Merge tag 'trace-v7.3-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/trace/linux-tracegrafted

Pull tracing fixes from Steven Rostedt:

 - Don't destroy user event fields when removal fails

   User event fields are destroyed before the event is removed from
   visibility. But that can fail leaving the still visible event with no
   fields. Move the destroying of the fields to after the event is
   successfully removed from visibility.

 - Initialize function graph state is fork before calling
   copy_exec_state()

   For non-CLONE_VM forks, copy_exec_state() allocates a new
   task_exec_state. If that allocation fails, ftrace_graph_exit_task()
   will free the tasks ret_stack pointer. Since that pointer is still
   using the parent's ret_stack, it mistakenly frees the parent's
   pointer too.

   Call ftrace_graph_init() on the task first which will NULL out the
   new tasks's ret_stack and if the copy fails, it will not free
   anything.

 - Remove FGRAPH_MAX_INDEX

   The macro FGRAPH_MAX_INDEX was added but never used. Remove it.

 - Save ent_size in function graph printing of nested functions

   The function graph tracer needs to look at the next event to see if
   the next event is the return of the current function entry. If it is,
   it prints a single line:

	ktime_get();

   Otherwise it prints it like a nested function:

	tick_nohz_irq_exit() {
	    ktime_get();
	    kcpustat_irq_exit();
	}

   In order to look at the next event, it must save the current event so
   that it has the information to print from it. It saves the event in
   the iterator descriptor called "ent". What it doesn't save is the
   ent_size of the event which is now used to know if the function graph
   arguments are to be printed. The peek doesn't save the size so the
   size used happens to be that of the size of the last event that was
   seen.

   Save the entry event size in the iterator descriptor so that the
   correct size is used.

 - Fix several errors with freeing data in the histogram code

   The histogram code had a lot of leaked or or incorrect accounting
   when failures happen. Correct them.

 - Fix histogram regression of .percent and .graph modifiers

   Up until 6.3 histogram values could have "percent" or "graph"
   modifiers that changed how they were printed. But a change that added
   restricting histograms values from being strings, stack traces and
   other modifiers inadvertently prevented them from using the percent
   and graph modifiers, which were legal use cases for values.

   Put back the percent and graph modifiers.

 - Fix various typos in the comments

 - Set the trace_clock before initializing a histogram with clock
   argument

   The histogram API allows the user to specific which trace clock to
   use via a "clock=" string. The histogram is set up first before the
   clock is checked. If the passed in clock is not valid, it exits
   without fully fixing up the histogram leaving it on the list and a
   use-after-free can trigger.

   Update the clock argument first and if it fails then exit gracefully
   before the histogram trigger is placed on any lists.

 - Restore :mod: trailer after parsing in ftrace_set_clr_event

   The function ftrace_set_clr_event() modifies the parse string and
   needs to put it back to what was passed in. It searches for ":mod:"
   via a strsep() but fails to put back the first ':' in the string.

   Add back the ':' in the passed in string.

 - Take trace_array reference when opening a tracer options file

   The options files are dynamically created and some tracers add their
   own options. When a tracer adds their own list of options, the
   trace_array holding them has an array to hold the list of options for
   each tracer. This array increases in size via a krealloc(), and the
   new entry gets a newly allocated array to hold the options of the new
   tracer being added.

   The element in each entry of the tracer's option array holds a
   pointer back to the trace_array, a pointer to the tracer it is
   associated to, a pointer to the flags of the option.

   The issue is that these arrays are freed when the trace_array is
   freed when its instance it represents is removed from the instances
   directory. There's a race that an open of one of these options files
   can happen when the instance is being removed.

   Add a new helper function to be called by the open function of the
   options file to iterate all existing trace_arrays under a lock and
   find the one that has the given option element in one of it's tracer
   arrays. If found, then update the associated trace_array's reference
   counter to keep it from being freed. If not found, have the open call
   return -ENODEV.

 - Disable interrupts when acquiring the lock in rb_wake_up_waiters()

   The function rb_wake_up_waiters() assumes it will be called in
   interrupt context and does not disable irqs when taking
   cpu_buffer->reader_lock, which can be called in hard interrupt
   context. The issue is in PREEMPT_RT, this function is called in
   thread context leaving this lock open to a deadlock.

   Take the lock with interrupts disabled.

 - Use rcu_assign_pointer() for tmp_ops filter hash

   The tmp_ops used in update_ftrace_direct_mod() assigns its
   filter_hash field directly, but that field is annotated as __rcu and
   sparse complains. Assign it with rcu_assign_pointer()

 - Fix use-after-free in enable_trigger_private_data_free()

   The trace_event_call is accessed through the event_trigger_data's
   trace_event_file pointer to put the trace_event_call on freeing. The
   issue is that the trace_event_file data may have been freed already
   causing a use-after-free. Add a field to the event_trigger_data that
   points directly to the trace_event_call so that it can decrement its
   reference directly without needing to go through the
   trace_event_file.

 - Fix accounting of buffer data remote headers

   trace_buffer_desc_size() and trace_remote_alloc_buffer() undercount
   the number of pages is needed for the asked for size as it doesn't
   take into account the meta data on each page. Add a helper function
   to do the calculation properly and use that in these functions.

 - Catch nr_page_va overflow in ring_buffer_desc sizing

   The number of pages per remote ring buffer is capped by
   ring_buffer_desc::nr_page_va (32 bits). A buffer_size large enough to
   overflow that field would silently allocate a descriptor smaller than
   what was asked for.

 - Do not resize the subbuf order if any per_cpu buffer is disabled

   The mmapping of ring buffers disables resizing the subbuffers, but it
   is done per-cpu whereas the subbuf size change is done for all the
   per_cpu buffers under the buffer->mutex. It could change the size of
   some while the mapping is happening on others. Have the resize of the
   subbuf order check all the per_cpu buffers under the lock to see if
   any of them is disabled before starting and causing an inconsistency
   between buffers that are being mapped.

* tag 'trace-v7.3-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/trace/linux-trace: (25 commits)
  ring-buffer: Check resize_disabled before publishing the new subbuf order
  tracing/remotes: Catch nr_page_va overflow in ring_buffer_desc sizing
  tracing/remotes: Account for ring buffer page header in size calculation
  tracing: Don't dereference trace_event_file in deferred trigger free
  ftrace: Use rcu_assign_pointer() for tmp_ops filter hash
  ring-buffer: Acquire the lock with irqsave in rb_wake_up_waiters()
  tracing: Take trace_array reference when opening a tracer options file
  tracing: Fix ring_buffer_read_page_size() kernel-doc
  tracing: Restore :mod: trailer after parsing in ftrace_set_clr_event()
  tracing: Fix memory corruption from a "STACKTRACE" histogram key
  tracing: Fix memory corruption from the histogram stacktrace modifier
  tracing: Undo the registration when enabling the histogram trigger fails
  tracing: Take the reference before publishing the named histogram trigger
  tracing: Set the trace clock before registering the histogram trigger
  tracing: Fix typo "preceeded" in comment
  tracing: Fix typo "availabe" in comment
  tracing: Let histogram values keep the percent and graph modifiers
  tracing: Keep the entry count when the histogram stats allocation fails
  tracing: Free histogram the field rejected for a bad modifier
  tracing: Free histogram the var ref when its initialization fails
  ...

