xref: /linux/arch/s390/boot/startup.c (revision c2a96b7f187fb6a455836d4a6e113947ff11de97)
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/string.h>
3 #include <linux/elf.h>
4 #include <asm/page-states.h>
5 #include <asm/boot_data.h>
6 #include <asm/extmem.h>
7 #include <asm/sections.h>
8 #include <asm/maccess.h>
9 #include <asm/cpu_mf.h>
10 #include <asm/setup.h>
11 #include <asm/kasan.h>
12 #include <asm/kexec.h>
13 #include <asm/sclp.h>
14 #include <asm/diag.h>
15 #include <asm/uv.h>
16 #include <asm/abs_lowcore.h>
17 #include <asm/physmem_info.h>
18 #include "decompressor.h"
19 #include "boot.h"
20 #include "uv.h"
21 
22 struct vm_layout __bootdata_preserved(vm_layout);
23 unsigned long __bootdata_preserved(__abs_lowcore);
24 unsigned long __bootdata_preserved(__memcpy_real_area);
25 pte_t *__bootdata_preserved(memcpy_real_ptep);
26 unsigned long __bootdata_preserved(VMALLOC_START);
27 unsigned long __bootdata_preserved(VMALLOC_END);
28 struct page *__bootdata_preserved(vmemmap);
29 unsigned long __bootdata_preserved(vmemmap_size);
30 unsigned long __bootdata_preserved(MODULES_VADDR);
31 unsigned long __bootdata_preserved(MODULES_END);
32 unsigned long __bootdata_preserved(max_mappable);
33 
34 u64 __bootdata_preserved(stfle_fac_list[16]);
35 struct oldmem_data __bootdata_preserved(oldmem_data);
36 
37 struct machine_info machine;
38 
39 void error(char *x)
40 {
41 	sclp_early_printk("\n\n");
42 	sclp_early_printk(x);
43 	sclp_early_printk("\n\n -- System halted");
44 
45 	disabled_wait();
46 }
47 
48 static void detect_facilities(void)
49 {
50 	if (test_facility(8)) {
51 		machine.has_edat1 = 1;
52 		local_ctl_set_bit(0, CR0_EDAT_BIT);
53 	}
54 	if (test_facility(78))
55 		machine.has_edat2 = 1;
56 	if (test_facility(130))
57 		machine.has_nx = 1;
58 }
59 
60 static int cmma_test_essa(void)
61 {
62 	unsigned long reg1, reg2, tmp = 0;
63 	int rc = 1;
64 	psw_t old;
65 
66 	/* Test ESSA_GET_STATE */
67 	asm volatile(
68 		"	mvc	0(16,%[psw_old]),0(%[psw_pgm])\n"
69 		"	epsw	%[reg1],%[reg2]\n"
70 		"	st	%[reg1],0(%[psw_pgm])\n"
71 		"	st	%[reg2],4(%[psw_pgm])\n"
72 		"	larl	%[reg1],1f\n"
73 		"	stg	%[reg1],8(%[psw_pgm])\n"
74 		"	.insn	rrf,0xb9ab0000,%[tmp],%[tmp],%[cmd],0\n"
75 		"	la	%[rc],0\n"
76 		"1:	mvc	0(16,%[psw_pgm]),0(%[psw_old])\n"
77 		: [reg1] "=&d" (reg1),
78 		  [reg2] "=&a" (reg2),
79 		  [rc] "+&d" (rc),
80 		  [tmp] "=&d" (tmp),
81 		  "+Q" (get_lowcore()->program_new_psw),
82 		  "=Q" (old)
83 		: [psw_old] "a" (&old),
84 		  [psw_pgm] "a" (&get_lowcore()->program_new_psw),
85 		  [cmd] "i" (ESSA_GET_STATE)
86 		: "cc", "memory");
87 	return rc;
88 }
89 
90 static void cmma_init(void)
91 {
92 	if (!cmma_flag)
93 		return;
94 	if (cmma_test_essa()) {
95 		cmma_flag = 0;
96 		return;
97 	}
98 	if (test_facility(147))
99 		cmma_flag = 2;
100 }
101 
102 static void setup_lpp(void)
103 {
104 	get_lowcore()->current_pid = 0;
105 	get_lowcore()->lpp = LPP_MAGIC;
106 	if (test_facility(40))
107 		lpp(&get_lowcore()->lpp);
108 }
109 
110 #ifdef CONFIG_KERNEL_UNCOMPRESSED
111 static unsigned long mem_safe_offset(void)
112 {
113 	return (unsigned long)_compressed_start;
114 }
115 
116 static void deploy_kernel(void *output)
117 {
118 	void *uncompressed_start = (void *)_compressed_start;
