Messages in this thread | | | From | "Rafael J. Wysocki" <> | Subject | Re: [PATCH] acpi: Issue _OSC call for native thermal interrupt handling | Date | Wed, 16 Mar 2016 23:48:41 +0100 |
| |
On Wednesday, March 16, 2016 03:25:19 PM Srinivas Pandruvada wrote: > There are several reports of freeze on enabling HWP (Hardware PStates) > feature on Skylake based systems by Intel P states driver. The root > cause is identified as the HWP interrupts causing BIOS code to freeze. > HWP interrupts uses thermal LVT. > Linux natively handles thermal interrupts, but in Skylake based systems > SMM will take control of thermal interrupts. This is a problem for several > reasons: > - It is freezing in BIOS when tries to handle thermal interrupt, which > will require BIOS upgrade > - With SMM handling thermal we loose all the reporting features of > Linux arch/x86/kernel/cpu/mcheck/therm_throt driver > - Some thermal drivers like x86-package-temp driver depends on the thermal > threshold interrupts > - The HWP interrupts are useful for debugging and tuning performance > > So we need native handling of thermal interrupts. To inform SMM that > OS will handle thermal interrupts, we need to use _OSC under processor > scope very early in ACPI initialization flow. This needs to be done > before SMM code path looks for _OSC capabilities. The bit 12 of > _OSC in processor scope defines whether OS will handle thermal interrupts. > When bit 12 is set to 1, OS will handle thermal interrupts. > Refer to this document for details on _OSC > http://www.intel.com/content/www/us/en/standards/processor-vendor- > specific-acpi-specification.html > > This change introduces a new function acpi_early_processor_set_osc(), > which walks acpi name space and finds acpi processor object and > set capability via _OSC method to take over thermal LVT. > > Also this change writes HWP status bits to 0 to clear any HWP status > bits in intel_thermal_interrupt(). > > Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Has this version of the patch been tested on a system where the problem is reproducible?
> --- > arch/x86/kernel/cpu/mcheck/therm_throt.c | 8 ++++++ > drivers/acpi/acpi_processor.c | 42 +++++++++++++++++++++++++++++++- > drivers/acpi/bus.c | 3 +++ > drivers/acpi/internal.h | 2 ++ > 4 files changed, 54 insertions(+), 1 deletion(-) > > diff --git a/arch/x86/kernel/cpu/mcheck/therm_throt.c b/arch/x86/kernel/cpu/mcheck/therm_throt.c > index 0b445c2..bb331f6 100644 > --- a/arch/x86/kernel/cpu/mcheck/therm_throt.c > +++ b/arch/x86/kernel/cpu/mcheck/therm_throt.c > @@ -79,6 +79,8 @@ static atomic_t therm_throt_en = ATOMIC_INIT(0); > > static u32 lvtthmr_init __read_mostly; > > +static bool thermal_hwp_interrupt_support; > + > #ifdef CONFIG_SYSFS > #define define_therm_throt_device_one_ro(_name) \ > static DEVICE_ATTR(_name, 0444, \ > @@ -384,6 +386,9 @@ static void intel_thermal_interrupt(void) > { > __u64 msr_val; > > + if (thermal_hwp_interrupt_support) > + wrmsrl_safe(MSR_HWP_STATUS, 0); > + > rdmsrl(MSR_IA32_THERM_STATUS, msr_val); > > /* Check for violation of core thermal thresholds*/ > @@ -552,6 +557,9 @@ void intel_init_thermal(struct cpuinfo_x86 *c) > rdmsr(MSR_IA32_MISC_ENABLE, l, h); > wrmsr(MSR_IA32_MISC_ENABLE, l | MSR_IA32_MISC_ENABLE_TM1, h); > > + if (static_cpu_has(X86_FEATURE_HWP)) > + thermal_hwp_interrupt_support = true; > + > /* Unmask the thermal vector: */ > l = apic_read(APIC_LVTTHMR); > apic_write(APIC_LVTTHMR, l & ~APIC_LVT_MASKED); > diff --git a/drivers/acpi/acpi_processor.c b/drivers/acpi/acpi_processor.c > index 6979186..5a78279 100644 > --- a/drivers/acpi/acpi_processor.c > +++ b/drivers/acpi/acpi_processor.c > @@ -391,7 +391,6 @@ static int acpi_processor_add(struct acpi_device *device, > if (pr->id >= setup_max_cpus && pr->id != 0) > return 0; > #endif > - > BUG_ON(pr->id >= nr_cpu_ids); > > /* > @@ -491,6 +490,47 @@ static void acpi_processor_remove(struct acpi_device *device) > } > #endif /* CONFIG_ACPI_HOTPLUG_CPU */ >
It occurs to me that the whole thing is only necessary on x86, so maybe it can go under a #ifdef?
Or even to arch/x86/acpi/ (boot.c or a new file)?
> +static bool acpi_hwp_native_thermal_lvt_set; > +static acpi_status acpi_set_hwp_native_thermal_lvt_osc(acpi_handle handle, > + u32 lvl, void *context, > + void **rv) > +{ > + u8 sb_uuid_str[] = "4077A616-290C-47BE-9EBD-D87058713953"; > + u32 capbuf[2]; > + struct acpi_osc_context osc_context = { > + .uuid_str = sb_uuid_str, > + .rev = 1, > + .cap.length = 8, > + .cap.pointer = capbuf, > + }; > + > + if (acpi_hwp_native_thermal_lvt_set) > + return AE_OK; > + > + if (!static_cpu_has(X86_FEATURE_HWP)) > + return AE_OK;
This check can be made once in acpi_early_processor_set_osc().
> + > + capbuf[0] = 0x0000; > + capbuf[1] = 0x1000; /* set bit 12 */ > + > + if (ACPI_SUCCESS(acpi_run_osc(handle, &osc_context))) { > + acpi_hwp_native_thermal_lvt_set = true; > + kfree(osc_context.ret.pointer); > + } > + > + return AE_OK; > +} > + > +void acpi_early_processor_set_osc(void) > +{ > + acpi_walk_namespace(ACPI_TYPE_PROCESSOR, ACPI_ROOT_OBJECT, > + ACPI_UINT32_MAX, > + acpi_set_hwp_native_thermal_lvt_osc, > + NULL, NULL, NULL); > + acpi_get_devices(ACPI_PROCESSOR_DEVICE_HID, > + acpi_set_hwp_native_thermal_lvt_osc, NULL, NULL); > +} > + > /* > * The following ACPI IDs are known to be suitable for representing as > * processor devices. > diff --git a/drivers/acpi/bus.c b/drivers/acpi/bus.c > index 891c42d..7e73aea 100644 > --- a/drivers/acpi/bus.c > +++ b/drivers/acpi/bus.c > @@ -1005,6 +1005,9 @@ static int __init acpi_bus_init(void) > goto error1; > } > > + /* Set capability bits for _OSC under processor scope */ > + acpi_early_processor_set_osc(); > + > /* > * _OSC method may exist in module level code, > * so it must be run after ACPI_FULL_INITIALIZATION > diff --git a/drivers/acpi/internal.h b/drivers/acpi/internal.h > index 1e6833a..5c787ac 100644 > --- a/drivers/acpi/internal.h > +++ b/drivers/acpi/internal.h > @@ -138,6 +138,8 @@ void acpi_early_processor_set_pdc(void); > static inline void acpi_early_processor_set_pdc(void) {} > #endif > > +void acpi_early_processor_set_osc(void); > + > /* -------------------------------------------------------------------------- > Embedded Controller > -------------------------------------------------------------------------- */ >
Thanks, Rafael
| |