Move stack into the fastest DMA-able memory

Author: salkinium

src/modm/platform/core/cortex/ram.ld.in

@@ -8,7 +8,7 @@
 
 %% if vector_table_location == "ram"
 /* Round up the number of vector table entries to the nearest power-of-two and multiply by 8. */
-__vec_alignment = (1 << LOG2CEIL({{ number_of_interrupts + 16 }})) * 8;
+__vec_alignment = (1 << LOG2CEIL({{ highest_irq + 16 }})) * 8;
 /* compute the vector table offset from start of RAM */
 __vec_offset = ALIGN(TOTAL_STACK_SIZE, __vec_alignment);
 %% else

src/modm/platform/core/stm32/dccm.ld.in

@@ -6,6 +6,14 @@
 %% set table_copy = []
 %% set table_zero = []
 
+%% if vector_table_location == "ram"
+/* Round up the number of vector table entries to the nearest power-of-two and multiply by 8. */
+__vec_alignment = (1 << LOG2CEIL({{ highest_irq + 16 }})) * 8;
+/* compute the vector table offset from start of RAM */
+__vec_offset = ALIGN(TOTAL_STACK_SIZE, __vec_alignment);
+%% else
+__vec_offset = TOTAL_STACK_SIZE;
+%% endif
 
 SECTIONS
 {
@@ -16,6 +24,8 @@ SECTIONS
 {{ linker.section_rom("FLASH") }}
 
 
+{{ linker.section_stack(cont_ram_regions[0].name|upper, "__vec_offset - TOTAL_STACK_SIZE") }}
+
 %% if vector_table_location == "ram"
 {{ linker.section_vector_ram(cont_ram_regions[0].name|upper) }}
 %% endif
@@ -37,9 +47,6 @@ SECTIONS
 %% endif
 
 
-{{ linker.section_stack("CCM") }}
-
-	/* CCM can only be accessed by D-Bus, do not place .fastcode here! */
 {{ linker.section("CCM AT >FLASH", "fastdata", sections=["data_ccm"]) }}
 %% do table_copy.append("fastdata")
 

src/modm/platform/core/stm32/module.lb

@@ -70,6 +70,7 @@ def build(env):
 
 def post_build(env):
     env.substitutions = env.query("::cortex-m:linkerscript")
+    env.substitutions.update(env.query("::cortex-m:vector_table"))
     env.outbasepath = "modm/link"
 
     linkerscript = "../cortex/ram.ld.in"

src/modm/platform/core/stm32/module.md

@@ -105,6 +105,7 @@ SRAMs explicitly to free up the space in the lower sections.
 
 Some STM32F4 have a battery-backed backup SRAM and a single-cycle CCM that is
 only accessible to the core via the D-Code bus, thus the CCM is not DMA-able.
+Therefore the main stack is placed into SRAM, even though it is slower than CCM.
 
 ```
             ┌────────────────────────┐◄ __backup_end
@@ -133,17 +134,17 @@ only accessible to the core via the D-Code bus, thus the CCM is not DMA-able.
             │  .data_sram1           │
             │  .data                 │
             │  .fastcode             │
-   SRAM1    │ (.vector_ram)          │◄ only if remapped into RAM
+            │ (.vector_ram)          │◄ only if remapped into RAM
+            │  +PROCESS_STACK_SIZE   │◄ __process_stack_top
+   SRAM1    │  +MAIN_STACK_SIZE      │◄ __main_stack_top
 0x2000 0000 └────────────────────────┘◄ __sram1_start
 
             ┌────────────────────────┐◄ __ccm_end
-            │  +HEAP_CCM             │
-   D-Code   │  .noinit_ccm           │
-    only    │  .bss_ccm              │
-   access   │  .data_ccm             │
-            │  .fastdata             │
-            │  +PROCESS_STACK_SIZE   │◄ __process_stack_top
-    CCM     │  +MAIN_STACK_SIZE      │◄ __main_stack_top
+   D-Code   │  +HEAP_CCM             │
+    only    │  .noinit_ccm           │
+   access   │  .bss_ccm              │
+            │  .data_ccm             │
+    CCM     │  .fastdata             │
 0x1000 0000 └────────────────────────┘◄ __ccm_start
 
             ┌────────────────────────┐◄ __flash_end