/* -----------------------------------------------------------------------------
 *
 * (c) The GHC Team, 1998-2004
 *
 * Canned Heap-Check and Stack-Check sequences.
 *
 * This file is written in a subset of C--, extended with various
 * features specific to GHC.  It is compiled by GHC directly.  For the
 * syntax of .cmm files, see the parser in ghc/compiler/GHC/Cmm/Parser.y.
 *
 * ---------------------------------------------------------------------------*/

#include "Cmm.h"
#include "Updates.h"
#include "SMPClosureOps.h"

#if defined(__PIC__)
import pthread_mutex_unlock;
#endif
import AcquireSRWLockExclusive;
import ReleaseSRWLockExclusives;

#if !defined(UnregisterisedCompiler)
import CLOSURE g0;
import CLOSURE large_alloc_lim;
import CLOSURE stg_MSG_THROWTO_info;
import CLOSURE stg_MVAR_DIRTY_info;
import CLOSURE stg_WHITEHOLE_info;
import CLOSURE stg_arg_bitmaps;
import CLOSURE stg_block_putmvar_info;
import CLOSURE stg_block_readmvar_info;
import CLOSURE stg_block_takemvar_info;
import CLOSURE stg_block_throwto_info;
import CLOSURE stg_enter_info;
import CLOSURE stg_gc_fun_info;
import CLOSURE stg_gc_prim_p_ll_info;
import CLOSURE stg_gc_prim_pp_ll_info;
import CLOSURE stg_ret_d_info;
import CLOSURE stg_ret_f_info;
import CLOSURE stg_ret_l_info;
import CLOSURE stg_ret_n_info;
import CLOSURE stg_ret_p_info;
#endif

/* Stack/Heap Check Failure
 * ------------------------
 *
 * Both heap and stack check failures end up in the same place, so
 * that we can share the code for the failure case when a proc needs
 * both a stack check and a heap check (a common case).
 *
 * So when we get here, we have to tell the difference between a stack
 * check failure and a heap check failure.  The code for the checks
 * looks like this:

        if (Sp - 16 < SpLim) goto c1Tf;
        Hp = Hp + 16;
        if (Hp > HpLim) goto c1Th;
        ...
    c1Th:
        HpAlloc = 16;
        goto c1Tf;
    c1Tf: jump stg_gc_enter_1 ();

 * Note that Sp is not decremented by the check, whereas Hp is.  The
 * reasons for this seem to be largely historic, I can't think of a
 * good reason not to decrement Sp at the check too. (--SDM)
 *
 * Note that HpLim may be set to zero arbitrarily by the timer signal
 * or another processor to trigger a context switch via heap check
 * failure.
 *
 * The job of these fragments (stg_gc_enter_1 and friends) is to
 *   1. Leave no slop in the heap, so Hp must be retreated if it was
 *      incremented by the check.  No-slop is a requirement for LDV
 *      profiling, at least.
 *   2. If a heap check failed, try to grab another heap block from
 *      the nursery and continue.
 *   3. otherwise, return to the scheduler with StackOverflow,
 *      HeapOverflow, or ThreadYielding as appropriate.
 *
 * We can tell whether Hp was incremented, because HpAlloc is
 * non-zero: HpAlloc is required to be zero at all times unless a
 * heap-check just failed, which is why the stack-check failure case
 * does not set HpAlloc (see code fragment above).  So that covers (1).
 * HpAlloc is zeroed in LOAD_THREAD_STATE().
 *
 * If Hp > HpLim, then either (a) we have reached the end of the
 * current heap block, or (b) HpLim == 0 and we should yield.  Hence
 * check Hp > HpLim first, and then HpLim == 0 to decide whether to
 * return ThreadYielding or try to grab another heap block from the
 * nursery.
 *
 * If Hp <= HpLim, then this must be a StackOverflow.  The scheduler
 * will either increase the size of our stack, or raise an exception if
 * the stack is already too big.
 */

