diff --git a/Cargo.toml b/Cargo.toml
index fa5e775c..bbb6254b 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -17,7 +17,6 @@ autoexamples = true
 links = "cortex-m-rt" # Prevent multiple versions of cortex-m-rt being linked
 
 [dependencies]
-r0 = "1.0"
 cortex-m-rt-macros = { path = "macros", version = "=0.6.11" }
 # Note: Do not depend on `cortex-m` here. This crate is used for testing `cortex-m`, so we need to
 # avoid pulling in multiple versions of `cortex-m`.
diff --git a/asm.S b/asm.S
new file mode 100644
index 00000000..62541298
--- /dev/null
+++ b/asm.S
@@ -0,0 +1,105 @@
+  .cfi_sections .debug_frame
+
+  # Notes for function attributes:
+  # .type and .thumb_func are _both_ required, otherwise the Thumb mode bit
+  # will not be set and an invalid vector table is generated.
+  # LLD requires that section flags are set explicitly.
+
+  .section .HardFaultTrampoline, "ax"
+  .global HardFaultTrampoline
+  .type HardFaultTrampoline,%function
+  .thumb_func
+  .cfi_startproc
+  # HardFault exceptions are bounced through this trampoline which grabs the
+  # stack pointer at the time of the exception and passes it to the user's
+  # HardFault handler in r0.
+HardFaultTrampoline:
+  # Depending on the stack mode in EXC_RETURN, fetch stack pointer from
+  # PSP or MSP.
+  mov r0, lr
+  mov r1, #4
+  tst r0, r1
+  bne 0f
+  mrs r0, MSP
+  b HardFault
+0:
+  mrs r0, PSP
+  b HardFault
+  .cfi_endproc
+  .size HardFaultTrampoline, . - HardFaultTrampoline
+
+  .section .Reset, "ax"
+  .global Reset
+  .type Reset,%function
+  .thumb_func
+  .cfi_startproc
+  # Main entry point after reset. This jumps to the user __pre_init function,
+  # which cannot be called from Rust code without invoking UB, then
+  # initialises RAM. If the target has an FPU, it is enabled. Finally, jumps
+  # to the user main function.
+Reset:
+  # ARMv6-M does not initialise LR, but many tools expect it to be 0xFFFF_FFFF
+  # when reaching the first call frame, so we set it at startup.
+  # ARMv7-M and above initialise LR to 0xFFFF_FFFF at reset.
+  ldr r4,=0xffffffff
+  mov lr,r4
+
+  # Run user pre-init code, which must be executed immediately after startup,
+  # before the potentially time-consuming memory initialisation takes place.
+  # Example use cases include disabling default watchdogs or enabling RAM.
+  bl __pre_init
+
+  # Restore LR after calling __pre_init (r4 is preserved by subroutines).
+  mov lr,r4
+
+  # Initialise .bss memory. `__sbss` and `__ebss` come from the linker script.
+  ldr r0,=__sbss
+  ldr r1,=__ebss
+  mov r2,#0
+0:
+  cmp r1, r0
+  beq 1f
+  stm r0!, {r2}
+  b 0b
+1:
+
+  # Initialise .data memory. `__sdata`, `__sidata`, and `__edata` come from the
+  # linker script. Copy from r2 into r0 until r0 reaches r1.
+  ldr r0,=__sdata
+  ldr r1,=__edata
+  ldr r2,=__sidata
+2:
+  cmp r1, r0
+  beq 3f
+  # load 1 word from r2 to r3, inc r2
+  ldm r2!, {r3}
+  # store 1 word from r3 to r0, inc r0
+  stm r0!, {r3}
+  b 2b
+3:
+
