| Message ID | 20240429131647.50165-3-puranjay@kernel.org (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | bpf, arm64: Support per-cpu instruction | expand |
Puranjay Mohan <puranjay@kernel.org> writes: > As ARM64 JIT now implements BPF_MOV64_PERCPU_REG instruction, inline > bpf_get_smp_processor_id(). > > ARM64 uses the per-cpu variable cpu_number to store the cpu id. While implementing this on the RISC-V JIT[1], I realized that reading the cpu_number from the per-cpu variable is not the best way to do this on arm64. arm64 now has the cpu number in the thread_info and reading that would be more efficient. Implementation in this patch is emitting: ; int cpu = bpf_get_smp_processor_id(); mov x7, #0xffff8000ffffffff movk x7, #0x8207, lsl #16 movk x7, #0x2008 mrs x10, tpidr_el1 add x7, x7, x10 ldr w7, [x7] If we do this in the JIT like I did for RISC-V[1] We can emit: ; int cpu = bpf_get_smp_processor_id(); mrs x10, sp_el0 ldr w7, [x10, #24] This gives ~ 10% improvement on glob-arr-inc and hash-inc compared to 3-4 % which this patch is providing. I will send v5 using this approach. [1] https://lore.kernel.org/all/20240430175834.33152-3-puranjay@kernel.org/ > Here is how the BPF and ARM64 JITed assembly changes after this commit: > > BPF > ===== > BEFORE AFTER > -------- ------- > > int cpu = bpf_get_smp_processor_id(); int cpu = bpf_get_smp_processor_id(); > (85) call bpf_get_smp_processor_id#229032 (18) r0 = 0xffff800082072008 > (bf) r0 = &(void __percpu *)(r0) > (61) r0 = *(u32 *)(r0 +0) > > ARM64 JIT > =========== > > BEFORE AFTER > -------- ------- > > int cpu = bpf_get_smp_processor_id(); int cpu = bpf_get_smp_processor_id(); > mov x10, #0xfffffffffffff4d0 mov x7, #0xffff8000ffffffff > movk x10, #0x802b, lsl #16 movk x7, #0x8207, lsl #16 > movk x10, #0x8000, lsl #32 movk x7, #0x2008 > blr x10 mrs x10, tpidr_el1 > add x7, x0, #0x0 add x7, x7, x10 > ldr w7, [x7] > > Performance improvement using benchmark[1] > > BEFORE AFTER > -------- ------- > > glob-arr-inc : 23.817 ± 0.019M/s glob-arr-inc : 24.631 ± 0.027M/s [+ 3.41%] > arr-inc : 23.253 ± 0.019M/s arr-inc : 23.742 ± 0.023M/s [+ 2.10%] > hash-inc : 12.258 ± 0.010M/s hash-inc : 12.625 ± 0.004M/s [+ 3.00%] > > [1] https://github.com/anakryiko/linux/commit/8dec900975ef Thanks, Puranjay
diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c index 4e474ef44e9c..d0725b1c7bec 100644 --- a/kernel/bpf/verifier.c +++ b/kernel/bpf/verifier.c @@ -20273,19 +20273,33 @@ static int do_misc_fixups(struct bpf_verifier_env *env) goto next_insn; } -#ifdef CONFIG_X86_64 /* Implement bpf_get_smp_processor_id() inline. */ if (insn->imm == BPF_FUNC_get_smp_processor_id && prog->jit_requested && bpf_jit_supports_percpu_insn()) { /* BPF_FUNC_get_smp_processor_id inlining is an - * optimization, so if pcpu_hot.cpu_number is ever + * optimization, so if cpu_number_addr is ever * changed in some incompatible and hard to support * way, it's fine to back out this inlining logic */ - insn_buf[0] = BPF_MOV32_IMM(BPF_REG_0, (u32)(unsigned long)&pcpu_hot.cpu_number); - insn_buf[1] = BPF_MOV64_PERCPU_REG(BPF_REG_0, BPF_REG_0); - insn_buf[2] = BPF_LDX_MEM(BPF_W, BPF_REG_0, BPF_REG_0, 0); - cnt = 3; + u64 cpu_number_addr; + struct bpf_insn ld_insn[2] = { + BPF_LD_IMM64(BPF_REG_0, 0) + }; + +#if defined(CONFIG_X86_64) + cpu_number_addr = (u64)&pcpu_hot.cpu_number; +#elif defined(CONFIG_ARM64) + cpu_number_addr = (u64)&cpu_number; +#else + goto next_insn; +#endif + ld_insn[0].imm = (u32)cpu_number_addr; + ld_insn[1].imm = (u32)(cpu_number_addr >> 32); + insn_buf[0] = ld_insn[0]; + insn_buf[1] = ld_insn[1]; + insn_buf[2] = BPF_MOV64_PERCPU_REG(BPF_REG_0, BPF_REG_0); + insn_buf[3] = BPF_LDX_MEM(BPF_W, BPF_REG_0, BPF_REG_0, 0); + cnt = 4; new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); if (!new_prog) @@ -20296,7 +20310,6 @@ static int do_misc_fixups(struct bpf_verifier_env *env) insn = new_prog->insnsi + i + delta; goto next_insn; } -#endif /* Implement bpf_get_func_arg inline. */ if (prog_type == BPF_PROG_TYPE_TRACING && insn->imm == BPF_FUNC_get_func_arg) {
As ARM64 JIT now implements BPF_MOV64_PERCPU_REG instruction, inline bpf_get_smp_processor_id(). ARM64 uses the per-cpu variable cpu_number to store the cpu id. Here is how the BPF and ARM64 JITed assembly changes after this commit: BPF ===== BEFORE AFTER -------- ------- int cpu = bpf_get_smp_processor_id(); int cpu = bpf_get_smp_processor_id(); (85) call bpf_get_smp_processor_id#229032 (18) r0 = 0xffff800082072008 (bf) r0 = &(void __percpu *)(r0) (61) r0 = *(u32 *)(r0 +0) ARM64 JIT =========== BEFORE AFTER -------- ------- int cpu = bpf_get_smp_processor_id(); int cpu = bpf_get_smp_processor_id(); mov x10, #0xfffffffffffff4d0 mov x7, #0xffff8000ffffffff movk x10, #0x802b, lsl #16 movk x7, #0x8207, lsl #16 movk x10, #0x8000, lsl #32 movk x7, #0x2008 blr x10 mrs x10, tpidr_el1 add x7, x0, #0x0 add x7, x7, x10 ldr w7, [x7] Performance improvement using benchmark[1] BEFORE AFTER -------- ------- glob-arr-inc : 23.817 ± 0.019M/s glob-arr-inc : 24.631 ± 0.027M/s [+ 3.41%] arr-inc : 23.253 ± 0.019M/s arr-inc : 23.742 ± 0.023M/s [+ 2.10%] hash-inc : 12.258 ± 0.010M/s hash-inc : 12.625 ± 0.004M/s [+ 3.00%] [1] https://github.com/anakryiko/linux/commit/8dec900975ef Signed-off-by: Puranjay Mohan <puranjay@kernel.org> Acked-by: Andrii Nakryiko <andrii@kernel.org> --- kernel/bpf/verifier.c | 27 ++++++++++++++++++++------- 1 file changed, 20 insertions(+), 7 deletions(-)