xref: /linux/arch/s390/boot/startup.c (revision 8e07e0e3964ca4e23ce7b68e2096fe660a888942)
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/sections.h>
7 #include <asm/maccess.h>
8 #include <asm/cpu_mf.h>
9 #include <asm/setup.h>
10 #include <asm/kasan.h>
11 #include <asm/kexec.h>
12 #include <asm/sclp.h>
13 #include <asm/diag.h>
14 #include <asm/uv.h>
15 #include <asm/abs_lowcore.h>
16 #include <asm/physmem_info.h>
17 #include "decompressor.h"
18 #include "boot.h"
19 #include "uv.h"
20 
21 unsigned long __bootdata_preserved(__kaslr_offset);
22 unsigned long __bootdata_preserved(__abs_lowcore);
23 unsigned long __bootdata_preserved(__memcpy_real_area);
24 pte_t *__bootdata_preserved(memcpy_real_ptep);
25 unsigned long __bootdata_preserved(VMALLOC_START);
26 unsigned long __bootdata_preserved(VMALLOC_END);
27 struct page *__bootdata_preserved(vmemmap);
28 unsigned long __bootdata_preserved(vmemmap_size);
29 unsigned long __bootdata_preserved(MODULES_VADDR);
30 unsigned long __bootdata_preserved(MODULES_END);
31 unsigned long __bootdata_preserved(max_mappable);
32 unsigned long __bootdata(ident_map_size);
33 
34 u64 __bootdata_preserved(stfle_fac_list[16]);
35 u64 __bootdata_preserved(alt_stfle_fac_list[16]);
36 struct oldmem_data __bootdata_preserved(oldmem_data);
37 
38 struct machine_info machine;
39 
40 void error(char *x)
41 {
42 	sclp_early_printk("\n\n");
43 	sclp_early_printk(x);
44 	sclp_early_printk("\n\n -- System halted");
45 
46 	disabled_wait();
47 }
48 
49 static void detect_facilities(void)
50 {
51 	if (test_facility(8)) {
52 		machine.has_edat1 = 1;
53 		local_ctl_set_bit(0, CR0_EDAT_BIT);
54 	}
55 	if (test_facility(78))
56 		machine.has_edat2 = 1;
57 	if (test_facility(130))
58 		machine.has_nx = 1;
59 }
60 
61 static int cmma_test_essa(void)
62 {
63 	unsigned long reg1, reg2, tmp = 0;
64 	int rc = 1;
65 	psw_t old;
66 
67 	/* Test ESSA_GET_STATE */
68 	asm volatile(
69 		"	mvc	0(16,%[psw_old]),0(%[psw_pgm])\n"
70 		"	epsw	%[reg1],%[reg2]\n"
71 		"	st	%[reg1],0(%[psw_pgm])\n"
72 		"	st	%[reg2],4(%[psw_pgm])\n"
73 		"	larl	%[reg1],1f\n"
74 		"	stg	%[reg1],8(%[psw_pgm])\n"
75 		"	.insn	rrf,0xb9ab0000,%[tmp],%[tmp],%[cmd],0\n"
76 		"	la	%[rc],0\n"
77 		"1:	mvc	0(16,%[psw_pgm]),0(%[psw_old])\n"
78 		: [reg1] "=&d" (reg1),
79 		  [reg2] "=&a" (reg2),
80 		  [rc] "+&d" (rc),
81 		  [tmp] "=&d" (tmp),
82 		  "+Q" (S390_lowcore.program_new_psw),
83 		  "=Q" (old)
84 		: [psw_old] "a" (&old),
85 		  [psw_pgm] "a" (&S390_lowcore.program_new_psw),
86 		  [cmd] "i" (ESSA_GET_STATE)
87 		: "cc", "memory");
88 	return rc;
89 }
90 
91 static void cmma_init(void)
92 {
93 	if (!cmma_flag)
94 		return;
95 	if (cmma_test_essa()) {
96 		cmma_flag = 0;
97 		return;
98 	}
99 	if (test_facility(147))
100 		cmma_flag = 2;
101 }
102 
103 static void setup_lpp(void)
104 {
105 	S390_lowcore.current_pid = 0;
106 	S390_lowcore.lpp = LPP_MAGIC;
