diff --git a/src/asm_runtime.rs b/src/asm_runtime.rs
index 506d439..e384ae2 100644
--- a/src/asm_runtime.rs
+++ b/src/asm_runtime.rs
@@ -21,7 +21,6 @@ const DMA_32_BIT_MEMCPY: DmaControl =
   DmaControl::new().with_transfer_32bit(true).with_enabled(true);
 
 const DMA3_OFFSET: usize = DMA3_SRC.as_usize() - 0x0400_0000;
-const IME_OFFSET: usize = IME.as_usize() - 0x0400_0000;
 const WAITCNT_OFFSET: usize = WAITCNT.as_usize() - 0x0400_0000;
 
 // Proc-macros can't see the target being built for, so we use this declarative
@@ -125,56 +124,43 @@ core::arch::global_asm! {
   mgba_logging_enable_request = const MGBA_LOGGING_ENABLE_REQUEST,
 }
 
+// This handler DOES NOT allow nested interrupts at this time.
 core::arch::global_asm! {
   bracer::put_fn_in_section!(".text.gba_rom_header"),
   ".global __runtime_irq_handler",
   // On Entry: r0 = 0x0400_0000 (mmio_base)
+  // We're allowed to use the usual C ABI registers.
   "__runtime_irq_handler:",
 
   force_a32!{
-    /* swap IME off, user can turn it back on if they want */
-    "add r12, r0, #{ime_offset}",
-    "mov r3, #0",
-    "swp r3, r3, [r12]",
-
-    /* Read/Update IE and IF */
-    "ldr r0, [r12, #-8]",
-    "and r0, r0, r0, LSR #16",
-    "strh r0, [r12, #-6]",
-
-    /* Read/Update BIOS_IF */
-    "sub  r2, r12, #(0x208+8)",
-    "ldrh r1, [r2]",
-    "orr  r1, r1, r0",
-    "strh r1, [r2]",
-
-    /* Call the Rust fn pointer (if set), using System mode */
-    "ldr r1, ={RUST_IRQ_HANDLER}",
-    "ldr r1, [r1]",
-    bracer::when!(("r1" != "#0")[9] {
-      // Note(Lokathor): We are *SKIPPING* the part where we ensure that the
-      // System stack pointer is aligned to 8 during the call to the rust
-      // function. This is *technically* against the AAPCS ABI, but the GBA's
-      // ARMv4T CPU does not even support any instructions that require an
-      // alignment of 8. By not bothering to align the stack, we save about 5
-      // cycles total. Which is neat, but if this were on the DS (which has an
-      // ARMv5TE CPU) you'd want to ensure the aligned stack.
-
-      bracer::a32_read_spsr_to!("r2"),
-      bracer::a32_set_cpu_control!(System, irq_masked= false, fiq_masked= false),
-      "push {{r2, r3, r12, lr}}",
-      bracer::a32_fake_blx!("r1"),
-      "pop {{r2, r3, r12, lr}}",
-      bracer::a32_set_cpu_control!(Supervisor, irq_masked= true, fiq_masked= false),
-      bracer::a32_write_spsr_from!("r2")
+    /* A fox wizard told me how to do this one */
+    // handle MMIO interrupt system
+    "mov  r12, 0x04000000",     // load r12 with a 1 cycle value
+    "ldr  r0, [r12, #0x200]!",  // load IE_IF with r12 writeback
+    "and  r0, r0, r0, LSR #16", // bits = IE & IF
+    "strh r0, [r12, #2]",       // write16 to just IF
+    // handle BIOS IntrWait system
+    "ldr  r1, [r12, #-0x208]!", // load BIOS_IF_?? with r12 writeback
+    "orr  r1, r1, r0",          // mark `bits` as `has_occurred`
+    "strh r1, [r12]",           // write16 to just BIOS_IF
+
+    // Get the rust code handler fn pointer, call it if non-null.
+    "ldr r12, ={RUST_IRQ_HANDLER}",
+    "ldr r12, [r12]",
+    bracer::when!(("r12" != "#0")[1] {
+      bracer::a32_read_spsr_to!("r3"),
+      "push {{r3, lr}}",
+      bracer::a32_set_cpu_control!(System, irq_masked = true, fiq_masked = true),
+      bracer::a32_fake_blx!("r12"),
+      bracer::a32_set_cpu_control!(IRQ, irq_masked = true, fiq_masked = true),
+      "pop {{r3, lr}}",
+      bracer::a32_write_spsr_from!("r3"),
     }),
 
-    /* Restore initial IME setting and return */
-    "swp r3, r3, [r12]",
+    // return to the BIOS
     "bx lr",
   },
 
   // Define Our Constants
-  ime_offset = const IME_OFFSET,
   RUST_IRQ_HANDLER = sym crate::RUST_IRQ_HANDLER,
 }