| Message ID | 20230303-topic-rpmcc_sleep-v2-0-ae80a325fe94@linaro.org (mailing list archive) |
|---|---|
| Headers | show |
| Series | SMD RPMCC sleep preparations | expand |
On 8.03.2023 22:35, Konrad Dybcio wrote: > v1 -> v2: > - Use CLK_IS_CRITICAL instead of leaving a clk enable vote, expand macros > to do so > - Fix the keepalive clocks for 8998 & 660 (CNoC -> PNoC, it was > confusingly named cnoc_periph downstream) > - Introduce .determinte_rate to ensure we don't set keepalive clocks' > rates below 19.2 MHz > - Add a (!conditional!) way to test the ultimate goal of all these changes > by essentially enabling unused clk cleanup through a dt property (for > legacy reasons) > > v2 was tested on: > > - MSM8996 Sony Kagura (can disable unused) > - MSM8998 Sony Maple (can disable unused with OOT icc) > - SM6375 Sony PDX225 (can disable unused with OOT icc) > > v1: https://lore.kernel.org/r/20230303-topic-rpmcc_sleep-v1-0-d9cfaf9b27a7@linaro.org > > This series brings support for a couple of things necessary for the full > system idle on SMD RPM SoCs, namely unused clk shutdown and keepalive > votes (permanent active votes that are required on certain clocks for the > platform to function). > > Tested on MSM8996 and SM6375, does not seem to introduce any additional > regressions. > > Keepalive clocks for other platforms were gathered by digging in old > downstream kernels, please give them a test. I have an implementation of rpmcc-within-icc ready(ish) locally. Turns out some SoCs need a keepalive (19.2MHz, active-only) vote on clocks that are NOT governed by interconnect.. So before we can disable clocks, both will need to be implemented.. ugh... I was hoping we could avoid having it in rpmcc.. Konrad > > Signed-off-by: Konrad Dybcio <konrad.dybcio@linaro.org> > --- > Konrad Dybcio (11): > dt-bindings: clock: qcom,rpmcc: Add a way to enable unused clock cleanup > clk: qcom: smd-rpm_ Make __DEFINE_CLK_SMD_RPM_BRANCH_PREFIX accept flags > clk: qcom: smd-rpm: Make DEFINE_CLK_SMD_RPM_BRANCH_A accept flags > clk: qcom: smd-rpm: Make BI_TCXO_AO critical > clk: qcom: smd-rpm: Make __DEFINE_CLK_SMD_RPM_PREFIX accept flags > clk: qcom: smd-rpm: Separate out a macro for defining an AO clock > clk: qcom: smd-rpm: Add support for keepalive votes > clk: qcom: smd-rpm: Introduce DEFINE_CLK_SMD_RPM_BUS_KEEPALIVE > clk: qcom: smd-rpm: Hook up PCNoC_0 keep_alive > clk: qcom: smd-rpm: Hook up CNoC_1 and SNoC_2 keep_alive > arm64: dts: qcom: msm8996: Enable rpmcc unused clk disablement > > Shawn Guo (3): > clk: qcom: smd-rpm: Add .is_enabled hook > clk: qcom: smd-rpm: Add .is_prepared hook > clk: qcom: smd-rpm: Mark clock enabled in clk_smd_rpm_handoff() > > .../devicetree/bindings/clock/qcom,rpmcc.yaml | 6 + > arch/arm64/boot/dts/qcom/msm8996.dtsi | 1 + > drivers/clk/qcom/clk-smd-rpm.c | 133 +++++++++++++++------ > 3 files changed, 106 insertions(+), 34 deletions(-) > --- > base-commit: fc31900c948610e7b5c2f15fb7795832c8325327 > change-id: 20230303-topic-rpmcc_sleep-d67aad9f3012 > > Best regards,
On Thu, Apr 20, 2023 at 03:50:16AM +0200, Konrad Dybcio wrote: > On 8.03.2023 22:35, Konrad Dybcio wrote: > > Keepalive clocks for other platforms were gathered by digging in old > > downstream kernels, please give them a test. > I have an implementation of rpmcc-within-icc ready(ish) locally. Turns out > some SoCs need a keepalive (19.2MHz, active-only) vote on clocks that > are NOT governed by interconnect.. So before we can disable clocks, > both will need to be implemented.. ugh... I was hoping we could avoid > having it in rpmcc.. > Can you give an example? Which clocks are affected on which SoC? Thanks, Stephan
