layered: cpuset support, rename and bugfix

Author: likewhatevs

scheds/rust/scx_layered/examples/cpuset.json

@@ -5,7 +5,7 @@
       "matches": [
         [
           {
-            "CommPrefix": "geekbench"
+            "CgroupPrefix": "system.slice/docker"
           }
         ]
       ],

scheds/rust/scx_layered/src/bpf/intf.h

@@ -30,7 +30,7 @@ enum consts {
 	MAX_TASKS		= 131072,
 	MAX_PATH		= 4096,
 	MAX_NUMA_NODES		= 64,
-	MAX_CONTAINERS		= 64,
+	MAX_CPUSETS		= 64,
 	MAX_LLCS		= 64,
 	MAX_COMM		= 16,
 	MAX_LAYER_MATCH_ORS	= 32,

scheds/rust/scx_layered/src/bpf/main.bpf.c

@@ -34,11 +34,11 @@ const volatile u64 numa_cpumasks[MAX_NUMA_NODES][MAX_CPUS / 64];
 const volatile u32 llc_numa_id_map[MAX_LLCS];
 const volatile u32 cpu_llc_id_map[MAX_CPUS];
 const volatile u32 nr_layers = 1;
-const volatile u32 nr_containers = 1;
+const volatile u32 nr_cpusets = 1;
 const volatile u32 nr_nodes = 32;	/* !0 for veristat, set during init */
 const volatile u32 nr_llcs = 32;	/* !0 for veristat, set during init */
 const volatile bool smt_enabled = true;
-const volatile bool enable_container = true;
+const volatile bool enable_cpuset = true;
 const volatile bool has_little_cores = true;
 const volatile bool xnuma_preemption = false;
 const volatile s32 __sibling_cpu[MAX_CPUS];
@@ -54,7 +54,7 @@ const volatile u64 min_open_layer_disallow_preempt_after_ns;
 const volatile u64 lo_fb_wait_ns = 5000000;	/* !0 for veristat */
 const volatile u32 lo_fb_share_ppk = 128;	/* !0 for veristat */
 const volatile bool percpu_kthread_preempt = true;
-const volatile u64 cpuset_fakemasks[MAX_CONTAINERS][MAX_CPUS / 64];
+const volatile u64 cpuset_fakemasks[MAX_CPUSETS][MAX_CPUS / 64];
 
 /* Flag to enable or disable antistall feature */
 const volatile bool enable_antistall = true;
@@ -81,7 +81,7 @@ struct cpumask_box {
 
 struct {
     __uint(type, BPF_MAP_TYPE_PERCPU_ARRAY);
-    __uint(max_entries, MAX_CONTAINERS);
+    __uint(max_entries, MAX_CPUSETS);
     __type(key, u32);
     __type(value, struct cpumask_box);
 } cpuset_cpumask SEC(".maps");
@@ -1349,10 +1349,10 @@ void BPF_STRUCT_OPS(layered_enqueue, struct task_struct *p, u64 enq_flags)
 	 * without making the whole scheduler node aware and should only be used
 	 * with open layers on non-saturated machines to avoid possible stalls.
 	 */
-	if ((!taskc->all_cpus_allowed &&
-		!(layer->allow_node_aligned && taskc->cpus_node_aligned)) ||
-		!(enable_container && taskc->cpus_cpuset_aligned) ||
-		!layer->nr_cpus) {
+	if ((!taskc->all_cpus_allowed && 
+		!((layer->allow_node_aligned && taskc->cpus_node_aligned) || 
+			(enable_cpuset && taskc->cpus_cpuset_aligned))) 
+		|| !layer->nr_cpus) {
 
 		taskc->dsq_id = task_cpuc->lo_fb_dsq_id;
 		/*
@@ -2617,8 +2617,8 @@ static void refresh_cpus_flags(struct task_ctx *taskc,
 			break;
 		}
 	}
-	if (enable_container) {
-		bpf_for(container_id, 0, nr_containers) {
+	if (enable_cpuset) {
+		bpf_for(container_id, 0, nr_cpusets) {
 			struct cpumask_box* box;
 			box = bpf_map_lookup_elem(&cpuset_cpumask, &container_id);
 			if (!box || !box->mask) {
@@ -3338,8 +3338,8 @@ s32 BPF_STRUCT_OPS_SLEEPABLE(layered_init)
 
 
 
-	if (enable_container) {
-		bpf_for(i, 0, nr_containers) {
+	if (enable_cpuset) {
+		bpf_for(i, 0, nr_cpusets) {
 			cpumask = bpf_cpumask_create();
 
 			if (!cpumask)
@@ -3348,7 +3348,7 @@ s32 BPF_STRUCT_OPS_SLEEPABLE(layered_init)
 
 			bpf_for(j, 0, MAX_CPUS/64) {
 				// verifier
-				if (i < 0 || i >= MAX_CONTAINERS || j < 0 || j >= (MAX_CPUS / 64)) {
+				if (i < 0 || i >= MAX_CPUSETS || j < 0 || j >= (MAX_CPUS / 64)) {
     					bpf_cpumask_release(cpumask);
     					return -1;
 				}
@@ -3370,7 +3370,7 @@ s32 BPF_STRUCT_OPS_SLEEPABLE(layered_init)
 					if (tmp_cpuset_cpumask)
 						bpf_cpumask_release(tmp_cpuset_cpumask);
 					scx_bpf_error("cpumask is null");
-					return -1;
+					return -1; 
 				}
 				bpf_cpumask_copy(tmp_cpuset_cpumask, cast_mask(cpumask));
 

scheds/rust/scx_layered/src/main.rs

@@ -584,9 +584,9 @@ struct Opts {
     #[clap(long, default_value = "false")]
     disable_antistall: bool,
 
-    /// Enable container support
+    /// Enable cpuset support
     #[clap(long, default_value = "false")]
-    enable_container: bool,
+    enable_cpuset: bool,
 
     /// Maximum task runnable_at delay (in seconds) before antistall turns on
     #[clap(long, default_value = "3")]
@@ -1407,7 +1407,7 @@ impl<'a> Scheduler<'a> {
             let cpuset_cpumask_slice = &mut skel.maps.rodata_data.cpuset_fakemasks[i];
             cpuset_cpumask_slice.copy_from_slice(&cpumask_bitvec);
         }
-        skel.maps.rodata_data.nr_containers = cpusets.len() as u32;
+        skel.maps.rodata_data.nr_cpusets = cpusets.len() as u32;
         Ok(())
     }
 
@@ -1851,7 +1851,7 @@ impl<'a> Scheduler<'a> {
         skel.maps.rodata_data.lo_fb_wait_ns = opts.lo_fb_wait_us * 1000;
         skel.maps.rodata_data.lo_fb_share_ppk = ((opts.lo_fb_share * 1024.0) as u32).clamp(1, 1024);
         skel.maps.rodata_data.enable_antistall = !opts.disable_antistall;
-        skel.maps.rodata_data.enable_container = opts.enable_container;
+        skel.maps.rodata_data.enable_cpuset = opts.enable_cpuset;
         skel.maps.rodata_data.enable_gpu_support = opts.enable_gpu_support;
 
         for (cpu, sib) in topo.sibling_cpus().iter().enumerate() {
@@ -1920,7 +1920,7 @@ impl<'a> Scheduler<'a> {
         Self::init_layers(&mut skel, &layer_specs, &topo)?;
         Self::init_nodes(&mut skel, opts, &topo);
 
-        if opts.enable_container {
+        if opts.enable_cpuset {
             Self::init_cpusets(&mut skel, &topo)?;
         }