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/physmem_info.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_preserved(VMALLOC_START);
25 unsigned long __bootdata_preserved(VMALLOC_END);
26 struct page *__bootdata_preserved(vmemmap);
27 unsigned long __bootdata_preserved(vmemmap_size);
28 unsigned long __bootdata_preserved(MODULES_VADDR);
29 unsigned long __bootdata_preserved(MODULES_END);
30 unsigned long __bootdata_preserved(max_mappable);
31 unsigned long __bootdata(ident_map_size);
32
33 u64 __bootdata_preserved(stfle_fac_list[16]);
34 u64 __bootdata_preserved(alt_stfle_fac_list[16]);
35 struct oldmem_data __bootdata_preserved(oldmem_data);
36
37 struct machine_info machine;
38
error(char * x)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
detect_facilities(void)48 static void detect_facilities(void)
49 {
50 if (test_facility(8)) {
51 machine.has_edat1 = 1;
52 __ctl_set_bit(0, 23);
53 }
54 if (test_facility(78))
55 machine.has_edat2 = 1;
56 if (test_facility(130))
57 machine.has_nx = 1;
58 }
59
setup_lpp(void)60 static void setup_lpp(void)
61 {
62 S390_lowcore.current_pid = 0;
63 S390_lowcore.lpp = LPP_MAGIC;
64 if (test_facility(40))
65 lpp(&S390_lowcore.lpp);
66 }
67
68 #ifdef CONFIG_KERNEL_UNCOMPRESSED
mem_safe_offset(void)69 unsigned long mem_safe_offset(void)
70 {
71 return vmlinux.default_lma + vmlinux.image_size + vmlinux.bss_size;
72 }
73 #endif
74
rescue_initrd(unsigned long min,unsigned long max)75 static void rescue_initrd(unsigned long min, unsigned long max)
76 {
77 unsigned long old_addr, addr, size;
78
79 if (!IS_ENABLED(CONFIG_BLK_DEV_INITRD))
80 return;
81 if (!get_physmem_reserved(RR_INITRD, &addr, &size))
82 return;
83 if (addr >= min && addr + size <= max)
84 return;
85 old_addr = addr;
86 physmem_free(RR_INITRD);
87 addr = physmem_alloc_top_down(RR_INITRD, size, 0);
88 memmove((void *)addr, (void *)old_addr, size);
89 }
90
copy_bootdata(void)91 static void copy_bootdata(void)
92 {
93 if (__boot_data_end - __boot_data_start != vmlinux.bootdata_size)
94 error(".boot.data section size mismatch");
95 memcpy((void *)vmlinux.bootdata_off, __boot_data_start, vmlinux.bootdata_size);
96 if (__boot_data_preserved_end - __boot_data_preserved_start != vmlinux.bootdata_preserved_size)
97 error(".boot.preserved.data section size mismatch");
98 memcpy((void *)vmlinux.bootdata_preserved_off, __boot_data_preserved_start, vmlinux.bootdata_preserved_size);
99 }
100
handle_relocs(unsigned long offset)101 static void handle_relocs(unsigned long offset)
102 {
103 Elf64_Rela *rela_start, *rela_end, *rela;
104 int r_type, r_sym, rc;
105 Elf64_Addr loc, val;
106 Elf64_Sym *dynsym;
107
108 rela_start = (Elf64_Rela *) vmlinux.rela_dyn_start;
109 rela_end = (Elf64_Rela *) vmlinux.rela_dyn_end;
110 dynsym = (Elf64_Sym *) vmlinux.dynsym_start;
111 for (rela = rela_start; rela < rela_end; rela++) {
112 loc = rela->r_offset + offset;
113 val = rela->r_addend;
114 r_sym = ELF64_R_SYM(rela->r_info);
115 if (r_sym) {
116 if (dynsym[r_sym].st_shndx != SHN_UNDEF)
117 val += dynsym[r_sym].st_value + offset;
118 } else {
119 /*
120 * 0 == undefined symbol table index (STN_UNDEF),
121 * used for R_390_RELATIVE, only add KASLR offset
122 */
123 val += offset;
124 }
125 r_type = ELF64_R_TYPE(rela->r_info);
126 rc = arch_kexec_do_relocs(r_type, (void *) loc, val, 0);
127 if (rc)
128 error("Unknown relocation type");
129 }
130 }
131
132 /*
133 * Merge information from several sources into a single ident_map_size value.
134 * "ident_map_size" represents the upper limit of physical memory we may ever
135 * reach. It might not be all online memory, but also include standby (offline)
136 * memory. "ident_map_size" could be lower then actual standby or even online
137 * memory present, due to limiting factors. We should never go above this limit.
138 * It is the size of our identity mapping.
