| blob | 1ce12b1b6738aeff65608cb9db3125a8cc703e1f |
1 ;;;; function call for the x86 VM
3 ;;;; This software is part of the SBCL system. See the README file for
4 ;;;; more information.
5 ;;;;
6 ;;;; This software is derived from the CMU CL system, which was
7 ;;;; written at Carnegie Mellon University and released into the
8 ;;;; public domain. The software is in the public domain and is
9 ;;;; provided with absolutely no warranty. See the COPYING and CREDITS
10 ;;;; files for more information.
17 ;;; Make a passing location TN for a local call return PC.
18 ;;;
19 ;;; Always wire the return PC location to the stack in its standard
20 ;;; location.
24 sap-stack-sc-number return-pc-save-offset))
29 ;;; This is similar to MAKE-RETURN-PC-PASSING-LOCATION, but makes a
30 ;;; location to pass OLD-FP in.
31 ;;;
32 ;;; This is wired in both the standard and the local-call conventions,
33 ;;; because we want to be able to assume it's always there. Besides,
34 ;;; the x86 doesn't have enough registers to really make it profitable
35 ;;; to pass it in a register.
39 ocfp-save-offset))
44 ;;; Make the TNs used to hold OLD-FP and RETURN-PC within the current
45 ;;; function. We treat these specially so that the debugger can find
46 ;;; them at a known location.
47 ;;;
48 ;;; Without using a save-tn - which does not make much sense if it is
49 ;;; wired to the stack?
52 control-stack-sc-number
53 ocfp-save-offset)
54 physenv))
58 sap-stack-sc-number return-pc-save-offset)
59 physenv))
61 ;;; Make a TN for the standard argument count passing location. We only
62 ;;; need to make the standard location, since a count is never passed when we
63 ;;; are using non-standard conventions.
66 \f
67 ;;;; frame hackery
69 ;;; This is used for setting up the Old-FP in local call.
75 ;;; We don't have a separate NFP, so we don't need to do anything here.
80 nil))
82 ;;; Accessing a slot from an earlier stack frame is definite hackery.
106 :base frame-pointer
139 collect
140 `(,name
146 fp value res))
149 fp new-val value)))))))))
151 ancestor-frame-ref/double-float
154 ancestor-frame-ref/single-float
157 ancestor-frame-ref/complex-double-float
160 ancestor-frame-ref/complex-single-float
163 ancestor-frame-ref/signed-byte-64
166 ancestor-frame-ref/unsigned-byte-64
169 ancestor-frame-ref/unsigned-byte-64
172 ancestor-frame-ref/system-area-pointer
180 ;; Skip space for the function header.
185 ;; The start of the actual code.
186 ;; Save the return-pc.
200 ;;; This is emitted directly before either a known-call-local, call-local,
201 ;;; or a multiple-call-local. All it does is allocate stack space for the
202 ;;; callee (who has the same size stack as us).
213 ;;; Allocate a partial frame for passing stack arguments in a full
214 ;;; call. NARGS is the number of arguments passed. We allocate at
215 ;;; least 3 slots, because the XEP noise is going to want to use them
216 ;;; before it can extend the stack.
224 n-word-bytes))))
225 \f
226 ;;; Emit code needed at the return-point from an unknown-values call
227 ;;; for a fixed number of values. Values is the head of the TN-REF
228 ;;; list for the locations that the values are to be received into.
229 ;;; Nvals is the number of values that are to be received (should
230 ;;; equal the length of Values).
231 ;;;
232 ;;; If 0 or 1 values are expected, then we just emit an instruction to
233 ;;; reset the SP (which will only be executed when other than 1 value
234 ;;; is returned.)
235 ;;;
236 ;;; In the general case we have to do three things:
237 ;;; -- Default unsupplied register values. This need only be done
238 ;;; when a single value is returned, since register values are
239 ;;; defaulted by the called in the non-single case.
