Merge tag 'arc-5.15-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/vgupta/arc

Pull ARC updates from Vineet Gupta:
 "Finally a big pile of changes for ARC (atomics/mm). These are from our
  internal arc64 tree, preparing mainline for eventual arc64 support.
  I'm spreading them out to avoid tsunami of patches in one release.

   - MM rework:
       - Implement up to 4 paging levels
       - Enable STRICT_MM_TYPECHECK
       - switch pgtable_t back to 'struct page *'

   - Atomics rework / implement relaxed accessors

   - Retire legacy MMUv1,v2; ARC750 cores

   - A few other build errors, typos"

* tag 'arc-5.15-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/vgupta/arc: (33 commits)
  ARC: mm: vmalloc sync from kernel to user table to update PMD ...
  ARC: mm: support 4 levels of page tables
  ARC: mm: support 3 levels of page tables
  ARC: mm: switch to asm-generic/pgalloc.h
  ARC: mm: switch pgtable_t back to struct page *
  ARC: mm: hack to allow 2 level build with 4 level code
  ARC: mm: disintegrate pgtable.h into levels and flags
  ARC: mm: disintegrate mmu.h (arcv2 bits out)
  ARC: mm: move MMU specific bits out of entry code ...
  ARC: mm: move MMU specific bits out of ASID allocator
  ARC: mm: non-functional code movement/cleanup
  ARC: mm: pmd_populate* to use the canonical set_pmd (and drop pmd_set)
  ARC: ioremap: use more commonly used PAGE_KERNEL based uncached flag
  ARC: mm: Enable STRICT_MM_TYPECHECKS
  ARC: mm: Fixes to allow STRICT_MM_TYPECHECKS
  ARC: mm: move mmu/cache externs out to setup.h
  ARC: mm: remove tlb paranoid code
  ARC: mm: use SCRATCH_DATA0 register for caching pgdir in ARCv2 only
  ARC: retire MMUv1 and MMUv2 support
  ARC: retire ARC750 support
  ...

arch/arc/Kconfig

@@ -116,16 +116,9 @@ choice
 	default ARC_CPU_770 if ISA_ARCOMPACT
 	default ARC_CPU_HS if ISA_ARCV2
 
-if ISA_ARCOMPACT
-
-config ARC_CPU_750D
-	bool "ARC750D"
-	select ARC_CANT_LLSC
-	help
-	  Support for ARC750 core
-
 config ARC_CPU_770
 	bool "ARC770"
+	depends on ISA_ARCOMPACT
 	select ARC_HAS_SWAPE
 	help
 	  Support for ARC770 core introduced with Rel 4.10 (Summer 2011)
@@ -135,8 +128,6 @@ config ARC_CPU_770
 	  -Caches: New Prog Model, Region Flush
 	  -Insns: endian swap, load-locked/store-conditional, time-stamp-ctr
 
-endif #ISA_ARCOMPACT
-
 config ARC_CPU_HS
 	bool "ARC-HS"
 	depends on ISA_ARCV2
@@ -274,33 +265,17 @@ config ARC_DCCM_BASE
 
 choice
 	prompt "MMU Version"
-	default ARC_MMU_V3 if ARC_CPU_770
-	default ARC_MMU_V2 if ARC_CPU_750D
-	default ARC_MMU_V4 if ARC_CPU_HS
-
-if ISA_ARCOMPACT
-
-config ARC_MMU_V1
-	bool "MMU v1"
-	help
-	  Orig ARC700 MMU
-
-config ARC_MMU_V2
-	bool "MMU v2"
-	help
-	  Fixed the deficiency of v1 - possible thrashing in memcpy scenario
-	  when 2 D-TLB and 1 I-TLB entries index into same 2way set.
+	default ARC_MMU_V3 if ISA_ARCOMPACT
+	default ARC_MMU_V4 if ISA_ARCV2
 
 config ARC_MMU_V3
 	bool "MMU v3"
-	depends on ARC_CPU_770
+	depends on ISA_ARCOMPACT
 	help
 	  Introduced with ARC700 4.10: New Features
 	  Variable Page size (1k-16k), var JTLB size 128 x (2 or 4)
 	  Shared Address Spaces (SASID)
 
-endif
-
 config ARC_MMU_V4
 	bool "MMU v4"
 	depends on ISA_ARCV2
@@ -319,7 +294,6 @@ config ARC_PAGE_SIZE_8K
 
 config ARC_PAGE_SIZE_16K
 	bool "16KB"
-	depends on ARC_MMU_V3 || ARC_MMU_V4
 
 config ARC_PAGE_SIZE_4K
 	bool "4KB"
@@ -340,6 +314,10 @@ config ARC_HUGEPAGE_16M
 
 endchoice
 
+config PGTABLE_LEVELS
+	int "Number of Page table levels"
+	default 2
+
 config ARC_COMPACT_IRQ_LEVELS
 	depends on ISA_ARCOMPACT
 	bool "Setup Timer IRQ as high Priority"
@@ -563,9 +541,6 @@ config ARC_DW2_UNWIND
 	  If you don't debug the kernel, you can say N, but we may not be able
 	  to solve problems without frame unwind information
 
-config ARC_DBG_TLB_PARANOIA
-	bool "Paranoia Checks in Low Level TLB Handlers"
-
 config ARC_DBG_JUMP_LABEL
 	bool "Paranoid checks in Static Keys (jump labels) code"
 	depends on JUMP_LABEL

arch/arc/include/asm/atomic-llsc.h

+/* SPDX-License-Identifier: GPL-2.0-only */
+
+#ifndef _ASM_ARC_ATOMIC_LLSC_H
+#define _ASM_ARC_ATOMIC_LLSC_H
+
+#define arch_atomic_set(v, i) WRITE_ONCE(((v)->counter), (i))
+
+#define ATOMIC_OP(op, c_op, asm_op)					\
+static inline void arch_atomic_##op(int i, atomic_t *v)			\
+{									\
+	unsigned int val;						\
+									\
+	__asm__ __volatile__(						\
+	"1:	llock   %[val], [%[ctr]]		\n"		\
+	"	" #asm_op " %[val], %[val], %[i]	\n"		\
+	"	scond   %[val], [%[ctr]]		\n"		\
+	"	bnz     1b				\n"		\
+	: [val]	"=&r"	(val) /* Early clobber to prevent reg reuse */	\
+	: [ctr]	"r"	(&v->counter), /* Not "m": llock only supports reg direct addr mode */	\
+	  [i]	"ir"	(i)						\
+	: "cc");							\
+}									\
+
+#define ATOMIC_OP_RETURN(op, c_op, asm_op)				\
+static inline int arch_atomic_##op##_return_relaxed(int i, atomic_t *v)	\
+{									\
+	unsigned int val;						\
+									\
+	__asm__ __volatile__(						\
+	"1:	llock   %[val], [%[ctr]]		\n"		\
+	"	" #asm_op " %[val], %[val], %[i]	\n"		\
+	"	scond   %[val], [%[ctr]]		\n"		\
+	"	bnz     1b				\n"		\
+	: [val]	"=&r"	(val)						\
+	: [ctr]	"r"	(&v->counter),					\
+	  [i]	"ir"	(i)						\
+	: "cc");							\
+									\
+	return val;							\
+}
+
+#define arch_atomic_add_return_relaxed		arch_atomic_add_return_relaxed
+#define arch_atomic_sub_return_relaxed		arch_atomic_sub_return_relaxed
+
+#define ATOMIC_FETCH_OP(op, c_op, asm_op)				\
+static inline int arch_atomic_fetch_##op##_relaxed(int i, atomic_t *v)	\
+{									\
+	unsigned int val, orig;						\
+									\
+	__asm__ __volatile__(						\
+	"1:	llock   %[orig], [%[ctr]]		\n"		\
+	"	" #asm_op " %[val], %[orig], %[i]	\n"		\
+	"	scond   %[val], [%[ctr]]		\n"		\
+	"	bnz     1b				\n"		\
+	: [val]	"=&r"	(val),						\
+	  [orig] "=&r" (orig)						\
+	: [ctr]	"r"	(&v->counter),					\
+	  [i]	"ir"	(i)						\
+	: "cc");							\
+									\
+	return orig;							\
+}
+
+#define arch_atomic_fetch_add_relaxed		arch_atomic_fetch_add_relaxed
+#define arch_atomic_fetch_sub_relaxed		arch_atomic_fetch_sub_relaxed
+
+#define arch_atomic_fetch_and_relaxed		arch_atomic_fetch_and_relaxed
+#define arch_atomic_fetch_andnot_relaxed	arch_atomic_fetch_andnot_relaxed
+#define arch_atomic_fetch_or_relaxed		arch_atomic_fetch_or_relaxed
+#define arch_atomic_fetch_xor_relaxed		arch_atomic_fetch_xor_relaxed
+
+#define ATOMIC_OPS(op, c_op, asm_op)					\
+	ATOMIC_OP(op, c_op, asm_op)					\
+	ATOMIC_OP_RETURN(op, c_op, asm_op)				\
+	ATOMIC_FETCH_OP(op, c_op, asm_op)
+
+ATOMIC_OPS(add, +=, add)
+ATOMIC_OPS(sub, -=, sub)
+
+#undef ATOMIC_OPS
+#define ATOMIC_OPS(op, c_op, asm_op)					\
+	ATOMIC_OP(op, c_op, asm_op)					\
+	ATOMIC_FETCH_OP(op, c_op, asm_op)
+
+ATOMIC_OPS(and, &=, and)
+ATOMIC_OPS(andnot, &= ~, bic)
+ATOMIC_OPS(or, |=, or)
+ATOMIC_OPS(xor, ^=, xor)
+
+#define arch_atomic_andnot		arch_atomic_andnot
+
+#undef ATOMIC_OPS
+#undef ATOMIC_FETCH_OP
+#undef ATOMIC_OP_RETURN
+#undef ATOMIC_OP
+
+#endif

arch/arc/include/asm/atomic-spinlock.h

+/* SPDX-License-Identifier: GPL-2.0-only */
+
+#ifndef _ASM_ARC_ATOMIC_SPLOCK_H
+#define _ASM_ARC_ATOMIC_SPLOCK_H
+
+/*
+ * Non hardware assisted Atomic-R-M-W
+ * Locking would change to irq-disabling only (UP) and spinlocks (SMP)
+ */
+
+static inline void arch_atomic_set(atomic_t *v, int i)
+{
+	/*
+	 * Independent of hardware support, all of the atomic_xxx() APIs need
+	 * to follow the same locking rules to make sure that a "hardware"
+	 * atomic insn (e.g. LD) doesn't clobber an "emulated" atomic insn
+	 * sequence
+	 *
+	 * Thus atomic_set() despite being 1 insn (and seemingly atomic)
+	 * requires the locking.
+	 */
+	unsigned long flags;
+
+	atomic_ops_lock(flags);
+	WRITE_ONCE(v->counter, i);
+	atomic_ops_unlock(flags);
+}
+
+#define arch_atomic_set_release(v, i)	arch_atomic_set((v), (i))
+
+#define ATOMIC_OP(op, c_op, asm_op)					\
+static inline void arch_atomic_##op(int i, atomic_t *v)			\
+{									\
+	unsigned long flags;						\
+									\
+	atomic_ops_lock(flags);						\
+	v->counter c_op i;						\
+	atomic_ops_unlock(flags);					\
+}
+
+#define ATOMIC_OP_RETURN(op, c_op, asm_op)				\
+static inline int arch_atomic_##op##_return(int i, atomic_t *v)		\
+{									\
+	unsigned long flags;						\
+	unsigned int temp;						\
+									\
+	/*								\
+	 * spin lock/unlock provides the needed smp_mb() before/after	\
+	 */								\
+	atomic_ops_lock(flags);						\
+	temp = v->counter;						\
+	temp c_op i;							\
+	v->counter = temp;						\
+	atomic_ops_unlock(flags);					\
+									\
+	return temp;							\
+}
+
+#define ATOMIC_FETCH_OP(op, c_op, asm_op)				\
+static inline int arch_atomic_fetch_##op(int i, atomic_t *v)		\
+{									\
+	unsigned long flags;						\
+	unsigned int orig;						\
+									\
+	/*								\
+	 * spin lock/unlock provides the needed smp_mb() before/after	\
+	 */								\
+	atomic_ops_lock(flags);						\
+	orig = v->counter;						\
+	v->counter c_op i;						\
+	atomic_ops_unlock(flags);					\
+									\
+	return orig;							\
+}
+
+#define ATOMIC_OPS(op, c_op, asm_op)					\
+	ATOMIC_OP(op, c_op, asm_op)					\
+	ATOMIC_OP_RETURN(op, c_op, asm_op)				\
+	ATOMIC_FETCH_OP(op, c_op, asm_op)
+
+ATOMIC_OPS(add, +=, add)
+ATOMIC_OPS(sub, -=, sub)
+
+#undef ATOMIC_OPS
+#define ATOMIC_OPS(op, c_op, asm_op)					\
+	ATOMIC_OP(op, c_op, asm_op)					\
+	ATOMIC_FETCH_OP(op, c_op, asm_op)
+
+ATOMIC_OPS(and, &=, and)
+ATOMIC_OPS(andnot, &= ~, bic)
+ATOMIC_OPS(or, |=, or)
+ATOMIC_OPS(xor, ^=, xor)
+
+#define arch_atomic_andnot		arch_atomic_andnot
+#define arch_atomic_fetch_andnot	arch_atomic_fetch_andnot
+
+#undef ATOMIC_OPS
+#undef ATOMIC_FETCH_OP
+#undef ATOMIC_OP_RETURN
+#undef ATOMIC_OP
+
+#endif

