RT tasks by default run at the highest capacity/performance level. When
uclamp is selected this default behavior is retained by enforcing the
requested uclamp.min (p->uclamp_req[UCLAMP_MIN]) of the RT tasks to be
uclamp_none(UCLAMP_MAX), which is SCHED_CAPACITY_SCALE; the maximum
value.
This is also referred to as 'the default boost value of RT tasks'.
See commit 1a00d99997 ("sched/uclamp: Set default clamps for RT tasks").
On battery powered devices, it is desired to control this default
(currently hardcoded) behavior at runtime to reduce energy consumed by
RT tasks.
For example, a mobile device manufacturer where big.LITTLE architecture
is dominant, the performance of the little cores varies across SoCs, and
on high end ones the big cores could be too power hungry.
Given the diversity of SoCs, the new knob allows manufactures to tune
the best performance/power for RT tasks for the particular hardware they
run on.
They could opt to further tune the value when the user selects
a different power saving mode or when the device is actively charging.
The runtime aspect of it further helps in creating a single kernel image
that can be run on multiple devices that require different tuning.
Keep in mind that a lot of RT tasks in the system are created by the
kernel. On Android for instance I can see over 50 RT tasks, only
a handful of which created by the Android framework.
To control the default behavior globally by system admins and device
integrator, introduce the new sysctl_sched_uclamp_util_min_rt_default
to change the default boost value of the RT tasks.
I anticipate this to be mostly in the form of modifying the init script
of a particular device.
To avoid polluting the fast path with unnecessary code, the approach
taken is to synchronously do the update by traversing all the existing
tasks in the system. This could race with a concurrent fork(), which is
dealt with by introducing sched_post_fork() function which will ensure
the racy fork will get the right update applied.
Tested on Juno-r2 in combination with the RT capacity awareness [1].
By default an RT task will go to the highest capacity CPU and run at the
maximum frequency, which is particularly energy inefficient on high end
mobile devices because the biggest core[s] are 'huge' and power hungry.
With this patch the RT task can be controlled to run anywhere by
default, and doesn't cause the frequency to be maximum all the time.
Yet any task that really needs to be boosted can easily escape this
default behavior by modifying its requested uclamp.min value
(p->uclamp_req[UCLAMP_MIN]) via sched_setattr() syscall.
[1] 804d402fb6: ("sched/rt: Make RT capacity-aware")
Signed-off-by: Qais Yousef <qais.yousef@arm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20200716110347.19553-2-qais.yousef@arm.com
The following splat was caught when setting uclamp value of a task:
BUG: sleeping function called from invalid context at ./include/linux/percpu-rwsem.h:49
cpus_read_lock+0x68/0x130
static_key_enable+0x1c/0x38
__sched_setscheduler+0x900/0xad8
Fix by ensuring we enable the key outside of the critical section in
__sched_setscheduler()
Fixes: 46609ce227 ("sched/uclamp: Protect uclamp fast path code with static key")
Signed-off-by: Qais Yousef <qais.yousef@arm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20200716110347.19553-4-qais.yousef@arm.com
The following commit:
14533a16c4 ("thermal/cpu-cooling, sched/core: Move the arch_set_thermal_pressure() API to generic scheduler code")
moved the definition of arch_set_thermal_pressure() to sched/core.c, but
kept its declaration in linux/arch_topology.h. When building e.g. an x86
kernel with CONFIG_SCHED_THERMAL_PRESSURE=y, cpufreq_cooling.c ends up
getting the declaration of arch_set_thermal_pressure() from
include/linux/arch_topology.h, which is somewhat awkward.
On top of this, sched/core.c unconditionally defines
o The thermal_pressure percpu variable
o arch_set_thermal_pressure()
while arch_scale_thermal_pressure() does nothing unless redefined by the
architecture.
arch_*() functions are meant to be defined by architectures, so revert the
aforementioned commit and re-implement it in a way that keeps
arch_set_thermal_pressure() architecture-definable, and doesn't define the
thermal pressure percpu variable for kernels that don't need
it (CONFIG_SCHED_THERMAL_PRESSURE=n).
Signed-off-by: Valentin Schneider <valentin.schneider@arm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20200712165917.9168-2-valentin.schneider@arm.com
In slow path, when selecting idlest group, if both groups have type
group_has_spare, only idle_cpus count gets compared.
