xref: /openbmc/linux/arch/s390/boot/startup.c (revision e368cd72)
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/string.h>
3 #include <linux/elf.h>
4 #include <asm/boot_data.h>
5 #include <asm/sections.h>
6 #include <asm/cpu_mf.h>
7 #include <asm/setup.h>
8 #include <asm/kasan.h>
9 #include <asm/kexec.h>
10 #include <asm/sclp.h>
11 #include <asm/diag.h>
12 #include <asm/uv.h>
13 #include "compressed/decompressor.h"
14 #include "boot.h"
15 #include "uv.h"
16 
17 unsigned long __bootdata_preserved(__kaslr_offset);
18 unsigned long __bootdata_preserved(VMALLOC_START);
19 unsigned long __bootdata_preserved(VMALLOC_END);
20 struct page *__bootdata_preserved(vmemmap);
21 unsigned long __bootdata_preserved(vmemmap_size);
22 unsigned long __bootdata_preserved(MODULES_VADDR);
23 unsigned long __bootdata_preserved(MODULES_END);
24 unsigned long __bootdata(ident_map_size);
25 int __bootdata(is_full_image) = 1;
26 struct initrd_data __bootdata(initrd_data);
27 
28 u64 __bootdata_preserved(stfle_fac_list[16]);
29 u64 __bootdata_preserved(alt_stfle_fac_list[16]);
30 struct oldmem_data __bootdata_preserved(oldmem_data);
31 
32 void error(char *x)
33 {
34 	sclp_early_printk("\n\n");
35 	sclp_early_printk(x);
36 	sclp_early_printk("\n\n -- System halted");
37 
38 	disabled_wait();
39 }
40 
41 static void setup_lpp(void)
42 {
43 	S390_lowcore.current_pid = 0;
44 	S390_lowcore.lpp = LPP_MAGIC;
45 	if (test_facility(40))
46 		lpp(&S390_lowcore.lpp);
47 }
48 
49 #ifdef CONFIG_KERNEL_UNCOMPRESSED
50 unsigned long mem_safe_offset(void)
51 {
52 	return vmlinux.default_lma + vmlinux.image_size + vmlinux.bss_size;
53 }
54 #endif
55 
56 static void rescue_initrd(unsigned long addr)
57 {
58 	if (!IS_ENABLED(CONFIG_BLK_DEV_INITRD))
59 		return;
60 	if (!initrd_data.start || !initrd_data.size)
61 		return;
62 	if (addr <= initrd_data.start)
63 		return;
64 	memmove((void *)addr, (void *)initrd_data.start, initrd_data.size);
65 	initrd_data.start = addr;
66 }
67 
68 static void copy_bootdata(void)
69 {
70 	if (__boot_data_end - __boot_data_start != vmlinux.bootdata_size)
71 		error(".boot.data section size mismatch");
72 	memcpy((void *)vmlinux.bootdata_off, __boot_data_start, vmlinux.bootdata_size);
73 	if (__boot_data_preserved_end - __boot_data_preserved_start != vmlinux.bootdata_preserved_size)
74 		error(".boot.preserved.data section size mismatch");
75 	memcpy((void *)vmlinux.bootdata_preserved_off, __boot_data_preserved_start, vmlinux.bootdata_preserved_size);
76 }
77 
78 static void handle_relocs(unsigned long offset)
79 {
80 	Elf64_Rela *rela_start, *rela_end, *rela;
81 	int r_type, r_sym, rc;
82 	Elf64_Addr loc, val;
83 	Elf64_Sym *dynsym;
84 
85 	rela_start = (Elf64_Rela *) vmlinux.rela_dyn_start;
86 	rela_end = (Elf64_Rela *) vmlinux.rela_dyn_end;
87 	dynsym = (Elf64_Sym *) vmlinux.dynsym_start;
88 	for (rela = rela_start; rela < rela_end; rela++) {
89 		loc = rela->r_offset + offset;
90 		val = rela->r_addend;
91 		r_sym = ELF64_R_SYM(rela->r_info);
92 		if (r_sym) {
93 			if (dynsym[r_sym].st_shndx != SHN_UNDEF)
94 				val += dynsym[r_sym].st_value + offset;
95 		} else {
96 			/*
97 			 * 0 == undefined symbol table index (STN_UNDEF),
98 			 * used for R_390_RELATIVE, only add KASLR offset
99 			 */
100 			val += offset;
101 		}
102 		r_type = ELF64_R_TYPE(rela->r_info);
103 		rc = arch_kexec_do_relocs(r_type, (void *) loc, val, 0);
104 		if (rc)
105 			error("Unknown relocation type");
106 	}
107 }
108 
109 /*
110  * Merge information from several sources into a single ident_map_size value.
