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>
This commit is contained in:
Boyan Karatotev 2024-11-06 14:55:35 +00:00
parent 2590e819eb
commit 83ec7e452c
30 changed files with 437 additions and 1881 deletions

View file

@ -1185,7 +1185,6 @@ $(eval $(call assert_booleans,\
DYN_DISABLE_AUTH \
EL3_EXCEPTION_HANDLING \
ENABLE_AMU_AUXILIARY_COUNTERS \
ENABLE_AMU_FCONF \
AMU_RESTRICT_COUNTERS \
ENABLE_ASSERTIONS \
ENABLE_PIE \
@ -1360,7 +1359,6 @@ $(eval $(call add_defines,\
DISABLE_MTPMU \
ENABLE_FEAT_AMU \
ENABLE_AMU_AUXILIARY_COUNTERS \
ENABLE_AMU_FCONF \
AMU_RESTRICT_COUNTERS \
ENABLE_ASSERTIONS \
ENABLE_BTI \

View file

@ -127,7 +127,7 @@ void bl31_setup(u_register_t arg0, u_register_t arg1, u_register_t arg2,
void bl31_main(void)
{
/* Init registers that never change for the lifetime of TF-A */
cm_manage_extensions_el3();
cm_manage_extensions_el3(plat_my_core_pos());
/* Init per-world context registers for non-secure world */
manage_extensions_nonsecure_per_world();

View file

@ -20,15 +20,9 @@ known as group 1 counters), controlled by the ``ENABLE_AMU_AUXILIARY_COUNTERS``
build option.
As a security precaution, Trusted Firmware-A does not enable these by default.
Instead, platforms may configure their auxiliary counters through one of two
possible mechanisms:
- |FCONF|, controlled by the ``ENABLE_AMU_FCONF`` build option.
- A platform implementation of the ``plat_amu_topology`` function (the default).
See :ref:`Activity Monitor Unit (AMU) Bindings` for documentation on the |FCONF|
device tree bindings.
Instead, platforms must configure their auxiliary counters through the
``plat_amu_aux_enables`` platform hook.
--------------
*Copyright (c) 2021, Arm Limited. All rights reserved.*
*Copyright (c) 2021-2025, Arm Limited. All rights reserved.*

View file

@ -1,142 +0,0 @@
Activity Monitor Unit (AMU) Bindings
====================================
To support platform-defined Activity Monitor Unit (|AMU|) auxiliary counters
through FCONF, the ``HW_CONFIG`` device tree accepts several |AMU|-specific
nodes and properties.
Bindings
^^^^^^^^
.. contents::
:local:
``/cpus/cpus/cpu*`` node properties
"""""""""""""""""""""""""""""""""""
The ``cpu`` node has been augmented to support a handle to an associated |AMU|
view, which should describe the counters offered by the core.
+---------------+-------+---------------+-------------------------------------+
| Property name | Usage | Value type | Description |
+===============+=======+===============+=====================================+
| ``amu`` | O | ``<phandle>`` | If present, indicates that an |AMU| |
| | | | is available and its counters are |
| | | | described by the node provided. |
+---------------+-------+---------------+-------------------------------------+
``/cpus/amus`` node properties
""""""""""""""""""""""""""""""
The ``amus`` node describes the |AMUs| implemented by the cores in the system.
This node does not have any properties.
``/cpus/amus/amu*`` node properties
"""""""""""""""""""""""""""""""""""
An ``amu`` node describes the layout and meaning of the auxiliary counter
registers of one or more |AMUs|, and may be shared by multiple cores.
+--------------------+-------+------------+------------------------------------+
| Property name | Usage | Value type | Description |
+====================+=======+============+====================================+
| ``#address-cells`` | R | ``<u32>`` | Value shall be 1. Specifies that |
| | | | the ``reg`` property array of |
| | | | children of this node uses a |
| | | | single cell. |
+--------------------+-------+------------+------------------------------------+
| ``#size-cells`` | R | ``<u32>`` | Value shall be 0. Specifies that |
| | | | no size is required in the ``reg`` |
| | | | property in children of this node. |
+--------------------+-------+------------+------------------------------------+
``/cpus/amus/amu*/counter*`` node properties
""""""""""""""""""""""""""""""""""""""""""""
A ``counter`` node describes an auxiliary counter belonging to the parent |AMU|
view.
+-------------------+-------+-------------+------------------------------------+
| Property name | Usage | Value type | Description |
+===================+=======+=============+====================================+
| ``reg`` | R | array | Represents the counter register |
| | | | index, and must be a single cell. |
+-------------------+-------+-------------+------------------------------------+
| ``enable-at-el3`` | O | ``<empty>`` | The presence of this property |
| | | | indicates that this counter should |
| | | | be enabled prior to EL3 exit. |
+-------------------+-------+-------------+------------------------------------+
Example
^^^^^^^
An example system offering four cores made up of two clusters, where the cores
of each cluster share different |AMUs|, may use something like the following:
.. code-block::
cpus {
#address-cells = <2>;
#size-cells = <0>;
amus {
amu0: amu-0 {
#address-cells = <1>;
#size-cells = <0>;
counterX: counter@0 {
reg = <0>;
enable-at-el3;
};
counterY: counter@1 {
reg = <1>;
enable-at-el3;
};
};
amu1: amu-1 {
#address-cells = <1>;
#size-cells = <0>;
counterZ: counter@0 {
reg = <0>;
enable-at-el3;
};
};
};
cpu0@00000 {
...
amu = <&amu0>;
};
cpu1@00100 {
...
amu = <&amu0>;
};
cpu2@10000 {
...
amu = <&amu1>;
};
cpu3@10100 {
...
amu = <&amu1>;
};
}
In this situation, ``cpu0`` and ``cpu1`` (the two cores in the first cluster),
share the view of their AMUs defined by ``amu0``. Likewise, ``cpu2`` and
``cpu3`` (the two cores in the second cluster), share the view of their |AMUs|
defined by ``amu1``. This will cause ``counterX`` and ``counterY`` to be enabled
for both ``cpu0`` and ``cpu1``, and ``counterZ`` to be enabled for both ``cpu2``
and ``cpu3``.

View file

@ -145,5 +145,4 @@ Properties binding information
:maxdepth: 1
fconf_properties
amu-bindings
tb_fw_bindings

View file

@ -258,10 +258,6 @@ Common build options
(also known as group 1 counters). These are implementation-defined counters,
and as such require additional platform configuration. Default is 0.
- ``ENABLE_AMU_FCONF``: Enables configuration of the AMU through FCONF, which
allows platforms with auxiliary counters to describe them via the
``HW_CONFIG`` device tree blob. Default is 0.
- ``ENABLE_ASSERTIONS``: This option controls whether or not calls to ``assert()``
are compiled out. For debug builds, this option defaults to 1, and calls to
``assert()`` are left in place. For release builds, this option defaults to 0

View file

@ -104,28 +104,6 @@
};
};
amus {
amu: amu-0 {
#address-cells = <1>;
#size-cells = <0>;
mpmm_gear0: counter@0 {
reg = <0>;
enable-at-el3;
};
mpmm_gear1: counter@1 {
reg = <1>;
enable-at-el3;
};
mpmm_gear2: counter@2 {
reg = <2>;
enable-at-el3;
};
};
};
CPU0:cpu@0 {
device_type = "cpu";
compatible = "arm,armv8";
@ -134,7 +112,6 @@
clocks = <&scmi_dvfs 0>;
cpu-idle-states = <&CPU_SLEEP_0 &CLUSTER_SLEEP_0>;
capacity-dmips-mhz = <LIT_CAPACITY>;
amu = <&amu>;
};
CPU1:cpu@100 {
@ -145,7 +122,6 @@
clocks = <&scmi_dvfs 0>;
cpu-idle-states = <&CPU_SLEEP_0 &CLUSTER_SLEEP_0>;
capacity-dmips-mhz = <LIT_CAPACITY>;
amu = <&amu>;
};
CPU2:cpu@200 {
@ -154,7 +130,6 @@
reg = <0x200>;
enable-method = "psci";
cpu-idle-states = <&CPU_SLEEP_0 &CLUSTER_SLEEP_0>;
amu = <&amu>;
};
CPU3:cpu@300 {
@ -163,7 +138,6 @@
reg = <0x300>;
enable-method = "psci";
cpu-idle-states = <&CPU_SLEEP_0 &CLUSTER_SLEEP_0>;
amu = <&amu>;
};
CPU4:cpu@400 {
@ -174,7 +148,6 @@
clocks = <&scmi_dvfs 1>;
cpu-idle-states = <&CPU_SLEEP_0 &CLUSTER_SLEEP_0>;
capacity-dmips-mhz = <MID_CAPACITY>;
amu = <&amu>;
};
CPU5:cpu@500 {
@ -185,7 +158,6 @@
clocks = <&scmi_dvfs 1>;
cpu-idle-states = <&CPU_SLEEP_0 &CLUSTER_SLEEP_0>;
capacity-dmips-mhz = <MID_CAPACITY>;
amu = <&amu>;
};
CPU6:cpu@600 {
@ -194,7 +166,6 @@
reg = <0x600>;
enable-method = "psci";
cpu-idle-states = <&CPU_SLEEP_0 &CLUSTER_SLEEP_0>;
amu = <&amu>;
};
CPU7:cpu@700 {
@ -203,7 +174,6 @@
reg = <0x700>;
enable-method = "psci";
cpu-idle-states = <&CPU_SLEEP_0 &CLUSTER_SLEEP_0>;
amu = <&amu>;
};
};

View file

@ -123,7 +123,6 @@
enable-method = "psci";
clocks = <&scmi_dvfs 1>;
capacity-dmips-mhz = <MID_CAPACITY>;
amu = <&amu>;
};
CPU9:cpu@900 {
@ -133,7 +132,6 @@
enable-method = "psci";
clocks = <&scmi_dvfs 2>;
capacity-dmips-mhz = <BIG2_CAPACITY>;
amu = <&amu>;
};
CPU10:cpu@A00 {
@ -143,7 +141,6 @@
enable-method = "psci";
clocks = <&scmi_dvfs 2>;
capacity-dmips-mhz = <BIG2_CAPACITY>;
amu = <&amu>;
};
CPU11:cpu@B00 {
@ -153,7 +150,6 @@
enable-method = "psci";
clocks = <&scmi_dvfs 2>;
capacity-dmips-mhz = <BIG2_CAPACITY>;
amu = <&amu>;
};
CPU12:cpu@C00 {
@ -163,7 +159,6 @@
enable-method = "psci";
clocks = <&scmi_dvfs 3>;
capacity-dmips-mhz = <BIG_CAPACITY>;
amu = <&amu>;
};
CPU13:cpu@D00 {
@ -173,7 +168,6 @@
enable-method = "psci";
clocks = <&scmi_dvfs 3>;
capacity-dmips-mhz = <BIG_CAPACITY>;
amu = <&amu>;
};
#endif
};

View file

@ -761,7 +761,7 @@
/* AMCNTENSET0 definitions */
#define AMCNTENSET0_Pn_SHIFT U(0)
#define AMCNTENSET0_Pn_MASK U(0xffff)
#define AMCNTENSET0_Pn_MASK U(0xf)
/* AMCNTENSET1 definitions */
#define AMCNTENSET1_Pn_SHIFT U(0)
@ -769,7 +769,7 @@
/* AMCNTENCLR0 definitions */
#define AMCNTENCLR0_Pn_SHIFT U(0)
#define AMCNTENCLR0_Pn_MASK U(0xffff)
#define AMCNTENCLR0_Pn_MASK U(0xf)
/* AMCNTENCLR1 definitions */
#define AMCNTENCLR1_Pn_SHIFT U(0)

View file

@ -1,5 +1,5 @@
/*
* Copyright (c) 2019-2024, Arm Limited. All rights reserved.
* Copyright (c) 2019-2025, Arm Limited. All rights reserved.
*
* SPDX-License-Identifier: BSD-3-Clause
*/
@ -85,6 +85,10 @@ CREATE_FEATURE_PRESENT(feat_ttcnp, id_mmfr4, ID_MMFR4_CNP_SHIFT,
CREATE_FEATURE_FUNCS(feat_amu, id_pfr0, ID_PFR0_AMU_SHIFT,
ID_PFR0_AMU_MASK, ID_PFR0_AMU_V1, ENABLE_FEAT_AMU)
/* Auxiliary counters for FEAT_AMU */
CREATE_FEATURE_FUNCS(feat_amu_aux, amcfgr, AMCFGR_NCG_SHIFT,
AMCFGR_NCG_MASK, 1U, ENABLE_AMU_AUXILIARY_COUNTERS)
/* FEAT_AMUV1P1: AMU Extension v1.1 */
CREATE_FEATURE_FUNCS(feat_amuv1p1, id_pfr0, ID_PFR0_AMU_SHIFT,
ID_PFR0_AMU_MASK, ID_PFR0_AMU_V1P1, ENABLE_FEAT_AMUv1p1)

