diff --git a/src/pal/inc/unixasmmacrosarm.inc b/src/pal/inc/unixasmmacrosarm.inc
index 169a5a727fb6..1295967192be 100644
--- a/src/pal/inc/unixasmmacrosarm.inc
+++ b/src/pal/inc/unixasmmacrosarm.inc
@@ -152,9 +152,8 @@ C_FUNC(\Name\()_End):
         .endif
 
         PUSH_CALLEE_SAVED_REGISTERS
-        .setfp r7, sp, #(3 * 4)
+        PROLOG_STACK_SAVE_OFFSET r7, #12
         // let r7 point the saved r7 in the stack (clang FP style)
-        add r7, sp, #(3 * 4)
 
         alloc_stack     __PWTB_StackAlloc
 
@@ -208,6 +207,11 @@ C_FUNC(\Name\()_End):
         mov \Register, sp
 .endm
 
+.macro PROLOG_STACK_SAVE_OFFSET Register, Offset
+        .setfp \Register, sp, \Offset
+        add \Register, sp, \Offset
+.endm
+
 .macro EPILOG_STACK_FREE Size
         add sp, sp, \Size
 .endm
@@ -216,6 +220,10 @@ C_FUNC(\Name\()_End):
         mov sp, \Register
 .endm
 
+.macro EPILOG_STACK_RESTORE_OFFSET Register, Offset
+        sub sp, \Register, \Offset
+.endm
+
 .macro EPILOG_BRANCH Target
         b \Target
 .endm
diff --git a/src/vm/arm/asmhelpers.S b/src/vm/arm/asmhelpers.S
index 4f8a3d8de83a..715dce89512b 100644
--- a/src/vm/arm/asmhelpers.S
+++ b/src/vm/arm/asmhelpers.S
@@ -32,7 +32,7 @@
 //void CallDescrWorkerInternal(CallDescrData * pCallDescrData)//
         NESTED_ENTRY CallDescrWorkerInternal,_TEXT,NoHandler
         PROLOG_PUSH         "{r4,r5,r7,lr}"
-        PROLOG_STACK_SAVE   r7
+        PROLOG_STACK_SAVE_OFFSET   r7, #8
 
         mov     r5,r0 // save pCallDescrData in r5
 
@@ -129,7 +129,7 @@ LOCAL_LABEL(LReturnDone):
         vldm    sp, {d0-d3}
 #endif
 
-        EPILOG_STACK_RESTORE    r7
+        EPILOG_STACK_RESTORE_OFFSET   r7, #8
         EPILOG_POP              "{r4,r5,r7,pc}"
 
         NESTED_END CallDescrWorkerInternal,_TEXT
@@ -174,7 +174,7 @@ LOCAL_LABEL(LReturnDone):
         // Spill callee saved registers and return address.
         PROLOG_PUSH         "{r4-r11,lr}"
         
-        PROLOG_STACK_SAVE   r7
+        PROLOG_STACK_SAVE_OFFSET   r7, #12
 
         //
         // This is the code that would have to run to setup this frame
@@ -241,7 +241,7 @@ LOCAL_LABEL(GoodGSCookie):
         // 
         // epilog
         //
-        EPILOG_STACK_RESTORE    r7
+        EPILOG_STACK_RESTORE_OFFSET   r7, #12
         EPILOG_POP              "{r4-r11,lr}"
         bx lr
         
@@ -287,7 +287,8 @@ LOCAL_LABEL(LNullThis):
 //
         NESTED_ENTRY TheUMEntryPrestub,_TEXT,NoHandler
 
-        push {r0-r4,lr}
+        PROLOG_PUSH "{r0-r4,r7,r8,lr}" // add r8 to make stack aligned by 8B
+        PROLOG_STACK_SAVE_OFFSET r7, #20
         vpush {d0-d7}
 
         CHECK_STACK_ALIGNMENT
@@ -300,7 +301,7 @@ LOCAL_LABEL(LNullThis):
 
         // Epilog
         vpop {d0-d7}
-        pop {r0-r4,lr}
+        pop {r0-r4,r7,r8,lr}
         bx r12
 
         NESTED_END TheUMEntryPrestub,_TEXT
@@ -311,7 +312,7 @@ LOCAL_LABEL(LNullThis):
         NESTED_ENTRY UMThunkStub,_TEXT,NoHandler
         PROLOG_PUSH         "{r4,r5,r7,r11,lr}"
         push                {r0-r3,r12}
-        PROLOG_STACK_SAVE   r7
+        PROLOG_STACK_SAVE_OFFSET   r7, #28
 
         //GBLA UMThunkStub_HiddenArg // offset of saved UMEntryThunk *
         //GBLA UMThunkStub_StackArgs // offset of original stack args (total size of UMThunkStub frame)
@@ -374,7 +375,7 @@ LOCAL_LABEL(UMThunkStub_PostCall):
         mov                 r4, 0
         str                 r4, [r5, #Thread__m_fPreemptiveGCDisabled]
 
-        EPILOG_STACK_RESTORE r7
+        EPILOG_STACK_RESTORE_OFFSET    r7, #28
         free_stack           4 * 5
         EPILOG_POP           "{r4,r5,r7,r11,pc}"
 
@@ -424,7 +425,7 @@ LOCAL_LABEL(UMThunkStub_WrongAppDomain):
         NESTED_ENTRY UM2MThunk_WrapperHelper, _TEXT, NoHandler
 
         PROLOG_PUSH         "{r4-r7,r11,lr}"
-        PROLOG_STACK_SAVE   r7
+        PROLOG_STACK_SAVE_OFFSET   r7, #12
 
         CHECK_STACK_ALIGNMENT
 
@@ -480,7 +481,7 @@ LOCAL_LABEL(UM2MThunk_WrapperHelper_ArgumentsSetup):
         vldm                sp, {d0-d3}
 #endif
 
