[CRIU] [PATCH] pie: relocs -- Fix compilation on ARM

[Pavel Emelyanov]

On 06/09/2015 11:55 PM, Cyrill Gorcunov wrote:
> On Tue, Jun 09, 2015 at 11:49:38PM +0300, Pavel Emelyanov wrote:
>> On 06/09/2015 11:47 PM, Cyrill Gorcunov wrote:
>>> On Tue, Jun 09, 2015 at 11:26:06PM +0300, Pavel Emelyanov wrote:
>>>>
>>>> This routine is called two times with last two arguments being the same -- elf_relocs
>>>> and sizeof(elf_relocs). Why can't we stop treating them as arguments and just reference
>>>> in the elf_relocs_apply as (global) variables?
>>>
>>> They are generated dynamically and not present on the systems which use old scheme
>>> (arm and such), so I don't see a reason why to make code more messy and define them
>>> somewhere as a global for those who don't even touch them.
>>
>> Look at parasite_infect_siezed(). It calls elf_relocs_apply() and passes
>> elf_relocs and ARRAY_SIZE(of it) as two last arguments and this elf_relocs
>> is NOT local variable to parasite_infect_siezed(), it's global. Why not
>> just reference one directly in parasite_infect_siezed()?
> 
> Where to get it from? 

Where does parasite_infect_siezed() get it from?

> Do you propose to _allocate_ elf_relocs on the machines
> which do not use this function at all? What the point? Currently, when arm
> uses old scheme the preprocessor defines all things to nop, if we define
> some empty elf_relocs it won't be nop anymore, but eat some tiny byte of
> memory, so I fail to see what the benefit in defining global @elf_relocs.
> Or you mean to use @elf_relocs inside elf_relocs_apply itself cutting off
> them from the argumenst?
> .
>