| blob | 69c0f954b0bf1a8a053ad1342894164910bcfc63 |
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.
197 ;;; This is emitted directly before either a known-call-local, call-local,
198 ;;; or a multiple-call-local. All it does is allocate stack space for the
199 ;;; callee (who has the same size stack as us).
210 ;;; Allocate a partial frame for passing stack arguments in a full
211 ;;; call. NARGS is the number of arguments passed. We allocate at
212 ;;; least 3 slots, because the XEP noise is going to want to use them
213 ;;; before it can extend the stack.
221 \f
222 ;;; Emit code needed at the return-point from an unknown-values call
223 ;;; for a fixed number of values. Values is the head of the TN-REF
224 ;;; list for the locations that the values are to be received into.
225 ;;; Nvals is the number of values that are to be received (should
226 ;;; equal the length of Values).
227 ;;;
228 ;;; If 0 or 1 values are expected, then we just emit an instruction to
229 ;;; reset the SP (which will only be executed when other than 1 value
230 ;;; is returned.)
231 ;;;
232 ;;; In the general case we have to do three things:
233 ;;; -- Default unsupplied register values. This need only be done
234 ;;; when a single value is returned, since register values are
235 ;;; defaulted by the called in the non-single case.
236 ;;; -- Default unsupplied stack values. This needs to be done whenever
237 ;;; there are stack values.
238 ;;; -- Reset SP. This must be done whenever other than 1 value is
239 ;;; returned, regardless of the number of values desired.
249 register-arg-count)
252 ;; The function never returns, it may happen that the code
253 ;; ends right here leavig the :SINGLE-VALUE-RETURN note
254 ;; dangling. Let's emit a NOP.
265 ;; Default the unsupplied registers.
273 for count from 2 below register-arg-count
281 ;; The number of bytes depends on the relative jump instructions.
282 ;; Best case is 31+(n-3)*14, worst case is 35+(n-3)*18. For
283 ;; NVALS=6 that is 73/89 bytes, and for NVALS=7 that is 87/107
284 ;; bytes which is likely better than using the blt below.
290 ;; Branch off to the MV case.
292 ;; Do the single value case.
293 ;; Default the register args
298 ;; Fake other registers so it looks like we returned with all the
299 ;; registers filled in.
318 ;; There are stack args so the frame of the callee is
319 ;; still there, save RDX in its first slot temporalily.
333 ;; We are setting the first stack argument to NIL.
334 ;; The callee's stack frame is dead, save RDX by
335 ;; pushing it to the stack, it will end up at same
336 ;; place as in the (STOREW RDX-TN RBX-TN -1) case
337 ;; above.
348 ;; Branch off to the MV case.
350 ;; Default the register args, and set up the stack as if we
351 ;; entered the MV return point.
355 ;; Compute a pointer to where to put the [defaulted] stack values.
362 ;; Load RAX with NIL so we can quickly store it, and set up
363 ;; stuff for the loop.
367 ;; Jump into the default loop.
369 ;; The regs are defaulted. We need to copy any stack arguments,
370 ;; and then default the remaining stack arguments.
372 ;; Compute the number of stack arguments, and if it's zero or
373 ;; less, don't copy any stack arguments.
376 ;; Save EDI.
378 ;; Throw away any unwanted args.
383 ;; Save the number of stack values.
385 ;; Compute a pointer to where the stack args go.
389 ;; Save ESI, and compute a pointer to where the args come from.
395 ;; Do the copy.
400 ;; Restore RSI.
402 ;; Now we have to default the remaining args. Find out how many.
405 ;; If none, then just blow out of here.
409 ;; Load RAX with NIL for fast storing.
411 ;; Do the store.
415 ;; Restore EDI, and reset the stack.
421 \f
422 ;;;; unknown values receiving
424 ;;; Emit code needed at the return point for an unknown-values call
425 ;;; for an arbitrary number of values.
426 ;;;
427 ;;; We do the single and non-single cases with no shared code: there
428 ;;; doesn't seem to be any potential overlap, and receiving a single
429 ;;; value is more important efficiency-wise.