arch/arc/include/asm/atomic64-arcv2.h

+/* SPDX-License-Identifier: GPL-2.0-only */
+
+/*
+ * ARCv2 supports 64-bit exclusive load (LLOCKD) / store (SCONDD)
+ *  - The address HAS to be 64-bit aligned
+ */
+
+#ifndef _ASM_ARC_ATOMIC64_ARCV2_H
+#define _ASM_ARC_ATOMIC64_ARCV2_H
+
+typedef struct {
+	s64 __aligned(8) counter;
+} atomic64_t;
+
+#define ATOMIC64_INIT(a) { (a) }
+
+static inline s64 arch_atomic64_read(const atomic64_t *v)
+{
+	s64 val;
+
+	__asm__ __volatile__(
+	"	ldd   %0, [%1]	\n"
+	: "=r"(val)
+	: "r"(&v->counter));
+
+	return val;
+}
+
+static inline void arch_atomic64_set(atomic64_t *v, s64 a)
+{
+	/*
+	 * This could have been a simple assignment in "C" but would need
+	 * explicit volatile. Otherwise gcc optimizers could elide the store
+	 * which borked atomic64 self-test
+	 * In the inline asm version, memory clobber needed for exact same
+	 * reason, to tell gcc about the store.
+	 *
+	 * This however is not needed for sibling atomic64_add() etc since both
+	 * load/store are explicitly done in inline asm. As long as API is used
+	 * for each access, gcc has no way to optimize away any load/store
+	 */
+	__asm__ __volatile__(
+	"	std   %0, [%1]	\n"
+	:
+	: "r"(a), "r"(&v->counter)
+	: "memory");
+}
+
+#define ATOMIC64_OP(op, op1, op2)					\
+static inline void arch_atomic64_##op(s64 a, atomic64_t *v)		\
+{									\
+	s64 val;							\
+									\
+	__asm__ __volatile__(						\
+	"1:				\n"				\
+	"	llockd  %0, [%1]	\n"				\
+	"	" #op1 " %L0, %L0, %L2	\n"				\
+	"	" #op2 " %H0, %H0, %H2	\n"				\
+	"	scondd   %0, [%1]	\n"				\
+	"	bnz     1b		\n"				\
+	: "=&r"(val)							\
+	: "r"(&v->counter), "ir"(a)					\
+	: "cc");							\
+}									\
+
+#define ATOMIC64_OP_RETURN(op, op1, op2)		        	\
+static inline s64 arch_atomic64_##op##_return_relaxed(s64 a, atomic64_t *v)	\
+{									\
+	s64 val;							\
+									\
+	__asm__ __volatile__(						\
+	"1:				\n"				\
+	"	llockd   %0, [%1]	\n"				\
+	"	" #op1 " %L0, %L0, %L2	\n"				\
+	"	" #op2 " %H0, %H0, %H2	\n"				\
+	"	scondd   %0, [%1]	\n"				\
+	"	bnz     1b		\n"				\
+	: [val] "=&r"(val)						\
+	: "r"(&v->counter), "ir"(a)					\
+	: "cc");	/* memory clobber comes from smp_mb() */	\
+									\
+	return val;							\
+}
+
+#define arch_atomic64_add_return_relaxed	arch_atomic64_add_return_relaxed
+#define arch_atomic64_sub_return_relaxed	arch_atomic64_sub_return_relaxed
+
+#define ATOMIC64_FETCH_OP(op, op1, op2)		        		\
+static inline s64 arch_atomic64_fetch_##op##_relaxed(s64 a, atomic64_t *v)	\
+{									\
+	s64 val, orig;							\
+									\
+	__asm__ __volatile__(						\
+	"1:				\n"				\
+	"	llockd   %0, [%2]	\n"				\
+	"	" #op1 " %L1, %L0, %L3	\n"				\
+	"	" #op2 " %H1, %H0, %H3	\n"				\
+	"	scondd   %1, [%2]	\n"				\
+	"	bnz     1b		\n"				\
+	: "=&r"(orig), "=&r"(val)					\
+	: "r"(&v->counter), "ir"(a)					\
+	: "cc");	/* memory clobber comes from smp_mb() */	\
+									\
+	return orig;							\
+}
+
+#define arch_atomic64_fetch_add_relaxed		arch_atomic64_fetch_add_relaxed
+#define arch_atomic64_fetch_sub_relaxed		arch_atomic64_fetch_sub_relaxed
+
+#define arch_atomic64_fetch_and_relaxed		arch_atomic64_fetch_and_relaxed
+#define arch_atomic64_fetch_andnot_relaxed	arch_atomic64_fetch_andnot_relaxed
+#define arch_atomic64_fetch_or_relaxed		arch_atomic64_fetch_or_relaxed
+#define arch_atomic64_fetch_xor_relaxed		arch_atomic64_fetch_xor_relaxed
+
+#define ATOMIC64_OPS(op, op1, op2)					\
+	ATOMIC64_OP(op, op1, op2)					\
+	ATOMIC64_OP_RETURN(op, op1, op2)				\
+	ATOMIC64_FETCH_OP(op, op1, op2)
+
+ATOMIC64_OPS(add, add.f, adc)
+ATOMIC64_OPS(sub, sub.f, sbc)
+
+#undef ATOMIC64_OPS
+#define ATOMIC64_OPS(op, op1, op2)					\
+	ATOMIC64_OP(op, op1, op2)					\
+	ATOMIC64_FETCH_OP(op, op1, op2)
+
+ATOMIC64_OPS(and, and, and)
+ATOMIC64_OPS(andnot, bic, bic)
+ATOMIC64_OPS(or, or, or)
+ATOMIC64_OPS(xor, xor, xor)
+
+#define arch_atomic64_andnot		arch_atomic64_andnot
+
+#undef ATOMIC64_OPS
+#undef ATOMIC64_FETCH_OP
+#undef ATOMIC64_OP_RETURN
+#undef ATOMIC64_OP
+
+static inline s64
+arch_atomic64_cmpxchg(atomic64_t *ptr, s64 expected, s64 new)
+{
+	s64 prev;
+
+	smp_mb();
+
+	__asm__ __volatile__(
+	"1:	llockd  %0, [%1]	\n"
+	"	brne    %L0, %L2, 2f	\n"
+	"	brne    %H0, %H2, 2f	\n"
+	"	scondd  %3, [%1]	\n"
+	"	bnz     1b		\n"
+	"2:				\n"
+	: "=&r"(prev)
+	: "r"(ptr), "ir"(expected), "r"(new)
+	: "cc");	/* memory clobber comes from smp_mb() */
+
+	smp_mb();
+
+	return prev;
+}
+
+static inline s64 arch_atomic64_xchg(atomic64_t *ptr, s64 new)
+{
+	s64 prev;
+
+	smp_mb();
+
+	__asm__ __volatile__(
+	"1:	llockd  %0, [%1]	\n"
+	"	scondd  %2, [%1]	\n"
+	"	bnz     1b		\n"
+	"2:				\n"
+	: "=&r"(prev)
+	: "r"(ptr), "r"(new)
+	: "cc");	/* memory clobber comes from smp_mb() */
+
+	smp_mb();
+
+	return prev;
+}
+
+/**
+ * arch_atomic64_dec_if_positive - decrement by 1 if old value positive
+ * @v: pointer of type atomic64_t
+ *
+ * The function returns the old value of *v minus 1, even if
+ * the atomic variable, v, was not decremented.
+ */
+
+static inline s64 arch_atomic64_dec_if_positive(atomic64_t *v)
+{
+	s64 val;
+
+	smp_mb();
+
+	__asm__ __volatile__(
+	"1:	llockd  %0, [%1]	\n"
+	"	sub.f   %L0, %L0, 1	# w0 - 1, set C on borrow\n"
+	"	sub.c   %H0, %H0, 1	# if C set, w1 - 1\n"
+	"	brlt    %H0, 0, 2f	\n"
+	"	scondd  %0, [%1]	\n"
+	"	bnz     1b		\n"
+	"2:				\n"
+	: "=&r"(val)
+	: "r"(&v->counter)
+	: "cc");	/* memory clobber comes from smp_mb() */
+
+	smp_mb();
+
+	return val;
+}
+#define arch_atomic64_dec_if_positive arch_atomic64_dec_if_positive
+
+/**
+ * arch_atomic64_fetch_add_unless - add unless the number is a given value
+ * @v: pointer of type atomic64_t
+ * @a: the amount to add to v...
+ * @u: ...unless v is equal to u.
+ *
+ * Atomically adds @a to @v, if it was not @u.
+ * Returns the old value of @v
+ */
+static inline s64 arch_atomic64_fetch_add_unless(atomic64_t *v, s64 a, s64 u)
+{
+	s64 old, temp;
+
+	smp_mb();
+
+	__asm__ __volatile__(
+	"1:	llockd  %0, [%2]	\n"
+	"	brne	%L0, %L4, 2f	# continue to add since v != u \n"
+	"	breq.d	%H0, %H4, 3f	# return since v == u \n"
+	"2:				\n"
+	"	add.f   %L1, %L0, %L3	\n"
+	"	adc     %H1, %H0, %H3	\n"
+	"	scondd  %1, [%2]	\n"
+	"	bnz     1b		\n"
+	"3:				\n"
+	: "=&r"(old), "=&r" (temp)
+	: "r"(&v->counter), "r"(a), "r"(u)
+	: "cc");	/* memory clobber comes from smp_mb() */
+
+	smp_mb();
+
+	return old;
+}
+#define arch_atomic64_fetch_add_unless arch_atomic64_fetch_add_unless
+
+#endif

