setup_default_stack_size: set __default_stacksize unconditionally (#396)

[wasi-libc.git] / libc-top-half / musl / src / env / __init_tls.c

#ifdef __wasilibc_unmodified_upstream
#define SYSCALL_NO_TLS 1
#include <elf.h>
#endif
#include <limits.h>
#ifdef __wasilibc_unmodified_upstream
#include <sys/mman.h>
#endif
#include <string.h>
#include <stddef.h>
#include "pthread_impl.h"
#include "libc.h"
#include "atomic.h"
#include "syscall.h"

volatile int __thread_list_lock;

#ifndef __wasilibc_unmodified_upstream

/* These symbols are generated by wasm-ld. __stack_high/__stack_low
 * symbols are only available in LLVM v16 and higher, therefore they're
 * defined as weak symbols and if not available, __heap_base/__data_end
 * is used instead.
 *
 * TODO: remove usage of __heap_base/__data_end for stack size calculation
 * once we drop support for LLVM v15 and older.
 */
extern unsigned char __heap_base;
extern unsigned char __data_end;
extern unsigned char __global_base;
extern weak unsigned char __stack_high;
extern weak unsigned char __stack_low;

static inline void setup_default_stack_size()
{
	ptrdiff_t stack_size;

	if (&__stack_high)
		stack_size = &__stack_high - &__stack_low;
	else {
		unsigned char *sp;
		__asm__(
			".globaltype __stack_pointer, i32\n"
			"global.get __stack_pointer\n"
			"local.set %0\n"
			: "=r"(sp));
		stack_size = sp > &__global_base ? &__heap_base - &__data_end : (ptrdiff_t)&__global_base;
	}

	__default_stacksize =
		stack_size < DEFAULT_STACK_MAX ?
		stack_size : DEFAULT_STACK_MAX;
}

void __wasi_init_tp() {
	__init_tp((void *)__get_tp());
}
#endif

int __init_tp(void *p)
{
	pthread_t td = p;
	td->self = td;
#ifdef __wasilibc_unmodified_upstream
	int r = __set_thread_area(TP_ADJ(p));
	if (r < 0) return -1;
	if (!r) libc.can_do_threads = 1;
	td->detach_state = DT_JOINABLE;
	td->tid = __syscall(SYS_set_tid_address, &__thread_list_lock);
#else
	setup_default_stack_size();
	td->detach_state = DT_JOINABLE;
	/*
	 * Initialize the TID to a value which doesn't conflict with
	 * host-allocated TIDs, so that TID-based locks can work.
	 *
	 * Note:
	 * - Host-allocated TIDs range from 1 to 0x1fffffff. (inclusive)
	 * - __tl_lock and __lockfile uses TID 0 as "unlocked".
	 * - __lockfile relies on the fact the most significant two bits
	 *   of TIDs are 0.
	 */
	td->tid = 0x3fffffff;
#endif
	td->locale = &libc.global_locale;
	td->robust_list.head = &td->robust_list.head;
	td->sysinfo = __sysinfo;
	td->next = td->prev = td;
	return 0;
}

#ifdef __wasilibc_unmodified_upstream

static struct builtin_tls {
	char c;
	struct pthread pt;
	void *space[16];
} builtin_tls[1];
#define MIN_TLS_ALIGN offsetof(struct builtin_tls, pt)

static struct tls_module main_tls;
#endif

#ifndef __wasilibc_unmodified_upstream
extern void __wasm_init_tls(void*);
#endif

void *__copy_tls(unsigned char *mem)
{
#ifdef __wasilibc_unmodified_upstream
	pthread_t td;
	struct tls_module *p;
	size_t i;
	uintptr_t *dtv;

