Re: [PATCH v5 bpf-next] Small BPF verifier log improvements

[Daniel Borkmann <daniel@xxxxxxxxxxxxx>]

On 3/2/22 11:13 PM, Alexei Starovoitov wrote:
> On Wed, Mar 2, 2022 at 1:46 PM Daniel Borkmann <daniel@xxxxxxxxxxxxx> wrote:
> > On 3/2/22 10:23 PM, Alexei Starovoitov wrote:
> > > On Tue, Mar 01, 2022 at 02:27:45PM -0800, Mykola Lysenko wrote:
> > > >               .prog_type = BPF_PROG_TYPE_SCHED_CLS,
> > > >               .matches = {
> > > > -                    {6, "R3_w=inv(id=0,umax_value=255,var_off=(0x0; 0xff))"},
> > > > -                    {7, "R4_w=inv(id=1,umax_value=255,var_off=(0x0; 0xff))"},
> > > > -                    {8, "R4_w=inv(id=0,umax_value=255,var_off=(0x0; 0xff))"},
> > > > -                    {9, "R4_w=inv(id=1,umax_value=255,var_off=(0x0; 0xff))"},
> > > > -                    {10, "R4_w=inv(id=0,umax_value=510,var_off=(0x0; 0x1fe))"},
> > > > -                    {11, "R4_w=inv(id=1,umax_value=255,var_off=(0x0; 0xff))"},
> > > > -                    {12, "R4_w=inv(id=0,umax_value=1020,var_off=(0x0; 0x3fc))"},
> > > > -                    {13, "R4_w=inv(id=1,umax_value=255,var_off=(0x0; 0xff))"},
> > > > -                    {14, "R4_w=inv(id=0,umax_value=2040,var_off=(0x0; 0x7f8))"},
> > > > -                    {15, "R4_w=inv(id=0,umax_value=4080,var_off=(0x0; 0xff0))"},
> > > > +                    {6, "R3_w=scalar(umax=255,var_off=(0x0; 0xff))"},
> > > > +                    {7, "R4_w=scalar(id=1,umax=255,var_off=(0x0; 0xff))"},
> > > > +                    {8, "R4_w=scalar(umax=255,var_off=(0x0; 0xff))"},
> > > > +                    {9, "R4_w=scalar(id=1,umax=255,var_off=(0x0; 0xff))"},
> > > > +                    {10, "R4_w=scalar(umax=510,var_off=(0x0; 0x1fe))"},
> > > > +                    {11, "R4_w=scalar(id=1,umax=255,var_off=(0x0; 0xff))"},
> > > > +                    {12, "R4_w=scalar(umax=1020,var_off=(0x0; 0x3fc))"},
> > > > +                    {13, "R4_w=scalar(id=1,umax=255,var_off=(0x0; 0xff))"},
> > > > +                    {14, "R4_w=scalar(umax=2040,var_off=(0x0; 0x7f8))"},
> > > > +                    {15, "R4_w=scalar(umax=4080,var_off=(0x0; 0xff0))"},
> > >
> > > Sorry for the later review.
> > > Would "int" be more precise and less verbose than "scalar"?
> >
> > Could work as well, although in many places today we make use of the term "scalar",
> > so developers might be more familiar with it (and more consistent towards the
> > verifier type internals).
>
> I was focusing more on users who will see these logs and
> would have to interpret them.
> I suspect the ratio is 1 developer to 1000 users.
> The users have to fight the verifier quite a bit to make the program pass.
> I was thinking about "i64" too. That's what llvm is using,
> but it's less clear than "int", which should be obvious
> for users since they write progs in C.
> It's also shorter.
>
> I can live with "scalar" though.

Imho, 'scalar' fits best. 'i64' would still be better then 'int' as it could imply
different width. Anyway, can push it out in a bit.