On 2016/9/13 16:24, Lorenzo Pieralisi wrote:

On Tue, Sep 13, 2016 at 04:15:31PM +0800, Hanjun Guo wrote:

[...]

quoted

quoted

quoted

+static acpi_status __init iort_match_iommu_callback(struct
acpi_iort_node *node,
+                            void *context)
+{
+    int ret;
+    struct fwnode_handle *fwnode;
+
+    fwnode = iort_get_fwnode(node);
+
+    if (!fwnode)
+        return AE_NOT_FOUND;
+
+    ret = iort_add_smmu_platform_device(fwnode, node);
+    if (ret) {
+        pr_err("Error in platform device creation\n");
+        return AE_ERROR;
+    }
+
+    return AE_OK;
+}
+
+static void __init iort_smmu_init(void)
+{
+    iort_scan_node(ACPI_IORT_NODE_SMMU, iort_match_iommu_callback,
NULL);
+    iort_scan_node(ACPI_IORT_NODE_SMMU_V3, iort_match_iommu_callback,
NULL);

Since iort_scan_node() returns after the first successful match it finds,
only the first SMMU_V3 in my IORT is being enumerated. I think you need
to go back to the "iterator" like approach you had been using or make
iort_match_iommu_callback() always return a non-AE_OK value so the scan
continues and has a chance to visit all of the SMMU_V3 nodes.

Please use the updated version of IORT patch (aka Tomasz's v11)
then things will work fine.

Nate is right, I was too keen on using iort_scan_node(), it does
not really work here (unless as he said I return a value !AE_OK in
the callback, which is horrible), I reverted back to the iterator
approach and I can push out a fixed up branch if useful before next
posting.

Ah, sorry, I just noticed "the first SMMU_V3" which is pretty similar
with the second problem which is noticed by Nate...

Thanks
Hanjun