AP startup on PVH follows the PV method, so we need to add a hook in
order to diverge from bare metal.
Approved by: gibbs
Sponsored by: Citrix Systems R&D
amd64/amd64/machdep.c:
- Add hook for start_all_aps on native (using native_start_all_aps
defined in mp_machdep).
amd64/amd64/mp_machdep.c:
- Make some variables global because they will also be used by the
Xen PVH AP startup code.
- Use the start_all_aps hook to start APs.
- Rename start_all_aps to native_start_all_aps.
amd64/include/smp.h:
- Add declaration for native_start_all_aps.
x86/include/init.h:
- Declare start_all_aps hook in init_ops.
x86/xen/pv.c:
- Pick external declarations from mp_machdep.
- Introduce Xen PV code to start APs on PVH.
- Set start_all_aps init hook to use the Xen PVH implementation.
Move asm IPIs handlers to C code, so both Xen and native IPI handlers
share the same code.
Reviewed by: jhb
Approved by: gibbs
Sponsored by: Citrix Systems R&D
amd64/amd64/apic_vector.S:
i386/i386/apic_vector.s:
- Remove asm coded IPI handlers and instead call the newly introduced
C variants.
amd64/amd64/mp_machdep.c:
i386/i386/mp_machdep.c:
- Add C coded clones to the asm IPI handlers (moved from
x86/xen/hvm.c).
i386/include/smp.h:
amd64/include/smp.h:
- Add prototypes for the C IPI handlers.
x86/xen/hvm.c:
- Move the C IPI handlers to mp_machdep and call those in the Xen IPI
handlers.
i386/xen/mp_machdep.c:
- Add dummy IPI handlers to the i386 Xen PV port (this port doesn't
support SMP).
- Restore the pre-PCID TLB shootdown handlers for whole address space
and single page invalidation asm code, and assign the IPI handler to
them when PCID is not supported or disabled. Old handlers have
linear control flow. But, still use the common return sequence.
- Stop using pcpu for INVPCID descriptors in the invlrg handler. It
is enough to allocate descriptors on the stack. As result, two
SWAPGS instructions are shaved off from the code for Haswell+.
- Fix the reverted condition in invlrng for checking of the PCID
support [1], also in invlrng check that pmap is kernel pmap before
performing other tests. For the kernel pmap, which provides global
mappings, the INVLPG must be used for invalidation always.
- Save the pre-computed pmap' %CR3 register in the struct pmap. This
allows to remove several checks for pm_pcid validity when %CR3 is
reloaded [2].
Noted by: gibbs [1]
Discussed with: alc [2]
Tested by: pho, flo
Sponsored by: The FreeBSD Foundation
Intel CPUs. The feature tags TLB entries with the Id of the address
space and allows to avoid TLB invalidation on the context switch, it
is available only in the long mode. In the microbenchmarks, using the
PCID decreased latency of the context switches by ~30% on SandyBridge
class desktop CPUs, measured with the lat_ctx program from lmbench.
If available, use INVPCID instruction when a TLB entry in non-current
address space needs to be invalidated. The instruction is typically
available on the Haswell.
If needed, the use of PCID can be turned off with the
vm.pmap.pcid_enabled loader tunable set to 0. The state of the
feature is reported by the vm.pmap.pcid_enabled sysctl. The sysctl
vm.pmap.pcid_save_cnt reports the number of context switches which
avoided invalidating the TLB; compare with the total number of context
switches, available as sysctl vm.stats.sys.v_swtch.
Sponsored by: The FreeBSD Foundation
Reviewed by: alc
Tested by: pho, bf
The code has definitely been broken for SCHED_ULE, which is a default
scheduler. It may have been broken for SCHED_4BSD in more subtle ways,
e.g. with manually configured CPU affinities and for interrupt devilery
purposes.
We still provide a way to disable individual CPUs or all hyperthreading
"twin" CPUs before SMP startup. See the UPDATING entry for details.
Interaction between building CPU topology and disabling CPUs still
remains fuzzy: topology is first built using all availble CPUs and then
the disabled CPUs should be "subtracted" from it. That doesn't work
well if the resulting topology becomes non-uniform.
This work is done in cooperation with Attilio Rao who in addition to
reviewing also provided parts of code.
