Re: [PATCH v4 0/8] Allwinner H6 Mali GPU support

[Robin Murphy <robin.murphy@arm.com>]

On 2019-05-14 10:22 pm, Clément Péron wrote:
> Hi,
> 
> On Tue, 14 May 2019 at 17:17, Clément Péron <peron.clem@gmail.com> wrote:
>>
>> Hi,
>>
>> On Tue, 14 May 2019 at 12:29, Neil Armstrong <narmstrong@baylibre.com> wrote:
>>>
>>> Hi,
>>>
>>> On 13/05/2019 17:14, Daniel Vetter wrote:
>>>> On Sun, May 12, 2019 at 07:46:00PM +0200, peron.clem@gmail.com wrote:
>>>>> From: Clément Péron <peron.clem@gmail.com>
>>>>>
>>>>> Hi,
>>>>>
>>>>> The Allwinner H6 has a Mali-T720 MP2. The drivers are
>>>>> out-of-tree so this series only introduce the dt-bindings.
>>>>
>>>> We do have an in-tree midgard driver now (since 5.2). Does this stuff work
>>>> together with your dt changes here?
>>>
>>> No, but it should be easy to add.
>> I will give it a try and let you know.
> Added the bus_clock and a ramp delay to the gpu_vdd but the driver
> fail at probe.
> 
> [    3.052919] panfrost 1800000.gpu: clock rate = 432000000
> [    3.058278] panfrost 1800000.gpu: bus_clock rate = 100000000
> [    3.179772] panfrost 1800000.gpu: mali-t720 id 0x720 major 0x1
> minor 0x1 status 0x0
> [    3.187432] panfrost 1800000.gpu: features: 00000000,10309e40,
> issues: 00000000,21054400
> [    3.195531] panfrost 1800000.gpu: Features: L2:0x07110206
> Shader:0x00000000 Tiler:0x00000809 Mem:0x1 MMU:0x00002821 AS:0xf
> JS:0x7
> [    3.207178] panfrost 1800000.gpu: shader_present=0x3 l2_present=0x1
> [    3.238257] panfrost 1800000.gpu: Fatal error during GPU init
> [    3.244165] panfrost: probe of 1800000.gpu failed with error -12
> 
> The ENOMEM is coming from "panfrost_mmu_init"
> alloc_io_pgtable_ops(ARM_MALI_LPAE, &pfdev->mmu->pgtbl_cfg,
>                                           pfdev);
> 
> Which is due to a check in the pgtable alloc "cfg->ias != 48"
> arm-lpae io-pgtable: arm_mali_lpae_alloc_pgtable cfg->ias 33 cfg->oas 40
> 
> DRI stack is totally new for me, could you give me a little clue about
> this issue ?

Heh, this is probably the one bit which doesn't really count as "DRI stack".

That's merely a somewhat-conservative sanity check - I'm pretty sure it 
*should* be fine to change the test to "cfg->ias > 48" (io-pgtable 
itself ought to cope). You'll just get to be the first to actually test 
a non-48-bit configuration here :)

Robin.