119 
120 	if (output == uncompressed_start)
121 		return;
122 	memmove(output, uncompressed_start, vmlinux.image_size);
123 	memset(uncompressed_start, 0, vmlinux.image_size);
124 }
125 #endif
126 
127 static void rescue_initrd(unsigned long min, unsigned long max)
128 {
129 	unsigned long old_addr, addr, size;
130 
131 	if (!IS_ENABLED(CONFIG_BLK_DEV_INITRD))
132 		return;
133 	if (!get_physmem_reserved(RR_INITRD, &addr, &size))
134 		return;
135 	if (addr >= min && addr + size <= max)
136 		return;
137 	old_addr = addr;
138 	physmem_free(RR_INITRD);
139 	addr = physmem_alloc_top_down(RR_INITRD, size, 0);
140 	memmove((void *)addr, (void *)old_addr, size);
141 }
142 
143 static void copy_bootdata(void)
144 {
145 	if (__boot_data_end - __boot_data_start != vmlinux.bootdata_size)
146 		error(".boot.data section size mismatch");
147 	memcpy((void *)vmlinux.bootdata_off, __boot_data_start, vmlinux.bootdata_size);
148 	if (__boot_data_preserved_end - __boot_data_preserved_start != vmlinux.bootdata_preserved_size)
149 		error(".boot.preserved.data section size mismatch");
150 	memcpy((void *)vmlinux.bootdata_preserved_off, __boot_data_preserved_start, vmlinux.bootdata_preserved_size);
151 }
152 
153 static void kaslr_adjust_relocs(unsigned long min_addr, unsigned long max_addr,
154 				unsigned long offset, unsigned long phys_offset)
155 {
156 	int *reloc;
157 	long loc;
158 
159 	/* Adjust R_390_64 relocations */
160 	for (reloc = (int *)__vmlinux_relocs_64_start; reloc < (int *)__vmlinux_relocs_64_end; reloc++) {
161 		loc = (long)*reloc + phys_offset;
162 		if (loc < min_addr || loc > max_addr)
163 			error("64-bit relocation outside of kernel!\n");
164 		*(u64 *)loc += offset - __START_KERNEL;
165 	}
166 }
167 
168 static void kaslr_adjust_got(unsigned long offset)
169 {
170 	u64 *entry;
171 
172 	/*
173 	 * Adjust GOT entries, except for ones for undefined weak symbols
174 	 * that resolved to zero. This also skips the first three reserved
175 	 * entries on s390x that are zero.
176 	 */
177 	for (entry = (u64 *)vmlinux.got_start; entry < (u64 *)vmlinux.got_end; entry++) {
178 		if (*entry)
179 			*entry += offset - __START_KERNEL;
180 	}
181 }
182 
183 /*
184  * Merge information from several sources into a single ident_map_size value.
185  * "ident_map_size" represents the upper limit of physical memory we may ever
186  * reach. It might not be all online memory, but also include standby (offline)
187  * memory. "ident_map_size" could be lower then actual standby or even online
188  * memory present, due to limiting factors. We should never go above this limit.
189  * It is the size of our identity mapping.
190  *
191  * Consider the following factors:
192  * 1. max_physmem_end - end of physical memory online or standby.
193  *    Always >= end of the last online memory range (get_physmem_online_end()).
194  * 2. CONFIG_MAX_PHYSMEM_BITS - the maximum size of physical memory the
195  *    kernel is able to support.
196  * 3. "mem=" kernel command line option which limits physical memory usage.
197  * 4. OLDMEM_BASE which is a kdump memory limit when the kernel is executed as
198  *    crash kernel.
199  * 5. "hsa" size which is a memory limit when the kernel is executed during
200  *    zfcp/nvme dump.
