Skip to contentSkip to conversation

Add support to preserve boot config in the DT flow

PATCH In review Vidya Sagar 4 participants 29 messages Jan 10, 2024 – Apr 22, 2024

Subsystems (6)
lore
Share & tools
lore
#0 lore raw

v2 cover Add support to preserve boot config in the DT flow

Add support to preserve the boot configuration of the PCIe bridges per host bridge basis based on the presence of the DT flag "preserve-boot-config" in the respective host bridge node. The existing "linux,pci-probe-only" works at a system level and can't be used at a single host bridge granularity. Also, the support for preserving the boot configuration per host bridge basis is already present for the ACPI based boot flow and this patch series extends that support for the DT based boot flow.

V2:
* Addressed issues reported by kernel test robot <[email protected]>

Vidya Sagar (2):
dt-bindings: Add PCIe "preserve-boot-config" property
PCI: Add support for "preserve-boot-config" property

 Documentation/devicetree/bindings/pci/pci.txt |  4 ++++
 drivers/pci/controller/pci-host-common.c      |  5 ++++-
 drivers/pci/of.c                              | 18 ++++++++++++++++++
 drivers/pci/probe.c                           |  2 +-
 include/linux/of_pci.h                        |  6 ++++++
 5 files changed, 33 insertions(+), 2 deletions(-)
Signature · 1 line
2.25.1
#1 lore raw

PATCH v2 1/2 dt-bindings: Add PCIe "preserve-boot-config" property

The existing "linux,pci-probe-only" property applies at a system level and it is not possible to selectively convey individual host bridge's requirement w.r.t preserving the boot configuration done by the platform firmware to the kernel. "preserve-boot-config" addresses that concern and can be used to preserve the boot configuration for host bridges selectivey.

---
V2:
* None

 Documentation/devicetree/bindings/pci/pci.txt | 4 ++++
 1 file changed, 4 insertions(+)
Documentation/devicetree/bindings/pci/pci.txt +4 -0
index 6a8f2874a24d..0a5ff998cbe8 100644
@@ -32,6 +32,10 @@ driver implementation may support the following properties:
    root port to downstream device and host bridge drivers can do programming
    which depends on CLKREQ signal existence. For example, programming root port
    not to advertise ASPM L1 Sub-States support if there is no CLKREQ signal.
+- preserve-boot-config:
+   If present this property specifies that this host bridge is already
+   configured by the platform firmware and the OS doesn't need to reconfigure
+   it again.
 
 PCI-PCI Bridge properties
 -------------------------
Signature · 1 line
2.25.1
maintainer
#2 lore raw

Re: [PATCH V2 1/2] dt-bindings: Add PCIe "preserve-boot-config" property

Quoted 20 lines · Vidya Sagar

The existing "linux,pci-probe-only" property applies at a system level and it is not possible to selectively convey individual host bridge's requirement w.r.t preserving the boot configuration done by the platform firmware to the kernel. "preserve-boot-config" addresses that concern and can be used to preserve the boot configuration for host bridges selectivey.

---
V2:
* None

 Documentation/devicetree/bindings/pci/pci.txt | 4 ++++
 1 file changed, 4 insertions(+)
Documentation/devicetree/bindings/pci/pci.txt +0 -0
index 6a8f2874a24d..0a5ff998cbe8 100644

Documentation/devicetree/bindings/pci/pci.txt driver implementation may support the following properties: · line 38

Quoted 7 lines
Documentation/devicetree/bindings/pci/pci.txt +4 -0
@@ -32,6 +32,10 @@ driver implementation may support the following properties:
    root port to downstream device and host bridge drivers can do programming
    which depends on CLKREQ signal existence. For example, programming root port
    not to advertise ASPM L1 Sub-States support if there is no CLKREQ signal.
+- preserve-boot-config:
+   If present this property specifies that this host bridge is already
+   configured by the platform firmware and the OS doesn't need to reconfigure
+   it again.

Anything new must be a schema.

Don't create something new to workaround the problem. Extend the existing support. For this, make "linux,pci-probe-only" allowed in host bridge nodes. Yeah, there's the issue that this property sets a global flag, but I'd imagine the PCI maintainers would agree that some flags should/could be per bus.

Rob

#3 lore raw

PATCH v2 2/2 PCI: Add support for "preserve-boot-config" property

In reply to Vidya Sagar

Add support for "preserve-boot-config" property that can be used to selectively (i.e. per host bridge) instruct the kernel to preserve the boot time configuration done by the platform firmware.

---
V2:
* Addressed issues reported by kernel test robot <[email protected]>

 drivers/pci/controller/pci-host-common.c |  5 ++++-
 drivers/pci/of.c                         | 18 ++++++++++++++++++
 drivers/pci/probe.c                      |  2 +-
 include/linux/of_pci.h                   |  6 ++++++
 4 files changed, 29 insertions(+), 2 deletions(-)
drivers/pci/controller/pci-host-common.c +4 -1
index 6be3266cd7b5..d3475dc9ec44 100644
@@ -68,13 +68,16 @@ int pci_host_common_probe(struct platform_device *pdev)
 
 	of_pci_check_probe_only();
 
+	bridge->preserve_config =
+		of_pci_check_preserve_boot_config(dev->of_node);
+
 	/* Parse and map our Configuration Space windows */
 	cfg = gen_pci_init(dev, bridge, ops);
 	if (IS_ERR(cfg))
 		return PTR_ERR(cfg);
 
 	/* Do not reassign resources if probe only */
-	if (!pci_has_flag(PCI_PROBE_ONLY))
+	if (!(pci_has_flag(PCI_PROBE_ONLY) || bridge->preserve_config))
 		pci_add_flags(PCI_REASSIGN_ALL_BUS);
 
 	bridge->sysdata = cfg;
drivers/pci/of.c +18 -0
index 51e3dd0ea5ab..ed3c0dd9804e 100644
@@ -258,6 +258,24 @@ void of_pci_check_probe_only(void)
 }
 EXPORT_SYMBOL_GPL(of_pci_check_probe_only);
 
+/**
+ * of_pci_check_preserve_boot_config - Return true if the boot configuration
+ *                                     needs to be preserved
+ * @node: Device tree node with the domain information.
+ *
+ * This function looks for a property called "preserve-boot-config" for a given
+ * PCIe controller's node and returns true if found. Having this property
+ * for a PCIe controller ensures that the kernel doesn't re-enumerate and
+ * reconfigure the BAR resources that are already done by the platform firmware.
+ *
+ * Return: true if the property exists false otherwise.
+ */
+bool of_pci_check_preserve_boot_config(struct device_node *node)
+{
+	return of_property_read_bool(node, "preserve-boot-config");
+}
+EXPORT_SYMBOL_GPL(of_pci_check_preserve_boot_config);
+
 /**
  * devm_of_pci_get_host_bridge_resources() - Resource-managed parsing of PCI
  *                                           host bridge resources from DT
drivers/pci/probe.c +1 -1
index 795534589b98..79d0ac34f567 100644
@@ -3085,7 +3085,7 @@ int pci_host_probe(struct pci_host_bridge *bridge)
 	 * ioport_resource trees in either pci_bus_claim_resources()
 	 * or pci_bus_assign_resources().
 	 */
-	if (pci_has_flag(PCI_PROBE_ONLY)) {
+	if (pci_has_flag(PCI_PROBE_ONLY) || bridge->preserve_config) {
 		pci_bus_claim_resources(bus);
 	} else {
 		pci_bus_size_bridges(bus);
include/linux/of_pci.h +6 -0
index 29658c0ee71f..ba5532005125 100644
@@ -13,6 +13,7 @@ struct device_node *of_pci_find_child_device(struct device_node *parent,
 					     unsigned int devfn);
 int of_pci_get_devfn(struct device_node *np);
 void of_pci_check_probe_only(void);
+bool of_pci_check_preserve_boot_config(struct device_node *node);
 #else
 static inline struct device_node *of_pci_find_child_device(struct device_node *parent,
 					     unsigned int devfn)
@@ -26,6 +27,11 @@ static inline int of_pci_get_devfn(struct device_node *np)
 }
 
 static inline void of_pci_check_probe_only(void) { }
+
+static inline bool of_pci_check_preserve_boot_config(struct device_node *node)
+{
+	return false;
+}
 #endif
 
 #if IS_ENABLED(CONFIG_OF_IRQ)
Signature · 1 line
2.25.1
#4 lore raw

Re: [PATCH V2 2/2] PCI: Add support for "preserve-boot-config" property

Quoted 21 lines · Vidya Sagar

Add support for "preserve-boot-config" property that can be used to selectively (i.e. per host bridge) instruct the kernel to preserve the boot time configuration done by the platform firmware.

---
V2:
* Addressed issues reported by kernel test robot <[email protected]>

 drivers/pci/controller/pci-host-common.c |  5 ++++-
 drivers/pci/of.c                         | 18 ++++++++++++++++++
 drivers/pci/probe.c                      |  2 +-
 include/linux/of_pci.h                   |  6 ++++++
 4 files changed, 29 insertions(+), 2 deletions(-)
drivers/pci/controller/pci-host-common.c +0 -0
index 6be3266cd7b5..d3475dc9ec44 100644

drivers/pci/controller/pci-host-common.c int pci_host_common_probe(struct platform_device *pdev) · line 72

Quoted 5 lines
drivers/pci/controller/pci-host-common.c +2 -0
@@ -68,13 +68,16 @@ int pci_host_common_probe(struct platform_device *pdev)
 
 	of_pci_check_probe_only();
 
+	bridge->preserve_config =
+		of_pci_check_preserve_boot_config(dev->of_node);

Thanks for leveraging the existing "preserve_config" support for the ACPI _DSM. Is pci_host_common_probe() the best place for this? I think there are many DT platform drivers that do not use pci_host_common_probe(), so I wonder if there's a more generic place to put this.

I see Rob's concern about adding "preserve-boot-config" vs extending "linux,pci-probe-only" and I don't really have an opinion on that, although I do think the "pci-probe-only" name is not as descriptive as it could be. I guess somebody will argue that "preserve_config" could be more descriptive, too :)

Bjorn

#5 lore raw

Re: [PATCH V2 2/2] PCI: Add support for "preserve-boot-config" property

Quoted 28 lines · Bjorn Helgaas

External email: Use caution opening links or attachments

Quoted 24 lines · Vidya Sagar

Add support for "preserve-boot-config" property that can be used to selectively (i.e. per host bridge) instruct the kernel to preserve the boot time configuration done by the platform firmware.

---
V2:
* Addressed issues reported by kernel test robot <[email protected]>

  drivers/pci/controller/pci-host-common.c |  5 ++++-
  drivers/pci/of.c                         | 18 ++++++++++++++++++
  drivers/pci/probe.c                      |  2 +-
  include/linux/of_pci.h                   |  6 ++++++
  4 files changed, 29 insertions(+), 2 deletions(-)
drivers/pci/controller/pci-host-common.c +0 -0
index 6be3266cd7b5..d3475dc9ec44 100644
@@ -68,13 +68,16 @@ int pci_host_common_probe(struct platform_device *pdev)
 
       of_pci_check_probe_only();
 
Quoted 8 lines

drivers/pci/controller/pci-host-common.c int pci_host_common_probe(struct platform_device *pdev) · line 72

Quoted 2 lines
drivers/pci/controller/pci-host-common.c +2 -0
+     bridge->preserve_config =
+             of_pci_check_preserve_boot_config(dev->of_node);

Thanks for leveraging the existing "preserve_config" support for the ACPI _DSM. Is pci_host_common_probe() the best place for this? I think there are many DT platform drivers that do not use pci_host_common_probe(), so I wonder if there's a more generic place to put this.

My understanding is that pci_host_common_probe() is mainly used in systems where the firmware would have taken care of all the platform specific initialization and giving the ECAM and 'ranges' info through DT for the Linux kernel to go ahead and perform the enumeration. This is similar to ACPI way of handing over the system from firmware to the OS.

If PCIe controllers are getting initialized in the kernel itself, then pci_host_probe() is called directly from the respective host controller drivers which is the case with all the DesignWare based implementations including Tegra194 and Tegra234. In those systems, since the controllers themselves have come up and gotten initialized in the kernel, resource assignment has to happen anyway.

