xref: /xnu-10002.41.9/osfmk/arm64/arm_vm_init.c (revision 699cd48037512bf4380799317ca44ca453c82f57)
1 /*
2  * Copyright (c) 2007-2011 Apple Inc. All rights reserved.
3  *
4  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
5  *
6  * This file contains Original Code and/or Modifications of Original Code
7  * as defined in and that are subject to the Apple Public Source License
8  * Version 2.0 (the 'License'). You may not use this file except in
9  * compliance with the License. The rights granted to you under the License
10  * may not be used to create, or enable the creation or redistribution of,
11  * unlawful or unlicensed copies of an Apple operating system, or to
12  * circumvent, violate, or enable the circumvention or violation of, any
13  * terms of an Apple operating system software license agreement.
14  *
15  * Please obtain a copy of the License at
16  * http://www.opensource.apple.com/apsl/ and read it before using this file.
17  *
18  * The Original Code and all software distributed under the License are
19  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23  * Please see the License for the specific language governing rights and
24  * limitations under the License.
25  *
26  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
27  */
28 
29 #include <mach_kdp.h>
30 #include <debug.h>
31 
32 #include <kern/assert.h>
33 #include <kern/misc_protos.h>
34 #include <kern/monotonic.h>
35 #include <mach/vm_types.h>
36 #include <mach/vm_param.h>
37 #include <vm/vm_kern.h>
38 #include <vm/vm_page.h>
39 #include <vm/pmap.h>
40 
41 #include <machine/atomic.h>
42 #include <arm64/proc_reg.h>
43 #include <arm64/lowglobals.h>
44 #include <arm/cpu_data_internal.h>
45 #include <arm/misc_protos.h>
46 #include <pexpert/arm64/boot.h>
47 #include <pexpert/device_tree.h>
48 
49 #include <libkern/kernel_mach_header.h>
50 #include <libkern/section_keywords.h>
51 
52 #include <san/kasan.h>
53 
54 #if __ARM_KERNEL_PROTECT__
55 /*
56  * If we want to support __ARM_KERNEL_PROTECT__, we need a sufficient amount of
57  * mappable space preceeding the kernel (as we unmap the kernel by cutting the
58  * range covered by TTBR1 in half).  This must also cover the exception vectors.
59  */
60 static_assert(KERNEL_PMAP_HEAP_RANGE_START > ARM_KERNEL_PROTECT_EXCEPTION_START);
61 
62 /* The exception vectors and the kernel cannot share root TTEs. */
63 static_assert((KERNEL_PMAP_HEAP_RANGE_START & ~ARM_TT_ROOT_OFFMASK) > ARM_KERNEL_PROTECT_EXCEPTION_START);
64 
65 /*
66  * We must have enough space in the TTBR1_EL1 range to create the EL0 mapping of
67  * the exception vectors.
68  */
69 static_assert((((~ARM_KERNEL_PROTECT_EXCEPTION_START) + 1) * 2ULL) <= (ARM_TT_ROOT_SIZE + ARM_TT_ROOT_INDEX_MASK));
70 #endif /* __ARM_KERNEL_PROTECT__ */
71 
72 #define ARM_DYNAMIC_TABLE_XN (ARM_TTE_TABLE_PXN | ARM_TTE_TABLE_XN)
73 
74 #if KASAN
75 extern vm_offset_t shadow_pbase;
76 extern vm_offset_t shadow_ptop;
77 extern vm_offset_t physmap_vbase;
78 extern vm_offset_t physmap_vtop;
79 #endif
80 
81 /*
82  * We explicitly place this in const, as it is not const from a language
83  * perspective, but it is only modified before we actually switch away from
84  * the bootstrap page tables.
85  */
86 SECURITY_READ_ONLY_LATE(uint8_t) bootstrap_pagetables[BOOTSTRAP_TABLE_SIZE] __attribute__((aligned(ARM_PGBYTES)));
87 
88 /*
89  * Denotes the end of xnu.
90  */
91 extern void *last_kernel_symbol;
92 
93 extern void arm64_replace_bootstack(cpu_data_t*);
94 extern void PE_slide_devicetree(vm_offset_t);
95 
96 /*
97  * KASLR parameters
98  */
99 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_kernel_base;
100 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_kernel_top;
101 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_kext_base;
102 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_kext_top;
103 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_kernel_stext;
104 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_kernel_etext;
105 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_kernel_slide;
106 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_kernel_slid_base;
107 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_kernel_slid_top;
108 
109 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_prelink_stext;
110 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_prelink_etext;
111 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_prelink_sdata;
112 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_prelink_edata;
113 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_prelink_sinfo;
114 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_prelink_einfo;
115 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_slinkedit;
116 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_elinkedit;
117 
118 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_kernel_builtinkmod_text;
119 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_kernel_builtinkmod_text_end;
120 
121 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_kernelcache_base;
122 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_kernelcache_top;
123 
124 /* Used by <mach/arm/vm_param.h> */
125 SECURITY_READ_ONLY_LATE(unsigned long) gVirtBase;
126 SECURITY_READ_ONLY_LATE(unsigned long) gPhysBase;
127 SECURITY_READ_ONLY_LATE(unsigned long) gPhysSize;
128 SECURITY_READ_ONLY_LATE(unsigned long) gT0Sz = T0SZ_BOOT;
129 SECURITY_READ_ONLY_LATE(unsigned long) gT1Sz = T1SZ_BOOT;
130 
131 /* 23543331 - step 1 of kext / kernel __TEXT and __DATA colocation is to move
132  * all kexts before the kernel.  This is only for arm64 devices and looks
133  * something like the following:
134  * -- vmaddr order --
135  * 0xffffff8004004000 __PRELINK_TEXT
136  * 0xffffff8007004000 __TEXT (xnu)
137  * 0xffffff80075ec000 __DATA (xnu)
138  * 0xffffff80076dc000 __KLD (xnu)
139  * 0xffffff80076e0000 __LAST (xnu)
140  * 0xffffff80076e4000 __LINKEDIT (xnu)
141  * 0xffffff80076e4000 __PRELINK_DATA (not used yet)
142  * 0xffffff800782c000 __PRELINK_INFO
143  * 0xffffff80078e4000 -- End of kernelcache
144  */
145 
146 /* 24921709 - make XNU ready for KTRR
147  *
148  * Two possible kernel cache layouts, depending on which kcgen is being used.
149  * VAs increasing downwards.
150  * Old KCGEN:
151  *
152  * __PRELINK_TEXT
153  * __TEXT
154  * __DATA_CONST
155  * __TEXT_EXEC
156  * __KLD
157  * __LAST
158  * __DATA
159  * __PRELINK_DATA (expected empty)
160  * __LINKEDIT
161  * __PRELINK_INFO
162  *
163  * New kcgen:
164  *
165  * __PRELINK_TEXT    <--- First KTRR (ReadOnly) segment
166  * __PLK_DATA_CONST
167  * __PLK_TEXT_EXEC
168  * __TEXT
169  * __DATA_CONST
170  * __TEXT_EXEC
171  * __KLD
172  * __LAST            <--- Last KTRR (ReadOnly) segment
173  * __DATA
174  * __BOOTDATA (if present)
175  * __LINKEDIT
176  * __PRELINK_DATA (expected populated now)
177  * __PLK_LINKEDIT
178  * __PRELINK_INFO
179  *
180  */
181 
182 vm_offset_t mem_size;                             /* Size of actual physical memory present
183                                                    * minus any performance buffer and possibly
184                                                    * limited by mem_limit in bytes */
185 uint64_t    mem_actual;                           /* The "One True" physical memory size
186                                                    * actually, it's the highest physical
187                                                    * address + 1 */
188 uint64_t    max_mem;                              /* Size of physical memory (bytes), adjusted
189                                                    * by maxmem */
190 uint64_t    max_mem_actual;                       /* Actual size of physical memory (bytes),
191                                                    * adjusted by the maxmem boot-arg */
192 uint64_t    sane_size;                            /* Memory size to use for defaults
193                                                    * calculations */
194 /* This no longer appears to be used; kill it? */
195 addr64_t    vm_last_addr = VM_MAX_KERNEL_ADDRESS; /* Highest kernel
196                                                    * virtual address known
197                                                    * to the VM system */
198 
199 SECURITY_READ_ONLY_LATE(vm_offset_t)              segEXTRADATA;
200 SECURITY_READ_ONLY_LATE(unsigned long)            segSizeEXTRADATA;
201 
202 /* Trust cache portion of EXTRADATA (if within it) */
203 SECURITY_READ_ONLY_LATE(vm_offset_t)              segTRUSTCACHE;
204 SECURITY_READ_ONLY_LATE(unsigned long)            segSizeTRUSTCACHE;
205 
206 SECURITY_READ_ONLY_LATE(vm_offset_t)          segLOWESTTEXT;
207 SECURITY_READ_ONLY_LATE(vm_offset_t)          segLOWEST;
208 SECURITY_READ_ONLY_LATE(vm_offset_t)          segLOWESTRO;
209 SECURITY_READ_ONLY_LATE(vm_offset_t)          segHIGHESTRO;
210 
211 /* Only set when booted from MH_FILESET kernel collections */
212 SECURITY_READ_ONLY_LATE(vm_offset_t)          segLOWESTKC;
213 SECURITY_READ_ONLY_LATE(vm_offset_t)          segHIGHESTKC;
214 SECURITY_READ_ONLY_LATE(vm_offset_t)          segLOWESTROKC;
215 SECURITY_READ_ONLY_LATE(vm_offset_t)          segHIGHESTROKC;
216 SECURITY_READ_ONLY_LATE(vm_offset_t)          segLOWESTAuxKC;
217 SECURITY_READ_ONLY_LATE(vm_offset_t)          segHIGHESTAuxKC;
218 SECURITY_READ_ONLY_LATE(vm_offset_t)          segLOWESTROAuxKC;
219 SECURITY_READ_ONLY_LATE(vm_offset_t)          segHIGHESTROAuxKC;
220 SECURITY_READ_ONLY_LATE(vm_offset_t)          segLOWESTRXAuxKC;
221 SECURITY_READ_ONLY_LATE(vm_offset_t)          segHIGHESTRXAuxKC;
222 SECURITY_READ_ONLY_LATE(vm_offset_t)          segHIGHESTNLEAuxKC;
223 
224 SECURITY_READ_ONLY_LATE(static vm_offset_t)   segTEXTB;
225 SECURITY_READ_ONLY_LATE(static unsigned long) segSizeTEXT;
226 
227 #if XNU_MONITOR
228 SECURITY_READ_ONLY_LATE(vm_offset_t)          segPPLTEXTB;
229 SECURITY_READ_ONLY_LATE(unsigned long)        segSizePPLTEXT;
230 
231 SECURITY_READ_ONLY_LATE(vm_offset_t)          segPPLTRAMPB;
232 SECURITY_READ_ONLY_LATE(unsigned long)        segSizePPLTRAMP;
233 
234 SECURITY_READ_ONLY_LATE(vm_offset_t)          segPPLDATACONSTB;
235 SECURITY_READ_ONLY_LATE(unsigned long)        segSizePPLDATACONST;
236 SECURITY_READ_ONLY_LATE(void *)               pmap_stacks_start = NULL;
237 SECURITY_READ_ONLY_LATE(void *)               pmap_stacks_end = NULL;
238 #if HAS_GUARDED_IO_FILTER
239 SECURITY_READ_ONLY_LATE(void *)               iofilter_stacks_start = NULL;
240 SECURITY_READ_ONLY_LATE(void *)               iofilter_stacks_end = NULL;
241 #endif
242 #endif
243 
244 SECURITY_READ_ONLY_LATE(static vm_offset_t)   segDATACONSTB;
245 SECURITY_READ_ONLY_LATE(static unsigned long) segSizeDATACONST;
246 
247 SECURITY_READ_ONLY_LATE(vm_offset_t)   segTEXTEXECB;
248 SECURITY_READ_ONLY_LATE(unsigned long) segSizeTEXTEXEC;
249 
250 SECURITY_READ_ONLY_LATE(static vm_offset_t)   segDATAB;
251 SECURITY_READ_ONLY_LATE(static unsigned long) segSizeDATA;
252 
253 #if XNU_MONITOR
254 SECURITY_READ_ONLY_LATE(vm_offset_t)          segPPLDATAB;
255 SECURITY_READ_ONLY_LATE(unsigned long)        segSizePPLDATA;
256 #endif
257 
258 SECURITY_READ_ONLY_LATE(vm_offset_t)          segBOOTDATAB;
259 SECURITY_READ_ONLY_LATE(unsigned long)        segSizeBOOTDATA;
260 extern vm_offset_t                            intstack_low_guard;
261 extern vm_offset_t                            intstack_high_guard;
262 extern vm_offset_t                            excepstack_high_guard;
263 
264 SECURITY_READ_ONLY_LATE(vm_offset_t)          segLINKB;
265 SECURITY_READ_ONLY_LATE(static unsigned long) segSizeLINK;
266 
267 SECURITY_READ_ONLY_LATE(static vm_offset_t)   segKLDB;
268 SECURITY_READ_ONLY_LATE(unsigned long)        segSizeKLD;
269 SECURITY_READ_ONLY_LATE(static vm_offset_t)   segKLDDATAB;
270 SECURITY_READ_ONLY_LATE(static unsigned long) segSizeKLDDATA;
271 SECURITY_READ_ONLY_LATE(vm_offset_t)          segLASTB;
272 SECURITY_READ_ONLY_LATE(unsigned long)        segSizeLAST;
273 SECURITY_READ_ONLY_LATE(vm_offset_t)          segLASTDATACONSTB;
274 SECURITY_READ_ONLY_LATE(unsigned long)        segSizeLASTDATACONST;
275 
276 SECURITY_READ_ONLY_LATE(vm_offset_t)          sectHIBTEXTB;
277 SECURITY_READ_ONLY_LATE(unsigned long)        sectSizeHIBTEXT;
278 SECURITY_READ_ONLY_LATE(vm_offset_t)          segHIBDATAB;
279 SECURITY_READ_ONLY_LATE(unsigned long)        segSizeHIBDATA;
280 SECURITY_READ_ONLY_LATE(vm_offset_t)          sectHIBDATACONSTB;
281 SECURITY_READ_ONLY_LATE(unsigned long)        sectSizeHIBDATACONST;
282 
283 SECURITY_READ_ONLY_LATE(vm_offset_t)          segPRELINKTEXTB;
284 SECURITY_READ_ONLY_LATE(unsigned long)        segSizePRELINKTEXT;
285 
286 SECURITY_READ_ONLY_LATE(static vm_offset_t)   segPLKTEXTEXECB;
287 SECURITY_READ_ONLY_LATE(static unsigned long) segSizePLKTEXTEXEC;
288 
289 SECURITY_READ_ONLY_LATE(static vm_offset_t)   segPLKDATACONSTB;
290 SECURITY_READ_ONLY_LATE(static unsigned long) segSizePLKDATACONST;
291 
292 SECURITY_READ_ONLY_LATE(static vm_offset_t)   segPRELINKDATAB;
293 SECURITY_READ_ONLY_LATE(static unsigned long) segSizePRELINKDATA;
294 
295 SECURITY_READ_ONLY_LATE(static vm_offset_t)   segPLKLLVMCOVB = 0;
296 SECURITY_READ_ONLY_LATE(static unsigned long) segSizePLKLLVMCOV = 0;
297 
298 SECURITY_READ_ONLY_LATE(static vm_offset_t)   segPLKLINKEDITB;
299 SECURITY_READ_ONLY_LATE(static unsigned long) segSizePLKLINKEDIT;
300 
301 SECURITY_READ_ONLY_LATE(static vm_offset_t)   segPRELINKINFOB;
302 SECURITY_READ_ONLY_LATE(static unsigned long) segSizePRELINKINFO;
303 
304 /* Only set when booted from MH_FILESET primary kernel collection */
305 SECURITY_READ_ONLY_LATE(vm_offset_t)          segKCTEXTEXECB;
306 SECURITY_READ_ONLY_LATE(unsigned long)        segSizeKCTEXTEXEC;
307 SECURITY_READ_ONLY_LATE(static vm_offset_t)   segKCDATACONSTB;
308 SECURITY_READ_ONLY_LATE(static unsigned long) segSizeKCDATACONST;
309 SECURITY_READ_ONLY_LATE(static vm_offset_t)   segKCDATAB;
310 SECURITY_READ_ONLY_LATE(static unsigned long) segSizeKCDATA;
311 
312 SECURITY_READ_ONLY_LATE(static boolean_t) use_contiguous_hint = TRUE;
313 
314 SECURITY_READ_ONLY_LATE(int) PAGE_SHIFT_CONST;
315 
316 SECURITY_READ_ONLY_LATE(vm_offset_t) end_kern;
317 SECURITY_READ_ONLY_LATE(vm_offset_t) etext;
318 SECURITY_READ_ONLY_LATE(vm_offset_t) sdata;
319 SECURITY_READ_ONLY_LATE(vm_offset_t) edata;
320 
321 SECURITY_READ_ONLY_LATE(static vm_offset_t) auxkc_mh, auxkc_base, auxkc_right_above;
322 
323 vm_offset_t alloc_ptpage(boolean_t map_static);
324 SECURITY_READ_ONLY_LATE(vm_offset_t) ropage_next;
325 extern int dtrace_keep_kernel_symbols(void);
326 
327 /*
328  * Bootstrap the system enough to run with virtual memory.
