x86/kprobes: Fix crash when probing CS CALL instructionsgrafted

When using eBPF to probe CS CALL instructions within a function,
a crash can be triggered.

The eBPF tool probes offset 257 of the __hrtimer_run_queues()
function:

<__hrtimer_run_queues+249>:  nopl   0x0(%rax,%rax,1)
<__hrtimer_run_queues+254>:  mov    %r14,%rdi
<__hrtimer_run_queues+257>:  cs call <__x86_indirect_thunk_r12>
<__hrtimer_run_queues+263>:  mov    %eax,%r12d
<__hrtimer_run_queues+266>:  xchg   %ax,%ax
<__hrtimer_run_queues+268>:  mov    %r13,%rdi

Which triggers this crash:

  BUG: unable to handle page fault for address: 00000000000f41c9
  #PF: supervisor write access in kernel mode
  #PF: error_code(0x0002) - not-present page
  PGD 0 P4D 0
  Oops: 0002 [#1] SMP NOPTI
  CPU: 1 PID: 0 Comm: swapper/1 Kdump: loaded Tainted: P
  RIP: 0010:__hrtimer_run_queues+0x106/0x230

Note that __hrtimer_run_queues+0x106 is __hrtimer_run_queues+262, which is
at the 6th byte of the above CS CALL instruction. Since the CS CALL
instruction occupies 6 bytes, the exception occurred in the middle of that
call instruction.

The root cause is that when using eBPF tools to probe in the middle of a
function, a kprobe with INT3 is used as the underlying implementation.

During single-step emulation of the original CALL instruction,
int3_emulate_call() assumes that the probed CALL instruction is 5 bytes
long. However, the actual CS-prefixed CALL instruction occupies 6 bytes,
so it constructs an incorrect exception return address. When the CPU
returns from the kprobe handler, the next instruction to be executed is at
the address of the last byte of that CS CALL instruction. Coincidentally,
starting from that address, the CPU fetches and decodes a completely
different instruction, which ultimately triggers a kernel crash.

Fix the issue by using the actual instruction length obtained from
the instruction decoder when constructing the exception return
address, rather than relying on the hardcoded CALL_INSN_SIZE macro.

[ mingo: Refined the changelog ]

Fixes: 6256e668b7af ("x86/kprobes: Use int3 instead of debug trap for single-step")
Suggested-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
Signed-off-by: Jinke Han <jinkehan@didiglobal.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
Acked-by: Yafang Shao <laoar.shao@gmail.com>
Acked-by: Borislav Petkov <bp@alien8.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: https://patch.msgid.link/20260908073742.GA10517@didi-ThinkCentre-M920t-N000

9 files changed, 1622 insertions, 0 deletions

diff --git a/arch/parisc/lib/Makefile b/arch/parisc/lib/Makefile
new file mode 100644
index 000000000..d5975d1fb
--- /dev/null
+++ b/arch/parisc/lib/Makefile
@@ -0,0 +1,9 @@
+# SPDX-License-Identifier: GPL-2.0-only
+#
+# Makefile for parisc-specific library files
+#
+
+lib-y	:= lusercopy.o bitops.o io.o memset.o memcpy.o \
+	   ucmpdi2.o delay.o
+
+obj-y	:= iomap.o
diff --git a/arch/parisc/lib/bitops.c b/arch/parisc/lib/bitops.c
new file mode 100644
index 000000000..9df810050
--- /dev/null
+++ b/arch/parisc/lib/bitops.c
@@ -0,0 +1,75 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * bitops.c: atomic operations which got too long to be inlined all over
+ *      the place.
+ * 
+ * Copyright 1999 Philipp Rumpf (prumpf@tux.org)
+ * Copyright 2000 Grant Grundler (grundler@cup.hp.com)
+ */
+
+#include <linux/kernel.h>
+#include <linux/spinlock.h>
+#include <linux/atomic.h>
+
+#ifdef CONFIG_SMP
+arch_spinlock_t __atomic_hash[ATOMIC_HASH_SIZE] __lock_aligned = {
+	[0 ... (ATOMIC_HASH_SIZE-1)]  = __ARCH_SPIN_LOCK_UNLOCKED
+};
+#endif
+
+#ifdef CONFIG_64BIT
+unsigned long notrace __xchg64(unsigned long x, volatile unsigned long *ptr)
+{
+	unsigned long temp, flags;
+
+	_atomic_spin_lock_irqsave(ptr, flags);
+	temp = *ptr;
+	*ptr = x;
+	_atomic_spin_unlock_irqrestore(ptr, flags);
+	return temp;
+}
+#endif
+
+unsigned long notrace __xchg32(int x, volatile int *ptr)
+{
+	unsigned long flags;
+	long temp;
+
+	_atomic_spin_lock_irqsave(ptr, flags);
+	temp = (long) *ptr;	/* XXX - sign extension wanted? */
+	*ptr = x;
+	_atomic_spin_unlock_irqrestore(ptr, flags);
+	return (unsigned long)temp;
+}
+
+
+unsigned long notrace __xchg8(char x, volatile char *ptr)
+{
+	unsigned long flags;
+	long temp;
+
+	_atomic_spin_lock_irqsave(ptr, flags);
+	temp = (long) *ptr;	/* XXX - sign extension wanted? */
+	*ptr = x;
+	_atomic_spin_unlock_irqrestore(ptr, flags);
+	return (unsigned long)temp;
+}
+
+
+#define CMPXCHG(T)						\
+	T notrace __cmpxchg_##T(volatile T *ptr, T old, T new)	\
+	{							\
+		unsigned long flags;				\
+		T prev;						\
+								\
+		_atomic_spin_lock_irqsave(ptr, flags);		\
+		if ((prev = *ptr) == old)			\
+			*ptr = new;				\
+		_atomic_spin_unlock_irqrestore(ptr, flags);	\
+		return prev;					\
+	}
+
+CMPXCHG(u64)
+CMPXCHG(u32)
+CMPXCHG(u16)
+CMPXCHG(u8)
diff --git a/arch/parisc/lib/delay.c b/arch/parisc/lib/delay.c
new file mode 100644
index 000000000..66e506520
--- /dev/null
+++ b/arch/parisc/lib/delay.c
@@ -0,0 +1,72 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ *	Precise Delay Loops for parisc
+ *
+ *	based on code by:
+ *	Copyright (C) 1993 Linus Torvalds
+ *	Copyright (C) 1997 Martin Mares <mj@atrey.karlin.mff.cuni.cz>
+ *	Copyright (C) 2008 Jiri Hladky <hladky _dot_ jiri _at_ gmail _dot_ com>
+ *
+ *	parisc implementation:
+ *	Copyright (C) 2013 Helge Deller <deller@gmx.de>
+ */
+
+
+#include <linux/module.h>
+#include <linux/preempt.h>
+#include <linux/init.h>
+
+#include <asm/delay.h>
+#include <asm/special_insns.h>    /* for mfctl() */
+#include <asm/processor.h> /* for boot_cpu_data */
+
+/* CR16 based delay: */
+static void __cr16_delay(unsigned long __loops)
+{
+	/*
+	 * Note: Due to unsigned math, cr16 rollovers shouldn't be
+	 * a problem here. However, on 32 bit, we need to make sure
+	 * we don't pass in too big a value. The current default
+	 * value of MAX_UDELAY_MS should help prevent this.
