Revert "sh: Kill off now redundant local irq disabling."

This reverts commit 64a6d72213dd810dd55bd0a503c36150af41c3c3.

Unfortunately we can't use on_each_cpu() for all of the cache ops, as
some of them only require preempt disabling. This seems to be the same
issue that impacts the mips r4k caches, where this code was based on.
This fixes up a deadlock that showed up in some IRQ context cases.

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

4 files changed, 71 insertions, 33 deletions

diff --git a/arch/sh/mm/cache-sh2a.c b/arch/sh/mm/cache-sh2a.c
index d783361e3f0..975899d8356 100644
--- a/arch/sh/mm/cache-sh2a.c
+++ b/arch/sh/mm/cache-sh2a.c
@@ -102,10 +102,12 @@ static void sh2a_flush_icache_range(void *args)
 	struct flusher_data *data = args;
 	unsigned long start, end;
 	unsigned long v;
+	unsigned long flags;
 
 	start = data->addr1 & ~(L1_CACHE_BYTES-1);
 	end = (data->addr2 + L1_CACHE_BYTES-1) & ~(L1_CACHE_BYTES-1);
 
+	local_irq_save(flags);
 	jump_to_uncached();
 
 	for (v = start; v < end; v+=L1_CACHE_BYTES) {
@@ -120,10 +122,12 @@ static void sh2a_flush_icache_range(void *args)
 			}
 		}
 		/* I-Cache invalidate */
-		ctrl_outl(addr, CACHE_IC_ADDRESS_ARRAY | addr | 0x00000008);
+		ctrl_outl(addr,
+			  CACHE_IC_ADDRESS_ARRAY | addr | 0x00000008);
 	}
 
 	back_to_cached();
+	local_irq_restore(flags);
 }
 
 void __init sh2a_cache_init(void)
diff --git a/arch/sh/mm/cache-sh4.c b/arch/sh/mm/cache-sh4.c
index 70fb906419d..3ac4945cb49 100644
--- a/arch/sh/mm/cache-sh4.c
+++ b/arch/sh/mm/cache-sh4.c
@@ -48,44 +48,48 @@ static void sh4_flush_icache_range(void *args)
 	struct flusher_data *data = args;
 	int icacheaddr;
 	unsigned long start, end;
-	unsigned long v;
+	unsigned long flags, v;
 	int i;
 
 	start = data->addr1;
 	end = data->addr2;
 
-	/* If there are too many pages then just blow the caches */
-	if (((end - start) >> PAGE_SHIFT) >= MAX_ICACHE_PAGES) {
-		local_flush_cache_all(args);
-	} else {
-		/* selectively flush d-cache then invalidate the i-cache */
-		/* this is inefficient, so only use for small ranges */
-		start &= ~(L1_CACHE_BYTES-1);
-		end += L1_CACHE_BYTES-1;
-		end &= ~(L1_CACHE_BYTES-1);
-
-		jump_to_uncached();
-
-		for (v = start; v < end; v+=L1_CACHE_BYTES) {
-			__ocbwb(v);
-
-			icacheaddr = CACHE_IC_ADDRESS_ARRAY |
-				(v & cpu_data->icache.entry_mask);
-
-			for (i = 0; i < cpu_data->icache.ways;
-				i++, icacheaddr += cpu_data->icache.way_incr)
-				/* Clear i-cache line valid-bit */
-				ctrl_outl(0, icacheaddr);
-		}
+       /* If there are too many pages then just blow the caches */
+        if (((end - start) >> PAGE_SHIFT) >= MAX_ICACHE_PAGES) {
+                local_flush_cache_all(args);
+       } else {
+               /* selectively flush d-cache then invalidate the i-cache */
+               /* this is inefficient, so only use for small ranges */
+               start &= ~(L1_CACHE_BYTES-1);
+               end += L1_CACHE_BYTES-1;
+               end &= ~(L1_CACHE_BYTES-1);
+
+               local_irq_save(flags);
+               jump_to_uncached();
+
+               for (v = start; v < end; v+=L1_CACHE_BYTES) {
+                       asm volatile("ocbwb     %0"
+                                    : /* no output */
+                                    : "m" (__m(v)));
+
+                       icacheaddr = CACHE_IC_ADDRESS_ARRAY | (
+                                       v & cpu_data->icache.entry_mask);
+
+                       for (i = 0; i < cpu_data->icache.ways;
+                               i++, icacheaddr += cpu_data->icache.way_incr)
+                                       /* Clear i-cache line valid-bit */
+                                       ctrl_outl(0, icacheaddr);
+               }
 
