[PATCH v2 perf,bpf 00/11] perf annotation of BPF programs

[PATCH v2 perf,bpf 00/11] perf annotation of BPF programs

[Song Liu]

Changes v1 to v2:
1. Fix compilation error with different feature-disassembler-four-args;
2. Fix a segfault in perf-record;
3. Split patches 5/9 and 6/9 so that perf_env changes and perf.data changes
   are in separate patches.

This series enables annotation of BPF programs in perf.

perf tool gathers information via sys_bpf and (optionally) stores them in
perf.data as headers.

Patch 1/11 fixes a minor issue in kernel;
Patch 2/11 to 4/11 introduce new helper functions and use them in perf and
     bpftool;
Patch 5/11 to 8/11 saves information of bpf program in perf_env;
Patch 9/11 adds --bpf-event options to perf-top;
Patch 10/11 enables annotation of bpf programs based on information
     gathered in 5/11 to 8/11;
Patch 11/11 handles information of short living BPF program that are loaded
     during perf-record or perf-top.

Commands tested during developments are perf-top, perf-record, perf-report,
and perf-annotate.

===================== Note on patch dependency  ========================
This set has dependency in both bpf-next tree and tip/perf/core. Current
version is developed on bpf-next tree with the following commits
cherry-picked from tip/perf/core:

(from 1/10 to 10/10)
commit 76193a94522f ("perf, bpf: Introduce PERF_RECORD_KSYMBOL")
commit d764ac646491 ("tools headers uapi: Sync tools/include/uapi/linux/perf_event.h")
commit 6ee52e2a3fe4 ("perf, bpf: Introduce PERF_RECORD_BPF_EVENT")
commit df063c83aa2c ("tools headers uapi: Sync tools/include/uapi/linux/perf_event.h")
commit 9aa0bfa370b2 ("perf tools: Handle PERF_RECORD_KSYMBOL")
commit 45178a928a4b ("perf tools: Handle PERF_RECORD_BPF_EVENT")
commit 7b612e291a5a ("perf tools: Synthesize PERF_RECORD_* for loaded BPF programs")
commit a40b95bcd30c ("perf top: Synthesize BPF events for pre-existing loaded BPF programs")
commit 6934058d9fb6 ("bpf: Add module name [bpf] to ksymbols for bpf programs")
commit 811184fb6977 ("perf bpf: Fix synthesized PERF_RECORD_KSYMBOL/BPF_EVENT")
========================================================================

This set is also available at:

https://github.com/liu-song-6/linux/tree/bpf-annotation

Thanks!!

Song Liu (11):
  perf, bpf: consider events with attr.bpf_event as side-band events
  bpf: libbpf: introduce bpf_program__get_prog_info_linear()
  bpf: bpftool: use bpf_program__get_prog_info_linear() in
    prog.c:do_dump()
  perf, bpf: synthesize bpf events with
    bpf_program__get_prog_info_linear()
  perf, bpf: save bpf_prog_info in a rbtree in perf_env
  perf, bpf: save bpf_prog_info information as headers to perf.data
  perf, bpf: save btf in a rbtree in perf_env
  perf, bpf: save btf information as headers to perf.data
  perf-top: add option --bpf-event
  perf, bpf: enable annotation of bpf program
  perf, bpf: save information about short living bpf programs

 kernel/events/core.c        |   3 +-
 tools/bpf/bpftool/prog.c    | 266 ++++++---------------------
 tools/lib/bpf/libbpf.c      | 251 +++++++++++++++++++++++++
 tools/lib/bpf/libbpf.h      |  63 +++++++
 tools/lib/bpf/libbpf.map    |   3 +
 tools/perf/Makefile.config  |   6 +-
 tools/perf/builtin-record.c |  15 +-
 tools/perf/builtin-top.c    |  15 +-
 tools/perf/util/annotate.c  | 149 ++++++++++++++-
 tools/perf/util/bpf-event.c | 354 +++++++++++++++++++++++++++---------
 tools/perf/util/bpf-event.h |  48 ++++-
 tools/perf/util/dso.c       |   1 +
 tools/perf/util/dso.h       |  33 ++--
 tools/perf/util/env.c       | 148 +++++++++++++++
 tools/perf/util/env.h       |  17 ++
 tools/perf/util/evlist.c    |  20 ++
 tools/perf/util/evlist.h    |   2 +
 tools/perf/util/header.c    | 231 ++++++++++++++++++++++-
 tools/perf/util/header.h    |   2 +
 tools/perf/util/symbol.c    |   1 +
 20 files changed, 1316 insertions(+), 312 deletions(-)

--
2.17.1

[PATCH v2 perf,bpf 01/11] perf, bpf: consider events with attr.bpf_event as side-band events

[Song Liu]

Events with bpf_event should be considered as side-band event, as they
carry information about BPF programs.

Fixes: 6ee52e2a3fe4 ("perf, bpf: Introduce PERF_RECORD_BPF_EVENT")
Signed-off-by: Song Liu <songliubraving@fb.com>
---
 kernel/events/core.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/kernel/events/core.c b/kernel/events/core.c
index 0a8dab322111..9403bdda5f8c 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -4238,7 +4238,8 @@ static bool is_sb_event(struct perf_event *event)
 	if (attr->mmap || attr->mmap_data || attr->mmap2 ||
 	    attr->comm || attr->comm_exec ||
 	    attr->task || attr->ksymbol ||
-	    attr->context_switch)
+	    attr->context_switch ||
+	    attr->bpf_event)
 		return true;
 	return false;
 }
-- 
2.17.1

[PATCH v2 perf,bpf 02/11] bpf: libbpf: introduce bpf_program__get_prog_info_linear()

[Song Liu]

Currently, bpf_prog_info includes 9 arrays. The user has the option to
fetch any combination of these arrays. However, this requires a lot of
handling of these arrays. This work becomes more tricky when we need to
store bpf_prog_info to a file, because these arrays are allocated
independently.

This patch introduces struct bpf_prog_info_linear, which stores arrays
of bpf_prog_info in continues memory. Helper functions are introduced
to unify the work to get different information of bpf_prog_info.
Specifically, bpf_program__get_prog_info_linear() allows the user to
select which arrays to fetch, and handles details for the user.

Plesae see the comments before enum bpf_prog_info_array for more details
and examples.

Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Song Liu <songliubraving@fb.com>
---
 tools/lib/bpf/libbpf.c   | 251 +++++++++++++++++++++++++++++++++++++++
 tools/lib/bpf/libbpf.h   |  63 ++++++++++
 tools/lib/bpf/libbpf.map |   3 +
 3 files changed, 317 insertions(+)

diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c
index e3c39edfb9d3..1f808c555747 100644
--- a/tools/lib/bpf/libbpf.c
+++ b/tools/lib/bpf/libbpf.c
@@ -113,6 +113,11 @@ void libbpf_print(enum libbpf_print_level level, const char *format, ...)
 # define LIBBPF_ELF_C_READ_MMAP ELF_C_READ
 #endif
 
+static inline __u64 ptr_to_u64(const void *ptr)
+{
+	return (__u64) (unsigned long) ptr;
+}
+
 struct bpf_capabilities {
 	/* v4.14: kernel support for program & map names. */
 	__u32 name:1;
@@ -2979,3 +2984,249 @@ bpf_perf_event_read_simple(void *mmap_mem, size_t mmap_size, size_t page_size,
 	ring_buffer_write_tail(header, data_tail);
 	return ret;
 }
+
+struct bpf_prog_info_array_desc {
+	int	array_offset;	/* e.g. offset of jited_prog_insns */
+	int	count_offset;	/* e.g. offset of jited_prog_len */
+	int	size_offset;	/* > 0: offset of rec size,
+				 * < 0: fix size of -size_offset
+				 */
+};
+
+static struct bpf_prog_info_array_desc bpf_prog_info_array_desc[] = {
+	[BPF_PROG_INFO_JITED_INSNS] = {
+		offsetof(struct bpf_prog_info, jited_prog_insns),
+		offsetof(struct bpf_prog_info, jited_prog_len),
+		-1,
+	},
+	[BPF_PROG_INFO_XLATED_INSNS] = {
+		offsetof(struct bpf_prog_info, xlated_prog_insns),
+		offsetof(struct bpf_prog_info, xlated_prog_len),
+		-1,
+	},
+	[BPF_PROG_INFO_MAP_IDS] = {
+		offsetof(struct bpf_prog_info, map_ids),
+		offsetof(struct bpf_prog_info, nr_map_ids),
+		-(int)sizeof(__u32),
+	},
+	[BPF_PROG_INFO_JITED_KSYMS] = {
+		offsetof(struct bpf_prog_info, jited_ksyms),
+		offsetof(struct bpf_prog_info, nr_jited_ksyms),
+		-(int)sizeof(__u64),
+	},
+	[BPF_PROG_INFO_JITED_FUNC_LENS] = {
+		offsetof(struct bpf_prog_info, jited_func_lens),
+		offsetof(struct bpf_prog_info, nr_jited_func_lens),
+		-(int)sizeof(__u32),
+	},
+	[BPF_PROG_INFO_FUNC_INFO] = {
+		offsetof(struct bpf_prog_info, func_info),
+		offsetof(struct bpf_prog_info, nr_func_info),
+		offsetof(struct bpf_prog_info, func_info_rec_size),
+	},
+	[BPF_PROG_INFO_LINE_INFO] = {
+		offsetof(struct bpf_prog_info, line_info),
+		offsetof(struct bpf_prog_info, nr_line_info),
+		offsetof(struct bpf_prog_info, line_info_rec_size),
+	},
+	[BPF_PROG_INFO_JITED_LINE_INFO] = {
+		offsetof(struct bpf_prog_info, jited_line_info),
+		offsetof(struct bpf_prog_info, nr_jited_line_info),
+		offsetof(struct bpf_prog_info, jited_line_info_rec_size),
+	},
+	[BPF_PROG_INFO_PROG_TAGS] = {
+		offsetof(struct bpf_prog_info, prog_tags),
+		offsetof(struct bpf_prog_info, nr_prog_tags),
+		-(int)sizeof(__u8) * BPF_TAG_SIZE,
+	},
+
+};
+
+static __u32 bpf_prog_info_read_offset_u32(struct bpf_prog_info *info, int offset)
+{
+	__u32 *array = (__u32 *)info;
+
+	if (offset >= 0)
+		return array[offset / sizeof(__u32)];
+	return -(int)offset;
+}
+
+static __u64 bpf_prog_info_read_offset_u64(struct bpf_prog_info *info, int offset)
+{
+	__u64 *array = (__u64 *)info;
+
+	if (offset >= 0)
+		return array[offset / sizeof(__u64)];
+	return -(int)offset;
+}
+
+static void bpf_prog_info_set_offset_u32(struct bpf_prog_info *info, int offset,
+					 __u32 val)
+{
+	__u32 *array = (__u32 *)info;
+
+	if (offset >= 0)
+		array[offset / sizeof(__u32)] = val;
+}
+
+static void bpf_prog_info_set_offset_u64(struct bpf_prog_info *info, int offset,
+					 __u64 val)
+{
+	__u64 *array = (__u64 *)info;
+
+	if (offset >= 0)
+		array[offset / sizeof(__u64)] = val;
+}
+
+struct bpf_prog_info_linear *
+bpf_program__get_prog_info_linear(int fd, __u64 arrays)
+{
+	struct bpf_prog_info_linear *info_linear;
+	struct bpf_prog_info info = {};
+	__u32 info_len = sizeof(info);
+	__u32 data_len = 0;
+	int i, err;
+	void *ptr;
+
+	if (arrays >> BPF_PROG_INFO_LAST_ARRAY)
+		return ERR_PTR(-EINVAL);
+
+	/* step 1: get array dimensions */
+	err = bpf_obj_get_info_by_fd(fd, &info, &info_len);
+	if (err) {
+		pr_debug("can't get prog info: %s", strerror(errno));
+		return ERR_PTR(-EFAULT);
+	}
+
+	/* step 2: calculate total size of all arrays */
+	for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
+		bool include_array = (arrays & (1UL << i)) > 0;
+		struct bpf_prog_info_array_desc *desc;
+		__u32 count, size;
+
+		desc = bpf_prog_info_array_desc + i;
+
+		/* kernel is too old to support this field */
+		if (info_len < desc->array_offset + sizeof(__u32) ||
+		    info_len < desc->count_offset + sizeof(__u32) ||
+		    (desc->size_offset > 0 && info_len < desc->size_offset))
+			include_array = false;
+
+		if (!include_array) {
+			arrays &= ~(1UL << i);	/* clear the bit */
+			continue;
+		}
+
+		count = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
+		size = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
+
+		data_len += count * size;
+	}
+
+	/* step 3: allocate continuous memory */
+	data_len = roundup(data_len, sizeof(__u64));
+	info_linear = malloc(sizeof(struct bpf_prog_info_linear) + data_len);
+	if (!info_linear)
+		return ERR_PTR(-ENOMEM);
+
+	/* step 4: fill data to info_linear->info */
+	info_linear->arrays = arrays;
+	memset(&info_linear->info, 0, sizeof(info));
+	ptr = info_linear->data;
+
+	for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
+		struct bpf_prog_info_array_desc *desc;
+		__u32 count, size;
+
+		if ((arrays & (1UL << i)) == 0)
+			continue;
+
+		desc = bpf_prog_info_array_desc + i;
+		count = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
+		size = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
+		bpf_prog_info_set_offset_u32(&info_linear->info,
+					     desc->count_offset, count);
+		bpf_prog_info_set_offset_u32(&info_linear->info,
+					     desc->size_offset, size);
+		bpf_prog_info_set_offset_u64(&info_linear->info,
+					     desc->array_offset,
+					     ptr_to_u64(ptr));
+		ptr += count * size;
+	}
+
+	/* step 5: call syscall again to get required arrays */
+	err = bpf_obj_get_info_by_fd(fd, &info_linear->info, &info_len);
+	if (err) {
+		pr_debug("can't get prog info: %s", strerror(errno));
+		free(info_linear);
+		return ERR_PTR(-EFAULT);
+	}
+
+	/* step 6: verify the data */
+	for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
+		struct bpf_prog_info_array_desc *desc;
+		__u32 v1, v2;
+
+		if ((arrays & (1UL << i)) == 0)
+			continue;
+
+		desc = bpf_prog_info_array_desc + i;
+		v1 = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
+		v2 = bpf_prog_info_read_offset_u32(&info_linear->info,
+						   desc->count_offset);
+		if (v1 != v2)
+			pr_warning("%s: mismatch in element count\n", __func__);
+
+		v1 = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
+		v2 = bpf_prog_info_read_offset_u32(&info_linear->info,
+						   desc->size_offset);
+		if (v1 != v2)
+			pr_warning("%s: mismatch in rec size\n", __func__);
+	}
+
+	/* step 7: update info_len and data_len */
+	info_linear->info_len = sizeof(struct bpf_prog_info);
+	info_linear->data_len = data_len;
+
+	return info_linear;
+}
+
+void bpf_program__bpil_addr_to_offs(struct bpf_prog_info_linear *info_linear)
+{
+	int i;
+
+	for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
+		struct bpf_prog_info_array_desc *desc;
+		__u64 addr, offs;
+
+		if ((info_linear->arrays & (1UL << i)) == 0)
+			continue;
+
+		desc = bpf_prog_info_array_desc + i;
+		addr = bpf_prog_info_read_offset_u64(&info_linear->info,
+						     desc->array_offset);
+		offs = addr - ptr_to_u64(info_linear->data);
+		bpf_prog_info_set_offset_u64(&info_linear->info,
+					     desc->array_offset, offs);
+	}
+}
+
+void bpf_program__bpil_offs_to_addr(struct bpf_prog_info_linear *info_linear)
+{
+	int i;
+
+	for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
+		struct bpf_prog_info_array_desc *desc;
+		__u64 addr, offs;
+
+		if ((info_linear->arrays & (1UL << i)) == 0)
+			continue;
+
+		desc = bpf_prog_info_array_desc + i;
+		offs = bpf_prog_info_read_offset_u64(&info_linear->info,
+						     desc->array_offset);
+		addr = offs + ptr_to_u64(info_linear->data);
+		bpf_prog_info_set_offset_u64(&info_linear->info,
+					     desc->array_offset, addr);
+	}
+}
diff --git a/tools/lib/bpf/libbpf.h b/tools/lib/bpf/libbpf.h
index 69a7c25eaccc..4d357c26fbd2 100644
--- a/tools/lib/bpf/libbpf.h
+++ b/tools/lib/bpf/libbpf.h
@@ -372,6 +372,69 @@ LIBBPF_API bool bpf_probe_map_type(enum bpf_map_type map_type, __u32 ifindex);
 LIBBPF_API bool bpf_probe_helper(enum bpf_func_id id,
 				 enum bpf_prog_type prog_type, __u32 ifindex);
 
+/*
+ * Get bpf_prog_info in continuous memory
+ *
+ * struct bpf_prog_info has multiple arrays. The user has option to choose
+ * arrays to fetch from kernel. The following APIs provide uniform way to
+ * fetch these data. All arrays in bpf_prog_info are stored in singile
+ * continuous memory region. This makes it easy to store the info in a
+ * file.
+ *
+ * Before writing bpf_prog_info_linear to files, it is necessary to
+ * translate pointers bpf_prog_info to offsets. Helper functions
+ * bpf_program__bpil_addr_to_offs() and bpf_program__bpil_offs_to_addr()
+ * are introduced to switch between pointers and offsets.
+ *
+ * Examples:
+ *   # To fetch map_ids and prog_tags:
+ *   __u64 arrays = (1UL << BPF_PROG_INFO_MAP_IDS) |
+ *           (1UL << BPF_PROG_INFO_PROG_TAGS);
+ *   struct bpf_prog_info_linear *info_linear =
+ *           bpf_program__get_prog_info_linear(fd, arrays);
+ *
+ *   # To save data in file
+ *   bpf_program__bpil_addr_to_offs(info_linear);
+ *   write(f, info_linear, sizeof(*info_linear) + info_linear->data_len);
+ *
+ *   # To read data from file
+ *   read(f, info_linear, <proper_size>);
+ *   bpf_program__bpil_offs_to_addr(info_linear);
+ */
+enum bpf_prog_info_array {
+	BPF_PROG_INFO_FIRST_ARRAY = 0,
+	BPF_PROG_INFO_JITED_INSNS = 0,
+	BPF_PROG_INFO_XLATED_INSNS,
+	BPF_PROG_INFO_MAP_IDS,
+	BPF_PROG_INFO_JITED_KSYMS,
+	BPF_PROG_INFO_JITED_FUNC_LENS,
+	BPF_PROG_INFO_FUNC_INFO,
+	BPF_PROG_INFO_LINE_INFO,
+	BPF_PROG_INFO_JITED_LINE_INFO,
+	BPF_PROG_INFO_PROG_TAGS,
+	BPF_PROG_INFO_LAST_ARRAY,
+};
+
+struct bpf_prog_info_linear {
+	/* size of struct bpf_prog_info, when the tool is compiled */
+	__u32			info_len;
+	/* total bytes allocated for data, round up to 8 bytes */
+	__u32			data_len;
+	/* which arrays are included in data */
+	__u64			arrays;
+	struct bpf_prog_info	info;
+	__u8			data[];
+};
+
+LIBBPF_API struct bpf_prog_info_linear *
+bpf_program__get_prog_info_linear(int fd, __u64 arrays);
+
+LIBBPF_API void
+bpf_program__bpil_addr_to_offs(struct bpf_prog_info_linear *info_linear);
+
+LIBBPF_API void
+bpf_program__bpil_offs_to_addr(struct bpf_prog_info_linear *info_linear);
+
 #ifdef __cplusplus
 } /* extern "C" */
 #endif
diff --git a/tools/lib/bpf/libbpf.map b/tools/lib/bpf/libbpf.map
index 5fc8222209f8..5f2fceec6ea7 100644
--- a/tools/lib/bpf/libbpf.map
+++ b/tools/lib/bpf/libbpf.map
@@ -145,4 +145,7 @@ LIBBPF_0.0.2 {
 		btf_ext__new;
 		btf_ext__reloc_func_info;
 		btf_ext__reloc_line_info;
+		bpf_program__get_prog_info_linear;
+		bpf_program__bpil_addr_to_offs;
+		bpf_program__bpil_offs_to_addr;
 } LIBBPF_0.0.1;
-- 
2.17.1

[PATCH v2 perf,bpf 03/11] bpf: bpftool: use bpf_program__get_prog_info_linear() in prog.c:do_dump()

[Song Liu]

This patches uses bpf_program__get_prog_info_linear() to simplify the
logic in prog.c do_dump().

Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Song Liu <songliubraving@fb.com>
---
 tools/bpf/bpftool/prog.c | 266 +++++++++------------------------------
 1 file changed, 59 insertions(+), 207 deletions(-)

diff --git a/tools/bpf/bpftool/prog.c b/tools/bpf/bpftool/prog.c
index 0640e9bc0ada..206b820df7c2 100644
--- a/tools/bpf/bpftool/prog.c
+++ b/tools/bpf/bpftool/prog.c
@@ -393,41 +393,31 @@ static int do_show(int argc, char **argv)
 
 static int do_dump(int argc, char **argv)
 {
-	unsigned int finfo_rec_size, linfo_rec_size, jited_linfo_rec_size;
-	void *func_info = NULL, *linfo = NULL, *jited_linfo = NULL;
-	unsigned int nr_finfo, nr_linfo = 0, nr_jited_linfo = 0;
+	struct bpf_prog_info_linear *info_linear;
 	struct bpf_prog_linfo *prog_linfo = NULL;
-	unsigned long *func_ksyms = NULL;
-	struct bpf_prog_info info = {};
-	unsigned int *func_lens = NULL;
+	enum {DUMP_JITED, DUMP_XLATED} mode;
 	const char *disasm_opt = NULL;
-	unsigned int nr_func_ksyms;
-	unsigned int nr_func_lens;
+	struct bpf_prog_info *info;
 	struct dump_data dd = {};
-	__u32 len = sizeof(info);
+	void *func_info = NULL;
 	struct btf *btf = NULL;
-	unsigned int buf_size;
 	char *filepath = NULL;
 	bool opcodes = false;
 	bool visual = false;
 	char func_sig[1024];
 	unsigned char *buf;
 	bool linum = false;
-	__u32 *member_len;
-	__u64 *member_ptr;
+	__u32 member_len;
+	__u64 arrays;
 	ssize_t n;
-	int err;
 	int fd;
 
 	if (is_prefix(*argv, "jited")) {
 		if (disasm_init())
 			return -1;
-
-		member_len = &info.jited_prog_len;
-		member_ptr = &info.jited_prog_insns;
+		mode = DUMP_JITED;
 	} else if (is_prefix(*argv, "xlated")) {
-		member_len = &info.xlated_prog_len;
-		member_ptr = &info.xlated_prog_insns;
+		mode = DUMP_XLATED;
 	} else {
 		p_err("expected 'xlated' or 'jited', got: %s", *argv);
 		return -1;
@@ -466,175 +456,50 @@ static int do_dump(int argc, char **argv)
 		return -1;
 	}
 
-	err = bpf_obj_get_info_by_fd(fd, &info, &len);
-	if (err) {
-		p_err("can't get prog info: %s", strerror(errno));
-		return -1;
-	}
-
-	if (!*member_len) {
-		p_info("no instructions returned");
-		close(fd);
-		return 0;
-	}
+	if (mode == DUMP_JITED)
+		arrays = 1UL << BPF_PROG_INFO_JITED_INSNS;
+	else
+		arrays = 1UL << BPF_PROG_INFO_XLATED_INSNS;
 
-	buf_size = *member_len;
+	arrays |= 1UL << BPF_PROG_INFO_JITED_KSYMS;
+	arrays |= 1UL << BPF_PROG_INFO_JITED_FUNC_LENS;
+	arrays |= 1UL << BPF_PROG_INFO_FUNC_INFO;
+	arrays |= 1UL << BPF_PROG_INFO_LINE_INFO;
+	arrays |= 1UL << BPF_PROG_INFO_JITED_LINE_INFO;
 
-	buf = malloc(buf_size);
-	if (!buf) {
-		p_err("mem alloc failed");
-		close(fd);
+	info_linear = bpf_program__get_prog_info_linear(fd, arrays);
+	close(fd);
+	if (IS_ERR_OR_NULL(info_linear)) {
+		p_err("can't get prog info: %s", strerror(errno));
 		return -1;
 	}
 
-	nr_func_ksyms = info.nr_jited_ksyms;
-	if (nr_func_ksyms) {
-		func_ksyms = malloc(nr_func_ksyms * sizeof(__u64));
-		if (!func_ksyms) {
-			p_err("mem alloc failed");
-			close(fd);
-			goto err_free;
-		}
-	}
-
-	nr_func_lens = info.nr_jited_func_lens;
-	if (nr_func_lens) {
-		func_lens = malloc(nr_func_lens * sizeof(__u32));
-		if (!func_lens) {
-			p_err("mem alloc failed");
-			close(fd);
+	info = &info_linear->info;
+	if (mode == DUMP_JITED) {
+		if (info->jited_prog_len == 0) {
+			p_info("no instructions returned");
 			goto err_free;
 		}
-	}
-
-	nr_finfo = info.nr_func_info;
-	finfo_rec_size = info.func_info_rec_size;
-	if (nr_finfo && finfo_rec_size) {
-		func_info = malloc(nr_finfo * finfo_rec_size);
-		if (!func_info) {
-			p_err("mem alloc failed");
-			close(fd);
+		buf = (unsigned char *)(info->jited_prog_insns);
+		member_len = info->jited_prog_len;
+	} else {	/* DUMP_XLATED */
+		if (info->xlated_prog_len == 0) {
+			p_err("error retrieving insn dump: kernel.kptr_restrict set?");
 			goto err_free;
 		}
+		buf = (unsigned char *)info->xlated_prog_insns;
+		member_len = info->xlated_prog_len;
 	}
 
