lib: sbi_pmu: Simplify FW counters to reduce memory usage

Currently, we have 32 elements (i.e. SBI_PMU_FW_EVENT_MAX) array of
"struct sbi_pmu_fw_event" for each of 128 possible HARTs
(i.e. SBI_HARTMASK_MAX_BITS).

To reduce memory usage of OpenSBI, we update FW counter implementation
as follows:
1) Remove SBI_PMU_FW_EVENT_MAX
2) Remove "struct sbi_pmu_fw_event"
3) Create per-HART bitmap of XLEN bits to track FW counters
   which are started on each HART
4) Create per-HART uint64_t array to track values of FW
   counters on each HART.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Reviewed-by: Andrew Jones <ajones@ventanamicro.com>
Reviewed-by: Atish Patra <atishp@rivosinc.com>

lib: sbi_pmu: Firmware counters are always 64 bits wide

As-per SBI specification, all firmware counters are always 64 bits
wide so let us update the SBI PMU implementation to reflect this fact.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Reviewed-by: Andrew Jones <ajones@ventanamicro.com>
Reviewed-by: Atish Patra <atishp@rivosinc.com>

lib: sbi_pmu: Replace sbi_pmu_ctr_read() with sbi_pmu_ctr_fw_read()

The "read a firmware counter" SBI call should only work for firmware
counters so let us replace sbi_pmu_ctr_read() with sbi_pmu_ctr_fw_read()
which works only on firmware counters.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Reviewed-by: Andrew Jones <ajones@ventanamicro.com>
Reviewed-by: Atish Patra <atishp@rivosinc.com>

lib: sbi_pmu: Remove "event_idx" member from struct sbi_pmu_fw_event

The "event_idx" member of struct sbi_pmu_fw_event is not used
anywhere so let us remove it.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Reviewed-by: Andrew Jones <ajones@ventanamicro.com>
Reviewed-by: Atish Patra <atishp@rivosinc.com>

include: Remove sideleg and sedeleg

sideleg and sedeleg csrs are not part of riscv isa spec
anymore, these csrs were part of N extension which
is removed from the riscv isa specification.

These commits removed all traces of these csrs from
riscv spec (https://github.com/riscv/riscv-isa-manual) -

commit f8d27f805b65 ("Remove or downgrade more references to N extension (#674)")
commit b6cade07034d ("Remove N extension chapter for now")

Signed-off-by: Rahul Pathak <rpathak@ventanamicro.com>
Reviewed-by: Andrew Jones <ajones@ventanamicro.com>
Reviewed-by: Anup Patel <anup@brainfault.org>

lib: utils: serial: Add Cadence UART driver

Add Cadence UART driver

Signed-off-by: Jun Liang Tan <junliang.tan@linux.starfivetech.com>
Signed-off-by: Wei Liang Lim <weiliang.lim@linux.starfivetech.com>
Reviewed-by: Andrew Jones <ajones@ventanamicro.com>
Reviewed-by: Anup Patel <anup@brainfault.org>

lib: sbi: Use the official extension name for AIA M-mode CSRs

The arch review of AIA spec is completed and we now have official
extension names for AIA: Smaia (M-mode AIA CSRs) and Ssaia (S-mode
AIA CSRs).

Refer, section 1.6 of the latest AIA v0.3.1 stable specification at
https://github.com/riscv/riscv-aia/releases/download/0.3.1-draft.32/riscv-interrupts-032.pdf)

Based on above, we update generic library to use "Smaia" extension
name for AIA M-mode CSRs.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Reviewed-by: Andrew Jones <ajones@ventanamicro.com>
Reviewed-by: Atish Patra <atishp@rivosinc.com>

lib: sbi_illegal_insn: Fix FENCE.TSO emulation infinite trap loop

In case of missing "FENCE.TSO" instruction implementation,
opensbi can emulate the "FENCE.TSO" with "FENCE RW,RW", but
mepc was not incremented to continue from the next instruction
causing infinite trap.

Fixes: cb8271c8 ("lib: sbi_illegal_insn: Add emulation for fence.tso")
Signed-off-by: Rahul Pathak <rpathak@ventanamicro.com>
Reviewed-by: Andrew Jones <ajones@ventanamicro.com>
Reviewed-by: Xiang W <wxjstz@126.com>
Reviewed-by: Samuel Holland <samuel@sholland.org>

lib: sbi_trap: Set hstatus.GVA when going to HS-mode

The privileged spec specifies that on a trap to HS-mode, hstatus.GVA
should be set to 1 if stval is written with a guest virtual address, and
to 0 otherwise. Implement this by setting hstatus.GVA to trap->gva when
redirecting traps to HS-mode.

Signed-off-by: Vivian Wang <dramforever@live.com>
Reviewed-by: Andrew Jones <ajones@ventanamicro.com>
Reviewed-by: Anup Patel <anup@brainfault.org>

lib: sbi_trap: Set hypervisor CSRs for HS-mode

The hypervisor CSRs hstatus, htval, htinst should always be set if the
trap is to be taken in HS-mode, regardless of which mode it came from.

Signed-off-by: Vivian Wang <dramforever@live.com>
Reviewed-by: Andrew Jones <ajones@ventanamicro.com>
Reviewed-by: Anup Patel <anup@brainfault.org>

lib: sbi: Set gva when creating sbi_trap_info

In some cases the sbi_trap_info argument passed to sbi_trap_redirect is
created from scratch by filling its fields. Since we previously added a
gva field to struct sbi_trap_info, initialize gva in these cases also.

