1 // SPDX-License-Identifier: GPL-2.0
2 #define boot_fmt(fmt) "startup: " fmt
3 #include <linux/string.h>
4 #include <linux/elf.h>
5 #include <asm/page-states.h>
6 #include <asm/boot_data.h>
7 #include <asm/extmem.h>
8 #include <asm/sections.h>
9 #include <asm/maccess.h>
10 #include <asm/cpu_mf.h>
11 #include <asm/setup.h>
12 #include <asm/kasan.h>
13 #include <asm/kexec.h>
14 #include <asm/sclp.h>
15 #include <asm/diag.h>
16 #include <asm/uv.h>
17 #include <asm/abs_lowcore.h>
18 #include <asm/physmem_info.h>
19 #include "decompressor.h"
20 #include "boot.h"
21 #include "uv.h"
22
23 struct vm_layout __bootdata_preserved(vm_layout);
24 unsigned long __bootdata_preserved(__abs_lowcore);
25 unsigned long __bootdata_preserved(__memcpy_real_area);
26 pte_t *__bootdata_preserved(memcpy_real_ptep);
27 unsigned long __bootdata_preserved(VMALLOC_START);
28 unsigned long __bootdata_preserved(VMALLOC_END);
29 struct page *__bootdata_preserved(vmemmap);
30 unsigned long __bootdata_preserved(vmemmap_size);
31 unsigned long __bootdata_preserved(MODULES_VADDR);
32 unsigned long __bootdata_preserved(MODULES_END);
33 unsigned long __bootdata_preserved(max_mappable);
34 unsigned long __bootdata_preserved(page_noexec_mask);
35 unsigned long __bootdata_preserved(segment_noexec_mask);
36 unsigned long __bootdata_preserved(region_noexec_mask);
37 int __bootdata_preserved(relocate_lowcore);
38
39 u64 __bootdata_preserved(stfle_fac_list[16]);
40 struct oldmem_data __bootdata_preserved(oldmem_data);
41
42 struct machine_info machine;
43
error(char * x)44 void error(char *x)
45 {
46 boot_emerg("%s\n", x);
47 boot_emerg(" -- System halted\n");
48 disabled_wait();
49 }
50
detect_facilities(void)51 static void detect_facilities(void)
52 {
53 if (test_facility(8)) {
54 machine.has_edat1 = 1;
55 local_ctl_set_bit(0, CR0_EDAT_BIT);
56 }
57 if (test_facility(78))
58 machine.has_edat2 = 1;
59 page_noexec_mask = -1UL;
60 segment_noexec_mask = -1UL;
61 region_noexec_mask = -1UL;
62 if (!test_facility(130)) {
63 page_noexec_mask &= ~_PAGE_NOEXEC;
64 segment_noexec_mask &= ~_SEGMENT_ENTRY_NOEXEC;
65 region_noexec_mask &= ~_REGION_ENTRY_NOEXEC;
66 }
67 }
68
cmma_test_essa(void)69 static int cmma_test_essa(void)
70 {
71 unsigned long reg1, reg2, tmp = 0;
72 int rc = 1;
73 psw_t old;
74
75 /* Test ESSA_GET_STATE */
76 asm volatile(
77 " mvc 0(16,%[psw_old]),0(%[psw_pgm])\n"
78 " epsw %[reg1],%[reg2]\n"
79 " st %[reg1],0(%[psw_pgm])\n"
80 " st %[reg2],4(%[psw_pgm])\n"
81 " larl %[reg1],1f\n"
82 " stg %[reg1],8(%[psw_pgm])\n"
83 " .insn rrf,0xb9ab0000,%[tmp],%[tmp],%[cmd],0\n"
84 " la %[rc],0\n"
85 "1: mvc 0(16,%[psw_pgm]),0(%[psw_old])\n"
86 : [reg1] "=&d" (reg1),
87 [reg2] "=&a" (reg2),
88 [rc] "+&d" (rc),
89 [tmp] "+&d" (tmp),
90 "+Q" (get_lowcore()->program_new_psw),
91 "=Q" (old)
92 : [psw_old] "a" (&old),
93 [psw_pgm] "a" (&get_lowcore()->program_new_psw),
94 [cmd] "i" (ESSA_GET_STATE)
95 : "cc", "memory");
96 return rc;
97 }
98
cmma_init(void)99 static void cmma_init(void)
100 {
101 if (!cmma_flag)
