[Bug 285993] nvme device breakage in 14.2 STABLE n270867-25df691800f0

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=285993

--- Comment #11 from Warner Losh <imp@FreeBSD.org> ---
bus_child_present is a newbus dispatch method.
sys/dev/pci/pci_pci.c:    DEVMETHOD(bus_child_present,  pcib_child_present),
sys/dev/pccbb/pccbb_pci.c:      DEVMETHOD(bus_child_present,           
cbb_child_present),
is where it's implemented. Since there are no cardbus NVMe drive, the first one
is relevant.

int
pcib_child_present(device_t dev, device_t child)
{
#ifdef PCI_HP
        struct pcib_softc *sc = device_get_softc(dev);
        int retval;

        retval = bus_child_present(dev);
        if (retval != 0 && sc->flags & PCIB_HOTPLUG)
                retval = pcib_hotplug_present(sc);
        return (retval);
#else
        return (bus_child_present(dev));
#endif
}

which right away suggests you could work around this with a nooptions PCI_HP
kernel.

But that code says that if all my parents think the child is there, then check
its hotplug status:
/*
 * Returns -1 if the card is fully inserted, powered, and ready for
 * access.  Otherwise, returns 0.
 */
static int
pcib_hotplug_present(struct pcib_softc *sc)
{

        /* Card must be inserted. */
        if (!pcib_hotplug_inserted(sc))
                return (0);

        /* Require the Data Link Layer to be active. */
        if (!(sc->pcie_link_sta & PCIEM_LINK_STA_DL_ACTIVE))
                return (0);

        return (-1);
}

(yes, I checked, this is the right protocol to use).

So the next debugging point is to add code in pcib_hotplug_present to see which
of these conditions fails.

-- 
You are receiving this mail because:
You are the assignee for the bug.