diff --git a/.cargo/config b/.cargo/config
new file mode 100644
index 00000000..86281707
--- /dev/null
+++ b/.cargo/config
@@ -0,0 +1,2 @@
+[alias]
+xtask = "run --package xtask --bin xtask --"
diff --git a/.travis.yml b/.travis.yml
index bbe3d487..7f25eae8 100644
--- a/.travis.yml
+++ b/.travis.yml
@@ -72,7 +72,6 @@ before_install: set -e
 
 install:
   - bash ci/install.sh
-  - export PATH="$PATH:$PWD/gcc/bin"
 
 script:
   - bash ci/script.sh
diff --git a/Cargo.toml b/Cargo.toml
index 2e55f6ad..eddc6fab 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -21,6 +21,14 @@ volatile-register = "0.2.0"
 bitfield = "0.13.2"
 embedded-hal = "0.2.4"
 
+[features]
+cm7-r0p1 = []
+inline-asm = []
+linker-plugin-lto = []
+
+[workspace]
+members = ["xtask"]
+
 [package.metadata.docs.rs]
 targets = [
     "thumbv8m.main-none-eabihf",
@@ -31,7 +39,3 @@ targets = [
     "thumbv8m.base-none-eabi",
     "thumbv8m.main-none-eabi"
 ]
-
-[features]
-cm7-r0p1 = []
-inline-asm = []
diff --git a/asm-cm7-r0p1.s b/asm-cm7-r0p1.s
deleted file mode 100644
index 28c3384e..00000000
--- a/asm-cm7-r0p1.s
+++ /dev/null
@@ -1,35 +0,0 @@
-  .cfi_sections .debug_frame
-
-  .section .text.__basepri_max_cm7_r0p1
-  .global __basepri_max_cm7_r0p1
-  .syntax unified
-  .thumb_func
-  .cfi_startproc
-__basepri_max_cm7_r0p1:
-  mrs r1, PRIMASK
-  cpsid i
-  tst.w r1, #1
-  msr BASEPRI_MAX, r0
-  it ne
-  bxne lr
-  cpsie i
-  bx lr
-  .cfi_endproc
-  .size __basepri_max_cm7_r0p1, . - __basepri_max_cm7_r0p1
-
-  .section .text.__basepri_w_cm7_r0p1
-  .global __basepri_w_cm7_r0p1
-  .syntax unified
-  .thumb_func
-  .cfi_startproc
-__basepri_w_cm7_r0p1:
-  mrs r1, PRIMASK
-  cpsid i
-  tst.w r1, #1
-  msr BASEPRI, r0
-  it ne
-  bxne lr
-  cpsie i
-  bx lr
-  .cfi_endproc
-  .size __basepri_w_cm7_r0p1, . - __basepri_w_cm7_r0p1
diff --git a/asm-fpu.s b/asm-fpu.s
deleted file mode 100644
index 417d199a..00000000
--- a/asm-fpu.s
+++ /dev/null
@@ -1,21 +0,0 @@
-  .cfi_sections .debug_frame
-
-  .section .text.__get_FPSCR
-  .global __get_FPSCR
-  .thumb_func
-  .cfi_startproc
-__get_FPSCR:
-  vmrs r0, fpscr
-  bx lr
-  .cfi_endproc
-  .size __get_FPSCR, . - __get_FPSCR
-
-  .section .text.__set_FPSCR
-  .global __set_FPSCR
-  .thumb_func
-  .cfi_startproc
-__set_FPSCR:
-  vmsr fpscr, r0
-  bx lr
-  .cfi_endproc
-  .size __set_FPSCR, . - __set_FPSCR
diff --git a/asm-toolchain b/asm-toolchain
new file mode 100644
index 00000000..a36829b6
--- /dev/null
+++ b/asm-toolchain
@@ -0,0 +1 @@
+nightly-2020-08-26
diff --git a/asm-v7.s b/asm-v7.s
deleted file mode 100644
index 17d7110e..00000000
--- a/asm-v7.s
+++ /dev/null
@@ -1,78 +0,0 @@
-  .syntax unified
-  .cfi_sections .debug_frame
-
-  .section .text.__basepri_max
-  .global __basepri_max
-  .thumb_func
-  .cfi_startproc
-__basepri_max:
-  msr BASEPRI_MAX, r0
-  bx lr
-  .cfi_endproc
-  .size __basepri_max, . - __basepri_max
-
-  .section .text.__basepri_r
-  .global __basepri_r
-  .thumb_func
-  .cfi_startproc
-__basepri_r:
-  mrs r0, BASEPRI
-  bx lr
-  .cfi_endproc
-  .size __basepri_r, . - __basepri_r
-
-  .section .text.__basepri_w
-  .global __basepri_w
-  .thumb_func
-  .cfi_startproc
-__basepri_w:
-  msr BASEPRI, r0
-  bx lr
-  .cfi_endproc
-  .size __basepri_w, . - __basepri_w
-
-  .section .text.__faultmask
-  .global __faultmask
-  .thumb_func
-  .cfi_startproc
-__faultmask:
-  mrs r0, FAULTMASK
-  bx lr
-  .cfi_endproc
-  .size __faultmask, . - __faultmask
-
-  .section .text.__enable_icache
-  .global __enable_icache
-  .thumb_func
-  .cfi_startproc
-__enable_icache:
-  ldr r0, =0xE000ED14       @ CCR
-  mrs r2, PRIMASK           @ save critical nesting info
-  cpsid i                   @ mask interrupts
-  ldr r1, [r0]              @ read CCR
-  orr.w r1, r1, #(1 << 17)  @ Set bit 17, IC
-  str r1, [r0]              @ write it back
-  dsb                       @ ensure store completes
-  isb                       @ synchronize pipeline
-  msr PRIMASK, r2           @ unnest critical section
-  bx lr
