Handling MSI-X interrupts in the head stage for x86 platform

Handling MSI-X interrupts in the head stage for x86 platform

[Alexander Frolov]

Hi all!

I have a PCIe device in x86 motherboard that generate periodic MSI-X 
interrupts and I would like to handle them in the head stage, how it can 
be done?
Thank you! Appreciate any help!

Best,
    Alex

RE: Handling MSI-X interrupts in the head stage for x86 platform

[Meng, Fino]

>Sent: Monday, June 1, 2020 4:09 PM
>
>Hi all!
>
>I have a PCIe device in x86 motherboard that generate periodic MSI-X interrupts and I would like to handle them in the head
>stage, how it can be done?
>Thank you! Appreciate any help!
>
>Best,
>    Alex

Hi Alex,

Kernel/driver/serial/16550A.c is a pci device, and have less code than net/igb driver, 
Which specific device u want to enable? Does it have a vanilla kernel driver already?

BR / Fino (孟祥夫)
Intel – IOTG Developer Enabling

Re: Handling MSI-X interrupts in the head stage for x86 platform

[Alexander Frolov]

On 6/1/20 12:23 PM, Meng, Fino via Xenomai wrote:
>
>> Sent: Monday, June 1, 2020 4:09 PM
>>
>> Hi all!
>>
>> I have a PCIe device in x86 motherboard that generate periodic MSI-X interrupts and I would like to handle them in the head
>> stage, how it can be done?
>> Thank you! Appreciate any help!
>>
>> Best,
>>     Alex
> Hi Alex,
>
> Kernel/driver/serial/16550A.c is a pci device, and have less code than net/igb driver,
> Which specific device u want to enable? Does it have a vanilla kernel driver already?

Not sure that 16550A uses MSI-X interrupts. However, net/igb does. Is it 
only driver in Xenomai that supports MSI-X?


I am using FPGA based device for periodic interrupts which is totally 
custom. We have are own driver.

Nevertheless, I have tested its latency from root stage driver and it is 
appropriately good (not over than 60us). It seems that
msix irqs are not stalled anywhere. Is it possible?

Can I guarantee that there will be no condition when MSIX interrupt will 
be deferred by head stage on more than that? What latency
improvement I can expect from moving to RTDM based driver?

Thank you!

Re: Handling MSI-X interrupts in the head stage for x86 platform

[Jan Kiszka]

On 02.06.20 08:40, Alexander Frolov via Xenomai wrote:
> 
> 
> On 6/1/20 12:23 PM, Meng, Fino via Xenomai wrote:
>>
>>> Sent: Monday, June 1, 2020 4:09 PM
>>>
>>> Hi all!
>>>
>>> I have a PCIe device in x86 motherboard that generate periodic MSI-X
>>> interrupts and I would like to handle them in the head
>>> stage, how it can be done?
>>> Thank you! Appreciate any help!
>>>
>>> Best,
>>>     Alex
>> Hi Alex,
>>
>> Kernel/driver/serial/16550A.c is a pci device, and have less code than
>> net/igb driver,
>> Which specific device u want to enable? Does it have a vanilla kernel
>> driver already?
> 
> Not sure that 16550A uses MSI-X interrupts. However, net/igb does. Is it
> only driver in Xenomai that supports MSI-X?
> 
> 
> I am using FPGA based device for periodic interrupts which is totally
> custom. We have are own driver.
> 
> Nevertheless, I have tested its latency from root stage driver and it is
> appropriately good (not over than 60us). It seems that
> msix irqs are not stalled anywhere. Is it possible?
> 
> Can I guarantee that there will be no condition when MSIX interrupt will
> be deferred by head stage on more than that? What latency
> improvement I can expect from moving to RTDM based driver?
> 

Head stage defers according to priorities: There there is an interrupt
or a task for head stage pending, that will win. No guarantees beyond
what your own workload can guarantee. In addition, no guarantees
architecturally on what will happen in Linux (root stage) which could
defer your interrupt. If you want determinism, move it to head stage.

Jan

PS: MSI-X works as well as MSI or legacy INTx.

-- 
Siemens AG, Corporate Technology, CT RDA IOT SES-DE
Corporate Competence Center Embedded Linux