107 	if (test_facility(40))
108 		lpp(&S390_lowcore.lpp);
109 }
110 
111 #ifdef CONFIG_KERNEL_UNCOMPRESSED
112 unsigned long mem_safe_offset(void)
113 {
114 	return vmlinux.default_lma + vmlinux.image_size + vmlinux.bss_size;
115 }
116 #endif
117 
118 static void rescue_initrd(unsigned long min, unsigned long max)
119 {
120 	unsigned long old_addr, addr, size;
121 
122 	if (!IS_ENABLED(CONFIG_BLK_DEV_INITRD))
123 		return;
124 	if (!get_physmem_reserved(RR_INITRD, &addr, &size))
125 		return;
126 	if (addr >= min && addr + size <= max)
127 		return;
128 	old_addr = addr;
129 	physmem_free(RR_INITRD);
130 	addr = physmem_alloc_top_down(RR_INITRD, size, 0);
131 	memmove((void *)addr, (void *)old_addr, size);
132 }
133 
134 static void copy_bootdata(void)
135 {
136 	if (__boot_data_end - __boot_data_start != vmlinux.bootdata_size)
137 		error(".boot.data section size mismatch");
138 	memcpy((void *)vmlinux.bootdata_off, __boot_data_start, vmlinux.bootdata_size);
139 	if (__boot_data_preserved_end - __boot_data_preserved_start != vmlinux.bootdata_preserved_size)
140 		error(".boot.preserved.data section size mismatch");
141 	memcpy((void *)vmlinux.bootdata_preserved_off, __boot_data_preserved_start, vmlinux.bootdata_preserved_size);
142 }
143 
144 static void handle_relocs(unsigned long offset)
145 {
146 	Elf64_Rela *rela_start, *rela_end, *rela;
147 	int r_type, r_sym, rc;
148 	Elf64_Addr loc, val;
149 	Elf64_Sym *dynsym;
150 
151 	rela_start = (Elf64_Rela *) vmlinux.rela_dyn_start;
152 	rela_end = (Elf64_Rela *) vmlinux.rela_dyn_end;
153 	dynsym = (Elf64_Sym *) vmlinux.dynsym_start;
154 	for (rela = rela_start; rela < rela_end; rela++) {
155 		loc = rela->r_offset + offset;
156 		val = rela->r_addend;
157 		r_sym = ELF64_R_SYM(rela->r_info);
158 		if (r_sym) {
159 			if (dynsym[r_sym].st_shndx != SHN_UNDEF)
160 				val += dynsym[r_sym].st_value + offset;
161 		} else {
162 			/*
163 			 * 0 == undefined symbol table index (STN_UNDEF),
164 			 * used for R_390_RELATIVE, only add KASLR offset
165 			 */
166 			val += offset;
167 		}
168 		r_type = ELF64_R_TYPE(rela->r_info);
169 		rc = arch_kexec_do_relocs(r_type, (void *) loc, val, 0);
170 		if (rc)
171 			error("Unknown relocation type");
172 	}
173 }
174 
175 /*
176  * Merge information from several sources into a single ident_map_size value.
177  * "ident_map_size" represents the upper limit of physical memory we may ever
178  * reach. It might not be all online memory, but also include standby (offline)
179  * memory. "ident_map_size" could be lower then actual standby or even online
180  * memory present, due to limiting factors. We should never go above this limit.
181  * It is the size of our identity mapping.
182  *
183  * Consider the following factors:
184  * 1. max_physmem_end - end of physical memory online or standby.
185  *    Always >= end of the last online memory range (get_physmem_online_end()).
