Re: [PATCH] pci: generic host: make it more generic
On 11.11.14 11:33, Ming Lei wrote:
> This patch converts the generic host controller driver
> into a more generic one, and basically don't need
> platform's pcibios support, and use DT based generic
> APIs to parse resource and remap IO port.
>
> This patch has been tested on both ARMv7 and ARMv8
> VM, and in theroy it should support other ARCHs too.
>
> With this patch, virtio PCI block, network and scsi devices
> can work well on ARMv7 and ARMv8 VM.
>
> QEMU needs below patches for runing the test:
>
> - Rob Herring's "Add generic PCI host device" patches
> http://lists.gnu.org/archive/html/qemu-devel/2014-06/msg03482.html
> http://lists.gnu.org/archive/html/qemu-devel/2014-06/msg03483.html
> - Alvise Rigo's "Add Generic PCI host device update"
> http://marc.info/?l=qemu-devel=140506329920172=2
>
> For ARMv8, cpu model of "host" and "cortex-a57" is missed
> in Alvise Rigo's patchset, otherwise ARM64 VM can't boot.
>
> All these QEMU patches can be found in below tree:
>
>
> http://kernel.ubuntu.com/git?p=ming/qemu.git;a=shortlog;h=refs/heads/arm64-pci-test
>
> Signed-off-by: Ming Lei
> ---
> drivers/pci/host/Kconfig|2 +-
> drivers/pci/host/pci-host-generic.c | 196
> +++
> 2 files changed, 63 insertions(+), 135 deletions(-)
>
> diff --git a/drivers/pci/host/Kconfig b/drivers/pci/host/Kconfig
> index 3dc25fa..885034d 100644
> --- a/drivers/pci/host/Kconfig
> +++ b/drivers/pci/host/Kconfig
> @@ -50,7 +50,7 @@ config PCI_RCAR_GEN2_PCIE
>
> config PCI_HOST_GENERIC
> bool "Generic PCI host controller"
> - depends on ARM && OF
> + depends on OF
> help
> Say Y here if you want to support a simple generic PCI host
> controller, such as the one emulated by kvmtool.
> diff --git a/drivers/pci/host/pci-host-generic.c
> b/drivers/pci/host/pci-host-generic.c
> index 3d2076f..00d0ca3 100644
> --- a/drivers/pci/host/pci-host-generic.c
> +++ b/drivers/pci/host/pci-host-generic.c
> @@ -44,12 +44,14 @@ struct gen_pci {
> struct list_headresources;
> };
>
> +/* fake sysdata for cheating ARCH's pcibios code */
> +static char gen_sysdata[256];
> +
> static void __iomem *gen_pci_map_cfg_bus_cam(struct pci_bus *bus,
>unsigned int devfn,
>int where)
> {
> - struct pci_sys_data *sys = bus->sysdata;
> - struct gen_pci *pci = sys->private_data;
> + struct gen_pci *pci = dev_get_drvdata(bus->dev.parent->parent);
Just as a heads up, this breaks when you have PCI bridges in between. In
that case you need something like the ugly patch below on top.
Alex
diff --git a/drivers/pci/host/pci-host-generic.c
b/drivers/pci/host/pci-host-generic.c
index e580296..730a07b 100644
--- a/drivers/pci/host/pci-host-generic.c
+++ b/drivers/pci/host/pci-host-generic.c
@@ -47,11 +47,26 @@ struct gen_pci {
/* fake sysdata for cheating ARCH's pcibios code */
static chargen_sysdata[256];
+static struct gen_pci *gen_pci_get_drvdata(struct pci_bus *bus)
+{
+ struct device *dev = bus->dev.parent->parent;
+ struct gen_pci *pci;
+
+ while (dev) {
+ pci = dev_get_drvdata(dev);
+ if (pci)
+ return pci;
+ dev = dev->parent;
+ }
+
+ return NULL;
+}
+
static void __iomem *gen_pci_map_cfg_bus_cam(struct pci_bus *bus,
unsigned int devfn,
int where)
{
- struct gen_pci *pci = dev_get_drvdata(bus->dev.parent->parent);
+ struct gen_pci *pci = gen_pci_get_drvdata(bus);
resource_size_t idx = bus->number - pci->cfg.bus_range.start;
return pci->cfg.win[idx] + ((devfn << 8) | where);
@@ -66,7 +81,7 @@ static void __iomem *gen_pci_map_cfg_bus_ecam(struct
pci_bus *bus,
unsigned int devfn,
int where)
{
- struct gen_pci *pci = dev_get_drvdata(bus->dev.parent->parent);
+ struct gen_pci *pci = gen_pci_get_drvdata(bus);
resource_size_t idx = bus->number - pci->cfg.bus_range.start;
return pci->cfg.win[idx] + ((devfn << 12) | where);
@@ -81,7 +96,11 @@ static int gen_pci_config_read(struct pci_bus *bus,
unsigned int devfn,
int where, int size, u32 *val)
{
void __iomem *addr;
- struct gen_pci *pci = dev_get_drvdata(bus->dev.parent->parent);
+ struct gen_pci *pci = gen_pci_get_drvdata(bus);
+
+ WARN_ON(!pci);
+ if (!pci)
+ return PCIBIOS_DEVICE_NOT_FOUND;
addr = pci->cfg.ops->map_bus(bus, devfn, where);
@@ -103,7 +122,11 @@ static int gen_pci_config_write(struct pci_bus
*bus, unsigned int devfn,
int where, int size, u32 val)
{
void __iomem *addr;
-
Re: [PATCH] pci: generic host: make it more generic
On 11.11.14 11:33, Ming Lei wrote:
This patch converts the generic host controller driver
into a more generic one, and basically don't need
platform's pcibios support, and use DT based generic
APIs to parse resource and remap IO port.
This patch has been tested on both ARMv7 and ARMv8
VM, and in theroy it should support other ARCHs too.
With this patch, virtio PCI block, network and scsi devices
can work well on ARMv7 and ARMv8 VM.
QEMU needs below patches for runing the test:
- Rob Herring's Add generic PCI host device patches
http://lists.gnu.org/archive/html/qemu-devel/2014-06/msg03482.html
http://lists.gnu.org/archive/html/qemu-devel/2014-06/msg03483.html
- Alvise Rigo's Add Generic PCI host device update
http://marc.info/?l=qemu-develm=140506329920172w=2
For ARMv8, cpu model of host and cortex-a57 is missed
in Alvise Rigo's patchset, otherwise ARM64 VM can't boot.