102 return;
103 if (cmma_test_essa()) {
104 cmma_flag = 0;
105 return;
106 }
107 if (test_facility(147))
108 cmma_flag = 2;
109 }
110
setup_lpp(void)111 static void setup_lpp(void)
112 {
113 get_lowcore()->current_pid = 0;
114 get_lowcore()->lpp = LPP_MAGIC;
115 if (test_facility(40))
116 lpp(&get_lowcore()->lpp);
117 }
118
119 #ifdef CONFIG_KERNEL_UNCOMPRESSED
mem_safe_offset(void)120 static unsigned long mem_safe_offset(void)
121 {
122 return (unsigned long)_compressed_start;
123 }
124
deploy_kernel(void * output)125 static void deploy_kernel(void *output)
126 {
127 void *uncompressed_start = (void *)_compressed_start;
128
129 if (output == uncompressed_start)
130 return;
131 memmove(output, uncompressed_start, vmlinux.image_size);
132 memset(uncompressed_start, 0, vmlinux.image_size);
133 }
134 #endif
135
rescue_initrd(unsigned long min,unsigned long max)136 static void rescue_initrd(unsigned long min, unsigned long max)
137 {
138 unsigned long old_addr, addr, size;
139
140 if (!IS_ENABLED(CONFIG_BLK_DEV_INITRD))
141 return;
142 if (!get_physmem_reserved(RR_INITRD, &addr, &size))
143 return;
144 if (addr >= min && addr + size <= max)
145 return;
146 old_addr = addr;
147 physmem_free(RR_INITRD);
148 addr = physmem_alloc_or_die(RR_INITRD, size, 0);
149 memmove((void *)addr, (void *)old_addr, size);
150 }
151
copy_bootdata(void)152 static void copy_bootdata(void)
153 {
154 if (__boot_data_end - __boot_data_start != vmlinux.bootdata_size)
155 error(".boot.data section size mismatch");
156 memcpy((void *)vmlinux.bootdata_off, __boot_data_start, vmlinux.bootdata_size);
157 if (__boot_data_preserved_end - __boot_data_preserved_start != vmlinux.bootdata_preserved_size)
158 error(".boot.preserved.data section size mismatch");
159 memcpy((void *)vmlinux.bootdata_preserved_off, __boot_data_preserved_start, vmlinux.bootdata_preserved_size);
160 }
161
kaslr_adjust_relocs(unsigned long min_addr,unsigned long max_addr,unsigned long offset,unsigned long phys_offset)162 static void kaslr_adjust_relocs(unsigned long min_addr, unsigned long max_addr,
163 unsigned long offset, unsigned long phys_offset)
164 {
165 int *reloc;
166 long loc;
167
168 /* Adjust R_390_64 relocations */
169 for (reloc = (int *)__vmlinux_relocs_64_start; reloc < (int *)__vmlinux_relocs_64_end; reloc++) {
170 loc = (long)*reloc + phys_offset;
171 if (loc < min_addr || loc > max_addr)
172 error("64-bit relocation outside of kernel!\n");
173 *(u64 *)loc += offset;
174 }
175 }
176
kaslr_adjust_got(unsigned long offset)177 static void kaslr_adjust_got(unsigned long offset)
178 {
179 u64 *entry;
180
181 /*
182 * Adjust GOT entries, except for ones for undefined weak symbols
183 * that resolved to zero. This also skips the first three reserved
184 * entries on s390x that are zero.
185 */
186 for (entry = (u64 *)vmlinux.got_start; entry < (u64 *)vmlinux.got_end; entry++) {
187 if (*entry)
188 *entry += offset;
189 }
190 }
191
192 /*
193 * Merge information from several sources into a single ident_map_size value.
