Files
MobileGL/docs/Disaggregated/notes/p5e/reports/vi-v1.md
T
swung0x48 aba329cc53 [Docs] (Disaggregated): P6's plan and contract draft; the working notes move from a local directory into the repository
Two things land together: the plan and contract draft P6 needs before it starts, and the
working notes that until now existed on exactly one machine.

1. P6's plan and contract draft (neither started; not one line of code)

     docs/Disaggregated/P6-SPAWN-PLAN.md      the package plan
     docs/Disaggregated/P6-CONTRACT-DRAFT.md  the contract draft (moves to
                                              MG_Remote/CONTRACT-P6.md when c6 lands)

   The three findings that mattered, all written into the documents:

   a. P6 is SMALLER than its roadmap cell suggests. The transport primitives are already in
      the tree and already tested: SocketDoorbell including hangup detection and the two
      lost-wakeup fences (Doorbell.cpp:211-360, four cases in FdPassingTest), SCM_RIGHTS fd
      passing, ShmSegment::Adopt, and the SurfaceOp / WindowKind schema already defined in
      protocol.fbs. inproc uses a real second shm mapping rather than an alias, and
      SessionRings.h:22-28 says that was done so today is not the first time it runs. P6 is
      assembly plus a control plane plus a process, not "write a socket transport".

   b. The evidence for "P6 is only a transport swap" has EXPIRED. It was bought by P5c, whose
      audit found 59 direct memory accesses in a tree that already looked clean. P5e changed
      the tree again. So the first package, a6, is a read-only audit that writes no code, and
      every contract row that depends on it is marked PENDING a6 rather than decided in
      advance.

   c. P6 does not get a game running on the phone. ANativeWindow* is meaningless in the child,
      and protocol.fbs:186-196's own comment says Android transfers the window out of band -
      that band is P12. P6 lands pbuffer / surfaceless / offscreen, which is what HeadlessGL,
      the split-equivalent lane and trace replay need, and which is exactly what P6's own exit
      gate measures. A real window reaching the server is a named refusal. This gets its own
      section because "P6 is done" would otherwise read as "the game runs now".

   The contract adds rule G: a role may not name a process-local handle. It extends P5c's
   rule E (a role may not name the other role's memory) from memory to identity - EGLDisplay /
   EGLSurface / EGLContext / ANativeWindow* / fd number / pid / mapped address either cross as
   a token this contract defines or they do not cross, and a descriptor crosses only through
   SCM_RIGHTS, where the kernel renumbers it.

   What P5e left for P6 gets its own section. The latch semantics are already designed
   (ARCHITECTURE.md:504) and are copied unchanged. What is new is that after run-ahead the
   client no longer waits while the peer process can DIE: the client holds records that were
   published and will never be applied, and Android's freezer suspends the child without
   closing anything. So slow and dead must be distinguishable WITHOUT a timeout threshold.
   That is P6's only new design; everything else is carriage. It shares a predicate with E1's
   outstanding re-specification (ID-122), and the two should be designed together.

   Two gates carried over from P5e's lessons: G2 requires integration-spawn's test-name set to
   match integration-split's name for name, because a differing set means the lane is quietly
   running less; and the arm-proof gate requires every entry to record the child pid and
   transport=spawn in its OWN private log - ID-124 established that this kind of false green
   really happens, and a spawn lane that actually ran monolith would pass every other gate.
   S2's shape must be settled before it is written, and the document says why: E1 is the
   cautionary case.

2. The working notes move into the repository (docs/Disaggregated/notes/, 415 .md files)

   ROADMAP / CURRENT_STAGE_PROGRESS / MEASUREMENTS cite ruling numbers, audit line numbers and
   the provenance of gate figures from ~/w7/notes/, a directory that exists on one machine and
   is under no version control. A record cited hundreds of times that lives only locally is not
   a record, so it moves in whole, with its directory structure intact.

   Only .md moves. Scripts, run logs, census and gate result JSON, patches, APKs and nm dumps
   stay where they are: they are evidence DATA rather than documents - large, regenerated per
   run, and mostly carrying absolute paths and device serials that mean something only on the
   machine that produced them. The mapping and this split are written in
   docs/Disaggregated/notes/README.md.

   Two exceptions are recorded there too: P6's BRIEF has been promoted to a first-class
   document and keeps no copy under notes/, so the two cannot drift apart; and recovered/ holds
   material salvaged from failed workflow runs, which may duplicate the like-named phase
   directories and is kept because a few of those were the only complete version at the time.