111  * "ident_map_size" represents the upper limit of physical memory we may ever
112  * reach. It might not be all online memory, but also include standby (offline)
113  * memory. "ident_map_size" could be lower then actual standby or even online
114  * memory present, due to limiting factors. We should never go above this limit.
115  * It is the size of our identity mapping.
116  *
117  * Consider the following factors:
118  * 1. max_physmem_end - end of physical memory online or standby.
119  *    Always <= end of the last online memory block (get_mem_detect_end()).
120  * 2. CONFIG_MAX_PHYSMEM_BITS - the maximum size of physical memory the
121  *    kernel is able to support.
122  * 3. "mem=" kernel command line option which limits physical memory usage.
123  * 4. OLDMEM_BASE which is a kdump memory limit when the kernel is executed as
124  *    crash kernel.
125  * 5. "hsa" size which is a memory limit when the kernel is executed during
126  *    zfcp/nvme dump.
127  */
128 static void setup_ident_map_size(unsigned long max_physmem_end)
129 {
130 	unsigned long hsa_size;
131 
132 	ident_map_size = max_physmem_end;
133 	if (memory_limit)
134 		ident_map_size = min(ident_map_size, memory_limit);
135 	ident_map_size = min(ident_map_size, 1UL << MAX_PHYSMEM_BITS);
136 
137 #ifdef CONFIG_CRASH_DUMP
138 	if (oldmem_data.start) {
139 		kaslr_enabled = 0;
140 		ident_map_size = min(ident_map_size, oldmem_data.size);
141 	} else if (ipl_block_valid && is_ipl_block_dump()) {
142 		kaslr_enabled = 0;
143 		if (!sclp_early_get_hsa_size(&hsa_size) && hsa_size)
144 			ident_map_size = min(ident_map_size, hsa_size);
145 	}
146 #endif
147 }
148 
149 static void setup_kernel_memory_layout(void)
150 {
151 	bool vmalloc_size_verified = false;
152 	unsigned long vmemmap_off;
153 	unsigned long vspace_left;
154 	unsigned long rte_size;
155 	unsigned long pages;
156 	unsigned long vmax;
157 
158 	pages = ident_map_size / PAGE_SIZE;
159 	/* vmemmap contains a multiple of PAGES_PER_SECTION struct pages */
160 	vmemmap_size = SECTION_ALIGN_UP(pages) * sizeof(struct page);
161 
162 	/* choose kernel address space layout: 4 or 3 levels. */
163 	vmemmap_off = round_up(ident_map_size, _REGION3_SIZE);
164 	if (IS_ENABLED(CONFIG_KASAN) ||
165 	    vmalloc_size > _REGION2_SIZE ||
166 	    vmemmap_off + vmemmap_size + vmalloc_size + MODULES_LEN > _REGION2_SIZE)
167 		vmax = _REGION1_SIZE;
168 	else
169 		vmax = _REGION2_SIZE;
170 
171 	/* keep vmemmap_off aligned to a top level region table entry */
172 	rte_size = vmax == _REGION1_SIZE ? _REGION2_SIZE : _REGION3_SIZE;
173 	MODULES_END = vmax;
174 	if (is_prot_virt_host()) {
175 		/*
176 		 * forcing modules and vmalloc area under the ultravisor
177 		 * secure storage limit, so that any vmalloc allocation
178 		 * we do could be used to back secure guest storage.
179 		 */
180 		adjust_to_uv_max(&MODULES_END);
181 	}
182 
183 #ifdef CONFIG_KASAN
184 	if (MODULES_END < vmax) {
185 		/* force vmalloc and modules below kasan shadow */
186 		MODULES_END = min(MODULES_END, KASAN_SHADOW_START);
187 	} else {
188 		/*
189 		 * leave vmalloc and modules above kasan shadow but make
190 		 * sure they don't overlap with it
191 		 */
192 		vmalloc_size = min(vmalloc_size, vmax - KASAN_SHADOW_END - MODULES_LEN);
193 		vmalloc_size_verified = true;
194 		vspace_left = KASAN_SHADOW_START;
195 	}
196 #endif
197 	MODULES_VADDR = MODULES_END - MODULES_LEN;
198 	VMALLOC_END = MODULES_VADDR;
199 
200 	if (vmalloc_size_verified) {
201 		VMALLOC_START = VMALLOC_END - vmalloc_size;
202 	} else {
203 		vmemmap_off = round_up(ident_map_size, rte_size);
204 
205 		if (vmemmap_off + vmemmap_size > VMALLOC_END ||
206 		    vmalloc_size > VMALLOC_END - vmemmap_off - vmemmap_size) {
207 			/*
208 			 * allow vmalloc area to occupy up to 1/2 of
209 			 * the rest virtual space left.