201  */
202 static void setup_ident_map_size(unsigned long max_physmem_end)
203 {
204 	unsigned long hsa_size;
205 
206 	ident_map_size = max_physmem_end;
207 	if (memory_limit)
208 		ident_map_size = min(ident_map_size, memory_limit);
209 	ident_map_size = min(ident_map_size, 1UL << MAX_PHYSMEM_BITS);
210 
211 #ifdef CONFIG_CRASH_DUMP
212 	if (oldmem_data.start) {
213 		__kaslr_enabled = 0;
214 		ident_map_size = min(ident_map_size, oldmem_data.size);
215 	} else if (ipl_block_valid && is_ipl_block_dump()) {
216 		__kaslr_enabled = 0;
217 		if (!sclp_early_get_hsa_size(&hsa_size) && hsa_size)
218 			ident_map_size = min(ident_map_size, hsa_size);
219 	}
220 #endif
221 }
222 
223 #define FIXMAP_SIZE	round_up(MEMCPY_REAL_SIZE + ABS_LOWCORE_MAP_SIZE, sizeof(struct lowcore))
224 
225 static unsigned long get_vmem_size(unsigned long identity_size,
226 				   unsigned long vmemmap_size,
227 				   unsigned long vmalloc_size,
228 				   unsigned long rte_size)
229 {
230 	unsigned long max_mappable, vsize;
231 
232 	max_mappable = max(identity_size, MAX_DCSS_ADDR);
233 	vsize = round_up(SZ_2G + max_mappable, rte_size) +
234 		round_up(vmemmap_size, rte_size) +
235 		FIXMAP_SIZE + MODULES_LEN + KASLR_LEN;
236 	return size_add(vsize, vmalloc_size);
237 }
238 
239 static unsigned long setup_kernel_memory_layout(unsigned long kernel_size)
240 {
241 	unsigned long vmemmap_start;
242 	unsigned long kernel_start;
243 	unsigned long asce_limit;
244 	unsigned long rte_size;
245 	unsigned long pages;
246 	unsigned long vsize;
247 	unsigned long vmax;
248 
249 	pages = ident_map_size / PAGE_SIZE;
250 	/* vmemmap contains a multiple of PAGES_PER_SECTION struct pages */
251 	vmemmap_size = SECTION_ALIGN_UP(pages) * sizeof(struct page);
252 
253 	/* choose kernel address space layout: 4 or 3 levels. */
254 	BUILD_BUG_ON(!IS_ALIGNED(__START_KERNEL, THREAD_SIZE));
255 	BUILD_BUG_ON(!IS_ALIGNED(__NO_KASLR_START_KERNEL, THREAD_SIZE));
256 	BUILD_BUG_ON(__NO_KASLR_END_KERNEL > _REGION1_SIZE);
257 	vsize = get_vmem_size(ident_map_size, vmemmap_size, vmalloc_size, _REGION3_SIZE);
258 	if (IS_ENABLED(CONFIG_KASAN) || __NO_KASLR_END_KERNEL > _REGION2_SIZE ||
259 	    (vsize > _REGION2_SIZE && kaslr_enabled())) {
260 		asce_limit = _REGION1_SIZE;
261 		if (__NO_KASLR_END_KERNEL > _REGION2_SIZE) {
262 			rte_size = _REGION2_SIZE;
263 			vsize = get_vmem_size(ident_map_size, vmemmap_size, vmalloc_size, _REGION2_SIZE);
264 		} else {
265 			rte_size = _REGION3_SIZE;
266 		}
267 	} else {
268 		asce_limit = _REGION2_SIZE;
269 		rte_size = _REGION3_SIZE;
270 	}
271 
272 	/*
273 	 * Forcing modules and vmalloc area under the ultravisor
274 	 * secure storage limit, so that any vmalloc allocation
275 	 * we do could be used to back secure guest storage.
276 	 *
277 	 * Assume the secure storage limit always exceeds _REGION2_SIZE,
278 	 * otherwise asce_limit and rte_size would have been adjusted.
