content: support SR-IOV

PATCH Superseded Tiwei Bie 3 participants 9 messages May 22, 2018 – May 24, 2018

lore

Share & tools

lore

Tiwei Bie

author

2018-05-22 10:26 UTC #0 lore raw

PATCH v3 1/1 content: support SR-IOV

Allocate a feature bit for virtio devices which support SR-IOV.

Signed-off-by Tiwei Bie <[email protected]>

Fixes https://github.com/oasis-tcs/virtio-spec/issues/11

---
More details can be found from this thread:
https://patchwork.kernel.org/patch/10285541/

This patch needs below patch applied first:
https://github.com/oasis-tcs/virtio-spec/issues/10
https://lists.oasis-open.org/archives/virtio-dev/201805/msg00046.html

v2 -> v3:
- Improve the wording (Cornelia);

v1 -> v2:
- s/Reserve/Allocate/ (MST);
- Add a Fixes tag (MST);
- Be more explicit in driver requirement (MST);
- Remove the "device MAY fail" description (MST);
- Rebase on IO_BARRIER patch;

RFC -> v1:
- Mention PCI in the description (Cornelia);

 content.tex | 15 +++++++++++++--
 1 file changed, 13 insertions(+), 2 deletions(-)

content.tex +13 -2

index 95c243f..e9e6f9a 100644
@@ -95,10 +95,10 @@ Feature bits are allocated as follows:
 \begin{description}
 \item[0 to 23] Feature bits for the specific device type
 
-\item[24 to 36] Feature bits reserved for extensions to the queue and
+\item[24 to 37] Feature bits reserved for extensions to the queue and
   feature negotiation mechanisms
 
-\item[37 and above] Feature bits reserved for future extensions.
+\item[38 and above] Feature bits reserved for future extensions.
 \end{description}
 
 \begin{note}

@@ -5357,6 +5357,9 @@ Descriptors} and \ref{sec:Packed Virtqueues / Indirect Flag: Scatter-Gather Supp
   better performance.  This feature indicates whether
   a stronger form of barrier suitable for hardware
   devices is necessary.
+  \item[VIRTIO_F_SR_IOV(37)] This feature indicates that
+  the device supports Single Root I/O Virtualization.
+  Currently only PCI devices support this feature.
 \end{description}
 
 \drivernormative{\section}{Reserved Feature Bits}{Reserved Feature Bits}

@@ -5376,6 +5379,11 @@ A driver SHOULD accept VIRTIO_F_IO_BARRIER if it is offered.
 If VIRTIO_F_IO_BARRIER has been negotiated, a driver MUST use
 the barriers suitable for hardware devices.
 
+A driver SHOULD accept VIRTIO_F_SR_IOV if it is offered.
+If VIRTIO_F_SR_IOV has been negotiated, a driver can
+enable virtual functions through the device's PCI SR-IOV
+capability structure.
+
 \devicenormative{\section}{Reserved Feature Bits}{Reserved Feature Bits}
 
 A device MUST offer VIRTIO_F_VERSION_1.  A device MAY fail to operate further

@@ -5392,6 +5400,9 @@ buffers in the same order in which they have been available.
 A device MAY fail to operate further if VIRTIO_F_IO_BARRIER
 is not accepted.
 
+A device SHOULD offer VIRTIO_F_SR_IOV if it presents a PCI
+SR-IOV capability structure.
+
 \section{Legacy Interface: Reserved Feature Bits}\label{sec:Reserved Feature Bits / Legacy Interface: Reserved Feature Bits}
 
 Transitional devices MAY offer the following:

Signature · 6 lines

2.17.0


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Cornelia Huck

Reviewed-by on v3 1/1

2018-05-22 11:03 UTC #1 lore raw

Quoted 5 lines · On Tue, 22 May 2018 18:26:15 +0800 Tiwei Bie <[email protected]> wrote:

Allocate a feature bit for virtio devices which support SR-IOV.
Suggested-by Michael S. Tsirkin <[email protected]>
Signed-off-by Tiwei Bie <[email protected]>
Fixes https://github.com/oasis-tcs/virtio-spec/issues/11

Reviewed-by Cornelia Huck <[email protected]>

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]

Michael S. Tsirkin

2018-05-23 17:54 UTC #2 lore raw

In reply to Tiwei Bie

Quoted 47 lines · Tiwei Bie

Allocate a feature bit for virtio devices which support SR-IOV.
Suggested-by Michael S. Tsirkin <[email protected]>
Signed-off-by Tiwei Bie <[email protected]>
Fixes https://github.com/oasis-tcs/virtio-spec/issues/11
---
More details can be found from this thread:
https://patchwork.kernel.org/patch/10285541/
This patch needs below patch applied first:
https://github.com/oasis-tcs/virtio-spec/issues/10
https://lists.oasis-open.org/archives/virtio-dev/201805/msg00046.html
v2 -> v3:
- Improve the wording (Cornelia);
v1 -> v2:
- s/Reserve/Allocate/ (MST);
- Add a Fixes tag (MST);
- Be more explicit in driver requirement (MST);
- Remove the "device MAY fail" description (MST);
- Rebase on IO_BARRIER patch;
RFC -> v1:
- Mention PCI in the description (Cornelia);
 content.tex | 15 +++++++++++++--
 1 file changed, 13 insertions(+), 2 deletions(-)
content.tex +2 -2
index 95c243f..e9e6f9a 100644
@@ -95,10 +95,10 @@ Feature bits are allocated as follows:
 \begin{description}
 \item[0 to 23] Feature bits for the specific device type
 
