~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/Documentation/admin-guide/hw-vuln/core-scheduling.rst

Version: ~ [ linux-6.12-rc7 ] ~ [ linux-6.11.7 ] ~ [ linux-6.10.14 ] ~ [ linux-6.9.12 ] ~ [ linux-6.8.12 ] ~ [ linux-6.7.12 ] ~ [ linux-6.6.60 ] ~ [ linux-6.5.13 ] ~ [ linux-6.4.16 ] ~ [ linux-6.3.13 ] ~ [ linux-6.2.16 ] ~ [ linux-6.1.116 ] ~ [ linux-6.0.19 ] ~ [ linux-5.19.17 ] ~ [ linux-5.18.19 ] ~ [ linux-5.17.15 ] ~ [ linux-5.16.20 ] ~ [ linux-5.15.171 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.229 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.285 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.323 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.336 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.337 ] ~ [ linux-4.4.302 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.12 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 .. SPDX-License-Identifier: GPL-2.0
  2 
  3 ===============
  4 Core Scheduling
  5 ===============
  6 Core scheduling support allows userspace to define groups of tasks that can
  7 share a core. These groups can be specified either for security usecases (one
  8 group of tasks don't trust another), or for performance usecases (some
  9 workloads may benefit from running on the same core as they don't need the same
 10 hardware resources of the shared core, or may prefer different cores if they
 11 do share hardware resource needs). This document only describes the security
 12 usecase.
 13 
 14 Security usecase
 15 ----------------
 16 A cross-HT attack involves the attacker and victim running on different Hyper
 17 Threads of the same core. MDS and L1TF are examples of such attacks.  The only
 18 full mitigation of cross-HT attacks is to disable Hyper Threading (HT). Core
 19 scheduling is a scheduler feature that can mitigate some (not all) cross-HT
 20 attacks. It allows HT to be turned on safely by ensuring that only tasks in a
 21 user-designated trusted group can share a core. This increase in core sharing
 22 can also improve performance, however it is not guaranteed that performance
 23 will always improve, though that is seen to be the case with a number of real
 24 world workloads. In theory, core scheduling aims to perform at least as good as
 25 when Hyper Threading is disabled. In practice, this is mostly the case though
 26 not always: as synchronizing scheduling decisions across 2 or more CPUs in a
 27 core involves additional overhead - especially when the system is lightly
 28 loaded. When ``total_threads <= N_CPUS/2``, the extra overhead may cause core
 29 scheduling to perform more poorly compared to SMT-disabled, where N_CPUS is the
 30 total number of CPUs. Please measure the performance of your workloads always.
 31 
 32 Usage
 33 -----
 34 Core scheduling support is enabled via the ``CONFIG_SCHED_CORE`` config option.
 35 Using this feature, userspace defines groups of tasks that can be co-scheduled
 36 on the same core. The core scheduler uses this information to make sure that
 37 tasks that are not in the same group never run simultaneously on a core, while
 38 doing its best to satisfy the system's scheduling requirements.
 39 
 40 Core scheduling can be enabled via the ``PR_SCHED_CORE`` prctl interface.
 41 This interface provides support for the creation of core scheduling groups, as
 42 well as admission and removal of tasks from created groups::
 43 
 44     #include <sys/prctl.h>
 45 
 46     int prctl(int option, unsigned long arg2, unsigned long arg3,
 47             unsigned long arg4, unsigned long arg5);
 48 
 49 option:
 50     ``PR_SCHED_CORE``
 51 
 52 arg2:
 53     Command for operation, must be one off:
 54 
 55     - ``PR_SCHED_CORE_GET`` -- get core_sched cookie of ``pid``.
 56     - ``PR_SCHED_CORE_CREATE`` -- create a new unique cookie for ``pid``.
 57     - ``PR_SCHED_CORE_SHARE_TO`` -- push core_sched cookie to ``pid``.
 58     - ``PR_SCHED_CORE_SHARE_FROM`` -- pull core_sched cookie from ``pid``.
 59 
 60 arg3:
 61     ``pid`` of the task for which the operation applies.
