xref: /freebsd/sys/arm64/arm64/locore.S (revision 05f6f65c3bdaf241ec819babda666445bf8e9909)
1/*-
2 * Copyright (c) 2012-2014 Andrew Turner
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 *    notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 *    notice, this list of conditions and the following disclaimer in the
12 *    documentation and/or other materials provided with the distribution.
13 *
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24 * SUCH DAMAGE.
25 */
26
27#include "assym.inc"
28#include "opt_kstack_pages.h"
29#include <sys/elf_common.h>
30#include <sys/syscall.h>
31#include <machine/asm.h>
32#include <machine/armreg.h>
33#include <machine/cpu.h>
34#include <machine/hypervisor.h>
35#include <machine/param.h>
36#include <machine/pte.h>
37#include <machine/vm.h>
38#include <machine/vmparam.h>
39
40#define	VIRT_BITS	48
41
42#if PAGE_SIZE == PAGE_SIZE_16K
43/*
44 * The number of level 3 tables to create. 32 will allow for 1G of address
45 * space, the same as a single level 2 page with 4k pages.
46 */
47#define	L3_PAGE_COUNT	32
48#endif
49
50/*
51 * The size of our bootstrap stack.
52 */
53#define	BOOT_STACK_SIZE	(KSTACK_PAGES * PAGE_SIZE)
54
55	.globl	kernbase
56	.set	kernbase, KERNBASE
57
58/*
59 * We assume:
60 *  MMU      on with an identity map, or off
61 *  D-Cache: off
62 *  I-Cache: on or off
63 *  We are loaded at a 2MiB aligned address
64 */
65
66ENTRY(_start)
67	/* Enter the kernel exception level */
68	bl	enter_kernel_el
69
70	/* Set the context id */
71	msr	contextidr_el1, xzr
72
73	/* Get the virt -> phys offset */
74	bl	get_load_phys_addr
75
76	/*
77	 * At this point:
78	 * x28 = Our physical load address
79	 */
80
81	/* Create the page tables */
82	bl	create_pagetables
83
84	/*
85	 * At this point:
86	 * x27 = TTBR0 table
87	 * x26 = Kernel L1 table
88	 * x24 = TTBR1 table
89	 * x22 = PTE shareability attributes
90	 */
91
92	/* Enable the mmu */
93	bl	start_mmu
94
95	/* Load the new ttbr0 pagetable */
96	adrp	x27, pagetable_l0_ttbr0
97	add	x27, x27, :lo12:pagetable_l0_ttbr0
98
99	/* Jump to the virtual address space */
100	ldr	x15, .Lvirtdone
101	br	x15
102
103virtdone:
104	BTI_J
105
106	/* Set up the stack */
107	adrp	x25, initstack_end
108	add	x25, x25, :lo12:initstack_end
109	sub	sp, x25, #PCB_SIZE
110
111	/* Zero the BSS */
112	ldr	x15, .Lbss
113	ldr	x14, .Lend
1141:
115	stp	xzr, xzr, [x15], #16
116	cmp	x15, x14
117	b.lo	1b
118
119#if defined(PERTHREAD_SSP)
120	/* Set sp_el0 to the boot canary for early per-thread SSP to work */
121	adrp	x15, boot_canary
122	add	x15, x15, :lo12:boot_canary
123	msr	sp_el0, x15
124#endif
125
126	/* Backup the module pointer */
127	mov	x1, x0
128
129	sub	sp, sp, #BOOTPARAMS_SIZE
130	mov	x0, sp
131
132	str	x1,  [x0, #BP_MODULEP]
133	adrp	x25, initstack
134	add	x25, x25, :lo12:initstack
135	str	x25, [x0, #BP_KERN_STACK]
136	str	x27, [x0, #BP_KERN_TTBR0]
137	str	x23, [x0, #BP_BOOT_EL]
138
139	/* Set this before it's used in kasan_init_early */
140	adrp	x1, pmap_sh_attr
141	str	x22, [x1, :lo12:pmap_sh_attr]
142
143#ifdef KASAN
144	/* Save bootparams */
145	mov	x19, x0
146
147	/* Bootstrap an early shadow map for the boot stack. */
148	ldr	x0, [x0, #BP_KERN_STACK]
149	ldr	x1, =BOOT_STACK_SIZE
150	bl	kasan_init_early
151
152	/* Restore bootparams */
153	mov	x0, x19
154#endif
155
156	/* trace back starts here */
157	mov	fp, #0
158	/* Branch to C code */
159	bl	initarm
160	/* We are done with the boot params */
161	add	sp, sp, #BOOTPARAMS_SIZE
162
163	/*
164	 * Enable pointer authentication in the kernel. We set the keys for
165	 * thread0 in initarm so have to wait until it returns to enable it.
166	 * If we were to enable it in initarm then any authentication when
167	 * returning would fail as it was called with pointer authentication
168	 * disabled.