arch/arc/include/asm/bitops.h

@@ -14,188 +14,6 @@
 
 #include <linux/types.h>
 #include <linux/compiler.h>
-#include <asm/barrier.h>
-#ifndef CONFIG_ARC_HAS_LLSC
-#include <asm/smp.h>
-#endif
-
-#ifdef CONFIG_ARC_HAS_LLSC
-
-/*
- * Hardware assisted Atomic-R-M-W
- */
-
-#define BIT_OP(op, c_op, asm_op)					\
-static inline void op##_bit(unsigned long nr, volatile unsigned long *m)\
-{									\
-	unsigned int temp;						\
-									\
-	m += nr >> 5;							\
-									\
-	nr &= 0x1f;							\
-									\
-	__asm__ __volatile__(						\
-	"1:	llock       %0, [%1]		\n"			\
-	"	" #asm_op " %0, %0, %2	\n"				\
-	"	scond       %0, [%1]		\n"			\
-	"	bnz         1b			\n"			\
-	: "=&r"(temp)	/* Early clobber, to prevent reg reuse */	\
-	: "r"(m),	/* Not "m": llock only supports reg direct addr mode */	\
-	  "ir"(nr)							\
-	: "cc");							\
-}
-
-/*
- * Semantically:
- *    Test the bit
- *    if clear
- *        set it and return 0 (old value)
- *    else
- *        return 1 (old value).
- *
- * Since ARC lacks a equivalent h/w primitive, the bit is set unconditionally
- * and the old value of bit is returned
- */
-#define TEST_N_BIT_OP(op, c_op, asm_op)					\
-static inline int test_and_##op##_bit(unsigned long nr, volatile unsigned long *m)\
-{									\
-	unsigned long old, temp;					\
-									\
-	m += nr >> 5;							\
-									\
-	nr &= 0x1f;							\
-									\
-	/*								\
-	 * Explicit full memory barrier needed before/after as		\
-	 * LLOCK/SCOND themselves don't provide any such smenatic	\
-	 */								\
-	smp_mb();							\
-									\
-	__asm__ __volatile__(						\
-	"1:	llock       %0, [%2]	\n"				\
-	"	" #asm_op " %1, %0, %3	\n"				\
-	"	scond       %1, [%2]	\n"				\
-	"	bnz         1b		\n"				\
-	: "=&r"(old), "=&r"(temp)					\
-	: "r"(m), "ir"(nr)						\
-	: "cc");							\
-									\
-	smp_mb();							\
-									\
-	return (old & (1 << nr)) != 0;					\
-}
-
-#else /* !CONFIG_ARC_HAS_LLSC */
-
-/*
- * Non hardware assisted Atomic-R-M-W
- * Locking would change to irq-disabling only (UP) and spinlocks (SMP)
- *
- * There's "significant" micro-optimization in writing our own variants of
- * bitops (over generic variants)
- *
- * (1) The generic APIs have "signed" @nr while we have it "unsigned"
- *     This avoids extra code to be generated for pointer arithmatic, since
- *     is "not sure" that index is NOT -ve
- * (2) Utilize the fact that ARCompact bit fidding insn (BSET/BCLR/ASL) etc
- *     only consider bottom 5 bits of @nr, so NO need to mask them off.
- *     (GCC Quirk: however for constant @nr we still need to do the masking
- *             at compile time)
- */
-
-#define BIT_OP(op, c_op, asm_op)					\
-static inline void op##_bit(unsigned long nr, volatile unsigned long *m)\
-{									\
-	unsigned long temp, flags;					\
-	m += nr >> 5;							\
-									\
-	/*								\
-	 * spin lock/unlock provide the needed smp_mb() before/after	\
-	 */								\
-	bitops_lock(flags);						\
-									\
-	temp = *m;							\
-	*m = temp c_op (1UL << (nr & 0x1f));					\
-									\
-	bitops_unlock(flags);						\
-}
-
-#define TEST_N_BIT_OP(op, c_op, asm_op)					\
-static inline int test_and_##op##_bit(unsigned long nr, volatile unsigned long *m)\
-{									\
-	unsigned long old, flags;					\
-	m += nr >> 5;							\
-									\
-	bitops_lock(flags);						\
-									\
-	old = *m;							\
-	*m = old c_op (1UL << (nr & 0x1f));				\
-									\
-	bitops_unlock(flags);						\
-									\
-	return (old & (1UL << (nr & 0x1f))) != 0;			\
-}
-
-#endif
-
-/***************************************
- * Non atomic variants
- **************************************/
-
-#define __BIT_OP(op, c_op, asm_op)					\
-static inline void __##op##_bit(unsigned long nr, volatile unsigned long *m)	\
-{									\
-	unsigned long temp;						\
-	m += nr >> 5;							\
-									\
-	temp = *m;							\
-	*m = temp c_op (1UL << (nr & 0x1f));				\
-}
-
-#define __TEST_N_BIT_OP(op, c_op, asm_op)				\
-static inline int __test_and_##op##_bit(unsigned long nr, volatile unsigned long *m)\
-{									\
-	unsigned long old;						\
-	m += nr >> 5;							\
-									\
-	old = *m;							\
-	*m = old c_op (1UL << (nr & 0x1f));				\
-									\
-	return (old & (1UL << (nr & 0x1f))) != 0;			\
-}
-
-#define BIT_OPS(op, c_op, asm_op)					\
-									\
-	/* set_bit(), clear_bit(), change_bit() */			\
-	BIT_OP(op, c_op, asm_op)					\
-									\
-	/* test_and_set_bit(), test_and_clear_bit(), test_and_change_bit() */\
-	TEST_N_BIT_OP(op, c_op, asm_op)					\
-									\
-	/* __set_bit(), __clear_bit(), __change_bit() */		\
-	__BIT_OP(op, c_op, asm_op)					\
-									\
-	/* __test_and_set_bit(), __test_and_clear_bit(), __test_and_change_bit() */\
-	__TEST_N_BIT_OP(op, c_op, asm_op)
-
-BIT_OPS(set, |, bset)
-BIT_OPS(clear, & ~, bclr)
-BIT_OPS(change, ^, bxor)
-
-/*
- * This routine doesn't need to be atomic.
- */
-static inline int
-test_bit(unsigned int nr, const volatile unsigned long *addr)
-{
-	unsigned long mask;
-
-	addr += nr >> 5;
-
-	mask = 1UL << (nr & 0x1f);
-
-	return ((mask & *addr) != 0);
-}
 
 #ifdef CONFIG_ISA_ARCOMPACT
 
@@ -296,7 +114,7 @@ static inline __attribute__ ((const)) unsigned long __ffs(unsigned long word)
  * @result: [1-32]
  * fls(1) = 1, fls(0x80000000) = 32, fls(0) = 0
  */
-static inline __attribute__ ((const)) int fls(unsigned long x)
+static inline __attribute__ ((const)) int fls(unsigned int x)
 {
 	int n;
 
@@ -323,7 +141,7 @@ static inline __attribute__ ((const)) int __fls(unsigned long x)
  * ffs = Find First Set in word (LSB to MSB)
  * @result: [1-32], 0 if all 0's
  */
-static inline __attribute__ ((const)) int ffs(unsigned long x)
+static inline __attribute__ ((const)) int ffs(unsigned int x)
 {
 	int n;
 
@@ -368,6 +186,8 @@ static inline __attribute__ ((const)) unsigned long __ffs(unsigned long x)
 #include <asm-generic/bitops/fls64.h>
 #include <asm-generic/bitops/sched.h>
 #include <asm-generic/bitops/lock.h>
+#include <asm-generic/bitops/atomic.h>
+#include <asm-generic/bitops/non-atomic.h>
 
 #include <asm-generic/bitops/find.h>
 #include <asm-generic/bitops/le.h>

arch/arc/include/asm/cache.h

@@ -62,10 +62,6 @@
 #define ARCH_SLAB_MINALIGN	8
 #endif
 
-extern void arc_cache_init(void);
-extern char *arc_cache_mumbojumbo(int cpu_id, char *buf, int len);
-extern void read_decode_cache_bcr(void);
-
 extern int ioc_enable;
 extern unsigned long perip_base, perip_end;
 

arch/arc/include/asm/cmpxchg.h

@@ -6,6 +6,7 @@
 #ifndef __ASM_ARC_CMPXCHG_H
 #define __ASM_ARC_CMPXCHG_H
 
+#include <linux/build_bug.h>
 #include <linux/types.h>
 
 #include <asm/barrier.h>
@@ -13,146 +14,130 @@
 
 #ifdef CONFIG_ARC_HAS_LLSC
 
-static inline unsigned long
-__cmpxchg(volatile void *ptr, unsigned long expected, unsigned long new)
-{
-	unsigned long prev;
-
-	/*
-	 * Explicit full memory barrier needed before/after as
-	 * LLOCK/SCOND themselves don't provide any such semantics
+/*
+ * if (*ptr == @old)
+ *      *ptr = @new
  */
-	smp_mb();
-
-	__asm__ __volatile__(
-	"1:	llock   %0, [%1]	\n"
-	"	brne    %0, %2, 2f	\n"
-	"	scond   %3, [%1]	\n"
-	"	bnz     1b		\n"
-	"2:				\n"
-	: "=&r"(prev)	/* Early clobber, to prevent reg reuse */
-	: "r"(ptr),	/* Not "m": llock only supports reg direct addr mode */
-	  "ir"(expected),
-	  "r"(new)	/* can't be "ir". scond can't take LIMM for "b" */
-	: "cc", "memory"); /* so that gcc knows memory is being written here */
-
-	smp_mb();
-
-	return prev;
-}
-
-#else /* !CONFIG_ARC_HAS_LLSC */
-
-static inline unsigned long
-__cmpxchg(volatile void *ptr, unsigned long expected, unsigned long new)
-{
-	unsigned long flags;
-	int prev;
-	volatile unsigned long *p = ptr;
+#define __cmpxchg(ptr, old, new)					\
+({									\
+	__typeof__(*(ptr)) _prev;					\
+									\
+	__asm__ __volatile__(						\
+	"1:	llock  %0, [%1]	\n"					\
+	"	brne   %0, %2, 2f	\n"				\
+	"	scond  %3, [%1]	\n"					\
+	"	bnz     1b		\n"				\
+	"2:				\n"				\
+	: "=&r"(_prev)	/* Early clobber prevent reg reuse */		\
+	: "r"(ptr),	/* Not "m": llock only supports reg */		\
+	  "ir"(old),							\
+	  "r"(new)	/* Not "ir": scond can't take LIMM */		\
+	: "cc",								\
+	  "memory");	/* gcc knows memory is clobbered */		\
+									\
+	_prev;								\
+})
 
-	/*
-	 * spin lock/unlock provide the needed smp_mb() before/after
-	 */
-	atomic_ops_lock(flags);
-	prev = *p;
-	if (prev == expected)
-		*p = new;
-	atomic_ops_unlock(flags);
-	return prev;
-}
+#define arch_cmpxchg_relaxed(ptr, old, new)				\
+({									\
+	__typeof__(ptr) _p_ = (ptr);					\
+	__typeof__(*(ptr)) _o_ = (old);					\
+	__typeof__(*(ptr)) _n_ = (new);					\
+	__typeof__(*(ptr)) _prev_;					\
+									\
+	switch(sizeof((_p_))) {						\
+	case 4:								\
+		_prev_ = __cmpxchg(_p_, _o_, _n_);			\
+		break;							\
+	default:							\
+		BUILD_BUG();						\
+	}								\
+	_prev_;								\
+})
 
-#endif
+#else
 
-#define arch_cmpxchg(ptr, o, n) ({			\
-	(typeof(*(ptr)))__cmpxchg((ptr),		\
-				  (unsigned long)(o),	\
-				  (unsigned long)(n));	\
+#define arch_cmpxchg(ptr, old, new)				        \
+({									\
+	volatile __typeof__(ptr) _p_ = (ptr);				\
+	__typeof__(*(ptr)) _o_ = (old);					\
+	__typeof__(*(ptr)) _n_ = (new);					\
+	__typeof__(*(ptr)) _prev_;					\
+	unsigned long __flags;						\
+									\
+	BUILD_BUG_ON(sizeof(_p_) != 4);					\
+									\
+	/*								\
+	 * spin lock/unlock provide the needed smp_mb() before/after	\
+	 */								\
+	atomic_ops_lock(__flags);					\
+	_prev_ = *_p_;							\
+	if (_prev_ == _o_)						\
+		*_p_ = _n_;						\
+	atomic_ops_unlock(__flags);					\
+	_prev_;								\
 })
 
-/*
- * atomic_cmpxchg is same as cmpxchg
- *   LLSC: only different in data-type, semantics are exactly same
- *  !LLSC: cmpxchg() has to use an external lock atomic_ops_lock to guarantee
- *         semantics, and this lock also happens to be used by atomic_*()
- */
-#define arch_atomic_cmpxchg(v, o, n) ((int)arch_cmpxchg(&((v)->counter), (o), (n)))
-
+#endif
 
 /*
- * xchg (reg with memory) based on "Native atomic" EX insn
+ * xchg
  */
-static inline unsigned long __xchg(unsigned long val, volatile void *ptr,
-				   int size)
-{
-	extern unsigned long __xchg_bad_pointer(void);
-
-	switch (size) {
-	case 4:
-		smp_mb();
-
-		__asm__ __volatile__(
-		"	ex  %0, [%1]	\n"
-		: "+r"(val)
-		: "r"(ptr)
-		: "memory");
+#ifdef CONFIG_ARC_HAS_LLSC
 
-		smp_mb();
+#define __xchg(ptr, val)						\
+({									\
+	__asm__ __volatile__(						\
+	"	ex  %0, [%1]	\n"	/* set new value */	        \
+	: "+r"(val)							\
+	: "r"(ptr)							\
+	: "memory");							\
+	_val_;		/* get old value */				\
+})
 
-		return val;
-	}
-	return __xchg_bad_pointer();
-}
+#define arch_xchg_relaxed(ptr, val)					\
+({									\
+	__typeof__(ptr) _p_ = (ptr);					\
+	__typeof__(*(ptr)) _val_ = (val);				\
+									\
+	switch(sizeof(*(_p_))) {					\
+	case 4:								\
+		_val_ = __xchg(_p_, _val_);				\
+		break;							\
+	default:							\
+		BUILD_BUG();						\
+	}								\
+	_val_;								\
+})
 
-#define _xchg(ptr, with) ((typeof(*(ptr)))__xchg((unsigned long)(with), (ptr), \
-						 sizeof(*(ptr))))
+#else  /* !CONFIG_ARC_HAS_LLSC */
 
 /*
- * xchg() maps directly to ARC EX instruction which guarantees atomicity.
- * However in !LLSC config, it also needs to be use @atomic_ops_lock spinlock
- * due to a subtle reason:
- *  - For !LLSC, cmpxchg() needs to use that lock (see above) and there is lot
- *    of  kernel code which calls xchg()/cmpxchg() on same data (see llist.h)
- *    Hence xchg() needs to follow same locking rules.
- *
- * Technically the lock is also needed for UP (boils down to irq save/restore)
- * but we can cheat a bit since cmpxchg() atomic_ops_lock() would cause irqs to
- * be disabled thus can't possibly be interrupted/preempted/clobbered by xchg()
- * Other way around, xchg is one instruction anyways, so can't be interrupted
- * as such
+ * EX instructions is baseline and present in !LLSC too. But in this
+ * regime it still needs use @atomic_ops_lock spinlock to allow interop
+ * with cmpxchg() which uses spinlock in !LLSC
+ * (llist.h use xchg and cmpxchg on sama data)
  */
 
