drivers: pci: move PCI domain assignment to generic PCI code

The current logic used for PCI domain assignment in arm64 pci_bus_assign_domain_nr() is flawed in that, depending on the host controllers configuration for a platform and the respective initialization sequence, core code may end up allocating PCI domain numbers from both DT and the core code generic domain counter, which would result in PCI domain allocation aliases/errors.

This patch fixes the logic behind the PCI domain number assignment and moves the resulting code to generic PCI core code so that the same domain allocation logic is used on all platforms selecting

CONFIG_PCI_DOMAINS_GENERIC

instead of resorting to an arch specific implementation that might end up duplicating the PCI domain assignment logic wrongly.

Cc Arnd Bergmann <[email protected]>

Cc Liviu Dudau <[email protected]>

Cc Bjorn Helgaas <[email protected]>

Cc Rob Herring <[email protected]>

Cc Catalin Marinas <[email protected]>

Signed-off-by Lorenzo Pieralisi <[email protected]>

---
v1 => v2:

- Moved generic pci_bus_assign_domain_nr() code to PCI core instead of adding an OF layer API
- Updated commit log and code comments

 arch/arm64/kernel/pci.c | 22 ----------------------
 drivers/pci/pci.c       | 50 +++++++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 50 insertions(+), 22 deletions(-)

arch/arm64/kernel/pci.c +0 -22

index ce5836c..6f93c24 100644
@@ -46,25 +46,3 @@ int pcibios_add_device(struct pci_dev *dev)
 
 	return 0;
 }
-
-
-#ifdef CONFIG_PCI_DOMAINS_GENERIC
-static bool dt_domain_found = false;
-
-void pci_bus_assign_domain_nr(struct pci_bus *bus, struct device *parent)
-{
-	int domain = of_get_pci_domain_nr(parent->of_node);
-
-	if (domain >= 0) {
-		dt_domain_found = true;
-	} else if (dt_domain_found == true) {
-		dev_err(parent, "Node %s is missing \"linux,pci-domain\" property in DT\n",
-			parent->of_node->full_name);
-		return;
-	} else {
-		domain = pci_get_new_domain_nr();
-	}
-
-	bus->domain_nr = domain;
-}
-#endif

drivers/pci/pci.c +50 -0

index 625a4ac..2279414 100644
@@ -10,6 +10,8 @@
 #include <linux/kernel.h>
 #include <linux/delay.h>
 #include <linux/init.h>
+#include <linux/of.h>
+#include <linux/of_pci.h>
 #include <linux/pci.h>
 #include <linux/pm.h>
 #include <linux/slab.h>

@@ -4447,6 +4449,54 @@ int pci_get_new_domain_nr(void)
 {
 	return atomic_inc_return(&__domain_nr);
 }
+
+#ifdef CONFIG_PCI_DOMAINS_GENERIC
+
+void pci_bus_assign_domain_nr(struct pci_bus *bus, struct device *parent)
+{
+	static int use_dt_domains = -1;
+	int domain = of_get_pci_domain_nr(parent->of_node);
+
+	/*
+	 * Check DT domain and use_dt_domains values.
+	 *
+	 * If DT domain property is valid (domain >= 0) and
+	 * use_dt_domains != 0, the DT assignment is valid since this means
+	 * we have not previously allocated a domain number by using
+	 * pci_get_new_domain_nr(); we should also update use_dt_domains to
+	 * 1, to indicate that we have just assigned a domain number from
+	 * DT.
+	 *
+	 * If DT domain property value is not valid (ie domain < 0), and we
+	 * have not previously assigned a domain number from DT
+	 * (use_dt_domains != 1) we should assign a domain number by
+	 * using the:
+	 *
+	 * pci_get_new_domain_nr()
+	 *
+	 * API and update the use_dt_domains value to keep track of method we
+	 * are using to assign domain numbers (use_dt_domains = 0).
+	 *
+	 * All other combinations imply we have a platform that is trying
+	 * to mix domain numbers obtained from DT and pci_get_new_domain_nr(),
+	 * which is a recipe for domain mishandling and it is prevented by
+	 * invalidating the domain value (domain = -1) and printing a
+	 * corresponding error.
+	 */
+	if (domain >= 0 && use_dt_domains) {
+		use_dt_domains = 1;
+	} else if (domain < 0 && use_dt_domains != 1) {
+		use_dt_domains = 0;
+		domain = pci_get_new_domain_nr();
+	} else {
+		dev_err(parent, "Node %s has inconsistent \"linux,pci-domain\" property in DT\n",
+			parent->of_node->full_name);
+		domain = -1;
+	}
+
+	bus->domain_nr = domain;
+}
+#endif
 #endif
 
 /**

Signature · 1 line

2.1.2

Quoted 25 lines · Lorenzo Pieralisi

The current logic used for PCI domain assignment in arm64 pci_bus_assign_domain_nr() is flawed in that, depending on the host controllers configuration for a platform and the respective initialization sequence, core code may end up allocating PCI domain numbers from both DT and the core code generic domain counter, which would result in PCI domain allocation aliases/errors.

This patch fixes the logic behind the PCI domain number assignment and moves the resulting code to generic PCI core code so that the same domain allocation logic is used on all platforms selecting

CONFIG_PCI_DOMAINS_GENERIC

instead of resorting to an arch specific implementation that might end up duplicating the PCI domain assignment logic wrongly.

