[RFC PATCH 0/3] Pass resources to pci_create_bus() and fix MIPS PCI resources

[RFC PATCH 0/3] Pass resources to pci_create_bus() and fix MIPS PCI resources

[Deng-Cheng Zhu]

For MIPS PCI, use the resources-list style to set up root resources rather than
filling in pci_bus->resource[] array directly. This will hide some ugly
implementation details.

In addition, change the pci_create_bus() interface to pass in available
resources to get them settled down early. This is to avoid possible resource
conflicts while doing pci_scan_slot() in pci_scan_child_bus(). Note that
pcibios_fixup_bus() can get rid of such conflicts, but it's done AFTER scanning
slots.

Deng-Cheng Zhu (3):
  MIPS: PCI: Use pci_bus_remove_resources()/pci_bus_add_resource() to
    set up root resources
  PCI: Pass available resources into pci_create_bus()
  MIPS: PCI: Pass controller's resources to pci_create_bus() in
    pcibios_scanbus()

 arch/microblaze/pci/pci-common.c |    3 +-
 arch/mips/pci/pci.c              |   49 +++++++++++++++++++++++++++++++++++--
 arch/powerpc/kernel/pci-common.c |    3 +-
 arch/sparc/kernel/pci.c          |    3 +-
 arch/x86/pci/acpi.c              |    2 +-
 drivers/pci/probe.c              |   15 ++++++++---
 include/linux/pci.h              |    3 +-
 7 files changed, 66 insertions(+), 12 deletions(-)

[RFC PATCH 1/3] MIPS: PCI: Use pci_bus_remove_resources()/pci_bus_add_resource() to set up root resources

[Deng-Cheng Zhu]

Use this new style (instead of filling in the pci_bus->resource[] array
directly) to hide some ugly implementation details.

Signed-off-by: Deng-Cheng Zhu <dengcheng.zhu@gmail.com>
---
 arch/mips/pci/pci.c |   11 +++++++++--
 1 files changed, 9 insertions(+), 2 deletions(-)

diff --git a/arch/mips/pci/pci.c b/arch/mips/pci/pci.c
index 33bba7b..7473214 100644
--- a/arch/mips/pci/pci.c
+++ b/arch/mips/pci/pci.c
@@ -261,6 +261,14 @@ static void pcibios_fixup_device_resources(struct pci_dev *dev,
 	}
 }
 
+static void __devinit
+pcibios_setup_root_resources(struct pci_bus *bus, struct pci_controller *ctrl)
+{
+	pci_bus_remove_resources(bus);
+	pci_bus_add_resource(bus, ctrl->io_resource, 0);
+	pci_bus_add_resource(bus, ctrl->mem_resource, 0);
+}
+
 void __devinit pcibios_fixup_bus(struct pci_bus *bus)
 {
 	/* Propagate hose info into the subordinate devices.  */
@@ -270,8 +278,7 @@ void __devinit pcibios_fixup_bus(struct pci_bus *bus)
 	struct pci_dev *dev = bus->self;
 
 	if (!dev) {
-		bus->resource[0] = hose->io_resource;
-		bus->resource[1] = hose->mem_resource;
+		pcibios_setup_root_resources(bus, hose);
 	} else if (pci_probe_only &&
 		   (dev->class >> 8) == PCI_CLASS_BRIDGE_PCI) {
 		pci_read_bridge_bases(bus);
-- 
1.7.1

Re: [RFC PATCH 1/3] MIPS: PCI: Use pci_bus_remove_resources()/pci_bus_add_resource() to set up root resources

[Bjorn Helgaas]

