2019-06-20 23:10:50 +00:00
.. SPDX-License-Identifier: GPL-2.0
Spectre Side Channels
=====================
Spectre is a class of side channel attacks that exploit branch prediction
and speculative execution on modern CPUs to read memory, possibly
bypassing access controls. Speculative execution side channel exploits
do not modify memory but attempt to infer privileged data in the memory.
This document covers Spectre variant 1 and Spectre variant 2.
Affected processors
-------------------
Speculative execution side channel methods affect a wide range of modern
high performance processors, since most modern high speed processors
use branch prediction and speculative execution.
The following CPUs are vulnerable:
- Intel Core, Atom, Pentium, and Xeon processors
- AMD Phenom, EPYC, and Zen processors
- IBM POWER and zSeries processors
- Higher end ARM processors
- Apple CPUs
- Higher end MIPS CPUs
- Likely most other high performance CPUs. Contact your CPU vendor for details.
Whether a processor is affected or not can be read out from the Spectre
vulnerability files in sysfs. See :ref: `spectre_sys_info` .
Related CVEs
------------
The following CVE entries describe Spectre variants:
2019-08-03 19:21:54 +00:00
============= ======================= ==========================
2019-06-20 23:10:50 +00:00
CVE-2017-5753 Bounds check bypass Spectre variant 1
CVE-2017-5715 Branch target injection Spectre variant 2
2019-08-03 19:21:54 +00:00
CVE-2019-1125 Spectre v1 swapgs Spectre variant 1 (swapgs)
============= ======================= ==========================
2019-06-20 23:10:50 +00:00
Problem
-------
CPUs use speculative operations to improve performance. That may leave
traces of memory accesses or computations in the processor's caches,
buffers, and branch predictors. Malicious software may be able to
influence the speculative execution paths, and then use the side effects
of the speculative execution in the CPUs' caches and buffers to infer
privileged data touched during the speculative execution.
Spectre variant 1 attacks take advantage of speculative execution of
conditional branches, while Spectre variant 2 attacks use speculative
execution of indirect branches to leak privileged memory.
2022-02-28 17:23:16 +00:00
See :ref: `[1] <spec_ref1>` :ref: `[5] <spec_ref5>` :ref: `[6] <spec_ref6>`
:ref: `[7] <spec_ref7>` :ref: `[10] <spec_ref10>` :ref: `[11] <spec_ref11>` .
2019-06-20 23:10:50 +00:00
Spectre variant 1 (Bounds Check Bypass)
---------------------------------------
The bounds check bypass attack :ref: `[2] <spec_ref2>` takes advantage
of speculative execution that bypasses conditional branch instructions
used for memory access bounds check (e.g. checking if the index of an
array results in memory access within a valid range). This results in
memory accesses to invalid memory (with out-of-bound index) that are
done speculatively before validation checks resolve. Such speculative
memory accesses can leave side effects, creating side channels which
leak information to the attacker.
There are some extensions of Spectre variant 1 attacks for reading data
over the network, see :ref: `[12] <spec_ref12>` . However such attacks
are difficult, low bandwidth, fragile, and are considered low risk.
2019-08-03 19:21:54 +00:00
Note that, despite "Bounds Check Bypass" name, Spectre variant 1 is not
only about user-controlled array bounds checks. It can affect any
conditional checks. The kernel entry code interrupt, exception, and NMI
handlers all have conditional swapgs checks. Those may be problematic
in the context of Spectre v1, as kernel code can speculatively run with
a user GS.
2019-06-20 23:10:50 +00:00
Spectre variant 2 (Branch Target Injection)
-------------------------------------------
The branch target injection attack takes advantage of speculative
execution of indirect branches :ref: `[3] <spec_ref3>` . The indirect
branch predictors inside the processor used to guess the target of
indirect branches can be influenced by an attacker, causing gadget code
to be speculatively executed, thus exposing sensitive data touched by
the victim. The side effects left in the CPU's caches during speculative
execution can be measured to infer data values.
.. _poison_btb:
In Spectre variant 2 attacks, the attacker can steer speculative indirect
branches in the victim to gadget code by poisoning the branch target
buffer of a CPU used for predicting indirect branch addresses. Such
poisoning could be done by indirect branching into existing code,
with the address offset of the indirect branch under the attacker's
control. Since the branch prediction on impacted hardware does not
fully disambiguate branch address and uses the offset for prediction,
this could cause privileged code's indirect branch to jump to a gadget
code with the same offset.
The most useful gadgets take an attacker-controlled input parameter (such
as a register value) so that the memory read can be controlled. Gadgets
without input parameters might be possible, but the attacker would have
very little control over what memory can be read, reducing the risk of
the attack revealing useful data.
One other variant 2 attack vector is for the attacker to poison the
return stack buffer (RSB) :ref: `[13] <spec_ref13>` to cause speculative
subroutine return instruction execution to go to a gadget. An attacker's
imbalanced subroutine call instructions might "poison" entries in the
return stack buffer which are later consumed by a victim's subroutine
return instructions. This attack can be mitigated by flushing the return
stack buffer on context switch, or virtual machine (VM) exit.
On systems with simultaneous multi-threading (SMT), attacks are possible
from the sibling thread, as level 1 cache and branch target buffer
(BTB) may be shared between hardware threads in a CPU core. A malicious
program running on the sibling thread may influence its peer's BTB to
steer its indirect branch speculations to gadget code, and measure the
speculative execution's side effects left in level 1 cache to infer the
victim's data.
2022-02-16 19:57:02 +00:00
Yet another variant 2 attack vector is for the attacker to poison the
Branch History Buffer (BHB) to speculatively steer an indirect branch
to a specific Branch Target Buffer (BTB) entry, even if the entry isn't
associated with the source address of the indirect branch. Specifically,
the BHB might be shared across privilege levels even in the presence of
Enhanced IBRS.
Currently the only known real-world BHB attack vector is via
unprivileged eBPF. Therefore, it's highly recommended to not enable
unprivileged eBPF, especially when eIBRS is used (without retpolines).
For a full mitigation against BHB attacks, it's recommended to use
retpolines (or eIBRS combined with retpolines).
2019-06-20 23:10:50 +00:00
Attack scenarios
----------------
The following list of attack scenarios have been anticipated, but may
not cover all possible attack vectors.
