Skip to contentSkip to conversation

PCI: use pci_is_root_bus() to check whether it is a root bus

PATCH In review Wei Yang 4 participants 14 messages Sep 6, 2013 – Sep 11, 2013 1 message not in the archive PCI

lore
Share & tools
lore

Wei Yang

author
#0 lore raw

PATCH 1/3 PCI: use pci_is_root_bus() to check whether it is a root bus

In __pci_bus_size_bridges() we check whether a pci bus is a root bus by testing bus->self. As indicated by commit 79af72d7 ("PCI: pci_is_root_bus helper"), bus->self == NULL is not a proper way to check the pci root bus.

This patch changes it to pci_is_root_bus() to check whether it is a root bus.

---
 drivers/pci/setup-bus.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
drivers/pci/setup-bus.c +1 -1
index 520210f..989de3c 100644
@@ -1134,7 +1134,7 @@ void __ref __pci_bus_size_bridges(struct pci_bus *bus,
 	}
 
 	/* The root bus? */
-	if (!bus->self)
+	if (pci_is_root_bus(bus))
 		return;
 
 	switch (bus->self->class >> 8) {
Signature · 1 line
1.7.1
author
#1 lore raw

PATCH 2/3 PCI: Use spec name for the comment of PCIe capability field

According to the PCIe specification, bit 7:4 of PCI Express Capabilities Register in a PCI Express Capability Structure is used to identify the Device/Port type. If this field equals to 0x7, this PCIe device is a PCI Express to PCI/PCI-X Bridge. While the comment of this value in the code does not comply with the specification.

This patch changes the comment to "PCIE to PCI/PCI-X Bridge" instead of "PCI/PCI-X Bridge", which comply with the specification.

---
 include/uapi/linux/pci_regs.h |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
include/uapi/linux/pci_regs.h +1 -1
index c3cc01d..b82b2ff 100644
@@ -421,7 +421,7 @@
 #define  PCI_EXP_TYPE_ROOT_PORT 0x4	/* Root Port */
 #define  PCI_EXP_TYPE_UPSTREAM	0x5	/* Upstream Port */
 #define  PCI_EXP_TYPE_DOWNSTREAM 0x6	/* Downstream Port */
-#define  PCI_EXP_TYPE_PCI_BRIDGE 0x7	/* PCI/PCI-X Bridge */
+#define  PCI_EXP_TYPE_PCI_BRIDGE 0x7	/* PCIE to PCI/PCI-X Bridge */
 #define  PCI_EXP_TYPE_PCIE_BRIDGE 0x8	/* PCI/PCI-X to PCIE Bridge */
 #define  PCI_EXP_TYPE_RC_END	0x9	/* Root Complex Integrated Endpoint */
 #define  PCI_EXP_TYPE_RC_EC	0xa	/* Root Complex Event Collector */
Signature · 1 line
1.7.1
maintainer
#2 lore raw

Re: [PATCH 2/3] PCI: Use spec name for the comment of PCIe capability field

Quoted 30 lines · Wei Yang

According to the PCIe specification, bit 7:4 of PCI Express Capabilities Register in a PCI Express Capability Structure is used to identify the Device/Port type. If this field equals to 0x7, this PCIe device is a PCI Express to PCI/PCI-X Bridge. While the comment of this value in the code does not comply with the specification.

This patch changes the comment to "PCIE to PCI/PCI-X Bridge" instead of "PCI/PCI-X Bridge", which comply with the specification.

