Improve scheduler performance by flattening nonsensical topology layers
(layers with only one child don't serve any purpose).
This is especially relevant on non-AMD Zen systems after r322776. On my
dual core Intel laptop, this brings the kern.sched.topology_spec table down
from three levels to two.
Submitted by: jeff
Reviewed by: attilio
Sponsored by: Dell EMC Isilon
Rather than repeatedly nesting loops, separate concerns with a single loop
per call stack level. Use a table to drive the recursive routine. Handle
missing topology layers more gracefully (infer a single unit).
Analyze some additional optional layers which may be present on e.g. AMD Zen
systems (groups, aka dies, per package; and cachegroups, aka CCXes, per
group).
Display that additional information in the boot-time topology information,
when it is relevent (non-one).
Reviewed by: markj@, mjoras@ (earlier version)
Sponsored by: Dell EMC Isilon
Differential Revision: https://reviews.freebsd.org/D12019
The MFC will include a compat definition of smp_no_rendevous_barrier()
that calls smp_no_rendezvous_barrier().
Reviewed by: gnn, kib
MFC after: 1 week
Differential Revision: https://reviews.freebsd.org/D10313
On several Intel chipsets, diagnostic NMIs sent from BMC or NMIs
reporting hardware errors are broadcasted to all CPUs.
When kernel is configured to enter kdb on NMI, the outcome is
problematic, because each CPU tries to enter kdb. All CPUs are
executing NMI handlers, which set the latches disabling the nested NMI
delivery; this means that stop_cpus_hard(), used by kdb_enter() to
stop other cpus by broadcasting IPI_STOP_HARD NMI, cannot work. One
indication of this is the harmless but annoying diagnostic "timeout
stopping cpus".
Much more harming behaviour is that because all CPUs try to enter kdb,
and if ddb is used as debugger, all CPUs issue prompt on console and
race for the input, not to mention the simultaneous use of the ddb
shared state.
Try to fix this by introducing a pseudo-lock for simultaneous attempts
to handle NMIs. If one core happens to enter NMI trap handler, other
cores see it and simulate reception of the IPI_STOP_HARD. More,
generic_stop_cpus() avoids sending IPI_STOP_HARD and avoids waiting
for the acknowledgement, relying on the nmi handler on other cores
suspending and then restarting the CPU.
Since it is impossible to detect at runtime whether some stray NMI is
broadcast or unicast, add a knob for administrator (really developer)
to configure debugging NMI handling mode.
The updated patch was debugged with the help from Andrey Gapon (avg)
and discussed with him.
Sponsored by: The FreeBSD Foundation
MFC after: 2 weeks
Differential revision: https://reviews.freebsd.org/D8249
Previously, the code determined a topology of processing units
(hardware threads, cores, packages) and then deduced a cache topology
using certain assumptions. The new code builds a topology that
includes both processing units and caches using the information
provided by the hardware.
At the moment, the discovered full topology is used only to creeate
a scheduling topology for SCHED_ULE.
There is no KPI for other kernel uses.
Summary:
- based on APIC ID derivation rules for Intel and AMD CPUs
- can handle non-uniform topologies
- requires homogeneous APIC ID assignment (same bit widths for ID
components)
- topology for dual-node AMD CPUs may not be optimal
- topology for latest AMD CPU models may not be optimal as the code is
several years old
- supports only thread/package/core/cache nodes
Todo:
- AMD dual-node processors
- latest AMD processors
- NUMA nodes
- checking for homogeneity of the APIC ID assignment across packages
- more flexible cache placement within topology
- expose topology to userland, e.g., via sysctl nodes
Long term todo:
- KPI for CPU sharing and affinity with respect to various resources
(e.g., two logical processors may share the same FPU, etc)
Reviewed by: mav
Tested by: mav
MFC after: 1 month
Differential Revision: https://reviews.freebsd.org/D2728
variable during mp_start() which is too late. Move this to mp_setmaxid()
where other architectures set it and move x86 assertions to MI code.
Reviewed by: kib (x86 part)
done by the functions called on other CPUs, are visible to the caller.
Pair otherwise useless acquire on smp_rv_waiters[3] with a release add
to ensure synchronized with relation, which guarantees visibility.
