xref: /xnu-11417.121.6/osfmk/arm64/arm_vm_init.c (revision a1e26a70f38d1d7daa7b49b258e2f8538ad81650)
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_internal.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 
371 SECURITY_READ_ONLY_LATE(static ptov_table_entry)        ptov_table[PTOV_TABLE_SIZE];
372 SECURITY_READ_ONLY_LATE(static boolean_t)               kva_active = FALSE;
373 
374 #define ARM64_PAGE_UNGUARDED (0)
375 #define ARM64_PAGE_GUARDED   (1)
376 
377 /* "physical to kernel virtual" - given a physical address, return the corresponding physical aperture address */
378 vm_map_address_t
phystokv(pmap_paddr_t pa)379 phystokv(pmap_paddr_t pa)
380 {
381 
382 	for (size_t i = 0; (i < PTOV_TABLE_SIZE) && (ptov_table[i].len != 0); i++) {
383 		if ((pa >= ptov_table[i].pa) && (pa < (ptov_table[i].pa + ptov_table[i].len))) {
384 			return pa - ptov_table[i].pa + ptov_table[i].va;
385 		}
386 	}
387 	if (__improbable((pa < gPhysBase) || ((pa - gPhysBase) >= real_phys_size))) {
388 		panic("%s: illegal PA: 0x%llx; phys base 0x%llx, size 0x%llx", __func__,
389 		    (unsigned long long)pa, (unsigned long long)gPhysBase, (unsigned long long)real_phys_size);
390 	}
391 	return pa - gPhysBase + gVirtBase;
392 }
393 
394 vm_map_address_t
phystokv_range(pmap_paddr_t pa,vm_size_t * max_len)395 phystokv_range(pmap_paddr_t pa, vm_size_t *max_len)
396 {
397 
398 	vm_size_t len;
399 	for (size_t i = 0; (i < PTOV_TABLE_SIZE) && (ptov_table[i].len != 0); i++) {
400 		if ((pa >= ptov_table[i].pa) && (pa < (ptov_table[i].pa + ptov_table[i].len))) {
401 			len = ptov_table[i].len - (pa - ptov_table[i].pa);
402 			if (*max_len > len) {
403 				*max_len = len;
404 			}
405 			return pa - ptov_table[i].pa + ptov_table[i].va;
406 		}
407 	}
408 	len = PAGE_SIZE - (pa & PAGE_MASK);
409 	if (*max_len > len) {
410 		*max_len = len;
411 	}
412 	if (__improbable((pa < gPhysBase) || ((pa - gPhysBase) >= real_phys_size))) {
413 		panic("%s: illegal PA: 0x%llx; phys base 0x%llx, size 0x%llx", __func__,
414 		    (unsigned long long)pa, (unsigned long long)gPhysBase, (unsigned long long)real_phys_size);
415 	}
416 	return pa - gPhysBase + gVirtBase;
417 }
418 
419 vm_offset_t
ml_static_vtop(vm_offset_t va)420 ml_static_vtop(vm_offset_t va)
421 {
422 	for (size_t i = 0; (i < PTOV_TABLE_SIZE) && (ptov_table[i].len != 0); i++) {
423 		if ((va >= ptov_table[i].va) && (va < (ptov_table[i].va + ptov_table[i].len))) {
424 			return va - ptov_table[i].va + ptov_table[i].pa;
425 		}
426 	}
427 	if (__improbable((va < gVirtBase) || (((vm_address_t)(va) - gVirtBase) >= gPhysSize))) {
428 		panic("%s: illegal VA: %p; virt base 0x%llx, size 0x%llx", __func__,
429 		    (void*)va, (unsigned long long)gVirtBase, (unsigned long long)gPhysSize);
430 	}
431 	return (vm_address_t)(va) - gVirtBase + gPhysBase;
432 }
433 
434 /*
435  * This rounds the given address up to the nearest boundary for a PTE contiguous
436  * hint.
437  */
438 static vm_offset_t
round_up_pte_hint_address(vm_offset_t address)439 round_up_pte_hint_address(vm_offset_t address)
440 {
441 	vm_offset_t hint_size = ARM_PTE_SIZE << ARM_PTE_HINT_ENTRIES_SHIFT;
442 	return (address + (hint_size - 1)) & ~(hint_size - 1);
443 }
444 
445 /* allocate a page for a page table: we support static and dynamic mappings.
446  *
447  * returns a virtual address for the allocated page
448  *
449  * for static mappings, we allocate from the region ropagetable_begin to ro_pagetable_end-1,
450  * which is defined in the DATA_CONST segment and will be protected RNX when vm_prot_finalize runs.
451  *
452  * for dynamic mappings, we allocate from avail_start, which should remain RWNX.
453  */
454 
455 vm_offset_t
alloc_ptpage(boolean_t map_static)456 alloc_ptpage(boolean_t map_static)
457 {
458 	vm_offset_t vaddr;
459 
460 #if !(defined(KERNEL_INTEGRITY_KTRR) || defined(KERNEL_INTEGRITY_CTRR))
461 	map_static = FALSE;
462 #endif
463 
464 	if (!ropage_next) {
465 		ropage_next = (vm_offset_t)&ropagetable_begin;
466 	}
467 
468 	if (map_static) {
469 		assert(ropage_next < (vm_offset_t)&ropagetable_end);
470 
471 		vaddr = ropage_next;
472 		ropage_next += ARM_PGBYTES;
473 
474 		return vaddr;
475 	} else {
476 		vaddr = phystokv(avail_start);
477 		avail_start += ARM_PGBYTES;
478 
479 		return vaddr;
480 	}
481 }
482 
483 #if DEBUG
484 
485 void dump_kva_l2(vm_offset_t tt_base, tt_entry_t *tt, int indent, uint64_t *rosz_out, uint64_t *rwsz_out);
486 
487 void
dump_kva_l2(vm_offset_t tt_base,tt_entry_t * tt,int indent,uint64_t * rosz_out,uint64_t * rwsz_out)488 dump_kva_l2(vm_offset_t tt_base, tt_entry_t *tt, int indent, uint64_t *rosz_out, uint64_t *rwsz_out)
489 {
490 	unsigned int i;
491 	boolean_t cur_ro, prev_ro = 0;
492 	int start_entry = -1;
493 	tt_entry_t cur, prev = 0;
494 	pmap_paddr_t robegin = kvtophys((vm_offset_t)&ropagetable_begin);
495 	pmap_paddr_t roend = kvtophys((vm_offset_t)&ropagetable_end);
496 	boolean_t tt_static = kvtophys((vm_offset_t)tt) >= robegin &&
497 	    kvtophys((vm_offset_t)tt) < roend;
498 
499 	for (i = 0; i < TTE_PGENTRIES; i++) {
500 		int tte_type = tt[i] & ARM_TTE_TYPE_MASK;
501 		cur = tt[i] & ARM_TTE_TABLE_MASK;
502 
503 		if (tt_static) {
504 			/* addresses mapped by this entry are static if it is a block mapping,
505 			 * or the table was allocated from the RO page table region */
506 			cur_ro = (tte_type == ARM_TTE_TYPE_BLOCK) || (cur >= robegin && cur < roend);
507 		} else {
508 			cur_ro = 0;
509 		}
510 
511 		if ((cur == 0 && prev != 0) || (cur_ro != prev_ro && prev != 0)) { // falling edge
512 			uintptr_t start, end, sz;
513 
514 			start = (uintptr_t)start_entry << ARM_TT_L2_SHIFT;
515 			start += tt_base;
516 			end = ((uintptr_t)i << ARM_TT_L2_SHIFT) - 1;
517 			end += tt_base;
518 
519 			sz = end - start + 1;
520 			printf("%*s0x%08x_%08x-0x%08x_%08x %s (%luMB)\n",
521 			    indent * 4, "",
522 			    (uint32_t)(start >> 32), (uint32_t)start,
523 			    (uint32_t)(end >> 32), (uint32_t)end,
524 			    prev_ro ? "Static " : "Dynamic",
525 			    (sz >> 20));
526 
527 			if (prev_ro) {
528 				*rosz_out += sz;
529 			} else {
530 				*rwsz_out += sz;
531 			}
532 		}
533 
534 		if ((prev == 0 && cur != 0) || cur_ro != prev_ro) { // rising edge: set start
535 			start_entry = i;
536 		}
537 
538 		prev = cur;
539 		prev_ro = cur_ro;
540 	}
541 }
542 
543 void
dump_kva_space()544 dump_kva_space()
545 {
546 	uint64_t tot_rosz = 0, tot_rwsz = 0;
547 	int ro_ptpages, rw_ptpages;
548 	pmap_paddr_t robegin = kvtophys((vm_offset_t)&ropagetable_begin);
549 	pmap_paddr_t roend = kvtophys((vm_offset_t)&ropagetable_end);
550 	boolean_t root_static = kvtophys((vm_offset_t)cpu_tte) >= robegin &&
551 	    kvtophys((vm_offset_t)cpu_tte) < roend;
552 	uint64_t kva_base = ~((1ULL << (64 - T1SZ_BOOT)) - 1);
553 
554 	printf("Root page table: %s\n", root_static ? "Static" : "Dynamic");
555 
556 	for (unsigned int i = 0; i < TTE_PGENTRIES; i++) {
557 		pmap_paddr_t cur;
558 		boolean_t cur_ro;
559 		uintptr_t start, end;
560 		uint64_t rosz = 0, rwsz = 0;
561 
562 		if ((cpu_tte[i] & ARM_TTE_VALID) == 0) {
563 			continue;
564 		}
565 
566 		cur = cpu_tte[i] & ARM_TTE_TABLE_MASK;
567 		start = (uint64_t)i << ARM_TT_L1_SHIFT;
568 		start = start + kva_base;
569 		end = start + (ARM_TT_L1_SIZE - 1);
570 		cur_ro = cur >= robegin && cur < roend;
571 
572 		printf("0x%08x_%08x-0x%08x_%08x %s\n",
573 		    (uint32_t)(start >> 32), (uint32_t)start,
574 		    (uint32_t)(end >> 32), (uint32_t)end,
575 		    cur_ro ? "Static " : "Dynamic");
576 
577 		dump_kva_l2(start, (tt_entry_t*)phystokv(cur), 1, &rosz, &rwsz);
578 		tot_rosz += rosz;
579 		tot_rwsz += rwsz;
580 	}
581 
582 	printf("L2 Address space mapped: Static %lluMB Dynamic %lluMB Total %lluMB\n",
583 	    tot_rosz >> 20,
584 	    tot_rwsz >> 20,
585 	    (tot_rosz >> 20) + (tot_rwsz >> 20));
586 
587 	ro_ptpages = (int)((ropage_next - (vm_offset_t)&ropagetable_begin) >> ARM_PGSHIFT);
588 	rw_ptpages = (int)(lowGlo.lgStaticSize  >> ARM_PGSHIFT);
589 	printf("Pages used: static %d dynamic %d\n", ro_ptpages, rw_ptpages);
590 }
591 
592 #endif /* DEBUG */
593 
594 #if __ARM_KERNEL_PROTECT__ || XNU_MONITOR
595 /*
596  * arm_vm_map:
597  *   root_ttp: The kernel virtual address for the root of the target page tables
598  *   vaddr: The target virtual address
599  *   pte: A page table entry value (may be ARM_PTE_EMPTY)
600  *
601  * This function installs pte at vaddr in root_ttp.  Any page table pages needed
602  * to install pte will be allocated by this function.
603  */
604 static void
arm_vm_map(tt_entry_t * root_ttp,vm_offset_t vaddr,pt_entry_t pte)605 arm_vm_map(tt_entry_t * root_ttp, vm_offset_t vaddr, pt_entry_t pte)
606 {
607 	vm_offset_t ptpage = 0;
608 	tt_entry_t * ttp = root_ttp;
609 
610 	tt_entry_t * l1_ttep = NULL;
611 	tt_entry_t l1_tte = 0;
612 
613 	tt_entry_t * l2_ttep = NULL;
614 	tt_entry_t l2_tte = 0;
615 	pt_entry_t * ptep = NULL;
616 	pt_entry_t cpte = 0;
617 
618 	/*
619 	 * Walk the target page table to find the PTE for the given virtual
620 	 * address.  Allocate any page table pages needed to do this.
621 	 */
622 	l1_ttep = ttp + ((vaddr & ARM_TT_L1_INDEX_MASK) >> ARM_TT_L1_SHIFT);
623 	l1_tte = *l1_ttep;
624 
625 	if (l1_tte == ARM_TTE_EMPTY) {
626 		ptpage = alloc_ptpage(TRUE);
627 		bzero((void *)ptpage, ARM_PGBYTES);
628 		l1_tte = kvtophys(ptpage);
629 		l1_tte &= ARM_TTE_TABLE_MASK;
630 		l1_tte |= ARM_TTE_VALID | ARM_TTE_TYPE_TABLE | ARM_TTE_TABLE_AP(ARM_TTE_TABLE_AP_USER_NA);
631 		*l1_ttep = l1_tte;
632 		ptpage = 0;
633 	}
634 
635 	ttp = (tt_entry_t *)phystokv(l1_tte & ARM_TTE_TABLE_MASK);
636 
637 	l2_ttep = ttp + ((vaddr & ARM_TT_L2_INDEX_MASK) >> ARM_TT_L2_SHIFT);
638 	l2_tte = *l2_ttep;
639 
640 	if (l2_tte == ARM_TTE_EMPTY) {
641 		ptpage = alloc_ptpage(TRUE);
642 		bzero((void *)ptpage, ARM_PGBYTES);
643 		l2_tte = kvtophys(ptpage);
644 		l2_tte &= ARM_TTE_TABLE_MASK;
645 		l2_tte |= ARM_TTE_VALID | ARM_TTE_TYPE_TABLE;
646 		*l2_ttep = l2_tte;
647 		ptpage = 0;
648 	}
649 
650 	ttp = (tt_entry_t *)phystokv(l2_tte & ARM_TTE_TABLE_MASK);
651 
652 	ptep = ttp + ((vaddr & ARM_TT_L3_INDEX_MASK) >> ARM_TT_L3_SHIFT);
653 	cpte = *ptep;
654 
655 	/*
656 	 * If the existing PTE is not empty, then we are replacing a valid
657 	 * mapping.
