Re: [PATCH net-next v3 1/2] net: mscc: ocelot: Add support for tcam

[Vladimir Oltean]

On Thu, 23 Apr 2020 at 11:29, Horatiu Vultur
<horatiu.vultur@xxxxxxxxxxxxx> wrote:
>
> The 04/23/2020 00:26, Vladimir Oltean wrote:
> > EXTERNAL EMAIL: Do not click links or open attachments unless you know the content is safe
> >
> > Hi Horatiu,
> >
> > On Fri, 31 May 2019 at 10:18, Horatiu Vultur
> > <horatiu.vultur@xxxxxxxxxxxxx> wrote:
> > >
> > > Add ACL support using the TCAM. Using ACL it is possible to create rules
> > > in hardware to filter/redirect frames.
> > >
> > > Signed-off-by: Horatiu Vultur <horatiu.vultur@xxxxxxxxxxxxx>
> > > ---

[...]

> > > +
> > > +/* Calculate offsets for entry */
> > > +static void is2_data_get(struct vcap_data *data, int ix)
> > > +{
> > > +       u32 i, col, offset, count, cnt, base, width = vcap_is2.tg_width;
> > > +
> > > +       count = (data->tg_sw == VCAP_TG_HALF ? 2 : 4);
> > > +       col = (ix % 2);
> > > +       cnt = (vcap_is2.sw_count / count);
> > > +       base = (vcap_is2.sw_count - col * cnt - cnt);
> > > +       data->tg_value = 0;
> > > +       data->tg_mask = 0;
> > > +       for (i = 0; i < cnt; i++) {
> > > +               offset = ((base + i) * width);
> > > +               data->tg_value |= (data->tg_sw << offset);
> > > +               data->tg_mask |= GENMASK(offset + width - 1, offset);
> > > +       }
> > > +
> > > +       /* Calculate key/action/counter offsets */
> > > +       col = (count - col - 1);
> > > +       data->key_offset = (base * vcap_is2.entry_width) / vcap_is2.sw_count;
> > > +       data->counter_offset = (cnt * col * vcap_is2.counter_width);
> > > +       i = data->type;
> > > +       width = vcap_is2.action_table[i].width;
> > > +       cnt = vcap_is2.action_table[i].count;
> > > +       data->action_offset =
> > > +               (((cnt * col * width) / count) + vcap_is2.action_type_width);
> > > +}
> > > +

[...]

> > > +}
> > > +
> > > +static void is2_entry_set(struct ocelot *ocelot, int ix,
> > > +                         struct ocelot_ace_rule *ace)
> > > +{
> > > +       u32 val, msk, type, type_mask = 0xf, i, count;
> > > +       struct ocelot_ace_vlan *tag = &ace->vlan;
> > > +       struct ocelot_vcap_u64 payload = { 0 };
> > > +       struct vcap_data data = { 0 };
> > > +       int row = (ix / 2);
> > > +
> > > +       /* Read row */
> > > +       vcap_row_cmd(ocelot, row, VCAP_CMD_READ, VCAP_SEL_ALL);
> > > +       vcap_cache2entry(ocelot, &data);
> > > +       vcap_cache2action(ocelot, &data);
> > > +
> > > +       data.tg_sw = VCAP_TG_HALF;
> > > +       is2_data_get(&data, ix);
> > > +       data.tg = (data.tg & ~data.tg_mask);
> > > +       if (ace->prio != 0)
> > > +               data.tg |= data.tg_value;
> >
>
> Hi Vladimir,
>
> > This complicated piece of logic here populates the type-group for
> > subwords > 0 unconditionally, and the type-group for subword 0 only if
> > the ACE is enabled.
> >
> > tc filter add dev swp0 ingress protocol ip flower skip_sw src_ip
> > 192.168.1.1 action drop
> > [   34.172068] is2_entry_set: ace->prio 49152 data tg 0xaa
> > tc filter del dev swp0 ingress pref 49152
> > [   44.266662] is2_entry_set: ace->prio 0 data tg 0xa0
> >
> > What is the purpose of this? Why can't the entire data->tg be set to
> > zero when deleting it?
> I don't remember exactly but let me try:
>
> In case you have only one entry per row, then you could set the tg to
> have value 0. But in case you have 2 entries(use half keys), you need to
> set the tg to 0 only to the half entry that you delete.
>
> So for example if you have only 1 half entry at subword 1 then the tg
> should be 0xa0. Then when you add a new entry on the same row but at
> subword 0 then the tg should have the value 0xaa.
> The value 0xaa, comes from the fact that the type group for half entry
> is 0x2 and this needs to be set for each subword. And IS2 has 4 subwords
> therefore 0b10101010 = 0xaa.
>
> I hope this helps, if not I can look deeper in the code and see exactly.
>

Oh, right, so for half and quarter keys you need to not affect the
neighbour keys when modifying a row. That's exactly the information I
was looking for, thanks!

> > Is there any special meaning to a TCAM entry > with subword zero unused?
> >

[...]

>
> --
> /Horatiu

-Vladimir