Re: [PATCH] PCI: rcar: Add L1 link state fix into data abort hook

[Bjorn Helgaas <helgaas@kernel.org>]

On Mon, Sep 14, 2020 at 03:36:08AM +0200, Marek Vasut wrote:
> On 9/13/20 7:22 PM, Bjorn Helgaas wrote:
> 
> [...]
> 
> >> The R-Car PCIe controller is capable of handling L0s/L1 link states.
> >> While the controller can enter and exit L0s link state, and exit L1
> >> link state, without any additional action from the driver, to enter
> >> L1 link state, the driver must complete the link state transition by
> >> issuing additional commands to the controller.
> > 
> > So IIUC an R-Car Root Port can't enter L1 without help from the
> > driver?  That *sounds* like a hardware defect, since ASPM is supposed
> > to be "hardware-autonomous" once configured.
> 
> The driver needs to complete the switch from L0 to L1 link state, yes.
> 
> >> The problem is, this transition is not atomic. The controller sets
> >> PMEL1RX bit in PMSR register upon reception of PM_ENTER_L1 DLLP from
> >> the PCIe card, but then the controller enters some sort of inbetween
> >> state. The driver must detect this condition and complete the link
> >> state transition, by setting L1IATN bit in PMCTLR and waiting for
> >> the link state transition to complete.
> >>
> >> If a PCIe access happens inside of this window, where the controller
> >> is inbetween L0 and L1 link states, the access generates a fault and
> >> the ARM 'imprecise external abort' handler is invoked.
> > 
> > Let me see if I understand this.
> > 
> >   1) Endpoint sends PM_ENTER_L1
> >   2) R-Car Root Port receives PM_ENTER_L1, sets PMEL1RX
> >   3) R-Car driver does something to complete transition to L1
> > 
> > And if the Endpoint driver does an MMIO access between 2) and 3), it
> > causes an external abort?
> 
> Yes, because the link is in this inbetween state. So you need 3) to
> complete that transition.
> 
> > I couldn't find anything in the R-Car driver today that completes the
> > transition to L1.
> 
> That's because this patch adds it.

So with this patch, we finish the transition to L1 and immediately
return to L0.  That must mean that even in this in-between state we
save some power?

And the link may stay in that in-between state indefinitely, until the
CPU does an MMIO access or the device initiates a DMA?  But it will
never spend any time in L1 because the link never actually *gets* to
L1 until some event that will take it back to L0 occurs?

> > And the MMIO access will immediately bring the link back to L0, won't
> > it?
> 
> Yes
> 
> > Is there any benefit to L1 at all for this Root Port?  If it can't
> > enter L1 and just stays in the in-between state until some event that
> > will bring it *out* of L1, maybe we just need a quirk to stop
> > advertising support for L1 in the first place.
> 
> Power saving I would say.
> 
> >> Just like other PCI controller drivers, here we hook the fault handler,
> >> perform the fixup to help the controller enter L1 link state, and then
> >> restart the instruction which triggered the fault. Since the controller
> >> is in L1 link state now, the link can exit from L1 link state to L0 and
> >> successfully complete the access.
> >>
> >> Note that this fixup is applicable only to Aarch32 R-Car controllers,
> >> the Aarch64 R-Car perform the same fixup in TFA, see TFA commit
> >> 0969397f2 ("rcar_gen3: plat: Prevent PCIe hang during L1X config access")
> > 
> > TFA?  This doesn't seem to be an upstream Linux commit; can you give a
> > more specific reference?
> 
> This is the Trusted Firmware from ARM, in this case it behaves similar
> to ACPI. See e.g.:
> 
> https://github.com/ARM-software/arm-trusted-firmware/commit/0969397f295621aa26b3d14b76dd397d22be58bf

Thanks for that; please include it in the commit log for the next
version.

Bjorn