Merge #652

652: Remove use of basepri register on thumbv8m.base r=AfoHT a=neonquill

The basepri register appears to be aviable on thumbv8m.main but not thumbv8m.base. At the very least, attempting to compile against a Cortex-M23 based Microchip ATSAML10E16A generates an error:

```
error[E0432]: unresolved import `cortex_m::register::basepri`
  --> /Users/dwatson/.cargo/registry/src/github.com-1ecc6299db9ec823/cortex-m-rtic-1.1.3/src/export.rs:25:5
   |
25 | use cortex_m::register::basepri;
   |     ^^^^^^^^^^^^^^^^^^^^^^^^^^^ no `basepri` in `register`
```

I wasn't sure if it made more sense to replace the `armv7m` config flag with something related to basepri availability or to get closer to matching the cortex-m use of several architecture specific flags. In the end i chose to make the minimal change possible and just narrowed the existing `thumbv8m` check.

Context:
[cortex-m:src/register/mod.rs](https://github.com/rust-embedded/cortex-m/blob/4e908625204a1e95dd3fd5bdcd8d66d6bc11c3bc/src/register/mod.rs#L33):
```
#[cfg(all(not(armv6m), not(armv8m_base)))]
pub mod basepri;
```

[cortex-m:build.rs](https://github.com/rust-embedded/cortex-m/blob/4e908625204a1e95dd3fd5bdcd8d66d6bc11c3bc/build.rs#L21):
```
    } else if target.starts_with("thumbv8m.base") {
        println!("cargo:rustc-cfg=cortex_m");
        println!("cargo:rustc-cfg=armv8m");
        println!("cargo:rustc-cfg=armv8m_base");
```

Co-authored-by: David Watson <david@neonquill.com>

6 files changed, 144 insertions, 56 deletions

diff --git a/CHANGELOG.md b/CHANGELOG.md
index b131b307..7d1a9f3d 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -13,6 +13,8 @@ For each category, *Added*, *Changed*, *Fixed* add new entries at the top!
 
 ### Fixed
 
+- Distinguish between thumbv8m.base and thumbv8m.main for basepri usage.
+
 ### Changed
 
 ## [v1.1.3] - 2022-06-23
diff --git a/build.rs b/build.rs
index 7c518c90..ff9ebe35 100644
--- a/build.rs
+++ b/build.rs
@@ -7,15 +7,21 @@ fn main() {
         println!("cargo:rustc-cfg=rustc_is_nightly");
     }
 
-    if target.starts_with("thumbv6m") {
-        println!("cargo:rustc-cfg=armv6m");
-    }
-
+    // These targets all have know support for the BASEPRI register.
     if target.starts_with("thumbv7m")
         | target.starts_with("thumbv7em")
-        | target.starts_with("thumbv8m")
+        | target.starts_with("thumbv8m.main")
+    {
+        println!("cargo:rustc-cfg=have_basepri");
+
+    // These targets are all known to _not_ have the BASEPRI register.
+    } else if target.starts_with("thumb")
+        && !(target.starts_with("thumbv6m") | target.starts_with("thumbv8m.base"))
     {
-        println!("cargo:rustc-cfg=armv7m");
+        panic!(
+            "Unknown target '{}'. Need to update BASEPRI logic in build.rs.",
+            target
+        );
     }
 
     println!("cargo:rerun-if-changed=build.rs");
diff --git a/macros/src/codegen/assertions.rs b/macros/src/codegen/assertions.rs
index f6a098b5..66e54095 100644
--- a/macros/src/codegen/assertions.rs
+++ b/macros/src/codegen/assertions.rs
@@ -22,15 +22,16 @@ pub fn codegen(app: &App, analysis: &Analysis, extra: &Extra) -> Vec<TokenStream
     }
 
     let device = &extra.device;
-    let arm_v6_checks: Vec<_> = app
+    let chunks_name = util::priority_mask_chunks_ident();
+    let no_basepri_checks: Vec<_> = app
         .hardware_tasks
         .iter()
         .filter_map(|(_, task)| {
             if !util::is_exception(&task.args.binds) {
                 let interrupt_name = &task.args.binds;
                 Some(quote!(
-                    if (#device::Interrupt::#interrupt_name as u32) > 31 {
-                        ::core::panic!("An interrupt above value 31 is used while in armv6");
+                    if (#device::Interrupt::#interrupt_name as usize) >= (#chunks_name * 32) {
+                        ::core::panic!("An interrupt out of range is used while in armv6 or armv8m.base");
                     }
                 ))
             } else {
@@ -41,8 +42,8 @@ pub fn codegen(app: &App, analysis: &Analysis, extra: &Extra) -> Vec<TokenStream
 
