vgsp-dual: stage B+C minimal per-mdev separation (G1-G3, G5, G6, G7, G8a)

Goal 1.0: two mdevs + two QEMU guests running CUDA simultaneously.
Single-mdev behaviour is kept identical: the legacy guest-RAM record
still serves the first/only open mdev, the channel table degenerates to
the historical key, and the session flush fires at the same
last-channel-closed transition.

- G6: guest-RAM backing {file, path, ranges} per mdev (legacy record +
  up to 4 vgpu-bound records); debugfs gsp-phys-guest-ram accepts an
  mdev UUID as an extra first token and keeps the 4-token form; records
  released on mdev cleanup and physical release.
- G7: gsp_phys_gpa_to_hpa_vgpu(vgpu, gpa, *hpa); NULL keeps the
  historical first-open resolution. Forwarding operands pass the
  request's vgpu; the single-guest boot-ladder pin passes NULL.
- G1: the sema-service thread services every open mdev each pass.
- G2: channel table keyed by (vgpu, h_object); per-guest ranks;
  counter/scrub/opdef/pump snapshot+writeback matches per vgpu;
  scrubber records per guest.
- G3: per-guest session records drive the flush; the shared learned
  tables (staged G4) are only flushed when no guest has live channels.
- G5: hold/abort keyed by (vgpu, handle); dead-owner guards on the
  hold resolution paths; single-inflight serialization kept.
- G8a: guest-GPA pin refused while more than one mdev is open.

Build: make -j8 clean success (vgsp.ko). No module load. See
DUAL-REPORT.md for the per-item status and live-validation plan.

1 files changed, 19 insertions, 1 deletions

diff --git a/kernel/gsp_sema_service.h b/kernel/gsp_sema_service.h
index 6822bd5..c1dd763 100644
--- a/kernel/gsp_sema_service.h
+++ b/kernel/gsp_sema_service.h
@@ -18,8 +18,16 @@
  *
  * All addresses are guest-physical except gpfifo_va, which is the GPU VA of
  * the channel's GPFIFO ring (translated through the discovered VA->PA map).
+ *
+ * vgpu names the owning mdev (guest session). Two guests' stock allocators
+ * choose the same object handles, so the channel table is keyed by
+ * (vgpu, h_object): a registration can never overwrite another guest's slot,
+ * and ranks/session bookkeeping are computed per guest.
  */
+struct nv_vgpu_device;
+
 struct sema_channel_desc {
+	struct nv_vgpu_device *vgpu;	/* owning mdev (guest session) */
 	u32 h_object;
 	u32 engine_type;
 	u32 cid;		/* RM channel id (the doorbell token vector) */
@@ -33,7 +41,17 @@ struct sema_channel_desc {
 };
 
 void gsp_sema_service_channel_add(const struct sema_channel_desc *desc);
-void gsp_sema_service_channel_del(u32 h_object);
+void gsp_sema_service_channel_del(struct nv_vgpu_device *vgpu, u32 h_object);
+
+/*
+ * Release every piece of service state that belongs to one mdev before its
+ * guest session disappears: drop the mdev's channels (per-vgpu session end),
+ * clear a release hold that names it (never dereference a removed device),
+ * and free its per-guest session record. Called from the mdev cleanup path
+ * after the device's emulation thread stopped and before the mdev struct can
+ * be freed. No-op for a vgpu that never registered a channel.
+ */
+void gsp_sema_service_vgpu_cleanup(struct nv_vgpu_device *vgpu);
 
 /*
  * Register the guest PMA scrubber's channel-buffer words (the CeUtils