riscv32::swap: fix stack push/pop asm

src/arch/riscv32.rs

@@ -44,7 +44,7 @@
 use core::mem;
 use stack::Stack;
 
-pub const STACK_ALIGNMENT: usize = 16;
+pub const STACK_ALIGNMENT: usize = 4096;
 
 #[derive(Debug, Clone, Copy)]
 #[repr(transparent)]
@@ -159,6 +159,12 @@ pub unsafe fn swap(arg: usize, new_sp: StackPointer,
     // Just pass a dummy pointer if we aren't linking the stack
     &mut dummy
   };
+  let stack_limit = if let Some(new_stack) = new_stack {
+    new_stack.limit() as *mut usize
+  } else {
+    // Stack should not reach all the way to address 0
+    0 as *mut usize
+  };
 
   #[naked]
   unsafe extern "C" fn trampoline() {
@@ -168,12 +174,15 @@ pub unsafe fn swap(arg: usize, new_sp: StackPointer,
         # from the stack; unwinder uses them to find the CFA of the caller,
         # and so they have to have the correct value immediately after the
         # call instruction that invoked the trampoline.
-        sw      fp, -12(sp)
-        sw      ra, -16(sp)
         addi    sp, sp, -16
+        sw      fp, 4(sp)
+        sw      ra, 0(sp)
         .cfi_offset fp, 4
         .cfi_offset ra, 0
 
+        # Perform environment call to modify the PMP regions
+        ecall
+
         # Link the call stacks together by writing the current stack bottom
         # address to the CFA slot in the new stack.
         sw      sp, 0(a3)
@@ -182,13 +191,12 @@ pub unsafe fn swap(arg: usize, new_sp: StackPointer,
         or      a1, zero, sp
         # Load stack pointer of the new context.
         or      sp, zero, a2
-        # Deallocate the 16 bytes
-        addi    sp, sp, 16
 
         # Restore frame pointer and link register of the new context.
         # Load frame and instruction pointers of the new context.
-        lw      fp, -12(sp)
-        lw      ra, -16(sp)
+        lw      fp, 4(sp)
+        lw      ra, 0(sp)
+        addi    sp, sp, 16
 
         # Return into the new context.
         jr      ra
@@ -209,6 +217,7 @@ pub unsafe fn swap(arg: usize, new_sp: StackPointer,
       "{a0}" (arg)
       "{a2}" (new_sp.0)
       "{a3}" (new_cfa)
+      "{a7}" (stack_limit)
     :/*"zero",*/"ra",/*"sp","gp","tp",*/"t0","t1","t2",
     /*"fp",*/"s1",/*"a0", "a1"*/"a2", "a3", "a4", "a5",
       "a6", "a7", "s2",  "s3",  "s4", "s5", "s6", "s7",