| Message ID | 20230328215328.101990-1-srinivas.pandruvada@linux.intel.com (mailing list archive) |
|---|---|
| State | Changes Requested, archived |
| Headers | show |
| Series | platform/x86/intel-uncore-freq: Uncore frequency control via TPMI | expand |
Hi Srinivas, On 3/28/23 23:53, Srinivas Pandruvada wrote: > Implement support of uncore frequency control via TPMI (Topology Aware > Register and PM Capsule Interface). This driver provides the similar > functionality as the current uncore frequency driver using MSRs. > > The hardware interface to read/write is basically substitution of MSR > 0x620 and 0x621. There are specific MMIO offset and bits to get/set > minimum and maximum uncore ratio, similar to MSRs. > > The scope of the uncore MSRs is package/die. But new generation of CPUs > have more granular control at a cluster level. Each package/die can have > multiple power domains, which further can have multiple clusters. The > TPMI interface allows control at cluster level. > > The primary use case for uncore sysfs is to set maximum and minimum > uncore frequency to reduce power consumption or latency. The current > uncore sysfs control is per package/die. This is enough for the majority > of users as workload will move to different power domains as it moves > between different CPUs. > > The current uncore sysfs provides controls at package/die level. When > user sets maximum/minimum limits, the driver sets the same limits to > each cluster. > > Here number of power domains = number of resources in this aux device. > There are offsets and bits to discover number of clusters and offset for > each cluster level controls. > > The TPMI documentation can be downloaded from: > https://github.com/intel/tpmi_power_management > > Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com> Thanks 3 small remarks below (inline), if you can address those then v2 should be ready for merging. > --- > The fine grain control at cluster level patches are here: > https://github.com/spandruvada/linux-kernel/commit/0d66ea4ff76ea19127f2d207a7e17bb86846ca32 > https://github.com/spandruvada/linux-kernel/commit/cb5c2349a58318c04955821d6528cc8015541e65 > But not submitting to ease in review process as I posted too many > patches this cycle. As mentioned before, it helps if you can get fellow Intel folks to review. > .../x86/intel/uncore-frequency/Kconfig | 4 + > .../x86/intel/uncore-frequency/Makefile | 2 + > .../uncore-frequency/uncore-frequency-tpmi.c | 346 ++++++++++++++++++ > 3 files changed, 352 insertions(+) > create mode 100644 drivers/platform/x86/intel/uncore-frequency/uncore-frequency-tpmi.c > > diff --git a/drivers/platform/x86/intel/uncore-frequency/Kconfig b/drivers/platform/x86/intel/uncore-frequency/Kconfig > index 21b209124916..a56d55056927 100644 > --- a/drivers/platform/x86/intel/uncore-frequency/Kconfig > +++ b/drivers/platform/x86/intel/uncore-frequency/Kconfig > @@ -6,9 +6,13 @@ > menu "Intel Uncore Frequency Control" > depends on X86_64 || COMPILE_TEST > > +config INTEL_UNCORE_FREQ_CONTROL_TPMI > + tristate > + > config INTEL_UNCORE_FREQ_CONTROL > tristate "Intel Uncore frequency control driver" > depends on X86_64 > + select INTEL_UNCORE_FREQ_CONTROL_TPMI if INTEL_TPMI > help > This driver allows control of Uncore frequency limits on > supported server platforms. > diff --git a/drivers/platform/x86/intel/uncore-frequency/Makefile b/drivers/platform/x86/intel/uncore-frequency/Makefile > index e0f7968e8285..08ff57492b28 100644 > --- a/drivers/platform/x86/intel/uncore-frequency/Makefile > +++ b/drivers/platform/x86/intel/uncore-frequency/Makefile > @@ -7,3 +7,5 @@ obj-$(CONFIG_INTEL_UNCORE_FREQ_CONTROL) += intel-uncore-frequency.o > intel-uncore-frequency-y := uncore-frequency.o > obj-$(CONFIG_INTEL_UNCORE_FREQ_CONTROL) += intel-uncore-frequency-common.o > intel-uncore-frequency-common-y := uncore-frequency-common.o > +obj-$(CONFIG_INTEL_UNCORE_FREQ_CONTROL_TPMI) += intel-uncore-frequency-tpmi.o > +intel-uncore-frequency-tpmi-y := uncore-frequency-tpmi.o > diff --git a/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-tpmi.c b/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-tpmi.c > new file mode 100644 > index 000000000000..4150bb877ab5 > --- /dev/null > +++ b/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-tpmi.c > @@ -0,0 +1,346 @@ > +// SPDX-License-Identifier: GPL-2.0-only > +/* > + * intel-ufs-tpmi: Intel x86 platform uncore frequency scaling > + * > + * Copyright (c) 2023, Intel Corporation. > + * All Rights Reserved. > + * > + * The hardware interface to read/write is basically substitution of > + * MSR 0x620 and 0x621. > + * There are specific MMIO offset and bits to get/set minimum and > + * maximum uncore ratio, similar to MSRs. > + * The scope of the uncore MSRs was package scope. But TPMI allows > + * new gen CPUs to have multiple uncore controls at uncore-cluster > + * level. Each package can have multiple power domains which further > + * can have multiple clusters. > + * Here number of power domains = number of resources in this aux > + * device. There are offsets and bits to discover number of clusters > + * and offset for each cluster level controls. > + * > + */ > + > +#include <linux/auxiliary_bus.h> > +#include <linux/bitfield.h> > +#include <linux/bits.h> > +#include <linux/io.h> > +#include <linux/module.h> > +#include <linux/intel_tpmi.h> > + > +#include "uncore-frequency-common.h" > + > +#define UNCORE_HEADER_VERSION 1 > +#define UNCORE_HEADER_INDEX 0 > +#define UNCORE_FABRIC_CLUSTER_OFFSET 8 > + > +/* status + control + adv_ctl1 + adv_ctl2 */ > +#define UNCORE_FABRIC_CLUSTER_SIZE (4 * 8) > + > +#define UNCORE_STATUS_INDEX 0 > +#define UNCORE_CONTROL_INDEX 8 > + > +#define UNCORE_FREQ_KHZ_MULTIPLIER 100000 > + > +struct tpmi_uncore_struct; > + > +/* Information for each cluster */ > +struct tpmi_uncore_cluster_info { > + void __iomem *cluster_base; > + struct uncore_data uncore_data; > + struct tpmi_uncore_struct *uncore_root; > +}; > + > +/* Information for each power domain */ > +struct tpmi_uncore_power_domain_info { > + void __iomem *uncore_base; Maybe make this an: u8 __iomem *uncore_base; To avoid all the casts below ? Although I wonder if the casts are necessary at all, AFAIK void * arithmics are the same as u8 * arithmics, so things should work without the casts regardless ? Still just turning this into a u8 * is probably better. > + int ufs_header_ver; > + int cluster_count; > + struct tpmi_uncore_cluster_info *cluster_infos; > +}; > + > +/* Information for all power domains in a package */ > +struct tpmi_uncore_struct { > + int power_domain_count; > + struct tpmi_uncore_power_domain_info *pd_info; > + struct tpmi_uncore_cluster_info root_cluster; > +}; > + > +#define UNCORE_GENMASK_MIN_RATIO GENMASK_ULL(21, 15) > +#define UNCORE_GENMASK_MAX_RATIO GENMASK_ULL(14, 8) > + > +/* Helper function to read MMIO offset for max/min control frequency */ > +static void read_control_freq(struct tpmi_uncore_cluster_info *cluster_info, > + unsigned int *min, unsigned int *max) > +{ > + u64 control; > + > + control = readq((u8 __iomem *)cluster_info->cluster_base + UNCORE_CONTROL_INDEX); > + *max = FIELD_GET(UNCORE_GENMASK_MAX_RATIO, control) * UNCORE_FREQ_KHZ_MULTIPLIER; > + *min = FIELD_GET(UNCORE_GENMASK_MIN_RATIO, control) * UNCORE_FREQ_KHZ_MULTIPLIER; > +} > + > +#define UNCORE_MAX_RATIO 0x7F > + > +/* Callback for sysfs read for max/min frequencies. Called under mutex locks */ > +static int uncore_read_control_freq(struct uncore_data *data, unsigned int *min, > + unsigned int *max) > +{ > + struct tpmi_uncore_cluster_info *cluster_info; > + struct tpmi_uncore_struct *uncore_root; > + int i, _min = 0, _max = 0; > + > + cluster_info = container_of(data, struct tpmi_uncore_cluster_info, uncore_data); > + uncore_root = cluster_info->uncore_root; > + > + *min = UNCORE_MAX_RATIO * UNCORE_FREQ_KHZ_MULTIPLIER; > + *max = 0; > + > + /* > + * Get the max/min by looking at each cluster. Get the lowest > + * min and highest max. > + */ > + for (i = 0; i < uncore_root->power_domain_count; ++i) { > + int j; > + > + for (j = 0; j < uncore_root->pd_info[i].cluster_count; ++j) { > + read_control_freq(&uncore_root->pd_info[i].cluster_infos[j], > + &_min, &_max); > + if (*min > _min) > + *min = _min; > + if (*max < _max) > + *max = _max; > + } > + } > + > + return 0; > +} > + > +/* Helper function to write MMIO offset for max/min control frequency */ > +static void write_control_freq(struct tpmi_uncore_cluster_info *cluster_info, unsigned int input, > + unsigned int min_max) > +{ > + u64 control; > + > + control = readq((u8 __iomem *)cluster_info->cluster_base + UNCORE_CONTROL_INDEX); > + > + if (min_max) { > + control &= ~UNCORE_GENMASK_MAX_RATIO; > + control |= FIELD_PREP(UNCORE_GENMASK_MAX_RATIO, input); > + } else { > + control &= ~UNCORE_GENMASK_MIN_RATIO; > + control |= FIELD_PREP(UNCORE_GENMASK_MIN_RATIO, input); > + } > + > + writeq(control, ((u8 __iomem *)cluster_info->cluster_base + UNCORE_CONTROL_INDEX)); > +} > + > +/* Callback for sysfs write for max/min frequencies. Called under mutex locks */ > +static int uncore_write_control_freq(struct uncore_data *data, unsigned int input, > + unsigned int min_max) > +{ > + struct tpmi_uncore_cluster_info *cluster_info; > + struct tpmi_uncore_struct *uncore_root; > + int i; > + > + input /= UNCORE_FREQ_KHZ_MULTIPLIER; > + if (!input || input > UNCORE_MAX_RATIO) > + return -EINVAL; > + > + cluster_info = container_of(data, struct tpmi_uncore_cluster_info, uncore_data); > + uncore_root = cluster_info->uncore_root; > + > + /* Update each cluster in a package */ > + for (i = 0; i < uncore_root->power_domain_count; ++i) { > + int j; > + > + for (j = 0; j < uncore_root->pd_info[i].cluster_count; ++j) > + write_control_freq(&uncore_root->pd_info[i].cluster_infos[j], > + input, min_max); > + } > + > + return 0; > +} > + > +/* Callback for sysfs read for the current uncore frequency. Called under mutex locks */ > +static int uncore_read_freq(struct uncore_data *data, unsigned int *freq) > +{ > + return -ENODATA; > +} > + > +#define UNCORE_GENMASK_VERSION GENMASK_ULL(7, 0) > +#define UNCORE_LOCAL_FABRIC_CLUSTER_ID_MASK GENMASK_ULL(15, 8) > +#define UNCORE_CLUSTER_OFF_MASK GENMASK_ULL(7, 0) > +#define UNCORE_MAX_CLUSTER_PER_DOMAIN 8 > + > +static int tpmi_uncore_init(struct auxiliary_device *auxdev) > +{ > + struct intel_tpmi_plat_info *plat_info; > + struct tpmi_uncore_struct *tpmi_uncore; > + int ret, i, pkg = 0; > + int num_resources; > + > + /* Get number of power domains, which is equal to number of resources */ > + num_resources = tpmi_get_resource_count(auxdev); > + if (!num_resources) > + return -EINVAL; > + > + /* Register callbacks to uncore core */ > + ret = uncore_freq_common_init(uncore_read_control_freq, uncore_write_control_freq, > + uncore_read_freq); > + if (ret) > + return ret; > + > + /* Allocate uncore instance per package */ > + tpmi_uncore = devm_kzalloc(&auxdev->dev, sizeof(*tpmi_uncore), GFP_KERNEL); > + if (!tpmi_uncore) { > + ret = -ENOMEM; > + goto err_rem_common; > + } > + > + /* Allocate memory for all power domains in a package */ > + tpmi_uncore->pd_info = devm_kcalloc(&auxdev->dev, num_resources, > + sizeof(*tpmi_uncore->pd_info), > + GFP_KERNEL); > + if (!tpmi_uncore->pd_info) { > + ret = -ENOMEM; > + goto err_rem_common; > + } > + > + tpmi_uncore->power_domain_count = num_resources; > + > + /* Get the package ID from the TPMI core */ > + plat_info = tpmi_get_platform_data(auxdev); > + if (plat_info) > + pkg = plat_info->package_id; > + > + for (i = 0; i < num_resources; ++i) { > + struct tpmi_uncore_power_domain_info *pd_info; > + struct resource *res; > + u64 cluster_offset; > + u8 cluster_mask; > + int mask, j; > + u64 header; > + > + res = tpmi_get_resource_at_index(auxdev, i); > + if (!res) > + continue; > + > + pd_info = &tpmi_uncore->pd_info[i]; > + > + pd_info->uncore_base = devm_ioremap_resource(&auxdev->dev, res); > + if (IS_ERR(pd_info->uncore_base)) { > + ret = PTR_ERR(pd_info->uncore_base); > + pd_info->uncore_base = NULL; pd_info is part of tpmi_uncore and on error the: auxiliary_set_drvdata(auxdev, tpmi_uncore); call is skipped, so pd_info will never be reachable (and should get freed by devm). So AFAICT, there is no need to NULL uncore_base ? > + goto err_rem_common; > + } > + > + /* Check for version and skip this resource if there is mismatch */ > + header = readq(pd_info->uncore_base); > + pd_info->ufs_header_ver = header & UNCORE_GENMASK_VERSION; > + if (pd_info->ufs_header_ver != UNCORE_HEADER_VERSION) { > + dev_info(&auxdev->dev, "Uncore: Unsupported version:%d\n", > + pd_info->ufs_header_ver); > + continue; > + } > + > + /* Get Cluster ID Mask */ > + cluster_mask = FIELD_GET(UNCORE_LOCAL_FABRIC_CLUSTER_ID_MASK, header); > + if (!cluster_mask) { > + dev_info(&auxdev->dev, "Uncore: Invalid cluster mask:%x\n", cluster_mask); > + continue; > + } > + > + /* Find out number of clusters in this resource */ > + mask = 0x01; > + for (j = 0; j < UNCORE_MAX_CLUSTER_PER_DOMAIN; ++j) { > + if (cluster_mask & mask) > + pd_info->cluster_count++; > + mask <<= 1; > + } > + > + pd_info->cluster_infos = devm_kcalloc(&auxdev->dev, pd_info->cluster_count, > + sizeof(struct tpmi_uncore_cluster_info), > + GFP_KERNEL); This devm_kcalloc() call is missing error checking. > + > + /* > + * Each