Message ID | 20231018231624.1044633-17-david.e.box@linux.intel.com (mailing list archive) |
---|---|
State | Changes Requested, archived |
Headers | show |
Series | intel_pmc: Add telemetry API to read counters | expand |
On Wed, 18 Oct 2023, David E. Box wrote: > Add a "die_c6_us_show" debugfs attribute. Reads the counter value using > Intel Platform Monitoring Technology (PMT) driver API. This counter is > useful for determining the idle residency of CPUs in the compute tile. > > Signed-off-by: David E. Box <david.e.box@linux.intel.com> > --- > V4 - no change > > V3 - Split previous PATCH V2 13. Separates implementation (this patch) from > platform specific use (next patch) > > V2 - Remove use of __func__ > - Use HZ_PER_MHZ > - Fix missing newlines in printks > > drivers/platform/x86/intel/pmc/core.c | 55 +++++++++++++++++++++++++++ > drivers/platform/x86/intel/pmc/core.h | 4 ++ > 2 files changed, 59 insertions(+) > > diff --git a/drivers/platform/x86/intel/pmc/core.c b/drivers/platform/x86/intel/pmc/core.c > index fcb0dc702aea..02f3e909cf22 100644 > --- a/drivers/platform/x86/intel/pmc/core.c > +++ b/drivers/platform/x86/intel/pmc/core.c > @@ -20,6 +20,7 @@ > #include <linux/pci.h> > #include <linux/slab.h> > #include <linux/suspend.h> > +#include <linux/units.h> > > #include <asm/cpu_device_id.h> > #include <asm/intel-family.h> > @@ -27,6 +28,7 @@ > #include <asm/tsc.h> > > #include "core.h" > +#include "../pmt/telemetry.h" > > /* Maximum number of modes supported by platfoms that has low power mode capability */ > const char *pmc_lpm_modes[] = { > @@ -822,6 +824,47 @@ static int pmc_core_substate_req_regs_show(struct seq_file *s, void *unused) > } > DEFINE_SHOW_ATTRIBUTE(pmc_core_substate_req_regs); > > +static unsigned int pmc_core_get_crystal_freq(void) > +{ > + unsigned int eax_denominator, ebx_numerator, ecx_hz, edx; > + > + if (boot_cpu_data.cpuid_level < 0x15) > + return 0; > + > + eax_denominator = ebx_numerator = ecx_hz = edx = 0; > + > + /* CPUID 15H TSC/Crystal ratio, plus optionally Crystal Hz */ > + cpuid(0x15, &eax_denominator, &ebx_numerator, &ecx_hz, &edx); > + > + if (ebx_numerator == 0 || eax_denominator == 0) > + return 0; > + > + return ecx_hz; > +} > + > +static int pmc_core_die_c6_us_show(struct seq_file *s, void *unused) > +{ > + struct pmc_dev *pmcdev = s->private; > + u64 die_c6_res, count; > + int ret; > + > + if (!pmcdev->crystal_freq) { > + dev_warn_once(&pmcdev->pdev->dev, "Bad crystal frequency\n"); Isn't it more like crystal frequency is not provided rather than bad frequency? > + return -EINVAL; -EINVAL is not good value to return here since there was nothing wrong with the input. Maybe -ENXIO would be better. > + } > + > + ret = pmt_telem_read(pmcdev->punit_ep, pmcdev->die_c6_offset, > + &count, 1); > + if (ret) > + return ret; > + > + die_c6_res = div64_u64(count * HZ_PER_MHZ, pmcdev->crystal_freq); > + seq_printf(s, "%llu\n", die_c6_res); > + > + return 0; > +} > +DEFINE_SHOW_ATTRIBUTE(pmc_core_die_c6_us); > + > static int pmc_core_lpm_latch_mode_show(struct seq_file *s, void *unused) > { > struct pmc_dev *pmcdev = s->private; > @@ -1118,6 +1161,12 @@ static void pmc_core_dbgfs_register(struct pmc_dev *pmcdev) > pmcdev->dbgfs_dir, pmcdev, > &pmc_core_substate_req_regs_fops); > } > + > + if (pmcdev->has_die_c6) { > + debugfs_create_file("die_c6_us_show", 0444, > + pmcdev->dbgfs_dir, pmcdev, > + &pmc_core_die_c6_us_fops); > + } > } > > static const struct x86_cpu_id intel_pmc_core_ids[] = { > @@ -1212,6 +1261,10 @@ static void pmc_core_clean_structure(struct platform_device *pdev) > pci_dev_put(pmcdev->ssram_pcidev); > pci_disable_device(pmcdev->ssram_pcidev); > } > + > + if (pmcdev->punit_ep) > + pmt_telem_unregister_endpoint(pmcdev->punit_ep); > + > platform_set_drvdata(pdev, NULL); > mutex_destroy(&pmcdev->lock); > } > @@ -1232,6 +1285,8 @@ static int pmc_core_probe(struct platform_device *pdev) > if (!pmcdev) > return -ENOMEM; > > + pmcdev->crystal_freq = pmc_core_get_crystal_freq(); > + > platform_set_drvdata(pdev, pmcdev); > pmcdev->pdev = pdev; > > diff --git a/drivers/platform/x86/intel/pmc/core.h b/drivers/platform/x86/intel/pmc/core.h > index 85b6f6ae4995..6d7673145f90 100644 > --- a/drivers/platform/x86/intel/pmc/core.h > +++ b/drivers/platform/x86/intel/pmc/core.h > @@ -16,6 +16,8 @@ > #include <linux/bits.h> > #include <linux/platform_device.h> > > +struct telem_endpoint; > + This seems unrelated to the patch.
