x86/x86-64 PCI domain support
PATCH 1/1 x86/x86-64 PCI domain support
Now that the dust has settled and the prep work is upstream, adding PCI domain support to x86 is a lot more straightforward.
Targetted for 2.6.24.
commit c7b14cf6aa39e09aef89a0cd08c553dba1ee7e98
Author: Jeff Garzik <[email protected]> Date: Sat Sep 1 07:36:25 2007 -0400
PCI domain support for x86[-64]
aka "PCI segment" support, in ACPI lingo.
arch/i386/pci/acpi.c | 15 ++++++++++----- arch/i386/pci/common.c | 6 ++++-- arch/x86_64/Kconfig | 4 ++++ include/asm-i386/pci.h | 14 ++++++++++++++ include/asm-x86_64/pci.h | 14 ++++++++++++++ 5 files changed, 46 insertions(+), 7 deletions(-)
c7b14cf6aa39e09aef89a0cd08c553dba1ee7e98
arch/i386/pci/acpi.c +10 -5
@@ -11,6 +11,13 @@ struct pci_bus * __devinit pci_acpi_scan_root(struct acpi_device *device, int dostruct pci_sysdata *sd;int pxm; +#ifndef CONFIG_PCI_DOMAINS +if (domain !=0 ) { +printk (KERN_WARNING "PCI: Multiple domains not supported\n" ); +return NULL ; + } +#endif +/* Allocate per-root-bus (not per bus) arch-specific data. * TODO: leak; this memory is never freed. * It's arguable whether it's worth the trouble to care.
@@ -21,11 +28,9 @@ struct pci_bus * __devinit pci_acpi_scan_root(struct acpi_device *device, int doreturn NULL ; } -if (domain !=0 ) { -printk (KERN_WARNING "PCI: Multiple domains not supported\n" ); -kfree (sd); -return NULL ; - } +#ifdef CONFIG_PCI_DOMAINS + sd->domain = domain; +#endif sd->node =-1 ;
arch/i386/pci/common.c +4 -2
@@ -29,12 +29,14 @@ struct pci_raw_ops *raw_pci_ops;static int pci_read (struct pci_bus *bus,unsigned int devfn,int where,int size,u32 *value) { -return raw_pci_ops->read (0 , bus->number, devfn, where, size, value); +return raw_pci_ops->read (pci_domain_nr (bus), bus->number, + devfn, where, size, value); }static int pci_write (struct pci_bus *bus,unsigned int devfn,int where,int size,u32 value) { -return raw_pci_ops->write (0 , bus->number, devfn, where, size, value); +return raw_pci_ops->write (pci_domain_nr (bus), bus->number, + devfn, where, size, value); }struct pci_ops pci_root_ops = {
arch/x86_64/Kconfig +4 -0
@@ -732,6 +732,10 @@ config PCI_MMCONFIG bool "Support mmconfig PCI config space access" depends on PCI && ACPI +config PCI_DOMAINS + bool "PCI domain support" + depends on PCI + source "drivers/pci/pcie/Kconfig" source "drivers/pci/Kconfig"
include/asm-i386/pci.h +14 -0
@@ -5,12 +5,26 @@#ifdef __KERNEL__struct pci_sysdata { +int domain;/* PCI domain */ int node;/* NUMA node */ };/* scan a bus after allocating a pci_sysdata for it */ extern struct pci_bus *pci_scan_bus_with_sysdata (int busno); +#ifdef CONFIG_PCI_DOMAINS +static inline int pci_domain_nr (struct pci_bus *bus) +{ +struct pci_sysdata *sd = bus->sysdata; +return sd->domain; +} + +static inline int pci_proc_domain (struct pci_bus *bus) +{ +return pci_domain_nr (bus); +} +#endif /* CONFIG_PCI_DOMAINS */ +#include <linux/mm.h> /* for struct page */ /* Can be used to override the logic in pci_scan_bus for skipping
include/asm-x86_64/pci.h +14 -0
@@ -6,12 +6,26 @@ #ifdef __KERNEL__ struct pci_sysdata { + int domain; /* PCI domain */ int node; /* NUMA node */ void* iommu; /* IOMMU private data */ }; extern struct pci_bus *pci_scan_bus_with_sysdata(int busno); +#ifdef CONFIG_PCI_DOMAINS +static inline int pci_domain_nr(struct pci_bus *bus) +{ + struct pci_sysdata *sd = bus->sysdata; + return sd->domain; +} + +static inline int pci_proc_domain(struct pci_bus *bus) +{ + return pci_domain_nr(bus); +} +#endif /* CONFIG_PCI_DOMAINS */ + #ifdef CONFIG_CALGARY_IOMMU static inline void* pci_iommu(struct pci_bus *bus) {
arch/x86_64/Kconfig config PCI_MMCONFIG · line 737
Quoted 5 lines
arch/x86_64/Kconfig +3
-0+config PCI_DOMAINS + bool "PCI domain support" + depends on PCI
I don't think this should be a config option.
But there should be a pci=... option with Documentation to turn it off at runtime
include/asm-i386/pci.h line 25
Quoted 4 lines
include/asm-i386/pci.h +4
-0+static inline int pci_proc_domain (struct pci_bus *bus) +{ +return pci_domain_nr (bus); +}
The second function is redundant?