9 files changed, 1258 insertions, 0 deletions

diff --git a/arch/hexagon/lib/Makefile b/arch/hexagon/lib/Makefile
new file mode 100644
index 000000000..107894c09
--- /dev/null
+++ b/arch/hexagon/lib/Makefile
@@ -0,0 +1,6 @@
+# SPDX-License-Identifier: GPL-2.0-only
+#
+# Makefile for hexagon-specific library files.
+#
+obj-y = checksum.o memcpy.o memset.o memcpy_likely_aligned.o \
+         divsi3.o modsi3.o udivsi3.o  umodsi3.o
diff --git a/arch/hexagon/lib/checksum.c b/arch/hexagon/lib/checksum.c
new file mode 100644
index 000000000..ba50822a0
--- /dev/null
+++ b/arch/hexagon/lib/checksum.c
@@ -0,0 +1,178 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Checksum functions for Hexagon
+ *
+ * Copyright (c) 2010-2011, The Linux Foundation. All rights reserved.
+ */
+
+/*  This was derived from arch/alpha/lib/checksum.c  */
+
+
+#include <linux/module.h>
+#include <linux/string.h>
+
+#include <asm/byteorder.h>
+#include <net/checksum.h>
+#include <linux/uaccess.h>
+#include <asm/intrinsics.h>
+
+
+/*  Vector value operations  */
+#define SIGN(x, y)	((0x8000ULL*x)<<y)
+#define CARRY(x, y)	((0x0002ULL*x)<<y)
+#define SELECT(x, y)	((0x0001ULL*x)<<y)
+
+#define VR_NEGATE(a, b, c, d)	(SIGN(a, 48) + SIGN(b, 32) + SIGN(c, 16) \
+	+ SIGN(d, 0))
+#define VR_CARRY(a, b, c, d)	(CARRY(a, 48) + CARRY(b, 32) + CARRY(c, 16) \
+	+ CARRY(d, 0))
+#define VR_SELECT(a, b, c, d)	(SELECT(a, 48) + SELECT(b, 32) + SELECT(c, 16) \
+	+ SELECT(d, 0))
+
+
+/* optimized HEXAGON V3 intrinsic version */
+static inline unsigned short from64to16(u64 x)
+{
+	u64 sum;
+
+	sum = HEXAGON_P_vrmpyh_PP(x^VR_NEGATE(1, 1, 1, 1),
+			     VR_SELECT(1, 1, 1, 1));
+	sum += VR_CARRY(0, 0, 1, 0);
+	sum = HEXAGON_P_vrmpyh_PP(sum, VR_SELECT(0, 0, 1, 1));
+
+	return 0xFFFF & sum;
+}
+
+/*
+ * computes the checksum of the TCP/UDP pseudo-header
+ * returns a 16-bit checksum, already complemented.
+ */
+__sum16 csum_tcpudp_magic(__be32 saddr, __be32 daddr,
+			  __u32 len, __u8 proto, __wsum sum)
+{
+	return (__force __sum16)~from64to16(
+		(__force u64)saddr + (__force u64)daddr +
+		(__force u64)sum + ((len + proto) << 8));
+}
+
+__wsum csum_tcpudp_nofold(__be32 saddr, __be32 daddr,
+			  __u32 len, __u8 proto, __wsum sum)
+{
+	u64 result;
+
+	result = (__force u64)saddr + (__force u64)daddr +
+		 (__force u64)sum + ((len + proto) << 8);
+
+	/* Fold down to 32-bits so we don't lose in the typedef-less
+	   network stack.  */
+	/* 64 to 33 */
+	result = (result & 0xffffffffUL) + (result >> 32);
+	/* 33 to 32 */
+	result = (result & 0xffffffffUL) + (result >> 32);
+	return (__force __wsum)result;
+}
+EXPORT_SYMBOL(csum_tcpudp_nofold);
+
+/*
+ * Do a 64-bit checksum on an arbitrary memory area..
+ *
+ * This isn't a great routine, but it's not _horrible_ either. The
+ * inner loop could be unrolled a bit further, and there are better
+ * ways to do the carry, but this is reasonable.
+ */
+
+/* optimized HEXAGON intrinsic version, with over read fixed */
+unsigned int do_csum(const void *voidptr, int len)
+{
+	u64 sum0, sum1, x0, x1, *ptr8_o, *ptr8_e, *ptr8;
+	int i, start, mid, end, mask;
+	const char *ptr = voidptr;
+	unsigned short *ptr2;
+	unsigned int *ptr4;
+
+	if (len <= 0)
+		return 0;
+
+	start = 0xF & (16-(((int) ptr) & 0xF)) ;
+	mask  = 0x7fffffffUL >> HEXAGON_R_cl0_R(len);
+	start = start & mask ;
+
+	mid = len - start;
+	end = mid & 0xF;
+	mid = mid>>4;
+	sum0 = mid << 18;
+	sum1 = 0;
+
+	if (start & 1)
+		sum0 += (u64) (ptr[0] << 8);
+	ptr2 = (unsigned short *) &ptr[start & 1];
+	if (start & 2)
+		sum1 += (u64) ptr2[0];
+	ptr4 = (unsigned int *) &ptr[start & 3];
+	if (start & 4) {
+		sum0 = HEXAGON_P_vrmpyhacc_PP(sum0,
+			VR_NEGATE(0, 0, 1, 1)^((u64)ptr4[0]),
+			VR_SELECT(0, 0, 1, 1));
+		sum0 += VR_SELECT(0, 0, 1, 0);
+	}
+	ptr8 = (u64 *) &ptr[start & 7];
+	if (start & 8) {
+		sum1 = HEXAGON_P_vrmpyhacc_PP(sum1,
+			VR_NEGATE(1, 1, 1, 1)^(ptr8[0]),
+			VR_SELECT(1, 1, 1, 1));
+		sum1 += VR_CARRY(0, 0, 1, 0);
+	}
+	ptr8_o = (u64 *) (ptr + start);
+	ptr8_e = (u64 *) (ptr + start + 8);
+
+	if (mid) {
+		x0 = *ptr8_e; ptr8_e += 2;
+		x1 = *ptr8_o; ptr8_o += 2;
+		if (mid > 1)
+			for (i = 0; i < mid-1; i++) {
+				sum0 = HEXAGON_P_vrmpyhacc_PP(sum0,
+					x0^VR_NEGATE(1, 1, 1, 1),
+					VR_SELECT(1, 1, 1, 1));
+				sum1 = HEXAGON_P_vrmpyhacc_PP(sum1,
+					x1^VR_NEGATE(1, 1, 1, 1),
+					VR_SELECT(1, 1, 1, 1));
+				x0 = *ptr8_e; ptr8_e += 2;
+				x1 = *ptr8_o; ptr8_o += 2;
+			}
+		sum0 = HEXAGON_P_vrmpyhacc_PP(sum0, x0^VR_NEGATE(1, 1, 1, 1),
+			VR_SELECT(1, 1, 1, 1));
+		sum1 = HEXAGON_P_vrmpyhacc_PP(sum1, x1^VR_NEGATE(1, 1, 1, 1),
+			VR_SELECT(1, 1, 1, 1));
+	}
+
+	ptr4 = (unsigned int *) &ptr[start + (mid * 16) + (end & 8)];
+	if (end & 4) {
+		sum1 = HEXAGON_P_vrmpyhacc_PP(sum1,
+			VR_NEGATE(0, 0, 1, 1)^((u64)ptr4[0]),
+			VR_SELECT(0, 0, 1, 1));
+		sum1 += VR_SELECT(0, 0, 1, 0);
+	}