279 	 */
280 	vmax = adjust_to_uv_max(asce_limit);
281 #ifdef CONFIG_KASAN
282 	BUILD_BUG_ON(__NO_KASLR_END_KERNEL > KASAN_SHADOW_START);
283 	/* force vmalloc and modules below kasan shadow */
284 	vmax = min(vmax, KASAN_SHADOW_START);
285 #endif
286 	vsize = min(vsize, vmax);
287 	if (kaslr_enabled()) {
288 		unsigned long kernel_end, kaslr_len, slots, pos;
289 
290 		kaslr_len = max(KASLR_LEN, vmax - vsize);
291 		slots = DIV_ROUND_UP(kaslr_len - kernel_size, THREAD_SIZE);
292 		if (get_random(slots, &pos))
293 			pos = 0;
294 		kernel_end = vmax - pos * THREAD_SIZE;
295 		kernel_start = round_down(kernel_end - kernel_size, THREAD_SIZE);
296 	} else if (vmax < __NO_KASLR_END_KERNEL || vsize > __NO_KASLR_END_KERNEL) {
297 		kernel_start = round_down(vmax - kernel_size, THREAD_SIZE);
298 		decompressor_printk("The kernel base address is forced to %lx\n", kernel_start);
299 	} else {
300 		kernel_start = __NO_KASLR_START_KERNEL;
301 	}
302 	__kaslr_offset = kernel_start;
303 
304 	MODULES_END = round_down(kernel_start, _SEGMENT_SIZE);
305 	MODULES_VADDR = MODULES_END - MODULES_LEN;
306 	VMALLOC_END = MODULES_VADDR;
307 	if (IS_ENABLED(CONFIG_KMSAN))
308 		VMALLOC_END -= MODULES_LEN * 2;
309 
310 	/* allow vmalloc area to occupy up to about 1/2 of the rest virtual space left */
311 	vsize = (VMALLOC_END - FIXMAP_SIZE) / 2;
312 	vsize = round_down(vsize, _SEGMENT_SIZE);
313 	vmalloc_size = min(vmalloc_size, vsize);
314 	if (IS_ENABLED(CONFIG_KMSAN)) {
315 		/* take 2/3 of vmalloc area for KMSAN shadow and origins */
316 		vmalloc_size = round_down(vmalloc_size / 3, _SEGMENT_SIZE);
317 		VMALLOC_END -= vmalloc_size * 2;
318 	}
319 	VMALLOC_START = VMALLOC_END - vmalloc_size;
320 
321 	__memcpy_real_area = round_down(VMALLOC_START - MEMCPY_REAL_SIZE, PAGE_SIZE);
322 	__abs_lowcore = round_down(__memcpy_real_area - ABS_LOWCORE_MAP_SIZE,
323 				   sizeof(struct lowcore));
324 
325 	/* split remaining virtual space between 1:1 mapping & vmemmap array */
326 	pages = __abs_lowcore / (PAGE_SIZE + sizeof(struct page));
327 	pages = SECTION_ALIGN_UP(pages);
328 	/* keep vmemmap_start aligned to a top level region table entry */
329 	vmemmap_start = round_down(__abs_lowcore - pages * sizeof(struct page), rte_size);
330 	/* make sure identity map doesn't overlay with vmemmap */
331 	ident_map_size = min(ident_map_size, vmemmap_start);
332 	vmemmap_size = SECTION_ALIGN_UP(ident_map_size / PAGE_SIZE) * sizeof(struct page);
333 	/* make sure vmemmap doesn't overlay with absolute lowcore area */
334 	if (vmemmap_start + vmemmap_size > __abs_lowcore) {
335 		vmemmap_size = SECTION_ALIGN_DOWN(ident_map_size / PAGE_SIZE) * sizeof(struct page);
336 		ident_map_size = vmemmap_size / sizeof(struct page) * PAGE_SIZE;
337 	}
338 	vmemmap = (struct page *)vmemmap_start;
339 	/* maximum address for which linear mapping could be created (DCSS, memory) */
340 	BUILD_BUG_ON(MAX_DCSS_ADDR > (1UL << MAX_PHYSMEM_BITS));
341 	max_mappable = max(ident_map_size, MAX_DCSS_ADDR);
342 	max_mappable = min(max_mappable, vmemmap_start);
343 	__identity_base = round_down(vmemmap_start - max_mappable, rte_size);
344 
345 	return asce_limit;
346 }
347 
348 /*
349  * This function clears the BSS section of the decompressed Linux kernel and NOT the decompressor's.