     let const_check = quote! {
         const _CONST_CHECK: () = {
-            if rtic::export::is_armv6() {
-                #(#arm_v6_checks)*
+            if !rtic::export::have_basepri() {
+                #(#no_basepri_checks)*
             } else {
                 // TODO: Add armv7 checks here
             }
diff --git a/macros/src/codegen/shared_resources.rs b/macros/src/codegen/shared_resources.rs
index 08d77cc3..4a750070 100644
--- a/macros/src/codegen/shared_resources.rs
+++ b/macros/src/codegen/shared_resources.rs
@@ -118,6 +118,8 @@ pub fn codegen(
     let device = &extra.device;
     let mut uses_exceptions_with_resources = false;
 
+    let mut mask_ids = Vec::new();
+
     for (&priority, name) in interrupt_ids.chain(app.hardware_tasks.values().flat_map(|task| {
         if !util::is_exception(&task.args.binds) {
             Some((&task.args.priority, &task.args.binds))
@@ -147,12 +149,13 @@ pub fn codegen(
     })) {
         let v = prio_to_masks.entry(priority - 1).or_insert(Vec::new());
         v.push(quote!(#device::Interrupt::#name as u32));
+        mask_ids.push(quote!(#device::Interrupt::#name as u32));
     }
 
-    // Call rtic::export::create_mask([u32; N]), where the array is the list of shifts
+    // Call rtic::export::create_mask([Mask; N]), where the array is the list of shifts
 
     let mut mask_arr = Vec::new();
-    // NOTE: 0..3 assumes max 4 priority levels according to M0 spec
+    // NOTE: 0..3 assumes max 4 priority levels according to M0, M23 spec
     for i in 0..3 {
         let v = if let Some(v) = prio_to_masks.get(&i) {
             v.clone()
@@ -165,18 +168,26 @@ pub fn codegen(
         ));
     }
 
+    // Generate a constant for the number of chunks needed by Mask.
+    let chunks_name = util::priority_mask_chunks_ident();
+    mod_app.push(quote!(
+        #[doc(hidden)]
+        #[allow(non_upper_case_globals)]
+        const #chunks_name: usize = rtic::export::compute_mask_chunks([#(#mask_ids),*]);
+    ));
+
     let masks_name = util::priority_masks_ident();
     mod_app.push(quote!(
         #[doc(hidden)]
         #[allow(non_upper_case_globals)]
-        const #masks_name: [u32; 3] = [#(#mask_arr),*];
+        const #masks_name: [rtic::export::Mask<#chunks_name>; 3] = [#(#mask_arr),*];
     ));
 
     if uses_exceptions_with_resources {
         mod_app.push(quote!(
             #[doc(hidden)]
             #[allow(non_upper_case_globals)]
-            const __rtic_internal_V6_ERROR: () = rtic::export::v6_panic();
+            const __rtic_internal_V6_ERROR: () = rtic::export::no_basepri_panic();
         ));
     }
 
diff --git a/macros/src/codegen/util.rs b/macros/src/codegen/util.rs
index 0f3dca7c..0a3edc20 100644
--- a/macros/src/codegen/util.rs
+++ b/macros/src/codegen/util.rs
@@ -253,6 +253,11 @@ pub fn static_shared_resource_ident(name: &Ident) -> Ident {
     mark_internal_name(&format!("shared_resource_{}", name))
 }
 
+/// Generates an Ident for the number of 32 bit chunks used for Mask storage.
+pub fn priority_mask_chunks_ident() -> Ident {
+    mark_internal_name("MASK_CHUNKS")
+}
+
 pub fn priority_masks_ident() -> Ident {
     mark_internal_name("MASKS")
 }
diff --git a/src/export.rs b/src/export.rs
index a4d76b53..6f2a1b63 100644
--- a/src/export.rs
+++ b/src/export.rs
@@ -21,10 +21,43 @@ pub use rtic_monotonic as monotonic;
 pub type SCFQ<const N: usize> = Queue<u8, N>;
 pub type SCRQ<T, const N: usize> = Queue<(T, u8), N>;
 