Suggested-by: Andrew Jones <ajones@ventanamicro.com>
Signed-off-by: Vivian Wang <dramforever@live.com>
Reviewed-by: Andrew Jones <ajones@ventanamicro.com>
Reviewed-by: Anup Patel <anup@brainfault.org>

lib: sbi_trap: Save mstatus[h].GVA in trap->gva

The machine mode GVA field is available if the hypervisor extension is
implemented, and indicates if mtval is a guest virtual address. Add a
gva field to sbi_trap_info for this, and in __sbi_expected_trap_hext,
save mstatus[h].GVA to it, so that gva indicates if tval is a guest
virtual address. If the hypervisor extension is not implemented, always
set gva to 0.

Signed-off-by: Vivian Wang <dramforever@live.com>
Reviewed-by: Andrew Jones <ajones@ventanamicro.com>
Reviewed-by: Anup Patel <anup@brainfault.org>

include: Add mstatus[h].GVA encodings

The machine mode GVA field is in mstatus for RV64 and mstatush for RV32,
and is available if the hypervisor extension is available. If an
exception occurs, we may need to redirect the trap to HS-mode, in which
case, hstatus.GVA should be set to same as the machine mode GVA bit.

Add MSTATUS_GVA for RV64, MSTATUSH_GVA for RV32, and their SHIFT
encodings. The SHIFT index is helpful in assembly code, since field
extraction can be implemented in only one register. In pseudocode:

- For RV32: gva = (mstatus >> MSTATUS_GVA_SHIFT) & 1;
- For RV64: gva = (mstatush >> MSTATUSH_GVA_SHIFT) & 1;

Signed-off-by: Vivian Wang <dramforever@live.com>
Reviewed-by: Andrew Jones <ajones@ventanamicro.com>
Reviewed-by: Anup Patel <anup@brainfault.org>

Makefile: Fix typo related to object.mk

The "object.mk" name referred in top-level makefile should be
"objects.mk".

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Reviewed-by: Andrew Jones <ajones@ventanamicro.com>
Tested-by: Andrew Jones <ajones@ventanamicro.com>
Acked-by: Atish Patra <atishp@rivosinc.com>
Tested-by: Atish Patra <atishp@rivosinc.com>

docs: Update documentation for kconfig support

We update all documentation files to:
1) Remove references to platform specific config.mk file since it is
   has been removed.
2) Add details about platform specific configs/defconfig and Kconfig
   files mandatory for each platform.
3) Add required packages in top-level README.md
4) Fix typo releated to object.mk in docs/platform/platform.md

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Reviewed-by: Andrew Jones <ajones@ventanamicro.com>
Tested-by: Andrew Jones <ajones@ventanamicro.com>
Acked-by: Atish Patra <atishp@rivosinc.com>
Tested-by: Atish Patra <atishp@rivosinc.com>

platform: Remove redundant config.mk from all platforms

The options defined in config.mk can be specified in objects.mk of each
platform so let us remove config.mk from all platforms.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Tested-by: Andrew Jones <ajones@ventanamicro.com>
Acked-by: Atish Patra <atishp@rivosinc.com>
Tested-by: Atish Patra <atishp@rivosinc.com>

platform: generic: Use kconfig to set platform version and default name

The generic platform version and default name should be set based
on kconfig options so that users can override it.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Reviewed-by: Andrew Jones <ajones@ventanamicro.com>
Tested-by: Andrew Jones <ajones@ventanamicro.com>
Acked-by: Atish Patra <atishp@rivosinc.com>
Tested-by: Atish Patra <atishp@rivosinc.com>

platform: generic: Use kconfig for enabling/disabling overrides

We update generic platform to use kconfig for enabling/disabling
platform overrides. We also enable all platform overrides in generic
platform defconfig.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Tested-by: Andrew Jones <ajones@ventanamicro.com>
Acked-by: Atish Patra <atishp@rivosinc.com>
Tested-by: Atish Patra <atishp@rivosinc.com>

lib: utils/fdt: Use kconfig for enabling/disabling

We update FDT support makefile to use kconfig for enabling/disabling.
To avoid compilation errors, we also enable FDT for each platform.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Tested-by: Andrew Jones <ajones@ventanamicro.com>
Acked-by: Atish Patra <atishp@rivosinc.com>
Tested-by: Atish Patra <atishp@rivosinc.com>

lib: utils/gpio: Use kconfig for enabling/disabling drivers

We update gpio drivers makefile to use kconfig for enabling/disabling
drivers. To avoid compile errors, we also enable appropriate gpio
drivers for each platform.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Tested-by: Andrew Jones <ajones@ventanamicro.com>
Acked-by: Atish Patra <atishp@rivosinc.com>
Tested-by: Atish Patra <atishp@rivosinc.com>

lib: utils/i2c: Use kconfig for enabling/disabling drivers

We update i2c drivers makefile to use kconfig for enabling/disabling
drivers. To avoid compile errors, we also enable appropriate i2c
drivers for each platform.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Tested-by: Andrew Jones <ajones@ventanamicro.com>
Acked-by: Atish Patra <atishp@rivosinc.com>
Tested-by: Atish Patra <atishp@rivosinc.com>