1. A user process attacking the kernel
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2019-08-03 19:21:54 +00:00
Spectre variant 1
~~~~~~~~~~~~~~~~~
2019-06-20 23:10:50 +00:00
The attacker passes a parameter to the kernel via a register or
via a known address in memory during a syscall. Such parameter may
be used later by the kernel as an index to an array or to derive
a pointer for a Spectre variant 1 attack. The index or pointer
is invalid, but bound checks are bypassed in the code branch taken
for speculative execution. This could cause privileged memory to be
accessed and leaked.
For kernel code that has been identified where data pointers could
potentially be influenced for Spectre attacks, new "nospec" accessor
macros are used to prevent speculative loading of data.
2019-08-03 19:21:54 +00:00
Spectre variant 1 (swapgs)
~~~~~~~~~~~~~~~~~~~~~~~~~~
An attacker can train the branch predictor to speculatively skip the
swapgs path for an interrupt or exception. If they initialize
the GS register to a user-space value, if the swapgs is speculatively
skipped, subsequent GS-related percpu accesses in the speculation
window will be done with the attacker-controlled GS value. This
could cause privileged memory to be accessed and leaked.
For example:
::
if (coming from user space)
swapgs
mov %gs:<percpu_offset> , %reg
mov (%reg), %reg1
When coming from user space, the CPU can speculatively skip the
swapgs, and then do a speculative percpu load using the user GS
value. So the user can speculatively force a read of any kernel
value. If a gadget exists which uses the percpu value as an address
in another load/store, then the contents of the kernel value may
become visible via an L1 side channel attack.
A similar attack exists when coming from kernel space. The CPU can
speculatively do the swapgs, causing the user GS to get used for the
rest of the speculative window.
Spectre variant 2
~~~~~~~~~~~~~~~~~
A spectre variant 2 attacker can :ref: `poison <poison_btb>` the branch
2019-06-20 23:10:50 +00:00
target buffer (BTB) before issuing syscall to launch an attack.
After entering the kernel, the kernel could use the poisoned branch
target buffer on indirect jump and jump to gadget code in speculative
execution.
If an attacker tries to control the memory addresses leaked during
speculative execution, he would also need to pass a parameter to the
gadget, either through a register or a known address in memory. After
the gadget has executed, he can measure the side effect.
The kernel can protect itself against consuming poisoned branch
target buffer entries by using return trampolines (also known as
"retpoline") :ref: `[3] <spec_ref3>` :ref: `[9] <spec_ref9>` for all
indirect branches. Return trampolines trap speculative execution paths
to prevent jumping to gadget code during speculative execution.
x86 CPUs with Enhanced Indirect Branch Restricted Speculation
(Enhanced IBRS) available in hardware should use the feature to
mitigate Spectre variant 2 instead of retpoline. Enhanced IBRS is
more efficient than retpoline.
There may be gadget code in firmware which could be exploited with
Spectre variant 2 attack by a rogue user process. To mitigate such
attacks on x86, Indirect Branch Restricted Speculation (IBRS) feature
is turned on before the kernel invokes any firmware code.
2. A user process attacking another user process
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
A malicious user process can try to attack another user process,
either via a context switch on the same hardware thread, or from the
sibling hyperthread sharing a physical processor core on simultaneous
multi-threading (SMT) system.
Spectre variant 1 attacks generally require passing parameters
between the processes, which needs a data passing relationship, such
as remote procedure calls (RPC). Those parameters are used in gadget
code to derive invalid data pointers accessing privileged memory in
the attacked process.
Spectre variant 2 attacks can be launched from a rogue process by
:ref: `poisoning <poison_btb>` the branch target buffer. This can
influence the indirect branch targets for a victim process that either
runs later on the same hardware thread, or running concurrently on
a sibling hardware thread sharing the same physical core.
A user process can protect itself against Spectre variant 2 attacks
by using the prctl() syscall to disable indirect branch speculation
for itself. An administrator can also cordon off an unsafe process
from polluting the branch target buffer by disabling the process's
indirect branch speculation. This comes with a performance cost
from not using indirect branch speculation and clearing the branch
target buffer. When SMT is enabled on x86, for a process that has
indirect branch speculation disabled, Single Threaded Indirect Branch
Predictors (STIBP) :ref: `[4] <spec_ref4>` are turned on to prevent the
sibling thread from controlling branch target buffer. In addition,
the Indirect Branch Prediction Barrier (IBPB) is issued to clear the
branch target buffer when context switching to and from such process.
On x86, the return stack buffer is stuffed on context switch.
This prevents the branch target buffer from being used for branch
prediction when the return stack buffer underflows while switching to
a deeper call stack. Any poisoned entries in the return stack buffer
left by the previous process will also be cleared.
User programs should use address space randomization to make attacks
more difficult (Set /proc/sys/kernel/randomize_va_space = 1 or 2).
3. A virtualized guest attacking the host
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
The attack mechanism is similar to how user processes attack the
kernel. The kernel is entered via hyper-calls or other virtualization
exit paths.
For Spectre variant 1 attacks, rogue guests can pass parameters
(e.g. in registers) via hyper-calls to derive invalid pointers to
speculate into privileged memory after entering the kernel. For places
where such kernel code has been identified, nospec accessor macros
are used to stop speculative memory access.
For Spectre variant 2 attacks, rogue guests can :ref:`poison
<poison_btb>` the branch target buffer or return stack buffer, causing
the kernel to jump to gadget code in the speculative execution paths.
To mitigate variant 2, the host kernel can use return trampolines
for indirect branches to bypass the poisoned branch target buffer,
and flushing the return stack buffer on VM exit. This prevents rogue
guests from affecting indirect branching in the host kernel.
To protect host processes from rogue guests, host processes can have
indirect branch speculation disabled via prctl(). The branch target
buffer is cleared before context switching to such processes.
4. A virtualized guest attacking other guest
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
A rogue guest may attack another guest to get data accessible by the
other guest.
Spectre variant 1 attacks are possible if parameters can be passed
between guests. This may be done via mechanisms such as shared memory
or message passing. Such parameters could be used to derive data
pointers to privileged data in guest. The privileged data could be
accessed by gadget code in the victim's speculation paths.
Spectre variant 2 attacks can be launched from a rogue guest by
:ref: `poisoning <poison_btb>` the branch target buffer or the return
stack buffer. Such poisoned entries could be used to influence
speculation execution paths in the victim guest.
Linux kernel mitigates attacks to other guests running in the same
CPU hardware thread by flushing the return stack buffer on VM exit,
and clearing the branch target buffer before switching to a new guest.
