On 18.09.23 11:06, David Woodhouse wrote:
> From: David Woodhouse <[email protected]>
>
> The VMM may have work to do on behalf of the guest, and it's often
> desirable to use the cycles when the vCPUS are idle.
>
> When the vCPU uses HLT this works out OK because the VMM can run its
> tasks in a separate thread which gets scheduled when the in-kernel
> emulation of HLT schedules away. It isn't perfect, because it doesn't
> easily allow for handling both low-priority maintenance tasks when the
> VMM wants to wait until the vCPU is idle, and also for higher priority
> tasks where the VMM does want to preempt the vCPU. It can also lead to
> noisy neighbour effects, when a host has isn't necessarily sized to
> expect any given VMM to suddenly be contending for many *more* pCPUs
> than it has vCPUs.
>
> In addition, there are times when we need to expose MWAIT to a guest
> for compatibility with a previous environment. And MWAIT is much harder
> because it's very hard to emulate properly.
>
> There were attempts at doing so based on marking the target page read-
> only in MONITOR and triggering the wake when it takes a minor fault,
> but so far they haven't led to a working solution:
> https://www.contrib.andrew.cmu.edu/~somlo/OSXKVM/mwait.html
>
> So when a guest executes MWAIT, either we've disabled exit-on-mwait and
> the guest actually sits in non-root mode hogging the pCPU, or if we do
> enable exit-on-mwait the kernel just treats it as a NOP and bounces
> right back into the guest to busy-wait round its idle loop.
>
> For a start, we can stick a yield() into that busy-loop. The yield()
> has fairly poorly defined semantics, but it's better than *nothing* and
> does allow a VMM's thread-based I/O and maintenance tasks to run a
> *little* better.
>
> Better still, we can bounce all the way out to *userspace* on an MWAIT
> exit, and let the VMM perform some of its pending work right there and
> then in the vCPU thread before re-entering the vCPU. That's much nicer
> than yield(). The vCPU is still runnable, since we still don't have a
> *real* emulation of MWAIT, so the vCPU thread can do a *little* bit of
> work and then go back into the vCPU for another turn around the loop.
>
> And if we're going to do that kind of task processing for MWAIT-idle
> guests directly from the vCPU thread, it's neater to do it for HLT-idle
> guests that way too.
>
> For HLT, the vCPU *isn't* runnable; it'll be in KVM_MP_STATE_HALTED.
> The VMM can poll the mp_state and know when the vCPU should be run
> again. But not poll(), although we might want to hook up something like
> that (or just a signal or eventfd) for other reasons for VSM anyway.
> The VMM can also just do some work and then re-enter the vCPU without
> the corresponding bit set in the kvm_run struct.
>
> So, er, what does this patch do? Add a capability, define two bits for
> exiting to userspace on HLT or MWAIT — in the kvm_run struct rather
> than needing a separate ioctl to turn them on or off, so that the VMM
> can make the decision each time it enters the vCPU. Hook it up to
> (ab?)use the existing KVM_EXIT_HLT which was previously only used when
> the local APIC was emulated in userspace, and add a new KVM_EXIT_MWAIT.
>
> Fairly much untested.
>
> If this approach seems reasonable, of course I'll add test cases and
> proper documentation before posting it for real. This is the proof of
> concept before we even put it through testing to see what performance
> we get out of it especially for those obnoxious MWAIT-enabled guests.
>
> Signed-off-by: David Woodhouse <[email protected]>
IIUC you want to do work in a user space vCPU thread when the guest vCPU
is idle. As you pointed out above, KVM can not actually do much about
MWAIT: It basically busy loops and hogs the CPU.
The typical flow I would expect for "work in a vCPU thread" is:
0) vCPU runs. HLT/MWAIT is directly exposed to guest.
1) vCPU exits. Creates deferred work. Enables HLT/MWAIT trapping.
2) vCPU runs again
3) vCPU calls HLT/MWAIT. We exit to user space to finish work from 1
4) vCPU runs again without HLT/MWAIT trapping
That means on top (or instead?) of the bits you have below that indicate
"Should I exit to user space?", what you really need are bits that do
what enable_cap(KVM_CAP_X86_DISABLE_EXITS) does in light-weight: Disable
HLT/MWAIT trapping temporarily.