+	 */
+	u32 bclock, now, loops = __loops;
+	int cpu;
+
+	preempt_disable();
+	cpu = smp_processor_id();
+	bclock = mfctl(16);
+	for (;;) {
+		now = mfctl(16);
+		if ((now - bclock) >= loops)
+			break;
+
+		/* Allow RT tasks to run */
+		preempt_enable();
+		asm volatile("	nop\n");
+		barrier();
+		preempt_disable();
+
+		/*
+		 * It is possible that we moved to another CPU, and
+		 * since CR16's are per-cpu we need to calculate
+		 * that. The delay must guarantee that we wait "at
+		 * least" the amount of time. Being moved to another
+		 * CPU could make the wait longer but we just need to
+		 * make sure we waited long enough. Rebalance the
+		 * counter for this CPU.
+		 */
+		if (unlikely(cpu != smp_processor_id())) {
+			loops -= (now - bclock);
+			cpu = smp_processor_id();
+			bclock = mfctl(16);
+		}
+	}
+	preempt_enable();
+}
+
+
+void __udelay(unsigned long usecs)
+{
+	__cr16_delay(usecs * ((unsigned long)boot_cpu_data.cpu_hz / 1000000UL));
+}
+EXPORT_SYMBOL(__udelay);
diff --git a/arch/parisc/lib/io.c b/arch/parisc/lib/io.c
new file mode 100644
index 000000000..3c7e617f5
--- /dev/null
+++ b/arch/parisc/lib/io.c
@@ -0,0 +1,381 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * arch/parisc/lib/io.c
+ *
+ * Copyright (c) Matthew Wilcox 2001 for Hewlett-Packard
+ * Copyright (c) Randolph Chung 2001 <tausq@debian.org>
+ *
+ * IO accessing functions which shouldn't be inlined because they're too big
+ */
+
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <asm/io.h>
+
+/*
+ * Read COUNT 8-bit bytes from port PORT into memory starting at
+ * SRC.
+ */
+void insb (unsigned long port, void *dst, unsigned long count)
+{
+	unsigned char *p;
+
+	p = (unsigned char *)dst;
+
+	while (((unsigned long)p) & 0x3) {
+		if (!count)
+			return;
+		count--;
+		*p = inb(port);
+		p++;
+	}
+
+	while (count >= 4) {
+		unsigned int w;
+		count -= 4;
+		w = inb(port) << 24;
+		w |= inb(port) << 16;
+		w |= inb(port) << 8;
+		w |= inb(port);
+		*(unsigned int *) p = w;
+		p += 4;
+	}
+
+	while (count) {
+		--count;
+		*p = inb(port);
+		p++;
+	}
+}
+
+
+/*
+ * Read COUNT 16-bit words from port PORT into memory starting at
+ * SRC.  SRC must be at least short aligned.  This is used by the
+ * IDE driver to read disk sectors.  Performance is important, but
+ * the interfaces seems to be slow: just using the inlined version
+ * of the inw() breaks things.
+ */
+void insw (unsigned long port, void *dst, unsigned long count)
+{
+	unsigned int l = 0, l2;
+	unsigned char *p;
+
+	p = (unsigned char *)dst;
+
+	if (!count)
+		return;
+
+	switch (((unsigned long)p) & 0x3)
+	{
+	 case 0x00:			/* Buffer 32-bit aligned */
+		while (count>=2) {
+
+			count -= 2;
+			l = cpu_to_le16(inw(port)) << 16;
+			l |= cpu_to_le16(inw(port));
+			*(unsigned int *)p = l;
+			p += 4;
+		}
+		if (count) {
+			*(unsigned short *)p = cpu_to_le16(inw(port));
+		}
+		break;
+
+	 case 0x02:			/* Buffer 16-bit aligned */
+		*(unsigned short *)p = cpu_to_le16(inw(port));
+		p += 2;
+		count--;
+		while (count>=2) {
+
+			count -= 2;
+			l = cpu_to_le16(inw(port)) << 16;
+			l |= cpu_to_le16(inw(port));
+			*(unsigned int *)p = l;
+			p += 4;
+		}
+		if (count) {
+			*(unsigned short *)p = cpu_to_le16(inw(port));
+		}
+		break;
+
+	 case 0x01:			/* Buffer 8-bit aligned */
+	 case 0x03:
+		/* I don't bother with 32bit transfers
+		 * in this case, 16bit will have to do -- DE */
+		--count;
+
+		l = cpu_to_le16(inw(port));
+		*p = l >> 8;
+		p++;
+		while (count--)
+		{
+			l2 = cpu_to_le16(inw(port));
+			*(unsigned short *)p = (l & 0xff) << 8 | (l2 >> 8);
+			p += 2;
+			l = l2;
+		}
+		*p = l & 0xff;
+		break;
+	}
+}
+
+
+
+/*
+ * Read COUNT 32-bit words from port PORT into memory starting at
+ * SRC. Now works with any alignment in SRC. Performance is important,
+ * but the interfaces seems to be slow: just using the inlined version
+ * of the inl() breaks things.
+ */
+void insl (unsigned long port, void *dst, unsigned long count)
+{
+	unsigned int l = 0, l2;
+	unsigned char *p;
+
+	p = (unsigned char *)dst;
+
+	if (!count)
+		return;
+
+	switch (((unsigned long) dst) & 0x3)
+	{
+	 case 0x00:			/* Buffer 32-bit aligned */
+		while (count--)
+		{
+			*(unsigned int *)p = cpu_to_le32(inl(port));
+			p += 4;
+		}
+		break;
+
+	 case 0x02:			/* Buffer 16-bit aligned */
+		--count;
+
+		l = cpu_to_le32(inl(port));
+		*(unsigned short *)p = l >> 16;
+		p += 2;
+
+		while (count--)
+		{
+			l2 = cpu_to_le32(inl(port));
+			*(unsigned int *)p = (l & 0xffff) << 16 | (l2 >> 16);
+			p += 4;
+			l = l2;
+		}
+		*(unsigned short *)p = l & 0xffff;
+		break;
+	 case 0x01:			/* Buffer 8-bit aligned */
+		--count;
+
+		l = cpu_to_le32(inl(port));
+		*(unsigned char *)p = l >> 24;
+		p++;
+		*(unsigned short *)p = (l >> 8) & 0xffff;
+		p += 2;
+		while (count--)
+		{
+			l2 = cpu_to_le32(inl(port));
+			*(unsigned int *)p = (l & 0xff) << 24 | (l2 >> 8);
+			p += 4;
+			l = l2;
+		}
+		*p = l & 0xff;
+		break;
+	 case 0x03:			/* Buffer 8-bit aligned */
+		--count;
+
+		l = cpu_to_le32(inl(port));
+		*p = l >> 24;
+		p++;
+		while (count--)
+		{
+			l2 = cpu_to_le32(inl(port));
+			*(unsigned int *)p = (l & 0xffffff) << 8 | l2 >> 24;
+			p += 4;
+			l = l2;
+		}
+		*(unsigned short *)p = (l >> 8) & 0xffff;
+		p += 2;
+		*p = l & 0xff;
+		break;
+	}
+}
+
+
+/*
+ * Like insb but in the opposite direction.
+ * Don't worry as much about doing aligned memory transfers:
+ * doing byte reads the "slow" way isn't nearly as slow as
+ * doing byte writes the slow way (no r-m-w cycle).
