[SPARC64] mm: Do not flush TLB mm in tlb_finish_mmu()

It isn't needed any longer, as noted by Hugh Dickins.

We still need the flush routines, due to the one remaining
call site in hugetlb_prefault_arch_hook().  That can be
eliminated at some later point, however.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

arch/sparc64/kernel/smp.c

@@ -839,43 +839,29 @@ void smp_flush_tlb_all(void)
  *    questionable (in theory the big win for threads is the massive sharing of
  *    address space state across processors).
  */
+
+/* This currently is only used by the hugetlb arch pre-fault
+ * hook on UltraSPARC-III+ and later when changing the pagesize
+ * bits of the context register for an address space.
+ */
 void smp_flush_tlb_mm(struct mm_struct *mm)
 {
-        /*
-         * This code is called from two places, dup_mmap and exit_mmap. In the
-         * former case, we really need a flush. In the later case, the callers
-         * are single threaded exec_mmap (really need a flush), multithreaded
-         * exec_mmap case (do not need to flush, since the caller gets a new
-         * context via activate_mm), and all other callers of mmput() whence
-         * the flush can be optimized since the associated threads are dead and
-         * the mm is being torn down (__exit_mm and other mmput callers) or the
-         * owning thread is dissociating itself from the mm. The
-         * (atomic_read(&mm->mm_users) == 0) check ensures real work is done
-         * for single thread exec and dup_mmap cases. An alternate check might
-         * have been (current->mm != mm).
-         *                                              Kanoj Sarcar
-         */
-        if (atomic_read(&mm->mm_users) == 0)
-                return;
-
-	{
-		u32 ctx = CTX_HWBITS(mm->context);
-		int cpu = get_cpu();
+	u32 ctx = CTX_HWBITS(mm->context);
+	int cpu = get_cpu();
 
-		if (atomic_read(&mm->mm_users) == 1) {
-			mm->cpu_vm_mask = cpumask_of_cpu(cpu);
-			goto local_flush_and_out;
-		}
+	if (atomic_read(&mm->mm_users) == 1) {
+		mm->cpu_vm_mask = cpumask_of_cpu(cpu);
+		goto local_flush_and_out;
+	}
 
-		smp_cross_call_masked(&xcall_flush_tlb_mm,
-				      ctx, 0, 0,
-				      mm->cpu_vm_mask);
+	smp_cross_call_masked(&xcall_flush_tlb_mm,
+			      ctx, 0, 0,
+			      mm->cpu_vm_mask);
 
-	local_flush_and_out:
-		__flush_tlb_mm(ctx, SECONDARY_CONTEXT);
+local_flush_and_out:
+	__flush_tlb_mm(ctx, SECONDARY_CONTEXT);
 
-		put_cpu();
-	}
+	put_cpu();
 }
 
 void smp_flush_tlb_pending(struct mm_struct *mm, unsigned long nr, unsigned long *vaddrs)

include/asm-sparc64/tlb.h

@@ -78,11 +78,9 @@ static inline void tlb_finish_mmu(struct mmu_gather *mp, unsigned long start, un
 {
 	tlb_flush_mmu(mp);
 
-	if (mp->fullmm) {
-		if (CTX_VALID(mp->mm->context))
-			do_flush_tlb_mm(mp->mm);
+	if (mp->fullmm)
 		mp->fullmm = 0;
-	} else
+	else
 		flush_tlb_pending();
 
 	/* keep the page table cache within bounds */