[PATCH 04/18] arm64: alternatives: Enforce alignment of struct alt_instr

[Marc Zyngier]

On 06/12/17 15:18, Konrad Rzeszutek Wilk wrote:
> On Wed, Dec 06, 2017 at 02:57:29PM +0000, Marc Zyngier wrote:
>> On 06/12/17 14:48, Konrad Rzeszutek Wilk wrote:
>>> On Wed, Dec 06, 2017 at 02:38:25PM +0000, Marc Zyngier wrote:
>>>> We're playing a dangerous game with struct alt_instr, as we produce
>>>> it using assembly tricks, but parse them using the C structure.
>>>> We just assume that the respective alignments of the two will
>>>> be the same.
>>>>
>>>> But as we add more fields to this structure, the alignment requirements
>>>> of the structure may change, and lead to all kind of funky bugs.
>>>>
>>>> TO solve this, let's move the definition of struct alt_instr to its
>>>> own file, and use this to generate the alignment constraint from
>>>> asm-offsets.c. The various macros are then patched to take the
>>>> alignment into account.
>>>
>>> Would it be better to use .p2align as on 32-bit ARM you must
>>> have it 4-byte aligned. Or at least have and BUILD_BUG_ON
>>> to make sure the size can be divided by four??
>>>
>>> Oh wait. You are not even touching ARM-32, how come? The alternative
>>> code can run on ARM-32 ...
>>
>> How? Given that I haven't written yet, I'd be grateful if you could
>> share your time machine...
> 
> Oh! I assumed it would be there as the Xen variant runs on ARM-32 and
> it borrowed a bunch of code from Linux. 

KVM definitely runs on 32bit, but doesn't have any alternative facility,
at least not in the way arm64 does.

> Please disregard my comment. I will go back to tweaking the time machine.

Let me know once you've debugged it, I have a couple of issues to
address in the past...

Thanks,

	M.
-- 
Jazz is not dead. It just smells funny...