Re: [PATCH v3 03/11] iommufd: Introduce IOMMUFD_OBJ_VIOMMU and its related struct

[Zhangfei Gao]

On Thu, 10 Oct 2024 at 00:40, Nicolin Chen <nicolinc@xxxxxxxxxx> wrote:
>
> Add a new IOMMUFD_OBJ_VIOMMU with an iommufd_viommu structure to represent
> a slice of physical IOMMU device passed to or shared with a user space VM.
> This slice, now a vIOMMU object, is a group of virtualization resources of
> a physical IOMMU's, such as:
>  - Security namespace for guest owned ID, e.g. guest-controlled cache tags
>  - Access to a sharable nesting parent pagetable across physical IOMMUs
>  - Virtualization of various platforms IDs, e.g. RIDs and others
>  - Delivery of paravirtualized invalidation
>  - Direct assigned invalidation queues
>  - Direct assigned interrupts
>  - Non-affiliated event reporting
>
> Add a new viommu_alloc op in iommu_ops, for drivers to allocate their own
> vIOMMU structures. And this allocation also needs a free(), so add struct
> iommufd_viommu_ops.
>
> To simplify a vIOMMU allocation, provide a iommufd_viommu_alloc() helper.
> It's suggested that a driver should embed a core-level viommu structure in
> its driver-level viommu struct and call the iommufd_viommu_alloc() helper,
> meanwhile the driver can also implement a viommu ops:
>     struct my_driver_viommu {
>         struct iommufd_viommu core;
>         /* driver-owned properties/features */
>         ....
>     };
>
>     static const struct iommufd_viommu_ops my_driver_viommu_ops = {
>         .free = my_driver_viommu_free,
>         /* future ops for virtualization features */
>         ....
>     };
>
>     static struct iommufd_viommu my_driver_viommu_alloc(...)
>     {
>         struct my_driver_viommu *my_viommu =
>                 iommufd_viommu_alloc(ictx, my_driver_viommu, core,
>                                      my_driver_viommu_ops);
>         /* Init my_viommu and related HW feature */
>         ....
>         return &my_viommu->core;
>     }
>
>     static struct iommu_domain_ops my_driver_domain_ops = {
>         ....
>         .viommu_alloc = my_driver_viommu_alloc,
>     };
>
> To make the Kernel config work between a driver and the iommufd core, put
> the for-driver allocation helpers into a new viommu_api file building with
> CONFIG_IOMMUFD_DRIVER.
>
> Signed-off-by: Nicolin Chen <nicolinc@xxxxxxxxxx>

> diff --git a/drivers/iommu/iommufd/viommu_api.c b/drivers/iommu/iommufd/viommu_api.c
> new file mode 100644
> index 000000000000..c1731f080d6b
> --- /dev/null
> +++ b/drivers/iommu/iommufd/viommu_api.c
> @@ -0,0 +1,57 @@
> +// SPDX-License-Identifier: GPL-2.0-only
> +/* Copyright (c) 2024, NVIDIA CORPORATION & AFFILIATES
> + */
> +
> +#include "iommufd_private.h"
> +
> +struct iommufd_object *iommufd_object_alloc_elm(struct iommufd_ctx *ictx,
> +                                               size_t size,
> +                                               enum iommufd_object_type type)
> +{
> +       struct iommufd_object *obj;
> +       int rc;
> +
> +       obj = kzalloc(size, GFP_KERNEL_ACCOUNT);
> +       if (!obj)
> +               return ERR_PTR(-ENOMEM);
> +       obj->type = type;
> +       /* Starts out bias'd by 1 until it is removed from the xarray */
> +       refcount_set(&obj->shortterm_users, 1);
> +       refcount_set(&obj->users, 1);

here set refcont 1

iommufd_device_bind -> iommufd_object_alloc(ictx, idev,
IOMMUFD_OBJ_DEVICE): refcont -> 1
refcount_inc(&idev->obj.users); refcount -> 2
will cause iommufd_device_unbind fail.

May remove refcount_inc(&idev->obj.users) in iommufd_device_bind

Thanks