-\item[24 to 36] Feature bits reserved for extensions to the queue and
+\item[24 to 37] Feature bits reserved for extensions to the queue and
   feature negotiation mechanisms
 
-\item[37 and above] Feature bits reserved for future extensions.
+\item[38 and above] Feature bits reserved for future extensions.
 \end{description}
 
 \begin{note}

content.tex Descriptors} and \ref{sec:Packed Virtqueues / Indirect Flag: Scatter-Gather Supp · line 5362

Quoted 6 lines

content.tex +3 -0

@@ -5357,6 +5357,9 @@ Descriptors} and \ref{sec:Packed Virtqueues / Indirect Flag: Scatter-Gather Supp
   better performance.  This feature indicates whether
   a stronger form of barrier suitable for hardware
   devices is necessary.
+  \item[VIRTIO_F_SR_IOV(37)] This feature indicates that
+  the device supports Single Root I/O Virtualization.
+  Currently only PCI devices support this feature.

I guess the assumption is that all VFs and the PF are of the same type?

I feel it might be handy down the road to support mixing types. For this reason, to avoid binding a wrong driver to a VF, I propose that all VFs have this bit too, and require that drivers ignore VFs without this bit.

What do you think?

content.tex A driver SHOULD accept VIRTIO_F_IO_BARRIER if it is offered. · line 5385

Quoted 11 lines

content.tex +4 -0

 \end{description}
 
 \drivernormative{\section}{Reserved Feature Bits}{Reserved Feature Bits}

@@ -5376,6 +5379,11 @@ A driver SHOULD accept VIRTIO_F_IO_BARRIER if it is offered.
 If VIRTIO_F_IO_BARRIER has been negotiated, a driver MUST use
 the barriers suitable for hardware devices.
 
+A driver SHOULD accept VIRTIO_F_SR_IOV if it is offered.
+If VIRTIO_F_SR_IOV has been negotiated, a driver can
+enable virtual functions through the device's PCI SR-IOV
+capability structure.

I feel the last sentence isn't clear enough. How about

a driver MUST negotiate VIRTIO_F_SR_IOV and complete the feature negotiation (including setting the DRIVER_OK \field{status} bit) before enabling virtual functions through the device's PCI SR-IOV capability structure.

Quoted 24 lines

content.tex +4 -0

+
 \devicenormative{\section}{Reserved Feature Bits}{Reserved Feature Bits}
 
 A device MUST offer VIRTIO_F_VERSION_1.  A device MAY fail to operate further

@@ -5392,6 +5400,9 @@ buffers in the same order in which they have been available.
 A device MAY fail to operate further if VIRTIO_F_IO_BARRIER
 is not accepted.
 
+A device SHOULD offer VIRTIO_F_SR_IOV if it presents a PCI
+SR-IOV capability structure.
+
 \section{Legacy Interface: Reserved Feature Bits}\label{sec:Reserved Feature Bits / Legacy Interface: Reserved Feature Bits}
 
 Transitional devices MAY offer the following:

Signature · 7 lines

-- 
2.17.0


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]

Michael S. Tsirkin

2018-05-23 19:34 UTC #3 lore raw

Quoted 57 lines · Michael S. Tsirkin

Quoted 47 lines · Tiwei Bie
Allocate a feature bit for virtio devices which support SR-IOV.
Suggested-by Michael S. Tsirkin <[email protected]>
Signed-off-by Tiwei Bie <[email protected]>
Fixes https://github.com/oasis-tcs/virtio-spec/issues/11
---
More details can be found from this thread:
https://patchwork.kernel.org/patch/10285541/
This patch needs below patch applied first:
https://github.com/oasis-tcs/virtio-spec/issues/10
https://lists.oasis-open.org/archives/virtio-dev/201805/msg00046.html
v2 -> v3:
- Improve the wording (Cornelia);
v1 -> v2:
- s/Reserve/Allocate/ (MST);
- Add a Fixes tag (MST);
- Be more explicit in driver requirement (MST);
- Remove the "device MAY fail" description (MST);
- Rebase on IO_BARRIER patch;
RFC -> v1:
- Mention PCI in the description (Cornelia);
 content.tex | 15 +++++++++++++--
 1 file changed, 13 insertions(+), 2 deletions(-)
content.tex +2 -2
index 95c243f..e9e6f9a 100644
@@ -95,10 +95,10 @@ Feature bits are allocated as follows:
 \begin{description}
 \item[0 to 23] Feature bits for the specific device type
 
-\item[24 to 36] Feature bits reserved for extensions to the queue and
+\item[24 to 37] Feature bits reserved for extensions to the queue and
   feature negotiation mechanisms
 
-\item[37 and above] Feature bits reserved for future extensions.
+\item[38 and above] Feature bits reserved for future extensions.
 \end{description}
 
 \begin{note}
content.tex Descriptors} and \ref{sec:Packed Virtqueues / Indirect Flag: Scatter-Gather Supp · line 5362
Quoted 6 lines
content.tex +3 -0
@@ -5357,6 +5357,9 @@ Descriptors} and \ref{sec:Packed Virtqueues / Indirect Flag: Scatter-Gather Supp
   better performance.  This feature indicates whether
   a stronger form of barrier suitable for hardware
   devices is necessary.
+  \item[VIRTIO_F_SR_IOV(37)] This feature indicates that
+  the device supports Single Root I/O Virtualization.
+  Currently only PCI devices support this feature.
I guess the assumption is that all VFs and the PF are of the same type?