-#if !defined(CONFIG_ARC_HAS_LLSC) && defined(CONFIG_SMP)
-
-#define arch_xchg(ptr, with)		\
+#define arch_xchg(ptr, val)					        \
 ({									\
-	unsigned long flags;		\
-	typeof(*(ptr)) old_val;		\
+	__typeof__(ptr) _p_ = (ptr);					\
+	__typeof__(*(ptr)) _val_ = (val);				\
+									\
+	unsigned long __flags;						\
 									\
-	atomic_ops_lock(flags);		\
-	old_val = _xchg(ptr, with);	\
-	atomic_ops_unlock(flags);	\
-	old_val;			\
+	atomic_ops_lock(__flags);					\
+									\
+	__asm__ __volatile__(						\
+	"	ex  %0, [%1]	\n"					\
+	: "+r"(_val_)							\
+	: "r"(_p_)							\
+	: "memory");							\
+									\
+	atomic_ops_unlock(__flags);					\
+	_val_;								\
 })
 
-#else
-
-#define arch_xchg(ptr, with)  _xchg(ptr, with)
-
 #endif
 
-/*
- * "atomic" variant of xchg()
- * REQ: It needs to follow the same serialization rules as other atomic_xxx()
- * Since xchg() doesn't always do that, it would seem that following definition
- * is incorrect. But here's the rationale:
- *   SMP : Even xchg() takes the atomic_ops_lock, so OK.
- *   LLSC: atomic_ops_lock are not relevant at all (even if SMP, since LLSC
- *         is natively "SMP safe", no serialization required).
- *   UP  : other atomics disable IRQ, so no way a difft ctxt atomic_xchg()
- *         could clobber them. atomic_xchg() itself would be 1 insn, so it
- *         can't be clobbered by others. Thus no serialization required when
- *         atomic_xchg is involved.
- */
-#define arch_atomic_xchg(v, new) (arch_xchg(&((v)->counter), new))
-
 #endif

arch/arc/include/asm/entry-compact.h

@@ -126,19 +126,11 @@
  * to be saved again on kernel mode stack, as part of pt_regs.
  *-------------------------------------------------------------*/
 .macro PROLOG_FREEUP_REG	reg, mem
-#ifndef ARC_USE_SCRATCH_REG
-	sr  \reg, [ARC_REG_SCRATCH_DATA0]
-#else
 	st  \reg, [\mem]
-#endif
 .endm
 
 .macro PROLOG_RESTORE_REG	reg, mem
-#ifndef ARC_USE_SCRATCH_REG
-	lr  \reg, [ARC_REG_SCRATCH_DATA0]
-#else
 	ld  \reg, [\mem]
-#endif
 .endm
 
 /*--------------------------------------------------------------

arch/arc/include/asm/hugepage.h

@@ -58,14 +58,6 @@ static inline void set_pmd_at(struct mm_struct *mm, unsigned long addr,
 extern void update_mmu_cache_pmd(struct vm_area_struct *vma, unsigned long addr,
 				 pmd_t *pmd);
 
-/* Generic variants assume pgtable_t is struct page *, hence need for these */
-#define __HAVE_ARCH_PGTABLE_DEPOSIT
-extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
-				       pgtable_t pgtable);
-
-#define __HAVE_ARCH_PGTABLE_WITHDRAW
-extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp);
-
 #define __HAVE_ARCH_FLUSH_PMD_TLB_RANGE
 extern void flush_pmd_tlb_range(struct vm_area_struct *vma, unsigned long start,
 				unsigned long end);

arch/arc/include/asm/mmu-arcv2.h

+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (C) 2004, 2007-2010, 2011-2012, 2019-20 Synopsys, Inc. (www.synopsys.com)
+ *
+ * MMUv3 (arc700) / MMUv4 (archs) are software page walked and software managed.
+ * This file contains the TLB access registers and commands
+ */
+
+#ifndef _ASM_ARC_MMU_ARCV2_H
+#define _ASM_ARC_MMU_ARCV2_H
+
+/*
+ * TLB Management regs
+ */
+#define ARC_REG_MMU_BCR		0x06f
+
+#ifdef CONFIG_ARC_MMU_V3
+#define ARC_REG_TLBPD0		0x405
+#define ARC_REG_TLBPD1		0x406
+#define ARC_REG_TLBPD1HI	0	/* Dummy: allows common code */
+#define ARC_REG_TLBINDEX	0x407
+#define ARC_REG_TLBCOMMAND	0x408
+#define ARC_REG_PID		0x409
+#define ARC_REG_SCRATCH_DATA0	0x418
+#else
+#define ARC_REG_TLBPD0		0x460
+#define ARC_REG_TLBPD1		0x461
+#define ARC_REG_TLBPD1HI	0x463
+#define ARC_REG_TLBINDEX	0x464
+#define ARC_REG_TLBCOMMAND	0x465
+#define ARC_REG_PID		0x468
+#define ARC_REG_SCRATCH_DATA0	0x46c
+#endif
+
+/* Bits in MMU PID reg */
+#define __TLB_ENABLE		(1 << 31)
+#define __PROG_ENABLE		(1 << 30)
+#define MMU_ENABLE		(__TLB_ENABLE | __PROG_ENABLE)
+
+/* Bits in TLB Index reg */
+#define TLB_LKUP_ERR		0x80000000
+
+#ifdef CONFIG_ARC_MMU_V3
+#define TLB_DUP_ERR		(TLB_LKUP_ERR | 0x00000001)
+#else
+#define TLB_DUP_ERR		(TLB_LKUP_ERR | 0x40000000)
+#endif
+
+/*
+ * TLB Commands
+ */
+#define TLBWrite    		0x1
+#define TLBRead     		0x2
+#define TLBGetIndex 		0x3
+#define TLBProbe    		0x4
+#define TLBWriteNI		0x5  /* write JTLB without inv uTLBs */
+#define TLBIVUTLB		0x6  /* explicitly inv uTLBs */
+
+#ifdef CONFIG_ARC_MMU_V4
+#define TLBInsertEntry		0x7
+#define TLBDeleteEntry		0x8
+#endif
+
+/* Masks for actual TLB "PD"s */
+#define PTE_BITS_IN_PD0		(_PAGE_GLOBAL | _PAGE_PRESENT | _PAGE_HW_SZ)
+#define PTE_BITS_RWX		(_PAGE_EXECUTE | _PAGE_WRITE | _PAGE_READ)
+
+#define PTE_BITS_NON_RWX_IN_PD1	(PAGE_MASK_PHYS | _PAGE_CACHEABLE)
+
+#ifndef __ASSEMBLY__
+
+struct mm_struct;
+extern int pae40_exist_but_not_enab(void);
+
+static inline int is_pae40_enabled(void)
+{
+	return IS_ENABLED(CONFIG_ARC_HAS_PAE40);
+}
+
+static inline void mmu_setup_asid(struct mm_struct *mm, unsigned long asid)
+{
+	write_aux_reg(ARC_REG_PID, asid | MMU_ENABLE);
+}
+
+static inline void mmu_setup_pgd(struct mm_struct *mm, void *pgd)
+{
+	/* PGD cached in MMU reg to avoid 3 mem lookups: task->mm->pgd */
+#ifdef CONFIG_ISA_ARCV2
+	write_aux_reg(ARC_REG_SCRATCH_DATA0, (unsigned int)pgd);
+#endif
+}
+
+#else
+
+.macro ARC_MMU_REENABLE reg
+	lr \reg, [ARC_REG_PID]
+	or \reg, \reg, MMU_ENABLE
+	sr \reg, [ARC_REG_PID]
+.endm
+
+#endif /* !__ASSEMBLY__ */
+
+#endif

arch/arc/include/asm/mmu.h

@@ -7,98 +7,15 @@
 #define _ASM_ARC_MMU_H
 
 #ifndef __ASSEMBLY__
-#include <linux/threads.h>	/* NR_CPUS */
-#endif
-
-#if defined(CONFIG_ARC_MMU_V1)
-#define CONFIG_ARC_MMU_VER 1
-#elif defined(CONFIG_ARC_MMU_V2)
-#define CONFIG_ARC_MMU_VER 2
-#elif defined(CONFIG_ARC_MMU_V3)
-#define CONFIG_ARC_MMU_VER 3
-#elif defined(CONFIG_ARC_MMU_V4)
-#define CONFIG_ARC_MMU_VER 4
-#endif
-
-/* MMU Management regs */
-#define ARC_REG_MMU_BCR		0x06f
-#if (CONFIG_ARC_MMU_VER < 4)
-#define ARC_REG_TLBPD0		0x405
-#define ARC_REG_TLBPD1		0x406
-#define ARC_REG_TLBPD1HI	0	/* Dummy: allows code sharing with ARC700 */
-#define ARC_REG_TLBINDEX	0x407
-#define ARC_REG_TLBCOMMAND	0x408
-#define ARC_REG_PID		0x409
-#define ARC_REG_SCRATCH_DATA0	0x418
-#else
-#define ARC_REG_TLBPD0		0x460
-#define ARC_REG_TLBPD1		0x461
-#define ARC_REG_TLBPD1HI	0x463
-#define ARC_REG_TLBINDEX	0x464
-#define ARC_REG_TLBCOMMAND	0x465
-#define ARC_REG_PID		0x468
-#define ARC_REG_SCRATCH_DATA0	0x46c
-#endif
-
-#if defined(CONFIG_ISA_ARCV2) || !defined(CONFIG_SMP)
-#define	ARC_USE_SCRATCH_REG
-#endif
-
-/* Bits in MMU PID register */
-#define __TLB_ENABLE		(1 << 31)
-#define __PROG_ENABLE		(1 << 30)
-#define MMU_ENABLE		(__TLB_ENABLE | __PROG_ENABLE)
-
-/* Error code if probe fails */
-#define TLB_LKUP_ERR		0x80000000
-
-#if (CONFIG_ARC_MMU_VER < 4)
-#define TLB_DUP_ERR	(TLB_LKUP_ERR | 0x00000001)
-#else
-#define TLB_DUP_ERR	(TLB_LKUP_ERR | 0x40000000)
-#endif
-
-/* TLB Commands */
-#define TLBWrite    0x1
-#define TLBRead     0x2
-#define TLBGetIndex 0x3
-#define TLBProbe    0x4
-
-#if (CONFIG_ARC_MMU_VER >= 2)
-#define TLBWriteNI  0x5		/* write JTLB without inv uTLBs */
-#define TLBIVUTLB   0x6		/* explicitly inv uTLBs */
-#else
-#define TLBWriteNI  TLBWrite	/* Not present in hardware, fallback */
-#endif
-
-#if (CONFIG_ARC_MMU_VER >= 4)
-#define TLBInsertEntry	0x7
-#define TLBDeleteEntry	0x8
-#endif
 
-#ifndef __ASSEMBLY__
+#include <linux/threads.h>	/* NR_CPUS */
 
 typedef struct {
 	unsigned long asid[NR_CPUS];	/* 8 bit MMU PID + Generation cycle */
 } mm_context_t;
 
-#ifdef CONFIG_ARC_DBG_TLB_PARANOIA
-void tlb_paranoid_check(unsigned int mm_asid, unsigned long address);
-#else
-#define tlb_paranoid_check(a, b)
 #endif
 
-void arc_mmu_init(void);
-extern char *arc_mmu_mumbojumbo(int cpu_id, char *buf, int len);
-void read_decode_mmu_bcr(void);
-
-static inline int is_pae40_enabled(void)
-{
-	return IS_ENABLED(CONFIG_ARC_HAS_PAE40);
-}
-
-extern int pae40_exist_but_not_enab(void);
-
-#endif	/* !__ASSEMBLY__ */
+#include <asm/mmu-arcv2.h>
 
 #endif

arch/arc/include/asm/mmu_context.h

@@ -15,22 +15,23 @@
 #ifndef _ASM_ARC_MMU_CONTEXT_H
 #define _ASM_ARC_MMU_CONTEXT_H
 
