Re: [RFC Design Doc] Add vNVDIMM support for Xen

["Jan Beulich" <JBeulich@suse.com>]

>>> On 24.02.16 at 14:28, <haozhong.zhang@intel.com> wrote:
> On 02/18/16 10:17, Jan Beulich wrote:
>> >>> On 01.02.16 at 06:44, <haozhong.zhang@intel.com> wrote:
>> >  This design treats host NVDIMM devices as ordinary MMIO devices:
>> 
>> Wrt the cachability note earlier on, I assume you're aware that with
>> the XSA-154 changes we disallow any cachable mappings of MMIO
>> by default.
>>
> 
> EPT entries that map the host NVDIMM SPAs to guest will be the only
> mapping used for NVDIMM. If the memory type in the last level entries is
> always set to the same type reported by NFIT and the ipat bit is always
> set as well, I think it would not suffer from the cache-type
> inconsistency problem in XSA-154?

This assumes Xen knows the NVDIMM address ranges, which so
far you meant to keep out of Xen iirc. But yes, things surely can
be made work, I simply wanted to point out that there are some
caveats.

>> >  (1) Dom0 Linux NVDIMM driver is responsible to detect (through NFIT)
>> >      and drive host NVDIMM devices (implementing block device
>> >      interface). Namespaces and file systems on host NVDIMM devices
>> >      are handled by Dom0 Linux as well.
>> > 
>> >  (2) QEMU mmap(2) the pmem NVDIMM devices (/dev/pmem0) into its
>> >      virtual address space (buf).
>> > 
>> >  (3) QEMU gets the host physical address of buf, i.e. the host system
>> >      physical address that is occupied by /dev/pmem0, and calls Xen
>> >      hypercall XEN_DOMCTL_memory_mapping to map it to a DomU.
>> > 
>> >  (ACPI part is described in Section 3.3 later)
>> > 
>> >  Above (1)(2) have already been done in current QEMU. Only (3) is
>> >  needed to implement in QEMU. No change is needed in Xen for address
>> >  mapping in this design.
>> > 
>> >  Open: It seems no system call/ioctl is provided by Linux kernel to
>> >        get the physical address from a virtual address.
>> >        /proc/<qemu_pid>/pagemap provides information of mapping from
>> >        VA to PA. Is it an acceptable solution to let QEMU parse this
>> >        file to get the physical address?
>> > 
>> >  Open: For a large pmem, mmap(2) is very possible to not map all SPA
>> >        occupied by pmem at the beginning, i.e. QEMU may not be able to
>> >        get all SPA of pmem from buf (in virtual address space) when
>> >        calling XEN_DOMCTL_memory_mapping.
>> >        Can mmap flag MAP_LOCKED or mlock(2) be used to enforce the
>> >        entire pmem being mmaped?
>> 
>> A fundamental question I have here is: Why does qemu need to
>> map this at all? It shouldn't itself need to access those ranges,
>> since the guest is given direct access. It would seem quite a bit
>> more natural if qemu simply inquired to underlying GFN range(s)
>> and handed those to Xen for translation to MFNs and mapping
>> into guest space.
>>
> 
> The above design is more like a hack on the existing QEMU
> implementation for KVM without modifying the Dom0 kernel.
> 
> Maybe it's better to let QEMU to get SPAs from Dom0 kernel (NVDIMM
> driver) and then pass them to Xen for the address mapping:
> (1) QEMU passes fd of /dev/pmemN or file on /dev/pmemN to Dom0 kernel.
> (2) Dom0 kernel finds and returns all SPAs occupied by /dev/pmemN or
>     portions of /dev/pmemN where the file sits.
> (3) QEMU passes above SPAs, and GMFN where they will be mapped to Xen
>     which maps SPAs to GMFN.

Indeed, and that would also eliminate the second of your Opens
above.

>> > 3.3 Guest ACPI Emulation
>> > 
>> > 3.3.1 My Design
>> > 
>> >  Guest ACPI emulation is composed of two parts: building guest NFIT
>> >  and SSDT that defines ACPI namespace devices for NVDIMM, and
>> >  emulating guest _DSM.
>> > 
>> >  (1) Building Guest ACPI Tables
>> > 
>> >   This design reuses and extends hvmloader's existing mechanism that
>> >   loads passthrough ACPI tables from binary files to load NFIT and
>> >   SSDT tables built by QEMU:
>> >   1) Because the current QEMU does not building any ACPI tables when
>> >      it runs as the Xen device model, this design needs to patch QEMU
>> >      to build NFIT and SSDT (so far only NFIT and SSDT) in this case.
>> > 
>> >   2) QEMU copies NFIT and SSDT to the end of guest memory below
>> >      4G. The guest address and size of those tables are written into
>> >      xenstore (/local/domain/domid/hvmloader/dm-acpi/{address,length}).
>> > 
>> >   3) hvmloader is patched to probe and load device model passthrough
>> >      ACPI tables from above xenstore keys. The detected ACPI tables
>> >      are then appended to the end of existing guest ACPI tables just
>> >      like what current construct_passthrough_tables() does.
>> > 
>> >   Reasons for this design are listed below:
>> >   - NFIT and SSDT in question are quite self-contained, i.e. they do
>> >     not refer to other ACPI tables and not conflict with existing
>> >     guest ACPI tables in Xen. Therefore, it is safe to copy them from
>> >     QEMU and append to existing guest ACPI tables.
>> 
>> How is this not conflicting being guaranteed? In particular I don't
>> see how tables containing AML code and coming from different
>> sources won't possibly cause ACPI name space collisions.
>>
> 
> Really there is no effective mechanism to avoid ACPI name space
> collisions (and other kinds of conflicts) between ACPI tables loaded
> from QEMU and ACPI tables built by hvmloader. Because which ACPI tables
> are loaded is determined by developers, IMO it's developers'
> responsibility to avoid any collisions and conflicts with existing ACPI
> tables.

Right, but this needs to be spelled out and settled on at design
time (i.e. now), rather leaving things unspecified, awaiting the
first clash.

Jan