169	 */
170	bl	ptrauth_start
171
172	bl	mi_startup
173
174	/* We should not get here */
175	brk	0
176
177	.align 3
178.Lvirtdone:
179	.quad	virtdone
180.Lbss:
181	.quad	__bss_start
182.Lend:
183	.quad	__bss_end
184END(_start)
185
186#ifdef SMP
187/*
188 * void
189 * mpentry_psci(unsigned long)
190 *
191 * Called by a core when it is being brought online with psci.
192 * The data in x0 is passed straight to init_secondary.
193 */
194ENTRY(mpentry_psci)
195	mov	x26, xzr
196	b	mpentry_common
197END(mpentry_psci)
198
199/*
200 * void
201 * mpentry_spintable(void)
202 *
203 * Called by a core when it is being brought online with a spin-table.
204 * Reads the new CPU ID and passes this to init_secondary.
205 */
206ENTRY(mpentry_spintable)
207	ldr	x26, =spintable_wait
208	b	mpentry_common
209END(mpentry_spintable)
210
211/* Wait for the current CPU to be released */
212LENTRY(spintable_wait)
213	/* Read the affinity bits from mpidr_el1 */
214	mrs	x1, mpidr_el1
215	ldr	x2, =CPU_AFF_MASK
216	and	x1, x1, x2
217
218	adrp	x2, ap_cpuid
2191:
220	ldr	x0, [x2, :lo12:ap_cpuid]
221	cmp	x0, x1
222	b.ne	1b
223
224	str	xzr, [x2, :lo12:ap_cpuid]
225	dsb	sy
226	sev
227
228	ret
229LEND(mpentry_spintable)
230
231LENTRY(mpentry_common)
232	/* Disable interrupts */
233	msr	daifset, #DAIF_INTR
234
235	/* Enter the kernel exception level */
236	bl	enter_kernel_el
237
238	/* Set the context id */
239	msr	contextidr_el1, xzr
240
241	/* Load the kernel page table */
242	adrp	x24, pagetable_l0_ttbr1
243	add	x24, x24, :lo12:pagetable_l0_ttbr1
244	/* Load the identity page table */
245	adrp	x27, pagetable_l0_ttbr0_bootstrap
246	add	x27, x27, :lo12:pagetable_l0_ttbr0_bootstrap
247
248	/* Enable the mmu */
249	bl	start_mmu
250
251	/* Load the new ttbr0 pagetable */
252	adrp	x27, pagetable_l0_ttbr0
253	add	x27, x27, :lo12:pagetable_l0_ttbr0
254
255	/* Jump to the virtual address space */
256	ldr	x15, =mp_virtdone
257	br	x15
258
259mp_virtdone:
260	BTI_J
261
262	/*
263	 * Allow this CPU to wait until the kernel is ready for it,
264	 * e.g. with spin-table but each CPU uses the same release address
265	 */
266	cbz	x26, 1f
267	blr	x26
2681:
269
270	/* Start using the AP boot stack */
271	adrp	x4, bootstack
272	ldr	x4, [x4, :lo12:bootstack]
273	mov	sp, x4
274
275#if defined(PERTHREAD_SSP)
276	/* Set sp_el0 to the boot canary for early per-thread SSP to work */
277	adrp	x15, boot_canary
278	add	x15, x15, :lo12:boot_canary
279	msr	sp_el0, x15
280#endif
281
282	/* Load the kernel ttbr0 pagetable */
283	msr	ttbr0_el1, x27
284	isb
285
286	/* Invalidate the TLB */
287	tlbi	vmalle1
288	dsb	sy
289	isb
290
291	/*
292	 * Initialize the per-CPU pointer before calling into C code, for the
293	 * benefit of kernel sanitizers.
294	 */
295	adrp	x18, bootpcpu
296	ldr	x18, [x18, :lo12:bootpcpu]
297	msr	tpidr_el1, x18
298
299	b	init_secondary
300LEND(mpentry_common)
301#endif
302
303/*
304 * Enter the exception level the kernel will use:
305 *
306 *  - If in EL1 continue in EL1
307 *  - If the CPU supports FEAT_VHE then set HCR_E2H and HCR_TGE and continue
308 *    in EL2
309 *  - Configure EL2 to support running the kernel at EL1 and exit to that
310 */
311LENTRY(enter_kernel_el)
312#define	INIT_SCTLR_EL1	(SCTLR_LSMAOE | SCTLR_nTLSMD | SCTLR_EIS | \
313    SCTLR_TSCXT | SCTLR_EOS)
314	mrs	x23, CurrentEL
315	and	x23, x23, #(CURRENTEL_EL_MASK)
316	cmp	x23, #(CURRENTEL_EL_EL2)
317	b.eq	1f
318
319	ldr	x2, =INIT_SCTLR_EL1
320	msr	sctlr_el1, x2
321	/* SCTLR_EOS is set so eret is a context synchronizing event so we
322	 * need an isb here to ensure it's observed by later instructions,
323	 * but don't need it in the eret below.
