This patch series suggests fixes for several corner cases in the RISC-V vector ptrace implementation:
- init vector context with proper vlenb, to avoid reading zero vlenb by an early attached debugger
- follow gdbserver expectations and return ENODATA instead of EINVAL if vector extension is supported but not yet activated for the traced process
- validate input vector csr registers in ptrace, to maintain an accurate view of the tracee's vector context across multiple halt/resume debug cycles
For detailed description see the appropriate commit messages. A new test suite validate_v_ptrace is added to the tools/testing/selftests/riscv/vector to verify some of the vector ptrace functionality and corner cases.
So far tested on the following platforms: - test in QEMU rv32/rv64 - test on c908 (BananaPi CanMV K230D Zero) - test on c906 (MangoPi MQ Pro)
Previous versions: - v4: https://lore.kernel.org/linux-riscv/20251108194207.1257866-1-geomatsi@gmail.... - v3: https://lore.kernel.org/linux-riscv/20251025210655.43099-1-geomatsi@gmail.co... - v2: https://lore.kernel.org/linux-riscv/20250821173957.563472-1-geomatsi@gmail.c... - v1: https://lore.kernel.org/linux-riscv/20251007115840.2320557-1-geomatsi@gmail....
Changes in v5: - add support and minimal set of tests for XTheadVector
Changes in v4: The form 'vsetvli x0, x0, ...' can only be used if VLMAX remains unchanged, see spec 6.2. This condition was not met by the initial values in the selftests w.r.t. the initial zeroed context. QEMU accepted such values, but actual hardware (c908, BananaPi CanMV Zero board) did not, setting vill. So fix the selftests after testing on hardware: - replace 'vsetvli x0, x0, ...' by 'vsetvli rd, x0, ...' - fixed instruction returns VLMAX, so use it in checks as well - replace fixed vlenb == 16 in the syscall test
Changes in v3: Address the review comments by Andy Chiu and rework the approach: - drop forced vector context save entirely - perform strict validation of vector csr regs in ptrace
Changes in v2: - add thread_info flag to allow to force vector context save - force vector context save after vector ptrace to ensure valid vector context in the next ptrace operations - force vector context save on the first context switch after vector context init to get proper vlenb
---
Ilya Mamay (1): riscv: ptrace: return ENODATA for inactive vector extension
Sergey Matyukevich (8): riscv: vector: init vector context with proper vlenb riscv: csr: define vtype register elements riscv: ptrace: validate input vector csr registers selftests: riscv: test ptrace vector interface selftests: riscv: verify initial vector state with ptrace selftests: riscv: verify syscalls discard vector context selftests: riscv: verify ptrace rejects invalid vector csr inputs selftests: riscv: verify ptrace accepts valid vector csr values
arch/riscv/include/asm/csr.h | 17 + arch/riscv/kernel/ptrace.c | 98 +- arch/riscv/kernel/vector.c | 12 +- .../testing/selftests/riscv/vector/.gitignore | 2 + tools/testing/selftests/riscv/vector/Makefile | 10 +- .../selftests/riscv/vector/v_helpers.c | 23 + .../selftests/riscv/vector/v_helpers.h | 2 + .../riscv/vector/validate_v_ptrace.c | 919 ++++++++++++++++++ 8 files changed, 1075 insertions(+), 8 deletions(-) create mode 100644 tools/testing/selftests/riscv/vector/validate_v_ptrace.c
base-commit: 8f0b4cce4481fb22653697cced8d0d04027cb1e8