-#include <asm/arcregs.h>
-#include <asm/tlb.h>
 #include <linux/sched/mm.h>
 
+#include <asm/tlb.h>
 #include <asm-generic/mm_hooks.h>
 
-/*		ARC700 ASID Management
+/*		ARC ASID Management
+ *
+ * MMU tags TLBs with an 8-bit ASID, avoiding need to flush the TLB on
+ * context-switch.
  *
- * ARC MMU provides 8-bit ASID (0..255) to TAG TLB entries, allowing entries
- * with same vaddr (different tasks) to co-exit. This provides for
- * "Fast Context Switch" i.e. no TLB flush on ctxt-switch
+ * ASID is managed per cpu, so task threads across CPUs can have different
+ * ASID. Global ASID management is needed if hardware supports TLB shootdown
+ * and/or shared TLB across cores, which ARC doesn't.
  *
- * Linux assigns each task a unique ASID. A simple round-robin allocation
- * of H/w ASID is done using software tracker @asid_cpu.
- * When it reaches max 255, the allocation cycle starts afresh by flushing
- * the entire TLB and wrapping ASID back to zero.
+ * Each task is assigned unique ASID, with a simple round-robin allocator
+ * tracked in @asid_cpu. When 8-bit value rolls over,a new cycle is started
+ * over from 0, and TLB is flushed
  *
  * A new allocation cycle, post rollover, could potentially reassign an ASID
  * to a different task. Thus the rule is to refresh the ASID in a new cycle.
@@ -93,7 +94,7 @@ static inline void get_new_mmu_context(struct mm_struct *mm)
 	asid_mm(mm, cpu) = asid_cpu(cpu);
 
 set_hw:
-	write_aux_reg(ARC_REG_PID, hw_pid(mm, cpu) | MMU_ENABLE);
+	mmu_setup_asid(mm, hw_pid(mm, cpu));
 
 	local_irq_restore(flags);
 }
@@ -146,10 +147,7 @@ static inline void switch_mm(struct mm_struct *prev, struct mm_struct *next,
 	 */
 	cpumask_set_cpu(cpu, mm_cpumask(next));
 
-#ifdef ARC_USE_SCRATCH_REG
-	/* PGD cached in MMU reg to avoid 3 mem lookups: task->mm->pgd */
-	write_aux_reg(ARC_REG_SCRATCH_DATA0, next->pgd);
-#endif
+	mmu_setup_pgd(next, next->pgd);
 
 	get_new_mmu_context(next);
 }

arch/arc/include/asm/page.h

@@ -34,12 +34,35 @@ void copy_user_highpage(struct page *to, struct page *from,
 			unsigned long u_vaddr, struct vm_area_struct *vma);
 void clear_user_page(void *to, unsigned long u_vaddr, struct page *page);
 
-#undef STRICT_MM_TYPECHECKS
+typedef struct {
+	unsigned long pgd;
+} pgd_t;
+
+#define pgd_val(x)	((x).pgd)
+#define __pgd(x)	((pgd_t) { (x) })
+
+#if CONFIG_PGTABLE_LEVELS > 3
+
+typedef struct {
+	unsigned long pud;
+} pud_t;
+
+#define pud_val(x)      	((x).pud)
+#define __pud(x)        	((pud_t) { (x) })
+
+#endif
+
+#if CONFIG_PGTABLE_LEVELS > 2
+
+typedef struct {
+	unsigned long pmd;
+} pmd_t;
+
+#define pmd_val(x)	((x).pmd)
+#define __pmd(x)	((pmd_t) { (x) })
+
+#endif
 
-#ifdef STRICT_MM_TYPECHECKS
-/*
- * These are used to make use of C type-checking..
- */
 typedef struct {
 #ifdef CONFIG_ARC_HAS_PAE40
 	unsigned long long pte;
@@ -47,44 +70,19 @@ typedef struct {
 	unsigned long pte;
 #endif
 } pte_t;
-typedef struct {
-	unsigned long pgd;
-} pgd_t;
+
+#define pte_val(x)	((x).pte)
+#define __pte(x)	((pte_t) { (x) })
+
 typedef struct {
 	unsigned long pgprot;
 } pgprot_t;
 
-#define pte_val(x)      ((x).pte)
-#define pgd_val(x)      ((x).pgd)
 #define pgprot_val(x)	((x).pgprot)
-
-#define __pte(x)        ((pte_t) { (x) })
-#define __pgd(x)        ((pgd_t) { (x) })
 #define __pgprot(x)	((pgprot_t) { (x) })
-
 #define pte_pgprot(x)	__pgprot(pte_val(x))
 
-#else /* !STRICT_MM_TYPECHECKS */
-
-#ifdef CONFIG_ARC_HAS_PAE40
-typedef unsigned long long pte_t;
-#else
-typedef unsigned long pte_t;
-#endif
-typedef unsigned long pgd_t;
-typedef unsigned long pgprot_t;
-
-#define pte_val(x)	(x)
-#define pgd_val(x)	(x)
-#define pgprot_val(x)	(x)
-#define __pte(x)	(x)
-#define __pgd(x)	(x)
-#define __pgprot(x)	(x)
-#define pte_pgprot(x)	(x)
-
-#endif
-
-typedef pte_t * pgtable_t;
+typedef struct page *pgtable_t;
 
 /*
  * Use virt_to_pfn with caution:

arch/arc/include/asm/pgalloc.h

@@ -31,30 +31,32 @@
 
 #include <linux/mm.h>
 #include <linux/log2.h>
+#include <asm-generic/pgalloc.h>
 
 static inline void
 pmd_populate_kernel(struct mm_struct *mm, pmd_t *pmd, pte_t *pte)
 {
-	pmd_set(pmd, pte);
-}
-
-static inline void
-pmd_populate(struct mm_struct *mm, pmd_t *pmd, pgtable_t ptep)
-{
-	pmd_set(pmd, (pte_t *) ptep);
+	/*
+	 * The cast to long below is OK in 32-bit PAE40 regime with long long pte
+	 * Despite "wider" pte, the pte table needs to be in non-PAE low memory
+	 * as all higher levels can only hold long pointers.
+	 *
+	 * The cast itself is needed given simplistic definition of set_pmd()
+	 */
+	set_pmd(pmd, __pmd((unsigned long)pte));
 }
 
-static inline int __get_order_pgd(void)
+static inline void pmd_populate(struct mm_struct *mm, pmd_t *pmd, pgtable_t pte_page)
 {
-	return get_order(PTRS_PER_PGD * sizeof(pgd_t));
+	set_pmd(pmd, __pmd((unsigned long)page_address(pte_page)));
 }
 
 static inline pgd_t *pgd_alloc(struct mm_struct *mm)
 {
-	int num, num2;
-	pgd_t *ret = (pgd_t *) __get_free_pages(GFP_KERNEL, __get_order_pgd());
+	pgd_t *ret = (pgd_t *) __get_free_page(GFP_KERNEL);
 
 	if (ret) {
+		int num, num2;
 		num = USER_PTRS_PER_PGD + USER_KERNEL_GUTTER / PGDIR_SIZE;
 		memzero(ret, num * sizeof(pgd_t));
 
@@ -68,64 +70,27 @@ static inline pgd_t *pgd_alloc(struct mm_struct *mm)
 	return ret;
 }
 
-static inline void pgd_free(struct mm_struct *mm, pgd_t *pgd)
-{
-	free_pages((unsigned long)pgd, __get_order_pgd());
-}
-
+#if CONFIG_PGTABLE_LEVELS > 3
 
-/*
- * With software-only page-tables, addr-split for traversal is tweakable and
- * that directly governs how big tables would be at each level.
- * Further, the MMU page size is configurable.
- * Thus we need to programatically assert the size constraint
- * All of this is const math, allowing gcc to do constant folding/propagation.
- */
-
-static inline int __get_order_pte(void)
+static inline void p4d_populate(struct mm_struct *mm, p4d_t *p4dp, pud_t *pudp)
 {
-	return get_order(PTRS_PER_PTE * sizeof(pte_t));
+	set_p4d(p4dp, __p4d((unsigned long)pudp));
 }
 
-static inline pte_t *pte_alloc_one_kernel(struct mm_struct *mm)
-{
-	pte_t *pte;
+#define __pud_free_tlb(tlb, pmd, addr)  pud_free((tlb)->mm, pmd)
 
-	pte = (pte_t *) __get_free_pages(GFP_KERNEL | __GFP_ZERO,
-					 __get_order_pte());
+#endif
 
-	return pte;
-}
+#if CONFIG_PGTABLE_LEVELS > 2
 
-static inline pgtable_t
-pte_alloc_one(struct mm_struct *mm)
+static inline void pud_populate(struct mm_struct *mm, pud_t *pudp, pmd_t *pmdp)
 {
-	pgtable_t pte_pg;
-	struct page *page;
-
-	pte_pg = (pgtable_t)__get_free_pages(GFP_KERNEL, __get_order_pte());
-	if (!pte_pg)
-		return 0;
-	memzero((void *)pte_pg, PTRS_PER_PTE * sizeof(pte_t));
-	page = virt_to_page(pte_pg);
-	if (!pgtable_pte_page_ctor(page)) {
-		__free_page(page);
-		return 0;
-	}
-
-	return pte_pg;
+	set_pud(pudp, __pud((unsigned long)pmdp));
 }
 
-static inline void pte_free_kernel(struct mm_struct *mm, pte_t *pte)
-{
-	free_pages((unsigned long)pte, __get_order_pte()); /* takes phy addr */
-}
+#define __pmd_free_tlb(tlb, pmd, addr)  pmd_free((tlb)->mm, pmd)
 
-static inline void pte_free(struct mm_struct *mm, pgtable_t ptep)
-{
-	pgtable_pte_page_dtor(virt_to_page(ptep));
-	free_pages((unsigned long)ptep, __get_order_pte());
-}
+#endif
 
 #define __pte_free_tlb(tlb, pte, addr)  pte_free((tlb)->mm, pte)
 

arch/arc/include/asm/pgtable-bits-arcv2.h

+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (C) 2004, 2007-2010, 2011-2012 Synopsys, Inc. (www.synopsys.com)
+ */
+
+/*
+ * page table flags for software walked/managed MMUv3 (ARC700) and MMUv4 (HS)
+ * There correspond to the corresponding bits in the TLB
+ */
+
+#ifndef _ASM_ARC_PGTABLE_BITS_ARCV2_H
+#define _ASM_ARC_PGTABLE_BITS_ARCV2_H
+
+#ifdef CONFIG_ARC_CACHE_PAGES
+#define _PAGE_CACHEABLE		(1 << 0)  /* Cached (H) */
+#else
+#define _PAGE_CACHEABLE		0
+#endif
+
+#define _PAGE_EXECUTE		(1 << 1)  /* User Execute  (H) */
+#define _PAGE_WRITE		(1 << 2)  /* User Write    (H) */
+#define _PAGE_READ		(1 << 3)  /* User Read     (H) */
+#define _PAGE_ACCESSED		(1 << 4)  /* Accessed      (s) */
+#define _PAGE_DIRTY		(1 << 5)  /* Modified      (s) */
+#define _PAGE_SPECIAL		(1 << 6)
+#define _PAGE_GLOBAL		(1 << 8)  /* ASID agnostic (H) */
+#define _PAGE_PRESENT		(1 << 9)  /* PTE/TLB Valid (H) */
+
+#ifdef CONFIG_ARC_MMU_V4
+#define _PAGE_HW_SZ		(1 << 10)  /* Normal/super (H) */
+#else
+#define _PAGE_HW_SZ		0
+#endif
+
+/* Defaults for every user page */
+#define ___DEF		(_PAGE_PRESENT | _PAGE_CACHEABLE)
+
+/* Set of bits not changed in pte_modify */
+#define _PAGE_CHG_MASK	(PAGE_MASK_PHYS | _PAGE_ACCESSED | _PAGE_DIRTY | \
+							   _PAGE_SPECIAL)
+
+/* More Abbrevaited helpers */
+#define PAGE_U_NONE     __pgprot(___DEF)
+#define PAGE_U_R        __pgprot(___DEF | _PAGE_READ)
+#define PAGE_U_W_R      __pgprot(___DEF | _PAGE_READ | _PAGE_WRITE)
+#define PAGE_U_X_R      __pgprot(___DEF | _PAGE_READ | _PAGE_EXECUTE)
+#define PAGE_U_X_W_R    __pgprot(___DEF \
+				| _PAGE_READ | _PAGE_WRITE | _PAGE_EXECUTE)
+#define PAGE_KERNEL     __pgprot(___DEF | _PAGE_GLOBAL \
+				| _PAGE_READ | _PAGE_WRITE | _PAGE_EXECUTE)
+
+#define PAGE_SHARED	PAGE_U_W_R
+
+#define pgprot_noncached(prot)	(__pgprot(pgprot_val(prot) & ~_PAGE_CACHEABLE))
+
+/*
+ * Mapping of vm_flags (Generic VM) to PTE flags (arch specific)
+ *
+ * Certain cases have 1:1 mapping
+ *  e.g. __P101 means VM_READ, VM_EXEC and !VM_SHARED
+ *       which directly corresponds to  PAGE_U_X_R
+ *
+ * Other rules which cause the divergence from 1:1 mapping
+ *
+ *  1. Although ARC700 can do exclusive execute/write protection (meaning R
+ *     can be tracked independet of X/W unlike some other CPUs), still to
+ *     keep things consistent with other archs:
+ *      -Write implies Read:   W => R
+ *      -Execute implies Read: X => R
+ *
+ *  2. Pvt Writable doesn't have Write Enabled initially: Pvt-W => !W
+ *     This is to enable COW mechanism
+ */
+	/* xwr */
+#define __P000  PAGE_U_NONE
+#define __P001  PAGE_U_R
+#define __P010  PAGE_U_R	/* Pvt-W => !W */
+#define __P011  PAGE_U_R	/* Pvt-W => !W */
+#define __P100  PAGE_U_X_R	/* X => R */
+#define __P101  PAGE_U_X_R
+#define __P110  PAGE_U_X_R	/* Pvt-W => !W and X => R */
+#define __P111  PAGE_U_X_R	/* Pvt-W => !W */
+
+#define __S000  PAGE_U_NONE
+#define __S001  PAGE_U_R
+#define __S010  PAGE_U_W_R	/* W => R */
+#define __S011  PAGE_U_W_R
+#define __S100  PAGE_U_X_R	/* X => R */
+#define __S101  PAGE_U_X_R
+#define __S110  PAGE_U_X_W_R	/* X => R */
+#define __S111  PAGE_U_X_W_R
+
+#ifndef __ASSEMBLY__
+
+#define pte_write(pte)		(pte_val(pte) & _PAGE_WRITE)
+#define pte_dirty(pte)		(pte_val(pte) & _PAGE_DIRTY)
+#define pte_young(pte)		(pte_val(pte) & _PAGE_ACCESSED)
+#define pte_special(pte)	(pte_val(pte) & _PAGE_SPECIAL)
+
+#define PTE_BIT_FUNC(fn, op) \
+	static inline pte_t pte_##fn(pte_t pte) { pte_val(pte) op; return pte; }
+
+PTE_BIT_FUNC(mknotpresent,     &= ~(_PAGE_PRESENT));
+PTE_BIT_FUNC(wrprotect,	&= ~(_PAGE_WRITE));
+PTE_BIT_FUNC(mkwrite,	|= (_PAGE_WRITE));
+PTE_BIT_FUNC(mkclean,	&= ~(_PAGE_DIRTY));
+PTE_BIT_FUNC(mkdirty,	|= (_PAGE_DIRTY));
+PTE_BIT_FUNC(mkold,	&= ~(_PAGE_ACCESSED));
+PTE_BIT_FUNC(mkyoung,	|= (_PAGE_ACCESSED));
+PTE_BIT_FUNC(mkspecial,	|= (_PAGE_SPECIAL));
+PTE_BIT_FUNC(mkhuge,	|= (_PAGE_HW_SZ));
+
+static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
+{
+	return __pte((pte_val(pte) & _PAGE_CHG_MASK) | pgprot_val(newprot));
+}
+
+static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
+			      pte_t *ptep, pte_t pteval)
+{
+	set_pte(ptep, pteval);
+}
+
+void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
+		      pte_t *ptep);
+
+/* Encode swap {type,off} tuple into PTE
+ * We reserve 13 bits for 5-bit @type, keeping bits 12-5 zero, ensuring that
+ * PAGE_PRESENT is zero in a PTE holding swap "identifier"
+ */
+#define __swp_entry(type, off)		((swp_entry_t) \
+					{ ((type) & 0x1f) | ((off) << 13) })
+
+/* Decode a PTE containing swap "identifier "into constituents */
+#define __swp_type(pte_lookalike)	(((pte_lookalike).val) & 0x1f)
+#define __swp_offset(pte_lookalike)	((pte_lookalike).val >> 13)
+
+#define __pte_to_swp_entry(pte)		((swp_entry_t) { pte_val(pte) })
+#define __swp_entry_to_pte(x)		((pte_t) { (x).val })
+
+#define kern_addr_valid(addr)	(1)
+
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE
+#include <asm/hugepage.h>
+#endif
+
+#endif /* __ASSEMBLY__ */
+
+#endif