324	 */
325	isb
326
327	/* Ensure SPSR_EL1 and pstate are in sync. The only wat to set the
328	 * latter is to set the former and return from an exception with eret.
329	 */
330	mov	x2, #(PSR_DAIF | PSR_M_EL1h)
331	msr	spsr_el1, x2
332	msr	elr_el1, lr
333	eret
334
3351:
336	dsb	sy
337	/*
338	 * Set just the reserved bits in sctlr_el2. This will disable the
339	 * MMU which may have broken the kernel if we enter the kernel in
340	 * EL2, e.g. when using VHE.
341	 */
342	ldr	x2, =(SCTLR_EL2_RES1 | SCTLR_EL2_EIS | SCTLR_EL2_EOS)
343	msr	sctlr_el2, x2
344	isb
345
346	/* Configure the Hypervisor */
347	ldr	x2, =(HCR_RW | HCR_APK | HCR_API | HCR_E2H)
348	msr	hcr_el2, x2
349
350	/* Stash value of HCR_EL2 for later */
351	isb
352	mrs	x4, hcr_el2
353
354	/* Load the Virtualization Process ID Register */
355	mrs	x2, midr_el1
356	msr	vpidr_el2, x2
357
358	/* Load the Virtualization Multiprocess ID Register */
359	mrs	x2, mpidr_el1
360	msr	vmpidr_el2, x2
361
362	/* Set the initial sctlr_el1 */
363	ldr	x2, =INIT_SCTLR_EL1
364	msr	sctlr_el1, x2
365
366	/* Check if the E2H flag is set */
367	tst	x4, #HCR_E2H
368	b.eq	.Lno_vhe
369
370	/*
371	 * The kernel will be running in EL2, route exceptions here rather
372	 * than EL1.
373	 */
374	orr	x4, x4, #(HCR_TGE)
375	msr	hcr_el2, x4
376	isb
377
378	msr	SCTLR_EL12_REG, x2
379	mov	x2, xzr /* CPTR_EL2 is managed by vfp.c */
380	ldr	x3, =(CNTHCTL_E2H_EL1PCTEN | CNTHCTL_E2H_EL1PTEN)
381	ldr	x5, =(PSR_DAIF | PSR_M_EL2h)
382	b	.Ldone_vhe
383
384.Lno_vhe:
385	/* Hypervisor trap functions */
386	adrp	x2, hyp_stub_vectors
387	add	x2, x2, :lo12:hyp_stub_vectors
388	msr	vbar_el2, x2
389
390	ldr	x2, =(CPTR_RES1)
391	ldr	x3, =(CNTHCTL_EL1PCTEN | CNTHCTL_EL1PCEN)
392	ldr	x5, =(PSR_DAIF | PSR_M_EL1h)
393
394.Ldone_vhe:
395
396	msr	cptr_el2, x2
397	/* Enable access to the physical timers at EL1 */
398	msr	cnthctl_el2, x3
399	/* Set the return PSTATE */
400	msr	spsr_el2, x5
401
402	/*
403	 * Configure the Extended Hypervisor register. This is only valid if
404	 * FEAT_HCX is enabled.
405	 */
406	CHECK_CPU_FEAT(x2, ID_AA64MMFR1, HCX, 2f)
407	/* Extended Hypervisor Configuration */
408	mov x2, xzr
409	msr HCRX_EL2_REG, x2
410	isb
4112:
412
413	/* Don't trap to EL2 for CP15 traps */
414	msr	hstr_el2, xzr
415
416	/* Set the counter offset to a known value */
417	msr	cntvoff_el2, xzr
418
419	/* Zero vttbr_el2 so a hypervisor can tell the host and guest apart */
420	msr	vttbr_el2, xzr
421
422	/* Check the CPU supports GIC, and configure the CPU interface */
423	CHECK_CPU_FEAT(x2, ID_AA64PFR0, GIC, 3f)
424
425	mrs	x2, icc_sre_el2
426	orr	x2, x2, #ICC_SRE_EL2_EN	/* Enable access from insecure EL1 */
427	orr	x2, x2, #ICC_SRE_EL2_SRE	/* Enable system registers */
428	msr	icc_sre_el2, x2
4293:
430
431	/* Set the address to return to our return address */
432	msr	elr_el2, x30
433	isb
434
435	eret
436#undef INIT_SCTLR_EL1
437LEND(enter_kernel_el)
438
439/*
440 * Get the physical address the kernel was loaded at.
