[PATCH v6 0/3] VIRTIO-IOMMU: Set default granule to host page size

[PATCH v6 0/3] VIRTIO-IOMMU: Set default granule to host page size

[Eric Auger]

We used to set the default granule to 4kB but with VFIO assignment
it makes more sense to use the actual host page size.

Indeed when hotplugging a VFIO device protected by a virtio-iommu
on a 64kB/64kB host/guest config, we currently get a qemu crash:

"vfio: DMA mapping failed, unable to continue"

This is due to the hot-attached VFIO device calling
memory_region_iommu_set_page_size_mask() with 64kB granule
whereas the virtio-iommu granule was already frozen to 4kB on
machine init done.

Introduce a new granule property, set this latter to "host"
and introduce a new compat (that sets it to 4k for machine
types older than 9.0).

Note that the new default will prevent 4kB guest on 64kB host
because the granule will be set to 64kB which would be larger
than the guest page size. In that situation, the virtio-iommu
driver fails on viommu_domain_finalise() with
"granule 0x10000 larger than system page size 0x1000".

In that case 4K granule should be used.

To summarize, before the series, the support matrix (credit
to Jean-Philippe Brucker) was:

 Host | Guest | virtio-net | IGB passthrough
  4k  | 4k    | Y          | Y
  64k | 64k   | Y          | N
  64k | 4k    | Y          | N
  4k  | 64k   | Y          | Y

After the series:

 Host | Guest | virtio-net | IGB passthrough
  4k  | 4k    | Y          | Y
  64k | 64k   | Y          | Y
  64k | 4k    | 4K         | N
  4k  | 64k   | Y          | Y

The current limitation of global granule in the virtio-iommu
should be removed and turned into per domain granule. But
until we get this upgraded, this new default is probably
better because I don't think anyone is currently interested in
running a 4KB page size guest with virtio-iommu on a 64KB host.
However supporting 64kB guest on 64kB host with virtio-iommu and
VFIO looks a more important feature.

This series can be found at:
https://github.com/eauger/qemu/tree/granule-v4

Applied on top of
[PATCH v5 0/4] VIRTIO-IOMMU: Introduce an aw-bits option
https://lore.kernel.org/all/20240215084315.863897-1-eric.auger@redhat.com/

History:
v5 -> v6:
- remove one useless header inclusion
- collected Philippe & Zhenzhong's R-b

v4 -> v5:
- use -(n * KiB) (Philippe)
- remove code that can be automatically generated
  and add the new enum in qapi/virtio.json (Philippe).
- Improve commit msg on last patch and collected Philippe's R-b

v3 -> v4:
- Add 8K granule (Richard)

v2 -> v3
- introduce a dedicated granule option to handle the compat


Eric Auger (3):
  qdev: Add a granule_mode property
  virtio-iommu: Add a granule property
  virtio-iommu: Change the default granule to the host page size

 qapi/virtio.json                    | 18 ++++++++++++++++++
 include/hw/qdev-properties-system.h |  3 +++
 include/hw/virtio/virtio-iommu.h    |  2 ++
 hw/core/machine.c                   |  1 +
 hw/core/qdev-properties-system.c    | 14 ++++++++++++++
 hw/virtio/virtio-iommu.c            | 28 +++++++++++++++++++++++++---
 qemu-options.hx                     |  3 +++
 7 files changed, 66 insertions(+), 3 deletions(-)

-- 
2.41.0

[PATCH v6 1/3] qdev: Add a granule_mode property

[Eric Auger]

Introduce a new enum type property allowing to set an
IOMMU granule. Values are 4k, 8k, 16k, 64k and host.
This latter indicates the vIOMMU granule will match
the host page size.

A subsequent patch will add such a property to the
virtio-iommu device.

