Arch-neutral name for hardware interrupt support

Arch-neutral name for hardware interrupt support

[Andrew Cooper]

Hello,

On ARM, the GIC is a hard prerequisite for VMs.

I can't remember what the state of RISCV is, but IIRC there is still
some debate over how interrupts are expected to work under virt.

On x86, the story is very different.  PV have no hardware assistance,
while HVM hardware assistance depends on hardware support.  Therefore we
want to introduce a new CDF flag so we can control the setting per
domain, rather than globally as it is done now.

This brings us to the question of what a suitable architecture name
would be.

Hardware Virtual Interrupts is a little too close to Hardware Virtual
(Machine) Introspection, and Hardware Assisted Interrupts (mirroring
HAP) doesn't seem great either.

Thoughts / ideas / suggestions?

~Andrew

Re: Arch-neutral name for hardware interrupt support

[Stefano Stabellini]

[-- Attachment #1: Type: text/plain, Size: 876 bytes --]

+Bertrand

On Mon, 1 Nov 2021, Andrew Cooper wrote:
> Hello,
> 
> On ARM, the GIC is a hard prerequisite for VMs.
> 
> I can't remember what the state of RISCV is, but IIRC there is still
> some debate over how interrupts are expected to work under virt.
> 
> On x86, the story is very different.  PV have no hardware assistance,
> while HVM hardware assistance depends on hardware support.  Therefore we
> want to introduce a new CDF flag so we can control the setting per
> domain, rather than globally as it is done now.
> 
> This brings us to the question of what a suitable architecture name
> would be.
> 
> Hardware Virtual Interrupts is a little too close to Hardware Virtual
> (Machine) Introspection, and Hardware Assisted Interrupts (mirroring
> HAP) doesn't seem great either.
> 
> Thoughts / ideas / suggestions?

Maybe "hardware-enabled virtualized interrupts"

Re: Arch-neutral name for hardware interrupt support

[Bobby Eshleman]

On Mon, Nov 01, 2021 at 10:47:26AM +0000, Andrew Cooper wrote:
> Hello,
> 
> On ARM, the GIC is a hard prerequisite for VMs.
> 
> I can't remember what the state of RISCV is, but IIRC there is still
> some debate over how interrupts are expected to work under virt.
> 

Yes sir, that addition to the spec is not yet released (publically) AFAICT.

Re: Arch-neutral name for hardware interrupt support

[Alistair Francis]

On Tue, Nov 2, 2021 at 6:33 AM Stefano Stabellini
<sstabellini@kernel.org> wrote:
>
> +Bertrand
>
> On Mon, 1 Nov 2021, Andrew Cooper wrote:
> > Hello,
> >
> > On ARM, the GIC is a hard prerequisite for VMs.
> >
> > I can't remember what the state of RISCV is, but IIRC there is still
> > some debate over how interrupts are expected to work under virt.

We are getting there, the current draft is pretty stable:
https://github.com/riscv/riscv-aia/

> >
> > On x86, the story is very different.  PV have no hardware assistance,
> > while HVM hardware assistance depends on hardware support.  Therefore we
> > want to introduce a new CDF flag so we can control the setting per
> > domain, rather than globally as it is done now.
> >
> > This brings us to the question of what a suitable architecture name
> > would be.
> >
> > Hardware Virtual Interrupts is a little too close to Hardware Virtual
> > (Machine) Introspection, and Hardware Assisted Interrupts (mirroring
> > HAP) doesn't seem great either.
> >
> > Thoughts / ideas / suggestions?
>
> Maybe "hardware-enabled virtualized interrupts"

hardware-supported virtualised interrupts (HSV Interrupts)

Alistair

Re: Arch-neutral name for hardware interrupt support

[Bertrand Marquis]

Hi,

> On 2 Nov 2021, at 07:16, Alistair Francis <alistair23@gmail.com> wrote:
> 
> On Tue, Nov 2, 2021 at 6:33 AM Stefano Stabellini
> <sstabellini@kernel.org> wrote:
>> 
>> +Bertrand
>> 
>> On Mon, 1 Nov 2021, Andrew Cooper wrote:
>>> Hello,
>>> 
>>> On ARM, the GIC is a hard prerequisite for VMs.
>>> 
>>> I can't remember what the state of RISCV is, but IIRC there is still
>>> some debate over how interrupts are expected to work under virt.
> 
> We are getting there, the current draft is pretty stable:
> https://github.com/riscv/riscv-aia/
> 
>>> 
>>> On x86, the story is very different.  PV have no hardware assistance,
>>> while HVM hardware assistance depends on hardware support.  Therefore we
>>> want to introduce a new CDF flag so we can control the setting per
>>> domain, rather than globally as it is done now.
>>> 
>>> This brings us to the question of what a suitable architecture name
>>> would be.
>>> 
>>> Hardware Virtual Interrupts is a little too close to Hardware Virtual
>>> (Machine) Introspection, and Hardware Assisted Interrupts (mirroring
>>> HAP) doesn't seem great either.
>>> 
>>> Thoughts / ideas / suggestions?
>> 
>> Maybe "hardware-enabled virtualized interrupts"
> 
> hardware-supported virtualised interrupts (HSV Interrupts)

Both propositions are ok I think but I have a preference for supported.