658 	 */
659 	if (cpte != ARM_PTE_EMPTY) {
660 		panic("%s: cpte=%#llx is not empty, "
661 		    "vaddr=%#lx, pte=%#llx",
662 		    __FUNCTION__, cpte,
663 		    vaddr, pte);
664 	}
665 
666 	*ptep = pte;
667 }
668 
669 #endif // __ARM_KERNEL_PROTECT || XNU_MONITOR
670 
671 #if __ARM_KERNEL_PROTECT__
672 
673 /*
674  * arm_vm_kernel_el0_map:
675  *   vaddr: The target virtual address
676  *   pte: A page table entry value (may be ARM_PTE_EMPTY)
677  *
678  * This function installs pte at vaddr for the EL0 kernel mappings.
679  */
680 static void
arm_vm_kernel_el0_map(vm_offset_t vaddr,pt_entry_t pte)681 arm_vm_kernel_el0_map(vm_offset_t vaddr, pt_entry_t pte)
682 {
683 	/* Calculate where vaddr will be in the EL1 kernel page tables. */
684 	vm_offset_t kernel_pmap_vaddr = vaddr - ((ARM_TT_ROOT_INDEX_MASK + ARM_TT_ROOT_SIZE) / 2ULL);
685 	arm_vm_map(cpu_tte, kernel_pmap_vaddr, pte);
686 }
687 
688 /*
689  * arm_vm_kernel_el1_map:
690  *   vaddr: The target virtual address
691  *   pte: A page table entry value (may be ARM_PTE_EMPTY)
692  *
693  * This function installs pte at vaddr for the EL1 kernel mappings.
694  */
695 static void
arm_vm_kernel_el1_map(vm_offset_t vaddr,pt_entry_t pte)696 arm_vm_kernel_el1_map(vm_offset_t vaddr, pt_entry_t pte)
697 {
698 	arm_vm_map(cpu_tte, vaddr, pte);
699 }
700 
701 /*
702  * arm_vm_kernel_pte:
703  *   vaddr: The target virtual address
704  *
705  * This function returns the PTE value for the given vaddr from the kernel page
706  * tables.  If the region has been been block mapped, we return what an
707  * equivalent PTE value would be (as regards permissions and flags).  We also
708  * remove the HINT bit (as we are not necessarily creating contiguous mappings.
709  */
710 static pt_entry_t
arm_vm_kernel_pte(vm_offset_t vaddr)711 arm_vm_kernel_pte(vm_offset_t vaddr)
712 {
713 	tt_entry_t * ttp = cpu_tte;
714 	tt_entry_t * ttep = NULL;
715 	tt_entry_t tte = 0;
716 	pt_entry_t * ptep = NULL;
717 	pt_entry_t pte = 0;
718 
719 	ttep = ttp + ((vaddr & ARM_TT_L1_INDEX_MASK) >> ARM_TT_L1_SHIFT);
720 	tte = *ttep;
721 
722 	assert(tte & ARM_TTE_VALID);
723 
724 	if ((tte & ARM_TTE_TYPE_MASK) == ARM_TTE_TYPE_BLOCK) {
725 		/* This is a block mapping; return the equivalent PTE value. */
726 		pte = (pt_entry_t)(tte & ~ARM_TTE_TYPE_MASK);
727 		pte |= ARM_PTE_TYPE_VALID;
728 		pte |= vaddr & ((ARM_TT_L1_SIZE - 1) & ARM_PTE_PAGE_MASK);
729 		pte &= ~ARM_PTE_HINT_MASK;
730 		return pte;
731 	}
732 
733 	ttp = (tt_entry_t *)phystokv(tte & ARM_TTE_TABLE_MASK);
734 	ttep = ttp + ((vaddr & ARM_TT_L2_INDEX_MASK) >> ARM_TT_L2_SHIFT);
735 	tte = *ttep;
736 
737 	assert(tte & ARM_TTE_VALID);
738 
739 	if ((tte & ARM_TTE_TYPE_MASK) == ARM_TTE_TYPE_BLOCK) {
740 		/* This is a block mapping; return the equivalent PTE value. */
741 		pte = (pt_entry_t)(tte & ~ARM_TTE_TYPE_MASK);
742 		pte |= ARM_PTE_TYPE_VALID;
743 		pte |= vaddr & ((ARM_TT_L2_SIZE - 1) & ARM_PTE_PAGE_MASK);
744 		pte &= ~ARM_PTE_HINT_MASK;
745 		return pte;
746 	}
747 
748 	ttp = (tt_entry_t *)phystokv(tte & ARM_TTE_TABLE_MASK);
749 
750 	ptep = ttp + ((vaddr & ARM_TT_L3_INDEX_MASK) >> ARM_TT_L3_SHIFT);
751 	pte = *ptep;
752 	pte &= ~ARM_PTE_HINT_MASK;
753 	return pte;
754 }
755 
756 /*
757  * arm_vm_prepare_kernel_el0_mappings:
758  *   alloc_only: Indicates if PTE values should be copied from the EL1 kernel
759  *     mappings.
760  *
761  * This function expands the kernel page tables to support the EL0 kernel
762  * mappings, and conditionally installs the PTE values for the EL0 kernel
763  * mappings (if alloc_only is false).
764  */
765 static void
arm_vm_prepare_kernel_el0_mappings(bool alloc_only)766 arm_vm_prepare_kernel_el0_mappings(bool alloc_only)
767 {
768 	pt_entry_t pte = 0;
769 	vm_offset_t start = ((vm_offset_t)&ExceptionVectorsBase) & ~PAGE_MASK;
770 	vm_offset_t end = (((vm_offset_t)&ExceptionVectorsEnd) + PAGE_MASK) & ~PAGE_MASK;
771 	vm_offset_t cur = 0;
772 	vm_offset_t cur_fixed = 0;
773 
774 	/* Expand for/map the exceptions vectors in the EL0 kernel mappings. */
775 	for (cur = start, cur_fixed = ARM_KERNEL_PROTECT_EXCEPTION_START; cur < end; cur += ARM_PGBYTES, cur_fixed += ARM_PGBYTES) {
776 		/*
777 		 * We map the exception vectors at a different address than that
778 		 * of the kernelcache to avoid sharing page table pages with the
779 		 * kernelcache (as this may cause issues with TLB caching of
780 		 * page table pages.
781 		 */
782 		if (!alloc_only) {
783 			pte = arm_vm_kernel_pte(cur);
784 		}
785 
786 		arm_vm_kernel_el1_map(cur_fixed, pte);
787 		arm_vm_kernel_el0_map(cur_fixed, pte);
788 	}
789 
790 	__builtin_arm_dmb(DMB_ISH);
791 	__builtin_arm_isb(ISB_SY);
792 
793 	if (!alloc_only) {
794 		/*
795 		 * If we have created the alternate exception vector mappings,
796 		 * the boot CPU may now switch over to them.
797 		 */
798 		set_vbar_el1(ARM_KERNEL_PROTECT_EXCEPTION_START);
799 		__builtin_arm_isb(ISB_SY);
800 	}
801 }
802 
803 /*
804  * arm_vm_populate_kernel_el0_mappings:
805  *
806  * This function adds all required mappings to the EL0 kernel mappings.
807  */
808 static void
arm_vm_populate_kernel_el0_mappings(void)809 arm_vm_populate_kernel_el0_mappings(void)
810 {
811 	arm_vm_prepare_kernel_el0_mappings(FALSE);
812 }
813 
814 /*
815  * arm_vm_expand_kernel_el0_mappings:
816  *
817  * This function expands the kernel page tables to accomodate the EL0 kernel
818  * mappings.
819  */
820 static void
arm_vm_expand_kernel_el0_mappings(void)821 arm_vm_expand_kernel_el0_mappings(void)
822 {
823 	arm_vm_prepare_kernel_el0_mappings(TRUE);
824 }
825 #endif /* __ARM_KERNEL_PROTECT__ */
826 
827 #if defined(KERNEL_INTEGRITY_KTRR) || defined(KERNEL_INTEGRITY_CTRR)
828 extern void bootstrap_instructions;
829 
830 /*
831  * arm_replace_identity_map takes the V=P map that we construct in start.s
832  * and repurposes it in order to have it map only the page we need in order
833  * to turn on the MMU.  This prevents us from running into issues where
834  * KTRR will cause us to fault on executable block mappings that cross the
835  * KTRR boundary.
836  */
837 static void
arm_replace_identity_map(void)838 arm_replace_identity_map(void)
839 {
840 	vm_offset_t addr;
841 	pmap_paddr_t paddr;
842 
843 	pmap_paddr_t l1_ptp_phys = 0;
844 	tt_entry_t *l1_ptp_virt = NULL;
845 	tt_entry_t *tte1 = NULL;
846 	pmap_paddr_t l2_ptp_phys = 0;
847 	tt_entry_t *l2_ptp_virt = NULL;
848 	tt_entry_t *tte2 = NULL;
849 	pmap_paddr_t l3_ptp_phys = 0;
850 	pt_entry_t *l3_ptp_virt = NULL;
851 	pt_entry_t *ptep = NULL;
852 
853 	addr = ((vm_offset_t)&bootstrap_instructions) & ~ARM_PGMASK;
854 	paddr = kvtophys(addr);
855 
856 	/*
857 	 * Grab references to the V=P page tables, and allocate an L3 page.
858 	 */
859 	l1_ptp_phys = kvtophys((vm_offset_t)&bootstrap_pagetables);
860 	l1_ptp_virt = (tt_entry_t *)phystokv(l1_ptp_phys);
861 	tte1 = &l1_ptp_virt[L1_TABLE_INDEX(paddr)];
862 
863 	l2_ptp_virt = L2_TABLE_VA(tte1);
864 	l2_ptp_phys = (*tte1) & ARM_TTE_TABLE_MASK;
865 	tte2 = &l2_ptp_virt[L2_TABLE_INDEX(paddr)];
866 
867 	l3_ptp_virt = (pt_entry_t *)alloc_ptpage(TRUE);
868 	l3_ptp_phys = kvtophys((vm_offset_t)l3_ptp_virt);
869 	ptep = &l3_ptp_virt[L3_TABLE_INDEX(paddr)];
870 
871 	/*
872 	 * Replace the large V=P mapping with a mapping that provides only the
873 	 * mappings needed to turn on the MMU.
874 	 */
875 
876 	bzero(l1_ptp_virt, ARM_PGBYTES);
877 	*tte1 = ARM_TTE_BOOT_TABLE | (l2_ptp_phys & ARM_TTE_TABLE_MASK);
878 
879 	bzero(l2_ptp_virt, ARM_PGBYTES);
880 	*tte2 = ARM_TTE_BOOT_TABLE | (l3_ptp_phys & ARM_TTE_TABLE_MASK);
881 
882 	*ptep = (paddr & ARM_PTE_MASK) |
883 	    ARM_PTE_TYPE_VALID |
884 	    ARM_PTE_SH(SH_OUTER_MEMORY) |
885 	    ARM_PTE_ATTRINDX(CACHE_ATTRINDX_WRITEBACK) |
886 	    ARM_PTE_AF |
887 	    ARM_PTE_AP(AP_RONA) |
888 	    ARM_PTE_NX;
889 }
890 #endif /* defined(KERNEL_INTEGRITY_KTRR) || defined(KERNEL_INTEGRITY_CTRR) */
891 
892 tt_entry_t *arm_kva_to_tte(vm_offset_t);
893 
894 tt_entry_t *
arm_kva_to_tte(vm_offset_t va)895 arm_kva_to_tte(vm_offset_t va)
896 {
897 	tt_entry_t *tte1, *tte2;
898 	tte1 = cpu_tte + L1_TABLE_INDEX(va);
899 	tte2 = L2_TABLE_VA(tte1) + L2_TABLE_INDEX(va);
900 
901 	return tte2;
902 }
903 
904 #if XNU_MONITOR
905 
906 static inline pt_entry_t *
arm_kva_to_pte(vm_offset_t va)907 arm_kva_to_pte(vm_offset_t va)
908 {
909 	tt_entry_t *tte2 = arm_kva_to_tte(va);
910 	return L3_TABLE_VA(tte2) + L3_TABLE_INDEX(va);
911 }
912 
913 #endif
914 
915 #define ARM64_GRANULE_ALLOW_BLOCK (1 << 0)
916 #define ARM64_GRANULE_ALLOW_HINT (1 << 1)
917 
918 /**
919  * Updates a translation table entry (TTE) with the supplied value, unless doing so might render
920  * the pagetable region read-only before subsequent updates have finished.  In that case, the TTE
921  * value will be saved off for deferred processing.
922  *
923  * @param ttep address of the TTE to update
924  * @param entry the value to store in ttep
925  * @param pa the base physical address mapped by the TTE
926  * @param ttebase L3-page- or L2-block-aligned base virtual address of the pagetable region
927  * @param granule mask indicating whether L2 block or L3 hint mappings are allowed for this segment
928  * @param deferred_ttep_pair 2-element array of addresses of deferred TTEs
929  * @param deferred_tte_pair 2-element array containing TTE values for deferred assignment to
930  *        corresponding elements of deferred_ttep_pair
931  */
932 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)933 update_or_defer_tte(tt_entry_t *ttep, tt_entry_t entry, pmap_paddr_t pa, vm_map_address_t ttebase,
934     unsigned granule __unused, tt_entry_t **deferred_ttep_pair, tt_entry_t *deferred_tte_pair)
935 {
936 	/*
937 	 * If we're trying to assign an entry that maps the current TTE region (identified by ttebase),
938 	 * and the pagetable is already live (indicated by kva_active), defer assignment of the current
939 	 * entry and possibly the entry after it until all other mappings in the segment have been
940 	 * updated.  Otherwise we may end up immediately marking the pagetable region read-only
941 	 * leading to a fault later on a later assignment if we manage to outrun the TLB.  This can
942 	 * happen on KTRR/CTRR-enabled devices when marking segDATACONST read-only, as the pagetables
943 	 * that map that segment must come from the segment itself.  We therefore store the initial
944 	 * recursive TTE in deferred_ttep_pair[0] and its value in deferred_tte_pair[0].  We may also
945 	 * defer assignment of the TTE following that recursive TTE and store its value in
946 	 * deferred_tte_pair[1], because the TTE region following the current one may also contain
947 	 * pagetables and we must avoid marking that region read-only before updating those tables.