arch/arc/include/asm/pgtable-levels.h

+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (C) 2020 Synopsys, Inc. (www.synopsys.com)
+ */
+
+/*
+ * Helpers for implemenintg paging levels
+ */
+
+#ifndef _ASM_ARC_PGTABLE_LEVELS_H
+#define _ASM_ARC_PGTABLE_LEVELS_H
+
+#if CONFIG_PGTABLE_LEVELS == 2
+
+/*
+ * 2 level paging setup for software walked MMUv3 (ARC700) and MMUv4 (HS)
+ *
+ * [31]            32 bit virtual address              [0]
+ * -------------------------------------------------------
+ * |               | <---------- PGDIR_SHIFT ----------> |
+ * |               |                | <-- PAGE_SHIFT --> |
+ * -------------------------------------------------------
+ *       |                  |                |
+ *       |                  |                --> off in page frame
+ *       |                  ---> index into Page Table
+ *       ----> index into Page Directory
+ *
+ * Given software walk, the vaddr split is arbitrary set to 11:8:13
+ * However enabling of super page in a 2 level regime pegs PGDIR_SHIFT to
+ * super page size.
+ */
+
+#if defined(CONFIG_ARC_HUGEPAGE_16M)
+#define PGDIR_SHIFT		24
+#elif defined(CONFIG_ARC_HUGEPAGE_2M)
+#define PGDIR_SHIFT		21
+#else
+/*
+ * No Super page case
+ * Default value provides 11:8:13 (8K), 10:10:12 (4K)
+ * Limits imposed by pgtable_t only PAGE_SIZE long
+ * (so 4K page can only have 1K entries: or 10 bits)
+ */
+#ifdef CONFIG_ARC_PAGE_SIZE_4K
+#define PGDIR_SHIFT		22
+#else
+#define PGDIR_SHIFT		21
+#endif
+
+#endif
+
+#else /* CONFIG_PGTABLE_LEVELS != 2 */
+
+/*
+ * A default 3 level paging testing setup in software walked MMU
+ *   MMUv4 (8K page): <4> : <7> : <8> : <13>
+ * A default 4 level paging testing setup in software walked MMU
+ *   MMUv4 (8K page): <4> : <3> : <4> : <8> : <13>
+ */
+#define PGDIR_SHIFT		28
+#if CONFIG_PGTABLE_LEVELS > 3
+#define PUD_SHIFT		25
+#endif
+#if CONFIG_PGTABLE_LEVELS > 2
+#define PMD_SHIFT		21
+#endif
+
+#endif /* CONFIG_PGTABLE_LEVELS */
+
+#define PGDIR_SIZE		BIT(PGDIR_SHIFT)
+#define PGDIR_MASK		(~(PGDIR_SIZE - 1))
+#define PTRS_PER_PGD		BIT(32 - PGDIR_SHIFT)
+
+#if CONFIG_PGTABLE_LEVELS > 3
+#define PUD_SIZE		BIT(PUD_SHIFT)
+#define PUD_MASK		(~(PUD_SIZE - 1))
+#define PTRS_PER_PUD		BIT(PGDIR_SHIFT - PUD_SHIFT)
+#endif
+
+#if CONFIG_PGTABLE_LEVELS > 2
+#define PMD_SIZE		BIT(PMD_SHIFT)
+#define PMD_MASK		(~(PMD_SIZE - 1))
+#define PTRS_PER_PMD		BIT(PUD_SHIFT - PMD_SHIFT)
+#endif
+
+#define PTRS_PER_PTE		BIT(PMD_SHIFT - PAGE_SHIFT)
+
+#ifndef __ASSEMBLY__
+
+#if CONFIG_PGTABLE_LEVELS > 3
+#include <asm-generic/pgtable-nop4d.h>
+#elif CONFIG_PGTABLE_LEVELS > 2
+#include <asm-generic/pgtable-nopud.h>
+#else
+#include <asm-generic/pgtable-nopmd.h>
+#endif
+
+/*
+ * 1st level paging: pgd
+ */
+#define pgd_index(addr)		((addr) >> PGDIR_SHIFT)
+#define pgd_offset(mm, addr)	(((mm)->pgd) + pgd_index(addr))
+#define pgd_offset_k(addr)	pgd_offset(&init_mm, addr)
+#define pgd_ERROR(e) \
+	pr_crit("%s:%d: bad pgd %08lx.\n", __FILE__, __LINE__, pgd_val(e))
+
+#if CONFIG_PGTABLE_LEVELS > 3
+
+/* In 4 level paging, p4d_* macros work on pgd */
+#define p4d_none(x)		(!p4d_val(x))
+#define p4d_bad(x)		((p4d_val(x) & ~PAGE_MASK))
+#define p4d_present(x)		(p4d_val(x))
+#define p4d_clear(xp)		do { p4d_val(*(xp)) = 0; } while (0)
+#define p4d_pgtable(p4d)	((pud_t *)(p4d_val(p4d) & PAGE_MASK))
+#define p4d_page(p4d)		virt_to_page(p4d_pgtable(p4d))
+#define set_p4d(p4dp, p4d)	(*(p4dp) = p4d)
+
+/*
+ * 2nd level paging: pud
+ */
+#define pud_ERROR(e) \
+	pr_crit("%s:%d: bad pud %08lx.\n", __FILE__, __LINE__, pud_val(e))
+
+#endif
+
+#if CONFIG_PGTABLE_LEVELS > 2
+
+/*
+ * In 3 level paging, pud_* macros work on pgd
+ * In 4 level paging, pud_* macros work on pud
+ */
+#define pud_none(x)		(!pud_val(x))
+#define pud_bad(x)		((pud_val(x) & ~PAGE_MASK))
+#define pud_present(x)		(pud_val(x))
+#define pud_clear(xp)		do { pud_val(*(xp)) = 0; } while (0)
+#define pud_pgtable(pud)	((pmd_t *)(pud_val(pud) & PAGE_MASK))
+#define pud_page(pud)		virt_to_page(pud_pgtable(pud))
+#define set_pud(pudp, pud)	(*(pudp) = pud)
+
+/*
+ * 3rd level paging: pmd
+ */
+#define pmd_ERROR(e) \
+	pr_crit("%s:%d: bad pmd %08lx.\n", __FILE__, __LINE__, pmd_val(e))
+
+#define pmd_pfn(pmd)		((pmd_val(pmd) & PMD_MASK) >> PAGE_SHIFT)
+#define pfn_pmd(pfn,prot)	__pmd(((pfn) << PAGE_SHIFT) | pgprot_val(prot))
+#define mk_pmd(page,prot)	pfn_pmd(page_to_pfn(page),prot)
+
+#endif
+
+/*
+ * Due to the strange way generic pgtable level folding works, the pmd_* macros
+ *  - are valid even for 2 levels (which supposedly only has pgd - pte)
+ *  - behave differently for 2 vs. 3
+ * In 2  level paging        (pgd -> pte), pmd_* macros work on pgd
+ * In 3+ level paging (pgd -> pmd -> pte), pmd_* macros work on pmd
+ */
+#define pmd_none(x)		(!pmd_val(x))
+#define pmd_bad(x)		((pmd_val(x) & ~PAGE_MASK))
+#define pmd_present(x)		(pmd_val(x))
+#define pmd_clear(xp)		do { pmd_val(*(xp)) = 0; } while (0)
+#define pmd_page_vaddr(pmd)	(pmd_val(pmd) & PAGE_MASK)
+#define pmd_page(pmd)		virt_to_page(pmd_page_vaddr(pmd))
+#define set_pmd(pmdp, pmd)	(*(pmdp) = pmd)
+#define pmd_pgtable(pmd)	((pgtable_t) pmd_page_vaddr(pmd))
+
+/*
+ * 4th level paging: pte
+ */
+#define pte_ERROR(e) \
+	pr_crit("%s:%d: bad pte %08lx.\n", __FILE__, __LINE__, pte_val(e))
+
+#define pte_none(x)		(!pte_val(x))
+#define pte_present(x)		(pte_val(x) & _PAGE_PRESENT)
+#define pte_clear(mm,addr,ptep)	set_pte_at(mm, addr, ptep, __pte(0))
+#define pte_page(pte)		pfn_to_page(pte_pfn(pte))
+#define set_pte(ptep, pte)	((*(ptep)) = (pte))
+#define pte_pfn(pte)		(pte_val(pte) >> PAGE_SHIFT)
+#define pfn_pte(pfn, prot)	__pte(__pfn_to_phys(pfn) | pgprot_val(prot))
+#define mk_pte(page, prot)	pfn_pte(page_to_pfn(page), prot)
+
+#ifdef CONFIG_ISA_ARCV2
+#define pmd_leaf(x)		(pmd_val(x) & _PAGE_HW_SZ)
+#endif
+
+#endif	/* !__ASSEMBLY__ */
+
+#endif

arch/arc/include/asm/processor.h

@@ -93,7 +93,7 @@ extern unsigned int get_wchan(struct task_struct *p);
 #define VMALLOC_START	(PAGE_OFFSET - (CONFIG_ARC_KVADDR_SIZE << 20))
 
 /* 1 PGDIR_SIZE each for fixmap/pkmap, 2 PGDIR_SIZE gutter (see asm/highmem.h) */
-#define VMALLOC_SIZE	((CONFIG_ARC_KVADDR_SIZE << 20) - PGDIR_SIZE * 4)
+#define VMALLOC_SIZE	((CONFIG_ARC_KVADDR_SIZE << 20) - PMD_SIZE * 4)
 
