1 ;;; Guile RTL assembler
3 ;;; Copyright (C) 2001, 2009, 2010, 2012, 2013 Free Software Foundation, Inc.
5 ;;; This library is free software; you can redistribute it and/or
6 ;;; modify it under the terms of the GNU Lesser General Public
7 ;;; License as published by the Free Software Foundation; either
8 ;;; version 3 of the License, or (at your option) any later version.
10 ;;; This library is distributed in the hope that it will be useful,
11 ;;; but WITHOUT ANY WARRANTY; without even the implied warranty of
12 ;;; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 ;;; Lesser General Public License for more details.
15 ;;; You should have received a copy of the GNU Lesser General Public
16 ;;; License along with this library; if not, write to the Free Software
17 ;;; Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21 ;;; This module implements an assembler that creates an ELF image from
22 ;;; RTL assembly and macro-assembly. The input can be given in
23 ;;; s-expression form, like ((OP ARG ...) ...). Internally there is a
24 ;;; procedural interface, the emit-OP procedures, but that is not
25 ;;; currently exported.
27 ;;; "Primitive instructions" correspond to RTL VM operations.
28 ;;; Assemblers for primitive instructions are generated programmatically
29 ;;; from (rtl-instruction-list), which itself is derived from the VM
30 ;;; sources. There are also "macro-instructions" like "label" or
31 ;;; "load-constant" that expand to 0 or more primitive instructions.
33 ;;; The assembler also handles some higher-level tasks, like creating
34 ;;; the symbol table, other metadata sections, creating a constant table
35 ;;; for the whole compilation unit, and writing the dynamic section of
36 ;;; the ELF file along with the appropriate initialization routines.
38 ;;; Most compilers will want to use the trio of make-assembler,
39 ;;; emit-text, and link-assembly. That will result in the creation of
40 ;;; an ELF image as a bytevector, which can then be loaded using
41 ;;; load-thunk-from-memory, or written to disk as a .go file.
45 (define-module (system vm assembler)
46 #:use-module (system base target)
47 #:use-module (system vm instruction)
48 #:use-module (system vm dwarf)
49 #:use-module (system vm elf)
50 #:use-module (system vm linker)
51 #:use-module (system vm objcode)
52 #:use-module (rnrs bytevectors)
53 #:use-module (ice-9 binary-ports)
54 #:use-module (ice-9 vlist)
55 #:use-module (ice-9 match)
56 #:use-module (srfi srfi-1)
57 #:use-module (srfi srfi-4)
58 #:use-module (srfi srfi-9)
59 #:use-module (srfi srfi-11)
60 #:export (make-assembler
68 ;;; RTL code consists of 32-bit units, often subdivided in some way.
69 ;;; These helpers create one 32-bit unit from multiple components.
71 (define-inlinable (pack-u8-u24 x y)
73 (error "out of range" x))
76 (define-inlinable (pack-u8-s24 x y)
78 (error "out of range" x))
84 (else (error "out of range" y)))
87 (define-inlinable (pack-u1-u7-u24 x y z)
89 (error "out of range" x))
91 (error "out of range" y))
92 (logior x (ash y 1) (ash z 8)))
94 (define-inlinable (pack-u8-u12-u12 x y z)
96 (error "out of range" x))
98 (error "out of range" y))
99 (logior x (ash y 8) (ash z 20)))
101 (define-inlinable (pack-u8-u8-u16 x y z)
103 (error "out of range" x))
105 (error "out of range" y))
106 (logior x (ash y 8) (ash z 16)))
108 (define-inlinable (pack-u8-u8-u8-u8 x y z w)
110 (error "out of range" x))
112 (error "out of range" y))
114 (error "out of range" z))
115 (logior x (ash y 8) (ash z 16) (ash w 24)))
117 (define-syntax pack-flags
119 ;; Add clauses as needed.
120 ((pack-flags f1 f2) (logior (if f1 (ash 1 0) 0)
121 (if f2 (ash 2 0) 0)))))
123 ;;; Helpers to read and write 32-bit units in a buffer.
125 (define-syntax-rule (u32-ref buf n)
126 (bytevector-u32-native-ref buf (* n 4)))
128 (define-syntax-rule (u32-set! buf n val)
129 (bytevector-u32-native-set! buf (* n 4) val))
131 (define-syntax-rule (s32-ref buf n)
132 (bytevector-s32-native-ref buf (* n 4)))
134 (define-syntax-rule (s32-set! buf n val)
135 (bytevector-s32-native-set! buf (* n 4) val))
140 ;;; A <meta> entry collects metadata for one procedure. Procedures are
141 ;;; written as contiguous ranges of RTL code.
143 (define-syntax-rule (assert-match arg pattern kind)
145 (unless (match x (pattern #t) (_ #f))
146 (error (string-append "expected " kind) x))))
148 (define-record-type <meta>
149 (%make-meta label properties low-pc high-pc arities)
152 (properties meta-properties set-meta-properties!)
154 (high-pc meta-high-pc set-meta-high-pc!)
155 (arities meta-arities set-meta-arities!))
157 (define (make-meta label properties low-pc)
158 (assert-match label (? symbol?) "symbol")
159 (assert-match properties (((? symbol?) . _) ...) "alist with symbolic keys")
160 (%make-meta label properties low-pc #f '()))
162 (define (meta-name meta)
163 (assq-ref (meta-properties meta) 'name))
165 ;; Metadata for one <lambda-case>.
166 (define-record-type <arity>
167 (make-arity req opt rest kw-indices allow-other-keys?
173 (kw-indices arity-kw-indices)
174 (allow-other-keys? arity-allow-other-keys?)
175 (low-pc arity-low-pc)
176 (high-pc arity-high-pc set-arity-high-pc!))
178 (define-syntax *block-size* (identifier-syntax 32))
180 ;;; An assembler collects all of the words emitted during assembly, and
181 ;;; also maintains ancillary information such as the constant table, a
182 ;;; relocation list, and so on.
184 ;;; RTL code consists of 32-bit units. We emit RTL code using native
185 ;;; endianness. If we're targeting a foreign endianness, we byte-swap
186 ;;; the bytevector as a whole instead of conditionalizing each access.
188 (define-record-type <asm>
189 (make-asm cur idx start prev written
193 shstrtab next-section-number
197 ;; We write RTL code into what is logically a growable vector,
198 ;; implemented as a list of blocks. asm-cur is the current block, and
199 ;; asm-idx is the current index into that block, in 32-bit units.
201 (cur asm-cur set-asm-cur!)
202 (idx asm-idx set-asm-idx!)
204 ;; asm-start is an absolute position, indicating the offset of the
205 ;; beginning of an instruction (in u32 units). It is updated after
206 ;; writing all the words for one primitive instruction. It models the
207 ;; position of the instruction pointer during execution, given that
208 ;; the RTL VM updates the IP only at the end of executing the
209 ;; instruction, and is thus useful for computing offsets between two
210 ;; points in a program.
212 (start asm-start set-asm-start!)
214 ;; The list of previously written blocks.
216 (prev asm-prev set-asm-prev!)
218 ;; The number of u32 words written in asm-prev, which is the same as
219 ;; the offset of the current block.
221 (written asm-written set-asm-written!)
223 ;; An alist of symbol -> position pairs, indicating the labels defined
224 ;; in this compilation unit.
226 (labels asm-labels set-asm-labels!)
228 ;; A list of relocations needed by the program text. We use an
229 ;; internal representation for relocations, and handle textualn
230 ;; relative relocations in the assembler. Other kinds of relocations
231 ;; are later reified as linker relocations and resolved by the linker.
233 (relocs asm-relocs set-asm-relocs!)
235 ;; Target information.
237 (word-size asm-word-size)
238 (endianness asm-endianness)
240 ;; The constant table, as a vhash of object -> label. All constants
241 ;; get de-duplicated and written into separate sections -- either the
242 ;; .rodata section, for read-only data, or .data, for constants that
243 ;; need initialization at load-time (like symbols). Constants can
244 ;; depend on other constants (e.g. a symbol depending on a stringbuf),
245 ;; so order in this table is important.
247 (constants asm-constants set-asm-constants!)
249 ;; A list of RTL instructions needed to initialize the constants.
250 ;; Will run in a thunk with 2 local variables.
252 (inits asm-inits set-asm-inits!)
254 ;; The shstrtab, for section names.
256 (shstrtab asm-shstrtab set-asm-shstrtab!)
258 ;; The section number for the next section to be written.
260 (next-section-number asm-next-section-number set-asm-next-section-number!)
262 ;; A list of <meta>, corresponding to procedure metadata.
264 (meta asm-meta set-asm-meta!)
266 ;; A list of (pos . source) pairs, indicating source information. POS
267 ;; is relative to the beginning of the text section, and SOURCE is in
268 ;; the same format that source-properties returns.
270 (sources asm-sources set-asm-sources!))