-  .cfi_endproc
-  .size __enable_icache, . - __enable_icache
-
-  .section .text.__enable_dcache
-  .global __enable_dcache
-  .thumb_func
-  .cfi_startproc
-__enable_dcache:
-  ldr r0, =0xE000ED14       @ CCR
-  mrs r2, PRIMASK           @ save critical nesting info
-  cpsid i                   @ mask interrupts
-  ldr r1, [r0]              @ read CCR
-  orr.w r1, r1, #(1 << 16)  @ Set bit 16, DC
-  str r1, [r0]              @ write it back
-  dsb                       @ ensure store completes
-  isb                       @ synchronize pipeline
-  msr PRIMASK, r2           @ unnest critical section
-  bx lr
-  .cfi_endproc
-  .size __enable_dcache, . - __enable_dcache
diff --git a/asm-v8-main.s b/asm-v8-main.s
deleted file mode 100644
index 1fad1555..00000000
--- a/asm-v8-main.s
+++ /dev/null
@@ -1,43 +0,0 @@
-
-  .cfi_sections .debug_frame
-
-  .section .text.__msplim_r
-  .global __msplim_r
-  .thumb_func
-  .cfi_startproc
-__msplim_r:
-  mrs r0, MSPLIM
-  bx lr
-  .cfi_endproc
-  .size __msplim_r, . - __msplim_r
-
-  .section .text.__msplim_w
-  .global __msplim_w
-  .thumb_func
-  .cfi_startproc
-__msplim_w:
-  msr MSPLIM, r0
-  bx lr
-  .cfi_endproc
-  .size __msplim_w, . - __msplim_w
-
-  .section .text.__psplim_r
-  .global __psplim_r
-  .thumb_func
-  .cfi_startproc
-__psplim_r:
-  mrs r0, PSPLIM
-  bx lr
-  .cfi_endproc
-  .size __psplim_r, . - __psplim_r
-
-  .section .text.__psplim_w
-  .global __psplim_w
-  .thumb_func
-  .cfi_startproc
-__psplim_w:
-  msr PSPLIM, r0
-  bx lr
-  .cfi_endproc
-  .size __psplim_w, . - __psplim_w
-
diff --git a/asm-v8.s b/asm-v8.s
deleted file mode 100644
index 7d3a8c94..00000000
--- a/asm-v8.s
+++ /dev/null
@@ -1,42 +0,0 @@
-  .cfi_sections .debug_frame
-
-  .section .text.__tt
-  .global __tt
-  .thumb_func
-  .cfi_startproc
-__tt:
-  tt r0, r0
-  bx lr
-  .cfi_endproc
-  .size __tt, . - __tt
-
-  .section .text.__ttt
-  .global __ttt
-  .thumb_func
-  .cfi_startproc
-__ttt:
-  ttt r0, r0
-  bx lr
-  .cfi_endproc
-  .size __ttt, . - __ttt
-
-  .section .text.__tta
-  .global __tta
-  .thumb_func
-  .cfi_startproc
-__tta:
-  tta r0, r0
-  bx lr
-  .cfi_endproc
-  .size __tta, . - __tta
-
-
-  .section .text.__ttat
-  .global __ttat
-  .thumb_func
-  .cfi_startproc
-__ttat:
-  ttat r0, r0
-  bx lr
-  .cfi_endproc
-  .size __ttat, . - __ttat
diff --git a/asm.rs b/asm.rs
new file mode 100644
index 00000000..f5b0179b
--- /dev/null
+++ b/asm.rs
@@ -0,0 +1,346 @@
+//! Assembly stubs for the `cortex-m` crate.
+//!
+//! We use this file to precompile some assembly stubs into the static libraries you can find in
+//! `bin`. Apps using the `cortex-m` crate then link against those static libraries and don't need
+//! to build this file themselves.
+//!
+//! Nowadays the assembly stubs are no longer actual assembly files, but actually just this Rust
+//! file `asm.rs` that uses unstable inline assembly, coupled with the `xtask` tool to invoke rustc
+//! and build the files.
+//!
+//! Precompiling this to a static lib allows users to call assembly routines from stable Rust, but
+//! also perform [linker plugin LTO] with the precompiled artifacts to completely inline the
+//! assembly routines into their code, which brings the "outline assembly" on par with "real" inline
+//! assembly.
+//!
+//! For developers and contributors to `cortex-m`, this setup means that they don't have to install
+//! any binutils, assembler, or C compiler to hack on the crate. All they need is to run `cargo
+//! xtask assemble` to rebuild the archives from this file.
+//!
+//! Cool, right?
+//!
+//! # Rust version management
+//!
+//! Since inline assembly is still unstable, and we want to ensure that the created blobs are
+//! up-to-date in CI, we have to pin the nightly version we use for this. The nightly toolchain is
+//! stored in `asm-toolchain`.
+//!
+//! The `cargo xtask` automation will automatically install the `asm-toolchain` as well as all
+//! Cortex-M targets needed to generate the blobs.
+//!
