a question about reset16.lds

[ali hagigat]

SECTIONS {
    /* Trigger an error if I have an unuseable start address */
    _bogus = ASSERT(_start >= 0xffff0000, "_start too low. Please
decrease CONFIG_ROM_IMAGE_SIZE");
    _ROMTOP = 0xfffffff0;
    . = _ROMTOP;
    .reset . : {
    	*(.reset)
    	. = 15 ;
    	BYTE(0x00);
    }
}
------------------------------------------------------------------
What does (.=15;) mean? It seems a mistake! and can be: . = .+1;

[Patrick Georgi]

Am 01.02.2011 12:46, schrieb ali hagigat:
> SECTIONS {
>    /* Trigger an error if I have an unuseable start address */
>    _bogus = ASSERT(_start>= 0xffff0000, "_start too low. Please
> decrease CONFIG_ROM_IMAGE_SIZE");
>    _ROMTOP = 0xfffffff0;
>    . = _ROMTOP;
>    .reset . : {
>    	*(.reset)
>    	. = 15 ;
>    	BYTE(0x00);
>    }
> }
>
> What does (.=15;) mean? It seems a mistake! and can be: . = .+1;

. = 15; means "set the location pointer to 15". With the byte that's 
written right after that, this means that .reset is exactly 16 bytes.
If it's less, ". = 15" pads the section to be larger, if it would have 
to be larger, ". = 15" makes the linker fail (because the location 
pointer must not go backwards).

I guess that also answers why ". = .+1" is no adequate substitute.

Patrick

[ali hagigat]

Thank you for the reply.

Lets consider the lines of that linker script again! First the
location counter is set to 0xfffffff0 and the output .reset section
starts from VMA=0xfffffff0. I think we both agree on that so far.
But after .reset section, .=15 forces the location counter to be 15!!
It means the location counter suddenly jumps from the top of memory to
some where in the bottom! and then linker asks for storing 0x00 in the
address of 15! or at the beginning of memory.
.=15 does not set the size of .reset as you wrote.
My suggestion was .=.+1 instead of .=15, or set VMA to one byte after
the final .reset and then ask for writing 0x00 at the address of
0xffffffff.
We are using . in the linker script file, it means VMA (virtual memory
address)    .....

On Tue, Feb 1, 2011 at 8:15 PM, Patrick Georgi  wrote:
> Am 01.02.2011 12:46, schrieb ali hagigat:
> > SECTIONS {
> >        /* Trigger an error if I have an unuseable start address */
> >        _bogus = ASSERT(_start>= 0xffff0000, "_start too low. Please
> > decrease CONFIG_ROM_IMAGE_SIZE");
> >        _ROMTOP = 0xfffffff0;
> >        . = _ROMTOP;
> >        .reset . : {
> >                *(.reset)
> >                . = 15 ;
> >                BYTE(0x00);
> >        }
> > }
> >
> > What does (.=15;) mean? It seems a mistake! and can be: . = .+1;
>
> . = 15; means "set the location pointer to 15". With the byte that's written
> right after that, this means that .reset is exactly 16 bytes.
> If it's less, ". = 15" pads the section to be larger, if it would have to be
> larger, ". = 15" makes the linker fail (because the location pointer must
> not go backwards).
>
> I guess that also answers why ". = .+1" is no adequate substitute.
>
> Patrick

[Patrick Georgi]

Am 05.02.2011 05:55, schrieb ali hagigat:
> Lets consider the lines of that linker script again! First the
> location counter is set to 0xfffffff0 and the output .reset section
> starts from VMA=0xfffffff0. I think we both agree on that so far.
> But after .reset section, .=15 forces the location counter to be 15!!

Inside a section definition, the location pointer is relative to the
start of the section. The section's location isn't always known
beforehand (but you're right that it is in our case).
> It means the location counter suddenly jumps from the top of memory to
> some where in the bottom! and then linker asks for storing 0x00 in the
> address of 15! or at the beginning of memory.

That would lead to a linker error. If you try ". = 0" at the place, the
linker will fail: cannot move location counter backwards (from fffffffd
to fffffff0)
> .=15 does not set the size of .reset as you wrote.
> My suggestion was .=.+1 instead of .=15, or set VMA to one byte after
> the final .reset and then ask for writing 0x00 at the address of
> 0xffffffff.

This relies on knowledge that *(.reset) yields 15 bytes of data, and
might break on runtime if it's less, or on build time if it's more.
With ". = 15" you get deterministic behaviour on build time.

Patrick