Align initial-exec TLS segments to the p_vaddr % align.

This is continuation of D21163/r359634, which handled the alignment
for global mode.

Non-x86 arches are not handled, maintainers are welcomed.

Tested by:	emaste
Sponsored by:	The FreeBSD Foundation
MFC after:	2 weeks
Differential revision:	https://reviews.freebsd.org/D24366

libexec/rtld-elf/aarch64/rtld_machdep.h

@@ -72,9 +72,9 @@ Elf_Addr reloc_jmpslot(Elf_Addr *where, Elf_Addr target,
 
 #define	round(size, align)				\
 	(((size) + (align) - 1) & ~((align) - 1))
-#define	calculate_first_tls_offset(size, align) \
+#define	calculate_first_tls_offset(size, align, offset)	\
 	round(16, align)
-#define	calculate_tls_offset(prev_offset, prev_size, size, 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) \

libexec/rtld-elf/amd64/reloc.c

@@ -552,3 +552,33 @@ void *__tls_get_addr(tls_index *ti)
 
     return tls_get_addr_common(&segbase[1], ti->ti_module, ti->ti_offset);
 }
+
+size_t
+calculate_first_tls_offset(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);
+}
+
+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)
+{
+	return (off);
+}

libexec/rtld-elf/amd64/rtld_machdep.h

@@ -61,14 +61,6 @@ extern uint32_t cpu_stdext_feature2;
 	(((Elf_Addr (*)(uint32_t, uint32_t, uint32_t, uint32_t))ptr)( \
 	    cpu_feature, cpu_feature2, cpu_stdext_feature, cpu_stdext_feature2))
 
-#define round(size, align) \
-	(((size) + (align) - 1) & ~((align) - 1))
-#define calculate_first_tls_offset(size, align) \
-	round(size, align)
-#define calculate_tls_offset(prev_offset, prev_size, size, align) \
-	round((prev_offset) + (size), align)
-#define calculate_tls_end(off, size) 	(off)
-
 typedef struct {
     unsigned long ti_module;
     unsigned long ti_offset;
@@ -81,4 +73,8 @@ void *__tls_get_addr(tls_index *ti) __exported;
 
 #define md_abi_variant_hook(x)
 
+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

libexec/rtld-elf/arm/rtld_machdep.h

@@ -64,9 +64,9 @@ typedef struct {
 
 #define round(size, align) \
     (((size) + (align) - 1) & ~((align) - 1))
-#define calculate_first_tls_offset(size, align) \
+#define calculate_first_tls_offset(size, align, offset)	\
     round(8, align)
-#define calculate_tls_offset(prev_offset, prev_size, size, 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) \

libexec/rtld-elf/i386/reloc.c

@@ -543,3 +543,33 @@ void *__tls_get_addr(tls_index *ti)
 
     return tls_get_addr_common(&segbase[1], ti->ti_module, ti->ti_offset);
 }
+
+size_t
+calculate_first_tls_offset(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);
+}
+
+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)
+{
+	return (off);
+}

libexec/rtld-elf/i386/rtld_machdep.h

@@ -61,14 +61,6 @@ extern uint32_t cpu_stdext_feature2;
 	(((Elf_Addr (*)(uint32_t, uint32_t, uint32_t, uint32_t))(ptr))( \
 	    cpu_feature, cpu_feature2, cpu_stdext_feature, cpu_stdext_feature2))
 
-#define round(size, align) \
-	(((size) + (align) - 1) & ~((align) - 1))
-#define calculate_first_tls_offset(size, align) \
-	round(size, align)
-#define calculate_tls_offset(prev_offset, prev_size, size, align) \
-	round((prev_offset) + (size), align)
-#define calculate_tls_end(off, size) 	(off)
-
 typedef struct {
     unsigned long ti_module;
     unsigned long ti_offset;
@@ -82,4 +74,8 @@ void *__tls_get_addr(tls_index *ti) __exported;
 
 #define md_abi_variant_hook(x)
 
