feat: improve segment check logic

.github/workflows/codspeed.yml

@@ -92,8 +92,6 @@ jobs:
       - name: Run the benchmarks
         uses: CodSpeedHQ/action@main
         continue-on-error: ${{ matrix.valgrind != 'local' }}
-        env:
-          CODSPEED_PERF_ENABLED: false
         with:
           working-directory: bench
           mode: walltime

callgrind/tests/Makefile.am

@@ -10,6 +10,7 @@ EXTRA_DIST = \
 	ann1.post.exp ann1.stderr.exp ann1.vgtest \
 	ann2.post.exp ann2.stderr.exp ann2.vgtest \
 	clreq.vgtest clreq.stderr.exp \
+	find_debuginfo.vgtest find_debuginfo.stderr.exp find_debuginfo.post.exp \
 	runtime_obj_skip_py.vgtest runtime_obj_skip_py.stderr.exp runtime_obj_skip_py.post.exp \
 	runtime_obj_skip_py.py runtime_obj_skip_py_shim.c \
 	bug497723.stderr.exp bug497723.post.exp bug497723.vgtest \
@@ -30,7 +31,7 @@ EXTRA_DIST = \
 	inline-crossfile.vgtest inline-crossfile.stderr.exp inline-crossfile.stdout.exp inline-crossfile.post.exp \
 	inline-crossfile-helper1.h inline-crossfile-helper2.h filter_inline
 
-check_PROGRAMS = clreq simwork threads inline-samefile inline-crossfile
+check_PROGRAMS = clreq find_debuginfo simwork threads inline-samefile inline-crossfile
 
 AM_CFLAGS   += $(AM_FLAG_M3264_PRI)
 AM_CXXFLAGS += $(AM_FLAG_M3264_PRI)

callgrind/tests/find_debuginfo.c

@@ -0,0 +1,26 @@
+/* Exercise VG_(find_DebugInfo) including the filename fallback for
+   executable code that lives outside the section named ".text".
+
+   Functions placed in orphan sections "warm_code" and "cold_code" end up
+   in their own ELF sections (PROGBITS, AX) inside the same R-E LOAD as
+   .text but *outside* di->text_avma/text_size. Calls into them force
+   VG_(find_DebugInfo) past the primary text-range check and into the
+   am_find_nsegment + filename match path added in commit 938e424b1. */
+
+static int hot_fn(int x) { return x + 1; }
+
+__attribute__((noinline, section("warm_code")))
+static int warm_fn(int x) { return x * 3 + 7; }
+
+__attribute__((noinline, section("cold_code")))
+static int cold_fn(int x) { return x ^ 0x5a; }
+
+int main(void)
+{
+   int y = 0;
+   for (int i = 0; i < 1000; i++)
+      y = hot_fn(y);
+   y = warm_fn(y);
+   y = cold_fn(y);
+   return y & 0xff;
+}

callgrind/tests/find_debuginfo.post.exp

@@ -0,0 +1,3 @@
+find_debuginfo
+ld-linux-x86-64.so.2
+libc.so.6

callgrind/tests/find_debuginfo.stderr.exp

@@ -0,0 +1,6 @@
+
+
+Events    : Ir
+Collected :
+
+I   refs:

callgrind/tests/find_debuginfo.vgtest

@@ -0,0 +1,4 @@
+prog: find_debuginfo
+vgopts: --compress-strings=no --callgrind-out-file=callgrind.out.find_debuginfo
+post: sh -c 'sed -n "s|^ob=.*/||p" callgrind.out.find_debuginfo | grep -v "^vgpreload" | sort -u'
+cleanup: rm -f callgrind.out.find_debuginfo

coregrind/m_debuginfo/debuginfo.c

@@ -2621,21 +2621,28 @@ DebugInfo* VG_(find_DebugInfo) ( DiEpoch ep, Addr a )
       optimized binaries: .bolt.org.text + .text + .text.warm + .text.cold
       live in two separate R-E PT_LOAD segments). The text-range check above
       only covers the section named ".text", so addresses in the other
-      executable region are missed and end up attributed to "???". Ask the
-      address-space manager which file backs this address, and match it to
-      a DebugInfo by filename. */
+      executable region are missed and end up attributed to "???". Match
+      against the rx mappings recorded for each DebugInfo. find_rx_mapping
+      keeps a per-DI single-entry cache (last_rx_map) so the hot case is
+      O(1) after the first hit. */
    if (eq_DiEpoch(ep, VG_(current_DiEpoch)())) {
-      const NSegment* seg = VG_(am_find_nsegment)(a);
-      const HChar* filename;
-      if (seg != NULL && (filename = VG_(am_get_filename)(seg)) != NULL) {
-         for (di = debugInfo_list; di != NULL; di = di->next) {
-            if (!is_DI_valid_for_epoch(di, ep))
-               continue;
-            if (di->fsm.filename != NULL
-                && 0 == VG_(strcmp)(di->fsm.filename, filename)) {
-               return di;
-            }
-         }
+      for (di = debugInfo_list; di != NULL; di = di->next) {
+         if (!is_DI_valid_for_epoch(di, ep))
+            continue;
+
+         const DebugInfoMapping* map = ML_(find_rx_mapping)(di, a, a);
+         if (map == NULL)
+             continue;
+
+        if (VG_(clo_verbosity) >= 2) {
+            VG_(message)(Vg_DebugMsg,
+                "find_DebugInfo: rx-mapping fallback matched 0x%lx -> "
+                "DI %p (%s) [avma=0x%lx size=%lu]\n",
+                a, di,
+                di->fsm.filename ? di->fsm.filename : "(no filename)",
+                map->avma, map->size);
+        }
+        return di;
       }
    }
    return NULL;