139 *
140 * Consider the following factors:
141 * 1. max_physmem_end - end of physical memory online or standby.
142 * Always >= end of the last online memory range (get_physmem_online_end()).
143 * 2. CONFIG_MAX_PHYSMEM_BITS - the maximum size of physical memory the
144 * kernel is able to support.
145 * 3. "mem=" kernel command line option which limits physical memory usage.
146 * 4. OLDMEM_BASE which is a kdump memory limit when the kernel is executed as
147 * crash kernel.
148 * 5. "hsa" size which is a memory limit when the kernel is executed during
149 * zfcp/nvme dump.
150 */
setup_ident_map_size(unsigned long max_physmem_end)151 static void setup_ident_map_size(unsigned long max_physmem_end)
152 {
153 unsigned long hsa_size;
154
155 ident_map_size = max_physmem_end;
156 if (memory_limit)
157 ident_map_size = min(ident_map_size, memory_limit);
158 ident_map_size = min(ident_map_size, 1UL << MAX_PHYSMEM_BITS);
159
160 #ifdef CONFIG_CRASH_DUMP
161 if (oldmem_data.start) {
162 __kaslr_enabled = 0;
163 ident_map_size = min(ident_map_size, oldmem_data.size);
164 } else if (ipl_block_valid && is_ipl_block_dump()) {
165 __kaslr_enabled = 0;
166 if (!sclp_early_get_hsa_size(&hsa_size) && hsa_size)
167 ident_map_size = min(ident_map_size, hsa_size);
168 }
169 #endif
170 }
171
setup_kernel_memory_layout(void)172 static unsigned long setup_kernel_memory_layout(void)
173 {
174 unsigned long vmemmap_start;
175 unsigned long asce_limit;
176 unsigned long rte_size;
177 unsigned long pages;
178 unsigned long vsize;
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 vsize = round_up(ident_map_size, _REGION3_SIZE) + vmemmap_size +
187 MODULES_LEN + MEMCPY_REAL_SIZE + ABS_LOWCORE_MAP_SIZE;
188 vsize = size_add(vsize, vmalloc_size);
189 if (IS_ENABLED(CONFIG_KASAN) || (vsize > _REGION2_SIZE)) {
190 asce_limit = _REGION1_SIZE;
191 rte_size = _REGION2_SIZE;
192 } else {
193 asce_limit = _REGION2_SIZE;
194 rte_size = _REGION3_SIZE;
195 }
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 - MEMCPY_REAL_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 vsize = round_down(VMALLOC_END / 2, _SEGMENT_SIZE);
216 vmalloc_size = min(vmalloc_size, vsize);
217 VMALLOC_START = VMALLOC_END - vmalloc_size;
218
219 /* split remaining virtual space between 1:1 mapping & vmemmap array */
220 pages = VMALLOC_START / (PAGE_SIZE + sizeof(struct page));
221 pages = SECTION_ALIGN_UP(pages);
222 /* keep vmemmap_start aligned to a top level region table entry */
223 vmemmap_start = round_down(VMALLOC_START - pages * sizeof(struct page), rte_size);
224 vmemmap_start = min(vmemmap_start, 1UL << MAX_PHYSMEM_BITS);
225 /* maximum mappable address as seen by arch_get_mappable_range() */
226 max_mappable = vmemmap_start;
227 /* make sure identity map doesn't overlay with vmemmap */
228 ident_map_size = min(ident_map_size, vmemmap_start);
229 vmemmap_size = SECTION_ALIGN_UP(ident_map_size / PAGE_SIZE) * sizeof(struct page);
230 /* make sure vmemmap doesn't overlay with vmalloc area */
231 VMALLOC_START = max(vmemmap_start + vmemmap_size, VMALLOC_START);
232 vmemmap = (struct page *)vmemmap_start;
233
234 return asce_limit;
235 }
236
237 /*
238 * This function clears the BSS section of the decompressed Linux kernel and NOT the decompressor's.
239 */
clear_bss_section(unsigned long vmlinux_lma)240 static void clear_bss_section(unsigned long vmlinux_lma)
241 {
242 memset((void *)vmlinux_lma + vmlinux.image_size, 0, vmlinux.bss_size);
243 }
244
245 /*
246 * Set vmalloc area size to an 8th of (potential) physical memory
247 * size, unless size has been set by kernel command line parameter.
