/*
* %CopyrightBegin%
*
* Copyright Ericsson AB 2001-2013. All Rights Reserved.
*
* The contents of this file are subject to the Erlang Public License,
* Version 1.1, (the "License"); you may not use this file except in
* compliance with the License. You should have received a copy of the
* Erlang Public License along with this software. If not, it can be
* retrieved online at http://www.erlang.org/.
*
* Software distributed under the License is distributed on an "AS IS"
* basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See
* the License for the specific language governing rights and limitations
* under the License.
*
* %CopyrightEnd%
*/
/*
* hipe_mode_switch.c
*/
#ifdef HAVE_CONFIG_H
#include "config.h"
#endif
#include "sys.h"
#include "erl_vm.h"
#include "global.h"
#include "erl_process.h"
#include "beam_load.h" /* which includes beam_opcodes.h */
#include "beam_catches.h"
#include "hipe_mode_switch.h"
#include "bif.h"
#include "error.h"
#include "hipe_stack.h"
#include "hipe_bif0.h" /* hipe_mfa_info_table_init() */
#if defined(ERTS_ENABLE_LOCK_CHECK) && defined(ERTS_SMP)
# define ERTS_SMP_REQ_PROC_MAIN_LOCK(P) \
if ((P)) erts_proc_lc_require_lock((P), ERTS_PROC_LOCK_MAIN)
# define ERTS_SMP_UNREQ_PROC_MAIN_LOCK(P) \
if ((P)) erts_proc_lc_unrequire_lock((P), ERTS_PROC_LOCK_MAIN)
#else
# define ERTS_SMP_REQ_PROC_MAIN_LOCK(P)
# define ERTS_SMP_UNREQ_PROC_MAIN_LOCK(P)
#endif
/*
* Internal debug support.
* #define HIPE_DEBUG to the desired debug level:
* 0 no checks
* 1 check PCB consistency at mode-switches
* 2 log commands and results at mode-switches
* 3 log commands, results, and PCB contents at mode-switches
*
* TODO: check PCB consistency at native BIF calls
*/
int hipe_modeswitch_debug = 0;
#define HIPE_DEBUG 0
#if HIPE_DEBUG > 1 /* include DPRINTF() logging */
#define DPRINTF(fmt, args...) \
do { \
if (hipe_modeswitch_debug > 0) { \
printf("%s, line %u: " fmt "\r\n", __FUNCTION__, __LINE__ , ##args); \
fflush(stdout); \
} \
} while (0)
static const char *code_str(unsigned code)
{
static const char *cmd_str[] = {
"call from beam",
"return from beam",
"throw from beam",
"resume from beam",
"return to beam",
"call to beam",
"throw to beam",
"suspend to beam",
"wait from native",
"wait_timeout from native",
"trap from native",
"call closure from beam",
"call closure to beam",
};
unsigned cmd = code & 0xFF;
if (cmd < (sizeof(cmd_str)/sizeof(cmd_str[0])))
return cmd_str[cmd];
else
return "???";
}
#else /* HIPE_DEBUG > 1 */
#define DPRINTF(fmt, args...) do{}while(0)
#endif /* HIPE_DEBUG > 1 */
#if HIPE_DEBUG > 0 /* include HIPE_ASSERT and PCB checking */
static void __noreturn
hipe_abort(const char *expr, const char *file, unsigned line)
{
erl_exit(1, "ASSERTION FAILED, file %s, line %u: %s\r\n", file, line, expr);
}
#define HIPE_ASSERT3(expr, file, line) \
do { \
if (!(expr)) \
hipe_abort(#expr, file, line); \
} while (0)
#define HIPE_ASSERT(expr) HIPE_ASSERT3(expr, __FILE__, __LINE__)
void hipe_check_pcb(Process *p, const char *file, unsigned line)
{
#if HIPE_DEBUG > 2
if (hipe_modeswitch_debug > 0) {
printf("%s, line %u: p %p = {htop %p, stop %p, nstack %p, nsp %p, nstend %p}\r\n", file, line, p, p->htop, p->stop, p->hipe.nstack, p->hipe.nsp, p->hipe.nstend);
}
#endif
