Re: [PATCH] setup.c: introduce __pa_symbol() and get ride of CPHYSADDR()

[Franck Bui-Huu <vagabon.xyz@gmail.com>]

Thiemo Seufer wrote:
> Franck Bui-Huu wrote:
>> sorry to be ignorant of 64 bit kernels, but what's the point
>> to load them in KSEG0.
> 
> Smaller code with better performance.
> 

you mean we get smaller code _only_ by using the short 2 instructions
you described below ?

>>> and use short 2-instruction symbol references there.
>> do you mean "it allows to use only 2 'lui' instructions to load
>> a symbol address into a register" ?
> 
> It allows a 2-instruction "lui ; addiu" sequence instead of a
> 6-instruction "lui ; lui ; addiu ; addiu ; dsll32 ; addu" sequence.
> 
[snip]
>>
>> 	code_resource.start = virt_to_phys(&_text);
>> 	code_resource.end = virt_to_phys(&_etext) - 1;
>> 	data_resource.start = virt_to_phys(&_etext);
>> 	data_resource.end = virt_to_phys(&_edata) - 1;
>>
>> How does it work in this case ?
> 
> Those are addresses in 64-bit space, no special handling is needed
> there.

hm I'missing something there. Let's say that '&_text' is in KSEG0 and
is equal to 0xffffffff80000000. In this case virt_to_phys() returns
0x57ffffff80000000 (with PAGE_OFFSET = 0xa800000000000000). Is this 
physical address correct ??

> 
> The same doesn't hold for the initrd addresses supplied by the (32-bit)
> firmware. The firmware doesn't convert the kernel parameters to 64-bit
> values because the O2 kernel used to allow a pure 32-bit build, and the
> firmware can't find out what's actually inside the object file.
> 

This should be already handled by this code taken from setup.c:

static int __init rd_start_early(char *p)
{
        unsigned long start = memparse(p, &p);

#ifdef CONFIG_64BIT
        /* HACK: Guess if the sign extension was forgotten */
        if (start > 0x0000000080000000 && start < 0x00000000ffffffff)
                start |= 0xffffffff00000000UL;
#endif
        initrd_start = start;
        initrd_end += start;

        return 0;
}

Thanks
		Franck