1iTLB multihit
2=============
3
4iTLB multihit is an erratum where some processors may incur a machine check
5error, possibly resulting in an unrecoverable CPU lockup, when an
6instruction fetch hits multiple entries in the instruction TLB. This can
7occur when the page size is changed along with either the physical address
8or cache type. A malicious guest running on a virtualized system can
9exploit this erratum to perform a denial of service attack.
10
11
12Affected processors
13-------------------
14
15Variations of this erratum are present on most Intel Core and Xeon processor
16models. The erratum is not present on:
17
18   - non-Intel processors
19
20   - Some Atoms (Airmont, Bonnell, Goldmont, GoldmontPlus, Saltwell, Silvermont)
21
22   - Intel processors that have the PSCHANGE_MC_NO bit set in the
23     IA32_ARCH_CAPABILITIES MSR.
24
25
26Related CVEs
27------------
28
29The following CVE entry is related to this issue:
30
31   ==============  =================================================
32   CVE-2018-12207  Machine Check Error Avoidance on Page Size Change
33   ==============  =================================================
34
35
36Problem
37-------
38
39Privileged software, including OS and virtual machine managers (VMM), are in
40charge of memory management. A key component in memory management is the control
41of the page tables. Modern processors use virtual memory, a technique that creates
42the illusion of a very large memory for processors. This virtual space is split
43into pages of a given size. Page tables translate virtual addresses to physical
44addresses.
45
46To reduce latency when performing a virtual to physical address translation,
47processors include a structure, called TLB, that caches recent translations.
48There are separate TLBs for instruction (iTLB) and data (dTLB).
49
50Under this errata, instructions are fetched from a linear address translated
51using a 4 KB translation cached in the iTLB. Privileged software modifies the
52paging structure so that the same linear address using large page size (2 MB, 4
53MB, 1 GB) with a different physical address or memory type.  After the page
54structure modification but before the software invalidates any iTLB entries for
55the linear address, a code fetch that happens on the same linear address may
56cause a machine-check error which can result in a system hang or shutdown.
57
58
59Attack scenarios
60----------------
61
62Attacks against the iTLB multihit erratum can be mounted from malicious
63guests in a virtualized system.
64
65
66iTLB multihit system information
67--------------------------------
68
69The Linux kernel provides a sysfs interface to enumerate the current iTLB
70multihit status of the system:whether the system is vulnerable and which
71mitigations are active. The relevant sysfs file is:
72
73/sys/devices/system/cpu/vulnerabilities/itlb_multihit
74
75The possible values in this file are:
76
77.. list-table::
78
79     * - Not affected
80       - The processor is not vulnerable.
81     * - KVM: Mitigation: Split huge pages
82       - Software changes mitigate this issue.
83     * - KVM: Vulnerable
84       - The processor is vulnerable, but no mitigation enabled
85
86
87Enumeration of the erratum
88--------------------------------
89
90A new bit has been allocated in the IA32_ARCH_CAPABILITIES (PSCHANGE_MC_NO) msr
91and will be set on CPU's which are mitigated against this issue.
92
93   =======================================   ===========   ===============================
94   IA32_ARCH_CAPABILITIES MSR                Not present   Possibly vulnerable,check model
95   IA32_ARCH_CAPABILITIES[PSCHANGE_MC_NO]    '0'           Likely vulnerable,check model
96   IA32_ARCH_CAPABILITIES[PSCHANGE_MC_NO]    '1'           Not vulnerable
97   =======================================   ===========   ===============================
98
99
100Mitigation mechanism
101-------------------------
102
103This erratum can be mitigated by restricting the use of large page sizes to
104non-executable pages.  This forces all iTLB entries to be 4K, and removes
105the possibility of multiple hits.
106
107In order to mitigate the vulnerability, KVM initially marks all huge pages
108as non-executable. If the guest attempts to execute in one of those pages,
109the page is broken down into 4K pages, which are then marked executable.
110
111If EPT is disabled or not available on the host, KVM is in control of TLB
112flushes and the problematic situation cannot happen.  However, the shadow
113EPT paging mechanism used by nested virtualization is vulnerable, because
114the nested guest can trigger multiple iTLB hits by modifying its own
115(non-nested) page tables.  For simplicity, KVM will make large pages
116non-executable in all shadow paging modes.
117
118Mitigation control on the kernel command line and KVM - module parameter
119------------------------------------------------------------------------
120
121The KVM hypervisor mitigation mechanism for marking huge pages as
122non-executable can be controlled with a module parameter "nx_huge_pages=".
123The kernel command line allows to control the iTLB multihit mitigations at
124boot time with the option "kvm.nx_huge_pages=".
125
126The valid arguments for these options are:
127
128  ==========  ================================================================
129  force       Mitigation is enabled. In this case, the mitigation implements
130              non-executable huge pages in Linux kernel KVM module. All huge
131              pages in the EPT are marked as non-executable.
132              If a guest attempts to execute in one of those pages, the page is
133              broken down into 4K pages, which are then marked executable.
134
135  off	      Mitigation is disabled.
136
137  auto        Enable mitigation only if the platform is affected and the kernel
138              was not booted with the "mitigations=off" command line parameter.
139	      This is the default option.
140  ==========  ================================================================
141
142
143Mitigation selection guide
144--------------------------
145
1461. No virtualization in use
147^^^^^^^^^^^^^^^^^^^^^^^^^^^
148
149   The system is protected by the kernel unconditionally and no further
150   action is required.
151
1522. Virtualization with trusted guests
153^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
154
155   If the guest comes from a trusted source, you may assume that the guest will
156   not attempt to maliciously exploit these errata and no further action is
157   required.
158
1593. Virtualization with untrusted guests
160^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
161   If the guest comes from an untrusted source, the guest host kernel will need
162   to apply iTLB multihit mitigation via the kernel command line or kvm
163   module parameter.
164