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/maccess.h>
7 #include <asm/cpu_mf.h>
8 #include <asm/setup.h>
9 #include <asm/kasan.h>
10 #include <asm/kexec.h>
11 #include <asm/sclp.h>
12 #include <asm/diag.h>
13 #include <asm/uv.h>
14 #include <asm/abs_lowcore.h>
15 #include <asm/mem_detect.h>
16 #include "decompressor.h"
17 #include "boot.h"
18 #include "uv.h"
19
20 unsigned long __bootdata_preserved(__kaslr_offset);
21 unsigned long __bootdata_preserved(__abs_lowcore);
22 unsigned long __bootdata_preserved(__memcpy_real_area);
23 pte_t *__bootdata_preserved(memcpy_real_ptep);
24 unsigned long __bootdata(__amode31_base);
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(ident_map_size);
32 int __bootdata(is_full_image) = 1;
33 struct initrd_data __bootdata(initrd_data);
34
35 u64 __bootdata_preserved(stfle_fac_list[16]);
36 u64 __bootdata_preserved(alt_stfle_fac_list[16]);
37 struct oldmem_data __bootdata_preserved(oldmem_data);
38
39 struct machine_info machine;
40
error(char * x)41 void error(char *x)
42 {
43 sclp_early_printk("\n\n");
44 sclp_early_printk(x);
45 sclp_early_printk("\n\n -- System halted");
46
47 disabled_wait();
48 }
49
detect_facilities(void)50 static void detect_facilities(void)
51 {
52 if (test_facility(8)) {
53 machine.has_edat1 = 1;
54 __ctl_set_bit(0, 23);
55 }
56 if (test_facility(78))
57 machine.has_edat2 = 1;
58 if (!noexec_disabled && test_facility(130)) {
59 machine.has_nx = 1;
60 __ctl_set_bit(0, 20);
61 }
62 }
63
setup_lpp(void)64 static void setup_lpp(void)
65 {
66 S390_lowcore.current_pid = 0;
67 S390_lowcore.lpp = LPP_MAGIC;
68 if (test_facility(40))
69 lpp(&S390_lowcore.lpp);
70 }
71
72 #ifdef CONFIG_KERNEL_UNCOMPRESSED
mem_safe_offset(void)73 unsigned long mem_safe_offset(void)
74 {
75 return vmlinux.default_lma + vmlinux.image_size + vmlinux.bss_size;
76 }
77 #endif
78
rescue_initrd(unsigned long safe_addr)79 static unsigned long rescue_initrd(unsigned long safe_addr)
80 {
81 if (!IS_ENABLED(CONFIG_BLK_DEV_INITRD))
82 return safe_addr;
83 if (!initrd_data.start || !initrd_data.size)
84 return safe_addr;
85 if (initrd_data.start < safe_addr) {
86 memmove((void *)safe_addr, (void *)initrd_data.start, initrd_data.size);
87 initrd_data.start = safe_addr;
88 }
89 return initrd_data.start + initrd_data.size;
90 }
91
copy_bootdata(void)92 static void copy_bootdata(void)
93 {
94 if (__boot_data_end - __boot_data_start != vmlinux.bootdata_size)
95 error(".boot.data section size mismatch");
96 memcpy((void *)vmlinux.bootdata_off, __boot_data_start, vmlinux.bootdata_size);
97 if (__boot_data_preserved_end - __boot_data_preserved_start != vmlinux.bootdata_preserved_size)
98 error(".boot.preserved.data section size mismatch");
99 memcpy((void *)vmlinux.bootdata_preserved_off, __boot_data_preserved_start, vmlinux.bootdata_preserved_size);
100 }
101
handle_relocs(unsigned long offset)102 static void handle_relocs(unsigned long offset)
103 {
104 Elf64_Rela *rela_start, *rela_end, *rela;
105 int r_type, r_sym, rc;
106 Elf64_Addr loc, val;
107 Elf64_Sym *dynsym;
108
109 rela_start = (Elf64_Rela *) vmlinux.rela_dyn_start;
110 rela_end = (Elf64_Rela *) vmlinux.rela_dyn_end;
111 dynsym = (Elf64_Sym *) vmlinux.dynsym_start;
112 for (rela = rela_start; rela < rela_end; rela++) {
113 loc = rela->r_offset + offset;
