RE: [PATCH] bpf/arm64: support packet data load instructions

[Marat Khalili <marat.khalili@huawei.com>]

> > I suggest the following logic in pseudo-code:
> >
> > static void
> > emit_ld_mbuf(struct a64_jit_ctx *ctx, uint32_t op, uint8_t tmp1,
> > uint8_t tmp2,
> > 	uint8_t src, int32_t imm)
> > {
> > 	// ...
> >
> > 	/* r1 = off: for ABS use imm, for IND use src + imm */
> > 	if (mode == BPF_ABS) {
> > 		assert imm >= 0, "verified by verifier"
> > 		emit MOV W1, #<imm>
> > 	} else {
> > 		/* add signed imm to the source register */
> > 		emit(s) X1 = src + #<imm>
> > 		/* verify dynamically that offset is within the domain of
> > __rte_pktmbuf_read */
> > 		emit(s) jump_to_epilogue if X1 <s 0 || X1 > INT32_MAX
> > 	}
> >
> > 	// ...
> > }
> 
> I don't know if it's the sum of src+imm that determines special meaning, or it's the imm itself.
> If it's the imm itself, a simple fix would be to update the validator's .imm.max values for BPF_IND
> from UINT32_MAX to INT32_MAX.
> 

Even if we do it (I have no particular opinion), it won't prevent sum of 64-bit
register and immediate from being negative or greater than the range of values
supported by 32-bit argument of __rte_pktmbuf_read.