Cc Arnd Bergmann <[email protected]>

Cc Liviu Dudau <[email protected]>

Cc Bjorn Helgaas <[email protected]>

Cc Rob Herring <[email protected]>

Cc Catalin Marinas <[email protected]>

Signed-off-by Lorenzo Pieralisi <[email protected]>

---
v1 => v2:

Quoted 3 lines

- Moved generic pci_bus_assign_domain_nr() code to PCI core instead of adding an OF layer API
- Updated commit log and code comments

Is this approach ok with everyone ? I would need to have this patch queued so that I can rebase the ARM32 pcibios pci_sys_data domain removal on top of it, if everyone agrees of course.

Thanks !
Lorenzo

Quoted 107 lines

 arch/arm64/kernel/pci.c | 22 ----------------------
 drivers/pci/pci.c       | 50 +++++++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 50 insertions(+), 22 deletions(-)

arch/arm64/kernel/pci.c +0 -22

index ce5836c..6f93c24 100644
@@ -46,25 +46,3 @@ int pcibios_add_device(struct pci_dev *dev)
 
 	return 0;
 }
-
-
-#ifdef CONFIG_PCI_DOMAINS_GENERIC
-static bool dt_domain_found = false;
-
-void pci_bus_assign_domain_nr(struct pci_bus *bus, struct device *parent)
-{
-	int domain = of_get_pci_domain_nr(parent->of_node);
-
-	if (domain >= 0) {
-		dt_domain_found = true;
-	} else if (dt_domain_found == true) {
-		dev_err(parent, "Node %s is missing \"linux,pci-domain\" property in DT\n",
-			parent->of_node->full_name);
-		return;
-	} else {
-		domain = pci_get_new_domain_nr();
-	}
-
-	bus->domain_nr = domain;
-}
-#endif

drivers/pci/pci.c +50 -0

index 625a4ac..2279414 100644
@@ -10,6 +10,8 @@
 #include <linux/kernel.h>
 #include <linux/delay.h>
 #include <linux/init.h>
+#include <linux/of.h>
+#include <linux/of_pci.h>
 #include <linux/pci.h>
 #include <linux/pm.h>
 #include <linux/slab.h>

@@ -4447,6 +4449,54 @@ int pci_get_new_domain_nr(void)
 {
 	return atomic_inc_return(&__domain_nr);
 }
+
+#ifdef CONFIG_PCI_DOMAINS_GENERIC
+
+void pci_bus_assign_domain_nr(struct pci_bus *bus, struct device *parent)
+{
+	static int use_dt_domains = -1;
+	int domain = of_get_pci_domain_nr(parent->of_node);
+
+	/*
+	 * Check DT domain and use_dt_domains values.
+	 *
+	 * If DT domain property is valid (domain >= 0) and
+	 * use_dt_domains != 0, the DT assignment is valid since this means
+	 * we have not previously allocated a domain number by using
+	 * pci_get_new_domain_nr(); we should also update use_dt_domains to
+	 * 1, to indicate that we have just assigned a domain number from
+	 * DT.
+	 *
+	 * If DT domain property value is not valid (ie domain < 0), and we
+	 * have not previously assigned a domain number from DT
+	 * (use_dt_domains != 1) we should assign a domain number by
+	 * using the:
+	 *
+	 * pci_get_new_domain_nr()
+	 *
+	 * API and update the use_dt_domains value to keep track of method we
+	 * are using to assign domain numbers (use_dt_domains = 0).
+	 *
+	 * All other combinations imply we have a platform that is trying
+	 * to mix domain numbers obtained from DT and pci_get_new_domain_nr(),
+	 * which is a recipe for domain mishandling and it is prevented by
+	 * invalidating the domain value (domain = -1) and printing a
+	 * corresponding error.
+	 */
+	if (domain >= 0 && use_dt_domains) {
+		use_dt_domains = 1;
+	} else if (domain < 0 && use_dt_domains != 1) {
+		use_dt_domains = 0;
+		domain = pci_get_new_domain_nr();
+	} else {
+		dev_err(parent, "Node %s has inconsistent \"linux,pci-domain\" property in DT\n",
+			parent->of_node->full_name);
+		domain = -1;
+	}
+
+	bus->domain_nr = domain;
+}
+#endif
 #endif
 
 /**

Signature · 1 line

2.1.2

The current logic used for PCI domain assignment in arm64 pci_bus_assign_domain_nr() is flawed in that, depending on the host controllers configuration for a platform and the respective initialization sequence, core code may end up allocating PCI domain numbers from both DT and the core code generic domain counter, which would result in PCI domain allocation aliases/errors.

This patch fixes the logic behind the PCI domain number assignment and moves the resulting code to generic PCI core code so that the same domain allocation logic is used on all platforms selecting

CONFIG_PCI_DOMAINS_GENERIC

instead of resorting to an arch specific implementation that might end up duplicating the PCI domain assignment logic wrongly.