View file

@ -324,6 +324,23 @@ DEFINE_COPROCR_RW_FUNCS_64(amevcntr01, AMEVCNTR01)
DEFINE_COPROCR_RW_FUNCS_64(amevcntr02, AMEVCNTR02)
DEFINE_COPROCR_RW_FUNCS_64(amevcntr03, AMEVCNTR03)
DEFINE_COPROCR_RW_FUNCS_64(amevcntr10, AMEVCNTR10);
DEFINE_COPROCR_RW_FUNCS_64(amevcntr11, AMEVCNTR11);
DEFINE_COPROCR_RW_FUNCS_64(amevcntr12, AMEVCNTR12);
DEFINE_COPROCR_RW_FUNCS_64(amevcntr13, AMEVCNTR13);
DEFINE_COPROCR_RW_FUNCS_64(amevcntr14, AMEVCNTR14);
DEFINE_COPROCR_RW_FUNCS_64(amevcntr15, AMEVCNTR15);
DEFINE_COPROCR_RW_FUNCS_64(amevcntr16, AMEVCNTR16);
DEFINE_COPROCR_RW_FUNCS_64(amevcntr17, AMEVCNTR17);
DEFINE_COPROCR_RW_FUNCS_64(amevcntr18, AMEVCNTR18);
DEFINE_COPROCR_RW_FUNCS_64(amevcntr19, AMEVCNTR19);
DEFINE_COPROCR_RW_FUNCS_64(amevcntr1a, AMEVCNTR1A);
DEFINE_COPROCR_RW_FUNCS_64(amevcntr1b, AMEVCNTR1B);
DEFINE_COPROCR_RW_FUNCS_64(amevcntr1c, AMEVCNTR1C);
DEFINE_COPROCR_RW_FUNCS_64(amevcntr1d, AMEVCNTR1D);
DEFINE_COPROCR_RW_FUNCS_64(amevcntr1e, AMEVCNTR1E);
DEFINE_COPROCR_RW_FUNCS_64(amevcntr1f, AMEVCNTR1F);
/*
* TLBI operation prototypes
*/

View file

@ -311,6 +311,10 @@ CREATE_FEATURE_FUNCS(feat_gcs, id_aa64pfr1_el1, ID_AA64PFR1_EL1_GCS_SHIFT,
CREATE_FEATURE_FUNCS(feat_amu, id_aa64pfr0_el1, ID_AA64PFR0_AMU_SHIFT,
ID_AA64PFR0_AMU_MASK, 1U, ENABLE_FEAT_AMU)
/* Auxiliary counters for FEAT_AMU */
CREATE_FEATURE_FUNCS(feat_amu_aux, amcfgr_el0, AMCFGR_EL0_NCG_SHIFT,
AMCFGR_EL0_NCG_MASK, 1U, ENABLE_AMU_AUXILIARY_COUNTERS)
/* FEAT_AMUV1P1: AMU Extension v1.1 */
CREATE_FEATURE_FUNCS(feat_amuv1p1, id_aa64pfr0_el1, ID_AA64PFR0_AMU_SHIFT,
ID_AA64PFR0_AMU_MASK, ID_AA64PFR0_AMU_V1P1, ENABLE_FEAT_AMUv1p1)

View file

@ -564,6 +564,27 @@ DEFINE_RENAME_SYSREG_RW_FUNCS(amcntenclr0_el0, AMCNTENCLR0_EL0)
DEFINE_RENAME_SYSREG_RW_FUNCS(amcntenset0_el0, AMCNTENSET0_EL0)
DEFINE_RENAME_SYSREG_RW_FUNCS(amcntenclr1_el0, AMCNTENCLR1_EL0)
DEFINE_RENAME_SYSREG_RW_FUNCS(amcntenset1_el0, AMCNTENSET1_EL0)
DEFINE_RENAME_SYSREG_RW_FUNCS(amevcntr00_el0, AMEVCNTR00_EL0);
DEFINE_RENAME_SYSREG_RW_FUNCS(amevcntr01_el0, AMEVCNTR01_EL0);
DEFINE_RENAME_SYSREG_RW_FUNCS(amevcntr02_el0, AMEVCNTR02_EL0);
DEFINE_RENAME_SYSREG_RW_FUNCS(amevcntr03_el0, AMEVCNTR03_EL0);
DEFINE_RENAME_SYSREG_RW_FUNCS(amevcntr10_el0, AMEVCNTR10_EL0);
DEFINE_RENAME_SYSREG_RW_FUNCS(amevcntr11_el0, AMEVCNTR11_EL0);
DEFINE_RENAME_SYSREG_RW_FUNCS(amevcntr12_el0, AMEVCNTR12_EL0);
DEFINE_RENAME_SYSREG_RW_FUNCS(amevcntr13_el0, AMEVCNTR13_EL0);
DEFINE_RENAME_SYSREG_RW_FUNCS(amevcntr14_el0, AMEVCNTR14_EL0);
DEFINE_RENAME_SYSREG_RW_FUNCS(amevcntr15_el0, AMEVCNTR15_EL0);
DEFINE_RENAME_SYSREG_RW_FUNCS(amevcntr16_el0, AMEVCNTR16_EL0);
DEFINE_RENAME_SYSREG_RW_FUNCS(amevcntr17_el0, AMEVCNTR17_EL0);
DEFINE_RENAME_SYSREG_RW_FUNCS(amevcntr18_el0, AMEVCNTR18_EL0);
DEFINE_RENAME_SYSREG_RW_FUNCS(amevcntr19_el0, AMEVCNTR19_EL0);
DEFINE_RENAME_SYSREG_RW_FUNCS(amevcntr1a_el0, AMEVCNTR1A_EL0);
DEFINE_RENAME_SYSREG_RW_FUNCS(amevcntr1b_el0, AMEVCNTR1B_EL0);
DEFINE_RENAME_SYSREG_RW_FUNCS(amevcntr1c_el0, AMEVCNTR1C_EL0);
DEFINE_RENAME_SYSREG_RW_FUNCS(amevcntr1d_el0, AMEVCNTR1D_EL0);
DEFINE_RENAME_SYSREG_RW_FUNCS(amevcntr1e_el0, AMEVCNTR1E_EL0);
DEFINE_RENAME_SYSREG_RW_FUNCS(amevcntr1f_el0, AMEVCNTR1F_EL0);
DEFINE_RENAME_SYSREG_RW_FUNCS(pmblimitr_el1, PMBLIMITR_EL1)

View file

@ -1,5 +1,5 @@
/*
* Copyright (c) 2013-2024, Arm Limited and Contributors. All rights reserved.
* Copyright (c) 2013-2025, Arm Limited and Contributors. All rights reserved.
*
* SPDX-License-Identifier: BSD-3-Clause
*/
@ -41,7 +41,7 @@ void cm_init_context_by_index(unsigned int cpu_idx,
#ifdef __aarch64__
#if IMAGE_BL31
void cm_manage_extensions_el3(void);
void cm_manage_extensions_el3(unsigned int my_idx);
void manage_extensions_nonsecure_per_world(void);
void cm_el3_arch_init_per_world(per_world_context_t *per_world_ctx);
void cm_handle_asymmetric_features(void);
@ -95,7 +95,7 @@ static inline void cm_set_next_context(void *context)
#else
void *cm_get_next_context(void);
void cm_set_next_context(void *context);
static inline void cm_manage_extensions_el3(void) {}
static inline void cm_manage_extensions_el3(unsigned int cpu_idx) {}
static inline void manage_extensions_nonsecure_per_world(void) {}
static inline void cm_handle_asymmetric_features(void) {}
#endif /* __aarch64__ */

View file

@ -1,5 +1,5 @@
/*
* Copyright (c) 2017-2023, Arm Limited and Contributors. All rights reserved.
* Copyright (c) 2017-2025, Arm Limited and Contributors. All rights reserved.
*
* SPDX-License-Identifier: BSD-3-Clause
*/
@ -17,7 +17,7 @@
#if ENABLE_FEAT_AMU
#if __aarch64__
void amu_enable(cpu_context_t *ctx);
void amu_init_el3(void);
void amu_init_el3(unsigned int core_pos);
void amu_init_el2_unused(void);
void amu_enable_per_world(per_world_context_t *per_world_ctx);
#else
@ -29,7 +29,7 @@ void amu_enable(bool el2_unused);
void amu_enable(cpu_context_t *ctx)
{
}
void amu_init_el3(void)
void amu_init_el3(unsigned int core_pos)
{
}
void amu_init_el2_unused(void)
@ -45,28 +45,57 @@ static inline void amu_enable(bool el2_unused)
#endif /*__aarch64__ */
#endif /* ENABLE_FEAT_AMU */
/*
* Per-core list of the counters to be enabled. Value will be written into
* AMCNTENSET1_EL0 verbatim.
*/
#if ENABLE_AMU_AUXILIARY_COUNTERS
/*
* AMU data for a single core.
*/
struct amu_core {
uint16_t enable; /* Mask of auxiliary counters to enable */
};
extern uint16_t plat_amu_aux_enables[PLATFORM_CORE_COUNT];
#endif
/*
* Topological platform data specific to the AMU.
*/
struct amu_topology {
struct amu_core cores[PLATFORM_CORE_COUNT]; /* Per-core data */
};
#define CTX_AMU_GRP0_ALL U(4)
#define CTX_AMU_GRP1_ALL U(16)
#if !ENABLE_AMU_FCONF
/*
* Retrieve the platform's AMU topology. A `NULL` return value is treated as a
* non-fatal error, in which case no auxiliary counters will be enabled.
*/
const struct amu_topology *plat_amu_topology(void);
#endif /* ENABLE_AMU_FCONF */
#endif /* ENABLE_AMU_AUXILIARY_COUNTERS */
typedef struct amu_regs {
u_register_t grp0[CTX_AMU_GRP0_ALL];
#if ENABLE_AMU_AUXILIARY_COUNTERS
u_register_t grp1[CTX_AMU_GRP1_ALL];
#endif
} amu_regs_t;
static inline u_register_t read_amu_grp0_ctx_reg(amu_regs_t *ctx, size_t index)
{
return ctx->grp0[index];
}
static inline void write_amu_grp0_ctx_reg(amu_regs_t *ctx, size_t index, u_register_t val)
{
ctx->grp0[index] = val;
}
static inline uint16_t get_amu_aux_enables(size_t index)
{
#if ENABLE_AMU_AUXILIARY_COUNTERS
return plat_amu_aux_enables[index];
#else
return 0;
#endif
}
static inline u_register_t read_amu_grp1_ctx_reg(amu_regs_t *ctx, size_t index)
{
#if ENABLE_AMU_AUXILIARY_COUNTERS
return ctx->grp1[index];
#else
return 0;
#endif
}
static inline void write_amu_grp1_ctx_reg(amu_regs_t *ctx, size_t index, u_register_t val)
{
#if ENABLE_AMU_AUXILIARY_COUNTERS
ctx->grp1[index] = val;
#endif
}
#endif /* AMU_H */

View file

@ -1,20 +0,0 @@
/*
* Copyright (c) 2021, Arm Limited. All rights reserved.
*
* SPDX-License-Identifier: BSD-3-Clause
*/
#ifndef FCONF_AMU_GETTER_H
#define FCONF_AMU_GETTER_H
#include <lib/extensions/amu.h>
#define amu__config_getter(id) fconf_amu_config.id
struct fconf_amu_config {
const struct amu_topology *topology;
};
extern struct fconf_amu_config fconf_amu_config;
#endif /* FCONF_AMU_GETTER_H */

View file

@ -661,13 +661,13 @@ void cm_setup_context(cpu_context_t *ctx, const entry_point_info_t *ep)
/*******************************************************************************
* Enable architecture extensions for EL3 execution. This function only updates
* registers in-place which are expected to either never change or be
* overwritten by el3_exit.
* overwritten by el3_exit. Expects the core_pos of the current core as argument.
******************************************************************************/
#if IMAGE_BL31
void cm_manage_extensions_el3(void)
void cm_manage_extensions_el3(unsigned int my_idx)
{
if (is_feat_amu_supported()) {
amu_init_el3();
amu_init_el3(my_idx);
}
if (is_feat_sme_supported()) {
@ -803,6 +803,7 @@ static void manage_extensions_secure_per_world(void)
static void manage_extensions_nonsecure(cpu_context_t *ctx)
{
#if IMAGE_BL31
/* NOTE: registers are not context switched */
if (is_feat_amu_supported()) {
amu_enable(ctx);
}

