[PATCH 2/2] drm/ast: Support AST2500

[Emil Velikov]

On 17 February 2017 at 21:36, Benjamin Herrenschmidt
<benh at kernel.crashing.org> wrote:
> On Fri, 2017-02-17 at 21:27 +0000, Emil Velikov wrote:
>> > Heh ok. I don't want to change that POST code too much as I'm not
>> > equipped to test it, but I'll have a look later today.
>> >
>>
>> Not sure why you opted for splitting each suggestion in separate
>> email, but it seems to have lead to a [serious] bugfix to go
>> unnoticed.
>> Namely:
>
> Dunno why either. I think I was distracted doing too many things at
> once.
>
>> > > > +static bool ddr_test_2500(struct ast_private *ast)
>> > > > +{
>> > > > +       ast_moutdwm(ast, 0x1E6E0074, 0x0000FFFF);
>> > > > +       ast_moutdwm(ast, 0x1E6E007C, 0xFF00FF00);
>> > > > +       if (mmc_test_burst(ast, 0) < 0)
>> > > > +               return false;
>> > > > +       if (mmc_test_burst(ast, 1) < 0)
>> > > > +               return false;
>> > > > +       if (mmc_test_burst(ast, 2) < 0)
>> > > > +               return false;
>> > > > +       if (mmc_test_burst(ast, 3) < 0)
>> > > > +               return false;
>> > > > +       if (mmc_test_single_2500(ast, 0) < 0)
>> > > > +               return false;
>> > > > +       return false;
>> > >
>> > > Final return should be true... either things got funny with v2 or
>> > > the
>> > > this patch was never tested ?
>
> As I said, never tested, I don't have the means, I'm waiting for Aspeed
> to test it, hopefully monday. I can test the basic function but not
> POST. I'll send a respin anyway.
>
> Note that the POST patch is purposefully at the end of the series, it
> can wait. The reason is that that code is only useful if the BMC has
> no code running on it, not even u-boot, and thus its memory controller
> needs to be remotely initialized by the host.
>
> Most servers out there have something running on the BMC and all my
> POWER9 systems won't boot without something on the BMC making them do
> so :-)
>
> So the POST patch can be merged later once it has had more massaging
> and testing.
>
Heh, I tried to be subtle here and not point fingers, but seemingly
that came out confusing ;-)

Afaict the code is 'broken' since it was sent to the ML by Y.C. Chen,
thus they likely [seemingly] haven't tested the POST either.
It's not my call how (or which parts) it should go it. Merely pointing
out what seems like bugs/issues.

Regards,
Emil