Re: [PATCH 1/4] x86/thread_info: introduce ->ftrace_int3_stack member

[Andy Lutomirski]

> On Apr 28, 2019, at 2:22 PM, Nicolai Stange <nstange@xxxxxxx> wrote:
> 
> Steven Rostedt <rostedt@xxxxxxxxxxx> writes:
> 
>> On Sun, 28 Apr 2019 10:41:10 -0700
>> Andy Lutomirski <luto@xxxxxxxxxxxxxx> wrote:
>> 
>> 
>>>> Note that at any given point
>>>> in time, there can be at most four such call insn emulations pending:
>>>> namely at most one per "process", "irq", "softirq" and "nmi" context.
>>>> 
>>> 
>>> Thatâs quite an assumption. I think your list should also contain
>>> exception, exceptions nested inside that exception, and machine
>>> check, at the very least.  Iâm also wondering why irq and softirq are
>>> treated separately.
> 
> You're right, I missed the machine check case.
> 
> 
>> 4 has usually been the context count we choose. But I guess in theory,
>> if we get exceptions then I could potentially be more.
> 
> After having seen the static_call discussion, I'm in no way defending
> this limited approach here, but out of curiosity: can the code between
> the push onto the stack from ftrace_int3_handler() and the subsequent
> pop from the stub actually trigger an (non-#MC) exception? There's an
> iret inbetween, but that can fault only when returning to user space,
> correct?

IRET doesnât do any fancy masking, so #DB, NMI and regular IRQs should all be possible.

> 
> 
>> As for irq vs softirq, an interrupt can preempt a softirq. Interrupts
>> are enabled while softirqs are running. When sofirqs run, softirqs are
>> disabled to prevent nested softirqs. But interrupts are enabled again,
>> and another interrupt may come in while a softirq is executing.
>> 
> 
> Thanks,
> 
> Nicolai
> 
> 
> -- 
> SUSE Linux GmbH, GF: Felix ImendÃrffer, Jane Smithard, Graham Norton,
> HRB 21284 (AG NÃrnberg)