[thirdparty/gcc.git] / libgcc / config / xtensa / lib2funcs.S

/* Assembly functions for libgcc2.
   Copyright (C) 2001-2022 Free Software Foundation, Inc.
   Contributed by Bob Wilson (bwilson@tensilica.com) at Tensilica.

This file is part of GCC.

GCC is free software; you can redistribute it and/or modify it under
the terms of the GNU General Public License as published by the Free
Software Foundation; either version 3, or (at your option) any later
version.

GCC is distributed in the hope that it will be useful, but WITHOUT ANY
WARRANTY; without even the implied warranty of MERCHANTABILITY or
FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
for more details.

Under Section 7 of GPL version 3, you are granted additional
permissions described in the GCC Runtime Library Exception, version
3.1, as published by the Free Software Foundation.

You should have received a copy of the GNU General Public License and
a copy of the GCC Runtime Library Exception along with this program;
see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see
<http://www.gnu.org/licenses/>.  */

#include "xtensa-config.h"

/* __xtensa_libgcc_window_spill: This function flushes out all but the
   current register window.  This is used to set up the stack so that
   arbitrary frames can be accessed.  */

#if XCHAL_HAVE_WINDOWED && !__XTENSA_CALL0_ABI__
	.align	4
	.global	__xtensa_libgcc_window_spill
	.type	__xtensa_libgcc_window_spill,@function
__xtensa_libgcc_window_spill:
	entry	sp, 48
#if XCHAL_NUM_AREGS > 16
	call12	1f
	retw
	.align	4
1:
	.rept	(XCHAL_NUM_AREGS - 24) / 12
	_entry	sp, 48
	mov	a12, a0
	.endr
	_entry	sp, 16
#if XCHAL_NUM_AREGS % 12 == 0
	mov	a4, a4
#elif XCHAL_NUM_AREGS % 12 == 4
	mov	a8, a8
#elif XCHAL_NUM_AREGS % 12 == 8
	mov	a12, a12
#endif
	retw
#else
	mov	a8, a8
	retw
#endif
	.size	__xtensa_libgcc_window_spill, .-__xtensa_libgcc_window_spill
#endif


/* __xtensa_nonlocal_goto: This code does all the hard work of a
   nonlocal goto on Xtensa.  It is here in the library to avoid the
   code size bloat of generating it in-line.  There are two
   arguments:

	a2 = frame pointer for the procedure containing the label
	a3 = goto handler address

  This function never returns to its caller but instead goes directly
  to the address of the specified goto handler.  */

#if XCHAL_HAVE_WINDOWED && !__XTENSA_CALL0_ABI__
	.align	4
	.global	__xtensa_nonlocal_goto
	.type	__xtensa_nonlocal_goto,@function
__xtensa_nonlocal_goto:
	entry	sp, 32

	/* Flush registers.  */
	call8	__xtensa_libgcc_window_spill

	/* Because the save area for a0-a3 is stored one frame below
	   the one identified by a2, the only way to restore those
	   registers is to unwind the stack.  If alloca() were never
	   called, we could just unwind until finding the sp value
	   matching a2.  However, a2 is a frame pointer, not a stack
	   pointer, and may not be encountered during the unwinding.
	   The solution is to unwind until going _past_ the value
	   given by a2.  This involves keeping three stack pointer
	   values during the unwinding:

		next = sp of frame N-1
		cur = sp of frame N
		prev = sp of frame N+1

	   When next > a2, the desired save area is stored relative
	   to prev.  At this point, cur will be the same as a2
	   except in the alloca() case.

	   Besides finding the values to be restored to a0-a3, we also
	   need to find the current window size for the target
	   function.  This can be extracted from the high bits of the
	   return address, initially in a0.  As the unwinding
	   proceeds, the window size is taken from the value of a0
	   saved _two_ frames below the current frame.  */

	addi	a5, sp, -16	/* a5 = prev - save area */
	l32i	a6, a5, 4
	addi	a6, a6, -16	/* a6 = cur - save area */
	mov	a8, a0		/* a8 = return address (for window size) */
	j	.Lfirstframe

.Lnextframe:
	l32i	a8, a5, 0	/* next return address (for window size) */
	mov	a5, a6		/* advance prev */
	addi	a6, a7, -16	/* advance cur */
.Lfirstframe:
	l32i	a7, a6, 4	/* a7 = next */
	bgeu	a2, a7, .Lnextframe

