| blob | db1ceee74b78c626c578c30c9132402130f60812 |
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.
13 \f
14 ;;;; interfaces to IR2 conversion
16 ;;; Return a wired TN describing the N'th full call argument passing
17 ;;; location.
35 ;;; Make a passing location TN for a local call return PC.
36 ;;;
37 ;;; Always wire the return PC location to the stack in its standard
38 ;;; location.
42 sap-stack-sc-number return-pc-save-offset))
47 ;;; This is similar to MAKE-RETURN-PC-PASSING-LOCATION, but makes a
48 ;;; location to pass OLD-FP in.
49 ;;;
50 ;;; This is wired in both the standard and the local-call conventions,
51 ;;; because we want to be able to assume it's always there. Besides,
52 ;;; the x86 doesn't have enough registers to really make it profitable
53 ;;; to pass it in a register.
57 ocfp-save-offset))
62 ;;; Make the TNs used to hold OLD-FP and RETURN-PC within the current
63 ;;; function. We treat these specially so that the debugger can find
64 ;;; them at a known location.
65 ;;;
66 ;;; Without using a save-tn - which does not make much sense if it is
67 ;;; wired to the stack?
70 control-stack-sc-number
71 ocfp-save-offset)
72 physenv))
76 sap-stack-sc-number return-pc-save-offset)
77 physenv))
79 ;;; Make a TN for the standard argument count passing location. We only
80 ;;; need to make the standard location, since a count is never passed when we
81 ;;; are using non-standard conventions.
85 ;;; Make a TN to hold the number-stack frame pointer. This is allocated
86 ;;; once per component, and is component-live.
96 ;;; Return a list of TNs that can be used to represent an unknown-values
97 ;;; continuation within a function.
102 ;;; This function is called by the ENTRY-ANALYZE phase, allowing
103 ;;; VM-dependent initialization of the IR2-COMPONENT structure. We
104 ;;; push placeholder entries in the CONSTANTS to leave room for
105 ;;; additional noise in the code object header.
112 \f
113 ;;;; frame hackery
115 ;;; This is used for setting up the Old-FP in local call.
121 ;;; We don't have a separate NFP, so we don't need to do anything here.
126 nil))
128 ;;; Accessing a slot from an earlier stack frame is definite hackery.
152 :base frame-pointer
185 collect
186 `(,name
192 fp value res))
195 fp new-val value)))))))))
197 ancestor-frame-ref/double-float
200 ancestor-frame-ref/single-float
203 ancestor-frame-ref/complex-double-float
206 ancestor-frame-ref/complex-single-float
209 ancestor-frame-ref/signed-byte-64
212 ancestor-frame-ref/unsigned-byte-64
215 ancestor-frame-ref/unsigned-byte-64
218 ancestor-frame-ref/system-area-pointer
226 ;; Skip space for the function header.
231 ;; The start of the actual code.
232 ;; Save the return-pc.
243 ;;; This is emitted directly before either a known-call-local, call-local,
244 ;;; or a multiple-call-local. All it does is allocate stack space for the
245 ;;; callee (who has the same size stack as us).
256 ;;; Allocate a partial frame for passing stack arguments in a full
257 ;;; call. NARGS is the number of arguments passed. We allocate at
258 ;;; least 3 slots, because the XEP noise is going to want to use them
259 ;;; before it can extend the stack.
267 \f
268 ;;; Emit code needed at the return-point from an unknown-values call
269 ;;; for a fixed number of values. Values is the head of the TN-REF
270 ;;; list for the locations that the values are to be received into.
271 ;;; Nvals is the number of values that are to be received (should
272 ;;; equal the length of Values).
273 ;;;
274 ;;; If 0 or 1 values are expected, then we just emit an instruction to
275 ;;; reset the SP (which will only be executed when other than 1 value
276 ;;; is returned.)
277 ;;;
278 ;;; In the general case we have to do three things:
279 ;;; -- Default unsupplied register values. This need only be done
280 ;;; when a single value is returned, since register values are
281 ;;; defaulted by the called in the non-single case.
282 ;;; -- Default unsupplied stack values. This needs to be done whenever
283 ;;; there are stack values.
284 ;;; -- Reset SP. This must be done whenever other than 1 value is
285 ;;; returned, regardless of the number of values desired.
295 register-arg-count)
298 ;; The function never returns, it may happen that the code
299 ;; ends right here leavig the :SINGLE-VALUE-RETURN note
300 ;; dangling. Let's emit a NOP.
311 ;; Default the unsupplied registers.
319 for count from 2 below register-arg-count
327 ;; The number of bytes depends on the relative jump instructions.
