startup.c

  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/diag288.h>
 10#include <asm/maccess.h>
 11#include <asm/machine.h>
 12#include <asm/sysinfo.h>
 13#include <asm/cpu_mf.h>
 14#include <asm/setup.h>
 15#include <asm/timex.h>
 16#include <asm/kasan.h>
 17#include <asm/kexec.h>
 18#include <asm/sclp.h>
 19#include <asm/diag.h>
 20#include <asm/uv.h>
 21#include <asm/abs_lowcore.h>
 22#include <asm/physmem_info.h>
 23#include <asm/stacktrace.h>
 24#include <asm/asm-offsets.h>
 25#include <asm/arch-stackprotector.h>
 26#include "decompressor.h"
 27#include "boot.h"
 28#include "uv.h"
 29
 30struct vm_layout __bootdata_preserved(vm_layout);
 31unsigned long __bootdata_preserved(__abs_lowcore);
 32unsigned long __bootdata_preserved(__memcpy_real_area);
 33pte_t *__bootdata_preserved(memcpy_real_ptep);
 34unsigned long __bootdata_preserved(VMALLOC_START);
 35unsigned long __bootdata_preserved(VMALLOC_END);
 36struct page *__bootdata_preserved(vmemmap);
 37unsigned long __bootdata_preserved(vmemmap_size);
 38unsigned long __bootdata_preserved(MODULES_VADDR);
 39unsigned long __bootdata_preserved(MODULES_END);
 40unsigned long __bootdata_preserved(max_mappable);
 41unsigned long __bootdata_preserved(page_noexec_mask);
 42unsigned long __bootdata_preserved(segment_noexec_mask);
 43unsigned long __bootdata_preserved(region_noexec_mask);
 44union tod_clock __bootdata_preserved(tod_clock_base);
 45u64 __bootdata_preserved(clock_comparator_max) = -1UL;
 46
 47u64 __bootdata_preserved(stfle_fac_list[16]);
 48struct oldmem_data __bootdata_preserved(oldmem_data);
 49
 50static char sysinfo_page[PAGE_SIZE] __aligned(PAGE_SIZE);
 51
 52static void detect_machine_type(void)
 53{
 54	struct sysinfo_3_2_2 *vmms = (struct sysinfo_3_2_2 *)&sysinfo_page;
 55
 56	/* Check current-configuration-level */
 57	if (stsi(NULL, 0, 0, 0) <= 2) {
 58		set_machine_feature(MFEATURE_LPAR);
 59		return;
 60	}
 61	/* Get virtual-machine cpu information. */
 62	if (stsi(vmms, 3, 2, 2) || !vmms->count)
 63		return;
 64	/* Detect known hypervisors */
 65	if (!memcmp(vmms->vm[0].cpi, "\xd2\xe5\xd4", 3))
 66		set_machine_feature(MFEATURE_KVM);
 67	else if (!memcmp(vmms->vm[0].cpi, "\xa9\x61\xe5\xd4", 4))
 68		set_machine_feature(MFEATURE_VM);
 69}
 70
 71static void detect_diag288(void)
 72{
 73	/* "BEGIN" in EBCDIC character set */
 74	static const char cmd[] = "\xc2\xc5\xc7\xc9\xd5";
 75	unsigned long action, len;
 76
 77	action = machine_is_vm() ? (unsigned long)cmd : LPARWDT_RESTART;
 78	len = machine_is_vm() ? sizeof(cmd) : 0;
 79	if (__diag288(WDT_FUNC_INIT, MIN_INTERVAL, action, len))
 80		return;
 81	__diag288(WDT_FUNC_CANCEL, 0, 0, 0);
 82	set_machine_feature(MFEATURE_DIAG288);
 83}
 84
 85static void detect_diag9c(void)
 86{
 87	unsigned int cpu;
 88	int rc = 1;
 89
 90	cpu = stap();
 91	asm_inline volatile(
 92		"	diag	%[cpu],%%r0,0x9c\n"
 93		"0:	lhi	%[rc],0\n"
 94		"1:\n"
 95		EX_TABLE(0b, 1b)
 96		: [rc] "+d" (rc)
 97		: [cpu] "d" (cpu)
 98		: "cc", "memory");
 99	if (!rc)
100		set_machine_feature(MFEATURE_DIAG9C);
101}
102
103static void reset_tod_clock(void)
104{
105	union tod_clock clk;
106
107	if (store_tod_clock_ext_cc(&clk) == 0)
108		return;
109	/* TOD clock not running. Set the clock to Unix Epoch. */
110	if (set_tod_clock(TOD_UNIX_EPOCH) || store_tod_clock_ext_cc(&clk))