View file

@ -1,5 +1,5 @@
/*
* Copyright (c) 2017-2021, Arm Limited and Contributors. All rights reserved.
* Copyright (c) 2017-2025, Arm Limited and Contributors. All rights reserved.
*
* SPDX-License-Identifier: BSD-3-Clause
*/
@ -8,7 +8,6 @@
#include <cdefs.h>
#include <stdbool.h>
#include "../amu_private.h"
#include <arch.h>
#include <arch_features.h>
#include <arch_helpers.h>
@ -18,51 +17,7 @@
#include <plat/common/platform.h>
struct amu_ctx {
uint64_t group0_cnts[AMU_GROUP0_MAX_COUNTERS];
#if ENABLE_AMU_AUXILIARY_COUNTERS
uint64_t group1_cnts[AMU_GROUP1_MAX_COUNTERS];
#endif
uint16_t group0_enable;
#if ENABLE_AMU_AUXILIARY_COUNTERS
uint16_t group1_enable;
#endif
};
static struct amu_ctx amu_ctxs_[PLATFORM_CORE_COUNT];
CASSERT((sizeof(amu_ctxs_[0].group0_enable) * CHAR_BIT) <= AMU_GROUP0_MAX_COUNTERS,
amu_ctx_group0_enable_cannot_represent_all_group0_counters);
#if ENABLE_AMU_AUXILIARY_COUNTERS
CASSERT((sizeof(amu_ctxs_[0].group1_enable) * CHAR_BIT) <= AMU_GROUP1_MAX_COUNTERS,
amu_ctx_group1_enable_cannot_represent_all_group1_counters);
#endif
static inline __unused void write_hcptr_tam(uint32_t value)
{
write_hcptr((read_hcptr() & ~TAM_BIT) |
((value << TAM_SHIFT) & TAM_BIT));
}
static inline __unused void write_amcr_cg1rz(uint32_t value)
{
write_amcr((read_amcr() & ~AMCR_CG1RZ_BIT) |
((value << AMCR_CG1RZ_SHIFT) & AMCR_CG1RZ_BIT));
}
static inline __unused uint32_t read_amcfgr_ncg(void)
{
return (read_amcfgr() >> AMCFGR_NCG_SHIFT) &
AMCFGR_NCG_MASK;
}
static inline __unused uint32_t read_amcgcr_cg0nc(void)
{
return (read_amcgcr() >> AMCGCR_CG0NC_SHIFT) &
AMCGCR_CG0NC_MASK;
}
amu_regs_t amu_ctx[PLATFORM_CORE_COUNT];
static inline __unused uint32_t read_amcgcr_cg1nc(void)
{
@ -70,134 +25,31 @@ static inline __unused uint32_t read_amcgcr_cg1nc(void)
AMCGCR_CG1NC_MASK;
}
static inline __unused uint32_t read_amcntenset0_px(void)
{
return (read_amcntenset0() >> AMCNTENSET0_Pn_SHIFT) &
AMCNTENSET0_Pn_MASK;
}
static inline __unused uint32_t read_amcntenset1_px(void)
{
return (read_amcntenset1() >> AMCNTENSET1_Pn_SHIFT) &
AMCNTENSET1_Pn_MASK;
}
static inline __unused void write_amcntenset0_px(uint32_t px)
{
uint32_t value = read_amcntenset0();
value &= ~AMCNTENSET0_Pn_MASK;
value |= (px << AMCNTENSET0_Pn_SHIFT) &
AMCNTENSET0_Pn_MASK;
write_amcntenset0(value);
}
static inline __unused void write_amcntenset1_px(uint32_t px)
{
uint32_t value = read_amcntenset1();
value &= ~AMCNTENSET1_Pn_MASK;
value |= (px << AMCNTENSET1_Pn_SHIFT) &
AMCNTENSET1_Pn_MASK;
write_amcntenset1(value);
}
static inline __unused void write_amcntenclr0_px(uint32_t px)
{
uint32_t value = read_amcntenclr0();
value &= ~AMCNTENCLR0_Pn_MASK;
value |= (px << AMCNTENCLR0_Pn_SHIFT) & AMCNTENCLR0_Pn_MASK;
write_amcntenclr0(value);
}
static inline __unused void write_amcntenclr1_px(uint32_t px)
{
uint32_t value = read_amcntenclr1();
value &= ~AMCNTENCLR1_Pn_MASK;
value |= (px << AMCNTENCLR1_Pn_SHIFT) & AMCNTENCLR1_Pn_MASK;
write_amcntenclr1(value);
}
#if ENABLE_AMU_AUXILIARY_COUNTERS
static __unused bool amu_group1_supported(void)
{
return read_amcfgr_ncg() > 0U;
}
#endif
/*
* Enable counters. This function is meant to be invoked by the context
* management library before exiting from EL3.
*/
void amu_enable(bool el2_unused)
{
uint32_t amcfgr_ncg; /* Number of counter groups */
uint32_t amcgcr_cg0nc; /* Number of group 0 counters */
uint32_t amcntenset0_px = 0x0; /* Group 0 enable mask */
uint32_t amcntenset1_px = 0x0; /* Group 1 enable mask */
if (el2_unused) {
/*
* HCPTR.TAM: Set to zero so any accesses to the Activity
* Monitor registers do not trap to EL2.
*/
write_hcptr_tam(0U);
write_hcptr(read_hcptr() & ~TAM_BIT);
}
/*
* Retrieve the number of architected counters. All of these counters
* are enabled by default.
*/
/* Architecture is currently pinned to 4 */
assert((read_amcgcr() & AMCGCR_CG0NC_MASK) == CTX_AMU_GRP0_ALL);
amcgcr_cg0nc = read_amcgcr_cg0nc();
amcntenset0_px = (UINT32_C(1) << (amcgcr_cg0nc)) - 1U;
assert(amcgcr_cg0nc <= AMU_AMCGCR_CG0NC_MAX);
/*
* The platform may opt to enable specific auxiliary counters. This can
* be done via the common FCONF getter, or via the platform-implemented
* function.
*/
#if ENABLE_AMU_AUXILIARY_COUNTERS
const struct amu_topology *topology;
#if ENABLE_AMU_FCONF
topology = FCONF_GET_PROPERTY(amu, config, topology);
#else
topology = plat_amu_topology();
#endif /* ENABLE_AMU_FCONF */
if (topology != NULL) {
/* Enable all architected counters by default */
write_amcntenset0(AMCNTENSET0_Pn_MASK);
if (is_feat_amu_aux_supported()) {
unsigned int core_pos = plat_my_core_pos();
amcntenset1_el0_px = topology->cores[core_pos].enable;
} else {
ERROR("AMU: failed to generate AMU topology\n");
}
#endif /* ENABLE_AMU_AUXILIARY_COUNTERS */
/*
* Enable the requested counters.
*/
write_amcntenset0_px(amcntenset0_px);
amcfgr_ncg = read_amcfgr_ncg();
if (amcfgr_ncg > 0U) {
write_amcntenset1_px(amcntenset1_px);
#if !ENABLE_AMU_AUXILIARY_COUNTERS
VERBOSE("AMU: auxiliary counters detected but support is disabled\n");
#endif
/* Something went wrong if we're trying to write higher bits */
assert((get_amu_aux_enables(core_pos) & ~AMCNTENSET1_Pn_MASK) == 0);
write_amcntenset1(get_amu_aux_enables(core_pos));
}
/* Bail out if FEAT_AMUv1p1 features are not present. */
@ -214,180 +66,177 @@ void amu_enable(bool el2_unused)
* mapped view are unaffected.
*/
VERBOSE("AMU group 1 counter access restricted.\n");
write_amcr_cg1rz(1U);
write_amcr(read_amcr() | 1U);
#else
write_amcr_cg1rz(0U);
write_amcr(0);
#endif
}
/* Read the group 0 counter identified by the given `idx`. */
static uint64_t amu_group0_cnt_read(unsigned int idx)
{
assert(is_feat_amu_supported());
assert(idx < read_amcgcr_cg0nc());
return amu_group0_cnt_read_internal(idx);
}
/* Write the group 0 counter identified by the given `idx` with `val` */
static void amu_group0_cnt_write(unsigned int idx, uint64_t val)
{
assert(is_feat_amu_supported());
assert(idx < read_amcgcr_cg0nc());
amu_group0_cnt_write_internal(idx, val);
isb();
}
#if ENABLE_AMU_AUXILIARY_COUNTERS
/* Read the group 1 counter identified by the given `idx` */
static uint64_t amu_group1_cnt_read(unsigned int idx)
{
assert(is_feat_amu_supported());
assert(amu_group1_supported());
assert(idx < read_amcgcr_cg1nc());
return amu_group1_cnt_read_internal(idx);
}
/* Write the group 1 counter identified by the given `idx` with `val` */
static void amu_group1_cnt_write(unsigned int idx, uint64_t val)
{
assert(is_feat_amu_supported());
assert(amu_group1_supported());
assert(idx < read_amcgcr_cg1nc());
amu_group1_cnt_write_internal(idx, val);
isb();
}
#endif
static void *amu_context_save(const void *arg)
{
uint32_t i;
unsigned int core_pos;
struct amu_ctx *ctx;
uint32_t amcgcr_cg0nc; /* Number of group 0 counters */
#if ENABLE_AMU_AUXILIARY_COUNTERS
uint32_t amcfgr_ncg; /* Number of counter groups */
uint32_t amcgcr_cg1nc; /* Number of group 1 counters */
#endif
if (!is_feat_amu_supported()) {
return (void *)0;
}
core_pos = plat_my_core_pos();
ctx = &amu_ctxs_[core_pos];
unsigned int core_pos = *(unsigned int *)arg;
amu_regs_t *ctx = &amu_ctx[core_pos];
amcgcr_cg0nc = read_amcgcr_cg0nc();
#if ENABLE_AMU_AUXILIARY_COUNTERS
amcfgr_ncg = read_amcfgr_ncg();
amcgcr_cg1nc = (amcfgr_ncg > 0U) ? read_amcgcr_cg1nc() : 0U;
#endif
/*
* Disable all AMU counters.
*/
ctx->group0_enable = read_amcntenset0_px();
write_amcntenclr0_px(ctx->group0_enable);
#if ENABLE_AMU_AUXILIARY_COUNTERS
if (amcfgr_ncg > 0U) {
ctx->group1_enable = read_amcntenset1_px();
write_amcntenclr1_px(ctx->group1_enable);
/* Disable all counters so we can write to them safely later */
write_amcntenclr0(AMCNTENCLR0_Pn_MASK);
if (is_feat_amu_aux_supported()) {
write_amcntenclr1(get_amu_aux_enables(core_pos));
}
#endif
/*
* Save the counters to the local context.
*/
isb(); /* Ensure counters have been stopped */
for (i = 0U; i < amcgcr_cg0nc; i++) {
ctx->group0_cnts[i] = amu_group0_cnt_read(i);
}
write_amu_grp0_ctx_reg(ctx, 0, read64_amevcntr00());
write_amu_grp0_ctx_reg(ctx, 1, read64_amevcntr01());
write_amu_grp0_ctx_reg(ctx, 2, read64_amevcntr02());
write_amu_grp0_ctx_reg(ctx, 3, read64_amevcntr03());
#if ENABLE_AMU_AUXILIARY_COUNTERS
for (i = 0U; i < amcgcr_cg1nc; i++) {
ctx->group1_cnts[i] = amu_group1_cnt_read(i);
if (is_feat_amu_aux_supported()) {
uint8_t num_counters = read_amcgcr_cg1nc();
switch (num_counters) {
case 0x10:
write_amu_grp1_ctx_reg(ctx, 0xf, read64_amevcntr1f());
__fallthrough;
case 0x0f:
write_amu_grp1_ctx_reg(ctx, 0xe, read64_amevcntr1e());
__fallthrough;
case 0x0e:
write_amu_grp1_ctx_reg(ctx, 0xd, read64_amevcntr1d());
__fallthrough;
case 0x0d:
write_amu_grp1_ctx_reg(ctx, 0xc, read64_amevcntr1c());
__fallthrough;
case 0x0c:
write_amu_grp1_ctx_reg(ctx, 0xb, read64_amevcntr1b());
__fallthrough;
case 0x0b:
write_amu_grp1_ctx_reg(ctx, 0xa, read64_amevcntr1a());
__fallthrough;
case 0x0a:
write_amu_grp1_ctx_reg(ctx, 0x9, read64_amevcntr19());
__fallthrough;
case 0x09:
write_amu_grp1_ctx_reg(ctx, 0x8, read64_amevcntr18());
__fallthrough;
case 0x08:
write_amu_grp1_ctx_reg(ctx, 0x7, read64_amevcntr17());
__fallthrough;
case 0x07:
write_amu_grp1_ctx_reg(ctx, 0x6, read64_amevcntr16());
__fallthrough;
case 0x06:
write_amu_grp1_ctx_reg(ctx, 0x5, read64_amevcntr15());
__fallthrough;
case 0x05:
write_amu_grp1_ctx_reg(ctx, 0x4, read64_amevcntr14());
__fallthrough;
case 0x04:
write_amu_grp1_ctx_reg(ctx, 0x3, read64_amevcntr13());
__fallthrough;
case 0x03:
write_amu_grp1_ctx_reg(ctx, 0x2, read64_amevcntr12());
__fallthrough;
case 0x02:
write_amu_grp1_ctx_reg(ctx, 0x1, read64_amevcntr11());
__fallthrough;
case 0x01:
write_amu_grp1_ctx_reg(ctx, 0x0, read64_amevcntr10());
__fallthrough;
case 0x00:
break;
default:
assert(0); /* something is wrong */
}
}
#endif
return (void *)0;
}
static void *amu_context_restore(const void *arg)
{
uint32_t i;
unsigned int core_pos;
struct amu_ctx *ctx;
uint32_t amcfgr_ncg; /* Number of counter groups */
uint32_t amcgcr_cg0nc; /* Number of group 0 counters */
#if ENABLE_AMU_AUXILIARY_COUNTERS
uint32_t amcgcr_cg1nc; /* Number of group 1 counters */
#endif
if (!is_feat_amu_supported()) {
return (void *)0;
}
core_pos = plat_my_core_pos();
ctx = &amu_ctxs_[core_pos];
unsigned int core_pos = *(unsigned int *)arg;
amu_regs_t *ctx = &amu_ctx[core_pos];
amcfgr_ncg = read_amcfgr_ncg();
amcgcr_cg0nc = read_amcgcr_cg0nc();
write64_amevcntr00(read_amu_grp0_ctx_reg(ctx, 0));
write64_amevcntr01(read_amu_grp0_ctx_reg(ctx, 1));
write64_amevcntr02(read_amu_grp0_ctx_reg(ctx, 2));
write64_amevcntr03(read_amu_grp0_ctx_reg(ctx, 3));
#if ENABLE_AMU_AUXILIARY_COUNTERS
amcgcr_cg1nc = (amcfgr_ncg > 0U) ? read_amcgcr_cg1nc() : 0U;
#endif
if (is_feat_amu_aux_supported()) {
uint8_t num_counters = read_amcgcr_cg1nc();
/*
* Sanity check that all counters were disabled when the context was
* previously saved.
*/
assert(read_amcntenset0_px() == 0U);
if (amcfgr_ncg > 0U) {
assert(read_amcntenset1_px() == 0U);
switch (num_counters) {
case 0x10:
write64_amevcntr1f(read_amu_grp1_ctx_reg(ctx, 0xf));
__fallthrough;
case 0x0f:
write64_amevcntr1e(read_amu_grp1_ctx_reg(ctx, 0xe));
__fallthrough;
case 0x0e:
write64_amevcntr1d(read_amu_grp1_ctx_reg(ctx, 0xd));
__fallthrough;
case 0x0d:
write64_amevcntr1c(read_amu_grp1_ctx_reg(ctx, 0xc));
__fallthrough;
case 0x0c:
write64_amevcntr1b(read_amu_grp1_ctx_reg(ctx, 0xb));
__fallthrough;
case 0x0b:
write64_amevcntr1a(read_amu_grp1_ctx_reg(ctx, 0xa));
__fallthrough;
case 0x0a:
write64_amevcntr19(read_amu_grp1_ctx_reg(ctx, 0x9));
__fallthrough;
case 0x09:
write64_amevcntr18(read_amu_grp1_ctx_reg(ctx, 0x8));
__fallthrough;
case 0x08:
write64_amevcntr17(read_amu_grp1_ctx_reg(ctx, 0x7));
__fallthrough;
case 0x07:
write64_amevcntr16(read_amu_grp1_ctx_reg(ctx, 0x6));
__fallthrough;
case 0x06:
write64_amevcntr15(read_amu_grp1_ctx_reg(ctx, 0x5));
__fallthrough;
case 0x05:
write64_amevcntr14(read_amu_grp1_ctx_reg(ctx, 0x4));
__fallthrough;
case 0x04:
write64_amevcntr13(read_amu_grp1_ctx_reg(ctx, 0x3));
__fallthrough;
case 0x03:
write64_amevcntr12(read_amu_grp1_ctx_reg(ctx, 0x2));
__fallthrough;
case 0x02:
write64_amevcntr11(read_amu_grp1_ctx_reg(ctx, 0x1));
__fallthrough;
case 0x01:
write64_amevcntr10(read_amu_grp1_ctx_reg(ctx, 0x0));
__fallthrough;
case 0x00:
break;
default:
assert(0); /* something is wrong */
}
}
/*
* Restore the counter values from the local context.
*/
for (i = 0U; i < amcgcr_cg0nc; i++) {
amu_group0_cnt_write(i, ctx->group0_cnts[i]);
/* now enable them again */
write_amcntenset0(AMCNTENSET0_Pn_MASK);
if (is_feat_amu_aux_supported()) {
write_amcntenset1(get_amu_aux_enables(core_pos));
}
#if ENABLE_AMU_AUXILIARY_COUNTERS
for (i = 0U; i < amcgcr_cg1nc; i++) {
amu_group1_cnt_write(i, ctx->group1_cnts[i]);
}
#endif
/*
* Re-enable counters that were disabled during context save.
*/
write_amcntenset0_px(ctx->group0_enable);
#if ENABLE_AMU_AUXILIARY_COUNTERS
if (amcfgr_ncg > 0U) {
write_amcntenset1_px(ctx->group1_enable);
}
#endif
isb();
return (void *)0;
}

