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. This makes hv tsc page based sched_clock continuous and ensures
that post resume, it starts from where it left off during suspend.
Override x86_platform.save_sched_clock_state and
x86_platform.restore_sched_clock_state routines to correct this as soon
as possible.
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 e5313f1c54 ("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: 1349401ff1 ("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>
Reviewed-by: Michael Kelley <mhklinux@outlook.com>
Link: https://lore.kernel.org/r/20240917053917.76787-1-namjain@linux.microsoft.com
Signed-off-by: Wei Liu <wei.liu@kernel.org>
Message-ID: <20240917053917.76787-1-namjain@linux.microsoft.com>
In a TDX VM without paravisor, currently the default timer is the Hyper-V
timer, which depends on the slow VM Reference Counter MSR: the Hyper-V TSC
page is not enabled in such a VM because the VM uses Invariant TSC as a
better clocksource and it's challenging to mark the Hyper-V TSC page shared
in very early boot.
Lower the rating of the Hyper-V timer so the local APIC timer becomes the
the default timer in such a VM, and print a warning in case Invariant TSC
is unavailable in such a VM. This change should cause no perceivable
performance difference.
Cc: stable@vger.kernel.org # 6.6+
Reviewed-by: Roman Kisel <romank@linux.microsoft.com>
Signed-off-by: Dexuan Cui <decui@microsoft.com>
Reviewed-by: Michael Kelley <mhklinux@outlook.com>
Link: https://lore.kernel.org/r/20240621061614.8339-1-decui@microsoft.com
Signed-off-by: Wei Liu <wei.liu@kernel.org>
Message-ID: <20240621061614.8339-1-decui@microsoft.com>
The HV_REGISTER_ are used as arguments to hv_set/get_register(), which
delegate to arch-specific mechanisms for getting/setting synthetic
Hyper-V MSRs.
On arm64, HV_REGISTER_ defines are synthetic VP registers accessed via
the get/set vp registers hypercalls. The naming matches the TLFS
document, although these register names are not specific to arm64.
However, on x86 the prefix HV_REGISTER_ indicates Hyper-V MSRs accessed
via rdmsrl()/wrmsrl(). This is not consistent with the TLFS doc, where
HV_REGISTER_ is *only* used for used for VP register names used by
the get/set register hypercalls.
To fix this inconsistency and prevent future confusion, change the
arch-generic aliases used by callers of hv_set/get_register() to have
the prefix HV_MSR_ instead of HV_REGISTER_.
Use the prefix HV_X64_MSR_ for the x86-only Hyper-V MSRs. On x86, the
generic HV_MSR_'s point to the corresponding HV_X64_MSR_.
Move the arm64 HV_REGISTER_* defines to the asm-generic hyperv-tlfs.h,
since these are not specific to arm64. On arm64, the generic HV_MSR_'s
point to the corresponding HV_REGISTER_.
While at it, rename hv_get/set_registers() and related functions to
hv_get/set_msr(), hv_get/set_nested_msr(), etc. These are only used for
Hyper-V MSRs and this naming makes that clear.
Signed-off-by: Nuno Das Neves <nunodasneves@linux.microsoft.com>
Reviewed-by: Wei Liu <wei.liu@kernel.org>
Reviewed-by: Michael Kelley <mhklinux@outlook.com>
Link: https://lore.kernel.org/r/1708440933-27125-1-git-send-email-nunodasneves@linux.microsoft.com
Signed-off-by: Wei Liu <wei.liu@kernel.org>
Message-ID: <1708440933-27125-1-git-send-email-nunodasneves@linux.microsoft.com>
Pull scheduler updates from Ingo Molnar:
"Scheduler SMP load-balancer improvements:
- Avoid unnecessary migrations within SMT domains on hybrid systems.
Problem:
On hybrid CPU systems, (processors with a mixture of
higher-frequency SMT cores and lower-frequency non-SMT cores),
under the old code lower-priority CPUs pulled tasks from the
higher-priority cores if more than one SMT sibling was busy -
resulting in many unnecessary task migrations.
Solution:
The new code improves the load balancer to recognize SMT cores
with more than one busy sibling and allows lower-priority CPUs
to pull tasks, which avoids superfluous migrations and lets
lower-priority cores inspect all SMT siblings for the busiest
queue.
