acpica: Add a thread safe acpi_get_name

[andrew]

Summary

The current acpi_name uses a static buffer to store the node name. This
is unsafe when two threads call this as the second will trash the
value of the first.

Add a version where the caller passes in the buffer and length so they
are safe from other threads.

Sponsored by: Arm Ltd

[obiwac]

sys/dev/acpica/acpi.c

4330–4333

maybe update this comment to point to acpi_get_name() instead

4356–4358

I know our acpi code is currently a little inconsistent regarding this, but I do think i'd prefer if this returned -1 on failure and 0 on success

[markj]

sys/dev/acpica/acpi.c

4356–4358

Even better IMO would just be to handle this like the existing acpi_name() does, i.e., copy in some string like "(unknown)". What is a caller supposed to do if the operation fails?

[andrew]

sys/dev/acpica/acpi.c

4356–4358

I'm planning on using it to clean up handling of MSIs on a non-PCI buses. In this case we need to know the path name, but can handle there not being one as we can return EINVAL:

static int
nexus_acpi_pcib_alloc_msi(device_t dev, device_t child, int count, int maxcount,
    int *irqs)
{
       char name[64];
       intptr_t msi_parent;
       u_int devid;
       int error;

       if (!acpi_get_name(acpi_get_handle(child), name, sizeof(name)))
               return (ENODEV);

       error = acpi_iort_map_named_msi(name, 0, &msi_parent, &devid);
       if (error != 0)
               return (error);

       return (intr_alloc_msi(dev, child, msi_parent, count, maxcount, irqs));
}

It looks like most of the acpi_name callers pass the result to a printf-like function so having unknown is ok there.