xref: /rk3399_ARM-atf/lib/aarch64/misc_helpers.S (revision b4cf974a3256275fe2c03d8eaaf07a5e5b337cfc)
1/*
2 * Copyright (c) 2013-2018, 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#include <xlat_tables_defs.h>
11
12	.globl	get_afflvl_shift
13	.globl	mpidr_mask_lower_afflvls
14	.globl	eret
15	.globl	smc
16
17	.globl	zero_normalmem
18	.globl	zeromem
19	.globl	zeromem16
20	.globl	memcpy16
21
22	.globl	disable_mmu_el1
23	.globl	disable_mmu_el3
24	.globl	disable_mmu_icache_el1
25	.globl	disable_mmu_icache_el3
26
27	.globl	fixup_gdt_reloc
28
29#if SUPPORT_VFP
30	.globl	enable_vfp
31#endif
32
33func get_afflvl_shift
34	cmp	x0, #3
35	cinc	x0, x0, eq
36	mov	x1, #MPIDR_AFFLVL_SHIFT
37	lsl	x0, x0, x1
38	ret
39endfunc get_afflvl_shift
40
41func mpidr_mask_lower_afflvls
42	cmp	x1, #3
43	cinc	x1, x1, eq
44	mov	x2, #MPIDR_AFFLVL_SHIFT
45	lsl	x2, x1, x2
46	lsr	x0, x0, x2
47	lsl	x0, x0, x2
48	ret
49endfunc mpidr_mask_lower_afflvls
50
51
52func eret
53	eret
54endfunc eret
55
56
57func smc
58	smc	#0
59endfunc smc
60
61/* -----------------------------------------------------------------------
62 * void zero_normalmem(void *mem, unsigned int length);
63 *
64 * Initialise a region in normal memory to 0. This functions complies with the
65 * AAPCS and can be called from C code.
66 *
67 * NOTE: MMU must be enabled when using this function as it can only operate on
68 *       normal memory. It is intended to be mainly used from C code when MMU
69 *       is usually enabled.
70 * -----------------------------------------------------------------------
71 */
72.equ	zero_normalmem, zeromem_dczva
73
74/* -----------------------------------------------------------------------
75 * void zeromem(void *mem, unsigned int length);
76 *
77 * Initialise a region of device memory to 0. This functions complies with the
78 * AAPCS and can be called from C code.
79 *
80 * NOTE: When data caches and MMU are enabled, zero_normalmem can usually be
81 *       used instead for faster zeroing.
82 *
83 * -----------------------------------------------------------------------
84 */
85func zeromem
86	/* x2 is the address past the last zeroed address */
87	add	x2, x0, x1
88	/*
89	 * Uses the fallback path that does not use DC ZVA instruction and
90	 * therefore does not need enabled MMU
91	 */
92	b	.Lzeromem_dczva_fallback_entry
93endfunc zeromem
94
95/* -----------------------------------------------------------------------
96 * void zeromem_dczva(void *mem, unsigned int length);
97 *
98 * Fill a region of normal memory of size "length" in bytes with null bytes.
99 * MMU must be enabled and the memory be of
100 * normal type. This is because this function internally uses the DC ZVA
101 * instruction, which generates an Alignment fault if used on any type of
102 * Device memory (see section D3.4.9 of the ARMv8 ARM, issue k). When the MMU
103 * is disabled, all memory behaves like Device-nGnRnE memory (see section
104 * D4.2.8), hence the requirement on the MMU being enabled.
105 * NOTE: The code assumes that the block size as defined in DCZID_EL0
106 *       register is at least 16 bytes.
107 *
108 * -----------------------------------------------------------------------
109 */
110func zeromem_dczva
111
112	/*
113	 * The function consists of a series of loops that zero memory one byte
114	 * at a time, 16 bytes at a time or using the DC ZVA instruction to
115	 * zero aligned block of bytes, which is assumed to be more than 16.
116	 * In the case where the DC ZVA instruction cannot be used or if the
117	 * first 16 bytes loop would overflow, there is fallback path that does
118	 * not use DC ZVA.
119	 * Note: The fallback path is also used by the zeromem function that
120	 *       branches to it directly.
