--- /dev/null
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Virtual I/O topology
+ *
+ * The Virtual I/O Translation Table (VIOT) describes the topology of
+ * para-virtual IOMMUs and the endpoints they manage. The OS uses it to
+ * initialize devices in the right order, preventing endpoints from issuing DMA
+ * before their IOMMU is ready.
+ *
+ * When binding a driver to a device, before calling the device driver's probe()
+ * method, the driver infrastructure calls dma_configure(). At that point the
+ * VIOT driver looks for an IOMMU associated to the device in the VIOT table.
+ * If an IOMMU exists and has been initialized, the VIOT driver initializes the
+ * device's IOMMU fwspec, allowing the DMA infrastructure to invoke the IOMMU
+ * ops when the device driver configures DMA mappings. If an IOMMU exists and
+ * hasn't yet been initialized, VIOT returns -EPROBE_DEFER to postpone probing
+ * the device until the IOMMU is available.
+ */
+#define pr_fmt(fmt) "ACPI: VIOT: " fmt
+
+#include <linux/acpi_viot.h>
+#include <linux/dma-iommu.h>
+#include <linux/fwnode.h>
+#include <linux/iommu.h>
+#include <linux/list.h>
+#include <linux/pci.h>
+#include <linux/platform_device.h>
+
+struct viot_iommu {
+       /* Node offset within the table */
+       unsigned int                    offset;
+       struct fwnode_handle            *fwnode;
+       struct list_head                list;
+};
+
+struct viot_endpoint {
+       union {
+               /* PCI range */
+               struct {
+                       u16             segment_start;
+                       u16             segment_end;
+                       u16             bdf_start;
+                       u16             bdf_end;
+               };
+               /* MMIO */
+               u64                     address;
+       };
+       u32                             endpoint_id;
+       struct viot_iommu               *viommu;
+       struct list_head                list;
+};
+
+static struct acpi_table_viot *viot;
+static LIST_HEAD(viot_iommus);
+static LIST_HEAD(viot_pci_ranges);
+static LIST_HEAD(viot_mmio_endpoints);
+
+static int __init viot_check_bounds(const struct acpi_viot_header *hdr)
+{
+       struct acpi_viot_header *start, *end, *hdr_end;
+
+       start = ACPI_ADD_PTR(struct acpi_viot_header, viot,
+                            max_t(size_t, sizeof(*viot), viot->node_offset));
+       end = ACPI_ADD_PTR(struct acpi_viot_header, viot, viot->header.length);
+       hdr_end = ACPI_ADD_PTR(struct acpi_viot_header, hdr, sizeof(*hdr));
+
+       if (hdr < start || hdr_end > end) {
+               pr_err(FW_BUG "Node pointer overflows\n");
+               return -EOVERFLOW;
+       }
+       if (hdr->length < sizeof(*hdr)) {
+               pr_err(FW_BUG "Empty node\n");
+               return -EINVAL;
+       }
+       return 0;
+}
+
+static int __init viot_get_pci_iommu_fwnode(struct viot_iommu *viommu,
+                                           u16 segment, u16 bdf)
+{
+       struct pci_dev *pdev;
+       struct fwnode_handle *fwnode;
+
+       pdev = pci_get_domain_bus_and_slot(segment, PCI_BUS_NUM(bdf),
+                                          bdf & 0xff);
+       if (!pdev) {
+               pr_err("Could not find PCI IOMMU\n");
+               return -ENODEV;
+       }
+
+       fwnode = pdev->dev.fwnode;
+       if (!fwnode) {
+               /*
+                * PCI devices aren't necessarily described by ACPI. Create a
+                * fwnode so the IOMMU subsystem can identify this device.