- Implement the 'runnable boosting' feature in the EAS balancer:
consider CPU contention in frequency, EAS max util & load-balance
busiest CPU selection.
This improves CPU utilization for certain workloads, while leaves
other key workloads unchanged.
Scheduler infrastructure improvements:
- Rewrite the scheduler topology setup code by consolidating it into
the build_sched_topology() helper function and building it
dynamically on the fly.
- Resolve the local_clock() vs. noinstr complications by rewriting
the code: provide separate sched_clock_noinstr() and
local_clock_noinstr() functions to be used in instrumentation code,
and make sure it is all instrumentation-safe.
Fixes:
- Fix a kthread_park() race with wait_woken()
- Fix misc wait_task_inactive() bugs unearthed by the -rt merge:
- Fix UP PREEMPT bug by unifying the SMP and UP implementations
- Fix task_struct::saved_state handling
- Fix various rq clock update bugs, unearthed by turning on the rq
clock debugging code.
- Fix the PSI WINDOW_MIN_US trigger limit, which was easy to trigger
by creating enough cgroups, by removing the warnign and restricting
window size triggers to PSI file write-permission or
CAP_SYS_RESOURCE.
- Propagate SMT flags in the topology when removing degenerate domain
- Fix grub_reclaim() calculation bug in the deadline scheduler code
- Avoid resetting the min update period when it is unnecessary, in
psi_trigger_destroy().
- Don't balance a task to its current running CPU in load_balance(),
which was possible on certain NUMA topologies with overlapping
groups.
- Fix the sched-debug printing of rq->nr_uninterruptible
Cleanups:
- Address various -Wmissing-prototype warnings, as a preparation to
(maybe) enable this warning in the future.
- Remove unused code
- Mark more functions __init
- Fix shadow-variable warnings"
* tag 'sched-core-2023-06-27' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (50 commits)
sched/core: Avoid multiple calling update_rq_clock() in __cfsb_csd_unthrottle()
sched/core: Avoid double calling update_rq_clock() in __balance_push_cpu_stop()
sched/core: Fixed missing rq clock update before calling set_rq_offline()
sched/deadline: Update GRUB description in the documentation
sched/deadline: Fix bandwidth reclaim equation in GRUB
sched/wait: Fix a kthread_park race with wait_woken()
sched/topology: Mark set_sched_topology() __init
sched/fair: Rename variable cpu_util eff_util
arm64/arch_timer: Fix MMIO byteswap
sched/fair, cpufreq: Introduce 'runnable boosting'
sched/fair: Refactor CPU utilization functions
cpuidle: Use local_clock_noinstr()
sched/clock: Provide local_clock_noinstr()
x86/tsc: Provide sched_clock_noinstr()
clocksource: hyper-v: Provide noinstr sched_clock()
clocksource: hyper-v: Adjust hv_read_tsc_page_tsc() to avoid special casing U64_MAX
x86/vdso: Fix gettimeofday masking
math64: Always inline u128 version of mul_u64_u64_shr()
s390/time: Provide sched_clock_noinstr()
loongarch: Provide noinstr sched_clock_read()
...
Current code assigns either the Hyper-V TSC page or MSR-based ref counter
as the sched clock. This may be sub-optimal in two cases. First, if there
is hardware support to ensure consistent TSC frequency across live
migrations and Hyper-V is using that support, the raw TSC is a faster
source of time than the Hyper-V TSC page. Second, the MSR-based ref
counter is relatively slow because reads require a trap to the hypervisor.
As such, it should never be used as the sched clock. The native sched
clock based on the raw TSC or jiffies is much better.
Rework the sched clock setup so it is set to the TSC page only if
Hyper-V indicates that the TSC may have inconsistent frequency across
live migrations. Also, remove the code that sets the sched clock to
the MSR-based ref counter. In the cases where it is not set, the sched
clock will then be the native sched clock.
As part of the rework, always enable both the TSC page clocksource and
the MSR-based ref counter clocksource. Set the ratings so the TSC page
clocksource is preferred. While the MSR-based ref counter clocksource
is unlikely to ever be the default, having it available for manual
selection is convenient for development purposes.
