[RFC v5 0/8] PCI: hv: Support host bridge probing on ARM64

Hi Bjorn, Arnd and Marc,

This is the v5 for the preparation of virtual PCI support on Hyper-V
ARM64, Previous versions:

v1:	https://lore.kernel.org/lkml/20210319161956.2838291-1-boqun.feng@gmail.com/
v2:	https://lore.kernel.org/lkml/20210503144635.2297386-1-boqun.feng@gmail.com/
v3:	https://lore.kernel.org/lkml/20210609163211.3467449-1-boqun.feng@gmail.com/
v4:	https://lore.kernel.org/lkml/20210714102737.198432-1-boqun.feng@gmail.com/

Changes since last version:

*	Rebase to 5.14-rc2

*	Wording changes (Capitalization and more explanation on why) as
	suggested by Bjorn.

*	Split the patch #3 in previous into two patches as suggested by
	Bjorn.

The basic problem we need to resolve is that ARM64 is an arch with
PCI_DOMAINS_GENERIC=y, so the bus sysdata is pci_config_window. However,
Hyper-V PCI provides a paravirtualized PCI interface, so there is no
actual pci_config_window for a PCI host bridge, so no information can be
retrieve from the pci_config_window of a Hyper-V virtual PCI bus. Also
there is no corresponding ACPI device for the Hyper-V PCI root bridge,
which introduces a special case when trying to find the ACPI device from
the sysdata (see patch #3).

With this patchset, we could enable the virtual PCI on Hyper-V ARM64
guest with other code under development.

Comments and suggestions are welcome.

Regards,
Boqun

Arnd Bergmann (1):
  PCI: hv: Generify PCI probing

Boqun Feng (7):
  PCI: Introduce domain_nr in pci_host_bridge
  PCI: Support populating MSI domains of root buses via bridges
  arm64: PCI: Restructure pcibios_root_bridge_prepare()
  arm64: PCI: Support root bridge preparation for Hyper-V
  PCI: hv: Set ->domain_nr of pci_host_bridge at probing time
  PCI: hv: Set up MSI domain at bridge probing time
  PCI: hv: Turn on the host bridge probing on ARM64

 arch/arm64/kernel/pci.c             | 29 +++++++---
 drivers/pci/controller/pci-hyperv.c | 86 +++++++++++++++++------------
 drivers/pci/probe.c                 | 12 +++-
 include/linux/pci.h                 | 10 ++++
 4 files changed, 92 insertions(+), 45 deletions(-)

-- 
2.30.2

Since PCI_HYPERV depends on PCI_MSI_IRQ_DOMAIN which selects
GENERIC_MSI_IRQ_DOMAIN, we can use dev_set_msi_domain() to set up the
MSI domain at probing time, and this works for both x86 and ARM64.

Therefore use it as the preparation for ARM64 Hyper-V PCI support.

As a result, no longer need to maintain ->fwnode in x86 specific
pci_sysdata, and make hv_pcibus_device own it instead.

Signed-off-by: Boqun Feng <redacted>
---
 drivers/pci/controller/pci-hyperv.c | 13 ++++++++-----
 1 file changed, 8 insertions(+), 5 deletions(-)

diff --git a/drivers/pci/controller/pci-hyperv.c b/drivers/pci/controller/pci-hyperv.c
index 5741b1dd3c14..e6276aaa4659 100644
--- a/drivers/pci/controller/pci-hyperv.c
+++ b/drivers/pci/controller/pci-hyperv.c

@@ -450,6 +450,7 @@ enum hv_pcibus_state {
 struct hv_pcibus_device {
 	struct pci_sysdata sysdata;
 	struct pci_host_bridge *bridge;
+	struct fwnode_handle *fwnode;
 	/* Protocol version negotiated with the host */
 	enum pci_protocol_version_t protocol_version;
 	enum hv_pcibus_state state;

@@ -1565,7 +1566,7 @@ static int hv_pcie_init_irq_domain(struct hv_pcibus_device *hbus)
 	hbus->msi_info.handler = handle_edge_irq;
 	hbus->msi_info.handler_name = "edge";
 	hbus->msi_info.data = hbus;
-	hbus->irq_domain = pci_msi_create_irq_domain(hbus->sysdata.fwnode,
+	hbus->irq_domain = pci_msi_create_irq_domain(hbus->fwnode,
 						     &hbus->msi_info,
 						     x86_vector_domain);
 	if (!hbus->irq_domain) {

@@ -1574,6 +1575,8 @@ static int hv_pcie_init_irq_domain(struct hv_pcibus_device *hbus)
 		return -ENODEV;
 	}
 
+	dev_set_msi_domain(&hbus->bridge->dev, hbus->irq_domain);
+
 	return 0;
 }
 

@@ -3118,9 +3121,9 @@ static int hv_pci_probe(struct hv_device *hdev,
 		goto unmap;
 	}
 
-	hbus->sysdata.fwnode = irq_domain_alloc_named_fwnode(name);
+	hbus->fwnode = irq_domain_alloc_named_fwnode(name);
 	kfree(name);
-	if (!hbus->sysdata.fwnode) {
+	if (!hbus->fwnode) {
 		ret = -ENOMEM;
 		goto unmap;
 	}

