powerpc/pseries: Add PFO support to the VIO bus
authorKent Yoder <key@linux.vnet.ibm.com>
Thu, 12 Apr 2012 05:08:22 +0000 (05:08 +0000)
committerBenjamin Herrenschmidt <benh@kernel.crashing.org>
Mon, 14 May 2012 00:49:09 +0000 (10:49 +1000)
Add support for the Platform Facilities Option (PFO) to the VIO bus.
These devices have a separate root node in OpenFirmware which
requires additional parsing to map into the existing VIO device
structure fields. This adds the interface for PFO device drivers to
make synchronous hypervisor calls.

Signed-off-by: Robert Jennings <rcj@linux.vnet.ibm.com>
Signed-off-by: Kent Yoder <key@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
arch/powerpc/include/asm/vio.h
arch/powerpc/kernel/vio.c

index 6bfd5ffe1d4fd287141523e10d1589d89472b636..b19adf751dd954c492b7afc062a76df896ef6b61 100644 (file)
 
 struct iommu_table;
 
+/*
+ * Platform Facilities Option (PFO)-specific data
+ */
+
+/* Starting unit address for PFO devices on the VIO BUS */
+#define VIO_BASE_PFO_UA        0x50000000
+
+/**
+ * vio_pfo_op - PFO operation parameters
+ *
+ * @flags: h_call subfunctions and modifiers
+ * @in: Input data block logical real address
+ * @inlen: If non-negative, the length of the input data block.  If negative,
+ *     the length of the input data descriptor list in bytes.
+ * @out: Output data block logical real address
+ * @outlen: If non-negative, the length of the input data block.  If negative,
+ *     the length of the input data descriptor list in bytes.
+ * @csbcpb: Logical real address of the 4k naturally-aligned storage block
+ *     containing the CSB & optional FC field specific CPB
+ * @timeout: # of milliseconds to retry h_call, 0 for no timeout.
+ * @hcall_err: pointer to return the h_call return value, else NULL
+ */
+struct vio_pfo_op {
+       u64 flags;
+       s64 in;
+       s64 inlen;
+       s64 out;
+       s64 outlen;
+       u64 csbcpb;
+       void *done;
+       unsigned long handle;
+       unsigned int timeout;
+       long hcall_err;
+};
+
+/* End PFO specific data */
+
+enum vio_dev_family {
+       VDEVICE,        /* The OF node is a child of /vdevice */
+       PFO,            /* The OF node is a child of /ibm,platform-facilities */
+};
+
 /**
  * vio_dev - This structure is used to describe virtual I/O devices.
  *
@@ -58,6 +100,7 @@ struct vio_dev {
        const char *name;
        const char *type;
        uint32_t unit_address;
+       uint32_t resource_id;
        unsigned int irq;
        struct {
                size_t desired;
@@ -65,6 +108,7 @@ struct vio_dev {
                size_t allocated;
                atomic_t allocs_failed;
        } cmo;
+       enum vio_dev_family family;
        struct device dev;
 };
 
@@ -95,6 +139,8 @@ extern void vio_cmo_set_dev_desired(struct vio_dev *viodev, size_t desired);
 
 extern void __devinit vio_unregister_device(struct vio_dev *dev);
 
+extern int vio_h_cop_sync(struct vio_dev *vdev, struct vio_pfo_op *op);
+
 struct device_node;
 
 extern struct vio_dev *vio_register_device_node(
index a3a99901c8ecf4a1929815329a72a54146017603..cb87301ccd55a63a46f3069efe3b860e3da5c3c8 100644 (file)
@@ -14,7 +14,9 @@
  *      2 of the License, or (at your option) any later version.
  */
 
