[Trim down CC list, otherwise lkml will drop it, Thanks for David Miller to bring attention to me ] On Sat, Feb 2, 2013 at 1:50 PM, Bjorn Helgaas <bhelgaas@xxxxxxxxxx> wrote: > On Sun, Jan 27, 2013 at 12:23 PM, Yinghai Lu <yinghai@xxxxxxxxxx> wrote: >> Now we have pci_root_buses list, and there is lots of iteration with >> list_of_each of it, that is not safe after we add pci root bus hotplug >> support after booting stage. >> >> Add pci_get_next_host_bridge and use bus_find_device in driver core to >> iterate host bridge and the same time get root bus. >> >> We replace searching root bus with searching host_bridge, >> as host_bridge->bus is the root bus. >> After those replacing, we even could kill pci_root_buses list. > > These are the problems I think you're fixing: > > 1) pci_find_next_bus() is not safe because even though it holds > pci_bus_sem while walking the pci_root_buses list, it doesn't hold a > reference on the bus it returns. The bus could be removed while the > caller is using it. > > 2) "list_for_each_entry(bus, &pci_root_buses, node)" is not safe > because hotplug might modify the pci_root_buses list. Replacing that > with for_each_pci_host_bridge() solves that problem by using > bus_find_device(), which is built on klists, which are designed for > that problem. > > 3) pci_find_next_bus() claims to iterate through all known PCI buses, > but in fact only iterates through root buses. > > So far, so good. Those are problems we need to fix. > > Your solution is to introduce for_each_pci_host_bridge() as an > iterator through the known host bridges. There are two scenarios > where we use something like this: > > 1) We want to perform an operation on every known host bridge. > > 2) We want to initialize something for every host bridge. > > In my opinion, the only instance of scenario 1) is bus_rescan_store(), > where we want to rescan all PCI host bridges. > > In every other case, we're doing some kind of initialization of all > the host bridges. For these cases, for_each_pci_host_bridge() is the > wrong solution because it doesn't work for hot-added bridges. I think > these cases should be changed to use pcibios_root_bridge_prepare() or > something something else called in the host bridge add path. Yes, will check if those for_pci_host_bridge could be converted to adding calling in pcibios_root_bridge_prepare one by one. Thanks Yinghai -- 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
