On Mon, May 19, 2025 at 09:05:57AM -0500, Bjorn Helgaas wrote: > On Mon, May 05, 2025 at 01:39:39PM -0400, Jim Quinlan wrote: > > Hello, > > > > I recently rebased to the latest Linux master > > > > ebd297a2affa Linus.Torvalds Merge tag 'net-6.15-rc5' of > > git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net > > > > and noticed that PCI is broken for > > "drivers/pci/controller/pcie-brcmstb.c" I've bisected this to the > > following commit > > > > 2489eeb777af PCI/pwrctrl: Skip scanning for the device further if > > pwrctrl device is created > > > > which is part of the series [1]. The driver in pcie-brcmstb.c is > > expecting the add_bus() method to be invoked twice per boot-up, but > > the second call does not happen. Not only does this code in > > brcm_pcie_add_bus() turn on regulators, it also subsequently initiates > > PCIe linkup. > > > > If I revert the aforementioned commit, all is well. > > > > FWIW, I have included the relevant sections of the PCIe DT we use at [2]. > > Mani, Bartosz, where are we at with this? The 2489eeb777af commit log > doesn't mention a problem fixed by that commit; it sounds more like an > optimization -- just avoiding an unnecessary scan. > > 2489eeb777af appeared in v6.15-rc1, so there's still time to revert it > before v6.15 if that's the right way to fix this regression. > We need to find out what is happening in the pcie-brcmstb driver first. From Jim's report, it looks like the driver expects add_bus() callback to be invoked twice, which seems weird. If the fix takes time, then we can revert 2489eeb777af in the meantime. - Mani -- மணிவண்ணன் சதாசிவம்
