According to the old PCIe spec, PDS would always have to come up with or before DLLLA. For various reasons, not all OEMs followed this requirement. As a result, in PCIe 4.0(*), there is a new way to disable in-band presence reporting, such that PDS always reports the status of the out-of-band presence (if implemented). For us, this means it is legal to get a Card Present event after the link is active, and the driver was loaded. This causes an erroneous removal of the device driver, followed by immediate re-probing. In a fully PCIe-compliant world, we could leave inband presence enabled and not require any software changes. However, certain OEMs simply disable inband presence without implementing the requisite query and control bits. I present two solutions to resolve this. The first and last patch in the series are required. Of the remaining two, only one is required to complete the series. Since we don't yet have a go/no-go on which method to use, both solutions are included: - 2/4: Try to wait for PDS _before_ loading the driver - 3/4: Load as usual, and recognize the delayed PDS event as such (*) ECN was approved in Nov 2018, and is normative spec text. A lot of the leaked PCIe 4.0 specs do not have this change. Changes since v2: * Dropped [RFC] from title * Addressed style issue found by Lukas Alexandru Gagniuc (4): PCI: hotplug: Add support for disabling in-band presence PCI: pciehp: Do not turn off slot if presence comes up after link PCI: hotplug: Wait for PDS when in-band presence is disabled PCI: hotplug: Add quirk For Dell nvme pcie switches drivers/pci/hotplug/pciehp.h | 1 + drivers/pci/hotplug/pciehp_ctrl.c | 24 ++++++++++++++ drivers/pci/hotplug/pciehp_hpc.c | 52 ++++++++++++++++++++++++++++++- include/linux/pci.h | 1 + include/uapi/linux/pci_regs.h | 2 ++ 5 files changed, 79 insertions(+), 1 deletion(-) -- 2.20.1
