Re: [PATCH v2] xen: don't require virtio with grants for non-PV guests

From: Juergen Gross
Date: Fri Jun 17 2022 - 01:49:11 EST


On 17.06.22 02:03, Stefano Stabellini wrote:
On Thu, 16 Jun 2022, Oleksandr wrote:
On 16.06.22 11:56, Juergen Gross wrote:
On 16.06.22 09:31, Oleksandr wrote:

On 16.06.22 08:37, Juergen Gross wrote:


Hello Juergen

Commit fa1f57421e0b ("xen/virtio: Enable restricted memory access using
Xen grant mappings") introduced a new requirement for using virtio
devices: the backend now needs to support the VIRTIO_F_ACCESS_PLATFORM
feature.

This is an undue requirement for non-PV guests, as those can be operated
with existing backends without any problem, as long as those backends
are running in dom0.

Per default allow virtio devices without grant support for non-PV
guests.

Add a new config item to always force use of grants for virtio.

Fixes: fa1f57421e0b ("xen/virtio: Enable restricted memory access using
Xen grant mappings")
Reported-by: Viresh Kumar <viresh.kumar@xxxxxxxxxx>
Signed-off-by: Juergen Gross <jgross@xxxxxxxx>
---
V2:
- remove command line parameter (Christoph Hellwig)
---
  drivers/xen/Kconfig | 9 +++++++++
  include/xen/xen.h   | 2 +-
  2 files changed, 10 insertions(+), 1 deletion(-)

diff --git a/drivers/xen/Kconfig b/drivers/xen/Kconfig
index bfd5f4f706bc..a65bd92121a5 100644
--- a/drivers/xen/Kconfig
+++ b/drivers/xen/Kconfig
@@ -355,4 +355,13 @@ config XEN_VIRTIO
        If in doubt, say n.
+config XEN_VIRTIO_FORCE_GRANT
+    bool "Require Xen virtio support to use grants"
+    depends on XEN_VIRTIO
+    help
+      Require virtio for Xen guests to use grant mappings.
+      This will avoid the need to give the backend the right to map all
+      of the guest memory. This will need support on the backend side
+      (e.g. qemu or kernel, depending on the virtio device types used).
+
  endmenu
diff --git a/include/xen/xen.h b/include/xen/xen.h
index 0780a81e140d..4d4188f20337 100644
--- a/include/xen/xen.h
+++ b/include/xen/xen.h
@@ -56,7 +56,7 @@ extern u64 xen_saved_max_mem_size;
  static inline void xen_set_restricted_virtio_memory_access(void)
  {
-    if (IS_ENABLED(CONFIG_XEN_VIRTIO) && xen_domain())
+    if (IS_ENABLED(CONFIG_XEN_VIRTIO_FORCE_GRANT) || xen_pv_domain())
          platform_set(PLATFORM_VIRTIO_RESTRICTED_MEM_ACCESS);


Looks like, the flag will be *always* set for paravirtualized guests even
if CONFIG_XEN_VIRTIO disabled.

Maybe we should clarify the check?


if (IS_ENABLED(CONFIG_XEN_VIRTIO_FORCE_GRANT) ||
IS_ENABLED(CONFIG_XEN_VIRTIO) && xen_pv_domain())

     platform_set(PLATFORM_VIRTIO_RESTRICTED_MEM_ACCESS);


Yes, we should. I had the function in grant-dma-ops.c in V1, and could drop
the
CONFIG_XEN_VIRTIO dependency for that reason.

I'll wait for more comments before sending V3, though.

ok



Please note, I am happy with current patch and it works in my Arm64 based
environment.

Just one moment to consider.


As it was already mentioned earlier in current thread the
PLATFORM_VIRTIO_RESTRICTED_MEM_ACCESS (former
arch_has_restricted_virtio_memory_access()) is not per device but about the
whole guest. Being set it makes VIRTIO_F_ACCESS_PLATFORM and
VIRTIO_F_VERSION_1 features mandatory for *all* virtio devices in the guest.

