Re: [PATCH v6 0/9] PCI: dwc: MSI-X feature

[Lucas Stach <l.stach@pengutronix.de>]

Am Dienstag, den 13.02.2018, 12:29 +0000 schrieb Lorenzo Pieralisi:
> On Mon, Feb 12, 2018 at 06:14:29PM +0000, Gustavo Pimentel wrote:
> > Hi Lorenzo,
> > 
> > On 12/02/2018 12:23, Lorenzo Pieralisi wrote:
> > > On Fri, Jan 26, 2018 at 04:35:35PM +0000, Gustavo Pimentel wrote:
> > > > Replaces the use of IRQ domain hierarchy by the IRQ chained
> > > > used by
> > > > pcie-designware and each SoC specific driver in order to allow
> > > > new and
> > > > complex features like MSI-X.
> > > > 
> > > > Adds Synopsys Root Complex support for MSI-X feature.
> > > 
> > > Hi Gustavo,
> > > 
> > > please CC me for future patch versions. I have a question for
> > > you,
> > > the series definitely looks like it is going in the right
> > > direction
> > > but I do not understand the cover letter.
> > > 
> > > - You are not replacing the IRQ domain hierarchy, you are
> > > actually
> > >   introducing it properly. It may be just nomenclature but I
> > > though
> > >   I would mention that.
> > > - I really do not understand why this series is advertised as
> > >   "implementing MSI-X". I want to understand why MSI-X it is not
> > >   currently supported and why this series would enable it, I
> > >   honestly do not understand the reasoning behind this, it looks
> > >   a bit fuzzy.
> > > 
> > > Please note I am just asking clarifications on the $SUBJECT for
> > > my own
> > > understanding, the series is definitely the right thing to do and
> > > I am
> > > happy to target v4.17 for it.
> > 
> > Marc Zyngier also pointed out like you that the description is
> > confusing and
> > does not correspond to what is done in the code and now I also see
> > it now.
> > 
> > Please fell free to review and point out something wrong, only in
> > this way can I
> > evolve. :)
> 
> Sure. Let's start by discussing this:
> 
> https://patchwork.kernel.org/patch/5708521/
> 
> If it is a HW issue your patches can't solve it. If it is a SW issue
> patch above (and current mainline code - commit 19c5392eb1c1) is
> wrong.
> 
> Before updating the code I want to understand what the problem is
> with
> MSI-X in current mainline.
> 
> To be as clear as I can: this series is doing the *right* thing, I
> want
> to understand what the current driver is doing with MSIs and why, so
> that we can help you update the code correctly.

It's partly a software issue and partly a wrong understanding of MSI-X
on my side at the time. The DW hardware doesn't support different MSI
target addresses for individual MSIs, but that's just a optional
feature of MSI-X and not required to implement them correctly. Enabling
of MSI-X on the DW hardware via a proper software implementation is
fine.

Regards,
Lucas