On Mon, Jan 7, 2019 at 11:05 AM Greg KH <gregkh@xxxxxxxxxxxxxxxxxxx> wrote:
>
> On Mon, Jan 07, 2019 at 10:55:28AM +0100, Gerhard Khüny wrote:
> > On Mon, Jan 7, 2019 at 10:05 AM Greg KH <gregkh@xxxxxxxxxxxxxxxxxxx> wrote:
> > >
> > > On Mon, Jan 07, 2019 at 09:57:56AM +0100, Gerhard Khüny wrote:
> > > > Hello,
> > > >
> > > > I have a question regarding the function
> > > > serial_pci_is_class_communication() in 8250_pci:
> > > >
> > > > static int serial_pci_is_class_communication(struct pci_dev *dev) {
> > > >
> > > > /*
> > > > * If it is not a communications device or the programming
> > > > * interface is greater than 6, give up.
> > > > */
> > > > if ((((dev->class >> 8) != PCI_CLASS_COMMUNICATION_SERIAL) &&
> > > > ((dev->class >> 8) != PCI_CLASS_COMMUNICATION_MULTISERIAL) &&
> > > > ((dev->class >> 8) != PCI_CLASS_COMMUNICATION_MODEM)) ||
> > > > (dev->class & 0xff) > 6)
> > > > return -ENODEV;
> > > >
> > > > return 0;
> > > >
> > > > }
> > > >
> > > > We have implemented a serial interface in a FPGA. The existing driver
> > > > 8250_pci is working perfect for us. Only the check using the function
> > > > serial_pci_is_class_communication does not work since the PCI class is
> > > > 0x05, subclass is 0x00 and programming interface is 0x00 for our
> > > > device.
> > >
> > > Then fix your PCI device identifier structure :)
> > >
> > The problem is that the FPGA holds additional functionality (e.g. .
> > RTC, Hex-switch,..). The device therefore is not a real serial device
> > but it has implemented a serial interface.
>
> Ok, then you need to add a pci id to the serial driver.
>
This is what I did.
> > > > My question is, if it is the intention to exclude PCI class memory
> > > > devices implementing a serial driver? The comment in the code could
> > > > say something different.
> > >
> > > If your device says it is a memory device and not a serial device, how
> > > could the kernel "know" it is really a serial device?
> > >
> > Because the init function returns no error.
> > Why not allow memory devices also?
>
> Because the generic pci memory device class is not a serial device.
>
> Again, please refer to the PCI specification for the details. You can
> not have a generic class device suddenly be attached to a driver that is
> not the same as that class device. For example, would you want your USB
> keyboard driver to suddenly attach to all USB modem devices just because
> one single vendor decided to implement keyboard functionality in a modem
> device but lie about what the class type was?
>
> Just add a new pci vendor/device id to the driver and you should be
> fine. Or fix your pci device structure to not claim it is something
> other than what it really is.
I totally agree that our device does not say what it really is. But the PCI
structure can't be changed. Since the plausibility check was introduced recently
I try to find an solution which is mainline as much as possible.
Thanks
Gerhard