Signed-off-by: Michael Kelley <mikelley@microsoft.com>
Reviewed-by: Dexuan Cui <decui@microsoft.com>
Signed-off-by: Daniel Lezcano <daniel.lezcano@linaro.org>
Link: https://lore.kernel.org/r/1687201360-16003-1-git-send-email-mikelley@microsoft.com
Currently hv_read_tsc_page_tsc() (ab)uses the (valid) time value of
U64_MAX as an error return. This breaks the clean wrap-around of the
clock.
Modify the function signature to return a boolean state and provide
another u64 pointer to store the actual time on success. This obviates
the need to steal one time value and restores the full counter width.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Michael Kelley <mikelley@microsoft.com>
Tested-by: Michael Kelley <mikelley@microsoft.com> # Hyper-V
Link: https://lore.kernel.org/r/20230519102715.775630881@infradead.org
EXPORT_SYMBOL and __init is a bad combination because the .init.text
section is freed up after the initialization. Hence, modules cannot
use symbols annotated __init. The access to a freed symbol may end up
with kernel panic.
modpost used to detect it, but it has been broken for a decade.
Recently, I fixed modpost so it started to warn it again, then this
showed up in linux-next builds.
There are two ways to fix it:
- Remove __init
- Remove EXPORT_SYMBOL
I chose the latter for this case because the only in-tree call-site,
arch/x86/kernel/cpu/mshyperv.c is never compiled as modular.
(CONFIG_HYPERVISOR_GUEST is boolean)
Fixes: dd2cb34861 ("clocksource/drivers: Continue making Hyper-V clocksource ISA agnostic")
Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Reviewed-by: Michael Kelley <mikelley@microsoft.com>
Link: https://lore.kernel.org/r/20220606050238.4162200-1-masahiroy@kernel.org
Signed-off-by: Wei Liu <wei.liu@kernel.org>
Hyper-V clock/timer code in hyperv_timer.c is mostly independent from
other VMbus drivers, but building for ARM64 without hyperv_timer.c
shows some remaining entanglements. A default implementation of
hv_read_reference_counter can just read a Hyper-V synthetic register
and be independent of hyperv_timer.c, so move this code out and into
hv_common.c. Then it can be used by the timesync driver even if
hyperv_timer.c isn't built on a particular architecture. If
hyperv_timer.c *is* built, it can override with a faster implementation.
Also provide stubs for stimer functions called by the VMbus driver when
hyperv_timer.c isn't built.
No functional changes.
Signed-off-by: Michael Kelley <mikelley@microsoft.com>
Link: https://lore.kernel.org/r/1626220906-22629-1-git-send-email-mikelley@microsoft.com
Signed-off-by: Wei Liu <wei.liu@kernel.org>
Pull Hyper-V updates from Wei Liu:
- VMBus enhancement
- Free page reporting support for Hyper-V balloon driver
- Some patches for running Linux as Arm64 Hyper-V guest
- A few misc clean-up patches
* tag 'hyperv-next-signed-20210426' of git://git.kernel.org/pub/scm/linux/kernel/git/hyperv/linux: (30 commits)
drivers: hv: Create a consistent pattern for checking Hyper-V hypercall status
x86/hyperv: Move hv_do_rep_hypercall to asm-generic
video: hyperv_fb: Add ratelimit on error message
Drivers: hv: vmbus: Increase wait time for VMbus unload
Drivers: hv: vmbus: Initialize unload_event statically
Drivers: hv: vmbus: Check for pending channel interrupts before taking a CPU offline
Drivers: hv: vmbus: Drivers: hv: vmbus: Introduce CHANNELMSG_MODIFYCHANNEL_RESPONSE
Drivers: hv: vmbus: Introduce and negotiate VMBus protocol version 5.3
Drivers: hv: vmbus: Use after free in __vmbus_open()
Drivers: hv: vmbus: remove unused function
Drivers: hv: vmbus: Remove unused linux/version.h header
x86/hyperv: remove unused linux/version.h header
x86/Hyper-V: Support for free page reporting
x86/hyperv: Fix unused variable 'hi' warning in hv_apic_read
x86/hyperv: Fix unused variable 'msr_val' warning in hv_qlock_wait
hv: hyperv.h: a few mundane typo fixes
drivers: hv: Fix EXPORT_SYMBOL and tab spaces issue
Drivers: hv: vmbus: Drop error message when 'No request id available'
asm-generic/hyperv: Add missing function prototypes per -W1 warnings
clocksource/drivers/hyper-v: Move handling of STIMER0 interrupts
...
