On Fri, Aug 20, 2021 at 7:12 PM Bjorn Helgaas <helgaas@xxxxxxxxxx> wrote:
>
> [+cc Shanker, Amey]
>
> On Fri, Aug 20, 2021 at 06:12:39PM +0200, Rafael J. Wysocki wrote:
> > From: Rafael J. Wysocki <rafael.j.wysocki@xxxxxxxxx>
> >
> > On some systems, in order to get to the deepest low-power state of
> > the platform (which may be necessary to save significant enough
> > amounts of energy while suspended to idle. for example), devices on
> > the PCI bus exposed by the VMD driver need to be power-managed via
> > ACPI. However, the layout of the ACPI namespace below the VMD
> > controller device object does not reflect the layout of the PCI bus
> > under the VMD host bridge, so in order to identify the ACPI companion
> > objects for the devices on that bus, it is necessary to use a special
> > _ADR encoding on the ACPI side. In other words, acpi_pci_find_companion()
> > does not work for these devices, so it needs to be amended with a
> > special lookup logic specific to the VMD bus.
> >
> > Address this issue by allowing the VMD driver to temporarily install
> > an ACPI companion lookup hook containing the code matching the devices
> > on the VMD PCI bus with the corresponding objects in the ACPI
> > namespace.
> >
> > Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@xxxxxxxxx>
> > Tested-by: Wendy Wang <wendy.wang@xxxxxxxxx>
> > ---
> > drivers/pci/controller/vmd.c | 48 ++++++++++++++++++++++++++
> > drivers/pci/host-bridge.c | 1
> > drivers/pci/pci-acpi.c | 78 +++++++++++++++++++++++++++++++++++++++++++
> > include/linux/pci-acpi.h | 3 +
> > 4 files changed, 130 insertions(+)
> >
> > Index: linux-pm/drivers/pci/controller/vmd.c
> > ===================================================================
> > --- linux-pm.orig/drivers/pci/controller/vmd.c
> > +++ linux-pm/drivers/pci/controller/vmd.c
> > @@ -11,6 +11,7 @@
> > #include <linux/module.h>
> > #include <linux/msi.h>
> > #include <linux/pci.h>
> > +#include <linux/pci-acpi.h>
> > #include <linux/pci-ecam.h>
> > #include <linux/srcu.h>
> > #include <linux/rculist.h>
> > @@ -447,6 +448,49 @@ static struct pci_ops vmd_ops = {
> > .write = vmd_pci_write,
> > };
> >
> > +#ifdef CONFIG_ACPI
> > +static struct acpi_device *vmd_acpi_find_companion(struct pci_dev *pci_dev)
> > +{
> > + struct pci_host_bridge *bridge;
> > + u32 busnr, addr;
> > +
> > + if (pci_dev->bus->ops != &vmd_ops)
> > + return NULL;
> > +
> > + bridge = pci_find_host_bridge(pci_dev->bus);
> > + busnr = pci_dev->bus->number - bridge->bus->number;
> > + addr = (busnr << 24) | ((u32)pci_dev->devfn << 16) | 0x8000FFFFU;
> > +
> > + dev_dbg(&pci_dev->dev, "Looking for ACPI companion (address 0x%x)\n",
> > + addr);
> > +
> > + return acpi_find_child_device(ACPI_COMPANION(bridge->dev.parent), addr,
> > + false);
> > +}
> > +
> > +static bool hook_installed;
> > +
> > +static void vmd_acpi_begin(void)
> > +{
> > + if (pci_acpi_set_companion_lookup_hook(vmd_acpi_find_companion))
> > + return;
> > +
> > + hook_installed = true;
> > +}
> > +
> > +static void vmd_acpi_end(void)
> > +{
> > + if (!hook_installed)
> > + return;
> > +
> > + pci_acpi_clear_companion_lookup_hook();
> > + hook_installed = false;
> > +}
> > +#else
> > +static inline void vmd_acpi_begin(void) { }
> > +static inline void vmd_acpi_end(void) { }
> > +#endif /* CONFIG_ACPI */
> > +
> > static void vmd_attach_resources(struct vmd_dev *vmd)
> > {
> > vmd->dev->resource[VMD_MEMBAR1].child = &vmd->resources[1];
> > @@ -747,6 +791,8 @@ static int vmd_enable_domain(struct vmd_
> > if (vmd->irq_domain)
> > dev_set_msi_domain(&vmd->bus->dev, vmd->irq_domain);
> >
> > + vmd_acpi_begin();
> > +
> > pci_scan_child_bus(vmd->bus);
> > pci_assign_unassigned_bus_resources(vmd->bus);
>
> This affects the same path as Shanker's patch to set the
> ACPI_COMPANION earlier, in pci_setup_device() instead of in
> acpi_bind_one() [1].