Also, please keep in mind that you still would need a fallback mechanism
to run your "deferred work" even when the guest does not call HLT/MWAIT,
like a regular timer in your main thread.
On top of all this, I'm not sure it's more efficient to do the trap to
the vCPU thread compared to just creating a separate real thread. Your
main problem is the emulatability of MWAIT because that leaves "no time"
to do deferred work. But then again, if your deferred work is so complex
that it needs more than a few ms (which you can always steal from the
vCPU thread, especiall with yield()), you'll need to start implementing
time slicing of that work in user space next - and basically rebuild
your own scheduler there. Ugh.
IMHO the real core value of this idea would be in a vcpu_run bit that on
VCPU_RUN can toggle between HLT/MWAIT intercept on and off. The actual
trap to user space, you're most likely better off with a separate thread.
Alex
>
> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> index a6582c1fd8b9..8f931539114a 100644
> --- a/arch/x86/kvm/x86.c
> +++ b/arch/x86/kvm/x86.c
> @@ -2128,9 +2128,23 @@ static int kvm_emulate_monitor_mwait(struct kvm_vcpu *vcpu, const char *insn)
> pr_warn_once("%s instruction emulated as NOP!\n", insn);
> return kvm_emulate_as_nop(vcpu);
> }
> +
> int kvm_emulate_mwait(struct kvm_vcpu *vcpu)
> {
> - return kvm_emulate_monitor_mwait(vcpu, "MWAIT");
> + int ret = kvm_emulate_monitor_mwait(vcpu, "MWAIT");
> +
> + if (ret && kvm_userspace_exit(vcpu, KVM_EXIT_MWAIT)) {
> + vcpu->run->exit_reason = KVM_EXIT_MWAIT;
> + ret = 0;
> + } else {
> + /*
> + * Calling yield() has poorly defined semantics, but the
> + * guest is in a busy loop and it's the best we can do
> + * without a full emulation of MONITOR/MWAIT.
> + */
> + yield();
> + }
> + return ret;
> }
> EXPORT_SYMBOL_GPL(kvm_emulate_mwait);
>
> @@ -4554,6 +4568,9 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
> r |= KVM_X86_DISABLE_EXITS_MWAIT;
> }
> break;
> + case KVM_CAP_X86_USERSPACE_EXITS:
> + r = KVM_X86_USERSPACE_VALID_EXITS;
> + break;
> case KVM_CAP_X86_SMM:
> if (!IS_ENABLED(CONFIG_KVM_SMM))
> break;
> @@ -9643,11 +9660,11 @@ static int __kvm_emulate_halt(struct kvm_vcpu *vcpu, int state, int reason)
> ++vcpu->stat.halt_exits;
> if (lapic_in_kernel(vcpu)) {
> vcpu->arch.mp_state = state;
> - return 1;
> - } else {
> - vcpu->run->exit_reason = reason;
> - return 0;
> + if (!kvm_userspace_exit(vcpu, reason))
> + return 1;
> }
> + vcpu->run->exit_reason = reason;
> + return 0;
> }
>
> int kvm_emulate_halt_noskip(struct kvm_vcpu *vcpu)
> diff --git a/arch/x86/kvm/x86.h b/arch/x86/kvm/x86.h
> index 1e7be1f6ab29..ce10a809151c 100644
> --- a/arch/x86/kvm/x86.h
> +++ b/arch/x86/kvm/x86.h
> @@ -430,6 +430,19 @@ static inline bool kvm_notify_vmexit_enabled(struct kvm *kvm)
> return kvm->arch.notify_vmexit_flags & KVM_X86_NOTIFY_VMEXIT_ENABLED;
> }
>
> +static inline bool kvm_userspace_exit(struct kvm_vcpu *vcpu, int reason)
> +{
> + if (reason == KVM_EXIT_HLT &&
> + (vcpu->run->userspace_exits & KVM_X86_USERSPACE_EXIT_HLT))
