7852ebc088
Before llvm20, (void)__sync_fetch_and_add(...) always generates locked xadd insns. In linux kernel upstream discussion [1], it is found that for arm64 architecture, the original semantics of (void)__sync_fetch_and_add(...), i.e., __atomic_fetch_add(...), is preferred in order for jit to emit proper native barrier insns. In llvm commits [2] and [3], (void)__sync_fetch_and_add(...) will generate the following insns: - for cpu v1/v2: locked xadd insns to keep backward compatibility - for cpu v3/v4: __atomic_fetch_add() insns To ensure proper barrier semantics for (void)__sync_fetch_and_add(...), cpu v3/v4 is recommended. This patch enables cpu=v3 as the default cpu version. For users wanting to use cpu v1, -mcpu=v1 needs to be explicitly added to clang/llc command line. [1] https://lore.kernel.org/bpf/ZqqiQQWRnz7H93Hc@google.com/T/#mb68d67bc8f39e35a0c3db52468b9de59b79f021f [2] https://github.com/llvm/llvm-project/pull/101428 [3] https://github.com/llvm/llvm-project/pull/106494
65 lines
2.7 KiB
LLVM
65 lines
2.7 KiB
LLVM
; RUN: llc < %s -march=bpfel -mcpu=v1 | FileCheck -check-prefixes=CHECK,EL %s
|
|
; RUN: llc < %s -march=bpfeb -mcpu=v1 | FileCheck -check-prefixes=CHECK,EB %s
|
|
|
|
%struct.bpf_map_def = type { i32, i32, i32, i32 }
|
|
%struct.__sk_buff = type opaque
|
|
%struct.routing_key_2 = type { [6 x i8] }
|
|
|
|
@routing = global %struct.bpf_map_def { i32 1, i32 6, i32 12, i32 1024 }, section "maps", align 4
|
|
@routing_miss_0 = global %struct.bpf_map_def { i32 1, i32 1, i32 12, i32 1 }, section "maps", align 4
|
|
@test1 = global %struct.bpf_map_def { i32 2, i32 4, i32 8, i32 1024 }, section "maps", align 4
|
|
@test1_miss_4 = global %struct.bpf_map_def { i32 2, i32 1, i32 8, i32 1 }, section "maps", align 4
|
|
@_license = global [4 x i8] c"GPL\00", section "license", align 1
|
|
@llvm.used = appending global [6 x ptr] [ptr @_license, ptr @ebpf_filter, ptr @routing, ptr @routing_miss_0, ptr @test1, ptr @test1_miss_4], section "llvm.metadata"
|
|
|
|
; Function Attrs: nounwind uwtable
|
|
define i32 @ebpf_filter(ptr nocapture readnone %ebpf_packet) #0 section "socket1" {
|
|
|
|
; EL: r1 = 11033905661445 ll
|
|
; EB: r1 = 361984551142686720 ll
|
|
; CHECK: *(u64 *)(r10 - 8) = r1
|
|
|
|
; CHECK: r1 = 0
|
|
; CHECK-DAG: *(u16 *)(r10 + 24) = r1
|
|
; CHECK-DAG: *(u16 *)(r10 + 22) = r1
|
|
; CHECK-DAG: *(u16 *)(r10 + 20) = r1
|
|
; CHECK-DAG: *(u16 *)(r10 + 18) = r1
|
|
; CHECK-DAG: *(u16 *)(r10 + 16) = r1
|
|
; CHECK-DAG: *(u16 *)(r10 + 14) = r1
|
|
; CHECK-DAG: *(u16 *)(r10 + 12) = r1
|
|
; CHECK-DAG: *(u16 *)(r10 + 10) = r1
|
|
; CHECK-DAG: *(u16 *)(r10 + 8) = r1
|
|
; CHECK-DAG: *(u16 *)(r10 + 6) = r1
|
|
; CHECK-DAG: *(u16 *)(r10 + 4) = r1
|
|
; CHECK-DAG: *(u16 *)(r10 + 2) = r1
|
|
; CHECK-DAG: *(u16 *)(r10 + 0) = r1
|
|
; CHECK-DAG: *(u16 *)(r10 + 26) = r1
|
|
|
|
; CHECK: r2 = r10
|
|
; CHECK: r2 += -8
|
|
; CHECK: r1 = routing
|
|
; CHECK: call bpf_map_lookup_elem
|
|
; CHECK: exit
|
|
%key = alloca %struct.routing_key_2, align 8
|
|
store i8 5, ptr %key, align 1
|
|
%1 = getelementptr inbounds %struct.routing_key_2, ptr %key, i64 0, i32 0, i64 1
|
|
store i8 6, ptr %1, align 1
|
|
%2 = getelementptr inbounds %struct.routing_key_2, ptr %key, i64 0, i32 0, i64 2
|
|
store i8 7, ptr %2, align 1
|
|
%3 = getelementptr inbounds %struct.routing_key_2, ptr %key, i64 0, i32 0, i64 3
|
|
store i8 8, ptr %3, align 1
|
|
%4 = getelementptr inbounds %struct.routing_key_2, ptr %key, i64 0, i32 0, i64 4
|
|
store i8 9, ptr %4, align 1
|
|
%5 = getelementptr inbounds %struct.routing_key_2, ptr %key, i64 0, i32 0, i64 5
|
|
store i8 10, ptr %5, align 1
|
|
%6 = getelementptr inbounds %struct.routing_key_2, ptr %key, i64 1, i32 0, i64 0
|
|
call void @llvm.memset.p0.i64(ptr %6, i8 0, i64 30, i1 false)
|
|
%7 = call i32 (ptr, ptr, ...) @bpf_map_lookup_elem(ptr nonnull @routing, ptr nonnull %key) #3
|
|
ret i32 undef
|
|
}
|
|
|
|
; Function Attrs: nounwind argmemonly
|
|
declare void @llvm.memset.p0.i64(ptr nocapture, i8, i64, i1) #1
|
|
|
|
declare i32 @bpf_map_lookup_elem(...) #2
|