Re: [PATCH][RFC] more initcall cleanups
On Wed, 16 May 2001, Alexander Viro wrote:
> a) I2C stuff got converted to module_init() nicely. That took
> a lot of cruft away.
Definitely makes sense.
> b) init order is preserved. However, that worked only because
> none of the i2c initialization functions touch stuff from random.c
I don't think init order is preserved.
> c) I had to put i2c before char to preserve the ordering.
> Not a big deal, but I'm somewhat at loss here - how to do it without
> really ugly drivers/Makefile. Suggestions?
This part is not necessary. The ordering of subdir-y only has to do with
compilation ordering, not with link order. The link order for
drivers/char/* is set explicitly in the toplevel Makefile.
> d) if we set CONFIG_I2C to 'y' we don't include drivers/i2c
> into subdir-m. However, having i2c-core in kernel and the rest done as
> modules is OK with i2c itself. And I seriously suspect that it's
> a common situation. Am I right assuming that correct way to deal with
> that is to put i2c into mod-subdirs?
Yes, that's exactly what mod-subdirs is for.
> e) looks like rand_initialize() is in the same class as handling
> VFS/VM/etc. caches - global infrastructure. It definitely should be
> called before all other drivers. Notice that old device_init() was asking
> for trouble - it called parport_init() before chr_dev_init() (which calls
> rand_initialize()), so if somebody would try to feed some entropy into
> pool during the parport_init() we would get an interesting (and hard
> to understand) problem. Maybe we should move the call into basic_setup()?
I suppose I'm not the right one to comment on this, but it seems to make
sense to me.
> diff -urN S5-pre3-init-0/drivers/Makefile S5-pre3-init/drivers/Makefile
> --- S5-pre3-init-0/drivers/Makefile Wed May 16 16:26:35 2001
> +++ S5-pre3-init/drivers/Makefile Wed May 16 20:47:52 2001
> @@ -9,8 +9,13 @@
> mod-subdirs := dio mtd sbus video macintosh usb input telephony sgi i2o ide \
> scsi md ieee1394 pnp isdn atm fc4 net/hamradio i2c acpi
>
> -subdir-y := parport char block net sound misc media cdrom
> -subdir-m := $(subdir-y)
> +subdir-y := parport
> +subdir-m := parport
> +
> +subdir-$(CONFIG_I2C) += i2c
> +
> +subdir-y += char block net sound misc media cdrom
> +subdir-m += char block net sound misc media cdrom
>
>
> subdir-$(CONFIG_DIO) += dio
> @@ -40,7 +45,6 @@
>
> # CONFIG_HAMRADIO can be set without CONFIG_NETDEVICE being set -- ch
> subdir-$(CONFIG_HAMRADIO)+= net/hamradio
> -subdir-$(CONFIG_I2C) += i2c
> subdir-$(CONFIG_ACPI)+= acpi
>
> include $(TOPDIR)/Rules.make
As stated above, this change won't affect vmlinux at all.
> diff -urN S5-pre3-init-0/drivers/char/random.c S5-pre3-init/drivers/char/random.c
> --- S5-pre3-init-0/drivers/char/random.c Wed May 16 16:26:36 2001
> +++ S5-pre3-init/drivers/char/random.cWed May 16 20:40:12 2001
> @@ -1380,7 +1380,7 @@
> }
> }
>
> -void __init rand_initialize(void)
> +static int __init rand_initialize(void)
> {
> int i;
>
> @@ -1404,7 +1404,10 @@
> memset(_timer_state, 0, sizeof(struct timer_rand_state));
> memset(_timer_state, 0, sizeof(struct timer_rand_state));
> extract_timer_state.dont_count_entropy = 1;
> + return 0;
> }
> +
> +__initcall(rand_initialize);
>
> void rand_initialize_irq(int irq)
> {
prototype should be removed from include/linux/random.h
--Kai
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [EMAIL PROTECTED]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH][RFC] more initcall cleanups
On Wed, 16 May 2001, Alexander Viro wrote:
a) I2C stuff got converted to module_init() nicely. That took
a lot of cruft away.