lib: utils/irqchip: Use kconfig for enabling/disabling drivers

We update irqchip drivers makefile to use kconfig for enabling/disabling
drivers. To avoid compile errors, we also enable appropriate irqchip
drivers for each platform.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Tested-by: Andrew Jones <ajones@ventanamicro.com>
Acked-by: Atish Patra <atishp@rivosinc.com>
Tested-by: Atish Patra <atishp@rivosinc.com>

lib: utils/ipi: Use kconfig for enabling/disabling drivers

We update ipi drivers makefile to use kconfig for enabling/disabling
drivers. To avoid compile errors, we also enable appropriate ipi
drivers for each platform.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Tested-by: Andrew Jones <ajones@ventanamicro.com>
Acked-by: Atish Patra <atishp@rivosinc.com>
Tested-by: Atish Patra <atishp@rivosinc.com>

lib: utils/timer: Use kconfig for enabling/disabling drivers

We update timer drivers makefile to use kconfig for enabling/disabling
drivers. To avoid compile errors, we also enable appropriate timer
drivers for each platform.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Tested-by: Andrew Jones <ajones@ventanamicro.com>
Acked-by: Atish Patra <atishp@rivosinc.com>
Tested-by: Atish Patra <atishp@rivosinc.com>

lib: utils/sys: Use kconfig for enabling/disabling drivers

We update system drivers makefile to use kconfig for enabling/disabling
drivers. To avoid compile errors, we also enable appropriate system
drivers for each platform.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Tested-by: Andrew Jones <ajones@ventanamicro.com>
Acked-by: Atish Patra <atishp@rivosinc.com>
Tested-by: Atish Patra <atishp@rivosinc.com>

lib: utils/reset: Use kconfig for enabling/disabling drivers

We update reset drivers makefile to use kconfig for enabling/disabling
drivers. To avoid compile errors, we also enable appropriate reset
drivers for each platform.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Tested-by: Andrew Jones <ajones@ventanamicro.com>
Acked-by: Atish Patra <atishp@rivosinc.com>
Tested-by: Atish Patra <atishp@rivosinc.com>

lib: utils/serial: Use kconfig for enabling/disabling drivers

We update serial drivers makefile to use kconfig for enabling/disabling
drivers. To avoid compile errors, we also enable appropriate serial
drivers for each platform.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Tested-by: Andrew Jones <ajones@ventanamicro.com>
Acked-by: Atish Patra <atishp@rivosinc.com>
Tested-by: Atish Patra <atishp@rivosinc.com>

Makefile: Compile lib/utils sources separately for each platform

Currently, if same build directory is used to compile two different
platforms then lib/utils objects are shared for these platforms.

We will be having platform specific configs to enable/disable drivers
in lib/utils and select compile time options for lib/utils sources.
This means lib/utils sources will now be compiled in a platform
specific way.

To tackle above, we update top-level Makefile as follows:
1) Don't create libsbiutils.a anymore because this can't be shared
   between platforms.
2) Compile lib/utils sources separately for each platform.
3) Add comments showing which make rules are for lib/sbi, lib/utils,
   firmware, and platform sources.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Tested-by: Andrew Jones <ajones@ventanamicro.com>
Acked-by: Atish Patra <atishp@rivosinc.com>
Tested-by: Atish Patra <atishp@rivosinc.com>

Makefile: Add initial kconfig support for each platform

We extend the top-level makefile to allow kconfig based configuration
for each platform where each platform has it's own set of configs with
"defconfig" being the default config.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Tested-by: Andrew Jones <ajones@ventanamicro.com>
Acked-by: Atish Patra <atishp@rivosinc.com>
Tested-by: Atish Patra <atishp@rivosinc.com>

scripts: Add Kconfiglib v14.1.0 under scripts directory

We adopt Kconfiglib v14.1.0 sources under scripts directory so that
top-level OpenSBI makefile can directly use Kconfiglib scripts without
expecting users to install a particular version of Kconfiglib on their
build system.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Tested-by: Andrew Jones <ajones@ventanamicro.com>
Acked-by: Atish Patra <atishp@rivosinc.com>
Tested-by: Atish Patra <atishp@rivosinc.com>

lib: sbi: Fix printf handling of long long

Read long long arguments directly using va_arg. Remove original hack for
RV32 that read a long long arg as two long args.

This un-breaks the case on RV64 where e.g. the long long is followed by
an odd number of ints:

    sbi_printf("%d %lld", (int) 1, (long long) 2LL);

Also remove the acnt variable, which is now unused.

Signed-off-by: dramforever <dramforever@live.com>
Reviewed-by: Andrew Jones <ajones@ventanamicro.com>
Reviewed-by: Anup Patel <anup@brainfault.org>

lib: utils/serial: Ensure baudrate is non-zero before using

RISC-V doesn't generate exceptions on divide-by-zero, but the result,
all bits set, is not likely what people expect either. In all cases
where we divide by baudrate there's a chance it's zero (when the DT
it came from is "bad"). To avoid difficult to debug situations, leave
baudrate dependent registers alone when baudrate is zero, as, also in
all cases, it appears we can skip initialization of those registers
and still [hopefully] have a functioning UART.

Signed-off-by: Andrew Jones <ajones@ventanamicro.com>
Reviewed-by: Anup Patel <anup@brainfault.org>

lib: serial: Clean up coding style in sifive-uart.c

Signed-off-by: Andrew Jones <ajones@ventanamicro.com>
Reviewed-by: Xiang W <wxjstz@126.com>
Reviewed-by: Anup Patel <anup@brainfault.org>

lib: utils/serial: Initialize platform_uart_data to zero

While it doesn't look like there are any current cases of using
uninitialized data, let's zero all the UART data members to be
safe. Zero may not actually be better than a random number in
some cases, so all structure members should still be validated
before use, but at least zero is usually easier to debug than
some random stack garbage...