329  * Map the kernel's code and data, and allocate the system page table.
330  * Page_size must already be set.
331  *
332  * Parameters:
333  * first_avail: first available physical page -
334  *              after kernel page tables
335  * avail_start: PA of first physical page
336  * avail_end:   PA of last physical page
337  */
338 SECURITY_READ_ONLY_LATE(vm_offset_t)     first_avail;
339 SECURITY_READ_ONLY_LATE(vm_offset_t)     static_memory_end;
340 SECURITY_READ_ONLY_LATE(pmap_paddr_t)    avail_start;
341 SECURITY_READ_ONLY_LATE(pmap_paddr_t)    avail_end;
342 SECURITY_READ_ONLY_LATE(pmap_paddr_t)    real_avail_end;
343 SECURITY_READ_ONLY_LATE(unsigned long)   real_phys_size;
344 SECURITY_READ_ONLY_LATE(vm_map_address_t) physmap_base = (vm_map_address_t)0;
345 SECURITY_READ_ONLY_LATE(vm_map_address_t) physmap_end = (vm_map_address_t)0;
346 
347 /**
348  * First physical address freely available to xnu.
349  */
350 SECURITY_READ_ONLY_LATE(addr64_t) first_avail_phys = 0;
351 
352 /*
353  * Bounds of the kernelcache; used for accounting.
354  */
355 SECURITY_READ_ONLY_LATE(vm_offset_t) arm_vm_kernelcache_phys_start;
356 SECURITY_READ_ONLY_LATE(vm_offset_t) arm_vm_kernelcache_phys_end;
357 
358 #if __ARM_KERNEL_PROTECT__
359 extern void ExceptionVectorsBase;
360 extern void ExceptionVectorsEnd;
361 #endif /* __ARM_KERNEL_PROTECT__ */
362 
363 typedef struct {
364 	pmap_paddr_t pa;
365 	vm_map_address_t va;
366 	vm_size_t len;
367 } ptov_table_entry;
368 
369 #define PTOV_TABLE_SIZE 8
370 SECURITY_READ_ONLY_LATE(static ptov_table_entry)        ptov_table[PTOV_TABLE_SIZE];
371 SECURITY_READ_ONLY_LATE(static boolean_t)               kva_active = FALSE;
372 
373 
374 vm_map_address_t
phystokv(pmap_paddr_t pa)375 phystokv(pmap_paddr_t pa)
376 {
377 
378 	for (size_t i = 0; (i < PTOV_TABLE_SIZE) && (ptov_table[i].len != 0); i++) {
379 		if ((pa >= ptov_table[i].pa) && (pa < (ptov_table[i].pa + ptov_table[i].len))) {
380 			return pa - ptov_table[i].pa + ptov_table[i].va;
381 		}
382 	}
383 	if (__improbable((pa < gPhysBase) || ((pa - gPhysBase) >= real_phys_size))) {
384 		panic("%s: illegal PA: 0x%llx; phys base 0x%llx, size 0x%llx", __func__,
385 		    (unsigned long long)pa, (unsigned long long)gPhysBase, (unsigned long long)real_phys_size);
386 	}
387 	return pa - gPhysBase + gVirtBase;
388 }
389 
390 vm_map_address_t
phystokv_range(pmap_paddr_t pa,vm_size_t * max_len)391 phystokv_range(pmap_paddr_t pa, vm_size_t *max_len)
392 {
393 
394 	vm_size_t len;
395 	for (size_t i = 0; (i < PTOV_TABLE_SIZE) && (ptov_table[i].len != 0); i++) {
396 		if ((pa >= ptov_table[i].pa) && (pa < (ptov_table[i].pa + ptov_table[i].len))) {
397 			len = ptov_table[i].len - (pa - ptov_table[i].pa);
398 			if (*max_len > len) {
399 				*max_len = len;
400 			}
401 			return pa - ptov_table[i].pa + ptov_table[i].va;
402 		}
403 	}
404 	len = PAGE_SIZE - (pa & PAGE_MASK);
405 	if (*max_len > len) {
406 		*max_len = len;
407 	}
408 	if (__improbable((pa < gPhysBase) || ((pa - gPhysBase) >= real_phys_size))) {
409 		panic("%s: illegal PA: 0x%llx; phys base 0x%llx, size 0x%llx", __func__,
410 		    (unsigned long long)pa, (unsigned long long)gPhysBase, (unsigned long long)real_phys_size);
411 	}
412 	return pa - gPhysBase + gVirtBase;
413 }
414 
415 vm_offset_t
ml_static_vtop(vm_offset_t va)416 ml_static_vtop(vm_offset_t va)
417 {
418 	for (size_t i = 0; (i < PTOV_TABLE_SIZE) && (ptov_table[i].len != 0); i++) {
419 		if ((va >= ptov_table[i].va) && (va < (ptov_table[i].va + ptov_table[i].len))) {
420 			return va - ptov_table[i].va + ptov_table[i].pa;
421 		}
422 	}
423 	if (__improbable((va < gVirtBase) || (((vm_address_t)(va) - gVirtBase) >= gPhysSize))) {
424 		panic("%s: illegal VA: %p; virt base 0x%llx, size 0x%llx", __func__,
425 		    (void*)va, (unsigned long long)gVirtBase, (unsigned long long)gPhysSize);
426 	}
427 	return (vm_address_t)(va) - gVirtBase + gPhysBase;
428 }
429 
430 /*
431  * This rounds the given address up to the nearest boundary for a PTE contiguous
432  * hint.
433  */
434 static vm_offset_t
round_up_pte_hint_address(vm_offset_t address)435 round_up_pte_hint_address(vm_offset_t address)
436 {
437 	vm_offset_t hint_size = ARM_PTE_SIZE << ARM_PTE_HINT_ENTRIES_SHIFT;
438 	return (address + (hint_size - 1)) & ~(hint_size - 1);
439 }
440 
441 /* allocate a page for a page table: we support static and dynamic mappings.
442  *
443  * returns a virtual address for the allocated page
444  *
445  * for static mappings, we allocate from the region ropagetable_begin to ro_pagetable_end-1,
446  * which is defined in the DATA_CONST segment and will be protected RNX when vm_prot_finalize runs.
447  *
448  * for dynamic mappings, we allocate from avail_start, which should remain RWNX.
449  */
450 
451 vm_offset_t
alloc_ptpage(boolean_t map_static)452 alloc_ptpage(boolean_t map_static)
453 {
454 	vm_offset_t vaddr;
455 
456 #if !(defined(KERNEL_INTEGRITY_KTRR) || defined(KERNEL_INTEGRITY_CTRR))
457 	map_static = FALSE;
458 #endif
459 
460 	if (!ropage_next) {
461 		ropage_next = (vm_offset_t)&ropagetable_begin;
462 	}
463 
464 	if (map_static) {
465 		assert(ropage_next < (vm_offset_t)&ropagetable_end);
466 
467 		vaddr = ropage_next;
468 		ropage_next += ARM_PGBYTES;
469 
470 		return vaddr;
471 	} else {
472 		vaddr = phystokv(avail_start);
473 		avail_start += ARM_PGBYTES;
474 
475 		return vaddr;
476 	}
477 }
478 
479 #if DEBUG
480 
481 void dump_kva_l2(vm_offset_t tt_base, tt_entry_t *tt, int indent, uint64_t *rosz_out, uint64_t *rwsz_out);
482 
483 void
dump_kva_l2(vm_offset_t tt_base,tt_entry_t * tt,int indent,uint64_t * rosz_out,uint64_t * rwsz_out)484 dump_kva_l2(vm_offset_t tt_base, tt_entry_t *tt, int indent, uint64_t *rosz_out, uint64_t *rwsz_out)
485 {
486 	unsigned int i;
487 	boolean_t cur_ro, prev_ro = 0;
488 	int start_entry = -1;
489 	tt_entry_t cur, prev = 0;
490 	pmap_paddr_t robegin = kvtophys((vm_offset_t)&ropagetable_begin);
491 	pmap_paddr_t roend = kvtophys((vm_offset_t)&ropagetable_end);
492 	boolean_t tt_static = kvtophys((vm_offset_t)tt) >= robegin &&
493 	    kvtophys((vm_offset_t)tt) < roend;
494 
495 	for (i = 0; i < TTE_PGENTRIES; i++) {
496 		int tte_type = tt[i] & ARM_TTE_TYPE_MASK;
497 		cur = tt[i] & ARM_TTE_TABLE_MASK;
498 
499 		if (tt_static) {
500 			/* addresses mapped by this entry are static if it is a block mapping,
501 			 * or the table was allocated from the RO page table region */
502 			cur_ro = (tte_type == ARM_TTE_TYPE_BLOCK) || (cur >= robegin && cur < roend);
503 		} else {
504 			cur_ro = 0;
505 		}
506 
507 		if ((cur == 0 && prev != 0) || (cur_ro != prev_ro && prev != 0)) { // falling edge
508 			uintptr_t start, end, sz;
509 
510 			start = (uintptr_t)start_entry << ARM_TT_L2_SHIFT;
511 			start += tt_base;
512 			end = ((uintptr_t)i << ARM_TT_L2_SHIFT) - 1;
513 			end += tt_base;
514 
515 			sz = end - start + 1;
516 			printf("%*s0x%08x_%08x-0x%08x_%08x %s (%luMB)\n",
517 			    indent * 4, "",
518 			    (uint32_t)(start >> 32), (uint32_t)start,
519 			    (uint32_t)(end >> 32), (uint32_t)end,
520 			    prev_ro ? "Static " : "Dynamic",
521 			    (sz >> 20));
522 
523 			if (prev_ro) {
524 				*rosz_out += sz;
525 			} else {
526 				*rwsz_out += sz;
527 			}
528 		}
529 
530 		if ((prev == 0 && cur != 0) || cur_ro != prev_ro) { // rising edge: set start
531 			start_entry = i;
532 		}
533 
534 		prev = cur;
535 		prev_ro = cur_ro;
536 	}
537 }
538 
539 void
dump_kva_space()540 dump_kva_space()
541 {
542 	uint64_t tot_rosz = 0, tot_rwsz = 0;
543 	int ro_ptpages, rw_ptpages;
544 	pmap_paddr_t robegin = kvtophys((vm_offset_t)&ropagetable_begin);
545 	pmap_paddr_t roend = kvtophys((vm_offset_t)&ropagetable_end);
546 	boolean_t root_static = kvtophys((vm_offset_t)cpu_tte) >= robegin &&
547 	    kvtophys((vm_offset_t)cpu_tte) < roend;
548 	uint64_t kva_base = ~((1ULL << (64 - T1SZ_BOOT)) - 1);
549 
550 	printf("Root page table: %s\n", root_static ? "Static" : "Dynamic");
551 
552 	for (unsigned int i = 0; i < TTE_PGENTRIES; i++) {
553 		pmap_paddr_t cur;
554 		boolean_t cur_ro;
555 		uintptr_t start, end;
556 		uint64_t rosz = 0, rwsz = 0;
557 
558 		if ((cpu_tte[i] & ARM_TTE_VALID) == 0) {
559 			continue;
560 		}
561 
562 		cur = cpu_tte[i] & ARM_TTE_TABLE_MASK;
563 		start = (uint64_t)i << ARM_TT_L1_SHIFT;
564 		start = start + kva_base;
565 		end = start + (ARM_TT_L1_SIZE - 1);
566 		cur_ro = cur >= robegin && cur < roend;
567 
568 		printf("0x%08x_%08x-0x%08x_%08x %s\n",
569 		    (uint32_t)(start >> 32), (uint32_t)start,
570 		    (uint32_t)(end >> 32), (uint32_t)end,
571 		    cur_ro ? "Static " : "Dynamic");
572 
573 		dump_kva_l2(start, (tt_entry_t*)phystokv(cur), 1, &rosz, &rwsz);
574 		tot_rosz += rosz;
575 		tot_rwsz += rwsz;
576 	}
577 
578 	printf("L2 Address space mapped: Static %lluMB Dynamic %lluMB Total %lluMB\n",
579 	    tot_rosz >> 20,
580 	    tot_rwsz >> 20,
581 	    (tot_rosz >> 20) + (tot_rwsz >> 20));
582 
583 	ro_ptpages = (int)((ropage_next - (vm_offset_t)&ropagetable_begin) >> ARM_PGSHIFT);
584 	rw_ptpages = (int)(lowGlo.lgStaticSize  >> ARM_PGSHIFT);
585 	printf("Pages used: static %d dynamic %d\n", ro_ptpages, rw_ptpages);
586 }
587 
588 #endif /* DEBUG */
589 
590 #if __ARM_KERNEL_PROTECT__ || XNU_MONITOR
591 /*
592  * arm_vm_map:
593  *   root_ttp: The kernel virtual address for the root of the target page tables
594  *   vaddr: The target virtual address
595  *   pte: A page table entry value (may be ARM_PTE_EMPTY)
596  *
597  * This function installs pte at vaddr in root_ttp.  Any page table pages needed
598  * to install pte will be allocated by this function.
599  */
600 static void
arm_vm_map(tt_entry_t * root_ttp,vm_offset_t vaddr,pt_entry_t pte)601 arm_vm_map(tt_entry_t * root_ttp, vm_offset_t vaddr, pt_entry_t pte)
602 {
603 	vm_offset_t ptpage = 0;
604 	tt_entry_t * ttp = root_ttp;
605 
606 	tt_entry_t * l1_ttep = NULL;
607 	tt_entry_t l1_tte = 0;
608 
609 	tt_entry_t * l2_ttep = NULL;
610 	tt_entry_t l2_tte = 0;
611 	pt_entry_t * ptep = NULL;
612 	pt_entry_t cpte = 0;
613 
614 	/*
615 	 * Walk the target page table to find the PTE for the given virtual
616 	 * address.  Allocate any page table pages needed to do this.
617 	 */
618 	l1_ttep = ttp + ((vaddr & ARM_TT_L1_INDEX_MASK) >> ARM_TT_L1_SHIFT);
619 	l1_tte = *l1_ttep;
620 
621 	if (l1_tte == ARM_TTE_EMPTY) {
622 		ptpage = alloc_ptpage(TRUE);
623 		bzero((void *)ptpage, ARM_PGBYTES);
624 		l1_tte = kvtophys(ptpage);
625 		l1_tte &= ARM_TTE_TABLE_MASK;
626 		l1_tte |= ARM_TTE_VALID | ARM_TTE_TYPE_TABLE | ARM_TTE_TABLE_AP(ARM_TTE_TABLE_AP_USER_NA);
627 		*l1_ttep = l1_tte;
628 		ptpage = 0;
629 	}
630 
631 	ttp = (tt_entry_t *)phystokv(l1_tte & ARM_TTE_TABLE_MASK);
632 
633 	l2_ttep = ttp + ((vaddr & ARM_TT_L2_INDEX_MASK) >> ARM_TT_L2_SHIFT);
634 	l2_tte = *l2_ttep;
635 
636 	if (l2_tte == ARM_TTE_EMPTY) {
637 		ptpage = alloc_ptpage(TRUE);
638 		bzero((void *)ptpage, ARM_PGBYTES);
639 		l2_tte = kvtophys(ptpage);
640 		l2_tte &= ARM_TTE_TABLE_MASK;
641 		l2_tte |= ARM_TTE_VALID | ARM_TTE_TYPE_TABLE;
642 		*l2_ttep = l2_tte;
643 		ptpage = 0;
644 	}
645 
646 	ttp = (tt_entry_t *)phystokv(l2_tte & ARM_TTE_TABLE_MASK);
647 
648 	ptep = ttp + ((vaddr & ARM_TT_L3_INDEX_MASK) >> ARM_TT_L3_SHIFT);
649 	cpte = *ptep;
650 
651 	/*
652 	 * If the existing PTE is not empty, then we are replacing a valid
653 	 * mapping.
654 	 */
655 	if (cpte != ARM_PTE_EMPTY) {
656 		panic("%s: cpte=%#llx is not empty, "
657 		    "vaddr=%#lx, pte=%#llx",
658 		    __FUNCTION__, cpte,
659 		    vaddr, pte);
660 	}
661 
662 	*ptep = pte;
663 }
664 
665 #endif // __ARM_KERNEL_PROTECT || XNU_MONITOR
666 
667 #if __ARM_KERNEL_PROTECT__
668 
669 /*
670  * arm_vm_kernel_el0_map:
671  *   vaddr: The target virtual address
672  *   pte: A page table entry value (may be ARM_PTE_EMPTY)
673  *
674  * This function installs pte at vaddr for the EL0 kernel mappings.
