在 2021/3/5 11:12, HongJieDeng 写道:
From: Hongjie Deng <deng.hong...@h3c.com>

We need more stack space, xori/ori no longer apply when
  _THREAD_MASK exceeds 16 bits

Signed-off-by: Hongjie Deng <deng.hong...@h3c.com>
---
  arch/mips/include/asm/stackframe.h | 8 ++++++++
  arch/mips/kernel/genex.S           | 6 ++++++
  2 files changed, 14 insertions(+)

diff --git a/arch/mips/include/asm/stackframe.h 
b/arch/mips/include/asm/stackframe.h
index aa430a6..6ebc39f 100644
--- a/arch/mips/include/asm/stackframe.h
+++ b/arch/mips/include/asm/stackframe.h
@@ -278,8 +278,16 @@
                sll     k0, 3           /* extract cu0 bit */
                bltz    k0, 9f
+#if _THREAD_MASK < (1 << 16 - 1)
                ori     $28, sp, _THREAD_MASK
                xori    $28, _THREAD_MASK
+#else
+               li      $28,   _THREAD_MASK
+               or  $28,  sp,  $28
+               li  $24, _THREAD_MASK
+               xor   $28, $28, $24

Hi Hongjie,

li breaks MIPS64, please try PTR_LI instead.
Also loading immediate cost a lot on MIPS,
you may save a unnecessary li by:

+               PTR_LI  v1,   _THREAD_MASK
+               or      gp,  sp,  v1
+               xor     gp,  gp,  v1

Also clobber v0/v1 seems much more reasonable than $24 here....

+#endif
+
  #ifdef CONFIG_CPU_CAVIUM_OCTEON
                .set    mips64
                pref    0, 0($28)       /* Prefetch the current pointer */
diff --git a/arch/mips/kernel/genex.S b/arch/mips/kernel/genex.S
index bcce32a..5ea4fe4 100644
--- a/arch/mips/kernel/genex.S
+++ b/arch/mips/kernel/genex.S
@@ -662,8 +662,14 @@ isrdhwr:
  #endif
        MTC0    k0, CP0_EPC
        /* I hope three instructions between MTC0 and ERET are enough... */
+#if    _THREAD_MASK  < (1 << 16 - 1)
        ori     k1, _THREAD_MASK
        xori    k1, _THREAD_MASK
+#else
+       li  $24 ,_THREAD_MASK
+       or   k1, k1, $24
+       xor   k1, k1, $24
+#endif
Ditto.

Actually another possible solution could be `.set at=v1` around it
and let aeembler to decide if it needs a extra register.

Thanks.

- Jiaxun
        LONG_L  v1, TI_TP_VALUE(k1)
        .set    push
        .set    arch=r4000

Reply via email to