Skip to contentSkip to conversation

Initialisation Sequence Framework

PATCH New Graeme Russ 2 participants 10 messages Apr 22, 2012 – May 3, 2012

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

cover Initialisation Sequence Framework

Well here it is (finally) - My Initialisation Sequence Framework

This series is limited to x86 and is controlled by CONFIG_INIT_FUNC so no other arch should be effected. I don't have any cross compilers, so I would appreciate if others could confirm the zero impact assertion

Asbestos suit on

Fire away ;)

Regards,

Graeme

Graeme Russ (3):
init_func: Add fundamental framework
init_func: Add x86 support
init_func: Use for eNET board

 Makefile                         |   34 +-
 arch/x86/cpu/cpu.c               |    3 +
 arch/x86/cpu/sc520/sc520_sdram.c |    3 +
 arch/x86/cpu/sc520/sc520_timer.c |    2 +
 arch/x86/cpu/u-boot.lds          |   23 +
 arch/x86/lib/board.c             |   10 +-
 arch/x86/lib/init_helpers.c      |   14 +
 arch/x86/lib/init_wrappers.c     |   11 +-
 arch/x86/lib/pcat_interrupts.c   |    2 +
 arch/x86/lib/relocate.c          |    4 +
 board/eNET/eNET.c                |    5 +
 common/Makefile                  |    2 +
 common/console.c                 |    3 +
 common/env_dataflash.c           |    1 +
 common/env_eeprom.c              |    1 +
 common/env_fat.c                 |    1 +
 common/env_flash.c               |    2 +
 common/env_mgdisk.c              |    1 +
 common/env_mmc.c                 |    1 +
 common/env_nand.c                |    1 +
 common/env_nowhere.c             |    1 +
 common/env_nvram.c               |    1 +
 common/env_onenand.c             |    1 +
 common/env_sf.c                  |    1 +
 common/serial.c                  |    2 +
 common/stdio.c                   |    2 +
 config.mk                        |    2 +
 doc/README.INIT_FUNC             |   65 ++
 include/configs/eNET.h           |    1 +
 include/init_func.h              |   37 +
 tools/Makefile                   |    6 +
 tools/mkinitseq.c                | 1512 ++++++++++++++++++++++++++++++++++++++
 32 files changed, 1746 insertions(+), 9 deletions(-)
 create mode 100644 doc/README.INIT_FUNC
 create mode 100644 include/init_func.h
 create mode 100644 tools/mkinitseq.c
Signature · 1 line
1.7.7.6
#1 lore raw

PATCH 1/3 init_func: Add fundamental framework

---
 Makefile             |   34 ++-
 common/Makefile      |    2 +
 config.mk            |    2 +
 doc/README.INIT_FUNC |   65 +++
 include/init_func.h  |   37 ++
 tools/Makefile       |    6 +
 tools/mkinitseq.c    | 1512 ++++++++++++++++++++++++++++++++++++++++++++++++++
 7 files changed, 1657 insertions(+), 1 deletions(-)
 create mode 100644 doc/README.INIT_FUNC
 create mode 100644 include/init_func.h
 create mode 100644 tools/mkinitseq.c
Makefile +33 -1
index cdd4294..62d9f81 100644
@@ -467,8 +467,40 @@ GEN_UBOOT = \
 			-Map u-boot.map -o u-boot
 endif
 
+ifeq ($(CONFIG_INIT_FUNC),y)
+INIT_SEQ = $(obj)init_seq.o
+
+GEN_UBOOT_INIT = \
+		UNDEF_SYM=`$(OBJDUMP) -x $(LIBBOARD) $(LIBS) | \
+		sed  -n -e 's/.*\($(SYM_PREFIX)__u_boot_cmd_.*\)/-u\1/p'|sort|uniq`;\
+		cd $(LNDIR) && $(LD) $(LDFLAGS) $(LDFLAGS_$(@F)) $$UNDEF_SYM $(__OBJS) \
+			--start-group $(__LIBS) --end-group \
+			-Map u-boot-init.map -o u-boot-init
+
+$(obj)u-boot-init.lds: $(LDSCRIPT)
+		$(CPP) $(CPPFLAGS) $(LDPPFLAGS) -ansi -D__ASSEMBLY__ -DMAKE_INIT_LDS -P - <$^ >$@
+
+$(obj)u-boot-init:	depend \
+		$(SUBDIR_TOOLS) $(OBJS) $(LIBBOARD) $(LIBS) $(obj)u-boot-init.lds
+		$(GEN_UBOOT_INIT)
+
+$(obj)u-boot-init.bin:	$(obj)u-boot-init
+		$(OBJCOPY) -j .initfuncs -O binary $< $@
+
+$(obj)init_seq.c:	depend \
+		$(obj)u-boot-init.bin
+		$(obj)tools/mkinitseq $(obj)u-boot-init.bin $(obj)common/init_seq.c
+
+$(obj)init_seq.o:	depend \
+		$(obj)init_seq.c
+		$(MAKE) BUILD_INIT_SEQ=y -C common all
+else
+INIT_SEQ =
+endif
+
 $(obj)u-boot:	depend \
-		$(SUBDIR_TOOLS) $(OBJS) $(LIBBOARD) $(LIBS) $(LDSCRIPT) $(obj)u-boot.lds
+		$(SUBDIR_TOOLS) $(OBJS) $(LIBBOARD) $(LIBS) $(LDSCRIPT) \
+		$(INIT_SEQ) $(obj)u-boot.lds
 		$(GEN_UBOOT)
 ifeq ($(CONFIG_KALLSYMS),y)
 		smap=`$(call SYSTEM_MAP,u-boot) | \
common/Makefile +2 -0
index d9f10f3..02a4485 100644
@@ -195,6 +195,8 @@ COBJS-y += dlmalloc.o
 COBJS-y += memsize.o
 COBJS-y += stdio.o
 
+# initialisation sequence (second build stage)
+COBJS-$(BUILD_INIT_SEQ) += init_seq.o
 
 COBJS	:= $(sort $(COBJS-y))
 XCOBJS	:= $(sort $(XCOBJS-y))
config.mk +2 -0
index fa33e62..2124b84 100644
@@ -257,6 +257,8 @@ ifneq ($(CONFIG_SYS_TEXT_BASE),)
 LDFLAGS_u-boot += -Ttext $(CONFIG_SYS_TEXT_BASE)
 endif
 
+LDFLAGS_u-boot-init += -T $(obj)u-boot-init.lds $(LDFLAGS_FINAL)
+
 LDFLAGS_u-boot-spl += -T $(obj)u-boot-spl.lds $(LDFLAGS_FINAL)
 ifneq ($(CONFIG_SPL_TEXT_BASE),)
 LDFLAGS_u-boot-spl += -Ttext $(CONFIG_SPL_TEXT_BASE)
doc/README.INIT_FUNC new file +65 -0

65 lines not on this page · raw message

include/init_func.h new file +37 -0
new file mode 100644
index 0000000..e4366b8
@@ -0,0 +1,37 @@
+#ifndef __INIT_FUNC_H__
+#define __INIT_FUNC_H__
+
+/*
+ * The requirements for any new initalization function is simple: it is
+ * a function with no parameters which returns an integer return code,
+ * where 0 means "continue" and != 0 means "fatal error, hang the system"
+ */
+typedef int (init_fnc_t) (void);
+
+extern init_fnc_t *init_sequence_f[];
+extern init_fnc_t *init_sequence_f_r[];
+extern init_fnc_t *init_sequence_r[];
+
+#ifdef CONFIG_INIT_FUNC
+#include <linux/compiler.h>
+
+#define INIT_FUNC(fn, grp, man_reqs, pre_reqs, pst_reqs) \
+	static const char __init_func_ ## fn[] __used \
+	__attribute__((__section__(".initfuncs"))) = \
+	"(f:" #fn ":" #grp ":" #man_reqs " | " #pre_reqs " | " #pst_reqs ")\n";
+
+#define SKIP_INIT(fn_or_group) \
+	static const char __skip_init_ ## fn_or_group[] __used \
+	__attribute__((__section__(".initfuncs"))) = \
+	"(s:" #fn_or_group ")\n";
+
+#define REPLACE_INIT(old_fn_or_group, new_fn_or_group) \
+	static const char __replace_init_ ## old_fn_or_group[] __used \
+	__attribute__((__section__(".initfuncs"))) = \
+	"(r:" #old_fn_or_group ":" #new_fn_or_group ")\n";
+#else
+#define INIT_FUNC(fn, group, man_reqs, pre_reqs, post_reqs)
+#define SKIP_INIT(fn_or_group)
+#define REPLACE_INIT(old_fn_or_group, new_fn_or_group)
+#endif
+#endif /* !__INIT_FUNC_H__ */
tools/Makefile +6 -0
index 8993fdd..e6ba6ef 100644
@@ -68,6 +68,7 @@ BIN_FILES-$(CONFIG_CMD_LOADS) += img2srec$(SFX)
 BIN_FILES-$(CONFIG_XWAY_SWAP_BYTES) += xway-swap-bytes$(SFX)
 BIN_FILES-y += mkenvimage$(SFX)
 BIN_FILES-y += mkimage$(SFX)
+BIN_FILES-$(CONFIG_INIT_FUNC) += mkinitseq$(SFX)
 BIN_FILES-$(CONFIG_SMDK5250) += mksmdk5250spl$(SFX)
 BIN_FILES-$(CONFIG_MX28) += mxsboot$(SFX)
 BIN_FILES-$(CONFIG_NETCONSOLE) += ncb$(SFX)
@@ -95,6 +96,7 @@ NOPED_OBJ_FILES-y += imximage.o
 NOPED_OBJ_FILES-y += omapimage.o
 NOPED_OBJ_FILES-y += mkenvimage.o
 NOPED_OBJ_FILES-y += mkimage.o
+NOPED_OBJ_FILES-$(CONFIG_INIT_FUNC) += mkinitseq.o
 OBJ_FILES-$(CONFIG_SMDK5250) += mkexynosspl.o
 OBJ_FILES-$(CONFIG_MX28) += mxsboot.o
 OBJ_FILES-$(CONFIG_NETCONSOLE) += ncb.o
@@ -215,6 +217,10 @@ $(obj)mkimage$(SFX):	$(obj)aisimage.o \
 	$(HOSTCC) $(HOSTCFLAGS) $(HOSTLDFLAGS) -o $@ $^
 	$(HOSTSTRIP) $@
 
+$(obj)mkinitseq$(SFX):	$(obj)mkinitseq.o
+	$(HOSTCC) $(HOSTCFLAGS) $(HOSTLDFLAGS) -o $@ $^
+	$(HOSTSTRIP) $@
+
 $(obj)mk$(BOARD)spl$(SFX):	$(obj)mkexynosspl.o
 	$(HOSTCC) $(HOSTCFLAGS) $(HOSTLDFLAGS) -o $@ $^
 	$(HOSTSTRIP) $@
tools/mkinitseq.c new file +1512 -0

1,512 lines not on this page · raw message

Signature · 1 line
1.7.7.6
#2 lore raw

PATCH 1/3 init_func: Add fundamental framework

Dear Graeme Russ,

Quoted 15 lines
---
 Makefile             |   34 ++-
 common/Makefile      |    2 +
 config.mk            |    2 +
 doc/README.INIT_FUNC |   65 +++
 include/init_func.h  |   37 ++
 tools/Makefile       |    6 +
 tools/mkinitseq.c    | 1512
++++++++++++++++++++++++++++++++++++++++++++++++++ 7 files changed, 1657
insertions(+), 1 deletions(-)
 create mode 100644 doc/README.INIT_FUNC
 create mode 100644 include/init_func.h
 create mode 100644 tools/mkinitseq.c

[...]