   CI's citation lint takes docs/Disaggregated/*.md only (non-recursive), and notes/ is
   deliberately outside it: those notes' file:line citations were written against their own
   code heads, and resolving them against today's HEAD would fail wholesale - which is not an
   error, it is their correct state as a historical record. To check one, use the revision it
   declares (--rev).

3. Inconsistencies fixed in passing

   - README.md's status line still said P5e was in progress while the stage table already said
     closed.
   - CURRENT_STAGE_PROGRESS's next-step item 1 still listed the exit gate as pending, though
     (a) and (b) were run in the previous commit; rewritten to closed plus the one item that
     actually remains (E1).
   - ROADMAP's P6 row still had the pre-phase plan in three of its cells; it now carries the
     package order, the scope boundary and the two new gates.

Checked with check_doc_citations.py: 125 citations across docs/Disaggregated/*.md, 0 problems.
2026-09-18 23:31:39 -04:00

236 lines
15 KiB
Markdown
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
# vi — VAO / vertex input / index, v1
Branch `p5e/vi`, head **112c1d8d**, base `f6cfcbd3`. Three commits, no trailers. Tree clean.
## 1 What changed, per file (line ranges at the BASE head)
**`DirectGLES.cpp`** — `:490` `VertexInputReadsRecords()` (arm selector: `Transport != Monolith` AND
the family bit, ruling 1 / ID-81) + `AnyClientSideVertexArrayInRecord()` (reads BOTH views: a null
`Res` is "client-sourced" only for an attribute the CSO declares ENABLED, since the window also
carries nulls for disabled slots below the high-water mark). `:786`
`SyncVaoAttributeBuffersByRecord(memo, epoch)` beside the existing `...ByHandle` — same memo, same
three-value key, walk over `st.VertexBuffers[Start..+Count).Res`, dedupe on `{slot,gen}`,
`EnsureBufferResourceForHandle(nullptr, Res)`, probe with `frontend == nullptr` — plus the two
named source decisions `ResolveDrawVertexBuffersFromRecord` / `ResolveDrawIndexBufferFromRecord`
(§3: they exist so a unit case drives the production decision, not a copy of it). `:807, :827-831`
the null-VAO early return no longer fires on the record arm and the dispatch takes it first.
`:896-941` the IBO arm reads `st.IndexBuffer.Res` + `st.IndexBufferSerial`; the legacy and
push-monolith arms unchanged inside a new brace block. `:1602-1639` `ResolveVaoTwin(SharedPtr)`
keeps its body and LOSES its `MGPipeFrontendKeyedRegistryScope` (§4.4's `:1609`). `:1655` new
`SyncCurrentVAOFromRecords(twin)`. `:1667-1694` `SyncCurrentVertexAttributeValues` reads neither
`GetBoundVertexArray` nor `GetConfigVersion`; memo keyed `{elementsHandle, ContentSerial,
activeMask}` over `rec->Attributes[i].Enabled`. `:4770-4795` `PrepareForDraw`'s VAO lines only, and
`AccessorCalls` is 1 per draw on the record arm (it was 2 and one of the two reads is gone — the
ledger has to say so). `:6446-6461` `RefuseElementArrayBufferFromTheFrontend` in both restart
helpers. `:6718-6788` the two client-array arms folded into one file-local `static`
`SyncClientSideVertexArraysForDrawArrays`; both scopes (`:6729`, `:6775`) deleted, twin by handle.
**`Managers.h`** — `ResolvedDrawBuffers::iboSerial` (push-only); `Entry::frontend`/`::attribIndex`
re-documented as monolith glue / diagnostics; `PendingAttribValueMask` gains `{elementsHandle,
elementsSerial}` (push-only); `SyncToBackendFromApplier` private → public (§4.1: "the object
parameter is deleted from the transport overload, not defaulted to null");
`BufferMutationEpochBumpsOffTheApplyThread()`; the two `Resolve*FromRecord` declarations.
**`Managers.cpp`** — `BumpBufferMutationEpoch` only (§4b); `SyncToBackendFromApplier`'s body is
UNCHANGED and is this family's existence proof.
**`EmitTables.{h,cpp}`** — `RemoteDrawBindings::ClientVertexArrays`, answered inside
`ReadDrawBindings`' one binding read; `PlanDrawInfo` sets `kDrawClientArrays` for every draw shape
(a `glDrawElements` can fetch its VERTICES from client memory); `EmitDrawRecord` and the reduced
