Re: [RFC] pci: using new interrupt API to enable MSI and MSI-X

[Joao Pinto <Joao.Pinto@synopsys.com>]

Hi Mark,

Às 3:57 PM de 5/5/2017, Marc Zyngier escreveu:
> Joao,
> 
> On 05/05/17 15:11, Joao Pinto wrote:
>> Hello,
>> I am currently adding the support for both MSI and MSI-x in pcie-designware and
>> I am now facing a dificulty.
>>
>> My test endpoint is a USB 3.1 MSI / MSI-X capable and I tested that with
>> the changes introduced by this patch we are able to enable MSI and MSI-X
>> in this endpoint (depends on the usage of the MSI_FLAG_PCI_MSIX flag).
>>
>> The problem I am facing now is that Intc for the USB 3.1 Endpoint is completely
>> bogus (524288) when it should be 1, and so I am not receiving any interrupts
>> from the endpoint.
> 
> It is not bogus at all. It is computed from the PCI requester ID in 
> pci_msi_domain_calc_hwirq. What you're seeing is the PCI/MDI domain 
> view of that interrupt, which is completely virtual.
> 
> The real thing happens in your own irqdomain, where the hwirq for IRQ46 
> is probably 1 (only you can know that). As for why it doesn't work, see 
> below:
> 

snip (...)

>  static struct irq_chip dw_msi_irq_chip = {
>  	.name = "PCI-MSI",
> -	.irq_enable = pci_msi_unmask_irq,
> -	.irq_disable = pci_msi_mask_irq,
> -	.irq_mask = pci_msi_mask_irq,
> -	.irq_unmask = pci_msi_unmask_irq,
> +	.irq_mask = dw_msi_mask_irq,
> +	.irq_unmask = dw_msi_unmask_irq,
>  };
> 
> I haven't dug any further, but this should be fixed first.
> 
> Thanks,
> 
> 	M.
> 

Thanks for the help! I will check it out!

Joao