Hi Sinan, On Sat, Jan 21, 2017 at 08:35:30PM -0500, Sinan Kaya wrote: > When I boot the system, I see the ASPM is working fine. After hotplug removal > and insertion, I see that the ASPM is disabled. > > I do not have any ASPM related kernel command line parameter and > the policy is default. > > ASPM gets configured by the firmware before the OS boots by default > unless powersave policy is passed to the kernel command line. > > Is there a particular reason why we are not re-configuring ASPM after > hotplug insertion? > > I think default policy applies to power-up values but not for > the hotplug case. > > With similar reasoning, maximum payload size gets configured by > the firmware before boot. I see that the OS is re-configuring the > MPS settings following hotplug. > > I feel like ASPM should have followed a similar policy. Huh, I was just wondering whether ASPM would work on hot-added devices. I think the ASPM enumeration path is ugly because we only call it for bridges, not for endpoints (pci_scan_slot() calls pcie_aspm_init_link_state() for "bus->self"). I'd like it better if we called it for *every* device from pci_configure_device() and let ASPM internally figure out whether this is a bridge or endpoint. It looks like the intent is that pcie_aspm_init_link_state() allocates dev->link_state and configures ASPM on the link. The hotplug remove *should* call pcie_aspm_exit_link_state() on the bridge to free dev->link_state, and the hotplug add should call pcie_aspm_init_link_state() again to reallocate dev->link_state and configure ASPM again. But I'm not sure the remove path is correct. It looks like pci_stop_dev() calls pcie_aspm_exit_link_state(dev) on the device we're removing, not on the bridge. If that's the case, we won't free the dev->link_state, and when we call pcie_aspm_init_link_state() after the hotplug add, it will see that dev->link_state is already set and won't do anything. You could add some instrumentation to pcie_aspm_init_link_state() and pcie_aspm_exit_link_state() to see what devices we call them for, and whether they actually do what they should. Bjorn -- To unsubscribe from this list: send the line "unsubscribe linux-pci" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