The question is “Do we want/need to lift this restriction for some devices
(which backends are trusted so can access all guest memory) at the same time”?
Copy here the original Viresh's question for the convenience:

"I understand from your email that the backends need to offer the
VIRTIO_F_ACCESS_PLATFORM flag now, but should this requirement be a bit soft?
I mean shouldn't we allow both types of backends to run with the same kernel,
ones that offer this feature and others that don't? The ones that don't offer
the feature, should continue to work like they used to, i.e. without the
restricted memory access feature."

Technically this can be possible with HVM.

Let's imagine the following situation:

- Dom0 with backends which don't offer required features for some reason(s)

But running in Dom0 (trusted domain) these backends are not obliged to offer
it (yes they can offer the required features and support grant mappings for
the virtio, but this is not strictly necessary, as they are considered as
trusted so are allowed to access all guest memory).

- DomD with backend which do offer them and require grant mappings for the
virtio

If this is a valid and correct use-case, then we indeed need an ability to
control that per device, otherwise - what is written below doesn't really
matter.

I am wondering whether we can avoid using global
PLATFORM_VIRTIO_RESTRICTED_MEM_ACCESS for Xen guests at all? I assume that all
we need to do (when CONFIG_XEN_VIRTIO is enabled) is to make sure that *only*
Xen grant DMA devices in HVM guests and *all* devices in PV guests offer
required flags.

Below the diff how this could be done w/o an extra options (not completely
tested), although I realize it might look hackish, and a lot more effort is
needed to get it right. In my Arm64 based environment it works, I have tried
to run two backends, the first offered required features and the corresponding
device node had required property, but the second didn’t and there was no
property.

diff --git a/arch/arm/xen/enlighten.c b/arch/arm/xen/enlighten.c
index 1f9c3ba..07eb69f 100644
--- a/arch/arm/xen/enlighten.c
+++ b/arch/arm/xen/enlighten.c
@@ -443,8 +443,6 @@ static int __init xen_guest_init(void)
        if (!xen_domain())
                return 0;

-       xen_set_restricted_virtio_memory_access();
-
        if (!acpi_disabled)
                xen_acpi_guest_init();
        else
diff --git a/arch/x86/xen/enlighten_hvm.c b/arch/x86/xen/enlighten_hvm.c
index 8b71b1d..517a9d8 100644
--- a/arch/x86/xen/enlighten_hvm.c
+++ b/arch/x86/xen/enlighten_hvm.c
@@ -195,8 +195,6 @@ static void __init xen_hvm_guest_init(void)
        if (xen_pv_domain())
                return;

-       xen_set_restricted_virtio_memory_access();
-
        init_hvm_pv_info();

        reserve_shared_info();
diff --git a/arch/x86/xen/enlighten_pv.c b/arch/x86/xen/enlighten_pv.c
index 30d24fe..ca85d14 100644
--- a/arch/x86/xen/enlighten_pv.c
+++ b/arch/x86/xen/enlighten_pv.c
@@ -108,8 +108,6 @@ static DEFINE_PER_CPU(struct tls_descs, shadow_tls_desc);

 static void __init xen_pv_init_platform(void)
 {
-       xen_set_restricted_virtio_memory_access();
-
        populate_extra_pte(fix_to_virt(FIX_PARAVIRT_BOOTMAP));

        set_fixmap(FIX_PARAVIRT_BOOTMAP, xen_start_info->shared_info);
diff --git a/drivers/virtio/virtio.c b/drivers/virtio/virtio.c
index 371e16b..875690a 100644
--- a/drivers/virtio/virtio.c
+++ b/drivers/virtio/virtio.c
@@ -167,6 +167,11 @@ void virtio_add_status(struct virtio_device *dev,
unsigned int status)
 }
 EXPORT_SYMBOL_GPL(virtio_add_status);

+int __weak device_has_restricted_virtio_memory_access(struct device *dev)
+{
+       return platform_has(PLATFORM_VIRTIO_RESTRICTED_MEM_ACCESS);
+}
+
 /* Do some validation, then set FEATURES_OK */
 static int virtio_features_ok(struct virtio_device *dev)
 {
@@ -174,7 +179,7 @@ static int virtio_features_ok(struct virtio_device *dev)

        might_sleep();