-#[cfg(armv7m)]
+/// Mask is used to store interrupt masks on systems without a BASEPRI register (M0, M0+, M23).
+/// It needs to be large enough to cover all the relevant interrupts in use.
+/// For M0/M0+ there are only 32 interrupts so we only need one u32 value.
+/// For M23 there can be as many as 480 interrupts.
+/// Rather than providing space for all possible interrupts, we just detect the highest interrupt in
+/// use at compile time and allocate enough u32 chunks to cover them.
+#[derive(Copy, Clone)]
+pub struct Mask<const M: usize>([u32; M]);
+
+impl<const M: usize> core::ops::BitOrAssign for Mask<M> {
+    fn bitor_assign(&mut self, rhs: Self) {
+        for i in 0..M {
+            self.0[i] |= rhs.0[i];
+        }
+    }
+}
+
+#[cfg(not(have_basepri))]
+impl<const M: usize> Mask<M> {
+    /// Set a bit inside a Mask.
+    const fn set_bit(mut self, bit: u32) -> Self {
+        let block = bit / 32;
+
+        if block as usize >= M {
+            panic!("Generating masks for thumbv6/thumbv8m.base failed! Are you compiling for thumbv6 on an thumbv7 MCU or using an unsupported thumbv8m.base MCU?");
+        }
+
+        let offset = bit - (block * 32);
+        self.0[block as usize] |= 1 << offset;
+        self
+    }
+}
+
+#[cfg(have_basepri)]
 use cortex_m::register::basepri;
 
-#[cfg(armv7m)]
+#[cfg(have_basepri)]
 #[inline(always)]
 pub fn run<F>(priority: u8, f: F)
 where
@@ -41,7 +74,7 @@ where
     }
 }
 
-#[cfg(not(armv7m))]
+#[cfg(not(have_basepri))]
 #[inline(always)]
 pub fn run<F>(_priority: u8, f: F)
 where
@@ -105,15 +138,15 @@ impl Priority {
 }
 
 /// Const helper to check architecture
-pub const fn is_armv6() -> bool {
-    #[cfg(not(armv6m))]
+pub const fn have_basepri() -> bool {
+    #[cfg(have_basepri)]
     {
-        false
+        true
     }
 
-    #[cfg(armv6m)]
+    #[cfg(not(have_basepri))]
     {
-        true
+        false
     }
 }
 
@@ -172,14 +205,14 @@ where
 /// Total OH of per task is max 2 clock cycles, negligible in practice
 /// but can in theory be fixed.
 ///
-#[cfg(armv7m)]
+#[cfg(have_basepri)]
 #[inline(always)]
-pub unsafe fn lock<T, R>(
+pub unsafe fn lock<T, R, const M: usize>(
     ptr: *mut T,
     priority: &Priority,
     ceiling: u8,
     nvic_prio_bits: u8,
-    _mask: &[u32; 3],
+    _mask: &[Mask<M>; 3],
     f: impl FnOnce(&mut T) -> R,
 ) -> R {
     let current = priority.get();
@@ -247,14 +280,14 @@ pub unsafe fn lock<T, R>(
 /// - Temporary lower exception priority
 ///
 /// These possible solutions are set goals for future work
-#[cfg(not(armv7m))]
+#[cfg(not(have_basepri))]
 #[inline(always)]
-pub unsafe fn lock<T, R>(
+pub unsafe fn lock<T, R, const M: usize>(
     ptr: *mut T,
     priority: &Priority,
     ceiling: u8,
     _nvic_prio_bits: u8,
-    masks: &[u32; 3],
+    masks: &[Mask<M>; 3],
     f: impl FnOnce(&mut T) -> R,
 ) -> R {
     let current = priority.get();
@@ -288,28 +321,38 @@ pub unsafe fn lock<T, R>(
     }
 }
 
