Imported Upstream version 2.0.0.11353upstream/2.0.0.11353

15 files changed, 568 insertions, 355 deletions

diff --git a/src/pal/src/arch/amd64/callsignalhandlerwrapper.S b/src/pal/src/arch/amd64/callsignalhandlerwrapper.S
new file mode 100644
index 0000000000..8260591c30
--- /dev/null
+++ b/src/pal/src/arch/amd64/callsignalhandlerwrapper.S
@@ -0,0 +1,31 @@
+// Licensed to the .NET Foundation under one or more agreements.
+// The .NET Foundation licenses this file to you under the MIT license.
+// See the LICENSE file in the project root for more information.
+
+.intel_syntax noprefix
+#include "unixasmmacros.inc"
+#include "asmconstants.h"
+
+.macro CALL_SIGNAL_HANDLER_WRAPPER Alignment
+
+.globl C_FUNC(SignalHandlerWorkerReturnOffset\Alignment)
+C_FUNC(SignalHandlerWorkerReturnOffset\Alignment):
+    .int LOCAL_LABEL(SignalHandlerWorkerReturn\Alignment)-C_FUNC(CallSignalHandlerWrapper\Alignment)
+
+// This function is never called, only a fake stack frame will be setup to have a return
+// address set to SignalHandlerWorkerReturn during SIGSEGV handling.
+// It enables the unwinder to unwind stack from the handling code to the actual failure site.
+NESTED_ENTRY CallSignalHandlerWrapper\Alignment, _TEXT, NoHandler
+    .cfi_def_cfa_offset (128 + 8 + \Alignment) // red zone + return address + alignment
+    .cfi_offset rip, -(128 + 8 + \Alignment)
+    push_nonvol_reg rbp
+    call    EXTERNAL_C_FUNC(signal_handler_worker)
+LOCAL_LABEL(SignalHandlerWorkerReturn\Alignment):
+    pop rbp
+    ret
+NESTED_END CallSignalHandlerWrapper\Alignment, _TEXT
+
+.endm
+
+CALL_SIGNAL_HANDLER_WRAPPER 0
+CALL_SIGNAL_HANDLER_WRAPPER 8
diff --git a/src/pal/src/arch/amd64/optimizedtls.cpp b/src/pal/src/arch/amd64/optimizedtls.cpp
deleted file mode 100644
index cd89db6b0a..0000000000
--- a/src/pal/src/arch/amd64/optimizedtls.cpp
+++ /dev/null
@@ -1,237 +0,0 @@
-// Licensed to the .NET Foundation under one or more agreements.
-// The .NET Foundation licenses this file to you under the MIT license.
-// See the LICENSE file in the project root for more information.
-
-/*++
-
-
-
-Module Name:
-
-    optimizedtls.cpp
-
-Abstract:
-
-    Implementation of platform-specific Thread local storage functions.
-
-
-
---*/
-
-#include "pal/thread.hpp"
-#include "pal/malloc.hpp"
-
-#include <pthread.h>
-
-#include "pal/dbgmsg.h"
-#include "pal/misc.h"
-#include "pal/debug.h"
-
-#include <stddef.h>
-
-using namespace CorUnix;
-
-SET_DEFAULT_DEBUG_CHANNEL(THREAD);
-
-#if defined(USE_OPTIMIZEDTLSGETTER)
-
-#define PAL_safe_offsetof(s,m) ((size_t)((ptrdiff_t)&(char&)(((s *)64)->m))-64)
-
-/*++
-Function:
-    CorUnix::TLSMakeOptimizedGetter
-
-    Creates a platform-optimized version of TlsGetValue compiled
-    for a particular index.
-
-    Generates the hot part of CorUnix::InternalGetCurrentThread
-    as a chunk of highly optimized machine-specific code at runtime.
-
-    Check the difference between CorUnix::InternalGetCurrentThread and
-    CorUnix::InternalGetCurrentThreadSlow to see the C/C++ code that matches
-    the code generated by this function.
---*/
-PAL_POPTIMIZEDTLSGETTER
-CorUnix::TLSMakeOptimizedGetter(
-        IN CPalThread* pThread,
-        IN DWORD dwTlsIndex)
-{
-#ifdef BIT64
-#pragma unused(pThread, dwTlsIndex)
-    ERROR("TLSMakeOptimizedGetter not rewritten for amd64 yet.");
-    return NULL;
-#else
-    PAL_POPTIMIZEDTLSGETTER Ret = NULL;
-    BYTE* p;
-    int i = 0;
-
-#ifdef __APPLE__
-#define TLS_OPTIMIZED_GETTER_SIZE 118
-#else
-#define TLS_OPTIMIZED_GETTER_SIZE 115
-#endif
-
-    p = (BYTE*)InternalMalloc(pThread, TLS_OPTIMIZED_GETTER_SIZE * sizeof(BYTE));
-
-    if (p == NULL)
-    {
-        return Ret;
-    }
-
-    // Need to preserve %ecx, %edx, and %esi registers as specified in
-    // GetThreadGeneric(void) in vm/amd64/asmhelpers.s
-    p[i++] = 0x51; // push %ecx
-    p[i++] = 0x52; // push %edx
-    p[i++] = 0x89; // mov %esp,%eax // %eax = sp;
-    p[i++] = 0xe0;
-    p[i++] = 0xc1; // shr $0x11,%eax // sp >> 17;
-    p[i++] = 0xe8;
-    p[i++] = 0x11;
-    p[i++] = 0x89; // mov %eax,%edx // key = sp >> 17;
-    p[i++] = 0xc2;
-    p[i++] = 0xc1; // sar $0x7,%edx // key >> 7;
-    p[i++] = 0xfa;
-    p[i++] = 0x07;
-    p[i++] = 0x29; // sub %edx,%eax // key -= key >> 7;
-    p[i++] = 0xd0;
-    p[i++] = 0x89; // mov %eax,%edx
-    p[i++] = 0xc2;
-    p[i++] = 0xc1; // sar $0x5,%edx // key >> 5;
-    p[i++] = 0xfa;
-    p[i++] = 0x05;
-    p[i++] = 0x29; // sub %edx,%eax // key -= key >> 5;
-    p[i++] = 0xd0;
-    p[i++] = 0x89; // mov %eax,%edx
-    p[i++] = 0xc2;
-    p[i++] = 0xc1; // sar $0x3,%edx // key >> 3;
-    p[i++] = 0xfa;
-    p[i++] = 0x03;
-    p[i++] = 0x29; // sub %edx,%eax // key -= key >> 3;
-    p[i++] = 0xd0;
-    p[i++] = 0x25; // and $0xff,%eax // key &= 0xFF;
-    p[i++] = 0xff;
-    p[i++] = 0x00;
-    p[i++] = 0x00;
-    p[i++] = 0x00;
-    p[i++] = 0x8b; // mov (flush_counter),%ecx // %ecx = counter = flush_counter;
-    p[i++] = 0x0d;
-    *((DWORD*) &p[i]) = (DWORD)&flush_counter;
-    i += sizeof(DWORD);
-    p[i++] = 0x8b; // mov (thread_hints,%eax,4),%eax // %edx = pThread = thread_hints[key];
-    p[i++] = 0x14;
-    p[i++] = 0x85;
-    *((DWORD*) &p[i]) = (DWORD)&thread_hints;
-    i += sizeof(DWORD);
-    p[i++] = 0x39; // cmp %esp,offsetof(CPalThread,tlsInfo)+offsetof(CThreadTLSInfo,minStack)(%edx)
-                   // if ((size_t)pThread->tlsInfo.minStack <= sp)
-    p[i++] = 0xa2;
-    *((DWORD*) &p[i]) = (DWORD)(PAL_safe_offsetof(CPalThread,tlsInfo)+PAL_safe_offsetof(CThreadTLSInfo,minStack));
-    i += sizeof(DWORD);
-    p[i++] = 0x77; // ja CallInternalGetCurrentThreadSlow:
-    p[i++] = 0x19;
-    p[i++] = 0x3b; // cmp offsetof(CPalThread,tlsInfo)+offsetof(CThreadTLSInfo,maxStack)(%edx),%esp
-                   // if (sp < (size_t)pThread->tlsInfo.maxStack)
-    p[i++] = 0xa2;
-    *((DWORD*) &p[i]) = (DWORD)(PAL_safe_offsetof(CPalThread,tlsInfo)+PAL_safe_offsetof(CThreadTLSInfo,maxStack));
-    i += sizeof(DWORD);
-    p[i++] = 0x73; // jae CallInternalGetCurrentThreadSlow:
-    p[i++] = 0x11;
-    p[i++] = 0x39; // cmp (flush_counter),%ecx // if (counter == flush_counter)
-    p[i++] = 0x0d;
-    *((DWORD*) &p[i]) = (DWORD)&flush_counter;
-    i += sizeof(DWORD);
-    p[i++] = 0x75; // jne CallInternalGetCurrentThreadSlow:
-    p[i++] = 0x09;
-    if (dwTlsIndex != THREAD_OBJECT_TLS_INDEX)
-    {
-        p[i++] = 0x8b; // mov offsetof(pThread->tlsSlots[dwTlsIndex])(%edx),%eax // %eax = pThread->tlsSlots[dwTlsIndex];
-        p[i++] = 0x82;
-        *((DWORD*) &p[i]) = (DWORD)(PAL_safe_offsetof(CPalThread,tlsInfo)+PAL_safe_offsetof(CThreadTLSInfo,tlsSlots[dwTlsIndex]));
-        i += sizeof(DWORD);
-    }
-    else
-    {
-        p[i++] = 0x89; // mov %edx,%eax // %eax = pThread;
-        p[i++] = 0xd0;
-        p[i++] = 0x90; // nop
-        p[i++] = 0x90; // nop
-        p[i++] = 0x90; // nop
-        p[i++] = 0x90; // nop
-    }
-    p[i++] = 0x5a; // pop %edx
-    p[i++] = 0x59; // pop %ecx
-    p[i++] = 0xc3; // ret
-    // CallInternalGetCurrentThreadSlow:
-    p[i++] = 0x5a; // pop %edx
-    p[i++] = 0x59; // pop %ecx
-    p[i++] = 0x8d; // lea (thread_hints,%eax,4),%eax // %eax = &thread_hints[key];
-    p[i++] = 0x04;
-    p[i++] = 0x85;
-    *((DWORD*) &p[i]) = (DWORD)&thread_hints;
-    i += sizeof(DWORD);
-    p[i++] = 0x55; // push %ebp
-    p[i++] = 0x89; // mov %esp,%ebp
-    p[i++] = 0xe5;
-    p[i++] = 0x51; // push %ecx
-    p[i++] = 0x89; // mov %esp,%ecx // this is the reference esp - need to match the reference esp used in the fast path.
-    p[i++] = 0xe1;
-    p[i++] = 0x52; // push %edx
-#ifdef __APPLE__
-    // establish 16-byte stack alignment
-    p[i++] = 0x83; // subl $8,%esp
-    p[i++] = 0xec;
-    p[i++] = 0x08;
-#endif
-    p[i++] = 0x50; // push %eax // store &thread_hints[key] on stack as 2nd argument;
-    p[i++] = 0x51; // push %ecx // reference esp - The 1st argument for call to InternalGetCurrentThreadSlow.
-    p[i++] = 0xe8; // call InternalGetCurrentThreadSlow
-    *((DWORD*) &p[i]) = (DWORD)&InternalGetCurrentThreadSlow - (DWORD)(&p[i+sizeof(DWORD)]);
-    i += sizeof(DWORD);
-#ifdef __APPLE__
-    p[i++] = 0x83; // addl $16,%esp
-    p[i++] = 0xc4;
-    p[i++] = 0x10;
-#else
-    p[i++] = 0x83; // addl $8,%esp
-    p[i++] = 0xc4;
-    p[i++] = 0x08;
-#endif
-    if (dwTlsIndex != THREAD_OBJECT_TLS_INDEX)
-    {
-        p[i++] = 0x8b; // mov offsetof(pThread->tlsSlots[dwTlsIndex])(%eax),%eax  // %eax = pThread->tlsSlots[dwTlsIndex];
-        p[i++] = 0x80;
-        *((DWORD*) &p[i]) = (DWORD)(PAL_safe_offsetof(CPalThread,tlsInfo)+PAL_safe_offsetof(CThreadTLSInfo,tlsSlots[dwTlsIndex]));
-        i += sizeof(DWORD);
-    }
-    p[i++] = 0x5a; // pop %edx
-    p[i++] = 0x59; // pop %ecx
-    p[i++] = 0xc9; // leave
-    p[i++] = 0xc3; // ret
-
-    if (i > TLS_OPTIMIZED_GETTER_SIZE)
-    {
-        ASSERT("Invalid TLS_OPTIMIZED_GETTER_SIZE %d\n", i);
-    }
-
-    DBG_FlushInstructionCache(p, TLS_OPTIMIZED_GETTER_SIZE * sizeof(BYTE));
-
-    Ret = (PAL_POPTIMIZEDTLSGETTER)p;
-
-    return Ret;
-#endif // BIT64 else
-}
-
-/*++
-Function:
-    TLSFreeOptimizedGetter
-
-    Frees a function created by MakeOptimizedTlsGetter().
---*/
-VOID
-CorUnix::TLSFreeOptimizedGetter(
-        IN PAL_POPTIMIZEDTLSGETTER pOptimizedTlsGetter)
-{
-    InternalFree(InternalGetCurrentThread(), (void *)pOptimizedTlsGetter);
-}
-
-#endif // USE_OPTIMIZEDTLSGETTER
diff --git a/src/pal/src/arch/amd64/signalhandlerhelper.cpp b/src/pal/src/arch/amd64/signalhandlerhelper.cpp
new file mode 100644
index 0000000000..8789f5a622
--- /dev/null
+++ b/src/pal/src/arch/amd64/signalhandlerhelper.cpp
@@ -0,0 +1,70 @@
+// Licensed to the .NET Foundation under one or more agreements.
+// The .NET Foundation licenses this file to you under the MIT license.
+// See the LICENSE file in the project root for more information.
+
+#include "pal/dbgmsg.h"
+SET_DEFAULT_DEBUG_CHANNEL(EXCEPT); // some headers have code with asserts, so do this first
+
+#include "pal/palinternal.h"
+#include "pal/context.h"
+#include "pal/signal.hpp"
+#include "pal/utils.h"
+#include <sys/ucontext.h>
+
+/*++
+Function :
+    signal_handler_worker
+
+    Handles signal on the original stack where the signal occured. 
+    Invoked via setcontext.
+
+Parameters :
+    POSIX signal handler parameter list ("man sigaction" for details)
+    returnPoint - context to which the function returns if the common_signal_handler returns
+
+    (no return value)
+--*/
+void ExecuteHandlerOnOriginalStack(int code, siginfo_t *siginfo, void *context, SignalHandlerWorkerReturnPoint* returnPoint)
+{
+    ucontext_t *ucontext = (ucontext_t *)context;
+    size_t faultSp = (size_t)MCREG_Rsp(ucontext->uc_mcontext);
+
+    _ASSERTE(IS_ALIGNED(faultSp, 8));
+
+    size_t fakeFrameReturnAddress;
+
+    if (IS_ALIGNED(faultSp, 16))
+    {
+        fakeFrameReturnAddress = (size_t)SignalHandlerWorkerReturnOffset0 + (size_t)CallSignalHandlerWrapper0;
+    }
+    else
+    {
+        fakeFrameReturnAddress = (size_t)SignalHandlerWorkerReturnOffset8 + (size_t)CallSignalHandlerWrapper8;
+    }
+
+    // preserve 128 bytes long red zone and align stack pointer
+    size_t* sp = (size_t*)ALIGN_DOWN(faultSp - 128, 16);
+