+#include <linux/cpu.h>
 #include <linux/types.h>
+#include <linux/delay.h>
 #include <linux/stat.h>
 #include <linux/device.h>
 #include <linux/init.h>
@@ -709,13 +711,26 @@ static int vio_cmo_bus_probe(struct vio_dev *viodev)
        struct vio_driver *viodrv = to_vio_driver(dev->driver);
        unsigned long flags;
        size_t size;
+       bool dma_capable = false;
+
+       /* A device requires entitlement if it has a DMA window property */
+       switch (viodev->family) {
+       case VDEVICE:
+               if (of_get_property(viodev->dev.of_node,
+                                       "ibm,my-dma-window", NULL))
+                       dma_capable = true;
+               break;
+       case PFO:
+               dma_capable = false;
+               break;
+       default:
+               dev_warn(dev, "unknown device family: %d\n", viodev->family);
+               BUG();
+               break;
+       }
 
-       /*
-        * Check to see that device has a DMA window and configure
-        * entitlement for the device.
-        */
-       if (of_get_property(viodev->dev.of_node,
-                           "ibm,my-dma-window", NULL)) {
+       /* Configure entitlement for the device. */
+       if (dma_capable) {
                /* Check that the driver is CMO enabled and get desired DMA */
                if (!viodrv->get_desired_dma) {
                        dev_err(dev, "%s: device driver does not support CMO\n",
@@ -1050,6 +1065,94 @@ static void vio_cmo_sysfs_init(void) { }
 EXPORT_SYMBOL(vio_cmo_entitlement_update);
 EXPORT_SYMBOL(vio_cmo_set_dev_desired);
 
+
+/*
+ * Platform Facilities Option (PFO) support
+ */
+
+/**
+ * vio_h_cop_sync - Perform a synchronous PFO co-processor operation
+ *
+ * @vdev - Pointer to a struct vio_dev for device
+ * @op - Pointer to a struct vio_pfo_op for the operation parameters
+ *
+ * Calls the hypervisor to synchronously perform the PFO operation
+ * described in @op.  In the case of a busy response from the hypervisor,
+ * the operation will be re-submitted indefinitely unless a non-zero timeout
+ * is specified or an error occurs. The timeout places a limit on when to
+ * stop re-submitting a operation, the total time can be exceeded if an
+ * operation is in progress.
+ *
+ * If op->hcall_ret is not NULL, this will be set to the return from the
+ * last h_cop_op call or it will be 0 if an error not involving the h_call
+ * was encountered.
+ *
+ * Returns:
+ *     0 on success,
+ *     -EINVAL if the h_call fails due to an invalid parameter,
+ *     -E2BIG if the h_call can not be performed synchronously,
+ *     -EBUSY if a timeout is specified and has elapsed,
+ *     -EACCES if the memory area for data/status has been rescinded, or
+ *     -EPERM if a hardware fault has been indicated
+ */
+int vio_h_cop_sync(struct vio_dev *vdev, struct vio_pfo_op *op)
+{
+       struct device *dev = &vdev->dev;
+       unsigned long deadline = 0;
+       long hret = 0;
+       int ret = 0;
+
+       if (op->timeout)
+               deadline = jiffies + msecs_to_jiffies(op->timeout);
+
+       while (true) {
+               hret = plpar_hcall_norets(H_COP, op->flags,
+                               vdev->resource_id,
+                               op->in, op->inlen, op->out,
+                               op->outlen, op->csbcpb);
+
+               if (hret == H_SUCCESS ||
+                   (hret != H_NOT_ENOUGH_RESOURCES &&
+                    hret != H_BUSY && hret != H_RESOURCE) ||
+                   (op->timeout && time_after(deadline, jiffies)))
+                       break;
+
+               dev_dbg(dev, "%s: hcall ret(%ld), retrying.\n", __func__, hret);
+       }
+
+       switch (hret) {
+       case H_SUCCESS:
+               ret = 0;
+               break;
+       case H_OP_MODE:
+       case H_TOO_BIG:
+               ret = -E2BIG;
+               break;
+       case H_RESCINDED:
+               ret = -EACCES;
+               break;
+       case H_HARDWARE:
+               ret = -EPERM;
+               break;
+       case H_NOT_ENOUGH_RESOURCES:
+       case H_RESOURCE:
+       case H_BUSY:
+               ret = -EBUSY;
+               break;
+       default:
+               ret = -EINVAL;
+               break;
+       }
+
+       if (ret)
+               dev_dbg(dev, "%s: Sync h_cop_op failure (ret:%d) (hret:%ld)\n",
+                               __func__, ret, hret);
+
+       op->hcall_err = hret;
+       return ret;
+}
+EXPORT_SYMBOL(vio_h_cop_sync);
+
 static struct iommu_table *vio_build_iommu_table(struct vio_dev *dev)
 {
        const unsigned char *dma_window;
@@ -1211,35 +1314,87 @@ static void __devinit vio_dev_release(struct device *dev)
 struct vio_dev *vio_register_device_node(struct device_node *of_node)
 {
        struct vio_dev *viodev;
+       struct device_node *parent_node;
        const unsigned int *unit_address;
+       const unsigned int *pfo_resid = NULL;
+       enum vio_dev_family family;
+       const char *of_node_name = of_node->name ? of_node->name : "<unknown>";
 