Quoted 6 lines

I feel it might be handy down the road to support mixing types. For this reason, to avoid binding a wrong driver to a VF, I propose that all VFs have this bit too, and require that drivers ignore VFs without this bit.
What do you think?

Thinking more about it, I can see how this might interfere with passing VFs through to legacy nested guests. How about reversing it then?

Require that drivers MUST NOT negotiate VIRTIO_F_SR_IOV if device does not have an SRIOV capability or is not a PCI device, in particular a VF.

And say a device without SRIOV cap SHOULD NOT expose this bit.

Quoted 14 lines

content.tex A driver SHOULD accept VIRTIO_F_IO_BARRIER if it is offered. · line 5385

Quoted 11 lines

content.tex +4 -0

 \end{description}
 
 \drivernormative{\section}{Reserved Feature Bits}{Reserved Feature Bits}

@@ -5376,6 +5379,11 @@ A driver SHOULD accept VIRTIO_F_IO_BARRIER if it is offered.
 If VIRTIO_F_IO_BARRIER has been negotiated, a driver MUST use
 the barriers suitable for hardware devices.
 
+A driver SHOULD accept VIRTIO_F_SR_IOV if it is offered.
+If VIRTIO_F_SR_IOV has been negotiated, a driver can
+enable virtual functions through the device's PCI SR-IOV
+capability structure.

I feel the last sentence isn't clear enough. How about

Quoted 29 lines

Quoted 24 lines

content.tex +4 -0

+
 \devicenormative{\section}{Reserved Feature Bits}{Reserved Feature Bits}
 
 A device MUST offer VIRTIO_F_VERSION_1.  A device MAY fail to operate further

@@ -5392,6 +5400,9 @@ buffers in the same order in which they have been available.
 A device MAY fail to operate further if VIRTIO_F_IO_BARRIER
 is not accepted.
 
+A device SHOULD offer VIRTIO_F_SR_IOV if it presents a PCI
+SR-IOV capability structure.
+
 \section{Legacy Interface: Reserved Feature Bits}\label{sec:Reserved Feature Bits / Legacy Interface: Reserved Feature Bits}
 
 Transitional devices MAY offer the following:

Signature · 7 lines

-- 
2.17.0


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]

Tiwei Bie

author

2018-05-24 00:06 UTC #4 lore raw

Quoted 65 lines · Michael S. Tsirkin

Quoted 57 lines · Michael S. Tsirkin
Quoted 47 lines · Tiwei Bie
Allocate a feature bit for virtio devices which support SR-IOV.
Suggested-by Michael S. Tsirkin <[email protected]>
Signed-off-by Tiwei Bie <[email protected]>
Fixes https://github.com/oasis-tcs/virtio-spec/issues/11
---
More details can be found from this thread:
https://patchwork.kernel.org/patch/10285541/
This patch needs below patch applied first:
https://github.com/oasis-tcs/virtio-spec/issues/10
https://lists.oasis-open.org/archives/virtio-dev/201805/msg00046.html
v2 -> v3:
- Improve the wording (Cornelia);
v1 -> v2:
- s/Reserve/Allocate/ (MST);
- Add a Fixes tag (MST);
- Be more explicit in driver requirement (MST);
- Remove the "device MAY fail" description (MST);
- Rebase on IO_BARRIER patch;
RFC -> v1:
- Mention PCI in the description (Cornelia);
 content.tex | 15 +++++++++++++--
 1 file changed, 13 insertions(+), 2 deletions(-)
content.tex +2 -2
index 95c243f..e9e6f9a 100644
@@ -95,10 +95,10 @@ Feature bits are allocated as follows:
 \begin{description}
 \item[0 to 23] Feature bits for the specific device type
 
-\item[24 to 36] Feature bits reserved for extensions to the queue and
+\item[24 to 37] Feature bits reserved for extensions to the queue and
   feature negotiation mechanisms
 
-\item[37 and above] Feature bits reserved for future extensions.
+\item[38 and above] Feature bits reserved for future extensions.
 \end{description}
 
 \begin{note}
content.tex Descriptors} and \ref{sec:Packed Virtqueues / Indirect Flag: Scatter-Gather Supp · line 5362
Quoted 6 lines
content.tex +3 -0
@@ -5357,6 +5357,9 @@ Descriptors} and \ref{sec:Packed Virtqueues / Indirect Flag: Scatter-Gather Supp
   better performance.  This feature indicates whether
   a stronger form of barrier suitable for hardware
   devices is necessary.
+  \item[VIRTIO_F_SR_IOV(37)] This feature indicates that
+  the device supports Single Root I/O Virtualization.
+  Currently only PCI devices support this feature.
I guess the assumption is that all VFs and the PF are of the same type?
Quoted 6 lines
I feel it might be handy down the road to support mixing types. For this reason, to avoid binding a wrong driver to a VF, I propose that all VFs have this bit too, and require that drivers ignore VFs without this bit.
What do you think?

Quoted 7 lines

Thinking more about it, I can see how this might interfere with passing VFs through to legacy nested guests. How about reversing it then?
Require that drivers MUST NOT negotiate VIRTIO_F_SR_IOV if device does not have an SRIOV capability or is not a PCI device, in particular a VF.

