xref: /openbmc/qemu/docs/system/arm/cpu-features.rst (revision 2f44bea9)
1Arm CPU Features
2================
3
4CPU features are optional features that a CPU of supporting type may
5choose to implement or not.  In QEMU, optional CPU features have
6corresponding boolean CPU proprieties that, when enabled, indicate
7that the feature is implemented, and, conversely, when disabled,
8indicate that it is not implemented. An example of an Arm CPU feature
9is the Performance Monitoring Unit (PMU).  CPU types such as the
10Cortex-A15 and the Cortex-A57, which respectively implement Arm
11architecture reference manuals ARMv7-A and ARMv8-A, may both optionally
12implement PMUs.  For example, if a user wants to use a Cortex-A15 without
13a PMU, then the ``-cpu`` parameter should contain ``pmu=off`` on the QEMU
14command line, i.e. ``-cpu cortex-a15,pmu=off``.
15
16As not all CPU types support all optional CPU features, then whether or
17not a CPU property exists depends on the CPU type.  For example, CPUs
18that implement the ARMv8-A architecture reference manual may optionally
19support the AArch32 CPU feature, which may be enabled by disabling the
20``aarch64`` CPU property.  A CPU type such as the Cortex-A15, which does
21not implement ARMv8-A, will not have the ``aarch64`` CPU property.
22
23QEMU's support may be limited for some CPU features, only partially
24supporting the feature or only supporting the feature under certain
25configurations.  For example, the ``aarch64`` CPU feature, which, when
26disabled, enables the optional AArch32 CPU feature, is only supported
27when using the KVM accelerator and when running on a host CPU type that
28supports the feature.  While ``aarch64`` currently only works with KVM,
29it could work with TCG.  CPU features that are specific to KVM are
30prefixed with "kvm-" and are described in "KVM VCPU Features".
31
32CPU Feature Probing
33===================
34
35Determining which CPU features are available and functional for a given
36CPU type is possible with the ``query-cpu-model-expansion`` QMP command.
37Below are some examples where ``scripts/qmp/qmp-shell`` (see the top comment
38block in the script for usage) is used to issue the QMP commands.
39
401. Determine which CPU features are available for the ``max`` CPU type
41   (Note, we started QEMU with qemu-system-aarch64, so ``max`` is
42   implementing the ARMv8-A reference manual in this case)::
43
44      (QEMU) query-cpu-model-expansion type=full model={"name":"max"}
45      { "return": {
46        "model": { "name": "max", "props": {
47        "sve1664": true, "pmu": true, "sve1792": true, "sve1920": true,
48        "sve128": true, "aarch64": true, "sve1024": true, "sve": true,
49        "sve640": true, "sve768": true, "sve1408": true, "sve256": true,
50        "sve1152": true, "sve512": true, "sve384": true, "sve1536": true,
51        "sve896": true, "sve1280": true, "sve2048": true
52      }}}}
53
54We see that the ``max`` CPU type has the ``pmu``, ``aarch64``, ``sve``, and many
55``sve<N>`` CPU features.  We also see that all the CPU features are
56enabled, as they are all ``true``.  (The ``sve<N>`` CPU features are all
57optional SVE vector lengths (see "SVE CPU Properties").  While with TCG
58all SVE vector lengths can be supported, when KVM is in use it's more
59likely that only a few lengths will be supported, if SVE is supported at
60all.)
61
62(2) Let's try to disable the PMU::
63
64      (QEMU) query-cpu-model-expansion type=full model={"name":"max","props":{"pmu":false}}
65      { "return": {
66        "model": { "name": "max", "props": {
67        "sve1664": true, "pmu": false, "sve1792": true, "sve1920": true,
68        "sve128": true, "aarch64": true, "sve1024": true, "sve": true,
69        "sve640": true, "sve768": true, "sve1408": true, "sve256": true,
70        "sve1152": true, "sve512": true, "sve384": true, "sve1536": true,
71        "sve896": true, "sve1280": true, "sve2048": true
72      }}}}
73
74We see it worked, as ``pmu`` is now ``false``.
75
76(3) Let's try to disable ``aarch64``, which enables the AArch32 CPU feature::
77
78      (QEMU) query-cpu-model-expansion type=full model={"name":"max","props":{"aarch64":false}}
79      {"error": {
80       "class": "GenericError", "desc":
81       "'aarch64' feature cannot be disabled unless KVM is enabled and 32-bit EL1 is supported"
82      }}
83
84It looks like this feature is limited to a configuration we do not
85currently have.