-       /* we need the 'device_type' property, in order to match with drivers */
-       if (of_node->type == NULL) {
-               printk(KERN_WARNING "%s: node %s missing 'device_type'\n",
-                               __func__,
-                               of_node->name ? of_node->name : "<unknown>");
+       /*
+        * Determine if this node is a under the /vdevice node or under the
+        * /ibm,platform-facilities node.  This decides the device's family.
+        */
+       parent_node = of_get_parent(of_node);
+       if (parent_node) {
+               if (!strcmp(parent_node->full_name, "/ibm,platform-facilities"))
+                       family = PFO;
+               else if (!strcmp(parent_node->full_name, "/vdevice"))
+                       family = VDEVICE;
+               else {
+                       pr_warn("%s: parent(%s) of %s not recognized.\n",
+                                       __func__,
+                                       parent_node->full_name,
+                                       of_node_name);
+                       of_node_put(parent_node);
+                       return NULL;
+               }
+               of_node_put(parent_node);
+       } else {
+               pr_warn("%s: could not determine the parent of node %s.\n",
+                               __func__, of_node_name);
                return NULL;
        }
 
-       unit_address = of_get_property(of_node, "reg", NULL);
-       if (unit_address == NULL) {
-               printk(KERN_WARNING "%s: node %s missing 'reg'\n",
-                               __func__,
-                               of_node->name ? of_node->name : "<unknown>");
-               return NULL;
+       if (family == PFO) {
+               if (of_get_property(of_node, "interrupt-controller", NULL)) {
+                       pr_debug("%s: Skipping the interrupt controller %s.\n",
+                                       __func__, of_node_name);
+                       return NULL;
+               }
        }
 
        /* allocate a vio_dev for this node */
        viodev = kzalloc(sizeof(struct vio_dev), GFP_KERNEL);
-       if (viodev == NULL)
+       if (viodev == NULL) {
+               pr_warn("%s: allocation failure for VIO device.\n", __func__);
                return NULL;
+       }
 
-       viodev->irq = irq_of_parse_and_map(of_node, 0);
+       /* we need the 'device_type' property, in order to match with drivers */
+       viodev->family = family;
+       if (viodev->family == VDEVICE) {
+               if (of_node->type != NULL)
+                       viodev->type = of_node->type;
+               else {
+                       pr_warn("%s: node %s is missing the 'device_type' "
+                                       "property.\n", __func__, of_node_name);
+                       goto out;
+               }
+
+               unit_address = of_get_property(of_node, "reg", NULL);
+               if (unit_address == NULL) {
+                       pr_warn("%s: node %s missing 'reg'\n",
+                                       __func__, of_node_name);
+                       goto out;
+               }
+               dev_set_name(&viodev->dev, "%x", *unit_address);
+               viodev->irq = irq_of_parse_and_map(of_node, 0);
+               viodev->unit_address = *unit_address;
+       } else {
+               /* PFO devices need their resource_id for submitting COP_OPs
+                * This is an optional field for devices, but is required when
+                * performing synchronous ops */
+               pfo_resid = of_get_property(of_node, "ibm,resource-id", NULL);
+               if (pfo_resid != NULL)
+                       viodev->resource_id = *pfo_resid;
+
+               unit_address = NULL;
+               dev_set_name(&viodev->dev, "%s", of_node_name);
+               viodev->type = of_node_name;
+               viodev->irq = 0;
+       }
 