+    // Build fake stack frame to enable the stack unwinder to unwind from signal_handler_worker to the faulting instruction
+    *--sp = (size_t)MCREG_Rip(ucontext->uc_mcontext);
+    *--sp = (size_t)MCREG_Rbp(ucontext->uc_mcontext);
+    size_t fp = (size_t)sp;
+    *--sp = fakeFrameReturnAddress;
+
+    // Switch the current context to the signal_handler_worker and the original stack
+    CONTEXT context2;
+    RtlCaptureContext(&context2);
+
+    // We don't care about the other registers state since the stack unwinding restores
+    // them for the target frame directly from the signal context.
+    context2.Rsp = (size_t)sp;
+    context2.Rbx = (size_t)faultSp;
+    context2.Rbp = (size_t)fp;
+    context2.Rip = (size_t)signal_handler_worker;
+    context2.Rdi = code;
+    context2.Rsi = (size_t)siginfo;
+    context2.Rdx = (size_t)context;
+    context2.Rcx = (size_t)returnPoint;
+
+    RtlRestoreContext(&context2, NULL);
+}
diff --git a/src/pal/src/arch/arm/callsignalhandlerwrapper.S b/src/pal/src/arch/arm/callsignalhandlerwrapper.S
new file mode 100644
index 0000000000..266e4fdfe9
--- /dev/null
+++ b/src/pal/src/arch/arm/callsignalhandlerwrapper.S
@@ -0,0 +1,32 @@
+// Licensed to the .NET Foundation under one or more agreements.
+// The .NET Foundation licenses this file to you under the MIT license.
+// See the LICENSE file in the project root for more information.
+
+#include "unixasmmacros.inc"
+#include "asmconstants.h"
+
+.syntax unified
+.thumb
+
+.macro CALL_SIGNAL_HANDLER_WRAPPER Alignment
+
+.globl C_FUNC(SignalHandlerWorkerReturnOffset\Alignment)
+C_FUNC(SignalHandlerWorkerReturnOffset\Alignment):
+    .int LOCAL_LABEL(SignalHandlerWorkerReturn\Alignment)-C_FUNC(CallSignalHandlerWrapper\Alignment)
+
+// This function is never called, only a fake stack frame will be setup to have a return
+// address set to SignalHandlerWorkerReturn during SIGSEGV handling.
+// It enables the unwinder to unwind stack from the handling code to the actual failure site.
+NESTED_ENTRY CallSignalHandlerWrapper\Alignment, _TEXT, NoHandler
+    sub     sp, sp, #(8 + \Alignment) // red zone + alignment
+    stmfd   sp!, {r7, lr}
+    bl      EXTERNAL_C_FUNC(signal_handler_worker)
+LOCAL_LABEL(SignalHandlerWorkerReturn\Alignment):
+    ldmfd   sp!, {r7, lr}
+    bx      lr
+NESTED_END CallSignalHandlerWrapper\Alignment, _TEXT
+
+.endm
+
+CALL_SIGNAL_HANDLER_WRAPPER 0
+CALL_SIGNAL_HANDLER_WRAPPER 4
diff --git a/src/pal/src/arch/arm/signalhandlerhelper.cpp b/src/pal/src/arch/arm/signalhandlerhelper.cpp
new file mode 100644
index 0000000000..e1ad460905
--- /dev/null
+++ b/src/pal/src/arch/arm/signalhandlerhelper.cpp
@@ -0,0 +1,70 @@
+// Licensed to the .NET Foundation under one or more agreements.
+// The .NET Foundation licenses this file to you under the MIT license.
+// See the LICENSE file in the project root for more information.
+
+#include "pal/dbgmsg.h"
+SET_DEFAULT_DEBUG_CHANNEL(EXCEPT); // some headers have code with asserts, so do this first
+
+#include "pal/palinternal.h"
+#include "pal/context.h"
+#include "pal/signal.hpp"
+#include "pal/utils.h"
+#include <sys/ucontext.h>
+
+/*++
+Function :
+    signal_handler_worker
+
+    Handles signal on the original stack where the signal occured. 
+    Invoked via setcontext.
+
+Parameters :
+    POSIX signal handler parameter list ("man sigaction" for details)
+    returnPoint - context to which the function returns if the common_signal_handler returns
+
+    (no return value)
+--*/
+void ExecuteHandlerOnOriginalStack(int code, siginfo_t *siginfo, void *context, SignalHandlerWorkerReturnPoint* returnPoint)
+{
+    ucontext_t *ucontext = (ucontext_t *)context;
+    size_t faultSp = (size_t)MCREG_Sp(ucontext->uc_mcontext);
+
+    _ASSERTE(IS_ALIGNED(faultSp, 4));
+
+    size_t fakeFrameReturnAddress;
+
+    if (IS_ALIGNED(faultSp, 8))
+    {
+        fakeFrameReturnAddress = (size_t)SignalHandlerWorkerReturnOffset0 + (size_t)CallSignalHandlerWrapper0;
+    }
+    else
+    {
+        fakeFrameReturnAddress = (size_t)SignalHandlerWorkerReturnOffset4 + (size_t)CallSignalHandlerWrapper4;
+    }
+
+    // preserve 8 bytes long red zone and align stack pointer
+    size_t* sp = (size_t*)ALIGN_DOWN(faultSp - 8, 8);
+
+    // Build fake stack frame to enable the stack unwinder to unwind from signal_handler_worker to the faulting instruction
+    // pushed LR
+    *--sp = (size_t)MCREG_Pc(ucontext->uc_mcontext);
+    // pushed frame pointer
+    *--sp = (size_t)MCREG_R7(ucontext->uc_mcontext); 
+
+    // Switch the current context to the signal_handler_worker and the original stack
+    CONTEXT context2;
+    RtlCaptureContext(&context2);
+
+    // We don't care about the other registers state since the stack unwinding restores
+    // them for the target frame directly from the signal context.
+    context2.Sp = (size_t)sp;
+    context2.R7 = (size_t)sp; // Fp and Sp are the same
+    context2.Lr = fakeFrameReturnAddress;
+    context2.Pc = (size_t)signal_handler_worker;
+    context2.R0 = code;
+    context2.R1 = (size_t)siginfo;
+    context2.R2 = (size_t)context;
+    context2.R3 = (size_t)returnPoint;
+
+    RtlRestoreContext(&context2, NULL);
+}
diff --git a/src/pal/src/arch/arm64/asmconstants.h b/src/pal/src/arch/arm64/asmconstants.h
new file mode 100644
index 0000000000..b2bf74461f
--- /dev/null
+++ b/src/pal/src/arch/arm64/asmconstants.h
@@ -0,0 +1,95 @@
+// Licensed to the .NET Foundation under one or more agreements.
+// The .NET Foundation licenses this file to you under the MIT license.
+// See the LICENSE file in the project root for more information.
+
+#ifndef __PAL_ARM64_ASMCONSTANTS_H__