	/* At this point, prev (a5) points to the save area with the saved
	   values of a0-a3.  Copy those values into the save area at the
	   current sp so they will be reloaded when the return from this
	   function underflows.  We don't have to worry about exceptions
	   while updating the current save area, because the windows have
	   already been flushed.  */

	addi	a4, sp, -16	/* a4 = save area of this function */
	l32i	a6, a5, 0
	l32i	a7, a5, 4
	s32i	a6, a4, 0
	s32i	a7, a4, 4
	l32i	a6, a5, 8
	l32i	a7, a5, 12
	s32i	a6, a4, 8
	s32i	a7, a4, 12

	/* Set return address to goto handler.  Use the window size bits
	   from the return address two frames below the target.  */
	extui	a8, a8, 30, 2	/* get window size from return addr. */
	slli	a3, a3, 2	/* get goto handler addr. << 2 */
	ssai	2
	src	a0, a8, a3	/* combine them with a funnel shift */

	retw
	.size	__xtensa_nonlocal_goto, .-__xtensa_nonlocal_goto
#endif


/* __xtensa_sync_caches: This function is called after writing a trampoline
   on the stack to force all the data writes to memory and invalidate the
   instruction cache. a2 is the address of the new trampoline.

   After the trampoline data is written out, it must be flushed out of
   the data cache into memory.  We use DHWB in case we have a writeback
   cache.  At least one DHWB instruction is needed for each data cache
   line which may be touched by the trampoline.  An ISYNC instruction
   must follow the DHWBs.

   We have to flush the i-cache to make sure that the new values get used.
   At least one IHI instruction is needed for each i-cache line which may
   be touched by the trampoline.  An ISYNC instruction is also needed to
   make sure that the modified instructions are loaded into the instruction
   fetch buffer.  */

/* Use the maximum trampoline size.  Flushing a bit extra is OK.  */
#define TRAMPOLINE_SIZE 60

