Code freeze exception for Xen 4.9 ARM ITS emulation?

Code freeze exception for Xen 4.9 ARM ITS emulation?

[Andre Przywara]

Hello!

The development of the ARM ITS emulation support has taken more time
than anticipated and I won't be able to address and fix all outstanding
comments until the official code freeze date.

However the first part of the series is in a good shape and should be
ready to be merged in time.

I was wondering if I could be granted a brief code freeze exception for
the remaining part of the series?
I would address the outstanding issues until Thursday, April 13th and
would post (a) new version(s) on the ML by then.
Needless to say, that regardless of this, I will be promptly reacting
and fixing and potential regressions introduced by the series.

The idea would be to split the series and merge the first patches by the
code freeze.
I would then address the remaining comments and the second part of the
series, consisting of patches to new files, could be merged by next
week.
I tuned the common code to make sure we only enter new code paths
if the code is compiled in and the machine actually has the ITS
hardware, to minimize the risk of breaking existing setups.


Rationale for the exception:
------------------------------
- This feature is a technical preview, disabled by default and
requiring the EXPERT Kconfig option to be actually compiled in.
- There is a strong interest from various parties to have the ITS
emulation support in the Xen 4.9 release. Without the ITS a number of
ARM servers will not be able to use MSIs (even in Dom0), which would
prevent some systems from using PCI devices at all.
- Xen's architecture requires the ITS emulation code to be usable for
Dom0 already, so this is not about device passthrough to DomUs, but
about to get PCI devices running. At the moment some servers can't use
any network or block storage devices in Dom0, which would be fixed by
this series.
- The technical preview would allow various people to easily test this
code, helping to gain confidence in this and to find potential bugs.


Potential risks:
----------------
- The code touches existing GIC emulation code, which is used by all ARM
machines so far. However the ITS is an extension to the GICv3 interrupt
controller, so the new code will be mostly adding functions, and changes
existing code only in a very few places. Most (if not all) of those
points can be secured by a gating variable (d->arch.vgic.has_its), which
would never be set when the code is not configured in or the machine is
not using the ITS.

Please let me know if you need further information or I can be of more
assistance!

Thank you very much!

Cheers,
Andre.

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
https://lists.xen.org/xen-devel

Re: Code freeze exception for Xen 4.9 ARM ITS emulation?

[Stefano Stabellini]

On Fri, 7 Apr 2017, Andre Przywara wrote:
> Hello!
> 
> The development of the ARM ITS emulation support has taken more time
> than anticipated and I won't be able to address and fix all outstanding
> comments until the official code freeze date.
> 
> However the first part of the series is in a good shape and should be
> ready to be merged in time.

Patches 1-9 are committed.


> I was wondering if I could be granted a brief code freeze exception for
> the remaining part of the series?
> I would address the outstanding issues until Thursday, April 13th and
> would post (a) new version(s) on the ML by then.
> Needless to say, that regardless of this, I will be promptly reacting
> and fixing and potential regressions introduced by the series.

The remaining patches are not bad, but there are still few holes. I'll
have to see the new versions with those holes plugged before I can make
a decision on whether the series is appropriate for 4.9. The only thing
I can say today is that it looks difficult but possible.

Of course, even if I think it's acceptable, for the code to make it into
Xen 4.9 past the code freeze, Julien will have to assess the risks, make
up his mind and, as release manager, he will have the final say.


> The idea would be to split the series and merge the first patches by the
> code freeze.
> I would then address the remaining comments and the second part of the
> series, consisting of patches to new files, could be merged by next
> week.
> I tuned the common code to make sure we only enter new code paths
> if the code is compiled in and the machine actually has the ITS
> hardware, to minimize the risk of breaking existing setups.
> 
> 
> Rationale for the exception:
> ------------------------------
> - This feature is a technical preview, disabled by default and
> requiring the EXPERT Kconfig option to be actually compiled in.
> - There is a strong interest from various parties to have the ITS
> emulation support in the Xen 4.9 release. Without the ITS a number of
> ARM servers will not be able to use MSIs (even in Dom0), which would
> prevent some systems from using PCI devices at all.
> - Xen's architecture requires the ITS emulation code to be usable for
> Dom0 already, so this is not about device passthrough to DomUs, but
> about to get PCI devices running. At the moment some servers can't use
> any network or block storage devices in Dom0, which would be fixed by
> this series.
> - The technical preview would allow various people to easily test this
> code, helping to gain confidence in this and to find potential bugs.
> 
> 
> Potential risks:
> ----------------
> - The code touches existing GIC emulation code, which is used by all ARM
> machines so far. However the ITS is an extension to the GICv3 interrupt
> controller, so the new code will be mostly adding functions, and changes
> existing code only in a very few places. Most (if not all) of those
> points can be secured by a gating variable (d->arch.vgic.has_its), which
> would never be set when the code is not configured in or the machine is
> not using the ITS.
> 
> Please let me know if you need further information or I can be of more
> assistance!
> 
> Thank you very much!
> 
> Cheers,
> Andre.
> 

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
https://lists.xen.org/xen-devel

Re: Code freeze exception for Xen 4.9 ARM ITS emulation?

[Julien Grall]

On 07/04/17 18:34, Andre Przywara wrote:
> Hello!

Hi Andre,

> The development of the ARM ITS emulation support has taken more time
> than anticipated and I won't be able to address and fix all outstanding
> comments until the official code freeze date.
>
> However the first part of the series is in a good shape and should be
> ready to be merged in time.
>
> I was wondering if I could be granted a brief code freeze exception for
> the remaining part of the series?

Thank you for your contributions. I spent yesterday reviewing the latest 
version (v8) and discovered potential locking issues in the series.

My understanding is they are not easily solvable and would require some 
rework of the vGIC. So I am afraid to say that ITS will not get merged 
for Xen 4.9. But keep in mind you can continue to work on this series 
and get it applied once next development window opens.

Cheers,

-- 
Julien Grall

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
https://lists.xen.org/xen-devel

Re: Code freeze exception for Xen 4.9 ARM ITS emulation?

[André Przywara]

On 13/04/17 14:29, Julien Grall wrote:

Hi Julien,

>> The development of the ARM ITS emulation support has taken more time
>> than anticipated and I won't be able to address and fix all outstanding
>> comments until the official code freeze date.
>>
>> However the first part of the series is in a good shape and should be
>> ready to be merged in time.
>>
>> I was wondering if I could be granted a brief code freeze exception for
>> the remaining part of the series?
> 
> Thank you for your contributions. I spent yesterday reviewing the latest
> version (v8) and discovered potential locking issues in the series.

I see and honestly don't know an easy solution for this nasty problem.

> My understanding is they are not easily solvable and would require some
> rework of the vGIC. So I am afraid to say that ITS will not get merged
> for Xen 4.9. But keep in mind you can continue to work on this series
> and get it applied once next development window opens.

OK, thanks for your and Stefano's massive efforts in this matter anyway
and for your support and help.

I am looking forward to getting this code in the right shape in the next
weeks.

Cheers,
Andre.


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
https://lists.xen.org/xen-devel