[PATCH v2] PCI: iproc: Use the EROM outbound window on BCMA

[PATCH v2] PCI: iproc: Use the EROM outbound window on BCMA

[Semih Baskan]

The PCIe outbound window base on Northstar depends on the PCIe Gen2 core
revision. Revision 0x01 uses 0x08000000, 0x40000000 and 0x48000000 for
controllers 0 to 2, while revision 0x07 (NS-B0) uses 0x08000000,
0x20000000 and 0x28000000. Broadcom's own driver branches on the core
revision for exactly this reason.

bcm-ns.dtsi is shared by every Northstar SoC, so it cannot carry a value
that is correct on both. Commit 767012397976 ("ARM: dts: BCM5301X:
Describe PCIe controllers fully") gave the controllers a ranges property.
The commit shipped in v7.1.

With that property present, two things go wrong with this driver:

devm_pci_alloc_host_bridge() parses those ranges and requests them, then
this driver adds its own window and requests the whole list a second
time, so every controller fails to probe with -EBUSY.

When the DT window is used, it is only correct on core revision 0x07. On
revision 0x01 it points at an address the hardware does not decode, and
the first MMIO access to a BAR takes an imprecise external abort.

The enumeration ROM reports the correct base for the revision actually
present, and bcma already provides it as addr_s[0]. Drop any memory
window that came from the device tree and use that instead, requesting
only the window this driver owns. This makes the driver correct whether
or not the DT describes a window.

When a dropped window does not match what the EROM reports, print a
warning naming both. The mismatch means the devicetree describes a
window the hardware does not decode, and that should be fixed in the
dts rather than ignored silently.

The same commit also added compatible = "brcm,iproc-pcie", so these
nodes now match pcie-iproc-platform. With CONFIG_PCIE_IPROC_PLATFORM
enabled, which is the default on ARCH_BCM_IPROC, that driver binds them
first and this driver's probe fails inside devm_pci_alloc_host_bridge().
This patch fixes the configurations where the BCMA driver is the one in
use; OpenWrt builds that way, with PCIE_IPROC_PLATFORM disabled. The
platform path takes the DT window as-is and has the same wrong address
on core revision 0x01, so that side needs a devicetree fix either way.

Tested on an ASUS RT-N18U (BCM47081) and a Linksys EA9200 (BCM4709),
both core revision 0x01.

Fixes: 767012397976 ("ARM: dts: BCM5301X: Describe PCIe controllers fully")
Tested-by: Rani Hod <rani.hod@gmail.com>
Cc: stable@vger.kernel.org # v7.1+
Signed-off-by: Semih Baskan <strst.gs@gmail.com>
---
v1 -> v2: print a warning for every devicetree memory window that does
not match the EROM window. Requested by Arnd Bergmann:
https://lore.kernel.org/all/d05ffeca-f289-42dd-b454-5a7c7741c6d5@app.fastmail.com/

v1: https://lore.kernel.org/all/20260727140939.389-1-strst.gs@gmail.com/
Applies to v7.2-rc5 and unchanged through v7.2-rc6; the file has no
commits in between.

 drivers/pci/controller/pcie-iproc-bcma.c | 17 ++++++++++++++++-
 1 file changed, 16 insertions(+), 1 deletion(-)

diff --git a/drivers/pci/controller/pcie-iproc-bcma.c b/drivers/pci/controller/pcie-iproc-bcma.c
index 593418c2b..06a471f4a 100644
--- a/drivers/pci/controller/pcie-iproc-bcma.c
+++ b/drivers/pci/controller/pcie-iproc-bcma.c
@@ -36,6 +36,7 @@ static int iproc_bcma_pcie_probe(struct bcma_device *bdev)
 	struct device *dev = &bdev->dev;
 	struct iproc_pcie *pcie;
 	struct pci_host_bridge *bridge;
+	struct resource_entry *win, *tmp;
 	int ret;
 