If SMT is used, Spectre variant 2 attacks from an untrusted guest
in the sibling hyperthread can be mitigated by the administrator,
by turning off the unsafe guest's indirect branch speculation via
prctl(). A guest can also protect itself by turning on microcode
based mitigations (such as IBPB or STIBP on x86) within the guest.
.. _spectre_sys_info:
Spectre system information
--------------------------
The Linux kernel provides a sysfs interface to enumerate the current
mitigation status of the system for Spectre: whether the system is
vulnerable, and which mitigations are active.
The sysfs file showing Spectre variant 1 mitigation status is:
/sys/devices/system/cpu/vulnerabilities/spectre_v1
The possible values in this file are:
2019-08-03 19:21:54 +00:00
.. list-table ::
* - 'Not affected'
- The processor is not vulnerable.
* - 'Vulnerable: __user pointer sanitization and usercopy barriers only; no swapgs barriers'
- The swapgs protections are disabled; otherwise it has
protection in the kernel on a case by case base with explicit
pointer sanitation and usercopy LFENCE barriers.
* - 'Mitigation: usercopy/swapgs barriers and __user pointer sanitization'
- Protection in the kernel on a case by case base with explicit
pointer sanitation, usercopy LFENCE barriers, and swapgs LFENCE
barriers.
2019-06-20 23:10:50 +00:00
However, the protections are put in place on a case by case basis,
and there is no guarantee that all possible attack vectors for Spectre
variant 1 are covered.
The spectre_v2 kernel file reports if the kernel has been compiled with
retpoline mitigation or if the CPU has hardware mitigation, and if the
CPU has support for additional process-specific mitigation.
This file also reports CPU features enabled by microcode to mitigate
attack between user processes:
1. Indirect Branch Prediction Barrier (IBPB) to add additional
isolation between processes of different users.
2. Single Thread Indirect Branch Predictors (STIBP) to add additional
isolation between CPU threads running on the same core.
These CPU features may impact performance when used and can be enabled
per process on a case-by-case base.
The sysfs file showing Spectre variant 2 mitigation status is:
/sys/devices/system/cpu/vulnerabilities/spectre_v2
The possible values in this file are:
- Kernel status:
2022-02-16 19:57:02 +00:00
======================================== =================================
'Not affected' The processor is not vulnerable
'Mitigation: None' Vulnerable, no mitigation
'Mitigation: Retpolines' Use Retpoline thunks
'Mitigation: LFENCE' Use LFENCE instructions
'Mitigation: Enhanced IBRS' Hardware-focused mitigation
'Mitigation: Enhanced IBRS + Retpolines' Hardware-focused + Retpolines
'Mitigation: Enhanced IBRS + LFENCE' Hardware-focused + LFENCE
======================================== =================================
2019-06-20 23:10:50 +00:00
- Firmware status: Show if Indirect Branch Restricted Speculation (IBRS) is
used to protect against Spectre variant 2 attacks when calling firmware (x86 only).
========== =============================================================
'IBRS_FW' Protection against user program attacks when calling firmware
========== =============================================================
- Indirect branch prediction barrier (IBPB) status for protection between
processes of different users. This feature can be controlled through
prctl() per process, or through kernel command line options. This is
an x86 only feature. For more details see below.
=================== ========================================================
'IBPB: disabled' IBPB unused
'IBPB: always-on' Use IBPB on all tasks
'IBPB: conditional' Use IBPB on SECCOMP or indirect branch restricted tasks
=================== ========================================================
- Single threaded indirect branch prediction (STIBP) status for protection
between different hyper threads. This feature can be controlled through
prctl per process, or through kernel command line options. This is x86
only feature. For more details see below.
==================== ========================================================
'STIBP: disabled' STIBP unused
'STIBP: forced' Use STIBP on all tasks
'STIBP: conditional' Use STIBP on SECCOMP or indirect branch restricted tasks
==================== ========================================================
- Return stack buffer (RSB) protection status:
============= ===========================================
'RSB filling' Protection of RSB on context switch enabled
============= ===========================================
x86/speculation: Add RSB VM Exit protections
tl;dr: The Enhanced IBRS mitigation for Spectre v2 does not work as
documented for RET instructions after VM exits. Mitigate it with a new
one-entry RSB stuffing mechanism and a new LFENCE.
== Background ==
Indirect Branch Restricted Speculation (IBRS) was designed to help
mitigate Branch Target Injection and Speculative Store Bypass, i.e.
Spectre, attacks. IBRS prevents software run in less privileged modes
from affecting branch prediction in more privileged modes. IBRS requires
the MSR to be written on every privilege level change.
To overcome some of the performance issues of IBRS, Enhanced IBRS was
introduced. eIBRS is an "always on" IBRS, in other words, just turn
it on once instead of writing the MSR on every privilege level change.
When eIBRS is enabled, more privileged modes should be protected from
less privileged modes, including protecting VMMs from guests.
== Problem ==
Here's a simplification of how guests are run on Linux' KVM:
void run_kvm_guest(void)
{
// Prepare to run guest
VMRESUME();
// Clean up after guest runs
}
The execution flow for that would look something like this to the
processor:
1. Host-side: call run_kvm_guest()
2. Host-side: VMRESUME
3. Guest runs, does "CALL guest_function"
4. VM exit, host runs again
5. Host might make some "cleanup" function calls
6. Host-side: RET from run_kvm_guest()
Now, when back on the host, there are a couple of possible scenarios of
post-guest activity the host needs to do before executing host code:
* on pre-eIBRS hardware (legacy IBRS, or nothing at all), the RSB is not
touched and Linux has to do a 32-entry stuffing.
* on eIBRS hardware, VM exit with IBRS enabled, or restoring the host
IBRS=1 shortly after VM exit, has a documented side effect of flushing
the RSB except in this PBRSB situation where the software needs to stuff
the last RSB entry "by hand".
IOW, with eIBRS supported, host RET instructions should no longer be
influenced by guest behavior after the host retires a single CALL
instruction.
However, if the RET instructions are "unbalanced" with CALLs after a VM
exit as is the RET in #6, it might speculatively use the address for the
instruction after the CALL in #3 as an RSB prediction. This is a problem
since the (untrusted) guest controls this address.
Balanced CALL/RET instruction pairs such as in step #5 are not affected.