328 ;; Best case is 31+(n-3)*14, worst case is 35+(n-3)*18. For
329 ;; NVALS=6 that is 73/89 bytes, and for NVALS=7 that is 87/107
330 ;; bytes which is likely better than using the blt below.
335 ;; Branch off to the MV case.
337 ;; Do the single value case.
338 ;; Default the register args
344 ;; Fake other registers so it looks like we returned with all the
345 ;; registers filled in.
365 ;; There are stack args so the frame of the callee is
366 ;; still there, save RDX in its first slot temporalily.
380 ;; We are setting the first stack argument to NIL.
381 ;; The callee's stack frame is dead, save RDX by
382 ;; pushing it to the stack, it will end up at same
383 ;; place as in the (STOREW RDX-TN RBX-TN -1) case
384 ;; above.
395 ;; Branch off to the MV case.
397 ;; Default the register args, and set up the stack as if we
398 ;; entered the MV return point.
402 ;; Compute a pointer to where to put the [defaulted] stack values.
409 ;; Load RAX with NIL so we can quickly store it, and set up
410 ;; stuff for the loop.
414 ;; Jump into the default loop.
416 ;; The regs are defaulted. We need to copy any stack arguments,
417 ;; and then default the remaining stack arguments.
419 ;; Compute the number of stack arguments, and if it's zero or
420 ;; less, don't copy any stack arguments.
423 ;; Save EDI.
425 ;; Throw away any unwanted args.
430 ;; Save the number of stack values.
432 ;; Compute a pointer to where the stack args go.
436 ;; Save ESI, and compute a pointer to where the args come from.
442 ;; Do the copy.
447 ;; Restore RSI.
449 ;; Now we have to default the remaining args. Find out how many.
452 ;; If none, then just blow out of here.
456 ;; Load RAX with NIL for fast storing.
458 ;; Do the store.
462 ;; Restore EDI, and reset the stack.
468 \f
469 ;;;; unknown values receiving
471 ;;; Emit code needed at the return point for an unknown-values call
472 ;;; for an arbitrary number of values.
473 ;;;
474 ;;; We do the single and non-single cases with no shared code: there
475 ;;; doesn't seem to be any potential overlap, and receiving a single
476 ;;; value is more important efficiency-wise.
477 ;;;
478 ;;; When there is a single value, we just push it on the stack,
479 ;;; returning the old SP and 1.
480 ;;;
481 ;;; When there is a variable number of values, we move all of the
482 ;;; argument registers onto the stack, and return ARGS and NARGS.
483 ;;;
484 ;;; ARGS and NARGS are TNs wired to the named locations. We must
485 ;;; explicitly allocate these TNs, since their lifetimes overlap with
486 ;;; the results start and count. (Also, it's nice to be able to target
487 ;;; them.)
504 ;; The stack frame is burnt and RETurned from if there are no
505 ;; stack values. In this case quickly reallocate sufficient space.
507 register-arg-count)
514 ;; FIXME: This can't be efficient, but LEA (my first choice)
515 ;; doesn't do subtraction.
520 ;; dtc: this writes the registers onto the stack even if they are
521 ;; not needed, only the number specified in rcx are used and have
522 ;; stack allocated to them. No harm is done.
525 for i downfrom -1
534 ;;; VOP that can be inherited by unknown values receivers. The main thing this
535 ;;; handles is allocation of the result temporaries.
539 values-start)
542 nvals)
545 \f
546 ;;;; local call with unknown values convention return
549 #+nil
553 #+nil
565 ;;; The local call convention doesn't fit that well with x86-style
566 ;;; calls. Emit a header for local calls to pop the return address
567 ;;; in the right place.
578 ;;; Non-TR local call for a fixed number of values passed according to
579 ;;; the unknown values convention.
580 ;;;
581 ;;; FP is the frame pointer in install before doing the call.
582 ;;;
583 ;;; NFP would be the number-stack frame pointer if we had a separate
584 ;;; number stack.
585 ;;;
586 ;;; Args are the argument passing locations, which are specified only
587 ;;; to terminate their lifetimes in the caller.
588 ;;;
589 ;;; VALUES are the return value locations (wired to the standard
590 ;;; passing locations). NVALS is the number of values received.
591 ;;;
592 ;;; Save is the save info, which we can ignore since saving has been
593 ;;; done.
594 ;;;
595 ;;; TARGET is a continuation pointing to the start of the called
596 ;;; function.
614 ;;; Non-TR local call for a variable number of return values passed according
615 ;;; to the unknown values convention. The results are the start of the values
616 ;;; glob and the number of values received.
