igb_uio: issue FLR during open and release of device file
[dpdk.git] / lib / librte_eal / linuxapp / igb_uio / igb_uio.c
index f5617d2..07a19a3 100644 (file)
 
 #include "compat.h"
 
-#ifdef RTE_PCI_CONFIG
-#define PCI_SYS_FILE_BUF_SIZE      10
-#define PCI_DEV_CAP_REG            0xA4
-#define PCI_DEV_CTRL_REG           0xA8
-#define PCI_DEV_CAP_EXT_TAG_MASK   0x20
-#define PCI_DEV_CTRL_EXT_TAG_SHIFT 8
-#define PCI_DEV_CTRL_EXT_TAG_MASK  (1 << PCI_DEV_CTRL_EXT_TAG_SHIFT)
-#endif
-
 /**
  * A structure describing the private information for a uio device.
  */
@@ -58,7 +49,7 @@ struct rte_uio_pci_dev {
        enum rte_intr_mode mode;
 };
 
-static char *intr_mode = NULL;
+static char *intr_mode;
 static enum rte_intr_mode igbuio_intr_mode_preferred = RTE_INTR_MODE_MSIX;
 
 /* sriov sysfs */
@@ -90,109 +81,10 @@ store_max_vfs(struct device *dev, struct device_attribute *attr,
        return err ? err : count;
 }
 
-#ifdef RTE_PCI_CONFIG
-static ssize_t
-show_extended_tag(struct device *dev, struct device_attribute *attr, char *buf)
-{
-       struct pci_dev *pci_dev = to_pci_dev(dev);
-       uint32_t val = 0;
-
-       pci_read_config_dword(pci_dev, PCI_DEV_CAP_REG, &val);
-       if (!(val & PCI_DEV_CAP_EXT_TAG_MASK)) /* Not supported */
-               return snprintf(buf, PCI_SYS_FILE_BUF_SIZE, "%s\n", "invalid");
-
-       val = 0;
-       pci_bus_read_config_dword(pci_dev->bus, pci_dev->devfn,
-                                       PCI_DEV_CTRL_REG, &val);
-
-       return snprintf(buf, PCI_SYS_FILE_BUF_SIZE, "%s\n",
-               (val & PCI_DEV_CTRL_EXT_TAG_MASK) ? "on" : "off");
-}
-
-static ssize_t
-store_extended_tag(struct device *dev,
-                  struct device_attribute *attr,
-                  const char *buf,
-                  size_t count)
-{
-       struct pci_dev *pci_dev = to_pci_dev(dev);
-       uint32_t val = 0, enable;
-
-       if (strncmp(buf, "on", 2) == 0)
-               enable = 1;
-       else if (strncmp(buf, "off", 3) == 0)
-               enable = 0;
-       else
-               return -EINVAL;
-
-       pci_cfg_access_lock(pci_dev);
-       pci_bus_read_config_dword(pci_dev->bus, pci_dev->devfn,
-                                       PCI_DEV_CAP_REG, &val);
-       if (!(val & PCI_DEV_CAP_EXT_TAG_MASK)) { /* Not supported */
-               pci_cfg_access_unlock(pci_dev);
-               return -EPERM;
-       }
-
-       val = 0;
-       pci_bus_read_config_dword(pci_dev->bus, pci_dev->devfn,
-                                       PCI_DEV_CTRL_REG, &val);
-       if (enable)
-               val |= PCI_DEV_CTRL_EXT_TAG_MASK;
-       else
-               val &= ~PCI_DEV_CTRL_EXT_TAG_MASK;
-       pci_bus_write_config_dword(pci_dev->bus, pci_dev->devfn,
-                                       PCI_DEV_CTRL_REG, val);
-       pci_cfg_access_unlock(pci_dev);
-
-       return count;
-}
-
-static ssize_t
-show_max_read_request_size(struct device *dev,
-                          struct device_attribute *attr,
-                          char *buf)
-{
-       struct pci_dev *pci_dev = to_pci_dev(dev);
-       int val = pcie_get_readrq(pci_dev);
-
-       return snprintf(buf, PCI_SYS_FILE_BUF_SIZE, "%d\n", val);
-}
-
-static ssize_t
-store_max_read_request_size(struct device *dev,
-                           struct device_attribute *attr,
-                           const char *buf,
-                           size_t count)
-{
-       struct pci_dev *pci_dev = to_pci_dev(dev);
-       unsigned long size = 0;
-       int ret;
-
-       if (0 != kstrtoul(buf, 0, &size))
-               return -EINVAL;
-
-       ret = pcie_set_readrq(pci_dev, (int)size);
-       if (ret < 0)
-               return ret;
-
-       return count;
-}
-#endif
-
 static DEVICE_ATTR(max_vfs, S_IRUGO | S_IWUSR, show_max_vfs, store_max_vfs);
-#ifdef RTE_PCI_CONFIG
-static DEVICE_ATTR(extended_tag, S_IRUGO | S_IWUSR, show_extended_tag,
-       store_extended_tag);
-static DEVICE_ATTR(max_read_request_size, S_IRUGO | S_IWUSR,
-       show_max_read_request_size, store_max_read_request_size);
-#endif
 
