* header with widely used value definitions
* handle all asm-related things in one file (Makefile.asm)
* move some asm bits from Makefile.build there
(rule %.s:%.c)
* add script to generate headers from assembles output
(hopefully better output, MIPS testing/joining to all arch
probably needed)
rfc-by: Oleg Verych
---
See next patches, to know how to use it
include/asm-generic/asm-values.h | 7 ++++
scripts/Makefile.asm | 26 ++++++++++++++++++
scripts/Makefile.build | 6 ----
scripts/mkasm-values.sh | 55 +++++++++++++++++++++++++++++++++++++
4 files changed, 88 insertions(+), 6 deletions(-)
Index: linux-2.6.22-rc4-mm2/scripts/Makefile.asm
===================================================================
--- /dev/null 1970-01-01 00:00:00.000000000 +0000
+++ linux-2.6.22-rc4-mm2/scripts/Makefile.asm 2007-06-12 23:43:41.070624500 +0200
@@ -0,0 +1,26 @@
+#
+# Help generate definitions needed by assembly language modules.
+#
+ifndef asm-values_h
+ asm-values := $(shell test -e $(srctree)/$(src)/asm-values.c && echo ok)
+ ifneq ($(asm-values),)
+ asm-values_h := asm-values.h
+ asm-values := asm-values.s
+ endif
+endif
+
+always += $(asm-values_h)
+targets += $(asm-values_h) $(asm-values)
+mkasm-values := $(srctree)/scripts/mkasm-values.sh
+
+quiet_cmd_mkasm-values = GEN $@
+ cmd_mkasm-values = $(CONFIG_SHELL) $(mkasm-values) $< $@ $(2)
+
+quiet_cmd_cc_s_c = CC $(quiet_modtag) $@
+ cmd_cc_s_c = $(CC) $(c_flags) -fverbose-asm -S -o $@ $<
+
+$(obj)/asm-values.h: $(obj)/asm-values.s
+ $(call cmd,mkasm-values,$(ASM_NS))
+
+$(obj)/%.s: $(src)/%.c FORCE
+ $(call if_changed_dep,cc_s_c)
Index: linux-2.6.22-rc4-mm2/scripts/Makefile.build
===================================================================
--- linux-2.6.22-rc4-mm2.orig/scripts/Makefile.build 2007-06-13 00:45:54.195930250 +0200
+++ linux-2.6.22-rc4-mm2/scripts/Makefile.build 2007-06-12 22:10:31.481297000 +0200
@@ -145,10 +145,4 @@ $(multi-objs-y:.o=.s) : modname = $(mo
$(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 $@ $<
-
-$(obj)/%.s: $(src)/%.c FORCE
- $(call if_changed_dep,cc_s_c)
-
quiet_cmd_cc_i_c = CPP $(quiet_modtag) $@
cmd_cc_i_c = $(CPP) $(c_flags) -o $@ $<
Index: linux-2.6.22-rc4-mm2/scripts/mkasm-values.sh
===================================================================
--- /dev/null 1970-01-01 00:00:00.000000000 +0000
+++ linux-2.6.22-rc4-mm2/scripts/mkasm-values.sh 2007-06-12 21:49:05.096903000 +0200
@@ -0,0 +1,55 @@
+#!/bin/sh
+
+# Input file "*.s", made by Kbuild from a "*.c" source, where values of
+# interest are calculated, is processed in a suitable output header file.
+#
+# $1 - input filename;
+# $2 - output filename;
+# $3 - generate (private) file with given namespace (optional)
+
+set -e
+
+[ -z "$1" ] || [ -z "$2" ] && {
+ echo "$0:
+
+Two parameters needed. Exiting.