633 \f
634 ;;;; local call with known values return
636 ;;; Non-TR local call with known return locations. Known-value return
637 ;;; works just like argument passing in local call.
638 ;;;
639 ;;; Note: we can't use normal load-tn allocation for the fixed args,
640 ;;; since all registers may be tied up by the more operand. Instead,
641 ;;; we use MAYBE-LOAD-STACK-TN.
657 \f
658 ;;; From Douglas Crosher
659 ;;; Return from known values call. We receive the return locations as
660 ;;; arguments to terminate their lifetimes in the returning function. We
661 ;;; restore FP and CSP and jump to the Return-PC.
672 ;; Zot all of the stack except for the old-fp and return-pc.
676 \f
677 ;;;; full call
678 ;;;
679 ;;; There is something of a cross-product effect with full calls.
680 ;;; Different versions are used depending on whether we know the
681 ;;; number of arguments or the name of the called function, and
682 ;;; whether we want fixed values, unknown values, or a tail call.
683 ;;;
684 ;;; In full call, the arguments are passed creating a partial frame on
685 ;;; the stack top and storing stack arguments into that frame. On
686 ;;; entry to the callee, this partial frame is pointed to by FP.
688 ;;; This macro helps in the definition of full call VOPs by avoiding
689 ;;; code replication in defining the cross-product VOPs.
690 ;;;
691 ;;; NAME is the name of the VOP to define.
692 ;;;
693 ;;; NAMED is true if the first argument is an fdefinition object whose
694 ;;; definition is to be called.
695 ;;;
696 ;;; RETURN is either :FIXED, :UNKNOWN or :TAIL:
697 ;;; -- If :FIXED, then the call is for a fixed number of values, returned in
698 ;;; the standard passing locations (passed as result operands).
699 ;;; -- If :UNKNOWN, then the result values are pushed on the stack, and the
700 ;;; result values are specified by the Start and Count as in the
701 ;;; unknown-values continuation representation.
702 ;;; -- If :TAIL, then do a tail-recursive call. No values are returned.
703 ;;; The Old-Fp and Return-PC are passed as the second and third arguments.
704 ;;;
705 ;;; In non-tail calls, the pointer to the stack arguments is passed as
706 ;;; the last fixed argument. If Variable is false, then the passing
707 ;;; locations are passed as a more arg. Variable is true if there are
708 ;;; a variable number of arguments passed on the stack. Variable
709 ;;; cannot be specified with :TAIL return. TR variable argument call
710 ;;; is implemented separately.
711 ;;;
712 ;;; In tail call with fixed arguments, the passing locations are
713 ;;; passed as a more arg, but there is no new-FP, since the arguments
714 ;;; have been set up in the current frame.
720 (:args
742 (:info
746 step-instrumenting)
748 (:ignore
752 ;; We pass either the fdefn object (for named call) or
753 ;; the actual function object (for unnamed call) in
754 ;; RAX. With named call, closure-tramp will replace it
755 ;; with the real function and invoke the real function
756 ;; for closures. Non-closures do not need this value,
757 ;; so don't care what shows up in it.
758 (:temporary
759 (:sc descriptor-reg
760 :offset rax-offset
763 rax)
765 ;; We pass the number of arguments in RCX.
768 ;; With variable call, we have to load the
769 ;; register-args out of the (new) stack frame before
770 ;; doing the call. Therefore, we have to tell the
771 ;; lifetime stuff that we need to use them.
785 old-fp-tmp)))
792 15
794 ;; This has to be done before the frame pointer is
795 ;; changed! RAX stores the 'lexical environment' needed
796 ;; for closures.
801 ;; For variable call, compute the number of
802 ;; arguments and move some of the arguments to
803 ;; registers.
805 ;; Compute the number of arguments.
811 ;; Move the necessary args to registers,
812 ;; this moves them all even if they are
813 ;; not all needed.
816 for index downfrom -1
824 ;; return to so might as well use that clue.
825 ;; This seems really important to the
826 ;; implementation of things like
827 ;; (without-interrupts ...)
828 ;;
829 ;; dtc; Could be doing a tail call from a
830 ;; known-local-call etc in which the old-fp
831 ;; or ret-pc are in regs or in non-standard
832 ;; places. If the passing location were
833 ;; wired to the stack in standard locations
834 ;; then these moves will be un-necessary;
835 ;; this is probably best for the x86.
840 ;; FIXME: FORMAT T for stale
841 ;; diagnostic output (several of
842 ;; them around here), ick
846 rbp-tn
851 rbp-tn
854 ;; For tail call, we have to push the
855 ;; return-pc so that it looks like we CALLed
856 ;; despite the fact that we are going to JMP.