675  */
676 static void
arm_vm_kernel_el0_map(vm_offset_t vaddr,pt_entry_t pte)677 arm_vm_kernel_el0_map(vm_offset_t vaddr, pt_entry_t pte)
678 {
679 	/* Calculate where vaddr will be in the EL1 kernel page tables. */
680 	vm_offset_t kernel_pmap_vaddr = vaddr - ((ARM_TT_ROOT_INDEX_MASK + ARM_TT_ROOT_SIZE) / 2ULL);
681 	arm_vm_map(cpu_tte, kernel_pmap_vaddr, pte);
682 }
683 
684 /*
685  * arm_vm_kernel_el1_map:
686  *   vaddr: The target virtual address
687  *   pte: A page table entry value (may be ARM_PTE_EMPTY)
688  *
689  * This function installs pte at vaddr for the EL1 kernel mappings.
690  */
691 static void
arm_vm_kernel_el1_map(vm_offset_t vaddr,pt_entry_t pte)692 arm_vm_kernel_el1_map(vm_offset_t vaddr, pt_entry_t pte)
693 {
694 	arm_vm_map(cpu_tte, vaddr, pte);
695 }
696 
697 /*
698  * arm_vm_kernel_pte:
699  *   vaddr: The target virtual address
700  *
701  * This function returns the PTE value for the given vaddr from the kernel page
702  * tables.  If the region has been been block mapped, we return what an
703  * equivalent PTE value would be (as regards permissions and flags).  We also
704  * remove the HINT bit (as we are not necessarily creating contiguous mappings.
705  */
706 static pt_entry_t
arm_vm_kernel_pte(vm_offset_t vaddr)707 arm_vm_kernel_pte(vm_offset_t vaddr)
708 {
709 	tt_entry_t * ttp = cpu_tte;
710 	tt_entry_t * ttep = NULL;
711 	tt_entry_t tte = 0;
712 	pt_entry_t * ptep = NULL;
713 	pt_entry_t pte = 0;
714 
715 	ttep = ttp + ((vaddr & ARM_TT_L1_INDEX_MASK) >> ARM_TT_L1_SHIFT);
716 	tte = *ttep;
717 
718 	assert(tte & ARM_TTE_VALID);
719 
720 	if ((tte & ARM_TTE_TYPE_MASK) == ARM_TTE_TYPE_BLOCK) {
721 		/* This is a block mapping; return the equivalent PTE value. */
722 		pte = (pt_entry_t)(tte & ~ARM_TTE_TYPE_MASK);
723 		pte |= ARM_PTE_TYPE_VALID;
724 		pte |= vaddr & ((ARM_TT_L1_SIZE - 1) & ARM_PTE_PAGE_MASK);
725 		pte &= ~ARM_PTE_HINT_MASK;
726 		return pte;
727 	}
728 
729 	ttp = (tt_entry_t *)phystokv(tte & ARM_TTE_TABLE_MASK);
730 	ttep = ttp + ((vaddr & ARM_TT_L2_INDEX_MASK) >> ARM_TT_L2_SHIFT);
731 	tte = *ttep;
732 
733 	assert(tte & ARM_TTE_VALID);
734 
735 	if ((tte & ARM_TTE_TYPE_MASK) == ARM_TTE_TYPE_BLOCK) {
736 		/* This is a block mapping; return the equivalent PTE value. */
737 		pte = (pt_entry_t)(tte & ~ARM_TTE_TYPE_MASK);
738 		pte |= ARM_PTE_TYPE_VALID;
739 		pte |= vaddr & ((ARM_TT_L2_SIZE - 1) & ARM_PTE_PAGE_MASK);
740 		pte &= ~ARM_PTE_HINT_MASK;
741 		return pte;
742 	}
743 
744 	ttp = (tt_entry_t *)phystokv(tte & ARM_TTE_TABLE_MASK);
745 
746 	ptep = ttp + ((vaddr & ARM_TT_L3_INDEX_MASK) >> ARM_TT_L3_SHIFT);
747 	pte = *ptep;
748 	pte &= ~ARM_PTE_HINT_MASK;
749 	return pte;
750 }
751 
752 /*
753  * arm_vm_prepare_kernel_el0_mappings:
754  *   alloc_only: Indicates if PTE values should be copied from the EL1 kernel
755  *     mappings.
756  *
757  * This function expands the kernel page tables to support the EL0 kernel
758  * mappings, and conditionally installs the PTE values for the EL0 kernel
759  * mappings (if alloc_only is false).
760  */
761 static void
arm_vm_prepare_kernel_el0_mappings(bool alloc_only)762 arm_vm_prepare_kernel_el0_mappings(bool alloc_only)
763 {
764 	pt_entry_t pte = 0;
765 	vm_offset_t start = ((vm_offset_t)&ExceptionVectorsBase) & ~PAGE_MASK;
766 	vm_offset_t end = (((vm_offset_t)&ExceptionVectorsEnd) + PAGE_MASK) & ~PAGE_MASK;
767 	vm_offset_t cur = 0;
768 	vm_offset_t cur_fixed = 0;
769 
770 	/* Expand for/map the exceptions vectors in the EL0 kernel mappings. */
771 	for (cur = start, cur_fixed = ARM_KERNEL_PROTECT_EXCEPTION_START; cur < end; cur += ARM_PGBYTES, cur_fixed += ARM_PGBYTES) {
772 		/*
773 		 * We map the exception vectors at a different address than that
774 		 * of the kernelcache to avoid sharing page table pages with the
775 		 * kernelcache (as this may cause issues with TLB caching of
776 		 * page table pages.
777 		 */
778 		if (!alloc_only) {
779 			pte = arm_vm_kernel_pte(cur);
780 		}
781 
782 		arm_vm_kernel_el1_map(cur_fixed, pte);
783 		arm_vm_kernel_el0_map(cur_fixed, pte);
784 	}
785 
786 	__builtin_arm_dmb(DMB_ISH);
787 	__builtin_arm_isb(ISB_SY);
788 
789 	if (!alloc_only) {
790 		/*
791 		 * If we have created the alternate exception vector mappings,
792 		 * the boot CPU may now switch over to them.
793 		 */
794 		set_vbar_el1(ARM_KERNEL_PROTECT_EXCEPTION_START);
795 		__builtin_arm_isb(ISB_SY);
796 	}
797 }
798 
799 /*
800  * arm_vm_populate_kernel_el0_mappings:
801  *
802  * This function adds all required mappings to the EL0 kernel mappings.
803  */
804 static void
arm_vm_populate_kernel_el0_mappings(void)805 arm_vm_populate_kernel_el0_mappings(void)
806 {
807 	arm_vm_prepare_kernel_el0_mappings(FALSE);
808 }
809 
810 /*
811  * arm_vm_expand_kernel_el0_mappings:
812  *
813  * This function expands the kernel page tables to accomodate the EL0 kernel
814  * mappings.
815  */
816 static void
arm_vm_expand_kernel_el0_mappings(void)817 arm_vm_expand_kernel_el0_mappings(void)
818 {
819 	arm_vm_prepare_kernel_el0_mappings(TRUE);
820 }
821 #endif /* __ARM_KERNEL_PROTECT__ */
822 
823 #if defined(KERNEL_INTEGRITY_KTRR) || defined(KERNEL_INTEGRITY_CTRR)
824 extern void bootstrap_instructions;
825 
826 /*
827  * arm_replace_identity_map takes the V=P map that we construct in start.s
828  * and repurposes it in order to have it map only the page we need in order
829  * to turn on the MMU.  This prevents us from running into issues where
830  * KTRR will cause us to fault on executable block mappings that cross the
831  * KTRR boundary.
832  */
833 static void
arm_replace_identity_map(void)834 arm_replace_identity_map(void)
835 {
836 	vm_offset_t addr;
837 	pmap_paddr_t paddr;
838 
839 	pmap_paddr_t l1_ptp_phys = 0;
840 	tt_entry_t *l1_ptp_virt = NULL;
841 	tt_entry_t *tte1 = NULL;
842 	pmap_paddr_t l2_ptp_phys = 0;
843 	tt_entry_t *l2_ptp_virt = NULL;
844 	tt_entry_t *tte2 = NULL;
845 	pmap_paddr_t l3_ptp_phys = 0;
846 	pt_entry_t *l3_ptp_virt = NULL;
847 	pt_entry_t *ptep = NULL;
848 
849 	addr = ((vm_offset_t)&bootstrap_instructions) & ~ARM_PGMASK;
850 	paddr = kvtophys(addr);
851 
852 	/*
853 	 * Grab references to the V=P page tables, and allocate an L3 page.
854 	 */
855 	l1_ptp_phys = kvtophys((vm_offset_t)&bootstrap_pagetables);
856 	l1_ptp_virt = (tt_entry_t *)phystokv(l1_ptp_phys);
857 	tte1 = &l1_ptp_virt[L1_TABLE_INDEX(paddr)];
858 
859 	l2_ptp_virt = L2_TABLE_VA(tte1);
860 	l2_ptp_phys = (*tte1) & ARM_TTE_TABLE_MASK;
861 	tte2 = &l2_ptp_virt[L2_TABLE_INDEX(paddr)];
862 
863 	l3_ptp_virt = (pt_entry_t *)alloc_ptpage(TRUE);
864 	l3_ptp_phys = kvtophys((vm_offset_t)l3_ptp_virt);
865 	ptep = &l3_ptp_virt[L3_TABLE_INDEX(paddr)];
866 
867 	/*
868 	 * Replace the large V=P mapping with a mapping that provides only the
869 	 * mappings needed to turn on the MMU.
870 	 */
871 
872 	bzero(l1_ptp_virt, ARM_PGBYTES);
873 	*tte1 = ARM_TTE_BOOT_TABLE | (l2_ptp_phys & ARM_TTE_TABLE_MASK);
874 
875 	bzero(l2_ptp_virt, ARM_PGBYTES);
876 	*tte2 = ARM_TTE_BOOT_TABLE | (l3_ptp_phys & ARM_TTE_TABLE_MASK);
877 
878 	*ptep = (paddr & ARM_PTE_MASK) |
879 	    ARM_PTE_TYPE_VALID |
880 	    ARM_PTE_SH(SH_OUTER_MEMORY) |
881 	    ARM_PTE_ATTRINDX(CACHE_ATTRINDX_WRITEBACK) |
882 	    ARM_PTE_AF |
883 	    ARM_PTE_AP(AP_RONA) |
884 	    ARM_PTE_NX;
885 }
886 #endif /* defined(KERNEL_INTEGRITY_KTRR) || defined(KERNEL_INTEGRITY_CTRR) */
887 
888 tt_entry_t *arm_kva_to_tte(vm_offset_t);
889 
890 tt_entry_t *
arm_kva_to_tte(vm_offset_t va)891 arm_kva_to_tte(vm_offset_t va)
892 {
893 	tt_entry_t *tte1, *tte2;
894 	tte1 = cpu_tte + L1_TABLE_INDEX(va);
895 	tte2 = L2_TABLE_VA(tte1) + L2_TABLE_INDEX(va);
896 
897 	return tte2;
898 }
899 
900 #if XNU_MONITOR
901 
902 static inline pt_entry_t *
arm_kva_to_pte(vm_offset_t va)903 arm_kva_to_pte(vm_offset_t va)
904 {
905 	tt_entry_t *tte2 = arm_kva_to_tte(va);
906 	return L3_TABLE_VA(tte2) + L3_TABLE_INDEX(va);
907 }
908 
909 #endif
910 
911 #define ARM64_GRANULE_ALLOW_BLOCK (1 << 0)
912 #define ARM64_GRANULE_ALLOW_HINT (1 << 1)
913 
914 /**
915  * Updates a translation table entry (TTE) with the supplied value, unless doing so might render
916  * the pagetable region read-only before subsequent updates have finished.  In that case, the TTE
917  * value will be saved off for deferred processing.
918  *
919  * @param ttep address of the TTE to update
920  * @param entry the value to store in ttep
921  * @param pa the base physical address mapped by the TTE
922  * @param ttebase L3-page- or L2-block-aligned base virtual address of the pagetable region
923  * @param granule mask indicating whether L2 block or L3 hint mappings are allowed for this segment
924  * @param deferred_ttep_pair 2-element array of addresses of deferred TTEs
925  * @param deferred_tte_pair 2-element array containing TTE values for deferred assignment to
926  *        corresponding elements of deferred_ttep_pair
927  */
928 static void
update_or_defer_tte(tt_entry_t * ttep,tt_entry_t entry,pmap_paddr_t pa,vm_map_address_t ttebase,unsigned granule __unused,tt_entry_t ** deferred_ttep_pair,tt_entry_t * deferred_tte_pair)929 update_or_defer_tte(tt_entry_t *ttep, tt_entry_t entry, pmap_paddr_t pa, vm_map_address_t ttebase,
930     unsigned granule __unused, tt_entry_t **deferred_ttep_pair, tt_entry_t *deferred_tte_pair)
931 {
932 	/*
933 	 * If we're trying to assign an entry that maps the current TTE region (identified by ttebase),
934 	 * and the pagetable is already live (indicated by kva_active), defer assignment of the current
935 	 * entry and possibly the entry after it until all other mappings in the segment have been
936 	 * updated.  Otherwise we may end up immediately marking the pagetable region read-only
937 	 * leading to a fault later on a later assignment if we manage to outrun the TLB.  This can
938 	 * happen on KTRR/CTRR-enabled devices when marking segDATACONST read-only, as the pagetables
939 	 * that map that segment must come from the segment itself.  We therefore store the initial
940 	 * recursive TTE in deferred_ttep_pair[0] and its value in deferred_tte_pair[0].  We may also
941 	 * defer assignment of the TTE following that recursive TTE and store its value in
942 	 * deferred_tte_pair[1], because the TTE region following the current one may also contain
943 	 * pagetables and we must avoid marking that region read-only before updating those tables.
944 	 *
945 	 * We require that such recursive mappings must exist in regions that can be mapped with L2
946 	 * block entries if they are sufficiently large.  This is what allows us to assume that no
947 	 * more than 2 deferred TTEs will be required, because:
948 	 * 	--If more than 2 adjacent L3 PTEs were required to map our pagetables, that would mean
949 	 * 	  we would have at least one full L3 pagetable page and would instead use an L2 block.
950 	 *	--If more than 2 adjacent L2 blocks were required to map our pagetables, that would
951 	 * 	  mean we would have at least one full L2-block-sized region of TTEs and something
952 	 *	  is very wrong because no segment should be that large.
953 	 */
954 	if ((deferred_ttep_pair != NULL) && (deferred_ttep_pair[0] != NULL) && (ttep == (deferred_ttep_pair[0] + 1))) {
955 		assert(deferred_tte_pair[1] == 0);
956 		deferred_ttep_pair[1] = ttep;
957 		deferred_tte_pair[1] = entry;
958 	} else if (kva_active && (phystokv(pa) == ttebase)) {
959 		assert(deferred_ttep_pair != NULL);
960 		assert(granule & ARM64_GRANULE_ALLOW_BLOCK);
961 		if (deferred_ttep_pair[0] == NULL) {
962 			deferred_ttep_pair[0] = ttep;
963 			deferred_tte_pair[0] = entry;
964 		} else {
965 			assert(deferred_ttep_pair[1] == NULL);
966 			deferred_ttep_pair[1] = ttep;
967 			deferred_tte_pair[1] = entry;
968 		}
969 	} else {
970 		*ttep = entry;
971 	}
972 }
973 
974 
975 /*
976  * arm_vm_page_granular_helper updates protections at the L3 level.  It will (if
977  * neccessary) allocate a page for the L3 table and update the corresponding L2
978  * entry.  Then, it will iterate over the L3 table, updating protections as necessary.
979  * This expects to be invoked on a L2 entry or sub L2 entry granularity, so this should
980  * not be invoked from a context that does not do L2 iteration separately (basically,
981  * don't call this except from arm_vm_page_granular_prot).
982  *
983  * unsigned granule: 0 => force to page granule, or a combination of
984  * ARM64_GRANULE_* flags declared above.
