| Message ID | 20230405132120.59886-4-quentin@isovalent.com (mailing list archive) |
|---|---|
| State | Accepted |
| Commit | 9fd496848b1c4cd04fee5f152bb7bcec11e1b901 |
| Delegated to: | BPF |
| Headers | show |
| Series | bpftool: Add inline annotations when dumping program CFGs | expand |
Quentin Monnet <quentin@isovalent.com> writes: > diff --git a/tools/bpf/bpftool/btf_dumper.c b/tools/bpf/bpftool/btf_dumper.c > index e7f6ec3a8f35..583aa843df92 100644 > --- a/tools/bpf/bpftool/btf_dumper.c > +++ b/tools/bpf/bpftool/btf_dumper.c > @@ -821,3 +821,37 @@ void btf_dump_linfo_json(const struct btf *btf, > BPF_LINE_INFO_LINE_COL(linfo->line_col)); > } > } > + > +static void dotlabel_puts(const char *s) > +{ > + for (; *s; ++s) { > + switch (*s) { > + case '\\': > + case '"': > + case '{': > + case '}': > + case '<': > + case '>': > + case '|': > + case ' ': > + putchar('\\'); > + __fallthrough; Is __fallthrough correct? I see the following compile error on s390 in linux-next (20230412): CC btf_dumper.o btf_dumper.c: In function ‘dotlabel_puts’: btf_dumper.c:838:25: error: ‘__fallthrough’ undeclared (first use in this function); did you mean ‘fallthrough’? 838 | __fallthrough; | ^~~~~~~~~~~~~ removing the two underscores fixes this. > + default: > + putchar(*s); > + } > + } > +} Thanks, Sven
2023-04-12 08:04 UTC+0200 ~ Sven Schnelle <svens@linux.ibm.com> > Quentin Monnet <quentin@isovalent.com> writes: > >> diff --git a/tools/bpf/bpftool/btf_dumper.c b/tools/bpf/bpftool/btf_dumper.c >> index e7f6ec3a8f35..583aa843df92 100644 >> --- a/tools/bpf/bpftool/btf_dumper.c >> +++ b/tools/bpf/bpftool/btf_dumper.c >> @@ -821,3 +821,37 @@ void btf_dump_linfo_json(const struct btf *btf, >> BPF_LINE_INFO_LINE_COL(linfo->line_col)); >> } >> } >> + >> +static void dotlabel_puts(const char *s) >> +{ >> + for (; *s; ++s) { >> + switch (*s) { >> + case '\\': >> + case '"': >> + case '{': >> + case '}': >> + case '<': >> + case '>': >> + case '|': >> + case ' ': >> + putchar('\\'); >> + __fallthrough; > > Is __fallthrough correct? I see the following compile error on s390 in > linux-next (20230412): > > CC btf_dumper.o > btf_dumper.c: In function ‘dotlabel_puts’: > btf_dumper.c:838:25: error: ‘__fallthrough’ undeclared (first use in this function); did you mean ‘fallthrough’? > 838 | __fallthrough; > | ^~~~~~~~~~~~~ > > removing the two underscores fixes this. I thought so? Perf seems to use the double underscores as well. Just "fallthrough" does not seem to be the right fix anyway, it gives me an error similar to yours on x86_64 with "fallthrough" undeclared. The definition should be pulled from tools/include/linux/compiler.h (and .../compiler-gcc.h). I thought this file would be at least included from bpftool's main.h, in turn included in btf_dumper.c. Looking at the chain of inclusions, on my system I get the following path: $ CFLAGS=-H make btf_dumper.o [...] . /root/dev/linux/tools/include/linux/bitops.h [...] .. /root/dev/linux/tools/include/linux/bits.h [...] ... /root/dev/linux/tools/include/linux/build_bug.h .... /root/dev/linux/tools/include/linux/compiler.h ..... /root/dev/linux/tools/include/linux/compiler_types.h ...... /root/dev/linux/tools/include/linux/compiler-gcc.h [...] What do you get on your side? If you add "#include <linux/compiler.h>" to btf_dumper.c directly, does it fix the issue? Quentin