+//! [linker plugin LTO]: https://doc.rust-lang.org/stable/rustc/linker-plugin-lto.html
+
+#![feature(asm)]
+#![no_std]
+#![crate_type = "staticlib"]
+#![deny(warnings)]
+
+#[no_mangle]
+pub unsafe extern "C" fn __bkpt() {
+    asm!("bkpt");
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn __control_r() -> u32 {
+    let r;
+    asm!("mrs {}, CONTROL", out(reg) r);
+    r
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn __control_w(w: u32) {
+    asm!("msr CONTROL, {}", in(reg) w);
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn __cpsid() {
+    asm!("cpsid i");
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn __cpsie() {
+    asm!("cpsie i");
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn __delay(cyc: u32) {
+    asm!("
+    1:
+        nop
+        subs {}, #1
+        bne 1b
+        // Branch to 1 instead of __delay does not generate R_ARM_THM_JUMP8 relocation, which breaks
+        // linking on the thumbv6m-none-eabi target
+    ", in(reg) cyc);
+}
+
+// FIXME do we need compiler fences here or should we expect them in the caller?
+#[no_mangle]
+pub unsafe extern "C" fn __dmb() {
+    asm!("dmb 0xF");
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn __dsb() {
+    asm!("dsb 0xF");
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn __isb() {
+    asm!("isb 0xF");
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn __msp_r() -> u32 {
+    let r;
+    asm!("mrs {}, MSP", out(reg) r);
+    r
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn __msp_w(val: u32) {
+    asm!("msr MSP, {}", in(reg) val);
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn __nop() {
+    // NOTE: This is a `pure` asm block, but applying that option allows the compiler to eliminate
+    // the nop entirely (or to collapse multiple subsequent ones). Since the user probably wants N
+    // nops when they call `nop` N times, let's not add that option.
+    asm!("nop");
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn __primask() -> u32 {
+    // FIXME: rename to __primask_r
+    let r;
+    asm!("mrs {}, PRIMASK", out(reg) r);
+    r
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn __psp_r() -> u32 {
+    let r;
+    asm!("mrs {}, PSP", out(reg) r);
+    r
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn __psp_w(val: u32) {
+    asm!("msr PSP, {}", in(reg) val);
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn __sev() {
+    asm!("sev");
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn __udf() {
+    asm!("udf #0");
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn __wfe() {
+    asm!("wfe");
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn __wfi() {
+    asm!("wfi");
+}
+
+#[cfg(armv7m)]
+pub mod v7m {
+    #[no_mangle]
+    pub unsafe extern "C" fn __basepri_max(val: u8) {
+        asm!("msr BASEPRI_MAX, {}", in(reg) val);
+    }
+
+    #[no_mangle]
+    pub unsafe extern "C" fn __basepri_r() -> u8 {
+        let r;
+        asm!("mrs {}, BASEPRI", out(reg) r);
+        r
+    }
+
+    #[no_mangle]
+    pub unsafe extern "C" fn __basepri_w(val: u8) {
+        asm!("msr BASEPRI, {}", in(reg) val);
+    }
+
+    #[no_mangle]
+    pub unsafe extern "C" fn __faultmask() -> u32 {
+        let r;
+        asm!("mrs {}, FAULTMASK", out(reg) r);
+        r
+    }
+
+    // FIXME: compiler_fences necessary?
+    #[no_mangle]
+    pub unsafe extern "C" fn __enable_icache() {
+        asm!(
+            "
+            ldr r0, =0xE000ED14       @ CCR
+            mrs r2, PRIMASK           @ save critical nesting info
+            cpsid i                   @ mask interrupts
+            ldr r1, [r0]              @ read CCR
+            orr.w r1, r1, #(1 << 17)  @ Set bit 17, IC
+            str r1, [r0]              @ write it back
+            dsb                       @ ensure store completes
+            isb                       @ synchronize pipeline
+            msr PRIMASK, r2           @ unnest critical section
+            ",
+            out("r0") _,
+            out("r1") _,
+            out("r2") _,
+        );
+    }
+
+    #[no_mangle]
+    pub unsafe extern "C" fn __enable_dcache() {
+        asm!(
+            "
+            ldr r0, =0xE000ED14       @ CCR
+            mrs r2, PRIMASK           @ save critical nesting info
+            cpsid i                   @ mask interrupts
+            ldr r1, [r0]              @ read CCR
+            orr.w r1, r1, #(1 << 16)  @ Set bit 16, DC
+            str r1, [r0]              @ write it back
+            dsb                       @ ensure store completes
+            isb                       @ synchronize pipeline
+            msr PRIMASK, r2           @ unnest critical section
+            ",
+            out("r0") _,
+            out("r1") _,
+            out("r2") _,
+        );
+    }
+}
+
+#[cfg(armv7em)]
+mod v7em {
+    #[no_mangle]
+    pub unsafe extern "C" fn __basepri_max_cm7_r0p1(val: u8) {
+        asm!(
+            "
+            mrs r1, PRIMASK
+            cpsid i
+            tst.w r1, #1
+            msr BASEPRI_MAX, {}
+            it ne
+            bxne lr
+            cpsie i
+            ",
+            in(reg) val,
+            out("r1") _,
+        );
+    }
+
+    #[no_mangle]
+    pub unsafe extern "C" fn __basepri_w_cm7_r0p1(val: u8) {
+        asm!(
+            "
+            mrs r1, PRIMASK
+            cpsid i
+            tst.w r1, #1
+            msr BASEPRI, {}
+            it ne
+            bxne lr
+            cpsie i
+            ",
+            in(reg) val,
+            out("r1") _,
+        );
+    }
+}
+
+/// Baseline and Mainline.
