xref: /linux/tools/lib/python/kdoc/kdoc_parser.py (revision 46d9c16115cf046ce564f0230126cae785f73826)
1#!/usr/bin/env python3
2# SPDX-License-Identifier: GPL-2.0
3# Copyright(c) 2025: Mauro Carvalho Chehab <mchehab@kernel.org>.
4#
5# pylint: disable=C0301,C0302,R0904,R0912,R0913,R0914,R0915,R0917,R1702
6
7"""
8Classes and functions related to reading a C language source or header FILE
9and extract embedded documentation comments from it.
10"""
11
12import sys
13import re
14from pprint import pformat
15
16from kdoc.c_lex import CTokenizer, tokenizer_set_log
17from kdoc.kdoc_re import KernRe
18from kdoc.kdoc_item import KdocItem
19
20#
21# Regular expressions used to parse kernel-doc markups at KernelDoc class.
22#
23# Let's declare them in lowercase outside any class to make it easier to
24# convert from the Perl script.
25#
26# As those are evaluated at the beginning, no need to cache them
27#
28
29# Allow whitespace at end of comment start.
30doc_start = KernRe(r'^/\*\*\s*$', cache=False)
31
32doc_end = KernRe(r'\*/', cache=False)
33doc_com = KernRe(r'\s*\*\s*', cache=False)
34doc_com_body = KernRe(r'\s*\* ?', cache=False)
35doc_decl = doc_com + KernRe(r'(\w+)', cache=False)
36
37# @params and a strictly limited set of supported section names
38# Specifically:
39#   Match @word:
40#         @...:
41#         @{section-name}:
42# while trying to not match literal block starts like "example::"
43#
44known_section_names = 'description|context|returns?|notes?|examples?'
45known_sections = KernRe(known_section_names, flags = re.I)
46doc_sect = doc_com + \
47    KernRe(r'\s*(@[.\w]+|@\.\.\.|' + known_section_names + r')\s*:([^:].*)?$',
48           flags=re.I, cache=False)
49
50doc_content = doc_com_body + KernRe(r'(.*)', cache=False)
51doc_inline_start = KernRe(r'^\s*/\*\*\s*$', cache=False)
52doc_inline_sect = KernRe(r'\s*\*\s*(@\s*[\w][\w\.]*\s*):(.*)', cache=False)
53doc_inline_end = KernRe(r'^\s*\*/\s*$', cache=False)
54doc_inline_oneline = KernRe(r'^\s*/\*\*\s*(@\s*[\w][\w\.]*\s*):\s*(.*)\s*\*/\s*$', cache=False)
55
56export_symbol = KernRe(r'^\s*EXPORT_SYMBOL(_GPL)?\s*\(\s*(\w+)\s*\)\s*', cache=False)
57export_symbol_ns = KernRe(r'^\s*EXPORT_SYMBOL_NS(_GPL)?\s*\(\s*(\w+)\s*,\s*"\S+"\)\s*', cache=False)
58
59type_param = KernRe(r"@(\w*((\.\w+)|(->\w+))*(\.\.\.)?)", cache=False)
60
61#
62# Tests for the beginning of a kerneldoc block in its various forms.
63#
64doc_block = doc_com + KernRe(r'DOC:\s*(.*)?', cache=False)
65doc_begin_data = KernRe(r"^\s*\*?\s*(struct|union|enum|typedef|var)\b\s*(\w*)", cache = False)
66doc_begin_func = KernRe(str(doc_com) +			# initial " * '
67                        r"(?:\w+\s*\*\s*)?" + 		# type (not captured)
68                        r'(?:define\s+)?' + 		# possible "define" (not captured)
69                        r'(\w+)\s*(?:\(\w*\))?\s*' +	# name and optional "(...)"
70                        r'(?:[-:].*)?$',		# description (not captured)
71                        cache = False)
72
73#
74# Ancillary functions
75#
76
77multi_space = KernRe(r'\s\s+')
78def trim_whitespace(s):
79    """
80    A little helper to get rid of excess white space.
81    """
82    return multi_space.sub(' ', s.strip())
83
84def trim_private_members(text):
85    """
86    Remove ``struct``/``enum`` members that have been marked "private".
87    """
88
89    tokens = CTokenizer(text)
90    return str(tokens)
91
92class state:
93    """
94    States used by the parser's state machine.
95    """
96
97    # Parser states
98    NORMAL        = 0        #: Normal code.
99    NAME          = 1        #: Looking for function name.
100    DECLARATION   = 2        #: We have seen a declaration which might not be done.
101    BODY          = 3        #: The body of the comment.
102    SPECIAL_SECTION = 4      #: Doc section ending with a blank line.
103    PROTO         = 5        #: Scanning prototype.
104    DOCBLOCK      = 6        #: Documentation block.
105    INLINE_NAME   = 7        #: Gathering doc outside main block.
106    INLINE_TEXT   = 8	     #: Reading the body of inline docs.
107
108    #: Names for each parser state.
109    name = [
110        "NORMAL",
111        "NAME",
112        "DECLARATION",
113        "BODY",
114        "SPECIAL_SECTION",
115        "PROTO",
116        "DOCBLOCK",
117        "INLINE_NAME",
118        "INLINE_TEXT",
119    ]
120
121
122SECTION_DEFAULT = "Description"  #: Default section.
123
124class KernelEntry:
125    """
126    Encapsulates a Kernel documentation entry.
127    """
128
129    def __init__(self, config, fname, ln):
130        self.config = config
131        self.fname = fname
132
133        self._contents = []
134        self.prototype = ""
135
136        self.warnings = []
137
138        self.parameterlist = []
139        self.parameterdescs = {}
140        self.parametertypes = {}
141        self.parameterdesc_start_lines = {}
142
143        self.sections_start_lines = {}
144        self.sections = {}
145
146        self.anon_struct_union = False
147
148        self.leading_space = None
149
150        self.fname = fname
151
152        # State flags
153        self.brcount = 0
154        self.declaration_start_line = ln + 1
155
156    #
157    # Management of section contents
158    #
159    def add_text(self, text):
160        """Add a new text to the entry contents list."""
161        self._contents.append(text)
162
163    def contents(self):
164        """Returns a string with all content texts that were added."""
165        return '\n'.join(self._contents) + '\n'
166
167    # TODO: rename to emit_message after removal of kernel-doc.pl
168    def emit_msg(self, ln, msg, *, warning=True):
169        """Emit a message."""
170
171        log_msg = f"{self.fname}:{ln} {msg}"
172
173        if not warning:
174            self.config.log.info(log_msg)
175            return
176
177        # Delegate warning output to output logic, as this way it
178        # will report warnings/info only for symbols that are output
179
180        self.warnings.append(log_msg)
181        return
182
183    def begin_section(self, line_no, title = SECTION_DEFAULT, dump = False):
184        """
185        Begin a new section.
186        """
187        if dump:
188            self.dump_section(start_new = True)
189        self.section = title
190        self.new_start_line = line_no
191
192    def dump_section(self, start_new=True):
193        """
194        Dumps section contents to arrays/hashes intended for that purpose.
195        """
196        #
197        # If we have accumulated no contents in the default ("description")
198        # section, don't bother.