PR: kern/145385
Discussed with: gcooper, ambrisko, mdf, sbruno
Reviewed by: attilio
Tested by: pho, pluknet
X-MFC after: never
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
function on the possibility of a thread to not preempt.
As this function is very tied to x86 (interrupts disabled checkings)
it is not intended to be used in MI code.
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
This information can be very valuable for CPU sleep-time (and respectively
idle power consumption) optimization.
Add counters for timer-related IPIs.
Reviewed by: jhb@ (previous version)
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)
With the arrival of 128+ cores it is necessary to handle more than that.
One of the first thing to change is the support for cpumask_t that needs
to handle more than 32 bits masking (which happens now). Some places,
however, still assume that cpumask_t is a 32 bits mask.
Fix that situation by using always correctly cpumask_t when needed.
While here, remove the part under STOP_NMI for the Xen support as it
is broken in any case.
Additively make ipi_nmi_pending as static.
Reviewed by: jhb, kmacy
Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com>
topology of nehalem/corei7 based systems.
- Remove the cpu_cores/cpu_logical detection from identcpu.
- Describe the layout of the system in cpu_mp_announce().
Sponsored by: Nokia
This code is heavily inspired by Takanori Watanabe's experimental SMP patch
for i386 and large portion was shamelessly cut and pasted from Peter Wemm's
AP boot code.
all to date and the latter also is only used in ia64 and powerpc
code which no longer serves a real purpose after bring-up and just
can be removed as well. Note that architectures like sun4u also
provide no means of implementing IPI'ing a CPU itself natively
in the first place.
Suggested by: jhb
Reviewed by: arch, grehan, jhb
tree structure that encodes the level of cache sharing and other
properties.
- Provide several convenience functions for creating one and two level
cpu trees as well as a default flat topology. The system now always
has some topology.
- On i386 and amd64 create a seperate level in the hierarchy for HTT
and multi-core cpus. This will allow the scheduler to intelligently
load balance non-uniform cores. Presently we don't detect what level
of the cache hierarchy is shared at each level in the topology.
- Add a mechanism for testing common topologies that have more information
than the MD code is able to provide via the kern.smp.topology tunable.
This should be considered a debugging tool only and not a stable api.
Sponsored by: Nokia
In particular:
- smp_tlb_mtx is no longer used, so it is axed.
- smp rendezvous lock isn't really a leaf spin-mutex. Its bad placement in
the table, however, has been the source of a false positive LOR reporting
with the dt_lock. However, smp rendezvous lock would have had sched_lock
there for older lock, so it wasn't still a leaf lock.
- allpmaps is only used in ia32 architecture, so it is inserted in the
appropriate stub.
Addictionally:
- kse_zombie_lock is no longer present, so its definition is axed out.
- zombie_lock doesn't need to have an exported symbol, so just let's it be
declared as static.
Tested by: kris
Approved by: jeff (mentor)
Approved by: re
passing a pointer to an opaque clockframe structure and requiring the
MD code to supply CLKF_FOO() macros to extract needed values out of the
opaque structure, just pass the needed values directly. In practice this
means passing the pair (usermode, pc) to hardclock() and profclock() and
passing the boolean (usermode) to hardclock_cpu() and hardclock_process().
Other details:
- Axe clockframe and CLKF_FOO() macros on all architectures. Basically,
all the archs were taking a trapframe and converting it into a clockframe
one way or another. Now they can just extract the PC and usermode values
directly out of the trapframe and pass it to fooclock().
- Renamed hardclock_process() to hardclock_cpu() as the latter is more
accurate.
- On Alpha, we now run profclock() at hz (profhz == hz) rather than at
the slower stathz.
- On Alpha, for the TurboLaser machines that don't have an 8254
timecounter, call hardclock() directly. This removes an extra
conditional check from every clock interrupt on Alpha on the BSP.
There is probably room for even further pruning here by changing Alpha
to use the simplified timecounter we use on x86 with the lapic timer
since we don't get interrupts from the 8254 on Alpha anyway.
- On x86, clkintr() shouldn't ever be called now unless using_lapic_timer
is false, so add a KASSERT() to that affect and remove a condition
to slightly optimize the non-lapic case.
- Change prototypeof arm_handler_execute() so that it's first arg is a
trapframe pointer rather than a void pointer for clarity.