-	linfo_rec_size = info.line_info_rec_size;
-	if (info.nr_line_info && linfo_rec_size && info.btf_id) {
-		nr_linfo = info.nr_line_info;
-		linfo = malloc(nr_linfo * linfo_rec_size);
-		if (!linfo) {
-			p_err("mem alloc failed");
-			close(fd);
-			goto err_free;
-		}
-	}
-
-	jited_linfo_rec_size = info.jited_line_info_rec_size;
-	if (info.nr_jited_line_info &&
-	    jited_linfo_rec_size &&
-	    info.nr_jited_ksyms &&
-	    info.nr_jited_func_lens &&
-	    info.btf_id) {
-		nr_jited_linfo = info.nr_jited_line_info;
-		jited_linfo = malloc(nr_jited_linfo * jited_linfo_rec_size);
-		if (!jited_linfo) {
-			p_err("mem alloc failed");
-			close(fd);
-			goto err_free;
-		}
-	}
-
-	memset(&info, 0, sizeof(info));
-
-	*member_ptr = ptr_to_u64(buf);
-	*member_len = buf_size;
-	info.jited_ksyms = ptr_to_u64(func_ksyms);
-	info.nr_jited_ksyms = nr_func_ksyms;
-	info.jited_func_lens = ptr_to_u64(func_lens);
-	info.nr_jited_func_lens = nr_func_lens;
-	info.nr_func_info = nr_finfo;
-	info.func_info_rec_size = finfo_rec_size;
-	info.func_info = ptr_to_u64(func_info);
-	info.nr_line_info = nr_linfo;
-	info.line_info_rec_size = linfo_rec_size;
-	info.line_info = ptr_to_u64(linfo);
-	info.nr_jited_line_info = nr_jited_linfo;
-	info.jited_line_info_rec_size = jited_linfo_rec_size;
-	info.jited_line_info = ptr_to_u64(jited_linfo);
-
-	err = bpf_obj_get_info_by_fd(fd, &info, &len);
-	close(fd);
-	if (err) {
-		p_err("can't get prog info: %s", strerror(errno));
-		goto err_free;
-	}
-
-	if (*member_len > buf_size) {
-		p_err("too many instructions returned");
-		goto err_free;
-	}
-
-	if (info.nr_jited_ksyms > nr_func_ksyms) {
-		p_err("too many addresses returned");
-		goto err_free;
-	}
-
-	if (info.nr_jited_func_lens > nr_func_lens) {
-		p_err("too many values returned");
-		goto err_free;
-	}
-
-	if (info.nr_func_info != nr_finfo) {
-		p_err("incorrect nr_func_info %d vs. expected %d",
-		      info.nr_func_info, nr_finfo);
-		goto err_free;
-	}
-
-	if (info.func_info_rec_size != finfo_rec_size) {
-		p_err("incorrect func_info_rec_size %d vs. expected %d",
-		      info.func_info_rec_size, finfo_rec_size);
-		goto err_free;
-	}
-
-	if (linfo && info.nr_line_info != nr_linfo) {
-		p_err("incorrect nr_line_info %u vs. expected %u",
-		      info.nr_line_info, nr_linfo);
-		goto err_free;
-	}
-
-	if (info.line_info_rec_size != linfo_rec_size) {
-		p_err("incorrect line_info_rec_size %u vs. expected %u",
-		      info.line_info_rec_size, linfo_rec_size);
-		goto err_free;
-	}
-
-	if (jited_linfo && info.nr_jited_line_info != nr_jited_linfo) {
-		p_err("incorrect nr_jited_line_info %u vs. expected %u",
-		      info.nr_jited_line_info, nr_jited_linfo);
-		goto err_free;
-	}
-
-	if (info.jited_line_info_rec_size != jited_linfo_rec_size) {
-		p_err("incorrect jited_line_info_rec_size %u vs. expected %u",
-		      info.jited_line_info_rec_size, jited_linfo_rec_size);
-		goto err_free;
-	}
-
-	if ((member_len == &info.jited_prog_len &&
-	     info.jited_prog_insns == 0) ||
-	    (member_len == &info.xlated_prog_len &&
-	     info.xlated_prog_insns == 0)) {
-		p_err("error retrieving insn dump: kernel.kptr_restrict set?");
-		goto err_free;
-	}
-
-	if (info.btf_id && btf__get_from_id(info.btf_id, &btf)) {
+	if (info->btf_id && btf__get_from_id(info->btf_id, &btf)) {
 		p_err("failed to get btf");
 		goto err_free;
 	}
 
-	if (nr_linfo) {
-		prog_linfo = bpf_prog_linfo__new(&info);
+	func_info = (void *)info->func_info;
+
+	if (info->nr_line_info) {
+		prog_linfo = bpf_prog_linfo__new(info);
 		if (!prog_linfo)
 			p_info("error in processing bpf_line_info.  continue without it.");
 	}
@@ -647,9 +512,9 @@ static int do_dump(int argc, char **argv)
 			goto err_free;
 		}
 
-		n = write(fd, buf, *member_len);
+		n = write(fd, buf, member_len);
 		close(fd);
-		if (n != *member_len) {
+		if (n != member_len) {
 			p_err("error writing output file: %s",
 			      n < 0 ? strerror(errno) : "short write");
 			goto err_free;
@@ -657,19 +522,19 @@ static int do_dump(int argc, char **argv)
 
 		if (json_output)
 			jsonw_null(json_wtr);
-	} else if (member_len == &info.jited_prog_len) {
+	} else if (mode == DUMP_JITED) {
 		const char *name = NULL;
 
-		if (info.ifindex) {
-			name = ifindex_to_bfd_params(info.ifindex,
-						     info.netns_dev,
-						     info.netns_ino,
+		if (info->ifindex) {
+			name = ifindex_to_bfd_params(info->ifindex,
+						     info->netns_dev,
+						     info->netns_ino,
 						     &disasm_opt);
 			if (!name)
 				goto err_free;
 		}
 
-		if (info.nr_jited_func_lens && info.jited_func_lens) {
+		if (info->nr_jited_func_lens && info->jited_func_lens) {
 			struct kernel_sym *sym = NULL;
 			struct bpf_func_info *record;
 			char sym_name[SYM_MAX_NAME];
@@ -677,17 +542,16 @@ static int do_dump(int argc, char **argv)
 			__u64 *ksyms = NULL;
 			__u32 *lens;
 			__u32 i;
-
-			if (info.nr_jited_ksyms) {
+			if (info->nr_jited_ksyms) {
 				kernel_syms_load(&dd);
-				ksyms = (__u64 *) info.jited_ksyms;
+				ksyms = (__u64 *) info->jited_ksyms;
 			}
 
 			if (json_output)
 				jsonw_start_array(json_wtr);
 
-			lens = (__u32 *) info.jited_func_lens;
-			for (i = 0; i < info.nr_jited_func_lens; i++) {
+			lens = (__u32 *) info->jited_func_lens;
+			for (i = 0; i < info->nr_jited_func_lens; i++) {
 				if (ksyms) {
 					sym = kernel_syms_search(&dd, ksyms[i]);
 					if (sym)
@@ -699,7 +563,7 @@ static int do_dump(int argc, char **argv)
 				}
 
 				if (func_info) {
-					record = func_info + i * finfo_rec_size;
+					record = func_info + i * info->func_info_rec_size;
 					btf_dumper_type_only(btf, record->type_id,
 							     func_sig,
 							     sizeof(func_sig));
@@ -736,49 +600,37 @@ static int do_dump(int argc, char **argv)
 			if (json_output)
 				jsonw_end_array(json_wtr);
 		} else {
-			disasm_print_insn(buf, *member_len, opcodes, name,
+			disasm_print_insn(buf, member_len, opcodes, name,
 					  disasm_opt, btf, NULL, 0, 0, false);
 		}
 	} else if (visual) {
 		if (json_output)
 			jsonw_null(json_wtr);
 		else
-			dump_xlated_cfg(buf, *member_len);
+			dump_xlated_cfg(buf, member_len);
 	} else {
 		kernel_syms_load(&dd);
-		dd.nr_jited_ksyms = info.nr_jited_ksyms;
-		dd.jited_ksyms = (__u64 *) info.jited_ksyms;
+		dd.nr_jited_ksyms = info->nr_jited_ksyms;
+		dd.jited_ksyms = (__u64 *) info->jited_ksyms;
 		dd.btf = btf;
 		dd.func_info = func_info;
-		dd.finfo_rec_size = finfo_rec_size;
+		dd.finfo_rec_size = info->func_info_rec_size;
 		dd.prog_linfo = prog_linfo;
 
 		if (json_output)
-			dump_xlated_json(&dd, buf, *member_len, opcodes,
+			dump_xlated_json(&dd, buf, member_len, opcodes,
 					 linum);
 		else
-			dump_xlated_plain(&dd, buf, *member_len, opcodes,
+			dump_xlated_plain(&dd, buf, member_len, opcodes,
 					  linum);
 		kernel_syms_destroy(&dd);
 	}
 
-	free(buf);
-	free(func_ksyms);
-	free(func_lens);
-	free(func_info);
-	free(linfo);
-	free(jited_linfo);
-	bpf_prog_linfo__free(prog_linfo);
+	free(info_linear);
 	return 0;
 
 err_free:
-	free(buf);
-	free(func_ksyms);
-	free(func_lens);
-	free(func_info);
-	free(linfo);
-	free(jited_linfo);
-	bpf_prog_linfo__free(prog_linfo);
+	free(info_linear);
 	return -1;
 }
 
-- 
2.17.1

[PATCH v2 perf,bpf 04/11] perf, bpf: synthesize bpf events with bpf_program__get_prog_info_linear()

[Song Liu]

With bpf_program__get_prog_info_linear, we can simplify the logic that
synthesizes bpf events.

This patch doesn't change the behavior of the code.

Signed-off-by: Song Liu <songliubraving@fb.com>
---
 tools/perf/util/bpf-event.c | 118 ++++++++++++------------------------
 1 file changed, 40 insertions(+), 78 deletions(-)

diff --git a/tools/perf/util/bpf-event.c b/tools/perf/util/bpf-event.c
index 796ef793f4ce..e6dfb95029e5 100644
--- a/tools/perf/util/bpf-event.c
+++ b/tools/perf/util/bpf-event.c
@@ -3,7 +3,9 @@
 #include <stdlib.h>
 #include <bpf/bpf.h>
 #include <bpf/btf.h>
+#include <bpf/libbpf.h>
 #include <linux/btf.h>
+#include <linux/err.h>
 #include "bpf-event.h"
 #include "debug.h"
 #include "symbol.h"
@@ -49,99 +51,62 @@ static int perf_event__synthesize_one_bpf_prog(struct perf_tool *tool,
 {
 	struct ksymbol_event *ksymbol_event = &event->ksymbol_event;
 	struct bpf_event *bpf_event = &event->bpf_event;
-	u32 sub_prog_cnt, i, func_info_rec_size = 0;
-	u8 (*prog_tags)[BPF_TAG_SIZE] = NULL;
-	struct bpf_prog_info info = { .type = 0, };
-	u32 info_len = sizeof(info);
-	void *func_infos = NULL;
-	u64 *prog_addrs = NULL;
+	struct bpf_prog_info_linear *info_linear;
+	struct bpf_prog_info *info;
 	struct btf *btf = NULL;
-	u32 *prog_lens = NULL;
 	bool has_btf = false;
-	char errbuf[512];
+	u32 sub_prog_cnt, i;
 	int err = 0;
+	u64 arrays;
 
-	/* Call bpf_obj_get_info_by_fd() to get sizes of arrays */
-	err = bpf_obj_get_info_by_fd(fd, &info, &info_len);
+	arrays = 1UL << BPF_PROG_INFO_JITED_KSYMS;
+	arrays |= 1UL << BPF_PROG_INFO_JITED_FUNC_LENS;
+	arrays |= 1UL << BPF_PROG_INFO_FUNC_INFO;
+	arrays |= 1UL << BPF_PROG_INFO_PROG_TAGS;
 
-	if (err) {
-		pr_debug("%s: failed to get BPF program info: %s, aborting\n",
-			 __func__, str_error_r(errno, errbuf, sizeof(errbuf)));
+	info_linear = bpf_program__get_prog_info_linear(fd, arrays);
+	if (IS_ERR_OR_NULL(info_linear)) {
+		info_linear = NULL;
+		pr_debug("%s: failed to get BPF program info. aborting\n", __func__);
 		return -1;
 	}
-	if (info_len < offsetof(struct bpf_prog_info, prog_tags)) {
+
+	if (info_linear->info_len < offsetof(struct bpf_prog_info, prog_tags)) {
 		pr_debug("%s: the kernel is too old, aborting\n", __func__);
 		return -2;
 	}
 
+	info = &info_linear->info;
+
 	/* number of ksyms, func_lengths, and tags should match */
-	sub_prog_cnt = info.nr_jited_ksyms;
-	if (sub_prog_cnt != info.nr_prog_tags ||
-	    sub_prog_cnt != info.nr_jited_func_lens)
+	sub_prog_cnt = info->nr_jited_ksyms;
+	if (sub_prog_cnt != info->nr_prog_tags ||
+	    sub_prog_cnt != info->nr_jited_func_lens)
 		return -1;
 
 	/* check BTF func info support */
-	if (info.btf_id && info.nr_func_info && info.func_info_rec_size) {
+	if (info->btf_id && info->nr_func_info && info->func_info_rec_size) {
 		/* btf func info number should be same as sub_prog_cnt */
-		if (sub_prog_cnt != info.nr_func_info) {
+		if (sub_prog_cnt != info->nr_func_info) {
 			pr_debug("%s: mismatch in BPF sub program count and BTF function info count, aborting\n", __func__);
-			return -1;
-		}
-		if (btf__get_from_id(info.btf_id, &btf)) {
-			pr_debug("%s: failed to get BTF of id %u, aborting\n", __func__, info.btf_id);
-			return -1;
+			err = -1;
+			goto out;
 		}
-		func_info_rec_size = info.func_info_rec_size;
-		func_infos = calloc(sub_prog_cnt, func_info_rec_size);
-		if (!func_infos) {
-			pr_debug("%s: failed to allocate memory for func_infos, aborting\n", __func__);
-			return -1;
+		if (btf__get_from_id(info->btf_id, &btf)) {
+			pr_debug("%s: failed to get BTF of id %u, aborting\n", __func__, info->btf_id);
+			err = -1;
+			btf = NULL;
+			goto out;
 		}
 		has_btf = true;
 	}
 
-	/*
-	 * We need address, length, and tag for each sub program.
-	 * Allocate memory and call bpf_obj_get_info_by_fd() again
-	 */
-	prog_addrs = calloc(sub_prog_cnt, sizeof(u64));
-	if (!prog_addrs) {
-		pr_debug("%s: failed to allocate memory for prog_addrs, aborting\n", __func__);
-		goto out;
-	}
-	prog_lens = calloc(sub_prog_cnt, sizeof(u32));
-	if (!prog_lens) {
-		pr_debug("%s: failed to allocate memory for prog_lens, aborting\n", __func__);
-		goto out;
-	}
-	prog_tags = calloc(sub_prog_cnt, BPF_TAG_SIZE);
-	if (!prog_tags) {
-		pr_debug("%s: failed to allocate memory for prog_tags, aborting\n", __func__);
-		goto out;
-	}
-
-	memset(&info, 0, sizeof(info));
-	info.nr_jited_ksyms = sub_prog_cnt;
-	info.nr_jited_func_lens = sub_prog_cnt;
-	info.nr_prog_tags = sub_prog_cnt;
-	info.jited_ksyms = ptr_to_u64(prog_addrs);
-	info.jited_func_lens = ptr_to_u64(prog_lens);
-	info.prog_tags = ptr_to_u64(prog_tags);
-	info_len = sizeof(info);
-	if (has_btf) {
-		info.nr_func_info = sub_prog_cnt;
-		info.func_info_rec_size = func_info_rec_size;
-		info.func_info = ptr_to_u64(func_infos);
-	}
-
-	err = bpf_obj_get_info_by_fd(fd, &info, &info_len);
-	if (err) {
-		pr_debug("%s: failed to get BPF program info, aborting\n", __func__);
-		goto out;
-	}
-
 	/* Synthesize PERF_RECORD_KSYMBOL */
 	for (i = 0; i < sub_prog_cnt; i++) {
+		u8 (*prog_tags)[BPF_TAG_SIZE] = (void *)(info->prog_tags);
+		__u32 *prog_lens = (__u32 *)(info->jited_func_lens);
+		__u64 *prog_addrs = (__u64 *)(info->jited_ksyms);
+		void *func_infos = (void *)(info->func_info);
 		const struct bpf_func_info *finfo;
 		const char *short_name = NULL;
 		const struct btf_type *t;
@@ -163,13 +128,13 @@ static int perf_event__synthesize_one_bpf_prog(struct perf_tool *tool,
 					 KSYM_NAME_LEN - name_len,
 					 prog_tags[i], BPF_TAG_SIZE);
 		if (has_btf) {
-			finfo = func_infos + i * info.func_info_rec_size;
+			finfo = func_infos + i * info->func_info_rec_size;
 			t = btf__type_by_id(btf, finfo->type_id);
 			short_name = btf__name_by_offset(btf, t->name_off);
 		} else if (i == 0 && sub_prog_cnt == 1) {
 			/* no subprog */
-			if (info.name[0])
-				short_name = info.name;
+			if (info->name[0])
+				short_name = info->name;
 		} else
 			short_name = "F";
 		if (short_name)
@@ -195,9 +160,9 @@ static int perf_event__synthesize_one_bpf_prog(struct perf_tool *tool,
 			},
 			.type = PERF_BPF_EVENT_PROG_LOAD,
 			.flags = 0,
-			.id = info.id,
+			.id = info->id,
 		};
-		memcpy(bpf_event->tag, prog_tags[i], BPF_TAG_SIZE);
+		memcpy(bpf_event->tag, info->tag, BPF_TAG_SIZE);
 		memset((void *)event + event->header.size, 0, machine->id_hdr_size);
 		event->header.size += machine->id_hdr_size;
 		err = perf_tool__process_synth_event(tool, event,
@@ -205,10 +170,7 @@ static int perf_event__synthesize_one_bpf_prog(struct perf_tool *tool,
 	}
 
 out:
-	free(prog_tags);
-	free(prog_lens);
-	free(prog_addrs);
-	free(func_infos);
+	free(info_linear);
 	free(btf);
 	return err ? -1 : 0;
 }
-- 
2.17.1

[PATCH v2 perf,bpf 05/11] perf, bpf: save bpf_prog_info in a rbtree in perf_env

[Song Liu]

bpf_prog_info contains information necessary to annotate bpf programs.
This patch saves bpf_prog_info for bpf programs loaded in the system.

Signed-off-by: Song Liu <songliubraving@fb.com>
---
 tools/perf/builtin-record.c |  2 +-
 tools/perf/builtin-top.c    |  2 +-
 tools/perf/util/bpf-event.c | 39 +++++++++++++----
 tools/perf/util/bpf-event.h | 15 +++++--
 tools/perf/util/env.c       | 83 +++++++++++++++++++++++++++++++++++++
 tools/perf/util/env.h       | 14 +++++++
 6 files changed, 142 insertions(+), 13 deletions(-)

diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index 88ea11d57c6f..2355e0a9eda0 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -1083,7 +1083,7 @@ static int record__synthesize(struct record *rec, bool tail)
 		return err;
 	}
 
-	err = perf_event__synthesize_bpf_events(tool, process_synthesized_event,
+	err = perf_event__synthesize_bpf_events(session, process_synthesized_event,
 						machine, opts);
 	if (err < 0)
 		pr_warning("Couldn't synthesize bpf events.\n");
diff --git a/tools/perf/builtin-top.c b/tools/perf/builtin-top.c
index 5a486d4de56e..27d8d42e0a4d 100644
--- a/tools/perf/builtin-top.c
+++ b/tools/perf/builtin-top.c
@@ -1216,7 +1216,7 @@ static int __cmd_top(struct perf_top *top)
 
 	init_process_thread(top);
 
-	ret = perf_event__synthesize_bpf_events(&top->tool, perf_event__process,
+	ret = perf_event__synthesize_bpf_events(top->session, perf_event__process,
 						&top->session->machines.host,
 						&top->record_opts);
 	if (ret < 0)
diff --git a/tools/perf/util/bpf-event.c b/tools/perf/util/bpf-event.c
index e6dfb95029e5..ead599bc4f4e 100644
--- a/tools/perf/util/bpf-event.c
+++ b/tools/perf/util/bpf-event.c
@@ -1,15 +1,13 @@
 // SPDX-License-Identifier: GPL-2.0
 #include <errno.h>
 #include <stdlib.h>
-#include <bpf/bpf.h>
-#include <bpf/btf.h>
-#include <bpf/libbpf.h>
-#include <linux/btf.h>
 #include <linux/err.h>
 #include "bpf-event.h"
 #include "debug.h"
 #include "symbol.h"
 #include "machine.h"
+#include "env.h"
+#include "session.h"
 
 #define ptr_to_u64(ptr)    ((__u64)(unsigned long)(ptr))
 
@@ -42,7 +40,7 @@ int machine__process_bpf_event(struct machine *machine __maybe_unused,
  *   -1 for failures;
  *   -2 for lack of kernel support.
  */
-static int perf_event__synthesize_one_bpf_prog(struct perf_tool *tool,
+static int perf_event__synthesize_one_bpf_prog(struct perf_session *session,
 					       perf_event__handler_t process,
 					       struct machine *machine,
 					       int fd,
@@ -52,17 +50,29 @@ static int perf_event__synthesize_one_bpf_prog(struct perf_tool *tool,
 	struct ksymbol_event *ksymbol_event = &event->ksymbol_event;
 	struct bpf_event *bpf_event = &event->bpf_event;
 	struct bpf_prog_info_linear *info_linear;
+	struct perf_tool *tool = session->tool;
+	struct bpf_prog_info_node *info_node;
 	struct bpf_prog_info *info;
 	struct btf *btf = NULL;
 	bool has_btf = false;
+	struct perf_env *env;
 	u32 sub_prog_cnt, i;
 	int err = 0;
 	u64 arrays;
 
+	/*
+	 * for perf-record and perf-report use header.env;
+	 * otherwise, use global perf_env.
+	 */
+	env = session->data ? &session->header.env : &perf_env;
+
 	arrays = 1UL << BPF_PROG_INFO_JITED_KSYMS;
 	arrays |= 1UL << BPF_PROG_INFO_JITED_FUNC_LENS;
 	arrays |= 1UL << BPF_PROG_INFO_FUNC_INFO;
 	arrays |= 1UL << BPF_PROG_INFO_PROG_TAGS;
+	arrays |= 1UL << BPF_PROG_INFO_JITED_INSNS;
+	arrays |= 1UL << BPF_PROG_INFO_LINE_INFO;
+	arrays |= 1UL << BPF_PROG_INFO_JITED_LINE_INFO;
 
 	info_linear = bpf_program__get_prog_info_linear(fd, arrays);
 	if (IS_ERR_OR_NULL(info_linear)) {
@@ -151,8 +161,8 @@ static int perf_event__synthesize_one_bpf_prog(struct perf_tool *tool,
 						     machine, process);
 	}
 
-	/* Synthesize PERF_RECORD_BPF_EVENT */
 	if (opts->bpf_event) {
+		/* Synthesize PERF_RECORD_BPF_EVENT */
 		*bpf_event = (struct bpf_event){
 			.header = {
 				.type = PERF_RECORD_BPF_EVENT,
@@ -165,6 +175,19 @@ static int perf_event__synthesize_one_bpf_prog(struct perf_tool *tool,
 		memcpy(bpf_event->tag, info->tag, BPF_TAG_SIZE);
 		memset((void *)event + event->header.size, 0, machine->id_hdr_size);
 		event->header.size += machine->id_hdr_size;
+
+		/* save bpf_prog_info to env */
+		info_node = malloc(sizeof(struct bpf_prog_info_node));
+		if (info_node) {
+			info_node->info_linear = info_linear;
+			perf_env__insert_bpf_prog_info(env, info_node);
+			info_linear = NULL;
+		}
+
+		/*
+		 * process after saving bpf_prog_info to env, so that
+		 * required information is ready for look up
+		 */
 		err = perf_tool__process_synth_event(tool, event,
 						     machine, process);
 	}
@@ -175,7 +198,7 @@ static int perf_event__synthesize_one_bpf_prog(struct perf_tool *tool,
 	return err ? -1 : 0;
 }
 
-int perf_event__synthesize_bpf_events(struct perf_tool *tool,
+int perf_event__synthesize_bpf_events(struct perf_session *session,
 				      perf_event__handler_t process,
 				      struct machine *machine,
 				      struct record_opts *opts)
@@ -209,7 +232,7 @@ int perf_event__synthesize_bpf_events(struct perf_tool *tool,
 			continue;
 		}
 
-		err = perf_event__synthesize_one_bpf_prog(tool, process,
+		err = perf_event__synthesize_one_bpf_prog(session, process,
 							  machine, fd,
 							  event, opts);
 		close(fd);
diff --git a/tools/perf/util/bpf-event.h b/tools/perf/util/bpf-event.h
index 7890067e1a37..11e6730b6105 100644
--- a/tools/perf/util/bpf-event.h
+++ b/tools/perf/util/bpf-event.h
@@ -3,19 +3,28 @@
 #define __PERF_BPF_EVENT_H
 
 #include <linux/compiler.h>
+#include <bpf/bpf.h>
+#include <bpf/btf.h>
+#include <bpf/libbpf.h>
+#include <linux/btf.h>
+#include <linux/rbtree.h>
 #include "event.h"
 
 struct machine;
 union perf_event;
 struct perf_sample;
-struct perf_tool;
 struct record_opts;
 
+struct bpf_prog_info_node {
+	struct bpf_prog_info_linear	*info_linear;
+	struct rb_node			rb_node;
+};
+
 #ifdef HAVE_LIBBPF_SUPPORT
 int machine__process_bpf_event(struct machine *machine, union perf_event *event,
 			       struct perf_sample *sample);
 
-int perf_event__synthesize_bpf_events(struct perf_tool *tool,
+int perf_event__synthesize_bpf_events(struct perf_session *session,
 				      perf_event__handler_t process,
 				      struct machine *machine,
 				      struct record_opts *opts);
@@ -27,7 +36,7 @@ static inline int machine__process_bpf_event(struct machine *machine __maybe_unu
 	return 0;
 }
 
-static inline int perf_event__synthesize_bpf_events(struct perf_tool *tool __maybe_unused,
+static inline int perf_event__synthesize_bpf_events(struct perf_session *session __maybe_unused,
 						    perf_event__handler_t process __maybe_unused,
 						    struct machine *machine __maybe_unused,
 						    struct record_opts *opts __maybe_unused)
diff --git a/tools/perf/util/env.c b/tools/perf/util/env.c
index 4c23779e271a..665b6fe3c7b2 100644
--- a/tools/perf/util/env.c
+++ b/tools/perf/util/env.c
@@ -8,10 +8,86 @@
 
 struct perf_env perf_env;
 
