| Message ID | 20230201080227.473547-2-jun.nie@linaro.org (mailing list archive) |
|---|---|
| State | Handled Elsewhere, archived |
| Headers | show |
| Series | [1/2] dt-bindings: interconnect: Add Qualcomm CCI dt-bindings | expand |
On 01/02/2023 09:02, Jun Nie wrote: > Cache Coherent Interconnect (CCI) is used by some Qualcomm SoCs. This > driver is introduced so that its freqency can be adjusted. And regulator > associated with opp table can be also adjusted accordingly which is > shared with cpu cluster. Please use scripts/get_maintainers.pl to get a list of necessary people and lists to CC. It might happen, that command when run on an older kernel, gives you outdated entries. Therefore please be sure you base your patches on recent Linux kernel. You need to Cc Qualcomm subarch maintainers. > > Signed-off-by: Jun Nie <jun.nie@linaro.org> > --- > drivers/devfreq/Kconfig | 9 +++ > drivers/devfreq/Makefile | 1 + > drivers/devfreq/qcom-cci.c | 162 +++++++++++++++++++++++++++++++++++++ Who is going to maintain this file/driver? > 3 files changed, 172 insertions(+) > create mode 100644 drivers/devfreq/qcom-cci.c > (...) > + > +static int qcom_cci_probe(struct platform_device *pdev) > +{ > + struct device *dev = &pdev->dev; > + struct qcom_cci *priv; > + const char *gov = DEVFREQ_GOV_USERSPACE; > + struct device_node *np = dev->of_node; > + struct nvmem_cell *speedbin_nvmem; > + int ret; > + u32 version; > + > + priv = devm_kzalloc(dev, sizeof(*priv), GFP_KERNEL); > + if (!priv) > + return -ENOMEM; > + > + priv->clk = devm_clk_get(dev, NULL); > + if (IS_ERR(priv->clk)) { > + ret = PTR_ERR(priv->clk); > + dev_err(dev, "failed to fetch clk: %d\n", ret); > + return ret; All these are just return dev_err_probe > + } > + platform_set_drvdata(pdev, priv); > + > + /* Check whether we have profiled speed version per chip */ > + speedbin_nvmem = of_nvmem_cell_get(np, NULL); > + if (IS_ERR(speedbin_nvmem)) > + return PTR_ERR(speedbin_nvmem); > + > + version = qcom_get_dev_version(speedbin_nvmem); > + dev_info(dev, "%s: set opp table version 0x%x\n", __func__, version); Drop __func__. > + > + nvmem_cell_put(speedbin_nvmem); > + ret = dev_pm_opp_set_supported_hw(dev, &version, 1); > + if (ret) { > + dev_err(dev, "Failed to set supported hardware\n"); return dev_err_probe > + return ret; > + } > + > + ret = dev_pm_opp_of_add_table(dev); > + if (ret < 0) { > + dev_err(dev, "failed to get OPP table\n"); > + return ret; return dev_err_probe > + } > + > + priv->profile.target = qcom_cci_target; > + priv->profile.exit = qcom_cci_exit; > + priv->profile.get_cur_freq = qcom_cci_get_cur_freq; > + priv->profile.initial_freq = clk_get_rate(priv->clk); > + > + priv->devfreq = devm_devfreq_add_device(dev, &priv->profile, > + gov, NULL); > + if (IS_ERR(priv->devfreq)) { > + ret = PTR_ERR(priv->devfreq); > + dev_err(dev, "failed to add devfreq device: %d\n", ret); ret = dev_err_probe > + goto err; > + } > + > + return 0; > + > +err: > + dev_pm_opp_of_remove_table(dev); > + return ret; > +} > + > +static const struct of_device_id qcom_cci_of_match[] = { > + { .compatible = "qcom,msm8939-cci"}, > + { /* sentinel */ }, > +}; > +MODULE_DEVICE_TABLE(of, qcom_cci_of_match); > + > +static struct platform_driver qcom_cci_platdrv = { > + .probe = qcom_cci_probe, > + .driver = { > + .name = "qcom-cci-devfreq", > + .of_match_table = qcom_cci_of_match, > + }, > +}; > +module_platform_driver(qcom_cci_platdrv); > + > +MODULE_DESCRIPTION("QCOM cci frequency scaling driver"); > +MODULE_AUTHOR("Jun Nie <jun.nie@linaro.org>"); > +MODULE_LICENSE("GPL"); Best regards, Krzysztof
On 01/02/2023 08:02, Jun Nie wrote: > Cache Coherent Interconnect (CCI) is used by some Qualcomm SoCs. This > driver is introduced so that its freqency can be adjusted. And regulator > associated with opp table can be also adjusted accordingly which is > shared with cpu cluster. > > Signed-off-by: Jun Nie <jun.nie@linaro.org> Nice work. Thanks for taking the time to follow up on this. I had a look at the 4.19 out-of-tree version of this we have and this code looks just about right. With Krzysztof's comments addressed please add my Reviewed-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org>
Hi, On 23. 2. 1. 17:02, Jun Nie wrote: > Cache Coherent Interconnect (CCI) is used by some Qualcomm SoCs. This > driver is introduced so that its freqency can be adjusted. And regulator > associated with opp table can be also adjusted accordingly which is > shared with cpu cluster. > > Signed-off-by: Jun Nie <jun.nie@linaro.org> > --- > drivers/devfreq/Kconfig | 9 +++ > drivers/devfreq/Makefile | 1 + > drivers/devfreq/qcom-cci.c | 162 +++++++++++++++++++++++++++++++++++++ > 3 files changed, 172 insertions(+) > create mode 100644 drivers/devfreq/qcom-cci.c > > diff --git a/drivers/devfreq/Kconfig b/drivers/devfreq/Kconfig > index 9754d8b31621..6f3f1872f3fb 100644 > --- a/drivers/devfreq/Kconfig > +++ b/drivers/devfreq/Kconfig > @@ -130,6 +130,15 @@ config ARM_MEDIATEK_CCI_DEVFREQ > buck voltages and update a proper CCI frequency. Use the notification > to get the regulator status. > > +config ARM_QCOM_CCI_DEVFREQ > + tristate "QUALCOMM CCI DEVFREQ Driver" > + depends on ARCH_QCOM || COMPILE_TEST > + select DEVFREQ_GOV_PASSIVE > + help > + This adds a devfreq driver for Qualcomm Cache Coherent Interconnect which > + shares the same regulator with the cpu cluster. This driver can track a > + proper regulator state while CCI frequency is updated. Maybe, this driver use the passive governor because as this description, the regulator is shared with cpu cluster. But, as I commented below, you use the 'userspace' governor in probe. is it right? > + > config ARM_RK3399_DMC_DEVFREQ > tristate "ARM RK3399 DMC DEVFREQ Driver" > depends on (ARCH_ROCKCHIP && HAVE_ARM_SMCCC) || \ > diff --git a/drivers/devfreq/Makefile b/drivers/devfreq/Makefile > index bf40d04928d0..f2526d8c168d 100644 > --- a/drivers/devfreq/Makefile > +++ b/drivers/devfreq/Makefile > @@ -12,6 +12,7 @@ obj-$(CONFIG_ARM_EXYNOS_BUS_DEVFREQ) += exynos-bus.o > obj-$(CONFIG_ARM_IMX_BUS_DEVFREQ) += imx-bus.o > obj-$(CONFIG_ARM_IMX8M_DDRC_DEVFREQ) += imx8m-ddrc.o > obj-$(CONFIG_ARM_MEDIATEK_CCI_DEVFREQ) += mtk-cci-devfreq.o > +obj-$(CONFIG_ARM_QCOM_CCI_DEVFREQ) += qcom-cci.o > obj-$(CONFIG_ARM_RK3399_DMC_DEVFREQ) += rk3399_dmc.o > obj-$(CONFIG_ARM_SUN8I_A33_MBUS_DEVFREQ) += sun8i-a33-mbus.o > obj-$(CONFIG_ARM_TEGRA_DEVFREQ) += tegra30-devfreq.o > diff --git a/drivers/devfreq/qcom-cci.c b/drivers/devfreq/qcom-cci.c > new file mode 100644 > index 000000000000..21b5f133cddc > --- /dev/null > +++ b/drivers/devfreq/qcom-cci.c > @@ -0,0 +1,162 @@ > +// SPDX-License-Identifier: GPL-2.0 > +/* > + * Copyright 2023 Linaro Ltd. > + */ > + > +#include <linux/clk.h> > +#include <linux/devfreq.h> > +#include <linux/device.h> > +#include <linux/module.h> > +#include <linux/nvmem-consumer.h> > +#include <linux/of_device.h> > +#include <linux/pm_opp.h> > +#include <linux/platform_device.h> > +#include <linux/slab.h> > + > +#define SPEED_PVS(s, p) ((s << 16) | p) What it meaning of PVS? Could you add the comment for 'PVS' and 's', 'p'? > + > +struct qcom_cci { > + struct devfreq_dev_profile profile; > + struct devfreq *devfreq; > + struct clk *clk; > +}; > + > +static int qcom_cci_target(struct device *dev, > + unsigned long *freq, u32 flags) Actually, this line is not long. You can type it on one line as following: static int qcom_cci_target(struct device *dev, unsigned long *freq, u32 flags) > +{ > + struct dev_pm_opp *new_opp; > + int ret; As I mentioned belwo, this local variable is not needed if just return PTR_ERR(new_opp). > + > + new_opp = devfreq_recommended_opp(dev, freq, flags); > + if (IS_ERR(new_opp)) { > + ret = PTR_ERR(new_opp); > + dev_err(dev, "failed to get recommended opp: %d\n", ret); > + return ret; Better to add 'return PTR_ERR(new_opp)' without 'ret' local variable. > + } > + dev_pm_opp_put(new_opp); > + > + return dev_pm_opp_set_rate(dev, *freq); > +} > + > +static int qcom_cci_get_cur_freq(struct device *dev, unsigned long *freq) > +{ > + struct qcom_cci *priv = dev_get_drvdata(dev); > + > + *freq = clk_get_rate(priv->clk); > + > + return 0; > +} > + > +static int qcom_get_dev_version(struct nvmem_cell *speedbin_nvmem) > +{ > + int speed = 0, pvs = 0; > + u8 *speedbin; > + size_t len; > + > + speedbin = nvmem_cell_read(speedbin_nvmem, &len); > + if (IS_ERR(speedbin)) > + return PTR_ERR(speedbin); > + > + speed = (speedbin[0xc] >> 2) & 0x7; > + pvs = (speedbin[0x3] >> 5 & 0x1) | ((speedbin[0x6] >> 2 & 0x3) << 1); Actually, 0xc, 0x3, 0x7, 0x1 and so on. It is impossible to understand the meaning of this hex value. Plesae add the constant defintion for the readability. > + kfree(speedbin); > + > + /* Convert speedbin and PVS into version bit map */ > + switch (SPEED_PVS(speed, pvs)) { > + case SPEED_PVS(0, 0): return 0x1; > + case SPEED_PVS(2, 0): return 0x2; > + case SPEED_PVS(2, 2): return 0x4; > + case SPEED_PVS(4, 0): return 0x8; > + case SPEED_PVS(5, 0): return 0x10; > + case SPEED_PVS(5, 6): return 0x20; > + default: > + return 0x1;> + } > +} > + > +static void qcom_cci_exit(struct device *dev) > +{ > + dev_pm_opp_of_remove_table(dev); > +} > + > +static int qcom_cci_probe(struct platform_device *pdev) > +{ > + struct device *dev = &pdev->dev; > + struct qcom_cci *priv; > + const char *gov = DEVFREQ_GOV_USERSPACE; Even if you select 'DEVFREQ_GOV_PASSIVE' on Kconfig, Is it right that this driver use the userspace governor? > + struct device_node *np = dev->of_node; > + struct nvmem_cell *speedbin_nvmem; > + int ret; > + u32 version; > + > + priv = devm_kzalloc(dev, sizeof(*priv), GFP_KERNEL); > + if (!priv) > + return -ENOMEM; > + > + priv->clk = devm_clk_get(dev, NULL); > + if (IS_ERR(priv->clk)) { > + ret = PTR_ERR(priv->clk); > + dev_err(dev, "failed to fetch clk: %d\n", ret); > + return ret; You can use dev_err_probe as following: return dev_err_probe(dev, ret, "failed to fetch clk: %d\n", ret); > + } > + platform_set_drvdata(pdev, priv); > + > + /* Check whether we have profiled speed version per chip */ > + speedbin_nvmem = of_nvmem_cell_get(np, NULL); > + if (IS_ERR(speedbin_nvmem)) > + return PTR_ERR(speedbin_nvmem); I recommend that you need to add the fail log with dev_err. > + > + version = qcom_get_dev_version(speedbin_nvmem); > + dev_info(dev, "%s: set opp table version 0x%x\n", __func__, version); Don't need to use '__func__' because dev_info will print the module name. It is enough. > + > + nvmem_cell_put(speedbin_nvmem); > + ret = dev_pm_opp_set_supported_hw(dev, &version, 1); > + if (ret) { > + dev_err(dev, "Failed to set supported hardware\n"); > + return ret; > + } > + > + ret = dev_pm_opp_of_add_table(dev); > + if (ret < 0) { > + dev_err(dev, "failed to get OPP table\n"); > + return ret; > + } > + > + priv->profile.target = qcom_cci_target; > + priv->profile.exit = qcom_cci_exit; > + priv->profile.get_cur_freq = qcom_cci_get_cur_freq; > + priv->profile.initial_freq = clk_get_rate(priv->clk); > + > + priv->devfreq = devm_devfreq_add_device(dev, &priv->profile, > + gov, NULL); > + if (IS_ERR(priv->devfreq)) { > + ret = PTR_ERR(priv->devfreq); > + dev_err(dev, "failed to add devfreq device: %d\n", ret); > + goto err; Need to goto 'err_remove_opp_table' instead of 'err'. > + } > + > + return 0; > + > +err: > + dev_pm_opp_of_remove_table(dev); > + return ret; For more correct exception handling, need to change it as following: err_remove_opp_table: dev_pm_opp_of_remove_table(dev); err: return ret; > +} > + > +static const struct of_device_id qcom_cci_of_match[] = { > + { .compatible = "qcom,msm8939-cci"}, > + { /* sentinel */ }, > +}; > +MODULE_DEVICE_TABLE(of, qcom_cci_of_match); > + > +static struct platform_driver qcom_cci_platdrv = { > + .probe = qcom_cci_probe, > + .driver = { > + .name = "qcom-cci-devfreq", > + .of_match_table = qcom_cci_of_match, > + }, > +}; > +module_platform_driver(qcom_cci_platdrv); > + > +MODULE_DESCRIPTION("QCOM cci frequency scaling driver"); cci is the abbreviation. You need to use the captical letter as following: MODULE_DESCRIPTION("QCOM CCI frequency scaling driver"); > +MODULE_AUTHOR("Jun Nie <jun.nie@linaro.org>"); > +MODULE_LICENSE("GPL");
On 01/02/2023 10:02, Jun Nie wrote: > Cache Coherent Interconnect (CCI) is used by some Qualcomm SoCs. This > driver is introduced so that its freqency can be adjusted. And regulator > associated with opp table can be also adjusted accordingly which is > shared with cpu cluster. > > Signed-off-by: Jun Nie <jun.nie@linaro.org> > --- > drivers/devfreq/Kconfig | 9 +++ > drivers/devfreq/Makefile | 1 + > drivers/devfreq/qcom-cci.c | 162 +++++++++++++++++++++++++++++++++++++ > 3 files changed, 172 insertions(+) > create mode 100644 drivers/devfreq/qcom-cci.c Could you please describe in some additional details what are you trying to achieve? Should the CCI frequency be scaled manually or does it follow the cluster frequency? Do clusters vote on the CCI frequency? I'm inclined to ask if it is possible to shift this to the cpufreq OPP tables?
