lib: sbi_ecall: Remove mcause, scratch and hartid parameters

We remove mcause, scratch and hartid parameters from various
functions for ecall handling because we can always get current
HART id and current scratch pointer using just one CSR access.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

lib: sbi_misaligned_ldst: Remove mcause, scratch and hartid parameters

We remove mcause, scratch and hartid parameters from various functions
for misaligned load/store handling because we can always get current
HART id and current scratch pointer using just one CSR access.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

lib: sbi_illegal_insn: Remove mcause, scratch and hartid parameters

We remove mcause, scratch and hartid parameters from various
functions for illegal instruction handling because we can always
get current HART id and current scratch pointer using just one
CSR access.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

lib: sbi_emulate_csr: Remove scratch and hartid parameter

We remove scratch and hartid parameter from various functions
for CSR emulation because we can always get current HART id
and current scratch pointer using just one CSR access.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

lib: sbi_trap: Remove scratch parameter from sbi_trap_redirect()

The scratch parameter of sbi_trap_redirect() is not used hence we
remove it.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

lib: Remove scratch parameter from unpriv load/store functions

The scratch parameter of unpriv load/store functions is now redundant
hence we remove it.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

lib: Optimize unpriv load/store implementation

This patch optimize unpriv load/store implementation by having
dedicated unpriv trap handler (just like KVM RISC-V).

As a result of this optimization:
1. We have reduced roughly 13+ instruction in all unpriv load/store
   functions. The reduced instruction also include two function calls.
2. Per-HART trap info pointer in scratch space is now redundant
   hence removed.
3. The sbi_trap_handler() is now much cleaner because we don't have
   to handle unpriv load/store traps.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

lib: utils: Improve fdt_cpu_fixup() implementation

Currently, the fdt_cpu_fixup() implementation assumes:
1. We have one CPU DT for each HART under /cpus DT node
2. The CPU DT nodes are named sequentially (i.e cpu@0,
   cpu@1, ...) which is not true for discontinuous and
   sparse HART ids (i.e. cpu@0, cpu@4, cpu@5). Generally,
   CPU DT node are named based on HART id and not HART
   index

If any of the above assumptions are violated then the
fdt_cpu_fixup() will not work.

This improves fdt_cpu_fixup() implementation and makes
it independent of above assumptions.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

scripts: Extend create-binary-archive.sh for unified binary tar ball

This patch extends/revamps create-binary-archive.sh to allow creating
unified binary tar ball containing both RV32 and RV64 images. Users
can also create "RV32 only" or "RV64 only" tar balls as well.

Few example commands are as follows:
./scripts/create-binary-archive.sh (Build both RV32 & RV64 images)
./scripts/create-binary-archive.sh -x 32 (Build RV32 only images)
./scripts/create-binary-archive.sh -x 64 (Build RV64 only images)

Signed-off-by: Anup Patel <anup.patel@wdc.com>

lib: prevent coldboot_lottery from overflowing

HSM_STOP will trigger multiple executions of sbi_init, atomic_add_return may
trigger coldboot_lottery overflow

Signed-off-by: Xiang Wang <merle@hardenedlinux.org>
Reviewed-by: Anup Patel <anup.patel@wdc.com>

platform: openpiton: Read the device configurations from device tree

OpenPiton is designed to run on different FPGAs with different
configurations. Use the fdt parser to retrieve the required values
from device tree instead of using fixed values.

Signed-off-by: Atish Patra <atish.patra@wdc.com>
Tested-by: Jonathan Balkind <jbalkind@cs.princeton.edu>
Reviewed-by: Anup Patel <anup.patel@wdc.com>

lib: utils: Move fdt fixup helper routines to a different file

FDT helper file contain both fdt fixup and parsing functions.

Split the fixup related functions to a separate file for a better code
organization.

