[IA64] ia64: add ar.k0 usage note

Update comment about how ar.k0 is used.  Make the initialization the
same as in start_secondary() (no functional change, just make it look
more similar).

Signed-off-by: Bjorn Helgaas <bjorn.helgaas@hp.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

1 files changed, 16 insertions, 13 deletions

diff --git a/arch/ia64/kernel/setup.c b/arch/ia64/kernel/setup.c
index 1f5c26dbe705..e256b114bf4e 100644
--- a/arch/ia64/kernel/setup.c
+++ b/arch/ia64/kernel/setup.c
@@ -244,28 +244,31 @@ find_initrd (void)
 static void __init
 io_port_init (void)
 {
-        extern unsigned long ia64_iobase;
         unsigned long phys_iobase;
 
         /*
-         *  Set `iobase' to the appropriate address in region 6 (uncached access range).
+         * Set `iobase' based on the EFI memory map or, failing that, the
+         * value firmware left in ar.k0.
          *
-         *  The EFI memory map is the "preferred" location to get the I/O port space base,
-         *  rather the relying on AR.KR0. This should become more clear in future SAL
-         *  specs. We'll fall back to getting it out of AR.KR0 if no appropriate entry is
-         *  found in the memory map.
+         * Note that in ia32 mode, IN/OUT instructions use ar.k0 to compute
+         * the port's virtual address, so ia32_load_state() loads it with a
+         * user virtual address.  But in ia64 mode, glibc uses the
+         * *physical* address in ar.k0 to mmap the appropriate area from
+         * /dev/mem, and the inX()/outX() interfaces use MMIO.  In both
+         * cases, user-mode can only use the legacy 0-64K I/O port space.
+         *
+         * ar.k0 is not involved in kernel I/O port accesses, which can use
+         * any of the I/O port spaces and are done via MMIO using the
+         * virtual mmio_base from the appropriate io_space[].
          */
         phys_iobase = efi_get_iobase();
-        if (phys_iobase)
-                /* set AR.KR0 since this is all we use it for anyway */
-                ia64_set_kr(IA64_KR_IO_BASE, phys_iobase);
-        else {
+        if (!phys_iobase) {
                 phys_iobase = ia64_get_kr(IA64_KR_IO_BASE);
-                printk(KERN_INFO "No I/O port range found in EFI memory map, falling back "
-                       "to AR.KR0\n");
-                printk(KERN_INFO "I/O port base = 0x%lx\n", phys_iobase);
+                printk(KERN_INFO "No I/O port range found in EFI memory map, "
+                        "falling back to AR.KR0 (0x%lx)\n", phys_iobase);
         }
         ia64_iobase = (unsigned long) ioremap(phys_iobase, 0);
+        ia64_set_kr(IA64_KR_IO_BASE, __pa(ia64_iobase));
 
         /* setup legacy IO port space */
         io_space[0].mmio_base = ia64_iobase;