On 20.04.2023 09:56, Stephan Gerhold wrote: > On Thu, Apr 20, 2023 at 03:50:16AM +0200, Konrad Dybcio wrote: >> On 8.03.2023 22:35, Konrad Dybcio wrote: >>> Keepalive clocks for other platforms were gathered by digging in old >>> downstream kernels, please give them a test. >> I have an implementation of rpmcc-within-icc ready(ish) locally. Turns out >> some SoCs need a keepalive (19.2MHz, active-only) vote on clocks that >> are NOT governed by interconnect.. So before we can disable clocks, >> both will need to be implemented.. ugh... I was hoping we could avoid >> having it in rpmcc.. >> > > Can you give an example? Which clocks are affected on which SoC? msm8998/sdm660 and PNoC Konrad > > Thanks, > Stephan
On Thu, Apr 20, 2023 at 11:36:24AM +0200, Konrad Dybcio wrote: > On 20.04.2023 09:56, Stephan Gerhold wrote: > > On Thu, Apr 20, 2023 at 03:50:16AM +0200, Konrad Dybcio wrote: > >> On 8.03.2023 22:35, Konrad Dybcio wrote: > >>> Keepalive clocks for other platforms were gathered by digging in old > >>> downstream kernels, please give them a test. > >> I have an implementation of rpmcc-within-icc ready(ish) locally. Turns out > >> some SoCs need a keepalive (19.2MHz, active-only) vote on clocks that > >> are NOT governed by interconnect.. So before we can disable clocks, > >> both will need to be implemented.. ugh... I was hoping we could avoid > >> having it in rpmcc.. > > Can you give an example? Which clocks are affected on which SoC? > msm8998/sdm660 and PNoC I don't see a PNoC for 8998/660, do you mean the "cnoc_periph_clk" downstream? Like the other NoCs it seems to be a RPM_BUS_CLK_TYPE, which means it does fit best into interconnect in my opinion. From a quick grep I don't see any usage of it in msm-4.4 downstream other than the active-only keepalive vote. So maybe you could just send that vote once in icc_rpm_smd and then ignore that clock (don't expose it at all)? Thanks, Stephan
On 20.04.2023 12:04, Stephan Gerhold wrote: > On Thu, Apr 20, 2023 at 11:36:24AM +0200, Konrad Dybcio wrote: >> On 20.04.2023 09:56, Stephan Gerhold wrote: >>> On Thu, Apr 20, 2023 at 03:50:16AM +0200, Konrad Dybcio wrote: >>>> On 8.03.2023 22:35, Konrad Dybcio wrote: >>>>> Keepalive clocks for other platforms were gathered by digging in old >>>>> downstream kernels, please give them a test. >>>> I have an implementation of rpmcc-within-icc ready(ish) locally. Turns out >>>> some SoCs need a keepalive (19.2MHz, active-only) vote on clocks that >>>> are NOT governed by interconnect.. So before we can disable clocks, >>>> both will need to be implemented.. ugh... I was hoping we could avoid >>>> having it in rpmcc.. >>> Can you give an example? Which clocks are affected on which SoC? >> msm8998/sdm660 and PNoC > > I don't see a PNoC for 8998/660, do you mean the "cnoc_periph_clk" It's the same, but Qualcomm kept changing the name every kernel release, so that's why we have 50 defines for the same thing upstream :( > downstream? Like the other NoCs it seems to be a RPM_BUS_CLK_TYPE, which > means it does fit best into interconnect in my opinion. From a quick > grep I don't see any usage of it in msm-4.4 downstream other than the > active-only keepalive vote. So maybe you could just send that vote once > in icc_rpm_smd and then ignore that clock (don't expose it at all)? Hm, perhaps that does sound like a good idea! As far as I understand, it's governed internally.. Older SoCs had a separate PNoC fabric exposed. Konrad > > Thanks, > Stephan