350  */
351 static void clear_bss_section(unsigned long kernel_start)
352 {
353 	memset((void *)kernel_start + vmlinux.image_size, 0, vmlinux.bss_size);
354 }
355 
356 /*
357  * Set vmalloc area size to an 8th of (potential) physical memory
358  * size, unless size has been set by kernel command line parameter.
359  */
360 static void setup_vmalloc_size(void)
361 {
362 	unsigned long size;
363 
364 	if (vmalloc_size_set)
365 		return;
366 	size = round_up(ident_map_size / 8, _SEGMENT_SIZE);
367 	vmalloc_size = max(size, vmalloc_size);
368 }
369 
370 static void kaslr_adjust_vmlinux_info(long offset)
371 {
372 	vmlinux.bootdata_off += offset;
373 	vmlinux.bootdata_preserved_off += offset;
374 	vmlinux.got_start += offset;
375 	vmlinux.got_end += offset;
376 	vmlinux.init_mm_off += offset;
377 	vmlinux.swapper_pg_dir_off += offset;
378 	vmlinux.invalid_pg_dir_off += offset;
379 #ifdef CONFIG_KASAN
380 	vmlinux.kasan_early_shadow_page_off += offset;
381 	vmlinux.kasan_early_shadow_pte_off += offset;
382 	vmlinux.kasan_early_shadow_pmd_off += offset;
383 	vmlinux.kasan_early_shadow_pud_off += offset;
384 	vmlinux.kasan_early_shadow_p4d_off += offset;
385 #endif
386 }
387 
388 static void fixup_vmlinux_info(void)
389 {
390 	vmlinux.entry -= __START_KERNEL;
391 	kaslr_adjust_vmlinux_info(-__START_KERNEL);
392 }
393 
394 void startup_kernel(void)
395 {
396 	unsigned long kernel_size = vmlinux.image_size + vmlinux.bss_size;
397 	unsigned long nokaslr_offset_phys, kaslr_large_page_offset;
398 	unsigned long amode31_lma = 0;
399 	unsigned long max_physmem_end;
400 	unsigned long asce_limit;
401 	unsigned long safe_addr;
402 	psw_t psw;
403 
404 	fixup_vmlinux_info();
405 	setup_lpp();
406 
407 	/*
408 	 * Non-randomized kernel physical start address must be _SEGMENT_SIZE
409 	 * aligned (see blow).
410 	 */
411 	nokaslr_offset_phys = ALIGN(mem_safe_offset(), _SEGMENT_SIZE);
412 	safe_addr = PAGE_ALIGN(nokaslr_offset_phys + kernel_size);
413 
414 	/*
415 	 * Reserve decompressor memory together with decompression heap,
416 	 * buffer and memory which might be occupied by uncompressed kernel
417 	 * (if KASLR is off or failed).
418 	 */
419 	physmem_reserve(RR_DECOMPRESSOR, 0, safe_addr);
420 	if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && parmarea.initrd_size)
421 		physmem_reserve(RR_INITRD, parmarea.initrd_start, parmarea.initrd_size);
422 	oldmem_data.start = parmarea.oldmem_base;
423 	oldmem_data.size = parmarea.oldmem_size;
424 
425 	store_ipl_parmblock();
426 	read_ipl_report();
427 	uv_query_info();
428 	sclp_early_read_info();
429 	setup_boot_command_line();
430 	parse_boot_command_line();
431 	detect_facilities();
432 	cmma_init();
433 	sanitize_prot_virt_host();
434 	max_physmem_end = detect_max_physmem_end();
435 	setup_ident_map_size(max_physmem_end);
436 	setup_vmalloc_size();
437 	asce_limit = setup_kernel_memory_layout(kernel_size);
438 	/* got final ident_map_size, physmem allocations could be performed now */
439 	physmem_set_usable_limit(ident_map_size);
440 	detect_physmem_online_ranges(max_physmem_end);
441 	save_ipl_cert_comp_list();
442 	rescue_initrd(safe_addr, ident_map_size);
443 
444 	/*
445 	 * __kaslr_offset_phys must be _SEGMENT_SIZE aligned, so the lower
446 	 * 20 bits (the offset within a large page) are zero. Copy the last
447 	 * 20 bits of __kaslr_offset, which is THREAD_SIZE aligned, to
448 	 * __kaslr_offset_phys.