STIMER0 interrupts are most naturally modeled as per-cpu IRQs. But
because x86/x64 doesn't have per-cpu IRQs, the core STIMER0 interrupt
handling machinery is done in code under arch/x86 and Linux IRQs are
not used. Adding support for ARM64 means adding equivalent code
using per-cpu IRQs under arch/arm64.
A better model is to treat per-cpu IRQs as the normal path (which it is
for modern architectures), and the x86/x64 path as the exception. Do this
by incorporating standard Linux per-cpu IRQ allocation into the main
SITMER0 driver code, and bypass it in the x86/x64 exception case. For
x86/x64, special case code is retained under arch/x86, but no STIMER0
interrupt handling code is needed under arch/arm64.
No functional change.
Signed-off-by: Michael Kelley <mikelley@microsoft.com>
Acked-by: Daniel Lezcano <daniel.lezcano@linaro.org>
Link: https://lore.kernel.org/r/1614721102-2241-11-git-send-email-mikelley@microsoft.com
Signed-off-by: Wei Liu <wei.liu@kernel.org>
On x86/x64, the TSC clocksource is available in a Hyper-V VM only if
Hyper-V provides the TSC_INVARIANT flag. The rating on the Hyper-V
Reference TSC page clocksource is currently set so that it will not
override the TSC clocksource in this case. Alternatively, if the TSC
clocksource is not available, then the Hyper-V clocksource is used.
But on ARM64, the Hyper-V Reference TSC page clocksource should
override the ARM arch counter, since the Hyper-V clocksource provides
scaling and offsetting during live migrations that is not provided
for the ARM arch counter.
To get the needed behavior for both x86/x64 and ARM64, tweak the
logic by defaulting the Hyper-V Reference TSC page clocksource
rating to a large value that will always override. If the Hyper-V
TSC_INVARIANT flag is set, then reduce the rating so that it will not
override the TSC.
While the logic for getting there is slightly different, the net
result in the normal cases is no functional change.
Signed-off-by: Michael Kelley <mikelley@microsoft.com>
Acked-by: Daniel Lezcano <daniel.lezcano@linaro.org>
Link: https://lore.kernel.org/r/1614721102-2241-10-git-send-email-mikelley@microsoft.com
Signed-off-by: Wei Liu <wei.liu@kernel.org>
While the Hyper-V Reference TSC code is architecture neutral, the
pv_ops.time.sched_clock() function is implemented for x86/x64, but not
for ARM64. Current code calls a utility function under arch/x86 (and
coming, under arch/arm64) to handle the difference.
Change this approach to handle the difference inline based on whether
GENERIC_SCHED_CLOCK is present. The new approach removes code under
arch/* since the difference is tied more to the specifics of the Linux
implementation than to the architecture.
No functional change.
Signed-off-by: Michael Kelley <mikelley@microsoft.com>
Reviewed-by: Boqun Feng <boqun.feng@gmail.com>
Acked-by: Daniel Lezcano <daniel.lezcano@linaro.org>
Link: https://lore.kernel.org/r/1614721102-2241-9-git-send-email-mikelley@microsoft.com
Signed-off-by: Wei Liu <wei.liu@kernel.org>
While the driver for the Hyper-V Reference TSC and STIMERs is architecture
neutral, vDSO is implemented for x86/x64, but not for ARM64. Current code
calls into utility functions under arch/x86 (and coming, under arch/arm64)
to handle the difference.
Change this approach to handle the difference inline based on whether
VDSO_CLOCK_MODE_HVCLOCK is present. The new approach removes code under
arch/* since the difference is tied more to the specifics of the Linux
implementation than to the architecture.
No functional change.
Signed-off-by: Michael Kelley <mikelley@microsoft.com>
Reviewed-by: Boqun Feng <boqun.feng@gmail.com>
Acked-by: Daniel Lezcano <daniel.lezcano@linaro.org>
Link: https://lore.kernel.org/r/1614721102-2241-8-git-send-email-mikelley@microsoft.com
Signed-off-by: Wei Liu <wei.liu@kernel.org>
