Use MSR to map framebuffer as WC (write-combining) = huge speed diff on real HW

2026-03-19 19:34:31 +01:00
parent 6a82d581fb
commit 424b4c4632
6 changed files with 133 additions and 2 deletions
@@ -202,9 +202,9 @@ void paging_init(struct boot_context boot_ctx)
    uint64_t fb_size = fb->pitch * fb->height;
    uint64_t fb_pages = (fb_size + PAGE_SIZE-1)/PAGE_SIZE;

-    // Map the framebuffer (with cache-disable & write-through)
+    // Map the framebuffer (PWT set, and no PCD means PAT1 [Write-Combining] for this region)
    for (uint64_t i=0; i<fb_pages; i++) {
-        paging_map_page(kernel_pml4, fb_virt+i*PAGE_SIZE, fb_phys+i*PAGE_SIZE, PTE_WRITABLE | PTE_PCD | PTE_PWT);
+        paging_map_page(kernel_pml4, fb_virt+i*PAGE_SIZE, fb_phys+i*PAGE_SIZE, PTE_WRITABLE | PTE_PWT);
        page_count++;
    }
    DEBUG("Mapped %u pages for framebuffer", page_count);