fix: reduce idle CPU wakeups and deferred-work polling

kernel/src/Timekeeping/ApicTimer.cpp

@@ -41,8 +41,8 @@ namespace Timekeeping {
     static constexpr uint32_t DIVIDE_BY_16 = 0x03;
 
     // The BSP keeps a 1 ms tick for timekeeping and sleep deadlines.
-    // APs use a coarser 10 ms scheduler tick to avoid waking idle cores
-    // 1000 times per second with no useful work to do.
+    // Running APs use a 10 ms scheduler tick. Idle APs mask it entirely and
+    // rely on reschedule IPIs, avoiding periodic package wakeups.
     static constexpr uint32_t BSP_TICK_INTERVAL_MS = 1;
     static constexpr uint32_t BSP_TIMER_HZ = 1000 / BSP_TICK_INTERVAL_MS;
     static constexpr uint32_t AP_TICK_INTERVAL_MS = 10;
@@ -220,8 +220,28 @@ namespace Timekeeping {
         // identical. This avoids PIT contention during AP boot.
         if (g_ticksPerMs == 0) return;
 
-        // Configure a coarser periodic timer on APs. The scheduler still gets
-        // a 10 ms time slice, but idle APs stop taking 1000 timer interrupts/sec.
+        // Configure the 10 ms scheduler timer for running APs. Their idle loop
+        // masks it after initialization and rearms it when dispatching work.
+        ProgramTimer(true, AP_TICK_INTERVAL_MS);
+    }
+
+    void ApicTimerEnterApIdle() {
+        auto* cpu = Smp::GetCurrentCpuData();
+        if (cpu == nullptr || cpu->cpuIndex == 0 || g_ticksPerMs == 0) return;
+
+        uint32_t lvt = Hal::LocalApic::ReadRegister(Hal::LocalApic::REG_TIMER_LVT);
+        if ((lvt & LVT_MASKED) == 0) {
+            Hal::LocalApic::WriteRegister(Hal::LocalApic::REG_TIMER_LVT,
+                                          lvt | LVT_MASKED);
+        }
+    }
+
+    void ApicTimerLeaveApIdle() {
+        auto* cpu = Smp::GetCurrentCpuData();
+        if (cpu == nullptr || cpu->cpuIndex == 0 || g_ticksPerMs == 0) return;
+
+        // Reprogram the initial count as well as unmasking. A deep idle state
+        // may have stopped the local timer at an arbitrary point in its period.
         ProgramTimer(true, AP_TICK_INTERVAL_MS);
     }
 
@@ -239,19 +259,24 @@ namespace Timekeeping {
         bool wasReserved = cpu->reservedForKernelWork;
         cpu->reservedForKernelWork = true;
 
-        // Drain USB hot-plug deferred work from any idle core, not just the BSP.
-        if (Drivers::USB::Xhci::HasDeferredWork()) {
+        // Drain USB work only when the MSI path has actually queued something.
+        // Bluetooth shares this controller, so service its protocol queues in
+        // the same pass after xHCI has delivered completion callbacks.
+        bool usbWork = Drivers::USB::Xhci::HasDeferredWork();
+        if (usbWork) {
             Drivers::USB::Xhci::ProcessDeferredWork();
         }
 
         // NIC hard IRQs only acknowledge/mask and queue RX work. Dispatching
         // Ethernet/TCP/UDP here keeps process-context IPC mutexes out of IRQs.
-        Drivers::Net::E1000::ProcessDeferredWork();
-        Drivers::Net::E1000E::ProcessDeferredWork();
+        if (Drivers::Net::E1000::HasDeferredWork())
+            Drivers::Net::E1000::ProcessDeferredWork();
+        if (Drivers::Net::E1000E::HasDeferredWork())
+            Drivers::Net::E1000E::ProcessDeferredWork();
 
         // HDA completion IRQs only acknowledge/mask. Resampling and DMA-ring
         // refill are far too expensive for hard interrupt context.
-        if (cpu->cpuIndex == 0) {
+        if (cpu->cpuIndex == 0 && Drivers::Audio::IntelHda::HasDeferredWork()) {
             Drivers::Audio::IntelHda::ProcessDeferredWork();
         }
 
@@ -260,7 +285,10 @@ namespace Timekeeping {
         // seconds and used to stall kmain before the first process spawned.
         // Cheap no-op unless an adapter is waiting; self-claiming, and safe
         // to preempt (the scheduler saves/resumes the idle context).
-        Drivers::USB::Bluetooth::ServiceDeferredInit();
+        bool bluetoothWork = usbWork ||
+            Drivers::USB::Bluetooth::HasDeferredWork();
+        if (bluetoothWork)
+            Drivers::USB::Bluetooth::ServiceDeferredInit();
 
         // Service Bluetooth inbound traffic (the headset's SDP/AVRCP queries,
         // AVDTP commands, ACL flow-control credits) whenever a core idles.
@@ -269,13 +297,16 @@ namespace Timekeeping {
         // writes got silence (observed: Bose re-dialing SDP during playback,
         // queries never answered).  Self-serializing and a cheap no-op when
         // the adapter is down.
-        Drivers::USB::Bluetooth::ServiceEvents();
+        if (bluetoothWork)
+            Drivers::USB::Bluetooth::ServiceEvents();
 
         // Wi-Fi mirrors the Bluetooth split: the firmware load needs the
         // ramdisk, and the RX/notification ring must be drained outside hard
         // interrupt context (the MSI handler only latches a flag).
-        Drivers::Net::Wifi::ServiceDeferredInit();
-        Drivers::Net::Wifi::ServiceEvents();
+        if (Drivers::Net::Wifi::HasDeferredWork()) {
+            Drivers::Net::Wifi::ServiceDeferredInit();
+            Drivers::Net::Wifi::ServiceEvents();
+        }
 
         // Thermal policy records transitions during BSP maintenance; print
         // them from this explicitly non-interrupt idle path.

kernel/src/Timekeeping/ApicTimer.hpp

@@ -11,9 +11,14 @@ namespace Timekeeping {
     // Initialize the APIC timer: calibrate against PIT, start periodic interrupts
     void ApicTimerInitialize();
 
-    // Initialize the APIC timer on an AP (calibrate + start, no IRQ handler registration)
+    // Initialize the scheduler timer on an AP using the BSP calibration.
     void ApicTimerInitializeAP();
 
+    // Idle APs are woken for runnable work by the reschedule IPI, so their
+    // periodic scheduler timer can remain masked until a process is dispatched.
+    void ApicTimerEnterApIdle();
+    void ApicTimerLeaveApIdle();
+
     // Reinitialize the APIC timer after S3 resume using the previously
     // calibrated tick rate. Skips PIT calibration and IRQ registration
     // (both survive in RAM). Only reprograms the timer hardware registers.