SiFive FU740 PCI driver fails on 6.18-rc1

[Bjorn Helgaas]

[+cc FU740 driver folks, Conor, regressions]

On Mon, Oct 13, 2025 at 12:14:54AM -0700, Ron Economos wrote:
> The SiFive FU740 PCI driver fails on the HiFive Unmatched board with Linux
> 6.18-rc1. The error message is:
> 
> [    3.166624] fu740-pcie e00000000.pcie: host bridge /soc/pcie at e00000000
> ranges:
> [    3.166706] fu740-pcie e00000000.pcie:       IO
> 0x0060080000..0x006008ffff -> 0x0060080000
> [    3.166767] fu740-pcie e00000000.pcie:      MEM
> 0x0060090000..0x007fffffff -> 0x0060090000
> [    3.166805] fu740-pcie e00000000.pcie:      MEM
> 0x2000000000..0x3fffffffff -> 0x2000000000
> [    3.166950] fu740-pcie e00000000.pcie: ECAM at [mem
> 0xdf0000000-0xdffffffff] for [bus 00-ff]
> [    3.579500] fu740-pcie e00000000.pcie: No iATU regions found
> [    3.579552] fu740-pcie e00000000.pcie: Failed to configure iATU in ECAM
> mode
> [    3.579655] fu740-pcie e00000000.pcie: probe with driver fu740-pcie
> failed with error -22
> 
> The normal message (on Linux 6.17.2) is:
> 
> [    3.381487] fu740-pcie e00000000.pcie: host bridge /soc/pcie at e00000000
> ranges:
> [    3.381584] fu740-pcie e00000000.pcie:       IO
> 0x0060080000..0x006008ffff -> 0x0060080000
> [    3.381682] fu740-pcie e00000000.pcie:      MEM
> 0x0060090000..0x007fffffff -> 0x0060090000
> [    3.381724] fu740-pcie e00000000.pcie:      MEM
> 0x2000000000..0x3fffffffff -> 0x2000000000
> [    3.484809] fu740-pcie e00000000.pcie: iATU: unroll T, 8 ob, 8 ib, align
> 4K, limit 4096G
> [    3.683678] fu740-pcie e00000000.pcie: PCIe Gen.1 x8 link up
> [    3.883674] fu740-pcie e00000000.pcie: PCIe Gen.3 x8 link up
> [    3.987678] fu740-pcie e00000000.pcie: PCIe Gen.3 x8 link up
> [    3.988164] fu740-pcie e00000000.pcie: PCI host bridge to bus 0000:00
> 
> Reverting the following commits solves the issue.
> 
> 0da48c5b2fa731b21bc523c82d927399a1e508b0 PCI: dwc: Support ECAM mechanism by
> enabling iATU 'CFG Shift Feature'
> 
> 4660e50cf81800f82eeecf743ad1e3e97ab72190 PCI: qcom: Prepare for the DWC ECAM
> enablement
> 
> f6fd357f7afbeb34a633e5688a23b9d7eb49d558 PCI: dwc: Prepare the driver for
> enabling ECAM mechanism using iATU 'CFG Shift Feature'

As Conor pointed out, we can't fix a code regression with a DT change.

#regzbot introduced: f6fd357f7afb ("PCI: dwc: Prepare the driver for enabling ECAM mechanism using iATU 'CFG Shift Feature'")