 62 
 63 arg4:
 64     ``pid_type`` for which the operation applies. It is one of
 65     ``PR_SCHED_CORE_SCOPE_``-prefixed macro constants.  For example, if arg4
 66     is ``PR_SCHED_CORE_SCOPE_THREAD_GROUP``, then the operation of this command
 67     will be performed for all tasks in the task group of ``pid``.
 68 
 69 arg5:
 70     userspace pointer to an unsigned long long for storing the cookie returned
 71     by ``PR_SCHED_CORE_GET`` command. Should be 0 for all other commands.
 72 
 73 In order for a process to push a cookie to, or pull a cookie from a process, it
 74 is required to have the ptrace access mode: `PTRACE_MODE_READ_REALCREDS` to the
 75 process.
 76 
 77 Building hierarchies of tasks
 78 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 79 The simplest way to build hierarchies of threads/processes which share a
 80 cookie and thus a core is to rely on the fact that the core-sched cookie is
 81 inherited across forks/clones and execs, thus setting a cookie for the
 82 'initial' script/executable/daemon will place every spawned child in the
 83 same core-sched group.
 84 
 85 Cookie Transferral
 86 ~~~~~~~~~~~~~~~~~~
 87 Transferring a cookie between the current and other tasks is possible using
 88 PR_SCHED_CORE_SHARE_FROM and PR_SCHED_CORE_SHARE_TO to inherit a cookie from a
 89 specified task or a share a cookie with a task. In combination this allows a
 90 simple helper program to pull a cookie from a task in an existing core
 91 scheduling group and share it with already running tasks.
 92 
 93 Design/Implementation
 94 ---------------------
 95 Each task that is tagged is assigned a cookie internally in the kernel. As
 96 mentioned in `Usage`_, tasks with the same cookie value are assumed to trust
 97 each other and share a core.
 98 
 99 The basic idea is that, every schedule event tries to select tasks for all the
100 siblings of a core such that all the selected tasks running on a core are
101 trusted (same cookie) at any point in time. Kernel threads are assumed trusted.
102 The idle task is considered special, as it trusts everything and everything
103 trusts it.
104 
105 During a schedule() event on any sibling of a core, the highest priority task on
106 the sibling's core is picked and assigned to the sibling calling schedule(), if
107 the sibling has the task enqueued. For rest of the siblings in the core,
108 highest priority task with the same cookie is selected if there is one runnable
109 in their individual run queues. If a task with same cookie is not available,
110 the idle task is selected.  Idle task is globally trusted.
111 
112 Once a task has been selected for all the siblings in the core, an IPI is sent to
113 siblings for whom a new task was selected. Siblings on receiving the IPI will
114 switch to the new task immediately. If an idle task is selected for a sibling,
115 then the sibling is considered to be in a `forced idle` state. I.e., it may
116 have tasks on its on runqueue to run, however it will still have to run idle.
117 More on this in the next section.
118 
119 Forced-idling of hyperthreads
120 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
121 The scheduler tries its best to find tasks that trust each other such that all
122 tasks selected to be scheduled are of the highest priority in a core.  However,
123 it is possible that some runqueues had tasks that were incompatible with the
124 highest priority ones in the core. Favoring security over fairness, one or more
125 siblings could be forced to select a lower priority task if the highest
126 priority task is not trusted with respect to the core wide highest priority
127 task.  If a sibling does not have a trusted task to run, it will be forced idle
128 by the scheduler (idle thread is scheduled to run).
129 
130 When the highest priority task is selected to run, a reschedule-IPI is sent to
131 the sibling to force it into idle. This results in 4 cases which need to be
132 considered depending on whether a VM or a regular usermode process was running
133 on either HT::
134 
135           HT1 (attack)            HT2 (victim)
136    A      idle -> user space      user space -> idle
137    B      idle -> user space      guest -> idle
138    C      idle -> guest           user space -> idle
139    D      idle -> guest           guest -> idle
140 
141 Note that for better performance, we do not wait for the destination CPU
142 (victim) to enter idle mode. This is because the sending of the IPI would bring
143 the destination CPU immediately into kernel mode from user space, or VMEXIT
144 in the case of guests. At best, this would only leak some scheduler metadata
145 which may not be worth protecting. It is also possible that the IPI is received
146 too late on some architectures, but this has not been observed in the case of
147 x86.