Quoted 6 lines

I see Rob's concern about adding "preserve-boot-config" vs extending "linux,pci-probe-only" and I don't really have an opinion on that, although I do think the "pci-probe-only" name is not as descriptive as it could be. I guess somebody will argue that "preserve_config" could be more descriptive, too :)

Honestly I would have liked to repurpose of_pci_check_probe_only() API to look for "preserve-boot-config" in the respective PCIe controller's DT node and not "linux,pci-probe-only" in the chosen entry, had it not for the single usage of of_pci_check_probe_only() in arch/powerpc /platforms/pseries/setup.c file.
Also FWIW, "linux,pci-probe-only" is not documented anywhere.

Since there is at least one user for of_pci_check_probe_only(), and combining with the fact that the scope where "linux,pci-probe-only" and "preserve-boot-config" are used (i.e. chosen entry Vs individual PCIe controller node), I prefer to have it as a separate option. Rob, please let me know if you have any strong objections to that?

Quoted 2 lines

Bjorn

maintainer
#6 lore raw

Re: [PATCH V2 2/2] PCI: Add support for "preserve-boot-config" property

Quoted 58 lines · Vidya Sagar
Quoted 28 lines · Bjorn Helgaas

External email: Use caution opening links or attachments

Quoted 24 lines · Vidya Sagar

Add support for "preserve-boot-config" property that can be used to selectively (i.e. per host bridge) instruct the kernel to preserve the boot time configuration done by the platform firmware.

---
V2:
* Addressed issues reported by kernel test robot <[email protected]>

  drivers/pci/controller/pci-host-common.c |  5 ++++-
  drivers/pci/of.c                         | 18 ++++++++++++++++++
  drivers/pci/probe.c                      |  2 +-
  include/linux/of_pci.h                   |  6 ++++++
  4 files changed, 29 insertions(+), 2 deletions(-)
drivers/pci/controller/pci-host-common.c +0 -0
index 6be3266cd7b5..d3475dc9ec44 100644
@@ -68,13 +68,16 @@ int pci_host_common_probe(struct platform_device *pdev)
 
       of_pci_check_probe_only();
 
Quoted 8 lines

drivers/pci/controller/pci-host-common.c int pci_host_common_probe(struct platform_device *pdev) · line 72

Quoted 2 lines
drivers/pci/controller/pci-host-common.c +2 -0
+     bridge->preserve_config =
+             of_pci_check_preserve_boot_config(dev->of_node);

Thanks for leveraging the existing "preserve_config" support for the ACPI _DSM. Is pci_host_common_probe() the best place for this? I think there are many DT platform drivers that do not use pci_host_common_probe(), so I wonder if there's a more generic place to put this.

My understanding is that pci_host_common_probe() is mainly used in systems where the firmware would have taken care of all the platform specific initialization and giving the ECAM and 'ranges' info through DT for the Linux kernel to go ahead and perform the enumeration. This is similar to ACPI way of handing over the system from firmware to the OS.

If PCIe controllers are getting initialized in the kernel itself, then pci_host_probe() is called directly from the respective host controller drivers which is the case with all the DesignWare based implementations including Tegra194 and Tegra234. In those systems, since the controllers themselves have come up and gotten initialized in the kernel, resource assignment has to happen anyway.

Quoted 6 lines

I see Rob's concern about adding "preserve-boot-config" vs extending "linux,pci-probe-only" and I don't really have an opinion on that, although I do think the "pci-probe-only" name is not as descriptive as it could be. I guess somebody will argue that "preserve_config" could be more descriptive, too :)

Quoted 6 lines

Honestly I would have liked to repurpose of_pci_check_probe_only() API to look for "preserve-boot-config" in the respective PCIe controller's DT node and not "linux,pci-probe-only" in the chosen entry, had it not for the single usage of of_pci_check_probe_only() in arch/powerpc /platforms/pseries/setup.c file.
Also FWIW, "linux,pci-probe-only" is not documented anywhere.

Yes, it is[1].

Quoted 6 lines

Since there is at least one user for of_pci_check_probe_only(), and combining with the fact that the scope where "linux,pci-probe-only" and "preserve-boot-config" are used (i.e. chosen entry Vs individual PCIe controller node), I prefer to have it as a separate option. Rob, please let me know if you have any strong objections to that?

Didn't I already object?

What's the concern with existing users? There shouldn't be any. If "linux,pci-probe-only" appeared in a bridge node, it would have been ignored and now would be honored.

Rob

[1] https://github.com/devicetree-org/dt-schema/blob/main/dtschema/schemas/chosen.yaml#L140

#7 lore raw

Re: [PATCH V2 2/2] PCI: Add support for "preserve-boot-config" property

In reply to Vidya Sagar
Quoted 41 lines · Vidya Sagar
Quoted 25 lines · Bjorn Helgaas
Quoted 24 lines · Vidya Sagar

Add support for "preserve-boot-config" property that can be used to selectively (i.e. per host bridge) instruct the kernel to preserve the boot time configuration done by the platform firmware.

---
V2:
* Addressed issues reported by kernel test robot <[email protected]>

  drivers/pci/controller/pci-host-common.c |  5 ++++-
  drivers/pci/of.c                         | 18 ++++++++++++++++++
  drivers/pci/probe.c                      |  2 +-
  include/linux/of_pci.h                   |  6 ++++++
  4 files changed, 29 insertions(+), 2 deletions(-)
drivers/pci/controller/pci-host-common.c +0 -0
index 6be3266cd7b5..d3475dc9ec44 100644
@@ -68,13 +68,16 @@ int pci_host_common_probe(struct platform_device *pdev)
 
       of_pci_check_probe_only();
 
Quoted 8 lines

drivers/pci/controller/pci-host-common.c int pci_host_common_probe(struct platform_device *pdev) · line 72

Quoted 2 lines
drivers/pci/controller/pci-host-common.c +2 -0
+     bridge->preserve_config =
+             of_pci_check_preserve_boot_config(dev->of_node);

Thanks for leveraging the existing "preserve_config" support for the ACPI _DSM. Is pci_host_common_probe() the best place for this? I think there are many DT platform drivers that do not use pci_host_common_probe(), so I wonder if there's a more generic place to put this.

My understanding is that pci_host_common_probe() is mainly used in systems where the firmware would have taken care of all the platform specific initialization and giving the ECAM and 'ranges' info through DT for the Linux kernel to go ahead and perform the enumeration. This is similar to ACPI way of handing over the system from firmware to the OS.

Quoted 6 lines

If PCIe controllers are getting initialized in the kernel itself, then pci_host_probe() is called directly from the respective host controller drivers which is the case with all the DesignWare based implementations including Tegra194 and Tegra234. In those systems, since the controllers themselves have come up and gotten initialized in the kernel, resource assignment has to happen anyway.

acpi_pci_root_create() sets "preserve_config" based on the DSM_PCI_PRESERVE_BOOT_CONFIG _DSM for all ACPI host bridges.

Similarly, I think we should set "preserve_config" based on the DT "preserve-boot-config" property for *all* DT-based host bridges, regardless of where the controller init happens.

acpi_pci_root_create
pci_create_root_bus
pci_alloc_host_bridge
pci_register_host_bridge
acpi_evaluate_dsm_typed(DSM_PCI_PRESERVE_BOOT_CONFIG) <-- pci_scan_child_bus

pci_host_common_probe
+ of_pci_check_preserve_boot_config <-- proposed
pci_host_probe
pci_scan_root_bus_bridge
pci_register_host_bridge
pci_set_bus_of_node
pci_scan_child_bus

Maybe we should do both in pci_register_host_bridge()? E.g., make a function that sets "preserve_config" based on either the ACPI _DSM or the DT property, whichever is appropriate, and call it from pci_register_host_bridge()?

Bjorn

#8 lore raw

Re: [PATCH V2 2/2] PCI: Add support for "preserve-boot-config" property

In reply to Vidya Sagar
Quoted 9 lines · Vidya Sagar

Add support for "preserve-boot-config" property that can be used to selectively (i.e. per host bridge) instruct the kernel to preserve the boot time configuration done by the platform firmware.

---
V2:
* Addressed issues reported by kernel test robot <[email protected]>

I don't think the lkp Reported-by adds useful information in this case. I have no idea what lkp actually reported, but I don't think it reported a bug that is fixed by this patch.

#9 lore raw

PATCH v3 1/1 PCI: Add support for preserving boot configuration

In reply to Vidya Sagar

Add support for preserving the boot configuration done by the platform firmware per host bridge basis, based on the presence of 'linux,pci-probe-only' property in the respective PCIe host bridge device-tree node. It also unifies the ACPI and DT based boot flows in this regard.

---
V3:
* Unified ACPI and DT flows as part of addressing Bjorn's review comments

V2:
* Addressed issues reported by kernel test robot <[email protected]>

 drivers/acpi/pci_root.c                  | 12 -------
 drivers/pci/controller/pci-host-common.c |  4 ---
 drivers/pci/of.c                         | 22 ++++++++++++
 drivers/pci/probe.c                      | 46 ++++++++++++++++++------
 include/linux/of_pci.h                   |  6 ++++
 5 files changed, 63 insertions(+), 27 deletions(-)
drivers/acpi/pci_root.c +0 -12
index 84030804a763..ddc2b3e89111 100644
@@ -1008,7 +1008,6 @@ struct pci_bus *acpi_pci_root_create(struct acpi_pci_root *root,
 	int node = acpi_get_node(device->handle);
 	struct pci_bus *bus;
 	struct pci_host_bridge *host_bridge;
-	union acpi_object *obj;
 
 	info->root = root;
 	info->bridge = device;
@@ -1050,17 +1049,6 @@ struct pci_bus *acpi_pci_root_create(struct acpi_pci_root *root,
 	if (!(root->osc_ext_control_set & OSC_CXL_ERROR_REPORTING_CONTROL))
 		host_bridge->native_cxl_error = 0;
 
-	/*
-	 * Evaluate the "PCI Boot Configuration" _DSM Function.  If it
-	 * exists and returns 0, we must preserve any PCI resource
-	 * assignments made by firmware for this host bridge.
-	 */
-	obj = acpi_evaluate_dsm(ACPI_HANDLE(bus->bridge), &pci_acpi_dsm_guid, 1,
-				DSM_PCI_PRESERVE_BOOT_CONFIG, NULL);
-	if (obj && obj->type == ACPI_TYPE_INTEGER && obj->integer.value == 0)
-		host_bridge->preserve_config = 1;
-	ACPI_FREE(obj);
-
 	acpi_dev_power_up_children_with_adr(device);
 
 	pci_scan_child_bus(bus);
drivers/pci/controller/pci-host-common.c +0 -4
index 6be3266cd7b5..e2602e38ae45 100644
@@ -73,10 +73,6 @@ int pci_host_common_probe(struct platform_device *pdev)
 	if (IS_ERR(cfg))
 		return PTR_ERR(cfg);
 
-	/* Do not reassign resources if probe only */
-	if (!pci_has_flag(PCI_PROBE_ONLY))
-		pci_add_flags(PCI_REASSIGN_ALL_BUS);
-
 	bridge->sysdata = cfg;
 	bridge->ops = (struct pci_ops *)&ops->pci_ops;
 	bridge->msi_domain = true;
drivers/pci/of.c +22 -0
index 51e3dd0ea5ab..7b553dd83587 100644
@@ -258,6 +258,28 @@ void of_pci_check_probe_only(void)
 }
 EXPORT_SYMBOL_GPL(of_pci_check_probe_only);
 
+/**
+ * of_pci_bridge_check_probe_only - Return true if the boot configuration
+ *                                  needs to be preserved
+ * @node: Device tree node with the domain information.
+ *
+ * This function looks for "linux,pci-probe-only" property for a given
+ * PCIe controller's node and returns true if found. Having this property
+ * for a PCIe controller ensures that the kernel doesn't re-enumerate and
+ * reconfigure the BAR resources that are already done by the platform firmware.
+ * NOTE: The scope of "linux,pci-probe-only" defined within a PCIe bridge device
+ *       is limited to the hierarchy under that particular bridge device. whereas
+ *       the scope of "linux,pci-probe-only" defined within chosen node is
+ *       system wide.
+ *
+ * Return: true if the property exists false otherwise.
+ */
+bool of_pci_bridge_check_probe_only(struct device_node *node)
+{
+	return of_property_read_bool(node, "linux,pci-probe-only");
+}
+EXPORT_SYMBOL_GPL(of_pci_bridge_check_probe_only);
+
 /**
  * devm_of_pci_get_host_bridge_resources() - Resource-managed parsing of PCI
  *                                           host bridge resources from DT
drivers/pci/probe.c +35 -11

73 lines not on this page · raw message

include/linux/of_pci.h +6 -0
index 29658c0ee71f..9e045de3be44 100644
@@ -13,6 +13,7 @@ struct device_node *of_pci_find_child_device(struct device_node *parent,
 					     unsigned int devfn);
 int of_pci_get_devfn(struct device_node *np);
 void of_pci_check_probe_only(void);
+bool of_pci_bridge_check_probe_only(struct device_node *node);
 #else
 static inline struct device_node *of_pci_find_child_device(struct device_node *parent,
 					     unsigned int devfn)
@@ -26,6 +27,11 @@ static inline int of_pci_get_devfn(struct device_node *np)
 }
 
 static inline void of_pci_check_probe_only(void) { }
+
+static inline bool of_pci_bridge_check_probe_only(struct device_node *node)
+{
+	return false;
+}
 #endif
 
 #if IS_ENABLED(CONFIG_OF_IRQ)
Signature · 1 line
2.25.1
#10 lore raw

Re: [PATCH V3] PCI: Add support for preserving boot configuration

Quoted 9 lines · Vidya Sagar

Add support for preserving the boot configuration done by the platform firmware per host bridge basis, based on the presence of 'linux,pci-probe-only' property in the respective PCIe host bridge device-tree node. It also unifies the ACPI and DT based boot flows in this regard.

