Skip to content

compositor-cpu-attack-plan.md

Latest commit

 

History

History
68 lines (46 loc) · 2.49 KB

compositor-cpu-attack-plan.md

File metadata and controls

68 lines (46 loc) · 2.49 KB

Compositor CPU Reduction Attack Plan

Baseline: BWM at 70% CPU, 155Hz loop, 6451us/iter, 13 syscalls/iter, 50% idle polling.

Priority 1: Event-Driven Compositor Wait [DONE]

Actual savings: ~14% CPU (34% -> 20%) Status: Merged (PR #254)

New syscall op=23 compositor_wait(timeout_ms, last_registry_gen) that:

  • Blocks in kernel until woken by mark_window_dirty, mouse, or registry change
  • Returns bitmask: bit0=dirty, bit1=mouse, bit2=registry + packed registry_gen
  • USB HID mouse handler wakes compositor on movement
  • REGISTRY_GENERATION atomic bumped on window register
  • Removed duplicate blocking from op=16 composite_windows

BWM main loop restructured: only blits when DIRTY flag set, only discovers windows when REGISTRY flag set, no idle polling.

Priority 2: MAP_SHARED Client Windows + Occluded Blit [DONE]

Actual savings: ~36% CPU (70% -> 34%) Status: Merged (PR #253)

  • op=21 map_window_buffer: maps client window physical pages into BWM read-only
  • op=22 check_window_dirty: lightweight generation check without pixel copy
  • Occluded blit: span-based row clipping skips pixels covered by higher-z windows
  • Eliminates: read_window_buffer syscalls, kernel page copies, pixel_cache, z-repair

Priority 3: Strip Vestigial Composite Allocations

Expected savings: 3-5% CPU Status: Planned

handle_composite_windows (op=16) allocates Vec and clones page_phys_addrs per window every frame. In MAP_SHARED path, page_phys_addrs is NEVER used by virgl_composite_windows -- it's vestigial. Also registered_windows() allocates a new Vec every call.

Fix: Remove Vec construction, pass only bg_dirty + dirty_rect to GPU driver. Replace registered_windows() with direct-write to output buffer.

Files:

  • kernel/src/syscall/graphics.rs (composite_windows handler)
  • kernel/src/drivers/virtio/gpu_pci.rs virgl_composite_windows signature

Priority 4: Reduce clock_gettime Calls [DONE]

Status: Done (part of PR #253)

Reduced from 5x now_monotonic() per iteration to 2 (start + end).

Priority 5: Batch Window Discovery [DONE]

Status: Done (part of PR #254)

REGISTRY_GENERATION atomic in kernel, compositor_wait returns generation. BWM only calls list_windows when registry changed.

Actual Results

After Attack CPU FPS
Baseline 70% ~100
+Priority 2 (MAP_SHARED + occluded blit) 34% ~130
+Priority 1 (compositor_wait) ~20% ~186

Total reduction: 70% -> 20% (71% reduction), FPS: 100 -> 186 (86% increase).