public inbox for gdb-patches@sourceware.org
 help / color / mirror / Atom feed
* [PATCH][gdb/symtab] Relocate call_site_htab
@ 2021-09-30  9:56 Tom de Vries
  2021-09-30 14:26 ` Simon Marchi
  0 siblings, 1 reply; 7+ messages in thread
From: Tom de Vries @ 2021-09-30  9:56 UTC (permalink / raw)
  To: gdb-patches; +Cc: Tom Tromey, Pedro Alves

[ CC-ing maintainers that reviewed previous submission. ]

Hi,

When running test-case gdb.arch/amd64-entry-value-inline.exp with target board
unix/-no-pie/-fno-PIE we have:
...
(gdb) continue^M
Continuing.^M
^M
Breakpoint 2, fn2 (y=y@entry=25, x=x@entry=6) at \
  gdb.arch/amd64-entry-value-inline.c:32^M
32            y = -2 + x;       /* break-here */^M
(gdb) PASS: gdb.arch/amd64-entry-value-inline.exp: \
  continue to breakpoint: break-here
p y^M
$1 = 25^M
(gdb) PASS: gdb.arch/amd64-entry-value-inline.exp: p y
...

But with target board unix/-pie/-fPIE we have instead:
...
p y^M
$1 = <optimized out>^M
(gdb) FAIL: gdb.arch/amd64-entry-value-inline.exp: p y
...

The test-case uses a .S file, which was generated using gcc 4.8.0, but I can
reproduce the same problem using the original C file and gcc 4.8.5.

The problem is that in order to access the value, call_site information is
accessed, which is both:
- unrelocated, and
- accessed as if it were relocated.

I've submitted an attempt at fixing this before, trying to handle this at all
points where the information is used (
https://sourceware.org/pipermail/gdb-patches/2019-August/159631.html ).

Instead, fix this more reliably by relocating the call_site information.

This fixes for me all remaining regressions for unix/-pie/-fPIE vs
unix/-no-pie/-fno-PIE (not counting ada compilation FAILs).

Tested on x86_64-linux.

Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=24892

Any comments?

Thanks,
- Tom

[gdb/symtab] Relocate call_site_htab

---
 gdb/objfiles.c | 58 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 58 insertions(+)

diff --git a/gdb/objfiles.c b/gdb/objfiles.c
index b65fa8820ca..ceff0fcfba4 100644
--- a/gdb/objfiles.c
+++ b/gdb/objfiles.c
@@ -626,6 +626,60 @@ relocate_one_symbol (struct symbol *sym, struct objfile *objfile,
     }
 }
 
+/* Relocate call_site S using offset DELTA.  */
+
+static void
+call_site_relocate (struct call_site *s, CORE_ADDR delta)
+{
+  s->pc += delta;
+  if (FIELD_LOC_KIND (s->target) == FIELD_LOC_KIND_PHYSADDR)
+    FIELD_STATIC_PHYSADDR (s->target) += delta;
+}
+
+/* Relocate HTAB, which is a COMPUNIT_CALL_SITE_HTAB using offset DELTA.  */
+
+static void
+compunit_call_site_htab_relocate (htab_t htab, CORE_ADDR delta)
+{
+  /* Changing the pc field changes the hashcode, so we can't just update the
+     elements.  Instead, we move them to this var, and then reinsert them.  */
+  std::vector<struct call_site *> tmp;
+
+  /* Copy elements to tmp.  */
+  auto visitor_func
+    = [] (void **slot, void *info) -> int
+    {
+      /* Copy element to tmp.  */
+      struct call_site *s = (struct call_site *) *slot;
+      std::vector<struct call_site *> *tmp_ptr
+      = (std::vector<struct call_site *> *)info;
+      tmp_ptr->push_back (s);
+
+      /* Keep going.  */
+      return 1;
+    };
+  htab_traverse (htab, visitor_func, &tmp);
+
+  /* Make hashtable empty.  This does not destroy the elements because the
+     hashtable is created with del_f == nullptr.  */
+  htab_empty (htab);
+
+  /* Relocate and reinsert elements.  */
+  for (struct call_site *s : tmp) {
+    /* Relocate element.  */
+    call_site_relocate (s, delta);
+
+    /* Reinsert element.  */
+    struct call_site call_site_local;
+    call_site_local.pc = s->pc;
+    void **slot
+      = htab_find_slot (htab, &call_site_local, INSERT);
+    gdb_assert (slot != NULL);
+    gdb_assert (*slot == NULL);
+    *slot = s;
+  }
+}
+
 /* Relocate OBJFILE to NEW_OFFSETS.  There should be OBJFILE->NUM_SECTIONS
    entries in new_offsets.  SEPARATE_DEBUG_OBJFILE is not touched here.
    Return non-zero iff any change happened.  */
@@ -697,6 +751,10 @@ objfile_relocate1 (struct objfile *objfile,
 		relocate_one_symbol (sym, objfile, delta);
 	      }
 	  }
+
+	if (COMPUNIT_CALL_SITE_HTAB (cust) != nullptr)
+	  compunit_call_site_htab_relocate (COMPUNIT_CALL_SITE_HTAB (cust),
+					    delta[block_line_section]);
       }
   }
 

^ permalink raw reply	[flat|nested] 7+ messages in thread
end of thread, other threads:[~2021-10-01 12:37 UTC | newest]

Thread overview: 7+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-09-30  9:56 [PATCH][gdb/symtab] Relocate call_site_htab Tom de Vries
2021-09-30 14:26 ` Simon Marchi
2021-09-30 18:14   ` Tom Tromey
2021-09-30 23:47   ` Tom de Vries
2021-10-01  1:15     ` Simon Marchi
2021-10-01  8:25       ` Tom de Vries
2021-10-01 12:37         ` Tom de Vries

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).