Re: [PATCH v2 bpf-next 08/13] bpf: Add instructions for atomic_[cmp]xchg

[Yonghong Song]

On 11/27/20 9:57 AM, Brendan Jackman wrote:
> This adds two atomic opcodes, both of which include the BPF_FETCH
> flag. XCHG without the BPF_FETCh flag would naturally encode

BPF_FETCH

> atomic_set. This is not supported because it would be of limited
> value to userspace (it doesn't imply any barriers). CMPXCHG without
> BPF_FETCH woulud be an atomic compare-and-write. We don't have such
> an operation in the kernel so it isn't provided to BPF either.
>
> There are two significant design decisions made for the CMPXCHG
> instruction:
>
>   - To solve the issue that this operation fundamentally has 3
>     operands, but we only have two register fields. Therefore the
>     operand we compare against (the kernel's API calls it 'old') is
>     hard-coded to be R0. x86 has similar design (and A64 doesn't
>     have this problem).
>
>     A potential alternative might be to encode the other operand's
>     register number in the immediate field.
>
>   - The kernel's atomic_cmpxchg returns the old value, while the C11
>     userspace APIs return a boolean indicating the comparison
>     result. Which should BPF do? A64 returns the old value. x86 returns
>     the old value in the hard-coded register (and also sets a
>     flag). That means return-old-value is easier to JIT.
>
> Signed-off-by: Brendan Jackman <jackmanb@xxxxxxxxxx>
> ---
>   arch/x86/net/bpf_jit_comp.c    |  8 ++++++++
>   include/linux/filter.h         | 20 ++++++++++++++++++++
>   include/uapi/linux/bpf.h       |  4 +++-
>   kernel/bpf/core.c              | 20 ++++++++++++++++++++
>   kernel/bpf/disasm.c            | 15 +++++++++++++++
>   kernel/bpf/verifier.c          | 19 +++++++++++++++++--
>   tools/include/linux/filter.h   | 20 ++++++++++++++++++++
>   tools/include/uapi/linux/bpf.h |  4 +++-
>   8 files changed, 106 insertions(+), 4 deletions(-)
[...]
> diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
> index cd4c03b25573..c8311cc114ec 100644
> --- a/kernel/bpf/verifier.c
> +++ b/kernel/bpf/verifier.c
> @@ -3601,10 +3601,13 @@ static int check_mem_access(struct bpf_verifier_env *env, int insn_idx, u32 regn
>   static int check_atomic(struct bpf_verifier_env *env, int insn_idx, struct bpf_insn *insn)
>   {
>   	int err;
> +	int load_reg;
>   
>   	switch (insn->imm) {
>   	case BPF_ADD:
>   	case BPF_ADD | BPF_FETCH:
> +	case BPF_XCHG:
> +	case BPF_CMPXCHG:
>   		break;
>   	default:
>   		verbose(env, "BPF_ATOMIC uses invalid atomic opcode %02x\n", insn->imm);
> @@ -3626,6 +3629,13 @@ static int check_atomic(struct bpf_verifier_env *env, int insn_idx, struct bpf_i
>   	if (err)
>   		return err;
>   
> +	if (insn->imm == BPF_CMPXCHG) {
> +		/* check src3 operand */

better comment about what src3 means here?

> +		err = check_reg_arg(env, BPF_REG_0, SRC_OP);
> +		if (err)
> +			return err;
> +	}
> +
>   	if (is_pointer_value(env, insn->src_reg)) {
>   		verbose(env, "R%d leaks addr into mem\n", insn->src_reg);
>   		return -EACCES;
> @@ -3656,8 +3666,13 @@ static int check_atomic(struct bpf_verifier_env *env, int insn_idx, struct bpf_i
>   	if (!(insn->imm & BPF_FETCH))
>   		return 0;
>   
> -	/* check and record load of old value into src reg  */
> -	err = check_reg_arg(env, insn->src_reg, DST_OP);
> +	if (insn->imm == BPF_CMPXCHG)
> +		load_reg = BPF_REG_0;
> +	else
> +		load_reg = insn->src_reg;
> +
> +	/* check and record load of old value */
> +	err = check_reg_arg(env, load_reg, DST_OP);
>   	if (err)
>   		return err;
>   
[...]