It seems like early_printk can be configured into
a few architectures but also appear not to be used.
$ git grep -w "early_printk"
arch/arm/kernel/Makefile:obj-$(CONFIG_EARLY_PRINTK) += early_printk.o
arch/arm/kernel/early_printk.c: * linux/arch/arm/kernel/early_printk.c
arch/arm/mach-socfpga/socfpga.c: early_printk("Early printk initialized\n");
arch/blackfin/kernel/Makefile:obj-$(CONFIG_EARLY_PRINTK) += early_printk.o
arch/blackfin/kernel/bfin_dma.c:#include <asm/early_printk.h>
arch/blackfin/kernel/early_printk.c: * derived from arch/x86/kernel/early_printk.c
arch/blackfin/kernel/early_printk.c:#include <asm/early_printk.h>
arch/blackfin/kernel/setup.c:#include <asm/early_printk.h>
arch/blackfin/kernel/shadow_console.c:#include <asm/early_printk.h>
arch/blackfin/mm/init.c:#include <asm/early_printk.h>
arch/ia64/sn/kernel/setup.c: * IO on SN2 is done via SAL calls, early_printk won't work without this.
arch/m68k/kernel/Makefile:obj-$(CONFIG_EARLY_PRINTK) += early_printk.o
arch/microblaze/kernel/Makefile:obj-$(CONFIG_EARLY_PRINTK) += early_printk.o
arch/microblaze/kernel/cpu/cpuinfo-pvr-full.c: early_printk("ERROR: Microblaze " x "-different for PVR and DTS\n");
arch/microblaze/kernel/cpu/cpuinfo-static.c: early_printk("ERROR: Microblaze " x "-different for kernel and DTS\n");
arch/microblaze/kernel/setup.c:/* initialize device tree for usage in early_printk */
arch/mips/ath25/Makefile:obj-$(CONFIG_EARLY_PRINTK) += early_printk.o
arch/mips/ath79/Makefile:obj-$(CONFIG_EARLY_PRINTK) += early_printk.o
arch/mips/bcm63xx/Makefile:obj-$(CONFIG_EARLY_PRINTK) += early_printk.o
arch/mips/include/asm/mach-lantiq/falcon/lantiq_soc.h: * during early_printk no ioremap possible at this early stage
arch/mips/include/asm/mach-lantiq/xway/lantiq_soc.h: * during early_printk no ioremap is possible
arch/mips/kernel/Makefile:obj-$(CONFIG_EARLY_PRINTK) += early_printk.o
arch/mips/lantiq/Makefile:obj-$(CONFIG_EARLY_PRINTK) += early_printk.o
arch/mips/loongson/common/Makefile:obj-$(CONFIG_EARLY_PRINTK) += early_printk.o
arch/mips/ralink/Makefile:obj-$(CONFIG_EARLY_PRINTK) += early_printk.o
arch/tile/kernel/Makefile:obj-$(CONFIG_EARLY_PRINTK) += early_printk.o
arch/tile/kernel/early_printk.c: early_printk("Kernel panic - not syncing: %pV", &vaf);
arch/um/kernel/Makefile:obj-$(CONFIG_EARLY_PRINTK) += early_printk.o
arch/unicore32/kernel/Makefile:obj-$(CONFIG_EARLY_PRINTK) += early_printk.o
arch/unicore32/kernel/early_printk.c: * linux/arch/unicore32/kernel/early_printk.c
arch/x86/kernel/Makefile:obj-$(CONFIG_EARLY_PRINTK) += early_printk.o
arch/x86/kernel/e820.c: early_printk(msg);
arch/x86/kernel/head64.c: early_printk("Kernel alive\n");
arch/x86/kernel/head_64.S: call early_printk
arch/x86/platform/efi/Makefile:obj-$(CONFIG_EARLY_PRINTK_EFI) += early_printk.o
drivers/tty/serial/8250/8250_early.c: * and on early_printk.c by Andi Kleen.
drivers/tty/serial/sn_console.c: * synchronous (raw) and asynchronous (buffered). initially, early_printk
drivers/tty/serial/sn_console.c: /* without early_printk, we may be invoked late enough to race
drivers/tty/serial/sn_console.c: /* early_printk invocation may have done this for us */
include/linux/printk.h:void early_printk(const char *fmt, ...);
include/linux/printk.h:void early_printk(const char *s, ...) { }
kernel/events/core.c: early_printk("perf interrupt took too long (%lld > %lld), lowering "
kernel/locking/lockdep.c: * We cannot printk in early bootup code. Not even early_printk()
kernel/printk/printk.c:asmlinkage __visible void early_printk(const char *fmt, ...)