449 	 *
450 	 * With this the last 20 bits of __kaslr_offset_phys and __kaslr_offset
451 	 * are identical, which is required to allow for large mappings of the
452 	 * kernel image.
453 	 */
454 	kaslr_large_page_offset = __kaslr_offset & ~_SEGMENT_MASK;
455 	if (kaslr_enabled()) {
456 		unsigned long end = ident_map_size - kaslr_large_page_offset;
457 
458 		__kaslr_offset_phys = randomize_within_range(kernel_size, _SEGMENT_SIZE, 0, end);
459 	}
460 	if (!__kaslr_offset_phys)
461 		__kaslr_offset_phys = nokaslr_offset_phys;
462 	__kaslr_offset_phys |= kaslr_large_page_offset;
463 	kaslr_adjust_vmlinux_info(__kaslr_offset_phys);
464 	physmem_reserve(RR_VMLINUX, __kaslr_offset_phys, kernel_size);
465 	deploy_kernel((void *)__kaslr_offset_phys);
466 
467 	/* vmlinux decompression is done, shrink reserved low memory */
468 	physmem_reserve(RR_DECOMPRESSOR, 0, (unsigned long)_decompressor_end);
469 
470 	/*
471 	 * In case KASLR is enabled the randomized location of .amode31
472 	 * section might overlap with .vmlinux.relocs section. To avoid that
473 	 * the below randomize_within_range() could have been called with
474 	 * __vmlinux_relocs_64_end as the lower range address. However,
475 	 * .amode31 section is written to by the decompressed kernel - at
476 	 * that time the contents of .vmlinux.relocs is not needed anymore.
477 	 * Conversly, .vmlinux.relocs is read only by the decompressor, even
478 	 * before the kernel started. Therefore, in case the two sections
479 	 * overlap there is no risk of corrupting any data.
480 	 */
481 	if (kaslr_enabled())
482 		amode31_lma = randomize_within_range(vmlinux.amode31_size, PAGE_SIZE, 0, SZ_2G);
483 	if (!amode31_lma)
484 		amode31_lma = __kaslr_offset_phys - vmlinux.amode31_size;
485 	physmem_reserve(RR_AMODE31, amode31_lma, vmlinux.amode31_size);
486 
487 	/*
488 	 * The order of the following operations is important:
489 	 *
490 	 * - kaslr_adjust_relocs() must follow clear_bss_section() to establish
491 	 *   static memory references to data in .bss to be used by setup_vmem()
492 	 *   (i.e init_mm.pgd)
493 	 *
494 	 * - setup_vmem() must follow kaslr_adjust_relocs() to be able using
495 	 *   static memory references to data in .bss (i.e init_mm.pgd)
496 	 *
497 	 * - copy_bootdata() must follow setup_vmem() to propagate changes
498 	 *   to bootdata made by setup_vmem()
499 	 */
500 	clear_bss_section(__kaslr_offset_phys);
501 	kaslr_adjust_relocs(__kaslr_offset_phys, __kaslr_offset_phys + vmlinux.image_size,
502 			    __kaslr_offset, __kaslr_offset_phys);
503 	kaslr_adjust_got(__kaslr_offset);
504 	setup_vmem(__kaslr_offset, __kaslr_offset + kernel_size, asce_limit);
505 	copy_bootdata();
506 
507 	/*
508 	 * Save KASLR offset for early dumps, before vmcore_info is set.
509 	 * Mark as uneven to distinguish from real vmcore_info pointer.
510 	 */
511 	get_lowcore()->vmcore_info = __kaslr_offset_phys ? __kaslr_offset_phys | 0x1UL : 0;
512 
513 	/*
514 	 * Jump to the decompressed kernel entry point and switch DAT mode on.
515 	 */
516 	psw.addr = __kaslr_offset + vmlinux.entry;
517 	psw.mask = PSW_KERNEL_BITS;
518 	__load_psw(psw);
519 }
520