/*
 * Copyright (c) 2016 Cadence Design Systems, Inc.
 * Copyright (c) 2022 Intel Corporation
 * SPDX-License-Identifier: Apache-2.0
 */
#include <xtensa/coreasm.h>
#include <zsr.h>

/* WINDOW OVERFLOW AND UNDERFLOW EXCEPTION VECTORS AND ALLOCA EXCEPTION
 * HANDLER
 *
 * Here is the code for each window overflow/underflow exception vector and
 * (interspersed) efficient code for handling the alloca exception cause.
 * Window exceptions are handled entirely in the vector area and are very tight
 * for performance. The alloca exception is also handled entirely in the window
 * vector area so comes at essentially no cost in code size. Users should never
 * need to modify them and Cadence Design Systems recommends they do not.
 *
 * Window handlers go at predetermined vector locations according to the Xtensa
 * hardware configuration, which is ensured by their placement in a special
 * section known to the Xtensa linker support package (LSP). Since their
 * offsets in that section are always the same, the LSPs do not define a
 * section per vector.
 *
 * These things are coded for XEA2 only (XEA1 is not supported).
 *
 * Note on Underflow Handlers:
 *
 * The underflow handler for returning from call[i+1] to call[i] must preserve
 * all the registers from call[i+1]'s window. In particular, a0 and a1 must be
 * preserved because the RETW instruction will be reexecuted (and may even
 * underflow if an intervening exception has flushed call[i]'s registers).
 * Registers a2 and up may contain return values.
 */

#if XCHAL_HAVE_WINDOWED

    .section .WindowVectors.text, "ax"

/* Window Overflow Exception for Call4.
 *
 * Invoked if a call[i] referenced a register (a4-a15)
 * that contains data from ancestor call[j];
 * call[j] had done a call4 to call[j+1].
 * On entry here:
 *	window rotated to call[j] start point;
 *	    a0-a3 are registers to be saved;
 *	    a4-a15 must be preserved;
 *	    a5 is call[j+1]'s stack pointer.
 */

    .org    0x0
    .global _WindowOverflow4
_WindowOverflow4:

    s32e    a0, a5, -16     /* save a0 to call[j+1]'s stack frame */
    s32e    a1, a5, -12     /* save a1 to call[j+1]'s stack frame */
    s32e    a2, a5,  -8     /* save a2 to call[j+1]'s stack frame */
    s32e    a3, a5,  -4     /* save a3 to call[j+1]'s stack frame */
    rfwo                    /* rotates back to call[i] position */

/* Window Underflow Exception for Call4
 *
 * Invoked by RETW returning from call[i+1] to call[i]
 * where call[i]'s registers must be reloaded (not live in ARs);
 * where call[i] had done a call4 to call[i+1].
 * On entry here:
 * 	    window rotated to call[i] start point;
 * 	    a0-a3 are undefined, must be reloaded with call[i].reg[0..3];
 * 	    a4-a15 must be preserved (they are call[i+1].reg[0..11]);
 * 	    a5 is call[i+1]'s stack pointer.
 */

    .org    0x40
    .global _WindowUnderflow4
_WindowUnderflow4:

    l32e    a0, a5, -16     /* restore a0 from call[i+1]'s stack frame */
    l32e    a1, a5, -12     /* restore a1 from call[i+1]'s stack frame */
    l32e    a2, a5,  -8     /* restore a2 from call[i+1]'s stack frame */
    l32e    a3, a5,  -4     /* restore a3 from call[i+1]'s stack frame */
    rfwu

/* Handle alloca exception generated by interruptee executing 'movsp'.
 * This uses space between the window vectors, so is essentially
 * "free".  All interruptee's regs are intact except a0 which is saved
 * in $ZSR_ALLOCA (assigned at build time, see gen_zsr.py for
 * details), and PS.EXCM has been set by the exception hardware (can't
 * be interrupted).  The fact the alloca exception was taken means the
 * registers associated with the base-save area have been spilled and
 * will be restored by the underflow handler, so those 4 registers are
 * available for scratch.  The code is optimized to avoid unaligned
 * branches and minimize cache misses.
 */