985  */
986 
987 static void
arm_vm_page_granular_helper(vm_offset_t start,vm_offset_t _end,vm_offset_t va,pmap_paddr_t pa_offset,int pte_prot_APX,int pte_prot_XN,unsigned granule,tt_entry_t ** deferred_ttep_pair,tt_entry_t * deferred_tte_pair)988 arm_vm_page_granular_helper(vm_offset_t start, vm_offset_t _end, vm_offset_t va, pmap_paddr_t pa_offset,
989     int pte_prot_APX, int pte_prot_XN, unsigned granule,
990     tt_entry_t **deferred_ttep_pair, tt_entry_t *deferred_tte_pair)
991 {
992 	if (va & ARM_TT_L2_OFFMASK) { /* ragged edge hanging over a ARM_TT_L2_SIZE  boundary */
993 		tt_entry_t *tte2;
994 		tt_entry_t tmplate;
995 		pmap_paddr_t pa;
996 		pt_entry_t *ppte, ptmp;
997 		addr64_t ppte_phys;
998 		unsigned i;
999 
1000 		va &= ~ARM_TT_L2_OFFMASK;
1001 		pa = va - gVirtBase + gPhysBase - pa_offset;
1002 
1003 		if (pa >= real_avail_end) {
1004 			return;
1005 		}
1006 
1007 		tte2 = arm_kva_to_tte(va);
1008 
1009 		assert(_end >= va);
1010 		tmplate = *tte2;
1011 
1012 		if (ARM_TTE_TYPE_TABLE == (tmplate & ARM_TTE_TYPE_MASK)) {
1013 			/* pick up the existing page table. */
1014 			ppte = (pt_entry_t *)phystokv((tmplate & ARM_TTE_TABLE_MASK));
1015 		} else {
1016 			// TTE must be reincarnated with page level mappings.
1017 
1018 			// ... but we don't want to break up blocks on live
1019 			// translation tables.
1020 			assert(!kva_active);
1021 
1022 			ppte = (pt_entry_t*)alloc_ptpage(pa_offset == 0);
1023 			bzero(ppte, ARM_PGBYTES);
1024 			ppte_phys = kvtophys((vm_offset_t)ppte);
1025 
1026 			*tte2 = pa_to_tte(ppte_phys) | ARM_TTE_TYPE_TABLE | ARM_TTE_VALID;
1027 		}
1028 
1029 		vm_offset_t len = _end - va;
1030 		if ((pa + len) > real_avail_end) {
1031 			_end -= (pa + len - real_avail_end);
1032 		}
1033 		assert((start - gVirtBase + gPhysBase - pa_offset) >= gPhysBase);
1034 
1035 		/* Round up to the nearest PAGE_SIZE boundary when creating mappings:
1036 		 * PAGE_SIZE may be a multiple of ARM_PGBYTES, and we don't want to leave
1037 		 * a ragged non-PAGE_SIZE-aligned edge. */
1038 		vm_offset_t rounded_end = round_page(_end);
1039 		/* Apply the desired protections to the specified page range */
1040 		for (i = 0; i <= (ARM_TT_L3_INDEX_MASK >> ARM_TT_L3_SHIFT); i++) {
1041 			if ((start <= va) && (va < rounded_end)) {
1042 				ptmp = pa | ARM_PTE_AF | ARM_PTE_SH(SH_OUTER_MEMORY) | ARM_PTE_TYPE;
1043 				ptmp = ptmp | ARM_PTE_ATTRINDX(CACHE_ATTRINDX_DEFAULT);
1044 				ptmp = ptmp | ARM_PTE_AP(pte_prot_APX);
1045 				ptmp = ptmp | ARM_PTE_NX;
1046 #if __ARM_KERNEL_PROTECT__
1047 				ptmp = ptmp | ARM_PTE_NG;
1048 #endif /* __ARM_KERNEL_PROTECT__ */
1049 
1050 				if (pte_prot_XN) {
1051 					ptmp = ptmp | ARM_PTE_PNX;
1052 				}
1053 
1054 				/*
1055 				 * If we can, apply the contiguous hint to this range.  The hint is
1056 				 * applicable if the current address falls within a hint-sized range that will
1057 				 * be fully covered by this mapping request.
1058 				 */
1059 				if ((va >= round_up_pte_hint_address(start)) && (round_up_pte_hint_address(va + 1) <= _end) &&
1060 				    (granule & ARM64_GRANULE_ALLOW_HINT) && use_contiguous_hint) {
1061 					assert((va & ((1 << ARM_PTE_HINT_ADDR_SHIFT) - 1)) == ((pa & ((1 << ARM_PTE_HINT_ADDR_SHIFT) - 1))));
1062 					ptmp |= ARM_PTE_HINT;
1063 					/* Do not attempt to reapply the hint bit to an already-active mapping.
1064 					 * This very likely means we're attempting to change attributes on an already-active mapping,
1065 					 * which violates the requirement of the hint bit.*/
1066 					assert(!kva_active || (ppte[i] == ARM_PTE_TYPE_FAULT));
1067 				}
1068 				/*
1069 				 * Do not change the contiguous bit on an active mapping.  Even in a single-threaded
1070 				 * environment, it's possible for prefetch to produce a TLB conflict by trying to pull in
1071 				 * a hint-sized entry on top of one or more existing page-sized entries.  It's also useful
1072 				 * to make sure we're not trying to unhint a sub-range of a larger hinted range, which
1073 				 * could produce a later TLB conflict.
1074 				 */
1075 				assert(!kva_active || (ppte[i] == ARM_PTE_TYPE_FAULT) || ((ppte[i] & ARM_PTE_HINT) == (ptmp & ARM_PTE_HINT)));
1076 
1077 				update_or_defer_tte(&ppte[i], ptmp, pa, (vm_map_address_t)ppte, granule, deferred_ttep_pair, deferred_tte_pair);
1078 			}
1079 
1080 			va += ARM_PGBYTES;
1081 			pa += ARM_PGBYTES;
1082 		}
1083 	}
1084 }
1085 
1086 /*
1087  * arm_vm_page_granular_prot updates protections by iterating over the L2 entries and
1088  * changing them.  If a particular chunk necessitates L3 entries (for reasons of
1089  * alignment or length, or an explicit request that the entry be fully expanded), we
1090  * hand off to arm_vm_page_granular_helper to deal with the L3 chunk of the logic.
1091  */
1092 static void
arm_vm_page_granular_prot(vm_offset_t start,unsigned long size,pmap_paddr_t pa_offset,int tte_prot_XN,int pte_prot_APX,int pte_prot_XN,unsigned granule)1093 arm_vm_page_granular_prot(vm_offset_t start, unsigned long size, pmap_paddr_t pa_offset,
1094     int tte_prot_XN, int pte_prot_APX, int pte_prot_XN,
1095     unsigned granule)
1096 {
1097 	tt_entry_t *deferred_ttep_pair[2] = {NULL};
1098 	tt_entry_t deferred_tte_pair[2] = {0};
1099 	vm_offset_t _end = start + size;
1100 	vm_offset_t align_start = (start + ARM_TT_L2_OFFMASK) & ~ARM_TT_L2_OFFMASK;
1101 
1102 	if (size == 0x0UL) {
1103 		return;
1104 	}
1105 
1106 	if (align_start > _end) {
1107 		align_start = _end;
1108 	}
1109 
1110 	arm_vm_page_granular_helper(start, align_start, start, pa_offset, pte_prot_APX, pte_prot_XN, granule, deferred_ttep_pair, deferred_tte_pair);
1111 
1112 	while ((_end - align_start) >= ARM_TT_L2_SIZE) {
1113 		if (!(granule & ARM64_GRANULE_ALLOW_BLOCK)) {
1114 			arm_vm_page_granular_helper(align_start, align_start + ARM_TT_L2_SIZE, align_start + 1, pa_offset,
1115 			    pte_prot_APX, pte_prot_XN, granule, deferred_ttep_pair, deferred_tte_pair);
1116 		} else {
1117 			pmap_paddr_t pa = align_start - gVirtBase + gPhysBase - pa_offset;
1118 			assert((pa & ARM_TT_L2_OFFMASK) == 0);
1119 			tt_entry_t *tte2;
1120 			tt_entry_t tmplate;
1121 
1122 			tte2 = arm_kva_to_tte(align_start);
1123 
1124 			if ((pa >= gPhysBase) && (pa < real_avail_end)) {
1125 				tmplate = (pa & ARM_TTE_BLOCK_L2_MASK) | ARM_TTE_TYPE_BLOCK
1126 				    | ARM_TTE_VALID | ARM_TTE_BLOCK_AF | ARM_TTE_BLOCK_NX
1127 				    | ARM_TTE_BLOCK_AP(pte_prot_APX) | ARM_TTE_BLOCK_SH(SH_OUTER_MEMORY)
1128 				    | ARM_TTE_BLOCK_ATTRINDX(CACHE_ATTRINDX_WRITEBACK);
1129 
1130 #if __ARM_KERNEL_PROTECT__
1131 				tmplate = tmplate | ARM_TTE_BLOCK_NG;
1132 #endif /* __ARM_KERNEL_PROTECT__ */
1133 				if (tte_prot_XN) {
1134 					tmplate = tmplate | ARM_TTE_BLOCK_PNX;
1135 				}
1136 
1137 				update_or_defer_tte(tte2, tmplate, pa, (vm_map_address_t)tte2 & ~ARM_TT_L2_OFFMASK,
1138 				    granule, deferred_ttep_pair, deferred_tte_pair);
1139 			}
1140 		}
1141 		align_start += ARM_TT_L2_SIZE;
1142 	}
1143 
1144 	if (align_start < _end) {
1145 		arm_vm_page_granular_helper(align_start, _end, _end, pa_offset, pte_prot_APX, pte_prot_XN, granule, deferred_ttep_pair, deferred_tte_pair);
1146 	}
1147 
1148 	if (deferred_ttep_pair[0] != NULL) {
1149 #if DEBUG || DEVELOPMENT
1150 		/*
1151 		 * Flush the TLB to catch bugs that might cause us to prematurely revoke write access from the pagetable page.
1152 		 * These bugs may otherwise be hidden by TLB entries in most cases, resulting in very rare panics.
1153 		 * Note that we always flush the TLB at the end of arm_vm_prot_finalize().
1154 		 */
1155 		flush_mmu_tlb();
1156 #endif
1157 		/*
1158 		 * The first TTE in the pair is a recursive mapping of the pagetable region, so we must update it last
1159 		 * to avoid potentially marking deferred_pte_pair[1] read-only.
1160 		 */
1161 		if (deferred_tte_pair[1] != 0) {
1162 			os_atomic_store(deferred_ttep_pair[1], deferred_tte_pair[1], release);
1163 		}
1164 		os_atomic_store(deferred_ttep_pair[0], deferred_tte_pair[0], release);
1165 	}
1166 }
1167 
1168 static inline void
arm_vm_page_granular_RNX(vm_offset_t start,unsigned long size,unsigned granule)1169 arm_vm_page_granular_RNX(vm_offset_t start, unsigned long size, unsigned granule)
1170 {
1171 	arm_vm_page_granular_prot(start, size, 0, 1, AP_RONA, 1, granule);
1172 }
1173 
1174 static inline void
arm_vm_page_granular_ROX(vm_offset_t start,unsigned long size,unsigned granule)1175 arm_vm_page_granular_ROX(vm_offset_t start, unsigned long size, unsigned granule)
1176 {
1177 	arm_vm_page_granular_prot(start, size, 0, 0, AP_RONA, 0, granule);
1178 }
1179 
1180 static inline void
arm_vm_page_granular_RWNX(vm_offset_t start,unsigned long size,unsigned granule)1181 arm_vm_page_granular_RWNX(vm_offset_t start, unsigned long size, unsigned granule)
1182 {
1183 	arm_vm_page_granular_prot(start, size, 0, 1, AP_RWNA, 1, granule);
1184 }
1185 
1186 // Populate seg...AuxKC and fixup AuxKC permissions
1187 static bool
arm_vm_auxkc_init(void)1188 arm_vm_auxkc_init(void)
1189 {
1190 	if (auxkc_mh == 0 || auxkc_base == 0) {
1191 		return false; // no auxKC.
1192 	}
1193 
1194 	/* Fixup AuxKC and populate seg*AuxKC globals used below */
1195 	arm_auxkc_init((void*)auxkc_mh, (void*)auxkc_base);
1196 
1197 	if (segLOWESTAuxKC != segLOWEST) {
1198 		panic("segLOWESTAuxKC (%p) not equal to segLOWEST (%p). auxkc_mh: %p, auxkc_base: %p",
1199 		    (void*)segLOWESTAuxKC, (void*)segLOWEST,
1200 		    (void*)auxkc_mh, (void*)auxkc_base);
1201 	}
1202 
1203 	/*
1204 	 * The AuxKC LINKEDIT segment needs to be covered by the RO region but is excluded
1205 	 * from the RO address range returned by kernel_collection_adjust_mh_addrs().
1206 	 * Ensure the highest non-LINKEDIT address in the AuxKC is the current end of
1207 	 * its RO region before extending it.
1208 	 */
1209 	assert(segHIGHESTROAuxKC == segHIGHESTNLEAuxKC);
1210 	assert(segHIGHESTAuxKC >= segHIGHESTROAuxKC);
1211 	if (segHIGHESTAuxKC > segHIGHESTROAuxKC) {
1212 		segHIGHESTROAuxKC = segHIGHESTAuxKC;
1213 	}
1214 
1215 	/*
1216 	 * The AuxKC RO region must be right below the device tree/trustcache so that it can be covered
1217 	 * by CTRR, and the AuxKC RX region must be within the RO region.
1218 	 */
1219 	assert(segHIGHESTROAuxKC == auxkc_right_above);
1220 	assert(segHIGHESTRXAuxKC <= segHIGHESTROAuxKC);
1221 	assert(segLOWESTRXAuxKC <= segHIGHESTRXAuxKC);
1222 	assert(segLOWESTROAuxKC <= segLOWESTRXAuxKC);
1223 	assert(segLOWESTAuxKC <= segLOWESTROAuxKC);
1224 
1225 	if (segHIGHESTRXAuxKC < segLOWEST) {
1226 		arm_vm_page_granular_RNX(segHIGHESTRXAuxKC, segLOWEST - segHIGHESTRXAuxKC, 0);
1227 	}
1228 	if (segLOWESTRXAuxKC < segHIGHESTRXAuxKC) {
1229 		arm_vm_page_granular_ROX(segLOWESTRXAuxKC, segHIGHESTRXAuxKC - segLOWESTRXAuxKC, 0); // Refined in OSKext::readPrelinkedExtensions
1230 	}
1231 	if (segLOWESTROAuxKC < segLOWESTRXAuxKC) {
1232 		arm_vm_page_granular_RNX(segLOWESTROAuxKC, segLOWESTRXAuxKC - segLOWESTROAuxKC, 0);
1233 	}
1234 	if (segLOWESTAuxKC < segLOWESTROAuxKC) {
1235 		arm_vm_page_granular_RWNX(segLOWESTAuxKC, segLOWESTROAuxKC - segLOWESTAuxKC, 0);
1236 	}
1237 
1238 	return true;
1239 }
1240 
1241 void
arm_vm_prot_init(__unused boot_args * args)1242 arm_vm_prot_init(__unused boot_args * args)
1243 {
1244 	segLOWESTTEXT = UINT64_MAX;
1245 	if (segSizePRELINKTEXT && (segPRELINKTEXTB < segLOWESTTEXT)) {
1246 		segLOWESTTEXT = segPRELINKTEXTB;
1247 	}
1248 	assert(segSizeTEXT);
1249 	if (segTEXTB < segLOWESTTEXT) {
1250 		segLOWESTTEXT = segTEXTB;
1251 	}
1252 	assert(segLOWESTTEXT < UINT64_MAX);
1253 
1254 	segEXTRADATA = 0;
1255 	segSizeEXTRADATA = 0;
1256 	segTRUSTCACHE = 0;
1257 	segSizeTRUSTCACHE = 0;
1258 
1259 	segLOWEST = segLOWESTTEXT;
1260 	segLOWESTRO = segLOWESTTEXT;
1261 
1262 	if (segLOWESTKC && segLOWESTKC < segLOWEST) {
1263 		/*
1264 		 * kernel collections have segments below the kernel. In particular the collection mach header
1265 		 * is below PRELINK_TEXT and is not covered by any other segments already tracked.
