RE: [PATCH] clocksource: hyper-v: Fix hv tsc page based sched_clock for hibernation

[Michael Kelley]

From: Naman Jain <namjain@xxxxxxxxxxxxxxxxxxx> 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
> 1) 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@xxxxxxxxxxxxxxx
> Fixes: 1349401ff1aa ("clocksource/drivers/hyper-v: Suspend/resume Hyper-V
> clocksource for hibernation")
> Co-developed-by: Dexuan Cui <decui@xxxxxxxxxxxxx>
> Signed-off-by: Dexuan Cui <decui@xxxxxxxxxxxxx>
> Signed-off-by: Naman Jain <namjain@xxxxxxxxxxxxxxxxxxx>
> ---
>  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