CVE-2026-98949

In the Linux kernel, the following vulnerability has been resolved:

drm/imagination: Fix reference and vm_bo handling in remap()

When a map overlaps part of an existing mapping, pvr_vm_gpuva_remap() splits the mapping into prev/next parts covering what the request did not take, instead of creating something new. It gets two things wrong.

• A GEM reference is taken for each part but it's never needed and never dropped, so it leaks one reference per split (remap-next-in-2m in tests/imagination/pvr_vm_map.c):

  CRITICAL: 33558528 bytes of shmem still held after close

• The parts still belong to the original object being split, but pvr_vm_gpuva_remap() links them to ctx->gpuvm_bo, the new object's vm_bo:

  prev_va --obj--> BO_A BO_B <--obj-- vm_bo (ctx->gpuvm_bo) _link_/ (mismatch: BO_A != BO_B)

  drm_gpuva_link() catches the mismatch:

  WARNING: drivers/gpu/drm/drm_gpuvm.c:2108 at drm_gpuva_link+0x2ec/0x310 drm_WARN_ON(obj != vm_bo->obj) Call trace: drm_gpuva_link pvr_vm_gpuva_remap __drm_gpuvm_sm_map pvr_vm_map pvr_ioctl_vm_map

  Link them to op->remap.unmap->va->vm_bo instead.

The locking of the GPUVA lists of the other objects touched by a split is not addressed here; it is handled by switching the GPUVM to immediate mode.

In the Linux kernel, the following vulnerability has been resolved:

drm/imagination: Fix reference and vm_bo handling in remap()

When a map overlaps part of an existing mapping, pvr_vm_gpuva_remap() splits the mapping into prev/next parts covering what the request did not take, instead of creating something new. It gets two things wrong.

• A GEM reference is taken for each part but it's never needed and never dropped, so it leaks one reference per split (remap-next-in-2m in tests/imagination/pvr_vm_map.c):

  CRITICAL: 33558528 bytes of shmem still held after close

• The parts still belong to the original object being split, but pvr_vm_gpuva_remap() links them to ctx->gpuvm_bo, the new object's vm_bo:

  prev_va --obj--> BO_A BO_B <--obj-- vm_bo (ctx->gpuvm_bo) _link_/ (mismatch: BO_A != BO_B)

  drm_gpuva_link() catches the mismatch:

  WARNING: drivers/gpu/drm/drm_gpuvm.c:2108 at drm_gpuva_link+0x2ec/0x310 drm_WARN_ON(obj != vm_bo->obj) Call trace: drm_gpuva_link pvr_vm_gpuva_remap __drm_gpuvm_sm_map pvr_vm_map pvr_ioctl_vm_map

  Link them to op->remap.unmap->va->vm_bo instead.

The locking of the GPUVA lists of the other objects touched by a split is not addressed here; it is handled by switching the GPUVM to immediate mode.

In the Linux kernel, the following vulnerability has been resolved:

drm/imagination: Fix reference and vm_bo handling in remap()

When a map overlaps part of an existing mapping, pvr_vm_gpuva_remap() splits the mapping into prev/next parts covering what the request did not take, instead of creating something new. It gets two things wrong.

• A GEM reference is taken for each part but it's never needed and never dropped, so it leaks one reference per split (remap-next-in-2m in tests/imagination/pvr_vm_map.c):

  CRITICAL: 33558528 bytes of shmem still held after close

• The parts still belong to the original object being split, but pvr_vm_gpuva_remap() links them to ctx->gpuvm_bo, the new object's vm_bo:

  prev_va --obj--> BO_A BO_B <--obj-- vm_bo (ctx->gpuvm_bo) _link_/ (mismatch: BO_A != BO_B)

  drm_gpuva_link() catches the mismatch:

  WARNING: drivers/gpu/drm/drm_gpuvm.c:2108 at drm_gpuva_link+0x2ec/0x310 drm_WARN_ON(obj != vm_bo->obj) Call trace: drm_gpuva_link pvr_vm_gpuva_remap __drm_gpuvm_sm_map pvr_vm_map pvr_ioctl_vm_map

  Link them to op->remap.unmap->va->vm_bo instead.

The locking of the GPUVA lists of the other objects touched by a split is not addressed here; it is handled by switching the GPUVM to immediate mode.

In the Linux kernel, the following vulnerability has been resolved:

drm/imagination: Fix reference and vm_bo handling in remap()

When a map overlaps part of an existing mapping, pvr_vm_gpuva_remap() splits the mapping into prev/next parts covering what the request did not take, instead of creating something new. It gets two things wrong.