441 */
442LENTRY(get_load_phys_addr)
443	/* Load the offset of get_load_phys_addr from KERNBASE */
444	ldr	x28, =(get_load_phys_addr - KERNBASE)
445	/* Load the physical address of get_load_phys_addr */
446	adr	x29, get_load_phys_addr
447	/* Find the physical address of KERNBASE, i.e. our load address */
448	sub	x28, x29, x28
449	ret
450LEND(get_load_phys_addr)
451
452/*
453 * This builds the page tables containing the identity map, and the kernel
454 * virtual map.
455 *
456 * It relys on:
457 *  We were loaded to an address that is on a 2MiB boundary
458 *  All the memory must not cross a 1GiB boundaty
459 *  x28 contains the physical address we were loaded from
460 *
461 *  There are 7 or 8 pages before that address for the page tables
462 *   The pages used are:
463 *    - The Kernel L3 tables (only for 16k kernel)
464 *    - The Kernel L2 table
465 *    - The Kernel L1 table
466 *    - The Kernel L0 table             (TTBR1)
467 *    - The identity (PA = VA) L2 table
468 *    - The identity (PA = VA) L1 table
469 *    - The identity (PA = VA) L0 table (Early TTBR0)
470 *    - The Kernel empty L0 table       (Late TTBR0)
471 */
472LENTRY(create_pagetables)
473	/* Save the Link register */
474	mov	x5, x30
475
476	/* Clean the page table */
477	adrp	x6, pagetable
478	add	x6, x6, :lo12:pagetable
479	mov	x26, x6
480	adrp	x27, pagetable_end
481	add	x27, x27, :lo12:pagetable_end
4821:
483	stp	xzr, xzr, [x6], #16
484	stp	xzr, xzr, [x6], #16
485	stp	xzr, xzr, [x6], #16
486	stp	xzr, xzr, [x6], #16
487	cmp	x6, x27
488	b.lo	1b
489
490	/*
491	 * Find the shareability attribute we should use. If FEAT_LPA2 is
492	 * enabled then the shareability field is moved from the page table
493	 * to tcr_el1 and the bits in the page table are reused by the
494	 * address field.
495	 */
496#if PAGE_SIZE == PAGE_SIZE_4K
497#define	LPA2_MASK	ID_AA64MMFR0_TGran4_MASK
498#define	LPA2_VAL	ID_AA64MMFR0_TGran4_LPA2
499#elif PAGE_SIZE == PAGE_SIZE_16K
500#define	LPA2_MASK	ID_AA64MMFR0_TGran16_MASK
501#define	LPA2_VAL	ID_AA64MMFR0_TGran16_LPA2
502#else
503#error Unsupported page size
504#endif
505	mrs	x6, id_aa64mmfr0_el1
506	mov	x7, LPA2_VAL
507	and	x6, x6, LPA2_MASK
508	cmp	x6, x7
509	ldr	x22, =(ATTR_SH(ATTR_SH_IS))
510	csel	x22, xzr, x22, eq
511#undef LPA2_MASK
512#undef LPA2_VAL
513
514	/*
515	 * Build the TTBR1 maps.
516	 */
517
518	/* Find the size of the kernel */
519	mov	x6, #(KERNBASE)
520
521#if defined(LINUX_BOOT_ABI)
522	/* X19 is used as 'map FDT data' flag */
523	mov	x19, xzr
524
525	/* No modules or FDT pointer ? */
526	cbz	x0, booti_no_fdt
527
528	/*
529	 * Test if x0 points to modules descriptor(virtual address) or
530	 * to FDT (physical address)
531	 */
532	cmp	x0, x6		/* x6 is #(KERNBASE) */
533	b.lo	booti_fdt
534#endif
535
536	/* Booted with modules pointer */
537	/* Find modulep - begin */
538	sub	x8, x0, x6
539	/*
540	 * Add space for the module data. When PAGE_SIZE is 4k this will
541	 * add at least 2 level 2 blocks (2 * 2MiB). When PAGE_SIZE is
542	 * larger it will be at least as large as we use smaller level 3
543	 * pages.
544	 */
545	ldr	x7, =((6 * 1024 * 1024) - 1)
546	add	x8, x8, x7
547	b	common
548
549#if defined(LINUX_BOOT_ABI)
550booti_fdt:
551	/* Booted by U-Boot booti with FDT data */
552	/* Set 'map FDT data' flag */
553	mov	x19, #1
554
555booti_no_fdt:
556	/* Booted by U-Boot booti without FTD data */
557	/* Find the end - begin */
558	ldr     x7, .Lend
559	sub     x8, x7, x6
560
561	/*
562	 * Add one 2MiB page for copy of FDT data (maximum FDT size),
563	 * one for metadata and round up
564	 */
565	ldr	x7, =(3 * L2_SIZE - 1)
566	add	x8, x8, x7
567#endif
568
569common:
570#if PAGE_SIZE != PAGE_SIZE_4K
571	/*
572	 * Create L3 and L3C pages. The kernel will be loaded at a 2M aligned
573	 * address, enabling the creation of L3C pages. However, when the page
574	 * size is larger than 4k, L2 blocks are too large to map the kernel
575	 * with 2M alignment.
