Update yjit.md about perf

doc/yjit/yjit.md

@@ -479,11 +479,16 @@ perf script --fields +pid > /tmp/test.perf
 
 You can also profile the number of cycles consumed by code generated by each YJIT function.
 
+#### Building perf with Python support
+
+`misc/yjit_perf.py` is a script for `perf script -s` command.
+By default, it doesn't support Python scripts, so you need to build `perf` from source.
+
 ```bash
 # Build perf from source for Python support
-# [Optional] libelf-dev libunwind-dev libaudit-dev libslang2-dev libdw-dev
-sudo apt-get install libpython3-dev python3-pip flex libtraceevent-dev
-git clone https://github.com/torvalds/linux
+sudo apt-get install libpython3-dev python3-pip flex libtraceevent-dev \
+  libelf-dev libunwind-dev libaudit-dev libslang2-dev libdw-dev
+git clone --depth=1 https://github.com/torvalds/linux
 cd linux/tools/perf
 make
 make install
@@ -499,3 +504,28 @@ PERF=record ruby --yjit-perf=codegen -Iharness-perf benchmarks/lobsters/benchmar
 # Aggregate results
 perf script -s ../ruby/misc/yjit_perf.py
 ```
+
+### Using perf without Python support
+
+`misc/yjit_perf.py` also has an extra interface to support `perf` without Python support.
+You need to run `perf script` (instead of `perf script -s`) first, and feed the output to `misc/yjit_perf.py`.
+
+This interface exists primarily for production environments. You can run `perf script` in production,
+export the output locally, and run the script using it.
+
+```bash
+# Install perf
+apt-get install linux-tools-common linux-tools-generic linux-tools-`uname -r`
+
+# [Optional] Allow running perf without sudo
+echo 0 | sudo tee /proc/sys/kernel/kptr_restrict
+echo -1 | sudo tee /proc/sys/kernel/perf_event_paranoid
+
+# Profile Ruby with --yjit-perf=codegen
+cd ../yjit-bench
+PERF=record ruby --yjit-perf=codegen -Iharness-perf benchmarks/lobsters/benchmark.rb
+
+# Aggregate results
+perf script > /tmp/perf.txt
+perf script /tmp/perf.txt
+```