I think driver can accept this feature as long as it's able to handle the SR-IOV capability and there is no need for it to check whether the device has the SR-IOV capability. And device should make sure that it won't offer this feature if it doesn't present this capability.

How about changing the driver requirement to:

A driver SHOULD accept VIRTIO_F_SR_IOV if it is offered. If VIRTIO_F_SR_IOV has been negotiated, a driver can enable virtual functions through the device's PCI SR-IOV capability structure. A driver MUST negotiate VIRTIO_F_SR_IOV and complete the feature negotiation (including setting the DRIVER_OK \field{status} bit) before enabling virtual functions through the device's PCI SR-IOV capability structure.

Quoted 3 lines

And say a device without SRIOV cap SHOULD NOT expose this bit.

No problem. How about:

A device SHOULD offer VIRTIO_F_SR_IOV if it presents a PCI SR-IOV capability structure. A device SHOULD NOT offer VIRTIO_F_SR_IOV if it doesn't presents a PCI SR-IOV capability structure.

Quoted 38 lines

content.tex A driver SHOULD accept VIRTIO_F_IO_BARRIER if it is offered. · line 5385

Quoted 13 lines

content.tex +4 -0

 \end{description}
 
 \drivernormative{\section}{Reserved Feature Bits}{Reserved Feature Bits}

@@ -5376,6 +5379,11 @@ A driver SHOULD accept VIRTIO_F_IO_BARRIER if it is offered.
 If VIRTIO_F_IO_BARRIER has been negotiated, a driver MUST use
 the barriers suitable for hardware devices.
 
+A driver SHOULD accept VIRTIO_F_SR_IOV if it is offered.
+If VIRTIO_F_SR_IOV has been negotiated, a driver can
+enable virtual functions through the device's PCI SR-IOV
+capability structure.

I feel the last sentence isn't clear enough. How about

Quoted 14 lines

content.tex +4 -0

+
 \devicenormative{\section}{Reserved Feature Bits}{Reserved Feature Bits}
 
 A device MUST offer VIRTIO_F_VERSION_1.  A device MAY fail to operate further

@@ -5392,6 +5400,9 @@ buffers in the same order in which they have been available.
 A device MAY fail to operate further if VIRTIO_F_IO_BARRIER
 is not accepted.
 
+A device SHOULD offer VIRTIO_F_SR_IOV if it presents a PCI
+SR-IOV capability structure.
+
 \section{Legacy Interface: Reserved Feature Bits}\label{sec:Reserved Feature Bits / Legacy Interface: Reserved Feature Bits}
 
 Transitional devices MAY offer the following:

Quoted 7 lines

Signature · 6 lines

2.17.0


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]

Michael S. Tsirkin

2018-05-24 13:44 UTC #5 lore raw

Quoted 70 lines · Tiwei Bie

Quoted 69 lines · Michael S. Tsirkin
Quoted 57 lines · Michael S. Tsirkin
Quoted 47 lines · Tiwei Bie
Allocate a feature bit for virtio devices which support SR-IOV.
Suggested-by Michael S. Tsirkin <[email protected]>
Signed-off-by Tiwei Bie <[email protected]>
Fixes https://github.com/oasis-tcs/virtio-spec/issues/11
---
More details can be found from this thread:
https://patchwork.kernel.org/patch/10285541/
This patch needs below patch applied first:
https://github.com/oasis-tcs/virtio-spec/issues/10
https://lists.oasis-open.org/archives/virtio-dev/201805/msg00046.html
v2 -> v3:
- Improve the wording (Cornelia);
v1 -> v2:
- s/Reserve/Allocate/ (MST);
- Add a Fixes tag (MST);
- Be more explicit in driver requirement (MST);
- Remove the "device MAY fail" description (MST);
- Rebase on IO_BARRIER patch;
RFC -> v1:
- Mention PCI in the description (Cornelia);
 content.tex | 15 +++++++++++++--
 1 file changed, 13 insertions(+), 2 deletions(-)
content.tex +2 -2
index 95c243f..e9e6f9a 100644
@@ -95,10 +95,10 @@ Feature bits are allocated as follows:
 \begin{description}
 \item[0 to 23] Feature bits for the specific device type
 
-\item[24 to 36] Feature bits reserved for extensions to the queue and
+\item[24 to 37] Feature bits reserved for extensions to the queue and
   feature negotiation mechanisms
 
-\item[37 and above] Feature bits reserved for future extensions.
+\item[38 and above] Feature bits reserved for future extensions.
 \end{description}
 
 \begin{note}
content.tex Descriptors} and \ref{sec:Packed Virtqueues / Indirect Flag: Scatter-Gather Supp · line 5362
Quoted 6 lines
content.tex +3 -0
@@ -5357,6 +5357,9 @@ Descriptors} and \ref{sec:Packed Virtqueues / Indirect Flag: Scatter-Gather Supp
   better performance.  This feature indicates whether
   a stronger form of barrier suitable for hardware
   devices is necessary.
+  \item[VIRTIO_F_SR_IOV(37)] This feature indicates that
+  the device supports Single Root I/O Virtualization.
+  Currently only PCI devices support this feature.
I guess the assumption is that all VFs and the PF are of the same type?
Quoted 6 lines
I feel it might be handy down the road to support mixing types. For this reason, to avoid binding a wrong driver to a VF, I propose that all VFs have this bit too, and require that drivers ignore VFs without this bit.
What do you think?
Thinking more about it, I can see how this might interfere with passing VFs through to legacy nested guests. How about reversing it then?