1266 		 */
1267 		arm_vm_page_granular_RNX(segLOWESTKC, segLOWEST - segLOWESTKC, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT);
1268 		segLOWEST = segLOWESTKC;
1269 		if (segLOWESTROKC && segLOWESTROKC < segLOWESTRO) {
1270 			segLOWESTRO = segLOWESTROKC;
1271 		}
1272 		if (segHIGHESTROKC && segHIGHESTROKC > segHIGHESTRO) {
1273 			segHIGHESTRO = segHIGHESTROKC;
1274 		}
1275 	}
1276 
1277 	DTEntry memory_map;
1278 	int err;
1279 
1280 	// Device Tree portion of EXTRADATA
1281 	if (SecureDTIsLockedDown()) {
1282 		segEXTRADATA = (vm_offset_t)PE_state.deviceTreeHead;
1283 		segSizeEXTRADATA = PE_state.deviceTreeSize;
1284 	}
1285 
1286 	// Trust Caches portion of EXTRADATA
1287 	{
1288 		DTMemoryMapRange const *trustCacheRange;
1289 		unsigned int trustCacheRangeSize;
1290 
1291 		err = SecureDTLookupEntry(NULL, "chosen/memory-map", &memory_map);
1292 		assert(err == kSuccess);
1293 
1294 		err = SecureDTGetProperty(memory_map, "TrustCache", (void const **)&trustCacheRange, &trustCacheRangeSize);
1295 		if (err == kSuccess) {
1296 			if (trustCacheRangeSize != sizeof(DTMemoryMapRange)) {
1297 				panic("Unexpected /chosen/memory-map/TrustCache property size %u != %zu", trustCacheRangeSize, sizeof(DTMemoryMapRange));
1298 			}
1299 
1300 			vm_offset_t const trustCacheRegion = phystokv(trustCacheRange->paddr);
1301 			if (trustCacheRegion < segLOWEST) {
1302 				if (segEXTRADATA != 0) {
1303 					if (trustCacheRegion != segEXTRADATA + segSizeEXTRADATA) {
1304 						panic("Unexpected location of TrustCache region: %#lx != %#lx",
1305 						    trustCacheRegion, segEXTRADATA + segSizeEXTRADATA);
1306 					}
1307 					segSizeEXTRADATA += trustCacheRange->length;
1308 				} else {
1309 					// Not all devices support CTRR device trees.
1310 					segEXTRADATA = trustCacheRegion;
1311 					segSizeEXTRADATA = trustCacheRange->length;
1312 				}
1313 			}
1314 #if !(DEVELOPMENT || DEBUG)
1315 			else {
1316 				panic("TrustCache region is in an unexpected place: %#lx > %#lx", trustCacheRegion, segLOWEST);
1317 			}
1318 #endif
1319 			segTRUSTCACHE = trustCacheRegion;
1320 			segSizeTRUSTCACHE = trustCacheRange->length;
1321 		}
1322 	}
1323 
1324 	if (segSizeEXTRADATA != 0) {
1325 		if (segEXTRADATA <= segLOWEST) {
1326 			segLOWEST = segEXTRADATA;
1327 			if (segEXTRADATA <= segLOWESTRO) {
1328 				segLOWESTRO = segEXTRADATA;
1329 			}
1330 		} else {
1331 			panic("EXTRADATA is in an unexpected place: %#lx > %#lx", segEXTRADATA, segLOWEST);
1332 		}
1333 
1334 		arm_vm_page_granular_RNX(segEXTRADATA, segSizeEXTRADATA, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT);
1335 	}
1336 
1337 	const DTMemoryMapRange *auxKC_range, *auxKC_header_range;
1338 	unsigned int auxKC_range_size, auxKC_header_range_size;
1339 
1340 	err = SecureDTGetProperty(memory_map, "AuxKC", (const void**)&auxKC_range,
1341 	    &auxKC_range_size);
1342 	if (err != kSuccess) {
1343 		goto noAuxKC;
1344 	}
1345 	assert(auxKC_range_size == sizeof(DTMemoryMapRange));
1346 	err = SecureDTGetProperty(memory_map, "AuxKC-mach_header",
1347 	    (const void**)&auxKC_header_range, &auxKC_header_range_size);
1348 	if (err != kSuccess) {
1349 		goto noAuxKC;
1350 	}
1351 	assert(auxKC_header_range_size == sizeof(DTMemoryMapRange));
1352 
1353 	if (auxKC_header_range->paddr == 0 || auxKC_range->paddr == 0) {
1354 		goto noAuxKC;
1355 	}
1356 
1357 	auxkc_mh = phystokv(auxKC_header_range->paddr);
1358 	auxkc_base = phystokv(auxKC_range->paddr);
1359 
1360 	if (auxkc_base < segLOWEST) {
1361 		auxkc_right_above = segLOWEST;
1362 		segLOWEST = auxkc_base;
1363 	} else {
1364 		panic("auxkc_base (%p) not below segLOWEST (%p)", (void*)auxkc_base, (void*)segLOWEST);
1365 	}
1366 
1367 	/* Map AuxKC RWNX initially so that arm_vm_auxkc_init can traverse
1368 	 * it and apply fixups (after we're off the bootstrap translation
1369 	 * tables).
1370 	 */
1371 	arm_vm_page_granular_RWNX(auxkc_base, auxKC_range->length, 0);
1372 
1373 noAuxKC:
1374 	/* Map coalesced kext TEXT segment RWNX for now */
1375 	arm_vm_page_granular_RWNX(segPRELINKTEXTB, segSizePRELINKTEXT, ARM64_GRANULE_ALLOW_BLOCK); // Refined in OSKext::readPrelinkedExtensions
1376 
1377 	/* Map coalesced kext DATA_CONST segment RWNX (could be empty) */
1378 	arm_vm_page_granular_RWNX(segPLKDATACONSTB, segSizePLKDATACONST, ARM64_GRANULE_ALLOW_BLOCK); // Refined in OSKext::readPrelinkedExtensions
1379 
1380 	/* Map coalesced kext TEXT_EXEC segment RX (could be empty) */
1381 	arm_vm_page_granular_ROX(segPLKTEXTEXECB, segSizePLKTEXTEXEC, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT); // Refined in OSKext::readPrelinkedExtensions
1382 
1383 	/* if new segments not present, set space between PRELINK_TEXT and xnu TEXT to RWNX
1384 	 * otherwise we no longer expect any space between the coalesced kext read only segments and xnu rosegments
1385 	 */
1386 	if (!segSizePLKDATACONST && !segSizePLKTEXTEXEC) {
1387 		if (segSizePRELINKTEXT) {
1388 			arm_vm_page_granular_RWNX(segPRELINKTEXTB + segSizePRELINKTEXT, segTEXTB - (segPRELINKTEXTB + segSizePRELINKTEXT),
1389 			    ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT);
1390 		}
1391 	} else {
1392 		/*
1393 		 * If we have the new segments, we should still protect the gap between kext
1394 		 * read-only pages and kernel read-only pages, in the event that this gap
1395 		 * exists.
1396 		 */
1397 		if ((segPLKDATACONSTB + segSizePLKDATACONST) < segTEXTB) {
1398 			arm_vm_page_granular_RWNX(segPLKDATACONSTB + segSizePLKDATACONST, segTEXTB - (segPLKDATACONSTB + segSizePLKDATACONST),
1399 			    ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT);
1400 		}
1401 	}
1402 
1403 	/*
1404 	 * Protection on kernel text is loose here to allow shenanigans early on.  These
1405 	 * protections are tightened in arm_vm_prot_finalize().  This is necessary because
1406 	 * we currently patch LowResetVectorBase in cpu.c.
1407 	 *
1408 	 * TEXT segment contains mach headers and other non-executable data. This will become RONX later.
1409 	 */
1410 	arm_vm_page_granular_RNX(segTEXTB, segSizeTEXT, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT);
1411 
1412 	/* Can DATACONST start out and stay RNX?
1413 	 * NO, stuff in this segment gets modified during startup (viz. mac_policy_init()/mac_policy_list)
1414 	 * Make RNX in prot_finalize
1415 	 */
1416 	arm_vm_page_granular_RWNX(segDATACONSTB, segSizeDATACONST, ARM64_GRANULE_ALLOW_BLOCK);
1417 
1418 	arm_vm_page_granular_ROX(segTEXTEXECB, segSizeTEXTEXEC, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT);
1419 
1420 #if XNU_MONITOR
1421 	arm_vm_page_granular_ROX(segPPLTEXTB, segSizePPLTEXT, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT);
1422 	arm_vm_page_granular_ROX(segPPLTRAMPB, segSizePPLTRAMP, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT);
1423 	arm_vm_page_granular_RNX(segPPLDATACONSTB, segSizePPLDATACONST, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT);
1424 #endif
1425 
1426 	/* DATA segment will remain RWNX */
1427 	arm_vm_page_granular_RWNX(segDATAB, segSizeDATA, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT);
1428 #if XNU_MONITOR
1429 	arm_vm_page_granular_RWNX(segPPLDATAB, segSizePPLDATA, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT);
1430 #endif
1431 
1432 	arm_vm_page_granular_RWNX(segHIBDATAB, segSizeHIBDATA, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT);
1433 
1434 	arm_vm_page_granular_RWNX(segBOOTDATAB, segSizeBOOTDATA, 0);
1435 	arm_vm_page_granular_RNX((vm_offset_t)&intstack_low_guard, PAGE_MAX_SIZE, 0);
1436 	arm_vm_page_granular_RNX((vm_offset_t)&intstack_high_guard, PAGE_MAX_SIZE, 0);
1437 	arm_vm_page_granular_RNX((vm_offset_t)&excepstack_high_guard, PAGE_MAX_SIZE, 0);
1438 
1439 	arm_vm_page_granular_ROX(segKLDB, segSizeKLD, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT);
1440 	arm_vm_page_granular_RNX(segKLDDATAB, segSizeKLDDATA, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT);
1441 	arm_vm_page_granular_RWNX(segLINKB, segSizeLINK, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT);
1442 	arm_vm_page_granular_RWNX(segPLKLINKEDITB, segSizePLKLINKEDIT, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT); // Coalesced kext LINKEDIT segment
1443 	arm_vm_page_granular_ROX(segLASTB, segSizeLAST, ARM64_GRANULE_ALLOW_BLOCK); // __LAST may be empty, but we cannot assume this
1444 	if (segLASTDATACONSTB) {
1445 		arm_vm_page_granular_RWNX(segLASTDATACONSTB, segSizeLASTDATACONST, ARM64_GRANULE_ALLOW_BLOCK); // __LASTDATA_CONST may be empty, but we cannot assume this
1446 	}
1447 	arm_vm_page_granular_RWNX(segPRELINKDATAB, segSizePRELINKDATA, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT); // Prelink __DATA for kexts (RW data)
1448 
1449 	if (segSizePLKLLVMCOV > 0) {
1450 		arm_vm_page_granular_RWNX(segPLKLLVMCOVB, segSizePLKLLVMCOV, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT); // LLVM code coverage data
1451 	}
1452 	arm_vm_page_granular_RWNX(segPRELINKINFOB, segSizePRELINKINFO, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT); /* PreLinkInfoDictionary */
1453 
1454 	/* Record the bounds of the kernelcache. */
1455 	vm_kernelcache_base = segLOWEST;
1456 	vm_kernelcache_top = end_kern;
1457 }
1458 
1459 /*
1460  * return < 0 for a < b
1461  *          0 for a == b
1462  *        > 0 for a > b
1463  */
1464 typedef int (*cmpfunc_t)(const void *a, const void *b);
1465 
1466 extern void
1467 qsort(void *a, size_t n, size_t es, cmpfunc_t cmp);
1468 
1469 static int
cmp_ptov_entries(const void * a,const void * b)1470 cmp_ptov_entries(const void *a, const void *b)
1471 {
1472 	const ptov_table_entry *entry_a = a;
1473 	const ptov_table_entry *entry_b = b;
1474 	// Sort in descending order of segment length
1475 	if (entry_a->len < entry_b->len) {
1476 		return 1;
1477 	} else if (entry_a->len > entry_b->len) {
1478 		return -1;
1479 	} else {
1480 		return 0;
1481 	}
1482 }
1483 
1484 SECURITY_READ_ONLY_LATE(static unsigned int) ptov_index = 0;
1485 
1486 #define ROUND_L1(addr) (((addr) + ARM_TT_L1_OFFMASK) & ~(ARM_TT_L1_OFFMASK))
1487 #define ROUND_TWIG(addr) (((addr) + ARM_TT_TWIG_OFFMASK) & ~(ARM_TT_TWIG_OFFMASK))
1488 
1489 static void
arm_vm_physmap_slide(ptov_table_entry * temp_ptov_table,vm_map_address_t orig_va,vm_size_t len,int pte_prot_APX,unsigned granule)1490 arm_vm_physmap_slide(ptov_table_entry *temp_ptov_table, vm_map_address_t orig_va, vm_size_t len, int pte_prot_APX, unsigned granule)
1491 {
1492 	pmap_paddr_t pa_offset;
1493 
1494 	if (__improbable(ptov_index >= PTOV_TABLE_SIZE)) {
1495 		panic("%s: PTOV table limit exceeded; segment va = 0x%llx, size = 0x%llx", __func__,
1496 		    (unsigned long long)orig_va, (unsigned long long)len);
1497 	}
1498 	assert((orig_va & ARM_PGMASK) == 0);
1499 	temp_ptov_table[ptov_index].pa = orig_va - gVirtBase + gPhysBase;
1500 	if (ptov_index == 0) {
1501 		temp_ptov_table[ptov_index].va = physmap_base;
1502 	} else {
1503 		temp_ptov_table[ptov_index].va = temp_ptov_table[ptov_index - 1].va + temp_ptov_table[ptov_index - 1].len;
1504 	}
1505 	if (granule & ARM64_GRANULE_ALLOW_BLOCK) {
1506 		vm_map_address_t orig_offset = temp_ptov_table[ptov_index].pa & ARM_TT_TWIG_OFFMASK;
1507 		vm_map_address_t new_offset = temp_ptov_table[ptov_index].va & ARM_TT_TWIG_OFFMASK;
1508 		if (new_offset < orig_offset) {
1509 			temp_ptov_table[ptov_index].va += (orig_offset - new_offset);
1510 		} else if (new_offset > orig_offset) {
1511 			temp_ptov_table[ptov_index].va = ROUND_TWIG(temp_ptov_table[ptov_index].va) + orig_offset;
1512 		}
1513 	}
1514 	assert((temp_ptov_table[ptov_index].va & ARM_PGMASK) == 0);
1515 	temp_ptov_table[ptov_index].len = round_page(len);
1516 	pa_offset = temp_ptov_table[ptov_index].va - orig_va;
1517 	arm_vm_page_granular_prot(temp_ptov_table[ptov_index].va, temp_ptov_table[ptov_index].len, pa_offset, 1, pte_prot_APX, 1, granule);
1518 	++ptov_index;
1519 }
1520 
1521 #if XNU_MONITOR
1522 
1523 SECURITY_READ_ONLY_LATE(static boolean_t) keep_linkedit = FALSE;
1524 
1525 static void
arm_vm_physmap_init(boot_args * args)1526 arm_vm_physmap_init(boot_args *args)
1527 {
1528 	ptov_table_entry temp_ptov_table[PTOV_TABLE_SIZE];
1529 	bzero(temp_ptov_table, sizeof(temp_ptov_table));
1530 
1531 	// This is memory that will either be handed back to the VM layer via ml_static_mfree(),
1532 	// or will be available for general-purpose use.   Physical aperture mappings for this memory
1533 	// must be at page granularity, so that PPL ownership or cache attribute changes can be reflected
1534 	// in the physical aperture mappings.
1535 
1536 	// Slid region between gPhysBase and beginning of protected text
1537 	arm_vm_physmap_slide(temp_ptov_table, gVirtBase, segLOWEST - gVirtBase, AP_RWNA, 0);
1538 
1539 	// kext bootstrap segments
1540 #if !defined(KERNEL_INTEGRITY_KTRR) && !defined(KERNEL_INTEGRITY_CTRR)
1541 	/* __KLD,__text is covered by the rorgn */
1542 	arm_vm_physmap_slide(temp_ptov_table, segKLDB, segSizeKLD, AP_RONA, 0);
1543 #endif
1544 	arm_vm_physmap_slide(temp_ptov_table, segKLDDATAB, segSizeKLDDATA, AP_RONA, 0);
1545 
1546 	// Early-boot data
1547 	arm_vm_physmap_slide(temp_ptov_table, segBOOTDATAB, segSizeBOOTDATA, AP_RONA, 0);
1548 
1549 	PE_parse_boot_argn("keepsyms", &keep_linkedit, sizeof(keep_linkedit));
1550 #if CONFIG_DTRACE
1551 	if (dtrace_keep_kernel_symbols()) {
1552 		keep_linkedit = TRUE;
1553 	}
1554 #endif /* CONFIG_DTRACE */
1555 #if KASAN_DYNAMIC_BLACKLIST
1556 	/* KASAN's dynamic blacklist needs to query the LINKEDIT segment at runtime.  As such, the
1557 	 * kext bootstrap code will not jettison LINKEDIT on kasan kernels, so don't bother to relocate it. */
1558 	keep_linkedit = TRUE;
1559 #endif
1560 	if (!keep_linkedit) {
1561 		// Kernel LINKEDIT
1562 		arm_vm_physmap_slide(temp_ptov_table, segLINKB, segSizeLINK, AP_RWNA, 0);
1563 
1564 		if (segSizePLKLINKEDIT) {
1565 			// Prelinked kernel LINKEDIT
1566 			arm_vm_physmap_slide(temp_ptov_table, segPLKLINKEDITB, segSizePLKLINKEDIT, AP_RWNA, 0);
1567 		}
1568 	}
1569 
1570 	// Prelinked kernel plists
1571 	arm_vm_physmap_slide(temp_ptov_table, segPRELINKINFOB, segSizePRELINKINFO, AP_RWNA, 0);
1572 
1573 	// Device tree (if not locked down), ramdisk, boot args
1574 	arm_vm_physmap_slide(temp_ptov_table, end_kern, (args->topOfKernelData - gPhysBase + gVirtBase) - end_kern, AP_RWNA, 0);
1575 	if (!SecureDTIsLockedDown()) {
1576 		PE_slide_devicetree(temp_ptov_table[ptov_index - 1].va - end_kern);
1577 	}
1578 
1579 	// Remainder of physical memory
1580 	arm_vm_physmap_slide(temp_ptov_table, (args->topOfKernelData - gPhysBase + gVirtBase),
1581 	    real_avail_end - args->topOfKernelData, AP_RWNA, 0);
1582 
1583 	assert((temp_ptov_table[ptov_index - 1].va + temp_ptov_table[ptov_index - 1].len) <= physmap_end);
1584 
1585 	// Sort in descending order of segment length.  LUT traversal is linear, so largest (most likely used)
1586 	// segments should be placed earliest in the table to optimize lookup performance.
