| Age | Commit message (Collapse) | Author |
|---|
|
These C source files should not be marked as executable.
Signed-off-by: Samuel Holland <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
This optimization creates a correctness issue, as it prevents supervisor
software from restricting VS-mode or U-mode access to the time CSR for
its own purposes.
Closes: https://github.com/riscv-software-src/opensbi/issues/370
Fixes: ebc8ebc0f846 ("lib: sbi: Improve HPM CSR read/write emulation")
Signed-off-by: Samuel Holland <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
Add support for controlling the pointer masking mode on harts which
support the Smnpm extension. This extension can only exist on harts
where XLEN >= 64 bits. This implementation selects the mode with the
smallest PMLEN that satisfies the caller's requested lower bound.
Signed-off-by: Samuel Holland <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
Writes to the low half CSR should not affect the high half of the value.
Make this separation explicit by writing to the delta in memory as two
adjacent XLEN-sized values.
Fixes: 1e9f88889f8b ("lib: Emulate HTIMEDELTA CSR for platforms not having TIME CSR")
Signed-off-by: Samuel Holland <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
This removes the compile-time limit on the number of domains. It also
reduces firmware size by about 200 bytes by removing the lookup table.
Signed-off-by: Samuel Holland <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
In `csr_read_allowed` and `csr_write_allowed` macros, has already
converted second param to `ulong`. So delete redundant `ulong`
where uses csr_read/write_allowed macros.
Signed-off-by: Zhang RunMin <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
This parameters was a remnant of a previous version, remove it now that
it is unused.
Signed-off-by: Clément Léger <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
- Completed TODO in `system_opcode_insn` to ensure CSR read/write
instruction handling.
- Refactored to use new macros `GET_RS1_NUM` and `GET_CSR_NUM`.
- Updated `GET_RM` macro and replaced hardcoded funct3 values with
constants (`CSRRW`, `CSRRS`, `CSRRC`, etc.).
- Removed redundant `GET_RM` from `riscv_fp.h`.
- Improved validation and error handling for CSR instructions.
This patch enhances the clarity and correctness of CSR handling
in `system_opcode_insn`.
Signed-off-by: Dongdong Zhang <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
This change adds a simple implementation of sbi_aligned_alloc(), for future use
in allocating aligned memory for SMMTT tables.
Signed-off-by: Gregor Haas <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
In the next commit, we'll add a new sbi_memalign() function. In order to
allocate aligned memory, we'll sometimes need to allocate from the middle of a
heap block, effectively splitting it in two. Allocating from the beginning of a
heap block in the nonaligned case more closely matches this behavior, reducing
the complexity of understanding the heap implementation.
Signed-off-by: Gregor Haas <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
The upcoming SMMTT implementation will require some larger contiguous memory
regions for the memory tracking tables. We plan to specify the memory region
for these tables as a reserved-memory node in the device tree, and then
dynamically allocate individual tables out of this region. These changes to the
SBI heap allocator will allow us to explicitly create and allocate from a
dedicated heap tied to the table memory region.
Signed-off-by: Gregor Haas <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
pmp_get() may return an error if the given entry, given by the caller
of is_pmp_entry_mapped(), is invalid. This results in the output
parameters for pmp_get() being uninitialized. To avoid using garbage
values, check the result and return early if necessary.
This issue is not being hit because at the moment
is_pmp_entry_mapped() is only being called from a single site with a
valid hardcoded value.
Signed-off-by: Carlos López <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
The fwt_hart_state struct inciludes a flexible array member, so its
allocation size will be that of the struct itself, plus that of each
of the members in the array. When calculating this size, instead of
taking the size of the struct, the size of a pointer to it was taken,
which is incorrect. Luckily, this happenned to not produce memory
corruption because the size of the non-flexible members of the struct
is the same as the size of a pointer.
Signed-off-by: Carlos López <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
In several dbtr functions, we first check that the dbtr trigger is not
NULL and that its state is what we expect. However, it only makes
sense to perform the second check if the dbtr trigger is not NULL.