194 * "ident_map_size" represents the upper limit of physical memory we may ever
195 * reach. It might not be all online memory, but also include standby (offline)
196 * memory or memory areas reserved for other means (e.g., memory devices such as
197 * virtio-mem).
198 *
199 * "ident_map_size" could be lower then actual standby/reserved or even online
200 * memory present, due to limiting factors. We should never go above this limit.
201 * It is the size of our identity mapping.
202 *
203 * Consider the following factors:
204 * 1. max_physmem_end - end of physical memory online, standby or reserved.
205 * Always >= end of the last online memory range (get_physmem_online_end()).
206 * 2. CONFIG_MAX_PHYSMEM_BITS - the maximum size of physical memory the
207 * kernel is able to support.
208 * 3. "mem=" kernel command line option which limits physical memory usage.
209 * 4. OLDMEM_BASE which is a kdump memory limit when the kernel is executed as
210 * crash kernel.
211 * 5. "hsa" size which is a memory limit when the kernel is executed during
212 * zfcp/nvme dump.
213 */
setup_ident_map_size(unsigned long max_physmem_end)214 static void setup_ident_map_size(unsigned long max_physmem_end)
215 {
216 unsigned long hsa_size;
217
218 ident_map_size = max_physmem_end;
219 if (memory_limit)
220 ident_map_size = min(ident_map_size, memory_limit);
221 ident_map_size = min(ident_map_size, 1UL << MAX_PHYSMEM_BITS);
222
223 #ifdef CONFIG_CRASH_DUMP
224 if (oldmem_data.start) {
225 __kaslr_enabled = 0;
226 ident_map_size = min(ident_map_size, oldmem_data.size);
227 boot_debug("kdump memory limit: 0x%016lx\n", oldmem_data.size);
228 } else if (ipl_block_valid && is_ipl_block_dump()) {
229 __kaslr_enabled = 0;
230 if (!sclp_early_get_hsa_size(&hsa_size) && hsa_size) {
231 ident_map_size = min(ident_map_size, hsa_size);
232 boot_debug("Stand-alone dump limit: 0x%016lx\n", hsa_size);
233 }
234 }
235 #endif
236 boot_debug("Identity map size: 0x%016lx\n", ident_map_size);
237 }
238
239 #define FIXMAP_SIZE round_up(MEMCPY_REAL_SIZE + ABS_LOWCORE_MAP_SIZE, sizeof(struct lowcore))
240
get_vmem_size(unsigned long identity_size,unsigned long vmemmap_size,unsigned long vmalloc_size,unsigned long rte_size)241 static unsigned long get_vmem_size(unsigned long identity_size,
242 unsigned long vmemmap_size,
243 unsigned long vmalloc_size,
244 unsigned long rte_size)
245 {
246 unsigned long max_mappable, vsize;
247
248 max_mappable = max(identity_size, MAX_DCSS_ADDR);
249 vsize = round_up(SZ_2G + max_mappable, rte_size) +
250 round_up(vmemmap_size, rte_size) +
251 FIXMAP_SIZE + MODULES_LEN + KASLR_LEN;
252 if (IS_ENABLED(CONFIG_KMSAN))
253 vsize += MODULES_LEN * 2;
254 return size_add(vsize, vmalloc_size);
255 }
256
setup_kernel_memory_layout(unsigned long kernel_size)257 static unsigned long setup_kernel_memory_layout(unsigned long kernel_size)
258 {
259 unsigned long vmemmap_start;
260 unsigned long kernel_start;
261 unsigned long asce_limit;
262 unsigned long rte_size;
263 unsigned long pages;