+"
+ exit 1
+}
+
+NS=`echo $3 | tr a-z A-Z`
+case $NS in
+ MIPS)
+SED_SCRIPT='
+/^@@@/{
+s|^@@@||;
+s| #.*$||;
+p;
+}' ;;
+ *)
+TAB="`printf '\t'`"
+SED_SCRIPT="
+/^->/{
+s|^->||;
+s|^\([^ ]*\) [\$#]*\([^ ]*\) \([^$TAB]*\).*|#define \1 \2$TAB/* \3 */|;
+p;
+}" ;;
+esac
+
+NS=__ASM_VALUES_${NS:=H}__
+
+exec 1>$2
+echo "\
+#if !defined($NS)
+#define $NS
+
+/*
+ * $2
+ * was generated from
+ * $1
+ */
+"
+sed -ne "$SED_SCRIPT" $1
+
+echo "
+#endif"
Index: linux-2.6.22-rc4-mm2/include/asm-generic/asm-values.h
===================================================================
--- /dev/null 1970-01-01 00:00:00.000000000 +0000
+++ linux-2.6.22-rc4-mm2/include/asm-generic/asm-values.h 2007-06-12 23:33:34.176696000 +0200
@@ -0,0 +1,7 @@
+#define DEFINE(sym, val) \
+ asm volatile("\n->" #sym " %0 " #val : : "i" (val))
+
+#define BLANK() asm volatile("\n->" : : )
+
+#define OFFSET(sym, str, mem) \
+ DEFINE(sym, offsetof(struct str, mem));
--
Jun 13, 2007 at 01:36:51AM +0200, asm-values patch set:
> * header with widely used value definitions
> * handle all asm-related things in one file (Makefile.asm)
> * move some asm bits from Makefile.build there
> (rule %.s:%.c)
> * add script to generate headers from assembles output
> (hopefully better output, MIPS testing/joining to all arch
> probably needed)
>
> rfc-by: Oleg Verych
> ---
So, is it another not so juicy lets-break-it-all stuff from me, or it just
doesn't apply for you?
I thought update all that asm-offsets stuff just like that. I want to
know, why it does not fit.
Thanks.
____
On Thu, Jun 14, 2007 at 09:41:43AM +0200, Oleg Verych wrote:
> Jun 13, 2007 at 01:36:51AM +0200, asm-values patch set:
> > * header with widely used value definitions
> > * handle all asm-related things in one file (Makefile.asm)
> > * move some asm bits from Makefile.build there
> > (rule %.s:%.c)
> > * add script to generate headers from assembles output
> > (hopefully better output, MIPS testing/joining to all arch
> > probably needed)
> >
> > rfc-by: Oleg Verych
> > ---
>
> So, is it another not so juicy lets-break-it-all stuff from me, or it just
> doesn't apply for you?
I have not yet looked at it. A few other items pending that I need to finish
off before next merge window has priority.
Sam
On Thu, Jun 14, 2007 at 11:12:25AM +0200, Sam Ravnborg wrote:
> On Thu, Jun 14, 2007 at 09:41:43AM +0200, Oleg Verych wrote:
> > Jun 13, 2007 at 01:36:51AM +0200, asm-values patch set:
> > > * header with widely used value definitions
> > > * handle all asm-related things in one file (Makefile.asm)
> > > * move some asm bits from Makefile.build there
> > > (rule %.s:%.c)
> > > * add script to generate headers from assembles output
> > > (hopefully better output, MIPS testing/joining to all arch
> > > probably needed)
> > >
> > > rfc-by: Oleg Verych
> > > ---
> >
> > So, is it another not so juicy lets-break-it-all stuff from me, or it just
> > doesn't apply for you?
>
> I have not yet looked at it. A few other items pending that I need to finish
> off before next merge window has priority.
That's my view about asm-offsets cleanup/generalizing. I just need ack or
nack from, for example arch maintainers if this is acceptable at all or
anybody, actually interested in this kind of things. Not cool stuff, but
still quite straight and easy to finish and merge.
____
On Thu, Jun 14, 2007 at 01:16:21PM +0200, Oleg Verych wrote:
> On Thu, Jun 14, 2007 at 11:12:25AM +0200, Sam Ravnborg wrote:
> > On Thu, Jun 14, 2007 at 09:41:43AM +0200, Oleg Verych wrote:
> > > Jun 13, 2007 at 01:36:51AM +0200, asm-values patch set:
> > > > * header with widely used value definitions
> > > > * handle all asm-related things in one file (Makefile.asm)
> > > > * move some asm bits from Makefile.build there
> > > > (rule %.s:%.c)
> > > > * add script to generate headers from assembles output
> > > > (hopefully better output, MIPS testing/joining to all arch
> > > > probably needed)
> > > >
> > > > rfc-by: Oleg Verych
> > > > ---
> > >
> > > So, is it another not so juicy lets-break-it-all stuff from me, or it just
> > > doesn't apply for you?
