xtensa: use callx0 opcode in fast_coprocessor

Instead of emulating call0 in fast_coprocessor use that opcode directly.
Use 'ret' instead of 'jx a0'.

Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>

diff --git a/arch/xtensa/kernel/coprocessor.S b/arch/xtensa/kernel/coprocessor.S
index c7b9f12896f20a6870588cd50e9849e11c992dcf..8bcbabbff38a745d3140d3609233fd1b1f48e667 100644 (file)
--- a/arch/xtensa/kernel/coprocessor.S
+++ b/arch/xtensa/kernel/coprocessor.S
@@ -30,7 +30,7 @@
                .align 4;                                               \
        .Lsave_cp_regs_cp##x:                                           \
                xchal_cp##x##_store a2 a3 a4 a5 a6;                     \
-               jx      a0;                                             \
+               ret;                                                    \
        .endif
 
 #define SAVE_CP_REGS_TAB(x)                                            \
@@ -47,7 +47,7 @@
                .align 4;                                               \
        .Lload_cp_regs_cp##x:                                           \
                xchal_cp##x##_load a2 a3 a4 a5 a6;                      \
-               jx      a0;                                             \
+               ret;                                                    \
        .endif
 
 #define LOAD_CP_REGS_TAB(x)                                            \
@@ -163,21 +163,20 @@ ENTRY(fast_coprocessor)
        s32i    a5, a4, THREAD_CPENABLE
 
        /*
-        * Get context save area and 'call' save routine. 
+        * Get context save area and call save routine.
         * (a4 still holds previous owner (thread_info), a3 CP number)
         */
 
        movi    a5, .Lsave_cp_regs_jump_table
-       movi    a0, 2f                  # a0: 'return' address
        addx8   a3, a3, a5              # a3: coprocessor number
        l32i    a2, a3, 4               # a2: xtregs offset
        l32i    a3, a3, 0               # a3: jump address
        add     a2, a2, a4
-       jx      a3
+       callx0  a3
 
        /* Note that only a0 and a1 were preserved. */
 
-2:     rsr     a3, exccause
+       rsr     a3, exccause
        addi    a3, a3, -EXCCAUSE_COPROCESSOR0_DISABLED
        movi    a0, coprocessor_owner
        addx4   a0, a3, a0
@@ -187,19 +186,18 @@ ENTRY(fast_coprocessor)
 1:     GET_THREAD_INFO (a4, a1)
        s32i    a4, a0, 0
 
-       /* Get context save area and 'call' load routine. */
+       /* Get context save area and call load routine. */
 
        movi    a5, .Lload_cp_regs_jump_table
-       movi    a0, 1f
        addx8   a3, a3, a5
        l32i    a2, a3, 4               # a2: xtregs offset
        l32i    a3, a3, 0               # a3: jump address
        add     a2, a2, a4
-       jx      a3
+       callx0  a3
 
        /* Restore all registers and return from exception handler. */
 
-1:     l32i    a6, a1, PT_AREG6
+       l32i    a6, a1, PT_AREG6
        l32i    a5, a1, PT_AREG5
        l32i    a4, a1, PT_AREG4