+/**
+ * of_pci_bridge_check_probe_only - Return true if the boot configuration
+ * needs to be preserved

I don't like the "check_probe_only" name because it's a boolean function but the name doesn't tell me what a true/false return value means. Something like "preserve_resources" would be better. If you want "probe_only", even removing the "check" would help.

Quoted 6 lines

+ * @node: Device tree node with the domain information.
+ *
+ * This function looks for "linux,pci-probe-only" property for a given
+ * PCIe controller's node and returns true if found. Having this property
+ * for a PCIe controller ensures that the kernel doesn't re-enumerate and
+ * reconfigure the BAR resources that are already done by the platform firmware.

This is generic PCI, not PCIe-specific (also in commit log and comment below).

I think "enumeration" specifically refers to discovering what devices are present, and the kernel always does that, so drop that part. Reconfiguring BARs and bridge windows is what we want to prevent.

drivers/pci/of.c void of_pci_check_probe_only(void) · line 281

Quoted 13 lines

+ * NOTE: The scope of "linux,pci-probe-only" defined within a PCIe bridge device

+ *       is limited to the hierarchy under that particular bridge device. whereas
+ *       the scope of "linux,pci-probe-only" defined within chosen node is
+ *       system wide.

+ *
+ * Return: true if the property exists false otherwise.
+ */

drivers/pci/of.c +5 -0
+bool of_pci_bridge_check_probe_only(struct device_node *node)
+{
+	return of_property_read_bool(node, "linux,pci-probe-only");
+}
+EXPORT_SYMBOL_GPL(of_pci_bridge_check_probe_only);

Why does this need to be exported for modules and exposed via include/linux/pci.h?

drivers/pci/probe.c static void pci_set_bus_msi_domain(struct pci_bus *bus) · line 883

Quoted 3 lines
drivers/pci/probe.c +3 -0
+static void pci_check_config_preserve(struct pci_host_bridge *host_bridge)
+{
+	if (&host_bridge->dev) {

Checking &host_bridge->dev doesn't seem like the right way to determine whether this is an ACPI host bridge.

Quoted 13 lines
drivers/pci/probe.c +13 -0
+		union acpi_object *obj;
+
+		/*
+		 * Evaluate the "PCI Boot Configuration" _DSM Function.  If it
+		 * exists and returns 0, we must preserve any PCI resource
+		 * assignments made by firmware for this host bridge.
+		 */
+		obj = acpi_evaluate_dsm(ACPI_HANDLE(&host_bridge->dev), &pci_acpi_dsm_guid, 1,
+					DSM_PCI_PRESERVE_BOOT_CONFIG, NULL);
+		if (obj && obj->type == ACPI_TYPE_INTEGER && obj->integer.value == 0)
+			host_bridge->preserve_config = 1;
+		ACPI_FREE(obj);
+	}
#11 lore raw

Re: [PATCH V3] PCI: Add support for preserving boot configuration

Quoted 12 lines · Bjorn Helgaas

External email: Use caution opening links or attachments

Quoted 8 lines · Vidya Sagar

Add support for preserving the boot configuration done by the platform firmware per host bridge basis, based on the presence of 'linux,pci-probe-only' property in the respective PCIe host bridge device-tree node. It also unifies the ACPI and DT based boot flows in this regard.
+/**
+ * of_pci_bridge_check_probe_only - Return true if the boot configuration
+ * needs to be preserved

Quoted 4 lines

I don't like the "check_probe_only" name because it's a boolean function but the name doesn't tell me what a true/false return value means. Something like "preserve_resources" would be better. If you want "probe_only", even removing the "check" would help.

I'll change it in the next patch.

Quoted 6 lines
Quoted 6 lines

+ * @node: Device tree node with the domain information.
+ *
+ * This function looks for "linux,pci-probe-only" property for a given
+ * PCIe controller's node and returns true if found. Having this property
+ * for a PCIe controller ensures that the kernel doesn't re-enumerate and
+ * reconfigure the BAR resources that are already done by the platform firmware.

Quoted 6 lines

This is generic PCI, not PCIe-specific (also in commit log and comment below).

I think "enumeration" specifically refers to discovering what devices are present, and the kernel always does that, so drop that part. Reconfiguring BARs and bridge windows is what we want to prevent.

Agree and I'll address it in the next patch.

Quoted 6 lines
Quoted 6 lines

+ * NOTE: The scope of "linux,pci-probe-only" defined within a PCIe bridge device

+ *       is limited to the hierarchy under that particular bridge device. whereas
+ *       the scope of "linux,pci-probe-only" defined within chosen node is
+ *       system wide.

+ *
+ * Return: true if the property exists false otherwise.

Quoted 8 lines

drivers/pci/of.c void of_pci_check_probe_only(void) · line 281

Quoted 6 lines

+ */

drivers/pci/of.c +5 -0
+bool of_pci_bridge_check_probe_only(struct device_node *node)
+{
+     return of_property_read_bool(node, "linux,pci-probe-only");
+}
+EXPORT_SYMBOL_GPL(of_pci_bridge_check_probe_only);

Why does this need to be exported for modules and exposed via include/linux/pci.h?

On a second through, I think it is not required to be exported. I'll address this also in the next patch.

Quoted 5 lines

drivers/pci/probe.c static void pci_set_bus_msi_domain(struct pci_bus *bus) · line 883

Quoted 3 lines
drivers/pci/probe.c +3 -0
+static void pci_check_config_preserve(struct pci_host_bridge *host_bridge)
+{
+     if (&host_bridge->dev) {

Checking &host_bridge->dev doesn't seem like the right way to determine whether this is an ACPI host bridge.

Honestly, I couldn't find a clear way to differentiate between an ACPI based host bridge and a DT based host bridge. Hence, the current code tries to get the information using both ways and since a system can only be either ACPI or DT based, but one at a time, preserve_config will be set only once (assuming the system wants it to be set). Let me know if there is a better approach for this?

I was looking at the way 'external_facing' gets set in both the boot flows and I see that there is no common place for it and the respective flows have their functions separately i.e. pci_acpi_set_external_facing() for ACPI and pci_set_bus_of_node() for DT.

Thanks,
Vidya Sagar

Quoted 13 lines
Quoted 13 lines
drivers/pci/probe.c +13 -0
+             union acpi_object *obj;
+
+             /*
+              * Evaluate the "PCI Boot Configuration" _DSM Function.  If it
+              * exists and returns 0, we must preserve any PCI resource
+              * assignments made by firmware for this host bridge.
+              */
+             obj = acpi_evaluate_dsm(ACPI_HANDLE(&host_bridge->dev), &pci_acpi_dsm_guid, 1,
+                                     DSM_PCI_PRESERVE_BOOT_CONFIG, NULL);
+             if (obj && obj->type == ACPI_TYPE_INTEGER && obj->integer.value == 0)
+                     host_bridge->preserve_config = 1;
+             ACPI_FREE(obj);
+     }
#12 lore raw

Re: [PATCH V3] PCI: Add support for preserving boot configuration

Quoted 6 lines · Vidya Sagar
Quoted 5 lines · Bjorn Helgaas
Quoted 1 line · Vidya Sagar

+ if (&host_bridge->dev) {

Checking &host_bridge->dev doesn't seem like the right way to determine whether this is an ACPI host bridge.

BTW, I think this condition is *always* true, since it's testing the address of a member of a struct.

Quoted 6 lines

Honestly, I couldn't find a clear way to differentiate between an ACPI based host bridge and a DT based host bridge. Hence, the current code tries to get the information using both ways and since a system can only be either ACPI or DT based, but one at a time, preserve_config will be set only once (assuming the system wants it to be set). Let me know if there is a better approach for this?

I'm not sure ACPI and DT will always be mutually exclusive; I think we're headed toward some combinations, e.g.,
https://lore.kernel.org/linux-pci/[email protected]/

But I think "if (ACPI_HANDLE(&host_bridge->dev))" would work.

Bjorn

#13 lore raw

PATCH v4 1/1 PCI: Add support for preserving boot configuration

In reply to Vidya Sagar

Add support for preserving the boot configuration done by the platform firmware per host bridge basis, based on the presence of 'linux,pci-probe-only' property in the respective PCI host bridge device-tree node. It also unifies the ACPI and DT based boot flows in this regard.

---
V4:
* Addressed Bjorn's review comments

V3:
* Unified ACPI and DT flows as part of addressing Bjorn's review comments

V2:
* Addressed issues reported by kernel test robot <[email protected]>

 drivers/acpi/pci_root.c                  | 12 -------
 drivers/pci/controller/pci-host-common.c |  4 ---
 drivers/pci/of.c                         | 21 +++++++++++
 drivers/pci/probe.c                      | 46 ++++++++++++++++++------
 include/linux/of_pci.h                   |  6 ++++
 5 files changed, 62 insertions(+), 27 deletions(-)
drivers/acpi/pci_root.c +0 -12
index 84030804a763..ddc2b3e89111 100644
@@ -1008,7 +1008,6 @@ struct pci_bus *acpi_pci_root_create(struct acpi_pci_root *root,
 	int node = acpi_get_node(device->handle);
 	struct pci_bus *bus;
 	struct pci_host_bridge *host_bridge;
-	union acpi_object *obj;
 
 	info->root = root;
 	info->bridge = device;
@@ -1050,17 +1049,6 @@ struct pci_bus *acpi_pci_root_create(struct acpi_pci_root *root,
 	if (!(root->osc_ext_control_set & OSC_CXL_ERROR_REPORTING_CONTROL))
 		host_bridge->native_cxl_error = 0;
 
-	/*
-	 * Evaluate the "PCI Boot Configuration" _DSM Function.  If it
-	 * exists and returns 0, we must preserve any PCI resource
-	 * assignments made by firmware for this host bridge.
-	 */
-	obj = acpi_evaluate_dsm(ACPI_HANDLE(bus->bridge), &pci_acpi_dsm_guid, 1,
-				DSM_PCI_PRESERVE_BOOT_CONFIG, NULL);
-	if (obj && obj->type == ACPI_TYPE_INTEGER && obj->integer.value == 0)
-		host_bridge->preserve_config = 1;
-	ACPI_FREE(obj);
-
 	acpi_dev_power_up_children_with_adr(device);
 
 	pci_scan_child_bus(bus);
drivers/pci/controller/pci-host-common.c +0 -4
index 6be3266cd7b5..e2602e38ae45 100644
@@ -73,10 +73,6 @@ int pci_host_common_probe(struct platform_device *pdev)
 	if (IS_ERR(cfg))
 		return PTR_ERR(cfg);
 
-	/* Do not reassign resources if probe only */
-	if (!pci_has_flag(PCI_PROBE_ONLY))
-		pci_add_flags(PCI_REASSIGN_ALL_BUS);
-
 	bridge->sysdata = cfg;
 	bridge->ops = (struct pci_ops *)&ops->pci_ops;
 	bridge->msi_domain = true;
drivers/pci/of.c +21 -0
index 51e3dd0ea5ab..f0f1156040a5 100644
@@ -258,6 +258,27 @@ void of_pci_check_probe_only(void)
 }
 EXPORT_SYMBOL_GPL(of_pci_check_probe_only);
 
