When reporting some complex trailing statements we report only the
starting line of the error, that tends to imply the shown line is in error
and confuse the reader.  As we do know where the actual error is report
that line too with an appropriate gap marker where applicable.

    #ERROR: trailing statements should be on next line
    #1: FILE: Z202.c:1:
    +       for (pbh = page_buffers(bh->b_page); pbh != bh;
    +               pbh = pbh->b_this_page, key++);
    #ERROR: trailing statements should be on next line
    #4: FILE: Z202.c:4:
    +       for (pbh = page_buffers(bh->b_page);
    [...]
    +               pbh = pbh->b_this_page, key++);
Signed-off-by: NAndy Whitcroft <apw@shadowen.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

When we want to confirm an export is directly after its definition we need
to allow for DEFINE_ style macros.  Add these to the execeptions.
Refactor the exceptions.
Signed-off-by: NAndy Whitcroft <apw@shadowen.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

We are missing 'simple' values which include square brackets.  Refactor to
ensure we handle nesting correctly and detect these simple forms.
Signed-off-by: NAndy Whitcroft <apw@shadowen.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

When we hit and #else or #elif we know we are meeting an alternative piece
of code.  All bets are off on indent if we did not see the open of the
control so stop checking.
Signed-off-by: NAndy Whitcroft <apw@shadowen.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Within the type checker we have a number of common kernel types which must
be implemented as typedefs.  Pull those out so that we can use the same
expressions to trigger exclusions.
Signed-off-by: NAndy Whitcroft <apw@shadowen.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

We are likely going to have 24 bit types.  Expand the type matcher to
match any size.
Signed-off-by: NAndy Whitcroft <apw@shadowen.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

We often see macros which define structure members, these are not complex
and necessarily do not have braces or brackets.  For example:

    #define _PLIST_HEAD_INIT(head)                      \
        .prio_list = LIST_HEAD_INIT((head).prio_list),  \
        .node_list = LIST_HEAD_INIT((head).node_list)
Signed-off-by: NAndy Whitcroft <apw@shadowen.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

If we have sufficient context detect and handle do without braces ({).
Else these incorrectly trigger a trailing statements error for the
associated while.
Signed-off-by: NAndy Whitcroft <apw@shadowen.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

A label is not a candidate for a possible type.  Exclude them.
Signed-off-by: NAndy Whitcroft <apw@shadowen.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

A do without braces '{' may trigger a false possible type 'do' and then
this may be interpreted as an external definition of foo():

	do
		foo();
	while (bar);

Add do to the type exclusions.  Fix up tests so we can check for them.
Signed-off-by: NAndy Whitcroft <apw@shadowen.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Signed-off-by: NAndy Whitcroft <apw@shadowen.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

We should only apply source checks to lines within hunks.  Checks which
are anchored in the context may falsly trigger in the commentory.  Ensure
they only match within valid hunk lines.
Signed-off-by: NAndy Whitcroft <apw@shadowen.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Correct calculation of the number of lines of condition where we have
suspect indent.
Signed-off-by: NAndy Whitcroft <apw@shadowen.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Firmware may be included in the kernel as .ihex files.  These are
inherantly text, but not source.  The line ending checks are applicable to
these kinds of file, allow just these checks to apply to all files.
Signed-off-by: NAndy Whitcroft <apw@shadowen.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Ensure that a close comment cannot incorrectly trigger in the middle of a
string.  Reported by Jaswinder Singh.
Signed-off-by: NAndy Whitcroft <apw@shadowen.org>
Cc: Jaswinder Singh <jaswinder@infradead.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

When the last hunk of a patch is short it will trigger errors from
checkpatch:

    Use of uninitialized value in pattern match (m//)
	    at /usr/local/bin/checkpatch.pl line 394.
    Use of uninitialized value in concatenation (.) or string
	    at /usr/local/bin/checkpatch.pl line 397.
    Use of uninitialized value in pattern match (m//)

Avoid touching beyond the last line.  Reported by Julien Brunel.
Signed-off-by: NAndy Whitcroft <apw@shadowen.org>
Cc: Julien Brunel <brunel@diku.dk>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

We should only be checking changes lines for the trailing statement check
on case/default statements.
Signed-off-by: NAndy Whitcroft <apw@shadowen.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Signed-off-by: NAndy Whitcroft <apw@shadowen.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

So that we eat our own dog food ensure the indent checks apply to perl
too.
Signed-off-by: NAndy Whitcroft <apw@shadowen.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Report the correct lines for single statement blocks.  Currently we are
reporting the right number of lines, but not skipping the negative lines.
Signed-off-by: NAndy Whitcroft <apw@shadowen.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Ensure that we handle literal %'s correctly when adjacent to a %Lx.

