diff --git a/lld/test/COFF/Inputs/tlssup-32.ll b/lld/test/COFF/Inputs/tlssup-32.ll
new file mode 100644
--- /dev/null
+++ b/lld/test/COFF/Inputs/tlssup-32.ll
@@ -0,0 +1,31 @@
+; We manually create these here if we're not linking against
+; the CRT which would usually provide these.
+
+target triple = "i686-pc-windows-msvc"
+
+%IMAGE_TLS_DIRECTORY32 = type {
+    i32,    ; StartAddressOfRawData
+    i32,    ; EndAddressOfRawData
+    i32,    ; AddressOfIndex
+    i32,    ; AddressOfCallBacks
+    i32,    ; SizeOfZeroFill
+    i32     ; Characteristics
+}
+
+@_tls_start = global i8 zeroinitializer, section ".tls"
+@_tls_end = global i8 zeroinitializer, section ".tls$ZZZ"
+@_tls_index = global i32 0
+
+@_tls_used = global %IMAGE_TLS_DIRECTORY32 {
+    i32 ptrtoint (i8* @_tls_start to i32),
+    i32 ptrtoint (i8* @_tls_end to i32),
+    i32 ptrtoint (i32* @_tls_index to i32),
+    i32 0,
+    i32 0,
+    i32 0
+}, section ".rdata$T"
+
+; MSVC target uses a direct offset (0x58) for x86-64 but expects
+; __tls_array to hold the offset (0x2C) on x86.
+module asm ".global __tls_array"
+module asm "__tls_array = 44"
\ No newline at end of file
diff --git a/lld/test/COFF/Inputs/tlssup-64.ll b/lld/test/COFF/Inputs/tlssup-64.ll
new file mode 100644
--- /dev/null
+++ b/lld/test/COFF/Inputs/tlssup-64.ll
@@ -0,0 +1,26 @@
+; We manually create these here if we're not linking against
+; the CRT which would usually provide these.
+
+target triple = "x86_64-pc-windows-msvc"
+
+%IMAGE_TLS_DIRECTORY64 = type {
+    i64,    ; StartAddressOfRawData
+    i64,    ; EndAddressOfRawData
+    i64,    ; AddressOfIndex
+    i64,    ; AddressOfCallBacks
+    i32,    ; SizeOfZeroFill
+    i32     ; Characteristics
+}
+
+@_tls_start = global i8 zeroinitializer, section ".tls"
+@_tls_end = global i8 zeroinitializer, section ".tls$ZZZ"
+@_tls_index = global i64 0
+
+@_tls_used = global %IMAGE_TLS_DIRECTORY64 {
+    i64 ptrtoint (i8* @_tls_start to i64),
+    i64 ptrtoint (i8* @_tls_end to i64),
+    i64 ptrtoint (i64* @_tls_index to i64),
+    i64 0,
+    i32 0,
+    i32 0
+}, section ".rdata$T"
\ No newline at end of file
diff --git a/lld/test/COFF/tls-alignment-32.ll b/lld/test/COFF/tls-alignment-32.ll
new file mode 100644
--- /dev/null
+++ b/lld/test/COFF/tls-alignment-32.ll
@@ -0,0 +1,23 @@
+; This test is to make sure that the necessary alignment for thread locals
+; gets reflected in the TLS Directory of the generated executable on x86.
+;
+; aligned_thread_local specifies 'align 32' and so the generated
+; exe should reflect that with a value of IMAGE_SCN_ALIGN_32BYTES
+; in the Characteristics field of the IMAGE_TLS_DIRECTORY
+
+; RUN: llc -filetype=obj %S/Inputs/tlssup-32.ll -o %t.tlssup.obj
+; RUN: llc -filetype=obj %s -o %t.obj
+; RUN: lld-link %t.tlssup.obj %t.obj -entry:main -nodefaultlib -out:%t.exe
+; RUN: llvm-readobj --coff-tls-directory %t.exe | FileCheck %s
+
+; CHECK: TLSDirectory {
+; CHECK: Characteristics [ (0x0)
+
+target triple = "i686-pc-windows-msvc"
+
+@aligned_thread_local = thread_local global i32 42, align 32
+
+define i32 @main() {
+  %t = load i32, i32* @aligned_thread_local
+  ret i32 %t
+}
diff --git a/lld/test/COFF/tls-alignment-64.ll b/lld/test/COFF/tls-alignment-64.ll
new file mode 100644
--- /dev/null
+++ b/lld/test/COFF/tls-alignment-64.ll
@@ -0,0 +1,23 @@
+; This test is to make sure that the necessary alignment for thread locals
+; gets reflected in the TLS Directory of the generated executable on x86-64.
+;
+; aligned_thread_local specifies 'align 64' and so the generated
+; exe should reflect that with a value of IMAGE_SCN_ALIGN_64BYTES
+; in the Characteristics field of the IMAGE_TLS_DIRECTORY
+
+; RUN: llc -filetype=obj %S/Inputs/tlssup-64.ll -o %t.tlssup.obj
+; RUN: llc -filetype=obj %s -o %t.obj
+; RUN: lld-link %t.tlssup.obj %t.obj -entry:main -nodefaultlib -out:%t.exe
+; RUN: llvm-readobj --coff-tls-directory %t.exe | FileCheck %s
+
+; CHECK: TLSDirectory {
+; CHECK: Characteristics [ (0x0)
+
+target triple = "x86_64-pc-windows-msvc"
+
+@aligned_thread_local = thread_local global i32 42, align 64
+
+define i32 @main() {
+  %t = load i32, i32* @aligned_thread_local
+  ret i32 %t
+}