]> git.saurik.com Git - apple/xnu.git/blob - osfmk/arm64/arm_vm_init.c
projects / apple / xnu.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
xnu-7195.101.1.tar.gz
[apple/xnu.git] / osfmk / arm64 / arm_vm_init.c
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_debug.h>
30 #include <mach_kdp.h>
31 #include <debug.h>
32
33 #include <kern/assert.h>
34 #include <kern/misc_protos.h>
35 #include <kern/monotonic.h>
36 #include <mach/vm_types.h>
37 #include <mach/vm_param.h>
38 #include <vm/vm_kern.h>
39 #include <vm/vm_page.h>
40 #include <vm/pmap.h>
41
42 #include <machine/atomic.h>
43 #include <arm64/proc_reg.h>
44 #include <arm64/lowglobals.h>
45 #include <arm/cpu_data_internal.h>
46 #include <arm/misc_protos.h>
47 #include <pexpert/arm64/boot.h>
48 #include <pexpert/device_tree.h>
49
50 #include <libkern/kernel_mach_header.h>
51 #include <libkern/section_keywords.h>
52
53 #include <san/kasan.h>
54
55 #if __ARM_KERNEL_PROTECT__
56 /*
57 * If we want to support __ARM_KERNEL_PROTECT__, we need a sufficient amount of
58 * mappable space preceeding the kernel (as we unmap the kernel by cutting the
59 * range covered by TTBR1 in half). This must also cover the exception vectors.
60 */
61 static_assert(KERNEL_PMAP_HEAP_RANGE_START > ARM_KERNEL_PROTECT_EXCEPTION_START);
62
63 /* The exception vectors and the kernel cannot share root TTEs. */
64 static_assert((KERNEL_PMAP_HEAP_RANGE_START & ~ARM_TT_ROOT_OFFMASK) > ARM_KERNEL_PROTECT_EXCEPTION_START);
65
66 /*
67 * We must have enough space in the TTBR1_EL1 range to create the EL0 mapping of
68 * the exception vectors.
69 */
70 static_assert((((~ARM_KERNEL_PROTECT_EXCEPTION_START) + 1) * 2ULL) <= (ARM_TT_ROOT_SIZE + ARM_TT_ROOT_INDEX_MASK));
71 #endif /* __ARM_KERNEL_PROTECT__ */
72
73 #define ARM_DYNAMIC_TABLE_XN (ARM_TTE_TABLE_PXN | ARM_TTE_TABLE_XN)
74
75 #if KASAN
76 extern vm_offset_t shadow_pbase;
77 extern vm_offset_t shadow_ptop;
78 extern vm_offset_t physmap_vbase;
79 extern vm_offset_t physmap_vtop;
80 #endif
81
82 /*
83 * We explicitly place this in const, as it is not const from a language
84 * perspective, but it is only modified before we actually switch away from
85 * the bootstrap page tables.
86 */
87 SECURITY_READ_ONLY_LATE(uint8_t) bootstrap_pagetables[BOOTSTRAP_TABLE_SIZE] __attribute__((aligned(ARM_PGBYTES)));
88
89 /*
90 * Denotes the end of xnu.
91 */
92 extern void *last_kernel_symbol;
93
94 extern void arm64_replace_bootstack(cpu_data_t*);
95 extern void PE_slide_devicetree(vm_offset_t);
96
97 /*
98 * KASLR parameters
99 */
100 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_kernel_base;
101 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_kernel_top;
102 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_kext_base;
103 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_kext_top;
104 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_kernel_stext;
105 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_kernel_etext;
106 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_kernel_slide;
107 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_kernel_slid_base;
108 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_kernel_slid_top;
109
110 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_prelink_stext;
111 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_prelink_etext;
112 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_prelink_sdata;
113 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_prelink_edata;
114 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_prelink_sinfo;
115 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_prelink_einfo;
116 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_slinkedit;
117 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_elinkedit;
118
119 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_kernel_builtinkmod_text;
120 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_kernel_builtinkmod_text_end;
121
122 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_kernelcache_base;
123 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_kernelcache_top;
124
125 /* Used by <mach/arm/vm_param.h> */
126 SECURITY_READ_ONLY_LATE(unsigned long) gVirtBase;
127 SECURITY_READ_ONLY_LATE(unsigned long) gPhysBase;
128 SECURITY_READ_ONLY_LATE(unsigned long) gPhysSize;
129 SECURITY_READ_ONLY_LATE(unsigned long) gT0Sz = T0SZ_BOOT;
130 SECURITY_READ_ONLY_LATE(unsigned long) gT1Sz = T1SZ_BOOT;
131
132 /* 23543331 - step 1 of kext / kernel __TEXT and __DATA colocation is to move
133 * all kexts before the kernel. This is only for arm64 devices and looks
134 * something like the following:
135 * -- vmaddr order --
136 * 0xffffff8004004000 __PRELINK_TEXT
137 * 0xffffff8007004000 __TEXT (xnu)
138 * 0xffffff80075ec000 __DATA (xnu)
139 * 0xffffff80076dc000 __KLD (xnu)
140 * 0xffffff80076e0000 __LAST (xnu)
141 * 0xffffff80076e4000 __LINKEDIT (xnu)
142 * 0xffffff80076e4000 __PRELINK_DATA (not used yet)
143 * 0xffffff800782c000 __PRELINK_INFO
144 * 0xffffff80078e4000 -- End of kernelcache
145 */
146
147 /* 24921709 - make XNU ready for KTRR
148 *
149 * Two possible kernel cache layouts, depending on which kcgen is being used.
150 * VAs increasing downwards.
151 * Old KCGEN:
152 *
153 * __PRELINK_TEXT
154 * __TEXT
155 * __DATA_CONST
156 * __TEXT_EXEC
157 * __KLD
158 * __LAST
159 * __DATA
160 * __PRELINK_DATA (expected empty)
161 * __LINKEDIT
162 * __PRELINK_INFO
163 *
164 * New kcgen:
165 *
166 * __PRELINK_TEXT <--- First KTRR (ReadOnly) segment
167 * __PLK_DATA_CONST
168 * __PLK_TEXT_EXEC
169 * __TEXT
170 * __DATA_CONST
171 * __TEXT_EXEC
172 * __KLD
173 * __LAST <--- Last KTRR (ReadOnly) segment
174 * __DATA
175 * __BOOTDATA (if present)
176 * __LINKEDIT
177 * __PRELINK_DATA (expected populated now)
178 * __PLK_LINKEDIT
179 * __PRELINK_INFO
180 *
181 */
182
183 vm_offset_t mem_size; /* Size of actual physical memory present
184 * minus any performance buffer and possibly
185 * limited by mem_limit in bytes */
186 uint64_t mem_actual; /* The "One True" physical memory size
187 * actually, it's the highest physical
188 * address + 1 */
189 uint64_t max_mem; /* Size of physical memory (bytes), adjusted
190 * by maxmem */
191 uint64_t max_mem_actual; /* Actual size of physical memory (bytes),
192 * adjusted by the maxmem boot-arg */
193 uint64_t sane_size; /* Memory size to use for defaults
194 * calculations */
195 /* This no longer appears to be used; kill it? */
196 addr64_t vm_last_addr = VM_MAX_KERNEL_ADDRESS; /* Highest kernel
197 * virtual address known
198 * to the VM system */
199
200 SECURITY_READ_ONLY_LATE(vm_offset_t) segEXTRADATA;
201 SECURITY_READ_ONLY_LATE(unsigned long) segSizeEXTRADATA;
202
203 SECURITY_READ_ONLY_LATE(vm_offset_t) segLOWESTTEXT;
204 SECURITY_READ_ONLY_LATE(vm_offset_t) segLOWEST;
205 SECURITY_READ_ONLY_LATE(vm_offset_t) segLOWESTRO;
206 SECURITY_READ_ONLY_LATE(vm_offset_t) segHIGHESTRO;
207
208 /* Only set when booted from MH_FILESET kernel collections */
209 SECURITY_READ_ONLY_LATE(vm_offset_t) segLOWESTKC;
210 SECURITY_READ_ONLY_LATE(vm_offset_t) segHIGHESTKC;
211 SECURITY_READ_ONLY_LATE(vm_offset_t) segLOWESTROKC;
212 SECURITY_READ_ONLY_LATE(vm_offset_t) segHIGHESTROKC;
213 SECURITY_READ_ONLY_LATE(vm_offset_t) segLOWESTAuxKC;
214 SECURITY_READ_ONLY_LATE(vm_offset_t) segHIGHESTAuxKC;
215 SECURITY_READ_ONLY_LATE(vm_offset_t) segLOWESTROAuxKC;
216 SECURITY_READ_ONLY_LATE(vm_offset_t) segHIGHESTROAuxKC;
217 SECURITY_READ_ONLY_LATE(vm_offset_t) segLOWESTRXAuxKC;
218 SECURITY_READ_ONLY_LATE(vm_offset_t) segHIGHESTRXAuxKC;
219 SECURITY_READ_ONLY_LATE(vm_offset_t) segHIGHESTNLEAuxKC;
220
221 SECURITY_READ_ONLY_LATE(static vm_offset_t) segTEXTB;
222 SECURITY_READ_ONLY_LATE(static unsigned long) segSizeTEXT;
223
224 #if XNU_MONITOR
225 SECURITY_READ_ONLY_LATE(vm_offset_t) segPPLTEXTB;
226 SECURITY_READ_ONLY_LATE(unsigned long) segSizePPLTEXT;
227
228 SECURITY_READ_ONLY_LATE(vm_offset_t) segPPLTRAMPB;
229 SECURITY_READ_ONLY_LATE(unsigned long) segSizePPLTRAMP;
230
231 SECURITY_READ_ONLY_LATE(vm_offset_t) segPPLDATACONSTB;
232 SECURITY_READ_ONLY_LATE(unsigned long) segSizePPLDATACONST;
233 SECURITY_READ_ONLY_LATE(void *) pmap_stacks_start = NULL;
234 SECURITY_READ_ONLY_LATE(void *) pmap_stacks_end = NULL;
235 #endif
236
237 SECURITY_READ_ONLY_LATE(static vm_offset_t) segDATACONSTB;
238 SECURITY_READ_ONLY_LATE(static unsigned long) segSizeDATACONST;
239
240 SECURITY_READ_ONLY_LATE(vm_offset_t) segTEXTEXECB;
241 SECURITY_READ_ONLY_LATE(unsigned long) segSizeTEXTEXEC;
242
243 SECURITY_READ_ONLY_LATE(static vm_offset_t) segDATAB;
244 SECURITY_READ_ONLY_LATE(static unsigned long) segSizeDATA;
245
246 #if XNU_MONITOR
247 SECURITY_READ_ONLY_LATE(vm_offset_t) segPPLDATAB;
248 SECURITY_READ_ONLY_LATE(unsigned long) segSizePPLDATA;
249 #endif
250
251 SECURITY_READ_ONLY_LATE(vm_offset_t) segBOOTDATAB;
252 SECURITY_READ_ONLY_LATE(unsigned long) segSizeBOOTDATA;
253 extern vm_offset_t intstack_low_guard;
254 extern vm_offset_t intstack_high_guard;
255 extern vm_offset_t excepstack_high_guard;
256
257 SECURITY_READ_ONLY_LATE(vm_offset_t) segLINKB;
258 SECURITY_READ_ONLY_LATE(static unsigned long) segSizeLINK;
259
260 SECURITY_READ_ONLY_LATE(static vm_offset_t) segKLDB;
261 SECURITY_READ_ONLY_LATE(unsigned long) segSizeKLD;
262 SECURITY_READ_ONLY_LATE(static vm_offset_t) segKLDDATAB;
263 SECURITY_READ_ONLY_LATE(static unsigned long) segSizeKLDDATA;
264 SECURITY_READ_ONLY_LATE(vm_offset_t) segLASTB;
265 SECURITY_READ_ONLY_LATE(unsigned long) segSizeLAST;
266 SECURITY_READ_ONLY_LATE(vm_offset_t) segLASTDATACONSTB;
267 SECURITY_READ_ONLY_LATE(unsigned long) segSizeLASTDATACONST;
268
269 SECURITY_READ_ONLY_LATE(vm_offset_t) sectHIBTEXTB;
270 SECURITY_READ_ONLY_LATE(unsigned long) sectSizeHIBTEXT;
271 SECURITY_READ_ONLY_LATE(vm_offset_t) segHIBDATAB;
272 SECURITY_READ_ONLY_LATE(unsigned long) segSizeHIBDATA;
273 SECURITY_READ_ONLY_LATE(vm_offset_t) sectHIBDATACONSTB;
274 SECURITY_READ_ONLY_LATE(unsigned long) sectSizeHIBDATACONST;
275
276 SECURITY_READ_ONLY_LATE(vm_offset_t) segPRELINKTEXTB;
277 SECURITY_READ_ONLY_LATE(unsigned long) segSizePRELINKTEXT;
278
279 SECURITY_READ_ONLY_LATE(static vm_offset_t) segPLKTEXTEXECB;
280 SECURITY_READ_ONLY_LATE(static unsigned long) segSizePLKTEXTEXEC;
281
282 SECURITY_READ_ONLY_LATE(static vm_offset_t) segPLKDATACONSTB;
283 SECURITY_READ_ONLY_LATE(static unsigned long) segSizePLKDATACONST;
284
285 SECURITY_READ_ONLY_LATE(static vm_offset_t) segPRELINKDATAB;
286 SECURITY_READ_ONLY_LATE(static unsigned long) segSizePRELINKDATA;
287
288 SECURITY_READ_ONLY_LATE(static vm_offset_t) segPLKLLVMCOVB = 0;
289 SECURITY_READ_ONLY_LATE(static unsigned long) segSizePLKLLVMCOV = 0;
290
291 SECURITY_READ_ONLY_LATE(static vm_offset_t) segPLKLINKEDITB;
292 SECURITY_READ_ONLY_LATE(static unsigned long) segSizePLKLINKEDIT;
293
294 SECURITY_READ_ONLY_LATE(static vm_offset_t) segPRELINKINFOB;
295 SECURITY_READ_ONLY_LATE(static unsigned long) segSizePRELINKINFO;
296
297 /* Only set when booted from MH_FILESET primary kernel collection */
298 SECURITY_READ_ONLY_LATE(vm_offset_t) segKCTEXTEXECB;
299 SECURITY_READ_ONLY_LATE(unsigned long) segSizeKCTEXTEXEC;
300 SECURITY_READ_ONLY_LATE(static vm_offset_t) segKCDATACONSTB;
301 SECURITY_READ_ONLY_LATE(static unsigned long) segSizeKCDATACONST;
302 SECURITY_READ_ONLY_LATE(static vm_offset_t) segKCDATAB;
303 SECURITY_READ_ONLY_LATE(static unsigned long) segSizeKCDATA;
304
305 SECURITY_READ_ONLY_LATE(static boolean_t) use_contiguous_hint = TRUE;
306
307 SECURITY_READ_ONLY_LATE(int) PAGE_SHIFT_CONST;
308
309 SECURITY_READ_ONLY_LATE(vm_offset_t) end_kern;
310 SECURITY_READ_ONLY_LATE(vm_offset_t) etext;
311 SECURITY_READ_ONLY_LATE(vm_offset_t) sdata;
312 SECURITY_READ_ONLY_LATE(vm_offset_t) edata;
313
314 SECURITY_READ_ONLY_LATE(static vm_offset_t) auxkc_mh, auxkc_base, auxkc_right_above;
315
316 vm_offset_t alloc_ptpage(boolean_t map_static);
317 SECURITY_READ_ONLY_LATE(vm_offset_t) ropage_next;
318
319 /*
320 * Bootstrap the system enough to run with virtual memory.