On Mon, 2023-10-23 at 19:31 +0300, Ilpo Järvinen wrote: > On Wed, 18 Oct 2023, David E. Box wrote: > > > Add a "die_c6_us_show" debugfs attribute. Reads the counter value using > > Intel Platform Monitoring Technology (PMT) driver API. This counter is > > useful for determining the idle residency of CPUs in the compute tile. > > > > Signed-off-by: David E. Box <david.e.box@linux.intel.com> > > --- > > V4 - no change > > > > V3 - Split previous PATCH V2 13. Separates implementation (this patch) from > > platform specific use (next patch) > > > > V2 - Remove use of __func__ > > - Use HZ_PER_MHZ > > - Fix missing newlines in printks > > > > drivers/platform/x86/intel/pmc/core.c | 55 +++++++++++++++++++++++++++ > > drivers/platform/x86/intel/pmc/core.h | 4 ++ > > 2 files changed, 59 insertions(+) > > > > diff --git a/drivers/platform/x86/intel/pmc/core.c > > b/drivers/platform/x86/intel/pmc/core.c > > index fcb0dc702aea..02f3e909cf22 100644 > > --- a/drivers/platform/x86/intel/pmc/core.c > > +++ b/drivers/platform/x86/intel/pmc/core.c > > @@ -20,6 +20,7 @@ > > #include <linux/pci.h> > > #include <linux/slab.h> > > #include <linux/suspend.h> > > +#include <linux/units.h> > > > > #include <asm/cpu_device_id.h> > > #include <asm/intel-family.h> > > @@ -27,6 +28,7 @@ > > #include <asm/tsc.h> > > > > #include "core.h" > > +#include "../pmt/telemetry.h" > > > > /* Maximum number of modes supported by platfoms that has low power mode > > capability */ > > const char *pmc_lpm_modes[] = { > > @@ -822,6 +824,47 @@ static int pmc_core_substate_req_regs_show(struct > > seq_file *s, void *unused) > > } > > DEFINE_SHOW_ATTRIBUTE(pmc_core_substate_req_regs); > > > > +static unsigned int pmc_core_get_crystal_freq(void) > > +{ > > + unsigned int eax_denominator, ebx_numerator, ecx_hz, edx; > > + > > + if (boot_cpu_data.cpuid_level < 0x15) > > + return 0; > > + > > + eax_denominator = ebx_numerator = ecx_hz = edx = 0; > > + > > + /* CPUID 15H TSC/Crystal ratio, plus optionally Crystal Hz */ > > + cpuid(0x15, &eax_denominator, &ebx_numerator, &ecx_hz, &edx); > > + > > + if (ebx_numerator == 0 || eax_denominator == 0) > > + return 0; > > + > > + return ecx_hz; > > +} > > + > > +static int pmc_core_die_c6_us_show(struct seq_file *s, void *unused) > > +{ > > + struct pmc_dev *pmcdev = s->private; > > + u64 die_c6_res, count; > > + int ret; > > + > > + if (!pmcdev->crystal_freq) { > > + dev_warn_once(&pmcdev->pdev->dev, "Bad crystal > > frequency\n"); > > Isn't it more like crystal frequency is not provided rather than bad > frequency? > > > + return -EINVAL; > > -EINVAL is not good value to return here since there was nothing wrong > with the input. Maybe -ENXIO would be better. Will make these changes. > > > + } > > + > > + ret = pmt_telem_read(pmcdev->punit_ep, pmcdev->die_c6_offset, > > + &count, 1); > > + if (ret) > > + return ret; > > + > > + die_c6_res = div64_u64(count * HZ_PER_MHZ, pmcdev->crystal_freq); > > + seq_printf(s, "%llu\n", die_c6_res); > > + > > + return 0; > > +} > > +DEFINE_SHOW_ATTRIBUTE(pmc_core_die_c6_us); > > + > > static int