On Wed, Aug 24, 2011 at 12:24 AM, Deng-Cheng Zhu
<dengcheng.zhu@gmail.com> wrote:
> Use this new style (instead of filling in the pci_bus->resource[] array
> directly) to hide some ugly implementation details.
>
> Signed-off-by: Deng-Cheng Zhu <dengcheng.zhu@gmail.com>
> ---
>  arch/mips/pci/pci.c |   11 +++++++++--
>  1 files changed, 9 insertions(+), 2 deletions(-)
>
> diff --git a/arch/mips/pci/pci.c b/arch/mips/pci/pci.c
> index 33bba7b..7473214 100644
> --- a/arch/mips/pci/pci.c
> +++ b/arch/mips/pci/pci.c
> @@ -261,6 +261,14 @@ static void pcibios_fixup_device_resources(struct pci_dev *dev,
>        }
>  }
>
> +static void __devinit
> +pcibios_setup_root_resources(struct pci_bus *bus, struct pci_controller *ctrl)
> +{
> +       pci_bus_remove_resources(bus);
> +       pci_bus_add_resource(bus, ctrl->io_resource, 0);
> +       pci_bus_add_resource(bus, ctrl->mem_resource, 0);
> +}
> +
>  void __devinit pcibios_fixup_bus(struct pci_bus *bus)
>  {
>        /* Propagate hose info into the subordinate devices.  */
> @@ -270,8 +278,7 @@ void __devinit pcibios_fixup_bus(struct pci_bus *bus)
>        struct pci_dev *dev = bus->self;
>
>        if (!dev) {
> -               bus->resource[0] = hose->io_resource;
> -               bus->resource[1] = hose->mem_resource;
> +               pcibios_setup_root_resources(bus, hose);
>        } else if (pci_probe_only &&
>                   (dev->class >> 8) == PCI_CLASS_BRIDGE_PCI) {
>                pci_read_bridge_bases(bus);

I don't understand this patch.

Wouldn't it be enough to have [PATCH 2/3], which adds the
pci_create_bus() argument with nobody using it yet, then [PATCH 3/3],
which makes mips supply the new argument, and add a hunk to [PATCH
3/3] that completely removes the bus->resource fixups in
pcibios_fixup_bus() at the same time?

Bjorn

Re: [RFC PATCH 1/3] MIPS: PCI: Use pci_bus_remove_resources()/pci_bus_add_resource() to set up root resources

[Deng-Cheng Zhu]

2011/8/24 Bjorn Helgaas <bhelgaas@google.com>:
> Wouldn't it be enough to have [PATCH 2/3], which adds the
> pci_create_bus() argument with nobody using it yet, then [PATCH 3/3],
> which makes mips supply the new argument, and add a hunk to [PATCH
> 3/3] that completely removes the bus->resource fixups in
> pcibios_fixup_bus() at the same time?
>
> Bjorn
>

Do you mean to merge this patch to [PATCH 3/3]? OK, that's good!


Deng-Cheng

Re: [RFC PATCH 1/3] MIPS: PCI: Use pci_bus_remove_resources()/pci_bus_add_resource() to set up root resources

[Bjorn Helgaas]

On Thu, Aug 25, 2011 at 12:41 AM, Deng-Cheng Zhu
<dengcheng.zhu@gmail.com> wrote:
> 2011/8/24 Bjorn Helgaas <bhelgaas@google.com>:
>> Wouldn't it be enough to have [PATCH 2/3], which adds the
>> pci_create_bus() argument with nobody using it yet, then [PATCH 3/3],
>> which makes mips supply the new argument, and add a hunk to [PATCH
>> 3/3] that completely removes the bus->resource fixups in
>> pcibios_fixup_bus() at the same time?
>>
>> Bjorn
>>
>
> Do you mean to merge this patch to [PATCH 3/3]? OK, that's good!

No, I just mean that I don't see why you need this patch at all.  If
you pass the list of bus resources to pci_create_bus(), there's no
need to fix anything up later.  Or am I missing something?

Bjorn

Re: [RFC PATCH 1/3] MIPS: PCI: Use pci_bus_remove_resources()/pci_bus_add_resource() to set up root resources

[Deng-Cheng Zhu]

2011/8/25 Bjorn Helgaas <bhelgaas@google.com>:
> No, I just mean that I don't see why you need this patch at all.  If
> you pass the list of bus resources to pci_create_bus(), there's no
> need to fix anything up later.  Or am I missing something?