948 	 *
949 	 * We require that such recursive mappings must exist in regions that can be mapped with L2
950 	 * block entries if they are sufficiently large.  This is what allows us to assume that no
951 	 * more than 2 deferred TTEs will be required, because:
952 	 * 	--If more than 2 adjacent L3 PTEs were required to map our pagetables, that would mean
953 	 * 	  we would have at least one full L3 pagetable page and would instead use an L2 block.
954 	 *	--If more than 2 adjacent L2 blocks were required to map our pagetables, that would
955 	 * 	  mean we would have at least one full L2-block-sized region of TTEs and something
956 	 *	  is very wrong because no segment should be that large.
957 	 */
958 	if ((deferred_ttep_pair != NULL) && (deferred_ttep_pair[0] != NULL) && (ttep == (deferred_ttep_pair[0] + 1))) {
959 		assert(deferred_tte_pair[1] == 0);
960 		deferred_ttep_pair[1] = ttep;
961 		deferred_tte_pair[1] = entry;
962 	} else if (kva_active && (phystokv(pa) == ttebase)) {
963 		assert(deferred_ttep_pair != NULL);
964 		assert(granule & ARM64_GRANULE_ALLOW_BLOCK);
965 		if (deferred_ttep_pair[0] == NULL) {
966 			deferred_ttep_pair[0] = ttep;
967 			deferred_tte_pair[0] = entry;
968 		} else {
969 			assert(deferred_ttep_pair[1] == NULL);
970 			deferred_ttep_pair[1] = ttep;
971 			deferred_tte_pair[1] = entry;
972 		}
973 	} else {
974 		*ttep = entry;
975 	}
976 }
977 
978 /*
979  * arm_vm_page_granular_helper updates protections at the L3 level.  It will (if
980  * neccessary) allocate a page for the L3 table and update the corresponding L2
981  * entry.  Then, it will iterate over the L3 table, updating protections as necessary.
982  * This expects to be invoked on a L2 entry or sub L2 entry granularity, so this should
983  * not be invoked from a context that does not do L2 iteration separately (basically,
984  * don't call this except from arm_vm_page_granular_prot).
985  *
986  * unsigned granule: 0 => force to page granule, or a combination of
987  * ARM64_GRANULE_* flags declared above.
988  *
989  * unsigned int guarded => flag indicating whether this range should be
990  * considered an ARM "guarded" page. This enables BTI enforcement for a region.
991  */
992 
993 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,__unused unsigned int guarded,tt_entry_t ** deferred_ttep_pair,tt_entry_t * deferred_tte_pair)994 arm_vm_page_granular_helper(vm_offset_t start, vm_offset_t _end, vm_offset_t va, pmap_paddr_t pa_offset,
995     int pte_prot_APX, int pte_prot_XN, unsigned granule, __unused unsigned int guarded,
996     tt_entry_t **deferred_ttep_pair, tt_entry_t *deferred_tte_pair)
997 {
998 	if (va & ARM_TT_L2_OFFMASK) { /* ragged edge hanging over a ARM_TT_L2_SIZE  boundary */
999 		tt_entry_t *tte2;
1000 		tt_entry_t tmplate;
1001 		pmap_paddr_t pa;
1002 		pt_entry_t *ppte, ptmp;
1003 		addr64_t ppte_phys;
1004 		unsigned i;
1005 
1006 		va &= ~ARM_TT_L2_OFFMASK;
1007 		pa = va - gVirtBase + gPhysBase - pa_offset;
1008 
1009 		if (pa >= real_avail_end) {
1010 			return;
1011 		}
1012 
1013 		tte2 = arm_kva_to_tte(va);
1014 
1015 		assert(_end >= va);
1016 		tmplate = *tte2;
1017 
1018 		if (ARM_TTE_TYPE_TABLE == (tmplate & ARM_TTE_TYPE_MASK)) {
1019 			/* pick up the existing page table. */
1020 			ppte = (pt_entry_t *)phystokv((tmplate & ARM_TTE_TABLE_MASK));
1021 		} else {
1022 			// TTE must be reincarnated with page level mappings.
1023 
1024 			// ... but we don't want to break up blocks on live
1025 			// translation tables.
1026 			assert(!kva_active);
1027 
1028 			ppte = (pt_entry_t*)alloc_ptpage(pa_offset == 0);
1029 			bzero(ppte, ARM_PGBYTES);
1030 			ppte_phys = kvtophys((vm_offset_t)ppte);
1031 
1032 			*tte2 = pa_to_tte(ppte_phys) | ARM_TTE_TYPE_TABLE | ARM_TTE_VALID;
1033 		}
1034 
1035 		vm_offset_t len = _end - va;
1036 		if ((pa + len) > real_avail_end) {
1037 			_end -= (pa + len - real_avail_end);
1038 		}
1039 		assert((start - gVirtBase + gPhysBase - pa_offset) >= gPhysBase);
1040 
1041 		/* Round up to the nearest PAGE_SIZE boundary when creating mappings:
1042 		 * PAGE_SIZE may be a multiple of ARM_PGBYTES, and we don't want to leave
1043 		 * a ragged non-PAGE_SIZE-aligned edge. */
1044 		vm_offset_t rounded_end = round_page(_end);
1045 		/* Apply the desired protections to the specified page range */
1046 		for (i = 0; i <= (ARM_TT_L3_INDEX_MASK >> ARM_TT_L3_SHIFT); i++) {
1047 			if ((start <= va) && (va < rounded_end)) {
1048 				ptmp = pa | ARM_PTE_AF | ARM_PTE_SH(SH_OUTER_MEMORY) | ARM_PTE_TYPE_VALID;
1049 				ptmp = ptmp | ARM_PTE_ATTRINDX(CACHE_ATTRINDX_DEFAULT);
1050 				ptmp = ptmp | ARM_PTE_AP(pte_prot_APX);
1051 				ptmp = ptmp | ARM_PTE_NX;
1052 #if __ARM_KERNEL_PROTECT__
1053 				ptmp = ptmp | ARM_PTE_NG;
1054 #endif /* __ARM_KERNEL_PROTECT__ */
1055 
1056 				if (pte_prot_XN) {
1057 					ptmp = ptmp | ARM_PTE_PNX;
1058 				}
1059 
1060 				/*
1061 				 * If we can, apply the contiguous hint to this range.  The hint is
1062 				 * applicable if the current address falls within a hint-sized range that will
1063 				 * be fully covered by this mapping request.
1064 				 */
1065 				if ((va >= round_up_pte_hint_address(start)) && (round_up_pte_hint_address(va + 1) <= _end) &&
1066 				    (granule & ARM64_GRANULE_ALLOW_HINT) && use_contiguous_hint) {
1067 					assert((va & ((1 << ARM_PTE_HINT_ADDR_SHIFT) - 1)) == ((pa & ((1 << ARM_PTE_HINT_ADDR_SHIFT) - 1))));
1068 					ptmp |= ARM_PTE_HINT;
1069 					/* Do not attempt to reapply the hint bit to an already-active mapping.
1070 					 * This very likely means we're attempting to change attributes on an already-active mapping,
1071 					 * which violates the requirement of the hint bit.*/
1072 					assert(!kva_active || (ppte[i] == ARM_PTE_EMPTY));
1073 				}
1074 
1075 #if BTI_ENFORCED
1076 				/*
1077 				 * Set the 'guarded page' flag to enable ARM BTI enforcement.
1078 				 */
1079 				if (guarded) {
1080 					ptmp |= ARM_PTE_GP;
1081 				}
1082 #endif /* BTI_ENFORCED */
1083 				/*
1084 				 * Do not change the contiguous bit on an active mapping.  Even in a single-threaded
1085 				 * environment, it's possible for prefetch to produce a TLB conflict by trying to pull in
1086 				 * a hint-sized entry on top of one or more existing page-sized entries.  It's also useful
1087 				 * to make sure we're not trying to unhint a sub-range of a larger hinted range, which
1088 				 * could produce a later TLB conflict.
1089 				 */
1090 				assert(!kva_active || (ppte[i] == ARM_PTE_EMPTY) || ((ppte[i] & ARM_PTE_HINT) == (ptmp & ARM_PTE_HINT)));
1091 
1092 				update_or_defer_tte(&ppte[i], ptmp, pa, (vm_map_address_t)ppte, granule, deferred_ttep_pair, deferred_tte_pair);
1093 			}
1094 
1095 			va += ARM_PGBYTES;
1096 			pa += ARM_PGBYTES;
1097 		}
1098 	}
1099 }
1100 
1101 /*
1102  * arm_vm_page_granular_prot updates protections by iterating over the L2 entries and
1103  * changing them.  If a particular chunk necessitates L3 entries (for reasons of
1104  * alignment or length, or an explicit request that the entry be fully expanded), we
1105  * hand off to arm_vm_page_granular_helper to deal with the L3 chunk of the logic.
1106  */
1107 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,unsigned int guarded)1108 arm_vm_page_granular_prot(vm_offset_t start, unsigned long size, pmap_paddr_t pa_offset,
1109     int tte_prot_XN, int pte_prot_APX, int pte_prot_XN,
1110     unsigned granule, unsigned int guarded)
1111 {
1112 	tt_entry_t *deferred_ttep_pair[2] = {NULL};
1113 	tt_entry_t deferred_tte_pair[2] = {0};
1114 	vm_offset_t _end = start + size;
1115 	vm_offset_t align_start = (start + ARM_TT_L2_OFFMASK) & ~ARM_TT_L2_OFFMASK;
1116 
1117 	if (size == 0x0UL) {
1118 		return;
1119 	}
1120 
1121 	if (align_start > _end) {
1122 		align_start = _end;
1123 	}
1124 
1125 	arm_vm_page_granular_helper(start, align_start, start, pa_offset, pte_prot_APX, pte_prot_XN, granule, guarded, deferred_ttep_pair, deferred_tte_pair);
1126 
1127 	while ((_end - align_start) >= ARM_TT_L2_SIZE) {
1128 		if (!(granule & ARM64_GRANULE_ALLOW_BLOCK)) {
1129 			arm_vm_page_granular_helper(align_start, align_start + ARM_TT_L2_SIZE, align_start + 1, pa_offset,
1130 			    pte_prot_APX, pte_prot_XN, granule, guarded, deferred_ttep_pair, deferred_tte_pair);
1131 		} else {
1132 			pmap_paddr_t pa = align_start - gVirtBase + gPhysBase - pa_offset;
1133 			assert((pa & ARM_TT_L2_OFFMASK) == 0);
1134 			tt_entry_t *tte2;
1135 			tt_entry_t tmplate;
1136 
1137 			tte2 = arm_kva_to_tte(align_start);
1138 
1139 			if ((pa >= gPhysBase) && (pa < real_avail_end)) {
1140 				tmplate = (pa & ARM_TTE_BLOCK_L2_MASK) | ARM_TTE_TYPE_BLOCK
1141 				    | ARM_TTE_VALID | ARM_TTE_BLOCK_AF | ARM_TTE_BLOCK_NX
1142 				    | ARM_TTE_BLOCK_AP(pte_prot_APX) | ARM_TTE_BLOCK_SH(SH_OUTER_MEMORY)
1143 				    | ARM_TTE_BLOCK_ATTRINDX(CACHE_ATTRINDX_WRITEBACK);
1144 
1145 #if __ARM_KERNEL_PROTECT__
1146 				tmplate = tmplate | ARM_TTE_BLOCK_NG;
1147 #endif /* __ARM_KERNEL_PROTECT__ */
1148 				if (tte_prot_XN) {
1149 					tmplate = tmplate | ARM_TTE_BLOCK_PNX;
1150 				}
1151 
1152 				update_or_defer_tte(tte2, tmplate, pa, (vm_map_address_t)tte2 & ~ARM_TT_L2_OFFMASK,
1153 				    granule, deferred_ttep_pair, deferred_tte_pair);
1154 			}
1155 		}
1156 		align_start += ARM_TT_L2_SIZE;
1157 	}
1158 
1159 	if (align_start < _end) {
1160 		arm_vm_page_granular_helper(align_start, _end, _end, pa_offset, pte_prot_APX, pte_prot_XN, granule, guarded, deferred_ttep_pair, deferred_tte_pair);
1161 	}
1162 
1163 	if (deferred_ttep_pair[0] != NULL) {
1164 #if DEBUG || DEVELOPMENT
1165 		/*
1166 		 * Flush the TLB to catch bugs that might cause us to prematurely revoke write access from the pagetable page.
1167 		 * These bugs may otherwise be hidden by TLB entries in most cases, resulting in very rare panics.
1168 		 * Note that we always flush the TLB at the end of arm_vm_prot_finalize().
1169 		 */
1170 		flush_mmu_tlb();
1171 #endif
1172 		/*
1173 		 * The first TTE in the pair is a recursive mapping of the pagetable region, so we must update it last
1174 		 * to avoid potentially marking deferred_pte_pair[1] read-only.
1175 		 */
1176 		if (deferred_tte_pair[1] != 0) {
1177 			os_atomic_store(deferred_ttep_pair[1], deferred_tte_pair[1], release);
1178 		}
1179 		os_atomic_store(deferred_ttep_pair[0], deferred_tte_pair[0], release);
1180 	}
1181 }
1182 
1183 static inline void
arm_vm_page_granular_RNX(vm_offset_t start,unsigned long size,unsigned granule)1184 arm_vm_page_granular_RNX(vm_offset_t start, unsigned long size, unsigned granule)
1185 {
1186 	arm_vm_page_granular_prot(start, size, 0, 1, AP_RONA, 1, granule, ARM64_PAGE_UNGUARDED);
1187 }
1188 
1189 static inline void
arm_vm_page_granular_ROX(vm_offset_t start,unsigned long size,unsigned granule,unsigned int guarded)1190 arm_vm_page_granular_ROX(vm_offset_t start, unsigned long size, unsigned granule, unsigned int guarded)
1191 {
1192 	arm_vm_page_granular_prot(start, size, 0, 0, AP_RONA, 0, granule, guarded);
1193 }
1194 
1195 static inline void
arm_vm_page_granular_RWNX(vm_offset_t start,unsigned long size,unsigned granule)1196 arm_vm_page_granular_RWNX(vm_offset_t start, unsigned long size, unsigned granule)
1197 {
1198 	arm_vm_page_granular_prot(start, size, 0, 1, AP_RWNA, 1, granule, ARM64_PAGE_UNGUARDED);
1199 }
1200 
1201 // Populate seg...AuxKC and fixup AuxKC permissions
1202 static bool
arm_vm_auxkc_init(void)1203 arm_vm_auxkc_init(void)
1204 {
1205 	if (auxkc_mh == 0 || auxkc_base == 0) {
1206 		return false; // no auxKC.