As a result, if multiple tasks are created in a tight loop,
and go back to sleep immediately
(while waiting for all tasks to be created),
they may be scheduled on the same core, because CPU is back to idle
when the new fork happen.
For example:
sudo perf record -e sched:sched_wakeup_new -- \
sysbench threads --threads=4 run
...
total number of events: 61582
...
sudo perf script
sysbench 129378 [006] 74586.633466: sched:sched_wakeup_new:
sysbench:129380 [120] success=1 CPU:007
sysbench 129378 [006] 74586.634718: sched:sched_wakeup_new:
sysbench:129381 [120] success=1 CPU:007
sysbench 129378 [006] 74586.635957: sched:sched_wakeup_new:
sysbench:129382 [120] success=1 CPU:007
sysbench 129378 [006] 74586.637183: sched:sched_wakeup_new:
sysbench:129383 [120] success=1 CPU:007
This may have negative impact on performance for workloads with frequent
creation of multiple threads.
In this patch we are using group_util to select idlest group if both groups
have equal number of idle_cpus. Comparing the number of idle cpu is
not enough in this case, because the newly forked thread sleeps
immediately and before we select the cpu for the next one.
This is shown in the trace where the same CPU7 is selected for
all wakeup_new events.
That's why, looking at utilization when there is the same number of
CPU is a good way to see where the previous task was placed. Using
nr_running doesn't solve the problem because the newly forked task is not
running and the cpu would not have been idle in this case and an idle
CPU would have been selected instead.
With this patch newly created tasks would be better distributed.
With this patch:
sudo perf record -e sched:sched_wakeup_new -- \
sysbench threads --threads=4 run
...
total number of events: 74401
...
sudo perf script
sysbench 129455 [006] 75232.853257: sched:sched_wakeup_new:
sysbench:129457 [120] success=1 CPU:008
sysbench 129455 [006] 75232.854489: sched:sched_wakeup_new:
sysbench:129458 [120] success=1 CPU:009
sysbench 129455 [006] 75232.855732: sched:sched_wakeup_new:
sysbench:129459 [120] success=1 CPU:010
sysbench 129455 [006] 75232.856980: sched:sched_wakeup_new:
sysbench:129460 [120] success=1 CPU:011
We tested this patch with following benchmarks:
master: 'commit b3a9e3b962 ("Linux 5.8-rc1")'
100 iterations of: perf bench -f simple futex wake -s -t 128 -w 1
Lower result is better
| | BASELINE | +PATCH | DELTA (%) |
|---------|------------|----------|-------------|
| mean | 0.33 | 0.313 | +5.152 |
| std (%) | 10.433 | 7.563 | |
100 iterations of: sysbench threads --threads=8 run
Higher result is better
| | BASELINE | +PATCH | DELTA (%) |
|---------|------------|----------|-------------|
| mean | 5235.02 | 5863.73 | +12.01 |
| std (%) | 8.166 | 10.265 | |
100 iterations of: sysbench mutex --mutex-num=1 --threads=8 run
Lower result is better
| | BASELINE | +PATCH | DELTA (%) |
|---------|------------|----------|-------------|
| mean | 0.413 | 0.404 | +2.179 |
| std (%) | 3.791 | 1.816 | |
Signed-off-by: Peter Puhov <peter.puhov@linaro.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20200714125941.4174-1-peter.puhov@linaro.org
There is apparently one site that violates the rule that only current
and ttwu() will modify task->state, namely ptrace_{,un}freeze_traced()
will change task->state for a remote task.
Oleg explains:
"TASK_TRACED/TASK_STOPPED was always protected by siglock. In
particular, ttwu(__TASK_TRACED) must be always called with siglock
held. That is why ptrace_freeze_traced() assumes it can safely do
s/TASK_TRACED/__TASK_TRACED/ under spin_lock(siglock)."
This breaks the ordering scheme introduced by commit:
dbfb089d36 ("sched: Fix loadavg accounting race")
Specifically, the reload not matching no longer implies we don't have
to block.
Simply things by noting that what we need is a LOAD->STORE ordering
and this can be provided by a control dependency.
So replace:
prev_state = prev->state;
raw_spin_lock(&rq->lock);
smp_mb__after_spinlock(); /* SMP-MB */
if (... && prev_state && prev_state == prev->state)
deactivate_task();
with:
prev_state = prev->state;
if (... && prev_state) /* CTRL-DEP */
deactivate_task();
Since that already implies the 'prev->state' load must be complete
before allowing the 'prev->on_rq = 0' store to become visible.