Signed-off-by: Andrew Jones <ajones@ventanamicro.com>
Reviewed-by: Anup Patel <anup@brainfault.org>

lib: utils/fdt: Factor out common uart node code

Factor out the common code used by the fdt UART node parsers,
allowing us to drop duplicate code.

Signed-off-by: Andrew Jones <ajones@ventanamicro.com>
Reviewed-by: Anup Patel <anup@brainfault.org>

lib: sbi: Fix fw_event_map initialization

fw_event_map represents array of firmware events. It should initialized
for maximum number of firmware events not counters.

Reviewed-by: Andrew Jones <ajones@ventanamicro.com>
Signed-off-by: Atish Patra <atishp@rivosinc.com>
Reviewed-by: Anup Patel <anup@brainfault.org>

lib: sbi: Fix possible buffer overrun in counter validation

The active_events array is accessed with counter ID passed from the supervisor
software before the counter ID bound check. This may cause a buffer overrun
if a supervisor passes an invalid counter ID.

Fix this by moving the access part after the bound check.

Reported-by: Andrew Jones <ajones@ventanamicro.com>
Reviewed-by: Andrew Jones <ajones@ventanamicro.com>
Signed-off-by: Atish Patra <atishp@rivosinc.com>
Reviewed-by: Anup Patel <anup@brainfault.org>

lib: sbi: Add the bound check for events during config match

Currently, there is no sanity check for firmware event code. We don't see
any issue as Linux kernel driver does a bound check on firmware events
already. However, OpenSBI can't assume sane supervisor mode software
always. Thus, an invalid event idx can cause a buffer overflow error.
For hardware events, the match will fail for invalid event code anyways.
However, a search is unecessary if event code is invalid.

Add a event ID validation function to solve the issue.

Signed-off-by: Atish Patra <atishp@rivosinc.com>
Reviewed-by: Andrew Jones <ajones@ventanamicro.com>
Reviewed-by: Anup Patel <anup@brainfault.org>

lib: sbi: Fix counter index sanity check

The current implementation computes the possible counter range
by doing a left shift of counter base. However, this may overflow depending
on the counter base value. In case of overflow, the highest counter id
may be computed incorrectly. As per the SBI specification, the respective
function should return an error if any of the counter is not valid.

Fix the counter index check by avoiding left shifting while doing the
sanity checks. Without the shift, the implementation just iterates
over the counter mask and computes the correct counter index by adding
the base to it.

Reviewed-by: Andrew Jones <ajones@ventanamicro.com>
Signed-off-by: Atish Patra <atishp@rivosinc.com>
Reviewed-by: Anup Patel <anup@brainfault.org>

lib: sbi: Fix sbi_snprintf

printc would happily write to 'out' even when 'out_len' was zero,
potentially overflowing buffers. Rework printc to not do that and
also ensure the null byte is written at the last position when
necessary, as stated in the snprintf man page. Also, panic if
sprintf or snprintf are called with NULL output strings (except
the special case of snprintf having a NULL output string and
a zero output size, allowing it to be used to get the number of
characters that would have been written). Finally, rename a
goto label which clashed with 'out'.

Fixes: 9e8ff05cb61f ("Initial commit.")
Signed-off-by: Andrew Jones <ajones@ventanamicro.com>
Reviewed-by: Anup Patel <anup@brainfault.org>

lib: utils/serial: Update Shakti UART based on latest implementation

The age old version of Shakti UART was upgraded long back, but we missed
updating the driver in OpenSBI. The old version of UART is not supported
anymore, hence removed the inline comment which is also outdated now.

Signed-off-by: Prasanna T <ptprasanna@gmail.com>
Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Reviewed-by: Xiang W <wxjstz@126.com>

lib: sbi_hart: Shorten the code to set MPV bit

MPV bit is set when the value of next_virt boolean variable equals
true. Since the value of next_virt is either 0 or 1, we can set
MPV bit without if-else logic.

Signed-off-by: Che-Chia Chang <alvinga@andestech.com>
Reviewed-by: Andrew Jones <ajones@ventanamicro.com>
Reviewed-by: Anup Patel <anup@brainfault.org>

lib: sbi: add check for ipi device for hsm start

If the ecall SBI_EXT_HSM_HART_START is called it might try to wake the
secondary hart using sbi_ipi_raw_send() to send an IPI to the hart.
This can fail if there is no IPI device but no error is returned from
sbi_ipi_raw_send() so the ecall returns as if the action completed and
the caller continues without noticing (in the case of Linux it just hangs
waiting for the secondary hart to become active)

Fix this by changing sbi_ipi_raw_send() to return and error, and if an
error is returned, then return it via SBI_EXT_HSM_HART_START call.

Signed-off-by: Ben Dooks <ben.dooks@sifive.com>
Reviewed-by: Anup Patel <anup@brainfault.org>

lib: sbi_timer: Added a conditional wait function which can timeout

Motivation: Suppose a peripheral needs to be configured to transmit
data. There is an SFR bit which indicates that the peripheral is ready
to transmit. The firmware should check the bit and will only transmit
data when the peripheral is ready. When the firmware starts polling the
SFR, the peripheral could be busy transmitting/receiving other data so
the firmware must wait till that completes. Assuming that there is no
other way, the firmware shouldn't wait indefinitely.