210 			 */
211 			vmalloc_size = min(vmalloc_size, VMALLOC_END / 2);
212 		}
213 		VMALLOC_START = VMALLOC_END - vmalloc_size;
214 		vspace_left = VMALLOC_START;
215 	}
216 
217 	pages = vspace_left / (PAGE_SIZE + sizeof(struct page));
218 	pages = SECTION_ALIGN_UP(pages);
219 	vmemmap_off = round_up(vspace_left - pages * sizeof(struct page), rte_size);
220 	/* keep vmemmap left most starting from a fresh region table entry */
221 	vmemmap_off = min(vmemmap_off, round_up(ident_map_size, rte_size));
222 	/* take care that identity map is lower then vmemmap */
223 	ident_map_size = min(ident_map_size, vmemmap_off);
224 	vmemmap_size = SECTION_ALIGN_UP(ident_map_size / PAGE_SIZE) * sizeof(struct page);
225 	VMALLOC_START = max(vmemmap_off + vmemmap_size, VMALLOC_START);
226 	vmemmap = (struct page *)vmemmap_off;
227 }
228 
229 /*
230  * This function clears the BSS section of the decompressed Linux kernel and NOT the decompressor's.
231  */
232 static void clear_bss_section(void)
233 {
234 	memset((void *)vmlinux.default_lma + vmlinux.image_size, 0, vmlinux.bss_size);
235 }
236 
237 /*
238  * Set vmalloc area size to an 8th of (potential) physical memory
239  * size, unless size has been set by kernel command line parameter.
240  */
241 static void setup_vmalloc_size(void)
242 {
243 	unsigned long size;
244 
245 	if (vmalloc_size_set)
246 		return;
247 	size = round_up(ident_map_size / 8, _SEGMENT_SIZE);
248 	vmalloc_size = max(size, vmalloc_size);
249 }
250 
251 static void offset_vmlinux_info(unsigned long offset)
252 {
253 	vmlinux.default_lma += offset;
254 	*(unsigned long *)(&vmlinux.entry) += offset;
255 	vmlinux.bootdata_off += offset;
256 	vmlinux.bootdata_preserved_off += offset;
257 	vmlinux.rela_dyn_start += offset;
258 	vmlinux.rela_dyn_end += offset;
259 	vmlinux.dynsym_start += offset;
260 }
261 
262 void startup_kernel(void)
263 {
264 	unsigned long random_lma;
265 	unsigned long safe_addr;
266 	void *img;
267 
268 	initrd_data.start = parmarea.initrd_start;
269 	initrd_data.size = parmarea.initrd_size;
270 	oldmem_data.start = parmarea.oldmem_base;
271 	oldmem_data.size = parmarea.oldmem_size;
272 
273 	setup_lpp();
274 	store_ipl_parmblock();
275 	safe_addr = mem_safe_offset();
276 	safe_addr = read_ipl_report(safe_addr);
277 	uv_query_info();
278 	rescue_initrd(safe_addr);
279 	sclp_early_read_info();
280 	setup_boot_command_line();
281 	parse_boot_command_line();
282 	sanitize_prot_virt_host();
283 	setup_ident_map_size(detect_memory());
284 	setup_vmalloc_size();
285 	setup_kernel_memory_layout();
286 
287 	if (IS_ENABLED(CONFIG_RANDOMIZE_BASE) && kaslr_enabled) {
288 		random_lma = get_random_base(safe_addr);
289 		if (random_lma) {
290 			__kaslr_offset = random_lma - vmlinux.default_lma;
291 			img = (void *)vmlinux.default_lma;
292 			offset_vmlinux_info(__kaslr_offset);
293 		}
294 	}
295 
296 	if (!IS_ENABLED(CONFIG_KERNEL_UNCOMPRESSED)) {
297 		img = decompress_kernel();
298 		memmove((void *)vmlinux.default_lma, img, vmlinux.image_size);
299 	} else if (__kaslr_offset)
300 		memcpy((void *)vmlinux.default_lma, img, vmlinux.image_size);
301 
302 	clear_bss_section();
303 	copy_bootdata();
304 	if (IS_ENABLED(CONFIG_RELOCATABLE))
305 		handle_relocs(__kaslr_offset);
306 
307 	if (__kaslr_offset) {
308 		/*
309 		 * Save KASLR offset for early dumps, before vmcore_info is set.
310 		 * Mark as uneven to distinguish from real vmcore_info pointer.
311 		 */
312 		S390_lowcore.vmcore_info = __kaslr_offset | 0x1UL;
313 		/* Clear non-relocated kernel */
314 		if (IS_ENABLED(CONFIG_KERNEL_UNCOMPRESSED))
315 			memset(img, 0, vmlinux.image_size);
316 	}
317 	vmlinux.entry();
318 }
319