All these QEMU patches can be found in below tree:
http://kernel.ubuntu.com/git?p=ming/qemu.git;a=shortlog;h=refs/heads/arm64-pci-test
Signed-off-by: Ming Lei ming@canonical.com
---
drivers/pci/host/Kconfig|2 +-
drivers/pci/host/pci-host-generic.c | 196
+++
2 files changed, 63 insertions(+), 135 deletions(-)
diff --git a/drivers/pci/host/Kconfig b/drivers/pci/host/Kconfig
index 3dc25fa..885034d 100644
--- a/drivers/pci/host/Kconfig
+++ b/drivers/pci/host/Kconfig
@@ -50,7 +50,7 @@ config PCI_RCAR_GEN2_PCIE
config PCI_HOST_GENERIC
bool Generic PCI host controller
- depends on ARM OF
+ depends on OF
help
Say Y here if you want to support a simple generic PCI host
controller, such as the one emulated by kvmtool.
diff --git a/drivers/pci/host/pci-host-generic.c
b/drivers/pci/host/pci-host-generic.c
index 3d2076f..00d0ca3 100644
--- a/drivers/pci/host/pci-host-generic.c
+++ b/drivers/pci/host/pci-host-generic.c
@@ -44,12 +44,14 @@ struct gen_pci {
struct list_headresources;
};
+/* fake sysdata for cheating ARCH's pcibios code */
+static char gen_sysdata[256];
+
static void __iomem *gen_pci_map_cfg_bus_cam(struct pci_bus *bus,
unsigned int devfn,
int where)
{
- struct pci_sys_data *sys = bus-sysdata;
- struct gen_pci *pci = sys-private_data;
+ struct gen_pci *pci = dev_get_drvdata(bus-dev.parent-parent);
Just as a heads up, this breaks when you have PCI bridges in between. In
that case you need something like the ugly patch below on top.
Alex
diff --git a/drivers/pci/host/pci-host-generic.c
b/drivers/pci/host/pci-host-generic.c
index e580296..730a07b 100644
--- a/drivers/pci/host/pci-host-generic.c
+++ b/drivers/pci/host/pci-host-generic.c
@@ -47,11 +47,26 @@ struct gen_pci {
/* fake sysdata for cheating ARCH's pcibios code */
static chargen_sysdata[256];
+static struct gen_pci *gen_pci_get_drvdata(struct pci_bus *bus)
+{
+ struct device *dev = bus-dev.parent-parent;
+ struct gen_pci *pci;
+
+ while (dev) {
+ pci = dev_get_drvdata(dev);
+ if (pci)
+ return pci;
+ dev = dev-parent;
+ }
+
+ return NULL;
+}
+
static void __iomem *gen_pci_map_cfg_bus_cam(struct pci_bus *bus,
unsigned int devfn,
int where)
{
- struct gen_pci *pci = dev_get_drvdata(bus-dev.parent-parent);
+ struct gen_pci *pci = gen_pci_get_drvdata(bus);
resource_size_t idx = bus-number - pci-cfg.bus_range.start;
return pci-cfg.win[idx] + ((devfn 8) | where);
@@ -66,7 +81,7 @@ static void __iomem *gen_pci_map_cfg_bus_ecam(struct
pci_bus *bus,
unsigned int devfn,
int where)
{
- struct gen_pci *pci = dev_get_drvdata(bus-dev.parent-parent);
+ struct gen_pci *pci = gen_pci_get_drvdata(bus);
resource_size_t idx = bus-number - pci-cfg.bus_range.start;
return pci-cfg.win[idx] + ((devfn 12) | where);
@@ -81,7 +96,11 @@ static int gen_pci_config_read(struct pci_bus *bus,
unsigned int devfn,
int where, int size, u32 *val)
{
void __iomem *addr;
- struct gen_pci *pci = dev_get_drvdata(bus-dev.parent-parent);
+ struct gen_pci *pci = gen_pci_get_drvdata(bus);
+
+ WARN_ON(!pci);
+ if (!pci)
+ return PCIBIOS_DEVICE_NOT_FOUND;
addr = pci-cfg.ops-map_bus(bus, devfn, where);
@@ -103,7 +122,11 @@ static int gen_pci_config_write(struct pci_bus
*bus, unsigned int devfn,
int where, int size, u32 val)
{
void __iomem *addr;
- struct gen_pci *pci = dev_get_drvdata(bus-dev.parent-parent);
+ struct
Re: [PATCH] pci: generic host: make it more generic
On Wed, Nov 12, 2014 at 1:58 AM, Ming Lei wrote: > On Wed, Nov 12, 2014 at 10:47 AM, Bjorn Helgaas wrote: >> On Tue, Nov 11, 2014 at 7:12 PM, Ming Lei wrote: >>> ... >>> Do you think it is doable to introduce a option(from module >>> parameter, or device tree, ...) to let the driver and PCI core >>> ignore/bypass all pcibios handling for the generic bus? >> >> No, I don't think so. > > Could you explain a bit? In VM world, it might be reasonable to > claim that there is no pcibios service or pcibios service needn't. > > IMO it isn't easy to make a cross-arch PCI generic controller > driver if pcibios handling has to be involved. I'm not going to add kernel or module parameters to change the way the basic code works. We need a way to factor the differences so the PCI core code is the same for all arches, and the differences are hidden in the arch-specific code. That's basically what the pcibios interfaces do. I suspect they were originally named "pcibios" because there's a "PCI BIOS" spec that tells you how to make BIOS calls to enumerate devices, access config space, manage interrupts, etc. But "pcibios" is now just the name of interfaces to arch-specific functionality. Most arches don't have a BIOS, and most pcibios_*() functions don't make BIOS calls. If you have a new arch that has no "pcibios service" (I don't know exactly what you mean by that), you can still implement the pcibios_*() interfaces used by the PCI core. If some of these interfaces don't actually do anything on your arch, that's fine; they can be no-ops. Bjorn -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH] pci: generic host: make it more generic
On Wed, Nov 12, 2014 at 8:32 PM, Lorenzo Pieralisi wrote: > On Wed, Nov 12, 2014 at 10:51:31AM +, Ming Lei wrote: >> On Wed, Nov 12, 2014 at 5:58 PM, Lorenzo Pieralisi >> wrote: >> > On Wed, Nov 12, 2014 at 02:12:12AM +, Ming Lei wrote: >> >> On Wed, Nov 12, 2014 at 2:24 AM, Lorenzo Pieralisi >> >> wrote: >> >> > On Tue, Nov 11, 2014 at 02:02:20PM +, Ming Lei wrote: >> >> > >> >> > [...] >> >> > >> >> >> > http://lists.infradead.org/pipermail/linux-arm-kernel/2014-October/296535.html >> >> >> > >> >> >> > Patch above is already queued and applies most of the changes you >> >> >> > have posted >> >> >> > above. >> >> >> >> >> >> I should have looked at list first before writing the patch, :-( >> >> >> >> >> >> Could you make it more generic so that other ARCHs(at least ARM64) >> >> >> can benefit from it too? >> >> > >> >> > That's our current goal, we are carrying out clean-ups to remove arch >> >> > dependency and move code to the generic layer PCI layer. >> >> >> >> OK, I am glad to test them after you post them out. >> >> >> >> At least, the 'struct gen_pci' pointer can be put in driver data >> >> of the platform_device now, which may remove dependency on >> >> pci_sysdata in this driver. But ARCH's pcibios code still may >> >> access 'pci_sysdata', do you have patches or solution to handle >> >> this issue? >> > >> > I will consider adding struct gen_pci pointer as platform device >> > driver data, we are working on removing pci_sys_data dependency in the ARM >> > pcibios backend, and I do not think we are far from achieving that. >> >> Other ARCHs has sort of pcibios backend too, so that said it still >> depends on ARM or ARM64, doesn't it? > > Yes, so ? If you think there is a way to make this host controller work I mean suppose the driver/device is only for ARM and ARM64, it should be easy to support, and it may not be necessary to remove pci_sys_data dependency in the ARM, such as: - make a different code path for ARM64 - or like this patch to fake sysdata for ARM - or what ever > on all arches you fancy using it on with no arch dependency in a proper > way step up to the plate and post the code, we will duly review it. The only way I thought of is to bypass pcibios handling for the generic host controller, but it may not be agreed. Thanks, Ming Lei -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH] pci: generic host: make it more generic