On 8.03.2023 22:35, Konrad Dybcio wrote: > v1 -> v2: > - Use CLK_IS_CRITICAL instead of leaving a clk enable vote, expand macros > to do so > - Fix the keepalive clocks for 8998 & 660 (CNoC -> PNoC, it was > confusingly named cnoc_periph downstream) > - Introduce .determinte_rate to ensure we don't set keepalive clocks' > rates below 19.2 MHz > - Add a (!conditional!) way to test the ultimate goal of all these changes > by essentially enabling unused clk cleanup through a dt property (for > legacy reasons) > > v2 was tested on: > > - MSM8996 Sony Kagura (can disable unused) > - MSM8998 Sony Maple (can disable unused with OOT icc) > - SM6375 Sony PDX225 (can disable unused with OOT icc) > > v1: https://lore.kernel.org/r/20230303-topic-rpmcc_sleep-v1-0-d9cfaf9b27a7@linaro.org > > This series brings support for a couple of things necessary for the full > system idle on SMD RPM SoCs, namely unused clk shutdown and keepalive > votes (permanent active votes that are required on certain clocks for the > platform to function). > > Tested on MSM8996 and SM6375, does not seem to introduce any additional > regressions. > > Keepalive clocks for other platforms were gathered by digging in old > downstream kernels, please give them a test. > > Signed-off-by: Konrad Dybcio <konrad.dybcio@linaro.org> > --- > Konrad Dybcio (11): > dt-bindings: clock: qcom,rpmcc: Add a way to enable unused clock cleanup > clk: qcom: smd-rpm_ Make __DEFINE_CLK_SMD_RPM_BRANCH_PREFIX accept flags > clk: qcom: smd-rpm: Make DEFINE_CLK_SMD_RPM_BRANCH_A accept flags > clk: qcom: smd-rpm: Make BI_TCXO_AO critical Stephen, parallel to all of the discussions, would you be willing to take patches 4-6 as they are? XO_A being critical is something that won't hurt without the rest. Konrad > clk: qcom: smd-rpm: Make __DEFINE_CLK_SMD_RPM_PREFIX accept flags > clk: qcom: smd-rpm: Separate out a macro for defining an AO clock > clk: qcom: smd-rpm: Add support for keepalive votes > clk: qcom: smd-rpm: Introduce DEFINE_CLK_SMD_RPM_BUS_KEEPALIVE > clk: qcom: smd-rpm: Hook up PCNoC_0 keep_alive > clk: qcom: smd-rpm: Hook up CNoC_1 and SNoC_2 keep_alive > arm64: dts: qcom: msm8996: Enable rpmcc unused clk disablement > > Shawn Guo (3): > clk: qcom: smd-rpm: Add .is_enabled hook > clk: qcom: smd-rpm: Add .is_prepared hook > clk: qcom: smd-rpm: Mark clock enabled in clk_smd_rpm_handoff() > > .../devicetree/bindings/clock/qcom,rpmcc.yaml | 6 + > arch/arm64/boot/dts/qcom/msm8996.dtsi | 1 + > drivers/clk/qcom/clk-smd-rpm.c | 133 +++++++++++++++------ > 3 files changed, 106 insertions(+), 34 deletions(-) > --- > base-commit: fc31900c948610e7b5c2f15fb7795832c8325327 > change-id: 20230303-topic-rpmcc_sleep-d67aad9f3012 > > Best regards,
Quoting Konrad Dybcio (2023-04-20 08:57:53) > > > > Konrad Dybcio (11): > > dt-bindings: clock: qcom,rpmcc: Add a way to enable unused clock cleanup > > > clk: qcom: smd-rpm_ Make __DEFINE_CLK_SMD_RPM_BRANCH_PREFIX accept flags > > clk: qcom: smd-rpm: Make DEFINE_CLK_SMD_RPM_BRANCH_A accept flags > > clk: qcom: smd-rpm: Make BI_TCXO_AO critical > Stephen, parallel to all of the discussions, would you be willing to > take patches 4-6 as they are? XO_A being critical is something that > won't hurt without the rest. Sure, can you resend just those in a series?