272 (define-inlinable (fresh-block)
273 (make-u32vector *block-size*))
275 (define* (make-assembler #:key (word-size (target-word-size))
276 (endianness (target-endianness)))
277 "Create an assembler for a given target @var{word-size} and
278 @var{endianness}, falling back to appropriate values for the configured
280 (make-asm (fresh-block) 0 0 '() 0
281 (make-hash-table) '()
284 (make-string-table) 1
287 (define (intern-section-name! asm string)
288 "Add a string to the section name table (shstrtab)."
289 (string-table-intern! (asm-shstrtab asm) string))
291 (define-inlinable (asm-pos asm)
292 "The offset of the next word to be written into the code buffer, in
294 (+ (asm-idx asm) (asm-written asm)))
296 (define (allocate-new-block asm)
297 "Close off the current block, and arrange for the next word to be
298 written to a fresh block."
299 (let ((new (fresh-block)))
300 (set-asm-prev! asm (cons (asm-cur asm) (asm-prev asm)))
301 (set-asm-written! asm (asm-pos asm))
302 (set-asm-cur! asm new)
303 (set-asm-idx! asm 0)))
305 (define-inlinable (emit asm u32)
306 "Emit one 32-bit word into the instruction stream. Assumes that there
307 is space for the word, and ensures that there is space for the next
309 (u32-set! (asm-cur asm) (asm-idx asm) u32)
310 (set-asm-idx! asm (1+ (asm-idx asm)))
311 (if (= (asm-idx asm) *block-size*)
312 (allocate-new-block asm)))
314 (define-inlinable (make-reloc type label base word)
315 "Make an internal relocation of type @var{type} referencing symbol
316 @var{label}, @var{word} words after position @var{start}. @var{type}
317 may be x8-s24, indicating a 24-bit relative label reference that can be
318 fixed up by the assembler, or s32, indicating a 32-bit relative
319 reference that needs to be fixed up by the linker."
320 (list type label base word))
322 (define-inlinable (reset-asm-start! asm)
323 "Reset the asm-start after writing the words for one instruction."
324 (set-asm-start! asm (asm-pos asm)))
326 (define (record-label-reference asm label)
327 "Record an x8-s24 local label reference. This value will get patched
328 up later by the assembler."
329 (let* ((start (asm-start asm))
331 (reloc (make-reloc 'x8-s24 label start (- pos start))))
332 (set-asm-relocs! asm (cons reloc (asm-relocs asm)))))
334 (define* (record-far-label-reference asm label #:optional (offset 0))
335 "Record an s32 far label reference. This value will get patched up
336 later by the linker."
337 (let* ((start (- (asm-start asm) offset))
339 (reloc (make-reloc 's32 label start (- pos start))))
340 (set-asm-relocs! asm (cons reloc (asm-relocs asm)))))
346 ;;; Primitive assemblers are defined by expanding `assembler' for each
347 ;;; opcode in `(rtl-instruction-list)'.
350 (eval-when (expand compile load eval)
351 (define (id-append ctx a b)
352 (datum->syntax ctx (symbol-append (syntax->datum a) (syntax->datum b)))))
354 (define-syntax assembler
356 (define-syntax op-case
359 ((_ asm name ((type arg ...) code ...) clause ...)
360 #`(if (eq? name 'type)
361 (with-syntax (((arg ...) (generate-temporaries #'(arg ...))))
364 (op-case asm name clause ...)))
366 #'(error "unmatched name" name)))))
368 (define (pack-first-word asm opcode type)
369 (with-syntax ((opcode opcode))
375 (emit asm (pack-u8-u24 opcode arg)))
377 (record-label-reference asm label)
380 (emit asm (pack-u8-u8-u16 opcode a (object-address imm))))
382 (emit asm (pack-u8-u12-u12 opcode a b)))
384 (emit asm (pack-u8-u8-u8-u8 opcode a b c))))))
386 (define (pack-tail-word asm type)
390 (emit asm (pack-u8-u24 a b)))
392 (record-label-reference asm label)
395 (emit asm (pack-u8-u8-u16 a b (object-address imm))))
397 (emit asm (pack-u8-u12-u12 a b c)))
398 ((U8_U8_U8_U8 a b c d)
399 (emit asm (pack-u8-u8-u8-u8 a b c d)))
403 (let ((val (object-address imm)))
404 (unless (zero? (ash val -32))
405 (error "FIXME: enable truncation of negative fixnums when cross-compiling"))
408 (unless (= (asm-word-size asm) 8)
409 (error "make-long-immediate unavailable for this target"))
410 (emit asm (ash (object-address imm) -32))
411 (emit asm (logand (object-address imm) (1- (ash 1 32)))))
414 (record-far-label-reference asm label)
417 (record-far-label-reference asm label)
420 (record-far-label-reference asm label)
423 (record-far-label-reference asm label
424 (* offset (/ (asm-word-size asm) 4)))
427 (emit asm (pack-u8-u24 0 a)))
429 (emit asm (pack-u8-u12-u12 0 a b)))
431 (record-label-reference asm label)
434 (record-label-reference asm label)
435 (emit asm (pack-u1-u7-u24 (if a 1 0) 0 0)))
436 ((B1_U7_L24 a b label)
437 (record-label-reference asm label)
438 (emit asm (pack-u1-u7-u24 (if a 1 0) b 0)))
440 (emit asm (pack-u1-u7-u24 (if a 1 0) 0 0)))
442 (emit asm (pack-u1-u7-u24 (if a 1 0) 0 b)))))
445 ((_ name opcode word0 word* ...)
446 (with-syntax ((((formal0 ...)
448 (pack-first-word #'asm
449 (syntax->datum #'opcode)
450 (syntax->datum #'word0)))
453 (map (lambda (word) (pack-tail-word #'asm word))
454 (syntax->datum #'(word* ...)))))
455 #'(lambda (asm formal0 ... formal* ... ...)
456 (unless (asm? asm) (error "not an asm"))
459 (reset-asm-start! asm)))))))
461 (define assemblers (make-hash-table))
463 (define-syntax define-assembler
466 ((_ name opcode kind arg ...)
467 (with-syntax ((emit (id-append #'name #'emit- #'name)))
470 (let ((emit (assembler name opcode arg ...)))
471 (hashq-set! assemblers 'name emit)
475 (define-syntax visit-opcodes
478 ((visit-opcodes macro arg ...)
479 (with-syntax (((inst ...)
480 (map (lambda (x) (datum->syntax #'macro x))
481 (rtl-instruction-list))))
483 (macro arg ... . inst)
486 (visit-opcodes define-assembler)
488 (define (emit-text asm instructions)
489 "Assemble @var{instructions} using the assembler @var{asm}.
490 @var{instructions} is a sequence of RTL instructions, expressed as a
491 list of lists. This procedure can be called many times before calling
492 @code{link-assembly}."
493 (for-each (lambda (inst)
494 (apply (or (hashq-ref assemblers (car inst))
495 (error 'bad-instruction inst))
503 ;;; The constant table records a topologically sorted set of literal
504 ;;; constants used by a program. For example, a pair uses its car and
505 ;;; cdr, a string uses its stringbuf, etc.
507 ;;; Some things we want to add to the constant table are not actually
508 ;;; Scheme objects: for example, stringbufs, cache cells for toplevel
509 ;;; references, or cache cells for non-closure procedures. For these we
510 ;;; define special record types and add instances of those record types
514 (define-inlinable (immediate? x)
515 "Return @code{#t} if @var{x} is immediate, and @code{#f} otherwise."
516 (not (zero? (logand (object-address x) 6))))
518 (define-record-type <stringbuf>
519 (make-stringbuf string)
521 (string stringbuf-string))
523 (define-record-type <static-procedure>
524 (make-static-procedure code)
526 (code static-procedure-code))
528 (define-record-type <uniform-vector-backing-store>
529 (make-uniform-vector-backing-store bytes element-size)
530 uniform-vector-backing-store?
531 (bytes uniform-vector-backing-store-bytes)
532 (element-size uniform-vector-backing-store-element-size))
534 (define-record-type <cache-cell>
535 (make-cache-cell scope key)
537 (scope cache-cell-scope)
538 (key cache-cell-key))
540 (define (simple-vector? obj)
542 (equal? (array-shape obj) (list (list 0 (1- (vector-length obj)))))))
544 (define (simple-uniform-vector? obj)
546 (symbol? (array-type obj))
547 (equal? (array-shape obj) (list (list 0 (1- (array-length obj)))))))
549 (define (statically-allocatable? x)
550 "Return @code{#t} if a non-immediate constant can be allocated
551 statically, and @code{#f} if it would need some kind of runtime
553 (or (pair? x) (string? x) (stringbuf? x) (static-procedure? x) (array? x)))
555 (define (intern-constant asm obj)
556 "Add an object to the constant table, and return a label that can be
557 used to reference it. If the object is already present in the constant
558 table, its existing label is used directly."