+	ptr2 = (unsigned short *) &ptr[start + (mid * 16) + (end & 12)];
+	if (end & 2)
+		sum0 += (u64) ptr2[0];
+
+	if (end & 1)
+		sum1 += (u64) ptr[start + (mid * 16) + (end & 14)];
+
+	ptr8 = (u64 *) &ptr[start + (mid * 16)];
+	if (end & 8) {
+		sum0 = HEXAGON_P_vrmpyhacc_PP(sum0,
+			VR_NEGATE(1, 1, 1, 1)^(ptr8[0]),
+			VR_SELECT(1, 1, 1, 1));
+		sum0 += VR_CARRY(0, 0, 1, 0);
+	}
+	sum0 = HEXAGON_P_vrmpyh_PP((sum0+sum1)^VR_NEGATE(0, 0, 0, 1),
+		VR_SELECT(0, 0, 1, 1));
+	sum0 += VR_NEGATE(0, 0, 0, 1);
+	sum0 = HEXAGON_P_vrmpyh_PP(sum0, VR_SELECT(0, 0, 1, 1));
+
+	if (start & 1)
+		sum0 = (sum0 << 8) | (0xFF & (sum0 >> 8));
+
+	return 0xFFFF & sum0;
+}
diff --git a/arch/hexagon/lib/divsi3.S b/arch/hexagon/lib/divsi3.S
new file mode 100644
index 000000000..783e09424
--- /dev/null
+++ b/arch/hexagon/lib/divsi3.S
@@ -0,0 +1,67 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (c) 2021, The Linux Foundation. All rights reserved.
+ */
+
+#include <linux/linkage.h>
+
+SYM_FUNC_START(__hexagon_divsi3)
+        {
+                p0 = cmp.gt(r0,#-1)
+                p1 = cmp.gt(r1,#-1)
+                r3:2 = vabsw(r1:0)
+        }
+        {
+                p3 = xor(p0,p1)
+                r4 = sub(r2,r3)
+                r6 = cl0(r2)
+                p0 = cmp.gtu(r3,r2)
+        }
+        {
+                r0 = mux(p3,#-1,#1)
+                r7 = cl0(r3)
+                p1 = cmp.gtu(r3,r4)
+        }
+        {
+                r0 = mux(p0,#0,r0)
+                p0 = or(p0,p1)
+                if (p0.new) jumpr:nt r31
+                r6 = sub(r7,r6)
+        }
+        {
+                r7 = r6
+                r5:4 = combine(#1,r3)
+                r6 = add(#1,lsr(r6,#1))
+                p0 = cmp.gtu(r6,#4)
+        }
+        {
+                r5:4 = vaslw(r5:4,r7)
+                if (!p0) r6 = #3
+        }
+        {
+                loop0(1f,r6)
+                r7:6 = vlsrw(r5:4,#1)
+                r1:0 = #0
+        }
+        .falign
+1:
+        {
+                r5:4 = vlsrw(r5:4,#2)
+                if (!p0.new) r0 = add(r0,r5)
+                if (!p0.new) r2 = sub(r2,r4)
+                p0 = cmp.gtu(r4,r2)
+        }
+        {
+                r7:6 = vlsrw(r7:6,#2)
+                if (!p0.new) r0 = add(r0,r7)
+                if (!p0.new) r2 = sub(r2,r6)
+                p0 = cmp.gtu(r6,r2)
+        }:endloop0
+        {
+                if (!p0) r0 = add(r0,r7)
+        }
+        {
+                if (p3) r0 = sub(r1,r0)
+                jumpr r31
+        }
+SYM_FUNC_END(__hexagon_divsi3)
diff --git a/arch/hexagon/lib/memcpy.S b/arch/hexagon/lib/memcpy.S
new file mode 100644
index 000000000..f8b3c02a8
--- /dev/null
+++ b/arch/hexagon/lib/memcpy.S
@@ -0,0 +1,529 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (c) 2010-2011, The Linux Foundation. All rights reserved.
+ */
+
+/*
+ * Description
+ *
+ *   library function for memcpy where length bytes are copied from
+ *   ptr_in to ptr_out. ptr_out is returned unchanged.
+ *   Allows any combination of alignment on input and output pointers
+ *   and length from 0 to 2^32-1
+ *
+ * Restrictions
+ *   The arrays should not overlap, the program will produce undefined output
+ *   if they do.
+ *   For blocks less than 16 bytes a byte by byte copy is performed. For
+ *   8byte alignments, and length multiples, a dword copy is performed up to
+ *   96bytes
+ * History
+ *
+ *   DJH  5/15/09 Initial version 1.0
+ *   DJH  6/ 1/09 Version 1.1 modified ABI to inlcude R16-R19
+ *   DJH  7/12/09 Version 1.2 optimized codesize down to 760 was 840
+ *   DJH 10/14/09 Version 1.3 added special loop for aligned case, was
+ *                            overreading bloated codesize back up to 892
+ *   DJH  4/20/10 Version 1.4 fixed Ldword_loop_epilog loop to prevent loads
+ *                            occurring if only 1 left outstanding, fixes bug
+ *                            # 3888, corrected for all alignments. Peeled off
+ *                            1 32byte chunk from kernel loop and extended 8byte
+ *                            loop at end to solve all combinations and prevent
+ *                            over read.  Fixed Ldword_loop_prolog to prevent
+ *                            overread for blocks less than 48bytes. Reduced
+ *                            codesize to 752 bytes
+ *   DJH  4/21/10 version 1.5 1.4 fix broke code for input block ends not
+ *                            aligned to dword boundaries,underwriting by 1
+ *                            byte, added detection for this and fixed. A
+ *                            little bloat.
+ *   DJH  4/23/10 version 1.6 corrected stack error, R20 was not being restored
+ *                            always, fixed the error of R20 being modified
+ *                            before it was being saved
+ * Natural c model