+/**
+ * of_pci_bridge_preserve_resources - Return true if the boot configuration
+ *                                    needs to be preserved
+ * @node: Device tree node with the domain information.
+ *
+ * This function looks for "linux,pci-probe-only" property for a given
+ * PCI controller's node and returns true if found. Having this property
+ * for a PCI controller ensures that the kernel doesn't reconfigure the
+ * BARs and bridge windows that are already done by the platform firmware.
+ * NOTE: The scope of "linux,pci-probe-only" defined within a PCI bridge device
+ *       is limited to the hierarchy under that particular bridge device. whereas
+ *       the scope of "linux,pci-probe-only" defined within chosen node is
+ *       system wide.
+ *
+ * Return: true if the property exists false otherwise.
+ */
+bool of_pci_bridge_preserve_resources(struct device_node *node)
+{
+	return of_property_read_bool(node, "linux,pci-probe-only");
+}
+
 /**
  * devm_of_pci_get_host_bridge_resources() - Resource-managed parsing of PCI
  *                                           host bridge resources from DT
drivers/pci/probe.c +35 -11

73 lines not on this page · raw message

include/linux/of_pci.h +6 -0
index 29658c0ee71f..3f3909a5d55d 100644
@@ -13,6 +13,7 @@ struct device_node *of_pci_find_child_device(struct device_node *parent,
 					     unsigned int devfn);
 int of_pci_get_devfn(struct device_node *np);
 void of_pci_check_probe_only(void);
+bool of_pci_bridge_preserve_resources(struct device_node *node);
 #else
 static inline struct device_node *of_pci_find_child_device(struct device_node *parent,
 					     unsigned int devfn)
@@ -26,6 +27,11 @@ static inline int of_pci_get_devfn(struct device_node *np)
 }
 
 static inline void of_pci_check_probe_only(void) { }
+
+static inline bool of_pci_bridge_preserve_resources(struct device_node *node)
+{
+	return false;
+}
 #endif
 
 #if IS_ENABLED(CONFIG_OF_IRQ)
Signature · 1 line
2.25.1
maintainer
#14 lore raw

Re: [PATCH V4] PCI: Add support for preserving boot configuration

Quoted 86 lines · Vidya Sagar

Add support for preserving the boot configuration done by the platform firmware per host bridge basis, based on the presence of 'linux,pci-probe-only' property in the respective PCI host bridge device-tree node. It also unifies the ACPI and DT based boot flows in this regard.

---
V4:
* Addressed Bjorn's review comments

V3:
* Unified ACPI and DT flows as part of addressing Bjorn's review comments

V2:
* Addressed issues reported by kernel test robot <[email protected]>

 drivers/acpi/pci_root.c                  | 12 -------
 drivers/pci/controller/pci-host-common.c |  4 ---
 drivers/pci/of.c                         | 21 +++++++++++
 drivers/pci/probe.c                      | 46 ++++++++++++++++++------
 include/linux/of_pci.h                   |  6 ++++
 5 files changed, 62 insertions(+), 27 deletions(-)
drivers/acpi/pci_root.c +0 -12
index 84030804a763..ddc2b3e89111 100644
@@ -1008,7 +1008,6 @@ struct pci_bus *acpi_pci_root_create(struct acpi_pci_root *root,
 	int node = acpi_get_node(device->handle);
 	struct pci_bus *bus;
 	struct pci_host_bridge *host_bridge;
-	union acpi_object *obj;
 
 	info->root = root;
 	info->bridge = device;
@@ -1050,17 +1049,6 @@ struct pci_bus *acpi_pci_root_create(struct acpi_pci_root *root,
 	if (!(root->osc_ext_control_set & OSC_CXL_ERROR_REPORTING_CONTROL))
 		host_bridge->native_cxl_error = 0;
 
-	/*
-	 * Evaluate the "PCI Boot Configuration" _DSM Function.  If it
-	 * exists and returns 0, we must preserve any PCI resource
-	 * assignments made by firmware for this host bridge.
-	 */
-	obj = acpi_evaluate_dsm(ACPI_HANDLE(bus->bridge), &pci_acpi_dsm_guid, 1,
-				DSM_PCI_PRESERVE_BOOT_CONFIG, NULL);
-	if (obj && obj->type == ACPI_TYPE_INTEGER && obj->integer.value == 0)
-		host_bridge->preserve_config = 1;
-	ACPI_FREE(obj);
-
 	acpi_dev_power_up_children_with_adr(device);
 
 	pci_scan_child_bus(bus);
drivers/pci/controller/pci-host-common.c +0 -4
index 6be3266cd7b5..e2602e38ae45 100644
@@ -73,10 +73,6 @@ int pci_host_common_probe(struct platform_device *pdev)
 	if (IS_ERR(cfg))
 		return PTR_ERR(cfg);
 
-	/* Do not reassign resources if probe only */
-	if (!pci_has_flag(PCI_PROBE_ONLY))
-		pci_add_flags(PCI_REASSIGN_ALL_BUS);
-
 	bridge->sysdata = cfg;
 	bridge->ops = (struct pci_ops *)&ops->pci_ops;
 	bridge->msi_domain = true;
drivers/pci/of.c +9 -0
index 51e3dd0ea5ab..f0f1156040a5 100644
@@ -258,6 +258,27 @@ void of_pci_check_probe_only(void)
 }
 EXPORT_SYMBOL_GPL(of_pci_check_probe_only);
 
+/**
+ * of_pci_bridge_preserve_resources - Return true if the boot configuration
+ *                                    needs to be preserved
+ * @node: Device tree node with the domain information.
+ *
+ * This function looks for "linux,pci-probe-only" property for a given
+ * PCI controller's node and returns true if found. Having this property
+ * for a PCI controller ensures that the kernel doesn't reconfigure the
+ * BARs and bridge windows that are already done by the platform firmware.

drivers/pci/of.c void of_pci_check_probe_only(void) · line 279

Quoted 10 lines
drivers/pci/of.c +10 -0
+ * NOTE: The scope of "linux,pci-probe-only" defined within a PCI bridge device
+ *       is limited to the hierarchy under that particular bridge device. whereas
+ *       the scope of "linux,pci-probe-only" defined within chosen node is
+ *       system wide.
+ *
+ * Return: true if the property exists false otherwise.
+ */
+bool of_pci_bridge_preserve_resources(struct device_node *node)
+{
+	return of_property_read_bool(node, "linux,pci-probe-only");

This is the wrong type. The existing "linux,pci-probe-only" is a u32 and non-zero value means probe-only. This would return true for 'linux,pci-probe-only = <0>'.

Also, this should also check chosen. If you make this work accepting NULL for node, then of_pci_check_probe_only() can be re-implemented to use it.

Rob

maintainer
#15 lore raw

Re: [PATCH V4] PCI: Add support for preserving boot configuration

In reply to Vidya Sagar
Quoted 17 lines · Vidya Sagar

Add support for preserving the boot configuration done by the platform firmware per host bridge basis, based on the presence of 'linux,pci-probe-only' property in the respective PCI host bridge device-tree node. It also unifies the ACPI and DT based boot flows in this regard.

---
V4:
* Addressed Bjorn's review comments

V3:
* Unified ACPI and DT flows as part of addressing Bjorn's review comments

V2:
* Addressed issues reported by kernel test robot <[email protected]>

Quoted 6 lines
 drivers/acpi/pci_root.c                  | 12 -------
 drivers/pci/controller/pci-host-common.c |  4 ---
 drivers/pci/of.c                         | 21 +++++++++++
 drivers/pci/probe.c                      | 46 ++++++++++++++++++------
 include/linux/of_pci.h                   |  6 ++++
 5 files changed, 62 insertions(+), 27 deletions(-)

One more thing.

drivers/pci/probe.c int pci_host_probe(struct pci_host_bridge *bridge) · line 3111

Quoted 7 lines
drivers/pci/probe.c +3 -0
@@ -3080,20 +3106,18 @@ int pci_host_probe(struct pci_host_bridge *bridge)
 
 	bus = bridge->bus;
 
+	/* If we must preserve the resource configuration, claim now */
+	if (pci_has_flag(PCI_PROBE_ONLY) || bridge->preserve_config)
+		pci_bus_claim_resources(bus);

No reason to check PCI_PROBE_ONLY if you set preserve_config based on /chosen as well. IOW, we should deprecate PCI_PROBE_ONLY flag in favor of the per host bridge setting.

Rob

#16 lore raw

PATCH v5 1/1 PCI: Add support for preserving boot configuration

In reply to Vidya Sagar

Add support for preserving the boot configuration done by the platform firmware per host bridge basis, based on the presence of 'linux,pci-probe-only' property in the respective PCI host bridge device-tree node. It also unifies the ACPI and DT based boot flows in this regard.

---
V5:
* Addressed Rob's review comments

V4:
* Addressed Bjorn's review comments

V3:
* Unified ACPI and DT flows as part of addressing Bjorn's review comments

V2:
* Addressed issues reported by kernel test robot <[email protected]>

 drivers/acpi/pci_root.c                  | 12 -----
 drivers/pci/controller/pci-host-common.c |  4 --
 drivers/pci/of.c                         | 57 +++++++++++++++++++-----
 drivers/pci/probe.c                      | 46 ++++++++++++++-----
 include/linux/of_pci.h                   |  6 +++
 5 files changed, 88 insertions(+), 37 deletions(-)
drivers/acpi/pci_root.c +0 -12
index 84030804a763..ddc2b3e89111 100644
@@ -1008,7 +1008,6 @@ struct pci_bus *acpi_pci_root_create(struct acpi_pci_root *root,
 	int node = acpi_get_node(device->handle);
 	struct pci_bus *bus;
 	struct pci_host_bridge *host_bridge;
-	union acpi_object *obj;
 
 	info->root = root;
 	info->bridge = device;
@@ -1050,17 +1049,6 @@ struct pci_bus *acpi_pci_root_create(struct acpi_pci_root *root,
 	if (!(root->osc_ext_control_set & OSC_CXL_ERROR_REPORTING_CONTROL))
 		host_bridge->native_cxl_error = 0;
 
-	/*
-	 * Evaluate the "PCI Boot Configuration" _DSM Function.  If it
-	 * exists and returns 0, we must preserve any PCI resource
-	 * assignments made by firmware for this host bridge.
-	 */
-	obj = acpi_evaluate_dsm(ACPI_HANDLE(bus->bridge), &pci_acpi_dsm_guid, 1,
-				DSM_PCI_PRESERVE_BOOT_CONFIG, NULL);
-	if (obj && obj->type == ACPI_TYPE_INTEGER && obj->integer.value == 0)
-		host_bridge->preserve_config = 1;
-	ACPI_FREE(obj);
-
 	acpi_dev_power_up_children_with_adr(device);
 
 	pci_scan_child_bus(bus);
drivers/pci/controller/pci-host-common.c +0 -4
index 6be3266cd7b5..e2602e38ae45 100644
@@ -73,10 +73,6 @@ int pci_host_common_probe(struct platform_device *pdev)
 	if (IS_ERR(cfg))
 		return PTR_ERR(cfg);
 
-	/* Do not reassign resources if probe only */
-	if (!pci_has_flag(PCI_PROBE_ONLY))
-		pci_add_flags(PCI_REASSIGN_ALL_BUS);
-
 	bridge->sysdata = cfg;
 	bridge->ops = (struct pci_ops *)&ops->pci_ops;
 	bridge->msi_domain = true;
drivers/pci/of.c +47 -10

71 lines not on this page · raw message

drivers/pci/probe.c +35 -11

73 lines not on this page · raw message

include/linux/of_pci.h +6 -0
index 29658c0ee71f..3f3909a5d55d 100644
@@ -13,6 +13,7 @@ struct device_node *of_pci_find_child_device(struct device_node *parent,
 					     unsigned int devfn);
 int of_pci_get_devfn(struct device_node *np);
 void of_pci_check_probe_only(void);
+bool of_pci_bridge_preserve_resources(struct device_node *node);
 #else
 static inline struct device_node *of_pci_find_child_device(struct device_node *parent,
 					     unsigned int devfn)
@@ -26,6 +27,11 @@ static inline int of_pci_get_devfn(struct device_node *np)
 }
 
 static inline void of_pci_check_probe_only(void) { }
+
+static inline bool of_pci_bridge_preserve_resources(struct device_node *node)
+{
+	return false;
+}
 #endif
 
 #if IS_ENABLED(CONFIG_OF_IRQ)
Signature · 1 line
2.25.1
maintainer Reviewed-by on v5 1/1
#17 lore raw

Re: [PATCH V5] PCI: Add support for preserving boot configuration

Quoted 20 lines · Vidya Sagar

Add support for preserving the boot configuration done by the platform firmware per host bridge basis, based on the presence of 'linux,pci-probe-only' property in the respective PCI host bridge device-tree node. It also unifies the ACPI and DT based boot flows in this regard.