240 ;;; -- Default unsupplied stack values. This needs to be done whenever
241 ;;; there are stack values.
242 ;;; -- Reset SP. This must be done whenever other than 1 value is
243 ;;; returned, regardless of the number of values desired.
253 register-arg-count)
256 ;; The function never returns, it may happen that the code
257 ;; ends right here leavig the :SINGLE-VALUE-RETURN note
258 ;; dangling. Let's emit a NOP.
269 ;; Default the unsupplied registers.
277 for count from 2 below register-arg-count
285 ;; The number of bytes depends on the relative jump instructions.
286 ;; Best case is 31+(n-3)*14, worst case is 35+(n-3)*18. For
287 ;; NVALS=6 that is 73/89 bytes, and for NVALS=7 that is 87/107
288 ;; bytes which is likely better than using the blt below.
294 ;; Branch off to the MV case.
296 ;; Do the single value case.
297 ;; Default the register args
302 ;; Fake other registers so it looks like we returned with all the
303 ;; registers filled in.
322 ;; There are stack args so the frame of the callee is
323 ;; still there, save RDX in its first slot temporalily.
337 ;; We are setting the first stack argument to NIL.
338 ;; The callee's stack frame is dead, save RDX by
339 ;; pushing it to the stack, it will end up at same
340 ;; place as in the (STOREW RDX-TN RBX-TN -1) case
341 ;; above.
353 ;; Branch off to the MV case.
355 ;; Default the register args, and set up the stack as if we
356 ;; entered the MV return point.
360 ;; Compute a pointer to where to put the [defaulted] stack values.
367 ;; Load RAX with NIL so we can quickly store it, and set up
368 ;; stuff for the loop.
371 ;; Jump into the default loop.
373 ;; The regs are defaulted. We need to copy any stack arguments,
374 ;; and then default the remaining stack arguments.
376 ;; Compute the number of stack arguments, and if it's zero or
377 ;; less, don't copy any stack arguments.
380 ;; Save EDI.
382 ;; Throw away any unwanted args.
387 ;; Save the number of stack values.
389 ;; Compute a pointer to where the stack args go.
393 ;; Save ESI, and compute a pointer to where the args come from.
399 ;; Do the copy.
403 ;; Restore RSI.
405 ;; Now we have to default the remaining args. Find out how many.
408 ;; If none, then just blow out of here.
412 ;; Load RAX with NIL for fast storing.
414 ;; Do the store.
418 ;; Restore EDI, and reset the stack.
424 \f
425 ;;;; unknown values receiving
427 ;;; Emit code needed at the return point for an unknown-values call
428 ;;; for an arbitrary number of values.
429 ;;;
430 ;;; We do the single and non-single cases with no shared code: there
431 ;;; doesn't seem to be any potential overlap, and receiving a single
432 ;;; value is more important efficiency-wise.
433 ;;;
434 ;;; When there is a single value, we just push it on the stack,
435 ;;; returning the old SP and 1.
436 ;;;
437 ;;; When there is a variable number of values, we move all of the
438 ;;; argument registers onto the stack, and return ARGS and NARGS.
439 ;;;
440 ;;; ARGS and NARGS are TNs wired to the named locations. We must
441 ;;; explicitly allocate these TNs, since their lifetimes overlap with
442 ;;; the results start and count. (Also, it's nice to be able to target
443 ;;; them.)
460 ;; The stack frame is burnt and RETurned from if there are no
461 ;; stack values. In this case quickly reallocate sufficient space.
463 register-arg-count)
470 ;; FIXME: This can't be efficient, but LEA (my first choice)
471 ;; doesn't do subtraction.
476 ;; dtc: this writes the registers onto the stack even if they are
477 ;; not needed, only the number specified in rcx are used and have
478 ;; stack allocated to them. No harm is done.
481 for i downfrom -1
490 ;;; VOP that can be inherited by unknown values receivers. The main thing this
491 ;;; handles is allocation of the result temporaries.