+ * ===============
+ * void * memcpy(char * ptr_out, char * ptr_in, int length) {
+ *   int i;
+ *   if(length) for(i=0; i < length; i++) { ptr_out[i] = ptr_in[i]; }
+ *   return(ptr_out);
+ * }
+ *
+ * Optimized memcpy function
+ * =========================
+ * void * memcpy(char * ptr_out, char * ptr_in, int len) {
+ *   int i, prolog, kernel, epilog, mask;
+ *   u8 offset;
+ *   s64 data0, dataF8, data70;
+ *
+ *   s64 * ptr8_in;
+ *   s64 * ptr8_out;
+ *   s32 * ptr4;
+ *   s16 * ptr2;
+ *
+ *   offset = ((int) ptr_in) & 7;
+ *   ptr8_in = (s64 *) &ptr_in[-offset];   //read in the aligned pointers
+ *
+ *   data70 = *ptr8_in++;
+ *   dataF8 = *ptr8_in++;
+ *
+ *   data0 = HEXAGON_P_valignb_PPp(dataF8, data70, offset);
+ *
+ *   prolog = 32 - ((int) ptr_out);
+ *   mask  = 0x7fffffff >> HEXAGON_R_cl0_R(len);
+ *   prolog = prolog & mask;
+ *   kernel = len - prolog;
+ *   epilog = kernel & 0x1F;
+ *   kernel = kernel>>5;
+ *
+ *   if (prolog & 1) { ptr_out[0] = (u8) data0; data0 >>= 8; ptr_out += 1;}
+ *   ptr2 = (s16 *) &ptr_out[0];
+ *   if (prolog & 2) { ptr2[0] = (u16) data0;  data0 >>= 16; ptr_out += 2;}
+ *   ptr4 = (s32 *) &ptr_out[0];
+ *   if (prolog & 4) { ptr4[0] = (u32) data0;  data0 >>= 32; ptr_out += 4;}
+ *
+ *   offset = offset + (prolog & 7);
+ *   if (offset >= 8) {
+ *     data70 = dataF8;
+ *     dataF8 = *ptr8_in++;
+ *   }
+ *   offset = offset & 0x7;
+ *
+ *   prolog = prolog >> 3;
+ *   if (prolog) for (i=0; i < prolog; i++) {
+ *       data0 = HEXAGON_P_valignb_PPp(dataF8, data70, offset);
+ *       ptr8_out = (s64 *) &ptr_out[0]; *ptr8_out = data0; ptr_out += 8;
+ *       data70 = dataF8;
+ *       dataF8 = *ptr8_in++;
+ *   }
+ *   if(kernel) { kernel -= 1; epilog += 32; }
+ *   if(kernel) for(i=0; i < kernel; i++) {
+ *       data0 = HEXAGON_P_valignb_PPp(dataF8, data70, offset);
+ *       ptr8_out = (s64 *) &ptr_out[0]; *ptr8_out = data0; ptr_out += 8;
+ *       data70 = *ptr8_in++;
+ *
+ *       data0 = HEXAGON_P_valignb_PPp(data70, dataF8, offset);
+ *       ptr8_out = (s64 *) &ptr_out[0]; *ptr8_out = data0; ptr_out += 8;
+ *       dataF8 = *ptr8_in++;
+ *
+ *       data0 = HEXAGON_P_valignb_PPp(dataF8, data70, offset);
+ *       ptr8_out = (s64 *) &ptr_out[0]; *ptr8_out = data0; ptr_out += 8;
+ *       data70 = *ptr8_in++;
+ *
+ *       data0 = HEXAGON_P_valignb_PPp(data70, dataF8, offset);
+ *       ptr8_out = (s64 *) &ptr_out[0]; *ptr8_out = data0; ptr_out += 8;
+ *       dataF8 = *ptr8_in++;
+ *   }
+ *   epilogdws = epilog >> 3;
+ *   if (epilogdws) for (i=0; i < epilogdws; i++) {
+ *       data0 = HEXAGON_P_valignb_PPp(dataF8, data70, offset);
+ *       ptr8_out = (s64 *) &ptr_out[0]; *ptr8_out = data0; ptr_out += 8;
+ *       data70 = dataF8;
+ *       dataF8 = *ptr8_in++;
+ *   }
+ *   data0 = HEXAGON_P_valignb_PPp(dataF8, data70, offset);
+ *
+ *   ptr4 = (s32 *) &ptr_out[0];
+ *   if (epilog & 4) { ptr4[0] = (u32) data0; data0 >>= 32; ptr_out += 4;}
+ *   ptr2 = (s16 *) &ptr_out[0];
+ *   if (epilog & 2) { ptr2[0] = (u16) data0; data0 >>= 16; ptr_out += 2;}
+ *   if (epilog & 1) { *ptr_out++ = (u8) data0; }
+ *
+ *   return(ptr_out - length);
+ * }
+ *
+ * Codesize : 784 bytes
+ */
+
+
+#define ptr_out		R0	/*  destination  pounter  */
+#define ptr_in		R1	/*  source pointer  */
+#define len		R2	/*  length of copy in bytes  */
+
+#define data70		R13:12	/*  lo 8 bytes of non-aligned transfer  */
+#define dataF8		R11:10	/*  hi 8 bytes of non-aligned transfer  */
+#define ldata0		R7:6	/*  even 8 bytes chunks  */
+#define ldata1		R25:24	/*  odd 8 bytes chunks  */
+#define data1		R7	/*  lower 8 bytes of ldata1  */
+#define data0		R6	/*  lower 8 bytes of ldata0  */
+
+#define ifbyte		p0	/*  if transfer has bytes in epilog/prolog  */
+#define ifhword		p0	/*  if transfer has shorts in epilog/prolog  */
+#define ifword		p0	/*  if transfer has words in epilog/prolog  */
+#define noprolog	p0	/*  no prolog, xfer starts at 32byte  */
+#define nokernel	p1	/*  no 32byte multiple block in the transfer  */
+#define noepilog	p0	/*  no epilog, xfer ends on 32byte boundary  */
+#define align		p2	/*  alignment of input rel to 8byte boundary  */
+#define kernel1		p0	/*  kernel count == 1  */
+
+#define dalign		R25	/*  rel alignment of input to output data  */
+#define star3		R16	/*  number bytes in prolog - dwords  */
+#define rest		R8	/*  length - prolog bytes  */
+#define back		R7	/*  nr bytes > dword boundary in src block  */
+#define epilog		R3	/*  bytes in epilog  */
+#define inc		R15:14	/*  inc kernel by -1 and defetch ptr by 32  */
+#define kernel		R4	/*  number of 32byte chunks in kernel  */
+#define ptr_in_p_128	R5	/*  pointer for prefetch of input data  */
+#define mask		R8	/*  mask used to determine prolog size  */
+#define shift		R8	/*  used to work a shifter to extract bytes  */
+#define shift2		R5	/*  in epilog to workshifter to extract bytes */