560 (intern-constant asm obj))
561 (define (field dst n obj)
562 (let ((src (recur obj)))
564 (if (statically-allocatable? obj)
565 `((static-patch! ,dst ,n ,src))
566 `((static-ref 1 ,src)
567 (static-set! 1 ,dst ,n)))
569 (define (intern obj label)
572 (append (field label 0 (car obj))
573 (field label 1 (cdr obj))))
574 ((simple-vector? obj)
575 (let lp ((i 0) (inits '()))
576 (if (< i (vector-length obj))
578 (append-reverse (field label (1+ i) (vector-ref obj i))
581 ((stringbuf? obj) '())
582 ((static-procedure? obj)
583 `((static-patch! ,label 1 ,(static-procedure-code obj))))
584 ((cache-cell? obj) '())
586 `((make-non-immediate 1 ,(recur (symbol->string obj)))
588 (static-set! 1 ,label 0)))
590 `((static-patch! ,label 1 ,(recur (make-stringbuf obj)))))
592 `((static-ref 1 ,(recur (keyword->symbol obj)))
593 (symbol->keyword 1 1)
594 (static-set! 1 ,label 0)))
596 `((make-non-immediate 1 ,(recur (number->string obj)))
598 (static-set! 1 ,label 0)))
599 ((uniform-vector-backing-store? obj) '())
600 ((simple-uniform-vector? obj)
601 `((static-patch! ,label 2
602 ,(recur (make-uniform-vector-backing-store
603 (uniform-array->bytevector obj)
605 ;; Bitvectors are addressed in
608 (uniform-vector-element-size obj)))))))
610 (error "don't know how to intern" obj))))
612 ((immediate? obj) #f)
613 ((vhash-assoc obj (asm-constants asm)) => cdr)
615 ;; Note that calling intern may mutate asm-constants and
616 ;; asm-constant-inits.
617 (let* ((label (gensym "constant"))
618 (inits (intern obj label)))
619 (set-asm-constants! asm (vhash-cons obj label (asm-constants asm)))
620 (set-asm-inits! asm (append-reverse inits (asm-inits asm)))
623 (define (intern-non-immediate asm obj)
624 "Intern a non-immediate into the constant table, and return its
626 (when (immediate? obj)
627 (error "expected a non-immediate" obj))
628 (intern-constant asm obj))
630 (define (intern-cache-cell asm scope key)
631 "Intern a cache cell into the constant table, and return its label.
632 If there is already a cache cell with the given scope and key, it is
634 (intern-constant asm (make-cache-cell scope key)))
636 ;; Return the label of the cell that holds the module for a scope.
637 (define (intern-module-cache-cell asm scope)
638 "Intern a cache cell for a module, and return its label."
639 (intern-cache-cell asm scope #t))
645 ;;; Macro assemblers bridge the gap between primitive instructions and
646 ;;; some higher-level operations.
649 (define-syntax define-macro-assembler
652 ((_ (name arg ...) body body* ...)
653 (with-syntax ((emit (id-append #'name #'emit- #'name)))
656 (let ((emit (lambda (arg ...) body body* ...)))
657 (hashq-set! assemblers 'name emit)
661 (define-macro-assembler (load-constant asm dst obj)
664 (let ((bits (object-address obj)))
666 ((and (< dst 256) (zero? (ash bits -16)))
667 (emit-make-short-immediate asm dst obj))
668 ((zero? (ash bits -32))
669 (emit-make-long-immediate asm dst obj))
671 (emit-make-long-long-immediate asm dst obj)))))
672 ((statically-allocatable? obj)
673 (emit-make-non-immediate asm dst (intern-non-immediate asm obj)))
675 (emit-static-ref asm dst (intern-non-immediate asm obj)))))
677 (define-macro-assembler (load-static-procedure asm dst label)
678 (let ((loc (intern-constant asm (make-static-procedure label))))
679 (emit-make-non-immediate asm dst loc)))
681 (define-syntax-rule (define-tc7-macro-assembler name tc7)
682 (define-macro-assembler (name asm slot invert? label)
683 (emit-br-if-tc7 asm slot invert? tc7 label)))
685 ;; Keep in sync with tags.h. Part of Guile's ABI. Currently unused
686 ;; macro assemblers are commented out. See also
687 ;; *branching-primcall-arities* in (language cps primitives), the set of
688 ;; macro-instructions in assembly.scm, and
689 ;; disassembler.scm:code-annotation.
691 ;; FIXME: Define all tc7 values in Scheme in one place, derived from
693 (define-tc7-macro-assembler br-if-symbol 5)
694 (define-tc7-macro-assembler br-if-variable 7)
695 (define-tc7-macro-assembler br-if-vector 13)
696 ;(define-tc7-macro-assembler br-if-weak-vector 13)
697 (define-tc7-macro-assembler br-if-string 21)
698 ;(define-tc7-macro-assembler br-if-heap-number 23)
699 ;(define-tc7-macro-assembler br-if-stringbuf 39)
700 (define-tc7-macro-assembler br-if-bytevector 77)
701 ;(define-tc7-macro-assembler br-if-pointer 31)
702 ;(define-tc7-macro-assembler br-if-hashtable 29)
703 ;(define-tc7-macro-assembler br-if-fluid 37)
704 ;(define-tc7-macro-assembler br-if-dynamic-state 45)
705 ;(define-tc7-macro-assembler br-if-frame 47)
706 ;(define-tc7-macro-assembler br-if-objcode 53)
707 ;(define-tc7-macro-assembler br-if-vm 55)
708 ;(define-tc7-macro-assembler br-if-vm-cont 71)
709 ;(define-tc7-macro-assembler br-if-rtl-program 69)
710 ;(define-tc7-macro-assembler br-if-program 79)
711 ;(define-tc7-macro-assembler br-if-weak-set 85)
712 ;(define-tc7-macro-assembler br-if-weak-table 87)
713 ;(define-tc7-macro-assembler br-if-array 93)
714 (define-tc7-macro-assembler br-if-bitvector 95)
715 ;(define-tc7-macro-assembler br-if-port 125)
716 ;(define-tc7-macro-assembler br-if-smob 127)
718 (define-macro-assembler (begin-program asm label properties)
719 (emit-label asm label)
720 (let ((meta (make-meta label properties (asm-start asm))))
721 (set-asm-meta! asm (cons meta (asm-meta asm)))))
723 (define-macro-assembler (end-program asm)
724 (let ((meta (car (asm-meta asm))))
725 (set-meta-high-pc! meta (asm-start asm))
726 (set-meta-arities! meta (reverse (meta-arities meta)))))
728 (define-macro-assembler (begin-standard-arity asm req nlocals alternate)
729 (emit-begin-opt-arity asm req '() #f nlocals alternate))
731 (define-macro-assembler (begin-opt-arity asm req opt rest nlocals alternate)
732 (emit-begin-kw-arity asm req opt rest '() #f nlocals alternate))
734 (define-macro-assembler (begin-kw-arity asm req opt rest kw-indices
735 allow-other-keys? nlocals alternate)
736 (assert-match req ((? symbol?) ...) "list of symbols")
737 (assert-match opt ((? symbol?) ...) "list of symbols")
738 (assert-match rest (or #f (? symbol?)) "#f or symbol")
739 (assert-match kw-indices (((? keyword?) . (? integer?)) ...)
740 "alist of keyword -> integer")
741 (assert-match allow-other-keys? (? boolean?) "boolean")
742 (assert-match nlocals (? integer?) "integer")
743 (assert-match alternate (or #f (? symbol?)) "#f or symbol")
744 (let* ((meta (car (asm-meta asm)))
745 (arity (make-arity req opt rest kw-indices allow-other-keys?
747 ;; The procedure itself is in slot 0, in the standard calling
748 ;; convention. For procedure prologues, nreq includes the
749 ;; procedure, so here we add 1.
750 (nreq (1+ (length req)))
752 (rest? (->bool rest)))
753 (set-meta-arities! meta (cons arity (meta-arities meta)))
755 ((or allow-other-keys? (pair? kw-indices))
756 (emit-kw-prelude asm nreq nopt rest? kw-indices allow-other-keys?
