{"id":"CVE-2026-99120","kind":"vulnerability","aliases":["CVE-2026-99120"],"related":[],"title":"Linux KVM on AMD: unexpected shadow-stack bit crashes host kernel with out-of-bounds access","description":"In the Linux kernel, the following vulnerability has been resolved:\n\nKVM: x86/mmu: Cope with the SS bit being set in #NPF exit codes\n\nWhen running Hyper-V from Windows 11, KVM is seeing the SS bit set\nin nested page fault error codes produced for shadowed NPT.  The AMD\nmanual is unclear about when SS is set, including whether it can be set\nwhen SupervisorShadowStackEn (bit 6) is not set in the MISC_CTL field\nof the VMCB.\n\nNevertheless, the unexpected SS bit causes out of bounds accesses in\npermission_fault(), seen as either WARNs or UBSAN reports:\n\n  kernel: UBSAN: array-index-out-of-bounds in arch/x86/kvm/mmu.h:225:27\n  kernel: index 34 is out of range for type 'u16 [16]'\n\n(in this case, the page fault error code is 34*2=0x44, i.e. SS|U; the\nreport shows SS|U|W happening as well).  Fix the handling of unknown\nerror codes by dropping SS and (with a WARN) any bits not among those\nthat permission_fault() expects to receive.\n\nAt the same time, given the uncertainty about when hardware sets SS,\nfollow the processor's behavior when constructing the fault's error\ncode, and preserve the SS bit as passed to FNAME(walk_addr_generic).\nWhile the behavior is surprising, there's a possibility that Hyper-V\nexpects it (Hyper-V does not even enable CET unless it finds MBEC/GMET!),\nso retain the information when reflecting the fault to L1 instead of\nunconditionally discarding it.\n\nBisection shows that this started happening with commit 687ee95c1b6d (\"KVM:\nnSVM: enable GMET for guests\"), even though SHSTK was available to guests\nsince Linux 6.18.  However, this is not because bit 1 of MISC_CTL (GMET)\nsomehow enables reporting the SS bit; rather, it's simply that Hyper-V\nnever uses CET unless GMET is also present.  The WARN is reproducible\neven with gmet=0, with some hacking of kvm-unit-tests's cet.c testcase\nto ensure that the first access to the shadow stack triggers a page fault.","summary":"When Hyper-V runs as a guest under KVM on AMD, nested page fault error codes can carry an unexpected SS bit, which causes out-of-bounds array accesses in the KVM MMU code. The result is a kernel WARN or UBSAN report, and the sources do not describe any attacker-controlled code execution.","severity":"unknown","cvss":null,"epss":null,"exploited":false,"exploited_at":null,"products":[],"ecosystems":[],"tags":["amd","hyper-v","infrastructure","kvm","linux","nested-virtualization"],"not_affected_if":["intel-only-hosts","no-nested-virtualization"],"how_to_check":"- Check whether the host runs KVM on AMD hardware with nested virtualization, for example `grep -m1 'model name' /proc/cpuinfo` and `lsmod | grep kvm_amd`.\n- Look for kernel log lines containing `UBSAN: array-index-out-of-bounds` or WARN traces in `arch/x86/kvm/mmu.h` with `dmesg | grep -i kvm`.\n- Compare the running kernel with `uname -r` against the fixed kernel version, which the sources do not state: unknown.","fix":"- Install a Linux kernel that includes the upstream commit resolving this KVM x86/mmu issue; the exact fixed version is unknown from the sources.\n- Reboot the host after updating the kernel package.","status":"open","published_at":"2026-10-28T11:17:31+00:00","first_seen_at":"2026-10-28T11:32:34+00:00","triaged_at":"2026-10-28T11:37:35+00:00","updated_at":"2026-10-28T11:37:35+00:00","category":"infrastructure","confidence":0.5,"triage":{"at":"2026-10-28T11:37:35Z","model":"claude-haiku-5-5","output":{"fix":"- Install a Linux kernel that includes the upstream commit resolving this KVM x86/mmu issue; the exact fixed version is unknown from the sources.\n- Reboot the host after updating the kernel package.","reach":"unknown","title":"Linux KVM on AMD: unexpected shadow-stack bit crashes host kernel with out-of-bounds access","impact":"low","summary":"When Hyper-V runs as a guest under KVM on AMD, nested page fault error codes can carry an unexpected SS bit, which causes out-of-bounds array accesses in the KVM MMU code. The result is a kernel WARN or UBSAN report, and the sources do not describe any attacker-controlled code execution.","category":"infrastructure","nicknames":[],"who_cares":["linux","kvm","amd","nested-virtualization","hyper-v"],"confidence":0.5,"needs_auth":"unknown","worst_case":"Kernel warnings or undefined-behavior reports on the hypervisor host","analyst_note":"Find the fixed kernel version and upstream commit for this KVM fix at https://nvd.nist.gov/vuln/detail/CVE-2026-99120 and the kernel.org stable advisories.","how_to_check":"- Check whether the host runs KVM on AMD hardware with nested virtualization, for example `grep -m1 'model name' /proc/cpuinfo` and `lsmod | grep kvm_amd`.