---
V5:
* Addressed Rob's review comments

V4:
* Addressed Bjorn's review comments

V3:
* Unified ACPI and DT flows as part of addressing Bjorn's review comments

V2:
* Addressed issues reported by kernel test robot <[email protected]>

Quoted 7 lines
 drivers/acpi/pci_root.c                  | 12 -----
 drivers/pci/controller/pci-host-common.c |  4 --
 drivers/pci/of.c                         | 57 +++++++++++++++++++-----
 drivers/pci/probe.c                      | 46 ++++++++++++++-----
 include/linux/of_pci.h                   |  6 +++
 5 files changed, 88 insertions(+), 37 deletions(-)
#18 lore raw

Re: [PATCH V5] PCI: Add support for preserving boot configuration

Thanks Rob for the review.
Bjorn, do you have any further comments for this patch?

Quoted 33 lines · Rob Herring

External email: Use caution opening links or attachments

Quoted 20 lines · Vidya Sagar

Add support for preserving the boot configuration done by the platform firmware per host bridge basis, based on the presence of 'linux,pci-probe-only' property in the respective PCI host bridge device-tree node. It also unifies the ACPI and DT based boot flows in this regard.

---
V5:
* Addressed Rob's review comments

V4:
* Addressed Bjorn's review comments

V3:
* Unified ACPI and DT flows as part of addressing Bjorn's review comments

V2:
* Addressed issues reported by kernel test robot <[email protected]>

Quoted 7 lines
  drivers/acpi/pci_root.c                  | 12 -----
  drivers/pci/controller/pci-host-common.c |  4 --
  drivers/pci/of.c                         | 57 +++++++++++++++++++-----
  drivers/pci/probe.c                      | 46 ++++++++++++++-----
  include/linux/of_pci.h                   |  6 +++
  5 files changed, 88 insertions(+), 37 deletions(-)
#19 lore raw

Re: [PATCH V5] PCI: Add support for preserving boot configuration

In reply to Vidya Sagar
Quoted 6 lines · Vidya Sagar

Add support for preserving the boot configuration done by the platform firmware per host bridge basis, based on the presence of 'linux,pci-probe-only' property in the respective PCI host bridge device-tree node. It also unifies the ACPI and DT based boot flows in this regard.

Quoted 6 lines
 drivers/acpi/pci_root.c                  | 12 -----
 drivers/pci/controller/pci-host-common.c |  4 --
 drivers/pci/of.c                         | 57 +++++++++++++++++++-----
 drivers/pci/probe.c                      | 46 ++++++++++++++-----
 include/linux/of_pci.h                   |  6 +++
 5 files changed, 88 insertions(+), 37 deletions(-)

What does this apply to? I tried v6.9-rc1:

  $ git checkout -b wip/2404-vidya-preserve-boot-v5 v6.9-rc1
  Switched to a new branch 'wip/2404-vidya-preserve-boot-v5'
  $ git am m/v5_20240401_vidyas_pci_add_support_for_preserving_boot_configuration.mbx
  Applying: PCI: Add support for preserving boot configuration
  error: patch failed: drivers/acpi/pci_root.c:1050
  error: drivers/acpi/pci_root.c: patch does not apply
  Patch failed at 0001 PCI: Add support for preserving boot configuration
Quoted 50 lines
drivers/acpi/pci_root.c +0 -12
index 84030804a763..ddc2b3e89111 100644
@@ -1008,7 +1008,6 @@ struct pci_bus *acpi_pci_root_create(struct acpi_pci_root *root,
 	int node = acpi_get_node(device->handle);
 	struct pci_bus *bus;
 	struct pci_host_bridge *host_bridge;
-	union acpi_object *obj;
 
 	info->root = root;
 	info->bridge = device;
@@ -1050,17 +1049,6 @@ struct pci_bus *acpi_pci_root_create(struct acpi_pci_root *root,
 	if (!(root->osc_ext_control_set & OSC_CXL_ERROR_REPORTING_CONTROL))
 		host_bridge->native_cxl_error = 0;
 
-	/*
-	 * Evaluate the "PCI Boot Configuration" _DSM Function.  If it
-	 * exists and returns 0, we must preserve any PCI resource
-	 * assignments made by firmware for this host bridge.
-	 */
-	obj = acpi_evaluate_dsm(ACPI_HANDLE(bus->bridge), &pci_acpi_dsm_guid, 1,
-				DSM_PCI_PRESERVE_BOOT_CONFIG, NULL);
-	if (obj && obj->type == ACPI_TYPE_INTEGER && obj->integer.value == 0)
-		host_bridge->preserve_config = 1;
-	ACPI_FREE(obj);
-
 	acpi_dev_power_up_children_with_adr(device);
 
 	pci_scan_child_bus(bus);
drivers/pci/controller/pci-host-common.c +0 -4
index 6be3266cd7b5..e2602e38ae45 100644
@@ -73,10 +73,6 @@ int pci_host_common_probe(struct platform_device *pdev)
 	if (IS_ERR(cfg))
 		return PTR_ERR(cfg);
 
-	/* Do not reassign resources if probe only */
-	if (!pci_has_flag(PCI_PROBE_ONLY))
-		pci_add_flags(PCI_REASSIGN_ALL_BUS);
-
 	bridge->sysdata = cfg;
 	bridge->ops = (struct pci_ops *)&ops->pci_ops;
 	bridge->msi_domain = true;
drivers/pci/of.c +0 -0
index 51e3dd0ea5ab..e6da3654f9ac 100644

drivers/pci/of.c EXPORT_SYMBOL_GPL(of_get_pci_domain_nr); · line 242

Quoted 5 lines
drivers/pci/of.c +1 -1
@@ -239,24 +239,61 @@ EXPORT_SYMBOL_GPL(of_get_pci_domain_nr);
  */
 void of_pci_check_probe_only(void)
 {
-	u32 val;
+	bool is_preserve_config = of_pci_bridge_preserve_resources(of_chosen);

No need for a bool here:

  if (of_pci_bridge_preserve_resources(of_chosen))
    pci_add_flags(PCI_PROBE_ONLY);

I think it would make more sense to add of_pci_bridge_preserve_resources() *above* of_pci_check_probe_only() since the usual order is to define functions earlier in the file than the calls.

drivers/pci/of.c EXPORT_SYMBOL_GPL(of_get_pci_domain_nr); · line 247

Quoted 5 lines
drivers/pci/of.c +5 -0
+
+	if (is_preserve_config)
+		pci_add_flags(PCI_PROBE_ONLY);
+	else
+		pci_clear_flags(PCI_PROBE_ONLY);

Not related to *this* patch, but I see that of_pci_check_probe_only() already clears PCI_PROBE_ONLY (added by f81c11af617c ("of/pci: Add of_pci_check_probe_only to parse "linux,pci-probe-only"").

I'm concerned about clearing PCI_PROBE_ONLY because some platforms set this unconditionally, and I don't think they necessarily have "linux,pci-probe-only" in DT.

Apparently none of them currently calls of_pci_check_probe_only() so PCI_PROBE_ONLY remains set, but clearing it here feels like a landmine waiting for somebody to move this into a unified call path.

I guess if we were to drop pci_clear_flags(), that should be a separate patch in case it breaks something.

Quoted 102 lines
drivers/pci/of.c +40 -9
+	pr_info("PROBE_ONLY %s\n", is_preserve_config ? "enabled" : "disabled");
+}
+EXPORT_SYMBOL_GPL(of_pci_check_probe_only);
+
+/**
+ * of_pci_bridge_preserve_resources - Return true if the boot configuration
+ *                                    needs to be preserved
+ * @node: Device tree node.
+ *
+ * This function looks for "linux,pci-probe-only" property for a given
+ * PCI controller's node and returns true if found. It will also look in the
+ * chosen node if the property is not found in the given controller's node.
+ * Having this property ensures that the kernel doesn't reconfigure the
+ * BARs and bridge windows that are already done by the platform firmware.
+ *
+ * Return: true if the property exists false otherwise.
+ */
+bool of_pci_bridge_preserve_resources(struct device_node *node)
+{
+	u32 val = 0;
 	int ret;
 
-	ret = of_property_read_u32(of_chosen, "linux,pci-probe-only", &val);
+	if (!node) {
+		pr_warn("device node is NULL, trying with of_chosen\n");
+		node = of_chosen;
+	}
+
+retry:
+	ret = of_property_read_u32(node, "linux,pci-probe-only", &val);
 	if (ret) {
-		if (ret == -ENODATA || ret == -EOVERFLOW)
-			pr_warn("linux,pci-probe-only without valid value, ignoring\n");
-		return;
+		if (ret == -ENODATA || ret == -EOVERFLOW) {
+			pr_warn("Incorrect value for linux,pci-probe-only in %pOF, ignoring\n", node);
+			return false;
+		}
+		if (ret == -EINVAL) {
+			if (node == of_chosen)
+				return false;
+
+			node = of_chosen;
+			goto retry;
+		}
 	}
 
 	if (val)
-		pci_add_flags(PCI_PROBE_ONLY);
+		return true;
 	else
-		pci_clear_flags(PCI_PROBE_ONLY);
-
-	pr_info("PROBE_ONLY %s\n", val ? "enabled" : "disabled");
+		return false;
 }
-EXPORT_SYMBOL_GPL(of_pci_check_probe_only);
 
 /**
  * devm_of_pci_get_host_bridge_resources() - Resource-managed parsing of PCI
drivers/pci/probe.c +23 -0
index 795534589b98..b0e0226a8da8 100644
@@ -15,6 +15,7 @@
 #include <linux/cpumask.h>
 #include <linux/aer.h>
 #include <linux/acpi.h>
+#include <linux/pci-acpi.h>
 #include <linux/hypervisor.h>
 #include <linux/irqdomain.h>
 #include <linux/pm_runtime.h>
@@ -877,6 +878,28 @@ static void pci_set_bus_msi_domain(struct pci_bus *bus)
 	dev_set_msi_domain(&bus->dev, d);
 }
 
+static void pci_check_config_preserve(struct pci_host_bridge *host_bridge)
+{
+	if (ACPI_HANDLE(&host_bridge->dev)) {
+		union acpi_object *obj;
+
+		/*
+		 * Evaluate the "PCI Boot Configuration" _DSM Function.  If it
+		 * exists and returns 0, we must preserve any PCI resource
+		 * assignments made by firmware for this host bridge.
+		 */
+		obj = acpi_evaluate_dsm(ACPI_HANDLE(&host_bridge->dev), &pci_acpi_dsm_guid, 1,
+					DSM_PCI_PRESERVE_BOOT_CONFIG, NULL);
+		if (obj && obj->type == ACPI_TYPE_INTEGER && obj->integer.value == 0)
+			host_bridge->preserve_config = 1;
+		ACPI_FREE(obj);
+	}
+
+	if (host_bridge->dev.parent && host_bridge->dev.parent->of_node)
+		host_bridge->preserve_config =
+			of_pci_bridge_preserve_resources(host_bridge->dev.parent->of_node);
+}
+
 static int pci_register_host_bridge(struct pci_host_bridge *bridge)
 {
 	struct device *parent = bridge->dev.parent;

drivers/pci/probe.c static int pci_register_host_bridge(struct pci_host_bridge *bridge) · line 998

Quoted 5 lines
drivers/pci/probe.c +2 -0
@@ -971,6 +994,9 @@ static int pci_register_host_bridge(struct pci_host_bridge *bridge)
 	if (nr_node_ids > 1 && pcibus_to_node(bus) == NUMA_NO_NODE)
 		dev_warn(&bus->dev, "Unknown NUMA node; performance will be reduced\n");
 
+	/* Check if the boot configuration by FW needs to be preserved */
+	pci_check_config_preserve(bridge);

I really have an allergic reaction to functions named "..._check_..." We can tell that the function has something to do with preserving configuration, and it's void so obviously the side effects are the important thing, but there's no clue in the caller about what the side effects are.