+ */
+void outsb(unsigned long port, const void * src, unsigned long count)
+{
+	const unsigned char *p;
+
+	p = (const unsigned char *)src;
+	while (count) {
+		count--;
+		outb(*p, port);
+		p++;
+	}
+}
+
+/*
+ * Like insw but in the opposite direction.  This is used by the IDE
+ * driver to write disk sectors.  Performance is important, but the
+ * interfaces seems to be slow: just using the inlined version of the
+ * outw() breaks things.
+ */
+void outsw (unsigned long port, const void *src, unsigned long count)
+{
+	unsigned int l = 0, l2;
+	const unsigned char *p;
+
+	p = (const unsigned char *)src;
+
+	if (!count)
+		return;
+
+	switch (((unsigned long)p) & 0x3)
+	{
+	 case 0x00:			/* Buffer 32-bit aligned */
+		while (count>=2) {
+			count -= 2;
+			l = *(unsigned int *)p;
+			p += 4;
+			outw(le16_to_cpu(l >> 16), port);
+			outw(le16_to_cpu(l & 0xffff), port);
+		}
+		if (count) {
+			outw(le16_to_cpu(*(unsigned short*)p), port);
+		}
+		break;
+
+	 case 0x02:			/* Buffer 16-bit aligned */
+
+		outw(le16_to_cpu(*(unsigned short*)p), port);
+		p += 2;
+		count--;
+
+		while (count>=2) {
+			count -= 2;
+			l = *(unsigned int *)p;
+			p += 4;
+			outw(le16_to_cpu(l >> 16), port);
+			outw(le16_to_cpu(l & 0xffff), port);
+		}
+		if (count) {
+			outw(le16_to_cpu(*(unsigned short *)p), port);
+		}
+		break;
+
+	 case 0x01:			/* Buffer 8-bit aligned */
+		/* I don't bother with 32bit transfers
+		 * in this case, 16bit will have to do -- DE */
+
+		l  = *p << 8;
+		p++;
+		count--;
+		while (count)
+		{
+			count--;
+			l2 = *(unsigned short *)p;
+			p += 2;
+			outw(le16_to_cpu(l | l2 >> 8), port);
+		        l = l2 << 8;
+		}
+		l2 = *(unsigned char *)p;
+		outw (le16_to_cpu(l | l2>>8), port);
+		break;
+
+	}
+}
+
+
+/*
+ * Like insl but in the opposite direction.  This is used by the IDE
+ * driver to write disk sectors.  Works with any alignment in SRC.
+ *  Performance is important, but the interfaces seems to be slow:
+ * just using the inlined version of the outl() breaks things.
+ */
+void outsl (unsigned long port, const void *src, unsigned long count)
+{
+	unsigned int l = 0, l2;
+	const unsigned char *p;
+
+	p = (const unsigned char *)src;
+
+	if (!count)
+		return;
+
+	switch (((unsigned long)p) & 0x3)
+	{
+	 case 0x00:			/* Buffer 32-bit aligned */
+		while (count--)
+		{
+			outl(le32_to_cpu(*(unsigned int *)p), port);
+			p += 4;
+		}
+		break;
+
+	 case 0x02:			/* Buffer 16-bit aligned */
+		--count;
+
+		l = *(unsigned short *)p;
+		p += 2;
+
+		while (count--)
+		{
+			l2 = *(unsigned int *)p;
+			p += 4;
+			outl (le32_to_cpu(l << 16 | l2 >> 16), port);
+			l = l2;
+		}
+		l2 = *(unsigned short *)p;
+		outl (le32_to_cpu(l << 16 | l2), port);
+		break;
+	 case 0x01:			/* Buffer 8-bit aligned */
+		--count;
+
+		l = *p << 24;
+		p++;
+		l |= *(unsigned short *)p << 8;
+		p += 2;
+
+		while (count--)
+		{
+			l2 = *(unsigned int *)p;
+			p += 4;
+			outl (le32_to_cpu(l | l2 >> 24), port);
+			l = l2 << 8;
+		}
+		l2 = *p;
+		outl (le32_to_cpu(l | l2), port);
+		break;
+	 case 0x03:			/* Buffer 8-bit aligned */
+		--count;
+
+		l = *p << 24;
+		p++;
+
+		while (count--)
+		{
+			l2 = *(unsigned int *)p;
+			p += 4;
+			outl (le32_to_cpu(l | l2 >> 8), port);
+			l = l2 << 24;
+		}
+		l2 = *(unsigned short *)p << 16;
+		p += 2;
+		l2 |= *p;
+		outl (le32_to_cpu(l | l2), port);
+		break;
+	}
+}
+
+EXPORT_SYMBOL(insb);
+EXPORT_SYMBOL(insw);
+EXPORT_SYMBOL(insl);
+EXPORT_SYMBOL(outsb);
+EXPORT_SYMBOL(outsw);
+EXPORT_SYMBOL(outsl);
diff --git a/arch/parisc/lib/iomap.c b/arch/parisc/lib/iomap.c
new file mode 100644
index 000000000..915c0c4da
--- /dev/null
+++ b/arch/parisc/lib/iomap.c
@@ -0,0 +1,551 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * iomap.c - Implement iomap interface for PA-RISC
+ * Copyright (c) 2004 Matthew Wilcox
+ */
+
+#include <linux/ioport.h>
+#include <linux/pci.h>
+#include <linux/export.h>
+#include <asm/io.h>
+
+/*
+ * The iomap space on 32-bit PA-RISC is intended to look like this:
+ * 00000000-7fffffff virtual mapped IO
+ * 80000000-8fffffff ISA/EISA port space that can't be virtually mapped
+ * 90000000-9fffffff Dino port space
+ * a0000000-afffffff Astro port space
+ * b0000000-bfffffff PAT port space
+ * c0000000-cfffffff non-swapped memory IO
+ * f0000000-ffffffff legacy IO memory pointers
+ *
+ * For the moment, here's what it looks like:
+ * 80000000-8fffffff All ISA/EISA port space
+ * f0000000-ffffffff legacy IO memory pointers
+ *
+ * On 64-bit, everything is extended, so:
+ * 8000000000000000-8fffffffffffffff All ISA/EISA port space
+ * f000000000000000-ffffffffffffffff legacy IO memory pointers
+ */
+
+/*
+ * Technically, this should be 'if (VMALLOC_START < addr < VMALLOC_END),
+ * but that's slow and we know it'll be within the first 2GB.
