fix(gpu): changing the mode of a GPU leaves the other devices of its slot in their VM

scripts/gpu_tpu/add_gpu_vm.sh

@@ -458,12 +458,14 @@ _is_pci_slot_assigned_to_vm() {
 
     local pci_full="$1"
     local vmid="$2"
-    local slot_base
-    slot_base="${pci_full#0000:}"
-    slot_base="${slot_base%.*}"   # 01:00
+    local bdf slot_base
+    bdf="${pci_full#0000:}"       # 01:00.0
+    slot_base="${bdf%.*}"         # 01:00
 
+    # The device itself or its whole slot; another function at the same slot
+    # is a different device.
     qm config "$vmid" 2>/dev/null \
-        | grep -qE "^hostpci[0-9]+:.*(0000:)?${slot_base}(\\.[0-7])?([,[:space:]]|$)"
+        | grep -qE "^hostpci[0-9]+:.*(0000:)?(${bdf}|${slot_base})([,;[:space:]]|$)"
 }
 
 # Match a specific PCI function when possible.
@@ -1453,7 +1455,9 @@ check_switch_mode() {
         local vmid
         vmid=$(basename "$conf" .conf)
         [[ "$vmid" == "$SELECTED_VMID" ]] && continue  # same target VM, no conflict
-        if grep -qE "hostpci[0-9]+:.*${pci_slot}" "$conf"; then
+        # Another VM has the GPU when it has the function of the GPU itself;
+        # a neighbour at the same slot, as a USB controller, is not the GPU.
+        if [[ -n "$(_vm_gpu_hostpci_indexes "$conf" "$SELECTED_GPU_PCI" "${IOMMU_DEVICES[@]}")" ]]; then
             vm_src_id="$vmid"
             vm_src_name=$(grep "^name:" "$conf" 2>/dev/null | awk '{print $2}')
             break
@@ -2008,12 +2012,14 @@ cleanup_vm_config() {
     local src_conf="/etc/pve/qemu-server/${SWITCH_VM_SRC}.conf"
     if [[ -f "$src_conf" ]]; then
         msg_info "$(translate 'Removing GPU from VM') ${SWITCH_VM_SRC}..."
-        # Precise regex: slot must be followed by ".<function>" and a
-        # delimiter. Kept in sync with switch_gpu_mode.sh. A looser
-        # ".*${pci_slot}" would match the slot as a substring and wipe
-        # unrelated hostpci entries (e.g. slot "00:02" matching inside
-        # a dGPU BDF 0000:02:00.0).
-        sed -E -i "/^hostpci[0-9]+:[[:space:]]*(0000:)?${pci_slot}\.[0-7]([,[:space:]]|$)/d" "$src_conf"
+        # Only the entries that carry the GPU are removed: its own function,
+        # the audio at its slot and the devices of its IOMMU group. Other
+        # functions at the same slot stay in the source VM.
+        local _gpu_idx
+        while IFS= read -r _gpu_idx; do
+            [[ -n "$_gpu_idx" ]] || continue
+            _vm_remove_hostpci_index "$SWITCH_VM_SRC" "$_gpu_idx" "$LOG_FILE"
+        done < <(_vm_gpu_hostpci_indexes "$src_conf" "$SELECTED_GPU_PCI" "${IOMMU_DEVICES[@]}")
         msg_ok "$(translate 'GPU removed from VM') ${SWITCH_VM_SRC}" | tee -a "$screen_capture"
 
         # Cascade cleanup: detect audio companions orphaned in the

scripts/gpu_tpu/switch_gpu_mode.sh

@@ -85,6 +85,7 @@ declare -a SELECTED_GPU_IDX=()
 
 declare -a SELECTED_IOMMU_IDS=()
 declare -a SELECTED_PCI_SLOTS=()
+declare -a SELECTED_GPU_BDFS=()
 declare -a SELECTED_NVIDIA_BDFS=()
 declare -a SELECTED_LEGACY_IOMMU_IDS=()
 