+#define __PAL_ARM64_ASMCONSTANTS_H__
+
+#define CONTEXT_ARM64   0x00400000L
+
+#define CONTEXT_CONTROL_BIT (0)
+#define CONTEXT_INTEGER_BIT (1)
+#define CONTEXT_FLOATING_POINT_BIT (2)
+#define CONTEXT_DEBUG_REGISTERS_BIT (3)
+
+#define CONTEXT_CONTROL (CONTEXT_ARM64 | (1L << CONTEXT_CONTROL_BIT))
+#define CONTEXT_INTEGER (CONTEXT_ARM64 | (1 << CONTEXT_INTEGER_BIT))
+#define CONTEXT_FLOATING_POINT  (CONTEXT_ARM64 | (1 << CONTEXT_FLOATING_POINT_BIT))
+#define CONTEXT_DEBUG_REGISTERS (CONTEXT_ARM64 | (1 << CONTEXT_DEBUG_REGISTERS_BIT))
+
+#define CONTEXT_FULL (CONTEXT_CONTROL | CONTEXT_INTEGER | CONTEXT_FLOATING_POINT)
+
+
+#define CONTEXT_ContextFlags 0
+#define CONTEXT_Cpsr         CONTEXT_ContextFlags+4
+#define CONTEXT_X0           CONTEXT_Cpsr+4
+#define CONTEXT_X1           CONTEXT_X0+8
+#define CONTEXT_X2           CONTEXT_X1+8
+#define CONTEXT_X3           CONTEXT_X2+8
+#define CONTEXT_X4           CONTEXT_X3+8
+#define CONTEXT_X5           CONTEXT_X4+8
+#define CONTEXT_X6           CONTEXT_X5+8
+#define CONTEXT_X7           CONTEXT_X6+8
+#define CONTEXT_X8           CONTEXT_X7+8
+#define CONTEXT_X9           CONTEXT_X8+8
+#define CONTEXT_X10          CONTEXT_X9+8
+#define CONTEXT_X11          CONTEXT_X10+8
+#define CONTEXT_X12          CONTEXT_X11+8
+#define CONTEXT_X13          CONTEXT_X12+8
+#define CONTEXT_X14          CONTEXT_X13+8
+#define CONTEXT_X15          CONTEXT_X14+8
+#define CONTEXT_X16          CONTEXT_X15+8
+#define CONTEXT_X17          CONTEXT_X16+8
+#define CONTEXT_X18          CONTEXT_X17+8
+#define CONTEXT_X19          CONTEXT_X18+8
+#define CONTEXT_X20          CONTEXT_X19+8
+#define CONTEXT_X21          CONTEXT_X20+8
+#define CONTEXT_X22          CONTEXT_X21+8
+#define CONTEXT_X23          CONTEXT_X22+8
+#define CONTEXT_X24          CONTEXT_X23+8
+#define CONTEXT_X25          CONTEXT_X24+8
+#define CONTEXT_X26          CONTEXT_X25+8
+#define CONTEXT_X27          CONTEXT_X26+8
+#define CONTEXT_X28          CONTEXT_X27+8
+#define CONTEXT_Fp           CONTEXT_X28+8
+#define CONTEXT_Lr           CONTEXT_Fp+8
+#define CONTEXT_Sp           CONTEXT_Lr+8
+#define CONTEXT_Pc           CONTEXT_Sp+8
+#define CONTEXT_NEON_OFFSET  CONTEXT_Pc+8
+#define CONTEXT_V0           0
+#define CONTEXT_V1           CONTEXT_V0+16
+#define CONTEXT_V2           CONTEXT_V1+16
+#define CONTEXT_V3           CONTEXT_V2+16
+#define CONTEXT_V4           CONTEXT_V3+16
+#define CONTEXT_V5           CONTEXT_V4+16
+#define CONTEXT_V6           CONTEXT_V5+16
+#define CONTEXT_V7           CONTEXT_V6+16
+#define CONTEXT_V8           CONTEXT_V7+16
+#define CONTEXT_V9           CONTEXT_V8+16
+#define CONTEXT_V10          CONTEXT_V9+16
+#define CONTEXT_V11          CONTEXT_V10+16
+#define CONTEXT_V12          CONTEXT_V11+16
+#define CONTEXT_V13          CONTEXT_V12+16
+#define CONTEXT_V14          CONTEXT_V13+16
+#define CONTEXT_V15          CONTEXT_V14+16
+#define CONTEXT_V16          CONTEXT_V15+16
+#define CONTEXT_V17          CONTEXT_V16+16
+#define CONTEXT_V18          CONTEXT_V17+16
+#define CONTEXT_V19          CONTEXT_V18+16
+#define CONTEXT_V20          CONTEXT_V19+16
+#define CONTEXT_V21          CONTEXT_V20+16
+#define CONTEXT_V22          CONTEXT_V21+16
+#define CONTEXT_V23          CONTEXT_V22+16
+#define CONTEXT_V24          CONTEXT_V23+16
+#define CONTEXT_V25          CONTEXT_V24+16
+#define CONTEXT_V26          CONTEXT_V25+16
+#define CONTEXT_V27          CONTEXT_V26+16
+#define CONTEXT_V28          CONTEXT_V27+16
+#define CONTEXT_V29          CONTEXT_V28+16
+#define CONTEXT_V30          CONTEXT_V29+16
+#define CONTEXT_V31          CONTEXT_V30+16
+#define CONTEXT_FLOAT_CONTROL_OFFSET  CONTEXT_V31
+#define CONTEXT_Fpcr         0
+#define CONTEXT_Fpsr         CONTEXT_Fpcr+4
+
+#endif
diff --git a/src/pal/src/arch/arm64/callsignalhandlerwrapper.S b/src/pal/src/arch/arm64/callsignalhandlerwrapper.S
new file mode 100644
index 0000000000..90fb602479
--- /dev/null
+++ b/src/pal/src/arch/arm64/callsignalhandlerwrapper.S
@@ -0,0 +1,32 @@
+// Licensed to the .NET Foundation under one or more agreements.
+// The .NET Foundation licenses this file to you under the MIT license.
+// See the LICENSE file in the project root for more information.
+
+#include "unixasmmacros.inc"
+#include "asmconstants.h"
+
+.macro CALL_SIGNAL_HANDLER_WRAPPER Alignment
+
+.globl C_FUNC(SignalHandlerWorkerReturnOffset\Alignment)
+C_FUNC(SignalHandlerWorkerReturnOffset\Alignment):
+    .int LOCAL_LABEL(SignalHandlerWorkerReturn\Alignment)-C_FUNC(CallSignalHandlerWrapper\Alignment)
+
+// This function is never called, only a fake stack frame will be setup to have a return
+// address set to SignalHandlerWorkerReturn during SIGSEGV handling.
+// It enables the unwinder to unwind stack from the handling code to the actual failure site.
+NESTED_ENTRY CallSignalHandlerWrapper\Alignment, _TEXT, NoHandler
+__StackAllocationSize = (128 + 8 + 8 + \Alignment) // red zone + fp + lr + alignment 
+    PROLOG_STACK_ALLOC __StackAllocationSize
+    .cfi_adjust_cfa_offset __StackAllocationSize
+    PROLOG_SAVE_REG_PAIR fp, lr, 0
+    bl      EXTERNAL_C_FUNC(signal_handler_worker)
+LOCAL_LABEL(SignalHandlerWorkerReturn\Alignment):
+    EPILOG_RESTORE_REG_PAIR fp, lr, 0
+    EPILOG_STACK_FREE __StackAllocationSize
+    ret
+NESTED_END CallSignalHandlerWrapper\Alignment, _TEXT
+
+.endm
+
+CALL_SIGNAL_HANDLER_WRAPPER 0
+CALL_SIGNAL_HANDLER_WRAPPER 8
diff --git a/src/pal/src/arch/arm64/context2.S b/src/pal/src/arch/arm64/context2.S
index a64e62c94d..e62a9ac4d9 100644
--- a/src/pal/src/arch/arm64/context2.S
+++ b/src/pal/src/arch/arm64/context2.S
@@ -8,87 +8,7 @@
 //
 