I'd prefer something like:

  bridge->preserve_config = pci_must_preserve_config(bridge);

where pci_must_preserve_config() has no side effects, returns bool, and the action is at the caller.

Quoted 38 lines
drivers/pci/probe.c +9 -11
 	/* Coalesce contiguous windows */
 	resource_list_for_each_entry_safe(window, n, &resources) {
 		if (list_is_last(&window->node, &resources))
@@ -3080,20 +3106,18 @@ int pci_host_probe(struct pci_host_bridge *bridge)
 
 	bus = bridge->bus;
 
+	/* If we must preserve the resource configuration, claim now */
+	if (bridge->preserve_config)
+		pci_bus_claim_resources(bus);
+
 	/*
-	 * We insert PCI resources into the iomem_resource and
-	 * ioport_resource trees in either pci_bus_claim_resources()
-	 * or pci_bus_assign_resources().
+	 * Assign whatever was left unassigned. If we didn't claim above,
+	 * this will reassign everything.
 	 */
-	if (pci_has_flag(PCI_PROBE_ONLY)) {
-		pci_bus_claim_resources(bus);
-	} else {
-		pci_bus_size_bridges(bus);
-		pci_bus_assign_resources(bus);
+	pci_assign_unassigned_root_bus_resources(bus);
 
-		list_for_each_entry(child, &bus->children, node)
-			pcie_bus_configure_settings(child);
-	}
+	list_for_each_entry(child, &bus->children, node)
+		pcie_bus_configure_settings(child);
 
 	pci_bus_add_devices(bus);
 	return 0;
include/linux/of_pci.h +0 -0
index 29658c0ee71f..3f3909a5d55d 100644

include/linux/of_pci.h struct device_node *of_pci_find_child_device(struct device_node *parent, · line 16

Quoted 4 lines
include/linux/of_pci.h +1 -0
@@ -13,6 +13,7 @@ struct device_node *of_pci_find_child_device(struct device_node *parent,
 					     unsigned int devfn);
 int of_pci_get_devfn(struct device_node *np);
 void of_pci_check_probe_only(void);
+bool of_pci_bridge_preserve_resources(struct device_node *node);

This looks like it should be in drivers/pci/pci.h since it's not used outside drivers/pci/.

Quoted 18 lines
include/linux/of_pci.h +5 -0
 #else
 static inline struct device_node *of_pci_find_child_device(struct device_node *parent,
 					     unsigned int devfn)
@@ -26,6 +27,11 @@ static inline int of_pci_get_devfn(struct device_node *np)
 }
 
 static inline void of_pci_check_probe_only(void) { }
+
+static inline bool of_pci_bridge_preserve_resources(struct device_node *node)
+{
+	return false;
+}
 #endif
 
 #if IS_ENABLED(CONFIG_OF_IRQ)
Signature · 1 line
2.25.1
#20 lore raw

Re: [PATCH V5] PCI: Add support for preserving boot configuration

Replied inline...

Quoted 17 lines · Bjorn Helgaas

External email: Use caution opening links or attachments

Quoted 11 lines · Vidya Sagar

Add support for preserving the boot configuration done by the platform firmware per host bridge basis, based on the presence of 'linux,pci-probe-only' property in the respective PCI host bridge device-tree node. It also unifies the ACPI and DT based boot flows in this regard.

  drivers/acpi/pci_root.c                  | 12 -----
  drivers/pci/controller/pci-host-common.c |  4 --
  drivers/pci/of.c                         | 57 +++++++++++++++++++-----
  drivers/pci/probe.c                      | 46 ++++++++++++++-----
  include/linux/of_pci.h                   |  6 +++
  5 files changed, 88 insertions(+), 37 deletions(-)

What does this apply to? I tried v6.9-rc1:

Quoted 8 lines
   $ git checkout -b wip/2404-vidya-preserve-boot-v5 v6.9-rc1
   Switched to a new branch 'wip/2404-vidya-preserve-boot-v5'
   $ git am m/v5_20240401_vidyas_pci_add_support_for_preserving_boot_configuration.mbx
   Applying: PCI: Add support for preserving boot configuration
   error: patch failed: drivers/acpi/pci_root.c:1050
   error: drivers/acpi/pci_root.c: patch does not apply
   Patch failed at 0001 PCI: Add support for preserving boot configuration

I'm going to rebase this patch on top of linux-next and send.

Quoted 57 lines
Quoted 50 lines
drivers/acpi/pci_root.c +0 -12
index 84030804a763..ddc2b3e89111 100644
@@ -1008,7 +1008,6 @@ struct pci_bus *acpi_pci_root_create(struct acpi_pci_root *root,
       int node = acpi_get_node(device->handle);
       struct pci_bus *bus;
       struct pci_host_bridge *host_bridge;
-     union acpi_object *obj;
 
       info->root = root;
       info->bridge = device;
@@ -1050,17 +1049,6 @@ struct pci_bus *acpi_pci_root_create(struct acpi_pci_root *root,
       if (!(root->osc_ext_control_set & OSC_CXL_ERROR_REPORTING_CONTROL))
               host_bridge->native_cxl_error = 0;
 
-     /*
-      * Evaluate the "PCI Boot Configuration" _DSM Function.  If it
-      * exists and returns 0, we must preserve any PCI resource
-      * assignments made by firmware for this host bridge.
-      */
-     obj = acpi_evaluate_dsm(ACPI_HANDLE(bus->bridge), &pci_acpi_dsm_guid, 1,
-                             DSM_PCI_PRESERVE_BOOT_CONFIG, NULL);
-     if (obj && obj->type == ACPI_TYPE_INTEGER && obj->integer.value == 0)
-             host_bridge->preserve_config = 1;
-     ACPI_FREE(obj);
-
       acpi_dev_power_up_children_with_adr(device);
 
       pci_scan_child_bus(bus);
drivers/pci/controller/pci-host-common.c +0 -4
index 6be3266cd7b5..e2602e38ae45 100644
@@ -73,10 +73,6 @@ int pci_host_common_probe(struct platform_device *pdev)
       if (IS_ERR(cfg))
               return PTR_ERR(cfg);
 
-     /* Do not reassign resources if probe only */
-     if (!pci_has_flag(PCI_PROBE_ONLY))
-             pci_add_flags(PCI_REASSIGN_ALL_BUS);
-
       bridge->sysdata = cfg;
       bridge->ops = (struct pci_ops *)&ops->pci_ops;
       bridge->msi_domain = true;
drivers/pci/of.c +0 -0
index 51e3dd0ea5ab..e6da3654f9ac 100644

drivers/pci/of.c EXPORT_SYMBOL_GPL(of_get_pci_domain_nr); · line 242

Quoted 5 lines
drivers/pci/of.c +1 -1
@@ -239,24 +239,61 @@ EXPORT_SYMBOL_GPL(of_get_pci_domain_nr);
   */
  void of_pci_check_probe_only(void)
  {
-     u32 val;
+     bool is_preserve_config = of_pci_bridge_preserve_resources(of_chosen);

No need for a bool here:

Quoted 7 lines
   if (of_pci_bridge_preserve_resources(of_chosen))
     pci_add_flags(PCI_PROBE_ONLY);

I think it would make more sense to add of_pci_bridge_preserve_resources() *above* of_pci_check_probe_only() since the usual order is to define functions earlier in the file than the calls.

I'll take care of it in the next patch.

Quoted 13 lines

drivers/pci/of.c EXPORT_SYMBOL_GPL(of_get_pci_domain_nr); · line 247

Quoted 5 lines
drivers/pci/of.c +5 -0
+
+     if (is_preserve_config)
+             pci_add_flags(PCI_PROBE_ONLY);
+     else
+             pci_clear_flags(PCI_PROBE_ONLY);

Not related to *this* patch, but I see that of_pci_check_probe_only() already clears PCI_PROBE_ONLY (added by f81c11af617c ("of/pci: Add of_pci_check_probe_only to parse "linux,pci-probe-only"").

I'm concerned about clearing PCI_PROBE_ONLY because some platforms set this unconditionally, and I don't think they necessarily have "linux,pci-probe-only" in DT.

Quoted 6 lines

Apparently none of them currently calls of_pci_check_probe_only() so PCI_PROBE_ONLY remains set, but clearing it here feels like a landmine waiting for somebody to move this into a unified call path.

I guess if we were to drop pci_clear_flags(), that should be a separate patch in case it breaks something.

As far as the current patch is concerned, since it is not going to do anything different compared to what already exists, I'll leave it at it for now and ponder over moving pci_clear_flags() to a different patch later.

Quoted 114 lines
Quoted 103 lines
drivers/pci/of.c +40 -9
+     pr_info("PROBE_ONLY %s\n", is_preserve_config ? "enabled" : "disabled");
+}
+EXPORT_SYMBOL_GPL(of_pci_check_probe_only);
+
+/**
+ * of_pci_bridge_preserve_resources - Return true if the boot configuration
+ *                                    needs to be preserved
+ * @node: Device tree node.
+ *
+ * This function looks for "linux,pci-probe-only" property for a given
+ * PCI controller's node and returns true if found. It will also look in the
+ * chosen node if the property is not found in the given controller's node.
+ * Having this property ensures that the kernel doesn't reconfigure the
+ * BARs and bridge windows that are already done by the platform firmware.
+ *
+ * Return: true if the property exists false otherwise.
+ */
+bool of_pci_bridge_preserve_resources(struct device_node *node)
+{
+     u32 val = 0;
       int ret;
 
-     ret = of_property_read_u32(of_chosen, "linux,pci-probe-only", &val);
+     if (!node) {
+             pr_warn("device node is NULL, trying with of_chosen\n");
+             node = of_chosen;
+     }
+
+retry:
+     ret = of_property_read_u32(node, "linux,pci-probe-only", &val);
       if (ret) {
-             if (ret == -ENODATA || ret == -EOVERFLOW)
-                     pr_warn("linux,pci-probe-only without valid value, ignoring\n");
-             return;
+             if (ret == -ENODATA || ret == -EOVERFLOW) {
+                     pr_warn("Incorrect value for linux,pci-probe-only in %pOF, ignoring\n", node);
+                     return false;
+             }
+             if (ret == -EINVAL) {
+                     if (node == of_chosen)
+                             return false;
+
+                     node = of_chosen;
+                     goto retry;
+             }
       }
 
       if (val)
-             pci_add_flags(PCI_PROBE_ONLY);
+             return true;
       else
-             pci_clear_flags(PCI_PROBE_ONLY);
-
-     pr_info("PROBE_ONLY %s\n", val ? "enabled" : "disabled");
+             return false;
  }
-EXPORT_SYMBOL_GPL(of_pci_check_probe_only);
 
  /**
   * devm_of_pci_get_host_bridge_resources() - Resource-managed parsing of PCI
drivers/pci/probe.c +23 -0
index 795534589b98..b0e0226a8da8 100644
@@ -15,6 +15,7 @@
  #include <linux/cpumask.h>
  #include <linux/aer.h>
  #include <linux/acpi.h>
+#include <linux/pci-acpi.h>
  #include <linux/hypervisor.h>
  #include <linux/irqdomain.h>
  #include <linux/pm_runtime.h>
@@ -877,6 +878,28 @@ static void pci_set_bus_msi_domain(struct pci_bus *bus)
       dev_set_msi_domain(&bus->dev, d);
  }
 
+static void pci_check_config_preserve(struct pci_host_bridge *host_bridge)
+{
+     if (ACPI_HANDLE(&host_bridge->dev)) {
+             union acpi_object *obj;
+
+             /*
+              * Evaluate the "PCI Boot Configuration" _DSM Function.  If it
+              * exists and returns 0, we must preserve any PCI resource
+              * assignments made by firmware for this host bridge.
+              */
+             obj = acpi_evaluate_dsm(ACPI_HANDLE(&host_bridge->dev), &pci_acpi_dsm_guid, 1,
+                                     DSM_PCI_PRESERVE_BOOT_CONFIG, NULL);
+             if (obj && obj->type == ACPI_TYPE_INTEGER && obj->integer.value == 0)
+                     host_bridge->preserve_config = 1;
+             ACPI_FREE(obj);
+     }
+
+     if (host_bridge->dev.parent && host_bridge->dev.parent->of_node)
+             host_bridge->preserve_config =
+                     of_pci_bridge_preserve_resources(host_bridge->dev.parent->of_node);
+}
+
  static int pci_register_host_bridge(struct pci_host_bridge *bridge)
  {
       struct device *parent = bridge->dev.parent;

drivers/pci/probe.c static int pci_register_host_bridge(struct pci_host_bridge *bridge) · line 998

Quoted 5 lines
drivers/pci/probe.c +2 -0
@@ -971,6 +994,9 @@ static int pci_register_host_bridge(struct pci_host_bridge *bridge)
       if (nr_node_ids > 1 && pcibus_to_node(bus) == NUMA_NO_NODE)
               dev_warn(&bus->dev, "Unknown NUMA node; performance will be reduced\n");
 
+     /* Check if the boot configuration by FW needs to be preserved */
+     pci_check_config_preserve(bridge);

I really have an allergic reaction to functions named "..._check_..." We can tell that the function has something to do with preserving configuration, and it's void so obviously the side effects are the important thing, but there's no clue in the caller about what the side effects are.