 		back_to_cached();
+		local_irq_restore(flags);
 	}
 }
 
 static inline void flush_cache_4096(unsigned long start,
 				    unsigned long phys)
 {
-	unsigned long exec_offset = 0;
+	unsigned long flags, exec_offset = 0;
 
 	/*
 	 * All types of SH-4 require PC to be in P2 to operate on the I-cache.
@@ -95,8 +99,10 @@ static inline void flush_cache_4096(unsigned long start,
 	    (start < CACHE_OC_ADDRESS_ARRAY))
 		exec_offset = 0x20000000;
 
+	local_irq_save(flags);
 	__flush_cache_4096(start | SH_CACHE_ASSOC,
 			   P1SEGADDR(phys), exec_offset);
+	local_irq_restore(flags);
 }
 
 /*
@@ -130,8 +136,9 @@ static void sh4_flush_dcache_page(void *arg)
 /* TODO: Selective icache invalidation through IC address array.. */
 static void __uses_jump_to_uncached flush_icache_all(void)
 {
-	unsigned long ccr;
+	unsigned long flags, ccr;
 
+	local_irq_save(flags);
 	jump_to_uncached();
 
 	/* Flush I-cache */
@@ -143,7 +150,9 @@ static void __uses_jump_to_uncached flush_icache_all(void)
 	 * back_to_cached() will take care of the barrier for us, don't add
 	 * another one!
 	 */
+
 	back_to_cached();
+	local_irq_restore(flags);
 }
 
 static inline void flush_dcache_all(void)
diff --git a/arch/sh/mm/cache-sh5.c b/arch/sh/mm/cache-sh5.c
index 2f9dd6df00a..467ff8e260f 100644
--- a/arch/sh/mm/cache-sh5.c
+++ b/arch/sh/mm/cache-sh5.c
@@ -34,22 +34,28 @@ static inline void
 sh64_setup_dtlb_cache_slot(unsigned long eaddr, unsigned long asid,
 			   unsigned long paddr)
 {
+	local_irq_disable();
 	sh64_setup_tlb_slot(dtlb_cache_slot, eaddr, asid, paddr);
 }
 
 static inline void sh64_teardown_dtlb_cache_slot(void)
 {
 	sh64_teardown_tlb_slot(dtlb_cache_slot);
+	local_irq_enable();
 }
 
 static inline void sh64_icache_inv_all(void)
 {
 	unsigned long long addr, flag, data;
+	unsigned long flags;
 
 	addr = ICCR0;
 	flag = ICCR0_ICI;
 	data = 0;
 
+	/* Make this a critical section for safety (probably not strictly necessary.) */
+	local_irq_save(flags);
+
 	/* Without %1 it gets unexplicably wrong */
 	__asm__ __volatile__ (
 		"getcfg	%3, 0, %0\n\t"
@@ -58,6 +64,8 @@ static inline void sh64_icache_inv_all(void)
 		"synci"
 		: "=&r" (data)
 		: "0" (data), "r" (flag), "r" (addr));
+
+	local_irq_restore(flags);
 }
 
