summaryrefslogtreecommitdiff
path: root/scripts/dtc/dtc-parser.y
diff options
context:
space:
mode:
authorStephen Warren <swarren@nvidia.com>2012-09-28 21:25:59 +0000
committerRob Herring <rob.herring@calxeda.com>2012-10-01 11:11:35 -0500
commitcd296721a9645f9f28800a072490fa15458d1fb7 (patch)
tree492b9a268a48af07844fbbd39519f95676ee73fe /scripts/dtc/dtc-parser.y
parentacc2097934b5403b97f95763fe99fc115b818061 (diff)
dtc: import latest upstream dtc
This updates scripts/dtc to commit 317a5d9 "dtc: zero out new label objects" from git://git.jdl.com/software/dtc.git. This adds features such as: * /bits/ syntax for cell data. * Math expressions within cell data. * The ability to delete properties or nodes. * Support for #line directives in the input file, which allows the use of cpp on *.dts. * -i command-line option (/include/ path) * -W/-E command-line options for error/warning control. * Removal of spew to STDOUT containing the filename being compiled. * Many additions to the libfdt API. Signed-off-by: Stephen Warren <swarren@nvidia.com> Acked-by: Jon Loeliger <jdl@jdl.com> Signed-off-by: Rob Herring <rob.herring@calxeda.com>
Diffstat (limited to 'scripts/dtc/dtc-parser.y')
-rw-r--r--scripts/dtc/dtc-parser.y255
1 files changed, 221 insertions, 34 deletions
diff --git a/scripts/dtc/dtc-parser.y b/scripts/dtc/dtc-parser.y
index 5e84a67fc1d2..f412460f94d7 100644
--- a/scripts/dtc/dtc-parser.y
+++ b/scripts/dtc/dtc-parser.y
@@ -34,6 +34,7 @@ extern struct boot_info *the_boot_info;
extern int treesource_error;
static unsigned long long eval_literal(const char *s, int base, int bits);
+static unsigned char eval_char_literal(const char *s);
%}
%union {
@@ -44,19 +45,28 @@ static unsigned long long eval_literal(const char *s, int base, int bits);
uint8_t byte;
struct data data;
- uint64_t addr;
- cell_t cell;
+ struct {
+ struct data data;
+ int bits;
+ } array;
+
struct property *prop;
struct property *proplist;
struct node *node;
struct node *nodelist;
struct reserve_info *re;
+ uint64_t integer;
}
%token DT_V1
%token DT_MEMRESERVE
+%token DT_LSHIFT DT_RSHIFT DT_LE DT_GE DT_EQ DT_NE DT_AND DT_OR
+%token DT_BITS
+%token DT_DEL_PROP
+%token DT_DEL_NODE
%token <propnodename> DT_PROPNODENAME
%token <literal> DT_LITERAL
+%token <literal> DT_CHAR_LITERAL
%token <cbase> DT_BASE
%token <byte> DT_BYTE
%token <data> DT_STRING
@@ -68,9 +78,7 @@ static unsigned long long eval_literal(const char *s, int base, int bits);
%type <data> propdataprefix
%type <re> memreserve
%type <re> memreserves
-%type <addr> addr
-%type <data> celllist
-%type <cell> cellval
+%type <array> arrayprefix
%type <data> bytestring
%type <prop> propdef
%type <proplist> proplist
@@ -80,6 +88,21 @@ static unsigned long long eval_literal(const char *s, int base, int bits);
%type <node> subnode
%type <nodelist> subnodes
+%type <integer> integer_prim
+%type <integer> integer_unary
+%type <integer> integer_mul
+%type <integer> integer_add
+%type <integer> integer_shift
+%type <integer> integer_rela
+%type <integer> integer_eq
+%type <integer> integer_bitand
+%type <integer> integer_bitxor