+#define prolog		R15	/*  bytes in  prolog  */
+#define epilogdws	R15	/*  number dwords in epilog  */
+#define shiftb		R14	/*  used to extract bytes  */
+#define offset		R9	/*  same as align in reg  */
+#define ptr_out_p_32	R17	/*  pointer to output dczero  */
+#define align888	R14	/*  if simple dword loop can be used  */
+#define len8		R9	/*  number of dwords in length  */
+#define over		R20	/*  nr of bytes > last inp buf dword boundary */
+
+#define ptr_in_p_128kernel	R5:4	/*  packed fetch pointer & kernel cnt */
+
+	.section .text
+	.p2align 4
+        .global memcpy
+        .type memcpy, @function
+memcpy:
+{
+	p2 = cmp.eq(len, #0);		/*  =0 */
+	align888 = or(ptr_in, ptr_out);	/*  %8 < 97 */
+	p0 = cmp.gtu(len, #23);		/*  %1, <24 */
+	p1 = cmp.eq(ptr_in, ptr_out);	/*  attempt to overwrite self */
+}
+{
+	p1 = or(p2, p1);
+	p3 = cmp.gtu(len, #95);		/*  %8 < 97 */
+	align888 = or(align888, len);	/*  %8 < 97 */
+	len8 = lsr(len, #3);		/*  %8 < 97 */
+}
+{
+	dcfetch(ptr_in);		/*  zero/ptrin=ptrout causes fetch */
+	p2 = bitsclr(align888, #7);	/*  %8 < 97  */
+	if(p1) jumpr r31;		/*  =0  */
+}
+{
+	p2 = and(p2,!p3);			/*  %8 < 97  */
+	if (p2.new) len = add(len, #-8);	/*  %8 < 97  */
+	if (p2.new) jump:NT .Ldwordaligned; 	/*  %8 < 97  */
+}
+{
+	if(!p0) jump .Lbytes23orless;	/*  %1, <24  */
+	mask.l = #LO(0x7fffffff);
+	/*  all bytes before line multiples of data  */
+	prolog = sub(#0, ptr_out);
+}
+{
+	/*  save r31 on stack, decrement sp by 16  */
+	allocframe(#24);
+	mask.h = #HI(0x7fffffff);
+	ptr_in_p_128 = add(ptr_in, #32);
+	back = cl0(len);
+}
+{
+	memd(sp+#0) = R17:16;		/*  save r16,r17 on stack6  */
+	r31.l = #LO(.Lmemcpy_return);	/*  set up final return pointer  */
+	prolog &= lsr(mask, back);
+	offset = and(ptr_in, #7);
+}
+{
+	memd(sp+#8) = R25:24;		/*  save r25,r24 on stack  */
+	dalign = sub(ptr_out, ptr_in);
+	r31.h = #HI(.Lmemcpy_return);	/*  set up final return pointer  */
+}
+{
+	/*  see if there if input buffer end if aligned  */
+	over = add(len, ptr_in);
+	back = add(len, offset);
+	memd(sp+#16) = R21:20;		/*  save r20,r21 on stack  */
+}
+{
+	noprolog = bitsclr(prolog, #7);
+	prolog = and(prolog, #31);
+	dcfetch(ptr_in_p_128);
+	ptr_in_p_128 = add(ptr_in_p_128, #32);
+}
+{
+	kernel = sub(len, prolog);
+	shift = asl(prolog, #3);
+	star3 = and(prolog, #7);
+	ptr_in = and(ptr_in, #-8);
+}
+{
+	prolog = lsr(prolog, #3);
+	epilog = and(kernel, #31);
+	ptr_out_p_32 = add(ptr_out, prolog);
+	over = and(over, #7);
+}
+{
+	p3 = cmp.gtu(back, #8);
+	kernel = lsr(kernel, #5);
+	dcfetch(ptr_in_p_128);
+	ptr_in_p_128 = add(ptr_in_p_128, #32);
+}
+{
+	p1 = cmp.eq(prolog, #0);
+	if(!p1.new) prolog = add(prolog, #1);
+	dcfetch(ptr_in_p_128);	/*  reserve the line 64bytes on  */
+	ptr_in_p_128 = add(ptr_in_p_128, #32);
+}
+{
+	nokernel = cmp.eq(kernel,#0);
+	dcfetch(ptr_in_p_128);	/* reserve the line 64bytes on  */
+	ptr_in_p_128 = add(ptr_in_p_128, #32);
+	shiftb = and(shift, #8);
+}
+{
+	dcfetch(ptr_in_p_128);		/*  reserve the line 64bytes on  */
+	ptr_in_p_128 = add(ptr_in_p_128, #32);
+	if(nokernel) jump .Lskip64;
+	p2 = cmp.eq(kernel, #1);	/*  skip ovr if kernel == 0  */
+}
+{
+	dczeroa(ptr_out_p_32);
+	/*  don't advance pointer  */
+	if(!p2) ptr_out_p_32 = add(ptr_out_p_32, #32);
+}
+{
+	dalign = and(dalign, #31);
+	dczeroa(ptr_out_p_32);
+}
+.Lskip64:
+{
+	data70 = memd(ptr_in++#16);
+	if(p3) dataF8 = memd(ptr_in+#8);
+	if(noprolog) jump .Lnoprolog32;
+	align = offset;
+}
+/*  upto initial 7 bytes  */
+{
+	ldata0 = valignb(dataF8, data70, align);
+	ifbyte = tstbit(shift,#3);
+	offset = add(offset, star3);
+}
+{
+	if(ifbyte) memb(ptr_out++#1) = data0;
+	ldata0 = lsr(ldata0, shiftb);
+	shiftb = and(shift, #16);
+	ifhword = tstbit(shift,#4);
+}
+{
+	if(ifhword) memh(ptr_out++#2) = data0;
+	ldata0 = lsr(ldata0, shiftb);
+	ifword = tstbit(shift,#5);
+	p2 = cmp.gtu(offset, #7);
+}
+{
+	if(ifword) memw(ptr_out++#4) = data0;
+	if(p2) data70 = dataF8;
+	if(p2) dataF8 = memd(ptr_in++#8);	/*  another 8 bytes  */
+	align = offset;
+}
+.Lnoprolog32:
+{
+	p3 = sp1loop0(.Ldword_loop_prolog, prolog)
+	rest = sub(len, star3);	/*  whats left after the loop  */
+	p0 = cmp.gt(over, #0);
+}
+	if(p0) rest = add(rest, #16);
+.Ldword_loop_prolog:
+{
+	if(p3) memd(ptr_out++#8) = ldata0;
+	ldata0 = valignb(dataF8, data70, align);
+	p0 = cmp.gt(rest, #16);
+}
+{
+	data70 = dataF8;
+	if(p0) dataF8 = memd(ptr_in++#8);
+	rest = add(rest, #-8);
+}:endloop0
+.Lkernel:
+{
+	/*  kernel is at least 32bytes  */
+	p3 = cmp.gtu(kernel, #0);
+	/*  last itn. remove edge effects  */
+	if(p3.new) kernel = add(kernel, #-1);
+	/*  dealt with in last dword loop  */
+	if(p3.new) epilog = add(epilog, #32);
+}
+{
+	nokernel = cmp.eq(kernel, #0);		/*  after adjustment, recheck */
+	if(nokernel.new) jump:NT .Lepilog;	/*  likely not taken  */
+	inc = combine(#32, #-1);