Cc Arnd Bergmann <[email protected]>

Cc Liviu Dudau <[email protected]>

Cc Bjorn Helgaas <[email protected]>

Cc Rob Herring <[email protected]>

Cc Catalin Marinas <[email protected]>

Signed-off-by Lorenzo Pieralisi <[email protected]>

drivers/pci/pci.c +39 -0

+
+#ifdef CONFIG_PCI_DOMAINS_GENERIC
+
+void pci_bus_assign_domain_nr(struct pci_bus *bus, struct device *parent)
+{
+	static int use_dt_domains = -1;
+	int domain = of_get_pci_domain_nr(parent->of_node);
+
+	/*
+	 * Check DT domain and use_dt_domains values.
+	 *
+	 * If DT domain property is valid (domain >= 0) and
+	 * use_dt_domains != 0, the DT assignment is valid since this means
+	 * we have not previously allocated a domain number by using
+	 * pci_get_new_domain_nr(); we should also update use_dt_domains to
+	 * 1, to indicate that we have just assigned a domain number from
+	 * DT.
+	 *
+	 * If DT domain property value is not valid (ie domain < 0), and we
+	 * have not previously assigned a domain number from DT
+	 * (use_dt_domains != 1) we should assign a domain number by
+	 * using the:
+	 *
+	 * pci_get_new_domain_nr()
+	 *
+	 * API and update the use_dt_domains value to keep track of method we
+	 * are using to assign domain numbers (use_dt_domains = 0).
+	 *
+	 * All other combinations imply we have a platform that is trying
+	 * to mix domain numbers obtained from DT and pci_get_new_domain_nr(),
+	 * which is a recipe for domain mishandling and it is prevented by
+	 * invalidating the domain value (domain = -1) and printing a
+	 * corresponding error.
+	 */
+	if (domain >= 0 && use_dt_domains) {
+		use_dt_domains = 1;
+	} else if (domain < 0 && use_dt_domains != 1) {
+		use_dt_domains = 0;
+		domain = pci_get_new_domain_nr();

Quoted 16 lines · On Monday 10 November 2014 16:41:46 Lorenzo Pieralisi wrote:

The current logic used for PCI domain assignment in arm64 pci_bus_assign_domain_nr() is flawed in that, depending on the host controllers configuration for a platform and the respective initialization sequence, core code may end up allocating PCI domain numbers from both DT and the core code generic domain counter, which would result in PCI domain allocation aliases/errors.

This patch fixes the logic behind the PCI domain number assignment and moves the resulting code to generic PCI core code so that the same domain allocation logic is used on all platforms selecting

CONFIG_PCI_DOMAINS_GENERIC

instead of resorting to an arch specific implementation that might end up duplicating the PCI domain assignment logic wrongly.

Quoted 6 lines

Cc Arnd Bergmann <[email protected]>

Cc Liviu Dudau <[email protected]>

Cc Bjorn Helgaas <[email protected]>

Cc Rob Herring <[email protected]>

Cc Catalin Marinas <[email protected]>

Signed-off-by Lorenzo Pieralisi <[email protected]>

The general approach seems good to me,

Acked-by Arnd Bergmann <[email protected]>

I would suggest one simplification:

Quoted 46 lines

drivers/pci/pci.c +46 -0

+
+#ifdef CONFIG_PCI_DOMAINS_GENERIC
+
+void pci_bus_assign_domain_nr(struct pci_bus *bus, struct device *parent)
+{
+	static int use_dt_domains = -1;
+	int domain = of_get_pci_domain_nr(parent->of_node);
+
+	/*
+	 * Check DT domain and use_dt_domains values.
+	 *
+	 * If DT domain property is valid (domain >= 0) and
+	 * use_dt_domains != 0, the DT assignment is valid since this means
+	 * we have not previously allocated a domain number by using
+	 * pci_get_new_domain_nr(); we should also update use_dt_domains to
+	 * 1, to indicate that we have just assigned a domain number from
+	 * DT.
+	 *
+	 * If DT domain property value is not valid (ie domain < 0), and we
+	 * have not previously assigned a domain number from DT
+	 * (use_dt_domains != 1) we should assign a domain number by
+	 * using the:
+	 *
+	 * pci_get_new_domain_nr()
+	 *
+	 * API and update the use_dt_domains value to keep track of method we
+	 * are using to assign domain numbers (use_dt_domains = 0).
+	 *
+	 * All other combinations imply we have a platform that is trying
+	 * to mix domain numbers obtained from DT and pci_get_new_domain_nr(),
+	 * which is a recipe for domain mishandling and it is prevented by
+	 * invalidating the domain value (domain = -1) and printing a
+	 * corresponding error.
+	 */
+	if (domain >= 0 && use_dt_domains) {
+		use_dt_domains = 1;
+	} else if (domain < 0 && use_dt_domains != 1) {
+		use_dt_domains = 0;
+		domain = pci_get_new_domain_nr();
+	} else {
+		dev_err(parent, "Node %s has inconsistent \"linux,pci-domain\" property in DT\n",
+			parent->of_node->full_name);
+		domain = -1;
+	}
+
+	bus->domain_nr = domain;

drivers/pci/pci.c int pci_get_new_domain_nr(void) · line 4500

Quoted 4 lines

drivers/pci/pci.c +2 -0

+}
+#endif
 #endif

Since this is now in the file in which it gets called, you can mark the function itself as 'static' and remove the extern declaration and inline wrapper from the header file. You can also avoid the #ifdef by doing

Quoted 4 lines

Since this is now in the file in which it gets called, you can mark the function itself as 'static' and remove the extern declaration and inline wrapper from the header file. You can also avoid the #ifdef by doing

It is not, it is in driver/pci/pci.c, it is called in probe.c.