@@ -3198,7 +3201,7 @@ static int hv_pci_probe(struct hv_device *hdev,
 free_irq_domain:
 	irq_domain_remove(hbus->irq_domain);
 free_fwnode:
-	irq_domain_free_fwnode(hbus->sysdata.fwnode);
+	irq_domain_free_fwnode(hbus->fwnode);
 unmap:
 	iounmap(hbus->cfg_addr);
 free_config:

@@ -3314,7 +3317,7 @@ static int hv_pci_remove(struct hv_device *hdev)
 	hv_free_config_window(hbus);
 	hv_pci_free_bridge_windows(hbus);
 	irq_domain_remove(hbus->irq_domain);
-	irq_domain_free_fwnode(hbus->sysdata.fwnode);
+	irq_domain_free_fwnode(hbus->fwnode);
 
 	hv_put_dom_num(hbus->bridge->domain_nr);
 

-- 
2.30.2

Restructure the pcibios_root_bridge_prepare() as the preparation for
supporting cases when no real ACPI device is related to the PCI host
bridge.

No functional change.

Signed-off-by: Boqun Feng <redacted>
---
 arch/arm64/kernel/pci.c | 19 ++++++++++++-------
 1 file changed, 12 insertions(+), 7 deletions(-)

diff --git a/arch/arm64/kernel/pci.c b/arch/arm64/kernel/pci.c
index 1006ed2d7c60..5148ae242780 100644
--- a/arch/arm64/kernel/pci.c
+++ b/arch/arm64/kernel/pci.c

@@ -82,14 +82,19 @@ int acpi_pci_bus_find_domain_nr(struct pci_bus *bus)
 
 int pcibios_root_bridge_prepare(struct pci_host_bridge *bridge)
 {
-	if (!acpi_disabled) {
-		struct pci_config_window *cfg = bridge->bus->sysdata;
-		struct acpi_device *adev = to_acpi_device(cfg->parent);
-		struct device *bus_dev = &bridge->bus->dev;
+	struct pci_config_window *cfg;
+	struct acpi_device *adev;
+	struct device *bus_dev;
 
-		ACPI_COMPANION_SET(&bridge->dev, adev);
-		set_dev_node(bus_dev, acpi_get_node(acpi_device_handle(adev)));
-	}
+	if (acpi_disabled)
+		return 0;
+
+	cfg = bridge->bus->sysdata;
+	adev = to_acpi_device(cfg->parent);
+	bus_dev = &bridge->bus->dev;
+
+	ACPI_COMPANION_SET(&bridge->dev, adev);
+	set_dev_node(bus_dev, acpi_get_node(acpi_device_handle(adev)));
 
 	return 0;
 }

-- 
2.30.2

Currently we retrieve the PCI domain number of the host bridge from the
bus sysdata (or pci_config_window if PCI_DOMAINS_GENERIC=y). Actually
we have the information at PCI host bridge probing time, and it makes
sense that we store it into pci_host_bridge. One benefit of doing so is
the requirement for supporting PCI on Hyper-V for ARM64, because the
host bridge of Hyper-V doesn't have pci_config_window, whereas ARM64 is
a PCI_DOMAINS_GENERIC=y arch, so we cannot retrieve the PCI domain
number from pci_config_window on ARM64 Hyper-V guest.

As the preparation for ARM64 Hyper-V PCI support, we introduce the
domain_nr in pci_host_bridge and a sentinel value to allow drivers to
set domain numbers properly at probing time. Currently
CONFIG_PCI_DOMAINS_GENERIC=y archs are only users of this
newly-introduced field.

Signed-off-by: Boqun Feng <redacted>
---
 drivers/pci/probe.c |  6 +++++-
 include/linux/pci.h | 10 ++++++++++
 2 files changed, 15 insertions(+), 1 deletion(-)

diff --git a/drivers/pci/probe.c b/drivers/pci/probe.c
index 79177ac37880..60c50d4f156f 100644
--- a/drivers/pci/probe.c
+++ b/drivers/pci/probe.c

@@ -594,6 +594,7 @@ static void pci_init_host_bridge(struct pci_host_bridge *bridge)
 	bridge->native_pme = 1;
 	bridge->native_ltr = 1;
 	bridge->native_dpc = 1;
+	bridge->domain_nr = PCI_DOMAIN_NR_NOT_SET;
 
 	device_initialize(&bridge->dev);
 }

@@ -898,7 +899,10 @@ static int pci_register_host_bridge(struct pci_host_bridge *bridge)
 	bus->ops = bridge->ops;
 	bus->number = bus->busn_res.start = bridge->busnr;
 #ifdef CONFIG_PCI_DOMAINS_GENERIC
-	bus->domain_nr = pci_bus_find_domain_nr(bus, parent);
+	if (bridge->domain_nr == PCI_DOMAIN_NR_NOT_SET)
+		bus->domain_nr = pci_bus_find_domain_nr(bus, parent);
+	else
+		bus->domain_nr = bridge->domain_nr;
 #endif
 
 	b = pci_find_bus(pci_domain_nr(bus), bridge->busnr);

diff --git a/include/linux/pci.h b/include/linux/pci.h
index 540b377ca8f6..2c413a64d168 100644
--- a/include/linux/pci.h
+++ b/include/linux/pci.h

