separate __tls_get_addr implementation from dynamic linker/init_tls

such separation serves multiple purposes:

- by having the common path for __tls_get_addr alone in its own
  function with a tail call to the slow case, code generation is
  greatly improved.

- by having __tls_get_addr in it own file, it can be replaced on a
  per-arch basis as needed, for optimization or ABI-specific purposes.

- by removing __tls_get_addr from __init_tls.c, a few bytes of code
  are shaved off of static binaries (which are unlikely to use this
  function unless the linker messed up).

src/env/__init_tls.c

@@ -53,11 +53,6 @@ void *__copy_tls(unsigned char *mem)
 	return td;
 }
 
-void *__tls_get_addr(size_t *v)
-{
-	return (char *)__pthread_self()->dtv[1]+v[1];
-}
-
 #if ULONG_MAX == 0xffffffff
 typedef Elf32_Phdr Phdr;
 #else

src/ldso/dynlink.c

@@ -1031,17 +1031,15 @@ void *__copy_tls(unsigned char *mem)
 	return td;
 }
 
-void *__tls_get_addr(size_t *v)
+void *__tls_get_new(size_t *v)
 {
 	pthread_t self = __pthread_self();
-	if (v[0]<=(size_t)self->dtv[0])
-		return (char *)self->dtv[v[0]]+v[1];
 
 	/* Block signals to make accessing new TLS async-signal-safe */
 	sigset_t set;
-	pthread_sigmask(SIG_BLOCK, SIGALL_SET, &set);
+	__block_all_sigs(&set);
 	if (v[0]<=(size_t)self->dtv[0]) {
-		pthread_sigmask(SIG_SETMASK, &set, 0);
+		__restore_sigs(&set);
 		return (char *)self->dtv[v[0]]+v[1];
 	}
 
@@ -1074,7 +1072,7 @@ void *__tls_get_addr(size_t *v)
 		memcpy(mem, p->tls_image, p->tls_len);
 		if (p->tls_id == v[0]) break;
 	}
-	pthread_sigmask(SIG_SETMASK, &set, 0);
+	__restore_sigs(&set);
 	return mem + v[1];
 }
 
@@ -1442,6 +1440,8 @@ static int invalid_dso_handle(void *h)
 	return 1;
 }
 
+void *__tls_get_addr(size_t *);
+
 static void *do_dlsym(struct dso *p, const char *s, void *ra)
 {
 	size_t i;

src/thread/__tls_get_addr.c

@@ -0,0 +1,17 @@
+#include <stddef.h>
+#include "pthread_impl.h"
+#include "libc.h"
+
+void *__tls_get_new(size_t *) ATTR_LIBC_VISIBILITY;
+
+void *__tls_get_addr(size_t *v)
+{
+	pthread_t self = __pthread_self();
+#ifdef SHARED
+	if (v[0]<=(size_t)self->dtv[0])
+		return (char *)self->dtv[v[0]]+v[1];
+	return __tls_get_new(v);
+#else
+	return (char *)self->dtv[1]+v[1];
+#endif
+}