Maybe I can move the function to probe.c, but this would leave the domain handling in two separate files.

I can't remove the #ifdeffery in that domain_nr in pci_bus is #ifdeffed too, unless I remove that #ifdef and I compile it in all the time.

The current logic used for PCI domain assignment in arm64 pci_bus_assign_domain_nr() is flawed in that, depending on the host controllers configuration for a platform and the respective initialization sequence, core code may end up allocating PCI domain numbers from both DT and the core code generic domain counter, which would result in PCI domain allocation aliases/errors.

This patch fixes the logic behind the PCI domain number assignment and moves the resulting code to generic PCI core code so that the same domain allocation logic is used on all platforms selecting

CONFIG_PCI_DOMAINS_GENERIC

instead of resorting to an arch specific implementation that might end up duplicating the PCI domain assignment logic wrongly.

Cc Arnd Bergmann <[email protected]>

Cc Liviu Dudau <[email protected]>

Cc Bjorn Helgaas <[email protected]>

Cc Rob Herring <[email protected]>

Cc Catalin Marinas <[email protected]>

Signed-off-by Lorenzo Pieralisi <[email protected]>

---
v1 => v2:

- Moved generic pci_bus_assign_domain_nr() code to PCI core instead of adding an OF layer API
- Updated commit log and code comments

 arch/arm64/kernel/pci.c | 22 ----------------------
 drivers/pci/pci.c       | 50 +++++++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 50 insertions(+), 22 deletions(-)

arch/arm64/kernel/pci.c +0 -22

index ce5836c..6f93c24 100644
@@ -46,25 +46,3 @@ int pcibios_add_device(struct pci_dev *dev)
 
 	return 0;
 }
-
-
-#ifdef CONFIG_PCI_DOMAINS_GENERIC
-static bool dt_domain_found = false;
-
-void pci_bus_assign_domain_nr(struct pci_bus *bus, struct device *parent)
-{
-	int domain = of_get_pci_domain_nr(parent->of_node);
-
-	if (domain >= 0) {
-		dt_domain_found = true;
-	} else if (dt_domain_found == true) {
-		dev_err(parent, "Node %s is missing \"linux,pci-domain\" property in DT\n",
-			parent->of_node->full_name);
-		return;
-	} else {
-		domain = pci_get_new_domain_nr();
-	}
-
-	bus->domain_nr = domain;
-}
-#endif

drivers/pci/pci.c +50 -0

index 625a4ac..2279414 100644
@@ -10,6 +10,8 @@
 #include <linux/kernel.h>
 #include <linux/delay.h>
 #include <linux/init.h>
+#include <linux/of.h>
+#include <linux/of_pci.h>
 #include <linux/pci.h>
 #include <linux/pm.h>
 #include <linux/slab.h>

@@ -4447,6 +4449,54 @@ int pci_get_new_domain_nr(void)
 {
 	return atomic_inc_return(&__domain_nr);
 }
+
+#ifdef CONFIG_PCI_DOMAINS_GENERIC
+
+void pci_bus_assign_domain_nr(struct pci_bus *bus, struct device *parent)
+{
+	static int use_dt_domains = -1;
+	int domain = of_get_pci_domain_nr(parent->of_node);
+
+	/*
+	 * Check DT domain and use_dt_domains values.
+	 *
+	 * If DT domain property is valid (domain >= 0) and
+	 * use_dt_domains != 0, the DT assignment is valid since this means
+	 * we have not previously allocated a domain number by using
+	 * pci_get_new_domain_nr(); we should also update use_dt_domains to
+	 * 1, to indicate that we have just assigned a domain number from
+	 * DT.
+	 *
+	 * If DT domain property value is not valid (ie domain < 0), and we
+	 * have not previously assigned a domain number from DT
+	 * (use_dt_domains != 1) we should assign a domain number by
+	 * using the:
+	 *
+	 * pci_get_new_domain_nr()
+	 *
+	 * API and update the use_dt_domains value to keep track of method we
+	 * are using to assign domain numbers (use_dt_domains = 0).
+	 *
+	 * All other combinations imply we have a platform that is trying
+	 * to mix domain numbers obtained from DT and pci_get_new_domain_nr(),
+	 * which is a recipe for domain mishandling and it is prevented by
+	 * invalidating the domain value (domain = -1) and printing a
+	 * corresponding error.
+	 */
+	if (domain >= 0 && use_dt_domains) {
+		use_dt_domains = 1;
+	} else if (domain < 0 && use_dt_domains != 1) {
+		use_dt_domains = 0;
+		domain = pci_get_new_domain_nr();
+	} else {
+		dev_err(parent, "Node %s has inconsistent \"linux,pci-domain\" property in DT\n",
+			parent->of_node->full_name);
+		domain = -1;
+	}
+
+	bus->domain_nr = domain;
+}
+#endif
 #endif
 
 /**

Hi Lorenzo,
You should send this to Bjorn instead of cc. Other, why put the OF related
function in PCI core. Why not move it in drivers/of/of_pci.c ?

Quoted 132 lines

The current logic used for PCI domain assignment in arm64 pci_bus_assign_domain_nr() is flawed in that, depending on the host controllers configuration for a platform and the respective initialization sequence, core code may end up allocating PCI domain numbers from both DT and the core code generic domain counter, which would result in PCI domain allocation aliases/errors.