Signed-off-by: Atish Patra <atish.patra@wdc.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>

platform: fpga/common: Add a fdt parsing helper functions

Different DT based platforms from the sam family may reuse IP blocks with
different configurations. These different configurations can be obtained
by parsing the device tree.

Add a FDT parser framework that can parse various device configurations from
device tree. Currently, the parsing algorithms doesn't cover all the use cases
or possible combination of DT configurations. It will be improved over time.

Signed-off-by: Atish Patra <atish.patra@wdc.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>

platform: Add OpenPiton platform support

OpenPiton is a research platform from Princeton University [1].

"OpenPiton is the world's first open source, general purpose,
multithreaded manycore processor. It is a tiled manycore
framework scalable from one to 1/2 billion cores."

Add OpenSBI support for OpenPiton. As it is based on ariane core,
it reuses the platform code from arine project.

[1]. https://github.com/PrincetonUniversity/openpiton

Signed-off-by: Atish Patra <atish.patra@wdc.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>

platform: fpga/ariane: Remove redundant plic address macros

All the common PLIC specific macros are already defined in plic.c.

Remove it from platform code. While at it, Fix the other coding style
issues.

Signed-off-by: Atish Patra <atish.patra@wdc.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>

platform: Move ariane standalone fpga project to its own project

Signed-off-by: Atish Patra <atish.patra@wdc.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>

lib: Check MSIP bit after returning from WFI

Commit 71d2b837c46e (lib: Move all coldboot wait APIs to sbi_init.c)
caused a regression while moving the code from sbi_hart.c to sbi_init.c.

As per original commit text, WFI can be implemented as a NOP according
to the RISC-V privilege specification. Software should ensure that
relevant interrupt pending bits are set. Otherwise, loop back to WFI.
Fix the regression by applying the original patch to sbi_init.c.

Fixes: 71d2b837c46e ("lib: Move all coldboot wait APIs to sbi_init.c")

Signed-off-by: Atish Patra <atish.patra@wdc.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>

lib: Fix sbi_ecall_register_extension to prevent extension IDs overlap

The original code does not prevent the following scenarios:
> sbi_ecall_register_extension(ext1); /* extension id (70-80) */
> sbi_ecall_register_extension(ext2); /* extension id (50-100) */

Signed-off-by: Xiang Wang <merle@hardenedlinux.org>
Reviewed-by: Anup Patel <anup.patel@wdc.com>

platform: sifive/fu540: Remove FU540_ENABLED_HART_MASK option

The FU540_ENABLED_HART_MASK compile time option was added for initial
bring-up on SiFive Unleashed. This option is redundant now because
disabled_hart_mask is already removed. Based on this rationale, we
remove FU540_ENABLED_HART_MASK compile time option.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

include: sbi_platform: Introduce HART index to HART id table

A platform can have discontinuous and/or sparse HART ids so we
cannot always assume a set of HARTs with continuous HART ids.

This patch adds support for discontinuous and sparse HART ids by
introducing HART index to HART id table. This table has platform
hart_count entries and it maps HART index to HART id.

The HART index to HART id table has only two restrictions:
1. HART index < sbi_platform hart_count
2. HART id < SBI_HARTMASK_MAX_BITS

Example1:
Let's say we have a platform with 2 HART ids 11 and 22, for such a
a platform:
hart_count = 2
hart_index2id[0] = 11
hart_index2id[1] = 22

Example2:
Let's say we have a platform with 5 HARTs ids 0, 1, 2, 3, and 4
but out of these HART with id 0 is not usable so for such a platform:
hart_count = 5
hart_index2id[0] = -1U
hart_index2id[1] = 1
hart_index2id[2] = 2
hart_index2id[3] = 3
hart_index2id[4] = 4
OR
hart_count = 4
hart_index2id[0] = 1
hart_index2id[1] = 2
hart_index2id[2] = 3
hart_index2id[3] = 4