View file

@ -1,271 +0,0 @@
/*
* Copyright (c) 2021, Arm Limited and Contributors. All rights reserved.
*
* SPDX-License-Identifier: BSD-3-Clause
*/
#include <arch.h>
#include <assert_macros.S>
#include <asm_macros.S>
.globl amu_group0_cnt_read_internal
.globl amu_group0_cnt_write_internal
.globl amu_group1_cnt_read_internal
.globl amu_group1_cnt_write_internal
.globl amu_group1_set_evtype_internal
/*
* uint64_t amu_group0_cnt_read_internal(int idx);
*
* Given `idx`, read the corresponding AMU counter
* and return it in `r0` and `r1`.
*/
func amu_group0_cnt_read_internal
#if ENABLE_ASSERTIONS
/* `idx` should be between [0, 3] */
mov r1, r0
lsr r1, r1, #2
cmp r1, #0
ASM_ASSERT(eq)
#endif
/*
* Given `idx` calculate address of ldcopr16/bx lr instruction pair
* in the table below.
*/
adr r1, 1f
lsl r0, r0, #3 /* each ldcopr16/bx lr sequence is 8 bytes */
add r1, r1, r0
bx r1
1:
ldcopr16 r0, r1, AMEVCNTR00 /* index 0 */
bx lr
ldcopr16 r0, r1, AMEVCNTR01 /* index 1 */
bx lr
ldcopr16 r0, r1, AMEVCNTR02 /* index 2 */
bx lr
ldcopr16 r0, r1, AMEVCNTR03 /* index 3 */
bx lr
endfunc amu_group0_cnt_read_internal
/*
* void amu_group0_cnt_write_internal(int idx, uint64_t val);
*
* Given `idx`, write `val` to the corresponding AMU counter.
* `idx` is passed in `r0` and `val` is passed in `r2` and `r3`.
* `r1` is used as a scratch register.
*/
func amu_group0_cnt_write_internal
#if ENABLE_ASSERTIONS
/* `idx` should be between [0, 3] */
mov r1, r0
lsr r1, r1, #2
cmp r1, #0
ASM_ASSERT(eq)
#endif
/*
* Given `idx` calculate address of stcopr16/bx lr instruction pair
* in the table below.
*/
adr r1, 1f
lsl r0, r0, #3 /* each stcopr16/bx lr sequence is 8 bytes */
add r1, r1, r0
bx r1
1:
stcopr16 r2, r3, AMEVCNTR00 /* index 0 */
bx lr
stcopr16 r2, r3, AMEVCNTR01 /* index 1 */
bx lr
stcopr16 r2, r3, AMEVCNTR02 /* index 2 */
bx lr
stcopr16 r2, r3, AMEVCNTR03 /* index 3 */
bx lr
endfunc amu_group0_cnt_write_internal
#if ENABLE_AMU_AUXILIARY_COUNTERS
/*
* uint64_t amu_group1_cnt_read_internal(int idx);
*
* Given `idx`, read the corresponding AMU counter
* and return it in `r0` and `r1`.
*/
func amu_group1_cnt_read_internal
#if ENABLE_ASSERTIONS
/* `idx` should be between [0, 15] */
mov r1, r0
lsr r1, r1, #4
cmp r1, #0
ASM_ASSERT(eq)
#endif
/*
* Given `idx` calculate address of ldcopr16/bx lr instruction pair
* in the table below.
*/
adr r1, 1f
lsl r0, r0, #3 /* each ldcopr16/bx lr sequence is 8 bytes */
add r1, r1, r0
bx r1
1:
ldcopr16 r0, r1, AMEVCNTR10 /* index 0 */
bx lr
ldcopr16 r0, r1, AMEVCNTR11 /* index 1 */
bx lr
ldcopr16 r0, r1, AMEVCNTR12 /* index 2 */
bx lr
ldcopr16 r0, r1, AMEVCNTR13 /* index 3 */
bx lr
ldcopr16 r0, r1, AMEVCNTR14 /* index 4 */
bx lr
ldcopr16 r0, r1, AMEVCNTR15 /* index 5 */
bx lr
ldcopr16 r0, r1, AMEVCNTR16 /* index 6 */
bx lr
ldcopr16 r0, r1, AMEVCNTR17 /* index 7 */
bx lr
ldcopr16 r0, r1, AMEVCNTR18 /* index 8 */
bx lr
ldcopr16 r0, r1, AMEVCNTR19 /* index 9 */
bx lr
ldcopr16 r0, r1, AMEVCNTR1A /* index 10 */
bx lr
ldcopr16 r0, r1, AMEVCNTR1B /* index 11 */
bx lr
ldcopr16 r0, r1, AMEVCNTR1C /* index 12 */
bx lr
ldcopr16 r0, r1, AMEVCNTR1D /* index 13 */
bx lr
ldcopr16 r0, r1, AMEVCNTR1E /* index 14 */
bx lr
ldcopr16 r0, r1, AMEVCNTR1F /* index 15 */
bx lr
endfunc amu_group1_cnt_read_internal
/*
* void amu_group1_cnt_write_internal(int idx, uint64_t val);
*
* Given `idx`, write `val` to the corresponding AMU counter.
* `idx` is passed in `r0` and `val` is passed in `r2` and `r3`.
* `r1` is used as a scratch register.
*/
func amu_group1_cnt_write_internal
#if ENABLE_ASSERTIONS
/* `idx` should be between [0, 15] */
mov r1, r0
lsr r1, r1, #4
cmp r1, #0
ASM_ASSERT(eq)
#endif
/*
* Given `idx` calculate address of ldcopr16/bx lr instruction pair
* in the table below.
*/
adr r1, 1f
lsl r0, r0, #3 /* each stcopr16/bx lr sequence is 8 bytes */
add r1, r1, r0
bx r1
1:
stcopr16 r2, r3, AMEVCNTR10 /* index 0 */
bx lr
stcopr16 r2, r3, AMEVCNTR11 /* index 1 */
bx lr
stcopr16 r2, r3, AMEVCNTR12 /* index 2 */
bx lr
stcopr16 r2, r3, AMEVCNTR13 /* index 3 */
bx lr
stcopr16 r2, r3, AMEVCNTR14 /* index 4 */
bx lr
stcopr16 r2, r3, AMEVCNTR15 /* index 5 */
bx lr
stcopr16 r2, r3, AMEVCNTR16 /* index 6 */
bx lr
stcopr16 r2, r3, AMEVCNTR17 /* index 7 */
bx lr
stcopr16 r2, r3, AMEVCNTR18 /* index 8 */
bx lr
stcopr16 r2, r3, AMEVCNTR19 /* index 9 */
bx lr
stcopr16 r2, r3, AMEVCNTR1A /* index 10 */
bx lr
stcopr16 r2, r3, AMEVCNTR1B /* index 11 */
bx lr
stcopr16 r2, r3, AMEVCNTR1C /* index 12 */
bx lr
stcopr16 r2, r3, AMEVCNTR1D /* index 13 */
bx lr
stcopr16 r2, r3, AMEVCNTR1E /* index 14 */
bx lr
stcopr16 r2, r3, AMEVCNTR1F /* index 15 */
bx lr
endfunc amu_group1_cnt_write_internal
/*
* void amu_group1_set_evtype_internal(int idx, unsigned int val);
*
* Program the AMU event type register indexed by `idx`
* with the value `val`.
*/
func amu_group1_set_evtype_internal
#if ENABLE_ASSERTIONS
/* `idx` should be between [0, 15] */
mov r2, r0
lsr r2, r2, #4
cmp r2, #0
ASM_ASSERT(eq)
/* val should be between [0, 65535] */
mov r2, r1
lsr r2, r2, #16
cmp r2, #0
ASM_ASSERT(eq)
#endif
/*
* Given `idx` calculate address of stcopr/bx lr instruction pair
* in the table below.
*/
adr r2, 1f
lsl r0, r0, #3 /* each stcopr/bx lr sequence is 8 bytes */
add r2, r2, r0
bx r2
1:
stcopr r1, AMEVTYPER10 /* index 0 */
bx lr
stcopr r1, AMEVTYPER11 /* index 1 */
bx lr
stcopr r1, AMEVTYPER12 /* index 2 */
bx lr
stcopr r1, AMEVTYPER13 /* index 3 */
bx lr
stcopr r1, AMEVTYPER14 /* index 4 */
bx lr
stcopr r1, AMEVTYPER15 /* index 5 */
bx lr
stcopr r1, AMEVTYPER16 /* index 6 */
bx lr
stcopr r1, AMEVTYPER17 /* index 7 */
bx lr
stcopr r1, AMEVTYPER18 /* index 8 */
bx lr
stcopr r1, AMEVTYPER19 /* index 9 */
bx lr
stcopr r1, AMEVTYPER1A /* index 10 */
bx lr
stcopr r1, AMEVTYPER1B /* index 11 */
bx lr
stcopr r1, AMEVTYPER1C /* index 12 */
bx lr
stcopr r1, AMEVTYPER1D /* index 13 */
bx lr
stcopr r1, AMEVTYPER1E /* index 14 */
bx lr
stcopr r1, AMEVTYPER1F /* index 15 */
bx lr
endfunc amu_group1_set_evtype_internal
#endif

