Age | Commit message (Collapse) | Author |
|
git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip
Pull x86 exception handling updates from Ingo Molnar:
- Clean up & simplify AP exception handling setup.
- Consolidate the disjoint IDT setup code living in idt_setup_traps()
and idt_setup_ist_traps() into a single idt_setup_traps()
initialization function and call it before cpu_init().
* tag 'x86-apic-2021-06-28' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
x86/idt: Rework IDT setup for boot CPU
x86/cpu: Init AP exception handling from cpu_init_secondary()
|
|
Signed-off-by: Ingo Molnar <mingo@kernel.org>
|
|
SEV-ES guests require properly setup task register with which the TSS
descriptor in the GDT can be located so that the IST-type #VC exception
handler which they need to function properly, can be executed.
This setup needs to happen before attempting to load microcode in
ucode_cpu_init() on secondary CPUs which can cause such #VC exceptions.
Simplify the machinery by running that exception setup from a new function
cpu_init_secondary() and explicitly call cpu_init_exception_handling() for
the boot CPU before cpu_init(). The latter prepares for fixing and
simplifying the exception/IST setup on the boot CPU.
There should be no functional changes resulting from this patch.
[ tglx: Reworked it so cpu_init_exception_handling() stays seperate ]
Signed-off-by: Borislav Petkov <bp@suse.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Lai Jiangshan <laijs@linux.alibaba.com>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/r/87k0o6gtvu.ffs@nanos.tec.linutronix.de
|
|
generations
Some AMD Ryzen generations has different calculation method on maximum
performance. 255 is not for all ASICs, some specific generations should use 166
as the maximum performance. Otherwise, it will report incorrect frequency value
like below:
~ → lscpu | grep MHz
CPU MHz: 3400.000
CPU max MHz: 7228.3198
CPU min MHz: 2200.0000
[ mingo: Tidied up whitespace use. ]
[ Alexander Monakov <amonakov@ispras.ru>: fix 225 -> 255 typo. ]
Fixes: 41ea667227ba ("x86, sched: Calculate frequency invariance for AMD systems")
Fixes: 3c55e94c0ade ("cpufreq: ACPI: Extend frequency tables to cover boost frequencies")
Reported-by: Jason Bagavatsingham <jason.bagavatsingham@gmail.com>
Fixed-by: Alexander Monakov <amonakov@ispras.ru>
Reviewed-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Signed-off-by: Huang Rui <ray.huang@amd.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Tested-by: Jason Bagavatsingham <jason.bagavatsingham@gmail.com>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/r/20210425073451.2557394-1-ray.huang@amd.com
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=211791
Signed-off-by: Ingo Molnar <mingo@kernel.org>
|
|
As pointed out by commit
de9b8f5dcbd9 ("sched: Fix crash trying to dequeue/enqueue the idle thread")
init_idle() can and will be invoked more than once on the same idle
task. At boot time, it is invoked for the boot CPU thread by
sched_init(). Then smp_init() creates the threads for all the secondary
CPUs and invokes init_idle() on them.
As the hotplug machinery brings the secondaries to life, it will issue
calls to idle_thread_get(), which itself invokes init_idle() yet again.
In this case it's invoked twice more per secondary: at _cpu_up(), and at
bringup_cpu().
Given smp_init() already initializes the idle tasks for all *possible*
CPUs, no further initialization should be required. Now, removing
init_idle() from idle_thread_get() exposes some interesting expectations
with regards to the idle task's preempt_count: the secondary startup always
issues a preempt_disable(), requiring some reset of the preempt count to 0
between hot-unplug and hotplug, which is currently served by
idle_thread_get() -> idle_init().
Given the idle task is supposed to have preemption disabled once and never
see it re-enabled, it seems that what we actually want is to initialize its
preempt_count to PREEMPT_DISABLED and leave it there. Do that, and remove
init_idle() from idle_thread_get().
Secondary startups were patched via coccinelle:
@begone@
@@
-preempt_disable();
...
cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
Signed-off-by: Valentin Schneider <valentin.schneider@arm.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Link: https://lore.kernel.org/r/20210512094636.2958515-1-valentin.schneider@arm.com
|
|
Signed-off-by: Wan Jiabing <wanjiabing@vivo.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/r/20210427063835.9039-1-wanjiabing@vivo.com
|
|
git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip
Pull x86 updates from Borislav Petkov:
- Turn the stack canary into a normal __percpu variable on 32-bit which
gets rid of the LAZY_GS stuff and a lot of code.
- Add an insn_decode() API which all users of the instruction decoder
should preferrably use. Its goal is to keep the details of the
instruction decoder away from its users and simplify and streamline
how one decodes insns in the kernel. Convert its users to it.
- kprobes improvements and fixes
- Set the maximum DIE per package variable on Hygon
- Rip out the dynamic NOP selection and simplify all the machinery
around selecting NOPs. Use the simplified NOPs in objtool now too.
- Add Xeon Sapphire Rapids to list of CPUs that support PPIN
- Simplify the retpolines by folding the entire thing into an
alternative now that objtool can handle alternatives with stack ops.
Then, have objtool rewrite the call to the retpoline with the
alternative which then will get patched at boot time.
- Document Intel uarch per models in intel-family.h
- Make Sub-NUMA Clustering topology the default and Cluster-on-Die the
exception on Intel.
* tag 'x86_core_for_v5.13' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (53 commits)
x86, sched: Treat Intel SNC topology as default, COD as exception
x86/cpu: Comment Skylake server stepping too
x86/cpu: Resort and comment Intel models
objtool/x86: Rewrite retpoline thunk calls
objtool: Skip magical retpoline .altinstr_replacement
objtool: Cache instruction relocs
objtool: Keep track of retpoline call sites
objtool: Add elf_create_undef_symbol()
objtool: Extract elf_symbol_add()
objtool: Extract elf_strtab_concat()
objtool: Create reloc sections implicitly
objtool: Add elf_create_reloc() helper
objtool: Rework the elf_rebuild_reloc_section() logic
objtool: Fix static_call list generation
objtool: Handle per arch retpoline naming
objtool: Correctly handle retpoline thunk calls
x86/retpoline: Simplify retpolines
x86/alternatives: Optimize optimize_nops()
x86: Add insn_decode_kernel()
x86/kprobes: Move 'inline' to the beginning of the kprobe_is_ss() declaration
...
|
|
git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip
Pull misc x86 cleanups from Borislav Petkov:
"Trivial cleanups and fixes all over the place"
* tag 'x86_cleanups_for_v5.13' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
MAINTAINERS: Remove me from IDE/ATAPI section
x86/pat: Do not compile stubbed functions when X86_PAT is off
x86/asm: Ensure asm/proto.h can be included stand-alone
x86/platform/intel/quark: Fix incorrect kernel-doc comment syntax in files
x86/msr: Make locally used functions static
x86/cacheinfo: Remove unneeded dead-store initialization
x86/process/64: Move cpu_current_top_of_stack out of TSS
tools/turbostat: Unmark non-kernel-doc comment
x86/syscalls: Fix -Wmissing-prototypes warnings from COND_SYSCALL()
x86/fpu/math-emu: Fix function cast warning
x86/msr: Fix wr/rdmsr_safe_regs_on_cpu() prototypes
x86: Fix various typos in comments, take #2
x86: Remove unusual Unicode characters from comments
x86/kaslr: Return boolean values from a function returning bool
x86: Fix various typos in comments
x86/setup: Remove unused RESERVE_BRK_ARRAY()
stacktrace: Move documentation for arch_stack_walk_reliable() to header
x86: Remove duplicate TSC DEADLINE MSR definitions
|
|
Commit 1340ccfa9a9a ("x86,sched: Allow topologies where NUMA nodes
share an LLC") added a vendor and model specific check to never
call topology_sane() for Intel Skylake Server systems where NUMA
nodes share an LLC.
Intel Ice Lake and Sapphire Rapids CPUs also enumerate an LLC that is
shared by multiple NUMA nodes. The LLC on these CPUs is shared for
off-package data access but private to the NUMA node for on-package
access. Rather than managing a list of allowable SNC topologies, make
this SNC topology the default, and treat Intel's Cluster-On-Die (COD)
topology as the exception.
In SNC mode, Sky Lake, Ice Lake, and Sapphire Rapids servers do not
emit this warning:
sched: CPU #3's llc-sibling CPU #0 is not on the same node! [node: 1 != 0]. Ignoring dependency.
Suggested-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Alison Schofield <alison.schofield@intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Dave Hansen <dave.hansen@linux.intel.com>
Cc: stable@vger.kernel.org
Link: https://lkml.kernel.org/r/20210310190233.31752-1-alison.schofield@intel.com
|
|
Commit 8cdddd182bd7 ("ACPI: processor: Fix CPU0 wakeup in
acpi_idle_play_dead()") tried to fix CPU0 hotplug breakage by copying
wakeup_cpu0() + start_cpu0() logic from hlt_play_dead()//mwait_play_dead()
into acpi_idle_play_dead(). The problem is that these functions are not
exported to modules so when CONFIG_ACPI_PROCESSOR=m build fails.
The issue could've been fixed by exporting both wakeup_cpu0()/start_cpu0()
(the later from assembly) but it seems putting the whole pattern into a
new function and exporting it instead is better.
Reported-by: kernel test robot <lkp@intel.com>
Fixes: 8cdddd182bd7 ("CPI: processor: Fix CPU0 wakeup in acpi_idle_play_dead()")
Cc: <stable@vger.kernel.org> # 5.10+
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
|
|
Commit 496121c02127 ("ACPI: processor: idle: Allow probing on platforms
with one ACPI C-state") broke CPU0 hotplug on certain systems, e.g.
I'm observing the following on AWS Nitro (e.g r5b.xlarge but other
instance types are affected as well):
# echo 0 > /sys/devices/system/cpu/cpu0/online
# echo 1 > /sys/devices/system/cpu/cpu0/online
<10 seconds delay>
-bash: echo: write error: Input/output error
In fact, the above mentioned commit only revealed the problem and did
not introduce it. On x86, to wakeup CPU an NMI is being used and
hlt_play_dead()/mwait_play_dead() loops are prepared to handle it:
/*
* If NMI wants to wake up CPU0, start CPU0.
*/
if (wakeup_cpu0())
start_cpu0();
cpuidle_play_dead() -> acpi_idle_play_dead() (which is now being called on
systems where it wasn't called before the above mentioned commit) serves
the same purpose but it doesn't have a path for CPU0. What happens now on
wakeup is:
- NMI is sent to CPU0
- wakeup_cpu0_nmi() works as expected
- we get back to while (1) loop in acpi_idle_play_dead()
- safe_halt() puts CPU0 to sleep again.
The straightforward/minimal fix is add the special handling for CPU0 on x86
and that's what the patch is doing.
Fixes: 496121c02127 ("ACPI: processor: idle: Allow probing on platforms with one ACPI C-state")
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Cc: 5.10+ <stable@vger.kernel.org> # 5.10+
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
|
|
Fix ~144 single-word typos in arch/x86/ code comments.
Doing this in a single commit should reduce the churn.
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Bjorn Helgaas <bhelgaas@google.com>
Cc: linux-kernel@vger.kernel.org
|
|
If the maximum performance level taken for computing the
arch_max_freq_ratio value used in the x86 scale-invariance code is
higher than the one corresponding to the cpuinfo.max_freq value
coming from the acpi_cpufreq driver, the scale-invariant utilization
falls below 100% even if the CPU runs at cpuinfo.max_freq or slightly
faster, which causes the schedutil governor to select a frequency
below cpuinfo.max_freq. That frequency corresponds to a frequency
table entry below the maximum performance level necessary to get to
the "boost" range of CPU frequencies which prevents "boost"
frequencies from being used in some workloads.
While this issue is related to scale-invariance, it may be amplified
by commit db865272d9c4 ("cpufreq: Avoid configuring old governors as
default with intel_pstate") from the 5.10 development cycle which
made it extremely easy to default to schedutil even if the preferred
driver is acpi_cpufreq as long as intel_pstate is built too, because
the mere presence of the latter effectively removes the ondemand
governor from the defaults. Distro kernels are likely to include
both intel_pstate and acpi_cpufreq on x86, so their users who cannot
use intel_pstate or choose to use acpi_cpufreq may easily be
affectecd by this issue.
If CPPC is available, it can be used to address this issue by
extending the frequency tables created by acpi_cpufreq to cover the
entire available frequency range (including "boost" frequencies) for
each CPU, but if CPPC is not there, acpi_cpufreq has no idea what
the maximum "boost" frequency is and the frequency tables created by
it cannot be extended in a meaningful way, so in that case make it
ask the arch scale-invariance code to to use the "nominal" performance
level for CPU utilization scaling in order to avoid the issue at hand.
Fixes: db865272d9c4 ("cpufreq: Avoid configuring old governors as default with intel_pstate")
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Reviewed-by: Giovanni Gherdovich <ggherdovich@suse.cz>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
|
|
On x86 scale invariace tends to be disabled during resume from
suspend-to-RAM, because the MPERF or APERF MSR values are not as
expected then due to updates taking place after the platform
firmware has been invoked to complete the suspend transition.
That, of course, is not desirable, especially if the schedutil
scaling governor is in use, because the lack of scale invariance
causes it to be less reliable.
To counter that effect, modify init_freq_invariance() to register
a syscore_ops object for scale invariance with the ->resume callback
pointing to init_counter_refs() which will run on the CPU starting
the resume transition (the other CPUs will be taken care of the
"online" operations taking place later).
