ARC: [SMP] Enable icache coherency

icaches are not snooped hence not cohrent in SMP setups which means
kernel has to do cross core calls to ensure the same.

The leaf routine __ic_line_inv_vaddr() now does cross core calls.

__sync_icache_dcache() is affected due to this:

* local dcache line flushed ahead of remote icache inv requests
* can't disable interrupts anymore, since
      __ic_line_inv_vaddr()->on_each_cpu() can deadlock.

| WARNING: CPU: 0 PID: 1 at kernel/smp.c:374
| smp_call_function_many+0x25a/0x2c4()
|
|  init_kprobes+0x90/0xc8
|     register_kprobe+0x1d6/0x510
|	__sync_icache_dcache+0x28/0x80
|
|	    DISABLE IRQ
|
|	    __ic_line_inv_vaddr
|		on_each_cpu
|		     smp_call_function_many+0x25a/0x2c4   --> WARN
|			__ic_line_inv_vaddr_local
|	    __dc_line_op

* TODO: Needs to use mask of relevant CPUs to avoid broadcasting

Signed-off-by: Noam Camus <noamc@ezchip.com>
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
diff --git a/arch/arc/mm/cache_arc700.c b/arch/arc/mm/cache_arc700.c
index 1f676c4..353b202 100644
--- a/arch/arc/mm/cache_arc700.c
+++ b/arch/arc/mm/cache_arc700.c
@@ -389,7 +389,7 @@
 /***********************************************************
  * Machine specific helper for per line I-Cache invalidate.
  */
-static void __ic_line_inv_vaddr(unsigned long paddr, unsigned long vaddr,
+static void __ic_line_inv_vaddr_local(unsigned long paddr, unsigned long vaddr,
 				unsigned long sz)
 {
 	unsigned long flags;
@@ -405,6 +405,23 @@
 	read_aux_reg(ARC_REG_IC_CTRL);	/* blocks */
 }
 
+struct ic_line_inv_vaddr_ipi {
+	unsigned long paddr, vaddr;
+	int sz;
+};
+
+static void __ic_line_inv_vaddr_helper(void *info)
+{
+        struct ic_line_inv_vaddr_ipi *ic_inv = (struct ic_line_inv_vaddr_ipi*) info;
+        __ic_line_inv_vaddr_local(ic_inv->paddr, ic_inv->vaddr, ic_inv->sz);
+}
+
+static void __ic_line_inv_vaddr(unsigned long paddr, unsigned long vaddr,
+				unsigned long sz)
+{
+	struct ic_line_inv_vaddr_ipi ic_inv = { paddr, vaddr , sz};
+	on_each_cpu(__ic_line_inv_vaddr_helper, &ic_inv, 1);
+}
 #else
 
 #define __ic_entire_inv()
@@ -553,12 +570,8 @@
  */
 void __sync_icache_dcache(unsigned long paddr, unsigned long vaddr, int len)
 {
-	unsigned long flags;
-
-	local_irq_save(flags);
-	__ic_line_inv_vaddr(paddr, vaddr, len);
 	__dc_line_op(paddr, vaddr, len, OP_FLUSH_N_INV);
-	local_irq_restore(flags);
+	__ic_line_inv_vaddr(paddr, vaddr, len);
 }
 
 /* wrapper to compile time eliminate alignment checks in flush loop */