tools: durable guest counter enumeration/injection and VA reader

The counter enumeration/injection workflow lived in the guest's /tmp and was
lost on every restart. Make it durable under tools/guest-diag:

- sem_counters.py: translate a process's per-channel u32 completion counters
  (16-byte stride in the /dev/nvidiactl counter page) to guest PAs through
  /proc/<pid>/pagemap; --pairs emits handle,va,pa triples.
- sem_slots.py: read/write the counter slots via /proc/<pid>/mem (fallback
  for the host counter publish).
- read_va.py: hex dump a process VA (UVM sysmem GPU VAs are usually valid
  CPU VAs in the owning process); --u32 for method streams.
- inject_counters.sh: host-side one-shot: push the tool, take the GR handles
  from the live chan_dump, enumerate, then register each counter with
  sema_counter and pin its VA with sema_map_add.
- push_guest_tools.sh: push the new tools after every VM start.
- launch_qemu_direct.sh: parameterized mdev UUID/paths instead of a stale
  hardcoded one; committed so it survives checkouts.

1 files changed, 68 insertions, 0 deletions

diff --git a/tools/guest-diag/inject_counters.sh b/tools/guest-diag/inject_counters.sh
new file mode 100755
index 0000000..a2a8371
--- /dev/null
+++ b/tools/guest-diag/inject_counters.sh
@@ -0,0 +1,68 @@
+#!/bin/bash
+# inject_counters.sh - register the running CUDA client's per-channel GPU
+# completion counters with the host sema service (the PART 45/46 manual
+# procedure, made durable). The service then publishes each GR channel's
+# consumed GPFIFO count into the client's counter, which is the completion
+# signal its trackers poll.
+#
+# Usage: tools/guest-diag/inject_counters.sh [guest-pid] [page-va] [count]
+#   guest-pid  default: pgrep -x cdt in the guest
+#   page-va    default: 0x20400f000 (observed counter page VA)
+#   count      default: 20 (one per GR channel)
+#
+# Handle order: GR channels (eng=0) from the host channel dump, ascending, are
+# zipped with ascending counter index (the observed 1:1 layout). Override the
+# list with GRS="0x...,0x...,...".
+set -euo pipefail
+
+HERE="$(cd "$(dirname "$0")" && pwd)"
+VMEXEC="$HERE/../../scripts/vgpu_vm.py"
+PARAM=/sys/module/open_iov_vfio/parameters
+
+PID="${1:-}"
+PAGE_VA="${2:-0x20400f000}"
+COUNT="${3:-20}"
+
+if [ -z "$PID" ]; then
+    PID=$(timeout 20 "$VMEXEC" exec "pgrep -x cdt" 2>/dev/null \
+        | grep -E '^[0-9]+$' | head -1 || true)
+fi
+if [ -z "$PID" ]; then
+    echo "ERROR: no cdt pid in the guest (pass one: $0 <pid> [page-va] [count])" >&2
+    exit 1
+fi
+
+# Push the enumeration tool (the guest's /tmp is wiped on boot).
+b64=$(base64 -w0 "$HERE/sem_counters.py")
+timeout 30 "$VMEXEC" exec "echo '$b64' | base64 -d > /tmp/sem_counters.py" >/dev/null
+
+# Channel handles: GR channels from the live channel dump, ascending.
+if [ -z "${GRS:-}" ]; then
+    echo 1 | sudo tee "$PARAM/chan_dump" >/dev/null
+    sleep 1
+    GRS=$(sudo dmesg | grep -o 'sema-chan-dump: h=0x[0-9a-f]* eng=0' \
+        | grep -o '0x[0-9a-f]*' | sort -u | paste -sd,)
+fi
+if [ -z "$GRS" ]; then
+    echo "ERROR: no GR channel handles (chan_dump empty; is a VM attached?)" >&2
+    exit 1
+fi
+
+PAIRS=$(timeout 30 "$VMEXEC" exec \
+    "python3 /tmp/sem_counters.py $PID $PAGE_VA --count=$COUNT --pairs $GRS" \
+    2>/dev/null | grep -E '^0x[0-9a-f]+,0x[0-9a-f]+,[0-9]+$')
+if [ -z "$PAIRS" ]; then
+    echo "ERROR: enumeration returned no pairs (pagemap access?)" >&2
+    exit 1
+fi
+
+while IFS=, read -r h va pa; do
+    # printf: sysfs writes carry a trailing newline that kstrtoull tolerates
+    # for both fields; use it anyway (no newline is also fine).
+    printf '%s,%s' "$h" "$pa" | sudo tee "$PARAM/sema_counter" >/dev/null
+    # Pin the counter VA->PA too: a push that releases the counter then
+    # resolves and applies through the normal release path, and pins survive
+    # the periodic map rebuilds.
+    printf '%s,%s' "$va" "$pa" | sudo tee "$PARAM/sema_map_add" >/dev/null
+    echo "registered h=$h counter va=$va pa=$pa"
+done <<< "$PAIRS"