== Solution ==
The PBRSB issue affects a wide variety of Intel processors which
support eIBRS. But not all of them need mitigation. Today,
X86_FEATURE_RSB_VMEXIT triggers an RSB filling sequence that mitigates
PBRSB. Systems setting RSB_VMEXIT need no further mitigation - i.e.,
eIBRS systems which enable legacy IBRS explicitly.
However, such systems (X86_FEATURE_IBRS_ENHANCED) do not set RSB_VMEXIT
and most of them need a new mitigation.
Therefore, introduce a new feature flag X86_FEATURE_RSB_VMEXIT_LITE
which triggers a lighter-weight PBRSB mitigation versus RSB_VMEXIT.
The lighter-weight mitigation performs a CALL instruction which is
immediately followed by a speculative execution barrier (INT3). This
steers speculative execution to the barrier -- just like a retpoline
-- which ensures that speculation can never reach an unbalanced RET.
Then, ensure this CALL is retired before continuing execution with an
LFENCE.
In other words, the window of exposure is opened at VM exit where RET
behavior is troublesome. While the window is open, force RSB predictions
sampling for RET targets to a dead end at the INT3. Close the window
with the LFENCE.
There is a subset of eIBRS systems which are not vulnerable to PBRSB.
Add these systems to the cpu_vuln_whitelist[] as NO_EIBRS_PBRSB.
Future systems that aren't vulnerable will set ARCH_CAP_PBRSB_NO.
[ bp: Massage, incorporate review comments from Andy Cooper. ]
Signed-off-by: Daniel Sneddon <daniel.sneddon@linux.intel.com>
Co-developed-by: Pawan Gupta <pawan.kumar.gupta@linux.intel.com>
Signed-off-by: Pawan Gupta <pawan.kumar.gupta@linux.intel.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
2022-08-02 22:47:01 +00:00
- EIBRS Post-barrier Return Stack Buffer (PBRSB) protection status:
=========================== =======================================================
'PBRSB-eIBRS: SW sequence' CPU is affected and protection of RSB on VMEXIT enabled
'PBRSB-eIBRS: Vulnerable' CPU is vulnerable
'PBRSB-eIBRS: Not affected' CPU is not affected by PBRSB
=========================== =======================================================
2019-06-20 23:10:50 +00:00
Full mitigation might require a microcode update from the CPU
vendor. When the necessary microcode is not available, the kernel will
report vulnerability.
Turning on mitigation for Spectre variant 1 and Spectre variant 2
-----------------------------------------------------------------
1. Kernel mitigation
^^^^^^^^^^^^^^^^^^^^
2019-08-03 19:21:54 +00:00
Spectre variant 1
~~~~~~~~~~~~~~~~~
2019-06-20 23:10:50 +00:00
For the Spectre variant 1, vulnerable kernel code (as determined
by code audit or scanning tools) is annotated on a case by case
basis to use nospec accessor macros for bounds clipping :ref:`[2]
<spec_ref2>` to avoid any usable disclosure gadgets. However, it may
not cover all attack vectors for Spectre variant 1.
2019-08-03 19:21:54 +00:00
Copy-from-user code has an LFENCE barrier to prevent the access_ok()
check from being mis-speculated. The barrier is done by the
barrier_nospec() macro.
For the swapgs variant of Spectre variant 1, LFENCE barriers are
added to interrupt, exception and NMI entry where needed. These
barriers are done by the FENCE_SWAPGS_KERNEL_ENTRY and
FENCE_SWAPGS_USER_ENTRY macros.
Spectre variant 2
~~~~~~~~~~~~~~~~~
2019-06-20 23:10:50 +00:00
For Spectre variant 2 mitigation, the compiler turns indirect calls or
jumps in the kernel into equivalent return trampolines (retpolines)
:ref: `[3] <spec_ref3>` :ref: `[9] <spec_ref9>` to go to the target
addresses. Speculative execution paths under retpolines are trapped
in an infinite loop to prevent any speculative execution jumping to
a gadget.
To turn on retpoline mitigation on a vulnerable CPU, the kernel
needs to be compiled with a gcc compiler that supports the
-mindirect-branch=thunk-extern -mindirect-branch-register options.
If the kernel is compiled with a Clang compiler, the compiler needs
to support -mretpoline-external-thunk option. The kernel config
CONFIG_RETPOLINE needs to be turned on, and the CPU needs to run with
the latest updated microcode.
On Intel Skylake-era systems the mitigation covers most, but not all,
cases. See :ref: `[3] <spec_ref3>` for more details.
2023-02-27 06:05:41 +00:00
On CPUs with hardware mitigation for Spectre variant 2 (e.g. IBRS
or enhanced IBRS on x86), retpoline is automatically disabled at run time.
Systems which support enhanced IBRS (eIBRS) enable IBRS protection once at
boot, by setting the IBRS bit, and they're automatically protected against
Spectre v2 variant attacks, including cross-thread branch target injections
on SMT systems (STIBP). In other words, eIBRS enables STIBP too.
Legacy IBRS systems clear the IBRS bit on exit to userspace and
therefore explicitly enable STIBP for that
2019-06-20 23:10:50 +00:00
The retpoline mitigation is turned on by default on vulnerable
CPUs. It can be forced on or off by the administrator
via the kernel command line and sysfs control files. See
:ref: `spectre_mitigation_control_command_line` .
On x86, indirect branch restricted speculation is turned on by default
before invoking any firmware code to prevent Spectre variant 2 exploits
using the firmware.
2021-12-30 17:19:40 +00:00
Using kernel address space randomization (CONFIG_RANDOMIZE_BASE=y
2019-06-20 23:10:50 +00:00
and CONFIG_SLAB_FREELIST_RANDOM=y in the kernel configuration) makes
attacks on the kernel generally more difficult.
2. User program mitigation
^^^^^^^^^^^^^^^^^^^^^^^^^^
User programs can mitigate Spectre variant 1 using LFENCE or "bounds
clipping". For more details see :ref: `[2] <spec_ref2>` .
For Spectre variant 2 mitigation, individual user programs
can be compiled with return trampolines for indirect branches.
This protects them from consuming poisoned entries in the branch
2023-02-27 06:05:41 +00:00
target buffer left by malicious software.
On legacy IBRS systems, at return to userspace, implicit STIBP is disabled
because the kernel clears the IBRS bit. In this case, the userspace programs
can disable indirect branch speculation via prctl() (See
:ref: `Documentation/userspace-api/spec_ctrl.rst <set_spec_ctrl>` ).