• A GEM reference is taken for each part but it's never needed and never dropped, so it leaks one reference per split (remap-next-in-2m in tests/imagination/pvr_vm_map.c):

  CRITICAL: 33558528 bytes of shmem still held after close

• The parts still belong to the original object being split, but pvr_vm_gpuva_remap() links them to ctx->gpuvm_bo, the new object's vm_bo:

  prev_va --obj--> BO_A BO_B <--obj-- vm_bo (ctx->gpuvm_bo) _link_/ (mismatch: BO_A != BO_B)

  drm_gpuva_link() catches the mismatch:

  WARNING: drivers/gpu/drm/drm_gpuvm.c:2108 at drm_gpuva_link+0x2ec/0x310 drm_WARN_ON(obj != vm_bo->obj) Call trace: drm_gpuva_link pvr_vm_gpuva_remap __drm_gpuvm_sm_map pvr_vm_map pvr_ioctl_vm_map

  Link them to op->remap.unmap->va->vm_bo instead.

The locking of the GPUVA lists of the other objects touched by a split is not addressed here; it is handled by switching the GPUVM to immediate mode.

In the Linux kernel, the following vulnerability has been resolved:

drm/imagination: Fix reference and vm_bo handling in remap()

When a map overlaps part of an existing mapping, pvr_vm_gpuva_remap() splits the mapping into prev/next parts covering what the request did not take, instead of creating something new. It gets two things wrong.

• A GEM reference is taken for each part but it's never needed and never dropped, so it leaks one reference per split (remap-next-in-2m in tests/imagination/pvr_vm_map.c):

  CRITICAL: 33558528 bytes of shmem still held after close

• The parts still belong to the original object being split, but pvr_vm_gpuva_remap() links them to ctx->gpuvm_bo, the new object's vm_bo:

  prev_va --obj--> BO_A BO_B <--obj-- vm_bo (ctx->gpuvm_bo) _link_/ (mismatch: BO_A != BO_B)

  drm_gpuva_link() catches the mismatch:

  WARNING: drivers/gpu/drm/drm_gpuvm.c:2108 at drm_gpuva_link+0x2ec/0x310 drm_WARN_ON(obj != vm_bo->obj) Call trace: drm_gpuva_link pvr_vm_gpuva_remap __drm_gpuvm_sm_map pvr_vm_map pvr_ioctl_vm_map

  Link them to op->remap.unmap->va->vm_bo instead.

The locking of the GPUVA lists of the other objects touched by a split is not addressed here; it is handled by switching the GPUVM to immediate mode.

In the Linux kernel, the following vulnerability has been resolved: drm/imagination: Fix reference and vm_bo handling in remap() When a map overlaps part of an existing mapping, pvr_vm_gpuva_remap() splits the mapping into prev/next parts covering what the request did not take, instead of creating something new. It gets two things wrong. - A GEM reference is taken for each part but it's never needed and never dropped, so it leaks one reference per split (remap-next-in-2m in tests/imagination/pvr_vm_map.c): CRITICAL: 33558528 bytes of shmem still held after close - The parts still belong to the original object being split, but pvr_vm_gpuva_remap() links them to ctx->gpuvm_bo, the new object's vm_bo: prev_va --obj--> BO_A BO_B <--obj-- vm_bo (ctx->gpuvm_bo) _link_/ (mismatch: BO_A != BO_B) drm_gpuva_link() catches the mismatch: WARNING: drivers/gpu/drm/drm_gpuvm.c:2108 at drm_gpuva_link+0x2ec/0x310 drm_WARN_ON(obj != vm_bo->obj) Call trace: drm_gpuva_link pvr_vm_gpuva_remap __drm_gpuvm_sm_map pvr_vm_map pvr_ioctl_vm_map Link them to op->remap.unmap->va->vm_bo instead. The locking of the GPUVA lists of the other objects touched by a split is not addressed here; it is handled by switching the GPUVM to immediate mode.

The SUSE Linux Enterprise 15 SP7 kernel was updated to fix various security issues:

The following security issues were fixed:

• CVE-2025-68219: cifs: fix memory leak in smb3_fs_context_parse_param error path (bsc#1255222).
• CVE-2026-31708: smb: client: fix OOB read in smb2_ioctl_query_info QUERY_INFO path (bsc#1264022).
• CVE-2026-31709: smb: client: validate the whole DACL before rewriting it in cifsacl (bsc#1264024).
• CVE-2026-43350: smb: client: require a full NFS mode SID before reading mode bits (bsc#1264985).
• CVE-2026-46155: smb/client: fix out-of-bounds read in smb2_compound_op() (bsc#1267720).
• CVE-2026-64370: posix-cpu-timers: Fix pid refcount leak in do_cpu_nanosleep() error path (bsc#1273805).
• CVE-2026-64380: smb: client: harden POSIX SID length parsing (bsc#1273557).
• CVE-2026-64430: NTB: epf: Avoid calling pci_irq_vector() from hardirq context (bsc#1274257).
• CVE-2026-64435: audit: Fix data races of skb_queue_len() readers on audit_queue (bsc#1274080).
• CVE-2026-68186: binfmt_misc: set have_execfd only once the interpreter is opened (bsc#1274949).
• CVE-2026-68436: drm/amd/display: use kvzalloc to allocate struct dc (bsc#1276275).
• CVE-2026-68447: drm/amdkfd: clamp v9 CRIU control stack checkpoint copy to BO size (bsc#1276280).
• CVE-2026-68460: f2fs: fix potential deadlock in f2fs_balance_fs() (bsc#1276774).
• CVE-2026-72010: cgroup/cpuset: rebind mm mempolicy to effective_mems, not mems_allowed (bsc#1276776).
• CVE-2026-72073: mmc: vub300: fix use-after-free on probe failure (bsc#1277157).
• CVE-2026-72110: bpf,fork: wipe ->bpf_storage before bailouts that access it (bsc#1277357).
• CVE-2026-72111: bpf: Reset register bounds before narrowing retval range in check_mem_access() (bsc#1277389).
• CVE-2026-72250: netfilter: nf_conntrack_reasm: guard mac_header adjustment after IPv6 defrag (bsc#1277663).
• CVE-2026-72253: netfilter: nf_conntrack_sip: validate skb_dst() before accessing it (bsc#1277665).
• CVE-2026-72318: cifs: validate DFS referral string offsets (bsc#1275939).
• CVE-2026-72405: net: udp_tunnel: prevent double queueing in udp_tunnel_nic_device_sync (bsc#1278033).
• CVE-2026-72416: netfilter: nft_compat: ebtables emulation must reject non-bridge targets (bsc#1278041).
• CVE-2026-72428: bpf: Fix stack slot index in nospec checks (bsc#1278050).
• CVE-2026-74368: wifi: ath12k: fix memory leak in ath12k_wifi7_dp_rx_h_verify_tkip_mic() (bsc#1277103).
• CVE-2026-74472: ublk: reset kernel-owned dev_info fields in ublk_ctrl_add_dev() (bsc#1277785).
• CVE-2026-74613: vsock/virtio: avoid refilling the RX queue after teardown (bsc#1277903).
• CVE-2026-74620: net/sched: act_gact, act_police: range check the fallback control action (bsc#1277044).
• CVE-2026-74624: netfilter: nf_conntrack: defer invalid log until after unlock (bsc#1277599).
• CVE-2026-74715: net: netfilter: Make ct zone opts configurable for bpf ct helpers (bsc#1277890).
• CVE-2026-74746: netfilter: flowtable: publish GC-visible tuple last (bsc#1278101).
• CVE-2026-74748: netfilter: ipset: fix refcount race between list:set GC and swap (bsc#1277578).
• CVE-2026-80572: Input: byd - synchronize timer deletion before freeing private data (bsc#1277338).
• CVE-2026-80602: perf/x86/amd/lbr: Fix kernel address leakage (bsc#1278123).
• CVE-2026-80637: netfilter: synproxy: fix unaligned memory access in timestamp adjustment (bsc#1278343).
• CVE-2026-80717: sctp: validate Adaptation Indication parameter length (bsc#1277836).
• CVE-2026-80755: selinux: reject a permission value exceeding the class permission count (bsc#1278415).
• CVE-2026-80757: selinux: reject a class permission count below its inherited common (bsc#1278711).
• CVE-2026-80808: ext4: stop retrying saturated xattr cache entries (bsc#1279795).
• CVE-2026-80809: ocfs2: fix missing metadata reservation for large xattrs (bsc#1279606).
• CVE-2026-80855: fuse: fix invalidate lock leak on open O_TRUNC DAX failure (bsc#1279564).
• CVE-2026-80879: ocfs2: fix circular locking dependency…

In the Linux kernel, the following vulnerability has been resolved: drm/imagination: Fix reference and vm_bo handling in remap() When a map overlaps part of an existing mapping, pvr_vm_gpuva_remap() splits the mapping into prev/next parts covering what the request did not take, instead of creating something new. It gets two things wrong. - A GEM reference is taken for each part but it's never needed and never dropped, so it leaks one reference per split (remap-next-in-2m in tests/imagination/pvr_vm_map.c): CRITICAL: 33558528 bytes of shmem still held after close - The parts still belong to the original object being split, but pvr_vm_gpuva_remap() links them to ctx->gpuvm_bo, the new object's vm_bo: prev_va --obj--> BO_A BO_B <--obj-- vm_bo (ctx->gpuvm_bo) _link_/ (mismatch: BO_A != BO_B) drm_gpuva_link() catches the mismatch: WARNING: drivers/gpu/drm/drm_gpuvm.c:2108 at drm_gpuva_link+0x2ec/0x310 drm_WARN_ON(obj != vm_bo->obj) Call trace: drm_gpuva_link pvr_vm_gpuva_remap __drm_gpuvm_sm_map pvr_vm_map pvr_ioctl_vm_map Link them to op->remap.unmap->va->vm_bo instead. The locking of the GPUVA lists of the other objects touched by a split is not addressed here; it is handled by switching the GPUVM to immediate mode.