576	 */
577#define	PTE_SHIFT	L3_SHIFT
578#define	BUILD_PTE_FUNC	build_l3_page_pagetable
579#else
580#define	PTE_SHIFT	L2_SHIFT
581#define	BUILD_PTE_FUNC	build_l2_block_pagetable
582#endif
583
584	/* Get the number of blocks/pages to allocate, rounded down */
585	lsr	x10, x8, #(PTE_SHIFT)
586
587	/* Create the kernel space PTE table */
588	mov	x6, x26
589	mov	x7, #(ATTR_S1_IDX(VM_MEMATTR_WRITE_BACK))
590	mov	x8, #(KERNBASE)
591	mov	x9, x28
592	bl	BUILD_PTE_FUNC
593
594#undef PTE_SHIFT
595#undef BUILD_PTE_FUNC
596
597#if PAGE_SIZE != PAGE_SIZE_4K
598	/* Move to the l2 table */
599	ldr	x9, =(PAGE_SIZE * L3_PAGE_COUNT)
600	add	x26, x26, x9
601
602	/* Link the l2 -> l3 table */
603	mov	x9, x6
604	mov	x6, x26
605	bl	link_l2_pagetable
606#endif
607
608	/* Move to the l1 table */
609	add	x26, x26, #PAGE_SIZE
610
611	/* Link the l1 -> l2 table */
612	mov	x9, x6
613	mov	x6, x26
614	bl	link_l1_pagetable
615
616	/* Move to the l0 table */
617	add	x24, x26, #PAGE_SIZE
618
619	/* Link the l0 -> l1 table */
620	mov	x9, x6
621	mov	x6, x24
622	mov	x10, #1
623	bl	link_l0_pagetable
624
625	/*
626	 * Build the TTBR0 maps.  As TTBR0 maps, they must specify ATTR_S1_nG.
627	 * They are only needed early on, so the VA = PA map is uncached.
628	 */
629	add	x27, x24, #PAGE_SIZE
630
631	mov	x6, x27		/* The initial page table */
632
633	/* Create the VA = PA map */
634	mov	x7, #(ATTR_S1_nG | ATTR_S1_IDX(VM_MEMATTR_WRITE_BACK))
635	adrp	x16, _start
636	and	x16, x16, #(~L2_OFFSET)
637	mov	x9, x16		/* PA start */
638	mov	x8, x16		/* VA start (== PA start) */
639	mov	x10, #1
640	bl	build_l2_block_pagetable
641
642#if defined(SOCDEV_PA)
643	/* Create a table for the UART */
644	mov	x7, #(ATTR_S1_nG | ATTR_S1_IDX(VM_MEMATTR_DEVICE))
645	ldr	x9, =(L2_SIZE)
646	add	x16, x16, x9	/* VA start */
647	mov	x8, x16
648
649	/* Store the socdev virtual address */
650	add	x17, x8, #(SOCDEV_PA & L2_OFFSET)
651	adrp	x9, socdev_va
652	str	x17, [x9, :lo12:socdev_va]
653
654	mov	x9, #(SOCDEV_PA & ~L2_OFFSET)	/* PA start */
655	mov	x10, #1
656	bl	build_l2_block_pagetable
657#endif
658
659#if defined(LINUX_BOOT_ABI)
660	/* Map FDT data ? */
661	cbz	x19, 1f
662
663	/* Create the mapping for FDT data (2 MiB max) */
664	mov	x7, #(ATTR_S1_nG | ATTR_S1_IDX(VM_MEMATTR_WRITE_BACK))
665	ldr	x9, =(L2_SIZE)
666	add	x16, x16, x9	/* VA start */
667	mov	x8, x16
668	mov	x9, x0			/* PA start */
669	/* Update the module pointer to point at the allocated memory */
670	and	x0, x0, #(L2_OFFSET)	/* Keep the lower bits */
671	add	x0, x0, x8		/* Add the aligned virtual address */
672
673	mov	x10, #1
674	bl	build_l2_block_pagetable
675
6761:
677#endif
678
679	/* Move to the l1 table */
680	add	x27, x27, #PAGE_SIZE
681
682	/* Link the l1 -> l2 table */
683	mov	x9, x6
684	mov	x6, x27
685	bl	link_l1_pagetable
686
687	/* Move to the l0 table */
688	add	x27, x27, #PAGE_SIZE
689
690	/* Link the l0 -> l1 table */
691	mov	x9, x6
692	mov	x6, x27
693	mov	x10, #1
694	bl	link_l0_pagetable
695
696	/* Restore the Link register */
697	mov	x30, x5
698	ret
699LEND(create_pagetables)
700
701/*
702 * Builds an L0 -> L1 table descriptor
703 *
704 *  x6  = L0 table
705 *  x8  = Virtual Address
706 *  x9  = L1 PA (trashed)
707 *  x10 = Entry count (trashed)
708 *  x11, x12 and x13 are trashed
709 */
710LENTRY(link_l0_pagetable)
711	/*
712	 * Link an L0 -> L1 table entry.
