s390/mm,kvm: flush gmap address space with IDTE

The __tlb_flush_mm() helper uses a global flush if the mm struct
has a gmap structure attached to it. Replace the global flush with
two individual flushes by means of the IDTE instruction if only a
single gmap is attached the the mm.
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

arch/s390/include/asm/mmu.h

@@ -12,6 +12,7 @@ typedef struct {
 	struct list_head pgtable_list;
 	spinlock_t gmap_lock;
 	struct list_head gmap_list;
+	unsigned long gmap_asce;
 	unsigned long asce;
 	unsigned long asce_limit;
 	unsigned long vdso_base;

arch/s390/include/asm/mmu_context.h

@@ -21,6 +21,7 @@ static inline int init_new_context(struct task_struct *tsk,
 	INIT_LIST_HEAD(&mm->context.gmap_list);
 	cpumask_clear(&mm->context.cpu_attach_mask);
 	atomic_set(&mm->context.flush_count, 0);
+	mm->context.gmap_asce = 0;
 	mm->context.flush_mm = 0;
 #ifdef CONFIG_PGSTE
 	mm->context.alloc_pgste = page_table_allocate_pgste;

arch/s390/include/asm/tlbflush.h

@@ -60,18 +60,25 @@ static inline void __tlb_flush_full(struct mm_struct *mm)
 	preempt_enable();
 }
 
-/*
- * Flush TLB entries for a specific ASCE on all CPUs. Should never be used
- * when more than one asce (e.g. gmap) ran on this mm.
- */
-static inline void __tlb_flush_asce(struct mm_struct *mm, unsigned long asce)
+static inline void __tlb_flush_mm(struct mm_struct *mm)
 {
+	unsigned long gmap_asce;
+
+	/*
+	 * If the machine has IDTE we prefer to do a per mm flush
+	 * on all cpus instead of doing a local flush if the mm
+	 * only ran on the local cpu.
+	 */
 	preempt_disable();
 	atomic_inc(&mm->context.flush_count);
-	if (MACHINE_HAS_IDTE)
-		__tlb_flush_idte(asce);
-	else
-		__tlb_flush_global();
+	gmap_asce = READ_ONCE(mm->context.gmap_asce);
+	if (MACHINE_HAS_IDTE && gmap_asce != -1UL) {
+		if (gmap_asce)
+			__tlb_flush_idte(gmap_asce);
+		__tlb_flush_idte(mm->context.asce);
+	} else {
+		__tlb_flush_full(mm);
+	}
 	/* Reset TLB flush mask */
 	cpumask_copy(mm_cpumask(mm), &mm->context.cpu_attach_mask);
 	atomic_dec(&mm->context.flush_count);
@@ -92,7 +99,7 @@ static inline void __tlb_flush_kernel(void)
 /*
  * Flush TLB entries for a specific ASCE on all CPUs.
  */
-static inline void __tlb_flush_asce(struct mm_struct *mm, unsigned long asce)
+static inline void __tlb_flush_mm(struct mm_struct *mm)
 {
 	__tlb_flush_local();
 }
@@ -103,19 +110,6 @@ static inline void __tlb_flush_kernel(void)
 }
 #endif
 
-static inline void __tlb_flush_mm(struct mm_struct * mm)
-{
-	/*
-	 * If the machine has IDTE we prefer to do a per mm flush
-	 * on all cpus instead of doing a local flush if the mm
-	 * only ran on the local cpu.
-	 */
-	if (MACHINE_HAS_IDTE && list_empty(&mm->context.gmap_list))
-		__tlb_flush_asce(mm, mm->context.asce);
-	else
-		__tlb_flush_full(mm);
-}
-
 static inline void __tlb_flush_mm_lazy(struct mm_struct * mm)
 {
 	if (mm->context.flush_mm) {

arch/s390/mm/gmap.c

@@ -94,6 +94,7 @@ static struct gmap *gmap_alloc(unsigned long limit)
 struct gmap *gmap_create(struct mm_struct *mm, unsigned long limit)
 {
 	struct gmap *gmap;
+	unsigned long gmap_asce;
 
 	gmap = gmap_alloc(limit);
 	if (!gmap)
@@ -101,6 +102,11 @@ struct gmap *gmap_create(struct mm_struct *mm, unsigned long limit)
 	gmap->mm = mm;
 	spin_lock(&mm->context.gmap_lock);
 	list_add_rcu(&gmap->list, &mm->context.gmap_list);
+	if (list_is_singular(&mm->context.gmap_list))
+		gmap_asce = gmap->asce;
+	else
+		gmap_asce = -1UL;
+	WRITE_ONCE(mm->context.gmap_asce, gmap_asce);
 	spin_unlock(&mm->context.gmap_lock);
 	return gmap;
 }
@@ -230,6 +236,7 @@ EXPORT_SYMBOL_GPL(gmap_put);
 void gmap_remove(struct gmap *gmap)
 {
 	struct gmap *sg, *next;
+	unsigned long gmap_asce;
 
 	/* Remove all shadow gmaps linked to this gmap */
 	if (!list_empty(&gmap->children)) {
@@ -243,6 +250,14 @@ void gmap_remove(struct gmap *gmap)
 	/* Remove gmap from the pre-mm list */
 	spin_lock(&gmap->mm->context.gmap_lock);
 	list_del_rcu(&gmap->list);
+	if (list_empty(&gmap->mm->context.gmap_list))
+		gmap_asce = 0;
+	else if (list_is_singular(&gmap->mm->context.gmap_list))
+		gmap_asce = list_first_entry(&gmap->mm->context.gmap_list,
+					     struct gmap, list)->asce;
+	else
+		gmap_asce = -1UL;
+	WRITE_ONCE(gmap->mm->context.gmap_asce, gmap_asce);
 	spin_unlock(&gmap->mm->context.gmap_lock);
 	synchronize_rcu();
 	/* Put reference */