186  * 2. CONFIG_MAX_PHYSMEM_BITS - the maximum size of physical memory the
187  *    kernel is able to support.
188  * 3. "mem=" kernel command line option which limits physical memory usage.
189  * 4. OLDMEM_BASE which is a kdump memory limit when the kernel is executed as
190  *    crash kernel.
191  * 5. "hsa" size which is a memory limit when the kernel is executed during
192  *    zfcp/nvme dump.
193  */
194 static void setup_ident_map_size(unsigned long max_physmem_end)
195 {
196 	unsigned long hsa_size;
197 
198 	ident_map_size = max_physmem_end;
199 	if (memory_limit)
200 		ident_map_size = min(ident_map_size, memory_limit);
201 	ident_map_size = min(ident_map_size, 1UL << MAX_PHYSMEM_BITS);
202 
203 #ifdef CONFIG_CRASH_DUMP
204 	if (oldmem_data.start) {
205 		__kaslr_enabled = 0;
206 		ident_map_size = min(ident_map_size, oldmem_data.size);
207 	} else if (ipl_block_valid && is_ipl_block_dump()) {
208 		__kaslr_enabled = 0;
209 		if (!sclp_early_get_hsa_size(&hsa_size) && hsa_size)
210 			ident_map_size = min(ident_map_size, hsa_size);
211 	}
212 #endif
213 }
214 
215 static unsigned long setup_kernel_memory_layout(void)
216 {
217 	unsigned long vmemmap_start;
218 	unsigned long asce_limit;
219 	unsigned long rte_size;
220 	unsigned long pages;
221 	unsigned long vsize;
222 	unsigned long vmax;
223 
224 	pages = ident_map_size / PAGE_SIZE;
225 	/* vmemmap contains a multiple of PAGES_PER_SECTION struct pages */
226 	vmemmap_size = SECTION_ALIGN_UP(pages) * sizeof(struct page);
227 
228 	/* choose kernel address space layout: 4 or 3 levels. */
229 	vsize = round_up(ident_map_size, _REGION3_SIZE) + vmemmap_size +
230 		MODULES_LEN + MEMCPY_REAL_SIZE + ABS_LOWCORE_MAP_SIZE;
231 	vsize = size_add(vsize, vmalloc_size);
232 	if (IS_ENABLED(CONFIG_KASAN) || (vsize > _REGION2_SIZE)) {
233 		asce_limit = _REGION1_SIZE;
234 		rte_size = _REGION2_SIZE;
235 	} else {
236 		asce_limit = _REGION2_SIZE;
237 		rte_size = _REGION3_SIZE;
238 	}
239 
240 	/*
241 	 * Forcing modules and vmalloc area under the ultravisor
242 	 * secure storage limit, so that any vmalloc allocation
243 	 * we do could be used to back secure guest storage.