199        #
200        if self.section == SECTION_DEFAULT and not self._contents:
201            return
202        name = self.section
203        contents = self.contents()
204
205        if type_param.match(name):
206            name = type_param.group(1)
207
208            self.parameterdescs[name] = contents
209            self.parameterdesc_start_lines[name] = self.new_start_line
210
211            self.new_start_line = 0
212
213        else:
214            if name in self.sections and self.sections[name] != "":
215                # Only warn on user-specified duplicate section names
216                if name != SECTION_DEFAULT:
217                    self.emit_msg(self.new_start_line,
218                                  f"duplicate section name '{name}'")
219                # Treat as a new paragraph - add a blank line
220                self.sections[name] += '\n' + contents
221            else:
222                self.sections[name] = contents
223                self.sections_start_lines[name] = self.new_start_line
224                self.new_start_line = 0
225
226#        self.config.log.debug("Section: %s : %s", name, pformat(vars(self)))
227
228        if start_new:
229            self.section = SECTION_DEFAULT
230            self._contents = []
231
232python_warning = False
233
234class KernelDoc:
235    """
236    Read a C language source or header FILE and extract embedded
237    documentation comments.
238    """
239
240    #: Name of context section.
241    section_context = "Context"
242
243    #: Name of return section.
244    section_return = "Return"
245
246    #: String to write when a parameter is not described.
247    undescribed = "-- undescribed --"
248
249    def __init__(self, config, fname, xforms, store_src=False):
250        """Initialize internal variables"""
251
252        self.fname = fname
253        self.config = config
254        self.xforms = xforms
255        self.store_src = store_src
256
257        tokenizer_set_log(self.config.log, f"{self.fname}: CMatch: ")
258
259        # Initial state for the state machines
260        self.state = state.NORMAL
261
262        # Store entry currently being processed
263        self.entry = None
264
265        # Place all potential outputs into an array
266        self.entries = []
267
268        #
269        # We need Python 3.7 for its "dicts remember the insertion
270        # order" guarantee
271        #
272        global python_warning
273        if (not python_warning and
274            sys.version_info.major == 3 and sys.version_info.minor < 7):
275
276            self.emit_msg(0,
277                          'Python 3.7 or later is required for correct results')
278            python_warning = True
279
280    def emit_msg(self, ln, msg, *, warning=True):
281        """Emit a message"""
282
283        if self.entry:
284            self.entry.emit_msg(ln, msg, warning=warning)
285            return
286
287        log_msg = f"{self.fname}:{ln} {msg}"
288
289        if warning:
290            self.config.log.warning(log_msg)
291        else:
292            self.config.log.info(log_msg)
293
294    def dump_section(self, start_new=True):
295        """
296        Dump section contents to arrays/hashes intended for that purpose.
297        """
298
299        if self.entry:
300            self.entry.dump_section(start_new)
301
302    # TODO: rename it to store_declaration after removal of kernel-doc.pl
303    def output_declaration(self, dtype, name, **args):
304        """
305        Store the entry into an entry array.
306
307        The actual output and output filters will be handled elsewhere.
308        """
309
310        item = KdocItem(name, self.fname, dtype,
311                        self.entry.declaration_start_line, **args)
312        item.warnings = self.entry.warnings
313
314        # Drop empty sections
315        # TODO: improve empty sections logic to emit warnings
316        sections = self.entry.sections
317        for section in ["Description", "Return"]:
318            if section in sections and not sections[section].rstrip():
319                del sections[section]
320        item.set_sections(sections, self.entry.sections_start_lines)
321        item.set_params(self.entry.parameterlist, self.entry.parameterdescs,
322                        self.entry.parametertypes,
323                        self.entry.parameterdesc_start_lines)
324        self.entries.append(item)
325
326        self.config.log.debug("Output: %s:%s = %s", dtype, name, pformat(args))
327
328    def emit_unused_warnings(self):
329        """
330        When the parser fails to produce a valid entry, it places some
331        warnings under `entry.warnings` that will be discarded when resetting
332        the state.
333
334        Ensure that those warnings are not lost.
335
336        .. note::
337
338              Because we are calling `config.warning()` here, those
339              warnings are not filtered by the `-W` parameters: they will all
340              be produced even when `-Wreturn`, `-Wshort-desc`, and/or
341              `-Wcontents-before-sections` are used.
342
343              Allowing those warnings to be filtered is complex, because it
344              would require storing them in a buffer and then filtering them
345              during the output step of the code, depending on the
346              selected symbols.
347        """
348        if self.entry and self.entry not in self.entries:
349            for log_msg in self.entry.warnings:
350                self.config.warning(log_msg)
351
352    def reset_state(self, ln):
353        """
354        Ancillary routine to create a new entry. It initializes all
355        variables used by the state machine.
356        """
357
358        self.emit_unused_warnings()
359
360        self.entry = KernelEntry(self.config, self.fname, ln)
361
362        # State flags
363        self.state = state.NORMAL
364
365    def push_parameter(self, ln, decl_type, param, dtype,
366                       org_arg, declaration_name):
367        """
368        Store parameters and their descriptions at self.entry.
369        """
370
371        if self.entry.anon_struct_union and dtype == "" and param == "}":
372            return  # Ignore the ending }; from anonymous struct/union
373
374        self.entry.anon_struct_union = False
375
376        param = KernRe(r'[\[\)].*').sub('', param, count=1)
377
378        #
379        # Look at various "anonymous type" cases.
380        #
381        if dtype == '':
382            if param.endswith("..."):
383                named_variadic = len(param) > 3
384                if named_variadic: # there is a name provided, use that
385                    #
386                    # If the user documented the parameter using the
387                    # ``@name...:`` form, the description is stored in
388                    # parameterdescs under the unstripped key.  Migrate
389                    # it to the stripped key so the user's text is not
390                    # silently dropped during output, and so the new
391                    # excess-parameter check in check_sections() does
392                    # not flag the unstripped key as orphaned.
393                    #
394                    orig = self.entry.parameterdescs.pop(param, None)
395                    param = param[:-3]
396                    if orig is not None and \
397                       not self.entry.parameterdescs.get(param):
398                        self.entry.parameterdescs[param] = orig
399                if not self.entry.parameterdescs.get(param):
400                    #
401                    # For a named variadic (e.g. ``args...``), emit the
402                    # standard "not described" warning before auto-filling
403                    # so a missing or mistyped ``@<name>:`` doc tag does
404                    # not go undetected.  The bare ``...`` form has no
405                    # natural name for the user to document and so always
406                    # gets the auto-generated text.
407                    #
408                    if named_variadic and decl_type == 'function':
409                        self.emit_msg(ln,
410                                      f"function parameter '{param}' "
411                                      f"not described in "
412                                      f"'{declaration_name}'")
413                    self.entry.parameterdescs[param] = "variable arguments"
414
415            elif (not param) or param == "void":
416                param = "void"
417                self.entry.parameterdescs[param] = "no arguments"
418
419            elif param in ["struct", "union"]:
420                # Handle unnamed (anonymous) union or struct
421                dtype = param
422                param = "{unnamed_" + param + "}"
423                self.entry.parameterdescs[param] = "anonymous\n"
424                self.entry.anon_struct_union = True
425
426        # Warn if parameter has no description
427        # (but ignore ones starting with # as these are not parameters
428        # but inline preprocessor statements)
429        if param not in self.entry.parameterdescs and not param.startswith("#"):
430            self.entry.parameterdescs[param] = self.undescribed
431
432            if "." not in param:
433                if decl_type == 'function':
434                    dname = f"{decl_type} parameter"
435                else:
436                    dname = f"{decl_type} member"
437
438                self.emit_msg(ln,
439                              f"{dname} '{param}' not described in '{declaration_name}'")
440
441        # Strip spaces from param so that it is one continuous string on
442        # parameterlist. This fixes a problem where check_sections()
443        # cannot find a parameter like "addr[6 + 2]" because it actually
444        # appears as "addr[6", "+", "2]" on the parameter list.
445        # However, it's better to maintain the param string unchanged for
446        # output, so just weaken the string compare in check_sections()
447        # to ignore "[blah" in a parameter string.
