Re: [RFT PATCH v3 16/27] irqchip/apple-aic: Add support for the Apple Interrupt Controller

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Hi Will,

On 29/03/2021 21.04, Will Deacon wrote:
One CPU still needs to be able to mutate the flags of another CPU to fire an
IPI; AIUI the per-cpu ops are *not* atomic for concurrent access by multiple
CPUs, and in fact there is no API for that, only for "this CPU".

Huh, I really thought we had an API for that, but you're right. Oh well! But
I'd still suggest a per-cpu atomic_t in that case, rather than the array.

Yeah, after digging into the per-cpu stuff earlier and understanding how it works, I agree that a per-cpu atomic makes sense here. Switched it to that (which simplified out a bunch of smp_processor_id() calls too). Thanks!

I think a more idiomatic (and portable) way to do this would be to use
the relaxed accessors, but with smp_mb__after_atomic() between them. Do you
have a good reason for _not_ doing it like that?

Not particularly, other than symmetry with the case below.

I think it would be better not to rely on arm64-specific ordering unless
there's a good reason to.

Sounds reasonable, I'll switch to the barrier version.

We do need the return data here, and the release semantics (or another
barrier before it). But the read below can be made relaxed and a barrier
used instead, and then the same patern above except with a plain
atomic_or().

Yes, I think using atomic_fetch_or() followed by atomic_read() would be
best (obviously with the relevant comments!)

atomic_fetch_or_release is sufficient here (atomic_fetch_or is stronger; atomic_fetch_or_relaxed would not be strong enough as this needs to be ordered after any writes prior to sending the IPI; in this case release semantics also make logical sense).

It is ordered, right? As the comment says, it "needs to be ordered after the
aic_ic_write() above". atomic_fetch_andnot() is *supposed* to be fully
ordered and that should include against the writel_relaxed() on
AIC_IPI_FLAG. On ARM it turns out it's not quite fully ordered, but the
acquire semantics of the read half are sufficient for this case, as they
guarantee the flags are always read after the FIQ has been ACKed.

Sorry, I missed that the answer to my question was already written in the
comment. However, I'm still a bit unsure about whether the memory barriers
give you what you need here. The barrier in atomic_fetch_andnot() will
order the previous aic_ic_write(AIC_IPI_ACK) for the purposes of other
CPUs reading those locations, but it doesn't say anything about when the
interrupt controller actually changes state after the Ack.

Given that the AIC is mapped Device-nGnRnE, the Arm ARM offers:

   | Additionally, for Device-nGnRnE memory, a read or write of a Location
   | in a Memory-mapped peripheral that exhibits side-effects is complete
   | only when the read or write both:
   |
   | * Can begin to affect the state of the Memory-mapped peripheral.
   | * Can trigger all associated side-effects, whether they affect other
   |   peripheral devices, PEs, or memory.

so without AIC documentation I can't tell whether completion of the Ack write
just begins the process of an Ack (in which case we might need something like
a read-back), or whether the write response back from the AIC only occurs once
the Ack has taken effect. Any ideas?

Ahh, you're talking about latency within AIC itself... I obviously don't have an authoritative answer to this, though the hardware designer in me wants to say this really ought to be single-cycle type stuff that isn't internally pipelined in a way that would create races.

I tried to set up an SMP test case for the atomic-to-AIC sequence in m1n1, but unfortunately I couldn't hit the race window in deliberately racy code (i.e. ack after clearing flags) without widening it even further with at least one dummy load in between, and of course I didn't experience any races with the proper code either.

What I can say is that a simple set IPI; ack IPI (in adjacent str instructions) sequence always yields a cleared IPI, and the converse always yields a set IPI. So if there is latency to the operations it seems it would at least be the same for sets and acks and would imply readbacks block, which should still yield equivalently correct results. But of course this is a single-CPU test, so it is not fully representative of what could happen in an SMP scenario.

At this point all I can say is I'm inclined to shrug and say we have no evidence of this being something that can happen, and it shouldn't in sane hardware, and hope for the best :-)

Thanks,
--
Hector Martin (marcan@xxxxxxxxx)
Public Key: https://mrcn.st/pub




[Index of Archives]     [Device Tree Compilter]     [Device Tree Spec]     [Linux Driver Backports]     [Video for Linux]     [Linux USB Devel]     [Linux PCI Devel]     [Linux Audio Users]     [Linux Kernel]     [Linux SCSI]     [XFree86]     [Yosemite Backpacking]


  Powered by Linux