diff --git a/lib/libc/gen/tls.c b/lib/libc/gen/tls.c
--- a/lib/libc/gen/tls.c
+++ b/lib/libc/gen/tls.c
@@ -81,20 +81,6 @@
 #error TLS_TCB_ALIGN undefined for target architecture
 #endif
 
-#if defined(__aarch64__) || defined(__arm__) || defined(__mips__) || \
-    defined(__powerpc__) || defined(__riscv)
-#define TLS_VARIANT_I
-#endif
-#if defined(__i386__) || defined(__amd64__)
-#define TLS_VARIANT_II
-#endif
-
-#if defined(__mips__) || defined(__powerpc__) || defined(__riscv)
-#define DTV_OFFSET 0x8000
-#else
-#define DTV_OFFSET 0
-#endif
-
 #ifndef PIC
 
 static size_t libc_tls_static_space;
@@ -297,7 +283,7 @@
 
 		/* Adjust the DTV. */
 		dtv = tcb[0];
-		dtv[2] = (Elf_Addr)(tls + DTV_OFFSET);
+		dtv[2] = (Elf_Addr)(tls + TLS_DTV_OFFSET);
 	} else {
 		dtv = __je_bootstrap_malloc(3 * sizeof(Elf_Addr));
 		if (dtv == NULL) {
@@ -308,7 +294,7 @@
 		tcb[0] = dtv;
 		dtv[0] = 1;		/* Generation. */
 		dtv[1] = 1;		/* Segments count. */
-		dtv[2] = (Elf_Addr)(tls + DTV_OFFSET);
+		dtv[2] = (Elf_Addr)(tls + TLS_DTV_OFFSET);
 
 		if (libc_tls_init_size > 0)
 			memcpy(tls, libc_tls_init, libc_tls_init_size);
diff --git a/libexec/rtld-elf/aarch64/rtld_machdep.h b/libexec/rtld-elf/aarch64/rtld_machdep.h
--- a/libexec/rtld-elf/aarch64/rtld_machdep.h
+++ b/libexec/rtld-elf/aarch64/rtld_machdep.h
@@ -76,7 +76,6 @@
 	round(16, align)
 #define	calculate_tls_offset(prev_offset, prev_size, size, align, offset) \
 	round(prev_offset + prev_size, align)
-#define	calculate_tls_end(off, size) 	((off) + (size))
 #define calculate_tls_post_size(align) \
 	round(TLS_TCB_SIZE, align) - TLS_TCB_SIZE
 
@@ -93,6 +92,7 @@
 
 #define md_abi_variant_hook(x)
 
+#define	TLS_VARIANT_I	1
 #define	TLS_DTV_OFFSET	0
 
 #endif
diff --git a/libexec/rtld-elf/amd64/reloc.c b/libexec/rtld-elf/amd64/reloc.c
--- a/libexec/rtld-elf/amd64/reloc.c
+++ b/libexec/rtld-elf/amd64/reloc.c
@@ -548,31 +548,22 @@
 }
 
 size_t
-calculate_first_tls_offset(size_t size, size_t align, size_t offset)
+calculate_tls_offset(size_t prev_offset, size_t prev_size __unused,
+    size_t size, size_t align, size_t offset)
 {
 	size_t res;
 
-	res = roundup(size, align);
-	offset &= align - 1;
-	if (offset != 0)
-		res += align - offset;
-	return (res);
+        /*
+	 * res is the smallest integer satisfying res - prev_offset >= size
+         * and (-res) % p_align = p_vaddr % p_align (= p_offset % p_align).
+	 */
+        res = prev_offset + size + align - 1;
+        res -= (res + offset) & (align - 1);
+        return (res);
 }
 
 size_t
-calculate_tls_offset(size_t prev_offset, size_t prev_size __unused, size_t size,
-    size_t align, size_t offset)
-{
-	size_t res;
-
-	res = roundup(prev_offset + size, align);
-	offset &= align - 1;
-	if (offset != 0)
-		res += align - offset;
-	return (res);
-}
-size_t
-calculate_tls_end(size_t off, size_t size __unused)
+calculate_first_tls_offset(size_t size, size_t align, size_t offset)
 {
-	return (off);
+	return (calculate_tls_offset(0, 0, size, align, offset));
 }
diff --git a/libexec/rtld-elf/amd64/rtld_machdep.h b/libexec/rtld-elf/amd64/rtld_machdep.h
--- a/libexec/rtld-elf/amd64/rtld_machdep.h
+++ b/libexec/rtld-elf/amd64/rtld_machdep.h
@@ -73,10 +73,10 @@
 
 #define md_abi_variant_hook(x)
 
