[src/trunk]: src/sys/arch/aarch64/aarch64 Just a few optimizations.

[ryo <ryo%NetBSD.org@localhost>]

details:   https://anonhg.NetBSD.org/src/rev/799a20964559
branches:  trunk
changeset: 953006:799a20964559
user:      ryo <ryo%NetBSD.org@localhost>
date:      Tue Feb 23 14:50:33 2021 +0000

description:
Just a few optimizations.

- in cpu_idle(), ci_intr_depth is always 0, so there is no need to fetch for increment or conditional branch.
- curcpu() is immutable in idle lwp, there is no need to consider KPREEMPT. Therefore, get curcpu() first and keep using it.
- add more comment.

diffstat:

 sys/arch/aarch64/aarch64/idle_machdep.S |  32 +++++++++++++++++++-------------
 1 files changed, 19 insertions(+), 13 deletions(-)

diffs (64 lines):

diff -r c360ff5c4cb0 -r 799a20964559 sys/arch/aarch64/aarch64/idle_machdep.S
--- a/sys/arch/aarch64/aarch64/idle_machdep.S   Tue Feb 23 14:27:27 2021 +0000
+++ b/sys/arch/aarch64/aarch64/idle_machdep.S   Tue Feb 23 14:50:33 2021 +0000
@@ -1,4 +1,4 @@
-/* $NetBSD: idle_machdep.S,v 1.8 2021/02/21 23:37:09 jmcneill Exp $ */
+/* $NetBSD: idle_machdep.S,v 1.9 2021/02/23 14:50:33 ryo Exp $ */
 
 /*-
  * Copyright (c) 2014 The NetBSD Foundation, Inc.
@@ -36,7 +36,7 @@
 #include <aarch64/locore.h>
 #include "assym.h"
 
-RCSID("$NetBSD: idle_machdep.S,v 1.8 2021/02/21 23:37:09 jmcneill Exp $");
+RCSID("$NetBSD: idle_machdep.S,v 1.9 2021/02/23 14:50:33 ryo Exp $");
 
 #ifdef ARM_INTR_IMPL
 #include ARM_INTR_IMPL
@@ -75,28 +75,34 @@
        adr     x0, 1f
        str     x0, [sp, #TF_PC]                /* CLKF_PC refer to tf_pc */
 
+       /*
+        * "idle/N" lwp is allocated on a per-CPU basis,
+        * curcpu() always return the same, and there is no need to
+        * consider KPREEMPT. safe even with interrupt enabled.
+        */
        mrs     x1, tpidr_el1                   /* get curlwp */
-       ldr     x1, [x1, #L_CPU]                /* get curcpu */
-       ldr     w28, [x1, #CI_INTR_DEPTH]       /* w28 = ci->ci_intr_depth */
-       add     w2, w28, #1                     /* w2 = intr_depth + 1 */
+       ldr     x28, [x1, #L_CPU]               /* get curcpu */
 
+       mov     w2, #1
        mov     x0, sp                          /* get pointer to trapframe */
 
        DISABLE_INTERRUPT
-       wfi
+       /*
+        * assert(ci->ci_intr_depth == 0),
+        * therefore, ci->ci_intr_depth++ would be definitely 1.
+        */
+       str     w2, [x28, #CI_INTR_DEPTH]       /* ci->ci_intr_depth = 1 */
 
-       str     w2, [x1, #CI_INTR_DEPTH]        /* ci->ci_intr_depth++ */
+       wfi
        bl      ARM_IRQ_HANDLER                 /* irqhandler(trapframe) */
 1:
-       mrs     x1, tpidr_el1                   /* get curlwp */
-       ldr     x1, [x1, #L_CPU]                /* get curcpu */
-       str     w28, [x1, #CI_INTR_DEPTH]       /* ci->ci_intr_depth = old */
+       /* x28 is curcpu() */
+       str     wzr, [x28, #CI_INTR_DEPTH]      /* ci->ci_intr_depth = 0 */
 
 #if defined(__HAVE_FAST_SOFTINTS) && !defined(__HAVE_PIC_FAST_SOFTINTS)
-       cbnz    w28, 1f                         /* Skip if intr_depth > 0 */
-       ldr     w3, [x1, #CI_SOFTINTS]          /* Get pending softint mask */
+       ldr     w3, [x28, #CI_SOFTINTS]         /* Get pending softint mask */
        /* CPL should be 0 */
-       ldr     w2, [x1, #CI_CPL]               /* Get current priority level */
+       ldr     w2, [x28, #CI_CPL]              /* Get current priority level */
        lsr     w3, w3, w2                      /* shift mask by cpl */
        cbz     w3, 1f
        bl      _C_LABEL(dosoftints)            /* dosoftints() */