1207 	}
1208 
1209 	/* Fixup AuxKC and populate seg*AuxKC globals used below */
1210 	arm_auxkc_init((void*)auxkc_mh, (void*)auxkc_base);
1211 
1212 	if (segLOWESTAuxKC != segLOWEST) {
1213 		panic("segLOWESTAuxKC (%p) not equal to segLOWEST (%p). auxkc_mh: %p, auxkc_base: %p",
1214 		    (void*)segLOWESTAuxKC, (void*)segLOWEST,
1215 		    (void*)auxkc_mh, (void*)auxkc_base);
1216 	}
1217 
1218 	/*
1219 	 * The AuxKC LINKEDIT segment needs to be covered by the RO region but is excluded
1220 	 * from the RO address range returned by kernel_collection_adjust_mh_addrs().
1221 	 * Ensure the highest non-LINKEDIT address in the AuxKC is the current end of
1222 	 * its RO region before extending it.
1223 	 */
1224 	assert(segHIGHESTROAuxKC == segHIGHESTNLEAuxKC);
1225 	assert(segHIGHESTAuxKC >= segHIGHESTROAuxKC);
1226 	if (segHIGHESTAuxKC > segHIGHESTROAuxKC) {
1227 		segHIGHESTROAuxKC = segHIGHESTAuxKC;
1228 	}
1229 
1230 	/*
1231 	 * The AuxKC RO region must be right below the device tree/trustcache so that it can be covered
1232 	 * by CTRR, and the AuxKC RX region must be within the RO region.
1233 	 */
1234 	assert(segHIGHESTROAuxKC == auxkc_right_above);
1235 	assert(segHIGHESTRXAuxKC <= segHIGHESTROAuxKC);
1236 	assert(segLOWESTRXAuxKC <= segHIGHESTRXAuxKC);
1237 	assert(segLOWESTROAuxKC <= segLOWESTRXAuxKC);
1238 	assert(segLOWESTAuxKC <= segLOWESTROAuxKC);
1239 
1240 	if (segHIGHESTRXAuxKC < segLOWEST) {
1241 		arm_vm_page_granular_RNX(segHIGHESTRXAuxKC, segLOWEST - segHIGHESTRXAuxKC, 0);
1242 	}
1243 	if (segLOWESTRXAuxKC < segHIGHESTRXAuxKC) {
1244 		/*
1245 		 * We cannot mark auxKC text as guarded because doing so would enforce
1246 		 * BTI on oblivious third-party kexts and break ABI compatibility.
1247 		 * Doing this defeats the purpose of BTI (branches to these pages are
1248 		 * unchecked!) but given both the relative rarity and the diversity of
1249 		 * third-party kexts, we expect that this is likely impractical to
1250 		 * exploit in practice.
1251 		 */
1252 		arm_vm_page_granular_ROX(segLOWESTRXAuxKC, segHIGHESTRXAuxKC - segLOWESTRXAuxKC, 0, ARM64_PAGE_UNGUARDED); // Refined in OSKext::readPrelinkedExtensions
1253 	}
1254 	if (segLOWESTROAuxKC < segLOWESTRXAuxKC) {
1255 		arm_vm_page_granular_RNX(segLOWESTROAuxKC, segLOWESTRXAuxKC - segLOWESTROAuxKC, 0);
1256 	}
1257 	if (segLOWESTAuxKC < segLOWESTROAuxKC) {
1258 		arm_vm_page_granular_RWNX(segLOWESTAuxKC, segLOWESTROAuxKC - segLOWESTAuxKC, 0);
1259 	}
1260 
1261 	return true;
1262 }
1263 
1264 void
arm_vm_prot_init(__unused boot_args * args)1265 arm_vm_prot_init(__unused boot_args * args)
1266 {
1267 	segLOWESTTEXT = UINT64_MAX;
1268 	if (segSizePRELINKTEXT && (segPRELINKTEXTB < segLOWESTTEXT)) {
1269 		segLOWESTTEXT = segPRELINKTEXTB;
1270 	}
1271 	assert(segSizeTEXT);
1272 	if (segTEXTB < segLOWESTTEXT) {
1273 		segLOWESTTEXT = segTEXTB;
1274 	}
1275 	assert(segLOWESTTEXT < UINT64_MAX);
1276 
1277 	segEXTRADATA = 0;
1278 	segSizeEXTRADATA = 0;
1279 	segTRUSTCACHE = 0;
1280 	segSizeTRUSTCACHE = 0;
1281 
1282 	segLOWEST = segLOWESTTEXT;
1283 	segLOWESTRO = segLOWESTTEXT;
1284 
1285 	if (segLOWESTKC && segLOWESTKC < segLOWEST) {
1286 		/*
1287 		 * kernel collections have segments below the kernel. In particular the collection mach header
1288 		 * is below PRELINK_TEXT and is not covered by any other segments already tracked.
1289 		 */
1290 		arm_vm_page_granular_RNX(segLOWESTKC, segLOWEST - segLOWESTKC, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT);
1291 		segLOWEST = segLOWESTKC;
1292 		if (segLOWESTROKC && segLOWESTROKC < segLOWESTRO) {
1293 			segLOWESTRO = segLOWESTROKC;
1294 		}
1295 		if (segHIGHESTROKC && segHIGHESTROKC > segHIGHESTRO) {
1296 			segHIGHESTRO = segHIGHESTROKC;
1297 		}
1298 	}
1299 
1300 	DTEntry memory_map;
1301 	int err;
1302 
1303 	// Device Tree portion of EXTRADATA
1304 	if (SecureDTIsLockedDown()) {
1305 		segEXTRADATA = (vm_offset_t)PE_state.deviceTreeHead;
1306 		segSizeEXTRADATA = PE_state.deviceTreeSize;
1307 	}
1308 
1309 	// Trust Caches portion of EXTRADATA
1310 	{
1311 		DTMemoryMapRange const *trustCacheRange;
1312 		unsigned int trustCacheRangeSize;
1313 
1314 		err = SecureDTLookupEntry(NULL, "chosen/memory-map", &memory_map);
1315 		assert(err == kSuccess);
1316 
1317 		err = SecureDTGetProperty(memory_map, "TrustCache", (void const **)&trustCacheRange, &trustCacheRangeSize);
1318 		if (err == kSuccess) {
1319 			if (trustCacheRangeSize != sizeof(DTMemoryMapRange)) {
1320 				panic("Unexpected /chosen/memory-map/TrustCache property size %u != %zu", trustCacheRangeSize, sizeof(DTMemoryMapRange));
1321 			}
1322 
1323 			vm_offset_t const trustCacheRegion = phystokv(trustCacheRange->paddr);
1324 			if (trustCacheRegion < segLOWEST) {
1325 				if (segEXTRADATA != 0) {
1326 					if (trustCacheRegion != segEXTRADATA + segSizeEXTRADATA) {
1327 						panic("Unexpected location of TrustCache region: %#lx != %#lx",
1328 						    trustCacheRegion, segEXTRADATA + segSizeEXTRADATA);
1329 					}
1330 					segSizeEXTRADATA += trustCacheRange->length;
1331 				} else {
1332 					// Not all devices support CTRR device trees.
1333 					segEXTRADATA = trustCacheRegion;
1334 					segSizeEXTRADATA = trustCacheRange->length;
1335 				}
1336 			}
1337 #if !(DEVELOPMENT || DEBUG)
1338 			else {
1339 				panic("TrustCache region is in an unexpected place: %#lx > %#lx", trustCacheRegion, segLOWEST);
1340 			}
1341 #endif
1342 			segTRUSTCACHE = trustCacheRegion;
1343 			segSizeTRUSTCACHE = trustCacheRange->length;
1344 		}
1345 	}
1346 
1347 	if (segSizeEXTRADATA != 0) {
1348 		if (segEXTRADATA <= segLOWEST) {
1349 			segLOWEST = segEXTRADATA;
1350 			if (segEXTRADATA <= segLOWESTRO) {
1351 				segLOWESTRO = segEXTRADATA;
1352 			}
1353 		} else {
1354 			panic("EXTRADATA is in an unexpected place: %#lx > %#lx", segEXTRADATA, segLOWEST);
1355 		}
1356 
1357 		arm_vm_page_granular_RNX(segEXTRADATA, segSizeEXTRADATA, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT);
1358 	}
1359 
1360 	const DTMemoryMapRange *auxKC_range, *auxKC_header_range;
1361 	unsigned int auxKC_range_size, auxKC_header_range_size;
1362 
1363 	err = SecureDTGetProperty(memory_map, "AuxKC", (const void**)&auxKC_range,
1364 	    &auxKC_range_size);
1365 	if (err != kSuccess) {
1366 		goto noAuxKC;
1367 	}
1368 	assert(auxKC_range_size == sizeof(DTMemoryMapRange));
1369 	err = SecureDTGetProperty(memory_map, "AuxKC-mach_header",
1370 	    (const void**)&auxKC_header_range, &auxKC_header_range_size);
1371 	if (err != kSuccess) {
1372 		goto noAuxKC;
1373 	}
1374 	assert(auxKC_header_range_size == sizeof(DTMemoryMapRange));
1375 
1376 	if (auxKC_header_range->paddr == 0 || auxKC_range->paddr == 0) {
1377 		goto noAuxKC;
1378 	}
1379 
1380 	auxkc_mh = phystokv(auxKC_header_range->paddr);
1381 	auxkc_base = phystokv(auxKC_range->paddr);
1382 
1383 	if (auxkc_base < segLOWEST) {
1384 		auxkc_right_above = segLOWEST;
1385 		segLOWEST = auxkc_base;
1386 	} else {
1387 		panic("auxkc_base (%p) not below segLOWEST (%p)", (void*)auxkc_base, (void*)segLOWEST);
1388 	}
1389 
1390 	/* Map AuxKC RWNX initially so that arm_vm_auxkc_init can traverse
1391 	 * it and apply fixups (after we're off the bootstrap translation
1392 	 * tables).
1393 	 */
1394 	arm_vm_page_granular_RWNX(auxkc_base, auxKC_range->length, 0);
1395 
1396 noAuxKC:
1397 	/* Map coalesced kext TEXT segment RWNX for now */
1398 	arm_vm_page_granular_RWNX(segPRELINKTEXTB, segSizePRELINKTEXT, ARM64_GRANULE_ALLOW_BLOCK); // Refined in OSKext::readPrelinkedExtensions
1399 
1400 	/* Map coalesced kext DATA_CONST segment RWNX (could be empty) */
1401 	arm_vm_page_granular_RWNX(segPLKDATACONSTB, segSizePLKDATACONST, ARM64_GRANULE_ALLOW_BLOCK); // Refined in OSKext::readPrelinkedExtensions
1402 
1403 	/* Map coalesced kext TEXT_EXEC segment RX (could be empty) */
1404 	arm_vm_page_granular_ROX(segPLKTEXTEXECB, segSizePLKTEXTEXEC, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT, ARM64_PAGE_GUARDED); // Refined in OSKext::readPrelinkedExtensions
1405 
1406 	/* if new segments not present, set space between PRELINK_TEXT and xnu TEXT to RWNX
1407 	 * otherwise we no longer expect any space between the coalesced kext read only segments and xnu rosegments
1408 	 */
1409 	if (!segSizePLKDATACONST && !segSizePLKTEXTEXEC) {
1410 		if (segSizePRELINKTEXT) {
1411 			arm_vm_page_granular_RWNX(segPRELINKTEXTB + segSizePRELINKTEXT, segTEXTB - (segPRELINKTEXTB + segSizePRELINKTEXT),
1412 			    ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT);
1413 		}
1414 	} else {
1415 		/*
1416 		 * If we have the new segments, we should still protect the gap between kext
1417 		 * read-only pages and kernel read-only pages, in the event that this gap
1418 		 * exists.
1419 		 */
1420 		if ((segPLKDATACONSTB + segSizePLKDATACONST) < segTEXTB) {
1421 			arm_vm_page_granular_RWNX(segPLKDATACONSTB + segSizePLKDATACONST, segTEXTB - (segPLKDATACONSTB + segSizePLKDATACONST),
1422 			    ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT);
1423 		}
1424 	}
1425 
1426 	/*
1427 	 * Protection on kernel text is loose here to allow shenanigans early on.  These
1428 	 * protections are tightened in arm_vm_prot_finalize().  This is necessary because
1429 	 * we currently patch LowResetVectorBase in cpu.c.
1430 	 *
1431 	 * TEXT segment contains mach headers and other non-executable data. This will become RONX later.
1432 	 */
1433 	arm_vm_page_granular_RNX(segTEXTB, segSizeTEXT, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT);
1434 
1435 	/* Can DATACONST start out and stay RNX?