713	 */
714	/* Find the table index */
715	lsr	x11, x8, #L0_SHIFT
716	and	x11, x11, #L0_ADDR_MASK
717
718	/* Build the L0 block entry */
719	mov	x12, #L0_TABLE
720	orr	x12, x12, #(TATTR_UXN_TABLE | TATTR_AP_TABLE_NO_EL0)
721
722	/* Only use the output address bits */
723	lsr	x9, x9, #PAGE_SHIFT
7241:	orr	x13, x12, x9, lsl #PAGE_SHIFT
725
726	/* Store the entry */
727	str	x13, [x6, x11, lsl #3]
728
729	sub	x10, x10, #1
730	add	x11, x11, #1
731	add	x9, x9, #1
732	cbnz	x10, 1b
733
734	ret
735LEND(link_l0_pagetable)
736
737/*
738 * Builds an L1 -> L2 table descriptor
739 *
740 *  x6  = L1 table
741 *  x8  = Virtual Address
742 *  x9  = L2 PA (trashed)
743 *  x11, x12 and x13 are trashed
744 */
745LENTRY(link_l1_pagetable)
746	/*
747	 * Link an L1 -> L2 table entry.
748	 */
749	/* Find the table index */
750	lsr	x11, x8, #L1_SHIFT
751	and	x11, x11, #Ln_ADDR_MASK
752
753	/* Build the L1 block entry */
754	mov	x12, #L1_TABLE
755
756	/* Only use the output address bits */
757	lsr	x9, x9, #PAGE_SHIFT
758	orr	x13, x12, x9, lsl #PAGE_SHIFT
759
760	/* Store the entry */
761	str	x13, [x6, x11, lsl #3]
762
763	ret
764LEND(link_l1_pagetable)
765
766/*
767 * Builds count 2 MiB page table entry
768 *  x6  = L2 table
769 *  x7  = Block attributes
770 *  x8  = VA start
771 *  x9  = PA start (trashed)
772 *  x10 = Entry count (trashed)
773 *  x11, x12 and x13 are trashed
774 */
775LENTRY(build_l2_block_pagetable)
776	/*
777	 * Build the L2 table entry.
778	 */
779	/* Find the table index */
780	lsr	x11, x8, #L2_SHIFT
781	and	x11, x11, #Ln_ADDR_MASK
782
783	/* Build the L2 block entry */
784	orr	x12, x7, #L2_BLOCK
785	orr	x12, x12, #(ATTR_AF)
786	orr	x12, x12, #(ATTR_S1_UXN)
787#ifdef __ARM_FEATURE_BTI_DEFAULT
788	orr	x12, x12, #(ATTR_S1_GP)
789#endif
790	/* Set the shareability attribute */
791	orr	x12, x12, x22
792
793	/* Only use the output address bits */
794	lsr	x9, x9, #L2_SHIFT
795
796	/* Set the physical address for this virtual address */
7971:	orr	x13, x12, x9, lsl #L2_SHIFT
798
799	/* Store the entry */
800	str	x13, [x6, x11, lsl #3]
801
802	sub	x10, x10, #1
803	add	x11, x11, #1
804	add	x9, x9, #1
805	cbnz	x10, 1b
806
807	ret
808LEND(build_l2_block_pagetable)
809
810#if PAGE_SIZE != PAGE_SIZE_4K
811/*
812 * Builds an L2 -> L3 table descriptor
813 *
814 *  x6  = L2 table
815 *  x8  = Virtual Address
816 *  x9  = L3 PA (trashed)
817 *  x11, x12 and x13 are trashed
818 */
819LENTRY(link_l2_pagetable)
820	/*
821	 * Link an L2 -> L3 table entry.
822	 */
823	/* Find the table index */
824	lsr	x11, x8, #L2_SHIFT
825	and	x11, x11, #Ln_ADDR_MASK
826
827	/* Build the L1 block entry */
828	mov	x12, #L2_TABLE
829
830	/* Only use the output address bits */
831	lsr	x9, x9, #PAGE_SHIFT
832	orr	x13, x12, x9, lsl #PAGE_SHIFT
833
834	/* Store the entry */
835	str	x13, [x6, x11, lsl #3]
836
837	ret
838LEND(link_l2_pagetable)
839
840/*
841 * Builds count level 3 page table entries. Uses ATTR_CONTIGUOUS to create
842 * large page (L3C) mappings when the current VA and remaining count allow
843 * it.