pmc_core_lpm_latch_mode_show(struct seq_file *s, void *unused) > > { > > struct pmc_dev *pmcdev = s->private; > > @@ -1118,6 +1161,12 @@ static void pmc_core_dbgfs_register(struct pmc_dev > > *pmcdev) > > pmcdev->dbgfs_dir, pmcdev, > > &pmc_core_substate_req_regs_fops); > > } > > + > > + if (pmcdev->has_die_c6) { > > + debugfs_create_file("die_c6_us_show", 0444, > > + pmcdev->dbgfs_dir, pmcdev, > > + &pmc_core_die_c6_us_fops); > > + } > > } > > > > static const struct x86_cpu_id intel_pmc_core_ids[] = { > > @@ -1212,6 +1261,10 @@ static void pmc_core_clean_structure(struct > > platform_device *pdev) > > pci_dev_put(pmcdev->ssram_pcidev); > > pci_disable_device(pmcdev->ssram_pcidev); > > } > > + > > + if (pmcdev->punit_ep) ... > > + pmt_telem_unregister_endpoint(pmcdev->punit_ep); > > + > > platform_set_drvdata(pdev, NULL); > > mutex_destroy(&pmcdev->lock); > > } > > @@ -1232,6 +1285,8 @@ static int pmc_core_probe(struct platform_device > > *pdev) > > if (!pmcdev) > > return -ENOMEM; > > > > + pmcdev->crystal_freq = pmc_core_get_crystal_freq(); > > + > > platform_set_drvdata(pdev, pmcdev); > > pmcdev->pdev = pdev; > > > > diff --git a/drivers/platform/x86/intel/pmc/core.h > > b/drivers/platform/x86/intel/pmc/core.h > > index 85b6f6ae4995..6d7673145f90 100644 > > --- a/drivers/platform/x86/intel/pmc/core.h > > +++ b/drivers/platform/x86/intel/pmc/core.h > > @@ -16,6 +16,8 @@ > > #include <linux/bits.h> > > #include <linux/platform_device.h> > > > > +struct telem_endpoint; > > + > > This seems unrelated to the patch. This was missing when "struct telem_endpoint *punit_ep" was declared in this header in an earlier patch that went upstream. But punit_ep does not get used yet until this patch as shown above. Will clarify in the changelog. David >
diff --git a/drivers/platform/x86/intel/pmc/core.c b/drivers/platform/x86/intel/pmc/core.c index fcb0dc702aea..02f3e909cf22 100644 --- a/drivers/platform/x86/intel/pmc/core.c +++ b/drivers/platform/x86/intel/pmc/core.c @@ -20,6 +20,7 @@ #include <linux/pci.h> #include <linux/slab.h> #include <linux/suspend.h> +#include <linux/units.h> #include <asm/cpu_device_id.h> #include <asm/intel-family.h> @@ -27,6 +28,7 @@ #include <asm/tsc.h> #include "core.h" +#include "../pmt/telemetry.h" /* Maximum number of modes supported by platfoms that has low power mode capability */ const char *pmc_lpm_modes[] = { @@ -822,6 +824,47 @@ static int pmc_core_substate_req_regs_show(struct seq_file *s, void *unused) } DEFINE_SHOW_ATTRIBUTE(pmc_core_substate_req_regs); +static unsigned int pmc_core_get_crystal_freq(void) +{ + unsigned int eax_denominator, ebx_numerator, ecx_hz, edx; + + if (boot_cpu_data.cpuid_level < 0x15) + return 0; + + eax_denominator = ebx_numerator = ecx_hz = edx = 0; + + /* CPUID 15H TSC/Crystal ratio, plus optionally Crystal Hz */ + cpuid(0x15, &eax_denominator, &ebx_numerator, &ecx_hz, &edx); + + if (ebx_numerator == 0 || eax_denominator == 0) + return 0; + + return ecx_hz; +} + +static int pmc_core_die_c6_us_show(struct seq_file *s, void *unused) +{ + struct pmc_dev *pmcdev = s->private; + u64 