diff --git a/arch/arm/pthread_arch.h b/arch/arm/pthread_arch.h
index 5f96f2b07d36c0162aa2545c44d9d03e3911166a..43a1c012f466dcd41bc6444edfd4c265c722f7cd 100644
--- a/arch/arm/pthread_arch.h
+++ b/arch/arm/pthread_arch.h
@@ -1,6 +1,9 @@
-typedef pthread_t (*__pthread_self_func_t)(void) __attribute__((const));
+typedef char *(*__ptr_func_t)(void) __attribute__((const));
 
-#define __pthread_self ((__pthread_self_func_t)0xffff0fe0)
+#define __pthread_self() \
+	((pthread_t)(((__ptr_func_t)0xffff0fe0)()+8-sizeof(struct pthread)))
+
+#define TLS_ABOVE_TP
+#define TP_ADJ(p) ((char *)(p) + sizeof(struct pthread) - 8)
 
-#define CANCEL_REG_SP 16
 #define CANCEL_REG_IP 18
diff --git a/arch/arm/reloc.h b/arch/arm/reloc.h
index c37af07bb8d84f6cfdcb71f276bff12b79d3dc46..b41314de8bd8c55eadcb09585ba1ae8d0ff97120 100644
--- a/arch/arm/reloc.h
+++ b/arch/arm/reloc.h
@@ -34,8 +34,8 @@ static inline void do_single_reloc(
 		break;
 	case R_ARM_TLS_TPOFF32:
 		*reloc_addr += def.sym
-			? def.sym->st_value - def.dso->tls_offset
-			: 0 - self->tls_offset;
+			? def.sym->st_value + def.dso->tls_offset + 8
+			: self->tls_offset + 8;
 		break;
 	}
 }
diff --git a/arch/i386/pthread_arch.h b/arch/i386/pthread_arch.h
index 0ea0aaca4253d362972148e1e7f30ef92fa02b80..1c06c7641df683b9ffe14c0e91f0b1da192abbe0 100644
--- a/arch/i386/pthread_arch.h
+++ b/arch/i386/pthread_arch.h
@@ -5,5 +5,6 @@ static inline struct pthread *__pthread_self()
 	return self;
 }
 
-#define CANCEL_REG_SP 7
+#define TP_ADJ(p) (p)
+
 #define CANCEL_REG_IP 14
diff --git a/arch/microblaze/pthread_arch.h b/arch/microblaze/pthread_arch.h
index 6c0ab5ad5a1dd6a404250fd84752d8971f084106..259d3d615ab1fb30755d48a03730c1a5d3005eff 100644
--- a/arch/microblaze/pthread_arch.h
+++ b/arch/microblaze/pthread_arch.h
@@ -5,4 +5,6 @@ static inline struct pthread *__pthread_self()
 	return self;
 }
 
+#define TP_ADJ(p) (p)
+
 #define CANCEL_REG_IP 32
diff --git a/arch/mips/pthread_arch.h b/arch/mips/pthread_arch.h
index 77b7330dacfd1445d4635d64bb0c6ece1a87897c..f8e35ae46377d5f89ed3eede40dcc1cff83c2747 100644
--- a/arch/mips/pthread_arch.h
+++ b/arch/mips/pthread_arch.h
@@ -1,12 +1,16 @@
 static inline struct pthread *__pthread_self()
 {
-	struct pthread *self;
 #ifdef __clang__
-	__asm__ __volatile__ (".word 0x7c03e83b ; move %0, $3" : "=r" (self) : : "$3" );
+	char *tp;
+	__asm__ __volatile__ (".word 0x7c03e83b ; move %0, $3" : "=r" (tp) : : "$3" );
 #else
-	__asm__ __volatile__ (".word 0x7c03e83b" : "=v" (self) );
+	register char *tp __asm__("$3");
+	__asm__ __volatile__ (".word 0x7c03e83b" : "=r" (tp) );
 #endif
-	return self;
+	return (pthread_t)(tp - 0x7000 - sizeof(struct pthread));
 }
 