On Wed, Nov 12, 2014 at 10:51:31AM +, Ming Lei wrote: > On Wed, Nov 12, 2014 at 5:58 PM, Lorenzo Pieralisi > wrote: > > On Wed, Nov 12, 2014 at 02:12:12AM +, Ming Lei wrote: > >> On Wed, Nov 12, 2014 at 2:24 AM, Lorenzo Pieralisi > >> wrote: > >> > On Tue, Nov 11, 2014 at 02:02:20PM +, Ming Lei wrote: > >> > > >> > [...] > >> > > >> >> > http://lists.infradead.org/pipermail/linux-arm-kernel/2014-October/296535.html > >> >> > > >> >> > Patch above is already queued and applies most of the changes you > >> >> > have posted > >> >> > above. > >> >> > >> >> I should have looked at list first before writing the patch, :-( > >> >> > >> >> Could you make it more generic so that other ARCHs(at least ARM64) > >> >> can benefit from it too? > >> > > >> > That's our current goal, we are carrying out clean-ups to remove arch > >> > dependency and move code to the generic layer PCI layer. > >> > >> OK, I am glad to test them after you post them out. > >> > >> At least, the 'struct gen_pci' pointer can be put in driver data > >> of the platform_device now, which may remove dependency on > >> pci_sysdata in this driver. But ARCH's pcibios code still may > >> access 'pci_sysdata', do you have patches or solution to handle > >> this issue? > > > > I will consider adding struct gen_pci pointer as platform device > > driver data, we are working on removing pci_sys_data dependency in the ARM > > pcibios backend, and I do not think we are far from achieving that. > > Other ARCHs has sort of pcibios backend too, so that said it still > depends on ARM or ARM64, doesn't it? Yes, so ? If you think there is a way to make this host controller work on all arches you fancy using it on with no arch dependency in a proper way step up to the plate and post the code, we will duly review it. Thanks, Lorenzo -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH] pci: generic host: make it more generic
On Wed, Nov 12, 2014 at 5:58 PM, Lorenzo Pieralisi wrote: > On Wed, Nov 12, 2014 at 02:12:12AM +, Ming Lei wrote: >> On Wed, Nov 12, 2014 at 2:24 AM, Lorenzo Pieralisi >> wrote: >> > On Tue, Nov 11, 2014 at 02:02:20PM +, Ming Lei wrote: >> > >> > [...] >> > >> >> > http://lists.infradead.org/pipermail/linux-arm-kernel/2014-October/296535.html >> >> > >> >> > Patch above is already queued and applies most of the changes you have >> >> > posted >> >> > above. >> >> >> >> I should have looked at list first before writing the patch, :-( >> >> >> >> Could you make it more generic so that other ARCHs(at least ARM64) >> >> can benefit from it too? >> > >> > That's our current goal, we are carrying out clean-ups to remove arch >> > dependency and move code to the generic layer PCI layer. >> >> OK, I am glad to test them after you post them out. >> >> At least, the 'struct gen_pci' pointer can be put in driver data >> of the platform_device now, which may remove dependency on >> pci_sysdata in this driver. But ARCH's pcibios code still may >> access 'pci_sysdata', do you have patches or solution to handle >> this issue? > > I will consider adding struct gen_pci pointer as platform device > driver data, we are working on removing pci_sys_data dependency in the ARM > pcibios backend, and I do not think we are far from achieving that. Other ARCHs has sort of pcibios backend too, so that said it still depends on ARM or ARM64, doesn't it? Thanks, -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH] pci: generic host: make it more generic
On Wed, Nov 12, 2014 at 02:12:12AM +, Ming Lei wrote: > On Wed, Nov 12, 2014 at 2:24 AM, Lorenzo Pieralisi > wrote: > > On Tue, Nov 11, 2014 at 02:02:20PM +, Ming Lei wrote: > > > > [...] > > > >> > http://lists.infradead.org/pipermail/linux-arm-kernel/2014-October/296535.html > >> > > >> > Patch above is already queued and applies most of the changes you have > >> > posted > >> > above. > >> > >> I should have looked at list first before writing the patch, :-( > >> > >> Could you make it more generic so that other ARCHs(at least ARM64) > >> can benefit from it too? > > > > That's our current goal, we are carrying out clean-ups to remove arch > > dependency and move code to the generic layer PCI layer. > > OK, I am glad to test them after you post them out. > > At least, the 'struct gen_pci' pointer can be put in driver data > of the platform_device now, which may remove dependency on > pci_sysdata in this driver. But ARCH's pcibios code still may > access 'pci_sysdata', do you have patches or solution to handle > this issue? I will consider adding struct gen_pci pointer as platform device driver data, we are working on removing pci_sys_data dependency in the ARM pcibios backend, and I do not think we are far from achieving that. Thanks, Lorenzo -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH] pci: generic host: make it more generic
On Wed, Nov 12, 2014 at 10:47 AM, Bjorn Helgaas wrote: > On Tue, Nov 11, 2014 at 7:12 PM, Ming Lei wrote: >> ... >> Do you think it is doable to introduce a option(from module >> parameter, or device tree, ...) to let the driver and PCI core >> ignore/bypass all pcibios handling for the generic bus? > > No, I don't think so. Could you explain a bit? In VM world, it might be reasonable to claim that there is no pcibios service or pcibios service needn't. IMO it isn't easy to make a cross-arch PCI generic controller driver if pcibios handling has to be involved. Thanks, Ming Lei -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH] pci: generic host: make it more generic
On Wed, Nov 12, 2014 at 10:47 AM, Bjorn Helgaas bhelg...@google.com wrote: On Tue, Nov 11, 2014 at 7:12 PM, Ming Lei ming@canonical.com wrote: ... Do you think it is doable to introduce a option(from module parameter, or device tree, ...) to let the driver and PCI core ignore/bypass all pcibios handling for the generic bus? No, I don't think so. Could you explain a bit? In VM world, it might be reasonable to claim that there is no pcibios service or pcibios service needn't. IMO it isn't easy to make a cross-arch PCI generic controller driver if pcibios handling has to be involved. Thanks, Ming Lei -- To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH] pci: generic host: make it more generic