Well, doing the root resource fixups in here is a *paranoid* way. It's to
deal with the 'unlikely' circumstance where controller_resources() returns
the NULL pointer in pcibios_scanbus() due to memory allocation failure.
Most of the time (always) it's nothing more than repeating the resource
list setup. But maybe we can do something like this:

if (unlikely(!dev && list_empty(&bus->resources))
        pcibios_setup_root_resources(bus, hose);


What do you think?


Deng-Cheng

Re: [RFC PATCH 1/3] MIPS: PCI: Use pci_bus_remove_resources()/pci_bus_add_resource() to set up root resources

[Bjorn Helgaas]

On Thu, Aug 25, 2011 at 5:14 PM, Deng-Cheng Zhu <dengcheng.zhu@gmail.com> wrote:
> 2011/8/25 Bjorn Helgaas <bhelgaas@google.com>:
>> No, I just mean that I don't see why you need this patch at all.  If
>> you pass the list of bus resources to pci_create_bus(), there's no
>> need to fix anything up later.  Or am I missing something?
>
> Well, doing the root resource fixups in here is a *paranoid* way. It's to
> deal with the 'unlikely' circumstance where controller_resources() returns
> the NULL pointer in pcibios_scanbus() due to memory allocation failure.
> Most of the time (always) it's nothing more than repeating the resource
> list setup. But maybe we can do something like this:
>
> if (unlikely(!dev && list_empty(&bus->resources))
>        pcibios_setup_root_resources(bus, hose);
>
>
> What do you think?

I don't think it's worth it.  Just check for failure of
controller_resources(), and if it fails, skip the pci_create_bus()
call.  You've already printed a message (you might add the PCI
domain/bus number).

Bjorn

[RFC PATCH 2/3] PCI: Pass available resources into pci_create_bus()

[Deng-Cheng Zhu]

Currently, after pci_create_bus(), resources available on the bus could be
handled by pci_scan_child_bus(). The problem is that, in
pci_scan_child_bus(), before calling arch-dependent pcibios_fixup_bus(),
PCI quirks will probably conflict (while doing pci_claim_resource()) with
resources like system controller's I/O resource that have not yet been
added to the bus. So, add those resources right before returning the newly
created bus in pci_create_bus().

A related discussion thread can be found here:
http://www.spinics.net/lists/mips/msg41654.html

Signed-off-by: Deng-Cheng Zhu <dengcheng.zhu@gmail.com>
---
 arch/microblaze/pci/pci-common.c |    3 ++-
 arch/powerpc/kernel/pci-common.c |    3 ++-
 arch/sparc/kernel/pci.c          |    3 ++-
 arch/x86/pci/acpi.c              |    2 +-
 drivers/pci/probe.c              |   15 +++++++++++----
 include/linux/pci.h              |    3 ++-
 6 files changed, 20 insertions(+), 9 deletions(-)

diff --git a/arch/microblaze/pci/pci-common.c b/arch/microblaze/pci/pci-common.c
index 4cfae20..9c35aa6 100644
--- a/arch/microblaze/pci/pci-common.c
+++ b/arch/microblaze/pci/pci-common.c
@@ -1581,7 +1581,8 @@ static void __devinit pcibios_scan_phb(struct pci_controller *hose)
 		 node ? node->full_name : "<NO NAME>");
 
 	/* Create an empty bus for the toplevel */
-	bus = pci_create_bus(hose->parent, hose->first_busno, hose->ops, hose);
+	bus = pci_create_bus(hose->parent, hose->first_busno,
+			     hose->ops, hose, NULL);
 	if (bus == NULL) {
 		printk(KERN_ERR "Failed to create bus for PCI domain %04x\n",
 		       hose->global_number);
diff --git a/arch/powerpc/kernel/pci-common.c b/arch/powerpc/kernel/pci-common.c
index 32656f1..2ede26a 100644
--- a/arch/powerpc/kernel/pci-common.c
+++ b/arch/powerpc/kernel/pci-common.c
@@ -1703,7 +1703,8 @@ void __devinit pcibios_scan_phb(struct pci_controller *hose)
 		 node ? node->full_name : "<NO NAME>");
 