Reviewed by: alc
Sponsored by: The FreeBSD Foundation
MFC after: 3 weeks
These changes prevent sysctl(8) from returning proper output,
such as:
1) no output from sysctl(8)
2) erroneously returning ENOMEM with tools like truss(1)
or uname(1)
truss: can not get etype: Cannot allocate memory
there is an environment variable which shall initialize the SYSCTL
during early boot. This works for all SYSCTL types both statically and
dynamically created ones, except for the SYSCTL NODE type and SYSCTLs
which belong to VNETs. A new flag, CTLFLAG_NOFETCH, has been added to
be used in the case a tunable sysctl has a custom initialisation
function allowing the sysctl to still be marked as a tunable. The
kernel SYSCTL API is mostly the same, with a few exceptions for some
special operations like iterating childrens of a static/extern SYSCTL
node. This operation should probably be made into a factored out
common macro, hence some device drivers use this. The reason for
changing the SYSCTL API was the need for a SYSCTL parent OID pointer
and not only the SYSCTL parent OID list pointer in order to quickly
generate the sysctl path. The motivation behind this patch is to avoid
parameter loading cludges inside the OFED driver subsystem. Instead of
adding special code to the OFED driver subsystem to post-load tunables
into dynamically created sysctls, we generalize this in the kernel.
Other changes:
- Corrected a possibly incorrect sysctl name from "hw.cbb.intr_mask"
to "hw.pcic.intr_mask".
- Removed redundant TUNABLE statements throughout the kernel.
- Some minor code rewrites in connection to removing not needed
TUNABLE statements.
- Added a missing SYSCTL_DECL().
- Wrapped two very long lines.
- Avoid malloc()/free() inside sysctl string handling, in case it is
called to initialize a sysctl from a tunable, hence malloc()/free() is
not ready when sysctls from the sysctl dataset are registered.
- Bumped FreeBSD version to indicate SYSCTL API change.
MFC after: 2 weeks
Sponsored by: Mellanox Technologies
the cpufreq code. Replace its use with smp_started. There's at least
one userland tool that still looks at the kern.smp.active sysctl, so
preserve it but point it to smp_started as well.
Discussed with: peter, jhb
MFC after: 3 days
Obtained from: Netflix
revision 255744.
sys/kern/subr_smp.c:
IPI_SUSPEND is only available on amd64 and i386. Protect
new uses of this constant with #ifdefs to avoid impacting
other platforms.
Approved by: re (blanket Xen)
amd64 and i386.
Submitted by: Roger Pau Monné
Sponsored by: Citrix Systems R&D
Reviewed by: gibbs
Approved by: re (blanket Xen)
MFC after: 2 weeks
sys/amd64/amd64/mp_machdep.c:
sys/amd64/include/cpu.h:
sys/i386/i386/mp_machdep.c:
sys/i386/include/cpu.h:
- Introduce two new CPU hooks for initialization and resume
purposes. This allows us to get rid of the XENHVM ifdefs in
mp_machdep, and also sets some hooks into common code that can be
used by other hypervisor implementations.
sys/amd64/conf/XENHVM:
sys/i386/conf/XENHVM:
- Remove these configs now that GENERIC has builtin support for Xen
HVM.
sys/kern/subr_smp.c:
- Make sure there are no pending IPIs when suspending a system.
sys/x86/xen/hvm.c:
- Add cpu init and resume vectors that are called from mp_machdep
using the new hooks.
- Only clear the vcpu_info mapping data on resume. It is already
clear for the BSP on a cold boot and is set correctly as APs
are started.
- Gate xen_hvm_init_cpu only to systems running under Xen.
sys/x86/xen/xen_intr.c:
- Gate the setup of event channels only to systems running under Xen.
Xen PVHVM guest.
Submitted by: Roger Pau Monné
Sponsored by: Citrix Systems R&D
Reviewed by: gibbs
Approved by: re (blanket Xen)
MFC after: 2 weeks
sys/amd64/amd64/mp_machdep.c:
sys/i386/i386/mp_machdep.c:
- Make sure that are no MMU related IPIs pending on migration.
- Reset pending IPI_BITMAP on resume.
