zephyr/arch/riscv/Kconfig

389 lines
12 KiB
Plaintext

# Copyright (c) 2016 Jean-Paul Etienne <fractalclone@gmail.com>
# SPDX-License-Identifier: Apache-2.0
menu "RISCV Options"
depends on RISCV
config ARCH
string
default "riscv"
config FLOAT_HARD
bool "Hard-float calling convention"
default y
depends on FPU
help
This option enables the hard-float calling convention.
config RISCV_GP
bool "RISC-V global pointer relative addressing"
default n
help
Use global pointer relative addressing for small globals declared
anywhere in the executable. It can benefit performance and reduce
the code size.
Note: To support this feature, RISC-V SoC needs to initialize
global pointer at program start or earlier than any instruction
using GP relative addressing.
config RISCV_ALWAYS_SWITCH_THROUGH_ECALL
bool "Do not use mret outside a trap handler context"
depends on MULTITHREADING
help
Use mret instruction only when in a trap handler.
This is for RISC-V implementations that require every mret to be
balanced with an ecall. This is not required by the RISC-V spec
and most people should say n here to minimize context switching
overhead.
config RISCV_EXCEPTION_STACK_TRACE
bool
default y
imply THREAD_STACK_INFO
help
Internal config to enable runtime stack traces on fatal exceptions.
menu "RISCV Processor Options"
config INCLUDE_RESET_VECTOR
bool "Jumps to __initialize directly"
help
Select 'y' here to use the Zephyr provided default implementation that
jumps to `__initialize` directly. Otherwise a SOC needs to provide its
custom `__reset` routine.
config RISCV_PRIVILEGED
bool
select ARCH_HAS_RAMFUNC_SUPPORT if XIP
help
Option selected by SoCs implementing the RISC-V privileged ISA.
config RISCV_SOC_HAS_ISR_STACKING
bool
depends on !USERSPACE
help
Enable low-level SOC-specific hardware stacking / unstacking
operations during ISR. This hidden option needs to be selected by SoC
if this feature is supported.
Some SOCs implement a mechanism for which, on interrupt handling,
part of the context is automatically saved by the hardware on the
stack according to a custom ESF format. The same part of the context
is automatically restored by hardware on mret.
Enabling this option requires that the SoC provides a
soc_isr_stacking.h header which defines the following:
- SOC_ISR_SW_STACKING: macro guarded by _ASMLANGUAGE called by the
IRQ wrapper assembly code on ISR entry to save in the ESF the
remaining part of the context not pushed already on the stack by
the hardware.
- SOC_ISR_SW_UNSTACKING: macro guarded by _ASMLANGUAGE called by the
IRQ wrapper assembly code on ISR exit to restore the part of the
context from the ESF that won't be restored by hardware on mret.
- SOC_ISR_STACKING_ESF_DECLARE: structure declaration for the ESF
guarded by !_ASMLANGUAGE. The ESF should be defined to account for
the hardware stacked registers in the proper order as they are
saved on the stack by the hardware, and the registers saved by the
software macros. The structure must be called 'struct arch_esf'.
config RISCV_SOC_HAS_CUSTOM_IRQ_HANDLING
bool
help
This allows the SoC to overwrite the irq handling. If enabled, the
function __soc_handle_all_irqs has to be implemented. It shall service
and clear all pending interrupts.
config RISCV_SOC_HAS_CUSTOM_IRQ_LOCK_OPS
bool
help
Hidden option to allow SoC to overwrite arch_irq_lock(),
arch_irq_unlock() and arch_irq_unlocked() functions with
platform-specific versions named z_soc_irq_lock(), z_soc_irq_unlock()
and z_soc_irq_unlocked().
Enable this hidden option and specialize the z_soc_* functions when
the RISC-V SoC needs to do something different and more than reading and
writing the mstatus register to lock and unlock the IRQs.
config RISCV_SOC_HAS_CUSTOM_SYS_IO
bool
help
Hidden option to allow SoC to overwrite sys_read*(), sys_write*() functions with
platform-specific versions named z_soc_sys_read*() and z_soc_sys_write*().
Enable this hidden option and specialize the z_soc_* functions when
the RISC-V SoC needs to do something different and more than reading and
writing the registers.
config RISCV_SOC_CONTEXT_SAVE
bool "SOC-based context saving in IRQ handlers"
select RISCV_SOC_OFFSETS
help
Enable low-level SOC-specific context management, for SOCs
with extra state that must be saved when entering an
interrupt/exception, and restored on exit. If unsure, leave
this at the default value.
