On 19.04.2023 18:18, Luca Weiss wrote: > On Mittwoch, 19. April 2023 18:12:04 CEST Konrad Dybcio wrote: >> On 19.04.2023 18:00, Luca Weiss wrote: >>> Hi Konrad, >>> >>> On Montag, 30. Jänner 2023 21:37:29 CEST Luca Weiss wrote: >>>> On Montag, 30. Jänner 2023 19:42:51 CET Konrad Dybcio wrote: >>>>> On 30.01.2023 19:36, Luca Weiss wrote: >>>>>> On Montag, 30. Jänner 2023 19:30:04 CET Konrad Dybcio wrote: >>>>>>> On 30.01.2023 19:20, luca@xxxxxxxxx wrote: >>>>>>>> From: Craig Tatlor <ctatlor97@xxxxxxxxx> >>>>>>>> >>>>>>>> The qfprom actually starts at 0xfc4b8000 instead of 0xfc4bc000 as >>>>>>>> defined previously. Adjust the tsens offsets accordingly. >>>>>>>> >>>>>>>> [luca@xxxxxxxxx: extract to standalone patch] >>>>>>>> >>>>>>>> Fixes: c59ffb519357 ("arm: dts: msm8974: Add thermal zones, tsens and >>>>>>>> qfprom nodes") Signed-off-by: Craig Tatlor <ctatlor97@xxxxxxxxx> >>>>>>>> Signed-off-by: Luca Weiss <luca@xxxxxxxxx> >>>>>>>> --- >>>>>>> >>>>>>> Isn't this a raw vs ecc-corrected values problem? >>>>>> >>>>>> Not quite sure what you mean. >>>>> >>>>> The QFPROM is split into two parts: one where raw values >>>>> are stored, and the other one where ECC-corrected copies >>>>> of them reside. Usually it's at offset of 0x4000. We should >>>>> generally be using the ECC-corrected ones, because.. well.. >>>>> they are ECC-corrected.. You may want to check if the >>>>> fuse you're adding reads the same value at +0x4000. >>>> >>>> Yeah that actually seems to work... >>>> >>>> But downstream's using this +0x4000 only for tsens it seems >>>> >>>> <0xfc4bc000 0x1000> as "tsens_eeprom_physical" >>>> >>>> qcom,clock-krait-8974 is using this: >>>> <0xfc4b80b0 0x08> as "efuse" >>>> >>>> Also seems HDMI driver is using a mix for HDCP stuff >>>> >>>> drivers/video/msm/mdss/mdss_hdmi_util.h: >>>> /* QFPROM Registers for HDMI/HDCP */ >>>> #define QFPROM_RAW_FEAT_CONFIG_ROW0_LSB (0x000000F8) >>>> #define QFPROM_RAW_FEAT_CONFIG_ROW0_MSB (0x000000FC) >>>> #define HDCP_KSV_LSB (0x000060D8) >>>> #define HDCP_KSV_MSB (0x000060DC) >>>> >>>> Any clue why Qualcomm used it this way in downstream? I'd rather not >>>> deviate too much if not for a good reason... >>> >>> Any comments on the above? >> >> This thread got burried to deep in the mailbox! >> >> I see two reasons why they could be using the uncorrected region: >> - their generators are messed up in general >> >> - they may have had an early chip revision once where there were >> problems with this and their generators were messed up to >> accommodate for it and everybody forgot to fix that >> >> No other good explanations as far as I'm aware! > > So, resolution is to use the offsets as declared in downstream, so take this > patch to have the full range available? No, the correct resolution to "fix QFPROM reg" would be to increase the size to 0x7000-0x4000 = 0x3000, as we should be using the ECC-corrected entries. Konrad > > Regards > Luca > > >