s390/mm: add guest ASCE TLB flush optimization

Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

arch/s390/include/asm/pgtable.h

@@ -953,9 +953,11 @@ static inline pte_t pte_mkhuge(pte_t pte)
 #define	IPTE_LOCAL	1
 
 #define IPTE_NODAT	0x400
+#define IPTE_GUEST_ASCE	0x800
 
 static inline void __ptep_ipte(unsigned long address, pte_t *ptep,
-			       unsigned long opt, int local)
+			       unsigned long opt, unsigned long asce,
+			       int local)
 {
 	unsigned long pto = (unsigned long) ptep;
 
@@ -969,6 +971,7 @@ static inline void __ptep_ipte(unsigned long address, pte_t *ptep,
 	}
 
 	/* Invalidate ptes with options + TLB flush of the ptes */
+	opt = opt | (asce & _ASCE_ORIGIN);
 	asm volatile(
 		"	.insn	rrf,0xb2210000,%[r1],%[r2],%[r3],%[m4]"
 		: [r2] "+a" (address), [r3] "+a" (opt)
@@ -1355,34 +1358,59 @@ static inline void __pmdp_csp(pmd_t *pmdp)
 
 #define IDTE_PTOA	0x0800
 #define IDTE_NODAT	0x1000
+#define IDTE_GUEST_ASCE	0x2000
 
 static inline void __pmdp_idte(unsigned long addr, pmd_t *pmdp,
-			       unsigned long opt, int local)
+			       unsigned long opt, unsigned long asce,
+			       int local)
 {
 	unsigned long sto;
 
 	sto = (unsigned long) pmdp - pmd_index(addr) * sizeof(pmd_t);
+	if (__builtin_constant_p(opt) && opt == 0) {
+		/* flush without guest asce */
 		asm volatile(
 			"	.insn	rrf,0xb98e0000,%[r1],%[r2],0,%[m4]"
 			: "+m" (*pmdp)
-		: [r1] "a" (sto), [r2] "a" ((addr & HPAGE_MASK) | opt),
+			: [r1] "a" (sto), [r2] "a" ((addr & HPAGE_MASK)),
 			  [m4] "i" (local)
 			: "cc" );
+	} else {
+		/* flush with guest asce */
+		asm volatile(
+			"	.insn	rrf,0xb98e0000,%[r1],%[r2],%[r3],%[m4]"
+			: "+m" (*pmdp)
+			: [r1] "a" (sto), [r2] "a" ((addr & HPAGE_MASK) | opt),
+			  [r3] "a" (asce), [m4] "i" (local)
+			: "cc" );
+	}
 }
 
 static inline void __pudp_idte(unsigned long addr, pud_t *pudp,
-			       unsigned long opt, int local)
+			       unsigned long opt, unsigned long asce,
+			       int local)
 {
 	unsigned long r3o;
 
 	r3o = (unsigned long) pudp - pud_index(addr) * sizeof(pud_t);
 	r3o |= _ASCE_TYPE_REGION3;
+	if (__builtin_constant_p(opt) && opt == 0) {
+		/* flush without guest asce */
 		asm volatile(
 			"	.insn	rrf,0xb98e0000,%[r1],%[r2],0,%[m4]"
 			: "+m" (*pudp)
-		: [r1] "a" (r3o), [r2] "a" ((addr & PUD_MASK) | opt),
+			: [r1] "a" (r3o), [r2] "a" ((addr & PUD_MASK)),
 			  [m4] "i" (local)
+			: "cc");
+	} else {
+		/* flush with guest asce */
+		asm volatile(
+			"	.insn	rrf,0xb98e0000,%[r1],%[r2],%[r3],%[m4]"
+			: "+m" (*pudp)
+			: [r1] "a" (r3o), [r2] "a" ((addr & PUD_MASK) | opt),
+			  [r3] "a" (asce), [m4] "i" (local)
 			: "cc" );
+	}
 }
 
 pmd_t pmdp_xchg_direct(struct mm_struct *, unsigned long, pmd_t *, pmd_t);

arch/s390/include/asm/tlbflush.h

@@ -23,6 +23,8 @@ static inline void __tlb_flush_idte(unsigned long asce)
 	unsigned long opt;
 
 	opt = IDTE_PTOA;
+	if (MACHINE_HAS_TLB_GUEST)
+		opt |= IDTE_GUEST_ASCE;
 	/* Global TLB flush for the mm */
 	asm volatile(
 		"	.insn	rrf,0xb98e0000,0,%0,%1,0"

arch/s390/mm/pageattr.c

@@ -328,7 +328,7 @@ static void ipte_range(pte_t *pte, unsigned long address, int nr)
 		return;
 	}
 	for (i = 0; i < nr; i++) {
-		__ptep_ipte(address, pte, 0, IPTE_GLOBAL);
+		__ptep_ipte(address, pte, 0, 0, IPTE_GLOBAL);
 		address += PAGE_SIZE;
 		pte++;
 	}

