[Q] x86 - boot/header.S

[Q] x86 - boot/header.S

[Cyrill Gorcunov]

Hi Peter, Sam,

could you take a look on x86/boot/header.S:280 please?

# Zero the bss
	movw	$__bss_start, %di
	movw	$_end+3, %cx
	xorl	%eax, %eax
	subw	%di, %cx
	shrw	$2, %cx
	rep; stosl

I wonder why is $_end there instead of $__bss_stop?
Well, accroding to vmlinux_32.lsd both _end and __bss_stop
are the same BUT __bss_stop is more convenient methink.
Would it be usefull to change?

		- Cyrill -

Re: [Q] x86 - boot/header.S

[Yinghai Lu]

On Sat, Feb 23, 2008 at 12:20 AM, Cyrill Gorcunov <gorcunov@gmail.com> wrote:
> Hi Peter, Sam,
>
>  could you take a look on x86/boot/header.S:280 please?
>
>  # Zero the bss
>         movw    $__bss_start, %di
>         movw    $_end+3, %cx
>         xorl    %eax, %eax
>         subw    %di, %cx
>         shrw    $2, %cx
>         rep; stosl
>
>  I wonder why is $_end there instead of $__bss_stop?
>  Well, accroding to vmlinux_32.lsd both _end and __bss_stop
>  are the same BUT __bss_stop is more convenient methink.
>  Would it be usefull to change?

we should have head32.c like head64.c
and x86_32_start_kernel.

Eric's patch long time ago...

YH

Re: [Q] x86 - boot/header.S

[Cyrill Gorcunov]

[Yinghai Lu - Sat, Feb 23, 2008 at 12:44:49AM -0800]
| On Sat, Feb 23, 2008 at 12:20 AM, Cyrill Gorcunov <gorcunov@gmail.com> wrote:
| > Hi Peter, Sam,
| >
| >  could you take a look on x86/boot/header.S:280 please?
| >
| >  # Zero the bss
| >         movw    $__bss_start, %di
| >         movw    $_end+3, %cx
| >         xorl    %eax, %eax
| >         subw    %di, %cx
| >         shrw    $2, %cx
| >         rep; stosl
| >
| >  I wonder why is $_end there instead of $__bss_stop?
| >  Well, accroding to vmlinux_32.lsd both _end and __bss_stop
| >  are the same BUT __bss_stop is more convenient methink.
| >  Would it be usefull to change?
| 
| we should have head32.c like head64.c
| and x86_32_start_kernel.
| 
| Eric's patch long time ago...
| 
| YH
| 

Hi Yinghai,

thanks for reply BUT that is not the point (or maybe I miss something).

Look, we only have head64.c - there is no head32.c at all. Both
vmlinux_32/64.lds defines _end exactly the same as __bss_stop. So in
code which DO fillup bss section with zeros the prefered name is
__bss_stop. The only thing I'm trying to say that it would be clean
naming scheme and I think it would help for further review - instead
of searching all over x86 files to find _end definition __bss_stop
tell us WHAT we are zeroing from the code.

		- Cyrill -

Re: [Q] x86 - boot/header.S

[Sam Ravnborg]

On Sat, Feb 23, 2008 at 12:07:39PM +0300, Cyrill Gorcunov wrote:
> [Yinghai Lu - Sat, Feb 23, 2008 at 12:44:49AM -0800]
> | On Sat, Feb 23, 2008 at 12:20 AM, Cyrill Gorcunov <gorcunov@gmail.com> wrote:
> | > Hi Peter, Sam,
> | >
> | >  could you take a look on x86/boot/header.S:280 please?
> | >
> | >  # Zero the bss
> | >         movw    $__bss_start, %di
> | >         movw    $_end+3, %cx
> | >         xorl    %eax, %eax
> | >         subw    %di, %cx
> | >         shrw    $2, %cx
> | >         rep; stosl
> | >
> | >  I wonder why is $_end there instead of $__bss_stop?
> | >  Well, accroding to vmlinux_32.lsd both _end and __bss_stop
> | >  are the same BUT __bss_stop is more convenient methink.
> | >  Would it be usefull to change?
> | 
> | we should have head32.c like head64.c
> | and x86_32_start_kernel.
> | 
> | Eric's patch long time ago...
> | 
> | YH
> | 
> 
> Hi Yinghai,
> 
> thanks for reply BUT that is not the point (or maybe I miss something).
> 
> Look, we only have head64.c - there is no head32.c at all. Both
> vmlinux_32/64.lds defines _end exactly the same as __bss_stop. So in
> code which DO fillup bss section with zeros the prefered name is
> __bss_stop. The only thing I'm trying to say that it would be clean
> naming scheme and I think it would help for further review - instead
> of searching all over x86 files to find _end definition __bss_stop
> tell us WHAT we are zeroing from the code.

We should introduce head32.c and then we can use the exact same function
as 64 bit does:

/* Don't add a printk in there. printk relies on the PDA which is not initialized
   yet. */
static void __init clear_bss(void)
{
        memset(__bss_start, 0,
               (unsigned long) __bss_stop - (unsigned long) __bss_start);
}

This answers your Q.