Fixes: dbfb089d36 ("sched: Fix loadavg accounting race")
Reported-by: Jiri Slaby <jirislaby@kernel.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Oleg Nesterov <oleg@redhat.com>
Tested-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Tested-by: Christian Brauner <christian.brauner@ubuntu.com>
Pull perf tooling fixes from Arnaldo Carvalho de Melo:
- Update hashmap.h from libbpf and kvm.h from x86's kernel UAPI.
- Set opt->set in libsubcmd's OPT_CALLBACK_SET(). This fixes
'perf record --switch-output-event event-name' usage"
* tag 'perf-tools-fixes-2020-07-19' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux:
tools arch kvm: Sync kvm headers with the kernel sources
perf tools: Sync hashmap.h with libbpf's
libsubcmd: Fix OPT_CALLBACK_SET()
Pull x86 fixes from Thomas Gleixner:
"A pile of fixes for x86:
- Fix the I/O bitmap invalidation on XEN PV, which was overlooked in
the recent ioperm/iopl rework. This caused the TSS and XEN's I/O
bitmap to get out of sync.
- Use the proper vectors for HYPERV.
- Make disabling of stack protector for the entry code work with GCC
builds which enable stack protector by default. Removing the option
is not sufficient, it needs an explicit -fno-stack-protector to
shut it off.
- Mark check_user_regs() noinstr as it is called from noinstr code.
The missing annotation causes it to be placed in the text section
which makes it instrumentable.
- Add the missing interrupt disable in exc_alignment_check()
- Fixup a XEN_PV build dependency in the 32bit entry code
- A few fixes to make the Clang integrated assembler happy
- Move EFI stub build to the right place for out of tree builds
- Make prepare_exit_to_usermode() static. It's not longer called from
ASM code"
* tag 'x86-urgent-2020-07-19' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
x86/boot: Don't add the EFI stub to targets
x86/entry: Actually disable stack protector
x86/ioperm: Fix io bitmap invalidation on Xen PV
x86: math-emu: Fix up 'cmp' insn for clang ias
x86/entry: Fix vectors to IDTENTRY_SYSVEC for CONFIG_HYPERV
x86/entry: Add compatibility with IAS
x86/entry/common: Make prepare_exit_to_usermode() static
x86/entry: Mark check_user_regs() noinstr
x86/traps: Disable interrupts in exc_aligment_check()
x86/entry/32: Fix XEN_PV build dependency
Pull timer fixes from Thomas Gleixner:
"Two fixes for the timer wheel:
- A timer which is already expired at enqueue time can set the
base->next_expiry value backwards. As a consequence base->clk can
be set back as well. This can lead to timers expiring early. Add a
sanity check to prevent this.
- When a timer is queued with an expiry time beyond the wheel
capacity then it should be queued in the bucket of the last wheel
level which is expiring last.
The code adjusted the expiry time to the maximum wheel capacity,
which is only correct when the wheel clock is 0. Aside of that the
check whether the delta is larger than wheel capacity does not
check the delta, it checks the expiry value itself. As a result
timers can expire at random.
Fix this by checking the right variable and adjust expiry time so
it becomes base->clock plus capacity which places it into the
outmost bucket in the last wheel level"
* tag 'timers-urgent-2020-07-19' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
timer: Fix wheel index calculation on last level
timer: Prevent base->clk from moving backward
Pull scheduler fixes from Thomas Gleixner:
"A set of scheduler fixes:
- Plug a load average accounting race which was introduced with a
recent optimization casing load average to show bogus numbers.
- Fix the rseq CPU id initialization for new tasks. sched_fork() does
not update the rseq CPU id so the id is the stale id of the parent
task, which can cause user space data corruption.
- Handle a 0 return value of task_h_load() correctly in the load
balancer, which does not decrease imbalance and therefore pulls
until the maximum number of loops is reached, which might be all
tasks just created by a fork bomb"
* tag 'sched-urgent-2020-07-19' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
sched/fair: handle case of task_h_load() returning 0
sched: Fix unreliable rseq cpu_id for new tasks
sched: Fix loadavg accounting race
Pull irq fixes from Thomas Gleixner:
"Two fixes for the interrupt subsystem:
- Make the handling of the firmware node consistent and do not free
the node after the domain has been created successfully. The core
code stores a pointer to it which can lead to a use after free or
double free.