1436 	 * NO, stuff in this segment gets modified during startup (viz. mac_policy_init()/mac_policy_list)
1437 	 * Make RNX in prot_finalize
1438 	 */
1439 	arm_vm_page_granular_RWNX(segDATACONSTB, segSizeDATACONST, ARM64_GRANULE_ALLOW_BLOCK);
1440 
1441 	arm_vm_page_granular_ROX(segTEXTEXECB, segSizeTEXTEXEC, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT, ARM64_PAGE_GUARDED);
1442 
1443 #if XNU_MONITOR
1444 	arm_vm_page_granular_ROX(segPPLTEXTB, segSizePPLTEXT, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT, ARM64_PAGE_UNGUARDED);
1445 	arm_vm_page_granular_ROX(segPPLTRAMPB, segSizePPLTRAMP, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT, ARM64_PAGE_UNGUARDED);
1446 	arm_vm_page_granular_RNX(segPPLDATACONSTB, segSizePPLDATACONST, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT);
1447 #endif
1448 
1449 	/* DATA segment will remain RWNX */
1450 	arm_vm_page_granular_RWNX(segDATAB, segSizeDATA, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT);
1451 #if XNU_MONITOR
1452 	arm_vm_page_granular_RWNX(segPPLDATAB, segSizePPLDATA, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT);
1453 #endif
1454 
1455 	arm_vm_page_granular_RWNX(segHIBDATAB, segSizeHIBDATA, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT);
1456 
1457 	arm_vm_page_granular_RWNX(segBOOTDATAB, segSizeBOOTDATA, 0);
1458 	arm_vm_page_granular_RNX((vm_offset_t)&intstack_low_guard, PAGE_MAX_SIZE, 0);
1459 	arm_vm_page_granular_RNX((vm_offset_t)&intstack_high_guard, PAGE_MAX_SIZE, 0);
1460 	arm_vm_page_granular_RNX((vm_offset_t)&excepstack_high_guard, PAGE_MAX_SIZE, 0);
1461 
1462 	arm_vm_page_granular_ROX(segKLDB, segSizeKLD, 0, ARM64_PAGE_GUARDED);
1463 	arm_vm_page_granular_RNX(segKLDDATAB, segSizeKLDDATA, 0);
1464 	arm_vm_page_granular_RWNX(segLINKB, segSizeLINK, 0);
1465 	arm_vm_page_granular_RWNX(segPLKLINKEDITB, segSizePLKLINKEDIT, 0); // Coalesced kext LINKEDIT segment
1466 	arm_vm_page_granular_ROX(segLASTB, segSizeLAST, ARM64_GRANULE_ALLOW_BLOCK, ARM64_PAGE_GUARDED); // __LAST may be empty, but we cannot assume this
1467 	if (segLASTDATACONSTB) {
1468 		arm_vm_page_granular_RWNX(segLASTDATACONSTB, segSizeLASTDATACONST, ARM64_GRANULE_ALLOW_BLOCK); // __LASTDATA_CONST may be empty, but we cannot assume this
1469 	}
1470 	arm_vm_page_granular_RWNX(segPRELINKDATAB, segSizePRELINKDATA, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT); // Prelink __DATA for kexts (RW data)
1471 
1472 	if (segSizePLKLLVMCOV > 0) {
1473 		arm_vm_page_granular_RWNX(segPLKLLVMCOVB, segSizePLKLLVMCOV, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT); // LLVM code coverage data
1474 	}
1475 	arm_vm_page_granular_RWNX(segPRELINKINFOB, segSizePRELINKINFO, 0); /* PreLinkInfoDictionary */
1476 
1477 	/* Record the bounds of the kernelcache. */
1478 	vm_kernelcache_base = segLOWEST;
1479 	vm_kernelcache_top = end_kern;
1480 }
1481 
1482 /*
1483  * return < 0 for a < b
1484  *          0 for a == b
1485  *        > 0 for a > b
1486  */
1487 typedef int (*cmpfunc_t)(const void *a, const void *b);
1488 
1489 extern void
1490 qsort(void *a, size_t n, size_t es, cmpfunc_t cmp);
1491 
1492 static int
cmp_ptov_entries(const void * a,const void * b)1493 cmp_ptov_entries(const void *a, const void *b)
1494 {
1495 	const ptov_table_entry *entry_a = a;
1496 	const ptov_table_entry *entry_b = b;
1497 	// Sort in descending order of segment length
1498 	if (entry_a->len < entry_b->len) {
1499 		return 1;
1500 	} else if (entry_a->len > entry_b->len) {
1501 		return -1;
1502 	} else {
1503 		return 0;
1504 	}
1505 }
1506 
1507 SECURITY_READ_ONLY_LATE(static unsigned int) ptov_index = 0;
1508 
1509 #define ROUND_L1(addr) (((addr) + ARM_TT_L1_OFFMASK) & ~(ARM_TT_L1_OFFMASK))
1510 #define ROUND_TWIG(addr) (((addr) + ARM_TT_TWIG_OFFMASK) & ~(ARM_TT_TWIG_OFFMASK))
1511 
1512 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)1513 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)
1514 {
1515 	pmap_paddr_t pa_offset;
1516 
1517 	if (__improbable(ptov_index >= PTOV_TABLE_SIZE)) {
1518 		panic("%s: PTOV table limit exceeded; segment va = 0x%llx, size = 0x%llx", __func__,
1519 		    (unsigned long long)orig_va, (unsigned long long)len);
1520 	}
1521 	assert((orig_va & ARM_PGMASK) == 0);
1522 	temp_ptov_table[ptov_index].pa = orig_va - gVirtBase + gPhysBase;
1523 	if (ptov_index == 0) {
1524 		temp_ptov_table[ptov_index].va = physmap_base;
1525 	} else {
1526 		temp_ptov_table[ptov_index].va = temp_ptov_table[ptov_index - 1].va + temp_ptov_table[ptov_index - 1].len;
1527 	}
1528 	if (granule & ARM64_GRANULE_ALLOW_BLOCK) {
1529 		vm_map_address_t orig_offset = temp_ptov_table[ptov_index].pa & ARM_TT_TWIG_OFFMASK;
1530 		vm_map_address_t new_offset = temp_ptov_table[ptov_index].va & ARM_TT_TWIG_OFFMASK;
1531 		if (new_offset < orig_offset) {
1532 			temp_ptov_table[ptov_index].va += (orig_offset - new_offset);
1533 		} else if (new_offset > orig_offset) {
1534 			temp_ptov_table[ptov_index].va = ROUND_TWIG(temp_ptov_table[ptov_index].va) + orig_offset;
1535 		}
1536 	}
1537 	assert((temp_ptov_table[ptov_index].va & ARM_PGMASK) == 0);
1538 	temp_ptov_table[ptov_index].len = round_page(len);
1539 	pa_offset = temp_ptov_table[ptov_index].va - orig_va;
1540 	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, ARM64_PAGE_UNGUARDED);
1541 	++ptov_index;
1542 }
1543 
1544 #if XNU_MONITOR
1545 
1546 SECURITY_READ_ONLY_LATE(static boolean_t) keep_linkedit = FALSE;
1547 
1548 static void
arm_vm_physmap_init(boot_args * args)1549 arm_vm_physmap_init(boot_args *args)
1550 {
1551 	ptov_table_entry temp_ptov_table[PTOV_TABLE_SIZE];
1552 	bzero(temp_ptov_table, sizeof(temp_ptov_table));
1553 
1554 	// This is memory that will either be handed back to the VM layer via ml_static_mfree(),
1555 	// or will be available for general-purpose use.   Physical aperture mappings for this memory
1556 	// must be at page granularity, so that PPL ownership or cache attribute changes can be reflected
1557 	// in the physical aperture mappings.
1558 
1559 	// Slid region between gPhysBase and beginning of protected text
1560 	arm_vm_physmap_slide(temp_ptov_table, gVirtBase, segLOWEST - gVirtBase, AP_RWNA, 0);
1561 
1562 	// kext bootstrap segments
1563 #if !defined(KERNEL_INTEGRITY_KTRR) && !defined(KERNEL_INTEGRITY_CTRR)
1564 	/* __KLD,__text is covered by the rorgn */
1565 	arm_vm_physmap_slide(temp_ptov_table, segKLDB, segSizeKLD, AP_RONA, 0);
1566 #endif
1567 	arm_vm_physmap_slide(temp_ptov_table, segKLDDATAB, segSizeKLDDATA, AP_RONA, 0);
1568 
1569 	// Early-boot data
1570 	arm_vm_physmap_slide(temp_ptov_table, segBOOTDATAB, segSizeBOOTDATA, AP_RONA, 0);
1571 
1572 	PE_parse_boot_argn("keepsyms", &keep_linkedit, sizeof(keep_linkedit));
1573 #if CONFIG_DTRACE
1574 	if (dtrace_keep_kernel_symbols()) {
1575 		keep_linkedit = TRUE;
1576 	}
1577 #endif /* CONFIG_DTRACE */
1578 #if KASAN_DYNAMIC_BLACKLIST
1579 	/* KASAN's dynamic blacklist needs to query the LINKEDIT segment at runtime.  As such, the
1580 	 * kext bootstrap code will not jettison LINKEDIT on kasan kernels, so don't bother to relocate it. */
1581 	keep_linkedit = TRUE;
1582 #endif
1583 	if (!keep_linkedit) {
1584 		// Kernel LINKEDIT
1585 		arm_vm_physmap_slide(temp_ptov_table, segLINKB, segSizeLINK, AP_RWNA, 0);
1586 
1587 		if (segSizePLKLINKEDIT) {
1588 			// Prelinked kernel LINKEDIT
1589 			arm_vm_physmap_slide(temp_ptov_table, segPLKLINKEDITB, segSizePLKLINKEDIT, AP_RWNA, 0);
1590 		}
1591 	}
1592 
1593 	// Prelinked kernel plists
1594 	arm_vm_physmap_slide(temp_ptov_table, segPRELINKINFOB, segSizePRELINKINFO, AP_RWNA, 0);
1595 
1596 	// Device tree (if not locked down), ramdisk, boot args
1597 	arm_vm_physmap_slide(temp_ptov_table, end_kern, (args->topOfKernelData - gPhysBase + gVirtBase) - end_kern, AP_RWNA, 0);
1598 	if (!SecureDTIsLockedDown()) {
1599 		PE_slide_devicetree(temp_ptov_table[ptov_index - 1].va - end_kern);
1600 	}
1601 
1602 	// Remainder of physical memory
1603 	arm_vm_physmap_slide(temp_ptov_table, (args->topOfKernelData - gPhysBase + gVirtBase),
1604 	    real_avail_end - args->topOfKernelData, AP_RWNA, 0);
1605 
1606 
1607 
1608 	assert((temp_ptov_table[ptov_index - 1].va + temp_ptov_table[ptov_index - 1].len) <= physmap_end);
1609 
1610 	// Sort in descending order of segment length.  LUT traversal is linear, so largest (most likely used)
1611 	// segments should be placed earliest in the table to optimize lookup performance.
1612 	qsort(temp_ptov_table, PTOV_TABLE_SIZE, sizeof(temp_ptov_table[0]), cmp_ptov_entries);
1613 
1614 	memcpy(ptov_table, temp_ptov_table, sizeof(ptov_table));
1615 }
1616 
1617 #else
1618 
1619 static void
arm_vm_physmap_init(boot_args * args)1620 arm_vm_physmap_init(boot_args *args)
1621 {
1622 	ptov_table_entry temp_ptov_table[PTOV_TABLE_SIZE];
1623 	bzero(temp_ptov_table, sizeof(temp_ptov_table));
1624 
1625 	// Will be handed back to VM layer through ml_static_mfree() in arm_vm_prot_finalize()
1626 	arm_vm_physmap_slide(temp_ptov_table, gVirtBase, segLOWEST - gVirtBase, AP_RWNA,
1627 	    ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT);
1628 
1629 	arm_vm_page_granular_RWNX(end_kern, phystokv(args->topOfKernelData) - end_kern,
1630 	    ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT); /* Device Tree (if not locked down), RAM Disk (if present), bootArgs */
1631 
1632 	arm_vm_physmap_slide(temp_ptov_table, (args->topOfKernelData - gPhysBase + gVirtBase),
1633 	    real_avail_end - args->topOfKernelData, AP_RWNA, ARM64_GRANULE_ALLOW_BLOCK | ARM64_GRANULE_ALLOW_HINT); // rest of physmem
1634 
1635 	assert((temp_ptov_table[ptov_index - 1].va + temp_ptov_table[ptov_index - 1].len) <= physmap_end);
1636 
1637 	// Sort in descending order of segment length.  LUT traversal is linear, so largest (most likely used)
1638 	// segments should be placed earliest in the table to optimize lookup performance.
1639 	qsort(temp_ptov_table, PTOV_TABLE_SIZE, sizeof(temp_ptov_table[0]), cmp_ptov_entries);
1640 
1641 	memcpy(ptov_table, temp_ptov_table, sizeof(ptov_table));
1642 }
1643 
1644 #endif // XNU_MONITOR
1645 
1646 void
arm_vm_prot_finalize(boot_args * args __unused)1647 arm_vm_prot_finalize(boot_args * args __unused)
1648 {
1649 	/*
1650 	 * At this point, we are far enough along in the boot process that it will be
1651 	 * safe to free up all of the memory preceeding the kernel.  It may in fact
1652 	 * be safe to do this earlier.
1653 	 *
1654 	 * This keeps the memory in the V-to-P mapping, but advertises it to the VM
1655 	 * as usable.
1656 	 */
1657 
1658 	/*
1659 	 * if old style PRELINK segment exists, free memory before it, and after it before XNU text
1660 	 * otherwise we're dealing with a new style kernel cache, so we should just free the
1661 	 * memory before PRELINK_TEXT segment, since the rest of the KEXT read only data segments
1662 	 * should be immediately followed by XNU's TEXT segment
1663 	 */
1664 
1665 	ml_static_mfree(phystokv(gPhysBase), segLOWEST - gVirtBase);
1666 
1667 	/*
1668 	 * KTRR support means we will be mucking with these pages and trying to
1669 	 * protect them; we cannot free the pages to the VM if we do this.