495 values-start)
498 nvals)
501 \f
502 ;;;; local call with unknown values convention return
505 #+nil
509 #+nil
521 ;;; The local call convention doesn't fit that well with x86-style
522 ;;; calls. Emit a header for local calls to pop the return address
523 ;;; in the right place.
534 ;;; Non-TR local call for a fixed number of values passed according to
535 ;;; the unknown values convention.
536 ;;;
537 ;;; FP is the frame pointer in install before doing the call.
538 ;;;
539 ;;; NFP would be the number-stack frame pointer if we had a separate
540 ;;; number stack.
541 ;;;
542 ;;; Args are the argument passing locations, which are specified only
543 ;;; to terminate their lifetimes in the caller.
544 ;;;
545 ;;; VALUES are the return value locations (wired to the standard
546 ;;; passing locations). NVALS is the number of values received.
547 ;;;
548 ;;; Save is the save info, which we can ignore since saving has been
549 ;;; done.
550 ;;;
551 ;;; TARGET is a continuation pointing to the start of the called
552 ;;; function.
570 ;;; Non-TR local call for a variable number of return values passed according
571 ;;; to the unknown values convention. The results are the start of the values
572 ;;; glob and the number of values received.
589 \f
590 ;;;; local call with known values return
592 ;;; Non-TR local call with known return locations. Known-value return
593 ;;; works just like argument passing in local call.
594 ;;;
595 ;;; Note: we can't use normal load-tn allocation for the fixed args,
596 ;;; since all registers may be tied up by the more operand. Instead,
597 ;;; we use MAYBE-LOAD-STACK-TN.
613 \f
614 ;;; From Douglas Crosher
615 ;;; Return from known values call. We receive the return locations as
616 ;;; arguments to terminate their lifetimes in the returning function. We
617 ;;; restore FP and CSP and jump to the Return-PC.
628 ;; Zot all of the stack except for the old-fp and return-pc.
632 \f
633 ;;;; full call
634 ;;;
635 ;;; There is something of a cross-product effect with full calls.
636 ;;; Different versions are used depending on whether we know the
637 ;;; number of arguments or the name of the called function, and
638 ;;; whether we want fixed values, unknown values, or a tail call.
639 ;;;
640 ;;; In full call, the arguments are passed creating a partial frame on
641 ;;; the stack top and storing stack arguments into that frame. On
642 ;;; entry to the callee, this partial frame is pointed to by FP.
644 ;;; This macro helps in the definition of full call VOPs by avoiding
645 ;;; code replication in defining the cross-product VOPs.
646 ;;;
647 ;;; NAME is the name of the VOP to define.
648 ;;;
649 ;;; NAMED is true if the first argument is an fdefinition object whose
650 ;;; definition is to be called.
651 ;;;
652 ;;; RETURN is either :FIXED, :UNKNOWN or :TAIL:
653 ;;; -- If :FIXED, then the call is for a fixed number of values, returned in
654 ;;; the standard passing locations (passed as result operands).
655 ;;; -- If :UNKNOWN, then the result values are pushed on the stack, and the
656 ;;; result values are specified by the Start and Count as in the
657 ;;; unknown-values continuation representation.
658 ;;; -- If :TAIL, then do a tail-recursive call. No values are returned.
659 ;;; The Old-Fp and Return-PC are passed as the second and third arguments.
660 ;;;
661 ;;; In non-tail calls, the pointer to the stack arguments is passed as
662 ;;; the last fixed argument. If Variable is false, then the passing
663 ;;; locations are passed as a more arg. Variable is true if there are
664 ;;; a variable number of arguments passed on the stack. Variable
665 ;;; cannot be specified with :TAIL return. TR variable argument call
666 ;;; is implemented separately.
667 ;;;
668 ;;; In tail call with fixed arguments, the passing locations are
669 ;;; passed as a more arg, but there is no new-FP, since the arguments
670 ;;; have been set up in the current frame.