	%Lx	bad
	%%Lx	good
	%%%Lx	bad
Signed-off-by: NAndy Whitcroft <apw@shadowen.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

We should skip over and check the lines which follow preprocessor
statements, labels, and blank lines.  These all have legitimate reasons to
be indented differently.
Signed-off-by: NAndy Whitcroft <apw@shadowen.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

We are currently only reporting syspect indents if the conditional is
modified but the indent missmatch could be generated by the body changing,
make sure we catch both.  Also only report the first line of the body, and
more importantly make sure we report the raw copy of the line.  Finally
report the indent levels to make it easier to understand what is wrong.
Signed-off-by: NAndy Whitcroft <apw@shadowen.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Absolute references to kernel source files are generally only useful
locally to the originator of the patch.  Check for any such references and
report them.
Signed-off-by: NAndy Whitcroft <apw@shadowen.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

It is much more likely that an architecture file will want to directly
include asm header files.  Reduce this WARNING to a CHECK when the
referencing file is in the arch directory.
Signed-off-by: NAndy Whitcroft <apw@shadowen.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

It is possible to have other include/asm paths within the tree which are
not subject to the do not edit checks.  Ignore those.
Signed-off-by: NAndy Whitcroft <apw@shadowen.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

We are not counting the lines in the block correctly which causes the
comment scan to stop prematurly and thus miss comments which end at the
end of the block.  Fix this up.
Signed-off-by: NAndy Whitcroft <apw@shadowen.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Only pull in new extension lines where the current contents ends with a \.
Signed-off-by: NAndy Whitcroft <apw@shadowen.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Add the cacheline alignment modifiers to the attribute lists.
Signed-off-by: NAndy Whitcroft <apw@shadowen.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Add support for direct testing of the attribute matcher, add basic tests
for it.
Signed-off-by: NAndy Whitcroft <apw@shadowen.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

It is a common and sane idiom to allow a single return on the end of a
case statement:

	switch (...) {
	case foo:	return bar;
	}

Add an exception for this.
Signed-off-by: NAndy Whitcroft <apw@shadowen.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Labels have different indent rules and must be ignored when checking the
conditional indent levels.  Also correct identify labels in single
statement conditionals.
Signed-off-by: NAndy Whitcroft <apw@shadowen.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

It is possible to use double ampersand (&&) in unary context where it
means the address of a goto label.  Handle spacing for it.
Signed-off-by: NAndy Whitcroft <apw@shadowen.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

It is wholy reasonable to have square brackets representing array slices
in braces on the same line.  These should be spaced.
Signed-off-by: NAndy Whitcroft <apw@shadowen.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Signed-off-by: NAndy Whitcroft <apw@shadowen.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

When checking spacing for pointer checks the type cannot start in the
middle of a word, ie. this is not 'int * bar':

	x = fooint * bar;
Signed-off-by: NAndy Whitcroft <apw@shadowen.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Ensure we ignore comments in complex macro detection else we incorrectly
report this:

	#define PFM_GROUP_PERM_ANY     -1      /* any user/group */
Signed-off-by: NAndy Whitcroft <apw@shadowen.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Now that we have a variants system, move to using that to carry the
unary/binary designation for +, -, &, and *.
Signed-off-by: NAndy Whitcroft <apw@shadowen.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Add checks for the question mark colon operator spacing, and also check
the other uses of colon.  Colon means a number of things:

 - it introduces the else part of the ?: operator,
 - it terminates a goto label,
 - it terminates the case value,
 - it separates the identifier from the bit size on bit fields, and
 - it is used to introduce option types in asm().
Signed-off-by: NAndy Whitcroft <apw@shadowen.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Add support for multiple modifiers such as:

	int __one __two foo;

Also handle trailing known modifiers when defecting modifiers:

	int __one foo __read_mostly;
Signed-off-by: NAndy Whitcroft <apw@shadowen.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>