Quoted 7 lines
doc/README.INIT_FUNC new file +1 -0
@@ -0,0 +1,65 @@
+The INIT_FUNC macro allows initialisation functions (i.e. functions which

are +executed before the main loop) to be easily added to the init sequence +

doc/README.INIT_FUNC new file +1 -0
+

doc/README.INIT_FUNC line 14

Quoted 10 lines
doc/README.INIT_FUNC new file +10 -0
+Specifying an Initialisation Function and is Dependencies
+---------------------------------------------------------
+The format of the INIT_FUNC macro is:
+
+INIT_FUNC(fn, grp, man_reqs, pre_reqs, pst_reqs)
+
+fn is the name of the init function to call. This function must have the
+following prototype:
+
+int foo(void);

What if I want to pass some data to such a func ? Clearly, I can think of this being doable, but extra hard.

Quoted 19 lines
doc/README.INIT_FUNC new file +2 -0
+
+Each init function must return 0 to indicate success - any other return

value +indicates failure and the init sequence will stop

doc/README.INIT_FUNC new file +3 -0
+
+grp is the name of the group that the init function belongs to. grp may be
+the same as fn for any individual init function, but between init

functions, +fn and grp must be unique.

doc/README.INIT_FUNC new file +2 -0
+
+The purpose of groups is to allow functions to be grouped together so

other +functions can specify the group as a whole as a dependency rather than having +to list every function in the group in the dependency list

doc/README.INIT_FUNC new file +2 -0
+
+man_reqs is a space seperated list of functions or groups that MUST exist

and +MUST run BEFORE fn

doc/README.INIT_FUNC new file +2 -0
+
+pre_reqs is a space seperated list of functions or groups that MAY exist

and +(if they do) MUST run BEFORE fn

doc/README.INIT_FUNC new file +2 -0
+
+pst_reqs is a space seperated list of functions or groups that MAY exist
Quoted 1 line

and +(if they do) MUST run AFTER fn

What's the point? Can't you create a kind of proxy object that the pst_reqs will have as a pre_req ?

Maybe you should create this:

INIT_FUNC(fn, grp, prereqs, postreqs) and for each function from prereqs and postreqs, specify per-function attributes via the GCC __attribute__(()) directive, like if the function must run before something or may run before something etc?

Quoted 13 lines
doc/README.INIT_FUNC new file +4 -0
+
+Skipping or Replacing a Function or Group
+-----------------------------------------
+Occassionally, a board may provide a completely seperate implementation

for +an initialisation function that is provided in the common arch, SoC or +common code.

doc/README.INIT_FUNC new file +4 -0
+
+SKIP_INIT(fn_or_group)
+
+After the initialisation function dependencies are calculated, all

functions +and groups listed in any SKIP_INITs are removed - This may result in +dependent functions being removed - It is up to the board code developer +to ensure suitable replacements are in place

doc/README.INIT_FUNC line 54

Quoted 7 lines
doc/README.INIT_FUNC new file +7 -0
+
+REPLACE_INIT(old_fn_or_group, new_fn_or_group)
+
+Like SKIP_INIT but replaces on function with another (or one group with
+another)
+
+Example: In the SoC code yoy may have

Yoy :)

tools/mkinitseq.c

Quoted 91 lines
doc/README.INIT_FUNC new file +11 -0
+
+INIT_FUNC(init_cpu_f, RESET, , , );
+
+In the board code, you may want a slightly tweaked version, so you might
+have:
+
+int my_new_init_cpu_f(void)
+{
+	...
+}
+REPLACE_INIT(init_cpu_f, my_new_init_cpu_f);
include/init_func.h new file +37 -0
new file mode 100644
index 0000000..e4366b8
@@ -0,0 +1,37 @@
+#ifndef __INIT_FUNC_H__
+#define __INIT_FUNC_H__
+
+/*
+ * The requirements for any new initalization function is simple: it is
+ * a function with no parameters which returns an integer return code,
+ * where 0 means "continue" and != 0 means "fatal error, hang the system"
+ */
+typedef int (init_fnc_t) (void);
+
+extern init_fnc_t *init_sequence_f[];
+extern init_fnc_t *init_sequence_f_r[];
+extern init_fnc_t *init_sequence_r[];
+
+#ifdef CONFIG_INIT_FUNC
+#include <linux/compiler.h>
+
+#define INIT_FUNC(fn, grp, man_reqs, pre_reqs, pst_reqs) \
+	static const char __init_func_ ## fn[] __used \
+	__attribute__((__section__(".initfuncs"))) = \
+	"(f:" #fn ":" #grp ":" #man_reqs " | " #pre_reqs " | " #pst_reqs ")\n";
+
+#define SKIP_INIT(fn_or_group) \
+	static const char __skip_init_ ## fn_or_group[] __used \
+	__attribute__((__section__(".initfuncs"))) = \
+	"(s:" #fn_or_group ")\n";
+
+#define REPLACE_INIT(old_fn_or_group, new_fn_or_group) \
+	static const char __replace_init_ ## old_fn_or_group[] __used \
+	__attribute__((__section__(".initfuncs"))) = \
+	"(r:" #old_fn_or_group ":" #new_fn_or_group ")\n";
+#else
+#define INIT_FUNC(fn, group, man_reqs, pre_reqs, post_reqs)
+#define SKIP_INIT(fn_or_group)
+#define REPLACE_INIT(old_fn_or_group, new_fn_or_group)
+#endif
+#endif /* !__INIT_FUNC_H__ */
tools/Makefile +6 -0
index 8993fdd..e6ba6ef 100644
@@ -68,6 +68,7 @@ BIN_FILES-$(CONFIG_CMD_LOADS) += img2srec$(SFX)
 BIN_FILES-$(CONFIG_XWAY_SWAP_BYTES) += xway-swap-bytes$(SFX)
 BIN_FILES-y += mkenvimage$(SFX)
 BIN_FILES-y += mkimage$(SFX)
+BIN_FILES-$(CONFIG_INIT_FUNC) += mkinitseq$(SFX)
 BIN_FILES-$(CONFIG_SMDK5250) += mksmdk5250spl$(SFX)
 BIN_FILES-$(CONFIG_MX28) += mxsboot$(SFX)
 BIN_FILES-$(CONFIG_NETCONSOLE) += ncb$(SFX)
@@ -95,6 +96,7 @@ NOPED_OBJ_FILES-y += imximage.o
 NOPED_OBJ_FILES-y += omapimage.o
 NOPED_OBJ_FILES-y += mkenvimage.o
 NOPED_OBJ_FILES-y += mkimage.o
+NOPED_OBJ_FILES-$(CONFIG_INIT_FUNC) += mkinitseq.o
 OBJ_FILES-$(CONFIG_SMDK5250) += mkexynosspl.o
 OBJ_FILES-$(CONFIG_MX28) += mxsboot.o
 OBJ_FILES-$(CONFIG_NETCONSOLE) += ncb.o
@@ -215,6 +217,10 @@ $(obj)mkimage$(SFX):	$(obj)aisimage.o \
 	$(HOSTCC) $(HOSTCFLAGS) $(HOSTLDFLAGS) -o $@ $^
 	$(HOSTSTRIP) $@
 
+$(obj)mkinitseq$(SFX):	$(obj)mkinitseq.o
+	$(HOSTCC) $(HOSTCFLAGS) $(HOSTLDFLAGS) -o $@ $^
+	$(HOSTSTRIP) $@
+
 $(obj)mk$(BOARD)spl$(SFX):	$(obj)mkexynosspl.o
 	$(HOSTCC) $(HOSTCFLAGS) $(HOSTLDFLAGS) -o $@ $^
 	$(HOSTSTRIP) $@
tools/mkinitseq.c new file +0 -0
new file mode 100644
index 0000000..b150de4
@@ -0,0 +1,1512 @@

Ok, this is the worst part. I think this could be reimplemented in shell ;-)

Quoted 27 lines
tools/mkinitseq.c new file +22 -0
+/*
+ * (C) Copyright 2012
+ * Graeme Russ <[email protected]>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License as
+ * published by the Free Software Foundation; either version 2 of
+ * the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston,
+ * MA 02111-1307 USA
+ */
+
+/**
+ * container_of - cast a member of a structure out to the containing

structure + * @ptr: the pointer to the member.
+ * @type: the type of the container struct this is embedded in.
+ * @member: the name of the member within the struct.
+ *
+ */

tools/mkinitseq.c line 30

Quoted 3 lines
tools/mkinitseq.c new file +3 -0
+#define container_of(ptr, type, member) ({			\
+	const typeof( ((type *)0)->member ) *__mptr = (ptr);	\
+	(type *)( (char *)__mptr - offsetof(type,member) );})

Don't we already have this defined in uboot ?

[...]

#3 lore raw

PATCH 1/3 init_func: Add fundamental framework

Hi Marek,

Thanks for taking a look

Quoted 9 lines
Quoted 9 lines

+The INIT_FUNC macro allows initialisation functions (i.e. functions which are +executed before the main loop) to be easily added to the init sequence +

doc/README.INIT_FUNC new file +6 -0
+
+Specifying an Initialisation Function and is Dependencies
+---------------------------------------------------------
+The format of the INIT_FUNC macro is:
+
+INIT_FUNC(fn, grp, man_reqs, pre_reqs, pst_reqs)
Quoted 8 lines

doc/README.INIT_FUNC line 14

Quoted 5 lines
doc/README.INIT_FUNC new file +5 -0
+
+fn is the name of the init function to call. This function must have the
+following prototype:
+
+int foo(void);

What if I want to pass some data to such a func ? Clearly, I can think of this being doable, but extra hard.