 #include "unixasmmacros.inc"
-
-#define CONTEXT_ARM64   0x00400000L
-
-#define CONTEXT_CONTROL (CONTEXT_ARM64 | 0x1L)
-#define CONTEXT_INTEGER (CONTEXT_ARM64 | 0x2L)
-#define CONTEXT_FLOATING_POINT  (CONTEXT_ARM64 | 0x4L)
-#define CONTEXT_DEBUG_REGISTERS (CONTEXT_ARM64 | 0x8L)
-
-#define CONTEXT_FULL (CONTEXT_CONTROL | CONTEXT_INTEGER | CONTEXT_FLOATING_POINT)
-
-#define CONTEXT_ContextFlags 0
-#define CONTEXT_Cpsr         CONTEXT_ContextFlags+4
-#define CONTEXT_X0           CONTEXT_Cpsr+4
-#define CONTEXT_X1           CONTEXT_X0+8
-#define CONTEXT_X2           CONTEXT_X1+8
-#define CONTEXT_X3           CONTEXT_X2+8
-#define CONTEXT_X4           CONTEXT_X3+8
-#define CONTEXT_X5           CONTEXT_X4+8
-#define CONTEXT_X6           CONTEXT_X5+8
-#define CONTEXT_X7           CONTEXT_X6+8
-#define CONTEXT_X8           CONTEXT_X7+8
-#define CONTEXT_X9           CONTEXT_X8+8
-#define CONTEXT_X10          CONTEXT_X9+8
-#define CONTEXT_X11          CONTEXT_X10+8
-#define CONTEXT_X12          CONTEXT_X11+8
-#define CONTEXT_X13          CONTEXT_X12+8
-#define CONTEXT_X14          CONTEXT_X13+8
-#define CONTEXT_X15          CONTEXT_X14+8
-#define CONTEXT_X16          CONTEXT_X15+8
-#define CONTEXT_X17          CONTEXT_X16+8
-#define CONTEXT_X18          CONTEXT_X17+8
-#define CONTEXT_X19          CONTEXT_X18+8
-#define CONTEXT_X20          CONTEXT_X19+8
-#define CONTEXT_X21          CONTEXT_X20+8
-#define CONTEXT_X22          CONTEXT_X21+8
-#define CONTEXT_X23          CONTEXT_X22+8
-#define CONTEXT_X24          CONTEXT_X23+8
-#define CONTEXT_X25          CONTEXT_X24+8
-#define CONTEXT_X26          CONTEXT_X25+8
-#define CONTEXT_X27          CONTEXT_X26+8
-#define CONTEXT_X28          CONTEXT_X27+8
-#define CONTEXT_Fp           CONTEXT_X28+8
-#define CONTEXT_Lr           CONTEXT_Fp+8
-#define CONTEXT_Sp           CONTEXT_Lr+8
-#define CONTEXT_Pc           CONTEXT_Sp+8
-#define CONTEXT_NEON_OFFSET  CONTEXT_Pc+8
-#define CONTEXT_V0           0
-#define CONTEXT_V1           CONTEXT_V0+16
-#define CONTEXT_V2           CONTEXT_V1+16
-#define CONTEXT_V3           CONTEXT_V2+16
-#define CONTEXT_V4           CONTEXT_V3+16
-#define CONTEXT_V5           CONTEXT_V4+16
-#define CONTEXT_V6           CONTEXT_V5+16
-#define CONTEXT_V7           CONTEXT_V6+16
-#define CONTEXT_V8           CONTEXT_V7+16
-#define CONTEXT_V9           CONTEXT_V8+16
-#define CONTEXT_V10          CONTEXT_V9+16
-#define CONTEXT_V11          CONTEXT_V10+16
-#define CONTEXT_V12          CONTEXT_V11+16
-#define CONTEXT_V13          CONTEXT_V12+16
-#define CONTEXT_V14          CONTEXT_V13+16
-#define CONTEXT_V15          CONTEXT_V14+16
-#define CONTEXT_V16          CONTEXT_V15+16
-#define CONTEXT_V17          CONTEXT_V16+16
-#define CONTEXT_V18          CONTEXT_V17+16
-#define CONTEXT_V19          CONTEXT_V18+16
-#define CONTEXT_V20          CONTEXT_V19+16
-#define CONTEXT_V21          CONTEXT_V20+16
-#define CONTEXT_V22          CONTEXT_V21+16
-#define CONTEXT_V23          CONTEXT_V22+16
-#define CONTEXT_V24          CONTEXT_V23+16
-#define CONTEXT_V25          CONTEXT_V24+16
-#define CONTEXT_V26          CONTEXT_V25+16
-#define CONTEXT_V27          CONTEXT_V26+16
-#define CONTEXT_V28          CONTEXT_V27+16
-#define CONTEXT_V29          CONTEXT_V28+16
-#define CONTEXT_V30          CONTEXT_V29+16
-#define CONTEXT_V31          CONTEXT_V30+16
-#define CONTEXT_FLOAT_CONTROL_OFFSET  CONTEXT_V31
-#define CONTEXT_Fpcr         0
-#define CONTEXT_Fpsr         CONTEXT_Fpcr+4
+#include "asmconstants.h"
 