430 ;;;
431 ;;; When there is a single value, we just push it on the stack,
432 ;;; returning the old SP and 1.
433 ;;;
434 ;;; When there is a variable number of values, we move all of the
435 ;;; argument registers onto the stack, and return ARGS and NARGS.
436 ;;;
437 ;;; ARGS and NARGS are TNs wired to the named locations. We must
438 ;;; explicitly allocate these TNs, since their lifetimes overlap with
439 ;;; the results start and count. (Also, it's nice to be able to target
440 ;;; them.)
457 ;; The stack frame is burnt and RETurned from if there are no
458 ;; stack values. In this case quickly reallocate sufficient space.
460 register-arg-count)
467 ;; FIXME: This can't be efficient, but LEA (my first choice)
468 ;; doesn't do subtraction.
473 ;; dtc: this writes the registers onto the stack even if they are
474 ;; not needed, only the number specified in rcx are used and have
475 ;; stack allocated to them. No harm is done.
478 for i downfrom -1
487 ;;; VOP that can be inherited by unknown values receivers. The main thing this
488 ;;; handles is allocation of the result temporaries.
492 values-start)
495 nvals)
498 \f
499 ;;;; local call with unknown values convention return
502 #+nil
506 #+nil
518 ;;; The local call convention doesn't fit that well with x86-style
519 ;;; calls. Emit a header for local calls to pop the return address
520 ;;; in the right place.
531 ;;; Non-TR local call for a fixed number of values passed according to
532 ;;; the unknown values convention.
533 ;;;
534 ;;; FP is the frame pointer in install before doing the call.
535 ;;;
536 ;;; NFP would be the number-stack frame pointer if we had a separate
537 ;;; number stack.
538 ;;;
539 ;;; Args are the argument passing locations, which are specified only
540 ;;; to terminate their lifetimes in the caller.
541 ;;;
542 ;;; VALUES are the return value locations (wired to the standard
543 ;;; passing locations). NVALS is the number of values received.
544 ;;;
545 ;;; Save is the save info, which we can ignore since saving has been
546 ;;; done.
547 ;;;
548 ;;; TARGET is a continuation pointing to the start of the called
549 ;;; function.
567 ;;; Non-TR local call for a variable number of return values passed according
568 ;;; to the unknown values convention. The results are the start of the values
569 ;;; glob and the number of values received.
586 \f
587 ;;;; local call with known values return
589 ;;; Non-TR local call with known return locations. Known-value return
590 ;;; works just like argument passing in local call.
591 ;;;
592 ;;; Note: we can't use normal load-tn allocation for the fixed args,
593 ;;; since all registers may be tied up by the more operand. Instead,
594 ;;; we use MAYBE-LOAD-STACK-TN.
610 \f
611 ;;; From Douglas Crosher
612 ;;; Return from known values call. We receive the return locations as
613 ;;; arguments to terminate their lifetimes in the returning function. We
614 ;;; restore FP and CSP and jump to the Return-PC.
625 ;; Zot all of the stack except for the old-fp and return-pc.
629 \f
630 ;;;; full call
631 ;;;
632 ;;; There is something of a cross-product effect with full calls.
633 ;;; Different versions are used depending on whether we know the
634 ;;; number of arguments or the name of the called function, and
635 ;;; whether we want fixed values, unknown values, or a tail call.
636 ;;;
637 ;;; In full call, the arguments are passed creating a partial frame on
638 ;;; the stack top and storing stack arguments into that frame. On
639 ;;; entry to the callee, this partial frame is pointed to by FP.
641 ;;; This macro helps in the definition of full call VOPs by avoiding
642 ;;; code replication in defining the cross-product VOPs.
643 ;;;
644 ;;; NAME is the name of the VOP to define.
645 ;;;
646 ;;; NAMED is true if the first argument is an fdefinition object whose
647 ;;; definition is to be called.
648 ;;;
649 ;;; RETURN is either :FIXED, :UNKNOWN or :TAIL:
650 ;;; -- If :FIXED, then the call is for a fixed number of values, returned in
651 ;;; the standard passing locations (passed as result operands).