I already resubmitted Eric's original patch to introduce head32.c as I also
liked the cleanup in two Makefiles.

	Sam

Re: [Q] x86 - boot/header.S

[Cyrill Gorcunov]

[Sam Ravnborg - Sat, Feb 23, 2008 at 10:18:17AM +0100]
| On Sat, Feb 23, 2008 at 12:07:39PM +0300, Cyrill Gorcunov wrote:
| > [Yinghai Lu - Sat, Feb 23, 2008 at 12:44:49AM -0800]
| > | On Sat, Feb 23, 2008 at 12:20 AM, Cyrill Gorcunov <gorcunov@gmail.com> wrote:
| > | > Hi Peter, Sam,
| > | >
| > | >  could you take a look on x86/boot/header.S:280 please?
| > | >
| > | >  # Zero the bss
| > | >         movw    $__bss_start, %di
| > | >         movw    $_end+3, %cx
| > | >         xorl    %eax, %eax
| > | >         subw    %di, %cx
| > | >         shrw    $2, %cx
| > | >         rep; stosl
| > | >
| > | >  I wonder why is $_end there instead of $__bss_stop?
| > | >  Well, accroding to vmlinux_32.lsd both _end and __bss_stop
| > | >  are the same BUT __bss_stop is more convenient methink.
| > | >  Would it be usefull to change?
| > | 
| > | we should have head32.c like head64.c
| > | and x86_32_start_kernel.
| > | 
| > | Eric's patch long time ago...
| > | 
| > | YH
| > | 
| > 
| > Hi Yinghai,
| > 
| > thanks for reply BUT that is not the point (or maybe I miss something).
| > 
| > Look, we only have head64.c - there is no head32.c at all. Both
| > vmlinux_32/64.lds defines _end exactly the same as __bss_stop. So in
| > code which DO fillup bss section with zeros the prefered name is
| > __bss_stop. The only thing I'm trying to say that it would be clean
| > naming scheme and I think it would help for further review - instead
| > of searching all over x86 files to find _end definition __bss_stop
| > tell us WHAT we are zeroing from the code.
| 
| We should introduce head32.c and then we can use the exact same function
| as 64 bit does:
| 
| /* Don't add a printk in there. printk relies on the PDA which is not initialized
|    yet. */
| static void __init clear_bss(void)
| {
|         memset(__bss_start, 0,
|                (unsigned long) __bss_stop - (unsigned long) __bss_start);
| }
| 
| This answers your Q.
| 
| I already resubmitted Eric's original patch to introduce head32.c as I also
| liked the cleanup in two Makefiles.
| 
| 	Sam
| 

thanks Sam, i just saw your patch.

		- Cyrill -

Re: [Q] x86 - boot/header.S

[H. Peter Anvin]

Cyrill Gorcunov wrote:
> Hi Peter, Sam,
> 
> could you take a look on x86/boot/header.S:280 please?
> 
> # Zero the bss
> 	movw	$__bss_start, %di
> 	movw	$_end+3, %cx
> 	xorl	%eax, %eax
> 	subw	%di, %cx
> 	shrw	$2, %cx
> 	rep; stosl
> 
> I wonder why is $_end there instead of $__bss_stop?
> Well, accroding to vmlinux_32.lsd both _end and __bss_stop
> are the same BUT __bss_stop is more convenient methink.
> Would it be usefull to change?

x86/boot/header.S goes with x86/boot/setup.ld and no other linker script.

	-hpa

Re: [Q] x86 - boot/header.S

[H. Peter Anvin]

Cyrill Gorcunov wrote:
> Look, we only have head64.c - there is no head32.c at all.

That's true to some degree, but head64.c contains a bunch of stuff which 
lives in setup_32.c in the 32-bit world.

	-hpa

Re: [Q] x86 - boot/header.S

[Cyrill Gorcunov]

[H. Peter Anvin - Sun, Feb 24, 2008 at 06:15:52PM -0800]
> Cyrill Gorcunov wrote:
>> Hi Peter, Sam,
>> could you take a look on x86/boot/header.S:280 please?
>> # Zero the bss
>> 	movw	$__bss_start, %di
>> 	movw	$_end+3, %cx
>> 	xorl	%eax, %eax
>> 	subw	%di, %cx
>> 	shrw	$2, %cx
>> 	rep; stosl
>> I wonder why is $_end there instead of $__bss_stop?
>> Well, accroding to vmlinux_32.lsd both _end and __bss_stop
>> are the same BUT __bss_stop is more convenient methink.
>> Would it be usefull to change?
>
> x86/boot/header.S goes with x86/boot/setup.ld and no other linker script.
>
> 	-hpa
>

indeed... :( anyway, setup.ld has the definition of __bss_stop too
though in this case __bss_stop is not equal to _end BUT
[__bss_start;__bss_stop] do cover *bss section anyway.

According to Sam's last post it will not be a problem anymore
'cause of memset further usage.

Thanks for comments, Peter.

		- Cyrill -