xref: /rk3399_ARM-atf/lib/aarch64/misc_helpers.S (revision 1dcc28cfbac5dae3992ad9581f9ea68f6cb339c1)
1/*
2 * Copyright (c) 2013-2017, ARM Limited and Contributors. All rights reserved.
3 *
4 * SPDX-License-Identifier: BSD-3-Clause
5 */
6
7#include <arch.h>
8#include <asm_macros.S>
9#include <assert_macros.S>
10
11	.globl	get_afflvl_shift
12	.globl	mpidr_mask_lower_afflvls
13	.globl	eret
14	.globl	smc
15
16	.globl	zero_normalmem
17	.globl	zeromem
18	.globl	zeromem16
19	.globl	memcpy16
20
21	.globl	disable_mmu_el1
22	.globl	disable_mmu_el3
23	.globl	disable_mmu_icache_el1
24	.globl	disable_mmu_icache_el3
25
26#if SUPPORT_VFP
27	.globl	enable_vfp
28#endif
29
30func get_afflvl_shift
31	cmp	x0, #3
32	cinc	x0, x0, eq
33	mov	x1, #MPIDR_AFFLVL_SHIFT
34	lsl	x0, x0, x1
35	ret
36endfunc get_afflvl_shift
37
38func mpidr_mask_lower_afflvls
39	cmp	x1, #3
40	cinc	x1, x1, eq
41	mov	x2, #MPIDR_AFFLVL_SHIFT
42	lsl	x2, x1, x2
43	lsr	x0, x0, x2
44	lsl	x0, x0, x2
45	ret
46endfunc mpidr_mask_lower_afflvls
47
48
49func eret
50	eret
51endfunc eret
52
53
54func smc
55	smc	#0
56endfunc smc
57
58/* -----------------------------------------------------------------------
59 * void zero_normalmem(void *mem, unsigned int length);
60 *
61 * Initialise a region in normal memory to 0. This functions complies with the
62 * AAPCS and can be called from C code.
63 *
64 * NOTE: MMU must be enabled when using this function as it can only operate on
65 *       normal memory. It is intended to be mainly used from C code when MMU
66 *       is usually enabled.
67 * -----------------------------------------------------------------------
68 */
69.equ	zero_normalmem, zeromem_dczva
70
71/* -----------------------------------------------------------------------
72 * void zeromem(void *mem, unsigned int length);
73 *
74 * Initialise a region of device memory to 0. This functions complies with the
75 * AAPCS and can be called from C code.
76 *
77 * NOTE: When data caches and MMU are enabled, zero_normalmem can usually be
78 *       used instead for faster zeroing.
79 *
80 * -----------------------------------------------------------------------
81 */
82func zeromem
83	/* x2 is the address past the last zeroed address */
84	add	x2, x0, x1
85	/*
86	 * Uses the fallback path that does not use DC ZVA instruction and
87	 * therefore does not need enabled MMU
88	 */
89	b	.Lzeromem_dczva_fallback_entry
90endfunc zeromem
91
92/* -----------------------------------------------------------------------
93 * void zeromem_dczva(void *mem, unsigned int length);
94 *
95 * Fill a region of normal memory of size "length" in bytes with null bytes.
96 * MMU must be enabled and the memory be of
97 * normal type. This is because this function internally uses the DC ZVA
98 * instruction, which generates an Alignment fault if used on any type of
99 * Device memory (see section D3.4.9 of the ARMv8 ARM, issue k). When the MMU
100 * is disabled, all memory behaves like Device-nGnRnE memory (see section
101 * D4.2.8), hence the requirement on the MMU being enabled.
102 * NOTE: The code assumes that the block size as defined in DCZID_EL0
103 *       register is at least 16 bytes.
104 *
105 * -----------------------------------------------------------------------
106 */
107func zeromem_dczva
108
109	/*
110	 * The function consists of a series of loops that zero memory one byte
111	 * at a time, 16 bytes at a time or using the DC ZVA instruction to
112	 * zero aligned block of bytes, which is assumed to be more than 16.
113	 * In the case where the DC ZVA instruction cannot be used or if the
114	 * first 16 bytes loop would overflow, there is fallback path that does
115	 * not use DC ZVA.
116	 * Note: The fallback path is also used by the zeromem function that
117	 *       branches to it directly.