Definitely makes sense.
b) init order is preserved. However, that worked only because
none of the i2c initialization functions touch stuff from random.c
I don't think init order is preserved.
c) I had to put i2c before char to preserve the ordering.
Not a big deal, but I'm somewhat at loss here - how to do it without
really ugly drivers/Makefile. Suggestions?
This part is not necessary. The ordering of subdir-y only has to do with
compilation ordering, not with link order. The link order for
drivers/char/* is set explicitly in the toplevel Makefile.
d) if we set CONFIG_I2C to 'y' we don't include drivers/i2c
into subdir-m. However, having i2c-core in kernel and the rest done as
modules is OK with i2c itself. And I seriously suspect that it's
a common situation. Am I right assuming that correct way to deal with
that is to put i2c into mod-subdirs?
Yes, that's exactly what mod-subdirs is for.
e) looks like rand_initialize() is in the same class as handling
VFS/VM/etc. caches - global infrastructure. It definitely should be
called before all other drivers. Notice that old device_init() was asking
for trouble - it called parport_init() before chr_dev_init() (which calls
rand_initialize()), so if somebody would try to feed some entropy into
pool during the parport_init() we would get an interesting (and hard
to understand) problem. Maybe we should move the call into basic_setup()?
I suppose I'm not the right one to comment on this, but it seems to make
sense to me.
diff -urN S5-pre3-init-0/drivers/Makefile S5-pre3-init/drivers/Makefile
--- S5-pre3-init-0/drivers/Makefile Wed May 16 16:26:35 2001
+++ S5-pre3-init/drivers/Makefile Wed May 16 20:47:52 2001
@@ -9,8 +9,13 @@
mod-subdirs := dio mtd sbus video macintosh usb input telephony sgi i2o ide \
scsi md ieee1394 pnp isdn atm fc4 net/hamradio i2c acpi
-subdir-y := parport char block net sound misc media cdrom
-subdir-m := $(subdir-y)
+subdir-y := parport
+subdir-m := parport
+
+subdir-$(CONFIG_I2C) += i2c
+
+subdir-y += char block net sound misc media cdrom
+subdir-m += char block net sound misc media cdrom
subdir-$(CONFIG_DIO) += dio
@@ -40,7 +45,6 @@
# CONFIG_HAMRADIO can be set without CONFIG_NETDEVICE being set -- ch
subdir-$(CONFIG_HAMRADIO)+= net/hamradio
-subdir-$(CONFIG_I2C) += i2c
subdir-$(CONFIG_ACPI)+= acpi
include $(TOPDIR)/Rules.make
As stated above, this change won't affect vmlinux at all.
diff -urN S5-pre3-init-0/drivers/char/random.c S5-pre3-init/drivers/char/random.c
--- S5-pre3-init-0/drivers/char/random.c Wed May 16 16:26:36 2001
+++ S5-pre3-init/drivers/char/random.cWed May 16 20:40:12 2001
@@ -1380,7 +1380,7 @@
}
}
-void __init rand_initialize(void)
+static int __init rand_initialize(void)
{
int i;
@@ -1404,7 +1404,10 @@
memset(mouse_timer_state, 0, sizeof(struct timer_rand_state));
memset(extract_timer_state, 0, sizeof(struct timer_rand_state));
extract_timer_state.dont_count_entropy = 1;
+ return 0;
}
+
+__initcall(rand_initialize);
void rand_initialize_irq(int irq)
{
prototype should be removed from include/linux/random.h
--Kai
-
To unsubscribe from this list: send the line unsubscribe linux-kernel in
the body of a message to [EMAIL PROTECTED]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
[PATCH][RFC] more initcall cleanups
Linus, I've done a bit more cleaning the device initialization
up (beginning of chr_dev_init()) and results were, well, interesting.
a) I2C stuff got converted to module_init() nicely. That took
a lot of cruft away.
b) init order is preserved. However, that worked only because
none of the i2c initialization functions touch stuff from random.c
c) I had to put i2c before char to preserve the ordering.