121	 *
122	 *              +---------+   zeromem_dczva
123	 *              |  entry  |
124	 *              +----+----+
125	 *                   |
126	 *                   v
127	 *              +---------+
128	 *              | checks  |>o-------+ (If any check fails, fallback)
129	 *              +----+----+         |
130	 *                   |              |---------------+
131	 *                   v              | Fallback path |
132	 *            +------+------+       |---------------+
133	 *            | 1 byte loop |       |
134	 *            +------+------+ .Lzeromem_dczva_initial_1byte_aligned_end
135	 *                   |              |
136	 *                   v              |
137	 *           +-------+-------+      |
138	 *           | 16 bytes loop |      |
139	 *           +-------+-------+      |
140	 *                   |              |
141	 *                   v              |
142	 *            +------+------+ .Lzeromem_dczva_blocksize_aligned
143	 *            | DC ZVA loop |       |
144	 *            +------+------+       |
145	 *       +--------+  |              |
146	 *       |        |  |              |
147	 *       |        v  v              |
148	 *       |   +-------+-------+ .Lzeromem_dczva_final_16bytes_aligned
149	 *       |   | 16 bytes loop |      |
150	 *       |   +-------+-------+      |
151	 *       |           |              |
152	 *       |           v              |
153	 *       |    +------+------+ .Lzeromem_dczva_final_1byte_aligned
154	 *       |    | 1 byte loop |       |
155	 *       |    +-------------+       |
156	 *       |           |              |
157	 *       |           v              |
158	 *       |       +---+--+           |
159	 *       |       | exit |           |
160	 *       |       +------+           |
161	 *       |			    |
162	 *       |           +--------------+    +------------------+ zeromem
163	 *       |           |  +----------------| zeromem function |
164	 *       |           |  |                +------------------+
165	 *       |           v  v
166	 *       |    +-------------+ .Lzeromem_dczva_fallback_entry
167	 *       |    | 1 byte loop |
168	 *       |    +------+------+
169	 *       |           |
170	 *       +-----------+
171	 */
172
173	/*
174	 * Readable names for registers
175	 *
176	 * Registers x0, x1 and x2 are also set by zeromem which
177	 * branches into the fallback path directly, so cursor, length and
178	 * stop_address should not be retargeted to other registers.
179	 */
180	cursor       .req x0 /* Start address and then current address */
181	length       .req x1 /* Length in bytes of the region to zero out */
182	/* Reusing x1 as length is never used after block_mask is set */
183	block_mask   .req x1 /* Bitmask of the block size read in DCZID_EL0 */
184	stop_address .req x2 /* Address past the last zeroed byte */
185	block_size   .req x3 /* Size of a block in bytes as read in DCZID_EL0 */
186	tmp1         .req x4
187	tmp2         .req x5
188
189#if ENABLE_ASSERTIONS
190	/*
191	 * Check for M bit (MMU enabled) of the current SCTLR_EL(1|3)
192	 * register value and panic if the MMU is disabled.
193	 */
194#if defined(IMAGE_BL1) || defined(IMAGE_BL31) || (defined(IMAGE_BL2) && BL2_AT_EL3)
195	mrs	tmp1, sctlr_el3
196#else
197	mrs	tmp1, sctlr_el1
198#endif
199
200	tst	tmp1, #SCTLR_M_BIT
201	ASM_ASSERT(ne)
202#endif /* ENABLE_ASSERTIONS */
203
204	/* stop_address is the address past the last to zero */
205	add	stop_address, cursor, length
206
207	/*
208	 * Get block_size = (log2(<block size>) >> 2) (see encoding of
209	 * dczid_el0 reg)
210	 */
211	mrs	block_size, dczid_el0
212
213	/*
214	 * Select the 4 lowest bits and convert the extracted log2(<block size
215	 * in words>) to <block size in bytes>
216	 */
217	ubfx	block_size, block_size, #0, #4
218	mov	tmp2, #(1 << 2)
219	lsl	block_size, tmp2, block_size
220
221#if ENABLE_ASSERTIONS
222	/*
223	 * Assumes block size is at least 16 bytes to avoid manual realignment
224	 * of the cursor at the end of the DCZVA loop.
225	 */
226	cmp	block_size, #16
227	ASM_ASSERT(hs)
228#endif
229	/*
230	 * Not worth doing all the setup for a region less than a block and
231	 * protects against zeroing a whole block when the area to zero is
232	 * smaller than that. Also, as it is assumed that the block size is at
233	 * least 16 bytes, this also protects the initial aligning loops from
234	 * trying to zero 16 bytes when length is less than 16.