---
 include/uapi/linux/pci_regs.h |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
include/uapi/linux/pci_regs.h +1 -1
index c3cc01d..b82b2ff 100644
@@ -421,7 +421,7 @@
 #define  PCI_EXP_TYPE_ROOT_PORT 0x4	/* Root Port */
 #define  PCI_EXP_TYPE_UPSTREAM	0x5	/* Upstream Port */
 #define  PCI_EXP_TYPE_DOWNSTREAM 0x6	/* Downstream Port */
-#define  PCI_EXP_TYPE_PCI_BRIDGE 0x7	/* PCI/PCI-X Bridge */
+#define  PCI_EXP_TYPE_PCI_BRIDGE 0x7	/* PCIE to PCI/PCI-X Bridge */
 #define  PCI_EXP_TYPE_PCIE_BRIDGE 0x8	/* PCI/PCI-X to PCIE Bridge */
 #define  PCI_EXP_TYPE_RC_END	0x9	/* Root Complex Integrated Endpoint */
 #define  PCI_EXP_TYPE_RC_EC	0xa	/* Root Complex Event Collector */
Signature · 1 line
1.7.1

A similar change is already in Linus' tree:

http://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/include/uapi/linux/pci_regs.h?id=fbf501c347b2eea8451a615bd823b6b91a1a8eed

I guess several of us found that comment confusing :)

author
#3 lore raw

Re: [PATCH 2/3] PCI: Use spec name for the comment of PCIe capability field

Quoted 34 lines · Bjorn Helgaas
Quoted 30 lines · Wei Yang

According to the PCIe specification, bit 7:4 of PCI Express Capabilities Register in a PCI Express Capability Structure is used to identify the Device/Port type. If this field equals to 0x7, this PCIe device is a PCI Express to PCI/PCI-X Bridge. While the comment of this value in the code does not comply with the specification.

This patch changes the comment to "PCIE to PCI/PCI-X Bridge" instead of "PCI/PCI-X Bridge", which comply with the specification.

---
 include/uapi/linux/pci_regs.h |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
include/uapi/linux/pci_regs.h +1 -1
index c3cc01d..b82b2ff 100644
@@ -421,7 +421,7 @@
 #define  PCI_EXP_TYPE_ROOT_PORT 0x4	/* Root Port */
 #define  PCI_EXP_TYPE_UPSTREAM	0x5	/* Upstream Port */
 #define  PCI_EXP_TYPE_DOWNSTREAM 0x6	/* Downstream Port */
-#define  PCI_EXP_TYPE_PCI_BRIDGE 0x7	/* PCI/PCI-X Bridge */
+#define  PCI_EXP_TYPE_PCI_BRIDGE 0x7	/* PCIE to PCI/PCI-X Bridge */
 #define  PCI_EXP_TYPE_PCIE_BRIDGE 0x8	/* PCI/PCI-X to PCIE Bridge */
 #define  PCI_EXP_TYPE_RC_END	0x9	/* Root Complex Integrated Endpoint */
 #define  PCI_EXP_TYPE_RC_EC	0xa	/* Root Complex Event Collector */
Signature · 1 line
1.7.1

A similar change is already in Linus' tree:

Quoted 3 lines

http://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/include/uapi/linux/pci_regs.h?id=fbf501c347b2eea8451a615bd823b6b91a1a8eed

I guess several of us found that comment confusing :)

Thanks, I didn't notice this :-)

Signature · 2 lines
Richard Yang
Help you, Help me
author
#4 lore raw

PATCH 3/3 PCI: Pass full info for window alignment

In reply to Wei Yang

When calculating the window_alignment(), type information like IORESOURCE_MEM and IORESOURCE_PREFETCH is not enough for some platform.

As on powernv platform, one prefetchable window could be IORESOURCE_MEM_64 or not. The platform will calculate the alignment based on this information.

This patch passes the full info for window alignment.