+#ifdef HAS_FPU
+  # Conditionally enable the FPU.
+  # Address of SCB.CPACR.
+  ldr r0, =0xE000ED88
+  # Enable access to CP10 and CP11 from both privileged and unprivileged mode.
+  ldr r1, =(0b1111 << 20)
+  # RMW.
+  ldr r2, [r0]
+  orr r2, r2, r1
+  str r2, [r0]
+  # Barrier is required on some processors.
+  dsb
+  isb
+#endif
+
+4:
+  # Jump to user main function. We use bl for the extended range, but the
+  # user main function may not return.
+  bl main
+
+  # Trap on return.
+  udf
+
+  .cfi_endproc
+  .size Reset, . - Reset
diff --git a/asm.s b/asm.s
deleted file mode 100644
index 58ed2741..00000000
--- a/asm.s
+++ /dev/null
@@ -1,67 +0,0 @@
-  .cfi_sections .debug_frame
-
-  # LLD requires that the section flags are explicitly set here
-  .section .HardFaultTrampoline, "ax"
-  .global HardFaultTrampoline
-  # .type and .thumb_func are both required; otherwise its Thumb bit does not
-  # get set and an invalid vector table is generated
-  .type HardFaultTrampoline,%function
-  .thumb_func
-  .cfi_startproc
-HardFaultTrampoline:
-  # depending on the stack mode in EXC_RETURN, fetch stack pointer from
-  # PSP or MSP
-  mov r0, lr
-  mov r1, #4
-  tst r0, r1
-  bne 0f
-  mrs r0, MSP
-  b HardFault
-0:
-  mrs r0, PSP
-  b HardFault
-  .cfi_endproc
-  .size HardFaultTrampoline, . - HardFaultTrampoline
-
-  .section .text.FpuTrampoline, "ax"
-  .global FpuTrampoline
-  # .type and .thumb_func are both required; otherwise its Thumb bit does not
-  # get set and an invalid vector table is generated
-  .type FpuTrampoline,%function
-  .thumb_func
-  .cfi_startproc
-  # This enables the FPU and jumps to the main function.
-FpuTrampoline:
-  # Address of SCB.CPACR.
-  ldr r0, =0xE000ED88
-  # Enable access to CP10 and CP11 from both privileged and unprivileged mode.
-  ldr r1, =(0b1111 << 20)
-  # RMW.
-  ldr r2, [r0]
-  orr r2, r2, r1
-  str r2, [r0]
-  # Barrier is required on some processors.
-  dsb
-  isb
-  # Hand execution over to `main`.
-  bl main
-  # Note: `main` must not return. `bl` is used only because it has a wider range than `b`.
-  .cfi_endproc
-  .size FpuTrampoline, . - FpuTrampoline
-
-  # ARMv6-M leaves LR in an unknown state on Reset
-  # this trampoline sets LR before it's pushed onto the stack by Reset
-  .section .PreResetTrampoline, "ax"
-  .global PreResetTrampoline
-  # .type and .thumb_func are both required; otherwise its Thumb bit does not
-  # get set and an invalid vector table is generated
-  .type PreResetTrampoline,%function
-  .thumb_func
-  .cfi_startproc
-PreResetTrampoline:
-  # set LR to the initial value used by the ARMv7-M (0xFFFF_FFFF)
-  ldr r0,=0xffffffff
-  mov lr,r0
-  b Reset
-  .cfi_endproc
-  .size PreResetTrampoline, . - PreResetTrampoline
diff --git a/assemble.sh b/assemble.sh
index cdb32054..b914fedd 100755
--- a/assemble.sh
+++ b/assemble.sh
@@ -9,21 +9,25 @@ crate=cortex-m-rt
 # remove existing blobs because otherwise this will append object files to the old blobs
 rm -f bin/*.a
 