+ */
+#ifdef CONFIG_64BIT
+#define INDIRECT_ADDR(addr)	(((unsigned long)(addr) & 1UL<<63) != 0)
+#define ADDR_TO_REGION(addr)    (((unsigned long)addr >> 60) & 7)
+#define IOPORT_MAP_BASE		(8UL << 60)
+#else
+#define INDIRECT_ADDR(addr)     (((unsigned long)(addr) & 1UL<<31) != 0)
+#define ADDR_TO_REGION(addr)    (((unsigned long)addr >> 28) & 7)
+#define IOPORT_MAP_BASE		(8UL << 28)
+#endif
+
+struct iomap_ops {
+	unsigned int (*read8)(const void __iomem *);
+	unsigned int (*read16)(const void __iomem *);
+	unsigned int (*read16be)(const void __iomem *);
+	unsigned int (*read32)(const void __iomem *);
+	unsigned int (*read32be)(const void __iomem *);
+#ifdef CONFIG_64BIT
+	u64 (*read64)(const void __iomem *);
+	u64 (*read64be)(const void __iomem *);
+#endif
+	void (*write8)(u8, void __iomem *);
+	void (*write16)(u16, void __iomem *);
+	void (*write16be)(u16, void __iomem *);
+	void (*write32)(u32, void __iomem *);
+	void (*write32be)(u32, void __iomem *);
+#ifdef CONFIG_64BIT
+	void (*write64)(u64, void __iomem *);
+	void (*write64be)(u64, void __iomem *);
+#endif
+	void (*read8r)(const void __iomem *, void *, unsigned long);
+	void (*read16r)(const void __iomem *, void *, unsigned long);
+	void (*read32r)(const void __iomem *, void *, unsigned long);
+	void (*write8r)(void __iomem *, const void *, unsigned long);
+	void (*write16r)(void __iomem *, const void *, unsigned long);
+	void (*write32r)(void __iomem *, const void *, unsigned long);
+};
+
+/* Generic ioport ops.  To be replaced later by specific dino/elroy/wax code */
+
+#define ADDR2PORT(addr) ((unsigned long __force)(addr) & 0xffffff)
+
+static unsigned int ioport_read8(const void __iomem *addr)
+{
+	return inb(ADDR2PORT(addr));
+}
+
+static unsigned int ioport_read16(const void __iomem *addr)
+{
+	return inw(ADDR2PORT(addr));
+}
+
+static unsigned int ioport_read32(const void __iomem *addr)
+{
+	return inl(ADDR2PORT(addr));
+}
+
+static void ioport_write8(u8 datum, void __iomem *addr)
+{
+	outb(datum, ADDR2PORT(addr));
+}
+
+static void ioport_write16(u16 datum, void __iomem *addr)
+{
+	outw(datum, ADDR2PORT(addr));
+}
+
+static void ioport_write32(u32 datum, void __iomem *addr)
+{
+	outl(datum, ADDR2PORT(addr));
+}
+
+static void ioport_read8r(const void __iomem *addr, void *dst, unsigned long count)
+{
+	insb(ADDR2PORT(addr), dst, count);
+}
+
+static void ioport_read16r(const void __iomem *addr, void *dst, unsigned long count)
+{
+	insw(ADDR2PORT(addr), dst, count);
+}
+
+static void ioport_read32r(const void __iomem *addr, void *dst, unsigned long count)
+{
+	insl(ADDR2PORT(addr), dst, count);
+}
+
+static void ioport_write8r(void __iomem *addr, const void *s, unsigned long n)
+{
+	outsb(ADDR2PORT(addr), s, n);
+}
+
+static void ioport_write16r(void __iomem *addr, const void *s, unsigned long n)
+{
+	outsw(ADDR2PORT(addr), s, n);
+}
+
+static void ioport_write32r(void __iomem *addr, const void *s, unsigned long n)
+{
+	outsl(ADDR2PORT(addr), s, n);
+}
+
+static const struct iomap_ops ioport_ops = {
+	.read8 = ioport_read8,
+	.read16 = ioport_read16,
+	.read16be = ioport_read16,
+	.read32 = ioport_read32,
+	.read32be = ioport_read32,
+	.write8 = ioport_write8,
+	.write16 = ioport_write16,
+	.write16be = ioport_write16,
+	.write32 = ioport_write32,
+	.write32be = ioport_write32,
+	.read8r = ioport_read8r,
+	.read16r = ioport_read16r,
+	.read32r = ioport_read32r,
+	.write8r = ioport_write8r,
+	.write16r = ioport_write16r,
+	.write32r = ioport_write32r,
+};
+
+/* Legacy I/O memory ops */
+
+static unsigned int iomem_read8(const void __iomem *addr)
+{
+	return readb(addr);
+}
+
+static unsigned int iomem_read16(const void __iomem *addr)
+{
+	return readw(addr);
+}
+
+static unsigned int iomem_read16be(const void __iomem *addr)
+{
+	return __raw_readw(addr);
+}
+
+static unsigned int iomem_read32(const void __iomem *addr)
+{
+	return readl(addr);
+}
+
+static unsigned int iomem_read32be(const void __iomem *addr)
+{
+	return __raw_readl(addr);
+}
+
+#ifdef CONFIG_64BIT
+static u64 iomem_read64(const void __iomem *addr)
+{
+	return readq(addr);
+}
+
+static u64 iomem_read64be(const void __iomem *addr)
+{
+	return __raw_readq(addr);
+}
+#endif
+
+static void iomem_write8(u8 datum, void __iomem *addr)
+{
+	writeb(datum, addr);
+}
+
+static void iomem_write16(u16 datum, void __iomem *addr)
+{
+	writew(datum, addr);
+}
+
+static void iomem_write16be(u16 datum, void __iomem *addr)
+{
+	__raw_writew(datum, addr);
+}
+
+static void iomem_write32(u32 datum, void __iomem *addr)
+{
+	writel(datum, addr);
+}
+
+static void iomem_write32be(u32 datum, void __iomem *addr)
+{
+	__raw_writel(datum, addr);
+}
+
+#ifdef CONFIG_64BIT
+static void iomem_write64(u64 datum, void __iomem *addr)
+{
+	writeq(datum, addr);
+}
+
+static void iomem_write64be(u64 datum, void __iomem *addr)
+{
+	__raw_writeq(datum, addr);
+}
+#endif
+
+static void iomem_read8r(const void __iomem *addr, void *dst, unsigned long count)
+{
+	while (count--) {
+		*(u8 *)dst = __raw_readb(addr);
+		dst++;
+	}
+}
+
+static void iomem_read16r(const void __iomem *addr, void *dst, unsigned long count)
+{
+	while (count--) {
+		*(u16 *)dst = __raw_readw(addr);
+		dst += 2;
+	}
+}
+
+static void iomem_read32r(const void __iomem *addr, void *dst, unsigned long count)
+{
+	while (count--) {
+		*(u32 *)dst = __raw_readl(addr);
+		dst += 4;
+	}
+}
+
+static void iomem_write8r(void __iomem *addr, const void *s, unsigned long n)
+{
+	while (n--) {
+		__raw_writeb(*(u8 *)s, addr);
+		s++;
+	}
+}
+
+static void iomem_write16r(void __iomem *addr, const void *s, unsigned long n)