View file

@ -1,5 +1,5 @@
/*
* Copyright (c) 2017-2023, Arm Limited and Contributors. All rights reserved.
* Copyright (c) 2017-2025, Arm Limited and Contributors. All rights reserved.
*
* SPDX-License-Identifier: BSD-3-Clause
*/
@ -10,179 +10,37 @@
#include <stdbool.h>
#include <stdint.h>
#include "../amu_private.h"
#include <arch.h>
#include <arch_features.h>
#include <arch_helpers.h>
#include <common/debug.h>
#include <lib/el3_runtime/pubsub_events.h>
#include <lib/extensions/amu.h>
#include <lib/utils_def.h>
#include <platform_def.h>
#include <plat/common/platform.h>
amu_regs_t amu_ctx[PLATFORM_CORE_COUNT];
#if ENABLE_AMU_FCONF
# include <lib/fconf/fconf.h>
# include <lib/fconf/fconf_amu_getter.h>
#endif
struct amu_ctx {
uint64_t group0_cnts[AMU_GROUP0_MAX_COUNTERS];
#if ENABLE_AMU_AUXILIARY_COUNTERS
uint64_t group1_cnts[AMU_GROUP1_MAX_COUNTERS];
#endif
/* Architected event counter 1 does not have an offset register */
uint64_t group0_voffsets[AMU_GROUP0_MAX_COUNTERS - 1U];
#if ENABLE_AMU_AUXILIARY_COUNTERS
uint64_t group1_voffsets[AMU_GROUP1_MAX_COUNTERS];
#endif
uint16_t group0_enable;
#if ENABLE_AMU_AUXILIARY_COUNTERS
uint16_t group1_enable;
#endif
};
static struct amu_ctx amu_ctxs_[PLATFORM_CORE_COUNT];
CASSERT((sizeof(amu_ctxs_[0].group0_enable) * CHAR_BIT) <= AMU_GROUP0_MAX_COUNTERS,
amu_ctx_group0_enable_cannot_represent_all_group0_counters);
#if ENABLE_AMU_AUXILIARY_COUNTERS
CASSERT((sizeof(amu_ctxs_[0].group1_enable) * CHAR_BIT) <= AMU_GROUP1_MAX_COUNTERS,
amu_ctx_group1_enable_cannot_represent_all_group1_counters);
#endif
static inline __unused uint64_t read_hcr_el2_amvoffen(void)
{
return (read_hcr_el2() & HCR_AMVOFFEN_BIT) >>
HCR_AMVOFFEN_SHIFT;
}
static inline __unused void write_cptr_el2_tam(uint64_t value)
{
write_cptr_el2((read_cptr_el2() & ~CPTR_EL2_TAM_BIT) |
((value << CPTR_EL2_TAM_SHIFT) & CPTR_EL2_TAM_BIT));
}
static inline __unused void ctx_write_scr_el3_amvoffen(cpu_context_t *ctx, uint64_t amvoffen)
{
uint64_t value = read_ctx_reg(get_el3state_ctx(ctx), CTX_SCR_EL3);
value &= ~SCR_AMVOFFEN_BIT;
value |= (amvoffen << SCR_AMVOFFEN_SHIFT) & SCR_AMVOFFEN_BIT;
write_ctx_reg(get_el3state_ctx(ctx), CTX_SCR_EL3, value);
}
static inline __unused void write_hcr_el2_amvoffen(uint64_t value)
{
write_hcr_el2((read_hcr_el2() & ~HCR_AMVOFFEN_BIT) |
((value << HCR_AMVOFFEN_SHIFT) & HCR_AMVOFFEN_BIT));
}
static inline __unused void write_amcr_el0_cg1rz(uint64_t value)
{
write_amcr_el0((read_amcr_el0() & ~AMCR_CG1RZ_BIT) |
((value << AMCR_CG1RZ_SHIFT) & AMCR_CG1RZ_BIT));
}
static inline __unused uint64_t read_amcfgr_el0_ncg(void)
{
return (read_amcfgr_el0() >> AMCFGR_EL0_NCG_SHIFT) &
AMCFGR_EL0_NCG_MASK;
}
static inline __unused uint64_t read_amcgcr_el0_cg0nc(void)
{
return (read_amcgcr_el0() >> AMCGCR_EL0_CG0NC_SHIFT) &
AMCGCR_EL0_CG0NC_MASK;
}
static inline __unused uint64_t read_amcg1idr_el0_voff(void)
{
return (read_amcg1idr_el0() >> AMCG1IDR_VOFF_SHIFT) &
AMCG1IDR_VOFF_MASK;
}
static inline __unused uint64_t read_amcgcr_el0_cg1nc(void)
static inline uint8_t read_amcgcr_el0_cg1nc(void)
{
return (read_amcgcr_el0() >> AMCGCR_EL0_CG1NC_SHIFT) &
AMCGCR_EL0_CG1NC_MASK;
}
static inline __unused uint64_t read_amcntenset0_el0_px(void)
{
return (read_amcntenset0_el0() >> AMCNTENSET0_EL0_Pn_SHIFT) &
AMCNTENSET0_EL0_Pn_MASK;
}
static inline __unused uint64_t read_amcntenset1_el0_px(void)
{
return (read_amcntenset1_el0() >> AMCNTENSET1_EL0_Pn_SHIFT) &
AMCNTENSET1_EL0_Pn_MASK;
}
static inline __unused void write_amcntenset0_el0_px(uint64_t px)
{
uint64_t value = read_amcntenset0_el0();
value &= ~AMCNTENSET0_EL0_Pn_MASK;
value |= (px << AMCNTENSET0_EL0_Pn_SHIFT) & AMCNTENSET0_EL0_Pn_MASK;
write_amcntenset0_el0(value);
}
static inline __unused void write_amcntenset1_el0_px(uint64_t px)
{
uint64_t value = read_amcntenset1_el0();
value &= ~AMCNTENSET1_EL0_Pn_MASK;
value |= (px << AMCNTENSET1_EL0_Pn_SHIFT) & AMCNTENSET1_EL0_Pn_MASK;
write_amcntenset1_el0(value);
}
static inline __unused void write_amcntenclr0_el0_px(uint64_t px)
{
uint64_t value = read_amcntenclr0_el0();
value &= ~AMCNTENCLR0_EL0_Pn_MASK;
value |= (px << AMCNTENCLR0_EL0_Pn_SHIFT) & AMCNTENCLR0_EL0_Pn_MASK;
write_amcntenclr0_el0(value);
}
static inline __unused void write_amcntenclr1_el0_px(uint64_t px)
{
uint64_t value = read_amcntenclr1_el0();
value &= ~AMCNTENCLR1_EL0_Pn_MASK;
value |= (px << AMCNTENCLR1_EL0_Pn_SHIFT) & AMCNTENCLR1_EL0_Pn_MASK;
write_amcntenclr1_el0(value);
}
#if ENABLE_AMU_AUXILIARY_COUNTERS
static __unused bool amu_group1_supported(void)
{
return read_amcfgr_el0_ncg() > 0U;
}
#endif
/*
* Enable counters. This function is meant to be invoked by the context
* management library before exiting from EL3.
*/
void amu_enable(cpu_context_t *ctx)
{
/* Initialize FEAT_AMUv1p1 features if present. */
if (is_feat_amuv1p1_supported()) {
el3_state_t *state = get_el3state_ctx(ctx);
u_register_t reg;
/*
* Set SCR_EL3.AMVOFFEN to one so that accesses to virtual
* offset registers at EL2 do not trap to EL3
*/
ctx_write_scr_el3_amvoffen(ctx, 1U);
reg = read_ctx_reg(state, CTX_SCR_EL3);
reg |= SCR_AMVOFFEN_BIT;
write_ctx_reg(state, CTX_SCR_EL3, reg);
}
}
@ -198,47 +56,19 @@ void amu_enable_per_world(per_world_context_t *per_world_ctx)
per_world_ctx->ctx_cptr_el3 = cptr_el3;
}
void amu_init_el3(void)
void amu_init_el3(unsigned int core_pos)
{
uint64_t group0_impl_ctr = read_amcgcr_el0_cg0nc();
uint64_t group0_en_mask = (1 << (group0_impl_ctr)) - 1U;
uint64_t num_ctr_groups = read_amcfgr_el0_ncg();
/* architecture is currently pinned to 4 */
assert((read_amcgcr_el0() & AMCGCR_EL0_CG0NC_MASK) == CTX_AMU_GRP0_ALL);
/* Enable all architected counters by default */
write_amcntenset0_el0_px(group0_en_mask);
#if ENABLE_AMU_AUXILIARY_COUNTERS
if (num_ctr_groups > 0U) {
uint64_t amcntenset1_el0_px = 0x0; /* Group 1 enable mask */
const struct amu_topology *topology;
/*
* The platform may opt to enable specific auxiliary counters.
* This can be done via the common FCONF getter, or via the
* platform-implemented function.
*/
#if ENABLE_AMU_FCONF
topology = FCONF_GET_PROPERTY(amu, config, topology);
#else
topology = plat_amu_topology();
#endif /* ENABLE_AMU_FCONF */
if (topology != NULL) {
unsigned int core_pos = plat_my_core_pos();
amcntenset1_el0_px = topology->cores[core_pos].enable;
} else {
ERROR("AMU: failed to generate AMU topology\n");
write_amcntenset0_el0(AMCNTENSET0_EL0_Pn_MASK);
if (is_feat_amu_aux_supported()) {
/* something went wrong if we're trying to write higher bits */
assert((get_amu_aux_enables(core_pos) & ~AMCNTENSET1_EL0_Pn_MASK) == 0);
write_amcntenset1_el0(get_amu_aux_enables(core_pos));
}
write_amcntenset1_el0_px(amcntenset1_el0_px);
}
#else /* ENABLE_AMU_AUXILIARY_COUNTERS */
if (num_ctr_groups > 0U) {
VERBOSE("AMU: auxiliary counters detected but support is disabled\n");
}
#endif /* ENABLE_AMU_AUXILIARY_COUNTERS */
if (is_feat_amuv1p1_supported()) {
#if AMU_RESTRICT_COUNTERS
/*
@ -249,9 +79,10 @@ void amu_init_el3(void)
* zero. Reads from the memory mapped view are unaffected.
*/
VERBOSE("AMU group 1 counter access restricted.\n");
write_amcr_el0_cg1rz(1U);
write_amcr_el0(AMCR_CG1RZ_BIT);
#else
write_amcr_el0_cg1rz(0U);
/* HDBG = 0 in both cases */
write_amcr_el0(0);
#endif
}
}
@ -262,230 +93,93 @@ void amu_init_el2_unused(void)
* CPTR_EL2.TAM: Set to zero so any accesses to the Activity Monitor
* registers do not trap to EL2.
*/
write_cptr_el2_tam(0U);
write_cptr_el2(read_cptr_el2() & ~CPTR_EL2_TAM_BIT);
/* Initialize FEAT_AMUv1p1 features if present. */
if (is_feat_amuv1p1_supported()) {
/* Make sure virtual offsets are disabled if EL2 not used. */
write_hcr_el2_amvoffen(0U);
/* Make sure virtual offsets are disabled */
write_hcr_el2(read_hcr_el2() & ~HCR_AMVOFFEN_BIT);
}
}
/* Read the group 0 counter identified by the given `idx`. */
static uint64_t amu_group0_cnt_read(unsigned int idx)
{
assert(is_feat_amu_supported());
assert(idx < read_amcgcr_el0_cg0nc());
return amu_group0_cnt_read_internal(idx);
}
/* Write the group 0 counter identified by the given `idx` with `val` */
static void amu_group0_cnt_write(unsigned int idx, uint64_t val)
{
assert(is_feat_amu_supported());
assert(idx < read_amcgcr_el0_cg0nc());
amu_group0_cnt_write_internal(idx, val);
isb();
}
/*
* Unlike with auxiliary counters, we cannot detect at runtime whether an
* architected counter supports a virtual offset. These are instead fixed
* according to FEAT_AMUv1p1, but this switch will need to be updated if later
* revisions of FEAT_AMU add additional architected counters.
*/
static bool amu_group0_voffset_supported(uint64_t idx)
{
switch (idx) {
case 0U:
case 2U:
case 3U:
return true;
case 1U:
return false;
default:
ERROR("AMU: can't set up virtual offset for unknown "
"architected counter %" PRIu64 "!\n", idx);
panic();
}
}
/*
* Read the group 0 offset register for a given index. Index must be 0, 2,
* or 3, the register for 1 does not exist.
*
* Using this function requires FEAT_AMUv1p1 support.
*/
static uint64_t amu_group0_voffset_read(unsigned int idx)
{
assert(is_feat_amuv1p1_supported());
assert(idx < read_amcgcr_el0_cg0nc());
assert(idx != 1U);
return amu_group0_voffset_read_internal(idx);
}
/*
* Write the group 0 offset register for a given index. Index must be 0, 2, or
* 3, the register for 1 does not exist.
*
* Using this function requires FEAT_AMUv1p1 support.
*/
static void amu_group0_voffset_write(unsigned int idx, uint64_t val)
{
assert(is_feat_amuv1p1_supported());
assert(idx < read_amcgcr_el0_cg0nc());
assert(idx != 1U);
amu_group0_voffset_write_internal(idx, val);
