On Fri, Jul 26, 2024 at 07:29:03PM +0530, Siddharth Vadapalli wrote:
Since the configuration of INTx (Legacy Interrupt) is not supported, set the .map_irq and .swizzle_irq callbacks to NULL. This fixes the error: of_irq_parse_pci: failed with rc=-22 when the pcieport driver attempts to setup INTx for the Host Bridge via "pci_assign_irq()". The device-tree node of the Host Bridge doesn't contain Legacy Interrupts. As a result, Legacy Interrupts are searched for in the MSI Interrupt Parent of the Host Bridge with the help of "of_irq_parse_raw()". Since the MSI Interrupt Parent of the Host Bridge uses 3 interrupt cells while Legacy Interrupts only use 1, the search for Legacy Interrupts is terminated prematurely by the "of_irq_parse_raw()" function with the -EINVAL error code (rc=-22).
Fixes: f3e25911a430 ("PCI: j721e: Add TI J721E PCIe driver") Cc: stable@vger.kernel.org Reported-by: Andrew Halaney ahalaney@redhat.com Tested-by: Andrew Halaney ahalaney@redhat.com Signed-off-by: Siddharth Vadapalli s-vadapalli@ti.com
TBH, I'm not comfortable with this change. Because, the INTx compatibility _should_ come from the platform description (in this case DT) and not the driver. The default map_irq() function is supposed to check the existence of INTx and correctly report it. In this case the issue is, the platform is not supporting INTx, but of_irq_parse_pci() reports a dubious error:
'of_irq_parse_pci: failed with rc=-22'
instead of the actual error:
`of_irq_parse_pci: no interrupt-map found, INTx interrupts not available'
So I would say that the fix is in of_irq_parse_pci() implementation. of_irq_parse_pci() is supposed to find whether the PCIe controller is supporting INTx or not by parsing the 'interrupt-{map/extended}' properties till host bridge/controller node. But this API along with of_irq_parse_raw(), just walks up the tree till the top level interrupt controller and checks for INTx, which just feels wrong (that's why you are getting -EINVAL because of #interrupt-cells mismatch).
I looked into the implementation over the weekend, but I'm not quite sure how to fix it though.
Rob, perhaps you have any idea?
- Mani
Hello,
This patch is based on commit 1722389b0d86 Merge tag 'net-6.11-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net of Mainline Linux.
v1: https://lore.kernel.org/r/20240724065048.285838-1-s-vadapalli@ti.com Changes since v1:
- Added "Cc: stable@vger.kernel.org" in the commit message.
- Based on Bjorn's feedback at: https://lore.kernel.org/r/20240724162304.GA802428@bhelgaas/ the $subject and commit message have been updated. Additionally, the comment in the driver has also been updated to specify "INTx" instead of "Legacy Interrupts".
- Collected Tested-by tag from Andrew Halaney ahalaney@redhat.com: https://lore.kernel.org/r/vj6vtjphpmqv6hcblaalr2m4bwjujjwvom6ca4bjdzcmgazyaa...
Patch has been tested on J721e-EVM and J784S4-EVM.
Regards, Siddharth.
drivers/pci/controller/cadence/pci-j721e.c | 4 ++++ 1 file changed, 4 insertions(+)
diff --git a/drivers/pci/controller/cadence/pci-j721e.c b/drivers/pci/controller/cadence/pci-j721e.c index 85718246016b..eaa6cfeb03c7 100644 --- a/drivers/pci/controller/cadence/pci-j721e.c +++ b/drivers/pci/controller/cadence/pci-j721e.c @@ -417,6 +417,10 @@ static int j721e_pcie_probe(struct platform_device *pdev) if (!bridge) return -ENOMEM;
/* INTx is not supported */bridge->map_irq = NULL;bridge->swizzle_irq = NULL;- if (!data->byte_access_allowed) bridge->ops = &cdns_ti_pcie_host_ops; rc = pci_host_bridge_priv(bridge);
-- 2.40.1