Fixes: e2b0d619b400 ("x86, sched: check for counters overflow in frequency invariant accounting")
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Giovanni Gherdovich <ggherdovich@suse.cz>
Link: https://lkml.kernel.org/r/1803209.Mvru99baaF@kreacher
|
|
git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip
Pull x86 apic updates from Thomas Gleixner:
"Yet another large set of x86 interrupt management updates:
- Simplification and distangling of the MSI related functionality
- Let IO/APIC construct the RTE entries from an MSI message instead
of having IO/APIC specific code in the interrupt remapping drivers
- Make the retrieval of the parent interrupt domain (vector or remap
unit) less hardcoded and use the relevant irqdomain callbacks for
selection.
- Allow the handling of more than 255 CPUs without a virtualized
IOMMU when the hypervisor supports it. This has made been possible
by the above modifications and also simplifies the existing
workaround in the HyperV specific virtual IOMMU.
- Cleanup of the historical timer_works() irq flags related
inconsistencies"
* tag 'x86-apic-2020-12-14' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (42 commits)
x86/ioapic: Cleanup the timer_works() irqflags mess
iommu/hyper-v: Remove I/O-APIC ID check from hyperv_irq_remapping_select()
iommu/amd: Fix IOMMU interrupt generation in X2APIC mode
iommu/amd: Don't register interrupt remapping irqdomain when IR is disabled
iommu/amd: Fix union of bitfields in intcapxt support
x86/ioapic: Correct the PCI/ISA trigger type selection
x86/ioapic: Use I/O-APIC ID for finding irqdomain, not index
x86/hyperv: Enable 15-bit APIC ID if the hypervisor supports it
x86/kvm: Enable 15-bit extension when KVM_FEATURE_MSI_EXT_DEST_ID detected
iommu/hyper-v: Disable IRQ pseudo-remapping if 15 bit APIC IDs are available
x86/apic: Support 15 bits of APIC ID in MSI where available
x86/ioapic: Handle Extended Destination ID field in RTE
iommu/vt-d: Simplify intel_irq_remapping_select()
x86: Kill all traces of irq_remapping_get_irq_domain()
x86/ioapic: Use irq_find_matching_fwspec() to find remapping irqdomain
x86/hpet: Use irq_find_matching_fwspec() to find remapping irqdomain
iommu/hyper-v: Implement select() method on remapping irqdomain
iommu/vt-d: Implement select() method on remapping irqdomain
iommu/amd: Implement select() method on remapping irqdomain
x86/apic: Add select() method on vector irqdomain
...
|
|
git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip
Pull scheduler updates from Thomas Gleixner:
- migrate_disable/enable() support which originates from the RT tree
and is now a prerequisite for the new preemptible kmap_local() API
which aims to replace kmap_atomic().
- A fair amount of topology and NUMA related improvements
- Improvements for the frequency invariant calculations
- Enhanced robustness for the global CPU priority tracking and decision
making
- The usual small fixes and enhancements all over the place
* tag 'sched-core-2020-12-14' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (61 commits)
sched/fair: Trivial correction of the newidle_balance() comment
sched/fair: Clear SMT siblings after determining the core is not idle
sched: Fix kernel-doc markup
x86: Print ratio freq_max/freq_base used in frequency invariance calculations
x86, sched: Use midpoint of max_boost and max_P for frequency invariance on AMD EPYC
x86, sched: Calculate frequency invariance for AMD systems
irq_work: Optimize irq_work_single()
smp: Cleanup smp_call_function*()
irq_work: Cleanup
sched: Limit the amount of NUMA imbalance that can exist at fork time
sched/numa: Allow a floating imbalance between NUMA nodes
sched: Avoid unnecessary calculation of load imbalance at clone time
sched/numa: Rename nr_running and break out the magic number
sched: Make migrate_disable/enable() independent of RT
sched/topology: Condition EAS enablement on FIE support
arm64: Rebuild sched domains on invariance status changes
sched/topology,schedutil: Wrap sched domains rebuild
sched/uclamp: Allow to reset a task uclamp constraint value
sched/core: Fix typos in comments
Documentation: scheduler: fix information on arch SD flags, sched_domain and sched_debug
...
|
|
The value freq_max/freq_base is a fundamental component of frequency
invariance calculations. It may come from a variety of sources such as MSRs
or ACPI data, tracking it down when troubleshooting a system could be
non-trivial. It is worth saving it in the kernel logs.
# dmesg | grep 'Estimated ratio of average max'
[ 14.024036] smpboot: Estimated ratio of average max frequency by base frequency (times 1024): 1289
Signed-off-by: Giovanni Gherdovich <ggherdovich@suse.cz>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lkml.kernel.org/r/20201112182614.10700-4-ggherdovich@suse.cz
|
|
AMD EPYC
Frequency invariant accounting calculations need the ratio
freq_curr/freq_max, but freq_max is unknown as it depends on dynamic power
allocation between cores: AMD EPYC CPUs implement "Core Performance Boost".
Three candidates are considered to estimate this value:
- maximum non-boost frequency
- maximum boost frequency
- the mid point between the above two
Experimental data on an AMD EPYC Zen2 machine slightly favors the third
option, which is applied with this patch.
The analysis uses the ondemand cpufreq governor as baseline, and compares
it with schedutil in a number of configurations. Using the freq_max value
described above offers a moderate advantage in performance and efficiency:
sugov-max (freq_max=max_boost) performs the worst on tbench: less
throughput and reduced efficiency than the other invariant-schedutil
options (see "Data Overview" below). Consider that tbench is generally a
problematic case as no schedutil version currently is better than ondemand.
sugov-P0 (freq_max=max_P) is the worst on dbench, while the other sugov's
can surpass ondemand with less filesystem latency and slightly increased
efficiency.
1. DATA OVERVIEW
2. DETAILED PERFORMANCE TABLES
3. POWER CONSUMPTION TABLE
1. DATA OVERVIEW
================
sugov-noinv : non-invariant schedutil governor
sugov-max : invariant schedutil, freq_max=max_boost
sugov-mid : invariant schedutil, freq_max=midpoint
sugov-P0 : invariant schedutil, freq_max=max_P
perfgov : performance governor
driver : acpi_cpufreq
machine : AMD EPYC 7742 (Zen2, aka "Rome"), dual socket,
128 cores / 256 threads, SATA SSD storage, 250G of memory,
XFS filesystem
Benchmarks are described in the next section.
Tilde (~) means the value is the same as baseline.
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
ondemand perfgov sugov-noinv sugov-max sugov-mid sugov-P0 better if
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
PERFORMANCE RATIOS
tbench 1.00 1.44 0.90 0.87 0.93 0.93 higher
dbench 1.00 0.91 0.95 0.94 0.94 1.06 lower
kernbench 1.00 0.93 ~ ~ ~ 0.97 lower
gitsource 1.00 0.66 0.97 0.96 ~ 0.95 lower
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
PERFORMANCE-PER-WATT RATIOS
tbench 1.00 1.16 0.84 0.84 0.88 0.85 higher
dbench 1.00 1.03 1.02 1.02 1.02 0.93 higher
kernbench 1.00 1.05 ~ ~ ~ ~ higher
gitsource 1.00 1.46 1.04 1.04 ~ 1.05 higher
2. DETAILED PERFORMANCE TABLES
==============================
Benchmark : tbench4 (i.e. dbench4 over the network, actually loopback)
Varying parameter : number of clients
Unit : MB/sec (higher is better)
5.9.0-ondemand (BASELINE) 5.9.0-perfgov 5.9.0-sugov-noinv
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
Hmean 1 427.19 +- 0.16% ( ) 778.35 +- 0.10% ( 82.20%) 346.92 +- 0.14% ( -18.79%)
Hmean 2 853.82 +- 0.09% ( ) 1536.23 +- 0.03% ( 79.93%) 694.36 +- 0.05% ( -18.68%)
Hmean 4 1657.54 +- 0.12% ( ) 2938.18 +- 0.12% ( 77.26%) 1362.81 +- 0.11% ( -17.78%)
Hmean 8 3301.87 +- 0.06% ( ) 5679.10 +- 0.04% ( 72.00%) 2693.35 +- 0.04% ( -18.43%)
Hmean 16 6139.65 +- 0.05% ( ) 9498.81 +- 0.04% ( 54.71%) 4889.97 +- 0.17% ( -20.35%)
Hmean 32 11170.28 +- 0.09% ( ) 17393.25 +- 0.08% ( 55.71%) 9104.55 +- 0.09% ( -18.49%)
Hmean 64 19322.97 +- 0.17% ( ) 31573.91 +- 0.08% ( 63.40%) 18552.52 +- 0.40% ( -3.99%)
Hmean 128 30383.71 +- 0.11% ( ) 37416.91 +- 0.15% ( 23.15%) 25938.70 +- 0.41% ( -14.63%)
Hmean 256 31143.96 +- 0.41% ( ) 30908.76 +- 0.88% ( -0.76%) 29754.32 +- 0.24% ( -4.46%)
Hmean 512 30858.49 +- 0.26% ( ) 38524.60 +- 1.19% ( 24.84%) 42080.39 +- 0.56% ( 36.37%)
Hmean 1024 39187.37 +- 0.19% ( ) 36213.86 +- 0.26% ( -7.59%) 39555.98 +- 0.12% ( 0.94%)
5.9.0-sugov-max 5.9.0-sugov-mid 5.9.0-sugov-P0
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
Hmean 1 352.59 +- 1.03% ( -17.46%) 352.08 +- 0.75% ( -17.58%) 352.31 +- 1.48% ( -17.53%)
Hmean 2 697.32 +- 0.08% ( -18.33%) 700.16 +- 0.20% ( -18.00%) 696.79 +- 0.06% ( -18.39%)
Hmean 4 1369.88 +- 0.04% ( -17.35%) 1369.72 +- 0.07% ( -17.36%) 1365.91 +- 0.05% ( -17.59%)
Hmean 8 2696.79 +- 0.04% ( -18.33%) 2711.06 +- 0.04% ( -17.89%) 2715.10 +- 0.61% ( -17.77%)
Hmean 16 4725.03 +- 0.03% ( -23.04%) 4875.65 +- 0.02% ( -20.59%) 4953.05 +- 0.28% ( -19.33%)
Hmean 32 9231.65 +- 0.10% ( -17.36%) 8704.89 +- 0.27% ( -22.07%) 10562.02 +- 0.36% ( -5.45%)
Hmean 64 15364.27 +- 0.19% ( -20.49%) 17786.64 +- 0.15% ( -7.95%) 19665.40 +- 0.22% ( 1.77%)
Hmean 128 42100.58 +- 0.13% ( 38.56%) 34946.28 +- 0.13% ( 15.02%) 38635.79 +- 0.06% ( 27.16%)
Hmean 256 30660.23 +- 1.08% ( -1.55%) 32307.67 +- 0.54% ( 3.74%) 31153.27 +- 0.12% ( 0.03%)
Hmean 512 24604.32 +- 0.14% ( -20.27%) 40408.50 +- 1.10% ( 30.95%) 38800.29 +- 1.23% ( 25.74%)
Hmean 1024 35535.47 +- 0.28% ( -9.32%) 41070.38 +- 2.56% ( 4.81%) 31308.29 +- 2.52% ( -20.11%)
Benchmark : dbench (filesystem stressor)
Varying parameter : number of clients
Unit : seconds (lower is better)
NOTE-1: This dbench version measures the average latency of a set of filesystem
operations, as we found the traditional dbench metric (throughput) to be
misleading.
NOTE-2: Due to high variability, we partition the original dataset and apply
statistical bootrapping (a resampling method). Accuracy is reported in the
form of 95% confidence intervals.