-       if (platform_has(PLATFORM_VIRTIO_RESTRICTED_MEM_ACCESS)) {
+       if (device_has_restricted_virtio_memory_access(dev->dev.parent)) {
                if (!virtio_has_feature(dev, VIRTIO_F_VERSION_1)) {
                        dev_warn(&dev->dev,
                                 "device must provide VIRTIO_F_VERSION_1\n");
diff --git a/drivers/xen/grant-dma-ops.c b/drivers/xen/grant-dma-ops.c
index 6586152..da938f6 100644
--- a/drivers/xen/grant-dma-ops.c
+++ b/drivers/xen/grant-dma-ops.c
@@ -11,6 +11,7 @@
 #include <linux/dma-map-ops.h>
 #include <linux/of.h>
 #include <linux/pfn.h>
+#include <linux/virtio_config.h>
 #include <linux/xarray.h>
 #include <xen/xen.h>
 #include <xen/grant_table.h>
@@ -286,6 +287,11 @@ bool xen_is_grant_dma_device(struct device *dev)
        return has_iommu;
 }

+int device_has_restricted_virtio_memory_access(struct device *dev)
+{
+       return (xen_pv_domain() || xen_is_grant_dma_device(dev));
+}
+
 void xen_grant_setup_dma_ops(struct device *dev)
 {
        struct xen_grant_dma_data *data;
diff --git a/include/linux/virtio_config.h b/include/linux/virtio_config.h
index 7949829..b3a455b 100644
--- a/include/linux/virtio_config.h
+++ b/include/linux/virtio_config.h
@@ -559,4 +559,6 @@ static inline void virtio_cwrite64(struct virtio_device
*vdev,
_r;                                                     \
        })

+int device_has_restricted_virtio_memory_access(struct device *dev);
+
 #endif /* _LINUX_VIRTIO_CONFIG_H */
diff --git a/include/xen/xen.h b/include/xen/xen.h
index 0780a81..a99bab8 100644
--- a/include/xen/xen.h
+++ b/include/xen/xen.h
@@ -52,14 +52,6 @@ bool xen_biovec_phys_mergeable(const struct bio_vec *vec1,
 extern u64 xen_saved_max_mem_size;
 #endif

-#include <linux/platform-feature.h>
-
-static inline void xen_set_restricted_virtio_memory_access(void)
-{
-       if (IS_ENABLED(CONFIG_XEN_VIRTIO) && xen_domain())
- platform_set(PLATFORM_VIRTIO_RESTRICTED_MEM_ACCESS);
-}
-
 #ifdef CONFIG_XEN_UNPOPULATED_ALLOC
 int xen_alloc_unpopulated_pages(unsigned int nr_pages, struct page **pages);
 void xen_free_unpopulated_pages(unsigned int nr_pages, struct page **pages);
(END)


I think when x86 HVM gains required support (via ACPI or other means) to
communicate the x86's alternative of "xen,grant-dma" then
xen_is_grant_dma_device() will be just extended to handle that.

Yeah I like this approach:

- on ARM it bases the setting of PLATFORM_VIRTIO_RESTRICTED_MEM_ACCESS
on "xen,grant-dma", as it should be
- it goes beyond my suggestion and it is capable of doing that
per-device, which is awesome
- on x86, it always enables for PV guests as they have no other choice

On top of this we could add a command line option or kconfig option to
force-enable it as well for the benefit of x86/HVM, but I would make
that option x86 specific.

In the end the proper solution would be a per-device setting, as Christoph
already said.

So basically I think we can rip out the PLATFORM_VIRTIO_RESTRICTED_MEM_ACCESS
flag again (which would mean we could rip out the whole platform feature
support again). Instead we should have a platform specific callback in virtio
which replaces the test for PLATFORM_VIRTIO_RESTRICTED_MEM_ACCESS. The callback
would have the virtio device as a parameter.

This callback would be pre-initialized with a function returning always
"false". SEV, TDX and /390 PV could replace it with a function returning
always "true". When CONFIG_XEN_VIRTIO_FORCE_GRANT is set, Xen guests would
return always "true", otherwise they can check whether e.g. "xen,grant-dma"
was set for the device in the device table and return "true" if this is the
case. This scheme would IMO cover all needs.


Juergen

Attachment: OpenPGP_0xB0DE9DD628BF132F.asc
Description: OpenPGP public key

Attachment: OpenPGP_signature
Description: OpenPGP digital signature