While exploring uretprobe syscall and trampoline for ARM64, we observed
a slight performance gain for Redis benchmark using uretprobe syscall.
This patchset aims to further improve the performance of uretprobe by
optimizing the management of struct return_instance data.
In details, uretprobe utilizes dynamically allocated memory for struct
return_instance data. These data track the call chain of instrumented
functions. This approach is not efficient, especially considering the
inherent locality of function invocation.
This patchset proposes a rework of the return_instances management. It
replaces dynamic memory allocation with a statically allocated array.
This approach leverages the stack-style usage of return_instance and
remove the need for kamlloc/kfree operations.
This patch has been tested on Kunpeng916 (Hi1616), 4 NUMA nodes, 64
cores @ 2.4GHz. Redis benchmarks show a throughput gain by 2% for Redis
GET and SET commands:
------------------------------------------------------------------
Test case | No uretprobes | uretprobes | uretprobes
| | (current) | (optimized)
==================================================================
Redis SET (RPS) | 47025 | 40619 (-13.6%) | 41529 (-11.6%)
------------------------------------------------------------------
Redis GET (RPS) | 46715 | 41426 (-11.3%) | 42306 (-9.4%)
------------------------------------------------------------------
Liao Chang (2):
uprobes: Optimize the return_instance related routines
selftests/bpf: Add uretprobe test for return_instance management
include/linux/uprobes.h | 10 +-
kernel/events/uprobes.c | 162 +++++++++++-------
.../bpf/prog_tests/uretprobe_depth.c | 150 ++++++++++++++++
.../selftests/bpf/progs/uretprobe_depth.c | 19 ++
4 files changed, 274 insertions(+), 67 deletions(-)
create mode 100644 tools/testing/selftests/bpf/prog_tests/uretprobe_depth.c
create mode 100644 tools/testing/selftests/bpf/progs/uretprobe_depth.c
--
2.34.1