86
87(4) Let's disable ``sve`` and see what happens to all the optional SVE
88    vector lengths::
89
90      (QEMU) query-cpu-model-expansion type=full model={"name":"max","props":{"sve":false}}
91      { "return": {
92        "model": { "name": "max", "props": {
93        "sve1664": false, "pmu": true, "sve1792": false, "sve1920": false,
94        "sve128": false, "aarch64": true, "sve1024": false, "sve": false,
95        "sve640": false, "sve768": false, "sve1408": false, "sve256": false,
96        "sve1152": false, "sve512": false, "sve384": false, "sve1536": false,
97        "sve896": false, "sve1280": false, "sve2048": false
98      }}}}
99
100As expected they are now all ``false``.
101
102(5) Let's try probing CPU features for the Cortex-A15 CPU type::
103
104      (QEMU) query-cpu-model-expansion type=full model={"name":"cortex-a15"}
105      {"return": {"model": {"name": "cortex-a15", "props": {"pmu": true}}}}
106
107Only the ``pmu`` CPU feature is available.
108
109A note about CPU feature dependencies
110-------------------------------------
111
112It's possible for features to have dependencies on other features. I.e.
113it may be possible to change one feature at a time without error, but
114when attempting to change all features at once an error could occur
115depending on the order they are processed.  It's also possible changing
116all at once doesn't generate an error, because a feature's dependencies
117are satisfied with other features, but the same feature cannot be changed
118independently without error.  For these reasons callers should always
119attempt to make their desired changes all at once in order to ensure the
120collection is valid.
121
122A note about CPU models and KVM
123-------------------------------
124
125Named CPU models generally do not work with KVM.  There are a few cases
126that do work, e.g. using the named CPU model ``cortex-a57`` with KVM on a
127seattle host, but mostly if KVM is enabled the ``host`` CPU type must be
128used.  This means the guest is provided all the same CPU features as the
129host CPU type has.  And, for this reason, the ``host`` CPU type should
130enable all CPU features that the host has by default.  Indeed it's even
131a bit strange to allow disabling CPU features that the host has when using
132the ``host`` CPU type, but in the absence of CPU models it's the best we can
133do if we want to launch guests without all the host's CPU features enabled.
134
135Enabling KVM also affects the ``query-cpu-model-expansion`` QMP command.  The
136affect is not only limited to specific features, as pointed out in example
137(3) of "CPU Feature Probing", but also to which CPU types may be expanded.
138When KVM is enabled, only the ``max``, ``host``, and current CPU type may be
139expanded.  This restriction is necessary as it's not possible to know all
140CPU types that may work with KVM, but it does impose a small risk of users
141experiencing unexpected errors.  For example on a seattle, as mentioned
142above, the ``cortex-a57`` CPU type is also valid when KVM is enabled.
143Therefore a user could use the ``host`` CPU type for the current type, but
144then attempt to query ``cortex-a57``, however that query will fail with our
145restrictions.  This shouldn't be an issue though as management layers and
146users have been preferring the ``host`` CPU type for use with KVM for quite
147some time.  Additionally, if the KVM-enabled QEMU instance running on a
148seattle host is using the ``cortex-a57`` CPU type, then querying ``cortex-a57``
149will work.
150
151Using CPU Features
152==================
153
154After determining which CPU features are available and supported for a
155given CPU type, then they may be selectively enabled or disabled on the
156QEMU command line with that CPU type::
157
158  $ qemu-system-aarch64 -M virt -cpu max,pmu=off,sve=on,sve128=on,sve256=on
159
160The example above disables the PMU and enables the first two SVE vector
161lengths for the ``max`` CPU type.  Note, the ``sve=on`` isn't actually
162necessary, because, as we observed above with our probe of the ``max`` CPU
163type, ``sve`` is already on by default.  Also, based on our probe of
164defaults, it would seem we need to disable many SVE vector lengths, rather
165than only enabling the two we want.  This isn't the case, because, as
166disabling many SVE vector lengths would be quite verbose, the ``sve<N>`` CPU
167properties have special semantics (see "SVE CPU Property Parsing
168Semantics").
169
170KVM VCPU Features
171=================
172
173KVM VCPU features are CPU features that are specific to KVM, such as
174paravirt features or features that enable CPU virtualization extensions.
175The features' CPU properties are only available when KVM is enabled and
176are named with the prefix "kvm-".  KVM VCPU features may be probed,
177enabled, and disabled in the same way as other CPU features.  Below is
178the list of KVM VCPU features and their descriptions.
179
180  kvm-no-adjvtime          By default kvm-no-adjvtime is disabled.  This
181                           means that by default the virtual time
182                           adjustment is enabled (vtime is not *not*
183                           adjusted).