5.9.0-ondemand (BASELINE) 5.9.0-perfgov 5.9.0-sugov-noinv
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
SubAmean 1 98.79 +- 0.92 ( ) 83.36 +- 0.82 ( 15.62%) 84.82 +- 0.92 ( 14.14%)
SubAmean 2 116.00 +- 0.89 ( ) 102.12 +- 0.77 ( 11.96%) 109.63 +- 0.89 ( 5.49%)
SubAmean 4 149.90 +- 1.03 ( ) 132.12 +- 0.91 ( 11.86%) 143.90 +- 1.15 ( 4.00%)
SubAmean 8 182.41 +- 1.13 ( ) 159.86 +- 0.93 ( 12.36%) 165.82 +- 1.03 ( 9.10%)
SubAmean 16 237.83 +- 1.23 ( ) 219.46 +- 1.14 ( 7.72%) 229.28 +- 1.19 ( 3.59%)
SubAmean 32 334.34 +- 1.49 ( ) 309.94 +- 1.42 ( 7.30%) 321.19 +- 1.36 ( 3.93%)
SubAmean 64 576.61 +- 2.16 ( ) 540.75 +- 2.00 ( 6.22%) 551.27 +- 1.99 ( 4.39%)
SubAmean 128 1350.07 +- 4.14 ( ) 1205.47 +- 3.20 ( 10.71%) 1280.26 +- 3.75 ( 5.17%)
SubAmean 256 3444.42 +- 7.97 ( ) 3698.00 +- 27.43 ( -7.36%) 3494.14 +- 7.81 ( -1.44%)
SubAmean 2048 39457.89 +- 29.01 ( ) 34105.33 +- 41.85 ( 13.57%) 39688.52 +- 36.26 ( -0.58%)
5.9.0-sugov-max 5.9.0-sugov-mid 5.9.0-sugov-P0
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
SubAmean 1 85.68 +- 1.04 ( 13.27%) 84.16 +- 0.84 ( 14.81%) 83.99 +- 0.90 ( 14.99%)
SubAmean 2 108.42 +- 0.95 ( 6.54%) 109.91 +- 1.39 ( 5.24%) 112.06 +- 0.91 ( 3.39%)
SubAmean 4 136.90 +- 1.04 ( 8.67%) 137.59 +- 0.93 ( 8.21%) 136.55 +- 0.95 ( 8.91%)
SubAmean 8 163.15 +- 0.96 ( 10.56%) 166.07 +- 1.02 ( 8.96%) 165.81 +- 0.99 ( 9.10%)
SubAmean 16 224.86 +- 1.12 ( 5.45%) 223.83 +- 1.06 ( 5.89%) 230.66 +- 1.19 ( 3.01%)
SubAmean 32 320.51 +- 1.38 ( 4.13%) 322.85 +- 1.49 ( 3.44%) 321.96 +- 1.46 ( 3.70%)
SubAmean 64 553.25 +- 1.93 ( 4.05%) 554.19 +- 2.08 ( 3.89%) 562.26 +- 2.22 ( 2.49%)
SubAmean 128 1264.35 +- 3.72 ( 6.35%) 1256.99 +- 3.46 ( 6.89%) 2018.97 +- 18.79 ( -49.55%)
SubAmean 256 3466.25 +- 8.25 ( -0.63%) 3450.58 +- 8.44 ( -0.18%) 5032.12 +- 38.74 ( -46.09%)
SubAmean 2048 39133.10 +- 45.71 ( 0.82%) 39905.95 +- 34.33 ( -1.14%) 53811.86 +-193.04 ( -36.38%)
Benchmark : kernbench (kernel compilation)
Varying parameter : number of jobs
Unit : seconds (lower is better)
5.9.0-ondemand (BASELINE) 5.9.0-perfgov 5.9.0-sugov-noinv
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
Amean 2 471.71 +- 26.61% ( ) 409.88 +- 16.99% ( 13.11%) 430.63 +- 0.18% ( 8.71%)
Amean 4 211.87 +- 0.58% ( ) 194.03 +- 0.74% ( 8.42%) 215.33 +- 0.64% ( -1.63%)
Amean 8 109.79 +- 1.27% ( ) 101.43 +- 1.53% ( 7.61%) 111.05 +- 1.95% ( -1.15%)
Amean 16 59.50 +- 1.28% ( ) 55.61 +- 1.35% ( 6.55%) 59.65 +- 1.78% ( -0.24%)
Amean 32 34.94 +- 1.22% ( ) 32.36 +- 1.95% ( 7.41%) 35.44 +- 0.63% ( -1.43%)
Amean 64 22.58 +- 0.38% ( ) 20.97 +- 1.28% ( 7.11%) 22.41 +- 1.73% ( 0.74%)
Amean 128 17.72 +- 0.44% ( ) 16.68 +- 0.32% ( 5.88%) 17.65 +- 0.96% ( 0.37%)
Amean 256 16.44 +- 0.53% ( ) 15.76 +- 0.32% ( 4.18%) 16.76 +- 0.60% ( -1.93%)
Amean 512 16.54 +- 0.21% ( ) 15.62 +- 0.41% ( 5.53%) 16.84 +- 0.85% ( -1.83%)
5.9.0-sugov-max 5.9.0-sugov-mid 5.9.0-sugov-P0
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
Amean 2 421.30 +- 0.24% ( 10.69%) 419.26 +- 0.15% ( 11.12%) 414.38 +- 0.33% ( 12.15%)
Amean 4 217.81 +- 5.53% ( -2.80%) 211.63 +- 0.99% ( 0.12%) 208.43 +- 0.47% ( 1.63%)
Amean 8 108.80 +- 0.43% ( 0.90%) 108.48 +- 1.44% ( 1.19%) 108.59 +- 3.08% ( 1.09%)
Amean 16 58.84 +- 0.74% ( 1.12%) 58.37 +- 0.94% ( 1.91%) 57.78 +- 0.78% ( 2.90%)
Amean 32 34.04 +- 2.00% ( 2.59%) 34.28 +- 1.18% ( 1.91%) 33.98 +- 2.21% ( 2.75%)
Amean 64 22.22 +- 1.69% ( 1.60%) 22.27 +- 1.60% ( 1.38%) 22.25 +- 1.41% ( 1.47%)
Amean 128 17.55 +- 0.24% ( 0.97%) 17.53 +- 0.94% ( 1.04%) 17.49 +- 0.43% ( 1.30%)
Amean 256 16.51 +- 0.46% ( -0.40%) 16.48 +- 0.48% ( -0.19%) 16.44 +- 1.21% ( 0.00%)
Amean 512 16.50 +- 0.35% ( 0.19%) 16.35 +- 0.42% ( 1.14%) 16.37 +- 0.33% ( 0.99%)
Benchmark : gitsource (time to run the git unit test suite)
Varying parameter : none
Unit : seconds (lower is better)
5.9.0-ondemand (BASELINE) 5.9.0-perfgov 5.9.0-sugov-noinv
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
Amean 1035.76 +- 0.30% ( ) 688.21 +- 0.04% ( 33.56%) 1003.85 +- 0.14% ( 3.08%)
5.9.0-sugov-max 5.9.0-sugov-mid 5.9.0-sugov-P0
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
Amean 995.82 +- 0.08% ( 3.86%) 1011.98 +- 0.03% ( 2.30%) 986.87 +- 0.19% ( 4.72%)
3. POWER CONSUMPTION TABLE
==========================
Average power consumption (watts).
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
ondemand perfgov sugov-noinv sugov-max sugov-mid sugov-P0
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
tbench4 227.25 281.83 244.17 236.76 241.50 247.99
dbench4 151.97 161.87 157.08 158.10 158.06 153.73
kernbench 162.78 167.22 162.90 164.19 164.65 164.72
gitsource 133.65 139.00 133.04 134.43 134.18 134.32
Signed-off-by: Giovanni Gherdovich <ggherdovich@suse.cz>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lkml.kernel.org/r/20201112182614.10700-3-ggherdovich@suse.cz
|
|
This is the first pass in creating the ability to calculate the
frequency invariance on AMD systems. This approach uses the CPPC
highest performance and nominal performance values that range from
0 - 255 instead of a high and base frquency. This is because we do
not have the ability on AMD to get a highest frequency value.
On AMD systems the highest performance and nominal performance
vaues do correspond to the highest and base frequencies for the system
so using them should produce an appropriate ratio but some tweaking
is likely necessary.
Due to CPPC being initialized later in boot than when the frequency
invariant calculation is currently made, I had to create a callback
from the CPPC init code to do the calculation after we have CPPC
data.
Special thanks to "kernel test robot <lkp@intel.com>" for reporting that
compilation of drivers/acpi/cppc_acpi.c is conditional to
CONFIG_ACPI_CPPC_LIB, not just CONFIG_ACPI.
[ ggherdovich@suse.cz: made safe under CPU hotplug, edited changelog. ]
Signed-off-by: Nathan Fontenot <nathan.fontenot@amd.com>
Signed-off-by: Giovanni Gherdovich <ggherdovich@suse.cz>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lkml.kernel.org/r/20201112182614.10700-2-ggherdovich@suse.cz
|
|
The call to rcu_cpu_starting() in mtrr_ap_init() is not early enough
in the CPU-hotplug onlining process, which results in lockdep splats
as follows:
=============================
WARNING: suspicious RCU usage
5.9.0+ #268 Not tainted
-----------------------------
kernel/kprobes.c:300 RCU-list traversed in non-reader section!!
other info that might help us debug this:
RCU used illegally from offline CPU!
rcu_scheduler_active = 1, debug_locks = 1
no locks held by swapper/1/0.
stack backtrace:
CPU: 1 PID: 0 Comm: swapper/1 Not tainted 5.9.0+ #268
Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.10.2-1ubuntu1 04/01/2014
Call Trace:
dump_stack+0x77/0x97
__is_insn_slot_addr+0x15d/0x170
kernel_text_address+0xba/0xe0
? get_stack_info+0x22/0xa0
__kernel_text_address+0x9/0x30
show_trace_log_lvl+0x17d/0x380
? dump_stack+0x77/0x97
dump_stack+0x77/0x97
__lock_acquire+0xdf7/0x1bf0
lock_acquire+0x258/0x3d0
? vprintk_emit+0x6d/0x2c0
_raw_spin_lock+0x27/0x40
? vprintk_emit+0x6d/0x2c0
vprintk_emit+0x6d/0x2c0
printk+0x4d/0x69
start_secondary+0x1c/0x100
secondary_startup_64_no_verify+0xb8/0xbb
This is avoided by moving the call to rcu_cpu_starting up near
the beginning of the start_secondary() function. Note that the
raw_smp_processor_id() is required in order to avoid calling into lockdep
before RCU has declared the CPU to be watched for readers.
Link: https://lore.kernel.org/lkml/160223032121.7002.1269740091547117869.tip-bot2@tip-bot2/
Reported-by: Qian Cai <cai@redhat.com>
Suggested-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
|
|
apic::irq_dest_mode is actually a boolean, but defined as u32 and named in
a way which does not explain what it means.
Make it a boolean and rename it to 'dest_mode_logical'
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/r/20201024213535.443185-9-dwmw2@infradead.org
|
|
struct apic has two members which store information about the destination
mode: dest_logical and irq_dest_mode.
dest_logical contains a mask which was historically used to set the
destination mode in IPI messages. Over time the usage was reduced and the
logical/physical functions were seperated.
There are only a few places which still use 'dest_logical' but they can
use 'irq_dest_mode' instead.
irq_dest_mode is actually a boolean where 0 means physical destination mode
and 1 means logical destination mode. Of course the name does not reflect
the functionality. This will be cleaned up in a subsequent change.
Remove apic::dest_logical and fixup the remaining users.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/r/20201024213535.443185-8-dwmw2@infradead.org
|
|
The IDT on 64-bit contains vectors which use paranoid_entry() and/or IST
stacks. To make these vectors work, the TSS and the getcpu GDT entry need
to be set up before the IDT is loaded.
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Signed-off-by: Borislav Petkov <bp@suse.de>
Link: https://lkml.kernel.org/r/20200907131613.12703-68-joro@8bytes.org
|
|
There is a race when taking a CPU offline. Current code looks like this:
native_cpu_disable()
{
...
apic_soft_disable();
/*
* Any existing set bits for pending interrupt to
* this CPU are preserved and will be sent via IPI
* to another CPU by fixup_irqs().
*/
cpu_disable_common();
{
....
/*
* Race window happens here. Once local APIC has been
* disabled any new interrupts from the device to
* the old CPU are lost
*/
fixup_irqs(); // Too late to capture anything in IRR.
...
}
}
The fix is to disable the APIC *after* cpu_disable_common().
Testing was done with a USB NIC that provided a source of frequent
interrupts. A script migrated interrupts to a specific CPU and
then took that CPU offline.
Fixes: 60dcaad5736f ("x86/hotplug: Silence APIC and NMI when CPU is dead")
Reported-by: Evan Green <evgreen@chromium.org>
Signed-off-by: Ashok Raj <ashok.raj@intel.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Tested-by: Evan Green <evgreen@chromium.org>
Reviewed-by: Evan Green <evgreen@chromium.org>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/lkml/875zdarr4h.fsf@nanos.tec.linutronix.de/
Link: https://lore.kernel.org/r/1598501530-45821-1-git-send-email-ashok.raj@intel.com
|
|
git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip
Pull x86 cpu updates from Ingo Molar:
- prepare for Intel's new SERIALIZE instruction
- enable split-lock debugging on more CPUs
- add more Intel CPU models
- optimize stack canary initialization a bit
- simplify the Spectre logic a bit
* tag 'x86-cpu-2020-08-03' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
x86/cpu: Refactor sync_core() for readability
x86/cpu: Relocate sync_core() to sync_core.h
x86/cpufeatures: Add enumeration for SERIALIZE instruction
x86/split_lock: Enable the split lock feature on Sapphire Rapids and Alder Lake CPUs
x86/cpu: Add Lakefield, Alder Lake and Rocket Lake models to the to Intel CPU family
x86/stackprotector: Pre-initialize canary for secondary CPUs
x86/speculation: Merge one test in spectre_v2_user_select_mitigation()
|
|
The idle tasks created for each secondary CPU already have a random stack
canary generated by fork(). Copy the canary to the percpu variable before
starting the secondary CPU which removes the need to call
boot_init_stack_canary().
Signed-off-by: Brian Gerst <brgerst@gmail.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Link: https://lkml.kernel.org/r/20200617225624.799335-1-brgerst@gmail.com
|
|
Be defensive against the case where the processor reports a base_freq
larger than turbo_freq (the ratio would be zero).
Fixes: 1567c3e3467c ("x86, sched: Add support for frequency invariance")
Signed-off-by: Giovanni Gherdovich <ggherdovich@suse.cz>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Link: https://lkml.kernel.org/r/20200531182453.15254-4-ggherdovich@suse.cz
|
|
There may be CPUs that support turbo boost but don't declare any turbo
ratio, i.e. their MSR_TURBO_RATIO_LIMIT is all zeroes. In that condition
scale-invariant calculations can't be performed.