@@ -688,6 +689,7 @@ check_sriov_and_block_if_needed() {
 collect_selected_iommu_ids() {
   SELECTED_IOMMU_IDS=()
   SELECTED_PCI_SLOTS=()
+  SELECTED_GPU_BDFS=()
   SELECTED_NVIDIA_BDFS=()
   SELECTED_LEGACY_IOMMU_IDS=()
 
@@ -699,6 +701,7 @@ collect_selected_iommu_ids() {
     slot="${pci#0000:}"
     slot="${slot%.*}"
     SELECTED_PCI_SLOTS+=("$slot")
+    SELECTED_GPU_BDFS+=("$pci")
 
     local -a group_bdfs=()
     mapfile -t group_bdfs < <(_get_iommu_group_bdfs "$pci")
@@ -894,9 +897,12 @@ detect_affected_vms_for_selected() {
   local conf
   for conf in /etc/pve/qemu-server/*.conf; do
     [[ -f "$conf" ]] || continue
-    local matched=false slot
-    for slot in "${SELECTED_PCI_SLOTS[@]}"; do
-      if grep -qE "hostpci[0-9]+:.*(0000:)?${slot}(\\.[0-7])?([,[:space:]]|$)" "$conf"; then
+    # A VM has the GPU when it has the function of the GPU itself. Another
+    # function at the same slot, as a USB controller, is a device of its own.
+    local matched=false gpu
+    for gpu in "${SELECTED_GPU_BDFS[@]}"; do
+      # shellcheck disable=SC2046
+      if [[ -n "$(_vm_gpu_hostpci_indexes "$conf" "$gpu" $(_get_iommu_group_bdfs "$gpu"))" ]]; then
         matched=true
         break
       fi
@@ -978,19 +984,16 @@ apply_vm_action_for_lxc_mode() {
     fi
 
     if [[ "$VM_ACTION" == "remove_gpu_keep_onboot" && -f "$conf" ]]; then
-      # Primary cleanup: strip hostpci lines whose BDF matches any of
-      # the GPU's selected slots. Matches both the PF function (.0) and
-      # any sibling audio or HDMI codec that shares the slot (typical
-      # for discrete NVIDIA/AMD cards where .1 is the HDMI audio).
-      #
-      # Precise regex: the slot must be followed by ".<function>" and
-      # either a delimiter or end-of-line. A looser ".*${slot}" would
-      # match by pure substring and delete unrelated hostpci entries —
-      # e.g. slot "00:02" would match inside "0000:02:00.0" (a dGPU at
-      # 02:00) and wipe both the iGPU and the unrelated dGPU.
-      local slot
-      for slot in "${SELECTED_PCI_SLOTS[@]}"; do
-        sed -E -i "/^hostpci[0-9]+:[[:space:]]*(0000:)?${slot}\.[0-7]([,[:space:]]|$)/d" "$conf"
+      # Primary cleanup: remove the entries that carry the GPU — its own
+      # function, the audio at its slot and the devices of its IOMMU
+      # group. Other functions at the same slot stay in the VM.
+      local gpu _gpu_idx
+      for gpu in "${SELECTED_GPU_BDFS[@]}"; do
+        while IFS= read -r _gpu_idx; do
+          [[ -n "$_gpu_idx" ]] || continue
+          _vm_remove_hostpci_index "$vmid" "$_gpu_idx" "$LOG_FILE"
+        # shellcheck disable=SC2046
+        done < <(_vm_gpu_hostpci_indexes "$conf" "$gpu" $(_get_iommu_group_bdfs "$gpu"))
       done
       msg_ok "$(translate 'GPU removed from VM config') ${vmid}" | tee -a "$screen_capture"
 

scripts/gpu_tpu/switch_gpu_mode_direct.sh

@@ -58,6 +58,7 @@ declare -a SELECTED_GPU_IDX=()
 
 declare -a SELECTED_IOMMU_IDS=()
 declare -a SELECTED_PCI_SLOTS=()
+declare -a SELECTED_GPU_BDFS=()
 declare -a SELECTED_NVIDIA_BDFS=()
 declare -a SELECTED_LEGACY_IOMMU_IDS=()
 
