vfio/pci: Add ioeventfd support

The ioeventfd here is actually irqfd handling of an ioeventfd such as
supported in KVM.  A user is able to pre-program a device write to
occur when the eventfd triggers.  This is yet another instance of
eventfd-irqfd triggering between KVM and vfio.  The impetus for this
is high frequency writes to pages which are virtualized in QEMU.
Enabling this near-direct write path for selected registers within
the virtualized page can improve performance and reduce overhead.
Specifically this is initially targeted at NVIDIA graphics cards where
the driver issues a write to an MMIO register within a virtualized
region in order to allow the MSI interrupt to re-trigger.

Reviewed-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Alex Williamson <alex.williamson@redhat.com>

drivers/vfio/pci/vfio_pci.c

@@ -305,6 +305,7 @@ static void vfio_pci_disable(struct vfio_pci_device *vdev)
 {
 	struct pci_dev *pdev = vdev->pdev;
 	struct vfio_pci_dummy_resource *dummy_res, *tmp;
+	struct vfio_pci_ioeventfd *ioeventfd, *ioeventfd_tmp;
 	int i, bar;
 
 	/* Stop the device from further DMA */
@@ -314,6 +315,15 @@ static void vfio_pci_disable(struct vfio_pci_device *vdev)
 				VFIO_IRQ_SET_ACTION_TRIGGER,
 				vdev->irq_type, 0, 0, NULL);
 
+	/* Device closed, don't need mutex here */
+	list_for_each_entry_safe(ioeventfd, ioeventfd_tmp,
+				 &vdev->ioeventfds_list, next) {
+		vfio_virqfd_disable(&ioeventfd->virqfd);
+		list_del(&ioeventfd->next);
+		kfree(ioeventfd);
+	}
+	vdev->ioeventfds_nr = 0;
+
 	vdev->virq_disabled = false;
 
 	for (i = 0; i < vdev->num_regions; i++)
@@ -1012,6 +1022,28 @@ static long vfio_pci_ioctl(void *device_data,
 
 		kfree(groups);
 		return ret;
+	} else if (cmd == VFIO_DEVICE_IOEVENTFD) {
+		struct vfio_device_ioeventfd ioeventfd;
+		int count;
+
+		minsz = offsetofend(struct vfio_device_ioeventfd, fd);
+
+		if (copy_from_user(&ioeventfd, (void __user *)arg, minsz))
+			return -EFAULT;
+
+		if (ioeventfd.argsz < minsz)
+			return -EINVAL;
+
+		if (ioeventfd.flags & ~VFIO_DEVICE_IOEVENTFD_SIZE_MASK)
+			return -EINVAL;
+
+		count = ioeventfd.flags & VFIO_DEVICE_IOEVENTFD_SIZE_MASK;
+
+		if (hweight8(count) != 1 || ioeventfd.fd < -1)
+			return -EINVAL;
+
+		return vfio_pci_ioeventfd(vdev, ioeventfd.offset,
+					  ioeventfd.data, count, ioeventfd.fd);
 	}
 
 	return -ENOTTY;
@@ -1174,6 +1206,8 @@ static int vfio_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 	vdev->irq_type = VFIO_PCI_NUM_IRQS;
 	mutex_init(&vdev->igate);
 	spin_lock_init(&vdev->irqlock);
+	mutex_init(&vdev->ioeventfds_lock);
+	INIT_LIST_HEAD(&vdev->ioeventfds_list);
 
 	ret = vfio_add_group_dev(&pdev->dev, &vfio_pci_ops, vdev);
 	if (ret) {
@@ -1215,6 +1249,7 @@ static void vfio_pci_remove(struct pci_dev *pdev)
 
 	vfio_iommu_group_put(pdev->dev.iommu_group, &pdev->dev);
 	kfree(vdev->region);
+	mutex_destroy(&vdev->ioeventfds_lock);
 	kfree(vdev);
 
 	if (vfio_pci_is_vga(pdev)) {

drivers/vfio/pci/vfio_pci_private.h

@@ -29,6 +29,19 @@
 #define PCI_CAP_ID_INVALID		0xFF	/* default raw access */
 #define PCI_CAP_ID_INVALID_VIRT		0xFE	/* default virt access */
 