652 ;;; -- If :UNKNOWN, then the result values are pushed on the stack, and the
653 ;;; result values are specified by the Start and Count as in the
654 ;;; unknown-values continuation representation.
655 ;;; -- If :TAIL, then do a tail-recursive call. No values are returned.
656 ;;; The Old-Fp and Return-PC are passed as the second and third arguments.
657 ;;;
658 ;;; In non-tail calls, the pointer to the stack arguments is passed as
659 ;;; the last fixed argument. If Variable is false, then the passing
660 ;;; locations are passed as a more arg. Variable is true if there are
661 ;;; a variable number of arguments passed on the stack. Variable
662 ;;; cannot be specified with :TAIL return. TR variable argument call
663 ;;; is implemented separately.
664 ;;;
665 ;;; In tail call with fixed arguments, the passing locations are
666 ;;; passed as a more arg, but there is no new-FP, since the arguments
667 ;;; have been set up in the current frame.
673 (:args
695 (:info
699 step-instrumenting)
701 (:ignore
705 ;; We pass either the fdefn object (for named call) or
706 ;; the actual function object (for unnamed call) in
707 ;; RAX. With named call, closure-tramp will replace it
708 ;; with the real function and invoke the real function
709 ;; for closures. Non-closures do not need this value,
710 ;; so don't care what shows up in it.
711 (:temporary
712 (:sc descriptor-reg
713 :offset rax-offset
716 rax)
718 ;; We pass the number of arguments in RCX.
721 ;; With variable call, we have to load the
722 ;; register-args out of the (new) stack frame before
723 ;; doing the call. Therefore, we have to tell the
724 ;; lifetime stuff that we need to use them.
738 old-fp-tmp)))
745 15
747 ;; This has to be done before the frame pointer is
748 ;; changed! RAX stores the 'lexical environment' needed
749 ;; for closures.
754 ;; For variable call, compute the number of
755 ;; arguments and move some of the arguments to
756 ;; registers.
758 ;; Compute the number of arguments.
764 ;; Move the necessary args to registers,
765 ;; this moves them all even if they are
766 ;; not all needed.
769 for index downfrom -1
777 ;; return to so might as well use that clue.
778 ;; This seems really important to the
779 ;; implementation of things like
780 ;; (without-interrupts ...)
781 ;;
782 ;; dtc; Could be doing a tail call from a
783 ;; known-local-call etc in which the old-fp
784 ;; or ret-pc are in regs or in non-standard
785 ;; places. If the passing location were
786 ;; wired to the stack in standard locations
787 ;; then these moves will be un-necessary;
788 ;; this is probably best for the x86.
793 ;; FIXME: FORMAT T for stale
794 ;; diagnostic output (several of
795 ;; them around here), ick
799 rbp-tn
804 rbp-tn
807 ;; For tail call, we have to push the
808 ;; return-pc so that it looks like we CALLed
809 ;; despite the fact that we are going to JMP.
811 ))
813 ;; For non-tail call, we have to save our
814 ;; frame pointer and install the new frame
815 ;; pointer. We can't load stack tns after this
816 ;; point.
818 ;; here which doesn't leave room for the
819 ;; ofp/ret stuff.
821 ;; The variable args are on the stack and
822 ;; become the frame, but there may be <3
823 ;; args and 3 stack slots are assumed
824 ;; allocate on the call. So need to ensure
825 ;; there are at least 3 slots. This hack
826 ;; just adds 3 more.
830 ;; Bias the new-fp for use as an fp
834 ;; Save the fp
839 )))
845 DONE
853 n-word-bytes)
854 other-pointer-lowtag)
856 fun-pointer-lowtag))))
858 (:fixed
860 (:unknown
863 node)))
876 ;;; This is defined separately, since it needs special code that BLT's
877 ;;; the arguments down. All the real work is done in the assembly
878 ;;; routine. We just set things up so that it can find what it needs.
890 ;; Move these into the passing locations if they are not already there.