235	 */
236	cmp	length, block_size
237	b.lo	.Lzeromem_dczva_fallback_entry
238
239	/*
240	 * Calculate the bitmask of the block alignment. It will never
241	 * underflow as the block size is between 4 bytes and 2kB.
242	 * block_mask = block_size - 1
243	 */
244	sub	block_mask, block_size, #1
245
246	/*
247	 * length alias should not be used after this point unless it is
248	 * defined as a register other than block_mask's.
249	 */
250	 .unreq length
251
252	/*
253	 * If the start address is already aligned to zero block size, go
254	 * straight to the cache zeroing loop. This is safe because at this
255	 * point, the length cannot be smaller than a block size.
256	 */
257	tst	cursor, block_mask
258	b.eq	.Lzeromem_dczva_blocksize_aligned
259
260	/*
261	 * Calculate the first block-size-aligned address. It is assumed that
262	 * the zero block size is at least 16 bytes. This address is the last
263	 * address of this initial loop.
264	 */
265	orr	tmp1, cursor, block_mask
266	add	tmp1, tmp1, #1
267
268	/*
269	 * If the addition overflows, skip the cache zeroing loops. This is
270	 * quite unlikely however.
271	 */
272	cbz	tmp1, .Lzeromem_dczva_fallback_entry
273
274	/*
275	 * If the first block-size-aligned address is past the last address,
276	 * fallback to the simpler code.
277	 */
278	cmp	tmp1, stop_address
279	b.hi	.Lzeromem_dczva_fallback_entry
280
281	/*
282	 * If the start address is already aligned to 16 bytes, skip this loop.
283	 * It is safe to do this because tmp1 (the stop address of the initial
284	 * 16 bytes loop) will never be greater than the final stop address.
285	 */
286	tst	cursor, #0xf
287	b.eq	.Lzeromem_dczva_initial_1byte_aligned_end
288
289	/* Calculate the next address aligned to 16 bytes */
290	orr	tmp2, cursor, #0xf
291	add	tmp2, tmp2, #1
292	/* If it overflows, fallback to the simple path (unlikely) */
293	cbz	tmp2, .Lzeromem_dczva_fallback_entry
294	/*
295	 * Next aligned address cannot be after the stop address because the
296	 * length cannot be smaller than 16 at this point.
297	 */
298
299	/* First loop: zero byte per byte */
3001:
301	strb	wzr, [cursor], #1
302	cmp	cursor, tmp2
303	b.ne	1b
304.Lzeromem_dczva_initial_1byte_aligned_end:
305
306	/*
307	 * Second loop: we need to zero 16 bytes at a time from cursor to tmp1
308	 * before being able to use the code that deals with block-size-aligned
309	 * addresses.
310	 */
311	cmp	cursor, tmp1
312	b.hs	2f
3131:
314	stp	xzr, xzr, [cursor], #16
315	cmp	cursor, tmp1
316	b.lo	1b
3172:
318
319	/*
320	 * Third loop: zero a block at a time using DC ZVA cache block zeroing
321	 * instruction.
322	 */
323.Lzeromem_dczva_blocksize_aligned:
324	/*
325	 * Calculate the last block-size-aligned address. If the result equals
326	 * to the start address, the loop will exit immediately.
327	 */
328	bic	tmp1, stop_address, block_mask
329
330	cmp	cursor, tmp1
331	b.hs	2f
3321:
333	/* Zero the block containing the cursor */
334	dc	zva, cursor
335	/* Increment the cursor by the size of a block */
336	add	cursor, cursor, block_size
337	cmp	cursor, tmp1
338	b.lo	1b
3392:
340
341	/*
342	 * Fourth loop: zero 16 bytes at a time and then byte per byte the
343	 * remaining area
344	 */
345.Lzeromem_dczva_final_16bytes_aligned:
346	/*
347	 * Calculate the last 16 bytes aligned address. It is assumed that the
348	 * block size will never be smaller than 16 bytes so that the current
349	 * cursor is aligned to at least 16 bytes boundary.
350	 */
351	bic	tmp1, stop_address, #15
352
353	cmp	cursor, tmp1
354	b.hs	2f
3551:
356	stp	xzr, xzr, [cursor], #16
357	cmp	cursor, tmp1
358	b.lo	1b
3592:
360
361	/* Fifth and final loop: zero byte per byte */
362.Lzeromem_dczva_final_1byte_aligned:
363	cmp	cursor, stop_address
364	b.eq	2f
3651:
366	strb	wzr, [cursor], #1
367	cmp	cursor, stop_address
368	b.ne	1b
3692:
370	ret
371
372	/* Fallback for unaligned start addresses */
373.Lzeromem_dczva_fallback_entry:
374	/*
375	 * If the start address is already aligned to 16 bytes, skip this loop.
