[PATCH 0/4] Retrieve information about DDR from SMEM

[Dmitry Baryshkov]

On Fri, Apr 11, 2025 at 12:52:32PM +0200, Konrad Dybcio wrote:
> On 4/11/25 12:50 PM, Dmitry Baryshkov wrote:
> > On Fri, Apr 11, 2025 at 12:03:03PM +0200, Konrad Dybcio wrote:
> >> On 4/11/25 11:57 AM, Dmitry Baryshkov wrote:
> >>> On Fri, 11 Apr 2025 at 12:49, Konrad Dybcio
> >>> <konrad.dybcio at oss.qualcomm.com> wrote:
> >>>>
> >>>> On 4/9/25 5:49 PM, Konrad Dybcio wrote:
> >>>>> On 4/9/25 5:44 PM, Dmitry Baryshkov wrote:
> >>>>>> On 09/04/2025 17:47, Konrad Dybcio wrote:
> >>>>>>> SMEM allows the OS to retrieve information about the DDR memory.
> >>>>>>> Among that information, is a semi-magic value called 'HBB', or Highest
> >>>>>>> Bank address Bit, which multimedia drivers (for hardware like Adreno
> >>>>>>> and MDSS) must retrieve in order to program the IP blocks correctly.
> >>>>>>>
> >>>>>>> This series introduces an API to retrieve that value, uses it in the
> >>>>>>> aforementioned programming sequences and exposes available DDR
> >>>>>>> frequencies in debugfs (to e.g. pass to aoss_qmp debugfs). More
> >>>>>>> information can be exposed in the future, as needed.
> >>>>>>
> >>>>>> I know that for some platforms HBB differs between GPU and DPU (as it's being programmed currently). Is there a way to check, which values are we going to program:
> >>>>>>
> >>>>>> - SM6115, SM6350, SM6375 (13 vs 14)
> >>>>
> >>>> SM6350 has INFO_V3
> >>>> SM6375 has INFO_V3_WITH_14_FREQS
> >>>
> >>> I'm not completely sure what you mean here. I pointed out that these
> >>> platforms disagreed upon the HBB value between the DPU/msm_mdss.c and
> >>> a6xx_gpu.c.
> >>> In some cases (a610/SM6115 and a619/SM6350) that was intentional to
> >>> fix screen corruption issues. I don't remember if it was the case for
> >>> QCM2290 or not.
> >>
> >> As I said below, I couldn't get a good answer yet, as the magic value
> >> is not provided explicitly and I'll hopefully be able to derive it from
> >> the available data
> > 
> > I see...
> > Is this data even supposed to be poked into? The foo_WITH_bar types
> > doesn't sound like a very stable API.
> 
> Yeah, it was designed with both the producer and consumer being part
> of a single codebase, always having the data structures in sync..

I feel somewhat worried about parsing those structures then. But... the
only viable alternative is to have an in-kernel list of possible
platform configurations and parse the /memory at foo/ddr_device_type
property.

-- 
With best wishes
Dmitry