2010-06-05 18:10:23

by Mike Frysinger

[permalink] [raw]
Subject: [PATCH 1/2] vmlinux.lds.h: allow people to set a smaller rootfs alignment

Add a new INIT_RAMFS() define to compliment the existing INIT_RAM_FS, but
this one lets people control the alignment.

Signed-off-by: Mike Frysinger <[email protected]>
---
include/asm-generic/vmlinux.lds.h | 7 ++++---
1 files changed, 4 insertions(+), 3 deletions(-)

diff --git a/include/asm-generic/vmlinux.lds.h b/include/asm-generic/vmlinux.lds.h
index 48c5299..9ca268f 100644
--- a/include/asm-generic/vmlinux.lds.h
+++ b/include/asm-generic/vmlinux.lds.h
@@ -619,14 +619,15 @@
VMLINUX_SYMBOL(__security_initcall_end) = .;

#ifdef CONFIG_BLK_DEV_INITRD
-#define INIT_RAM_FS \
- . = ALIGN(PAGE_SIZE); \
+#define INIT_RAMFS(align) \
+ . = ALIGN(align); \
VMLINUX_SYMBOL(__initramfs_start) = .; \
*(.init.ramfs) \
VMLINUX_SYMBOL(__initramfs_end) = .;
#else
-#define INIT_RAM_FS
+#define INIT_RAMFS(align)
#endif
+#define INIT_RAM_FS INIT_RAMFS(PAGE_SIZE)

/*
* Default discarded sections.
--
1.7.1


2010-06-05 18:10:25

by Mike Frysinger

[permalink] [raw]
Subject: [PATCH 2/2] Blackfin: use smaller rootfs alignment

No need for the rootfs to be aligned to a page since we load it up and
free it during boot.

Signed-off-by: Mike Frysinger <[email protected]>
---
arch/blackfin/kernel/vmlinux.lds.S | 2 +-
1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/arch/blackfin/kernel/vmlinux.lds.S b/arch/blackfin/kernel/vmlinux.lds.S
index 984c781..622d780 100644
--- a/arch/blackfin/kernel/vmlinux.lds.S
+++ b/arch/blackfin/kernel/vmlinux.lds.S
@@ -158,7 +158,7 @@ SECTIONS
INIT_CALLS
CON_INITCALL
SECURITY_INITCALL
- INIT_RAM_FS
+ INIT_RAMFS(4)

. = ALIGN(4);
___per_cpu_load = .;
--
1.7.1

2010-06-07 23:27:57

by Tim Abbott

[permalink] [raw]
Subject: Re: [PATCH 1/2] vmlinux.lds.h: allow people to set a smaller rootfs alignment

Hi Mike,

Both patches look good. Please consider them:

Acked-by: Tim Abbott <[email protected]>

-Tim Abbott

On Sat, 5 Jun 2010, Mike Frysinger wrote:

> Add a new INIT_RAMFS() define to compliment the existing INIT_RAM_FS, but
> this one lets people control the alignment.
>
> Signed-off-by: Mike Frysinger <[email protected]>
> ---
> include/asm-generic/vmlinux.lds.h | 7 ++++---
> 1 files changed, 4 insertions(+), 3 deletions(-)
>
> diff --git a/include/asm-generic/vmlinux.lds.h b/include/asm-generic/vmlinux.lds.h
> index 48c5299..9ca268f 100644
> --- a/include/asm-generic/vmlinux.lds.h
> +++ b/include/asm-generic/vmlinux.lds.h
> @@ -619,14 +619,15 @@
> VMLINUX_SYMBOL(__security_initcall_end) = .;
>
> #ifdef CONFIG_BLK_DEV_INITRD
> -#define INIT_RAM_FS \
> - . = ALIGN(PAGE_SIZE); \
> +#define INIT_RAMFS(align) \
> + . = ALIGN(align); \
> VMLINUX_SYMBOL(__initramfs_start) = .; \
> *(.init.ramfs) \
> VMLINUX_SYMBOL(__initramfs_end) = .;
> #else
> -#define INIT_RAM_FS
> +#define INIT_RAMFS(align)
> #endif
> +#define INIT_RAM_FS INIT_RAMFS(PAGE_SIZE)
>
> /*
> * Default discarded sections.
> --
> 1.7.1
>
>

