Re: [RFC -tip] x86: introduce ENTRY(KPROBE)_X86 assembly helpers to catch unbalanced declaration

[Alexander van Heukelum <heukelum@mailshack.com>]

On Sun, Nov 23, 2008 at 10:48:28PM +0300, Cyrill Gorcunov wrote:
> [Ingo Molnar - Sun, Nov 23, 2008 at 08:31:34PM +0100]
> ...
> | > 
> | > Just got an error in implementation -- we have to support nested
> | > ENTRY without problem. Will check. What a surprise :-)
> | 
> | do you mean:
> | 
> |   ENTRY(system_call)
> |   ENTRY(system_call_after_swapgs)
> |   ...
> |   END(system_call)
> | 
> | that's more of a bug - system_call_after_swapgs is not a real entry 
> | point, we just need the label of it. Perhaps something like __ENTRY() 
> | for that case would be enough.
> | 
> | nor is this one real:
> | 
> |   ENTRY(interrupt)
> |   ENTRY(irq_entries_start)
> |   ...
> |   END(irq_entries_start)
> |   END(interrupt)
> | 
> | do we really need .irq_entries_start?
> | 
> | I think in general we should define a flat hierarchy of entries.
> | 
> | 	Ingo
> | 
> 
> Yeah, I meant these cases. I don't think we really need irq_entries_start
> (didn't find any mention of them in tree). In case of system_call_after_swapgs
> I'm not that sure, but since xen use it as a plain jmp (at least now) it
> could be converted to a plain label. Ingo, I'll continue tomorrow evening --
> have some other things to be done :)

The problem is that ENTRY(interrupt) is done in init.rodata, and
ENTRY(irq_entries_start) is done in .text. So inside the .S-file,
they are nested, but in the .o-file they are separate. Instead of
removing ENTRY(irq_entries_start), I think we should just expand to:


	.section .init.rodata,"a"
	.p2align 5
	.global interrupt
interrupt:

and

	size interrupt, .-interrupt

But the only importance I can think of is that this keeps both
the "interrupt" array and irq_entries_start visible in debugging
information.

Alternatively, we could probably do away with the interrupt
array entirely. We _know_ how the irq stubs are structured and
irq_entries_start is in principle enough information to reconstruct
all information in interrupt.

Anyhow, I too think that supporting nested ENTRY/END in .S files is
unnecessary.

Alexander

> 		- Cyrill -