114 val = rela->r_addend;
115 r_sym = ELF64_R_SYM(rela->r_info);
116 if (r_sym) {
117 if (dynsym[r_sym].st_shndx != SHN_UNDEF)
118 val += dynsym[r_sym].st_value + offset;
119 } else {
120 /*
121 * 0 == undefined symbol table index (STN_UNDEF),
122 * used for R_390_RELATIVE, only add KASLR offset
123 */
124 val += offset;
125 }
126 r_type = ELF64_R_TYPE(rela->r_info);
127 rc = arch_kexec_do_relocs(r_type, (void *) loc, val, 0);
128 if (rc)
129 error("Unknown relocation type");
130 }
131 }
132
133 /*
134 * Merge information from several sources into a single ident_map_size value.
135 * "ident_map_size" represents the upper limit of physical memory we may ever
136 * reach. It might not be all online memory, but also include standby (offline)
137 * memory. "ident_map_size" could be lower then actual standby or even online
138 * memory present, due to limiting factors. We should never go above this limit.
139 * It is the size of our identity mapping.
140 *
141 * Consider the following factors:
142 * 1. max_physmem_end - end of physical memory online or standby.
143 * Always <= end of the last online memory block (get_mem_detect_end()).
144 * 2. CONFIG_MAX_PHYSMEM_BITS - the maximum size of physical memory the
145 * kernel is able to support.
146 * 3. "mem=" kernel command line option which limits physical memory usage.
147 * 4. OLDMEM_BASE which is a kdump memory limit when the kernel is executed as
148 * crash kernel.
149 * 5. "hsa" size which is a memory limit when the kernel is executed during
150 * zfcp/nvme dump.
151 */
setup_ident_map_size(unsigned long max_physmem_end)152 static void setup_ident_map_size(unsigned long max_physmem_end)
153 {
154 unsigned long hsa_size;
155
156 ident_map_size = max_physmem_end;
157 if (memory_limit)
158 ident_map_size = min(ident_map_size, memory_limit);
159 ident_map_size = min(ident_map_size, 1UL << MAX_PHYSMEM_BITS);
160
161 #ifdef CONFIG_CRASH_DUMP
162 if (oldmem_data.start) {
163 kaslr_enabled = 0;
164 ident_map_size = min(ident_map_size, oldmem_data.size);
165 } else if (ipl_block_valid && is_ipl_block_dump()) {
166 kaslr_enabled = 0;
167 if (!sclp_early_get_hsa_size(&hsa_size) && hsa_size)
168 ident_map_size = min(ident_map_size, hsa_size);
169 }
170 #endif
171 }
172
setup_kernel_memory_layout(void)173 static unsigned long setup_kernel_memory_layout(void)
174 {
175 unsigned long vmemmap_start;
176 unsigned long asce_limit;
177 unsigned long rte_size;
178 unsigned long pages;
179 unsigned long vmax;
180
181 pages = ident_map_size / PAGE_SIZE;
182 /* vmemmap contains a multiple of PAGES_PER_SECTION struct pages */
183 vmemmap_size = SECTION_ALIGN_UP(pages) * sizeof(struct page);
184
185 /* choose kernel address space layout: 4 or 3 levels. */
186 vmemmap_start = round_up(ident_map_size, _REGION3_SIZE);
187 if (IS_ENABLED(CONFIG_KASAN) ||
188 vmalloc_size > _REGION2_SIZE ||
189 vmemmap_start + vmemmap_size + vmalloc_size + MODULES_LEN >
190 _REGION2_SIZE) {
191 asce_limit = _REGION1_SIZE;
192 rte_size = _REGION2_SIZE;
193 } else {
194 asce_limit = _REGION2_SIZE;
195 rte_size = _REGION3_SIZE;
196 }
197 /*
198 * forcing modules and vmalloc area under the ultravisor
199 * secure storage limit, so that any vmalloc allocation
200 * we do could be used to back secure guest storage.