@@ -526,6 +526,15 @@ static inline int pci_channel_offline(struct pci_dev *pdev)
 	return (pdev->error_state != pci_channel_io_normal);
 }
 
+/*
+* Currently in ACPI spec, for each PCI host bridge, PCI Segment Group number is
+* limited to a 16-bit value, therefore (int)-1 is not a valid PCI domain number,
+* and can be used as a sentinel value indicating >domain_nr is not set by the
+* driver (and CONFIG_PCI_DOMAINS_GENERIC=y archs will set it with
+* pci_bus_find_domain_nr()).
+*/
+#define PCI_DOMAIN_NR_NOT_SET (-1)
+
 struct pci_host_bridge {
 	struct device	dev;
 	struct pci_bus	*bus;		/* Root bus */

@@ -533,6 +542,7 @@ struct pci_host_bridge {
 	struct pci_ops	*child_ops;
 	void		*sysdata;
 	int		busnr;
+	int		domain_nr;
 	struct list_head windows;	/* resource_entry */
 	struct list_head dma_ranges;	/* dma ranges resource list */
 	u8 (*swizzle_irq)(struct pci_dev *, u8 *); /* Platform IRQ swizzler */

-- 
2.30.2

Currently at root bridge preparation, the corresponding ACPI device will
be set as the companion, however for a Hyper-V virtual PCI root bridge,
there is no corresponding ACPI device, because a Hyper-V virtual PCI
root bridge is discovered via VMBus rather than ACPI table. In order to
support this, we need to make pcibios_root_bridge_prepare() work with
cfg->parent being NULL.

Use a NULL pointer as the ACPI device if there is no corresponding ACPI
device, and this is fine because: 1) ACPI_COMPANION_SET() can work with
the second parameter being NULL, 2) semantically, if a NULL pointer is
set via ACPI_COMPANION_SET(), ACPI_COMPANION() (the read API for this
field) will return NULL, and since ACPI_COMPANION() may return NULL, so
users must have handled the cases where it returns NULL, and 3) since
there is no corresponding ACPI device, it would be wrong to use any
other value here.

Signed-off-by: Boqun Feng <redacted>
---
 arch/arm64/kernel/pci.c | 12 +++++++++++-
 1 file changed, 11 insertions(+), 1 deletion(-)

diff --git a/arch/arm64/kernel/pci.c b/arch/arm64/kernel/pci.c
index 5148ae242780..2276689b5411 100644
--- a/arch/arm64/kernel/pci.c
+++ b/arch/arm64/kernel/pci.c

@@ -90,7 +90,17 @@ int pcibios_root_bridge_prepare(struct pci_host_bridge *bridge)
 		return 0;
 
 	cfg = bridge->bus->sysdata;
-	adev = to_acpi_device(cfg->parent);
+
+	/*
+	 * On Hyper-V there is no corresponding ACPI device for a root bridge,
+	 * therefore ->parent is set as NULL by the driver. And set 'adev' as
+	 * NULL in this case because there is no proper ACPI device.
+	 */
+	if (!cfg->parent)
+		adev = NULL;
+	else
+		adev = to_acpi_device(cfg->parent);
+
 	bus_dev = &bridge->bus->dev;
 
 	ACPI_COMPANION_SET(&bridge->dev, adev);

-- 
2.30.2

Now we have everything we need, just provide a proper sysdata type for
the bus to use on ARM64 and everything else works.

Signed-off-by: Boqun Feng <redacted>
---
 drivers/pci/controller/pci-hyperv.c | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/drivers/pci/controller/pci-hyperv.c b/drivers/pci/controller/pci-hyperv.c
index e6276aaa4659..62dbe98d1fe1 100644
--- a/drivers/pci/controller/pci-hyperv.c
+++ b/drivers/pci/controller/pci-hyperv.c

@@ -40,6 +40,7 @@
 #include <linux/kernel.h>
 #include <linux/module.h>
 #include <linux/pci.h>
+#include <linux/pci-ecam.h>
 #include <linux/delay.h>
 #include <linux/semaphore.h>
 #include <linux/irqdomain.h>

@@ -448,7 +449,11 @@ enum hv_pcibus_state {
 };
 
 struct hv_pcibus_device {
+#ifdef CONFIG_X86
 	struct pci_sysdata sysdata;
+#elif defined(CONFIG_ARM64)
+	struct pci_config_window sysdata;
+#endif
 	struct pci_host_bridge *bridge;
 	struct fwnode_handle *fwnode;
 	/* Protocol version negotiated with the host */

@@ -3075,7 +3080,9 @@ static int hv_pci_probe(struct hv_device *hdev,
 			 dom_req, dom);
 
 	hbus->bridge->domain_nr = dom;
+#ifdef CONFIG_X86
 	hbus->sysdata.domain = dom;
+#endif
 
 	hbus->hdev = hdev;
 	INIT_LIST_HEAD(&hbus->children);

-- 
2.30.2

Currently, at probing time, the MSI domains of root buses are populated
if either the information of MSI domain is available from firmware (DT
or ACPI), or arch-specific sysdata is used to pass the fwnode of the MSI
domain. These two conditions don't cover all, e.g. Hyper-V virtual PCI
on ARM64, which doesn't have the MSI information in the firmware and
couldn't use arch-specific sysdata because running on an architecture
with PCI_DOMAINS_GENERIC=y.

