Re: [Qemu-devel] [RFC PATCH 0/8] Add Generic PCI host device update

[alvise rigo]

On Wed, Nov 5, 2014 at 11:23 AM, Claudio Fontana
<address@hidden> wrote:
> Hi Alvise,

Hi Claudio,

>
> On 11.07.2014 09:21, Alvise Rigo wrote:
>> This patch series is based on the previous work [1] and [2] by Rob
>> Herring and it tries to enhance this work on these points:
>
> do your patches need to be applied on top of Rob's?

Yes, definitively.

>
> I ask because I cannot see the patch "hw/arm/virt: Add generic PCI host 
> device" and among these,
> as wll as the "hw/pci-host: add a generic PCI host".

It seems to me that it's common practice in this mailing list to not
include the patches upon which the work is based.
What I could do for the next version is to set up a repository with
all the dependences included.

>
> I think those two need modifications as well, as they hardcode addresses, and 
> also try to register
> the PCI bus as a PCIE bus: does it really provide a PCI-Express? (probably 
> harmless but still would benefit from review).

This patches should not be compatible with PCI-Express devices, since
for them the ECAM access mechanism is required.
The initial idea was to add a second bus for PCI-Express cards,
however there wouldn't have been devices to use/test it, so I put this
on hold.
In any case, I will consider it for the next iteration.

Regards,
alvise

>
>>
>> - Some of the hardcoded values have been moved to an header file.  This
>>   header file is also used to share some device structures with the
>>   mach-virt machine.
>
> Some additional hardcoded addresses have been also introduced with your 
> changes though.
> (I'll post comments to the the patches in the series momentarily).
>
>> - The interrupt-map dt node generation has been revisited; it is now
>>   done after the generic devices init so that it's possible to attach
>>   PCI devices by mean of the qdev infrastructure. This allows to have
>>   several devices in the PCI bus, with the current limitation of one
>>   interrupt per PCI slot.
>>
>> Probably the most objectionable part of these patches regards the way
>> some data and definitions have been shared between the machine and the
>> device code; a better solution is still under evaluation. Any advice on
>> this and on the rest of the work is highly appreciated.
>>
>> This work has been tested attaching several PCI devices to the mach-virt
>> platform.  The tested devices are: virtio-blk-pci, virtio-net-pci,
>> lsi53c895a and pci-ohci (all attached at the same time).
>> Even if the original work was not changed in its core functionalities, I
>> couldn't reproduce the malfunctioning of the LSI SCSI mentioned in [1].
>> After attaching several qcow2 images, formatting and filling them, I
>> didn't notice anything wrong. Am I missing something?
>>
>> Thank you,
>> alvise
>>
>> [1]
>> "[Qemu-devel] [RFC PATCH 1/2] hw/pci-host: add a generic PCI host"
>> http://lists.gnu.org/archive/html/qemu-devel/2014-06/msg03482.html
>> [2]
>> "[Qemu-devel] [RFC PATCH 2/2] hw/arm/virt: Add generic PCI host device"
>> http://lists.gnu.org/archive/html/qemu-devel/2014-06/msg03483.html
>>
>> Alvise Rigo (8):
>>   mach-virt: move GIC inside mach-virt structure
>>   mach-virt: improve PCI memory topology definition
>>   QEMUMachine: finalize_dt function
>>   generic_pci: create header file
>>   generic_pci: create own map irq function
>>   generic_pci: generate dt node after devices init
>>   generic_pci: realize device with machine data
>>   generic_pci: add interrupt map structures
>>
>>  hw/arm/virt.c                     | 110 +++++++++++++++---------
>>  hw/pci-host/generic-pci.c         | 173 
>> +++++++++++++++++++++++++++++---------
>>  include/hw/boards.h               |   4 +
>>  include/hw/pci-host/pci_generic.h |  66 +++++++++++++++
>>  vl.c                              |   5 ++
>>  5 files changed, 277 insertions(+), 81 deletions(-)
>>  create mode 100644 include/hw/pci-host/pci_generic.h
>>
>
>