On Mon, Apr 8, 2019 at 5:47 AM Daniel Bristot de Oliveira <bris...@redhat.com> wrote: > > On 4/4/19 2:01 AM, Andy Lutomirski wrote: > >> To resolve this problem, the set/unset of the IRQ/NMI context needs to > >> be done before the execution of the first C execution, and after its > >> return. By doing so, and using this method to identify the context in the > >> trace recursion protection, no more events are lost. > > I would much rather do the opposite: completely remove context > > tracking from the asm and, instead, stick it into the C code. We'd > > need to make sure that the C code is totally immune from tracing, > > kprobes, etc, but it would be a nice cleanup. And then you could fix > > this bug in C! > > > > > > Humm... what we could do to have things in C is to set the variable right at > the > begin of the C handler, e.g., do_IRQ(), and right before the return. > > But by doing this we would have a problem with two things: > > 1) irq handler itself (e.g., do_IRQ()) > 2) functions/tracepoints that might run before and after the handler execution > (e.g., preemptirq tracer), but still in the IRQ context. > > We can work around the first case by checking if (the function is in the > __irq_entry .text section) in the recursion control. > > The second case would still be a problem. For instance, the preemptirq: > tracepoints in the preemptirq tracer would be "dropped" in the case of a > miss-identification of a recursion. > > Thinking aloud: should we try to move the preemptirq tracers to the C part?
I think we should try to move as much as possible to the C part.
