RE: [ACPI Code First ECN v2]: Generic Port, performace data for hotplug memory

[Vikram Sethi <vsethi@nvidia.com>]

Hi Dan, 
> From: Williams, Dan J <dan.j.williams@intel.com>
> While the platform BIOS is able to describe the performance
> characteristics of CXL memory that is present at boot, it is unable to
> statically enumerate the performance of CXL memory hot inserted
> post-boot. The OS can enumerate most of the characteristics from link
> registers and CDAT, but the performance from the CPU to the host
> bridge, for example, is not enumerated by PCIE or CXL. Introduce an
> ACPI mechanism for this purpose.
> 
> ---
> 
> # Title: Introduce a Generic Port for hotplug memory buses like CXL
> 
> # Status: Draft v2
> 
> # Document: ACPI Specification 6.4
> 
> # License
> SPDX-License Identifier: CC-BY-4.0
> 
> # Submitter:
> * Sponsor: Dan Williams, Intel
> * Creators/Contributors:
>     * Mahesh Natu, Intel
>     * Chet Douglas, Intel
>     * Deepak Shivakumar, Intel
>     * Jonathan Cameron, Huawei
> 
> # Changes since v1
>   * Rename Generic Target to Generic Port and make a new distinct SRAT
>     type independent of Generic Initiator (Jonathan)
>   * Clarify that this new "Port" concept is not limited to CXL. It is a
>     generic way to describe the performance of static paths to
>     dynamically added system memory (Mahesh)
> 
> # Summary of the Change
> Introduce a new "Generic Port" type to the SRAT to describe the
> performance from CPU and other initiator domains to the root of a CXL
> topology, or any other topology that might dynamically add system memory
> behind the "Port". This is in support of, but not limited to, the OS
> being able to enumerate the performance topology for dynamically added /
> discovered CXL Memory Device endpoints.
> 
> # Benefits of the Change
> Consider the case of a system with a set of CXL Host Bridges (ACPI0016),
> and some endpoints attached at boot. In that scenario the platform
> firmware is able to enumerate those devices, enumerate and map CXL
> memory into the system physical memory address space, and generate the
> typical static SRAT/SLIT/HMAT set of tables describing CXL attached
> memory. Now, consider the case where devices are dynamically added and
> enumerated post boot, i.e. post generation of the static memory tables.
> In this scenario platform firmware is unable to perform the end-to-end
> enumeration necessary to populate SRAT and HMAT for the endpoints that
> may be hot-inserted behind those bridges post power-on. The
> address-range is unknown so SRAT can not be pre-populated in the typical
> way that hotplug system memory is enumerated. Even if a static address
> range was set aside for future hotplug the performance is unknown (no
> CDAT nor interleave configuration) so HMAT can not be pre-populated.
> 
> However, what is known to platform firmware that generates the
> SRAT/SLIT/HMAT and is the performance characteristics of the path
> between CPU and Generic Initiators to the Generic Port (e.g. CXL Host
> Bridge). With the addition of a Generic Port proximity domain to the
> SRAT then the SLIT and HMAT can enumerate the platform-static component
> of a given edge in the platform-performance topology graph. It enables
> the OS to build out a performance mapping for system memory address
> ranges dynamically discovered, or provisioned, behind a Generic Port.
> The OS mapping takes into account the Generic Port performance (as
> either an initiator or a target), the interleave configuration, and the
> bus enumerable performance characteristics (link latency, bandwidth,
> switch traversals) to supplement the static HMAT data enumerated at
> boot.
> 
> # Impact of the Change
> A new SRAT type requires non-conforming system software to ignore the
> new type in the SRAT, ignore any coordinate in the SLIT that includes
> the associated port's proximity domain, and ignore any coordinate in the
> HMAT that includes the port's proximity domain as either an initiator or
> a target.
> 
> In contrast, conforming system software need only consult the Generic
> Port data to optionally extend the enumeration and distinguish Port
> attached initiators and memory targets from the existing set of
> enumerated proximity domains.
> 
> A conforming implementation also has the option to ignore the Generic Port
> contribution to the performance,
> 
> in either a row, or col  to be considered by system software that parses
> SRAT, SLIT, and HMAT. Given that the OS still needs to dynamically
> enumerate and instantiate the memory ranges and initiators behind the
> Generic Port. The assumption is that operating systems that do not
> support native CXL enumeration will ignore this data in the HMAT, while
> CXL native enumeration aware environments will use this fragment of the
> performance path to calculate the performance characteristics.
> 
> # References
> * Compute Express Link Specification v2.0,
> <https://www.computeexpresslink.org/>
> 
> # Detailed Description of the Change
> 
> * Section 5.2.16 System Resource Affinity Table (SRAT) add another
>   bullet for Generic Ports:
> 
>         * generic ports (e.g. host bridges that can dynamically discover
>           new initiators and instantiate new memory range targets)
> 
> * Add new section 5.2.16.7 Generic Port Affinity Structure:
>   The Generic Port Affinity Structure provides an association between a
>   proximity domain number and a device handle representing a Generic
>   Port (e.g. CXL Host Bridge, or similar device that hosts a dynamic
>   topology of memory ranges and/or initiators).
> 
>   Support of Generic Port Affinity Structures by an OSPM is optional.
> 
> * Add a table describing the Generic Port Affinity Structure (Table
>   5.xx):
> 
> 
> | Field  | Byte Length | Byte Offset | Description                  |
> | :----- | :---        | :---        | :--------------------------- |
> | Type   | 1           | 0           | 6 Generic Port Structure     |
> | Length | 1           | 1           | 32                           |
> | Reserved | 1         | 2           | Reserved and must be zero    |
> | Device Handle Type | 1 | 3 | Device Handle Type: See 5.2.16.6 Generic Initiator
> Affinity Structure for the possible device handle types and their format. |
> | Proximity Domain | 4 | 4 | The proximity domain to identify the performance of
> this port in the HMAT. |
> | Device Handle | 16   | 8           | Device Handle of the Generic Port, see Table
> 5.57 and 5.58 for a description of this field. |
> | Flags  | 4           | 24          | See table 5.59 for a description of this field. |
> | Reserved | 4         | 28          | Reserved and must be zero.   |
> 
> * Replace all instances of "Initiator" with "Initiator / Port" in "Table
>   5.59 Flags - Generic Initiator Affinity Structure", including the
>   table name.

