| Message ID | 20210608135503.12135-4-y.karadz@gmail.com (mailing list archive) |
|---|---|
| State | Rejected |
| Headers | show |
| Series | [1/4] libtracefs: Fix enable_disable_all() return value | expand |
[ Sending again, but this time with "Reply-all" to include the mailing list ] On Tue, 8 Jun 2021 16:55:03 +0300 "Yordan Karadzhov (VMware)" <y.karadz@gmail.com> wrote: > Currently the return value of the callback function can be used to > stop pulling data from the trace buffer, however this return value > is lost and the user has no idea if tracefs_iterate_raw_events() > terminated because there was no more data or because this was > requested from the callback function. If we propagate the return > value of the callback, this can be used in cases like the one below: > > int callback(struct tep_event *event, struct tep_record *record, > int cpu, void *py_func) > { > .... > > return (something) ? 0 : 1 > } > > int main() > { > int ret; > > .... > > while(ret == 0) > ret = tracefs_iterate_raw_events(tep, instance, > NULL, 0, > callback, NULL); > > .... > > Here the user can effectively terminate the pulling the data > from inside the callback. With this change, the user can't tell if returned -1 due to an error or because the callback ended early, (which is not considered an error). The proper way for an application to handle this, is to pass in a context structure, and have the callback set a value if it exits early or not. I've done this already for needing the same information. No change to libtracefs is needed. This functionality is already available with the current design: struct my_struct { bool stopped; }; int callback(struct tep_event *event, struct tep_record *record, int cpu, void *data) { struct my_struct *my_data = data; [..] if (condition) { my_data->stopped = true; return 1; } return 0; } int main() { struct my_struct my_data = { .stopped = false }; int ret = 0; while (!ret && !my_data.stopped) ret = tracefs_iterate_raw_events(tep, instance, NULL, 0, callback, &my_data); } Now, if you are using this within python, and you want the python wrapper to pass data as well, you just need to add that to the struct: struct my_struct { bool stopped; func *python_callback; void *python_data; } int callback(struct tep_event *event, struct tep_record *record, int cpu, void *data) { struct my_struct *my_data = data; int ret; [..] ret = my_data->python_callback(event, record, cpu, data->python_data); if (ret) { my_data->stopped = true; return 1; } return 0; } int python_iterator(pthyon_callback, python_data) { struct my_struct my_data = { .stopped = false }; int ret = 0; my_data.python_data = python_data; my_data.python_callback = python_callback; while (!ret && !my_data.stopped) ret = tracefs_iterate_raw_events(tep, instance, NULL, 0, callback, &my_data); } So this patch isn't needed. -- Steve
On 9.06.21 г. 20:00, Steven Rostedt wrote: > [ Sending again, but this time with "Reply-all" to include the mailing list ] > > On Tue, 8 Jun 2021 16:55:03 +0300 > "Yordan Karadzhov (VMware)" <y.karadz@gmail.com> wrote: > >> Currently the return value of the callback function can be used to >> stop pulling data from the trace buffer, however this return value >> is lost and the user has no idea if tracefs_iterate_raw_events() >> terminated because there was no more data or because this was >> requested from the callback function. If we propagate the return >> value of the callback, this can be used in cases like the one below: >> >> int callback(struct tep_event *event, struct tep_record *record, >> int cpu, void *py_func) >> { >> .... >> >> return (something) ? 0 : 1 >> } >> >> int main() >> { >> int ret; >> >> .... >> >> while(ret == 0) >> ret = tracefs_iterate_raw_events(tep, instance, >> NULL, 0, >> callback, NULL); >> >> .... >> >> Here the user can effectively terminate the pulling the data >> from inside the callback. > > With this change, the user can't tell if returned -1 due to an error or > because the callback ended early, (which is not considered an error). > > The proper way for an application to handle this, is to pass in a > context structure, and have the callback set a value if it exits early > or not. I've done this already for needing the same information. > > No change to libtracefs is needed. This functionality is already > available with the current design: > > struct my_struct { > bool stopped; > }; > > int callback(struct tep_event *event, struct tep_record *record, > int cpu, void *data) > { > struct my_struct *my_data = data; > > [..] > if (condition) { > my_data->stopped = true; > return 1; > } > return 0; > } > > int main() > { > struct my_struct my_data = { .stopped = false }; > int ret = 0; > > while (!ret && !my_data.stopped) > ret = tracefs_iterate_raw_events(tep, instance, NULL, 0, > callback, &my_data); > } > > Now, if you are using this within python, and you want the python > wrapper to pass data as well, you just need to add that to the struct: > > struct my_struct { > bool stopped; > func *python_callback; > void *python_data; > } > > int callback(struct tep_event *event, struct tep_record *record, > int cpu, void *data) > { > struct my_struct *my_data = data; > int ret; > > [..] > ret = my_data->python_callback(event, record, cpu, data->python_data); > > if (ret) { > my_data->stopped = true; > return 1; > } > return 0; > } > > int python_iterator(pthyon_callback, python_data) > { > struct my_struct my_data = { .stopped = false }; > int ret = 0; > > my_data.python_data = python_data; > my_data.python_callback = python_callback; > > while (!ret && !my_data.stopped) > ret = tracefs_iterate_raw_events(tep, instance, NULL, 0, > callback, &my_data); > } > > So this patch isn't needed. I agree. Your solution is better. Thanks! Yordan > > -- Steve >
diff --git a/src/tracefs-events.c b/src/tracefs-events.c index 0fef64f..0a87d28 100644 --- a/src/tracefs-events.c +++ b/src/tracefs-events.c @@ -131,7 +131,8 @@ static int read_cpu_pages(struct tep_handle *tep, struct cpu_iterate *cpus, int } if (j < count) { if (callback(cpus[j].event, &cpus[j].record, cpus[j].cpu, callback_context)) - break; + return -1; + cpus[j].event = NULL; read_next_record(tep, cpus + j); } else {
Currently the return value of the callback function can be used to stop pulling data from the trace buffer, however this return value is lost and the user has no idea if tracefs_iterate_raw_events() terminated because there was no more data or because this was requested from the callback function. If we propagate the return value of the callback, this can be used in cases like the one below: int callback(struct tep_event *event, struct tep_record *record, int cpu, void *py_func) { .... return (something) ? 0 : 1 } int main() { int ret; .... while(ret == 0) ret = tracefs_iterate_raw_events(tep, instance, NULL, 0, callback, NULL); .... Here the user can effectively terminate the pulling the data from inside the callback. Signed-off-by: Yordan Karadzhov (VMware) <y.karadz@gmail.com> --- src/tracefs-events.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)