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
49 lines
1.9 KiB
LLVM
49 lines
1.9 KiB
LLVM
; RUN: llc < %s -march=bpfel -mcpu=v1 | FileCheck %s
|
|
|
|
%struct.bpf_context = type { i64, i64, i64, i64, i64, i64, i64 }
|
|
%struct.sk_buff = type { i64, i64, i64, i64, i64, i64, i64 }
|
|
%struct.net_device = type { i64, i64, i64, i64, i64, i64, i64 }
|
|
|
|
@bpf_prog1.devname = private unnamed_addr constant [3 x i8] c"lo\00", align 1
|
|
@bpf_prog1.fmt = private unnamed_addr constant [15 x i8] c"skb %x dev %x\0A\00", align 1
|
|
|
|
; Function Attrs: nounwind uwtable
|
|
define i32 @bpf_prog1(ptr nocapture %ctx) #0 section "events/net/netif_receive_skb" {
|
|
%devname = alloca [3 x i8], align 1
|
|
%fmt = alloca [15 x i8], align 1
|
|
%1 = getelementptr inbounds [3 x i8], ptr %devname, i64 0, i64 0
|
|
call void @llvm.memcpy.p0.p0.i64(ptr %1, ptr @bpf_prog1.devname, i64 3, i1 false)
|
|
%2 = getelementptr inbounds %struct.bpf_context, ptr %ctx, i64 0, i32 0
|
|
%3 = load i64, ptr %2, align 8
|
|
%4 = inttoptr i64 %3 to ptr
|
|
%5 = getelementptr inbounds %struct.sk_buff, ptr %4, i64 0, i32 2
|
|
%6 = bitcast ptr %5 to ptr
|
|
%7 = call ptr inttoptr (i64 4 to ptr)(ptr %6) #1
|
|
%8 = call i32 inttoptr (i64 9 to ptr)(ptr %7, ptr %1, i32 2) #1
|
|
%9 = icmp eq i32 %8, 0
|
|
br i1 %9, label %10, label %13
|
|
|
|
; <label>:10 ; preds = %0
|
|
%11 = getelementptr inbounds [15 x i8], ptr %fmt, i64 0, i64 0
|
|
call void @llvm.memcpy.p0.p0.i64(ptr %11, ptr @bpf_prog1.fmt, i64 15, i1 false)
|
|
%12 = call i32 (ptr, i32, ...) inttoptr (i64 11 to ptr)(ptr %11, i32 15, ptr %4, ptr %7) #1
|
|
; CHECK-LABEL: bpf_prog1:
|
|
; CHECK: call 4
|
|
; CHECK: call 9
|
|
; CHECK: if r0 != 0
|
|
; CHECK: r1 = 622884453
|
|
; CHECK: *(u32 *)(r10 - 16) = r1
|
|
; CHECK: r1 = 7214898703899978611 ll
|
|
; CHECK: *(u64 *)(r10 - 24) = r1
|
|
; CHECK: call 11
|
|
; CHECK: r0 = 0
|
|
; CHECK: exit
|
|
br label %13
|
|
|
|
; <label>:13 ; preds = %10, %0
|
|
ret i32 0
|
|
}
|
|
|
|
; Function Attrs: nounwind
|
|
declare void @llvm.memcpy.p0.p0.i64(ptr nocapture, ptr nocapture, i64, i1) #1
|