This used to "work" because the pointer was not stored when the
initial code was written, but at some point later it was required
to store it. Of course nobody noticed that the existing users break
that way.
- Handle affinity setting on inactive interrupts correctly when
hierarchical irq domains are enabled.
When interrupts are inactive with the modern hierarchical irqdomain
design, the interrupt chips are not necessarily in a state where
affinity changes can be handled. The legacy irq chip design allowed
this because interrupts are immediately fully initialized at
allocation time. X86 has a hacky workaround for this, but other
implementations do not.
This cased malfunction on GIC-V3. Instead of playing whack a mole
to find all affected drivers, change the core code to store the
requested affinity setting and then establish it when the interrupt
is allocated, which makes the X86 hack go away"
* tag 'irq-urgent-2020-07-19' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
genirq/affinity: Handle affinity setting on inactive interrupts correctly
irqdomain/treewide: Keep firmware node unconditionally allocated
Pull USB fixes from Greg KH:
"Here are a few small USB fixes, and one thunderbolt fix, for 5.8-rc6.
Nothing huge in here, just the normal collection of gadget, dwc2/3,
serial, and other minor USB driver fixes and id additions. Full
details are in the shortlog.
All of these have been in linux-next for a while with no reported
issues"
* tag 'usb-5.8-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb:
USB: serial: iuu_phoenix: fix memory corruption
USB: c67x00: fix use after free in c67x00_giveback_urb
usb: gadget: function: fix missing spinlock in f_uac1_legacy
usb: gadget: udc: atmel: fix uninitialized read in debug printk
usb: gadget: udc: atmel: remove outdated comment in usba_ep_disable()
usb: dwc2: Fix shutdown callback in platform
usb: cdns3: trace: fix some endian issues
usb: cdns3: ep0: fix some endian issues
usb: gadget: udc: gr_udc: fix memleak on error handling path in gr_ep_init()
usb: gadget: fix langid kernel-doc warning in usbstring.c
usb: dwc3: pci: add support for the Intel Jasper Lake
usb: dwc3: pci: add support for the Intel Tiger Lake PCH -H variant
usb: chipidea: core: add wakeup support for extcon
USB: serial: option: add Quectel EG95 LTE modem
thunderbolt: Fix path indices used in USB3 tunnel discovery
USB: serial: ch341: add new Product ID for CH340
USB: serial: option: add GosunCn GM500 series
USB: serial: cypress_m8: enable Simply Automated UPB PIM
Pull dma-mapping fixes from Christoph Hellwig:
"Ensure we always have fully addressable memory in the dma coherent
pool (Nicolas Saenz Julienne)"
* tag 'dma-mapping-5.8-6' of git://git.infradead.org/users/hch/dma-mapping:
dma-pool: do not allocate pool memory from CMA
dma-pool: make sure atomic pool suits device
dma-pool: introduce dma_guess_pool()
dma-pool: get rid of dma_in_atomic_pool()
dma-direct: provide function to check physical memory area validity
vmlinux-objs-y is added to targets, which currently means that the EFI
stub gets added to the targets as well. It shouldn't be added since it
is built elsewhere.
This confuses Makefile.build which interprets the EFI stub as a target
$(obj)/$(objtree)/drivers/firmware/efi/libstub/lib.a
and will create drivers/firmware/efi/libstub/ underneath
arch/x86/boot/compressed, to hold this supposed target, if building
out-of-tree. [0]
Fix this by pulling the stub out of vmlinux-objs-y into efi-obj-y.
[0] See scripts/Makefile.build near the end:
# Create directories for object files if they do not exist
Signed-off-by: Arvind Sankar <nivedita@alum.mit.edu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Masahiro Yamada <masahiroy@kernel.org>
Acked-by: Ard Biesheuvel <ardb@kernel.org>
Link: https://lkml.kernel.org/r/20200715032631.1562882-1-nivedita@alum.mit.edu
Qais Yousef
Miaohe Lin
王文虎
Qinglang Miao
Valentin Schneider
Randy Dunlap
Muchun Song
Peter Puhov
Paul Gortmaker
Peter Zijlstra
Linus Torvalds
Arvind Sankar