264 unsigned long vsize;
265 unsigned long vmax;
266
267 pages = ident_map_size / PAGE_SIZE;
268 /* vmemmap contains a multiple of PAGES_PER_SECTION struct pages */
269 vmemmap_size = SECTION_ALIGN_UP(pages) * sizeof(struct page);
270
271 /* choose kernel address space layout: 4 or 3 levels. */
272 BUILD_BUG_ON(!IS_ALIGNED(TEXT_OFFSET, THREAD_SIZE));
273 BUILD_BUG_ON(!IS_ALIGNED(__NO_KASLR_START_KERNEL, THREAD_SIZE));
274 BUILD_BUG_ON(__NO_KASLR_END_KERNEL > _REGION1_SIZE);
275 vsize = get_vmem_size(ident_map_size, vmemmap_size, vmalloc_size, _REGION3_SIZE);
276 boot_debug("vmem size estimated: 0x%016lx\n", vsize);
277 if (IS_ENABLED(CONFIG_KASAN) || __NO_KASLR_END_KERNEL > _REGION2_SIZE ||
278 (vsize > _REGION2_SIZE && kaslr_enabled())) {
279 asce_limit = _REGION1_SIZE;
280 if (__NO_KASLR_END_KERNEL > _REGION2_SIZE) {
281 rte_size = _REGION2_SIZE;
282 vsize = get_vmem_size(ident_map_size, vmemmap_size, vmalloc_size, _REGION2_SIZE);
283 } else {
284 rte_size = _REGION3_SIZE;
285 }
286 } else {
287 asce_limit = _REGION2_SIZE;
288 rte_size = _REGION3_SIZE;
289 }
290
291 /*
292 * Forcing modules and vmalloc area under the ultravisor
293 * secure storage limit, so that any vmalloc allocation
294 * we do could be used to back secure guest storage.
295 *
296 * Assume the secure storage limit always exceeds _REGION2_SIZE,
297 * otherwise asce_limit and rte_size would have been adjusted.
298 */
299 vmax = adjust_to_uv_max(asce_limit);
300 boot_debug("%d level paging 0x%016lx vmax\n", vmax == _REGION1_SIZE ? 4 : 3, vmax);
301 #ifdef CONFIG_KASAN
302 BUILD_BUG_ON(__NO_KASLR_END_KERNEL > KASAN_SHADOW_START);
303 boot_debug("KASAN shadow area: 0x%016lx-0x%016lx\n", KASAN_SHADOW_START, KASAN_SHADOW_END);
304 /* force vmalloc and modules below kasan shadow */
305 vmax = min(vmax, KASAN_SHADOW_START);
306 #endif
307 vsize = min(vsize, vmax);
308 if (kaslr_enabled()) {
309 unsigned long kernel_end, kaslr_len, slots, pos;
310
311 kaslr_len = max(KASLR_LEN, vmax - vsize);
312 slots = DIV_ROUND_UP(kaslr_len - kernel_size, THREAD_SIZE);
313 if (get_random(slots, &pos))
314 pos = 0;
315 kernel_end = vmax - pos * THREAD_SIZE;
316 kernel_start = round_down(kernel_end - kernel_size, THREAD_SIZE);
317 boot_debug("Randomization range: 0x%016lx-0x%016lx\n", vmax - kaslr_len, vmax);
318 boot_debug("kernel image: 0x%016lx-0x%016lx (kaslr)\n", kernel_start,
319 kernel_size + kernel_size);
320 } else if (vmax < __NO_KASLR_END_KERNEL || vsize > __NO_KASLR_END_KERNEL) {
321 kernel_start = round_down(vmax - kernel_size, THREAD_SIZE);
322 boot_debug("kernel image: 0x%016lx-0x%016lx (constrained)\n", kernel_start,
323 kernel_start + kernel_size);
324 } else {
325 kernel_start = __NO_KASLR_START_KERNEL;
326 boot_debug("kernel image: 0x%016lx-0x%016lx (nokaslr)\n", kernel_start,
327 kernel_start + kernel_size);
328 }
329 __kaslr_offset = kernel_start;
330 boot_debug("__kaslr_offset: 0x%016lx\n", __kaslr_offset);
331
332 MODULES_END = round_down(kernel_start, _SEGMENT_SIZE);
333 MODULES_VADDR = MODULES_END - MODULES_LEN;