The function sbi_timer_waitms_until() will constantly check whether a
certain condition is satisfied, or timeout occurs. It can be used for
the cases when a timeout is required.

Signed-off-by: Adnan Rahman Chowdhury <adnan.chowdhury@sifive.com>
Reviewed-by: Xiang W <wxjstz@126.com>
Reviewed-by: Anup Patel <anup@brainfault.org>

include: Bump-up version to 1.1

This patch updates OpenSBI version to 1.1 as part of
release preparation.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>

lib: pmu: allow to use the highest available counter

OpenSBI explicitly assumes that there is no pmu hardware counter with
index 1: hardware uses that bit for TM control. So OpenSBI filters
out that index in sanity checks. However OpenSBI also excludes that
counter when reports total amount of hardware counters to Linux. As
a result, Linux uses incomplete counters mask excluding the highest
available counter.

Return accurate number of counters, update the firmware counter
starting index, fix range checks that include num_hw_ctrs.

The simple test is to make sure that there is no counter multiplexing
in the following command:

$ perf stat -e \
r8000000000000000,r8000000000000001,r8000000000000002,r8000000000000003, \
r8000000000000004,r8000000000000005,r8000000000000006,r8000000000000007, \
r8000000000000008,r8000000000000009,r800000000000000a,r800000000000000b, \
r800000000000000c,r800000000000000d,r800000000000000e,r800000000000000f  \
ls

Note that 16 firmware events with 16 counters won't require multiplexing.

Signed-off-by: Sergey Matyukevich <geomatsi@gmail.com>
Signed-off-by: Atish Patra <atishp@rivosinc.com>
Reviewed-by: Atish Patra <atishp@rivosinc.com>

include: sbi: Change spec version to 1.0

Now that SBI v1.0 specification is ratified, we change spec verion
implemented by OpenSBI to v1.0.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Reviewed-by: Atish Patra <atishp@rivosinc.com>

lib: utils/timer: Remove Allwinner D1 CLINT compatibles

The allwinner,sun20i-d1-clint compatible string is not documented in any
official binding, so it should not be used by drivers.

The MSWI in the D1 CLINT is compatible with the ACLINT specification, so
it can take advantage of generic driver support. However, that is only
possible if the MSWI and MTIMER are split into separate DT nodes. This
means the final binding for this device is likely to be incompatible
with what is implemented here.

Remove this compatible string from the driver to prevent it from
appearing in a stable version and causing future issues.

Signed-off-by: Samuel Holland <samuel@sholland.org>
Reviewed-by: Anup Patel <anup@brainfault.org>

lib: irqchip/plic: fix typo in plic_warm_irqchip_init

The second invocation of plic_context_init() incorrectly calls the
function with m_cntx_id instead of s_cntx_id. This breaks systems which
only have 1 external interrupt per hart.

Fixes: 8c362e7 ("lib: irqchip/plic: Factor out a context init function")
Signed-off-by: Jan Remes <jan.remes@codasip.com>
Reviewed-by: Anup Patel <anup@brainfault.org>
Reviewed-by: Samuel Holland <samuel@sholland.org>

lib: utils/timer: Add a separate compatible for the D1 CLINT

The CLINT in the Allwinner D1 SoC apparently does not support 64-bit
MMIO access. A property was added to support this quirk (and that
property was copied to the ACLINT MTIMER code). However, since this
difference in behavior makes the D1 CLINT incompatible with the SiFive
CLINT's programming interface, a better solution is to use a separate
compatible string.

Signed-off-by: Samuel Holland <samuel@sholland.org>
Reviewed-by: Anup Patel <anup@brainfault.org>

lib: utils: Remove CSRs that set/clear an IMSIC interrupt file bits

Based on architecture review committee feedback, the [m|s|vs]seteienum,
[m|s|vs]clreienum, [m|s|vs]seteipnum, and [m|s|vs]clreipnum CSRs are
removed in the latest AIA draft v0.3.0 specification.
(Refer, https://github.com/riscv/riscv-aia/releases/tag/0.3.0-draft.31)

These CSRs were mostly for software convenience and software can always
use [m|s|vs]iselect and [m|s|vs]ireg CSRs to update the IMSIC interrupt
file bits.

We update the IMSIC programming as-per above to match the latest AIA
draft specification.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Reviewed-by: Atish Patra <atishp@rivosinc.com>

lib: sbi: Fix tval and tinst for sbi_get_insn()

We should not change trap->tval to mepc because mtval already points to
the faulting portion of the emulated instruction fetch, which is also
what stval is expected to be.

In addition, htinst is only allowed to be zero for instruction access
faults or page faults, and is only allowed to be zero or a
psuedoinstruction for instruction guest-page faults. Fix trap->tinst for
these cases.

Signed-off-by: dramforever <dramforever@live.com>
Reviewed-by: Anup Patel <anup@brainfault.org>

lib: sbi: Fixup tinst for exceptions in sbi_misaligned_*()

If there is an exception while emulating a misaligned load/store, fixup
uptrap.tinst before redirecting. Otherwise, HS-mode software may receive
an htinst describing the lbu/sb instruction that faulted during
emulation[1].