On 01/02/2023 11:32, Dmitry Baryshkov wrote: > On 01/02/2023 10:02, Jun Nie wrote: >> Cache Coherent Interconnect (CCI) is used by some Qualcomm SoCs. This >> driver is introduced so that its freqency can be adjusted. And regulator >> associated with opp table can be also adjusted accordingly which is >> shared with cpu cluster. >> >> Signed-off-by: Jun Nie <jun.nie@linaro.org> >> --- >> drivers/devfreq/Kconfig | 9 +++ >> drivers/devfreq/Makefile | 1 + >> drivers/devfreq/qcom-cci.c | 162 +++++++++++++++++++++++++++++++++++++ >> 3 files changed, 172 insertions(+) >> create mode 100644 drivers/devfreq/qcom-cci.c > > Could you please describe in some additional details what are you trying > to achieve? Should the CCI frequency be scaled manually or does it > follow the cluster frequency? Do clusters vote on the CCI frequency? > > I'm inclined to ask if it is possible to shift this to the cpufreq OPP > tables? > Might not be so easy to just append CCI opps to the cluster frequency opps cci_cache: qcom,cci { compatible = "qcom,msm8939-cci"; clock-names = "devfreq_clk"; clocks = <&apcs2>; governor = "cpufreq"; operating-points-v2 = <&cci_opp_table>; power-domains = <&cpr>; power-domain-names = "cpr"; nvmem-cells = <&cpr_efuse_speedbin_pvs>; nvmem-cell-names = "cpr_efuse_speedbin_pvs"; }; devfreq-cpufreq { cci-cpufreq { target-dev = <&cci_cache>; cpu-to-dev-map-0 = < 200000 200000000 >, < 345600 200000000 >, < 400000 200000000 >, < 533330 297600000 >, < 800000 297600000 >, < 960000 297600000 >, < 1113600 297000000 >, < 1344000 595200000 >, < 1459200 595200000 >, < 1497600 595200000 >, < 1651200 595200000 >; cpu-to-dev-map-4 = < 200000 200000000 >, < 249600 200000000 >, < 499200 297600000 >, < 800000 297600000 >, < 998400 595200000 >, < 1113600 595200000 >; }; }; cci_opp_table: cci-opp-table { compatible = "operating-points-v2"; opp-200000000 { opp-hz = /bits/ 64 <200000000>; opp-supported-hw = <0x3f>; required-opps = <&cpr_opp3>; }; opp-297600000 { opp-hz = /bits/ 64 <297600000>; opp-supported-hw = <0x3f>; required-opps = <&cpr_opp12>; }; opp-400000000-cpr14 { opp-hz = /bits/ 64 <400000000>; opp-supported-hw = <0x1>; required-opps = <&cpr_opp14>; }; opp-400000000-cpr15 { opp-hz = /bits/ 64 <400000000>; opp-supported-hw = <0x3e>; required-opps = <&cpr_opp15>; }; opp-595200000 { opp-hz = /bits/ 64 <595200000>; opp-supported-hw = <0x3f>; required-opps = <&cpr_opp17>; }; }; cpr_opp_table: cpr-opp-table { compatible = "operating-points-v2-qcom-level"; cpr_opp1: opp1 { opp-hz = /bits/ 64 <200000000>; opp-level = <1>; qcom,opp-fuse-level = <1>; }; cpr_opp2: opp2 { opp-hz = /bits/ 64 <345600000>; opp-level = <2>; qcom,opp-fuse-level = <1>; }; cpr_opp3: opp3 { opp-hz = /bits/ 64 <400000000>; opp-level = <3>; qcom,opp-fuse-level = <1>; }; cpr_opp4: opp4 { opp-hz = /bits/ 64 <422400000>; opp-level = <4>; qcom,opp-fuse-level = <2>; }; cpr_opp5: opp5 { opp-hz = /bits/ 64 <499200000>; opp-level = <5>; qcom,opp-fuse-level = <2>; }; cpr_opp6: opp6 { opp-hz = /bits/ 64 <533330000>; opp-level = <6>; qcom,opp-fuse-level = <2>; }; cpr_opp7: opp7 { opp-hz = /bits/ 64 <652800000>; opp-level = <7>; qcom,opp-fuse-level = <2>; }; cpr_opp8: opp8 { opp-hz = /bits/ 64 <729600000>; opp-level = <8>; qcom,opp-fuse-level = <2>; }; cpr_opp9: opp9 { opp-hz = /bits/ 64 <800000000>; opp-level = <9>; qcom,opp-fuse-level = <2>; }; cpr_opp10: opp10 { opp-hz = /bits/ 64 <806400000>; opp-level = <10>; qcom,opp-fuse-level = <2>; }; cpr_opp11: opp11 { opp-hz = /bits/ 64 <883200000>; opp-level = <11>; qcom,opp-fuse-level = <2>; }; cpr_opp12: opp12 { opp-hz = /bits/ 64 <960000000>; opp-level = <12>; qcom,opp-fuse-level = <2>; }; }; --- bod
On Wed, 1 Feb 2023 at 13:46, Bryan O'Donoghue <bryan.odonoghue@linaro.org> wrote: > > On 01/02/2023 11:32, Dmitry Baryshkov wrote: > > On 01/02/2023 10:02, Jun Nie wrote: > >> Cache Coherent Interconnect (CCI) is used by some Qualcomm SoCs. This > >> driver is introduced so that its freqency can be adjusted. And regulator > >> associated with opp table can be also adjusted accordingly which is > >> shared with cpu cluster. > >> > >> Signed-off-by: Jun Nie <jun.nie@linaro.org> > >> --- > >> drivers/devfreq/Kconfig | 9 +++ > >> drivers/devfreq/Makefile | 1 + > >> drivers/devfreq/qcom-cci.c | 162 +++++++++++++++++++++++++++++++++++++ > >> 3 files changed, 172 insertions(+) > >> create mode 100644 drivers/devfreq/qcom-cci.c > > > > Could you please describe in some additional details what are you trying > > to achieve? Should the CCI frequency be scaled manually or does it > > follow the cluster frequency? Do clusters vote on the CCI frequency? > > > > I'm inclined to ask if it is possible to shift this to the cpufreq OPP > > tables? > > > > Might not be so easy to just append CCI opps to the cluster frequency opps > > cci_cache: qcom,cci { > compatible = "qcom,msm8939-cci"; > clock-names = "devfreq_clk"; > clocks = <&apcs2>; > governor = "cpufreq"; > operating-points-v2 = <&cci_opp_table>; > power-domains = <&cpr>; > power-domain-names = "cpr"; > nvmem-cells = <&cpr_efuse_speedbin_pvs>; > nvmem-cell-names = "cpr_efuse_speedbin_pvs"; > }; > > devfreq-cpufreq { > cci-cpufreq { > target-dev = <&cci_cache>; > cpu-to-dev-map-0 = > < 200000 200000000 >, > < 345600 200000000 >, > < 400000 200000000 >, > < 533330 297600000 >, > < 800000 297600000 >, > < 960000 297600000 >, > < 1113600 297000000 >, > < 1344000 595200000 >, > < 1459200 595200000 >, > < 1497600 595200000 >, > < 1651200 595200000 >; > cpu-to-dev-map-4 = > < 200000 200000000 >, > < 249600 200000000 >, > < 499200 297600000 >, > < 800000 297600000 >, > < 998400 595200000 >, > < 1113600 595200000 >; These should map to existing opp entries. I ended up doing the interconnect driver that maps a clock to the interconnect. Then I can use it in the cpu opp tables. > }; > }; > > cci_opp_table: cci-opp-table { > compatible = "operating-points-v2"; > > opp-200000000 { > opp-hz = /bits/ 64 <200000000>; > opp-supported-hw = <0x3f>; > required-opps = <&cpr_opp3>; And these should probably map to max(cpu's CPR opp, CCI's CPR opp). > }; > > opp-297600000 { > opp-hz = /bits/ 64 <297600000>; > opp-supported-hw = <0x3f>; > required-opps = <&cpr_opp12>; > }; > > opp-400000000-cpr14 { > opp-hz = /bits/ 64 <400000000>; > opp-supported-hw = <0x1>; > required-opps = <&cpr_opp14>; > }; > > opp-400000000-cpr15 { > opp-hz = /bits/ 64 <400000000>; > opp-supported-hw = <0x3e>; > required-opps = <&cpr_opp15>; > }; > > opp-595200000 { > opp-hz = /bits/ 64 <595200000>; > opp-supported-hw = <0x3f>; > required-opps = <&cpr_opp17>; > }; > }; > > > cpr_opp_table: cpr-opp-table { > compatible = "operating-points-v2-qcom-level"; > > cpr_opp1: opp1 { > opp-hz = /bits/ 64 <200000000>; > opp-level = <1>; > qcom,opp-fuse-level = <1>; > }; > cpr_opp2: opp2 { > opp-hz = /bits/ 64 <345600000>; > opp-level = <2>; > qcom,opp-fuse-level = <1>; > }; > cpr_opp3: opp3 { > opp-hz = /bits/ 64 <400000000>; > opp-level = <3>; > qcom,opp-fuse-level = <1>; > }; > cpr_opp4: opp4 { > opp-hz = /bits/ 64 <422400000>; > opp-level = <4>; > qcom,opp-fuse-level = <2>; > }; > cpr_opp5: opp5 { > opp-hz = /bits/ 64 <499200000>; > opp-level = <5>; > qcom,opp-fuse-level = <2>; > }; > cpr_opp6: opp6 { > opp-hz = /bits/ 64 <533330000>; > opp-level = <6>; > qcom,opp-fuse-level = <2>; > }; > cpr_opp7: opp7 { > opp-hz = /bits/ 64 <652800000>; > opp-level = <7>; > qcom,opp-fuse-level = <2>; > }; > cpr_opp8: opp8 { > opp-hz = /bits/ 64 <729600000>; > opp-level = <8>; > qcom,opp-fuse-level = <2>; > }; > cpr_opp9: opp9 { > opp-hz = /bits/ 64 <800000000>; > opp-level = <9>; > qcom,opp-fuse-level = <2>; > }; > cpr_opp10: opp10 { > opp-hz = /bits/ 64 <806400000>; > opp-level = <10>; > qcom,opp-fuse-level = <2>; > }; > cpr_opp11: opp11 { > opp-hz = /bits/ 64 <883200000>; > opp-level = <11>; > qcom,opp-fuse-level = <2>; > }; > cpr_opp12: opp12 { > opp-hz = /bits/ 64 <960000000>; > opp-level = <12>; > qcom,opp-fuse-level = <2>; > }; > }; > > --- > bod
On 01/02/2023 13:41, Dmitry Baryshkov wrote: >> cci-cpufreq { >> target-dev = <&cci_cache>; >> cpu-to-dev-map-0 = >> < 200000 200000000 >, >> < 345600 200000000 >, >> < 400000 200000000 >, >> < 533330 297600000 >, >> < 800000 297600000 >, >> < 960000 297600000 >, >> < 1113600 297000000 >, >> < 1344000 595200000 >, >> < 1459200 595200000 >, >> < 1497600 595200000 >, >> < 1651200 595200000 >; >> cpu-to-dev-map-4 = >> < 200000 200000000 >, >> < 249600 200000000 >, >> < 499200 297600000 >, >> < 800000 297600000 >, >> < 998400 595200000 >, >> < 1113600 595200000 >; > These should map to existing opp entries. > > I ended up doing the interconnect driver that maps a clock to the > interconnect. Then I can use it in the cpu opp tables. > Can you point us at what it is you are proposing ? --- bod
On 01/02/2023 16:45, Bryan O'Donoghue wrote: > On 01/02/2023 13:41, Dmitry Baryshkov wrote: >>> cci-cpufreq { >>> target-dev = <&cci_cache>; >>> cpu-to-dev-map-0 = >>> < 200000 200000000 >, >>> < 345600 200000000 >, >>> < 400000 200000000 >, >>> < 533330 297600000 >, >>> < 800000 297600000 >, >>> < 960000 297600000 >, >>> < 1113600 297000000 >, >>> < 1344000 595200000 >, >>> < 1459200 595200000 >, >>> < 1497600 595200000 >, >>> < 1651200 595200000 >; >>> cpu-to-dev-map-4 = >>> < 200000 200000000 >, >>> < 249600 200000000 >, >>> < 499200 297600000 >, >>> < 800000 297600000 >, >>> < 998400 595200000 >, >>> < 1113600 595200000 >; >> These should map to existing opp entries. >> >> I ended up doing the interconnect driver that maps a clock to the >> interconnect. Then I can use it in the cpu opp tables. >> > > Can you point us at what it is you are proposing ? https://patchwork.kernel.org/project/linux-arm-msm/patch/20230120061417.2623751-9-dmitry.baryshkov@linaro.org/ > > --- > bod
> > Signed-off-by: Jun Nie <jun.nie@linaro.org> > > --- > > drivers/devfreq/Kconfig | 9 +++ > > drivers/devfreq/Makefile | 1 + > > drivers/devfreq/qcom-cci.c | 162 +++++++++++++++++++++++++++++++++++++ > > Who is going to maintain this file/driver? > I will add myself as maintainer of this file. All other comments will be addressed in next version. Thanks! - Jun
On 01/02/2023 14:58, Dmitry Baryshkov wrote: > On 01/02/2023 16:45, Bryan O'Donoghue wrote: >> On 01/02/2023 13:41, Dmitry Baryshkov wrote: >>>> cci-cpufreq { >>>> target-dev = <&cci_cache>; >>>> cpu-to-dev-map-0 = >>>> < 200000 200000000 >, >>>> < 345600 200000000 >, >>>> < 400000 200000000 >, >>>> < 533330 297600000 >, >>>> < 800000 297600000 >, >>>> < 960000 297600000 >, >>>> < 1113600 297000000 >, >>>> < 1344000 595200000 >, >>>> < 1459200 595200000 >, >>>> < 1497600 595200000 >, >>>> < 1651200 595200000 >; >>>> cpu-to-dev-map-4 = >>>> < 200000 200000000 >, >>>> < 249600 200000000 >, >>>> < 499200 297600000 >, >>>> < 800000 297600000 >, >>>> < 998400 595200000 >, >>>> < 1113600 595200000 >; >>> These should map to existing opp entries. >>> >>> I ended up doing the interconnect driver that maps a clock to the >>> interconnect. Then I can use it in the cpu opp tables. >>> >> >> Can you point us at what it is you are proposing ? > > https://patchwork.kernel.org/project/linux-arm-msm/patch/20230120061417.2623751-9-dmitry.baryshkov@linaro.org/ > Is there no driver code too ?
Chanwoo Choi <cwchoi00@gmail.com> 于2023年2月1日周三 19:02写道： > > Hi, > > On 23. 2. 1. 17:02, Jun Nie wrote: > > Cache Coherent Interconnect (CCI) is used by some Qualcomm SoCs. This > > driver is introduced so that its freqency can be adjusted. And regulator > > associated with opp table can be also adjusted accordingly which is > > shared with cpu cluster. > > > > Signed-off-by: Jun Nie <jun.nie@linaro.org> > > --- > > drivers/devfreq/Kconfig | 9 +++ > > drivers/devfreq/Makefile | 1 + > > drivers/devfreq/qcom-cci.c | 162 +++++++++++++++++++++++++++++++++++++ > > 3 files changed, 172 insertions(+) > > create mode 100644 drivers/devfreq/qcom-cci.c > > > > diff --git a/drivers/devfreq/Kconfig b/drivers/devfreq/Kconfig > > index 9754d8b31621..6f3f1872f3fb 100644 > > --- a/drivers/devfreq/Kconfig > > +++ b/drivers/devfreq/Kconfig > > @@ -130,6 +130,15 @@ config ARM_MEDIATEK_CCI_DEVFREQ > > buck voltages and update a proper CCI frequency. Use the notification > > to get the regulator status. > > > > +config ARM_QCOM_CCI_DEVFREQ > > + tristate "QUALCOMM CCI DEVFREQ Driver" > > + depends on ARCH_QCOM || COMPILE_TEST > > + select DEVFREQ_GOV_PASSIVE > > + help > > + This adds a devfreq driver for Qualcomm Cache Coherent Interconnect which > > + shares the same regulator with the cpu cluster. This driver can track a > > + proper regulator state while CCI frequency is updated. > > Maybe, this driver use the passive governor because as this description, > the regulator is shared with cpu cluster. But, as I commented below, > you use the 'userspace' governor in probe. is it right? > > > + > > config ARM_RK3399_DMC_DEVFREQ > > tristate "ARM RK3399 DMC DEVFREQ Driver" > > depends on (ARCH_ROCKCHIP && HAVE_ARM_SMCCC) || \ > > diff --git a/drivers/devfreq/Makefile b/drivers/devfreq/Makefile > > index bf40d04928d0..f2526d8c168d 100644 > > --- a/drivers/devfreq/Makefile > > +++ b/drivers/devfreq/Makefile > > @@ -12,6 +12,7 @@ obj-$(CONFIG_ARM_EXYNOS_BUS_DEVFREQ) += exynos-bus.o > > obj-$(CONFIG_ARM_IMX_BUS_DEVFREQ) += imx-bus.o > > obj-$(CONFIG_ARM_IMX8M_DDRC_DEVFREQ) += imx8m-ddrc.o > > obj-$(CONFIG_ARM_MEDIATEK_CCI_DEVFREQ) += mtk-cci-devfreq.o > > +obj-$(CONFIG_ARM_QCOM_CCI_DEVFREQ) += qcom-cci.o > > obj-$(CONFIG_ARM_RK3399_DMC_DEVFREQ) += rk3399_dmc.o > > obj-$(CONFIG_ARM_SUN8I_A33_MBUS_DEVFREQ) += sun8i-a33-mbus.o > > obj-$(CONFIG_ARM_TEGRA_DEVFREQ) += tegra30-devfreq.o > > diff --git a/drivers/devfreq/qcom-cci.c b/drivers/devfreq/qcom-cci.c > > new file mode 100644 > > index 000000000000..21b5f133cddc > > --- /dev/null > > +++ b/drivers/devfreq/qcom-cci.c > > @@ -0,0 +1,162 @@ > > +// SPDX-License-Identifier: GPL-2.0 > > +/* > > + * Copyright 2023 Linaro Ltd. > > + */ > > + > > +#include <linux/clk.h> > > +#include <linux/devfreq.h> > > +#include <linux/device.h> > > +#include <linux/module.h> > > +#include <linux/nvmem-consumer.h> > > +#include <linux/of_device.h> > > +#include <linux/pm_opp.h> > > +#include <linux/platform_device.h> > > +#include <linux/slab.h> > > + > > +#define SPEED_PVS(s, p) ((s << 16) | p) > > What it meaning of PVS? Could you add the comment for 'PVS' and 's', 'p'? > This is what I have from Qualcomm code. Maybe others have more info about it. Process Voltage Scaling(PVS) Tables defines the voltage and frequency value based on the msm-id in SMEM and speedbin blown in the efuse combination. > > + > > +struct qcom_cci { > > + struct devfreq_dev_profile profile; > > + struct devfreq *devfreq; > > + struct clk *clk; > > +}; > > + > > +static int qcom_cci_target(struct device *dev, > > + unsigned long *freq, u32 flags) > > Actually, this line is not long. You can type it on one line as following: > > static int qcom_cci_target(struct device *dev, unsigned long *freq, u32 flags) > > > +{ > > + struct dev_pm_opp *new_opp; > > + int ret; > > As I mentioned belwo, this local variable is not needed > if just return PTR_ERR(new_opp). > > > + > > + new_opp = devfreq_recommended_opp(dev, freq, flags); > > + if (IS_ERR(new_opp)) { > > + ret = PTR_ERR(new_opp); > > + dev_err(dev, "failed to get recommended opp: %d\n", ret); > > + return ret; > > Better to add 'return PTR_ERR(new_opp)' without 'ret' local variable. > > > + } > > + dev_pm_opp_put(new_opp); > > + > > + return dev_pm_opp_set_rate(dev, *freq); > > +} > > + > > +static int qcom_cci_get_cur_freq(struct device *dev, unsigned long *freq) > > +{ > > + struct qcom_cci *priv = dev_get_drvdata(dev); > > + > > + *freq = clk_get_rate(priv->clk); > > + > > + return 0; > > +} > > + > > +static int qcom_get_dev_version(struct nvmem_cell *speedbin_nvmem) > > +{ > > + int speed = 0, pvs = 0; > > + u8 *speedbin; > > + size_t len; > > + > > + speedbin = nvmem_cell_read(speedbin_nvmem, &len); > > + if (IS_ERR(speedbin)) > > + return PTR_ERR(speedbin); > > + > > + speed = (speedbin[0xc] >> 2) & 0x7; > > + pvs = (speedbin[0x3] >> 5 & 0x1) | ((speedbin[0x6] >> 2 & 0x3) << 1); > > Actually, 0xc, 0x3, 0x7, 0x1 and so on. It is impossible to understand > the meaning of this hex value. Plesae add the constant defintion > for the readability. > Thanks for pointing it. The bit wise manipulation will be converted into nvmem offset and bit position in next version. So driver will read out value directly without bit shift/mask. Other comments will be adopted in next version. Thanks! - Jun
Dmitry Baryshkov <dmitry.baryshkov@linaro.org> 于2023年2月1日周三 19:32写道： > > On 01/02/2023 10:02, Jun Nie wrote: > > Cache Coherent Interconnect (CCI) is used by some Qualcomm SoCs. This > > driver is introduced so that its freqency can be adjusted. And regulator > > associated with opp table can be also adjusted accordingly which is > > shared with cpu cluster. > > > > Signed-off-by: Jun Nie <jun.nie@linaro.org> > > --- > > drivers/devfreq/Kconfig | 9 +++ > > drivers/devfreq/Makefile | 1 + > > drivers/devfreq/qcom-cci.c | 162 +++++++++++++++++++++++++++++++++++++ > > 3 files changed, 172 insertions(+) > > create mode 100644 drivers/devfreq/qcom-cci.c > > Could you please describe in some additional details what are you trying > to achieve? Should the CCI frequency be scaled manually or does it > follow the cluster frequency? Do clusters vote on the CCI frequency? > > I'm inclined to ask if it is possible to shift this to the cpufreq OPP > tables? > > -- > With best wishes > Dmitry > The plan is to add CCI opp node as required-opp property in CPU opp, so that CPU will vote CCI frequency when scaling CPU frequency. The interconnect bandwidth side is not addressed yet. This patch only address CCI frequency and power voltage requirement with help of device tree. - Jun
On 01/02/2023 17:17, Bryan O'Donoghue wrote: > On 01/02/2023 14:58, Dmitry Baryshkov wrote: >> On 01/02/2023 16:45, Bryan O'Donoghue wrote: >>> On 01/02/2023 13:41, Dmitry Baryshkov wrote: >>>>> cci-cpufreq { >>>>> target-dev = <&cci_cache>; >>>>> cpu-to-dev-map-0 = >>>>> < 200000 200000000 >, >>>>> < 345600 200000000 >, >>>>> < 400000 200000000 >, >>>>> < 533330 297600000 >, >>>>> < 800000 297600000 >, >>>>> < 960000 297600000 >, >>>>> < 1113600 297000000 >, >>>>> < 1344000 595200000 >, >>>>> < 1459200 595200000 >, >>>>> < 1497600 595200000 >, >>>>> < 1651200 595200000 >; >>>>> cpu-to-dev-map-4 = >>>>> < 200000 200000000 >, >>>>> < 249600 200000000 >, >>>>> < 499200 297600000 >, >>>>> < 800000 297600000 >, >>>>> < 998400 595200000 >, >>>>> < 1113600 595200000 >; >>>> These should map to existing opp entries. >>>> >>>> I ended up doing the interconnect driver that maps a clock to the >>>> interconnect. Then I can use it in the cpu opp tables. >>>> >>> >>> Can you point us at what it is you are proposing ? >> >> https://patchwork.kernel.org/project/linux-arm-msm/patch/20230120061417.2623751-9-dmitry.baryshkov@linaro.org/ >> > Is there no driver code too ? There are two parts, one is the 'CBF clock' driver, which just provides a clock, another part actually connects the clock and interconnect. Initially I implemented it as a part of the CBF driver (see https://patchwork.kernel.org/project/linux-arm-msm/patch/20230120061417.2623751-5-dmitry.baryshkov@linaro.org/), next revision will move the interconnect part to drivers/interconnect.