- Use KCOUNT macro in profclock() to lookup the kernel profiling bucket.
Tested on: alpha, amd64, arm, i386, ia64, sparc64
Reviewed by: bde (mostly)
- Move PUSH_FRAME and POP_FRAME to asmacros.h and use PUSH_FRAME in
atpic entry points.
- Move PCPU_* asm macros out of the middle of the asm profiling macros.
- Pass IRQ vector argument as an int rather than void * to reduce diffs
with i386.
- EOI the lapic in C for the lapic timer handler.
- GC unused Xcpuast function.
- Split IPI_STOP handling code of ipi_nmi_handler() out into a
cpustop_handler() function and call it from Xcpustop rather than
duplicating all the logic in assembly.
- Fixup the list of symbols with interrupt frames in ddb traces.
Xatpic_fastintr* have never existed on amd64, and the lapic timer
handler and various IPI handlers were missing.
- Use trapframe instead of intrframe for interrupt entry points (on amd64
the interrupt vector was already a separate argument, so the two frames
were already identical) and GC intrframe.
Submitted by: peter (3)
IPI_STOP IPIs.
- Change the i386 and amd64 MD IPI code to send an NMI if STOP_NMI is
enabled if an attempt is made to send an IPI_STOP IPI. If the kernel
option is enabled, there is also a sysctl to change the behavior at
runtime (debug.stop_cpus_with_nmi which defaults to enabled). This
includes removing stop_cpus_nmi() and making ipi_nmi_selected() a
private function for i386 and amd64.
- Fix ipi_all(), ipi_all_but_self(), and ipi_self() on i386 and amd64 to
properly handle bitmapped IPIs as well as IPI_STOP IPIs when STOP_NMI is
enabled.
- Fix ipi_nmi_handler() to execute the restart function on the first CPU
that is restarted making use of atomic_readandclear() rather than
assuming that the BSP is always included in the set of restarted CPUs.
Also, the NMI handler didn't clear the function pointer meaning that
subsequent stop and restarts could execute the function again.
- Define a new macro HAVE_STOPPEDPCBS on i386 and amd64 to control the use
of stoppedpcbs[] and always enable it for i386 and amd64 instead of
being dependent on KDB_STOP_NMI. It works fine in both the NMI and
non-NMI cases.
a regular IPI vector, but this vector is blocked when interrupts are disabled.
With "options KDB_STOP_NMI" and debug.kdb.stop_cpus_with_nmi set, KDB will
send an NMI to each CPU instead. The code also has a context-stuffing
feature which helps ddb extract the state of processes running on the
stopped CPUs.
KDB_STOP_NMI is only useful with SMP and complains if SMP is not defined.
This feature only applies to i386 and amd64 at the moment, but could be
used on other architectures with the appropriate MD bits.
Submitted by: ups
to help the AMD cpus (which have a hardware tlb flush filter). I held
off to see what the 64 bit Intel cpus did, but it doesn't seem to help
much there either. Oh well, store it in the Attic.
- This is heavily derived from John Baldwin's apic/pci cleanup on i386.
- I have completely rewritten or drastically cleaned up some other parts.
(in particular, bootstrap)
- This is still a WIP. It seems that there are some highly bogus bioses
on nVidia nForce3-150 boards. I can't stress how broken these boards
are. I have a workaround in mind, but right now the Asus SK8N is broken.
The Gigabyte K8NPro (nVidia based) is also mind-numbingly hosed.
- Most of my testing has been with SCHED_ULE. SCHED_4BSD works.
- the apic and acpi components are 'standard'.
- If you have an nVidia nForce3-150 board, you are stuck with 'device
atpic' in addition, because they somehow managed to forget to connect the
8254 timer to the apic, even though its in the same silicon! ARGH!
This directly violates the ACPI spec.
a heavily stripped down FreeBSD/i386 (brutally stripped down actually) to
attempt to get a stable base to start from. There is a lot missing still.
Worth noting:
- The kernel runs at 1GB in order to cheat with the pmap code. pmap uses
a variation of the PAE code in order to avoid having to worry about 4
levels of page tables yet.
- It boots in 64 bit "long mode" with a tiny trampoline embedded in the
i386 loader. This simplifies locore.s greatly.