244 	 */
245 	vmax = adjust_to_uv_max(asce_limit);
246 #ifdef CONFIG_KASAN
247 	/* force vmalloc and modules below kasan shadow */
248 	vmax = min(vmax, KASAN_SHADOW_START);
249 #endif
250 	__memcpy_real_area = round_down(vmax - MEMCPY_REAL_SIZE, PAGE_SIZE);
251 	__abs_lowcore = round_down(__memcpy_real_area - ABS_LOWCORE_MAP_SIZE,
252 				   sizeof(struct lowcore));
253 	MODULES_END = round_down(__abs_lowcore, _SEGMENT_SIZE);
254 	MODULES_VADDR = MODULES_END - MODULES_LEN;
255 	VMALLOC_END = MODULES_VADDR;
256 
257 	/* allow vmalloc area to occupy up to about 1/2 of the rest virtual space left */
258 	vmalloc_size = min(vmalloc_size, round_down(VMALLOC_END / 2, _REGION3_SIZE));
259 	VMALLOC_START = VMALLOC_END - vmalloc_size;
260 
261 	/* split remaining virtual space between 1:1 mapping & vmemmap array */
262 	pages = VMALLOC_START / (PAGE_SIZE + sizeof(struct page));
263 	pages = SECTION_ALIGN_UP(pages);
264 	/* keep vmemmap_start aligned to a top level region table entry */
265 	vmemmap_start = round_down(VMALLOC_START - pages * sizeof(struct page), rte_size);
266 	vmemmap_start = min(vmemmap_start, 1UL << MAX_PHYSMEM_BITS);
267 	/* maximum mappable address as seen by arch_get_mappable_range() */
268 	max_mappable = vmemmap_start;
269 	/* make sure identity map doesn't overlay with vmemmap */
270 	ident_map_size = min(ident_map_size, vmemmap_start);
271 	vmemmap_size = SECTION_ALIGN_UP(ident_map_size / PAGE_SIZE) * sizeof(struct page);
272 	/* make sure vmemmap doesn't overlay with vmalloc area */
273 	VMALLOC_START = max(vmemmap_start + vmemmap_size, VMALLOC_START);
274 	vmemmap = (struct page *)vmemmap_start;
275 
276 	return asce_limit;
277 }
278 
279 /*
280  * This function clears the BSS section of the decompressed Linux kernel and NOT the decompressor's.
281  */
282 static void clear_bss_section(unsigned long vmlinux_lma)
283 {
284 	memset((void *)vmlinux_lma + vmlinux.image_size, 0, vmlinux.bss_size);
285 }
286 
287 /*
288  * Set vmalloc area size to an 8th of (potential) physical memory
289  * size, unless size has been set by kernel command line parameter.
290  */
291 static void setup_vmalloc_size(void)
292 {
293 	unsigned long size;
294 
295 	if (vmalloc_size_set)
296 		return;
297 	size = round_up(ident_map_size / 8, _SEGMENT_SIZE);
298 	vmalloc_size = max(size, vmalloc_size);
299 }
300 
301 static void offset_vmlinux_info(unsigned long offset)
302 {
303 	*(unsigned long *)(&vmlinux.entry) += offset;
304 	vmlinux.bootdata_off += offset;
305 	vmlinux.bootdata_preserved_off += offset;
306 	vmlinux.rela_dyn_start += offset;
307 	vmlinux.rela_dyn_end += offset;
308 	vmlinux.dynsym_start += offset;
309 	vmlinux.init_mm_off += offset;
310 	vmlinux.swapper_pg_dir_off += offset;
311 	vmlinux.invalid_pg_dir_off += offset;
312 #ifdef CONFIG_KASAN
313 	vmlinux.kasan_early_shadow_page_off += offset;
314 	vmlinux.kasan_early_shadow_pte_off += offset;
315 	vmlinux.kasan_early_shadow_pmd_off += offset;
316 	vmlinux.kasan_early_shadow_pud_off += offset;
317 	vmlinux.kasan_early_shadow_p4d_off += offset;
318 #endif
319 }
320 
321 void startup_kernel(void)
322 {
323 	unsigned long max_physmem_end;
324 	unsigned long vmlinux_lma = 0;
325 	unsigned long amode31_lma = 0;
326 	unsigned long asce_limit;
327 	unsigned long safe_addr;
328 	void *img;
329 	psw_t psw;
330 
331 	setup_lpp();
332 	safe_addr = mem_safe_offset();
333 
334 	/*
335 	 * Reserve decompressor memory together with decompression heap, buffer and
336 	 * memory which might be occupied by uncompressed kernel at default 1Mb
337 	 * position (if KASLR is off or failed).