kernel/printk/printk.c: * early_printk) - sometimes before setup_arch() completes - be careful
kernel/printk/printk.c: * There are two types of consoles - bootconsoles (early_printk) and
These seem to the only uses:
arch/arm/mach-socfpga/socfpga.c: early_printk("Early printk initialized\n");
[]
arch/microblaze/kernel/cpu/cpuinfo-pvr-full.c: early_printk("ERROR: Microblaze " x "-different for PVR and DTS\n");
arch/microblaze/kernel/cpu/cpuinfo-static.c: early_printk("ERROR: Microblaze " x "-different for kernel and DTS\n");
[]
arch/tile/kernel/early_printk.c: early_printk("Kernel panic - not syncing: %pV", &vaf);
[]
arch/x86/kernel/e820.c: early_printk(msg);
arch/x86/kernel/head64.c: early_printk("Kernel alive\n");
arch/x86/kernel/head_64.S: call early_printk
[]
kernel/events/core.c: early_printk("perf interrupt took too long (%lld > %lld), lowering "
So blackfin, m68k, and mips seems to have it possible to enable,
but also don't appear at first glance to use it,
Is early_printk really used by these architectures?
Should it be removed?
On Wed, Dec 17, 2014 at 12:58 PM, Joe Perches <[email protected]> wrote:
> It seems like early_printk can be configured into
> a few architectures but also appear not to be used.
>
> $ git grep -w "early_printk"
[snip]
> arch/mips/kernel/Makefile:obj-$(CONFIG_EARLY_PRINTK) += early_printk.o
Nowadays I try to use OF_EARLYCON whenever possible, but when that has
been unavailable, I have used arch/mips/kernel/early_printk.c to get
console output before the serial driver is initialized. It runs very
early in the boot sequence and has very few dependencies, which makes
it useful for board bringup.
At least on MIPS, the EARLY_PRINTK implementation registers itself as
a console and works with standard printk() calls. It doesn't rely on
arch/driver code explicitly calling early_printk().
Side note: looking through kernel/printk/printk.c it looks like
there's a space missing in the description string:
MODULE_PARM_DESC(ignore_loglevel, "ignore loglevel setting, to"
"print all kernel messages to the console.");
But since CONFIG_PRINTK is a bool option, I don't know if the
description string actually shows up anywhere. Should it be converted
into a comment?
Kevin Cernekee <[email protected]> writes:
> On Wed, Dec 17, 2014 at 12:58 PM, Joe Perches <[email protected]> wrote:
>> It seems like early_printk can be configured into
>> a few architectures but also appear not to be used.
>>
>> $ git grep -w "early_printk"
> [snip]
>> arch/mips/kernel/Makefile:obj-$(CONFIG_EARLY_PRINTK) += early_printk.o
>
> Nowadays I try to use OF_EARLYCON whenever possible, but when that has
> been unavailable, I have used arch/mips/kernel/early_printk.c to get
> console output before the serial driver is initialized. It runs very
> early in the boot sequence and has very few dependencies, which makes
> it useful for board bringup.
EARLY_PRINTK on MIPS is enabled just after prom_init(), well before
OF_EARLYCON is usable. I'd prefer if it stayed that way.
--
M?ns Rullg?rd
[email protected]
On Wed, 17 Dec 2014, Joe Perches wrote:
> It seems like early_printk can be configured into
> a few architectures but also appear not to be used.
>
> $ git grep -w "early_printk"
...
> These seem to the only uses:
...
> So blackfin, m68k, and mips seems to have it possible to enable,
> but also don't appear at first glance to use it,
Hint: CONFIG_EARLY_PRINTK covers far more than early_printk()
> Is early_printk really used by these architectures?
> Should it be removed?
Sure, if you have a good reason to remove working functionality.
Thanks,
tglx
On Fri, 2014-12-19 at 00:33 +0100, Thomas Gleixner wrote:
> On Wed, 17 Dec 2014, Joe Perches wrote:
> > It seems like early_printk can be configured into
> > a few architectures but also appear not to be used.