-       dev_set_name(&viodev->dev, "%x", *unit_address);
        viodev->name = of_node->name;
-       viodev->type = of_node->type;
-       viodev->unit_address = *unit_address;
        viodev->dev.of_node = of_node_get(of_node);
 
        if (firmware_has_feature(FW_FEATURE_CMO))
@@ -1267,16 +1422,51 @@ struct vio_dev *vio_register_device_node(struct device_node *of_node)
        }
 
        return viodev;
+
+out:   /* Use this exit point for any return prior to device_register */
+       kfree(viodev);
+
+       return NULL;
 }
 EXPORT_SYMBOL(vio_register_device_node);
 
+/*
+ * vio_bus_scan_for_devices - Scan OF and register each child device
+ * @root_name - OF node name for the root of the subtree to search.
+ *             This must be non-NULL
+ *
+ * Starting from the root node provide, register the device node for
+ * each child beneath the root.
+ */
+static void vio_bus_scan_register_devices(char *root_name)
+{
+       struct device_node *node_root, *node_child;
+
+       if (!root_name)
+               return;
+
+       node_root = of_find_node_by_name(NULL, root_name);
+       if (node_root) {
+
+               /*
+                * Create struct vio_devices for each virtual device in
+                * the device tree. Drivers will associate with them later.
+                */
+               node_child = of_get_next_child(node_root, NULL);
+               while (node_child) {
+                       vio_register_device_node(node_child);
+                       node_child = of_get_next_child(node_root, node_child);
+               }
+               of_node_put(node_root);
+       }
+}
+
 /**
  * vio_bus_init: - Initialize the virtual IO bus
  */
 static int __init vio_bus_init(void)
 {
        int err;
-       struct device_node *node_vroot;
 
        if (firmware_has_feature(FW_FEATURE_CMO))
                vio_cmo_sysfs_init();
@@ -1301,19 +1491,8 @@ static int __init vio_bus_init(void)
        if (firmware_has_feature(FW_FEATURE_CMO))
                vio_cmo_bus_init();
 
-       node_vroot = of_find_node_by_name(NULL, "vdevice");
-       if (node_vroot) {
-               struct device_node *of_node;
-
-               /*
-                * Create struct vio_devices for each virtual device in
-                * the device tree. Drivers will associate with them later.
-                */
-               for (of_node = node_vroot->child; of_node != NULL;
-                               of_node = of_node->sibling)
-                       vio_register_device_node(of_node);
-               of_node_put(node_vroot);
-       }
+       vio_bus_scan_register_devices("vdevice");
+       vio_bus_scan_register_devices("ibm,platform-facilities");
 
        return 0;
 }
@@ -1436,12 +1615,28 @@ struct vio_dev *vio_find_node(struct device_node *vnode)
 {
        const uint32_t *unit_address;
        char kobj_name[20];
+       struct device_node *vnode_parent;
+       const char *dev_type;
+
+       vnode_parent = of_get_parent(vnode);
+       if (!vnode_parent)
+               return NULL;
+
+       dev_type = of_get_property(vnode_parent, "device_type", NULL);
+       of_node_put(vnode_parent);
+       if (!dev_type)
+               return NULL;
 
        /* construct the kobject name from the device node */
-       unit_address = of_get_property(vnode, "reg", NULL);
-       if (!unit_address)
+       if (!strcmp(dev_type, "vdevice")) {
+               unit_address = of_get_property(vnode, "reg", NULL);
+               if (!unit_address)
+                       return NULL;
+               snprintf(kobj_name, sizeof(kobj_name), "%x", *unit_address);
+       } else if (!strcmp(dev_type, "ibm,platform-facilities"))
+               snprintf(kobj_name, sizeof(kobj_name), "%s", vnode->name);
+       else
                return NULL;
-       snprintf(kobj_name, sizeof(kobj_name), "%x", *unit_address);
 
        return vio_find_name(kobj_name);
 }