321 * Map the kernel's code and data, and allocate the system page table.
322 * Page_size must already be set.
323 *
324 * Parameters:
325 * first_avail: first available physical page -
326 * after kernel page tables
327 * avail_start: PA of first physical page
328 * avail_end: PA of last physical page
329 */
330 SECURITY_READ_ONLY_LATE(vm_offset_t) first_avail;
331 SECURITY_READ_ONLY_LATE(vm_offset_t) static_memory_end;
332 SECURITY_READ_ONLY_LATE(pmap_paddr_t) avail_start;
333 SECURITY_READ_ONLY_LATE(pmap_paddr_t) avail_end;
334 SECURITY_READ_ONLY_LATE(pmap_paddr_t) real_avail_end;
335 SECURITY_READ_ONLY_LATE(unsigned long) real_phys_size;
336 SECURITY_READ_ONLY_LATE(vm_map_address_t) physmap_base = (vm_map_address_t)0;
337 SECURITY_READ_ONLY_LATE(vm_map_address_t) physmap_end = (vm_map_address_t)0;
338
339 #if __ARM_KERNEL_PROTECT__
340 extern void ExceptionVectorsBase;
341 extern void ExceptionVectorsEnd;
342 #endif /* __ARM_KERNEL_PROTECT__ */
343
344 typedef struct {
345 pmap_paddr_t pa;
346 vm_map_address_t va;
347 vm_size_t len;
348 } ptov_table_entry;
349
350 #define PTOV_TABLE_SIZE 8
351 SECURITY_READ_ONLY_LATE(static ptov_table_entry) ptov_table[PTOV_TABLE_SIZE];
352 SECURITY_READ_ONLY_LATE(static boolean_t) kva_active = FALSE;
353
354
355 vm_map_address_t
356 phystokv(pmap_paddr_t pa)
357 {
358 for (size_t i = 0; (i < PTOV_TABLE_SIZE) && (ptov_table[i].len != 0); i++) {
359 if ((pa >= ptov_table[i].pa) && (pa < (ptov_table[i].pa + ptov_table[i].len))) {
360 return pa - ptov_table[i].pa + ptov_table[i].va;
361 }
362 }
363 assertf((pa - gPhysBase) < real_phys_size, "%s: illegal PA: 0x%llx", __func__, (uint64_t)pa);
364 return pa - gPhysBase + gVirtBase;
365 }
366
367 vm_map_address_t
368 phystokv_range(pmap_paddr_t pa, vm_size_t *max_len)
369 {
370 vm_size_t len;
371 for (size_t i = 0; (i < PTOV_TABLE_SIZE) && (ptov_table[i].len != 0); i++) {
372 if ((pa >= ptov_table[i].pa) && (pa < (ptov_table[i].pa + ptov_table[i].len))) {
373 len = ptov_table[i].len - (pa - ptov_table[i].pa);
374 if (*max_len > len) {
375 *max_len = len;
376 }
377 return pa - ptov_table[i].pa + ptov_table[i].va;
378 }
379 }
380 len = PAGE_SIZE - (pa & PAGE_MASK);
381 if (*max_len > len) {
382 *max_len = len;
383 }
384 assertf((pa - gPhysBase) < real_phys_size, "%s: illegal PA: 0x%llx", __func__, (uint64_t)pa);
385 return pa - gPhysBase + gVirtBase;
386 }
387
388 vm_offset_t
389 ml_static_vtop(vm_offset_t va)
390 {
391 for (size_t i = 0; (i < PTOV_TABLE_SIZE) && (ptov_table[i].len != 0); i++) {
392 if ((va >= ptov_table[i].va) && (va < (ptov_table[i].va + ptov_table[i].len))) {
393 return va - ptov_table[i].va + ptov_table[i].pa;
394 }
395 }
396 assertf(((vm_address_t)(va) - gVirtBase) < gPhysSize, "%s: illegal VA: %p", __func__, (void*)va);
397 return (vm_address_t)(va) - gVirtBase + gPhysBase;
398 }
399
400 /*
401 * This rounds the given address up to the nearest boundary for a PTE contiguous
402 * hint.
403 */
404 static vm_offset_t
405 round_up_pte_hint_address(vm_offset_t address)
406 {
407 vm_offset_t hint_size = ARM_PTE_SIZE << ARM_PTE_HINT_ENTRIES_SHIFT;
408 return (address + (hint_size - 1)) & ~(hint_size - 1);
409 }
410
411 /* allocate a page for a page table: we support static and dynamic mappings.
412 *
413 * returns a virtual address for the allocated page
414 *
415 * for static mappings, we allocate from the region ropagetable_begin to ro_pagetable_end-1,
416 * which is defined in the DATA_CONST segment and will be protected RNX when vm_prot_finalize runs.
417 *
418 * for dynamic mappings, we allocate from avail_start, which should remain RWNX.
419 */
420
421 vm_offset_t
422 alloc_ptpage(boolean_t map_static)
423 {
424 vm_offset_t vaddr;
425
426 #if !(defined(KERNEL_INTEGRITY_KTRR) || defined(KERNEL_INTEGRITY_CTRR))
427 map_static = FALSE;
428 #endif
429
430 if (!ropage_next) {
431 ropage_next = (vm_offset_t)&ropagetable_begin;
432 }
433
434 if (map_static) {
435 assert(ropage_next < (vm_offset_t)&ropagetable_end);
436
437 vaddr = ropage_next;
438 ropage_next += ARM_PGBYTES;
439
440 return vaddr;
441 } else {
442 vaddr = phystokv(avail_start);
443 avail_start += ARM_PGBYTES;
444
445 return vaddr;
446 }
447 }
448
449 #if DEBUG
450
451 void dump_kva_l2(vm_offset_t tt_base, tt_entry_t *tt, int indent, uint64_t *rosz_out, uint64_t *rwsz_out);
452
453 void
454 dump_kva_l2(vm_offset_t tt_base, tt_entry_t *tt, int indent, uint64_t *rosz_out, uint64_t *rwsz_out)
455 {
456 unsigned int i;
457 boolean_t cur_ro, prev_ro = 0;
458 int start_entry = -1;
459 tt_entry_t cur, prev = 0;
460 pmap_paddr_t robegin = kvtophys((vm_offset_t)&ropagetable_begin);
461 pmap_paddr_t roend = kvtophys((vm_offset_t)&ropagetable_end);
462 boolean_t tt_static = kvtophys((vm_offset_t)tt) >= robegin &&
463 kvtophys((vm_offset_t)tt) < roend;
464
465 for (i = 0; i < TTE_PGENTRIES; i++) {
466 int tte_type = tt[i] & ARM_TTE_TYPE_MASK;
467 cur = tt[i] & ARM_TTE_TABLE_MASK;
468
469 if (tt_static) {
470 /* addresses mapped by this entry are static if it is a block mapping,
471 * or the table was allocated from the RO page table region */
472 cur_ro = (tte_type == ARM_TTE_TYPE_BLOCK) || (cur >= robegin && cur < roend);
473 } else {
474 cur_ro = 0;
475 }
476
477 if ((cur == 0 && prev != 0) || (cur_ro != prev_ro && prev != 0)) { // falling edge
478 uintptr_t start, end, sz;
479
480 start = (uintptr_t)start_entry << ARM_TT_L2_SHIFT;
481 start += tt_base;
482 end = ((uintptr_t)i << ARM_TT_L2_SHIFT) - 1;
483 end += tt_base;
484
485 sz = end - start + 1;
486 printf("%*s0x%08x_%08x-0x%08x_%08x %s (%luMB)\n",
487 indent * 4, "",
488 (uint32_t)(start >> 32), (uint32_t)start,
489 (uint32_t)(end >> 32), (uint32_t)end,
490 prev_ro ? "Static " : "Dynamic",
491 (sz >> 20));
492
493 if (prev_ro) {
494 *rosz_out += sz;
495 } else {
496 *rwsz_out += sz;
497 }
498 }
499
500 if ((prev == 0 && cur != 0) || cur_ro != prev_ro) { // rising edge: set start
501 start_entry = i;
502 }
503
504 prev = cur;
505 prev_ro = cur_ro;
506 }
507 }
508
509 void
510 dump_kva_space()
511 {
512 uint64_t tot_rosz = 0, tot_rwsz = 0;
513 int ro_ptpages, rw_ptpages;
514 pmap_paddr_t robegin = kvtophys((vm_offset_t)&ropagetable_begin);
515 pmap_paddr_t roend = kvtophys((vm_offset_t)&ropagetable_end);
516 boolean_t root_static = kvtophys((vm_offset_t)cpu_tte) >= robegin &&
517 kvtophys((vm_offset_t)cpu_tte) < roend;
518 uint64_t kva_base = ~((1ULL << (64 - T1SZ_BOOT)) - 1);
519
520 printf("Root page table: %s\n", root_static ? "Static" : "Dynamic");
521
522 for (unsigned int i = 0; i < TTE_PGENTRIES; i++) {
523 pmap_paddr_t cur;
524 boolean_t cur_ro;
525 uintptr_t start, end;
526 uint64_t rosz = 0, rwsz = 0;
527
528 if ((cpu_tte[i] & ARM_TTE_VALID) == 0) {
529 continue;
530 }
531
532 cur = cpu_tte[i] & ARM_TTE_TABLE_MASK;
533 start = (uint64_t)i << ARM_TT_L1_SHIFT;
534 start = start + kva_base;
535 end = start + (ARM_TT_L1_SIZE - 1);
536 cur_ro = cur >= robegin && cur < roend;
537
538 printf("0x%08x_%08x-0x%08x_%08x %s\n",
539 (uint32_t)(start >> 32), (uint32_t)start,
540 (uint32_t)(end >> 32), (uint32_t)end,
541 cur_ro ? "Static " : "Dynamic");
542
543 dump_kva_l2(start, (tt_entry_t*)phystokv(cur), 1, &rosz, &rwsz);
544 tot_rosz += rosz;
545 tot_rwsz += rwsz;
546 }
547
548 printf("L2 Address space mapped: Static %lluMB Dynamic %lluMB Total %lluMB\n",
549 tot_rosz >> 20,
550 tot_rwsz >> 20,
551 (tot_rosz >> 20) + (tot_rwsz >> 20));
552
553 ro_ptpages = (int)((ropage_next - (vm_offset_t)&ropagetable_begin) >> ARM_PGSHIFT);
554 rw_ptpages = (int)(lowGlo.lgStaticSize >> ARM_PGSHIFT);
555 printf("Pages used: static %d dynamic %d\n", ro_ptpages, rw_ptpages);
556 }
557
558 #endif /* DEBUG */
559
560 #if __ARM_KERNEL_PROTECT__ || XNU_MONITOR
561 /*
562 * arm_vm_map:
563 * root_ttp: The kernel virtual address for the root of the target page tables
564 * vaddr: The target virtual address
565 * pte: A page table entry value (may be ARM_PTE_EMPTY)
566 *
567 * This function installs pte at vaddr in root_ttp. Any page table pages needed
568 * to install pte will be allocated by this function.
569 */
570 static void
571 arm_vm_map(tt_entry_t * root_ttp, vm_offset_t vaddr, pt_entry_t pte)
572 {
573 vm_offset_t ptpage = 0;
574 tt_entry_t * ttp = root_ttp;
575
576 tt_entry_t * l1_ttep = NULL;
577 tt_entry_t l1_tte = 0;
578
579 tt_entry_t * l2_ttep = NULL;
580 tt_entry_t l2_tte = 0;
581 pt_entry_t * ptep = NULL;
582 pt_entry_t cpte = 0;
583
584 /*
585 * Walk the target page table to find the PTE for the given virtual
586 * address. Allocate any page table pages needed to do this.
587 */
588 l1_ttep = ttp + ((vaddr & ARM_TT_L1_INDEX_MASK) >> ARM_TT_L1_SHIFT);
589 l1_tte = *l1_ttep;
590
591 if (l1_tte == ARM_TTE_EMPTY) {
592 ptpage = alloc_ptpage(TRUE);
593 bzero((void *)ptpage, ARM_PGBYTES);
594 l1_tte = kvtophys(ptpage);
595 l1_tte &= ARM_TTE_TABLE_MASK;
596 l1_tte |= ARM_TTE_VALID | ARM_TTE_TYPE_TABLE;
597 *l1_ttep = l1_tte;
598 ptpage = 0;
599 }
600
601 ttp = (tt_entry_t *)phystokv(l1_tte & ARM_TTE_TABLE_MASK);
602
603 l2_ttep = ttp + ((vaddr & ARM_TT_L2_INDEX_MASK) >> ARM_TT_L2_SHIFT);
604 l2_tte = *l2_ttep;
605
606 if (l2_tte == ARM_TTE_EMPTY) {
607 ptpage = alloc_ptpage(TRUE);
608 bzero((void *)ptpage, ARM_PGBYTES);
609 l2_tte = kvtophys(ptpage);
610 l2_tte &= ARM_TTE_TABLE_MASK;
611 l2_tte |= ARM_TTE_VALID | ARM_TTE_TYPE_TABLE;
612 *l2_ttep = l2_tte;
613 ptpage = 0;
614 }
615
616 ttp = (tt_entry_t *)phystokv(l2_tte & ARM_TTE_TABLE_MASK);
617
618 ptep = ttp + ((vaddr & ARM_TT_L3_INDEX_MASK) >> ARM_TT_L3_SHIFT);
619 cpte = *ptep;
620
621 /*
622 * If the existing PTE is not empty, then we are replacing a valid
623 * mapping.