To support populating MSI domains of the root buses at the probing when
neither of the above condition is true, the ->msi_domain of the
corresponding bridge device is used: in pci_host_bridge_msi_domain(),
which should return the MSI domain of the root bus, the ->msi_domain of
the corresponding bridge is fetched first as a potential value of the
MSI domain of the root bus.

In order to use the approach to populate MSI domains, the driver needs
to dev_set_msi_domain() on the bridge before calling
pci_register_host_bridge(), and makes sure GENERIC_MSI_IRQ_DOMAIN=y.

Another advantage of this new approach is providing an arch-independent
way to populate MSI domains, which allows sharing the driver code as
much as possible between architectures.

Originally-by: Arnd Bergmann [off-list ref]
Signed-off-by: Boqun Feng <redacted>
---
 drivers/pci/probe.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/drivers/pci/probe.c b/drivers/pci/probe.c
index 60c50d4f156f..ea7f2a57e2f5 100644
--- a/drivers/pci/probe.c
+++ b/drivers/pci/probe.c

@@ -829,11 +829,15 @@ static struct irq_domain *pci_host_bridge_msi_domain(struct pci_bus *bus)
 {
 	struct irq_domain *d;
 
+	/* If the host bridge driver sets a MSI domain of the bridge, use it */
+	d = dev_get_msi_domain(bus->bridge);
+
 	/*
 	 * Any firmware interface that can resolve the msi_domain
 	 * should be called from here.
 	 */
-	d = pci_host_bridge_of_msi_domain(bus);
+	if (!d)
+		d = pci_host_bridge_of_msi_domain(bus);
 	if (!d)
 		d = pci_host_bridge_acpi_msi_domain(bus);
 

-- 
2.30.2

From: Arnd Bergmann <arnd@arndb.de>

In order to support ARM64 Hyper-V PCI, we need to set up the bridge at
probing time because ARM64 is a PCI_DOMAIN_GENERIC=y arch and we don't
have pci_config_window (ARM64 sysdata) for a PCI root bus on Hyper-V, so
it's impossible to retrieve the information (e.g. PCI domains, MSI
domains) from bus sysdata on ARM64 after creation.

Originally in create_root_hv_pci_bus(), pci_create_root_bus() is used to
create the root bus and the corresponding bridge based on x86 sysdata.
Now we create a bridge first and then call pci_scan_root_bus_bridge(),
which allows us to do the necessary set-ups for the bridge.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Boqun Feng <redacted>
---
 drivers/pci/controller/pci-hyperv.c | 57 +++++++++++++++--------------
 1 file changed, 30 insertions(+), 27 deletions(-)

diff --git a/drivers/pci/controller/pci-hyperv.c b/drivers/pci/controller/pci-hyperv.c
index a53bd8728d0d..8d42da5dd1d4 100644
--- a/drivers/pci/controller/pci-hyperv.c
+++ b/drivers/pci/controller/pci-hyperv.c

@@ -449,6 +449,7 @@ enum hv_pcibus_state {
 
 struct hv_pcibus_device {
 	struct pci_sysdata sysdata;
+	struct pci_host_bridge *bridge;
 	/* Protocol version negotiated with the host */
 	enum pci_protocol_version_t protocol_version;
 	enum hv_pcibus_state state;

@@ -464,8 +465,6 @@ struct hv_pcibus_device {
 	spinlock_t device_list_lock;	/* Protect lists below */
 	void __iomem *cfg_addr;
 
-	struct list_head resources_for_children;
-
 	struct list_head children;
 	struct list_head dr_list;
 

@@ -1797,7 +1796,7 @@ static void hv_pci_assign_slots(struct hv_pcibus_device *hbus)
 
 		slot_nr = PCI_SLOT(wslot_to_devfn(hpdev->desc.win_slot.slot));
 		snprintf(name, SLOT_NAME_SIZE, "%u", hpdev->desc.ser);
-		hpdev->pci_slot = pci_create_slot(hbus->pci_bus, slot_nr,
+		hpdev->pci_slot = pci_create_slot(hbus->bridge->bus, slot_nr,
 					  name, NULL);
 		if (IS_ERR(hpdev->pci_slot)) {
 			pr_warn("pci_create slot %s failed\n", name);