byte in the register point to status and control > + * registers belonging to cluster id 0-8. > + */ > + cluster_offset = readq((u8 __iomem *)pd_info->uncore_base + > + UNCORE_FABRIC_CLUSTER_OFFSET); > + > + for (j = 0; j < pd_info->cluster_count; ++j) { > + struct tpmi_uncore_cluster_info *cluster_info; > + > + /* Get the offset for this cluster */ > + mask = (cluster_offset & UNCORE_CLUSTER_OFF_MASK); > + /* Offset in QWORD, so change to bytes */ > + mask <<= 3; > + > + cluster_info = &pd_info->cluster_infos[j]; > + > + cluster_info->cluster_base = (u8 __iomem *)pd_info->uncore_base + mask; > + > + cluster_info->uncore_data.package_id = pkg; > + /* There are no dies like Cascade Lake */ > + cluster_info->uncore_data.die_id = 0; > + > + /* Point to next cluster offset */ > + cluster_offset >>= UNCORE_MAX_CLUSTER_PER_DOMAIN; > + } > + } > + > + auxiliary_set_drvdata(auxdev, tpmi_uncore); > + > + tpmi_uncore->root_cluster.uncore_root = tpmi_uncore; > + tpmi_uncore->root_cluster.uncore_data.package_id = pkg; > + ret = uncore_freq_add_entry(&tpmi_uncore->root_cluster.uncore_data, 0); > + if (ret) > + goto err_rem_common; > + > + return 0; > + > +err_rem_common: > + uncore_freq_common_exit(); > + > + return ret; > +} > + > +static int tpmi_uncore_remove(struct auxiliary_device *auxdev) > +{ > + struct tpmi_uncore_struct *tpmi_uncore = auxiliary_get_drvdata(auxdev); > + > + uncore_freq_remove_die_entry(&tpmi_uncore->root_cluster.uncore_data); > + > + uncore_freq_common_exit(); > + > + return 0; > +} > + > +static int intel_uncore_probe(struct auxiliary_device *auxdev, const struct auxiliary_device_id *id) > +{ > + return tpmi_uncore_init(auxdev); > +} > + > +static void intel_uncore_remove(struct auxiliary_device *auxdev) > +{ > + tpmi_uncore_remove(auxdev); > +} > + > +static const struct auxiliary_device_id intel_uncore_id_table[] = { > + { .name = "intel_vsec.tpmi-uncore" }, > + {} > +}; > +MODULE_DEVICE_TABLE(auxiliary, intel_uncore_id_table); > + > +static struct auxiliary_driver intel_uncore_aux_driver = { > + .id_table = intel_uncore_id_table, > + .remove = intel_uncore_remove, > + .probe = intel_uncore_probe, > +}; > + > +module_auxiliary_driver(intel_uncore_aux_driver); > + > +MODULE_IMPORT_NS(INTEL_TPMI); > +MODULE_IMPORT_NS(INTEL_UNCORE_FREQUENCY); > +MODULE_DESCRIPTION("Intel TPMI UFS Driver"); > +MODULE_LICENSE("GPL"); Regards, Hans
Hi Hans, > > [...] > > The TPMI documentation can be downloaded from: > > https://github.com/intel/tpmi_power_management > > > > Signed-off-by: Srinivas Pandruvada > > <srinivas.pandruvada@linux.intel.com> > > Thanks 3 small remarks below (inline), if you can address those then > v2 should be ready for merging. Sure. Will send an update. > > > --- > > The fine grain control at cluster level patches are here: > > https://github.com/spandruvada/linux-kernel/commit/0d66ea4ff76ea19127f2d207a7e17bb86846ca32 > > https://github.com/spandruvada/linux-kernel/commit/cb5c2349a58318c04955821d6528cc8015541e65 > > But not submitting to ease in review process as I posted too many > > patches this cycle. > > As mentioned before, it helps if you can get fellow Intel folks to > review. I will add to the next version with reviewed-by and tested-by. > > > .../x86/intel/uncore-frequency/Kconfig | 4 + > > .../x86/intel/uncore-frequency/Makefile | 2 + > > .../uncore-frequency/uncore-frequency-tpmi.c | 346 > > ++++++++++++++++++ > > [...] > > +/* Information for each power domain */ > > +struct tpmi_uncore_power_domain_info { > > + void __iomem *uncore_base; > > Maybe make this an: > > u8 __iomem *uncore_base; > > To avoid all the casts below ? > > Although I wonder if the casts are necessary at all, AFAIK void * > arithmics are the same as u8 * arithmics, so things should work > without the casts regardless ? > You are right. I already don't cast in SST, I missed this one. > Still just turning this into a u8 * is probably better. I will change to u8* > > [...] > > + pd_info->uncore_base = > > devm_ioremap_resource(&auxdev->dev, res); > > + if (IS_ERR(pd_info->uncore_base)) { > > + ret = PTR_ERR(pd_info->uncore_base); > > + pd_info->uncore_base = NULL; > > pd_info is part of tpmi_uncore and on error the: > > auxiliary_set_drvdata(auxdev, tpmi_uncore); > > call is skipped, so pd_info will never be reachable (and should get > freed by devm). > So AFAICT, there is no need to NULL uncore_base ? Correct. Let me think of case when firmware messed