On 2015/6/10 0:12, Lorenzo Pieralisi wrote:

On Mon, Jun 08, 2015 at 05:20:46PM +0100, Jiang Liu wrote:

[...]

quoted

+static int acpi_pci_probe_root_resources(struct acpi_pci_root_info *info)
+{
+	int ret;
+	struct list_head *list = &info->resources;
+	struct acpi_device *device = info->bridge;
+	struct resource_entry *entry, *tmp;
+	unsigned long flags;
+
+	flags = IORESOURCE_IO | IORESOURCE_MEM | IORESOURCE_MEM_8AND16BIT;

Is IORESOURCE_MEM_8AND16BIT required because of some pending patches
that will change ACPI resource filtering ? It does not seem to make
a difference in the mainline code, AFAICT.

Hi Lorenzo,
	Sorry, the 'IORESOURCE_MEM_8AND16BIT' is leaked into this patch
from another bugfix patch. It should be removed.

quoted

+static void acpi_pci_root_release_info(struct pci_host_bridge *bridge)
+{
+	struct resource *res;
+	struct resource_entry *entry;
+
+	resource_list_for_each_entry(entry, &bridge->windows) {
+		res = entry->res;
+		if (res->parent &&
+		    (res->flags & (IORESOURCE_MEM | IORESOURCE_IO)))
+			release_resource(res);
+	}

It is a question: is this loop necessary given that we are already
releasing resources in __acpi_pci_root_release_info() ?

Function pci_create_root_bus() moves resources from info->resources list
onto bridge->windows list, so an ACPI resource will be either on
info->resources or bridge->windows. Thus we need to deal with both
info->resources and bridge->windows.

quoted

+	__acpi_pci_root_release_info(bridge->release_data);
+}
+
+struct pci_bus *acpi_pci_root_create(struct acpi_pci_root *root,
+				     struct acpi_pci_root_ops *ops,
+				     struct acpi_pci_root_info *info,
+				     void *sysdata, int segment, int node)

I do not think you need to pass segment and node, they clutter the
function signature when you can retrieve them from root, I would
make them local variables and use root->segment and acpi_get_node
in the function body to retrieve them.

On x86, node and segment may be overridden under certain conditions.
For example, segment will always be 0 if 'pci_ignore_seg' is set.

quoted

+{
+	int ret, busnum = root->secondary.start;
+	struct acpi_device *device = root->device;
+	struct pci_bus *bus;
+
+	info->root = root;
+	info->bridge = device;
+	info->ops = ops;
+	INIT_LIST_HEAD(&info->resources);
+	snprintf(info->name, sizeof(info->name), "PCI Bus %04x:%02x",
+		 segment, busnum);
+
+	if (ops->init_info && ops->init_info(info))
+		goto out_release_info;
+	ret = acpi_pci_probe_root_resources(info);
+	if (ops->prepare_resources)
+		ret = ops->prepare_resources(info, ret);

You go through this ret passing song and dance because we may want to
call prepare_resources even if acpi_pci_probe_root_resource failed (on
x86), correct ? I will have a further look at x86 and ia64 if we
can consolidate these ops function hooks even further.

Yes. X86 uses flag 'pci_use_crs' to choose ACPI parsed resources or
other method parsed resources. This provides user a way to work
around some bios issues.
Thanks!
Gerry