The idea is that no changes are being made to the existing init methodology.

Quoted 24 lines
Quoted 19 lines
(quoted diff) +2 -0
+
+Each init function must return 0 to indicate success - any other return

value +indicates failure and the init sequence will stop

doc/README.INIT_FUNC new file +3 -0
+
+grp is the name of the group that the init function belongs to. grp may be
+the same as fn for any individual init function, but between init

functions, +fn and grp must be unique.

(quoted diff) +2 -0
+
+The purpose of groups is to allow functions to be grouped together so

other +functions can specify the group as a whole as a dependency rather than having +to list every function in the group in the dependency list

(quoted diff) +2 -0
+
+man_reqs is a space seperated list of functions or groups that MUST exist

and +MUST run BEFORE fn

(quoted diff) +2 -0
+
+pre_reqs is a space seperated list of functions or groups that MAY exist

and +(if they do) MUST run BEFORE fn

(quoted diff) +2 -0
+
+pst_reqs is a space seperated list of functions or groups that MAY exist
Quoted 1 line

and +(if they do) MUST run AFTER fn

What's the point? Can't you create a kind of proxy object that the pst_reqs will have as a pre_req ?

Quoted 6 lines

Maybe you should create this:

INIT_FUNC(fn, grp, prereqs, postreqs) and for each function from prereqs and postreqs, specify per-function attributes via the GCC __attribute__(()) directive, like if the function must run before something or may run before something etc?

Eep, I would like to see that implemented - I'm sure it would be 'interesting'

The way INIT_FUNC and friends work is to simply build a list of static strings (which we just happen to shove in a seperate section so they can be filtered in/out based on the link stage)

Quoted 14 lines
Quoted 14 lines
doc/README.INIT_FUNC new file +4 -0
+
+Skipping or Replacing a Function or Group
+-----------------------------------------
+Occassionally, a board may provide a completely seperate implementation

for +an initialisation function that is provided in the common arch, SoC or +common code.

doc/README.INIT_FUNC new file +4 -0
+
+SKIP_INIT(fn_or_group)
+
+After the initialisation function dependencies are calculated, all

functions +and groups listed in any SKIP_INITs are removed - This may result in +dependent functions being removed - It is up to the board code developer +to ensure suitable replacements are in place
+

Quoted 8 lines

doc/README.INIT_FUNC line 54

Quoted 6 lines
doc/README.INIT_FUNC new file +6 -0
+REPLACE_INIT(old_fn_or_group, new_fn_or_group)
+
+Like SKIP_INIT but replaces on function with another (or one group with
+another)
+
+Example: In the SoC code yoy may have

Yoy :)

Yes. The ultimate goal is to remove a heap of '#ifdef mess' and delegate the selection of initialiasation functions to where is is most appropriate. CPU init in ARCH code with board specific init in board code with the ARCH code 100% unaware of what the board is doing. This will also make it a lot easier for vendors to implement multi-arch solutions :)

Quoted 11 lines
Quoted 11 lines
doc/README.INIT_FUNC new file +11 -0
+
+INIT_FUNC(init_cpu_f, RESET, , , );
+
+In the board code, you may want a slightly tweaked version, so you might
+have:
+
+int my_new_init_cpu_f(void)
+{
+ ? ? ...
+}
+REPLACE_INIT(init_cpu_f, my_new_init_cpu_f);

[snip]

Quoted 8 lines

tools/mkinitseq.c

Quoted 6 lines
tools/mkinitseq.c new file +0 -0
new file mode 100644
index 0000000..b150de4
@@ -0,0 +1,1512 @@

Ok, this is the worst part. I think this could be reimplemented in shell ;-)

Be my guest :) - Have a really good look at what is happening behind the scenes - Lots of list processing and checking. In particular, the processing of the REPLACE and SKIP macros would be particularly nasty.

I have no problem with this being done in shell code. BUT if I have to do so in order for it to be accepted, then I'll bin this patch series right now. I'm more than happy to integrate somebody elses shell-based tool into the series. Sorry to be so blunt, but I do not have the time or energy to re-implement this myself.

My argument is that this is a black-box tool like gcc/ld. Once we prove it to do what it does correctly, we can 'leave it alone'(tm) and test cases are fairly trivial. You can even mock-up the input file in vi :)

And is a shell based implementation going to be any easier to understand, modify, debug, etc?

Quoted 24 lines
Quoted 24 lines
tools/mkinitseq.c new file +22 -0
+/*
+ * (C) Copyright 2012
+ * Graeme Russ <[email protected]>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License as
+ * published by the Free Software Foundation; either version 2 of
+ * the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. ?See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston,
+ * MA 02111-1307 USA
+ */
+
+/**
+ * container_of - cast a member of a structure out to the containing

structure + * @ptr: ? the pointer to the member.
+ * @type: ? ?the type of the container struct this is embedded in.

Quoted 8 lines
Quoted 6 lines

+ * @member: ?the name of the member within the struct.
+ *
+ */
+#define container_of(ptr, type, member) ({ ? ? ? ? ? ? ? ? ? \
+ ? ? const typeof( ((type *)0)->member ) *__mptr = (ptr); ? ?\
+ ? ? (type *)( (char *)__mptr - offsetof(type,member) );})

Don't we already have this defined in uboot ?

mkinitseq is host-code so it only brings in host headers (not U-Boot headers) - If you don't have Linux kernel headers installed, you can't get this macro. Actually, even though I do have them, I had a hard time getting this macro in, so I gave up - Any hints?

As an aside, I wonder if the kernel headers are required for the list macros?

Regards,

Graeme

#4 lore raw

PATCH 1/3 init_func: Add fundamental framework

Dear Graeme Russ,

Quoted 22 lines

Hi Marek,

Thanks for taking a look

Quoted 9 lines
Quoted 9 lines

+The INIT_FUNC macro allows initialisation functions (i.e. functions which are +executed before the main loop) to be easily added to the init sequence +

doc/README.INIT_FUNC new file +6 -0
+
+Specifying an Initialisation Function and is Dependencies
+---------------------------------------------------------
+The format of the INIT_FUNC macro is:
+
+INIT_FUNC(fn, grp, man_reqs, pre_reqs, pst_reqs)
Quoted 8 lines

Review

Quoted 5 lines
(quoted diff) +2 -0
+
+fn is the name of the init function to call. This function must have

the +following prototype:

(quoted diff) +2 -0
+
+int foo(void);

What if I want to pass some data to such a func ? Clearly, I can think of this being doable, but extra hard.

Quoted 2 lines

The idea is that no changes are being made to the existing init methodology.

Well ... what if I want to pass eg. pdata?

Quoted 27 lines
Quoted 27 lines
Quoted 20 lines
(quoted diff) +2 -0
+
+Each init function must return 0 to indicate success - any other return

value +indicates failure and the init sequence will stop

(quoted diff) +2 -0
+
+grp is the name of the group that the init function belongs to. grp may

be +the same as fn for any individual init function, but between init functions, +fn and grp must be unique.

(quoted diff) +2 -0
+
+The purpose of groups is to allow functions to be grouped together so

other +functions can specify the group as a whole as a dependency rather than having +to list every function in the group in the dependency list

(quoted diff) +2 -0
+
+man_reqs is a space seperated list of functions or groups that MUST

exist and +MUST run BEFORE fn

(quoted diff) +2 -0
+
+pre_reqs is a space seperated list of functions or groups that MAY

exist and +(if they do) MUST run BEFORE fn

(quoted diff) +2 -0
+
+pst_reqs is a space seperated list of functions or groups that MAY
Quoted 1 line

exist and +(if they do) MUST run AFTER fn

What's the point? Can't you create a kind of proxy object that the pst_reqs will have as a pre_req ?

Maybe you should create this:

Quoted 7 lines
Quoted 4 lines

INIT_FUNC(fn, grp, prereqs, postreqs) and for each function from prereqs and postreqs, specify per-function attributes via the GCC __attribute__(()) directive, like if the function must run before something or may run before something etc?

Eep, I would like to see that implemented - I'm sure it would be 'interesting'

Pervy and sadistic at least :-)

Quoted 3 lines

The way INIT_FUNC and friends work is to simply build a list of static strings (which we just happen to shove in a seperate section so they can be filtered in/out based on the link stage)

Yes, I got the understanding of it. I was more bothered by the man_reqs and pre_reqs, what if we soon need functions that are executed under another weird condition? And if you look at it the other way -- if you need function that's executed conditionally, why not wrap the condition into the pre_req (or some proxy object, whatever).

Quoted 24 lines
Quoted 15 lines
Quoted 15 lines
doc/README.INIT_FUNC new file +4 -0
+
+Skipping or Replacing a Function or Group
+-----------------------------------------
+Occassionally, a board may provide a completely seperate implementation

for +an initialisation function that is provided in the common arch, SoC or +common code.

doc/README.INIT_FUNC new file +4 -0
+
+SKIP_INIT(fn_or_group)
+
+After the initialisation function dependencies are calculated, all

functions +and groups listed in any SKIP_INITs are removed - This may result in +dependent functions being removed - It is up to the board code developer +to ensure suitable replacements are in place +

Quoted 8 lines

doc/README.INIT_FUNC line 54

Quoted 6 lines
doc/README.INIT_FUNC new file +6 -0
+REPLACE_INIT(old_fn_or_group, new_fn_or_group)
+
+Like SKIP_INIT but replaces on function with another (or one group with
+another)
+
+Example: In the SoC code yoy may have

Yoy :)

Quoted 5 lines

Yes. The ultimate goal is to remove a heap of '#ifdef mess' and delegate the selection of initialiasation functions to where is is most appropriate. CPU init in ARCH code with board specific init in board code with the ARCH code 100% unaware of what the board is doing. This will also make it a lot easier for vendors to implement multi-arch solutions :)

Correct, alongside kbuild and driver model, we'll make an awesome bootloader. But fix that s/yoy/you/ please ;-)

Quoted 24 lines
Quoted 11 lines
Quoted 11 lines
doc/README.INIT_FUNC new file +4 -0
+
+INIT_FUNC(init_cpu_f, RESET, , , );
+
+In the board code, you may want a slightly tweaked version, so you

might +have:

doc/README.INIT_FUNC new file +6 -0
+
+int my_new_init_cpu_f(void)
+{
+     ...
+}
+REPLACE_INIT(init_cpu_f, my_new_init_cpu_f);

[snip]

Quoted 9 lines

tools/mkinitseq.c

Quoted 6 lines
tools/mkinitseq.c new file +0 -0
new file mode 100644
index 0000000..b150de4
@@ -0,0 +1,1512 @@

Ok, this is the worst part. I think this could be reimplemented in shell ;-)

Quoted 3 lines

Be my guest :) - Have a really good look at what is happening behind the scenes - Lots of list processing and checking. In particular, the processing of the REPLACE and SKIP macros would be particularly nasty.

