diff --git a/esphome/components/esp8266/boards.py b/esphome/components/esp8266/boards.py
index e5a9628e90..2646682766 100644
--- a/esphome/components/esp8266/boards.py
+++ b/esphome/components/esp8266/boards.py
@@ -362,9 +362,10 @@ BOARDS = {
 }
 
 """
-ESP8266_BOARD_BUILD generate with:
+ESP8266_BOARD_BUILD generate with (v4.2.1 is the platform version the
+native toolchain mirrors; regenerate against the tag when bumping it):
 
-git clone https://github.com/platformio/platform-espressif8266
+git clone -b v4.2.1 https://github.com/platformio/platform-espressif8266
 python3 - <<'EOF'
 import json, glob, os
 for f in sorted(glob.glob("platform-espressif8266/boards/*.json")):
diff --git a/esphome/components/esp8266/build_surgery.py b/esphome/components/esp8266/build_surgery.py
index c0034f7bad..47f9042f08 100644
--- a/esphome/components/esp8266/build_surgery.py
+++ b/esphome/components/esp8266/build_surgery.py
@@ -58,21 +58,33 @@ _TESTING_SEGMENT_SIZES = {
 
 
 def _segment_line_re(segment_name: str) -> re.Pattern[str]:
-    """The MEMORY line for one segment: ``<seg> : org = 0x..., len = 0x...``."""
+    """The MEMORY line for one segment: ``<seg> : org = 0x..., len = 0x...``.
+
+    Anchored to the start of the line so a name never matches inside a
+    longer one (``ram0_0_seg`` must not read ``dram0_0_seg``). The size
+    group stops at the hex digits, leaving any ``ul`` suffix (from the
+    preprocessed ``MMU_IRAM_SIZE``) in place.
+    """
     return re.compile(
-        rf"({segment_name}\s*:\s*org\s*=\s*0x[0-9a-fA-F]+\s*,\s*len\s*=\s*)"
-        r"(0x[0-9a-fA-F]+)"
+        rf"(^[ \t]*{re.escape(segment_name)}"
+        r"\s*:\s*org\s*=\s*0x[0-9a-fA-F]+\s*,\s*len\s*=\s*)"
+        r"(0x[0-9a-fA-F]+)",
+        re.MULTILINE,
     )
 
 
 def apply_testing_memory_patches(content: str, segments: Collection[str]) -> str:
     """Enlarge the named memory segments so grouped CI test builds can link.
 
-    Each caller passes the segments its linker script defines; a segment
-    that fails to match raises, since a silently kept real memory limit
-    would fail grouped builds far from the cause.
+    Each caller passes the segments its linker script defines: the
+    generated common ld carries ``iram1_0_seg``; the flash ld carries
+    ``dram0_0_seg`` and ``irom0_0_seg``. A segment that fails to match
+    raises, since a silently kept real memory limit would fail grouped
+    builds far from the cause.
     """
     for segment in segments:
+        if segment not in _TESTING_SEGMENT_SIZES:
+            raise RuntimeError(f"Unknown testing-mode segment {segment!r}")
         content, count = _segment_line_re(segment).subn(
             rf"\g<1>{_TESTING_SEGMENT_SIZES[segment]}", content
         )
diff --git a/tests/unit_tests/components/esp8266/test_build_surgery.py b/tests/unit_tests/components/esp8266/test_build_surgery.py
index 64e36f1591..e62a2270fb 100644
--- a/tests/unit_tests/components/esp8266/test_build_surgery.py
+++ b/tests/unit_tests/components/esp8266/test_build_surgery.py
@@ -24,16 +24,26 @@ _COMMON_LD_SNIPPET = """\
   } >dram0_0_seg :dram0_0_phdr
 """
 
+# Shaped like the real SDK flash ld scripts: no iram1_0_seg (that lives in
+# the generated common ld only)
 _FLASH_LD_SNIPPET = """\
 MEMORY
 {
   dport0_0_seg :                        org = 0x3FF00000, len = 0x10
   dram0_0_seg :                         org = 0x3FFE8000, len = 0x14000
-  iram1_0_seg :                         org = 0x40100000, len = 0x8000
   irom0_0_seg :                         org = 0x40201010, len = 0xfeff0
 }
 """
 
+# Shaped like the preprocessed common ld: MMU_IRAM_SIZE expands with a ul
+# suffix the patcher must leave in place
+_COMMON_LD_MEMORY_SNIPPET = """\
+MEMORY
+{
+  iram1_0_seg :                         org = 0x40100000, len = 0x8000ul
+}
+"""
+
 
 def test_relocate_ratetable_inserts_after_data_start() -> None:
     patched = relocate_ratetable(_COMMON_LD_SNIPPET)
@@ -52,15 +62,32 @@ def test_relocate_ratetable_requires_anchor() -> None:
 
 def test_testing_memory_patches_enlarge_segments() -> None:
     patched = apply_testing_memory_patches(
-        _FLASH_LD_SNIPPET, ("iram1_0_seg", "dram0_0_seg", "irom0_0_seg")
+        _FLASH_LD_SNIPPET, ("dram0_0_seg", "irom0_0_seg")
     )
-    assert segment_length(patched, "iram1_0_seg") == 0x200000
     assert segment_length(patched, "dram0_0_seg") == 0x200000
     assert segment_length(patched, "irom0_0_seg") == 0x2000000
     # Untouched segments keep their sizes
     assert segment_length(patched, "dport0_0_seg") == 0x10
 
 
+def test_testing_memory_patches_keep_ul_suffix() -> None:
+    """The common ld's preprocessed sizes carry a ul suffix; the patch must
+    replace only the hex digits, as testing_mode.py.script does."""
+    patched = apply_testing_memory_patches(_COMMON_LD_MEMORY_SNIPPET, ("iram1_0_seg",))
+    assert "len = 0x200000ul" in patched
+    assert segment_length(patched, "iram1_0_seg") == 0x200000
+
+
+def test_segment_length_requires_whole_name() -> None:
+    """A name must match its own line, never inside a longer segment name."""
+    assert segment_length(_FLASH_LD_SNIPPET, "ram0_0_seg") is None
+
+
+def test_testing_memory_patches_unknown_segment_raises() -> None:
+    with pytest.raises(RuntimeError, match="Unknown testing-mode segment"):
+        apply_testing_memory_patches("MEMORY { }", ("bogus_seg",))
+
+
 def test_segment_length() -> None:
     assert segment_length(_FLASH_LD_SNIPPET, "irom0_0_seg") == 0xFEFF0
     assert segment_length(_FLASH_LD_SNIPPET, "missing_seg") is None