+#[cfg(armv8m)]
+pub mod v8m {
+    #[no_mangle]
+    pub unsafe extern "C" fn __tt(mut target: u32) -> u32 {
+        asm!("tt {target}, {target}", target = inout(reg) target);
+        target
+    }
+
+    #[no_mangle]
+    pub unsafe extern "C" fn __ttt(mut target: u32) -> u32 {
+        asm!("ttt {target}, {target}", target = inout(reg) target);
+        target
+    }
+
+    #[no_mangle]
+    pub unsafe extern "C" fn __tta(mut target: u32) -> u32 {
+        asm!("tta {target}, {target}", target = inout(reg) target);
+        target
+    }
+
+    #[no_mangle]
+    pub unsafe extern "C" fn __ttat(mut target: u32) -> u32 {
+        asm!("ttat {target}, {target}", target = inout(reg) target);
+        target
+    }
+}
+
+/// Mainline only.
+#[cfg(armv8m_main)]
+pub mod v8m_main {
+    #[no_mangle]
+    pub unsafe extern "C" fn __msplim_r() -> u32 {
+        let r;
+        asm!("mrs {}, MSPLIM", out(reg) r);
+        r
+    }
+
+    #[no_mangle]
+    pub unsafe extern "C" fn __msplim_w(val: u32) {
+        asm!("msr MSPLIM, {}", in(reg) val);
+    }
+
+    #[no_mangle]
+    pub unsafe extern "C" fn __psplim_r() -> u32 {
+        let r;
+        asm!("mrs {}, PSPLIM", out(reg) r);
+        r
+    }
+
+    #[no_mangle]
+    pub unsafe extern "C" fn __psplim_w(val: u32) {
+        asm!("msr PSPLIM, {}", in(reg) val);
+    }
+}
+
+/// All targets with FPU.
+#[cfg(has_fpu)]
+pub mod fpu {
+    #[no_mangle]
+    pub unsafe extern "C" fn __get_FPSCR() -> u32 {
+        let r;
+        asm!("vmrs {}, fpscr", out(reg) r);
+        r
+    }
+
+    #[no_mangle]
+    pub unsafe extern "C" fn __set_FPSCR(val: u32) {
+        asm!("vmsr fpscr, {}", in(reg) val);
+    }
+}
+
+/// We *must* define a panic handler here, even though nothing here should ever be able to panic.
+///
+/// We prove that nothing will ever panic by calling a function that doesn't exist. If the panic
+/// handler gets linked in, this causes a linker error. We always build this file with optimizations
+/// enabled, but even without them the panic handler should never be linked in.
+#[panic_handler]
+fn panic(_: &core::panic::PanicInfo) -> ! {
+    extern "C" {
+        #[link_name = "cortex-m internal error: panic handler not optimized out, please file an \
+        issue at https://github.com/rust-embedded/cortex-m"]
+        fn __cortex_m_should_not_panic() -> !;
+    }
+
+    unsafe {
+        __cortex_m_should_not_panic();
+    }
+}
diff --git a/asm.s b/asm.s
deleted file mode 100644
index 3d1a54bb..00000000
--- a/asm.s
+++ /dev/null
@@ -1,192 +0,0 @@
-  .cfi_sections .debug_frame
-
-  .section .text.__bkpt
-  .global __bkpt
-  .thumb_func
-  .cfi_startproc
-__bkpt:
-  bkpt
-  bx lr
-  .cfi_endproc
-  .size __bkpt, . - __bkpt
-
-  .section .text.__control_r
-  .global __control_r
-  .thumb_func
-  .cfi_startproc
-__control_r:
-  mrs r0, CONTROL
-  bx lr
-  .cfi_endproc
-  .size __control_r, . - __control_r
-
-  .section .text.__control_w
-  .global __control_w
-  .thumb_func
-  .cfi_startproc
-__control_w:
-  msr CONTROL, r0
-  bx lr
-  .cfi_endproc
-  .size __control_w, . - __control_w
-
-
-  .section .text.__cpsid
-  .global __cpsid
-  .thumb_func
-  .cfi_startproc
-__cpsid:
-  cpsid i
-  bx lr
-  .cfi_endproc
-  .size __cpsid, . - __cpsid
-
-  .section .text.__cpsie
-  .global __cpsie
-  .thumb_func
-  .cfi_startproc
-__cpsie:
-  cpsie i
-  bx lr
-  .cfi_endproc
-  .size __cpsie, . - __cpsie
-
-  .section .text.__delay
-  .global __delay
-  .syntax unified
-  .thumb_func
-  .cfi_startproc
-__delay:
-1:
-  nop
-  subs r0, #1
-  bne 1b  // Branch to 1 instead of __delay does not generate R_ARM_THM_JUMP8 relocation, which breaks linking on the thumbv6m-none-eabi target
-  bx lr
-  .cfi_endproc
-  .size __delay, . - __delay
-
-  .section .text.__dmb
-  .global __dmb
-  .thumb_func
-  .cfi_startproc
-__dmb:
-  dmb 0xF
-  bx lr
-  .cfi_endproc
-  .size __dmb, . - __dmb
-
-  .section .text.__dsb
-  .global __dsb
-  .thumb_func
-  .cfi_startproc
-__dsb:
-  dsb 0xF
-  bx lr
-  .cfi_endproc
-  .size __dsb, . - __dsb
-
-  .section .text.__isb
-  .global __isb
-  .thumb_func
-  .cfi_startproc
-__isb:
-  isb 0xF
-  bx lr
-  .cfi_endproc
-  .size __isb, . - __isb
-
-  .section .text.__msp_r
-  .global __msp_r
-  .thumb_func
-  .cfi_startproc
-__msp_r:
-  mrs r0, MSP
-  bx lr
-  .cfi_endproc
-  .size __msp_r, . - __msp_r
-
-  .section .text.__msp_w
-  .global __msp_w
-  .thumb_func
-__msp_w:
-  msr MSP, r0
-  bx lr
-  .size __msp_w, . - __msp_w
-
-  .section .text.__nop
-  .global __nop
-  .thumb_func
-  .cfi_startproc
-__nop:
-  bx lr
-  .cfi_endproc
-  .size __nop, . - __nop
-
-  .section .text.__primask
-  .global __primask
-  .thumb_func
-  .cfi_startproc
-__primask:
-  mrs r0, PRIMASK
-  bx lr
-  .cfi_endproc
-  .size __primask, . - __primask
-
-  .section .text.__psp_r
-  .global __psp_r
-  .thumb_func
-  .cfi_startproc
-__psp_r:
-  mrs r0, PSP
-  bx lr
-  .cfi_endproc
-  .size __psp_r, . - __psp_r
-
-  .section .text.__psp_w
-  .global __psp_w
-  .thumb_func
-__psp_w:
-  msr PSP, r0
-  bx lr
-  .size __psp_w, . - __psp_w
-
-  .section .text.__sev
-  .global __sev
-  .thumb_func
-  .cfi_startproc
-__sev:
-  sev
-  bx lr
-  .cfi_endproc
-  .size __sev, . - __sev
-
-
-  .section .text.__udf
-  .global __udf
-  .thumb_func
-  .cfi_startproc
-__udf:
-  udf
-  .cfi_endproc
-  .size __udf, . - __udf
-
-  .section .text.__wfe
-  .global __wfe
-  .thumb_func
-  .cfi_startproc
-__wfe:
-  wfe
-  bx lr
-  .cfi_endproc
-  .size __wfe, . - __wfe
-
-
-  .section .text.__wfi
-  .global __wfi
-  .thumb_func
-  .cfi_startproc
-__wfi:
-  wfi
-  bx lr
-  .cfi_endproc
-  .size __wfi, . - __wfi
diff --git a/assemble.sh b/assemble.sh
deleted file mode 100755
index 070481a7..00000000
--- a/assemble.sh
+++ /dev/null
@@ -1,47 +0,0 @@
-#!/usr/bin/env bash
-
-set -euxo pipefail
-
-# cflags taken from cc 1.0.22
-
-crate=cortex-m
-
-# remove existing blobs because otherwise this will append object files to the old blobs
-rm -f bin/*.a
-
-pwd=$(pwd)
-flags=(-g --debug-prefix-map "$pwd=.")
-
-arm-none-eabi-as "${flags[@]}" -march=armv6s-m asm.s -o bin/$crate.o
-ar crs bin/thumbv6m-none-eabi.a bin/$crate.o
-
-arm-none-eabi-as "${flags[@]}" -march=armv7-m asm.s -o bin/$crate.o
-arm-none-eabi-as "${flags[@]}" -march=armv7-m asm-v7.s -o bin/$crate-v7.o
-ar crs bin/thumbv7m-none-eabi.a bin/$crate.o bin/$crate-v7.o
-
-arm-none-eabi-as "${flags[@]}" -march=armv7e-m asm.s -o bin/$crate.o
-arm-none-eabi-as "${flags[@]}" -march=armv7e-m asm-fpu.s -mfpu=fpv4-sp-d16 -o bin/$crate-v7-fpu.o
-arm-none-eabi-as "${flags[@]}" -march=armv7e-m asm-cm7-r0p1.s -o bin/$crate-cm7-r0p1.o
-arm-none-eabi-as "${flags[@]}" -march=armv7e-m asm-v7.s -o bin/$crate-v7.o
-ar crs bin/thumbv7em-none-eabi.a bin/$crate.o bin/$crate-v7.o bin/$crate-cm7-r0p1.o
-ar crs bin/thumbv7em-none-eabihf.a bin/$crate.o bin/$crate-v7.o bin/$crate-cm7-r0p1.o bin/$crate-v7-fpu.o
-
-arm-none-eabi-as "${flags[@]}" -march=armv8-m.base asm.s -o bin/$crate.o
-arm-none-eabi-as "${flags[@]}" -march=armv8-m.base asm-v8.s -o bin/$crate-v8.o
-ar crs bin/thumbv8m.base-none-eabi.a bin/$crate.o bin/$crate-v8.o
-
-arm-none-eabi-as "${flags[@]}" -march=armv8-m.main asm.s -o bin/$crate.o
-arm-none-eabi-as "${flags[@]}" -march=armv8-m.main asm-v7.s -o bin/$crate-v7.o
-arm-none-eabi-as "${flags[@]}" -march=armv8-m.main asm-v8.s -o bin/$crate-v8.o
-arm-none-eabi-as "${flags[@]}" -march=armv8-m.main asm-v8-main.s -o bin/$crate-v8-main.o
-arm-none-eabi-as "${flags[@]}" -march=armv8-m.main asm-fpu.s -mfpu=fpv5-sp-d16 -o bin/$crate-v8-fpu.o
-ar crs bin/thumbv8m.main-none-eabi.a bin/$crate.o bin/$crate-v7.o bin/$crate-v8.o bin/$crate-v8-main.o
-ar crs bin/thumbv8m.main-none-eabihf.a bin/$crate.o bin/$crate-v7.o bin/$crate-v8.o bin/$crate-v8-main.o bin/$crate-v8-fpu.o
-
-rm bin/$crate.o
-rm bin/$crate-v7.o
-rm bin/$crate-v7-fpu.o
-rm bin/$crate-v8-fpu.o
-rm bin/$crate-cm7-r0p1.o
-rm bin/$crate-v8.o
-rm bin/$crate-v8-main.o
diff --git a/bin/thumbv6m-none-eabi-lto.a b/bin/thumbv6m-none-eabi-lto.a
new file mode 100644
index 00000000..4d2f02cf
Binary files /dev/null and b/bin/thumbv6m-none-eabi-lto.a differ
diff --git a/bin/thumbv6m-none-eabi.a b/bin/thumbv6m-none-eabi.a
index b2f72b4c..beedd73b 100644
Binary files a/bin/thumbv6m-none-eabi.a and b/bin/thumbv6m-none-eabi.a differ
diff --git a/bin/thumbv7em-none-eabi-lto.a b/bin/thumbv7em-none-eabi-lto.a
new file mode 100644
index 00000000..7b8a4f87
Binary files /dev/null and b/bin/thumbv7em-none-eabi-lto.a differ
diff --git a/bin/thumbv7em-none-eabi.a b/bin/thumbv7em-none-eabi.a
index 5d5c48fb..588e5cdb 100644
Binary files a/bin/thumbv7em-none-eabi.a and b/bin/thumbv7em-none-eabi.a differ
diff --git a/bin/thumbv7em-none-eabihf-lto.a b/bin/thumbv7em-none-eabihf-lto.a