Fixes: 1567c3e3467c ("x86, sched: Add support for frequency invariance")
Suggested-by: Ricardo Neri <ricardo.neri-calderon@linux.intel.com>
Signed-off-by: Giovanni Gherdovich <ggherdovich@suse.cz>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Tested-by: Ricardo Neri <ricardo.neri-calderon@linux.intel.com>
Link: https://lkml.kernel.org/r/20200531182453.15254-3-ggherdovich@suse.cz
|
|
The product mcnt * arch_max_freq_ratio can overflows u64.
For context, a large value for arch_max_freq_ratio would be 5000,
corresponding to a turbo_freq/base_freq ratio of 5 (normally it's more like
1500-2000). A large increment frequency for the MPERF counter would be 5GHz
(the base clock of all CPUs on the market today is less than that). With
these figures, a CPU would need to go without a scheduler tick for around 8
days for the u64 overflow to happen. It is unlikely, but the check is
warranted.
Under similar conditions, the difference acnt of two consecutive APERF
readings can overflow as well.
In these circumstances is appropriate to disable frequency invariant
accounting: the feature relies on measures of the clock frequency done at
every scheduler tick, which need to be "fresh" to be at all meaningful.
A note on i386: prior to version 5.1, the GCC compiler didn't have the
builtin function __builtin_mul_overflow. In these GCC versions the macro
check_mul_overflow needs __udivdi3() to do (u64)a/b, which the kernel
doesn't provide. For this reason this change fails to build on i386 if
GCC<5.1, and we protect the entire frequency invariant code behind
CONFIG_X86_64 (special thanks to "kbuild test robot" <lkp@intel.com>).
Fixes: 1567c3e3467c ("x86, sched: Add support for frequency invariance")
Signed-off-by: Giovanni Gherdovich <ggherdovich@suse.cz>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Link: https://lkml.kernel.org/r/20200531182453.15254-2-ggherdovich@suse.cz
|
|
The replacement of <asm/pgrable.h> with <linux/pgtable.h> made the include
of the latter in the middle of asm includes. Fix this up with the aid of
the below script and manual adjustments here and there.
import sys
import re
if len(sys.argv) is not 3:
print "USAGE: %s <file> <header>" % (sys.argv[0])
sys.exit(1)
hdr_to_move="#include <linux/%s>" % sys.argv[2]
moved = False
in_hdrs = False
with open(sys.argv[1], "r") as f:
lines = f.readlines()
for _line in lines:
line = _line.rstrip('
')
if line == hdr_to_move:
continue
if line.startswith("#include <linux/"):
in_hdrs = True
elif not moved and in_hdrs:
moved = True
print hdr_to_move
print line
Signed-off-by: Mike Rapoport <rppt@linux.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Cain <bcain@codeaurora.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Chris Zankel <chris@zankel.net>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Greentime Hu <green.hu@gmail.com>
Cc: Greg Ungerer <gerg@linux-m68k.org>
Cc: Guan Xuetao <gxt@pku.edu.cn>
Cc: Guo Ren <guoren@kernel.org>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Helge Deller <deller@gmx.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Ley Foon Tan <ley.foon.tan@intel.com>
Cc: Mark Salter <msalter@redhat.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Matt Turner <mattst88@gmail.com>
Cc: Max Filippov <jcmvbkbc@gmail.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Michal Simek <monstr@monstr.eu>
Cc: Nick Hu <nickhu@andestech.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Richard Weinberger <richard@nod.at>
Cc: Rich Felker <dalias@libc.org>
Cc: Russell King <linux@armlinux.org.uk>
Cc: Stafford Horne <shorne@gmail.com>
Cc: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Vincent Chen <deanbo422@gmail.com>
Cc: Vineet Gupta <vgupta@synopsys.com>
Cc: Will Deacon <will@kernel.org>
Cc: Yoshinori Sato <ysato@users.sourceforge.jp>
Link: http://lkml.kernel.org/r/20200514170327.31389-4-rppt@kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
The include/linux/pgtable.h is going to be the home of generic page table
manipulation functions.
Start with moving asm-generic/pgtable.h to include/linux/pgtable.h and
make the latter include asm/pgtable.h.
Signed-off-by: Mike Rapoport <rppt@linux.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Cain <bcain@codeaurora.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Chris Zankel <chris@zankel.net>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Greentime Hu <green.hu@gmail.com>
Cc: Greg Ungerer <gerg@linux-m68k.org>
Cc: Guan Xuetao <gxt@pku.edu.cn>
Cc: Guo Ren <guoren@kernel.org>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Helge Deller <deller@gmx.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Ley Foon Tan <ley.foon.tan@intel.com>
Cc: Mark Salter <msalter@redhat.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Matt Turner <mattst88@gmail.com>
Cc: Max Filippov <jcmvbkbc@gmail.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Michal Simek <monstr@monstr.eu>
Cc: Nick Hu <nickhu@andestech.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Richard Weinberger <richard@nod.at>
Cc: Rich Felker <dalias@libc.org>
Cc: Russell King <linux@armlinux.org.uk>
Cc: Stafford Horne <shorne@gmail.com>
Cc: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Vincent Chen <deanbo422@gmail.com>
Cc: Vineet Gupta <vgupta@synopsys.com>
Cc: Will Deacon <will@kernel.org>
Cc: Yoshinori Sato <ysato@users.sourceforge.jp>
Link: http://lkml.kernel.org/r/20200514170327.31389-3-rppt@kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip
Pull x86 cleanups from Ingo Molnar:
"Misc cleanups, with an emphasis on removing obsolete/dead code"
* tag 'x86-cleanups-2020-06-01' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
x86/spinlock: Remove obsolete ticket spinlock macros and types
x86/mm: Drop deprecated DISCONTIGMEM support for 32-bit
x86/apb_timer: Drop unused declaration and macro
x86/apb_timer: Drop unused TSC calibration
x86/io_apic: Remove unused function mp_init_irq_at_boot()
x86/mm: Stop printing BRK addresses
x86/audit: Fix a -Wmissing-prototypes warning for ia32_classify_syscall()
x86/nmi: Remove edac.h include leftover
mm: Remove MPX leftovers
x86/mm/mmap: Fix -Wmissing-prototypes warnings
x86/early_printk: Remove unused includes
crash_dump: Remove no longer used saved_max_pfn
x86/smpboot: Remove the last ICPU() macro
|
|
git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip
Pull SMP updates from Ingo Molnar:
"Misc cleanups in the SMP hotplug and cross-call code"
* tag 'smp-core-2020-06-01' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
cpu/hotplug: Remove __freeze_secondary_cpus()
cpu/hotplug: Remove disable_nonboot_cpus()
cpu/hotplug: Fix a typo in comment "broadacasted"->"broadcasted"
smp: Use smp_call_func_t in on_each_cpu()
|
|
... or the odyssey of trying to disable the stack protector for the
function which generates the stack canary value.
The whole story started with Sergei reporting a boot crash with a kernel
built with gcc-10:
Kernel panic — not syncing: stack-protector: Kernel stack is corrupted in: start_secondary
CPU: 1 PID: 0 Comm: swapper/1 Not tainted 5.6.0-rc5—00235—gfffb08b37df9 #139
Hardware name: Gigabyte Technology Co., Ltd. To be filled by O.E.M./H77M—D3H, BIOS F12 11/14/2013
Call Trace:
dump_stack
panic
? start_secondary
__stack_chk_fail
start_secondary
secondary_startup_64
-—-[ end Kernel panic — not syncing: stack—protector: Kernel stack is corrupted in: start_secondary
This happens because gcc-10 tail-call optimizes the last function call
in start_secondary() - cpu_startup_entry() - and thus emits a stack
canary check which fails because the canary value changes after the
boot_init_stack_canary() call.
To fix that, the initial attempt was to mark the one function which
generates the stack canary with:
__attribute__((optimize("-fno-stack-protector"))) ... start_secondary(void *unused)
however, using the optimize attribute doesn't work cumulatively
as the attribute does not add to but rather replaces previously
supplied optimization options - roughly all -fxxx options.
The key one among them being -fno-omit-frame-pointer and thus leading to
not present frame pointer - frame pointer which the kernel needs.
The next attempt to prevent compilers from tail-call optimizing
the last function call cpu_startup_entry(), shy of carving out
start_secondary() into a separate compilation unit and building it with
-fno-stack-protector, was to add an empty asm("").
This current solution was short and sweet, and reportedly, is supported
by both compilers but we didn't get very far this time: future (LTO?)
optimization passes could potentially eliminate this, which leads us
to the third attempt: having an actual memory barrier there which the
compiler cannot ignore or move around etc.
That should hold for a long time, but hey we said that about the other
two solutions too so...
Reported-by: Sergei Trofimovich <slyfox@gentoo.org>
Signed-off-by: Borislav Petkov <bp@suse.de>
Tested-by: Kalle Valo <kvalo@codeaurora.org>
Cc: <stable@vger.kernel.org>
Link: https://lkml.kernel.org/r/20200314164451.346497-1-slyfox@gentoo.org
|
|
The single user could have called freeze_secondary_cpus() directly.
Since this function was a source of confusion, remove it as it's
just a pointless wrapper.
While at it, rename enable_nonboot_cpus() to thaw_secondary_cpus() to
preserve the naming symmetry.
Done automatically via:
git grep -l enable_nonboot_cpus | xargs sed -i 's/enable_nonboot_cpus/thaw_secondary_cpus/g'
Signed-off-by: Qais Yousef <qais.yousef@arm.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net>
Link: https://lkml.kernel.org/r/20200430114004.17477-1-qais.yousef@arm.com
|
|
Improve readability of the function intel_set_max_freq_ratio() by moving
the check for KNL CPUs there, together with checks for GLM and SKX.
Signed-off-by: Giovanni Gherdovich <ggherdovich@suse.cz>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Link: https://lkml.kernel.org/r/20200416054745.740-5-ggherdovich@suse.cz
|
|
The static key arch_scale_freq_key only needs to be enabled once (at
boot). This change fixes a bug by which the key was enabled every time cpu0
is started, even as a secondary CPU during cpu hotplug. Secondary CPUs are
started from the idle thread: setting a static key from there means
acquiring a lock and may result in sleeping in the idle task, causing CPU
lockup.
Another consequence of this change is that init_counter_refs() is now
called on each CPU correctly; previously the function on_each_cpu() was
used, but it was called at boot when the only online cpu is cpu0.
[ggherdovich@suse.cz: Tested and wrote changelog]
Fixes: 1567c3e3467c ("x86, sched: Add support for frequency invariance")
Reported-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Giovanni Gherdovich <ggherdovich@suse.cz>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Link: https://lkml.kernel.org/r/20200416054745.740-4-ggherdovich@suse.cz
|
|
If a CPU has less than 4 physical cores, MSR_TURBO_RATIO_LIMIT will
rightfully report that the 4C turbo ratio is zero. In such cases, use the
1C turbo ratio instead for frequency invariance calculations.
Fixes: 1567c3e3467c ("x86, sched: Add support for frequency invariance")
Reported-by: Like Xu <like.xu@linux.intel.com>
Reported-by: Neil Rickert <nwr10cst-oslnx@yahoo.com>
Signed-off-by: Giovanni Gherdovich <ggherdovich@suse.cz>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Tested-by: Dave Kleikamp <dave.kleikamp@oracle.com>
Link: https://lkml.kernel.org/r/20200416054745.740-3-ggherdovich@suse.cz
|
|
Some hypervisors such as VMWare ESXi 5.5 advertise support for
X86_FEATURE_APERFMPERF but then fill all MSR's with zeroes. In particular,
MSR_PLATFORM_INFO set to zero tricks the code that wants to know the base
clock frequency of the CPU (highest non-turbo frequency), producing a
division by zero when computing the ratio turbo_freq/base_freq necessary
for frequency invariant accounting.
It is to be noted that even if MSR_PLATFORM_INFO contained the appropriate
data, APERF and MPERF are constantly zero on ESXi 5.5, thus freq-invariance
couldn't be done in principle (not that it would make a lot of sense in a
VM anyway). The real problem is advertising X86_FEATURE_APERFMPERF. This
appears to be fixed in more recent versions: ESXi 6.7 doesn't advertise
that feature.
Fixes: 1567c3e3467c ("x86, sched: Add support for frequency invariance")
Signed-off-by: Giovanni Gherdovich <ggherdovich@suse.cz>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Link: https://lkml.kernel.org/r/20200416054745.740-2-ggherdovich@suse.cz
|
|
Now all is using the shiny new macros.