@@ -1827,7 +1826,7 @@ static void hv_pci_remove_slots(struct hv_pcibus_device *hbus)
 static void hv_pci_assign_numa_node(struct hv_pcibus_device *hbus)
 {
 	struct pci_dev *dev;
-	struct pci_bus *bus = hbus->pci_bus;
+	struct pci_bus *bus = hbus->bridge->bus;
 	struct hv_pci_dev *hv_dev;
 
 	list_for_each_entry(dev, &bus->devices, bus_list) {

@@ -1850,21 +1849,22 @@ static void hv_pci_assign_numa_node(struct hv_pcibus_device *hbus)
  */
 static int create_root_hv_pci_bus(struct hv_pcibus_device *hbus)
 {
-	/* Register the device */
-	hbus->pci_bus = pci_create_root_bus(&hbus->hdev->device,
-					    0, /* bus number is always zero */
-					    &hv_pcifront_ops,
-					    &hbus->sysdata,
-					    &hbus->resources_for_children);
-	if (!hbus->pci_bus)
-		return -ENODEV;
+	int error;
+	struct pci_host_bridge *bridge = hbus->bridge;
+
+	bridge->dev.parent = &hbus->hdev->device;
+	bridge->sysdata = &hbus->sysdata;
+	bridge->ops = &hv_pcifront_ops;
+
+	error = pci_scan_root_bus_bridge(bridge);
+	if (error)
+		return error;
 
 	pci_lock_rescan_remove();
-	pci_scan_child_bus(hbus->pci_bus);
 	hv_pci_assign_numa_node(hbus);
-	pci_bus_assign_resources(hbus->pci_bus);
+	pci_bus_assign_resources(bridge->bus);
 	hv_pci_assign_slots(hbus);
-	pci_bus_add_devices(hbus->pci_bus);
+	pci_bus_add_devices(bridge->bus);
 	pci_unlock_rescan_remove();
 	hbus->state = hv_pcibus_installed;
 	return 0;

@@ -2127,7 +2127,7 @@ static void pci_devices_present_work(struct work_struct *work)
 		 * because there may have been changes.
 		 */
 		pci_lock_rescan_remove();
-		pci_scan_child_bus(hbus->pci_bus);
+		pci_scan_child_bus(hbus->bridge->bus);
 		hv_pci_assign_numa_node(hbus);
 		hv_pci_assign_slots(hbus);
 		pci_unlock_rescan_remove();

@@ -2295,8 +2295,8 @@ static void hv_eject_device_work(struct work_struct *work)
 	/*
 	 * Ejection can come before or after the PCI bus has been set up, so
 	 * attempt to find it and tear down the bus state, if it exists.  This
-	 * must be done without constructs like pci_domain_nr(hbus->pci_bus)
-	 * because hbus->pci_bus may not exist yet.
+	 * must be done without constructs like pci_domain_nr(hbus->bridge->bus)
+	 * because hbus->bridge->bus may not exist yet.
 	 */
 	wslot = wslot_to_devfn(hpdev->desc.win_slot.slot);
 	pdev = pci_get_domain_bus_and_slot(hbus->sysdata.domain, 0, wslot);

@@ -2662,8 +2662,7 @@ static int hv_pci_allocate_bridge_windows(struct hv_pcibus_device *hbus)
 		/* Modify this resource to become a bridge window. */
 		hbus->low_mmio_res->flags |= IORESOURCE_WINDOW;
 		hbus->low_mmio_res->flags &= ~IORESOURCE_BUSY;
-		pci_add_resource(&hbus->resources_for_children,
-				 hbus->low_mmio_res);
+		pci_add_resource(&hbus->bridge->windows, hbus->low_mmio_res);
 	}
 
 	if (hbus->high_mmio_space) {

@@ -2682,8 +2681,7 @@ static int hv_pci_allocate_bridge_windows(struct hv_pcibus_device *hbus)
 		/* Modify this resource to become a bridge window. */
 		hbus->high_mmio_res->flags |= IORESOURCE_WINDOW;
 		hbus->high_mmio_res->flags &= ~IORESOURCE_BUSY;
-		pci_add_resource(&hbus->resources_for_children,
-				 hbus->high_mmio_res);
+		pci_add_resource(&hbus->bridge->windows, hbus->high_mmio_res);
 	}
 
 	return 0;

@@ -3002,6 +3000,7 @@ static void hv_put_dom_num(u16 dom)
 static int hv_pci_probe(struct hv_device *hdev,
 			const struct hv_vmbus_device_id *dev_id)
 {
+	struct pci_host_bridge *bridge;
 	struct hv_pcibus_device *hbus;
 	u16 dom_req, dom;
 	char *name;

@@ -3014,6 +3013,10 @@ static int hv_pci_probe(struct hv_device *hdev,
 	 */
 	BUILD_BUG_ON(sizeof(*hbus) > HV_HYP_PAGE_SIZE);
 
+	bridge = devm_pci_alloc_host_bridge(&hdev->device, 0);
+	if (!bridge)
+		return -ENOMEM;
+
 	/*
 	 * With the recent 59bb47985c1d ("mm, sl[aou]b: guarantee natural
 	 * alignment for kmalloc(power-of-two)"), kzalloc() is able to allocate

@@ -3035,6 +3038,8 @@ static int hv_pci_probe(struct hv_device *hdev,
 	hbus = kzalloc(HV_HYP_PAGE_SIZE, GFP_KERNEL);
 	if (!hbus)
 		return -ENOMEM;
+
+	hbus->bridge = bridge;
 	hbus->state = hv_pcibus_init;
 	hbus->wslot_res_allocated = -1;
 