+                */
+               fwnode = acpi_alloc_fwnode_static();
+               if (!fwnode) {
+                       pci_dev_put(pdev);
+                       return -ENOMEM;
+               }
+               set_primary_fwnode(&pdev->dev, fwnode);
+       }
+       viommu->fwnode = pdev->dev.fwnode;
+       pci_dev_put(pdev);
+       return 0;
+}
+
+static int __init viot_get_mmio_iommu_fwnode(struct viot_iommu *viommu,
+                                            u64 address)
+{
+       struct acpi_device *adev;
+       struct resource res = {
+               .start  = address,
+               .end    = address,
+               .flags  = IORESOURCE_MEM,
+       };
+
+       adev = acpi_resource_consumer(&res);
+       if (!adev) {
+               pr_err("Could not find MMIO IOMMU\n");
+               return -EINVAL;
+       }
+       viommu->fwnode = &adev->fwnode;
+       return 0;
+}
+
+static struct viot_iommu * __init viot_get_iommu(unsigned int offset)
+{
+       int ret;
+       struct viot_iommu *viommu;
+       struct acpi_viot_header *hdr = ACPI_ADD_PTR(struct acpi_viot_header,
+                                                   viot, offset);
+       union {
+               struct acpi_viot_virtio_iommu_pci pci;
+               struct acpi_viot_virtio_iommu_mmio mmio;
+       } *node = (void *)hdr;
+
+       list_for_each_entry(viommu, &viot_iommus, list)
+               if (viommu->offset == offset)
+                       return viommu;
+
+       if (viot_check_bounds(hdr))
+               return NULL;
+
+       viommu = kzalloc(sizeof(*viommu), GFP_KERNEL);
+       if (!viommu)
+               return NULL;
+
+       viommu->offset = offset;
+       switch (hdr->type) {
+       case ACPI_VIOT_NODE_VIRTIO_IOMMU_PCI:
+               if (hdr->length < sizeof(node->pci))
+                       goto err_free;
+
+               ret = viot_get_pci_iommu_fwnode(viommu, node->pci.segment,
+                                               node->pci.bdf);
+               break;
+       case ACPI_VIOT_NODE_VIRTIO_IOMMU_MMIO:
+               if (hdr->length < sizeof(node->mmio))
+                       goto err_free;
+
+               ret = viot_get_mmio_iommu_fwnode(viommu,
+                                                node->mmio.base_address);
+               break;
+       default:
+               ret = -EINVAL;
+       }
+       if (ret)
+               goto err_free;
+
+       list_add(&viommu->list, &viot_iommus);
+       return viommu;
+
+err_free:
+       kfree(viommu);
+       return NULL;
+}
+
+static int __init viot_parse_node(const struct acpi_viot_header *hdr)
+{
+       int ret = -EINVAL;
+       struct list_head *list;
+       struct viot_endpoint *ep;
+       union {
+               struct acpi_viot_mmio mmio;
+               struct acpi_viot_pci_range pci;
+       } *node = (void *)hdr;
+
+       if (viot_check_bounds(hdr))
+               return -EINVAL;
+
+       if (hdr->type == ACPI_VIOT_NODE_VIRTIO_IOMMU_PCI ||
+           hdr->type == ACPI_VIOT_NODE_VIRTIO_IOMMU_MMIO)
+               return 0;
+
+       ep = kzalloc(sizeof(*ep), GFP_KERNEL);
+       if (!ep)
+               return -ENOMEM;
+
+       switch (hdr->type) {
+       case ACPI_VIOT_NODE_PCI_RANGE:
+               if (hdr->length < sizeof(node->pci)) {
+                       pr_err(FW_BUG "Invalid PCI node size\n");
+                       goto err_free;
+               }
+
+               ep->segment_start = node->pci.segment_start;
+               ep->segment_end = node->pci.segment_end;
+               ep->bdf_start = node->pci.bdf_start;
+               ep->bdf_end = node->pci.bdf_end;
+               ep->endpoint_id = node->pci.endpoint_start;
+               ep->viommu = viot_get_iommu(node->pci.output_node);
+               list = &viot_pci_ranges;
+               break;
+       case ACPI_VIOT_NODE_MMIO:
+               if (hdr->length < sizeof(node->mmio)) {
+                       pr_err(FW_BUG "Invalid MMIO node size\n");
+                       goto err_free;
+               }
+
+               ep->address = node->mmio.base_address;
+               ep->endpoint_id = node->mmio.endpoint;
+               ep->viommu = viot_get_iommu(node->mmio.output_node);
+               list = &viot_mmio_endpoints;
+               break;
+       default:
+               pr_warn("Unsupported node %x\n", hdr->type);
+               ret = 0;
+               goto err_free;
+       }
+
+       if (!ep->viommu) {
+               pr_warn("No IOMMU node found\n");
+               /*
+                * A future version of the table may use the node for other
+                * purposes. Keep parsing.