Signed-off-by: Eric Auger <eric.auger@redhat.com>
Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Zhenzhong Duan <zhenzhong.duan@intel.com>

---
v5 -> v6
- remove #include "hw/virtio/virtio-iommu.h" (Zhenzhong)

v4 -> v5
- remove code that can be automatically generated
  and add the new enum in qapi/virtio.json (Philippe).
  Added Phild's SOB. low case needs to be used due to
  the Jason generation.

v3 -> v4:
- Add 8K
---
 qapi/virtio.json                    | 18 ++++++++++++++++++
 include/hw/qdev-properties-system.h |  3 +++
 hw/core/qdev-properties-system.c    | 14 ++++++++++++++
 3 files changed, 35 insertions(+)

diff --git a/qapi/virtio.json b/qapi/virtio.json
index a79013fe89..95745fdfd7 100644
--- a/qapi/virtio.json
+++ b/qapi/virtio.json
@@ -957,3 +957,21 @@
 
 { 'struct': 'DummyVirtioForceArrays',
   'data': { 'unused-iothread-vq-mapping': ['IOThreadVirtQueueMapping'] } }
+
+##
+# @GranuleMode:
+#
+# @4k: granule page size of 4KiB
+#
+# @8k: granule page size of 8KiB
+#
+# @16k: granule page size of 16KiB
+#
+# @64k: granule page size of 64KiB
+#
+# @host: granule matches the host page size
+#
+# Since: 9.0
+##
+{ 'enum': 'GranuleMode',
+  'data': [ '4k', '8k', '16k', '64k', 'host' ] }
diff --git a/include/hw/qdev-properties-system.h b/include/hw/qdev-properties-system.h
index 06c359c190..626be87dd3 100644
--- a/include/hw/qdev-properties-system.h
+++ b/include/hw/qdev-properties-system.h
@@ -8,6 +8,7 @@ extern const PropertyInfo qdev_prop_macaddr;
 extern const PropertyInfo qdev_prop_reserved_region;
 extern const PropertyInfo qdev_prop_multifd_compression;
 extern const PropertyInfo qdev_prop_mig_mode;
+extern const PropertyInfo qdev_prop_granule_mode;
 extern const PropertyInfo qdev_prop_losttickpolicy;
 extern const PropertyInfo qdev_prop_blockdev_on_error;
 extern const PropertyInfo qdev_prop_bios_chs_trans;
@@ -47,6 +48,8 @@ extern const PropertyInfo qdev_prop_iothread_vq_mapping_list;
 #define DEFINE_PROP_MIG_MODE(_n, _s, _f, _d) \
     DEFINE_PROP_SIGNED(_n, _s, _f, _d, qdev_prop_mig_mode, \
                        MigMode)
+#define DEFINE_PROP_GRANULE_MODE(_n, _s, _f, _d) \
+    DEFINE_PROP_SIGNED(_n, _s, _f, _d, qdev_prop_granule_mode, GranuleMode)
 #define DEFINE_PROP_LOSTTICKPOLICY(_n, _s, _f, _d) \
     DEFINE_PROP_SIGNED(_n, _s, _f, _d, qdev_prop_losttickpolicy, \
                         LostTickPolicy)
diff --git a/hw/core/qdev-properties-system.c b/hw/core/qdev-properties-system.c
index 1a396521d5..b45e90edb2 100644
--- a/hw/core/qdev-properties-system.c
+++ b/hw/core/qdev-properties-system.c
@@ -679,6 +679,20 @@ const PropertyInfo qdev_prop_mig_mode = {
     .set_default_value = qdev_propinfo_set_default_value_enum,
 };
 
+/* --- GranuleMode --- */
+
+QEMU_BUILD_BUG_ON(sizeof(GranuleMode) != sizeof(int));
+
+const PropertyInfo qdev_prop_granule_mode = {
+    .name = "GranuleMode",
+    .description = "granule_mode values, "
+                   "4k, 8k, 16k, 64k, host",
+    .enum_table = &GranuleMode_lookup,
+    .get = qdev_propinfo_get_enum,
+    .set = qdev_propinfo_set_enum,
+    .set_default_value = qdev_propinfo_set_default_value_enum,
+};
+
 /* --- Reserved Region --- */
 
 /*
-- 
2.41.0

[PATCH v6 2/3] virtio-iommu: Add a granule property

[Eric Auger]

This allows to choose which granule will be used by
default by the virtio-iommu. Current page size mask
default is qemu_target_page_mask so this translates
into a 4K granule.