893 ;; And jump to the assembly routine.
895 \f
896 ;;;; unknown values return
898 ;;; Return a single-value using the Unknown-Values convention.
899 ;;;
900 ;;; pfw--get wired-tn conflicts sometimes if register sc specd for args
901 ;;; having problems targeting args to regs -- using temps instead.
902 ;;;
903 ;;; First off, modifying the return-pc defeats the branch-prediction
904 ;;; optimizations on modern CPUs quite handily. Second, we can do all
905 ;;; this without needing a temp register. Fixed the latter, at least.
906 ;;; -- AB 2006/Feb/04
914 ;; Drop stack above old-fp
916 ;; Clear the multiple-value return flag
918 ;; Restore the old frame pointer
920 ;; And return.
923 ;;; Do unknown-values return of a fixed (other than 1) number of
924 ;;; values. The VALUES are required to be set up in the standard
925 ;;; passing locations. NVALS is the number of values returned.
926 ;;;
927 ;;; Basically, we just load RCX with the number of values returned and
928 ;;; RBX with a pointer to the values, set RSP to point to the end of
929 ;;; the values, and jump directly to return-pc.
936 ;; In the case of other than one value, we need these registers to
937 ;; tell the caller where they are and how many there are.
940 ;; We need to stretch the lifetime of return-pc past the argument
941 ;; registers so that we can default the argument registers without
942 ;; trashing return-pc.
953 ;; This is handled in RETURN-SINGLE.
955 ;; Establish the values pointer and values count.
961 ;; Pre-default any argument register that need it.
968 ;; Set the multiple value return flag.
970 ;; And away we go. Except that return-pc is still on the
971 ;; stack and we've changed the stack pointer. So we have to
972 ;; tell it to index off of RBX instead of RBP.
978 ;; Some values are on the stack after RETURN-PC and OLD-FP,
979 ;; can't return normally and some slots of the frame will
980 ;; be used as temporaries by the receiver.
981 ;;
982 ;; Clear as much of the stack as possible, but not past the
983 ;; old frame address.
994 ;;; Do unknown-values return of an arbitrary number of values (passed
995 ;;; on the stack.) We check for the common case of a single return
996 ;;; value, and do that inline using the normal single value return
997 ;;; convention. Otherwise, we branch off to code that calls an
998 ;;; assembly-routine.
999 ;;;
1000 ;;; The assembly routine takes the following args:
1001 ;;; RCX -- number of values to find there.
1002 ;;; RSI -- pointer to where to find the values.
1018 ;; Check for the single case.
1022 ;; Return with one value.
1024 ;; Clear the stack until ocfp.
1026 ;; clear the multiple-value return flag
1028 ;; Out of here.
1031 ;; Nope, not the single case. Jump to the assembly routine.
1036 \f
1037 ;;;; XEP hackery
1039 ;;; Get the lexical environment from its passing location.
1045 ;; Get result.
1048 ;;; Copy a &MORE arg from the argument area to the end of the current
1049 ;;; frame. FIXED is the number of non-&MORE arguments.
1056 ;; Avoid the copy if there are no more args.
1061 ;; verify-arg-count will do a CMP
1067 ;; Create a negated copy of the number of arguments to allow us to
1068 ;; use EA calculations in order to do scaled subtraction.
1072 ;; Allocate the space on the stack.
1073 ;; stack = rbp + sp->fp-offset - (max 3 frame-size) - (nargs - fixed)
1074 ;; if we'd move SP backward, swap the meaning of rsp and source;
1075 ;; otherwise, we'd be accessing values below SP, and that's no good
1076 ;; if a signal interrupts this code sequence. In that case, store
1077 ;; the final value in rsp after the stack-stack memmove loop.
1079 rsp-tn
1080 source)
1087 ;; Now: nargs>=1 && nargs>fixed
1090 ;; the code above only moves the final value of rsp in
1091 ;; rsp directly if that condition is satisfied. Currently,
1092 ;; r-a-c is 3, so the aver is OK. If the calling convention
1093 ;; ever changes, the logic above with LEA will have to be
1094 ;; adjusted.