\n- Look for kernel log lines containing `UBSAN: array-index-out-of-bounds` or WARN traces in `arch/x86/kvm/mmu.h` with `dmesg | grep -i kvm`.\n- Compare the running kernel with `uname -r` against the fixed kernel version, which the sources do not state: unknown.","needs_analyst":true,"not_affected_if":["intel-only-hosts","no-nested-virtualization"],"who_is_affected":"Hosts running KVM on AMD processors that run nested guests, particularly Hyper-V on Windows 11 as an L1 guest, are the reported case. The sources say the SS bit appears in nested page faults with shadowed NPT and that the bug was triggered with gmet=0 as well. Teams that do not run KVM on AMD with nested virtualization can skip it.","user_interaction":"no","needs_special_setup":"yes","severity_if_unknown":"low"},"cost_usd":0.000803,"input_tokens":3971,"output_tokens":811,"severity_before":"unknown"},"analyst_status":null,"analyst_note":null,"search":"'-2026':20A,23A '-99120':21A,24A '0':406C '0x44':225C '1':372C '11':122C '16':213C '2':224C '225':201C '27':202C '34':205C,223C '6':159C '6.18':365C '687ee95c1b6d':349C 'access':18A,65B,180C,423C 'addr':297C 'also':397C 'amd':4A,25B,46B,141C 'among':255C 'arch/x86/kvm/mmu.h':200C 'array':64B,194C 'array-index-out-of-bound':193C 'attack':88B 'attacker-control':87B 'avail':360C 'behavior':279C,301C 'bisect':341C 'bit':9A,57B,108C,128C,158C,175C,253C,291C,371C,382C 'bound':17A,63B,179C,198C 'carri':53B 'case':216C 'caus':59B,176C 'cet':321C,393C 'cet.c':416C 'code':51B,70B,90B,114C,135C,221C,244C,286C 'commit':348C 'construct':281C 'control':89B 'cope':104C 'crash':10A 'ctl':166C,375C 'cve':19A,22A 'describ':85B 'discard':339C 'drop':246C 'either':186C 'enabl':320C,352C,378C 'ensur':419C 'error':50B,134C,220C,243C,285C 'even':319C,356C,403C 'execut':91B 'exit':113C 'expect':260C,312C 'fault':49B,133C,183C,219C,259C,283C,333C,431C 'field':167C 'find':324C 'first':422C 'fix':238C 'fname':295C 'follow':97C,275C 'generic':298C 'given':267C 'gmet':353C,376C,395C,405C 'guest':42B,355C,362C 'hack':409C 'handl':240C 'happen':235C,346C 'hardwar':272C 'host':11A 'howev':366C 'hyper':27B,37B,118C,310C,315C,389C 'hyper-v':26B,36B,117C,309C,314C,388C 'i.e':226C 'includ':150C 'index':195C,204C 'inform':329C 'infrastructur':29B 'instead':336C 'kernel':12A,75B,95C,191C,203C 'kvm':2A,30B,44B,68B,102C,123C,350C,412C 'kvm-unit-test':411C 'l1':335C 'linux':1A,31B,94C,364C 'manual':142C 'mbec/gmet':325C 'misc':165C,374C 'mmu':69B 'nest':33B,47B,131C 'nested-virtu':32B 'never':391C 'nevertheless':171C 'npf':112C 'npt':139C 'nsvm':351C 'out-of-bound':14A,60B 'page':48B,132C,218C,430C 'pass':293C 'permiss':182C,258C 'possibl':307C 'present':398C 'preserv':288C 'processor':277C 'produc':136C 'rang':209C 'rather':383C 'receiv':262C 'reflect':331C 'report':79B,190C,230C,379C 'reproduc':402C 'resolv':101C 'result':72B 'retain':327C 'run':39B,116C 'see':125C 'seen':184C 'set':110C,129C,149C,155C,162C,273C 'shadow':7A,138C,426C 'shadow-stack':6A 'show':231C,342C 'shstk':358C 'simpli':386C 'sinc':363C 'somehow':377C 'sourc':82B 'ss':56B,107C,127C,147C,174C,227C,232C,247C,274C,290C,381C 'stack':8A,427C 'start':345C 'supervisorshadowstacken':157C 'surpris':303C 'test':414C 'testcas':417C 'though':357C 'time':266C 'trigger':428C 'type':211C 'u':228C,233C 'u16':212C 'ubsan':78B,189C,192C 'uncertainti':269C 'unclear':144C 'uncondit':338C 'unexpect':5A,55B,173C 'unit':413C 'unknown':242C 'unless':322C,394C 'use':392C 'v':28B,38B,119C,311C,316C,390C 'virtual':34B 'vmcb':170C 'vulner':98C 'w':234C 'walk':296C 'warn':76B,187C,251C,400C 'well':237C 'whether':151C 'window':121C 'x86/mmu':103C","nicknames":[],"mitigations":[],"who_affected":"Hosts running KVM on AMD processors that run nested guests, particularly Hyper-V on Windows 11 as an L1 guest, are the reported case. The sources say the SS bit appears in nested page faults with shadowed NPT and that the bug was triggered with gmet=0 as well. Teams that do not run KVM on AMD with nested virtualization can skip it.","tier":"C","verified":{},"analyst_revisit_at":null,"first_public_at":"2026-10-28T11:17:31+00:00","kev":null,"cvss_vector":null,"cvss_by":null,"poc_count":0,"sources":[{"source":"nvd","external_id":"CVE-2026-99120","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-99120","severity":null,"published_at":"2026-10-28T11:17:31+00:00","modified_at":"2026-10-28T11:17:31+00:00","first_seen_at":"2026-10-28T11:32:34+00:00","fetched_at":"2026-10-28T11:32:34+00:00","note":null}],"documents":[],"notice":"Free to read at a human pace. An API with keys and an MCP server are in preparation; early access: hello@infrastack.ai"}