Othwerwise we will dereference a NULL pointer. Thus, change the
condition so that it shortcuts to the first check if necessary.
Signed-off-by: Carlos López <[email protected]>
Reviewed-By: Anup Patel <[email protected]>
|
|
The console device is registered by platform only in early_init()
callback so any prints before this point will be lost. Introduce an
early console buffer for caching prints before platform early_init().
For crashes before platform early_init(), users can simply dump the
contents of the console_early_buffer[] string using a debugger. The
relative address of the console_early_buffer[] string can be found
using following two commands:
CONSOLE_EARLY_FIFO_ADDR=`${CROSS_COMPILE}objdump -D \
build/platform/generic/firmware/fw_dynamic.elf | \
grep "<console_early_fifo>:" | awk '{print $1}'`
${CROSS_COMPILE}objdump -R build/platform/generic/firmware/fw_dynamic.elf | \
grep $CONSOLE_EARLY_FIFO_ADDR | awk '{print $3}'
Signed-off-by: Anup Patel <[email protected]>
Reviewed-By: Himanshu Chauhan <[email protected]>
|
|
Extend sbi_fifo_enqueue() to allow forceful queueing by droping
data from the tail.
Signed-off-by: Anup Patel <[email protected]>
Reviewed-By: Himanshu Chauhan <[email protected]>
|
|
Don't use sbi_memcpy() for basic data types in fifo enqueue/dequeue
instead use direct type-cast and assignment.
Signed-off-by: Anup Patel <[email protected]>
Reviewed-By: Himanshu Chauhan <[email protected]>
|
|
Now that all platforms have been updated to initialize serial console
device in early_init(), the sbi_console_init() and console_init()
platform callback are redundant hence remove them.
Signed-off-by: Anup Patel <[email protected]>
Reviewed-By: Himanshu Chauhan <[email protected]>
|
|
Attempting to access the menvcfg CSR raises an illegal instruction
exception on hardware which implements Sm1p11 or older.
Fixes: e9ee9678ba50 ("lib: sbi: fwft: add support for SBI_FWFT_PTE_AD_HW_UPDATING")
Signed-off-by: Samuel Holland <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
This patch introduces a new math test suite to the SBI unit
tests. The changes include:
* Updating lib/sbi/tests/objects.mk to include
math_test_suite and sbi_math_test.o.
* Adding a new file lib/sbi/tests/sbi_math_test.c which
contains tests for log2roundup function using various cases.
The addition of this test suite ensures that mathematical
functions are verified and work as expected.
Signed-off-by: Dongdong Zhang <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
This patch improves the readability of the SBI unit test
output by adding color-coded status messages.
Adding ANSI color codes for green (pass) and red (fail)
in sbi_unit_test.c.
Now, the test results will be displayed in green for passed
tests and red for failed tests, making it easier to quickly
distinguish between them.
Signed-off-by: Dongdong Zhang <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
We would like to clean any files generated by the carray
scripts by just searching for the filename as the current
make system turns f.carray into f.o. Change to make the
make system turn f.carray into f.carray.o
note, command to go through .mk files changing the .o
in the .mk files is:
find . -type f -name "*.carray" | xargs -t -I fname /bin/bash -x -c ' fn=`basename -s .carray fname`; echo "$fn"; sed -i `dirname fname `/objects.mk -e s/"$fn".o/"$fn".carray.o/g'
Link: https://patchwork.ozlabs.org/project/opensbi/patch/[email protected]/
Reported-by: Ivan Orlov <[email protected]>
Suggested-by: Andrew Jones <[email protected]>
Signed-off-by: Ben Dooks <[email protected]>
Reviewed-by: Andrew Jones <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
Illegal CSR accesses from lower privilege modes are delegated to S-mode
and do not necessarily indicate a bug. Supervisor software may want to
emulate some CSRs, or may intentionally disable access to certain
existing CSRs, and thus will expect traps when those CSRs are accessed.