-arm-none-eabi-as -march=armv6s-m asm.s -o bin/$crate.o
+arm-none-eabi-gcc -c -march=armv6s-m asm.S -o bin/$crate.o
 ar crs bin/thumbv6m-none-eabi.a bin/$crate.o
 
-arm-none-eabi-as -march=armv7-m asm.s -o bin/$crate.o
+arm-none-eabi-gcc -c -march=armv7-m asm.S -o bin/$crate.o
 ar crs bin/thumbv7m-none-eabi.a bin/$crate.o
 
-arm-none-eabi-as -march=armv7e-m asm.s -o bin/$crate.o
+arm-none-eabi-gcc -c -march=armv7e-m asm.S -o bin/$crate.o
 ar crs bin/thumbv7em-none-eabi.a bin/$crate.o
+
+arm-none-eabi-gcc -c -march=armv7e-m asm.S -DHAS_FPU -o bin/$crate.o
 ar crs bin/thumbv7em-none-eabihf.a bin/$crate.o
 
-arm-none-eabi-as -march=armv8-m.base asm.s -o bin/$crate.o
+arm-none-eabi-gcc -c -march=armv8-m.base asm.S -o bin/$crate.o
 ar crs bin/thumbv8m.base-none-eabi.a bin/$crate.o
 
-arm-none-eabi-as -march=armv8-m.main asm.s -o bin/$crate.o
+arm-none-eabi-gcc -c -march=armv8-m.main asm.S -o bin/$crate.o
 ar crs bin/thumbv8m.main-none-eabi.a bin/$crate.o
+
+arm-none-eabi-gcc -c -march=armv8-m.main -DHAS_FPU asm.S -o bin/$crate.o
 ar crs bin/thumbv8m.main-none-eabihf.a bin/$crate.o
 
 rm bin/$crate.o
diff --git a/bin/thumbv6m-none-eabi.a b/bin/thumbv6m-none-eabi.a
index 65684dad..3ac07773 100644
Binary files a/bin/thumbv6m-none-eabi.a and b/bin/thumbv6m-none-eabi.a differ
diff --git a/bin/thumbv7em-none-eabi.a b/bin/thumbv7em-none-eabi.a
index c4e1f47a..d38ee469 100644
Binary files a/bin/thumbv7em-none-eabi.a and b/bin/thumbv7em-none-eabi.a differ
diff --git a/bin/thumbv7em-none-eabihf.a b/bin/thumbv7em-none-eabihf.a
index c4e1f47a..a782dce7 100644
Binary files a/bin/thumbv7em-none-eabihf.a and b/bin/thumbv7em-none-eabihf.a differ
diff --git a/bin/thumbv7m-none-eabi.a b/bin/thumbv7m-none-eabi.a
index ed96942a..038af9fe 100644
Binary files a/bin/thumbv7m-none-eabi.a and b/bin/thumbv7m-none-eabi.a differ
diff --git a/bin/thumbv8m.base-none-eabi.a b/bin/thumbv8m.base-none-eabi.a
index f1c7734a..ad383aa9 100644
Binary files a/bin/thumbv8m.base-none-eabi.a and b/bin/thumbv8m.base-none-eabi.a differ
diff --git a/bin/thumbv8m.main-none-eabi.a b/bin/thumbv8m.main-none-eabi.a
index cb216dcf..ef6e77a8 100644
Binary files a/bin/thumbv8m.main-none-eabi.a and b/bin/thumbv8m.main-none-eabi.a differ
diff --git a/bin/thumbv8m.main-none-eabihf.a b/bin/thumbv8m.main-none-eabihf.a
index cb216dcf..44380fb3 100644
Binary files a/bin/thumbv8m.main-none-eabihf.a and b/bin/thumbv8m.main-none-eabihf.a differ
diff --git a/build.rs b/build.rs
index 20f32b88..6758d9ef 100644
--- a/build.rs
+++ b/build.rs
@@ -18,8 +18,6 @@ fn main() {
 
     let out_dir = PathBuf::from(env::var("OUT_DIR").unwrap());
 
-    has_fpu(&target);
-
     if target.starts_with("thumbv") {
         fs::copy(
             format!("bin/{}.a", target),
@@ -92,9 +90,3 @@ handlers.");
     println!("cargo:rerun-if-changed=build.rs");
     println!("cargo:rerun-if-changed=link.x.in");
 }
-
-fn has_fpu(target: &str) {
-    if target.ends_with("eabihf") {
-        println!("cargo:rustc-cfg=has_fpu");
-    }
-}
diff --git a/link.x.in b/link.x.in
index f1a921d0..78fa8255 100644
--- a/link.x.in
+++ b/link.x.in
@@ -23,8 +23,9 @@
 INCLUDE memory.x
 