-#[cfg(not(armv7m))]
+#[cfg(not(have_basepri))]
 #[inline(always)]
-fn compute_mask(from_prio: u8, to_prio: u8, masks: &[u32; 3]) -> u32 {
-    let mut res = 0;
+fn compute_mask<const M: usize>(from_prio: u8, to_prio: u8, masks: &[Mask<M>; 3]) -> Mask<M> {
+    let mut res = Mask([0; M]);
     masks[from_prio as usize..to_prio as usize]
         .iter()
-        .for_each(|m| res |= m);
+        .for_each(|m| res |= *m);
     res
 }
 
 // enables interrupts
-#[cfg(not(armv7m))]
+#[cfg(not(have_basepri))]
 #[inline(always)]
-unsafe fn set_enable_mask(mask: u32) {
-    (*NVIC::PTR).iser[0].write(mask)
+unsafe fn set_enable_mask<const M: usize>(mask: Mask<M>) {
+    for i in 0..M {
+        // This check should involve compile time constants and be optimized out.
+        if mask.0[i] != 0 {
+            (*NVIC::PTR).iser[i].write(mask.0[i]);
+        }
+    }
 }
 
 // disables interrupts
-#[cfg(not(armv7m))]
+#[cfg(not(have_basepri))]
 #[inline(always)]
-unsafe fn clear_enable_mask(mask: u32) {
-    (*NVIC::PTR).icer[0].write(mask)
+unsafe fn clear_enable_mask<const M: usize>(mask: Mask<M>) {
+    for i in 0..M {
+        // This check should involve compile time constants and be optimized out.
+        if mask.0[i] != 0 {
+            (*NVIC::PTR).icer[i].write(mask.0[i]);
+        }
+    }
 }
 
 #[inline]
@@ -318,36 +361,56 @@ pub fn logical2hw(logical: u8, nvic_prio_bits: u8) -> u8 {
     ((1 << nvic_prio_bits) - logical) << (8 - nvic_prio_bits)
 }
 
-#[cfg(not(armv6m))]
-pub const fn create_mask<const N: usize>(_: [u32; N]) -> u32 {
-    0
+#[cfg(have_basepri)]
+pub const fn create_mask<const N: usize, const M: usize>(_: [u32; N]) -> Mask<M> {
+    Mask([0; M])
 }
 
-#[cfg(armv6m)]
-pub const fn create_mask<const N: usize>(list_of_shifts: [u32; N]) -> u32 {
-    let mut mask = 0;
+#[cfg(not(have_basepri))]
+pub const fn create_mask<const N: usize, const M: usize>(list_of_shifts: [u32; N]) -> Mask<M> {
+    let mut mask = Mask([0; M]);
     let mut i = 0;
 
     while i < N {
         let shift = list_of_shifts[i];
         i += 1;
-
-        if shift > 31 {
-            panic!("Generating masks for thumbv6 failed! Are you compiling for thumbv6 on an thumbv7 MCU?");
-        }
-
-        mask |= 1 << shift;
+        mask = mask.set_bit(shift);
     }
 
     mask
 }
 
-#[cfg(not(armv6m))]
-pub const fn v6_panic() {
+#[cfg(have_basepri)]
+pub const fn compute_mask_chunks<const L: usize>(_: [u32; L]) -> usize {
+    0
+}
+
+/// Compute the number of u32 chunks needed to store the Mask value.
+/// On M0, M0+ this should always end up being 1.
+/// On M23 we will pick a number that allows us to store the highest index used by the code.
+/// This means the amount of overhead will vary based on the actually interrupts used by the code.
+#[cfg(not(have_basepri))]
+pub const fn compute_mask_chunks<const L: usize>(ids: [u32; L]) -> usize {
+    let mut max: usize = 0;
+    let mut i = 0;
+
+    while i < L {
+        let id = ids[i] as usize;
+        i += 1;
+
+        if id > max {
+            max = id;
+        }
+    }
+    (max + 32) / 32
+}
+
+#[cfg(have_basepri)]
+pub const fn no_basepri_panic() {
     // For non-v6 all is fine
 }
 
-#[cfg(armv6m)]
-pub const fn v6_panic() {
-    panic!("Exceptions with shared resources are not allowed when compiling for thumbv6. Use local resources or `#[lock_free]` shared resources");
+#[cfg(not(have_basepri))]
+pub const fn no_basepri_panic() {
+    panic!("Exceptions with shared resources are not allowed when compiling for thumbv6 or thumbv8m.base. Use local resources or `#[lock_free]` shared resources");
 }