Hello Lorenzo, Keith, Bjorn These are VMD fixes for 4.20 First one looks to fix an orphaned struct resource issue when removing the VMD module Second patch enables firmware-first error handling on a VMD domain. Because the endpoint is somewhat divorced from the VMD domain, the driver doesn't use a standard ACPI method of signaling firmware-first on the domain. The idea behind firmware-first in VMD is that the endpoint will advertise its support using the interface bit in the endpoint. The driver enables SMI system error messages in the root port control register. The firmware is allowed to handle the interrupt as it sees fit, but returns control to the VMD driver by synthesizing an MSI message. The kernel native AER driver may be used for further processing, so is left enabled. If the firmware doesn't need the native driver, of course it can handle and clear all errors and the subsequent MSI will be eventually dropped as it goes through the VMD ISR. At this time, I'm unaware of how the errors will be represented in the error status headers. In the past they could be seen as a combination of the VMD endpoint device's bus number and the erroring device's bus number, but could be different depending on the type of error and if the erroring endpoint was lost (completion timeout). Currently the native driver handles these errors by scanning every device on the bus for error status and there's no plan to change that mechanism for future VMD devices. Follow-up to: https://patchwork.ozlabs.org/patch/966128/ https://patchwork.ozlabs.org/patch/964254/ Jon Derrick (2): PCI/VMD: Detach resources after stopping root bus PCI/VMD: Set up firmware-first if capable drivers/pci/controller/vmd.c | 32 +++++++++++++++++++++++++++++++- 1 file changed, 31 insertions(+), 1 deletion(-) -- 1.8.3.1