On 4/25/23 20:35, Stephen Boyd wrote: > Quoting Konrad Dybcio (2023-04-20 08:57:53) >> >>> Konrad Dybcio (11): >>> dt-bindings: clock: qcom,rpmcc: Add a way to enable unused clock cleanup >>> clk: qcom: smd-rpm_ Make __DEFINE_CLK_SMD_RPM_BRANCH_PREFIX accept flags >>> clk: qcom: smd-rpm: Make DEFINE_CLK_SMD_RPM_BRANCH_A accept flags >>> clk: qcom: smd-rpm: Make BI_TCXO_AO critical >> Stephen, parallel to all of the discussions, would you be willing to >> take patches 4-6 as they are? XO_A being critical is something that >> won't hurt without the rest. > Sure, can you resend just those in a series? Thanks, I'll do that after Connect! Konrad
v1 -> v2: - Use CLK_IS_CRITICAL instead of leaving a clk enable vote, expand macros to do so - Fix the keepalive clocks for 8998 & 660 (CNoC -> PNoC, it was confusingly named cnoc_periph downstream) - Introduce .determinte_rate to ensure we don't set keepalive clocks' rates below 19.2 MHz - Add a (!conditional!) way to test the ultimate goal of all these changes by essentially enabling unused clk cleanup through a dt property (for legacy reasons) v2 was tested on: - MSM8996 Sony Kagura (can disable unused) - MSM8998 Sony Maple (can disable unused with OOT icc) - SM6375 Sony PDX225 (can disable unused with OOT icc) v1: https://lore.kernel.org/r/20230303-topic-rpmcc_sleep-v1-0-d9cfaf9b27a7@linaro.org This series brings support for a couple of things necessary for the full system idle on SMD RPM SoCs, namely unused clk shutdown and keepalive votes (permanent active votes that are required on certain clocks for the platform to function). Tested on MSM8996 and SM6375, does not seem to introduce any additional regressions. Keepalive clocks for other platforms were gathered by digging in old downstream kernels, please give them a test. Signed-off-by: Konrad Dybcio <konrad.dybcio@linaro.org> --- Konrad Dybcio (11): dt-bindings: clock: qcom,rpmcc: Add a way to enable unused clock cleanup clk: qcom: smd-rpm_ Make __DEFINE_CLK_SMD_RPM_BRANCH_PREFIX accept flags clk: qcom: smd-rpm: Make DEFINE_CLK_SMD_RPM_BRANCH_A accept flags clk: qcom: smd-rpm: Make BI_TCXO_AO critical clk: qcom: smd-rpm: Make __DEFINE_CLK_SMD_RPM_PREFIX accept flags clk: qcom: smd-rpm: Separate out a macro for defining an AO clock clk: qcom: smd-rpm: Add support for keepalive votes clk: qcom: smd-rpm: Introduce DEFINE_CLK_SMD_RPM_BUS_KEEPALIVE clk: qcom: smd-rpm: Hook up PCNoC_0 keep_alive clk: qcom: smd-rpm: Hook up CNoC_1 and SNoC_2 keep_alive arm64: dts: qcom: msm8996: Enable rpmcc unused clk disablement Shawn Guo (3): clk: qcom: smd-rpm: Add .is_enabled hook clk: qcom: smd-rpm: Add .is_prepared hook clk: qcom: smd-rpm: Mark clock enabled in clk_smd_rpm_handoff() .../devicetree/bindings/clock/qcom,rpmcc.yaml | 6 + arch/arm64/boot/dts/qcom/msm8996.dtsi | 1 + drivers/clk/qcom/clk-smd-rpm.c | 133 +++++++++++++++------ 3 files changed, 106 insertions(+), 34 deletions(-) --- base-commit: fc31900c948610e7b5c2f15fb7795832c8325327 change-id: 20230303-topic-rpmcc_sleep-d67aad9f3012 Best regards,