676 (:args
680 ;; If immobile-space is in use, then named call does not require
681 ;; a register unless the caller is NOT in immobile space,
682 ;; in which case the register is needed because there is no
683 ;; absolute addressing mode for jmp/call.
704 (:info
707 ;; Intuitively you might want FUN to be the first codegen arg,
708 ;; but that won't work, because EMIT-ARG-MOVES wants the
709 ;; passing locs in (FIRST (vop-codegen-info vop)).
712 step-instrumenting)
714 (:ignore
718 ;; We pass either the fdefn object (for named call) or
719 ;; the actual function object (for unnamed call) in
720 ;; RAX. With named call, closure-tramp will replace it
721 ;; with the real function and invoke the real function
722 ;; for closures. Non-closures do not need this value,
723 ;; so don't care what shows up in it.
728 ;; We pass the number of arguments in RCX.
731 ;; With variable call, we have to load the
732 ;; register-args out of the (new) stack frame before
733 ;; doing the call. Therefore, we have to tell the
734 ;; lifetime stuff that we need to use them.
748 old-fp-tmp)))
755 15
760 ;; This has to be done before the frame pointer is
761 ;; changed! RAX stores the 'lexical environment' needed
762 ;; for closures.
767 ;; For variable call, compute the number of
768 ;; arguments and move some of the arguments to
769 ;; registers.
771 ;; Compute the number of arguments.
777 ;; Move the necessary args to registers,
778 ;; this moves them all even if they are
779 ;; not all needed.
782 for index downfrom -1
790 ;; return to so might as well use that clue.
791 ;; This seems really important to the
792 ;; implementation of things like
793 ;; (without-interrupts ...)
794 ;;
795 ;; dtc; Could be doing a tail call from a
796 ;; known-local-call etc in which the old-fp
797 ;; or ret-pc are in regs or in non-standard
798 ;; places. If the passing location were
799 ;; wired to the stack in standard locations
800 ;; then these moves will be un-necessary;
801 ;; this is probably best for the x86.
806 ;; FIXME: FORMAT T for stale
807 ;; diagnostic output (several of
808 ;; them around here), ick
812 rbp-tn
817 rbp-tn
820 ;; For tail call, we have to push the
821 ;; return-pc so that it looks like we CALLed
822 ;; despite the fact that we are going to JMP.
824 ))
826 ;; For non-tail call, we have to save our
827 ;; frame pointer and install the new frame
828 ;; pointer. We can't load stack tns after this
829 ;; point.
831 ;; here which doesn't leave room for the
832 ;; ofp/ret stuff.
834 ;; The variable args are on the stack and
835 ;; become the frame, but there may be <3
836 ;; args and 3 stack slots are assumed
837 ;; allocate on the call. So need to ensure
838 ;; there are at least 3 slots. This hack
839 ;; just adds 3 more.
843 ;; Bias the new-fp for use as an fp
847 ;; Save the fp
852 )))
858 DONE
868 ;; RAX-TN was not declared as a temp var,
869 ;; however it's sole purpose at this point is
870 ;; for function call, so even if it was used
871 ;; to compute a stack argument, it's free now.
872 ;; If the call hits the undefined fun trap,
873 ;; RAX will get loaded regardless.
875 rax-tn))))
881 n-word-bytes)
882 other-pointer-lowtag)
884 fun-pointer-lowtag)))))
887 (:fixed
889 (:unknown
892 node)))
905 ;;; This is defined separately, since it needs special code that BLT's
906 ;;; the arguments down. All the real work is done in the assembly
907 ;;; routine. We just set things up so that it can find what it needs.
919 ;; Move these into the passing locations if they are not already there.
922 ;; And jump to the assembly routine.
924 \f
925 ;;;; unknown values return
927 ;;; Return a single-value using the Unknown-Values convention.
928 ;;;
929 ;;; pfw--get wired-tn conflicts sometimes if register sc specd for args
930 ;;; having problems targeting args to regs -- using temps instead.