+void perf_env__insert_bpf_prog_info(struct perf_env *env,
+				    struct bpf_prog_info_node *info_node)
+{
+	__u32 prog_id = info_node->info_linear->info.id;
+	struct bpf_prog_info_node *node;
+	struct rb_node *parent = NULL;
+	struct rb_node **p;
+
+	down_write(&env->bpf_info_lock);
+	p = &env->bpf_prog_infos.rb_node;
+
+	while (*p != NULL) {
+		parent = *p;
+		node = rb_entry(parent, struct bpf_prog_info_node, rb_node);
+		if (prog_id < node->info_linear->info.id) {
+			p = &(*p)->rb_left;
+		} else if (prog_id > node->info_linear->info.id) {
+			p = &(*p)->rb_right;
+		} else {
+			pr_debug("duplicated bpf prog info %u\n", prog_id);
+			up_write(&env->bpf_info_lock);
+			return;
+		}
+	}
+
+	rb_link_node(&info_node->rb_node, parent, p);
+	rb_insert_color(&info_node->rb_node, &env->bpf_prog_infos);
+	up_write(&env->bpf_info_lock);
+}
+
+struct bpf_prog_info_node *perf_env__find_bpf_prog_info(struct perf_env *env,
+							__u32 prog_id)
+{
+	struct bpf_prog_info_node *node = NULL;
+	struct rb_node *n;
+
+	down_read(&env->bpf_info_lock);
+	n = env->bpf_prog_infos.rb_node;
+
+	while (n) {
+		node = rb_entry(n, struct bpf_prog_info_node, rb_node);
+		if (prog_id < node->info_linear->info.id)
+			n = n->rb_left;
+		else if (prog_id > node->info_linear->info.id)
+			n = n->rb_right;
+		else
+			break;
+	}
+
+	up_read(&env->bpf_info_lock);
+	return node;
+}
+
+/* purge data in bpf_prog_infos tree */
+static void purge_bpf_info(struct perf_env *env)
+{
+	struct rb_root *root;
+	struct rb_node *next;
+
+	down_write(&env->bpf_info_lock);
+
+	root = &env->bpf_prog_infos;
+	next = rb_first(root);
+
+	while (next) {
+		struct bpf_prog_info_node *node;
+
+		node = rb_entry(next, struct bpf_prog_info_node, rb_node);
+		next = rb_next(&node->rb_node);
+		rb_erase_init(&node->rb_node, root);
+		free(node);
+	}
+	up_write(&env->bpf_info_lock);
+}
+
 void perf_env__exit(struct perf_env *env)
 {
 	int i;
 
+	purge_bpf_info(env);
 	zfree(&env->hostname);
 	zfree(&env->os_release);
 	zfree(&env->version);
@@ -38,6 +114,12 @@ void perf_env__exit(struct perf_env *env)
 	zfree(&env->memory_nodes);
 }
 
+static void init_bpf_rb_trees(struct perf_env *env)
+{
+	env->bpf_prog_infos = RB_ROOT;
+	init_rwsem(&env->bpf_info_lock);
+}
+
 int perf_env__set_cmdline(struct perf_env *env, int argc, const char *argv[])
 {
 	int i;
@@ -59,6 +141,7 @@ int perf_env__set_cmdline(struct perf_env *env, int argc, const char *argv[])
 
 	env->nr_cmdline = argc;
 
+	init_bpf_rb_trees(env);
 	return 0;
 out_free:
 	zfree(&env->cmdline_argv);
diff --git a/tools/perf/util/env.h b/tools/perf/util/env.h
index d01b8355f4ca..a6d25e91bc98 100644
--- a/tools/perf/util/env.h
+++ b/tools/perf/util/env.h
@@ -3,7 +3,10 @@
 #define __PERF_ENV_H
 
 #include <linux/types.h>
+#include <linux/rbtree.h>
 #include "cpumap.h"
+#include "rwsem.h"
+#include "bpf-event.h"
 
 struct cpu_topology_map {
 	int	socket_id;
@@ -64,6 +67,13 @@ struct perf_env {
 	struct memory_node	*memory_nodes;
 	unsigned long long	 memory_bsize;
 	u64                     clockid_res_ns;
+
+	/*
+	 * bpf_info_lock protects bpf rbtrees. This is needed because the
+	 * trees are accessed by different threads in perf-top
+	 */
+	struct rw_semaphore	bpf_info_lock;
+	struct rb_root		bpf_prog_infos;
 };
 
 extern struct perf_env perf_env;
@@ -80,4 +90,8 @@ const char *perf_env__arch(struct perf_env *env);
 const char *perf_env__raw_arch(struct perf_env *env);
 int perf_env__nr_cpus_avail(struct perf_env *env);
 
+void perf_env__insert_bpf_prog_info(struct perf_env *env,
+				    struct bpf_prog_info_node *info_node);
+struct bpf_prog_info_node *perf_env__find_bpf_prog_info(struct perf_env *env,
+							__u32 prog_id);
 #endif /* __PERF_ENV_H */
-- 
2.17.1

[PATCH v2 perf,bpf 06/11] perf, bpf: save bpf_prog_info information as headers to perf.data

[Song Liu]

This patch enables perf-record to save bpf_prog_info information as
headers to perf.data. A new header type HEADER_BPF_PROG_INFO is
introduced for this data.

Signed-off-by: Song Liu <songliubraving@fb.com>
---
 tools/perf/util/header.c | 134 ++++++++++++++++++++++++++++++++++++++-
 tools/perf/util/header.h |   1 +
 2 files changed, 134 insertions(+), 1 deletion(-)

diff --git a/tools/perf/util/header.c b/tools/perf/util/header.c
index dec6d218c31c..2ae76a9d06f6 100644
--- a/tools/perf/util/header.c
+++ b/tools/perf/util/header.c
@@ -1080,6 +1080,51 @@ static int write_clockid(struct feat_fd *ff,
 			sizeof(ff->ph->env.clockid_res_ns));
 }
 
+static int write_bpf_prog_info(struct feat_fd *ff,
+			       struct perf_evlist *evlist __maybe_unused)
+{
+	struct perf_env *env = &ff->ph->env;
+	struct rb_root *root;
+	struct rb_node *next;
+	u32 count = 0;
+	int ret;
+
+	down_read(&env->bpf_info_lock);
+
+	root = &env->bpf_prog_infos;
+	next = rb_first(root);
+	while (next) {
+		++count;
+		next = rb_next(next);
+	}
+
+	ret = do_write(ff, &count, sizeof(count));
+
+	if (ret < 0)
+		goto out;
+
+	next = rb_first(root);
+	while (next) {
+		struct bpf_prog_info_node *node;
+		size_t len;
+
+		node = rb_entry(next, struct bpf_prog_info_node, rb_node);
+		next = rb_next(&node->rb_node);
+		len = sizeof(struct bpf_prog_info_linear) +
+			node->info_linear->data_len;
+
+		/* before writing to file, translate address to offset */
+		bpf_program__bpil_addr_to_offs(node->info_linear);
+		ret = do_write(ff, node->info_linear, len);
+		bpf_program__bpil_offs_to_addr(node->info_linear);
+		if (ret < 0)
+			goto out;
+	}
+out:
+	up_read(&env->bpf_info_lock);
+	return ret;
+}
+
 static int cpu_cache_level__sort(const void *a, const void *b)
 {
 	struct cpu_cache_level *cache_a = (struct cpu_cache_level *)a;
@@ -1560,6 +1605,29 @@ static void print_clockid(struct feat_fd *ff, FILE *fp)
 		ff->ph->env.clockid_res_ns * 1000);
 }
 
+static void print_bpf_prog_info(struct feat_fd *ff, FILE *fp)
+{
+	struct perf_env *env = &ff->ph->env;
+	struct rb_root *root;
+	struct rb_node *next;
+
+	down_read(&env->bpf_info_lock);
+
+	root = &env->bpf_prog_infos;
+	next = rb_first(root);
+
+	while (next) {
+		struct bpf_prog_info_node *node;
+
+		node = rb_entry(next, struct bpf_prog_info_node, rb_node);
+		next = rb_next(&node->rb_node);
+		fprintf(fp, "# bpf_prog_info of id %u\n",
+			node->info_linear->info.id);
+	}
+
+	up_read(&env->bpf_info_lock);
+}
+
 static void free_event_desc(struct perf_evsel *events)
 {
 	struct perf_evsel *evsel;
@@ -2592,6 +2660,69 @@ static int process_clockid(struct feat_fd *ff,
 	return 0;
 }
 
+static int process_bpf_prog_info(struct feat_fd *ff,
+				 void *data __maybe_unused)
+{
+	struct bpf_prog_info_linear *info_linear;
+	struct bpf_prog_info_node *info_node;
+	struct perf_env *env = &ff->ph->env;
+	u32 count, i;
+	int err = -1;
+
+	if (do_read_u32(ff, &count))
+		return -1;
+
+	down_write(&env->bpf_info_lock);
+
+	for (i = 0; i < count; ++i) {
+		u32 info_len, data_len;
+
+		info_linear = NULL;
+		info_node = NULL;
+		if (do_read_u32(ff, &info_len))
+			goto out;
+		if (do_read_u32(ff, &data_len))
+			goto out;
+
+		if (info_len > sizeof(struct bpf_prog_info)) {
+			pr_warning("detected invalid bpf_prog_info\n");
+			goto out;
+		}
+
+		info_linear = malloc(sizeof(struct bpf_prog_info_linear) +
+				     data_len);
+		if (!info_linear)
+			goto out;
+		info_linear->info_len = sizeof(struct bpf_prog_info);
+		info_linear->data_len = data_len;
+		if (do_read_u64(ff, (u64 *)(&info_linear->arrays)))
+			goto out;
+		if (__do_read(ff, &info_linear->info, info_len))
+			goto out;
+		if (info_len < sizeof(struct bpf_prog_info))
+			memset(((void *)(&info_linear->info)) + info_len, 0,
+			       sizeof(struct bpf_prog_info) - info_len);
+
+		if (__do_read(ff, info_linear->data, data_len))
+			goto out;
+		info_node = malloc(sizeof(struct bpf_prog_info_node));
+		if (!info_node)
+			goto out;
+
+		/* after reading from file, translate offset to address */
+		bpf_program__bpil_offs_to_addr(info_linear);
+		info_node->info_linear = info_linear;
+		perf_env__insert_bpf_prog_info(env, info_node);
+	}
+
+	return 0;
+out:
+	free(info_linear);
+	free(info_node);
+	up_write(&env->bpf_info_lock);
+	return err;
+}
+
 struct feature_ops {
 	int (*write)(struct feat_fd *ff, struct perf_evlist *evlist);
 	void (*print)(struct feat_fd *ff, FILE *fp);
@@ -2651,7 +2782,8 @@ static const struct feature_ops feat_ops[HEADER_LAST_FEATURE] = {
 	FEAT_OPN(CACHE,		cache,		true),
 	FEAT_OPR(SAMPLE_TIME,	sample_time,	false),
 	FEAT_OPR(MEM_TOPOLOGY,	mem_topology,	true),
-	FEAT_OPR(CLOCKID,       clockid,        false)
+	FEAT_OPR(CLOCKID,       clockid,        false),
+	FEAT_OPR(BPF_PROG_INFO, bpf_prog_info,  false)
 };
 
 struct header_print_data {
diff --git a/tools/perf/util/header.h b/tools/perf/util/header.h
index 0d553ddca0a3..0785c91b4c3a 100644
--- a/tools/perf/util/header.h
+++ b/tools/perf/util/header.h
@@ -39,6 +39,7 @@ enum {
 	HEADER_SAMPLE_TIME,
 	HEADER_MEM_TOPOLOGY,
 	HEADER_CLOCKID,
+	HEADER_BPF_PROG_INFO,
 	HEADER_LAST_FEATURE,
 	HEADER_FEAT_BITS	= 256,
 };
-- 
2.17.1

[PATCH v2 perf,bpf 07/11] perf, bpf: save btf in a rbtree in perf_env

[Song Liu]

btf contains information necessary to annotate bpf programs. This patch
saves btf for bpf programs loaded in the system.

Signed-off-by: Song Liu <songliubraving@fb.com>
---
 tools/perf/util/bpf-event.c | 22 +++++++++++++
 tools/perf/util/bpf-event.h |  7 ++++
 tools/perf/util/env.c       | 65 +++++++++++++++++++++++++++++++++++++
 tools/perf/util/env.h       |  3 ++
 4 files changed, 97 insertions(+)

diff --git a/tools/perf/util/bpf-event.c b/tools/perf/util/bpf-event.c
index ead599bc4f4e..37a5b8134e00 100644
--- a/tools/perf/util/bpf-event.c
+++ b/tools/perf/util/bpf-event.c
@@ -30,6 +30,27 @@ int machine__process_bpf_event(struct machine *machine __maybe_unused,
 	return 0;
 }
 
+static int perf_fetch_btf(struct perf_env *env, u32 btf_id, struct btf *btf)
+{
+	struct btf_node *node;
+	u32 data_size;
+	const void *data;
+
+	data = btf__get_raw_data(btf, &data_size);
+
+	node = malloc(data_size + sizeof(struct btf_node));
+
+	if (!node)
+		return -1;
+
+	node->id = btf_id;
+	node->data_size = data_size;
+	memcpy(node->data, data, data_size);
+
+	perf_env__insert_btf(env, node);
+	return 0;
+}
+
 /*
  * Synthesize PERF_RECORD_KSYMBOL and PERF_RECORD_BPF_EVENT for one bpf
  * program. One PERF_RECORD_BPF_EVENT is generated for the program. And
@@ -109,6 +130,7 @@ static int perf_event__synthesize_one_bpf_prog(struct perf_session *session,
 			goto out;
 		}
 		has_btf = true;
+		perf_fetch_btf(env, info->btf_id, btf);
 	}
 
 	/* Synthesize PERF_RECORD_KSYMBOL */
diff --git a/tools/perf/util/bpf-event.h b/tools/perf/util/bpf-event.h
index 11e6730b6105..60ce24e4e5c6 100644
--- a/tools/perf/util/bpf-event.h
+++ b/tools/perf/util/bpf-event.h
@@ -20,6 +20,13 @@ struct bpf_prog_info_node {
 	struct rb_node			rb_node;
 };
 
+struct btf_node {
+	struct rb_node	rb_node;
+	u32		id;
+	u32		data_size;
+	char		data[];
+};
+
 #ifdef HAVE_LIBBPF_SUPPORT
 int machine__process_bpf_event(struct machine *machine, union perf_event *event,
 			       struct perf_sample *sample);
diff --git a/tools/perf/util/env.c b/tools/perf/util/env.c
index 665b6fe3c7b2..6f9e3d4b94bc 100644
--- a/tools/perf/util/env.c
+++ b/tools/perf/util/env.c
@@ -61,6 +61,57 @@ struct bpf_prog_info_node *perf_env__find_bpf_prog_info(struct perf_env *env,
 	return node;
 }
 
+void perf_env__insert_btf(struct perf_env *env, struct btf_node *btf_node)
+{
+	struct rb_node *parent = NULL;
+	__u32 btf_id = btf_node->id;
+	struct btf_node *node;
+	struct rb_node **p;
+
+	down_write(&env->bpf_info_lock);
+	p = &env->btfs.rb_node;
+
+	while (*p != NULL) {
+		parent = *p;
+		node = rb_entry(parent, struct btf_node, rb_node);
+		if (btf_id < node->id) {
+			p = &(*p)->rb_left;
+		} else if (btf_id > node->id) {
+			p = &(*p)->rb_right;
+		} else {
+			pr_debug("duplicated btf %u\n", btf_id);
+			up_write(&env->bpf_info_lock);
+			return;
+		}
+	}
+
+	rb_link_node(&btf_node->rb_node, parent, p);
+	rb_insert_color(&btf_node->rb_node, &env->btfs);
+	up_write(&env->bpf_info_lock);
+}
+
+struct btf_node *perf_env__find_btf(struct perf_env *env, __u32 btf_id)
+{
+	struct btf_node *node = NULL;
+	struct rb_node *n;
+
+	down_read(&env->bpf_info_lock);
+	n = env->btfs.rb_node;
+
+	while (n) {
+		node = rb_entry(n, struct btf_node, rb_node);
+		if (btf_id < node->id)
+			n = n->rb_left;
+		else if (btf_id > node->id)
+			n = n->rb_right;
+		else
+			break;
+	}
+
+	up_read(&env->bpf_info_lock);
+	return node;
+}
+
 /* purge data in bpf_prog_infos tree */
 static void purge_bpf_info(struct perf_env *env)
 {
@@ -80,6 +131,19 @@ static void purge_bpf_info(struct perf_env *env)
 		rb_erase_init(&node->rb_node, root);
 		free(node);
 	}
+
+	root = &env->btfs;
+	next = rb_first(root);
+
+	while (next) {
+		struct btf_node *node;
+
+		node = rb_entry(next, struct btf_node, rb_node);
+		next = rb_next(&node->rb_node);
+		rb_erase_init(&node->rb_node, root);
+		free(node);
+	}
+
 	up_write(&env->bpf_info_lock);
 }
 
@@ -117,6 +181,7 @@ void perf_env__exit(struct perf_env *env)
 static void init_bpf_rb_trees(struct perf_env *env)
 {
 	env->bpf_prog_infos = RB_ROOT;
+	env->btfs = RB_ROOT;
 	init_rwsem(&env->bpf_info_lock);
 }
 
diff --git a/tools/perf/util/env.h b/tools/perf/util/env.h
index a6d25e91bc98..f0c74255423c 100644
--- a/tools/perf/util/env.h
+++ b/tools/perf/util/env.h
@@ -74,6 +74,7 @@ struct perf_env {
 	 */
 	struct rw_semaphore	bpf_info_lock;
 	struct rb_root		bpf_prog_infos;
+	struct rb_root		btfs;
 };
 
 extern struct perf_env perf_env;
@@ -94,4 +95,6 @@ void perf_env__insert_bpf_prog_info(struct perf_env *env,
 				    struct bpf_prog_info_node *info_node);
 struct bpf_prog_info_node *perf_env__find_bpf_prog_info(struct perf_env *env,
 							__u32 prog_id);
+void perf_env__insert_btf(struct perf_env *env, struct btf_node *btf_node);
+struct btf_node *perf_env__find_btf(struct perf_env *env, __u32 btf_id);
 #endif /* __PERF_ENV_H */
-- 
2.17.1

[PATCH v2 perf,bpf 08/11] perf, bpf: save btf information as headers to perf.data

[Song Liu]

This patch enables perf-record to save btf information as headers to
perf.data A new header type HEADER_BTF is introduced for this data.

Signed-off-by: Song Liu <songliubraving@fb.com>
---
 tools/perf/util/header.c | 99 +++++++++++++++++++++++++++++++++++++++-
 tools/perf/util/header.h |  1 +
 2 files changed, 99 insertions(+), 1 deletion(-)

diff --git a/tools/perf/util/header.c b/tools/perf/util/header.c
index 2ae76a9d06f6..3f1562afe8e5 100644
--- a/tools/perf/util/header.c
+++ b/tools/perf/util/header.c
@@ -1125,6 +1125,45 @@ static int write_bpf_prog_info(struct feat_fd *ff,
 	return ret;
 }
 
+static int write_btf(struct feat_fd *ff,
+		     struct perf_evlist *evlist __maybe_unused)
+{
+	struct perf_env *env = &ff->ph->env;
+	struct rb_root *root;
+	struct rb_node *next;
+	u32 count = 0;
+	int ret;
+
+	down_read(&env->bpf_info_lock);
+
+	root = &env->btfs;
+	next = rb_first(root);
+	while (next) {
+		++count;
+		next = rb_next(next);
+	}
+
+	ret = do_write(ff, &count, sizeof(count));
+
+	if (ret < 0)
+		goto out;
+
+	next = rb_first(root);
+	while (next) {
+		struct btf_node *node;
+
+		node = rb_entry(next, struct btf_node, rb_node);
+		next = rb_next(&node->rb_node);
+		ret = do_write(ff, node,
+			       sizeof(struct btf_node) + node->data_size);
+		if (ret < 0)
+			goto out;
+	}
+out:
+	up_read(&env->bpf_info_lock);
+	return ret;
+}
+
 static int cpu_cache_level__sort(const void *a, const void *b)
 {
 	struct cpu_cache_level *cache_a = (struct cpu_cache_level *)a;
@@ -1628,6 +1667,28 @@ static void print_bpf_prog_info(struct feat_fd *ff, FILE *fp)
 	up_read(&env->bpf_info_lock);
 }
 
+static void print_btf(struct feat_fd *ff, FILE *fp)
+{
+	struct perf_env *env = &ff->ph->env;
+	struct rb_root *root;
+	struct rb_node *next;
+
+	down_read(&env->bpf_info_lock);
+
+	root = &env->btfs;
+	next = rb_first(root);
+
+	while (next) {
+		struct btf_node *node;
+
+		node = rb_entry(next, struct btf_node, rb_node);
+		next = rb_next(&node->rb_node);
+		fprintf(fp, "# bpf_prog_info of id %u\n", node->id);
+	}
+
+	up_read(&env->bpf_info_lock);
+}
+
 static void free_event_desc(struct perf_evsel *events)
 {
 	struct perf_evsel *evsel;
@@ -2723,6 +2784,41 @@ static int process_bpf_prog_info(struct feat_fd *ff,
 	return err;
 }
 
+static int process_btf(struct feat_fd *ff, void *data __maybe_unused)
+{
+	struct perf_env *env = &ff->ph->env;
+	u32 count, i;
+
+	if (do_read_u32(ff, &count))
+		return -1;
+
+	down_write(&env->bpf_info_lock);
+
+	for (i = 0; i < count; ++i) {
+		struct btf_node btf_node;
+		struct btf_node *node;
+
+		if (__do_read(ff, &btf_node, sizeof(struct btf_node)))
+			return -1;
+
+		node = malloc(sizeof(struct btf_node) + btf_node.data_size);
+		if (!node)
+			return -1;
+
+		node->id = btf_node.id;
+		node->data_size = btf_node.data_size;
+
+		if (__do_read(ff, node->data, btf_node.data_size)) {
+			free(node);
+			return -1;
+		}
+		perf_env__insert_btf(env, node);
+	}
+
+	up_write(&env->bpf_info_lock);
+	return 0;
+}
+
 struct feature_ops {
 	int (*write)(struct feat_fd *ff, struct perf_evlist *evlist);
 	void (*print)(struct feat_fd *ff, FILE *fp);
@@ -2783,7 +2879,8 @@ static const struct feature_ops feat_ops[HEADER_LAST_FEATURE] = {
 	FEAT_OPR(SAMPLE_TIME,	sample_time,	false),
 	FEAT_OPR(MEM_TOPOLOGY,	mem_topology,	true),
 	FEAT_OPR(CLOCKID,       clockid,        false),
-	FEAT_OPR(BPF_PROG_INFO, bpf_prog_info,  false)
+	FEAT_OPR(BPF_PROG_INFO, bpf_prog_info,  false),
+	FEAT_OPR(BTF,           btf,            false)
 };
 
 struct header_print_data {
diff --git a/tools/perf/util/header.h b/tools/perf/util/header.h
index 0785c91b4c3a..ba51d8e43c53 100644
--- a/tools/perf/util/header.h
+++ b/tools/perf/util/header.h
@@ -40,6 +40,7 @@ enum {
 	HEADER_MEM_TOPOLOGY,
 	HEADER_CLOCKID,
 	HEADER_BPF_PROG_INFO,
+	HEADER_BTF,
 	HEADER_LAST_FEATURE,
 	HEADER_FEAT_BITS	= 256,
 };
-- 
2.17.1

[PATCH v2 perf,bpf 09/11] perf-top: add option --bpf-event

[Song Liu]

bpf events are only tracked when opts->bpf_event is enabled. This patch
adds command line flag to enable this for perf-top.

Signed-off-by: Song Liu <songliubraving@fb.com>
---
 tools/perf/builtin-top.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/tools/perf/builtin-top.c b/tools/perf/builtin-top.c
index 27d8d42e0a4d..5271d7211b9c 100644
--- a/tools/perf/builtin-top.c
+++ b/tools/perf/builtin-top.c
@@ -1492,6 +1492,7 @@ int cmd_top(int argc, const char **argv)
 		    "Display raw encoding of assembly instructions (default)"),
 	OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel,
 		    "Enable kernel symbol demangling"),
+	OPT_BOOLEAN(0, "bpf-event", &opts->bpf_event, "record bpf events"),
 	OPT_STRING(0, "objdump", &top.annotation_opts.objdump_path, "path",
 		    "objdump binary to use for disassembly and annotations"),
 	OPT_STRING('M', "disassembler-style", &top.annotation_opts.disassembler_style, "disassembler style",
-- 
2.17.1

[PATCH v2 perf,bpf 10/11] perf, bpf: enable annotation of bpf program

[Song Liu]

This patch enables the annotation of bpf program.

A new dso type DSO_BINARY_TYPE__BPF_PROG_INFO is introduced to for BPF
programs. In symbol__disassemble(), DSO_BINARY_TYPE__BPF_PROG_INFO dso
calls into a new function symbol__disassemble_bpf(), where annotation
line information is filled based bpf_prog_info and btf saved in given
perf_env.

symbol__disassemble_bpf() uses libbfd to disassemble bpf programs.