448
449        self.entry.parameterlist.append(param)
450        org_arg = KernRe(r'\s\s+').sub(' ', org_arg)
451        self.entry.parametertypes[param] = org_arg
452
453
454    def create_parameter_list(self, ln, decl_type, args,
455                              splitter, declaration_name):
456        """
457        Creates a list of parameters, storing them at self.entry.
458        """
459
460        # temporarily replace all commas inside function pointer definition
461        arg_expr = KernRe(r'(\([^\),]+),')
462        while arg_expr.search(args):
463            args = arg_expr.sub(r"\1#", args)
464
465        for arg in args.split(splitter):
466            # Ignore argument attributes
467            arg = KernRe(r'\sPOS0?\s').sub(' ', arg)
468
469            # Replace '[at_least ' with '[static '.  This allows sphinx to parse
470            # array parameter declarations like 'char A[at_least 4]', where
471            # 'at_least' is #defined to 'static' by the kernel headers.
472            arg = arg.replace('[at_least ', '[static ')
473
474            # Strip leading/trailing spaces
475            arg = arg.strip()
476            arg = KernRe(r'\s+').sub(' ', arg, count=1)
477
478            if arg.startswith('#'):
479                # Treat preprocessor directive as a typeless variable just to fill
480                # corresponding data structures "correctly". Catch it later in
481                # output_* subs.
482
483                # Treat preprocessor directive as a typeless variable
484                self.push_parameter(ln, decl_type, arg, "",
485                                    "", declaration_name)
486            #
487            # The pointer-to-function case.
488            #
489            elif KernRe(r'\(.+\)\s*\(').search(arg):
490                arg = arg.replace('#', ',')
491                r = KernRe(r'[^\(]+\(\*?\s*'  # Everything up to "(*"
492                           r'([\w\[\].]*)'    # Capture the name and possible [array]
493                           r'\s*\)')	      # Make sure the trailing ")" is there
494                if r.match(arg):
495                    param = r.group(1)
496                else:
497                    self.emit_msg(ln, f"Invalid param: {arg}")
498                    param = arg
499                dtype = arg.replace(param, '')
500                self.push_parameter(ln, decl_type, param, dtype, arg, declaration_name)
501            #
502            # The array-of-pointers case.  Dig the parameter name out from the middle
503            # of the declaration.
504            #
505            elif KernRe(r'\(.+\)\s*\[').search(arg):
506                r = KernRe(r'[^\(]+\(\s*\*\s*'		# Up to "(" and maybe "*"
507                           r'([\w.]*?)'			# The actual pointer name
508                           r'\s*(\[\s*\w+\s*\]\s*)*\)') # The [array portion]
509                if r.match(arg):
510                    param = r.group(1)
511                else:
512                    self.emit_msg(ln, f"Invalid param: {arg}")
513                    param = arg
514                dtype = arg.replace(param, '')
515                self.push_parameter(ln, decl_type, param, dtype, arg, declaration_name)
516            elif arg:
517                #
518                # Clean up extraneous spaces and split the string at commas; the first
519                # element of the resulting list will also include the type information.
520                #
521                arg = KernRe(r'\s*:\s*').sub(":", arg)
522                arg = KernRe(r'\s*\[').sub('[', arg)
523                args = KernRe(r'\s*,\s*').split(arg)
524                args[0] = re.sub(r'(\*+)\s*', r' \1', args[0])
525                #
526                # args[0] has a string of "type a".  If "a" includes an [array]
527                # declaration, we want to not be fooled by any white space inside
528                # the brackets, so detect and handle that case specially.
529                #
530                r = KernRe(r'^([^[\]]*\s+)(.*)$')
531                if r.match(args[0]):
532                    args[0] = r.group(2)
533                    dtype = r.group(1)
534                else:
535                    # No space in args[0]; this seems wrong but preserves previous behavior
536                    dtype = ''
537
538                bitfield_re = KernRe(r'(.*?):(\w+)')
539                for param in args:
540                    #
541                    # For pointers, shift the star(s) from the variable name to the
542                    # type declaration.
543                    #
544                    r = KernRe(r'^(\*+)\s*(.*)')
545                    if r.match(param):
546                        self.push_parameter(ln, decl_type, r.group(2),
547                                            f"{dtype} {r.group(1)}",
548                                            arg, declaration_name)
549                    #
550                    # Perform a similar shift for bitfields.
551                    #
552                    elif bitfield_re.search(param):
553                        if dtype != "":  # Skip unnamed bit-fields
554                            self.push_parameter(ln, decl_type, bitfield_re.group(1),
555                                                f"{dtype}:{bitfield_re.group(2)}",
556                                                arg, declaration_name)
557                    else:
558                        self.push_parameter(ln, decl_type, param, dtype,
559                                            arg, declaration_name)
560
561    def check_sections(self, ln, decl_name, decl_type):
562        """
563        Check for errors inside sections, emitting warnings if not found
564        parameters are described.
565        """
566        for section in self.entry.sections:
567            if section not in self.entry.parameterlist and \
568               not known_sections.search(section):
569                if decl_type == 'function':
570                    dname = f"{decl_type} parameter"
571                else:
572                    dname = f"{decl_type} member"
573                self.emit_msg(ln,
574                              f"Excess {dname} '{section}' description in '{decl_name}'")
575
576        #
577        # Check that documented parameter names (from doc comments, including
578        # inline ``/** @member: */`` tags) actually match real members in
579        # the declaration.  This catches mismatched or stale kernel-doc
580        # member tags that don't correspond to any actual struct/union
581        # member or function parameter.
582        #
583        for param_name, desc in self.entry.parameterdescs.items():
584            # Skip auto-generated entries from push_parameter()
585            if desc == self.undescribed:
586                continue
587            if desc in ("no arguments", "anonymous\n", "variable arguments"):
588                continue
589            if param_name.startswith("{unnamed_"):
590                continue
591            if param_name in self.entry.parameterlist:
592                continue
593
594            if decl_type == 'function':
595                dname = f"{decl_type} parameter"
596            else:
597                dname = f"{decl_type} member"
598            self.emit_msg(ln,
599                          f"Excess {dname} '{param_name}' description in '{decl_name}'")
600
601    def check_return_section(self, ln, declaration_name, return_type):
602        """
603        If the function doesn't return void, warns about the lack of a
604        return description.
605        """
606
607        if not self.config.wreturn:
608            return
609
610        # Ignore an empty return type (It's a macro)
611        # Ignore functions with a "void" return type (but not "void *")
612        if not return_type or KernRe(r'void\s*\w*\s*$').search(return_type):
613            return
614
615        if not self.entry.sections.get("Return", None):
616            self.emit_msg(ln,
617                          f"No description found for return value of '{declaration_name}'")
618
619    def split_struct_proto(self, proto):
620        """
621        Split apart a structure prototype; returns (struct|union, name,
622        members) or ``None``.
623        """
624
625        type_pattern = r'(struct|union)'
626        qualifiers = [
627            "__attribute__",
628            "__packed",
629            "__aligned",
630            "____cacheline_aligned_in_smp",
631            "____cacheline_aligned",
632        ]
633        definition_body = r'\{(.*)\}\s*' + "(?:" + '|'.join(qualifiers) + ")?"
634
635        r = KernRe(type_pattern + r'\s+(\w+)\s*' + definition_body)
636        if r.search(proto):
637            return (r.group(1), r.group(2), r.group(3))
638        else:
639            r = KernRe(r'typedef\s+' + type_pattern + r'\s*' + definition_body + r'\s*(\w+)\s*;')
640            if r.search(proto):
641                return (r.group(1), r.group(3), r.group(2))
642        return None
643
644    def rewrite_struct_members(self, members):
645        """
646        Process ``struct``/``union`` members from the most deeply nested
647        outward.