Thanks for the pointer.
> I think this should still work correctly since vmd_acpi_begin()
> inserts the VMD-specific hook in acpi_pci_find_companion() before we
> call pci_scan_child_bus() to enumerate devices below VMD.
>
> So when Shanker's new pci_set_acpi_fwnode() calls
> acpi_pci_find_companion(), it should get to vmd_acpi_find_companion()
> even though it happens earlier than acpi_bind_one().
>
> Just FYI so you can double check.
Yes, it should still work.
> > @@ -760,6 +806,8 @@ static int vmd_enable_domain(struct vmd_
> >
> > pci_bus_add_devices(vmd->bus);
> >
> > + vmd_acpi_end();
> > +
> > WARN(sysfs_create_link(&vmd->dev->dev.kobj, &vmd->bus->dev.kobj,
> > "domain"), "Can't create symlink to domain\n");
> > return 0;
> > Index: linux-pm/drivers/pci/host-bridge.c
> > ===================================================================
> > --- linux-pm.orig/drivers/pci/host-bridge.c
> > +++ linux-pm/drivers/pci/host-bridge.c
> > @@ -23,6 +23,7 @@ struct pci_host_bridge *pci_find_host_br
> >
> > return to_pci_host_bridge(root_bus->bridge);
> > }
> > +EXPORT_SYMBOL_GPL(pci_find_host_bridge);
> >
> > struct device *pci_get_host_bridge_device(struct pci_dev *dev)
> > {
> > Index: linux-pm/drivers/pci/pci-acpi.c
> > ===================================================================
> > --- linux-pm.orig/drivers/pci/pci-acpi.c
> > +++ linux-pm/drivers/pci/pci-acpi.c
> > @@ -1159,6 +1159,72 @@ void acpi_pci_remove_bus(struct pci_bus
> > }
> >
> > /* ACPI bus type */
> > +
> > +
> > +DEFINE_STATIC_KEY_FALSE(pci_acpi_companion_lookup_key);
>
> I don't know anything about static keys, but it looks like it
> complicates the code a fair amount, and the doc [2] suggests it's for
> performance-sensitive fast path code.
>
> Is this a path that needs that much optimization? Or is this static
> key dance something that's *always* needed when we change a function
> pointer at run-time?
Well, taking the mutex and checking if the hook is present every time
a PCI device is enumerated sounds to me like unnecessary overhead, but
functionally it can work without the static branch just fine.
Using an rwsem instead of the mutex may reduce the overhead, though.
> > +static DEFINE_MUTEX(pci_acpi_companion_lookup_mtx);
> > +static struct acpi_device *(*pci_acpi_find_companion_hook)(struct pci_dev *);
> > +
> > +/**
> > + * pci_acpi_set_companion_lookup_hook - Set ACPI companion lookup callback.
> > + * @func: ACPI companion lookup callback pointer or NULL.
> > + *
> > + * Set a special ACPI companion lookup callback for PCI devices whose companion
> > + * objects in the ACPI namespace have _ADR with non-standard bus-device-function
> > + * encodings.
> > + *
> > + * Return 0 on success or a negative error code on failure (in which case no
> > + * changes are made).
> > + *
> > + * The caller is responsible for the appropriate ordering of the invocations of
> > + * this function with respect to the enumeration of the PCI devices needing the
> > + * callback installed by it.