Signed-off-by: Song Liu <songliubraving@fb.com>
---
 tools/perf/Makefile.config  |   6 +-
 tools/perf/util/annotate.c  | 149 +++++++++++++++++++++++++++++++++++-
 tools/perf/util/bpf-event.c |  48 ++++++++++++
 tools/perf/util/bpf-event.h |   4 +
 tools/perf/util/dso.c       |   1 +
 tools/perf/util/dso.h       |  33 +++++---
 tools/perf/util/symbol.c    |   1 +
 7 files changed, 229 insertions(+), 13 deletions(-)

diff --git a/tools/perf/Makefile.config b/tools/perf/Makefile.config
index b441c88cafa1..ab223239f1fb 100644
--- a/tools/perf/Makefile.config
+++ b/tools/perf/Makefile.config
@@ -701,7 +701,7 @@ else
 endif
 
 ifeq ($(feature-libbfd), 1)
-  EXTLIBS += -lbfd
+  EXTLIBS += -lbfd -lopcodes
 else
   # we are on a system that requires -liberty and (maybe) -lz
   # to link against -lbfd; test each case individually here
@@ -796,6 +796,10 @@ ifdef HAVE_KVM_STAT_SUPPORT
     CFLAGS += -DHAVE_KVM_STAT_SUPPORT
 endif
 
+ifeq ($(feature-disassembler-four-args), 1)
+    CFLAGS += -DDISASM_FOUR_ARGS_SIGNATURE
+endif
+
 ifeq (${IS_64_BIT}, 1)
   ifndef NO_PERF_READ_VDSO32
     $(call feature_check,compile-32)
diff --git a/tools/perf/util/annotate.c b/tools/perf/util/annotate.c
index 70de8f6b3aee..078017d31ca9 100644
--- a/tools/perf/util/annotate.c
+++ b/tools/perf/util/annotate.c
@@ -22,6 +22,7 @@
 #include "annotate.h"
 #include "evsel.h"
 #include "evlist.h"
+#include "bpf-event.h"
 #include "block-range.h"
 #include "string2.h"
 #include "arch/common.h"
@@ -29,6 +30,9 @@
 #include <pthread.h>
 #include <linux/bitops.h>
 #include <linux/kernel.h>
+#include <bfd.h>
+#include <dis-asm.h>
+#include <bpf/libbpf.h>
 
 /* FIXME: For the HE_COLORSET */
 #include "ui/browser.h"
@@ -1672,6 +1676,147 @@ static int dso__disassemble_filename(struct dso *dso, char *filename, size_t fil
 	return 0;
 }
 
+static void get_exec_path(char *tpath, size_t size)
+{
+	const char *path = "/proc/self/exe";
+	ssize_t len;
+
+	len = readlink(path, tpath, size - 1);
+	assert(len > 0);
+	tpath[len] = 0;
+}
+
+static int symbol__disassemble_bpf(struct symbol *sym,
+				   struct annotate_args *args)
+{
+	struct annotation *notes = symbol__annotation(sym);
+	struct annotation_options *opts = args->options;
+	struct bpf_prog_info_linear *info_linear;
+	struct bpf_prog_linfo *prog_linfo = NULL;
+	struct bpf_prog_info_node *info_node;
+	int len = sym->end - sym->start;
+	disassembler_ftype disassemble;
+	struct map *map = args->ms.map;
+	struct disassemble_info info;
+	struct dso *dso = map->dso;
+	int pc = 0, count, sub_id;
+	struct btf *btf = NULL;
+	char tpath[PATH_MAX];
+	size_t buf_size;
+	int nr_skip = 0;
+	__u64 arrays;
+	char *buf;
+	bfd *bfdf;
+	FILE *s;
+
+	if (dso->binary_type != DSO_BINARY_TYPE__BPF_PROG_INFO)
+		return -1;
+
+	pr_debug("%s: handling sym %s addr %lx len %lx\n", __func__,
+		 sym->name, sym->start, sym->end - sym->start);
+
+	memset(tpath, 0, sizeof(tpath));
+	get_exec_path(tpath, sizeof(tpath));
+
+	bfdf = bfd_openr(tpath, NULL);
+	assert(bfdf);
+	assert(bfd_check_format(bfdf, bfd_object));
+
+	s = open_memstream(&buf, &buf_size);
+	init_disassemble_info(&info, s,
+			      (fprintf_ftype) fprintf);
+
+	info.arch = bfd_get_arch(bfdf);
+	info.mach = bfd_get_mach(bfdf);
+
+	arrays = 1UL << BPF_PROG_INFO_JITED_INSNS;
+	arrays |= 1UL << BPF_PROG_INFO_JITED_KSYMS;
+	arrays |= 1UL << BPF_PROG_INFO_LINE_INFO;
+	arrays |= 1UL << BPF_PROG_INFO_FUNC_INFO;
+
+	info_node = perf_env__find_bpf_prog_info(dso->bpf_prog.env,
+						 dso->bpf_prog.id);
+	if (!info_node)
+		return -1;
+	info_linear = info_node->info_linear;
+	sub_id = dso->bpf_prog.sub_id;
+
+	info.buffer = (void *)(info_linear->info.jited_prog_insns);
+	info.buffer_length = info_linear->info.jited_prog_len;
+
+	if (info_linear->info.nr_line_info)
+		prog_linfo = bpf_prog_linfo__new(&info_linear->info);
+	prog_linfo = prog_linfo;
+
+	if (info_linear->info.btf_id) {
+		struct btf_node *node;
+
+		node = perf_env__find_btf(dso->bpf_prog.env,
+					  info_linear->info.btf_id);
+		if (node)
+			btf = btf__new((__u8 *)(node->data),
+				       node->data_size);
+	}
+
+	disassemble_init_for_target(&info);
+
+#ifdef DISASM_FOUR_ARGS_SIGNATURE
+	disassemble = disassembler(info.arch,
+				   bfd_big_endian(bfdf),
+				   info.mach,
+				   bfdf);
+#else
+	disassemble = disassembler(bfdf);
+#endif
+	assert(disassemble);
+
+	fflush(s);
+	do {
+		const struct bpf_line_info *linfo = NULL;
+		struct disasm_line *dl;
+		size_t prev_buf_size;
+		const char *srcline;
+		u64 addr;
+
+		addr = pc + ((u64 *)(info_linear->info.jited_ksyms))[sub_id];
+		count = disassemble(pc, &info);
+
+		linfo = bpf_prog_linfo__lfind_addr_func(prog_linfo, addr, sub_id,
+							nr_skip);
+
+		if (linfo) {
+			srcline = btf__name_by_offset(btf, linfo->line_off);
+			nr_skip++;
+		} else
+			srcline = NULL;
+
+		fprintf(s, "\n");
+		prev_buf_size = buf_size;
+		fflush(s);
+
+		if (!opts->hide_src_code && srcline) {
+			args->offset = -1;
+			args->line = strdup(srcline);
+			args->line_nr = 0;
+			args->ms.sym  = sym;
+			dl = disasm_line__new(args);
+			annotation_line__add(&dl->al, &notes->src->source);
+		}
+
+		args->offset = pc;
+		args->line = buf + prev_buf_size;
+		args->line_nr = 0;
+		args->ms.sym  = sym;
+		dl = disasm_line__new(args);
+		annotation_line__add(&dl->al, &notes->src->source);
+
+		pc += count;
+	} while (count > 0 && pc < len);
+
+	bfd_close(bfdf);
+	return 0;
+}
+
 static int symbol__disassemble(struct symbol *sym, struct annotate_args *args)
 {
 	struct annotation_options *opts = args->options;
@@ -1699,7 +1844,9 @@ static int symbol__disassemble(struct symbol *sym, struct annotate_args *args)
 	pr_debug("annotating [%p] %30s : [%p] %30s\n",
 		 dso, dso->long_name, sym, sym->name);
 
-	if (dso__is_kcore(dso)) {
+	if (dso->binary_type == DSO_BINARY_TYPE__BPF_PROG_INFO) {
+		return symbol__disassemble_bpf(sym, args);
+	} else if (dso__is_kcore(dso)) {
 		kce.kcore_filename = symfs_filename;
 		kce.addr = map__rip_2objdump(map, sym->start);
 		kce.offs = sym->start;
diff --git a/tools/perf/util/bpf-event.c b/tools/perf/util/bpf-event.c
index 37a5b8134e00..4f347d61ed96 100644
--- a/tools/perf/util/bpf-event.c
+++ b/tools/perf/util/bpf-event.c
@@ -21,12 +21,60 @@ static int snprintf_hex(char *buf, size_t size, unsigned char *data, size_t len)
 	return ret;
 }
 
+static int machine__process_bpf_event_load(struct machine *machine __maybe_unused,
+					   union perf_event *event,
+					   struct perf_sample *sample __maybe_unused)
+{
+	struct bpf_prog_info_linear *info_linear;
+	struct bpf_prog_info_node *info_node;
+	struct perf_env *env = machine->env;
+	int id = event->bpf_event.id;
+	unsigned int i;
+
+	/* perf-record, no need to handle bpf-event */
+	if (env == NULL)
+		return 0;
+
+	info_node = perf_env__find_bpf_prog_info(env, id);
+	if (!info_node)
+		return 0;
+	info_linear = info_node->info_linear;
+
+	for (i = 0; i < info_linear->info.nr_jited_ksyms; i++) {
+		u64 *addrs = (u64 *)(info_linear->info.jited_ksyms);
+		u64 addr = addrs[i];
+		struct map *map;
+
+		map = map_groups__find(&machine->kmaps, addr);
+
+		if (map) {
+			map->dso->binary_type = DSO_BINARY_TYPE__BPF_PROG_INFO;
+			map->dso->bpf_prog.id = id;
+			map->dso->bpf_prog.sub_id = i;
+			map->dso->bpf_prog.env = env;
+		}
+	}
+	return 0;
+}
+
 int machine__process_bpf_event(struct machine *machine __maybe_unused,
 			       union perf_event *event,
 			       struct perf_sample *sample __maybe_unused)
 {
 	if (dump_trace)
 		perf_event__fprintf_bpf_event(event, stdout);
+
+	switch (event->bpf_event.type) {
+	case PERF_BPF_EVENT_PROG_LOAD:
+		return machine__process_bpf_event_load(machine, event, sample);
+
+	case PERF_BPF_EVENT_PROG_UNLOAD:
+		break;
+	default:
+		pr_debug("unexpected bpf_event type of %d\n",
+			 event->bpf_event.type);
+		break;
+	}
 	return 0;
 }
 
diff --git a/tools/perf/util/bpf-event.h b/tools/perf/util/bpf-event.h
index 60ce24e4e5c6..c4f0f1395ea5 100644
--- a/tools/perf/util/bpf-event.h
+++ b/tools/perf/util/bpf-event.h
@@ -2,6 +2,10 @@
 #ifndef __PERF_BPF_EVENT_H
 #define __PERF_BPF_EVENT_H
 
+#include <bpf/bpf.h>
+#include <bpf/btf.h>
+#include <bpf/libbpf.h>
+#include <linux/btf.h>
 #include <linux/compiler.h>
 #include <bpf/bpf.h>
 #include <bpf/btf.h>
diff --git a/tools/perf/util/dso.c b/tools/perf/util/dso.c
index 62c8cf622607..1798192bf0f9 100644
--- a/tools/perf/util/dso.c
+++ b/tools/perf/util/dso.c
@@ -181,6 +181,7 @@ int dso__read_binary_type_filename(const struct dso *dso,
 	case DSO_BINARY_TYPE__KALLSYMS:
 	case DSO_BINARY_TYPE__GUEST_KALLSYMS:
 	case DSO_BINARY_TYPE__JAVA_JIT:
+	case DSO_BINARY_TYPE__BPF_PROG_INFO:
 	case DSO_BINARY_TYPE__NOT_FOUND:
 		ret = -1;
 		break;
diff --git a/tools/perf/util/dso.h b/tools/perf/util/dso.h
index 8c8a7abe809d..f20d319463f1 100644
--- a/tools/perf/util/dso.h
+++ b/tools/perf/util/dso.h
@@ -14,6 +14,8 @@
 #include "namespaces.h"
 #include "build-id.h"
 
+struct perf_env;
+
 enum dso_binary_type {
 	DSO_BINARY_TYPE__KALLSYMS = 0,
 	DSO_BINARY_TYPE__GUEST_KALLSYMS,
@@ -34,6 +36,7 @@ enum dso_binary_type {
 	DSO_BINARY_TYPE__KCORE,
 	DSO_BINARY_TYPE__GUEST_KCORE,
 	DSO_BINARY_TYPE__OPENEMBEDDED_DEBUGINFO,
+	DSO_BINARY_TYPE__BPF_PROG_INFO,
 	DSO_BINARY_TYPE__NOT_FOUND,
 };
 
@@ -177,17 +180,25 @@ struct dso {
 	struct auxtrace_cache *auxtrace_cache;
 	int		 comp;
 
-	/* dso data file */
-	struct {
-		struct rb_root	 cache;
-		int		 fd;
-		int		 status;
-		u32		 status_seen;
-		size_t		 file_size;
-		struct list_head open_entry;
-		u64		 debug_frame_offset;
-		u64		 eh_frame_hdr_offset;
-	} data;
+	union {
+		/* dso data file */
+		struct {
+			struct rb_root	 cache;
+			int		 fd;
+			int		 status;
+			u32		 status_seen;
+			size_t		 file_size;
+			struct list_head open_entry;
+			u64		 debug_frame_offset;
+			u64		 eh_frame_hdr_offset;
+		} data;
+		/* bpf prog information */
+		struct {
+			u32		id;
+			u32		sub_id;
+			struct perf_env	*env;
+		} bpf_prog;
+	};
 
 	union { /* Tool specific area */
 		void	 *priv;
diff --git a/tools/perf/util/symbol.c b/tools/perf/util/symbol.c
index 48efad6d0f90..33ae59e89da2 100644
--- a/tools/perf/util/symbol.c
+++ b/tools/perf/util/symbol.c
@@ -1441,6 +1441,7 @@ static bool dso__is_compatible_symtab_type(struct dso *dso, bool kmod,
 	case DSO_BINARY_TYPE__BUILD_ID_CACHE_DEBUGINFO:
 		return true;
 
+	case DSO_BINARY_TYPE__BPF_PROG_INFO:
 	case DSO_BINARY_TYPE__NOT_FOUND:
 	default:
 		return false;
-- 
2.17.1

[PATCH v2 perf,bpf 11/11] perf, bpf: save information about short living bpf programs

[Song Liu]

To annotate bpf programs in perf, it is necessary to save information in
bpf_prog_info and btf. For short living bpf program, it is necessary to
save these information before it is unloaded.

This patch saves these information in a separate thread. This thread
creates its own evlist, that only tracks bpf events. This evlists uses
ring buffer with very low watermark for lower latency. When bpf load
events are received, this thread tries to gather information via sys_bpf
and save it in perf_env.

Signed-off-by: Song Liu <songliubraving@fb.com>
---
 tools/perf/builtin-record.c |  13 ++++
 tools/perf/builtin-top.c    |  12 ++++
 tools/perf/util/bpf-event.c | 129 ++++++++++++++++++++++++++++++++++++
 tools/perf/util/bpf-event.h |  22 ++++++
 tools/perf/util/evlist.c    |  20 ++++++
 tools/perf/util/evlist.h    |   2 +
 6 files changed, 198 insertions(+)

diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index 2355e0a9eda0..46abb44aaaab 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -1106,6 +1106,8 @@ static int __cmd_record(struct record *rec, int argc, const char **argv)
 	struct perf_data *data = &rec->data;
 	struct perf_session *session;
 	bool disabled = false, draining = false;
+	struct bpf_event_poll_args poll_args;
+	bool bpf_thread_running = false;
 	int fd;
 
 	atexit(record__sig_exit);
@@ -1206,6 +1208,14 @@ static int __cmd_record(struct record *rec, int argc, const char **argv)
 		goto out_child;
 	}
 
+	if (rec->opts.bpf_event) {
+		poll_args.env = &session->header.env;
+		poll_args.target = &rec->opts.target;
+		poll_args.done = &done;
+		if (bpf_event__start_polling_thread(&poll_args) == 0)
+			bpf_thread_running = true;
+	}
+
 	err = record__synthesize(rec, false);
 	if (err < 0)
 		goto out_child;
@@ -1456,6 +1466,9 @@ static int __cmd_record(struct record *rec, int argc, const char **argv)
 
 out_delete_session:
 	perf_session__delete(session);
+
+	if (bpf_thread_running)
+		bpf_event__stop_polling_thread(&poll_args);
 	return status;
 }
 
diff --git a/tools/perf/builtin-top.c b/tools/perf/builtin-top.c
index 5271d7211b9c..2586ee081967 100644
--- a/tools/perf/builtin-top.c
+++ b/tools/perf/builtin-top.c
@@ -1524,10 +1524,12 @@ int cmd_top(int argc, const char **argv)
 			"number of thread to run event synthesize"),
 	OPT_END()
 	};
+	struct bpf_event_poll_args poll_args;
 	const char * const top_usage[] = {
 		"perf top [<options>]",
 		NULL
 	};
+	bool bpf_thread_running = false;
 	int status = hists__init();
 
 	if (status < 0)
@@ -1652,8 +1654,18 @@ int cmd_top(int argc, const char **argv)
 		signal(SIGWINCH, winch_sig);
 	}
 
+	if (top.record_opts.bpf_event) {
+		poll_args.env = &perf_env;
+		poll_args.target = target;
+		poll_args.done = &done;
+		if (bpf_event__start_polling_thread(&poll_args) == 0)
+			bpf_thread_running = true;
+	}
 	status = __cmd_top(&top);
 
+	if (bpf_thread_running)
+		bpf_event__stop_polling_thread(&poll_args);
+
 out_delete_evlist:
 	perf_evlist__delete(top.evlist);
 
diff --git a/tools/perf/util/bpf-event.c b/tools/perf/util/bpf-event.c
index 4f347d61ed96..0caf137c515b 100644
--- a/tools/perf/util/bpf-event.c
+++ b/tools/perf/util/bpf-event.c
@@ -8,6 +8,7 @@
 #include "machine.h"
 #include "env.h"
 #include "session.h"
+#include "evlist.h"
 
 #define ptr_to_u64(ptr)    ((__u64)(unsigned long)(ptr))
 
@@ -316,3 +317,131 @@ int perf_event__synthesize_bpf_events(struct perf_session *session,
 	free(event);
 	return err;
 }
+
+static void perf_env_add_bpf_info(struct perf_env *env, u32 id)
+{
+	struct bpf_prog_info_linear *info_linear;
+	struct bpf_prog_info_node *info_node;
+	struct btf *btf = NULL;
+	u64 arrays;
+	u32 btf_id;
+	int fd;
+
+	fd = bpf_prog_get_fd_by_id(id);
+	if (fd < 0)
+		return;
+
+	arrays = 1UL << BPF_PROG_INFO_JITED_KSYMS;
+	arrays |= 1UL << BPF_PROG_INFO_JITED_FUNC_LENS;
+	arrays |= 1UL << BPF_PROG_INFO_FUNC_INFO;
+	arrays |= 1UL << BPF_PROG_INFO_PROG_TAGS;
+	arrays |= 1UL << BPF_PROG_INFO_JITED_INSNS;
+	arrays |= 1UL << BPF_PROG_INFO_LINE_INFO;
+	arrays |= 1UL << BPF_PROG_INFO_JITED_LINE_INFO;
+
+	info_linear = bpf_program__get_prog_info_linear(fd, arrays);
+	if (IS_ERR_OR_NULL(info_linear)) {
+		pr_debug("%s: failed to get BPF program info. aborting\n", __func__);
+		goto out;
+	}
+
+	btf_id = info_linear->info.btf_id;
+
+	info_node = malloc(sizeof(struct bpf_prog_info_node));
+	if (info_node) {
+		info_node->info_linear = info_linear;
+		perf_env__insert_bpf_prog_info(env, info_node);
+	} else
+		free(info_linear);
+
+	if (btf_id == 0)
+		goto out;
+
+	if (btf__get_from_id(btf_id, &btf)) {
+		pr_debug("%s: failed to get BTF of id %u, aborting\n",
+			 __func__, btf_id);
+		goto out;
+	}
+	perf_fetch_btf(env, btf_id, btf);
+
+out:
+	free(btf);
+	close(fd);
+}
+
+static void *bpf_poll_thread(void *arg)
+{
+	struct bpf_event_poll_args *args = arg;
+	int i;
+
+	while (!*(args->done)) {
+		perf_evlist__poll(args->evlist, 1000);
+
+		for (i = 0; i < args->evlist->nr_mmaps; i++) {
+			struct perf_mmap *map = &args->evlist->mmap[i];
+			union perf_event *event;
+
+			if (perf_mmap__read_init(map))
+				continue;
+			while ((event = perf_mmap__read_event(map)) != NULL) {
+				pr_debug("processing vip event of type %d\n",
+					 event->header.type);
+				switch (event->header.type) {
+				case PERF_RECORD_BPF_EVENT:
+					if (event->bpf_event.type != PERF_BPF_EVENT_PROG_LOAD)
+						break;
+					perf_env_add_bpf_info(args->env, event->bpf_event.id);
+					break;
+				default:
+					break;
+				}
+				perf_mmap__consume(map);
+			}
+			perf_mmap__read_done(map);
+		}
+	}
+	return NULL;
+}
+
+pthread_t poll_thread;
+
+int bpf_event__start_polling_thread(struct bpf_event_poll_args *args)
+{
+	struct perf_evsel *counter;
+
+	args->evlist = perf_evlist__new();
+
+	if (args->evlist == NULL)
+		return -1;
+
+	if (perf_evlist__create_maps(args->evlist, args->target))
+		return -1;
+
+	if (perf_evlist__add_bpf_tracker(args->evlist))
+		return -1;
+
+	evlist__for_each_entry(args->evlist, counter) {
+		if (perf_evsel__open(counter, args->evlist->cpus,
+				     args->evlist->threads) < 0)
+			return -1;
+	}
+
+	if (perf_evlist__mmap(args->evlist, UINT_MAX))
+		return -1;
+
+	evlist__for_each_entry(args->evlist, counter) {
+		if (perf_evsel__enable(counter))
+			return -1;
+	}
+
+	if (pthread_create(&poll_thread, NULL, bpf_poll_thread, args))
+		return -1;
+
+	return 0;
+}
+
+void bpf_event__stop_polling_thread(struct bpf_event_poll_args *args)
+{
+	pthread_join(poll_thread, NULL);
+	perf_evlist__exit(args->evlist);
+}
diff --git a/tools/perf/util/bpf-event.h b/tools/perf/util/bpf-event.h
index c4f0f1395ea5..61914827c1e3 100644
--- a/tools/perf/util/bpf-event.h
+++ b/tools/perf/util/bpf-event.h
@@ -12,12 +12,17 @@
 #include <bpf/libbpf.h>
 #include <linux/btf.h>
 #include <linux/rbtree.h>
+#include <pthread.h>
+#include <api/fd/array.h>
 #include "event.h"
 
 struct machine;
 union perf_event;
+struct perf_env;
 struct perf_sample;
 struct record_opts;
+struct evlist;
+struct target;
 
 struct bpf_prog_info_node {
 	struct bpf_prog_info_linear	*info_linear;
@@ -31,6 +36,13 @@ struct btf_node {
 	char		data[];
 };
 
+struct bpf_event_poll_args {
+	struct perf_env		*env;
+	struct perf_evlist	*evlist;
+	struct target		*target;
+	volatile int		*done;
+};
+
 #ifdef HAVE_LIBBPF_SUPPORT
 int machine__process_bpf_event(struct machine *machine, union perf_event *event,
 			       struct perf_sample *sample);
@@ -39,6 +51,8 @@ int perf_event__synthesize_bpf_events(struct perf_session *session,
 				      perf_event__handler_t process,
 				      struct machine *machine,
 				      struct record_opts *opts);
+int bpf_event__start_polling_thread(struct bpf_event_poll_args *args);
+void bpf_event__stop_polling_thread(struct bpf_event_poll_args *args);
 #else
 static inline int machine__process_bpf_event(struct machine *machine __maybe_unused,
 					     union perf_event *event __maybe_unused,
@@ -54,5 +68,13 @@ static inline int perf_event__synthesize_bpf_events(struct perf_session *session
 {
 	return 0;
 }
+
+static inline int bpf_event__start_polling_thread(struct bpf_event_poll_args *args __maybe_unused)
+{
+	return 0;
+}
+void bpf_event__stop_polling_thread(struct bpf_event_poll_args *args __maybe_unused)
+{
+}
 #endif // HAVE_LIBBPF_SUPPORT
 #endif
diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c
index 8c902276d4b4..612c079579ce 100644
--- a/tools/perf/util/evlist.c
+++ b/tools/perf/util/evlist.c
@@ -271,6 +271,26 @@ int perf_evlist__add_dummy(struct perf_evlist *evlist)
 	return 0;
 }
 
+int perf_evlist__add_bpf_tracker(struct perf_evlist *evlist)
+{
+	struct perf_event_attr attr = {
+		.type	          = PERF_TYPE_SOFTWARE,
+		.config           = PERF_COUNT_SW_DUMMY,
+		.watermark        = 1,
+		.bpf_event        = 1,
+		.wakeup_watermark = 1,
+		.size	   = sizeof(attr), /* to capture ABI version */
+	};
+	struct perf_evsel *evsel = perf_evsel__new_idx(&attr,
+						       evlist->nr_entries);
+
+	if (evsel == NULL)
+		return -ENOMEM;
+
+	perf_evlist__add(evlist, evsel);
+	return 0;
+}
+
 static int perf_evlist__add_attrs(struct perf_evlist *evlist,
 				  struct perf_event_attr *attrs, size_t nr_attrs)
 {
diff --git a/tools/perf/util/evlist.h b/tools/perf/util/evlist.h
index 868294491194..a2d22715188e 100644
--- a/tools/perf/util/evlist.h
+++ b/tools/perf/util/evlist.h
@@ -84,6 +84,8 @@ int __perf_evlist__add_default_attrs(struct perf_evlist *evlist,
 
 int perf_evlist__add_dummy(struct perf_evlist *evlist);
 
+int perf_evlist__add_bpf_tracker(struct perf_evlist *evlist);
+
 int perf_evlist__add_newtp(struct perf_evlist *evlist,
 			   const char *sys, const char *name, void *handler);
 
-- 
2.17.1

Re: [PATCH v2 perf,bpf 01/11] perf, bpf: consider events with attr.bpf_event as side-band events

[Arnaldo Carvalho de Melo]

Em Thu, Feb 14, 2019 at 03:58:40PM -0800, Song Liu escreveu:
> Events with bpf_event should be considered as side-band event, as they
> carry information about BPF programs.
> 
> Fixes: 6ee52e2a3fe4 ("perf, bpf: Introduce PERF_RECORD_BPF_EVENT")
> Signed-off-by: Song Liu <songliubraving@fb.com>
> ---
>  kernel/events/core.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/kernel/events/core.c b/kernel/events/core.c
> index 0a8dab322111..9403bdda5f8c 100644
> --- a/kernel/events/core.c
> +++ b/kernel/events/core.c
> @@ -4238,7 +4238,8 @@ static bool is_sb_event(struct perf_event *event)
>  	if (attr->mmap || attr->mmap_data || attr->mmap2 ||
>  	    attr->comm || attr->comm_exec ||
>  	    attr->task || attr->ksymbol ||
> -	    attr->context_switch)
> +	    attr->context_switch ||
> +	    attr->bpf_event)

What about attr->ksymbol?