624 */
625 if (cpte != ARM_PTE_EMPTY) {
626 panic("%s: cpte=%#llx is not empty, "
627 "vaddr=%#lx, pte=%#llx",
628 __FUNCTION__, cpte,
629 vaddr, pte);
630 }
631
632 *ptep = pte;
633 }
634
635 #endif // __ARM_KERNEL_PROTECT || XNU_MONITOR
636
637 #if __ARM_KERNEL_PROTECT__
638
639 /*
640 * arm_vm_kernel_el0_map:
641 * vaddr: The target virtual address
642 * pte: A page table entry value (may be ARM_PTE_EMPTY)
643 *
644 * This function installs pte at vaddr for the EL0 kernel mappings.
645 */
646 static void
647 arm_vm_kernel_el0_map(vm_offset_t vaddr, pt_entry_t pte)
648 {
649 /* Calculate where vaddr will be in the EL1 kernel page tables. */
650 vm_offset_t kernel_pmap_vaddr = vaddr - ((ARM_TT_ROOT_INDEX_MASK + ARM_TT_ROOT_SIZE) / 2ULL);
651 arm_vm_map(cpu_tte, kernel_pmap_vaddr, pte);
652 }
653
654 /*
655 * arm_vm_kernel_el1_map:
656 * vaddr: The target virtual address
657 * pte: A page table entry value (may be ARM_PTE_EMPTY)
658 *
659 * This function installs pte at vaddr for the EL1 kernel mappings.
660 */
661 static void
662 arm_vm_kernel_el1_map(vm_offset_t vaddr, pt_entry_t pte)
663 {
664 arm_vm_map(cpu_tte, vaddr, pte);
665 }
666
667 /*
668 * arm_vm_kernel_pte:
669 * vaddr: The target virtual address
670 *
671 * This function returns the PTE value for the given vaddr from the kernel page
672 * tables. If the region has been been block mapped, we return what an
673 * equivalent PTE value would be (as regards permissions and flags). We also
674 * remove the HINT bit (as we are not necessarily creating contiguous mappings.
675 */
676 static pt_entry_t
677 arm_vm_kernel_pte(vm_offset_t vaddr)
678 {
679 tt_entry_t * ttp = cpu_tte;
680 tt_entry_t * ttep = NULL;
681 tt_entry_t tte = 0;
682 pt_entry_t * ptep = NULL;
683 pt_entry_t pte = 0;
684
685 ttep = ttp + ((vaddr & ARM_TT_L1_INDEX_MASK) >> ARM_TT_L1_SHIFT);
686 tte = *ttep;
687
688 assert(tte & ARM_TTE_VALID);
689
690 if ((tte & ARM_TTE_TYPE_MASK) == ARM_TTE_TYPE_BLOCK) {
691 /* This is a block mapping; return the equivalent PTE value. */
692 pte = (pt_entry_t)(tte & ~ARM_TTE_TYPE_MASK);
693 pte |= ARM_PTE_TYPE_VALID;
694 pte |= vaddr & ((ARM_TT_L1_SIZE - 1) & ARM_PTE_PAGE_MASK);
695 pte &= ~ARM_PTE_HINT_MASK;
696 return pte;
697 }
698
699 ttp = (tt_entry_t *)phystokv(tte & ARM_TTE_TABLE_MASK);
700 ttep = ttp + ((vaddr & ARM_TT_L2_INDEX_MASK) >> ARM_TT_L2_SHIFT);
701 tte = *ttep;
702
703 assert(tte & ARM_TTE_VALID);
704
705 if ((tte & ARM_TTE_TYPE_MASK) == ARM_TTE_TYPE_BLOCK) {
706 /* This is a block mapping; return the equivalent PTE value. */
707 pte = (pt_entry_t)(tte & ~ARM_TTE_TYPE_MASK);
708 pte |= ARM_PTE_TYPE_VALID;
709 pte |= vaddr & ((ARM_TT_L2_SIZE - 1) & ARM_PTE_PAGE_MASK);
710 pte &= ~ARM_PTE_HINT_MASK;
711 return pte;
712 }
713
714 ttp = (tt_entry_t *)phystokv(tte & ARM_TTE_TABLE_MASK);
715
716 ptep = ttp + ((vaddr & ARM_TT_L3_INDEX_MASK) >> ARM_TT_L3_SHIFT);
717 pte = *ptep;
718 pte &= ~ARM_PTE_HINT_MASK;
719 return pte;
720 }
721
722 /*
723 * arm_vm_prepare_kernel_el0_mappings:
724 * alloc_only: Indicates if PTE values should be copied from the EL1 kernel
725 * mappings.
726 *
727 * This function expands the kernel page tables to support the EL0 kernel
728 * mappings, and conditionally installs the PTE values for the EL0 kernel
729 * mappings (if alloc_only is false).
730 */
731 static void
732 arm_vm_prepare_kernel_el0_mappings(bool alloc_only)
733 {
734 pt_entry_t pte = 0;
735 vm_offset_t start = ((vm_offset_t)&ExceptionVectorsBase) & ~PAGE_MASK;
736 vm_offset_t end = (((vm_offset_t)&ExceptionVectorsEnd) + PAGE_MASK) & ~PAGE_MASK;
737 vm_offset_t cur = 0;
738 vm_offset_t cur_fixed = 0;
739
740 /* Expand for/map the exceptions vectors in the EL0 kernel mappings. */
741 for (cur = start, cur_fixed = ARM_KERNEL_PROTECT_EXCEPTION_START; cur < end; cur += ARM_PGBYTES, cur_fixed += ARM_PGBYTES) {
742 /*
743 * We map the exception vectors at a different address than that
744 * of the kernelcache to avoid sharing page table pages with the
745 * kernelcache (as this may cause issues with TLB caching of
746 * page table pages.
747 */
748 if (!alloc_only) {
749 pte = arm_vm_kernel_pte(cur);
750 }
751
752 arm_vm_kernel_el1_map(cur_fixed, pte);
753 arm_vm_kernel_el0_map(cur_fixed, pte);
754 }
755
756 __builtin_arm_dmb(DMB_ISH);
757 __builtin_arm_isb(ISB_SY);
758
759 if (!alloc_only) {
760 /*
761 * If we have created the alternate exception vector mappings,
762 * the boot CPU may now switch over to them.
763 */
764 set_vbar_el1(ARM_KERNEL_PROTECT_EXCEPTION_START);
765 __builtin_arm_isb(ISB_SY);
766 }
767 }
768
769 /*
770 * arm_vm_populate_kernel_el0_mappings:
771 *
772 * This function adds all required mappings to the EL0 kernel mappings.
773 */
774 static void
775 arm_vm_populate_kernel_el0_mappings(void)
776 {
777 arm_vm_prepare_kernel_el0_mappings(FALSE);
778 }
779
780 /*
781 * arm_vm_expand_kernel_el0_mappings:
782 *
783 * This function expands the kernel page tables to accomodate the EL0 kernel
784 * mappings.
785 */
786 static void
787 arm_vm_expand_kernel_el0_mappings(void)
788 {
789 arm_vm_prepare_kernel_el0_mappings(TRUE);
790 }
791 #endif /* __ARM_KERNEL_PROTECT__ */
792
793 #if defined(KERNEL_INTEGRITY_KTRR) || defined(KERNEL_INTEGRITY_CTRR)
794 extern void bootstrap_instructions;
795
796 /*
797 * arm_replace_identity_map takes the V=P map that we construct in start.s
798 * and repurposes it in order to have it map only the page we need in order
799 * to turn on the MMU. This prevents us from running into issues where
800 * KTRR will cause us to fault on executable block mappings that cross the
801 * KTRR boundary.
802 */
803 static void
804 arm_replace_identity_map(void)
805 {
806 vm_offset_t addr;
807 pmap_paddr_t paddr;
808
809 pmap_paddr_t l1_ptp_phys = 0;
810 tt_entry_t *l1_ptp_virt = NULL;
811 tt_entry_t *tte1 = NULL;
812 pmap_paddr_t l2_ptp_phys = 0;
813 tt_entry_t *l2_ptp_virt = NULL;
814 tt_entry_t *tte2 = NULL;
815 pmap_paddr_t l3_ptp_phys = 0;
816 pt_entry_t *l3_ptp_virt = NULL;
817 pt_entry_t *ptep = NULL;
818
819 addr = ((vm_offset_t)&bootstrap_instructions) & ~ARM_PGMASK;
820 paddr = kvtophys(addr);
821
822 /*
823 * Grab references to the V=P page tables, and allocate an L3 page.
824 */
825 l1_ptp_phys = kvtophys((vm_offset_t)&bootstrap_pagetables);
826 l1_ptp_virt = (tt_entry_t *)phystokv(l1_ptp_phys);
827 tte1 = &l1_ptp_virt[L1_TABLE_INDEX(paddr)];
828
829 l2_ptp_virt = L2_TABLE_VA(tte1);
830 l2_ptp_phys = (*tte1) & ARM_TTE_TABLE_MASK;
831 tte2 = &l2_ptp_virt[L2_TABLE_INDEX(paddr)];
832
833 l3_ptp_virt = (pt_entry_t *)alloc_ptpage(TRUE);
834 l3_ptp_phys = kvtophys((vm_offset_t)l3_ptp_virt);
835 ptep = &l3_ptp_virt[L3_TABLE_INDEX(paddr)];
836
837 /*
838 * Replace the large V=P mapping with a mapping that provides only the
839 * mappings needed to turn on the MMU.
840 */
841
842 bzero(l1_ptp_virt, ARM_PGBYTES);
843 *tte1 = ARM_TTE_BOOT_TABLE | (l2_ptp_phys & ARM_TTE_TABLE_MASK);
844
845 bzero(l2_ptp_virt, ARM_PGBYTES);
846 *tte2 = ARM_TTE_BOOT_TABLE | (l3_ptp_phys & ARM_TTE_TABLE_MASK);
847
848 *ptep = (paddr & ARM_PTE_MASK) |
849 ARM_PTE_TYPE_VALID |
850 ARM_PTE_SH(SH_OUTER_MEMORY) |
851 ARM_PTE_ATTRINDX(CACHE_ATTRINDX_WRITEBACK) |
852 ARM_PTE_AF |
853 ARM_PTE_AP(AP_RONA) |
854 ARM_PTE_NX;
855 }
856 #endif /* defined(KERNEL_INTEGRITY_KTRR) || defined(KERNEL_INTEGRITY_CTRR) */
857
858 tt_entry_t *arm_kva_to_tte(vm_offset_t);
859
860 tt_entry_t *
861 arm_kva_to_tte(vm_offset_t va)
862 {
863 tt_entry_t *tte1, *tte2;
864 tte1 = cpu_tte + L1_TABLE_INDEX(va);
865 tte2 = L2_TABLE_VA(tte1) + L2_TABLE_INDEX(va);
866
867 return tte2;
868 }
869
870 #if XNU_MONITOR
871
872 static inline pt_entry_t *
873 arm_kva_to_pte(vm_offset_t va)
874 {
875 tt_entry_t *tte2 = arm_kva_to_tte(va);
876 return L3_TABLE_VA(tte2) + L3_TABLE_INDEX(va);
877 }
878
879 #endif
880
881 #define ARM64_GRANULE_ALLOW_BLOCK (1 << 0)
882 #define ARM64_GRANULE_ALLOW_HINT (1 << 1)
883
884 /*
885 * arm_vm_page_granular_helper updates protections at the L3 level. It will (if
886 * neccessary) allocate a page for the L3 table and update the corresponding L2
887 * entry. Then, it will iterate over the L3 table, updating protections as necessary.
888 * This expects to be invoked on a L2 entry or sub L2 entry granularity, so this should
889 * not be invoked from a context that does not do L2 iteration separately (basically,
890 * don't call this except from arm_vm_page_granular_prot).
891 *
892 * unsigned granule: 0 => force to page granule, or a combination of
893 * ARM64_GRANULE_* flags declared above.
894 */
895
896 static void
897 arm_vm_page_granular_helper(vm_offset_t start, vm_offset_t _end, vm_offset_t va, pmap_paddr_t pa_offset,
898 int pte_prot_APX, int pte_prot_XN, unsigned granule,
899 pt_entry_t **deferred_pte, pt_entry_t *deferred_ptmp)
900 {
901 if (va & ARM_TT_L2_OFFMASK) { /* ragged edge hanging over a ARM_TT_L2_SIZE boundary */
902 tt_entry_t *tte2;
903 tt_entry_t tmplate;
904 pmap_paddr_t pa;
905 pt_entry_t *ppte, *recursive_pte = NULL, ptmp, recursive_ptmp = 0;
906 addr64_t ppte_phys;
907 unsigned i;
908
909 va &= ~ARM_TT_L2_OFFMASK;
910 pa = va - gVirtBase + gPhysBase - pa_offset;
911
912 if (pa >= real_avail_end) {
913 return;
914 }
915
916 tte2 = arm_kva_to_tte(va);
917
918 assert(_end >= va);
919 tmplate = *tte2;
920
921 if (ARM_TTE_TYPE_TABLE == (tmplate & ARM_TTE_TYPE_MASK)) {
922 /* pick up the existing page table. */
923 ppte = (pt_entry_t *)phystokv((tmplate & ARM_TTE_TABLE_MASK));
924 } else {
925 // TTE must be reincarnated with page level mappings.