#define PRE_RETURN(why,what_next)                       \
  StgTSO_what_next(CurrentTSO) = what_next::I16;        \
  StgRegTable_rRet(BaseReg) = why;                      \
  R1 = BaseReg;

/* Remember that the return address is *removed* when returning to a
 * ThreadRunGHC thread.
 */

stg_gc_noregs
{
    W_ ret;

    DEBUG_ONLY(foreign "C" heapCheckFail());
    if (Hp > HpLim) {
        Hp = Hp - HpAlloc/*in bytes*/;
        if (HpLim == 0) {
                ret = ThreadYielding;
                goto sched;
        }
        if (HpAlloc <= BLOCK_SIZE
            && bdescr_link(CurrentNursery) != NULL) {
            HpAlloc = 0;
            CLOSE_NURSERY();
            Capability_total_allocated(MyCapability()) =
              Capability_total_allocated(MyCapability()) +
              %zx64(BYTES_TO_WDS(bdescr_free(CurrentNursery) -
                                 bdescr_start(CurrentNursery)));
            CurrentNursery = bdescr_link(CurrentNursery);
            bdescr_free(CurrentNursery) = bdescr_start(CurrentNursery);
            OPEN_NURSERY();

            CInt context_switch, interrupt;
            context_switch = %relaxed Capability_context_switch(MyCapability());
            interrupt = %relaxed Capability_interrupt(MyCapability());

            if (context_switch != 0 :: CInt ||
                interrupt != 0 :: CInt ||
                (StgTSO_alloc_limit(CurrentTSO) `lt` (0::I64) &&
                 (TO_W_(StgTSO_flags(CurrentTSO)) & TSO_ALLOC_LIMIT) != 0)) {
                ret = ThreadYielding;
                goto sched;
            } else {
                jump %ENTRY_CODE(Sp(0)) [];
            }
        } else {
            ret = HeapOverflow;
            goto sched;
        }
    } else {
        if (CHECK_GC()) {
            ret = HeapOverflow;
        } else {
            ret = StackOverflow;
        }
    }
  sched:
    PRE_RETURN(ret,ThreadRunGHC);
    jump stg_returnToSched [R1];
}

#define HP_GENERIC                              \
    PRE_RETURN(HeapOverflow, ThreadRunGHC)      \
    jump stg_returnToSched [R1];

#define BLOCK_GENERIC                           \
    PRE_RETURN(ThreadBlocked,  ThreadRunGHC)    \
    jump stg_returnToSched [R1];

#define YIELD_GENERIC                           \
    PRE_RETURN(ThreadYielding, ThreadRunGHC)    \
    jump stg_returnToSched [R1];

#define BLOCK_BUT_FIRST(c)                      \
    PRE_RETURN(ThreadBlocked, ThreadRunGHC)     \
    R2 = c;                                     \
    jump stg_returnToSchedButFirst [R1,R2,R3];

#define YIELD_TO_INTERPRETER                    \
    PRE_RETURN(ThreadYielding, ThreadInterpret) \
    jump stg_returnToSchedNotPaused [R1];

/* -----------------------------------------------------------------------------
   Heap checks in thunks/functions.

   In these cases, node always points to the function closure.  This gives
   us an easy way to return to the function: just leave R1 on the top of
   the stack, and have the scheduler enter it to return.

   There are canned sequences for 'n' pointer values in registers.
   -------------------------------------------------------------------------- */

INFO_TABLE_RET ( stg_enter, RET_SMALL, W_ info_ptr, P_ closure )
    return (/* no return values */)
{
    ENTER(closure);
}

__stg_gc_enter_1 (P_ node)
{
    jump stg_gc_noregs (stg_enter_info, node) ();
}

/* -----------------------------------------------------------------------------
   Canned heap checks for primitives.