2019-06-20 23:10:50 +00:00
On x86, this will turn on STIBP to guard against attacks from the
sibling thread when the user program is running, and use IBPB to
flush the branch target buffer when switching to/from the program.
Restricting indirect branch speculation on a user program will
also prevent the program from launching a variant 2 attack
x86: change default to spec_store_bypass_disable=prctl spectre_v2_user=prctl
Switch the kernel default of SSBD and STIBP to the ones with
CONFIG_SECCOMP=n (i.e. spec_store_bypass_disable=prctl
spectre_v2_user=prctl) even if CONFIG_SECCOMP=y.
Several motivations listed below:
- If SMT is enabled the seccomp jail can still attack the rest of the
system even with spectre_v2_user=seccomp by using MDS-HT (except on
XEON PHI where MDS can be tamed with SMT left enabled, but that's a
special case). Setting STIBP become a very expensive window dressing
after MDS-HT was discovered.
- The seccomp jail cannot attack the kernel with spectre-v2-HT
regardless (even if STIBP is not set), but with MDS-HT the seccomp
jail can attack the kernel too.
- With spec_store_bypass_disable=prctl the seccomp jail can attack the
other userland (guest or host mode) using spectre-v2-HT, but the
userland attack is already mitigated by both ASLR and pid namespaces
for host userland and through virt isolation with libkrun or
kata. (if something if somebody is worried about spectre-v2-HT it's
best to mount proc with hidepid=2,gid=proc on workstations where not
all apps may run under container runtimes, rather than slowing down
all seccomp jails, but the best is to add pid namespaces to the
seccomp jail). As opposed MDS-HT is not mitigated and the seccomp
jail can still attack all other host and guest userland if SMT is
enabled even with spec_store_bypass_disable=seccomp.
- If full security is required then MDS-HT must also be mitigated with
nosmt and then spectre_v2_user=prctl and spectre_v2_user=seccomp
would become identical.
- Setting spectre_v2_user=seccomp is overall lower priority than to
setting javascript.options.wasm false in about:config to protect
against remote wasm MDS-HT, instead of worrying about Spectre-v2-HT
and STIBP which again is already statistically well mitigated by
other means in userland and it's fully mitigated in kernel with
retpolines (unlike the wasm assist call with MDS-HT).
- SSBD is needed to prevent reading the JIT memory and the primary
user being the OpenJDK. However the primary user of SSBD wouldn't be
covered by spec_store_bypass_disable=seccomp because it doesn't use
seccomp and the primary user also explicitly declined to set
PR_SET_SPECULATION_CTRL+PR_SPEC_STORE_BYPASS despite it easily
could. In fact it would need to set it only when the sandboxing
mechanism is enabled for javaws applets, but it still declined it by
declaring security within the same user address space as an
untenable objective for their JIT, even in the sandboxing case where
performance would be a lesser concern (for the record: I kind of
disagree in not setting PR_SPEC_STORE_BYPASS in the sandbox case and
I prefer to run javaws through a wrapper that sets
PR_SPEC_STORE_BYPASS if I need). In turn it can be inferred that
even if the primary user of SSBD would use seccomp, they would
invoke it with SECCOMP_FILTER_FLAG_SPEC_ALLOW by now.
- runc/crun already set SECCOMP_FILTER_FLAG_SPEC_ALLOW by default, k8s
and podman have a default json seccomp allowlist that cannot be
slowed down, so for the #1 seccomp user this change is already a
noop.
- systemd/sshd or other apps that use seccomp, if they really need
STIBP or SSBD, they need to explicitly set the
PR_SET_SPECULATION_CTRL by now. The stibp/ssbd seccomp blind
catch-all approach was done probably initially with a wishful
thinking objective to pretend to have a peace of mind that it could
magically fix it all. That was wishful thinking before MDS-HT was
discovered, but after MDS-HT has been discovered it become just
window dressing.
- For qemu "-sandbox" seccomp jail it wouldn't make sense to set STIBP
or SSBD. SSBD doesn't help with KVM because there's no JIT (if it's
needed with TCG it should be an opt-in with
PR_SET_SPECULATION_CTRL+PR_SPEC_STORE_BYPASS and it shouldn't
slowdown KVM for nothing). For qemu+KVM STIBP would be even more
window dressing than it is for all other apps, because in the
qemu+KVM case there's not only the MDS attack to worry about with
SMT enabled. Even after disabling SMT, there's still a theoretical
spectre-v2 attack possible within the same thread context from guest
mode to host ring3 that the host kernel retpoline mitigation has no
theoretical chance to mitigate. On some kernels a
ibrs-always/ibrs-retpoline opt-in model is provided that will
enabled IBRS in the qemu host ring3 userland which fixes this
theoretical concern. Only after enabling IBRS in the host userland
it would then make sense to proceed and worry about STIBP and an
attack on the other host userland, but then again SMT would need to
be disabled for full security anyway, so that would render STIBP
again a noop.
- last but not the least: the lack of "spec_store_bypass_disable=prctl
spectre_v2_user=prctl" means the moment a guest boots and
sshd/systemd runs, the guest kernel will write to SPEC_CTRL MSR
which will make the guest vmexit forever slower, forcing KVM to
issue a very slow rdmsr instruction at every vmexit. So the end
result is that SPEC_CTRL MSR is only available in GCE. Most other
public cloud providers don't expose SPEC_CTRL, which means that not
only STIBP/SSBD isn't available, but IBPB isn't available either
(which would cause no overhead to the guest or the hypervisor
because it's write only and requires no reading during vmexit). So
the current default already net loss in security (missing IBPB)
which means most public cloud providers cannot achieve a fully
secure guest with nosmt (and nosmt is enough to fully mitigate
MDS-HT). It also means GCE and is unfairly penalized in performance
because it provides the option to enable full security in the guest
as an opt-in (i.e. nosmt and IBPB). So this change will allow all
cloud providers to expose SPEC_CTRL without incurring into any
hypervisor slowdown and at the same time it will remove the unfair
penalization of GCE performance for doing the right thing and it'll
allow to get full security with nosmt with IBPB being available (and
STIBP becoming meaningless).