931 ;;;
932 ;;; First off, modifying the return-pc defeats the branch-prediction
933 ;;; optimizations on modern CPUs quite handily. Second, we can do all
934 ;;; this without needing a temp register. Fixed the latter, at least.
935 ;;; -- AB 2006/Feb/04
943 ;; Drop stack above old-fp
945 ;; Clear the multiple-value return flag
947 ;; Restore the old frame pointer
949 ;; And return.
952 ;;; Do unknown-values return of a fixed (other than 1) number of
953 ;;; values. The VALUES are required to be set up in the standard
954 ;;; passing locations. NVALS is the number of values returned.
955 ;;;
956 ;;; Basically, we just load RCX with the number of values returned and
957 ;;; RBX with a pointer to the values, set RSP to point to the end of
958 ;;; the values, and jump directly to return-pc.
965 ;; In the case of other than one value, we need these registers to
966 ;; tell the caller where they are and how many there are.
969 ;; We need to stretch the lifetime of return-pc past the argument
970 ;; registers so that we can default the argument registers without
971 ;; trashing return-pc.
982 ;; This is handled in RETURN-SINGLE.
984 ;; Establish the values pointer and values count.
990 ;; Pre-default any argument register that need it.
997 ;; Set the multiple value return flag.
999 ;; And away we go. Except that return-pc is still on the
1000 ;; stack and we've changed the stack pointer. So we have to
1001 ;; tell it to index off of RBX instead of RBP.
1007 ;; Some values are on the stack after RETURN-PC and OLD-FP,
1008 ;; can't return normally and some slots of the frame will
1009 ;; be used as temporaries by the receiver.
1010 ;;
1011 ;; Clear as much of the stack as possible, but not past the
1012 ;; old frame address.
1023 ;;; Do unknown-values return of an arbitrary number of values (passed
1024 ;;; on the stack.) We check for the common case of a single return
1025 ;;; value, and do that inline using the normal single value return
1026 ;;; convention. Otherwise, we branch off to code that calls an
1027 ;;; assembly-routine.
1028 ;;;
1029 ;;; The assembly routine takes the following args:
1030 ;;; RCX -- number of values to find there.
1031 ;;; RSI -- pointer to where to find the values.
1047 ;; Check for the single case.
1051 ;; Return with one value.
1053 ;; Clear the stack until ocfp.
1055 ;; clear the multiple-value return flag
1057 ;; Out of here.
1060 ;; Nope, not the single case. Jump to the assembly routine.
1065 \f
1066 ;;;; XEP hackery
1068 ;;; Get the lexical environment from its passing location.
1074 ;; Get result.
1077 ;;; Copy a &MORE arg from the argument area to the end of the current
1078 ;;; frame. FIXED is the number of non-&MORE arguments.
1085 ;; Avoid the copy if there are no more args.
1090 ;; verify-arg-count will do a CMP
1096 ;; Create a negated copy of the number of arguments to allow us to
1097 ;; use EA calculations in order to do scaled subtraction.
1101 ;; Allocate the space on the stack.
1102 ;; stack = rbp + sp->fp-offset - frame-size - (nargs - fixed)
1103 ;; if we'd move SP backward, swap the meaning of rsp and source;
1104 ;; otherwise, we'd be accessing values below SP, and that's no good
1105 ;; if a signal interrupts this code sequence. In that case, store
1106 ;; the final value in rsp after the stack-stack memmove loop.
1108 rsp-tn
1109 source)
1116 ;; Now: nargs>=1 && nargs>fixed
1119 ;; the code above only moves the final value of rsp in
1120 ;; rsp directly if that condition is satisfied. Currently,
1121 ;; r-a-c is 3, so the aver is OK. If the calling convention
1122 ;; ever changes, the logic above with LEA will have to be
1123 ;; adjusted.
1125 ;; We must stop when we run out of stack args, not when we
1126 ;; run out of more args.