   We can't use stg_gc_fun because primitives are not functions, so
   these fragments let us save some boilerplate heap-check-failure
   code in a few common cases.
   -------------------------------------------------------------------------- */

stg_gc_prim (W_ fun)
{
    call stg_gc_noregs ();
    jump fun();
}

stg_gc_prim_p (P_ arg, W_ fun)
{
    call stg_gc_noregs ();
    jump fun(arg);
}

stg_gc_prim_pp (P_ arg1, P_ arg2, W_ fun)
{
    call stg_gc_noregs ();
    jump fun(arg1,arg2);
}

stg_gc_prim_n (W_ arg, W_ fun)
{
    call stg_gc_noregs ();
    jump fun(arg);
}

INFO_TABLE_RET(stg_gc_prim_p_ll, RET_SMALL, W_ info, P_ arg, W_ fun)
    /* explicit stack */
{
    W_ fun;
    P_ arg;
    fun = Sp(2);
    arg = Sp(1);
    Sp_adj(3);
    R1 = arg;
    jump fun [R1];
}

stg_gc_prim_p_ll
{
    W_ fun;
    P_ arg;
    fun = R2;
    arg = R1;
    Sp_adj(-3);
    Sp(2) = fun;
    Sp(1) = arg;
    Sp(0) = stg_gc_prim_p_ll_info;
    jump stg_gc_noregs [];
}

INFO_TABLE_RET(stg_gc_prim_pp_ll, RET_SMALL, W_ info, P_ arg1, P_ arg2, W_ fun)
    /* explicit stack */
{
    W_ fun;
    P_ arg1, arg2;
    fun = Sp(3);
    arg2 = Sp(2);
    arg1 = Sp(1);
    Sp_adj(4);
    R1 = arg1;
    R2 = arg2;
    jump fun [R1, R2];
}

stg_gc_prim_pp_ll
{
    W_ fun;
    P_ arg1, arg2;
    fun = R3;
    arg1 = R1;
    arg2 = R2;
    Sp_adj(-4);
    Sp(3) = fun;
    Sp(2) = arg2;
    Sp(1) = arg1;
    Sp(0) = stg_gc_prim_pp_ll_info;
    jump stg_gc_noregs [];
}

/* -----------------------------------------------------------------------------
   Info tables for returning values of various types.  These are used
   when we want to push a frame on the stack that will return a value
   to the frame underneath it.
   -------------------------------------------------------------------------- */

INFO_TABLE_RET ( stg_ret_v, RET_SMALL, W_ info_ptr )
    return (/* no return values */)
{
    return ();
}

INFO_TABLE_RET ( stg_ret_p, RET_SMALL, W_ info_ptr, P_ ptr )
    return (/* no return values */)
{
    return (ptr);
}

INFO_TABLE_RET ( stg_ret_n, RET_SMALL, W_ info_ptr, W_ nptr )
    return (/* no return values */)
{
    return (nptr);
}

INFO_TABLE_RET ( stg_ret_f, RET_SMALL, W_ info_ptr, F_ f )
    return (/* no return values */)
{
    return (f);
}

INFO_TABLE_RET ( stg_ret_d, RET_SMALL, W_ info_ptr, D_ d )
    return (/* no return values */)
{
    return (d);
}

INFO_TABLE_RET ( stg_ret_l, RET_SMALL, W_ info_ptr, L_ l )
    return (/* no return values */)
{
    return (l);
}

/* -----------------------------------------------------------------------------
   Canned heap-check failures for case alts, where we have some values
   in registers or on the stack according to the NativeReturn
   convention.
   -------------------------------------------------------------------------- */


/*-- void return ------------------------------------------------------------ */

/*-- R1 is a GC pointer, but we don't enter it ----------------------- */

stg_gc_unpt_r1 return (P_ ptr) /* NB. return convention */
{
    jump stg_gc_noregs (stg_ret_p_info, ptr) ();
}

/*-- R1 is unboxed -------------------------------------------------- */

stg_gc_unbx_r1 return (W_ nptr) /* NB. return convention */
{
    jump stg_gc_noregs (stg_ret_n_info, nptr) ();
}

