diff --git a/.circleci/config.yml b/.circleci/config.yml
index aefe8fc0f934a..7c0d945d74065 100644
--- a/.circleci/config.yml
+++ b/.circleci/config.yml
@@ -68,7 +68,6 @@ commands:
             $PYTHON_BIN ./embuilder.py build ALL
             $PYTHON_BIN tests/runner.py test_hello_world
       - run:
-
           name: embuilder (LTO)
           command: |
             $PYTHON_BIN ./embuilder.py build MINIMAL --lto
diff --git a/ChangeLog.md b/ChangeLog.md
index 59f3ca256fd19..d8f987350d0ac 100644
--- a/ChangeLog.md
+++ b/ChangeLog.md
@@ -17,6 +17,8 @@ See docs/process.md for how version tagging works.
 
 Current Trunk
 -------------
+- A new `PRINTF_LONG_DOUBLE` option allows printf to print long doubles at full
+  float128 precision. (#11130)
 - `emscripten_async_queue_on_thread` has been renamed to
   `emscripten_dispatch_to_thread` which no longer implies that it is async -
   the operation is in fact only async if it is sent to another thread, while it
diff --git a/src/settings.js b/src/settings.js
index a8ab4d1060a31..924921c7cca43 100644
--- a/src/settings.js
+++ b/src/settings.js
@@ -1773,6 +1773,12 @@ var LLD_REPORT_UNDEFINED = 0;
 // programs. This which matches the behaviour of gcc/g++ and clang/clang++.
 var DEFAULT_TO_CXX = 1;
 
+// While LLVM's wasm32 has long double = float128, we don't support printing
+// that at full precision by default. Instead we print as 64-bit doubles, which
+// saves libc code size. You can flip this option on to get a libc with full
+// long double printing precision.
+var PRINTF_LONG_DOUBLE = 0;
+
 //===========================================
 // Internal, used for testing only, from here
 //===========================================
diff --git a/system/lib/libc/musl/src/stdio/vfprintf.c b/system/lib/libc/musl/src/stdio/vfprintf.c
index 83a1a6a964195..40b9388cfb8c9 100644
--- a/system/lib/libc/musl/src/stdio/vfprintf.c
+++ b/system/lib/libc/musl/src/stdio/vfprintf.c
@@ -9,11 +9,12 @@
 #include <math.h>
 #include <float.h>
 
+#ifndef EMSCRIPTEN_PRINTF_LONG_DOUBLE
 // XXX EMSCRIPTEN - while wasm32 has long double = float128, we don't support
-//                  printing it at full precision; instead, we lower to
+//                  printing at full precision by default. instead, we lower to
 //                  64-bit double. These macros makes our changes a little less
 //                  invasive.
-#define long_double double
+typedef double long_double;
 #undef LDBL_TRUE_MIN
 #define LDBL_TRUE_MIN DBL_DENORM_MIN
 #undef LDBL_MIN
@@ -36,6 +37,10 @@
 #define LDBL_MAX_10_EXP DBL_MAX_10_EXP
 #undef frexpl
 #define frexpl(x, exp) frexp(x, exp)
+#else // EMSCRIPTEN_FULL_LONG_DOUBLE_PRINTING
+// XXX EMSCRIPTEN - full long double printing support
+typedef long double long_double;
+#endif
 
 /* Some useful macros */
 
@@ -244,9 +249,9 @@ typedef char compiler_defines_long_double_incorrectly[9-(int)sizeof(long_double)
 //                  get it linked in
 //                  also use a double argument here, as mentioned before,
 //                  we print float128s at double precision 
-typedef int (*fmt_fp_t)(FILE *f, double y, int w, int p, int fl, int t);
+typedef int (*fmt_fp_t)(FILE *f, long_double y, int w, int p, int fl, int t);
 
-static int fmt_fp(FILE *f, double y, int w, int p, int fl, int t)
+static int fmt_fp(FILE *f, long_double y, int w, int p, int fl, int t)
 {
 	uint32_t big[(LDBL_MANT_DIG+28)/29 + 1          // mantissa expansion
 		+ (LDBL_MAX_EXP+LDBL_MANT_DIG+28+8)/9]; // exponent expansion
diff --git a/tests/test_other.py b/tests/test_other.py
index 8b44480e72d26..c281d46722687 100644
--- a/tests/test_other.py
+++ b/tests/test_other.py
@@ -10055,6 +10055,35 @@ def test(code):
     self.assertGreater(lf, both - 100)
     self.assertLess(lf, both - 50)
 