isb();
}
#if ENABLE_AMU_AUXILIARY_COUNTERS
/* Read the group 1 counter identified by the given `idx` */
static uint64_t amu_group1_cnt_read(unsigned int idx)
{
assert(is_feat_amu_supported());
assert(amu_group1_supported());
assert(idx < read_amcgcr_el0_cg1nc());
return amu_group1_cnt_read_internal(idx);
}
/* Write the group 1 counter identified by the given `idx` with `val` */
static void amu_group1_cnt_write(unsigned int idx, uint64_t val)
{
assert(is_feat_amu_supported());
assert(amu_group1_supported());
assert(idx < read_amcgcr_el0_cg1nc());
amu_group1_cnt_write_internal(idx, val);
isb();
}
/*
* Read the group 1 offset register for a given index.
*
* Using this function requires FEAT_AMUv1p1 support.
*/
static uint64_t amu_group1_voffset_read(unsigned int idx)
{
assert(is_feat_amuv1p1_supported());
assert(amu_group1_supported());
assert(idx < read_amcgcr_el0_cg1nc());
assert((read_amcg1idr_el0_voff() & (UINT64_C(1) << idx)) != 0U);
return amu_group1_voffset_read_internal(idx);
}
/*
* Write the group 1 offset register for a given index.
*
* Using this function requires FEAT_AMUv1p1 support.
*/
static void amu_group1_voffset_write(unsigned int idx, uint64_t val)
{
assert(is_feat_amuv1p1_supported());
assert(amu_group1_supported());
assert(idx < read_amcgcr_el0_cg1nc());
assert((read_amcg1idr_el0_voff() & (UINT64_C(1) << idx)) != 0U);
amu_group1_voffset_write_internal(idx, val);
isb();
}
#endif
static void *amu_context_save(const void *arg)
{
uint64_t i, j;
unsigned int core_pos;
struct amu_ctx *ctx;
uint64_t hcr_el2_amvoffen = 0; /* AMU virtual offsets enabled */
uint64_t amcgcr_el0_cg0nc; /* Number of group 0 counters */
#if ENABLE_AMU_AUXILIARY_COUNTERS
uint64_t amcg1idr_el0_voff; /* Auxiliary counters with virtual offsets */
uint64_t amcfgr_el0_ncg; /* Number of counter groups */
uint64_t amcgcr_el0_cg1nc; /* Number of group 1 counters */
#endif
if (!is_feat_amu_supported()) {
return (void *)0;
}
core_pos = plat_my_core_pos();
ctx = &amu_ctxs_[core_pos];
unsigned int core_pos = *(unsigned int *)arg;
amu_regs_t *ctx = &amu_ctx[core_pos];
amcgcr_el0_cg0nc = read_amcgcr_el0_cg0nc();
if (is_feat_amuv1p1_supported()) {
hcr_el2_amvoffen = read_hcr_el2_amvoffen();
/* disable all counters so we can write them safely later */
write_amcntenclr0_el0(AMCNTENCLR0_EL0_Pn_MASK);
if (is_feat_amu_aux_supported()) {
write_amcntenclr1_el0(get_amu_aux_enables(core_pos));
}
#if ENABLE_AMU_AUXILIARY_COUNTERS
amcfgr_el0_ncg = read_amcfgr_el0_ncg();
amcgcr_el0_cg1nc = (amcfgr_el0_ncg > 0U) ? read_amcgcr_el0_cg1nc() : 0U;
amcg1idr_el0_voff = (hcr_el2_amvoffen != 0U) ? read_amcg1idr_el0_voff() : 0U;
#endif
isb();
/*
* Disable all AMU counters.
*/
write_amu_grp0_ctx_reg(ctx, 0, read_amevcntr00_el0());
write_amu_grp0_ctx_reg(ctx, 1, read_amevcntr01_el0());
write_amu_grp0_ctx_reg(ctx, 2, read_amevcntr02_el0());
write_amu_grp0_ctx_reg(ctx, 3, read_amevcntr03_el0());
ctx->group0_enable = read_amcntenset0_el0_px();
write_amcntenclr0_el0_px(ctx->group0_enable);
if (is_feat_amu_aux_supported()) {
uint8_t num_counters = read_amcgcr_el0_cg1nc();
#if ENABLE_AMU_AUXILIARY_COUNTERS
if (amcfgr_el0_ncg > 0U) {
ctx->group1_enable = read_amcntenset1_el0_px();
write_amcntenclr1_el0_px(ctx->group1_enable);
switch (num_counters) {
case 0x10:
write_amu_grp1_ctx_reg(ctx, 0xf, read_amevcntr1f_el0());
__fallthrough;
case 0x0f:
write_amu_grp1_ctx_reg(ctx, 0xe, read_amevcntr1e_el0());
__fallthrough;
case 0x0e:
write_amu_grp1_ctx_reg(ctx, 0xd, read_amevcntr1d_el0());
__fallthrough;
case 0x0d:
write_amu_grp1_ctx_reg(ctx, 0xc, read_amevcntr1c_el0());
__fallthrough;
case 0x0c:
write_amu_grp1_ctx_reg(ctx, 0xb, read_amevcntr1b_el0());
__fallthrough;
case 0x0b:
write_amu_grp1_ctx_reg(ctx, 0xa, read_amevcntr1a_el0());
__fallthrough;
case 0x0a:
write_amu_grp1_ctx_reg(ctx, 0x9, read_amevcntr19_el0());
__fallthrough;
case 0x09:
write_amu_grp1_ctx_reg(ctx, 0x8, read_amevcntr18_el0());
__fallthrough;
case 0x08:
write_amu_grp1_ctx_reg(ctx, 0x7, read_amevcntr17_el0());
__fallthrough;
case 0x07:
write_amu_grp1_ctx_reg(ctx, 0x6, read_amevcntr16_el0());
__fallthrough;
case 0x06:
write_amu_grp1_ctx_reg(ctx, 0x5, read_amevcntr15_el0());
__fallthrough;
case 0x05:
write_amu_grp1_ctx_reg(ctx, 0x4, read_amevcntr14_el0());
__fallthrough;
case 0x04:
write_amu_grp1_ctx_reg(ctx, 0x3, read_amevcntr13_el0());
__fallthrough;
case 0x03:
write_amu_grp1_ctx_reg(ctx, 0x2, read_amevcntr12_el0());
__fallthrough;
case 0x02:
write_amu_grp1_ctx_reg(ctx, 0x1, read_amevcntr11_el0());
__fallthrough;
case 0x01:
write_amu_grp1_ctx_reg(ctx, 0x0, read_amevcntr10_el0());
__fallthrough;
case 0x00:
break;
default:
assert(0); /* something is wrong */
}
#endif
/*
* Save the counters to the local context.
*/
isb(); /* Ensure counters have been stopped */
for (i = 0U; i < amcgcr_el0_cg0nc; i++) {
ctx->group0_cnts[i] = amu_group0_cnt_read(i);
}
#if ENABLE_AMU_AUXILIARY_COUNTERS
for (i = 0U; i < amcgcr_el0_cg1nc; i++) {
ctx->group1_cnts[i] = amu_group1_cnt_read(i);
}
#endif
/*
* Save virtual offsets for counters that offer them.
*/
if (hcr_el2_amvoffen != 0U) {
for (i = 0U, j = 0U; i < amcgcr_el0_cg0nc; i++) {
if (!amu_group0_voffset_supported(i)) {
continue; /* No virtual offset */
}
ctx->group0_voffsets[j++] = amu_group0_voffset_read(i);
}
#if ENABLE_AMU_AUXILIARY_COUNTERS
for (i = 0U, j = 0U; i < amcgcr_el0_cg1nc; i++) {
if ((amcg1idr_el0_voff >> i) & 1U) {
continue; /* No virtual offset */
}
ctx->group1_voffsets[j++] = amu_group1_voffset_read(i);
}
#endif
}
return (void *)0;
@ -493,94 +187,85 @@ static void *amu_context_save(const void *arg)
static void *amu_context_restore(const void *arg)
{
uint64_t i, j;
unsigned int core_pos;
struct amu_ctx *ctx;
uint64_t hcr_el2_amvoffen = 0; /* AMU virtual offsets enabled */
uint64_t amcgcr_el0_cg0nc; /* Number of group 0 counters */
#if ENABLE_AMU_AUXILIARY_COUNTERS
uint64_t amcfgr_el0_ncg; /* Number of counter groups */
uint64_t amcgcr_el0_cg1nc; /* Number of group 1 counters */
uint64_t amcg1idr_el0_voff; /* Auxiliary counters with virtual offsets */
#endif
if (!is_feat_amu_supported()) {
return (void *)0;
}
core_pos = plat_my_core_pos();
ctx = &amu_ctxs_[core_pos];
unsigned int core_pos = *(unsigned int *)arg;
amu_regs_t *ctx = &amu_ctx[core_pos];
amcgcr_el0_cg0nc = read_amcgcr_el0_cg0nc();
write_amevcntr00_el0(read_amu_grp0_ctx_reg(ctx, 0));
write_amevcntr01_el0(read_amu_grp0_ctx_reg(ctx, 1));
write_amevcntr02_el0(read_amu_grp0_ctx_reg(ctx, 2));
write_amevcntr03_el0(read_amu_grp0_ctx_reg(ctx, 3));
if (is_feat_amuv1p1_supported()) {
hcr_el2_amvoffen = read_hcr_el2_amvoffen();
if (is_feat_amu_aux_supported()) {
uint8_t num_counters = read_amcgcr_el0_cg1nc();
switch (num_counters) {
case 0x10:
write_amevcntr1f_el0(read_amu_grp1_ctx_reg(ctx, 0xf));
__fallthrough;
case 0x0f:
write_amevcntr1e_el0(read_amu_grp1_ctx_reg(ctx, 0xe));
__fallthrough;
case 0x0e:
write_amevcntr1d_el0(read_amu_grp1_ctx_reg(ctx, 0xd));
__fallthrough;
case 0x0d:
write_amevcntr1c_el0(read_amu_grp1_ctx_reg(ctx, 0xc));
__fallthrough;
case 0x0c:
write_amevcntr1b_el0(read_amu_grp1_ctx_reg(ctx, 0xb));
__fallthrough;
case 0x0b:
write_amevcntr1a_el0(read_amu_grp1_ctx_reg(ctx, 0xa));
__fallthrough;
case 0x0a:
write_amevcntr19_el0(read_amu_grp1_ctx_reg(ctx, 0x9));
__fallthrough;
case 0x09:
write_amevcntr18_el0(read_amu_grp1_ctx_reg(ctx, 0x8));
__fallthrough;
case 0x08:
write_amevcntr17_el0(read_amu_grp1_ctx_reg(ctx, 0x7));
__fallthrough;
case 0x07:
write_amevcntr16_el0(read_amu_grp1_ctx_reg(ctx, 0x6));
__fallthrough;
case 0x06:
write_amevcntr15_el0(read_amu_grp1_ctx_reg(ctx, 0x5));
__fallthrough;
case 0x05:
write_amevcntr14_el0(read_amu_grp1_ctx_reg(ctx, 0x4));
__fallthrough;
case 0x04:
write_amevcntr13_el0(read_amu_grp1_ctx_reg(ctx, 0x3));
__fallthrough;
case 0x03:
write_amevcntr12_el0(read_amu_grp1_ctx_reg(ctx, 0x2));
__fallthrough;
case 0x02:
write_amevcntr11_el0(read_amu_grp1_ctx_reg(ctx, 0x1));
__fallthrough;
case 0x01:
write_amevcntr10_el0(read_amu_grp1_ctx_reg(ctx, 0x0));
__fallthrough;
case 0x00:
break;
default:
assert(0); /* something is wrong */
}
}
#if ENABLE_AMU_AUXILIARY_COUNTERS
amcfgr_el0_ncg = read_amcfgr_el0_ncg();
amcgcr_el0_cg1nc = (amcfgr_el0_ncg > 0U) ? read_amcgcr_el0_cg1nc() : 0U;
amcg1idr_el0_voff = (hcr_el2_amvoffen != 0U) ? read_amcg1idr_el0_voff() : 0U;
#endif
/*
* Restore the counter values from the local context.
*/
for (i = 0U; i < amcgcr_el0_cg0nc; i++) {
amu_group0_cnt_write(i, ctx->group0_cnts[i]);
/* now enable them again */
write_amcntenset0_el0(AMCNTENSET0_EL0_Pn_MASK);
if (is_feat_amu_aux_supported()) {
write_amcntenset1_el0(get_amu_aux_enables(core_pos));
}
#if ENABLE_AMU_AUXILIARY_COUNTERS
for (i = 0U; i < amcgcr_el0_cg1nc; i++) {
amu_group1_cnt_write(i, ctx->group1_cnts[i]);
}
#endif
/*
* Restore virtual offsets for counters that offer them.
*/
if (hcr_el2_amvoffen != 0U) {
for (i = 0U, j = 0U; i < amcgcr_el0_cg0nc; i++) {
if (!amu_group0_voffset_supported(i)) {
continue; /* No virtual offset */
}
amu_group0_voffset_write(i, ctx->group0_voffsets[j++]);
}
#if ENABLE_AMU_AUXILIARY_COUNTERS
for (i = 0U, j = 0U; i < amcgcr_el0_cg1nc; i++) {
if ((amcg1idr_el0_voff >> i) & 1U) {
continue; /* No virtual offset */
}
amu_group1_voffset_write(i, ctx->group1_voffsets[j++]);
}
#endif
}
/*
* Re-enable counters that were disabled during context save.
*/
write_amcntenset0_el0_px(ctx->group0_enable);
#if ENABLE_AMU_AUXILIARY_COUNTERS
if (amcfgr_el0_ncg > 0) {
write_amcntenset1_el0_px(ctx->group1_enable);
}
#endif
#if ENABLE_MPMM
mpmm_enable();
#endif
isb();
return (void *)0;
}