HIPE_ASSERT3(p != NULL, file, line);
HIPE_ASSERT3(p->htop <= p->stop, file, line);
HIPE_ASSERT3(p->hipe.nstack <= p->hipe.nstend, file, line);
HIPE_ASSERT3(p->hipe.nsp >= p->hipe.nstack, file, line);
HIPE_ASSERT3(p->hipe.nsp <= p->hipe.nstend, file, line);
}
#define HIPE_CHECK_PCB(P) hipe_check_pcb((P), __FILE__, __LINE__)
#else /* HIPE_DEBUG > 0 */
#define HIPE_ASSERT(expr) do{}while(0)
#define HIPE_CHECK_PCB(P) do{}while(0)
#endif /* HIPE_DEBUG > 0 */
/* ensure that at least nwords words are available on the native stack */
static void hipe_check_nstack(Process *p, unsigned nwords);
#if defined(__sparc__)
#include "hipe_sparc_glue.h"
#elif defined(__i386__)
#include "hipe_x86_glue.h"
#elif defined(__x86_64__)
#include "hipe_amd64_glue.h"
#elif defined(__powerpc__) || defined(__ppc__) || defined(__powerpc64__)
#include "hipe_ppc_glue.h"
#elif defined(__arm__)
#include "hipe_arm_glue.h"
#endif
#define BeamOpCode(Op) ((Uint)BeamOp(Op))
Uint hipe_beam_pc_return[1]; /* needed in hipe_debug.c */
Uint hipe_beam_pc_throw[1]; /* needed in hipe_debug.c */
Uint hipe_beam_pc_resume[1]; /* needed by hipe_set_timeout() */
static Eterm hipe_beam_catch_throw;
void hipe_mode_switch_init(void)
{
hipe_arch_glue_init();
hipe_beam_pc_return[0] = BeamOpCode(op_hipe_trap_return);
hipe_beam_pc_throw[0] = BeamOpCode(op_hipe_trap_throw);
hipe_beam_pc_resume[0] = BeamOpCode(op_hipe_trap_resume);
hipe_beam_catch_throw =
make_catch(beam_catches_cons(hipe_beam_pc_throw, BEAM_CATCHES_NIL));
hipe_mfa_info_table_init();
}
void hipe_set_call_trap(Uint *bfun, void *nfun, int is_closure)
{
HIPE_ASSERT(bfun[-5] == BeamOpCode(op_i_func_info_IaaI));
bfun[0] =
is_closure
? BeamOpCode(op_hipe_trap_call_closure)
: BeamOpCode(op_hipe_trap_call);
bfun[-4] = (Uint)nfun;
}
void hipe_reserve_beam_trap_frame(Process *p, Eterm reg[], unsigned arity)
{
/* ensure that at least 2 words are available on the BEAM stack */
if ((p->stop - 2) < p->htop) {
DPRINTF("calling gc to reserve BEAM stack size");
p->fcalls -= erts_garbage_collect(p, 2, reg, arity);
ASSERT(!((p->stop - 2) < p->htop));
}
p->stop -= 2;
p->stop[0] = NIL;
p->stop[1] = NIL;
}
static __inline__ void
hipe_push_beam_trap_frame(Process *p, Eterm reg[], unsigned arity)
{
if (p->flags & F_DISABLE_GC) {
/* Trap frame already reserved */
ASSERT(p->stop[0] == NIL && p->stop[1] == NIL);
}
else {
if ((p->stop - 2) < p->htop) {
DPRINTF("calling gc to increase BEAM stack size");
p->fcalls -= erts_garbage_collect(p, 2, reg, arity);
ASSERT(!((p->stop - 2) < p->htop));
}
p->stop -= 2;
}
p->stop[1] = hipe_beam_catch_throw;
p->stop[0] = make_cp(p->cp);
++p->catches;
p->cp = hipe_beam_pc_return;
}
void hipe_unreserve_beam_trap_frame(Process *p)
{
ASSERT(p->stop[0] == NIL && p->stop[1] == NIL);
p->stop += 2;
}
static __inline__ void hipe_pop_beam_trap_frame(Process *p)
{
p->cp = cp_val(p->stop[0]);
--p->catches;
p->stop += 2;
}
Process *hipe_mode_switch(Process *p, unsigned cmd, Eterm reg[])
{
unsigned result;
#if NR_ARG_REGS > 5
/* When NR_ARG_REGS > 5, we need to protect the process' input
reduction count (which BEAM stores in def_arg_reg[5]) from
being clobbered by the arch glue code. */
Eterm reds_in = p->def_arg_reg[5];
#endif
#if NR_ARG_REGS > 4
Eterm o_reds = p->def_arg_reg[4];
#endif
p->i = NULL;