For example, Linux disables sstatus.VS by default in order to detect
when userspace first accesses vector register state; this includes the
CSRs defined by the V extesion. As a result, if the first vector
instruction in a process is a CSR access, OpenSBI will log the illegal
instruction exception, even though there is no unexpected or erroneous
behavior occurring.
Since the illegal instruction exception is delegated to S-mode, S-mode
software should be responsible for reporting the exception, not OpenSBI.
Signed-off-by: Samuel Holland <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
Current Debug Trigger SBI extension proposal suggests to activate
shmem area and obtain its physical address from S-mode software
in the following way:
: If both `shmem_phys_lo` and `shmem_phys_hi` parameters are not
: all-ones bitwise then `shmem_phys_lo` specifies the lower XLEN
: bits and `shmem_phys_hi` specifies the upper XLEN bits of the
: shared memory physical base address. The `shmem_phys_lo` MUST
: be `(XLEN / 8)` byte aligned and the size of shared memory is
: assumed to be `trig_max * (XLEN / 2)` bytes.
For more details see the current version of the proposal:
- https://lists.riscv.org/g/tech-debug/message/1302
On the other hand, on RV32, the M-mode can only access the first 4GB of
the physical address space because M-mode does not have MMU to access
full 34-bit physical address space. Similarly, on RV64, the M-mode can
only access memory addressed by 64 bits.
This commit checks shmem address in function sbi_dbtr_setup_shmem
to make sure that shmem_phys_hi part of the valid address is zero.
Besides, the macro DBTR_SHMEM_MAKE_PHYS is updated to take into
account only low XLEN part.
Signed-off-by: Sergey Matyukevich <[email protected]>
Reviewed-by: Himanshu Chauhan <[email protected]>
|
|
Error code returned by the ecall handles should use the defines from
sbi_ecall_interface.h rather than sbi_error.h.
Signed-off-by: Clément Léger <[email protected]>
Reviewed-by: Andrew Jones <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
As stated by the spec and pointed out by Andrew Jones, the value passed
for MISALIGNED_EXC_DELEG and PTE_AD_HW_UPDATING should be either 0 or 1.
Add check for these values and return SBI_EINVAL if not.
Signed-off-by: Clément Léger <[email protected]>
Reviewed-by: Andrew Jones <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
The SBI FWFT extension defines a set of function that can be called
to control the configuration of some platform features (misaligned
trap delegation, etc). This patch implements sbi_fwft_set() and
sbi_fwft_get() as defined in the specification [1].
Link: https://lists.riscv.org/g/tech-prs/message/924 [1]
Signed-off-by: Clément Léger <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
Add support for SBI_FWFT_PTE_AD_HW_UPDATING based on SVADU presence.
Signed-off-by: Clément Léger <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
Add support for SBI_FWFT_MISALIGNED_EXC_DELEG withing FWFT support. This
support allows to delegate misaligned accesses traps.
Signed-off-by: Clément Léger <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
This extension allows the software running in supervisor mode to control
the behavior of various features of the SBI [1]. Implement the support
for such extension.
Link: https://lists.riscv.org/g/tech-prs/message/924 [1]
Signed-off-by: Clément Léger <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
Add support for Svade and Svadu extensions. When both are present in the
device tree, the M-mode firmware should select the Svade extension
to comply with the RVA23 profile, which mandates Svade and lists Svadu as
an optional extension.
Signed-off-by: Yong-Xuan Wang <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
Save/restore sbi_trap_context during domain context switching to
ensure proper trap handling and isolation. This maintains correct
domain-specific state, avoiding context corruption.
Fixes: abea949721bc ("lib: sbi: Introduce trap context")
Signed-off-by: Yu Chien Peter Lin <[email protected]>
Reviewed-by: Alvin Chang <[email protected]>
Tested-by: Alvin Chang <[email protected]>
Reviewed-by: Yong Li <[email protected]>
Tested-by: Yong Li <[email protected]>
|
|
The spec states that a6, a7, flags and sepc are writable but the
implementation was not allowing that. Add support for these 4 writable
attributes.