This patch fixes the logic behind the PCI domain number assignment and moves the resulting code to generic PCI core code so that the same domain allocation logic is used on all platforms selecting

CONFIG_PCI_DOMAINS_GENERIC

instead of resorting to an arch specific implementation that might end up duplicating the PCI domain assignment logic wrongly.

Cc Arnd Bergmann <[email protected]>

Cc Liviu Dudau <[email protected]>

Cc Bjorn Helgaas <[email protected]>

Cc Rob Herring <[email protected]>

Cc Catalin Marinas <[email protected]>

Signed-off-by Lorenzo Pieralisi <[email protected]>

---
v1 => v2:

- Moved generic pci_bus_assign_domain_nr() code to PCI core instead of adding an OF layer API
- Updated commit log and code comments

 arch/arm64/kernel/pci.c | 22 ----------------------
 drivers/pci/pci.c       | 50 +++++++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 50 insertions(+), 22 deletions(-)

arch/arm64/kernel/pci.c +0 -22

index ce5836c..6f93c24 100644
@@ -46,25 +46,3 @@ int pcibios_add_device(struct pci_dev *dev)
 
 	return 0;
 }
-
-
-#ifdef CONFIG_PCI_DOMAINS_GENERIC
-static bool dt_domain_found = false;
-
-void pci_bus_assign_domain_nr(struct pci_bus *bus, struct device *parent)
-{
-	int domain = of_get_pci_domain_nr(parent->of_node);
-
-	if (domain >= 0) {
-		dt_domain_found = true;
-	} else if (dt_domain_found == true) {
-		dev_err(parent, "Node %s is missing \"linux,pci-domain\" property in DT\n",
-			parent->of_node->full_name);
-		return;
-	} else {
-		domain = pci_get_new_domain_nr();
-	}
-
-	bus->domain_nr = domain;
-}
-#endif

drivers/pci/pci.c +50 -0

index 625a4ac..2279414 100644
@@ -10,6 +10,8 @@
 #include <linux/kernel.h>
 #include <linux/delay.h>
 #include <linux/init.h>
+#include <linux/of.h>
+#include <linux/of_pci.h>
 #include <linux/pci.h>
 #include <linux/pm.h>
 #include <linux/slab.h>

@@ -4447,6 +4449,54 @@ int pci_get_new_domain_nr(void)
 {
 	return atomic_inc_return(&__domain_nr);
 }
+
+#ifdef CONFIG_PCI_DOMAINS_GENERIC
+
+void pci_bus_assign_domain_nr(struct pci_bus *bus, struct device *parent)
+{
+	static int use_dt_domains = -1;
+	int domain = of_get_pci_domain_nr(parent->of_node);
+
+	/*
+	 * Check DT domain and use_dt_domains values.
+	 *
+	 * If DT domain property is valid (domain >= 0) and
+	 * use_dt_domains != 0, the DT assignment is valid since this means
+	 * we have not previously allocated a domain number by using
+	 * pci_get_new_domain_nr(); we should also update use_dt_domains to
+	 * 1, to indicate that we have just assigned a domain number from
+	 * DT.
+	 *
+	 * If DT domain property value is not valid (ie domain < 0), and we
+	 * have not previously assigned a domain number from DT
+	 * (use_dt_domains != 1) we should assign a domain number by
+	 * using the:
+	 *
+	 * pci_get_new_domain_nr()
+	 *
+	 * API and update the use_dt_domains value to keep track of method we
+	 * are using to assign domain numbers (use_dt_domains = 0).
+	 *
+	 * All other combinations imply we have a platform that is trying
+	 * to mix domain numbers obtained from DT and pci_get_new_domain_nr(),
+	 * which is a recipe for domain mishandling and it is prevented by
+	 * invalidating the domain value (domain = -1) and printing a
+	 * corresponding error.
+	 */
+	if (domain >= 0 && use_dt_domains) {
+		use_dt_domains = 1;
+	} else if (domain < 0 && use_dt_domains != 1) {
+		use_dt_domains = 0;
+		domain = pci_get_new_domain_nr();
+	} else {
+		dev_err(parent, "Node %s has inconsistent \"linux,pci-domain\" property in DT\n",
+			parent->of_node->full_name);
+		domain = -1;
+	}
+
+	bus->domain_nr = domain;
+}
+#endif
 #endif
 
 /**

Signature · 2 lines

Thanks!
Yijing

Quoted 3 lines · Yijing Wang

Hi Lorenzo,
You should send this to Bjorn instead of cc. Other, why put the OF related
function in PCI core. Why not move it in drivers/of/of_pci.c ?

I did, you missed v1, and the problem is that with ACPI forthcoming we do not want to have domain assignment scattered in different places, but part of core code (ie a single function that prevents mixed initialization from DT/counter and so on).

Bjorn, are you fine with this patch ? Do you want me to resend it as part of the ARM 32 PCI domain refactoring [1] ? [1] depends on this patch going in first.

Thanks,
Lorenzo

[1] http://www.spinics.net/lists/arm-kernel/msg375423.html

Quoted 131 lines · Lorenzo Pieralisi

Quoted 130 lines

The current logic used for PCI domain assignment in arm64 pci_bus_assign_domain_nr() is flawed in that, depending on the host controllers configuration for a platform and the respective initialization sequence, core code may end up allocating PCI domain numbers from both DT and the core code generic domain counter, which would result in PCI domain allocation aliases/errors.