[1]: https://github.com/riscv-software-src/opensbi/issues/258

Signed-off-by: dramforever <dramforever@live.com>
Reviewed-by: Anup Patel <anup@brainfault.org>

include: sbi: Add mtinst/htinst psuedoinstructions

Add psuedoinstruction encodings written to mtinst/htinst for faults
caused by implicit memory access for VS-stage address translation

Signed-off-by: dramforever <dramforever@live.com>
Reviewed-by: Anup Patel <anup@brainfault.org>

platform: Add HSM implementation for Allwinner D1

Allwinner D1 contains a "PPU" power domain controller which can
automatically power down/up the CPU power domain. This power domain
includes the C906 core along with its CLINT and PLIC.

This HSM implementation supports non-retentive hart suspend by:
  1) Saving/restoring state that is lost during hart suspend,
  2) Performing cache maintenance before/after hart suspend,
  3) Configuring wakeup sources before hart suspend, and
  4) Asking the PPU to power down the hart when it enters WFI.

Since this HSM implementation is for a single-core SoC, it does not need
to worry about concurrency or saving multiple instances of state.

Reviewed-by: Anup Patel <anup@brainfault.org>
Signed-off-by: Samuel Holland <samuel@sholland.org>

lib: utils/irqchip: Add wrapper for T-HEAD PLIC delegation

The delegation bit is lost along with the rest of the PLIC state when
the CPU power domain in the Allwinner D1 is powered down, so the PLIC
needs to be re-delegated to S-mode during the hart resume path.

Reviewed-by: Anup Patel <anup@brainfault.org>
Signed-off-by: Samuel Holland <samuel@sholland.org>

lib: utils/irqchip: Add FDT wrappers for PLIC save/restore functions

These functions save/restore the state of the PLIC associated with the
current hart. The context save/restore functions only manipulate a
single context, since most likely the M-mode context is unused and does
not need to be saved.

Signed-off-by: Samuel Holland <samuel@sholland.org>
Reviewed-by: Anup Patel <anup@brainfault.org>

lib: irqchip/plic: Add priority save/restore helpers

These can be used by platform code to save the PLIC priority state, if
it would otherwise be lost during non-retentive suspend. The platform
is responsible for allocating all necessary storage.

As a space optimization, store the saved priority values as 8-bit
integers, since that is large enough to hold any priority value on the
relevant platforms.

Reviewed-by: Anup Patel <anup@brainfault.org>
Signed-off-by: Samuel Holland <samuel@sholland.org>

lib: irqchip/plic: Add context save/restore helpers

These can be used by platform code to save the PLIC context state, if
it would otherwise be lost during non-retentive suspend. The platform
is responsible for allocating all necessary storage.

Reviewed-by: Anup Patel <anup@brainfault.org>
Signed-off-by: Samuel Holland <samuel@sholland.org>

lib: irqchip/plic: Factor out a context init function

This simplifies both the callers and the callees by removing duplicated
code and consolidating the error handling. It also fixes two bugs in the
process:
  1) ie_words was one too large when plic->num_src was a multiple of 32.
  2) plic_set_ie takes a 32-bit mask, not a Boolean value, so the FPGA
     platforms previously only enabled one out of every 32 interrupts.

Reviewed-by: Anup Patel <anup@brainfault.org>
Signed-off-by: Samuel Holland <samuel@sholland.org>

lib: irqchip/plic: Constify plic_data pointers

None of the functions modify the passed-in plic_data, so mark it const.

Reviewed-by: Anup Patel <anup@brainfault.org>
Signed-off-by: Samuel Holland <samuel@sholland.org>

lib: sbi_hsm: Assume a consistent resume address

The suspend code needs to know the resume address for two reasons:
  1) Programming some hardware register or management firmware. Here we
     assume the hardware/firmware maintains its state between suspends,
     so it only needs to be programmed once at startup.
  2) When a non-retentive suspend request ends up being retentive, due
     to lack of hardware support, pending interrupt, or for some other
     reason. However, the behavior here is not platform-dependent, and
     this can be handled in the generic hart suspend function.

Since neither situation requires the platform-level suspend function to
know the resume address, stop passing it to that function. Instead,
handle the non-retentive to retentive situation generically.

Reviewed-by: Anup Patel <anup@brainfault.org>
Signed-off-by: Samuel Holland <samuel@sholland.org>

lib: sbi_hsm: Call a device hook during hart resume

Non-retentive suspend states may require platform-specific actions
during resume. For example, firmware may need to save and restore the
values of custom CSRs. Add a hook to support this.

Reviewed-by: Anup Patel <anup@brainfault.org>
Signed-off-by: Samuel Holland <samuel@sholland.org>

platform: generic: add overrides for vendor extensions

Allow the vendor_ext_check and vendor_ext_provider APIs of the
generic platform to be overridden by other platforms

Signed-off-by: Mayuresh Chitale <mchitale@ventanamicro.com>
Reviewed-by: Anup Patel <anup@brainfault.org>

lib: sbi_trap: Redirect exception based on hedeleg

HS-mode software can choose what exceptions to delegate to VS-mode using
the hedeleg CSR. Synthetic VS/VU-mode exceptions should also honor
hedeleg. They should be redirected to VS-mode if and only if delegated
by HS-mode.

Signed-off-by: dramforever <dramforever@live.com>
Reviewed-by: Xiang W <wxjstz@126.com>
Reviewed-by: Anup Patel <anup@brainfault.org>

lib: sbi_illegal_insn: Add emulation for fence.tso

While OpenC906 appears to properly decode `fence.tso` as a fence
instruction[1], the version of the C906 taped out in the Allwinner D1
does not, and raises illegal instruction.