2010-06-09 07:20:41

by Milton Miller

[permalink] [raw]
Subject: Re: [PATCH 1/2] vmlinux.lds.h: allow people to set a smaller rootfs alignment


On Sat, 5 Jun 2010 at around 14:14:45 -0400, Mike Frysinger wrote:
> Add a new INIT_RAMFS() define to compliment the existing INIT_RAM_FS, but
> this one lets people control the alignment.
>
> diff --git a/include/asm-generic/vmlinux.lds.h b/include/asm-generic/vmlinux.lds.h
> index 48c5299..9ca268f 100644
> --- a/include/asm-generic/vmlinux.lds.h
> +++ b/include/asm-generic/vmlinux.lds.h
> @@ -619,14 +619,15 @@
> VMLINUX_SYMBOL(__security_initcall_end) = .;
>
> #ifdef CONFIG_BLK_DEV_INITRD
> -#define INIT_RAM_FS \
> - . = ALIGN(PAGE_SIZE); \
> +#define INIT_RAMFS(align) \
> + . = ALIGN(align); \
> VMLINUX_SYMBOL(__initramfs_start) = .; \
> *(.init.ramfs) \
> VMLINUX_SYMBOL(__initramfs_end) = .;
> #else
> -#define INIT_RAM_FS
> +#define INIT_RAMFS(align)
> #endif
> +#define INIT_RAM_FS INIT_RAMFS(PAGE_SIZE)
>

[ and patch 2/2 calls the new macro with align 4, for only blackfin ].

Mike,
As you identified, the alignment for the __initramfs is too large.

As can be seen by the grammer of the INIT_RAM_FS in [1], the section
itself always requires alignment of exactly 4 [2].

I realize it requires a bit more auditing and cordination, but we
should really just change the macro to align to 4.

[1] Documentation/early-userspace/buffer-format.txt,
[2] for the general case; as compressed fragments are aligin(1)

milton

2010-06-09 07:43:17

by Mike Frysinger

[permalink] [raw]
Subject: Re: [PATCH 1/2] vmlinux.lds.h: allow people to set a smaller rootfs alignment

On Wed, Jun 9, 2010 at 03:20, Milton Miller wrote:
> On Sat,  5 Jun 2010 at around 14:14:45 -0400, Mike Frysinger wrote:
>> Add a new INIT_RAMFS() define to compliment the existing INIT_RAM_FS, but
>> this one lets people control the alignment.
>>
>> diff --git a/include/asm-generic/vmlinux.lds.h b/include/asm-generic/vmlinux.lds.h
>> index 48c5299..9ca268f 100644
>> --- a/include/asm-generic/vmlinux.lds.h
>> +++ b/include/asm-generic/vmlinux.lds.h
>> @@ -619,14 +619,15 @@
>>               VMLINUX_SYMBOL(__security_initcall_end) = .;
>>
>>  #ifdef CONFIG_BLK_DEV_INITRD
>> -#define INIT_RAM_FS                                                  \
>> -     . = ALIGN(PAGE_SIZE);                                           \
>> +#define INIT_RAMFS(align)                                            \
>> +     . = ALIGN(align);                                               \
>>       VMLINUX_SYMBOL(__initramfs_start) = .;                          \
>>       *(.init.ramfs)                                                  \
>>       VMLINUX_SYMBOL(__initramfs_end) = .;
>>  #else
>> -#define INIT_RAM_FS
>> +#define INIT_RAMFS(align)
>>  #endif
>> +#define INIT_RAM_FS INIT_RAMFS(PAGE_SIZE)
>
> [ and patch 2/2 calls the new macro with align 4, for only blackfin ].
>
> Mike,
> As you identified, the alignment for the __initramfs is too large.
>
> As can be seen by the grammer of the INIT_RAM_FS in [1], the section
> itself always requires alignment of exactly 4 [2].
>
> I realize it requires a bit more auditing and cordination, but we
> should really just change the macro to align to 4.
>
> [1] Documentation/early-userspace/buffer-format.txt,
> [2] for the general case; as compressed fragments are aligin(1)

the back story is to bring the alignment back to the way it was before
for Blackfin systems, not to drop it down to some min value. so dont
credit me with too much here ;).