926
927 // ... but we don't want to break up blocks on live
928 // translation tables.
929 assert(!kva_active);
930
931 ppte = (pt_entry_t*)alloc_ptpage(pa_offset == 0);
932 bzero(ppte, ARM_PGBYTES);
933 ppte_phys = kvtophys((vm_offset_t)ppte);
934
935 *tte2 = pa_to_tte(ppte_phys) | ARM_TTE_TYPE_TABLE | ARM_TTE_VALID;
936 }
937
938 vm_offset_t len = _end - va;
939 if ((pa + len) > real_avail_end) {
940 _end -= (pa + len - real_avail_end);
941 }
942 assert((start - gVirtBase + gPhysBase - pa_offset) >= gPhysBase);
943
944 /* Round up to the nearest PAGE_SIZE boundary when creating mappings:
945 * PAGE_SIZE may be a multiple of ARM_PGBYTES, and we don't want to leave
946 * a ragged non-PAGE_SIZE-aligned edge. */
947 vm_offset_t rounded_end = round_page(_end);
948 /* Apply the desired protections to the specified page range */
949 for (i = 0; i <= (ARM_TT_L3_INDEX_MASK >> ARM_TT_L3_SHIFT); i++) {
950 if ((start <= va) && (va < rounded_end)) {
951 ptmp = pa | ARM_PTE_AF | ARM_PTE_SH(SH_OUTER_MEMORY) | ARM_PTE_TYPE;
952 ptmp = ptmp | ARM_PTE_ATTRINDX(CACHE_ATTRINDX_DEFAULT);
953 ptmp = ptmp | ARM_PTE_AP(pte_prot_APX);
954 ptmp = ptmp | ARM_PTE_NX;
955 #if __ARM_KERNEL_PROTECT__
956 ptmp = ptmp | ARM_PTE_NG;
957 #endif /* __ARM_KERNEL_PROTECT__ */
958
959 if (pte_prot_XN) {
960 ptmp = ptmp | ARM_PTE_PNX;
961 }
962
963 /*
964 * If we can, apply the contiguous hint to this range. The hint is
965 * applicable if the current address falls within a hint-sized range that will
966 * be fully covered by this mapping request.
967 */
968 if ((va >= round_up_pte_hint_address(start)) && (round_up_pte_hint_address(va + 1) <= _end) &&
969 (granule & ARM64_GRANULE_ALLOW_HINT) && use_contiguous_hint) {
970 assert((va & ((1 << ARM_PTE_HINT_ADDR_SHIFT) - 1)) == ((pa & ((1 << ARM_PTE_HINT_ADDR_SHIFT) - 1))));
971 ptmp |= ARM_PTE_HINT;
972 /* Do not attempt to reapply the hint bit to an already-active mapping.
973 * This very likely means we're attempting to change attributes on an already-active mapping,
974 * which violates the requirement of the hint bit.*/
975 assert(!kva_active || (ppte[i] == ARM_PTE_TYPE_FAULT));
976 }
977 /*
978 * Do not change the contiguous bit on an active mapping. Even in a single-threaded
979 * environment, it's possible for prefetch to produce a TLB conflict by trying to pull in
980 * a hint-sized entry on top of one or more existing page-sized entries. It's also useful
981 * to make sure we're not trying to unhint a sub-range of a larger hinted range, which
982 * could produce a later TLB conflict.
983 */
984 assert(!kva_active || (ppte[i] == ARM_PTE_TYPE_FAULT) || ((ppte[i] & ARM_PTE_HINT) == (ptmp & ARM_PTE_HINT)));
985
986 /*
987 * If we reach an entry that maps the current pte page, delay updating it until the very end.
988 * Otherwise we might end up making the PTE page read-only, leading to a fault later on in
989 * this function if we manage to outrun the TLB. This can happen on KTRR-enabled devices when
990 * marking segDATACONST read-only. Mappings for this region may straddle a PT page boundary,
991 * so we must also defer assignment of the following PTE. We will assume that if the region
992 * were to require one or more full L3 pages, it would instead use L2 blocks where possible,
993 * therefore only requiring at most one L3 page at the beginning and one at the end.
994 */
995 if (kva_active && ((pt_entry_t*)(phystokv(pa)) == ppte)) {
996 assert(recursive_pte == NULL);
997 assert(granule & ARM64_GRANULE_ALLOW_BLOCK);
998 recursive_pte = &ppte[i];
999 recursive_ptmp = ptmp;
1000 } else if ((deferred_pte != NULL) && (&ppte[i] == &recursive_pte[1])) {
1001 assert(*deferred_pte == NULL);
1002 assert(deferred_ptmp != NULL);
1003 *deferred_pte = &ppte[i];
1004 *deferred_ptmp = ptmp;
1005 } else {
1006 ppte[i] = ptmp;
1007 }
1008 }
1009
1010 va += ARM_PGBYTES;
1011 pa += ARM_PGBYTES;
1012 }
1013 if (recursive_pte != NULL) {
1014 *recursive_pte = recursive_ptmp;
1015 }
1016 }
1017 }
1018
1019 /*
1020 * arm_vm_page_granular_prot updates protections by iterating over the L2 entries and
1021 * changing them. If a particular chunk necessitates L3 entries (for reasons of
1022 * alignment or length, or an explicit request that the entry be fully expanded), we
1023 * hand off to arm_vm_page_granular_helper to deal with the L3 chunk of the logic.
1024 */
1025 static void
1026 arm_vm_page_granular_prot(vm_offset_t start, unsigned long size, pmap_paddr_t pa_offset,
1027 int tte_prot_XN, int pte_prot_APX, int pte_prot_XN,
1028 unsigned granule)
1029 {
1030 pt_entry_t *deferred_pte = NULL, deferred_ptmp = 0;
1031 vm_offset_t _end = start + size;
1032 vm_offset_t align_start = (start + ARM_TT_L2_OFFMASK) & ~ARM_TT_L2_OFFMASK;
1033
1034 if (size == 0x0UL) {
1035 return;
1036 }
1037
1038 if (align_start > _end) {
1039 arm_vm_page_granular_helper(start, _end, start, pa_offset, pte_prot_APX, pte_prot_XN, granule, NULL, NULL);
1040 return;
1041 }
1042
1043 arm_vm_page_granular_helper(start, align_start, start, pa_offset, pte_prot_APX, pte_prot_XN, granule, &deferred_pte, &deferred_ptmp);
1044
1045 while ((_end - align_start) >= ARM_TT_L2_SIZE) {
1046 if (!(granule & ARM64_GRANULE_ALLOW_BLOCK)) {
1047 arm_vm_page_granular_helper(align_start, align_start + ARM_TT_L2_SIZE, align_start + 1, pa_offset,
1048 pte_prot_APX, pte_prot_XN, granule, NULL, NULL);
1049 } else {
1050 pmap_paddr_t pa = align_start - gVirtBase + gPhysBase - pa_offset;
1051 assert((pa & ARM_TT_L2_OFFMASK) == 0);
1052 tt_entry_t *tte2;
1053 tt_entry_t tmplate;
1054
1055 tte2 = arm_kva_to_tte(align_start);
1056
1057 if ((pa >= gPhysBase) && (pa < real_avail_end)) {
1058 tmplate = (pa & ARM_TTE_BLOCK_L2_MASK) | ARM_TTE_TYPE_BLOCK
1059 | ARM_TTE_VALID | ARM_TTE_BLOCK_AF | ARM_TTE_BLOCK_NX
1060 | ARM_TTE_BLOCK_AP(pte_prot_APX) | ARM_TTE_BLOCK_SH(SH_OUTER_MEMORY)
1061 | ARM_TTE_BLOCK_ATTRINDX(CACHE_ATTRINDX_WRITEBACK);
1062
1063 #if __ARM_KERNEL_PROTECT__
1064 tmplate = tmplate | ARM_TTE_BLOCK_NG;
1065 #endif /* __ARM_KERNEL_PROTECT__ */
1066 if (tte_prot_XN) {
1067 tmplate = tmplate | ARM_TTE_BLOCK_PNX;
1068 }
1069
1070 *tte2 = tmplate;
1071 }
1072 }
1073 align_start += ARM_TT_L2_SIZE;
1074 }
1075
1076 if (align_start < _end) {
1077 arm_vm_page_granular_helper(align_start, _end, _end, pa_offset, pte_prot_APX, pte_prot_XN, granule, &deferred_pte, &deferred_ptmp);
1078 }
1079
1080 if (deferred_pte != NULL) {
1081 *deferred_pte = deferred_ptmp;
1082 }
1083 }
1084
1085 static inline void
1086 arm_vm_page_granular_RNX(vm_offset_t start, unsigned long size, unsigned granule)
1087 {
1088 arm_vm_page_granular_prot(start, size, 0, 1, AP_RONA, 1, granule);
1089 }
1090
1091 static inline void
1092 arm_vm_page_granular_ROX(vm_offset_t start, unsigned long size, unsigned granule)
1093 {
1094 arm_vm_page_granular_prot(start, size, 0, 0, AP_RONA, 0, granule);
1095 }
1096
1097 static inline void
1098 arm_vm_page_granular_RWNX(vm_offset_t start, unsigned long size, unsigned granule)
1099 {
1100 arm_vm_page_granular_prot(start, size, 0, 1, AP_RWNA, 1, granule);
1101 }
1102
1103 /* used in the chosen/memory-map node, populated by iBoot. */
1104 typedef struct MemoryMapFileInfo {
1105 vm_offset_t paddr;
1106 size_t length;
1107 } MemoryMapFileInfo;
1108
1109 // Populate seg...AuxKC and fixup AuxKC permissions
1110 static bool
1111 arm_vm_auxkc_init(void)
1112 {
1113 if (auxkc_mh == 0 || auxkc_base == 0) {
1114 return false; // no auxKC.
1115 }
1116
1117 /* Fixup AuxKC and populate seg*AuxKC globals used below */
1118 arm_auxkc_init((void*)auxkc_mh, (void*)auxkc_base);
1119
1120 if (segLOWESTAuxKC != segLOWEST) {
1121 panic("segLOWESTAuxKC (%p) not equal to segLOWEST (%p). auxkc_mh: %p, auxkc_base: %p",
1122 (void*)segLOWESTAuxKC, (void*)segLOWEST,
1123 (void*)auxkc_mh, (void*)auxkc_base);
1124 }
1125
1126 /*
1127 * The AuxKC LINKEDIT segment needs to be covered by the RO region but is excluded
1128 * from the RO address range returned by kernel_collection_adjust_mh_addrs().
1129 * Ensure the highest non-LINKEDIT address in the AuxKC is the current end of
1130 * its RO region before extending it.
1131 */
1132 assert(segHIGHESTROAuxKC == segHIGHESTNLEAuxKC);
1133 assert(segHIGHESTAuxKC >= segHIGHESTROAuxKC);
1134 if (segHIGHESTAuxKC > segHIGHESTROAuxKC) {
1135 segHIGHESTROAuxKC = segHIGHESTAuxKC;
1136 }
1137
1138 /*
1139 * The AuxKC RO region must be right below the device tree/trustcache so that it can be covered
1140 * by CTRR, and the AuxKC RX region must be within the RO region.
1141 */
1142 assert(segHIGHESTROAuxKC == auxkc_right_above);
1143 assert(segHIGHESTRXAuxKC <= segHIGHESTROAuxKC);
1144 assert(segLOWESTRXAuxKC <= segHIGHESTRXAuxKC);
1145 assert(segLOWESTROAuxKC <= segLOWESTRXAuxKC);
1146 assert(segLOWESTAuxKC <= segLOWESTROAuxKC);
1147
1148 if (segHIGHESTRXAuxKC < segLOWEST) {
1149 arm_vm_page_granular_RNX(segHIGHESTRXAuxKC, segLOWEST - segHIGHESTRXAuxKC, 0);
1150 }
1151 if (segLOWESTRXAuxKC < segHIGHESTRXAuxKC) {
1152 arm_vm_page_granular_ROX(segLOWESTRXAuxKC, segHIGHESTRXAuxKC - segLOWESTRXAuxKC, 0); // Refined in OSKext::readPrelinkedExtensions
1153 }
1154 if (segLOWESTROAuxKC < segLOWESTRXAuxKC) {
1155 arm_vm_page_granular_RNX(segLOWESTROAuxKC, segLOWESTRXAuxKC - segLOWESTROAuxKC, 0);
1156 }
1157 if (segLOWESTAuxKC < segLOWESTROAuxKC) {
1158 arm_vm_page_granular_RWNX(segLOWESTAuxKC, segLOWESTROAuxKC - segLOWESTAuxKC, 0);
1159 }
1160
1161 return true;
1162 }
1163
1164 void
1165 arm_vm_prot_init(__unused boot_args * args)
1166 {
1167 segLOWESTTEXT = UINT64_MAX;
1168 if (segSizePRELINKTEXT && (segPRELINKTEXTB < segLOWESTTEXT)) {
1169 segLOWESTTEXT = segPRELINKTEXTB;
1170 }
1171 assert(segSizeTEXT);
1172 if (segTEXTB < segLOWESTTEXT) {
1173 segLOWESTTEXT = segTEXTB;
1174 }
1175 assert(segLOWESTTEXT < UINT64_MAX);
1176
1177 segEXTRADATA = segLOWESTTEXT;
1178 segSizeEXTRADATA = 0;
1179
1180 segLOWEST = segLOWESTTEXT;
1181 segLOWESTRO = segLOWESTTEXT;
1182
1183 if (segLOWESTKC && segLOWESTKC < segLOWEST) {
1184 /*
1185 * kernel collections have segments below the kernel. In particular the collection mach header
1186 * is below PRELINK_TEXT and is not covered by any other segments already tracked.