> >
> > $ git grep -w "early_printk"
> ...
> > These seem to the only uses:
> ...
> > So blackfin, m68k, and mips seems to have it possible to enable,
> > but also don't appear at first glance to use it,
>
> Hint: CONFIG_EARLY_PRINTK covers far more than early_printk()
I know this.
Note also I didn't specify CONFIG_EARLY_PRINTK,
just early_printk.
> > Is early_printk really used by these architectures?
> > Should it be removed?
>
> Sure, if you have a good reason to remove working functionality.
Unused.
Ideally, all direct early_printk() uses would go away.
This would just be a starting point to minimize code.
Joe Perches <[email protected]> writes:
> On Fri, 2014-12-19 at 00:33 +0100, Thomas Gleixner wrote:
>> On Wed, 17 Dec 2014, Joe Perches wrote:
>> > It seems like early_printk can be configured into
>> > a few architectures but also appear not to be used.
>> >
>> > $ git grep -w "early_printk"
>> ...
>> > These seem to the only uses:
>> ...
>> > So blackfin, m68k, and mips seems to have it possible to enable,
>> > but also don't appear at first glance to use it,
>>
>> Hint: CONFIG_EARLY_PRINTK covers far more than early_printk()
>
> I know this.
>
> Note also I didn't specify CONFIG_EARLY_PRINTK,
> just early_printk.
>
>> > Is early_printk really used by these architectures?
>> > Should it be removed?
>>
>> Sure, if you have a good reason to remove working functionality.
>
> Unused.
What exactly are you proposing to remove? I see no unused code related
to early printk (in any variant) under arch/mips.
--
M?ns Rullg?rd
[email protected]
On Fri, 2014-12-19 at 01:43 +0000, M?ns Rullg?rd wrote:
> What exactly are you proposing to remove?
Optionally compile out
kernel/printk/printk.c:early_printk()
even if CONFIG_EARLY_PRINTK is enabled.
> I see no unused code related
> to early printk (in any variant) under arch/mips.
I think there could be yet another CONFIG option
to specifically enable the early_printk function
for the arches that use it.
The kernel/printk/early_printk() function seems
used only by arm/microblaze/tile/x86.
On Wed, Dec 17, 2014 at 9:58 PM, Joe Perches <[email protected]> wrote:
> It seems like early_printk can be configured into
> a few architectures but also appear not to be used.
>
> $ git grep -w "early_printk"
[...]
> arch/m68k/kernel/Makefile:obj-$(CONFIG_EARLY_PRINTK) += early_printk.o
[...]
> These seem to the only uses:
>
> arch/arm/mach-socfpga/socfpga.c: early_printk("Early printk initialized\n");
> []
> arch/microblaze/kernel/cpu/cpuinfo-pvr-full.c: early_printk("ERROR: Microblaze " x "-different for PVR and DTS\n");
> arch/microblaze/kernel/cpu/cpuinfo-static.c: early_printk("ERROR: Microblaze " x "-different for kernel and DTS\n");
> []
> arch/tile/kernel/early_printk.c: early_printk("Kernel panic - not syncing: %pV", &vaf);
> []
> arch/x86/kernel/e820.c: early_printk(msg);
> arch/x86/kernel/head64.c: early_printk("Kernel alive\n");
> arch/x86/kernel/head_64.S: call early_printk
> []
> kernel/events/core.c: early_printk("perf interrupt took too long (%lld > %lld), lowering "
>
> So blackfin, m68k, and mips seems to have it possible to enable,
> but also don't appear at first glance to use it,
>
> Is early_printk really used by these architectures?
> Should it be removed?
The m68k "early_printk" match is not about the early_printk()
function, but about
the early printk functionality. Hence nothing to remove there.
Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected]
In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds
Joe Perches <[email protected]> writes:
> On Fri, 2014-12-19 at 01:43 +0000, M?ns Rullg?rd wrote:
>> What exactly are you proposing to remove?
>
> Optionally compile out
> kernel/printk/printk.c:early_printk()
> even if CONFIG_EARLY_PRINTK is enabled.
>
>> I see no unused code related
>> to early printk (in any variant) under arch/mips.
>
> I think there could be yet another CONFIG option
> to specifically enable the early_printk function
> for the arches that use it.
Why bother? On MIPS it would save 132 bytes of compiled code.
> The kernel/printk/early_printk() function seems
> used only by arm/microblaze/tile/x86.