Enabling this option requires that the SoC provide a
soc_context.h header which defines the following macros:
- SOC_ESF_MEMBERS: structure component declarations to
allocate space for. The last such declaration should not
end in a semicolon, for portability. The generic RISC-V
architecture code will allocate space for these members in
a "struct soc_esf" type (typedefed to soc_esf_t), which will
be available if arch.h is included.
- SOC_ESF_INIT: structure contents initializer for struct soc_esf
state. The last initialized member should not end in a comma.
The generic architecture IRQ wrapper will also call
\_\_soc_save_context and \_\_soc_restore_context routines at
ISR entry and exit, respectively. These should typically
be implemented in assembly. If they were C functions, they
would have these signatures:
``void __soc_save_context(soc_esf_t *state);``
``void __soc_restore_context(soc_esf_t *state);``
The calls obey standard calling conventions; i.e., the state
pointer address is in a0, and ra contains the return address.
config RISCV_SOC_OFFSETS
bool "SOC-based offsets"
help
Enabling this option requires that the SoC provide a soc_offsets.h
header which defines the following macros:
- GEN_SOC_OFFSET_SYMS(): a macro which expands to
GEN_OFFSET_SYM(soc_esf_t, soc_specific_member) calls
to ensure offset macros for SOC_ESF_MEMBERS are defined
in offsets.h. The last one should not end in a semicolon.
See gen_offset.h for more details.
config RISCV_HAS_PLIC
bool
depends on RISCV_PRIVILEGED
help
Does the SOC provide support for a Platform Level Interrupt Controller (PLIC).
config RISCV_HAS_CLIC
bool
depends on RISCV_PRIVILEGED
help
Does the SOC provide support for a Core-Local Interrupt Controller (CLIC).
config RISCV_SOC_EXCEPTION_FROM_IRQ
bool
help
Option selected by SoCs that require a custom mechanism to check if
an exception is the result of an interrupt or not. If selected,
__soc_is_irq() needs to be implemented by the SoC.
config RISCV_SOC_INTERRUPT_INIT
bool "SOC-based interrupt initialization"
help
Enable SOC-based interrupt initialization
(call soc_interrupt_init, within _IntLibInit when enabled)
config RISCV_MCAUSE_EXCEPTION_MASK
hex
default 0x7FFFFFFFFFFFFFFF if 64BIT
default 0x7FFFFFFF
help
Specify the bits to use for exception code in mcause register.
config RISCV_GENERIC_TOOLCHAIN
bool "Compile using generic riscv32 toolchain"
default y
help
Compile using generic riscv32 toolchain.
Allow SOCs that have custom extended riscv ISA to still
compile with generic riscv32 toolchain.
config GEN_ISR_TABLES
default y
config GEN_IRQ_VECTOR_TABLE
default n
config RISCV_RESERVED_IRQ_ISR_TABLES_OFFSET
int
default 0
depends on GEN_ISR_TABLES
help
On some RISCV platform the first interrupt vectors are primarly
intended for inter-hart interrupt signaling and so retained for that
purpose and not available. When this option is set, all the IRQ
vectors are shifted by this offset value when installed into the
software ISR table and the IRQ vector table. CONFIG_NUM_IRQS must be
properly sized to take into account this offset. This is a hidden
option which needs to be set per architecture and left alone.
config NUM_IRQS
int
config RV_BOOT_HART
int "Starting HART ID"
default 0
help
This option sets the starting HART ID for the SMP core.
For RISC-V systems such as MPFS and FU540 this would be set to 1 to
skip the E51 HART 0 as it is not usable in SMP configurations.
config RISCV_HART_MASK
int
default -1
help
Configures the mask for the HART ID.