No code changed:
# arch/x86/kernel/smpboot.o:
text data bss dec hex filename
16432 2649 40 19121 4ab1 smpboot.o.before
16432 2649 40 19121 4ab1 smpboot.o.after
md5:
a58104003b72c1de533095bc5a4c30a9 smpboot.o.before.asm
a58104003b72c1de533095bc5a4c30a9 smpboot.o.after.asm
Signed-off-by: Borislav Petkov <bp@suse.de>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: https://lkml.kernel.org/r/20200324185836.GI22931@zn.tnic
|
|
git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip
Pull x86 cleanups from Ingo Molnar:
"This topic tree contains more commits than usual:
- most of it are uaccess cleanups/reorganization by Al
- there's a bunch of prototype declaration (--Wmissing-prototypes)
cleanups
- misc other cleanups all around the map"
* 'x86-cleanups-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (36 commits)
x86/mm/set_memory: Fix -Wmissing-prototypes warnings
x86/efi: Add a prototype for efi_arch_mem_reserve()
x86/mm: Mark setup_emu2phys_nid() static
x86/jump_label: Move 'inline' keyword placement
x86/platform/uv: Add a missing prototype for uv_bau_message_interrupt()
kill uaccess_try()
x86: unsafe_put-style macro for sigmask
x86: x32_setup_rt_frame(): consolidate uaccess areas
x86: __setup_rt_frame(): consolidate uaccess areas
x86: __setup_frame(): consolidate uaccess areas
x86: setup_sigcontext(): list user_access_{begin,end}() into callers
x86: get rid of put_user_try in __setup_rt_frame() (both 32bit and 64bit)
x86: ia32_setup_rt_frame(): consolidate uaccess areas
x86: ia32_setup_frame(): consolidate uaccess areas
x86: ia32_setup_sigcontext(): lift user_access_{begin,end}() into the callers
x86/alternatives: Mark text_poke_loc_init() static
x86/cpu: Fix a -Wmissing-prototypes warning for init_ia32_feat_ctl()
x86/mm: Drop pud_mknotpresent()
x86: Replace setup_irq() by request_irq()
x86/configs: Slightly reduce defconfigs
...
|
|
git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip
Pull scheduler updates from Ingo Molnar:
"The main changes in this cycle are:
- Various NUMA scheduling updates: harmonize the load-balancer and
NUMA placement logic to not work against each other. The intended
result is better locality, better utilization and fewer migrations.
- Introduce Thermal Pressure tracking and optimizations, to improve
task placement on thermally overloaded systems.
- Implement frequency invariant scheduler accounting on (some) x86
CPUs. This is done by observing and sampling the 'recent' CPU
frequency average at ~tick boundaries. The CPU provides this data
via the APERF/MPERF MSRs. This hopefully makes our capacity
estimates more precise and keeps tasks on the same CPU better even
if it might seem overloaded at a lower momentary frequency. (As
usual, turbo mode is a complication that we resolve by observing
the maximum frequency and renormalizing to it.)
- Add asymmetric CPU capacity wakeup scan to improve capacity
utilization on asymmetric topologies. (big.LITTLE systems)
- PSI fixes and optimizations.
- RT scheduling capacity awareness fixes & improvements.
- Optimize the CONFIG_RT_GROUP_SCHED constraints code.
- Misc fixes, cleanups and optimizations - see the changelog for
details"
* 'sched-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (62 commits)
threads: Update PID limit comment according to futex UAPI change
sched/fair: Fix condition of avg_load calculation
sched/rt: cpupri_find: Trigger a full search as fallback
kthread: Do not preempt current task if it is going to call schedule()
sched/fair: Improve spreading of utilization
sched: Avoid scale real weight down to zero
psi: Move PF_MEMSTALL out of task->flags
MAINTAINERS: Add maintenance information for psi
psi: Optimize switching tasks inside shared cgroups
psi: Fix cpu.pressure for cpu.max and competing cgroups
sched/core: Distribute tasks within affinity masks
sched/fair: Fix enqueue_task_fair warning
thermal/cpu-cooling, sched/core: Move the arch_set_thermal_pressure() API to generic scheduler code
sched/rt: Remove unnecessary push for unfit tasks
sched/rt: Allow pulling unfitting task
sched/rt: Optimize cpupri_find() on non-heterogenous systems
sched/rt: Re-instate old behavior in select_task_rq_rt()
sched/rt: cpupri_find: Implement fallback mechanism for !fit case
sched/fair: Fix reordering of enqueue/dequeue_task_fair()
sched/fair: Fix runnable_avg for throttled cfs
...
|
|
The new macro set has a consistent namespace and uses C99 initializers
instead of the grufty C89 ones.
Get rid the of the local macro wrappers for consistency.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Borislav Petkov <bp@suse.de>
Reviewed-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Link: https://lkml.kernel.org/r/20200320131509.250559388@linutronix.de
|
|
Fix a couple of typos in code comments.
[ bp: While at it: s/IRQ's/IRQs/. ]
Signed-off-by: Martin Molnar <martin.molnar.programming@gmail.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Reviewed-by: Randy Dunlap <rdunlap@infradead.org>
Link: https://lkml.kernel.org/r/0819a044-c360-44a4-f0b6-3f5bafe2d35c@gmail.com
|
|
invariance
On some platforms such as the Dell XPS 13 laptop the firmware disables turbo
when the machine is disconnected from AC, and viceversa it enables it again
when it's reconnected. In these cases a _PPC ACPI notification is issued.
The scheduler needs to know freq_max for frequency-invariant calculations.
To account for turbo availability to come and go, record freq_max at boot as
if turbo was available and store it in a helper variable. Use a setter
function to swap between freq_base and freq_max every time turbo goes off or on.
Signed-off-by: Giovanni Gherdovich <ggherdovich@suse.cz>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Link: https://lkml.kernel.org/r/20200122151617.531-7-ggherdovich@suse.cz
|
|
The scheduler needs the ratio freq_curr/freq_max for frequency-invariant
accounting. On all ATOM CPUs prior to Goldmont, set freq_max to the 1-core
turbo ratio.
We intended to perform tests validating that this patch doesn't regress in
terms of energy efficiency, given that this is the primary concern on Atom
processors. Alas, we found out that turbostat doesn't support reading RAPL
interfaces on our test machine (Airmont), and we don't have external equipment
to measure power consumption; all we have is the performance results of the
benchmarks we ran.
Test machine:
Platform : Dell Wyse 3040 Thin Client[1]
CPU Model : Intel Atom x5-Z8350 (aka Cherry Trail, aka Airmont)
Fam/Mod/Ste : 6:76:4
Topology : 1 socket, 4 cores / 4 threads
Memory : 2G
Storage : onboard flash, XFS filesystem
[1] https://www.dell.com/en-us/work/shop/wyse-endpoints-and-software/wyse-3040-thin-client/spd/wyse-3040-thin-client
Base frequency and available turbo levels (MHz):
Min Operating Freq 266 |***
Low Freq Mode 800 |********
Base Freq 2400 |************************
4 Cores 2800 |****************************
3 Cores 2800 |****************************
2 Cores 3200 |********************************
1 Core 3200 |********************************
Tested kernels:
Baseline : v5.4-rc1, intel_pstate passive, schedutil
Comparison #1 : v5.4-rc1, intel_pstate active , powersave
Comparison #2 : v5.4-rc1, this patch, intel_pstate passive, schedutil
tbench, hackbench and kernbench performed the same under all three kernels;
dbench ran faster with intel_pstate/powersave and the git unit tests were a
lot faster with intel_pstate/powersave and invariant schedutil wrt the
baseline. Not that any of this is terrbily interesting anyway, one doesn't buy
an Atom system to go fast. Power consumption regressions aren't expected but
we lack the equipment to make that measurement. Turbostat seems to think that
reading RAPL on this machine isn't a good idea and we're trusting that
decision.
comparison ratio of performance with baseline; 1.00 means neutral,
lower is better:
I_PSTATE FREQ-INV
----------------------------------------
dbench 0.90 ~
kernbench 0.98 0.97
gitsource 0.63 0.43
Signed-off-by: Giovanni Gherdovich <ggherdovich@suse.cz>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Link: https://lkml.kernel.org/r/20200122151617.531-6-ggherdovich@suse.cz
|
|
The scheduler needs the ratio freq_curr/freq_max for frequency-invariant
accounting. On GOLDMONT (aka Apollo Lake), GOLDMONT_D (aka Denverton) and
GOLDMONT_PLUS CPUs (aka Gemini Lake) set freq_max to the highest frequency
reported by the CPU.
The encoding of turbo ratios for GOLDMONT* is identical to the one for
SKYLAKE_X, but we treat the Atom case apart because we want to set freq_max to
a higher value, thus the ratio freq_curr/freq_max to be lower, leading to more
conservative frequency selections (favoring power efficiency).
Signed-off-by: Giovanni Gherdovich <ggherdovich@suse.cz>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Link: https://lkml.kernel.org/r/20200122151617.531-5-ggherdovich@suse.cz
|
|
The scheduler needs the ratio freq_curr/freq_max for frequency-invariant
accounting. On Xeon Phi CPUs set freq_max to the second-highest frequency
reported by the CPU.
Xeon Phi CPUs such as Knights Landing and Knights Mill typically have either
one or two turbo frequencies; in the former case that's 100 MHz above the base
frequency, in the latter case the two levels are 100 MHz and 200 MHz above
base frequency.
We set freq_max to the second-highest frequency reported by the CPU. This
could be the base frequency (if only one turbo level is available) or the first
turbo level (if two levels are available). The rationale is to compromise
between power efficiency or performance -- going straight to max turbo would
favor efficiency and blindly using base freq would favor performance.
For reference, this is how MSR_TURBO_RATIO_LIMIT must be parsed on a Xeon Phi
to get the available frequencies (taken from a comment in turbostat's sources):
[0] -- Reserved
[7:1] -- Base value of number of active cores of bucket 1.
[15:8] -- Base value of freq ratio of bucket 1.
[20:16] -- +ve delta of number of active cores of bucket 2.
i.e. active cores of bucket 2 =
active cores of bucket 1 + delta
[23:21] -- Negative delta of freq ratio of bucket 2.
i.e. freq ratio of bucket 2 =
freq ratio of bucket 1 - delta
[28:24]-- +ve delta of number of active cores of bucket 3.
[31:29]-- -ve delta of freq ratio of bucket 3.
[36:32]-- +ve delta of number of active cores of bucket 4.
[39:37]-- -ve delta of freq ratio of bucket 4.
[44:40]-- +ve delta of number of active cores of bucket 5.
[47:45]-- -ve delta of freq ratio of bucket 5.
[52:48]-- +ve delta of number of active cores of bucket 6.
[55:53]-- -ve delta of freq ratio of bucket 6.
[60:56]-- +ve delta of number of active cores of bucket 7.
[63:61]-- -ve delta of freq ratio of bucket 7.
1. PERFORMANCE EVALUATION: TBENCH +5%
2. NEUTRAL BENCHMARKS (ALL OTHERS)
3. TEST SETUP
1. PERFORMANCE EVALUATION: TBENCH +5%
-------------------------------------
A performance evaluation was conducted on a Knights Mill machine (see "Test
Setup" below), were the frequency-invariance patch (on schedutil) is compared
to both non-invariant schedutil and active intel_pstate with powersave: all
three tested kernels behave the same performance-wise and with regard to power
consumption (performance per watt). The only notable difference is tbench:
comparison ratio of performance with baseline; 1.00 means neutral,
higher is better:
I_PSTATE FREQ-INV
----------------------------------------
tbench 1.04 1.05
performance-per-watt ratios with baseline; 1.00 means neutral, higher is better:
I_PSTATE FREQ-INV
----------------------------------------
tbench 1.03 1.04
which essentially means that frequency-invariant schedutil is 5% better than
baseline, the same as intel_pstate+powersave.
As the results above are averaged over the varying parameter, here the detailed
table.
Varying parameter : number of clients
Unit : MB/sec (higher is better)
5.2.0 vanilla (BASELINE) 5.2.0 intel_pstate 5.2.0 freq-inv
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
Hmean 1 49.06 +- 2.12% ( ) 51.66 +- 1.52% ( 5.30%) 52.87 +- 0.88% ( 7.76%)
Hmean 2 93.82 +- 0.45% ( ) 103.24 +- 0.70% ( 10.05%) 105.90 +- 0.70% ( 12.88%)
Hmean 4 192.46 +- 1.15% ( ) 215.95 +- 0.60% ( 12.21%) 215.78 +- 1.43% ( 12.12%)
Hmean 8 406.74 +- 2.58% ( ) 438.58 +- 0.36% ( 7.83%) 437.61 +- 0.97% ( 7.59%)
Hmean 16 857.70 +- 1.22% ( ) 890.26 +- 0.72% ( 3.80%) 889.11 +- 0.73% ( 3.66%)
Hmean 32 1760.10 +- 0.92% ( ) 1791.70 +- 0.44% ( 1.79%) 1787.95 +- 0.44% ( 1.58%)
Hmean 64 3183.50 +- 0.34% ( ) 3183.19 +- 0.36% ( -0.01%) 3187.53 +- 0.36% ( 0.13%)
Hmean 128 4830.96 +- 0.31% ( ) 4846.53 +- 0.30% ( 0.32%) 4855.86 +- 0.30% ( 0.52%)
Hmean 256 5467.98 +- 0.38% ( ) 5793.80 +- 0.28% ( 5.96%) 5821.94 +- 0.17% ( 6.47%)
Hmean 512 5398.10 +- 0.06% ( ) 5745.56 +- 0.08% ( 6.44%) 5503.68 +- 0.07% ( 1.96%)
Hmean 1024 5290.43 +- 0.63% ( ) 5221.07 +- 0.47% ( -1.31%) 5277.22 +- 0.80% ( -0.25%)
Hmean 1088 5139.71 +- 0.57% ( ) 5236.02 +- 0.71% ( 1.87%) 5190.57 +- 0.41% ( 0.99%)
2. NEUTRAL BENCHMARKS (ALL OTHERS)
----------------------------------
* pgbench (both read/write and read-only)
* NASA Parallel Benchmarks (NPB), MPI or OpenMP for message-passing
* hackbench
* netperf
* dbench
* kernbench
* gitsource (git unit test suite)
3. TEST SETUP
-------------
Test machine:
CPU Model : Intel Xeon Phi CPU 7255 @ 1.10GHz (a.k.a. Knights Mill)
Fam/Mod/Ste : 6:133:0
Topology : 1 socket, 68 cores / 272 threads
Memory : 96G
Storage : rotary, XFS filesystem
Max EFFICiency, BASE frequency and available turbo levels (MHz):
EFFIC 1000 |**********
BASE 1100 |***********
68C 1100 |***********
30C 1200 |************
Tested kernels:
Baseline : v5.2, intel_pstate passive, schedutil
Comparison #1 : v5.2, intel_pstate active , powersave
Comparison #2 : v5.2, this patch, intel_pstate passive, schedutil
Signed-off-by: Giovanni Gherdovich <ggherdovich@suse.cz>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Link: https://lkml.kernel.org/r/20200122151617.531-4-ggherdovich@suse.cz
|
|
The scheduler needs the ratio freq_curr/freq_max for frequency-invariant
accounting. On SKYLAKE_X CPUs set freq_max to the highest frequency that can
be sustained by a group of at least 4 cores.