Example to put things in prospective: the STIBP enabled in seccomp has
never been about protecting apps using seccomp like sshd from an
attack from a malicious userland, but to the contrary it has always
been about protecting the system from an attack from sshd, after a
successful remote network exploit against sshd. In fact initially it
wasn't obvious STIBP would work both ways (STIBP was about preventing
the task that runs with STIBP to be attacked with spectre-v2-HT, but
accidentally in the STIBP case it also prevents the attack in the
other direction). In the hypothetical case that sshd has been remotely
exploited the last concern should be STIBP being set, because it'll be
still possible to obtain info even from the kernel by using MDS if
nosmt wasn't set (and if it was set, STIBP is a noop in the first
place). As opposed kernel cannot leak anything with spectre-v2 HT
because of retpolines and the userland is mitigated by ASLR already
and ideally PID namespaces too. If something it'd be worth checking if
sshd run the seccomp thread under pid namespaces too if available in
the running kernel. SSBD also would be a noop for sshd, since sshd
uses no JIT. If sshd prefers to keep doing the STIBP window dressing
exercise, it still can even after this change of defaults by opting-in
with PR_SPEC_INDIRECT_BRANCH.
Ultimately setting SSBD and STIBP by default for all seccomp jails is
a bad sweet spot and bad default with more cons than pros that end up
reducing security in the public cloud (by giving an huge incentive to
not expose SPEC_CTRL which would be needed to get full security with
IBPB after setting nosmt in the guest) and by excessively hurting
performance to more secure apps using seccomp that end up having to
opt out with SECCOMP_FILTER_FLAG_SPEC_ALLOW.
The following is the verified result of the new default with SMT
enabled:
(gdb) print spectre_v2_user_stibp
$1 = SPECTRE_V2_USER_PRCTL
(gdb) print spectre_v2_user_ibpb
$2 = SPECTRE_V2_USER_PRCTL
(gdb) print ssb_mode
$3 = SPEC_STORE_BYPASS_PRCTL
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Kees Cook <keescook@chromium.org>
Link: https://lore.kernel.org/r/20201104235054.5678-1-aarcange@redhat.com
Acked-by: Josh Poimboeuf <jpoimboe@redhat.com>
Link: https://lore.kernel.org/lkml/AAA2EF2C-293D-4D5B-BFA6-FF655105CD84@redhat.com
Acked-by: Waiman Long <longman@redhat.com>
Link: https://lore.kernel.org/lkml/c0722838-06f7-da6b-138f-e0f26362f16a@redhat.com
2020-11-04 23:50:54 +00:00
on x86. Administrators can change that behavior via the kernel
command line and sysfs control files.
2019-06-20 23:10:50 +00:00
See :ref: `spectre_mitigation_control_command_line` .
Programs that disable their indirect branch speculation will have
more overhead and run slower.
User programs should use address space randomization
(/proc/sys/kernel/randomize_va_space = 1 or 2) to make attacks more
difficult.
3. VM mitigation
^^^^^^^^^^^^^^^^
Within the kernel, Spectre variant 1 attacks from rogue guests are
mitigated on a case by case basis in VM exit paths. Vulnerable code
uses nospec accessor macros for "bounds clipping", to avoid any
usable disclosure gadgets. However, this may not cover all variant
1 attack vectors.
For Spectre variant 2 attacks from rogue guests to the kernel, the
Linux kernel uses retpoline or Enhanced IBRS to prevent consumption of
poisoned entries in branch target buffer left by rogue guests. It also
flushes the return stack buffer on every VM exit to prevent a return
stack buffer underflow so poisoned branch target buffer could be used,
or attacker guests leaving poisoned entries in the return stack buffer.
To mitigate guest-to-guest attacks in the same CPU hardware thread,
the branch target buffer is sanitized by flushing before switching
to a new guest on a CPU.
The above mitigations are turned on by default on vulnerable CPUs.
To mitigate guest-to-guest attacks from sibling thread when SMT is
in use, an untrusted guest running in the sibling thread can have
its indirect branch speculation disabled by administrator via prctl().
The kernel also allows guests to use any microcode based mitigation
they choose to use (such as IBPB or STIBP on x86) to protect themselves.
.. _spectre_mitigation_control_command_line:
Mitigation control on the kernel command line
---------------------------------------------
Spectre variant 2 mitigation can be disabled or force enabled at the
kernel command line.
2019-08-03 19:21:54 +00:00
nospectre_v1
[X86,PPC] Disable mitigations for Spectre Variant 1
(bounds check bypass). With this option data leaks are
possible in the system.
2019-06-20 23:10:50 +00:00
nospectre_v2
[X86] Disable all mitigations for the Spectre variant 2
(indirect branch prediction) vulnerability. System may
allow data leaks with this option, which is equivalent
to spectre_v2=off.
spectre_v2=
[X86] Control mitigation of Spectre variant 2
(indirect branch speculation) vulnerability.
The default operation protects the kernel from
user space attacks.
on
unconditionally enable, implies
spectre_v2_user=on
off
unconditionally disable, implies
spectre_v2_user=off
auto
kernel detects whether your CPU model is
vulnerable
Selecting 'on' will, and 'auto' may, choose a
mitigation method at run time according to the
CPU, the available microcode, the setting of the
CONFIG_RETPOLINE configuration option, and the
compiler with which the kernel was built.
Selecting 'on' will also enable the mitigation
against user space to user space task attacks.
Selecting 'off' will disable both the kernel and
the user space protections.
Specific mitigations can also be selected manually:
2022-02-16 19:57:02 +00:00
retpoline auto pick between generic,lfence
retpoline,generic Retpolines
retpoline,lfence LFENCE; indirect branch
retpoline,amd alias for retpoline,lfence
2023-01-24 16:33:18 +00:00
eibrs Enhanced/Auto IBRS
eibrs,retpoline Enhanced/Auto IBRS + Retpolines
eibrs,lfence Enhanced/Auto IBRS + LFENCE
2022-08-30 12:36:14 +00:00
ibrs use IBRS to protect kernel
2019-06-20 23:10:50 +00:00
Not specifying this option is equivalent to
spectre_v2=auto.
In general the kernel by default selects
reasonable mitigations for the current CPU. To
disable Spectre variant 2 mitigations, boot with
spectre_v2=off. Spectre variant 1 mitigations
cannot be disabled.
2022-02-16 19:57:02 +00:00
For spectre_v2_user see Documentation/admin-guide/kernel-parameters.txt
2020-11-05 00:14:06 +00:00
2019-06-20 23:10:50 +00:00
Mitigation selection guide
--------------------------
1. Trusted userspace
^^^^^^^^^^^^^^^^^^^^
If all userspace applications are from trusted sources and do not
execute externally supplied untrusted code, then the mitigations can
be disabled.
