Alexei Starovoitov wrote:
> On Tue, Mar 24, 2020 at 10:38:56AM -0700, John Fastabend wrote:
> > -static void __reg_bound_offset32(struct bpf_reg_state *reg)
> > +static void __reg_combine_32_into_64(struct bpf_reg_state *reg)
> > {
> > - u64 mask = 0xffffFFFF;
> > - struct tnum range = tnum_range(reg->umin_value & mask,
> > - reg->umax_value & mask);
> > - struct tnum lo32 = tnum_cast(reg->var_off, 4);
> > - struct tnum hi32 = tnum_lshift(tnum_rshift(reg->var_off, 32), 32);
> > + /* special case when 64-bit register has upper 32-bit register
> > + * zeroed. Typically happens after zext or <<32, >>32 sequence
> > + * allowing us to use 32-bit bounds directly,
> > + */
> > + if (tnum_equals_const(tnum_clear_subreg(reg->var_off), 0)) {
> > + reg->umin_value = reg->u32_min_value;
> > + reg->umax_value = reg->u32_max_value;
> > + reg->smin_value = reg->s32_min_value;
> > + reg->smax_value = reg->s32_max_value;
>
> Looks like above will not be correct for negative s32_min/max.
> When upper 32-bit are cleared and we're processing jmp32
> we cannot set smax_value to s32_max_value.
> Consider if (w0 s< -5)
> s32_max_value == -5
> which is 0xfffffffb
> but upper 32 are zeros so smax_value should be (u64)0xfffffffb
> and not (s64)-5
Right, good catch. I'll use below logic here as well.
>
> We can be fancy and precise with this logic, but I would just use similar
> approach from zext_32_to_64() where the following:
> + if (reg->s32_min_value > 0)
> + reg->smin_value = reg->s32_min_value;
> + else
> + reg->smin_value = 0;
> + if (reg->s32_max_value > 0)
> + reg->smax_value = reg->s32_max_value;
> + else
> + reg->smax_value = U32_MAX;
> should work for this case too ?
>
> > + if (BPF_SRC(insn->code) == BPF_K) {
> > + pred = is_branch_taken(dst_reg, insn->imm, opcode, is_jmp32);
> > + } else if (src_reg->type == SCALAR_VALUE && is_jmp32 && tnum_is_const(tnum_subreg(src_reg->var_off))) {
> > + pred = is_branch_taken(dst_reg, tnum_subreg(src_reg->var_off).value, opcode, is_jmp32);
> > + } else if (src_reg->type == SCALAR_VALUE && !is_jmp32 && tnum_is_const(src_reg->var_off)) {
> > + pred = is_branch_taken(dst_reg, src_reg->var_off.value, opcode, is_jmp32);
> > + }
>
> pls wrap these lines. Way above normal.
+1
>
> The rest is awesome.
Thanks.