On Wed, Nov 12, 2014 at 02:12:12AM +, Ming Lei wrote: On Wed, Nov 12, 2014 at 2:24 AM, Lorenzo Pieralisi lorenzo.pieral...@arm.com wrote: On Tue, Nov 11, 2014 at 02:02:20PM +, Ming Lei wrote: [...] http://lists.infradead.org/pipermail/linux-arm-kernel/2014-October/296535.html Patch above is already queued and applies most of the changes you have posted above. I should have looked at list first before writing the patch, :-( Could you make it more generic so that other ARCHs(at least ARM64) can benefit from it too? That's our current goal, we are carrying out clean-ups to remove arch dependency and move code to the generic layer PCI layer. OK, I am glad to test them after you post them out. At least, the 'struct gen_pci' pointer can be put in driver data of the platform_device now, which may remove dependency on pci_sysdata in this driver. But ARCH's pcibios code still may access 'pci_sysdata', do you have patches or solution to handle this issue? I will consider adding struct gen_pci pointer as platform device driver data, we are working on removing pci_sys_data dependency in the ARM pcibios backend, and I do not think we are far from achieving that. Thanks, Lorenzo -- To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH] pci: generic host: make it more generic
On Wed, Nov 12, 2014 at 5:58 PM, Lorenzo Pieralisi lorenzo.pieral...@arm.com wrote: On Wed, Nov 12, 2014 at 02:12:12AM +, Ming Lei wrote: On Wed, Nov 12, 2014 at 2:24 AM, Lorenzo Pieralisi lorenzo.pieral...@arm.com wrote: On Tue, Nov 11, 2014 at 02:02:20PM +, Ming Lei wrote: [...] http://lists.infradead.org/pipermail/linux-arm-kernel/2014-October/296535.html Patch above is already queued and applies most of the changes you have posted above. I should have looked at list first before writing the patch, :-( Could you make it more generic so that other ARCHs(at least ARM64) can benefit from it too? That's our current goal, we are carrying out clean-ups to remove arch dependency and move code to the generic layer PCI layer. OK, I am glad to test them after you post them out. At least, the 'struct gen_pci' pointer can be put in driver data of the platform_device now, which may remove dependency on pci_sysdata in this driver. But ARCH's pcibios code still may access 'pci_sysdata', do you have patches or solution to handle this issue? I will consider adding struct gen_pci pointer as platform device driver data, we are working on removing pci_sys_data dependency in the ARM pcibios backend, and I do not think we are far from achieving that. Other ARCHs has sort of pcibios backend too, so that said it still depends on ARM or ARM64, doesn't it? Thanks, -- To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH] pci: generic host: make it more generic
On Wed, Nov 12, 2014 at 10:51:31AM +, Ming Lei wrote: On Wed, Nov 12, 2014 at 5:58 PM, Lorenzo Pieralisi lorenzo.pieral...@arm.com wrote: On Wed, Nov 12, 2014 at 02:12:12AM +, Ming Lei wrote: On Wed, Nov 12, 2014 at 2:24 AM, Lorenzo Pieralisi lorenzo.pieral...@arm.com wrote: On Tue, Nov 11, 2014 at 02:02:20PM +, Ming Lei wrote: [...] http://lists.infradead.org/pipermail/linux-arm-kernel/2014-October/296535.html Patch above is already queued and applies most of the changes you have posted above. I should have looked at list first before writing the patch, :-( Could you make it more generic so that other ARCHs(at least ARM64) can benefit from it too? That's our current goal, we are carrying out clean-ups to remove arch dependency and move code to the generic layer PCI layer. OK, I am glad to test them after you post them out. At least, the 'struct gen_pci' pointer can be put in driver data of the platform_device now, which may remove dependency on pci_sysdata in this driver. But ARCH's pcibios code still may access 'pci_sysdata', do you have patches or solution to handle this issue? I will consider adding struct gen_pci pointer as platform device driver data, we are working on removing pci_sys_data dependency in the ARM pcibios backend, and I do not think we are far from achieving that. Other ARCHs has sort of pcibios backend too, so that said it still depends on ARM or ARM64, doesn't it? Yes, so ? If you think there is a way to make this host controller work on all arches you fancy using it on with no arch dependency in a proper way step up to the plate and post the code, we will duly review it. Thanks, Lorenzo -- To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH] pci: generic host: make it more generic
On Wed, Nov 12, 2014 at 8:32 PM, Lorenzo Pieralisi lorenzo.pieral...@arm.com wrote: On Wed, Nov 12, 2014 at 10:51:31AM +, Ming Lei wrote: On Wed, Nov 12, 2014 at 5:58 PM, Lorenzo Pieralisi lorenzo.pieral...@arm.com wrote: On Wed, Nov 12, 2014 at 02:12:12AM +, Ming Lei wrote: On Wed, Nov 12, 2014 at 2:24 AM, Lorenzo Pieralisi lorenzo.pieral...@arm.com wrote: On Tue, Nov 11, 2014 at 02:02:20PM +, Ming Lei wrote: [...] http://lists.infradead.org/pipermail/linux-arm-kernel/2014-October/296535.html Patch above is already queued and applies most of the changes you have posted above. I should have looked at list first before writing the patch, :-( Could you make it more generic so that other ARCHs(at least ARM64) can benefit from it too? That's our current goal, we are carrying out clean-ups to remove arch dependency and move code to the generic layer PCI layer. OK, I am glad to test them after you post them out. At least, the 'struct gen_pci' pointer can be put in driver data of the platform_device now, which may remove dependency on pci_sysdata in this driver. But ARCH's pcibios code still may access 'pci_sysdata', do you have patches or solution to handle this issue? I will consider adding struct gen_pci pointer as platform device driver data, we are working on removing pci_sys_data dependency in the ARM pcibios backend, and I do not think we are far from achieving that. Other ARCHs has sort of pcibios backend too, so that said it still depends on ARM or ARM64, doesn't it? Yes, so ? If you think there is a way to make this host controller work I mean suppose the driver/device is only for ARM and ARM64, it should be easy to support, and it may not be necessary to remove pci_sys_data dependency in the ARM, such as: - make a different code path for ARM64 - or like this patch to fake sysdata for ARM - or what ever on all arches you fancy using it on with no arch dependency in a proper way step up to the plate and post the code, we will duly review it. The only way I thought of is to bypass pcibios handling for the generic host controller, but it may not be agreed. Thanks, Ming Lei -- To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH] pci: generic host: make it more generic