111		disabled_wait();
112	memset(&tod_clock_base, 0, sizeof(tod_clock_base));
113	tod_clock_base.tod = TOD_UNIX_EPOCH;
114	get_lowcore()->last_update_clock = TOD_UNIX_EPOCH;
115}
116
117static void detect_facilities(void)
118{
119	if (cpu_has_edat1())
120		local_ctl_set_bit(0, CR0_EDAT_BIT);
121	page_noexec_mask = -1UL;
122	segment_noexec_mask = -1UL;
123	region_noexec_mask = -1UL;
124	if (!cpu_has_nx()) {
125		page_noexec_mask &= ~_PAGE_NOEXEC;
126		segment_noexec_mask &= ~_SEGMENT_ENTRY_NOEXEC;
127		region_noexec_mask &= ~_REGION_ENTRY_NOEXEC;
128	}
129	if (IS_ENABLED(CONFIG_PCI) && test_facility(153))
130		set_machine_feature(MFEATURE_PCI_MIO);
131	reset_tod_clock();
132	if (test_facility(139) && (tod_clock_base.tod >> 63)) {
133		/* Enable signed clock comparator comparisons */
134		set_machine_feature(MFEATURE_SCC);
135		clock_comparator_max = -1UL >> 1;
136		local_ctl_set_bit(0, CR0_CLOCK_COMPARATOR_SIGN_BIT);
137	}
138	if (test_facility(50) && test_facility(73)) {
139		set_machine_feature(MFEATURE_TX);
140		local_ctl_set_bit(0, CR0_TRANSACTIONAL_EXECUTION_BIT);
141	}
142	if (cpu_has_vx())
143		local_ctl_set_bit(0, CR0_VECTOR_BIT);
144}
145
146static int cmma_test_essa(void)
147{
148	unsigned long tmp = 0;
149	int rc = 1;
150
151	/* Test ESSA_GET_STATE */
152	asm_inline volatile(
153		"	.insn	rrf,0xb9ab0000,%[tmp],%[tmp],%[cmd],0\n"
154		"0:	lhi	%[rc],0\n"
155		"1:\n"
156		EX_TABLE(0b, 1b)
157		: [rc] "+d" (rc), [tmp] "+d" (tmp)
158		: [cmd] "i" (ESSA_GET_STATE)
159		: "cc", "memory");
160	return rc;
161}
162
163static void cmma_init(void)
164{
165	if (!cmma_flag)
166		return;
167	if (cmma_test_essa()) {
168		cmma_flag = 0;
169		return;
170	}
171	if (test_facility(147))
172		cmma_flag = 2;
173}
174
175static void setup_lpp(void)
176{
177	get_lowcore()->current_pid = 0;
178	get_lowcore()->lpp = LPP_MAGIC;
179	if (test_facility(40))
180		lpp(&get_lowcore()->lpp);
181}
182
183#ifdef CONFIG_KERNEL_UNCOMPRESSED
184static unsigned long mem_safe_offset(void)
185{
186	return (unsigned long)_compressed_start;
187}
188
189static void deploy_kernel(void *output)
190{
191	void *uncompressed_start = (void *)_compressed_start;
192
193	if (output == uncompressed_start)
194		return;
195	memmove(output, uncompressed_start, vmlinux.image_size);
196	memset(uncompressed_start, 0, vmlinux.image_size);
197}
198#endif
199
200static void rescue_initrd(unsigned long min, unsigned long max)
201{
202	unsigned long old_addr, addr, size;
203
204	if (!IS_ENABLED(CONFIG_BLK_DEV_INITRD))
205		return;
206	if (!get_physmem_reserved(RR_INITRD, &addr, &size))
207		return;
208	if (addr >= min && addr + size <= max)
209		return;
210	old_addr = addr;
211	physmem_free(RR_INITRD);
212	addr = physmem_alloc_or_die(RR_INITRD, size, 0);
213	memmove((void *)addr, (void *)old_addr, size);
214}
215
216static void copy_bootdata(void)
217{
218	if (__boot_data_end - __boot_data_start != vmlinux.bootdata_size)
219		boot_panic(".boot.data section size mismatch\n");
220	memcpy((void *)vmlinux.bootdata_off, __boot_data_start, vmlinux.bootdata_size);
221	if (__boot_data_preserved_end - __boot_data_preserved_start != vmlinux.bootdata_preserved_size)
222		boot_panic(".boot.preserved.data section size mismatch\n");
223	memcpy((void *)vmlinux.bootdata_preserved_off, __boot_data_preserved_start, vmlinux.bootdata_preserved_size);
224}
