On Sun, Mar 24, 2013 at 12:06:49PM +0100, Thomas Petazzoni wrote:
> Andrew, Thierry,
>
> On Fri, 22 Mar 2013 09:30:27 +0000, Andrew Murray wrote:
>
> > I think this could work well. In the future if the use of an independent MSI
> > controller is required, then new DT bindings for host-bridges could use
> > phandles to reference independent MSI controllers as their providers of
> > MSIs. I guess this functionality can be built on top of what you have proposed
> > later as the need arises.
>
> On Marvell HW (at least Armada 370/XP), MSIs are handled by the
> main interrupt controller directly, or more precisely, managing the
> MSIs requires fiddling with registers that are part of the interrupt
> controller registers, and not part of the PCIe controller registers.
>
> Basically, when a MSI interrupt is raised, it corresponds to IRQ 1 on
> the main interrupt controller. Then, one has to read a register of the
> main interrupt controller to find out which MSI interrupt was actually
> triggered. So in our case, the MSI irq_chip really belongs to the
> interrupt controller driver, and not the PCIe driver. Also, the
> physical address to be added in the 'struct msi_msg' is the physical
> address of an interrupt controller register.
>
> Therefore, I'm not sure how to do the interaction between the PCIe
> driver and the interrupt controller driver.
>
> Suggestions?
That sounds very much like one of the use-cases that were discussed. The
easiest solution would probably be to add an API to look up an MSI chip
from a DT phandle, so that the PCIe controller's device node could have
it as a property, somewhat like this:
msi: interrupt-controller {
};
pcie-controller {
...
marvell,msi = <&msi>;
...
};
Then add some basic infrastructure to register the MSI chip with a
global list, call that from the interrupt controller initialization:
...
msi_chip_add(&msi);
...
And finally look it up from the PCIe controller driver:
node = of_parse_phandle(dev->of_node, "marvell,msi", 0);
if (node)
msi = of_find_msi_chip_by_node(node);
That's roughly what other subsystems do. I wrote something similar once
for backlight devices, though the registration step (msi_chip_add)
wasn't necessary there since backlight devices all go into a common
struct class so class_find_device() can be used instead of going through
a separate registry.
Thierry
Attachment:
pgpLfiXuv6qs0.pgp
Description: PGP signature