858 ))
860 ;; For non-tail call, we have to save our
861 ;; frame pointer and install the new frame
862 ;; pointer. We can't load stack tns after this
863 ;; point.
865 ;; here which doesn't leave room for the
866 ;; ofp/ret stuff.
868 ;; The variable args are on the stack and
869 ;; become the frame, but there may be <3
870 ;; args and 3 stack slots are assumed
871 ;; allocate on the call. So need to ensure
872 ;; there are at least 3 slots. This hack
873 ;; just adds 3 more.
877 ;; Bias the new-fp for use as an fp
881 ;; Save the fp
886 )))
892 DONE
900 n-word-bytes)
901 other-pointer-lowtag)
903 fun-pointer-lowtag))))
905 (:fixed
907 (:unknown
910 node)))
923 ;;; This is defined separately, since it needs special code that BLT's
924 ;;; the arguments down. All the real work is done in the assembly
925 ;;; routine. We just set things up so that it can find what it needs.
936 ;; Move these into the passing locations if they are not already there.
939 ;; And jump to the assembly routine.
942 \f
943 ;;;; unknown values return
945 ;;; Return a single-value using the Unknown-Values convention.
946 ;;;
947 ;;; pfw--get wired-tn conflicts sometimes if register sc specd for args
948 ;;; having problems targeting args to regs -- using temps instead.
949 ;;;
950 ;;; First off, modifying the return-pc defeats the branch-prediction
951 ;;; optimizations on modern CPUs quite handily. Second, we can do all
952 ;;; this without needing a temp register. Fixed the latter, at least.
953 ;;; -- AB 2006/Feb/04
961 ;; Drop stack above old-fp
963 ;; Clear the multiple-value return flag
965 ;; Restore the old frame pointer
967 ;; And return.
970 ;;; Do unknown-values return of a fixed (other than 1) number of
971 ;;; values. The VALUES are required to be set up in the standard
972 ;;; passing locations. NVALS is the number of values returned.
973 ;;;
974 ;;; Basically, we just load RCX with the number of values returned and
975 ;;; RBX with a pointer to the values, set RSP to point to the end of
976 ;;; the values, and jump directly to return-pc.
983 ;; In the case of other than one value, we need these registers to
984 ;; tell the caller where they are and how many there are.
987 ;; We need to stretch the lifetime of return-pc past the argument
988 ;; registers so that we can default the argument registers without
989 ;; trashing return-pc.
1000 ;; This is handled in RETURN-SINGLE.
1002 ;; Establish the values pointer and values count.
1008 ;; Pre-default any argument register that need it.
1015 ;; Set the multiple value return flag.
1017 ;; And away we go. Except that return-pc is still on the
1018 ;; stack and we've changed the stack pointer. So we have to
1019 ;; tell it to index off of RBX instead of RBP.
1025 ;; Some values are on the stack after RETURN-PC and OLD-FP,
1026 ;; can't return normally and some slots of the frame will
1027 ;; be used as temporaries by the receiver.
1028 ;;
1029 ;; Clear as much of the stack as possible, but not past the
1030 ;; old frame address.
1041 ;;; Do unknown-values return of an arbitrary number of values (passed
1042 ;;; on the stack.) We check for the common case of a single return
1043 ;;; value, and do that inline using the normal single value return
1044 ;;; convention. Otherwise, we branch off to code that calls an
1045 ;;; assembly-routine.
1046 ;;;
1047 ;;; The assembly routine takes the following args:
1048 ;;; RCX -- number of values to find there.
1049 ;;; RSI -- pointer to where to find the values.
1064 ;; Check for the single case.
1068 ;; Return with one value.
1070 ;; Clear the stack until ocfp.
1072 ;; clear the multiple-value return flag
1074 ;; Out of here.
1077 ;; Nope, not the single case. Jump to the assembly routine.
1083 \f
1084 ;;;; XEP hackery
1086 ;;; Get the lexical environment from its passing location.
1092 ;; Get result.
1095 ;;; Copy a &MORE arg from the argument area to the end of the current
1096 ;;; frame. FIXED is the number of non-&MORE arguments.
1103 ;; Avoid the copy if there are no more args.
1110 ;; Create a negated copy of the number of arguments to allow us to
1111 ;; use EA calculations in order to do scaled subtraction.
1115 ;; Allocate the space on the stack.
1116 ;; stack = rbp + sp->fp-offset - (max 3 frame-size) - (nargs - fixed)
1117 ;; if we'd move SP backward, swap the meaning of rsp and source;
1118 ;; otherwise, we'd be accessing values below SP, and that's no good
1119 ;; if a signal interrupts this code sequence. In that case, store
1120 ;; the final value in rsp after the stack-stack memmove loop.
