On 11/26/23 3:14 PM, Eduard Zingerman wrote:
On Sat, 2023-11-25 at 20:22 -0800, Yonghong Song wrote: [...]
--- a/tools/testing/selftests/bpf/progs/test_tunnel_kern.c +++ b/tools/testing/selftests/bpf/progs/test_tunnel_kern.c @@ -6,7 +6,10 @@ * modify it under the terms of version 2 of the GNU General Public * License as published by the Free Software Foundation. */ -#define BPF_NO_PRESERVE_ACCESS_INDEX +#if __has_attribute(preserve_static_offset) +struct __attribute__((preserve_static_offset)) erspan_md2; +struct __attribute__((preserve_static_offset)) erspan_metadata; +#endif #include "vmlinux.h"
[...]
int bpf_skb_get_fou_encap(struct __sk_buff *skb_ctx, @@ -174,9 +177,13 @@ int erspan_set_tunnel(struct __sk_buff *skb) __u8 hwid = 7; md.version = 2; +#if __has_attribute(preserve_static_offset) md.u.md2.dir = direction; md.u.md2.hwid = hwid & 0xf; md.u.md2.hwid_upper = (hwid >> 4) & 0x3; +#else
/* Change bit-field store to byte(s)-level stores. */+#endif #endif ret = bpf_skb_set_tunnel_opt(skb, &md, sizeof(md));
====
Eduard, could you double check whether this is a valid use case to solve this kind of issue with preserve_static_offset attribute?
Tbh I'm not sure. This test passes with preserve_static_offset because it suppresses preserve_access_index. In general clang translates bitfield access to a set of IR statements like:
C: struct foo { unsigned _; unsigned a:1; ... }; ... foo->a ...
IR: %a = getelementptr inbounds %struct.foo, ptr %0, i32 0, i32 1 %bf.load = load i8, ptr %a, align 4 %bf.clear = and i8 %bf.load, 1 %bf.cast = zext i8 %bf.clear to i32
With preserve_static_offset the getelementptr+load are replaced by a single statement which is preserved as-is till code generation, thus load with align 4 is preserved.
On the other hand, I'm not sure that clang guarantees that load or stores used for bitfield access would be always aligned according to verifier expectations.
I think it should be true. The frontend does alignment analysis based on types and (packed vs. unpacked) and assign each load/store with proper alignment (like 'align 4' in the above). 'align 4' truely means the load itself is 4-byte aligned. Otherwise, it will be very confusing for arch's which do not support unaligned memory access (e.g. BPF).
I think we should check if there are some clang knobs that prevent generation of unaligned memory access. I'll take a look.