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>
This commit is contained in:
Anup Patel
2020-03-19 17:13:16 +05:30
committed by Anup Patel
parent 160c88535f
commit 1de66d170e
7 changed files with 147 additions and 125 deletions

View File

@@ -16,9 +16,11 @@ struct sbi_scratch;
int sbi_hart_init(struct sbi_scratch *scratch, u32 hartid, bool cold_boot);
void *sbi_hart_get_trap_info(struct sbi_scratch *scratch);
void sbi_hart_set_trap_info(struct sbi_scratch *scratch, void *data);
extern void (*sbi_hart_unpriv_trap)(void);
static inline ulong sbi_hart_unpriv_trap_addr(void)
{
return (ulong)sbi_hart_unpriv_trap;
}
void sbi_hart_delegation_dump(struct sbi_scratch *scratch);
void sbi_hart_pmp_dump(struct sbi_scratch *scratch);