Quentin Monnet <quentin@isovalent.com> writes: > 2023-04-12 08:04 UTC+0200 ~ Sven Schnelle <svens@linux.ibm.com> >> Quentin Monnet <quentin@isovalent.com> writes: >> >>> diff --git a/tools/bpf/bpftool/btf_dumper.c b/tools/bpf/bpftool/btf_dumper.c >>> index e7f6ec3a8f35..583aa843df92 100644 >>> --- a/tools/bpf/bpftool/btf_dumper.c >>> +++ b/tools/bpf/bpftool/btf_dumper.c >>> @@ -821,3 +821,37 @@ void btf_dump_linfo_json(const struct btf *btf, >>> BPF_LINE_INFO_LINE_COL(linfo->line_col)); >>> } >>> } >>> + >>> +static void dotlabel_puts(const char *s) >>> +{ >>> + for (; *s; ++s) { >>> + switch (*s) { >>> + case '\\': >>> + case '"': >>> + case '{': >>> + case '}': >>> + case '<': >>> + case '>': >>> + case '|': >>> + case ' ': >>> + putchar('\\'); >>> + __fallthrough; >> >> Is __fallthrough correct? I see the following compile error on s390 in >> linux-next (20230412): >> >> CC btf_dumper.o >> btf_dumper.c: In function ‘dotlabel_puts’: >> btf_dumper.c:838:25: error: ‘__fallthrough’ undeclared (first use in this function); did you mean ‘fallthrough’? >> 838 | __fallthrough; >> | ^~~~~~~~~~~~~ >> >> removing the two underscores fixes this. > > I thought so? Perf seems to use the double underscores as well. Just > "fallthrough" does not seem to be the right fix anyway, it gives me an > error similar to yours on x86_64 with "fallthrough" undeclared. > > The definition should be pulled from tools/include/linux/compiler.h (and > .../compiler-gcc.h). I thought this file would be at least included from > bpftool's main.h, in turn included in btf_dumper.c. Looking at the chain > of inclusions, on my system I get the following path: > > $ CFLAGS=-H make btf_dumper.o > [...] > . /root/dev/linux/tools/include/linux/bitops.h > [...] > .. /root/dev/linux/tools/include/linux/bits.h > [...] > ... /root/dev/linux/tools/include/linux/build_bug.h > .... /root/dev/linux/tools/include/linux/compiler.h > ..... /root/dev/linux/tools/include/linux/compiler_types.h > ...... /root/dev/linux/tools/include/linux/compiler-gcc.h > [...] > > What do you get on your side? > > If you add "#include <linux/compiler.h>" to btf_dumper.c directly, does > it fix the issue? This seems to clash with: commit f7a858bffcddaaf70c71b6b656e7cc21b6107cec Author: Liam Howlett <liam.howlett@oracle.com> Date: Fri Nov 25 15:50:16 2022 +0000 tools: Rename __fallthrough to fallthrough Rename the fallthrough attribute to better align with the kernel version. Copy the definition from include/linux/compiler_attributes.h including the #else clause. Adding the #else clause allows the tools compiler.h header to drop the check for a definition entirely and keeps both definitions together. Change any __fallthrough statements to fallthrough anywhere it was used within perf. This allows other tools to use the same key word as the kernel. Which was also merged in linux-next.