1187 */
1188 arm_vm_page_granular_RNX(segLOWESTKC, segLOWEST - segLOWESTKC, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT);
1189 segLOWEST = segLOWESTKC;
1190 if (segLOWESTROKC && segLOWESTROKC < segLOWESTRO) {
1191 segLOWESTRO = segLOWESTROKC;
1192 }
1193 if (segHIGHESTROKC && segHIGHESTROKC > segHIGHESTRO) {
1194 segHIGHESTRO = segHIGHESTROKC;
1195 }
1196 }
1197
1198 DTEntry memory_map;
1199 MemoryMapFileInfo const *trustCacheRange;
1200 unsigned int trustCacheRangeSize;
1201 int err;
1202
1203 if (SecureDTIsLockedDown()) {
1204 segEXTRADATA = (vm_offset_t)PE_state.deviceTreeHead;
1205 segSizeEXTRADATA = PE_state.deviceTreeSize;
1206 }
1207
1208 err = SecureDTLookupEntry(NULL, "chosen/memory-map", &memory_map);
1209 assert(err == kSuccess);
1210
1211 err = SecureDTGetProperty(memory_map, "TrustCache", (void const **)&trustCacheRange, &trustCacheRangeSize);
1212 if (err == kSuccess) {
1213 assert(trustCacheRangeSize == sizeof(MemoryMapFileInfo));
1214
1215 if (segSizeEXTRADATA == 0) {
1216 segEXTRADATA = phystokv(trustCacheRange->paddr);
1217 segSizeEXTRADATA = trustCacheRange->length;
1218 } else {
1219 segSizeEXTRADATA += trustCacheRange->length;
1220 }
1221 }
1222
1223 if (segSizeEXTRADATA != 0) {
1224 if (segEXTRADATA <= segLOWEST) {
1225 segLOWEST = segEXTRADATA;
1226 if (segEXTRADATA <= segLOWESTRO) {
1227 segLOWESTRO = segEXTRADATA;
1228 }
1229 }
1230 #if !(DEBUG || DEVELOPMENT)
1231
1232
1233 else {
1234 panic("EXTRADATA is in an unexpected place: %#lx > %#lx", segEXTRADATA, segLOWEST);
1235 }
1236 #endif /* !(DEBUG || DEVELOPMENT) */
1237
1238 arm_vm_page_granular_RNX(segEXTRADATA, segSizeEXTRADATA, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT);
1239 }
1240
1241 const MemoryMapFileInfo *auxKC_range, *auxKC_header_range;
1242 unsigned int auxKC_range_size, auxKC_header_range_size;
1243
1244 err = SecureDTGetProperty(memory_map, "AuxKC", (const void**)&auxKC_range,
1245 &auxKC_range_size);
1246 if (err != kSuccess) {
1247 goto noAuxKC;
1248 }
1249 assert(auxKC_range_size == sizeof(MemoryMapFileInfo));
1250 err = SecureDTGetProperty(memory_map, "AuxKC-mach_header",
1251 (const void**)&auxKC_header_range, &auxKC_header_range_size);
1252 if (err != kSuccess) {
1253 goto noAuxKC;
1254 }
1255 assert(auxKC_header_range_size == sizeof(MemoryMapFileInfo));
1256
1257 auxkc_mh = phystokv(auxKC_header_range->paddr);
1258 auxkc_base = phystokv(auxKC_range->paddr);
1259 if (!auxkc_mh || !auxkc_base) {
1260 goto noAuxKC;
1261 }
1262
1263 if (auxkc_base < segLOWEST) {
1264 auxkc_right_above = segLOWEST;
1265 segLOWEST = auxkc_base;
1266 } else {
1267 panic("auxkc_base (%p) not below segLOWEST (%p)", (void*)auxkc_base, (void*)segLOWEST);
1268 }
1269
1270 /* Map AuxKC RWNX initially so that arm_vm_auxkc_init can traverse
1271 * it and apply fixups (after we're off the bootstrap translation
1272 * tables).
1273 */
1274 arm_vm_page_granular_RWNX(auxkc_base, auxKC_range->length, 0);
1275
1276 noAuxKC:
1277 /* Map coalesced kext TEXT segment RWNX for now */
1278 arm_vm_page_granular_RWNX(segPRELINKTEXTB, segSizePRELINKTEXT, ARM64_GRANULE_ALLOW_BLOCK); // Refined in OSKext::readPrelinkedExtensions
1279
1280 /* Map coalesced kext DATA_CONST segment RWNX (could be empty) */
1281 arm_vm_page_granular_RWNX(segPLKDATACONSTB, segSizePLKDATACONST, ARM64_GRANULE_ALLOW_BLOCK); // Refined in OSKext::readPrelinkedExtensions
1282
1283 /* Map coalesced kext TEXT_EXEC segment RX (could be empty) */
1284 arm_vm_page_granular_ROX(segPLKTEXTEXECB, segSizePLKTEXTEXEC, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT); // Refined in OSKext::readPrelinkedExtensions
1285
1286 /* if new segments not present, set space between PRELINK_TEXT and xnu TEXT to RWNX
1287 * otherwise we no longer expect any space between the coalesced kext read only segments and xnu rosegments
1288 */
1289 if (!segSizePLKDATACONST && !segSizePLKTEXTEXEC) {
1290 if (segSizePRELINKTEXT) {
1291 arm_vm_page_granular_RWNX(segPRELINKTEXTB + segSizePRELINKTEXT, segTEXTB - (segPRELINKTEXTB + segSizePRELINKTEXT),
1292 ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT);
1293 }
1294 } else {
1295 /*
1296 * If we have the new segments, we should still protect the gap between kext
1297 * read-only pages and kernel read-only pages, in the event that this gap
1298 * exists.
1299 */
1300 if ((segPLKDATACONSTB + segSizePLKDATACONST) < segTEXTB) {
1301 arm_vm_page_granular_RWNX(segPLKDATACONSTB + segSizePLKDATACONST, segTEXTB - (segPLKDATACONSTB + segSizePLKDATACONST),
1302 ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT);
1303 }
1304 }
1305
1306 /*
1307 * Protection on kernel text is loose here to allow shenanigans early on. These
1308 * protections are tightened in arm_vm_prot_finalize(). This is necessary because
1309 * we currently patch LowResetVectorBase in cpu.c.
1310 *
1311 * TEXT segment contains mach headers and other non-executable data. This will become RONX later.
1312 */
1313 arm_vm_page_granular_RNX(segTEXTB, segSizeTEXT, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT);
1314
1315 /* Can DATACONST start out and stay RNX?
1316 * NO, stuff in this segment gets modified during startup (viz. mac_policy_init()/mac_policy_list)
1317 * Make RNX in prot_finalize
1318 */
1319 arm_vm_page_granular_RWNX(segDATACONSTB, segSizeDATACONST, ARM64_GRANULE_ALLOW_BLOCK);
1320
1321 arm_vm_page_granular_ROX(segTEXTEXECB, segSizeTEXTEXEC, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT);
1322
1323 #if XNU_MONITOR
1324 arm_vm_page_granular_ROX(segPPLTEXTB, segSizePPLTEXT, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT);
1325 arm_vm_page_granular_ROX(segPPLTRAMPB, segSizePPLTRAMP, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT);
1326 arm_vm_page_granular_RNX(segPPLDATACONSTB, segSizePPLDATACONST, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT);
1327 #endif
1328
1329 /* DATA segment will remain RWNX */
1330 arm_vm_page_granular_RWNX(segDATAB, segSizeDATA, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT);
1331 #if XNU_MONITOR
1332 arm_vm_page_granular_RWNX(segPPLDATAB, segSizePPLDATA, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT);
1333 #endif
1334
1335 arm_vm_page_granular_RWNX(segHIBDATAB, segSizeHIBDATA, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT);
1336
1337 arm_vm_page_granular_RWNX(segBOOTDATAB, segSizeBOOTDATA, 0);
1338 arm_vm_page_granular_RNX((vm_offset_t)&intstack_low_guard, PAGE_MAX_SIZE, 0);
1339 arm_vm_page_granular_RNX((vm_offset_t)&intstack_high_guard, PAGE_MAX_SIZE, 0);
1340 arm_vm_page_granular_RNX((vm_offset_t)&excepstack_high_guard, PAGE_MAX_SIZE, 0);
1341
1342 arm_vm_page_granular_ROX(segKLDB, segSizeKLD, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT);
1343 arm_vm_page_granular_RNX(segKLDDATAB, segSizeKLDDATA, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT);
1344 arm_vm_page_granular_RWNX(segLINKB, segSizeLINK, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT);
1345 arm_vm_page_granular_RWNX(segPLKLINKEDITB, segSizePLKLINKEDIT, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT); // Coalesced kext LINKEDIT segment
1346 arm_vm_page_granular_ROX(segLASTB, segSizeLAST, ARM64_GRANULE_ALLOW_BLOCK); // __LAST may be empty, but we cannot assume this
1347 if (segLASTDATACONSTB) {
1348 arm_vm_page_granular_RWNX(segLASTDATACONSTB, segSizeLASTDATACONST, ARM64_GRANULE_ALLOW_BLOCK); // __LASTDATA_CONST may be empty, but we cannot assume this
1349 }
1350 arm_vm_page_granular_RWNX(segPRELINKDATAB, segSizePRELINKDATA, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT); // Prelink __DATA for kexts (RW data)
1351
1352 if (segSizePLKLLVMCOV > 0) {
1353 arm_vm_page_granular_RWNX(segPLKLLVMCOVB, segSizePLKLLVMCOV, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT); // LLVM code coverage data
1354 }
1355 arm_vm_page_granular_RWNX(segPRELINKINFOB, segSizePRELINKINFO, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT); /* PreLinkInfoDictionary */
1356
1357 /* Record the bounds of the kernelcache. */
1358 vm_kernelcache_base = segLOWEST;
1359 vm_kernelcache_top = end_kern;
1360 }
1361
1362 /*
1363 * return < 0 for a < b
1364 * 0 for a == b
1365 * > 0 for a > b
1366 */
1367 typedef int (*cmpfunc_t)(const void *a, const void *b);
1368
1369 extern void
1370 qsort(void *a, size_t n, size_t es, cmpfunc_t cmp);
1371
1372 static int
1373 cmp_ptov_entries(const void *a, const void *b)
1374 {
1375 const ptov_table_entry *entry_a = a;
1376 const ptov_table_entry *entry_b = b;
1377 // Sort in descending order of segment length
1378 if (entry_a->len < entry_b->len) {
1379 return 1;
1380 } else if (entry_a->len > entry_b->len) {
1381 return -1;
1382 } else {
1383 return 0;
1384 }
1385 }
1386
1387 SECURITY_READ_ONLY_LATE(static unsigned int) ptov_index = 0;
1388
1389 #define ROUND_L1(addr) (((addr) + ARM_TT_L1_OFFMASK) & ~(ARM_TT_L1_OFFMASK))
1390 #define ROUND_TWIG(addr) (((addr) + ARM_TT_TWIG_OFFMASK) & ~(ARM_TT_TWIG_OFFMASK))
1391
1392 static void
1393 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)
1394 {
1395 pmap_paddr_t pa_offset;
1396
1397 assert(ptov_index < PTOV_TABLE_SIZE);
1398 assert((orig_va & ARM_PGMASK) == 0);
1399 temp_ptov_table[ptov_index].pa = orig_va - gVirtBase + gPhysBase;
1400 if (ptov_index == 0) {
1401 temp_ptov_table[ptov_index].va = physmap_base;
1402 } else {
1403 temp_ptov_table[ptov_index].va = temp_ptov_table[ptov_index - 1].va + temp_ptov_table[ptov_index - 1].len;
1404 }
1405 if (granule & ARM64_GRANULE_ALLOW_BLOCK) {
1406 vm_map_address_t orig_offset = temp_ptov_table[ptov_index].pa & ARM_TT_TWIG_OFFMASK;
1407 vm_map_address_t new_offset = temp_ptov_table[ptov_index].va & ARM_TT_TWIG_OFFMASK;
1408 if (new_offset < orig_offset) {
1409 temp_ptov_table[ptov_index].va += (orig_offset - new_offset);
1410 } else if (new_offset > orig_offset) {
1411 temp_ptov_table[ptov_index].va = ROUND_TWIG(temp_ptov_table[ptov_index].va) + orig_offset;
1412 }
1413 }
1414 assert((temp_ptov_table[ptov_index].va & ARM_PGMASK) == 0);
1415 temp_ptov_table[ptov_index].len = round_page(len);
1416 pa_offset = temp_ptov_table[ptov_index].va - orig_va;
1417 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);
1418 ++ptov_index;
1419 }
1420
1421 #if XNU_MONITOR
1422
1423 SECURITY_READ_ONLY_LATE(static boolean_t) keep_linkedit = FALSE;
1424
1425 static void
1426 arm_vm_physmap_init(boot_args *args)
1427 {
1428 ptov_table_entry temp_ptov_table[PTOV_TABLE_SIZE];
1429 bzero(temp_ptov_table, sizeof(temp_ptov_table));
1430
1431 // This is memory that will either be handed back to the VM layer via ml_static_mfree(),
1432 // or will be available for general-purpose use. Physical aperture mappings for this memory
1433 // must be at page granularity, so that PPL ownership or cache attribute changes can be reflected
1434 // in the physical aperture mappings.