 	/* Create an empty bus for the toplevel */
-	bus = pci_create_bus(hose->parent, hose->first_busno, hose->ops, hose);
+	bus = pci_create_bus(hose->parent, hose->first_busno,
+			     hose->ops, hose, NULL);
 	if (bus == NULL) {
 		pr_err("Failed to create bus for PCI domain %04x\n",
 			hose->global_number);
diff --git a/arch/sparc/kernel/pci.c b/arch/sparc/kernel/pci.c
index 1e94f94..77c38bb 100644
--- a/arch/sparc/kernel/pci.c
+++ b/arch/sparc/kernel/pci.c
@@ -689,7 +689,8 @@ struct pci_bus * __devinit pci_scan_one_pbm(struct pci_pbm_info *pbm,
 
 	printk("PCI: Scanning PBM %s\n", node->full_name);
 
-	bus = pci_create_bus(parent, pbm->pci_first_busno, pbm->pci_ops, pbm);
+	bus = pci_create_bus(parent, pbm->pci_first_busno,
+			     pbm->pci_ops, pbm, NULL);
 	if (!bus) {
 		printk(KERN_ERR "Failed to create bus for %s\n",
 		       node->full_name);
diff --git a/arch/x86/pci/acpi.c b/arch/x86/pci/acpi.c
index c953302..bab2113 100644
--- a/arch/x86/pci/acpi.c
+++ b/arch/x86/pci/acpi.c
@@ -353,7 +353,7 @@ struct pci_bus * __devinit pci_acpi_scan_root(struct acpi_pci_root *root)
 		memcpy(bus->sysdata, sd, sizeof(*sd));
 		kfree(sd);
 	} else {
-		bus = pci_create_bus(NULL, busnum, &pci_root_ops, sd);
+		bus = pci_create_bus(NULL, busnum, &pci_root_ops, sd, NULL);
 		if (bus) {
 			get_current_resources(device, busnum, domain, bus);
 			bus->subordinate = pci_scan_child_bus(bus);
diff --git a/drivers/pci/probe.c b/drivers/pci/probe.c
index 8473727..7735fe7 100644
--- a/drivers/pci/probe.c
+++ b/drivers/pci/probe.c
@@ -1516,7 +1516,8 @@ unsigned int __devinit pci_scan_child_bus(struct pci_bus *bus)
 }
 
 struct pci_bus * pci_create_bus(struct device *parent,
-		int bus, struct pci_ops *ops, void *sysdata)
+		int bus, struct pci_ops *ops, void *sysdata,
+		struct pci_bus_resource *bus_res)
 {
 	int error;
 	struct pci_bus *b, *b2;
@@ -1570,8 +1571,14 @@ struct pci_bus * pci_create_bus(struct device *parent,
 	pci_create_legacy_files(b);
 
 	b->number = b->secondary = bus;
-	b->resource[0] = &ioport_resource;
-	b->resource[1] = &iomem_resource;
+
+	/* Add initial resources to the bus */
+	if (bus_res != NULL) {
+		list_add_tail(&b->resources, &bus_res->list);
+	} else {
+		pci_bus_add_resource(b, &ioport_resource, 0);
+		pci_bus_add_resource(b, &iomem_resource, 0);
+	}
 
 	return b;
 
@@ -1592,7 +1599,7 @@ struct pci_bus * __devinit pci_scan_bus_parented(struct device *parent,
 {
 	struct pci_bus *b;
 