I wanted to discuss the implications of a CXL host bridge implementation that 
does not set the "Architectural Transactions" bit/flag in the aforementioned 
Flags in Table 5.59.

Since the kernel would be expecting all "System RAM" to have equivalent 
Functional properties, if HDM cannot have all the same functionality, then in
the absence of ISA specific ACPI tables clarifying what architectural feature isn't
supported, the kernel may be forced to not online the HDM memory as system 
RAM. If there is more granular expression of what features are lacking in a ISA
Specific table (eg Memory Tagging/Memory Protection keys not supported),
the kernel could choose to not enable that feature in all of system RAM (if 
discrepancy discovered at boot), but still online the HDM as System RAM.

To that end, it may be useful to clarify this to host vendors by way of an 
Implementation note (ideally in the CXL specification). Something like:
"CXL hosts are encouraged to support all architectural features in HDM 
as they do in CPU attached memory to avoid either the memory from 
being onlined as System RAM, or the architectural feature being disabled. 
Hosts must indicate architectural parity for HDM in ACPI SRAT 
“Generic Port” flags “Architectural transactions” bit by setting it to 1. 
A port that sets this bit to 0 will need ISA specific ways/ACPI tables to 
describe which specific ISA features would not work in HDM, so an OS 
could disable that memory or that feature."

Thoughts?

Thanks,
Vikram