Skip to contentSkip to conversation

KVM call agenda for 2013-06-25

DISCUSSION Juan Quintela 4 participants 4 messages Jun 11, 2013 – Jun 25, 2013 KVM

lore
Share & tools
lore
#0 lore raw

Hi

Now we have moved to one call each other week. Please, send any topic that you are interested in covering.

Thanks, Juan.

PD. If you want to attend and you don't have the call details,
contact me.

#1 lore raw

I don't think my presence on the call is necessary, but I would appreciate it you put RDMA on the agenda.

The patches have been thoroughly bug-tested and reviewed.

- Michael

Quoted 10 lines · Juan Quintela

Hi

Now we have moved to one call each other week. Please, send any topic that you are interested in covering.

Thanks, Juan.

PD. If you want to attend and you don't have the call details,
contact me.

#2 lore raw
In reply to Juan Quintela
Quoted 5 lines · Juan Quintela

Hi

Now we have moved to one call each other week. Please, send any topic that you are interested in covering.

VFIO for device tree based platforms

Alex

Quoted 6 lines

Thanks, Juan.

PD. If you want to attend and you don't have the call details,
contact me.

#3 lore raw
In reply to Juan Quintela
Quoted 11 lines · On Tue, 11 Jun 2013 17:52:53 +0200 Juan Quintela <[email protected]> wrote:

Hi

Now we have moved to one call each other week. Please, send any topic that you are interested in covering.

Thanks, Juan.

PD. If you want to attend and you don't have the call details,
contact me.

Using static vs. dynamic properties vs. globals in particular case for CPU feature properties. Anthony suggested on IRC to use static properties for it but recently it was questioned why not used dynamic properties for it (afaerber):

1. static properties:

* using default values in static properties to define CPU models at class_init() time

* static properties could eventually evolve onto class properties, probably without much effort required.

* allows to simplify x86_cpu_initfn() and replace several custom property setters with a generic static property setters as result reducing code base and duplication.

2. introduce post_initfn() hook and move setting defaults of static properties and globals into it. So that property setters won't have to operate on partially initialized object instance. It also would allow to use dynamic properties in globals and compat_props.

3. using globals for simplifying cpu_model parsing and possibly getting rid of it in favor of -device FOO_CPU and eventually replacing cpu_init(cpu_model) with all its complexity by simple generic device_add() sequence.

as one of the steps to it, cpu_model "-cpu FOO_CPU,feat1=x,feat2=y" which is the template for N CPUs created in machine_init() and which is parsed by target specific parser in cpu_*_init(), could be internally transformed into a set of global properties, like:

      FOO_CPU.feat1=x FOO_CPU.feat2=y

then target specific parsers could be modified to CPU_CLASS hook that would keep compatibility code and make this remapping instead of parsing cpu_model string for each CPU and operating directly on CPU object instance. That would allow to treat CPUs as usual devices during hot-plug and use device-add. And open possibility to create individual CPUs on QEMU CLI using -device (which is important for migration and arbitrary CPU hot-add/remove) and heterogeneous CPUs machines.