die_c6_res, count; + int ret; + + if (!pmcdev->crystal_freq) { + dev_warn_once(&pmcdev->pdev->dev, "Bad crystal frequency\n"); + return -EINVAL; + } + + ret = pmt_telem_read(pmcdev->punit_ep, pmcdev->die_c6_offset, + &count, 1); + if (ret) + return ret; + + die_c6_res = div64_u64(count * HZ_PER_MHZ, pmcdev->crystal_freq); + seq_printf(s, "%llu\n", die_c6_res); + + return 0; +} +DEFINE_SHOW_ATTRIBUTE(pmc_core_die_c6_us); + static int pmc_core_lpm_latch_mode_show(struct seq_file *s, void *unused) { struct pmc_dev *pmcdev = s->private; @@ -1118,6 +1161,12 @@ static void pmc_core_dbgfs_register(struct pmc_dev *pmcdev) pmcdev->dbgfs_dir, pmcdev, &pmc_core_substate_req_regs_fops); } + + if (pmcdev->has_die_c6) { + debugfs_create_file("die_c6_us_show", 0444, + pmcdev->dbgfs_dir, pmcdev, + &pmc_core_die_c6_us_fops); + } } static const struct x86_cpu_id intel_pmc_core_ids[] = { @@ -1212,6 +1261,10 @@ static void pmc_core_clean_structure(struct platform_device *pdev) pci_dev_put(pmcdev->ssram_pcidev); pci_disable_device(pmcdev->ssram_pcidev); } + + if (pmcdev->punit_ep) + pmt_telem_unregister_endpoint(pmcdev->punit_ep); + platform_set_drvdata(pdev, NULL); mutex_destroy(&pmcdev->lock); } @@ -1232,6 +1285,8 @@ static int pmc_core_probe(struct platform_device *pdev) if (!pmcdev) return -ENOMEM; + pmcdev->crystal_freq = pmc_core_get_crystal_freq(); + platform_set_drvdata(pdev, pmcdev); pmcdev->pdev = pdev; diff --git a/drivers/platform/x86/intel/pmc/core.h b/drivers/platform/x86/intel/pmc/core.h index 85b6f6ae4995..6d7673145f90 100644 --- a/drivers/platform/x86/intel/pmc/core.h +++ b/drivers/platform/x86/intel/pmc/core.h @@ -16,6 +16,8 @@ #include <linux/bits.h> #include <linux/platform_device.h> +struct telem_endpoint; + #define SLP_S0_RES_COUNTER_MASK GENMASK(31, 0) #define PMC_BASE_ADDR_DEFAULT 0xFE000000 @@ -357,6 +359,7 @@ struct pmc { * @devs: pointer to an array of pmc pointers * @pdev: pointer to platform_device struct * @ssram_pcidev: pointer to pci device struct for the PMC SSRAM + * @crystal_freq: crystal frequency from cpuid * @dbgfs_dir: path to debugfs interface * @pmc_xram_read_bit: flag to indicate whether PMC XRAM shadow registers * used to read MPHY PG and PLL status are available @@ -374,6 +377,7 @@ struct pmc_dev { struct dentry *dbgfs_dir; struct platform_device *pdev; struct pci_dev *ssram_pcidev; + unsigned int crystal_freq; int pmc_xram_read_bit; struct mutex lock; /* generic mutex lock for PMC Core */
Add a "die_c6_us_show" debugfs attribute. Reads the counter value using Intel Platform Monitoring Technology (PMT) driver API. This counter is useful for determining the idle residency of CPUs in the compute tile. Signed-off-by: David E. Box <david.e.box@linux.intel.com> --- V4 - no change V3 - Split previous PATCH V2 13. Separates implementation (this patch) from platform specific use (next patch) V2 - Remove use of __func__ - Use HZ_PER_MHZ - Fix missing newlines in printks drivers/platform/x86/intel/pmc/core.c | 55 +++++++++++++++++++++++++++ drivers/platform/x86/intel/pmc/core.h | 4 ++ 2 files changed, 59 insertions(+)