+{
+	while (n--) {
+		__raw_writew(*(u16 *)s, addr);
+		s += 2;
+	}
+}
+
+static void iomem_write32r(void __iomem *addr, const void *s, unsigned long n)
+{
+	while (n--) {
+		__raw_writel(*(u32 *)s, addr);
+		s += 4;
+	}
+}
+
+static const struct iomap_ops iomem_ops = {
+	.read8 = iomem_read8,
+	.read16 = iomem_read16,
+	.read16be = iomem_read16be,
+	.read32 = iomem_read32,
+	.read32be = iomem_read32be,
+#ifdef CONFIG_64BIT
+	.read64 = iomem_read64,
+	.read64be = iomem_read64be,
+#endif
+	.write8 = iomem_write8,
+	.write16 = iomem_write16,
+	.write16be = iomem_write16be,
+	.write32 = iomem_write32,
+	.write32be = iomem_write32be,
+#ifdef CONFIG_64BIT
+	.write64 = iomem_write64,
+	.write64be = iomem_write64be,
+#endif
+	.read8r = iomem_read8r,
+	.read16r = iomem_read16r,
+	.read32r = iomem_read32r,
+	.write8r = iomem_write8r,
+	.write16r = iomem_write16r,
+	.write32r = iomem_write32r,
+};
+
+static const struct iomap_ops *iomap_ops[8] = {
+	[0] = &ioport_ops,
+	[7] = &iomem_ops
+};
+
+
+unsigned int ioread8(const void __iomem *addr)
+{
+	if (unlikely(INDIRECT_ADDR(addr)))
+		return iomap_ops[ADDR_TO_REGION(addr)]->read8(addr);
+	return *((u8 *)addr);
+}
+
+unsigned int ioread16(const void __iomem *addr)
+{
+	if (unlikely(INDIRECT_ADDR(addr)))
+		return iomap_ops[ADDR_TO_REGION(addr)]->read16(addr);
+	return le16_to_cpup((u16 *)addr);
+}
+
+unsigned int ioread16be(const void __iomem *addr)
+{
+	if (unlikely(INDIRECT_ADDR(addr)))
+		return iomap_ops[ADDR_TO_REGION(addr)]->read16be(addr);
+	return *((u16 *)addr);
+}
+
+unsigned int ioread32(const void __iomem *addr)
+{
+	if (unlikely(INDIRECT_ADDR(addr)))
+		return iomap_ops[ADDR_TO_REGION(addr)]->read32(addr);
+	return le32_to_cpup((u32 *)addr);
+}
+
+unsigned int ioread32be(const void __iomem *addr)
+{
+	if (unlikely(INDIRECT_ADDR(addr)))
+		return iomap_ops[ADDR_TO_REGION(addr)]->read32be(addr);
+	return *((u32 *)addr);
+}
+
+#ifdef CONFIG_64BIT
+u64 ioread64(const void __iomem *addr)
+{
+	if (unlikely(INDIRECT_ADDR(addr)))
+		return iomap_ops[ADDR_TO_REGION(addr)]->read64(addr);
+	return le64_to_cpup((u64 *)addr);
+}
+
+u64 ioread64be(const void __iomem *addr)
+{
+	if (unlikely(INDIRECT_ADDR(addr)))
+		return iomap_ops[ADDR_TO_REGION(addr)]->read64be(addr);
+	return *((u64 *)addr);
+}
+#endif
+
+void iowrite8(u8 datum, void __iomem *addr)
+{
+	if (unlikely(INDIRECT_ADDR(addr))) {
+		iomap_ops[ADDR_TO_REGION(addr)]->write8(datum, addr);
+	} else {
+		*((u8 *)addr) = datum;
+	}
+}
+
+void iowrite16(u16 datum, void __iomem *addr)
+{
+	if (unlikely(INDIRECT_ADDR(addr))) {
+		iomap_ops[ADDR_TO_REGION(addr)]->write16(datum, addr);
+	} else {
+		*((u16 *)addr) = cpu_to_le16(datum);
+	}
+}
+
+void iowrite16be(u16 datum, void __iomem *addr)
+{
+	if (unlikely(INDIRECT_ADDR(addr))) {
+		iomap_ops[ADDR_TO_REGION(addr)]->write16be(datum, addr);
+	} else {
+		*((u16 *)addr) = datum;
+	}
+}
+
+void iowrite32(u32 datum, void __iomem *addr)
+{
+	if (unlikely(INDIRECT_ADDR(addr))) {
+		iomap_ops[ADDR_TO_REGION(addr)]->write32(datum, addr);
+	} else {
+		*((u32 *)addr) = cpu_to_le32(datum);
+	}
+}
+
+void iowrite32be(u32 datum, void __iomem *addr)
+{
+	if (unlikely(INDIRECT_ADDR(addr))) {
+		iomap_ops[ADDR_TO_REGION(addr)]->write32be(datum, addr);
+	} else {
+		*((u32 *)addr) = datum;
+	}
+}
+
+#ifdef CONFIG_64BIT
+void iowrite64(u64 datum, void __iomem *addr)
+{
+	if (unlikely(INDIRECT_ADDR(addr))) {
+		iomap_ops[ADDR_TO_REGION(addr)]->write64(datum, addr);
+	} else {
+		*((u64 *)addr) = cpu_to_le64(datum);
+	}
+}
+
+void iowrite64be(u64 datum, void __iomem *addr)
+{
+	if (unlikely(INDIRECT_ADDR(addr))) {
+		iomap_ops[ADDR_TO_REGION(addr)]->write64be(datum, addr);
+	} else {
+		*((u64 *)addr) = datum;
+	}
+}
+#endif
+
+/* Repeating interfaces */
+
+void ioread8_rep(const void __iomem *addr, void *dst, unsigned long count)
+{
+	if (unlikely(INDIRECT_ADDR(addr))) {
+		iomap_ops[ADDR_TO_REGION(addr)]->read8r(addr, dst, count);
+	} else {
+		while (count--) {
+			*(u8 *)dst = *(u8 *)addr;
+			dst++;
+		}
+	}
+}
+
+void ioread16_rep(const void __iomem *addr, void *dst, unsigned long count)
+{
+	if (unlikely(INDIRECT_ADDR(addr))) {
+		iomap_ops[ADDR_TO_REGION(addr)]->read16r(addr, dst, count);
+	} else {
+		while (count--) {
+			*(u16 *)dst = *(u16 *)addr;
+			dst += 2;
+		}
+	}
+}
+
+void ioread32_rep(const void __iomem *addr, void *dst, unsigned long count)
+{
+	if (unlikely(INDIRECT_ADDR(addr))) {
+		iomap_ops[ADDR_TO_REGION(addr)]->read32r(addr, dst, count);
+	} else {
+		while (count--) {
+			*(u32 *)dst = *(u32 *)addr;
+			dst += 4;
+		}
+	}
+}
+
+void iowrite8_rep(void __iomem *addr, const void *src, unsigned long count)
+{
+	if (unlikely(INDIRECT_ADDR(addr))) {
+		iomap_ops[ADDR_TO_REGION(addr)]->write8r(addr, src, count);
+	} else {
+		while (count--) {
+			*(u8 *)addr = *(u8 *)src;
+			src++;
+		}
+	}
+}
+
+void iowrite16_rep(void __iomem *addr, const void *src, unsigned long count)
+{
+	if (unlikely(INDIRECT_ADDR(addr))) {
+		iomap_ops[ADDR_TO_REGION(addr)]->write16r(addr, src, count);
+	} else {
+		while (count--) {
+			*(u16 *)addr = *(u16 *)src;
+			src += 2;
+		}
+	}
+}
+
+void iowrite32_rep(void __iomem *addr, const void *src, unsigned long count)
+{
+	if (unlikely(INDIRECT_ADDR(addr))) {
+		iomap_ops[ADDR_TO_REGION(addr)]->write32r(addr, src, count);
+	} else {
+		while (count--) {
+			*(u32 *)addr = *(u32 *)src;
+			src += 4;
+		}
+	}
+}
+
+/* Mapping interfaces */
+
+void __iomem *ioport_map(unsigned long port, unsigned int nr)
+{
+	return (void __iomem *)(IOPORT_MAP_BASE | port);
+}
+