338 	 */
339 	physmem_reserve(RR_DECOMPRESSOR, 0, safe_addr);
340 	if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && parmarea.initrd_size)
341 		physmem_reserve(RR_INITRD, parmarea.initrd_start, parmarea.initrd_size);
342 	oldmem_data.start = parmarea.oldmem_base;
343 	oldmem_data.size = parmarea.oldmem_size;
344 
345 	store_ipl_parmblock();
346 	read_ipl_report();
347 	uv_query_info();
348 	sclp_early_read_info();
349 	setup_boot_command_line();
350 	parse_boot_command_line();
351 	detect_facilities();
352 	cmma_init();
353 	sanitize_prot_virt_host();
354 	max_physmem_end = detect_max_physmem_end();
355 	setup_ident_map_size(max_physmem_end);
356 	setup_vmalloc_size();
357 	asce_limit = setup_kernel_memory_layout();
358 	/* got final ident_map_size, physmem allocations could be performed now */
359 	physmem_set_usable_limit(ident_map_size);
360 	detect_physmem_online_ranges(max_physmem_end);
361 	save_ipl_cert_comp_list();
362 	rescue_initrd(safe_addr, ident_map_size);
363 
364 	if (kaslr_enabled()) {
365 		vmlinux_lma = randomize_within_range(vmlinux.image_size + vmlinux.bss_size,
366 						     THREAD_SIZE, vmlinux.default_lma,
367 						     ident_map_size);
368 		if (vmlinux_lma) {
369 			__kaslr_offset = vmlinux_lma - vmlinux.default_lma;
370 			offset_vmlinux_info(__kaslr_offset);
371 		}
372 	}
373 	vmlinux_lma = vmlinux_lma ?: vmlinux.default_lma;
374 	physmem_reserve(RR_VMLINUX, vmlinux_lma, vmlinux.image_size + vmlinux.bss_size);
375 
376 	if (!IS_ENABLED(CONFIG_KERNEL_UNCOMPRESSED)) {
377 		img = decompress_kernel();
378 		memmove((void *)vmlinux_lma, img, vmlinux.image_size);
379 	} else if (__kaslr_offset) {
380 		img = (void *)vmlinux.default_lma;
381 		memmove((void *)vmlinux_lma, img, vmlinux.image_size);
382 		memset(img, 0, vmlinux.image_size);
383 	}
384 
385 	/* vmlinux decompression is done, shrink reserved low memory */
386 	physmem_reserve(RR_DECOMPRESSOR, 0, (unsigned long)_decompressor_end);
387 	if (kaslr_enabled())
388 		amode31_lma = randomize_within_range(vmlinux.amode31_size, PAGE_SIZE, 0, SZ_2G);
389 	amode31_lma = amode31_lma ?: vmlinux.default_lma - vmlinux.amode31_size;
390 	physmem_reserve(RR_AMODE31, amode31_lma, vmlinux.amode31_size);
391 
392 	/*
393 	 * The order of the following operations is important:
394 	 *
395 	 * - handle_relocs() must follow clear_bss_section() to establish static
396 	 *   memory references to data in .bss to be used by setup_vmem()
397 	 *   (i.e init_mm.pgd)
398 	 *
399 	 * - setup_vmem() must follow handle_relocs() to be able using
400 	 *   static memory references to data in .bss (i.e init_mm.pgd)
401 	 *
402 	 * - copy_bootdata() must follow setup_vmem() to propagate changes to
403 	 *   bootdata made by setup_vmem()
404 	 */
405 	clear_bss_section(vmlinux_lma);
406 	handle_relocs(__kaslr_offset);
407 	setup_vmem(asce_limit);
408 	copy_bootdata();
409 
410 	/*
411 	 * Save KASLR offset for early dumps, before vmcore_info is set.
412 	 * Mark as uneven to distinguish from real vmcore_info pointer.
413 	 */
414 	S390_lowcore.vmcore_info = __kaslr_offset ? __kaslr_offset | 0x1UL : 0;
415 
416 	/*
417 	 * Jump to the decompressed kernel entry point and switch DAT mode on.
418 	 */
419 	psw.addr = vmlinux.entry;
420 	psw.mask = PSW_KERNEL_BITS;
421 	__load_psw(psw);
422 }
423