This patch fixes the logic behind the PCI domain number assignment and moves the resulting code to generic PCI core code so that the same domain allocation logic is used on all platforms selecting

CONFIG_PCI_DOMAINS_GENERIC

instead of resorting to an arch specific implementation that might end up duplicating the PCI domain assignment logic wrongly.

Cc Arnd Bergmann <[email protected]>

Cc Liviu Dudau <[email protected]>

Cc Bjorn Helgaas <[email protected]>

Cc Rob Herring <[email protected]>

Cc Catalin Marinas <[email protected]>

Signed-off-by Lorenzo Pieralisi <[email protected]>

---
v1 => v2:

- Moved generic pci_bus_assign_domain_nr() code to PCI core instead of adding an OF layer API
- Updated commit log and code comments

 arch/arm64/kernel/pci.c | 22 ----------------------
 drivers/pci/pci.c       | 50 +++++++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 50 insertions(+), 22 deletions(-)

arch/arm64/kernel/pci.c +0 -22

index ce5836c..6f93c24 100644
@@ -46,25 +46,3 @@ int pcibios_add_device(struct pci_dev *dev)
 
 	return 0;
 }
-
-
-#ifdef CONFIG_PCI_DOMAINS_GENERIC
-static bool dt_domain_found = false;
-
-void pci_bus_assign_domain_nr(struct pci_bus *bus, struct device *parent)
-{
-	int domain = of_get_pci_domain_nr(parent->of_node);
-
-	if (domain >= 0) {
-		dt_domain_found = true;
-	} else if (dt_domain_found == true) {
-		dev_err(parent, "Node %s is missing \"linux,pci-domain\" property in DT\n",
-			parent->of_node->full_name);
-		return;
-	} else {
-		domain = pci_get_new_domain_nr();
-	}
-
-	bus->domain_nr = domain;
-}
-#endif

drivers/pci/pci.c +50 -0

index 625a4ac..2279414 100644
@@ -10,6 +10,8 @@
 #include <linux/kernel.h>
 #include <linux/delay.h>
 #include <linux/init.h>
+#include <linux/of.h>
+#include <linux/of_pci.h>
 #include <linux/pci.h>
 #include <linux/pm.h>
 #include <linux/slab.h>

@@ -4447,6 +4449,54 @@ int pci_get_new_domain_nr(void)
 {
 	return atomic_inc_return(&__domain_nr);
 }
+
+#ifdef CONFIG_PCI_DOMAINS_GENERIC
+
+void pci_bus_assign_domain_nr(struct pci_bus *bus, struct device *parent)
+{
+	static int use_dt_domains = -1;
+	int domain = of_get_pci_domain_nr(parent->of_node);
+
+	/*
+	 * Check DT domain and use_dt_domains values.
+	 *
+	 * If DT domain property is valid (domain >= 0) and
+	 * use_dt_domains != 0, the DT assignment is valid since this means
+	 * we have not previously allocated a domain number by using
+	 * pci_get_new_domain_nr(); we should also update use_dt_domains to
+	 * 1, to indicate that we have just assigned a domain number from
+	 * DT.
+	 *
+	 * If DT domain property value is not valid (ie domain < 0), and we
+	 * have not previously assigned a domain number from DT
+	 * (use_dt_domains != 1) we should assign a domain number by
+	 * using the:
+	 *
+	 * pci_get_new_domain_nr()
+	 *
+	 * API and update the use_dt_domains value to keep track of method we
+	 * are using to assign domain numbers (use_dt_domains = 0).
+	 *
+	 * All other combinations imply we have a platform that is trying
+	 * to mix domain numbers obtained from DT and pci_get_new_domain_nr(),
+	 * which is a recipe for domain mishandling and it is prevented by
+	 * invalidating the domain value (domain = -1) and printing a
+	 * corresponding error.
+	 */
+	if (domain >= 0 && use_dt_domains) {
+		use_dt_domains = 1;
+	} else if (domain < 0 && use_dt_domains != 1) {
+		use_dt_domains = 0;
+		domain = pci_get_new_domain_nr();
+	} else {
+		dev_err(parent, "Node %s has inconsistent \"linux,pci-domain\" property in DT\n",
+			parent->of_node->full_name);
+		domain = -1;
+	}
+
+	bus->domain_nr = domain;
+}
+#endif
 #endif
 

Quoted 9 lines

Quoted 2 lines

drivers/pci/pci.c +0 -0

 /**

Signature · 2 lines

Thanks!
Yijing

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel at lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

The current logic used for PCI domain assignment in arm64 pci_bus_assign_domain_nr() is flawed in that, depending on the host controllers configuration for a platform and the respective initialization sequence, core code may end up allocating PCI domain numbers from both DT and the core code generic domain counter, which would result in PCI domain allocation aliases/errors.

This patch fixes the logic behind the PCI domain number assignment and moves the resulting code to generic PCI core code so that the same domain allocation logic is used on all platforms selecting

CONFIG_PCI_DOMAINS_GENERIC

instead of resorting to an arch specific implementation that might end up duplicating the PCI domain assignment logic wrongly.