That's what sed can do for you, can't it ?

Quoted 5 lines

I have no problem with this being done in shell code. BUT if I have to do so in order for it to be accepted, then I'll bin this patch series right now. I'm more than happy to integrate somebody elses shell-based tool into the series. Sorry to be so blunt, but I do not have the time or energy to re-implement this myself.

Ooh, I smell flame-fuel :-) I'd certainly prefer to see this in shell, since I believe the substitutions and reordering of text might be easier there.

Quoted 3 lines

My argument is that this is a black-box tool like gcc/ld. Once we prove it to do what it does correctly, we can 'leave it alone'(tm) and test cases are fairly trivial. You can even mock-up the input file in vi :)

Ewwwwww ... like gcc/ld ... or maybe like Windows (TM)(C)(R)(?) :-D

Quoted 2 lines

And is a shell based implementation going to be any easier to understand, modify, debug, etc?

I wonder :-)

Quoted 33 lines
Quoted 24 lines
Quoted 24 lines
tools/mkinitseq.c new file +22 -0
+/*
+ * (C) Copyright 2012
+ * Graeme Russ <[email protected]>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License as
+ * published by the Free Software Foundation; either version 2 of
+ * the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston,
+ * MA 02111-1307 USA
+ */
+
+/**
+ * container_of - cast a member of a structure out to the containing

structure + * @ptr: the pointer to the member.
+ * @type: the type of the container struct this is embedded in.

Quoted 8 lines

tools/mkinitseq.c line 30

Quoted 6 lines

+ * @member: the name of the member within the struct.
+ *
+ */

tools/mkinitseq.c new file +3 -0
+#define container_of(ptr, type, member) ({                   \
+     const typeof( ((type *)0)->member ) *__mptr = (ptr);    \
+     (type *)( (char *)__mptr - offsetof(type,member) );})

Don't we already have this defined in uboot ?

Quoted 4 lines

mkinitseq is host-code so it only brings in host headers (not U-Boot headers) - If you don't have Linux kernel headers installed, you can't get this macro. Actually, even though I do have them, I had a hard time getting this macro in, so I gave up - Any hints?

Reimplement this thing in shell :-)

Quoted 3 lines

As an aside, I wonder if the kernel headers are required for the list macros?

Nope.

Quoted 4 lines

Regards,

Graeme

#5 lore raw

PATCH 1/3 init_func: Add fundamental framework

Hi Marek,

Quoted 24 lines · Marek Vasut

Dear Graeme Russ,

Quoted 22 lines

Hi Marek,

Thanks for taking a look

Quoted 17 lines
Quoted 12 lines

+The INIT_FUNC macro allows initialisation functions (i.e. functions which are +executed before the main loop) to be easily added to the init sequence +

doc/README.INIT_FUNC new file +8 -0
+
+Specifying an Initialisation Function and is Dependencies
+---------------------------------------------------------
+The format of the INIT_FUNC macro is:
+
+INIT_FUNC(fn, grp, man_reqs, pre_reqs, pst_reqs)
+
+fn is the name of the init function to call. This function must have

the +following prototype:

Review

Quoted 2 lines
(quoted diff) +2 -0
+
+int foo(void);

What if I want to pass some data to such a func ? Clearly, I can think of this being doable, but extra hard.

Quoted 4 lines
Quoted 2 lines

The idea is that no changes are being made to the existing init methodology.

Well ... what if I want to pass eg. pdata?

Do what we do now - Global Data is the only way to pass data between initialisation functions. Sure, this _might_ change in the future (but I doubt it) but we can deal with that later

Quoted 40 lines
Quoted 26 lines
Quoted 26 lines
Quoted 19 lines
(quoted diff) +2 -0
+
+Each init function must return 0 to indicate success - any other return

value +indicates failure and the init sequence will stop

(quoted diff) +2 -0
+
+grp is the name of the group that the init function belongs to. grp may

be +the same as fn for any individual init function, but between init functions, +fn and grp must be unique.

(quoted diff) +2 -0
+
+The purpose of groups is to allow functions to be grouped together so

other +functions can specify the group as a whole as a dependency rather than having +to list every function in the group in the dependency list

(quoted diff) +2 -0
+
+man_reqs is a space seperated list of functions or groups that MUST

exist and +MUST run BEFORE fn

(quoted diff) +2 -0
+
+pre_reqs is a space seperated list of functions or groups that MAY

exist and +(if they do) MUST run BEFORE fn

(quoted diff) +2 -0
+
+pst_reqs is a space seperated list of functions or groups that MAY
Quoted 1 line

exist and +(if they do) MUST run AFTER fn

What's the point? Can't you create a kind of proxy object that the pst_reqs will have as a pre_req ?

Maybe you should create this:

Quoted 7 lines
Quoted 4 lines

INIT_FUNC(fn, grp, prereqs, postreqs) and for each function from prereqs and postreqs, specify per-function attributes via the GCC __attribute__(()) directive, like if the function must run before something or may run before something etc?

Eep, I would like to see that implemented - I'm sure it would be 'interesting'

Pervy and sadistic at least :-)

Quoted 3 lines

The way INIT_FUNC and friends work is to simply build a list of static strings (which we just happen to shove in a seperate section so they can be filtered in/out based on the link stage)

Quoted 5 lines

Yes, I got the understanding of it. I was more bothered by the man_reqs and pre_reqs, what if we soon need functions that are executed under another weird condition? And if you look at it the other way -- if you need function that's executed conditionally, why not wrap the condition into the pre_req (or some proxy object, whatever).

It's not about being executed conditionally - It's about 'I need to be initialised after 'X', but if 'X' does not exist, I can still initialise'

Network is one example - If you have a PCI network adapter, it must be intialised after PCI. But if the adapter is not PCI, you still want to be able to initialise it

Timer is another - Hopefully the timer infrastructure will soon be an arch independent API. But each arch (and potentially down to the board level) may have a role to play in getting the timer infrastucture running (FPGA, PLL, PIT intialiastion). So potentially, the common driver code would be intialised with:

INIT_FUNC(timer_api, timer, arch_timer, board_timer, );

So the arch_timer function MUST exist and it MUST run before timer_api. The board specific timer initialisation is optional. If the board has timer related code, it MUST run before timer_api but timer_api can still run if board_timer does not exist (i.e. all the timer init done at the arch level)

Now depending on the arch and board, board_timer may be either:

INIT_FUNC(board_timer, timer, arch_timer, , timer_api);

i.e. arch -> board -> api

or

INIT_FUNC(board_timer, timer, , , arch_timer timer_api);

i.e. board -> arch -> api

NOTE: In both of the above example, including timer_api as a post-req is redundant. The timer_api INIT_FUNC definition already mandated the order. This is not a problem and including or excluding it makes no difference to the output. Of course, including timer_api in the above examples means we could have done:

INIT_FUNC(timer_api, timer, arch_timer, , );

But adding the redundant references makes it clearer when you are looking at that bit of code. The 'black-box' tool will warn you if you created a circular reference and will print out what the init sequence is that created the circular reference.

Now, here is where the fun begins :) - Lets say you need timer support early, but you can't get the low level infrastructure for hi-res timers running until later...

board/vendor_me/common/timer.c
INIT_FUNC(me_timer_init, me_late_timer, me_pit_init, , ); INIT_FUNC(me_pit_init, me_late_timer, , fpga_init pll_init, me_timer_init);

(again, me_late_timer in the second INIT_FUNC is redundant)

board/vendor_me/board_a/timer.c INIT_FUNC(fpga_init, me_late_timer, , , me_pit_init);

board/vendor_me/board_b/timer.c INIT_FUNC(pll_init, me_late_timer, , , me_pit_init);

(again, me_pit_init in these INIT_FUNCs is redundant)

board/vendor_me/board_c/timer.c
/*
* This board has the PIT hard-wired to a crystal oscillator so there
* is now board_level init function - So we can actually initialise
* the PIT earlier :)
*/
int me_pit_early_init(void)

{
        return me_pit_init();
}

INIT_FUNC(me_pit_early_init, timer, arch_timer, , );

/*
* Don't need any late timer init - This will skip both me_timer_init and
* me_pit_init as they have been defined in the me_late_timer group */
INIT_SKIP(me_late_timer);

The idea is that me_late_timer() will perform some tweak to the timer API to redirect it to the hi-res clock source. The hi-res clock source is driven by a PIT common to all the boards manufacture by this vendor. But the PIT has a different raw clock source depending on a particular board Board A has an FPGA, board B has a PLL and board C uses a hard-wired crystal oscillator

Quoted 23 lines
Quoted 23 lines
Quoted 22 lines
Quoted 13 lines
doc/README.INIT_FUNC new file +4 -0
+
+Skipping or Replacing a Function or Group
+-----------------------------------------
+Occassionally, a board may provide a completely seperate implementation

for +an initialisation function that is provided in the common arch, SoC or +common code.

doc/README.INIT_FUNC new file +4 -0
+
+SKIP_INIT(fn_or_group)
+
+After the initialisation function dependencies are calculated, all

functions +and groups listed in any SKIP_INITs are removed - This may result in +dependent functions being removed - It is up to the board code developer +to ensure suitable replacements are in place

doc/README.INIT_FUNC line 54

Quoted 7 lines
doc/README.INIT_FUNC new file +7 -0
+
+REPLACE_INIT(old_fn_or_group, new_fn_or_group)
+
+Like SKIP_INIT but replaces on function with another (or one group with
+another)
+
+Example: In the SoC code yoy may have

Yoy :)

Quoted 8 lines
Quoted 5 lines

Yes. The ultimate goal is to remove a heap of '#ifdef mess' and delegate the selection of initialiasation functions to where is is most appropriate. CPU init in ARCH code with board specific init in board code with the ARCH code 100% unaware of what the board is doing. This will also make it a lot easier for vendors to implement multi-arch solutions :)

Correct, alongside kbuild and driver model, we'll make an awesome bootloader. But fix that s/yoy/you/ please ;-)

Ah, I see now... Will do

Quoted 24 lines
Quoted 24 lines
Quoted 11 lines
Quoted 11 lines
doc/README.INIT_FUNC new file +4 -0
+
+INIT_FUNC(init_cpu_f, RESET, , , );
+
+In the board code, you may want a slightly tweaked version, so you

might +have:

doc/README.INIT_FUNC new file +6 -0
+
+int my_new_init_cpu_f(void)
+{
+ ? ? ...
+}
+REPLACE_INIT(init_cpu_f, my_new_init_cpu_f);

[snip]

Quoted 9 lines

tools/mkinitseq.c

Quoted 6 lines
tools/mkinitseq.c new file +0 -0
new file mode 100644
index 0000000..b150de4
@@ -0,0 +1,1512 @@

Ok, this is the worst part. I think this could be reimplemented in shell ;-)

Quoted 5 lines
Quoted 3 lines

Be my guest :) - Have a really good look at what is happening behind the scenes - Lots of list processing and checking. In particular, the processing of the REPLACE and SKIP macros would be particularly nasty.