 #define VMALLOC_END	(VMALLOC_START + VMALLOC_SIZE)
 

arch/arc/include/asm/setup.h

@@ -2,8 +2,8 @@
 /*
  * Copyright (C) 2004, 2007-2010, 2011-2012 Synopsys, Inc. (www.synopsys.com)
  */
-#ifndef __ASMARC_SETUP_H
-#define __ASMARC_SETUP_H
+#ifndef __ASM_ARC_SETUP_H
+#define __ASM_ARC_SETUP_H
 
 
 #include <linux/types.h>
@@ -34,4 +34,12 @@ long __init arc_get_mem_sz(void);
 #define IS_AVAIL2(v, s, cfg)	IS_AVAIL1(v, s), IS_AVAIL1(v, IS_USED_CFG(cfg))
 #define IS_AVAIL3(v, v2, s)	IS_AVAIL1(v, s), IS_AVAIL1(v, IS_DISABLED_RUN(v2))
 
+extern void arc_mmu_init(void);
+extern char *arc_mmu_mumbojumbo(int cpu_id, char *buf, int len);
+extern void read_decode_mmu_bcr(void);
+
+extern void arc_cache_init(void);
+extern char *arc_cache_mumbojumbo(int cpu_id, char *buf, int len);
+extern void read_decode_cache_bcr(void);
+
 #endif /* __ASMARC_SETUP_H */

arch/arc/include/asm/smp.h

@@ -105,7 +105,6 @@ static inline const char *arc_platform_smp_cpuinfo(void)
 #include <asm/spinlock.h>
 
 extern arch_spinlock_t smp_atomic_ops_lock;
-extern arch_spinlock_t smp_bitops_lock;
 
 #define atomic_ops_lock(flags)	do {		\
 	local_irq_save(flags);			\
@@ -117,24 +116,11 @@ extern arch_spinlock_t smp_bitops_lock;
 	local_irq_restore(flags);		\
 } while (0)
 
-#define bitops_lock(flags)	do {		\
-	local_irq_save(flags);			\
-	arch_spin_lock(&smp_bitops_lock);	\
-} while (0)
-
-#define bitops_unlock(flags) do {		\
-	arch_spin_unlock(&smp_bitops_lock);	\
-	local_irq_restore(flags);		\
-} while (0)
-
 #else /* !CONFIG_SMP */
 
 #define atomic_ops_lock(flags)		local_irq_save(flags)
 #define atomic_ops_unlock(flags)	local_irq_restore(flags)
 
-#define bitops_lock(flags)		local_irq_save(flags)
-#define bitops_unlock(flags)		local_irq_restore(flags)
-
 #endif /* !CONFIG_SMP */
 
 #endif	/* !CONFIG_ARC_HAS_LLSC */

arch/arc/include/asm/tlb-mmu1.h

-/* SPDX-License-Identifier: GPL-2.0-only */
-/*
- * Copyright (C) 2004, 2007-2010, 2011-2012 Synopsys, Inc. (www.synopsys.com)
- */
-
-#ifndef __ASM_TLB_MMU_V1_H__
-#define __ASM_TLB_MMU_V1_H__
-
-#include <asm/mmu.h>
-
-#if defined(__ASSEMBLY__) && (CONFIG_ARC_MMU_VER == 1)
-
-.macro TLB_WRITE_HEURISTICS
-
-#define JH_HACK1
-#undef JH_HACK2
-#undef JH_HACK3
-
-#ifdef JH_HACK3
-; Calculate set index for 2-way MMU
-; -avoiding use of GetIndex from MMU
-;   and its unpleasant LFSR pseudo-random sequence
-;
-; r1 = TLBPD0 from TLB_RELOAD above
-;
-; -- jh_ex_way_set not cleared on startup
-;    didn't want to change setup.c
-;    hence extra instruction to clean
-;
-; -- should be in cache since in same line
-;    as r0/r1 saves above
-;
-ld  r0,[jh_ex_way_sel]  ; victim pointer
-and r0,r0,1         ; clean
-xor.f   r0,r0,1         ; flip
-st  r0,[jh_ex_way_sel]  ; store back
-asr r0,r1,12        ; get set # <<1, note bit 12=R=0
-or.nz   r0,r0,1         ; set way bit
-and r0,r0,0xff      ; clean
-sr  r0,[ARC_REG_TLBINDEX]
-#endif
-
-#ifdef JH_HACK2
-; JH hack #2
-;  Faster than hack #1 in non-thrash case, but hard-coded for 2-way MMU
-;  Slower in thrash case (where it matters) because more code is executed
-;  Inefficient due to two-register paradigm of this miss handler
-;
-/* r1 = data TLBPD0 at this point */
-lr      r0,[eret]               /* instruction address */
-xor     r0,r0,r1                /* compare set #       */
-and.f   r0,r0,0x000fe000        /* 2-way MMU mask      */
-bne     88f                     /* not in same set - no need to probe */
-
-lr      r0,[eret]               /* instruction address */
-and     r0,r0,PAGE_MASK         /* VPN of instruction address */
-; lr  r1,[ARC_REG_TLBPD0]     /* Data VPN+ASID - already in r1 from TLB_RELOAD*/
-and     r1,r1,0xff              /* Data ASID */
-or      r0,r0,r1                /* Instruction address + Data ASID */
-
-lr      r1,[ARC_REG_TLBPD0]     /* save TLBPD0 containing data TLB*/
-sr      r0,[ARC_REG_TLBPD0]     /* write instruction address to TLBPD0 */
-sr      TLBProbe, [ARC_REG_TLBCOMMAND] /* Look for instruction */
-lr      r0,[ARC_REG_TLBINDEX]   /* r0 = index where instruction is, if at all */
-sr      r1,[ARC_REG_TLBPD0]     /* restore TLBPD0 */
-
-xor     r0,r0,1                 /* flip bottom bit of data index */
-b.d     89f
-sr      r0,[ARC_REG_TLBINDEX]   /* and put it back */
-88:
-sr  TLBGetIndex, [ARC_REG_TLBCOMMAND]
-89:
-#endif
-
-#ifdef JH_HACK1
-;
-; Always checks whether instruction will be kicked out by dtlb miss
-;
-mov_s   r3, r1                  ; save PD0 prepared by TLB_RELOAD in r3
-lr      r0,[eret]               /* instruction address */
-and     r0,r0,PAGE_MASK         /* VPN of instruction address */
-bmsk    r1,r3,7                 /* Data ASID, bits 7-0 */
-or_s    r0,r0,r1                /* Instruction address + Data ASID */
-
-sr      r0,[ARC_REG_TLBPD0]     /* write instruction address to TLBPD0 */
-sr      TLBProbe, [ARC_REG_TLBCOMMAND] /* Look for instruction */
-lr      r0,[ARC_REG_TLBINDEX]   /* r0 = index where instruction is, if at all */
-sr      r3,[ARC_REG_TLBPD0]     /* restore TLBPD0 */
-
-sr      TLBGetIndex, [ARC_REG_TLBCOMMAND]
-lr      r1,[ARC_REG_TLBINDEX]   /* r1 = index where MMU wants to put data */
-cmp     r0,r1                   /* if no match on indices, go around */
-xor.eq  r1,r1,1                 /* flip bottom bit of data index */
-sr      r1,[ARC_REG_TLBINDEX]   /* and put it back */
-#endif
-
-.endm
-
-#endif
-
-#endif

arch/arc/kernel/entry-arcv2.S

@@ -10,6 +10,7 @@
 #include <asm/errno.h>
 #include <asm/arcregs.h>
 #include <asm/irqflags.h>
+#include <asm/mmu.h>
 
 ; A maximum number of supported interrupts in the core interrupt controller.
 ; This number is not equal to the maximum interrupt number (256) because

arch/arc/kernel/entry.S

@@ -101,11 +101,8 @@ ENTRY(EV_MachineCheck)
 	lr  r0, [efa]
 	mov r1, sp
 
-	; hardware auto-disables MMU, re-enable it to allow kernel vaddr
-	; access for say stack unwinding of modules for crash dumps
-	lr	r3, [ARC_REG_PID]
-	or	r3, r3, MMU_ENABLE
-	sr	r3, [ARC_REG_PID]
+	; MC excpetions disable MMU
+	ARC_MMU_REENABLE r3
 
 	lsr  	r3, r2, 8
 	bmsk 	r3, r3, 7

arch/arc/kernel/intc-compact.c

@@ -142,7 +142,7 @@ IRQCHIP_DECLARE(arc_intc, "snps,arc700-intc", init_onchip_IRQ);
  *    Time hard-ISR, timer_interrupt( ) calls spin_unlock_irq several times.
  *    Here local_irq_enable( ) shd not re-enable lower priority interrupts
  * -If called from soft-ISR, it must re-enable all interrupts
- *    soft ISR are low prioity jobs which can be very slow, thus all IRQs
+ *    soft ISR are low priority jobs which can be very slow, thus all IRQs
  *    must be enabled while they run.
  *    Now hardware context wise we may still be in L2 ISR (not done rtie)
  *    still we must re-enable both L1 and L2 IRQs

arch/arc/kernel/smp.c

@@ -29,10 +29,8 @@
 
 #ifndef CONFIG_ARC_HAS_LLSC
 arch_spinlock_t smp_atomic_ops_lock = __ARCH_SPIN_LOCK_UNLOCKED;
-arch_spinlock_t smp_bitops_lock = __ARCH_SPIN_LOCK_UNLOCKED;
 
 EXPORT_SYMBOL_GPL(smp_atomic_ops_lock);
-EXPORT_SYMBOL_GPL(smp_bitops_lock);
 #endif
 
 struct plat_smp_ops  __weak plat_smp_ops;
@@ -283,7 +281,7 @@ static void ipi_send_msg_one(int cpu, enum ipi_msg_type msg)
 	/*
 	 * Call the platform specific IPI kick function, but avoid if possible:
 	 * Only do so if there's no pending msg from other concurrent sender(s).
-	 * Otherwise, recevier will see this msg as well when it takes the
+	 * Otherwise, receiver will see this msg as well when it takes the
 	 * IPI corresponding to that msg. This is true, even if it is already in
 	 * IPI handler, because !@old means it has not yet dequeued the msg(s)
 	 * so @new msg can be a free-loader

arch/arc/kernel/stacktrace.c

@@ -149,7 +149,7 @@ arc_unwind_core(struct task_struct *tsk, struct pt_regs *regs,
 #else
 	/* On ARC, only Dward based unwinder works. fp based backtracing is
 	 * not possible (-fno-omit-frame-pointer) because of the way function
-	 * prelogue is setup (callee regs saved and then fp set and not other
+	 * prologue is setup (callee regs saved and then fp set and not other
 	 * way around
 	 */
 	pr_warn_once("CONFIG_ARC_DW2_UNWIND needs to be enabled\n");

arch/arc/mm/cache.c

@@ -205,92 +205,23 @@ void read_decode_cache_bcr(void)
 #define OP_INV_IC	0x4
 
 /*
- *		I-Cache Aliasing in ARC700 VIPT caches (MMU v1-v3)
+ * Cache Flush programming model
  *
- * ARC VIPT I-cache uses vaddr to index into cache and paddr to match the tag.
- * The orig Cache Management Module "CDU" only required paddr to invalidate a
- * certain line since it sufficed as index in Non-Aliasing VIPT cache-geometry.
- * Infact for distinct V1,V2,P: all of {V1-P},{V2-P},{P-P} would end up fetching
- * the exact same line.
+ * ARC700 MMUv3 I$ and D$ are both VIPT and can potentially alias.
+ * Programming model requires both paddr and vaddr irrespecive of aliasing
+ * considerations:
+ *  - vaddr in {I,D}C_IV?L
+ *  - paddr in {I,D}C_PTAG
  *
- * However for larger Caches (way-size > page-size) - i.e. in Aliasing config,
- * paddr alone could not be used to correctly index the cache.
+ * In HS38x (MMUv4), D$ is PIPT, I$ is VIPT and can still alias.
+ * Programming model is different for aliasing vs. non-aliasing I$
+ *  - D$ / Non-aliasing I$: only paddr in {I,D}C_IV?L
+ *  - Aliasing I$: same as ARC700 above (so MMUv3 routine used for MMUv4 I$)
  *
- * ------------------
- * MMU v1/v2 (Fixed Page Size 8k)
- * ------------------
- * The solution was to provide CDU with these additonal vaddr bits. These
- * would be bits [x:13], x would depend on cache-geometry, 13 comes from
- * standard page size of 8k.
- * H/w folks chose [17:13] to be a future safe range, and moreso these 5 bits
- * of vaddr could easily be "stuffed" in the paddr as bits [4:0] since the
- * orig 5 bits of paddr were anyways ignored by CDU line ops, as they
- * represent the offset within cache-line. The adv of using this "clumsy"
- * interface for additional info was no new reg was needed in CDU programming
- * model.
- *
- * 17:13 represented the max num of bits passable, actual bits needed were
- * fewer, based on the num-of-aliases possible.
- * -for 2 alias possibility, only bit 13 needed (32K cache)
- * -for 4 alias possibility, bits 14:13 needed (64K cache)
- *
- * ------------------
- * MMU v3
- * ------------------
- * This ver of MMU supports variable page sizes (1k-16k): although Linux will
- * only support 8k (default), 16k and 4k.
- * However from hardware perspective, smaller page sizes aggravate aliasing
- * meaning more vaddr bits needed to disambiguate the cache-line-op ;
- * the existing scheme of piggybacking won't work for certain configurations.
- * Two new registers IC_PTAG and DC_PTAG inttoduced.
- * "tag" bits are provided in PTAG, index bits in existing IVIL/IVDL/FLDL regs
- */
-
-static inline
-void __cache_line_loop_v2(phys_addr_t paddr, unsigned long vaddr,
-			  unsigned long sz, const int op, const int full_page)
-{
-	unsigned int aux_cmd;
-	int num_lines;
-
-	if (op == OP_INV_IC) {
-		aux_cmd = ARC_REG_IC_IVIL;
-	} else {
-		/* d$ cmd: INV (discard or wback-n-discard) OR FLUSH (wback) */
-		aux_cmd = op & OP_INV ? ARC_REG_DC_IVDL : ARC_REG_DC_FLDL;
-	}
-
-	/* Ensure we properly floor/ceil the non-line aligned/sized requests
-	 * and have @paddr - aligned to cache line and integral @num_lines.
-	 * This however can be avoided for page sized since:
-	 *  -@paddr will be cache-line aligned already (being page aligned)
-	 *  -@sz will be integral multiple of line size (being page sized).
+ *  - If PAE40 is enabled, independent of aliasing considerations, the higher
+ *    bits needs to be written into PTAG_HI
  */
-	if (!full_page) {
-		sz += paddr & ~CACHE_LINE_MASK;
-		paddr &= CACHE_LINE_MASK;
-		vaddr &= CACHE_LINE_MASK;
-	}
 