+	p3 = cmp.gtu(dalign, #24);
+}
+{
+	if(p3) jump .Lodd_alignment;
+}
+{
+	loop0(.Loword_loop_25to31, kernel);
+	kernel1 = cmp.gtu(kernel, #1);
+	rest = kernel;
+}
+	.falign
+.Loword_loop_25to31:
+{
+	dcfetch(ptr_in_p_128);	/*  prefetch 4 lines ahead  */
+	if(kernel1) ptr_out_p_32 = add(ptr_out_p_32, #32);
+}
+{
+	dczeroa(ptr_out_p_32);	/*  reserve the next 32bytes in cache  */
+	p3 = cmp.eq(kernel, rest);
+}
+{
+	/*  kernel -= 1  */
+	ptr_in_p_128kernel = vaddw(ptr_in_p_128kernel, inc);
+	/*  kill write on first iteration  */
+	if(!p3) memd(ptr_out++#8) = ldata1;
+	ldata1 = valignb(dataF8, data70, align);
+	data70 = memd(ptr_in++#8);
+}
+{
+	memd(ptr_out++#8) = ldata0;
+	ldata0 = valignb(data70, dataF8, align);
+	dataF8 = memd(ptr_in++#8);
+}
+{
+	memd(ptr_out++#8) = ldata1;
+	ldata1 = valignb(dataF8, data70, align);
+	data70 = memd(ptr_in++#8);
+}
+{
+	memd(ptr_out++#8) = ldata0;
+	ldata0 = valignb(data70, dataF8, align);
+	dataF8 = memd(ptr_in++#8);
+	kernel1 = cmp.gtu(kernel, #1);
+}:endloop0
+{
+	memd(ptr_out++#8) = ldata1;
+	jump .Lepilog;
+}
+.Lodd_alignment:
+{
+	loop0(.Loword_loop_00to24, kernel);
+	kernel1 = cmp.gtu(kernel, #1);
+	rest = add(kernel, #-1);
+}
+	.falign
+.Loword_loop_00to24:
+{
+	dcfetch(ptr_in_p_128);	/*  prefetch 4 lines ahead  */
+	ptr_in_p_128kernel = vaddw(ptr_in_p_128kernel, inc);
+	if(kernel1) ptr_out_p_32 = add(ptr_out_p_32, #32);
+}
+{
+	dczeroa(ptr_out_p_32);	/*  reserve the next 32bytes in cache  */
+}
+{
+	memd(ptr_out++#8) = ldata0;
+	ldata0 = valignb(dataF8, data70, align);
+	data70 = memd(ptr_in++#8);
+}
+{
+	memd(ptr_out++#8) = ldata0;
+	ldata0 = valignb(data70, dataF8, align);
+	dataF8 = memd(ptr_in++#8);
+}
+{
+	memd(ptr_out++#8) = ldata0;
+	ldata0 = valignb(dataF8, data70, align);
+	data70 = memd(ptr_in++#8);
+}
+{
+	memd(ptr_out++#8) = ldata0;
+	ldata0 = valignb(data70, dataF8, align);
+	dataF8 = memd(ptr_in++#8);
+	kernel1 = cmp.gtu(kernel, #1);
+}:endloop0
+.Lepilog:
+{
+	noepilog = cmp.eq(epilog,#0);
+	epilogdws = lsr(epilog, #3);
+	kernel = and(epilog, #7);
+}
+{
+	if(noepilog) jumpr r31;
+	if(noepilog) ptr_out = sub(ptr_out, len);
+	p3 = cmp.eq(epilogdws, #0);
+	shift2 = asl(epilog, #3);
+}
+{
+	shiftb = and(shift2, #32);
+	ifword = tstbit(epilog,#2);
+	if(p3) jump .Lepilog60;
+	if(!p3) epilog = add(epilog, #-16);
+}
+{
+	loop0(.Ldword_loop_epilog, epilogdws);
+	/*  stop criteria is lsbs unless = 0 then its 8  */
+	p3 = cmp.eq(kernel, #0);
+	if(p3.new) kernel= #8;
+	p1 = cmp.gt(over, #0);
+}
+	/*  if not aligned to end of buffer execute 1 more iteration  */
+	if(p1) kernel= #0;
+.Ldword_loop_epilog:
+{
+	memd(ptr_out++#8) = ldata0;
+	ldata0 = valignb(dataF8, data70, align);
+	p3 = cmp.gt(epilog, kernel);
+}
+{
+	data70 = dataF8;
+	if(p3) dataF8 = memd(ptr_in++#8);
+	epilog = add(epilog, #-8);
+}:endloop0
+/* copy last 7 bytes */
+.Lepilog60:
+{
+	if(ifword) memw(ptr_out++#4) = data0;
+	ldata0 = lsr(ldata0, shiftb);
+	ifhword = tstbit(epilog,#1);
+	shiftb = and(shift2, #16);
+}
+{
+	if(ifhword) memh(ptr_out++#2) = data0;
+	ldata0 = lsr(ldata0, shiftb);
+	ifbyte = tstbit(epilog,#0);
+	if(ifbyte.new) len = add(len, #-1);
+}
+{
+	if(ifbyte) memb(ptr_out) = data0;
+	ptr_out = sub(ptr_out, len);	/*  return dest pointer  */
+        jumpr r31;
+}
+/*  do byte copy for small n  */
+.Lbytes23orless:
+{
+	p3 = sp1loop0(.Lbyte_copy, len);
+	len = add(len, #-1);
+}
+.Lbyte_copy:
+{
+	data0 = memb(ptr_in++#1);
+	if(p3) memb(ptr_out++#1) = data0;
+}:endloop0
+{
+	memb(ptr_out) = data0;
+	ptr_out = sub(ptr_out, len);
+	jumpr r31;
+}
+/*  do dword copies for aligned in, out and length  */
+.Ldwordaligned:
+{
+	p3 = sp1loop0(.Ldword_copy, len8);
+}
+.Ldword_copy:
+{
+	if(p3) memd(ptr_out++#8) = ldata0;
+	ldata0 = memd(ptr_in++#8);
+}:endloop0
+{
+	memd(ptr_out) = ldata0;
+	ptr_out = sub(ptr_out, len);
+	jumpr r31;	/*  return to function caller  */
+}
+.Lmemcpy_return:
+	r21:20 = memd(sp+#16);	/*  restore r20+r21  */
+{
+	r25:24 = memd(sp+#8);	/*  restore r24+r25  */
+	r17:16 = memd(sp+#0);	/*  restore r16+r17  */
+}
+	deallocframe;	/*  restore r31 and incrment stack by 16  */
+	jumpr r31
diff --git a/arch/hexagon/lib/memcpy_likely_aligned.S b/arch/hexagon/lib/memcpy_likely_aligned.S
new file mode 100644
index 000000000..6a541fb90
--- /dev/null
+++ b/arch/hexagon/lib/memcpy_likely_aligned.S
@@ -0,0 +1,56 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (c) 2021, The Linux Foundation. All rights reserved.
+ */
+
+#include <linux/linkage.h>
+
+SYM_FUNC_START(__hexagon_memcpy_likely_aligned_min32bytes_mult8bytes)
+        {
+                p0 = bitsclr(r1,#7)
+                p0 = bitsclr(r0,#7)
+                if (p0.new) r5:4 = memd(r1)
+                if (p0.new) r7:6 = memd(r1+#8)
+        }
+        {
+                if (!p0) jump:nt .Lmemcpy_call
+                if (p0) r9:8 = memd(r1+#16)
+                if (p0) r11:10 = memd(r1+#24)