On Wed, Nov 12, 2014 at 1:58 AM, Ming Lei ming@canonical.com wrote: On Wed, Nov 12, 2014 at 10:47 AM, Bjorn Helgaas bhelg...@google.com wrote: On Tue, Nov 11, 2014 at 7:12 PM, Ming Lei ming@canonical.com wrote: ... Do you think it is doable to introduce a option(from module parameter, or device tree, ...) to let the driver and PCI core ignore/bypass all pcibios handling for the generic bus? No, I don't think so. Could you explain a bit? In VM world, it might be reasonable to claim that there is no pcibios service or pcibios service needn't. IMO it isn't easy to make a cross-arch PCI generic controller driver if pcibios handling has to be involved. I'm not going to add kernel or module parameters to change the way the basic code works. We need a way to factor the differences so the PCI core code is the same for all arches, and the differences are hidden in the arch-specific code. That's basically what the pcibios interfaces do. I suspect they were originally named pcibios because there's a PCI BIOS spec that tells you how to make BIOS calls to enumerate devices, access config space, manage interrupts, etc. But pcibios is now just the name of interfaces to arch-specific functionality. Most arches don't have a BIOS, and most pcibios_*() functions don't make BIOS calls. If you have a new arch that has no pcibios service (I don't know exactly what you mean by that), you can still implement the pcibios_*() interfaces used by the PCI core. If some of these interfaces don't actually do anything on your arch, that's fine; they can be no-ops. Bjorn -- To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH] pci: generic host: make it more generic
On Tue, Nov 11, 2014 at 7:12 PM, Ming Lei wrote: > ... > Do you think it is doable to introduce a option(from module > parameter, or device tree, ...) to let the driver and PCI core > ignore/bypass all pcibios handling for the generic bus? No, I don't think so. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH] pci: generic host: make it more generic
On Wed, Nov 12, 2014 at 2:24 AM, Lorenzo Pieralisi wrote: > On Tue, Nov 11, 2014 at 02:02:20PM +, Ming Lei wrote: > > [...] > >> > http://lists.infradead.org/pipermail/linux-arm-kernel/2014-October/296535.html >> > >> > Patch above is already queued and applies most of the changes you have >> > posted >> > above. >> >> I should have looked at list first before writing the patch, :-( >> >> Could you make it more generic so that other ARCHs(at least ARM64) >> can benefit from it too? > > That's our current goal, we are carrying out clean-ups to remove arch > dependency and move code to the generic layer PCI layer. OK, I am glad to test them after you post them out. At least, the 'struct gen_pci' pointer can be put in driver data of the platform_device now, which may remove dependency on pci_sysdata in this driver. But ARCH's pcibios code still may access 'pci_sysdata', do you have patches or solution to handle this issue? Do you think it is doable to introduce a option(from module parameter, or device tree, ...) to let the driver and PCI core ignore/bypass all pcibios handling for the generic bus? If that is OK, it can provide an approach to reach the goal a bit easy. In reality, generic host controller should be only used by VM, and VM's firmware shouldn't be very complicated and I think the option can meet most of demands. Thanks, Ming Lei -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH] pci: generic host: make it more generic
On Tue, Nov 11, 2014 at 02:02:20PM +, Ming Lei wrote: [...] > > http://lists.infradead.org/pipermail/linux-arm-kernel/2014-October/296535.html > > > > Patch above is already queued and applies most of the changes you have > > posted > > above. > > I should have looked at list first before writing the patch, :-( > > Could you make it more generic so that other ARCHs(at least ARM64) > can benefit from it too? That's our current goal, we are carrying out clean-ups to remove arch dependency and move code to the generic layer PCI layer. Lorenzo -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH] pci: generic host: make it more generic
On Tue, Nov 11, 2014 at 8:29 PM, Lorenzo Pieralisi
wrote:
> On Tue, Nov 11, 2014 at 10:33:59AM +, Ming Lei wrote:
>> This patch converts the generic host controller driver
>> into a more generic one, and basically don't need
>> platform's pcibios support, and use DT based generic
>> APIs to parse resource and remap IO port.
>>
>> This patch has been tested on both ARMv7 and ARMv8
>> VM, and in theroy it should support other ARCHs too.
>
> In theory, yes.
>
>> With this patch, virtio PCI block, network and scsi devices
>> can work well on ARMv7 and ARMv8 VM.
>>
>> QEMU needs below patches for runing the test:
>>
>> - Rob Herring's "Add generic PCI host device" patches
>> http://lists.gnu.org/archive/html/qemu-devel/2014-06/msg03482.html
>> http://lists.gnu.org/archive/html/qemu-devel/2014-06/msg03483.html
>> - Alvise Rigo's "Add Generic PCI host device update"
>> http://marc.info/?l=qemu-devel=140506329920172=2
>>
>> For ARMv8, cpu model of "host" and "cortex-a57" is missed
>> in Alvise Rigo's patchset, otherwise ARM64 VM can't boot.
>>
>> All these QEMU patches can be found in below tree:
>>
>>
>> http://kernel.ubuntu.com/git?p=ming/qemu.git;a=shortlog;h=refs/heads/arm64-pci-test
>>
>> Signed-off-by: Ming Lei
>> ---
>> drivers/pci/host/Kconfig|2 +-
>> drivers/pci/host/pci-host-generic.c | 196
>> +++
>> 2 files changed, 63 insertions(+), 135 deletions(-)
>>
>> diff --git a/drivers/pci/host/Kconfig b/drivers/pci/host/Kconfig
>> index 3dc25fa..885034d 100644
>> --- a/drivers/pci/host/Kconfig
>> +++ b/drivers/pci/host/Kconfig
>> @@ -50,7 +50,7 @@ config PCI_RCAR_GEN2_PCIE
>>
>> config PCI_HOST_GENERIC
>> bool "Generic PCI host controller"
>> - depends on ARM && OF
>> + depends on OF
>> help
>> Say Y here if you want to support a simple generic PCI host
>> controller, such as the one emulated by kvmtool.
>> diff --git a/drivers/pci/host/pci-host-generic.c
>> b/drivers/pci/host/pci-host-generic.c
>> index 3d2076f..00d0ca3 100644
>> --- a/drivers/pci/host/pci-host-generic.c
>> +++ b/drivers/pci/host/pci-host-generic.c
>> @@ -44,12 +44,14 @@ struct gen_pci {
>> struct list_headresources;
>> };
>>
>> +/* fake sysdata for cheating ARCH's pcibios code */
>> +static char gen_sysdata[256];
>
> I take this as a joke and it does not make me laugh.
One motivation of the patch is to remove ARCH or platform dependency,
that is why the above faked sysdata is introduced, but pci_sysdata is a
generic issue.