1127 ;; Number to copy = nargs-3
1128 ;; Save the original count of args.
1131 ;; Everything of interest in registers.
1134 ;; Number to copy = nargs-fixed
1138 ;; Initialize R8 to be the end of args.
1139 ;; Swap with SP if necessary to mirror the previous condition
1141 source
1142 rsp-tn)
1147 ;; src: rbp + temp + sp->fp
1148 ;; dst: rbp + temp + sp->fp + (fixed - [stack-size])
1154 ;; dst is lower than src, copy forward
1156 ;; We used to use REP MOVS here, but on modern x86 it performs
1157 ;; much worse than an explicit loop for small blocks.
1166 ;; dst is higher than src; copy backward
1173 temp)
1175 ;; done with the stack--stack copy. Reset RSP to its final
1176 ;; value
1178 DO-REGS
1180 ;; Here: nargs>=1 && nargs>fixed
1182 ;; Now we have to deposit any more args that showed up in
1183 ;; registers.
1186 ;; Store it relative to rbp
1199 ;; Don't deposit any more than there are.
1207 JUST-ALLOC-FRAME
1209 (* n-word-bytes
1213 DONE))
1257 ;;; Turn more arg (context, count) into a list.
1277 ;; Check to see whether there are no args, and just return NIL if so.
1283 ;; Set up the result.
1285 ;; Jump into the middle of the loop, 'cause that's where we want
1286 ;; to start.
1289 ;; Compute a pointer to the next cons.
1291 ;; Store a pointer to this cons in the CDR of the previous cons.
1294 ;; Grab one value and stash it in the car of this cons.
1298 ;; Go back for more.
1301 ;; NIL out the last cons.
1305 ;;; Return the location and size of the &MORE arg glob created by
1306 ;;; COPY-MORE-ARG. SUPPLIED is the total number of arguments supplied
1307 ;;; (originally passed in RCX). FIXED is the number of non-rest
1308 ;;; arguments.
1309 ;;;
1310 ;;; We must duplicate some of the work done by COPY-MORE-ARG, since at
1311 ;;; that time the environment is in a pretty brain-damaged state,
1312 ;;; preventing this info from being returned as values. What we do is
1313 ;;; compute supplied - fixed, and return a pointer that many words
1314 ;;; below the current stack top.
1327 ;; SP at this point points at the last arg pushed.
1328 ;; Point to the first more-arg, not above it.
1330 :index count
1344 ;; NOTE: copy-more-arg expects this to issue a CMP for min > 1
1366 ;; Signal an error about an untagged number.
1367 ;; These are pretty much boilerplate and could be generic except:
1368 ;; - the names of the SCs could differ between backends (or maybe not?)
1369 ;; - in the "/c" case, the older backends don't eval the errcode
1370 ;; And the 6 vops above ought to be generic too...
1371 ;; FIXME: there are still some occurrences of
1372 ;; note: doing signed word to integer coercion
1373 ;; in regard to SB-C::%TYPE-CHECK-ERROR. Figure out why.
1378 ;; Types are trees of symbols, so 'any-reg' is not
1379 ;; really possible.
1385 ;; cost is a smidgen less than type-check-error
1386 ;; otherwise this does not get selected.
1400 ;;; Single-stepping
1403 ;; We use different ways of representing whether stepping is on on
1404 ;; +SB-THREAD / -SB-THREAD: on +SB-THREAD, we use a slot in the
1405 ;; thread structure. On -SB-THREAD we use the value of a static
1406 ;; symbol. Things are done this way, since reading a thread-local
1407 ;; slot from a symbol would require an extra register on +SB-THREAD,
1408 ;; and reading a slot from a thread structure would require an extra
1409 ;; register on -SB-THREAD. While this isn't critical for x86-64,
1410 ;; it's more serious for x86.
1411 #!+sb-thread
1413 :base thread-base-tn
1416 #!-sb-thread
1431 DONE