[MG_Backend] (P7): when the orphaned wire buffer stores still parked exceed MOBILEGL_IPC_WIRE_DEFERRED_MB (64 by default), the server takes the host-access sync point mid-frame and reclaims them all

The defer-path sweep frees an orphan once its last submission has retired, but inside one long
frame a store that was drawn from is named by recorded, unsubmitted work, and nothing submits it
until the frame ends. The knob is read once through MG_Config::Ipc like its siblings; above it
the park calls WaitForSubmitIndex(GetSyncPointSubmitIndex(), flush) - the call
WaitForWireBufferHostAccess makes - and sweeps again, which empties the list because no entry can
be tagged past the sync point just waited out. A sync point that cannot complete dies through
the existing Fatal{ResourceUnavailable, "buffer-write-sync"} site, now one helper shared with the
host write (fatal_census: 79 abort sites, unchanged). 0 is the negative control.

MobileGL/Config.h

@@ -497,6 +497,18 @@ namespace MobileGL::MG_Config {
         // stage chunk budget (MGPipeStageChunkBytes, a quarter of this), and a record type with
         // no cut is Fatal{RingOverrun, "SEG_STAGE"} rather than allowed to exceed it.
         Uint32 StageMb = 32;
+        // MOBILEGL_IPC_WIRE_DEFERRED_MB (P7 wave 4 M2, ID-P7-32): the SERVER's budget, in MiB,
+        // for orphaned wire buffer stores - the old VkBuffer every glBufferData that crosses the
+        // wire leaves behind - that a GPU command recorded but not yet retired may still name.
+        // Stores no command names are destroyed at once and stores whose last submission has
+        // retired are destroyed at the next park; this bounds the REST. When the parked bytes
+        // exceed it after a sweep, the server flushes what it has recorded and waits for it
+        // (WaitForWireBufferHostAccess's sync point, mid-frame), which retires every one. It is
+        // not a frame count because a frame is not bounded: a snapshot-exiting pbuffer replay
+        // delivers one present for 1.3 M calls. 0 IS THE NEGATIVE CONTROL, not "unlimited by
+        // design": no forced sync, so a one-frame respecify-and-draw loop grows without bound
+        // and MagmaWireReclaimScenario's watermark case must go red.
+        Uint32 WireDeferredMb = 64;
         // MOBILEGL_IPC_SPIN_US: spin before parking on a doorbell, either direction.
         Uint32 SpinUs = 50;
         // MOBILEGL_IPC_PERSISTENT_BLOCK_KB: block granularity of the persistent-map push.

MobileGL/ConfigLoader.cpp

@@ -392,6 +392,9 @@ namespace MobileGL::MG_ConfigLoader {
         // are sanity, not policy.
         ipc.RingMb = QueryEnvUint32("MOBILEGL_IPC_RING_MB", 8, 1, 1024);
         ipc.StageMb = QueryEnvUint32("MOBILEGL_IPC_STAGE_MB", 32, 1, 4096);
+        // P7 wave 4 M2 (Config.h has the semantics). 0 is admitted ON PURPOSE as the negative
+        // control: no forced sync, and MagmaWireReclaimScenario's watermark case must go red.
+        ipc.WireDeferredMb = QueryEnvUint32("MOBILEGL_IPC_WIRE_DEFERRED_MB", 64, 0, 65536);
         ipc.SpinUs = QueryEnvUint32("MOBILEGL_IPC_SPIN_US", 50, 0, 1000000);
         // 0 is admitted ON PURPOSE and is the negative control of exit gate E3(a): it turns
         // the persistent-map push OFF, and PersistentCoherentMapScenario must go red.
@@ -462,10 +465,10 @@ namespace MobileGL::MG_ConfigLoader {
         if (MG_Config::Transport == MG_Config::TransportMode::Monolith) return;
         // One line, on the arm where these numbers decide behaviour, because every one of
         // them is a number a bug report has to quote.
-        MGLOG_I("Config: IPC ring=%uMiB stage=%uMiB spin=%uus persistent-block=%uKiB "
+        MGLOG_I("Config: IPC ring=%uMiB stage=%uMiB wire-deferred=%uMiB spin=%uus persistent-block=%uKiB "
                 "adopt-tier=%u verb-barrier=%u run-ahead=%u present-credit=%u strict=%d "
                 "audit=%d role-split-state=%d affinity='%s'",
-                ipc.RingMb, ipc.StageMb, ipc.SpinUs, ipc.PersistentBlockKb, ipc.AdoptTier,
+                ipc.RingMb, ipc.StageMb, ipc.WireDeferredMb, ipc.SpinUs, ipc.PersistentBlockKb, ipc.AdoptTier,
                 ipc.VerbBarrier, ipc.RunAhead, ipc.PresentCredit,
                 static_cast<int>(ipc.StrictErrors), static_cast<int>(ipc.Audit),
                 static_cast<int>(ipc.RoleSplitState), ipc.ServerAffinity.c_str());
@@ -474,6 +477,11 @@ namespace MobileGL::MG_ConfigLoader {
                     "expected to fail: the client still pulls 31 of 63 PipeInputs fields from a "
                     "live GLContext, so an unbarriered queue lets the server read future values");
         }
+        if (ipc.WireDeferredMb == 0) {
+            MGLOG_W("Config: MOBILEGL_IPC_WIRE_DEFERRED_MB=0 is the M2 NEGATIVE CONTROL: the "
+                    "server never forces a sync for orphaned wire buffer stores, so a long frame "
+                    "that respecifies and draws holds every one of them until the frame ends");
+        }
         if (ipc.PersistentBlockKb == 0) {
             MGLOG_W("Config: MOBILEGL_IPC_PERSISTENT_BLOCK_KB=0 is the E3(a) NEGATIVE CONTROL: "
                     "the persistent-map push is OFF and a coherent-map scenario must go red");

