aboutsummaryrefslogtreecommitdiff
path: root/ld/emultempl
diff options
context:
space:
mode:
authorAlan Modra <amodra@gmail.com>2001-08-20 00:22:41 +0000
committerAlan Modra <amodra@gmail.com>2001-08-20 00:22:41 +0000
commitae7fb08f1a002acd930c38897b9993b780d8798c (patch)
tree87762c70a8a75fd06e2d4e560da11eab4e12072c /ld/emultempl
parent415b1b13876006c4704a85e6dfc8bdda38a5056e (diff)
downloadgdb-ae7fb08f1a002acd930c38897b9993b780d8798c.zip
gdb-ae7fb08f1a002acd930c38897b9993b780d8798c.tar.gz
gdb-ae7fb08f1a002acd930c38897b9993b780d8798c.tar.bz2
* pe-dll.c (pe_dll_fill_sections): Correct type of "relax" param
passed to lang_size_sections. (pe_exe_fill_sections): Likewise. * emultempl/pe.em (output_prev_sec_find): Copied from elf32.em. (gld_${EMULATION_NAME}_place_orphan): Merge from elf32.em.
Diffstat (limited to 'ld/emultempl')
-rw-r--r--ld/emultempl/pe.em123
1 files changed, 89 insertions, 34 deletions
diff --git a/ld/emultempl/pe.em b/ld/emultempl/pe.em
index 6519eae..fce661e 100644
--- a/ld/emultempl/pe.em
+++ b/ld/emultempl/pe.em
@@ -108,6 +108,8 @@ static void gld_${EMULATION_NAME}_after_open PARAMS ((void));
static void gld_${EMULATION_NAME}_before_parse PARAMS ((void));
static void gld_${EMULATION_NAME}_after_parse PARAMS ((void));
static void gld_${EMULATION_NAME}_before_allocation PARAMS ((void));
+static asection *output_prev_sec_find
+ PARAMS ((lang_output_section_statement_type *));
static boolean gld_${EMULATION_NAME}_place_orphan
PARAMS ((lang_input_statement_type *, asection *));
static char *gld_${EMULATION_NAME}_get_script PARAMS ((int *));
@@ -1405,6 +1407,32 @@ gld_${EMULATION_NAME}_finish ()
}
+/* Find the last output section before given output statement.
+ Used by place_orphan. */
+
+static asection *
+output_prev_sec_find (os)
+ lang_output_section_statement_type *os;
+{
+ asection *s = (asection *) NULL;
+ lang_statement_union_type *u;
+ lang_output_section_statement_type *lookup;
+
+ for (u = lang_output_section_statement.head;
+ u != (lang_statement_union_type *) NULL;
+ u = lookup->next)
+ {
+ lookup = &u->output_section_statement;
+ if (lookup == os)
+ return s;
+
+ if (lookup->bfd_section != NULL && lookup->bfd_section->owner != NULL)
+ s = lookup->bfd_section;
+ }
+
+ return NULL;
+}
+
/* Place an orphan section.
We use this to put sections in a reasonable place in the file, and
@@ -1455,9 +1483,12 @@ gld_${EMULATION_NAME}_place_orphan (file, s)
lang_list_init (&add_child);
if (os != NULL
- && os->bfd_section != NULL
- && ((s->flags ^ os->bfd_section->flags) & (SEC_LOAD | SEC_ALLOC)) == 0)
+ && (os->bfd_section == NULL
+ || ((s->flags ^ os->bfd_section->flags)
+ & (SEC_LOAD | SEC_ALLOC)) == 0))
{
+ /* We already have an output section statement with this
+ name, and its bfd section, if any, has compatible flags. */
lang_add_section (&add_child, s, os, file);
}
else
@@ -1498,7 +1529,7 @@ gld_${EMULATION_NAME}_place_orphan (file, s)
/* Choose a unique name for the section. This will be needed if
the same section name appears in the input file with
- different loadable or allocateable characteristics. */
+ different loadable or allocatable characteristics. */
outsecname = xstrdup (hold_section_name);
if (bfd_get_section_by_name (output_bfd, outsecname) != NULL)
{
@@ -1590,28 +1621,30 @@ gld_${EMULATION_NAME}_place_orphan (file, s)
asection *snew, **pps;
snew = os->bfd_section;
- if (place->os->bfd_section != NULL || place->section != NULL)
+
+ /* Shuffle the bfd section list to make the output file look
+ neater. This is really only cosmetic. */
+ if (place->section == NULL)
{
- /* Shuffle the section to make the output file look neater. */
- if (place->section == NULL)
- {
-#if 0
- /* Finding the end of the list is a little tricky. We
- make a wild stab at it by comparing section flags. */
- flagword first_flags = place->os->bfd_section->flags;
- for (pps = &place->os->bfd_section->next;
- *pps != NULL && (*pps)->flags == first_flags;
- pps = &(*pps)->next)
- ;
- place->section = pps;
-#else
- /* Put orphans after the first section on the list. */
- place->section = &place->os->bfd_section->next;
-#endif
- }
+ asection *bfd_section = place->os->bfd_section;
+
+ /* If the output statement hasn't been used to place
+ any input sections (and thus doesn't have an output
+ bfd_section), look for the closest prior output statement
+ having an output section. */
+ if (bfd_section == NULL)
+ bfd_section = output_prev_sec_find (place->os);
+
+ if (bfd_section != NULL && bfd_section != snew)
+ place->section = &bfd_section->next;
+ }
+ if (place->section != NULL)
+ {
/* Unlink the section. */
- for (pps = &output_bfd->sections; *pps != snew; pps = &(*pps)->next)
+ for (pps = &output_bfd->sections;
+ *pps != snew;
+ pps = &(*pps)->next)
;
*pps = snew->next;
@@ -1619,21 +1652,43 @@ gld_${EMULATION_NAME}_place_orphan (file, s)
snew->next = *place->section;
*place->section = snew;
}
- place->section = &snew->next; /* Save the end of this list. */
- if (place->stmt == NULL)
- {
- /* Put the new statement list right at the head. */
- *add.tail = place->os->header.next;
- place->os->header.next = add.head;
- }
- else
+ /* Save the end of this list. Further ophans of this type will
+ follow the one we've just added. */
+ place->section = &snew->next;
+
+ /* The following is non-cosmetic. We try to put the output
+ statements in some sort of reasonable order here, because
+ they determine the final load addresses of the orphan
+ sections. In addition, placing output statements in the
+ wrong order may require extra segments. For instance,
+ given a typical situation of all read-only sections placed
+ in one segment and following that a segment containing all
+ the read-write sections, we wouldn't want to place an orphan
+ read/write section before or amongst the read-only ones. */
+ if (add.head != NULL)
{
- /* Put it after the last orphan statement we added. */
- *add.tail = *place->stmt;
- *place->stmt = add.head;
+ if (place->stmt == NULL)
+ {
+ /* Put the new statement list right at the head. */
+ *add.tail = place->os->header.next;
+ place->os->header.next = add.head;
+ }
+ else
+ {
+ /* Put it after the last orphan statement we added. */
+ *add.tail = *place->stmt;
+ *place->stmt = add.head;
+ }
+
+ /* Fix the global list pointer if we happened to tack our
+ new list at the tail. */
+ if (*old->tail == add.head)
+ old->tail = add.tail;
+
+ /* Save the end of this list. */
+ place->stmt = add.tail;
}
- place->stmt = add.tail; /* Save the end of this list. */
}
}