-	b = pci_create_bus(parent, bus, ops, sysdata);
+	b = pci_create_bus(parent, bus, ops, sysdata, NULL);
 	if (b)
 		b->subordinate = pci_scan_child_bus(b);
 	return b;
diff --git a/include/linux/pci.h b/include/linux/pci.h
index 8c230cb..5e1bacd 100644
--- a/include/linux/pci.h
+++ b/include/linux/pci.h
@@ -666,7 +666,8 @@ static inline struct pci_bus * __devinit pci_scan_bus(int bus, struct pci_ops *o
 	return root_bus;
 }
 struct pci_bus *pci_create_bus(struct device *parent, int bus,
-			       struct pci_ops *ops, void *sysdata);
+			       struct pci_ops *ops, void *sysdata,
+			       struct pci_bus_resource *bus_res);
 struct pci_bus *pci_add_new_bus(struct pci_bus *parent, struct pci_dev *dev,
 				int busnr);
 void pcie_update_link_speed(struct pci_bus *bus, u16 link_status);
-- 
1.7.1

Re: [RFC PATCH 2/3] PCI: Pass available resources into pci_create_bus()

[Bjorn Helgaas]

On Wed, Aug 24, 2011 at 12:24 AM, Deng-Cheng Zhu
<dengcheng.zhu@gmail.com> wrote:
> Currently, after pci_create_bus(), resources available on the bus could be
> handled by pci_scan_child_bus(). The problem is that, in
> pci_scan_child_bus(), before calling arch-dependent pcibios_fixup_bus(),
> PCI quirks will probably conflict (while doing pci_claim_resource()) with
> resources like system controller's I/O resource that have not yet been
> added to the bus. So, add those resources right before returning the newly
> created bus in pci_create_bus().

I like this approach a lot.  Thanks for working it up.  It's a nice
small change with very little impact to other architectures, and you
have a nice clear changelog.  You might mention something about the
fact that by default, the bus starts out with all of ioport_resource
and iomem_resource -- that will mean something to people who know how
host bridges work.

> diff --git a/drivers/pci/probe.c b/drivers/pci/probe.c
> index 8473727..7735fe7 100644
> --- a/drivers/pci/probe.c
> +++ b/drivers/pci/probe.c
> @@ -1516,7 +1516,8 @@ unsigned int __devinit pci_scan_child_bus(struct pci_bus *bus)
>  }
>
>  struct pci_bus * pci_create_bus(struct device *parent,
> -               int bus, struct pci_ops *ops, void *sysdata)
> +               int bus, struct pci_ops *ops, void *sysdata,
> +               struct pci_bus_resource *bus_res)
>  {
>        int error;
>        struct pci_bus *b, *b2;
> @@ -1570,8 +1571,14 @@ struct pci_bus * pci_create_bus(struct device *parent,
>        pci_create_legacy_files(b);
>
>        b->number = b->secondary = bus;
> -       b->resource[0] = &ioport_resource;
> -       b->resource[1] = &iomem_resource;
> +
> +       /* Add initial resources to the bus */
> +       if (bus_res != NULL) {
> +               list_add_tail(&b->resources, &bus_res->list);
> +       } else {
> +               pci_bus_add_resource(b, &ioport_resource, 0);
> +               pci_bus_add_resource(b, &iomem_resource, 0);
> +       }

Using pci_bus_add_resource() here *seems* like it should be the right
thing, but I don't think it will work correctly.

The problem is that struct pci_bus has both a table of resources
(bus->resource[]) *and* a list (bus->resources).
pci_bus_add_resource() always puts the new resource on the list, but
various arch code still references the table directly, e.g., sparc has
"pbus->resource[0] = &pbm->io_space" in pcibios_fixup_bus().

As written, I think this patch will break sparc because the host
bridge will end up with both pbm->io_space (in the table) and
ioport_resource (in the list).

I think something like this would work, though:

    if (bus_res)
        list_add_tail(&b->resources, &bus_res->list);
    else {
        b->resource[0] = &ioport_resource;
        b->resource[1] = &iomem_resource;
    }

Bjorn

Re: [RFC PATCH 2/3] PCI: Pass available resources into pci_create_bus()

[Deng-Cheng Zhu]

2011/8/24 Bjorn Helgaas <bhelgaas@google.com>:
> I like this approach a lot.  Thanks for working it up.  It's a nice
> small change with very little impact to other architectures, and you
> have a nice clear changelog.  You might mention something about the
> fact that by default, the bus starts out with all of ioport_resource
> and iomem_resource -- that will mean something to people who know how
> host bridges work.