225
226static void kaslr_adjust_relocs(unsigned long min_addr, unsigned long max_addr,
227				unsigned long offset, unsigned long phys_offset)
228{
229	int *reloc;
230	long loc;
231
232	/* Adjust R_390_64 relocations */
233	for (reloc = (int *)__vmlinux_relocs_64_start; reloc < (int *)__vmlinux_relocs_64_end; reloc++) {
234		loc = (long)*reloc + phys_offset;
235		if (loc < min_addr || loc > max_addr)
236			boot_panic("64-bit relocation outside of kernel!\n");
237		*(u64 *)loc += offset;
238	}
239}
240
241static void kaslr_adjust_got(unsigned long offset)
242{
243	u64 *entry;
244
245	/*
246	 * Adjust GOT entries, except for ones for undefined weak symbols
247	 * that resolved to zero. This also skips the first three reserved
248	 * entries on s390x that are zero.
249	 */
250	for (entry = (u64 *)vmlinux.got_start; entry < (u64 *)vmlinux.got_end; entry++) {
251		if (*entry)
252			*entry += offset;
253	}
254}
255
256/*
257 * Merge information from several sources into a single ident_map_size value.
258 * "ident_map_size" represents the upper limit of physical memory we may ever
259 * reach. It might not be all online memory, but also include standby (offline)
260 * memory or memory areas reserved for other means (e.g., memory devices such as
261 * virtio-mem).
262 *
263 * "ident_map_size" could be lower then actual standby/reserved or even online
264 * memory present, due to limiting factors. We should never go above this limit.
265 * It is the size of our identity mapping.
266 *
267 * Consider the following factors:
268 * 1. max_physmem_end - end of physical memory online, standby or reserved.
269 *    Always >= end of the last online memory range (get_physmem_online_end()).
270 * 2. CONFIG_MAX_PHYSMEM_BITS - the maximum size of physical memory the
271 *    kernel is able to support.
272 * 3. "mem=" kernel command line option which limits physical memory usage.
273 * 4. OLDMEM_BASE which is a kdump memory limit when the kernel is executed as
274 *    crash kernel.
275 * 5. "hsa" size which is a memory limit when the kernel is executed during
276 *    zfcp/nvme dump.
277 */
278static void setup_ident_map_size(unsigned long max_physmem_end)
279{
280	unsigned long hsa_size;
281
282	ident_map_size = max_physmem_end;
283	if (memory_limit)
284		ident_map_size = min(ident_map_size, memory_limit);
285	ident_map_size = min(ident_map_size, 1UL << MAX_PHYSMEM_BITS);
286
287#ifdef CONFIG_CRASH_DUMP
288	if (oldmem_data.start) {
289		__kaslr_enabled = 0;
290		ident_map_size = min(ident_map_size, oldmem_data.size);
291		boot_debug("kdump memory limit:  0x%016lx\n", oldmem_data.size);
292	} else if (ipl_block_valid && is_ipl_block_dump()) {
293		__kaslr_enabled = 0;
294		if (!sclp_early_get_hsa_size(&hsa_size) && hsa_size) {
295			ident_map_size = min(ident_map_size, hsa_size);
296			boot_debug("Stand-alone dump limit: 0x%016lx\n", hsa_size);
297		}
298	}
299#endif
300	boot_debug("Identity map size:   0x%016lx\n", ident_map_size);
301}
302
303#define FIXMAP_SIZE	round_up(MEMCPY_REAL_SIZE + ABS_LOWCORE_MAP_SIZE, sizeof(struct lowcore))
304
305static unsigned long get_vmem_size(unsigned long identity_size,
306				   unsigned long vmemmap_size,
307				   unsigned long vmalloc_size,
308				   unsigned long rte_size)
309{
310	unsigned long max_mappable, vsize;
311
312	max_mappable = max(identity_size, MAX_DCSS_ADDR);