+/* Cap maximum number of ioeventfds per device (arbitrary) */
+#define VFIO_PCI_IOEVENTFD_MAX		1000
+
+struct vfio_pci_ioeventfd {
+	struct list_head	next;
+	struct virqfd		*virqfd;
+	void __iomem		*addr;
+	uint64_t		data;
+	loff_t			pos;
+	int			bar;
+	int			count;
+};
+
 struct vfio_pci_irq_ctx {
 	struct eventfd_ctx	*trigger;
 	struct virqfd		*unmask;
@@ -92,9 +105,12 @@ struct vfio_pci_device {
 	bool			nointx;
 	struct pci_saved_state	*pci_saved_state;
 	int			refcnt;
+	int			ioeventfds_nr;
 	struct eventfd_ctx	*err_trigger;
 	struct eventfd_ctx	*req_trigger;
 	struct list_head	dummy_resources_list;
+	struct mutex		ioeventfds_lock;
+	struct list_head	ioeventfds_list;
 };
 
 #define is_intx(vdev) (vdev->irq_type == VFIO_PCI_INTX_IRQ_INDEX)
@@ -120,6 +136,9 @@ extern ssize_t vfio_pci_bar_rw(struct vfio_pci_device *vdev, char __user *buf,
 extern ssize_t vfio_pci_vga_rw(struct vfio_pci_device *vdev, char __user *buf,
 			       size_t count, loff_t *ppos, bool iswrite);
 
+extern long vfio_pci_ioeventfd(struct vfio_pci_device *vdev, loff_t offset,
+			       uint64_t data, int count, int fd);
+
 extern int vfio_pci_init_perm_bits(void);
 extern void vfio_pci_uninit_perm_bits(void);
 

drivers/vfio/pci/vfio_pci_rdwr.c

@@ -17,6 +17,7 @@
 #include <linux/pci.h>
 #include <linux/uaccess.h>
 #include <linux/io.h>
+#include <linux/vfio.h>
 #include <linux/vgaarb.h>
 
 #include "vfio_pci_private.h"
@@ -275,3 +276,113 @@ ssize_t vfio_pci_vga_rw(struct vfio_pci_device *vdev, char __user *buf,
 