Quoted 8 lines

Quoted 3 lines
Require that drivers MUST NOT negotiate VIRTIO_F_SR_IOV if device does not have an SRIOV capability or is not a PCI device, in particular a VF.
I think driver can accept this feature as long as it's able to handle the SR-IOV capability and there is no need for it to check whether the device has the SR-IOV capability.

So my point is this, VFs themselves do not have
this feature.

Should all of them have it? None of them?
I don't see what use it is to VFs, but maybe
we will come with a use down the road.

I propose we require that
1. drivers ignore this if there is
no SRIOV cap, and

2. that devices do not expose it.

This way if we come up with a use down the road, only new drivers will negotiate it.

Quoted 5 lines

And device should make sure that it won't
offer this feature if it doesn't present this capability. How about changing the driver requirement to:
A driver SHOULD accept VIRTIO_F_SR_IOV if it is offered.

This part won't address the issue above.

Quoted 10 lines

If VIRTIO_F_SR_IOV has been negotiated, a driver can enable virtual functions through the device's PCI SR-IOV capability structure. A driver MUST negotiate VIRTIO_F_SR_IOV and complete the feature negotiation (including setting the DRIVER_OK \field{status} bit) before enabling virtual functions through the device's PCI SR-IOV capability structure.
Quoted 1 line

Quoted 8 lines

Quoted 2 lines
And say a device without SRIOV cap SHOULD NOT expose this bit.
No problem. How about:
A device SHOULD offer VIRTIO_F_SR_IOV if it presents a PCI SR-IOV capability structure. A device SHOULD NOT offer VIRTIO_F_SR_IOV if it doesn't presents

doesn't present

Quoted 2 lines

a PCI SR-IOV
capability structure.

Assuming we teach drivers they should ignore it
if it is there without SRIOV, then this last one I'd make MUST NOT.

Quoted 38 lines

content.tex A driver SHOULD accept VIRTIO_F_IO_BARRIER if it is offered. · line 5385

Quoted 13 lines

content.tex +4 -0

 \end{description}
 
 \drivernormative{\section}{Reserved Feature Bits}{Reserved Feature Bits}

@@ -5376,6 +5379,11 @@ A driver SHOULD accept VIRTIO_F_IO_BARRIER if it is offered.
 If VIRTIO_F_IO_BARRIER has been negotiated, a driver MUST use
 the barriers suitable for hardware devices.
 
+A driver SHOULD accept VIRTIO_F_SR_IOV if it is offered.
+If VIRTIO_F_SR_IOV has been negotiated, a driver can
+enable virtual functions through the device's PCI SR-IOV
+capability structure.

I feel the last sentence isn't clear enough. How about

Quoted 14 lines

content.tex +4 -0

+
 \devicenormative{\section}{Reserved Feature Bits}{Reserved Feature Bits}
 
 A device MUST offer VIRTIO_F_VERSION_1.  A device MAY fail to operate further

@@ -5392,6 +5400,9 @@ buffers in the same order in which they have been available.
 A device MAY fail to operate further if VIRTIO_F_IO_BARRIER
 is not accepted.
 
+A device SHOULD offer VIRTIO_F_SR_IOV if it presents a PCI
+SR-IOV capability structure.
+
 \section{Legacy Interface: Reserved Feature Bits}\label{sec:Reserved Feature Bits / Legacy Interface: Reserved Feature Bits}
 
 Transitional devices MAY offer the following:

Quoted 7 lines

Signature · 6 lines

2.17.0


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]

Tiwei Bie

author

2018-05-24 15:15 UTC #6 lore raw

Quoted 75 lines · Michael S. Tsirkin

Quoted 74 lines · Tiwei Bie
Quoted 72 lines · Michael S. Tsirkin
Quoted 57 lines · Michael S. Tsirkin
Quoted 47 lines · Tiwei Bie
Allocate a feature bit for virtio devices which support SR-IOV.
Suggested-by Michael S. Tsirkin <[email protected]>
Signed-off-by Tiwei Bie <[email protected]>
Fixes https://github.com/oasis-tcs/virtio-spec/issues/11
---
More details can be found from this thread:
https://patchwork.kernel.org/patch/10285541/
This patch needs below patch applied first:
https://github.com/oasis-tcs/virtio-spec/issues/10
https://lists.oasis-open.org/archives/virtio-dev/201805/msg00046.html
v2 -> v3:
- Improve the wording (Cornelia);
v1 -> v2:
- s/Reserve/Allocate/ (MST);
- Add a Fixes tag (MST);
- Be more explicit in driver requirement (MST);
- Remove the "device MAY fail" description (MST);
- Rebase on IO_BARRIER patch;
RFC -> v1:
- Mention PCI in the description (Cornelia);
 content.tex | 15 +++++++++++++--
 1 file changed, 13 insertions(+), 2 deletions(-)
content.tex +2 -2
index 95c243f..e9e6f9a 100644
@@ -95,10 +95,10 @@ Feature bits are allocated as follows:
 \begin{description}
 \item[0 to 23] Feature bits for the specific device type
 