118	 *
119	 *              +---------+   zeromem_dczva
120	 *              |  entry  |
121	 *              +----+----+
122	 *                   |
123	 *                   v
124	 *              +---------+
125	 *              | checks  |>o-------+ (If any check fails, fallback)
126	 *              +----+----+         |
127	 *                   |              |---------------+
128	 *                   v              | Fallback path |
129	 *            +------+------+       |---------------+
130	 *            | 1 byte loop |       |
131	 *            +------+------+ .Lzeromem_dczva_initial_1byte_aligned_end
132	 *                   |              |
133	 *                   v              |
134	 *           +-------+-------+      |
135	 *           | 16 bytes loop |      |
136	 *           +-------+-------+      |
137	 *                   |              |
138	 *                   v              |
139	 *            +------+------+ .Lzeromem_dczva_blocksize_aligned
140	 *            | DC ZVA loop |       |
141	 *            +------+------+       |
142	 *       +--------+  |              |
143	 *       |        |  |              |
144	 *       |        v  v              |
145	 *       |   +-------+-------+ .Lzeromem_dczva_final_16bytes_aligned
146	 *       |   | 16 bytes loop |      |
147	 *       |   +-------+-------+      |
148	 *       |           |              |
149	 *       |           v              |
150	 *       |    +------+------+ .Lzeromem_dczva_final_1byte_aligned
151	 *       |    | 1 byte loop |       |
152	 *       |    +-------------+       |
153	 *       |           |              |
154	 *       |           v              |
155	 *       |       +---+--+           |
156	 *       |       | exit |           |
157	 *       |       +------+           |
158	 *       |			    |
159	 *       |           +--------------+    +------------------+ zeromem
160	 *       |           |  +----------------| zeromem function |
161	 *       |           |  |                +------------------+
162	 *       |           v  v
163	 *       |    +-------------+ .Lzeromem_dczva_fallback_entry
164	 *       |    | 1 byte loop |
165	 *       |    +------+------+
166	 *       |           |
167	 *       +-----------+
168	 */
169
170	/*
171	 * Readable names for registers
172	 *
173	 * Registers x0, x1 and x2 are also set by zeromem which
174	 * branches into the fallback path directly, so cursor, length and
175	 * stop_address should not be retargeted to other registers.
176	 */
177	cursor       .req x0 /* Start address and then current address */
178	length       .req x1 /* Length in bytes of the region to zero out */
179	/* Reusing x1 as length is never used after block_mask is set */
180	block_mask   .req x1 /* Bitmask of the block size read in DCZID_EL0 */
181	stop_address .req x2 /* Address past the last zeroed byte */
182	block_size   .req x3 /* Size of a block in bytes as read in DCZID_EL0 */
183	tmp1         .req x4
184	tmp2         .req x5
185
186#if ENABLE_ASSERTIONS
187	/*
188	 * Check for M bit (MMU enabled) of the current SCTLR_EL(1|3)
189	 * register value and panic if the MMU is disabled.
190	 */
191#if defined(IMAGE_BL1) || defined(IMAGE_BL31) || (defined(IMAGE_BL2) && BL2_AT_EL3)
192	mrs	tmp1, sctlr_el3
193#else
194	mrs	tmp1, sctlr_el1
195#endif
196
197	tst	tmp1, #SCTLR_M_BIT
198	ASM_ASSERT(ne)
199#endif /* ENABLE_ASSERTIONS */
200
201	/* stop_address is the address past the last to zero */
202	add	stop_address, cursor, length
203
204	/*
205	 * Get block_size = (log2(<block size>) >> 2) (see encoding of
206	 * dczid_el0 reg)
207	 */
208	mrs	block_size, dczid_el0
209
210	/*
211	 * Select the 4 lowest bits and convert the extracted log2(<block size
212	 * in words>) to <block size in bytes>
213	 */
214	ubfx	block_size, block_size, #0, #4
215	mov	tmp2, #(1 << 2)
216	lsl	block_size, tmp2, block_size
217
218#if ENABLE_ASSERTIONS
219	/*
220	 * Assumes block size is at least 16 bytes to avoid manual realignment
221	 * of the cursor at the end of the DCZVA loop.
222	 */
223	cmp	block_size, #16
224	ASM_ASSERT(hs)
225#endif
226	/*
227	 * Not worth doing all the setup for a region less than a block and
228	 * protects against zeroing a whole block when the area to zero is
229	 * smaller than that. Also, as it is assumed that the block size is at
230	 * least 16 bytes, this also protects the initial aligning loops from
231	 * trying to zero 16 bytes when length is less than 16.
232	 */
233	cmp	length, block_size
234	b.lo	.Lzeromem_dczva_fallback_entry
235
236	/*
237	 * Calculate the bitmask of the block alignment. It will never
238	 * underflow as the block size is between 4 bytes and 2kB.