/*-- F1 contains a float ------------------------------------------------- */

stg_gc_f1 return (F_ f)
{
    jump stg_gc_noregs (stg_ret_f_info, f) ();
}

/*-- D1 contains a double ------------------------------------------------- */

stg_gc_d1 return (D_ d)
{
    jump stg_gc_noregs (stg_ret_d_info, d) ();
}


/*-- L1 contains an int64 ------------------------------------------------- */

stg_gc_l1 return (L_ l)
{
    jump stg_gc_noregs (stg_ret_l_info, l) ();
}

/*-- Unboxed tuples with multiple pointers -------------------------------- */

stg_gc_pp return (P_ arg1, P_ arg2)
{
    call stg_gc_noregs();
    return (arg1,arg2);
}

stg_gc_ppp return (P_ arg1, P_ arg2, P_ arg3)
{
    call stg_gc_noregs();
    return (arg1,arg2,arg3);
}

stg_gc_pppp return (P_ arg1, P_ arg2, P_ arg3, P_ arg4)
{
    call stg_gc_noregs();
    return (arg1,arg2,arg3,arg4);
}

/* -----------------------------------------------------------------------------
   Generic function entry heap check code.

   At a function entry point, the arguments are as per the calling convention,
   i.e. some in regs and some on the stack.  There may or may not be
   a pointer to the function closure in R1 - if there isn't, then the heap
   check failure code in the function will arrange to load it.

   The function's argument types are described in its info table, so we
   can just jump to this bit of generic code to save away all the
   registers and return to the scheduler.

   This code arranges the stack like this:

         |        ....         |
         |        args         |
         +---------------------+
         |      f_closure      |
         +---------------------+
         |        size         |
         +---------------------+
         |   stg_gc_fun_info   |
         +---------------------+

   The size is the number of words of arguments on the stack, and is cached
   in the frame in order to simplify stack walking: otherwise the size of
   this stack frame would have to be calculated by looking at f's info table.

   -------------------------------------------------------------------------- */

__stg_gc_fun /* explicit stack */
{
    W_ size;
    W_ info;
    W_ type;

    info = %GET_FUN_INFO(UNTAG(R1));

    // cache the size
    type = TO_W_(StgFunInfoExtra_fun_type(info));
    if (type == ARG_GEN) {
        size = BITMAP_SIZE(StgFunInfoExtra_bitmap(info));
    } else {
        if (type == ARG_GEN_BIG) {
#if defined(TABLES_NEXT_TO_CODE)
            // bitmap field holds an offset
            size = StgLargeBitmap_size(
                      TO_W_(StgFunInfoExtraRev_bitmap_offset(info))
                      + %GET_ENTRY(UNTAG(R1)) /* ### */ );
#else
            size = StgLargeBitmap_size( StgFunInfoExtra_bitmap(info) );
#endif
        } else {
            size = BITMAP_SIZE(W_[stg_arg_bitmaps + WDS(type)]);
        }
    }

#if defined(NO_ARG_REGS)
    // we don't have to save any registers away
    Sp_adj(-3);
    Sp(2) = R1;
    Sp(1) = size;
    Sp(0) = stg_gc_fun_info;
    jump stg_gc_noregs [];
#else
    W_ type;
    type = TO_W_(StgFunInfoExtra_fun_type(info));
    // cache the size
    if (type == ARG_GEN || type == ARG_GEN_BIG) {
        // regs already saved by the heap check code
        Sp_adj(-3);
        Sp(2) = R1;
        Sp(1) = size;
        Sp(0) = stg_gc_fun_info;
        // DEBUG_ONLY(foreign "C" debugBelch("stg_fun_gc_gen(ARG_GEN)"););
        jump stg_gc_noregs [];
    } else {
        jump W_[stg_stack_save_entries + WDS(type)] [*]; // all regs live
            // jumps to stg_gc_noregs after saving stuff
    }
#endif /* !NO_ARG_REGS */
}


/* -----------------------------------------------------------------------------
   Generic Apply (return point)

   The dual to stg_fun_gc_gen (above): this fragment returns to the
   function, passing arguments in the stack and in registers
   appropriately.  The stack layout is given above.
   -------------------------------------------------------------------------- */