-\item[24 to 36] Feature bits reserved for extensions to the queue and
+\item[24 to 37] Feature bits reserved for extensions to the queue and
   feature negotiation mechanisms
 
-\item[37 and above] Feature bits reserved for future extensions.
+\item[38 and above] Feature bits reserved for future extensions.
 \end{description}
 
 \begin{note}
content.tex Descriptors} and \ref{sec:Packed Virtqueues / Indirect Flag: Scatter-Gather Supp · line 5362
Quoted 6 lines
content.tex +3 -0
@@ -5357,6 +5357,9 @@ Descriptors} and \ref{sec:Packed Virtqueues / Indirect Flag: Scatter-Gather Supp
   better performance.  This feature indicates whether
   a stronger form of barrier suitable for hardware
   devices is necessary.
+  \item[VIRTIO_F_SR_IOV(37)] This feature indicates that
+  the device supports Single Root I/O Virtualization.
+  Currently only PCI devices support this feature.
I guess the assumption is that all VFs and the PF are of the same type?
Quoted 6 lines
I feel it might be handy down the road to support mixing types. For this reason, to avoid binding a wrong driver to a VF, I propose that all VFs have this bit too, and require that drivers ignore VFs without this bit.
What do you think?
Thinking more about it, I can see how this might interfere with passing VFs through to legacy nested guests. How about reversing it then?
Require that drivers MUST NOT negotiate VIRTIO_F_SR_IOV if device does not have an SRIOV capability or is not a PCI device, in particular a VF.

Quoted 7 lines

Quoted 4 lines
I think driver can accept this feature as long as it's able to handle the SR-IOV capability and there is no need for it to check whether the device has the SR-IOV capability.
So my point is this, VFs themselves do not have
this feature.

Yeah. I also think VFs shouldn't present this feature.

Quoted 5 lines

Should all of them have it? None of them?
I don't see what use it is to VFs, but maybe
we will come with a use down the road.

Quoted 8 lines

I propose we require that
1. drivers ignore this if there is
no SRIOV cap, and
2. that devices do not expose it.
This way if we come up with a use down the road, only new drivers will negotiate it.

I got your point now. Thanks!

How about:

If VIRTIO_F_SR_IOV has been negotiated, a driver can enable virtual functions through the device's PCI SR-IOV capability structure. A driver MUST NOT negotiate VIRTIO_F_SR_IOV if the device does not have a PCI SR-IOV capability structure or is not a PCI device. A driver MUST negotiate VIRTIO_F_SR_IOV and complete the feature negotiation (including setting the DRIVER_OK \field{status} bit) before enabling virtual functions through the device's PCI SR-IOV capability structure.

Quoted 25 lines

Quoted 8 lines
And device should make sure that it won't
offer this feature if it doesn't present this capability. How about changing the driver requirement to:
A driver SHOULD accept VIRTIO_F_SR_IOV if it is offered.
This part won't address the issue above.
Quoted 13 lines
If VIRTIO_F_SR_IOV has been negotiated, a driver can enable virtual functions through the device's PCI SR-IOV capability structure. A driver MUST negotiate VIRTIO_F_SR_IOV and complete the feature negotiation (including setting the DRIVER_OK \field{status} bit) before enabling virtual functions through the device's PCI SR-IOV capability structure.
Quoted 3 lines
And say a device without SRIOV cap SHOULD NOT expose this bit.

Quoted 7 lines

Quoted 5 lines
No problem. How about:
A device SHOULD offer VIRTIO_F_SR_IOV if it presents a PCI SR-IOV capability structure. A device SHOULD NOT offer VIRTIO_F_SR_IOV if it doesn't presents
doesn't present

Right. Thanks for catching it!

Quoted 6 lines

Quoted 3 lines
a PCI SR-IOV
capability structure.
Assuming we teach drivers they should ignore it
if it is there without SRIOV, then this last one I'd make MUST NOT.

Okay, how about

A device SHOULD offer VIRTIO_F_SR_IOV if it is a PCI device and presents a PCI SR-IOV capability structure, otherwise it MUST NOT offer VIRTIO_F_SR_IOV.

Best regards,
Tiwei Bie

Quoted 43 lines

Quoted 43 lines
Quoted 43 lines
Quoted 43 lines
Quoted 7 lines
content.tex +0 -0
 \end{description}
 
 \drivernormative{\section}{Reserved Feature Bits}{Reserved Feature Bits}
content.tex A driver SHOULD accept VIRTIO_F_IO_BARRIER if it is offered. · line 5385
Quoted 7 lines
content.tex +4 -0
@@ -5376,6 +5379,11 @@ A driver SHOULD accept VIRTIO_F_IO_BARRIER if it is offered.
 If VIRTIO_F_IO_BARRIER has been negotiated, a driver MUST use
 the barriers suitable for hardware devices.
 
+A driver SHOULD accept VIRTIO_F_SR_IOV if it is offered.
+If VIRTIO_F_SR_IOV has been negotiated, a driver can
+enable virtual functions through the device's PCI SR-IOV
+capability structure.
I feel the last sentence isn't clear enough. How about
a driver MUST negotiate VIRTIO_F_SR_IOV and complete the feature negotiation (including setting the DRIVER_OK \field{status} bit) before enabling virtual functions through the device's PCI SR-IOV capability structure.
Quoted 21 lines
content.tex +4 -0
+
 \devicenormative{\section}{Reserved Feature Bits}{Reserved Feature Bits}
 
 A device MUST offer VIRTIO_F_VERSION_1.  A device MAY fail to operate further
@@ -5392,6 +5400,9 @@ buffers in the same order in which they have been available.
 A device MAY fail to operate further if VIRTIO_F_IO_BARRIER
 is not accepted.
 