For RISC-V systems with HART ID starting from non-zero value,
i.e. 128, 129, ..(0x80, 8x81, ..), this can be configured to 63 (0x7f)
such that we can extract the bits that start from 0.
config RISCV_PMP
bool "RISC-V PMP Support"
select THREAD_STACK_INFO
select CPU_HAS_MPU
select ARCH_HAS_USERSPACE
select ARCH_HAS_STACK_PROTECTION
select MPU
select SRAM_REGION_PERMISSIONS
select ARCH_MEM_DOMAIN_SYNCHRONOUS_API if USERSPACE
select ARCH_MEM_DOMAIN_DATA if USERSPACE
select THREAD_LOCAL_STORAGE if USERSPACE
select ARCH_MEM_DOMAIN_SUPPORTS_ISOLATED_STACKS
select MEM_DOMAIN_ISOLATED_STACKS
help
MCU implements Physical Memory Protection.
if RISCV_PMP
config PMP_SLOTS
int "Number of PMP slots"
default 8
help
This is the number of PMP entries implemented by the hardware.
Typical values are 8 or 16.
config PMP_NO_TOR
bool
help
Set this if TOR (Top Of Range) mode is not supported.
config PMP_NO_NA4
bool
help
Set this if NA4 (Naturally Aligned 4-byte) mode is not supported.
config PMP_NO_NAPOT
bool
help
Set this if NAPOT (Naturally Aligned Power Of Two) is not supported.
config PMP_POWER_OF_TWO_ALIGNMENT
bool "Enforce power-of-two alignment on PMP memory areas" if !PMP_NO_TOR
default y if TEST_USERSPACE
default y if (PMP_SLOTS = 8)
default y if PMP_NO_TOR
select MPU_REQUIRES_POWER_OF_TWO_ALIGNMENT
select GEN_PRIV_STACKS
help
This option reduces the PMP slot usage but increases
memory consumption. Useful when enabling userspace mode with
many memory domains and/or few PMP slots available.
config PMP_GRANULARITY
int "The granularity of PMP address matching"
default 8 if (PMP_NO_TOR && PMP_NO_NA4)
default 4
help
The granularity must be a power of 2 greater than or equal to 4
(ie 4, 8, 16, ...), but if neither TOR mode nor NA4 mode is
supported, the minimum granularity is 8.
endif #RISCV_PMP
config PMP_STACK_GUARD
def_bool y
depends on MULTITHREADING
depends on HW_STACK_PROTECTION
config PMP_STACK_GUARD_MIN_SIZE
int "Stack Guard area size"
depends on PMP_STACK_GUARD
default 1024 if 64BIT
default 512
help
The Hardware Stack Protection implements a guard area at the bottom
of the stack using the PMP to catch stack overflows by marking that
guard area not accessible.
This is the size of the guard area. This should be large enough to
catch any sudden jump in stack pointer decrement, plus some
wiggle room to accommodate the eventual overflow exception
stack usage.
# Implement the null pointer detection using the Physical Memory Protection
# (PMP) Unit.
config NULL_POINTER_EXCEPTION_DETECTION_PMP
bool "Use PMP for null pointer exception detection"
depends on RISCV_PMP
help
Null pointer dereference detection implemented
using PMP functionality.
if NULL_POINTER_EXCEPTION_DETECTION_PMP
config NULL_POINTER_EXCEPTION_REGION_SIZE
hex "Inaccessible region to implement null pointer detection"
default 0x10
help
Use a PMP slot to make region (starting at address 0x0) inaccessible for
detecting null pointer dereferencing (raising a CPU access fault).
Minimum is 4 bytes.
endif # NULL_POINTER_EXCEPTION_DETECTION_PMP
endmenu
config MAIN_STACK_SIZE
default 4096 if 64BIT
default 2048 if PMP_STACK_GUARD
config TEST_EXTRA_STACK_SIZE
default 1536
config CMSIS_THREAD_MAX_STACK_SIZE
default 1024 if 64BIT
config CMSIS_V2_THREAD_MAX_STACK_SIZE
default 1024 if 64BIT
config ARCH_IRQ_VECTOR_TABLE_ALIGN
default 256
config RISCV_TRAP_HANDLER_ALIGNMENT
int "Alignment of RISC-V trap handler in bytes"
default 64 if RISCV_HAS_CLIC
default 4
help
This value configures the alignment of RISC-V trap handling
code. The requirement for a particular alignment arises from
the format of MTVEC register which is RISC-V platform-specific.
The minimum alignment is 4 bytes according to the Spec.
config GEN_IRQ_VECTOR_TABLE
select RISCV_VECTORED_MODE if RISCV_PRIVILEGED
config ARCH_HAS_SINGLE_THREAD_SUPPORT
default y if !SMP
rsource "Kconfig.isa"
endmenu