Only perform a single TLB flush after identity mapping

This commit changes the BIOS bootloader to only flush the TLB once,
after it has identity-mapped every physical memory frame. This should be
a bit more efficient, as we don't perform a separate `invlpg` for every
page table entry we create, and instead only flush the entire thing by
reloading `CR3` when we're actually ready to use it.

This is based on a suggestion by @phil-opp, here:
rust-osdev#260 (comment)

This change doesn't actually seem to make all that big an impact in boot
times on QEMU v7.1 on its own, relative to PR rust-osdev#260, but it might make an
additional improvement on top of that PR.

src/bin/bios.rs

@@ -109,19 +109,28 @@ fn bootloader_main(
             PhysFrame::containing_address(PhysAddr::new(4096 * 512 * 512));
         let end_frame = PhysFrame::containing_address(PhysAddr::new(max_phys_addr - 1));
         for frame in PhysFrame::range_inclusive(start_frame, end_frame) {
-            unsafe {
+            let flusher = unsafe {
                 bootloader_page_table
                     .identity_map(
                         frame,
                         PageTableFlags::PRESENT | PageTableFlags::WRITABLE,
                         &mut frame_allocator,
                     )
                     .unwrap()
-                    .flush()
             };
+            // skip flushing the entry from the TLB for now, as we will
+            // flush the entire TLB at the end of the loop.
+            flusher.ignore();
         }
     }
 
+    // once all the physical memory is mapped, flush the TLB by reloading the
+    // CR3 register.
+    //
+    // we perform a single flush here rather than flushing each individual entry as
+    // it's mapped using `invlpg`, for efficiency.
+    x86_64::instructions::tlb::flush_all();
+
     let framebuffer_addr = PhysAddr::new(unsafe { VBEModeInfo_physbaseptr }.into());
     let mut error = None;
     let framebuffer_info = unsafe {