- Arnaldo

>  		return true;
>  	return false;
>  }
> -- 
> 2.17.1

Re: [PATCH v2 perf,bpf 05/11] perf, bpf: save bpf_prog_info in a rbtree in perf_env

[Arnaldo Carvalho de Melo]

Em Thu, Feb 14, 2019 at 04:00:06PM -0800, Song Liu escreveu:
> bpf_prog_info contains information necessary to annotate bpf programs.
> This patch saves bpf_prog_info for bpf programs loaded in the system.
> 
> Signed-off-by: Song Liu <songliubraving@fb.com>
> ---
>  tools/perf/builtin-record.c |  2 +-
>  tools/perf/builtin-top.c    |  2 +-
>  tools/perf/util/bpf-event.c | 39 +++++++++++++----
>  tools/perf/util/bpf-event.h | 15 +++++--
>  tools/perf/util/env.c       | 83 +++++++++++++++++++++++++++++++++++++
>  tools/perf/util/env.h       | 14 +++++++
>  6 files changed, 142 insertions(+), 13 deletions(-)
> 
> diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
> index 88ea11d57c6f..2355e0a9eda0 100644
> --- a/tools/perf/builtin-record.c
> +++ b/tools/perf/builtin-record.c
> @@ -1083,7 +1083,7 @@ static int record__synthesize(struct record *rec, bool tail)
>  		return err;
>  	}
>  
> -	err = perf_event__synthesize_bpf_events(tool, process_synthesized_event,
> +	err = perf_event__synthesize_bpf_events(session, process_synthesized_event,
>  						machine, opts);
>  	if (err < 0)
>  		pr_warning("Couldn't synthesize bpf events.\n");
> diff --git a/tools/perf/builtin-top.c b/tools/perf/builtin-top.c
> index 5a486d4de56e..27d8d42e0a4d 100644
> --- a/tools/perf/builtin-top.c
> +++ b/tools/perf/builtin-top.c
> @@ -1216,7 +1216,7 @@ static int __cmd_top(struct perf_top *top)
>  
>  	init_process_thread(top);
>  
> -	ret = perf_event__synthesize_bpf_events(&top->tool, perf_event__process,
> +	ret = perf_event__synthesize_bpf_events(top->session, perf_event__process,
>  						&top->session->machines.host,
>  						&top->record_opts);
>  	if (ret < 0)
> diff --git a/tools/perf/util/bpf-event.c b/tools/perf/util/bpf-event.c
> index e6dfb95029e5..ead599bc4f4e 100644
> --- a/tools/perf/util/bpf-event.c
> +++ b/tools/perf/util/bpf-event.c
> @@ -1,15 +1,13 @@
>  // SPDX-License-Identifier: GPL-2.0
>  #include <errno.h>
>  #include <stdlib.h>
> -#include <bpf/bpf.h>
> -#include <bpf/btf.h>
> -#include <bpf/libbpf.h>
> -#include <linux/btf.h>

I think you need these here, since in this C file you will use the
definitions for these structs, see further comments below.

>  #include <linux/err.h>
>  #include "bpf-event.h"
>  #include "debug.h"
>  #include "symbol.h"
>  #include "machine.h"
> +#include "env.h"
> +#include "session.h"
>  
>  #define ptr_to_u64(ptr)    ((__u64)(unsigned long)(ptr))
>  
> @@ -42,7 +40,7 @@ int machine__process_bpf_event(struct machine *machine __maybe_unused,
>   *   -1 for failures;
>   *   -2 for lack of kernel support.
>   */
> -static int perf_event__synthesize_one_bpf_prog(struct perf_tool *tool,
> +static int perf_event__synthesize_one_bpf_prog(struct perf_session *session,
>  					       perf_event__handler_t process,
>  					       struct machine *machine,
>  					       int fd,
> @@ -52,17 +50,29 @@ static int perf_event__synthesize_one_bpf_prog(struct perf_tool *tool,
>  	struct ksymbol_event *ksymbol_event = &event->ksymbol_event;
>  	struct bpf_event *bpf_event = &event->bpf_event;
>  	struct bpf_prog_info_linear *info_linear;
> +	struct perf_tool *tool = session->tool;
> +	struct bpf_prog_info_node *info_node;
>  	struct bpf_prog_info *info;
>  	struct btf *btf = NULL;
>  	bool has_btf = false;
> +	struct perf_env *env;
>  	u32 sub_prog_cnt, i;
>  	int err = 0;
>  	u64 arrays;
>  
> +	/*
> +	 * for perf-record and perf-report use header.env;
> +	 * otherwise, use global perf_env.
> +	 */
> +	env = session->data ? &session->header.env : &perf_env;
> +
>  	arrays = 1UL << BPF_PROG_INFO_JITED_KSYMS;
>  	arrays |= 1UL << BPF_PROG_INFO_JITED_FUNC_LENS;
>  	arrays |= 1UL << BPF_PROG_INFO_FUNC_INFO;
>  	arrays |= 1UL << BPF_PROG_INFO_PROG_TAGS;
> +	arrays |= 1UL << BPF_PROG_INFO_JITED_INSNS;
> +	arrays |= 1UL << BPF_PROG_INFO_LINE_INFO;
> +	arrays |= 1UL << BPF_PROG_INFO_JITED_LINE_INFO;
>  
>  	info_linear = bpf_program__get_prog_info_linear(fd, arrays);
>  	if (IS_ERR_OR_NULL(info_linear)) {
> @@ -151,8 +161,8 @@ static int perf_event__synthesize_one_bpf_prog(struct perf_tool *tool,
>  						     machine, process);
>  	}
>  
> -	/* Synthesize PERF_RECORD_BPF_EVENT */
>  	if (opts->bpf_event) {
> +		/* Synthesize PERF_RECORD_BPF_EVENT */
>  		*bpf_event = (struct bpf_event){
>  			.header = {
>  				.type = PERF_RECORD_BPF_EVENT,
> @@ -165,6 +175,19 @@ static int perf_event__synthesize_one_bpf_prog(struct perf_tool *tool,
>  		memcpy(bpf_event->tag, info->tag, BPF_TAG_SIZE);
>  		memset((void *)event + event->header.size, 0, machine->id_hdr_size);
>  		event->header.size += machine->id_hdr_size;
> +
> +		/* save bpf_prog_info to env */
> +		info_node = malloc(sizeof(struct bpf_prog_info_node));
> +		if (info_node) {
> +			info_node->info_linear = info_linear;
> +			perf_env__insert_bpf_prog_info(env, info_node);
> +			info_linear = NULL;
> +		}
> +
> +		/*
> +		 * process after saving bpf_prog_info to env, so that
> +		 * required information is ready for look up
> +		 */
>  		err = perf_tool__process_synth_event(tool, event,
>  						     machine, process);
>  	}
> @@ -175,7 +198,7 @@ static int perf_event__synthesize_one_bpf_prog(struct perf_tool *tool,
>  	return err ? -1 : 0;
>  }
>  
> -int perf_event__synthesize_bpf_events(struct perf_tool *tool,
> +int perf_event__synthesize_bpf_events(struct perf_session *session,
>  				      perf_event__handler_t process,
>  				      struct machine *machine,
>  				      struct record_opts *opts)
> @@ -209,7 +232,7 @@ int perf_event__synthesize_bpf_events(struct perf_tool *tool,
>  			continue;
>  		}
>  
> -		err = perf_event__synthesize_one_bpf_prog(tool, process,
> +		err = perf_event__synthesize_one_bpf_prog(session, process,
>  							  machine, fd,
>  							  event, opts);
>  		close(fd);
> diff --git a/tools/perf/util/bpf-event.h b/tools/perf/util/bpf-event.h
> index 7890067e1a37..11e6730b6105 100644
> --- a/tools/perf/util/bpf-event.h
> +++ b/tools/perf/util/bpf-event.h
> @@ -3,19 +3,28 @@
>  #define __PERF_BPF_EVENT_H
>  
>  #include <linux/compiler.h>
> +#include <bpf/bpf.h>
> +#include <bpf/btf.h>
> +#include <bpf/libbpf.h>
> +#include <linux/btf.h>

Are you sure you'll need all of these headers here? Perhaps just some
forward declarations will do?

In fact the only bpf or btf structure here seems to be
bpf_prog_info_linear, which needs ust a forward declaration.

Avoiding these includes reduces the build time, and since the build-test
target does many builds and I want to build this in many containers, we
should try to reduce the build time by using just what is needed in each
header and C file. Also during development it helps with not rebuilding
tons of things when something unrelated changes in a header, etc.

- Arnaldo

> +#include <linux/rbtree.h>
>  #include "event.h"
>  
>  struct machine;
>  union perf_event;
>  struct perf_sample;
> -struct perf_tool;
>  struct record_opts;
>  
> +struct bpf_prog_info_node {
> +	struct bpf_prog_info_linear	*info_linear;
> +	struct rb_node			rb_node;
> +};
> +
>  #ifdef HAVE_LIBBPF_SUPPORT
>  int machine__process_bpf_event(struct machine *machine, union perf_event *event,
>  			       struct perf_sample *sample);
>  
> -int perf_event__synthesize_bpf_events(struct perf_tool *tool,
> +int perf_event__synthesize_bpf_events(struct perf_session *session,
>  				      perf_event__handler_t process,
>  				      struct machine *machine,
>  				      struct record_opts *opts);
> @@ -27,7 +36,7 @@ static inline int machine__process_bpf_event(struct machine *machine __maybe_unu
>  	return 0;
>  }
>  
> -static inline int perf_event__synthesize_bpf_events(struct perf_tool *tool __maybe_unused,
> +static inline int perf_event__synthesize_bpf_events(struct perf_session *session __maybe_unused,
>  						    perf_event__handler_t process __maybe_unused,
>  						    struct machine *machine __maybe_unused,
>  						    struct record_opts *opts __maybe_unused)
> diff --git a/tools/perf/util/env.c b/tools/perf/util/env.c
> index 4c23779e271a..665b6fe3c7b2 100644
> --- a/tools/perf/util/env.c
> +++ b/tools/perf/util/env.c
> @@ -8,10 +8,86 @@
>  
>  struct perf_env perf_env;
>  
> +void perf_env__insert_bpf_prog_info(struct perf_env *env,
> +				    struct bpf_prog_info_node *info_node)
> +{
> +	__u32 prog_id = info_node->info_linear->info.id;
> +	struct bpf_prog_info_node *node;
> +	struct rb_node *parent = NULL;
> +	struct rb_node **p;
> +
> +	down_write(&env->bpf_info_lock);
> +	p = &env->bpf_prog_infos.rb_node;
> +
> +	while (*p != NULL) {
> +		parent = *p;
> +		node = rb_entry(parent, struct bpf_prog_info_node, rb_node);
> +		if (prog_id < node->info_linear->info.id) {
> +			p = &(*p)->rb_left;
> +		} else if (prog_id > node->info_linear->info.id) {
> +			p = &(*p)->rb_right;
> +		} else {
> +			pr_debug("duplicated bpf prog info %u\n", prog_id);
> +			up_write(&env->bpf_info_lock);
> +			return;
> +		}
> +	}
> +
> +	rb_link_node(&info_node->rb_node, parent, p);
> +	rb_insert_color(&info_node->rb_node, &env->bpf_prog_infos);
> +	up_write(&env->bpf_info_lock);
> +}
> +
> +struct bpf_prog_info_node *perf_env__find_bpf_prog_info(struct perf_env *env,
> +							__u32 prog_id)
> +{
> +	struct bpf_prog_info_node *node = NULL;
> +	struct rb_node *n;
> +
> +	down_read(&env->bpf_info_lock);
> +	n = env->bpf_prog_infos.rb_node;
> +
> +	while (n) {
> +		node = rb_entry(n, struct bpf_prog_info_node, rb_node);
> +		if (prog_id < node->info_linear->info.id)
> +			n = n->rb_left;
> +		else if (prog_id > node->info_linear->info.id)
> +			n = n->rb_right;
> +		else
> +			break;
> +	}
> +
> +	up_read(&env->bpf_info_lock);
> +	return node;
> +}
> +
> +/* purge data in bpf_prog_infos tree */
> +static void purge_bpf_info(struct perf_env *env)
> +{
> +	struct rb_root *root;
> +	struct rb_node *next;
> +
> +	down_write(&env->bpf_info_lock);
> +
> +	root = &env->bpf_prog_infos;
> +	next = rb_first(root);
> +
> +	while (next) {
> +		struct bpf_prog_info_node *node;
> +
> +		node = rb_entry(next, struct bpf_prog_info_node, rb_node);
> +		next = rb_next(&node->rb_node);
> +		rb_erase_init(&node->rb_node, root);
> +		free(node);
> +	}
> +	up_write(&env->bpf_info_lock);
> +}
> +
>  void perf_env__exit(struct perf_env *env)
>  {
>  	int i;
>  
> +	purge_bpf_info(env);
>  	zfree(&env->hostname);
>  	zfree(&env->os_release);
>  	zfree(&env->version);
> @@ -38,6 +114,12 @@ void perf_env__exit(struct perf_env *env)
>  	zfree(&env->memory_nodes);
>  }
>  
> +static void init_bpf_rb_trees(struct perf_env *env)
> +{
> +	env->bpf_prog_infos = RB_ROOT;
> +	init_rwsem(&env->bpf_info_lock);
> +}
> +
>  int perf_env__set_cmdline(struct perf_env *env, int argc, const char *argv[])
>  {
>  	int i;
> @@ -59,6 +141,7 @@ int perf_env__set_cmdline(struct perf_env *env, int argc, const char *argv[])
>  
>  	env->nr_cmdline = argc;
>  
> +	init_bpf_rb_trees(env);
>  	return 0;
>  out_free:
>  	zfree(&env->cmdline_argv);
> diff --git a/tools/perf/util/env.h b/tools/perf/util/env.h
> index d01b8355f4ca..a6d25e91bc98 100644
> --- a/tools/perf/util/env.h
> +++ b/tools/perf/util/env.h
> @@ -3,7 +3,10 @@
>  #define __PERF_ENV_H
>  
>  #include <linux/types.h>
> +#include <linux/rbtree.h>
>  #include "cpumap.h"
> +#include "rwsem.h"

You don't need the following header, just use a forward declaration for
the sole struct you use with a pointer:

struct bpf_prog_info_node;

> +#include "bpf-event.h"
>  
>  struct cpu_topology_map {
>  	int	socket_id;
> @@ -64,6 +67,13 @@ struct perf_env {
>  	struct memory_node	*memory_nodes;
>  	unsigned long long	 memory_bsize;
>  	u64                     clockid_res_ns;
> +
> +	/*
> +	 * bpf_info_lock protects bpf rbtrees. This is needed because the
> +	 * trees are accessed by different threads in perf-top
> +	 */
> +	struct rw_semaphore	bpf_info_lock;
> +	struct rb_root		bpf_prog_infos;

Please group this into a structure, i.e.:

	struct {
		struct rw_semaphore lock;
		struct rb_root	    infos;
	} bpf_progs;

>  };
>  
>  extern struct perf_env perf_env;
> @@ -80,4 +90,8 @@ const char *perf_env__arch(struct perf_env *env);
>  const char *perf_env__raw_arch(struct perf_env *env);
>  int perf_env__nr_cpus_avail(struct perf_env *env);
>  
> +void perf_env__insert_bpf_prog_info(struct perf_env *env,
> +				    struct bpf_prog_info_node *info_node);
> +struct bpf_prog_info_node *perf_env__find_bpf_prog_info(struct perf_env *env,
> +							__u32 prog_id);
>  #endif /* __PERF_ENV_H */
> -- 
> 2.17.1

Re: [PATCH v2 perf,bpf 08/11] perf, bpf: save btf information as headers to perf.data

[Arnaldo Carvalho de Melo]

Em Thu, Feb 14, 2019 at 04:00:09PM -0800, Song Liu escreveu:
> This patch enables perf-record to save btf information as headers to
> perf.data A new header type HEADER_BTF is introduced for this data.

Jiri,

	Wouldn't it be better for this HEADER_BTF to be introduced
already as an user space event, Song, see:

  tools/perf/util/event.h

and:

  tools/perf/util/event.c

perf_event__synthesize_cpu_map()

- Arnaldo

 
> Signed-off-by: Song Liu <songliubraving@fb.com>
> ---
>  tools/perf/util/header.c | 99 +++++++++++++++++++++++++++++++++++++++-
>  tools/perf/util/header.h |  1 +
>  2 files changed, 99 insertions(+), 1 deletion(-)
> 
> diff --git a/tools/perf/util/header.c b/tools/perf/util/header.c
> index 2ae76a9d06f6..3f1562afe8e5 100644
> --- a/tools/perf/util/header.c
> +++ b/tools/perf/util/header.c
> @@ -1125,6 +1125,45 @@ static int write_bpf_prog_info(struct feat_fd *ff,
>  	return ret;
>  }
>  
> +static int write_btf(struct feat_fd *ff,
> +		     struct perf_evlist *evlist __maybe_unused)
> +{
> +	struct perf_env *env = &ff->ph->env;
> +	struct rb_root *root;
> +	struct rb_node *next;
> +	u32 count = 0;
> +	int ret;
> +
> +	down_read(&env->bpf_info_lock);
> +
> +	root = &env->btfs;
> +	next = rb_first(root);
> +	while (next) {
> +		++count;
> +		next = rb_next(next);
> +	}
> +
> +	ret = do_write(ff, &count, sizeof(count));
> +
> +	if (ret < 0)
> +		goto out;
> +
> +	next = rb_first(root);
> +	while (next) {
> +		struct btf_node *node;
> +
> +		node = rb_entry(next, struct btf_node, rb_node);
> +		next = rb_next(&node->rb_node);
> +		ret = do_write(ff, node,
> +			       sizeof(struct btf_node) + node->data_size);
> +		if (ret < 0)
> +			goto out;
> +	}
> +out:
> +	up_read(&env->bpf_info_lock);
> +	return ret;
> +}
> +
>  static int cpu_cache_level__sort(const void *a, const void *b)
>  {
>  	struct cpu_cache_level *cache_a = (struct cpu_cache_level *)a;
> @@ -1628,6 +1667,28 @@ static void print_bpf_prog_info(struct feat_fd *ff, FILE *fp)
>  	up_read(&env->bpf_info_lock);
>  }
>  
> +static void print_btf(struct feat_fd *ff, FILE *fp)
> +{
> +	struct perf_env *env = &ff->ph->env;
> +	struct rb_root *root;
> +	struct rb_node *next;
> +
> +	down_read(&env->bpf_info_lock);
> +
> +	root = &env->btfs;
> +	next = rb_first(root);
> +
> +	while (next) {
> +		struct btf_node *node;
> +
> +		node = rb_entry(next, struct btf_node, rb_node);
> +		next = rb_next(&node->rb_node);
> +		fprintf(fp, "# bpf_prog_info of id %u\n", node->id);
> +	}
> +
> +	up_read(&env->bpf_info_lock);
> +}
> +
>  static void free_event_desc(struct perf_evsel *events)
>  {
>  	struct perf_evsel *evsel;
> @@ -2723,6 +2784,41 @@ static int process_bpf_prog_info(struct feat_fd *ff,
>  	return err;
>  }
>  
> +static int process_btf(struct feat_fd *ff, void *data __maybe_unused)
> +{
> +	struct perf_env *env = &ff->ph->env;
> +	u32 count, i;
> +
> +	if (do_read_u32(ff, &count))
> +		return -1;
> +
> +	down_write(&env->bpf_info_lock);
> +
> +	for (i = 0; i < count; ++i) {
> +		struct btf_node btf_node;
> +		struct btf_node *node;
> +
> +		if (__do_read(ff, &btf_node, sizeof(struct btf_node)))
> +			return -1;
> +
> +		node = malloc(sizeof(struct btf_node) + btf_node.data_size);
> +		if (!node)
> +			return -1;
> +
> +		node->id = btf_node.id;
> +		node->data_size = btf_node.data_size;
> +
> +		if (__do_read(ff, node->data, btf_node.data_size)) {
> +			free(node);
> +			return -1;
> +		}
> +		perf_env__insert_btf(env, node);
> +	}
> +
> +	up_write(&env->bpf_info_lock);
> +	return 0;
> +}
> +
>  struct feature_ops {
>  	int (*write)(struct feat_fd *ff, struct perf_evlist *evlist);
>  	void (*print)(struct feat_fd *ff, FILE *fp);
> @@ -2783,7 +2879,8 @@ static const struct feature_ops feat_ops[HEADER_LAST_FEATURE] = {
>  	FEAT_OPR(SAMPLE_TIME,	sample_time,	false),
>  	FEAT_OPR(MEM_TOPOLOGY,	mem_topology,	true),
>  	FEAT_OPR(CLOCKID,       clockid,        false),
> -	FEAT_OPR(BPF_PROG_INFO, bpf_prog_info,  false)
> +	FEAT_OPR(BPF_PROG_INFO, bpf_prog_info,  false),
> +	FEAT_OPR(BTF,           btf,            false)
>  };
>  
>  struct header_print_data {
> diff --git a/tools/perf/util/header.h b/tools/perf/util/header.h
> index 0785c91b4c3a..ba51d8e43c53 100644
> --- a/tools/perf/util/header.h
> +++ b/tools/perf/util/header.h
> @@ -40,6 +40,7 @@ enum {
>  	HEADER_MEM_TOPOLOGY,
>  	HEADER_CLOCKID,
>  	HEADER_BPF_PROG_INFO,
> +	HEADER_BTF,
>  	HEADER_LAST_FEATURE,
>  	HEADER_FEAT_BITS	= 256,
>  };
> -- 
> 2.17.1

Re: [PATCH v2 perf,bpf 09/11] perf-top: add option --bpf-event

[Arnaldo Carvalho de Melo]

Em Thu, Feb 14, 2019 at 04:00:10PM -0800, Song Liu escreveu:
> bpf events are only tracked when opts->bpf_event is enabled. This patch
> adds command line flag to enable this for perf-top.

Shouldn't this start as enabled and we just provide a way to disable it
for testing purposes? Normally perf top works system wide, and then we
want to get info about BPF events, I think.

- Arnaldo
 
> Signed-off-by: Song Liu <songliubraving@fb.com>
> ---
>  tools/perf/builtin-top.c | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/tools/perf/builtin-top.c b/tools/perf/builtin-top.c
> index 27d8d42e0a4d..5271d7211b9c 100644
> --- a/tools/perf/builtin-top.c
> +++ b/tools/perf/builtin-top.c
> @@ -1492,6 +1492,7 @@ int cmd_top(int argc, const char **argv)
>  		    "Display raw encoding of assembly instructions (default)"),
>  	OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel,
>  		    "Enable kernel symbol demangling"),
> +	OPT_BOOLEAN(0, "bpf-event", &opts->bpf_event, "record bpf events"),
>  	OPT_STRING(0, "objdump", &top.annotation_opts.objdump_path, "path",
>  		    "objdump binary to use for disassembly and annotations"),
>  	OPT_STRING('M', "disassembler-style", &top.annotation_opts.disassembler_style, "disassembler style",
> -- 
> 2.17.1

Re: [PATCH v2 perf,bpf 07/11] perf, bpf: save btf in a rbtree in perf_env

[Arnaldo Carvalho de Melo]

Em Thu, Feb 14, 2019 at 04:00:08PM -0800, Song Liu escreveu:
> btf contains information necessary to annotate bpf programs. This patch
> saves btf for bpf programs loaded in the system.
> 
> Signed-off-by: Song Liu <songliubraving@fb.com>
> ---
>  tools/perf/util/bpf-event.c | 22 +++++++++++++
>  tools/perf/util/bpf-event.h |  7 ++++
>  tools/perf/util/env.c       | 65 +++++++++++++++++++++++++++++++++++++
>  tools/perf/util/env.h       |  3 ++
>  4 files changed, 97 insertions(+)
> 
> diff --git a/tools/perf/util/bpf-event.c b/tools/perf/util/bpf-event.c
> index ead599bc4f4e..37a5b8134e00 100644
> --- a/tools/perf/util/bpf-event.c
> +++ b/tools/perf/util/bpf-event.c
> @@ -30,6 +30,27 @@ int machine__process_bpf_event(struct machine *machine __maybe_unused,
>  	return 0;
>  }
>  
> +static int perf_fetch_btf(struct perf_env *env, u32 btf_id, struct btf *btf)

Please use perf_env__fetch_bpf

> +{
> +	struct btf_node *node;
> +	u32 data_size;
> +	const void *data;
> +
> +	data = btf__get_raw_data(btf, &data_size);
> +
> +	node = malloc(data_size + sizeof(struct btf_node));
> +
> +	if (!node)
> +		return -1;
> +
> +	node->id = btf_id;
> +	node->data_size = data_size;
> +	memcpy(node->data, data, data_size);
> +
> +	perf_env__insert_btf(env, node);

Just like you did above :-)

> +	return 0;
> +}
> +
>  /*
>   * Synthesize PERF_RECORD_KSYMBOL and PERF_RECORD_BPF_EVENT for one bpf
>   * program. One PERF_RECORD_BPF_EVENT is generated for the program. And
> @@ -109,6 +130,7 @@ static int perf_event__synthesize_one_bpf_prog(struct perf_session *session,
>  			goto out;
>  		}
>  		has_btf = true;
> +		perf_fetch_btf(env, info->btf_id, btf);
>  	}
>  
>  	/* Synthesize PERF_RECORD_KSYMBOL */
> diff --git a/tools/perf/util/bpf-event.h b/tools/perf/util/bpf-event.h
> index 11e6730b6105..60ce24e4e5c6 100644
> --- a/tools/perf/util/bpf-event.h
> +++ b/tools/perf/util/bpf-event.h
> @@ -20,6 +20,13 @@ struct bpf_prog_info_node {
>  	struct rb_node			rb_node;
>  };
>  
> +struct btf_node {
> +	struct rb_node	rb_node;
> +	u32		id;
> +	u32		data_size;
> +	char		data[];
> +};
> +
>  #ifdef HAVE_LIBBPF_SUPPORT
>  int machine__process_bpf_event(struct machine *machine, union perf_event *event,
>  			       struct perf_sample *sample);
> diff --git a/tools/perf/util/env.c b/tools/perf/util/env.c
> index 665b6fe3c7b2..6f9e3d4b94bc 100644
> --- a/tools/perf/util/env.c
> +++ b/tools/perf/util/env.c
> @@ -61,6 +61,57 @@ struct bpf_prog_info_node *perf_env__find_bpf_prog_info(struct perf_env *env,
>  	return node;
>  }
>  
> +void perf_env__insert_btf(struct perf_env *env, struct btf_node *btf_node)
> +{
> +	struct rb_node *parent = NULL;
> +	__u32 btf_id = btf_node->id;
> +	struct btf_node *node;
> +	struct rb_node **p;
> +
> +	down_write(&env->bpf_info_lock);
> +	p = &env->btfs.rb_node;
> +
> +	while (*p != NULL) {
> +		parent = *p;
> +		node = rb_entry(parent, struct btf_node, rb_node);
> +		if (btf_id < node->id) {
> +			p = &(*p)->rb_left;
> +		} else if (btf_id > node->id) {
> +			p = &(*p)->rb_right;
> +		} else {
> +			pr_debug("duplicated btf %u\n", btf_id);

			goto out:

> +			up_write(&env->bpf_info_lock);
> +			return;
> +		}
> +	}
> +
> +	rb_link_node(&btf_node->rb_node, parent, p);
> +	rb_insert_color(&btf_node->rb_node, &env->btfs);
out:
> +	up_write(&env->bpf_info_lock);
> +}
> +
> +struct btf_node *perf_env__find_btf(struct perf_env *env, __u32 btf_id)
> +{
> +	struct btf_node *node = NULL;
> +	struct rb_node *n;
> +
> +	down_read(&env->bpf_info_lock);
> +	n = env->btfs.rb_node;
> +
> +	while (n) {
> +		node = rb_entry(n, struct btf_node, rb_node);
> +		if (btf_id < node->id)
> +			n = n->rb_left;
> +		else if (btf_id > node->id)
> +			n = n->rb_right;
> +		else
> +			break;
> +	}
> +
> +	up_read(&env->bpf_info_lock);
> +	return node;
> +}
> +
>  /* purge data in bpf_prog_infos tree */
>  static void purge_bpf_info(struct perf_env *env)

The above should've been perf_env__purge_bpf()

>  {
> @@ -80,6 +131,19 @@ static void purge_bpf_info(struct perf_env *env)
>  		rb_erase_init(&node->rb_node, root);
>  		free(node);
>  	}
> +
> +	root = &env->btfs;
> +	next = rb_first(root);
> +
> +	while (next) {
> +		struct btf_node *node;
> +
> +		node = rb_entry(next, struct btf_node, rb_node);
> +		next = rb_next(&node->rb_node);
> +		rb_erase_init(&node->rb_node, root);
> +		free(node);
> +	}
> +
>  	up_write(&env->bpf_info_lock);
>  }
>  
> @@ -117,6 +181,7 @@ void perf_env__exit(struct perf_env *env)
>  static void init_bpf_rb_trees(struct perf_env *env)
>  {
>  	env->bpf_prog_infos = RB_ROOT;
> +	env->btfs = RB_ROOT;
>  	init_rwsem(&env->bpf_info_lock);
>  }
>  
> diff --git a/tools/perf/util/env.h b/tools/perf/util/env.h
> index a6d25e91bc98..f0c74255423c 100644
> --- a/tools/perf/util/env.h
> +++ b/tools/perf/util/env.h
> @@ -74,6 +74,7 @@ struct perf_env {
>  	 */
>  	struct rw_semaphore	bpf_info_lock;
>  	struct rb_root		bpf_prog_infos;
> +	struct rb_root		btfs;
>  };
>  
>  extern struct perf_env perf_env;
> @@ -94,4 +95,6 @@ void perf_env__insert_bpf_prog_info(struct perf_env *env,
>  				    struct bpf_prog_info_node *info_node);
>  struct bpf_prog_info_node *perf_env__find_bpf_prog_info(struct perf_env *env,
>  							__u32 prog_id);
> +void perf_env__insert_btf(struct perf_env *env, struct btf_node *btf_node);
> +struct btf_node *perf_env__find_btf(struct perf_env *env, __u32 btf_id);
>  #endif /* __PERF_ENV_H */
> -- 
> 2.17.1

Re: [PATCH v2 perf,bpf 11/11] perf, bpf: save information about short living bpf programs

[Arnaldo Carvalho de Melo]

Em Thu, Feb 14, 2019 at 04:00:45PM -0800, Song Liu escreveu:
> To annotate bpf programs in perf, it is necessary to save information in
> bpf_prog_info and btf. For short living bpf program, it is necessary to
> save these information before it is unloaded.
> 
> This patch saves these information in a separate thread. This thread
> creates its own evlist, that only tracks bpf events. This evlists uses
> ring buffer with very low watermark for lower latency. When bpf load
> events are received, this thread tries to gather information via sys_bpf
> and save it in perf_env.
> 
> Signed-off-by: Song Liu <songliubraving@fb.com>
> ---
>  tools/perf/builtin-record.c |  13 ++++
>  tools/perf/builtin-top.c    |  12 ++++
>  tools/perf/util/bpf-event.c | 129 ++++++++++++++++++++++++++++++++++++
>  tools/perf/util/bpf-event.h |  22 ++++++
>  tools/perf/util/evlist.c    |  20 ++++++
>  tools/perf/util/evlist.h    |   2 +
>  6 files changed, 198 insertions(+)
> 
> diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
> index 2355e0a9eda0..46abb44aaaab 100644
> --- a/tools/perf/builtin-record.c
> +++ b/tools/perf/builtin-record.c
> @@ -1106,6 +1106,8 @@ static int __cmd_record(struct record *rec, int argc, const char **argv)
>  	struct perf_data *data = &rec->data;
>  	struct perf_session *session;
>  	bool disabled = false, draining = false;
> +	struct bpf_event_poll_args poll_args;
> +	bool bpf_thread_running = false;
>  	int fd;
>  
>  	atexit(record__sig_exit);
> @@ -1206,6 +1208,14 @@ static int __cmd_record(struct record *rec, int argc, const char **argv)
>  		goto out_child;
>  	}
>  
> +	if (rec->opts.bpf_event) {
> +		poll_args.env = &session->header.env;
> +		poll_args.target = &rec->opts.target;
> +		poll_args.done = &done;
> +		if (bpf_event__start_polling_thread(&poll_args) == 0)
> +			bpf_thread_running = true;
> +	}
> +
>  	err = record__synthesize(rec, false);
>  	if (err < 0)
>  		goto out_child;
> @@ -1456,6 +1466,9 @@ static int __cmd_record(struct record *rec, int argc, const char **argv)
>  
>  out_delete_session:
>  	perf_session__delete(session);
> +
> +	if (bpf_thread_running)
> +		bpf_event__stop_polling_thread(&poll_args);
>  	return status;
>  }
>  
> diff --git a/tools/perf/builtin-top.c b/tools/perf/builtin-top.c
> index 5271d7211b9c..2586ee081967 100644
> --- a/tools/perf/builtin-top.c
> +++ b/tools/perf/builtin-top.c
> @@ -1524,10 +1524,12 @@ int cmd_top(int argc, const char **argv)
>  			"number of thread to run event synthesize"),
>  	OPT_END()
>  	};
> +	struct bpf_event_poll_args poll_args;
>  	const char * const top_usage[] = {
>  		"perf top [<options>]",
>  		NULL
>  	};
> +	bool bpf_thread_running = false;
>  	int status = hists__init();
>  
>  	if (status < 0)
> @@ -1652,8 +1654,18 @@ int cmd_top(int argc, const char **argv)
>  		signal(SIGWINCH, winch_sig);
>  	}
>  
> +	if (top.record_opts.bpf_event) {
> +		poll_args.env = &perf_env;
> +		poll_args.target = target;
> +		poll_args.done = &done;
> +		if (bpf_event__start_polling_thread(&poll_args) == 0)
> +			bpf_thread_running = true;
> +	}
>  	status = __cmd_top(&top);
>  
> +	if (bpf_thread_running)
> +		bpf_event__stop_polling_thread(&poll_args);
> +
>  out_delete_evlist:
>  	perf_evlist__delete(top.evlist);
>  
> diff --git a/tools/perf/util/bpf-event.c b/tools/perf/util/bpf-event.c
> index 4f347d61ed96..0caf137c515b 100644
> --- a/tools/perf/util/bpf-event.c
> +++ b/tools/perf/util/bpf-event.c
> @@ -8,6 +8,7 @@
>  #include "machine.h"
>  #include "env.h"
>  #include "session.h"
> +#include "evlist.h"
>  
>  #define ptr_to_u64(ptr)    ((__u64)(unsigned long)(ptr))
>  
> @@ -316,3 +317,131 @@ int perf_event__synthesize_bpf_events(struct perf_session *session,
>  	free(event);
>  	return err;
>  }
> +
> +static void perf_env_add_bpf_info(struct perf_env *env, u32 id)
> +{
> +	struct bpf_prog_info_linear *info_linear;
> +	struct bpf_prog_info_node *info_node;
> +	struct btf *btf = NULL;
> +	u64 arrays;
> +	u32 btf_id;
> +	int fd;
> +
> +	fd = bpf_prog_get_fd_by_id(id);
> +	if (fd < 0)
> +		return;
> +
> +	arrays = 1UL << BPF_PROG_INFO_JITED_KSYMS;
> +	arrays |= 1UL << BPF_PROG_INFO_JITED_FUNC_LENS;
> +	arrays |= 1UL << BPF_PROG_INFO_FUNC_INFO;
> +	arrays |= 1UL << BPF_PROG_INFO_PROG_TAGS;
> +	arrays |= 1UL << BPF_PROG_INFO_JITED_INSNS;
> +	arrays |= 1UL << BPF_PROG_INFO_LINE_INFO;
> +	arrays |= 1UL << BPF_PROG_INFO_JITED_LINE_INFO;
> +
> +	info_linear = bpf_program__get_prog_info_linear(fd, arrays);
> +	if (IS_ERR_OR_NULL(info_linear)) {
> +		pr_debug("%s: failed to get BPF program info. aborting\n", __func__);
> +		goto out;
> +	}
> +
> +	btf_id = info_linear->info.btf_id;
> +
> +	info_node = malloc(sizeof(struct bpf_prog_info_node));
> +	if (info_node) {
> +		info_node->info_linear = info_linear;
> +		perf_env__insert_bpf_prog_info(env, info_node);
> +	} else
> +		free(info_linear);
> +
> +	if (btf_id == 0)
> +		goto out;
> +
> +	if (btf__get_from_id(btf_id, &btf)) {
> +		pr_debug("%s: failed to get BTF of id %u, aborting\n",
> +			 __func__, btf_id);
> +		goto out;
> +	}
> +	perf_fetch_btf(env, btf_id, btf);
> +
> +out:
> +	free(btf);
> +	close(fd);
> +}
> +
> +static void *bpf_poll_thread(void *arg)
> +{
> +	struct bpf_event_poll_args *args = arg;
> +	int i;
> +
> +	while (!*(args->done)) {
> +		perf_evlist__poll(args->evlist, 1000);
> +
> +		for (i = 0; i < args->evlist->nr_mmaps; i++) {
> +			struct perf_mmap *map = &args->evlist->mmap[i];
> +			union perf_event *event;
> +
> +			if (perf_mmap__read_init(map))
> +				continue;
> +			while ((event = perf_mmap__read_event(map)) != NULL) {
> +				pr_debug("processing vip event of type %d\n",
> +					 event->header.type);
> +				switch (event->header.type) {
> +				case PERF_RECORD_BPF_EVENT:
> +					if (event->bpf_event.type != PERF_BPF_EVENT_PROG_LOAD)
> +						break;
> +					perf_env_add_bpf_info(args->env, event->bpf_event.id);
> +					break;
> +				default:
> +					break;
> +				}
> +				perf_mmap__consume(map);
> +			}
> +			perf_mmap__read_done(map);
> +		}
> +	}
> +	return NULL;
> +}
> +
> +pthread_t poll_thread;
> +
> +int bpf_event__start_polling_thread(struct bpf_event_poll_args *args)
> +{
> +	struct perf_evsel *counter;
> +
> +	args->evlist = perf_evlist__new();
> +
> +	if (args->evlist == NULL)
> +		return -1;
> +
> +	if (perf_evlist__create_maps(args->evlist, args->target))
		goto out_delete_evlist;
> +
> +	if (perf_evlist__add_bpf_tracker(args->evlist))
		goto out_delete_evlist;
> +
> +	evlist__for_each_entry(args->evlist, counter) {
> +		if (perf_evsel__open(counter, args->evlist->cpus,
> +				     args->evlist->threads) < 0)
			goto out_delete_evlist;
> +	}
> +
> +	if (perf_evlist__mmap(args->evlist, UINT_MAX))
		goto out_delete_evlist;
> +
> +	evlist__for_each_entry(args->evlist, counter) {
> +		if (perf_evsel__enable(counter))
			goto out_delete_evlist;
> +	}
> +
> +	if (pthread_create(&poll_thread, NULL, bpf_poll_thread, args))
		goto out_delete_evlist;	
> +
> +	return 0;
out_delete_evlist:
	perf_evlist__delete(args->evlist);
	args->evlist = NULL;

	return -1;
> +}
> +
> +void bpf_event__stop_polling_thread(struct bpf_event_poll_args *args)
> +{
> +	pthread_join(poll_thread, NULL);
> +	perf_evlist__exit(args->evlist);
> +}
> diff --git a/tools/perf/util/bpf-event.h b/tools/perf/util/bpf-event.h
> index c4f0f1395ea5..61914827c1e3 100644
> --- a/tools/perf/util/bpf-event.h
> +++ b/tools/perf/util/bpf-event.h
> @@ -12,12 +12,17 @@
>  #include <bpf/libbpf.h>
>  #include <linux/btf.h>
>  #include <linux/rbtree.h>
> +#include <pthread.h>
> +#include <api/fd/array.h>
>  #include "event.h"
>  
>  struct machine;
>  union perf_event;
> +struct perf_env;
>  struct perf_sample;
>  struct record_opts;
> +struct evlist;
> +struct target;
>  
>  struct bpf_prog_info_node {
>  	struct bpf_prog_info_linear	*info_linear;
> @@ -31,6 +36,13 @@ struct btf_node {
>  	char		data[];
>  };
>  
> +struct bpf_event_poll_args {
> +	struct perf_env		*env;
> +	struct perf_evlist	*evlist;
> +	struct target		*target;
> +	volatile int		*done;
> +};
> +
>  #ifdef HAVE_LIBBPF_SUPPORT
>  int machine__process_bpf_event(struct machine *machine, union perf_event *event,
>  			       struct perf_sample *sample);
> @@ -39,6 +51,8 @@ int perf_event__synthesize_bpf_events(struct perf_session *session,
>  				      perf_event__handler_t process,
>  				      struct machine *machine,
>  				      struct record_opts *opts);
> +int bpf_event__start_polling_thread(struct bpf_event_poll_args *args);
> +void bpf_event__stop_polling_thread(struct bpf_event_poll_args *args);
>  #else
>  static inline int machine__process_bpf_event(struct machine *machine __maybe_unused,
>  					     union perf_event *event __maybe_unused,
> @@ -54,5 +68,13 @@ static inline int perf_event__synthesize_bpf_events(struct perf_session *session
>  {
>  	return 0;
>  }
> +
> +static inline int bpf_event__start_polling_thread(struct bpf_event_poll_args *args __maybe_unused)
> +{
> +	return 0;
> +}
> +void bpf_event__stop_polling_thread(struct bpf_event_poll_args *args __maybe_unused)
> +{
> +}
>  #endif // HAVE_LIBBPF_SUPPORT
>  #endif
> diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c
> index 8c902276d4b4..612c079579ce 100644
> --- a/tools/perf/util/evlist.c
> +++ b/tools/perf/util/evlist.c
> @@ -271,6 +271,26 @@ int perf_evlist__add_dummy(struct perf_evlist *evlist)
>  	return 0;
>  }
>  
> +int perf_evlist__add_bpf_tracker(struct perf_evlist *evlist)
> +{
> +	struct perf_event_attr attr = {
> +		.type	          = PERF_TYPE_SOFTWARE,
> +		.config           = PERF_COUNT_SW_DUMMY,
> +		.watermark        = 1,
> +		.bpf_event        = 1,
> +		.wakeup_watermark = 1,
> +		.size	   = sizeof(attr), /* to capture ABI version */
> +	};
> +	struct perf_evsel *evsel = perf_evsel__new_idx(&attr,
> +						       evlist->nr_entries);
> +
> +	if (evsel == NULL)
> +		return -ENOMEM;
> +
> +	perf_evlist__add(evlist, evsel);

You could use:

	struct perf_evlist *evlist = perf_evlist__new_dummy();
	if (evlist != NULL) {
		struct perf_evsel *evsel == perf_evlist__first(evlist);
		evsel->attr.bpf_event = evsel->attr.watermark = evsel->attr.wakeup_watermark = 1;
		return 0;
	}
	return -1;

Because in this case all you'll have in this evlist is the bpf tracker,
right? The add_bpf_tracker would be handy if we would want to have a
pre-existing evlist with some other events and wanted to add a bpf
tracker, no?

- Arnaldo

> +	return 0;
> +}
> +
>  static int perf_evlist__add_attrs(struct perf_evlist *evlist,
>  				  struct perf_event_attr *attrs, size_t nr_attrs)
>  {
> diff --git a/tools/perf/util/evlist.h b/tools/perf/util/evlist.h
> index 868294491194..a2d22715188e 100644
> --- a/tools/perf/util/evlist.h
> +++ b/tools/perf/util/evlist.h
> @@ -84,6 +84,8 @@ int __perf_evlist__add_default_attrs(struct perf_evlist *evlist,
>  
>  int perf_evlist__add_dummy(struct perf_evlist *evlist);
>  
> +int perf_evlist__add_bpf_tracker(struct perf_evlist *evlist);
> +
>  int perf_evlist__add_newtp(struct perf_evlist *evlist,
>  			   const char *sys, const char *name, void *handler);
>  
> -- 
> 2.17.1

Re: [PATCH v2 perf,bpf 01/11] perf, bpf: consider events with attr.bpf_event as side-band events

[Song Liu]

> On Feb 15, 2019, at 6:14 AM, Arnaldo Carvalho de Melo <acme@redhat.com> wrote:
> 
> Em Thu, Feb 14, 2019 at 03:58:40PM -0800, Song Liu escreveu:
>> Events with bpf_event should be considered as side-band event, as they
>> carry information about BPF programs.
>> 
>> Fixes: 6ee52e2a3fe4 ("perf, bpf: Introduce PERF_RECORD_BPF_EVENT")
>> Signed-off-by: Song Liu <songliubraving@fb.com>
>> ---
>> kernel/events/core.c | 3 ++-
>> 1 file changed, 2 insertions(+), 1 deletion(-)
>> 
>> diff --git a/kernel/events/core.c b/kernel/events/core.c
>> index 0a8dab322111..9403bdda5f8c 100644
>> --- a/kernel/events/core.c
>> +++ b/kernel/events/core.c
>> @@ -4238,7 +4238,8 @@ static bool is_sb_event(struct perf_event *event)
>> 	if (attr->mmap || attr->mmap_data || attr->mmap2 ||
>> 	    attr->comm || attr->comm_exec ||
>> 	    attr->task || attr->ksymbol ||
				^ here

>> -	    attr->context_switch)
>> +	    attr->context_switch ||
>> +	    attr->bpf_event)
> 
> What about attr->ksymbol?
> 
> - Arnaldo

It is already included up there. attr->bpf_event was missed initially 
because we didn't use events with bpf_event but not ksymbol. With this 
set, the polling thread only has bpf_event, so we need this fix. 

Thanks,
Song

Re: [PATCH v2 perf,bpf 05/11] perf, bpf: save bpf_prog_info in a rbtree in perf_env

[Song Liu]

> On Feb 15, 2019, at 6:21 AM, Arnaldo Carvalho de Melo <acme@redhat.com> wrote:
> 
> Em Thu, Feb 14, 2019 at 04:00:06PM -0800, Song Liu escreveu:
>> bpf_prog_info contains information necessary to annotate bpf programs.
>> This patch saves bpf_prog_info for bpf programs loaded in the system.
>> 
>> Signed-off-by: Song Liu <songliubraving@fb.com>
>> ---
>> tools/perf/builtin-record.c |  2 +-
>> tools/perf/builtin-top.c    |  2 +-
>> tools/perf/util/bpf-event.c | 39 +++++++++++++----
>> tools/perf/util/bpf-event.h | 15 +++++--
>> tools/perf/util/env.c       | 83 +++++++++++++++++++++++++++++++++++++
>> tools/perf/util/env.h       | 14 +++++++
>> 6 files changed, 142 insertions(+), 13 deletions(-)
>> 
>> diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
>> index 88ea11d57c6f..2355e0a9eda0 100644
>> --- a/tools/perf/builtin-record.c
>> +++ b/tools/perf/builtin-record.c
>> @@ -1083,7 +1083,7 @@ static int record__synthesize(struct record *rec, bool tail)
>> 		return err;
>> 	}
>> 
>> -	err = perf_event__synthesize_bpf_events(tool, process_synthesized_event,
>> +	err = perf_event__synthesize_bpf_events(session, process_synthesized_event,
>> 						machine, opts);
>> 	if (err < 0)
>> 		pr_warning("Couldn't synthesize bpf events.\n");
>> diff --git a/tools/perf/builtin-top.c b/tools/perf/builtin-top.c
>> index 5a486d4de56e..27d8d42e0a4d 100644
>> --- a/tools/perf/builtin-top.c
>> +++ b/tools/perf/builtin-top.c
>> @@ -1216,7 +1216,7 @@ static int __cmd_top(struct perf_top *top)
>> 
>> 	init_process_thread(top);
>> 
>> -	ret = perf_event__synthesize_bpf_events(&top->tool, perf_event__process,
>> +	ret = perf_event__synthesize_bpf_events(top->session, perf_event__process,
>> 						&top->session->machines.host,
>> 						&top->record_opts);
>> 	if (ret < 0)
>> diff --git a/tools/perf/util/bpf-event.c b/tools/perf/util/bpf-event.c
>> index e6dfb95029e5..ead599bc4f4e 100644
>> --- a/tools/perf/util/bpf-event.c
>> +++ b/tools/perf/util/bpf-event.c
>> @@ -1,15 +1,13 @@
>> // SPDX-License-Identifier: GPL-2.0
>> #include <errno.h>
>> #include <stdlib.h>
>> -#include <bpf/bpf.h>
>> -#include <bpf/btf.h>
>> -#include <bpf/libbpf.h>
>> -#include <linux/btf.h>
> 
> I think you need these here, since in this C file you will use the
> definitions for these structs, see further comments below.
> 
>> #include <linux/err.h>
>> #include "bpf-event.h"
>> #include "debug.h"
>> #include "symbol.h"
>> #include "machine.h"
>> +#include "env.h"
>> +#include "session.h"
>> 
>> #define ptr_to_u64(ptr)    ((__u64)(unsigned long)(ptr))
>> 
>> @@ -42,7 +40,7 @@ int machine__process_bpf_event(struct machine *machine __maybe_unused,
>>  *   -1 for failures;
>>  *   -2 for lack of kernel support.
>>  */
>> -static int perf_event__synthesize_one_bpf_prog(struct perf_tool *tool,
>> +static int perf_event__synthesize_one_bpf_prog(struct perf_session *session,
>> 					       perf_event__handler_t process,
>> 					       struct machine *machine,
>> 					       int fd,
>> @@ -52,17 +50,29 @@ static int perf_event__synthesize_one_bpf_prog(struct perf_tool *tool,
>> 	struct ksymbol_event *ksymbol_event = &event->ksymbol_event;
>> 	struct bpf_event *bpf_event = &event->bpf_event;
>> 	struct bpf_prog_info_linear *info_linear;
>> +	struct perf_tool *tool = session->tool;
>> +	struct bpf_prog_info_node *info_node;
>> 	struct bpf_prog_info *info;
>> 	struct btf *btf = NULL;
>> 	bool has_btf = false;
>> +	struct perf_env *env;
>> 	u32 sub_prog_cnt, i;
>> 	int err = 0;
>> 	u64 arrays;
>> 
>> +	/*
>> +	 * for perf-record and perf-report use header.env;
>> +	 * otherwise, use global perf_env.
>> +	 */
>> +	env = session->data ? &session->header.env : &perf_env;
>> +
>> 	arrays = 1UL << BPF_PROG_INFO_JITED_KSYMS;
>> 	arrays |= 1UL << BPF_PROG_INFO_JITED_FUNC_LENS;
>> 	arrays |= 1UL << BPF_PROG_INFO_FUNC_INFO;
>> 	arrays |= 1UL << BPF_PROG_INFO_PROG_TAGS;
>> +	arrays |= 1UL << BPF_PROG_INFO_JITED_INSNS;
>> +	arrays |= 1UL << BPF_PROG_INFO_LINE_INFO;
>> +	arrays |= 1UL << BPF_PROG_INFO_JITED_LINE_INFO;
>> 
>> 	info_linear = bpf_program__get_prog_info_linear(fd, arrays);
>> 	if (IS_ERR_OR_NULL(info_linear)) {
>> @@ -151,8 +161,8 @@ static int perf_event__synthesize_one_bpf_prog(struct perf_tool *tool,
>> 						     machine, process);
>> 	}
>> 
>> -	/* Synthesize PERF_RECORD_BPF_EVENT */
>> 	if (opts->bpf_event) {
>> +		/* Synthesize PERF_RECORD_BPF_EVENT */
>> 		*bpf_event = (struct bpf_event){
>> 			.header = {
>> 				.type = PERF_RECORD_BPF_EVENT,
>> @@ -165,6 +175,19 @@ static int perf_event__synthesize_one_bpf_prog(struct perf_tool *tool,
>> 		memcpy(bpf_event->tag, info->tag, BPF_TAG_SIZE);
>> 		memset((void *)event + event->header.size, 0, machine->id_hdr_size);
>> 		event->header.size += machine->id_hdr_size;
>> +
>> +		/* save bpf_prog_info to env */
>> +		info_node = malloc(sizeof(struct bpf_prog_info_node));
>> +		if (info_node) {
>> +			info_node->info_linear = info_linear;
>> +			perf_env__insert_bpf_prog_info(env, info_node);
>> +			info_linear = NULL;
>> +		}
>> +
>> +		/*
>> +		 * process after saving bpf_prog_info to env, so that
>> +		 * required information is ready for look up
>> +		 */
>> 		err = perf_tool__process_synth_event(tool, event,
>> 						     machine, process);
>> 	}
>> @@ -175,7 +198,7 @@ static int perf_event__synthesize_one_bpf_prog(struct perf_tool *tool,
>> 	return err ? -1 : 0;
>> }
>> 
>> -int perf_event__synthesize_bpf_events(struct perf_tool *tool,
>> +int perf_event__synthesize_bpf_events(struct perf_session *session,
>> 				      perf_event__handler_t process,
>> 				      struct machine *machine,
>> 				      struct record_opts *opts)
>> @@ -209,7 +232,7 @@ int perf_event__synthesize_bpf_events(struct perf_tool *tool,
>> 			continue;
>> 		}
>> 
>> -		err = perf_event__synthesize_one_bpf_prog(tool, process,
>> +		err = perf_event__synthesize_one_bpf_prog(session, process,
>> 							  machine, fd,
>> 							  event, opts);
>> 		close(fd);
>> diff --git a/tools/perf/util/bpf-event.h b/tools/perf/util/bpf-event.h
>> index 7890067e1a37..11e6730b6105 100644
>> --- a/tools/perf/util/bpf-event.h
>> +++ b/tools/perf/util/bpf-event.h
>> @@ -3,19 +3,28 @@
>> #define __PERF_BPF_EVENT_H
>> 
>> #include <linux/compiler.h>
>> +#include <bpf/bpf.h>
>> +#include <bpf/btf.h>
>> +#include <bpf/libbpf.h>
>> +#include <linux/btf.h>
> 
> Are you sure you'll need all of these headers here? Perhaps just some
> forward declarations will do?
> 
> In fact the only bpf or btf structure here seems to be
> bpf_prog_info_linear, which needs ust a forward declaration.
> 
> Avoiding these includes reduces the build time, and since the build-test
> target does many builds and I want to build this in many containers, we
> should try to reduce the build time by using just what is needed in each
> header and C file. Also during development it helps with not rebuilding
> tons of things when something unrelated changes in a header, etc.