On 01/02/2023 17:12, Dmitry Baryshkov wrote: > On 01/02/2023 17:17, Bryan O'Donoghue wrote: >> On 01/02/2023 14:58, Dmitry Baryshkov wrote: >>> On 01/02/2023 16:45, Bryan O'Donoghue wrote: >>>> On 01/02/2023 13:41, Dmitry Baryshkov wrote: >>>>>> cci-cpufreq { >>>>>> target-dev = <&cci_cache>; >>>>>> cpu-to-dev-map-0 = >>>>>> < 200000 200000000 >, >>>>>> < 345600 200000000 >, >>>>>> < 400000 200000000 >, >>>>>> < 533330 297600000 >, >>>>>> < 800000 297600000 >, >>>>>> < 960000 297600000 >, >>>>>> < 1113600 297000000 >, >>>>>> < 1344000 595200000 >, >>>>>> < 1459200 595200000 >, >>>>>> < 1497600 595200000 >, >>>>>> < 1651200 595200000 >; >>>>>> cpu-to-dev-map-4 = >>>>>> < 200000 200000000 >, >>>>>> < 249600 200000000 >, >>>>>> < 499200 297600000 >, >>>>>> < 800000 297600000 >, >>>>>> < 998400 595200000 >, >>>>>> < 1113600 595200000 >; >>>>> These should map to existing opp entries. >>>>> >>>>> I ended up doing the interconnect driver that maps a clock to the >>>>> interconnect. Then I can use it in the cpu opp tables. >>>>> >>>> >>>> Can you point us at what it is you are proposing ? >>> >>> https://patchwork.kernel.org/project/linux-arm-msm/patch/20230120061417.2623751-9-dmitry.baryshkov@linaro.org/ >>> >> Is there no driver code too ? > > There are two parts, one is the 'CBF clock' driver, which just provides > a clock, another part actually connects the clock and interconnect. > Initially I implemented it as a part of the CBF driver (see > https://patchwork.kernel.org/project/linux-arm-msm/patch/20230120061417.2623751-5-dmitry.baryshkov@linaro.org/), next revision will move the interconnect part to drivers/interconnect. > Ah so just to be clear - discussing with Dmitry - CCI has its own set of fuses. We have fusebin settings for clusterX and CCI. So, I think we agree this means a separate driver for cci is warranted. --- bod
On 01/02/2023 17:17, Jun Nie wrote: > Chanwoo Choi <cwchoi00@gmail.com> 于2023年2月1日周三 19:02写道： >> >> Hi, >> >> On 23. 2. 1. 17:02, Jun Nie wrote: >>> Cache Coherent Interconnect (CCI) is used by some Qualcomm SoCs. This >>> driver is introduced so that its freqency can be adjusted. And regulator >>> associated with opp table can be also adjusted accordingly which is >>> shared with cpu cluster. >>> >>> Signed-off-by: Jun Nie <jun.nie@linaro.org> >>> --- >>> drivers/devfreq/Kconfig | 9 +++ >>> drivers/devfreq/Makefile | 1 + >>> drivers/devfreq/qcom-cci.c | 162 +++++++++++++++++++++++++++++++++++++ >>> 3 files changed, 172 insertions(+) >>> create mode 100644 drivers/devfreq/qcom-cci.c >>> >>> diff --git a/drivers/devfreq/Kconfig b/drivers/devfreq/Kconfig >>> index 9754d8b31621..6f3f1872f3fb 100644 >>> --- a/drivers/devfreq/Kconfig >>> +++ b/drivers/devfreq/Kconfig >>> @@ -130,6 +130,15 @@ config ARM_MEDIATEK_CCI_DEVFREQ >>> buck voltages and update a proper CCI frequency. Use the notification >>> to get the regulator status. >>> >>> +config ARM_QCOM_CCI_DEVFREQ >>> + tristate "QUALCOMM CCI DEVFREQ Driver" >>> + depends on ARCH_QCOM || COMPILE_TEST >>> + select DEVFREQ_GOV_PASSIVE >>> + help >>> + This adds a devfreq driver for Qualcomm Cache Coherent Interconnect which >>> + shares the same regulator with the cpu cluster. This driver can track a >>> + proper regulator state while CCI frequency is updated. >> >> Maybe, this driver use the passive governor because as this description, >> the regulator is shared with cpu cluster. But, as I commented below, >> you use the 'userspace' governor in probe. is it right? >> >>> + >>> config ARM_RK3399_DMC_DEVFREQ >>> tristate "ARM RK3399 DMC DEVFREQ Driver" >>> depends on (ARCH_ROCKCHIP && HAVE_ARM_SMCCC) || \ >>> diff --git a/drivers/devfreq/Makefile b/drivers/devfreq/Makefile >>> index bf40d04928d0..f2526d8c168d 100644 >>> --- a/drivers/devfreq/Makefile >>> +++ b/drivers/devfreq/Makefile >>> @@ -12,6 +12,7 @@ obj-$(CONFIG_ARM_EXYNOS_BUS_DEVFREQ) += exynos-bus.o >>> obj-$(CONFIG_ARM_IMX_BUS_DEVFREQ) += imx-bus.o >>> obj-$(CONFIG_ARM_IMX8M_DDRC_DEVFREQ) += imx8m-ddrc.o >>> obj-$(CONFIG_ARM_MEDIATEK_CCI_DEVFREQ) += mtk-cci-devfreq.o >>> +obj-$(CONFIG_ARM_QCOM_CCI_DEVFREQ) += qcom-cci.o >>> obj-$(CONFIG_ARM_RK3399_DMC_DEVFREQ) += rk3399_dmc.o >>> obj-$(CONFIG_ARM_SUN8I_A33_MBUS_DEVFREQ) += sun8i-a33-mbus.o >>> obj-$(CONFIG_ARM_TEGRA_DEVFREQ) += tegra30-devfreq.o >>> diff --git a/drivers/devfreq/qcom-cci.c b/drivers/devfreq/qcom-cci.c >>> new file mode 100644 >>> index 000000000000..21b5f133cddc >>> --- /dev/null >>> +++ b/drivers/devfreq/qcom-cci.c >>> @@ -0,0 +1,162 @@ >>> +// SPDX-License-Identifier: GPL-2.0 >>> +/* >>> + * Copyright 2023 Linaro Ltd. >>> + */ >>> + >>> +#include <linux/clk.h> >>> +#include <linux/devfreq.h> >>> +#include <linux/device.h> >>> +#include <linux/module.h> >>> +#include <linux/nvmem-consumer.h> >>> +#include <linux/of_device.h> >>> +#include <linux/pm_opp.h> >>> +#include <linux/platform_device.h> >>> +#include <linux/slab.h> >>> + >>> +#define SPEED_PVS(s, p) ((s << 16) | p) >> >> What it meaning of PVS? Could you add the comment for 'PVS' and 's', 'p'? >> > This is what I have from Qualcomm code. Maybe others have more info about it. > Process Voltage Scaling(PVS) Tables defines the voltage and frequency > value based on the msm-id in SMEM and speedbin blown in the efuse > combination. > >>> + >>> +struct qcom_cci { >>> + struct devfreq_dev_profile profile; >>> + struct devfreq *devfreq; >>> + struct clk *clk; >>> +}; >>> + >>> +static int qcom_cci_target(struct device *dev, >>> + unsigned long *freq, u32 flags) >> >> Actually, this line is not long. You can type it on one line as following: >> >> static int qcom_cci_target(struct device *dev, unsigned long *freq, u32 flags) >> >>> +{ >>> + struct dev_pm_opp *new_opp; >>> + int ret; >> >> As I mentioned belwo, this local variable is not needed >> if just return PTR_ERR(new_opp). >> >>> + >>> + new_opp = devfreq_recommended_opp(dev, freq, flags); >>> + if (IS_ERR(new_opp)) { >>> + ret = PTR_ERR(new_opp); >>> + dev_err(dev, "failed to get recommended opp: %d\n", ret); >>> + return ret; >> >> Better to add 'return PTR_ERR(new_opp)' without 'ret' local variable. >> >>> + } >>> + dev_pm_opp_put(new_opp); >>> + >>> + return dev_pm_opp_set_rate(dev, *freq); >>> +} >>> + >>> +static int qcom_cci_get_cur_freq(struct device *dev, unsigned long *freq) >>> +{ >>> + struct qcom_cci *priv = dev_get_drvdata(dev); >>> + >>> + *freq = clk_get_rate(priv->clk); >>> + >>> + return 0; >>> +} >>> + >>> +static int qcom_get_dev_version(struct nvmem_cell *speedbin_nvmem) >>> +{ >>> + int speed = 0, pvs = 0; >>> + u8 *speedbin; >>> + size_t len; >>> + >>> + speedbin = nvmem_cell_read(speedbin_nvmem, &len); >>> + if (IS_ERR(speedbin)) >>> + return PTR_ERR(speedbin); >>> + >>> + speed = (speedbin[0xc] >> 2) & 0x7; >>> + pvs = (speedbin[0x3] >> 5 & 0x1) | ((speedbin[0x6] >> 2 & 0x3) << 1); >> >> Actually, 0xc, 0x3, 0x7, 0x1 and so on. It is impossible to understand >> the meaning of this hex value. Plesae add the constant defintion >> for the readability. >> > Thanks for pointing it. The bit wise manipulation will be converted > into nvmem offset and > bit position in next version. So driver will read out value directly > without bit shift/mask. > Other comments will be adopted in next version. Thanks! Please consider handling everything using the nvmem API. If PVS is split into two regions it might be better to parse them separately and then combine them in the driver(*), rather than performing such parsing in the driver. (*) while doing the rework of the tsens driver I also faced a similar issue. It might be nice to have 'combined' nvmem cells.