Cc Arnd Bergmann <[email protected]>

Cc Liviu Dudau <[email protected]>

Cc Bjorn Helgaas <[email protected]>

Cc Rob Herring <[email protected]>

Cc Catalin Marinas <[email protected]>

Signed-off-by Lorenzo Pieralisi <[email protected]>

---
v1 => v2:

- Moved generic pci_bus_assign_domain_nr() code to PCI core instead of adding an OF layer API
- Updated commit log and code comments

 arch/arm64/kernel/pci.c | 22 ----------------------
 drivers/pci/pci.c       | 50 +++++++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 50 insertions(+), 22 deletions(-)

arch/arm64/kernel/pci.c +0 -22

index ce5836c..6f93c24 100644
@@ -46,25 +46,3 @@ int pcibios_add_device(struct pci_dev *dev)
 
 	return 0;
 }
-
-
-#ifdef CONFIG_PCI_DOMAINS_GENERIC
-static bool dt_domain_found = false;
-
-void pci_bus_assign_domain_nr(struct pci_bus *bus, struct device *parent)
-{
-	int domain = of_get_pci_domain_nr(parent->of_node);
-
-	if (domain >= 0) {
-		dt_domain_found = true;
-	} else if (dt_domain_found == true) {
-		dev_err(parent, "Node %s is missing \"linux,pci-domain\" property in DT\n",
-			parent->of_node->full_name);
-		return;
-	} else {
-		domain = pci_get_new_domain_nr();
-	}
-
-	bus->domain_nr = domain;
-}
-#endif

drivers/pci/pci.c +50 -0

index 625a4ac..2279414 100644
@@ -10,6 +10,8 @@
 #include <linux/kernel.h>
 #include <linux/delay.h>
 #include <linux/init.h>
+#include <linux/of.h>
+#include <linux/of_pci.h>
 #include <linux/pci.h>
 #include <linux/pm.h>
 #include <linux/slab.h>

@@ -4447,6 +4449,54 @@ int pci_get_new_domain_nr(void)
 {
 	return atomic_inc_return(&__domain_nr);
 }
+
+#ifdef CONFIG_PCI_DOMAINS_GENERIC
+
+void pci_bus_assign_domain_nr(struct pci_bus *bus, struct device *parent)
+{
+	static int use_dt_domains = -1;
+	int domain = of_get_pci_domain_nr(parent->of_node);
+
+	/*
+	 * Check DT domain and use_dt_domains values.
+	 *
+	 * If DT domain property is valid (domain >= 0) and
+	 * use_dt_domains != 0, the DT assignment is valid since this means
+	 * we have not previously allocated a domain number by using
+	 * pci_get_new_domain_nr(); we should also update use_dt_domains to
+	 * 1, to indicate that we have just assigned a domain number from
+	 * DT.
+	 *
+	 * If DT domain property value is not valid (ie domain < 0), and we
+	 * have not previously assigned a domain number from DT
+	 * (use_dt_domains != 1) we should assign a domain number by
+	 * using the:
+	 *
+	 * pci_get_new_domain_nr()
+	 *
+	 * API and update the use_dt_domains value to keep track of method we
+	 * are using to assign domain numbers (use_dt_domains = 0).
+	 *
+	 * All other combinations imply we have a platform that is trying
+	 * to mix domain numbers obtained from DT and pci_get_new_domain_nr(),
+	 * which is a recipe for domain mishandling and it is prevented by
+	 * invalidating the domain value (domain = -1) and printing a
+	 * corresponding error.
+	 */
+	if (domain >= 0 && use_dt_domains) {
+		use_dt_domains = 1;
+	} else if (domain < 0 && use_dt_domains != 1) {
+		use_dt_domains = 0;
+		domain = pci_get_new_domain_nr();
+	} else {
+		dev_err(parent, "Node %s has inconsistent \"linux,pci-domain\" property in DT\n",
+			parent->of_node->full_name);
+		domain = -1;
+	}
+
+	bus->domain_nr = domain;
+}
+#endif
 #endif
 
 /**

Signature · 1 line

2.1.2

Quoted 16 lines · Lorenzo Pieralisi

The current logic used for PCI domain assignment in arm64 pci_bus_assign_domain_nr() is flawed in that, depending on the host controllers configuration for a platform and the respective initialization sequence, core code may end up allocating PCI domain numbers from both DT and the core code generic domain counter, which would result in PCI domain allocation aliases/errors.

This patch fixes the logic behind the PCI domain number assignment and moves the resulting code to generic PCI core code so that the same domain allocation logic is used on all platforms selecting

CONFIG_PCI_DOMAINS_GENERIC

instead of resorting to an arch specific implementation that might end up duplicating the PCI domain assignment logic wrongly.

Quoted 6 lines

Cc Arnd Bergmann <[email protected]>

Cc Liviu Dudau <[email protected]>

Cc Bjorn Helgaas <[email protected]>

Cc Rob Herring <[email protected]>

Cc Catalin Marinas <[email protected]>

Signed-off-by Lorenzo Pieralisi <[email protected]>

Applied with Liviu and Arnd's acks to pci/domain for v3.19, thanks.

It doesn't matter whether I'm in the to: or cc: list. The important thing is that it appears on the linux-pci list, because I use patchwork as my to-do list, and patchwork only vacuums up stuff from linux-pci.