758 ((or rest? (pair? opt))
759 (emit-opt-prelude asm nreq nopt rest? nlocals alternate))
761 (emit-standard-prelude asm nreq nlocals alternate)))))
763 (define-macro-assembler (end-arity asm)
764 (let ((arity (car (meta-arities (car (asm-meta asm))))))
765 (set-arity-high-pc! arity (asm-start asm))))
767 (define-macro-assembler (standard-prelude asm nreq nlocals alternate)
770 (emit-br-if-nargs-ne asm nreq alternate)
771 (emit-alloc-frame asm nlocals))
772 ((and (< nreq (ash 1 12)) (< (- nlocals nreq) (ash 1 12)))
773 (emit-assert-nargs-ee/locals asm nreq (- nlocals nreq)))
775 (emit-assert-nargs-ee asm nreq)
776 (emit-alloc-frame asm nlocals))))
778 (define-macro-assembler (opt-prelude asm nreq nopt rest? nlocals alternate)
780 (emit-br-if-nargs-lt asm nreq alternate)
781 (emit-assert-nargs-ge asm nreq))
784 (emit-bind-rest asm (+ nreq nopt)))
786 (emit-br-if-nargs-gt asm (+ nreq nopt) alternate))
788 (emit-assert-nargs-le asm (+ nreq nopt))))
789 (emit-alloc-frame asm nlocals))
791 (define-macro-assembler (kw-prelude asm nreq nopt rest? kw-indices
792 allow-other-keys? nlocals alternate)
794 (emit-br-if-nargs-lt asm nreq alternate)
795 (emit-assert-nargs-ge asm nreq))
796 (let ((ntotal (fold (lambda (kw ntotal)
798 (((? keyword?) . idx)
799 (max (1+ idx) ntotal))))
800 (+ nreq nopt) kw-indices)))
801 ;; FIXME: port 581f410f
802 (emit-bind-kwargs asm nreq
803 (pack-flags allow-other-keys? rest?)
806 (intern-constant asm kw-indices))
807 (emit-alloc-frame asm nlocals)))
809 (define-macro-assembler (label asm sym)
810 (hashq-set! (asm-labels asm) sym (asm-start asm)))
812 (define-macro-assembler (source asm source)
813 (set-asm-sources! asm (acons (asm-start asm) source (asm-sources asm))))
815 (define-macro-assembler (cache-current-module! asm module scope)
816 (let ((mod-label (intern-module-cache-cell asm scope)))
817 (emit-static-set! asm module mod-label 0)))
819 (define-macro-assembler (cached-toplevel-box asm dst scope sym bound?)
820 (let ((sym-label (intern-non-immediate asm sym))
821 (mod-label (intern-module-cache-cell asm scope))
822 (cell-label (intern-cache-cell asm scope sym)))
823 (emit-toplevel-box asm dst cell-label mod-label sym-label bound?)))
825 (define-macro-assembler (cached-module-box asm dst module-name sym public? bound?)
826 (let* ((sym-label (intern-non-immediate asm sym))
827 (key (cons public? module-name))
828 (mod-name-label (intern-constant asm key))
829 (cell-label (intern-cache-cell asm key sym)))
830 (emit-module-box asm dst cell-label mod-name-label sym-label bound?)))
836 ;;; Helper for linking objects.
839 (define (make-object asm name bv relocs labels . kwargs)
840 "Make a linker object. This helper handles interning the name in the
841 shstrtab, assigning the size, allocating a fresh index, and defining a
842 corresponding linker symbol for the start of the section."
843 (let ((name-idx (intern-section-name! asm (symbol->string name)))
844 (index (asm-next-section-number asm)))
845 (set-asm-next-section-number! asm (1+ index))
846 (make-linker-object (apply make-elf-section
849 #:size (bytevector-length bv)
852 (cons (make-linker-symbol name 0) labels))))
858 ;;; Linking the constant table. This code is somewhat intertwingled
859 ;;; with the intern-constant code above, as that procedure also
860 ;;; residualizes instructions to initialize constants at load time.
863 (define (write-immediate asm buf pos x)
864 (let ((val (object-address x))
865 (endianness (asm-endianness asm)))
866 (case (asm-word-size asm)
867 ((4) (bytevector-u32-set! buf pos val endianness))
868 ((8) (bytevector-u64-set! buf pos val endianness))
869 (else (error "bad word size" asm)))))
871 (define (emit-init-constants asm)
872 "If there is writable data that needs initialization at runtime, emit
873 a procedure to do that and return its label. Otherwise return
875 (let ((inits (asm-inits asm)))
876 (and (not (null? inits))
877 (let ((label (gensym "init-constants")))
879 `((begin-program ,label ())
880 (assert-nargs-ee/locals 1 1)
882 (load-constant 1 ,*unspecified*)
887 (define (link-data asm data name)
888 "Link the static data for a program into the @var{name} section (which
889 should be .data or .rodata), and return the resulting linker object.
890 @var{data} should be a vhash mapping objects to labels."
891 (define (align address alignment)
893 (modulo (- alignment (modulo address alignment)) alignment)))
895 (define tc7-vector 13)
896 (define stringbuf-shared-flag #x100)
897 (define stringbuf-wide-flag #x400)
898 (define tc7-stringbuf 39)
899 (define tc7-narrow-stringbuf
900 (+ tc7-stringbuf stringbuf-shared-flag))
901 (define tc7-wide-stringbuf
902 (+ tc7-stringbuf stringbuf-shared-flag stringbuf-wide-flag))
903 (define tc7-ro-string (+ 21 #x200))
904 (define tc7-rtl-program 69)
905 (define tc7-bytevector 77)
906 (define tc7-bitvector 95)
908 (let ((word-size (asm-word-size asm))
909 (endianness (asm-endianness asm)))
910 (define (byte-length x)
913 (let ((x (stringbuf-string x)))
915 (case (string-bytes-per-char x)
916 ((1) (1+ (string-length x)))
917 ((4) (* (1+ (string-length x)) 4))
918 (else (error "bad string bytes per char" x))))))
919 ((static-procedure? x)
926 (* (1+ (vector-length x)) word-size))
927 ((simple-uniform-vector? x)
929 ((uniform-vector-backing-store? x)
930 (bytevector-length (uniform-vector-backing-store-bytes x)))
934 (define (write-constant-reference buf pos x)
935 ;; The asm-inits will fix up any reference to a non-immediate.
936 (write-immediate asm buf pos (if (immediate? x) x #f)))
938 (define (write buf pos obj)
941 (let* ((x (stringbuf-string obj))
942 (len (string-length x))
943 (tag (if (= (string-bytes-per-char x) 1)
945 tc7-wide-stringbuf)))
948 (bytevector-u32-set! buf pos tag endianness)
949 (bytevector-u32-set! buf (+ pos 4) len endianness))
951 (bytevector-u64-set! buf pos tag endianness)
952 (bytevector-u64-set! buf (+ pos 8) len endianness))
954 (error "bad word size" asm)))
955 (let ((pos (+ pos (* word-size 2))))
956 (case (string-bytes-per-char x)
960 (let ((u8 (char->integer (string-ref x i))))
961 (bytevector-u8-set! buf (+ pos i) u8)
963 (bytevector-u8-set! buf (+ pos i) 0))))
967 (let ((u32 (char->integer (string-ref x i))))
968 (bytevector-u32-set! buf (+ pos (* i 4)) u32 endianness)
970 (bytevector-u32-set! buf (+ pos (* i 4)) 0 endianness))))
971 (else (error "bad string bytes per char" x))))))
973 ((static-procedure? obj)
976 (bytevector-u32-set! buf pos tc7-rtl-program endianness)
977 (bytevector-u32-set! buf (+ pos 4) 0 endianness))
979 (bytevector-u64-set! buf pos tc7-rtl-program endianness)
980 (bytevector-u64-set! buf (+ pos 8) 0 endianness))
981 (else (error "bad word size"))))
984 (write-immediate asm buf pos #f))
987 (let ((tag (logior tc7-ro-string (ash (string-length obj) 8))))
990 (bytevector-u32-set! buf pos tc7-ro-string endianness)
991 (write-immediate asm buf (+ pos 4) #f) ; stringbuf
992 (bytevector-u32-set! buf (+ pos 8) 0 endianness)
993 (bytevector-u32-set! buf (+ pos 12) (string-length obj) endianness))
995 (bytevector-u64-set! buf pos tc7-ro-string endianness)
996 (write-immediate asm buf (+ pos 8) #f) ; stringbuf
997 (bytevector-u64-set! buf (+ pos 16) 0 endianness)
998 (bytevector-u64-set! buf (+ pos 24) (string-length obj) endianness))
999 (else (error "bad word size")))))
1002 (write-constant-reference buf pos (car obj))
1003 (write-constant-reference buf (+ pos word-size) (cdr obj)))
1005 ((simple-vector? obj)
1006 (let* ((len (vector-length obj))
1007 (tag (logior tc7-vector (ash len 8))))
1009 ((4) (bytevector-u32-set! buf pos tag endianness))
1010 ((8) (bytevector-u64-set! buf pos tag endianness))
1011 (else (error "bad word size")))
1013 (when (< i (vector-length obj))
1014 (let ((pos (+ pos word-size (* i word-size)))
1015 (elt (vector-ref obj i)))
1016 (write-constant-reference buf pos elt)
1020 (write-immediate asm buf pos #f))
1023 (write-immediate asm buf pos #f))
1026 (write-immediate asm buf pos #f))
1028 ((simple-uniform-vector? obj)
1029 (let ((tag (if (bitvector? obj)
1031 (let ((type-code (uniform-vector-element-type-code obj)))
1032 (logior tc7-bytevector (ash type-code 7))))))
1035 (bytevector-u32-set! buf pos tag endianness)
1036 (bytevector-u32-set! buf (+ pos 4)
1037 (if (bitvector? obj)
1038 (bitvector-length obj)
1039 (bytevector-length obj))
1040 endianness) ; length
1041 (bytevector-u32-set! buf (+ pos 8) 0 endianness) ; pointer
1042 (write-immediate asm buf (+ pos 12) #f)) ; owner
1044 (bytevector-u64-set! buf pos tag endianness)
1045 (bytevector-u64-set! buf (+ pos 8)
1046 (if (bitvector? obj)
1047 (bitvector-length obj)
1048 (bytevector-length obj))
1049 endianness) ; length
1050 (bytevector-u64-set! buf (+ pos 16) 0 endianness) ; pointer
1051 (write-immediate asm buf (+ pos 24) #f)) ; owner
1052 (else (error "bad word size")))))
1054 ((uniform-vector-backing-store? obj)
1055 (let ((bv (uniform-vector-backing-store-bytes obj)))
1056 (bytevector-copy! bv 0 buf pos (bytevector-length bv))
1057 (unless (or (= 1 (uniform-vector-backing-store-element-size obj))
1058 (eq? endianness (native-endianness)))
1059 ;; Need to swap units of element-size bytes
1060 (error "FIXME: Implement byte order swap"))))
1063 (error "unrecognized object" obj))))
1066 ((vlist-null? data) #f)
1068 (let* ((byte-len (vhash-fold (lambda (k v len)
1069 (+ (byte-length k) (align len 8)))
1071 (buf (make-bytevector byte-len 0)))
1072 (let lp ((i 0) (pos 0) (symbols '()))
1073 (if (< i (vlist-length data))
1074 (let* ((pair (vlist-ref data i))
1076 (obj-label (cdr pair)))
1079 (align (+ (byte-length obj) pos) 8)
1080 (cons (make-linker-symbol obj-label pos) symbols)))
1081 (make-object asm name buf '() symbols
1083 ('.data (logior SHF_ALLOC SHF_WRITE))
1084 ('.rodata SHF_ALLOC))))))))))
1086 (define (link-constants asm)
1087 "Link sections to hold constants needed by the program text emitted
1090 Returns three values: an object for the .rodata section, an object for
1091 the .data section, and a label for an initialization procedure. Any of
1092 these may be @code{#f}."
