Fix issue #505

example/gpu/perthreadRuntimeDist/.gitignore

@@ -0,0 +1,2 @@
+/perthreadRuntimeDist
+/.output

example/gpu/perthreadRuntimeDist/Makefile

@@ -0,0 +1,145 @@
+# SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
+OUTPUT := .output
+CLANG ?= clang
+LIBBPF_SRC := $(abspath ../../../third_party/libbpf/src)
+BPFTOOL_SRC := $(abspath ../../../third_party/bpftool/src)
+LIBBPF_OBJ := $(abspath $(OUTPUT)/libbpf.a)
+BPFTOOL_OUTPUT ?= $(abspath $(OUTPUT)/bpftool)
+BPFTOOL ?= $(BPFTOOL_OUTPUT)/bootstrap/bpftool
+ARCH ?= $(shell uname -m | sed 's/x86_64/x86/' \
+			 | sed 's/arm.*/arm/' \
+			 | sed 's/aarch64/arm64/' \
+			 | sed 's/ppc64le/powerpc/' \
+			 | sed 's/mips.*/mips/' \
+			 | sed 's/riscv64/riscv/' \
+			 | sed 's/loongarch64/loongarch/')
+VMLINUX := ../../../third_party/vmlinux/$(ARCH)/vmlinux.h
+# Use our own libbpf API headers and Linux UAPI headers distributed with
+# libbpf to avoid dependency on system-wide headers, which could be missing or
+# outdated
+INCLUDES := -I$(OUTPUT) -I../../../third_party/libbpf/include/uapi -I$(dir $(VMLINUX))
+CFLAGS := -g -Wall
+ALL_LDFLAGS := $(LDFLAGS) $(EXTRA_LDFLAGS)
+
+APPS = perthreadRuntimeDist # minimal minimal_legacy uprobe kprobe fentry usdt sockfilter tc ksyscall
+
+CARGO ?= $(shell which cargo)
+ifeq ($(strip $(CARGO)),)
+BZS_APPS :=
+else
+BZS_APPS := # profile
+APPS += $(BZS_APPS)
+# Required by libblazesym
+ALL_LDFLAGS += -lrt -ldl -lpthread -lm
+endif
+
+# Get Clang's default includes on this system. We'll explicitly add these dirs
+# to the includes list when compiling with `-target bpf` because otherwise some
+# architecture-specific dirs will be "missing" on some architectures/distros -
+# headers such as asm/types.h, asm/byteorder.h, asm/socket.h, asm/sockios.h,
+# sys/cdefs.h etc. might be missing.
+#
+# Use '-idirafter': Don't interfere with include mechanics except where the
+# build would have failed anyways.
+CLANG_BPF_SYS_INCLUDES ?= $(shell $(CLANG) -v -E - </dev/null 2>&1 \
+	| sed -n '/<...> search starts here:/,/End of search list./{ s| \(/.*\)|-idirafter \1|p }')
+
+ifeq ($(V),1)
+	Q =
+	msg =
+else
+	Q = @
+	msg = @printf '  %-8s %s%s\n'					\
+		      "$(1)"						\
+		      "$(patsubst $(abspath $(OUTPUT))/%,%,$(2))"	\
+		      "$(if $(3), $(3))";
+	MAKEFLAGS += --no-print-directory
+endif
+
+define allow-override
+  $(if $(or $(findstring environment,$(origin $(1))),\
+            $(findstring command line,$(origin $(1)))),,\
+    $(eval $(1) = $(2)))
+endef
+
+$(call allow-override,CC,$(CROSS_COMPILE)cc)