844 *  x6  = L3 table
845 *  x7  = Block attributes
846 *  x8  = VA start
847 *  x9  = PA start (trashed)
848 *  x10 = Entry count (trashed)
849 *  x11, x12 and x13 are trashed
850 *
851 * VA start (x8) modulo L3C_SIZE must equal PA start (x9) modulo L3C_SIZE.
852 */
853LENTRY(build_l3_page_pagetable)
854	/*
855	 * Build the L3 table entry.
856	 */
857	/* Find the table index */
858	lsr	x11, x8, #L3_SHIFT
859	and	x11, x11, #Ln_ADDR_MASK
860
861	/* Build the L3 page entry */
862	orr	x12, x7, #L3_PAGE
863	orr	x12, x12, #(ATTR_AF)
864	orr	x12, x12, #(ATTR_S1_UXN)
865#ifdef __ARM_FEATURE_BTI_DEFAULT
866	orr	x12, x12, #(ATTR_S1_GP)
867#endif
868	/* Set the shareability attribute */
869	orr	x12, x12, x22
870
871	/* Only use the output address bits */
872	lsr	x9, x9, #L3_SHIFT
873
874	/* Check if an ATTR_CONTIGUOUS mapping is possible */
8751:	tst	x11, #(L3C_ENTRIES - 1)
876	b.ne	2f
877	cmp	x10, #L3C_ENTRIES
878	b.lo	3f
879	orr	x12, x12, #(ATTR_CONTIGUOUS)
880	b	2f
8813:	and	x12, x12, #(~ATTR_CONTIGUOUS)
882
883	/* Set the physical address for this virtual address */
8842:	orr	x13, x12, x9, lsl #L3_SHIFT
885
886	/* Store the entry */
887	str	x13, [x6, x11, lsl #3]
888
889	sub	x10, x10, #1
890	add	x11, x11, #1
891	add	x9, x9, #1
892	cbnz	x10, 1b
893
894	ret
895LEND(build_l3_page_pagetable)
896#endif
897
898LENTRY(start_mmu)
899	dsb	sy
900
901	/* Load the exception vectors */
902	ldr	x2, =exception_vectors
903	msr	vbar_el1, x2
904
905	/* Load ttbr0 and ttbr1 */
906	msr	ttbr0_el1, x27
907	msr	ttbr1_el1, x24
908	isb
909
910	/* Clear the Monitor Debug System control register */
911	msr	mdscr_el1, xzr
912
913	/* Invalidate the TLB */
914	tlbi	vmalle1is
915	dsb	ish
916	isb
917
918	ldr	x2, mair
919	msr	mair_el1, x2
920
921	/*
922	 * Setup TCR according to the PARange and ASIDBits fields
923	 * from ID_AA64MMFR0_EL1 and the HAFDBS field from the
924	 * ID_AA64MMFR1_EL1.  More precisely, set TCR_EL1.AS
925	 * to 1 only if the ASIDBits field equals 0b0010.
926	 */
927	ldr	x2, tcr
928
929	/* If x22 contains a non-zero value then LPA2 is not implemented */
930	cbnz	x22, .Lno_lpa2
931	ldr	x3, =(TCR_DS)
932	orr	x2, x2, x3
933.Lno_lpa2:
934
935	mrs	x3, id_aa64mmfr0_el1
936
937	/* Copy the bottom 3 bits from id_aa64mmfr0_el1 into TCR.IPS */
938	bfi	x2, x3, #(TCR_IPS_SHIFT), #(TCR_IPS_WIDTH)
939	and	x3, x3, #(ID_AA64MMFR0_ASIDBits_MASK)
940
941	/* Check if the HW supports 16 bit ASIDS */
942	cmp	x3, #(ID_AA64MMFR0_ASIDBits_16)
943	/* If so x3 == 1, else x3 == 0 */
944	cset	x3, eq
945	/* Set TCR.AS with x3 */
946	bfi	x2, x3, #(TCR_ASID_SHIFT), #(TCR_ASID_WIDTH)
947
948	/*
949	 * Check if the HW supports access flag updates, and set
950	 * TCR_EL1.HA accordingly. The TCR_EL1.HD flag to enable
951	 * HW management of dirty state is set in C code as it may
952	 * need to be disabled because of CPU errata.
953	 */
954	CHECK_CPU_FEAT(x3, ID_AA64MMFR1, HAFDBS, 1f)
955	orr	x2, x2, #(TCR_HA)
9561:
957
958	msr	tcr_el1, x2
959
960	/*
961	 * Setup SCTLR.