Quoted 6 lines

I'd prefer something like:

   bridge->preserve_config = pci_must_preserve_config(bridge);

where pci_must_preserve_config() has no side effects, returns bool, and the action is at the caller.

I'll take care of it in the next patch.

Quoted 31 lines
Quoted 31 lines
drivers/pci/probe.c +9 -11
       /* Coalesce contiguous windows */
       resource_list_for_each_entry_safe(window, n, &resources) {
               if (list_is_last(&window->node, &resources))
@@ -3080,20 +3106,18 @@ int pci_host_probe(struct pci_host_bridge *bridge)
 
       bus = bridge->bus;
 
+     /* If we must preserve the resource configuration, claim now */
+     if (bridge->preserve_config)
+             pci_bus_claim_resources(bus);
+
       /*
-      * We insert PCI resources into the iomem_resource and
-      * ioport_resource trees in either pci_bus_claim_resources()
-      * or pci_bus_assign_resources().
+      * Assign whatever was left unassigned. If we didn't claim above,
+      * this will reassign everything.
        */
-     if (pci_has_flag(PCI_PROBE_ONLY)) {
-             pci_bus_claim_resources(bus);
-     } else {
-             pci_bus_size_bridges(bus);
-             pci_bus_assign_resources(bus);
+     pci_assign_unassigned_root_bus_resources(bus);
 
-             list_for_each_entry(child, &bus->children, node)
-                     pcie_bus_configure_settings(child);
-     }
+     list_for_each_entry(child, &bus->children, node)
+             pcie_bus_configure_settings(child);
 
Quoted 13 lines

include/linux/of_pci.h struct device_node *of_pci_find_child_device(struct device_node *parent, · line 16

Quoted 11 lines
drivers/pci/probe.c +0 -0
       pci_bus_add_devices(bus);
       return 0;
include/linux/of_pci.h +1 -0
index 29658c0ee71f..3f3909a5d55d 100644
@@ -13,6 +13,7 @@ struct device_node *of_pci_find_child_device(struct device_node *parent,
                                            unsigned int devfn);
  int of_pci_get_devfn(struct device_node *np);
  void of_pci_check_probe_only(void);
+bool of_pci_bridge_preserve_resources(struct device_node *node);

This looks like it should be in drivers/pci/pci.h since it's not used outside drivers/pci/.

Agree. I'll take care of it in the next patch.

Quoted 18 lines
Quoted 18 lines
include/linux/of_pci.h +5 -1
  #else
  static inline struct device_node *of_pci_find_child_device(struct device_node *parent,
                                            unsigned int devfn)
@@ -26,6 +27,11 @@ static inline int of_pci_get_devfn(struct device_node *np)
  }
 
  static inline void of_pci_check_probe_only(void) { }
+
+static inline bool of_pci_bridge_preserve_resources(struct device_node *node)
+{
+     return false;
+}
  #endif
 
  #if IS_ENABLED(CONFIG_OF_IRQ)
--

2.25.1

#21 lore raw

PATCH v6 1/1 PCI: Add support for preserving boot configuration

In reply to Vidya Sagar

Add support for preserving the boot configuration done by the platform firmware per host bridge basis, based on the presence of 'linux,pci-probe-only' property in the respective PCI host bridge device-tree node. It also unifies the ACPI and DT based boot flows in this regard.

---
V5:
* Addressed Rob's review comments

V4:
* Addressed Bjorn's review comments

V3:
* Unified ACPI and DT flows as part of addressing Bjorn's review comments

V2:
* Addressed issues reported by kernel test robot <[email protected]>

 drivers/acpi/pci_root.c                  | 12 ------
 drivers/pci/controller/pci-host-common.c |  4 --
 drivers/pci/of.c                         | 54 +++++++++++++++++++-----
 drivers/pci/pci.h                        |  7 +++
 drivers/pci/probe.c                      | 48 ++++++++++++++++-----
 5 files changed, 88 insertions(+), 37 deletions(-)
drivers/acpi/pci_root.c +0 -12
index 58b89b8d950e..ddc2b3e89111 100644
@@ -1008,7 +1008,6 @@ struct pci_bus *acpi_pci_root_create(struct acpi_pci_root *root,
 	int node = acpi_get_node(device->handle);
 	struct pci_bus *bus;
 	struct pci_host_bridge *host_bridge;
-	union acpi_object *obj;
 
 	info->root = root;
 	info->bridge = device;
@@ -1050,17 +1049,6 @@ struct pci_bus *acpi_pci_root_create(struct acpi_pci_root *root,
 	if (!(root->osc_ext_control_set & OSC_CXL_ERROR_REPORTING_CONTROL))
 		host_bridge->native_cxl_error = 0;
 
-	/*
-	 * Evaluate the "PCI Boot Configuration" _DSM Function.  If it
-	 * exists and returns 0, we must preserve any PCI resource
-	 * assignments made by firmware for this host bridge.
-	 */
-	obj = acpi_evaluate_dsm_typed(ACPI_HANDLE(bus->bridge), &pci_acpi_dsm_guid, 1,
-				      DSM_PCI_PRESERVE_BOOT_CONFIG, NULL, ACPI_TYPE_INTEGER);
-	if (obj && obj->integer.value == 0)
-		host_bridge->preserve_config = 1;
-	ACPI_FREE(obj);
-
 	acpi_dev_power_up_children_with_adr(device);
 
 	pci_scan_child_bus(bus);
drivers/pci/controller/pci-host-common.c +0 -4
index 45b71806182d..c96b2de163b5 100644
@@ -73,10 +73,6 @@ int pci_host_common_probe(struct platform_device *pdev)
 	if (IS_ERR(cfg))
 		return PTR_ERR(cfg);
 
-	/* Do not reassign resources if probe only */
-	if (!pci_has_flag(PCI_PROBE_ONLY))
-		pci_add_flags(PCI_REASSIGN_ALL_BUS);
-
 	bridge->sysdata = cfg;
 	bridge->ops = (struct pci_ops *)&ops->pci_ops;
 	bridge->msi_domain = true;
drivers/pci/of.c +44 -10

71 lines not on this page · raw message

drivers/pci/pci.h +7 -0
index 17fed1846847..1da54ab18534 100644
@@ -650,6 +650,7 @@ int of_pci_get_max_link_speed(struct device_node *node);
 u32 of_pci_get_slot_power_limit(struct device_node *node,
 				u8 *slot_power_limit_value,
 				u8 *slot_power_limit_scale);
+bool of_pci_bridge_preserve_resources(struct device_node *node);
 int pci_set_of_node(struct pci_dev *dev);
 void pci_release_of_node(struct pci_dev *dev);
 void pci_set_bus_of_node(struct pci_bus *bus);
@@ -688,6 +689,12 @@ of_pci_get_slot_power_limit(struct device_node *node,
 	return 0;
 }
 
+static inline bool
+of_pci_bridge_preserve_resources(struct device_node *node)
+{
+	return false;
+}
+
 static inline int pci_set_of_node(struct pci_dev *dev) { return 0; }
 static inline void pci_release_of_node(struct pci_dev *dev) { }
 static inline void pci_set_bus_of_node(struct pci_bus *bus) { }
drivers/pci/probe.c +37 -11

75 lines not on this page · raw message

Signature · 1 line
2.25.1
#22 lore raw

Re: [PATCH V6] PCI: Add support for preserving boot configuration

Quoted 5 lines · Vidya Sagar

Add support for preserving the boot configuration done by the platform firmware per host bridge basis, based on the presence of 'linux,pci-probe-only' property in the respective PCI host bridge device-tree node. It also unifies the ACPI and DT based boot flows in this regard.

This looks great; my only comments are:

- It might make sense to move the _DSM evaluation to pci-acpi.c instead of probe.c.

- There are a few different things going on (move _DSM, look at DT host bridge in addition to of_chosen, integrate ACPI and DT flows, tweak usage of PCI_PROBE_ONLY vs bridge->preserve_config, use pci_assign_unassigned_root_bus_resources() when we didn't before), and I think they could be split into multiple patches to make it easier to read.

I took a stab at splitting this, but didn't quite complete it. I'll post that as v7-incomplete, and if you compare with v6, you'll see what's missing (the PCI_PROBE_ONLY and assign resources stuff -- there's some subtlety there that wasn't obvious to me).

Bjorn

#23 lore raw

cover PCI: Add support for preserving boot configuration

In reply to Vidya Sagar

From: Bjorn Helgaas <[email protected]>

This is Vidya's work; I started splitting it into separate patches.

This is based on v6
(https://lore.kernel.org/r/[email protected]) but is missing a few pieces, so it's not complete yet. I just wanted to show the idea of what I had in mind.

The missing pieces are:

- Set PCI_REASSIGN_ALL_BUS if !PCI_PROBE_ONLY

- Use bridge->preserve_config instead of PCI_PROBE_ONLY

- Change to use pci_assign_unassigned_root_bus_resources() instead of pci_bus_size_bridge() and pci_bus_assign_resources()

Bjorn Helgaas (3):
PCI: Move PRESERVE_BOOT_CONFIG _DSM evaluation to
pci_register_host_bridge()
PCI: of: Add of_pci_preserve_config() for per-host bridge support
PCI: Unify ACPI and DT 'preserve config' support

 drivers/acpi/pci_root.c | 12 ---------
 drivers/pci/of.c        | 60 ++++++++++++++++++++++++++++++++---------
 drivers/pci/pci-acpi.c  | 22 +++++++++++++++
 drivers/pci/pci.h       | 12 +++++++++
 drivers/pci/probe.c     | 14 ++++++++++
 5 files changed, 95 insertions(+), 25 deletions(-)
Signature · 1 line
2.34.1
#24 lore raw

PATCH 1/3 PCI: Move PRESERVE_BOOT_CONFIG _DSM evaluation to pci_register_host_bridge()

From: Bjorn Helgaas <[email protected]>

Move the PRESERVE_BOOT_CONFIG _DSM evaluation from acpi_pci_root_create() to pci_register_host_bridge().

This will help unify the ACPI _DSM path and the DT-based "linux,pci-probe-only" paths.

This should be safe because it happens earlier than it used to:

acpi_pci_root_create
pci_create_root_bus
pci_register_host_bridge
+ bridge->preserve_config = pci_preserve_config(bridge)
pci_acpi_preserve_config
+ acpi_evaluate_dsm_typed(DSM_PCI_PRESERVE_BOOT_CONFIG)
- acpi_evaluate_dsm_typed(DSM_PCI_PRESERVE_BOOT_CONFIG)

No functional change intended.