201 */
202 vmax = adjust_to_uv_max(asce_limit);
203 #ifdef CONFIG_KASAN
204 /* force vmalloc and modules below kasan shadow */
205 vmax = min(vmax, KASAN_SHADOW_START);
206 #endif
207 __memcpy_real_area = round_down(vmax - PAGE_SIZE, PAGE_SIZE);
208 __abs_lowcore = round_down(__memcpy_real_area - ABS_LOWCORE_MAP_SIZE,
209 sizeof(struct lowcore));
210 MODULES_END = round_down(__abs_lowcore, _SEGMENT_SIZE);
211 MODULES_VADDR = MODULES_END - MODULES_LEN;
212 VMALLOC_END = MODULES_VADDR;
213
214 /* allow vmalloc area to occupy up to about 1/2 of the rest virtual space left */
215 vmalloc_size = min(vmalloc_size, round_down(VMALLOC_END / 2, _REGION3_SIZE));
216 VMALLOC_START = VMALLOC_END - vmalloc_size;
217
218 /* split remaining virtual space between 1:1 mapping & vmemmap array */
219 pages = VMALLOC_START / (PAGE_SIZE + sizeof(struct page));
220 pages = SECTION_ALIGN_UP(pages);
221 /* keep vmemmap_start aligned to a top level region table entry */
222 vmemmap_start = round_down(VMALLOC_START - pages * sizeof(struct page), rte_size);
223 /* vmemmap_start is the future VMEM_MAX_PHYS, make sure it is within MAX_PHYSMEM */
224 vmemmap_start = min(vmemmap_start, 1UL << MAX_PHYSMEM_BITS);
225 /* make sure identity map doesn't overlay with vmemmap */
226 ident_map_size = min(ident_map_size, vmemmap_start);
227 vmemmap_size = SECTION_ALIGN_UP(ident_map_size / PAGE_SIZE) * sizeof(struct page);
228 /* make sure vmemmap doesn't overlay with vmalloc area */
229 VMALLOC_START = max(vmemmap_start + vmemmap_size, VMALLOC_START);
230 vmemmap = (struct page *)vmemmap_start;
231
232 return asce_limit;
233 }
234
235 /*
236 * This function clears the BSS section of the decompressed Linux kernel and NOT the decompressor's.
237 */
clear_bss_section(void)238 static void clear_bss_section(void)
239 {
240 memset((void *)vmlinux.default_lma + vmlinux.image_size, 0, vmlinux.bss_size);
241 }
242
243 /*
244 * Set vmalloc area size to an 8th of (potential) physical memory
245 * size, unless size has been set by kernel command line parameter.