+void ioport_unmap(void __iomem *addr)
+{
+	if (!INDIRECT_ADDR(addr)) {
+		iounmap(addr);
+	}
+}
+
+#ifdef CONFIG_PCI
+void pci_iounmap(struct pci_dev *dev, void __iomem * addr)
+{
+	if (!INDIRECT_ADDR(addr)) {
+		iounmap(addr);
+	}
+}
+EXPORT_SYMBOL(pci_iounmap);
+#endif
+
+EXPORT_SYMBOL(ioread8);
+EXPORT_SYMBOL(ioread16);
+EXPORT_SYMBOL(ioread16be);
+EXPORT_SYMBOL(ioread32);
+EXPORT_SYMBOL(ioread32be);
+#ifdef CONFIG_64BIT
+EXPORT_SYMBOL(ioread64);
+EXPORT_SYMBOL(ioread64be);
+#endif
+EXPORT_SYMBOL(iowrite8);
+EXPORT_SYMBOL(iowrite16);
+EXPORT_SYMBOL(iowrite16be);
+EXPORT_SYMBOL(iowrite32);
+EXPORT_SYMBOL(iowrite32be);
+#ifdef CONFIG_64BIT
+EXPORT_SYMBOL(iowrite64);
+EXPORT_SYMBOL(iowrite64be);
+#endif
+EXPORT_SYMBOL(ioread8_rep);
+EXPORT_SYMBOL(ioread16_rep);
+EXPORT_SYMBOL(ioread32_rep);
+EXPORT_SYMBOL(iowrite8_rep);
+EXPORT_SYMBOL(iowrite16_rep);
+EXPORT_SYMBOL(iowrite32_rep);
+EXPORT_SYMBOL(ioport_map);
+EXPORT_SYMBOL(ioport_unmap);
diff --git a/arch/parisc/lib/lusercopy.S b/arch/parisc/lib/lusercopy.S
new file mode 100644
index 000000000..b428d29e4
--- /dev/null
+++ b/arch/parisc/lib/lusercopy.S
@@ -0,0 +1,362 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ *    User Space Access Routines
+ *
+ *    Copyright (C) 2000-2002 Hewlett-Packard (John Marvin)
+ *    Copyright (C) 2000 Richard Hirst <rhirst with parisc-linux.org>
+ *    Copyright (C) 2001 Matthieu Delahaye <delahaym at esiee.fr>
+ *    Copyright (C) 2003 Randolph Chung <tausq with parisc-linux.org>
+ *    Copyright (C) 2017 Helge Deller <deller@gmx.de>
+ *    Copyright (C) 2017 John David Anglin <dave.anglin@bell.net>
+ */
+
+/*
+ * These routines still have plenty of room for optimization
+ * (word & doubleword load/store, dual issue, store hints, etc.).
+ */
+
+/*
+ * The following routines assume that space register 3 (sr3) contains
+ * the space id associated with the current users address space.
+ */
+
+
+	.text
+	
+#include <asm/assembly.h>
+#include <asm/errno.h>
+#include <linux/linkage.h>
+
+	/*
+	 * unsigned long lclear_user(void *to, unsigned long n)
+	 *
+	 * Returns 0 for success.
+	 * otherwise, returns number of bytes not transferred.
+	 */
+
+ENTRY_CFI(lclear_user)
+	comib,=,n   0,%r25,$lclu_done
+$lclu_loop:
+	addib,<>    -1,%r25,$lclu_loop
+1:	stbs,ma     %r0,1(%sr3,%r26)
+
+$lclu_done:
+	bv          %r0(%r2)
+	copy        %r25,%r28
+
+2:	b           $lclu_done
+	ldo         1(%r25),%r25
+
+	ASM_EXCEPTIONTABLE_ENTRY(1b,2b)
+ENDPROC_CFI(lclear_user)
+
+
+/*
+ * unsigned long pa_memcpy(void *dstp, const void *srcp, unsigned long len)
+ *
+ * Inputs:
+ * - sr1 already contains space of source region
+ * - sr2 already contains space of destination region
+ *
+ * Returns:
+ * - number of bytes that could not be copied.
+ *   On success, this will be zero.
+ *
+ * This code is based on a C-implementation of a copy routine written by
+ * Randolph Chung, which in turn was derived from the glibc.
+ *
+ * Several strategies are tried to try to get the best performance for various
+ * conditions. In the optimal case, we copy by loops that copy 32- or 16-bytes
+ * at a time using general registers.  Unaligned copies are handled either by
+ * aligning the destination and then using shift-and-write method, or in a few
+ * cases by falling back to a byte-at-a-time copy.
+ *
+ * Testing with various alignments and buffer sizes shows that this code is
+ * often >10x faster than a simple byte-at-a-time copy, even for strangely
+ * aligned operands. It is interesting to note that the glibc version of memcpy
+ * (written in C) is actually quite fast already. This routine is able to beat
+ * it by 30-40% for aligned copies because of the loop unrolling, but in some
+ * cases the glibc version is still slightly faster. This lends more
+ * credibility that gcc can generate very good code as long as we are careful.
+ *
+ * Possible optimizations:
+ * - add cache prefetching
+ * - try not to use the post-increment address modifiers; they may create
+ *   additional interlocks. Assumption is that those were only efficient on old
+ *   machines (pre PA8000 processors)
+ */
+
+	dst = arg0
+	src = arg1
+	len = arg2
+	end = arg3
+	t1  = r19
+	t2  = r20
+	t3  = r21
+	t4  = r22
+	srcspc = sr1
+	dstspc = sr2
+
+	t0 = r1
+	a1 = t1
+	a2 = t2
+	a3 = t3
+	a0 = t4
+
+	save_src = ret0
+	save_dst = ret1
+	save_len = r31
+
+ENTRY_CFI(pa_memcpy)
+	/* Last destination address */
+	add	dst,len,end
+
+	/* short copy with less than 16 bytes? */
+	cmpib,COND(>>=),n 15,len,.Lbyte_loop
+
+	/* same alignment? */
+	xor	src,dst,t0
+	extru	t0,31,2,t1
+	cmpib,<>,n  0,t1,.Lunaligned_copy
+
+#ifdef CONFIG_64BIT
+	/* only do 64-bit copies if we can get aligned. */
+	extru	t0,31,3,t1
+	cmpib,<>,n  0,t1,.Lalign_loop32
+
+	/* loop until we are 64-bit aligned */
+.Lalign_loop64:
+	extru	dst,31,3,t1
+	cmpib,=,n	0,t1,.Lcopy_loop_16_start
+20:	ldb,ma	1(srcspc,src),t1
+21:	stb,ma	t1,1(dstspc,dst)
+	b	.Lalign_loop64
+	ldo	-1(len),len
+
+	ASM_EXCEPTIONTABLE_ENTRY(20b,.Lcopy_done)
+	ASM_EXCEPTIONTABLE_ENTRY(21b,.Lcopy_done)
+
+.Lcopy_loop_16_start:
+	ldi	31,t0
+.Lcopy_loop_16:
+	cmpb,COND(>>=),n t0,len,.Lword_loop
+
+10:	ldd	0(srcspc,src),t1
+11:	ldd	8(srcspc,src),t2
+	ldo	16(src),src
+12:	std,ma	t1,8(dstspc,dst)
+13:	std,ma	t2,8(dstspc,dst)
+14:	ldd	0(srcspc,src),t1
+15:	ldd	8(srcspc,src),t2
+	ldo	16(src),src
+16:	std,ma	t1,8(dstspc,dst)
+17:	std,ma	t2,8(dstspc,dst)
+
+	ASM_EXCEPTIONTABLE_ENTRY(10b,.Lcopy_done)
+	ASM_EXCEPTIONTABLE_ENTRY(11b,.Lcopy16_fault)
+	ASM_EXCEPTIONTABLE_ENTRY(12b,.Lcopy_done)
+	ASM_EXCEPTIONTABLE_ENTRY(13b,.Lcopy_done)
