Re: [PATCH 5/8] x86, brk: Make extend_brk() available with va/pa.

[H. Peter Anvin]

Konrad Rzeszutek Wilk <konrad.wilk@xxxxxxxxxx> wrote:
>On Wed, Aug 21, 2013 at 02:35:36PM +0200, H. Peter Anvin wrote:
>> Global symbols are inaccessible in physical mode.
>
>Even if they are embedded in the assembler code and use
>GLOBAL(paging_enabled) ?

Yes, because the address is different in physical mode.  Think about it.  You could do a *function* like:

paging_enabled:
	mov (%esp),%edx
	xor %eax,%eax
	cmpl $PAGE_OFFSET,%edx
	setae %al
	ret

>> 
>> This is incidentally yet another example of "PV/weird platform
>violence", since in their absence it would be trivial to work around
>this by using segmentation.
>
>I don't follow why it could not.
>
>Why can't there be a __pa_symbol(paging_enabled) that is used. Won't
>that in effect allow you to check the contents of that 'global
>constant' even when you don't have paging enabled?

Yes.  But not once paging has been turned on.

>> >>As mentioned above, on 32bit before paging is enabled, we have to
>> >>access variables
>> >>with pa. So introduce a "bool is_phys" parameter to extend_brk(),
>and
>> >>convert va
>> >>to pa is it is true.
>> >
>> >Could you do it differently? Meaning have a global symbol
>> >(paging_enabled) which will be used by most of the functions you
>> >changed in this patch and the next ones? It would naturally be
>enabled
>> >when paging is on and __va addresses can be used. 
>> >
>> >That could also be used in the printk case to do a BUG_ON before
>paging
>> >is enabled on 32bit. Or perhaps use a different code path to deal
>with
>> >using __pa address. 
>> >
>> >? 

-- 
Sent from my mobile phone. Please excuse brevity and lack of formatting.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/