Re: [Xen-devel] [TESTDAY] Test report

[Julien Grall <julien.grall@xxxxxxx>]

Hello Edgar,

On 14/05/16 20:04, Edgar E. Iglesias wrote:
> On Sat, May 14, 2016 at 07:15:55PM +0100, Julien Grall wrote:
> You can see that it is used in the interrupt-map properties.
> IIUC, the interrupt lines connected to the pcie_intc controller
> are simply going to be combined into the "intx" line IRQ 116 on
> going from the pcie bridge towards the gic.
>
> I don't think we need to do much than to ignore the node but what
> we present to dom0 must look the same..
I think you are right. I expect the platform to only have one main 
interrupt controller (i.e the GIC), the others would be connected to the 
main one.
Therefore, the interrupts will have no meaning for the GIC and could be 
ignored.
FWIW, we already have a such check in handle_device.

> > > Disabling the pcie node for zynqmp boards gets dom0 to boot (obviously
> > > without PCIe support).
> > >
> > > Does it make sense to try to fix this problem this late inte the
> > > release cycle? (I can have a closer look and propose a possible fix
> > > for discussion)
> > I would try to fix it in Xen 4.7 if the patch is simple. Otherwise we could
> > backport it after the release.
> OK, great.
> I'll have a closer look too then.
>
> I was trying this, but it may be too permissive:
>
> diff --git a/xen/common/device_tree.c b/xen/common/device_tree.c
> index 0ed86a7..68cb162 100644
> --- a/xen/common/device_tree.c
> +++ b/xen/common/device_tree.c
> @@ -1176,6 +1176,15 @@ int dt_for_each_irq_map(const struct dt_device_node *dev,
>           for ( i = 0; i < pintsize; i++ )
>               dt_raw_irq.specifier[i] = dt_read_number(imap + i, 1);
>
> +        if (dt_raw_irq.controller != dt_interrupt_controller) {
Coding style.

> +            /* If this is not the main interrupt controller, we assume
> +             * it's part of a bus-bridge and ignore remapping IRQs for it.
> +             * Xen only supports one interrupt controller at the moment.  */
This comment is not true for every case. The interrupt controller may 
not be part of the bus-bridge.
One platform with multiple interrupt controller, I expect to always see 
one main controller, the others would be connected to the main one. So, 
the interrupt will have no meaning for the GIC.
> +            imap += pintsize;
> +            imaplen -= pintsize;
I would add a dt_printk to mention that the IRQ has been skipped.

> +            continue;
> +        }
Please send a formal patch to the ML. What would be the drawback if this 
patch doesn't reach Xen 4.7? I.e Are PCI devices necessary to run the 
platform?
Regards,

--
Julien Grall

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel