[+cc Lorenzo, beginning of thread: https://lore.kernel.org/r/20220127234917.GA150851@bhelgaas] On Thu, Feb 03, 2022 at 01:55:28PM +0100, Pali Rohár wrote: > On Thursday 03 February 2022 13:26:42 Pali Rohár wrote: > > On Thursday 27 January 2022 17:49:17 Bjorn Helgaas wrote: > > > On Thu, Jan 27, 2022 at 10:52:43PM +0000, bugzilla-daemon@xxxxxxxxxxxxxxxxxxx wrote: > > > > https://bugzilla.kernel.org/show_bug.cgi?id=215540 > > > > > > > > Bug ID: 215540 > > > > Summary: mvebu: no pcie devices detected on turris omnia > > > > (5.16.3 regression) > > > > Product: Drivers > > > > Version: 2.5 > > > > Kernel Version: 5.16.3 > > > > Hardware: ARM > > > > OS: Linux > > > > Tree: Mainline > > > > Status: NEW > > > > Severity: normal > > > > Priority: P1 > > > > Component: PCI > > > > Assignee: drivers_pci@xxxxxxxxxxxxxxxxxxxx > > > > Reporter: jpalus@xxxxxxxxxxxx > > > > Regression: No > > > > > > > > After kernel upgrade from 5.16.1 to 5.16.3 Turris Omnia (Armada 385) > > > > no longer detects pcie devices (wifi/msata). Haven't tried 5.16.2 > > > > but it doesn't seem to have any relevant changes, while 5.16.3 > > > > carries a few. > > I found another issue: Into stable tree was backported "modified" patch. > I'm not sure it is is source of this issue but looks like it is related. > > If you open mentioned problematic commit in web ui: > https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/?h=linux-5.16.y&id=7cde9bf0731688896831f90da9fe755f44a6d5e0 > > And compare it with patch which is on "Link:" line from commit message: > https://lore.kernel.org/r/20211125124605.25915-12-pali@xxxxxxxxxx > > You will see that diffs are different. In my original patch (which I > sent to ML) is: > > mvebu_pcie_setup_hw(port); > mvebu_pcie_set_local_dev_nr(port, 1); > + mvebu_pcie_set_local_bus_nr(port, 0); > > But in git web ui is: > > mvebu_pcie_setup_hw(port); > - mvebu_pcie_set_local_dev_nr(port, 1); > + mvebu_pcie_set_local_dev_nr(port, 0); > > I do not know how it could happen. But local **device** number must be > always set to 1 (see comment above code for explanation) and default > value of local **bus** number should be 0 (as is in my original patch). > > So above patch in stable tree is broken. I think current mainline is broken, too, isn't it? See below. > Bjorn & Greg: How do you want to handle this situation? Should I prepare > special patch for stable which fix it? Or something else? > > Anyway, do you know how it could happen that patch was incorrectly > auto-backported into stable? Differences between original and > wrongly-modified patch looks very similar (both "bus" and "dev" keywords > have same number of characters) and it was hard for me to see that there > are differences. So probably overlooking could happen or maybe git or > patch tools could do such small changes when doing backports? Your patch on the mailing list [1] contains: mvebu_pcie_setup_hw(port); mvebu_pcie_set_local_dev_nr(port, 1); + mvebu_pcie_set_local_bus_nr(port, 0); 91a8d79fc797 ("PCI: mvebu: Fix configuring secondary bus of PCIe Root Port via emulated bridge") [2] appeared in v5.17-rc1 and contains: mvebu_pcie_setup_hw(port); - mvebu_pcie_set_local_dev_nr(port, 1); + mvebu_pcie_set_local_dev_nr(port, 0); And this is the current state of mainline [3]. 91a8d79fc797 was backported to v5.16.3 as 7cde9bf07316 [4], which also contains: mvebu_pcie_setup_hw(port); - mvebu_pcie_set_local_dev_nr(port, 1); + mvebu_pcie_set_local_dev_nr(port, 0); So I think the problem was a merge error when we first applied this for mainline, and we just need to make a patch for mainline, apply it for v5.17, and mark it for stable. Bjorn [1] https://lore.kernel.org/r/20211125124605.25915-12-pali@xxxxxxxxxx [2] https://git.kernel.org/linus/91a8d79fc797 [3] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/pci/controller/pci-mvebu.c?id=v5.17-rc2#n1323 [4] https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/?h=linux-5.16.y&id=7cde9bf07316
