[PATCH RFCv3 01/14] arm64: introduce aarch64_insn_gen_comp_branch_imm()

[Will Deacon]

On Wed, Jul 16, 2014 at 10:19:31PM +0100, Zi Shen Lim wrote:
> On Wed, Jul 16, 2014 at 05:04:50PM +0100, Will Deacon wrote:
> > On Tue, Jul 15, 2014 at 07:24:59AM +0100, Zi Shen Lim wrote:
> [...]
> > > +enum aarch64_insn_register {
> > > +	AARCH64_INSN_REG_0  = 0,
> > > +	AARCH64_INSN_REG_1  = 1,
> > > +	AARCH64_INSN_REG_2  = 2,
> > > +	AARCH64_INSN_REG_3  = 3,
> > > +	AARCH64_INSN_REG_4  = 4,
> > > +	AARCH64_INSN_REG_5  = 5,
> > > +	AARCH64_INSN_REG_6  = 6,
> > > +	AARCH64_INSN_REG_7  = 7,
> > > +	AARCH64_INSN_REG_8  = 8,
> > > +	AARCH64_INSN_REG_9  = 9,
> > > +	AARCH64_INSN_REG_10 = 10,
> > > +	AARCH64_INSN_REG_11 = 11,
> > > +	AARCH64_INSN_REG_12 = 12,
> > > +	AARCH64_INSN_REG_13 = 13,
> > > +	AARCH64_INSN_REG_14 = 14,
> > > +	AARCH64_INSN_REG_15 = 15,
> > > +	AARCH64_INSN_REG_16 = 16,
> > > +	AARCH64_INSN_REG_17 = 17,
> > > +	AARCH64_INSN_REG_18 = 18,
> > > +	AARCH64_INSN_REG_19 = 19,
> > > +	AARCH64_INSN_REG_20 = 20,
> > > +	AARCH64_INSN_REG_21 = 21,
> > > +	AARCH64_INSN_REG_22 = 22,
> > > +	AARCH64_INSN_REG_23 = 23,
> > > +	AARCH64_INSN_REG_24 = 24,
> > > +	AARCH64_INSN_REG_25 = 25,
> > > +	AARCH64_INSN_REG_26 = 26,
> > > +	AARCH64_INSN_REG_27 = 27,
> > > +	AARCH64_INSN_REG_28 = 28,
> > > +	AARCH64_INSN_REG_29 = 29,
> > > +	AARCH64_INSN_REG_FP = 29, /* Frame pointer */
> > > +	AARCH64_INSN_REG_30 = 30,
> > > +	AARCH64_INSN_REG_LR = 30, /* Link register */
> > > +	AARCH64_INSN_REG_ZR = 31, /* Zero: as source register */
> > > +	AARCH64_INSN_REG_SP = 31  /* Stack pointer: as load/store base reg */
> > 
> > Can you just #define AARCH64_INSN_REG(x) instead, then have some magic
> > values like ARM64_REG_LR which are defined as the appropriate numbers?
> 
> I actually had something like what you mentioned in the beginning, but
> decided to go with the above - thinking that it's clearer to present
> the complete set of valid register definitions.
> 
> The #define can still be added for convenience, though I think it's also a
> potential source of errors - it's much easier to typo something like
> AARCH64_INSN_REG(32) and not get caught.

Fair enough, that's a good enough reason to leave it like it is.

> [...]
> > > +	switch (variant) {
> > > +	case AARCH64_INSN_VARIANT_32BIT:
> > > +		break;
> > > +	case AARCH64_INSN_VARIANT_64BIT:
> > > +		insn |= BIT(31);
> > 
> > FWIW, that bit (31) is referred to as the `SF' bit in the instruction
> > encodings (for Sixty-Four). You could have a #define for that to help people
> > match up the bitfield, if you like.
> 
> Something like this?
> 
> 	#define AARCH64_INSN_SF_BIT  BIT(31)
> 
> 	...
> 
> 	case AARCH64_INSN_VARIANT_64BIT:
> 		insn |= AARCH64_INSN_SF_BIT;
> 
> In the case of bitfield instruction, there's also an "N" bit.
> So something like this?
> 
> 	#define AARCH64_INSN_N_BIT  BIT(22)
> 
> 	...
> 
> 	case AARCH64_INSN_VARIANT_64BIT:
> 		insn |= AARCH64_INSN_SF_BIT | AARCH64_INSN_N_BIT;

Looks good.

> > 
> > > +		break;
> > > +	default:
> > > +		BUG_ON(1);
> > 
> > Is a BUG_ON justifiable here? Is there not a nicer way to fail?
> 
> In general, it'd be nice if we returned something like -EINVAL and
> have all callers handle failures. Today all code gen functions return
> the u32 instruction and there's no error handling by callers.
> I think following the precedence (aarch64_insn_gen_branch_imm())
> of failing with BUG_ON is a reasonable tradeoff.

Well, I don't necessarily agree with that BUG_ON, either :)
I take it eBPF doesn't have a `trap' instruction or similar? Otherwise, we
could generate that and avoid having to propagate errors directly to the
caller.

> In this case here, when we hit the default (failure) case, that means
> there's a serious error of attempting to use an unsupported
> variant. I think we're better off failing hard here than trying to
> arbitrarily "fallback" on a default choice.

It might be a serious error for BPF, but a BUG_ON brings down the entire
machine, which I think is unfortunate.

> 
> One potential option instead of switch (variant) is:
> 
> 	if (variant == AARCH64_INSN_VARIANT_64BIT)
> 		/* do something */
> 	else
> 		/* do something else */
> 
> which would be quite reasonable to do as we only have VARIANT_{32,64}BIT
> today.
> 
> However, consider the case where we add VARIANT_128BIT or other flavors
> in the future. The if/else option (basically defaulting to VARIANT_32BIT)
> would then make much less sense.

I don't think we need to worry about hypothetical extensions to the
instruction set at this stage.

Will