@@ -3071,7 +3076,6 @@ static int hv_pci_probe(struct hv_device *hdev,
 	hbus->hdev = hdev;
 	INIT_LIST_HEAD(&hbus->children);
 	INIT_LIST_HEAD(&hbus->dr_list);
-	INIT_LIST_HEAD(&hbus->resources_for_children);
 	spin_lock_init(&hbus->config_lock);
 	spin_lock_init(&hbus->device_list_lock);
 	spin_lock_init(&hbus->retarget_msi_interrupt_lock);

@@ -3295,9 +3299,9 @@ static int hv_pci_remove(struct hv_device *hdev)
 
 		/* Remove the bus from PCI's point of view. */
 		pci_lock_rescan_remove();
-		pci_stop_root_bus(hbus->pci_bus);
+		pci_stop_root_bus(hbus->bridge->bus);
 		hv_pci_remove_slots(hbus);
-		pci_remove_root_bus(hbus->pci_bus);
+		pci_remove_root_bus(hbus->bridge->bus);
 		pci_unlock_rescan_remove();
 	}
 

@@ -3307,7 +3311,6 @@ static int hv_pci_remove(struct hv_device *hdev)
 
 	iounmap(hbus->cfg_addr);
 	hv_free_config_window(hbus);
-	pci_free_resource_list(&hbus->resources_for_children);
 	hv_pci_free_bridge_windows(hbus);
 	irq_domain_remove(hbus->irq_domain);
 	irq_domain_free_fwnode(hbus->sysdata.fwnode);

@@ -3390,7 +3393,7 @@ static int hv_pci_restore_msi_msg(struct pci_dev *pdev, void *arg)
  */
 static void hv_pci_restore_msi_state(struct hv_pcibus_device *hbus)
 {
-	pci_walk_bus(hbus->pci_bus, hv_pci_restore_msi_msg, NULL);
+	pci_walk_bus(hbus->bridge->bus, hv_pci_restore_msi_msg, NULL);
 }
 
 static int hv_pci_resume(struct hv_device *hdev)

-- 
2.30.2

No functional change, just store and maintain the PCI domain number in
the ->domain_nr of pci_host_bridge. Note that we still need to keep
the copy of domain number in x86-specific pci_sysdata, because x86 is
not a PCI_DOMAINS_GENERIC=y architecture, so the ->domain_nr of
pci_host_bridge doesn't work for it yet.

Signed-off-by: Boqun Feng <redacted>
---
 drivers/pci/controller/pci-hyperv.c | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)

diff --git a/drivers/pci/controller/pci-hyperv.c b/drivers/pci/controller/pci-hyperv.c
index 8d42da5dd1d4..5741b1dd3c14 100644
--- a/drivers/pci/controller/pci-hyperv.c
+++ b/drivers/pci/controller/pci-hyperv.c

@@ -2299,7 +2299,7 @@ static void hv_eject_device_work(struct work_struct *work)
 	 * because hbus->bridge->bus may not exist yet.
 	 */
 	wslot = wslot_to_devfn(hpdev->desc.win_slot.slot);
-	pdev = pci_get_domain_bus_and_slot(hbus->sysdata.domain, 0, wslot);
+	pdev = pci_get_domain_bus_and_slot(hbus->bridge->domain_nr, 0, wslot);
 	if (pdev) {
 		pci_lock_rescan_remove();
 		pci_stop_and_remove_bus_device(pdev);

@@ -3071,6 +3071,7 @@ static int hv_pci_probe(struct hv_device *hdev,
 			 "PCI dom# 0x%hx has collision, using 0x%hx",
 			 dom_req, dom);
 
+	hbus->bridge->domain_nr = dom;
 	hbus->sysdata.domain = dom;
 
 	hbus->hdev = hdev;

@@ -3080,7 +3081,7 @@ static int hv_pci_probe(struct hv_device *hdev,
 	spin_lock_init(&hbus->device_list_lock);
 	spin_lock_init(&hbus->retarget_msi_interrupt_lock);
 	hbus->wq = alloc_ordered_workqueue("hv_pci_%x", 0,
-					   hbus->sysdata.domain);
+					   hbus->bridge->domain_nr);
 	if (!hbus->wq) {
 		ret = -ENOMEM;
 		goto free_dom;

@@ -3207,7 +3208,7 @@ static int hv_pci_probe(struct hv_device *hdev,
 destroy_wq:
 	destroy_workqueue(hbus->wq);
 free_dom:
-	hv_put_dom_num(hbus->sysdata.domain);
+	hv_put_dom_num(hbus->bridge->domain_nr);
 free_bus:
 	kfree(hbus);
 	return ret;

@@ -3315,7 +3316,7 @@ static int hv_pci_remove(struct hv_device *hdev)
 	irq_domain_remove(hbus->irq_domain);
 	irq_domain_free_fwnode(hbus->sysdata.fwnode);
 
-	hv_put_dom_num(hbus->sysdata.domain);
+	hv_put_dom_num(hbus->bridge->domain_nr);
 
 	kfree(hbus);
 	return ret;

-- 
2.30.2

On Tue, 20 Jul 2021 14:44:29 +0100,
Boqun Feng [off-list ref] wrote:

Am I the only one who find this rather odd? Nothing ever populates
this data structure on arm64, and its only purpose seems to serve as
an anchor to retrieve the hbus via container_of().

If that's indeed the case, I'd rather see an arch-specific to_hbus()
helper that uses another (preexisting) field as the anchor for arm64.

