[DTrace-devel] [PATCH 2/2] cg: fix dt_cg_check_fault()

[Eugene Loh]

Reviewed-by: Eugene Loh <eugene.loh at oracle.com>

but it wasn't all that bad, was it?  All the exit label had was a 
"return 0".  So the old code was in a sense just inlining that. So the 
old code wasn't observably incorrect.  It's just that this patch cleans 
things up?  I'm just trying to check my understanding of the situation.

On 5/11/23 01:24, Kris Van Hees via DTrace-devel wrote:
> The dt_cg_check_fault() function is used to detect faults reported
> by pre-compiled BPF code, and is meant to bypass further clause
> execution when a fault occurs by jumping to the exit label.  But it
> was not doing that at all.
>
> Signed-off-by: Kris Van Hees <kris.van.hees at oracle.com>
> ---
>   libdtrace/dt_cg.c | 5 +----
>   1 file changed, 1 insertion(+), 4 deletions(-)
>
> diff --git a/libdtrace/dt_cg.c b/libdtrace/dt_cg.c
> index 8a890043..bbbd5093 100644
> --- a/libdtrace/dt_cg.c
> +++ b/libdtrace/dt_cg.c
> @@ -1003,16 +1003,13 @@ dt_cg_check_fault(dt_pcb_t *pcb)
>   	dt_irlist_t	*dlp = &pcb->pcb_ir;
>   	dt_regset_t	*drp = pcb->pcb_regs;
>   	int		reg = dt_regset_alloc(drp);
> -	uint_t		lbl_ok = dt_irlist_label(dlp);
>   
>   	if (reg == -1)
>   		longjmp(yypcb->pcb_jmpbuf, EDT_NOREG);
>   	emit(dlp,  BPF_LOAD(BPF_DW, reg, BPF_REG_FP, DT_STK_DCTX));
>   	emit(dlp,  BPF_LOAD(BPF_DW, reg, reg, DCTX_MST));
>   	emit(dlp,  BPF_LOAD(BPF_DW, reg, reg, DMST_FAULT));
> -	emit(dlp,  BPF_BRANCH_IMM(BPF_JEQ, reg, 0, lbl_ok));
> -	emitl(dlp, lbl_ok,
> -		   BPF_NOP());
> +	emit(dlp,  BPF_BRANCH_IMM(BPF_JNE, reg, 0, pcb->pcb_exitlbl));
>   	dt_regset_free(drp, reg);
>   }
>   /*