/* Set current_function to undefined. stdlib hibernate tests rely on it. */
p->current = NULL;
DPRINTF("cmd == %#x (%s)", cmd, code_str(cmd));
HIPE_CHECK_PCB(p);
p->arity = 0;
switch (cmd & 0xFF) {
case HIPE_MODE_SWITCH_CMD_CALL: {
/* BEAM calls a native code function */
unsigned arity = cmd >> 8;
/* p->hipe.ncallee set in beam_emu */
if (p->cp == hipe_beam_pc_return) {
/* Native called BEAM, which now tailcalls native. */
hipe_pop_beam_trap_frame(p);
result = hipe_tailcall_to_native(p, arity, reg);
break;
}
DPRINTF("calling %#lx/%u", (long)p->hipe.ncallee, arity);
result = hipe_call_to_native(p, arity, reg);
break;
}
case HIPE_MODE_SWITCH_CMD_CALL_CLOSURE: {
/* BEAM calls a native code closure */
unsigned arity = cmd >> 8; /* #formals + #fvs (closure not counted) */
Eterm fun;
ErlFunThing *funp;
/* drop the fvs, move the closure, correct arity */
fun = reg[arity];
HIPE_ASSERT(is_fun(fun));
funp = (ErlFunThing*)fun_val(fun);
HIPE_ASSERT(funp->num_free <= arity);
arity -= funp->num_free; /* arity == #formals */
reg[arity] = fun;
++arity; /* correct for having added the closure */
/* HIPE_ASSERT(p->hipe.ncallee == (void(*)(void))funp->native_address); */
/* just like a normal call from now on */
/* p->hipe.ncallee set in beam_emu */
if (p->cp == hipe_beam_pc_return) {
/* Native called BEAM, which now tailcalls native. */
hipe_pop_beam_trap_frame(p);
result = hipe_tailcall_to_native(p, arity, reg);
break;
}
DPRINTF("calling %#lx/%u", (long)p->hipe.ncallee, arity);
result = hipe_call_to_native(p, arity, reg);
break;
}
case HIPE_MODE_SWITCH_CMD_THROW: {
/* BEAM just executed hipe_beam_pc_throw[] */
/* Native called BEAM, which now throws an exception back to native. */
DPRINTF("beam throws freason %#lx fvalue %#lx", p->freason, p->fvalue);
hipe_pop_beam_trap_frame(p);
do_throw_to_native:
p->def_arg_reg[0] = exception_tag[GET_EXC_CLASS(p->freason)];
hipe_find_handler(p);
result = hipe_throw_to_native(p);
break;
}
case HIPE_MODE_SWITCH_CMD_RETURN: {
/* BEAM just executed hipe_beam_pc_return[] */
/* Native called BEAM, which now returns back to native. */
/* pop trap frame off estack */
hipe_pop_beam_trap_frame(p);
p->def_arg_reg[0] = reg[0];
result = hipe_return_to_native(p);
break;
}
do_resume:
case HIPE_MODE_SWITCH_CMD_RESUME: {
/* BEAM just executed hipe_beam_pc_resume[] */
/* BEAM called native, which suspended. */
if (p->flags & F_TIMO) {
/* XXX: The process will immediately execute 'clear_timeout',
repeating these two statements. Remove them? */
p->flags &= ~F_TIMO;
JOIN_MESSAGE(p);
p->def_arg_reg[0] = 0; /* make_small(0)? */
} else
p->def_arg_reg[0] = 1; /* make_small(1)? */
result = hipe_return_to_native(p);
break;
}
default:
erl_exit(1, "hipe_mode_switch: cmd %#x\r\n", cmd);
}
do_return_from_native:
DPRINTF("result == %#x (%s)", result, code_str(result));
HIPE_CHECK_PCB(p);
switch (result) {
case HIPE_MODE_SWITCH_RES_RETURN: {
hipe_return_from_native(p);
reg[0] = p->def_arg_reg[0];
DPRINTF("returning with r(0) == %#lx", reg[0]);
break;
}
case HIPE_MODE_SWITCH_RES_THROW: {
DPRINTF("native throws freason %#lx fvalue %#lx", p->freason, p->fvalue);
hipe_throw_from_native(p);
break;
}
case HIPE_MODE_SWITCH_RES_TRAP: {
/*
* Native code called a BIF, which "failed" with a TRAP to BEAM.