+                */
+               ret = 0;
+               goto err_free;
+       }
+
+       list_add(&ep->list, list);
+       return 0;
+
+err_free:
+       kfree(ep);
+       return ret;
+}
+
+/**
+ * acpi_viot_init - Parse the VIOT table
+ *
+ * Parse the VIOT table, prepare the list of endpoints to be used during DMA
+ * setup of devices.
+ */
+void __init acpi_viot_init(void)
+{
+       int i;
+       acpi_status status;
+       struct acpi_table_header *hdr;
+       struct acpi_viot_header *node;
+
+       status = acpi_get_table(ACPI_SIG_VIOT, 0, &hdr);
+       if (ACPI_FAILURE(status)) {
+               if (status != AE_NOT_FOUND) {
+                       const char *msg = acpi_format_exception(status);
+
+                       pr_err("Failed to get table, %s\n", msg);
+               }
+               return;
+       }
+
+       viot = (void *)hdr;
+
+       node = ACPI_ADD_PTR(struct acpi_viot_header, viot, viot->node_offset);
+       for (i = 0; i < viot->node_count; i++) {
+               if (viot_parse_node(node))
+                       return;
+
+               node = ACPI_ADD_PTR(struct acpi_viot_header, node,
+                                   node->length);
+       }
+
+       acpi_put_table(hdr);
+}
+
+static int viot_dev_iommu_init(struct device *dev, struct viot_iommu *viommu,
+                              u32 epid)
+{
+       const struct iommu_ops *ops;
+
+       if (!viommu)
+               return -ENODEV;
+
+       /* We're not translating ourself */
+       if (viommu->fwnode == dev->fwnode)
+               return -EINVAL;
+
+       ops = iommu_ops_from_fwnode(viommu->fwnode);
+       if (!ops)
+               return IS_ENABLED(CONFIG_VIRTIO_IOMMU) ?
+                       -EPROBE_DEFER : -ENODEV;
+
+       return acpi_iommu_fwspec_init(dev, epid, viommu->fwnode, ops);
+}
+
+static int viot_pci_dev_iommu_init(struct pci_dev *pdev, u16 dev_id, void *data)
+{
+       u32 epid;
+       struct viot_endpoint *ep;
+       u32 domain_nr = pci_domain_nr(pdev->bus);
+
+       list_for_each_entry(ep, &viot_pci_ranges, list) {
+               if (domain_nr >= ep->segment_start &&
+                   domain_nr <= ep->segment_end &&
+                   dev_id >= ep->bdf_start &&
+                   dev_id <= ep->bdf_end) {
+                       epid = ((domain_nr - ep->segment_start) << 16) +
+                               dev_id - ep->bdf_start + ep->endpoint_id;
+
+                       /*
+                        * If we found a PCI range managed by the viommu, we're
+                        * the one that has to request ACS.
+                        */
+                       pci_request_acs();
+
+                       return viot_dev_iommu_init(&pdev->dev, ep->viommu,
+                                                  epid);
+               }
+       }
+       return -ENODEV;
+}
+
+static int viot_mmio_dev_iommu_init(struct platform_device *pdev)
+{
+       struct resource *mem;
+       struct viot_endpoint *ep;
+
+       mem = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+       if (!mem)
+               return -ENODEV;
+
+       list_for_each_entry(ep, &viot_mmio_endpoints, list) {
+               if (ep->address == mem->start)
+                       return viot_dev_iommu_init(&pdev->dev, ep->viommu,
+                                                  ep->endpoint_id);
+       }
+       return -ENODEV;
+}
+
+/**
+ * viot_iommu_configure - Setup IOMMU ops for an endpoint described by VIOT
+ * @dev: the endpoint
+ *
+ * Return: 0 on success, <0 on failure
+ */
+int viot_iommu_configure(struct device *dev)
+{
+       if (dev_is_pci(dev))
+               return pci_for_each_dma_alias(to_pci_dev(dev),
+                                             viot_pci_dev_iommu_init, NULL);
+       else if (dev_is_platform(dev))
+               return viot_mmio_dev_iommu_init(to_platform_device(dev));
+       return -ENODEV;
+}