+$(call allow-override,LD,$(CROSS_COMPILE)ld)
+
+.PHONY: all
+all: $(APPS) vec_add
+
+vec_add: vec_add.cu
+	@if command -v nvcc >/dev/null 2>&1; then \
+		nvcc -arch=sm_61 -cudart shared vec_add.cu -o vec_add -g; \
+	else \
+		echo "Warning: CUDA not found, skipping vec_add build"; \
+	fi
+
+.PHONY: clean
+clean:
+	$(call msg,CLEAN)
+	$(Q)rm -rf $(OUTPUT) $(APPS) vec_add
+
+$(OUTPUT) $(OUTPUT)/libbpf $(BPFTOOL_OUTPUT):
+	$(call msg,MKDIR,$@)
+	$(Q)mkdir -p $@
+
+# Build libbpf
+$(LIBBPF_OBJ): $(wildcard $(LIBBPF_SRC)/*.[ch] $(LIBBPF_SRC)/Makefile) | $(OUTPUT)/libbpf
+	$(call msg,LIB,$@)
+	$(Q)$(MAKE) -C $(LIBBPF_SRC) BUILD_STATIC_ONLY=1		      \
+		    OBJDIR=$(dir $@)/libbpf DESTDIR=$(dir $@)		      \
+		    INCLUDEDIR= LIBDIR= UAPIDIR=			      \
+		    install
+
+# Build bpftool
+$(BPFTOOL): | $(BPFTOOL_OUTPUT)
+	$(call msg,BPFTOOL,$@)
+	$(Q)$(MAKE) ARCH= CROSS_COMPILE= OUTPUT=$(BPFTOOL_OUTPUT)/ -C $(BPFTOOL_SRC) bootstrap
+
+
+$(LIBBLAZESYM_SRC)/target/release/libblazesym.a::
+	$(Q)cd $(LIBBLAZESYM_SRC) && $(CARGO) build --features=cheader,dont-generate-test-files --release
+
+$(LIBBLAZESYM_OBJ): $(LIBBLAZESYM_SRC)/target/release/libblazesym.a | $(OUTPUT)
+	$(call msg,LIB, $@)
+	$(Q)cp $(LIBBLAZESYM_SRC)/target/release/libblazesym.a $@
+
+$(LIBBLAZESYM_HEADER): $(LIBBLAZESYM_SRC)/target/release/libblazesym.a | $(OUTPUT)
+	$(call msg,LIB,$@)
+	$(Q)cp $(LIBBLAZESYM_SRC)/target/release/blazesym.h $@
+
+# Build BPF code
+$(OUTPUT)/%.bpf.o: %.bpf.c $(LIBBPF_OBJ) $(wildcard %.h) $(VMLINUX) | $(OUTPUT) $(BPFTOOL)
+	$(call msg,BPF,$@)
+	$(Q)$(CLANG) -Xlinker --export-dynamic -g -O2 -target bpf -D__TARGET_ARCH_$(ARCH)		      \
+		     $(INCLUDES) $(CLANG_BPF_SYS_INCLUDES)		      \
+		     -c $(filter %.c,$^) -o $(patsubst %.bpf.o,%.tmp.bpf.o,$@)
+	$(Q)$(BPFTOOL) gen object $@ $(patsubst %.bpf.o,%.tmp.bpf.o,$@)
+
+# Generate BPF skeletons
+$(OUTPUT)/%.skel.h: $(OUTPUT)/%.bpf.o | $(OUTPUT) $(BPFTOOL)
+	$(call msg,GEN-SKEL,$@)
+	$(Q)$(BPFTOOL) gen skeleton $< > $@
+
+# Build user-space code
+$(patsubst %,$(OUTPUT)/%.o,$(APPS)): %.o: %.skel.h
+
+$(OUTPUT)/%.o: %.c $(wildcard %.h) | $(OUTPUT)
+	$(call msg,CC,$@)
+	$(Q)$(CC) $(CFLAGS) $(INCLUDES) -c $(filter %.c,$^) -o $@
+
+$(patsubst %,$(OUTPUT)/%.o,$(BZS_APPS)): $(LIBBLAZESYM_HEADER)
+
+$(BZS_APPS): $(LIBBLAZESYM_OBJ)
+
+# Build application binary
+$(APPS): %: $(OUTPUT)/%.o $(LIBBPF_OBJ) | $(OUTPUT)
+	$(call msg,BINARY,$@)
+	$(Q)$(CC) $(CFLAGS) $^ $(ALL_LDFLAGS) -lelf -lz -o $@
+
+# delete failed targets
+.DELETE_ON_ERROR:
+
+# keep intermediate (.skel.h, .bpf.o, etc) targets
+.SECONDARY:

