch/x86/kernel/ftrace.c:536:7: error: ?__NR_syscall_max? undeclared
> > (first use in this function)
> > make[2]: *** [arch/x86/kernel/ftrace.o] Error 1
> > make[1]: *** [arch/x86/kernel] Error 2
> > make: *** [arch/x86] Error 2
> > 
> > 
> > That is the the first time, I encounter a 'NR_syscalls' usage, we error, b/c
> > unistd.h does not include 'asm/asm-offsets.h'. If I add an #include of
> > 'asm/asm-offsets.h' to ftrace.c the compile is fine. But the point of
> > this was to the 'NR_syscalls' definition just from the unistd.h file...
> > I'll see if I can come up with a simpler patch.
> 
> Then add the asm/asm-offests.h to the x86_64 define in unistd.h and have 
> the asm-offset_64.c only include the unistd_64.h file. Would that work?
> 
> -- Steve
> 

doesn't quite work, b/c the top of asm-offsets_64.c has a bunch of
includes, such as:  <linux/sched.h>, which in turn include unistd.h at
a point in time where asm/asm-offests.h does not exist.

However, a much simpler version of the previous patch I posted does do
the trick, below. I hope this version is more palatable?

thanks,

-Jason

Signed-off-by: Jason Baron <jbaron@redhat.com>

diff --git a/arch/x86/include/asm/unistd_64.h b/arch/x86/include/asm/unistd_64.h
index 900e161..b9f3c60 100644
--- a/arch/x86/include/asm/unistd_64.h
+++ b/arch/x86/include/asm/unistd_64.h
@@ -688,6 +688,12 @@ __SYSCALL(__NR_perf_counter_open, sys_perf_counter_open)
 #endif	/* __NO_STUBS */
 
 #ifdef __KERNEL__
+
+#ifndef COMPILE_OFFSETS
+#include <asm/asm-offsets.h>
+#define NR_syscalls (__NR_syscall_max + 1)
+#endif
+
 /*
  * "Conditional" syscalls
  *
diff --git a/arch/x86/kernel/asm-offsets_64.c b/arch/x86/kernel/asm-offsets_64.c
index 898ecc4..4a6aeed 100644
--- a/arch/x86/kernel/asm-offsets_64.c
+++ b/arch/x86/kernel/asm-offsets_64.c
@@ -3,6 +3,7 @@
  * This code generates raw asm output which is post-processed to extract
  * and format the required data.
  */
+#define COMPILE_OFFSETS
 
 #include <linux/crypto.h>
 #include <linux/sched.h>