184
185                           When virtual time adjustment is enabled each
186                           time the VM transitions back to running state
187                           the VCPU's virtual counter is updated to ensure
188                           stopped time is not counted.  This avoids time
189                           jumps surprising guest OSes and applications,
190                           as long as they use the virtual counter for
191                           timekeeping.  However it has the side effect of
192                           the virtual and physical counters diverging.
193                           All timekeeping based on the virtual counter
194                           will appear to lag behind any timekeeping that
195                           does not subtract VM stopped time.  The guest
196                           may resynchronize its virtual counter with
197                           other time sources as needed.
198
199                           Enable kvm-no-adjvtime to disable virtual time
200                           adjustment, also restoring the legacy (pre-5.0)
201                           behavior.
202
203  kvm-steal-time           Since v5.2, kvm-steal-time is enabled by
204                           default when KVM is enabled, the feature is
205                           supported, and the guest is 64-bit.
206
207                           When kvm-steal-time is enabled a 64-bit guest
208                           can account for time its CPUs were not running
209                           due to the host not scheduling the corresponding
210                           VCPU threads.  The accounting statistics may
211                           influence the guest scheduler behavior and/or be
212                           exposed to the guest userspace.
213
214TCG VCPU Features
215=================
216
217TCG VCPU features are CPU features that are specific to TCG.
218Below is the list of TCG VCPU features and their descriptions.
219
220  pauth                    Enable or disable ``FEAT_Pauth``, pointer
221                           authentication.  By default, the feature is
222                           enabled with ``-cpu max``.
223
224  pauth-impdef             When ``FEAT_Pauth`` is enabled, either the
225                           *impdef* (Implementation Defined) algorithm
226                           is enabled or the *architected* QARMA algorithm
227                           is enabled.  By default the impdef algorithm
228                           is disabled, and QARMA is enabled.
229
230                           The architected QARMA algorithm has good
231                           cryptographic properties, but can be quite slow
232                           to emulate.  The impdef algorithm used by QEMU
233                           is non-cryptographic but significantly faster.
234
235SVE CPU Properties
236==================
237
238There are two types of SVE CPU properties: ``sve`` and ``sve<N>``.  The first
239is used to enable or disable the entire SVE feature, just as the ``pmu``
240CPU property completely enables or disables the PMU.  The second type
241is used to enable or disable specific vector lengths, where ``N`` is the
242number of bits of the length.  The ``sve<N>`` CPU properties have special
243dependencies and constraints, see "SVE CPU Property Dependencies and
244Constraints" below.  Additionally, as we want all supported vector lengths
245to be enabled by default, then, in order to avoid overly verbose command
246lines (command lines full of ``sve<N>=off``, for all ``N`` not wanted), we
247provide the parsing semantics listed in "SVE CPU Property Parsing
248Semantics".
249
250SVE CPU Property Dependencies and Constraints
251---------------------------------------------
252
253  1) At least one vector length must be enabled when ``sve`` is enabled.
254
255  2) If a vector length ``N`` is enabled, then, when KVM is enabled, all
256     smaller, host supported vector lengths must also be enabled.  If
257     KVM is not enabled, then only all the smaller, power-of-two vector
258     lengths must be enabled.  E.g. with KVM if the host supports all
259     vector lengths up to 512-bits (128, 256, 384, 512), then if ``sve512``
260     is enabled, the 128-bit vector length, 256-bit vector length, and
261     384-bit vector length must also be enabled. Without KVM, the 384-bit
262     vector length would not be required.
263
264  3) If KVM is enabled then only vector lengths that the host CPU type
265     support may be enabled.  If SVE is not supported by the host, then
266     no ``sve*`` properties may be enabled.
267
268SVE CPU Property Parsing Semantics
269----------------------------------
270
271  1) If SVE is disabled (``sve=off``), then which SVE vector lengths
272     are enabled or disabled is irrelevant to the guest, as the entire
273     SVE feature is disabled and that disables all vector lengths for
274     the guest.  However QEMU will still track any ``sve<N>`` CPU
275     properties provided by the user.  If later an ``sve=on`` is provided,
276     then the guest will get only the enabled lengths.  If no ``sve=on``
277     is provided and there are explicitly enabled vector lengths, then
278     an error is generated.
279
280  2) If SVE is enabled (``sve=on``), but no ``sve<N>`` CPU properties are
281     provided, then all supported vector lengths are enabled, which when
282     KVM is not in use means including the non-power-of-two lengths, and,
283     when KVM is in use, it means all vector lengths supported by the host
284     processor.
285
286  3) If SVE is enabled, then an error is generated when attempting to
287     disable the last enabled vector length (see constraint (1) of "SVE
288     CPU Property Dependencies and Constraints").
289
290  4) If one or more vector lengths have been explicitly enabled and at
291     at least one of the dependency lengths of the maximum enabled length
292     has been explicitly disabled, then an error is generated (see
293     constraint (2) of "SVE CPU Property Dependencies and Constraints").