- Init vcpu_info on resume.
sys/amd64/include/intr_machdep.h:
sys/i386/include/intr_machdep.h:
sys/x86/acpica/acpi_wakeup.c:
sys/x86/x86/intr_machdep.c:
sys/x86/isa/atpic.c:
sys/x86/x86/io_apic.c:
sys/x86/x86/local_apic.c:
- Add a "suspend_cancelled" parameter to pic_resume(). For the
Xen PIC, restoration of interrupt services differs between
the aborted suspend and normal resume cases, so we must provide
this information.
sys/dev/acpica/acpi_timer.c:
sys/dev/xen/timer/timer.c:
sys/timetc.h:
- Don't swap out "suspend safe" timers across a suspend/resume
cycle. This includes the Xen PV and ACPI timers.
sys/dev/xen/control/control.c:
- Perform proper suspend/resume process for PVHVM:
- Suspend all APs before going into suspension, this allows us
to reset the vcpu_info on resume for each AP.
- Reset shared info page and callback on resume.
sys/dev/xen/timer/timer.c:
- Implement suspend/resume support for the PV timer. Since FreeBSD
doesn't perform a per-cpu resume of the timer, we need to call
smp_rendezvous in order to correctly resume the timer on each CPU.
sys/dev/xen/xenpci/xenpci.c:
- Don't reset the PCI interrupt on each suspend/resume.
sys/kern/subr_smp.c:
- When suspending a PVHVM domain make sure there are no MMU IPIs
in-flight, or we will get a lockup on resume due to the fact that
pending event channels are not carried over on migration.
- Implement a generic version of restart_cpus that can be used by
suspended and stopped cpus.
sys/x86/xen/hvm.c:
- Implement resume support for the hypercall page and shared info.
- Clear vcpu_info so it can be reset by APs when resuming from
suspension.
sys/dev/xen/xenpci/xenpci.c:
sys/x86/xen/hvm.c:
sys/x86/xen/xen_intr.c:
- Support UP kernel configurations.
sys/x86/xen/xen_intr.c:
- Properly rebind per-cpus VIRQs and IPIs on resume.
- Implement a function to ensure that all preempted threads have switched
back out at least once. Use this to make sure there are no stale
references to the old ktr_buf or the lock profiling buffers before
updating them.
Reviewed by: marius (sparc64 parts), attilio (earlier patch)
Sponsored by: EMC / Isilon Storage Division
suspend/resume procedures are minimized among them.
common:
- Add global cpuset suspended_cpus to indicate APs are suspended/resumed.
- Remove acpi_waketag and acpi_wakemap from acpivar.h (no longer used).
- Add some variables in acpi_wakecode.S in order to minimize the difference
among amd64 and i386.
- Disable load_cr3() because now CR3 is restored in resumectx().
amd64:
- Add suspend/resume related members (such as MSR) in PCB.
- Modify savectx() for above new PCB members.
- Merge acpi_switch.S into cpu_switch.S as resumectx().
i386:
- Merge(and remove) suspendctx() into savectx() in order to match with
amd64 code.
Reviewed by: attilio@, acpi@
Most part is merged from amd64.
- i386/acpica/acpi_wakecode.S
Replaced with amd64 code (from realmode to paging enabling code).
- i386/acpica/acpi_wakeup.c
Replaced with amd64 code (except for wakeup_pagetables stuff).
- i386/include/pcb.h
- i386/i386/genassym.c
Added PCB new members (CR0, CR2, CR4, DS, ED, FS, SS, GDT, IDT, LDT
and TR) needed for suspend/resume, not for context switch.
- i386/i386/swtch.s
Added suspendctx() and resumectx().
Note that savectx() was not changed and used for suspending (while
amd64 code uses it).
BSP and AP execute the same sequence, suspendctx(), acpi_wakecode()
and resumectx() for suspend/resume (in case of UP system also).
- i386/i386/apic_vector.s
Added cpususpend().
- i386/i386/mp_machdep.c
- i386/include/smp.h
Added cpususpend_handler().
- i386/include/apicvar.h
- kern/subr_smp.c
- sys/smp.h
Added IPI_SUSPEND and suspend_cpus().
- i386/i386/initcpu.c
- i386/i386/machdep.c
- i386/include/md_var.h
- pc98/pc98/machdep.c
Moved initializecpu() declarations to md_var.h.
MFC after: 3 days
The SYSCTL_NODE macro defines a list that stores all child-elements of
that node. If there's no SYSCTL_DECL macro anywhere else, there's no
reason why it shouldn't be static.
UP/!SMP case.