+A device SHOULD offer VIRTIO_F_SR_IOV if it presents a PCI
+SR-IOV capability structure.
+
 \section{Legacy Interface: Reserved Feature Bits}\label{sec:Reserved Feature Bits / Legacy Interface: Reserved Feature Bits}
 
 Transitional devices MAY offer the following:
Signature · 2 lines
-- 
2.17.0

Quoted 8 lines

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]

Michael S. Tsirkin

2018-05-24 15:20 UTC #7 lore raw

Quoted 104 lines · Tiwei Bie

Quoted 75 lines · Michael S. Tsirkin
Quoted 74 lines · Tiwei Bie
Quoted 72 lines · Michael S. Tsirkin
Quoted 57 lines · Michael S. Tsirkin
Quoted 47 lines · Tiwei Bie
Allocate a feature bit for virtio devices which support SR-IOV.
Suggested-by Michael S. Tsirkin <[email protected]>
Signed-off-by Tiwei Bie <[email protected]>
Fixes https://github.com/oasis-tcs/virtio-spec/issues/11
---
More details can be found from this thread:
https://patchwork.kernel.org/patch/10285541/
This patch needs below patch applied first:
https://github.com/oasis-tcs/virtio-spec/issues/10
https://lists.oasis-open.org/archives/virtio-dev/201805/msg00046.html
v2 -> v3:
- Improve the wording (Cornelia);
v1 -> v2:
- s/Reserve/Allocate/ (MST);
- Add a Fixes tag (MST);
- Be more explicit in driver requirement (MST);
- Remove the "device MAY fail" description (MST);
- Rebase on IO_BARRIER patch;
RFC -> v1:
- Mention PCI in the description (Cornelia);
 content.tex | 15 +++++++++++++--
 1 file changed, 13 insertions(+), 2 deletions(-)
content.tex +2 -2
index 95c243f..e9e6f9a 100644
@@ -95,10 +95,10 @@ Feature bits are allocated as follows:
 \begin{description}
 \item[0 to 23] Feature bits for the specific device type
 
-\item[24 to 36] Feature bits reserved for extensions to the queue and
+\item[24 to 37] Feature bits reserved for extensions to the queue and
   feature negotiation mechanisms
 
-\item[37 and above] Feature bits reserved for future extensions.
+\item[38 and above] Feature bits reserved for future extensions.
 \end{description}
 
 \begin{note}
content.tex Descriptors} and \ref{sec:Packed Virtqueues / Indirect Flag: Scatter-Gather Supp · line 5362
Quoted 6 lines
content.tex +3 -0
@@ -5357,6 +5357,9 @@ Descriptors} and \ref{sec:Packed Virtqueues / Indirect Flag: Scatter-Gather Supp
   better performance.  This feature indicates whether
   a stronger form of barrier suitable for hardware
   devices is necessary.
+  \item[VIRTIO_F_SR_IOV(37)] This feature indicates that
+  the device supports Single Root I/O Virtualization.
+  Currently only PCI devices support this feature.
I guess the assumption is that all VFs and the PF are of the same type?
Quoted 6 lines
I feel it might be handy down the road to support mixing types. For this reason, to avoid binding a wrong driver to a VF, I propose that all VFs have this bit too, and require that drivers ignore VFs without this bit.
What do you think?
Thinking more about it, I can see how this might interfere with passing VFs through to legacy nested guests. How about reversing it then?
Require that drivers MUST NOT negotiate VIRTIO_F_SR_IOV if device does not have an SRIOV capability or is not a PCI device, in particular a VF.
Quoted 7 lines
Quoted 4 lines
I think driver can accept this feature as long as it's able to handle the SR-IOV capability and there is no need for it to check whether the device has the SR-IOV capability.
So my point is this, VFs themselves do not have
this feature.
Yeah. I also think VFs shouldn't present this feature.
Quoted 5 lines
Should all of them have it? None of them?
I don't see what use it is to VFs, but maybe
we will come with a use down the road.
Quoted 8 lines
I propose we require that
1. drivers ignore this if there is
no SRIOV cap, and
2. that devices do not expose it.
This way if we come up with a use down the road, only new drivers will negotiate it.
I got your point now. Thanks!
How about:

Quoted 9 lines

If VIRTIO_F_SR_IOV has been negotiated, a driver can enable virtual functions through the device's PCI SR-IOV capability structure. A driver MUST NOT negotiate VIRTIO_F_SR_IOV if the device does not have a PCI SR-IOV capability structure or is not a PCI device. A driver MUST negotiate VIRTIO_F_SR_IOV and complete the feature negotiation (including setting the DRIVER_OK \field{status} bit) before enabling virtual functions through the device's PCI SR-IOV capability structure.

Sounds good.