new file mode 100644
index 00000000..4efadd80
Binary files /dev/null and b/bin/thumbv7em-none-eabihf-lto.a differ
diff --git a/bin/thumbv7em-none-eabihf.a b/bin/thumbv7em-none-eabihf.a
index 1570f837..fd08ba59 100644
Binary files a/bin/thumbv7em-none-eabihf.a and b/bin/thumbv7em-none-eabihf.a differ
diff --git a/bin/thumbv7m-none-eabi-lto.a b/bin/thumbv7m-none-eabi-lto.a
new file mode 100644
index 00000000..28e5860a
Binary files /dev/null and b/bin/thumbv7m-none-eabi-lto.a differ
diff --git a/bin/thumbv7m-none-eabi.a b/bin/thumbv7m-none-eabi.a
index 51ab7103..4559ad4a 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-lto.a b/bin/thumbv8m.base-none-eabi-lto.a
new file mode 100644
index 00000000..ee467925
Binary files /dev/null and b/bin/thumbv8m.base-none-eabi-lto.a differ
diff --git a/bin/thumbv8m.base-none-eabi.a b/bin/thumbv8m.base-none-eabi.a
index 33bcff24..26d25bd3 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-lto.a b/bin/thumbv8m.main-none-eabi-lto.a
new file mode 100644
index 00000000..e3aa0cf1
Binary files /dev/null and b/bin/thumbv8m.main-none-eabi-lto.a differ
diff --git a/bin/thumbv8m.main-none-eabi.a b/bin/thumbv8m.main-none-eabi.a
index 1b2418da..10bbf53f 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-lto.a b/bin/thumbv8m.main-none-eabihf-lto.a
new file mode 100644
index 00000000..1c25494b
Binary files /dev/null and b/bin/thumbv8m.main-none-eabihf-lto.a differ
diff --git a/bin/thumbv8m.main-none-eabihf.a b/bin/thumbv8m.main-none-eabihf.a
index 312dcd23..06517880 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 d53dea5c..dc9b3a02 100644
--- a/build.rs
+++ b/build.rs
@@ -7,10 +7,17 @@ fn main() {
     let name = env::var("CARGO_PKG_NAME").unwrap();
 
     if target.starts_with("thumb") {
+        let suffix = if env::var_os("CARGO_FEATURE_LINKER_PLUGIN_LTO").is_some() {
+            "-lto"
+        } else {
+            ""
+        };
+
         fs::copy(
-            format!("bin/{}.a", target),
+            format!("bin/{}{}.a", target, suffix),
             out_dir.join(format!("lib{}.a", name)),
-        ).unwrap();
+        )
+        .unwrap();
 
         println!("cargo:rustc-link-lib=static={}", name);
         println!("cargo:rustc-link-search={}", out_dir.display());
@@ -25,7 +32,7 @@ fn main() {
     } else if target.starts_with("thumbv7em-") {
         println!("cargo:rustc-cfg=cortex_m");
         println!("cargo:rustc-cfg=armv7m");
-        println!("cargo:rustc-cfg=armv7em");  // (not currently used)
+        println!("cargo:rustc-cfg=armv7em"); // (not currently used)
     } else if target.starts_with("thumbv8m.base") {
         println!("cargo:rustc-cfg=cortex_m");
         println!("cargo:rustc-cfg=armv8m");
diff --git a/check-blobs.sh b/check-blobs.sh
deleted file mode 100755
index 166b4a4c..00000000
--- a/check-blobs.sh
+++ /dev/null
@@ -1,21 +0,0 @@
-#!/usr/bin/env bash
-
-# Checks that the blobs are up to date with the committed assembly files
-
-set -euxo pipefail
-
-for lib in bin/*.a; do
-    filename=$(basename "$lib")
-    arm-none-eabi-objdump -Cd "$lib" > "bin/${filename%.a}.before"
-done
-
-./assemble.sh
-
-for lib in bin/*.a; do
-    filename=$(basename "$lib")
-    arm-none-eabi-objdump -Cd "$lib" > "bin/${filename%.a}.after"
-done
-
-for cksum in bin/*.after; do
-    diff -u "$cksum" "${cksum%.after}.before"
-done
diff --git a/ci/install.sh b/ci/install.sh
index 371cc8e4..fb3e52dc 100644
--- a/ci/install.sh
+++ b/ci/install.sh
@@ -6,10 +6,6 @@ main() {
             rustup target add $TARGET
             ;;
     esac
-
-    mkdir gcc
-
-    curl -L https://developer.arm.com/-/media/Files/downloads/gnu-rm/7-2018q2/gcc-arm-none-eabi-7-2018-q2-update-linux.tar.bz2?revision=bc2c96c0-14b5-4bb4-9f18-bceb4050fee7?product=GNU%20Arm%20Embedded%20Toolchain,64-bit,,Linux,7-2018-q2-update | tar --strip-components=1 -C gcc -xj
 }
 
 main
diff --git a/ci/script.sh b/ci/script.sh
index 7c309935..fc1e7547 100644
--- a/ci/script.sh
+++ b/ci/script.sh
@@ -29,7 +29,7 @@ main() {
     esac
 
     if [ $TARGET = x86_64-unknown-linux-gnu ]; then
-        ./check-blobs.sh
+        cargo xtask check-blobs
     fi
 
     if [ $TRAVIS_RUST_VERSION = nightly ]; then
diff --git a/src/lib.rs b/src/lib.rs
index 5dbb8f56..723816aa 100644
--- a/src/lib.rs
+++ b/src/lib.rs
@@ -24,6 +24,32 @@
 //!