648
649        Rewrite the members of a ``struct`` or ``union`` for easier formatting
650        later on. Among other things, this function will turn a member like::
651
652          struct { inner_members; } foo;
653
654        into::
655
656          struct foo; inner_members;
657        """
658
659        #
660        # The trick is in the ``^{`` below - it prevents a match of an outer
661        # ``struct``/``union`` until the inner one has been munged
662        # (removing the ``{`` in the process).
663        #
664        struct_members = KernRe(r'(struct|union)'   # 0: declaration type
665                                r'([^\{\};]+)' 	    # 1: possible name
666                                r'(\{)'
667                                r'([^\{\}]*)'       # 3: Contents of declaration
668                                r'(\})'
669                                r'([^\{\};]*)(;)')  # 5: Remaining stuff after declaration
670        tuples = struct_members.findall(members)
671        while tuples:
672            for t in tuples:
673                newmember = ""
674                oldmember = "".join(t) # Reconstruct the original formatting
675                dtype, name, lbr, content, rbr, rest, semi = t
676                #
677                # Pass through each field name, normalizing the form and formatting.
678                #
679                for s_id in rest.split(','):
680                    s_id = s_id.strip()
681                    newmember += f"{dtype} {s_id}; "
682                    #
683                    # Remove bitfield/array/pointer info, getting the bare name.
684                    #
685                    s_id = KernRe(r'[:\[].*').sub('', s_id)
686                    s_id = KernRe(r'^\s*\**(\S+)\s*').sub(r'\1', s_id)
687                    #
688                    # Pass through the members of this inner structure/union.
689                    #
690                    for arg in content.split(';'):
691                        arg = arg.strip()
692                        #
693                        # Look for (type)(*name)(args) - pointer to function
694                        #
695                        r = KernRe(r'^([^\(]+\(\*?\s*)([\w.]*)(\s*\).*)')
696                        if r.match(arg):
697                            dtype, name, extra = r.group(1), r.group(2), r.group(3)
698                            # Pointer-to-function
699                            if not s_id:
700                                # Anonymous struct/union
701                                newmember += f"{dtype}{name}{extra}; "
702                            else:
703                                newmember += f"{dtype}{s_id}.{name}{extra}; "
704                        #
705                        # Otherwise a non-function member.
706                        #
707                        else:
708                            #
709                            # Remove bitmap and array portions and spaces around commas
710                            #
711                            arg = KernRe(r':\s*\d+\s*').sub('', arg)
712                            arg = KernRe(r'\[.*\]').sub('', arg)
713                            arg = KernRe(r'\s*,\s*').sub(',', arg)
714                            #
715                            # Look for a normal decl - "type name[,name...]"
716                            #
717                            r = KernRe(r'(.*)\s+([\S+,]+)')
718                            if r.search(arg):
719                                for name in r.group(2).split(','):
720                                    name = KernRe(r'^\s*\**(\S+)\s*').sub(r'\1', name)
721                                    if not s_id:
722                                        # Anonymous struct/union
723                                        newmember += f"{r.group(1)} {name}; "
724                                    else:
725                                        newmember += f"{r.group(1)} {s_id}.{name}; "
726                            else:
727                                newmember += f"{arg}; "
728                #
729                # At the end of the s_id loop, replace the original declaration with
730                # the munged version.
731                #
732                members = members.replace(oldmember, newmember)
733            #
734            # End of the tuple loop - search again and see if there are outer members
735            # that now turn up.
736            #
737            tuples = struct_members.findall(members)
738        return members
739
740    def format_struct_decl(self, declaration):
741        """
742        Format the ``struct`` declaration into a standard form for inclusion
743        in the resulting docs.
744        """
745
746        #
747        # Insert newlines, get rid of extra spaces.
748        #
749        declaration = KernRe(r'([\{;])').sub(r'\1\n', declaration)
750        declaration = KernRe(r'\}\s+;').sub('};', declaration)
751        #
752        # Format inline enums with each member on its own line.
753        #
754        r = KernRe(r'(enum\s+\{[^\}]+),([^\n])')
755        while r.search(declaration):
756            declaration = r.sub(r'\1,\n\2', declaration)
757        #
758        # Now go through and supply the right number of tabs
759        # for each line.
760        #
761        def_args = declaration.split('\n')
762        level = 1
763        declaration = ""
764        for clause in def_args:
765            clause = KernRe(r'\s+').sub(' ', clause.strip(), count=1)
766            if clause:
767                if '}' in clause and level > 1:
768                    level -= 1
769                if not clause.startswith('#'):
770                    declaration += "\t" * level
771                declaration += "\t" + clause + "\n"
772                if "{" in clause and "}" not in clause:
773                    level += 1
774        return declaration
775
776
777    def dump_struct(self, ln, proto, source):
778        """
779        Store an entry for a ``struct`` or ``union``
780        """
781        #
782        # Do the basic parse to get the pieces of the declaration.
783        #
784        source = source
785        proto = trim_private_members(proto)
786        struct_parts = self.split_struct_proto(proto)
787        if not struct_parts:
788            self.emit_msg(ln, f"{proto} error: Cannot parse struct or union!")
789            return
790        decl_type, declaration_name, members = struct_parts
791
792        if self.entry.identifier != declaration_name:
793            self.emit_msg(ln, f"expecting prototype for {decl_type} {self.entry.identifier}. "
794                          f"Prototype was for {decl_type} {declaration_name} instead\n")
795            return
796        #
797        # Go through the list of members applying all of our transformations.
798        #
799        members = self.xforms.apply("struct", members)
800
801        #
802        # Deal with embedded struct and union members, and drop enums entirely.
803        #
804        declaration = members
805        members = self.rewrite_struct_members(members)
806        members = re.sub(r'(\{[^\{\}]*\})', '', members)
807        #
808        # Output the result and we are done.
809        #
810        self.create_parameter_list(ln, decl_type, members, ';',
811                                   declaration_name)
812        self.check_sections(ln, declaration_name, decl_type)
813        self.output_declaration(decl_type, declaration_name,
814                                source=source,
815                                definition=self.format_struct_decl(declaration),
816                                purpose=self.entry.declaration_purpose)
817
818    def dump_enum(self, ln, proto, source):
819        """
820        Store an ``enum`` inside self.entries array.
821        """
822        #
823        # Strip preprocessor directives.  Note that this depends on the
824        # trailing semicolon we added in process_proto_type().
825        #
826        source = source
827        proto = trim_private_members(proto)
828        proto = KernRe(r'#\s*((define|ifdef|if)\s+|endif)[^;]*;', flags=re.S).sub('', proto)
829        #
830        # Parse out the name and members of the enum.  Typedef form first.
831        #
832        r = KernRe(r'typedef\s+enum\s*\{(.*)\}\s*(\w*)\s*;')
833        if r.search(proto):
834            declaration_name = r.group(2)
835            members = r.group(1)
836        #
837        # Failing that, look for a straight enum
838        #
839        else:
840            r = KernRe(r'enum\s+(\w*)\s*\{(.*)\}')
841            if r.match(proto):
842                declaration_name = r.group(1)
843                members = r.group(2)
844        #
845        # OK, this isn't going to work.
846        #
847            else:
848                self.emit_msg(ln, f"{proto}: error: Cannot parse enum!")
849                return
850        #
851        # Make sure we found what we were expecting.
852        #
853        if self.entry.identifier != declaration_name:
854            if self.entry.identifier == "":
855                self.emit_msg(ln,
856                              f"{proto}: wrong kernel-doc identifier on prototype")
857            else:
858                self.emit_msg(ln,
859                              f"expecting prototype for enum {self.entry.identifier}. "
860                              f"Prototype was for enum {declaration_name} instead")
861            return
862
863        if not declaration_name:
864            declaration_name = "(anonymous)"
865        #
866        # Parse out the name of each enum member, and verify that we
867        # have a description for it.
