CP-20679: populate PGPU.compatibility_metadata

For Nvidia graphics processors, fetch the compatibility metadata
from the driver and store it in the PGPU.compatibility field (a
string-string map) under the "nvidia" key.

Signed-off-by: Thomas Sanders <[email protected]>

Author: thomassa

ocaml/xapi/jbuild

@@ -95,6 +95,7 @@ let () = Printf.ksprintf Jbuild_plugin.V1.send {|
    xcp.network
    xcp.v6
    xcp.memory
+   xcp.gpumon
    xcp.updates
    rrdd-plugin
    xenopsd

ocaml/xapi/xapi_gpumon.ml

@@ -37,3 +37,15 @@ module Gpumon = Daemon_manager.Make(struct
   end)
 
 let with_gpumon_stopped = Gpumon.with_daemon_stopped
+
+module Nvidia = struct
+  let key = "nvidia"
+
+  (* N.B. the pgpu must be in the local host where this function runs *)
+  let get_pgpu_compatibility_metadata ~__context ~pgpu =
+    let get = Gpumon_client.Client.Nvidia.get_pgpu_metadata in
+    let pci = Db.PGPU.get_PCI ~__context ~self:pgpu in
+    let address = Db.PCI.get_pci_id ~__context ~self:pci in
+    [key, get "xapi_gpumon" address]
+
+end (* Nvidia *)

ocaml/xapi/xapi_gpumon.mli

@@ -15,3 +15,18 @@
 (** Stop gpumon if it's running, perform f, then start gpumon if
   * no other threads which require gpumon to be stopped are running. *)
 val with_gpumon_stopped : ?timeout:float -> (unit -> 'a) -> 'a
+
+module Nvidia : sig
+  (** The key used in the metadata assoc-list *)
+  val key: string
+
+  (** Fetch metadata about the PGPU from the driver, and return
+   *  [(key, metadata)] where key is a fixed value and metadata
+   *  is the opaque string of data from the graphics driver.
+   *  IMPORTANT: This must be called on the host that has the GPU installed in it. *)
+  val get_pgpu_compatibility_metadata:
+    __context:Context.t ->
+    pgpu:[ `PGPU ] API.Ref.t ->
+    (string * string Gpumon_client.Client.RPCM.t) list
+
+end

ocaml/xapi/xapi_pci.mli

@@ -26,6 +26,10 @@ val is_class_of_kind : base_class -> int -> bool
 (** Get int value for PCI {class, vendor, device}_id *)
 val int_of_id : string -> int
 
+(** Get string value of int form of PCI {class, vendor, device}_id
+  * (the reverse of int_of_id) *)
+val id_of_int : int -> string
+
 (** Get an identifier for this PCI device **)
 val string_of_pci : __context:Context.t -> self:API.ref_PCI -> string
 

ocaml/xapi/xapi_pgpu.ml

@@ -29,6 +29,14 @@ let calculate_max_capacities ~__context ~pCI ~size ~supported_VGPU_types =
        vgpu_type, max_capacity)
     supported_VGPU_types
 
+let set_compatibility_metadata ~__context ~pgpu =
+  let pci = Db.PGPU.get_PCI ~__context ~self:pgpu in
+  if Db.PCI.get_vendor_id ~__context ~self:pci = Xapi_pci.id_of_int Xapi_vgpu_type.Nvidia.vendor_id then (
+     let () = Db.PGPU.set_compatibility_metadata ~__context ~self:pgpu
+       ~value:(Xapi_gpumon.Nvidia.get_pgpu_compatibility_metadata ~__context ~pgpu)
+     in ()
+  )
+
 let create ~__context ~pCI ~gPU_group ~host ~other_config
     ~supported_VGPU_types ~size ~dom0_access
     ~is_system_display_device =
@@ -46,6 +54,7 @@ let create ~__context ~pCI ~gPU_group ~host ~other_config
     ~self:pgpu ~value:supported_VGPU_types;
   Db.PGPU.set_enabled_VGPU_types ~__context
     ~self:pgpu ~value:supported_VGPU_types;
+  set_compatibility_metadata ~__context ~pgpu;
   debug "PGPU ref='%s' created (host = '%s')" (Ref.string_of pgpu) (Ref.string_of host);
   pgpu
 
@@ -182,6 +191,7 @@ let update_gpus ~__context =
           Db.PGPU.set_is_system_display_device ~__context
             ~self:rf
             ~value:is_system_display_device;
+          set_compatibility_metadata ~__context ~pgpu:rf;
           (rf, rc)
         with Not_found ->
           (* If a new PCI has appeared then we know this is a system boot.