294
295  5) When KVM is enabled, if the host does not support SVE, then an error
296     is generated when attempting to enable any ``sve*`` properties (see
297     constraint (3) of "SVE CPU Property Dependencies and Constraints").
298
299  6) When KVM is enabled, if the host does support SVE, then an error is
300     generated when attempting to enable any vector lengths not supported
301     by the host (see constraint (3) of "SVE CPU Property Dependencies and
302     Constraints").
303
304  7) If one or more ``sve<N>`` CPU properties are set ``off``, but no ``sve<N>``,
305     CPU properties are set ``on``, then the specified vector lengths are
306     disabled but the default for any unspecified lengths remains enabled.
307     When KVM is not enabled, disabling a power-of-two vector length also
308     disables all vector lengths larger than the power-of-two length.
309     When KVM is enabled, then disabling any supported vector length also
310     disables all larger vector lengths (see constraint (2) of "SVE CPU
311     Property Dependencies and Constraints").
312
313  8) If one or more ``sve<N>`` CPU properties are set to ``on``, then they
314     are enabled and all unspecified lengths default to disabled, except
315     for the required lengths per constraint (2) of "SVE CPU Property
316     Dependencies and Constraints", which will even be auto-enabled if
317     they were not explicitly enabled.
318
319  9) If SVE was disabled (``sve=off``), allowing all vector lengths to be
320     explicitly disabled (i.e. avoiding the error specified in (3) of
321     "SVE CPU Property Parsing Semantics"), then if later an ``sve=on`` is
322     provided an error will be generated.  To avoid this error, one must
323     enable at least one vector length prior to enabling SVE.
324
325SVE CPU Property Examples
326-------------------------
327
328  1) Disable SVE::
329
330     $ qemu-system-aarch64 -M virt -cpu max,sve=off
331
332  2) Implicitly enable all vector lengths for the ``max`` CPU type::
333
334     $ qemu-system-aarch64 -M virt -cpu max
335
336  3) When KVM is enabled, implicitly enable all host CPU supported vector
337     lengths with the ``host`` CPU type::
338
339     $ qemu-system-aarch64 -M virt,accel=kvm -cpu host
340
341  4) Only enable the 128-bit vector length::
342
343     $ qemu-system-aarch64 -M virt -cpu max,sve128=on
344
345  5) Disable the 512-bit vector length and all larger vector lengths,
346     since 512 is a power-of-two.  This results in all the smaller,
347     uninitialized lengths (128, 256, and 384) defaulting to enabled::
348
349     $ qemu-system-aarch64 -M virt -cpu max,sve512=off
350
351  6) Enable the 128-bit, 256-bit, and 512-bit vector lengths::
352
353     $ qemu-system-aarch64 -M virt -cpu max,sve128=on,sve256=on,sve512=on
354
355  7) The same as (6), but since the 128-bit and 256-bit vector
356     lengths are required for the 512-bit vector length to be enabled,
357     then allow them to be auto-enabled::
358
359     $ qemu-system-aarch64 -M virt -cpu max,sve512=on
360
361  8) Do the same as (7), but by first disabling SVE and then re-enabling it::
362
363     $ qemu-system-aarch64 -M virt -cpu max,sve=off,sve512=on,sve=on
364
365  9) Force errors regarding the last vector length::
366
367     $ qemu-system-aarch64 -M virt -cpu max,sve128=off
368     $ qemu-system-aarch64 -M virt -cpu max,sve=off,sve128=off,sve=on
369
370SVE CPU Property Recommendations
371--------------------------------
372
373The examples in "SVE CPU Property Examples" exhibit many ways to select
374vector lengths which developers may find useful in order to avoid overly
375verbose command lines.  However, the recommended way to select vector
376lengths is to explicitly enable each desired length.  Therefore only
377example's (1), (4), and (6) exhibit recommended uses of the properties.
378
379SVE User-mode Default Vector Length Property
380--------------------------------------------
381
382For qemu-aarch64, the cpu property ``sve-default-vector-length=N`` is
383defined to mirror the Linux kernel parameter file
384``/proc/sys/abi/sve_default_vector_length``.  The default length, ``N``,
385is in units of bytes and must be between 16 and 8192.
386If not specified, the default vector length is 64.
387
388If the default length is larger than the maximum vector length enabled,
389the actual vector length will be reduced.  Note that the maximum vector
390length supported by QEMU is 256.
391
392If this property is set to ``-1`` then the default vector length
393is set to the maximum possible length.
394