+#define	TLS_VARIANT_II	1
 #define	TLS_DTV_OFFSET	0
 
 size_t calculate_first_tls_offset(size_t size, size_t align, size_t offset);
 size_t calculate_tls_offset(size_t prev_offset, size_t prev_size, size_t size,
     size_t align, size_t offset);
-size_t calculate_tls_end(size_t off, size_t size);
 #endif
diff --git a/libexec/rtld-elf/arm/rtld_machdep.h b/libexec/rtld-elf/arm/rtld_machdep.h
--- a/libexec/rtld-elf/arm/rtld_machdep.h
+++ b/libexec/rtld-elf/arm/rtld_machdep.h
@@ -68,7 +68,6 @@
     round(8, align)
 #define calculate_tls_offset(prev_offset, prev_size, size, align, offset) \
     round(prev_offset + prev_size, align)
-#define calculate_tls_end(off, size)    ((off) + (size))
 #define calculate_tls_post_size(align) \
     round(TLS_TCB_SIZE, align) - TLS_TCB_SIZE
 	
@@ -86,6 +85,7 @@
 #define md_abi_variant_hook(x)
 #endif
 
+#define	TLS_VARIANT_I	1
 #define	TLS_DTV_OFFSET	0
 
 #endif
diff --git a/libexec/rtld-elf/i386/reloc.c b/libexec/rtld-elf/i386/reloc.c
--- a/libexec/rtld-elf/i386/reloc.c
+++ b/libexec/rtld-elf/i386/reloc.c
@@ -539,31 +539,22 @@
 }
 
 size_t
-calculate_first_tls_offset(size_t size, size_t align, size_t offset)
+calculate_tls_offset(size_t prev_offset, size_t prev_size __unused,
+    size_t size, size_t align, size_t offset)
 {
 	size_t res;
 
-	res = roundup(size, align);
-	offset &= align - 1;
-	if (offset != 0)
-		res += align - offset;
-	return (res);
+        /*
+	 * res is the smallest integer satisfying res - prev_offset >= size
+         * and (-res) % p_align = p_vaddr % p_align (= p_offset % p_align).
+	 */
+        res = prev_offset + size + align - 1;
+        res -= (res + offset) & (align - 1);
+        return (res);
 }
 
 size_t
-calculate_tls_offset(size_t prev_offset, size_t prev_size __unused, size_t size,
-    size_t align, size_t offset)
-{
-	size_t res;
-
-	res = roundup(prev_offset + size, align);
-	offset &= align - 1;
-	if (offset != 0)
-		res += align - offset;
-	return (res);
-}
-size_t
-calculate_tls_end(size_t off, size_t size __unused)
+calculate_first_tls_offset(size_t size, size_t align, size_t offset)
 {
-	return (off);
+	return (calculate_tls_offset(0, 0, size, align, offset));
 }
diff --git a/libexec/rtld-elf/i386/rtld_machdep.h b/libexec/rtld-elf/i386/rtld_machdep.h
--- a/libexec/rtld-elf/i386/rtld_machdep.h
+++ b/libexec/rtld-elf/i386/rtld_machdep.h
@@ -74,10 +74,10 @@
 
 #define md_abi_variant_hook(x)
 
+#define	TLS_VARIANT_II	1
 #define	TLS_DTV_OFFSET	0
 
 size_t calculate_first_tls_offset(size_t size, size_t align, size_t offset);
 size_t calculate_tls_offset(size_t prev_offset, size_t prev_size, size_t size,
     size_t align, size_t offset);
-size_t calculate_tls_end(size_t off, size_t size);
 #endif
diff --git a/libexec/rtld-elf/mips/rtld_machdep.h b/libexec/rtld-elf/mips/rtld_machdep.h
--- a/libexec/rtld-elf/mips/rtld_machdep.h
+++ b/libexec/rtld-elf/mips/rtld_machdep.h
@@ -68,7 +68,6 @@
     TLS_TCB_SIZE
 #define calculate_tls_offset(prev_offset, prev_size, size, align, offset) \
     round(prev_offset + prev_size, align)
-#define calculate_tls_end(off, size)    ((off) + (size))
 #define calculate_tls_post_size(align)  0
 
 extern void *__tls_get_addr(tls_index *ti);
@@ -78,6 +77,7 @@
 
 #define md_abi_variant_hook(x)
 
+#define	TLS_VARIANT_I	1
 #define	TLS_DTV_OFFSET	0
 
 #endif
diff --git a/libexec/rtld-elf/powerpc/rtld_machdep.h b/libexec/rtld-elf/powerpc/rtld_machdep.h
--- a/libexec/rtld-elf/powerpc/rtld_machdep.h
+++ b/libexec/rtld-elf/powerpc/rtld_machdep.h
@@ -73,6 +73,7 @@
  * TLS
  */
 