148 
149 Trust model
150 ~~~~~~~~~~~
151 Core scheduling maintains trust relationships amongst groups of tasks by
152 assigning them a tag that is the same cookie value.
153 When a system with core scheduling boots, all tasks are considered to trust
154 each other. This is because the core scheduler does not have information about
155 trust relationships until userspace uses the above mentioned interfaces, to
156 communicate them. In other words, all tasks have a default cookie value of 0.
157 and are considered system-wide trusted. The forced-idling of siblings running
158 cookie-0 tasks is also avoided.
159 
160 Once userspace uses the above mentioned interfaces to group sets of tasks, tasks
161 within such groups are considered to trust each other, but do not trust those
162 outside. Tasks outside the group also don't trust tasks within.
163 
164 Limitations of core-scheduling
165 ------------------------------
166 Core scheduling tries to guarantee that only trusted tasks run concurrently on a
167 core. But there could be small window of time during which untrusted tasks run
168 concurrently or kernel could be running concurrently with a task not trusted by
169 kernel.
170 
171 IPI processing delays
172 ~~~~~~~~~~~~~~~~~~~~~
173 Core scheduling selects only trusted tasks to run together. IPI is used to notify
174 the siblings to switch to the new task. But there could be hardware delays in
175 receiving of the IPI on some arch (on x86, this has not been observed). This may
176 cause an attacker task to start running on a CPU before its siblings receive the
177 IPI. Even though cache is flushed on entry to user mode, victim tasks on siblings
178 may populate data in the cache and micro architectural buffers after the attacker
179 starts to run and this is a possibility for data leak.
180 
181 Open cross-HT issues that core scheduling does not solve
182 --------------------------------------------------------
183 1. For MDS
184 ~~~~~~~~~~
185 Core scheduling cannot protect against MDS attacks between the siblings
186 running in user mode and the others running in kernel mode. Even though all
187 siblings run tasks which trust each other, when the kernel is executing
188 code on behalf of a task, it cannot trust the code running in the
189 sibling. Such attacks are possible for any combination of sibling CPU modes
190 (host or guest mode).
191 
192 2. For L1TF
193 ~~~~~~~~~~~
194 Core scheduling cannot protect against an L1TF guest attacker exploiting a
195 guest or host victim. This is because the guest attacker can craft invalid
196 PTEs which are not inverted due to a vulnerable guest kernel. The only
197 solution is to disable EPT (Extended Page Tables).
198 
199 For both MDS and L1TF, if the guest vCPU is configured to not trust each
200 other (by tagging separately), then the guest to guest attacks would go away.
201 Or it could be a system admin policy which considers guest to guest attacks as
202 a guest problem.
203 
204 Another approach to resolve these would be to make every untrusted task on the
205 system to not trust every other untrusted task. While this could reduce
206 parallelism of the untrusted tasks, it would still solve the above issues while
207 allowing system processes (trusted tasks) to share a core.
208 
209 3. Protecting the kernel (IRQ, syscall, VMEXIT)
210 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
211 Unfortunately, core scheduling does not protect kernel contexts running on
212 sibling hyperthreads from one another. Prototypes of mitigations have been posted
213 to LKML to solve this, but it is debatable whether such windows are practically
214 exploitable, and whether the performance overhead of the prototypes are worth
215 it (not to mention, the added code complexity).
216 
217 Other Use cases
218 ---------------
219 The main use case for Core scheduling is mitigating the cross-HT vulnerabilities
220 with SMT enabled. There are other use cases where this feature could be used:
221 
222 - Isolating tasks that needs a whole core: Examples include realtime tasks, tasks
223   that uses SIMD instructions etc.
224 - Gang scheduling: Requirements for a group of tasks that needs to be scheduled
225   together could also be realized using core scheduling. One example is vCPUs of
226   a VM.

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

sflogo.php