1670 	 */
1671 	if (!segSizePLKDATACONST && !segSizePLKTEXTEXEC && segSizePRELINKTEXT) {
1672 		/* If new segments not present, PRELINK_TEXT is not dynamically sized, free DRAM between it and xnu TEXT */
1673 		ml_static_mfree(segPRELINKTEXTB + segSizePRELINKTEXT, segTEXTB - (segPRELINKTEXTB + segSizePRELINKTEXT));
1674 	}
1675 
1676 	/* tighten permissions on kext read only data and code */
1677 	arm_vm_page_granular_RNX(segPRELINKTEXTB, segSizePRELINKTEXT, ARM64_GRANULE_ALLOW_BLOCK);
1678 	arm_vm_page_granular_RNX(segPLKDATACONSTB, segSizePLKDATACONST, ARM64_GRANULE_ALLOW_BLOCK);
1679 
1680 	cpu_stack_alloc(&BootCpuData);
1681 	arm64_replace_bootstack(&BootCpuData);
1682 	ml_static_mfree(phystokv(segBOOTDATAB - gVirtBase + gPhysBase), segSizeBOOTDATA);
1683 
1684 #if __ARM_KERNEL_PROTECT__
1685 	arm_vm_populate_kernel_el0_mappings();
1686 #endif /* __ARM_KERNEL_PROTECT__ */
1687 
1688 #if XNU_MONITOR
1689 #if !defined(KERNEL_INTEGRITY_KTRR) && !defined(KERNEL_INTEGRITY_CTRR)
1690 	/* __KLD,__text is covered by the rorgn */
1691 	for (vm_offset_t va = segKLDB; va < (segKLDB + segSizeKLD); va += ARM_PGBYTES) {
1692 		pt_entry_t *pte = arm_kva_to_pte(va);
1693 		*pte = ARM_PTE_EMPTY;
1694 	}
1695 #endif
1696 	for (vm_offset_t va = segKLDDATAB; va < (segKLDDATAB + segSizeKLDDATA); va += ARM_PGBYTES) {
1697 		pt_entry_t *pte = arm_kva_to_pte(va);
1698 		*pte = ARM_PTE_EMPTY;
1699 	}
1700 	/* Clear the original stack mappings; these pages should be mapped through ptov_table. */
1701 	for (vm_offset_t va = segBOOTDATAB; va < (segBOOTDATAB + segSizeBOOTDATA); va += ARM_PGBYTES) {
1702 		pt_entry_t *pte = arm_kva_to_pte(va);
1703 		*pte = ARM_PTE_EMPTY;
1704 	}
1705 	/* Clear the original PRELINKINFO mapping. This segment should be jettisoned during I/O Kit
1706 	 * initialization before we reach this point. */
1707 	for (vm_offset_t va = segPRELINKINFOB; va < (segPRELINKINFOB + segSizePRELINKINFO); va += ARM_PGBYTES) {
1708 		pt_entry_t *pte = arm_kva_to_pte(va);
1709 		*pte = ARM_PTE_EMPTY;
1710 	}
1711 	if (!keep_linkedit) {
1712 		for (vm_offset_t va = segLINKB; va < (segLINKB + segSizeLINK); va += ARM_PGBYTES) {
1713 			pt_entry_t *pte = arm_kva_to_pte(va);
1714 			*pte = ARM_PTE_EMPTY;
1715 		}
1716 		if (segSizePLKLINKEDIT) {
1717 			for (vm_offset_t va = segPLKLINKEDITB; va < (segPLKLINKEDITB + segSizePLKLINKEDIT); va += ARM_PGBYTES) {
1718 				pt_entry_t *pte = arm_kva_to_pte(va);
1719 				*pte = ARM_PTE_EMPTY;
1720 			}
1721 		}
1722 	}
1723 #endif /* XNU_MONITOR */
1724 
1725 #if defined(KERNEL_INTEGRITY_KTRR) || defined(KERNEL_INTEGRITY_CTRR)
1726 	/*
1727 	 * __LAST,__pinst should no longer be executable.
1728 	 */
1729 	arm_vm_page_granular_RNX(segLASTB, segSizeLAST, ARM64_GRANULE_ALLOW_BLOCK);
1730 
1731 	/* __LASTDATA_CONST should no longer be writable. */
1732 	if (segLASTDATACONSTB) {
1733 		arm_vm_page_granular_RNX(segLASTDATACONSTB, segSizeLASTDATACONST, ARM64_GRANULE_ALLOW_BLOCK);
1734 	}
1735 
1736 	/*
1737 	 * __KLD,__text should no longer be executable.
1738 	 */
1739 	arm_vm_page_granular_RNX(segKLDB, segSizeKLD, ARM64_GRANULE_ALLOW_BLOCK);
1740 
1741 	/*
1742 	 * Must wait until all other region permissions are set before locking down DATA_CONST
1743 	 * as the kernel static page tables live in DATA_CONST on KTRR enabled systems
1744 	 * and will become immutable.
1745 	 */
1746 #endif
1747 
1748 	arm_vm_page_granular_RNX(segDATACONSTB, segSizeDATACONST, ARM64_GRANULE_ALLOW_BLOCK);
1749 
1750 	__builtin_arm_dsb(DSB_ISH);
1751 	flush_mmu_tlb();
1752 }
1753 
1754 /*
1755  * TBI (top-byte ignore) is an ARMv8 feature for ignoring the top 8 bits of
1756  * address accesses. It can be enabled separately for TTBR0 (user) and
1757  * TTBR1 (kernel).
1758  */
1759 void
arm_set_kernel_tbi(void)1760 arm_set_kernel_tbi(void)
1761 {
1762 #if !__ARM_KERNEL_PROTECT__ && CONFIG_KERNEL_TBI
1763 	uint64_t old_tcr, new_tcr;
1764 
1765 	old_tcr = new_tcr = get_tcr();
1766 	/*
1767 	 * For kernel configurations that require TBI support on
1768 	 * PAC systems, we enable DATA TBI only.
1769 	 */
1770 	new_tcr |= TCR_TBI1_TOPBYTE_IGNORED;
1771 	new_tcr |= TCR_TBID1_ENABLE;
1772 
1773 	if (old_tcr != new_tcr) {
1774 		set_tcr(new_tcr);
1775 		sysreg_restore.tcr_el1 = new_tcr;
1776 	}
1777 #endif /* !__ARM_KERNEL_PROTECT__ && CONFIG_KERNEL_TBI */
1778 }
1779 
1780 static void
arm_set_user_tbi(void)1781 arm_set_user_tbi(void)
1782 {
1783 #if !__ARM_KERNEL_PROTECT__
1784 	uint64_t old_tcr, new_tcr;
1785 
1786 	old_tcr = new_tcr = get_tcr();
1787 	new_tcr |= TCR_TBI0_TOPBYTE_IGNORED;
1788 
1789 	if (old_tcr != new_tcr) {
1790 		set_tcr(new_tcr);
1791 		sysreg_restore.tcr_el1 = new_tcr;
1792 	}
1793 #endif /* !__ARM_KERNEL_PROTECT__ */
1794 }
1795 
1796 /*
1797  * Initialize and enter blank (invalid) page tables in a L1 translation table for a given VA range.
1798  *
1799  * This is a helper function used to build up the initial page tables for the kernel translation table.
1800  * With KERNEL_INTEGRITY we keep at least the root level of the kernel page table immutable, thus the need
1801  * to preallocate before machine_lockdown any L1 entries necessary during the entire kernel runtime.
1802  *
1803  * For a given VA range, if necessary, allocate new L2 translation tables and install the table entries in
1804  * the appropriate L1 table indexes. called before the translation table is active
1805  *
1806  * parameters:
1807  *
1808  * tt: virtual address of L1 translation table to modify
1809  * start: beginning of VA range
1810  * end: end of VA range
1811  * static_map: whether to allocate the new translation table page from read only memory
1812  * table_attrs: attributes of new table entry in addition to VALID and TYPE_TABLE attributes
1813  *
1814  */
1815 
1816 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)1817 init_ptpages(tt_entry_t *tt, vm_map_address_t start, vm_map_address_t end, bool static_map, uint64_t table_attrs)
1818 {
1819 	tt_entry_t *l1_tte;
1820 	vm_offset_t ptpage_vaddr;
1821 
1822 	l1_tte = tt + ((start & ARM_TT_L1_INDEX_MASK) >> ARM_TT_L1_SHIFT);
1823 
1824 	while (start < end) {
1825 		if (*l1_tte == ARM_TTE_EMPTY) {
1826 			/* Allocate a page and setup L1 Table TTE in L1 */
1827 			ptpage_vaddr = alloc_ptpage(static_map);
1828 			*l1_tte = (kvtophys(ptpage_vaddr) & ARM_TTE_TABLE_MASK) | ARM_TTE_TYPE_TABLE | ARM_TTE_VALID | table_attrs;
1829 			bzero((void *)ptpage_vaddr, ARM_PGBYTES);
1830 		}
1831 
1832 		if ((start + ARM_TT_L1_SIZE) < start) {
1833 			/* If this is the last L1 entry, it must cover the last mapping. */
1834 			break;
1835 		}
1836 
1837 		start += ARM_TT_L1_SIZE;
1838 		l1_tte++;
1839 	}
1840 }
1841 
1842 #define ARM64_PHYSMAP_SLIDE_RANGE (1ULL << 30) // 1 GB
1843 #define ARM64_PHYSMAP_SLIDE_MASK  (ARM64_PHYSMAP_SLIDE_RANGE - 1)
1844 
1845 void
arm_vm_init(uint64_t memory_size,boot_args * args)1846 arm_vm_init(uint64_t memory_size, boot_args * args)
1847 {
1848 	vm_map_address_t va_l1, va_l1_end;
1849 	tt_entry_t       *cpu_l1_tte;
1850 	vm_map_address_t va_l2, va_l2_end;
1851 	tt_entry_t       *cpu_l2_tte;
1852 	pmap_paddr_t     boot_ttep;
1853 	tt_entry_t       *boot_tte;
1854 	uint64_t         mem_segments;
1855 	vm_offset_t      ptpage_vaddr;
1856 	vm_map_address_t dynamic_memory_begin;
1857 
1858 	/*
1859 	 * Get the virtual and physical kernel-managed memory base from boot_args.
1860 	 */
1861 	gVirtBase = args->virtBase;
1862 	gPhysBase = args->physBase;
1863 #if KASAN
1864 	real_phys_size = args->memSize + (shadow_ptop - shadow_pbase);
1865 #else
1866 	real_phys_size = args->memSize;
1867 #endif
1868 	/*
1869 	 * Ensure the physical region we specify for the VM to manage ends on a
1870 	 * software page boundary.  Note that the software page size (PAGE_SIZE)
1871 	 * may be a multiple of the hardware page size specified in ARM_PGBYTES.
1872 	 * We must round the reported memory size down to the nearest PAGE_SIZE
1873 	 * boundary to ensure the VM does not try to manage a page it does not
1874 	 * completely own.  The KASAN shadow region, if present, is managed entirely
1875 	 * in units of the hardware page size and should not need similar treatment.
1876 	 */
1877 	gPhysSize = mem_size = ((gPhysBase + args->memSize) & ~PAGE_MASK) - gPhysBase;
1878 
1879 	mem_actual = args->memSizeActual ? args->memSizeActual : mem_size;
1880 
1881 	if ((memory_size != 0) && (mem_size > memory_size)) {
1882 		mem_size = memory_size;
1883 		max_mem_actual = memory_size;
1884 	} else {
1885 		max_mem_actual = mem_actual;
1886 	}
1887 #if !defined(ARM_LARGE_MEMORY)
1888 	if (mem_size >= ((VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS) / 2)) {
1889 		panic("Unsupported memory configuration %lx", mem_size);
1890 	}
1891 #endif
1892 
1893 #if defined(ARM_LARGE_MEMORY)
1894 	unsigned long physmap_l1_entries = ((real_phys_size + ARM64_PHYSMAP_SLIDE_RANGE) >> ARM_TT_L1_SHIFT) + 1;
1895 	physmap_base = VM_MIN_KERNEL_ADDRESS - (physmap_l1_entries << ARM_TT_L1_SHIFT);
1896 #else
1897 	physmap_base = phystokv(args->topOfKernelData);
1898 #endif
1899 
1900 	// Slide the physical aperture to a random page-aligned location within the slide range
1901 	uint64_t physmap_slide = early_random() & ARM64_PHYSMAP_SLIDE_MASK & ~((uint64_t)PAGE_MASK);
1902 	assert(physmap_slide < ARM64_PHYSMAP_SLIDE_RANGE);
1903 
1904 	physmap_base += physmap_slide;
1905 
1906 #if XNU_MONITOR
1907 	physmap_base = ROUND_TWIG(physmap_base);
1908 #if defined(ARM_LARGE_MEMORY)
1909 	static_memory_end = phystokv(args->topOfKernelData);
1910 #else
1911 	static_memory_end = physmap_base + mem_size;
1912 #endif // ARM_LARGE_MEMORY
1913 	physmap_end = physmap_base + real_phys_size;
1914 
1915 
1916 #else
1917 #if defined(ARM_LARGE_MEMORY)
1918 	/* For large memory systems with no PPL such as virtual machines */
1919 	static_memory_end = phystokv(args->topOfKernelData);
1920 	physmap_end = physmap_base + real_phys_size;
1921 #else
1922 	static_memory_end = physmap_base + mem_size + (PTOV_TABLE_SIZE * ARM_TT_TWIG_SIZE); // worst possible case for block alignment
1923 	physmap_end = physmap_base + real_phys_size + (PTOV_TABLE_SIZE * ARM_TT_TWIG_SIZE);
1924 #endif // ARM_LARGE_MEMORY
1925 #endif
1926 
1927 #if KASAN && !defined(ARM_LARGE_MEMORY)
1928 	/* add the KASAN stolen memory to the physmap */
1929 	dynamic_memory_begin = static_memory_end + (shadow_ptop - shadow_pbase);
1930 #else
1931 	dynamic_memory_begin = static_memory_end;
1932 #endif
1933 #if XNU_MONITOR
1934 	pmap_stacks_start = (void*)dynamic_memory_begin;
1935 	dynamic_memory_begin += PPL_STACK_REGION_SIZE;
1936 	pmap_stacks_end = (void*)dynamic_memory_begin;
1937 
1938 #if HAS_GUARDED_IO_FILTER
1939     iofilter_stacks_start = (void*)dynamic_memory_begin;
1940     dynamic_memory_begin += IOFILTER_STACK_REGION_SIZE;
1941     iofilter_stacks_end = (void*)dynamic_memory_begin;
1942 #endif
1943 #endif
1944 	if (dynamic_memory_begin > VM_MAX_KERNEL_ADDRESS) {
1945 		panic("Unsupported memory configuration %lx", mem_size);
1946 	}
1947 
1948 	boot_tte = (tt_entry_t *)&bootstrap_pagetables;
1949 	boot_ttep = kvtophys((vm_offset_t)boot_tte);
1950 
1951 #if DEVELOPMENT || DEBUG
1952 	/* Sanity check - assert that BOOTSTRAP_TABLE_SIZE is sufficiently-large to
1953 	 * hold our bootstrap mappings for any possible slide */
1954 	size_t bytes_mapped = dynamic_memory_begin - gVirtBase;
1955 	size_t l1_entries = 1 + ((bytes_mapped + ARM_TT_L1_SIZE - 1) / ARM_TT_L1_SIZE);
1956 	/* 1 L1 each for V=P and KVA, plus 1 page for each L2 */
1957 	size_t pages_used = 2 * (l1_entries + 1);
1958 	if (pages_used > BOOTSTRAP_TABLE_SIZE) {
1959 		panic("BOOTSTRAP_TABLE_SIZE too small for memory config");
1960 	}
1961 #endif
1962 
1963 	/*
1964 	 *  TTBR0 L1, TTBR0 L2 - 1:1 bootstrap mapping.