313	vsize = round_up(SZ_2G + max_mappable, rte_size) +
314		round_up(vmemmap_size, rte_size) +
315		FIXMAP_SIZE + MODULES_LEN + KASLR_LEN;
316	if (IS_ENABLED(CONFIG_KMSAN))
317		vsize += MODULES_LEN * 2;
318	return size_add(vsize, vmalloc_size);
319}
320
321static unsigned long setup_kernel_memory_layout(unsigned long kernel_size)
322{
323	unsigned long vmemmap_start;
324	unsigned long kernel_start;
325	unsigned long asce_limit;
326	unsigned long rte_size;
327	unsigned long pages;
328	unsigned long vsize;
329	unsigned long vmax;
330
331	pages = ident_map_size / PAGE_SIZE;
332	/* vmemmap contains a multiple of PAGES_PER_SECTION struct pages */
333	vmemmap_size = SECTION_ALIGN_UP(pages) * sizeof(struct page);
334
335	/* choose kernel address space layout: 4 or 3 levels. */
336	BUILD_BUG_ON(!IS_ALIGNED(TEXT_OFFSET, THREAD_SIZE));
337	BUILD_BUG_ON(!IS_ALIGNED(__NO_KASLR_START_KERNEL, THREAD_SIZE));
338	BUILD_BUG_ON(__NO_KASLR_END_KERNEL > _REGION1_SIZE);
339	vsize = get_vmem_size(ident_map_size, vmemmap_size, vmalloc_size, _REGION3_SIZE);
340	boot_debug("vmem size estimated: 0x%016lx\n", vsize);
341	if (IS_ENABLED(CONFIG_KASAN) || __NO_KASLR_END_KERNEL > _REGION2_SIZE ||
342	    (vsize > _REGION2_SIZE && kaslr_enabled())) {
343		asce_limit = _REGION1_SIZE;
344		if (__NO_KASLR_END_KERNEL > _REGION2_SIZE) {
345			rte_size = _REGION2_SIZE;
346			vsize = get_vmem_size(ident_map_size, vmemmap_size, vmalloc_size, _REGION2_SIZE);
347		} else {
348			rte_size = _REGION3_SIZE;
349		}
350	} else {
351		asce_limit = _REGION2_SIZE;
352		rte_size = _REGION3_SIZE;
353	}
354
355	/*
356	 * Forcing modules and vmalloc area under the ultravisor
357	 * secure storage limit, so that any vmalloc allocation
358	 * we do could be used to back secure guest storage.
359	 *
360	 * Assume the secure storage limit always exceeds _REGION2_SIZE,
361	 * otherwise asce_limit and rte_size would have been adjusted.
362	 */
363	vmax = adjust_to_uv_max(asce_limit);
364	boot_debug("%d level paging       0x%016lx vmax\n", vmax == _REGION1_SIZE ? 4 : 3, vmax);
365#ifdef CONFIG_KASAN
366	BUILD_BUG_ON(__NO_KASLR_END_KERNEL > KASAN_SHADOW_START);
367	boot_debug("KASAN shadow area:   0x%016lx-0x%016lx\n", KASAN_SHADOW_START, KASAN_SHADOW_END);
368	/* force vmalloc and modules below kasan shadow */
369	vmax = min(vmax, KASAN_SHADOW_START);
370#endif
371	vsize = min(vsize, vmax);
372	if (kaslr_enabled()) {
373		unsigned long kernel_end, kaslr_len, slots, pos;
374
375		kaslr_len = max(KASLR_LEN, vmax - vsize);
376		slots = DIV_ROUND_UP(kaslr_len - kernel_size, THREAD_SIZE);
377		if (get_random(slots, &pos))
378			pos = 0;
379		kernel_end = vmax - pos * THREAD_SIZE;
380		kernel_start = round_down(kernel_end - kernel_size, THREAD_SIZE);
381		boot_debug("Randomization range: 0x%016lx-0x%016lx\n", vmax - kaslr_len, vmax);
382		boot_debug("kernel image:        0x%016lx-0x%016lx (kaslr)\n", kernel_start,
383			   kernel_start + kernel_size);
384	} else if (vmax < __NO_KASLR_END_KERNEL || vsize > __NO_KASLR_END_KERNEL) {
385		kernel_start = round_down(vmax - kernel_size, THREAD_SIZE);
386		boot_debug("kernel image:        0x%016lx-0x%016lx (constrained)\n", kernel_start,
387			   kernel_start + kernel_size);
388	} else {
389		kernel_start = __NO_KASLR_START_KERNEL;
