Documentation/trace/fprobetrace.rst | 27 +- kernel/trace/trace.c | 3 +- kernel/trace/trace_fprobe.c | 247 ++++++++++++++---- .../test.d/dynevent/add_remove_fprobe.tc | 121 +++++++++ .../test.d/dynevent/fprobe_syntax_errors.tc | 13 + 5 files changed, 349 insertions(+), 62 deletions(-)
This series aims to extend fprobe with list-style filters and a clear entry/exist qualifier. Users can now specify a comma-separated symbol list with ! exclusions, and use a spec-level suffix to select probe type: - funcA*, !funcAB, funcC -> entry probes - funcA*, !funcAB, funcC:entry -> explicit entry - funcA*, !funcAB, funcC:exit -> return/exit across the whole list For compatibility, %return remains supported for single, literal symbols. When a list or wildcard is used, an explicit [GROUP/EVENT is required and autogeneration is disabled. Autogen names are kept for single-symbol specs, with wildcard sanitization. For list/wildcard forms we set ctx->funcname = NULL so BTF lookups are not attempted. The series moves parsing to the parse path, documents the new syntax, and adds selftests that accept valid list cases and reject empty tokens, stray commas, and %return mixed with lists or wildcards. Selftests also verify enable/disable flow and that entry+exit on the same set do not double-count attached functions. Help wanted: This is my first time contributing ftrace selftests. I would appreciate comments and recommendations on test structure and coverage. Basic coverage is included, but this likely needs broader testing across architectures. Feedback and additional test ideas are welcome. Changes since v2: - Introduce spec-level: :entry/:exit; reject %return with lists/wildcards - Require explict [GROUP/]EVENT for list/wildcard; keep autogen only for single literal. - Sanitize autogen names for single-symbol wildcards - Set ctx->funcname = NULL for list/wildcard to bypass BTF - Move list parsing out of __register_trace_fprobe() and into the parse path - Update docs and tracefs README and add dynevent selftests for accept/reject and enable/disable flow Link: https://lore.kernel.org/lkml/20250904103219.f4937968362bfff1ecd3f004@kernel.org/ Ryan Chung (5): docs: tracing: fprobe: document list filters and :entry/:exit tracing: fprobe: require explicit [GROUP/]EVENT for list/wildcard tracing: fprobe: support comma-separated symbols and :entry/:exit selftests/ftrace: dynevent: add reject cases for list/:entry/:exit selftests/ftrace: dynevent: add reject cases Documentation/trace/fprobetrace.rst | 27 +- kernel/trace/trace.c | 3 +- kernel/trace/trace_fprobe.c | 247 ++++++++++++++---- .../test.d/dynevent/add_remove_fprobe.tc | 121 +++++++++ .../test.d/dynevent/fprobe_syntax_errors.tc | 13 + 5 files changed, 349 insertions(+), 62 deletions(-) -- 2.43.0
Hi Ryan, Thanks for update! On Sun, 5 Oct 2025 08:46:54 +0900 Ryan Chung <seokwoo.chung130@gmail.com> wrote: > This series aims to extend fprobe with list-style filters and a clear > entry/exist qualifier. Users can now specify a comma-separated symbol > list with ! exclusions, and use a spec-level suffix to select probe > type: > > - funcA*, !funcAB, funcC -> entry probes > - funcA*, !funcAB, funcC:entry -> explicit entry > - funcA*, !funcAB, funcC:exit -> return/exit across the whole list Just a note, it should not accept spaces in the list. The space is the highest level delimiter. I hope actual implementation does not accept spaces. So something like: "funcA*,!funcAB,funcC" "funcA*,!funcAB,funcC:entry" "funcA*,!funcAB,funcC:exit" > > For compatibility, %return remains supported for single, literal > symbols. When a list or wildcard is used, an explicit [GROUP/EVENT is > required and autogeneration is disabled. Autogen names are kept for > single-symbol specs, with wildcard sanitization. For list/wildcard forms > we set ctx->funcname = NULL so BTF lookups are not attempted. OK. So "funcA*%return" and "funcA,funcB%return" will fail. > > The series moves parsing to the parse path, documents the new syntax, > and adds selftests that accept valid list cases and reject empty tokens, > stray commas, and %return mixed with lists or wildcards. Selftests also > verify enable/disable flow and that entry+exit on the same set do not > double-count attached functions. Thanks for adding selftests and document, that is important to maintain features. > > Help wanted: This is my first time contributing ftrace selftests. I > would appreciate comments and recommendations on test structure and > coverage. OK, let me review it. Thanks, > > Basic coverage is included, but this likely needs broader testing across > architectures. Feedback and additional test ideas are welcome. > > Changes since v2: > - Introduce spec-level: :entry/:exit; reject %return with > lists/wildcards > - Require explict [GROUP/]EVENT for