[PATCH 2/2] lib: sbi: Reduce stack usage of vector misaligned load/store emulation
Anup Patel <[email protected]> Sat, 13 Jun 2026 18:51:29 +0530
| Newsgroups | org.infradead.lists.opensbi |
|---|---|
| Message-ID | <[email protected]> |
The vector misaligned load/store emulation requires mask[] of 8192 bytes which cause stack corruptions since default stack size is also 8192 bytes. To address the above issue, allocate mask[] on-demand from heap for each hart and store the per-hart pointer in scratch space for reuse. Signed-off-by: Anup Patel <[email protected]> --- include/sbi/sbi_trap_ldst.h | 2 ++ lib/sbi/sbi_trap.c | 8 +++++ lib/sbi/sbi_trap_v_ldst.c | 61 +++++++++++++++++++++++++++++++++++-- 3 files changed, 68 insertions(+), 3 deletions(-) diff --git a/include/sbi/sbi_trap_ldst.h b/include/sbi/sbi_trap_ldst.h index a6a6c75b..65539035 100644 --- a/include/sbi/sbi_trap_ldst.h +++ b/include/sbi/sbi_trap_ldst.h @@ -37,4 +37,6 @@ int sbi_misaligned_v_ld_emulator(int rlen, union sbi_ldst_data *out_val, int sbi_misaligned_v_st_emulator(int wlen, union sbi_ldst_data in_val, struct sbi_trap_context *tcntx); +int sbi_misaligned_v_ldst_init(void); + #endif diff --git a/lib/sbi/sbi_trap.c b/lib/sbi/sbi_trap.c index 16774fd2..428d9296 100644 --- a/lib/sbi/sbi_trap.c +++ b/lib/sbi/sbi_trap.c @@ -459,5 +459,13 @@ static int delegate_traps(struct sbi_scratch *scratch) int sbi_trap_init(struct sbi_scratch *scratch, bool cold_boot) { + int rc; + + if (cold_boot) { + rc = sbi_misaligned_v_ldst_init(); + if (rc) + return rc; + } + return delegate_traps(scratch); } diff --git a/lib/sbi/sbi_trap_v_ldst.c b/lib/sbi/sbi_trap_v_ldst.c index f4d469dc..e642d4a2 100644 --- a/lib/sbi/sbi_trap_v_ldst.c +++ b/lib/sbi/sbi_trap_v_ldst.c @@ -12,15 +12,45 @@ #include <sbi/riscv_asm.h> #include <sbi/riscv_encoding.h> #include <sbi/sbi_error.h> +#include <sbi/sbi_heap.h> +#include <sbi/sbi_scratch.h> #include <sbi/sbi_trap_ldst.h> #include <sbi/sbi_trap.h> #include <sbi/sbi_unpriv.h> -#include <sbi/sbi_trap.h> #ifdef OPENSBI_CC_SUPPORT_VECTOR #define VLEN_MAX 65536 +struct misaligned_v_ldst_data { + uint8_t mask[VLEN_MAX / 8]; +}; + +static unsigned long misaligned_v_ldst_dataptr_offset; + +#define misaligned_v_ldst_get_data_ptr(__scratch) \ + sbi_scratch_read_type((__scratch), void *, misaligned_v_ldst_dataptr_offset) + +#define misaligned_v_ldst_set_data_ptr(__scratch, __ldst) \ + sbi_scratch_write_type((__scratch), void *, misaligned_v_ldst_dataptr_offset, (__ldst)) + +static struct misaligned_v_ldst_data *misaligned_v_ldst_thishart_data_ptr(void) +{ + struct sbi_scratch *scratch = sbi_scratch_thishart_ptr(); + struct misaligned_v_ldst_data *ptr; + + if (!misaligned_v_ldst_dataptr_offset) + return NULL; + + ptr = misaligned_v_ldst_get_data_ptr(scratch); + if (!ptr) { + ptr = sbi_zalloc(sizeof(*ptr)); + misaligned_v_ldst_set_data_ptr(scratch, ptr); + } + + return misaligned_v_ldst_get_data_ptr(sbi_scratch_thishart_ptr()); +} + static inline void set_vreg(ulong vlenb, ulong which, ulong pos, ulong size, const uint8_t *bytes) { @@ -142,6 +172,7 @@ int sbi_misaligned_v_ld_emulator(int rlen, union sbi_ldst_data *out_val, { const struct sbi_trap_info *orig_trap = &tcntx->trap; struct sbi_trap_regs *regs = &tcntx->regs; + struct misaligned_v_ldst_data *hart_data; struct sbi_trap_info uptrap; ulong insn = sbi_get_insn(regs->mepc, &uptrap); ulong vl = csr_read(CSR_VL); @@ -157,12 +188,17 @@ int sbi_misaligned_v_ld_emulator(int rlen, union sbi_ldst_data *out_val, ulong vlmul = GET_VLMUL(vtype); bool illegal = GET_MEW(insn); bool masked = IS_MASKED(insn); - uint8_t mask[VLEN_MAX / 8]; uint8_t bytes[8 * sizeof(uint64_t)]; ulong len = GET_LEN(view); ulong nf = GET_NF(insn); ulong vemul = GET_VEMUL(vlmul, view, vsew); ulong emul = GET_EMUL(vemul); + uint8_t *mask; + + hart_data = misaligned_v_ldst_thishart_data_ptr(); + if (!hart_data) + return SBI_ENOMEM; + mask = hart_data->mask; if (IS_UNIT_STRIDE_LOAD(insn) || IS_FAULT_ONLY_FIRST_LOAD(insn)) { stride = nf * len; @@ -242,6 +278,7 @@ int sbi_misaligned_v_st_emulator(int wlen, union sbi_ldst_data in_val, { const struct sbi_trap_info *orig_trap = &tcntx->trap; struct sbi_trap_regs *regs = &tcntx->regs; + struct misaligned_v_ldst_data *hart_data; struct sbi_trap_info uptrap; ulong insn = sbi_get_insn(regs->mepc, &uptrap); ulong vl = csr_read(CSR_VL); @@ -257,12 +294,17 @@ int sbi_misaligned_v_st_emulator(int wlen, union sbi_ldst_data in_val, ulong vlmul = GET_VLMUL(vtype); bool illegal = GET_MEW(insn); bool masked = IS_MASKED(insn); - uint8_t mask[VLEN_MAX / 8]; uint8_t bytes[8 * sizeof(uint64_t)]; ulong len = GET_LEN(view); ulong nf = GET_NF(insn); ulong vemul = GET_VEMUL(vlmul, view, vsew); ulong emul = GET_EMUL(vemul); + uint8_t *mask; + + hart_data = misaligned_v_ldst_thishart_data_ptr(); + if (!hart_data) + return SBI_ENOMEM; + mask = hart_data->mask; if (IS_UNIT_STRIDE_STORE(insn)) { stride = nf * len; @@ -330,6 +372,15 @@ int sbi_misaligned_v_st_emulator(int wlen, union sbi_ldst_data in_val, return vl; } + +int sbi_misaligned_v_ldst_init(void) +{ + misaligned_v_ldst_dataptr_offset = sbi_scratch_alloc_type_offset(void *); + if (!misaligned_v_ldst_dataptr_offset) + return SBI_ENOMEM; + + return 0; +} #else int sbi_misaligned_v_ld_emulator(int rlen, union sbi_ldst_data *out_val, struct sbi_trap_context *tcntx) @@ -341,4 +392,8 @@ int sbi_misaligned_v_st_emulator(int wlen, union sbi_ldst_data in_val, { return 0; } +int sbi_misaligned_v_ldst_init(void) +{ + return 0; +} #endif /* OPENSBI_CC_SUPPORT_VECTOR */ -- 2.43.0 -- opensbi mailing list [email protected] http://lists.infradead.org/mailman/listinfo/opensbi