On 10/20/23 10:19, Eric Auger wrote:
Hi,
On 10/20/23 07:48, Duan, Zhenzhong wrote:
Hi Cédric,
-----Original Message-----
From: Cédric Le Goater <c...@redhat.com>
Sent: Thursday, October 19, 2023 8:18 PM
Subject: Re: [PATCH v2 02/27] vfio: Introduce base object for VFIOContainer and
targetted interface
On 10/19/23 04:29, Duan, Zhenzhong wrote:
-----Original Message-----
From: Cédric Le Goater <c...@redhat.com>
Sent: Wednesday, October 18, 2023 4:04 PM
Subject: Re: [PATCH v2 02/27] vfio: Introduce base object for VFIOContainer
and
targetted interface
On 10/18/23 04:41, Duan, Zhenzhong wrote:
Hi Cédric,
-----Original Message-----
From: Cédric Le Goater <c...@redhat.com>
Sent: Tuesday, October 17, 2023 11:51 PM
Subject: Re: [PATCH v2 02/27] vfio: Introduce base object for VFIOContainer
and
targetted interface
On 10/16/23 10:31, Zhenzhong Duan wrote:
From: Eric Auger <eric.au...@redhat.com>
Introduce a dumb VFIOContainer base object and its targetted interface.
This is willingly not a QOM object because we don't want it to be
visible from the user interface. The VFIOContainer will be smoothly
populated in subsequent patches as well as interfaces.
No fucntional change intended.
Signed-off-by: Eric Auger <eric.au...@redhat.com>
Signed-off-by: Yi Liu <yi.l....@intel.com>
Signed-off-by: Yi Sun <yi.y....@linux.intel.com>
Signed-off-by: Zhenzhong Duan <zhenzhong.d...@intel.com>
---
include/hw/vfio/vfio-common.h | 8 +--
include/hw/vfio/vfio-container-base.h | 82
+++++++++++++++++++++++++++
2 files changed, 84 insertions(+), 6 deletions(-)
create mode 100644 include/hw/vfio/vfio-container-base.h
diff --git a/include/hw/vfio/vfio-common.h b/include/hw/vfio/vfio-
common.h
index 34648e518e..9651cf921c 100644
--- a/include/hw/vfio/vfio-common.h
+++ b/include/hw/vfio/vfio-common.h
@@ -30,6 +30,7 @@
#include <linux/vfio.h>
#endif
#include "sysemu/sysemu.h"
+#include "hw/vfio/vfio-container-base.h"
#define VFIO_MSG_PREFIX "vfio %s: "
@@ -81,6 +82,7 @@ typedef struct VFIOAddressSpace {
struct VFIOGroup;
typedef struct VFIOLegacyContainer {
+ VFIOContainer bcontainer;
That's the parent class, right ?
Right.
VFIOAddressSpace *space;
int fd; /* /dev/vfio/vfio, empowered by the attached groups */
MemoryListener listener;
@@ -200,12 +202,6 @@ typedef struct VFIODisplay {
} dmabuf;
} VFIODisplay;
-typedef struct {
- unsigned long *bitmap;
- hwaddr size;
- hwaddr pages;
-} VFIOBitmap;
-
void vfio_host_win_add(VFIOLegacyContainer *container,
hwaddr min_iova, hwaddr max_iova,
uint64_t iova_pgsizes);
diff --git a/include/hw/vfio/vfio-container-base.h b/include/hw/vfio/vfio-
container-base.h
new file mode 100644
index 0000000000..afc8543d22
--- /dev/null
+++ b/include/hw/vfio/vfio-container-base.h
@@ -0,0 +1,82 @@
+/*
+ * VFIO BASE CONTAINER
+ *
+ * Copyright (C) 2023 Intel Corporation.
+ * Copyright Red Hat, Inc. 2023
+ *
+ * Authors: Yi Liu <yi.l....@intel.com>
+ * Eric Auger <eric.au...@redhat.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+
+ * You should have received a copy of the GNU General Public License
along
+ * with this program; if not, see <http://www.gnu.org/licenses/>.