> >
> > I have not yet looked at it. A few other items pending that I need to finish
> > off before next merge window has priority.
>
> That's my view about asm-offsets cleanup/generalizing. I just need ack or
> nack from, for example arch maintainers if this is acceptable at all or
> anybody, actually interested in this kind of things. Not cool stuff, but
> still quite straight and easy to finish and merge.
I will for sure take a look.
But with less than 1 hour/day for Linux stuff this has been scheduled after
other stuff I need to do before next merge window.
Sam
Hi Oleg.
On Wed, Jun 13, 2007 at 01:36:51AM +0200, Oleg Verych wrote:
> * header with widely used value definitions
> * handle all asm-related things in one file (Makefile.asm)
> * move some asm bits from Makefile.build there
> (rule %.s:%.c)
> * add script to generate headers from assembles output
> (hopefully better output, MIPS testing/joining to all arch
> probably needed)
>
Took a look at this patch at last.
Your patch moves everything to Makefile.asm and uses a script
to generate the offset file.
Now where we go the step to use a script to generate the offset file
a much simpler approach would be to move both steps to the shell script.
So the script are called with the input file as one of the
arguments and it spits out the resulting file to stdout.
Doing it this way would allow the caller to have full control
on the filenames and the current call-site in top-level Kbuild
would still be much simpler than today.
Moving the bits from Makefile.build that is only used for asm-offset is
a nice cleanup and this should the script version support too.
I see no value in renaming from asm_offset to asm_value - please drop it.
Introducing the generic asm-values.h should wait and when you do
you should be preapred to update all architectures (ecasue otherwise it
will not happen).
Sorry for the late feedback.
Sam
On Sun, Sep 16, 2007 at 08:29:12PM +0200, Sam Ravnborg wrote:
> Hi Oleg.
Hallo. Nice, you are bringing it back. I'll try to have LKML-like
output this time, not a makefile mess and stuff:
[]
> I see no value in renaming from asm_offset to asm_value - please drop it.
> Introducing the generic asm-values.h should wait and when you do
> you should be preapred to update all architectures (ecasue otherwise it
> will not happen).
All archs, are using same syntax. But.
Interesting exception is MIPS, where tokens are organized and implemented
more nicely, that is where *asm-values* name coming from, actually. One
can see all that text, size, constant tokens in addition to just offsets.
So, if name change isn't worth, then OK; filename compatibility check, as
one in `linux/Kbuild', is easy to hack as well to remove.
But nice feature/tool with meaningful name and functionality is just
another thing.
Opinions?
____
On Sun, Sep 16, 2007 at 09:32:58PM +0200, Oleg Verych wrote:
> On Sun, Sep 16, 2007 at 08:29:12PM +0200, Sam Ravnborg wrote:
> > Hi Oleg.
>
> Hallo. Nice, you are bringing it back. I'll try to have LKML-like
> output this time, not a makefile mess and stuff:
>
> []
> > I see no value in renaming from asm_offset to asm_value - please drop it.
> > Introducing the generic asm-values.h should wait and when you do
> > you should be preapred to update all architectures (ecasue otherwise it
> > will not happen).
>
> All archs, are using same syntax. But.
>
> Interesting exception is MIPS, where tokens are organized and implemented
> more nicely, that is where *asm-values* name coming from, actually. One
> can see all that text, size, constant tokens in addition to just offsets.
>
> So, if name change isn't worth, then OK; filename compatibility check, as
> one in `linux/Kbuild', is easy to hack as well to remove.
>
> But nice feature/tool with meaningful name and functionality is just
> another thing.
>
> Opinions?
Everyone and their cousin these days knows that asm-offset is constants
generated from C and used in assembler.
If we benefit from more values in the asm-offset file - let's do it.
But there is no reason to change a name that has been used for this purpose
for as long as I can remember.
Sam