-	num_lines = DIV_ROUND_UP(sz, L1_CACHE_BYTES);
-
-	/* MMUv2 and before: paddr contains stuffed vaddrs bits */
-	paddr |= (vaddr >> PAGE_SHIFT) & 0x1F;
-
-	while (num_lines-- > 0) {
-		write_aux_reg(aux_cmd, paddr);
-		paddr += L1_CACHE_BYTES;
-	}
-}
-
-/*
- * For ARC700 MMUv3 I-cache and D-cache flushes
- *  - ARC700 programming model requires paddr and vaddr be passed in seperate
- *    AUX registers (*_IV*L and *_PTAG respectively) irrespective of whether the
- *    caches actually alias or not.
- * -  For HS38, only the aliasing I-cache configuration uses the PTAG reg
- *    (non aliasing I-cache version doesn't; while D-cache can't possibly alias)
- */
 static inline
 void __cache_line_loop_v3(phys_addr_t paddr, unsigned long vaddr,
 			  unsigned long sz, const int op, const int full_page)
@@ -350,17 +281,6 @@ void __cache_line_loop_v3(phys_addr_t paddr, unsigned long vaddr,
 #ifndef USE_RGN_FLSH
 
 /*
- * In HS38x (MMU v4), I-cache is VIPT (can alias), D-cache is PIPT
- * Here's how cache ops are implemented
- *
- *  - D-cache: only paddr needed (in DC_IVDL/DC_FLDL)
- *  - I-cache Non Aliasing: Despite VIPT, only paddr needed (in IC_IVIL)
- *  - I-cache Aliasing: Both vaddr and paddr needed (in IC_IVIL, IC_PTAG
- *    respectively, similar to MMU v3 programming model, hence
- *    __cache_line_loop_v3() is used)
- *
- * If PAE40 is enabled, independent of aliasing considerations, the higher bits
- * needs to be written into PTAG_HI
  */
 static inline
 void __cache_line_loop_v4(phys_addr_t paddr, unsigned long vaddr,
@@ -460,11 +380,9 @@ void __cache_line_loop_v4(phys_addr_t paddr, unsigned long vaddr,
 
 #endif
 
-#if (CONFIG_ARC_MMU_VER < 3)
-#define __cache_line_loop	__cache_line_loop_v2
-#elif (CONFIG_ARC_MMU_VER == 3)
+#ifdef CONFIG_ARC_MMU_V3
 #define __cache_line_loop	__cache_line_loop_v3
-#elif (CONFIG_ARC_MMU_VER > 3)
+#else
 #define __cache_line_loop	__cache_line_loop_v4
 #endif
 
@@ -1123,7 +1041,7 @@ void clear_user_page(void *to, unsigned long u_vaddr, struct page *page)
 	clear_page(to);
 	clear_bit(PG_dc_clean, &page->flags);
 }
-
+EXPORT_SYMBOL(clear_user_page);
 
 /**********************************************************************
  * Explicit Cache flush request from user space via syscall

arch/arc/mm/fault.c

@@ -33,27 +33,33 @@ noinline static int handle_kernel_vaddr_fault(unsigned long address)
 	pud_t *pud, *pud_k;
 	pmd_t *pmd, *pmd_k;
 
-	pgd = pgd_offset_fast(current->active_mm, address);
+	pgd = pgd_offset(current->active_mm, address);
 	pgd_k = pgd_offset_k(address);
 
-	if (!pgd_present(*pgd_k))
+	if (pgd_none (*pgd_k))
 		goto bad_area;
+	if (!pgd_present(*pgd))
+		set_pgd(pgd, *pgd_k);
 
 	p4d = p4d_offset(pgd, address);
 	p4d_k = p4d_offset(pgd_k, address);
-	if (!p4d_present(*p4d_k))
+	if (p4d_none(*p4d_k))
 		goto bad_area;
+	if (!p4d_present(*p4d))
+		set_p4d(p4d, *p4d_k);
 
 	pud = pud_offset(p4d, address);
 	pud_k = pud_offset(p4d_k, address);
-	if (!pud_present(*pud_k))
+	if (pud_none(*pud_k))
 		goto bad_area;
+	if (!pud_present(*pud))
+		set_pud(pud, *pud_k);
 
 	pmd = pmd_offset(pud, address);
 	pmd_k = pmd_offset(pud_k, address);
-	if (!pmd_present(*pmd_k))
+	if (pmd_none(*pmd_k))
 		goto bad_area;
-
+	if (!pmd_present(*pmd))
 		set_pmd(pmd, *pmd_k);
 
 	/* XXX: create the TLB entry here */

arch/arc/mm/init.c

@@ -189,6 +189,11 @@ void __init mem_init(void)
 {
 	memblock_free_all();
 	highmem_init();
+
+	BUILD_BUG_ON((PTRS_PER_PGD * sizeof(pgd_t)) > PAGE_SIZE);
+	BUILD_BUG_ON((PTRS_PER_PUD * sizeof(pud_t)) > PAGE_SIZE);
+	BUILD_BUG_ON((PTRS_PER_PMD * sizeof(pmd_t)) > PAGE_SIZE);
+	BUILD_BUG_ON((PTRS_PER_PTE * sizeof(pte_t)) > PAGE_SIZE);
 }
 
 #ifdef CONFIG_HIGHMEM

arch/arc/mm/ioremap.c

@@ -39,7 +39,8 @@ void __iomem *ioremap(phys_addr_t paddr, unsigned long size)
 	if (arc_uncached_addr_space(paddr))
 		return (void __iomem *)(u32)paddr;
 
-	return ioremap_prot(paddr, size, PAGE_KERNEL_NO_CACHE);
+	return ioremap_prot(paddr, size,
+			    pgprot_val(pgprot_noncached(PAGE_KERNEL)));
 }
 EXPORT_SYMBOL(ioremap);
 

arch/arc/mm/tlbex.S

@@ -39,7 +39,6 @@
 #include <asm/arcregs.h>
 #include <asm/cache.h>
 #include <asm/processor.h>
-#include <asm/tlb-mmu1.h>
 
 #ifdef CONFIG_ISA_ARCOMPACT
 ;-----------------------------------------------------------------
@@ -94,11 +93,6 @@ ex_saved_reg1:
 	st_s  r1, [r0, 4]
 	st_s  r2, [r0, 8]
 	st_s  r3, [r0, 12]
-
-	; VERIFY if the ASID in MMU-PID Reg is same as
-	; one in Linux data structures
-
-	tlb_paranoid_check_asm
 .endm
 
 .macro TLBMISS_RESTORE_REGS
@@ -148,53 +142,16 @@ ex_saved_reg1:
 #endif
 
 ;============================================================================
-;  Troubleshooting Stuff
+;TLB Miss handling Code
 ;============================================================================
 
-; Linux keeps ASID (Address Space ID) in task->active_mm->context.asid
-; When Creating TLB Entries, instead of doing 3 dependent loads from memory,
-; we use the MMU PID Reg to get current ASID.
-; In bizzare scenrios SW and HW ASID can get out-of-sync which is trouble.
-; So we try to detect this in TLB Mis shandler
-
-.macro tlb_paranoid_check_asm
-
-#ifdef CONFIG_ARC_DBG_TLB_PARANOIA
-
-	GET_CURR_TASK_ON_CPU  r3
-	ld r0, [r3, TASK_ACT_MM]
-	ld r0, [r0, MM_CTXT+MM_CTXT_ASID]
-	breq r0, 0, 55f	; Error if no ASID allocated
-
-	lr r1, [ARC_REG_PID]
-	and r1, r1, 0xFF
-
-	and r2, r0, 0xFF	; MMU PID bits only for comparison
-	breq r1, r2, 5f
-
-55:
-	; Error if H/w and S/w ASID don't match, but NOT if in kernel mode
-	lr  r2, [erstatus]
-	bbit0 r2, STATUS_U_BIT, 5f
-
-	; We sure are in troubled waters, Flag the error, but to do so
-	; need to switch to kernel mode stack to call error routine
-	GET_TSK_STACK_BASE   r3, sp
-
-	; Call printk to shoutout aloud
-	mov r2, 1
-	j print_asid_mismatch
-
-5:	; ASIDs match so proceed normally
-	nop
-
+#ifndef PMD_SHIFT
+#define PMD_SHIFT PUD_SHIFT
 #endif
 
-.endm
-
-;============================================================================
-;TLB Miss handling Code
-;============================================================================
+#ifndef PUD_SHIFT
+#define PUD_SHIFT PGDIR_SHIFT
+#endif
 
 ;-----------------------------------------------------------------------------
 ; This macro does the page-table lookup for the faulting address.
@@ -203,7 +160,7 @@ ex_saved_reg1:
 
 	lr  r2, [efa]
 
-#ifdef ARC_USE_SCRATCH_REG
+#ifdef CONFIG_ISA_ARCV2
 	lr  r1, [ARC_REG_SCRATCH_DATA0] ; current pgd
 #else
 	GET_CURR_TASK_ON_CPU  r1
@@ -216,6 +173,24 @@ ex_saved_reg1:
 	tst	r3, r3
 	bz	do_slow_path_pf         ; if no Page Table, do page fault
 
+#if CONFIG_PGTABLE_LEVELS > 3
+	lsr     r0, r2, PUD_SHIFT	; Bits for indexing into PUD
+	and	r0, r0, (PTRS_PER_PUD - 1)
+	ld.as	r1, [r3, r0]		; PMD entry
+	tst	r1, r1
+	bz	do_slow_path_pf
+	mov	r3, r1
+#endif
+
+#if CONFIG_PGTABLE_LEVELS > 2
+	lsr     r0, r2, PMD_SHIFT	; Bits for indexing into PMD
+	and	r0, r0, (PTRS_PER_PMD - 1)
+	ld.as	r1, [r3, r0]		; PMD entry
+	tst	r1, r1
+	bz	do_slow_path_pf
+	mov	r3, r1
+#endif
+
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 	and.f	0, r3, _PAGE_HW_SZ	; Is this Huge PMD (thp)
 	add2.nz	r1, r1, r0
@@ -279,7 +254,7 @@ ex_saved_reg1:
 ; Commit the TLB entry into MMU
 
 .macro COMMIT_ENTRY_TO_MMU
-#if (CONFIG_ARC_MMU_VER < 4)
+#ifdef CONFIG_ARC_MMU_V3
 
 	/* Get free TLB slot: Set = computed from vaddr, way = random */
 	sr  TLBGetIndex, [ARC_REG_TLBCOMMAND]
@@ -375,13 +350,6 @@ ENTRY(EV_TLBMissD)
 
 	CONV_PTE_TO_TLB
 
-#if (CONFIG_ARC_MMU_VER == 1)
-	; MMU with 2 way set assoc J-TLB, needs some help in pathetic case of
-	; memcpy where 3 parties contend for 2 ways, ensuing a livelock.
-	; But only for old MMU or one with Metal Fix
-	TLB_WRITE_HEURISTICS
-#endif
-
 	COMMIT_ENTRY_TO_MMU
 	TLBMISS_RESTORE_REGS
 EV_TLBMissD_fast_ret:	; additional label for VDK OS-kit instrumentation