Not a big deal, but I'm somewhat at loss here - how to do it without
really ugly drivers/Makefile. Suggestions?
d) if we set CONFIG_I2C to 'y' we don't include drivers/i2c
into subdir-m. However, having i2c-core in kernel and the rest done as
modules is OK with i2c itself. And I seriously suspect that it's
a common situation. Am I right assuming that correct way to deal with
that is to put i2c into mod-subdirs?
e) looks like rand_initialize() is in the same class as handling
VFS/VM/etc. caches - global infrastructure. It definitely should be
called before all other drivers. Notice that old device_init() was asking
for trouble - it called parport_init() before chr_dev_init() (which calls
rand_initialize()), so if somebody would try to feed some entropy into
pool during the parport_init() we would get an interesting (and hard
to understand) problem. Maybe we should move the call into basic_setup()?
Comments?
Very preliminary patch follows. Please, _don't_ apply it in that
form - drivers/Makefile is just too ugly.
Al
diff -urN S5-pre3-init-0/drivers/Makefile S5-pre3-init/drivers/Makefile
--- S5-pre3-init-0/drivers/Makefile Wed May 16 16:26:35 2001
+++ S5-pre3-init/drivers/Makefile Wed May 16 20:47:52 2001
@@ -9,8 +9,13 @@
mod-subdirs := dio mtd sbus video macintosh usb input telephony sgi i2o ide \
scsi md ieee1394 pnp isdn atm fc4 net/hamradio i2c acpi
-subdir-y :=parport char block net sound misc media cdrom
-subdir-m :=$(subdir-y)
+subdir-y :=parport
+subdir-m :=parport
+
+subdir-$(CONFIG_I2C) += i2c
+
+subdir-y +=char block net sound misc media cdrom
+subdir-m +=char block net sound misc media cdrom
subdir-$(CONFIG_DIO) += dio
@@ -40,7 +45,6 @@
# CONFIG_HAMRADIO can be set without CONFIG_NETDEVICE being set -- ch
subdir-$(CONFIG_HAMRADIO) += net/hamradio
-subdir-$(CONFIG_I2C) += i2c
subdir-$(CONFIG_ACPI) += acpi
include $(TOPDIR)/Rules.make
diff -urN S5-pre3-init-0/drivers/char/Makefile S5-pre3-init/drivers/char/Makefile