Rather than introduce more config complexity, you could try to remove
the 7 remaining uses of early_printk().
- arch/arm/mach-socfpga
Single early_printk("Early printk initialized\n") call serving no
apparent purpose can probably be safely deleted. Since there are no
other early_printk() calls, this information seems rather useless.
- arch/microblaze, arch/tile, arch/x86
These all do a register_console() for the early console, so regular
printk() should work. Moreover, x86 allows multiple early consoles,
but calling early_printk() will only output to the last one specified.
--
M?ns Rullg?rd
[email protected]
On Fri, 2014-12-19 at 10:33 +0000, M?ns Rullg?rd wrote:
> Joe Perches <[email protected]> writes:
> > The kernel/printk/early_printk() function seems
> > used only by arm/microblaze/tile/x86.
>
> Rather than introduce more config complexity, you could try to remove
> the 7 remaining uses of early_printk().
That's the general idea actually.
Dunno if that's possible at one go though.
cheers, Joe
On Wed 2014-12-17 12:58:47, Joe Perches wrote:
> It seems like early_printk can be configured into
> a few architectures but also appear not to be used.
>
> $ git grep -w "early_printk"
> arch/arm/kernel/Makefile:obj-$(CONFIG_EARLY_PRINTK) += early_printk.o
> arch/arm/kernel/early_printk.c: * linux/arch/arm/kernel/early_printk.c
> arch/arm/mach-socfpga/socfpga.c: early_printk("Early printk initialized\n");
> arch/blackfin/kernel/Makefile:obj-$(CONFIG_EARLY_PRINTK) += early_printk.o
> arch/blackfin/kernel/bfin_dma.c:#include <asm/early_printk.h>
> arch/blackfin/kernel/early_printk.c: * derived from arch/x86/kernel/early_printk.c
> arch/blackfin/kernel/early_printk.c:#include <asm/early_printk.h>
> arch/blackfin/kernel/setup.c:#include <asm/early_printk.h>
> arch/blackfin/kernel/shadow_console.c:#include <asm/early_printk.h>
> arch/blackfin/mm/init.c:#include <asm/early_printk.h>
> arch/ia64/sn/kernel/setup.c: * IO on SN2 is done via SAL calls, early_printk won't work without this.
> arch/m68k/kernel/Makefile:obj-$(CONFIG_EARLY_PRINTK) += early_printk.o
> arch/microblaze/kernel/Makefile:obj-$(CONFIG_EARLY_PRINTK) += early_printk.o
> arch/microblaze/kernel/cpu/cpuinfo-pvr-full.c: early_printk("ERROR: Microblaze " x "-different for PVR and DTS\n");
> arch/microblaze/kernel/cpu/cpuinfo-static.c: early_printk("ERROR: Microblaze " x "-different for kernel and DTS\n");
> arch/microblaze/kernel/setup.c:/* initialize device tree for usage in early_printk */
> arch/mips/ath25/Makefile:obj-$(CONFIG_EARLY_PRINTK) += early_printk.o
> arch/mips/ath79/Makefile:obj-$(CONFIG_EARLY_PRINTK) += early_printk.o
> arch/mips/bcm63xx/Makefile:obj-$(CONFIG_EARLY_PRINTK) += early_printk.o
> arch/mips/include/asm/mach-lantiq/falcon/lantiq_soc.h: * during early_printk no ioremap possible at this early stage
> arch/mips/include/asm/mach-lantiq/xway/lantiq_soc.h: * during early_printk no ioremap is possible
> arch/mips/kernel/Makefile:obj-$(CONFIG_EARLY_PRINTK) += early_printk.o
> arch/mips/lantiq/Makefile:obj-$(CONFIG_EARLY_PRINTK) += early_printk.o
> arch/mips/loongson/common/Makefile:obj-$(CONFIG_EARLY_PRINTK) += early_printk.o
> arch/mips/ralink/Makefile:obj-$(CONFIG_EARLY_PRINTK) += early_printk.o
> arch/tile/kernel/Makefile:obj-$(CONFIG_EARLY_PRINTK) += early_printk.o
> arch/tile/kernel/early_printk.c: early_printk("Kernel panic - not syncing: %pV", &vaf);
> arch/um/kernel/Makefile:obj-$(CONFIG_EARLY_PRINTK) += early_printk.o
> arch/unicore32/kernel/Makefile:obj-$(CONFIG_EARLY_PRINTK) += early_printk.o
> arch/unicore32/kernel/early_printk.c: * linux/arch/unicore32/kernel/early_printk.c
> arch/x86/kernel/Makefile:obj-$(CONFIG_EARLY_PRINTK) += early_printk.o
> arch/x86/kernel/e820.c: early_printk(msg);
> arch/x86/kernel/head64.c: early_printk("Kernel alive\n");
> arch/x86/kernel/head_64.S: call early_printk
> arch/x86/platform/efi/Makefile:obj-$(CONFIG_EARLY_PRINTK_EFI) += early_printk.o
> drivers/tty/serial/8250/8250_early.c: * and on early_printk.c by Andi Kleen.