The callbacks may be relying on this feature and having 2 different
ways to deal with them is not correct.
Reported by: rstone
Reviewed by: jhb
MFC after: 2 weeks
This is a followup to r222032 and a reimplementation of it.
While that revision fixed the race for the smp_rv_waiters[2] exit
sentinel, it still left a possibility for a target CPU to access
stale or wrong smp_rv_func_arg in smp_rv_teardown_func.
To fix this race the slave CPUs signal when they are really fully
done with the rendezvous and the master CPU waits until all slaves
are done.
Diagnosed by: kib
Reviewed by: jhb, mlaier, neel
Approved by: re (kib)
MFC after: 2 weeks
may be jointly referenced via the mask CTLFLAG_CAPRW. Sysctls with these
flags are available in Capsicum's capability mode; other sysctl nodes are
not.
Flag several useful sysctls as available in capability mode, such as memory
layout sysctls required by the run-time linker and malloc(3). Also expose
access to randomness and available kernel features.
A few sysctls are enabled to support name->MIB conversion; these may leak
information to capability mode by virtue of providing resolution on names
not flagged for access in capability mode. This is, generally, not a huge
problem, but might be something to resolve in the future. Flag these cases
with XXX comments.
Submitted by: jonathan
Sponsored by: Google, Inc.
... and also increase the timeout.
It's better to try to proceed somehow despite stuck CPUs than to hang
indefinitely. Especially so during shutdown and when entering kdb or panic.
Timeout value is still an aribitrary value.
Timeout diagnostic is just a printf; the work on something more
debuggable is planned by attilio. Need to be careful here as
stop_cpus_hard is called very early while enetering kdb and soon(-ish)
it may become called very early when entering panic.
Reviewed by: attilio
MFC after: 2 months
Specifically, a critical_exit() call that drops the nesting level to zero
has a brief window where the pending preemption flag is set and the
nesting level is set to zero. This is done purposefully to avoid races
where a preemption scheduled by an interrupt could be lost otherwise (see
revision 144777). However, this does mean that if an interrupt fires
during this window and enters and exits a critical section, it may preempt
from the interrupt context. This is generally fine as the interrupt code
is careful to arrange critical sections so that they are not exited until
it is safe to preempt (e.g. interrupts EOI'd and masked if necessary).
However, the SMP rendezvous IPI handler does not quite follow this rule,
and in general a rendezvous can never be preempted. Rendezvous handlers
are also not permitted to schedule threads to execute, so they will not
typically trigger preemptions. SMP rendezvous handlers may use
spinlocks (carefully) such as the rm_cleanIPI() handler used in rmlocks,
but using a spinlock also enters and exits a critical section. If the
interrupted top-half code is in the brief window of critical_exit() where
the nesting level is zero but a preemption is pending, then releasing the
spinlock can trigger a preemption. Because we know that SMP rendezvous
handlers can never schedule a thread, we know that a critical_exit() in
an SMP rendezvous handler will only preempt in this edge case. We also
know that the top-half thread will happily handle the deferred preemption
once the SMP rendezvous has completed, so the preemption will not be lost.
This makes it safe to employ a workaround where we use a nested critical
section in the SMP rendezvous code itself around rendezvous action
routines to prevent any preemptions during an SMP rendezvous. The
workaround intentionally avoids checking for a deferred preemption
when leaving the critical section on the assumption that if there is a
pending preemption it will be handled by the interrupted top-half code.
Submitted by: mlaier (variation specific to rm_cleanIPI())
Obtained from: Isilon
MFC after: 1 week
last CPU to to finish the rendezvous action may become visible to
different CPUs at different times. As a result, the CPU that initiated
the rendezvous may exit the rendezvous and drop the lock allowing another
rendezvous to be initiated on the same CPU or a different CPU. In that
case the exit sentinel may be cleared before all CPUs have noticed causing
those CPUs to hang forever.
Workaround this by using a generation count to notice when this race
occurs and to exit the rendezvous in that case.
The problem was independently diagnosted by mlaier@ and avg@ as well.
Submitted by: neel
Reviewed by: avg, mlaier
Obtained from: NetApp
MFC after: 1 week
cpuset_t objects.
That is going to offer the underlying support for a simple bump of
MAXCPU and then support for number of cpus > 32 (as it is today).