 static void sh64_icache_inv_kernel_range(unsigned long start, unsigned long end)
@@ -82,6 +90,7 @@ static void sh64_icache_inv_user_page(struct vm_area_struct *vma, unsigned long
 	   Also, eaddr is page-aligned. */
 	unsigned int cpu = smp_processor_id();
 	unsigned long long addr, end_addr;
+	unsigned long flags = 0;
 	unsigned long running_asid, vma_asid;
 	addr = eaddr;
 	end_addr = addr + PAGE_SIZE;
@@ -102,9 +111,10 @@ static void sh64_icache_inv_user_page(struct vm_area_struct *vma, unsigned long
 
 	running_asid = get_asid();
 	vma_asid = cpu_asid(cpu, vma->vm_mm);
-	if (running_asid != vma_asid)
+	if (running_asid != vma_asid) {
+		local_irq_save(flags);
 		switch_and_save_asid(vma_asid);
-
+	}
 	while (addr < end_addr) {
 		/* Worth unrolling a little */
 		__asm__ __volatile__("icbi %0,  0" : : "r" (addr));
@@ -113,9 +123,10 @@ static void sh64_icache_inv_user_page(struct vm_area_struct *vma, unsigned long
 		__asm__ __volatile__("icbi %0, 96" : : "r" (addr));
 		addr += 128;
 	}
-
-	if (running_asid != vma_asid)
+	if (running_asid != vma_asid) {
 		switch_and_save_asid(running_asid);
+		local_irq_restore(flags);
+	}
 }
 
 static void sh64_icache_inv_user_page_range(struct mm_struct *mm,
@@ -148,12 +159,16 @@ static void sh64_icache_inv_user_page_range(struct mm_struct *mm,
 		unsigned long eaddr;
 		unsigned long after_last_page_start;
 		unsigned long mm_asid, current_asid;
+		unsigned long flags = 0;
 
 		mm_asid = cpu_asid(smp_processor_id(), mm);
 		current_asid = get_asid();
 
-		if (mm_asid != current_asid)
+		if (mm_asid != current_asid) {
+			/* Switch ASID and run the invalidate loop under cli */
+			local_irq_save(flags);
 			switch_and_save_asid(mm_asid);
+		}
 
 		aligned_start = start & PAGE_MASK;
 		after_last_page_start = PAGE_SIZE + ((end - 1) & PAGE_MASK);
@@ -179,8 +194,10 @@ static void sh64_icache_inv_user_page_range(struct mm_struct *mm,
 			aligned_start = vma->vm_end; /* Skip to start of next region */
 		}
 
-		if (mm_asid != current_asid)
+		if (mm_asid != current_asid) {
 			switch_and_save_asid(current_asid);
+			local_irq_restore(flags);
+		}
 	}
 }
 
diff --git a/arch/sh/mm/cache-sh7705.c b/arch/sh/mm/cache-sh7705.c
index 9dc38660e3d..6293f57fa88 100644
--- a/arch/sh/mm/cache-sh7705.c
+++ b/arch/sh/mm/cache-sh7705.c
@@ -81,6 +81,7 @@ static void sh7705_flush_icache_range(void *args)
 static void __flush_dcache_page(unsigned long phys)
 {
 	unsigned long ways, waysize, addrstart;
+	unsigned long flags;
 
 	phys |= SH_CACHE_VALID;
 
@@ -97,6 +98,7 @@ static void __flush_dcache_page(unsigned long phys)
 	 * potential cache aliasing, therefore the optimisation is probably not
 	 * possible.
 	 */
+	local_irq_save(flags);
 	jump_to_uncached();
 
 	ways = current_cpu_data.dcache.ways;
@@ -124,6 +126,7 @@ static void __flush_dcache_page(unsigned long phys)
 	} while (--ways);
 
 	back_to_cached();
+	local_irq_restore(flags);
 }
 
 /*
@@ -142,9 +145,14 @@ static void sh7705_flush_dcache_page(void *page)
 
 static void sh7705_flush_cache_all(void *args)
 {
+	unsigned long flags;
+
+	local_irq_save(flags);
 	jump_to_uncached();
+
 	cache_wback_all();
 	back_to_cached();
+	local_irq_restore(flags);
 }
 
 /*