> + return true;
> +
> + if (reason == KVM_EXIT_MWAIT &&
> + (vcpu->run->userspace_exits & KVM_X86_USERSPACE_EXIT_MWAIT))
> + return true;
> +
> + return false;
> +}
> +
> enum kvm_intr_type {
> /* Values are arbitrary, but must be non-zero. */
> KVM_HANDLING_IRQ = 1,
> diff --git a/include/uapi/linux/kvm.h b/include/uapi/linux/kvm.h
> index 13065dd96132..43d94d49fc24 100644
> --- a/include/uapi/linux/kvm.h
> +++ b/include/uapi/linux/kvm.h
> @@ -264,6 +264,7 @@ struct kvm_xen_exit {
> #define KVM_EXIT_RISCV_SBI 35
> #define KVM_EXIT_RISCV_CSR 36
> #define KVM_EXIT_NOTIFY 37
> +#define KVM_EXIT_MWAIT 38
>
> /* For KVM_EXIT_INTERNAL_ERROR */
> /* Emulate instruction failed. */
> @@ -283,7 +284,8 @@ struct kvm_run {
> /* in */
> __u8 request_interrupt_window;
> __u8 immediate_exit;
> - __u8 padding1[6];
> + __u8 userspace_exits;
> + __u8 padding1[5];
>
> /* out */
> __u32 exit_reason;
> @@ -841,6 +843,11 @@ struct kvm_ioeventfd {
> KVM_X86_DISABLE_EXITS_PAUSE | \
> KVM_X86_DISABLE_EXITS_CSTATE)
>
> +#define KVM_X86_USERSPACE_EXIT_MWAIT (1 << 0)
> +#define KVM_X86_USERSPACE_EXIT_HLT (1 << 1)
> +#define KVM_X86_USERSPACE_VALID_EXITS (KVM_X86_USERSPACE_EXIT_MWAIT | \
> + KVM_X86_USERSPACE_EXIT_HLT)
> +
> /* for KVM_ENABLE_CAP */
> struct kvm_enable_cap {
> /* in */
> @@ -1192,6 +1199,7 @@ struct kvm_ppc_resize_hpt {
> #define KVM_CAP_COUNTER_OFFSET 227
> #define KVM_CAP_ARM_EAGER_SPLIT_CHUNK_SIZE 228
> #define KVM_CAP_ARM_SUPPORTED_BLOCK_SIZES 229
> +#define KVM_CAP_X86_USERSPACE_EXITS 230
>
> #ifdef KVM_CAP_IRQ_ROUTING
>
>
Amazon Development Center Germany GmbH
Krausenstr. 38
10117 Berlin
Geschaeftsfuehrung: Christian Schlaeger, Jonathan Weiss
Eingetragen am Amtsgericht Charlottenburg unter HRB 149173 B
Sitz: Berlin
Ust-ID: DE 289 237 879
On Mon, 2023-09-18 at 11:41 +0200, Alexander Graf wrote:
>
> IIUC you want to do work in a user space vCPU thread when the guest vCPU
> is idle. As you pointed out above, KVM can not actually do much about
> MWAIT: It basically busy loops and hogs the CPU.
Well.. I suspect what I *really* want is a decent way to emulate MWAIT
properly and let it actually sleep. Or failing that, to declare that we
can actually change the guest-visible experience when those guests are
migrated to KVM, and take away MWAIT completely.
> The typical flow I would expect for "work in a vCPU thread" is:
>
> 0) vCPU runs. HLT/MWAIT is directly exposed to guest.
> 1) vCPU exits. Creates deferred work. Enables HLT/MWAIT trapping.
That can happen, but it may also be a separate I/O thread which
receives an eventfd notification and finds that there is now work to be
done. If that work can be fairly much instantaneous, it can be done
immediately. Else it gets deferred to what we Linux hackers might think
of as a workqueue.
If all the vCPUs are in HLT when the work queue becomes non-empty, we'd
need to prod them *all* to change their exit-on-{HLT,MWAIT} status when
work becomes available, just in case one of them becomes idle and can
process the work "for free" using idle cycles.
