On 07/24/2015 12:13 AM, Marc Zyngier wrote:
Hi Suravee,
On 23/07/15 09:44, Suravee Suthikulpanit wrote:
On 7/17/15 22:55, Marc Zyngier wrote:
On 09/07/15 06:44, Suravee Suthikulpanit wrote:
Currently, in irq_find_host(), irq_domain_ops.match() uses struct device_node pointer to match a particular irq_domain. This will not be able to support ACPI where reference to GICv2m is in the MADT table (i.e the MSI frame).
This patch modifies irq_domain_ops.match(), to allow various types of irq_domain reference. It also introduces enum irq_domain_ref_types.
Along with the new reference type, it also introduces a new function, irq_find_domain(), which can be used to find a particular irqdomain using different reference types.
Signed-off-by: Suravee Suthikulpanit Suravee.Suthikulpanit@amd.com
Why has this patch been dropped from the v2? It has the best bit, and it is what I asked Hanjun to use.
But I see that none of the two series is using this approach, which is rather annoying.
Where are we going?
Thanks,
M.
In my V2, I didn't think this was needed with the new irq_domain_bus_token. I'll update this in the V3 and incorporate your changes from:
[PATCH v5 00/19] Per-device MSI domain & platform MSI [PATCH 0/5] Making the generic ACPI GSI layer irqdomain aware
domain_token and bus_token are orthogonal things. One is gives you a namespace, and the other allows different types for a given name.
Anyway, rebasing on top of these series should make it obvious. I've pushed out a branch (irq/gsi-irq-domain-v2) that contains the latest versions of these series.
Please synchronize with Hanjun to produce a *single* series that covers self-probing, GICv2m and GICv3 support.
Will do, thanks for your great help!
Hanjun