1587 	qsort(temp_ptov_table, PTOV_TABLE_SIZE, sizeof(temp_ptov_table[0]), cmp_ptov_entries);
1588 
1589 	memcpy(ptov_table, temp_ptov_table, sizeof(ptov_table));
1590 }
1591 
1592 #else
1593 
1594 static void
arm_vm_physmap_init(boot_args * args)1595 arm_vm_physmap_init(boot_args *args)
1596 {
1597 	ptov_table_entry temp_ptov_table[PTOV_TABLE_SIZE];
1598 	bzero(temp_ptov_table, sizeof(temp_ptov_table));
1599 
1600 	// Will be handed back to VM layer through ml_static_mfree() in arm_vm_prot_finalize()
1601 	arm_vm_physmap_slide(temp_ptov_table, gVirtBase, segLOWEST - gVirtBase, AP_RWNA,
1602 	    ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT);
1603 
1604 	arm_vm_page_granular_RWNX(end_kern, phystokv(args->topOfKernelData) - end_kern,
1605 	    ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT); /* Device Tree (if not locked down), RAM Disk (if present), bootArgs */
1606 
1607 	arm_vm_physmap_slide(temp_ptov_table, (args->topOfKernelData - gPhysBase + gVirtBase),
1608 	    real_avail_end - args->topOfKernelData, AP_RWNA, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT); // rest of physmem
1609 
1610 	assert((temp_ptov_table[ptov_index - 1].va + temp_ptov_table[ptov_index - 1].len) <= physmap_end);
1611 
1612 	// Sort in descending order of segment length.  LUT traversal is linear, so largest (most likely used)
1613 	// segments should be placed earliest in the table to optimize lookup performance.
1614 	qsort(temp_ptov_table, PTOV_TABLE_SIZE, sizeof(temp_ptov_table[0]), cmp_ptov_entries);
1615 
1616 	memcpy(ptov_table, temp_ptov_table, sizeof(ptov_table));
1617 }
1618 
1619 #endif // XNU_MONITOR
1620 
1621 void
arm_vm_prot_finalize(boot_args * args __unused)1622 arm_vm_prot_finalize(boot_args * args __unused)
1623 {
1624 	/*
1625 	 * At this point, we are far enough along in the boot process that it will be
1626 	 * safe to free up all of the memory preceeding the kernel.  It may in fact
1627 	 * be safe to do this earlier.
1628 	 *
1629 	 * This keeps the memory in the V-to-P mapping, but advertises it to the VM
1630 	 * as usable.
1631 	 */
1632 
1633 	/*
1634 	 * if old style PRELINK segment exists, free memory before it, and after it before XNU text
1635 	 * otherwise we're dealing with a new style kernel cache, so we should just free the
1636 	 * memory before PRELINK_TEXT segment, since the rest of the KEXT read only data segments
1637 	 * should be immediately followed by XNU's TEXT segment
1638 	 */
1639 
1640 	ml_static_mfree(phystokv(gPhysBase), segLOWEST - gVirtBase);
1641 
1642 	/*
1643 	 * KTRR support means we will be mucking with these pages and trying to
1644 	 * protect them; we cannot free the pages to the VM if we do this.
1645 	 */
1646 	if (!segSizePLKDATACONST && !segSizePLKTEXTEXEC && segSizePRELINKTEXT) {
1647 		/* If new segments not present, PRELINK_TEXT is not dynamically sized, free DRAM between it and xnu TEXT */
1648 		ml_static_mfree(segPRELINKTEXTB + segSizePRELINKTEXT, segTEXTB - (segPRELINKTEXTB + segSizePRELINKTEXT));
1649 	}
1650 
1651 	/* tighten permissions on kext read only data and code */
1652 	arm_vm_page_granular_RNX(segPRELINKTEXTB, segSizePRELINKTEXT, ARM64_GRANULE_ALLOW_BLOCK);
1653 	arm_vm_page_granular_RNX(segPLKDATACONSTB, segSizePLKDATACONST, ARM64_GRANULE_ALLOW_BLOCK);
1654 
1655 	cpu_stack_alloc(&BootCpuData);
1656 	arm64_replace_bootstack(&BootCpuData);
1657 	ml_static_mfree(phystokv(segBOOTDATAB - gVirtBase + gPhysBase), segSizeBOOTDATA);
1658 
1659 #if __ARM_KERNEL_PROTECT__
1660 	arm_vm_populate_kernel_el0_mappings();
1661 #endif /* __ARM_KERNEL_PROTECT__ */
1662 
1663 #if XNU_MONITOR
1664 #if !defined(KERNEL_INTEGRITY_KTRR) && !defined(KERNEL_INTEGRITY_CTRR)
1665 	/* __KLD,__text is covered by the rorgn */
1666 	for (vm_offset_t va = segKLDB; va < (segKLDB + segSizeKLD); va += ARM_PGBYTES) {
1667 		pt_entry_t *pte = arm_kva_to_pte(va);
1668 		*pte = ARM_PTE_EMPTY;
1669 	}
1670 #endif
1671 	for (vm_offset_t va = segKLDDATAB; va < (segKLDDATAB + segSizeKLDDATA); va += ARM_PGBYTES) {
1672 		pt_entry_t *pte = arm_kva_to_pte(va);
1673 		*pte = ARM_PTE_EMPTY;
1674 	}
1675 	/* Clear the original stack mappings; these pages should be mapped through ptov_table. */
1676 	for (vm_offset_t va = segBOOTDATAB; va < (segBOOTDATAB + segSizeBOOTDATA); va += ARM_PGBYTES) {
1677 		pt_entry_t *pte = arm_kva_to_pte(va);
1678 		*pte = ARM_PTE_EMPTY;
1679 	}
1680 	/* Clear the original PRELINKINFO mapping. This segment should be jettisoned during I/O Kit
1681 	 * initialization before we reach this point. */
1682 	for (vm_offset_t va = segPRELINKINFOB; va < (segPRELINKINFOB + segSizePRELINKINFO); va += ARM_PGBYTES) {
1683 		pt_entry_t *pte = arm_kva_to_pte(va);
1684 		*pte = ARM_PTE_EMPTY;
1685 	}
1686 	if (!keep_linkedit) {
1687 		for (vm_offset_t va = segLINKB; va < (segLINKB + segSizeLINK); va += ARM_PGBYTES) {
1688 			pt_entry_t *pte = arm_kva_to_pte(va);
1689 			*pte = ARM_PTE_EMPTY;
1690 		}
1691 		if (segSizePLKLINKEDIT) {
1692 			for (vm_offset_t va = segPLKLINKEDITB; va < (segPLKLINKEDITB + segSizePLKLINKEDIT); va += ARM_PGBYTES) {
1693 				pt_entry_t *pte = arm_kva_to_pte(va);
1694 				*pte = ARM_PTE_EMPTY;
1695 			}
1696 		}
1697 	}
1698 #endif /* XNU_MONITOR */
1699 
1700 #if defined(KERNEL_INTEGRITY_KTRR) || defined(KERNEL_INTEGRITY_CTRR)
1701 	/*
1702 	 * __LAST,__pinst should no longer be executable.
1703 	 */
1704 	arm_vm_page_granular_RNX(segLASTB, segSizeLAST, ARM64_GRANULE_ALLOW_BLOCK);
1705 
1706 	/* __LASTDATA_CONST should no longer be writable. */
1707 	if (segLASTDATACONSTB) {
1708 		arm_vm_page_granular_RNX(segLASTDATACONSTB, segSizeLASTDATACONST, ARM64_GRANULE_ALLOW_BLOCK);
1709 	}
1710 
1711 	/*
1712 	 * __KLD,__text should no longer be executable.
1713 	 */
1714 	arm_vm_page_granular_RNX(segKLDB, segSizeKLD, ARM64_GRANULE_ALLOW_BLOCK);
1715 
1716 	/*
1717 	 * Must wait until all other region permissions are set before locking down DATA_CONST
1718 	 * as the kernel static page tables live in DATA_CONST on KTRR enabled systems
1719 	 * and will become immutable.
1720 	 */
1721 #endif
1722 
1723 	arm_vm_page_granular_RNX(segDATACONSTB, segSizeDATACONST, ARM64_GRANULE_ALLOW_BLOCK);
1724 
1725 	__builtin_arm_dsb(DSB_ISH);
1726 	flush_mmu_tlb();
1727 }
1728 
1729 /*
1730  * TBI (top-byte ignore) is an ARMv8 feature for ignoring the top 8 bits of
1731  * address accesses. It can be enabled separately for TTBR0 (user) and
1732  * TTBR1 (kernel).
1733  */
1734 void
arm_set_kernel_tbi(void)1735 arm_set_kernel_tbi(void)
1736 {
1737 #if !__ARM_KERNEL_PROTECT__ && CONFIG_KERNEL_TBI
1738 	uint64_t old_tcr, new_tcr;
1739 
1740 	old_tcr = new_tcr = get_tcr();
1741 	/*
1742 	 * For kernel configurations that require TBI support on
1743 	 * PAC systems, we enable DATA TBI only.
1744 	 */
1745 	new_tcr |= TCR_TBI1_TOPBYTE_IGNORED;
1746 	new_tcr |= TCR_TBID1_ENABLE;
1747 
1748 	if (old_tcr != new_tcr) {
1749 		set_tcr(new_tcr);
1750 		sysreg_restore.tcr_el1 = new_tcr;
1751 	}
1752 #endif /* !__ARM_KERNEL_PROTECT__ && CONFIG_KERNEL_TBI */
1753 }
1754 
1755 static void
arm_set_user_tbi(void)1756 arm_set_user_tbi(void)
1757 {
1758 #if !__ARM_KERNEL_PROTECT__
1759 	uint64_t old_tcr, new_tcr;
1760 
1761 	old_tcr = new_tcr = get_tcr();
1762 	new_tcr |= TCR_TBI0_TOPBYTE_IGNORED;
1763 
1764 	if (old_tcr != new_tcr) {
1765 		set_tcr(new_tcr);
1766 		sysreg_restore.tcr_el1 = new_tcr;
1767 	}
1768 #endif /* !__ARM_KERNEL_PROTECT__ */
1769 }
1770 
1771 /*
1772  * Initialize and enter blank (invalid) page tables in a L1 translation table for a given VA range.
1773  *
1774  * This is a helper function used to build up the initial page tables for the kernel translation table.
1775  * With KERNEL_INTEGRITY we keep at least the root level of the kernel page table immutable, thus the need
1776  * to preallocate before machine_lockdown any L1 entries necessary during the entire kernel runtime.
1777  *
1778  * For a given VA range, if necessary, allocate new L2 translation tables and install the table entries in
1779  * the appropriate L1 table indexes. called before the translation table is active
1780  *
1781  * parameters:
1782  *
1783  * tt: virtual address of L1 translation table to modify
1784  * start: beginning of VA range
1785  * end: end of VA range
1786  * static_map: whether to allocate the new translation table page from read only memory
1787  * table_attrs: attributes of new table entry in addition to VALID and TYPE_TABLE attributes
1788  *
1789  */
1790 
1791 static void
init_ptpages(tt_entry_t * tt,vm_map_address_t start,vm_map_address_t end,bool static_map,uint64_t table_attrs)1792 init_ptpages(tt_entry_t *tt, vm_map_address_t start, vm_map_address_t end, bool static_map, uint64_t table_attrs)
1793 {
1794 	tt_entry_t *l1_tte;
1795 	vm_offset_t ptpage_vaddr;
1796 
1797 	l1_tte = tt + ((start & ARM_TT_L1_INDEX_MASK) >> ARM_TT_L1_SHIFT);
1798 
1799 	while (start < end) {
1800 		if (*l1_tte == ARM_TTE_EMPTY) {
1801 			/* Allocate a page and setup L1 Table TTE in L1 */
1802 			ptpage_vaddr = alloc_ptpage(static_map);
1803 			*l1_tte = (kvtophys(ptpage_vaddr) & ARM_TTE_TABLE_MASK) | ARM_TTE_TYPE_TABLE | ARM_TTE_VALID | table_attrs;
1804 			bzero((void *)ptpage_vaddr, ARM_PGBYTES);
1805 		}
1806 
1807 		if ((start + ARM_TT_L1_SIZE) < start) {
1808 			/* If this is the last L1 entry, it must cover the last mapping. */
1809 			break;
1810 		}
1811 
1812 		start += ARM_TT_L1_SIZE;
1813 		l1_tte++;
1814 	}
1815 }
1816 
1817 #define ARM64_PHYSMAP_SLIDE_RANGE (1ULL << 30) // 1 GB
1818 #define ARM64_PHYSMAP_SLIDE_MASK  (ARM64_PHYSMAP_SLIDE_RANGE - 1)
1819 
1820 void
arm_vm_init(uint64_t memory_size,boot_args * args)1821 arm_vm_init(uint64_t memory_size, boot_args * args)
1822 {
1823 	vm_map_address_t va_l1, va_l1_end;
1824 	tt_entry_t       *cpu_l1_tte;
1825 	vm_map_address_t va_l2, va_l2_end;
1826 	tt_entry_t       *cpu_l2_tte;
1827 	pmap_paddr_t     boot_ttep;
1828 	tt_entry_t       *boot_tte;
1829 	uint64_t         mem_segments;
1830 	vm_offset_t      ptpage_vaddr;
1831 	vm_map_address_t dynamic_memory_begin;
1832 
1833 	/*
1834 	 * Get the virtual and physical kernel-managed memory base from boot_args.
1835 	 */
1836 	gVirtBase = args->virtBase;
1837 	gPhysBase = args->physBase;
1838 #if KASAN
1839 	real_phys_size = args->memSize + (shadow_ptop - shadow_pbase);
1840 #else
1841 	real_phys_size = args->memSize;
1842 #endif
1843 	/*
1844 	 * Ensure the physical region we specify for the VM to manage ends on a
1845 	 * software page boundary.  Note that the software page size (PAGE_SIZE)
1846 	 * may be a multiple of the hardware page size specified in ARM_PGBYTES.
1847 	 * We must round the reported memory size down to the nearest PAGE_SIZE
1848 	 * boundary to ensure the VM does not try to manage a page it does not
1849 	 * completely own.  The KASAN shadow region, if present, is managed entirely
1850 	 * in units of the hardware page size and should not need similar treatment.