up one power domain base address (very unlikely as this is goes through several validation levels), may be continue to check next power domain instead of return. > > > + goto err_rem_common; > > + } > > + > > + /* Check for version and skip this resource if > > there is mismatch */ > > + header = readq(pd_info->uncore_base); > > + pd_info->ufs_header_ver = header & > > UNCORE_GENMASK_VERSION; > > + if (pd_info->ufs_header_ver != > > UNCORE_HEADER_VERSION) { > > + dev_info(&auxdev->dev, "Uncore: Unsupported > > version:%d\n", > > + pd_info->ufs_header_ver); > > + continue; > > + } > > + > > + /* Get Cluster ID Mask */ > > + cluster_mask = > > FIELD_GET(UNCORE_LOCAL_FABRIC_CLUSTER_ID_MASK, header); > > + if (!cluster_mask) { > > + dev_info(&auxdev->dev, "Uncore: Invalid > > cluster mask:%x\n", cluster_mask); > > + continue; > > + } > > + > > + /* Find out number of clusters in this resource */ > > + mask = 0x01; > > + for (j = 0; j < UNCORE_MAX_CLUSTER_PER_DOMAIN; ++j) > > { > > + if (cluster_mask & mask) > > + pd_info->cluster_count++; > > + mask <<= 1; > > + } > > + > > + pd_info->cluster_infos = devm_kcalloc(&auxdev->dev, > > pd_info->cluster_count, > > + sizeof(struct > > tpmi_uncore_cluster_info), > > + GFP_KERNEL); > > This devm_kcalloc() call is missing error checking. Thanks for pointing. Thanks, Srinivas > > > + > > + /* > > + * Each byte in the register point to status and > > control > > + * registers belonging to cluster id 0-8. > > + */ > > + cluster_offset = readq((u8 __iomem *)pd_info- > > >uncore_base + > > + UNCORE_FABRIC_CLUSTER_OFFSE > > T); > > + > > + for (j = 0; j < pd_info->cluster_count; ++j) { > > + struct tpmi_uncore_cluster_info > > *cluster_info; > > + > > + /* Get the offset for this cluster */ > > + mask = (cluster_offset & > > UNCORE_CLUSTER_OFF_MASK); > > + /* Offset in QWORD, so change to bytes */ > > + mask <<= 3; > > + > > + cluster_info = &pd_info->cluster_infos[j]; > > + > > + cluster_info->cluster_base = (u8 __iomem > > *)pd_info->uncore_base + mask; > > + > > + cluster_info->uncore_data.package_id = pkg; > > + /* There are no dies like Cascade Lake */ > > + cluster_info->uncore_data.die_id = 0; > > + > > + /* Point to next cluster offset */ > > + cluster_offset >>= > > UNCORE_MAX_CLUSTER_PER_DOMAIN; > > + } > > + } > > + > > + auxiliary_set_drvdata(auxdev, tpmi_uncore); > > + > > + tpmi_uncore->root_cluster.uncore_root = tpmi_uncore; > > + tpmi_uncore->root_cluster.uncore_data.package_id = pkg; > > + ret = uncore_freq_add_entry(&tpmi_uncore- > > >root_cluster.uncore_data, 0); > > + if (ret) > > + goto err_rem_common; > > + > > + return 0; > > + > > +err_rem_common: > > + uncore_freq_common_exit(); > > + > > + return ret; > > +} > > + > > +static int tpmi_uncore_remove(struct auxiliary_device *auxdev) > > +{ > > + struct tpmi_uncore_struct *tpmi_uncore = > > auxiliary_get_drvdata(auxdev); > > + > > + uncore_freq_remove_die_entry(&tpmi_uncore- > > >root_cluster.uncore_data); > > + > > + uncore_freq_common_exit(); > > + > > + return 0; > > +} > > + > > +static int intel_uncore_probe(struct auxiliary_device *auxdev, > > const struct auxiliary_device_id *id) > > +{ > > + return tpmi_uncore_init(auxdev); > > +} > > + > > +static void intel_uncore_remove(struct auxiliary_device *auxdev) > > +{ > > + tpmi_uncore_remove(auxdev); > > +} > > + > > +static const struct auxiliary_device_id intel_uncore_id_table[] = > > { > > + { .name = "intel_vsec.tpmi-uncore" }, > > + {} > > +}; > > +MODULE_DEVICE_TABLE(auxiliary, intel_uncore_id_table); > > + > > +static struct auxiliary_driver intel_uncore_aux_driver = { > > + .id_table = intel_uncore_id_table, > > + .remove = intel_uncore_remove, > > + .probe = intel_uncore_probe, > > +}; > > + > > +module_auxiliary_driver(intel_uncore_aux_driver); > > + > > +MODULE_IMPORT_NS(INTEL_TPMI); > > +MODULE_IMPORT_NS(INTEL_UNCORE_FREQUENCY); > > +MODULE_DESCRIPTION("Intel TPMI UFS Driver"); > > +MODULE_LICENSE("GPL"); > > > Regards, > > Hans > >
