Re: [PATCH 2/2] serial: 8250: Add support for 8250/16550 as MFD function

From: Greg Kroah-Hartman
Date: Tue May 21 2019 - 10:45:56 EST


On Tue, May 21, 2019 at 04:31:52PM +0200, Esben Haabendal wrote:
> Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx> writes:
>
> > On Tue, May 21, 2019 at 01:50:25PM +0200, Esben Haabendal wrote:
> >> Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx> writes:
> >>
> >> > On Tue, May 21, 2019 at 01:11:08PM +0200, Esben Haabendal wrote:
> >> >> Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx> writes:
> >> >>
> >> >> >> I will try ad hold back with this thread until you get back to it.
> >> >> >
> >> >> > Ok, I have no idea what is going on here, sorry. This is a really long
> >> >> > and meandering thread, and I can't even find the original patches in my
> >> >> > queue.
> >> >> >
> >> >> > So can you resend things and we can start over? :)
> >> >>
> >> >> Will do.
> >> >>
> >> >> > But note, using a mfd for a uart seems VERY odd to me...
> >> >>
> >> >> Ok. In my case, I have a pcie card with an fpga which includes 5 uart
> >> >> ports, 3 ethernet interfaces and a number of custom IP blocks.
> >> >> I believe that an mfd driver for that pcie card in that case.
> >> >
> >> > I believe you need to fix that fpga to expose individual pci devices
> >> > such that you can properly bind the individual devices to the expected
> >> > drivers :)
> >>
> >> Well, that is really out-of-scope of what I am doing here.
> >
> > Not really, if you have control over the fpga firmware (and odds are you
> > do), just fix that and instantly your device works with all kernels, no
> > need to change anything.
> >
> > Why not do this?
>
> Because I do not have control over fpga firmware.

Who does? Why did they create it this way if it can not be accessed by
an operating system as-is? Has it passed the PCI tests? Do you have a
link to where you can get this crazy device?

> >> > Seriously, who makes such a broken fpga device that goes against the PCI
> >> > spec that way? Well, not so much as "goes against it", as "ignores all
> >> > of the proper ideas of the past 20 years for working with PCI devices".
> >>
> >> Might be. But that is the firmware I have to work with here, and I
> >> still hope we can find a good solution for implementing a driver without
> >> having to maintain out-of-tree patches.
> >
> > As this hardware will not work on any operating system as-is, why not
> > fix the firmware to keep from having to support a one-off device that no
> > one else would be crazy enough to create? :)
>
> Clearly, someone has been crazy enough. Hopefully, we can be smart
> enough to make Linux fit to it.

Sometimes you need to go tell the hardware/firmware people not to do
foolish things. You can not always fix their problems in software.
Please push back on this.

thanks,

greg k-h