    .align  4
    .global _xt_alloca_exc
_xt_alloca_exc:

    rsr     a0, WINDOWBASE  /* grab WINDOWBASE before rotw changes it */
    rotw    -1      /* WINDOWBASE goes to a4, new a0-a3 are scratch */
    rsr     a2, PS
    extui   a3, a2, XCHAL_PS_OWB_SHIFT, XCHAL_PS_OWB_BITS
    xor     a3, a3, a4    /* bits changed from old to current windowbase */
    rsr     a4, ZSR_ALLOCA  /* restore original a0 (now in a4) */
    slli    a3, a3, XCHAL_PS_OWB_SHIFT
    xor     a2, a2, a3      /* flip changed bits in old window base */
    wsr     a2, PS          /* update PS.OWB to new window base */
    rsync

    _bbci.l a4, 31, _WindowUnderflow4
    rotw    -1              /* original a0 goes to a8 */
    _bbci.l a8, 30, _WindowUnderflow8
    rotw    -1
    j               _WindowUnderflow12

/* Window Overflow Exception for Call8
 *
 * Invoked if a call[i] referenced a register (a4-a15)
 * that contains data from ancestor call[j];
 * call[j] had done a call8 to call[j+1].
 * On entry here:
 * 	window rotated to call[j] start point;
 * 	    a0-a7 are registers to be saved;
 * 	    a8-a15 must be preserved;
 * 	    a9 is call[j+1]'s stack pointer.
 */

    .org    0x80
    .global _WindowOverflow8
_WindowOverflow8:

    s32e    a0, a9, -16     /* save a0 to call[j+1]'s stack frame */
    l32e    a0, a1, -12     /* a0 <- call[j-1]'s sp
                               (used to find end of call[j]'s frame) */
    s32e    a1, a9, -12     /* save a1 to call[j+1]'s stack frame */
    s32e    a2, a9,  -8     /* save a2 to call[j+1]'s stack frame */
    s32e    a3, a9,  -4     /* save a3 to call[j+1]'s stack frame */
    s32e    a4, a0, -32     /* save a4 to call[j]'s stack frame */
    s32e    a5, a0, -28     /* save a5 to call[j]'s stack frame */
    s32e    a6, a0, -24     /* save a6 to call[j]'s stack frame */
    s32e    a7, a0, -20     /* save a7 to call[j]'s stack frame */
    rfwo                    /* rotates back to call[i] position */

/*
 * Window Underflow Exception for Call8
 *
 * Invoked by RETW returning from call[i+1] to call[i]
 * where call[i]'s registers must be reloaded (not live in ARs);
 * where call[i] had done a call8 to call[i+1].
 * On entry here:
 * 	    window rotated to call[i] start point;
 * 	    a0-a7 are undefined, must be reloaded with call[i].reg[0..7];
 * 	    a8-a15 must be preserved (they are call[i+1].reg[0..7]);
 * 	    a9 is call[i+1]'s stack pointer.
 */

    .org    0xC0
    .global _WindowUnderflow8
_WindowUnderflow8:

    l32e    a0, a9, -16     /* restore a0 from call[i+1]'s stack frame */
    l32e    a1, a9, -12     /* restore a1 from call[i+1]'s stack frame */
    l32e    a2, a9,  -8     /* restore a2 from call[i+1]'s stack frame */
    l32e    a7, a1, -12     /* a7 <- call[i-1]'s sp
                               (used to find end of call[i]'s frame) */
    l32e    a3, a9,  -4     /* restore a3 from call[i+1]'s stack frame */
    l32e    a4, a7, -32     /* restore a4 from call[i]'s stack frame */
    l32e    a5, a7, -28     /* restore a5 from call[i]'s stack frame */
    l32e    a6, a7, -24     /* restore a6 from call[i]'s stack frame */
    l32e    a7, a7, -20     /* restore a7 from call[i]'s stack frame */
    rfwu