- There are still quite a few fragments of i386-specific code that have
not been translated yet, and some that I cheated and wrote dumb C
versions of (bcopy etc).
- It has both int 0x80 for syscalls (but using registers for argument
passing, as is native on the amd64 ABI), and the 'syscall' instruction
for syscalls. int 0x80 preserves all registers, 'syscall' does not.
- I have tried to minimize looking at the NetBSD code, except in a couple
of places (eg: to find which register they use to replace the trashed
%rcx register in the syscall instruction). As a result, there is not a
lot of similarity. I did look at NetBSD a few times while debugging to
get some ideas about what I might have done wrong in my first attempt.
as it could be and can do with some more cleanup. Currently its under
options LAZY_SWITCH. What this does is avoid %cr3 reloads for short
context switches that do not involve another user process. ie: we can
take an interrupt, switch to a kthread and return to the user without
explicitly flushing the tlb. However, this isn't as exciting as it could
be, the interrupt overhead is still high and too much blocks on Giant
still. There are some debug sysctls, for stats and for an on/off switch.
The main problem with doing this has been "what if the process that you're
running on exits while we're borrowing its address space?" - in this case
we use an IPI to give it a kick when we're about to reclaim the pmap.
Its not compiled in unless you add the LAZY_SWITCH option. I want to fix a
few more things and get some more feedback before turning it on by default.
This is NOT a replacement for Bosko's lazy interrupt stuff. This was more
meant for the kthread case, while his was for interrupts. Mine helps a
little for interrupts, but his helps a lot more.
The stats are enabled with options SWTCH_OPTIM_STATS - this has been a
pseudo-option for years, I just added a bunch of stuff to it.
One non-trivial change was to select a new thread before calling
cpu_switch() in the first place. This allows us to catch the silly
case of doing a cpu_switch() to the current process. This happens
uncomfortably often. This simplifies a bit of the asm code in cpu_switch
(no longer have to call choosethread() in the middle). This has been
implemented on i386 and (thanks to jake) sparc64. The others will come
soon. This is actually seperate to the lazy switch stuff.
Glanced at by: jake, jhb
This keeps the logical cpu's halted in the idle loop. By default
the logical cpu's are halted at startup. It is also possible to
halt any cpu in the idle loop now using machdep.hlt_cpus.
Examples of how to use this:
machdep.hlt_cpus=1 halt cpu0
machdep.hlt_cpus=2 halt cpu1
machdep.hlt_cpus=4 halt cpu2
machdep.hlt_cpus=3 halt cpu0,cpu1
Reviewed by: jhb, peter
statclock based on profhz when profiling is enabled MD, since most platforms
don't use this anyway. This removes the need for statclock_process, whose
only purpose was to subdivide profhz, and gets the profiling clock running
outside of sched_lock on platforms that implement suswintr.
Also changed the interface for starting and stopping the profiling clock to
do just that, instead of changing the rate of statclock, since they can now
be separate.
Reviewed by: jhb, tmm
Tested on: i386, sparc64
- It actually works this time, honest!
- Fine grained TLB shootdowns for SMP on i386. IPI's are very expensive,
so try and optimize things where possible.
- Introduce ranged shootdowns that can be done as a single IPI.
- PG_G support for i386
- Specific-cpu targeted shootdowns. For example, there is no sense in
globally purging the TLB cache for where we are stealing a page from
the local unshared process on the local cpu. Use pm_active to track
this.
- Add some instrumentation for the tlb shootdown code.
- Rip out SMP code from <machine/cpufunc.h>
- Try and fix some very bogus PG_G and PG_PS interactions that were bad
enough to cause vm86 bios calls to break. vm86 depended on our existing
bugs and this was the cause of the VESA panics last time.
- Fix the silly one-line error that caused the 'panic: bad pte' last time.
- Fix a couple of other silly one-line errors that should have caused more
pain than they did.
Some more work is needed:
- pmap_{zero,copy}_page[_idle]. These can be done without IPI's if we
have a hook in cpu_switch.
- The IPI handlers need some cleanup. I have a bogus %ds load that can
be avoided.
- APTD handling is rather bogus and appears to be a large source of
global TLB IPI shootdowns for no really good reason.