list/wildcard; keep autogen only for > single literal. > - Sanitize autogen names for single-symbol wildcards > - Set ctx->funcname = NULL for list/wildcard to bypass BTF > - Move list parsing out of __register_trace_fprobe() and into the parse > path > - Update docs and tracefs README and add dynevent selftests for > accept/reject and enable/disable flow > > Link: https://lore.kernel.org/lkml/20250904103219.f4937968362bfff1ecd3f004@kernel.org/ > > Ryan Chung (5): > docs: tracing: fprobe: document list filters and :entry/:exit > tracing: fprobe: require explicit [GROUP/]EVENT for list/wildcard > tracing: fprobe: support comma-separated symbols and :entry/:exit > selftests/ftrace: dynevent: add reject cases for list/:entry/:exit > selftests/ftrace: dynevent: add reject cases > > Documentation/trace/fprobetrace.rst | 27 +- > kernel/trace/trace.c | 3 +- > kernel/trace/trace_fprobe.c | 247 ++++++++++++++---- > .../test.d/dynevent/add_remove_fprobe.tc | 121 +++++++++ > .../test.d/dynevent/fprobe_syntax_errors.tc | 13 + > 5 files changed, 349 insertions(+), 62 deletions(-) > > -- > 2.43.0 > -- Masami Hiramatsu (Google) <mhiramat@kernel.org>
Hi. Thanks for the review. On Wed, Oct 08, 2025 at 09:51:11AM +0900, Masami Hiramatsu wrote: > Hi Ryan, > > Thanks for update! > > On Sun, 5 Oct 2025 08:46:54 +0900 > Ryan Chung <seokwoo.chung130@gmail.com> wrote: > > > This series aims to extend fprobe with list-style filters and a clear > > entry/exist qualifier. Users can now specify a comma-separated symbol > > list with ! exclusions, and use a spec-level suffix to select probe > > type: > > > > - funcA*, !funcAB, funcC -> entry probes > > - funcA*, !funcAB, funcC:entry -> explicit entry > > - funcA*, !funcAB, funcC:exit -> return/exit across the whole list > > > Just a note, it should not accept spaces in the list. The space > is the highest level delimiter. I hope actual implementation > does not accept spaces. So something like: > > "funcA*,!funcAB,funcC" > "funcA*,!funcAB,funcC:entry" > "funcA*,!funcAB,funcC:exit" > > I see. I will adjust the code so that the parser reject any whitespace. > > > > For compatibility, %return remains supported for single, literal > > symbols. When a list or wildcard is used, an explicit [GROUP/EVENT is > > required and autogeneration is disabled. Autogen names are kept for > > single-symbol specs, with wildcard sanitization. For list/wildcard forms > > we set ctx->funcname = NULL so BTF lookups are not attempted. > > OK. So "funcA*%return" and "funcA,funcB%return" will fail. > Yes. %return is only accepted for a single literal symbol. > > > > The series moves parsing to the parse path, documents the new syntax, > > and adds selftests that accept valid list cases and reject empty tokens, > > stray commas, and %return mixed with lists or wildcards. Selftests also > > verify enable/disable flow and that entry+exit on the same set do not > > double-count attached functions. > > Thanks for adding selftests and document, that is important to maintain > features. > > > > > Help wanted: This is my first time contributing ftrace selftests. I > > would appreciate comments and recommendations on test structure and > > coverage. > > OK, let me review it. > > Thanks, > Thank you. > > > > > Basic coverage is included, but this likely needs broader testing across > > architectures. Feedback and additional test ideas are welcome. > > > > Changes since v2: > > - Introduce spec-level: :entry/:exit; reject %return with > > lists/wildcards > > - Require explict [GROUP/]EVENT for list/wildcard; keep autogen only for > > single literal. > > - Sanitize autogen names for single-symbol wildcards > > - Set ctx->funcname = NULL for list/wildcard to bypass BTF > > - Move list parsing out of __register_trace_fprobe() and into the parse > > path > > - Update docs and tracefs README and add dynevent selftests for > > accept/reject and enable/disable flow > > > > Link: https://lore.kernel.org/lkml/20250904103219.f4937968362bfff1ecd3f004@kernel.org/ > > > > Ryan Chung (5): > > docs: tracing: fprobe: document list filters and :entry/:exit > > tracing: fprobe: require explicit [GROUP/]EVENT for list/wildcard > > tracing: fprobe: support comma-separated symbols and :entry/:exit > > selftests/ftrace: dynevent: add reject cases for list/:entry/:exit > > selftests/ftrace: dynevent: add reject cases > > > > Documentation/trace/fprobetrace.rst | 27 +- > > kernel/trace/trace.c | 3 +- > > kernel/trace/trace_fprobe.c | 247 ++++++++++++++---- > > .../test.d/dynevent/add_remove_fprobe.tc | 121 +++++++++ > > .../test.d/dynevent/fprobe_syntax_errors.tc | 13 + > > 5 files changed, 349 insertions(+), 62 deletions(-) > > > > -- > > 2.43.0 > > > > > -- > Masami Hiramatsu (Google) <mhiramat@kernel.org> Best regards, Ryan Chung
© 2016 - 2025 Red Hat, Inc.