Merge #322

322: rj/itcm r=jordens a=jordens

* close #315
* would profit from cortex-m-rtic 0.6 elevating the attributes to the actual ISR thus removing the veneer

Co-authored-by: Robert Jördens <rj@quartiq.de>

Cargo.lock

@@ -145,18 +145,15 @@ dependencies = [
 [[package]]
 name = "cortex-m-rt"
 version = "0.6.13"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "980c9d0233a909f355ed297ef122f257942de5e0a2cb1c39f60684b65bcb90fb"
+source = "git+https://github.com/rust-embedded/cortex-m-rt.git?rev=a2e3ad5#a2e3ad54478c6b98e519a1b0946395d790c0b6c7"
 dependencies = [
  "cortex-m-rt-macros",
- "r0",
 ]
 
 [[package]]
 name = "cortex-m-rt-macros"
-version = "0.1.8"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "4717562afbba06e760d34451919f5c3bf3ac15c7bb897e8b04862a7428378647"
+version = "0.6.11"
+source = "git+https://github.com/rust-embedded/cortex-m-rt.git?rev=a2e3ad5#a2e3ad54478c6b98e519a1b0946395d790c0b6c7"
 dependencies = [
  "proc-macro2",
  "quote",
@@ -588,12 +585,6 @@ dependencies = [
  "proc-macro2",
 ]
 
-[[package]]
-name = "r0"
-version = "0.2.2"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "e2a38df5b15c8d5c7e8654189744d8e396bddc18ad48041a500ce52d6948941f"
-
 [[package]]
 name = "rand"
 version = "0.8.3"

Cargo.toml

@@ -46,10 +46,12 @@ miniconf = "0.1.0"
 shared-bus = {version = "0.2.2", features = ["cortex-m"] }
 serde-json-core = "0.3"
 
+# rtt-target bump
 [dependencies.rtt-logger]
 git = "https://github.com/quartiq/rtt-logger.git"
 rev = "70b0eb5"
 
+# rewrite
 [dependencies.mcp23017]
 git = "https://github.com/lucazulian/mcp23017.git"
 rev = "523d71d"
@@ -58,6 +60,11 @@ rev = "523d71d"
 features = ["stm32h743v", "rt", "unproven", "ethernet", "quadspi"]
 version = "0.9.0"
 
+# link.x section start/end
+[patch.crates-io.cortex-m-rt]
+git = "https://github.com/rust-embedded/cortex-m-rt.git"
+rev = "a2e3ad5"
+
 [patch.crates-io.miniconf]
 git = "https://github.com/quartiq/miniconf.git"
 rev = "c6f2b28"

build.rs

@@ -0,0 +1,3 @@
+fn main() {
+    println!("cargo:rerun-if-changed=memory.x");
+}

memory.x

@@ -13,10 +13,6 @@ MEMORY
 }
 
 SECTIONS {
-  .itcm : ALIGN(8) {
-    *(.itcm .itcm.*);
-    . = ALIGN(8);
-    } > ITCM
   .axisram (NOLOAD) : ALIGN(8) {
     *(.axisram .axisram.*);
     . = ALIGN(8);
@@ -33,4 +29,18 @@ SECTIONS {
     *(.sram3 .sram3.*);
     . = ALIGN(4);
     } > SRAM3
-} INSERT AFTER .bss;
+  .itcm : ALIGN(8) {
+    . = ALIGN(8);
+    __sitcm = .;
+    *(.itcm .itcm.*);
+    . = ALIGN(8);
+    __eitcm = .;
+  } > ITCM AT>FLASH
+  __siitcm = LOADADDR(.itcm);
+} INSERT AFTER .uninit;
+
+ASSERT(__sitcm % 8 == 0 && __eitcm % 8 == 0, "
+BUG(cortex-m-rt): .itcm is not 8-byte aligned");
+
+ASSERT(__siitcm % 4 == 0, "
+BUG(cortex-m-rt): the LMA of .itcm is not 4-byte aligned");

src/bin/dual-iir.rs

@@ -120,6 +120,8 @@ const APP: () = {
     /// Because the ADC and DAC operate at the same rate, these two constraints actually implement
     /// the same time bounds, meeting one also means the other is also met.
     #[task(binds=DMA1_STR4, resources=[adcs, digital_inputs, dacs, iir_state, settings, telemetry], priority=2)]
+    #[inline(never)]
+    #[link_section = ".itcm.process"]
     fn process(c: process::Context) {
         let adc_samples = [
             c.resources.adcs.0.acquire_buffer(),

src/bin/lockin.rs

@@ -157,6 +157,8 @@ const APP: () = {
     /// It outputs either I/Q or power/phase on DAC0/DAC1. Data is normalized to full scale.
     /// PLL bandwidth, filter bandwidth, slope, and x/y or power/phase post-filters are available.
     #[task(binds=DMA1_STR4, resources=[adcs, dacs, lockin, timestamper, pll, settings, telemetry], priority=2)]
+    #[inline(never)]
+    #[link_section = ".itcm.process"]
     fn process(c: process::Context) {
         let adc_samples = [
             c.resources.adcs.0.acquire_buffer(),

src/hardware/configuration.rs

@@ -105,6 +105,52 @@ pub struct PounderDevices {
 /// Static storage for the ethernet DMA descriptor ring.
 static mut DES_RING: ethernet::DesRing = ethernet::DesRing::new();
 