INFO_TABLE_RET ( stg_gc_fun, RET_FUN )
    /* explicit stack */
{
    R1 = Sp(2);
    Sp_adj(3);
#if defined(NO_ARG_REGS)
    // Minor optimisation: there are no argument registers to load up,
    // so we can just jump straight to the function's entry point.
    jump %GET_ENTRY(UNTAG(R1)) [R1];
#else
    W_ info;
    W_ type;

    info = %GET_FUN_INFO(UNTAG(R1));
    type = TO_W_(StgFunInfoExtra_fun_type(info));
    if (type == ARG_GEN || type == ARG_GEN_BIG) {
        jump StgFunInfoExtra_slow_apply(info) [R1];
    } else {
        if (type == ARG_BCO) {
            // cover this case just to be on the safe side
            Sp_adj(-2);
            Sp(1) = R1;
            Sp(0) = stg_apply_interp_info;
            jump stg_yield_to_interpreter [];
        } else {
            jump W_[stg_ap_stack_entries + WDS(type)] [R1];
        }
    }
#endif
}

/* -----------------------------------------------------------------------------
   Yields
   -------------------------------------------------------------------------- */

stg_yield_noregs
{
    YIELD_GENERIC;
}

/* -----------------------------------------------------------------------------
   Yielding to the interpreter... top of stack says what to do next.
   -------------------------------------------------------------------------- */

stg_yield_to_interpreter
{
    YIELD_TO_INTERPRETER;
}

/* -----------------------------------------------------------------------------
   Blocks
   -------------------------------------------------------------------------- */

stg_block_noregs
{
    BLOCK_GENERIC;
}

/* -----------------------------------------------------------------------------
 * takeMVar/putMVar-specific blocks
 *
 * Stack layout for a thread blocked in takeMVar/readMVar:
 *
 *       ret. addr
 *       ptr to MVar   (R1)
 *       stg_block_takemvar_info (or stg_block_readmvar_info)
 *
 * Stack layout for a thread blocked in putMVar:
 *
 *       ret. addr
 *       ptr to Value  (R2)
 *       ptr to MVar   (R1)
 *       stg_block_putmvar_info
 *
 * See PrimOps.cmm for a description of the workings of take/putMVar.
 *
 * -------------------------------------------------------------------------- */

INFO_TABLE_RET ( stg_block_takemvar, RET_SMALL, W_ info_ptr, P_ mvar )
    return ()
{
    jump stg_takeMVarzh(mvar);
}

// code fragment executed just before we return to the scheduler
stg_block_takemvar_finally
{
    W_ r1, r3;
    r1 = R1;
    r3 = R3;
    unlockClosure(r3, stg_MVAR_DIRTY_info);
    R1 = r1;
    R3 = r3;
    jump StgReturn [R1];
}

// Stack usage covered by RESERVED_STACK_WORDS
stg_block_takemvar /* mvar passed in R1 */
{
    Sp_adj(-2);
    Sp(1) = R1;
    Sp(0) = stg_block_takemvar_info;
    R3 = R1; // mvar communicated to stg_block_takemvar_finally in R3
    BLOCK_BUT_FIRST(stg_block_takemvar_finally);
}

INFO_TABLE_RET ( stg_block_readmvar, RET_SMALL, W_ info_ptr, P_ mvar )
    return ()
{
    jump stg_readMVarzh(mvar);
}

// code fragment executed just before we return to the scheduler
stg_block_readmvar_finally
{
    W_ r1, r3;
    r1 = R1;
    r3 = R3;
    unlockClosure(r3, stg_MVAR_DIRTY_info);
    R1 = r1;
    R3 = r3;
    jump StgReturn [R1];
}

stg_block_readmvar /* mvar passed in R1 */
{
    Sp_adj(-2);
    Sp(1) = R1;
    Sp(0) = stg_block_readmvar_info;
    R3 = R1; // mvar communicated to stg_block_readmvar_finally in R3
    BLOCK_BUT_FIRST(stg_block_readmvar_finally);
}