> 2) vCPU runs again
> 3) vCPU calls HLT/MWAIT. We exit to user space to finish work from 1
> 4) vCPU runs again without HLT/MWAIT trapping
>
> That means on top (or instead?) of the bits you have below that indicate
> "Should I exit to user space?", what you really need are bits that do
> what enable_cap(KVM_CAP_X86_DISABLE_EXITS) does in light-weight: Disable
> HLT/MWAIT trapping temporarily.
If I do it that way, yes. A lightweight way to enable/disable the exits
even to kernel would be a nice to have. But it's a trade-off. For HLT
you'd get lower latency re-entering the vCPU at a cost of much higher
latency processing work if the vCPU was *already* in HLT.
We probably would want to stop burning power in the MWAIT loop though,
and let the pCPU sit in the guest in MWAIT if there really is nothing
else to do.
We're experimenting with various permutations.
> Also, please keep in mind that you still would need a fallback mechanism
> to run your "deferred work" even when the guest does not call HLT/MWAIT,
> like a regular timer in your main thread.
Yeah. In that case I think the ideal answer is that we let the kernel
scheduler sort it out. I was thinking of a model where we have I/O (or
workqueue) threads in *addition* to the userspace exits on idle. The
separate threads own the work (and a number of them are woken according
to the queue depth), and idle vCPUs *opportunistically* process work
items on top of that.
That approach alone would work fine with the existing HLT scheduling;
it's just MWAIT which is a pain because yield() doesn't really do much
(but as noted, it's better than *nothing*).
> On top of all this, I'm not sure it's more efficient to do the trap to
> the vCPU thread compared to just creating a separate real thread. Your
> main problem is the emulatability of MWAIT because that leaves "no time"
> to do deferred work. But then again, if your deferred work is so complex
> that it needs more than a few ms (which you can always steal from the
> vCPU thread, especiall with yield()), you'll need to start implementing
> time slicing of that work in user space next - and basically rebuild
> your own scheduler there. Ugh.
>
> IMHO the real core value of this idea would be in a vcpu_run bit that on
> VCPU_RUN can toggle between HLT/MWAIT intercept on and off. The actual
> trap to user space, you're most likely better off with a separate thread.
No, that's very much not the point. The problem is that yield() doesn't
work well enough — and isn't designed or guaranteed to do anything in
particular for most cases. It's better than *nothing* but we want the
opportunity to do the actual work right there in the *loop* of the
guest bouncing through MWAIT.
On 18.09.23 13:10, David Woodhouse wrote:
> On Mon, 2023-09-18 at 11:41 +0200, Alexander Graf wrote:
>> IIUC you want to do work in a user space vCPU thread when the guest vCPU
>> is idle. As you pointed out above, KVM can not actually do much about
>> MWAIT: It basically busy loops and hogs the CPU.
> Well.. I suspect what I *really* want is a decent way to emulate MWAIT
> properly and let it actually sleep. Or failing that, to declare that we
> can actually change the guest-visible experience when those guests are
> migrated to KVM, and take away MWAIT completely.
>
>> The typical flow I would expect for "work in a vCPU thread" is:
>>
>> 0) vCPU runs. HLT/MWAIT is directly exposed to guest.
>> 1) vCPU exits. Creates deferred work. Enables HLT/MWAIT trapping.
> That can happen, but it may also be a separate I/O thread which
> receives an eventfd notification and finds that there is now work to be
> done. If that work can be fairly much instantaneous, it can be done
> immediately. Else it gets deferred to what we Linux hackers might think
> of as a workqueue.
>
> If all the vCPUs are in HLT when the work queue becomes non-empty, we'd
> need to prod them *all* to change their exit-on-{HLT,MWAIT} status when
> work becomes available, just in case one of them becomes idle and can
> process the work "for free" using idle cycles.
>
>> 2) vCPU runs again
>> 3) vCPU calls HLT/MWAIT. We exit to user space to finish work from 1
>> 4) vCPU runs again without HLT/MWAIT trapping
>>
>> That means on top (or instead?) of the bits you have below that indicate
>> "Should I exit to user space?", what you really need are bits that do
>> what enable_cap(KVM_CAP_X86_DISABLE_EXITS) does in light-weight: Disable
>> HLT/MWAIT trapping temporarily.