 /* # Entry point = reset vector */
+EXTERN(__RESET_VECTOR);
+EXTERN(Reset);
 ENTRY(Reset);
-EXTERN(__RESET_VECTOR); /* depends on the `Reset` symbol */
 
 /* # Exception vectors */
 /* This is effectively weak aliasing at the linker level */
@@ -85,13 +86,15 @@ SECTIONS
   /* ### .text */
   .text _stext :
   {
-    /* place these 2 close to each other or the `b` instruction will fail to link */
-    *(.PreResetTrampoline);
     *(.Reset);
 
     *(.text .text.*);
+
+    /* The HardFaultTrampoline uses the `b` instruction to enter `HardFault`,
+       so must be placed close to it. */
     *(.HardFaultTrampoline);
     *(.HardFault.*);
+
     . = ALIGN(4); /* Pad .text to the alignment to workaround overlapping load section bug in old lld */
   } > FLASH
   . = ALIGN(4); /* Ensure __etext is aligned if something unaligned is inserted after .text */
diff --git a/src/lib.rs b/src/lib.rs
index ab4bc3f2..e5e290d7 100644
--- a/src/lib.rs
+++ b/src/lib.rs
@@ -192,11 +192,9 @@
 //!
 //! One will always find the following (unmangled) symbols in `cortex-m-rt` applications:
 //!
-//! - `Reset`. This is the reset handler. The microcontroller will executed this function upon
+//! - `Reset`. This is the reset handler. The microcontroller will execute this function upon
 //! booting. This function will call the user program entry point (cf. [`#[entry]`][attr-entry])
-//! using the `main` symbol so you may also find that symbol in your program; if you do, `main`
-//! will contain your application code. Some other times `main` gets inlined into `Reset` so you
-//! won't find it.
+//! using the `main` symbol so you will also find that symbol in your program.
 //!
 //! - `DefaultHandler`. This is the default handler. If not overridden using `#[exception] fn
 //! DefaultHandler(..` this will be an infinite loop.
@@ -212,8 +210,8 @@
 //! the initial value of the stack pointer; this is where the stack will be located -- the stack
 //! grows downwards towards smaller addresses.
 //!
-//! - `__RESET_VECTOR`. This is the reset vector, a pointer into the `Reset` handler. This vector is
-//! located in the `.vector_table` section after `__STACK_START`.
+//! - `__RESET_VECTOR`. This is the reset vector, a pointer to the `Reset` function. This vector
+//! is located in the `.vector_table` section after `__STACK_START`.
 //!
 //! - `__EXCEPTIONS`. This is the core exceptions portion of the vector table; it's an array of 14
 //! exception vectors, which includes exceptions like `HardFault` and `SysTick`. This array is
@@ -226,19 +224,11 @@
 //!
 //! - `__pre_init`. This is a function to be run before RAM is initialized. It defaults to an empty
 //! function. The function called can be changed by applying the [`#[pre_init]`][attr-pre_init]
-//! attribute to a function. The empty function is not optimized out by default, but if an empty
-//! function is passed to [`#[pre_init]`][attr-pre_init] the function call will be optimized out.
+//! attribute to a function.
 //!
 //! If you override any exception handler you'll find it as an unmangled symbol, e.g. `SysTick` or
 //! `SVCall`, in the output of `objdump`,
 //!
-//! If you are targeting the `thumbv7em-none-eabihf` target you'll also see a `ResetTrampoline`
-//! symbol in the output. To avoid the compiler placing FPU instructions before the FPU has been
-//! enabled (cf. `vpush`) `Reset` calls the function `ResetTrampoline` which is marked as
-//! `#[inline(never)]` and `ResetTrampoline` calls `main`. The compiler is free to inline `main`
-//! into `ResetTrampoline` but it can't inline `ResetTrampoline` into `Reset` -- the FPU is enabled
-//! in `Reset`.
-//!
 //! # Advanced usage
 //!
 //! ## Setting the program entry point
@@ -248,9 +238,9 @@
 //! guarantees.
 //!
 //! The `Reset` handler will call a symbol named `main` (unmangled) *after* initializing `.bss` and
-//! `.data`, and enabling the FPU (if the target is `thumbv7em-none-eabihf`). A function with the
-//! `entry` attribute will be set to have the export name "`main`"; in addition, its mutable
-//! statics are turned into safe mutable references (see [`#[entry]`][attr-entry] for details).
+//! `.data`, and enabling the FPU (if the target has an FPU). A function with the `entry` attribute
+//! will be set to have the export name "`main`"; in addition, its mutable statics are turned into
+//! safe mutable references (see [`#[entry]`][attr-entry] for details).
 //!
 //! The unmangled `main` symbol must have signature `extern "C" fn() -> !` or its invocation from
 //! `Reset`  will result in undefined behavior.
@@ -411,7 +401,7 @@
 //!         *(.ccmram .ccmram.*);
 //!         . = ALIGN(4);
 //!     } > CCMRAM
-//! } INSERT AFTER .bss;
+//! }
 //! ```
 //!
 //! You can then use something like this to place a variable into this specific section of memory:
@@ -438,7 +428,6 @@
 #![no_std]
 