arch/s390/mm/pgtable.c

@@ -35,9 +35,13 @@ static inline void ptep_ipte_local(struct mm_struct *mm, unsigned long addr,
 		asce = READ_ONCE(mm->context.gmap_asce);
 		if (asce == 0UL)
 			opt |= IPTE_NODAT;
-		__ptep_ipte(addr, ptep, opt, IPTE_LOCAL);
+		if (asce != -1UL) {
+			asce = asce ? : mm->context.asce;
+			opt |= IPTE_GUEST_ASCE;
+		}
+		__ptep_ipte(addr, ptep, opt, asce, IPTE_LOCAL);
 	} else {
-		__ptep_ipte(addr, ptep, 0, IPTE_LOCAL);
+		__ptep_ipte(addr, ptep, 0, 0, IPTE_LOCAL);
 	}
 }
 
@@ -51,9 +55,13 @@ static inline void ptep_ipte_global(struct mm_struct *mm, unsigned long addr,
 		asce = READ_ONCE(mm->context.gmap_asce);
 		if (asce == 0UL)
 			opt |= IPTE_NODAT;
-		__ptep_ipte(addr, ptep, opt, IPTE_GLOBAL);
+		if (asce != -1UL) {
+			asce = asce ? : mm->context.asce;
+			opt |= IPTE_GUEST_ASCE;
+		}
+		__ptep_ipte(addr, ptep, opt, asce, IPTE_GLOBAL);
 	} else {
-		__ptep_ipte(addr, ptep, 0, IPTE_GLOBAL);
+		__ptep_ipte(addr, ptep, 0, 0, IPTE_GLOBAL);
 	}
 }
 
@@ -326,18 +334,20 @@ static inline void pmdp_idte_local(struct mm_struct *mm,
 				   unsigned long addr, pmd_t *pmdp)
 {
 	if (MACHINE_HAS_TLB_GUEST)
-		__pmdp_idte(addr, pmdp, IDTE_NODAT, IDTE_LOCAL);
+		__pmdp_idte(addr, pmdp, IDTE_NODAT | IDTE_GUEST_ASCE,
+			    mm->context.asce, IDTE_LOCAL);
 	else
-		__pmdp_idte(addr, pmdp, 0, IDTE_LOCAL);
+		__pmdp_idte(addr, pmdp, 0, 0, IDTE_LOCAL);
 }
 
 static inline void pmdp_idte_global(struct mm_struct *mm,
 				    unsigned long addr, pmd_t *pmdp)
 {
 	if (MACHINE_HAS_TLB_GUEST)
-		__pmdp_idte(addr, pmdp, IDTE_NODAT, IDTE_GLOBAL);
+		__pmdp_idte(addr, pmdp, IDTE_NODAT | IDTE_GUEST_ASCE,
+			    mm->context.asce, IDTE_GLOBAL);
 	else if (MACHINE_HAS_IDTE)
-		__pmdp_idte(addr, pmdp, 0, IDTE_GLOBAL);
+		__pmdp_idte(addr, pmdp, 0, 0, IDTE_GLOBAL);
 	else
 		__pmdp_csp(pmdp);
 }
@@ -410,18 +420,20 @@ static inline void pudp_idte_local(struct mm_struct *mm,
 				   unsigned long addr, pud_t *pudp)
 {
 	if (MACHINE_HAS_TLB_GUEST)
-		__pudp_idte(addr, pudp, IDTE_NODAT, IDTE_LOCAL);
+		__pudp_idte(addr, pudp, IDTE_NODAT | IDTE_GUEST_ASCE,
+			    mm->context.asce, IDTE_LOCAL);
 	else
-		__pudp_idte(addr, pudp, 0, IDTE_LOCAL);
+		__pudp_idte(addr, pudp, 0, 0, IDTE_LOCAL);
 }
 
 static inline void pudp_idte_global(struct mm_struct *mm,
 				    unsigned long addr, pud_t *pudp)
 {
 	if (MACHINE_HAS_TLB_GUEST)
-		__pudp_idte(addr, pudp, IDTE_NODAT, IDTE_GLOBAL);
+		__pudp_idte(addr, pudp, IDTE_NODAT | IDTE_GUEST_ASCE,
+			    mm->context.asce, IDTE_GLOBAL);
 	else if (MACHINE_HAS_IDTE)
-		__pudp_idte(addr, pudp, 0, IDTE_GLOBAL);
+		__pudp_idte(addr, pudp, 0, 0, IDTE_GLOBAL);
 	else
 		/*
 		 * Invalid bit position is the same for pmd and pud, so we can