868        #
869        member_set = set()
870        members = KernRe(r'\([^;)]*\)').sub('', members)
871        for arg in members.split(','):
872            arg = KernRe(r'^\s*(\w+).*').sub(r'\1', arg)
873            if not arg.strip():
874                continue
875
876            self.entry.parameterlist.append(arg)
877            if arg not in self.entry.parameterdescs:
878                self.entry.parameterdescs[arg] = self.undescribed
879                self.emit_msg(ln,
880                              f"Enum value '{arg}' not described in enum '{declaration_name}'")
881            member_set.add(arg)
882        #
883        # Ensure that every described member actually exists in the enum.
884        #
885        for k in self.entry.parameterdescs:
886            if k not in member_set:
887                self.emit_msg(ln,
888                              f"Excess enum value '@{k}' description in '{declaration_name}'")
889
890        self.output_declaration('enum', declaration_name,
891                                source=source,
892                                purpose=self.entry.declaration_purpose)
893
894    def dump_var(self, ln, proto, source):
895        """
896        Store variables that are part of kAPI.
897        """
898        VAR_ATTRIBS = [
899            "extern",
900            "const",
901        ]
902        OPTIONAL_VAR_ATTR = r"^(?:\b(?:" +"|".join(VAR_ATTRIBS) +r")\b\s*)*"
903
904        #
905        # Store the full prototype before modifying it
906        #
907        source = source
908        full_proto = proto
909        declaration_name = None
910
911        #
912        # Handle macro definitions
913        #
914        macro_prefixes = [
915            KernRe(r"DEFINE_[\w_]+\s*\(([\w_]+)\)"),
916        ]
917
918        for r in macro_prefixes:
919            match = r.search(proto)
920            if match:
921                declaration_name = match.group(1)
922                break
923
924        #
925        # Drop comments and macros to have a pure C prototype
926        #
927        if not declaration_name:
928            proto = self.xforms.apply("var", proto)
929
930        proto = proto.rstrip()
931
932        #
933        # Variable name is at the end of the declaration
934        #
935
936        default_val = None
937
938        r= KernRe(OPTIONAL_VAR_ATTR + r"\s*[\w_\s]*\s+(?:\*+)?([\w_]+)\s*[\d\]\[]*\s*(=.*)?")
939        if r.match(proto):
940            if not declaration_name:
941                declaration_name = r.group(1)
942
943            default_val = r.group(2)
944        else:
945            r= KernRe(OPTIONAL_VAR_ATTR + r"(?:[\w_\s]*)?\s+(?:\*+)?(?:[\w_]+)\s*[\d\]\[]*\s*(=.*)?")
946
947            if r.match(proto):
948                default_val = r.group(1)
949        if not declaration_name:
950           self.emit_msg(ln,f"{proto}: can't parse variable")
951           return
952
953        if default_val:
954            default_val = default_val.lstrip("=").strip()
955
956        self.output_declaration("var", declaration_name,
957                                source=source,
958                                full_proto=full_proto,
959                                default_val=default_val,
960                                purpose=self.entry.declaration_purpose)
961
962    def dump_declaration(self, ln, prototype, source):
963        """
964        Store a data declaration inside self.entries array.
965        """
966
967        if self.entry.decl_type == "enum":
968            self.dump_enum(ln, prototype, source)
969        elif self.entry.decl_type == "typedef":
970            self.dump_typedef(ln, prototype, source)
971        elif self.entry.decl_type in ["union", "struct"]:
972            self.dump_struct(ln, prototype, source)
973        elif self.entry.decl_type == "var":
974            self.dump_var(ln, prototype, source)
975        else:
976            # This would be a bug
977            self.emit_message(ln, f'Unknown declaration type: {self.entry.decl_type}')
978
979    def dump_function(self, ln, prototype, source):
980        """
981        Store a function or function macro inside self.entries array.
982        """
983
984        source = source
985        found = func_macro = False
986        return_type = ''
987        decl_type = 'function'
988
989        #
990        # If we have a macro, remove the "#define" at the front.
991        #
992        new_proto = KernRe(r"^#\s*define\s+").sub("", prototype)
993        if new_proto != prototype:
994            prototype = new_proto
995            #
996            # Dispense with the simple "#define A B" case here; the key
997            # is the space after the name of the symbol being defined.
998            # NOTE that the seemingly misnamed "func_macro" indicates a
999            # macro *without* arguments.
1000            #
1001            r = KernRe(r'^(\w+)\s+')
1002            if r.search(prototype):
1003                return_type = ''
1004                declaration_name = r.group(1)
1005                func_macro = True
1006                found = True
1007        else:
1008            #
1009            # Apply the initial transformations.
1010            #
1011            prototype = self.xforms.apply("func", prototype)
1012
1013        # Yes, this truly is vile.  We are looking for:
1014        # 1. Return type (may be nothing if we're looking at a macro)
1015        # 2. Function name
1016        # 3. Function parameters.
1017        #
1018        # All the while we have to watch out for function pointer parameters
1019        # (which IIRC is what the two sections are for), C types (these
1020        # regexps don't even start to express all the possibilities), and
1021        # so on.
1022        #
1023        # If you mess with these regexps, it's a good idea to check that
1024        # the following functions' documentation still comes out right:
1025        # - parport_register_device (function pointer parameters)
1026        # - atomic_set (macro)
1027        # - pci_match_device, __copy_to_user (long return type)
1028
1029        name = r'\w+'
1030        type1 = r'(?:[\w\s]+)?'
1031        type2 = r'(?:[\w\s]+\*+)+'
1032        #
1033        # Attempt to match first on (args) with no internal parentheses; this
1034        # lets us easily filter out __acquires() and other post-args stuff.  If
1035        # that fails, just grab the rest of the line to the last closing
1036        # parenthesis.
1037        #
1038        proto_args = r'\(([^\(]*|.*)\)'
1039        #
1040        # (Except for the simple macro case) attempt to split up the prototype
1041        # in the various ways we understand.
1042        #
1043        if not found:
1044            patterns = [
1045                rf'^()({name})\s*{proto_args}',
1046                rf'^({type1})\s+({name})\s*{proto_args}',
1047                rf'^({type2})\s*({name})\s*{proto_args}',
1048            ]
1049
1050            for p in patterns:
1051                r = KernRe(p)
1052                if r.match(prototype):
1053                    return_type = r.group(1)
1054                    declaration_name = r.group(2)
1055                    args = r.group(3)
1056                    self.create_parameter_list(ln, decl_type, args, ',',
1057                                               declaration_name)
1058                    found = True
1059                    break
1060        #
1061        # Parsing done; make sure that things are as we expect.
1062        #
1063        if not found:
1064            self.emit_msg(ln,
1065                          f"cannot understand function prototype: '{prototype}'")
1066            return
1067        if self.entry.identifier != declaration_name:
1068            self.emit_msg(ln, f"expecting prototype for {self.entry.identifier}(). "
1069                          f"Prototype was for {declaration_name}() instead")
1070            return
1071        self.check_sections(ln, declaration_name, "function")
1072        self.check_return_section(ln, declaration_name, return_type)
1073        #
1074        # Store the result.
1075        #
1076        self.output_declaration(decl_type, declaration_name,
1077                                source=source,
1078                                typedef=('typedef' in return_type),
1079                                functiontype=return_type,
1080                                purpose=self.entry.declaration_purpose,
1081                                func_macro=func_macro)
1082
1083
1084    def dump_typedef(self, ln, proto, source):
1085        """
1086        Store a ``typedef`` inside self.entries array.
1087        """
1088        #
1089        # We start by looking for function typedefs.