2023-04-12 14:26 UTC+0200 ~ Sven Schnelle <svens@linux.ibm.com> > Quentin Monnet <quentin@isovalent.com> writes: > >> 2023-04-12 08:04 UTC+0200 ~ Sven Schnelle <svens@linux.ibm.com> >>> Quentin Monnet <quentin@isovalent.com> writes: >>> >>>> diff --git a/tools/bpf/bpftool/btf_dumper.c b/tools/bpf/bpftool/btf_dumper.c >>>> index e7f6ec3a8f35..583aa843df92 100644 >>>> --- a/tools/bpf/bpftool/btf_dumper.c >>>> +++ b/tools/bpf/bpftool/btf_dumper.c >>>> @@ -821,3 +821,37 @@ void btf_dump_linfo_json(const struct btf *btf, >>>> BPF_LINE_INFO_LINE_COL(linfo->line_col)); >>>> } >>>> } >>>> + >>>> +static void dotlabel_puts(const char *s) >>>> +{ >>>> + for (; *s; ++s) { >>>> + switch (*s) { >>>> + case '\\': >>>> + case '"': >>>> + case '{': >>>> + case '}': >>>> + case '<': >>>> + case '>': >>>> + case '|': >>>> + case ' ': >>>> + putchar('\\'); >>>> + __fallthrough; >>> >>> Is __fallthrough correct? I see the following compile error on s390 in >>> linux-next (20230412): >>> >>> CC btf_dumper.o >>> btf_dumper.c: In function ‘dotlabel_puts’: >>> btf_dumper.c:838:25: error: ‘__fallthrough’ undeclared (first use in this function); did you mean ‘fallthrough’? >>> 838 | __fallthrough; >>> | ^~~~~~~~~~~~~ >>> >>> removing the two underscores fixes this. >> >> I thought so? Perf seems to use the double underscores as well. Just >> "fallthrough" does not seem to be the right fix anyway, it gives me an >> error similar to yours on x86_64 with "fallthrough" undeclared. >> >> The definition should be pulled from tools/include/linux/compiler.h (and >> .../compiler-gcc.h). I thought this file would be at least included from >> bpftool's main.h, in turn included in btf_dumper.c. Looking at the chain >> of inclusions, on my system I get the following path: >> >> $ CFLAGS=-H make btf_dumper.o >> [...] >> . /root/dev/linux/tools/include/linux/bitops.h >> [...] >> .. /root/dev/linux/tools/include/linux/bits.h >> [...] >> ... /root/dev/linux/tools/include/linux/build_bug.h >> .... /root/dev/linux/tools/include/linux/compiler.h >> ..... /root/dev/linux/tools/include/linux/compiler_types.h >> ...... /root/dev/linux/tools/include/linux/compiler-gcc.h >> [...] >> >> What do you get on your side? >> >> If you add "#include <linux/compiler.h>" to btf_dumper.c directly, does >> it fix the issue? > > This seems to clash with: > > commit f7a858bffcddaaf70c71b6b656e7cc21b6107cec > Author: Liam Howlett <liam.howlett@oracle.com> > Date: Fri Nov 25 15:50:16 2022 +0000 > > tools: Rename __fallthrough to fallthrough > > Rename the fallthrough attribute to better align with the kernel > version. Copy the definition from include/linux/compiler_attributes.h > including the #else clause. Adding the #else clause allows the tools > compiler.h header to drop the check for a definition entirely and keeps > both definitions together. > > Change any __fallthrough statements to fallthrough anywhere it was used > within perf. > > This allows other tools to use the same key word as the kernel. > > Which was also merged in linux-next. Right, I was not aware of this commit. In that case, replacing with "fallthrough" in linux-next makes sense indeed. Thomas Richter just submitted that fix at https://lore.kernel.org/all/20230412123636.2358949-1-tmricht@linux.ibm.com/. Thanks! Quentin