-Andi
Quoted 6 lines · Andi Kleen
arch/x86_64/Kconfig config PCI_MMCONFIG · line 737
Quoted 4 lines
arch/x86_64/Kconfig +3
-0+config PCI_DOMAINS + bool "PCI domain support" + depends on PCII don't think this should be a config option.
CONFIG_PCI_DOMAINS is referenced in arch-neutral code, so the symbol _must_ be defined.
It can be hidden, perhaps, if that makes our x86 maintainer happy :)
Quoted 1 line
But there should be a pci=... option with Documentation to turn it off at runtime
OK
Quoted 6 lines
include/asm-i386/pci.h line 25
Quoted 4 lines
include/asm-i386/pci.h +4
-0+static inline int pci_proc_domain (struct pci_bus *bus) +{ +return pci_domain_nr (bus); +}The second function is redundant?
No, it's a hook we must implement, when CONFIG_PCI_DOMAINS is enabled.
Jeff
Quoted 4 lines
Quoted 2 lines
The second function is redundant?
No, it's a hook we must implement, when
CONFIG_PCI_DOMAINSis enabled.
Then the other function is redundant.
-Andi
Quoted 4 lines · Andi Kleen
Quoted 2 lines
Quoted 1 line
The second function is redundant?
No, it's a hook we must implement, when
CONFIG_PCI_DOMAINSis enabled.Then the other function is redundant.
No, both functions are required by the interface.
Jeff
Quoted 7 lines · On Sat, 01 Sep 2007 18:40:27 -0400 Jeff Garzik wrote:
Quoted 4 lines · Andi Kleen
Quoted 2 lines
Quoted 1 line
The second function is redundant?
No, it's a hook we must implement, when
CONFIG_PCI_DOMAINSis enabled.Then the other function is redundant.
No, both functions are required by the interface.
by what interface? and why, please? (instead of just stating "required")
---
~Randy
*** Remember to use Documentation/SubmitChecklist when testing your code ***
Quoted 9 lines · Randy Dunlap
Quoted 5 lines · On Sat, 01 Sep 2007 18:40:27 -0400 Jeff Garzik wrote:
Quoted 3 lines · Andi Kleen
Quoted 2 lines
Quoted 1 line
The second function is redundant?
No, it's a hook we must implement, when
CONFIG_PCI_DOMAINSis enabled.Then the other function is redundant.
No, both functions are required by the interface.
by what interface? and why, please? (instead of just stating "required")
grep for CONFIG_PCI_DOMAINS in arch code and include/linux/pci.h.
This is normal "arch" interface: you enable a define, and a group of functions is assumed to be present. Otherwise (!defined), a set of stub no-ops is activated for your arch.
I have implemented the [small] group of functions the code assumes to be present, when CONFIG_PCI_DOMAINS is enabled, like all the other arches that implement PCI domain support.
Jeff
Quoted 4 lines · Jeff Garzik
Quoted 3 lines · Randy Dunlap
Quoted 1 line · On Sat, 01 Sep 2007 18:40:27 -0400 Jeff Garzik wrote:
Andi Kleen wrote:
Quoted 8 lines
Quoted 6 lines
Quoted 4 lines
Quoted 3 lines
Quoted 2 lines
Quoted 1 line
The second function is redundant?
No, it's a hook we must implement, when
CONFIG_PCI_DOMAINSis enabled.Then the other function is redundant.
No, both functions are required by the interface.
by what interface? and why, please? (instead of just stating "required")
grep for
CONFIG_PCI_DOMAINSin arch code andinclude/linux/pci.h.
Thanks, I get it.
Quoted 3 lines
This is normal "arch" interface: you enable a define, and a group of functions is assumed to be present. Otherwise (!defined), a set of stub no-ops is activated for your arch.
I have implemented the [small] group of functions the code assumes to be present, when
CONFIG_PCI_DOMAINSis enabled, like all the other arches that implement PCI domain support.
+#ifdef CONFIG_PCI_DOMAINS
+static inline int pci_domain_nr(struct pci_bus *bus) +{
+ struct pci_sysdata *sd = bus->sysdata;
+ return sd->domain;
+}
+
+static inline int pci_proc_domain(struct pci_bus *bus) +{
+ return pci_domain_nr(bus);
+}
+#endif /* CONFIG_PCI_DOMAINS */
So if CONFIG_PCI_DOMAINS=y, proc_proc_domain() decides to print the domain nr in /proc iff the domain nr != 0 ? so that the /proc file format is different depending on the domain nr (0 vs. !0) ?
I suppose that is similar to what arch/powerpc/kernel/pci_64.c does with hose->buid. I think that I had rather see the domain nr printed whenever CONFIG_PCI_DOMAINS=y... oh well.
Signature · 2 lines
~Randy *** Remember to use Documentation/SubmitChecklist when testing your code ***
Quoted 5 lines · Jeff Garzik
Now that the dust has settled and the prep work is upstream, adding PCI edomain support to x86 is a lot more straightforward.
Targetted for 2.6.24.
Looks good to me.
Cheers,
Muli