1965 	 *  TTBR1 L1, TTBR1 L2 - kernel mapping
1966 	 */
1967 
1968 	/*
1969 	 * TODO: free bootstrap table memory back to allocator.
1970 	 * on large memory systems bootstrap tables could be quite large.
1971 	 * after bootstrap complete, xnu can warm start with a single 16KB page mapping
1972 	 * to trampoline to KVA. this requires only 3 pages to stay resident.
1973 	 */
1974 	first_avail_phys = avail_start = args->topOfKernelData;
1975 
1976 #if defined(KERNEL_INTEGRITY_KTRR) || defined(KERNEL_INTEGRITY_CTRR)
1977 	arm_replace_identity_map();
1978 #endif
1979 
1980 	/* Initialize invalid tte page */
1981 	invalid_tte = (tt_entry_t *)alloc_ptpage(TRUE);
1982 	invalid_ttep = kvtophys((vm_offset_t)invalid_tte);
1983 	bzero(invalid_tte, ARM_PGBYTES);
1984 
1985 	/*
1986 	 * Initialize l1 page table page
1987 	 */
1988 	cpu_tte = (tt_entry_t *)alloc_ptpage(TRUE);
1989 	cpu_ttep = kvtophys((vm_offset_t)cpu_tte);
1990 	bzero(cpu_tte, ARM_PGBYTES);
1991 	avail_end = gPhysBase + mem_size;
1992 	assert(!(avail_end & PAGE_MASK));
1993 
1994 #if KASAN
1995 	real_avail_end = gPhysBase + real_phys_size;
1996 #else
1997 	real_avail_end = avail_end;
1998 #endif
1999 
2000 	/*
2001 	 * Initialize l1 and l2 page table pages :
2002 	 *   map physical memory at the kernel base virtual address
2003 	 *   cover the kernel dynamic address range section
2004 	 *
2005 	 *   the so called physical aperture should be statically mapped
2006 	 */
2007 	init_ptpages(cpu_tte, gVirtBase, dynamic_memory_begin, TRUE, ARM_TTE_TABLE_AP(ARM_TTE_TABLE_AP_USER_NA));
2008 
2009 #if defined(ARM_LARGE_MEMORY)
2010 	/*
2011 	 * Initialize l1 page table pages :
2012 	 *   on large memory systems the physical aperture exists separately below
2013 	 *   the rest of the kernel virtual address space
2014 	 */
2015 	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));
2016 #endif
2017 
2018 
2019 #if __ARM_KERNEL_PROTECT__
2020 	/* Expand the page tables to prepare for the EL0 mappings. */
2021 	arm_vm_expand_kernel_el0_mappings();
2022 #endif /* __ARM_KERNEL_PROTECT__ */
2023 
2024 	/*
2025 	 * Now retrieve addresses for various segments from kernel mach-o header
2026 	 */
2027 	segPRELINKTEXTB  = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__PRELINK_TEXT", &segSizePRELINKTEXT);
2028 	segPLKDATACONSTB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__PLK_DATA_CONST", &segSizePLKDATACONST);
2029 	segPLKTEXTEXECB  = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__PLK_TEXT_EXEC", &segSizePLKTEXTEXEC);
2030 	segTEXTB         = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__TEXT", &segSizeTEXT);
2031 	segDATACONSTB    = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__DATA_CONST", &segSizeDATACONST);
2032 	segTEXTEXECB     = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__TEXT_EXEC", &segSizeTEXTEXEC);
2033 #if XNU_MONITOR
2034 	segPPLTEXTB      = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__PPLTEXT", &segSizePPLTEXT);
2035 	segPPLTRAMPB     = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__PPLTRAMP", &segSizePPLTRAMP);
2036 	segPPLDATACONSTB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__PPLDATA_CONST", &segSizePPLDATACONST);
2037 #endif
2038 	segDATAB         = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__DATA", &segSizeDATA);
2039 #if XNU_MONITOR
2040 	segPPLDATAB      = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__PPLDATA", &segSizePPLDATA);
2041 #endif
2042 
2043 	segBOOTDATAB     = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__BOOTDATA", &segSizeBOOTDATA);
2044 	segLINKB         = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__LINKEDIT", &segSizeLINK);
2045 	segKLDB          = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__KLD", &segSizeKLD);
2046 	segKLDDATAB      = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__KLDDATA", &segSizeKLDDATA);
2047 	segPRELINKDATAB  = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__PRELINK_DATA", &segSizePRELINKDATA);
2048 	segPRELINKINFOB  = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__PRELINK_INFO", &segSizePRELINKINFO);
2049 	segPLKLLVMCOVB   = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__PLK_LLVM_COV", &segSizePLKLLVMCOV);
2050 	segPLKLINKEDITB  = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__PLK_LINKEDIT", &segSizePLKLINKEDIT);
2051 	segLASTB         = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__LAST", &segSizeLAST);
2052 	segLASTDATACONSTB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__LASTDATA_CONST", &segSizeLASTDATACONST);
2053 
2054 	sectHIBTEXTB     = (vm_offset_t) getsectdatafromheader(&_mh_execute_header, "__TEXT_EXEC", "__hib_text", &sectSizeHIBTEXT);
2055 	sectHIBDATACONSTB = (vm_offset_t) getsectdatafromheader(&_mh_execute_header, "__DATA_CONST", "__hib_const", &sectSizeHIBDATACONST);
2056 	segHIBDATAB      = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, "__HIBDATA", &segSizeHIBDATA);
2057 
2058 	if (kernel_mach_header_is_in_fileset(&_mh_execute_header)) {
2059 		kernel_mach_header_t *kc_mh = PE_get_kc_header(KCKindPrimary);
2060 
2061 		// fileset has kext PLK_TEXT_EXEC under kernel collection TEXT_EXEC following kernel's LAST
2062 		segKCTEXTEXECB = (vm_offset_t) getsegdatafromheader(kc_mh,             "__TEXT_EXEC", &segSizeKCTEXTEXEC);
2063 		assert(segPLKTEXTEXECB && !segSizePLKTEXTEXEC);                        // kernel PLK_TEXT_EXEC must be empty
2064 
2065 		assert(segLASTB);                                                      // kernel LAST can be empty, but it must have
2066 		                                                                       // a valid address for computations below.
2067 
2068 		assert(segKCTEXTEXECB <= segLASTB);                                    // KC TEXT_EXEC must contain kernel LAST
2069 		assert(segKCTEXTEXECB + segSizeKCTEXTEXEC >= segLASTB + segSizeLAST);
2070 		segPLKTEXTEXECB = segLASTB + segSizeLAST;
2071 		segSizePLKTEXTEXEC = segSizeKCTEXTEXEC - (segPLKTEXTEXECB - segKCTEXTEXECB);
2072 
2073 		// fileset has kext PLK_DATA_CONST under kernel collection DATA_CONST following kernel's LASTDATA_CONST
2074 		segKCDATACONSTB = (vm_offset_t) getsegdatafromheader(kc_mh,            "__DATA_CONST", &segSizeKCDATACONST);
2075 		assert(segPLKDATACONSTB && !segSizePLKDATACONST);                      // kernel PLK_DATA_CONST must be empty
2076 		assert(segLASTDATACONSTB && segSizeLASTDATACONST);                     // kernel LASTDATA_CONST must be non-empty
2077 		assert(segKCDATACONSTB <= segLASTDATACONSTB);                          // KC DATA_CONST must contain kernel LASTDATA_CONST
2078 		assert(segKCDATACONSTB + segSizeKCDATACONST >= segLASTDATACONSTB + segSizeLASTDATACONST);
2079 		segPLKDATACONSTB = segLASTDATACONSTB + segSizeLASTDATACONST;
2080 		segSizePLKDATACONST = segSizeKCDATACONST - (segPLKDATACONSTB - segKCDATACONSTB);
2081 
2082 		// fileset has kext PRELINK_DATA under kernel collection DATA following kernel's empty PRELINK_DATA
2083 		segKCDATAB      = (vm_offset_t) getsegdatafromheader(kc_mh,            "__DATA", &segSizeKCDATA);
2084 		assert(segPRELINKDATAB && !segSizePRELINKDATA);                        // kernel PRELINK_DATA must be empty
2085 		assert(segKCDATAB <= segPRELINKDATAB);                                 // KC DATA must contain kernel PRELINK_DATA
2086 		assert(segKCDATAB + segSizeKCDATA >= segPRELINKDATAB + segSizePRELINKDATA);
2087 		segSizePRELINKDATA = segSizeKCDATA - (segPRELINKDATAB - segKCDATAB);
2088 
2089 		// fileset has consolidated PRELINK_TEXT, PRELINK_INFO and LINKEDIT at the kernel collection level
2090 		assert(segPRELINKTEXTB && !segSizePRELINKTEXT);                        // kernel PRELINK_TEXT must be empty
2091 		segPRELINKTEXTB = (vm_offset_t) getsegdatafromheader(kc_mh,            "__PRELINK_TEXT", &segSizePRELINKTEXT);
2092 		assert(segPRELINKINFOB && !segSizePRELINKINFO);                        // kernel PRELINK_INFO must be empty
2093 		segPRELINKINFOB = (vm_offset_t) getsegdatafromheader(kc_mh,            "__PRELINK_INFO", &segSizePRELINKINFO);
2094 		segLINKB        = (vm_offset_t) getsegdatafromheader(kc_mh,            "__LINKEDIT", &segSizeLINK);
2095 	}
2096 
2097 	(void) PE_parse_boot_argn("use_contiguous_hint", &use_contiguous_hint, sizeof(use_contiguous_hint));
2098 	assert(segSizePRELINKTEXT < 0x03000000); /* 23355738 */
2099 
2100 	/* if one of the new segments is present, the other one better be as well */
2101 	if (segSizePLKDATACONST || segSizePLKTEXTEXEC) {
2102 		assert(segSizePLKDATACONST && segSizePLKTEXTEXEC);
2103 	}
2104 
2105 	etext = (vm_offset_t) segTEXTB + segSizeTEXT;
2106 	sdata = (vm_offset_t) segDATAB;
2107 	edata = (vm_offset_t) segDATAB + segSizeDATA;
2108 	end_kern = round_page(segHIGHESTKC ? segHIGHESTKC : getlastkerneladdr()); /* Force end to next page */
2109 
2110 	vm_set_page_size();
2111 
2112 	vm_kernel_base = segTEXTB;
2113 	vm_kernel_top = (vm_offset_t) &last_kernel_symbol;
2114 	vm_kext_base = segPRELINKTEXTB;
2115 	vm_kext_top = vm_kext_base + segSizePRELINKTEXT;
2116 
2117 	vm_prelink_stext = segPRELINKTEXTB;
2118 	if (!segSizePLKTEXTEXEC && !segSizePLKDATACONST) {
2119 		vm_prelink_etext = segPRELINKTEXTB + segSizePRELINKTEXT;
2120 	} else {
2121 		vm_prelink_etext = segPRELINKTEXTB + segSizePRELINKTEXT + segSizePLKDATACONST + segSizePLKTEXTEXEC;
2122 	}
2123 	vm_prelink_sinfo = segPRELINKINFOB;
2124 	vm_prelink_einfo = segPRELINKINFOB + segSizePRELINKINFO;
2125 	vm_slinkedit = segLINKB;
2126 	vm_elinkedit = segLINKB + segSizeLINK;
2127 
2128 	vm_prelink_sdata = segPRELINKDATAB;
2129 	vm_prelink_edata = segPRELINKDATAB + segSizePRELINKDATA;
2130 
2131 	arm_vm_prot_init(args);
2132 
2133 	/*
2134 	 * Initialize the page tables for the low globals:
2135 	 *   cover this address range:
2136 	 *     LOW_GLOBAL_BASE_ADDRESS + 2MB
2137 	 */
2138 	va_l1 = va_l2 = LOW_GLOBAL_BASE_ADDRESS;
2139 	cpu_l1_tte = cpu_tte + ((va_l1 & ARM_TT_L1_INDEX_MASK) >> ARM_TT_L1_SHIFT);
2140 	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);
2141 	ptpage_vaddr = alloc_ptpage(TRUE);
2142 	*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;
2143 	bzero((void *)ptpage_vaddr, ARM_PGBYTES);
2144 
2145 	/*
2146 	 * Initialize l2 page table pages :
2147 	 *   cover this address range:
2148 	 *    KERNEL_DYNAMIC_ADDR - VM_MAX_KERNEL_ADDRESS
2149 	 */
2150 #if defined(ARM_LARGE_MEMORY)
2151 	/*
2152 	 * dynamic mapped memory outside the VM allocator VA range required to bootstrap VM system
2153 	 * don't expect to exceed 64GB, no sense mapping any more space between here and the VM heap range
2154 	 */
2155 	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));
2156 #else
2157 	/*
2158 	 * TODO: do these pages really need to come from RO memory?