1090        #
1091        typedef_type = r'typedef((?:\s+[\w*]+\b){0,7}\s+(?:\w+\b|\*+))\s*'
1092        typedef_ident = r'\*?\s*(\w\S+)\s*'
1093        typedef_args = r'\s*\((.*)\);'
1094
1095        source = source
1096
1097        typedef1 = KernRe(typedef_type + r'\(' + typedef_ident + r'\)' + typedef_args)
1098        typedef2 = KernRe(typedef_type + typedef_ident + typedef_args)
1099
1100        # Parse function typedef prototypes
1101        for r in [typedef1, typedef2]:
1102            if not r.match(proto):
1103                continue
1104
1105            return_type = r.group(1).strip()
1106            declaration_name = r.group(2)
1107            args = r.group(3)
1108
1109            if self.entry.identifier != declaration_name:
1110                self.emit_msg(ln,
1111                              f"expecting prototype for typedef {self.entry.identifier}. Prototype was for typedef {declaration_name} instead\n")
1112                return
1113
1114            self.create_parameter_list(ln, 'function', args, ',', declaration_name)
1115
1116            self.output_declaration('function', declaration_name,
1117                                    source=source,
1118                                    typedef=True,
1119                                    functiontype=return_type,
1120                                    purpose=self.entry.declaration_purpose)
1121            return
1122        #
1123        # Not a function, try to parse a simple typedef.
1124        #
1125        r = KernRe(r'typedef.*\s+(\w+)\s*;')
1126        if r.match(proto):
1127            declaration_name = r.group(1)
1128
1129            if self.entry.identifier != declaration_name:
1130                self.emit_msg(ln,
1131                              f"expecting prototype for typedef {self.entry.identifier}. Prototype was for typedef {declaration_name} instead\n")
1132                return
1133
1134            self.output_declaration('typedef', declaration_name,
1135                                    source=source,
1136                                    purpose=self.entry.declaration_purpose)
1137            return
1138
1139        self.emit_msg(ln, "error: Cannot parse typedef!")
1140
1141    @staticmethod
1142    def process_export(function_set, line):
1143        """
1144        process ``EXPORT_SYMBOL*`` tags
1145
1146        This method doesn't use any variable from the class, so declare it
1147        with a staticmethod decorator.
1148        """
1149
1150        # We support documenting some exported symbols with different
1151        # names.  A horrible hack.
1152        suffixes = [ '_noprof' ]
1153
1154        # Note: it accepts only one EXPORT_SYMBOL* per line, as having
1155        # multiple export lines would violate Kernel coding style.
1156
1157        if export_symbol.search(line):
1158            symbol = export_symbol.group(2)
1159        elif export_symbol_ns.search(line):
1160            symbol = export_symbol_ns.group(2)
1161        else:
1162            return False
1163        #
1164        # Found an export, trim out any special suffixes
1165        #
1166        for suffix in suffixes:
1167            # Be backward compatible with Python < 3.9
1168            if symbol.endswith(suffix):
1169                symbol = symbol[:-len(suffix)]
1170        function_set.add(symbol)
1171        return True
1172
1173    def process_normal(self, ln, line, source):
1174        """
1175        STATE_NORMAL: looking for the ``/**`` to begin everything.
1176        """
1177
1178        if not doc_start.match(line):
1179            return
1180
1181        # start a new entry
1182        self.reset_state(ln)
1183
1184        # next line is always the function name
1185        self.state = state.NAME
1186
1187    def process_name(self, ln, line, source):
1188        """
1189        STATE_NAME: Looking for the "name - description" line
1190        """
1191        #
1192        # Check for a DOC: block and handle them specially.
1193        #
1194        if doc_block.search(line):
1195
1196            if not doc_block.group(1):
1197                self.entry.begin_section(ln, "Introduction")
1198            else:
1199                self.entry.begin_section(ln, doc_block.group(1))
1200
1201            self.entry.identifier = self.entry.section
1202            self.state = state.DOCBLOCK
1203        #
1204        # Otherwise we're looking for a normal kerneldoc declaration line.
1205        #
1206        elif doc_decl.search(line):
1207            self.entry.identifier = doc_decl.group(1)
1208
1209            # Test for data declaration
1210            if doc_begin_data.search(line):
1211                self.entry.decl_type = doc_begin_data.group(1)
1212                self.entry.identifier = doc_begin_data.group(2)
1213            #
1214            # Look for a function description
1215            #
1216            elif doc_begin_func.search(line):
1217                self.entry.identifier = doc_begin_func.group(1)
1218                self.entry.decl_type = "function"
1219            #
1220            # We struck out.
1221            #
1222            else:
1223                self.emit_msg(ln,
1224                              f"This comment starts with '/**', but isn't a kernel-doc comment. Refer to Documentation/doc-guide/kernel-doc.rst\n{line}")
1225                self.state = state.NORMAL
1226                return
1227            #
1228            # OK, set up for a new kerneldoc entry.
1229            #
1230            self.state = state.BODY
1231            self.entry.identifier = self.entry.identifier.strip(" ")
1232            # if there's no @param blocks need to set up default section here
1233            self.entry.begin_section(ln + 1)
1234            #
1235            # Find the description portion, which *should* be there but
1236            # isn't always.
1237            # (We should be able to capture this from the previous parsing - someday)
1238            #
1239            r = KernRe("[-:](.*)")
1240            if r.search(line):
1241                self.entry.declaration_purpose = trim_whitespace(r.group(1))
1242                self.state = state.DECLARATION
1243            else:
1244                self.entry.declaration_purpose = ""
1245
1246            if not self.entry.declaration_purpose and self.config.wshort_desc:
1247                self.emit_msg(ln,
1248                              f"missing initial short description on line:\n{line}")
1249
1250            if not self.entry.identifier and self.entry.decl_type != "enum":
1251                self.emit_msg(ln,
1252                              f"wrong kernel-doc identifier on line:\n{line}")
1253                self.state = state.NORMAL
1254
1255            if self.config.verbose:
1256                self.emit_msg(ln,
1257                              f"Scanning doc for {self.entry.decl_type} {self.entry.identifier}",
1258                                  warning=False)
1259        #
1260        # Failed to find an identifier. Emit a warning
1261        #
1262        else:
1263            self.emit_msg(ln, f"Cannot find identifier on line:\n{line}")
1264
1265    def is_new_section(self, ln, line):
1266        """
1267        Helper function to determine if a new section is being started.
1268        """
1269        if doc_sect.search(line):
1270            self.state = state.BODY
1271            #
1272            # Pick out the name of our new section, tweaking it if need be.
1273            #
1274            newsection = doc_sect.group(1)
1275            if newsection.lower() == 'description':
1276                newsection = 'Description'
1277            elif newsection.lower() == 'context':
1278                newsection = 'Context'
1279                self.state = state.SPECIAL_SECTION
1280            elif newsection.lower() in ["@return", "@returns",
1281                                        "return", "returns"]:
1282                newsection = "Return"
1283                self.state = state.SPECIAL_SECTION
1284            elif newsection[0] == '@':
1285                self.state = state.SPECIAL_SECTION
1286            #
1287            # Initialize the contents, and get the new section going.