With HART index to HART id table in place, the hart_disabled()
callback is now redundant so we remove it as well.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

lib: Don't use sbi_platform_hart_count() API

We don't need to use sbi_platform_hart_count() in sbi_init and
sbi_scratch because checking sbi_platform_hart_disabled() or
return value of sbi_hartid_to_scratch() is sufficient.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

lib: sbi_hsm: Don't use sbi_platform_hart_disabled() API

Checking return value of sbi_hartid_to_scratch() is sufficient
so no need to explicitly check for disabled HART using the
sbi_platform_hart_disabled() API.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

lib: sbi_hsm: Remove scratch parameter from hart_started_mask() API

The scratch parameter in sbi_hsm_hart_started_mask() API is now
redundant hence removing it.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

lib: sbi_hsm: Don't use sbi_platform_hart_count() API

We remove usage of sbi_platform_hart_count() API from sbi_hsm
so that discontinuous and sparse HART ids can be supported.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>

lib: sbi_scratch: Don't set hartid_to_scratch table for disabled HART

As a step towards supporting discontinuous and sparse HART ids, we
don't set hartid_to_scratch table for disabled HARTs.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

include: sbi_platform: Improve sbi_platform_hart_disabled() API

The sbi_platform_hart_disabled() should return TRUE for HART id
greater than platform hart count.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>

lib: Handle failure of sbi_hartid_to_scratch() API

The sbi_hartid_to_scratch() API can fail for non-existent HARTs so
all uses of sbi_hartid_to_scratch() API should check return value.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

lib: sbi_tlb: Simplify sbi_tlb_entry_process() function

We remove redundant scratch parameter from sbi_tlb_entry_process()
function.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

lib: sbi_hsm: Simplify hart_get_state() and hart_started() APIs

We remove redundant scratch parameter from sbi_hsm_hart_get_state()
and sbi_hsm_hart_started() APIs.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

include: Simplify HART id to scratch macro

This patch simplify HART id to scratch macro as follows:
1. Remove current "scratch" pointer argument because now we
   use HART id to scratch table
2. Rename sbi_hart_id_to_scratch() to sbi_hartid_to_scratch()
   to have macro name consistent with the name of callback
   in struct sbi_scratch

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

lib: sbi_scratch: Introduce HART id to scratch table

Instead of calling hartid_to_scratch() callback every time when
we want sbi_scratch pointer from HART id, we create a table of
sbi_scratch pointers and use that to get sbi_scratch pointer.

As a result of HART id to scratch table, the conversion of
HART id to sbi_scratch pointer is just 2-3 instructions which
was 9 instructions previously.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

platform: Update to call general DT fix-up helper

Platform andes/ae350, ariane-fpga, qemu/virt and sifive/fu540 codes
have been updated to call this general DT fix-up helper.

Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>

lib: utils: Add a general device tree fix-up helper

This adds a general device tree fix-up helper to do all required
device tree fix-ups for a typical platform.

Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>

platform: sifive/fu540: Replace cpu0 node fix-up with the new helper

This replaces the FU540 specific cpu0 node "status" property fix-up
with the newly introduced generic fdt_cpu_fixup() helper.

Unlike previous logic, the helper routine does not test the "mmu-type"
property to determine which node we should fix up, instead it uses
sbi_platform_hart_disabled() API.

Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>

lib: utils: Add a fdt_cpu_fixup() helper

Add a helper routine to updates the "status" property of a CPU node
in the device tree to "disabled" if that hart is in disabled state.

Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>

platform: sifive/fu540: Remove "stdout-path" fix-up

As of today the upstream U-Boot & Linux kernel ships a device tree
that already has "stdout-path" properly set in the "/chosen" node.
This is the same with the QEMU 'sifive_u' machine. Hence the codes
to fix up the "stdout-path" in OpenSBI is not necessary.

Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

platform: sifive/fu540: Fix up DT for reserved memory

This calls fdt_reserved_memory_fixup() helper in the platform's
final_init() routine.

Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

platform: qemu/virt: Fix up DT for reserved memory

This calls fdt_reserved_memory_fixup() helper in the platform's
final_init() routine.

Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>

platform: ariane-fpga: Fix up DT for reserved memory

This calls fdt_reserved_memory_fixup() helper in the platform's
final_init() routine.

Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

platform: andes/ae350: Fix up DT for reserved memory

This calls fdt_reserved_memory_fixup() helper in the platform's
final_init() routine.

Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

lib: utils: Move PLIC DT fix up codes to fdt_helper.c

Now that we have a dedicated fdt_helper.c file for DT releated
helper routines, move plic_fdt_fixup() codes from plic.c to
fdt_helper.c and rename it to fdt_plic_fixup() at the same time,
to keep name consistency in the same file.

Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

platform: Clean up include header files

Adjust the order of include header files in alphabetical order in
platform codes. Also remove unnecessary inclusions.

Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

lib: utils: Add a fdt_reserved_memory_fixup() helper

Add a helper routine to insert a child node of the reserved memory
node in the device tree that describes the protected memory region
done by OpenSBI via PMP.

Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

libfdt: Compile fdt_addresses.c

Pull fdt_addresses.o in for fdt_address_cells() & fdt_size_cells().

Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

lib: sbi: Update pmp_get() to return decoded size directly

Currently pmp_get() returns the log2 length of the PMP memory
region size. The caller has to calculate the size based on that
and the same codes are duplicated.

Update this function to return decoded size directly.

Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

scripts: Cover sifive/fu540 in the 32-bit build

We now support building sifive/fu540 as a 32-bit platform. Add it
in the 32-bit platform list of binary archive script.

Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>

platform: Use one unified per-HART stack size macro for all platforms

As of today all platforms use 8KB of per-HART stack hence there is
no need for each platform to define its own macro or use the magic
number. Create one macro for all platforms. Platform still can use
its own version if needed.

Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>

platform: Set per-HART stack size to 8KB in the template platform codes

The template platform codes should set per-HART stack size to 8KB
to avoid possible mistakes of future platform ports.

Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>

platform: ariane-fpga: Set per-HART stack size to 8KB

Recent commit

  4a603eb ("platform: kendryte/k210: Set per-HART stack size to 8KB")

forgot to update ariane-fpga codes, and the following commit

  678c3c3 ("include: sbi_scratch: Set per-HART scratch size to 4KB")

changed the per-HART scratch size to 4KB, which potentially breaks
ariane-fpga platform.

This patch set per-HART stack size of ariane-fpga to 8KB for consistency.

Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>

Makefile: Fix distclean make target

The default install directory is not being removed correctly by
distclean make target due to recent changes. This patch fixes
distclean make target to fix default install directory removal.

Fixes: 82ae8e8fe2a3 ("makefile: Do setup of the install target
more flexible")
Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

include: Make sbi_current_hartid() as macro in riscv_asm.h

The sbi_current_hartid() being a regular function is quite
expensive because for callers it is a function call instead
of a direct CSR read. This patch converts sbi_current_hartid()
into a macro in riscv_asm.h.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

lib: sbi_hsm: Optimize sbi_hsm_hart_get_state() implementation

Now that sbi_hart_id_to_scratch() is optimized, we don't need
the "if ()" statement. Also, the hstate local variable is
redundant so we remove that as well.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

firmware: fw_base: Optimize _hartid_to_scratch() implementation

This patch optimizes _hartid_to_scratch() in following ways:
1. Use caller saved registers instead of callee saved registers
   so that we don't need to save/restore registers on stack
2. Remove second redundant mul instruction by re-arranging
   instructions

Overall, we reduce 9 instructions in _hartid_to_scratch()
implementation.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>

include: Make sbi_hart_id_to_scratch() as macro

The sbi_hart_id_to_scratch() just forwards call to firmware specific
hartid_to_scratch() callback so we make sbi_hart_id_to_scratch() as
macro in sbi_scratch.h instead of regular function in sbi_hart.c.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>

libfdt: Upgrade to v1.5.1 release

Sync with latest libfdt v1.5.1 release source codes.

Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>

libfdt: Add INT32_MAX and UINT32_MAX in libfdt_env.h

Add two macros in preparation to sync libfdt codes to latest v1.5.1
release from upstream.

Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>

sbi: Add definitions for true/false

We currently have TRUE/FALSE, but it could be convenient to have
true/false as well.

Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>

platform: ariane-fpga: Change license for ariane-fpga from GPL-2.0 to BSD-2

Signed-off-by: Panagiotis Peristerakis <perister@ics.forth.gr>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>

lib: sbi: Fix misaligned trap handling

Compile time checks of __riscv_compressed can only check if OpenSBI is
being compiled using compressed instructions or not. Checking this macro
does not indicate if an instruction that generated a misaligned trap is
a compressed instruction or not.

Since the misaligned trap handling code inspects instructions _C_ bits
to detect compressed instructions, we can remove all static checks on
__riscv_compressed and dissociate hanlding of misaligned traps and
OpenSBI compilation.

Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>

docs: Remove out-of-date documentation

Upstream U-Boot now have SMP support and doesn't require any additional
patches for HiFive Unleashed.

Update the documentation.

Signed-off-by: Atish Patra <atish.patra@wdc.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>

lib: sbi_hart: Remove HART available mask and related APIs

The HART available mask and related APIs are now totally redundant
because of more extensive HART state machine implemented by sbi_hsm.

Due to above, we remove HART available mask and related APIs.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>

lib: sbi_ipi: Use sbi_hsm_hart_started_mask() API

This patch replaces use of sbi_hart_available_mask() API with
sbi_hsm_hart_started_mask() API.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>

lib: sbi_system: Use sbi_hsm_hart_started_mask() API

This patch replaces use of sbi_hart_available_mask() API with
sbi_hsm_hart_started_mask() API.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>

lib: sbi_ecall_legacy: Use sbi_hsm_hart_started_mask() API

This patch replaces use of sbi_hart_available_mask() API with
sbi_hsm_hart_started_mask API.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>

lib: Introduce sbi_hsm_hart_started_mask() API

This patch introduce sbi_hsm_hart_started_mask() API as
a replacement of sbi_hart_available_mask() API.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>

include: Remove disabled_hart_mask from sbi_platform

The disabled_hard_mask in sbi_platform is only 64bits wide so we
cannot disable a HART with HARTID > 63. To tackle this, we remove
disabled_hart_mask and replace it with hart_disabled() platform
callback.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

lib: sbi_init: Use hartmask for coldboot wait

The coldboot_wait_bitmap is of fixed size and has __riscv_xlen
bits. This limits us to scale beyond __riscv_xlen HARTs hence
we replace coldboot_wait_bitmap with coldboot_wait_hmask which
is of type struct sbi_hartmask.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

lib: Move all coldboot wait APIs to sbi_init.c

The coldboot wait APIs are only used by sbi_init.c so no point in
having coldboot related code in sbi_hart.c.

As per-above rationale, we move all coldboot wait related APIs to
sbi_init.c as static/local functions.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>

lib: sbi_tlb: Use sbi_hartmask in sbi_tlb_info

Instead of using single ulong as source mask for sbi_tlb_info,
we use sbi_hartmask. This way sbi_tlb_info can easily scale
for large number of HARTs.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

lib: Introduce SBI_TLB_INFO_INIT() helper macro

We introduce SBI_TLB_INFO_INIT() helper macro to help easy
initialization of struct sbi_tlb_info which is passed to the
sbi_tlb_request() API.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

lib: sbi_init: Don't allow HARTID greater than SBI_HARTMASK_MAX_BITS

We only allow HARTs with HARTID less than SBI_HARTMASK_MAX_BITS in
sbi_init() function so that sbi_hartmask can safely used across
OpenSBI sources.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

include: Simple hartmask library

We add a simple hartmask library to create and maintain HART mask
in different parts of OpenSBI sources. This new library is build
as wrapper library on-top-of bitmap library.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

lib: Simple bitmap library

We add simple bitmap library which will help us create and maintain
bitmaps. It will also help us create a simple HART mask library.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

lib: serial: Fix coding style issues

This fixes various coding style issues found in the serial codes.
No functional changes.

Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>

lib: sbi: Fix coding style issues

This fixes various coding style issues found in the SBI codes.
No functional changes.

Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>

platform: ae350: Sort build objects in alphabetical order

Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>

lib: Sort build objects in alphabetical order

Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>

include: sbi_scratch: Set per-HART scratch size to 4KB

Currently, the per-HART scratch size is 256 bytes on RV32 and
512 bytes on RV64. This patch set per-HART scratch size to 4KB
(4096 bytes) for both RV32 and RV64 so that we don't run-out
of scratch space anytime soon.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

platform: kendryte/k210: Set per-HART stack size to 8KB

All platform except kendryte/k210 use 8KB of per-HART stack hence
this patch set per-HART stack size of kendryte/k210 to 8KB for
consistency.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

include: sbi_bitops: More useful bit operations

This patch extends our bit operation library with mechanism to:
1. Iteratively traverse bits
2. Set bit
3. Clear bit
4. Change bit
5. ... other helpful functions ...

Most the above is adopted from Xvisor sources.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

include: Move bits related defines and macros to sbi_bitops.h

The right location for all bits related defines and macros is
sbi_bitops.h hence this patch. With this patch, the sbi_bits.h
is redundant so we remove it.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

lib: Fix return type of sbi_hsm_hart_started()

The return type of sbi_hsm_hart_started() should be bool.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>

platform: sifive: fu540: allow sv32 as an mmu-type

There has already been a commit to master which added 32-bit specific
fdt/payload addresses for the sifive/fu540 platform [0]. This commit
introduces another change for using sifive/fu540 as a 32-bit platform.
On 32-bit platforms, cores with the SV32 MMU type should not be
disabled. For this reason, this commit also allows using this MMU type
on the sifive/fu540 platform.

Alternatively it would also be possible to only allow SV39 and SV48 if
`__riscv_xlen == 64` and SV32 if `__riscv_xlen == 32`. Removing the
check entirely would also be an option.

[0]: 66fb729a1e46a9a46e809f3b7867fef91477e494

Signed-off-by: Sören Tempel <tempel@uni-bremen.de>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>

makefile: Do setup of the install target more flexible

- Add possibility to setup include, libs, firmware and docs paths.

- Change the default installation paths for include, libs, firmware
  and docs to meet FHS [1].

[1] https://refspecs.linuxfoundation.org/FHS_3.0/fhs/index.html

Signed-off-by: Nikita Ermakov <coffe92@gmail.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>

lib: Check MSIP bit after returning from WFI

As per the RISC-V privilege specification, WFI can be implemented as
a NOP. Software should ensure that relevant interrupt pending bits
are set. Otherwise, loop back to WFI.

Signed-off-by: Atish Patra <atish.patra@wdc.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>

lib: Implement Hart State Management (HSM) SBI extension

This patch adds support HSM extension. The specification is available
at https://github.com/riscv/riscv-sbi-doc.

It allows to implement hart hotplug and fixed ordered hart booting in
supervisor.

Signed-off-by: Atish Patra <atish.patra@wdc.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>

lib: Add possible hart status values

SBI HSM extension defines possible hart status values in the
specification.

Define all possible status values. Add a helper function to
convert hart state to status because hart states are internal
to OpenSBI only and may not match the status values defined in
the specification.

Signed-off-by: Atish Patra <atish.patra@wdc.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>

lib: Implement hart hotplug

This patch adds support for hart hotplug in OpenSBI using a generic WFI
based approach. Hart hotplug can be achieved via SBI HSM extension which
allows supervisor mode software to start or stop any harts anytime.