1096 ;; We must stop when we run out of stack args, not when we
1097 ;; run out of more args.
1098 ;; Number to copy = nargs-3
1099 ;; Save the original count of args.
1102 ;; Everything of interest in registers.
1105 ;; Number to copy = nargs-fixed
1109 ;; Initialize R8 to be the end of args.
1110 ;; Swap with SP if necessary to mirror the previous condition
1112 source
1113 rsp-tn)
1118 ;; src: rbp + temp + sp->fp
1119 ;; dst: rbp + temp + sp->fp + (fixed - (max 3 [stack-size]))
1125 ;; dst is lower than src, copy forward
1127 ;; We used to use REP MOVS here, but on modern x86 it performs
1128 ;; much worse than an explicit loop for small blocks.
1137 ;; dst is higher than src; copy backward
1144 temp)
1146 ;; done with the stack--stack copy. Reset RSP to its final
1147 ;; value
1149 DO-REGS
1151 ;; Here: nargs>=1 && nargs>fixed
1153 ;; Now we have to deposit any more args that showed up in
1154 ;; registers.
1157 ;; Store it relative to rbp
1171 ;; Don't deposit any more than there are.
1179 JUST-ALLOC-FRAME
1186 DONE))
1230 ;;; Turn more arg (context, count) into a list.
1250 ;; Check to see whether there are no args, and just return NIL if so.
1256 ;; Set up the result.
1258 ;; Jump into the middle of the loop, 'cause that's where we want
1259 ;; to start.
1262 ;; Compute a pointer to the next cons.
1264 ;; Store a pointer to this cons in the CDR of the previous cons.
1267 ;; Grab one value and stash it in the car of this cons.
1271 ;; Go back for more.
1274 ;; NIL out the last cons.
1278 ;;; Return the location and size of the &MORE arg glob created by
1279 ;;; COPY-MORE-ARG. SUPPLIED is the total number of arguments supplied
1280 ;;; (originally passed in RCX). FIXED is the number of non-rest
1281 ;;; arguments.
1282 ;;;
1283 ;;; We must duplicate some of the work done by COPY-MORE-ARG, since at
1284 ;;; that time the environment is in a pretty brain-damaged state,
1285 ;;; preventing this info from being returned as values. What we do is
1286 ;;; compute supplied - fixed, and return a pointer that many words
1287 ;;; below the current stack top.
1300 ;; SP at this point points at the last arg pushed.
1301 ;; Point to the first more-arg, not above it.
1303 :index count
1317 ;; NOTE: copy-more-arg expects this to issue a CMP for min > 1
1339 ;; Signal an error about an untagged number.
1340 ;; These are pretty much boilerplate and could be generic except:
1341 ;; - the names of the SCs could differ between backends (or maybe not?)
1342 ;; - in the "/c" case, the older backends don't eval the errcode
1343 ;; And the 6 vops above ought to be generic too...
1344 ;; FIXME: there are still some occurrences of
1345 ;; note: doing signed word to integer coercion
1346 ;; in regard to SB-C::%TYPE-CHECK-ERROR. Figure out why.
1351 ;; Types are trees of symbols, so 'any-reg' is not
1352 ;; really possible.
1358 ;; cost is a smidgen less than type-check-error
1359 ;; otherwise this does not get selected.
1373 ;;; Single-stepping
1376 ;; We use different ways of representing whether stepping is on on
1377 ;; +SB-THREAD / -SB-THREAD: on +SB-THREAD, we use a slot in the
1378 ;; thread structure. On -SB-THREAD we use the value of a static
1379 ;; symbol. Things are done this way, since reading a thread-local
1380 ;; slot from a symbol would require an extra register on +SB-THREAD,
1381 ;; and reading a slot from a thread structure would require an extra
1382 ;; register on -SB-THREAD. While this isn't critical for x86-64,
1383 ;; it's more serious for x86.
1384 #!+sb-thread
1386 :base thread-base-tn
1389 #!-sb-thread
1404 DONE