+#define TLS_ABOVE_TP
+#define TP_ADJ(p) ((char *)(p) + sizeof(struct pthread) + 0x7000)
+
 #define CANCEL_REG_IP (3-(union {int __i; char __b;}){1}.__b)
diff --git a/arch/mips/reloc.h b/arch/mips/reloc.h
index 785489ed1077ec860e670aadb4836e16c7678f87..f5e9c77b0c63f4e9754b8fee95661d71c87e4b67 100644
--- a/arch/mips/reloc.h
+++ b/arch/mips/reloc.h
@@ -23,6 +23,17 @@ static inline void do_single_reloc(
 	case R_MIPS_COPY:
 		memcpy(reloc_addr, (void *)sym_val, sym_size);
 		break;
+	case R_MIPS_TLS_DTPMOD32:
+		*reloc_addr = def.dso ? def.dso->tls_id : self->tls_id;
+		break;
+	case R_MIPS_TLS_DTPREL32:
+		*reloc_addr += def.sym->st_value;
+		break;
+	case R_MIPS_TLS_TPREL32:
+		*reloc_addr += def.sym
+			? def.sym->st_value + def.dso->tls_offset - 0x7000
+			: self->tls_offset - 0x7000;
+		break;
 	}
 }
 
diff --git a/arch/x86_64/pthread_arch.h b/arch/x86_64/pthread_arch.h
index 836187f5654ea4087c754c512f3543163e3738ff..77c8387c57b698118bbd1af0da452274ca0d710f 100644
--- a/arch/x86_64/pthread_arch.h
+++ b/arch/x86_64/pthread_arch.h
@@ -5,5 +5,6 @@ static inline struct pthread *__pthread_self()
 	return self;
 }
 
-#define CANCEL_REG_SP 15
+#define TP_ADJ(p) (p)
+
 #define CANCEL_REG_IP 16
diff --git a/src/env/__init_tls.c b/src/env/__init_tls.c
index e70025d77a01b25176c2ffd0487cee103cf047a6..ab5992ce1c928eaccddc44bd8f301f2a5b98a981 100644
--- a/src/env/__init_tls.c
+++ b/src/env/__init_tls.c
@@ -15,11 +15,18 @@ void *__copy_tls(unsigned char *mem)
 	if (!image) return mem;
 	void **dtv = (void *)mem;
 	dtv[0] = (void *)1;
+#ifdef TLS_ABOVE_TP
+	mem += sizeof(void *) * 2;
+	mem += -((uintptr_t)mem + sizeof(struct pthread)) & (align-1);
+	td = (pthread_t)mem;
+	mem += sizeof(struct pthread);
+#else
 	mem += __libc.tls_size - sizeof(struct pthread);
 	mem -= (uintptr_t)mem & (align-1);
 	td = (pthread_t)mem;
 	td->dtv = dtv;
 	mem -= size;
+#endif
 	dtv[1] = mem;
 	memcpy(mem, image, len);
 	return td;
@@ -33,7 +40,7 @@ void *__tls_get_addr(size_t *v)
 static void *simple(void *p)
 {
 	*(void **)p = p;
-	return __set_thread_area(p) ? 0 : p;
+	return __set_thread_area(TP_ADJ(p)) ? 0 : p;
 }
 
 weak_alias(simple, __install_initial_tls);
diff --git a/src/ldso/dynlink.c b/src/ldso/dynlink.c
index 5353979e31b1ecf4b39b4db1ce3cad38b0613a3c..2ec8fb8c4985b92a782a3a692e3b3884178480f7 100644
--- a/src/ldso/dynlink.c
+++ b/src/ldso/dynlink.c
@@ -539,10 +539,16 @@ static struct dso *load_library(const char *name)
 		}
 		p->tls_id = ++tls_cnt;
 		tls_align = MAXP2(tls_align, p->tls_align);
+#ifdef TLS_ABOVE_TP
+		p->tls_offset = tls_offset + ( (tls_align-1) &
+			-(tls_offset + (uintptr_t)p->tls_image) );
+		tls_offset += p->tls_size;
+#else
 		tls_offset += p->tls_size + p->tls_align - 1;
 		tls_offset -= (tls_offset + (uintptr_t)p->tls_image)
 			& (p->tls_align-1);
 		p->tls_offset = tls_offset;
+#endif
 		p->new_dtv = (void *)(-sizeof(size_t) &
 			(uintptr_t)(p->name+strlen(p->name)+sizeof(size_t)));
 		p->new_tls = (void *)(p->new_dtv + n_th*(tls_cnt+1));
@@ -697,6 +703,18 @@ void *__copy_tls(unsigned char *mem)
 	void **dtv = (void *)mem;
 	dtv[0] = (void *)tls_cnt;
 