 static struct attribute *dev_attrs[] = {
        &dev_attr_max_vfs.attr,
-#ifdef RTE_PCI_CONFIG
-       &dev_attr_extended_tag.attr,
-       &dev_attr_max_read_request_size.attr,
-#endif
        NULL,
 };
 
@@ -278,6 +170,37 @@ igbuio_pci_irqhandler(int irq, struct uio_info *info)
        return IRQ_HANDLED;
 }
 
+/**
+ * This gets called while opening uio device file.
+ */
+static int
+igbuio_pci_open(struct uio_info *info, struct inode *inode)
+{
+       struct rte_uio_pci_dev *udev = info->priv;
+       struct pci_dev *dev = udev->pdev;
+
+       pci_reset_function(dev);
+
+       /* set bus master, which was cleared by the reset function */
+       pci_set_master(dev);
+
+       return 0;
+}
+
+static int
+igbuio_pci_release(struct uio_info *info, struct inode *inode)
+{
+       struct rte_uio_pci_dev *udev = info->priv;
+       struct pci_dev *dev = udev->pdev;
+
+       /* stop the device from further DMA */
+       pci_clear_master(dev);
+
+       pci_reset_function(dev);
+
+       return 0;
+}
+
 #ifdef CONFIG_XEN_DOM0
 static int
 igbuio_dom0_mmap_phys(struct uio_info *info, struct vm_area_struct *vma)
@@ -332,7 +255,7 @@ igbuio_pci_setup_iomem(struct pci_dev *dev, struct uio_info *info,
        unsigned long addr, len;
        void *internal_addr;
 
-       if (sizeof(info->mem) / sizeof(info->mem[0]) <= n)
+       if (n >= ARRAY_SIZE(info->mem))
                return -EINVAL;
 
        addr = pci_resource_start(dev, pci_bar);
@@ -357,7 +280,7 @@ igbuio_pci_setup_ioport(struct pci_dev *dev, struct uio_info *info,
 {
        unsigned long addr, len;
 
-       if (sizeof(info->port) / sizeof(info->port[0]) <= n)
+       if (n >= ARRAY_SIZE(info->port))
                return -EINVAL;
 
        addr = pci_resource_start(dev, pci_bar);
@@ -402,7 +325,7 @@ igbuio_setup_bars(struct pci_dev *dev, struct uio_info *info)
        iom = 0;
        iop = 0;
 
-       for (i = 0; i != sizeof(bar_names) / sizeof(bar_names[0]); i++) {
+       for (i = 0; i < ARRAY_SIZE(bar_names); i++) {
                if (pci_resource_len(dev, i) != 0 &&
                                pci_resource_start(dev, i) != 0) {
                        flags = pci_resource_flags(dev, i);
@@ -422,7 +345,7 @@ igbuio_setup_bars(struct pci_dev *dev, struct uio_info *info)
                }
        }
 
-       return (iom != 0) ? ret : -ENOENT;
+       return (iom != 0 || iop != 0) ? ret : -ENOENT;
 }
 
 #if LINUX_VERSION_CODE < KERNEL_VERSION(3, 8, 0)
@@ -433,7 +356,11 @@ static int
 igbuio_pci_probe(struct pci_dev *dev, const struct pci_device_id *id)
 {
        struct rte_uio_pci_dev *udev;
+#ifdef HAVE_PCI_ENABLE_MSIX
        struct msix_entry msix_entry;
+#endif
+       dma_addr_t map_dma_addr;
+       void *map_addr;
        int err;
 
        udev = kzalloc(sizeof(struct rte_uio_pci_dev), GFP_KERNEL);
@@ -450,16 +377,6 @@ igbuio_pci_probe(struct pci_dev *dev, const struct pci_device_id *id)
                goto fail_free;
        }
 
-       /*
-        * reserve device's PCI memory regions for use by this
-        * module
-        */
-       err = pci_request_regions(dev, "igb_uio");
-       if (err != 0) {
-               dev_err(&dev->dev, "Cannot request regions\n");
-               goto fail_disable;
-       }
-
        /* enable bus mastering on the device */
        pci_set_master(dev);
 