+	ASM_EXCEPTIONTABLE_ENTRY(14b,.Lcopy_done)
+	ASM_EXCEPTIONTABLE_ENTRY(15b,.Lcopy16_fault)
+	ASM_EXCEPTIONTABLE_ENTRY(16b,.Lcopy_done)
+	ASM_EXCEPTIONTABLE_ENTRY(17b,.Lcopy_done)
+
+	b	.Lcopy_loop_16
+	ldo	-32(len),len
+
+.Lword_loop:
+	cmpib,COND(>>=),n 3,len,.Lbyte_loop
+20:	ldw,ma	4(srcspc,src),t1
+21:	stw,ma	t1,4(dstspc,dst)
+	b	.Lword_loop
+	ldo	-4(len),len
+
+	ASM_EXCEPTIONTABLE_ENTRY(20b,.Lcopy_done)
+	ASM_EXCEPTIONTABLE_ENTRY(21b,.Lcopy_done)
+
+#endif /* CONFIG_64BIT */
+
+	/* loop until we are 32-bit aligned */
+.Lalign_loop32:
+	extru	dst,31,2,t1
+	cmpib,=,n	0,t1,.Lcopy_loop_8
+20:	ldb,ma	1(srcspc,src),t1
+21:	stb,ma	t1,1(dstspc,dst)
+	b	.Lalign_loop32
+	ldo	-1(len),len
+
+	ASM_EXCEPTIONTABLE_ENTRY(20b,.Lcopy_done)
+	ASM_EXCEPTIONTABLE_ENTRY(21b,.Lcopy_done)
+
+
+.Lcopy_loop_8:
+	cmpib,COND(>>=),n 15,len,.Lbyte_loop
+
+10:	ldw	0(srcspc,src),t1
+11:	ldw	4(srcspc,src),t2
+12:	stw,ma	t1,4(dstspc,dst)
+13:	stw,ma	t2,4(dstspc,dst)
+14:	ldw	8(srcspc,src),t1
+15:	ldw	12(srcspc,src),t2
+	ldo	16(src),src
+16:	stw,ma	t1,4(dstspc,dst)
+17:	stw,ma	t2,4(dstspc,dst)
+
+	ASM_EXCEPTIONTABLE_ENTRY(10b,.Lcopy_done)
+	ASM_EXCEPTIONTABLE_ENTRY(11b,.Lcopy8_fault)
+	ASM_EXCEPTIONTABLE_ENTRY(12b,.Lcopy_done)
+	ASM_EXCEPTIONTABLE_ENTRY(13b,.Lcopy_done)
+	ASM_EXCEPTIONTABLE_ENTRY(14b,.Lcopy_done)
+	ASM_EXCEPTIONTABLE_ENTRY(15b,.Lcopy8_fault)
+	ASM_EXCEPTIONTABLE_ENTRY(16b,.Lcopy_done)
+	ASM_EXCEPTIONTABLE_ENTRY(17b,.Lcopy_done)
+
+	b	.Lcopy_loop_8
+	ldo	-16(len),len
+
+.Lbyte_loop:
+	cmpclr,COND(<>) len,%r0,%r0
+	b,n	.Lcopy_done
+20:	ldb	0(srcspc,src),t1
+	ldo	1(src),src
+21:	stb,ma	t1,1(dstspc,dst)
+	b	.Lbyte_loop
+	ldo	-1(len),len
+
+	ASM_EXCEPTIONTABLE_ENTRY(20b,.Lcopy_done)
+	ASM_EXCEPTIONTABLE_ENTRY(21b,.Lcopy_done)
+
+.Lcopy_done:
+	bv	%r0(%r2)
+	sub	end,dst,ret0
+
+
+	/* src and dst are not aligned the same way. */
+	/* need to go the hard way */
+.Lunaligned_copy:
+	/* align until dst is 32bit-word-aligned */
+	extru	dst,31,2,t1
+	cmpib,=,n	0,t1,.Lcopy_dstaligned
+20:	ldb	0(srcspc,src),t1
+	ldo	1(src),src
+21:	stb,ma	t1,1(dstspc,dst)
+	b	.Lunaligned_copy
+	ldo	-1(len),len
+
+	ASM_EXCEPTIONTABLE_ENTRY(20b,.Lcopy_done)
+	ASM_EXCEPTIONTABLE_ENTRY(21b,.Lcopy_done)
+
+.Lcopy_dstaligned:
+
+	/* store src, dst and len in safe place */
+	copy	src,save_src
+	copy	dst,save_dst
+	copy	len,save_len
+
+	/* len now needs give number of words to copy */
+	SHRREG	len,2,len
+
+	/*
+	 * Copy from a not-aligned src to an aligned dst using shifts.
+	 * Handles 4 words per loop.
+	 */
+
+	depw,z src,28,2,t0
+	subi 32,t0,t0
+	mtsar t0
+	extru len,31,2,t0
+	cmpib,= 2,t0,.Lcase2
+	/* Make src aligned by rounding it down.  */
+	depi 0,31,2,src
+
+	cmpiclr,<> 3,t0,%r0
+	b,n .Lcase3
+	cmpiclr,<> 1,t0,%r0
+	b,n .Lcase1
+.Lcase0:
+	cmpb,COND(=) %r0,len,.Lcda_finish
+	nop
+
+1:	ldw,ma 4(srcspc,src), a3
+	ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcda_rdfault)
+1:	ldw,ma 4(srcspc,src), a0
+	ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcda_rdfault)
+	b,n .Ldo3
+.Lcase1:
+1:	ldw,ma 4(srcspc,src), a2
+	ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcda_rdfault)
+1:	ldw,ma 4(srcspc,src), a3
+	ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcda_rdfault)
+	ldo -1(len),len
+	cmpb,COND(=),n %r0,len,.Ldo0
+.Ldo4:
+1:	ldw,ma 4(srcspc,src), a0
+	ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcda_rdfault)
+	shrpw a2, a3, %sar, t0
+1:	stw,ma t0, 4(dstspc,dst)
+	ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcopy_done)
+.Ldo3:
+1:	ldw,ma 4(srcspc,src), a1
+	ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcda_rdfault)
+	shrpw a3, a0, %sar, t0
+1:	stw,ma t0, 4(dstspc,dst)
+	ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcopy_done)
+.Ldo2:
+1:	ldw,ma 4(srcspc,src), a2
+	ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcda_rdfault)
+	shrpw a0, a1, %sar, t0
+1:	stw,ma t0, 4(dstspc,dst)
+	ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcopy_done)
+.Ldo1:
+1:	ldw,ma 4(srcspc,src), a3
+	ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcda_rdfault)
+	shrpw a1, a2, %sar, t0
+1:	stw,ma t0, 4(dstspc,dst)
+	ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcopy_done)
+	ldo -4(len),len
+	cmpb,COND(<>) %r0,len,.Ldo4
+	nop
+.Ldo0:
+	shrpw a2, a3, %sar, t0
+1:	stw,ma t0, 4(dstspc,dst)
+	ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcopy_done)
+
+.Lcda_rdfault:
+.Lcda_finish:
+	/* calculate new src, dst and len and jump to byte-copy loop */
+	sub	dst,save_dst,t0
+	add	save_src,t0,src
+	b	.Lbyte_loop
+	sub	save_len,t0,len
+
+.Lcase3:
+1:	ldw,ma 4(srcspc,src), a0
+	ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcda_rdfault)
+1:	ldw,ma 4(srcspc,src), a1
+	ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcda_rdfault)
+	b .Ldo2
+	ldo 1(len),len
+.Lcase2:
+1:	ldw,ma 4(srcspc,src), a1
+	ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcda_rdfault)
+1:	ldw,ma 4(srcspc,src), a2
+	ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcda_rdfault)
+	b .Ldo1
+	ldo 2(len),len
+
+
+	/* fault exception fixup handlers: */
+#ifdef CONFIG_64BIT
+.Lcopy16_fault:
+	b	.Lcopy_done
+10:	std,ma	t1,8(dstspc,dst)
+	ASM_EXCEPTIONTABLE_ENTRY(10b,.Lcopy_done)
+#endif
+
+.Lcopy8_fault:
+	b	.Lcopy_done
+10:	stw,ma	t1,4(dstspc,dst)
+	ASM_EXCEPTIONTABLE_ENTRY(10b,.Lcopy_done)
+ENDPROC_CFI(pa_memcpy)
+
+	.end
diff --git a/arch/parisc/lib/memcpy.c b/arch/parisc/lib/memcpy.c
new file mode 100644
index 000000000..03165c82d
--- /dev/null
+++ b/arch/parisc/lib/memcpy.c
@@ -0,0 +1,73 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ *    Optimized memory copy routines.
