- 23 Sep, 2018 1 commit
-
-
Matheus Marchini authored
LLVM optimizes IR before turning it into BPF instructions. Sometimes the optimization might cut off important code if the IR was not built correctly. -d is not enough to find these issues, therefore this patch introduces more debug levels to also print the raw IR before it gets optimized.
-
- 21 Sep, 2018 3 commits
-
-
Brendan Gregg authored
usdt: resolve usdt arguments
-
Matheus Marchini authored
Now we can infer the path from current_attach_point_, which means we don't need to keep track of both references.
-
Matheus Marchini authored
This patch introduces initial support for USDT arguments through the argX builtins. For example: `./bpftrace -e 'usdt:bin:probe { printf("%s\n", str(arg0)); }'` Ref: https://github.com/iovisor/bpftrace/issues/33
-
- 19 Sep, 2018 2 commits
-
-
Brendan Gregg authored
fix attach to tracepoint regression
-
Matheus Marchini authored
Regression introduced by https://github.com/iovisor/bpftrace/pull/100. We can attach to multiple tracepoints within the same process without probem, which means we don't need to a suffix to it. Fixes: https://github.com/iovisor/bpftrace/issues/103
-
- 17 Sep, 2018 4 commits
-
-
Brendan Gregg authored
add system() call
-
williangaspar authored
-
williangaspar authored
-
williangaspar authored
-
- 16 Sep, 2018 6 commits
-
-
-
Brendan Gregg authored
attach to multiple identical probes
-
Matheus Marchini authored
-
Matheus Marchini authored
-
Matheus Marchini authored
-
Matheus Marchini authored
There were two problems keeping bpftrace from attaching to multiple identical probes in the same script: the first one was in our IR code generation, where we were creating multiple sections with the same name (since the secion name is determined by the probe's name). Since sections must have unique names, LLVM was not able to translate our IR to BPF. This was resolved by appending a `_#` to the section's name. The second problem was attaching to that probe: we can't use the same event name multiple times in "kprobe_events". bcc already handles multiple probes coming from *different* processes, since it also includes the PID on the event name. The solution was similar, we append `_#` to the event name before sending it to bcc. This required some strucutral changes, like keeping the `#` (index) in the AST Probe nodes and calling `BPFTrace::add_probe` during codegen instead of during semantic analysis (otherwise we wouldn't have the index for our Probes). Fixes: https://github.com/iovisor/bpftrace/issues/97
-
- 15 Sep, 2018 14 commits
-
-
Brendan Gregg authored
More Tools
-
Brendan Gregg authored
-
Brendan Gregg authored
support negative integers
-
Brendan Gregg authored
-
Brendan Gregg authored
-
Brendan Gregg authored
-
Brendan Gregg authored
-
Brendan Gregg authored
-
Brendan Gregg authored
-
Brendan Gregg authored
-
Brendan Gregg authored
-
Brendan Gregg authored
-
Brendan Gregg authored
-
Brendan Gregg authored
-
- 14 Sep, 2018 1 commit
-
-
Brendan Gregg authored
fix print() top logic
-
- 13 Sep, 2018 8 commits
-
-
Brendan Gregg authored
-
Brendan Gregg authored
add username builtin
-
williangaspar authored
-
williangaspar authored
-
-
Brendan Gregg authored
fix unterminated map.ident strings for print, clear, and zero
-
Brendan Gregg authored
fix printf with name on multiple events
-
Brendan Gregg authored
-
- 12 Sep, 2018 1 commit
-
-
Brendan Gregg authored
-