Re: [PATCH v2 5/6] drivers/hv/vmbus: Get the irq number from DeviceTree

[Roman Kisel <romank@linux.microsoft.com>]

On 5/17/2024 10:14 AM, Rob Herring wrote:
> On Tue, May 14, 2024 at 5:45 PM Roman Kisel <romank@linux.microsoft.com> wrote:
>>
>> The vmbus driver uses ACPI for interrupt assignment on
>> arm64 hence it won't function in the VTL mode where only
>> DeviceTree can be used.
>>
>> Update the vmbus driver to discover interrupt configuration
>> via DeviceTree.
>>
>> Signed-off-by: Roman Kisel <romank@linux.microsoft.com>
>> ---
>>   drivers/hv/vmbus_drv.c | 37 +++++++++++++++++++++++++++++++++++++
>>   1 file changed, 37 insertions(+)
>>
>> diff --git a/drivers/hv/vmbus_drv.c b/drivers/hv/vmbus_drv.c
>> index e25223cee3ab..52f01bd1c947 100644
>> --- a/drivers/hv/vmbus_drv.c
>> +++ b/drivers/hv/vmbus_drv.c
>> @@ -36,6 +36,7 @@
>>   #include <linux/syscore_ops.h>
>>   #include <linux/dma-map-ops.h>
>>   #include <linux/pci.h>
>> +#include <linux/of_irq.h>
> 
> If you are using this header in a driver, you are doing it wrong. We
> have common functions which work on both ACPI or DT, so use them if
> you have a need to support both.
> 
Understood, thank you! I'll look more for the examples. If you happen to 
have in mind the place where the idiomatic/more preferred approach is 
used, please let me know, would owe you a great debt of gratitude.


> Though my first question on a binding will be the same as on every
> 'hypervisor binding'.  Why can't you make your hypervisor interfaces
> discoverable? It's all s/w, not some h/w device which is fixed.
> 
I've taken a look at the related art. AWS's Firecracker, Intel's Cloud 
Hypervisor, Google's CrosVM, QEmu allow the guest use the 
well-established battle-tested generic approaches (ACPI, 
DeviceTree/OpenFirmware) of describing the virtual hardware and its 
resources rather than making the guest use their own specific 
interfaces. That holds true for the s/w devices like 
"vcpu-stall-detector" and VirtIO that do not have counterparts built as 
hardware, too.

Here, the guest needs to set up VMBus (the intra-partition communication 
transport) to be able to talk to the host partition. Receiving a message 
needs an interrupt service routine attached to the interrupt injected 
into the guest virtual processor, and DeviceTree lets provide the 
interrupt number. If a custom interface were used here, it'd look less 
expected due to others relying on ACPI and DT for configuring virtual 
devices and busses. A specialized interface would add more code (new 
code) instead of relying on the approach that is widely used.


> Rob

-- 
Thank you,
Roman