Handle this errata by emulating `fence.tso` as `fence rw, rw`.

[1]: https://github.com/T-head-Semi/openc906/blob/30827e7f/C906_RTL_FACTORY/gen_rtl/idu/rtl/aq_idu_id_decd.v#L2097

Signed-off-by: Samuel Holland <samuel@sholland.org>
Reviewed-by: Anup Patel <anup@brainfault.org>

lib: sbi_illegal_insn: Constify illegal_insn_table

This table does not need to be modified at runtime.

Signed-off-by: Samuel Holland <samuel@sholland.org>
Reviewed-by: Anup Patel <anup@brainfault.org>

lib: sbi_timer: Drop unnecessary get_platform_ticks wrapper

The device's timer_value callback is already the right prototype to use
for the get_time_val function pointer.

Signed-off-by: Samuel Holland <samuel@sholland.org>
Reviewed-by: Anup Patel <anup@brainfault.org>

lib: utils/fdt: Require match data to be const

Match data stores hardware attributes which do not change at runtime, so
it does not need to be mutable. Make it const.

Reviewed-by: Guo Ren <guoren@kernel.org>
Reviewed-by: Xiang W <wxjstz@126.com>
Reviewed-by: Anup Patel <anup@brainfault.org>
Signed-off-by: Samuel Holland <samuel@sholland.org>

lib: sbi: fix system_opcode_insn

If the csr's operation comes from M mode, it should not be forwarded
to low-privilege processing, this patch fixes this problem.

Signed-off-by: Xiang W <wxjstz@126.com>
Reviewed-by: Anup Patel <anup@brainfault.org>

lib: utils/fdt: rename fdt_parse_max_hart_id

The function returns the highest hart-id of the harts actually used in
the system (enabled). Change the name to reflect this fact.

Signed-off-by: Jan Remes <jan.remes@codasip.com>
Reviewed-by: Anup Patel <anup@brainfault.org>

platform: generic: check if CPU node is enabled

Ignore CPU nodes in FDT which are not enabled.

Signed-off-by: Jan Remes <jan.remes@codasip.com>
Reviewed-by: Xiang W <wxjstz@126.com>
Reviewed-by: Anup Patel <anup@brainfault.org>

lib: utils: check if CPU node is enabled

Ignore CPU nodes in FDT that are not enabled.

Signed-off-by: Jan Remes <jan.remes@codasip.com>
Reviewed-by: Anup Patel <anup@brainfault.org>

lib: utils/fdt: introduce fdt_node_is_enabled()

If an FDT node contains a "status" property and this property is not
"ok" or "okay", this node should be ignored. Introduce a function that
checks this.

Signed-off-by: Jan Remes <jan.remes@codasip.com>
Reviewed-by: Xiang W <wxjstz@126.com>
Reviewed-by: Anup Patel <anup@brainfault.org>

lib: utils/irqchip: fix typo when checking for CPU node

Fix typo in irqchip_imsic_update_hartid_table() when checking for CPU
node.

Signed-off-by: Jan Remes <jan.remes@codasip.com>
Reviewed-by: Anup Patel <anup@brainfault.org>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>

platform: generic: Move Sifive platform overrides into own directory

Let us move SiFive platform overrides for FU540 and FU740 into a separate
directory so better maintainability. Other SoC vendors can also create
their own directory under platform/generic.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Reviewed-by: Atish Patra <atishp@rivosinc.com>

platform: generic: Generate platform override module list at compile-time

Instead of having platform override module list hard-coded in the C source,
we generate it using carray.sh at compile-time.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Reviewed-by: Atish Patra <atishp@rivosinc.com>

lib: utils/gpio: Generate FDT gpio driver list at compile-time

Instead of having FDT gpio driver list hard-coded in the C source,
we generate it using carray.sh at compile-time.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Reviewed-by: Atish Patra <atishp@rivosinc.com>

lib: utils/i2c: Generate FDT i2c adapter driver list at compile-time

Instead of having FDT i2c adapter driver list hard-coded in the C source,
we generate it using carray.sh at compile-time.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Reviewed-by: Atish Patra <atishp@rivosinc.com>

lib: utils/ipi: Generate FDT ipi driver list at compile-time

Instead of having FDT ipi driver list hard-coded in the C source,
we generate it using carray.sh at compile-time.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Reviewed-by: Atish Patra <atishp@rivosinc.com>

lib: utils/irqchip: Generate FDT irqchip driver list at compile-time

Instead of having FDT irqchip driver list hard-coded in the C source,
we generate it using carray.sh at compile-time.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Reviewed-by: Atish Patra <atishp@rivosinc.com>

lib: utils/timer: Generate FDT timer driver list at compile-time

Instead of having FDT timer driver list hard-coded in the C source,
we generate it using carray.sh at compile-time.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Reviewed-by: Atish Patra <atishp@rivosinc.com>

lib: utils/serial: Generate FDT serial driver list at compile-time

Instead of having FDT serial driver list hard-coded in the C source,
we generate it using carray.sh at compile-time.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Reviewed-by: Atish Patra <atishp@rivosinc.com>

lib: utils/reset: Generate FDT reset driver list at compile-time

Instead of having FDT reset driver list hard-coded in the C source,
we generate it using carray.sh at compile-time.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Reviewed-by: Atish Patra <atishp@rivosinc.com>

Makefile: Add support for generating C array at compile time

Generating C array at compile time based on details provided by
objects.mk is a very useful feature which will help us compile
only a subset of drivers or modules.

