Re: [RFC PATCH] iommu/xen: Add Xen PV-IOMMU driver

[Jan Beulich <jbeulich@suse.com>]

On 17.06.2024 15:36, Teddy Astie wrote:
> Le 13/06/2024 à 16:32, Jan Beulich a écrit :
>> On 13.06.2024 15:50, Teddy Astie wrote:
>>> @@ -214,6 +215,38 @@ struct xen_add_to_physmap_range {
>>>   };
>>>   DEFINE_GUEST_HANDLE_STRUCT(xen_add_to_physmap_range);
>>>   
>>> +/*
>>> + * With some legacy devices, certain guest-physical addresses cannot safely
>>> + * be used for other purposes, e.g. to map guest RAM.  This hypercall
>>> + * enumerates those regions so the toolstack can avoid using them.
>>> + */
>>> +#define XENMEM_reserved_device_memory_map   27
>>> +struct xen_reserved_device_memory {
>>> +    xen_pfn_t start_pfn;
>>> +    xen_ulong_t nr_pages;
>>> +};
>>> +DEFINE_GUEST_HANDLE_STRUCT(xen_reserved_device_memory);
>>> +
>>> +struct xen_reserved_device_memory_map {
>>> +#define XENMEM_RDM_ALL 1 /* Request all regions (ignore dev union). */
>>> +    /* IN */
>>> +    uint32_t flags;
>>> +    /*
>>> +     * IN/OUT
>>> +     *
>>> +     * Gets set to the required number of entries when too low,
>>> +     * signaled by error code -ERANGE.
>>> +     */
>>> +    unsigned int nr_entries;
>>> +    /* OUT */
>>> +    GUEST_HANDLE(xen_reserved_device_memory) buffer;
>>> +    /* IN */
>>> +    union {
>>> +        struct physdev_pci_device pci;
>>> +    } dev;
>>> +};
>>> +DEFINE_GUEST_HANDLE_STRUCT(xen_reserved_device_memory_map);
>>
>> This is a tools-only (i.e. unstable) sub-function in Xen; even the comment
>> at the top says "toolstack". It is therefore not suitable for use in a
>> kernel.
>>
> IMO this comment actually describes how the toolstack uses the 
> hypercall, but I don't think it is actually reserved for toolstack use.

Well, the canonical version of the header is quite explicit about this,
by having the definition in a __XEN__ || __XEN_TOOLS__ section.

> Or maybe we should allow the kernel to use this hypercall as well.

That's an option to consider.

Jan