`EmitDrawArrays` path refuse under run-ahead.
**Tests** — `MG_Test/SanityTest.cpp` (+ pull skip twins), `MG_Test/Pipe/VertexInputEmitTest.cpp`,
new `MG_IntegrationTest/Scenarios/ClientVertexArrayScenario.cpp` with its `CMakeLists.txt` block and
its `Harness/SplitLogPaths.cmake.in` entry.
## 2 The kimi-audit rows of this family, 1–34
**RETIRED** (not read under a transport): **1** (`GetBoundVertexArray@PrepareForDraw`), **2**, **3**
(twin probe + mint), **5** (`GetConfigVersion`), **6** (the VAO handoff), **9**, **10**'s
`:1683`/`:1688` halves (`:1700` was already RECORD_SUPPLIED), **12**, **13** (the `Find(vao.get())`
probe is gone unconditionally; the VAO itself is taken only when the record says an enabled
attribute is client-sourced), **16**, **17**, **19**, **20**, **21**, **23**, **24**; and **25** /
**27** *for this family's callers* (`HandleOfBuffer` inside `IsBufferDrawClean` /
`EnsureBufferResource` is no longer reached from the vertex-input arm; sb/tx2/fb keep their own).
Row **18**'s frontend argument is gone (the record arm passes `nullptr`).
**LEFT BARRIERED**: **14** / **34** — `SyncClientSideAttributesForDrawArrays` still dereferences
`attrib.Offset` as a client pointer (§4a). **15** — `MultiDraw.cpp:96`, see R2.
Rows **4**, **7**, **8**, **22**, **26** are legacy-arm-only bodies and die with the legacy arm, not
with this package.
**NOT FOUND** as a live split read: **11** (the restart helpers were already unreachable under a
transport — the P5c substitution sets `serverElementBinding >= 0` before its null test; now they
refuse by name instead of being unreachable-by-reading), **28**, **29** (monolith-only already;
passing `nullptr` changes nothing), **33** (already `Transport == Monolith`-gated).
**NOT vi's**: **30**, **31**, **32** (sb's).
## 3 Red-once evidence, verbatim
**(1) Twin key.** `PrepareForDraw`'s transport arm reverted to
`ResolveVaoTwin(MGB_CTX->GetBoundVertexArray())`; both `DirectGLES.Split.TriangleScenario` entries
`Subprocess aborted`:
```
[08:02:07] [Linux mgl-srv-apply/FATAL]: MGPipe: Fatal{RoleViolation, "MGPipeSlots"} - the apply thread called MGPipeSlots().HandleOf. With an active transport the client slot allocator is client-only memory (CONTRACT-P5C §3.1, rule E; CONTRACT-P5E §4.4): a handle arrives already minted in a record, and a server that resolves or mints one off a frontend object's lifetime id is reading memory that will not exist on its side of a real split. A named exemption scope admits it only while the record being applied is BARRIERED (barriered=1) and, for Magma's P7 debt, only on a DirectVulkan server
```
Note `barriered=1`: **no id test hook was needed**. Deleting the scope is what makes the revert loud
today rather than after ra flips the wait rule.
**(2) The buffer walk.** `ResolveDrawVertexBuffersFromRecord` reverted to
`MG_State::pGLContext->GetBoundVertexArray()->GetAllAttributes()`:
```
../MobileGL/MG_Test/SanityTest.cpp:5088: Failure
Expected equality of these values:
count
Which is: 1
2u
Which is: 2
the walk did not resolve st.VertexBuffers[Start..+Count): two DISTINCT handles are named there (A twice and B), plus one client-memory array with no store to ensure
```
**(3) The index arm.** `ResolveDrawIndexBufferFromRecord` reverted to the VAO's element slot:
```
../MobileGL/MG_Test/SanityTest.cpp:5149: Failure
Expected equality of these values:
request.Res
Which is: 8-byte object <02-00 00-00 00-00 00-00>
publishedHandle
Which is: 8-byte object <01-00 00-00 00-00 00-00>
the index arm did not read st.IndexBuffer.Res; if it answered the VAO's element slot it is reading a binding the client has already changed
../MobileGL/MG_Test/SanityTest.cpp:5152: Failure
request.Serial
Which is: 0
11u
IndexBufferSerial did not travel with the handle
```
**(4) Client arrays under run-ahead.** `kMGPipeP5eRunAheadReady` flipped to `true` in
`MG_Backend/Init.cpp`, `ClientVertexArrayScenario` run in the split lane — exit 134 and, on the GL
thread's own log:
```