@@ -486,6 +403,8 @@ igbuio_pci_probe(struct pci_dev *dev, const struct pci_device_id *id)
        udev->info.version = "0.1";
        udev->info.handler = igbuio_pci_irqhandler;
        udev->info.irqcontrol = igbuio_pci_irqcontrol;
+       udev->info.open = igbuio_pci_open;
+       udev->info.release = igbuio_pci_release;
 #ifdef CONFIG_XEN_DOM0
        /* check if the driver run on Xen Dom0 */
        if (xen_initial_domain())
@@ -497,18 +416,28 @@ igbuio_pci_probe(struct pci_dev *dev, const struct pci_device_id *id)
        switch (igbuio_intr_mode_preferred) {
        case RTE_INTR_MODE_MSIX:
                /* Only 1 msi-x vector needed */
+#ifdef HAVE_PCI_ENABLE_MSIX
                msix_entry.entry = 0;
                if (pci_enable_msix(dev, &msix_entry, 1) == 0) {
                        dev_dbg(&dev->dev, "using MSI-X");
+                       udev->info.irq_flags = IRQF_NO_THREAD;
                        udev->info.irq = msix_entry.vector;
                        udev->mode = RTE_INTR_MODE_MSIX;
                        break;
                }
+#else
+               if (pci_alloc_irq_vectors(dev, 1, 1, PCI_IRQ_MSIX) == 1) {
+                       dev_dbg(&dev->dev, "using MSI-X");
+                       udev->info.irq = pci_irq_vector(dev, 0);
+                       udev->mode = RTE_INTR_MODE_MSIX;
+                       break;
+               }
+#endif
                /* fall back to INTX */
        case RTE_INTR_MODE_LEGACY:
                if (pci_intx_mask_supported(dev)) {
                        dev_dbg(&dev->dev, "using INTX");
-                       udev->info.irq_flags = IRQF_SHARED;
+                       udev->info.irq_flags = IRQF_SHARED | IRQF_NO_THREAD;
                        udev->info.irq = dev->irq;
                        udev->mode = RTE_INTR_MODE_LEGACY;
                        break;
@@ -541,6 +470,27 @@ igbuio_pci_probe(struct pci_dev *dev, const struct pci_device_id *id)
        dev_info(&dev->dev, "uio device registered with irq %lx\n",
                 udev->info.irq);
 
+       /*
+        * Doing a harmless dma mapping for attaching the device to
+        * the iommu identity mapping if kernel boots with iommu=pt.
+        * Note this is not a problem if no IOMMU at all.
+        */
+       map_addr = dma_alloc_coherent(&dev->dev, 1024, &map_dma_addr,
+                       GFP_KERNEL);
+       if (map_addr)
+               memset(map_addr, 0, 1024);
+
+       if (!map_addr)
+               dev_info(&dev->dev, "dma mapping failed\n");
+       else {
+               dev_info(&dev->dev, "mapping 1K dma=%#llx host=%p\n",
+                        (unsigned long long)map_dma_addr, map_addr);
+
+               dma_free_coherent(&dev->dev, 1024, map_addr, map_dma_addr);
+               dev_info(&dev->dev, "unmapping 1K dma=%#llx host=%p\n",
+                        (unsigned long long)map_dma_addr, map_addr);
+       }
+
        return 0;
 
 fail_remove_group:
@@ -549,8 +499,6 @@ fail_release_iomem:
        igbuio_pci_release_iomem(&udev->info);
        if (udev->mode == RTE_INTR_MODE_MSIX)
                pci_disable_msix(udev->pdev);
-       pci_release_regions(dev);
-fail_disable:
        pci_disable_device(dev);
 fail_free:
        kfree(udev);
@@ -561,24 +509,16 @@ fail_free:
 static void
 igbuio_pci_remove(struct pci_dev *dev)
 {
-       struct uio_info *info = pci_get_drvdata(dev);
-       struct rte_uio_pci_dev *udev;
-
-       if (info->priv == NULL) {
-               pr_notice("Not igbuio device\n");
-               return;
-       }
-       udev = info->priv;
+       struct rte_uio_pci_dev *udev = pci_get_drvdata(dev);
 
        sysfs_remove_group(&dev->dev.kobj, &dev_attr_grp);
-       uio_unregister_device(info);
-       igbuio_pci_release_iomem(info);
+       uio_unregister_device(&udev->info);
+       igbuio_pci_release_iomem(&udev->info);
        if (udev->mode == RTE_INTR_MODE_MSIX)
                pci_disable_msix(dev);
-       pci_release_regions(dev);
        pci_disable_device(dev);
        pci_set_drvdata(dev, NULL);
-       kfree(info);
+       kfree(udev);
 }
 
 static int