From: Naman Jain namjain@linux.microsoft.com Sent: Sunday, September 8, 2024 10:39 PM
read_hv_sched_clock_tsc() assumes that the Hyper-V clock counter is bigger than the variable hv_sched_clock_offset, which is cached during early boot, but depending on the timing this assumption may be false when a hibernated VM starts again (the clock counter starts from 0 again) and is resuming back (Note: hv_init_tsc_clocksource() is not called during hibernation/resume); consequently, read_hv_sched_clock_tsc() may return a negative integer (which is interpreted as a huge positive integer since the return type is u64) and new kernel messages are prefixed with huge timestamps before read_hv_sched_clock_tsc() grows big enough (which typically takes several seconds).
Fix the issue by saving the Hyper-V clock counter just before the suspend, and using it to correct the hv_sched_clock_offset in resume. Override x86_platform.save_sched_clock_state and x86_platform.restore_sched_clock_state so that we don't have to touch the common x86 code.
Note: if Invariant TSC is available, the issue doesn't happen because
- we don't register read_hv_sched_clock_tsc() for sched clock:
See commit e5313f1c5404 ("clocksource/drivers/hyper-v: Rework clocksource and sched clock setup"); 2) the common x86 code adjusts TSC similarly: see __restore_processor_state() -> tsc_verify_tsc_adjust(true) and x86_platform.restore_sched_clock_state().
Cc: stable@vger.kernel.org Fixes: 1349401ff1aa ("clocksource/drivers/hyper-v: Suspend/resume Hyper-V clocksource for hibernation") Co-developed-by: Dexuan Cui decui@microsoft.com Signed-off-by: Dexuan Cui decui@microsoft.com Signed-off-by: Naman Jain namjain@linux.microsoft.com
drivers/clocksource/hyperv_timer.c | 64 +++++++++++++++++++++++++++++- 1 file changed, 63 insertions(+), 1 deletion(-)
diff --git a/drivers/clocksource/hyperv_timer.c b/drivers/clocksource/hyperv_timer.c index b2a080647e41..7aa44b8aae2e 100644 --- a/drivers/clocksource/hyperv_timer.c +++ b/drivers/clocksource/hyperv_timer.c @@ -27,7 +27,10 @@ #include <asm/mshyperv.h>
static struct clock_event_device __percpu *hv_clock_event; -static u64 hv_sched_clock_offset __ro_after_init;
+/* Can have negative values, after resume from hibernation, so keep them s64 */ +static s64 hv_sched_clock_offset __read_mostly; +static s64 hv_sched_clock_offset_saved;
/*
- If false, we're using the old mechanism for stimer0 interrupts
@@ -51,6 +54,9 @@ static int stimer0_irq = -1; static int stimer0_message_sint; static __maybe_unused DEFINE_PER_CPU(long, stimer0_evt);
+static void (*old_save_sched_clock_state)(void); +static void (*old_restore_sched_clock_state)(void);
/*
- Common code for stimer0 interrupts coming via Direct Mode or
- as a VMbus message.
@@ -434,6 +440,39 @@ static u64 noinstr read_hv_sched_clock_tsc(void) (NSEC_PER_SEC / HV_CLOCK_HZ); }
+/*
- Hyper-V clock counter resets during hibernation. Save and restore clock
- offset during suspend/resume, while also considering the time passed
- before suspend. This is to make sure that sched_clock using hv tsc page
- based clocksource, proceeds from where it left off during suspend and
- it shows correct time for the timestamps of kernel messages after resume.
- */
+static void save_hv_clock_tsc_state(void) +{
- hv_sched_clock_offset_saved = hv_read_reference_counter();
+}
+static void restore_hv_clock_tsc_state(void) +{
- /*
* Time passed before suspend = hv_sched_clock_offset_saved* - hv_sched_clock_offset (old)** After Hyper-V clock counter resets, hv_sched_clock_offset needs a correction.** New time = hv_read_reference_counter() (future) - hv_sched_clock_offset(new)
* New time = Time passed before suspend + hv_read_reference_counter()(future)
* - hv_read_reference_counter() (now)** Solving the above two equations gives:** hv_sched_clock_offset (new) = hv_sched_clock_offset (old)* - hv_sched_clock_offset_saved* + hv_read_reference_counter() (now))*/- hv_sched_clock_offset -= hv_sched_clock_offset_saved -
hv_read_reference_counter(); +}
static void suspend_hv_clock_tsc(struct clocksource *arg) { union hv_reference_tsc_msr tsc_msr; @@ -456,6 +495,24 @@ static void resume_hv_clock_tsc(struct clocksource *arg) hv_set_msr(HV_MSR_REFERENCE_TSC, tsc_msr.as_uint64); }
+/*
- Functions to override save_sched_clock_state and restore_sched_clock_state
- functions of x86_platform. The Hyper-V clock counter is reset during
- suspend-resume and the offset used to measure time needs to be
- corrected, post resume.
- */
+static void hv_save_sched_clock_state(void) +{
- save_hv_clock_tsc_state();
- old_save_sched_clock_state();
+}
+static void hv_restore_sched_clock_state(void) +{
- restore_hv_clock_tsc_state();
- old_restore_sched_clock_state();
+}
#ifdef HAVE_VDSO_CLOCKMODE_HVCLOCK static int hv_cs_enable(struct clocksource *cs) { @@ -539,6 +596,11 @@ static void __init hv_init_tsc_clocksource(void)
hv_read_reference_counter = read_hv_clock_tsc;
- old_save_sched_clock_state = x86_platform.save_sched_clock_state;
- x86_platform.save_sched_clock_state = hv_save_sched_clock_state;
- old_restore_sched_clock_state = x86_platform.restore_sched_clock_state;
- x86_platform.restore_sched_clock_state = hv_restore_sched_clock_state;
This Hyper-V clocksource/timer driver has intentionally been kept instruction set architecture independent. See the comment at the top of the source code file. We've also avoided any "#ifdef x86" or similar, though it's OK to have #ifdef's on specific clock-related functionality like GENERIC_SCHED_CLOCK.
The reference to "x86_platform" violates the intended independence. The code to save-on-suspend and update-on-resume can probably stay in this module in generic form, but hooking the functions into the x86_platform function call mechanism should move to x86-specific code.
Michael
- /*
- TSC page mapping works differently in root compared to guest.
- In guest partition the guest PFN has to be passed to the
base-commit: da3ea35007d0af457a0afc87e84fddaebc4e0b63
2.25.1