I see. I will fix this in next version. 

> 
> - Arnaldo
> 
>> +#include <linux/rbtree.h>
>> #include "event.h"
>> 
>> struct machine;
>> union perf_event;
>> struct perf_sample;
>> -struct perf_tool;
>> struct record_opts;
>> 
>> +struct bpf_prog_info_node {
>> +	struct bpf_prog_info_linear	*info_linear;
>> +	struct rb_node			rb_node;
>> +};
>> +
>> #ifdef HAVE_LIBBPF_SUPPORT
>> int machine__process_bpf_event(struct machine *machine, union perf_event *event,
>> 			       struct perf_sample *sample);
>> 
>> -int perf_event__synthesize_bpf_events(struct perf_tool *tool,
>> +int perf_event__synthesize_bpf_events(struct perf_session *session,
>> 				      perf_event__handler_t process,
>> 				      struct machine *machine,
>> 				      struct record_opts *opts);
>> @@ -27,7 +36,7 @@ static inline int machine__process_bpf_event(struct machine *machine __maybe_unu
>> 	return 0;
>> }
>> 
>> -static inline int perf_event__synthesize_bpf_events(struct perf_tool *tool __maybe_unused,
>> +static inline int perf_event__synthesize_bpf_events(struct perf_session *session __maybe_unused,
>> 						    perf_event__handler_t process __maybe_unused,
>> 						    struct machine *machine __maybe_unused,
>> 						    struct record_opts *opts __maybe_unused)
>> diff --git a/tools/perf/util/env.c b/tools/perf/util/env.c
>> index 4c23779e271a..665b6fe3c7b2 100644
>> --- a/tools/perf/util/env.c
>> +++ b/tools/perf/util/env.c
>> @@ -8,10 +8,86 @@
>> 
>> struct perf_env perf_env;
>> 
>> +void perf_env__insert_bpf_prog_info(struct perf_env *env,
>> +				    struct bpf_prog_info_node *info_node)
>> +{
>> +	__u32 prog_id = info_node->info_linear->info.id;
>> +	struct bpf_prog_info_node *node;
>> +	struct rb_node *parent = NULL;
>> +	struct rb_node **p;
>> +
>> +	down_write(&env->bpf_info_lock);
>> +	p = &env->bpf_prog_infos.rb_node;
>> +
>> +	while (*p != NULL) {
>> +		parent = *p;
>> +		node = rb_entry(parent, struct bpf_prog_info_node, rb_node);
>> +		if (prog_id < node->info_linear->info.id) {
>> +			p = &(*p)->rb_left;
>> +		} else if (prog_id > node->info_linear->info.id) {
>> +			p = &(*p)->rb_right;
>> +		} else {
>> +			pr_debug("duplicated bpf prog info %u\n", prog_id);
>> +			up_write(&env->bpf_info_lock);
>> +			return;
>> +		}
>> +	}
>> +
>> +	rb_link_node(&info_node->rb_node, parent, p);
>> +	rb_insert_color(&info_node->rb_node, &env->bpf_prog_infos);
>> +	up_write(&env->bpf_info_lock);
>> +}
>> +
>> +struct bpf_prog_info_node *perf_env__find_bpf_prog_info(struct perf_env *env,
>> +							__u32 prog_id)
>> +{
>> +	struct bpf_prog_info_node *node = NULL;
>> +	struct rb_node *n;
>> +
>> +	down_read(&env->bpf_info_lock);
>> +	n = env->bpf_prog_infos.rb_node;
>> +
>> +	while (n) {
>> +		node = rb_entry(n, struct bpf_prog_info_node, rb_node);
>> +		if (prog_id < node->info_linear->info.id)
>> +			n = n->rb_left;
>> +		else if (prog_id > node->info_linear->info.id)
>> +			n = n->rb_right;
>> +		else
>> +			break;
>> +	}
>> +
>> +	up_read(&env->bpf_info_lock);
>> +	return node;
>> +}
>> +
>> +/* purge data in bpf_prog_infos tree */
>> +static void purge_bpf_info(struct perf_env *env)
>> +{
>> +	struct rb_root *root;
>> +	struct rb_node *next;
>> +
>> +	down_write(&env->bpf_info_lock);
>> +
>> +	root = &env->bpf_prog_infos;
>> +	next = rb_first(root);
>> +
>> +	while (next) {
>> +		struct bpf_prog_info_node *node;
>> +
>> +		node = rb_entry(next, struct bpf_prog_info_node, rb_node);
>> +		next = rb_next(&node->rb_node);
>> +		rb_erase_init(&node->rb_node, root);
>> +		free(node);
>> +	}
>> +	up_write(&env->bpf_info_lock);
>> +}
>> +
>> void perf_env__exit(struct perf_env *env)
>> {
>> 	int i;
>> 
>> +	purge_bpf_info(env);
>> 	zfree(&env->hostname);
>> 	zfree(&env->os_release);
>> 	zfree(&env->version);
>> @@ -38,6 +114,12 @@ void perf_env__exit(struct perf_env *env)
>> 	zfree(&env->memory_nodes);
>> }
>> 
>> +static void init_bpf_rb_trees(struct perf_env *env)
>> +{
>> +	env->bpf_prog_infos = RB_ROOT;
>> +	init_rwsem(&env->bpf_info_lock);
>> +}
>> +
>> int perf_env__set_cmdline(struct perf_env *env, int argc, const char *argv[])
>> {
>> 	int i;
>> @@ -59,6 +141,7 @@ int perf_env__set_cmdline(struct perf_env *env, int argc, const char *argv[])
>> 
>> 	env->nr_cmdline = argc;
>> 
>> +	init_bpf_rb_trees(env);
>> 	return 0;
>> out_free:
>> 	zfree(&env->cmdline_argv);
>> diff --git a/tools/perf/util/env.h b/tools/perf/util/env.h
>> index d01b8355f4ca..a6d25e91bc98 100644
>> --- a/tools/perf/util/env.h
>> +++ b/tools/perf/util/env.h
>> @@ -3,7 +3,10 @@
>> #define __PERF_ENV_H
>> 
>> #include <linux/types.h>
>> +#include <linux/rbtree.h>
>> #include "cpumap.h"
>> +#include "rwsem.h"
> 
> You don't need the following header, just use a forward declaration for
> the sole struct you use with a pointer:
> 
> struct bpf_prog_info_node;
> 
>> +#include "bpf-event.h"
>> 
>> struct cpu_topology_map {
>> 	int	socket_id;
>> @@ -64,6 +67,13 @@ struct perf_env {
>> 	struct memory_node	*memory_nodes;
>> 	unsigned long long	 memory_bsize;
>> 	u64                     clockid_res_ns;
>> +
>> +	/*
>> +	 * bpf_info_lock protects bpf rbtrees. This is needed because the
>> +	 * trees are accessed by different threads in perf-top
>> +	 */
>> +	struct rw_semaphore	bpf_info_lock;
>> +	struct rb_root		bpf_prog_infos;
> 
> Please group this into a structure, i.e.:
> 
> 	struct {
> 		struct rw_semaphore lock;
> 		struct rb_root	    infos;
> 	} bpf_progs;

Will fix this in next version. 

> 
>> };
>> 
>> extern struct perf_env perf_env;
>> @@ -80,4 +90,8 @@ const char *perf_env__arch(struct perf_env *env);
>> const char *perf_env__raw_arch(struct perf_env *env);
>> int perf_env__nr_cpus_avail(struct perf_env *env);
>> 
>> +void perf_env__insert_bpf_prog_info(struct perf_env *env,
>> +				    struct bpf_prog_info_node *info_node);
>> +struct bpf_prog_info_node *perf_env__find_bpf_prog_info(struct perf_env *env,
>> +							__u32 prog_id);
>> #endif /* __PERF_ENV_H */
>> -- 
>> 2.17.1

Re: [PATCH v2 perf,bpf 09/11] perf-top: add option --bpf-event

[Song Liu]

> On Feb 15, 2019, at 6:27 AM, Arnaldo Carvalho de Melo <acme@redhat.com> wrote:
> 
> Em Thu, Feb 14, 2019 at 04:00:10PM -0800, Song Liu escreveu:
>> bpf events are only tracked when opts->bpf_event is enabled. This patch
>> adds command line flag to enable this for perf-top.
> 
> Shouldn't this start as enabled and we just provide a way to disable it
> for testing purposes? Normally perf top works system wide, and then we
> want to get info about BPF events, I think.
> 
> - Arnaldo

We can enable it by default, and add a --no-bpf-event option to disable it
for testing. 

Thanks,
Song



>> Signed-off-by: Song Liu <songliubraving@fb.com>
>> ---
>> tools/perf/builtin-top.c | 1 +
>> 1 file changed, 1 insertion(+)
>> 
>> diff --git a/tools/perf/builtin-top.c b/tools/perf/builtin-top.c
>> index 27d8d42e0a4d..5271d7211b9c 100644
>> --- a/tools/perf/builtin-top.c
>> +++ b/tools/perf/builtin-top.c
>> @@ -1492,6 +1492,7 @@ int cmd_top(int argc, const char **argv)
>> 		    "Display raw encoding of assembly instructions (default)"),
>> 	OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel,
>> 		    "Enable kernel symbol demangling"),
>> +	OPT_BOOLEAN(0, "bpf-event", &opts->bpf_event, "record bpf events"),
>> 	OPT_STRING(0, "objdump", &top.annotation_opts.objdump_path, "path",
>> 		    "objdump binary to use for disassembly and annotations"),
>> 	OPT_STRING('M', "disassembler-style", &top.annotation_opts.disassembler_style, "disassembler style",
>> -- 
>> 2.17.1

Re: [PATCH v2 perf,bpf 11/11] perf, bpf: save information about short living bpf programs

[Song Liu]

> On Feb 15, 2019, at 6:41 AM, Arnaldo Carvalho de Melo <acme@redhat.com> wrote:
> 
> Em Thu, Feb 14, 2019 at 04:00:45PM -0800, Song Liu escreveu:
>> To annotate bpf programs in perf, it is necessary to save information in
>> bpf_prog_info and btf. For short living bpf program, it is necessary to
>> save these information before it is unloaded.
>> 
>> This patch saves these information in a separate thread. This thread
>> creates its own evlist, that only tracks bpf events. This evlists uses
>> ring buffer with very low watermark for lower latency. When bpf load
>> events are received, this thread tries to gather information via sys_bpf
>> and save it in perf_env.
>> 
>> Signed-off-by: Song Liu <songliubraving@fb.com>
>> ---
>> tools/perf/builtin-record.c |  13 ++++
>> tools/perf/builtin-top.c    |  12 ++++
>> tools/perf/util/bpf-event.c | 129 ++++++++++++++++++++++++++++++++++++
>> tools/perf/util/bpf-event.h |  22 ++++++
>> tools/perf/util/evlist.c    |  20 ++++++
>> tools/perf/util/evlist.h    |   2 +
>> 6 files changed, 198 insertions(+)
>> 
>> diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
>> index 2355e0a9eda0..46abb44aaaab 100644
>> --- a/tools/perf/builtin-record.c
>> +++ b/tools/perf/builtin-record.c
>> @@ -1106,6 +1106,8 @@ static int __cmd_record(struct record *rec, int argc, const char **argv)
>> 	struct perf_data *data = &rec->data;
>> 	struct perf_session *session;
>> 	bool disabled = false, draining = false;
>> +	struct bpf_event_poll_args poll_args;
>> +	bool bpf_thread_running = false;
>> 	int fd;
>> 
>> 	atexit(record__sig_exit);
>> @@ -1206,6 +1208,14 @@ static int __cmd_record(struct record *rec, int argc, const char **argv)
>> 		goto out_child;
>> 	}
>> 
>> +	if (rec->opts.bpf_event) {
>> +		poll_args.env = &session->header.env;
>> +		poll_args.target = &rec->opts.target;
>> +		poll_args.done = &done;
>> +		if (bpf_event__start_polling_thread(&poll_args) == 0)
>> +			bpf_thread_running = true;
>> +	}
>> +
>> 	err = record__synthesize(rec, false);
>> 	if (err < 0)
>> 		goto out_child;
>> @@ -1456,6 +1466,9 @@ static int __cmd_record(struct record *rec, int argc, const char **argv)
>> 
>> out_delete_session:
>> 	perf_session__delete(session);
>> +
>> +	if (bpf_thread_running)
>> +		bpf_event__stop_polling_thread(&poll_args);
>> 	return status;
>> }
>> 
>> diff --git a/tools/perf/builtin-top.c b/tools/perf/builtin-top.c
>> index 5271d7211b9c..2586ee081967 100644
>> --- a/tools/perf/builtin-top.c
>> +++ b/tools/perf/builtin-top.c
>> @@ -1524,10 +1524,12 @@ int cmd_top(int argc, const char **argv)
>> 			"number of thread to run event synthesize"),
>> 	OPT_END()
>> 	};
>> +	struct bpf_event_poll_args poll_args;
>> 	const char * const top_usage[] = {
>> 		"perf top [<options>]",
>> 		NULL
>> 	};
>> +	bool bpf_thread_running = false;
>> 	int status = hists__init();
>> 
>> 	if (status < 0)
>> @@ -1652,8 +1654,18 @@ int cmd_top(int argc, const char **argv)
>> 		signal(SIGWINCH, winch_sig);
>> 	}
>> 
>> +	if (top.record_opts.bpf_event) {
>> +		poll_args.env = &perf_env;
>> +		poll_args.target = target;
>> +		poll_args.done = &done;
>> +		if (bpf_event__start_polling_thread(&poll_args) == 0)
>> +			bpf_thread_running = true;
>> +	}
>> 	status = __cmd_top(&top);
>> 
>> +	if (bpf_thread_running)
>> +		bpf_event__stop_polling_thread(&poll_args);
>> +
>> out_delete_evlist:
>> 	perf_evlist__delete(top.evlist);
>> 
>> diff --git a/tools/perf/util/bpf-event.c b/tools/perf/util/bpf-event.c
>> index 4f347d61ed96..0caf137c515b 100644
>> --- a/tools/perf/util/bpf-event.c
>> +++ b/tools/perf/util/bpf-event.c
>> @@ -8,6 +8,7 @@
>> #include "machine.h"
>> #include "env.h"
>> #include "session.h"
>> +#include "evlist.h"
>> 
>> #define ptr_to_u64(ptr)    ((__u64)(unsigned long)(ptr))
>> 
>> @@ -316,3 +317,131 @@ int perf_event__synthesize_bpf_events(struct perf_session *session,
>> 	free(event);
>> 	return err;
>> }
>> +
>> +static void perf_env_add_bpf_info(struct perf_env *env, u32 id)
>> +{
>> +	struct bpf_prog_info_linear *info_linear;
>> +	struct bpf_prog_info_node *info_node;
>> +	struct btf *btf = NULL;
>> +	u64 arrays;
>> +	u32 btf_id;
>> +	int fd;
>> +
>> +	fd = bpf_prog_get_fd_by_id(id);
>> +	if (fd < 0)
>> +		return;
>> +
>> +	arrays = 1UL << BPF_PROG_INFO_JITED_KSYMS;
>> +	arrays |= 1UL << BPF_PROG_INFO_JITED_FUNC_LENS;
>> +	arrays |= 1UL << BPF_PROG_INFO_FUNC_INFO;
>> +	arrays |= 1UL << BPF_PROG_INFO_PROG_TAGS;
>> +	arrays |= 1UL << BPF_PROG_INFO_JITED_INSNS;
>> +	arrays |= 1UL << BPF_PROG_INFO_LINE_INFO;
>> +	arrays |= 1UL << BPF_PROG_INFO_JITED_LINE_INFO;
>> +
>> +	info_linear = bpf_program__get_prog_info_linear(fd, arrays);
>> +	if (IS_ERR_OR_NULL(info_linear)) {
>> +		pr_debug("%s: failed to get BPF program info. aborting\n", __func__);
>> +		goto out;
>> +	}
>> +
>> +	btf_id = info_linear->info.btf_id;
>> +
>> +	info_node = malloc(sizeof(struct bpf_prog_info_node));
>> +	if (info_node) {
>> +		info_node->info_linear = info_linear;
>> +		perf_env__insert_bpf_prog_info(env, info_node);
>> +	} else
>> +		free(info_linear);
>> +
>> +	if (btf_id == 0)
>> +		goto out;
>> +
>> +	if (btf__get_from_id(btf_id, &btf)) {
>> +		pr_debug("%s: failed to get BTF of id %u, aborting\n",
>> +			 __func__, btf_id);
>> +		goto out;
>> +	}
>> +	perf_fetch_btf(env, btf_id, btf);
>> +
>> +out:
>> +	free(btf);
>> +	close(fd);
>> +}
>> +
>> +static void *bpf_poll_thread(void *arg)
>> +{
>> +	struct bpf_event_poll_args *args = arg;
>> +	int i;
>> +
>> +	while (!*(args->done)) {
>> +		perf_evlist__poll(args->evlist, 1000);
>> +
>> +		for (i = 0; i < args->evlist->nr_mmaps; i++) {
>> +			struct perf_mmap *map = &args->evlist->mmap[i];
>> +			union perf_event *event;
>> +
>> +			if (perf_mmap__read_init(map))
>> +				continue;
>> +			while ((event = perf_mmap__read_event(map)) != NULL) {
>> +				pr_debug("processing vip event of type %d\n",
>> +					 event->header.type);
>> +				switch (event->header.type) {
>> +				case PERF_RECORD_BPF_EVENT:
>> +					if (event->bpf_event.type != PERF_BPF_EVENT_PROG_LOAD)
>> +						break;
>> +					perf_env_add_bpf_info(args->env, event->bpf_event.id);
>> +					break;
>> +				default:
>> +					break;
>> +				}
>> +				perf_mmap__consume(map);
>> +			}
>> +			perf_mmap__read_done(map);
>> +		}
>> +	}
>> +	return NULL;
>> +}
>> +
>> +pthread_t poll_thread;
>> +
>> +int bpf_event__start_polling_thread(struct bpf_event_poll_args *args)
>> +{
>> +	struct perf_evsel *counter;
>> +
>> +	args->evlist = perf_evlist__new();
>> +
>> +	if (args->evlist == NULL)
>> +		return -1;
>> +
>> +	if (perf_evlist__create_maps(args->evlist, args->target))
> 		goto out_delete_evlist;
>> +
>> +	if (perf_evlist__add_bpf_tracker(args->evlist))
> 		goto out_delete_evlist;
>> +
>> +	evlist__for_each_entry(args->evlist, counter) {
>> +		if (perf_evsel__open(counter, args->evlist->cpus,
>> +				     args->evlist->threads) < 0)
> 			goto out_delete_evlist;
>> +	}
>> +
>> +	if (perf_evlist__mmap(args->evlist, UINT_MAX))
> 		goto out_delete_evlist;
>> +
>> +	evlist__for_each_entry(args->evlist, counter) {
>> +		if (perf_evsel__enable(counter))
> 			goto out_delete_evlist;
>> +	}
>> +
>> +	if (pthread_create(&poll_thread, NULL, bpf_poll_thread, args))
> 		goto out_delete_evlist;	
>> +
>> +	return 0;
> out_delete_evlist:
> 	perf_evlist__delete(args->evlist);
> 	args->evlist = NULL;
> 
> 	return -1;
>> +}
>> +
>> +void bpf_event__stop_polling_thread(struct bpf_event_poll_args *args)
>> +{
>> +	pthread_join(poll_thread, NULL);
>> +	perf_evlist__exit(args->evlist);
>> +}
>> diff --git a/tools/perf/util/bpf-event.h b/tools/perf/util/bpf-event.h
>> index c4f0f1395ea5..61914827c1e3 100644
>> --- a/tools/perf/util/bpf-event.h
>> +++ b/tools/perf/util/bpf-event.h
>> @@ -12,12 +12,17 @@
>> #include <bpf/libbpf.h>
>> #include <linux/btf.h>
>> #include <linux/rbtree.h>
>> +#include <pthread.h>
>> +#include <api/fd/array.h>
>> #include "event.h"
>> 
>> struct machine;
>> union perf_event;
>> +struct perf_env;
>> struct perf_sample;
>> struct record_opts;
>> +struct evlist;
>> +struct target;
>> 
>> struct bpf_prog_info_node {
>> 	struct bpf_prog_info_linear	*info_linear;
>> @@ -31,6 +36,13 @@ struct btf_node {
>> 	char		data[];
>> };
>> 
>> +struct bpf_event_poll_args {
>> +	struct perf_env		*env;
>> +	struct perf_evlist	*evlist;
>> +	struct target		*target;
>> +	volatile int		*done;
>> +};
>> +
>> #ifdef HAVE_LIBBPF_SUPPORT
>> int machine__process_bpf_event(struct machine *machine, union perf_event *event,
>> 			       struct perf_sample *sample);
>> @@ -39,6 +51,8 @@ int perf_event__synthesize_bpf_events(struct perf_session *session,
>> 				      perf_event__handler_t process,
>> 				      struct machine *machine,
>> 				      struct record_opts *opts);
>> +int bpf_event__start_polling_thread(struct bpf_event_poll_args *args);
>> +void bpf_event__stop_polling_thread(struct bpf_event_poll_args *args);
>> #else
>> static inline int machine__process_bpf_event(struct machine *machine __maybe_unused,
>> 					     union perf_event *event __maybe_unused,
>> @@ -54,5 +68,13 @@ static inline int perf_event__synthesize_bpf_events(struct perf_session *session
>> {
>> 	return 0;
>> }
>> +
>> +static inline int bpf_event__start_polling_thread(struct bpf_event_poll_args *args __maybe_unused)
>> +{
>> +	return 0;
>> +}
>> +void bpf_event__stop_polling_thread(struct bpf_event_poll_args *args __maybe_unused)
>> +{
>> +}
>> #endif // HAVE_LIBBPF_SUPPORT
>> #endif
>> diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c
>> index 8c902276d4b4..612c079579ce 100644
>> --- a/tools/perf/util/evlist.c
>> +++ b/tools/perf/util/evlist.c
>> @@ -271,6 +271,26 @@ int perf_evlist__add_dummy(struct perf_evlist *evlist)
>> 	return 0;
>> }
>> 
>> +int perf_evlist__add_bpf_tracker(struct perf_evlist *evlist)
>> +{
>> +	struct perf_event_attr attr = {
>> +		.type	          = PERF_TYPE_SOFTWARE,
>> +		.config           = PERF_COUNT_SW_DUMMY,
>> +		.watermark        = 1,
>> +		.bpf_event        = 1,
>> +		.wakeup_watermark = 1,
>> +		.size	   = sizeof(attr), /* to capture ABI version */
>> +	};
>> +	struct perf_evsel *evsel = perf_evsel__new_idx(&attr,
>> +						       evlist->nr_entries);
>> +
>> +	if (evsel == NULL)
>> +		return -ENOMEM;
>> +
>> +	perf_evlist__add(evlist, evsel);
> 
> You could use:
> 
> 	struct perf_evlist *evlist = perf_evlist__new_dummy();
> 	if (evlist != NULL) {
> 		struct perf_evsel *evsel == perf_evlist__first(evlist);
> 		evsel->attr.bpf_event = evsel->attr.watermark = evsel->attr.wakeup_watermark = 1;
> 		return 0;
> 	}
> 	return -1;

This looks cleaner. Let me fix in next version. 

> 
> Because in this case all you'll have in this evlist is the bpf tracker,
> right? The add_bpf_tracker would be handy if we would want to have a
> pre-existing evlist with some other events and wanted to add a bpf
> tracker, no?

I think all we need is a side-band evlist instead of the main evlist. May
be we should call it side-band evlist, and make it more generic?