1288            #
1289            newcontents = doc_sect.group(2)
1290            if not newcontents:
1291                newcontents = ""
1292            self.dump_section()
1293            self.entry.begin_section(ln, newsection)
1294            self.entry.leading_space = None
1295
1296            self.entry.add_text(newcontents.lstrip())
1297            return True
1298        return False
1299
1300    def is_comment_end(self, ln, line):
1301        """
1302        Helper function to detect (and effect) the end of a kerneldoc comment.
1303        """
1304        if doc_end.search(line):
1305            self.dump_section()
1306
1307            # Look for doc_com + <text> + doc_end:
1308            r = KernRe(r'\s*\*\s*[a-zA-Z_0-9:.]+\*/')
1309            if r.match(line):
1310                self.emit_msg(ln, f"suspicious ending line: {line}")
1311
1312            self.entry.prototype = ""
1313            self.entry.new_start_line = ln + 1
1314
1315            self.state = state.PROTO
1316            return True
1317        return False
1318
1319
1320    def process_decl(self, ln, line, source):
1321        """
1322        STATE_DECLARATION: We've seen the beginning of a declaration.
1323        """
1324        if self.is_new_section(ln, line) or self.is_comment_end(ln, line):
1325            return
1326        #
1327        # Look for anything with the " * " line beginning.
1328        #
1329        if doc_content.search(line):
1330            cont = doc_content.group(1)
1331            #
1332            # A blank line means that we have moved out of the declaration
1333            # part of the comment (without any "special section" parameter
1334            # descriptions).
1335            #
1336            if cont == "":
1337                self.state = state.BODY
1338            #
1339            # Otherwise we have more of the declaration section to soak up.
1340            #
1341            else:
1342                self.entry.declaration_purpose = \
1343                    trim_whitespace(self.entry.declaration_purpose + ' ' + cont)
1344        else:
1345            # Unknown line, ignore
1346            self.emit_msg(ln, f"bad line: {line}")
1347
1348
1349    def process_special(self, ln, line, source):
1350        """
1351        STATE_SPECIAL_SECTION: a section ending with a blank line.
1352        """
1353        #
1354        # If we have hit a blank line (only the " * " marker), then this
1355        # section is done.
1356        #
1357        if KernRe(r"\s*\*\s*$").match(line):
1358            self.entry.begin_section(ln, dump = True)
1359            self.state = state.BODY
1360            return
1361        #
1362        # Not a blank line, look for the other ways to end the section.
1363        #
1364        if self.is_new_section(ln, line) or self.is_comment_end(ln, line):
1365            return
1366        #
1367        # OK, we should have a continuation of the text for this section.
1368        #
1369        if doc_content.search(line):
1370            cont = doc_content.group(1)
1371            #
1372            # If the lines of text after the first in a special section have
1373            # leading white space, we need to trim it out or Sphinx will get
1374            # confused.  For the second line (the None case), see what we
1375            # find there and remember it.
1376            #
1377            if self.entry.leading_space is None:
1378                r = KernRe(r'^(\s+)')
1379                if r.match(cont):
1380                    self.entry.leading_space = len(r.group(1))
1381                else:
1382                    self.entry.leading_space = 0
1383            #
1384            # Otherwise, before trimming any leading chars, be *sure*
1385            # that they are white space.  We should maybe warn if this
1386            # isn't the case.
1387            #
1388            for i in range(0, self.entry.leading_space):
1389                if cont[i] != " ":
1390                    self.entry.leading_space = i
1391                    break
1392            #
1393            # Add the trimmed result to the section and we're done.
1394            #
1395            self.entry.add_text(cont[self.entry.leading_space:])
1396        else:
1397            # Unknown line, ignore
1398            self.emit_msg(ln, f"bad line: {line}")
1399
1400    def process_body(self, ln, line, source):
1401        """
1402        STATE_BODY: the bulk of a kerneldoc comment.
1403        """
1404        if self.is_new_section(ln, line) or self.is_comment_end(ln, line):
1405            return
1406
1407        if doc_content.search(line):
1408            cont = doc_content.group(1)
1409            self.entry.add_text(cont)
1410        else:
1411            # Unknown line, ignore
1412            self.emit_msg(ln, f"bad line: {line}")
1413
1414    def process_inline_name(self, ln, line, source):
1415        """STATE_INLINE_NAME: beginning of docbook comments within a prototype."""
1416
1417        if doc_inline_sect.search(line):
1418            self.entry.begin_section(ln, doc_inline_sect.group(1))
1419            self.entry.add_text(doc_inline_sect.group(2).lstrip())
1420            self.state = state.INLINE_TEXT
1421        elif doc_inline_end.search(line):
1422            self.dump_section()
1423            self.state = state.PROTO
1424        elif doc_content.search(line):
1425            self.emit_msg(ln, f"Incorrect use of kernel-doc format: {line}")
1426            self.state = state.PROTO
1427
1428            #
1429            # Don't let it add partial comments at the code, as breaks the
1430            # logic meant to remove comments from prototypes.
1431            #
1432            self.process_proto_type(ln, "/**\n" + line, source)
1433        # else ... ??
1434
1435    def process_inline_text(self, ln, line, source):
1436        """STATE_INLINE_TEXT: docbook comments within a prototype."""
1437
1438        if doc_inline_end.search(line):
1439            self.dump_section()
1440            self.state = state.PROTO
1441        elif doc_content.search(line):
1442            self.entry.add_text(doc_content.group(1))
1443        # else ... ??
1444
1445    def syscall_munge(self, ln, proto):         # pylint: disable=W0613
1446        """
1447        Handle syscall definitions.
1448        """
1449
1450        is_void = False
1451
1452        # Strip newlines/CR's
1453        proto = re.sub(r'[\r\n]+', ' ', proto)
1454
1455        # Check if it's a SYSCALL_DEFINE0
1456        if 'SYSCALL_DEFINE0' in proto:
1457            is_void = True
1458
1459        # Replace SYSCALL_DEFINE with correct return type & function name
1460        proto = KernRe(r'SYSCALL_DEFINE.*\(').sub('long sys_', proto)
1461
1462        r = KernRe(r'long\s+(sys_.*?),')
1463        if r.search(proto):
1464            proto = KernRe(',').sub('(', proto, count=1)
1465        elif is_void:
1466            proto = KernRe(r'\)').sub('(void)', proto, count=1)
1467
1468        # Now delete all of the odd-numbered commas in the proto
1469        # so that argument types & names don't have a comma between them
1470        count = 0
1471        length = len(proto)
1472
1473        if is_void:
1474            length = 0  # skip the loop if is_void
1475
1476        for ix in range(length):
1477            if proto[ix] == ',':
1478                count += 1
1479                if count % 2 == 1:
1480                    proto = proto[:ix] + ' ' + proto[ix + 1:]
1481
1482        return proto
1483
1484    def tracepoint_munge(self, ln, proto):
1485        """
1486        Handle tracepoint definitions.
1487        """
1488
1489        tracepointname = None
1490        tracepointargs = None
1491
1492        # Match tracepoint name based on different patterns
1493        r = KernRe(r'TRACE_EVENT\((.*?),')
1494        if r.search(proto):
1495            tracepointname = r.group(1)
1496
1497        r = KernRe(r'DEFINE_SINGLE_EVENT\((.*?),')
1498        if r.search(proto):
1499            tracepointname = r.group(1)
1500
1501        r = KernRe(r'DEFINE_EVENT\((.*?),(.*?),')
1502        if r.search(proto):
1503            tracepointname = r.group(2)
1504
1505        if tracepointname:
1506            tracepointname = tracepointname.lstrip()
1507
1508        r = KernRe(r'TP_PROTO\((.*?)\)')
1509        if r.search(proto):
1510            tracepointargs = r.group(1)
1511
1512        if not tracepointname or not tracepointargs:
1513            self.emit_msg(ln,
1514                          f"Unrecognized tracepoint format:\n{proto}\n")
1515        else:
1516            proto = f"static inline void trace_{tracepointname}({tracepointargs})"
1517            self.entry.identifier = f"trace_{self.entry.identifier}"
1518
1519        return proto
1520
1521    def process_proto_function(self, ln, line, source):
1522        """Ancillary routine to process a function prototype."""
