Currently, Application Processors (non-boot CPUs) are started by
MD code at SI_SUB_CPU, but they are kept waiting in a "pen" until
SI_SUB_SMP at which point they are released to run kernel threads.
SI_SUB_SMP is one of the last SYSINIT levels, so APs don't enter
the scheduler and start running threads until fairly late in the
boot.
This change moves SI_SUB_SMP up to just before software interrupt
threads are created allowing the APs to start executing kernel
threads much sooner (before any devices are probed). This allows
several initialization routines that need to perform initialization
on all CPUs to now perform that initialization in one step rather
than having to defer the AP initialization to a second SYSINIT run
at SI_SUB_SMP. It also permits all CPUs to be available for
handling interrupts before any devices are probed.
This last feature fixes a problem on with interrupt vector exhaustion.
Specifically, in the old model all device interrupts were routed
onto the boot CPU during boot. Later after the APs were released at
SI_SUB_SMP, interrupts were redistributed across all CPUs.
However, several drivers for multiqueue hardware allocate N interrupts
per CPU in the system. In a system with many CPUs, just a few drivers
doing this could exhaust the available pool of interrupt vectors on
the boot CPU as each driver was allocating N * mp_ncpu vectors on the
boot CPU. Now, drivers will allocate interrupts on their desired CPUs
during boot meaning that only N interrupts are allocated from the boot
CPU instead of N * mp_ncpu.
Some other bits of code can also be simplified as smp_started is
now true much earlier and will now always be true for these bits of
code. This removes the need to treat the single-CPU boot environment
as a special case.
As a transition aid, the new behavior is available under a new kernel
option (EARLY_AP_STARTUP). This will allow the option to be turned off
if need be during initial testing. I plan to enable this on x86 by
default in a followup commit in the next few days and to have all
platforms moved over before 11.0. Once the transition is complete,
the option will be removed along with the !EARLY_AP_STARTUP code.
These changes have only been tested on x86. Other platform maintainers
are encouraged to port their architectures over as well. The main
things to check for are any uses of smp_started in MD code that can be
simplified and SI_SUB_SMP SYSINITs in MD code that can be removed in
the EARLY_AP_STARTUP case (e.g. the interrupt shuffling).
PR: kern/199321
Reviewed by: markj, gnn, kib
Sponsored by: Netflix
Arguably we should only be doing the probe/attach to children of
these devices as well.
Tested by: Michal Stanek <mst_semihalf.com> (arm64)
Differential Revision: https://reviews.freebsd.org/D6133
Both of the callers were expecting the input cap_set to be modified.
This fixes them to request cap_set to be updated with the returned buffer.
Reviewed by: jkim
Differential Revision: https://reviews.freebsd.org/D6040
Eventually with earlier AP startup this code will change to call the
startup function synchronously instead of queueing the task. Moving
the time we queue the task should be a no-op since taskqueue threads
don't start executing tasks until much later, but this reduces the diff
with the earlier AP startup patches.
Sponsored by: Netflix
This wrapper does not translate errors in the first word to ACPI
error status returns. Use this wrapper in the acpi_cpu(4) driver in
place of the existing _OSC code. While here, fix a bug where the wrong
count of words was passed when invoking _OSC.
Reviewed by: jkim
MFC after: 2 weeks
Differential Revision: https://reviews.freebsd.org/D6022
drivers, one for fdt, one for acpi. It then uses this to decide if it will
use fdt or acpi.
The GICv2 (interrupt controller) and Generic Timer drivers have been
updated to handle both cases.
As this is early code we still need FDT to find the kernel console, and
some parts are still missing, including PCI support.
Differential Revision: https://reviews.freebsd.org/D2463
Reviewed by: jhb, jkim, emaste
Obtained from: ABT Systems Ltd
Relnotes: Yes
Sponsored by: The FreeBSD Foundation
bridges only supported Intel Pentium and Pentium II era processors and there
is no reason for hardware virtualizations to emulate these quirks.
MFC after: 1 week
interacts with interrupts, query ACPI and use MWAIT for entrance into
Cx sleep states. Support C1 "I/O then halt" mode. See Intel'
document 302223-007 "Intelб╝ Processor Vendor-Specific ACPI Interface
Specification" for description.
Move the acpi_cpu_c1() function into x86/cpu_machdep.c and use
it instead of inlining "sti; hlt" sequence in several places.