1851 	 */
1852 	gPhysSize = mem_size = ((gPhysBase + args->memSize) & ~PAGE_MASK) - gPhysBase;
1853 
1854 	mem_actual = args->memSizeActual ? args->memSizeActual : mem_size;
1855 
1856 	if ((memory_size != 0) && (mem_size > memory_size)) {
1857 		mem_size = memory_size;
1858 		max_mem_actual = memory_size;
1859 	} else {
1860 		max_mem_actual = mem_actual;
1861 	}
1862 #if !defined(ARM_LARGE_MEMORY)
1863 	if (mem_size >= ((VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS) / 2)) {
1864 		panic("Unsupported memory configuration %lx", mem_size);
1865 	}
1866 #endif
1867 
1868 #if defined(ARM_LARGE_MEMORY)
1869 	unsigned long physmap_l1_entries = ((real_phys_size + ARM64_PHYSMAP_SLIDE_RANGE) >> ARM_TT_L1_SHIFT) + 1;
1870 	physmap_base = VM_MIN_KERNEL_ADDRESS - (physmap_l1_entries << ARM_TT_L1_SHIFT);
1871 #else
1872 	physmap_base = phystokv(args->topOfKernelData);
1873 #endif
1874 
1875 	// Slide the physical aperture to a random page-aligned location within the slide range
1876 	uint64_t physmap_slide = early_random() & ARM64_PHYSMAP_SLIDE_MASK & ~((uint64_t)PAGE_MASK);
1877 	assert(physmap_slide < ARM64_PHYSMAP_SLIDE_RANGE);
1878 
1879 	physmap_base += physmap_slide;
1880 
1881 #if XNU_MONITOR
1882 	physmap_base = ROUND_TWIG(physmap_base);
1883 #if defined(ARM_LARGE_MEMORY)
1884 	static_memory_end = phystokv(args->topOfKernelData);
1885 #else
1886 	static_memory_end = physmap_base + mem_size;
1887 #endif // ARM_LARGE_MEMORY
1888 	physmap_end = physmap_base + real_phys_size;
1889 #else
1890 #if defined(ARM_LARGE_MEMORY)
1891 	/* For large memory systems with no PPL such as virtual machines */
1892 	static_memory_end = phystokv(args->topOfKernelData);
1893 	physmap_end = physmap_base + real_phys_size;
1894 #else
1895 	static_memory_end = physmap_base + mem_size + (PTOV_TABLE_SIZE * ARM_TT_TWIG_SIZE); // worst possible case for block alignment
1896 	physmap_end = physmap_base + real_phys_size + (PTOV_TABLE_SIZE * ARM_TT_TWIG_SIZE);
1897 #endif // ARM_LARGE_MEMORY
1898 #endif
1899 
1900 #if KASAN && !defined(ARM_LARGE_MEMORY)
1901 	/* add the KASAN stolen memory to the physmap */
1902 	dynamic_memory_begin = static_memory_end + (shadow_ptop - shadow_pbase);
1903 #else
1904 	dynamic_memory_begin = static_memory_end;
1905 #endif
1906 #if XNU_MONITOR
1907 	pmap_stacks_start = (void*)dynamic_memory_begin;
1908 	dynamic_memory_begin += PPL_STACK_REGION_SIZE;
1909 	pmap_stacks_end = (void*)dynamic_memory_begin;
1910 
1911 #if HAS_GUARDED_IO_FILTER
1912     iofilter_stacks_start = (void*)dynamic_memory_begin;
1913     dynamic_memory_begin += IOFILTER_STACK_REGION_SIZE;
1914     iofilter_stacks_end = (void*)dynamic_memory_begin;
1915 #endif
1916 #endif
1917 	if (dynamic_memory_begin > VM_MAX_KERNEL_ADDRESS) {
1918 		panic("Unsupported memory configuration %lx", mem_size);
1919 	}
1920 
1921 	boot_tte = (tt_entry_t *)&bootstrap_pagetables;
1922 	boot_ttep = kvtophys((vm_offset_t)boot_tte);
1923 
1924 #if DEVELOPMENT || DEBUG
1925 	/* Sanity check - assert that BOOTSTRAP_TABLE_SIZE is sufficiently-large to
1926 	 * hold our bootstrap mappings for any possible slide */
1927 	size_t bytes_mapped = dynamic_memory_begin - gVirtBase;
1928 	size_t l1_entries = 1 + ((bytes_mapped + ARM_TT_L1_SIZE - 1) / ARM_TT_L1_SIZE);
1929 	/* 1 L1 each for V=P and KVA, plus 1 page for each L2 */
1930 	size_t pages_used = 2 * (l1_entries + 1);
1931 	if (pages_used > BOOTSTRAP_TABLE_SIZE) {
1932 		panic("BOOTSTRAP_TABLE_SIZE too small for memory config");
1933 	}
1934 #endif
1935 
1936 	/*
1937 	 *  TTBR0 L1, TTBR0 L2 - 1:1 bootstrap mapping.
1938 	 *  TTBR1 L1, TTBR1 L2 - kernel mapping
1939 	 */
1940 
1941 	/*
1942 	 * TODO: free bootstrap table memory back to allocator.
1943 	 * on large memory systems bootstrap tables could be quite large.
1944 	 * after bootstrap complete, xnu can warm start with a single 16KB page mapping
1945 	 * to trampoline to KVA. this requires only 3 pages to stay resident.
1946 	 */
1947 	first_avail_phys = avail_start = args->topOfKernelData;
1948 
1949 #if defined(KERNEL_INTEGRITY_KTRR) || defined(KERNEL_INTEGRITY_CTRR)
1950 	arm_replace_identity_map();
1951 #endif
1952 
1953 	/* Initialize invalid tte page */
1954 	invalid_tte = (tt_entry_t *)alloc_ptpage(TRUE);
1955 	invalid_ttep = kvtophys((vm_offset_t)invalid_tte);
1956 	bzero(invalid_tte, ARM_PGBYTES);
1957 
1958 	/*
1959 	 * Initialize l1 page table page
1960 	 */
1961 	cpu_tte = (tt_entry_t *)alloc_ptpage(TRUE);
1962 	cpu_ttep = kvtophys((vm_offset_t)cpu_tte);
1963 	bzero(cpu_tte, ARM_PGBYTES);
1964 	avail_end = gPhysBase + mem_size;
1965 	assert(!(avail_end & PAGE_MASK));
1966 
1967 #if KASAN
1968 	real_avail_end = gPhysBase + real_phys_size;
1969 #else
1970 	real_avail_end = avail_end;
1971 #endif
1972 
1973 	/*
1974 	 * Initialize l1 and l2 page table pages :
1975 	 *   map physical memory at the kernel base virtual address
1976 	 *   cover the kernel dynamic address range section
1977 	 *
1978 	 *   the so called physical aperture should be statically mapped
1979 	 */
1980 	init_ptpages(cpu_tte, gVirtBase, dynamic_memory_begin, TRUE, ARM_TTE_TABLE_AP(ARM_TTE_TABLE_AP_USER_NA));
1981 
1982 #if defined(ARM_LARGE_MEMORY)
1983 	/*
1984 	 * Initialize l1 page table pages :
1985 	 *   on large memory systems the physical aperture exists separately below
1986 	 *   the rest of the kernel virtual address space
1987 	 */
1988 	init_ptpages(cpu_tte, physmap_base, ROUND_L1(physmap_end), TRUE, ARM_DYNAMIC_TABLE_XN | ARM_TTE_TABLE_AP(ARM_TTE_TABLE_AP_USER_NA));
1989 #endif
1990 
1991 
1992 #if __ARM_KERNEL_PROTECT__
1993 	/* Expand the page tables to prepare for the EL0 mappings. */
1994 	arm_vm_expand_kernel_el0_mappings();
1995 #endif /* __ARM_KERNEL_PROTECT__ */
1996 
1997 	/*
1998 	 * Now retrieve addresses for various segments from kernel mach-o header
1999 	 */
2000 	segPRELINKTEXTB  = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__PRELINK_TEXT", &segSizePRELINKTEXT);
2001 	segPLKDATACONSTB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__PLK_DATA_CONST", &segSizePLKDATACONST);
2002 	segPLKTEXTEXECB  = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__PLK_TEXT_EXEC", &segSizePLKTEXTEXEC);
2003 	segTEXTB         = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__TEXT", &segSizeTEXT);
2004 	segDATACONSTB    = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__DATA_CONST", &segSizeDATACONST);
2005 	segTEXTEXECB     = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__TEXT_EXEC", &segSizeTEXTEXEC);
2006 #if XNU_MONITOR
2007 	segPPLTEXTB      = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__PPLTEXT", &segSizePPLTEXT);
2008 	segPPLTRAMPB     = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__PPLTRAMP", &segSizePPLTRAMP);
2009 	segPPLDATACONSTB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__PPLDATA_CONST", &segSizePPLDATACONST);
2010 #endif
2011 	segDATAB         = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__DATA", &segSizeDATA);
2012 #if XNU_MONITOR
2013 	segPPLDATAB      = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__PPLDATA", &segSizePPLDATA);
2014 #endif
2015 
2016 	segBOOTDATAB     = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__BOOTDATA", &segSizeBOOTDATA);
2017 	segLINKB         = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__LINKEDIT", &segSizeLINK);
2018 	segKLDB          = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__KLD", &segSizeKLD);
2019 	segKLDDATAB      = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__KLDDATA", &segSizeKLDDATA);
2020 	segPRELINKDATAB  = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__PRELINK_DATA", &segSizePRELINKDATA);
2021 	segPRELINKINFOB  = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__PRELINK_INFO", &segSizePRELINKINFO);
2022 	segPLKLLVMCOVB   = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__PLK_LLVM_COV", &segSizePLKLLVMCOV);
2023 	segPLKLINKEDITB  = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__PLK_LINKEDIT", &segSizePLKLINKEDIT);
2024 	segLASTB         = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__LAST", &segSizeLAST);
2025 	segLASTDATACONSTB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__LASTDATA_CONST", &segSizeLASTDATACONST);
2026 
2027 	sectHIBTEXTB     = (vm_offset_t) getsectdatafromheader(&_mh_execute_header, "__TEXT_EXEC", "__hib_text", &sectSizeHIBTEXT);
2028 	sectHIBDATACONSTB = (vm_offset_t) getsectdatafromheader(&_mh_execute_header, "__DATA_CONST", "__hib_const", &sectSizeHIBDATACONST);
2029 	segHIBDATAB      = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__HIBDATA", &segSizeHIBDATA);
2030 
2031 	if (kernel_mach_header_is_in_fileset(&_mh_execute_header)) {
2032 		kernel_mach_header_t *kc_mh = PE_get_kc_header(KCKindPrimary);
2033 
2034 		// fileset has kext PLK_TEXT_EXEC under kernel collection TEXT_EXEC following kernel's LAST
2035 		segKCTEXTEXECB = (vm_offset_t) getsegdatafromheader(kc_mh,             "__TEXT_EXEC", &segSizeKCTEXTEXEC);
2036 		assert(segPLKTEXTEXECB && !segSizePLKTEXTEXEC);                        // kernel PLK_TEXT_EXEC must be empty
2037 
2038 		assert(segLASTB);                                                      // kernel LAST can be empty, but it must have
2039 		                                                                       // a valid address for computations below.
2040 
2041 		assert(segKCTEXTEXECB <= segLASTB);                                    // KC TEXT_EXEC must contain kernel LAST
2042 		assert(segKCTEXTEXECB + segSizeKCTEXTEXEC >= segLASTB + segSizeLAST);
2043 		segPLKTEXTEXECB = segLASTB + segSizeLAST;
2044 		segSizePLKTEXTEXEC = segSizeKCTEXTEXEC - (segPLKTEXTEXECB - segKCTEXTEXECB);
2045 
2046 		// fileset has kext PLK_DATA_CONST under kernel collection DATA_CONST following kernel's LASTDATA_CONST
2047 		segKCDATACONSTB = (vm_offset_t) getsegdatafromheader(kc_mh,            "__DATA_CONST", &segSizeKCDATACONST);
2048 		assert(segPLKDATACONSTB && !segSizePLKDATACONST);                      // kernel PLK_DATA_CONST must be empty
2049 		assert(segLASTDATACONSTB && segSizeLASTDATACONST);                     // kernel LASTDATA_CONST must be non-empty
2050 		assert(segKCDATACONSTB <= segLASTDATACONSTB);                          // KC DATA_CONST must contain kernel LASTDATA_CONST
2051 		assert(segKCDATACONSTB + segSizeKCDATACONST >= segLASTDATACONSTB + segSizeLASTDATACONST);
2052 		segPLKDATACONSTB = segLASTDATACONSTB + segSizeLASTDATACONST;
2053 		segSizePLKDATACONST = segSizeKCDATACONST - (segPLKDATACONSTB - segKCDATACONSTB);
2054 
2055 		// fileset has kext PRELINK_DATA under kernel collection DATA following kernel's empty PRELINK_DATA
2056 		segKCDATAB      = (vm_offset_t) getsegdatafromheader(kc_mh,            "__DATA", &segSizeKCDATA);
2057 		assert(segPRELINKDATAB && !segSizePRELINKDATA);                        // kernel PRELINK_DATA must be empty
2058 		assert(segKCDATAB <= segPRELINKDATAB);                                 // KC DATA must contain kernel PRELINK_DATA
2059 		assert(segKCDATAB + segSizeKCDATA >= segPRELINKDATAB + segSizePRELINKDATA);
2060 		segSizePRELINKDATA = segSizeKCDATA - (segPRELINKDATAB - segKCDATAB);
2061 
2062 		// fileset has consolidated PRELINK_TEXT, PRELINK_INFO and LINKEDIT at the kernel collection level
2063 		assert(segPRELINKTEXTB && !segSizePRELINKTEXT);                        // kernel PRELINK_TEXT must be empty
2064 		segPRELINKTEXTB = (vm_offset_t) getsegdatafromheader(kc_mh,            "__PRELINK_TEXT", &segSizePRELINKTEXT);
2065 		assert(segPRELINKINFOB && !segSizePRELINKINFO);                        // kernel PRELINK_INFO must be empty
2066 		segPRELINKINFOB = (vm_offset_t) getsegdatafromheader(kc_mh,            "__PRELINK_INFO", &segSizePRELINKINFO);
2067 		segLINKB        = (vm_offset_t) getsegdatafromheader(kc_mh,            "__LINKEDIT", &segSizeLINK);
2068 	}
2069 
2070 	(void) PE_parse_boot_argn("use_contiguous_hint", &use_contiguous_hint, sizeof(use_contiguous_hint));
2071 	assert(segSizePRELINKTEXT < 0x03000000); /* 23355738 */
2072 
2073 	/* if one of the new segments is present, the other one better be as well */
2074 	if (segSizePLKDATACONST || segSizePLKTEXTEXEC) {
2075 		assert(segSizePLKDATACONST && segSizePLKTEXTEXEC);
2076 	}
2077 
2078 	etext = (vm_offset_t) segTEXTB + segSizeTEXT;
2079 	sdata = (vm_offset_t) segDATAB;
2080 	edata = (vm_offset_t) segDATAB + segSizeDATA;
2081 	end_kern = round_page(segHIGHESTKC ? segHIGHESTKC : getlastkerneladdr()); /* Force end to next page */
2082 
2083 	vm_set_page_size();
2084 
2085 	vm_kernel_base = segTEXTB;
2086 	vm_kernel_top = (vm_offset_t) &last_kernel_symbol;
2087 	vm_kext_base = segPRELINKTEXTB;
2088 	vm_kext_top = vm_kext_base + segSizePRELINKTEXT;
2089 
2090 	vm_prelink_stext = segPRELINKTEXTB;
2091 	if (!segSizePLKTEXTEXEC && !segSizePLKDATACONST) {
2092 		vm_prelink_etext = segPRELINKTEXTB + segSizePRELINKTEXT;
2093 	} else {
2094 		vm_prelink_etext = segPRELINKTEXTB + segSizePRELINKTEXT + segSizePLKDATACONST + segSizePLKTEXTEXEC;
2095 	}
2096 	vm_prelink_sinfo = segPRELINKINFOB;
2097 	vm_prelink_einfo = segPRELINKINFOB + segSizePRELINKINFO;
2098 	vm_slinkedit = segLINKB;
2099 	vm_elinkedit = segLINKB + segSizeLINK;
2100 
2101 	vm_prelink_sdata = segPRELINKDATAB;
2102 	vm_prelink_edata = segPRELINKDATAB + segSizePRELINKDATA;
2103 
2104 	arm_vm_prot_init(args);
2105 
2106 	/*
2107 	 * Initialize the page tables for the low globals:
2108 	 *   cover this address range:
2109 	 *     LOW_GLOBAL_BASE_ADDRESS + 2MB
2110 	 */
2111 	va_l1 = va_l2 = LOW_GLOBAL_BASE_ADDRESS;
2112 	cpu_l1_tte = cpu_tte + ((va_l1 & ARM_TT_L1_INDEX_MASK) >> ARM_TT_L1_SHIFT);
2113 	cpu_l2_tte = ((tt_entry_t *) phystokv(((*cpu_l1_tte) & ARM_TTE_TABLE_MASK))) + ((va_l2 & ARM_TT_L2_INDEX_MASK) >> ARM_TT_L2_SHIFT);
2114 	ptpage_vaddr = alloc_ptpage(TRUE);
2115 	*cpu_l2_tte = (kvtophys(ptpage_vaddr) & ARM_TTE_TABLE_MASK) | ARM_TTE_TYPE_TABLE | ARM_TTE_VALID | ARM_TTE_TABLE_PXN | ARM_TTE_TABLE_XN;
2116 	bzero((void *)ptpage_vaddr, ARM_PGBYTES);
2117 
2118 	/*
2119 	 * Initialize l2 page table pages :
2120 	 *   cover this address range:
2121 	 *    KERNEL_DYNAMIC_ADDR - VM_MAX_KERNEL_ADDRESS
2122 	 */
2123 #if defined(ARM_LARGE_MEMORY)
2124 	/*
2125 	 * dynamic mapped memory outside the VM allocator VA range required to bootstrap VM system
2126 	 * don't expect to exceed 64GB, no sense mapping any more space between here and the VM heap range
2127 	 */
2128 	init_ptpages(cpu_tte, dynamic_memory_begin, ROUND_L1(dynamic_memory_begin), FALSE, ARM_DYNAMIC_TABLE_XN | ARM_TTE_TABLE_AP(ARM_TTE_TABLE_AP_USER_NA));
2129 #else
2130 	/*
2131 	 * TODO: do these pages really need to come from RO memory?