#ifdef TLS_ABOVE_TP
	dtv = (uintptr_t*)(mem + libc.tls_size) - (libc.tls_cnt + 1);

	mem += -((uintptr_t)mem + sizeof(struct pthread)) & (libc.tls_align-1);
	td = (pthread_t)mem;
	mem += sizeof(struct pthread);

	for (i=1, p=libc.tls_head; p; i++, p=p->next) {
		dtv[i] = (uintptr_t)(mem + p->offset) + DTP_OFFSET;
		memcpy(mem + p->offset, p->image, p->len);
	}
#else
	dtv = (uintptr_t *)mem;

	mem += libc.tls_size - sizeof(struct pthread);
	mem -= (uintptr_t)mem & (libc.tls_align-1);
	td = (pthread_t)mem;

	for (i=1, p=libc.tls_head; p; i++, p=p->next) {
		dtv[i] = (uintptr_t)(mem - p->offset) + DTP_OFFSET;
		memcpy(mem - p->offset, p->image, p->len);
	}
#endif
	dtv[0] = libc.tls_cnt;
	td->dtv = dtv;
	return td;
#else
	size_t tls_align = __builtin_wasm_tls_align();
	volatile void* tls_base = __builtin_wasm_tls_base();
	mem += tls_align;
	mem -= (uintptr_t)mem & (tls_align-1);
	__wasm_init_tls(mem);
  	__asm__("local.get %0\n"
			"global.set __tls_base\n"
			:: "r"(tls_base));
	return mem;
#endif
}

#ifdef __wasilibc_unmodified_upstream
#if ULONG_MAX == 0xffffffff
typedef Elf32_Phdr Phdr;
#else
typedef Elf64_Phdr Phdr;
#endif

extern weak hidden const size_t _DYNAMIC[];

static void static_init_tls(size_t *aux)
{
	unsigned char *p;
	size_t n;
	Phdr *phdr, *tls_phdr=0;
	size_t base = 0;
	void *mem;

	for (p=(void *)aux[AT_PHDR],n=aux[AT_PHNUM]; n; n--,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_DYNAMIC && _DYNAMIC)
			base = (size_t)_DYNAMIC - phdr->p_vaddr;
		if (phdr->p_type == PT_TLS)
			tls_phdr = phdr;
		if (phdr->p_type == PT_GNU_STACK &&
		    phdr->p_memsz > __default_stacksize)
			__default_stacksize =
				phdr->p_memsz < DEFAULT_STACK_MAX ?
				phdr->p_memsz : DEFAULT_STACK_MAX;
	}

	if (tls_phdr) {
		main_tls.image = (void *)(base + tls_phdr->p_vaddr);
		main_tls.len = tls_phdr->p_filesz;
		main_tls.size = tls_phdr->p_memsz;
		main_tls.align = tls_phdr->p_align;
		libc.tls_cnt = 1;
		libc.tls_head = &main_tls;
	}

	main_tls.size += (-main_tls.size - (uintptr_t)main_tls.image)
		& (main_tls.align-1);
#ifdef TLS_ABOVE_TP
	main_tls.offset = GAP_ABOVE_TP;
	main_tls.offset += (-GAP_ABOVE_TP + (uintptr_t)main_tls.image)
		& (main_tls.align-1);
#else
	main_tls.offset = main_tls.size;
#endif
	if (main_tls.align < MIN_TLS_ALIGN) main_tls.align = MIN_TLS_ALIGN;

	libc.tls_align = main_tls.align;
	libc.tls_size = 2*sizeof(void *) + sizeof(struct pthread)
#ifdef TLS_ABOVE_TP
		+ main_tls.offset
#endif
		+ main_tls.size + main_tls.align
		+ MIN_TLS_ALIGN-1 & -MIN_TLS_ALIGN;

	if (libc.tls_size > sizeof builtin_tls) {
#ifndef SYS_mmap2
#define SYS_mmap2 SYS_mmap
#endif
		mem = (void *)__syscall(
			SYS_mmap2,
			0, libc.tls_size, PROT_READ|PROT_WRITE,
			MAP_ANONYMOUS|MAP_PRIVATE, -1, 0);
		/* -4095...-1 cast to void * will crash on dereference anyway,
		 * so don't bloat the init code checking for error codes and
		 * explicitly calling a_crash(). */
	} else {
		mem = builtin_tls;
	}

	/* Failure to initialize thread pointer is always fatal. */
	if (__init_tp(__copy_tls(mem)) < 0)
		a_crash();
}

weak_alias(static_init_tls, __init_tls);
#endif