[PATCH v4 1/6] ARM: shmobile: r8a7740 dtsi: Add L2 cache-controller node

[Geert Uytterhoeven]

Hi Sudeep,

[reviving this old thread, now we're two merge windows further]

On Fri, Aug 7, 2015 at 11:45 AM, Sudeep Holla <sudeep.holla at arm.com> wrote:
> On 06/08/15 17:21, Geert Uytterhoeven wrote:
>> On Wed, Aug 5, 2015 at 12:58 PM, Sudeep Holla <sudeep.holla at arm.com>
>> wrote:
>>> On 05/08/15 11:44, Geert Uytterhoeven wrote:
>>>> On Wed, Aug 5, 2015 at 11:34 AM, Sudeep Holla <sudeep.holla at arm.com>
>>>> wrote:
>
>
> [..]
>
>>>>> Any particular reason whey you need all this cache-* properties ? Is
>>>>
>>>> To describe the cache as good as possible.
>>>
>>> Why if you can probe it ? IMO DT is mostly useful to describe things
>>> that can't be probed/discovered using hardware.
>>>
>>>>> something broken on these SoCs ? We should be able to get most of these
>>>>> information from the SoC(reading some registers). It's good to avoid
>>>>> passing them via DT if they can be discovered from hardware.
>>>>
>>>> So we have all these documented properties in
>>>> Documentation/devicetree/bindings/arm/l2cc.txt, but they're not meant to
>>>> be used?
>>>
>>> No I didn't mean that, I just wanted to know if they can't be probed due
>>> to some hardware issue. It would avoid issues with wrong DTs especially
>>> if they are not so easy to upgrade.
>>
>> I think it works just fine without them.
>
> Yes, in general if you specify a value in DT that can be probed, its
> usually to override the probed value(useful if there is some h/w errata)...
>
>> Should I drop all cache-* properties marked optional in
>> Documentation/devicetree/bindings/arm/l2cc.txt?
>> That would be cache-size, cache-sets, cache-block-size, and
>> cache-line-size.
>
> ... however if you incorrect values by mistake, then it's problematic
> even if h/w provides correct value.
>
>> What about the L1 cache? I know Linux uses none of the d-cache-*
>> and i-cache-* properties.
>
> Same there, IIRC PPC use them, but on ARM I think so far the need has
> not arise.
>
> Just to re-iterate myself, I am not against adding them, but it's not
> really needed. I just wanted to know if there was any h/w issue.

AFAIK, there's nothing to be overridden. The cache seems to be configured in
the exact same way with and without cache-size, cache-sets, cache-block-size,
and cache-line-size.

With:

    L2C OF: override cache size: 262144 bytes (256KB)
    L2C OF: override line size: 32 bytes
    L2C OF: override way size: 32768 bytes (32KB)
    L2C OF: override associativity: 8
    L2C: DT/platform modifies aux control register: 0x02040000 -> 0x02440000
    L2C: DT/platform tries to modify or specify cache size
    L2C-310 erratum 769419 enabled
    L2C-310 enabling early BRESP for Cortex-A9
    L2C-310 full line of zeros enabled for Cortex-A9
    L2C-310 dynamic clock gating enabled, standby mode enabled
    L2C-310 cache controller enabled, 8 ways, 256 kB
    L2C-310: CACHE_ID 0x410000c7, AUX_CTRL 0x46440001

Without:

    L2C: DT/platform modifies aux control register: 0x02040000 -> 0x02440000
    L2C-310 erratum 769419 enabled
    L2C-310 enabling early BRESP for Cortex-A9
    L2C-310 full line of zeros enabled for Cortex-A9
    L2C-310 dynamic clock gating enabled, standby mode enabled
    L2C-310 cache controller enabled, 8 ways, 256 kB
    L2C-310: CACHE_ID 0x410000c7, AUX_CTRL 0x46440001

Hence I'll drop cache-size, cache-sets, cache-block-size, and cache-line-size,
for both unified L2 and L1 I/D caches.

Gr{oetje,eeting}s,

                        Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert at linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds