From 30c7e5b123673d5e570e238dbada2fb68a87212c Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 22 Dec 2017 10:20:11 +0100 Subject: x86/tsc: Allow TSC calibration without PIT Zhang Rui reported that a Surface Pro 4 will fail to boot with lapic=notscdeadline. Part of the problem is that that machine doesn't have a PIT. If, for some reason, the TSC init has to fall back to TSC calibration, it relies on the PIT to be present. Allow TSC calibration to reliably fall back to HPET. The below results in an accurate TSC measurement when forced on a IVB: tsc: Unable to calibrate against PIT tsc: No reference (HPET/PMTIMER) available tsc: Unable to calibrate against PIT tsc: using HPET reference calibration tsc: Detected 2792.451 MHz processor Signed-off-by: Peter Zijlstra (Intel) Signed-off-by: Thomas Gleixner Cc: len.brown@intel.com Cc: rui.zhang@intel.com Link: https://lkml.kernel.org/r/20171222092243.333145937@infradead.org --- arch/x86/include/asm/i8259.h | 5 +++++ arch/x86/kernel/tsc.c | 18 ++++++++++++++++++ 2 files changed, 23 insertions(+) (limited to 'arch') diff --git a/arch/x86/include/asm/i8259.h b/arch/x86/include/asm/i8259.h index c8376b40e882..5cdcdbd4d892 100644 --- a/arch/x86/include/asm/i8259.h +++ b/arch/x86/include/asm/i8259.h @@ -69,6 +69,11 @@ struct legacy_pic { extern struct legacy_pic *legacy_pic; extern struct legacy_pic null_legacy_pic; +static inline bool has_legacy_pic(void) +{ + return legacy_pic != &null_legacy_pic; +} + static inline int nr_legacy_irqs(void) { return legacy_pic->nr_legacy_irqs; diff --git a/arch/x86/kernel/tsc.c b/arch/x86/kernel/tsc.c index e169e85db434..a2c9dd8bfc6f 100644 --- a/arch/x86/kernel/tsc.c +++ b/arch/x86/kernel/tsc.c @@ -25,6 +25,7 @@ #include #include #include +#include unsigned int __read_mostly cpu_khz; /* TSC clocks / usec, not used here */ EXPORT_SYMBOL(cpu_khz); @@ -363,6 +364,20 @@ static unsigned long pit_calibrate_tsc(u32 latch, unsigned long ms, int loopmin) unsigned long tscmin, tscmax; int pitcnt; + if (!has_legacy_pic()) { + /* + * Relies on tsc_early_delay_calibrate() to have given us semi + * usable udelay(), wait for the same 50ms we would have with + * the PIT loop below. + */ + udelay(10 * USEC_PER_MSEC); + udelay(10 * USEC_PER_MSEC); + udelay(10 * USEC_PER_MSEC); + udelay(10 * USEC_PER_MSEC); + udelay(10 * USEC_PER_MSEC); + return ULONG_MAX; + } + /* Set the Gate high, disable speaker */ outb((inb(0x61) & ~0x02) | 0x01, 0x61); @@ -487,6 +502,9 @@ static unsigned long quick_pit_calibrate(void) u64 tsc, delta; unsigned long d1, d2; + if (!has_legacy_pic()) + return 0; + /* Set the Gate high, disable speaker */ outb((inb(0x61) & ~0x02) | 0x01, 0x61); -- cgit v1.2.3-70-g09d2 From 6d671e1b85c63e7a337ba76c1a154c091545cff8 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 22 Dec 2017 10:20:12 +0100 Subject: x86/time: Unconditionally register legacy timer interrupt Even without a PIC/PIT the legacy timer interrupt is required for HPET in legacy replacement mode. Signed-off-by: Peter Zijlstra (Intel) Signed-off-by: Thomas Gleixner Cc: len.brown@intel.com Cc: rui.zhang@intel.com Link: https://lkml.kernel.org/r/20171222092243.382623763@infradead.org --- arch/x86/kernel/time.c | 9 ++++++--- 1 file changed, 6 insertions(+), 3 deletions(-) (limited to 'arch') diff --git a/arch/x86/kernel/time.c b/arch/x86/kernel/time.c index 749d189f8cd4..774ebafa97c4 100644 --- a/arch/x86/kernel/time.c +++ b/arch/x86/kernel/time.c @@ -69,9 +69,12 @@ static struct irqaction irq0 = { static void __init setup_default_timer_irq(void) { - if (!nr_legacy_irqs()) - return; - setup_irq(0, &irq0); + /* + * Unconditionally register the legacy timer; even without legacy + * PIC/PIT we need this for the HPET0 in legacy replacement mode. + */ + if (setup_irq(0, &irq0)) + pr_info("Failed to register legacy timer interrupt\n"); } /* Default timer init function */ -- cgit v1.2.3-70-g09d2 From aa83c45762a242acce9b35020363225a7b59d7c9 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 22 Dec 2017 10:20:13 +0100 Subject: x86/tsc: Introduce early tsc clocksource Without TSC_KNOWN_FREQ the TSC clocksource is registered so late that the kernel first switches to the HPET. Using HPET on large CPU count machines is undesirable. Therefore register a tsc-early clocksource using the preliminary tsc_khz from quick calibration. Then when the final TSC calibration is done, it can switch to the tuned frequency. The only notably problem is that the real tsc clocksource must be marked with CLOCK_SOURCE_VALID_FOR_HRES, otherwise it will not be selected when unregistering tsc-early. tsc-early cannot be left registered, because then the clocksource code would fall back to it when we tsc clocksource is marked unstable later. Signed-off-by: Peter Zijlstra (Intel) Signed-off-by: Thomas Gleixner Cc: len.brown@intel.com Cc: rui.zhang@intel.com Cc: Len Brown Link: https://lkml.kernel.org/r/20171222092243.431585460@infradead.org --- arch/x86/kernel/tsc.c | 43 +++++++++++++++++++++++++++++++++---------- 1 file changed, 33 insertions(+), 10 deletions(-) (limited to 'arch') diff --git a/arch/x86/kernel/tsc.c b/arch/x86/kernel/tsc.c index a2c9dd8bfc6f..fb4302738410 100644 --- a/arch/x86/kernel/tsc.c +++ b/arch/x86/kernel/tsc.c @@ -1006,8 +1006,6 @@ static void __init detect_art(void) /* clocksource code */ -static struct clocksource clocksource_tsc; - static void tsc_resume(struct clocksource *cs) { tsc_verify_tsc_adjust(true); @@ -1058,12 +1056,31 @@ static void tsc_cs_tick_stable(struct clocksource *cs) /* * .mask MUST be CLOCKSOURCE_MASK(64). See comment above read_tsc() */ +static struct clocksource clocksource_tsc_early = { + .name = "tsc-early", + .rating = 299, + .read = read_tsc, + .mask = CLOCKSOURCE_MASK(64), + .flags = CLOCK_SOURCE_IS_CONTINUOUS | + CLOCK_SOURCE_MUST_VERIFY, + .archdata = { .vclock_mode = VCLOCK_TSC }, + .resume = tsc_resume, + .mark_unstable = tsc_cs_mark_unstable, + .tick_stable = tsc_cs_tick_stable, +}; + +/* + * Must mark VALID_FOR_HRES early such that when we unregister tsc_early + * this one will immediately take over. We will only register if TSC has + * been found good. + */ static struct clocksource clocksource_tsc = { .name = "tsc", .rating = 300, .read = read_tsc, .mask = CLOCKSOURCE_MASK(64), .flags = CLOCK_SOURCE_IS_CONTINUOUS | + CLOCK_SOURCE_VALID_FOR_HRES | CLOCK_SOURCE_MUST_VERIFY, .archdata = { .vclock_mode = VCLOCK_TSC }, .resume = tsc_resume, @@ -1187,8 +1204,8 @@ static void tsc_refine_calibration_work(struct work_struct *work) int cpu; /* Don't bother refining TSC on unstable systems */ - if (check_tsc_unstable()) - goto out; + if (tsc_unstable) + return; /* * Since the work is started early in boot, we may be @@ -1240,9 +1257,13 @@ static void tsc_refine_calibration_work(struct work_struct *work) set_cyc2ns_scale(tsc_khz, cpu, tsc_stop); out: + if (tsc_unstable) + return; + if (boot_cpu_has(X86_FEATURE_ART)) art_related_clocksource = &clocksource_tsc; clocksource_register_khz(&clocksource_tsc, tsc_khz); + clocksource_unregister(&clocksource_tsc_early); } @@ -1251,13 +1272,11 @@ static int __init init_tsc_clocksource(void) if (!boot_cpu_has(X86_FEATURE_TSC) || tsc_disabled > 0 || !tsc_khz) return 0; + if (check_tsc_unstable()) + return 0; + if (tsc_clocksource_reliable) clocksource_tsc.flags &= ~CLOCK_SOURCE_MUST_VERIFY; - /* lower the rating if we already know its unstable: */ - if (check_tsc_unstable()) { - clocksource_tsc.rating = 0; - clocksource_tsc.flags &= ~CLOCK_SOURCE_IS_CONTINUOUS; - } if (boot_cpu_has(X86_FEATURE_NONSTOP_TSC_S3)) clocksource_tsc.flags |= CLOCK_SOURCE_SUSPEND_NONSTOP; @@ -1270,6 +1289,7 @@ static int __init init_tsc_clocksource(void) if (boot_cpu_has(X86_FEATURE_ART)) art_related_clocksource = &clocksource_tsc; clocksource_register_khz(&clocksource_tsc, tsc_khz); + clocksource_unregister(&clocksource_tsc_early); return 0; } @@ -1374,9 +1394,12 @@ void __init tsc_init(void) check_system_tsc_reliable(); - if (unsynchronized_tsc()) + if (unsynchronized_tsc()) { mark_tsc_unstable("TSCs unsynchronized"); + return; + } + clocksource_register_khz(&clocksource_tsc_early, tsc_khz); detect_art(); } -- cgit v1.2.3-70-g09d2 From fe6daab1ee9dfe7f89974ee6c486cccb0f18a61d Mon Sep 17 00:00:00 2001 From: davidwang Date: Mon, 22 Jan 2018 18:14:17 +0800 Subject: x86/centaur: Mark TSC invariant Centaur CPU has a constant frequency TSC and that TSC does not stop in C-States. But because the corresponding TSC feature flags are not set for that CPU, the TSC is treated as not constant frequency and assumed to stop in C-States, which makes it an unreliable and unusable clock source. Setting those flags tells the kernel that the TSC is usable, so it will select it over HPET. The effect of this is that reading time stamps (from kernel or user space) will be faster and more efficent. Signed-off-by: davidwang Signed-off-by: Thomas Gleixner Cc: qiyuanwang@zhaoxin.com Cc: linux-pm@vger.kernel.org Cc: brucechang@via-alliance.com Cc: cooperyan@zhaoxin.com Cc: benjaminpan@viatech.com Link: https://lkml.kernel.org/r/1516616057-5158-1-git-send-email-davidwang@zhaoxin.com --- arch/x86/kernel/cpu/centaur.c | 4 ++++ drivers/acpi/processor_idle.c | 1 + 2 files changed, 5 insertions(+) (limited to 'arch') diff --git a/arch/x86/kernel/cpu/centaur.c b/arch/x86/kernel/cpu/centaur.c index 68bc6d9b3132..c578cd29c2d2 100644 --- a/arch/x86/kernel/cpu/centaur.c +++ b/arch/x86/kernel/cpu/centaur.c @@ -106,6 +106,10 @@ static void early_init_centaur(struct cpuinfo_x86 *c) #ifdef CONFIG_X86_64 set_cpu_cap(c, X86_FEATURE_SYSENTER32); #endif + if (c->x86_power & (1 << 8)) { + set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC); + set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC); + } } static void init_centaur(struct cpuinfo_x86 *c) diff --git a/drivers/acpi/processor_idle.c b/drivers/acpi/processor_idle.c index d50a7b6ccddd..5f0071c7e2e1 100644 --- a/drivers/acpi/processor_idle.c +++ b/drivers/acpi/processor_idle.c @@ -207,6 +207,7 @@ static void tsc_check_state(int state) switch (boot_cpu_data.x86_vendor) { case X86_VENDOR_AMD: case X86_VENDOR_INTEL: + case X86_VENDOR_CENTAUR: /* * AMD Fam10h TSC will tick in all * C/P/S0/S1 states when this bit is set. -- cgit v1.2.3-70-g09d2