Signed-off-by: Clément Léger <[email protected]>
Reviewed-by: Samuel Holland <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
Currently, the sbi_sse_exit() gets event in a loop but does not put
it back after use. This results in global events remaining locked
causing hangs on sub-sequent calls to sse_event_get() for global
events.
Fixes: c8cdf01d8f3a ("lib: sbi: Add support for Supervisor Software Events extension")
Signed-off-by: Anup Patel <[email protected]>
Reviewed-by: Himanshu Chauhan <[email protected]>
|
|
The inline function can simplify the code by setting some call
restrictions. This ensures logical smoothness
Signed-off-by: Xiang W <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
Modify the return value of the sbi_dbtr_shmem_disabled function to
bool to make the semantics clearer.
Signed-off-by: Xiang W <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
The previous code detected shmem_phys_hi and shmem_phys_lo as two
addresses. fix this bug
Signed-off-by: Xiang W <[email protected]>
Reviewed-by: Himanshu Chauhan <[email protected]>
|
|
Obtaining a 64-bit address under rv64 does not require combining
32-bit registers
Signed-off-by: Xiang W <[email protected]>
Reviewed-by: Himanshu Chauhan <[email protected]>
|
|
After getting hart_shmem_base, dbtr_thishart_state_ptr cannot be
empty. So remove check code.
Signed-off-by: Xiang W <[email protected]>
Reviewed-by: Himanshu Chauhan <[email protected]>
|
|
When only phys_hi is equal to SBI_DBTR_SHMEM_INVALID_ADDR, it may be
a legal address. The old code would modify the legal address.
Signed-off-by: Xiang W <[email protected]>
Reviewed-by: Himanshu Chauhan <[email protected]>
|
|
Add spinlock protection to avoid race condition on assigned_harts
during domain context switching. Also, rename/add variables for
accessing the corresponding domain of target/current context.
Signed-off-by: Alvin Chang <[email protected]>
Reviewed-by: Yu Chien Peter Lin <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
Although Mstatus.MPV is set, before executing mret, access to VS mode
registers should use the actual register addresses, not the pseudonyms
of S registers.
Signed-off-by: Pope B.Lei <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
When registering an SSE event, check for the handler_entry_pc to belong
to supervisor mode domain using sbi_domain_check_addr_range().
Signed-off-by: Clément Léger <[email protected]>
Reported-by: Samuel Holland <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
Rather then passing 1 to sbi_domain_check_addr_range() for supervisor
mode, use PRV_S.
Signed-off-by: Clément Léger <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
Move the enable callback call before sending the IPI. Even though the
event is locked and no race condition can happen, this is more logical.
Signed-off-by: Clément Léger <[email protected]>
Reported-by: Samuel Holland <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
The list loop below that check is actually not looping if the list is
empty so there was no need for this check.
Signed-off-by: Clément Léger <[email protected]>
Reported-by: Samuel Holland <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
Rather than checking 32bits overflow with some absolute value, check the
value to be different from the cast itself.
Signed-off-by: Clément Léger <[email protected]>
Reported-by: Samuel Holland <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
There was a naming incoherency between enabled events list lock/unlock.
Rename sse_hart_unlock() to sse_enabled_event_unlock() to be coherent
and reword comments above lock()/unlock() functions.
Signed-off-by: Clément Léger <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
Fix some errors spotted by Samuel while reviewing the SSE implementation.
Signed-off-by: Clément Léger <[email protected]>
Reported-by: Samuel Holland <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
Implement the test which covers some of the functions from the
`riscv_locks.h` file. This test consists of 3 test cases:
1) For lock/unlock functions
2) Unsuccessful trylock (the lock was previously taken)
3) Successful trylock (the lock is free and can be taken)
Signed-off-by: Ivan Orlov <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