Quoted 112 lines

---
v1 => v2:

- Moved generic pci_bus_assign_domain_nr() code to PCI core instead of adding an OF layer API
- Updated commit log and code comments

 arch/arm64/kernel/pci.c | 22 ----------------------
 drivers/pci/pci.c       | 50 +++++++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 50 insertions(+), 22 deletions(-)

arch/arm64/kernel/pci.c +0 -22

index ce5836c..6f93c24 100644
@@ -46,25 +46,3 @@ int pcibios_add_device(struct pci_dev *dev)
 
 	return 0;
 }
-
-
-#ifdef CONFIG_PCI_DOMAINS_GENERIC
-static bool dt_domain_found = false;
-
-void pci_bus_assign_domain_nr(struct pci_bus *bus, struct device *parent)
-{
-	int domain = of_get_pci_domain_nr(parent->of_node);
-
-	if (domain >= 0) {
-		dt_domain_found = true;
-	} else if (dt_domain_found == true) {
-		dev_err(parent, "Node %s is missing \"linux,pci-domain\" property in DT\n",
-			parent->of_node->full_name);
-		return;
-	} else {
-		domain = pci_get_new_domain_nr();
-	}
-
-	bus->domain_nr = domain;
-}
-#endif

drivers/pci/pci.c +50 -0

index 625a4ac..2279414 100644
@@ -10,6 +10,8 @@
 #include <linux/kernel.h>
 #include <linux/delay.h>
 #include <linux/init.h>
+#include <linux/of.h>
+#include <linux/of_pci.h>
 #include <linux/pci.h>
 #include <linux/pm.h>
 #include <linux/slab.h>

@@ -4447,6 +4449,54 @@ int pci_get_new_domain_nr(void)
 {
 	return atomic_inc_return(&__domain_nr);
 }
+
+#ifdef CONFIG_PCI_DOMAINS_GENERIC
+
+void pci_bus_assign_domain_nr(struct pci_bus *bus, struct device *parent)
+{
+	static int use_dt_domains = -1;
+	int domain = of_get_pci_domain_nr(parent->of_node);
+
+	/*
+	 * Check DT domain and use_dt_domains values.
+	 *
+	 * If DT domain property is valid (domain >= 0) and
+	 * use_dt_domains != 0, the DT assignment is valid since this means
+	 * we have not previously allocated a domain number by using
+	 * pci_get_new_domain_nr(); we should also update use_dt_domains to
+	 * 1, to indicate that we have just assigned a domain number from
+	 * DT.
+	 *
+	 * If DT domain property value is not valid (ie domain < 0), and we
+	 * have not previously assigned a domain number from DT
+	 * (use_dt_domains != 1) we should assign a domain number by
+	 * using the:
+	 *
+	 * pci_get_new_domain_nr()
+	 *
+	 * API and update the use_dt_domains value to keep track of method we
+	 * are using to assign domain numbers (use_dt_domains = 0).
+	 *
+	 * All other combinations imply we have a platform that is trying
+	 * to mix domain numbers obtained from DT and pci_get_new_domain_nr(),
+	 * which is a recipe for domain mishandling and it is prevented by
+	 * invalidating the domain value (domain = -1) and printing a
+	 * corresponding error.
+	 */
+	if (domain >= 0 && use_dt_domains) {
+		use_dt_domains = 1;
+	} else if (domain < 0 && use_dt_domains != 1) {
+		use_dt_domains = 0;
+		domain = pci_get_new_domain_nr();
+	} else {
+		dev_err(parent, "Node %s has inconsistent \"linux,pci-domain\" property in DT\n",
+			parent->of_node->full_name);
+		domain = -1;
+	}
+
+	bus->domain_nr = domain;
+}
+#endif
 #endif
 
 /**

Signature · 1 line

2.1.2

Hi Bjorn,

Quoted 24 lines · Bjorn Helgaas

Quoted 16 lines · Lorenzo Pieralisi

The current logic used for PCI domain assignment in arm64 pci_bus_assign_domain_nr() is flawed in that, depending on the host controllers configuration for a platform and the respective initialization sequence, core code may end up allocating PCI domain numbers from both DT and the core code generic domain counter, which would result in PCI domain allocation aliases/errors.

This patch fixes the logic behind the PCI domain number assignment and moves the resulting code to generic PCI core code so that the same domain allocation logic is used on all platforms selecting

CONFIG_PCI_DOMAINS_GENERIC

instead of resorting to an arch specific implementation that might end up duplicating the PCI domain assignment logic wrongly.

Quoted 6 lines

Cc Arnd Bergmann <[email protected]>

Cc Liviu Dudau <[email protected]>

Cc Bjorn Helgaas <[email protected]>

Cc Rob Herring <[email protected]>

Cc Catalin Marinas <[email protected]>

Signed-off-by Lorenzo Pieralisi <[email protected]>

Quoted 5 lines

Applied with Liviu and Arnd's acks to pci/domain for v3.19, thanks.

It doesn't matter whether I'm in the to: or cc: list. The important thing is that it appears on the linux-pci list, because I use patchwork as my to-do list, and patchwork only vacuums up stuff from linux-pci.

This patch has not been merged last cycle and it is a prerequisite for the ARM pcibios domain code removal patchset you merged in your pci/domain branch. Please consider pulling it there so that the ARM pcibios domain removal patchset can be based correctly on top of this patch, at the moment it can't compile (see kbuild test report).