From the changelog of commit 31e07522be56 ("tools/power turbostat: fix
decoding for GLM, DNV, SKX turbo-ratio limits"):
> Newer processors do not hard-code the the number of cpus in each bin
> to {1, 2, 3, 4, 5, 6, 7, 8} Rather, they can specify any number
> of CPUS in each of the 8 bins:
>
> eg.
>
> ...
> 37 * 100.0 = 3600.0 MHz max turbo 4 active cores
> 38 * 100.0 = 3700.0 MHz max turbo 3 active cores
> 39 * 100.0 = 3800.0 MHz max turbo 2 active cores
> 39 * 100.0 = 3900.0 MHz max turbo 1 active cores
>
> could now look something like this:
>
> ...
> 37 * 100.0 = 3600.0 MHz max turbo 16 active cores
> 38 * 100.0 = 3700.0 MHz max turbo 8 active cores
> 39 * 100.0 = 3800.0 MHz max turbo 4 active cores
> 39 * 100.0 = 3900.0 MHz max turbo 2 active cores
This encoding of turbo levels applies to both SKYLAKE_X and GOLDMONT/GOLDMONT_D,
but we treat these two classes in separate commits because their freq_max
values need to be different. For SKX we prefer a lower freq_max in the ratio
freq_curr/freq_max, allowing load and utilization to overshoot and the
schedutil governor to be more performance-oriented. Models from the Atom
series (such as GOLDMONT*) are handled in a forthcoming commit as they have to
favor power-efficiency over performance.
Results from a performance evaluation follow.
1. TEST SETUP
2. NEUTRAL BENCHMARKS
3. NON-NEUTRAL BENCHMARKS
4. DETAILED TABLES
1. TEST SETUP
-------------
Test machine:
CPU Model : Intel Xeon Platinum 8260L CPU @ 2.40GHz (a.k.a. Cascade Lake)
Fam/Mod/Ste : 6:85:6
Topology : 2 sockets, 24 cores / 48 threads each socket
Memory : 192G
Storage : SSD, XFS filesystem
Max EFFICiency, BASE frequency and available turbo levels (MHz):
EFFIC 1000 |**********
BASE 2400 |************************
24C 3100 |*******************************
20C 3300 |*********************************
16C 3600 |************************************
12C 3600 |************************************
8C 3600 |************************************
4C 3700 |*************************************
2C 3900 |***************************************
Tested kernels:
Baseline : v5.2, intel_pstate passive, schedutil
Comparison #1 : v5.2, intel_pstate active , powersave+HWP
Comparison #2 : v5.2, this patch, intel_pstate passive, schedutil
2. NEUTRAL BENCHMARKS
---------------------
* pgbench read/write
* NASA Parallel Benchmarks (NPB), MPI or OpenMP for message-passing
* hackbench
* netperf
3. NON-NEUTRAL BENCHMARKS
-------------------------
comparison ratio with baseline; 1.00 means neutral, higher is better:
I_PSTATE FREQ-INV
----------------------------------------
pgbench read-only 1.10 ~
tbench 1.82 1.14
comparison ratio with baseline; 1.00 means neutral, lower is better:
I_PSTATE FREQ-INV
----------------------------------------
dbench ~ 0.97
kernbench 0.88 0.78
gitsource[*] ~ 0.46
[*] "gitsource" consists in running git's unit tests
tilde (~) means 1.00, ie result identical to baseline
Performance per watt:
performance-per-watt ratios with baseline; 1.00 means neutral, higher is better:
I_PSTATE FREQ-INV
----------------------------------------
dbench 0.92 0.91
tbench 1.26 1.04
kernbench 0.95 0.96
gitsource 1.03 1.30
Similarly to earlier Xeons, measurable performance gains over non-invariant
schedutil are observed on dbench, tbench, kernel compilation and running the
git unit tests suite. Looking at the detailed tables show that the patch
scores the largest difference when the machine is lightly loaded. Power
efficiency suffers lightly on kernbench and a bit more on dbench, but largely
improves on gitsource (which also runs considerably faster). For reference, we
also report results using active intel_pstate with powersave and HWP; the
largest gap between non-invariant schedutil and intel_pstate+powersave is
still tbench, which runs 82% better and with 26% improved efficiency on the
latter configuration -- this divide isn't closed yet by frequency-invariant
schedutil.
4. DETAILED TABLES
------------------
Benchmark : tbench4 (i.e. dbench4 over the network, actually loopback)
Varying parameter : number of clients
Unit : MB/sec (higher is better)
5.2.0 vanilla (BASELINE) 5.2.0 intel_pstate/HWP 5.2.0 freq-inv
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
Hmean 1 183.56 +- 0.21% ( ) 516.12 +- 0.57% ( 181.18%) 185.59 +- 0.59% ( 1.11%)
Hmean 2 365.75 +- 0.25% ( ) 1015.14 +- 0.33% ( 177.55%) 402.59 +- 4.48% ( 10.07%)
Hmean 4 720.99 +- 0.44% ( ) 1951.75 +- 0.28% ( 170.70%) 738.39 +- 1.72% ( 2.41%)
Hmean 8 1449.93 +- 0.34% ( ) 3830.56 +- 0.24% ( 164.19%) 1750.36 +- 4.65% ( 20.72%)
Hmean 16 2874.26 +- 0.57% ( ) 7381.62 +- 0.53% ( 156.82%) 4348.35 +- 2.22% ( 51.29%)
Hmean 32 6116.17 +- 5.10% ( ) 13013.05 +- 0.08% ( 112.76%) 8980.35 +- 0.66% ( 46.83%)
Hmean 64 14485.04 +- 3.46% ( ) 17835.12 +- 0.35% ( 23.13%) 16540.73 +- 0.51% ( 14.19%)
Hmean 128 30779.16 +- 3.20% ( ) 32796.94 +- 2.13% ( 6.56%) 31512.58 +- 0.20% ( 2.38%)
Hmean 256 34664.66 +- 0.81% ( ) 34604.67 +- 0.46% ( -0.17%) 34943.70 +- 0.25% ( 0.80%)
Hmean 384 33957.51 +- 0.11% ( ) 34091.50 +- 0.14% ( 0.39%) 33921.41 +- 0.09% ( -0.11%)
Benchmark : kernbench (kernel compilation)
Varying parameter : number of jobs
Unit : seconds (lower is better)
5.2.0 vanilla (BASELINE) 5.2.0 intel_pstate/HWP 5.2.0 freq-inv
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
Amean 2 332.94 +- 0.40% ( ) 260.16 +- 0.45% ( 21.86%) 233.56 +- 0.21% ( 29.85%)
Amean 4 173.04 +- 0.43% ( ) 138.76 +- 0.03% ( 19.81%) 123.59 +- 0.11% ( 28.58%)
Amean 8 89.65 +- 0.20% ( ) 73.54 +- 0.09% ( 17.97%) 65.69 +- 0.10% ( 26.72%)
Amean 16 48.08 +- 1.41% ( ) 41.64 +- 1.61% ( 13.40%) 36.00 +- 1.80% ( 25.11%)
Amean 32 28.78 +- 0.72% ( ) 26.61 +- 1.99% ( 7.55%) 23.19 +- 1.68% ( 19.43%)
Amean 64 20.46 +- 1.85% ( ) 19.76 +- 0.35% ( 3.42%) 17.38 +- 0.92% ( 15.06%)
Amean 128 18.69 +- 1.70% ( ) 17.59 +- 1.04% ( 5.90%) 15.73 +- 1.40% ( 15.85%)
Amean 192 18.82 +- 1.01% ( ) 17.76 +- 0.77% ( 5.67%) 15.57 +- 1.80% ( 17.28%)
Benchmark : gitsource (time to run the git unit test suite)
Varying parameter : none
Unit : seconds (lower is better)
5.2.0 vanilla (BASELINE) 5.2.0 intel_pstate/HWP 5.2.0 freq-inv
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
Amean 792.49 +- 0.20% ( ) 779.35 +- 0.24% ( 1.66%) 427.14 +- 0.16% ( 46.10%)
Signed-off-by: Giovanni Gherdovich <ggherdovich@suse.cz>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Link: https://lkml.kernel.org/r/20200122151617.531-3-ggherdovich@suse.cz
|
|
Implement arch_scale_freq_capacity() for 'modern' x86. This function
is used by the scheduler to correctly account usage in the face of
DVFS.
The present patch addresses Intel processors specifically and has positive
performance and performance-per-watt implications for the schedutil cpufreq
governor, bringing it closer to, if not on-par with, the powersave governor
from the intel_pstate driver/framework.
Large performance gains are obtained when the machine is lightly loaded and
no regression are observed at saturation. The benchmarks with the largest
gains are kernel compilation, tbench (the networking version of dbench) and
shell-intensive workloads.
1. FREQUENCY INVARIANCE: MOTIVATION
* Without it, a task looks larger if the CPU runs slower
2. PECULIARITIES OF X86
* freq invariance accounting requires knowing the ratio freq_curr/freq_max
2.1 CURRENT FREQUENCY
* Use delta_APERF / delta_MPERF * freq_base (a.k.a "BusyMHz")
2.2 MAX FREQUENCY
* It varies with time (turbo). As an approximation, we set it to a
constant, i.e. 4-cores turbo frequency.
3. EFFECTS ON THE SCHEDUTIL FREQUENCY GOVERNOR
* The invariant schedutil's formula has no feedback loop and reacts faster
to utilization changes
4. KNOWN LIMITATIONS
* In some cases tasks can't reach max util despite how hard they try
5. PERFORMANCE TESTING
5.1 MACHINES
* Skylake, Broadwell, Haswell
5.2 SETUP
* baseline Linux v5.2 w/ non-invariant schedutil. Tested freq_max = 1-2-3-4-8-12
active cores turbo w/ invariant schedutil, and intel_pstate/powersave
5.3 BENCHMARK RESULTS
5.3.1 NEUTRAL BENCHMARKS
* NAS Parallel Benchmark (HPC), hackbench
5.3.2 NON-NEUTRAL BENCHMARKS
* tbench (10-30% better), kernbench (10-15% better),
shell-intensive-scripts (30-50% better)
* no regressions
5.3.3 SELECTION OF DETAILED RESULTS
5.3.4 POWER CONSUMPTION, PERFORMANCE-PER-WATT
* dbench (5% worse on one machine), kernbench (3% worse),
tbench (5-10% better), shell-intensive-scripts (10-40% better)
6. MICROARCH'ES ADDRESSED HERE
* Xeon Core before Scalable Performance processors line (Xeon Gold/Platinum
etc have different MSRs semantic for querying turbo levels)
7. REFERENCES
* MMTests performance testing framework, github.com/gormanm/mmtests
+-------------------------------------------------------------------------+
| 1. FREQUENCY INVARIANCE: MOTIVATION
+-------------------------------------------------------------------------+
For example; suppose a CPU has two frequencies: 500 and 1000 Mhz. When
running a task that would consume 1/3rd of a CPU at 1000 MHz, it would
appear to consume 2/3rd (or 66.6%) when running at 500 MHz, giving the
false impression this CPU is almost at capacity, even though it can go
faster [*]. In a nutshell, without frequency scale-invariance tasks look
larger just because the CPU is running slower.
[*] (footnote: this assumes a linear frequency/performance relation; which
everybody knows to be false, but given realities its the best approximation
we can make.)
+-------------------------------------------------------------------------+
| 2. PECULIARITIES OF X86
+-------------------------------------------------------------------------+
Accounting for frequency changes in PELT signals requires the computation of
the ratio freq_curr / freq_max. On x86 neither of those terms is readily
available.
2.1 CURRENT FREQUENCY
====================
Since modern x86 has hardware control over the actual frequency we run
at (because amongst other things, Turbo-Mode), we cannot simply use
the frequency as requested through cpufreq.
Instead we use the APERF/MPERF MSRs to compute the effective frequency
over the recent past. Also, because reading MSRs is expensive, don't
do so every time we need the value, but amortize the cost by doing it
every tick.
2.2 MAX FREQUENCY
=================
Obtaining freq_max is also non-trivial because at any time the hardware can
provide a frequency boost to a selected subset of cores if the package has
enough power to spare (eg: Turbo Boost). This means that the maximum frequency
available to a given core changes with time.
