Split syscalls headers and tools

[lttng-modules.git] / tools / syscalls / README

diff --git a/tools/syscalls/README b/tools/syscalls/README
new file mode 100644 (file)
index 0000000..a51531b
--- /dev/null
+++ b/tools/syscalls/README
@@ -0,0 +1,33 @@
+LTTng system call tracing
+
+1) lttng-syscall-extractor
+
+You need to build a kernel with CONFIG_FTRACE_SYSCALLS=y and
+CONFIG_KALLSYMS_ALL=y for extraction. Apply the linker patch to get your
+kernel to keep the system call metadata after boot.  Then build and load
+the LTTng syscall extractor module. The module will fail to load (this
+is expected). See the dmesg output for system call metadata.
+
+2) Generate system call TRACE_EVENT().
+
+Take the dmesg data and feed it to lttng-syscalls-generate-headers.sh from the
+tools/syscalls directory. See the script header for usage example. It should be
+run for both the integers and pointers types.
+
+After these are created, you just need to follow the new system call additions,
+no need to regenerate the whole thing, since system calls are only appended to.
+
+3) Override headers
+
+You need to provide override headers (even if they don't contain
+overrides) for each architecture. Those are meant to contain manual
+overrides for specific system calls for cases where the script do not
+generate the correct typing for specific system calls.
+
+4) Integrating the new architecture, updating architecture header
+
+Edit instrumentation/syscalls/syscalls_integers.h and
+instrumentation/syscalls/syscalls_pointers.h. Add the new architecture (e.g.
+CONFIG_X86_64) to the newly-generated headers (e.g. changing the current
+x86-64-syscalls_<CLASS>.h into x86-64-syscalls_<CLASS>.h), then rebuild
+lttng-modules.