* Prior to returning, the BIF stored (see BIF_TRAP<N>):
* the callee's address in p->i
* the callee's parameters in reg[0..2]
* the callee's arity in p->arity (for BEAM gc purposes)
*
* We need to remove the BIF's parameters from the native
* stack: to this end hipe_${ARCH}_glue.S stores the BIF's
* arity in p->hipe.narity.
*
* If the BIF emptied the stack (typically hibernate), p->hipe.nstack
* is NULL and there is no need to get rid of stacked parameters.
*/
unsigned int i, is_recursive = 0;
if (p->hipe.nstack != NULL) {
ASSERT(p->hipe.nsp != NULL);
is_recursive = hipe_trap_from_native_is_recursive(p);
}
else {
/* Some architectures (risc) need this re-reset of nsp as the
* BIF wrapper do not detect stack change and causes an obsolete
* stack pointer to be saved in p->hipe.nsp before return to us.
*/
p->hipe.nsp = NULL;
}
/* Schedule next process if current process was hibernated or is waiting
for messages */
if (p->flags & F_HIBERNATE_SCHED) {
p->flags &= ~F_HIBERNATE_SCHED;
goto do_schedule;
}
if (!(erts_smp_atomic32_read_acqb(&p->state) & ERTS_PSFLG_ACTIVE)) {
for (i = 0; i < p->arity; ++i)
p->arg_reg[i] = reg[i];
goto do_schedule;
}
if (is_recursive)
hipe_push_beam_trap_frame(p, reg, p->arity);
result = HIPE_MODE_SWITCH_RES_CALL;
break;
}
case HIPE_MODE_SWITCH_RES_CALL: {
/* Native code calls or tailcalls BEAM.
*
* p->i is the callee's BEAM code
* p->arity is the callee's arity
* p->def_arg_reg[] contains the register parameters
* p->hipe.nsp[] contains the stacked parameters
*/
if (hipe_call_from_native_is_recursive(p, reg)) {
/* BEAM called native, which now calls BEAM */
hipe_push_beam_trap_frame(p, reg, p->arity);
}
break;
}
case HIPE_MODE_SWITCH_RES_CALL_CLOSURE: {
/* Native code calls or tailcalls a closure in BEAM
*
* In native code a call to a closure of arity n looks like
* F(A1, ..., AN, Closure),
* The BEAM code for a closure expects to get:
* F(A1, ..., AN, FV1, ..., FVM, Closure)
* (Where Ai is argument i and FVj is free variable j)
*
* p->hipe.closure contains the closure
* p->def_arg_reg[] contains the register parameters
* p->hipe.nsp[] contains the stacked parameters
*/
ErlFunThing *closure;
unsigned num_free, arity, i, is_recursive;
HIPE_ASSERT(is_fun(p->hipe.closure));
closure = (ErlFunThing*)fun_val(p->hipe.closure);
num_free = closure->num_free;
arity = closure->fe->arity;
/* Store the arity in p->arity for the stack popping. */
/* Note: we already have the closure so only need to move arity
values to reg[]. However, there are arity+1 parameters in the
native code state that need to be removed. */
p->arity = arity+1; /* +1 for the closure */
/* Get parameters, don't do GC just yet. */
is_recursive = hipe_call_from_native_is_recursive(p, reg);
if ((Sint)closure->fe->address[-1] < 0) {
/* Unloaded. Let beam_emu.c:call_fun() deal with it. */
result = HIPE_MODE_SWITCH_RES_CALL_CLOSURE;
} else {
/* The BEAM code is present. Prepare to call it. */
/* Append the free vars after the actual parameters. */
for (i = 0; i < num_free; ++i)
reg[arity+i] = closure->env[i];
/* Update arity to reflect the new parameters. */
arity += i;
/* Make a call to the closure's BEAM code. */
p->i = closure->fe->address;
/* Change result code to the faster plain CALL type. */
result = HIPE_MODE_SWITCH_RES_CALL;
}
/* Append the closure as the last parameter. Don't increment arity. */
reg[arity] = p->hipe.closure;
if (is_recursive) {
/* BEAM called native, which now calls BEAM.