The approach taken in this change is to arbitrarily set freq_max to a constant
value at boot. The value chosen is the "4-cores (4C) turbo frequency" on most
microarchitectures, after evaluating the following candidates:
* 1-core (1C) turbo frequency (the fastest turbo state available)
* around base frequency (a.k.a. max P-state)
* something in between, such as 4C turbo
To interpret these options, consider that this is the denominator in
freq_curr/freq_max, and that ratio will be used to scale PELT signals such as
util_avg and load_avg. A large denominator will undershoot (util_avg looks a
bit smaller than it really is), viceversa with a smaller denominator PELT
signals will tend to overshoot. Given that PELT drives frequency selection
in the schedutil governor, we will have:
freq_max set to | effect on DVFS
--------------------+------------------
1C turbo | power efficiency (lower freq choices)
base freq | performance (higher util_avg, higher freq requests)
4C turbo | a bit of both
4C turbo proves to be a good compromise in a number of benchmarks (see below).
+-------------------------------------------------------------------------+
| 3. EFFECTS ON THE SCHEDUTIL FREQUENCY GOVERNOR
+-------------------------------------------------------------------------+
Once an architecture implements a frequency scale-invariant utilization (the
PELT signal util_avg), schedutil switches its frequency selection formula from
freq_next = 1.25 * freq_curr * util [non-invariant util signal]
to
freq_next = 1.25 * freq_max * util [invariant util signal]
where, in the second formula, freq_max is set to the 1C turbo frequency (max
turbo). The advantage of the second formula, whose usage we unlock with this
patch, is that freq_next doesn't depend on the current frequency in an
iterative fashion, but can jump to any frequency in a single update. This
absence of feedback in the formula makes it quicker to react to utilization
changes and more robust against pathological instabilities.
Compare it to the update formula of intel_pstate/powersave:
freq_next = 1.25 * freq_max * Busy%
where again freq_max is 1C turbo and Busy% is the percentage of time not spent
idling (calculated with delta_MPERF / delta_TSC); essentially the same as
invariant schedutil, and largely responsible for intel_pstate/powersave good
reputation. The non-invariant schedutil formula is derived from the invariant
one by approximating util_inv with util_raw * freq_curr / freq_max, but this
has limitations.
Testing shows improved performances due to better frequency selections when
the machine is lightly loaded, and essentially no change in behaviour at
saturation / overutilization.
+-------------------------------------------------------------------------+
| 4. KNOWN LIMITATIONS
+-------------------------------------------------------------------------+
It's been shown that it is possible to create pathological scenarios where a
CPU-bound task cannot reach max utilization, if the normalizing factor
freq_max is fixed to a constant value (see [Lelli-2018]).
If freq_max is set to 4C turbo as we do here, one needs to peg at least 5
cores in a package doing some busywork, and observe that none of those task
will ever reach max util (1024) because they're all running at less than the
4C turbo frequency.
While this concern still applies, we believe the performance benefit of
frequency scale-invariant PELT signals outweights the cost of this limitation.
[Lelli-2018]
https://lore.kernel.org/lkml/20180517150418.GF22493@localhost.localdomain/
+-------------------------------------------------------------------------+
| 5. PERFORMANCE TESTING
+-------------------------------------------------------------------------+
5.1 MACHINES
============
We tested the patch on three machines, with Skylake, Broadwell and Haswell
CPUs. The details are below, together with the available turbo ratios as
reported by the appropriate MSRs.
* 8x-SKYLAKE-UMA:
Single socket E3-1240 v5, Skylake 4 cores/8 threads
Max EFFiciency, BASE frequency and available turbo levels (MHz):
EFFIC 800 |********
BASE 3500 |***********************************
4C 3700 |*************************************
3C 3800 |**************************************
2C 3900 |***************************************
1C 3900 |***************************************
* 80x-BROADWELL-NUMA:
Two sockets E5-2698 v4, 2x Broadwell 20 cores/40 threads
Max EFFiciency, BASE frequency and available turbo levels (MHz):
EFFIC 1200 |************
BASE 2200 |**********************
8C 2900 |*****************************
7C 3000 |******************************
6C 3100 |*******************************
5C 3200 |********************************
4C 3300 |*********************************
3C 3400 |**********************************
2C 3600 |************************************
1C 3600 |************************************
* 48x-HASWELL-NUMA
Two sockets E5-2670 v3, 2x Haswell 12 cores/24 threads
Max EFFiciency, BASE frequency and available turbo levels (MHz):
EFFIC 1200 |************
BASE 2300 |***********************
12C 2600 |**************************
11C 2600 |**************************
10C 2600 |**************************
9C 2600 |**************************
8C 2600 |**************************
7C 2600 |**************************
6C 2600 |**************************
5C 2700 |***************************
4C 2800 |****************************
3C 2900 |*****************************
2C 3100 |*******************************
1C 3100 |*******************************
5.2 SETUP
=========
* The baseline is Linux v5.2 with schedutil (non-invariant) and the intel_pstate
driver in passive mode.
* The rationale for choosing the various freq_max values to test have been to
try all the 1-2-3-4C turbo levels (note that 1C and 2C turbo are identical
on all machines), plus one more value closer to base_freq but still in the
turbo range (8C turbo for both 80x-BROADWELL-NUMA and 48x-HASWELL-NUMA).
* In addition we've run all tests with intel_pstate/powersave for comparison.
* The filesystem is always XFS, the userspace is openSUSE Leap 15.1.
* 8x-SKYLAKE-UMA is capable of HWP (Hardware-Managed P-States), so the runs
with active intel_pstate on this machine use that.
This gives, in terms of combinations tested on each machine:
* 8x-SKYLAKE-UMA
* Baseline: Linux v5.2, non-invariant schedutil, intel_pstate passive
* intel_pstate active + powersave + HWP
* invariant schedutil, freq_max = 1C turbo
* invariant schedutil, freq_max = 3C turbo
* invariant schedutil, freq_max = 4C turbo
* both 80x-BROADWELL-NUMA and 48x-HASWELL-NUMA
* [same as 8x-SKYLAKE-UMA, but no HWP capable]
* invariant schedutil, freq_max = 8C turbo
(which on 48x-HASWELL-NUMA is the same as 12C turbo, or "all cores turbo")
5.3 BENCHMARK RESULTS
=====================
5.3.1 NEUTRAL BENCHMARKS
------------------------
Tests that didn't show any measurable difference in performance on any of the
test machines between non-invariant schedutil and our patch are:
* NAS Parallel Benchmarks (NPB) using either MPI or openMP for IPC, any
computational kernel
* flexible I/O (FIO)
* hackbench (using threads or processes, and using pipes or sockets)
5.3.2 NON-NEUTRAL BENCHMARKS
----------------------------
What follow are summary tables where each benchmark result is given a score.
* A tilde (~) means a neutral result, i.e. no difference from baseline.
* Scores are computed with the ratio result_new / result_baseline, so a tilde
means a score of 1.00.
* The results in the score ratio are the geometric means of results running
the benchmark with different parameters (eg: for kernbench: using 1, 2, 4,
... number of processes; for pgbench: varying the number of clients, and so
on).
* The first three tables show higher-is-better kind of tests (i.e. measured in
operations/second), the subsequent three show lower-is-better kind of tests
(i.e. the workload is fixed and we measure elapsed time, think kernbench).
* "gitsource" is a name we made up for the test consisting in running the
entire unit tests suite of the Git SCM and measuring how long it takes. We
take it as a typical example of shell-intensive serialized workload.
* In the "I_PSTATE" column we have the results for intel_pstate/powersave. Other
columns show invariant schedutil for different values of freq_max. 4C turbo
is circled as it's the value we've chosen for the final implementation.
80x-BROADWELL-NUMA (comparison ratio; higher is better)
+------+
I_PSTATE 1C 3C | 4C | 8C
pgbench-ro 1.14 ~ ~ | 1.11 | 1.14
pgbench-rw ~ ~ ~ | ~ | ~
netperf-udp 1.06 ~ 1.06 | 1.05 | 1.07
netperf-tcp ~ 1.03 ~ | 1.01 | 1.02
tbench4 1.57 1.18 1.22 | 1.30 | 1.56
+------+
8x-SKYLAKE-UMA (comparison ratio; higher is better)
+------+
I_PSTATE/HWP 1C 3C | 4C |
pgbench-ro ~ ~ ~ | ~ |
pgbench-rw ~ ~ ~ | ~ |
netperf-udp ~ ~ ~ | ~ |
netperf-tcp ~ ~ ~ | ~ |
tbench4 1.30 1.14 1.14 | 1.16 |
+------+
48x-HASWELL-NUMA (comparison ratio; higher is better)
+------+
I_PSTATE 1C 3C | 4C | 12C
pgbench-ro 1.15 ~ ~ | 1.06 | 1.16
pgbench-rw ~ ~ ~ | ~ | ~
netperf-udp 1.05 0.97 1.04 | 1.04 | 1.02
netperf-tcp 0.96 1.01 1.01 | 1.01 | 1.01
tbench4 1.50 1.05 1.13 | 1.13 | 1.25
+------+
In the table above we see that active intel_pstate is slightly better than our
4C-turbo patch (both in reference to the baseline non-invariant schedutil) on
read-only pgbench and much better on tbench. Both cases are notable in which
it shows that lowering our freq_max (to 8C-turbo and 12C-turbo on
80x-BROADWELL-NUMA and 48x-HASWELL-NUMA respectively) helps invariant
schedutil to get closer.
If we ignore active intel_pstate and focus on the comparison with baseline
alone, there are several instances of double-digit performance improvement.
80x-BROADWELL-NUMA (comparison ratio; lower is better)
+------+
I_PSTATE 1C 3C | 4C | 8C
dbench4 1.23 0.95 0.95 | 0.95 | 0.95
kernbench 0.93 0.83 0.83 | 0.83 | 0.82
gitsource 0.98 0.49 0.49 | 0.49 | 0.48
+------+
8x-SKYLAKE-UMA (comparison ratio; lower is better)
+------+
I_PSTATE/HWP 1C 3C | 4C |
dbench4 ~ ~ ~ | ~ |
kernbench ~ ~ ~ | ~ |
gitsource 0.92 0.55 0.55 | 0.55 |
+------+
48x-HASWELL-NUMA (comparison ratio; lower is better)
+------+
I_PSTATE 1C 3C | 4C | 8C
dbench4 ~ ~ ~ | ~ | ~
kernbench 0.94 0.90 0.89 | 0.90 | 0.90
gitsource 0.97 0.69 0.69 | 0.69 | 0.69
+------+
dbench is not very remarkable here, unless we notice how poorly active
intel_pstate is performing on 80x-BROADWELL-NUMA: 23% regression versus
non-invariant schedutil. We repeated that run getting consistent results. Out
of scope for the patch at hand, but deserving future investigation. Other than
that, we previously ran this campaign with Linux v5.0 and saw the patch doing
better on dbench a the time. We haven't checked closely and can only speculate
at this point.
On the NUMA boxes kernbench gets 10-15% improvements on average; we'll see in
the detailed tables that the gains concentrate on low process counts (lightly
loaded machines).
The test we call "gitsource" (running the git unit test suite, a long-running
single-threaded shell script) appears rather spectacular in this table (gains
of 30-50% depending on the machine). It is to be noted, however, that
gitsource has no adjustable parameters (such as the number of jobs in
kernbench, which we average over in order to get a single-number summary
score) and is exactly the kind of low-parallelism workload that benefits the
most from this patch. When looking at the detailed tables of kernbench or
tbench4, at low process or client counts one can see similar numbers.
5.3.3 SELECTION OF DETAILED RESULTS
-----------------------------------
Machine : 48x-HASWELL-NUMA
Benchmark : tbench4 (i.e. dbench4 over the network, actually loopback)
Varying parameter : number of clients
Unit : MB/sec (higher is better)
5.2.0 vanilla (BASELINE) 5.2.0 intel_pstate 5.2.0 1C-turbo
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
Hmean 1 126.73 +- 0.31% ( ) 315.91 +- 0.66% ( 149.28%) 125.03 +- 0.76% ( -1.34%)
Hmean 2 258.04 +- 0.62% ( ) 614.16 +- 0.51% ( 138.01%) 269.58 +- 1.45% ( 4.47%)
Hmean 4 514.30 +- 0.67% ( ) 1146.58 +- 0.54% ( 122.94%) 533.84 +- 1.99% ( 3.80%)
Hmean 8 1111.38 +- 2.52% ( ) 2159.78 +- 0.38% ( 94.33%) 1359.92 +- 1.56% ( 22.36%)
Hmean 16 2286.47 +- 1.36% ( ) 3338.29 +- 0.21% ( 46.00%) 2720.20 +- 0.52% ( 18.97%)
Hmean 32 4704.84 +- 0.35% ( ) 4759.03 +- 0.43% ( 1.15%) 4774.48 +- 0.30% ( 1.48%)
Hmean 64 7578.04 +- 0.27% ( ) 7533.70 +- 0.43% ( -0.59%) 7462.17 +- 0.65% ( -1.53%)
Hmean 128 6998.52 +- 0.16% ( ) 6987.59 +- 0.12% ( -0.16%) 6909.17 +- 0.14% ( -1.28%)
Hmean 192 6901.35 +- 0.25% ( ) 6913.16 +- 0.10% ( 0.17%) 6855.47 +- 0.21% ( -0.66%)
5.2.0 3C-turbo 5.2.0 4C-turbo 5.2.0 12C-turbo
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
Hmean 1 128.43 +- 0.28% ( 1.34%) 130.64 +- 3.81% ( 3.09%) 153.71 +- 5.89% ( 21.30%)
Hmean 2 311.70 +- 6.15% ( 20.79%) 281.66 +- 3.40% ( 9.15%) 305.08 +- 5.70% ( 18.23%)
Hmean 4 641.98 +- 2.32% ( 24.83%) 623.88 +- 5.28% ( 21.31%) 906.84 +- 4.65% ( 76.32%)
Hmean 8 1633.31 +- 1.56% ( 46.96%) 1714.16 +- 0.93% ( 54.24%) 2095.74 +- 0.47% ( 88.57%)
Hmean 16 3047.24 +- 0.42% ( 33.27%) 3155.02 +- 0.30% ( 37.99%) 3634.58 +- 0.15% ( 58.96%)
Hmean 32 4734.31 +- 0.60% ( 0.63%) 4804.38 +- 0.23% ( 2.12%) 4674.62 +- 0.27% ( -0.64%)
Hmean 64 7699.74 +- 0.35% ( 1.61%) 7499.72 +- 0.34% ( -1.03%) 7659.03 +- 0.25% ( 1.07%)
Hmean 128 6935.18 +- 0.15% ( -0.91%) 6942.54 +- 0.10% ( -0.80%) 7004.85 +- 0.12% ( 0.09%)
Hmean 192 6901.62 +- 0.12% ( 0.00%) 6856.93 +- 0.10% ( -0.64%) 6978.74 +- 0.10% ( 1.12%)
This is one of the cases where the patch still can't surpass active
intel_pstate, not even when freq_max is as low as 12C-turbo. Otherwise, gains are
visible up to 16 clients and the saturated scenario is the same as baseline.