if we look at the alignments before the unification, we can easily see
why Tim was a bit cautious and started the common value at PAGE_SIZE:
alpha: PAGE_SIZE
arm: 32
avr32: 32
blackfin: 4
cris: none (so default from the input sections == 4?)
frv: 4096 (== PAGE_SIZE?)
h8300: 4
ia64: none (so default from the input sections == 4?)
m32r: 4096 (== PAGE_SIZE?)
m68k: 8192 (== 2x PAGE_SIZE?)
microblaze: 4096 (== PAGE_SIZE?)
mips: PAGE_SIZE
mn10300: PAGE_SIZE
parisc: PAGE_SIZE
powerpc: PAGE_SIZE
s390: 0x100
score: PAGE_SIZE
sh: PAGE_SIZE
sparc: PAGE_SIZE
x86: PAGE_SIZE
xtensa: 4096 (== PAGE_SIZE?)

i have no problem changing the default to 4 since it works fine on
Blackfin systems. your analysis of the source alignment also seems
reasonable ... i assume that the 8 byte fields in the source cpio
arent referenced directly, otherwise it'd cause problems on 64bit
systems that require 64bit values to be naturally aligned.

beyond that, i dont see why any larger alignment values would be
required since this source archive isnt executed directly. it gets
expanded into a ramfs and then freed with the rest of the init stuff.
-mike

2010-06-09 08:20:36

by Geert Uytterhoeven

[permalink] [raw]
Subject: Re: [PATCH 1/2] vmlinux.lds.h: allow people to set a smaller rootfs alignment

On Wed, Jun 9, 2010 at 09:42, Mike Frysinger <[email protected]> wrote:
> if we look at the alignments before the unification, we can easily see
> why Tim was a bit cautious and started the common value at PAGE_SIZE:

> m68k: 8192 (== 2x PAGE_SIZE?)

For Sun 3, PAGE_SIZE is 8192, not 4096. So it's the maximum page size on m68k.

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

2010-06-11 05:50:42

by H. Peter Anvin

[permalink] [raw]
Subject: Re: [PATCH 1/2] vmlinux.lds.h: allow people to set a smaller rootfs alignment

On 06/09/2010 12:42 AM, Mike Frysinger wrote:
>
> i have no problem changing the default to 4 since it works fine on
> Blackfin systems. your analysis of the source alignment also seems
> reasonable ... i assume that the 8 byte fields in the source cpio
> arent referenced directly, otherwise it'd cause problems on 64bit
> systems that require 64bit values to be naturally aligned.
>
> beyond that, i dont see why any larger alignment values would be
> required since this source archive isnt executed directly. it gets
> expanded into a ramfs and then freed with the rest of the init stuff.
>

We have talked for a long time about incrementally freeing the initramfs
as it is decoded, which might help systems which are using most of their
memory for an initramfs image...

-hpa

--
H. Peter Anvin, Intel Open Source Technology Center
I work for Intel. I don't speak on their behalf.

2010-06-11 07:30:56

by Milton Miller

[permalink] [raw]
Subject: Re: [PATCH 1/2] vmlinux.lds.h: allow people to set a smaller rootfs alignment

On Thu, 10 Jun 2010 at around 22:50:34 -0700, H. Peter Anvin wrote:
> On 06/09/2010 12:42 AM, Mike Frysinger wrote:
> >
> > i have no problem changing the default to 4 since it works fine on
> > Blackfin systems. your analysis of the source alignment also seems
> > reasonable ... i assume that the 8 byte fields in the source cpio
> > arent referenced directly, otherwise it'd cause problems on 64bit
> > systems that require 64bit values to be naturally aligned.
> >
> > beyond that, i dont see why any larger alignment values would be
> > required since this source archive isnt executed directly. it gets
> > expanded into a ramfs and then freed with the rest of the init stuff.
> >
>
> We have talked for a long time about incrementally freeing the initramfs
> as it is decoded, which might help systems which are using most of their
> memory for an initramfs image...

at which time the author of said theoritcal patch that might possibly
need page alignment would only need to update vmlinux.lds.h and not all
the architectures vmlinux.lds.S files.

milton