In the acpi(4) man page, besides documenting the dev.cpu.N.cx_methods
sysctl, correct the names for dev.cpu.N.{cx_usage,cx_lowest,cx_supported}
sysctls.
Both jkim and avg have some other patches implementing the mwait
functionality; this work is unrelated. Linux does not rely on the
ACPI to provide correct tables describing Cx modes. Instead, the
driver has pre-defined knowledge of the CPU models, it was supplied by
Intel.
Tested by: pho (previous versions)
Sponsored by: The FreeBSD Foundation
accidentally enable non-existent states.
This bug was triggered if ACPI advertises the presence of a C2 state
which we fail to parse via acpi_PkgGas due to our lack of support for
FFixedHW resources, and causes an immediate panic when an attempt is
made to enter the (NULL) state.
One affected platform is the EC2 c4.8xlarge VM instance type; there
may be others.
MFC after: 1 week
Thanks to: jkim, @_msw_
may also halt in C2 and not just C3 (it seems that in some cases the BIOS
advertises its C3 state as a C2 state in _CST). Just play it safe and
disable both C2 and C3 states if a user forces the use of the TSC as the
timecounter on such CPUs.
PR: 192316
Differential Revision: https://reviews.freebsd.org/D1441
No objection from: jkim
MFC after: 1 week
Also disable a couple of ACPI devices that are not usable under Dom0.
To this end a couple of booleans are added that allow disabling ACPI
specific devices.
Sponsored by: Citrix Systems R&D
Reviewed by: jhb
x86/xen/xen_nexus.c:
- Return BUS_PROBE_SPECIFIC in the Xen Nexus attachement routine to
force the usage of the Xen Nexus.
- Attach the ACPI bus when running as Dom0.
dev/acpica/acpi_cpu.c:
dev/acpica/acpi_hpet.c:
dev/acpica/acpi_timer.c
- Add a variable that gates the addition of the devices.
x86/include/init.h:
- Declare variables that control the attachment of ACPI cpu, hpet and
timer devices.
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
that are being done by the OS.
For now this'll match up with the "wakeups"; although I'll dig deeper into
this to see if we can determine which sleep state the CPU managed to get
into. Most things I've seen these days only expose up to C2 or C3 via
ACPI even though the CPU goes all the way down to C6 or C7.
When CPU becomes idle, cpu_idleclock() calculates time to the next timer
event in order to reprogram hw timer. Return that time in sbintime_t to
the caller and pass it to acpi_cpu_idle(), where it can be used as one
more factor (quite precise) to extimate furter sleep time and choose
optimal sleep state. This is a preparatory change for further callout
improvements will be committed in the next days.
The commmit is not targeted for MFC.
... to avoid any races or inconsistencies.
This should fix a regression introduced in r243404.
Also, remove a stale comment that has not been true for quite a while
now.
Pointyhat to: avg
Teested by: trociny, emaste, dumbbell (earlier version)
MFC after: 1 week
... instead of the ever increasing ones.
Also, do free old resources when allocating new ones when cx states
change.
Tested by: Tom Lislegaard <Tom.Lislegaard@proact.no>
Obtained from: jkim
MFC after: 1 week
For C1 and C2 states use cpu_ticks() to measure sleep time instead of much
slower ACPI timer. We can't do it for C3, as TSC may stop there. But it is
less important there as wake up latency is high any way.
For C1 and C2 states do not check/clear bus mastering activity status, as
it is important only for C3. As side effect it can make CPU enter C2 instead
of C3 if last BM activity was two sleeps back (unlike one before), but
that may be even good because of collecting more statistics. Premature BM
wakeup from C3, entered because of overestimation, can easily be worse then
entering C2 from both performance and power consumption points of view.
Together on dual Xeon E5645 system on sequential 512 bytes read test this
change makes cpu_idle_acpi() as fast as simplest cpu_idle_hlt() and only
few percents slower then cpu_idle_mwait(), while deeper states are still
actively used during idle periods.
To help with diagnostics, add C-state type into dev.cpu.X.cx_supported.
Sponsored by: iXsystems, Inc.
... from a user-set persistent limit on the said level.
Allow to set the user-imposed limit below current deepest available level
as the available levels may be dynamically changed by ACPI platform
in both directions.
Allow "Cmax" as an input value for cx_lowest sysctls to mean that there
is not limit and OS can use all available C-states.
Retire global cpu_cx_count as it no longer serves any meaningful
purpose.
