DevKit7/libctru
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Use __tdata_align to align thread local storage

Browse Source
This commit is contained in:
Ian Chamberlain 2022-08-17 22:55:37 -04:00
parent 59cf50c201
commit 6b2d2fb1ab
No known key found for this signature in database
GPG Key ID: AE5484D09405AA60
3 changed files with 26 additions and 19 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
libctru/source/internal.h
View File

@ -8,6 +8,12 @@
#define THREADVARS_MAGIC 0x21545624 // !TV$
#define FS_OVERRIDE_MAGIC 0x21465324 // !FS$
extern const size_t __tdata_align;
extern const u8 __tdata_lma[];
extern const u8 __tdata_lma_end[];
extern u8 __tls_start[];
extern u8 __tls_end[];
// Keep this structure under 0x80 bytes
typedef struct
{
@ -48,3 +54,11 @@ static inline ThreadVars* getThreadVars(void)
}
void initThreadVars(struct Thread_tag *thread);
static inline size_t getThreadLocalStartOffset(size_t tls_tp) {
size_t align = 8;
if (__tdata_align > align) align = __tdata_align;
// ARM ELF TLS ABI mandates an 8-byte header, so we include an extra 8 bytes
// then add padding to align the .tdata properly
return (8 + (size_t)tls_tp + (__tdata_align - 1)) & ~(__tdata_align - 1);
}

13
libctru/source/system/syscalls.c
View File

@ -16,10 +16,6 @@
void __ctru_exit(int rc);
extern const u8 __tdata_lma[];
extern const u8 __tdata_lma_end[];
extern u8 __tls_start[];
struct _reent* __SYSCALL(getreent)()
{
ThreadVars* tv = getThreadVars();
@ -43,7 +39,7 @@ int __SYSCALL(clock_gettime)(clockid_t clock_id, struct timespec *tp) {
tp->tv_nsec = (ms_since_epoch % 1000) * 1000000;
}
}
else if (clock_id == CLOCK_MONOTONIC)
else if (clock_id == CLOCK_MONOTONIC)
{
if (tp != NULL)
{
@ -66,7 +62,7 @@ int __SYSCALL(clock_gettime)(clockid_t clock_id, struct timespec *tp) {
int __SYSCALL(clock_getres)(clockid_t clock_id, struct timespec *res) {
if (clock_id == CLOCK_REALTIME)
{
if (res != NULL)
if (res != NULL)
{
res->tv_sec = 0;
res->tv_nsec = 1000000;
@ -168,7 +164,7 @@ void initThreadVars(struct Thread_tag *thread)
tv->thread_ptr = thread;
#pragma GCC diagnostic push
#pragma GCC diagnostic ignored "-Warray-bounds"
tv->tls_tp = (thread != NULL ? (u8*)thread->stacktop : __tls_start) - 8; // Arm ELF TLS ABI mandates an 8-byte header
tv->tls_tp = (thread != NULL ? (u8*)thread->stacktop : __tls_start);
#pragma GCC diagnostic pop
tv->srv_blocking_policy = false;
@ -185,6 +181,7 @@ void __system_initSyscalls(void)
// Initialize thread vars for the main thread
initThreadVars(NULL);
u32 tls_size = __tdata_lma_end - __tdata_lma;
size_t tdata_start = getThreadLocalStartOffset((size_t)__tls_start);
if (tls_size)
memcpy(__tls_start, __tdata_lma, tls_size);
memcpy((void*)tdata_start, __tdata_lma, tls_size);
}

18
libctru/source/thread.c
View File

@ -3,11 +3,6 @@
#include <malloc.h>
#include <string.h>
extern const u8 __tdata_lma[];
extern const u8 __tdata_lma_end[];
extern u8 __tls_start[];
extern u8 __tls_end[];
static void __panic(void)
{
svcBreak(USERBREAK_PANIC);
@ -24,8 +19,8 @@ static void _thread_begin(void* arg)
Thread threadCreate(ThreadFunc entrypoint, void* arg, size_t stack_size, int prio, int core_id, bool detached)
{
size_t stackoffset = (sizeof(struct Thread_tag)+7)&~7;
size_t allocsize = stackoffset + ((stack_size+7)&~7);
size_t stackoffset = (sizeof(struct Thread_tag) + 7) & ~7;
size_t allocsize = getThreadLocalStartOffset(stackoffset + stack_size);
size_t tlssize = __tls_end-__tls_start;
size_t tlsloadsize = __tdata_lma_end-__tdata_lma;
size_t tbsssize = tlssize-tlsloadsize;
@ -35,19 +30,20 @@ Thread threadCreate(ThreadFunc entrypoint, void* arg, size_t stack_size, int pri
if ((allocsize-stackoffset) < stack_size) return NULL;
if ((allocsize+tlssize) < allocsize) return NULL;
Thread t = (Thread)memalign(8,allocsize+tlssize);
Thread t = (Thread)memalign(__tdata_align, allocsize + tlssize);
if (!t) return NULL;
t->ep = entrypoint;
t->arg = arg;
t->detached = detached;
t->finished = false;
t->stacktop = (u8*)t + allocsize;
t->stacktop = (u8*)t + stackoffset + stack_size;
void* tdata_start = (void*)getThreadLocalStartOffset((size_t)t->stacktop);
if (tlsloadsize)
memcpy(t->stacktop, __tdata_lma, tlsloadsize);
memcpy(tdata_start, __tdata_lma, tlsloadsize);
if (tbsssize)
memset((u8*)t->stacktop+tlsloadsize, 0, tbsssize);
memset(tdata_start + tlsloadsize, 0, tbsssize);
// Set up child thread's reent struct, inheriting standard file handles
_REENT_INIT_PTR(&t->reent);