Optimize clearing framebuffer

Author: jdreaver

README.md

@@ -106,8 +106,10 @@ make test
 
 ## TODO
 
-- framebuffer crate removal
-  - Fix page fault caused by `write-framebuffer hello` (I think page table changes broke this)
+- Fix page fault caused by `write-framebuffer hello` (I think page table changes broke this)
+  - It is just when `framebuffer.clear();` is called. When that is commented out stuff works fine.
+  - The address of the page fault changes. It this caused by some kind of interrupt during the clear?
+    - Are interrupts + task switches broken in kernel code? I would expect the prime number thing to break if so. Maybe just the shell for some reason?
 - VFS read/write code:
   - Change inode `write` API to be similar to read (based on blocks)
   - Nuke old block iteration code in ext2 now that write and read don't need it

kernel/src/graphics/framebuffer.rs

@@ -139,17 +139,11 @@ impl VESAFramebuffer32Bit {
     }
 
     pub(super) fn clear(&mut self) {
-        // for i in 0..(self.pitch * self.height_pixels) {
-        //     unsafe {
-        //         *self.address.add(i) = 0x00;
-        //     }
-        // }
-
-        // This is faster in debug mode, and it is neat, so I'm keeping it
-        // around.
         unsafe {
-            core::slice::from_raw_parts_mut(self.address, self.pitch * self.height_pixels)
-                .fill(0x00);
+            // N.B. `write_bytes` is a highly optimized way to write zeroes.
+            // Making a slice and doing `slice.fill(0)` is supposed to optimize
+            // to this, but it doesn't seem to when compiling in debug mode.
+            self.address.write_bytes(0, self.pitch * self.height_pixels);
         };
     }
 }