[08:03:18] [Linux MobileGLIntegra/FATAL]: MGPipe: Fatal{UnmigratedVerb, "DrawArrays+CLIENT_ARRAYS"}
```
**(ID-95 window pin, extra.)** `EmitVertexBuffers`' count narrowed by one:
```
../MobileGL/MG_Test/Pipe/VertexInputEmitTest.cpp:490: Failure
Expected: (i) < (window.Start + window.Count), actual: 7 vs 7
create_vertex_elements declares attribute 7 ENABLED but set_vertex_buffers' window ends at 7 - the same skip, at the other end. A narrower window is the sink's Fatal{ProtocolCorruption, "SetVertexBuffers.Count"}
```
Every revert was put back and re-verified green before the commits.
## 4 The two answers the gate asked for
**(a) Does any existing scenario exercise a client vertex array under split? NO.** Every
`glVertexAttribPointer` in `MG_IntegrationTest/Scenarios` is issued with `GL_ARRAY_BUFFER` bound, so
its last argument is a byte offset; `DoublePrecisionScenario` — the brief's candidate — uses
`glVertexAttribFormat`/`glVertexAttribBinding`/`glVertexAttribLFormat` against real buffer objects
(`DoublePrecisionScenario.cpp:944-959`). `ClientVertexArrayScenario` is therefore new and is the
only lane entry that can reach the refusal.
**A DEVIATION FROM THE BRIEF that needs the integrator's eye.** The brief makes the
`DrawArrays`/`MultiDrawArrays` arms *monolith-only*. **They cannot be, before ra lands.** Under split
LOCKSTEP a client vertex array is not refused (ID-82: lockstep unchanged), so a monolith-only arm
would silently drop the attribute upload and render wrong *today*. What landed instead: the arm
stays on both, resolves the twin BY HANDLE (so the probe and both scopes go either way), takes the
frontend VAO only when `AnyClientSideVertexArrayInRecord()` is true — which is what actually retires
`GetBoundVertexArray@DrawArrays` for ordinary draws — and aborts by name if it is ever reached from
an UNBARRIERED apply, which escalation (ii) of §2.1 and the client refusal both make unreachable.
The new scenario passes on all four arms (DirectGLES / DirectVulkan / DirectGLES.Split × 2 cases),
which is the evidence that the arm is load-bearing. `SyncClientSideAttributesForDrawArrays` is
likewise NOT made monolith-only.
**(b) Does any GL-thread path bump `CurrentBufferMutationEpoch` under a transport? YES — exactly one
class, and it is before the apply thread exists.** An unconditional refusal in
`BumpBufferMutationEpoch` took the whole split lane down in EGL bring-up, which is how this was
found rather than reasoned: `BackendObject_DirectGLES::Initialize()` runs on the APP thread by
design and ends in `RegisterBufferBackendOps()`, whose last act is a bump. At that moment
`ServerLoop::Detail::g_applyThreadKey == 0` — no record applied, no twin, no memo stamp for it to
slide under. What landed is the narrower pin: a bump off the apply thread *while one is running* is
counted (`BufferMutationEpochBumpsOffTheApplyThread()`) and **Fatal under
`MOBILEGL_IPC_STRICT_ERRORS`**; across the 113-entry split lane it never fires. Not Fatal by default
because `StagedShadowProductionTest`/`ServerLoopTest` drive the real op tables from the test thread
beside a started loop — 7 unit cases went red on the wider form, and policing a fixture is not what
this pin is for.
## 5 Gate
| step | result |
|---|---|
| `build` | rc=0 |
| `unit` | **2209/2209** (base 2206; +3, each with a pull skip twin) |
| `isplit` | **113/113** (base 111; +2 `ClientVertexArrayScenario` split entries) |
| `gens` | all 7 rc=0; `check_doc_citations.py --strict` rc=0 (its three printed lines are pre-existing `CONTRACT-P5.md` ambiguities) |
| `one`, the eight named scenarios | **274/274**, covering BOTH the monolith `DirectGLES.*` arm and `DirectGLES.Split.*`, so ruling 1's A/B is exercised and `ObjectSubsystemControlScenario`'s `0x0ff`/`0x17f` verdicts are unchanged |
| `strict` | red by design (110/113 abort), markers below |
**Strict markers, before → after** (per-entry private logs, ID-53; counts are log lines):
| `field@verb` | base (111) | after (113) | owner |
|---|---|---|---|
| `GetBoundVertexArray@DrawArrays` | **14** | **0 — GONE** | vi |
