[RFC v5 5/7] vfio-pci: Allow to mmap sub-page MMIO BARs if the mmio page is exclusive

[Yongji Xie]

Current vfio-pci implementation disallows to mmap
sub-page(size < PAGE_SIZE) MMIO BARs because
these BARs' mmio page may be shared with other
BARs.

But we should allow to mmap these sub-page MMIO
BARs if we can make sure these BARs' mmio page
will not be shared with other BARs.

To acheive that, we firstly need to enforce all
PCI MMIO BARs to be page aligned like the commit
"PCI: Add support for enforcing all MMIO BARs
to be page aligned" does. Most of PCI BARs
will be assigned into an exclusive page with a
hole. Then, we must make sure that hot-add
device's BAR will never be assigned into the
hole. So we add shadow resources and put them
into the hole in this patch. With these two
guarantees, I think it should be safe to mmap
sub-page BAR.

Signed-off-by: Yongji Xie <xyjxie at linux.vnet.ibm.com>
---
 drivers/vfio/pci/vfio_pci.c         |   58 ++++++++++++++++++++++++++++++-----
 drivers/vfio/pci/vfio_pci_private.h |    8 +++++
 2 files changed, 59 insertions(+), 7 deletions(-)

diff --git a/drivers/vfio/pci/vfio_pci.c b/drivers/vfio/pci/vfio_pci.c
index 98059df..c60d790 100644
--- a/drivers/vfio/pci/vfio_pci.c
+++ b/drivers/vfio/pci/vfio_pci.c
@@ -110,13 +110,47 @@ static inline bool vfio_pci_is_vga(struct pci_dev *pdev)
 	return (pdev->class >> 8) == PCI_CLASS_DISPLAY_VGA;
 }
 
+static bool vfio_pci_bar_mmap_supported(struct vfio_pci_device *vdev, int index)
+{
+	struct resource *res = vdev->pdev->resource + index;
+	struct vfio_pci_shadow_resource *shadow_res;
+
+	if (IS_ENABLED(CONFIG_VFIO_PCI_MMAP) && res->flags & IORESOURCE_MEM &&
+		resource_size(res) > 0) {
+		if (resource_size(res) >= PAGE_SIZE)
+			return true;
+
+		if (!(res->start & ~PAGE_MASK)) {
+			/*
+			 * Add shadow resource for sub-page bar whose mmio
+			 * page is exclusive in case that hot-add device's
+			 * bar is assigned into the mem hole.
+			 */
+			shadow_res = kzalloc(sizeof(*shadow_res), GFP_KERNEL);
+			shadow_res->resource.start = res->end + 1;
+			shadow_res->resource.end = res->start + PAGE_SIZE;
+			shadow_res->resource.flags = res->flags;
+			if (request_resource(res->parent,
+					&shadow_res->resource)) {
+				kfree(shadow_res);
+				return false;
+			}
+			shadow_res->index = index;
+			list_add(&shadow_res->res_next,
+					&vdev->shadow_resources_list);
+			return true;
+		}
+	}
+	return false;
+}
+
 static void vfio_pci_try_bus_reset(struct vfio_pci_device *vdev);
 static void vfio_pci_disable(struct vfio_pci_device *vdev);
 
 static int vfio_pci_enable(struct vfio_pci_device *vdev)
 {
 	struct pci_dev *pdev = vdev->pdev;
-	int ret;
+	int ret, bar;
 	u16 cmd;
 	u8 msix_pos;
 
@@ -183,12 +217,17 @@ static int vfio_pci_enable(struct vfio_pci_device *vdev)
 		}
 	}
 
+	for (bar = PCI_STD_RESOURCES; bar <= PCI_STD_RESOURCE_END; bar++) {
+		vdev->bar_mmap_supported[bar] =
+				vfio_pci_bar_mmap_supported(vdev, bar);
+	}
 	return 0;
 }
 
 static void vfio_pci_disable(struct vfio_pci_device *vdev)
 {
 	struct pci_dev *pdev = vdev->pdev;
+	struct vfio_pci_shadow_resource *shadow_res;
 	int i, bar;
 
 	/* Stop the device from further DMA */
@@ -217,6 +256,13 @@ static void vfio_pci_disable(struct vfio_pci_device *vdev)
 		vdev->barmap[bar] = NULL;
 	}
 
+	list_for_each_entry(shadow_res, &vdev->shadow_resources_list,
+				res_next) {
+		list_del(&shadow_res->res_next);
+		release_resource(&shadow_res->resource);
+		kfree(shadow_res);
+	}
+
 	vdev->needs_reset = true;
 
 	/*
@@ -587,9 +633,7 @@ static long vfio_pci_ioctl(void *device_data,
 
 			info.flags = VFIO_REGION_INFO_FLAG_READ |
 				     VFIO_REGION_INFO_FLAG_WRITE;
-			if (IS_ENABLED(CONFIG_VFIO_PCI_MMAP) &&
-			    pci_resource_flags(pdev, info.index) &
-			    IORESOURCE_MEM && info.size >= PAGE_SIZE) {
+			if (vdev->bar_mmap_supported[info.index]) {
 				info.flags |= VFIO_REGION_INFO_FLAG_MMAP;
 				if (info.index == vdev->msix_bar) {
 					ret = msix_sparse_mmap_cap(vdev, &caps);
@@ -1011,16 +1055,16 @@ static int vfio_pci_mmap(void *device_data, struct vm_area_struct *vma)
 		return -EINVAL;
 	if (index >= VFIO_PCI_ROM_REGION_INDEX)
 		return -EINVAL;
-	if (!(pci_resource_flags(pdev, index) & IORESOURCE_MEM))
+	if (!vdev->bar_mmap_supported[index])
 		return -EINVAL;
 
-	phys_len = pci_resource_len(pdev, index);
+	phys_len = PAGE_ALIGN(pci_resource_len(pdev, index));
 	req_len = vma->vm_end - vma->vm_start;
 	pgoff = vma->vm_pgoff &
 		((1U << (VFIO_PCI_OFFSET_SHIFT - PAGE_SHIFT)) - 1);
 	req_start = pgoff << PAGE_SHIFT;
 
-	if (phys_len < PAGE_SIZE || req_start + req_len > phys_len)
+	if (req_start + req_len > phys_len)
 		return -EINVAL;
 
 	if (index == vdev->msix_bar) {
diff --git a/drivers/vfio/pci/vfio_pci_private.h b/drivers/vfio/pci/vfio_pci_private.h
index 8a7d546..0ea4c62 100644
--- a/drivers/vfio/pci/vfio_pci_private.h
+++ b/drivers/vfio/pci/vfio_pci_private.h
@@ -57,9 +57,16 @@ struct vfio_pci_region {
 	u32				flags;
 };
 
+struct vfio_pci_shadow_resource {
+	struct resource		resource;
+	int			index;
+	struct list_head	res_next;
+};
+
 struct vfio_pci_device {
 	struct pci_dev		*pdev;
 	void __iomem		*barmap[PCI_STD_RESOURCE_END + 1];
+	bool			bar_mmap_supported[PCI_STD_RESOURCE_END + 1];
 	u8			*pci_config_map;
 	u8			*vconfig;
 	struct perm_bits	*msi_perm;
@@ -87,6 +94,7 @@ struct vfio_pci_device {
 	int			refcnt;
 	struct eventfd_ctx	*err_trigger;
 	struct eventfd_ctx	*req_trigger;
+	struct list_head	shadow_resources_list;
 };
 
 #define is_intx(vdev) (vdev->irq_type == VFIO_PCI_INTX_IRQ_INDEX)
-- 
1.7.9.5