On 01/02/2023 19:16, Bryan O'Donoghue wrote: > On 01/02/2023 17:12, Dmitry Baryshkov wrote: >> On 01/02/2023 17:17, Bryan O'Donoghue wrote: >>> On 01/02/2023 14:58, Dmitry Baryshkov wrote: >>>> On 01/02/2023 16:45, Bryan O'Donoghue wrote: >>>>> On 01/02/2023 13:41, Dmitry Baryshkov wrote: >>>>>>> cci-cpufreq { >>>>>>> target-dev = <&cci_cache>; >>>>>>> cpu-to-dev-map-0 = >>>>>>> < 200000 200000000 >, >>>>>>> < 345600 200000000 >, >>>>>>> < 400000 200000000 >, >>>>>>> < 533330 297600000 >, >>>>>>> < 800000 297600000 >, >>>>>>> < 960000 297600000 >, >>>>>>> < 1113600 297000000 >, >>>>>>> < 1344000 595200000 >, >>>>>>> < 1459200 595200000 >, >>>>>>> < 1497600 595200000 >, >>>>>>> < 1651200 595200000 >; >>>>>>> cpu-to-dev-map-4 = >>>>>>> < 200000 200000000 >, >>>>>>> < 249600 200000000 >, >>>>>>> < 499200 297600000 >, >>>>>>> < 800000 297600000 >, >>>>>>> < 998400 595200000 >, >>>>>>> < 1113600 595200000 >; >>>>>> These should map to existing opp entries. >>>>>> >>>>>> I ended up doing the interconnect driver that maps a clock to the >>>>>> interconnect. Then I can use it in the cpu opp tables. >>>>>> >>>>> >>>>> Can you point us at what it is you are proposing ? >>>> >>>> https://patchwork.kernel.org/project/linux-arm-msm/patch/20230120061417.2623751-9-dmitry.baryshkov@linaro.org/ >>>> >>> Is there no driver code too ? >> >> There are two parts, one is the 'CBF clock' driver, which just >> provides a clock, another part actually connects the clock and >> interconnect. Initially I implemented it as a part of the CBF driver >> (see >> https://patchwork.kernel.org/project/linux-arm-msm/patch/20230120061417.2623751-5-dmitry.baryshkov@linaro.org/), next revision will move the interconnect part to drivers/interconnect. >> > > Ah so just to be clear - discussing with Dmitry - CCI has its own set of > fuses. > > We have fusebin settings for clusterX and CCI. > > So, I think we agree this means a separate driver for cci is warranted. Yes.
Dmitry Baryshkov <dmitry.baryshkov@linaro.org> 于2023年2月1日周三 21:41写道： > > On Wed, 1 Feb 2023 at 13:46, Bryan O'Donoghue > <bryan.odonoghue@linaro.org> wrote: > > > > On 01/02/2023 11:32, Dmitry Baryshkov wrote: > > > On 01/02/2023 10:02, Jun Nie wrote: > > >> Cache Coherent Interconnect (CCI) is used by some Qualcomm SoCs. This > > >> driver is introduced so that its freqency can be adjusted. And regulator > > >> associated with opp table can be also adjusted accordingly which is > > >> shared with cpu cluster. > > >> > > >> Signed-off-by: Jun Nie <jun.nie@linaro.org> > > >> --- > > >> drivers/devfreq/Kconfig | 9 +++ > > >> drivers/devfreq/Makefile | 1 + > > >> drivers/devfreq/qcom-cci.c | 162 +++++++++++++++++++++++++++++++++++++ > > >> 3 files changed, 172 insertions(+) > > >> create mode 100644 drivers/devfreq/qcom-cci.c > > > > > > Could you please describe in some additional details what are you trying > > > to achieve? Should the CCI frequency be scaled manually or does it > > > follow the cluster frequency? Do clusters vote on the CCI frequency? > > > > > > I'm inclined to ask if it is possible to shift this to the cpufreq OPP > > > tables? > > > > > > > Might not be so easy to just append CCI opps to the cluster frequency opps > > > > cci_cache: qcom,cci { > > compatible = "qcom,msm8939-cci"; > > clock-names = "devfreq_clk"; > > clocks = <&apcs2>; > > governor = "cpufreq"; > > operating-points-v2 = <&cci_opp_table>; > > power-domains = <&cpr>; > > power-domain-names = "cpr"; > > nvmem-cells = <&cpr_efuse_speedbin_pvs>; > > nvmem-cell-names = "cpr_efuse_speedbin_pvs"; > > }; > > > > devfreq-cpufreq { > > cci-cpufreq { > > target-dev = <&cci_cache>; > > cpu-to-dev-map-0 = > > < 200000 200000000 >, > > < 345600 200000000 >, > > < 400000 200000000 >, > > < 533330 297600000 >, > > < 800000 297600000 >, > > < 960000 297600000 >, > > < 1113600 297000000 >, > > < 1344000 595200000 >, > > < 1459200 595200000 >, > > < 1497600 595200000 >, > > < 1651200 595200000 >; > > cpu-to-dev-map-4 = > > < 200000 200000000 >, > > < 249600 200000000 >, > > < 499200 297600000 >, > > < 800000 297600000 >, > > < 998400 595200000 >, > > < 1113600 595200000 >; > > These should map to existing opp entries. > > I ended up doing the interconnect driver that maps a clock to the > interconnect. Then I can use it in the cpu opp tables. > > > }; > > }; > > > > cci_opp_table: cci-opp-table { > > compatible = "operating-points-v2"; > > > > opp-200000000 { > > opp-hz = /bits/ 64 <200000000>; > > opp-supported-hw = <0x3f>; > > required-opps = <&cpr_opp3>; > > And these should probably map to max(cpu's CPR opp, CCI's CPR opp). The plan is opp framework to handle it when CPU opp requires both cpr power domain opp and CCI opp. While CCI opp will also requires specific cpr opp. Because CPU have a opp match table per pvs/speed, while CCI has another match table, it seems impossible to write the cpr opp requirements in the code statically. > > > }; > > > > opp-297600000 { > > opp-hz = /bits/ 64 <297600000>; > > opp-supported-hw = <0x3f>; > > required-opps = <&cpr_opp12>; > > }; > > > > opp-400000000-cpr14 { > > opp-hz = /bits/ 64 <400000000>; > > opp-supported-hw = <0x1>; > > required-opps = <&cpr_opp14>; > > }; > > > > opp-400000000-cpr15 { > > opp-hz = /bits/ 64 <400000000>; > > opp-supported-hw = <0x3e>; > > required-opps = <&cpr_opp15>; > > }; > > > > opp-595200000 { > > opp-hz = /bits/ 64 <595200000>; > > opp-supported-hw = <0x3f>; > > required-opps = <&cpr_opp17>; > > }; > > }; > > > > > > cpr_opp_table: cpr-opp-table { > > compatible = "operating-points-v2-qcom-level"; > > > > cpr_opp1: opp1 { > > opp-hz = /bits/ 64 <200000000>; > > opp-level = <1>; > > qcom,opp-fuse-level = <1>; > > }; > > cpr_opp2: opp2 { > > opp-hz = /bits/ 64 <345600000>; > > opp-level = <2>; > > qcom,opp-fuse-level = <1>; > > }; > > cpr_opp3: opp3 { > > opp-hz = /bits/ 64 <400000000>; > > opp-level = <3>; > > qcom,opp-fuse-level = <1>; > > }; > > cpr_opp4: opp4 { > > opp-hz = /bits/ 64 <422400000>; > > opp-level = <4>; > > qcom,opp-fuse-level = <2>; > > }; > > cpr_opp5: opp5 { > > opp-hz = /bits/ 64 <499200000>; > > opp-level = <5>; > > qcom,opp-fuse-level = <2>; > > }; > > cpr_opp6: opp6 { > > opp-hz = /bits/ 64 <533330000>; > > opp-level = <6>; > > qcom,opp-fuse-level = <2>; > > }; > > cpr_opp7: opp7 { > > opp-hz = /bits/ 64 <652800000>; > > opp-level = <7>; > > qcom,opp-fuse-level = <2>; > > }; > > cpr_opp8: opp8 { > > opp-hz = /bits/ 64 <729600000>; > > opp-level = <8>; > > qcom,opp-fuse-level = <2>; > > }; > > cpr_opp9: opp9 { > > opp-hz = /bits/ 64 <800000000>; > > opp-level = <9>; > > qcom,opp-fuse-level = <2>; > > }; > > cpr_opp10: opp10 { > > opp-hz = /bits/ 64 <806400000>; > > opp-level = <10>; > > qcom,opp-fuse-level = <2>; > > }; > > cpr_opp11: opp11 { > > opp-hz = /bits/ 64 <883200000>; > > opp-level = <11>; > > qcom,opp-fuse-level = <2>; > > }; > > cpr_opp12: opp12 { > > opp-hz = /bits/ 64 <960000000>; > > opp-level = <12>; > > qcom,opp-fuse-level = <2>; > > }; > > }; > > > > --- > > bod > > > > -- > With best wishes > Dmitry