+ *
+ *    Copyright (C) 2004 Randolph Chung <tausq@debian.org>
+ *    Copyright (C) 2013-2017 Helge Deller <deller@gmx.de>
+ *
+ *    Portions derived from the GNU C Library
+ *    Copyright (C) 1991, 1997, 2003 Free Software Foundation, Inc.
+ */
+
+#include <linux/module.h>
+#include <linux/compiler.h>
+#include <linux/uaccess.h>
+#include <linux/mm.h>
+
+#define get_user_space()	mfsp(SR_USER)
+#define get_kernel_space()	SR_KERNEL
+
+/* Returns 0 for success, otherwise, returns number of bytes not transferred. */
+extern unsigned long pa_memcpy(void *dst, const void *src,
+				unsigned long len);
+
+unsigned long raw_copy_to_user(void __user *dst, const void *src,
+			       unsigned long len)
+{
+	mtsp(get_kernel_space(), SR_TEMP1);
+	mtsp(get_user_space(), SR_TEMP2);
+	return pa_memcpy((void __force *)dst, src, len);
+}
+EXPORT_SYMBOL(raw_copy_to_user);
+
+unsigned long raw_copy_from_user(void *dst, const void __user *src,
+			       unsigned long len)
+{
+	unsigned long start = (unsigned long) src;
+	unsigned long end = start + len;
+	unsigned long newlen = len;
+
+	mtsp(get_user_space(), SR_TEMP1);
+	mtsp(get_kernel_space(), SR_TEMP2);
+
+	/* Check region is user accessible */
+	while (start < end) {
+		if (!prober_user(SR_TEMP1, start)) {
+			newlen = (start - (unsigned long) src);
+			break;
+		}
+		start += PAGE_SIZE;
+		/* align to page boundry which may have different permission */
+		start = PAGE_ALIGN_DOWN(start);
+	}
+	return len - newlen + pa_memcpy(dst, (void __force *)src, newlen);
+}
+EXPORT_SYMBOL(raw_copy_from_user);
+
+void * memcpy(void * dst,const void *src, size_t count)
+{
+	mtsp(get_kernel_space(), SR_TEMP1);
+	mtsp(get_kernel_space(), SR_TEMP2);
+	pa_memcpy(dst, src, count);
+	return dst;
+}
+
+EXPORT_SYMBOL(memcpy);
+
+bool copy_from_kernel_nofault_allowed(const void *unsafe_src, size_t size)
+{
+	if ((unsigned long)unsafe_src < PAGE_SIZE)
+		return false;
+	/* check for I/O space F_EXTEND(0xfff00000) access as well? */
+	return true;
+}
diff --git a/arch/parisc/lib/memset.c b/arch/parisc/lib/memset.c
new file mode 100644
index 000000000..133e48098
--- /dev/null
+++ b/arch/parisc/lib/memset.c
@@ -0,0 +1,72 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+#include <linux/types.h>
+#include <asm/string.h>
+
+#define OPSIZ (BITS_PER_LONG/8)
+typedef unsigned long op_t;
+
+void *
+memset (void *dstpp, int sc, size_t len)
+{
+  unsigned int c = sc;
+  long int dstp = (long int) dstpp;
+
+  if (len >= 8)
+    {
+      size_t xlen;
+      op_t cccc;
+
+      cccc = (unsigned char) c;
+      cccc |= cccc << 8;
+      cccc |= cccc << 16;
+      if (OPSIZ > 4)
+	/* Do the shift in two steps to avoid warning if long has 32 bits.  */
+	cccc |= (cccc << 16) << 16;
+
+      /* There are at least some bytes to set.
+	 No need to test for LEN == 0 in this alignment loop.  */
+      while (dstp % OPSIZ != 0)
+	{
+	  ((unsigned char *) dstp)[0] = c;
+	  dstp += 1;
+	  len -= 1;
+	}
+
+      /* Write 8 `op_t' per iteration until less than 8 `op_t' remain.  */
+      xlen = len / (OPSIZ * 8);
+      while (xlen > 0)
+	{
+	  ((op_t *) dstp)[0] = cccc;
+	  ((op_t *) dstp)[1] = cccc;
+	  ((op_t *) dstp)[2] = cccc;
+	  ((op_t *) dstp)[3] = cccc;
+	  ((op_t *) dstp)[4] = cccc;
+	  ((op_t *) dstp)[5] = cccc;
+	  ((op_t *) dstp)[6] = cccc;
+	  ((op_t *) dstp)[7] = cccc;
+	  dstp += 8 * OPSIZ;
+	  xlen -= 1;
+	}
+      len %= OPSIZ * 8;
+
+      /* Write 1 `op_t' per iteration until less than OPSIZ bytes remain.  */
+      xlen = len / OPSIZ;
+      while (xlen > 0)
+	{
+	  ((op_t *) dstp)[0] = cccc;
+	  dstp += OPSIZ;
+	  xlen -= 1;
+	}
+      len %= OPSIZ;
+    }
+
+  /* Write the last few bytes.  */
+  while (len > 0)
+    {
+      ((unsigned char *) dstp)[0] = c;
+      dstp += 1;
+      len -= 1;
+    }
+
+  return dstpp;
+}
diff --git a/arch/parisc/lib/ucmpdi2.c b/arch/parisc/lib/ucmpdi2.c
new file mode 100644
index 000000000..9d8b4dbae
--- /dev/null
+++ b/arch/parisc/lib/ucmpdi2.c
@@ -0,0 +1,27 @@
+// SPDX-License-Identifier: GPL-2.0
+#include <linux/module.h>
+#include <linux/libgcc.h>
+
+union ull_union {
+	unsigned long long ull;
+	struct {
+		unsigned int high;
+		unsigned int low;
+	} ui;
+};
+
+word_type __ucmpdi2(unsigned long long a, unsigned long long b)
+{
+	union ull_union au = {.ull = a};
+	union ull_union bu = {.ull = b};
+
+	if (au.ui.high < bu.ui.high)
+		return 0;
+	else if (au.ui.high > bu.ui.high)
+		return 2;
+	if (au.ui.low < bu.ui.low)
+		return 0;
+	else if (au.ui.low > bu.ui.low)
+		return 2;
+	return 1;
+}