1435
1436 // Slid region between gPhysBase and beginning of protected text
1437 arm_vm_physmap_slide(temp_ptov_table, gVirtBase, segLOWEST - gVirtBase, AP_RWNA, 0);
1438
1439 // kext bootstrap segments
1440 #if !defined(KERNEL_INTEGRITY_KTRR) && !defined(KERNEL_INTEGRITY_CTRR)
1441 /* __KLD,__text is covered by the rorgn */
1442 arm_vm_physmap_slide(temp_ptov_table, segKLDB, segSizeKLD, AP_RONA, 0);
1443 #endif
1444 arm_vm_physmap_slide(temp_ptov_table, segKLDDATAB, segSizeKLDDATA, AP_RONA, 0);
1445
1446 // Early-boot data
1447 arm_vm_physmap_slide(temp_ptov_table, segBOOTDATAB, segSizeBOOTDATA, AP_RONA, 0);
1448
1449 #if KASAN_DYNAMIC_BLACKLIST
1450 /* KASAN's dynamic blacklist needs to query the LINKEDIT segment at runtime. As such, the
1451 * kext bootstrap code will not jettison LINKEDIT on kasan kernels, so don't bother to relocate it. */
1452 keep_linkedit = TRUE;
1453 #else
1454 PE_parse_boot_argn("keepsyms", &keep_linkedit, sizeof(keep_linkedit));
1455 if (kernel_mach_header_is_in_fileset(&_mh_execute_header)) {
1456 keep_linkedit = TRUE;
1457 }
1458 #endif
1459 if (!keep_linkedit) {
1460 // Kernel LINKEDIT
1461 arm_vm_physmap_slide(temp_ptov_table, segLINKB, segSizeLINK, AP_RWNA, 0);
1462
1463 // Prelinked kernel LINKEDIT
1464 arm_vm_physmap_slide(temp_ptov_table, segPLKLINKEDITB, segSizePLKLINKEDIT, AP_RWNA, 0);
1465 }
1466
1467 // Prelinked kernel plists
1468 arm_vm_physmap_slide(temp_ptov_table, segPRELINKINFOB, segSizePRELINKINFO, AP_RWNA, 0);
1469
1470 // Device tree (if not locked down), ramdisk, boot args
1471 arm_vm_physmap_slide(temp_ptov_table, end_kern, (args->topOfKernelData - gPhysBase + gVirtBase) - end_kern, AP_RWNA, 0);
1472 if (!SecureDTIsLockedDown()) {
1473 PE_slide_devicetree(temp_ptov_table[ptov_index - 1].va - end_kern);
1474 }
1475
1476 // Remainder of physical memory
1477 arm_vm_physmap_slide(temp_ptov_table, (args->topOfKernelData - gPhysBase + gVirtBase),
1478 real_avail_end - args->topOfKernelData, AP_RWNA, 0);
1479
1480 assert((temp_ptov_table[ptov_index - 1].va + temp_ptov_table[ptov_index - 1].len) <= physmap_end);
1481
1482 // Sort in descending order of segment length. LUT traversal is linear, so largest (most likely used)
1483 // segments should be placed earliest in the table to optimize lookup performance.
1484 qsort(temp_ptov_table, PTOV_TABLE_SIZE, sizeof(temp_ptov_table[0]), cmp_ptov_entries);
1485
1486 memcpy(ptov_table, temp_ptov_table, sizeof(ptov_table));
1487 }
1488
1489 #else
1490
1491 static void
1492 arm_vm_physmap_init(boot_args *args)
1493 {
1494 ptov_table_entry temp_ptov_table[PTOV_TABLE_SIZE];
1495 bzero(temp_ptov_table, sizeof(temp_ptov_table));
1496
1497 // Will be handed back to VM layer through ml_static_mfree() in arm_vm_prot_finalize()
1498 arm_vm_physmap_slide(temp_ptov_table, gVirtBase, segLOWEST - gVirtBase, AP_RWNA,
1499 ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT);
1500
1501 arm_vm_page_granular_RWNX(end_kern, phystokv(args->topOfKernelData) - end_kern,
1502 ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT); /* Device Tree (if not locked down), RAM Disk (if present), bootArgs */
1503
1504 arm_vm_physmap_slide(temp_ptov_table, (args->topOfKernelData - gPhysBase + gVirtBase),
1505 real_avail_end - args->topOfKernelData, AP_RWNA, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT); // rest of physmem
1506
1507 assert((temp_ptov_table[ptov_index - 1].va + temp_ptov_table[ptov_index - 1].len) <= physmap_end);
1508
1509 // Sort in descending order of segment length. LUT traversal is linear, so largest (most likely used)
1510 // segments should be placed earliest in the table to optimize lookup performance.
1511 qsort(temp_ptov_table, PTOV_TABLE_SIZE, sizeof(temp_ptov_table[0]), cmp_ptov_entries);
1512
1513 memcpy(ptov_table, temp_ptov_table, sizeof(ptov_table));
1514 }
1515
1516 #endif // XNU_MONITOR
1517
1518 void
1519 arm_vm_prot_finalize(boot_args * args __unused)
1520 {
1521 /*
1522 * At this point, we are far enough along in the boot process that it will be
1523 * safe to free up all of the memory preceeding the kernel. It may in fact
1524 * be safe to do this earlier.
1525 *
1526 * This keeps the memory in the V-to-P mapping, but advertises it to the VM
1527 * as usable.
1528 */
1529
1530 /*
1531 * if old style PRELINK segment exists, free memory before it, and after it before XNU text
1532 * otherwise we're dealing with a new style kernel cache, so we should just free the
1533 * memory before PRELINK_TEXT segment, since the rest of the KEXT read only data segments
1534 * should be immediately followed by XNU's TEXT segment
1535 */
1536
1537 ml_static_mfree(phystokv(gPhysBase), segLOWEST - gVirtBase);
1538
1539 /*
1540 * KTRR support means we will be mucking with these pages and trying to
1541 * protect them; we cannot free the pages to the VM if we do this.
1542 */
1543 if (!segSizePLKDATACONST && !segSizePLKTEXTEXEC && segSizePRELINKTEXT) {
1544 /* If new segments not present, PRELINK_TEXT is not dynamically sized, free DRAM between it and xnu TEXT */
1545 ml_static_mfree(segPRELINKTEXTB + segSizePRELINKTEXT, segTEXTB - (segPRELINKTEXTB + segSizePRELINKTEXT));
1546 }
1547
1548 /* tighten permissions on kext read only data and code */
1549 arm_vm_page_granular_RNX(segPRELINKTEXTB, segSizePRELINKTEXT, ARM64_GRANULE_ALLOW_BLOCK);
1550 arm_vm_page_granular_RNX(segPLKDATACONSTB, segSizePLKDATACONST, ARM64_GRANULE_ALLOW_BLOCK);
1551
1552 cpu_stack_alloc(&BootCpuData);
1553 arm64_replace_bootstack(&BootCpuData);
1554 ml_static_mfree(phystokv(segBOOTDATAB - gVirtBase + gPhysBase), segSizeBOOTDATA);
1555
1556 #if __ARM_KERNEL_PROTECT__
1557 arm_vm_populate_kernel_el0_mappings();
1558 #endif /* __ARM_KERNEL_PROTECT__ */
1559
1560 #if XNU_MONITOR
1561 #if !defined(KERNEL_INTEGRITY_KTRR) && !defined(KERNEL_INTEGRITY_CTRR)
1562 /* __KLD,__text is covered by the rorgn */
1563 for (vm_offset_t va = segKLDB; va < (segKLDB + segSizeKLD); va += ARM_PGBYTES) {
1564 pt_entry_t *pte = arm_kva_to_pte(va);
1565 *pte = ARM_PTE_EMPTY;
1566 }
1567 #endif
1568 for (vm_offset_t va = segKLDDATAB; va < (segKLDDATAB + segSizeKLDDATA); va += ARM_PGBYTES) {
1569 pt_entry_t *pte = arm_kva_to_pte(va);
1570 *pte = ARM_PTE_EMPTY;
1571 }
1572 /* Clear the original stack mappings; these pages should be mapped through ptov_table. */
1573 for (vm_offset_t va = segBOOTDATAB; va < (segBOOTDATAB + segSizeBOOTDATA); va += ARM_PGBYTES) {
1574 pt_entry_t *pte = arm_kva_to_pte(va);
1575 *pte = ARM_PTE_EMPTY;
1576 }
1577 /* Clear the original PRELINKINFO mapping. This segment should be jettisoned during I/O Kit
1578 * initialization before we reach this point. */
1579 for (vm_offset_t va = segPRELINKINFOB; va < (segPRELINKINFOB + segSizePRELINKINFO); va += ARM_PGBYTES) {
1580 pt_entry_t *pte = arm_kva_to_pte(va);
1581 *pte = ARM_PTE_EMPTY;
1582 }
1583 if (!keep_linkedit) {
1584 for (vm_offset_t va = segLINKB; va < (segLINKB + segSizeLINK); va += ARM_PGBYTES) {
1585 pt_entry_t *pte = arm_kva_to_pte(va);
1586 *pte = ARM_PTE_EMPTY;
1587 }
1588 for (vm_offset_t va = segPLKLINKEDITB; va < (segPLKLINKEDITB + segSizePLKLINKEDIT); va += ARM_PGBYTES) {
1589 pt_entry_t *pte = arm_kva_to_pte(va);
1590 *pte = ARM_PTE_EMPTY;
1591 }
1592 }
1593 #endif /* XNU_MONITOR */
1594
1595 #if defined(KERNEL_INTEGRITY_KTRR) || defined(KERNEL_INTEGRITY_CTRR)
1596 /*
1597 * __LAST,__pinst should no longer be executable.
1598 */
1599 arm_vm_page_granular_RNX(segLASTB, segSizeLAST, ARM64_GRANULE_ALLOW_BLOCK);
1600
1601 /* __LASTDATA_CONST should no longer be writable. */
1602 if (segLASTDATACONSTB) {
1603 arm_vm_page_granular_RNX(segLASTDATACONSTB, segSizeLASTDATACONST, ARM64_GRANULE_ALLOW_BLOCK);
1604 }
1605
1606 /*
1607 * __KLD,__text should no longer be executable.
1608 */
1609 arm_vm_page_granular_RNX(segKLDB, segSizeKLD, ARM64_GRANULE_ALLOW_BLOCK);
1610
1611 /*
1612 * Must wait until all other region permissions are set before locking down DATA_CONST
1613 * as the kernel static page tables live in DATA_CONST on KTRR enabled systems
1614 * and will become immutable.
1615 */
1616 #endif
1617
1618 arm_vm_page_granular_RNX(segDATACONSTB, segSizeDATACONST, ARM64_GRANULE_ALLOW_BLOCK);
1619
1620 __builtin_arm_dsb(DSB_ISH);
1621 flush_mmu_tlb();
1622 }
1623
1624 #define TBI_USER 0x1
1625 #define TBI_KERNEL 0x2
1626
1627 /*
1628 * TBI (top-byte ignore) is an ARMv8 feature for ignoring the top 8 bits of
1629 * address accesses. It can be enabled separately for TTBR0 (user) and
1630 * TTBR1 (kernel). We enable it by default for user only.
1631 */
1632 static void
1633 set_tbi(void)
1634 {
1635 #if !__ARM_KERNEL_PROTECT__
1636 uint64_t old_tcr, new_tcr;
1637
1638 old_tcr = new_tcr = get_tcr();
1639 new_tcr |= TCR_TBI0_TOPBYTE_IGNORED;
1640
1641 if (old_tcr != new_tcr) {
1642 set_tcr(new_tcr);
1643 sysreg_restore.tcr_el1 = new_tcr;
1644 }
1645 #endif /* !__ARM_KERNEL_PROTECT__ */
1646 }
1647
1648 /*
1649 * Initialize and enter blank (invalid) page tables in a L1 translation table for a given VA range.
1650 *
1651 * This is a helper function used to build up the initial page tables for the kernel translation table.
1652 * With KERNEL_INTEGRITY we keep at least the root level of the kernel page table immutable, thus the need
1653 * to preallocate before machine_lockdown any L1 entries necessary during the entire kernel runtime.
1654 *
1655 * For a given VA range, if necessary, allocate new L2 translation tables and install the table entries in
1656 * the appropriate L1 table indexes. called before the translation table is active
1657 *
1658 * parameters:
1659 *
1660 * tt: virtual address of L1 translation table to modify
1661 * start: beginning of VA range
1662 * end: end of VA range
1663 * static_map: whether to allocate the new translation table page from read only memory
1664 * table_attrs: attributes of new table entry in addition to VALID and TYPE_TABLE attributes
1665 *
1666 */
1667
1668 static void
1669 init_ptpages(tt_entry_t *tt, vm_map_address_t start, vm_map_address_t end, bool static_map, uint64_t table_attrs)
1670 {
1671 tt_entry_t *l1_tte;
1672 vm_offset_t ptpage_vaddr;
1673
1674 l1_tte = tt + ((start & ARM_TT_L1_INDEX_MASK) >> ARM_TT_L1_SHIFT);
1675
1676 while (start < end) {
1677 if (*l1_tte == ARM_TTE_EMPTY) {
1678 /* Allocate a page and setup L1 Table TTE in L1 */
1679 ptpage_vaddr = alloc_ptpage(static_map);
1680 *l1_tte = (kvtophys(ptpage_vaddr) & ARM_TTE_TABLE_MASK) | ARM_TTE_TYPE_TABLE | ARM_TTE_VALID | table_attrs;
1681 bzero((void *)ptpage_vaddr, ARM_PGBYTES);
1682 }
1683
1684 if ((start + ARM_TT_L1_SIZE) < start) {
1685 /* If this is the last L1 entry, it must cover the last mapping. */
1686 break;
1687 }
1688
1689 start += ARM_TT_L1_SIZE;
1690 l1_tte++;
1691 }
1692 }
1693
1694 #define ARM64_PHYSMAP_SLIDE_RANGE (1ULL << 30) // 1 GB
1695 #define ARM64_PHYSMAP_SLIDE_MASK (ARM64_PHYSMAP_SLIDE_RANGE - 1)
1696
1697 void
1698 arm_vm_init(uint64_t memory_size, boot_args * args)
1699 {
1700 vm_map_address_t va_l1, va_l1_end;
1701 tt_entry_t *cpu_l1_tte;
1702 vm_map_address_t va_l2, va_l2_end;
1703 tt_entry_t *cpu_l2_tte;
1704 pmap_paddr_t boot_ttep;
1705 tt_entry_t *boot_tte;
1706 uint64_t mem_segments;
1707 vm_offset_t ptpage_vaddr;
1708 vm_map_address_t dynamic_memory_begin;
1709
1710 /*
1711 * Get the virtual and physical kernel-managed memory base from boot_args.
1712 */
1713 gVirtBase = args->virtBase;
1714 gPhysBase = args->physBase;
1715 #if KASAN
1716 real_phys_size = args->memSize + (shadow_ptop - shadow_pbase);
1717 #else
1718 real_phys_size = args->memSize;
1719 #endif
1720 /*
1721 * Ensure the physical region we specify for the VM to manage ends on a
1722 * software page boundary. Note that the software page size (PAGE_SIZE)
1723 * may be a multiple of the hardware page size specified in ARM_PGBYTES.