 // Incoming:
 //  x0: Context*
@@ -115,10 +35,8 @@ LEAF_ENTRY CONTEXT_CaptureContext, _TEXT
     ldr x2, [sp, 24]
     str w2, [x0, CONTEXT_Cpsr]
     stp fp, lr, [x0, CONTEXT_Fp]
-    add sp, sp, #32
-    mov x2, sp
+    add x2, sp, #32
     stp x2, lr, [x0, CONTEXT_Sp]
-    sub sp, sp, #32
 
 LOCAL_LABEL(Done_CONTEXT_CONTROL): 
     // we dont clobber x1 in the CONTEXT_CONTROL case
@@ -224,14 +142,8 @@ LEAF_ENTRY RtlRestoreContext, _TEXT
     // since we potentially clobber x0 below, we'll bank it in x16
     mov x16, x0
 
-    ldr w2, [x16, CONTEXT_ContextFlags]
-    // clangs assembler doesn't seem to support the mov Wx, imm32 yet
-    movz w3, #0x40, lsl #16
-    movk w3, #0x4
-    mov w4, w3
-    and w3, w2, w3
-    cmp w3, w4
-    b.ne LOCAL_LABEL(No_Restore_CONTEXT_FLOATING_POINT)
+    ldr w17, [x16, CONTEXT_ContextFlags]
+    tbz w17, #CONTEXT_FLOATING_POINT_BIT, LOCAL_LABEL(No_Restore_CONTEXT_FLOATING_POINT)
 
     add x16, x16,   CONTEXT_NEON_OFFSET
     ldp q0, q1,   [x16, CONTEXT_V0]
@@ -256,12 +168,7 @@ LEAF_ENTRY RtlRestoreContext, _TEXT
     sub x16, x16,   CONTEXT_NEON_OFFSET
 
 LOCAL_LABEL(No_Restore_CONTEXT_FLOATING_POINT):
-    movz w2, #0x40, lsl #16
-    movk w2, #0x2
-    mov w3, w2
-    and w2, w1, w2
-    cmp w2, w3
-    b.ne LOCAL_LABEL(No_Restore_CONTEXT_INTEGER)
+    tbz w17, #CONTEXT_INTEGER_BIT, LOCAL_LABEL(No_Restore_CONTEXT_INTEGER)
 
     ldp x0, x1,   [x16, CONTEXT_X0]
     ldp x2, x3,   [x16, CONTEXT_X2]
@@ -279,12 +186,7 @@ LOCAL_LABEL(No_Restore_CONTEXT_FLOATING_POINT):
     ldr x28,      [x16, CONTEXT_X28]
 
 LOCAL_LABEL(No_Restore_CONTEXT_INTEGER):
-    movz w2, #0x40, lsl #16
-    movk w2, #0x2
-    mov w3, w2
-    and w2, w1, w2
-    cmp w2, w3
-    b.ne LOCAL_LABEL(No_Restore_CONTEXT_CONTROL)
+    tbz w17, #CONTEXT_CONTROL_BIT, LOCAL_LABEL(No_Restore_CONTEXT_CONTROL)
 
     ldr w17, [x16, CONTEXT_Cpsr]
     msr nzcv, x17
@@ -293,8 +195,8 @@ LOCAL_LABEL(No_Restore_CONTEXT_INTEGER):
     mov sp, x17
     ldr x17, [x16, CONTEXT_Pc]
     br x17
-
+ 
 LOCAL_LABEL(No_Restore_CONTEXT_CONTROL):
-    ret
+   ret
 
 LEAF_END RtlRestoreContext, _TEXT
diff --git a/src/pal/src/arch/arm64/exceptionhelper.S b/src/pal/src/arch/arm64/exceptionhelper.S
index 4fdcfc5eb1..480846eb61 100644
--- a/src/pal/src/arch/arm64/exceptionhelper.S
+++ b/src/pal/src/arch/arm64/exceptionhelper.S
@@ -3,7 +3,30 @@
 // See the LICENSE file in the project root for more information.
 
 #include "unixasmmacros.inc"
+#include "asmconstants.h"
 