+/// Setup ITCM and load its code from flash.
+///
+/// For portability and maintainability this is implemented in Rust.
+/// Since this is implemented in Rust the compiler may assume that bss and data are set
+/// up already. There is no easy way to ensure this implementation will never need bss
+/// or data. Hence we can't safely run this as the cortex-m-rt `pre_init` hook before
+/// bss/data is setup.
+///
+/// Calling (through IRQ or directly) any code in ITCM before having called
+/// this method is undefined.
+fn load_itcm() {
+    extern "C" {
+        static mut __sitcm: u32;
+        static mut __eitcm: u32;
+        static mut __siitcm: u32;
+    }
+    use core::{ptr, slice, sync::atomic};
+
+    // NOTE(unsafe): Assuming the address symbols from the linker as well as
+    // the source instruction data are all valid, this is safe as it only
+    // copies linker-prepared data to where the code expects it to be.
+    // Calling it multiple times is safe as well.
+
+    unsafe {
+        // ITCM is enabled on reset on our CPU but might not be on others.
+        // Keep for completeness.
+        const ITCMCR: *mut u32 = 0xE000_EF90usize as _;
+        ptr::write_volatile(ITCMCR, ptr::read_volatile(ITCMCR) | 1);
+
+        // Ensure ITCM is enabled before loading.
+        atomic::fence(atomic::Ordering::SeqCst);
+
+        let len =
+            (&__eitcm as *const u32).offset_from(&__sitcm as *const _) as usize;
+        let dst = slice::from_raw_parts_mut(&mut __sitcm as *mut _, len);
+        let src = slice::from_raw_parts(&__siitcm as *const _, len);
+        // Load code into ITCM.
+        dst.copy_from_slice(src);
+    }
+
+    // Ensure ITCM is loaded before potentially executing any instructions from it.
+    atomic::fence(atomic::Ordering::SeqCst);
+    cortex_m::asm::dsb();
+    cortex_m::asm::isb();
+}
+
 /// Configure the stabilizer hardware for operation.
 ///
 /// # Args
@@ -157,9 +203,12 @@ pub fn setup(
         log::set_logger(&LOGGER)
             .map(|()| log::set_max_level(log::LevelFilter::Trace))
             .unwrap();
-        log::info!("starting...");
+        log::info!("Starting");
     }
 
+    // Before being able to call any code in ITCM, load that code from flash.
+    load_itcm();
+
     // Set up the system timer for RTIC scheduling.
     {
         let tim15 =
@@ -872,6 +921,7 @@ pub fn setup(
 
         #[cfg(feature = "pounder_v1_1")]
         let pounder_stamper = {
+            log::info!("Assuming Pounder v1.1 or later");
             let etr_pin = gpioa.pa0.into_alternate_af3();
 
             // The frequency in the constructor is dont-care, as we will modify the period + clock
@@ -934,13 +984,13 @@ pub fn setup(
         digital_inputs,
     };
 
+    // Enable the instruction cache.
+    core.SCB.enable_icache();
+
     // info!("Version {} {}", build_info::PKG_VERSION, build_info::GIT_VERSION.unwrap());
     // info!("Built on {}", build_info::BUILT_TIME_UTC);
     // info!("{} {}", build_info::RUSTC_VERSION, build_info::TARGET);
     log::info!("setup() complete");
 
-    // Enable the instruction cache.
-    core.SCB.enable_icache();
-
     (stabilizer, pounder)
 }

src/hardware/mod.rs

@@ -90,11 +90,11 @@ fn panic(info: &core::panic::PanicInfo) -> ! {
 }
 
 #[cortex_m_rt::exception]
-fn HardFault(ef: &cortex_m_rt::ExceptionFrame) -> ! {
+unsafe fn HardFault(ef: &cortex_m_rt::ExceptionFrame) -> ! {
     panic!("HardFault at {:#?}", ef);
 }
 
 #[cortex_m_rt::exception]
-fn DefaultHandler(irqn: i16) {
+unsafe fn DefaultHandler(irqn: i16) {
     panic!("Unhandled exception (IRQn = {})", irqn);
 }