Re: [PATCH] PCI: hotplug: Add extension driver for Ampere Altra hotplug LED control

From: Rafael J. Wysocki
Date: Fri Sep 29 2023 - 15:06:18 EST


On Thu, Sep 28, 2023 at 5:47 PM Bjorn Helgaas <helgaas@xxxxxxxxxx> wrote:
>
> On Wed, Sep 27, 2023 at 01:23:47PM -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.
> >
> > Signed-off-by: D Scott Phillips <scott@xxxxxxxxxxxxxxxxxxxxxx>
> > ---
> > drivers/pci/hotplug/Kconfig | 13 ++
> > drivers/pci/hotplug/Makefile | 3 +-
> > drivers/pci/hotplug/acpiphp_ampere_altra.c | 141 +++++++++++++++++++++
> > 3 files changed, 156 insertions(+), 1 deletion(-)
> > create mode 100644 drivers/pci/hotplug/acpiphp_ampere_altra.c
> >
> > diff --git a/drivers/pci/hotplug/Kconfig b/drivers/pci/hotplug/Kconfig
> > index 48113b210cf93..9fde600a9ad3e 100644
> > --- a/drivers/pci/hotplug/Kconfig
> > +++ b/drivers/pci/hotplug/Kconfig
> > @@ -61,6 +61,19 @@ config HOTPLUG_PCI_ACPI
> >
> > When in doubt, say N.
> >
> > +config HOTPLUG_PCI_ACPI_AMPERE_ALTRA
> > + tristate "ACPI PCI Hotplug driver Ampere Altra extensions"
> > + depends on HOTPLUG_PCI_ACPI
> > + depends on HAVE_ARM_SMCCC_DISCOVERY
> > + depends on m
>
> Why is this restricted to being a module? It's not unprecedented, but
> unless this only works as a module for some reason, I would leave that
> choice up to the user.
>
> > + help
> > + Say Y here if you have an Ampere Altra system.
> > +
> > + To compile this driver as a module, choose M here: the
> > + module will be called acpiphp_ampere_altra.
> > +
> > + When in doubt, say N.
> > +
> > config HOTPLUG_PCI_ACPI_IBM
> > tristate "ACPI PCI Hotplug driver IBM extensions"
> > depends on HOTPLUG_PCI_ACPI
> > diff --git a/drivers/pci/hotplug/Makefile b/drivers/pci/hotplug/Makefile
> > index 5196983220df6..29d7f6171b305 100644
> > --- a/drivers/pci/hotplug/Makefile
> > +++ b/drivers/pci/hotplug/Makefile
> > @@ -21,8 +21,9 @@ obj-$(CONFIG_HOTPLUG_PCI_RPA_DLPAR) += rpadlpar_io.o
> > obj-$(CONFIG_HOTPLUG_PCI_ACPI) += acpiphp.o
> > obj-$(CONFIG_HOTPLUG_PCI_S390) += s390_pci_hpc.o
> >
> > -# acpiphp_ibm extends acpiphp, so should be linked afterwards.
> > +# acpiphp_ibm extend acpiphp, so should be linked afterwards.
> >
> > +obj-$(CONFIG_HOTPLUG_PCI_ACPI_AMPERE_ALTRA) += acpiphp_ampere_altra.o
> > obj-$(CONFIG_HOTPLUG_PCI_ACPI_IBM) += acpiphp_ibm.o
> >
> > pci_hotplug-objs := pci_hotplug_core.o
> > diff --git a/drivers/pci/hotplug/acpiphp_ampere_altra.c b/drivers/pci/hotplug/acpiphp_ampere_altra.c
> > new file mode 100644
> > index 0000000000000..8692b939dea78
> > --- /dev/null
> > +++ b/drivers/pci/hotplug/acpiphp_ampere_altra.c
> > @@ -0,0 +1,141 @@
> > +// SPDX-License-Identifier: GPL-2.0
> > +/*
> > + * ACPI PCI Hot Plug Ampere Altra Extension
>
> Would be helpful to include a hint about what this module *does*.
> IIUC, it controls the attention indicator.
>
> > + *
> > + * Copyright (C) 2023 Ampere Computing LLC
> > + *
>
> Spurious blank line.
>
> > + */
> > +
> > +#define pr_fmt(fmt) "acpiphp_ampere_altra: " fmt
> > +
> > +#include <linux/init.h>
> > +#include <linux/module.h>
> > +#include <linux/pci.h>
> > +#include <linux/pci_hotplug.h>
> > +
> > +#include "acpiphp.h"
> > +
> > +#define HANDLE_OPEN 0xb0200000
> > +#define HANDLE_CLOSE 0xb0300000
> > +#define REQUEST 0xf0700000
> > +#define LED_CMD 0x00000004
> > +#define LED_ATTENTION 0x00000002
> > +#define LED_SET_ON 0x00000001
> > +#define LED_SET_OFF 0x00000002
> > +#define LED_SET_BLINK 0x00000003
> > +
> > +static const struct acpi_device_id acpi_ids[] = {
> > + {"AMPC0008", 0}, {}
> > +};
> > +MODULE_DEVICE_TABLE(acpi, acpi_ids);
> > +
> > +static u32 led_service_id[4];
> > +
> > +static int led_status(u8 status)
> > +{
> > + switch (status) {
> > + case 1: return LED_SET_ON;
> > + case 2: return LED_SET_BLINK;
> > + default: return LED_SET_OFF;
> > + }
> > +}
> > +
> > +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_domain_nr(bus) | ((root_port->devfn >> 3) << 4), 0, 0,
>
> PCI_SLOT(root_port->devfn)
>
> > + 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;
> > +}
> > +
> > +static int get_attention_status(struct hotplug_slot *slot, u8 *status)
> > +{
> > + return -EINVAL;
> > +}
> > +
> > +static struct acpiphp_attention_info ampere_altra_attn = {
> > + .set_attn = set_attention_status,
> > + .get_attn = get_attention_status,
> > + .owner = THIS_MODULE,
> > +};
> > +
> > +static acpi_status __init get_acpi_handle(acpi_handle handle, u32 level,
> > + void *context, void **return_value)
> > +{
> > + *(acpi_handle *)return_value = handle;
> > + return AE_CTRL_TERMINATE;
> > +}
> > +
> > +static int __init acpiphp_ampere_altra_init(void)
> > +{
> > + struct fwnode_handle *fwnode;
> > + acpi_handle leds_handle = NULL;
> > + struct acpi_device *leds;
> > + acpi_status status;
> > + int ret;
> > +
> > + status = acpi_get_devices("AMPC0008", get_acpi_handle, NULL,
> > + &leds_handle);
>
> Rafael, can you comment on whether we should use acpi_get_devices(),
> acpi_bus_register_driver(), acpi_acpi_scan_add_handler(), or something
> else here?

Personally, I would go for a platform driver, because the ACPI core
should create a platform device for this object.

acpi_get_devices() carries out a namespace walk that is costly and
entirely avoidable.

> I try to avoid pci_get_device() because it subverts the
> driver model (no hotplug support, no driver/device binding).
>
> I see Documentation/driver-api/acpi/scan_handlers.rst, but I'm not
> clear on when to use acpi_bus_register_driver() vs

Never.

> acpi_acpi_scan_add_handler().

When you don't want the ACPI core to create a platform device for your
ACPI device object. There are cases like that, but they are rare.

> I guess the only ACPI connection here is to retrieve the "uuid"
> property once, and there's no need for any ACPI services after that.

That's in agreement with my understanding.