On 15.07.20 13:51, Michael S. Tsirkin wrote:

On Wed, Jul 15, 2020 at 06:16:59PM +0800, Jason Wang wrote:

quoted

On 2020/7/15 下午5:50, Michael S. Tsirkin wrote:

quoted

On Wed, Jul 15, 2020 at 10:31:09AM +0200, Pierre Morel wrote:

quoted

If protected virtualization is active on s390, the virtio queues are
not accessible to the host, unless VIRTIO_F_IOMMU_PLATFORM has been
negotiated. Use the new arch_validate_virtio_features() interface to
fail probe if that's not the case, preventing a host error on access
attempt.

Signed-off-by: Pierre Morel <redacted>
Reviewed-by: Cornelia Huck <cohuck@redhat.com>
Acked-by: Halil Pasic <pasic@linux.ibm.com>
Acked-by: Christian Borntraeger <redacted>
---
  arch/s390/mm/init.c | 28 ++++++++++++++++++++++++++++
  1 file changed, 28 insertions(+)

diff --git a/arch/s390/mm/init.c b/arch/s390/mm/init.c
index 6dc7c3b60ef6..d39af6554d4f 100644
--- a/arch/s390/mm/init.c
+++ b/arch/s390/mm/init.c

@@ -45,6 +45,7 @@
  #include <asm/kasan.h>
  #include <asm/dma-mapping.h>
  #include <asm/uv.h>
+#include <linux/virtio_config.h>
  pgd_t swapper_pg_dir[PTRS_PER_PGD] __section(.bss..swapper_pg_dir);

@@ -161,6 +162,33 @@ bool force_dma_unencrypted(struct device *dev)
  	return is_prot_virt_guest();
  }
+/*
+ * arch_validate_virtio_features
+ * @dev: the VIRTIO device being added
+ *
+ * Return an error if required features are missing on a guest running
+ * with protected virtualization.
+ */
+int arch_validate_virtio_features(struct virtio_device *dev)
+{
+	if (!is_prot_virt_guest())
+		return 0;
+
+	if (!virtio_has_feature(dev, VIRTIO_F_VERSION_1)) {
+		dev_warn(&dev->dev,
+			 "legacy virtio not supported with protected virtualization\n");
+		return -ENODEV;
+	}
+
+	if (!virtio_has_feature(dev, VIRTIO_F_IOMMU_PLATFORM)) {
+		dev_warn(&dev->dev,
+			 "support for limited memory access required for protected virtualization\n");
+		return -ENODEV;
+	}
+
+	return 0;
+}
+
  /* protected virtualization */
  static void pv_init(void)
  {

What bothers me here is that arch code depends on virtio now.
It works even with a modular virtio when functions are inline,
but it seems fragile: e.g. it breaks virtio as an out of tree module,
since layout of struct virtio_device can change.

If you prefer that, we can simply create an arch/s390/kernel/virtio.c ?