| f396aec8 | 09-Sep-2025 |
Arvind Ram Prakash <arvind.ramprakash@arm.com> |
feat(cpufeat): add support for FEAT_IDTE3
This patch adds support for FEAT_IDTE3, which introduces support
for handling the trapping of Group 3 and Group 5 (only GMID_EL1)
registers to EL3 (unless t
feat(cpufeat): add support for FEAT_IDTE3
This patch adds support for FEAT_IDTE3, which introduces support
for handling the trapping of Group 3 and Group 5 (only GMID_EL1)
registers to EL3 (unless trapped to EL2). IDTE3 allows EL3 to
modify the view of ID registers for lower ELs, and this capability
is used to disable fields of ID registers tied to disabled features.
The ID registers are initially read as-is and stored in context.
Then, based on the feature enablement status for each world, if a
particular feature is disabled, its corresponding field in the
cached ID register is set to Res0. When lower ELs attempt to read
an ID register, the cached ID register value is returned. This
allows EL3 to prevent lower ELs from accessing feature-specific
system registers that are disabled in EL3, even though the hardware
implements them.
The emulated ID register values are stored primarily in per-world
context, except for certain debug-related ID registers such as
ID_AA64DFR0_EL1 and ID_AA64DFR1_EL1, which are stored in the
cpu_data and are unique to each PE. This is done to support feature
asymmetry that is commonly seen in debug features.
FEAT_IDTE3 traps all Group 3 ID registers in the range
op0 == 3, op1 == 0, CRn == 0, CRm == {2–7}, op2 == {0–7} and the
Group 5 GMID_EL1 register. However, only a handful of ID registers
contain fields used to detect features enabled in EL3. Hence, we
only cache those ID registers, while the rest are transparently
returned as is to the lower EL.
This patch updates the CREATE_FEATURE_FUNCS macro to generate
update_feat_xyz_idreg_field() functions that disable ID register
fields on a per-feature basis. The enabled_worlds scope is used to
disable ID register fields for security states where the feature is
not enabled.
This EXPERIMENTAL feature is controlled by the ENABLE_FEAT_IDTE3
build flag and is currently disabled by default.
Signed-off-by: Arvind Ram Prakash <arvind.ramprakash@arm.com>
Change-Id: I5f998eeab81bb48c7595addc5595313a9ebb96d5
show more ...
|
| 4779becd | 06-Aug-2025 |
Boyan Karatotev <boyan.karatotev@arm.com> |
refactor(el3-runtime): streamline cpu_data assembly offsets using the cpu_ops template
The cpu_data structure, just like cpu_ops, is collection of disparate
data that must be accessible from both C
refactor(el3-runtime): streamline cpu_data assembly offsets using the cpu_ops template
The cpu_data structure, just like cpu_ops, is collection of disparate
data that must be accessible from both C and assembly. Achieving this is
tricky as there is no way to export structure offsets from C directly so
they must be manually recreated with `#define`s and asserts. However,
the cpu_data structure is quite old and the assembly offsets are a
patchwork of additions and extremely difficult to reason with and
modify. In fact, certain currently unused builds with
ENABLE_RUNTIME_INSTRUMENTATION=1 fail to build.
To untangle this, convert the assembly offsets to the pattern used for
the cpu_ops structure. That is, first define the sizes of every member,
as generically as possible, and then chain their offsets one after the
other. To make sure this is always correct, add a CASSERT for the offset
of every member. This makes it easy to modify the structure and fixes
the build failures.
Change-Id: I61aeb55e9c494896663a3c719c10e3c072f56349
Signed-off-by: Boyan Karatotev <boyan.karatotev@arm.com>
show more ...
|
| 34a22a02 | 05-Aug-2025 |
Boyan Karatotev <boyan.karatotev@arm.com> |
refactor(el3-runtime): move context security states to context.h
The three security states (S, NS, RL) are architecturally quite
consistent - anything that uses them has the same numerical assignmen
refactor(el3-runtime): move context security states to context.h
The three security states (S, NS, RL) are architecturally quite
consistent - anything that uses them has the same numerical assignments
(0, 1, 2) and they are quite convenient for indexing. However, we're not
as consistent in tf-a and this is defined in a few places. Since
cpu_data has a dependency on the context management library, use its
security state convention in a few more places and take away this
responsibility from cpu_data.
Change-Id: Iec73b2be2eef91975554767557de72424d0031f1
Signed-off-by: Boyan Karatotev <boyan.karatotev@arm.com>
show more ...
|
| 0a580b51 | 15-Nov-2024 |
Boyan Karatotev <boyan.karatotev@arm.com> |
perf(cm): drop ZCR_EL3 saving and some ISBs and replace them with root context
SVE and SME aren't enabled symmetrically for all worlds, but EL3 needs
to context switch them nonetheless. Previously,
perf(cm): drop ZCR_EL3 saving and some ISBs and replace them with root context
SVE and SME aren't enabled symmetrically for all worlds, but EL3 needs
to context switch them nonetheless. Previously, this had to happen by
writing the enable bits just before reading/writing the relevant
context. But since the introduction of root context, this need not be
the case. We can have these enables always be present for EL3 and save
on some work (and ISBs!) on every context switch.
We can also hoist ZCR_EL3 to a never changing register, as we set its
value to be identical for every world, which happens to be the one we
want for EL3 too.
Change-Id: I3d950e72049a298008205ba32f230d5a5c02f8b0
Signed-off-by: Boyan Karatotev <boyan.karatotev@arm.com>
show more ...
|
| 83ec7e45 | 06-Nov-2024 |
Boyan Karatotev <boyan.karatotev@arm.com> |
perf(amu): greatly simplify AMU context management
The current code is incredibly resilient to updates to the spec and
has worked quite well so far. However, recent implementations expose a
weakness
perf(amu): greatly simplify AMU context management
The current code is incredibly resilient to updates to the spec and
has worked quite well so far. However, recent implementations expose a
weakness in that this is rather slow. A large part of it is written in
assembly, making it opaque to the compiler for optimisations. The
future proofness requires reading registers that are effectively
`volatile`, making it even harder for the compiler, as well as adding
lots of implicit barriers, making it hard for the microarchitecutre to
optimise as well.
We can make a few assumptions, checked by a few well placed asserts, and
remove a lot of this burden. For a start, at the moment there are 4
group 0 counters with static assignments. Contexting them is a trivial
affair that doesn't need a loop. Similarly, there can only be up to 16
group 1 counters. Contexting them is a bit harder, but we can do with a
single branch with a falling through switch. If/when both of these
change, we have a pair of asserts and the feature detection mechanism to
guard us against pretending that we support something we don't.
We can drop contexting of the offset registers. They are fully
accessible by EL2 and as such are its responsibility to preserve on
powerdown.
Another small thing we can do, is pass the core_pos into the hook.
The caller already knows which core we're running on, we don't need to
call this non-trivial function again.
Finally, knowing this, we don't really need the auxiliary AMUs to be
described by the device tree. Linux doesn't care at the moment, and any
information we need for EL3 can be neatly placed in a simple array.
All of this, combined with lifting the actual saving out of assembly,
reduces the instructions to save the context from 180 to 40, including a
lot fewer branches. The code is also much shorter and easier to read.
Also propagate to aarch32 so that the two don't diverge too much.
Change-Id: Ib62e6e9ba5be7fb9fb8965c8eee148d5598a5361
Signed-off-by: Boyan Karatotev <boyan.karatotev@arm.com>
show more ...
|