 extern crate cortex_m_rt_macros as macros;
-extern crate r0;
 
 use core::fmt;
 use core::sync::atomic::{self, Ordering};
@@ -701,8 +690,8 @@ pub use macros::exception;
 ///
 /// # Safety
 ///
-/// The function will be called before static variables are initialized. Any access of static
-/// variables will result in undefined behavior.
+/// The function will be called before memory is initialized, as soon as possible after reset. Any
+/// access of memory, including any static variables, will result in undefined behavior.
 ///
 /// **Warning**: Due to [rvalue static promotion][rfc1414] static variables may be accessed whenever
 /// taking a reference to a constant. This means that even trivial expressions such as `&1` in the
@@ -919,66 +908,12 @@ pub fn heap_start() -> *mut u32 {
     unsafe { &mut __sheap }
 }
 
-/* Entry point */
+// Entry point is Reset.
 #[doc(hidden)]
 #[link_section = ".vector_table.reset_vector"]
 #[no_mangle]
-#[cfg(not(armv6m))]
 pub static __RESET_VECTOR: unsafe extern "C" fn() -> ! = Reset;
 
-#[doc(hidden)]
-#[link_section = ".vector_table.reset_vector"]
-#[no_mangle]
-#[cfg(armv6m)]
-pub static __RESET_VECTOR: unsafe extern "C" fn() -> ! = PreResetTrampoline;
-
-#[doc(hidden)]
-#[link_section = ".Reset"]
-#[no_mangle]
-pub unsafe extern "C" fn Reset() -> ! {
-    extern "C" {
-
-        // These symbols come from `link.x`
-        static mut __sbss: u32;
-        static mut __ebss: u32;
-
-        static mut __sdata: u32;
-        static mut __edata: u32;
-        static __sidata: u32;
-    }
-
-    extern "Rust" {
-        // This symbol will be provided by the user via `#[pre_init]`
-        fn __pre_init();
-    }
-
-    __pre_init();
-
-    // Initialize RAM
-    r0::zero_bss(&mut __sbss, &mut __ebss);
-    r0::init_data(&mut __sdata, &mut __edata, &__sidata);
-
-    #[allow(clippy::match_single_binding)]
-    match () {
-        #[cfg(not(has_fpu))]
-        () => {
-            extern "C" {
-                // This symbol will be provided by the user via `#[entry]`
-                fn main() -> !;
-            }
-            main()
-        }
-        #[cfg(has_fpu)]
-        () => {
-            extern "C" {
-                fn FpuTrampoline() -> !;
-            }
-
-            FpuTrampoline()
-        }
-    }
-}
-
 #[allow(unused_variables)]
 #[doc(hidden)]
 #[link_section = ".HardFault.default"]
@@ -1038,8 +973,7 @@ pub enum Exception {
 pub use self::Exception as exception;
 
 extern "C" {
-    #[cfg(armv6m)]
-    fn PreResetTrampoline() -> !;
+    fn Reset() -> !;
 
     fn NonMaskableInt();