246 */
setup_vmalloc_size(void)247 static void setup_vmalloc_size(void)
248 {
249 unsigned long size;
250
251 if (vmalloc_size_set)
252 return;
253 size = round_up(ident_map_size / 8, _SEGMENT_SIZE);
254 vmalloc_size = max(size, vmalloc_size);
255 }
256
offset_vmlinux_info(unsigned long offset)257 static void offset_vmlinux_info(unsigned long offset)
258 {
259 vmlinux.default_lma += offset;
260 *(unsigned long *)(&vmlinux.entry) += offset;
261 vmlinux.bootdata_off += offset;
262 vmlinux.bootdata_preserved_off += offset;
263 vmlinux.rela_dyn_start += offset;
264 vmlinux.rela_dyn_end += offset;
265 vmlinux.dynsym_start += offset;
266 vmlinux.init_mm_off += offset;
267 vmlinux.swapper_pg_dir_off += offset;
268 vmlinux.invalid_pg_dir_off += offset;
269 }
270
reserve_amode31(unsigned long safe_addr)271 static unsigned long reserve_amode31(unsigned long safe_addr)
272 {
273 __amode31_base = PAGE_ALIGN(safe_addr);
274 return __amode31_base + vmlinux.amode31_size;
275 }
276
startup_kernel(void)277 void startup_kernel(void)
278 {
279 unsigned long max_physmem_end;
280 unsigned long random_lma;
281 unsigned long safe_addr;
282 unsigned long asce_limit;
283 void *img;
284 psw_t psw;
285
286 initrd_data.start = parmarea.initrd_start;
287 initrd_data.size = parmarea.initrd_size;
288 oldmem_data.start = parmarea.oldmem_base;
289 oldmem_data.size = parmarea.oldmem_size;
290
291 setup_lpp();
292 store_ipl_parmblock();
293 safe_addr = mem_safe_offset();
294 safe_addr = reserve_amode31(safe_addr);
295 safe_addr = read_ipl_report(safe_addr);
296 uv_query_info();
297 safe_addr = rescue_initrd(safe_addr);
298 sclp_early_read_info();
299 setup_boot_command_line();
300 parse_boot_command_line();
301 detect_facilities();
302 sanitize_prot_virt_host();
303 max_physmem_end = detect_memory(&safe_addr);
304 setup_ident_map_size(max_physmem_end);
305 setup_vmalloc_size();
306 asce_limit = setup_kernel_memory_layout();
307 mem_detect_set_usable_limit(ident_map_size);
308
309 if (IS_ENABLED(CONFIG_RANDOMIZE_BASE) && kaslr_enabled) {
310 random_lma = get_random_base(safe_addr);
311 if (random_lma) {
312 __kaslr_offset = random_lma - vmlinux.default_lma;
313 img = (void *)vmlinux.default_lma;
314 offset_vmlinux_info(__kaslr_offset);
315 }
316 }
317
318 if (!IS_ENABLED(CONFIG_KERNEL_UNCOMPRESSED)) {
319 img = decompress_kernel();
320 memmove((void *)vmlinux.default_lma, img, vmlinux.image_size);
321 } else if (__kaslr_offset)
322 memcpy((void *)vmlinux.default_lma, img, vmlinux.image_size);
323
324 /*
325 * The order of the following operations is important:
326 *
327 * - handle_relocs() must follow clear_bss_section() to establish static
328 * memory references to data in .bss to be used by setup_vmem()
329 * (i.e init_mm.pgd)
330 *
331 * - setup_vmem() must follow handle_relocs() to be able using
332 * static memory references to data in .bss (i.e init_mm.pgd)
333 *
334 * - copy_bootdata() must follow setup_vmem() to propagate changes to
335 * bootdata made by setup_vmem()
336 */
337 clear_bss_section();
338 handle_relocs(__kaslr_offset);
339 setup_vmem(asce_limit);
340 copy_bootdata();
341
342 if (__kaslr_offset) {
343 /*
344 * Save KASLR offset for early dumps, before vmcore_info is set.
345 * Mark as uneven to distinguish from real vmcore_info pointer.
346 */
347 S390_lowcore.vmcore_info = __kaslr_offset | 0x1UL;
348 /* Clear non-relocated kernel */
349 if (IS_ENABLED(CONFIG_KERNEL_UNCOMPRESSED))
350 memset(img, 0, vmlinux.image_size);
351 }
352
353 /*
354 * Jump to the decompressed kernel entry point and switch DAT mode on.
355 */
356 psw.addr = vmlinux.entry;
357 psw.mask = PSW_KERNEL_BITS;
358 __load_psw(psw);
359 }
360