I see speedups of between 1.5% and ~4% on buildworlds in a while 1 loop.
I expect to see a bigger difference when there is significant pageout
activity or the system otherwise has memory shortages.
I have backed out a few optimizations that I had been using over the last
few days in order to be a little more conservative. I'll revisit these
again over the next few days as the dust settles.
New option: DISABLE_PG_G - In case I missed something.
not removing tabs before "__P((", and not outdenting continuation lines
to preserve non-KNF lining up of code with parentheses. Switch to KNF
formatting and/or rewrap the whole prototype in some cases.
There is some unresolved badness that has been eluding me, particularly
affecting uniprocessor kernels. Turning off PG_G helped (which is a bad
sign) but didn't solve it entirely. Userland programs still crashed.
on for a while:
- fine grained TLB shootdown for SMP on i386
- ranged TLB shootdowns.. eg: specify a range of pages to shoot down with
a single IPI, since the IPI is very expensive. Adjust some callers
that used to trigger this inside tight loops to do a ranged shootdown
at the end instead.
- PG_G support for SMP on i386 (options ENABLE_PG_G)
- defer PG_G activation till after we decide what we are going to do with
PSE and the 4MB pages at the start of the kernel. This should solve
some rumored strangeness about stale PG_G entries getting stuck
underneath the 4MB pages.
- add some instrumentation for the fine TLB shootdown
- convert some asm instruction wrappers from functions to inlines. gcc
seems to do a fair bit better with this.
- [temporarily!] pessimize the tlb shootdown IPI handlers. I will fix
this again shortly.
This has been working fairly well for me for a while, but I have tweaked
it again prior to commit since my last major testing round. The only
outstanding problem that I know of is PG_G related, which is why there
is an option for it (not on by default for SMP). I have seen a world
speedups by a few percent (as much as 4 or 5% in one case) but I have
*not* accurately measured this - I am a bit sceptical of these numbers.
- Axe inlvtlb_ok as it was completely redundant with smp_active.
- Remove references to non-existent variable and non-existent file
in i386/include/smp.h.
- Don't perform initializations local to each CPU while holding the
ap boot lock on i386 while an AP bootstraps itself.
- Reorganize the AP startup code some to unify the latter half of the
functions to bring an AP up. Eventually this might be broken out into
a MI function in subr_smp.c.
been made machine independent and various other adjustments have been made
to support Alpha SMP.
- It splits the per-process portions of hardclock() and statclock() off
into hardclock_process() and statclock_process() respectively. hardclock()
and statclock() call the *_process() functions for the current process so
that UP systems will run as before. For SMP systems, it is simply necessary
to ensure that all other processors execute the *_process() functions when the
main clock functions are triggered on one CPU by an interrupt. For the alpha
4100, clock interrupts are delievered in a staggered broadcast fashion, so
we simply call hardclock/statclock on the boot CPU and call the *_process()
functions on the secondaries. For x86, we call statclock and hardclock as
usual and then call forward_hardclock/statclock in the MD code to send an IPI
to cause the AP's to execute forwared_hardclock/statclock which then call the
*_process() functions.
- forward_signal() and forward_roundrobin() have been reworked to be MI and to
involve less hackery. Now the cpu doing the forward sets any flags, etc. and
sends a very simple IPI_AST to the other cpu(s). AST IPIs now just basically
return so that they can execute ast() and don't bother with setting the
astpending or needresched flags themselves. This also removes the loop in
forward_signal() as sched_lock closes the race condition that the loop worked
around.
- need_resched(), resched_wanted() and clear_resched() have been changed to take
a process to act on rather than assuming curproc so that they can be used to
implement forward_roundrobin() as described above.
- Various other SMP variables have been moved to a MI subr_smp.c and a new
header sys/smp.h declares MI SMP variables and API's. The IPI API's from
machine/ipl.h have moved to machine/smp.h which is included by sys/smp.h.
- The globaldata_register() and globaldata_find() functions as well as the
SLIST of globaldata structures has become MI and moved into subr_smp.c.
Also, the globaldata list is only available if SMP support is compiled in.
Reviewed by: jake, peter
Looked over by: eivind
mpapic.c. This gives us the benefit of C type checking. These functions
are not called in any critical paths and are not used by the interrupt
routines.