View file

@ -1,389 +0,0 @@
/*
* Copyright (c) 2017-2021, Arm Limited and Contributors. All rights reserved.
*
* SPDX-License-Identifier: BSD-3-Clause
*/
#include <arch.h>
#include <assert_macros.S>
#include <asm_macros.S>
.globl amu_group0_cnt_read_internal
.globl amu_group0_cnt_write_internal
.globl amu_group1_cnt_read_internal
.globl amu_group1_cnt_write_internal
.globl amu_group1_set_evtype_internal
/* FEAT_AMUv1p1 virtualisation offset register functions */
.globl amu_group0_voffset_read_internal
.globl amu_group0_voffset_write_internal
.globl amu_group1_voffset_read_internal
.globl amu_group1_voffset_write_internal
/*
* uint64_t amu_group0_cnt_read_internal(int idx);
*
* Given `idx`, read the corresponding AMU counter
* and return it in `x0`.
*/
func amu_group0_cnt_read_internal
adr x1, 1f
#if ENABLE_ASSERTIONS
/*
* It can be dangerous to call this function with an
* out of bounds index. Ensure `idx` is valid.
*/
tst x0, #~3
ASM_ASSERT(eq)
#endif
/*
* Given `idx` calculate address of mrs/ret instruction pair
* in the table below.
*/
add x1, x1, x0, lsl #3 /* each mrs/ret sequence is 8 bytes */
#if ENABLE_BTI
add x1, x1, x0, lsl #2 /* + "bti j" instruction */
#endif
br x1
1: read AMEVCNTR00_EL0 /* index 0 */
read AMEVCNTR01_EL0 /* index 1 */
read AMEVCNTR02_EL0 /* index 2 */
read AMEVCNTR03_EL0 /* index 3 */
endfunc amu_group0_cnt_read_internal
/*
* void amu_group0_cnt_write_internal(int idx, uint64_t val);
*
* Given `idx`, write `val` to the corresponding AMU counter.
*/
func amu_group0_cnt_write_internal
adr x2, 1f
#if ENABLE_ASSERTIONS
/*
* It can be dangerous to call this function with an
* out of bounds index. Ensure `idx` is valid.
*/
tst x0, #~3
ASM_ASSERT(eq)
#endif
/*
* Given `idx` calculate address of mrs/ret instruction pair
* in the table below.
*/
add x2, x2, x0, lsl #3 /* each msr/ret sequence is 8 bytes */
#if ENABLE_BTI
add x2, x2, x0, lsl #2 /* + "bti j" instruction */
#endif
br x2
1: write AMEVCNTR00_EL0 /* index 0 */
write AMEVCNTR01_EL0 /* index 1 */
write AMEVCNTR02_EL0 /* index 2 */
write AMEVCNTR03_EL0 /* index 3 */
endfunc amu_group0_cnt_write_internal
#if ENABLE_AMU_AUXILIARY_COUNTERS
/*
* uint64_t amu_group1_cnt_read_internal(int idx);
*
* Given `idx`, read the corresponding AMU counter
* and return it in `x0`.
*/
func amu_group1_cnt_read_internal
adr x1, 1f
#if ENABLE_ASSERTIONS
/*
* It can be dangerous to call this function with an
* out of bounds index. Ensure `idx` is valid.
*/
tst x0, #~0xF
ASM_ASSERT(eq)
#endif
/*
* Given `idx` calculate address of mrs/ret instruction pair
* in the table below.
*/
add x1, x1, x0, lsl #3 /* each mrs/ret sequence is 8 bytes */
#if ENABLE_BTI
add x1, x1, x0, lsl #2 /* + "bti j" instruction */
#endif
br x1
1: read AMEVCNTR10_EL0 /* index 0 */
read AMEVCNTR11_EL0 /* index 1 */
read AMEVCNTR12_EL0 /* index 2 */
read AMEVCNTR13_EL0 /* index 3 */
read AMEVCNTR14_EL0 /* index 4 */
read AMEVCNTR15_EL0 /* index 5 */
read AMEVCNTR16_EL0 /* index 6 */
read AMEVCNTR17_EL0 /* index 7 */
read AMEVCNTR18_EL0 /* index 8 */
read AMEVCNTR19_EL0 /* index 9 */
read AMEVCNTR1A_EL0 /* index 10 */
read AMEVCNTR1B_EL0 /* index 11 */
read AMEVCNTR1C_EL0 /* index 12 */
read AMEVCNTR1D_EL0 /* index 13 */
read AMEVCNTR1E_EL0 /* index 14 */
read AMEVCNTR1F_EL0 /* index 15 */
endfunc amu_group1_cnt_read_internal
/*
* void amu_group1_cnt_write_internal(int idx, uint64_t val);
*
* Given `idx`, write `val` to the corresponding AMU counter.
*/
func amu_group1_cnt_write_internal
adr x2, 1f
#if ENABLE_ASSERTIONS
/*
* It can be dangerous to call this function with an
* out of bounds index. Ensure `idx` is valid.
*/
tst x0, #~0xF
ASM_ASSERT(eq)
#endif
/*
* Given `idx` calculate address of mrs/ret instruction pair
* in the table below.
*/
add x2, x2, x0, lsl #3 /* each msr/ret sequence is 8 bytes */
#if ENABLE_BTI
add x2, x2, x0, lsl #2 /* + "bti j" instruction */
#endif
br x2
1: write AMEVCNTR10_EL0 /* index 0 */
write AMEVCNTR11_EL0 /* index 1 */
write AMEVCNTR12_EL0 /* index 2 */
write AMEVCNTR13_EL0 /* index 3 */
write AMEVCNTR14_EL0 /* index 4 */
write AMEVCNTR15_EL0 /* index 5 */
write AMEVCNTR16_EL0 /* index 6 */
write AMEVCNTR17_EL0 /* index 7 */
write AMEVCNTR18_EL0 /* index 8 */
write AMEVCNTR19_EL0 /* index 9 */
write AMEVCNTR1A_EL0 /* index 10 */
write AMEVCNTR1B_EL0 /* index 11 */
write AMEVCNTR1C_EL0 /* index 12 */
write AMEVCNTR1D_EL0 /* index 13 */
write AMEVCNTR1E_EL0 /* index 14 */
write AMEVCNTR1F_EL0 /* index 15 */
endfunc amu_group1_cnt_write_internal
/*
* void amu_group1_set_evtype_internal(int idx, unsigned int val);
*
* Program the AMU event type register indexed by `idx`
* with the value `val`.
*/
func amu_group1_set_evtype_internal
adr x2, 1f
#if ENABLE_ASSERTIONS
/*
* It can be dangerous to call this function with an
* out of bounds index. Ensure `idx` is valid.
*/
tst x0, #~0xF
ASM_ASSERT(eq)
/* val should be between [0, 65535] */
tst x1, #~0xFFFF
ASM_ASSERT(eq)
#endif
/*
* Given `idx` calculate address of msr/ret instruction pair
* in the table below.
*/
add x2, x2, x0, lsl #3 /* each msr/ret sequence is 8 bytes */
#if ENABLE_BTI
add x2, x2, x0, lsl #2 /* + "bti j" instruction */
#endif
br x2
1: write AMEVTYPER10_EL0 /* index 0 */
write AMEVTYPER11_EL0 /* index 1 */
write AMEVTYPER12_EL0 /* index 2 */
write AMEVTYPER13_EL0 /* index 3 */
write AMEVTYPER14_EL0 /* index 4 */
write AMEVTYPER15_EL0 /* index 5 */
write AMEVTYPER16_EL0 /* index 6 */
write AMEVTYPER17_EL0 /* index 7 */
write AMEVTYPER18_EL0 /* index 8 */
write AMEVTYPER19_EL0 /* index 9 */
write AMEVTYPER1A_EL0 /* index 10 */
write AMEVTYPER1B_EL0 /* index 11 */
write AMEVTYPER1C_EL0 /* index 12 */
write AMEVTYPER1D_EL0 /* index 13 */
write AMEVTYPER1E_EL0 /* index 14 */
write AMEVTYPER1F_EL0 /* index 15 */
endfunc amu_group1_set_evtype_internal
#endif
/*
* Accessor functions for virtual offset registers added with FEAT_AMUv1p1
*/
/*
* uint64_t amu_group0_voffset_read_internal(int idx);
*
* Given `idx`, read the corresponding AMU virtual offset register
* and return it in `x0`.
*/
func amu_group0_voffset_read_internal
adr x1, 1f
#if ENABLE_ASSERTIONS
/*
* It can be dangerous to call this function with an
* out of bounds index. Ensure `idx` is valid.
*/
tst x0, #~3
ASM_ASSERT(eq)
/* Make sure idx != 1 since AMEVCNTVOFF01_EL2 does not exist */
cmp x0, #1
ASM_ASSERT(ne)
#endif
/*
* Given `idx` calculate address of mrs/ret instruction pair
* in the table below.
*/
add x1, x1, x0, lsl #3 /* each mrs/ret sequence is 8 bytes */
#if ENABLE_BTI
add x1, x1, x0, lsl #2 /* + "bti j" instruction */
#endif
br x1
1: read AMEVCNTVOFF00_EL2 /* index 0 */
.skip 8 /* AMEVCNTVOFF01_EL2 does not exist */
#if ENABLE_BTI
.skip 4
#endif
read AMEVCNTVOFF02_EL2 /* index 2 */
read AMEVCNTVOFF03_EL2 /* index 3 */
endfunc amu_group0_voffset_read_internal
/*
* void amu_group0_voffset_write_internal(int idx, uint64_t val);
*
* Given `idx`, write `val` to the corresponding AMU virtual offset register.
*/
func amu_group0_voffset_write_internal
adr x2, 1f
#if ENABLE_ASSERTIONS
/*
* It can be dangerous to call this function with an
* out of bounds index. Ensure `idx` is valid.
*/
tst x0, #~3
ASM_ASSERT(eq)
/* Make sure idx != 1 since AMEVCNTVOFF01_EL2 does not exist */
cmp x0, #1
ASM_ASSERT(ne)
#endif
/*
* Given `idx` calculate address of mrs/ret instruction pair
* in the table below.
*/
add x2, x2, x0, lsl #3 /* each msr/ret sequence is 8 bytes */
#if ENABLE_BTI
add x2, x2, x0, lsl #2 /* + "bti j" instruction */
#endif
br x2
1: write AMEVCNTVOFF00_EL2 /* index 0 */
.skip 8 /* AMEVCNTVOFF01_EL2 does not exist */
#if ENABLE_BTI
.skip 4
#endif
write AMEVCNTVOFF02_EL2 /* index 2 */
write AMEVCNTVOFF03_EL2 /* index 3 */
endfunc amu_group0_voffset_write_internal
#if ENABLE_AMU_AUXILIARY_COUNTERS
/*
* uint64_t amu_group1_voffset_read_internal(int idx);
*
* Given `idx`, read the corresponding AMU virtual offset register
* and return it in `x0`.
*/
func amu_group1_voffset_read_internal
adr x1, 1f
#if ENABLE_ASSERTIONS
/*
* It can be dangerous to call this function with an
* out of bounds index. Ensure `idx` is valid.
*/
tst x0, #~0xF
ASM_ASSERT(eq)
#endif
/*
* Given `idx` calculate address of mrs/ret instruction pair
* in the table below.
*/
add x1, x1, x0, lsl #3 /* each mrs/ret sequence is 8 bytes */
#if ENABLE_BTI
add x1, x1, x0, lsl #2 /* + "bti j" instruction */
#endif
br x1
1: read AMEVCNTVOFF10_EL2 /* index 0 */
read AMEVCNTVOFF11_EL2 /* index 1 */
read AMEVCNTVOFF12_EL2 /* index 2 */
read AMEVCNTVOFF13_EL2 /* index 3 */
read AMEVCNTVOFF14_EL2 /* index 4 */
read AMEVCNTVOFF15_EL2 /* index 5 */
read AMEVCNTVOFF16_EL2 /* index 6 */
read AMEVCNTVOFF17_EL2 /* index 7 */
read AMEVCNTVOFF18_EL2 /* index 8 */
read AMEVCNTVOFF19_EL2 /* index 9 */
read AMEVCNTVOFF1A_EL2 /* index 10 */
read AMEVCNTVOFF1B_EL2 /* index 11 */
read AMEVCNTVOFF1C_EL2 /* index 12 */
read AMEVCNTVOFF1D_EL2 /* index 13 */
read AMEVCNTVOFF1E_EL2 /* index 14 */
read AMEVCNTVOFF1F_EL2 /* index 15 */
endfunc amu_group1_voffset_read_internal
/*
* void amu_group1_voffset_write_internal(int idx, uint64_t val);
*
* Given `idx`, write `val` to the corresponding AMU virtual offset register.
*/
func amu_group1_voffset_write_internal
adr x2, 1f
#if ENABLE_ASSERTIONS
/*
* It can be dangerous to call this function with an
* out of bounds index. Ensure `idx` is valid.
*/
tst x0, #~0xF
ASM_ASSERT(eq)
#endif
/*
* Given `idx` calculate address of mrs/ret instruction pair
* in the table below.
*/
add x2, x2, x0, lsl #3 /* each msr/ret sequence is 8 bytes */
#if ENABLE_BTI
add x2, x2, x0, lsl #2 /* + "bti j" instruction */
#endif
br x2
1: write AMEVCNTVOFF10_EL2 /* index 0 */
write AMEVCNTVOFF11_EL2 /* index 1 */
write AMEVCNTVOFF12_EL2 /* index 2 */
write AMEVCNTVOFF13_EL2 /* index 3 */
write AMEVCNTVOFF14_EL2 /* index 4 */
write AMEVCNTVOFF15_EL2 /* index 5 */
write AMEVCNTVOFF16_EL2 /* index 6 */
write AMEVCNTVOFF17_EL2 /* index 7 */
write AMEVCNTVOFF18_EL2 /* index 8 */
write AMEVCNTVOFF19_EL2 /* index 9 */
write AMEVCNTVOFF1A_EL2 /* index 10 */
write AMEVCNTVOFF1B_EL2 /* index 11 */
write AMEVCNTVOFF1C_EL2 /* index 12 */
write AMEVCNTVOFF1D_EL2 /* index 13 */
write AMEVCNTVOFF1E_EL2 /* index 14 */
write AMEVCNTVOFF1F_EL2 /* index 15 */
endfunc amu_group1_voffset_write_internal
#endif