2. Protect sensitive programs
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
For security-sensitive programs that have secrets (e.g. crypto
keys), protection against Spectre variant 2 can be put in place by
disabling indirect branch speculation when the program is running
(See :ref: `Documentation/userspace-api/spec_ctrl.rst <set_spec_ctrl>` ).
3. Sandbox untrusted programs
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Untrusted programs that could be a source of attacks can be cordoned
off by disabling their indirect branch speculation when they are run
(See :ref: `Documentation/userspace-api/spec_ctrl.rst <set_spec_ctrl>` ).
This prevents untrusted programs from polluting the branch target
x86: change default to spec_store_bypass_disable=prctl spectre_v2_user=prctl
Switch the kernel default of SSBD and STIBP to the ones with
CONFIG_SECCOMP=n (i.e. spec_store_bypass_disable=prctl
spectre_v2_user=prctl) even if CONFIG_SECCOMP=y.
Several motivations listed below:
- If SMT is enabled the seccomp jail can still attack the rest of the
system even with spectre_v2_user=seccomp by using MDS-HT (except on
XEON PHI where MDS can be tamed with SMT left enabled, but that's a
special case). Setting STIBP become a very expensive window dressing
after MDS-HT was discovered.
- The seccomp jail cannot attack the kernel with spectre-v2-HT
regardless (even if STIBP is not set), but with MDS-HT the seccomp
jail can attack the kernel too.
- With spec_store_bypass_disable=prctl the seccomp jail can attack the
other userland (guest or host mode) using spectre-v2-HT, but the
userland attack is already mitigated by both ASLR and pid namespaces
for host userland and through virt isolation with libkrun or
kata. (if something if somebody is worried about spectre-v2-HT it's
best to mount proc with hidepid=2,gid=proc on workstations where not
all apps may run under container runtimes, rather than slowing down
all seccomp jails, but the best is to add pid namespaces to the
seccomp jail). As opposed MDS-HT is not mitigated and the seccomp
jail can still attack all other host and guest userland if SMT is
enabled even with spec_store_bypass_disable=seccomp.
- If full security is required then MDS-HT must also be mitigated with
nosmt and then spectre_v2_user=prctl and spectre_v2_user=seccomp
would become identical.
- Setting spectre_v2_user=seccomp is overall lower priority than to
setting javascript.options.wasm false in about:config to protect
against remote wasm MDS-HT, instead of worrying about Spectre-v2-HT
and STIBP which again is already statistically well mitigated by
other means in userland and it's fully mitigated in kernel with
retpolines (unlike the wasm assist call with MDS-HT).
- SSBD is needed to prevent reading the JIT memory and the primary
user being the OpenJDK. However the primary user of SSBD wouldn't be
covered by spec_store_bypass_disable=seccomp because it doesn't use
seccomp and the primary user also explicitly declined to set
PR_SET_SPECULATION_CTRL+PR_SPEC_STORE_BYPASS despite it easily
could. In fact it would need to set it only when the sandboxing
mechanism is enabled for javaws applets, but it still declined it by
declaring security within the same user address space as an
untenable objective for their JIT, even in the sandboxing case where
performance would be a lesser concern (for the record: I kind of
disagree in not setting PR_SPEC_STORE_BYPASS in the sandbox case and
I prefer to run javaws through a wrapper that sets
PR_SPEC_STORE_BYPASS if I need). In turn it can be inferred that
even if the primary user of SSBD would use seccomp, they would
invoke it with SECCOMP_FILTER_FLAG_SPEC_ALLOW by now.
- runc/crun already set SECCOMP_FILTER_FLAG_SPEC_ALLOW by default, k8s
and podman have a default json seccomp allowlist that cannot be
slowed down, so for the #1 seccomp user this change is already a
noop.
- systemd/sshd or other apps that use seccomp, if they really need
STIBP or SSBD, they need to explicitly set the
PR_SET_SPECULATION_CTRL by now. The stibp/ssbd seccomp blind
catch-all approach was done probably initially with a wishful
thinking objective to pretend to have a peace of mind that it could
magically fix it all. That was wishful thinking before MDS-HT was
discovered, but after MDS-HT has been discovered it become just
window dressing.
- For qemu "-sandbox" seccomp jail it wouldn't make sense to set STIBP
or SSBD. SSBD doesn't help with KVM because there's no JIT (if it's
needed with TCG it should be an opt-in with
PR_SET_SPECULATION_CTRL+PR_SPEC_STORE_BYPASS and it shouldn't
slowdown KVM for nothing). For qemu+KVM STIBP would be even more
window dressing than it is for all other apps, because in the
qemu+KVM case there's not only the MDS attack to worry about with
SMT enabled. Even after disabling SMT, there's still a theoretical
spectre-v2 attack possible within the same thread context from guest
mode to host ring3 that the host kernel retpoline mitigation has no
theoretical chance to mitigate. On some kernels a
ibrs-always/ibrs-retpoline opt-in model is provided that will
enabled IBRS in the qemu host ring3 userland which fixes this
theoretical concern. Only after enabling IBRS in the host userland
it would then make sense to proceed and worry about STIBP and an
attack on the other host userland, but then again SMT would need to
be disabled for full security anyway, so that would render STIBP
again a noop.
- last but not the least: the lack of "spec_store_bypass_disable=prctl
spectre_v2_user=prctl" means the moment a guest boots and
sshd/systemd runs, the guest kernel will write to SPEC_CTRL MSR
which will make the guest vmexit forever slower, forcing KVM to
issue a very slow rdmsr instruction at every vmexit. So the end
result is that SPEC_CTRL MSR is only available in GCE. Most other
public cloud providers don't expose SPEC_CTRL, which means that not
only STIBP/SSBD isn't available, but IBPB isn't available either
(which would cause no overhead to the guest or the hypervisor
because it's write only and requires no reading during vmexit). So
the current default already net loss in security (missing IBPB)
which means most public cloud providers cannot achieve a fully
secure guest with nosmt (and nosmt is enough to fully mitigate
MDS-HT). It also means GCE and is unfairly penalized in performance
because it provides the option to enable full security in the guest
as an opt-in (i.e. nosmt and IBPB). So this change will allow all
cloud providers to expose SPEC_CTRL without incurring into any
hypervisor slowdown and at the same time it will remove the unfair
penalization of GCE performance for doing the right thing and it'll
allow to get full security with nosmt with IBPB being available (and
STIBP becoming meaningless).