Need to put a trap-frame on the beam stack.
This may cause GC, which is safe now that
the arguments, free vars, and most
importantly the closure, all are in reg[]. */
hipe_push_beam_trap_frame(p, reg, arity+1);
}
break;
}
case HIPE_MODE_SWITCH_RES_SUSPEND: {
p->i = hipe_beam_pc_resume;
p->arity = 0;
goto do_schedule;
}
case HIPE_MODE_SWITCH_RES_WAIT:
case HIPE_MODE_SWITCH_RES_WAIT_TIMEOUT: {
/* same semantics, different debug trace messages */
#ifdef ERTS_SMP
/* XXX: BEAM has different entries for the locked and unlocked
cases. HiPE doesn't, so we must check dynamically. */
if (p->hipe_smp.have_receive_locks)
p->hipe_smp.have_receive_locks = 0;
else
erts_smp_proc_lock(p, ERTS_PROC_LOCKS_MSG_RECEIVE);
#endif
p->i = hipe_beam_pc_resume;
p->arity = 0;
erts_smp_atomic32_read_band_relb(&p->state,
~ERTS_PSFLG_ACTIVE);
erts_smp_proc_unlock(p, ERTS_PROC_LOCKS_MSG_RECEIVE);
do_schedule:
{
#if !(NR_ARG_REGS > 5)
int reds_in = p->def_arg_reg[5];
#endif
ERTS_SMP_UNREQ_PROC_MAIN_LOCK(p);
p = schedule(p, reds_in - p->fcalls);
ERTS_SMP_REQ_PROC_MAIN_LOCK(p);
#ifdef ERTS_SMP
p->hipe_smp.have_receive_locks = 0;
reg = p->scheduler_data->x_reg_array;
#endif
}
{
Eterm *argp;
int i;
argp = p->arg_reg;
for (i = p->arity; --i >= 0;)
reg[i] = argp[i];
}
{
#if !(NR_ARG_REGS > 5)
Eterm reds_in;
#endif
#if !(NR_ARG_REGS > 4)
Eterm o_reds;
#endif
reds_in = p->fcalls;
o_reds = 0;
if (ERTS_PROC_GET_SAVED_CALLS_BUF(p)) {
o_reds = reds_in;
reds_in = 0;
p->fcalls = 0;
}
p->def_arg_reg[4] = o_reds;
p->def_arg_reg[5] = reds_in;
if (p->i == hipe_beam_pc_resume) {
p->i = NULL;
p->arity = 0;
goto do_resume;
}
}
HIPE_CHECK_PCB(p);
result = HIPE_MODE_SWITCH_RES_CALL;
p->def_arg_reg[3] = result;
return p;
}
case HIPE_MODE_SWITCH_RES_APPLY: {
Eterm mfa[3], args;
unsigned int arity;
void *address;
hipe_pop_params(p, 3, &mfa[0]);
/* Unroll the arglist onto reg[]. */
args = mfa[2];
arity = 0;
while (is_list(args)) {
if (arity < 255) {
reg[arity++] = CAR(list_val(args));
args = CDR(list_val(args));
} else
goto do_apply_fail;
}
if (is_not_nil(args))
goto do_apply_fail;
/* find a native code entry point for {M,F,A} for a remote call */
address = hipe_get_remote_na(mfa[0], mfa[1], arity);
if (!address)
goto do_apply_fail;
p->hipe.ncallee = (void(*)(void)) address;
result = hipe_tailcall_to_native(p, arity, reg);
goto do_return_from_native;
do_apply_fail:
p->freason = BADARG;
goto do_throw_to_native;
}
default:
erl_exit(1, "hipe_mode_switch: result %#x\r\n", result);
}
HIPE_CHECK_PCB(p);
p->def_arg_reg[3] = result;
#if NR_ARG_REGS > 4
p->def_arg_reg[4] = o_reds;
#endif
#if NR_ARG_REGS > 5
p->def_arg_reg[5] = reds_in;
#endif
return p;
}
#define HIPE_INITIAL_NSTACK_SIZE 128
/* PRE: size is zero or a power of two */
static unsigned hipe_next_nstack_size(unsigned size)
{