--- S5-pre3-init-0/drivers/char/MakefileWed May 16 16:26:36 2001
+++ S5-pre3-init/drivers/char/Makefile Wed May 16 20:39:26 2001
@@ -16,7 +16,7 @@
O_TARGET := char.o
-obj-y += mem.o tty_io.o n_tty.o tty_ioctl.o raw.o pty.o misc.o random.o
+obj-y += random.o mem.o tty_io.o n_tty.o tty_ioctl.o raw.o pty.o misc.o
# All of the (potential) objects that export symbols.
# This list comes from 'grep -l EXPORT_SYMBOL *.[hc]'.
diff -urN S5-pre3-init-0/drivers/char/mem.c S5-pre3-init/drivers/char/mem.c
--- S5-pre3-init-0/drivers/char/mem.c Wed May 16 16:26:36 2001
+++ S5-pre3-init/drivers/char/mem.c Wed May 16 20:48:51 2001
@@ -26,9 +26,6 @@
#include
#include
-#ifdef CONFIG_I2C
-extern int i2c_init_all(void);
-#endif
#ifdef CONFIG_VIDEO_DEV
extern int videodev_init(void);
#endif
@@ -615,10 +612,6 @@
if (devfs_register_chrdev(MEM_MAJOR,"mem",_fops))
printk("unable to get major %d for memory devs\n", MEM_MAJOR);
memory_devfs_register();
- rand_initialize();
-#ifdef CONFIG_I2C
- i2c_init_all();
-#endif
#if defined (CONFIG_FB)
fbmem_init();
#endif
diff -urN S5-pre3-init-0/drivers/char/random.c S5-pre3-init/drivers/char/random.c
--- S5-pre3-init-0/drivers/char/random.cWed May 16 16:26:36 2001
+++ S5-pre3-init/drivers/char/random.c Wed May 16 20:40:12 2001
@@ -1380,7 +1380,7 @@
}
}
-void __init rand_initialize(void)
+static int __init rand_initialize(void)
{
int i;
@@ -1404,7 +1404,10 @@
memset(_timer_state, 0, sizeof(struct timer_rand_state));
memset(_timer_state, 0, sizeof(struct timer_rand_state));
extract_timer_state.dont_count_entropy = 1;
+ return 0;
}
+
+__initcall(rand_initialize);
void rand_initialize_irq(int irq)
{
diff -urN S5-pre3-init-0/drivers/i2c/i2c-algo-bit.c
S5-pre3-init/drivers/i2c/i2c-algo-bit.c
--- S5-pre3-init-0/drivers/i2c/i2c-algo-bit.c Sun Apr 1 23:56:45 2001
+++ S5-pre3-init/drivers/i2c/i2c-algo-bit.c Wed May 16 20:54:07 2001
@@ -607,7 +607,7 @@
return 0;
}
-int __init i2c_algo_bit_init (void)
+static int __init i2c_algo_bit_init (void)
{
printk("i2c-algo-bit.o: i2c bit algorithm module\n");
return 0;
@@ -618,7 +618,6 @@
EXPORT_SYMBOL(i2c_bit_add_bus);
EXPORT_SYMBOL(i2c_bit_del_bus);
-#ifdef MODULE
MODULE_AUTHOR("Simon G. Vogl <[EMAIL
[PATCH][RFC] more initcall cleanups
Linus, I've done a bit more cleaning the device initialization
up (beginning of chr_dev_init()) and results were, well, interesting.
a) I2C stuff got converted to module_init() nicely. That took
a lot of cruft away.
b) init order is preserved. However, that worked only because
none of the i2c initialization functions touch stuff from random.c
c) I had to put i2c before char to preserve the ordering.
Not a big deal, but I'm somewhat at loss here - how to do it without
really ugly drivers/Makefile. Suggestions?
d) if we set CONFIG_I2C to 'y' we don't include drivers/i2c
into subdir-m. However, having i2c-core in kernel and the rest done as
modules is OK with i2c itself. And I seriously suspect that it's
a common situation. Am I right assuming that correct way to deal with
that is to put i2c into mod-subdirs?
e) looks like rand_initialize() is in the same class as handling
VFS/VM/etc. caches - global infrastructure. It definitely should be
called before all other drivers. Notice that old device_init() was asking
for trouble - it called parport_init() before chr_dev_init() (which calls
rand_initialize()), so if somebody would try to feed some entropy into
pool during the parport_init() we would get an interesting (and hard
to understand) problem. Maybe we should move the call into basic_setup()?
Comments?
Very preliminary patch follows. Please, _don't_ apply it in that
form - drivers/Makefile is just too ugly.