That's what sed can do for you, can't it ?

So below - If you can show me (i.e. write the whole thing in shell) I might include it

Quoted 8 lines
Quoted 5 lines

I have no problem with this being done in shell code. BUT if I have to do so in order for it to be accepted, then I'll bin this patch series right now. I'm more than happy to integrate somebody elses shell-based tool into the series. Sorry to be so blunt, but I do not have the time or energy to re-implement this myself.

Ooh, I smell flame-fuel :-) I'd certainly prefer to see this in shell, since I believe the substitutions and reordering of text might be easier there.

See above - be my guest ;)

Quoted 5 lines
Quoted 3 lines

My argument is that this is a black-box tool like gcc/ld. Once we prove it to do what it does correctly, we can 'leave it alone'(tm) and test cases are fairly trivial. You can even mock-up the input file in vi :)

Ewwwwww ... like gcc/ld ... or maybe like Windows (TM)(C)(R)(?) :-D

No, like FLOSS gcc/ld :P

Quoted 4 lines
Quoted 2 lines

And is a shell based implementation going to be any easier to understand, modify, debug, etc?

I wonder :-)

Well, if someone manages to do it in shell, we will see - Take a look at main() - I have been very carefull to layout the way the tool operates in a very linear way but ther lists are global, so sharing data between the shell stages will be a pain (pipes?)

Quoted 33 lines
Quoted 33 lines
Quoted 32 lines
Quoted 22 lines
tools/mkinitseq.c new file +22 -0
+/*
+ * (C) Copyright 2012
+ * Graeme Russ <[email protected]>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License as
+ * published by the Free Software Foundation; either version 2 of
+ * the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. ?See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston,
+ * MA 02111-1307 USA
+ */
+
+/**
+ * container_of - cast a member of a structure out to the containing
Quoted 8 lines

structure + * @ptr: ? the pointer to the member.
+ * @type: ? ?the type of the container struct this is embedded in.
+ * @member: ?the name of the member within the struct.
+ *
+ */
+#define container_of(ptr, type, member) ({ ? ? ? ? ? ? ? ? ? \
+ ? ? const typeof( ((type *)0)->member ) *__mptr = (ptr); ? ?\
+ ? ? (type *)( (char *)__mptr - offsetof(type,member) );})

Don't we already have this defined in uboot ?

Quoted 6 lines
Quoted 4 lines

mkinitseq is host-code so it only brings in host headers (not U-Boot headers) - If you don't have Linux kernel headers installed, you can't get this macro. Actually, even though I do have them, I had a hard time getting this macro in, so I gave up - Any hints?

Reimplement this thing in shell :-)

You already know my answer to this ;)

Quoted 4 lines
Quoted 2 lines

As an aside, I wonder if the kernel headers are required for the list macros?

Nope.

Good :)

Regards,

Graeme

#6 lore raw

PATCH 1/3 init_func: Add fundamental framework

Dear Graeme Russ,

Quoted 27 lines

Hi Marek,

Quoted 24 lines · Marek Vasut

Dear Graeme Russ,

Quoted 22 lines

Hi Marek,

Thanks for taking a look

Quoted 17 lines
Quoted 12 lines

+The INIT_FUNC macro allows initialisation functions (i.e. functions which are +executed before the main loop) to be easily added to the init sequence +

doc/README.INIT_FUNC new file +8 -0
+
+Specifying an Initialisation Function and is Dependencies
+---------------------------------------------------------
+The format of the INIT_FUNC macro is:
+
+INIT_FUNC(fn, grp, man_reqs, pre_reqs, pst_reqs)
+
+fn is the name of the init function to call. This function must have

the +following prototype:

Review

Quoted 2 lines
(quoted diff) +2 -0
+
+int foo(void);

What if I want to pass some data to such a func ? Clearly, I can think of this being doable, but extra hard.

Quoted 8 lines
Quoted 4 lines
Quoted 2 lines

The idea is that no changes are being made to the existing init methodology.

Well ... what if I want to pass eg. pdata?

Do what we do now - Global Data is the only way to pass data between initialisation functions. Sure, this _might_ change in the future (but I doubt it) but we can deal with that later

Global data won't die, but they should shrink as much as possible ... so I don't like the idea of using global data that way.

Quoted 47 lines
Quoted 41 lines
Quoted 27 lines
Quoted 27 lines
Quoted 20 lines
(quoted diff) +2 -0
+
+Each init function must return 0 to indicate success - any other

return value +indicates failure and the init sequence will stop

(quoted diff) +2 -0
+
+grp is the name of the group that the init function belongs to. grp

may be +the same as fn for any individual init function, but between init functions, +fn and grp must be unique.

(quoted diff) +2 -0
+
+The purpose of groups is to allow functions to be grouped together

so other +functions can specify the group as a whole as a dependency rather than having +to list every function in the group in the dependency list +
+man_reqs is a space seperated list of functions or groups that MUST exist and +MUST run BEFORE fn

(quoted diff) +2 -0
+
+pre_reqs is a space seperated list of functions or groups that MAY

exist and +(if they do) MUST run BEFORE fn

(quoted diff) +2 -0
+
+pst_reqs is a space seperated list of functions or groups that MAY
Quoted 1 line

exist and +(if they do) MUST run AFTER fn

What's the point? Can't you create a kind of proxy object that the pst_reqs will have as a pre_req ?

Maybe you should create this:

Quoted 7 lines
Quoted 4 lines

INIT_FUNC(fn, grp, prereqs, postreqs) and for each function from prereqs and postreqs, specify per-function attributes via the GCC __attribute__(()) directive, like if the function must run before something or may run before something etc?

Eep, I would like to see that implemented - I'm sure it would be 'interesting'

Pervy and sadistic at least :-)

Quoted 3 lines

The way INIT_FUNC and friends work is to simply build a list of static strings (which we just happen to shove in a seperate section so they can be filtered in/out based on the link stage)

Quoted 5 lines

Yes, I got the understanding of it. I was more bothered by the man_reqs and pre_reqs, what if we soon need functions that are executed under another weird condition? And if you look at it the other way -- if you need function that's executed conditionally, why not wrap the condition into the pre_req (or some proxy object, whatever).

Quoted 6 lines

It's not about being executed conditionally - It's about 'I need to be initialised after 'X', but if 'X' does not exist, I can still initialise'

Network is one example - If you have a PCI network adapter, it must be intialised after PCI. But if the adapter is not PCI, you still want to be able to initialise it

So why don't we make function, that can depend on "OR-group" and/or "AND-group" then? Won't that be more logical? I mean, group where you must execute as much as possible, to be OR-group and group where you must execute everything to be AND-group.

Quoted 6 lines

Timer is another - Hopefully the timer infrastructure will soon be an arch independent API. But each arch (and potentially down to the board level) may have a role to play in getting the timer infrastucture running (FPGA, PLL, PIT intialiastion). So potentially, the common driver code would be intialised with:

Well, I hope to create unified timer api with the DM biz.

Quoted 6 lines

INIT_FUNC(timer_api, timer, arch_timer, board_timer, );

So the arch_timer function MUST exist and it MUST run before timer_api. The board specific timer initialisation is optional. If the board has timer related code, it MUST run before timer_api but timer_api can still run if board_timer does not exist (i.e. all the timer init done at the arch level)

Ok, I see. OR-groups won't fit in here?

Quoted 18 lines

Now depending on the arch and board, board_timer may be either:

INIT_FUNC(board_timer, timer, arch_timer, , timer_api);

i.e. arch -> board -> api

or

INIT_FUNC(board_timer, timer, , , arch_timer timer_api);

i.e. board -> arch -> api

NOTE: In both of the above example, including timer_api as a post-req is redundant. The timer_api INIT_FUNC definition already mandated the order. This is not a problem and including or excluding it makes no difference to the output. Of course, including timer_api in the above examples means we could have done:

Quoted 6 lines

INIT_FUNC(timer_api, timer, arch_timer, , );

But adding the redundant references makes it clearer when you are looking at that bit of code. The 'black-box' tool will warn you if you created a circular reference and will print out what the init sequence is that created the circular reference.

Neat.

Quoted 36 lines

Now, here is where the fun begins :) - Lets say you need timer support early, but you can't get the low level infrastructure for hi-res timers running until later...

board/vendor_me/common/timer.c
INIT_FUNC(me_timer_init, me_late_timer, me_pit_init, , ); INIT_FUNC(me_pit_init, me_late_timer, , fpga_init pll_init, me_timer_init);

(again, me_late_timer in the second INIT_FUNC is redundant)

board/vendor_me/board_a/timer.c INIT_FUNC(fpga_init, me_late_timer, , , me_pit_init);

board/vendor_me/board_b/timer.c INIT_FUNC(pll_init, me_late_timer, , , me_pit_init);

(again, me_pit_init in these INIT_FUNCs is redundant)

board/vendor_me/board_c/timer.c
/*
* This board has the PIT hard-wired to a crystal oscillator so there
* is now board_level init function - So we can actually initialise
* the PIT earlier :)
*/
int me_pit_early_init(void)

{
        return me_pit_init();
}

INIT_FUNC(me_pit_early_init, timer, arch_timer, , );

/*
* Don't need any late timer init - This will skip both me_timer_init and
* me_pit_init as they have been defined in the me_late_timer group */
INIT_SKIP(me_late_timer);

Quoted 6 lines

The idea is that me_late_timer() will perform some tweak to the timer API to redirect it to the hi-res clock source. The hi-res clock source is driven by a PIT common to all the boards manufacture by this vendor. But the PIT has a different raw clock source depending on a particular board Board A has an FPGA, board B has a PLL and board C uses a hard-wired crystal oscillator

I see.

