On Thu, 2025-05-22 at 18:19 +0200, Niklas Cassel wrote: [snip] > > I think the solution is to add a call to add a > pcie_bus_configure_settings() > call in pcie_do_recovery() / pci_host_recover_slots() / > pci_host_reset_slot() / > pcibios_reset_secondary_bus(). > > > Or possibly a: > list_for_each_entry(child, &bus->children, node) > pcie_bus_configure_settings(child); Hey Niklas, Thanks for writing this out in detail. Just to add to this, I have tested your suggestion, and it does work. That's is: ``` @@ -4990,6 +4991,8 @@ void __weak pcibios_reset_secondary_bus(struct pci_dev *dev) if (ret) pci_err(dev, "failed to reset slot: %d\n", ret); + list_for_each_entry(child, &dev->bus->children, node) + pcie_bus_configure_settings(child); return; } -- ``` To clarify, with this patch applied, between a Rock5B EP and Rock5B RC both CONFIG_PCIE_BUS_PERFORMANCE=y. When the host issues a sysfs initiated bus reset to the endpoint. Everything works as expected. Whereas before, DMA from EP to host would not work after the bus reset. Regards, Wilfred
