diff --git a/avr-hal-generic/Cargo.toml b/avr-hal-generic/Cargo.toml
index fae3aeb010..81a940ac65 100644
--- a/avr-hal-generic/Cargo.toml
+++ b/avr-hal-generic/Cargo.toml
@@ -18,3 +18,6 @@ features = ["unproven"]
 [dependencies.void]
 version = "1.0.2"
 default-features = false
+
+[build-dependencies]
+rustversion = "1.0"
diff --git a/avr-hal-generic/build.rs b/avr-hal-generic/build.rs
new file mode 100644
index 0000000000..b22938833f
--- /dev/null
+++ b/avr-hal-generic/build.rs
@@ -0,0 +1,16 @@
+fn main() {
+    println!("cargo:rerun-if-changed=build.rs");
+
+    maybe_enable_asm();
+}
+
+#[rustversion::before(1.59.0)]
+fn maybe_enable_asm() {
+    //
+}
+
+#[rustversion::since(1.59.0)]
+fn maybe_enable_asm() {
+    // https://github.com/rust-lang/rust/pull/92816
+    println!("cargo:rustc-cfg=avr_hal_asm_macro");
+}
diff --git a/avr-hal-generic/src/delay.rs b/avr-hal-generic/src/delay.rs
index cac1df5c95..1e413eb6a1 100644
--- a/avr-hal-generic/src/delay.rs
+++ b/avr-hal-generic/src/delay.rs
@@ -3,6 +3,9 @@
 use core::marker;
 use hal::blocking::delay;
 
+#[cfg(all(target_arch = "avr", avr_hal_asm_macro))]
+use core::arch::asm;
+
 /// A busy-loop delay implementation
 ///
 /// # Example
@@ -32,7 +35,18 @@ impl<SPEED> Delay<SPEED> {
 // based on https://github.com/arduino/ArduinoCore-avr/blob/master/cores/arduino/wiring.c
 
 cfg_if::cfg_if! {
-    if #[cfg(target_arch = "avr")] {
+    if #[cfg(all(target_arch = "avr", avr_hal_asm_macro))] {
+        fn busy_loop(c: u16) {
+            unsafe {
+                asm!(
+                    "1:",
+                    "sbiw {c}, 1",
+                    "brne 1b",
+                    c = in(reg_iw) c,
+                );
+            }
+        }
+    } else if #[cfg(target_arch = "avr")] {
         #[allow(unused_assignments)]
         fn busy_loop(mut c: u16) {
             unsafe {
@@ -81,9 +95,15 @@ impl delay::DelayUs<u16> for Delay<crate::clock::MHz20> {
 
         // for a one-microsecond delay, simply return.  the overhead
         // of the function call takes 18 (20) cycles, which is 1us
+        #[cfg(all(target_arch = "avr", avr_hal_asm_macro))]
+        unsafe {
+            asm!("nop", "nop", "nop", "nop");
+        }
+
+        #[cfg(all(target_arch = "avr", not(avr_hal_asm_macro)))]
         unsafe {
             llvm_asm!("nop\nnop\nnop\nnop" :::: "volatile");
-        } //just waiting 4 cycles
+        }
 
         if us <= 1 {
             return;
@@ -238,7 +258,7 @@ where
     }
 }
 
-impl <SPEED> delay::DelayMs<u8> for Delay<SPEED>
+impl<SPEED> delay::DelayMs<u8> for Delay<SPEED>
 where
     Delay<SPEED>: delay::DelayMs<u16>,
 {
diff --git a/avr-hal-generic/src/lib.rs b/avr-hal-generic/src/lib.rs
index 590ef7d1a9..1814e80234 100644
--- a/avr-hal-generic/src/lib.rs
+++ b/avr-hal-generic/src/lib.rs
@@ -1,5 +1,6 @@
 #![no_std]
-#![feature(llvm_asm)]
+#![cfg_attr(avr_hal_asm_macro, feature(asm_experimental_arch))]
+#![cfg_attr(not(avr_hal_asm_macro), feature(llvm_asm))]
 
 pub extern crate embedded_hal as hal;