Al
diff -urN S5-pre3-init-0/drivers/Makefile S5-pre3-init/drivers/Makefile
--- S5-pre3-init-0/drivers/Makefile Wed May 16 16:26:35 2001
+++ S5-pre3-init/drivers/Makefile Wed May 16 20:47:52 2001
@@ -9,8 +9,13 @@
mod-subdirs := dio mtd sbus video macintosh usb input telephony sgi i2o ide \
scsi md ieee1394 pnp isdn atm fc4 net/hamradio i2c acpi
-subdir-y :=parport char block net sound misc media cdrom
-subdir-m :=$(subdir-y)
+subdir-y :=parport
+subdir-m :=parport
+
+subdir-$(CONFIG_I2C) += i2c
+
+subdir-y +=char block net sound misc media cdrom
+subdir-m +=char block net sound misc media cdrom
subdir-$(CONFIG_DIO) += dio
@@ -40,7 +45,6 @@
# CONFIG_HAMRADIO can be set without CONFIG_NETDEVICE being set -- ch
subdir-$(CONFIG_HAMRADIO) += net/hamradio
-subdir-$(CONFIG_I2C) += i2c
subdir-$(CONFIG_ACPI) += acpi
include $(TOPDIR)/Rules.make
diff -urN S5-pre3-init-0/drivers/char/Makefile S5-pre3-init/drivers/char/Makefile
--- S5-pre3-init-0/drivers/char/MakefileWed May 16 16:26:36 2001
+++ S5-pre3-init/drivers/char/Makefile Wed May 16 20:39:26 2001
@@ -16,7 +16,7 @@
O_TARGET := char.o
-obj-y += mem.o tty_io.o n_tty.o tty_ioctl.o raw.o pty.o misc.o random.o
+obj-y += random.o mem.o tty_io.o n_tty.o tty_ioctl.o raw.o pty.o misc.o
# All of the (potential) objects that export symbols.
# This list comes from 'grep -l EXPORT_SYMBOL *.[hc]'.
diff -urN S5-pre3-init-0/drivers/char/mem.c S5-pre3-init/drivers/char/mem.c
--- S5-pre3-init-0/drivers/char/mem.c Wed May 16 16:26:36 2001
+++ S5-pre3-init/drivers/char/mem.c Wed May 16 20:48:51 2001
@@ -26,9 +26,6 @@
#include asm/io.h
#include asm/pgalloc.h
-#ifdef CONFIG_I2C
-extern int i2c_init_all(void);
-#endif
#ifdef CONFIG_VIDEO_DEV
extern int videodev_init(void);
#endif
@@ -615,10 +612,6 @@
if (devfs_register_chrdev(MEM_MAJOR,mem,memory_fops))
printk(unable to get major %d for memory devs\n, MEM_MAJOR);
memory_devfs_register();
- rand_initialize();
-#ifdef CONFIG_I2C
- i2c_init_all();
-#endif
#if defined (CONFIG_FB)
fbmem_init();
#endif
diff -urN S5-pre3-init-0/drivers/char/random.c S5-pre3-init/drivers/char/random.c
--- S5-pre3-init-0/drivers/char/random.cWed May 16 16:26:36 2001
+++ S5-pre3-init/drivers/char/random.c Wed May 16 20:40:12 2001
@@ -1380,7 +1380,7 @@
}
}
-void __init rand_initialize(void)
+static int __init rand_initialize(void)
{
int i;
@@ -1404,7 +1404,10 @@
memset(mouse_timer_state, 0, sizeof(struct timer_rand_state));
memset(extract_timer_state, 0, sizeof(struct timer_rand_state));
extract_timer_state.dont_count_entropy = 1;
+ return 0;
}
+
+__initcall(rand_initialize);
void rand_initialize_irq(int irq)
{
diff -urN S5-pre3-init-0/drivers/i2c/i2c-algo-bit.c
S5-pre3-init/drivers/i2c/i2c-algo-bit.c
--- S5-pre3-init-0/drivers/i2c/i2c-algo-bit.c Sun Apr 1 23:56:45 2001
+++ S5-pre3-init/drivers/i2c/i2c-algo-bit.c Wed May 16 20:54:07 2001
@@ -607,7 +607,7 @@
return 0;
}
-int __init i2c_algo_bit_init (void)
+static int __init i2c_algo_bit_init (void)
{
printk(i2c-algo-bit.o: i2c bit algorithm module\n);
return 0;
@@ -618,7 +618,6 @@
EXPORT_SYMBOL(i2c_bit_add_bus);
EXPORT_SYMBOL(i2c_bit_del_bus);
-#ifdef MODULE