 //! The disadvantage is that `inline-asm` requires a nightly toolchain.
 //!
+//! ## `cm7-r0p1`
+//!
+//! This feature enables workarounds for errata found on Cortex-M7 chips with revision r0p1. Some
+//! functions in this crate only work correctly on those chips if this Cargo feature is enabled
+//! (the functions are documented accordingly).
+//!
+//! ## `linker-plugin-lto`
+//!
+//! This feature links against prebuilt assembly blobs that are compatible with [Linker-Plugin LTO].
+//! This allows inlining assembly routines into the caller, even without the `inline-asm` feature,
+//! and works on stable Rust (but note the drawbacks below!).
+//!
+//! If you want to use this feature, you need to be aware of a few things:
+//!
+//! - You need to make sure that `-Clinker-plugin-lto` is passed to rustc. Please refer to the
+//!   [Linker-Plugin LTO] documentation for details.
+//!
+//! - You have to use a Rust version whose LLVM version is compatible with the toolchain in
+//!   `asm-toolchain`.
+//!
+//! - Due to a [Rust bug][rust-lang/rust#75940], this option does not work with optimization
+//!   levels `s` and `z`.
+//!
+//! [Linker-Plugin LTO]: https://doc.rust-lang.org/stable/rustc/linker-plugin-lto.html
+//! [rust-lang/rust#75940]: https://github.com/rust-lang/rust/issues/75940
+//!
 //! # Minimum Supported Rust Version (MSRV)
 //!
 //! This crate is guaranteed to compile on stable Rust 1.31 and up. It *might*
diff --git a/xtask/Cargo.toml b/xtask/Cargo.toml
new file mode 100644
index 00000000..9d35164d
--- /dev/null
+++ b/xtask/Cargo.toml
@@ -0,0 +1,9 @@
+[package]
+name = "xtask"
+version = "0.0.0"
+authors = ["The Cortex-M Team <cortex-m@teams.rust-embedded.org>"]
+edition = "2018"
+publish = false
+
+[dependencies]
+ar = "0.8.0"
diff --git a/xtask/src/main.rs b/xtask/src/main.rs
new file mode 100644
index 00000000..ab59f57b
--- /dev/null
+++ b/xtask/src/main.rs
@@ -0,0 +1,216 @@
+//! `cargo xtask` automation.
+//!
+//! Please refer to <https://github.com/matklad/cargo-xtask/> for an explanation of the concept.
+//!
+//! Also see the docs in `asm.rs`.
+
+use process::Stdio;
+use std::env::{self, current_dir};
+use std::{
+    collections::BTreeMap,
+    fs::{self, File},
+    process::{self, Command},
+};
+
+fn toolchain() -> String {
+    fs::read_to_string("asm-toolchain")
+        .unwrap()
+        .trim()
+        .to_string()
+}
+
+fn rustc() -> Command {
+    let mut cmd = Command::new("rustc");
+    cmd.arg(format!("+{}", toolchain()));
+    cmd
+}
+
+fn assemble_really(target: &str, cfgs: &[&str], plugin_lto: bool) {
+    let mut cmd = rustc();
+
+    // Set the codegen target.
+    cmd.arg("--target").arg(target);
+    // Set all the `--cfg` directives for the target.
+    cmd.args(cfgs.iter().map(|cfg| format!("--cfg={}", cfg)));
+
+    // We want some level of debuginfo to allow unwinding through the functions.
+    cmd.arg("-g");
+    // We always optimize the assembly shims. There's not really any reason not to.
+    cmd.arg("-O");
+
+    // rustc will usually add frame pointers by default to aid with debugging, but that is a high
+    // overhead for the tiny assembly routines.
+    cmd.arg("-Cforce-frame-pointers=no");
+
+    // We don't want any system-specific paths to show up since we ship the result to other users.
+    // Add `--remap-path-prefix $(pwd)=.`.
+    let mut dir = current_dir().unwrap().as_os_str().to_os_string();
+    dir.push("=.");
+    cmd.arg("--remap-path-prefix").arg(dir);
+
+    // We let rustc build a single object file, not a staticlib, since the latter pulls in loads of
+    // code that will never be used (`compiler_builtins` and `core::fmt`, etc.). We build the static
+    // archive by hand after compiling.