334 VMALLOC_END = MODULES_VADDR;
335 if (IS_ENABLED(CONFIG_KMSAN))
336 VMALLOC_END -= MODULES_LEN * 2;
337 boot_debug("modules area: 0x%016lx-0x%016lx\n", MODULES_VADDR, MODULES_END);
338
339 /* allow vmalloc area to occupy up to about 1/2 of the rest virtual space left */
340 vsize = (VMALLOC_END - FIXMAP_SIZE) / 2;
341 vsize = round_down(vsize, _SEGMENT_SIZE);
342 vmalloc_size = min(vmalloc_size, vsize);
343 if (IS_ENABLED(CONFIG_KMSAN)) {
344 /* take 2/3 of vmalloc area for KMSAN shadow and origins */
345 vmalloc_size = round_down(vmalloc_size / 3, _SEGMENT_SIZE);
346 VMALLOC_END -= vmalloc_size * 2;
347 }
348 VMALLOC_START = VMALLOC_END - vmalloc_size;
349 boot_debug("vmalloc area: 0x%016lx-0x%016lx\n", VMALLOC_START, VMALLOC_END);
350
351 __memcpy_real_area = round_down(VMALLOC_START - MEMCPY_REAL_SIZE, PAGE_SIZE);
352 boot_debug("memcpy real area: 0x%016lx-0x%016lx\n", __memcpy_real_area,
353 __memcpy_real_area + MEMCPY_REAL_SIZE);
354 __abs_lowcore = round_down(__memcpy_real_area - ABS_LOWCORE_MAP_SIZE,
355 sizeof(struct lowcore));
356 boot_debug("abs lowcore: 0x%016lx-0x%016lx\n", __abs_lowcore,
357 __abs_lowcore + ABS_LOWCORE_MAP_SIZE);
358
359 /* split remaining virtual space between 1:1 mapping & vmemmap array */
360 pages = __abs_lowcore / (PAGE_SIZE + sizeof(struct page));
361 pages = SECTION_ALIGN_UP(pages);
362 /* keep vmemmap_start aligned to a top level region table entry */
363 vmemmap_start = round_down(__abs_lowcore - pages * sizeof(struct page), rte_size);
364 /* make sure identity map doesn't overlay with vmemmap */
365 ident_map_size = min(ident_map_size, vmemmap_start);
366 vmemmap_size = SECTION_ALIGN_UP(ident_map_size / PAGE_SIZE) * sizeof(struct page);
367 /* make sure vmemmap doesn't overlay with absolute lowcore area */
368 if (vmemmap_start + vmemmap_size > __abs_lowcore) {
369 vmemmap_size = SECTION_ALIGN_DOWN(ident_map_size / PAGE_SIZE) * sizeof(struct page);
370 ident_map_size = vmemmap_size / sizeof(struct page) * PAGE_SIZE;
371 }
372 vmemmap = (struct page *)vmemmap_start;
373 /* maximum address for which linear mapping could be created (DCSS, memory) */
374 BUILD_BUG_ON(MAX_DCSS_ADDR > (1UL << MAX_PHYSMEM_BITS));
375 max_mappable = max(ident_map_size, MAX_DCSS_ADDR);
376 max_mappable = min(max_mappable, vmemmap_start);
377 #ifdef CONFIG_RANDOMIZE_IDENTITY_BASE
378 __identity_base = round_down(vmemmap_start - max_mappable, rte_size);
379 #endif
380 boot_debug("identity map: 0x%016lx-0x%016lx\n", __identity_base,
381 __identity_base + ident_map_size);
382
383 return asce_limit;
384 }
385
386 /*
387 * This function clears the BSS section of the decompressed Linux kernel and NOT the decompressor's.
388 */
clear_bss_section(unsigned long kernel_start)389 static void clear_bss_section(unsigned long kernel_start)
390 {
391 memset((void *)kernel_start + vmlinux.image_size, 0, vmlinux.bss_size);
392 }
393
394 /*
395 * Set vmalloc area size to an 8th of (potential) physical memory
396 * size, unless size has been set by kernel command line parameter.