390		boot_debug("kernel image:        0x%016lx-0x%016lx (nokaslr)\n", kernel_start,
391			   kernel_start + kernel_size);
392	}
393	__kaslr_offset = kernel_start;
394	boot_debug("__kaslr_offset:      0x%016lx\n", __kaslr_offset);
395
396	MODULES_END = round_down(kernel_start, _SEGMENT_SIZE);
397	MODULES_VADDR = MODULES_END - MODULES_LEN;
398	VMALLOC_END = MODULES_VADDR;
399	if (IS_ENABLED(CONFIG_KMSAN))
400		VMALLOC_END -= MODULES_LEN * 2;
401	boot_debug("modules area:        0x%016lx-0x%016lx\n", MODULES_VADDR, MODULES_END);
402
403	/* allow vmalloc area to occupy up to about 1/2 of the rest virtual space left */
404	vsize = (VMALLOC_END - FIXMAP_SIZE) / 2;
405	vsize = round_down(vsize, _SEGMENT_SIZE);
406	vmalloc_size = min(vmalloc_size, vsize);
407	if (IS_ENABLED(CONFIG_KMSAN)) {
408		/* take 2/3 of vmalloc area for KMSAN shadow and origins */
409		vmalloc_size = round_down(vmalloc_size / 3, _SEGMENT_SIZE);
410		VMALLOC_END -= vmalloc_size * 2;
411	}
412	VMALLOC_START = VMALLOC_END - vmalloc_size;
413	boot_debug("vmalloc area:        0x%016lx-0x%016lx\n", VMALLOC_START, VMALLOC_END);
414
415	__memcpy_real_area = round_down(VMALLOC_START - MEMCPY_REAL_SIZE, PAGE_SIZE);
416	boot_debug("memcpy real area:    0x%016lx-0x%016lx\n", __memcpy_real_area,
417		   __memcpy_real_area + MEMCPY_REAL_SIZE);
418	__abs_lowcore = round_down(__memcpy_real_area - ABS_LOWCORE_MAP_SIZE,
419				   sizeof(struct lowcore));
420	boot_debug("abs lowcore:         0x%016lx-0x%016lx\n", __abs_lowcore,
421		   __abs_lowcore + ABS_LOWCORE_MAP_SIZE);
422
423	/* split remaining virtual space between 1:1 mapping & vmemmap array */
424	pages = __abs_lowcore / (PAGE_SIZE + sizeof(struct page));
425	pages = SECTION_ALIGN_UP(pages);
426	/* keep vmemmap_start aligned to a top level region table entry */
427	vmemmap_start = round_down(__abs_lowcore - pages * sizeof(struct page), rte_size);
428	/* make sure identity map doesn't overlay with vmemmap */
429	ident_map_size = min(ident_map_size, vmemmap_start);
430	vmemmap_size = SECTION_ALIGN_UP(ident_map_size / PAGE_SIZE) * sizeof(struct page);
431	/* make sure vmemmap doesn't overlay with absolute lowcore area */
432	if (vmemmap_start + vmemmap_size > __abs_lowcore) {
433		vmemmap_size = SECTION_ALIGN_DOWN(ident_map_size / PAGE_SIZE) * sizeof(struct page);
434		ident_map_size = vmemmap_size / sizeof(struct page) * PAGE_SIZE;
435	}
436	vmemmap = (struct page *)vmemmap_start;
437	/* maximum address for which linear mapping could be created (DCSS, memory) */
438	BUILD_BUG_ON(MAX_DCSS_ADDR > (1UL << MAX_PHYSMEM_BITS));
439	max_mappable = max(ident_map_size, MAX_DCSS_ADDR);
440	max_mappable = min(max_mappable, vmemmap_start);
441#ifdef CONFIG_RANDOMIZE_IDENTITY_BASE
442	__identity_base = round_down(vmemmap_start - max_mappable, rte_size);
443#endif
444	boot_debug("identity map:        0x%016lx-0x%016lx\n", __identity_base,
445		   __identity_base + ident_map_size);
446
447	return asce_limit;
448}
449
450/*
451 * This function clears the BSS section of the decompressed Linux kernel and NOT the decompressor's.
452 */
453static void clear_bss_section(unsigned long kernel_start)
454{
455	memset((void *)kernel_start + vmlinux.image_size, 0, vmlinux.bss_size);
456}
457
458/*
459 * Set vmalloc area size to an 8th of (potential) physical memory
460 * size, unless size has been set by kernel command line parameter.