1093 (define (shareable? x)
1097 (and (immediate? (car x)) (immediate? (cdr x))))
1100 (or (= i (vector-length x))
1101 (and (immediate? (vector-ref x i))
1103 ((uniform-vector-backing-store? x) #t)
1105 (let* ((constants (asm-constants asm))
1106 (len (vlist-length constants)))
1111 (values (link-data asm ro '.rodata)
1112 (link-data asm rw '.data)
1113 (emit-init-constants asm))
1114 (let ((pair (vlist-ref constants i)))
1115 (if (shareable? (car pair))
1116 (lp (1+ i) (vhash-consq (car pair) (cdr pair) ro) rw)
1117 (lp (1+ i) ro (vhash-consq (car pair) (cdr pair) rw))))))))
1122 ;;; Linking program text.
1125 (define (process-relocs buf relocs labels)
1126 "Patch up internal x8-s24 relocations, and any s32 relocations that
1127 reference symbols in the text section. Return a list of linker
1128 relocations for references to symbols defined outside the text section."
1130 (lambda (reloc tail)
1132 ((type label base word)
1133 (let ((abs (hashq-ref labels label))
1134 (dst (+ base word)))
1138 (let ((rel (- abs base)))
1139 (s32-set! buf dst rel)
1141 (cons (make-linker-reloc 'rel32/4 (* dst 4) word label)
1145 (error "unbound near relocation" reloc))
1146 (let ((rel (- abs base))
1147 (u32 (u32-ref buf dst)))
1148 (u32-set! buf dst (pack-u8-s24 (logand u32 #xff) rel))
1150 (else (error "bad relocation kind" reloc)))))))
1154 (define (process-labels labels)
1155 "Define linker symbols for the label-offset map in @var{labels}.
1156 The offsets are expected to be expressed in words."
1157 (hash-map->list (lambda (label loc)
1158 (make-linker-symbol label (* loc 4)))
1161 (define (swap-bytes! buf)
1162 "Patch up the text buffer @var{buf}, swapping the endianness of each
1164 (unless (zero? (modulo (bytevector-length buf) 4))
1165 (error "unexpected length"))
1166 (let ((byte-len (bytevector-length buf)))
1168 (unless (= pos byte-len)
1169 (bytevector-u32-set!
1171 (bytevector-u32-ref buf pos (endianness big))
1172 (endianness little))
1175 (define (link-text-object asm)
1176 "Link the .rtl-text section, swapping the endianness of the bytes if
1178 (let ((buf (make-u32vector (asm-pos asm))))
1179 (let lp ((pos 0) (prev (reverse (asm-prev asm))))
1181 (let ((byte-size (* (asm-idx asm) 4)))
1182 (bytevector-copy! (asm-cur asm) 0 buf pos byte-size)
1183 (unless (eq? (asm-endianness asm) (native-endianness))
1185 (make-object asm '.rtl-text
1187 (process-relocs buf (asm-relocs asm)
1189 (process-labels (asm-labels asm))))
1190 (let ((len (* *block-size* 4)))
1191 (bytevector-copy! (car prev) 0 buf pos len)
1192 (lp (+ pos len) (cdr prev)))))))
1198 ;;; Linking other sections of the ELF file, like the dynamic segment,
1199 ;;; the symbol table, etc.
1202 (define (link-dynamic-section asm text rw rw-init)
1203 "Link the dynamic section for an ELF image with RTL text, given the
1204 writable data section @var{rw} needing fixup from the procedure with
1205 label @var{rw-init}. @var{rw-init} may be false. If @var{rw} is true,
1206 it will be added to the GC roots at runtime."
1207 (define-syntax-rule (emit-dynamic-section word-size %set-uword! reloc-type)
1208 (let* ((endianness (asm-endianness asm))
1209 (bv (make-bytevector (* word-size (if rw (if rw-init 12 10) 6)) 0))
1212 (%set-uword! bv (* i word-size) uword endianness)))
1216 (set! relocs (cons (make-linker-reloc 'reloc-type
1217 (* i word-size) 0 label)
1219 (%set-uword! bv (* i word-size) 0 endianness))))
1220 (set-uword! 0 DT_GUILE_RTL_VERSION)
1221 (set-uword! 1 #x02020000)
1222 (set-uword! 2 DT_GUILE_ENTRY)
1223 (set-label! 3 '.rtl-text)
1227 (set-uword! 4 DT_GUILE_GC_ROOT)
1228 (set-label! 5 '.data)
1229 (set-uword! 6 DT_GUILE_GC_ROOT_SZ)
1230 (set-uword! 7 (bytevector-length (linker-object-bv rw)))
1233 (set-uword! 8 DT_INIT) ; constants
1234 (set-label! 9 rw-init)
1235 (set-uword! 10 DT_NULL)
1238 (set-uword! 8 DT_NULL)
1241 (set-uword! 4 DT_NULL)
1243 (make-object asm '.dynamic bv relocs '()
1244 #:type SHT_DYNAMIC #:flags SHF_ALLOC)))
1245 (case (asm-word-size asm)
1246 ((4) (emit-dynamic-section 4 bytevector-u32-set! abs32/1))
1247 ((8) (emit-dynamic-section 8 bytevector-u64-set! abs64/1))
1248 (else (error "bad word size" asm))))
1250 (define (link-shstrtab asm)
1251 "Link the string table for the section headers."
1252 (intern-section-name! asm ".shstrtab")
1253 (make-object asm '.shstrtab
1254 (link-string-table! (asm-shstrtab asm))
1256 #:type SHT_STRTAB #:flags 0))
1258 (define (link-symtab text-section asm)
1259 (let* ((endianness (asm-endianness asm))
1260 (word-size (asm-word-size asm))
1261 (size (elf-symbol-len word-size))
1262 (meta (reverse (asm-meta asm)))
1264 (strtab (make-string-table))
1265 (bv (make-bytevector (* n size) 0)))
1266 (define (intern-string! name)
1267 (string-table-intern! strtab (if name (symbol->string name) "")))
1270 (let ((name (intern-string! (meta-name meta))))
1271 (write-elf-symbol bv (* n size) endianness word-size
1274 ;; Symbol value and size are measured in
1276 #:value (* 4 (meta-low-pc meta))
1277 #:size (* 4 (- (meta-high-pc meta)
1278 (meta-low-pc meta)))
1280 #:visibility STV_HIDDEN
1281 #:shndx (elf-section-index text-section)))))
1283 (let ((strtab (make-object asm '.strtab
1284 (link-string-table! strtab)
1286 #:type SHT_STRTAB #:flags 0)))
1287 (values (make-object asm '.symtab
1290 #:type SHT_SYMTAB #:flags 0 #:entsize size
1291 #:link (elf-section-index
1292 (linker-object-section strtab)))
1295 ;;; The .guile.arities section describes the arities that a function can
1296 ;;; have. It is in two parts: a sorted array of headers describing
1297 ;;; basic arities, and an array of links out to a string table (and in
1298 ;;; the case of keyword arguments, to the data section) for argument
1299 ;;; names. The whole thing is prefixed by a uint32 indicating the
1300 ;;; offset of the end of the headers array.