962	 */
963	ldr	x2, sctlr_set
964	ldr	x3, sctlr_clear
965	mrs	x1, sctlr_el1
966	bic	x1, x1, x3	/* Clear the required bits */
967	orr	x1, x1, x2	/* Set the required bits */
968	msr	sctlr_el1, x1
969	isb
970
971	ret
972
973	.align 3
974mair:
975	.quad	MAIR_ATTR(MAIR_DEVICE_nGnRnE, VM_MEMATTR_DEVICE_nGnRnE) | \
976		MAIR_ATTR(MAIR_NORMAL_NC, VM_MEMATTR_UNCACHEABLE)   |	\
977		MAIR_ATTR(MAIR_NORMAL_WB, VM_MEMATTR_WRITE_BACK)    |	\
978		MAIR_ATTR(MAIR_NORMAL_WT, VM_MEMATTR_WRITE_THROUGH) |	\
979		MAIR_ATTR(MAIR_DEVICE_nGnRE, VM_MEMATTR_DEVICE_nGnRE)
980tcr:
981#if PAGE_SIZE == PAGE_SIZE_4K
982#define	TCR_TG	(TCR_TG1_4K | TCR_TG0_4K)
983#elif PAGE_SIZE == PAGE_SIZE_16K
984#define	TCR_TG	(TCR_TG1_16K | TCR_TG0_16K)
985#else
986#error Unsupported page size
987#endif
988
989	.quad (TCR_TxSZ(64 - VIRT_BITS) | TCR_TG |			\
990	    TCR_SH1_IS | TCR_ORGN1_WBWA | TCR_IRGN1_WBWA |		\
991	    TCR_SH0_IS | TCR_ORGN0_WBWA | TCR_IRGN0_WBWA)
992sctlr_set:
993	/* Bits to set */
994	.quad (SCTLR_LSMAOE | SCTLR_nTLSMD | SCTLR_UCI | SCTLR_SPAN | \
995	    SCTLR_nTWE | SCTLR_nTWI | SCTLR_UCT | SCTLR_DZE | \
996	    SCTLR_I | SCTLR_SED | SCTLR_SA0 | SCTLR_SA | SCTLR_C | \
997	    SCTLR_M | SCTLR_CP15BEN | SCTLR_BT1 | SCTLR_BT0)
998sctlr_clear:
999	/* Bits to clear */
1000	.quad (SCTLR_EE | SCTLR_E0E | SCTLR_IESB | SCTLR_WXN | SCTLR_UMA | \
1001	    SCTLR_ITD | SCTLR_A)
1002LEND(start_mmu)
1003
1004ENTRY(abort)
1005	b abort
1006END(abort)
1007
1008.bss
1009	.align	PAGE_SHIFT
1010initstack:
1011	.space	BOOT_STACK_SIZE
1012initstack_end:
1013
1014	.section .init_pagetable, "aw", %nobits
1015	.align PAGE_SHIFT
1016	/*
1017	 * 6 initial tables (in the following order):
1018	 *           L2 for kernel (High addresses)
1019	 *           L1 for kernel
1020	 *           L0 for kernel
1021	 *           L1 bootstrap for user   (Low addresses)
1022	 *           L0 bootstrap for user
1023	 *           L0 for user
1024	 */
1025	.globl pagetable_l0_ttbr1
1026pagetable:
1027#if PAGE_SIZE != PAGE_SIZE_4K
1028	.space	(PAGE_SIZE * L3_PAGE_COUNT)
1029pagetable_l2_ttbr1:
1030#endif
1031	.space	PAGE_SIZE
1032pagetable_l1_ttbr1:
1033	.space	PAGE_SIZE
1034pagetable_l0_ttbr1:
1035	.space	PAGE_SIZE
1036pagetable_l2_ttbr0_bootstrap:
1037	.space	PAGE_SIZE
1038pagetable_l1_ttbr0_bootstrap:
1039	.space	PAGE_SIZE
1040pagetable_l0_ttbr0_bootstrap:
1041	.space	PAGE_SIZE
1042pagetable_l0_ttbr0:
1043	.space	PAGE_SIZE
1044pagetable_end:
1045
1046el2_pagetable:
1047	.space	PAGE_SIZE
1048
1049	.section .rodata, "a", %progbits
1050	.globl	aarch32_sigcode
1051	.align 2
1052aarch32_sigcode:
1053	.word 0xe1a0000d	// mov r0, sp
1054	.word 0xe2800040	// add r0, r0, #SIGF_UC
1055	.word 0xe59f700c	// ldr r7, [pc, #12]
1056	.word 0xef000000	// swi #0
1057	.word 0xe59f7008	// ldr r7, [pc, #8]
1058	.word 0xef000000	// swi #0
1059	.word 0xeafffffa	// b . - 16
1060	.word SYS_sigreturn
1061	.word SYS_exit
1062	.align	3
1063	.size aarch32_sigcode, . - aarch32_sigcode
1064aarch32_esigcode:
1065	.data
1066	.global sz_aarch32_sigcode
1067sz_aarch32_sigcode:
1068	.quad aarch32_esigcode - aarch32_sigcode
1069
1070GNU_PROPERTY_AARCH64_FEATURE_1_NOTE(GNU_PROPERTY_AARCH64_FEATURE_1_VAL)
1071