+ */
+
+#ifndef HW_VFIO_VFIO_BASE_CONTAINER_H
+#define HW_VFIO_VFIO_BASE_CONTAINER_H
+
+#include "exec/memory.h"
+#ifndef CONFIG_USER_ONLY
+#include "exec/hwaddr.h"
+#endif
+
+typedef struct VFIOContainer VFIOContainer;
+typedef struct VFIODevice VFIODevice;
+typedef struct VFIOIOMMUBackendOpsClass
VFIOIOMMUBackendOpsClass;
+
+typedef struct {
+ unsigned long *bitmap;
+ hwaddr size;
+ hwaddr pages;
+} VFIOBitmap;
+
+/*
+ * This is the base object for vfio container backends
+ */
+struct VFIOContainer {
+ VFIOIOMMUBackendOpsClass *ops;
This is unexpected.
I thought that an abstract QOM model for VFIOContainer was going
to be introduced with a VFIOContainerClass with the ops below
(VFIOIOMMUBackendOpsClass).
Then, we would call :
VFIOContainerClass *vcc = VFIO_CONTAINER_GET_CLASS(container);
to get the specific implementation for the current container.
I don't understand the VFIOIOMMUBackendOpsClass pointer and
TYPE_VFIO_IOMMU_BACKEND_OPS. It seems redundant.
The original implementation was abstract QOM model. But it wasn't
accepted,
see https://lore.kernel.org/all/YmuFv2s5TPuw7K%2Fu@yekko/ for details.
I see the idea was challenged, not rejected. I need to dig in further and this
will take time.
Thanks for help looking into it.
+David, Hi David, I'm digging into your concern of using QOM to abstract base
container and legacy VFIOContainer:
"The QOM class of things is visible to the user/config layer via QMP (and
sometimes command line).
It doesn't necessarily correspond to guest visible differences, but it often
does."
AIUI, while it's true when the QOM type includes TYPE_USER_CREATABLE
interface,
otherwise, user can't create an object of this type. Only difference is user
will
see
"object type '%s' isn't supported by object-add" instead of "invalid object
type: %s".
Is your expectation to not permit user to create this object or only want user
to see "invalid object type: %s".
If you mean the first, then I think QOM could be suitable if we don't include
TYPE_USER_CREATABLE interface?
I was imagining some kind of QOM hierarchy under the vfio device
with various QOM interfaces (similar to the ops) to define the
possible IOMMU backends. The fact that we use the IOMMUFD object
>from the command line made it more plausible. I might be mistaking.
Got your point.
This way we introduce a new QOM type "vfio-pci-iommufd" for iommufd support,
and vfio-pci keep same for legacy backend, e.g:
#qemu -object iommufd,id=iommufd0 \
-device vfio-pci-iommufd,iommufd=iommufd0,id=vfio0... \
-device vfio-pci,id=vfio1...
you would need to do that for all types for vfio devices, ap, ccw,
platform. Looks heavy to me. Why would we need to use a different
vfio-pci-* device while we could switch the iommu backend according to
the "iommufd" prop presence. The initial discussion was about QOMyfying
the container instead.
yes.
I took a closer look at the first part which adds the backend ops,
including patch 19 adding the iommufd backend, not saying that I have
identified all the dark corners.
A QOM-like design would have introduced a VFIOLegacyContainer,
inheriting from VFIOContainer (same for VFIOIOMMUFDContainer) with a
VFIOContainerClass to implement the specific backend ops.
VFIOspaprContainer would have made sense also.
But QOM doesn't seem well adapted for the current needs. So let's try
a simpler approach. It seems that VFIOIOMMUBackendOpsClass is
useless. IMO, it could be a callbacks structure like we have for
memory regions initialized with vfio_container_init(). This would
remove some noise around the QOM typeinfo definitions.
'struct vfio_iommu_ops' reads/sounds like a good name.
Can we try that in a v3 ? It should not be such an earthquake.
spapr has some singularities which would be good to isolate in a
vfio_iommu_spapr_ops to remove all the VFIO_SPAPR_TCE_*_IOMMU code in
container.c. vfio_legacy_{add,del}_section_window are SPAPR specific.
FYI, I did some adjustements bc of the recent introduction of iova_ranges
in my branch :
https://github.com/legoater/qemu/commits/vfio-8.2
Thanks,
C.