1302 ;;; The arity headers array is a packed array of structures of the form:
1304 ;;; struct arity_header {
1305 ;;; uint32_t low_pc;
1306 ;;; uint32_t high_pc;
1307 ;;; uint32_t offset;
1313 ;;; All of the offsets and addresses are 32 bits. We can expand in the
1314 ;;; future to use 64-bit offsets if appropriate, but there are other
1315 ;;; aspects of RTL that constrain us to a total image that fits in 32
1316 ;;; bits, so for the moment we'll simplify the problem space.
1318 ;;; The following flags values are defined:
1321 ;;; #x2: allow-other-keys?
1322 ;;; #x4: has-keyword-args?
1323 ;;; #x8: is-case-lambda?
1325 ;;; Functions with a single arity specify their number of required and
1326 ;;; optional arguments in nreq and nopt, and do not have the
1327 ;;; is-case-lambda? flag set. Their "offset" member links to an array
1328 ;;; of pointers into the associated .guile.arities.strtab string table,
1329 ;;; identifying the argument names. This offset is relative to the
1330 ;;; start of the .guile.arities section. Links for required arguments
1331 ;;; are first, in order, as uint32 values. Next follow the optionals,
1332 ;;; then the rest link if has-rest? is set, then a link to the "keyword
1333 ;;; indices" literal if has-keyword-args? is set. Unlike the other
1334 ;;; links, the kw-indices link points into the data section, and is
1335 ;;; relative to the ELF image as a whole.
1337 ;;; Functions with no arities have no arities information present in the
1338 ;;; .guile.arities section.
1340 ;;; Functions with multiple arities are preceded by a header with
1341 ;;; is-case-lambda? set. All other fields are 0, except low-pc and
1342 ;;; high-pc which should be the bounds of the whole function. Headers
1343 ;;; for the individual arities follow. In this way the whole headers
1344 ;;; array is sorted in increasing low-pc order, and case-lambda clauses
1345 ;;; are contained within the [low-pc, high-pc] of the case-lambda
1348 ;; Length of the prefix to the arities section, in bytes.
1349 (define arities-prefix-len 4)
1351 ;; Length of an arity header, in bytes.
1352 (define arity-header-len (* 6 4))
1354 ;; The offset of "offset" within arity header, in bytes.
1355 (define arity-header-offset-offset (* 2 4))
1357 (define-syntax-rule (pack-arity-flags has-rest? allow-other-keys?
1358 has-keyword-args? is-case-lambda?)
1359 (logior (if has-rest? (ash 1 0) 0)
1360 (if allow-other-keys? (ash 1 1) 0)
1361 (if has-keyword-args? (ash 1 2) 0)
1362 (if is-case-lambda? (ash 1 3) 0)))
1364 (define (meta-arities-size meta)
1365 (define (lambda-size arity)
1367 (* 4 ;; name pointers
1368 (+ (length (arity-req arity))
1369 (length (arity-opt arity))
1370 (if (arity-rest arity) 1 0)
1371 (if (pair? (arity-kw-indices arity)) 1 0)))))
1372 (define (case-lambda-size arities)
1374 arity-header-len ;; case-lambda header
1375 (map lambda-size arities))) ;; the cases
1376 (match (meta-arities meta)
1378 ((arity) (lambda-size arity))
1379 (arities (case-lambda-size arities))))
1381 (define (write-arity-headers metas bv endianness)
1382 (define (write-arity-header* pos low-pc high-pc flags nreq nopt)
1383 (bytevector-u32-set! bv pos (* low-pc 4) endianness)
1384 (bytevector-u32-set! bv (+ pos 4) (* high-pc 4) endianness)
1385 (bytevector-u32-set! bv (+ pos 8) 0 endianness) ; offset
1386 (bytevector-u32-set! bv (+ pos 12) flags endianness)
1387 (bytevector-u32-set! bv (+ pos 16) nreq endianness)
1388 (bytevector-u32-set! bv (+ pos 20) nopt endianness))
1389 (define (write-arity-header pos arity)
1390 (write-arity-header* pos (arity-low-pc arity)
1391 (arity-high-pc arity)
1392 (pack-arity-flags (arity-rest arity)
1393 (arity-allow-other-keys? arity)
1394 (pair? (arity-kw-indices arity))
1396 (length (arity-req arity))
1397 (length (arity-opt arity))))
1398 (let lp ((metas metas) (pos arities-prefix-len) (offsets '()))
1401 ;; Fill in the prefix.
1402 (bytevector-u32-set! bv 0 pos endianness)
1403 (values pos (reverse offsets)))
1405 (match (meta-arities meta)
1406 (() (lp metas pos offsets))
1408 (write-arity-header pos arity)
1410 (+ pos arity-header-len)
1411 (acons arity (+ pos arity-header-offset-offset) offsets)))
1413 ;; Write a case-lambda header, then individual arities.
1414 ;; The case-lambda header's offset link is 0.
1415 (write-arity-header* pos (meta-low-pc meta) (meta-high-pc meta)
1416 (pack-arity-flags #f #f #f #t) 0 0)
1417 (let lp* ((arities arities) (pos (+ pos arity-header-len))
1420 (() (lp metas pos offsets))
1422 (write-arity-header pos arity)
1424 (+ pos arity-header-len)
1426 (+ pos arity-header-offset-offset)
1429 (define (write-arity-links asm bv pos arity-offset-pairs strtab)
1430 (define (write-symbol sym pos)
1431 (bytevector-u32-set! bv pos
1432 (string-table-intern! strtab (symbol->string sym))
1433 (asm-endianness asm))
1435 (define (write-kw-indices pos kw-indices)
1436 ;; FIXME: Assert that kw-indices is already interned.
1437 (make-linker-reloc 'abs32/1 pos 0
1438 (intern-constant asm kw-indices)))
1439 (let lp ((pos pos) (pairs arity-offset-pairs) (relocs '()))
1442 (unless (= pos (bytevector-length bv))
1443 (error "expected to fully fill the bytevector"
1444 pos (bytevector-length bv)))
1446 (((arity . offset) . pairs)
1447 (bytevector-u32-set! bv offset pos (asm-endianness asm))
1448 (let ((pos (fold write-symbol
1450 (append (arity-req arity)
1453 ((arity-rest arity) => list)
1455 (match (arity-kw-indices arity)
1456 (() (lp pos pairs relocs))
1460 (cons (write-kw-indices pos kw-indices) relocs)))))))))
1462 (define (link-arities asm)
1463 (let* ((endianness (asm-endianness asm))
1464 (metas (reverse (asm-meta asm)))
1465 (size (fold (lambda (meta size)
1466 (+ size (meta-arities-size meta)))
1469 (strtab (make-string-table))
1470 (bv (make-bytevector size 0)))
1471 (let ((kw-indices-relocs
1474 (write-arity-headers metas bv endianness))
1475 (lambda (pos arity-offset-pairs)
1476 (write-arity-links asm bv pos arity-offset-pairs strtab)))))
1477 (let ((strtab (make-object asm '.guile.arities.strtab
1478 (link-string-table! strtab)
1480 #:type SHT_STRTAB #:flags 0)))
1481 (values (make-object asm '.guile.arities
1483 kw-indices-relocs '()
1484 #:type SHT_PROGBITS #:flags 0
1485 #:link (elf-section-index
1486 (linker-object-section strtab)))
1490 ;;; The .guile.docstrs section is a packed, sorted array of (pc, str)
1491 ;;; values. Pc and str are both 32 bits wide. (Either could change to
1492 ;;; 64 bits if appropriate in the future.) Pc is the address of the
1493 ;;; entry to a program, relative to the start of the text section, in
1494 ;;; bytes, and str is an index into the associated .guile.docstrs.strtab
1495 ;;; string table section.
1498 ;; The size of a docstrs entry, in bytes.
