Re: [PATCH v3 04/27] arm/MSI: Save MSI chip in pci_sys_data
From: Bjorn Helgaas <bhelgaas@google.com>
Date: 2014-10-23 05:36:03
Also in:
linux-arch, linux-arm-kernel, linux-iommu, linux-mips, linux-pci, linux-s390, lkml, sparclinux
On Wed, Oct 15, 2014 at 11:06:52AM +0800, Yijing Wang wrote:
quoted hunk ↗ jump to hunk
Saving msi chip in pci_sys_data can make pci bus and devices don't need to know msi chip detail, it also make pci enumeration code be decoupled from msi chip. In fact, all pci devices under the same pci hostbridge share same msi chip. So msi chip should be seen as one of resources or attributes to be initialized in pci host bridge driver. Currently, pci hostbridge drivers create pci_host_bridge in pci_create_root_bus(), and pass arch specific pci sysdata to core pci scan functions. So pci arch sysdata is good place to save msi chip. Signed-off-by: Yijing Wang <redacted> --- arch/arm/include/asm/mach/pci.h | 6 ++++++ arch/arm/include/asm/pci.h | 9 +++++++++ arch/arm/kernel/bios32.c | 3 +++ drivers/pci/msi.c | 6 ++++++ include/linux/pci.h | 9 +++++++++ 5 files changed, 33 insertions(+), 0 deletions(-)diff --git a/arch/arm/include/asm/mach/pci.h b/arch/arm/include/asm/mach/pci.h index 7fc4278..59b0d87 100644 --- a/arch/arm/include/asm/mach/pci.h +++ b/arch/arm/include/asm/mach/pci.h@@ -22,6 +22,9 @@ struct hw_pci { #ifdef CONFIG_PCI_DOMAINS int domain; #endif +#ifdef CONFIG_PCI_MSI + struct msi_chip *msi_chip; +#endif struct pci_ops *ops; int nr_controllers; void **private_data;@@ -47,6 +50,9 @@ struct pci_sys_data { #ifdef CONFIG_PCI_DOMAINS int domain; #endif +#ifdef CONFIG_PCI_MSI + struct msi_chip *msi_chip; +#endif struct list_head node; int busnr; /* primary bus number */ u64 mem_offset; /* bus->cpu memory mapping offset */diff --git a/arch/arm/include/asm/pci.h b/arch/arm/include/asm/pci.h index 7e95d85..b562c09 100644 --- a/arch/arm/include/asm/pci.h +++ b/arch/arm/include/asm/pci.h@@ -31,6 +31,15 @@ static inline int pci_proc_domain(struct pci_bus *bus) } #endif /* CONFIG_PCI_DOMAINS */ +#ifdef CONFIG_PCI_MSI +static inline struct msi_chip *pci_msi_chip(struct pci_bus *bus) +{ + struct pci_sys_data *root = bus->sysdata; + + return root->msi_chip; +} +#endif + /* * The PCI address space does equal the physical memory address space. * The networking and block device layers use this boolean for bouncediff --git a/arch/arm/kernel/bios32.c b/arch/arm/kernel/bios32.c index 17a26c1..a19038d 100644 --- a/arch/arm/kernel/bios32.c +++ b/arch/arm/kernel/bios32.c@@ -471,6 +471,9 @@ static void pcibios_init_hw(struct device *parent, struct hw_pci *hw, #ifdef CONFIG_PCI_DOMAINS sys->domain = hw->domain; #endif +#ifdef CONFIG_PCI_MSI + sys->msi_chip = hw->msi_chip; +#endif sys->busnr = busnr; sys->swizzle = hw->swizzle; sys->map_irq = hw->map_irq;diff --git a/drivers/pci/msi.c b/drivers/pci/msi.c index 22e413c..f11108c 100644 --- a/drivers/pci/msi.c +++ b/drivers/pci/msi.c@@ -35,6 +35,9 @@ int __weak arch_setup_msi_irq(struct pci_dev *dev, struct msi_desc *desc) struct msi_chip *chip = dev->bus->msi; int err; + if (!chip) + chip = pci_msi_chip(dev->bus); + if (!chip || !chip->setup_irq) return -EINVAL;@@ -50,6 +53,9 @@ void __weak arch_teardown_msi_irq(unsigned int irq) struct msi_desc *entry = irq_get_msi_desc(irq); struct msi_chip *chip = entry->dev->bus->msi; + if (!chip) + chip = pci_msi_chip(entry->dev->bus); + if (!chip || !chip->teardown_irq) return;diff --git a/include/linux/pci.h b/include/linux/pci.h index 9cd2721..7a48b40 100644 --- a/include/linux/pci.h +++ b/include/linux/pci.h@@ -1433,6 +1433,15 @@ static inline int pci_get_new_domain_nr(void) { return -ENOSYS; } #include <asm/pci.h> +/* Just avoid compile error, will be clean up later */ +#ifdef CONFIG_PCI_MSI + +#ifndef pci_msi_chip +#define pci_msi_chip(bus) NULL +#endif +#endif
I don't like the mixture of ARM changes and PCI core changes in the same
patch. Can you split this into a core patch that does something like this:
struct msi_chip * __weak pcibios_msi_controller(struct pci_bus *bus)
{
return NULL;
}
struct msi_chip *pci_msi_controller(struct pci_bus *bus)
{
msi_chip *controller = bus->msi;
if (controller)
return controller;
return pcibios_msi_controller(bus);
}
followed by an ARM patch that puts the msi_chip pointer in struct hw_pci
and implements pcibios_msi_controller()?
I know you're trying to *remove* weak functions, and this adds one, but
this section of the series is more about getting rid of the ARM
pcibios_add_bus() because all it was used for was setting the bus->msi
pointer.
Eventually we might have a way to stash an MSI controller pointer in the
generic pci_host_bridge struct, and then the pcibios_msi_controller()
interface could go away.
+ /* these helpers provide future and backwards compatibility * for accessing popular PCI BAR info */ #define pci_resource_start(dev, bar) ((dev)->resource[(bar)].start) -- 1.7.1