	.text
	.align	4
	.global	__xtensa_sync_caches
	.type	__xtensa_sync_caches,@function
__xtensa_sync_caches:
#if XCHAL_HAVE_WINDOWED && !__XTENSA_CALL0_ABI__
	entry 	sp, 32
#endif
#if XCHAL_DCACHE_SIZE > 0
	/* Flush the trampoline from the data cache.  */
	extui	a4, a2, 0, XCHAL_DCACHE_LINEWIDTH
	addi	a4, a4, TRAMPOLINE_SIZE
	addi	a4, a4, (1 << XCHAL_DCACHE_LINEWIDTH) - 1
	srli	a4, a4, XCHAL_DCACHE_LINEWIDTH
	mov	a3, a2
.Ldcache_loop:
	dhwb	a3, 0
	addi	a3, a3, (1 << XCHAL_DCACHE_LINEWIDTH)
	addi	a4, a4, -1
	bnez	a4, .Ldcache_loop
	isync
#endif
#if XCHAL_ICACHE_SIZE > 0
	/* Invalidate the corresponding lines in the instruction cache.  */
	extui	a4, a2, 0, XCHAL_ICACHE_LINEWIDTH
	addi	a4, a4, TRAMPOLINE_SIZE
	addi	a4, a4, (1 << XCHAL_ICACHE_LINEWIDTH) - 1
	srli	a4, a4, XCHAL_ICACHE_LINEWIDTH
.Licache_loop:
	ihi	a2, 0
	addi	a2, a2, (1 << XCHAL_ICACHE_LINEWIDTH)
	addi	a4, a4, -1
	bnez	a4, .Licache_loop
#endif
	isync
#if XCHAL_HAVE_WINDOWED && !__XTENSA_CALL0_ABI__
	retw
#else
	ret
#endif
	.size	__xtensa_sync_caches, .-__xtensa_sync_caches
Commit	Line	Data
03984308	1	/* Assembly functions for libgcc2.
7adcbafe	2	Copyright (C) 2001-2022 Free Software Foundation, Inc.
03984308 BW	3	Contributed by Bob Wilson (bwilson@tensilica.com) at Tensilica.
	4
	5	This file is part of GCC.
	6
	7	GCC is free software; you can redistribute it and/or modify it under
	8	the terms of the GNU General Public License as published by the Free
748086b7	9	Software Foundation; either version 3, or (at your option) any later
03984308 BW	10	version.
	11
	12	GCC is distributed in the hope that it will be useful, but WITHOUT ANY
	13	WARRANTY; without even the implied warranty of MERCHANTABILITY or
	14	FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
	15	for more details.
	16
748086b7 JJ	17	Under Section 7 of GPL version 3, you are granted additional
	18	permissions described in the GCC Runtime Library Exception, version
	19	3.1, as published by the Free Software Foundation.
	20
	21	You should have received a copy of the GNU General Public License and
	22	a copy of the GCC Runtime Library Exception along with this program;
	23	see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
	24	<http://www.gnu.org/licenses/>. */
03984308	25
e677f70c	26	#include "xtensa-config.h"
03984308	27
57174693 BW	28	/* __xtensa_libgcc_window_spill: This function flushes out all but the
57174693 BW	29	current register window. This is used to set up the stack so that
ff482c8d	30	arbitrary frames can be accessed. */
03984308	31
590e2636	32	#if XCHAL_HAVE_WINDOWED && !__XTENSA_CALL0_ABI__
57174693 BW	33	.align 4
	34	.global __xtensa_libgcc_window_spill
	35	.type __xtensa_libgcc_window_spill,@function
03984308	36	__xtensa_libgcc_window_spill:
b6ac5f62 MF	37	entry sp, 48
	38	#if XCHAL_NUM_AREGS > 16
	39	call12 1f
	40	retw
	41	.align 4
	42	1:
	43	.rept (XCHAL_NUM_AREGS - 24) / 12
	44	_entry sp, 48
	45	mov a12, a0
	46	.endr
	47	_entry sp, 16
	48	#if XCHAL_NUM_AREGS % 12 == 0
	49	mov a4, a4
	50	#elif XCHAL_NUM_AREGS % 12 == 4
	51	mov a8, a8
	52	#elif XCHAL_NUM_AREGS % 12 == 8
	53	mov a12, a12
	54	#endif
	55	retw
	56	#else
	57	mov a8, a8
03984308	58	retw
b6ac5f62	59	#endif
76f6e118	60	.size __xtensa_libgcc_window_spill, .-__xtensa_libgcc_window_spill
590e2636	61	#endif
03984308 BW	62
	63
	64	/* __xtensa_nonlocal_goto: This code does all the hard work of a
	65	nonlocal goto on Xtensa. It is here in the library to avoid the
	66	code size bloat of generating it in-line. There are two
	67	arguments:
	68
	69	a2 = frame pointer for the procedure containing the label
	70	a3 = goto handler address
	71
	72	This function never returns to its caller but instead goes directly
	73	to the address of the specified goto handler. */
	74
590e2636	75	#if XCHAL_HAVE_WINDOWED && !__XTENSA_CALL0_ABI__
03984308 BW	76	.align 4
	77	.global __xtensa_nonlocal_goto
	78	.type __xtensa_nonlocal_goto,@function
	79	__xtensa_nonlocal_goto:
	80	entry sp, 32
	81
76f6e118	82	/* Flush registers. */
b6ac5f62	83	call8 __xtensa_libgcc_window_spill
03984308 BW	84
	85	/* Because the save area for a0-a3 is stored one frame below
	86	the one identified by a2, the only way to restore those
	87	registers is to unwind the stack. If alloca() were never
	88	called, we could just unwind until finding the sp value
	89	matching a2. However, a2 is a frame pointer, not a stack
	90	pointer, and may not be encountered during the unwinding.
	91	The solution is to unwind until going _past_ the value
	92	given by a2. This involves keeping three stack pointer
	93	values during the unwinding:
	94
	95	next = sp of frame N-1
	96	cur = sp of frame N
	97	prev = sp of frame N+1
	98
