Cleanup version of the xelink support

Author: hsyrja

cmd/gpu_fakedev/configs/2x4-PVC-xelink.json

@@ -0,0 +1,13 @@
+{
+	"Info": "2x 4 tile 4 GiB PVC [Ponte Vecchio] GPUs",
+	"DevCount": 2,
+        "TilesPerDev": 4,
+        "DevsPerNode": 1,
+	"DevMemSize": 4294967296,
+	"Capabilities": {
+		"platform": "fake_PVC",
+                "connections": "0.1-0.0_0.2-0.0_0.3-0.0_1.0-0.0_1.1-0.0_1.2-0.0_1.3-0.0_0.2-0.1_0.3-0.1_1.0-0.1_1.1-0.1_1.2-0.1_1.3-0.1_0.3-0.2_1.0-0.2_1.1-0.2_1.2-0.2_1.3-0.2_1.0-0.3_1.1-0.3_1.2-0.3_1.3-0.3_1.1-1.0_1.2-1.0_1.3-1.0_1.2-1.1_1.3-1.1_1.3-1.2",
+                "connection-topology": "RAW"
+	}
+}
+

cmd/gpu_fakedev/configs/8x2-PVC-xelink.json

@@ -0,0 +1,12 @@
+{
+	"Info": "8x 4 GiB PVC [Ponte Vecchio] GPUs",
+	"DevCount": 8,
+        "TilesPerDev": 2,
+        "DevsPerNode": 2,
+	"DevMemSize": 4294967296,
+	"Capabilities": {
+		"platform": "fake_PVC",
+                "connections": "",
+                "connection-topology": "FULL"
+	}
+}

cmd/gpu_fakedev/gpu_fakedev.go

@@ -1,4 +1,4 @@
-// Copyright 2021-2022 Intel Corporation. All Rights Reserved.
+// Copyright 2021-2023 Intel Corporation. All Rights Reserved.
 //
 // Licensed under the Apache License, Version 2.0 (the "License");
 // you may not use this file except in compliance with the License.
@@ -44,6 +44,7 @@ import (
 	"os"
 	"path/filepath"
 	"strconv"
+	"strings"
 
 	"golang.org/x/sys/unix"
 )
@@ -58,9 +59,11 @@ const (
 	devfsPath  = "dev"
 	mib        = 1024.0 * 1024.0
 	// null device major, minor on linux.
-	devNullMajor = 1
-	devNullMinor = 3
-	devNullType  = unix.S_IFCHR
+	devNullMajor    = 1
+	devNullMinor    = 3
+	devNullType     = unix.S_IFCHR
+	maxK8sLabelSize = 63
+	fullyConnected  = "FULL"
 )
 
 var verbose bool
@@ -270,6 +273,8 @@ func generateDriFiles(opts genOptions) {
 	log.Printf("Generating fake DRI device(s) sysfs, debugfs and devfs content under '%s' & '%s'",
 		sysfsPath, devfsPath)
 
+	makeXelinkSideCar(opts.Capabilities["connection-topology"], opts.DevCount, opts.TilesPerDev, opts.Capabilities["connections"])
+
 	opts.dirs, opts.files = 0, 0
 	for i := 0; i < opts.DevCount; i++ {
 		if err := addSysfsDriTree(sysfsPath, &opts, i); err != nil {
@@ -291,6 +296,88 @@ func generateDriFiles(opts genOptions) {
 	log.Printf("Done, created %d dirs, %d devices and %d files.", opts.dirs, opts.devs, opts.files)
 }
 
+func makeXelinkSideCar(topology string, gpus, tiles int, connections string) {
+	if topology != fullyConnected {
+		log.Printf("XELINK: generate xelink sidecar label file, using (GPUs: %d, Tiles: %d)", gpus, tiles)
+	} else {
+		log.Printf("XELINK: generate xelink sidecar label file, using (GPUs: %d, Tiles: %d, Topology: %s)", gpus, tiles, topology)
+	}
+
+	if topology == fullyConnected {
+		saveSideCarFile(buildConnectionList(gpus, tiles))
+	} else {
+		saveSideCarFile(connections)
+	}
+}
+
+func buildConnectionList(gpus, tiles int) string {
+	mm := 0
+
+	var nodes = make([]string, 0)
+
+	for mm < gpus {
+		nn := 0
+		for nn < tiles {
+			nodes = append(nodes, fmt.Sprintf("%d.%d", mm, nn))
+			nn++
+		}
+		mm++
+	}
+
+	var links = make(map[string]struct{}, 0)
+
+	var smap = make([]string, 0)
+
+	for _, from := range nodes {
+		for _, to := range nodes {
+			// no self links, TODO ignore in-gpu xelinks
+			if to == from {
+				continue
+			}
+
+			link := fmt.Sprintf("%s-%s", to, from)
+
+			reverselink := fmt.Sprintf("%s-%s", from, to)
+			if _, exists := links[reverselink]; !exists {
+				links[link] = struct{}{}
+
+				smap = append(smap, link)
+			}
+		}
+	}
+
+	return fmt.Sprintf("%s", strings.Join(smap, "_"))
+}
+
+func saveSideCarFile(connections string) {
+	f, err := os.Create("xpum-sidecar-labels.txt")
+	if err != nil {
+		panic(err)
+	}
+	defer f.Close()
+
+	// Write first line without Z prefix
+	line := fmt.Sprintf("xpumanager.intel.com/xe-links=%s", connections[:min(len(connections), maxK8sLabelSize)])
+	fmt.Println(line)
+
+	if _, err := f.WriteString(line + "\n"); err != nil {
+		panic(err)
+	}
+
+	index := 2
+
+	// Write next lines with Z prefix
+	for i := maxK8sLabelSize; i < len(connections); i += (maxK8sLabelSize - 1) {
+		line := fmt.Sprintf("xpumanager.intel.com/xe-links%d=Z%s", index, connections[i:min(len(connections), i+maxK8sLabelSize-1)])
+		fmt.Println(line)
+
+		if _, err := f.WriteString(line + "\n"); err != nil {
+			panic(err)
+		}
+		index++
+	}
+}
+
 // getOptions parses options from given JSON file, validates and returns them.
 func getOptions(name string) genOptions {
 	if name == "" {