>
>> static void __iomem *gen_pci_map_cfg_bus_cam(struct pci_bus *bus,
>>unsigned int devfn,
>>int where)
>> {
>> - struct pci_sys_data *sys = bus->sysdata;
>> - struct gen_pci *pci = sys->private_data;
>> + struct gen_pci *pci = dev_get_drvdata(bus->dev.parent->parent);
>> resource_size_t idx = bus->number - pci->cfg.bus_range.start;
>>
>> return pci->cfg.win[idx] + ((devfn << 8) | where);
>> @@ -64,8 +66,7 @@ static void __iomem *gen_pci_map_cfg_bus_ecam(struct
>> pci_bus *bus,
>> unsigned int devfn,
>> int where)
>> {
>> - struct pci_sys_data *sys = bus->sysdata;
>> - struct gen_pci *pci = sys->private_data;
>> + struct gen_pci *pci = dev_get_drvdata(bus->dev.parent->parent);
>> resource_size_t idx = bus->number - pci->cfg.bus_range.start;
>>
>> return pci->cfg.win[idx] + ((devfn << 12) | where);
>> @@ -80,8 +81,7 @@ static int gen_pci_config_read(struct pci_bus *bus,
>> unsigned int devfn,
>> int where, int size, u32 *val)
>> {
>> void __iomem *addr;
>> - struct pci_sys_data *sys = bus->sysdata;
>> - struct gen_pci *pci = sys->private_data;
>> + struct gen_pci *pci = dev_get_drvdata(bus->dev.parent->parent);
>>
>> addr = pci->cfg.ops->map_bus(bus, devfn, where);
>>
>> @@ -103,8 +103,7 @@ static int gen_pci_config_write(struct pci_bus *bus,
>> unsigned int devfn,
>>int where, int size, u32 val)
>> {
>> void __iomem *addr;
>> - struct pci_sys_data *sys = bus->sysdata;
>> - struct gen_pci *pci = sys->private_data;
>> + struct gen_pci *pci = dev_get_drvdata(bus->dev.parent->parent);
>>
>> addr = pci->cfg.ops->map_bus(bus, devfn, where);
>>
>> @@ -138,45 +137,6 @@ static const struct of_device_id gen_pci_of_match[] = {
>> };
>> MODULE_DEVICE_TABLE(of, gen_pci_of_match);
>>
>> -static int gen_pci_calc_io_offset(struct device *dev,
>> - struct of_pci_range *range,
>> - struct resource *res,
>> - resource_size_t *offset)
>> -{
>> - static atomic_t wins = ATOMIC_INIT(0);
>> - int err, idx, max_win;
>> - unsigned int window;
>> -
>> -
Re: [PATCH] pci: generic host: make it more generic
On Tue, Nov 11, 2014 at 10:33:59AM +, Ming Lei wrote:
> This patch converts the generic host controller driver
> into a more generic one, and basically don't need
> platform's pcibios support, and use DT based generic
> APIs to parse resource and remap IO port.
>
> This patch has been tested on both ARMv7 and ARMv8
> VM, and in theroy it should support other ARCHs too.
In theory, yes.
> With this patch, virtio PCI block, network and scsi devices
> can work well on ARMv7 and ARMv8 VM.
>
> QEMU needs below patches for runing the test:
>
> - Rob Herring's "Add generic PCI host device" patches
> http://lists.gnu.org/archive/html/qemu-devel/2014-06/msg03482.html
> http://lists.gnu.org/archive/html/qemu-devel/2014-06/msg03483.html
> - Alvise Rigo's "Add Generic PCI host device update"
> http://marc.info/?l=qemu-devel=140506329920172=2
>
> For ARMv8, cpu model of "host" and "cortex-a57" is missed
> in Alvise Rigo's patchset, otherwise ARM64 VM can't boot.
>
> All these QEMU patches can be found in below tree:
>
>
> http://kernel.ubuntu.com/git?p=ming/qemu.git;a=shortlog;h=refs/heads/arm64-pci-test
>
> Signed-off-by: Ming Lei
> ---
> drivers/pci/host/Kconfig|2 +-
> drivers/pci/host/pci-host-generic.c | 196
> +++
> 2 files changed, 63 insertions(+), 135 deletions(-)
>
> diff --git a/drivers/pci/host/Kconfig b/drivers/pci/host/Kconfig
> index 3dc25fa..885034d 100644
> --- a/drivers/pci/host/Kconfig
> +++ b/drivers/pci/host/Kconfig
> @@ -50,7 +50,7 @@ config PCI_RCAR_GEN2_PCIE
>
> config PCI_HOST_GENERIC
> bool "Generic PCI host controller"
> - depends on ARM && OF
> + depends on OF
> help
> Say Y here if you want to support a simple generic PCI host
> controller, such as the one emulated by kvmtool.
> diff --git a/drivers/pci/host/pci-host-generic.c
> b/drivers/pci/host/pci-host-generic.c
> index 3d2076f..00d0ca3 100644
> --- a/drivers/pci/host/pci-host-generic.c
> +++ b/drivers/pci/host/pci-host-generic.c
> @@ -44,12 +44,14 @@ struct gen_pci {
> struct list_headresources;
> };
>
> +/* fake sysdata for cheating ARCH's pcibios code */
> +static char gen_sysdata[256];
I take this as a joke and it does not make me laugh.
> static void __iomem *gen_pci_map_cfg_bus_cam(struct pci_bus *bus,
>unsigned int devfn,
>int where)
> {
> - struct pci_sys_data *sys = bus->sysdata;
> - struct gen_pci *pci = sys->private_data;
> + struct gen_pci *pci = dev_get_drvdata(bus->dev.parent->parent);
> resource_size_t idx = bus->number - pci->cfg.bus_range.start;
>
> return pci->cfg.win[idx] + ((devfn << 8) | where);
> @@ -64,8 +66,7 @@ static void __iomem *gen_pci_map_cfg_bus_ecam(struct
> pci_bus *bus,
> unsigned int devfn,
> int where)
> {
> - struct pci_sys_data *sys = bus->sysdata;
> - struct gen_pci *pci = sys->private_data;
> + struct gen_pci *pci = dev_get_drvdata(bus->dev.parent->parent);
> resource_size_t idx = bus->number - pci->cfg.bus_range.start;
>
> return pci->cfg.win[idx] + ((devfn << 12) | where);
> @@ -80,8 +81,7 @@ static int gen_pci_config_read(struct pci_bus *bus,
> unsigned int devfn,
> int where, int size, u32 *val)
> {
> void __iomem *addr;
> - struct pci_sys_data *sys = bus->sysdata;
> - struct gen_pci *pci = sys->private_data;
> + struct gen_pci *pci = dev_get_drvdata(bus->dev.parent->parent);
>
> addr = pci->cfg.ops->map_bus(bus, devfn, where);
>
> @@ -103,8 +103,7 @@ static int gen_pci_config_write(struct pci_bus *bus,
> unsigned int devfn,
>int where, int size, u32 val)
> {
> void __iomem *addr;
> - struct pci_sys_data *sys = bus->sysdata;
> - struct gen_pci *pci = sys->private_data;
> + struct gen_pci *pci = dev_get_drvdata(bus->dev.parent->parent);
>
> addr = pci->cfg.ops->map_bus(bus, devfn, where);
>
> @@ -138,45 +137,6 @@ static const struct of_device_id gen_pci_of_match[] = {
> };
> MODULE_DEVICE_TABLE(of, gen_pci_of_match);
>
> -static int gen_pci_calc_io_offset(struct device *dev,
> - struct of_pci_range *range,
> - struct resource *res,
> - resource_size_t *offset)
> -{
> - static atomic_t wins = ATOMIC_INIT(0);
> - int err, idx, max_win;
> - unsigned int window;
> -
> - if (!PAGE_ALIGNED(range->cpu_addr))
> - return -EINVAL;
> -
> - max_win = (IO_SPACE_LIMIT + 1) / SZ_64K;
> - idx = atomic_inc_return();
> - if (idx > max_win)
> - return -ENOSPC;
> -
> - window = (idx - 1) * SZ_64K;
> - err = pci_ioremap_io(window, range->cpu_addr);
> -
Re: [PATCH] pci: generic host: make it more generic
On Tue, Nov 11, 2014 at 10:33:59AM +, Ming Lei wrote:
This patch converts the generic host controller driver
into a more generic one, and basically don't need
platform's pcibios support, and use DT based generic
APIs to parse resource and remap IO port.