@@ -510,6 +511,7 @@ validate_vm_mode_blocked_ids() {
 collect_selected_iommu_ids() {
   SELECTED_IOMMU_IDS=()
   SELECTED_PCI_SLOTS=()
+  SELECTED_GPU_BDFS=()
   SELECTED_NVIDIA_BDFS=()
   SELECTED_LEGACY_IOMMU_IDS=()
 
@@ -521,6 +523,7 @@ collect_selected_iommu_ids() {
     slot="${pci#0000:}"
     slot="${slot%.*}"
     SELECTED_PCI_SLOTS+=("$slot")
+    SELECTED_GPU_BDFS+=("$pci")
 
     local -a group_bdfs=()
     mapfile -t group_bdfs < <(_get_iommu_group_bdfs "$pci")
@@ -721,9 +724,12 @@ detect_affected_vms_for_selected() {
   local conf
   for conf in /etc/pve/qemu-server/*.conf; do
     [[ -f "$conf" ]] || continue
-    local matched=false slot
-    for slot in "${SELECTED_PCI_SLOTS[@]}"; do
-      if grep -qE "hostpci[0-9]+:.*(0000:)?${slot}(\\.[0-7])?([,[:space:]]|$)" "$conf"; then
+    # A VM has the GPU when it has the function of the GPU itself. Another
+    # function at the same slot, as a USB controller, is a device of its own.
+    local matched=false gpu
+    for gpu in "${SELECTED_GPU_BDFS[@]}"; do
+      # shellcheck disable=SC2046
+      if [[ -n "$(_vm_gpu_hostpci_indexes "$conf" "$gpu" $(_get_iommu_group_bdfs "$gpu"))" ]]; then
         matched=true
         break
       fi
@@ -806,17 +812,16 @@ apply_vm_action_for_lxc_mode() {
     fi
 
     if [[ "$VM_ACTION" == "remove_gpu_keep_onboot" && -f "$conf" ]]; then
-      # Primary cleanup: strip hostpci lines whose BDF matches any of
-      # the GPU's selected slots. Matches both the PF function (.0) and
-      # sibling audio/HDMI codecs (.1, typical for discrete cards).
-      #
-      # Precise regex: the slot must be followed by ".<function>" and a
-      # delimiter. Kept in sync with switch_gpu_mode.sh — a looser
-      # substring match would wipe unrelated hostpci entries (e.g. slot
-      # "00:02" matching as a substring inside a dGPU BDF 0000:02:00.0).
-      local slot
-      for slot in "${SELECTED_PCI_SLOTS[@]}"; do
-        sed -E -i "/^hostpci[0-9]+:[[:space:]]*(0000:)?${slot}\.[0-7]([,[:space:]]|$)/d" "$conf"
+      # Primary cleanup: remove the entries that carry the GPU — its own
+      # function, the audio at its slot and the devices of its IOMMU
+      # group. Other functions at the same slot stay in the VM.
+      local gpu _gpu_idx
+      for gpu in "${SELECTED_GPU_BDFS[@]}"; do
+        while IFS= read -r _gpu_idx; do
+          [[ -n "$_gpu_idx" ]] || continue
+          _vm_remove_hostpci_index "$vmid" "$_gpu_idx" "$LOG_FILE"
+        # shellcheck disable=SC2046
+        done < <(_vm_gpu_hostpci_indexes "$conf" "$gpu" $(_get_iommu_group_bdfs "$gpu"))
       done
       msg_ok "$(translate 'GPU removed from VM config') ${vmid}" | tee -a "$screen_capture"