	99	When next > a2, the desired save area is stored relative
	100	to prev. At this point, cur will be the same as a2
	101	except in the alloca() case.
	102
	103	Besides finding the values to be restored to a0-a3, we also
	104	need to find the current window size for the target
	105	function. This can be extracted from the high bits of the
	106	return address, initially in a0. As the unwinding
	107	proceeds, the window size is taken from the value of a0
ff482c8d	108	saved _two_ frames below the current frame. */
03984308	109
76f6e118	110	addi a5, sp, -16 /* a5 = prev - save area */
03984308	111	l32i a6, a5, 4
76f6e118 BW	112	addi a6, a6, -16 /* a6 = cur - save area */
76f6e118 BW	113	mov a8, a0 /* a8 = return address (for window size) */
03984308 BW	114	j .Lfirstframe
03984308 BW	115
76f6e118 BW	116	.Lnextframe:
	117	l32i a8, a5, 0 /* next return address (for window size) */
	118	mov a5, a6 /* advance prev */
	119	addi a6, a7, -16 /* advance cur */
	120	.Lfirstframe:
	121	l32i a7, a6, 4 /* a7 = next */
9e8c2a98	122	bgeu a2, a7, .Lnextframe
03984308 BW	123
	124	/* At this point, prev (a5) points to the save area with the saved
	125	values of a0-a3. Copy those values into the save area at the
	126	current sp so they will be reloaded when the return from this
	127	function underflows. We don't have to worry about exceptions
	128	while updating the current save area, because the windows have
ff482c8d	129	already been flushed. */
03984308	130
76f6e118	131	addi a4, sp, -16 /* a4 = save area of this function */
03984308 BW	132	l32i a6, a5, 0
	133	l32i a7, a5, 4
	134	s32i a6, a4, 0
	135	s32i a7, a4, 4
	136	l32i a6, a5, 8
76f6e118	137	l32i a7, a5, 12
03984308 BW	138	s32i a6, a4, 8
03984308 BW	139	s32i a7, a4, 12
76f6e118	140
03984308	141	/* Set return address to goto handler. Use the window size bits
ff482c8d	142	from the return address two frames below the target. */
76f6e118 BW	143	extui a8, a8, 30, 2 /* get window size from return addr. */
76f6e118 BW	144	slli a3, a3, 2 /* get goto handler addr. << 2 */
03984308	145	ssai 2
76f6e118	146	src a0, a8, a3 /* combine them with a funnel shift */
03984308 BW	147
03984308 BW	148	retw
76f6e118	149	.size __xtensa_nonlocal_goto, .-__xtensa_nonlocal_goto
590e2636	150	#endif
03984308 BW	151
	152
	153	/* __xtensa_sync_caches: This function is called after writing a trampoline
	154	on the stack to force all the data writes to memory and invalidate the
	155	instruction cache. a2 is the address of the new trampoline.
	156
	157	After the trampoline data is written out, it must be flushed out of
	158	the data cache into memory. We use DHWB in case we have a writeback
	159	cache. At least one DHWB instruction is needed for each data cache
	160	line which may be touched by the trampoline. An ISYNC instruction
	161	must follow the DHWBs.
	162
	163	We have to flush the i-cache to make sure that the new values get used.
	164	At least one IHI instruction is needed for each i-cache line which may
	165	be touched by the trampoline. An ISYNC instruction is also needed to
	166	make sure that the modified instructions are loaded into the instruction
ff482c8d	167	fetch buffer. */
13fb3a61	168
7f0ee694	169	/* Use the maximum trampoline size. Flushing a bit extra is OK. */
85333688	170	#define TRAMPOLINE_SIZE 60
03984308 BW	171
	172	.text
	173	.align 4
	174	.global __xtensa_sync_caches
	175	.type __xtensa_sync_caches,@function
	176	__xtensa_sync_caches:
590e2636	177	#if XCHAL_HAVE_WINDOWED && !__XTENSA_CALL0_ABI__
03984308	178	entry sp, 32
590e2636	179	#endif
1839841c	180	#if XCHAL_DCACHE_SIZE > 0
76f6e118	181	/* Flush the trampoline from the data cache. */
03984308 BW	182	extui a4, a2, 0, XCHAL_DCACHE_LINEWIDTH
	183	addi a4, a4, TRAMPOLINE_SIZE
	184	addi a4, a4, (1 << XCHAL_DCACHE_LINEWIDTH) - 1
	185	srli a4, a4, XCHAL_DCACHE_LINEWIDTH
	186	mov a3, a2
	187	.Ldcache_loop:
	188	dhwb a3, 0
	189	addi a3, a3, (1 << XCHAL_DCACHE_LINEWIDTH)
	190	addi a4, a4, -1
	191	bnez a4, .Ldcache_loop
	192	isync
76f6e118	193	#endif
03984308	194	#if XCHAL_ICACHE_SIZE > 0
76f6e118	195	/* Invalidate the corresponding lines in the instruction cache. */
03984308 BW	196	extui a4, a2, 0, XCHAL_ICACHE_LINEWIDTH
	197	addi a4, a4, TRAMPOLINE_SIZE
	198	addi a4, a4, (1 << XCHAL_ICACHE_LINEWIDTH) - 1
	199	srli a4, a4, XCHAL_ICACHE_LINEWIDTH
	200	.Licache_loop:
	201	ihi a2, 0
	202	addi a2, a2, (1 << XCHAL_ICACHE_LINEWIDTH)
	203	addi a4, a4, -1
	204	bnez a4, .Licache_loop
03984308	205	#endif
9502767b	206	isync
590e2636	207	#if XCHAL_HAVE_WINDOWED && !__XTENSA_CALL0_ABI__
03984308	208	retw
590e2636 MF	209	#else
	210	ret
	211	#endif
76f6e118	212	.size __xtensa_sync_caches, .-__xtensa_sync_caches