+//////////////////////////////////////////////////////////////////////////
+//
+// This function creates a stack frame right below the target frame, restores all callee
+// saved registers, SP, and LR from the passed in context.
+// Then it uses the ThrowExceptionHelper to throw the passed in exception from that context.
+// EXTERN_C void ThrowExceptionFromContextInternal(CONTEXT* context, PAL_SEHException* ex);
 LEAF_ENTRY ThrowExceptionFromContextInternal, _TEXT
-    EMIT_BREAKPOINT
+    // Save the FP & LR to the stack so that the unwind can work at the instruction after
+    // loading the FP from the context, but before loading the SP from the context.
+    stp fp, lr,  [sp, -16]!
+
+    ldp x19,x20, [x0, #(CONTEXT_X19)]
+    ldp x21,x22, [x0, #(CONTEXT_X21)]
+    ldp x23,x24, [x0, #(CONTEXT_X23)]
+    ldp x24,x25, [x0, #(CONTEXT_X24)]
+    ldp x26,x27, [x0, #(CONTEXT_X26)]
+    ldp x28,fp,  [x0, #(CONTEXT_X28)]
+    ldr lr,      [x0, #(CONTEXT_Pc)]
+    ldr x2,      [x0, #(CONTEXT_Sp)]
+    mov sp, x2
+
+    // The PAL_SEHException pointer
+    mov x0, x1
+    b   EXTERNAL_C_FUNC(ThrowExceptionHelper)
 LEAF_END ThrowExceptionFromContextInternal, _TEXT
diff --git a/src/pal/src/arch/arm64/signalhandlerhelper.cpp b/src/pal/src/arch/arm64/signalhandlerhelper.cpp
new file mode 100644
index 0000000000..c35c629ab3
--- /dev/null
+++ b/src/pal/src/arch/arm64/signalhandlerhelper.cpp
@@ -0,0 +1,67 @@
+// Licensed to the .NET Foundation under one or more agreements.
+// The .NET Foundation licenses this file to you under the MIT license.
+// See the LICENSE file in the project root for more information.
+
+#include "pal/dbgmsg.h"
+SET_DEFAULT_DEBUG_CHANNEL(EXCEPT); // some headers have code with asserts, so do this first
+
+#include "pal/palinternal.h"
+#include "pal/context.h"
+#include "pal/signal.hpp"
+#include "pal/utils.h"
+#include <sys/ucontext.h>
+
+/*++
+Function :
+    signal_handler_worker
+
+    Handles signal on the original stack where the signal occured. 
+    Invoked via setcontext.
+
+Parameters :
+    POSIX signal handler parameter list ("man sigaction" for details)
+    returnPoint - context to which the function returns if the common_signal_handler returns
+
+    (no return value)
+--*/
+void ExecuteHandlerOnOriginalStack(int code, siginfo_t *siginfo, void *context, SignalHandlerWorkerReturnPoint* returnPoint)
+{
+    ucontext_t *ucontext = (ucontext_t *)context;
+    size_t faultSp = (size_t)MCREG_Sp(ucontext->uc_mcontext);
+    _ASSERTE(IS_ALIGNED(faultSp, 8));
+
+    size_t fakeFrameReturnAddress;
+
+    if (IS_ALIGNED(faultSp, 16))
+    {
+        fakeFrameReturnAddress = (size_t)SignalHandlerWorkerReturnOffset0 + (size_t)CallSignalHandlerWrapper0;
+    }
+    else
+    {
+        fakeFrameReturnAddress = (size_t)SignalHandlerWorkerReturnOffset8 + (size_t)CallSignalHandlerWrapper8;
+    }
+
+    // preserve 128 bytes long red zone and align stack pointer
+    size_t* sp = (size_t*)ALIGN_DOWN(faultSp - 128, 16);
+
+    // Build fake stack frame to enable the stack unwinder to unwind from signal_handler_worker to the faulting instruction
+    // pushed LR
+    *--sp = (size_t)MCREG_Pc(ucontext->uc_mcontext);
+    // pushed frame pointer
+    *--sp = (size_t)MCREG_Fp(ucontext->uc_mcontext); 
+
+    // Switch the current context to the signal_handler_worker and the original stack
+    CONTEXT context2;
+    RtlCaptureContext(&context2);
+
+    context2.Sp = (size_t)sp;
+    context2.Fp = (size_t)sp;
+    context2.Lr = fakeFrameReturnAddress;
+    context2.Pc = (size_t)signal_handler_worker;
+    context2.X0 = code;
+    context2.X1 = (size_t)siginfo;
+    context2.X2 = (size_t)context;
+    context2.X3 = (size_t)returnPoint;
+
+    RtlRestoreContext(&context2, NULL);
+}
diff --git a/src/pal/src/arch/i386/asmconstants.h b/src/pal/src/arch/i386/asmconstants.h
index ff763ef16b..d947cb8bcd 100644
--- a/src/pal/src/arch/i386/asmconstants.h
+++ b/src/pal/src/arch/i386/asmconstants.h
@@ -28,3 +28,4 @@
 #define CONTEXT_Xmm5 CONTEXT_Xmm4+16
 #define CONTEXT_Xmm6 CONTEXT_Xmm5+16
 #define CONTEXT_Xmm7 CONTEXT_Xmm6+16
+#define CONTEXT_ResumeEsp CONTEXT_ExtendedRegisters+512
diff --git a/src/pal/src/arch/i386/callsignalhandlerwrapper.S b/src/pal/src/arch/i386/callsignalhandlerwrapper.S
new file mode 100644
index 0000000000..26f06d9886
--- /dev/null
+++ b/src/pal/src/arch/i386/callsignalhandlerwrapper.S
@@ -0,0 +1,47 @@
+// Licensed to the .NET Foundation under one or more agreements.
+// The .NET Foundation licenses this file to you under the MIT license.
+// See the LICENSE file in the project root for more information.
+
+.intel_syntax noprefix
+#include "unixasmmacros.inc"
+#include "asmconstants.h"
+
+.macro CALL_SIGNAL_HANDLER_WRAPPER Alignment
+
+.globl C_FUNC(SignalHandlerWorkerReturnOffset\Alignment)
+C_FUNC(SignalHandlerWorkerReturnOffset\Alignment):
+    .int LOCAL_LABEL(SignalHandlerWorkerReturn\Alignment)-C_FUNC(CallSignalHandlerWrapper\Alignment)
+
+// This function is never called, only a fake stack frame will be setup to have a return
+// address set to SignalHandlerWorkerReturn during SIGSEGV handling.
+// It enables the unwinder to unwind stack from the handling code to the actual failure site.
+NESTED_ENTRY CallSignalHandlerWrapper\Alignment, _TEXT, NoHandler
+
+    .cfi_def_cfa_offset (4 + \Alignment) // return address + stack alignment
+    .cfi_offset eip, -(4 + \Alignment)
+    push ebp
+    .cfi_adjust_cfa_offset 4
+    .cfi_rel_offset ebp, 0
+    mov ebp, esp
+    .cfi_def_cfa_register ebp
+    // Align stack
+    sub  esp, 8
+    // Simulate arguments pushing
+    push eax
+    push eax
+    push eax
+    push eax
+    call    EXTERNAL_C_FUNC(signal_handler_worker)
+LOCAL_LABEL(SignalHandlerWorkerReturn\Alignment):
+    add esp, 4 * 4 + 8
+    pop ebp
+    ret
+
+NESTED_END CallSignalHandlerWrapper\Alignment, _TEXT
+
+.endm
+
+CALL_SIGNAL_HANDLER_WRAPPER 0
+CALL_SIGNAL_HANDLER_WRAPPER 4
+CALL_SIGNAL_HANDLER_WRAPPER 8
+CALL_SIGNAL_HANDLER_WRAPPER 12
diff --git a/src/pal/src/arch/i386/context2.S b/src/pal/src/arch/i386/context2.S
index 11aba5e647..6c31b074cc 100644
--- a/src/pal/src/arch/i386/context2.S
+++ b/src/pal/src/arch/i386/context2.S
@@ -42,6 +42,7 @@ LEAF_ENTRY CONTEXT_CaptureContext, _TEXT
     mov   [eax + CONTEXT_Ebp], ebp
     lea   ebx, [esp + 12]
     mov   [eax + CONTEXT_Esp], ebx
+    mov   [eax + CONTEXT_ResumeEsp], ebx
     mov   ebx, [esp + 8]
     mov   [eax + CONTEXT_Eip], ebx
 