-        EPILOG_STACK_RESTORE r7
+        EPILOG_STACK_RESTORE_OFFSET r7, #12
         EPILOG_POP          "{r4-r7,r11,pc}"
 
         NESTED_END UM2MThunk_WrapperHelper, _TEXT
@@ -516,7 +517,8 @@ ThePreStubPatchLabel:
 // The call in ndirect import precode points to this function.
         NESTED_ENTRY NDirectImportThunk, _TEXT, NoHandler
 
-        push {r0-r4,lr}                                 // Spill general argument registers, return address and
+        PROLOG_PUSH "{r0-r4,r7,r8,lr}"                  // Spill general argument registers, return address and
+        PROLOG_STACK_SAVE_OFFSET r7, #20
                                                         // arbitrary register to keep stack aligned
         vpush {d0-d7}                                   // Spill floating point argument registers
 
@@ -527,7 +529,7 @@ ThePreStubPatchLabel:
         mov     r12, r0
 
         vpop {d0-d7}
-        pop {r0-r4,lr}
+        pop {r0-r4,r7,r8,lr}
 
         // If we got back from NDirectImportWorker, the MD has been successfully
         // linked. Proceed to execute the original DLL call.
@@ -676,8 +678,8 @@ LOCAL_LABEL(Done):
         NESTED_ENTRY RedirectedHandledJITCaseFor\reason\()_Stub, _TEXT, NoHandler
 
         PROLOG_PUSH "{r7,lr}"   // return address
-        alloc_stack 4           // stack slot to save the CONTEXT *
         PROLOG_STACK_SAVE r7
+        alloc_stack 4           // stack slot to save the CONTEXT *
 
         //REDIRECTSTUB_SP_OFFSET_CONTEXT is defined in asmconstants.h
         //If CONTEXT is not saved at 0 offset from SP it must be changed as well.
@@ -793,7 +795,8 @@ LOCAL_LABEL(stackProbe_loop):
     NESTED_ENTRY VirtualMethodFixupStub, _TEXT, NoHandler
 
     // Save arguments and return address
-    push {r0-r3, lr}
+    PROLOG_PUSH "{r0-r3, r7,r8, lr}" // keep increase by 8B for alignment
+    PROLOG_STACK_SAVE_OFFSET r7, #20
 
     // Align stack
     alloc_stack  SIZEOF__FloatArgumentRegisters + 4
@@ -820,7 +823,7 @@ LOCAL_LABEL(stackProbe_loop):
     // pop the stack and restore original register state
     vldm               sp, {d0-d7}
     free_stack SIZEOF__FloatArgumentRegisters + 4
-    pop {r0-r3, lr}
+    pop {r0-r3, r7,r8, lr}
 
     PATCH_LABEL VirtualMethodFixupPatchLabel
 
@@ -906,7 +909,8 @@ LOCAL_LABEL(stackProbe_loop):
 //
     NESTED_ENTRY JIT_RareDisableHelper, _TEXT, NoHandler
 
-    PROLOG_PUSH "{r0-r1, r11, lr}" // save integer return value
+    PROLOG_PUSH "{r0-r1, r7,r8, r11, lr}" // save integer return value
+    PROLOG_STACK_SAVE_OFFSET r7, #8
     vpush {d0-d3}                  // floating point return value
 
     CHECK_STACK_ALIGNMENT
@@ -914,7 +918,7 @@ LOCAL_LABEL(stackProbe_loop):
     bl          C_FUNC(JIT_RareDisableHelperWorker)
 
     vpop {d0-d3}
-    EPILOG_POP "{r0-r1, r11, pc}"
+    EPILOG_POP "{r0-r1, r7,r8, r11, pc}"
 
     NESTED_END JIT_RareDisableHelper, _TEXT
 
diff --git a/src/vm/arm/ehhelpers.S b/src/vm/arm/ehhelpers.S
index dfe1b11e0d31..c32a06b7ee6c 100644
--- a/src/vm/arm/ehhelpers.S
+++ b/src/vm/arm/ehhelpers.S
@@ -47,6 +47,8 @@ OFFSET_OF_FRAME=(4 + SIZEOF__GSCookie)
         //
         // Runtime check for 8-byte alignment. 
         PROLOG_STACK_SAVE r7
+        // We lose stack unwindability here by configuring fp(r7) incorrectely
+        // here.
         and r0, r7, #4
         sub sp, sp, r0
 
@@ -99,6 +101,7 @@ OFFSET_OF_FRAME=(4 + SIZEOF__GSCookie)
         NESTED_ENTRY CallEHFunclet, _TEXT, NoHandler
 
         PROLOG_PUSH  "{r4-r11, lr}"
+        PROLOG_STACK_SAVE_OFFSET  r7, #12
         alloc_stack  4
 
         // On entry:
@@ -124,7 +127,8 @@ OFFSET_OF_FRAME=(4 + SIZEOF__GSCookie)
         // frame pointer for accessing the locals in the parent method.
         NESTED_ENTRY CallEHFilterFunclet, _TEXT, NoHandler
 
-        PROLOG_PUSH  "{lr}"
+        PROLOG_PUSH  "{r7, lr}"
+        PROLOG_STACK_SAVE r7
         alloc_stack  4
 
         // On entry:
@@ -140,6 +144,6 @@ OFFSET_OF_FRAME=(4 + SIZEOF__GSCookie)
         blx r2
 
         free_stack   4
-        EPILOG_POP   "{pc}"
+        EPILOG_POP   "{r7, pc}"
 
         NESTED_END CallEHFilterFunclet, _TEXT