239	 * block_mask = block_size - 1
240	 */
241	sub	block_mask, block_size, #1
242
243	/*
244	 * length alias should not be used after this point unless it is
245	 * defined as a register other than block_mask's.
246	 */
247	 .unreq length
248
249	/*
250	 * If the start address is already aligned to zero block size, go
251	 * straight to the cache zeroing loop. This is safe because at this
252	 * point, the length cannot be smaller than a block size.
253	 */
254	tst	cursor, block_mask
255	b.eq	.Lzeromem_dczva_blocksize_aligned
256
257	/*
258	 * Calculate the first block-size-aligned address. It is assumed that
259	 * the zero block size is at least 16 bytes. This address is the last
260	 * address of this initial loop.
261	 */
262	orr	tmp1, cursor, block_mask
263	add	tmp1, tmp1, #1
264
265	/*
266	 * If the addition overflows, skip the cache zeroing loops. This is
267	 * quite unlikely however.
268	 */
269	cbz	tmp1, .Lzeromem_dczva_fallback_entry
270
271	/*
272	 * If the first block-size-aligned address is past the last address,
273	 * fallback to the simpler code.
274	 */
275	cmp	tmp1, stop_address
276	b.hi	.Lzeromem_dczva_fallback_entry
277
278	/*
279	 * If the start address is already aligned to 16 bytes, skip this loop.
280	 * It is safe to do this because tmp1 (the stop address of the initial
281	 * 16 bytes loop) will never be greater than the final stop address.
282	 */
283	tst	cursor, #0xf
284	b.eq	.Lzeromem_dczva_initial_1byte_aligned_end
285
286	/* Calculate the next address aligned to 16 bytes */
287	orr	tmp2, cursor, #0xf
288	add	tmp2, tmp2, #1
289	/* If it overflows, fallback to the simple path (unlikely) */
290	cbz	tmp2, .Lzeromem_dczva_fallback_entry
291	/*
292	 * Next aligned address cannot be after the stop address because the
293	 * length cannot be smaller than 16 at this point.
294	 */
295
296	/* First loop: zero byte per byte */
2971:
298	strb	wzr, [cursor], #1
299	cmp	cursor, tmp2
300	b.ne	1b
301.Lzeromem_dczva_initial_1byte_aligned_end:
302
303	/*
304	 * Second loop: we need to zero 16 bytes at a time from cursor to tmp1
305	 * before being able to use the code that deals with block-size-aligned
306	 * addresses.
307	 */
308	cmp	cursor, tmp1
309	b.hs	2f
3101:
311	stp	xzr, xzr, [cursor], #16
312	cmp	cursor, tmp1
313	b.lo	1b
3142:
315
316	/*
317	 * Third loop: zero a block at a time using DC ZVA cache block zeroing
318	 * instruction.
319	 */
320.Lzeromem_dczva_blocksize_aligned:
321	/*
322	 * Calculate the last block-size-aligned address. If the result equals
323	 * to the start address, the loop will exit immediately.
324	 */
325	bic	tmp1, stop_address, block_mask
326
327	cmp	cursor, tmp1
328	b.hs	2f
3291:
330	/* Zero the block containing the cursor */
331	dc	zva, cursor
332	/* Increment the cursor by the size of a block */
333	add	cursor, cursor, block_size
334	cmp	cursor, tmp1
335	b.lo	1b
3362:
337
338	/*
339	 * Fourth loop: zero 16 bytes at a time and then byte per byte the
340	 * remaining area
341	 */
342.Lzeromem_dczva_final_16bytes_aligned:
343	/*
344	 * Calculate the last 16 bytes aligned address. It is assumed that the
345	 * block size will never be smaller than 16 bytes so that the current
346	 * cursor is aligned to at least 16 bytes boundary.
347	 */
348	bic	tmp1, stop_address, #15
349
350	cmp	cursor, tmp1
351	b.hs	2f
3521:
353	stp	xzr, xzr, [cursor], #16
354	cmp	cursor, tmp1
355	b.lo	1b
3562:
357
358	/* Fifth and final loop: zero byte per byte */
359.Lzeromem_dczva_final_1byte_aligned:
360	cmp	cursor, stop_address
361	b.eq	2f
3621:
363	strb	wzr, [cursor], #1
364	cmp	cursor, stop_address
365	b.ne	1b
3662:
367	ret
368
369	/* Fallback for unaligned start addresses */
370.Lzeromem_dczva_fallback_entry:
371	/*
372	 * If the start address is already aligned to 16 bytes, skip this loop.