example/gpu/perthreadRuntimeDist/perthreadRuntimeDist.bpf.c

@@ -0,0 +1,62 @@
+#include "vmlinux.h"
+#include <bpf/bpf_helpers.h>
+#include <bpf/bpf_core_read.h>
+#include "perthreadRuntimeDist.h"
+
+
+struct {
+    __uint(type, BPF_MAP_TYPE_PERF_EVENT_ARRAY);
+    __uint(key_size, sizeof(__u32));
+    __uint(value_size, sizeof(__u32));
+} events SEC(".maps");
+
+// 用于记录 start time
+struct {
+    __uint(type, BPF_MAP_TYPE_HASH);
+    __uint(max_entries, 8192);
+    __type(key, u32);
+    __type(value, u64);
+} start SEC(".maps");
+
+char LICENSE[] SEC("license") = "Dual BSD/GPL";
+
+// EXT helper index — 必须存在且可见
+static const u64 (*bpf_get_globaltimer)(void) = (void *)502;
+
+// GPU kernel entry
+SEC("kprobe/cudaLaunchKernel")
+int cuda__kernel_entry(struct pt_regs *ctx)
+{
+    u32 tid = bpf_get_current_pid_tgid();
+
+    u64 start_cycles = bpf_get_globaltimer(); // bpf_ktime_get_ns()?
+
+    bpf_map_update_elem(&start, &tid, &start_cycles, BPF_ANY);
+
+    return 0;
+}
+
+SEC("kretprobe/cudaLaunchKernel")
+int cuda__kernel_exit(struct pt_regs *ctx)
+{
+    u32 tid = bpf_get_current_pid_tgid();
+    u64 *start_cycles = bpf_map_lookup_elem(&start, &tid);
+
+    if (!start_cycles)
+        return 0;
+
+    u64 end_cycles = bpf_get_globaltimer();
+
+    struct event_t evt = {
+        .tid = tid,
+        .cycles = end_cycles - *start_cycles,
+    };
+
+    bpf_perf_event_output(ctx, &events,
+                          BPF_F_CURRENT_CPU,
+                          &evt,
+                          sizeof(evt));
+
+    bpf_map_delete_elem(&start, &tid);
+    return 0;
+}

example/gpu/perthreadRuntimeDist/perthreadRuntimeDist.c

@@ -0,0 +1,61 @@
+#include <stdio.h>
+#include <unistd.h>
+#include <bpf/libbpf.h>
+#include "perthreadRuntimeDist.skel.h"
+#include "perthreadRuntimeDist.h"
+
+static void handle_event(void *ctx, int cpu, void *data, __u32 size)
+{
+    struct event_t *e = data;
+    printf("[CPU %d] tid=%u cycles=%llu ns\n",
+           cpu, e->tid, e->cycles);
+}
+
+static void handle_lost(void *ctx, int cpu, __u64 lost)
+{
+    printf("LOST %llu events on CPU %d\n", lost, cpu);
+}
+
+int main()
+{
+    struct perthreadRuntimeDist_bpf *skel;
+    struct perf_buffer *pb;
+    int events_fd;
+
+    skel = perthreadRuntimeDist_bpf__open();
+    if (!skel) {
+        fprintf(stderr, "Failed to open BPF skeleton\n");
+        return 1;
+    }
+
+    if (perthreadRuntimeDist_bpf__load(skel)) {
+        fprintf(stderr, "Failed to load BPF skeleton\n");
+        return 1;
+    }
+
+    if (perthreadRuntimeDist_bpf__attach(skel)) {
+        fprintf(stderr, "Failed to attach BPF skeleton\n");
+        return 1;
+    }
+
+    printf("BPF attached successfully\n");
+
+    events_fd = bpf_map__fd(skel->maps.events);
+    pb = perf_buffer__new(events_fd, 16 /*buffer pages*/,
+                         handle_event, handle_lost, NULL, NULL);
+
+    if (!pb) {
+        fprintf(stderr, "Failed to open perf buffer\n");
+        return 1;
+    }
+
+    printf("Collecting data...\n");
+
+    while (1) {
+        int err = perf_buffer__poll(pb, 100 /*ms*/);
+        if (err < 0)
+            printf("perf_buffer__poll() error %d\n", err);
+    }
+
+    return 0;
+}

example/gpu/perthreadRuntimeDist/perthreadRuntimeDist.h

@@ -0,0 +1,9 @@
+#ifndef __PERTHREAD_RUNTIME_DIST_H__
+#define __PERTHREAD_RUNTIME_DIST_H__
+
+struct event_t {
+    __u32 tid;
+    __u64 cycles;
+};
+
+#endif