> If I do it that way, yes. A lightweight way to enable/disable the exits
> even to kernel would be a nice to have. But it's a trade-off. For HLT
> you'd get lower latency re-entering the vCPU at a cost of much higher
> latency processing work if the vCPU was *already* in HLT.
>
> We probably would want to stop burning power in the MWAIT loop though,
> and let the pCPU sit in the guest in MWAIT if there really is nothing
> else to do.
>
> We're experimenting with various permutations.
>
>> Also, please keep in mind that you still would need a fallback mechanism
>> to run your "deferred work" even when the guest does not call HLT/MWAIT,
>> like a regular timer in your main thread.
> Yeah. In that case I think the ideal answer is that we let the kernel
> scheduler sort it out. I was thinking of a model where we have I/O (or
> workqueue) threads in *addition* to the userspace exits on idle. The
> separate threads own the work (and a number of them are woken according
> to the queue depth), and idle vCPUs *opportunistically* process work
> items on top of that.
>
> That approach alone would work fine with the existing HLT scheduling;
> it's just MWAIT which is a pain because yield() doesn't really do much
> (but as noted, it's better than *nothing*).
>
>> On top of all this, I'm not sure it's more efficient to do the trap to
>> the vCPU thread compared to just creating a separate real thread. Your
>> main problem is the emulatability of MWAIT because that leaves "no time"
>> to do deferred work. But then again, if your deferred work is so complex
>> that it needs more than a few ms (which you can always steal from the
>> vCPU thread, especiall with yield()), you'll need to start implementing
>> time slicing of that work in user space next - and basically rebuild
>> your own scheduler there. Ugh.
>>
>> IMHO the real core value of this idea would be in a vcpu_run bit that on
>> VCPU_RUN can toggle between HLT/MWAIT intercept on and off. The actual
>> trap to user space, you're most likely better off with a separate thread.
> No, that's very much not the point. The problem is that yield() doesn't
> work well enough — and isn't designed or guaranteed to do anything in
> particular for most cases. It's better than *nothing* but we want the
> opportunity to do the actual work right there in the *loop* of the
> guest bouncing through MWAIT.
The problem with MWAIT is that you don't really know when it's done.
You could find out by making MONITOR'ed pages(!) read-only so you can
wake up any target vCPU that's in MWAIT, but that's considerably
expensive if you want to do it well.
You could also burn one VM/system wide CPU that does nothing but waits
for changes in any MONITOR'ed cache line. Doable with less power
consumption if you use TSX I guess. But probably not what you want either.
Another alternative would be to make guests PV aware, so they understand
you don't actually do MWAIT and give you a hypercall every time they
modify whatever anyone would want to monitor (such as
thread_info->flags). But that requires new guest kernels. I don't think
you want to wait for that :).
So in a nutshell, emulating MWAIT properly is just super difficult. If
you have even the remotest chance to get away with doing HLT instead,
I'd take that. In that model, an I/O thread that schedules over idle
threads becomes natural.
Alex
Amazon Development Center Germany GmbH
Krausenstr. 38
10117 Berlin
Geschaeftsfuehrung: Christian Schlaeger, Jonathan Weiss
Eingetragen am Amtsgericht Charlottenburg unter HRB 149173 B
Sitz: Berlin
Ust-ID: DE 289 237 879
On Mon, Sep 18, 2023 at 01:59:50PM +0200, Alexander Graf wrote:
> The problem with MWAIT is that you don't really know when it's done.
This isn't really a problem. MWAIT is allowed (expected even) to return
early.
REP;NOP is a valid implementation of MWAIT.
MWAIT must not delay waking (much) after either:
- write to monitored address
- interrupt pending
But it doesn't say anything about not waking up sooner.
Now, obviously on real hardware you prefer if MWAIT were to also do the
whole C-state thing and safe your some actual power, but this is virt,
real hardware is not a concern and wakeup-timeliness also not much.
IIRC the ARM64 WFE thing has a 10khz timer or something it wakes from if
nothing else. So I suppose what I'm saying is that: nanosleep(100000)
might be a suitable MWAIT implementation.
It's virt, it sucks anyway :-)