Quoted 59 lines
Quoted 30 lines
Quoted 23 lines
Quoted 14 lines
Quoted 14 lines
doc/README.INIT_FUNC new file +4 -0
+
+Skipping or Replacing a Function or Group
+-----------------------------------------
+Occassionally, a board may provide a completely seperate

implementation for +an initialisation function that is provided in the common arch, SoC or +common code.

doc/README.INIT_FUNC new file +4 -0
+
+SKIP_INIT(fn_or_group)
+
+After the initialisation function dependencies are calculated, all

functions +and groups listed in any SKIP_INITs are removed - This may result in +dependent functions being removed - It is up to the board code developer +to ensure suitable replacements are in place +

Quoted 8 lines

Review

Quoted 6 lines
doc/README.INIT_FUNC new file +3 -0
+REPLACE_INIT(old_fn_or_group, new_fn_or_group)
+
+Like SKIP_INIT but replaces on function with another (or one group

with +another)

(quoted diff) +2 -0
+
+Example: In the SoC code yoy may have

Yoy :)

Quoted 6 lines

Yes. The ultimate goal is to remove a heap of '#ifdef mess' and delegate the selection of initialiasation functions to where is is most appropriate. CPU init in ARCH code with board specific init in board code with the ARCH code 100% unaware of what the board is doing. This will also make it a lot easier for vendors to implement multi-arch solutions :)

Quoted 2 lines

Correct, alongside kbuild and driver model, we'll make an awesome bootloader. But fix that s/yoy/you/ please ;-)

Ah, I see now... Will do

Quoted 24 lines
Quoted 24 lines
Quoted 11 lines
Quoted 11 lines
doc/README.INIT_FUNC new file +4 -0
+
+INIT_FUNC(init_cpu_f, RESET, , , );
+
+In the board code, you may want a slightly tweaked version, so you

might +have:

doc/README.INIT_FUNC new file +6 -0
+
+int my_new_init_cpu_f(void)
+{
+     ...
+}
+REPLACE_INIT(init_cpu_f, my_new_init_cpu_f);

[snip]

Quoted 9 lines

tools/mkinitseq.c

Quoted 6 lines
tools/mkinitseq.c new file +0 -0
new file mode 100644
index 0000000..b150de4
@@ -0,0 +1,1512 @@

Ok, this is the worst part. I think this could be reimplemented in shell ;-)

Quoted 8 lines
Quoted 5 lines
Quoted 3 lines

Be my guest :) - Have a really good look at what is happening behind the scenes - Lots of list processing and checking. In particular, the processing of the REPLACE and SKIP macros would be particularly nasty.

That's what sed can do for you, can't it ?

So below - If you can show me (i.e. write the whole thing in shell) I might include it

I'll think about it ;-)

Quoted 12 lines
Quoted 9 lines
Quoted 5 lines

I have no problem with this being done in shell code. BUT if I have to do so in order for it to be accepted, then I'll bin this patch series right now. I'm more than happy to integrate somebody elses shell-based tool into the series. Sorry to be so blunt, but I do not have the time or energy to re-implement this myself.

Ooh, I smell flame-fuel :-) I'd certainly prefer to see this in shell, since I believe the substitutions and reordering of text might be easier there.

See above - be my guest ;)

Quoted 7 lines
Quoted 5 lines
Quoted 3 lines

My argument is that this is a black-box tool like gcc/ld. Once we prove it to do what it does correctly, we can 'leave it alone'(tm) and test cases are fairly trivial. You can even mock-up the input file in vi :)

Ewwwwww ... like gcc/ld ... or maybe like Windows (TM)(C)(R)(?) :-D

No, like FLOSS gcc/ld :P

:-)

Quoted 9 lines
Quoted 4 lines
Quoted 2 lines

And is a shell based implementation going to be any easier to understand, modify, debug, etc?

I wonder :-)

Well, if someone manages to do it in shell, we will see - Take a look at main() - I have been very carefull to layout the way the tool operates in a very linear way but ther lists are global, so sharing data between the shell stages will be a pain (pipes?)

Bash can do locals ;-)

Quoted 52 lines
Quoted 33 lines
Quoted 33 lines
Quoted 32 lines
Quoted 27 lines
tools/mkinitseq.c new file +22 -0
+/*
+ * (C) Copyright 2012
+ * Graeme Russ <[email protected]>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License as
+ * published by the Free Software Foundation; either version 2 of
+ * the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston,
+ * MA 02111-1307 USA
+ */
+
+/**
+ * container_of - cast a member of a structure out to the containing

structure + * @ptr: the pointer to the member.
+ * @type: the type of the container struct this is embedded in.
+ * @member: the name of the member within the struct.
+ *
+ */

tools/mkinitseq.c line 30

Quoted 3 lines
tools/mkinitseq.c new file +3 -0
+#define container_of(ptr, type, member) ({                   \
+     const typeof( ((type *)0)->member ) *__mptr = (ptr);    \
+     (type *)( (char *)__mptr - offsetof(type,member) );})

Don't we already have this defined in uboot ?

Quoted 6 lines
Quoted 4 lines

mkinitseq is host-code so it only brings in host headers (not U-Boot headers) - If you don't have Linux kernel headers installed, you can't get this macro. Actually, even though I do have them, I had a hard time getting this macro in, so I gave up - Any hints?

Reimplement this thing in shell :-)

You already know my answer to this ;)

Quoted 4 lines
Quoted 2 lines

As an aside, I wonder if the kernel headers are required for the list macros?

Nope.

Good :)

Regards,

Graeme

#7 lore raw

PATCH 2/3 init_func: Add x86 support

In reply to Graeme Russ
---
 arch/x86/cpu/u-boot.lds |   23 +++++++++++++++++++++++
 arch/x86/lib/board.c    |   10 +++-------
 2 files changed, 26 insertions(+), 7 deletions(-)
arch/x86/cpu/u-boot.lds +23 -0
index fe28030..b88fd57 100644
@@ -28,6 +28,27 @@ ENTRY(_start)
 
 SECTIONS
 {
+#ifdef MAKE_INIT_LDS
+	.initfuncs : { KEEP(*(.initfuncs*)) }
+
+	/DISCARD/ : { *(.text*) }
+	/DISCARD/ : { *(.debug*) }
+	/DISCARD/ : { *(.u_boot_cmd*) }
+	/DISCARD/ : { *(.rodata*) }
+	/DISCARD/ : { *(.data*) }
+	/DISCARD/ : { *(.dynsym*) }
+	/DISCARD/ : { *(.hash*) }
+	/DISCARD/ : { *(.bss*) }
+	/DISCARD/ : { *(.bss) }
+	/DISCARD/ : { *(.rel.dyn*) }
+	/DISCARD/ : { *(.rel.dyn) }
+	/DISCARD/ : { *(.dynstr*) }
+	/DISCARD/ : { *(.dynamic*) }
+	/DISCARD/ : { *(.plt*) }
+	/DISCARD/ : { *(.interp*) }
+	/DISCARD/ : { *(.gnu*) }
+	/DISCARD/ : { *(.comment*) }
+#else
 	. = CONFIG_SYS_TEXT_BASE;	/* Location of bootcode in flash */
 	__text_start = .;
 	.text  : { *(.text*); }
@@ -72,6 +93,7 @@ SECTIONS
 	/DISCARD/ : { *(.plt*) }
 	/DISCARD/ : { *(.interp*) }
 	/DISCARD/ : { *(.gnu*) }
+	/DISCARD/ : { *(.initfuncs*) }
 
 	/* 16bit realmode trampoline code */
 	.realmode REALMODE_BASE : AT ( LOADADDR(.rel.dyn) + SIZEOF(.rel.dyn) ) { KEEP(*(.realmode)) }
@@ -94,4 +116,5 @@ SECTIONS
 
 	. = RESET_VEC_LOC;
 	.resetvec : AT (CONFIG_SYS_TEXT_BASE + (CONFIG_SYS_MONITOR_LEN - RESET_SEG_SIZE + RESET_VEC_LOC)) { KEEP(*(.resetvec)); }
+#endif
 }
arch/x86/lib/board.c +3 -7
index 5f0b62c..5789581 100644
@@ -32,6 +32,7 @@
  */
 
 #include <common.h>
+#include <init_func.h>
 #include <watchdog.h>
 #include <stdio_dev.h>
 #include <asm/u-boot-x86.h>
@@ -77,13 +78,7 @@
  *      - All remaining initialisation
  */
 
-/*
- * The requirements for any new initalization function is simple: it is
- * a function with no parameters which returns an integer return code,
- * where 0 means "continue" and != 0 means "fatal error, hang the system"
- */
-typedef int (init_fnc_t) (void);
-
+#ifndef CONFIG_INIT_FUNC
 /*
  * init_sequence_f is the list of init functions which are run when U-Boot
  * is executing from Flash with a limited 'C' environment. The following
@@ -206,6 +201,7 @@ init_fnc_t *init_sequence_r[] = {
 #endif
 	NULL,
 };
+#endif
 
 static void do_init_loop(init_fnc_t **init_fnc_ptr)
 {
Signature · 1 line
1.7.7.6
#8 lore raw

PATCH 3/3 init_func: Use for eNET board

In reply to Graeme Russ
---
 arch/x86/cpu/cpu.c               |    3 +++
 arch/x86/cpu/sc520/sc520_sdram.c |    3 +++
 arch/x86/cpu/sc520/sc520_timer.c |    2 ++
 arch/x86/lib/init_helpers.c      |   14 ++++++++++++++
 arch/x86/lib/init_wrappers.c     |   11 ++++++++++-
 arch/x86/lib/pcat_interrupts.c   |    2 ++
 arch/x86/lib/relocate.c          |    4 ++++
 board/eNET/eNET.c                |    5 +++++
 common/console.c                 |    3 +++
 common/env_dataflash.c           |    1 +
 common/env_eeprom.c              |    1 +
 common/env_fat.c                 |    1 +
 common/env_flash.c               |    2 ++
 common/env_mgdisk.c              |    1 +
 common/env_mmc.c                 |    1 +
 common/env_nand.c                |    1 +
 common/env_nowhere.c             |    1 +
 common/env_nvram.c               |    1 +
 common/env_onenand.c             |    1 +
 common/env_sf.c                  |    1 +
 common/serial.c                  |    2 ++
 common/stdio.c                   |    2 ++
 include/configs/eNET.h           |    1 +
 23 files changed, 63 insertions(+), 1 deletions(-)
arch/x86/cpu/cpu.c +3 -0
index e9bb0d7..263df28 100644
@@ -33,6 +33,7 @@
  */
 
