On 08/06/2020 13:32, Christian Hewitt wrote:
>
>> On 8 Jun 2020, at 12:58 pm, Neil Armstrong <[email protected]> wrote:
>>
>>> The current value is taken from Amlogic's 4.9 bsp kernel which appears
>>> to use the wrong ID. For comparison, here's before/after:
>>>
>>> [ 0.152237] soc soc0: Amlogic Meson SM1 (Unknown) Revision 2b:c (10:2)
>>> Detected
>>> [ 0.152463] soc soc0: Amlogic Meson SM1 (S905X3) Revision 2b:c (10:2)
>>> Detected
>>
>> For SM1 we have weird values, here are the one we gen on KernelCI:
>> SEI610 (S905D3):
>> Amlogic Meson SM1 (Unknown) Revision 2b:c (50:2)
>>
>> VIM3L (S905D3)
>> Amlogic Meson SM1 (Unknown) Revision 2b:c (4:2)
>>
>> And you report now:
>> Odroid-C4 (S905X4)
>> Amlogic Meson SM1 (Unknown) Revision 2b:c (10:2)
>>
>> And at the time you reported:
>> VIM3L (S905D3)
>> Amlogic Meson SM1 (S905D3) Revision 2b:c (b0:2)
>>
>> So I'm quite lost here….
>
> In [1] you added the SM1 SoC ID and S905X3 ID { "S905X3", 0x2b, 0x5, 0xf }
> which matches the BSP kernel source.
>
> In [2] I added a second S905X3 ID { "S905X3", 0x2b, 0xb0, 0xf2 } for VIM3L,
> but
> this was a mistake (the chip is S905D3) so [3] corrected it to { "S905D3",
> 0x2b,
> 0xb0, 0xf0 } which matches my VIM3L board. If the VIM3L’s in KernelCI now show
> "2b:c (4:2)” it looks there has been a package change (silicon fixes?) and we
> need to add another S905D3 package ID.
>
> In the last week I helped three different users install LE on S905X3 Android
> box devices, all of which declare "2b:c (10:2)” as the SoC ID, which matches
> the only S905X3 device I own (Odroid C4). So I assumed the BSP is wrong and
> sent this fix patch.
>
> If SEI610 is S905X3 (implied by the original SM1 submission from you) KernelCI
> output shows the current ID is wrong (‘Unknown’) so we probably need to fix it
> with { "S905X3", 0x2b, 0x50, 0xf0 } then add new S905D3 (VIM3L) and S905X3 (as
> per this patch) IDs.
>
> If SEI610 is S905D3 (as stated in this thread) the fix in this patch is still
> correct, but we also need to send a second patch to add two new S905D3 package
> IDs (SEI610 + VIM3L).
>
> I’ll ping you off-list to discuss.
OK, we have 5 different values :
S905X3 05 0000 0101 (first SEI610 support, early samples)
S905X3 10 0001 0000 (recent S905X3 android boxes, inc Odroid-C4)
S905X3 50 0101 0000 (SEI610 kernelci report, new SEI610 revision)
S905D3 04 0000 0100 (VIM3L in kernelci)
S905D3 b0 1011 0000 (VIM3L)
Seems the early S905X3/S905D3 samples used the low 4 bits to setup the revision.
And the mass-production ones uses the higher 4 bits.
So we can detect all these with the following bits & mask:
/* Early SM1 Samples */
{ "S905X3", 0x2b, 0x5 0xf5 },
{ "S905D3", 0x2b, 0x4 0xf5 },
/* Mass production SM1 */
{ "S905X3", 0x2b, 0x10, 0x3f },
{ "S905D3", 0x2b, 0x30 0x3f },
Neil
>
> Christian
>
> [1] c9cc9bec36d0 ("soc: amlogic: meson-gx-socinfo: Add SM1 and S905X3 ID”)
> [2] 1d7c541b8a5b ("soc: amlogic: meson-gx-socinfo: Add S905X3 ID for VIM3L”)
> [3] fdfc6997bd08 ("soc: amlogic: meson-gx-socinfo: Fix S905D3 ID for VIM3L”)
>
>
