joy/android_packages_modules_Connectivity
Watch
1
Fork
You've already forked android_packages_modules_Connectivity
 0
Code Issues Pull requests Projects Releases Wiki Activity

bpf: define READ_ONCE and WRITE_ONCE

Browse source 
These are convenience macros which do
(const|volatile) cast for atomic operations.

I've run into a case where two 32-bit writes get optimized
by the compiler into a single 64-bit write, which then fails
verification.  WRITE_ONCE() avoids that by virtue of the volatile annotation.

Technically I think the same thing could happen on read,
so add READ_ONCE too.

Test: N/A
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: Iade089aab3522916bd7b431d4064f89a21bc78b0
This commit is contained in:
Maciej Żenczykowski 2025-01-30 14:43:29 -08:00
commit 645b2d9755
1 changed files with 16 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

16
bpf/headers/include/bpf_helpers.h
View file

@ -419,6 +419,22 @@ static void (*bpf_ringbuf_submit_unsafe)(const void* data, __u64 flags) = (void*
DEFINE_BPF_MAP_UGM(the_map, TYPE, KeyType, ValueType, num_entries, \
DEFAULT_BPF_MAP_UID, gid, 0660)
// idea from Linux include/linux/compiler_types.h (eBPF is always a 64-bit arch)
#define NATIVE_WORD(t) ((sizeof(t) == 1) || (sizeof(t) == 2) || (sizeof(t) == 4) || (sizeof(t) == 8))
// simplified from Linux include/asm-generic/rwonce.h
#define READ_ONCE(x) \
({ \
_Static_assert(NATIVE_WORD(x), "READ_ONCE requires a native word size"); \
(*(const volatile typeof(x) *)&(x)) \
})
#define WRITE_ONCE(x, value) \
do { \
_Static_assert(NATIVE_WORD(x), "WRITE_ONCE requires a native word size"); \
*(volatile typeof(x) *)&(x) = (value); \
} while (0)
// LLVM eBPF builtins: they directly generate BPF_LD_ABS/BPF_LD_IND (skb may be ignored?)
unsigned long long load_byte(void* skb, unsigned long long off) asm("llvm.bpf.load.byte");
unsigned long long load_half(void* skb, unsigned long long off) asm("llvm.bpf.load.half");