Re: [PATCH v4 7/7] soc: renesas: Add L2 cache management for RZ/Five SoC

[Samuel Holland]

On 11/28/22 06:08, Lad, Prabhakar wrote:
> Hi Geert,
> 
> On Sun, Nov 27, 2022 at 9:55 AM Geert Uytterhoeven <geert@xxxxxxxxxxxxxx> wrote:
>>
>> Hi Prabhakar,
>>
>> On Sat, Nov 26, 2022 at 10:10 PM Lad, Prabhakar
>> <prabhakar.csengg@xxxxxxxxx> wrote:
>>> On Fri, Nov 25, 2022 at 7:43 PM Samuel Holland <samuel@xxxxxxxxxxxx> wrote:
>>>> On 11/24/22 11:22, Prabhakar wrote:
>>>>> From: Lad Prabhakar <prabhakar.mahadev-lad.rj@xxxxxxxxxxxxxx>
>>>>>
>>>>> On the AX45MP core, cache coherency is a specification option so it may
>>>>> not be supported. In this case DMA will fail. As a workaround, firstly we
>>>>> allocate a global dma coherent pool from which DMA allocations are taken
>>>>> and marked as non-cacheable + bufferable using the PMA region as specified
>>>>> in the device tree. Synchronization callbacks are implemented to
>>>>> synchronize when doing DMA transactions.
>>>>>
>>>>> The Andes AX45MP core has a Programmable Physical Memory Attributes (PMA)
>>>>> block that allows dynamic adjustment of memory attributes in the runtime.
>>>>> It contains a configurable amount of PMA entries implemented as CSR
>>>>> registers to control the attributes of memory locations in interest.
>>>>>
>>>>> Below are the memory attributes supported:
>>>>> * Device, Non-bufferable
>>>>> * Device, bufferable
>>>>> * Memory, Non-cacheable, Non-bufferable
>>>>> * Memory, Non-cacheable, Bufferable
>>>>> * Memory, Write-back, No-allocate
>>>>> * Memory, Write-back, Read-allocate
>>>>> * Memory, Write-back, Write-allocate
>>>>> * Memory, Write-back, Read and Write-allocate
>>>>>
>>>>> This patch adds support to configure the memory attributes of the memory
>>>>> regions as passed from the l2 cache node and exposes the cache management
>>>>> ops.
>>>>
>>>> Forgive my ignorance, but why do you need both a DMA pool and explicit
>>>> cache maintenance? Wouldn't the purpose of marking a memory region as
>>>> permanently non-cacheable be to avoid cache maintenance? And likewise,
>>>> if you are doing cache maintenance anyway, why does it matter if/how the
>>>> memory is cacheable?
>>>>
>>> "Memory, Non-cacheable, Bufferable" raises an AXI signal for
>>> transactions hence needing SW implementation for cache maintenance.
>>>
>>>>> More info about PMA (section 10.3):
>>>>> Link: http://www.andestech.com/wp-content/uploads/AX45MP-1C-Rev.-5.0.0-Datasheet.pdf
>>>>>
>>>>> Signed-off-by: Lad Prabhakar <prabhakar.mahadev-lad.rj@xxxxxxxxxxxxxx>
>>
>>>>> +static int ax45mp_configure_pma_regions(struct device_node *np)
>>>>> +{
>>>>> +     const char *propname = "andestech,pma-regions";
>>>>> +     u32 start, size, flags;
>>>>> +     unsigned int entry_id;
>>>>> +     unsigned int i;
>>>>> +     int count;
>>>>> +     int ret;
>>>>> +
>>>>> +     count = of_property_count_elems_of_size(np, propname, sizeof(u32) * 3);
>>>>> +     if (count < 0)
>>>>> +             return count;
>>>>> +
>>>>> +     if (count > AX45MP_MAX_PMA_REGIONS)
>>>>> +             return -EINVAL;
>>>>> +
>>>>> +     for (i = 0, entry_id = 0 ; entry_id < count ; i += 3, entry_id++) {
>>>>> +             of_property_read_u32_index(np, propname, i, &start);
>>>>> +             of_property_read_u32_index(np, propname, i + 1, &size);
>>>>> +             of_property_read_u32_index(np, propname, i + 2, &flags);
>>>>> +             ret = ax45mp_sbi_set_pma(start, size, flags, entry_id);
>>>>> +             if (!ret)
>>>>> +                     pr_err("Failed to setup PMA region 0x%x - 0x%x flags: 0x%x",
>>>>> +                            start, start + size, flags);
>>>>> +     }
>>>>> +
>>>>> +     return 0;
>>>>> +}
>>>>
>>>> If firmware support is required to set up these PMA regions, why is
>>>> Linux doing this at all? The firmware has access to the devicetree as
>>>> well. It can set this up before entering S-mode, and then you don't need
>>>> to expose this capability via an SBI extension. In fact, firmware could
>>>> generate the reserved-memory node based on these regions at runtime (or
>>>> vice versa).
>>>>
>>> That's a good point. I'll do some research on this and get back.
>>>
>>> Btw are there any existing examples where the firmware adds DT nodes?
>>
>> /memory, reserved-memory, optee on ARM, RPC status on R-Car Gen3/4, ...
>>
> On the TF-A we pass the FDT blob to u-boot and this does the magic.
> 
> On the RISC-V what would be the correct approach?
> - We setup the PMA regions in OpenSBI
> - We provide a vendor specific EXT to check if the PMA is setup
> - In u-boot ft_board_setup() callback add the reserved-memory node
> 
> Does the above approach sound good or is there a better approach I'm missing?

My suggestion was to fix up the DT in OpenSBI itself. See
lib/utils/fdt/fdt_fixup.c in the OpenSBI source tree. There is also a
platform hook for this. Then OpenSBI passes the FDT to U-Boot, and
U-Boot passes it on to Linux. No SBI extension is needed in that case.

If you optionally want your U-Boot to support loading a replacement FDT
from disk, then ft_board_setup() would need to copy the reserved-memory
nodes from U-Boot's control FDT to the loaded FDT. But this logic is the
same for all reserved-memory nodes, including the one OpenSBI adds
already. U-Boot has some code for this copying which you could reuse.

Regards,
Samuel