diff --git a/drivers/platform/x86/intel/uncore-frequency/Kconfig b/drivers/platform/x86/intel/uncore-frequency/Kconfig index 21b209124916..a56d55056927 100644 --- a/drivers/platform/x86/intel/uncore-frequency/Kconfig +++ b/drivers/platform/x86/intel/uncore-frequency/Kconfig @@ -6,9 +6,13 @@ menu "Intel Uncore Frequency Control" depends on X86_64 || COMPILE_TEST +config INTEL_UNCORE_FREQ_CONTROL_TPMI + tristate + config INTEL_UNCORE_FREQ_CONTROL tristate "Intel Uncore frequency control driver" depends on X86_64 + select INTEL_UNCORE_FREQ_CONTROL_TPMI if INTEL_TPMI help This driver allows control of Uncore frequency limits on supported server platforms. diff --git a/drivers/platform/x86/intel/uncore-frequency/Makefile b/drivers/platform/x86/intel/uncore-frequency/Makefile index e0f7968e8285..08ff57492b28 100644 --- a/drivers/platform/x86/intel/uncore-frequency/Makefile +++ b/drivers/platform/x86/intel/uncore-frequency/Makefile @@ -7,3 +7,5 @@ obj-$(CONFIG_INTEL_UNCORE_FREQ_CONTROL) += intel-uncore-frequency.o intel-uncore-frequency-y := uncore-frequency.o obj-$(CONFIG_INTEL_UNCORE_FREQ_CONTROL) += intel-uncore-frequency-common.o intel-uncore-frequency-common-y := uncore-frequency-common.o +obj-$(CONFIG_INTEL_UNCORE_FREQ_CONTROL_TPMI) += intel-uncore-frequency-tpmi.o +intel-uncore-frequency-tpmi-y := uncore-frequency-tpmi.o diff --git a/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-tpmi.c b/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-tpmi.c new file mode 100644 index 000000000000..4150bb877ab5 --- /dev/null +++ b/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-tpmi.c @@ -0,0 +1,346 @@ +// SPDX-License-Identifier: GPL-2.0-only +/* + * intel-ufs-tpmi: Intel x86 platform uncore frequency scaling + * + * Copyright (c) 2023, Intel Corporation. + * All Rights Reserved. + * + * The hardware interface to read/write is basically substitution of + * MSR 0x620 and 0x621. + * There are specific MMIO offset and bits to get/set minimum and + * maximum uncore ratio, similar to MSRs. + * The scope of the uncore MSRs was package scope. But TPMI allows + * new gen CPUs to have multiple uncore controls at uncore-cluster + * level. Each package can have multiple power domains which further + * can have multiple clusters. + * Here number of power domains = number of resources in this aux + * device. There are offsets and bits to discover number of clusters + * and offset for each cluster level controls. + * + */ + +#include <linux/auxiliary_bus.h> +#include <linux/bitfield.h> +#include <linux/bits.h> +#include <linux/io.h> +#include <linux/module.h> +#include <linux/intel_tpmi.h> + +#include "uncore-frequency-common.h" + +#define UNCORE_HEADER_VERSION 1 +#define UNCORE_HEADER_INDEX 0 +#define UNCORE_FABRIC_CLUSTER_OFFSET 8 + +/* status + control + adv_ctl1 + adv_ctl2 */ +#define UNCORE_FABRIC_CLUSTER_SIZE (4 * 8) + +#define UNCORE_STATUS_INDEX 0 +#define UNCORE_CONTROL_INDEX 8 + +#define UNCORE_FREQ_KHZ_MULTIPLIER 100000 + +struct tpmi_uncore_struct; + +/* Information for each cluster */ +struct tpmi_uncore_cluster_info { + void __iomem *cluster_base; + struct uncore_data uncore_data; + struct tpmi_uncore_struct *uncore_root; +}; + +/* Information for each power domain */ +struct tpmi_uncore_power_domain_info { + void __iomem *uncore_base; + int ufs_header_ver; + int cluster_count; + struct tpmi_uncore_cluster_info *cluster_infos; +}; + +/* Information for all power domains in a package */ +struct tpmi_uncore_struct { + int power_domain_count; + struct tpmi_uncore_power_domain_info *pd_info; + struct tpmi_uncore_cluster_info root_cluster; +}; + +#define UNCORE_GENMASK_MIN_RATIO GENMASK_ULL(21, 15) +#define UNCORE_GENMASK_MAX_RATIO GENMASK_ULL(14, 8) + +/* Helper function to read MMIO offset for max/min control frequency */ +static void read_control_freq(struct tpmi_uncore_cluster_info *cluster_info, + unsigned int *min, unsigned int *max) +{ + u64 control; + + control = readq((u8 __iomem *)cluster_info->cluster_base + UNCORE_CONTROL_INDEX); + *max = FIELD_GET(UNCORE_GENMASK_MAX_RATIO, control) * UNCORE_FREQ_KHZ_MULTIPLIER; + *min = FIELD_GET(UNCORE_GENMASK_MIN_RATIO, control) * UNCORE_FREQ_KHZ_MULTIPLIER; +} + +#define UNCORE_MAX_RATIO 0x7F + +/* Callback for sysfs read for max/min frequencies. Called under mutex locks */ +static int uncore_read_control_freq(struct uncore_data *data, unsigned int *min, + unsigned int *max) +{ + struct tpmi_uncore_cluster_info *cluster_info; + struct tpmi_uncore_struct *uncore_root; + int i, _min = 0, _max = 0; + + cluster_info = container_of(data, struct tpmi_uncore_cluster_info, uncore_data); + uncore_root = cluster_info->uncore_root; + + *min = UNCORE_MAX_RATIO * UNCORE_FREQ_KHZ_MULTIPLIER; + *max = 0; + + /* + * Get the max/min by looking at each cluster. Get the lowest + * min and highest max. + */ + for (i = 0; i < uncore_root->power_domain_count; ++i) { + int j; + + for (j = 0; j < uncore_root->pd_info[i].cluster_count; ++j) { + read_control_freq(&uncore_root->pd_info[i].cluster_infos[j], + &_min, &_max); + if (*min > _min) + *min = _min; + if (*max < _max) + *max = _max; + } + } + + return 0; +} + +/* Helper function to write MMIO offset for max/min control frequency */ +static void write_control_freq(struct tpmi_uncore_cluster_info *cluster_info, unsigned int input, + unsigned int min_max) +{ + u64 control; + + control = readq((u8 __iomem *)cluster_info->cluster_base + UNCORE_CONTROL_INDEX); + + if (min_max) { + control &= ~UNCORE_GENMASK_MAX_RATIO; + control |= FIELD_PREP(UNCORE_GENMASK_MAX_RATIO, input); + } else { + control &= ~UNCORE_GENMASK_MIN_RATIO; + control |= FIELD_PREP(UNCORE_GENMASK_MIN_RATIO, input); + } + + writeq(control, ((u8 __iomem *)cluster_info->cluster_base + UNCORE_CONTROL_INDEX)); +} + +/* Callback for sysfs write for max/min frequencies. Called under mutex locks */ +static int uncore_write_control_freq(struct uncore_data *data, unsigned int input, + unsigned int min_max) +{ + struct tpmi_uncore_cluster_info *cluster_info; + struct tpmi_uncore_struct *uncore_root; + int i; + + input /= UNCORE_FREQ_KHZ_MULTIPLIER; + if (!input || input > UNCORE_MAX_RATIO) + return -EINVAL; + + cluster_info = container_of(data, struct tpmi_uncore_cluster_info, uncore_data); + uncore_root = cluster_info->uncore_root; + + /* Update each cluster in a package */ + for (i = 0; i < uncore_root->power_domain_count; ++i) { + int j; + + for (j = 0; j < uncore_root->pd_info[i].cluster_count; ++j) + write_control_freq(&uncore_root->pd_info[i].cluster_infos[j], + input, min_max); + } + + return 0; +} + +/* Callback for sysfs read for the current uncore frequency. Called under mutex locks */ +static int uncore_read_freq(struct uncore_data *data, unsigned int *freq) +{ + return -ENODATA; +} + +#define UNCORE_GENMASK_VERSION GENMASK_ULL(7, 0) +#define UNCORE_LOCAL_FABRIC_CLUSTER_ID_MASK GENMASK_ULL(15, 8) +#define UNCORE_CLUSTER_OFF_MASK GENMASK_ULL(7, 0) +#define UNCORE_MAX_CLUSTER_PER_DOMAIN 8 + +static int tpmi_uncore_init(struct auxiliary_device *auxdev) +{ + struct intel_tpmi_plat_info *plat_info; + struct tpmi_uncore_struct *tpmi_uncore; + int ret, i, pkg = 0; + int num_resources; + + /* Get number of power domains, which is equal to number of resources */ + num_resources = tpmi_get_resource_count(auxdev); + if (!num_resources) + return -EINVAL; + + /* Register callbacks to uncore core */ + ret = uncore_freq_common_init(uncore_read_control_freq, uncore_write_control_freq, + uncore_read_freq); + if (ret) + return ret; + + /* Allocate uncore instance per package */ + tpmi_uncore = devm_kzalloc(&auxdev->dev, sizeof(*tpmi_uncore), GFP_KERNEL); + if (!tpmi_uncore) { + ret = -ENOMEM; + goto err_rem_common; + } + + /* Allocate memory for all power domains in a package */ + tpmi_uncore->pd_info = devm_kcalloc(&auxdev->dev, num_resources, + sizeof(*tpmi_uncore->pd_info), + GFP_KERNEL); + if (!tpmi_uncore->pd_info) { + ret = -ENOMEM; + goto err_rem_common; + } + + tpmi_uncore->power_domain_count = num_resources; + + /* Get the package ID from the TPMI core */ + plat_info = tpmi_get_platform_data(auxdev); + if (plat_info) + pkg = plat_info->package_id; + + for (i = 0; i < num_resources; ++i) { + struct tpmi_uncore_power_domain_info *pd_info; + struct resource *res; + u64 cluster_offset; + u8 cluster_mask; + int mask, j; + u64 header; + + res = tpmi_get_resource_at_index(auxdev, i); + if (!res) + continue; + + pd_info = &tpmi_uncore->pd_info[i]; + + pd_info->uncore_base = devm_ioremap_resource(&auxdev->dev, res); + if (IS_ERR(pd_info->uncore_base)) { + ret = PTR_ERR(pd_info->uncore_base); + pd_info->uncore_base = NULL; + goto err_rem_common; + } + + /* Check for version and skip this resource if there is mismatch */ + header = readq(pd_info->uncore_base); + pd_info->ufs_header_ver = header & UNCORE_GENMASK_VERSION; + if (pd_info->ufs_header_ver != UNCORE_HEADER_VERSION) { + dev_info(&auxdev->dev, "Uncore: Unsupported version:%d\n", + pd_info->ufs_header_ver); + continue; + } + + /* Get Cluster ID Mask */ + cluster_mask = FIELD_GET(UNCORE_LOCAL_FABRIC_CLUSTER_ID_MASK, header); + if (!cluster_mask) { + dev_info(&auxdev->dev, "Uncore: Invalid cluster mask:%x\n", cluster_mask); + continue; + } + + /* Find out number of clusters in this resource */ + mask = 0x01; + for (j = 0; j < UNCORE_MAX_CLUSTER_PER_DOMAIN; ++j) { + if (cluster_mask & mask) + pd_info->cluster_count++; + mask <<= 1; + } + + pd_info->cluster_infos = devm_kcalloc(&auxdev->dev, pd_info->cluster_count, + sizeof(struct tpmi_uncore_cluster_info), + GFP_KERNEL); + + /* + * Each byte in the register point to status and control + * registers belonging to cluster id 0-8. + */ + cluster_offset = readq((u8 __iomem *)pd_info->uncore_base + + UNCORE_FABRIC_CLUSTER_OFFSET); + + for (j = 0; j < pd_info->cluster_count; ++j) { + struct tpmi_uncore_cluster_info *cluster_info; + + /* Get the offset for this cluster */ + mask = (cluster_offset & UNCORE_CLUSTER_OFF_MASK); + /* Offset in QWORD, so change to bytes */ + mask <<= 3; + + cluster_info = &pd_info->cluster_infos[j]; + + cluster_info->cluster_base = (u8 __iomem *)pd_info->uncore_base + mask; + + cluster_info->uncore_data.package_id = pkg; + /* There are no dies like Cascade Lake */ + cluster_info->uncore_data.die_id = 0; + + /* Point to next cluster offset */ + cluster_offset >>= UNCORE_MAX_CLUSTER_PER_DOMAIN; + } + } + + auxiliary_set_drvdata(auxdev, tpmi_uncore); + + tpmi_uncore->root_cluster.uncore_root = tpmi_uncore; + tpmi_uncore->root_cluster.uncore_data.package_id = pkg; + ret = uncore_freq_add_entry(&tpmi_uncore->root_cluster.uncore_data, 0); + if (ret) + goto err_rem_common; + + return 0; + +err_rem_common: + uncore_freq_common_exit(); + + return ret; +} + +static int tpmi_uncore_remove(struct auxiliary_device *auxdev) +{ + struct tpmi_uncore_struct *tpmi_uncore = auxiliary_get_drvdata(auxdev); + + uncore_freq_remove_die_entry(&tpmi_uncore->root_cluster.uncore_data); + + uncore_freq_common_exit(); + + return 0; +} + +static int intel_uncore_probe(struct auxiliary_device *auxdev, const struct auxiliary_device_id *id) +{ + return tpmi_uncore_init(auxdev); +} + +static void intel_uncore_remove(struct auxiliary_device *auxdev) +{ + tpmi_uncore_remove(auxdev); +} + +static const struct auxiliary_device_id intel_uncore_id_table[] = { + { .name = "intel_vsec.tpmi-uncore" }, + {} +}; +MODULE_DEVICE_TABLE(auxiliary, intel_uncore_id_table); + +static struct auxiliary_driver intel_uncore_aux_driver = { + .id_table = intel_uncore_id_table, + .remove = intel_uncore_remove, + .probe = intel_uncore_probe, +}; + +module_auxiliary_driver(intel_uncore_aux_driver); + +MODULE_IMPORT_NS(INTEL_TPMI); +MODULE_IMPORT_NS(INTEL_UNCORE_FREQUENCY); +MODULE_DESCRIPTION("Intel TPMI UFS Driver"); +MODULE_LICENSE("GPL");
Implement support of uncore frequency control via TPMI (Topology Aware Register and PM Capsule Interface). This driver provides the similar functionality as the current uncore frequency driver using MSRs. The hardware interface to read/write is basically substitution of MSR 0x620 and 0x621. There are specific MMIO offset and bits to get/set minimum and maximum uncore ratio, similar to MSRs. The scope of the uncore MSRs is package/die. But new generation of CPUs have more granular control at a cluster level. Each package/die can have multiple power domains, which further can have multiple clusters. The TPMI interface allows control at cluster level. The primary use case for uncore sysfs is to set maximum and minimum uncore frequency to reduce power consumption or latency. The current uncore sysfs control is per package/die. This is enough for the majority of users as workload will move to different power domains as it moves between different CPUs. The current uncore sysfs provides controls at package/die level. When user sets maximum/minimum limits, the driver sets the same limits to each cluster. Here number of power domains = number of resources in this aux device. There are offsets and bits to discover number of clusters and offset for each cluster level controls. The TPMI documentation can be downloaded from: https://github.com/intel/tpmi_power_management Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com> --- The fine grain control at cluster level patches are here: https://github.com/spandruvada/linux-kernel/commit/0d66ea4ff76ea19127f2d207a7e17bb86846ca32 https://github.com/spandruvada/linux-kernel/commit/cb5c2349a58318c04955821d6528cc8015541e65 But not submitting to ease in review process as I posted too many patches this cycle. .../x86/intel/uncore-frequency/Kconfig | 4 + .../x86/intel/uncore-frequency/Makefile | 2 + .../uncore-frequency/uncore-frequency-tpmi.c | 346 ++++++++++++++++++ 3 files changed, 352 insertions(+) create mode 100644 drivers/platform/x86/intel/uncore-frequency/uncore-frequency-tpmi.c