View file

@ -1,24 +1,13 @@
#
# Copyright (c) 2021, Arm Limited. All rights reserved.
# Copyright (c) 2021-2025, Arm Limited. All rights reserved.
#
# SPDX-License-Identifier: BSD-3-Clause
#
include lib/fconf/fconf.mk
AMU_SOURCES := lib/extensions/amu/${ARCH}/amu.c \
lib/extensions/amu/${ARCH}/amu_helpers.S
AMU_SOURCES := lib/extensions/amu/${ARCH}/amu.c
ifneq (${ENABLE_AMU_AUXILIARY_COUNTERS},0)
ifeq (${ENABLE_FEAT_AMU},0)
$(error AMU auxiliary counter support (`ENABLE_AMU_AUXILIARY_COUNTERS`) requires AMU support (`ENABLE_FEAT_AMU`))
$(error "ENABLE_AMU_AUXILIARY_COUNTERS requires ENABLE_FEAT_AMU")
endif
endif
ifneq (${ENABLE_AMU_FCONF},0)
ifeq (${ENABLE_AMU_AUXILIARY_COUNTERS},0)
$(error AMU FCONF support (`ENABLE_AMU_FCONF`) is not necessary when auxiliary counter support (`ENABLE_AMU_AUXILIARY_COUNTERS`) is disabled)
endif
AMU_SOURCES += ${FCONF_AMU_SOURCES}
endif

View file

@ -1,38 +0,0 @@
/*
* Copyright (c) 2017-2021, Arm Limited and Contributors. All rights reserved.
*
* SPDX-License-Identifier: BSD-3-Clause
*/
#ifndef AMU_PRIVATE_H
#define AMU_PRIVATE_H
#include <stdint.h>
#include <lib/cassert.h>
#include <lib/extensions/amu.h>
#include <lib/utils_def.h>
#include <platform_def.h>
#define AMU_GROUP0_MAX_COUNTERS U(16)
#define AMU_GROUP1_MAX_COUNTERS U(16)
#define AMU_AMCGCR_CG0NC_MAX U(16)
uint64_t amu_group0_cnt_read_internal(unsigned int idx);
void amu_group0_cnt_write_internal(unsigned int idx, uint64_t val);
uint64_t amu_group1_cnt_read_internal(unsigned int idx);
void amu_group1_cnt_write_internal(unsigned int idx, uint64_t val);
void amu_group1_set_evtype_internal(unsigned int idx, unsigned int val);
#if __aarch64__
uint64_t amu_group0_voffset_read_internal(unsigned int idx);
void amu_group0_voffset_write_internal(unsigned int idx, uint64_t val);
uint64_t amu_group1_voffset_read_internal(unsigned int idx);
void amu_group1_voffset_write_internal(unsigned int idx, uint64_t val);
#endif
#endif /* AMU_PRIVATE_H */

View file

@ -11,6 +11,3 @@ FCONF_SOURCES += ${FDT_WRAPPERS_SOURCES}
FCONF_DYN_SOURCES := lib/fconf/fconf_dyn_cfg_getter.c
FCONF_DYN_SOURCES += ${FDT_WRAPPERS_SOURCES}
FCONF_AMU_SOURCES := lib/fconf/fconf_amu_getter.c
FCONF_AMU_SOURCES += ${FDT_WRAPPERS_SOURCES}

View file

@ -1,142 +0,0 @@
/*
* Copyright (c) 2021, Arm Limited. All rights reserved.
*
* SPDX-License-Identifier: BSD-3-Clause
*/
#include <stddef.h>
#include <stdint.h>
#include <common/debug.h>
#include <common/fdt_wrappers.h>
#include <lib/fconf/fconf.h>
#include <lib/fconf/fconf_amu_getter.h>
#include <libfdt.h>
#include <plat/common/platform.h>
struct fconf_amu_config fconf_amu_config;
static struct amu_topology fconf_amu_topology_;
/*
* Populate the core-specific AMU structure with information retrieved from a
* device tree.
*
* Returns `0` on success, or a negative integer representing an error code.
*/
static int fconf_populate_amu_cpu_amu(const void *fdt, int parent,
struct amu_core *amu)
{
int ret = 0;
int node = 0;
fdt_for_each_subnode(node, fdt, parent) {
const char *name;
const char *value;
int len;
uintptr_t idx = 0U;
name = fdt_get_name(fdt, node, &len);
if (strncmp(name, "counter@", 8) != 0) {
continue;
}
ret = fdt_get_reg_props_by_index(fdt, node, 0, &idx, NULL);
if (ret < 0) {
break;
}
value = fdt_getprop(fdt, node, "enable-at-el3", &len);
if ((value == NULL) && (len != -FDT_ERR_NOTFOUND)) {
break;
}
if (len != -FDT_ERR_NOTFOUND) {
amu->enable |= (1 << idx);
}
}
if ((node < 0) && (node != -FDT_ERR_NOTFOUND)) {
return node;
}
return ret;
}
/*
* Within a `cpu` node, attempt to dereference the `amu` property, and populate
* the AMU information for the core.
*
* Returns `0` on success, or a negative integer representing an error code.
*/
static int fconf_populate_amu_cpu(const void *fdt, int node, uintptr_t mpidr)
{
int ret;
int idx;
uint32_t amu_phandle;
struct amu_core *amu;
ret = fdt_read_uint32(fdt, node, "amu", &amu_phandle);
if (ret < 0) {
if (ret == -FDT_ERR_NOTFOUND) {
ret = 0;
}
return ret;
}
node = fdt_node_offset_by_phandle(fdt, amu_phandle);
if (node < 0) {
return node;
}
idx = plat_core_pos_by_mpidr(mpidr);
if (idx < 0) {
return -FDT_ERR_BADVALUE;
}
amu = &fconf_amu_topology_.cores[idx];
return fconf_populate_amu_cpu_amu(fdt, node, amu);
}
/*
* Populates the global `amu_topology` structure based on what's described by
* the hardware configuration device tree blob.
*
* The device tree is expected to provide an `amu` property for each `cpu` node,
* like so:
*
* cpu@0 {
* amu = <&cpu0_amu>;
* };
*
* amus {
* cpu0_amu: amu-0 {
* counters {
* #address-cells = <2>;
* #size-cells = <0>;
*
* counter@x,y {
* reg = <x y>; // Group x, counter y
* };
* };
* };
* };
*/
static int fconf_populate_amu(uintptr_t config)
{
int ret = fdtw_for_each_cpu(
(const void *)config, fconf_populate_amu_cpu);
if (ret == 0) {
fconf_amu_config.topology = &fconf_amu_topology_;
} else {
ERROR("FCONF: failed to parse AMU information: %d\n", ret);
}
return ret;
}
FCONF_REGISTER_POPULATOR(HW_CONFIG, amu, fconf_populate_amu);

View file

@ -972,7 +972,7 @@ void psci_warmboot_entrypoint(void)
psci_power_state_t state_info = { {PSCI_LOCAL_STATE_RUN} };
/* Init registers that never change for the lifetime of TF-A */
cm_manage_extensions_el3();
cm_manage_extensions_el3(cpu_idx);
/*
* Verify that we have been explicitly turned ON or resumed from

View file

@ -42,12 +42,13 @@ static void psci_cpu_suspend_to_standby_finish(unsigned int end_pwrlvl,
* This function does generic and platform specific suspend to power down
* operations.
******************************************************************************/
static void psci_suspend_to_pwrdown_start(unsigned int end_pwrlvl,
static void psci_suspend_to_pwrdown_start(unsigned int idx,
unsigned int end_pwrlvl,
unsigned int max_off_lvl,
const entry_point_info_t *ep,
const psci_power_state_t *state_info)
{
PUBLISH_EVENT(psci_suspend_pwrdown_start);
PUBLISH_EVENT_ARG(psci_suspend_pwrdown_start, &idx);
#if PSCI_OS_INIT_MODE
#ifdef PLAT_MAX_CPU_SUSPEND_PWR_LVL
@ -223,7 +224,7 @@ int psci_cpu_suspend_start(unsigned int idx,
#endif
#endif
max_off_lvl = psci_find_max_off_lvl(state_info);
psci_suspend_to_pwrdown_start(end_pwrlvl, max_off_lvl, ep, state_info);
psci_suspend_to_pwrdown_start(idx, end_pwrlvl, end_pwrlvl, ep, state_info);
}
/*
@ -382,5 +383,5 @@ void psci_cpu_suspend_to_powerdown_finish(unsigned int cpu_idx, unsigned int max
/* This loses its meaning when not suspending, reset so it's correct for OFF */
psci_set_suspend_pwrlvl(PLAT_MAX_PWR_LVL);
PUBLISH_EVENT(psci_suspend_pwrdown_finish);
PUBLISH_EVENT_ARG(psci_suspend_pwrdown_finish, &cpu_idx);
}

View file

@ -291,7 +291,6 @@ endif
# Feature flags for supporting Activity monitor extensions.
ENABLE_FEAT_AMU ?= 0
ENABLE_AMU_AUXILIARY_COUNTERS ?= 0
ENABLE_AMU_FCONF ?= 0
AMU_RESTRICT_COUNTERS ?= 1
# Build option to enable MPAM for lower ELs.

View file

@ -32,7 +32,6 @@ ENABLE_SME_FOR_SWD := 1
ENABLE_TRBE_FOR_NS := 1
ENABLE_SYS_REG_TRACE_FOR_NS := 1
ENABLE_FEAT_AMU := 1
ENABLE_AMU_FCONF := 1
ENABLE_AMU_AUXILIARY_COUNTERS := 1
ENABLE_MPMM := 1
ENABLE_FEAT_MTE2 := 2

View file

@ -72,6 +72,18 @@ static scmi_channel_plat_info_t tc_scmi_plat_info = {
};
#endif
/* the bottom 3 AMU group 1 counters */
#define MPMM_GEARS ((1 << 0) | (1 << 1) | (1 << 2))
uint16_t plat_amu_aux_enables[PLATFORM_CORE_COUNT] = {
MPMM_GEARS, MPMM_GEARS, MPMM_GEARS, MPMM_GEARS,
MPMM_GEARS, MPMM_GEARS, MPMM_GEARS, MPMM_GEARS,
#if PLATFORM_CORE_COUNT == 14
MPMM_GEARS, MPMM_GEARS, MPMM_GEARS, MPMM_GEARS,
MPMM_GEARS, MPMM_GEARS
#endif
};
#if (TARGET_PLATFORM == 3) || (TARGET_PLATFORM == 4)
static void enable_ns_mcn_pmu(void)
{