This patch has been tested on both ARMv7 and ARMv8
VM, and in theroy it should support other ARCHs too.
In theory, yes.
With this patch, virtio PCI block, network and scsi devices
can work well on ARMv7 and ARMv8 VM.
QEMU needs below patches for runing the test:
- Rob Herring's Add generic PCI host device patches
http://lists.gnu.org/archive/html/qemu-devel/2014-06/msg03482.html
http://lists.gnu.org/archive/html/qemu-devel/2014-06/msg03483.html
- Alvise Rigo's Add Generic PCI host device update
http://marc.info/?l=qemu-develm=140506329920172w=2
For ARMv8, cpu model of host and cortex-a57 is missed
in Alvise Rigo's patchset, otherwise ARM64 VM can't boot.
All these QEMU patches can be found in below tree:
http://kernel.ubuntu.com/git?p=ming/qemu.git;a=shortlog;h=refs/heads/arm64-pci-test
Signed-off-by: Ming Lei ming@canonical.com
---
drivers/pci/host/Kconfig|2 +-
drivers/pci/host/pci-host-generic.c | 196
+++
2 files changed, 63 insertions(+), 135 deletions(-)
diff --git a/drivers/pci/host/Kconfig b/drivers/pci/host/Kconfig
index 3dc25fa..885034d 100644
--- a/drivers/pci/host/Kconfig
+++ b/drivers/pci/host/Kconfig
@@ -50,7 +50,7 @@ config PCI_RCAR_GEN2_PCIE
config PCI_HOST_GENERIC
bool Generic PCI host controller
- depends on ARM OF
+ depends on OF
help
Say Y here if you want to support a simple generic PCI host
controller, such as the one emulated by kvmtool.
diff --git a/drivers/pci/host/pci-host-generic.c
b/drivers/pci/host/pci-host-generic.c
index 3d2076f..00d0ca3 100644
--- a/drivers/pci/host/pci-host-generic.c
+++ b/drivers/pci/host/pci-host-generic.c
@@ -44,12 +44,14 @@ struct gen_pci {
struct list_headresources;
};
+/* fake sysdata for cheating ARCH's pcibios code */
+static char gen_sysdata[256];
I take this as a joke and it does not make me laugh.
static void __iomem *gen_pci_map_cfg_bus_cam(struct pci_bus *bus,
unsigned int devfn,
int where)
{
- struct pci_sys_data *sys = bus-sysdata;
- struct gen_pci *pci = sys-private_data;
+ struct gen_pci *pci = dev_get_drvdata(bus-dev.parent-parent);
resource_size_t idx = bus-number - pci-cfg.bus_range.start;
return pci-cfg.win[idx] + ((devfn 8) | where);
@@ -64,8 +66,7 @@ static void __iomem *gen_pci_map_cfg_bus_ecam(struct
pci_bus *bus,
unsigned int devfn,
int where)
{
- struct pci_sys_data *sys = bus-sysdata;
- struct gen_pci *pci = sys-private_data;
+ struct gen_pci *pci = dev_get_drvdata(bus-dev.parent-parent);
resource_size_t idx = bus-number - pci-cfg.bus_range.start;
return pci-cfg.win[idx] + ((devfn 12) | where);
@@ -80,8 +81,7 @@ static int gen_pci_config_read(struct pci_bus *bus,
unsigned int devfn,
int where, int size, u32 *val)
{
void __iomem *addr;
- struct pci_sys_data *sys = bus-sysdata;
- struct gen_pci *pci = sys-private_data;
+ struct gen_pci *pci = dev_get_drvdata(bus-dev.parent-parent);
addr = pci-cfg.ops-map_bus(bus, devfn, where);
@@ -103,8 +103,7 @@ static int gen_pci_config_write(struct pci_bus *bus,
unsigned int devfn,
int where, int size, u32 val)
{
void __iomem *addr;
- struct pci_sys_data *sys = bus-sysdata;
- struct gen_pci *pci = sys-private_data;
+ struct gen_pci *pci = dev_get_drvdata(bus-dev.parent-parent);
addr = pci-cfg.ops-map_bus(bus, devfn, where);
@@ -138,45 +137,6 @@ static const struct of_device_id gen_pci_of_match[] = {
};
MODULE_DEVICE_TABLE(of, gen_pci_of_match);
-static int gen_pci_calc_io_offset(struct device *dev,
- struct of_pci_range *range,
- struct resource *res,
- resource_size_t *offset)
-{
- static atomic_t wins = ATOMIC_INIT(0);
- int err, idx, max_win;
- unsigned int window;
-
- if (!PAGE_ALIGNED(range-cpu_addr))
- return -EINVAL;
-
- max_win = (IO_SPACE_LIMIT + 1) / SZ_64K;
- idx = atomic_inc_return(wins);
- if (idx max_win)
- return -ENOSPC;
-
- window = (idx - 1) * SZ_64K;
- err = pci_ioremap_io(window, range-cpu_addr);
- if (err)
- return err;
-
- of_pci_range_to_resource(range, dev-of_node, res);
- res-start = window;
-
Re: [PATCH] pci: generic host: make it more generic
On Tue, Nov 11, 2014 at 8:29 PM, Lorenzo Pieralisi
lorenzo.pieral...@arm.com wrote:
On Tue, Nov 11, 2014 at 10:33:59AM +, Ming Lei wrote:
This patch converts the generic host controller driver
into a more generic one, and basically don't need
platform's pcibios support, and use DT based generic
APIs to parse resource and remap IO port.
This patch has been tested on both ARMv7 and ARMv8
VM, and in theroy it should support other ARCHs too.
In theory, yes.
With this patch, virtio PCI block, network and scsi devices
can work well on ARMv7 and ARMv8 VM.