Thanks! And I'll add this info to the patch description.

> Using pci_bus_add_resource() here *seems* like it should be the right
> thing, but I don't think it will work correctly.
>
> The problem is that struct pci_bus has both a table of resources
> (bus->resource[]) *and* a list (bus->resources).
> pci_bus_add_resource() always puts the new resource on the list, but
> various arch code still references the table directly, e.g., sparc has
> "pbus->resource[0] = &pbm->io_space" in pcibios_fixup_bus().
>
> As written, I think this patch will break sparc because the host
> bridge will end up with both pbm->io_space (in the table) and
> ioport_resource (in the list).

Good catch! I overlooked this point.

> I think something like this would work, though:
>
>    if (bus_res)
>        list_add_tail(&b->resources, &bus_res->list);
>    else {
>        b->resource[0] = &ioport_resource;
>        b->resource[1] = &iomem_resource;
>    }

Yes, it should work.


Thanks again for your time,

Deng-Cheng

[RFC PATCH 3/3] MIPS: PCI: Pass controller's resources to pci_create_bus() in pcibios_scanbus()

[Deng-Cheng Zhu]

Use the new interface of pci_create_bus() so that system controller's
resources are added to the root bus upon bus creation, thereby avoiding
conflicts with PCI quirks before pcibios_fixup_bus() gets the chance to do
right things in pci_scan_child_bus().

Signed-off-by: Deng-Cheng Zhu <dengcheng.zhu@gmail.com>
---
 arch/mips/pci/pci.c |   38 +++++++++++++++++++++++++++++++++++++-
 1 files changed, 37 insertions(+), 1 deletions(-)

diff --git a/arch/mips/pci/pci.c b/arch/mips/pci/pci.c
index 7473214..a5ff6bc 100644
--- a/arch/mips/pci/pci.c
+++ b/arch/mips/pci/pci.c
@@ -76,11 +76,41 @@ pcibios_align_resource(void *data, const struct resource *res,
 	return start;
 }
 
+static struct pci_bus_resource *
+controller_resources(const struct pci_controller *ctrl)
+{
+	struct pci_bus_resource *mem_res, *io_res;
+
+	mem_res = kzalloc(sizeof(struct pci_bus_resource), GFP_KERNEL);
+	if (!mem_res)
+		goto err_out;
+
+	mem_res->res = ctrl->mem_resource;
+	mem_res->flags = 0;
+	INIT_LIST_HEAD(&mem_res->list);
+
+	io_res = kzalloc(sizeof(struct pci_bus_resource), GFP_KERNEL);
+	if (!io_res) {
+		kfree(mem_res);
+		goto err_out;
+	}
+
+	io_res->res = ctrl->io_resource;
+	io_res->flags = 0;
+	list_add(&io_res->list, &mem_res->list);
+
+	return mem_res;
+err_out:
+	printk(KERN_ERR "Can't allocate PCI bus resource.\n");
+	return NULL;
+}
+
 static void __devinit pcibios_scanbus(struct pci_controller *hose)
 {
 	static int next_busno;
 	static int need_domain_info;
 	struct pci_bus *bus;
+	struct pci_bus_resource *bus_res;
 
 	if (!hose->iommu)
 		PCI_DMA_BUS_IS_PHYS = 1;
@@ -88,7 +118,13 @@ static void __devinit pcibios_scanbus(struct pci_controller *hose)
 	if (hose->get_busno && pci_probe_only)
 		next_busno = (*hose->get_busno)();
 
-	bus = pci_scan_bus(next_busno, hose->pci_ops, hose);
+	bus_res = controller_resources(hose);
+	bus = pci_create_bus(NULL, next_busno, hose->pci_ops, hose, bus_res);
+	if (bus) {
+		bus->subordinate = pci_scan_child_bus(bus);
+		pci_bus_add_devices(bus);
+	}
+
 	hose->bus = bus;
 
 	need_domain_info = need_domain_info || hose->index;
-- 
1.7.1