+#define	TLS_VARIANT_I	1
 #define TLS_TP_OFFSET	0x7000
 #define TLS_DTV_OFFSET	0x8000
 #define TLS_TCB_SIZE	8
@@ -83,7 +84,6 @@
     TLS_TCB_SIZE
 #define calculate_tls_offset(prev_offset, prev_size, size, align, offset) \
     round(prev_offset + prev_size, align)
-#define calculate_tls_end(off, size)    ((off) + (size))
 #define calculate_tls_post_size(align)  0
  
 typedef struct {
diff --git a/libexec/rtld-elf/powerpc64/rtld_machdep.h b/libexec/rtld-elf/powerpc64/rtld_machdep.h
--- a/libexec/rtld-elf/powerpc64/rtld_machdep.h
+++ b/libexec/rtld-elf/powerpc64/rtld_machdep.h
@@ -65,6 +65,7 @@
  * TLS
  */
 
+#define	TLS_VARIANT_I	1
 #define TLS_TP_OFFSET	0x7000
 #define TLS_DTV_OFFSET	0x8000
 #define TLS_TCB_SIZE	16
@@ -75,7 +76,6 @@
     TLS_TCB_SIZE
 #define calculate_tls_offset(prev_offset, prev_size, size, align, offset) \
     round(prev_offset + prev_size, align)
-#define calculate_tls_end(off, size)    ((off) + (size))
 #define calculate_tls_post_size(align)  0
  
 typedef struct {
diff --git a/libexec/rtld-elf/riscv/rtld_machdep.h b/libexec/rtld-elf/riscv/rtld_machdep.h
--- a/libexec/rtld-elf/riscv/rtld_machdep.h
+++ b/libexec/rtld-elf/riscv/rtld_machdep.h
@@ -82,6 +82,7 @@
 /*
  * TLS
  */
+#define	TLS_VARIANT_I	1
 #define	TLS_TP_OFFSET	0x0
 #define	TLS_DTV_OFFSET	0x800
 #define	TLS_TCB_SIZE	16
@@ -92,7 +93,6 @@
     TLS_TCB_SIZE
 #define calculate_tls_offset(prev_offset, prev_size, size, align, offset) \
     round(prev_offset + prev_size, align)
-#define calculate_tls_end(off, size)    ((off) + (size))
 #define calculate_tls_post_size(align)  0
 
 typedef struct {
diff --git a/libexec/rtld-elf/rtld.c b/libexec/rtld-elf/rtld.c
--- a/libexec/rtld-elf/rtld.c
+++ b/libexec/rtld-elf/rtld.c
@@ -4939,8 +4939,7 @@
 	return (tls_get_addr_slow(dtvp, index, offset, false));
 }
 
-#if defined(__aarch64__) || defined(__arm__) || defined(__mips__) || \
-    defined(__powerpc__) || defined(__riscv)
+#ifdef TLS_VARIANT_I
 
 /*
  * Return pointer to allocated TLS block
@@ -5071,9 +5070,9 @@
     free_aligned(get_tls_block_ptr(tcb, tcbsize));
 }
 
-#endif
+#endif	/* TLS_VARIANT_I */
 
-#if defined(__i386__) || defined(__amd64__)
+#ifdef TLS_VARIANT_II
 
 /*
  * Allocate Static TLS using the Variant II method.
@@ -5179,7 +5178,7 @@
     free((void*) dtv);
 }
 
-#endif
+#endif	/* TLS_VARIANT_II */
 
 /*
  * Allocate TLS block for module with given index.
@@ -5227,6 +5226,11 @@
 	off = calculate_tls_offset(tls_last_offset, tls_last_size,
 	  obj->tlssize, obj->tlsalign, obj->tlspoffset);
 
+    obj->tlsoffset = off;
+#ifdef TLS_VARIANT_I
+    off += obj->tlssize;
+#endif
+
     /*
      * If we have already fixed the size of the static TLS block, we
      * must stay within that size. When allocating the static TLS, we
@@ -5234,13 +5238,13 @@
      * loading modules which use static TLS.
      */
     if (tls_static_space != 0) {
-	if (calculate_tls_end(off, obj->tlssize) > tls_static_space)
+	if (off > tls_static_space)
 	    return false;
     } else if (obj->tlsalign > tls_static_max_align) {
 	    tls_static_max_align = obj->tlsalign;
     }
 
-    tls_last_offset = obj->tlsoffset = off;
+    tls_last_offset = off;
     tls_last_size = obj->tlssize;
     obj->tls_done = true;
 
@@ -5257,8 +5261,11 @@
      * simplistic workaround to allow libGL.so.1 to be loaded and
      * unloaded multiple times.
      */
-    if (calculate_tls_end(obj->tlsoffset, obj->tlssize)
-	== calculate_tls_end(tls_last_offset, tls_last_size)) {
+    size_t off = obj->tlsoffset;
+#ifdef TLS_VARIANT_I
+    off += obj->tlssize;
+#endif
+    if (off == tls_last_offset) {
 	tls_last_offset -= obj->tlssize;
 	tls_last_size = 0;
     }