---
 drivers/acpi/pci_root.c | 12 ------------
 drivers/pci/pci-acpi.c  | 22 ++++++++++++++++++++++
 drivers/pci/pci.h       |  5 +++++
 drivers/pci/probe.c     | 11 +++++++++++
 4 files changed, 38 insertions(+), 12 deletions(-)
drivers/acpi/pci_root.c +0 -12
index 58b89b8d950e..ddc2b3e89111 100644
@@ -1008,7 +1008,6 @@ struct pci_bus *acpi_pci_root_create(struct acpi_pci_root *root,
 	int node = acpi_get_node(device->handle);
 	struct pci_bus *bus;
 	struct pci_host_bridge *host_bridge;
-	union acpi_object *obj;
 
 	info->root = root;
 	info->bridge = device;
@@ -1050,17 +1049,6 @@ struct pci_bus *acpi_pci_root_create(struct acpi_pci_root *root,
 	if (!(root->osc_ext_control_set & OSC_CXL_ERROR_REPORTING_CONTROL))
 		host_bridge->native_cxl_error = 0;
 
-	/*
-	 * Evaluate the "PCI Boot Configuration" _DSM Function.  If it
-	 * exists and returns 0, we must preserve any PCI resource
-	 * assignments made by firmware for this host bridge.
-	 */
-	obj = acpi_evaluate_dsm_typed(ACPI_HANDLE(bus->bridge), &pci_acpi_dsm_guid, 1,
-				      DSM_PCI_PRESERVE_BOOT_CONFIG, NULL, ACPI_TYPE_INTEGER);
-	if (obj && obj->integer.value == 0)
-		host_bridge->preserve_config = 1;
-	ACPI_FREE(obj);
-
 	acpi_dev_power_up_children_with_adr(device);
 
 	pci_scan_child_bus(bus);
drivers/pci/pci-acpi.c +22 -0
index 004575091596..9cc447da9475 100644
@@ -119,6 +119,28 @@ phys_addr_t acpi_pci_root_get_mcfg_addr(acpi_handle handle)
 	return (phys_addr_t)mcfg_addr;
 }
 
+bool pci_acpi_preserve_config(struct pci_host_bridge *host_bridge)
+{
+	if (ACPI_HANDLE(&host_bridge->dev)) {
+		union acpi_object *obj;
+
+		/*
+		 * Evaluate the "PCI Boot Configuration" _DSM Function.  If it
+		 * exists and returns 0, we must preserve any PCI resource
+		 * assignments made by firmware for this host bridge.
+		 */
+		obj = acpi_evaluate_dsm_typed(ACPI_HANDLE(&host_bridge->dev),
+					      &pci_acpi_dsm_guid,
+					      1, DSM_PCI_PRESERVE_BOOT_CONFIG,
+					      NULL, ACPI_TYPE_INTEGER);
+		if (obj && obj->integer.value == 0)
+			return true;
+		ACPI_FREE(obj);
+	}
+
+	return false;
+}
+
 /* _HPX PCI Setting Record (Type 0); same as _HPP */
 struct hpx_type0 {
 	u32 revision;		/* Not present in _HPP */
drivers/pci/pci.h +5 -0
index 17fed1846847..180d3907b543 100644
@@ -734,6 +734,7 @@ static inline void pci_restore_aer_state(struct pci_dev *dev) { }
 #endif
 
 #ifdef CONFIG_ACPI
+bool pci_acpi_preserve_config(struct pci_host_bridge *bridge);
 int pci_acpi_program_hp_params(struct pci_dev *dev);
 extern const struct attribute_group pci_dev_acpi_attr_group;
 void pci_set_acpi_fwnode(struct pci_dev *dev);
@@ -747,6 +748,10 @@ int acpi_pci_wakeup(struct pci_dev *dev, bool enable);
 bool acpi_pci_need_resume(struct pci_dev *dev);
 pci_power_t acpi_pci_choose_state(struct pci_dev *pdev);
 #else
+static inline bool pci_acpi_preserve_config(struct pci_host_bridge *bridge)
+{
+	return false;
+}
 static inline int pci_dev_acpi_reset(struct pci_dev *dev, bool probe)
 {
 	return -ENOTTY;
drivers/pci/probe.c +11 -0
index 1325fbae2f28..ee086d029450 100644
@@ -889,6 +889,14 @@ static void pci_set_bus_msi_domain(struct pci_bus *bus)
 	dev_set_msi_domain(&bus->dev, d);
 }
 
+static bool pci_preserve_config(struct pci_host_bridge *host_bridge)
+{
+	if (pci_acpi_preserve_config(host_bridge))
+		return true;
+
+	return false;
+}
+
 static int pci_register_host_bridge(struct pci_host_bridge *bridge)
 {
 	struct device *parent = bridge->dev.parent;
@@ -983,6 +991,9 @@ static int pci_register_host_bridge(struct pci_host_bridge *bridge)
 	if (nr_node_ids > 1 && pcibus_to_node(bus) == NUMA_NO_NODE)
 		dev_warn(&bus->dev, "Unknown NUMA node; performance will be reduced\n");
 
+	/* Check if the boot configuration by FW needs to be preserved */
+	bridge->preserve_config = pci_preserve_config(bridge);
+
 	/* Coalesce contiguous windows */
 	resource_list_for_each_entry_safe(window, n, &resources) {
 		if (list_is_last(&window->node, &resources))
Signature · 1 line
2.34.1
#25 lore raw

Re: [PATCH v7-incomplete 1/3] PCI: Move PRESERVE_BOOT_CONFIG _DSM evaluation to pci_register_host_bridge()

Sun, Apr 21, 2024 at 02:09:12PM -0500, Bjorn Helgaas kirjoitti:

Quoted 7 lines

From: Bjorn Helgaas <[email protected]>

Move the PRESERVE_BOOT_CONFIG _DSM evaluation from acpi_pci_root_create() to pci_register_host_bridge().

This will help unify the ACPI _DSM path and the DT-based "linux,pci-probe-only" paths.

...

drivers/pci/pci-acpi.c phys_addr_t acpi_pci_root_get_mcfg_addr(acpi_handle handle) · line 124

Quoted 3 lines
drivers/pci/pci-acpi.c +3 -0
+bool pci_acpi_preserve_config(struct pci_host_bridge *host_bridge)
+{
+	if (ACPI_HANDLE(&host_bridge->dev)) {

Wouldn't the below looks nicer if you invert the conditional?

	handle = ACPI_HANDLE(...);

	if (!handle)
		return false;

	...
Quoted 18 lines
drivers/pci/pci-acpi.c +18 -0
+		union acpi_object *obj;
+
+		/*
+		 * Evaluate the "PCI Boot Configuration" _DSM Function.  If it
+		 * exists and returns 0, we must preserve any PCI resource
+		 * assignments made by firmware for this host bridge.
+		 */
+		obj = acpi_evaluate_dsm_typed(ACPI_HANDLE(&host_bridge->dev),
+					      &pci_acpi_dsm_guid,
+					      1, DSM_PCI_PRESERVE_BOOT_CONFIG,
+					      NULL, ACPI_TYPE_INTEGER);
+		if (obj && obj->integer.value == 0)
+			return true;
+		ACPI_FREE(obj);
+	}
+
+	return false;
+}
Signature · 2 lines
With Best Regards,
Andy Shevchenko
#26 lore raw

PATCH 2/3 PCI: of: Add of_pci_preserve_config() for per-host bridge support

In reply to Bjorn Helgaas

From: Bjorn Helgaas <[email protected]>

Add of_pci_preserve_config() to look for the "linux,pci-probe-only" property under a specified node. If it's not found there, look under "of_chosen" in addition.

If the caller didn't specify a node, look under "of_chosen".

With a future patch, this will support "linux,pci-probe-only" on a per host bridge basis based on the presence of the property in the respective PCI host bridge DT node.

Implement of_pci_check_probe_only() using of_pci_preserve_config().

---
 drivers/pci/of.c  | 60 +++++++++++++++++++++++++++++++++++++----------
 drivers/pci/pci.h |  7 ++++++
 2 files changed, 54 insertions(+), 13 deletions(-)
drivers/pci/of.c +47 -13
index 51e3dd0ea5ab..d21c0bed36f3 100644
@@ -233,28 +233,62 @@ int of_get_pci_domain_nr(struct device_node *node)
 }
 EXPORT_SYMBOL_GPL(of_get_pci_domain_nr);
 
+/**
+ * of_pci_preserve_config - Return true if the boot configuration needs to
+ *                          be preserved
+ * @node: Device tree node.
+ *
+ * This function looks for "linux,pci-probe-only" property for a given
+ * PCI controller's node and returns true if found. It will also look in the
+ * chosen node if the property is not found in the given controller's node.
+ * Having this property ensures that the kernel doesn't reconfigure the
+ * BARs and bridge windows that are already done by the platform firmware.
+ *
+ * Return: true if the property exists false otherwise.
+ */
+bool of_pci_preserve_config(struct device_node *node)
+{
+	u32 val = 0;
+	int ret;
+
+	if (!node) {
+		pr_warn("device node is NULL, trying with of_chosen\n");
+		node = of_chosen;
+	}
+
+retry:
+	ret = of_property_read_u32(node, "linux,pci-probe-only", &val);
+	if (ret) {
+		if (ret == -ENODATA || ret == -EOVERFLOW) {
+			pr_warn("Incorrect value for linux,pci-probe-only in %pOF, ignoring\n",
+				node);
+			return false;
+		}
+		if (ret == -EINVAL) {
+			if (node == of_chosen)
+				return false;
+
+			node = of_chosen;
+			goto retry;
+		}
+	}
+
+	if (val)
+		return true;
+	else
+		return false;
+}
+
 /**
  * of_pci_check_probe_only - Setup probe only mode if linux,pci-probe-only
  *                           is present and valid
  */
 void of_pci_check_probe_only(void)
 {
-	u32 val;
-	int ret;
-
-	ret = of_property_read_u32(of_chosen, "linux,pci-probe-only", &val);
-	if (ret) {
-		if (ret == -ENODATA || ret == -EOVERFLOW)
-			pr_warn("linux,pci-probe-only without valid value, ignoring\n");
-		return;
-	}
-
-	if (val)
+	if (of_pci_preserve_config(of_chosen))
 		pci_add_flags(PCI_PROBE_ONLY);
 	else
 		pci_clear_flags(PCI_PROBE_ONLY);
-
-	pr_info("PROBE_ONLY %s\n", val ? "enabled" : "disabled");
 }
 EXPORT_SYMBOL_GPL(of_pci_check_probe_only);
 
drivers/pci/pci.h +7 -0
index 180d3907b543..feedbedd65df 100644
@@ -650,6 +650,7 @@ int of_pci_get_max_link_speed(struct device_node *node);
 u32 of_pci_get_slot_power_limit(struct device_node *node,
 				u8 *slot_power_limit_value,
 				u8 *slot_power_limit_scale);
+bool of_pci_preserve_config(struct device_node *node);
 int pci_set_of_node(struct pci_dev *dev);
 void pci_release_of_node(struct pci_dev *dev);
 void pci_set_bus_of_node(struct pci_bus *bus);
@@ -688,6 +689,12 @@ of_pci_get_slot_power_limit(struct device_node *node,
 	return 0;
 }
 
+static inline bool
+of_pci_preserve_config(struct device_node *node)
+{
+	return false;
+}
+
 static inline int pci_set_of_node(struct pci_dev *dev) { return 0; }
 static inline void pci_release_of_node(struct pci_dev *dev) { }
 static inline void pci_set_bus_of_node(struct pci_bus *bus) { }
Signature · 1 line
2.34.1
#27 lore raw

PATCH 3/3 PCI: Unify ACPI and DT 'preserve config' support

In reply to Bjorn Helgaas

From: Bjorn Helgaas <[email protected]>

---
 drivers/pci/probe.c | 3 +++
 1 file changed, 3 insertions(+)
drivers/pci/probe.c +3 -0
index ee086d029450..2c232c22d6af 100644
@@ -894,6 +894,9 @@ static bool pci_preserve_config(struct pci_host_bridge *host_bridge)
 	if (pci_acpi_preserve_config(host_bridge))
 		return true;
 
+	if (host_bridge->dev.parent && host_bridge->dev.parent->of_node)
+		return of_pci_preserve_config(host_bridge->dev.parent->of_node);
+
 	return false;
 }
 
Signature · 1 line
2.34.1
#28 lore raw

Re: [PATCH v7-incomplete 3/3] PCI: Unify ACPI and DT 'preserve config' support

Sun, Apr 21, 2024 at 02:09:14PM -0500, Bjorn Helgaas kirjoitti:

...

drivers/pci/probe.c static bool pci_preserve_config(struct pci_host_bridge *host_bridge) · line 898

Quoted 2 lines
drivers/pci/probe.c +2 -0
+	if (host_bridge->dev.parent && host_bridge->dev.parent->of_node)
+		return of_pci_preserve_config(host_bridge->dev.parent->of_node);

If you need fwnode or of_node from struct device, please avoid direct dereference, we have

dev_fwnode()
of_fwnode_handle()
dev_of_node()

Signature · 2 lines
With Best Regards,
Andy Shevchenko