From 17e1f522e4e9e04d13c4ca9495a2f4f1bdc58c36 Mon Sep 17 00:00:00 2001 From: Arthur Hanson Rasmusson Date: Mon, 28 Sep 2026 07:34:34 -0400 Subject: sema-svc: apply releases promptly and decode compute-push releases The content sweep now: - decodes a release whose method stream continues with any INC_METHOD, not only the CE LAUNCH_DMA group, so the releases at the head of a compute/GR channel's first push are applied too; - remembers releases whose target VA had no translation (deferred list) and retries them every service pass, so a VA->PA injection lands them immediately instead of after the next full sweep; - re-arms the fast sliced sweep for a few seconds after an injection; - treats a channel whose USERD/ring cannot be resolved as stalled, so the sweep keeps looking for the releases its unreadable pushes carry. tools: add per-cdt client-pin helper (sem pages + /dev/nvidia0 windows), sem-page scanner and memprobe fixes (short-read guard, --brief). Live-verified: injecting a client sem page applied 12 deferred releases within a second; the service kept applying (89 -> 202) while cdt ran. --- tools/guest-diag/pin_client_sems.sh | 66 +++++++++++++++++++++++++++++++++++++ 1 file changed, 66 insertions(+) create mode 100755 tools/guest-diag/pin_client_sems.sh (limited to 'tools/guest-diag/pin_client_sems.sh') diff --git a/tools/guest-diag/pin_client_sems.sh b/tools/guest-diag/pin_client_sems.sh new file mode 100755 index 0000000..82d2789 --- /dev/null +++ b/tools/guest-diag/pin_client_sems.sh @@ -0,0 +1,66 @@ +#!/bin/bash +# pin_client_sems.sh - bridge the CUDA client's device mappings into the +# emulation for the live cdt process. +# +# Two things are pinned, both per-process (a new cdt maps new physical pages +# at the same VAs, so re-run this for every cdt run): +# 1. the /dev/nvidia0 window: the driver's channel rings/USERDs (the GR +# channels' GPFIFO VAs live here), so the pump can run their pushes; +# 2. the semaphore pages libcuda's trackers poll (found by scanning the +# client's memory for sem references, then translated through the guest +# pagemap). +# +# Usage: tools/guest-diag/pin_client_sems.sh [iterations] [sleep-secs] +set -u +cd "$(dirname "$(readlink -f "$0")")/../.." +VMEXEC=./scripts/vgpu_vm.py +ITER=${1:-1} +SLEEP=${2:-10} + +PID=$(timeout 20 $VMEXEC exec --timeout 15 'pgrep -x cdt | head -1' 2>/dev/null | tail -1) +[ -n "$PID" ] || { echo "no cdt pid"; exit 1; } +echo "cdt pid=$PID" + +B64=$(gzip -9 < tools/guest-diag/sempages.py | base64 -w0) +timeout 30 $VMEXEC exec --timeout 25 "echo '$B64' | base64 -d | gunzip > /tmp/sempages.py" >/dev/null 2>&1 +B64=$(gzip -9 < tools/guest-diag/memprobe.py | base64 -w0) +timeout 30 $VMEXEC exec --timeout 25 "echo '$B64' | base64 -d | gunzip > /tmp/memprobe.py" >/dev/null 2>&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 + +pin_range() { # pin_range