397 */
setup_vmalloc_size(void)398 static void setup_vmalloc_size(void)
399 {
400 unsigned long size;
401
402 if (vmalloc_size_set)
403 return;
404 size = round_up(ident_map_size / 8, _SEGMENT_SIZE);
405 vmalloc_size = max(size, vmalloc_size);
406 }
407
kaslr_adjust_vmlinux_info(long offset)408 static void kaslr_adjust_vmlinux_info(long offset)
409 {
410 vmlinux.bootdata_off += offset;
411 vmlinux.bootdata_preserved_off += offset;
412 vmlinux.got_start += offset;
413 vmlinux.got_end += offset;
414 vmlinux.init_mm_off += offset;
415 vmlinux.swapper_pg_dir_off += offset;
416 vmlinux.invalid_pg_dir_off += offset;
417 vmlinux.alt_instructions += offset;
418 vmlinux.alt_instructions_end += offset;
419 #ifdef CONFIG_KASAN
420 vmlinux.kasan_early_shadow_page_off += offset;
421 vmlinux.kasan_early_shadow_pte_off += offset;
422 vmlinux.kasan_early_shadow_pmd_off += offset;
423 vmlinux.kasan_early_shadow_pud_off += offset;
424 vmlinux.kasan_early_shadow_p4d_off += offset;
425 #endif
426 }
427
startup_kernel(void)428 void startup_kernel(void)
429 {
430 unsigned long vmlinux_size = vmlinux.image_size + vmlinux.bss_size;
431 unsigned long nokaslr_text_lma, text_lma = 0, amode31_lma = 0;
432 unsigned long kernel_size = TEXT_OFFSET + vmlinux_size;
433 unsigned long kaslr_large_page_offset;
434 unsigned long max_physmem_end;
435 unsigned long asce_limit;
436 unsigned long safe_addr;
437 psw_t psw;
438
439 setup_lpp();
440 store_ipl_parmblock();
441 uv_query_info();
442 setup_boot_command_line();
443 parse_boot_command_line();
444
445 /*
446 * Non-randomized kernel physical start address must be _SEGMENT_SIZE
447 * aligned (see blow).
448 */
449 nokaslr_text_lma = ALIGN(mem_safe_offset(), _SEGMENT_SIZE);
450 safe_addr = PAGE_ALIGN(nokaslr_text_lma + vmlinux_size);
451
452 /*
453 * Reserve decompressor memory together with decompression heap,
454 * buffer and memory which might be occupied by uncompressed kernel
455 * (if KASLR is off or failed).
456 */
457 physmem_reserve(RR_DECOMPRESSOR, 0, safe_addr);
458 if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && parmarea.initrd_size)
459 physmem_reserve(RR_INITRD, parmarea.initrd_start, parmarea.initrd_size);
460 oldmem_data.start = parmarea.oldmem_base;
461 oldmem_data.size = parmarea.oldmem_size;
462
463 read_ipl_report();
464 sclp_early_read_info();
465 detect_facilities();
466 cmma_init();
467 sanitize_prot_virt_host();
468 max_physmem_end = detect_max_physmem_end();
469 setup_ident_map_size(max_physmem_end);
470 setup_vmalloc_size();
471 asce_limit = setup_kernel_memory_layout(kernel_size);
472 /* got final ident_map_size, physmem allocations could be performed now */
473 physmem_set_usable_limit(ident_map_size);
474 detect_physmem_online_ranges(max_physmem_end);
475 save_ipl_cert_comp_list();
476 rescue_initrd(safe_addr, ident_map_size);
477
478 /*
479 * __kaslr_offset_phys must be _SEGMENT_SIZE aligned, so the lower
480 * 20 bits (the offset within a large page) are zero. Copy the last
481 * 20 bits of __kaslr_offset, which is THREAD_SIZE aligned, to
482 * __kaslr_offset_phys.
483 *
484 * With this the last 20 bits of __kaslr_offset_phys and __kaslr_offset
485 * are identical, which is required to allow for large mappings of the
486 * kernel image.