373	 */
374	tst	cursor, #0xf
375	b.eq	.Lzeromem_dczva_final_16bytes_aligned
376
377	/* Calculate the next address aligned to 16 bytes */
378	orr	tmp1, cursor, #15
379	add	tmp1, tmp1, #1
380	/* If it overflows, fallback to byte per byte zeroing */
381	cbz	tmp1, .Lzeromem_dczva_final_1byte_aligned
382	/* If the next aligned address is after the stop address, fall back */
383	cmp	tmp1, stop_address
384	b.hs	.Lzeromem_dczva_final_1byte_aligned
385
386	/* Fallback entry loop: zero byte per byte */
3871:
388	strb	wzr, [cursor], #1
389	cmp	cursor, tmp1
390	b.ne	1b
391
392	b	.Lzeromem_dczva_final_16bytes_aligned
393
394	.unreq	cursor
395	/*
396	 * length is already unreq'ed to reuse the register for another
397	 * variable.
398	 */
399	.unreq	stop_address
400	.unreq	block_size
401	.unreq	block_mask
402	.unreq	tmp1
403	.unreq	tmp2
404endfunc zeromem_dczva
405
406/* --------------------------------------------------------------------------
407 * void memcpy16(void *dest, const void *src, unsigned int length)
408 *
409 * Copy length bytes from memory area src to memory area dest.
410 * The memory areas should not overlap.
411 * Destination and source addresses must be 16-byte aligned.
412 * --------------------------------------------------------------------------
413 */
414func memcpy16
415#if ENABLE_ASSERTIONS
416	orr	x3, x0, x1
417	tst	x3, #0xf
418	ASM_ASSERT(eq)
419#endif
420/* copy 16 bytes at a time */
421m_loop16:
422	cmp	x2, #16
423	b.lo	m_loop1
424	ldp	x3, x4, [x1], #16
425	stp	x3, x4, [x0], #16
426	sub	x2, x2, #16
427	b	m_loop16
428/* copy byte per byte */
429m_loop1:
430	cbz	x2, m_end
431	ldrb	w3, [x1], #1
432	strb	w3, [x0], #1
433	subs	x2, x2, #1
434	b.ne	m_loop1
435m_end:
436	ret
437endfunc memcpy16
438
439/* ---------------------------------------------------------------------------
440 * Disable the MMU at EL3
441 * ---------------------------------------------------------------------------
442 */
443
444func disable_mmu_el3
445	mov	x1, #(SCTLR_M_BIT | SCTLR_C_BIT)
446do_disable_mmu_el3:
447	mrs	x0, sctlr_el3
448	bic	x0, x0, x1
449	msr	sctlr_el3, x0
450	isb	/* ensure MMU is off */
451	dsb	sy
452	ret
453endfunc disable_mmu_el3
454
455
456func disable_mmu_icache_el3
457	mov	x1, #(SCTLR_M_BIT | SCTLR_C_BIT | SCTLR_I_BIT)
458	b	do_disable_mmu_el3
459endfunc disable_mmu_icache_el3
460
461/* ---------------------------------------------------------------------------
462 * Disable the MMU at EL1
463 * ---------------------------------------------------------------------------
464 */
465
466func disable_mmu_el1
467	mov	x1, #(SCTLR_M_BIT | SCTLR_C_BIT)
468do_disable_mmu_el1:
469	mrs	x0, sctlr_el1
470	bic	x0, x0, x1
471	msr	sctlr_el1, x0
472	isb	/* ensure MMU is off */
473	dsb	sy
474	ret
475endfunc disable_mmu_el1
476
477
478func disable_mmu_icache_el1
479	mov	x1, #(SCTLR_M_BIT | SCTLR_C_BIT | SCTLR_I_BIT)
480	b	do_disable_mmu_el1
481endfunc disable_mmu_icache_el1
482
483/* ---------------------------------------------------------------------------
484 * Enable the use of VFP at EL3
485 * ---------------------------------------------------------------------------
486 */
487#if SUPPORT_VFP
488func enable_vfp
489	mrs	x0, cpacr_el1
490	orr	x0, x0, #CPACR_VFP_BITS
491	msr	cpacr_el1, x0
492	mrs	x0, cptr_el3
493	mov	x1, #AARCH64_CPTR_TFP
494	bic	x0, x0, x1
495	msr	cptr_el3, x0
496	isb
497	ret
498endfunc enable_vfp
499#endif
500