diff --git a/drivers/devfreq/Kconfig b/drivers/devfreq/Kconfig index 9754d8b31621..6f3f1872f3fb 100644 --- a/drivers/devfreq/Kconfig +++ b/drivers/devfreq/Kconfig @@ -130,6 +130,15 @@ config ARM_MEDIATEK_CCI_DEVFREQ buck voltages and update a proper CCI frequency. Use the notification to get the regulator status. +config ARM_QCOM_CCI_DEVFREQ + tristate "QUALCOMM CCI DEVFREQ Driver" + depends on ARCH_QCOM || COMPILE_TEST + select DEVFREQ_GOV_PASSIVE + help + This adds a devfreq driver for Qualcomm Cache Coherent Interconnect which + shares the same regulator with the cpu cluster. This driver can track a + proper regulator state while CCI frequency is updated. + config ARM_RK3399_DMC_DEVFREQ tristate "ARM RK3399 DMC DEVFREQ Driver" depends on (ARCH_ROCKCHIP && HAVE_ARM_SMCCC) || \ diff --git a/drivers/devfreq/Makefile b/drivers/devfreq/Makefile index bf40d04928d0..f2526d8c168d 100644 --- a/drivers/devfreq/Makefile +++ b/drivers/devfreq/Makefile @@ -12,6 +12,7 @@ obj-$(CONFIG_ARM_EXYNOS_BUS_DEVFREQ) += exynos-bus.o obj-$(CONFIG_ARM_IMX_BUS_DEVFREQ) += imx-bus.o obj-$(CONFIG_ARM_IMX8M_DDRC_DEVFREQ) += imx8m-ddrc.o obj-$(CONFIG_ARM_MEDIATEK_CCI_DEVFREQ) += mtk-cci-devfreq.o +obj-$(CONFIG_ARM_QCOM_CCI_DEVFREQ) += qcom-cci.o obj-$(CONFIG_ARM_RK3399_DMC_DEVFREQ) += rk3399_dmc.o obj-$(CONFIG_ARM_SUN8I_A33_MBUS_DEVFREQ) += sun8i-a33-mbus.o obj-$(CONFIG_ARM_TEGRA_DEVFREQ) += tegra30-devfreq.o diff --git a/drivers/devfreq/qcom-cci.c b/drivers/devfreq/qcom-cci.c new file mode 100644 index 000000000000..21b5f133cddc --- /dev/null +++ b/drivers/devfreq/qcom-cci.c @@ -0,0 +1,162 @@ +// SPDX-License-Identifier: GPL-2.0 +/* + * Copyright 2023 Linaro Ltd. + */ + +#include <linux/clk.h> +#include <linux/devfreq.h> +#include <linux/device.h> +#include <linux/module.h> +#include <linux/nvmem-consumer.h> +#include <linux/of_device.h> +#include <linux/pm_opp.h> +#include <linux/platform_device.h> +#include <linux/slab.h> + +#define SPEED_PVS(s, p) ((s << 16) | p) + +struct qcom_cci { + struct devfreq_dev_profile profile; + struct devfreq *devfreq; + struct clk *clk; +}; + +static int qcom_cci_target(struct device *dev, + unsigned long *freq, u32 flags) +{ + struct dev_pm_opp *new_opp; + int ret; + + new_opp = devfreq_recommended_opp(dev, freq, flags); + if (IS_ERR(new_opp)) { + ret = PTR_ERR(new_opp); + dev_err(dev, "failed to get recommended opp: %d\n", ret); + return ret; + } + dev_pm_opp_put(new_opp); + + return dev_pm_opp_set_rate(dev, *freq); +} + +static int qcom_cci_get_cur_freq(struct device *dev, unsigned long *freq) +{ + struct qcom_cci *priv = dev_get_drvdata(dev); + + *freq = clk_get_rate(priv->clk); + + return 0; +} + +static int qcom_get_dev_version(struct nvmem_cell *speedbin_nvmem) +{ + int speed = 0, pvs = 0; + u8 *speedbin; + size_t len; + + speedbin = nvmem_cell_read(speedbin_nvmem, &len); + if (IS_ERR(speedbin)) + return PTR_ERR(speedbin); + + speed = (speedbin[0xc] >> 2) & 0x7; + pvs = (speedbin[0x3] >> 5 & 0x1) | ((speedbin[0x6] >> 2 & 0x3) << 1); + kfree(speedbin); + + /* Convert speedbin and PVS into version bit map */ + switch (SPEED_PVS(speed, pvs)) { + case SPEED_PVS(0, 0): return 0x1; + case SPEED_PVS(2, 0): return 0x2; + case SPEED_PVS(2, 2): return 0x4; + case SPEED_PVS(4, 0): return 0x8; + case SPEED_PVS(5, 0): return 0x10; + case SPEED_PVS(5, 6): return 0x20; + default: + return 0x1; + } +} + +static void qcom_cci_exit(struct device *dev) +{ + dev_pm_opp_of_remove_table(dev); +} + +static int qcom_cci_probe(struct platform_device *pdev) +{ + struct device *dev = &pdev->dev; + struct qcom_cci *priv; + const char *gov = DEVFREQ_GOV_USERSPACE; + struct device_node *np = dev->of_node; + struct nvmem_cell *speedbin_nvmem; + int ret; + u32 version; + + priv = devm_kzalloc(dev, sizeof(*priv), GFP_KERNEL); + if (!priv) + return -ENOMEM; + + priv->clk = devm_clk_get(dev, NULL); + if (IS_ERR(priv->clk)) { + ret = PTR_ERR(priv->clk); + dev_err(dev, "failed to fetch clk: %d\n", ret); + return ret; + } + platform_set_drvdata(pdev, priv); + + /* Check whether we have profiled speed version per chip */ + speedbin_nvmem = of_nvmem_cell_get(np, NULL); + if (IS_ERR(speedbin_nvmem)) + return PTR_ERR(speedbin_nvmem); + + version = qcom_get_dev_version(speedbin_nvmem); + dev_info(dev, "%s: set opp table version 0x%x\n", __func__, version); + + nvmem_cell_put(speedbin_nvmem); + ret = dev_pm_opp_set_supported_hw(dev, &version, 1); + if (ret) { + dev_err(dev, "Failed to set supported hardware\n"); + return ret; + } + + ret = dev_pm_opp_of_add_table(dev); + if (ret < 0) { + dev_err(dev, "failed to get OPP table\n"); + return ret; + } + + priv->profile.target = qcom_cci_target; + priv->profile.exit = qcom_cci_exit; + priv->profile.get_cur_freq = qcom_cci_get_cur_freq; + priv->profile.initial_freq = clk_get_rate(priv->clk); + + priv->devfreq = devm_devfreq_add_device(dev, &priv->profile, + gov, NULL); + if (IS_ERR(priv->devfreq)) { + ret = PTR_ERR(priv->devfreq); + dev_err(dev, "failed to add devfreq device: %d\n", ret); + goto err; + } + + return 0; + +err: + dev_pm_opp_of_remove_table(dev); + return ret; +} + +static const struct of_device_id qcom_cci_of_match[] = { + { .compatible = "qcom,msm8939-cci"}, + { /* sentinel */ }, +}; +MODULE_DEVICE_TABLE(of, qcom_cci_of_match); + +static struct platform_driver qcom_cci_platdrv = { + .probe = qcom_cci_probe, + .driver = { + .name = "qcom-cci-devfreq", + .of_match_table = qcom_cci_of_match, + }, +}; +module_platform_driver(qcom_cci_platdrv); + +MODULE_DESCRIPTION("QCOM cci frequency scaling driver"); +MODULE_AUTHOR("Jun Nie <jun.nie@linaro.org>"); +MODULE_LICENSE("GPL");
Cache Coherent Interconnect (CCI) is used by some Qualcomm SoCs. This driver is introduced so that its freqency can be adjusted. And regulator associated with opp table can be also adjusted accordingly which is shared with cpu cluster. Signed-off-by: Jun Nie <jun.nie@linaro.org> --- drivers/devfreq/Kconfig | 9 +++ drivers/devfreq/Makefile | 1 + drivers/devfreq/qcom-cci.c | 162 +++++++++++++++++++++++++++++++++++++ 3 files changed, 172 insertions(+) create mode 100644 drivers/devfreq/qcom-cci.c