return size ? size * 2 : HIPE_INITIAL_NSTACK_SIZE;
}
#if 0 && defined(HIPE_NSTACK_GROWS_UP)
#define hipe_nstack_avail(p) ((p)->hipe.nstend - (p)->hipe.nsp)
void hipe_inc_nstack(Process *p)
{
Eterm *old_nstack = p->hipe.nstack;
unsigned old_size = p->hipe.nstend - old_nstack;
unsigned new_size = hipe_next_nstack_size(old_size);
Eterm *new_nstack = erts_realloc(ERTS_ALC_T_HIPE,
(char *) old_nstack,
new_size*sizeof(Eterm));
p->hipe.nstend = new_nstack + new_size;
if (new_nstack != old_nstack) {
p->hipe.nsp = new_nstack + (p->hipe.nsp - old_nstack);
p->hipe.nstack = new_nstack;
if (p->hipe.nstgraylim)
p->hipe.nstgraylim =
new_nstack + (p->hipe.nstgraylim - old_nstack);
if (p->hipe.nstblacklim)
p->hipe.nstblacklim =
new_nstack + (p->hipe.nstblacklim - old_nstack);
}
}
#endif
#if defined(HIPE_NSTACK_GROWS_DOWN)
#define hipe_nstack_avail(p) ((unsigned)((p)->hipe.nsp - (p)->hipe.nstack))
void hipe_inc_nstack(Process *p)
{
unsigned old_size = p->hipe.nstend - p->hipe.nstack;
unsigned new_size = hipe_next_nstack_size(old_size);
Eterm *new_nstack = erts_alloc(ERTS_ALC_T_HIPE, new_size*sizeof(Eterm));
unsigned used_size = p->hipe.nstend - p->hipe.nsp;
sys_memcpy(new_nstack+new_size-used_size, p->hipe.nsp, used_size*sizeof(Eterm));
if (p->hipe.nstgraylim)
p->hipe.nstgraylim = new_nstack + new_size - (p->hipe.nstend - p->hipe.nstgraylim);
if (p->hipe.nstblacklim)
p->hipe.nstblacklim = new_nstack + new_size - (p->hipe.nstend - p->hipe.nstblacklim);
if (p->hipe.nstack)
erts_free(ERTS_ALC_T_HIPE, p->hipe.nstack);
p->hipe.nstack = new_nstack;
p->hipe.nstend = new_nstack + new_size;
p->hipe.nsp = new_nstack + new_size - used_size;
}
#endif
void hipe_empty_nstack(Process *p)
{
if (p->hipe.nstack) {
erts_free(ERTS_ALC_T_HIPE, p->hipe.nstack);
}
p->hipe.nstgraylim = NULL;
p->hipe.nsp = NULL;
p->hipe.nstack = NULL;
p->hipe.nstend = NULL;
}
static void hipe_check_nstack(Process *p, unsigned nwords)
{
while (hipe_nstack_avail(p) < nwords)
hipe_inc_nstack(p);
}
void hipe_set_closure_stub(ErlFunEntry *fe, unsigned num_free)
{
unsigned arity;
arity = fe->arity;
fe->native_address = (Eterm*) hipe_closure_stub_address(arity);
}
Eterm hipe_build_stacktrace(Process *p, struct StackTrace *s)
{
int depth, i;
Uint heap_size;
Eterm *hp, *hp_end, mfa, m, f, head, *next_p, next;
const void *ra;
unsigned int a;
depth = s->depth;
if (depth < 1)
return NIL;
heap_size = 7 * depth; /* each [{M,F,A,[]}|_] is 2+5 == 7 words */
hp = HAlloc(p, heap_size);
hp_end = hp + heap_size;
head = NIL;
next_p = &head;
for (i = 0; i < depth; ++i) {
ra = (const void*)s->trace[i];
if (!hipe_find_mfa_from_ra(ra, &m, &f, &a))
continue;
mfa = TUPLE4(hp, m, f, make_small(a), NIL);
hp += 5;
next = CONS(hp, mfa, NIL);
*next_p = next;
next_p = &CDR(list_val(next));
hp += 2;
}
HRelease(p, hp_end, hp);
return head;
}