Quoted 43 lines

Quoted 26 lines
Quoted 9 lines
And device should make sure that it won't
offer this feature if it doesn't present this capability. How about changing the driver requirement to:
A driver SHOULD accept VIRTIO_F_SR_IOV if it is offered.
This part won't address the issue above.
Quoted 13 lines
If VIRTIO_F_SR_IOV has been negotiated, a driver can enable virtual functions through the device's PCI SR-IOV capability structure. A driver MUST negotiate VIRTIO_F_SR_IOV and complete the feature negotiation (including setting the DRIVER_OK \field{status} bit) before enabling virtual functions through the device's PCI SR-IOV capability structure.
Quoted 3 lines
And say a device without SRIOV cap SHOULD NOT expose this bit.
Quoted 7 lines
Quoted 5 lines
No problem. How about:
A device SHOULD offer VIRTIO_F_SR_IOV if it presents a PCI SR-IOV capability structure. A device SHOULD NOT offer VIRTIO_F_SR_IOV if it doesn't presents
doesn't present
Right. Thanks for catching it!
Quoted 6 lines
Quoted 3 lines
a PCI SR-IOV
capability structure.
Assuming we teach drivers they should ignore it
if it is there without SRIOV, then this last one I'd make MUST NOT.

Quoted 8 lines

Okay, how about
A device SHOULD offer VIRTIO_F_SR_IOV if it is a PCI device and presents a PCI SR-IOV capability structure, otherwise it MUST NOT offer VIRTIO_F_SR_IOV.
Best regards,
Tiwei Bie

Sounds good.

Quoted 43 lines

Quoted 43 lines
Quoted 43 lines
Quoted 43 lines
Quoted 43 lines
Quoted 7 lines
content.tex +0 -0
 \end{description}
 
 \drivernormative{\section}{Reserved Feature Bits}{Reserved Feature Bits}
content.tex A driver SHOULD accept VIRTIO_F_IO_BARRIER if it is offered. · line 5385
Quoted 7 lines
content.tex +4 -0
@@ -5376,6 +5379,11 @@ A driver SHOULD accept VIRTIO_F_IO_BARRIER if it is offered.
 If VIRTIO_F_IO_BARRIER has been negotiated, a driver MUST use
 the barriers suitable for hardware devices.
 
+A driver SHOULD accept VIRTIO_F_SR_IOV if it is offered.
+If VIRTIO_F_SR_IOV has been negotiated, a driver can
+enable virtual functions through the device's PCI SR-IOV
+capability structure.
I feel the last sentence isn't clear enough. How about
a driver MUST negotiate VIRTIO_F_SR_IOV and complete the feature negotiation (including setting the DRIVER_OK \field{status} bit) before enabling virtual functions through the device's PCI SR-IOV capability structure.
Quoted 21 lines
content.tex +4 -0
+
 \devicenormative{\section}{Reserved Feature Bits}{Reserved Feature Bits}
 
 A device MUST offer VIRTIO_F_VERSION_1.  A device MAY fail to operate further
@@ -5392,6 +5400,9 @@ buffers in the same order in which they have been available.
 A device MAY fail to operate further if VIRTIO_F_IO_BARRIER
 is not accepted.
 
+A device SHOULD offer VIRTIO_F_SR_IOV if it presents a PCI
+SR-IOV capability structure.
+
 \section{Legacy Interface: Reserved Feature Bits}\label{sec:Reserved Feature Bits / Legacy Interface: Reserved Feature Bits}
 
 Transitional devices MAY offer the following:
Signature · 2 lines
-- 
2.17.0

Quoted 8 lines

Quoted 8 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]

Tiwei Bie

author

2018-05-24 15:26 UTC #8 lore raw

Quoted 6 lines · Michael S. Tsirkin

Quoted 5 lines · Tiwei Bie
Quoted 4 lines · Michael S. Tsirkin
Quoted 3 lines · Tiwei Bie
Quoted 2 lines · Michael S. Tsirkin
Quoted 1 line · Michael S. Tsirkin
On Tue, May 22, 2018 at 06:26:15PM +0800, Tiwei Bie wrote:

[...]

Quoted 36 lines

Quoted 24 lines
Quoted 3 lines
So my point is this, VFs themselves do not have this feature.
Yeah. I also think VFs shouldn't present this feature.
Quoted 5 lines
Should all of them have it? None of them? I don't see what use it is to VFs, but maybe we will come with a use down the road.
Quoted 8 lines
I propose we require that
1. drivers ignore this if there is
no SRIOV cap, and
2. that devices do not expose it.
This way if we come up with a use down the road, only new drivers will negotiate it.
I got your point now. Thanks!
How about:
Quoted 9 lines
If VIRTIO_F_SR_IOV has been negotiated, a driver can enable virtual functions through the device's PCI SR-IOV capability structure. A driver MUST NOT negotiate VIRTIO_F_SR_IOV if the device does not have a PCI SR-IOV capability structure or is not a PCI device. A driver MUST negotiate VIRTIO_F_SR_IOV and complete the feature negotiation (including setting the DRIVER_OK \field{status} bit) before enabling virtual functions through the device's PCI SR-IOV capability structure.
Sounds good.

[...]

Quoted 6 lines

Quoted 6 lines
Quoted 3 lines
Assuming we teach drivers they should ignore it
if it is there without SRIOV, then this last one I'd make MUST NOT.
Okay, how about

Quoted 8 lines

Quoted 6 lines
A device SHOULD offer VIRTIO_F_SR_IOV if it is a PCI device and presents a PCI SR-IOV capability structure, otherwise it MUST NOT offer VIRTIO_F_SR_IOV.
Best regards,
Tiwei Bie
Sounds good.

Thanks a lot! I'll send a new version.

Best regards,
Tiwei Bie

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]