Reviewed by: jhb, gianni, sbruno
Tested by: sbruno, Vitaly Magerya <vmagerya@gmail.com>
MFC after: 2 weeks
although by default only C1 is enabled (cx_lowest=0) and enabling deeper
states goes through acpi_cpu_set_cx_lowest which re-evaluates cpu_non_c3
MFC after: 2 weeks
cpu_non_c3 is already evaluated in acpi_cpu_cx_cst and in
acpi_cpu_set_cx_lowest.
Besides acpi_cpu_cx_list is not protected by any locking.
As a result also move setting of cpu_can_deep_sleep to more appropriate
places.
MFC after: 2 weeks
Adjust power_profile script to handle the new world order as well.
Some vendors are opting out of a C2 state and only defining C1 & C3. This
leads the acpi_cpu display to indicate that the machine supports C1 & C2
which is caused by the (mis)use of the index of the cx_state array as the
ACPI_STATE_CX value.
e.g. the code was pretending that cx_state[i] would
always convert to i by subtracting 1.
cx_state[2] == ACPI_STATE_C3
cx_state[1] == ACPI_STATE_C2
cx_state[0] == ACPI_STATE_C1
however, on certain machines this would lead to
cx_state[1] == ACPI_STATE_C3
cx_state[0] == ACPI_STATE_C1
This didn't break anything but led to a display of:
* dev.cpu.0.cx_supported: C1/1 C2/96
Instead of
* dev.cpu.0.cx_supported: C1/1 C3/96
MFC after: 2 weeks
processor objects. Instead of forcing the new-bus CPU objects to use
a unit number equal to pc_cpuid, adjust acpi_pcpu_get_id() to honor the
MADT IDs by default. As with the previous change, setting
debug.acpi.cpu_unordered to 1 in the loader will revert to the old
behavior.
Tested by: jimharris
MFC after: 1 month
one. Interestingly, these are actually the default for quite some time
(bus_generic_driver_added(9) since r52045 and bus_generic_print_child(9)
since r52045) but even recently added device drivers do this unnecessarily.
Discussed with: jhb, marcel
- While at it, use DEVMETHOD_END.
Discussed with: jhb
- Also while at it, use __FBSDID.
processors unless the invariant TSC bit of CPUID is set. Intel processors
may stop incrementing TSC when DPSLP# pin is asserted, according to Intel
processor manuals, i. e., TSC timecounter is useless if the processor can
enter deep sleep state (C3/C4). This problem was accidentally uncovered by
r222869, which increased timecounter quality of P-state invariant TSC, e.g.,
for Core2 Duo T5870 (Family 6, Model f) and Atom N270 (Family 6, Model 1c).
Reported by: Fabian Keil (freebsd-listen at fabiankeil dot de)
Ian FREISLICH (ianf at clue dot co dot za)
Tested by: Fabian Keil (freebsd-listen at fabiankeil dot de)
- Core2 Duo T5870 (C3 state available/enabled)
jkim - Xeon X5150 (C3 state unavailable)
safer for i386 because it can be easily over 4 GHz now. More worse, it can
be easily changed by user with 'machdep.tsc_freq' tunable (directly) or
cpufreq(4) (indirectly). Note it is intentionally not used in performance
critical paths to avoid performance regression (but we should, in theory).
Alternatively, we may add "virtual TSC" with lower frequency if maximum
frequency overflows 32 bits (and ignore possible incoherency as we do now).
function always returned the nominal frequency instead of current frequency
because we use RDTSC instruction to calculate difference in CPU ticks, which
is supposedly constant for the case. Now we support cpu_get_nominal_mhz()
for the case, instead. Note it should be just enough for most usage cases
because cpu_est_clockrate() is often times abused to find maximum frequency
of the processor.
ACPI specification sates that if P_LVL2_LAT > 100, then a system doesn't
support C2; if P_LVL3_LAT > 1000, then C3 is not supported.
But there are no such rules for Cx state data returned by _CST. If a
state is not supported it should not be included into the return
package. In other words, any latency value returned by _CST is valid,
it's up to the OS and/or user to decide whether to use it.
Submitted by: nork
Suggested by: mav
MFC after: 1 week
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
This reflects actual type used to store and compare child device orders.
Change is mostly done via a Coccinelle (soon to be devel/coccinelle)
semantic patch.
Verified by LINT+modules kernel builds.
Followup to: r212213
MFC after: 10 days