+#ifdef TLS_ABOVE_TP
+	mem += sizeof(void *) * (tls_cnt+1);
+	mem += -((uintptr_t)mem + sizeof(struct pthread)) & (tls_align-1);
+	td = (pthread_t)mem;
+	mem += sizeof(struct pthread);
+
+	for (p=head; p; p=p->next) {
+		if (!p->tls_id) continue;
+		dtv[p->tls_id] = mem + p->tls_offset;
+		memcpy(dtv[p->tls_id], p->tls_image, p->tls_len);
+	}
+#else
 	mem += libc.tls_size - sizeof(struct pthread);
 	mem -= (uintptr_t)mem & (tls_align-1);
 	td = (pthread_t)mem;
@@ -706,6 +724,7 @@ void *__copy_tls(unsigned char *mem)
 		dtv[p->tls_id] = mem - p->tls_offset;
 		memcpy(dtv[p->tls_id], p->tls_image, p->tls_len);
 	}
+#endif
 	td->dtv = dtv;
 	return td;
 }
@@ -753,9 +772,12 @@ void *__tls_get_addr(size_t *v)
 
 static void update_tls_size()
 {
-	size_t below_tp = (1+tls_cnt) * sizeof(void *) + tls_offset;
-	size_t above_tp = sizeof(struct pthread) + tls_align;
-	libc.tls_size = ALIGN(below_tp + above_tp, tls_align);
+	libc.tls_size = ALIGN(
+		(1+tls_cnt) * sizeof(void *) +
+		tls_offset +
+		sizeof(struct pthread) +
+		tls_align * 2,
+	tls_align);
 }
 
 void *__dynlink(int argc, char **argv)
@@ -866,9 +888,16 @@ void *__dynlink(int argc, char **argv)
 	}
 	if (app->tls_size) {
 		app->tls_id = tls_cnt = 1;
+#ifdef TLS_ABOVE_TP
+		app->tls_offset = 0;
+		tls_offset = app->tls_size
+			+ ( -((uintptr_t)app->tls_image + app->tls_size)
+			& (app->tls_align-1) );
+#else
 		tls_offset = app->tls_offset = app->tls_size
 			+ ( -((uintptr_t)app->tls_image + app->tls_size)
 			& (app->tls_align-1) );
+#endif
 		tls_align = MAXP2(tls_align, app->tls_align);
 	}
 	app->global = 1;
diff --git a/src/thread/arm/tls.s b/src/thread/arm/tls.s
new file mode 100644
index 0000000000000000000000000000000000000000..59736ac279c07150dc208971fe4a2c3337b459f3
--- /dev/null
+++ b/src/thread/arm/tls.s
@@ -0,0 +1,4 @@
+.global __aeabi_read_tp
+.type __aeabi_read_tp,%function
+__aeabi_read_tp:
+	ldr pc,=0xffff0fe0
diff --git a/src/thread/pthread_create.c b/src/thread/pthread_create.c
index 627ff5fbbe087cea6782eb8a9703aa389061ff5e..e67616e7464475187a138017cc9a09d428453721 100644
--- a/src/thread/pthread_create.c
+++ b/src/thread/pthread_create.c
@@ -147,7 +147,7 @@ int pthread_create(pthread_t *restrict res, const pthread_attr_t *restrict attr,
 	new->canary = self->canary;
 
 	a_inc(&libc.threads_minus_1);
-	ret = __clone(start, stack, flags, new, &new->tid, new, &new->tid);
+	ret = __clone(start, stack, flags, new, &new->tid, TP_ADJ(new), &new->tid);
 
 	__release_ptc();
 
diff --git a/src/thread/pthread_self.c b/src/thread/pthread_self.c
index a93d8129353103d5b0a63be15636bbe7b1d8a558..23dbaa5eb805ccbcf65d9e1772b9ab6cabb52853 100644
--- a/src/thread/pthread_self.c
+++ b/src/thread/pthread_self.c
@@ -10,7 +10,7 @@ static int init_main_thread()
 {
 	__syscall(SYS_rt_sigprocmask, SIG_UNBLOCK,
 		SIGPT_SET, 0, __SYSCALL_SSLEN);
-	if (__set_thread_area(main_thread) < 0) return -1;
+	if (__set_thread_area(TP_ADJ(main_thread)) < 0) return -1;
 	main_thread->canceldisable = libc.canceldisable;
 	main_thread->tsd = (void **)__pthread_tsd_main;
 	main_thread->errno_ptr = __errno_location();