1724 * We must round the reported memory size down to the nearest PAGE_SIZE
1725 * boundary to ensure the VM does not try to manage a page it does not
1726 * completely own. The KASAN shadow region, if present, is managed entirely
1727 * in units of the hardware page size and should not need similar treatment.
1728 */
1729 gPhysSize = mem_size = ((gPhysBase + args->memSize) & ~PAGE_MASK) - gPhysBase;
1730
1731 mem_actual = args->memSizeActual ? args->memSizeActual : mem_size;
1732
1733 if ((memory_size != 0) && (mem_size > memory_size)) {
1734 mem_size = memory_size;
1735 max_mem_actual = memory_size;
1736 } else {
1737 max_mem_actual = mem_actual;
1738 }
1739 if (mem_size >= ((VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS) / 2)) {
1740 panic("Unsupported memory configuration %lx\n", mem_size);
1741 }
1742
1743 #if defined(ARM_LARGE_MEMORY)
1744 unsigned long physmap_l1_entries = ((real_phys_size + ARM64_PHYSMAP_SLIDE_RANGE) >> ARM_TT_L1_SHIFT) + 1;
1745 physmap_base = VM_MIN_KERNEL_ADDRESS - (physmap_l1_entries << ARM_TT_L1_SHIFT);
1746 #else
1747 physmap_base = phystokv(args->topOfKernelData);
1748 #endif
1749
1750 // Slide the physical aperture to a random page-aligned location within the slide range
1751 uint64_t physmap_slide = early_random() & ARM64_PHYSMAP_SLIDE_MASK & ~((uint64_t)PAGE_MASK);
1752 assert(physmap_slide < ARM64_PHYSMAP_SLIDE_RANGE);
1753
1754 physmap_base += physmap_slide;
1755
1756 #if XNU_MONITOR
1757 physmap_base = ROUND_TWIG(physmap_base);
1758 #if defined(ARM_LARGE_MEMORY)
1759 static_memory_end = phystokv(args->topOfKernelData);
1760 #else
1761 static_memory_end = physmap_base + mem_size;
1762 #endif // ARM_LARGE_MEMORY
1763 physmap_end = physmap_base + real_phys_size;
1764 #else
1765 static_memory_end = physmap_base + mem_size + (PTOV_TABLE_SIZE * ARM_TT_TWIG_SIZE); // worst possible case for block alignment
1766 physmap_end = physmap_base + real_phys_size + (PTOV_TABLE_SIZE * ARM_TT_TWIG_SIZE);
1767 #endif
1768
1769 #if KASAN && !defined(ARM_LARGE_MEMORY)
1770 /* add the KASAN stolen memory to the physmap */
1771 dynamic_memory_begin = static_memory_end + (shadow_ptop - shadow_pbase);
1772 #else
1773 dynamic_memory_begin = static_memory_end;
1774 #endif
1775 #if XNU_MONITOR
1776 pmap_stacks_start = (void*)dynamic_memory_begin;
1777 dynamic_memory_begin += PPL_STACK_REGION_SIZE;
1778 pmap_stacks_end = (void*)dynamic_memory_begin;
1779 #endif
1780 if (dynamic_memory_begin > VM_MAX_KERNEL_ADDRESS) {
1781 panic("Unsupported memory configuration %lx\n", mem_size);
1782 }
1783
1784 boot_tte = (tt_entry_t *)&bootstrap_pagetables;
1785 boot_ttep = kvtophys((vm_offset_t)boot_tte);
1786
1787 #if DEVELOPMENT || DEBUG
1788 /* Sanity check - assert that BOOTSTRAP_TABLE_SIZE is sufficiently-large to
1789 * hold our bootstrap mappings for any possible slide */
1790 size_t bytes_mapped = dynamic_memory_begin - gVirtBase;
1791 size_t l1_entries = 1 + ((bytes_mapped + ARM_TT_L1_SIZE - 1) / ARM_TT_L1_SIZE);
1792 /* 1 L1 each for V=P and KVA, plus 1 page for each L2 */
1793 size_t pages_used = 2 * (l1_entries + 1);
1794 if (pages_used > BOOTSTRAP_TABLE_SIZE) {
1795 panic("BOOTSTRAP_TABLE_SIZE too small for memory config\n");
1796 }
1797 #endif
1798
1799 /*
1800 * TTBR0 L1, TTBR0 L2 - 1:1 bootstrap mapping.
1801 * TTBR1 L1, TTBR1 L2 - kernel mapping
1802 */
1803
1804 /*
1805 * TODO: free bootstrap table memory back to allocator.
1806 * on large memory systems bootstrap tables could be quite large.
1807 * after bootstrap complete, xnu can warm start with a single 16KB page mapping
1808 * to trampoline to KVA. this requires only 3 pages to stay resident.
1809 */
1810 avail_start = args->topOfKernelData;
1811
1812 #if defined(KERNEL_INTEGRITY_KTRR) || defined(KERNEL_INTEGRITY_CTRR)
1813 arm_replace_identity_map();
1814 #endif
1815
1816 /* Initialize invalid tte page */
1817 invalid_tte = (tt_entry_t *)alloc_ptpage(TRUE);
1818 invalid_ttep = kvtophys((vm_offset_t)invalid_tte);
1819 bzero(invalid_tte, ARM_PGBYTES);
1820
1821 /*
1822 * Initialize l1 page table page
1823 */
1824 cpu_tte = (tt_entry_t *)alloc_ptpage(TRUE);
1825 cpu_ttep = kvtophys((vm_offset_t)cpu_tte);
1826 bzero(cpu_tte, ARM_PGBYTES);
1827 avail_end = gPhysBase + mem_size;
1828 assert(!(avail_end & PAGE_MASK));
1829
1830 #if KASAN
1831 real_avail_end = gPhysBase + real_phys_size;
1832 #else
1833 real_avail_end = avail_end;
1834 #endif
1835
1836 /*
1837 * Initialize l1 and l2 page table pages :
1838 * map physical memory at the kernel base virtual address
1839 * cover the kernel dynamic address range section
1840 *
1841 * the so called physical aperture should be statically mapped
1842 */
1843 init_ptpages(cpu_tte, gVirtBase, dynamic_memory_begin, TRUE, 0);
1844
1845 #if defined(ARM_LARGE_MEMORY)
1846 /*
1847 * Initialize l1 page table pages :
1848 * on large memory systems the physical aperture exists separately below
1849 * the rest of the kernel virtual address space
1850 */
1851 init_ptpages(cpu_tte, physmap_base, ROUND_L1(physmap_end), TRUE, ARM_DYNAMIC_TABLE_XN);
1852 #endif
1853
1854
1855 #if __ARM_KERNEL_PROTECT__
1856 /* Expand the page tables to prepare for the EL0 mappings. */
1857 arm_vm_expand_kernel_el0_mappings();
1858 #endif /* __ARM_KERNEL_PROTECT__ */
1859
1860 /*
1861 * Now retrieve addresses for various segments from kernel mach-o header
1862 */
1863 segPRELINKTEXTB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__PRELINK_TEXT", &segSizePRELINKTEXT);
1864 segPLKDATACONSTB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__PLK_DATA_CONST", &segSizePLKDATACONST);
1865 segPLKTEXTEXECB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__PLK_TEXT_EXEC", &segSizePLKTEXTEXEC);
1866 segTEXTB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__TEXT", &segSizeTEXT);
1867 segDATACONSTB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__DATA_CONST", &segSizeDATACONST);
1868 segTEXTEXECB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__TEXT_EXEC", &segSizeTEXTEXEC);
1869 #if XNU_MONITOR
1870 segPPLTEXTB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__PPLTEXT", &segSizePPLTEXT);
1871 segPPLTRAMPB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__PPLTRAMP", &segSizePPLTRAMP);
1872 segPPLDATACONSTB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__PPLDATA_CONST", &segSizePPLDATACONST);
1873 #endif
1874 segDATAB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__DATA", &segSizeDATA);
1875 #if XNU_MONITOR
1876 segPPLDATAB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__PPLDATA", &segSizePPLDATA);
1877 #endif
1878
1879 segBOOTDATAB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__BOOTDATA", &segSizeBOOTDATA);
1880 segLINKB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__LINKEDIT", &segSizeLINK);
1881 segKLDB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__KLD", &segSizeKLD);
1882 segKLDDATAB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__KLDDATA", &segSizeKLDDATA);
1883 segPRELINKDATAB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__PRELINK_DATA", &segSizePRELINKDATA);
1884 segPRELINKINFOB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__PRELINK_INFO", &segSizePRELINKINFO);
1885 segPLKLLVMCOVB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__PLK_LLVM_COV", &segSizePLKLLVMCOV);
1886 segPLKLINKEDITB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__PLK_LINKEDIT", &segSizePLKLINKEDIT);
1887 segLASTB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__LAST", &segSizeLAST);
1888 segLASTDATACONSTB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__LASTDATA_CONST", &segSizeLASTDATACONST);
1889
1890 sectHIBTEXTB = (vm_offset_t) getsectdatafromheader(&_mh_execute_header, "__TEXT_EXEC", "__hib_text", &sectSizeHIBTEXT);
1891 sectHIBDATACONSTB = (vm_offset_t) getsectdatafromheader(&_mh_execute_header, "__DATA_CONST", "__hib_const", &sectSizeHIBDATACONST);
1892 segHIBDATAB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__HIBDATA", &segSizeHIBDATA);
1893
1894 if (kernel_mach_header_is_in_fileset(&_mh_execute_header)) {
1895 kernel_mach_header_t *kc_mh = PE_get_kc_header(KCKindPrimary);
1896
1897 // fileset has kext PLK_TEXT_EXEC under kernel collection TEXT_EXEC following kernel's LAST
1898 segKCTEXTEXECB = (vm_offset_t) getsegdatafromheader(kc_mh, "__TEXT_EXEC", &segSizeKCTEXTEXEC);
1899 assert(segPLKTEXTEXECB && !segSizePLKTEXTEXEC); // kernel PLK_TEXT_EXEC must be empty
1900
1901 assert(segLASTB); // kernel LAST can be empty, but it must have
1902 // a valid address for computations below.
1903
1904 assert(segKCTEXTEXECB <= segLASTB); // KC TEXT_EXEC must contain kernel LAST
1905 assert(segKCTEXTEXECB + segSizeKCTEXTEXEC >= segLASTB + segSizeLAST);
1906 segPLKTEXTEXECB = segLASTB + segSizeLAST;
1907 segSizePLKTEXTEXEC = segSizeKCTEXTEXEC - (segPLKTEXTEXECB - segKCTEXTEXECB);
1908
1909 // fileset has kext PLK_DATA_CONST under kernel collection DATA_CONST following kernel's LASTDATA_CONST
1910 segKCDATACONSTB = (vm_offset_t) getsegdatafromheader(kc_mh, "__DATA_CONST", &segSizeKCDATACONST);
1911 assert(segPLKDATACONSTB && !segSizePLKDATACONST); // kernel PLK_DATA_CONST must be empty
1912 assert(segLASTDATACONSTB && segSizeLASTDATACONST); // kernel LASTDATA_CONST must be non-empty
1913 assert(segKCDATACONSTB <= segLASTDATACONSTB); // KC DATA_CONST must contain kernel LASTDATA_CONST
1914 assert(segKCDATACONSTB + segSizeKCDATACONST >= segLASTDATACONSTB + segSizeLASTDATACONST);
1915 segPLKDATACONSTB = segLASTDATACONSTB + segSizeLASTDATACONST;
1916 segSizePLKDATACONST = segSizeKCDATACONST - (segPLKDATACONSTB - segKCDATACONSTB);
1917
1918 // fileset has kext PRELINK_DATA under kernel collection DATA following kernel's empty PRELINK_DATA
1919 segKCDATAB = (vm_offset_t) getsegdatafromheader(kc_mh, "__DATA", &segSizeKCDATA);
1920 assert(segPRELINKDATAB && !segSizePRELINKDATA); // kernel PRELINK_DATA must be empty
1921 assert(segKCDATAB <= segPRELINKDATAB); // KC DATA must contain kernel PRELINK_DATA
1922 assert(segKCDATAB + segSizeKCDATA >= segPRELINKDATAB + segSizePRELINKDATA);
1923 segSizePRELINKDATA = segSizeKCDATA - (segPRELINKDATAB - segKCDATAB);
1924
1925 // fileset has consolidated PRELINK_TEXT, PRELINK_INFO and LINKEDIT at the kernel collection level
1926 assert(segPRELINKTEXTB && !segSizePRELINKTEXT); // kernel PRELINK_TEXT must be empty
1927 segPRELINKTEXTB = (vm_offset_t) getsegdatafromheader(kc_mh, "__PRELINK_TEXT", &segSizePRELINKTEXT);
1928 assert(segPRELINKINFOB && !segSizePRELINKINFO); // kernel PRELINK_INFO must be empty
1929 segPRELINKINFOB = (vm_offset_t) getsegdatafromheader(kc_mh, "__PRELINK_INFO", &segSizePRELINKINFO);
1930 segLINKB = (vm_offset_t) getsegdatafromheader(kc_mh, "__LINKEDIT", &segSizeLINK);
1931 }
1932
1933 (void) PE_parse_boot_argn("use_contiguous_hint", &use_contiguous_hint, sizeof(use_contiguous_hint));
1934 assert(segSizePRELINKTEXT < 0x03000000); /* 23355738 */
1935
1936 /* if one of the new segments is present, the other one better be as well */
1937 if (segSizePLKDATACONST || segSizePLKTEXTEXEC) {
1938 assert(segSizePLKDATACONST && segSizePLKTEXTEXEC);
1939 }
1940
1941 etext = (vm_offset_t) segTEXTB + segSizeTEXT;
1942 sdata = (vm_offset_t) segDATAB;
1943 edata = (vm_offset_t) segDATAB + segSizeDATA;
1944 end_kern = round_page(segHIGHESTKC ? segHIGHESTKC : getlastaddr()); /* Force end to next page */
1945
1946 vm_set_page_size();
1947
1948 vm_kernel_base = segTEXTB;
1949 vm_kernel_top = (vm_offset_t) &last_kernel_symbol;
1950 vm_kext_base = segPRELINKTEXTB;
1951 vm_kext_top = vm_kext_base + segSizePRELINKTEXT;
1952
1953 vm_prelink_stext = segPRELINKTEXTB;
1954 if (!segSizePLKTEXTEXEC && !segSizePLKDATACONST) {
1955 vm_prelink_etext = segPRELINKTEXTB + segSizePRELINKTEXT;
1956 } else {
1957 vm_prelink_etext = segPRELINKTEXTB + segSizePRELINKTEXT + segSizePLKDATACONST + segSizePLKTEXTEXEC;
1958 }
1959 vm_prelink_sinfo = segPRELINKINFOB;
1960 vm_prelink_einfo = segPRELINKINFOB + segSizePRELINKINFO;
1961 vm_slinkedit = segLINKB;
1962 vm_elinkedit = segLINKB + segSizeLINK;
1963
1964 vm_prelink_sdata = segPRELINKDATAB;
1965 vm_prelink_edata = segPRELINKDATAB + segSizePRELINKDATA;
1966
1967 arm_vm_prot_init(args);
1968
1969 vm_page_kernelcache_count = (unsigned int) (atop_64(end_kern - segLOWEST));
1970
1971 /*
1972 * Initialize the page tables for the low globals:
1973 * cover this address range:
1974 * LOW_GLOBAL_BASE_ADDRESS + 2MB
1975 */
1976 va_l1 = va_l2 = LOW_GLOBAL_BASE_ADDRESS;
1977 cpu_l1_tte = cpu_tte + ((va_l1 & ARM_TT_L1_INDEX_MASK) >> ARM_TT_L1_SHIFT);
1978 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);
1979 ptpage_vaddr = alloc_ptpage(TRUE);
1980 *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;
1981 bzero((void *)ptpage_vaddr, ARM_PGBYTES);
1982
1983 /*
1984 * Initialize l2 page table pages :
1985 * cover this address range:
1986 * KERNEL_DYNAMIC_ADDR - VM_MAX_KERNEL_ADDRESS
1987 */
1988 #if defined(ARM_LARGE_MEMORY)
1989 /*
1990 * dynamic mapped memory outside the VM allocator VA range required to bootstrap VM system
1991 * don't expect to exceed 64GB, no sense mapping any more space between here and the VM heap range
1992 */
1993 init_ptpages(cpu_tte, dynamic_memory_begin, ROUND_L1(dynamic_memory_begin), FALSE, ARM_DYNAMIC_TABLE_XN);
1994 #else
1995 /*
1996 * TODO: do these pages really need to come from RO memory?