Example to put things in prospective: the STIBP enabled in seccomp has
never been about protecting apps using seccomp like sshd from an
attack from a malicious userland, but to the contrary it has always
been about protecting the system from an attack from sshd, after a
successful remote network exploit against sshd. In fact initially it
wasn't obvious STIBP would work both ways (STIBP was about preventing
the task that runs with STIBP to be attacked with spectre-v2-HT, but
accidentally in the STIBP case it also prevents the attack in the
other direction). In the hypothetical case that sshd has been remotely
exploited the last concern should be STIBP being set, because it'll be
still possible to obtain info even from the kernel by using MDS if
nosmt wasn't set (and if it was set, STIBP is a noop in the first
place). As opposed kernel cannot leak anything with spectre-v2 HT
because of retpolines and the userland is mitigated by ASLR already
and ideally PID namespaces too. If something it'd be worth checking if
sshd run the seccomp thread under pid namespaces too if available in
the running kernel. SSBD also would be a noop for sshd, since sshd
uses no JIT. If sshd prefers to keep doing the STIBP window dressing
exercise, it still can even after this change of defaults by opting-in
with PR_SPEC_INDIRECT_BRANCH.
Ultimately setting SSBD and STIBP by default for all seccomp jails is
a bad sweet spot and bad default with more cons than pros that end up
reducing security in the public cloud (by giving an huge incentive to
not expose SPEC_CTRL which would be needed to get full security with
IBPB after setting nosmt in the guest) and by excessively hurting
performance to more secure apps using seccomp that end up having to
opt out with SECCOMP_FILTER_FLAG_SPEC_ALLOW.
The following is the verified result of the new default with SMT
enabled:
(gdb) print spectre_v2_user_stibp
$1 = SPECTRE_V2_USER_PRCTL
(gdb) print spectre_v2_user_ibpb
$2 = SPECTRE_V2_USER_PRCTL
(gdb) print ssb_mode
$3 = SPEC_STORE_BYPASS_PRCTL
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Kees Cook <keescook@chromium.org>
Link: https://lore.kernel.org/r/20201104235054.5678-1-aarcange@redhat.com
Acked-by: Josh Poimboeuf <jpoimboe@redhat.com>
Link: https://lore.kernel.org/lkml/AAA2EF2C-293D-4D5B-BFA6-FF655105CD84@redhat.com
Acked-by: Waiman Long <longman@redhat.com>
Link: https://lore.kernel.org/lkml/c0722838-06f7-da6b-138f-e0f26362f16a@redhat.com
2020-11-04 23:50:54 +00:00
buffer. This behavior can be changed via the kernel command line
and sysfs control files. See
2019-06-20 23:10:50 +00:00
:ref: `spectre_mitigation_control_command_line` .
3. High security mode
^^^^^^^^^^^^^^^^^^^^^
All Spectre variant 2 mitigations can be forced on
at boot time for all programs (See the "on" option in
:ref: `spectre_mitigation_control_command_line` ). This will add
overhead as indirect branch speculations for all programs will be
restricted.
On x86, branch target buffer will be flushed with IBPB when switching
to a new program. STIBP is left on all the time to protect programs
against variant 2 attacks originating from programs running on
sibling threads.
Alternatively, STIBP can be used only when running programs
whose indirect branch speculation is explicitly disabled,
while IBPB is still used all the time when switching to a new
program to clear the branch target buffer (See "ibpb" option in
:ref: `spectre_mitigation_control_command_line` ). This "ibpb" option
has less performance cost than the "on" option, which leaves STIBP
on all the time.
References on Spectre
---------------------
Intel white papers:
.. _spec_ref1:
[1] `Intel analysis of speculative execution side channels <https://newsroom.intel.com/wp-content/uploads/sites/11/2018/01/Intel-Analysis-of-Speculative-Execution-Side-Channels.pdf> `_ .
.. _spec_ref2:
[2] `Bounds check bypass <https://software.intel.com/security-software-guidance/software-guidance/bounds-check-bypass> `_ .
.. _spec_ref3:
[3] `Deep dive: Retpoline: A branch target injection mitigation <https://software.intel.com/security-software-guidance/insights/deep-dive-retpoline-branch-target-injection-mitigation> `_ .
.. _spec_ref4:
[4] `Deep Dive: Single Thread Indirect Branch Predictors <https://software.intel.com/security-software-guidance/insights/deep-dive-single-thread-indirect-branch-predictors> `_ .
AMD white papers:
.. _spec_ref5:
[5] `AMD64 technology indirect branch control extension <https://developer.amd.com/wp-content/resources/Architecture_Guidelines_Update_Indirect_Branch_Control.pdf> `_ .
.. _spec_ref6:
2022-02-28 17:23:16 +00:00
[6] `Software techniques for managing speculation on AMD processors <https://developer.amd.com/wp-content/resources/Managing-Speculation-on-AMD-Processors.pdf> `_ .
2019-06-20 23:10:50 +00:00
ARM white papers:
.. _spec_ref7:
[7] `Cache speculation side-channels <https://developer.arm.com/support/arm-security-updates/speculative-processor-vulnerability/download-the-whitepaper> `_ .
.. _spec_ref8:
[8] `Cache speculation issues update <https://developer.arm.com/support/arm-security-updates/speculative-processor-vulnerability/latest-updates/cache-speculation-issues-update> `_ .
Google white paper:
.. _spec_ref9:
[9] `Retpoline: a software construct for preventing branch-target-injection <https://support.google.com/faqs/answer/7625886> `_ .
MIPS white paper:
.. _spec_ref10:
[10] `MIPS: response on speculative execution and side channel vulnerabilities <https://www.mips.com/blog/mips-response-on-speculative-execution-and-side-channel-vulnerabilities/> `_ .
Academic papers:
.. _spec_ref11:
[11] `Spectre Attacks: Exploiting Speculative Execution <https://spectreattack.com/spectre.pdf> `_ .
.. _spec_ref12:
[12] `NetSpectre: Read Arbitrary Memory over Network <https://arxiv.org/abs/1807.10535> `_ .
.. _spec_ref13:
[13] `Spectre Returns! Speculation Attacks using the Return Stack Buffer <https://www.usenix.org/system/files/conference/woot18/woot18-paper-koruyeh.pdf> `_ .