+                p0 = cmp.gtu(r2,#64)
+        }
+        {
+                if (p0) jump:nt .Lmemcpy_call
+                if (!p0) memd(r0) = r5:4
+                if (!p0) memd(r0+#8) = r7:6
+                p0 = cmp.gtu(r2,#32)
+        }
+        {
+                p1 = cmp.gtu(r2,#40)
+                p2 = cmp.gtu(r2,#48)
+                if (p0) r13:12 = memd(r1+#32)
+                if (p1.new) r15:14 = memd(r1+#40)
+        }
+        {
+                memd(r0+#16) = r9:8
+                memd(r0+#24) = r11:10
+        }
+        {
+                if (p0) memd(r0+#32) = r13:12
+                if (p1) memd(r0+#40) = r15:14
+                if (!p2) jumpr:t r31
+        }
+        {
+                p0 = cmp.gtu(r2,#56)
+                r5:4 = memd(r1+#48)
+                if (p0.new) r7:6 = memd(r1+#56)
+        }
+        {
+                memd(r0+#48) = r5:4
+                if (p0) memd(r0+#56) = r7:6
+                jumpr r31
+        }
+
+.Lmemcpy_call:
+        jump memcpy
+
+SYM_FUNC_END(__hexagon_memcpy_likely_aligned_min32bytes_mult8bytes)
diff --git a/arch/hexagon/lib/memset.S b/arch/hexagon/lib/memset.S
new file mode 100644
index 000000000..e67304e3f
--- /dev/null
+++ b/arch/hexagon/lib/memset.S
@@ -0,0 +1,302 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (c) 2011, The Linux Foundation. All rights reserved.
+ */
+
+
+/* HEXAGON assembly optimized memset */
+/* Replaces the standard library function memset */
+
+
+        .macro HEXAGON_OPT_FUNC_BEGIN name
+	.text
+	.p2align 4
+	.globl \name
+	.type  \name, @function
+\name:
+	.endm
+
+	.macro HEXAGON_OPT_FUNC_FINISH name
+	.size  \name, . - \name
+	.endm
+
+/* FUNCTION: memset (v2 version) */
+#if __HEXAGON_ARCH__ < 3
+HEXAGON_OPT_FUNC_BEGIN memset
+	{
+		r6 = #8
+		r7 = extractu(r0, #3 , #0)
+		p0 = cmp.eq(r2, #0)
+		p1 = cmp.gtu(r2, #7)
+	}
+	{
+		r4 = vsplatb(r1)
+		r8 = r0           /* leave r0 intact for return val  */
+		r9 = sub(r6, r7)  /* bytes until double alignment  */
+		if p0 jumpr r31   /* count == 0, so return  */
+	}
+	{
+		r3 = #0
+		r7 = #0
+		p0 = tstbit(r9, #0)
+		if p1 jump 2f /* skip byte loop */
+	}
+
+/* less than 8 bytes to set, so just set a byte at a time and return  */
+
+		loop0(1f, r2) /* byte loop */
+	.falign
+1: /* byte loop */
+	{
+		memb(r8++#1) = r4
+	}:endloop0
+		jumpr r31
+	.falign
+2: /* skip byte loop */
+	{
+		r6 = #1
+		p0 = tstbit(r9, #1)
+		p1 = cmp.eq(r2, #1)
+		if !p0 jump 3f /* skip initial byte store */
+	}
+	{
+		memb(r8++#1) = r4
+		r3:2 = sub(r3:2, r7:6)
+		if p1 jumpr r31
+	}
+	.falign
+3: /* skip initial byte store */
+	{
+		r6 = #2
+		p0 = tstbit(r9, #2)
+		p1 = cmp.eq(r2, #2)
+		if !p0 jump 4f /* skip initial half store */
+	}
+	{
+		memh(r8++#2) = r4
+		r3:2 = sub(r3:2, r7:6)
+		if p1 jumpr r31
+	}
+	.falign
+4: /* skip initial half store */
+	{
+		r6 = #4
+		p0 = cmp.gtu(r2, #7)
+		p1 = cmp.eq(r2, #4)
+		if !p0 jump 5f /* skip initial word store */
+	}
+	{
+		memw(r8++#4) = r4
+		r3:2 = sub(r3:2, r7:6)
+		p0 = cmp.gtu(r2, #11)
+		if p1 jumpr r31
+	}
+	.falign
+5: /* skip initial word store */
+	{
+		r10 = lsr(r2, #3)
+		p1 = cmp.eq(r3, #1)
+		if !p0 jump 7f /* skip double loop */
+	}
+	{
+		r5 = r4
+		r6 = #8
+		loop0(6f, r10) /* double loop */
+	}
+
+/* set bytes a double word at a time  */
+
+	.falign
+6: /* double loop */
+	{
+		memd(r8++#8) = r5:4
+		r3:2 = sub(r3:2, r7:6)
+		p1 = cmp.eq(r2, #8)
+	}:endloop0
+	.falign
+7: /* skip double loop */
+	{
+		p0 = tstbit(r2, #2)
+		if p1 jumpr r31
+	}
+	{
+		r6 = #4
+		p0 = tstbit(r2, #1)
+		p1 = cmp.eq(r2, #4)
+		if !p0 jump 8f /* skip final word store */
+	}
+	{
+		memw(r8++#4) = r4
+		r3:2 = sub(r3:2, r7:6)
+		if p1 jumpr r31
+	}
+	.falign
+8: /* skip final word store */
+	{
+		p1 = cmp.eq(r2, #2)
+		if !p0 jump 9f /* skip final half store */
+	}
+	{
+		memh(r8++#2) = r4
+		if p1 jumpr r31
+	}
+	.falign
+9: /* skip final half store */
+	{
+		memb(r8++#1) = r4
+		jumpr r31
+	}
+HEXAGON_OPT_FUNC_FINISH memset
+#endif
+
+
+/*  FUNCTION: memset (v3 and higher version)  */
+#if __HEXAGON_ARCH__ >= 3
+HEXAGON_OPT_FUNC_BEGIN memset
+	{
+		r7=vsplatb(r1)
+		r6 = r0
+		if (r2==#0) jump:nt .L1
+	}
+	{
+		r5:4=combine(r7,r7)
+		p0 = cmp.gtu(r2,#8)
+		if (p0.new) jump:nt .L3
+	}
+	{
+		r3 = r0
+		loop0(.L47,r2)
+	}
+	.falign
+.L47:
+	{
+		memb(r3++#1) = r1
+	}:endloop0 /* start=.L47 */
+		jumpr r31
+.L3:
+	{
+		p0 = tstbit(r0,#0)
+		if (!p0.new) jump:nt .L8
+		p1 = cmp.eq(r2, #1)
+	}
+	{
+		r6 = add(r0, #1)
+		r2 = add(r2,#-1)
+		memb(r0) = r1
+		if (p1) jump .L1
+	}
+.L8:
+	{
+		p0 = tstbit(r6,#1)
+		if (!p0.new) jump:nt .L10
+	}
+	{
+		r2 = add(r2,#-2)
+		memh(r6++#2) = r7
+		p0 = cmp.eq(r2, #2)
+		if (p0.new) jump:nt .L1
+	}
+.L10:
+	{
+		p0 = tstbit(r6,#2)
+		if (!p0.new) jump:nt .L12
+	}
+	{
+		r2 = add(r2,#-4)
+		memw(r6++#4) = r7
+		p0 = cmp.eq(r2, #4)