> > + */
> > +int pci_acpi_set_companion_lookup_hook(struct acpi_device *(*func)(struct pci_dev *))
> > +{
> > + int ret;
> > +
> > + if (!func)
> > + return -EINVAL;
> > +
> > + mutex_lock(&pci_acpi_companion_lookup_mtx);
> > +
> > + if (pci_acpi_find_companion_hook) {
> > + ret = -EBUSY;
> > + } else {
> > + pci_acpi_find_companion_hook = func;
> > + static_branch_enable(&pci_acpi_companion_lookup_key);
> > + ret = 0;
> > + }
> > +
> > + mutex_unlock(&pci_acpi_companion_lookup_mtx);
> > +
> > + return ret;
> > +}
> > +EXPORT_SYMBOL_GPL(pci_acpi_set_companion_lookup_hook);
> > +
> > +/**
> > + * pci_acpi_clear_companion_lookup_hook - Clear ACPI companion lookup callback.
> > + *
> > + * Clear the special ACPI companion lookup callback previously set by
> > + * pci_acpi_set_companion_lookup_hook(). Block until the last running instance
> > + * of the callback returns before clearing it.
> > + *
> > + * The caller is responsible for the appropriate ordering of the invocations of
> > + * this function with respect to the enumeration of the PCI devices needing the
> > + * callback cleared by it.
> > + */
> > +void pci_acpi_clear_companion_lookup_hook(void)
> > +{
> > + mutex_lock(&pci_acpi_companion_lookup_mtx);
> > +
> > + pci_acpi_find_companion_hook = NULL;
> > + static_branch_disable(&pci_acpi_companion_lookup_key);
> > +
> > + mutex_unlock(&pci_acpi_companion_lookup_mtx);
> > +}
> > +EXPORT_SYMBOL_GPL(pci_acpi_clear_companion_lookup_hook);
> > +
> > static struct acpi_device *acpi_pci_find_companion(struct device *dev)
> > {
> > struct pci_dev *pci_dev = to_pci_dev(dev);
> > @@ -1166,6 +1232,18 @@ static struct acpi_device *acpi_pci_find
> > bool check_children;
> > u64 addr;
> >
> > + if (static_branch_unlikely(&pci_acpi_companion_lookup_key)) {
> > + mutex_lock(&pci_acpi_companion_lookup_mtx);
> > +
> > + adev = pci_acpi_find_companion_hook ?
> > + pci_acpi_find_companion_hook(pci_dev) : NULL;
> > +
> > + mutex_unlock(&pci_acpi_companion_lookup_mtx);
> > +
> > + if (adev)
> > + return adev;
> > + }
> > +
> > check_children = pci_is_bridge(pci_dev);
> > /* Please ref to ACPI spec for the syntax of _ADR */
> > addr = (PCI_SLOT(pci_dev->devfn) << 16) | PCI_FUNC(pci_dev->devfn);
> > Index: linux-pm/include/linux/pci-acpi.h
> > ===================================================================
> > --- linux-pm.orig/include/linux/pci-acpi.h
> > +++ linux-pm/include/linux/pci-acpi.h
> > @@ -122,6 +122,9 @@ static inline void pci_acpi_add_edr_noti
> > static inline void pci_acpi_remove_edr_notifier(struct pci_dev *pdev) { }
> > #endif /* CONFIG_PCIE_EDR */
> >
> > +int pci_acpi_set_companion_lookup_hook(struct acpi_device *(*func)(struct pci_dev *));
> > +void pci_acpi_clear_companion_lookup_hook(void);
> > +
> > #else /* CONFIG_ACPI */
> > static inline void acpi_pci_add_bus(struct pci_bus *bus) { }
> > static inline void acpi_pci_remove_bus(struct pci_bus *bus) { }
>
> [1] https://lore.kernel.org/r/20210817180937.3123-8-ameynarkhede03@xxxxxxxxx
> [2] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/Documentation/staging/static-keys.rst?id=v5.13