461 */
462static void setup_vmalloc_size(void)
463{
464	unsigned long size;
465
466	if (vmalloc_size_set)
467		return;
468	size = round_up(ident_map_size / 8, _SEGMENT_SIZE);
469	vmalloc_size = max(size, vmalloc_size);
470}
471
472static void kaslr_adjust_vmlinux_info(long offset)
473{
474	vmlinux.bootdata_off += offset;
475	vmlinux.bootdata_preserved_off += offset;
476	vmlinux.got_start += offset;
477	vmlinux.got_end += offset;
478	vmlinux.init_mm_off += offset;
479	vmlinux.swapper_pg_dir_off += offset;
480	vmlinux.invalid_pg_dir_off += offset;
481	vmlinux.alt_instructions += offset;
482	vmlinux.alt_instructions_end += offset;
483#ifdef CONFIG_STACKPROTECTOR
484	vmlinux.stack_prot_start += offset;
485	vmlinux.stack_prot_end += offset;
486#endif
487#ifdef CONFIG_KASAN
488	vmlinux.kasan_early_shadow_page_off += offset;
489	vmlinux.kasan_early_shadow_pte_off += offset;
490	vmlinux.kasan_early_shadow_pmd_off += offset;
491	vmlinux.kasan_early_shadow_pud_off += offset;
492	vmlinux.kasan_early_shadow_p4d_off += offset;
493#endif
494}
495
496void startup_kernel(void)
497{
498	unsigned long vmlinux_size = vmlinux.image_size + vmlinux.bss_size;
499	unsigned long nokaslr_text_lma, text_lma = 0, amode31_lma = 0;
500	unsigned long kernel_size = TEXT_OFFSET + vmlinux_size;
501	unsigned long kaslr_large_page_offset;
502	unsigned long max_physmem_end;
503	unsigned long asce_limit;
504	unsigned long safe_addr;
505	psw_t psw;
506
507	setup_lpp();
508	store_ipl_parmblock();
509	uv_query_info();
510	setup_boot_command_line();
511	parse_boot_command_line();
512
513	/*
514	 * Non-randomized kernel physical start address must be _SEGMENT_SIZE
515	 * aligned (see blow).
516	 */
517	nokaslr_text_lma = ALIGN(mem_safe_offset(), _SEGMENT_SIZE);
518	safe_addr = PAGE_ALIGN(nokaslr_text_lma + vmlinux_size);
519
520	/*
521	 * Reserve decompressor memory together with decompression heap,
522	 * buffer and memory which might be occupied by uncompressed kernel
523	 * (if KASLR is off or failed).
524	 */
525	physmem_reserve(RR_DECOMPRESSOR, 0, safe_addr);
526	if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && parmarea.initrd_size)
527		physmem_reserve(RR_INITRD, parmarea.initrd_start, parmarea.initrd_size);
528	oldmem_data.start = parmarea.oldmem_base;
529	oldmem_data.size = parmarea.oldmem_size;
530
531	read_ipl_report();
532	sclp_early_read_info();
533	sclp_early_detect_machine_features();
534	detect_facilities();
535	detect_diag9c();
536	detect_machine_type();
537	/* detect_diag288() needs machine type */
538	detect_diag288();
539	cmma_init();
540	sanitize_prot_virt_host();
541	max_physmem_end = detect_max_physmem_end();
542	setup_ident_map_size(max_physmem_end);
543	setup_vmalloc_size();
544	asce_limit = setup_kernel_memory_layout(kernel_size);
545	/* got final ident_map_size, physmem allocations could be performed now */
546	physmem_set_usable_limit(ident_map_size);
547	detect_physmem_online_ranges(max_physmem_end);
548	save_ipl_cert_comp_list();
549	rescue_initrd(safe_addr, ident_map_size);
550
551	/*
552	 * __kaslr_offset_phys must be _SEGMENT_SIZE aligned, so the lower
553	 * 20 bits (the offset within a large page) are zero. Copy the last
554	 * 20 bits of __kaslr_offset, which is THREAD_SIZE aligned, to
555	 * __kaslr_offset_phys.
556	 *
557	 * With this the last 20 bits of __kaslr_offset_phys and __kaslr_offset
558	 * are identical, which is required to allow for large mappings of the
559	 * kernel image.