+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

libexec/rtld-elf/mips/rtld_machdep.h

@@ -66,9 +66,9 @@ typedef struct {
 
 #define round(size, align) \
     (((size) + (align) - 1) & ~((align) - 1))
-#define calculate_first_tls_offset(size, align) \
+#define calculate_first_tls_offset(size, align, offset)	\
     TLS_TCB_SIZE
-#define calculate_tls_offset(prev_offset, prev_size, size, 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)  0

libexec/rtld-elf/powerpc/rtld_machdep.h

@@ -79,9 +79,9 @@ void _rtld_powerpc_pltcall(void);
 
 #define round(size, align) \
     (((size) + (align) - 1) & ~((align) - 1))
-#define calculate_first_tls_offset(size, align) \
+#define calculate_first_tls_offset(size, align, offset)	\
     TLS_TCB_SIZE
-#define calculate_tls_offset(prev_offset, prev_size, size, 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)  0

libexec/rtld-elf/powerpc64/rtld_machdep.h

@@ -71,9 +71,9 @@ extern u_long cpu_features2; /* r4 */
 
 #define round(size, align) \
     (((size) + (align) - 1) & ~((align) - 1))
-#define calculate_first_tls_offset(size, align) \
+#define calculate_first_tls_offset(size, align, offset)	\
     TLS_TCB_SIZE
-#define calculate_tls_offset(prev_offset, prev_size, size, 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)  0

libexec/rtld-elf/riscv/rtld_machdep.h

@@ -88,9 +88,9 @@ Elf_Addr reloc_jmpslot(Elf_Addr *where, Elf_Addr target,
 
 #define round(size, align) \
     (((size) + (align) - 1) & ~((align) - 1))
-#define calculate_first_tls_offset(size, align) \
+#define calculate_first_tls_offset(size, align, offset)	\
     TLS_TCB_SIZE
-#define calculate_tls_offset(prev_offset, prev_size, size, 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)  0

libexec/rtld-elf/rtld.c

@@ -4979,13 +4979,13 @@ allocate_tls(Obj_Entry *objs, void *oldtls, size_t tcbsize, size_t tcbalign)
     ralign = tcbalign;
     if (tls_static_max_align > ralign)
 	    ralign = tls_static_max_align;
-    size = round(tls_static_space, ralign) + round(tcbsize, ralign);
+    size = roundup(tls_static_space, ralign) + roundup(tcbsize, ralign);
 
     assert(tcbsize >= 2*sizeof(Elf_Addr));
     tls = malloc_aligned(size, ralign, 0 /* XXX */);
     dtv = xcalloc(tls_max_index + 2, sizeof(Elf_Addr));
 
-    segbase = (Elf_Addr)(tls + round(tls_static_space, ralign));
+    segbase = (Elf_Addr)(tls + roundup(tls_static_space, ralign));
     ((Elf_Addr*)segbase)[0] = segbase;
     ((Elf_Addr*)segbase)[1] = (Elf_Addr) dtv;
 
@@ -5051,7 +5051,7 @@ free_tls(void *tls, size_t tcbsize  __unused, size_t tcbalign)
     ralign = tcbalign;
     if (tls_static_max_align > ralign)
 	    ralign = tls_static_max_align;
-    size = round(tls_static_space, ralign);
+    size = roundup(tls_static_space, ralign);
 
     dtv = ((Elf_Addr**)tls)[1];
     dtvsize = dtv[1];
@@ -5109,10 +5109,11 @@ allocate_tls_offset(Obj_Entry *obj)
     }
 
     if (tls_last_offset == 0)
-	off = calculate_first_tls_offset(obj->tlssize, obj->tlsalign);
+	off = calculate_first_tls_offset(obj->tlssize, obj->tlsalign,
+	  obj->tlspoffset);
     else
 	off = calculate_tls_offset(tls_last_offset, tls_last_size,
-				   obj->tlssize, obj->tlsalign);
+	  obj->tlssize, obj->tlsalign, obj->tlspoffset);
 
     /*
      * If we have already fixed the size of the static TLS block, we