1499 (define docstr-size 8)
1501 (define (link-docstrs asm)
1502 (define (find-docstrings)
1503 (filter-map (lambda (meta)
1504 (define (is-documentation? pair)
1505 (eq? (car pair) 'documentation))
1506 (let* ((props (meta-properties meta))
1507 (tail (find-tail is-documentation? props)))
1509 (not (find-tail is-documentation? (cdr tail)))
1510 (string? (cdar tail))
1511 (cons (* 4 (meta-low-pc meta)) (cdar tail)))))
1512 (reverse (asm-meta asm))))
1513 (let* ((endianness (asm-endianness asm))
1514 (docstrings (find-docstrings))
1515 (strtab (make-string-table))
1516 (bv (make-bytevector (* (length docstrings) docstr-size) 0)))
1517 (fold (lambda (pair pos)
1520 (bytevector-u32-set! bv pos pc endianness)
1521 (bytevector-u32-set! bv (+ pos 4)
1522 (string-table-intern! strtab string)
1524 (+ pos docstr-size))))
1527 (let ((strtab (make-object asm '.guile.docstrs.strtab
1528 (link-string-table! strtab)
1530 #:type SHT_STRTAB #:flags 0)))
1531 (values (make-object asm '.guile.docstrs
1534 #:type SHT_PROGBITS #:flags 0
1535 #:link (elf-section-index
1536 (linker-object-section strtab)))
1540 ;;; The .guile.procprops section is a packed, sorted array of (pc, addr)
1541 ;;; values. Pc and addr are both 32 bits wide. (Either could change to
1542 ;;; 64 bits if appropriate in the future.) Pc is the address of the
1543 ;;; entry to a program, relative to the start of the text section, and
1544 ;;; addr is the address of the associated properties alist, relative to
1545 ;;; the start of the ELF image.
1547 ;;; Since procedure properties are stored in the data sections, we need
1548 ;;; to link the procedures property section first. (Note that this
1549 ;;; constraint does not apply to the arities section, which may
1550 ;;; reference the data sections via the kw-indices literal, because
1551 ;;; assembling the text section already makes sure that the kw-indices
1555 ;; The size of a procprops entry, in bytes.
1556 (define procprops-size 8)
1558 (define (link-procprops asm)
1559 (define (assoc-remove-one alist key value-pred)
1562 ((((? (lambda (x) (eq? x key))) . value) . alist)
1563 (if (value-pred value)
1565 (acons key value alist)))
1567 (acons k v (assoc-remove-one alist key value-pred)))))
1568 (define (props-without-name-or-docstring meta)
1570 (assoc-remove-one (meta-properties meta) 'name (lambda (x) #t))
1573 (define (find-procprops)
1574 (filter-map (lambda (meta)
1575 (let ((props (props-without-name-or-docstring meta)))
1577 (cons (meta-low-pc meta) props))))
1578 (reverse (asm-meta asm))))
1579 (let* ((endianness (asm-endianness asm))
1580 (procprops (find-procprops))
1581 (bv (make-bytevector (* (length procprops) procprops-size) 0)))
1582 (let lp ((procprops procprops) (pos 0) (relocs '()))
1585 (make-object asm '.guile.procprops
1588 #:type SHT_PROGBITS #:flags 0))
1589 (((pc . props) . procprops)
1590 (bytevector-u32-set! bv pos pc endianness)
1592 (+ pos procprops-size)
1593 (cons (make-linker-reloc 'abs32/1 (+ pos 4) 0
1594 (intern-constant asm props))
1598 ;;; The DWARF .debug_info, .debug_abbrev, .debug_str, and .debug_loc
1599 ;;; sections provide line number and local variable liveness
1600 ;;; information. Their format is defined by the DWARF
1604 (define (asm-language asm)
1605 ;; FIXME: Plumb language through to the assembler.
1608 ;; -> 5 values: .debug_info, .debug_abbrev, .debug_str, .debug_loc, .debug_lines
1609 (define (link-debug asm)
1610 (define (put-s8 port val)
1611 (let ((bv (make-bytevector 1)))
1612 (bytevector-s8-set! bv 0 val)
1613 (put-bytevector port bv)))
1615 (define (put-u16 port val)
1616 (let ((bv (make-bytevector 2)))
1617 (bytevector-u16-set! bv 0 val (asm-endianness asm))
1618 (put-bytevector port bv)))
1620 (define (put-u32 port val)
1621 (let ((bv (make-bytevector 4)))
1622 (bytevector-u32-set! bv 0 val (asm-endianness asm))
1623 (put-bytevector port bv)))
1625 (define (put-u64 port val)
1626 (let ((bv (make-bytevector 8)))
1627 (bytevector-u64-set! bv 0 val (asm-endianness asm))
1628 (put-bytevector port bv)))
1630 (define (put-uleb128 port val)
1632 (let ((next (ash val -7)))
1636 (put-u8 port (logior #x80 (logand val #x7f)))
1639 (define (put-sleb128 port val)
1641 (if (<= 0 (+ val 64) 127)
1642 (put-u8 port (logand val #x7f))
1644 (put-u8 port (logior #x80 (logand val #x7f)))
1645 (lp (ash val -7))))))
1647 (define (port-position port)
1648 (seek port 0 SEEK_CUR))
1650 (define (meta->subprogram-die meta)
1654 => (lambda (name) `((name ,(symbol->string name)))))
1657 (low-pc ,(meta-label meta))
1658 (high-pc ,(* 4 (- (meta-high-pc meta) (meta-low-pc meta)))))))
1660 (define (make-compile-unit-die asm)
1662 (@ (producer ,(string-append "Guile " (version)))
1663 (language ,(asm-language asm))
1665 (high-pc ,(* 4 (asm-pos asm)))
1667 ,@(map meta->subprogram-die (reverse (asm-meta asm)))))
1669 (let-values (((die-port get-die-bv) (open-bytevector-output-port))
1671 ((abbrev-port get-abbrev-bv) (open-bytevector-output-port))
1672 ;; (tag has-kids? attrs forms) -> code
1673 ((abbrevs) vlist-null)
1674 ((strtab) (make-string-table))
1675 ((line-port get-line-bv) (open-bytevector-output-port))
1678 ((files) vlist-null))
1680 (define (write-abbrev code tag has-children? attrs forms)
1681 (put-uleb128 abbrev-port code)
1682 (put-uleb128 abbrev-port (tag-name->code tag))
1683 (put-u8 abbrev-port (children-name->code (if has-children? 'yes 'no)))
1684 (for-each (lambda (attr form)
1685 (put-uleb128 abbrev-port (attribute-name->code attr))
1686 (put-uleb128 abbrev-port (form-name->code form)))
1688 (put-uleb128 abbrev-port 0)
1689 (put-uleb128 abbrev-port 0))
1691 (define (intern-abbrev tag has-children? attrs forms)
1692 (let ((key (list tag has-children? attrs forms)))
1693 (match (vhash-assoc key abbrevs)
1695 (#f (let ((code (1+ (vlist-length abbrevs))))
1696 (set! abbrevs (vhash-cons key code abbrevs))
1697 (write-abbrev code tag has-children? attrs forms)
1700 (define (intern-file file)
1701 (match (vhash-assoc file files)
1703 (#f (let ((code (1+ (vlist-length files))))
1704 (set! files (vhash-cons file code files))
1707 (define (write-sources)
1708 ;; Choose line base and line range values that will allow for an
1709 ;; address advance range of 16 words. The special opcode range is
1710 ;; from 10 to 255, so 246 values.
1714 (let lp ((sources (asm-sources asm)) (out '()))
1716 (((pc . s) . sources)
1717 (let ((file (assq-ref s 'filename))
1718 (line (assq-ref s 'line))
1719 (col (assq-ref s 'column)))
1721 ;; Guile line and column numbers are 0-indexed, but
1722 ;; they are 1-indexed for DWARF.
1724 (if file (intern-file file) 0)
1729 ;; Compilation unit header for .debug_line. We write in
1730 ;; DWARF 2 format because more tools understand it than DWARF
1731 ;; 4, which incompatibly adds another field to this header.
1733 (put-u32 line-port 0) ; Length; will patch later.
1734 (put-u16 line-port 2) ; DWARF 2 format.
1735 (put-u32 line-port 0) ; Prologue length; will patch later.
1736 (put-u8 line-port 4) ; Minimum instruction length: 4 bytes.
1737 (put-u8 line-port 1) ; Default is-stmt: true.
1739 (put-s8 line-port base) ; Line base. See the DWARF standard.
1740 (put-u8 line-port range) ; Line range. See the DWARF standard.
1741 (put-u8 line-port 10) ; Opcode base: the first "special" opcode.
1743 ;; A table of the number of uleb128 arguments taken by each
1744 ;; of the standard opcodes.
1745 (put-u8 line-port 0) ; 1: copy
1746 (put-u8 line-port 1) ; 2: advance-pc
1747 (put-u8 line-port 1) ; 3: advance-line
1748 (put-u8 line-port 1) ; 4: set-file
1749 (put-u8 line-port 1) ; 5: set-column
1750 (put-u8 line-port 0) ; 6: negate-stmt
1751 (put-u8 line-port 0) ; 7: set-basic-block
1752 (put-u8 line-port 0) ; 8: const-add-pc
1753 (put-u8 line-port 1) ; 9: fixed-advance-pc
1755 ;; Include directories, as a zero-terminated sequence of
1756 ;; nul-terminated strings. Nothing, for the moment.