@@ -82,7 +83,7 @@ LOCAL_LABEL(Done_CONTEXT_EXTENDED_REGISTERS):
     // Restore
     pop   ebx
     pop   eax
-    ret   4
+    ret
 LEAF_END CONTEXT_CaptureContext, _TEXT
 
 LEAF_ENTRY RtlCaptureContext, _TEXT
@@ -114,7 +115,7 @@ LOCAL_LABEL(Done_Restore_CONTEXT_FLOATING_POINT):
 LOCAL_LABEL(Done_Restore_CONTEXT_EXTENDED_REGISTERS):
 
     // Restore Stack
-    mov   esp, [eax + CONTEXT_Esp]
+    mov   esp, [eax + CONTEXT_ResumeEsp]
 
     // Create a minimal frame
     push  DWORD PTR [eax + CONTEXT_Eip]
diff --git a/src/pal/src/arch/i386/exceptionhelper.S b/src/pal/src/arch/i386/exceptionhelper.S
index 2061be26f8..bf44124479 100644
--- a/src/pal/src/arch/i386/exceptionhelper.S
+++ b/src/pal/src/arch/i386/exceptionhelper.S
@@ -19,11 +19,14 @@
 
 LEAF_ENTRY ThrowExceptionFromContextInternal, _TEXT
         push  ebp
-        mov   eax, [esp + 12] // ebx: PAL_SEHException *
-        mov   ebx, [esp + 8]  // eax: CONTEXT *
+        mov   ecx, [esp + 12] // ecx: PAL_SEHException * (first argument for ThrowExceptionHelper)
+        mov   eax, [esp + 8]  // ebx: CONTEXT *
 
-        mov   ebp, [ebx + CONTEXT_Ebp]
-        mov   esp, [ebx + CONTEXT_Esp]
+        mov   ebp, [eax + CONTEXT_Ebp]
+        mov   esp, [eax + CONTEXT_ResumeEsp]
+        mov   ebx, [eax + CONTEXT_Ebx]
+        mov   esi, [eax + CONTEXT_Esi]
+        mov   edi, [eax + CONTEXT_Edi]
 
         // The ESP is re-initialized as the target frame's value, so the current function's
         // CFA is now right at the ESP.
@@ -33,11 +36,9 @@ LEAF_ENTRY ThrowExceptionFromContextInternal, _TEXT
         // the EBP is no longer saved in the current stack frame.
         .cfi_restore ebp
 
-        // Store PAL_SEHException as the first argument
-        push    eax
-
         // Store return address to the stack
-        mov     ebx, [ebx + CONTEXT_Eip]
-        push    ebx
+        mov     eax, [eax + CONTEXT_Eip]
+        push    eax
         jmp     EXTERNAL_C_FUNC(ThrowExceptionHelper)
+
 LEAF_END ThrowExceptionFromContextInternal, _TEXT
diff --git a/src/pal/src/arch/i386/signalhandlerhelper.cpp b/src/pal/src/arch/i386/signalhandlerhelper.cpp
new file mode 100644
index 0000000000..3369abe093
--- /dev/null
+++ b/src/pal/src/arch/i386/signalhandlerhelper.cpp
@@ -0,0 +1,78 @@
+// Licensed to the .NET Foundation under one or more agreements.
+// The .NET Foundation licenses this file to you under the MIT license.
+// See the LICENSE file in the project root for more information.
+
+#include "pal/dbgmsg.h"
+SET_DEFAULT_DEBUG_CHANNEL(EXCEPT); // some headers have code with asserts, so do this first
+
+#include "pal/palinternal.h"
+#include "pal/context.h"
+#include "pal/signal.hpp"
+#include "pal/utils.h"
+#include <sys/ucontext.h>
+
+/*++
+Function :
+    signal_handler_worker
+
+    Handles signal on the original stack where the signal occured. 
+    Invoked via setcontext.
+
+Parameters :
+    POSIX signal handler parameter list ("man sigaction" for details)
+    returnPoint - context to which the function returns if the common_signal_handler returns
+
+    (no return value)
+--*/
+void ExecuteHandlerOnOriginalStack(int code, siginfo_t *siginfo, void *context, SignalHandlerWorkerReturnPoint* returnPoint)
+{
+    ucontext_t *ucontext = (ucontext_t *)context;
+    size_t faultSp = (size_t)MCREG_Esp(ucontext->uc_mcontext);
+
+    _ASSERTE(IS_ALIGNED(faultSp, 4));
+
+    size_t fakeFrameReturnAddress;
+
+    switch (faultSp & 0xc)
+    {
+        case 0x0:
+            fakeFrameReturnAddress = (size_t)SignalHandlerWorkerReturnOffset0 + (size_t)CallSignalHandlerWrapper0;
+            break;
+        case 0x4:
+            fakeFrameReturnAddress = (size_t)SignalHandlerWorkerReturnOffset4 + (size_t)CallSignalHandlerWrapper4;
+            break;
+        case 0x8:
+            fakeFrameReturnAddress = (size_t)SignalHandlerWorkerReturnOffset8 + (size_t)CallSignalHandlerWrapper8;
+            break;
+        case 0xc:
+            fakeFrameReturnAddress = (size_t)SignalHandlerWorkerReturnOffset12 + (size_t)CallSignalHandlerWrapper12;
+            break;
+    }
+
+    size_t* sp = (size_t*)ALIGN_DOWN(faultSp, 16);
+
+    // Build fake stack frame to enable the stack unwinder to unwind from signal_handler_worker to the faulting instruction
+    *--sp = (size_t)MCREG_Eip(ucontext->uc_mcontext);
+    *--sp = (size_t)MCREG_Ebp(ucontext->uc_mcontext);
+    size_t fp = (size_t)sp;
+    // Align stack
+    sp -= 2; 
+    *--sp = (size_t)returnPoint;
+    *--sp = (size_t)context;
+    *--sp = (size_t)siginfo;
+    *--sp = code;
+    *--sp = fakeFrameReturnAddress;
+
+    // Switch the current context to the signal_handler_worker and the original stack
+    CONTEXT context2;
+    RtlCaptureContext(&context2);
+
+    // We don't care about the other registers state since the stack unwinding restores
+    // them for the target frame directly from the signal context.
+    context2.Esp = (size_t)sp;
+    context2.ResumeEsp = (size_t)sp;
+    context2.Ebp = (size_t)fp;
+    context2.Eip = (size_t)signal_handler_worker;
+
+    RtlRestoreContext(&context2, NULL);
+}