---
 drivers/pci/setup-bus.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
drivers/pci/setup-bus.c +1 -1
index 989de3c..8781eb1 100644
@@ -980,7 +980,7 @@ static int pbus_size_mem(struct pci_bus *bus, unsigned long mask,
 	}
 
 	min_align = calculate_mem_align(aligns, max_order);
-	min_align = max(min_align, window_alignment(bus, b_res->flags & mask));
+	min_align = max(min_align, window_alignment(bus, b_res->flags));
 	size0 = calculate_memsize(size, min_size, 0, resource_size(b_res), min_align);
 	if (children_add_size > add_size)
 		add_size = children_add_size;
Signature · 1 line
1.7.1
maintainer
#5 lore raw

Re: [PATCH 3/3] PCI: Pass full info for window alignment

[+cc Gavin]

Quoted 18 lines · Wei Yang

When calculating the window_alignment(), type information like IORESOURCE_MEM and IORESOURCE_PREFETCH is not enough for some platform.

As on powernv platform, one prefetchable window could be IORESOURCE_MEM_64 or not. The platform will calculate the alignment based on this information.

This patch passes the full info for window alignment.

---
 drivers/pci/setup-bus.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
drivers/pci/setup-bus.c +0 -0
index 989de3c..8781eb1 100644

drivers/pci/setup-bus.c static int pbus_size_mem(struct pci_bus *bus, unsigned long mask, · line 983

Quoted 5 lines
drivers/pci/setup-bus.c +1 -1
@@ -980,7 +980,7 @@ static int pbus_size_mem(struct pci_bus *bus, unsigned long mask,
 	}
 
 	min_align = calculate_mem_align(aligns, max_order);
-	min_align = max(min_align, window_alignment(bus, b_res->flags & mask));
+	min_align = max(min_align, window_alignment(bus, b_res->flags));

This makes sense to me, but since Gavin added this window_alignment() call recently with the mask, I'd like him to ack this before I apply it.

Quoted 6 lines
drivers/pci/setup-bus.c +0 -0
 	size0 = calculate_memsize(size, min_size, 0, resource_size(b_res), min_align);
 	if (children_add_size > add_size)
 		add_size = children_add_size;
Signature · 1 line
1.7.1
maintainer
#6 lore raw
In reply to Wei Yang

[+cc Kenji, Alex]

Quoted 7 lines · Wei Yang

In __pci_bus_size_bridges() we check whether a pci bus is a root bus by testing bus->self. As indicated by commit 79af72d7 ("PCI: pci_is_root_bus helper"), bus->self == NULL is not a proper way to check the pci root bus.

This patch changes it to pci_is_root_bus() to check whether it is a root bus.

I think this is a good change, even if only on the grounds of consistency.

Did you trip over a case where a root bus has bus->self != NULL? I'd like to know more details about the case where:

  (bus->parent == NULL) && (bus->self != NULL)

I'm sure that situation exists, or Kenji and Alex would not have made the change in 79af72d7, but I don't know the details.

I'd like to know the details so I can recognize similar problems elsewhere.

Quoted 21 lines
---
 drivers/pci/setup-bus.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
drivers/pci/setup-bus.c +1 -1
index 520210f..989de3c 100644
@@ -1134,7 +1134,7 @@ void __ref __pci_bus_size_bridges(struct pci_bus *bus,
 	}
 
 	/* The root bus? */
-	if (!bus->self)
+	if (pci_is_root_bus(bus))
 		return;
 
 	switch (bus->self->class >> 8) {
Signature · 1 line
1.7.1
author
#7 lore raw
Quoted 11 lines · Bjorn Helgaas

[+cc Kenji, Alex]

Quoted 7 lines · Wei Yang

In __pci_bus_size_bridges() we check whether a pci bus is a root bus by testing bus->self. As indicated by commit 79af72d7 ("PCI: pci_is_root_bus helper"), bus->self == NULL is not a proper way to check the pci root bus.

This patch changes it to pci_is_root_bus() to check whether it is a root bus.

Quoted 7 lines

I think this is a good change, even if only on the grounds of consistency.

Did you trip over a case where a root bus has bus->self != NULL? I'd like to know more details about the case where:

(bus->parent == NULL) && (bus->self != NULL)

Currently no. I believe the two platforms I have met, x86 and powerpc, don't have this status.

Quoted 3 lines

I'm sure that situation exists, or Kenji and Alex would not have made the change in 79af72d7, but I don't know the details.