1997 * With legacy 3 level table systems we never mapped more than a single L1 entry so this may be dead code
1998 */
1999 init_ptpages(cpu_tte, dynamic_memory_begin, VM_MAX_KERNEL_ADDRESS, TRUE, ARM_DYNAMIC_TABLE_XN);
2000 #endif
2001
2002 #if KASAN
2003 /* record the extent of the physmap */
2004 physmap_vbase = physmap_base;
2005 physmap_vtop = physmap_end;
2006 kasan_init();
2007 #endif /* KASAN */
2008
2009 #if MONOTONIC
2010 mt_early_init();
2011 #endif /* MONOTONIC */
2012
2013 set_tbi();
2014
2015 arm_vm_physmap_init(args);
2016 set_mmu_ttb_alternate(cpu_ttep & TTBR_BADDR_MASK);
2017
2018 ml_enable_monitor();
2019
2020 set_mmu_ttb(invalid_ttep & TTBR_BADDR_MASK);
2021
2022 flush_mmu_tlb();
2023 #if defined(HAS_VMSA_LOCK)
2024 vmsa_lock();
2025 #endif
2026 kva_active = TRUE;
2027 // global table pointers may need to be different due to physical aperture remapping
2028 cpu_tte = (tt_entry_t*)(phystokv(cpu_ttep));
2029 invalid_tte = (tt_entry_t*)(phystokv(invalid_ttep));
2030
2031 // From here on out, we're off the bootstrap translation tables.
2032
2033
2034 /* AuxKC initialization has to be deferred until this point, since
2035 * the AuxKC may not have been fully mapped in the bootstrap
2036 * tables, if it spilled downwards into the prior L2 block.
2037 *
2038 * Now that its mapping set up by arm_vm_prot_init() is active,
2039 * we can traverse and fix it up.
2040 */
2041
2042 if (arm_vm_auxkc_init()) {
2043 if (segLOWESTROAuxKC < segLOWESTRO) {
2044 segLOWESTRO = segLOWESTROAuxKC;
2045 }
2046 if (segHIGHESTROAuxKC > segHIGHESTRO) {
2047 segHIGHESTRO = segHIGHESTROAuxKC;
2048 }
2049 if (segLOWESTRXAuxKC < segLOWESTTEXT) {
2050 segLOWESTTEXT = segLOWESTRXAuxKC;
2051 }
2052 assert(segLOWEST == segLOWESTAuxKC);
2053
2054 // The preliminary auxKC mapping has been broken up.
2055 flush_mmu_tlb();
2056 }
2057
2058 sane_size = mem_size - (avail_start - gPhysBase);
2059 max_mem = mem_size;
2060 vm_kernel_slid_base = segLOWESTTEXT;
2061 vm_kernel_slid_top = vm_prelink_einfo;
2062 // vm_kernel_slide is set by arm_init()->arm_slide_rebase_and_sign_image()
2063 vm_kernel_stext = segTEXTB;
2064
2065 if (kernel_mach_header_is_in_fileset(&_mh_execute_header)) {
2066 // fileset has kext TEXT before kernel DATA_CONST
2067 assert(segTEXTEXECB == segTEXTB + segSizeTEXT);
2068 vm_kernel_etext = segTEXTB + segSizeTEXT + segSizeTEXTEXEC;
2069 } else {
2070 assert(segDATACONSTB == segTEXTB + segSizeTEXT);
2071 assert(segTEXTEXECB == segDATACONSTB + segSizeDATACONST);
2072 vm_kernel_etext = segTEXTB + segSizeTEXT + segSizeDATACONST + segSizeTEXTEXEC;
2073 }
2074
2075 dynamic_memory_begin = ROUND_TWIG(dynamic_memory_begin);
2076 #if defined(KERNEL_INTEGRITY_CTRR) && defined(CONFIG_XNUPOST)
2077 // reserve a 32MB region without permission overrides to use later for a CTRR unit test
2078 {
2079 extern vm_offset_t ctrr_test_page;
2080 tt_entry_t *new_tte;
2081
2082 ctrr_test_page = dynamic_memory_begin;
2083 dynamic_memory_begin += ARM_TT_L2_SIZE;
2084 cpu_l1_tte = cpu_tte + ((ctrr_test_page & ARM_TT_L1_INDEX_MASK) >> ARM_TT_L1_SHIFT);
2085 assert((*cpu_l1_tte) & ARM_TTE_VALID);
2086 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);
2087 assert((*cpu_l2_tte) == ARM_TTE_EMPTY);
2088 new_tte = (tt_entry_t *)alloc_ptpage(FALSE);
2089 bzero(new_tte, ARM_PGBYTES);
2090 *cpu_l2_tte = (kvtophys((vm_offset_t)new_tte) & ARM_TTE_TABLE_MASK) | ARM_TTE_TYPE_TABLE | ARM_TTE_VALID;
2091 }
2092 #endif /* defined(KERNEL_INTEGRITY_CTRR) && defined(CONFIG_XNUPOST) */
2093 #if XNU_MONITOR
2094 for (vm_offset_t cur = (vm_offset_t)pmap_stacks_start; cur < (vm_offset_t)pmap_stacks_end; cur += ARM_PGBYTES) {
2095 arm_vm_map(cpu_tte, cur, ARM_PTE_EMPTY);
2096 }
2097 #endif
2098 pmap_bootstrap(dynamic_memory_begin);
2099
2100 disable_preemption();
2101
2102 /*
2103 * Initialize l3 page table pages :
2104 * cover this address range:
2105 * 2MB + FrameBuffer size + 10MB for each 256MB segment
2106 */
2107
2108 mem_segments = (mem_size + 0x0FFFFFFF) >> 28;
2109
2110 va_l1 = dynamic_memory_begin;
2111 va_l1_end = va_l1 + ((2 + (mem_segments * 10)) << 20);
2112 va_l1_end += round_page(args->Video.v_height * args->Video.v_rowBytes);
2113 va_l1_end = (va_l1_end + 0x00000000007FFFFFULL) & 0xFFFFFFFFFF800000ULL;
2114
2115 cpu_l1_tte = cpu_tte + ((va_l1 & ARM_TT_L1_INDEX_MASK) >> ARM_TT_L1_SHIFT);
2116
2117 while (va_l1 < va_l1_end) {
2118 va_l2 = va_l1;
2119
2120 if (((va_l1 & ~ARM_TT_L1_OFFMASK) + ARM_TT_L1_SIZE) < va_l1) {
2121 /* If this is the last L1 entry, it must cover the last mapping. */
2122 va_l2_end = va_l1_end;
2123 } else {
2124 va_l2_end = MIN((va_l1 & ~ARM_TT_L1_OFFMASK) + ARM_TT_L1_SIZE, va_l1_end);
2125 }
2126
2127 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);
2128
2129 while (va_l2 < va_l2_end) {
2130 pt_entry_t * ptp;
2131 pmap_paddr_t ptp_phys;
2132
2133 /* Allocate a page and setup L3 Table TTE in L2 */
2134 ptp = (pt_entry_t *) alloc_ptpage(FALSE);
2135 ptp_phys = (pmap_paddr_t)kvtophys((vm_offset_t)ptp);
2136
2137 bzero(ptp, ARM_PGBYTES);
2138 pmap_init_pte_page(kernel_pmap, ptp, va_l2, 3, TRUE);
2139
2140 *cpu_l2_tte = (pa_to_tte(ptp_phys)) | ARM_TTE_TYPE_TABLE | ARM_TTE_VALID | ARM_DYNAMIC_TABLE_XN;
2141
2142 va_l2 += ARM_TT_L2_SIZE;
2143 cpu_l2_tte++;
2144 }
2145
2146 va_l1 = va_l2_end;
2147 cpu_l1_tte++;
2148 }
2149
2150 #if defined(KERNEL_INTEGRITY_KTRR) || defined(KERNEL_INTEGRITY_CTRR)
2151 /*
2152 * In this configuration, the bootstrap mappings (arm_vm_init) and
2153 * the heap mappings occupy separate L1 regions. Explicitly set up
2154 * the heap L1 allocations here.
2155 */
2156 #if defined(ARM_LARGE_MEMORY)
2157 init_ptpages(cpu_tte, KERNEL_PMAP_HEAP_RANGE_START & ~ARM_TT_L1_OFFMASK, VM_MAX_KERNEL_ADDRESS, FALSE, ARM_DYNAMIC_TABLE_XN);
2158 #else // defined(ARM_LARGE_MEMORY)
2159 va_l1 = VM_MIN_KERNEL_ADDRESS & ~ARM_TT_L1_OFFMASK;
2160 init_ptpages(cpu_tte, VM_MIN_KERNEL_ADDRESS & ~ARM_TT_L1_OFFMASK, VM_MAX_KERNEL_ADDRESS, FALSE, ARM_DYNAMIC_TABLE_XN);
2161 #endif // defined(ARM_LARGE_MEMORY)
2162 #endif // defined(KERNEL_INTEGRITY_KTRR) || defined(KERNEL_INTEGRITY_CTRR)
2163
2164 /*
2165 * Initialize l3 page table pages :
2166 * cover this address range:
2167 * ((VM_MAX_KERNEL_ADDRESS & CPUWINDOWS_BASE_MASK) - PE_EARLY_BOOT_VA) to VM_MAX_KERNEL_ADDRESS
2168 */
2169 va_l1 = (VM_MAX_KERNEL_ADDRESS & CPUWINDOWS_BASE_MASK) - PE_EARLY_BOOT_VA;
2170 va_l1_end = VM_MAX_KERNEL_ADDRESS;
2171
2172 cpu_l1_tte = cpu_tte + ((va_l1 & ARM_TT_L1_INDEX_MASK) >> ARM_TT_L1_SHIFT);
2173
2174 while (va_l1 < va_l1_end) {
2175 va_l2 = va_l1;
2176
2177 if (((va_l1 & ~ARM_TT_L1_OFFMASK) + ARM_TT_L1_SIZE) < va_l1) {
2178 /* If this is the last L1 entry, it must cover the last mapping. */
2179 va_l2_end = va_l1_end;
2180 } else {
2181 va_l2_end = MIN((va_l1 & ~ARM_TT_L1_OFFMASK) + ARM_TT_L1_SIZE, va_l1_end);
2182 }
2183
2184 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);
2185
2186 while (va_l2 < va_l2_end) {
2187 pt_entry_t * ptp;
2188 pmap_paddr_t ptp_phys;
2189
2190 /* Allocate a page and setup L3 Table TTE in L2 */
2191 ptp = (pt_entry_t *) alloc_ptpage(FALSE);
2192 ptp_phys = (pmap_paddr_t)kvtophys((vm_offset_t)ptp);
2193
2194 bzero(ptp, ARM_PGBYTES);
2195 pmap_init_pte_page(kernel_pmap, ptp, va_l2, 3, TRUE);
2196
2197 *cpu_l2_tte = (pa_to_tte(ptp_phys)) | ARM_TTE_TYPE_TABLE | ARM_TTE_VALID | ARM_DYNAMIC_TABLE_XN;
2198
2199 va_l2 += ARM_TT_L2_SIZE;
2200 cpu_l2_tte++;
2201 }
2202
2203 va_l1 = va_l2_end;
2204 cpu_l1_tte++;
2205 }
2206
2207
2208 /*
2209 * Adjust avail_start so that the range that the VM owns
2210 * starts on a PAGE_SIZE aligned boundary.
2211 */
2212 avail_start = (avail_start + PAGE_MASK) & ~PAGE_MASK;
2213
2214 #if XNU_MONITOR
2215 pmap_static_allocations_done();
2216 #endif
2217 first_avail = avail_start;
2218 patch_low_glo_static_region(args->topOfKernelData, avail_start - args->topOfKernelData);
2219 enable_preemption();
2220 }
mirror of XNU