2132 	 * With legacy 3 level table systems we never mapped more than a single L1 entry so this may be dead code
2133 	 */
2134 	init_ptpages(cpu_tte, dynamic_memory_begin, VM_MAX_KERNEL_ADDRESS, TRUE, ARM_DYNAMIC_TABLE_XN | ARM_TTE_TABLE_AP(ARM_TTE_TABLE_AP_USER_NA));
2135 #endif
2136 
2137 #if KASAN
2138 	/* record the extent of the physmap */
2139 	physmap_vbase = physmap_base;
2140 	physmap_vtop = physmap_end;
2141 	kasan_init();
2142 #endif /* KASAN */
2143 
2144 #if MONOTONIC
2145 	mt_early_init();
2146 #endif /* MONOTONIC */
2147 
2148 	arm_set_user_tbi();
2149 
2150 	arm_vm_physmap_init(args);
2151 	set_mmu_ttb_alternate(cpu_ttep & TTBR_BADDR_MASK);
2152 
2153 	ml_enable_monitor();
2154 
2155 	set_mmu_ttb(invalid_ttep & TTBR_BADDR_MASK);
2156 
2157 	flush_mmu_tlb();
2158 	kva_active = TRUE;
2159 	// global table pointers may need to be different due to physical aperture remapping
2160 	cpu_tte = (tt_entry_t*)(phystokv(cpu_ttep));
2161 	invalid_tte = (tt_entry_t*)(phystokv(invalid_ttep));
2162 
2163 	// From here on out, we're off the bootstrap translation tables.
2164 
2165 
2166 	/* AuxKC initialization has to be deferred until this point, since
2167 	 * the AuxKC may not have been fully mapped in the bootstrap
2168 	 * tables, if it spilled downwards into the prior L2 block.
2169 	 *
2170 	 * Now that its mapping set up by arm_vm_prot_init() is active,
2171 	 * we can traverse and fix it up.
2172 	 */
2173 
2174 	/* Calculate the physical bounds of the kernelcache; using
2175 	 * gVirtBase/gPhysBase math to do this directly is generally a bad idea
2176 	 * as the physmap is no longer physically contiguous.  However, this is
2177 	 * done here as segLOWEST and end_kern are both virtual addresses the
2178 	 * bootstrap physmap, and because kvtophys references the page tables
2179 	 * (at least at the time this comment was written), meaning that at
2180 	 * least end_kern may not point to a valid mapping on some kernelcache
2181 	 * configurations, so kvtophys would report a physical address of 0.
2182 	 *
2183 	 * Long term, the kernelcache should probably be described in terms of
2184 	 * multiple physical ranges, as there is no strong guarantee or
2185 	 * requirement that the kernelcache will always be physically
2186 	 * contiguous.
2187 	 */
2188 	arm_vm_kernelcache_phys_start = segLOWEST - gVirtBase + gPhysBase;
2189 	arm_vm_kernelcache_phys_end = end_kern - gVirtBase + gPhysBase;;
2190 
2191 	/* Calculate the number of pages that belong to the kernelcache. */
2192 	vm_page_kernelcache_count = (unsigned int) (atop_64(arm_vm_kernelcache_phys_end - arm_vm_kernelcache_phys_start));
2193 
2194 	if (arm_vm_auxkc_init()) {
2195 		if (segLOWESTROAuxKC < segLOWESTRO) {
2196 			segLOWESTRO = segLOWESTROAuxKC;
2197 		}
2198 		if (segHIGHESTROAuxKC > segHIGHESTRO) {
2199 			segHIGHESTRO = segHIGHESTROAuxKC;
2200 		}
2201 		if (segLOWESTRXAuxKC < segLOWESTTEXT) {
2202 			segLOWESTTEXT = segLOWESTRXAuxKC;
2203 		}
2204 		assert(segLOWEST == segLOWESTAuxKC);
2205 
2206 		// The preliminary auxKC mapping has been broken up.
2207 		flush_mmu_tlb();
2208 	}
2209 
2210 	sane_size = mem_size - (avail_start - gPhysBase);
2211 	max_mem = mem_size;
2212 	vm_kernel_slid_base = segLOWESTTEXT;
2213 	// vm_kernel_slide is set by arm_init()->arm_slide_rebase_and_sign_image()
2214 	vm_kernel_stext = segTEXTB;
2215 
2216 	if (kernel_mach_header_is_in_fileset(&_mh_execute_header)) {
2217 		vm_kernel_etext = segTEXTEXECB + segSizeTEXTEXEC;
2218 		vm_kernel_slid_top = vm_slinkedit;
2219 	} else {
2220 		assert(segDATACONSTB == segTEXTB + segSizeTEXT);
2221 		assert(segTEXTEXECB == segDATACONSTB + segSizeDATACONST);
2222 		vm_kernel_etext = segTEXTB + segSizeTEXT + segSizeDATACONST + segSizeTEXTEXEC;
2223 		vm_kernel_slid_top = vm_prelink_einfo;
2224 	}
2225 
2226 	dynamic_memory_begin = ROUND_TWIG(dynamic_memory_begin);
2227 #if defined(KERNEL_INTEGRITY_CTRR) && defined(CONFIG_XNUPOST)
2228 	// reserve a 32MB region without permission overrides to use later for a CTRR unit test
2229 	{
2230 		extern vm_offset_t ctrr_test_page;
2231 		tt_entry_t *new_tte;
2232 
2233 		ctrr_test_page = dynamic_memory_begin;
2234 		dynamic_memory_begin += ARM_TT_L2_SIZE;
2235 		cpu_l1_tte = cpu_tte + ((ctrr_test_page & ARM_TT_L1_INDEX_MASK) >> ARM_TT_L1_SHIFT);
2236 		assert((*cpu_l1_tte) & ARM_TTE_VALID);
2237 		cpu_l2_tte = ((tt_entry_t *) phystokv(((*cpu_l1_tte) & ARM_TTE_TABLE_MASK))) + ((ctrr_test_page & ARM_TT_L2_INDEX_MASK) >> ARM_TT_L2_SHIFT);
2238 		assert((*cpu_l2_tte) == ARM_TTE_EMPTY);
2239 		new_tte = (tt_entry_t *)alloc_ptpage(FALSE);
2240 		bzero(new_tte, ARM_PGBYTES);
2241 		*cpu_l2_tte = (kvtophys((vm_offset_t)new_tte) & ARM_TTE_TABLE_MASK) | ARM_TTE_TYPE_TABLE | ARM_TTE_VALID;
2242 	}
2243 #endif /* defined(KERNEL_INTEGRITY_CTRR) && defined(CONFIG_XNUPOST) */
2244 #if XNU_MONITOR
2245 	for (vm_offset_t cur = (vm_offset_t)pmap_stacks_start; cur < (vm_offset_t)pmap_stacks_end; cur += ARM_PGBYTES) {
2246 		arm_vm_map(cpu_tte, cur, ARM_PTE_EMPTY);
2247 	}
2248 #if HAS_GUARDED_IO_FILTER
2249     for (vm_offset_t cur = (vm_offset_t)iofilter_stacks_start; cur < (vm_offset_t)iofilter_stacks_end; cur += ARM_PGBYTES) {
2250         arm_vm_map(cpu_tte, cur, ARM_PTE_EMPTY);
2251     }
2252 #endif
2253 #endif
2254 	pmap_bootstrap(dynamic_memory_begin);
2255 
2256 	disable_preemption();
2257 
2258 	/*
2259 	 * Initialize l3 page table pages :
2260 	 *   cover this address range:
2261 	 *    2MB + FrameBuffer size + 10MB for each 256MB segment
2262 	 */
2263 
2264 	mem_segments = (mem_size + 0x0FFFFFFF) >> 28;
2265 
2266 	va_l1 = dynamic_memory_begin;
2267 	va_l1_end = va_l1 + ((2 + (mem_segments * 10)) << 20);
2268 	va_l1_end += round_page(args->Video.v_height * args->Video.v_rowBytes);
2269 	va_l1_end = (va_l1_end + 0x00000000007FFFFFULL) & 0xFFFFFFFFFF800000ULL;
2270 
2271 	cpu_l1_tte = cpu_tte + ((va_l1 & ARM_TT_L1_INDEX_MASK) >> ARM_TT_L1_SHIFT);
2272 
2273 	while (va_l1 < va_l1_end) {
2274 		va_l2 = va_l1;
2275 
2276 		if (((va_l1 & ~ARM_TT_L1_OFFMASK) + ARM_TT_L1_SIZE) < va_l1) {
2277 			/* If this is the last L1 entry, it must cover the last mapping. */
2278 			va_l2_end = va_l1_end;
2279 		} else {
2280 			va_l2_end = MIN((va_l1 & ~ARM_TT_L1_OFFMASK) + ARM_TT_L1_SIZE, va_l1_end);
2281 		}
2282 
2283 		cpu_l2_tte = ((tt_entry_t *) phystokv(((*cpu_l1_tte) & ARM_TTE_TABLE_MASK))) + ((va_l2 & ARM_TT_L2_INDEX_MASK) >> ARM_TT_L2_SHIFT);
2284 
2285 		while (va_l2 < va_l2_end) {
2286 			pt_entry_t *    ptp;
2287 			pmap_paddr_t    ptp_phys;
2288 
2289 			/* Allocate a page and setup L3 Table TTE in L2 */
2290 			ptp = (pt_entry_t *) alloc_ptpage(FALSE);
2291 			ptp_phys = (pmap_paddr_t)kvtophys((vm_offset_t)ptp);
2292 
2293 			bzero(ptp, ARM_PGBYTES);
2294 			pmap_init_pte_page(kernel_pmap, ptp, va_l2, 3, TRUE);
2295 
2296 			*cpu_l2_tte = (pa_to_tte(ptp_phys)) | ARM_TTE_TYPE_TABLE | ARM_TTE_VALID | ARM_DYNAMIC_TABLE_XN;
2297 
2298 			va_l2 += ARM_TT_L2_SIZE;
2299 			cpu_l2_tte++;
2300 		}
2301 
2302 		va_l1 = va_l2_end;
2303 		cpu_l1_tte++;
2304 	}
2305 
2306 #if defined(KERNEL_INTEGRITY_KTRR) || defined(KERNEL_INTEGRITY_CTRR)
2307 	/*
2308 	 * In this configuration, the bootstrap mappings (arm_vm_init) and
2309 	 * the heap mappings occupy separate L1 regions.  Explicitly set up
2310 	 * the heap L1 allocations here.
2311 	 */
2312 #if defined(ARM_LARGE_MEMORY)
2313 	init_ptpages(cpu_tte, KERNEL_PMAP_HEAP_RANGE_START & ~ARM_TT_L1_OFFMASK, VM_MAX_KERNEL_ADDRESS, FALSE, ARM_DYNAMIC_TABLE_XN | ARM_TTE_TABLE_AP(ARM_TTE_TABLE_AP_USER_NA));
2314 #else // defined(ARM_LARGE_MEMORY)
2315 	va_l1 = VM_MIN_KERNEL_ADDRESS & ~ARM_TT_L1_OFFMASK;
2316 	init_ptpages(cpu_tte, VM_MIN_KERNEL_ADDRESS & ~ARM_TT_L1_OFFMASK, VM_MAX_KERNEL_ADDRESS, FALSE, ARM_DYNAMIC_TABLE_XN | ARM_TTE_TABLE_AP(ARM_TTE_TABLE_AP_USER_NA));
2317 #endif // defined(ARM_LARGE_MEMORY)
2318 #else
2319 #if defined(ARM_LARGE_MEMORY)
2320 	/* For large memory systems with no KTRR/CTRR such as virtual machines */
2321 	init_ptpages(cpu_tte, KERNEL_PMAP_HEAP_RANGE_START & ~ARM_TT_L1_OFFMASK, VM_MAX_KERNEL_ADDRESS, FALSE, ARM_DYNAMIC_TABLE_XN | ARM_TTE_TABLE_AP(ARM_TTE_TABLE_AP_USER_NA));
2322 #endif
2323 #endif // defined(KERNEL_INTEGRITY_KTRR) || defined(KERNEL_INTEGRITY_CTRR)
2324 
2325 	/*
2326 	 * Initialize l3 page table pages :
2327 	 *   cover this address range:
2328 	 *   ((VM_MAX_KERNEL_ADDRESS & CPUWINDOWS_BASE_MASK) - PE_EARLY_BOOT_VA) to VM_MAX_KERNEL_ADDRESS
2329 	 */
2330 	va_l1 = (VM_MAX_KERNEL_ADDRESS & CPUWINDOWS_BASE_MASK) - PE_EARLY_BOOT_VA;
2331 	va_l1_end = VM_MAX_KERNEL_ADDRESS;
2332 
2333 	cpu_l1_tte = cpu_tte + ((va_l1 & ARM_TT_L1_INDEX_MASK) >> ARM_TT_L1_SHIFT);
2334 
2335 	while (va_l1 < va_l1_end) {
2336 		va_l2 = va_l1;
2337 
2338 		if (((va_l1 & ~ARM_TT_L1_OFFMASK) + ARM_TT_L1_SIZE) < va_l1) {
2339 			/* If this is the last L1 entry, it must cover the last mapping. */
2340 			va_l2_end = va_l1_end;
2341 		} else {
2342 			va_l2_end = MIN((va_l1 & ~ARM_TT_L1_OFFMASK) + ARM_TT_L1_SIZE, va_l1_end);
2343 		}
2344 
2345 		cpu_l2_tte = ((tt_entry_t *) phystokv(((*cpu_l1_tte) & ARM_TTE_TABLE_MASK))) + ((va_l2 & ARM_TT_L2_INDEX_MASK) >> ARM_TT_L2_SHIFT);
2346 
2347 		while (va_l2 < va_l2_end) {
2348 			pt_entry_t *    ptp;
2349 			pmap_paddr_t    ptp_phys;
2350 
2351 			/* Allocate a page and setup L3 Table TTE in L2 */
2352 			ptp = (pt_entry_t *) alloc_ptpage(FALSE);
2353 			ptp_phys = (pmap_paddr_t)kvtophys((vm_offset_t)ptp);
2354 
2355 			bzero(ptp, ARM_PGBYTES);
2356 			pmap_init_pte_page(kernel_pmap, ptp, va_l2, 3, TRUE);
2357 
2358 			*cpu_l2_tte = (pa_to_tte(ptp_phys)) | ARM_TTE_TYPE_TABLE | ARM_TTE_VALID | ARM_DYNAMIC_TABLE_XN;
2359 
2360 			va_l2 += ARM_TT_L2_SIZE;
2361 			cpu_l2_tte++;
2362 		}
2363 
2364 		va_l1 = va_l2_end;
2365 		cpu_l1_tte++;
2366 	}
2367 
2368 
2369 	/*
2370 	 * Adjust avail_start so that the range that the VM owns
2371 	 * starts on a PAGE_SIZE aligned boundary.
2372 	 */
2373 	avail_start = (avail_start + PAGE_MASK) & ~PAGE_MASK;
2374 
2375 #if XNU_MONITOR
2376 	pmap_static_allocations_done();
2377 #endif
2378 	first_avail = avail_start;
2379 	patch_low_glo_static_region(args->topOfKernelData, avail_start - args->topOfKernelData);
2380 	enable_preemption();
2381 }
2382 
2383 /*
2384  * Returns true if the address is within __TEXT, __TEXT_EXEC or __DATA_CONST
2385  * segment range. This is what [vm_kernel_stext, vm_kernel_etext) range used to
2386  * cover. The segments together may not be continuous anymore and so individual
2387  * intervals are inspected.
2388  */
2389 bool
kernel_text_contains(vm_offset_t addr)2390 kernel_text_contains(vm_offset_t addr)
2391 {
2392 	if (segTEXTB <= addr && addr < (segTEXTB + segSizeTEXT)) {
2393 		return true;
2394 	}
2395 	if (segTEXTEXECB <= addr && addr < (segTEXTEXECB + segSizeTEXTEXEC)) {
2396 		return true;
2397 	}
2398 	return segDATACONSTB <= addr && addr < (segDATACONSTB + segSizeDATACONST);
2399 }
2400