Add two new BPF instructions for dealing with per-CPU memory.
One, BPF_LDX | BPF_ADDR_PERCPU | BPF_DW (where BPF_ADD_PERCPU is unused
0xe0 opcode), resolved provided per-CPU address (offset) to an absolute
address where per-CPU data resides for "this" CPU. This is the most universal,
and, strictly speaking, the only per-CPU BPF instruction necessary.
I also added BPF_LDX | BPF_MEM_PERCPU | BPF_{B,H,W,DW} (BPF_MEM_PERCPU using
another unused 0xc0 opcode), which can be considered an optimization
instruction, which allows to *read* per-CPU data up to 8 bytes in one
instruction, without having to first resolve the address and then
dereferencing the memory. This one is used in inlining of
bpf_get_smp_processor_id(), but it would be fine to implement the latter with
BPF_ADD_PERCPU, followed by normal BPF_LDX | BPF_MEM, so I'm fine dropping
this one, if requested.
This instructions are currently supported by x86-64 BPF JIT, but it would be
great if this was added for other arches ASAP, of course.
In either case, we also implement inlining for three cases:
- bpf_get_smp_processor_id(), which allows to avoid unnecessary trivial
function call, saving a bit of performance and also not polluting LBR
records with unnecessary function call/return records;
- PERCPU_ARRAY's bpf_map_lookup_elem() is completely inlined, bringing its
performance to implementing per-CPU data structures using global variables
in BPF (which is an awesome improvement, see benchmarks below);
- PERCPU_HASH's bpf_map_lookup_elem() is partially inlined, just like the
same for non-PERCPU HASH map; this still saves a bit of overhead.
To validate performance benefits, I hacked together a tiny benchmark doing
only bpf_map_lookup_elem() and incrementing the value by 1 for PERCPU_ARRAY
(arr-inc benchmark below) and PERCPU_HASH (hash-inc benchmark below) maps. To
establish a baseline, I also implemented logic similar to PERCPU_ARRAY based
on global variable array using bpf_get_smp_processor_id() to index array for
current CPU (glob-arr-inc benchmark below).
BEFORE
======
glob-arr-inc : 163.685 ± 0.092M/s
arr-inc : 138.096 ± 0.160M/s
hash-inc : 66.855 ± 0.123M/s
AFTER
=====
glob-arr-inc : 173.921 ± 0.039M/s (+6%)
arr-inc : 170.729 ± 0.210M/s (+23.7%)
hash-inc : 68.673 ± 0.070M/s (+2.7%)
As can be seen, PERCPU_HASH gets a modest +2.7% improvement, while global
array-based gets a nice +6% due to inlining of bpf_get_smp_processor_id().
But what's really important is that arr-inc benchmark basically catches up
with glob-arr-inc, resulting in +23.7% improvement. This means that in
practice it won't be necessary to avoid PERCPU_ARRAY anymore if performance is
critical (e.g., high-frequent stats collection, which is often a practical use
for PERCPU_ARRAY today).
Andrii Nakryiko (4):
bpf: add internal-only per-CPU LDX instructions
bpf: inline bpf_get_smp_processor_id() helper
bpf: inline bpf_map_lookup_elem() for PERCPU_ARRAY maps
bpf: inline bpf_map_lookup_elem() helper for PERCPU_HASH map
arch/x86/net/bpf_jit_comp.c | 29 +++++++++++++++++++++++++++++
include/linux/filter.h | 27 +++++++++++++++++++++++++++
kernel/bpf/arraymap.c | 33 +++++++++++++++++++++++++++++++++
kernel/bpf/core.c | 5 +++++
kernel/bpf/disasm.c | 33 ++++++++++++++++++++++++++-------
kernel/bpf/hashtab.c | 21 +++++++++++++++++++++
kernel/bpf/verifier.c | 17 +++++++++++++++++
7 files changed, 158 insertions(+), 7 deletions(-)
--
2.43.0