487 */
488 kaslr_large_page_offset = __kaslr_offset & ~_SEGMENT_MASK;
489 if (kaslr_enabled()) {
490 unsigned long size = vmlinux_size + kaslr_large_page_offset;
491
492 text_lma = randomize_within_range(size, _SEGMENT_SIZE, TEXT_OFFSET, ident_map_size);
493 }
494 if (!text_lma)
495 text_lma = nokaslr_text_lma;
496 text_lma |= kaslr_large_page_offset;
497
498 /*
499 * [__kaslr_offset_phys..__kaslr_offset_phys + TEXT_OFFSET] region is
500 * never accessed via the kernel image mapping as per the linker script:
501 *
502 * . = TEXT_OFFSET;
503 *
504 * Therefore, this region could be used for something else and does
505 * not need to be reserved. See how it is skipped in setup_vmem().
506 */
507 __kaslr_offset_phys = text_lma - TEXT_OFFSET;
508 kaslr_adjust_vmlinux_info(__kaslr_offset_phys);
509 physmem_reserve(RR_VMLINUX, text_lma, vmlinux_size);
510 deploy_kernel((void *)text_lma);
511
512 /* vmlinux decompression is done, shrink reserved low memory */
513 physmem_reserve(RR_DECOMPRESSOR, 0, (unsigned long)_decompressor_end);
514
515 /*
516 * In case KASLR is enabled the randomized location of .amode31
517 * section might overlap with .vmlinux.relocs section. To avoid that
518 * the below randomize_within_range() could have been called with
519 * __vmlinux_relocs_64_end as the lower range address. However,
520 * .amode31 section is written to by the decompressed kernel - at
521 * that time the contents of .vmlinux.relocs is not needed anymore.
522 * Conversely, .vmlinux.relocs is read only by the decompressor, even
523 * before the kernel started. Therefore, in case the two sections
524 * overlap there is no risk of corrupting any data.
525 */
526 if (kaslr_enabled()) {
527 unsigned long amode31_min;
528
529 amode31_min = (unsigned long)_decompressor_end;
530 amode31_lma = randomize_within_range(vmlinux.amode31_size, PAGE_SIZE, amode31_min, SZ_2G);
531 }
532 if (!amode31_lma)
533 amode31_lma = text_lma - vmlinux.amode31_size;
534 physmem_reserve(RR_AMODE31, amode31_lma, vmlinux.amode31_size);
535
536 /*
537 * The order of the following operations is important:
538 *
539 * - kaslr_adjust_relocs() must follow clear_bss_section() to establish
540 * static memory references to data in .bss to be used by setup_vmem()
541 * (i.e init_mm.pgd)
542 *
543 * - setup_vmem() must follow kaslr_adjust_relocs() to be able using
544 * static memory references to data in .bss (i.e init_mm.pgd)
545 *
546 * - copy_bootdata() must follow setup_vmem() to propagate changes
547 * to bootdata made by setup_vmem()
548 */
549 clear_bss_section(text_lma);
550 kaslr_adjust_relocs(text_lma, text_lma + vmlinux.image_size,
551 __kaslr_offset, __kaslr_offset_phys);
552 kaslr_adjust_got(__kaslr_offset);
553 setup_vmem(__kaslr_offset, __kaslr_offset + kernel_size, asce_limit);
554 dump_physmem_reserved();
555 copy_bootdata();
556 __apply_alternatives((struct alt_instr *)_vmlinux_info.alt_instructions,
557 (struct alt_instr *)_vmlinux_info.alt_instructions_end,
558 ALT_CTX_EARLY);
559
560 /*
561 * Save KASLR offset for early dumps, before vmcore_info is set.
562 * Mark as uneven to distinguish from real vmcore_info pointer.
563 */
564 get_lowcore()->vmcore_info = __kaslr_offset_phys ? __kaslr_offset_phys | 0x1UL : 0;
565
566 /*
567 * Jump to the decompressed kernel entry point and switch DAT mode on.
568 */
569 psw.addr = __kaslr_offset + vmlinux.entry;
570 psw.mask = PSW_KERNEL_BITS;
571 boot_debug("Starting kernel at: 0x%016lx\n", psw.addr);
572 __load_psw(psw);
573 }
574