Cheers
Bertrand

> 
> Alistair

Re: Arch-neutral name for hardware interrupt support

[Roger Pau Monné]

On Mon, Nov 01, 2021 at 10:47:26AM +0000, Andrew Cooper wrote:
> Hello,
> 
> On ARM, the GIC is a hard prerequisite for VMs.
> 
> I can't remember what the state of RISCV is, but IIRC there is still
> some debate over how interrupts are expected to work under virt.
> 
> On x86, the story is very different.  PV have no hardware assistance,
> while HVM hardware assistance depends on hardware support.  Therefore we
> want to introduce a new CDF flag so we can control the setting per
> domain, rather than globally as it is done now.
> 
> This brings us to the question of what a suitable architecture name
> would be.
> 
> Hardware Virtual Interrupts is a little too close to Hardware Virtual
> (Machine) Introspection, and Hardware Assisted Interrupts (mirroring
> HAP) doesn't seem great either.
> 
> Thoughts / ideas / suggestions?

Is this going to cover both hardware assisted interrupt
controller and assisted interrupt delivery (ie: APIC virtualization
and posted interrupts in x86 speak) or just one of them?

Thanks, Roger.

Re: Arch-neutral name for hardware interrupt support

[Jan Beulich]

On 01.11.2021 11:47, Andrew Cooper wrote:
> Hello,
> 
> On ARM, the GIC is a hard prerequisite for VMs.
> 
> I can't remember what the state of RISCV is, but IIRC there is still
> some debate over how interrupts are expected to work under virt.
> 
> On x86, the story is very different.  PV have no hardware assistance,
> while HVM hardware assistance depends on hardware support.  Therefore we
> want to introduce a new CDF flag so we can control the setting per
> domain, rather than globally as it is done now.
> 
> This brings us to the question of what a suitable architecture name
> would be.
> 
> Hardware Virtual Interrupts is a little too close to Hardware Virtual
> (Machine) Introspection, and Hardware Assisted Interrupts (mirroring
> HAP) doesn't seem great either.
> 
> Thoughts / ideas / suggestions?

As per the subject of your mail simply "hwint"? But, as Roger says, a
good name certainly depends on the scope of what exactly you want to
cover.

Jan

Re: Arch-neutral name for hardware interrupt support

[Jane Malalane]

[-- Attachment #1: Type: text/plain, Size: 1192 bytes --]

On 02/11/2021 08:51, Roger Pau Monné wrote:
> On Mon, Nov 01, 2021 at 10:47:26AM +0000, Andrew Cooper wrote:
>> Hello,
>>
>> On ARM, the GIC is a hard prerequisite for VMs.
>>
>> I can't remember what the state of RISCV is, but IIRC there is still
>> some debate over how interrupts are expected to work under virt.
>>
>> On x86, the story is very different.  PV have no hardware assistance,
>> while HVM hardware assistance depends on hardware support.  Therefore we
>> want to introduce a new CDF flag so we can control the setting per
>> domain, rather than globally as it is done now.
>>
>> This brings us to the question of what a suitable architecture name
>> would be.
>>
>> Hardware Virtual Interrupts is a little too close to Hardware Virtual
>> (Machine) Introspection, and Hardware Assisted Interrupts (mirroring
>> HAP) doesn't seem great either.
>>
>> Thoughts / ideas / suggestions?
> Is this going to cover both hardware assisted interrupt
> controller and assisted interrupt delivery (ie: APIC virtualization
> and posted interrupts in x86 speak) or just one of them?
>
> Thanks, Roger.

It's not going to cover IOMMU posted interrupts, only CPU posted interrupts.


Jane.

[-- Attachment #2: Type: text/html, Size: 1605 bytes --]

Re: Arch-neutral name for hardware interrupt support

[Roger Pau Monné]

On Wed, Nov 03, 2021 at 10:16:26AM +0000, Jane Malalane wrote:
> On 02/11/2021 08:51, Roger Pau Monné wrote:
> > On Mon, Nov 01, 2021 at 10:47:26AM +0000, Andrew Cooper wrote:
> > > Hello,
> > > 
> > > On ARM, the GIC is a hard prerequisite for VMs.
> > > 
> > > I can't remember what the state of RISCV is, but IIRC there is still
> > > some debate over how interrupts are expected to work under virt.
> > > 
> > > On x86, the story is very different.  PV have no hardware assistance,
> > > while HVM hardware assistance depends on hardware support.  Therefore we
> > > want to introduce a new CDF flag so we can control the setting per
> > > domain, rather than globally as it is done now.
> > > 
> > > This brings us to the question of what a suitable architecture name
> > > would be.
> > > 
> > > Hardware Virtual Interrupts is a little too close to Hardware Virtual
> > > (Machine) Introspection, and Hardware Assisted Interrupts (mirroring
> > > HAP) doesn't seem great either.
> > > 
> > > Thoughts / ideas / suggestions?
> > Is this going to cover both hardware assisted interrupt
> > controller and assisted interrupt delivery (ie: APIC virtualization
> > and posted interrupts in x86 speak) or just one of them?
> > 
> > Thanks, Roger.
> 
> It's not going to cover IOMMU posted interrupts, only CPU posted interrupts.

IMO it needs to be something like Hardware {assisted,virtual}
interrupt controller, or some such, to differentiate from IOMMU posted
interrupts, which will likely use Hardware {assisted,virtual}
interrupts. So:

 * Hardware {assisted,virtual} interrupt controller: all APIC related
   hardware assisted emulation.
 * Hardware {assisted,virtual} interrupts: IOMMU posted interrupts.

I prefer `assisted` rather than `virtual`, but I'm not a native
speaker so my taste could go against what most prefer.

Regards, Roger.