 #include <common.h>
+#include <init_func.h>
 #include <command.h>
 #include <asm/processor.h>
 #include <asm/processor-flags.h>
@@ -137,6 +138,7 @@ int x86_cpu_init_f(void)
 	return 0;
 }
 int cpu_init_f(void) __attribute__((weak, alias("x86_cpu_init_f")));
+INIT_FUNC(cpu_init_f, cpu_f, RESET, , SDRAM);
 
 int x86_cpu_init_r(void)
 {
@@ -145,6 +147,7 @@ int x86_cpu_init_r(void)
 	return 0;
 }
 int cpu_init_r(void) __attribute__((weak, alias("x86_cpu_init_r")));
+INIT_FUNC(cpu_init_r, cpu_r, mem_malloc_init_r, , );
 
 void x86_enable_caches(void)
 {
arch/x86/cpu/sc520/sc520_sdram.c +3 -0
index 9dc1334..7033e18 100644
@@ -22,6 +22,7 @@
  */
 
 #include <common.h>
+#include <init_func.h>
 #include <asm/io.h>
 #include <asm/processor-flags.h>
 #include <asm/arch/sc520.h>
@@ -51,6 +52,7 @@ int dram_init_f(void)
 
 	return 0;
 }
+INIT_FUNC(dram_init_f, dram_f, console_init_f, , SDRAM);
 
 static inline void sc520_dummy_write(void)
 {
@@ -477,3 +479,4 @@ int dram_init(void)
 
 	return 0;
 }
+INIT_FUNC(dram_init, dram_r, , board_early_init_r, );
arch/x86/cpu/sc520/sc520_timer.c +2 -0
index 41f121f..7b25c08 100644
@@ -25,6 +25,7 @@
  */
 
 #include <common.h>
+#include <init_func.h>
 #include <asm/io.h>
 #include <asm/interrupt.h>
 #include <asm/arch/sc520.h>
@@ -69,6 +70,7 @@ int timer_init(void)
 
 	return 0;
 }
+INIT_FUNC(timer_init, timer, interrupt_init, , );
 
 /* Allow boards to override udelay implementation */
 void __udelay(unsigned long usec)
arch/x86/lib/init_helpers.c +14 -0
index 9f4dee0..aa88048 100644
@@ -21,6 +21,7 @@
  * MA 02111-1307 USA
  */
 #include <common.h>
+#include <init_func.h>
 #include <command.h>
 #include <stdio_dev.h>
 #include <version.h>
@@ -50,6 +51,7 @@ int display_banner(void)
 
 	return 0;
 }
+INIT_FUNC(display_banner, banner, timer_init, , );
 
 int display_dram_config(void)
 {
@@ -64,12 +66,14 @@ int display_dram_config(void)
 
 	return 0;
 }
+INIT_FUNC(display_dram_config, display_dram, display_banner, , );
 
 int init_baudrate_f(void)
 {
 	gd->baudrate = getenv_ulong("baudrate", 10, CONFIG_BAUDRATE);
 	return 0;
 }
+INIT_FUNC(init_baudrate_f, display_dram, env_init, , );
 
 int calculate_relocation_address(void)
 {
@@ -106,6 +110,7 @@ int calculate_relocation_address(void)
 
 	return 0;
 }
+INIT_FUNC(calculate_relocation_address, calc_reloc, dram_init_f, , );
 
 int copy_gd_to_ram_f_r(void)
 {
@@ -131,12 +136,14 @@ int copy_gd_to_ram_f_r(void)
 
 	return 0;
 }
+INIT_FUNC(copy_gd_to_ram_f_r, copy_gd, SDRAM, , RELOC);
 
 int init_cache_f_r(void)
 {
 	/* Initialise the CPU cache(s) */
 	return init_cache();
 }
+INIT_FUNC(init_cache_f_r, cache_f_r, copy_gd_to_ram_f_r, , RELOC);
 
 int set_reloc_flag_r(void)
 {
@@ -144,6 +151,7 @@ int set_reloc_flag_r(void)
 
 	return 0;
 }
+INIT_FUNC(set_reloc_flag_r, reloc_flag, RELOC, , );
 
 int mem_malloc_init_r(void)
 {
@@ -152,6 +160,7 @@ int mem_malloc_init_r(void)
 
 	return 0;
 }
+INIT_FUNC(mem_malloc_init_r, malloc, init_bd_struct_r, , );
 
 bd_t bd_data;
 
@@ -162,6 +171,7 @@ int init_bd_struct_r(void)
 
 	return 0;
 }
+INIT_FUNC(init_bd_struct_r, bd_struct, set_reloc_flag_r, , );
 
 #ifndef CONFIG_SYS_NO_FLASH
 int flash_init_r(void)
@@ -177,6 +187,7 @@ int flash_init_r(void)
 
 	return 0;
 }
+INIT_FUNC(flash_init_r, flash_r, RELOC, serial_initialize_r display_dram_config, );
 #endif
 
 int init_ip_address_r(void)
@@ -186,6 +197,7 @@ int init_ip_address_r(void)
 
 	return 0;
 }
+INIT_FUNC(init_ip_address_r, ip_address, env_relocate_r, , pci_init_r stdio_init);
 
 #ifdef CONFIG_STATUS_LED
 int status_led_set_r(void)
@@ -207,6 +219,7 @@ int set_bootfile_r(void)
 
 	return 0;
 }
+INIT_FUNC(set_bootfile_r, set_bootfile, set_load_addr_r, , );
 
 int set_load_addr_r(void)
 {
@@ -215,3 +228,4 @@ int set_load_addr_r(void)
 
 	return 0;
 }
+INIT_FUNC(set_load_addr_r, set_load_addr, enable_interrupts_r, , );
arch/x86/lib/init_wrappers.c +10 -1
index 71449fe..87ca713 100644
@@ -21,6 +21,7 @@
  * MA 02111-1307 USA
  */
 #include <common.h>
+#include <init_func.h>
 #include <serial.h>
 #include <kgdb.h>
 #include <scsi.h>
@@ -35,6 +36,7 @@ int serial_initialize_r(void)
 
 	return 0;
 }
+INIT_FUNC(serial_initialize_r, serial_multi, RELOC, display_dram_config, );
 
 int env_relocate_r(void)
 {
@@ -43,7 +45,7 @@ int env_relocate_r(void)
 
 	return 0;
 }
-
+INIT_FUNC(env_relocate_r, env_reloc, RELOC env_init, display_dram_config serial_initialize_r flash_init_r, );
 
 int pci_init_r(void)
 {
@@ -52,6 +54,7 @@ int pci_init_r(void)
 
 	return 0;
 }
+INIT_FUNC(pci_init_r, pci, cpu_r, init_ip_address_r env_relocate_r, stdio_init);
 
 int jumptable_init_r(void)
 {
@@ -59,6 +62,7 @@ int jumptable_init_r(void)
 
 	return 0;
 }
+INIT_FUNC(jumptable_init_r, jumptable, stdio_init, , );
 
 int pcmcia_init_r(void)
 {
@@ -83,6 +87,7 @@ int enable_interrupts_r(void)
 
 	return 0;
 }
+INIT_FUNC(enable_interrupts_r, enable_interrupts, , kgdb_init_r pci_init_r misc_init_r console_init_r, );
 
 int eth_initialize_r(void)
 {
@@ -91,7 +96,9 @@ int eth_initialize_r(void)
 
 	return 0;
 }
+INIT_FUNC(eth_initialize_r, eth, pci, bb_miiphy_init_r doc_init_r scsi_init_r ide_init_r set_bootfile_r set_load_addr_r, );
 
+#ifdef CONFIG_RESET_PHY_R
 int reset_phy_r(void)
 {
 #ifdef DEBUG
@@ -101,6 +108,8 @@ int reset_phy_r(void)
 
 	return 0;
 }
+INIT_FUNC(reset_phy_r, phy_reset, eth, eth_initialize_r, );
+#endif
 
 int ide_init_r(void)
 {
arch/x86/lib/pcat_interrupts.c +2 -0
index 5dac498..a403b7b 100644
@@ -31,6 +31,7 @@
  */
 
 #include <common.h>
+#include <init_func.h>
 #include <asm/io.h>
 #include <asm/i8259.h>
 #include <asm/ibmpc.h>
@@ -82,6 +83,7 @@ int interrupt_init(void)
 
 	return 0;
 }
+INIT_FUNC(interrupt_init, pcat_interrupt, , dram_init, );
 
 void mask_irq(int irq)
 {
arch/x86/lib/relocate.c +4 -0
index 200baab..69b0fd6 100644
@@ -32,6 +32,7 @@
  */
 
 #include <common.h>
+#include <init_func.h>
 #include <malloc.h>
 #include <asm/u-boot-x86.h>
 #include <asm/relocate.h>
@@ -45,6 +46,7 @@ int copy_uboot_to_ram(void)
 
 	return 0;
 }
+INIT_FUNC(copy_uboot_to_ram, copy_to_ram, SDRAM, init_cache_f_r, RELOC);
 
 int clear_bss(void)
 {
@@ -55,6 +57,7 @@ int clear_bss(void)
 
 	return 0;
 }
+INIT_FUNC(clear_bss, clear_bss, SDRAM, copy_uboot_to_ram, RELOC);
 
 int do_elf_reloc_fixups(void)
 {
@@ -89,3 +92,4 @@ int do_elf_reloc_fixups(void)
 
 	return 0;
 }
+INIT_FUNC(do_elf_reloc_fixups, elf_reloc, clear_bss, , RELOC);
board/eNET/eNET.c +5 -0
index 2f26470..19bbbe4 100644
@@ -22,6 +22,7 @@
  */
 
 #include <common.h>
+#include <init_func.h>
 #include <asm/io.h>
 #include <asm/arch/sc520.h>
 #include <net.h>
@@ -106,6 +107,7 @@ int board_early_init_f(void)
 
 	return 0;
 }
+INIT_FUNC(board_early_init_f, board_early_f, cpu_init_f, , SDRAM);
 
 static void enet_setup_pars(void)
 {
@@ -161,6 +163,7 @@ int board_early_init_r(void)
 
 	return 0;
 }
+INIT_FUNC(board_early_init_r, board_early_r, cpu_init_r, , pci);
 
 void show_boot_progress(int val)
 {
@@ -186,6 +189,8 @@ int last_stage_init(void)
 
 	return 0;
 }
+INIT_FUNC(last_stage_init, last_stage, , eth_initialize_r reset_phy_r, );
+
 
 ulong board_flash_get_legacy(ulong base, int banknum, flash_info_t *info)
 {
common/console.c +3 -0
index 1177f7d..01e6450 100644
@@ -22,6 +22,7 @@
  */
 