Agree, willing to hear about some backgroups about the original patch.

Quoted 25 lines

I'd like to know the details so I can recognize similar problems elsewhere.

Quoted 21 lines
---
 drivers/pci/setup-bus.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
drivers/pci/setup-bus.c +1 -1
index 520210f..989de3c 100644
@@ -1134,7 +1134,7 @@ void __ref __pci_bus_size_bridges(struct pci_bus *bus,
 	}
 
 	/* The root bus? */
-	if (!bus->self)
+	if (pci_is_root_bus(bus))
 		return;
 
 	switch (bus->self->class >> 8) {
Signature · 1 line
1.7.1
Signature · 2 lines
Richard Yang
Help you, Help me
author
#8 lore raw
In reply to Bjorn Helgaas
Quoted 11 lines · Bjorn Helgaas

[+cc Kenji, Alex]

Quoted 7 lines · Wei Yang

In __pci_bus_size_bridges() we check whether a pci bus is a root bus by testing bus->self. As indicated by commit 79af72d7 ("PCI: pci_is_root_bus helper"), bus->self == NULL is not a proper way to check the pci root bus.

This patch changes it to pci_is_root_bus() to check whether it is a root bus.

Quoted 7 lines

I think this is a good change, even if only on the grounds of consistency.

Did you trip over a case where a root bus has bus->self != NULL? I'd like to know more details about the case where:

(bus->parent == NULL) && (bus->self != NULL)

I found one case that (bus->self == NULL) and this bus is not root bus.

Not sure, this case will meet your requirement?

Signature · 2 lines
Richard Yang
Help you, Help me
maintainer
#9 lore raw
Quoted 15 lines · Wei Yang
Quoted 14 lines · Bjorn Helgaas

[+cc Kenji, Alex]

Quoted 7 lines · Wei Yang

In __pci_bus_size_bridges() we check whether a pci bus is a root bus by testing bus->self. As indicated by commit 79af72d7 ("PCI: pci_is_root_bus helper"), bus->self == NULL is not a proper way to check the pci root bus.

This patch changes it to pci_is_root_bus() to check whether it is a root bus.

I think this is a good change, even if only on the grounds of consistency.

Quoted 8 lines
Quoted 4 lines

Did you trip over a case where a root bus has bus->self != NULL? I'd like to know more details about the case where:

(bus->parent == NULL) && (bus->self != NULL)

I found one case that (bus->self == NULL) and this bus is not root bus.

Not sure, this case will meet your requirement?

I'm definitely interested in that case as well. Can you include a complete "lspci -vv" output for the case where this happens?

I suspect this happens with SR-IOV when we create "virtual" buses, i.e., in virtfn_add_bus(). I think maybe I'll update the comment at pci_is_root_bus() with a note about why we want to use it instead of testing for "bus->self == NULL".

I think we still have other tree traversal issues related to those virtual buses, e.g., the ones I mentioned here [1]. But those are a separate problem.

Bjorn

[1] http://lkml.kernel.org/r/CAErSpo5sFfr=O-Pp=PyaxGauaEajaTr2aK-EQ_rTVUk1zyz8cA@mail.gmail.com

author
#10 lore raw
Quoted 7 lines · Bjorn Helgaas
Quoted 6 lines · Wei Yang
Quoted 5 lines · Bjorn Helgaas

[+cc Kenji, Alex]

Did you trip over a case where a root bus has bus->self != NULL? I'd like to know more details about the case where:

Quoted 8 lines
Quoted 5 lines
Quoted 1 line

(bus->parent == NULL) && (bus->self != NULL)

I found one case that (bus->self == NULL) and this bus is not root bus.

Not sure, this case will meet your requirement?

I'm definitely interested in that case as well. Can you include a complete "lspci -vv" output for the case where this happens?

I don't *see* it. I notice this during the code reading.