diff --git a/tools/bpf/bpftool/btf_dumper.c b/tools/bpf/bpftool/btf_dumper.c index e7f6ec3a8f35..583aa843df92 100644 --- a/tools/bpf/bpftool/btf_dumper.c +++ b/tools/bpf/bpftool/btf_dumper.c @@ -821,3 +821,37 @@ void btf_dump_linfo_json(const struct btf *btf, BPF_LINE_INFO_LINE_COL(linfo->line_col)); } } + +static void dotlabel_puts(const char *s) +{ + for (; *s; ++s) { + switch (*s) { + case '\\': + case '"': + case '{': + case '}': + case '<': + case '>': + case '|': + case ' ': + putchar('\\'); + __fallthrough; + default: + putchar(*s); + } + } +} + +void btf_dump_linfo_dotlabel(const struct btf *btf, + const struct bpf_line_info *linfo) +{ + const char *line = btf__name_by_offset(btf, linfo->line_off); + + if (!line || !strlen(line)) + return; + line = ltrim(line); + + printf("; "); + dotlabel_puts(line); + printf("\\l\\\n"); +} diff --git a/tools/bpf/bpftool/cfg.c b/tools/bpf/bpftool/cfg.c index 1951219a9af7..9fdc1f0cdd6e 100644 --- a/tools/bpf/bpftool/cfg.c +++ b/tools/bpf/bpftool/cfg.c @@ -380,7 +380,8 @@ static void cfg_destroy(struct cfg *cfg) } } -static void draw_bb_node(struct func_node *func, struct bb_node *bb) +static void +draw_bb_node(struct func_node *func, struct bb_node *bb, struct dump_data *dd) { const char *shape; @@ -398,13 +399,9 @@ static void draw_bb_node(struct func_node *func, struct bb_node *bb) printf("EXIT"); } else { unsigned int start_idx; - struct dump_data dd = {}; - - printf("{"); - kernel_syms_load(&dd); + printf("{\\\n"); start_idx = bb->head - func->start; - dump_xlated_for_graph(&dd, bb->head, bb->tail, start_idx); - kernel_syms_destroy(&dd); + dump_xlated_for_graph(dd, bb->head, bb->tail, start_idx); printf("}"); } @@ -430,12 +427,12 @@ static void draw_bb_succ_edges(struct func_node *func, struct bb_node *bb) } } -static void func_output_bb_def(struct func_node *func) +static void func_output_bb_def(struct func_node *func, struct dump_data *dd) { struct bb_node *bb; list_for_each_entry(bb, &func->bbs, l) { - draw_bb_node(func, bb); + draw_bb_node(func, bb, dd); } } @@ -455,7 +452,7 @@ static void func_output_edges(struct func_node *func) func_idx, ENTRY_BLOCK_INDEX, func_idx, EXIT_BLOCK_INDEX); } -static void cfg_dump(struct cfg *cfg) +static void cfg_dump(struct cfg *cfg, struct dump_data *dd) { struct func_node *func; @@ -463,14 +460,14 @@ static void cfg_dump(struct cfg *cfg) list_for_each_entry(func, &cfg->funcs, l) { printf("subgraph \"cluster_%d\" {\n\tstyle=\"dashed\";\n\tcolor=\"black\";\n\tlabel=\"func_%d ()\";\n", func->idx, func->idx); - func_output_bb_def(func); + func_output_bb_def(func, dd); func_output_edges(func); printf("}\n"); } printf("}\n"); } -void dump_xlated_cfg(void *buf, unsigned int len) +void dump_xlated_cfg(struct dump_data *dd, void *buf, unsigned int len) { struct bpf_insn *insn = buf; struct cfg cfg; @@ -479,7 +476,7 @@ void dump_xlated_cfg(void *buf, unsigned int len) if (cfg_build(&cfg, insn, len)) return; - cfg_dump(&cfg); + cfg_dump(&cfg, dd); cfg_destroy(&cfg); } diff --git a/tools/bpf/bpftool/cfg.h b/tools/bpf/bpftool/cfg.h index e144257ea6d2..909d17e6d4c2 100644 --- a/tools/bpf/bpftool/cfg.h +++ b/tools/bpf/bpftool/cfg.h @@ -4,6 +4,8 @@ #ifndef __BPF_TOOL_CFG_H #define __BPF_TOOL_CFG_H -void dump_xlated_cfg(void *buf, unsigned int len); +#include "xlated_dumper.h" + +void dump_xlated_cfg(struct dump_data *dd, void *buf, unsigned int len); #endif /* __BPF_TOOL_CFG_H */ diff --git a/tools/bpf/bpftool/main.h b/tools/bpf/bpftool/main.h index 0ef373cef4c7..e9ee514b22d4 100644 --- a/tools/bpf/bpftool/main.h +++ b/tools/bpf/bpftool/main.h @@ -229,6 +229,8 @@ void btf_dump_linfo_plain(const struct btf *btf, const char *prefix, bool linum); void btf_dump_linfo_json(const struct btf *btf, const struct bpf_line_info *linfo, bool linum); +void btf_dump_linfo_dotlabel(const struct btf *btf, + const struct bpf_line_info *linfo); struct nlattr; struct ifinfomsg; diff --git a/tools/bpf/bpftool/prog.c b/tools/bpf/bpftool/prog.c index afbe3ec342c8..d855118f0d96 100644 --- a/tools/bpf/bpftool/prog.c +++ b/tools/bpf/bpftool/prog.c @@ -840,11 +840,6 @@ prog_dump(struct bpf_prog_info *info, enum dump_mode mode, false)) goto exit_free; } - } else if (visual) { - if (json_output) - jsonw_null(json_wtr); - else - dump_xlated_cfg(buf, member_len); } else { kernel_syms_load(&dd); dd.nr_jited_ksyms = info->nr_jited_ksyms; @@ -854,12 +849,14 @@ prog_dump(struct bpf_prog_info *info, enum dump_mode mode, 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, - linum); + if (json_output && visual) + jsonw_null(json_wtr); + else if (json_output) + dump_xlated_json(&dd, buf, member_len, opcodes, linum); + else if (visual) + dump_xlated_cfg(&dd, buf, member_len); else - dump_xlated_plain(&dd, buf, member_len, opcodes, - linum); + dump_xlated_plain(&dd, buf, member_len, opcodes, linum); kernel_syms_destroy(&dd); } diff --git a/tools/bpf/bpftool/xlated_dumper.c b/tools/bpf/bpftool/xlated_dumper.c index 3daa05d9bbb7..2ee83561b945 100644 --- a/tools/bpf/bpftool/xlated_dumper.c +++ b/tools/bpf/bpftool/xlated_dumper.c @@ -369,21 +369,51 @@ void dump_xlated_for_graph(struct dump_data *dd, void *buf_start, void *buf_end, .cb_imm = print_imm, .private_data = dd, }; + const struct bpf_prog_linfo *prog_linfo = dd->prog_linfo; + const struct bpf_line_info *last_linfo = NULL; + struct bpf_func_info *record = dd->func_info; struct bpf_insn *insn_start = buf_start; struct bpf_insn *insn_end = buf_end; struct bpf_insn *cur = insn_start; + struct btf *btf = dd->btf; bool double_insn = false; + char func_sig[1024]; for (; cur <= insn_end; cur++) { + unsigned int insn_off; + if (double_insn) { double_insn = false; continue; } double_insn = cur->code == (BPF_LD | BPF_IMM | BPF_DW); - printf("% 4d: ", (int)(cur - insn_start + start_idx)); + insn_off = (unsigned int)(cur - insn_start + start_idx); + if (btf && record) { + if (record->insn_off == insn_off) { + btf_dumper_type_only(btf, record->type_id, + func_sig, + sizeof(func_sig)); + if (func_sig[0] != '\0') + printf("; %s:\\l\\\n", func_sig); + record = (void *)record + dd->finfo_rec_size; + } + } + + if (prog_linfo) { + const struct bpf_line_info *linfo; + + linfo = bpf_prog_linfo__lfind(prog_linfo, insn_off, 0); + if (linfo && linfo != last_linfo) { + btf_dump_linfo_dotlabel(btf, linfo); + last_linfo = linfo; + } + } + + printf("%d: ", insn_off); print_bpf_insn(&cbs, cur, true); + if (cur != insn_end) - printf(" | "); + printf("| "); } }
We support dumping the control flow graph of loaded programs to the DOT format with bpftool, but so far this feature wouldn't display the source code lines available through BTF along with the eBPF bytecode. Let's add support for these annotations, to make it easier to read the graph. In prog.c, we move the call to dump_xlated_cfg() in order to pass and use the full struct dump_data, instead of creating a minimal one in draw_bb_node(). We pass the pointer to this struct down to dump_xlated_for_graph() in xlated_dumper.c, where most of the logics is added. We deal with BTF mostly like we do for plain or JSON output, except that we cannot use a "nr_skip" value to skip a given number of linfo records (we don't process the BPF instructions linearly, and apart from the root of the graph we don't know how many records we should skip, so we just store the last linfo and make sure the new one we find is different before printing it). When printing the source instructions to the label of a DOT graph node, there are a few subtleties to address. We want some special newline markers, and there are some characters that we must escape. To deal with them, we introduce a new dedicated function btf_dump_linfo_dotlabel() in btf_dumper.c. We'll reuse this function in a later commit to format the filepath, line, and column references as well. Signed-off-by: Quentin Monnet <quentin@isovalent.com> --- tools/bpf/bpftool/btf_dumper.c | 34 +++++++++++++++++++++++++++++++ tools/bpf/bpftool/cfg.c | 23 +++++++++------------ tools/bpf/bpftool/cfg.h | 4 +++- tools/bpf/bpftool/main.h | 2 ++ tools/bpf/bpftool/prog.c | 17 +++++++--------- tools/bpf/bpftool/xlated_dumper.c | 34 +++++++++++++++++++++++++++++-- 6 files changed, 88 insertions(+), 26 deletions(-)