> drivers/tty/serial/sn_console.c: * synchronous (raw) and asynchronous (buffered). initially, early_printk
> drivers/tty/serial/sn_console.c: /* without early_printk, we may be invoked late enough to race
> drivers/tty/serial/sn_console.c: /* early_printk invocation may have done this for us */
> include/linux/printk.h:void early_printk(const char *fmt, ...);
> include/linux/printk.h:void early_printk(const char *s, ...) { }
> kernel/events/core.c: early_printk("perf interrupt took too long (%lld > %lld), lowering "
> kernel/locking/lockdep.c: * We cannot printk in early bootup code. Not even early_printk()
> kernel/printk/printk.c:asmlinkage __visible void early_printk(const char *fmt, ...)
> kernel/printk/printk.c: * early_printk) - sometimes before setup_arch() completes - be careful
> kernel/printk/printk.c: * There are two types of consoles - bootconsoles (early_printk) and
>
> These seem to the only uses:
>
> arch/arm/mach-socfpga/socfpga.c: early_printk("Early printk initialized\n");
> []
> arch/microblaze/kernel/cpu/cpuinfo-pvr-full.c: early_printk("ERROR: Microblaze " x "-different for PVR and DTS\n");
> arch/microblaze/kernel/cpu/cpuinfo-static.c: early_printk("ERROR: Microblaze " x "-different for kernel and DTS\n");
> []
> arch/tile/kernel/early_printk.c: early_printk("Kernel panic - not syncing: %pV", &vaf);
> []
> arch/x86/kernel/e820.c: early_printk(msg);
> arch/x86/kernel/head64.c: early_printk("Kernel alive\n");
> arch/x86/kernel/head_64.S: call early_printk
> []
> kernel/events/core.c: early_printk("perf interrupt took too long (%lld > %lld), lowering "
>
> So blackfin, m68k, and mips seems to have it possible to enable,
> but also don't appear at first glance to use it,
>
> Is early_printk really used by these architectures?
> Should it be removed?
early_printk() is for debugging. Actually, feel free to remove:
arch/arm/mach-socfpga/socfpga.c: early_printk("Early printk initialized\n");
Its debugging output I added and should have been removed. So... early_printk()s
_should_ be unused in the mainline code, but functionality should be
kept, so that when someone needs it, he can enable it.
Pavel
--
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html
On Fri 2014-12-19 10:33:42, M?ns Rullg?rd wrote:
> Joe Perches <[email protected]> writes:
>
> > On Fri, 2014-12-19 at 01:43 +0000, M?ns Rullg?rd wrote:
> >> What exactly are you proposing to remove?
> >
> > Optionally compile out
> > kernel/printk/printk.c:early_printk()
> > even if CONFIG_EARLY_PRINTK is enabled.
> >
> >> I see no unused code related
> >> to early printk (in any variant) under arch/mips.
> >
> > I think there could be yet another CONFIG option
> > to specifically enable the early_printk function
> > for the arches that use it.
>
> Why bother? On MIPS it would save 132 bytes of compiled code.
>
> > The kernel/printk/early_printk() function seems
> > used only by arm/microblaze/tile/x86.
>
> Rather than introduce more config complexity, you could try to remove
> the 7 remaining uses of early_printk().
>
> - arch/arm/mach-socfpga
> Single early_printk("Early printk initialized\n") call serving no
> apparent purpose can probably be safely deleted. Since there are no
> other early_printk() calls, this information seems rather useless.
Feel free to do that, but please keep early_printk() available...
...so that I can debug the socfpga early boot when needed.
Pavel
--
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html