On Wed, Oct 25, 2023 at 10:41:46AM -0700, D Scott Phillips wrote:
> Bjorn Helgaas <helgaas@xxxxxxxxxx> writes:
>
> > On Fri, Sep 29, 2023 at 05:20:36PM -0700, D Scott Phillips wrote:
> >> On Ampere Altra, PCIe hotplug is handled through ACPI. A side interface is
> >> also present to request system firmware control of attention LEDs. Add an
> >> ACPI PCI Hotplug companion driver to support attention LED control.
> >> ...
> >
> >> + arm_smccc_smc(REQUEST, LED_CMD, led_status(status), LED_ATTENTION,
> >> + pci_domain_nr(bus) | (PCI_SLOT(root_port->devfn) << 4), 0, 0,
> >
> > pci_domain_nr() returns "int" (normally 32 bits), but since this is an
> > ACPI system, the domain comes from _SEG, which is defined to be 16
> > bits (ACPI r6.5, sec 6.5.6).
> >
> > So it looks like ORing in the "slot << 4" clobbers the upper 12 bits
> > of _SEG.
> >
> > Is this code doing the right thing?
>
> Hi Bjorn,
>
> on these Altra platforms _SEG is limited within 0-11. I can add an `&
> 0xf` to pci_domain_nr() to make it clear that the segment number is
> encoded down into 4 bits in the smc request.
If the following looks OK, we're all set. I put these on pci/hotplug
for v6.7, thanks!
+static int set_attention_status(struct hotplug_slot *slot, u8 status)
+{
+ struct arm_smccc_res res;
+ struct pci_bus *bus;
+ struct pci_dev *root_port;
+ unsigned long flags;
+ u32 handle;
+ int ret = 0;
+
+ bus = slot->pci_slot->bus;
+ root_port = pcie_find_root_port(bus->self);
+ if (!root_port)
+ return -ENODEV;
+
+ local_irq_save(flags);
+ arm_smccc_smc(HANDLE_OPEN, led_service_id[0], led_service_id[1],
+ led_service_id[2], led_service_id[3], 0, 0, 0, &res);
+ if (res.a0) {
+ ret = -ENODEV;
+ goto out;
+ }
+ handle = res.a1 & 0xffff0000;
+
+ arm_smccc_smc(REQUEST, LED_CMD, led_status(status), LED_ATTENTION,
+ PCI_SLOT(root_port->devfn) << 4 | pci_domain_nr(bus) & 0xf,
+ 0, 0, handle, &res);
+ if (res.a0)
+ ret = -ENODEV;
+
+ arm_smccc_smc(HANDLE_CLOSE, handle, 0, 0, 0, 0, 0, 0, &res);
+
+ out:
+ local_irq_restore(flags);
+ return ret;
+}