So yes, this happens when a virtual bus is created for SR-IOV.

Sounds this is the *only* case for (bus->self == NULl)?

Quoted 5 lines

I suspect this happens with SR-IOV when we create "virtual" buses, i.e., in virtfn_add_bus(). I think maybe I'll update the comment at pci_is_root_bus() with a note about why we want to use it instead of testing for "bus->self == NULL".

Quoted 7 lines

I think we still have other tree traversal issues related to those virtual buses, e.g., the ones I mentioned here [1]. But those are a separate problem.

Bjorn

[1] http://lkml.kernel.org/r/CAErSpo5sFfr=O-Pp=PyaxGauaEajaTr2aK-EQ_rTVUk1zyz8cA@mail.gmail.com

Here comes another problem I met recently.

On some platforms, like powernv, each pci-dev has a dev-tree node to represent it. The dev-tree node is set by pci_set_of_node() during pci_scan_device().

In pci_set_of_node(), it searchs for the node with the same devfn in its parent. Here comes the question, who should be the VF's parent. Would be a conflict for the devfn between PF and VF?

Signature · 2 lines
Richard Yang
Help you, Help me
#11 lore raw
Quoted 6 lines · Wei Yang

On some platforms, like powernv, each pci-dev has a dev-tree node to represent it. The dev-tree node is set by pci_set_of_node() during pci_scan_device().

In pci_set_of_node(), it searchs for the node with the same devfn in its parent. Here comes the question, who should be the VF's parent. Would be a conflict for the devfn between PF and VF?

I think it makes no sense to create OFW nodes for VFs under powernv anyway ...

(which is another reason why we need urgently to disconnect the EEH data from the device node ...)

Cheers,
Ben.

author
#12 lore raw
Quoted 4 lines · Benjamin Herrenschmidt
Quoted 3 lines · Wei Yang

On some platforms, like powernv, each pci-dev has a dev-tree node to represent it. The dev-tree node is set by pci_set_of_node() during pci_scan_device().

Quoted 8 lines
Quoted 3 lines

In pci_set_of_node(), it searchs for the node with the same devfn in its parent. Here comes the question, who should be the VF's parent. Would be a conflict for the devfn between PF and VF?

I think it makes no sense to create OFW nodes for VFs under powernv anyway ...

(which is another reason why we need urgently to disconnect the EEH data from the device node ...)

Ok, got it.

Quoted 4 lines

Cheers,
Ben.

Signature · 2 lines
Richard Yang
Help you, Help me

A message that is not in the archive (sent off-list or not yet received): CA+C+4OYxqgExkepUix5wU4VPRRUcnAFPQJcbUYcLHRcAmNgqgw@mail.gmail.com

#14 lore raw

Resending because the first mail got rejected by vger.

Quoted 18 lines · Bjorn Helgaas

[+cc Kenji, Alex]

Quoted 7 lines · Wei Yang

In __pci_bus_size_bridges() we check whether a pci bus is a root bus by testing bus->self. As indicated by commit 79af72d7 ("PCI: pci_is_root_bus helper"), bus->self == NULL is not a proper way to check the pci root bus.

This patch changes it to pci_is_root_bus() to check whether it is a root bus.

I think this is a good change, even if only on the grounds of consistency.

Did you trip over a case where a root bus has bus->self != NULL? I'd like to know more details about the case where:

Quoted 7 lines
  (bus->parent == NULL) && (bus->self != NULL)

I'm sure that situation exists, or Kenji and Alex would not have made the change in 79af72d7, but I don't know the details.

I'd like to know the details so I can recognize similar problems elsewhere.

It has been quite a long time, and so I had to google around in the archives to try and remind myself why we made this change. Unfortunately, I couldn't find anything public...

Reaching back into my unreliable human memory though, I truly seem to recall that this interface was created out of paranoia/future-proofing and not because we had tripped over any actual issues we had found in the wild at that time.

Sorry for not being more helpful.

/ac