Thanks,

	M.

-- 
Without deviation from the norm, progress is not possible.

On Tue, Jul 20, 2021 at 03:38:26PM +0100, Marc Zyngier wrote:

On Tue, 20 Jul 2021 14:44:29 +0100,
Boqun Feng [off-list ref] wrote:

quoted

Now we have everything we need, just provide a proper sysdata type for
the bus to use on ARM64 and everything else works.

Signed-off-by: Boqun Feng <redacted>
---
 drivers/pci/controller/pci-hyperv.c | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/drivers/pci/controller/pci-hyperv.c b/drivers/pci/controller/pci-hyperv.c
index e6276aaa4659..62dbe98d1fe1 100644
--- a/drivers/pci/controller/pci-hyperv.c
+++ b/drivers/pci/controller/pci-hyperv.c

@@ -40,6 +40,7 @@
 #include <linux/kernel.h>
 #include <linux/module.h>
 #include <linux/pci.h>
+#include <linux/pci-ecam.h>
 #include <linux/delay.h>
 #include <linux/semaphore.h>
 #include <linux/irqdomain.h>

@@ -448,7 +449,11 @@ enum hv_pcibus_state {
 };
 
 struct hv_pcibus_device {
+#ifdef CONFIG_X86
 	struct pci_sysdata sysdata;
+#elif defined(CONFIG_ARM64)
+	struct pci_config_window sysdata;
+#endif

Am I the only one who find this rather odd? Nothing ever populates
this data structure on arm64, and its only purpose seems to serve as
an anchor to retrieve the hbus via container_of().

This field will also be used as the ->sysdata of pci_bus and
pci_host_bridge, and some of the PCI core code touches. Although I made
this field as all zeroed and make sure PCI core can handle (patch #4).

If that's indeed the case, I'd rather see an arch-specific to_hbus()
helper that uses another (preexisting) field as the anchor for arm64.

I did a quick look, but I didn't find another field works: the field
needs to be placed inside hv_pcibus_device and the address can be
retrieved via pci_bus. I'm open to any suggestion in case that I missed
something.

Regards,
Boqun

Thanks,

	M.

-- 
Without deviation from the norm, progress is not possible.

On Tue, 20 Jul 2021 15:59:41 +0100,
Boqun Feng [off-list ref] wrote:

On Tue, Jul 20, 2021 at 03:38:26PM +0100, Marc Zyngier wrote:

quoted

On Tue, 20 Jul 2021 14:44:29 +0100,
Boqun Feng [off-list ref] wrote:

quoted

Now we have everything we need, just provide a proper sysdata type for
the bus to use on ARM64 and everything else works.

Signed-off-by: Boqun Feng <redacted>
---
 drivers/pci/controller/pci-hyperv.c | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/drivers/pci/controller/pci-hyperv.c b/drivers/pci/controller/pci-hyperv.c
index e6276aaa4659..62dbe98d1fe1 100644
--- a/drivers/pci/controller/pci-hyperv.c
+++ b/drivers/pci/controller/pci-hyperv.c

@@ -40,6 +40,7 @@
 #include <linux/kernel.h>
 #include <linux/module.h>
 #include <linux/pci.h>
+#include <linux/pci-ecam.h>
 #include <linux/delay.h>
 #include <linux/semaphore.h>
 #include <linux/irqdomain.h>

@@ -448,7 +449,11 @@ enum hv_pcibus_state {
 };
 
 struct hv_pcibus_device {
+#ifdef CONFIG_X86
 	struct pci_sysdata sysdata;
+#elif defined(CONFIG_ARM64)
+	struct pci_config_window sysdata;
+#endif

Am I the only one who find this rather odd? Nothing ever populates
this data structure on arm64, and its only purpose seems to serve as
an anchor to retrieve the hbus via container_of().

This field will also be used as the ->sysdata of pci_bus and
pci_host_bridge, and some of the PCI core code touches. Although I made
this field as all zeroed and make sure PCI core can handle (patch #4).

Huh, I see. I missed this particular nugget. This is so convoluted...

quoted

If that's indeed the case, I'd rather see an arch-specific to_hbus()
helper that uses another (preexisting) field as the anchor for arm64.

I did a quick look, but I didn't find another field works: the field
needs to be placed inside hv_pcibus_device and the address can be
retrieved via pci_bus. I'm open to any suggestion in case that I missed
something.

No, the above pretty much kills my suggestion.

Thanks for the explanation,

	M.

-- 
Without deviation from the norm, progress is not possible.

On Tue, Jul 20, 2021 at 09:44:22PM +0800, Boqun Feng wrote:

Currently we retrieve the PCI domain number of the host bridge from the
bus sysdata (or pci_config_window if PCI_DOMAINS_GENERIC=y). Actually
we have the information at PCI host bridge probing time, and it makes
sense that we store it into pci_host_bridge. One benefit of doing so is
the requirement for supporting PCI on Hyper-V for ARM64, because the
host bridge of Hyper-V doesn't have pci_config_window, whereas ARM64 is
a PCI_DOMAINS_GENERIC=y arch, so we cannot retrieve the PCI domain
number from pci_config_window on ARM64 Hyper-V guest.

As the preparation for ARM64 Hyper-V PCI support, we introduce the
domain_nr in pci_host_bridge and a sentinel value to allow drivers to
set domain numbers properly at probing time. Currently
CONFIG_PCI_DOMAINS_GENERIC=y archs are only users of this
newly-introduced field.

Signed-off-by: Boqun Feng <redacted>

Once all the issues are ironed out, Lorenzo should probably merge this
since it's primarily Hyper-V stuff, but I'm OK with this part:

Acked-by: Bjorn Helgaas <bhelgaas@google.com>

But fix the comment issue below.

Fix comment indentation (follow style of other comments in this file).

s/>domain_nr/->domain_nr/

On Tue, Jul 20, 2021 at 05:49:25PM -0500, Bjorn Helgaas wrote:

On Tue, Jul 20, 2021 at 09:44:22PM +0800, Boqun Feng wrote:

quoted

Currently we retrieve the PCI domain number of the host bridge from the
bus sysdata (or pci_config_window if PCI_DOMAINS_GENERIC=y). Actually
we have the information at PCI host bridge probing time, and it makes
sense that we store it into pci_host_bridge. One benefit of doing so is
the requirement for supporting PCI on Hyper-V for ARM64, because the
host bridge of Hyper-V doesn't have pci_config_window, whereas ARM64 is
a PCI_DOMAINS_GENERIC=y arch, so we cannot retrieve the PCI domain
number from pci_config_window on ARM64 Hyper-V guest.

As the preparation for ARM64 Hyper-V PCI support, we introduce the
domain_nr in pci_host_bridge and a sentinel value to allow drivers to
set domain numbers properly at probing time. Currently
CONFIG_PCI_DOMAINS_GENERIC=y archs are only users of this
newly-introduced field.

Signed-off-by: Boqun Feng <redacted>

Once all the issues are ironed out, Lorenzo should probably merge this
since it's primarily Hyper-V stuff, but I'm OK with this part:

Acked-by: Bjorn Helgaas <bhelgaas@google.com>

Thanks!

But fix the comment issue below.

Just send a v5.1 for this patch with the comment fixed and your
Acked-by.

Regards,
Boqun

quoted

---
 drivers/pci/probe.c |  6 +++++-
 include/linux/pci.h | 10 ++++++++++
 2 files changed, 15 insertions(+), 1 deletion(-)

diff --git a/drivers/pci/probe.c b/drivers/pci/probe.c
index 79177ac37880..60c50d4f156f 100644
--- a/drivers/pci/probe.c
+++ b/drivers/pci/probe.c

@@ -594,6 +594,7 @@ static void pci_init_host_bridge(struct pci_host_bridge *bridge)
 	bridge->native_pme = 1;
 	bridge->native_ltr = 1;
 	bridge->native_dpc = 1;
+	bridge->domain_nr = PCI_DOMAIN_NR_NOT_SET;
 
 	device_initialize(&bridge->dev);
 }

@@ -898,7 +899,10 @@ static int pci_register_host_bridge(struct pci_host_bridge *bridge)
 	bus->ops = bridge->ops;
 	bus->number = bus->busn_res.start = bridge->busnr;
 #ifdef CONFIG_PCI_DOMAINS_GENERIC
-	bus->domain_nr = pci_bus_find_domain_nr(bus, parent);
+	if (bridge->domain_nr == PCI_DOMAIN_NR_NOT_SET)
+		bus->domain_nr = pci_bus_find_domain_nr(bus, parent);
+	else
+		bus->domain_nr = bridge->domain_nr;
 #endif
 
 	b = pci_find_bus(pci_domain_nr(bus), bridge->busnr);

diff --git a/include/linux/pci.h b/include/linux/pci.h
index 540b377ca8f6..2c413a64d168 100644
--- a/include/linux/pci.h
+++ b/include/linux/pci.h

@@ -526,6 +526,15 @@ static inline int pci_channel_offline(struct pci_dev *pdev)
 	return (pdev->error_state != pci_channel_io_normal);
 }
 
+/*
+* Currently in ACPI spec, for each PCI host bridge, PCI Segment Group number is
+* limited to a 16-bit value, therefore (int)-1 is not a valid PCI domain number,
+* and can be used as a sentinel value indicating >domain_nr is not set by the
+* driver (and CONFIG_PCI_DOMAINS_GENERIC=y archs will set it with
+* pci_bus_find_domain_nr()).
+*/

Fix comment indentation (follow style of other comments in this file).

s/>domain_nr/->domain_nr/

quoted

+#define PCI_DOMAIN_NR_NOT_SET (-1)
+
 struct pci_host_bridge {
 	struct device	dev;
 	struct pci_bus	*bus;		/* Root bus */

@@ -533,6 +542,7 @@ struct pci_host_bridge {
 	struct pci_ops	*child_ops;
 	void		*sysdata;
 	int		busnr;
+	int		domain_nr;
 	struct list_head windows;	/* resource_entry */
 	struct list_head dma_ranges;	/* dma ranges resource list */
 	u8 (*swizzle_irq)(struct pci_dev *, u8 *); /* Platform IRQ swizzler */

-- 
2.30.2