Signed-off-by: Eric Auger <eric.auger@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Zhenzhong Duan <zhenzhong.duan@intel.com>

---
v4 -> v5:
- use -(n * KiB) (Phild)

v3 -> v4:
- granule_mode introduction moved to that patch
---
 include/hw/virtio/virtio-iommu.h |  2 ++
 hw/virtio/virtio-iommu.c         | 28 +++++++++++++++++++++++++---
 qemu-options.hx                  |  3 +++
 3 files changed, 30 insertions(+), 3 deletions(-)

diff --git a/include/hw/virtio/virtio-iommu.h b/include/hw/virtio/virtio-iommu.h
index 5fbe4677c2..f2785f7997 100644
--- a/include/hw/virtio/virtio-iommu.h
+++ b/include/hw/virtio/virtio-iommu.h
@@ -24,6 +24,7 @@
 #include "hw/virtio/virtio.h"
 #include "hw/pci/pci.h"
 #include "qom/object.h"
+#include "qapi/qapi-types-virtio.h"
 
 #define TYPE_VIRTIO_IOMMU "virtio-iommu-device"
 #define TYPE_VIRTIO_IOMMU_PCI "virtio-iommu-pci"
@@ -67,6 +68,7 @@ struct VirtIOIOMMU {
     Notifier machine_done;
     bool granule_frozen;
     uint8_t aw_bits;
+    GranuleMode granule_mode;
 };
 
 #endif
diff --git a/hw/virtio/virtio-iommu.c b/hw/virtio/virtio-iommu.c
index 2ec5ef3cd1..33e0520bc8 100644
--- a/hw/virtio/virtio-iommu.c
+++ b/hw/virtio/virtio-iommu.c
@@ -29,6 +29,7 @@
 #include "sysemu/reset.h"
 #include "sysemu/sysemu.h"
 #include "qemu/reserved-region.h"
+#include "qemu/units.h"
 #include "qapi/error.h"
 #include "qemu/error-report.h"
 #include "trace.h"
@@ -1115,8 +1116,8 @@ static int virtio_iommu_notify_flag_changed(IOMMUMemoryRegion *iommu_mr,
 }
 
 /*
- * The default mask (TARGET_PAGE_MASK) is the smallest supported guest granule,
- * for example 0xfffffffffffff000. When an assigned device has page size
+ * The default mask depends on the "granule" property. For example, with
+ * 4K granule, it is -(4 * KiB). When an assigned device has page size
  * restrictions due to the hardware IOMMU configuration, apply this restriction
  * to the mask.
  */
@@ -1313,7 +1314,26 @@ static void virtio_iommu_device_realize(DeviceState *dev, Error **errp)
      * in vfio realize
      */
     s->config.bypass = s->boot_bypass;
-    s->config.page_size_mask = qemu_target_page_mask();
+
+    switch (s->granule_mode) {
+    case GRANULE_MODE_4K:
+        s->config.page_size_mask = -(4 * KiB);
+        break;
+    case GRANULE_MODE_8K:
+        s->config.page_size_mask = -(8 * KiB);
+        break;
+    case GRANULE_MODE_16K:
+        s->config.page_size_mask = -(16 * KiB);
+        break;
+    case GRANULE_MODE_64K:
+        s->config.page_size_mask = -(64 * KiB);
+        break;
+    case GRANULE_MODE_HOST:
+        s->config.page_size_mask = qemu_real_host_page_mask();
+        break;
+    default:
+        error_setg(errp, "Unsupported granule mode");
+    }
     if (s->aw_bits < 32 || s->aw_bits > 64) {
         error_setg(errp, "aw-bits must be within [32,64]");
     }
@@ -1527,6 +1547,8 @@ static Property virtio_iommu_properties[] = {
                      TYPE_PCI_BUS, PCIBus *),
     DEFINE_PROP_BOOL("boot-bypass", VirtIOIOMMU, boot_bypass, true),
     DEFINE_PROP_UINT8("aw-bits", VirtIOIOMMU, aw_bits, 0),
+    DEFINE_PROP_GRANULE_MODE("granule", VirtIOIOMMU, granule_mode,
+                             GRANULE_MODE_4K),
     DEFINE_PROP_END_OF_LIST(),
 };
 