2159 	 * With legacy 3 level table systems we never mapped more than a single L1 entry so this may be dead code
2160 	 */
2161 	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));
2162 #endif
2163 
2164 #if KASAN
2165 	/* record the extent of the physmap */
2166 	physmap_vbase = physmap_base;
2167 	physmap_vtop = physmap_end;
2168 	kasan_init();
2169 #endif /* KASAN */
2170 
2171 #if CONFIG_CPU_COUNTERS
2172 	mt_early_init();
2173 #endif /* CONFIG_CPU_COUNTERS */
2174 
2175 	arm_set_user_tbi();
2176 
2177 	arm_vm_physmap_init(args);
2178 	set_mmu_ttb_alternate(cpu_ttep & TTBR_BADDR_MASK);
2179 
2180 	ml_enable_monitor();
2181 
2182 	set_mmu_ttb(invalid_ttep & TTBR_BADDR_MASK);
2183 
2184 	flush_mmu_tlb();
2185 	kva_active = TRUE;
2186 	// global table pointers may need to be different due to physical aperture remapping
2187 	cpu_tte = (tt_entry_t*)(phystokv(cpu_ttep));
2188 	invalid_tte = (tt_entry_t*)(phystokv(invalid_ttep));
2189 
2190 	// From here on out, we're off the bootstrap translation tables.
2191 
2192 
2193 	/* AuxKC initialization has to be deferred until this point, since
2194 	 * the AuxKC may not have been fully mapped in the bootstrap
2195 	 * tables, if it spilled downwards into the prior L2 block.
2196 	 *
2197 	 * Now that its mapping set up by arm_vm_prot_init() is active,
2198 	 * we can traverse and fix it up.
2199 	 */
2200 
2201 	/* Calculate the physical bounds of the kernelcache; using
2202 	 * gVirtBase/gPhysBase math to do this directly is generally a bad idea
2203 	 * as the physmap is no longer physically contiguous.  However, this is
2204 	 * done here as segLOWEST and end_kern are both virtual addresses the
2205 	 * bootstrap physmap, and because kvtophys references the page tables
2206 	 * (at least at the time this comment was written), meaning that at
2207 	 * least end_kern may not point to a valid mapping on some kernelcache
2208 	 * configurations, so kvtophys would report a physical address of 0.
2209 	 *
2210 	 * Long term, the kernelcache should probably be described in terms of
2211 	 * multiple physical ranges, as there is no strong guarantee or
2212 	 * requirement that the kernelcache will always be physically
2213 	 * contiguous.
2214 	 */
2215 	arm_vm_kernelcache_phys_start = segLOWEST - gVirtBase + gPhysBase;
2216 	arm_vm_kernelcache_phys_end = end_kern - gVirtBase + gPhysBase;;
2217 
2218 	/* Calculate the number of pages that belong to the kernelcache. */
2219 	vm_page_kernelcache_count = (unsigned int) (atop_64(arm_vm_kernelcache_phys_end - arm_vm_kernelcache_phys_start));
2220 
2221 	if (arm_vm_auxkc_init()) {
2222 		if (segLOWESTROAuxKC < segLOWESTRO) {
2223 			segLOWESTRO = segLOWESTROAuxKC;
2224 		}
2225 		if (segHIGHESTROAuxKC > segHIGHESTRO) {
2226 			segHIGHESTRO = segHIGHESTROAuxKC;
2227 		}
2228 		if (segLOWESTRXAuxKC < segLOWESTTEXT) {
2229 			segLOWESTTEXT = segLOWESTRXAuxKC;
2230 		}
2231 		assert(segLOWEST == segLOWESTAuxKC);
2232 
2233 		// The preliminary auxKC mapping has been broken up.
2234 		flush_mmu_tlb();
2235 	}
2236 
2237 	sane_size = mem_size - (avail_start - gPhysBase);
2238 	max_mem = mem_size;
2239 	vm_kernel_slid_base = segLOWESTTEXT;
2240 	// vm_kernel_slide is set by arm_init()->arm_slide_rebase_and_sign_image()
2241 	vm_kernel_stext = segTEXTB;
2242 
2243 	if (kernel_mach_header_is_in_fileset(&_mh_execute_header)) {
2244 		vm_kernel_etext = segTEXTEXECB + segSizeTEXTEXEC;
2245 		vm_kernel_slid_top = vm_slinkedit;
2246 	} else {
2247 		assert(segDATACONSTB == segTEXTB + segSizeTEXT);
2248 		assert(segTEXTEXECB == segDATACONSTB + segSizeDATACONST);
2249 		vm_kernel_etext = segTEXTB + segSizeTEXT + segSizeDATACONST + segSizeTEXTEXEC;
2250 		vm_kernel_slid_top = vm_prelink_einfo;
2251 	}
2252 
2253 	dynamic_memory_begin = ROUND_TWIG(dynamic_memory_begin);
2254 #if defined(KERNEL_INTEGRITY_CTRR) && defined(CONFIG_XNUPOST)
2255 	// reserve a 32MB region without permission overrides to use later for a CTRR unit test
2256 	{
2257 		extern vm_offset_t ctrr_test_page;
2258 		tt_entry_t *new_tte;
2259 
2260 		ctrr_test_page = dynamic_memory_begin;
2261 		dynamic_memory_begin += ARM_TT_L2_SIZE;
2262 		cpu_l1_tte = cpu_tte + ((ctrr_test_page & ARM_TT_L1_INDEX_MASK) >> ARM_TT_L1_SHIFT);
2263 		assert((*cpu_l1_tte) & ARM_TTE_VALID);
2264 		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);
2265 		assert((*cpu_l2_tte) == ARM_TTE_EMPTY);
2266 		new_tte = (tt_entry_t *)alloc_ptpage(FALSE);
2267 		bzero(new_tte, ARM_PGBYTES);
2268 		*cpu_l2_tte = (kvtophys((vm_offset_t)new_tte) & ARM_TTE_TABLE_MASK) | ARM_TTE_TYPE_TABLE | ARM_TTE_VALID;
2269 	}
2270 #endif /* defined(KERNEL_INTEGRITY_CTRR) && defined(CONFIG_XNUPOST) */
2271 #if XNU_MONITOR
2272 	for (vm_offset_t cur = (vm_offset_t)pmap_stacks_start; cur < (vm_offset_t)pmap_stacks_end; cur += ARM_PGBYTES) {
2273 		arm_vm_map(cpu_tte, cur, ARM_PTE_EMPTY);
2274 	}
2275 #if HAS_GUARDED_IO_FILTER
2276     for (vm_offset_t cur = (vm_offset_t)iofilter_stacks_start; cur < (vm_offset_t)iofilter_stacks_end; cur += ARM_PGBYTES) {
2277         arm_vm_map(cpu_tte, cur, ARM_PTE_EMPTY);
2278     }
2279 #endif
2280 #endif
2281 	pmap_bootstrap(dynamic_memory_begin);
2282 
2283 	disable_preemption();
2284 
2285 	/*
2286 	 * Initialize l3 page table pages :
2287 	 *   cover this address range:
2288 	 *    2MB + FrameBuffer size + 10MB for each 256MB segment
2289 	 */
2290 
2291 	mem_segments = (mem_size + 0x0FFFFFFF) >> 28;
2292 
2293 	va_l1 = dynamic_memory_begin;
2294 	va_l1_end = va_l1 + ((2 + (mem_segments * 10)) << 20);
2295 	va_l1_end += round_page(args->Video.v_height * args->Video.v_rowBytes);
2296 	va_l1_end = (va_l1_end + 0x00000000007FFFFFULL) & 0xFFFFFFFFFF800000ULL;
2297 
2298 	cpu_l1_tte = cpu_tte + ((va_l1 & ARM_TT_L1_INDEX_MASK) >> ARM_TT_L1_SHIFT);
2299 
2300 	while (va_l1 < va_l1_end) {
2301 		va_l2 = va_l1;
2302 
2303 		if (((va_l1 & ~ARM_TT_L1_OFFMASK) + ARM_TT_L1_SIZE) < va_l1) {
2304 			/* If this is the last L1 entry, it must cover the last mapping. */
2305 			va_l2_end = va_l1_end;
2306 		} else {
2307 			va_l2_end = MIN((va_l1 & ~ARM_TT_L1_OFFMASK) + ARM_TT_L1_SIZE, va_l1_end);
2308 		}
2309 
2310 		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);
2311 
2312 		while (va_l2 < va_l2_end) {
2313 			pt_entry_t *    ptp;
2314 			pmap_paddr_t    ptp_phys;
2315 
2316 			/* Allocate a page and setup L3 Table TTE in L2 */
2317 			ptp = (pt_entry_t *) alloc_ptpage(FALSE);
2318 			ptp_phys = (pmap_paddr_t)kvtophys((vm_offset_t)ptp);
2319 
2320 			bzero(ptp, ARM_PGBYTES);
2321 			pmap_init_pte_page(kernel_pmap, ptp, va_l2, 3, TRUE);
2322 
2323 			*cpu_l2_tte = (pa_to_tte(ptp_phys)) | ARM_TTE_TYPE_TABLE | ARM_TTE_VALID | ARM_DYNAMIC_TABLE_XN;
2324 
2325 			va_l2 += ARM_TT_L2_SIZE;
2326 			cpu_l2_tte++;
2327 		}
2328 
2329 		va_l1 = va_l2_end;
2330 		cpu_l1_tte++;
2331 	}
2332 
2333 #if defined(KERNEL_INTEGRITY_KTRR) || defined(KERNEL_INTEGRITY_CTRR)
2334 	/*
2335 	 * In this configuration, the bootstrap mappings (arm_vm_init) and
2336 	 * the heap mappings occupy separate L1 regions.  Explicitly set up
2337 	 * the heap L1 allocations here.
2338 	 */
2339 #if defined(ARM_LARGE_MEMORY)
2340 	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));
2341 #else // defined(ARM_LARGE_MEMORY)
2342 	va_l1 = VM_MIN_KERNEL_ADDRESS & ~ARM_TT_L1_OFFMASK;
2343 	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));
2344 #endif // defined(ARM_LARGE_MEMORY)
2345 #else
2346 #if defined(ARM_LARGE_MEMORY)
2347 	/* For large memory systems with no KTRR/CTRR such as virtual machines */
2348 	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));
2349 #endif
2350 #endif // defined(KERNEL_INTEGRITY_KTRR) || defined(KERNEL_INTEGRITY_CTRR)
2351 
2352 	/*
2353 	 * Initialize l3 page table pages :
2354 	 *   cover this address range:
2355 	 *   ((VM_MAX_KERNEL_ADDRESS & CPUWINDOWS_BASE_MASK) - PE_EARLY_BOOT_VA) to VM_MAX_KERNEL_ADDRESS
2356 	 */
2357 	va_l1 = (VM_MAX_KERNEL_ADDRESS & CPUWINDOWS_BASE_MASK) - PE_EARLY_BOOT_VA;
2358 	va_l1_end = VM_MAX_KERNEL_ADDRESS;
2359 
2360 	cpu_l1_tte = cpu_tte + ((va_l1 & ARM_TT_L1_INDEX_MASK) >> ARM_TT_L1_SHIFT);
2361 
2362 	while (va_l1 < va_l1_end) {
2363 		va_l2 = va_l1;
2364 
2365 		if (((va_l1 & ~ARM_TT_L1_OFFMASK) + ARM_TT_L1_SIZE) < va_l1) {
2366 			/* If this is the last L1 entry, it must cover the last mapping. */
2367 			va_l2_end = va_l1_end;
2368 		} else {
2369 			va_l2_end = MIN((va_l1 & ~ARM_TT_L1_OFFMASK) + ARM_TT_L1_SIZE, va_l1_end);
2370 		}
2371 
2372 		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);
2373 
2374 		while (va_l2 < va_l2_end) {
2375 			pt_entry_t *    ptp;
2376 			pmap_paddr_t    ptp_phys;
2377 
2378 			/* Allocate a page and setup L3 Table TTE in L2 */
2379 			ptp = (pt_entry_t *) alloc_ptpage(FALSE);
2380 			ptp_phys = (pmap_paddr_t)kvtophys((vm_offset_t)ptp);
2381 
2382 			bzero(ptp, ARM_PGBYTES);
2383 			pmap_init_pte_page(kernel_pmap, ptp, va_l2, 3, TRUE);
2384 
2385 			*cpu_l2_tte = (pa_to_tte(ptp_phys)) | ARM_TTE_TYPE_TABLE | ARM_TTE_VALID | ARM_DYNAMIC_TABLE_XN;
2386 
2387 			va_l2 += ARM_TT_L2_SIZE;
2388 			cpu_l2_tte++;
2389 		}
2390 
2391 		va_l1 = va_l2_end;
2392 		cpu_l1_tte++;
2393 	}
2394 
2395 
2396 	/*
2397 	 * Adjust avail_start so that the range that the VM owns
2398 	 * starts on a PAGE_SIZE aligned boundary.
2399 	 */
2400 	avail_start = (avail_start + PAGE_MASK) & ~PAGE_MASK;
2401 
2402 #if XNU_MONITOR
2403 	pmap_static_allocations_done();
2404 #endif
2405 	first_avail = avail_start;
2406 	patch_low_glo_static_region(args->topOfKernelData, avail_start - args->topOfKernelData);
2407 	enable_preemption();
2408 }
2409 
2410 /*
2411  * Returns true if the address is within __TEXT, __TEXT_EXEC or __DATA_CONST
2412  * segment range. This is what [vm_kernel_stext, vm_kernel_etext) range used to
2413  * cover. The segments together may not be continuous anymore and so individual
2414  * intervals are inspected.
2415  */
2416 bool
kernel_text_contains(vm_offset_t addr)2417 kernel_text_contains(vm_offset_t addr)
2418 {
2419 	if (segTEXTB <= addr && addr < (segTEXTB + segSizeTEXT)) {
2420 		return true;
2421 	}
2422 	if (segTEXTEXECB <= addr && addr < (segTEXTEXECB + segSizeTEXTEXEC)) {
2423 		return true;
2424 	}
2425 	return segDATACONSTB <= addr && addr < (segDATACONSTB + segSizeDATACONST);
2426 }
2427