560	 */
561	kaslr_large_page_offset = __kaslr_offset & ~_SEGMENT_MASK;
562	if (kaslr_enabled()) {
563		unsigned long size = vmlinux_size + kaslr_large_page_offset;
564
565		text_lma = randomize_within_range(size, _SEGMENT_SIZE, TEXT_OFFSET, ident_map_size);
566	}
567	if (!text_lma)
568		text_lma = nokaslr_text_lma;
569	text_lma |= kaslr_large_page_offset;
570
571	/*
572	 * [__kaslr_offset_phys..__kaslr_offset_phys + TEXT_OFFSET] region is
573	 * never accessed via the kernel image mapping as per the linker script:
574	 *
575	 *	. = TEXT_OFFSET;
576	 *
577	 * Therefore, this region could be used for something else and does
578	 * not need to be reserved. See how it is skipped in setup_vmem().
579	 */
580	__kaslr_offset_phys = text_lma - TEXT_OFFSET;
581	kaslr_adjust_vmlinux_info(__kaslr_offset_phys);
582	physmem_reserve(RR_VMLINUX, text_lma, vmlinux_size);
583	deploy_kernel((void *)text_lma);
584
585	/* vmlinux decompression is done, shrink reserved low memory */
586	physmem_reserve(RR_DECOMPRESSOR, 0, (unsigned long)_decompressor_end);
587
588	/*
589	 * In case KASLR is enabled the randomized location of .amode31
590	 * section might overlap with .vmlinux.relocs section. To avoid that
591	 * the below randomize_within_range() could have been called with
592	 * __vmlinux_relocs_64_end as the lower range address. However,
593	 * .amode31 section is written to by the decompressed kernel - at
594	 * that time the contents of .vmlinux.relocs is not needed anymore.
595	 * Conversely, .vmlinux.relocs is read only by the decompressor, even
596	 * before the kernel started. Therefore, in case the two sections
597	 * overlap there is no risk of corrupting any data.
598	 */
599	if (kaslr_enabled()) {
600		unsigned long amode31_min;
601
602		amode31_min = (unsigned long)_decompressor_end;
603		amode31_lma = randomize_within_range(vmlinux.amode31_size, PAGE_SIZE, amode31_min, SZ_2G);
604	}
605	if (!amode31_lma)
606		amode31_lma = text_lma - vmlinux.amode31_size;
607	physmem_reserve(RR_AMODE31, amode31_lma, vmlinux.amode31_size);
608
609	/*
610	 * The order of the following operations is important:
611	 *
612	 * - kaslr_adjust_relocs() must follow clear_bss_section() to establish
613	 *   static memory references to data in .bss to be used by setup_vmem()
614	 *   (i.e init_mm.pgd)
615	 *
616	 * - setup_vmem() must follow kaslr_adjust_relocs() to be able using
617	 *   static memory references to data in .bss (i.e init_mm.pgd)
618	 *
619	 * - copy_bootdata() must follow setup_vmem() to propagate changes
620	 *   to bootdata made by setup_vmem()
621	 */
622	clear_bss_section(text_lma);
623	kaslr_adjust_relocs(text_lma, text_lma + vmlinux.image_size,
624			    __kaslr_offset, __kaslr_offset_phys);
625	kaslr_adjust_got(__kaslr_offset);
626	setup_vmem(__kaslr_offset, __kaslr_offset + kernel_size, asce_limit);
627	dump_physmem_reserved();
628	copy_bootdata();
629	__apply_alternatives((struct alt_instr *)_vmlinux_info.alt_instructions,
630			     (struct alt_instr *)_vmlinux_info.alt_instructions_end,
631			     ALT_CTX_EARLY);
632	stack_protector_apply_early(text_lma);
633
634	/*
635	 * Save KASLR offset for early dumps, before vmcore_info is set.
636	 * Mark as uneven to distinguish from real vmcore_info pointer.
637	 */
638	get_lowcore()->vmcore_info = __kaslr_offset_phys ? __kaslr_offset_phys | 0x1UL : 0;
639
640	/*
641	 * Jump to the decompressed kernel entry point and switch DAT mode on.
642	 */
643	psw.addr = __kaslr_offset + vmlinux.entry;
644	psw.mask = PSW_KERNEL_BITS;
645	boot_debug("Starting kernel at:  0x%016lx\n", psw.addr);
646	jump_to_kernel(&psw);
647}