The scores in the summary table from the previous sections are ratios of
geometric means of the results over different clients, as seen in this table.
Machine : 80x-BROADWELL-NUMA
Benchmark : kernbench (kernel compilation)
Varying parameter : number of jobs
Unit : seconds (lower is better)
5.2.0 vanilla (BASELINE) 5.2.0 intel_pstate 5.2.0 1C-turbo
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
Amean 2 379.68 +- 0.06% ( ) 330.20 +- 0.43% ( 13.03%) 285.93 +- 0.07% ( 24.69%)
Amean 4 200.15 +- 0.24% ( ) 175.89 +- 0.22% ( 12.12%) 153.78 +- 0.25% ( 23.17%)
Amean 8 106.20 +- 0.31% ( ) 95.54 +- 0.23% ( 10.03%) 86.74 +- 0.10% ( 18.32%)
Amean 16 56.96 +- 1.31% ( ) 53.25 +- 1.22% ( 6.50%) 48.34 +- 1.73% ( 15.13%)
Amean 32 34.80 +- 2.46% ( ) 33.81 +- 0.77% ( 2.83%) 30.28 +- 1.59% ( 12.99%)
Amean 64 26.11 +- 1.63% ( ) 25.04 +- 1.07% ( 4.10%) 22.41 +- 2.37% ( 14.16%)
Amean 128 24.80 +- 1.36% ( ) 23.57 +- 1.23% ( 4.93%) 21.44 +- 1.37% ( 13.55%)
Amean 160 24.85 +- 0.56% ( ) 23.85 +- 1.17% ( 4.06%) 21.25 +- 1.12% ( 14.49%)
5.2.0 3C-turbo 5.2.0 4C-turbo 5.2.0 8C-turbo
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
Amean 2 284.08 +- 0.13% ( 25.18%) 283.96 +- 0.51% ( 25.21%) 285.05 +- 0.21% ( 24.92%)
Amean 4 153.18 +- 0.22% ( 23.47%) 154.70 +- 1.64% ( 22.71%) 153.64 +- 0.30% ( 23.24%)
Amean 8 87.06 +- 0.28% ( 18.02%) 86.77 +- 0.46% ( 18.29%) 86.78 +- 0.22% ( 18.28%)
Amean 16 48.03 +- 0.93% ( 15.68%) 47.75 +- 1.99% ( 16.17%) 47.52 +- 1.61% ( 16.57%)
Amean 32 30.23 +- 1.20% ( 13.14%) 30.08 +- 1.67% ( 13.57%) 30.07 +- 1.67% ( 13.60%)
Amean 64 22.59 +- 2.02% ( 13.50%) 22.63 +- 0.81% ( 13.32%) 22.42 +- 0.76% ( 14.12%)
Amean 128 21.37 +- 0.67% ( 13.82%) 21.31 +- 1.15% ( 14.07%) 21.17 +- 1.93% ( 14.63%)
Amean 160 21.68 +- 0.57% ( 12.76%) 21.18 +- 1.74% ( 14.77%) 21.22 +- 1.00% ( 14.61%)
The patch outperform active intel_pstate (and baseline) by a considerable
margin; the summary table from the previous section says 4C turbo and active
intel_pstate are 0.83 and 0.93 against baseline respectively, so 4C turbo is
0.83/0.93=0.89 against intel_pstate (~10% better on average). There is no
noticeable difference with regard to the value of freq_max.
Machine : 8x-SKYLAKE-UMA
Benchmark : gitsource (time to run the git unit test suite)
Varying parameter : none
Unit : seconds (lower is better)
5.2.0 vanilla 5.2.0 intel_pstate/hwp 5.2.0 1C-turbo
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
Amean 858.85 +- 1.16% ( ) 791.94 +- 0.21% ( 7.79%) 474.95 ( 44.70%)
5.2.0 3C-turbo 5.2.0 4C-turbo
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
Amean 475.26 +- 0.20% ( 44.66%) 474.34 +- 0.13% ( 44.77%)
In this test, which is of interest as representing shell-intensive
(i.e. fork-intensive) serialized workloads, invariant schedutil outperforms
intel_pstate/powersave by a whopping 40% margin.
5.3.4 POWER CONSUMPTION, PERFORMANCE-PER-WATT
---------------------------------------------
The following table shows average power consumption in watt for each
benchmark. Data comes from turbostat (package average), which in turn is read
from the RAPL interface on CPUs. We know the patch affects CPU frequencies so
it's reasonable to ignore other power consumers (such as memory or I/O). Also,
we don't have a power meter available in the lab so RAPL is the best we have.
turbostat sampled average power every 10 seconds for the entire duration of
each benchmark. We took all those values and averaged them (i.e. with don't
have detail on a per-parameter granularity, only on whole benchmarks).
80x-BROADWELL-NUMA (power consumption, watts)
+--------+
BASELINE I_PSTATE 1C 3C | 4C | 8C
pgbench-ro 130.01 142.77 131.11 132.45 | 134.65 | 136.84
pgbench-rw 68.30 60.83 71.45 71.70 | 71.65 | 72.54
dbench4 90.25 59.06 101.43 99.89 | 101.10 | 102.94
netperf-udp 65.70 69.81 66.02 68.03 | 68.27 | 68.95
netperf-tcp 88.08 87.96 88.97 88.89 | 88.85 | 88.20
tbench4 142.32 176.73 153.02 163.91 | 165.58 | 176.07
kernbench 92.94 101.95 114.91 115.47 | 115.52 | 115.10
gitsource 40.92 41.87 75.14 75.20 | 75.40 | 75.70
+--------+
8x-SKYLAKE-UMA (power consumption, watts)
+--------+
BASELINE I_PSTATE/HWP 1C 3C | 4C |
pgbench-ro 46.49 46.68 46.56 46.59 | 46.52 |
pgbench-rw 29.34 31.38 30.98 31.00 | 31.00 |
dbench4 27.28 27.37 27.49 27.41 | 27.38 |
netperf-udp 22.33 22.41 22.36 22.35 | 22.36 |
netperf-tcp 27.29 27.29 27.30 27.31 | 27.33 |
tbench4 41.13 45.61 43.10 43.33 | 43.56 |
kernbench 42.56 42.63 43.01 43.01 | 43.01 |
gitsource 13.32 13.69 17.33 17.30 | 17.35 |
+--------+
48x-HASWELL-NUMA (power consumption, watts)
+--------+
BASELINE I_PSTATE 1C 3C | 4C | 12C
pgbench-ro 128.84 136.04 129.87 132.43 | 132.30 | 134.86
pgbench-rw 37.68 37.92 37.17 37.74 | 37.73 | 37.31
dbench4 28.56 28.73 28.60 28.73 | 28.70 | 28.79
netperf-udp 56.70 60.44 56.79 57.42 | 57.54 | 57.52
netperf-tcp 75.49 75.27 75.87 76.02 | 76.01 | 75.95
tbench4 115.44 139.51 119.53 123.07 | 123.97 | 130.22
kernbench 83.23 91.55 95.58 95.69 | 95.72 | 96.04
gitsource 36.79 36.99 39.99 40.34 | 40.35 | 40.23
+--------+
A lower power consumption isn't necessarily better, it depends on what is done
with that energy. Here are tables with the ratio of performance-per-watt on
each machine and benchmark. Higher is always better; a tilde (~) means a
neutral ratio (i.e. 1.00).
80x-BROADWELL-NUMA (performance-per-watt ratios; higher is better)
+------+
I_PSTATE 1C 3C | 4C | 8C
pgbench-ro 1.04 1.06 0.94 | 1.07 | 1.08
pgbench-rw 1.10 0.97 0.96 | 0.96 | 0.97
dbench4 1.24 0.94 0.95 | 0.94 | 0.92
netperf-udp ~ 1.02 1.02 | ~ | 1.02
netperf-tcp ~ 1.02 ~ | ~ | 1.02
tbench4 1.26 1.10 1.06 | 1.12 | 1.26
kernbench 0.98 0.97 0.97 | 0.97 | 0.98
gitsource ~ 1.11 1.11 | 1.11 | 1.13
+------+
8x-SKYLAKE-UMA (performance-per-watt ratios; higher is better)
+------+
I_PSTATE/HWP 1C 3C | 4C |
pgbench-ro ~ ~ ~ | ~ |
pgbench-rw 0.95 0.97 0.96 | 0.96 |
dbench4 ~ ~ ~ | ~ |
netperf-udp ~ ~ ~ | ~ |
netperf-tcp ~ ~ ~ | ~ |
tbench4 1.17 1.09 1.08 | 1.10 |
kernbench ~ ~ ~ | ~ |
gitsource 1.06 1.40 1.40 | 1.40 |
+------+
48x-HASWELL-NUMA (performance-per-watt ratios; higher is better)
+------+
I_PSTATE 1C 3C | 4C | 12C
pgbench-ro 1.09 ~ 1.09 | 1.03 | 1.11
pgbench-rw ~ 0.86 ~ | ~ | 0.86
dbench4 ~ 1.02 1.02 | 1.02 | ~
netperf-udp ~ 0.97 1.03 | 1.02 | ~
netperf-tcp 0.96 ~ ~ | ~ | ~
tbench4 1.24 ~ 1.06 | 1.05 | 1.11
kernbench 0.97 0.97 0.98 | 0.97 | 0.96
gitsource 1.03 1.33 1.32 | 1.32 | 1.33
+------+
These results are overall pleasing: in plenty of cases we observe
performance-per-watt improvements. The few regressions (read/write pgbench and
dbench on the Broadwell machine) are of small magnitude. kernbench loses a few
percentage points (it has a 10-15% performance improvement, but apparently the
increase in power consumption is larger than that). tbench4 and gitsource, which
benefit the most from the patch, keep a positive score in this table which is
a welcome surprise; that suggests that in those particular workloads the
non-invariant schedutil (and active intel_pstate, too) makes some rather
suboptimal frequency selections.
+-------------------------------------------------------------------------+
| 6. MICROARCH'ES ADDRESSED HERE
+-------------------------------------------------------------------------+
The patch addresses Xeon Core processors that use MSR_PLATFORM_INFO and
MSR_TURBO_RATIO_LIMIT to advertise their base frequency and turbo frequencies
respectively. This excludes the recent Xeon Scalable Performance processors
line (Xeon Gold, Platinum etc) whose MSRs have to be parsed differently.
Subsequent patches will address:
* Xeon Scalable Performance processors and Atom Goldmont/Goldmont Plus
* Xeon Phi (Knights Landing, Knights Mill)
* Atom Silvermont
+-------------------------------------------------------------------------+
| 7. REFERENCES
+-------------------------------------------------------------------------+
Tests have been run with the help of the MMTests performance testing
framework, see github.com/gormanm/mmtests. The configuration file names for
the benchmark used are:
db-pgbench-timed-ro-small-xfs
db-pgbench-timed-rw-small-xfs
io-dbench4-async-xfs
network-netperf-unbound
network-tbench
scheduler-unbound
workload-kerndevel-xfs
workload-shellscripts-xfs
hpc-nas-c-class-mpi-full-xfs
hpc-nas-c-class-omp-full
All those benchmarks are generally available on the web:
pgbench: https://www.postgresql.org/docs/10/pgbench.html
netperf: https://hewlettpackard.github.io/netperf/
dbench/tbench: https://dbench.samba.org/
gitsource: git unit test suite, github.com/git/git
NAS Parallel Benchmarks: https://www.nas.nasa.gov/publications/npb.html
hackbench: https://people.redhat.com/mingo/cfs-scheduler/tools/hackbench.c
Suggested-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Giovanni Gherdovich <ggherdovich@suse.cz>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Acked-by: Doug Smythies <dsmythies@telus.net>
Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Link: https://lkml.kernel.org/r/20200122151617.531-2-ggherdovich@suse.cz
|