QEMU needs below patches for runing the test:
- Rob Herring's Add generic PCI host device patches
http://lists.gnu.org/archive/html/qemu-devel/2014-06/msg03482.html
http://lists.gnu.org/archive/html/qemu-devel/2014-06/msg03483.html
- Alvise Rigo's Add Generic PCI host device update
http://marc.info/?l=qemu-develm=140506329920172w=2
For ARMv8, cpu model of host and cortex-a57 is missed
in Alvise Rigo's patchset, otherwise ARM64 VM can't boot.
All these QEMU patches can be found in below tree:
http://kernel.ubuntu.com/git?p=ming/qemu.git;a=shortlog;h=refs/heads/arm64-pci-test
Signed-off-by: Ming Lei ming@canonical.com
---
drivers/pci/host/Kconfig|2 +-
drivers/pci/host/pci-host-generic.c | 196
+++
2 files changed, 63 insertions(+), 135 deletions(-)
diff --git a/drivers/pci/host/Kconfig b/drivers/pci/host/Kconfig
index 3dc25fa..885034d 100644
--- a/drivers/pci/host/Kconfig
+++ b/drivers/pci/host/Kconfig
@@ -50,7 +50,7 @@ config PCI_RCAR_GEN2_PCIE
config PCI_HOST_GENERIC
bool Generic PCI host controller
- depends on ARM OF
+ depends on OF
help
Say Y here if you want to support a simple generic PCI host
controller, such as the one emulated by kvmtool.
diff --git a/drivers/pci/host/pci-host-generic.c
b/drivers/pci/host/pci-host-generic.c
index 3d2076f..00d0ca3 100644
--- a/drivers/pci/host/pci-host-generic.c
+++ b/drivers/pci/host/pci-host-generic.c
@@ -44,12 +44,14 @@ struct gen_pci {
struct list_headresources;
};
+/* fake sysdata for cheating ARCH's pcibios code */
+static char gen_sysdata[256];
I take this as a joke and it does not make me laugh.
One motivation of the patch is to remove ARCH or platform dependency,
that is why the above faked sysdata is introduced, but pci_sysdata is a
generic issue.
static void __iomem *gen_pci_map_cfg_bus_cam(struct pci_bus *bus,
unsigned int devfn,
int where)
{
- struct pci_sys_data *sys = bus-sysdata;
- struct gen_pci *pci = sys-private_data;
+ struct gen_pci *pci = dev_get_drvdata(bus-dev.parent-parent);
resource_size_t idx = bus-number - pci-cfg.bus_range.start;
return pci-cfg.win[idx] + ((devfn 8) | where);
@@ -64,8 +66,7 @@ static void __iomem *gen_pci_map_cfg_bus_ecam(struct
pci_bus *bus,
unsigned int devfn,
int where)
{
- struct pci_sys_data *sys = bus-sysdata;
- struct gen_pci *pci = sys-private_data;
+ struct gen_pci *pci = dev_get_drvdata(bus-dev.parent-parent);
resource_size_t idx = bus-number - pci-cfg.bus_range.start;
return pci-cfg.win[idx] + ((devfn 12) | where);
@@ -80,8 +81,7 @@ static int gen_pci_config_read(struct pci_bus *bus,
unsigned int devfn,
int where, int size, u32 *val)
{
void __iomem *addr;
- struct pci_sys_data *sys = bus-sysdata;
- struct gen_pci *pci = sys-private_data;
+ struct gen_pci *pci = dev_get_drvdata(bus-dev.parent-parent);
addr = pci-cfg.ops-map_bus(bus, devfn, where);
@@ -103,8 +103,7 @@ static int gen_pci_config_write(struct pci_bus *bus,
unsigned int devfn,
int where, int size, u32 val)
{
void __iomem *addr;
- struct pci_sys_data *sys = bus-sysdata;
- struct gen_pci *pci = sys-private_data;
+ struct gen_pci *pci = dev_get_drvdata(bus-dev.parent-parent);
addr = pci-cfg.ops-map_bus(bus, devfn, where);
@@ -138,45 +137,6 @@ static const struct of_device_id gen_pci_of_match[] = {
};
MODULE_DEVICE_TABLE(of, gen_pci_of_match);
-static int gen_pci_calc_io_offset(struct device *dev,
- struct of_pci_range *range,
- struct resource *res,
- resource_size_t *offset)
-{
- static atomic_t wins = ATOMIC_INIT(0);
- int err, idx, max_win;
- unsigned int window;
-
- if (!PAGE_ALIGNED(range-cpu_addr))
- return -EINVAL;
-
- max_win = (IO_SPACE_LIMIT + 1) / SZ_64K;
- idx = atomic_inc_return(wins);
- if (idx max_win)
- return -ENOSPC;
-
- window
Re: [PATCH] pci: generic host: make it more generic
On Tue, Nov 11, 2014 at 02:02:20PM +, Ming Lei wrote: [...] http://lists.infradead.org/pipermail/linux-arm-kernel/2014-October/296535.html Patch above is already queued and applies most of the changes you have posted above. I should have looked at list first before writing the patch, :-( Could you make it more generic so that other ARCHs(at least ARM64) can benefit from it too? That's our current goal, we are carrying out clean-ups to remove arch dependency and move code to the generic layer PCI layer. Lorenzo -- To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH] pci: generic host: make it more generic
On Wed, Nov 12, 2014 at 2:24 AM, Lorenzo Pieralisi lorenzo.pieral...@arm.com wrote: On Tue, Nov 11, 2014 at 02:02:20PM +, Ming Lei wrote: [...] http://lists.infradead.org/pipermail/linux-arm-kernel/2014-October/296535.html Patch above is already queued and applies most of the changes you have posted above. I should have looked at list first before writing the patch, :-( Could you make it more generic so that other ARCHs(at least ARM64) can benefit from it too? That's our current goal, we are carrying out clean-ups to remove arch dependency and move code to the generic layer PCI layer. OK, I am glad to test them after you post them out. At least, the 'struct gen_pci' pointer can be put in driver data of the platform_device now, which may remove dependency on pci_sysdata in this driver. But ARCH's pcibios code still may access 'pci_sysdata', do you have patches or solution to handle this issue? Do you think it is doable to introduce a option(from module parameter, or device tree, ...) to let the driver and PCI core ignore/bypass all pcibios handling for the generic bus? If that is OK, it can provide an approach to reach the goal a bit easy. In reality, generic host controller should be only used by VM, and VM's firmware shouldn't be very complicated and I think the option can meet most of demands. Thanks, Ming Lei -- To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH] pci: generic host: make it more generic
On Tue, Nov 11, 2014 at 7:12 PM, Ming Lei ming@canonical.com wrote: ... Do you think it is doable to introduce a option(from module parameter, or device tree, ...) to let the driver and PCI core ignore/bypass all pcibios handling for the generic bus? No, I don't think so. -- To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
