TLS (GNU/C11 thread-local storage) support for static-linked programs

the design for TLS in dynamic-linked programs is mostly complete too,
but I have not yet implemented it. cost is nonzero but still low for
programs which do not use TLS and/or do not use threads (a few hundred
bytes of new code, plus dependency on memcpy). i believe it can be
made smaller at some point by merging __init_tls and __init_security
into __libc_start_main and avoiding duplicate auxv-parsing code.

at the same time, I've also slightly changed the logic pthread_create
uses to allocate guard pages to ensure that guard pages are not
counted towards commit charge.

src/env/__init_tls.c

@@ -0,0 +1,71 @@
+#include <elf.h>
+#include <limits.h>
+#include "pthread_impl.h"
+#include "libc.h"
+#include "atomic.h"
+
+#ifndef SHARED
+
+static void *image;
+static size_t len, bss, align;
+
+void *__copy_tls(unsigned char *mem, size_t cnt)
+{
+	mem += ((uintptr_t)image - (uintptr_t)mem) & (align-1);
+	memcpy(mem, image, len);
+	return mem + len + bss;
+}
+
+static void *simple(void *p)
+{
+	*(void **)p = p;
+	return __set_thread_area(p) ? 0 : p;
+}
+
+weak_alias(simple, __install_initial_tls);
+
+void *__mmap(void *, size_t, int, int, int, off_t);
+
+#if ULONG_MAX == 0xffffffff
+typedef Elf32_Phdr Phdr;
+#else
+typedef Elf64_Phdr Phdr;
+#endif
+
+#define AUX_CNT 6
+
+void __init_tls(size_t *auxv)
+{
+	size_t i, aux[AUX_CNT] = { 0 };
+	unsigned char *p, *mem;
+	size_t n, d;
+	Phdr *phdr, *tls_phdr=0;
+	size_t base = 0;
+
+	for (; auxv[0]; auxv+=2) if (auxv[0]<AUX_CNT) aux[auxv[0]] = auxv[1];
+	p = (void *)aux[AT_PHDR];
+	for (p=(void *)aux[AT_PHDR]; aux[AT_PHNUM]--; p+=aux[AT_PHENT]) {
+		phdr = (void *)p;
+		if (phdr->p_type == PT_PHDR)
+			base = aux[AT_PHDR] - phdr->p_vaddr;
+		if (phdr->p_type == PT_TLS)
+			tls_phdr = phdr;
+	}
+	if (!tls_phdr) return;
+
+	libc.tls_size = len+bss+align+4*sizeof(size_t)+sizeof(struct pthread);
+
+	image = (void *)(base + tls_phdr->p_vaddr);
+	len = tls_phdr->p_filesz;
+	bss = tls_phdr->p_memsz - len;
+	align = tls_phdr->p_align;
+	if (align < 4*sizeof(size_t)) align = 4*sizeof(size_t);
+	mem = __mmap(0, libc.tls_size, PROT_READ|PROT_WRITE,
+		MAP_ANONYMOUS|MAP_PRIVATE, -1, 0);
+	if (mem == MAP_FAILED) a_crash();
+
+	if (!__install_initial_tls(__copy_tls(mem, 0))) a_crash();
+}
+#else
+void __init_tls(size_t *auxv) { }
+#endif

src/env/__libc_start_main.c

@@ -1,5 +1,6 @@
 #include "libc.h"
 
+void __init_tls(size_t *);
 void __init_security(size_t *);
 
 int __libc_start_main(
@@ -15,6 +16,7 @@ int __libc_start_main(
 	libc.ldso_fini = ldso_fini;
 	libc.fini = fini;
 
+	__init_tls((void *)auxv);
 	__init_security((void *)auxv);
 
 	/* Execute constructors (static) linked into the application */

src/internal/libc.h

@@ -16,6 +16,7 @@ struct __libc {
 	int canceldisable;
 	FILE *ofl_head;
 	int ofl_lock[2];
+	volatile size_t tls_size, tls_cnt;
 };
 
 extern size_t __hwcap;

src/ldso/dynlink.c

@@ -17,6 +17,9 @@
 #include <pthread.h>
 #include <ctype.h>
 #include <dlfcn.h>
+#include "pthread_impl.h"
+#include "libc.h"
+#undef libc
 
 static int errflag;
 static char errbuf[128];
@@ -788,6 +791,8 @@ void *__dynlink(int argc, char **argv)
 	debug.state = 0;
 	_dl_debug_state();
 