376	 */
377	tst	cursor, #0xf
378	b.eq	.Lzeromem_dczva_final_16bytes_aligned
379
380	/* Calculate the next address aligned to 16 bytes */
381	orr	tmp1, cursor, #15
382	add	tmp1, tmp1, #1
383	/* If it overflows, fallback to byte per byte zeroing */
384	cbz	tmp1, .Lzeromem_dczva_final_1byte_aligned
385	/* If the next aligned address is after the stop address, fall back */
386	cmp	tmp1, stop_address
387	b.hs	.Lzeromem_dczva_final_1byte_aligned
388
389	/* Fallback entry loop: zero byte per byte */
3901:
391	strb	wzr, [cursor], #1
392	cmp	cursor, tmp1
393	b.ne	1b
394
395	b	.Lzeromem_dczva_final_16bytes_aligned
396
397	.unreq	cursor
398	/*
399	 * length is already unreq'ed to reuse the register for another
400	 * variable.
401	 */
402	.unreq	stop_address
403	.unreq	block_size
404	.unreq	block_mask
405	.unreq	tmp1
406	.unreq	tmp2
407endfunc zeromem_dczva
408
409/* --------------------------------------------------------------------------
410 * void memcpy16(void *dest, const void *src, unsigned int length)
411 *
412 * Copy length bytes from memory area src to memory area dest.
413 * The memory areas should not overlap.
414 * Destination and source addresses must be 16-byte aligned.
415 * --------------------------------------------------------------------------
416 */
417func memcpy16
418#if ENABLE_ASSERTIONS
419	orr	x3, x0, x1
420	tst	x3, #0xf
421	ASM_ASSERT(eq)
422#endif
423/* copy 16 bytes at a time */
424m_loop16:
425	cmp	x2, #16
426	b.lo	m_loop1
427	ldp	x3, x4, [x1], #16
428	stp	x3, x4, [x0], #16
429	sub	x2, x2, #16
430	b	m_loop16
431/* copy byte per byte */
432m_loop1:
433	cbz	x2, m_end
434	ldrb	w3, [x1], #1
435	strb	w3, [x0], #1
436	subs	x2, x2, #1
437	b.ne	m_loop1
438m_end:
439	ret
440endfunc memcpy16
441
442/* ---------------------------------------------------------------------------
443 * Disable the MMU at EL3
444 * ---------------------------------------------------------------------------
445 */
446
447func disable_mmu_el3
448	mov	x1, #(SCTLR_M_BIT | SCTLR_C_BIT)
449do_disable_mmu_el3:
450	mrs	x0, sctlr_el3
451	bic	x0, x0, x1
452	msr	sctlr_el3, x0
453	isb	/* ensure MMU is off */
454	dsb	sy
455	ret
456endfunc disable_mmu_el3
457
458
459func disable_mmu_icache_el3
460	mov	x1, #(SCTLR_M_BIT | SCTLR_C_BIT | SCTLR_I_BIT)
461	b	do_disable_mmu_el3
462endfunc disable_mmu_icache_el3
463
464/* ---------------------------------------------------------------------------
465 * Disable the MMU at EL1
466 * ---------------------------------------------------------------------------
467 */
468
469func disable_mmu_el1
470	mov	x1, #(SCTLR_M_BIT | SCTLR_C_BIT)
471do_disable_mmu_el1:
472	mrs	x0, sctlr_el1
473	bic	x0, x0, x1
474	msr	sctlr_el1, x0
475	isb	/* ensure MMU is off */
476	dsb	sy
477	ret
478endfunc disable_mmu_el1
479
480
481func disable_mmu_icache_el1
482	mov	x1, #(SCTLR_M_BIT | SCTLR_C_BIT | SCTLR_I_BIT)
483	b	do_disable_mmu_el1
484endfunc disable_mmu_icache_el1
485
486/* ---------------------------------------------------------------------------
487 * Enable the use of VFP at EL3
488 * ---------------------------------------------------------------------------
489 */
490#if SUPPORT_VFP
491func enable_vfp
492	mrs	x0, cpacr_el1
493	orr	x0, x0, #CPACR_VFP_BITS
494	msr	cpacr_el1, x0
495	mrs	x0, cptr_el3
496	mov	x1, #AARCH64_CPTR_TFP
497	bic	x0, x0, x1
498	msr	cptr_el3, x0
499	isb
500	ret
501endfunc enable_vfp
502#endif
503
504/* ---------------------------------------------------------------------------
505 * Helper to fixup Global Descriptor table (GDT) and dynamic relocations
506 * (.rela.dyn) at runtime.