MobileGL/MG_Backend/DirectVulkan/Renderer/VkBufferManager.cpp

@@ -365,10 +365,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                 uploaded = StagedWireRangeCopy(*resource, bytes, static_cast<SizeT>(offset),
                                                 static_cast<SizeT>(size));
             }
-            if (!uploaded && !WaitForWireBufferHostAccess(*resource)) {
-                MGLOG_F("Magma: Fatal{ResourceUnavailable, \"buffer-write-sync\"}");
-                std::abort();
-            }
+            if (!uploaded && !WaitForWireBufferHostAccess(*resource)) WireBufferSyncFatal("host-write");
         }
         if (!uploaded) {
             uploaded = resource->buffer.Upload(bytes, size, offset);
@@ -606,14 +603,38 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         // pbuffer replay that snapshot-exits delivers ONE present for 1.3 M calls (ID-P7-32) -
         // so the only cadence that tracks the workload is the workload itself.
         SweepDeferredWireReleases();
+        EnforceWireDeferredWatermark();
+    }
+
+    void VkBufferManager::EnforceWireDeferredWatermark() {
+        const Uint64 budget = static_cast<Uint64>(MG_Config::Ipc.WireDeferredMb) * 1024u * 1024u;
+        if (budget == 0 || m_deferredWireBytes <= budget) return;
+        // What is left after the sweep is named by work that is recorded but unsubmitted, or
+        // submitted but unretired. Every entry is tagged at or below the sync point taken HERE,
+        // so waiting it out - the host-access wait's own call, which flushes the recording first
+        // - proves all of them dead. A mid-frame flush is what WaitForWireBufferHostAccess and a
+        // glClientWaitSync already do to this command stream; it costs one submission per
+        // budget's worth of orphans, not one per glBufferData.
+        if (pVulkanRenderer == nullptr ||
+            !pVulkanRenderer->WaitForSubmitIndex(pVulkanRenderer->GetSyncPointSubmitIndex(), UINT64_MAX, true)) {
+            WireBufferSyncFatal("deferred-watermark");
+        }
+        SweepDeferredWireReleases();
+    }
+
+    void VkBufferManager::WireBufferSyncFatal(const char* site) {
+        MGLOG_F("Magma: Fatal{ResourceUnavailable, \"buffer-write-sync\"} {site=%s}", site);
+        std::abort();
     }
 
     SizeT VkBufferManager::SweepDeferredWireReleases() {
-        if (m_deferredWireReleases.empty() || pVulkanRenderer == nullptr) return 0;
+        if (m_deferredWireReleases.empty()) return 0;
         SizeT dead = 0;
-        if (pVulkanRenderer->IsSubmitIndexComplete(pVulkanRenderer->GetSyncPointSubmitIndex())) {
-            // Idle: nothing recorded, every submission retired. Every parked store is dead,
-            // including one tagged for a batch that was abandoned instead of submitted.
+        if (pVulkanRenderer == nullptr ||
+            pVulkanRenderer->IsSubmitIndexComplete(pVulkanRenderer->GetSyncPointSubmitIndex())) {
+            // Idle: nothing recorded, every submission retired - or no renderer, so no queue.
+            // Every parked store is dead, including one tagged for a batch that was abandoned
+            // instead of submitted.
             dead = m_deferredWireReleases.size();
         } else {
             // The list is in park order and GetSyncPointSubmitIndex() never decreases, so the

MobileGL/MG_Backend/DirectVulkan/Renderer/VkBufferManager.h

@@ -294,13 +294,22 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             // running total has to stay exact as entries leave.
             Uint64 bytes = 0;
         };
-        // Park a wire store, then sweep. `lastUseSerial` is the releasing resource's, captured
-        // before the caller resets it.
+        // Park a wire store, sweep, then hold the watermark. `lastUseSerial` is the releasing
+        // resource's, captured before the caller resets it.
         void DeferWireRelease(VkBufferObject&& buffer, Uint64 lastUseSerial);
         // Destroy every parked store proven dead by the rules above. Returns how many.
         SizeT SweepDeferredWireReleases();
+        // MOBILEGL_IPC_WIRE_DEFERRED_MB (Config.h has the semantics): when the parked bytes
+        // still exceed the budget after a sweep, take the sync point WaitForWireBufferHostAccess
+        // takes - flush what is recorded, wait for it - and sweep again, which retires every
+        // parked store, because none can be tagged past the sync point it just waited out.
+        void EnforceWireDeferredWatermark();
         // Teardown: the caller has proven the device idle (Shutdown / RecreateTransientArenas).
         void DestroyAllDeferredWireReleases();
+        // THE one Fatal{ResourceUnavailable, "buffer-write-sync"} site (rule I: no second abort
+        // for the census to count), shared by the host write that cannot wait for the GPU and
+        // the watermark that cannot. `site` says which.
+        [[noreturn]] static void WireBufferSyncFatal(const char* site);
 #endif
         Bool InitializeTransientArenas();
         static VkBufferUsageFlags GetVkBufferUsage(BufferKind kind);