Thanks,
Song

> 
> - Arnaldo
> 
>> +	return 0;
>> +}
>> +
>> static int perf_evlist__add_attrs(struct perf_evlist *evlist,
>> 				  struct perf_event_attr *attrs, size_t nr_attrs)
>> {
>> diff --git a/tools/perf/util/evlist.h b/tools/perf/util/evlist.h
>> index 868294491194..a2d22715188e 100644
>> --- a/tools/perf/util/evlist.h
>> +++ b/tools/perf/util/evlist.h
>> @@ -84,6 +84,8 @@ int __perf_evlist__add_default_attrs(struct perf_evlist *evlist,
>> 
>> int perf_evlist__add_dummy(struct perf_evlist *evlist);
>> 
>> +int perf_evlist__add_bpf_tracker(struct perf_evlist *evlist);
>> +
>> int perf_evlist__add_newtp(struct perf_evlist *evlist,
>> 			   const char *sys, const char *name, void *handler);
>> 
>> -- 
>> 2.17.1

Re: [PATCH v2 perf,bpf 08/11] perf, bpf: save btf information as headers to perf.data

[Song Liu]

> On Feb 15, 2019, at 6:26 AM, Arnaldo Carvalho de Melo <acme@redhat.com> wrote:
> 
> Em Thu, Feb 14, 2019 at 04:00:09PM -0800, Song Liu escreveu:
>> This patch enables perf-record to save btf information as headers to
>> perf.data A new header type HEADER_BTF is introduced for this data.
> 
> Jiri,
> 
> 	Wouldn't it be better for this HEADER_BTF to be introduced
> already as an user space event, Song, see:
> 
>  tools/perf/util/event.h
> 
> and:
> 
>  tools/perf/util/event.c
> 
> perf_event__synthesize_cpu_map()
> 
> - Arnaldo
> 

BTF would be short living for short living BPF programs. I guess 
saving them as header is easier than merging them with samples. 

What's the benefit of saving them as user space events?

Thanks,
Song


> 
>> Signed-off-by: Song Liu <songliubraving@fb.com>
>> ---
>> tools/perf/util/header.c | 99 +++++++++++++++++++++++++++++++++++++++-
>> tools/perf/util/header.h |  1 +
>> 2 files changed, 99 insertions(+), 1 deletion(-)
>> 
>> diff --git a/tools/perf/util/header.c b/tools/perf/util/header.c
>> index 2ae76a9d06f6..3f1562afe8e5 100644
>> --- a/tools/perf/util/header.c
>> +++ b/tools/perf/util/header.c
>> @@ -1125,6 +1125,45 @@ static int write_bpf_prog_info(struct feat_fd *ff,
>> 	return ret;
>> }
>> 
>> +static int write_btf(struct feat_fd *ff,
>> +		     struct perf_evlist *evlist __maybe_unused)
>> +{
>> +	struct perf_env *env = &ff->ph->env;
>> +	struct rb_root *root;
>> +	struct rb_node *next;
>> +	u32 count = 0;
>> +	int ret;
>> +
>> +	down_read(&env->bpf_info_lock);
>> +
>> +	root = &env->btfs;
>> +	next = rb_first(root);
>> +	while (next) {
>> +		++count;
>> +		next = rb_next(next);
>> +	}
>> +
>> +	ret = do_write(ff, &count, sizeof(count));
>> +
>> +	if (ret < 0)
>> +		goto out;
>> +
>> +	next = rb_first(root);
>> +	while (next) {
>> +		struct btf_node *node;
>> +
>> +		node = rb_entry(next, struct btf_node, rb_node);
>> +		next = rb_next(&node->rb_node);
>> +		ret = do_write(ff, node,
>> +			       sizeof(struct btf_node) + node->data_size);
>> +		if (ret < 0)
>> +			goto out;
>> +	}
>> +out:
>> +	up_read(&env->bpf_info_lock);
>> +	return ret;
>> +}
>> +
>> static int cpu_cache_level__sort(const void *a, const void *b)
>> {
>> 	struct cpu_cache_level *cache_a = (struct cpu_cache_level *)a;
>> @@ -1628,6 +1667,28 @@ static void print_bpf_prog_info(struct feat_fd *ff, FILE *fp)
>> 	up_read(&env->bpf_info_lock);
>> }
>> 
>> +static void print_btf(struct feat_fd *ff, FILE *fp)
>> +{
>> +	struct perf_env *env = &ff->ph->env;
>> +	struct rb_root *root;
>> +	struct rb_node *next;
>> +
>> +	down_read(&env->bpf_info_lock);
>> +
>> +	root = &env->btfs;
>> +	next = rb_first(root);
>> +
>> +	while (next) {
>> +		struct btf_node *node;
>> +
>> +		node = rb_entry(next, struct btf_node, rb_node);
>> +		next = rb_next(&node->rb_node);
>> +		fprintf(fp, "# bpf_prog_info of id %u\n", node->id);
>> +	}
>> +
>> +	up_read(&env->bpf_info_lock);
>> +}
>> +
>> static void free_event_desc(struct perf_evsel *events)
>> {
>> 	struct perf_evsel *evsel;
>> @@ -2723,6 +2784,41 @@ static int process_bpf_prog_info(struct feat_fd *ff,
>> 	return err;
>> }
>> 
>> +static int process_btf(struct feat_fd *ff, void *data __maybe_unused)
>> +{
>> +	struct perf_env *env = &ff->ph->env;
>> +	u32 count, i;
>> +
>> +	if (do_read_u32(ff, &count))
>> +		return -1;
>> +
>> +	down_write(&env->bpf_info_lock);
>> +
>> +	for (i = 0; i < count; ++i) {
>> +		struct btf_node btf_node;
>> +		struct btf_node *node;
>> +
>> +		if (__do_read(ff, &btf_node, sizeof(struct btf_node)))
>> +			return -1;
>> +
>> +		node = malloc(sizeof(struct btf_node) + btf_node.data_size);
>> +		if (!node)
>> +			return -1;
>> +
>> +		node->id = btf_node.id;
>> +		node->data_size = btf_node.data_size;
>> +
>> +		if (__do_read(ff, node->data, btf_node.data_size)) {
>> +			free(node);
>> +			return -1;
>> +		}
>> +		perf_env__insert_btf(env, node);
>> +	}
>> +
>> +	up_write(&env->bpf_info_lock);
>> +	return 0;
>> +}
>> +
>> struct feature_ops {
>> 	int (*write)(struct feat_fd *ff, struct perf_evlist *evlist);
>> 	void (*print)(struct feat_fd *ff, FILE *fp);
>> @@ -2783,7 +2879,8 @@ static const struct feature_ops feat_ops[HEADER_LAST_FEATURE] = {
>> 	FEAT_OPR(SAMPLE_TIME,	sample_time,	false),
>> 	FEAT_OPR(MEM_TOPOLOGY,	mem_topology,	true),
>> 	FEAT_OPR(CLOCKID,       clockid,        false),
>> -	FEAT_OPR(BPF_PROG_INFO, bpf_prog_info,  false)
>> +	FEAT_OPR(BPF_PROG_INFO, bpf_prog_info,  false),
>> +	FEAT_OPR(BTF,           btf,            false)
>> };
>> 
>> struct header_print_data {
>> diff --git a/tools/perf/util/header.h b/tools/perf/util/header.h
>> index 0785c91b4c3a..ba51d8e43c53 100644
>> --- a/tools/perf/util/header.h
>> +++ b/tools/perf/util/header.h
>> @@ -40,6 +40,7 @@ enum {
>> 	HEADER_MEM_TOPOLOGY,
>> 	HEADER_CLOCKID,
>> 	HEADER_BPF_PROG_INFO,
>> +	HEADER_BTF,
>> 	HEADER_LAST_FEATURE,
>> 	HEADER_FEAT_BITS	= 256,
>> };
>> -- 
>> 2.17.1

Re: [PATCH v2 perf,bpf 08/11] perf, bpf: save btf information as headers to perf.data

[Arnaldo Carvalho de Melo]

Em Fri, Feb 15, 2019 at 05:25:01PM +0000, Song Liu escreveu:
> > On Feb 15, 2019, at 6:26 AM, Arnaldo Carvalho de Melo <acme@redhat.com> wrote:
> > Em Thu, Feb 14, 2019 at 04:00:09PM -0800, Song Liu escreveu:
> >> This patch enables perf-record to save btf information as headers to
> >> perf.data A new header type HEADER_BTF is introduced for this data.

> > 	Wouldn't it be better for this HEADER_BTF to be introduced
> > already as an user space event, Song, see:

> >  tools/perf/util/event.h

> > and:
 
> >  tools/perf/util/event.c
 
> > perf_event__synthesize_cpu_map()
 
> BTF would be short living for short living BPF programs. I guess 
> saving them as header is easier than merging them with samples. 
 
> What's the benefit of saving them as user space events?

When we work with pipe mode, i.e.:

	perf record -o - | perf report -i -

and other combinations (with 'perf script', 'perf inject', etc), we need
a way to pass the headers to the other side, and the way was via user
space events.

This is something Stephane and Jiri have been discussing recently,
probably they have more justifications, Stephane, Jiri?

- Arnaldo

Re: [PATCH v2 perf,bpf 08/11] perf, bpf: save btf information as headers to perf.data

[Song Liu]

> On Feb 15, 2019, at 9:40 AM, Arnaldo Carvalho de Melo <acme@kernel.org> wrote:
> 
> Em Fri, Feb 15, 2019 at 05:25:01PM +0000, Song Liu escreveu:
>>> On Feb 15, 2019, at 6:26 AM, Arnaldo Carvalho de Melo <acme@redhat.com> wrote:
>>> Em Thu, Feb 14, 2019 at 04:00:09PM -0800, Song Liu escreveu:
>>>> This patch enables perf-record to save btf information as headers to
>>>> perf.data A new header type HEADER_BTF is introduced for this data.
> 
>>> 	Wouldn't it be better for this HEADER_BTF to be introduced
>>> already as an user space event, Song, see:
> 
>>> tools/perf/util/event.h
> 
>>> and:
> 
>>> tools/perf/util/event.c
> 
>>> perf_event__synthesize_cpu_map()
> 
>> BTF would be short living for short living BPF programs. I guess 
>> saving them as header is easier than merging them with samples. 
> 
>> What's the benefit of saving them as user space events?
> 
> When we work with pipe mode, i.e.:
> 
> 	perf record -o - | perf report -i -
> 
> and other combinations (with 'perf script', 'perf inject', etc), we need
> a way to pass the headers to the other side, and the way was via user
> space events.
> 
> This is something Stephane and Jiri have been discussing recently,
> probably they have more justifications, Stephane, Jiri?
> 
> - Arnaldo

I see. In this case, we will need some synchronization between main
thread and the polling thread, as they are both writing to the same
pipe. 

Thanks,
Song

Re: [PATCH v2 perf,bpf 01/11] perf, bpf: consider events with attr.bpf_event as side-band events

[Arnaldo Carvalho de Melo]

Em Fri, Feb 15, 2019 at 05:02:02PM +0000, Song Liu escreveu:
> 
> 
> > On Feb 15, 2019, at 6:14 AM, Arnaldo Carvalho de Melo <acme@redhat.com> wrote:
> > 
> > Em Thu, Feb 14, 2019 at 03:58:40PM -0800, Song Liu escreveu:
> >> Events with bpf_event should be considered as side-band event, as they
> >> carry information about BPF programs.
> >> 
> >> Fixes: 6ee52e2a3fe4 ("perf, bpf: Introduce PERF_RECORD_BPF_EVENT")
> >> Signed-off-by: Song Liu <songliubraving@fb.com>
> >> ---
> >> kernel/events/core.c | 3 ++-
> >> 1 file changed, 2 insertions(+), 1 deletion(-)
> >> 
> >> diff --git a/kernel/events/core.c b/kernel/events/core.c
> >> index 0a8dab322111..9403bdda5f8c 100644
> >> --- a/kernel/events/core.c
> >> +++ b/kernel/events/core.c
> >> @@ -4238,7 +4238,8 @@ static bool is_sb_event(struct perf_event *event)
> >> 	if (attr->mmap || attr->mmap_data || attr->mmap2 ||
> >> 	    attr->comm || attr->comm_exec ||
> >> 	    attr->task || attr->ksymbol ||
> 				^ here

how come, I looked after context_switch, that came before, have to
replace my glasses... :-)
> 
> >> -	    attr->context_switch)
> >> +	    attr->context_switch ||
> >> +	    attr->bpf_event)
> > 
> > What about attr->ksymbol?
 
> It is already included up there. attr->bpf_event was missed initially 
> because we didn't use events with bpf_event but not ksymbol. With this 
> set, the polling thread only has bpf_event, so we need this fix. 

Ok, I'll get this one merged asap.

- Arnaldo

Re: [PATCH v2 perf,bpf 09/11] perf-top: add option --bpf-event

[Arnaldo Carvalho de Melo]

Em Fri, Feb 15, 2019 at 05:06:19PM +0000, Song Liu escreveu:
> > On Feb 15, 2019, at 6:27 AM, Arnaldo Carvalho de Melo <acme@redhat.com> wrote:
> > Em Thu, Feb 14, 2019 at 04:00:10PM -0800, Song Liu escreveu:
> >> bpf events are only tracked when opts->bpf_event is enabled. This patch
> >> adds command line flag to enable this for perf-top.
 
> > Shouldn't this start as enabled and we just provide a way to disable it
> > for testing purposes? Normally perf top works system wide, and then we
> > want to get info about BPF events, I think.

> We can enable it by default, and add a --no-bpf-event option to disable it
> for testing. 

Excellent, thanks.

- Arnaldo

Re: [PATCH v2 perf,bpf 11/11] perf, bpf: save information about short living bpf programs

[Arnaldo Carvalho de Melo]

Em Fri, Feb 15, 2019 at 05:13:01PM +0000, Song Liu escreveu:
> > On Feb 15, 2019, at 6:41 AM, Arnaldo Carvalho de Melo <acme@redhat.com> wrote:
> > Em Thu, Feb 14, 2019 at 04:00:45PM -0800, Song Liu escreveu:
> >> +pthread_t poll_thread;
> >> +
> >> +int bpf_event__start_polling_thread(struct bpf_event_poll_args *args)
> >> +{
> >> +	struct perf_evsel *counter;
> >> +
> >> +	args->evlist = perf_evlist__new();
> >> +
> >> +	if (args->evlist == NULL)
> >> +		return -1;
> >> +
> >> +	if (perf_evlist__create_maps(args->evlist, args->target))
> > 		goto out_delete_evlist;
> >> +
> >> +	if (perf_evlist__add_bpf_tracker(args->evlist))
> > 		goto out_delete_evlist;
> >> +
> >> +	evlist__for_each_entry(args->evlist, counter) {
> >> +		if (perf_evsel__open(counter, args->evlist->cpus,
> >> +				     args->evlist->threads) < 0)
> > 			goto out_delete_evlist;
> >> +	}
> >> +
> >> +	if (perf_evlist__mmap(args->evlist, UINT_MAX))
> > 		goto out_delete_evlist;
> >> +
> >> +	evlist__for_each_entry(args->evlist, counter) {
> >> +		if (perf_evsel__enable(counter))
> > 			goto out_delete_evlist;
> >> +	}
> >> +
> >> +	if (pthread_create(&poll_thread, NULL, bpf_poll_thread, args))
> > 		goto out_delete_evlist;	
> >> +
> >> +	return 0;
> > out_delete_evlist:
> > 	perf_evlist__delete(args->evlist);
> > 	args->evlist = NULL;

Have you seen the error handling suggestion above?

> >> +int perf_evlist__add_bpf_tracker(struct perf_evlist *evlist)
> >> +{
> >> +	struct perf_event_attr attr = {
> >> +		.type	          = PERF_TYPE_SOFTWARE,
> >> +		.config           = PERF_COUNT_SW_DUMMY,
> >> +		.watermark        = 1,
> >> +		.bpf_event        = 1,
> >> +		.wakeup_watermark = 1,
> >> +		.size	   = sizeof(attr), /* to capture ABI version */
> >> +	};
> >> +	struct perf_evsel *evsel = perf_evsel__new_idx(&attr,
> >> +						       evlist->nr_entries);
> >> +
> >> +	if (evsel == NULL)
> >> +		return -ENOMEM;
> >> +
> >> +	perf_evlist__add(evlist, evsel);

> > You could use:
 
> > 	struct perf_evlist *evlist = perf_evlist__new_dummy();
> > 	if (evlist != NULL) {
> > 		struct perf_evsel *evsel == perf_evlist__first(evlist);
> > 		evsel->attr.bpf_event = evsel->attr.watermark = evsel->attr.wakeup_watermark = 1;
> > 		return 0;
> > 	}
> > 	return -1;
 
> This looks cleaner. Let me fix in next version. 

> > Because in this case all you'll have in this evlist is the bpf tracker,
> > right? The add_bpf_tracker would be handy if we would want to have a
> > pre-existing evlist with some other events and wanted to add a bpf
> > tracker, no?

> I think all we need is a side-band evlist instead of the main evlist. May
> be we should call it side-band evlist, and make it more generic?

Sure, you could for instance have something like:

	struct perf_event_attr attr = {
		.watermark        = 1,
		.bpf_event        = 1,
		.wakeup_watermark = 1,
	}
 	struct perf_evlist *evlist = perf_evlist__new_side_band(&attr);


And the other details will be set by it, i.e. the .config

	.type	          = PERF_TYPE_SOFTWARE,
	.config           = PERF_COUNT_SW_DUMMY,
	.size	   = sizeof(attr), /* to capture ABI version */

And the idx arg.

- Arnaldo

Re: [PATCH v2 perf,bpf 11/11] perf, bpf: save information about short living bpf programs

[Song Liu]

> On Feb 15, 2019, at 10:09 AM, Arnaldo Carvalho de Melo <acme@redhat.com> wrote:
> 
> Em Fri, Feb 15, 2019 at 05:13:01PM +0000, Song Liu escreveu:
>>> On Feb 15, 2019, at 6:41 AM, Arnaldo Carvalho de Melo <acme@redhat.com> wrote:
>>> Em Thu, Feb 14, 2019 at 04:00:45PM -0800, Song Liu escreveu:
>>>> +pthread_t poll_thread;
>>>> +
>>>> +int bpf_event__start_polling_thread(struct bpf_event_poll_args *args)
>>>> +{
>>>> +	struct perf_evsel *counter;
>>>> +
>>>> +	args->evlist = perf_evlist__new();
>>>> +
>>>> +	if (args->evlist == NULL)
>>>> +		return -1;
>>>> +
>>>> +	if (perf_evlist__create_maps(args->evlist, args->target))
>>> 		goto out_delete_evlist;
>>>> +
>>>> +	if (perf_evlist__add_bpf_tracker(args->evlist))
>>> 		goto out_delete_evlist;
>>>> +
>>>> +	evlist__for_each_entry(args->evlist, counter) {
>>>> +		if (perf_evsel__open(counter, args->evlist->cpus,
>>>> +				     args->evlist->threads) < 0)
>>> 			goto out_delete_evlist;
>>>> +	}
>>>> +
>>>> +	if (perf_evlist__mmap(args->evlist, UINT_MAX))
>>> 		goto out_delete_evlist;
>>>> +
>>>> +	evlist__for_each_entry(args->evlist, counter) {
>>>> +		if (perf_evsel__enable(counter))
>>> 			goto out_delete_evlist;
>>>> +	}
>>>> +
>>>> +	if (pthread_create(&poll_thread, NULL, bpf_poll_thread, args))
>>> 		goto out_delete_evlist;	
>>>> +
>>>> +	return 0;
>>> out_delete_evlist:
>>> 	perf_evlist__delete(args->evlist);
>>> 	args->evlist = NULL;
> 
> Have you seen the error handling suggestion above?

Yes! I will include these changes, as always. :)

> 
>>>> +int perf_evlist__add_bpf_tracker(struct perf_evlist *evlist)
>>>> +{
>>>> +	struct perf_event_attr attr = {
>>>> +		.type	          = PERF_TYPE_SOFTWARE,
>>>> +		.config           = PERF_COUNT_SW_DUMMY,
>>>> +		.watermark        = 1,
>>>> +		.bpf_event        = 1,
>>>> +		.wakeup_watermark = 1,
>>>> +		.size	   = sizeof(attr), /* to capture ABI version */
>>>> +	};
>>>> +	struct perf_evsel *evsel = perf_evsel__new_idx(&attr,
>>>> +						       evlist->nr_entries);
>>>> +
>>>> +	if (evsel == NULL)
>>>> +		return -ENOMEM;
>>>> +
>>>> +	perf_evlist__add(evlist, evsel);
> 
>>> You could use:
> 
>>> 	struct perf_evlist *evlist = perf_evlist__new_dummy();
>>> 	if (evlist != NULL) {
>>> 		struct perf_evsel *evsel == perf_evlist__first(evlist);
>>> 		evsel->attr.bpf_event = evsel->attr.watermark = evsel->attr.wakeup_watermark = 1;
>>> 		return 0;
>>> 	}
>>> 	return -1;
> 
>> This looks cleaner. Let me fix in next version. 
> 
>>> Because in this case all you'll have in this evlist is the bpf tracker,
>>> right? The add_bpf_tracker would be handy if we would want to have a
>>> pre-existing evlist with some other events and wanted to add a bpf
>>> tracker, no?
> 
>> I think all we need is a side-band evlist instead of the main evlist. May
>> be we should call it side-band evlist, and make it more generic?
> 
> Sure, you could for instance have something like:
> 
> 	struct perf_event_attr attr = {
> 		.watermark        = 1,
> 		.bpf_event        = 1,
> 		.wakeup_watermark = 1,
> 	}
> 	struct perf_evlist *evlist = perf_evlist__new_side_band(&attr);
> 
> 
> And the other details will be set by it, i.e. the .config
> 
> 	.type	          = PERF_TYPE_SOFTWARE,
> 	.config           = PERF_COUNT_SW_DUMMY,
> 	.size	   = sizeof(attr), /* to capture ABI version */
> 
> And the idx arg.
> 
> - Arnaldo

This looks good. Let me revise the patch in that direction.

Thanks,
Song

Re: [PATCH v2 perf,bpf 08/11] perf, bpf: save btf information as headers to perf.data

[Arnaldo Carvalho de Melo]

Em Fri, Feb 15, 2019 at 05:47:58PM +0000, Song Liu escreveu:
> 
> 
> > On Feb 15, 2019, at 9:40 AM, Arnaldo Carvalho de Melo <acme@kernel.org> wrote:
> > 
> > Em Fri, Feb 15, 2019 at 05:25:01PM +0000, Song Liu escreveu:
> >>> On Feb 15, 2019, at 6:26 AM, Arnaldo Carvalho de Melo <acme@redhat.com> wrote:
> >>> Em Thu, Feb 14, 2019 at 04:00:09PM -0800, Song Liu escreveu:
> >>>> This patch enables perf-record to save btf information as headers to
> >>>> perf.data A new header type HEADER_BTF is introduced for this data.
> > 
> >>> 	Wouldn't it be better for this HEADER_BTF to be introduced
> >>> already as an user space event, Song, see:
> > 
> >>> tools/perf/util/event.h
> > 
> >>> and:
> > 
> >>> tools/perf/util/event.c
> > 
> >>> perf_event__synthesize_cpu_map()
> > 
> >> BTF would be short living for short living BPF programs. I guess 
> >> saving them as header is easier than merging them with samples. 
> > 
> >> What's the benefit of saving them as user space events?
> > 
> > When we work with pipe mode, i.e.:
> > 
> > 	perf record -o - | perf report -i -
> > 
> > and other combinations (with 'perf script', 'perf inject', etc), we need
> > a way to pass the headers to the other side, and the way was via user
> > space events.
> > 
> > This is something Stephane and Jiri have been discussing recently,
> > probably they have more justifications, Stephane, Jiri?
> > 
> > - Arnaldo
> 
> I see. In this case, we will need some synchronization between main
> thread and the polling thread, as they are both writing to the same
> pipe. 

So, the whole context is that we need to have 'perf record' to start a
thread per CPU and then read the already per-cpu mmap buffers in the
matching thread, with the right affinity, numa settings to have the
record phase not cause contention, etc, so it ends up dumping one stream
per CPU in a separate file in a 'perf.data' directory instead of a
perf.data file.

Jiri is working on that, so, if you dump one more stream into that
directory, it would, at post processing time, be ordered together with
the other stream, the per-cpu ones.

- Arnaldo

Re: [PATCH v2 perf,bpf 08/11] perf, bpf: save btf information as headers to perf.data

[Song Liu]

> On Feb 15, 2019, at 10:20 AM, Arnaldo Carvalho de Melo <acme@kernel.org> wrote:
> 
> Em Fri, Feb 15, 2019 at 05:47:58PM +0000, Song Liu escreveu:
>> 
>> 
>>> On Feb 15, 2019, at 9:40 AM, Arnaldo Carvalho de Melo <acme@kernel.org> wrote:
>>> 
>>> Em Fri, Feb 15, 2019 at 05:25:01PM +0000, Song Liu escreveu:
>>>>> On Feb 15, 2019, at 6:26 AM, Arnaldo Carvalho de Melo <acme@redhat.com> wrote:
>>>>> Em Thu, Feb 14, 2019 at 04:00:09PM -0800, Song Liu escreveu:
>>>>>> This patch enables perf-record to save btf information as headers to
>>>>>> perf.data A new header type HEADER_BTF is introduced for this data.
>>> 
>>>>> 	Wouldn't it be better for this HEADER_BTF to be introduced
>>>>> already as an user space event, Song, see:
>>> 
>>>>> tools/perf/util/event.h
>>> 
>>>>> and:
>>> 
>>>>> tools/perf/util/event.c
>>> 
>>>>> perf_event__synthesize_cpu_map()
>>> 
>>>> BTF would be short living for short living BPF programs. I guess 
>>>> saving them as header is easier than merging them with samples. 
>>> 
>>>> What's the benefit of saving them as user space events?
>>> 
>>> When we work with pipe mode, i.e.:
>>> 
>>> 	perf record -o - | perf report -i -
>>> 
>>> and other combinations (with 'perf script', 'perf inject', etc), we need
>>> a way to pass the headers to the other side, and the way was via user
>>> space events.
>>> 
>>> This is something Stephane and Jiri have been discussing recently,
>>> probably they have more justifications, Stephane, Jiri?
>>> 
>>> - Arnaldo
>> 
>> I see. In this case, we will need some synchronization between main
>> thread and the polling thread, as they are both writing to the same
>> pipe. 
> 
> So, the whole context is that we need to have 'perf record' to start a
> thread per CPU and then read the already per-cpu mmap buffers in the
> matching thread, with the right affinity, numa settings to have the
> record phase not cause contention, etc, so it ends up dumping one stream
> per CPU in a separate file in a 'perf.data' directory instead of a
> perf.data file.
> 
> Jiri is working on that, so, if you dump one more stream into that
> directory, it would, at post processing time, be ordered together with
> the other stream, the per-cpu ones.
> 
> - Arnaldo

I see. This solution looks great. 

For this set, how about I keep this part as-is (at least for v3)? In 
this case, it will goes to the header file after Jiri's change. Once 
Jiri's work is done, I will move them into per-cpu files. 

Thanks,
Song