Re: [v2] PCI: imx: make msi work without pcieportbus

[Lucas Stach]

Am Donnerstag, den 13.12.2018, 12:34 +0000 schrieb Lorenzo Pieralisi:
> On Thu, Dec 13, 2018 at 11:07:16AM +0100, Lucas Stach wrote:
> > Am Donnerstag, den 13.12.2018, 09:57 +0000 schrieb Richard Zhu:
> > > Hi Lucas:
> > > 
> > > > -----Original Message-----
> > > > > > From: Lucas Stach [mailto:l.stach@xxxxxxxxxxxxxx]
> > > > 
> > > > Sent: 2018å12æ13æ 17:19
> > > > > > > > To: Richard Zhu <hongxing.zhu@xxxxxxx>; bhelgaas@xxxxxxxxxx;
> > > > 
> > > > lorenzo.pieralisi@xxxxxxx; andrew.smirnov@xxxxxxxxx
> > > > > > > > Cc: linux-pci@xxxxxxxxxxxxxxx; linux-arm-kernel@xxxxxxxxxxxxxxxxxxx;
> > > > 
> > > > linux-kernel@xxxxxxxxxxxxxxx
> > > > Subject: Re: [v2] PCI: imx: make msi work without pcieportbus
> > > > 
> > > > Hi Richard,
> > > > 
> > > > Am Donnerstag, den 13.12.2018, 08:02 +0000 schrieb Richard Zhu:
> > > > > MSI_EN of iMX PCIe RC would be asserted when PCIEPORTBUS driver is
> > > > > selected.
> > > > > Thus, the MSI works fine on iMX PCIe before.
> > > > > Assert it unconditionally when MSI is supported.
> > > > > Otherwise, the MSI wouldn't be triggered although the EP is present
> > > > > and the MSIs are assigned.
> > > > 
> > > > Thanks for digging into this issue. This seems like the right way forward.
> > > > However, did you test this with devices using legacy IRQs?
> > > > I.e. booting with "nomsi" on the kernel command line to see if legacy IRQs still
> > > > work if this bit is set, or if we need to avoid setting this when the user
> > > > explicitly requests to disable MSIs?
> > > > 
> > > > Regards,
> > > > Lucas
> > > > 
> > > 
> > > [Richard Zhu] Thanks for your review.
> > > The Legacy INTx is broken.
> > > The MSI_EN bit shouldn't be asserted when the user explicitly requests to disable MSIs.
> > 
> > Okay, so this patch should be extended with a check for
> > pci_msi_enabled() to see if the user explicitly want legacy IRQs.
> > 
> > > BTW, regarding to Baruch's comments, it seems that all the (IS_ENABLED(CONFIG_PCI_MSI) check in
> > > Âthe dwc host drivers are not required anymore, since the depends on PCI_MSI_IRQ_DOMAIN, right?
> > 
> > That's correct. This is mostly a historical artifact from the time when
> > Âwe were able to build without MSI support. Those checks could be
> > cleaned up now that we depend on the MSI options.
> 
> The question is whether we should really depend on PCI_MSI_IRQ_DOMAIN
> option, should we ?

It certainly cuts down on the combinatorial space when testing those
changes. IMHO this is a good thing, so as long as there is no
compelling reason for making this dependency configurable I'm totally
fine with keeping it the way it is now.

Regards,
Lucas