+%type <integer> integer_bitor
+%type <integer> integer_and
+%type <integer> integer_or
+%type <integer> integer_trinary
+%type <integer> integer_expr
+
%%
sourcefile:
@@ -102,7 +125,7 @@ memreserves:
;
memreserve:
- DT_MEMRESERVE addr addr ';'
+ DT_MEMRESERVE integer_prim integer_prim ';'
{
$$ = build_reserve_entry($2, $3);
}
@@ -113,13 +136,6 @@ memreserve:
}
;
-addr:
- DT_LITERAL
- {
- $$ = eval_literal($1, 0, 64);
- }
- ;
-
devicetree:
'/' nodedef
{
@@ -139,6 +155,17 @@ devicetree:
print_error("label or path, '%s', not found", $2);
$$ = $1;
}
+ | devicetree DT_DEL_NODE DT_REF ';'
+ {
+ struct node *target = get_node_by_ref($1, $3);
+
+ if (!target)
+ print_error("label or path, '%s', not found", $3);
+ else
+ delete_node(target);
+
+ $$ = $1;
+ }
;
nodedef:
@@ -168,6 +195,10 @@ propdef:
{
$$ = build_property($1, empty_data);
}
+ | DT_DEL_PROP DT_PROPNODENAME ';'
+ {
+ $$ = build_property_delete($2);
+ }
| DT_LABEL propdef
{
add_label(&$2->labels, $1);
@@ -180,9 +211,9 @@ propdata:
{
$$ = data_merge($1, $2);
}
- | propdataprefix '<' celllist '>'
+ | propdataprefix arrayprefix '>'
{
- $$ = data_merge($1, $3);
+ $$ = data_merge($1, $2.data);
}
| propdataprefix '[' bytestring ']'
{
@@ -192,7 +223,7 @@ propdata:
{
$$ = data_add_marker($1, REF_PATH, $2);
}
- | propdataprefix DT_INCBIN '(' DT_STRING ',' addr ',' addr ')'
+ | propdataprefix DT_INCBIN '(' DT_STRING ',' integer_prim ',' integer_prim ')'
{
FILE *f = srcfile_relative_open($4.val, NULL);
struct data d;
@@ -240,31 +271,154 @@ propdataprefix:
}
;
-celllist:
- /* empty */
+arrayprefix:
+ DT_BITS DT_LITERAL '<'
+ {
+ $$.data = empty_data;
+ $$.bits = eval_literal($2, 0, 7);
+
+ if (($$.bits != 8) &&
+ ($$.bits != 16) &&
+ ($$.bits != 32) &&
+ ($$.bits != 64))
+ {
+ print_error("Only 8, 16, 32 and 64-bit elements"
+ " are currently supported");
+ $$.bits = 32;
+ }
+ }
+ | '<'
+ {
+ $$.data = empty_data;
+ $$.bits = 32;
+ }
+ | arrayprefix integer_prim
+ {
+ if ($1.bits < 64) {
+ uint64_t mask = (1ULL << $1.bits) - 1;
+ /*
+ * Bits above mask must either be all zero
+ * (positive within range of mask) or all one
+ * (negative and sign-extended). The second
+ * condition is true if when we set all bits
+ * within the mask to one (i.e. | in the
+ * mask), all bits are one.
+ */
+ if (($2 > mask) && (($2 | mask) != -1ULL))
+ print_error(
+ "integer value out of range "
+ "%016lx (%d bits)", $1.bits);
+ }
+
+ $$.data = data_append_integer($1.data, $2, $1.bits);
+ }
+ | arrayprefix DT_REF
+ {
+ uint64_t val = ~0ULL >> (64 - $1.bits);
+
+ if ($1.bits == 32)
+ $1.data = data_add_marker($1.data,
+ REF_PHANDLE,
+ $2);
+ else
+ print_error("References are only allowed in "
+ "arrays with 32-bit elements.");
+
+ $$.data = data_append_integer($1.data, val, $1.bits);
+ }
+ | arrayprefix DT_LABEL
{
- $$ = empty_data;
+ $$.data = data_add_marker($1.data, LABEL, $2);
}
- | celllist cellval
+ ;
+
+integer_prim:
+ DT_LITERAL
{
- $$ = data_append_cell($1, $2);
+ $$ = eval_literal($1, 0, 64);
}
- | celllist DT_REF
+ | DT_CHAR_LITERAL
{
- $$ = data_append_cell(data_add_marker($1, REF_PHANDLE,
- $2), -1);