INFO_TABLE_RET( stg_block_putmvar, RET_SMALL, W_ info_ptr,
                P_ mvar, P_ val )
    return ()
{
    jump stg_putMVarzh(mvar, val);
}

// code fragment executed just before we return to the scheduler
stg_block_putmvar_finally
{
    W_ r1, r3;
    r1 = R1;
    r3 = R3;
    unlockClosure(r3, stg_MVAR_DIRTY_info);
    R1 = r1;
    R3 = r3;
    jump StgReturn [R1];
}

stg_block_putmvar (P_ mvar, P_ val)
{
    push (stg_block_putmvar_info, mvar, val) {
      R3 = R1; // mvar communicated to stg_block_putmvar_finally in R3
      BLOCK_BUT_FIRST(stg_block_putmvar_finally);
   }
}

stg_block_blackhole (P_ node)
{
    Sp_adj(-2);
    Sp(1) = node;
    Sp(0) = stg_enter_info;
    BLOCK_GENERIC;
}

INFO_TABLE_RET ( stg_block_throwto, RET_SMALL, W_ info_ptr,
                 P_ tso, P_ exception )
    return ()
{
    jump stg_killThreadzh(tso, exception);
}

stg_block_throwto_finally
{
    // unlock the throwto message, but only if it wasn't already
    // unlocked.  It may have been unlocked if we revoked the message
    // due to an exception being raised during threadPaused().
    if (StgHeader_info(StgTSO_block_info(CurrentTSO)) == stg_WHITEHOLE_info) {
        W_ r1;
        r1 = R1;
        unlockClosure(StgTSO_block_info(CurrentTSO), stg_MSG_THROWTO_info);
        R1 = r1;
    }
    jump StgReturn [R1];
}

stg_block_throwto (P_ tso, P_ exception)
{
    push (stg_block_throwto_info, tso, exception) {
       BLOCK_BUT_FIRST(stg_block_throwto_finally);
    }
}

#if defined(mingw32_HOST_OS)
INFO_TABLE_RET ( stg_block_async, RET_SMALL, W_ info_ptr, W_ ares )
    return ()
{
    W_ len, errC;

    len = TO_W_(StgAsyncIOResult_len(ares));
    errC = TO_W_(StgAsyncIOResult_errCode(ares));
    ccall free(ares "ptr");
    return (len, errC);
}

stg_block_async
{
    Sp_adj(-2);
    Sp(0) = stg_block_async_info;
    BLOCK_GENERIC;
}

/* Used by threadDelay implementation; it would be desirable to get rid of
 * this free()'ing void return continuation.
 */
INFO_TABLE_RET ( stg_block_async_void, RET_SMALL, W_ info_ptr, W_ ares )
    return ()
{
    ccall free(ares "ptr");
    return ();
}

stg_block_async_void
{
    Sp_adj(-2);
    Sp(0) = stg_block_async_void_info;
    BLOCK_GENERIC;
}

#endif


/* -----------------------------------------------------------------------------
   STM-specific waiting
   -------------------------------------------------------------------------- */

stg_block_stmwait
{
    // When blocking on an MVar we have to be careful to only release
    // the lock on the MVar at the very last moment (using
    // BLOCK_BUT_FIRST()), since when we release the lock another
    // Capability can wake up the thread, which modifies its stack and
    // other state.  This is not a problem for STM, because STM
    // wakeups are non-destructive; the waker simply calls
    // tryWakeupThread() which sends a message to the owner
    // Capability.  So the moment we release this lock we might start
    // getting wakeup messages, but that's perfectly harmless.
    //
    // Furthermore, we *must* release these locks, just in case an
    // exception is raised in this thread by
    // maybePerformBlockedException() while exiting to the scheduler,
    // which will abort the transaction, which needs to obtain a lock
    // on all the TVars to remove the thread from the queues.
    //
    ccall stmWaitUnlock(MyCapability() "ptr", R3 "ptr");
    BLOCK_GENERIC;
}