 	bridge = devm_pci_alloc_host_bridge(dev, sizeof(*pcie));
@@ -59,8 +60,22 @@ static int iproc_bcma_pcie_probe(struct bcma_device *bdev)
 	pcie->mem.end = bdev->addr_s[0] + SZ_128M - 1;
 	pcie->mem.name = "PCIe MEM space";
 	pcie->mem.flags = IORESOURCE_MEM;
+
+	resource_list_for_each_entry_safe(win, tmp, &bridge->windows) {
+		if (resource_type(win->res) != IORESOURCE_MEM)
+			continue;
+
+		if (win->res->start != pcie->mem.start ||
+		    win->res->end != pcie->mem.end)
+			dev_warn(dev, "DT window %pR does not match EROM window %pR, using EROM\n",
+				 win->res, &pcie->mem);
+
+		devm_release_resource(dev, win->res);
+		resource_list_destroy_entry(win);
+	}
+
 	pci_add_resource(&bridge->windows, &pcie->mem);
-	ret = devm_request_pci_bus_resources(dev, &bridge->windows);
+	ret = devm_request_resource(dev, &iomem_resource, &pcie->mem);
 	if (ret)
 		return ret;
 
-- 
2.43.0

Re: [PATCH v2] PCI: iproc: Use the EROM outbound window on BCMA

[Bjorn Helgaas]

On Fri, Aug 07, 2026 at 06:57:26AM +0300, Semih Baskan wrote:
> The PCIe outbound window base on Northstar depends on the PCIe Gen2 core
> revision. Revision 0x01 uses 0x08000000, 0x40000000 and 0x48000000 for
> controllers 0 to 2, while revision 0x07 (NS-B0) uses 0x08000000,
> 0x20000000 and 0x28000000. Broadcom's own driver branches on the core
> revision for exactly this reason.
> 
> bcm-ns.dtsi is shared by every Northstar SoC, so it cannot carry a value
> that is correct on both. Commit 767012397976 ("ARM: dts: BCM5301X:
> Describe PCIe controllers fully") gave the controllers a ranges property.
> The commit shipped in v7.1.
> 
> With that property present, two things go wrong with this driver:

It looks like you're going to present two problems next.  They should
be bullets so it's obvious what the first is and what the second is,
e.g.,

  - devm_pci_alloc_host_bridge() parses ...

  - When the DT window ...

(or whatever the second item is, I'm not sure where it starts)

> devm_pci_alloc_host_bridge() parses those ranges and requests them, then
> this driver adds its own window and requests the whole list a second
> time, so every controller fails to probe with -EBUSY.
> 
> When the DT window is used, it is only correct on core revision 0x07. On
> revision 0x01 it points at an address the hardware does not decode, and
> the first MMIO access to a BAR takes an imprecise external abort.
> 
> The enumeration ROM reports the correct base for the revision actually
> present, and bcma already provides it as addr_s[0]. Drop any memory
> window that came from the device tree and use that instead, requesting
> only the window this driver owns. This makes the driver correct whether
> or not the DT describes a window.
> 
> When a dropped window does not match what the EROM reports, print a
> warning naming both. The mismatch means the devicetree describes a
> window the hardware does not decode, and that should be fixed in the
> dts rather than ignored silently.
> 
> The same commit also added compatible = "brcm,iproc-pcie", so these
> nodes now match pcie-iproc-platform. With CONFIG_PCIE_IPROC_PLATFORM
> enabled, which is the default on ARCH_BCM_IPROC, that driver binds them
> first and this driver's probe fails inside devm_pci_alloc_host_bridge().
> This patch fixes the configurations where the BCMA driver is the one in
> use; OpenWrt builds that way, with PCIE_IPROC_PLATFORM disabled. The
> platform path takes the DT window as-is and has the same wrong address
> on core revision 0x01, so that side needs a devicetree fix either way.
> 
> Tested on an ASUS RT-N18U (BCM47081) and a Linksys EA9200 (BCM4709),
> both core revision 0x01.
> 
> Fixes: 767012397976 ("ARM: dts: BCM5301X: Describe PCIe controllers fully")
> Tested-by: Rani Hod <rani.hod@gmail.com>
> Cc: stable@vger.kernel.org # v7.1+
> Signed-off-by: Semih Baskan <strst.gs@gmail.com>
> ---
> v1 -> v2: print a warning for every devicetree memory window that does
> not match the EROM window. Requested by Arnd Bergmann:
> https://lore.kernel.org/all/d05ffeca-f289-42dd-b454-5a7c7741c6d5@app.fastmail.com/
> 
> v1: https://lore.kernel.org/all/20260727140939.389-1-strst.gs@gmail.com/
> Applies to v7.2-rc5 and unchanged through v7.2-rc6; the file has no
> commits in between.

FWIW, there's no need to rebase for new -rcs.  We always apply PCI
patches on -rc1 anyway.