Any platform wishes to implement platform specific hart hotplug must
implement both hart_start and hart_stop in addition to enable platform
feature SBI_PLATFORM_HAS_HART_HOTPLUG.

Signed-off-by: Atish Patra <atish.patra@wdc.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>

lib: Add a new platform feature to bringup secondary harts

A platform may have a specific method to bring-up secondary harts for the
first time but may rely on generic WFI based approach for hart hotplug in
absence of a platform specific hart hotplug method.

Define a platform feature flag for such platforms. The platform needs to
implement platform specific bring-up method in hart_start and not define
hart_stop method in this case. They must only define
SBI_PLATFORM_HAS_HART_SECONDARY_BOOT.

SBI_PLATFORM_HAS_HART_HOTPLUG should only be defined when the platform
intend to support both hart_start and hart_stop and do not intend to rely
on generic WFI based approach.

Signed-off-by: Atish Patra <atish.patra@wdc.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>

include: Bump-up version to 0.6

This patch updates OpenSBI version to 0.6 as part of
release preparation.

Signed-off-by: Anup Patel <anup.patel@wdc.com>

lib: Simplify the for-loop in sbi_ipi_send_many()

We don't need to separately call sbi_ipi_send() for current HART
in sbi_ipi_send_many(). Instead, we can simplify the for-loop in
sbi_ipi_send_many() and call sbi_ipi_send() for all HARTs in the
for-loop itself.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>

lib: Use available hart mask for correct hbase value

As per the latest SBI specification, all online harts should receive
IPI if hbase is set to -1.

Set the target mask to all available hart mask if hbase is -1.

Signed-off-by: Atish Patra <atish.patra@wdc.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>

lib: Fix typo in atomic exchange functions

There is a typo in atomic operations code which prevents the
usage of riscv atomic instructions even if it is supported.

Fix the typo.

Signed-off-by: Atish Patra <atish.patra@wdc.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>

platform: qemu: virt: Correct the typo in config.mk

It should be "aligned".

Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>

platform: sifive: fu540: Add 32-bit specific fdt/payload addresses

For testing 32-bit SiFive specific drivers with QEMU riscv32, add
32-bit specific FW_JUMP_FDT_ADDR and FW_PAYLOAD_OFFSET.

Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>

lib: Print interrupt and exception delegation in boot prints

We print MIDELEG and MEDELEG CSRs as part of boot prints so that
boot log shows the interrupts and exceptions delegated to S-mode.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

lib: Don't check MIDELEG and MEDELEG at end of delegate_traps()

The MIDELEG and MEDELEG CSR checks at end of delegate_traps() were
added for initial bring-up on SiFive Unleashed and QEMU. These
checks are not required any more and in-future these checks can
cause failures because some of the MIDELEG/MEDELEG bits will be
hard-wired to 0 or 1.

For related discussion, refer github issue:
https://github.com/riscv/opensbi/issues/157

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>

lib: utils: htif: Fix 32-bit build

When building 32-bit OpenSBI images, we get:

  lib/utils/sys/htif.c: In function '__check_fromhost':
  lib/utils/sys/htif.c:12:31: error: left shift count >= width of type
                                     [-Werror=shift-count-overflow]
   #define HTIF_DATA_MASK  ((1UL << HTIF_DATA_BITS) - 1)
                                 ^~

Fixes: c2f23cc6edd7 ("platform: Add Spike initial support")
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>

Revert "lib: Use __builtin_ctzl() in pmp_get()"

This reverts commit 897b8fbdd92fcfad194417d348b8dad16ab0e17a.

We are seeing compile errors using newlib based GCC cross-toolchain
so we restore back old ctz() implementation.

Signed-off-by: Anup Patel <anup.patel@wdc.com>

doc: thead-c910: Fix doc styles

- make title underline the same length as the title itself
- satisfy the 80 character per line rule as much as possible

Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>