| `GetProgramForDraw@DrawArrays` | 0 | **14 (new)** | pg — the next field in *those same 14 entries* |
| `GetFramebufferBindingSlot@Clear` | 39 | 41 | fb (+2 = my two new lane entries, which clear first) |
| `GetImageTextureBinding@BindImageTexture` | 13 | 13 | fb |
| `GetTextureUnitObject@Clear` | 9 | 9 | tx2 |
| `GetTextureObject@CopyImageSubData` | 6 | 6 | tx2 (barriered, allowlisted) |
| `GetProgramForDispatch@DispatchCompute` | 3 | 3 | pg |
| `ValidateProgramName@ShaderStorageBlockBinding` | 2 | 2 | pg (barriered) |
This family's only visible field is retired; the field that surfaced behind it is pg's.
## 6 G1 / monolith impact
The pull build is **not rebuilt here** (the gate configures only `build-split`); the delta is named
rather than measured. Everything is under `#if MOBILEGL_BUILD_DISAGGREGATED` or
`#if MOBILEGL_PIPE_PUSH` except: one new **file-local `static`** function
(`SyncClientSideVertexArraysForDrawArrays`) re-homing two byte-identical inline blocks from the two
`DrawArrays` entry points (no exported symbol); three constant-folded branches
(`VertexInputReadsRecords()` is `return false` there, so the null-VAO guard, `PrepareForDraw`'s
selector and the `AccessorCalls` count all fold) plus two default-constructed null `SharedPtr`
locals; and `RefuseElementArrayBufferFromTheFrontend` compiling to `(void)entry;`. No struct layout
moves — `iboSerial` and the `PendingAttribValueMask` fields are `MOBILEGL_PIPE_PUSH`-only. G2/G14:
the new unit cases have pull skip twins; the new scenario's ctest entries appear in both arms
symmetrically. **Please re-run the pull configure at landing.**
## 7 Seams assumed, and what I need
1. **ra — `ClientSession::RunAheadArmed()`** does not exist yet. `EmitTables.cpp`'s
`RunAheadWouldSkipTheWait(session)` spells the same conjunction `MGPipeTypes.h:148` states
(`BarrierArmed() && Ipc.RunAhead && Caps().HasCap(kCapRunAheadApply)`). **ra should collapse that
body into the latch**: a duplicated predicate that can disagree with the wait rule about which
arm the session is on is the one failure mode here. Inert until the caps bit is published.
2. **tx2 — `CheckUnitWindows(st)`.** I pinned the PRODUCER half of ID-95 / ruling 19 only; the sink
check and its `Fatal{ProtocolCorruption, "SetVertexBuffers.Count"}` are tx2's. My test names that
exact string so the two halves meet.
3. **pg.** `SyncCurrentVertexAttributeValues` still reads `program->GetActiveAttributeLocationMask()`
and `GetAttribType(location)` (kimi row 106, S1's R3). My half is done; it stays a frontend reader
— a BARRIERED one — until pg carries the mask and types in the `ShaderCso` record.
`GetProgramForDraw@DrawArrays` is now the marker those 14 entries die on, so pg will see it.
**Rulings needed:** **R1** — the "monolith-only" wording for the two `DrawArrays` arms and for
`SyncClientSideAttributesForDrawArrays` (§4a); confirm my reasoning or tell me to take the
regression. **R2** — `MultiDraw.cpp:96 BoundIndexBuffer()` / `:105 BoundIndexBufferId()` (kimi rows
15, 128) read `MGB_CTX->GetBoundVertexArray()` on every multi-draw with no transport arm, and
`MultiDraw.cpp` is in no package's scope in BRIEF-P5E; it is a two-line change of exactly §5.1's
shape. **Left as-is** — widen vi's range or take it. **R3** — `BumpBufferMutationEpoch` is in
`Managers.cpp` outside my declared range; the gate asked for the pin and there was nowhere else for
it. **R4** — `SyncToBackendFromApplier` moved private → public in `Managers.h` (a declaration, not a
body), for the reason §4.1 gives.
## 8 What I did not do
- `ResolvedDrawBuffers`' deletion (S1 R7) — trailing by instruction, measurement first. It is now
cheap to measure: the memo's entries are `{handle, resource, bindingIndex}` over a window that is
1–3 entries on the measured workload.
- Staging client vertex arrays (P8).
- A `RemoteClientTest` unit twin for `PlanDrawInfo`'s `kDrawClientArrays` flag — that file is not in
vi's scope; the flag is covered end to end by the scenario's red-once instead.
- No pull-build configure (§6).