1523
1524        # strip C99-style comments to end of line
1525        line = KernRe(r"//.*$", re.S).sub('', line)
1526        #
1527        # Soak up the line's worth of prototype text, stopping at { or ; if present.
1528        #
1529        if KernRe(r'\s*#\s*define').match(line):
1530            self.entry.prototype = line
1531        elif not line.startswith('#'):   # skip other preprocessor stuff
1532            r = KernRe(r'([^\{]*)')
1533            if r.match(line):
1534                self.entry.prototype += r.group(1) + " "
1535        #
1536        # If we now have the whole prototype, clean it up and declare victory.
1537        #
1538        if '{' in line or ';' in line or KernRe(r'\s*#\s*define').match(line):
1539            # strip comments and surrounding spaces
1540            self.entry.prototype = KernRe(r'/\*.*\*/').sub('', self.entry.prototype).strip()
1541            #
1542            # Handle self.entry.prototypes for function pointers like:
1543            #       int (*pcs_config)(struct foo)
1544            # by turning it into
1545            #	    int pcs_config(struct foo)
1546            #
1547            r = KernRe(r'^(\S+\s+)\(\s*\*(\S+)\)')
1548            self.entry.prototype = r.sub(r'\1\2', self.entry.prototype)
1549            #
1550            # Handle special declaration syntaxes
1551            #
1552            if 'SYSCALL_DEFINE' in self.entry.prototype:
1553                self.entry.prototype = self.syscall_munge(ln,
1554                                                          self.entry.prototype)
1555            else:
1556                r = KernRe(r'TRACE_EVENT|DEFINE_EVENT|DEFINE_SINGLE_EVENT')
1557                if r.search(self.entry.prototype):
1558                    self.entry.prototype = self.tracepoint_munge(ln,
1559                                                                 self.entry.prototype)
1560            #
1561            # ... and we're done
1562            #
1563            self.dump_function(ln, self.entry.prototype, source)
1564            self.reset_state(ln)
1565
1566    def process_proto_type(self, ln, line, source):
1567        """
1568        Ancillary routine to process a type.
1569        """
1570
1571        # Strip C99-style comments and surrounding whitespace
1572        line = KernRe(r"//.*$", re.S).sub('', line).strip()
1573        if not line:
1574            return # nothing to see here
1575
1576        # To distinguish preprocessor directive from regular declaration later.
1577        if line.startswith('#'):
1578            line += ";"
1579        #
1580        # Split the declaration on any of { } or ;, and accumulate pieces
1581        # until we hit a semicolon while not inside {brackets}
1582        #
1583        r = KernRe(r'(.*?)([{};])')
1584        for chunk in r.split(line):
1585            if chunk:  # Ignore empty matches
1586                self.entry.prototype += chunk
1587                #
1588                # This cries out for a match statement ... someday after we can
1589                # drop Python 3.9 ...
1590                #
1591                if chunk == '{':
1592                    self.entry.brcount += 1
1593                elif chunk == '}':
1594                    self.entry.brcount -= 1
1595                elif chunk == ';' and self.entry.brcount <= 0:
1596                    self.dump_declaration(ln, self.entry.prototype, source)
1597                    self.reset_state(ln)
1598                    return
1599        #
1600        # We hit the end of the line while still in the declaration; put
1601        # in a space to represent the newline.
1602        #
1603        self.entry.prototype += ' '
1604
1605    def process_proto(self, ln, line, source):
1606        """STATE_PROTO: reading a function/whatever prototype."""
1607
1608        if doc_inline_oneline.search(line):
1609            self.entry.begin_section(ln, doc_inline_oneline.group(1))
1610            self.entry.add_text(doc_inline_oneline.group(2))
1611            self.dump_section()
1612
1613        elif doc_inline_start.search(line):
1614            self.state = state.INLINE_NAME
1615
1616        elif self.entry.decl_type == 'function':
1617            self.process_proto_function(ln, line, source)
1618
1619        else:
1620            self.process_proto_type(ln, line, source)
1621
1622    def process_docblock(self, ln, line, source):
1623        """STATE_DOCBLOCK: within a ``DOC:`` block."""
1624
1625        if doc_end.search(line):
1626            self.dump_section()
1627            self.output_declaration("doc", self.entry.identifier,
1628                                    source=source)
1629            self.reset_state(ln)
1630
1631        elif doc_content.search(line):
1632            self.entry.add_text(doc_content.group(1))
1633
1634    def parse_export(self):
1635        """
1636        Parses ``EXPORT_SYMBOL*`` macros from a single Kernel source file.
1637        """
1638
1639        export_table = set()
1640
1641        try:
1642            with open(self.fname, "r", encoding="utf8",
1643                      errors="backslashreplace") as fp:
1644
1645                for line in fp:
1646                    self.process_export(export_table, line)
1647
1648        except IOError:
1649            return None
1650
1651        return export_table
1652
1653    #: The state/action table telling us which function to invoke in each state.
1654    state_actions = {
1655        state.NORMAL:			process_normal,
1656        state.NAME:			process_name,
1657        state.BODY:			process_body,
1658        state.DECLARATION:		process_decl,
1659        state.SPECIAL_SECTION:		process_special,
1660        state.INLINE_NAME:		process_inline_name,
1661        state.INLINE_TEXT:		process_inline_text,
1662        state.PROTO:			process_proto,
1663        state.DOCBLOCK:			process_docblock,
1664        }
1665
1666    def parse_kdoc(self):
1667        """
1668        Open and process each line of a C source file.
1669        The parsing is controlled via a state machine, and the line is passed
1670        to a different process function depending on the state. The process
1671        function may update the state as needed.
1672
1673        Besides parsing kernel-doc tags, it also parses export symbols.
1674        """
1675
1676        prev = ""
1677        prev_ln = None
1678        export_table = set()
1679        self.state = state.NORMAL
1680        source = ""
1681
1682        try:
1683            with open(self.fname, "r", encoding="utf8",
1684                      errors="backslashreplace") as fp:
1685                for ln, line in enumerate(fp):
1686
1687                    line = line.expandtabs().strip("\n")
1688
1689                    # Group continuation lines on prototypes
1690                    if self.state == state.PROTO:
1691                        if line.endswith("\\"):
1692                            prev += line.rstrip("\\")
1693                            if not prev_ln:
1694                                prev_ln = ln
1695                            continue
1696
1697                        if prev:
1698                            ln = prev_ln
1699                            line = prev + line
1700                            prev = ""
1701                            prev_ln = None
1702
1703                    self.config.log.debug("%d %s: %s",
1704                                          ln, state.name[self.state],
1705                                          line)
1706
1707                    if self.store_src:
1708                        if source and self.state == state.NORMAL:
1709                            source = ""
1710                        elif self.state != state.NORMAL:
1711                            source += line + "\n"
1712
1713                    # This is an optimization over the original script.
1714                    # There, when export_file was used for the same file,
1715                    # it was read twice. Here, we use the already-existing
1716                    # loop to parse exported symbols as well.
1717                    #
1718                    if (self.state != state.NORMAL) or \
1719                       not self.process_export(export_table, line):
1720                        prev_state = self.state
1721                        # Hand this line to the appropriate state handler
1722                        self.state_actions[self.state](self, ln, line, source)
1723                        if prev_state == state.NORMAL and self.state != state.NORMAL:
1724                            source += line + "\n"
1725
1726            self.emit_unused_warnings()
1727
1728        except OSError:
1729            self.config.log.error(f"Error: Cannot open file {self.fname}")
1730
1731        return export_table, self.entries
1732