+  @parameterized({
+    'normal': ([], '''\
+0.000051 => -5.123719529365189373493194580078e-05
+0.000051 => -5.123719300544352718866300544498e-05
+0.000051 => -5.123719300544352718866300544498e-05
+'''),
+    'full_long_double': (['-s', 'PRINTF_LONG_DOUBLE'], '''\
+0.000051 => -5.123719529365189373493194580078e-05
+0.000051 => -5.123719300544352718866300544498e-05
+0.000051 => -5.123719300544352710023893104250e-05
+'''),
+  })
+  @no_fastcomp('float128 is wasm backend only')
+  def test_long_double_printing(self, args, expected):
+    create_test_file('src.cpp', r'''
+#include <stdio.h>
+
+int main(void) {
+    float f = 5.123456789e-5;
+    double d = 5.123456789e-5;
+    long double ld = 5.123456789e-5;
+    printf("%f => %.30e\n", f, f  / (f - 1));
+    printf("%f => %.30e\n", d, d / (d - 1));
+    printf("%Lf => %.30Le\n", ld, ld / (ld - 1));
+}
+    ''')
+    run_process([PYTHON, EMCC, 'src.cpp'] + args)
+    self.assertContained(expected, run_js('a.out.js'))
+
   # Tests that passing -s MALLOC=none will not include system malloc() to the build.
   def test_malloc_none(self):
     stderr = self.expect_fail([PYTHON, EMCC, path_from_root('tests', 'malloc_none.c'), '-s', 'MALLOC=none'])
diff --git a/tools/system_libs.py b/tools/system_libs.py
index 3091e14668b1b..99c072845d4d3 100755
--- a/tools/system_libs.py
+++ b/tools/system_libs.py
@@ -814,6 +814,20 @@ def get_files(self):
     return libc_files
 
 
+class libprintf_long_double(libc):
+  name = 'libprintf_long_double'
+
+  cflags = ['-DEMSCRIPTEN_PRINTF_LONG_DOUBLE']
+
+  def get_files(self):
+    return files_in_path(
+        path_components=['system', 'lib', 'libc', 'musl', 'src', 'stdio'],
+        filenames=['vfprintf.c'])
+
+  def can_build(self):
+    return super(libprintf_long_double, self).can_build() and shared.Settings.WASM_BACKEND
+
+
 class libsockets(MuslInternalLibrary, MTLibrary):
   name = 'libsockets'
 
@@ -1571,6 +1585,15 @@ def add_library(lib):
 
     sanitize = shared.Settings.USE_LSAN or shared.Settings.USE_ASAN or shared.Settings.UBSAN_RUNTIME
 
+    # JS math must come before anything else, so that it overrides the normal
+    # libc math.
+    if shared.Settings.JS_MATH:
+      add_library(system_libs_map['libjsmath'])
+
+    # to override the normal libc printf, we must come before it
+    if shared.Settings.PRINTF_LONG_DOUBLE:
+      add_library(system_libs_map['libprintf_long_double'])
+
     add_library(system_libs_map['libc'])
     add_library(system_libs_map['libcompiler_rt'])
     if not shared.Settings.WASM_BACKEND and not shared.Settings.MINIMAL_RUNTIME:
@@ -1633,11 +1656,6 @@ def add_library(lib):
   # header methods. To avoid that, we link libc++abi last.
   libs_to_link.sort(key=lambda x: x[0].endswith('libc++abi.bc'))
 
-  # JS math must come before anything else, so that it overrides the normal
-  # libc math.
-  if shared.Settings.JS_MATH:
-    libs_to_link = [(system_libs_map['libjsmath'].get_path(), True)] + libs_to_link
-
   # When LINKABLE is set the entire link command line is wrapped in --whole-archive by
   # Building.link_ldd.  And since --whole-archive/--no-whole-archive processing does not nest we
   # shouldn't add any extra `--no-whole-archive` or we will undo the intent of Building.link_ldd.