Minimal collection for running on Xen/ARM64

[ehem]

For people who didn't figure out the end goal of all the pull requests, here is a proof of concept with the pieces gathered together. Large amounts of other cleanup was found during implementation, hopefully those get in too.

The presently tested and known working bootloader is the ArmVirtXen build of Tianocore/EDK2. Due to a change in Xen at present one patch is required. Hopefully a similar patch will be integrated soon, but presently that extra is needed.

Configure the VM with kernel = "XEN_EFI.fi" and a FreeBSD mini-memstick build as a disk. Then everything will boot as expected.

Likely a bunch of squashing is needed, this though is a bit tricky to ensure Julien Grall gets appropriate credit. Though the Xen implementation was heavily rewritten.

Issues:

• Hooking pic_init_secondary(). Possibly add pass for devices not in PIC hierarchy Intrng: adjusting pic_init_secondary for further use cases #1280. Possibly declare hypervisor PIC root. Mostly advice on which direction to use.
• Fix intr_isrc_deregister() to release the event. Presently fix INTRNG interrupt release #1281.
• Approach to handler addition/description. Presently split INTRNG main file into two files #1285 (first two commits).

[ehem]

Commit cecb5a1 was tested and confirmed to work.

[ehem]

Before being rebased, 54bf574 was tested and confirmed to work. The delta is simply removing the stray clock interrupt fix and pulling 2 of the intermediates.

Mostly there are 3 problems in need of fixes. First, since Xen interrupts are very dynamic, release needs to work #1281. Second, some sort of per-processor hook is needed; pic_init_secondary() is basically ideal, but the current implementation is problematic, #1280.

Lastly, INTRNG's interface is appropriate for many PICs. Problem is due to being an unusual bus, this is well suited to a lower-level interface, #1285. Trying to work with the existing interface would be quite expensive.

Then simply becomes an issue of figuring out how much squashing should be done (while @jgrall did the original implementation, this has rather a lot of changes to get it into shape).

[ehem]

So instead of #1280, #1363 also functions to provide access to the appropriate hooks. For sys/arm64/include/intr.h use:

+#ifdef XENHVM
+#define INTR_ROOT_XEN	2
+#define INTR_ROOT_NUM	3
+#else
+
 #define INTR_ROOT_NUM	2
+#endif

In sys/dev/xen/intrng/xen_arch_intr.c add:

+	i = intr_pic_claim_root_num(dev, ofw_bus_get_node(dev),
+	    (intr_irq_filter_t *)panic, NULL, INTR_ROOT_XEN);
+	if (i != 0)
+		return (i);

After the intr_pic_register() call. This is somewhat hack-ish as we're barely using INTRNG, but it does seem kind of appropriate as it does amount to a disconnected PIC.

[markjdb]

I just looked at intrng, not very much at Xen-specific bits yet.

[ehem]

I'm expecting the per-processor startup hook to change in light of other recent commits. Two issues with INTRNG remain though:

1. Since Xen events are rather more dynamic than hardware interrupts, released events need some sort of handling. The obvious approach is intr_isrc_deregister(), problem is this hadn't previously been tested and it leaks the struct intr_event.
2. Since they're a specialized type of interrupt, not all of INTRNG's services are needed. In particular the lower level interface of intr_add_handler() (rename isrc_add_handler()) and intr_describe() (break intr_describe_irq() apart) is rather superior.