Right now, cpumask_t is an int, 32 bits on all our supported architecture.
cpumask_t on the other side is implemented as an array of longs, and
easilly extendible by definition.
The architectures touched by this commit are the following:
- amd64
- i386
- pc98
- arm
- ia64
- XEN
while the others are still missing.
Userland is believed to be fully converted with the changes contained
here.
Some technical notes:
- This commit may be considered an ABI nop for all the architectures
different from amd64 and ia64 (and sparc64 in the future)
- per-cpu members, which are now converted to cpuset_t, needs to be
accessed avoiding migration, because the size of cpuset_t should be
considered unknown
- size of cpuset_t objects is different from kernel and userland (this is
primirally done in order to leave some more space in userland to cope
with KBI extensions). If you need to access kernel cpuset_t from the
userland please refer to example in this patch on how to do that
correctly (kgdb may be a good source, for example).
- Support for other architectures is going to be added soon
- Only MAXCPU for amd64 is bumped now
The patch has been tested by sbruno and Nicholas Esborn on opteron
4 x 12 pack CPUs. More testing on big SMP is expected to came soon.
pluknet tested the patch with his 8-ways on both amd64 and i386.
Tested by: pluknet, sbruno, gianni, Nicholas Esborn
Reviewed by: jeff, jhb, sbruno
This is based on the same approach as used in panic().
In theory parallel execution of generic_stop_cpus() could lead to two CPUs
stopping each other and everyone else, and thus a total system halt.
Also, in theory, we should have some smarter locking here, because two
(or more CPUs) could be stopping unrelated sets of CPUs.
But in practice, it seems, this function is only used to stop
"all other" CPUs.
Additionally, I took this opportunity to make amd64-specific suspend_cpus()
function use generic_stop_cpus() instead of rolling out essentially
duplicate code.
This code is based on code by Sandvine Incorporated.
Suggested by: mdf
Reviewed by: jhb, jkim (earlier version)
MFC after: 2 weeks
number of CPUs detection.
However, that was not mention at all, the problem was not reported, the
patch has not been MFCed and the fix is mostly improper.
Fix the original overflow (caused when 32 CPUs must be detected) by
just using a different mathematical computation (it also makes more
explicit the size of operands involved, which is good in the moment
waiting for a more complete support for a large number of CPUs).
PR: kern/148698
Submitted by: Joe Landers <jlanders at vmware dot com>
Tested by: gianni
MFC after: 10 days
IPI to a specific CPU by its cpuid. Replace calls to ipi_selected() that
constructed a mask for a single CPU with calls to ipi_cpu() instead. This
will matter more in the future when we transition from cpumask_t to
cpuset_t for CPU masks in which case building a CPU mask is more expensive.
Submitted by: peter, sbruno
Reviewed by: rookie
Obtained from: Yahoo! (x86)
MFC after: 1 month
am now able to run 32 cores ok.. but I still will hang
on buildworld with a NFS problem. I suspect I am missing
a patch for the netlogic rge driver.
JC check and see if I am missing anything except your
core-mask changes
Obtained from: JC
has proven to have a good effect when entering KDB by using a NMI,
but it completely violates all the good rules about interrupts
disabled while holding a spinlock in other occasions. This can be the
cause of deadlocks on events where a normal IPI_STOP is expected.
* Adds an new IPI called IPI_STOP_HARD on all the supported architectures.
This IPI is responsible for sending a stop message among CPUs using a
privileged channel when disponible. In other cases it just does match a
normal IPI_STOP.
Right now the IPI_STOP_HARD functionality uses a NMI on ia32 and amd64
architectures, while on the other has a normal IPI_STOP effect. It is
responsibility of maintainers to eventually implement an hard stop
when necessary and possible.
* Use the new IPI facility in order to implement a new userend SMP kernel
function called stop_cpus_hard(). That is specular to stop_cpu() but
it does use the privileged channel for the stopping facility.
* Let KDB use the newly introduced function stop_cpus_hard() and leave
stop_cpus() for all the other cases
* Disable interrupts on CPU0 when starting the process of APs suspension.
* Style cleanup and comments adding
This patch should fix the reboot/shutdown deadlocks many users are
constantly reporting on mailing lists.
Please don't forget to update your config file with the STOP_NMI
option removal
Reviewed by: jhb
Tested by: pho, bz, rink
Approved by: re (kib)