+	/* Stand-in until real TLS support is added to dynamic linker */
+	__libc.tls_size = sizeof(struct pthread) + 4*sizeof(size_t);
 	if (ssp_used) __init_ssp(auxv);
 
 	do_init_fini(tail);
@@ -802,6 +807,11 @@ void *__dynlink(int argc, char **argv)
 	return (void *)aux[AT_ENTRY];
 }
 
+void *__copy_tls(unsigned char *mem, size_t cnt)
+{
+	return mem;
+}
+
 void *dlopen(const char *file, int mode)
 {
 	struct dso *volatile p, *orig_tail = tail, *next;

src/thread/pthread_create.c

@@ -84,6 +84,8 @@ static void init_file_lock(FILE *f)
 	if (f && f->lock<0) f->lock = 0;
 }
 
+void *__copy_tls(unsigned char *, size_t);
+
 int pthread_create(pthread_t *restrict res, const pthread_attr_t *restrict attr, void *(*entry)(void *), void *restrict arg)
 {
 	int ret;
@@ -92,6 +94,8 @@ int pthread_create(pthread_t *restrict res, const pthread_attr_t *restrict attr,
 	struct pthread *self = pthread_self(), *new;
 	unsigned char *map, *stack, *tsd;
 	unsigned flags = 0x7d8f00;
+	size_t tls_cnt = libc.tls_cnt;
+	size_t tls_size = libc.tls_size;
 
 	if (!self) return ENOSYS;
 	if (!libc.threaded) {
@@ -109,15 +113,24 @@ int pthread_create(pthread_t *restrict res, const pthread_attr_t *restrict attr,
 	} else {
 		if (attr) {
 			guard = ROUND(attr->_a_guardsize + DEFAULT_GUARD_SIZE);
-			size = guard + ROUND(attr->_a_stacksize + DEFAULT_STACK_SIZE);
+			size = guard + ROUND(attr->_a_stacksize
+				+ DEFAULT_STACK_SIZE + tls_size);
 		}
 		size += __pthread_tsd_size;
-		map = mmap(0, size, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANON, -1, 0);
-		if (map == MAP_FAILED) return EAGAIN;
-		if (guard) mprotect(map, guard, PROT_NONE);
+		if (guard) {
+			map = mmap(0, size, PROT_NONE, MAP_PRIVATE|MAP_ANON, -1, 0);
+			if (map == MAP_FAILED) return EAGAIN;
+			if (mprotect(map+guard, size-guard, PROT_READ|PROT_WRITE)) {
+				munmap(map, size);
+				return EAGAIN;
+			}
+		} else {
+			map = mmap(0, size, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANON, -1, 0);
+			if (map == MAP_FAILED) return EAGAIN;
+		}
 		tsd = map + size - __pthread_tsd_size;
 	}
-	new = (void *)(tsd - sizeof *new - PAGE_SIZE%sizeof *new);
+	new = __copy_tls(tsd - tls_size, tls_cnt);
 	new->map_base = map;
 	new->map_size = size;
 	new->pid = self->pid;

src/thread/pthread_self.c

@@ -1,6 +1,6 @@
 #include "pthread_impl.h"
 
-static struct pthread main_thread;
+static struct pthread *main_thread = &(struct pthread){0};
 
 /* pthread_key_create.c overrides this */
 static const void *dummy[1] = { 0 };
@@ -10,14 +10,14 @@ 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;
-	main_thread.canceldisable = libc.canceldisable;
-	main_thread.tsd = (void **)__pthread_tsd_main;
-	main_thread.errno_ptr = __errno_location();
-	main_thread.self = &main_thread;
-	main_thread.tid = main_thread.pid =
-		__syscall(SYS_set_tid_address, &main_thread.tid);
-	libc.main_thread = &main_thread;
+	if (__set_thread_area(main_thread) < 0) return -1;
+	main_thread->canceldisable = libc.canceldisable;
+	main_thread->tsd = (void **)__pthread_tsd_main;
+	main_thread->errno_ptr = __errno_location();
+	main_thread->self = main_thread;
+	main_thread->tid = main_thread->pid =
+		__syscall(SYS_set_tid_address, &main_thread->tid);
+	libc.main_thread = main_thread;
 	return 0;
 }
 
@@ -35,3 +35,9 @@ pthread_t __pthread_self_def()
 
 weak_alias(__pthread_self_def, pthread_self);
 weak_alias(__pthread_self_def, __pthread_self_init);
+
+void *__install_initial_tls(void *p)
+{
+	main_thread = p;
+	return __pthread_self_def();
+}