 #include <common.h>
+#include <init_func.h>
 #include <stdarg.h>
 #include <malloc.h>
 #include <stdio_dev.h>
@@ -588,6 +589,7 @@ int console_init_f(void)
 
 	return 0;
 }
+INIT_FUNC(console_init_f, console_f, serial_init, , SDRAM);
 
 void stdio_print_current_devices(void)
 {
@@ -775,5 +777,6 @@ int console_init_r(void)
 
 	return 0;
 }
+INIT_FUNC(console_init_r, console_r, stdio_init, jumptable_init_r, );
 
 #endif /* CONFIG_SYS_CONSOLE_IS_IN_ENV */
common/env_dataflash.c +1 -0
index 3c5af37..423e0ad 100644
@@ -114,3 +114,4 @@ int env_init(void)
 
 	return 0;
 }
+INIT_FUNC(env_init, env_init, cpu_init_f, board_early_init_f, SDRAM);
common/env_eeprom.c +1 -0
index b66bba2..75b4e11 100644
@@ -289,3 +289,4 @@ int env_init(void)
 	return 0;
 }
 #endif
+INIT_FUNC(env_init, env_init, cpu_init_f, board_early_init_f, SDRAM);
common/env_fat.c +1 -0
index bad92aa..d218094 100644
@@ -53,6 +53,7 @@ int env_init(void)
 
 	return 0;
 }
+INIT_FUNC(env_init, env_init, cpu_init_f, board_early_init_f, SDRAM);
 
 #ifdef CONFIG_CMD_SAVEENV
 int saveenv(void)
common/env_flash.c +2 -0
index aa970d4..5714b5b 100644
@@ -27,6 +27,7 @@
 /* #define DEBUG */
 
 #include <common.h>
+#include <init_func.h>
 #include <command.h>
 #include <environment.h>
 #include <linux/stddef.h>
@@ -117,6 +118,7 @@ int env_init(void)
 
 	return 0;
 }
+INIT_FUNC(env_init, env_init, cpu_init_f, board_early_init_f, SDRAM);
 
 #ifdef CMD_SAVEENV
 int saveenv(void)
common/env_mgdisk.c +1 -0
index d00e141..e5ffaad 100644
@@ -74,3 +74,4 @@ int env_init(void)
 
 	return 0;
 }
+INIT_FUNC(env_init, env_init, cpu_init_f, board_early_init_f, SDRAM);
common/env_mmc.c +1 -0
index 0c58ae1..dc8a21a 100644
@@ -62,6 +62,7 @@ int env_init(void)
 
 	return 0;
 }
+INIT_FUNC(env_init, env_init, cpu_init_f, board_early_init_f, SDRAM);
 
 static int init_mmc_for_env(struct mmc *mmc)
 {
common/env_nand.c +1 -0
index e8daec9..10e84f0 100644
@@ -134,6 +134,7 @@ int env_init(void)
 
 	return 0;
 }
+INIT_FUNC(env_init, env_init, cpu_init_f, board_early_init_f, SDRAM);
 
 #ifdef CMD_SAVEENV
 /*
common/env_nowhere.c +1 -0
index 18fcf2c..04d35e5 100644
@@ -49,3 +49,4 @@ int env_init(void)
 
 	return 0;
 }
+INIT_FUNC(env_init, env_init, cpu_init_f, board_early_init_f, SDRAM);
common/env_nvram.c +1 -0
index 6483db3..0487de0 100644
@@ -134,3 +134,4 @@ int env_init(void)
 
 	return 0;
 }
+INIT_FUNC(env_init, env_init, cpu_init_f, board_early_init_f, SDRAM);
common/env_onenand.c +1 -0
index 652665a..ec98de5 100644
@@ -135,3 +135,4 @@ int env_init(void)
 
 	return 0;
 }
+INIT_FUNC(env_init, env_init, cpu_init_f, board_early_init_f, SDRAM);
common/env_sf.c +1 -0
index bbd472f..84fd3b2 100644
@@ -349,3 +349,4 @@ int env_init(void)
 
 	return 0;
 }
+INIT_FUNC(env_init, env_init, cpu_init_f, board_early_init_f, SDRAM);
common/serial.c +2 -0
index 75cc1bb..920c5d3 100644
@@ -22,6 +22,7 @@
  */
 
 #include <common.h>
+#include <init_func.h>
 #include <serial.h>
 #include <stdio_dev.h>
 #include <post.h>
@@ -190,6 +191,7 @@ int serial_init(void)
 {
 	return get_current()->init();
 }
+INIT_FUNC(serial_init, serial_f, init_baudrate_f, , SDRAM);
 
 void serial_setbrg(void)
 {
common/stdio.c +2 -0
index 1bf9ba0..6472259 100644
@@ -22,6 +22,7 @@
  */
 
 #include <config.h>
+#include <init_func.h>
 #include <common.h>
 #include <stdarg.h>
 #include <malloc.h>
@@ -242,3 +243,4 @@ int stdio_init (void)
 
 	return (0);
 }
+INIT_FUNC(stdio_init, stdio, env_relocate_r, , );
include/configs/eNET.h +1 -0
index d5c9cad..57cecff 100644
@@ -29,6 +29,7 @@
 #ifndef __CONFIG_H
 #define __CONFIG_H
 
+#define CONFIG_INIT_FUNC
 /*
  * High Level Configuration Options
  * (easy to change)
Signature · 1 line
1.7.7.6
#9 lore raw

cover Initialisation Sequence Framework

In reply to Graeme Russ

Dear Graeme Russ,

Quoted 7 lines

Well here it is (finally) - My Initialisation Sequence Framework

This series is limited to x86 and is controlled by CONFIG_INIT_FUNC so no other arch should be effected. I don't have any cross compilers, so I would appreciate if others could confirm the zero impact assertion

Asbestos suit on

Oh this is gonna be the flamewar of the year now :-E~

Gasoline ready, asbestos pants - check, flamethrower on ... let's go!

                                         )  (  (    (
                                         (  )  () @@  )  (( (
                                     (      (  )( @@  (  )) ) (
                                   (    (  ( ()( /---\   (()( (
     _______                            )  ) )(@ !O O! )@@  ( ) ) )
    <   ____)                      ) (  ( )( ()@ \ o / (@@@@@ ( ()( )
 /--|  |(  o|                     (  )  ) ((@@(@@ !o! @@@@(@@@@@)() (
|   >   \___|                      ) ( @)@@)@ /---\-/---\ )@@@@@()( )
|  /---------+                    (@@@@)@@@( // /-----\ \\ @@@)@@@@@(  .
| |    \ =========______/|@@@@@@@@@@@@@(@@@ // @ /---\ @ \\ @(@@@(@@@ .  .
|  \   \\=========------\|@@@@@@@@@@@@@@@@@ O @@@ /-\ @@@ O @@(@@)@@ @   .
|   \   \----+--\-)))           @@@@@@@@@@ !! @@@@ % @@@@ !! @@)@@@ .. .
|   |\______|_)))/             .    @@@@@@ !! @@ /---\ @@ !! @@(@@@ @ . .
 \__==========           *        .    @@ /MM  /\O   O/\  MM\ @@@@@@@. .
    |   |-\   \          (       .      @ !!!  !! \-/ !!  !!! @@@@@ .
    |   |  \   \          )      .     .  @@@@ !!     !!  .(. @.  .. .
    |   |   \   \        (    /   .(  . \)). ( |O  )( O! @@@@ . )      .
    |   |   /   /         ) (      )).  ((  .) !! ((( !! @@ (. ((. .   .
    |   |  /   /   ()  ))   ))   .( ( ( ) ). ( !!  )( !! ) ((   ))  ..
    |   |_<   /   ( ) ( (  ) )   (( )  )).) ((/ |  (  | \(  )) ((. ).

____<_____\\__\__(___)_))_((_(____))__(_(___.oooO_____Oooo.(_(_)_)((_

Artist: P.r.i.m.a.l
Borrowed from: http://www.ascii-art.de/ascii/def/flamethrower.txt
Copyright ? 1997-2004 Andreas Freise

Quoted 40 lines

Fire away ;)

Regards,

Graeme

Graeme Russ (3):
init_func: Add fundamental framework
init_func: Add x86 support
init_func: Use for eNET board

 Makefile                         |   34 +-
 arch/x86/cpu/cpu.c               |    3 +
 arch/x86/cpu/sc520/sc520_sdram.c |    3 +
 arch/x86/cpu/sc520/sc520_timer.c |    2 +
 arch/x86/cpu/u-boot.lds          |   23 +
 arch/x86/lib/board.c             |   10 +-
 arch/x86/lib/init_helpers.c      |   14 +
 arch/x86/lib/init_wrappers.c     |   11 +-
 arch/x86/lib/pcat_interrupts.c   |    2 +
 arch/x86/lib/relocate.c          |    4 +
 board/eNET/eNET.c                |    5 +
 common/Makefile                  |    2 +
 common/console.c                 |    3 +
 common/env_dataflash.c           |    1 +
 common/env_eeprom.c              |    1 +
 common/env_fat.c                 |    1 +
 common/env_flash.c               |    2 +
 common/env_mgdisk.c              |    1 +
 common/env_mmc.c                 |    1 +
 common/env_nand.c                |    1 +
 common/env_nowhere.c             |    1 +
 common/env_nvram.c               |    1 +
 common/env_onenand.c             |    1 +
 common/env_sf.c                  |    1 +
 common/serial.c                  |    2 +
 common/stdio.c                   |    2 +
 config.mk                        |    2 +
 doc/README.INIT_FUNC             |   65 ++
 include/configs/eNET.h           |    1 +
Quoted 8 lines
 include/init_func.h              |   37 +
 tools/Makefile                   |    6 +
 tools/mkinitseq.c                | 1512
++++++++++++++++++++++++++++++++++++++ 32 files changed, 1746
insertions(+), 9 deletions(-)
 create mode 100644 doc/README.INIT_FUNC
 create mode 100644 include/init_func.h
 create mode 100644 tools/mkinitseq.c

Best regards,
Marek Vasut