1757 (put-u8 line-port 0)
1759 ;; File table. For each file that contributes to this
1760 ;; compilation unit, a nul-terminated file name string, and a
1761 ;; uleb128 for each of directory the file was found in, the
1762 ;; modification time, and the file's size in bytes. We pass
1763 ;; zero for the latter three fields.
1768 (put-bytevector line-port (string->utf8 file))
1769 (put-u8 line-port 0)
1770 (put-uleb128 line-port 0) ; directory
1771 (put-uleb128 line-port 0) ; mtime
1772 (put-uleb128 line-port 0))) ; size
1776 (put-u8 line-port 0) ; 0 byte terminating file list.
1778 ;; Patch prologue length.
1779 (let ((offset (port-position line-port)))
1780 (seek line-port 6 SEEK_SET)
1781 (put-u32 line-port (- offset 10))
1782 (seek line-port offset SEEK_SET))
1784 ;; Now write the statement program.
1786 (define (extended-op opcode payload-len)
1787 (put-u8 line-port 0) ; extended op
1788 (put-uleb128 line-port (1+ payload-len)) ; payload-len + opcode
1789 (put-uleb128 line-port opcode))
1790 (define (set-address sym)
1791 (define (add-reloc! kind)
1793 (cons (make-linker-reloc kind
1794 (port-position line-port)
1798 (match (asm-word-size asm)
1801 (add-reloc! 'abs32/1)
1802 (put-u32 line-port 0))
1805 (add-reloc! 'abs64/1)
1806 (put-u64 line-port 0))))
1807 (define (end-sequence pc)
1808 (let ((pc-inc (- (asm-pos asm) pc)))
1809 (put-u8 line-port 2) ; advance-pc
1810 (put-uleb128 line-port pc-inc))
1812 (define (advance-pc pc-inc line-inc)
1813 (let ((spec (+ (- line-inc base) (* pc-inc range) 10)))
1815 ((or (< line-inc base) (>= line-inc (+ base range)))
1816 (advance-line line-inc)
1817 (advance-pc pc-inc 0))
1819 (put-u8 line-port spec))
1821 (put-u8 line-port 8) ; const-advance-pc
1822 (advance-pc (- pc-inc (floor/ (- 255 10) range))
1825 (put-u8 line-port 2) ; advance-pc
1826 (put-uleb128 line-port pc-inc)
1827 (advance-pc 0 line-inc)))))
1828 (define (advance-line inc)
1829 (put-u8 line-port 3)
1830 (put-sleb128 line-port inc))
1831 (define (set-file file)
1832 (put-u8 line-port 4)
1833 (put-uleb128 line-port file))
1834 (define (set-column col)
1835 (put-u8 line-port 5)
1836 (put-uleb128 line-port col))
1838 (set-address '.rtl-text)
1840 (let lp ((in out) (pc 0) (file 1) (line 1) (col 0))
1844 ;; There was no source info in the first place. Set
1845 ;; file register to 0 before adding final row.
1848 (((pc* file* line* col*) . in*)
1850 ((and (eqv? file file*) (eqv? line line*) (eqv? col col*))
1851 (lp in* pc file line col))
1853 (unless (eqv? col col*)
1855 (unless (eqv? file file*)
1857 (advance-pc (- pc* pc) (- line* line))
1858 (lp in* pc* file* line* col*)))))))))))
1860 (define (compute-code attr val)
1862 ('name (string-table-intern! strtab val))
1865 ('producer (string-table-intern! strtab val))
1866 ('language (language-name->code val))
1869 (define (exact-integer? val)
1870 (and (number? val) (integer? val) (exact? val)))
1872 (define (choose-form attr val code)
1874 ((string? val) 'strp)
1875 ((eq? attr 'stmt-list) 'sec-offset)
1876 ((exact-integer? code)
1878 ((< code 0) 'sleb128)
1879 ((<= code #xff) 'data1)
1880 ((<= code #xffff) 'data2)
1881 ((<= code #xffffffff) 'data4)
1882 ((<= code #xffffffffffffffff) 'data8)
1884 ((symbol? val) 'addr)
1885 (else (error "unhandled case" attr val code))))
1887 (define (add-die-relocation! kind sym)
1889 (cons (make-linker-reloc kind (port-position die-port) 0 sym)
1892 (define (write-value code form)
1894 ('data1 (put-u8 die-port code))
1895 ('data2 (put-u16 die-port code))
1896 ('data4 (put-u32 die-port code))
1897 ('data8 (put-u64 die-port code))
1898 ('uleb128 (put-uleb128 die-port code))
1899 ('sleb128 (put-sleb128 die-port code))
1901 (match (asm-word-size asm)
1903 (add-die-relocation! 'abs32/1 code)
1904 (put-u32 die-port 0))
1906 (add-die-relocation! 'abs64/1 code)
1907 (put-u64 die-port 0))))
1908 ('sec-offset (put-u32 die-port code))
1909 ('strp (put-u32 die-port code))))
1911 (define (write-die die)
1913 ((tag ('@ (attrs vals) ...) children ...)
1914 (let* ((codes (map compute-code attrs vals))
1915 (forms (map choose-form attrs vals codes))
1916 (has-children? (not (null? children)))
1917 (abbrev-code (intern-abbrev tag has-children? attrs forms)))
1918 (put-uleb128 die-port abbrev-code)
1919 (for-each write-value codes forms)
1921 (for-each write-die children)
1922 (put-uleb128 die-port 0))))))
1924 ;; Compilation unit header.
1925 (put-u32 die-port 0) ; Length; will patch later.
1926 (put-u16 die-port 4) ; DWARF 4.
1927 (put-u32 die-port 0) ; Abbrevs offset.
1928 (put-u8 die-port (asm-word-size asm)) ; Address size.
1930 (write-die (make-compile-unit-die asm))
1932 ;; Terminate the abbrevs list.
1933 (put-uleb128 abbrev-port 0)
1937 (values (let ((bv (get-die-bv)))
1938 ;; Patch DWARF32 length.
1939 (bytevector-u32-set! bv 0 (- (bytevector-length bv) 4)
1940 (asm-endianness asm))
1941 (make-object asm '.debug_info bv die-relocs '()
1942 #:type SHT_PROGBITS #:flags 0))
1943 (make-object asm '.debug_abbrev (get-abbrev-bv) '() '()
1944 #:type SHT_PROGBITS #:flags 0)
1945 (make-object asm '.debug_str (link-string-table! strtab) '() '()
1946 #:type SHT_PROGBITS #:flags 0)
1947 (make-object asm '.debug_loc #vu8() '() '()
1948 #:type SHT_PROGBITS #:flags 0)
1949 (let ((bv (get-line-bv)))
1950 ;; Patch DWARF32 length.
1951 (bytevector-u32-set! bv 0 (- (bytevector-length bv) 4)
1952 (asm-endianness asm))
1953 (make-object asm '.debug_line bv line-relocs '()
1954 #:type SHT_PROGBITS #:flags 0)))))
1956 (define (link-objects asm)
1957 (let*-values (;; Link procprops before constants, because it probably
1958 ;; interns more constants.
1959 ((procprops) (link-procprops asm))
1960 ((ro rw rw-init) (link-constants asm))
1961 ;; Link text object after constants, so that the
1962 ;; constants initializer gets included.
1963 ((text) (link-text-object asm))
1964 ((dt) (link-dynamic-section asm text rw rw-init))
1965 ((symtab strtab) (link-symtab (linker-object-section text) asm))
1966 ((arities arities-strtab) (link-arities asm))
1967 ((docstrs docstrs-strtab) (link-docstrs asm))
1968 ((dinfo dabbrev dstrtab dloc dline) (link-debug asm))
1969 ;; This needs to be linked last, because linking other
1970 ;; sections adds entries to the string table.
1971 ((shstrtab) (link-shstrtab asm)))
1973 (list text ro rw dt symtab strtab arities arities-strtab
1974 docstrs docstrs-strtab procprops
1975 dinfo dabbrev dstrtab dloc dline
1982 ;;; High-level public interfaces.
1985 (define* (link-assembly asm #:key (page-aligned? #t))
1986 "Produce an ELF image from the code and data emitted into @var{asm}.
1987 The result is a bytevector, by default linked so that read-only and
1988 writable data are on separate pages. Pass @code{#:page-aligned? #f} to
1989 disable this behavior."
1990 (link-elf (link-objects asm) #:page-aligned? page-aligned?))
1992 (define (assemble-program instructions)
1993 "Take the sequence of instructions @var{instructions}, assemble them
1994 into RTL code, link an image, and load that image from memory. Returns
1996 (let ((asm (make-assembler)))
1997 (emit-text asm instructions)
1998 (load-thunk-from-memory (link-assembly asm #:page-aligned? #f))))