sh: Kill off machvec IRQ hinting.

Everything is using sparseirq these days, so we have no need to
arbitrarily size nr_irqs ahead of time. The legacy IRQ pre-allocation
likewise has no meaning for us, so that's killed off too. We now depend
on nr_irqs expansion by the generic hardirq layer instead.

It's also worth noting that the majority of boards had completely bogus
values for their nr_irqs relative to their CPU and configurations, so
this ends up correcting behaviour for quite a few platforms.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>

arch/sh/include/asm/irq.h

@@ -5,12 +5,15 @@
 #include <asm/machvec.h>
 
 /*
- * A sane default based on a reasonable vector table size, platforms are
- * advised to cap this at the hard limit that they're interested in
- * through the machvec.
+ * Only legacy non-sparseirq platforms have to set a reasonably sane
+ * value here. sparseirq platforms allocate their irq_descs on the fly,
+ * so will expand automatically based on the number of registered IRQs.
  */
-#define NR_IRQS			512
-#define NR_IRQS_LEGACY		8	/* Legacy external IRQ0-7 */
+#ifdef CONFIG_SPARSE_IRQ
+# define NR_IRQS		8
+#else
+# define NR_IRQS		512
+#endif
 
 /*
  * This is a special IRQ number for indicating that no IRQ has been

arch/sh/include/asm/machvec.h

@@ -17,7 +17,6 @@
 struct sh_machine_vector {
 	void (*mv_setup)(char **cmdline_p);
 	const char *mv_name;
-	int mv_nr_irqs;
 
 	int (*mv_irq_demux)(int irq);
 	void (*mv_init_irq)(void);