+    cmd.arg("--emit=obj");
+
+    if plugin_lto {
+        // Make artifacts compatible with Linker-Plugin LTO (and incompatible with everything else).
+        cmd.arg("-Clinker-plugin-lto");
+    }
+
+    let file_stub = if plugin_lto {
+        format!("{}-lto", target)
+    } else {
+        target.to_string()
+    };
+
+    let obj_file = format!("bin/{}.o", file_stub);
+
+    // Pass output and input file.
+    cmd.arg("-o").arg(&obj_file);
+    cmd.arg("asm.rs");
+
+    println!("{:?}", cmd);
+    let status = cmd.status().unwrap();
+    assert!(status.success());
+
+    // Archive `target.o` -> `bin/target.a`.
+    let mut builder = ar::Builder::new(File::create(format!("bin/{}.a", file_stub)).unwrap());
+
+    // Use `append`, not `append_path`, to avoid adding any filesystem metadata (modification times,
+    // etc.).
+    let file = fs::read(&obj_file).unwrap();
+    builder
+        .append(
+            &ar::Header::new(obj_file.as_bytes().to_vec(), file.len() as u64),
+            &*file,
+        )
+        .unwrap();
+
+    fs::remove_file(&obj_file).unwrap();
+}
+
+fn assemble(target: &str, cfgs: &[&str]) {
+    assemble_really(target, cfgs, false);
+    assemble_really(target, cfgs, true);
+}
+
+// `--target` -> `--cfg` list (mirrors what `build.rs` does).
+static TARGETS: &[(&str, &[&str])] = &[
+    ("thumbv6m-none-eabi", &[]),
+    ("thumbv7m-none-eabi", &["armv7m"]),
+    ("thumbv7em-none-eabi", &["armv7m", "armv7em"]),
+    ("thumbv7em-none-eabihf", &["armv7m", "armv7em", "has_fpu"]),
+    ("thumbv8m.base-none-eabi", &["armv8m", "armv8m_base"]),
+    (
+        "thumbv8m.main-none-eabi",
+        &["armv7m", "armv8m", "armv8m_main"],
+    ),
+    (
+        "thumbv8m.main-none-eabihf",
+        &["armv7m", "armv8m", "armv8m_main", "has_fpu"],
+    ),
+];
+
+fn assemble_blobs() {
+    let mut cmd = rustc();
+    cmd.arg("-V");
+    cmd.stdout(Stdio::null());
+    let status = cmd.status().unwrap();
+
+    if !status.success() {
+        let toolchain = toolchain();
+        println!(
+            "asm toolchain {} does not seem to be installed. installing it now.",
+            toolchain
+        );
+
+        let mut rustup = Command::new("rustup");
+        let status = rustup.arg("install").arg(&toolchain).status().unwrap();
+        assert!(status.success(), "rustup command failed: {:?}", rustup);
+
+        let mut rustup = Command::new("rustup");
+        let status = rustup
+            .arg("target")
+            .arg("add")
+            .args(TARGETS.iter().map(|(target, _)| *target))
+            .arg("--toolchain")
+            .arg(toolchain)
+            .status()
+            .unwrap();
+        assert!(status.success(), "rustup command failed: {:?}", rustup);
+    }
+
+    for (target, cfgs) in TARGETS {
+        println!("building artifacts for {}", target);
+        assemble(target, cfgs);
+    }
+}
+
+fn check_blobs() {
+    // Load each `.a` file in `bin` into memory.
+    let mut files_before = BTreeMap::new();
+    for entry in fs::read_dir("bin").unwrap() {
+        let entry = entry.unwrap();
+        if entry.path().extension().unwrap() == "a" {
+            files_before.insert(
+                entry
+                    .path()
+                    .file_name()
+                    .unwrap()
+                    .to_str()
+                    .unwrap()
+                    .to_string(),
+                fs::read(entry.path()).unwrap(),
+            );
+        }
+    }
+
+    assemble_blobs();
+
+    let mut files_after = BTreeMap::new();
+    for entry in fs::read_dir("bin").unwrap() {
+        let entry = entry.unwrap();
+        if entry.path().extension().unwrap() == "a" {
+            files_after.insert(
+                entry
+                    .path()
+                    .file_name()
+                    .unwrap()
+                    .to_str()
+                    .unwrap()
+                    .to_string(),
+                fs::read(entry.path()).unwrap(),
+            );
+        }
+    }
+
+    // Ensure they contain the same files.
+    let before = files_before.keys().collect::<Vec<_>>();
+    let after = files_after.keys().collect::<Vec<_>>();
+    assert_eq!(before, after);
+
+    for ((file, before), (_, after)) in files_before.iter().zip(files_after.iter()) {
+        if before != after {
+            panic!("{} differs between rebuilds", file);
+        }
+    }
+
+    println!("Blobs identical.");
+}
+
+fn main() {
+    let subcommand = env::args().skip(1).next();
+    match subcommand.as_deref() {
+        Some("assemble") => assemble_blobs(),
+        Some("check-blobs") => check_blobs(),
+        _ => {
+            eprintln!("usage: cargo xtask <subcommand>");
+            eprintln!();
+            eprintln!("subcommands:");
+            eprintln!("    assemble     Reassemble the pre-built artifacts");
+            eprintln!("    check-blobs  Check that the pre-built artifacts are up-to-date and reproducible");
+            process::exit(1);
+        }
+    }
+}