+ $$ = eval_char_literal($1);
}
- | celllist DT_LABEL
+ | '(' integer_expr ')'
{
- $$ = data_add_marker($1, LABEL, $2);
+ $$ = $2;
}
;
-cellval:
- DT_LITERAL
- {
- $$ = eval_literal($1, 0, 32);
- }
+integer_expr:
+ integer_trinary
+ ;
+
+integer_trinary:
+ integer_or
+ | integer_or '?' integer_expr ':' integer_trinary { $$ = $1 ? $3 : $5; }
+ ;
+
+integer_or:
+ integer_and
+ | integer_or DT_OR integer_and { $$ = $1 || $3; }
+ ;
+
+integer_and:
+ integer_bitor
+ | integer_and DT_AND integer_bitor { $$ = $1 && $3; }
+ ;
+
+integer_bitor:
+ integer_bitxor
+ | integer_bitor '|' integer_bitxor { $$ = $1 | $3; }
+ ;
+
+integer_bitxor:
+ integer_bitand
+ | integer_bitxor '^' integer_bitand { $$ = $1 ^ $3; }
+ ;
+
+integer_bitand:
+ integer_eq
+ | integer_bitand '&' integer_eq { $$ = $1 & $3; }
+ ;
+
+integer_eq:
+ integer_rela
+ | integer_eq DT_EQ integer_rela { $$ = $1 == $3; }
+ | integer_eq DT_NE integer_rela { $$ = $1 != $3; }
+ ;
+
+integer_rela:
+ integer_shift
+ | integer_rela '<' integer_shift { $$ = $1 < $3; }
+ | integer_rela '>' integer_shift { $$ = $1 > $3; }
+ | integer_rela DT_LE integer_shift { $$ = $1 <= $3; }
+ | integer_rela DT_GE integer_shift { $$ = $1 >= $3; }
+ ;
+
+integer_shift:
+ integer_shift DT_LSHIFT integer_add { $$ = $1 << $3; }
+ | integer_shift DT_RSHIFT integer_add { $$ = $1 >> $3; }
+ | integer_add
+ ;
+
+integer_add:
+ integer_add '+' integer_mul { $$ = $1 + $3; }
+ | integer_add '-' integer_mul { $$ = $1 - $3; }
+ | integer_mul
+ ;
+
+integer_mul:
+ integer_mul '*' integer_unary { $$ = $1 * $3; }
+ | integer_mul '/' integer_unary { $$ = $1 / $3; }
+ | integer_mul '%' integer_unary { $$ = $1 % $3; }
+ | integer_unary
+ ;
+
+integer_unary:
+ integer_prim
+ | '-' integer_unary { $$ = -$2; }
+ | '~' integer_unary { $$ = ~$2; }
+ | '!' integer_unary { $$ = !$2; }
;
bytestring:
@@ -303,6 +457,10 @@ subnode:
{
$$ = name_node($2, $1);
}
+ | DT_DEL_NODE DT_PROPNODENAME ';'
+ {
+ $$ = name_node(build_node_delete(), $2);
+ }
| DT_LABEL subnode
{
add_label(&$2->labels, $1);
@@ -334,12 +492,41 @@ static unsigned long long eval_literal(const char *s, int base, int bits)
errno = 0;
val = strtoull(s, &e, base);
- if (*e)
- print_error("bad characters in literal");
- else if ((errno == ERANGE)
+ if (*e) {
+ size_t uls = strspn(e, "UL");
+ if (e[uls])
+ print_error("bad characters in literal");
+ }
+ if ((errno == ERANGE)
|| ((bits < 64) && (val >= (1ULL << bits))))
print_error("literal out of range");
else if (errno != 0)
print_error("bad literal");
return val;
}
+
+static unsigned char eval_char_literal(const char *s)
+{
+ int i = 1;
+ char c = s[0];
+
+ if (c == '\0')
+ {
+ print_error("empty character literal");
+ return 0;
+ }
+
+ /*
+ * If the first character in the character literal is a \ then process
+ * the remaining characters as an escape encoding. If the first
+ * character is neither an escape or a terminator it should be the only
+ * character in the literal and will be returned.
+ */
+ if (c == '\\')
+ c = get_escape_char(s, &i);
+
+ if (s[i] != '\0')
+ print_error("malformed character literal");
+
+ return c;
+}