diff --git a/qemu-options.hx b/qemu-options.hx
index 3b670758b0..c7b43b67d5 100644
--- a/qemu-options.hx
+++ b/qemu-options.hx
@@ -1179,6 +1179,9 @@ SRST
     ``aw-bits=val`` (val between 32 and 64, default depends on machine)
         This decides the address width of IOVA address space. It defaults
         to 39 bits on q35 machines and 48 bits on ARM virt machines.
+    ``granule=val`` (possible values are 4K, 8K, 16K, 64K and host)
+        This decides the default granule to be be exposed by the
+        virtio-iommu. If host, the granule matches the host page size.
 
 ERST
 
-- 
2.41.0

[PATCH v6 3/3] virtio-iommu: Change the default granule to the host page size

[Eric Auger]

We used to set the default granule to 4KB but with VFIO assignment
it makes more sense to use the actual host page size.

Indeed when hotplugging a VFIO device protected by a virtio-iommu
on a 64kB/64kB host/guest config, we current get a qemu crash:

"vfio: DMA mapping failed, unable to continue"

This is due to the hot-attached VFIO device calling
memory_region_iommu_set_page_size_mask() with 64kB granule
whereas the virtio-iommu granule was already frozen to 4KB on
machine init done.

Set the granule property to "host" and introduce a new compat.
The page size mask used before 9.0 was qemu_target_page_mask().
Since the virtio-iommu currently only supports x86_64 and aarch64,
this matched a 4KB granule.

Note that the new default will prevent 4kB guest on 64kB host
because the granule will be set to 64kB which would be larger
than the guest page size. In that situation, the virtio-iommu
driver fails on viommu_domain_finalise() with
"granule 0x10000 larger than system page size 0x1000".

In that case the workaround is to request 4K granule.

The current limitation of global granule in the virtio-iommu
should be removed and turned into per domain granule. But
until we get this upgraded, this new default is probably
better because I don't think anyone is currently interested in
running a 4KB page size guest with virtio-iommu on a 64KB host.
However supporting 64kB guest on 64kB host with virtio-iommu and
VFIO looks a more important feature.

Signed-off-by: Eric Auger <eric.auger@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>

---

v4 -> v5
- use low case, mandated by the jason qapi
---
 hw/core/machine.c        | 1 +
 hw/virtio/virtio-iommu.c | 2 +-
 2 files changed, 2 insertions(+), 1 deletion(-)

diff --git a/hw/core/machine.c b/hw/core/machine.c
index 70ac96954c..56f38b6579 100644
--- a/hw/core/machine.c
+++ b/hw/core/machine.c
@@ -35,6 +35,7 @@
 
 GlobalProperty hw_compat_8_2[] = {
     { TYPE_VIRTIO_IOMMU_PCI, "aw-bits", "64" },
+    { TYPE_VIRTIO_IOMMU_PCI, "granule", "4k" },
 };
 const size_t hw_compat_8_2_len = G_N_ELEMENTS(hw_compat_8_2);
 
diff --git a/hw/virtio/virtio-iommu.c b/hw/virtio/virtio-iommu.c
index 33e0520bc8..6831446e29 100644
--- a/hw/virtio/virtio-iommu.c
+++ b/hw/virtio/virtio-iommu.c
@@ -1548,7 +1548,7 @@ static Property virtio_iommu_properties[] = {
     DEFINE_PROP_BOOL("boot-bypass", VirtIOIOMMU, boot_bypass, true),
     DEFINE_PROP_UINT8("aw-bits", VirtIOIOMMU, aw_bits, 0),
     DEFINE_PROP_GRANULE_MODE("granule", VirtIOIOMMU, granule_mode,
-                             GRANULE_MODE_4K),
+                             GRANULE_MODE_HOST),
     DEFINE_PROP_END_OF_LIST(),
 };
 
-- 
2.41.0