507 *
508 * This function is meant to be used when the firmware is compiled with -fpie
509 * and linked with -pie options. We rely on the linker script exporting
510 * appropriate markers for start and end of the section. For GOT, we
511 * expect __GOT_START__ and __GOT_END__. Similarly for .rela.dyn, we expect
512 * __RELA_START__ and __RELA_END__.
513 *
514 * The function takes the limits of the memory to apply fixups to as
515 * arguments (which is usually the limits of the relocable BL image).
516 *   x0 -  the start of the fixup region
517 *   x1 -  the limit of the fixup region
518 * These addresses have to be page (4KB aligned).
519 * ---------------------------------------------------------------------------
520 */
521func fixup_gdt_reloc
522	mov	x6, x0
523	mov	x7, x1
524
525	/* Test if the limits are 4K aligned */
526#if ENABLE_ASSERTIONS
527	orr	x0, x0, x1
528	tst	x0, #(PAGE_SIZE - 1)
529	ASM_ASSERT(eq)
530#endif
531	/*
532	 * Calculate the offset based on return address in x30.
533	 * Assume that this funtion is called within a page of the start of
534	 * of fixup region.
535	 */
536	and	x2, x30, #~(PAGE_SIZE - 1)
537	sub	x0, x2, x6	/* Diff(S) = Current Address - Compiled Address */
538
539	adrp	x1, __GOT_START__
540	add	x1, x1, :lo12:__GOT_START__
541	adrp	x2, __GOT_END__
542	add	x2, x2, :lo12:__GOT_END__
543
544	/*
545	 * GOT is an array of 64_bit addresses which must be fixed up as
546	 * new_addr = old_addr + Diff(S).
547	 * The new_addr is the address currently the binary is executing from
548	 * and old_addr is the address at compile time.
549	 */
5501:
551	ldr	x3, [x1]
552	/* Skip adding offset if address is < lower limit */
553	cmp	x3, x6
554	b.lo	2f
555	/* Skip adding offset if address is >= upper limit */
556	cmp	x3, x7
557	b.ge	2f
558	add	x3, x3, x0
559	str	x3, [x1]
5602:
561	add	x1, x1, #8
562	cmp	x1, x2
563	b.lo	1b
564
565	/* Starting dynamic relocations. Use adrp/adr to get RELA_START and END */
566	adrp	x1, __RELA_START__
567	add	x1, x1, :lo12:__RELA_START__
568	adrp	x2, __RELA_END__
569	add	x2, x2, :lo12:__RELA_END__
570	/*
571	 * According to ELF-64 specification, the RELA data structure is as
572	 * follows:
573	 *	typedef struct
574	 * 	{
575	 *		Elf64_Addr r_offset;
576	 *		Elf64_Xword r_info;
577	 *		Elf64_Sxword r_addend;
578	 *	} Elf64_Rela;
579	 *
580	 * r_offset is address of reference
581	 * r_info is symbol index and type of relocation (in this case
582	 * 0x403 which corresponds to R_AARCH64_RELATIV).
583	 * r_addend is constant part of expression.
584	 *
585	 * Size of Elf64_Rela structure is 24 bytes.
586	 */
5871:
588	/* Assert that the relocation type is R_AARCH64_RELATIV */
589#if ENABLE_ASSERTIONS
590	ldr	x3, [x1, #8]
591	cmp	x3, #0x403
592	ASM_ASSERT(eq)
593#endif
594	ldr	x3, [x1]	/* r_offset */
595	add	x3, x0, x3
596	ldr	x4, [x1, #16]	/* r_addend */
597
598	/* Skip adding offset if r_addend is < lower limit */
599	cmp	x4, x6
600	b.lo	2f
601	/* Skip adding offset if r_addend entry is >= upper limit */
602	cmp	x4, x7
603	b.ge	2f
604
605	add	x4, x0, x4	/* Diff(S) + r_addend */
606	str	x4, [x3]
607
6082:	add	x1, x1, #24
609	cmp	x1, x2
610	b.lo	1b
611
612	ret
613endfunc fixup_gdt_reloc
614