1122 rsp-tn
1123 source)
1130 ;; Now: nargs>=1 && nargs>fixed
1132 ;; Save the original count of args.
1136 ;; the code above only moves the final value of rsp in
1137 ;; rsp directly if that condition is satisfied. Currently,
1138 ;; r-a-c is 3, so the aver is OK. If the calling convention
1139 ;; ever changes, the logic above with LEA will have to be
1140 ;; adjusted.
1142 ;; We must stop when we run out of stack args, not when we
1143 ;; run out of more args.
1144 ;; Number to copy = nargs-3
1146 ;; Everything of interest in registers.
1149 ;; Number to copy = nargs-fixed
1152 ;; Initialize R8 to be the end of args.
1153 ;; Swap with SP if necessary to mirror the previous condition
1155 source
1156 rsp-tn)
1161 ;; src: rbp + temp + sp->fp
1162 ;; dst: rbp + temp + sp->fp + (fixed - (max 3 [stack-size]))
1168 ;; dst is lower than src, copy forward
1170 ;; We used to use REP MOVS here, but on modern x86 it performs
1171 ;; much worse than an explicit loop for small blocks.
1180 ;; dst is higher than src; copy backward
1187 temp)
1189 ;; done with the stack--stack copy. Reset RSP to its final
1190 ;; value
1192 DO-REGS
1194 ;; Here: nargs>=1 && nargs>fixed
1196 ;; Now we have to deposit any more args that showed up in
1197 ;; registers.
1200 ;; Store it relative to rbp
1214 ;; Don't deposit any more than there are.
1222 JUST-ALLOC-FRAME
1229 DONE))
1273 ;;; Turn more arg (context, count) into a list.
1293 ;; Check to see whether there are no args, and just return NIL if so.
1299 ;; Set up the result.
1301 ;; Jump into the middle of the loop, 'cause that's where we want
1302 ;; to start.
1305 ;; Compute a pointer to the next cons.
1307 ;; Store a pointer to this cons in the CDR of the previous cons.
1310 ;; Grab one value and stash it in the car of this cons.
1314 ;; Go back for more.
1317 ;; NIL out the last cons.
1321 ;;; Return the location and size of the &MORE arg glob created by
1322 ;;; COPY-MORE-ARG. SUPPLIED is the total number of arguments supplied
1323 ;;; (originally passed in RCX). FIXED is the number of non-rest
1324 ;;; arguments.
1325 ;;;
1326 ;;; We must duplicate some of the work done by COPY-MORE-ARG, since at
1327 ;;; that time the environment is in a pretty brain-damaged state,
1328 ;;; preventing this info from being returned as values. What we do is
1329 ;;; compute supplied - fixed, and return a pointer that many words
1330 ;;; below the current stack top.
1343 ;; SP at this point points at the last arg pushed.
1344 ;; Point to the first more-arg, not above it.
1346 :index count
1376 ;;; Various other error signallers.
1384 control-stack constant)))
1385 args))
1393 object type)
1395 object layout)
1402 ;; Signal an error about an untagged number.
1403 ;; These are pretty much boilerplate and could be generic except:
1404 ;; - the names of the SCs could differ between backends (or maybe not?)
1405 ;; - in the "/c" case, the older backends don't eval the errcode
1406 ;; And the 6 vops above ought to be generic too...
1407 ;; FIXME: there are still some occurrences of
1408 ;; note: doing signed word to integer coercion
1409 ;; in regard to SB-C::%TYPE-CHECK-ERROR. Figure out why.
1414 ;; Types are trees of symbols, so 'any-reg' is not
1415 ;; really possible.
1420 ;; cost is a smidgen less than type-check-error
1421 ;; otherwise this does not get selected.
1435 ;;; Single-stepping
1438 ;; We use different ways of representing whether stepping is on on
1439 ;; +SB-THREAD / -SB-THREAD: on +SB-THREAD, we use a slot in the
1440 ;; thread structure. On -SB-THREAD we use the value of a static
1441 ;; symbol. Things are done this way, since reading a thread-local
1442 ;; slot from a symbol would require an extra register on +SB-THREAD,
1443 ;; and reading a slot from a thread structure would require an extra
1444 ;; register on -SB-THREAD. While this isn't critical for x86-64,
1445 ;; it's more serious for x86.
1446 #!+sb-thread
1448 :base thread-base-tn
1450 nil-value)
1451 #!-sb-thread
1457 nil-value))
1466 DONE