On 4/30/25 6:19 PM, neil.armstr...@linaro.org wrote:
> On 30/04/2025 17:36, Konrad Dybcio wrote:
>> On 4/30/25 4:49 PM, neil.armstr...@linaro.org wrote:
>>> On 30/04/2025 15:09, Konrad Dybcio wrote:
>>>> On 4/30/25 2:49 PM, neil.armstr...@linaro.org wrote:
>>>>> On 30/04/2025 14:35, Konrad Dybcio wrote:
>>>>>> On 4/30/25 2:26 PM, neil.armstr...@linaro.org wrote:
>>>>>>> Hi,
>>>>>>>
>>>>>>> On 30/04/2025 13:34, Konrad Dybcio wrote:
>>>>>>>> From: Konrad Dybcio <konrad.dyb...@linaro.org>
>>>>>>>>
>>>>>>>> Add speebin data for A740, as found on SM8550 and derivative SoCs.
>>>>>>>>
>>>>>>>> For non-development SoCs it seems that "everything except FC_AC, FC_AF
>>>>>>>> should be speedbin 1", but what the values are for said "everything"
>>>>>>>> are
>>>>>>>> not known, so that's an exercise left to the user..
>>>>>>>>
>>>>>>>> Reviewed-by: Dmitry Baryshkov <dmitry.barysh...@linaro.org>
>>>>>>>> Signed-off-by: Konrad Dybcio <konrad.dyb...@linaro.org>
>>>>>>>> ---
>>>>>>>> drivers/gpu/drm/msm/adreno/a6xx_catalog.c | 8 ++++++++
>>>>>>>> 1 file changed, 8 insertions(+)
>>>>>>>>
>>>>>>>> diff --git a/drivers/gpu/drm/msm/adreno/a6xx_catalog.c
>>>>>>>> b/drivers/gpu/drm/msm/adreno/a6xx_catalog.c
>>>>>>>> index
>>>>>>>> 53e2ff4406d8f0afe474aaafbf0e459ef8f4577d..61daa331567925e529deae5e25d6fb63a8ba8375
>>>>>>>> 100644
>>>>>>>> --- a/drivers/gpu/drm/msm/adreno/a6xx_catalog.c
>>>>>>>> +++ b/drivers/gpu/drm/msm/adreno/a6xx_catalog.c
>>>>>>>> @@ -11,6 +11,9 @@
>>>>>>>> #include "a6xx.xml.h"
>>>>>>>> #include "a6xx_gmu.xml.h"
>>>>>>>> +#include <linux/soc/qcom/smem.h>
>>>>>>>> +#include <linux/soc/qcom/socinfo.h>
>>>>>>>> +
>>>>>>>> static const struct adreno_reglist a612_hwcg[] = {
>>>>>>>> {REG_A6XX_RBBM_CLOCK_CNTL_SP0, 0x22222222},
>>>>>>>> {REG_A6XX_RBBM_CLOCK_CNTL2_SP0, 0x02222220},
>>>>>>>> @@ -1431,6 +1434,11 @@ static const struct adreno_info a7xx_gpus[] = {
>>>>>>>> },
>>>>>>>> .address_space_size = SZ_16G,
>>>>>>>> .preempt_record_size = 4192 * SZ_1K,
>>>>>>>> + .speedbins = ADRENO_SPEEDBINS(
>>>>>>>> + { ADRENO_SKU_ID(SOCINFO_FC_AC), 0 },
>>>>>>>> + { ADRENO_SKU_ID(SOCINFO_FC_AF), 0 },
>>>>>>>> + /* Other feature codes (on prod SoCs) should match to
>>>>>>>> speedbin 1 */
>>>>>>>
>>>>>>> I'm trying to understand this sentence. because reading patch 4, when
>>>>>>> there's no match
>>>>>>> devm_pm_opp_set_supported_hw() is simply never called so how can it
>>>>>>> match speedbin 1 ?
>>>>>>
>>>>>> What I'm saying is that all other entries that happen to be possibly
>>>>>> added down the line are expected to be speedbin 1 (i.e. BIT(1))
>>>>>>
>>>>>>> Before this change the fallback was speedbin = BIT(0), but this
>>>>>>> disappeared.
>>>>>>
>>>>>> No, the default was to allow speedbin mask ~(0U)
>>>>>
>>>>> Hmm no:
>>>>>
>>>>> supp_hw = fuse_to_supp_hw(info, speedbin);
>>>>>
>>>>> if (supp_hw == UINT_MAX) {
>>>>> DRM_DEV_ERROR(dev,
>>>>> "missing support for speed-bin: %u. Some OPPs may not be
>>>>> supported by hardware\n",
>>>>> speedbin);
>>>>> supp_hw = BIT(0); /* Default */
>>>>> }
>>>>>
>>>>> ret = devm_pm_opp_set_supported_hw(dev, &supp_hw, 1);
>>>>> if (ret)
>>>>> return ret;
>>>>
>>>> Right, that's my own code even..
>>>>
>>>> in any case, the kernel can't know about the speed bins that aren't
>>>> defined and here we only define bin0, which doesn't break things
>>>>
>>>> the kernel isn't aware about hw with bin1 with or without this change
>>>> so it effectively doesn't matter
>>>
>>> But it's regression for the other platforms, where before an unknown SKU
>>> mapped to supp_hw=BIT(0)
>>>
>>> Not calling devm_pm_opp_set_supported_hw() is a major regression,
>>> if the opp-supported-hw is present, the OPP will be rejected:
>>
>> A comment in patch 4 explains that. We can either be forwards or backwards
>> compatible (i.e. accept a limited amount of
>> speedbin_in_driver x speedbin_in_dt combinations)
>
> I have a hard time understanding the change, please be much more verbose
> in the cover letter and commit messages.
>
> The fact that you do such a large change in the speedbin policy in patch 4
> makes it hard to understand why it's needed in the first place.
>
> Finally I'm very concerned that "old" SM8550 DT won't work on new kernels,
> this is frankly unacceptable, and this should be addressed in the first
> place.
>
> The nvmem situation was much simple, where we considered we added the nvmem
> property at the same time as opp-supported-hw in OPPs, but it's no more the
> case.
>
> So I think the OPP API should probably be extended to address this situation
> first, since if we do not have the opp-supported-hw in OPPs, all OPPs are
> safe.
>
> So this code:
> count = of_property_count_u32_elems(np, "opp-supported-hw");
> if (count <= 0 || count % levels) {
> dev_err(dev, "%s: Invalid opp-supported-hw property (%d)\n",
> __func__, count);
> return false;
> }
> should return true in this specific case, like a supported_hw_failsafe mode.
Not really. opp-supported-hws = <BIT(0)> usually translates to the *fastest*
bin in our case, so perhaps that change I made previously to default to it
wasn't the wisest. In other words, all slower SKUs that weren't added to the
kernel catalog & dt are potentially getting overclocked, which is no bueno.
That is not always the case, but it most certainly has been for a number of
years.
Old DTs in this case would be DTs lacking opp-supported-hw with the kernel
having speedbin tables. The inverse ("too new DTs") case translates into
"someone put some unexpected stuff in dt and the kernel has no idea what
to do with it".
In this context, old DTs would continue to work after patch 4, as the first
early return in adreno_set_speedbin() takes care of that.
Konrad
