Add USDTs for task switch and allocation

contrib/bpftrace/rt_all.bt

@@ -0,0 +1,15 @@
+#!/usr/bin/env bpftrace
+
+BEGIN
+{
+    printf("Tracing Julia Task events... Hit Ctrl-C to end.\n");
+}
+usdt:usr/lib/libjulia-internal.so:julia:rt__ctx__switch
+{
+    printf("Task switch from %x to %x (with PTLS %x)\n", arg0, arg1, arg2);
+}
+
+usdt:usr/lib/libjulia-internal.so:julia:rt__new__task
+{
+    printf("Task %x created by parent %x (with PTLS %x)\n", arg1, arg0, arg2);
+}

doc/src/devdocs/probes.md

@@ -90,10 +90,15 @@ the probe handler.
 3. `julia:gc__mark__begin`: Beginning the mark phase
 4. `julia:gc__mark_end(scanned_bytes, perm_scanned)`: Mark phase ended
 5. `julia:gc__sweep_begin(full)`: Starting sweep
-6. `julia:gc__sweep_end()`: Sweep phase finished
+6. `julia:gc__sweep_end`: Sweep phase finished
 7. `julia:gc__end`: GC is finished, other threads continue work
 8. `julia:gc__finalizer`: Initial GC thread has finished running finalizers
 
+### Task runtime probes
+
+1. `julia:rt__ctx__switch(last, next, ptls)`: Switching from task `last` to task `next` on thread associated with PTLS `ptls`.
+2. `julia:rt__new__task(parent, child, ptls)`: Task `parent` created task `child` on thread associated with PTLS `ptls`.
+
 #### GC stop-the-world latency
 
 An example `bpftrace` script is given in `contrib/gc_stop_the_world_latency.bt`

src/julia_internal.h

@@ -1532,6 +1532,8 @@ uint16_t __gnu_f2h_ieee(float param) JL_NOTSAFEPOINT;
 #define JL_PROBE_GC_SWEEP_END() do ; while (0)
 #define JL_PROBE_GC_END() do ; while (0)
 #define JL_PROBE_GC_FINALIZER() do ; while (0)
+#define JL_PROBE_RT_CTX_SWITCH(last, next, ptls) do ; while (0)
+#define JL_PROBE_RT_NEW_TASK(parent, child, ptls) do ; while (0)
 
 #define JL_PROBE_GC_BEGIN_ENABLED() (0)
 #define JL_PROBE_GC_STOP_THE_WORLD_ENABLED() (0)
@@ -1541,6 +1543,8 @@ uint16_t __gnu_f2h_ieee(float param) JL_NOTSAFEPOINT;
 #define JL_PROBE_GC_SWEEP_END_ENABLED()  (0)
 #define JL_PROBE_GC_END_ENABLED() (0)
 #define JL_PROBE_GC_FINALIZER_ENABLED() (0)
+#define JL_PROBE_RT_CTX_SWITCH_ENABLED() (0)
+#define JL_PROBE_RT_NEW_TASK_ENABLED() (0)
 #endif
 
 #endif

src/task.c

@@ -370,6 +370,8 @@ static void ctx_switch(jl_task_t *lastt)
     // none of these locks should be held across a task switch
     assert(ptls->locks.len == 0);
 
+    JL_PROBE_RT_CTX_SWITCH(lastt, t, ptls);
+
 #ifdef _COMPILER_TSAN_ENABLED_
     if (lastt->ctx.tsan_state != __tsan_get_current_fiber()) {
         // Something went really wrong - don't even assume that we can
@@ -753,6 +755,7 @@ JL_DLLEXPORT jl_task_t *jl_new_task(jl_function_t *start, jl_value_t *completion
 {
     jl_task_t *ct = jl_current_task;
     jl_task_t *t = (jl_task_t*)jl_gc_alloc(ct->ptls, sizeof(jl_task_t), jl_task_type);
+    JL_PROBE_RT_NEW_TASK(ct, t, ct->ptls);
     t->copy_stack = 0;
     if (ssize == 0) {
         // stack size unspecified; use default

src/uprobes.d

@@ -9,6 +9,9 @@ provider julia {
     probe gc__sweep__end();
     probe gc__end();
     probe gc__finalizer();
+
+    probe rt__ctx__switch(jl_task_t *last, jl_task_t *next, jl_ptls_t ptls);
+    probe rt__new__task(jl_task_t *parent, jl_task_t *child, jl_ptls_t ptls);
 };
 
 #pragma D attributes Evolving/Evolving/Common provider julia provider