/*
 * Window Overflow Exception for Call12
 *
 * Invoked if a call[i] referenced a register (a4-a15)
 * that contains data from ancestor call[j];
 * call[j] had done a call12 to call[j+1].
 * On entry here:
 * 	window rotated to call[j] start point;
 * 	    a0-a11 are registers to be saved;
 * 	    a12-a15 must be preserved;
 * 	    a13 is call[j+1]'s stack pointer.
 */

    .org    0x100
    .global _WindowOverflow12
_WindowOverflow12:

    s32e    a0,  a13, -16   /* save a0 to call[j+1]'s stack frame */
    l32e    a0,  a1,  -12   /* a0 <- call[j-1]'s sp
                               (used to find end of call[j]'s frame) */
    s32e    a1,  a13, -12   /* save a1 to call[j+1]'s stack frame */
    s32e    a2,  a13,  -8   /* save a2 to call[j+1]'s stack frame */
    s32e    a3,  a13,  -4   /* save a3 to call[j+1]'s stack frame */
    s32e    a4,  a0,  -48   /* save a4 to end of call[j]'s stack frame */
    s32e    a5,  a0,  -44   /* save a5 to end of call[j]'s stack frame */
    s32e    a6,  a0,  -40   /* save a6 to end of call[j]'s stack frame */
    s32e    a7,  a0,  -36   /* save a7 to end of call[j]'s stack frame */
    s32e    a8,  a0,  -32   /* save a8 to end of call[j]'s stack frame */
    s32e    a9,  a0,  -28   /* save a9 to end of call[j]'s stack frame */
    s32e    a10, a0,  -24   /* save a10 to end of call[j]'s stack frame */
    s32e    a11, a0,  -20   /* save a11 to end of call[j]'s stack frame */
    rfwo                    /* rotates back to call[i] position */

/*
 * Window Underflow Exception for Call12
 *
 * Invoked by RETW returning from call[i+1] to call[i]
 * where call[i]'s registers must be reloaded (not live in ARs);
 * where call[i] had done a call12 to call[i+1].
 * On entry here:
 *	window rotated to call[i] start point;
 *	a0-a11 are undefined, must be reloaded with call[i].reg[0..11];
 *	a12-a15 must be preserved (they are call[i+1].reg[0..3]);
 *	a13 is call[i+1]'s stack pointer.
 */

    .org 0x140
    .global _WindowUnderflow12
_WindowUnderflow12:

    l32e    a0,  a13, -16 /* restore a0 from call[i+1]'s stack frame */
    l32e    a1,  a13, -12 /* restore a1 from call[i+1]'s stack frame */
    l32e    a2,  a13,  -8 /* restore a2 from call[i+1]'s stack frame */
    l32e    a11, a1,  -12 /* a11 <- call[i-1]'s sp
                           * (used to find end of call[i]'s frame) */
    l32e    a3,  a13,  -4 /* restore a3 from call[i+1]'s stack frame */
    l32e    a4,  a11, -48 /* restore a4 from end of call[i]'s stack frame */
    l32e    a5,  a11, -44 /* restore a5 from end of call[i]'s stack frame */
    l32e    a6,  a11, -40 /* restore a6 from end of call[i]'s stack frame */
    l32e    a7,  a11, -36 /* restore a7 from end of call[i]'s stack frame */
    l32e    a8,  a11, -32 /* restore a8 from end of call[i]'s stack frame */
    l32e    a9,  a11, -28 /* restore a9 from end of call[i]'s stack frame */
    l32e    a10, a11, -24 /* restore a10 from end of call[i]'s stack
    		       * frame */
    l32e    a11, a11, -20   /* restore a11 from end of call[i]'s stack
    			 * frame */
    rfwu

#endif /* XCHAL_HAVE_WINDOWED */