kernel: enlarge the pin table and its probe window; tools: bridge client mappings into the service

The pin table became the primary VA->PA store once push-buffer regions and
bridged client mappings started adding thousands of pages per boot; at
8192 slots with an 8-probe window, inserts were silently dropped and
lookups missed at half load. 32768 slots and a 32-probe window keep the
mapped set intact.

tools/guest-diag/{guest_pmap_all.py,bridge_sems.sh} translate the CUDA
client's /dev/nvidia* mappings through the guest's own pagemap and inject
each present page into the service as a sysmem pin. With the client's
semaphore pages pinned the service applies their releases into the page
the client polls (verified: applied 67 -> 109, misses stop growing for
those VAs), where before every write went to the vidmem shadow.

1 files changed, 42 insertions, 0 deletions

diff --git a/tools/guest-diag/bridge_sems.sh b/tools/guest-diag/bridge_sems.sh
new file mode 100755
index 0000000..067a9ee
--- /dev/null
+++ b/tools/guest-diag/bridge_sems.sh
@@ -0,0 +1,42 @@
+#!/bin/bash
+# bridge_sems.sh - bridge the guest's own page table view into the host service.
+#
+# The emulation cannot (yet) translate the CUDA client's sysmem semaphores
+# (guest RAM pages mapped at VAs above the vidmem window base), so releases
+# for them are counted as misses. This helper translates the client's
+# /dev/nvidia* mappings through the guest's /proc/<pid>/pagemap (the guest
+# kernel owns the mapping) and injects each present page into the service as
+# a sysmem pin.
+#
+# Usage: tools/guest-diag/bridge_sems.sh [iterations] [prefix-hex]
+set -u
+cd "$(dirname "$0")/../.."
+
+ITER=${1:-1}
+START=${2:-0}
+VMEXEC=./scripts/vgpu_vm.py
+
+for i in $(seq 1 "$ITER"); do
+    pid=$(timeout 20 $VMEXEC exec --timeout 15 'pgrep -x cdt | head -1' 2>/dev/null | tail -1)
+    [ -n "$pid" ] || { echo "no cdt pid"; exit 1; }
+
+    b64=$(base64 -w0 tools/guest-diag/guest_pmap_all.py)
+    timeout 30 $VMEXEC exec --timeout 25 "echo '$b64' | base64 -d > /tmp/pmapall.py" >/dev/null 2>&1
+
+    cursor=$START
+    total=0
+    for chunk in $(seq 1 40); do
+        mapfile -t lines < <(timeout 60 $VMEXEC exec --timeout 50 "sudo python3 /tmp/pmapall.py $pid $cursor 100" 2>/dev/null | grep -aE '^0x[0-9a-f]+,0x[0-9a-f]+$')
+        [ "${#lines[@]}" -gt 0 ] || break
+        for l in "${lines[@]}"; do
+            va=${l%%,*}
+            pa=${l##*,}
+            echo "$va,$pa,0x1000,1" | sudo tee /sys/module/open_iov_vfio/parameters/sema_map_add >/dev/null
+        done
+        total=$((total + ${#lines[@]}))
+        last_va=${lines[-1]%%,*}
+        cursor=$(printf '0x%x' $(( 16#${last_va#0x} + 0x1000 )))
+    done
+    echo "iter $i: pinned $total pages (pid $pid)"
+    sleep 2
+done
\ No newline at end of file