We add a bash script (carray.sh) which takes array details and
object/variable list from command-line to generate a C source
containing array of object/variable pointers. We also extend
top-level makefile to use carray.sh whenever specified through
objects.mk.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Reviewed-by: Atish Patra <atishp@rivosinc.com>

Makefile: Allow generated C source to be anywhere in build directory

The generated C source could be anywhere within build directory so
let us update the make rule to comple generated C source accordingly.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Reviewed-by: Atish Patra <atishp@rivosinc.com>

lib: sbi_platform: Add callback to populate HART extensions

We add platform specific extensions_init() callback which allows
platforms to populate HART extensions for each HART. For example,
the generic platform can populate HART extensions from HART ISA
string described in DeviceTree.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Reviewed-by: Atish Patra <atishp@rivosinc.com>

lib: sbi: Add sbi_hart_update_extension() function

We add sbi_hart_update_extension() function which allow platforms
to enable/disable hart extensions.

Signed-off-by: Anup Patel <anup@brainfault.org>
Reviewed-by: Atish Patra <atishp@rivosinc.com>

lib: sbi: Detect hart features only once for each hart

Currently, the hart_detect_features() is called everytime a hart
is stopped and started again which is unnecessary work.

We update hart_detect_features() to detect hart features only
once for each hart.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Reviewed-by: Atish Patra <atishp@rivosinc.com>

lib: sbi: Convert hart features into hart extensions

Since past few years, we have been using "hart features" in OpenSBI
to represent all optionalities and multi-letter extensions defined
by the RISC-V specifications.

The RISC-V profiles specification has taken a different approach and
started assigning extension names for all optionalities which did not
have any extension name previously.
(Refer, https://github.com/riscv/riscv-profiles/blob/main/profiles.adoc)

Inspired from the RISC-V profiles specification, we convert OpenSBI
hart features into hart extensions. Going forward, we align the
extension naming with RISC-V profiles specification. Currently, only
"time CSR" and "AIA CSR" have not been assigned extension name but
for everything else we have a name.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Reviewed-by: Atish Patra <atishp@rivosinc.com>

lib: sbi: Fix AIA feature detection

The AIA feature detection uses unnecessary goto which is not need
and AIA case in sbi_hart_feature_id2string() does not break. This
patch fixes both issues in AIA feature detection.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Reviewed-by: Atish Patra <atishp@rivosinc.com>

lib: sbi: Remove MENVCFG hart feature

If a hart implements privileged spec v1.12 (or higher) then we can
safely assume that menvcfg CSR is present and we don't need MENVCFG
as a hart feature.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Reviewed-by: Atish Patra <atishp@rivosinc.com>

lib: sbi: Remove MCOUNTINHIBT hart feature

If a hart implements privileged spec v1.11 (or higher) then we can
safely assume that mcountinhibit CSR is present and we don't need
MCOUNTINHIBT as a hart feature.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Reviewed-by: Atish Patra <atishp@rivosinc.com>

lib: sbi: Remove MCOUNTEREN and SCOUNTEREN hart features

If a hart implements privileged spec v1.10 (or higher) then we can
safely assume that [m|s]counteren CSR are present and we don't need
MCOUNTEREN and SCOUNTEREN as hart features.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Reviewed-by: Atish Patra <atishp@rivosinc.com>

lib: sbi: Update the name of ISA string printed at boot time

The ISA string printed at boot time is not the complete ISA string
representing all single letter and multi-letter extensions rather
it is base ISA string derived from misa CSR so let us update the
boot print accordingly.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Reviewed-by: Atish Patra <atishp@rivosinc.com>

lib: sbi: Remove 's' and 'u' from misa_string() output

Both 's' and 'u' are not treated as ISA extensions since these are
privilege modes so let's remove it from misa_string() output.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Reviewed-by: Atish Patra <atishp@rivosinc.com>

lib: sbi: Detect and print privileged spec version

It is possible to guess privileged spec versions based on the CSRs
that where introduced in different privileged spec versions. In
future, if we are not able guess privileged spec version then we
can have platform provide it.

We add privileged spec version as per-hart feature and try to guess
it based on presence of mcounteren, mcountinhibit, and menvcfg CSRs.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Reviewed-by: Atish Patra <atishp@rivosinc.com>

lib: sbi: Fix mhpmeventh access for rv32 in absence of sscofpmf

MHPMEVENT3H-31H are defined in sscofpmf extension. Thus, they should be
accessed only if sscofpmf is present.

Signed-off-by: Atish Patra <atishp@rivosinc.com>
Reviewed-by: Anup Patel <anup@brainfault.org>

include: sbi: Define SBI_PMU_HW_EVENT_MAX to 256

Increase maximum number of PMU hardware events that can be mapped
by OpenSBI to 256

Signed-off-by: Jun Liang Tan <junliang.tan@linux.starfivetech.com>
Signed-off-by: Wei Liang Lim <weiliang.lim@linux.starfivetech.com>
Reviewed-by: Anup Patel <anup@brainfault.org>

lib: sbi: Fix mstatus_init() for RV32 when Sscofpmf is not available

The mhpmevent3h to mhpmevent31h CSRs are available on RV32 only when
Sscofpmf extension is available so mstatus_init() should set this
CSRs only when Sscofpmf extension is available.

Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Reviewed-by: Atish Patra <atishp@rivosinc.com>