+		if (p0.new) jump:nt .L1
+	}
+.L12:
+	{
+		p0 = cmp.gtu(r2,#127)
+		if (!p0.new) jump:nt .L14
+	}
+		r3 = and(r6,#31)
+		if (r3==#0) jump:nt .L17
+	{
+		memd(r6++#8) = r5:4
+		r2 = add(r2,#-8)
+	}
+		r3 = and(r6,#31)
+		if (r3==#0) jump:nt .L17
+	{
+		memd(r6++#8) = r5:4
+		r2 = add(r2,#-8)
+	}
+		r3 = and(r6,#31)
+		if (r3==#0) jump:nt .L17
+	{
+		memd(r6++#8) = r5:4
+		r2 = add(r2,#-8)
+	}
+.L17:
+	{
+		r3 = lsr(r2,#5)
+		if (r1!=#0) jump:nt .L18
+	}
+	{
+		r8 = r3
+		r3 = r6
+		loop0(.L46,r3)
+	}
+	.falign
+.L46:
+	{
+		dczeroa(r6)
+		r6 = add(r6,#32)
+		r2 = add(r2,#-32)
+	}:endloop0 /* start=.L46 */
+.L14:
+	{
+		p0 = cmp.gtu(r2,#7)
+		if (!p0.new) jump:nt .L28
+		r8 = lsr(r2,#3)
+	}
+		loop0(.L44,r8)
+	.falign
+.L44:
+	{
+		memd(r6++#8) = r5:4
+		r2 = add(r2,#-8)
+	}:endloop0 /* start=.L44 */
+.L28:
+	{
+		p0 = tstbit(r2,#2)
+		if (!p0.new) jump:nt .L33
+	}
+	{
+		r2 = add(r2,#-4)
+		memw(r6++#4) = r7
+	}
+.L33:
+	{
+		p0 = tstbit(r2,#1)
+		if (!p0.new) jump:nt .L35
+	}
+	{
+		r2 = add(r2,#-2)
+		memh(r6++#2) = r7
+	}
+.L35:
+		p0 = cmp.eq(r2,#1)
+		if (p0) memb(r6) = r1
+.L1:
+		jumpr r31
+.L18:
+		loop0(.L45,r3)
+	.falign
+.L45:
+		dczeroa(r6)
+	{
+		memd(r6++#8) = r5:4
+		r2 = add(r2,#-32)
+	}
+		memd(r6++#8) = r5:4
+		memd(r6++#8) = r5:4
+	{
+		memd(r6++#8) = r5:4
+	}:endloop0 /* start=.L45  */
+		jump .L14
+HEXAGON_OPT_FUNC_FINISH memset
+#endif
diff --git a/arch/hexagon/lib/modsi3.S b/arch/hexagon/lib/modsi3.S
new file mode 100644
index 000000000..9ea1c86ef
--- /dev/null
+++ b/arch/hexagon/lib/modsi3.S
@@ -0,0 +1,46 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (c) 2021, The Linux Foundation. All rights reserved.
+ */
+
+#include <linux/linkage.h>
+
+SYM_FUNC_START(__hexagon_modsi3)
+        {
+                p2 = cmp.ge(r0,#0)
+                r2 = abs(r0)
+                r1 = abs(r1)
+        }
+        {
+                r3 = cl0(r2)
+                r4 = cl0(r1)
+                p0 = cmp.gtu(r1,r2)
+        }
+        {
+                r3 = sub(r4,r3)
+                if (p0) jumpr r31
+        }
+        {
+                p1 = cmp.eq(r3,#0)
+                loop0(1f,r3)
+                r0 = r2
+                r2 = lsl(r1,r3)
+        }
+        .falign
+1:
+        {
+                p0 = cmp.gtu(r2,r0)
+                if (!p0.new) r0 = sub(r0,r2)
+                r2 = lsr(r2,#1)
+                if (p1) r1 = #0
+        }:endloop0
+        {
+                p0 = cmp.gtu(r2,r0)
+                if (!p0.new) r0 = sub(r0,r1)
+                if (p2) jumpr r31
+        }
+        {
+                r0 = neg(r0)
+                jumpr r31
+        }
+SYM_FUNC_END(__hexagon_modsi3)
diff --git a/arch/hexagon/lib/udivsi3.S b/arch/hexagon/lib/udivsi3.S
new file mode 100644
index 000000000..477f27b93
--- /dev/null
+++ b/arch/hexagon/lib/udivsi3.S
@@ -0,0 +1,38 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (c) 2021, The Linux Foundation. All rights reserved.
+ */
+
+#include <linux/linkage.h>
+
+SYM_FUNC_START(__hexagon_udivsi3)
+        {
+                r2 = cl0(r0)
+                r3 = cl0(r1)
+                r5:4 = combine(#1,#0)
+                p0 = cmp.gtu(r1,r0)
+        }
+        {
+                r6 = sub(r3,r2)
+                r4 = r1
+                r1:0 = combine(r0,r4)
+                if (p0) jumpr r31
+        }
+        {
+                r3:2 = vlslw(r5:4,r6)
+                loop0(1f,r6)
+        }
+        .falign
+1:
+        {
+                p0 = cmp.gtu(r2,r1)
+                if (!p0.new) r1 = sub(r1,r2)
+                if (!p0.new) r0 = add(r0,r3)
+                r3:2 = vlsrw(r3:2,#1)
+        }:endloop0
+        {
+                p0 = cmp.gtu(r2,r1)
+                if (!p0.new) r0 = add(r0,r3)
+                jumpr r31
+        }
+SYM_FUNC_END(__hexagon_udivsi3)
diff --git a/arch/hexagon/lib/umodsi3.S b/arch/hexagon/lib/umodsi3.S
new file mode 100644
index 000000000..280bf06a5
--- /dev/null
+++ b/arch/hexagon/lib/umodsi3.S
@@ -0,0 +1,36 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (c) 2021, The Linux Foundation. All rights reserved.
+ */
+
+#include <linux/linkage.h>
+
+SYM_FUNC_START(__hexagon_umodsi3)
+        {
+                r2 = cl0(r0)
+                r3 = cl0(r1)
+                p0 = cmp.gtu(r1,r0)
+        }
+        {
+                r2 = sub(r3,r2)
+                if (p0) jumpr r31
+        }
+        {
+                loop0(1f,r2)
+                p1 = cmp.eq(r2,#0)
+                r2 = lsl(r1,r2)
+        }
+        .falign
+1:
+        {
+                p0 = cmp.gtu(r2,r0)
+                if (!p0.new) r0 = sub(r0,r2)
+                r2 = lsr(r2,#1)
+                if (p1) r1 = #0
+        }:endloop0
+        {
+                p0 = cmp.gtu(r2,r0)
+                if (!p0.new) r0 = sub(r0,r1)
+                jumpr r31
+        }
+SYM_FUNC_END(__hexagon_umodsi3)