 	return done;
 }
+
+static int vfio_pci_ioeventfd_handler(void *opaque, void *unused)
+{
+	struct vfio_pci_ioeventfd *ioeventfd = opaque;
+
+	switch (ioeventfd->count) {
+	case 1:
+		vfio_iowrite8(ioeventfd->data, ioeventfd->addr);
+		break;
+	case 2:
+		vfio_iowrite16(ioeventfd->data, ioeventfd->addr);
+		break;
+	case 4:
+		vfio_iowrite32(ioeventfd->data, ioeventfd->addr);
+		break;
+#ifdef iowrite64
+	case 8:
+		vfio_iowrite64(ioeventfd->data, ioeventfd->addr);
+		break;
+#endif
+	}
+
+	return 0;
+}
+
+long vfio_pci_ioeventfd(struct vfio_pci_device *vdev, loff_t offset,
+			uint64_t data, int count, int fd)
+{
+	struct pci_dev *pdev = vdev->pdev;
+	loff_t pos = offset & VFIO_PCI_OFFSET_MASK;
+	int ret, bar = VFIO_PCI_OFFSET_TO_INDEX(offset);
+	struct vfio_pci_ioeventfd *ioeventfd;
+
+	/* Only support ioeventfds into BARs */
+	if (bar > VFIO_PCI_BAR5_REGION_INDEX)
+		return -EINVAL;
+
+	if (pos + count > pci_resource_len(pdev, bar))
+		return -EINVAL;
+
+	/* Disallow ioeventfds working around MSI-X table writes */
+	if (bar == vdev->msix_bar &&
+	    !(pos + count <= vdev->msix_offset ||
+	      pos >= vdev->msix_offset + vdev->msix_size))
+		return -EINVAL;
+
+#ifndef iowrite64
+	if (count == 8)
+		return -EINVAL;
+#endif
+
+	ret = vfio_pci_setup_barmap(vdev, bar);
+	if (ret)
+		return ret;
+
+	mutex_lock(&vdev->ioeventfds_lock);
+
+	list_for_each_entry(ioeventfd, &vdev->ioeventfds_list, next) {
+		if (ioeventfd->pos == pos && ioeventfd->bar == bar &&
+		    ioeventfd->data == data && ioeventfd->count == count) {
+			if (fd == -1) {
+				vfio_virqfd_disable(&ioeventfd->virqfd);
+				list_del(&ioeventfd->next);
+				vdev->ioeventfds_nr--;
+				kfree(ioeventfd);
+				ret = 0;
+			} else
+				ret = -EEXIST;
+
+			goto out_unlock;
+		}
+	}
+
+	if (fd < 0) {
+		ret = -ENODEV;
+		goto out_unlock;
+	}
+
+	if (vdev->ioeventfds_nr >= VFIO_PCI_IOEVENTFD_MAX) {
+		ret = -ENOSPC;
+		goto out_unlock;
+	}
+
+	ioeventfd = kzalloc(sizeof(*ioeventfd), GFP_KERNEL);
+	if (!ioeventfd) {
+		ret = -ENOMEM;
+		goto out_unlock;
+	}
+
+	ioeventfd->addr = vdev->barmap[bar] + pos;
+	ioeventfd->data = data;
+	ioeventfd->pos = pos;
+	ioeventfd->bar = bar;
+	ioeventfd->count = count;
+
+	ret = vfio_virqfd_enable(ioeventfd, vfio_pci_ioeventfd_handler,
+				 NULL, NULL, &ioeventfd->virqfd, fd);
+	if (ret) {
+		kfree(ioeventfd);
+		goto out_unlock;
+	}
+
+	list_add(&ioeventfd->next, &vdev->ioeventfds_list);
+	vdev->ioeventfds_nr++;
+
+out_unlock:
+	mutex_unlock(&vdev->ioeventfds_lock);
+
+	return ret;
+}

include/uapi/linux/vfio.h

@@ -575,6 +575,33 @@ struct vfio_device_gfx_plane_info {
 
 #define VFIO_DEVICE_GET_GFX_DMABUF _IO(VFIO_TYPE, VFIO_BASE + 15)
 
+/**
+ * VFIO_DEVICE_IOEVENTFD - _IOW(VFIO_TYPE, VFIO_BASE + 16,
+ *                              struct vfio_device_ioeventfd)
+ *
+ * Perform a write to the device at the specified device fd offset, with
+ * the specified data and width when the provided eventfd is triggered.
+ * vfio bus drivers may not support this for all regions, for all widths,
+ * or at all.  vfio-pci currently only enables support for BAR regions,
+ * excluding the MSI-X vector table.
+ *
+ * Return: 0 on success, -errno on failure.
+ */
+struct vfio_device_ioeventfd {
+	__u32	argsz;
+	__u32	flags;
+#define VFIO_DEVICE_IOEVENTFD_8		(1 << 0) /* 1-byte write */
+#define VFIO_DEVICE_IOEVENTFD_16	(1 << 1) /* 2-byte write */
+#define VFIO_DEVICE_IOEVENTFD_32	(1 << 2) /* 4-byte write */
+#define VFIO_DEVICE_IOEVENTFD_64	(1 << 3) /* 8-byte write */
+#define VFIO_DEVICE_IOEVENTFD_SIZE_MASK	(0xf)
+	__u64	offset;			/* device fd offset of write */
+	__u64	data;			/* data to be written */
+	__s32	fd;			/* -1 for de-assignment */
+};
+
+#define VFIO_DEVICE_IOEVENTFD		_IO(VFIO_TYPE, VFIO_BASE + 16)
+
 /* -------- API for Type1 VFIO IOMMU -------- */
 
 /**