248 */
setup_vmalloc_size(void)249 static void setup_vmalloc_size(void)
250 {
251 unsigned long size;
252
253 if (vmalloc_size_set)
254 return;
255 size = round_up(ident_map_size / 8, _SEGMENT_SIZE);
256 vmalloc_size = max(size, vmalloc_size);
257 }
258
offset_vmlinux_info(unsigned long offset)259 static void offset_vmlinux_info(unsigned long offset)
260 {
261 *(unsigned long *)(&vmlinux.entry) += offset;
262 vmlinux.bootdata_off += offset;
263 vmlinux.bootdata_preserved_off += offset;
264 vmlinux.rela_dyn_start += offset;
265 vmlinux.rela_dyn_end += offset;
266 vmlinux.dynsym_start += offset;
267 vmlinux.init_mm_off += offset;
268 vmlinux.swapper_pg_dir_off += offset;
269 vmlinux.invalid_pg_dir_off += offset;
270 #ifdef CONFIG_KASAN
271 vmlinux.kasan_early_shadow_page_off += offset;
272 vmlinux.kasan_early_shadow_pte_off += offset;
273 vmlinux.kasan_early_shadow_pmd_off += offset;
274 vmlinux.kasan_early_shadow_pud_off += offset;
275 vmlinux.kasan_early_shadow_p4d_off += offset;
276 #endif
277 }
278
startup_kernel(void)279 void startup_kernel(void)
280 {
281 unsigned long max_physmem_end;
282 unsigned long vmlinux_lma = 0;
283 unsigned long amode31_lma = 0;
284 unsigned long asce_limit;
285 unsigned long safe_addr;
286 void *img;
287 psw_t psw;
288
289 setup_lpp();
290 safe_addr = mem_safe_offset();
291
292 /*
293 * Reserve decompressor memory together with decompression heap, buffer and
294 * memory which might be occupied by uncompressed kernel at default 1Mb
295 * position (if KASLR is off or failed).
296 */
297 physmem_reserve(RR_DECOMPRESSOR, 0, safe_addr);
298 if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && parmarea.initrd_size)
299 physmem_reserve(RR_INITRD, parmarea.initrd_start, parmarea.initrd_size);
300 oldmem_data.start = parmarea.oldmem_base;
301 oldmem_data.size = parmarea.oldmem_size;
302
303 store_ipl_parmblock();
304 read_ipl_report();
305 uv_query_info();
306 sclp_early_read_info();
307 setup_boot_command_line();
308 parse_boot_command_line();
309 detect_facilities();
310 sanitize_prot_virt_host();
311 max_physmem_end = detect_max_physmem_end();
312 setup_ident_map_size(max_physmem_end);
313 setup_vmalloc_size();
314 asce_limit = setup_kernel_memory_layout();
315 /* got final ident_map_size, physmem allocations could be performed now */
316 physmem_set_usable_limit(ident_map_size);
317 detect_physmem_online_ranges(max_physmem_end);
318 save_ipl_cert_comp_list();
319 rescue_initrd(safe_addr, ident_map_size);
320
321 if (kaslr_enabled()) {
322 vmlinux_lma = randomize_within_range(vmlinux.image_size + vmlinux.bss_size,
323 THREAD_SIZE, vmlinux.default_lma,
324 ident_map_size);
325 if (vmlinux_lma) {
326 __kaslr_offset = vmlinux_lma - vmlinux.default_lma;
327 offset_vmlinux_info(__kaslr_offset);
328 }
329 }
330 vmlinux_lma = vmlinux_lma ?: vmlinux.default_lma;
331 physmem_reserve(RR_VMLINUX, vmlinux_lma, vmlinux.image_size + vmlinux.bss_size);
332
333 if (!IS_ENABLED(CONFIG_KERNEL_UNCOMPRESSED)) {
334 img = decompress_kernel();
335 memmove((void *)vmlinux_lma, img, vmlinux.image_size);
336 } else if (__kaslr_offset) {
337 img = (void *)vmlinux.default_lma;
338 memmove((void *)vmlinux_lma, img, vmlinux.image_size);
339 memset(img, 0, vmlinux.image_size);
340 }
341
342 /* vmlinux decompression is done, shrink reserved low memory */
343 physmem_reserve(RR_DECOMPRESSOR, 0, (unsigned long)_decompressor_end);
344 if (kaslr_enabled())
345 amode31_lma = randomize_within_range(vmlinux.amode31_size, PAGE_SIZE, 0, SZ_2G);
346 amode31_lma = amode31_lma ?: vmlinux.default_lma - vmlinux.amode31_size;
347 physmem_reserve(RR_AMODE31, amode31_lma, vmlinux.amode31_size);
348
349 /*
350 * The order of the following operations is important:
351 *
352 * - handle_relocs() must follow clear_bss_section() to establish static
353 * memory references to data in .bss to be used by setup_vmem()
354 * (i.e init_mm.pgd)
355 *
356 * - setup_vmem() must follow handle_relocs() to be able using
357 * static memory references to data in .bss (i.e init_mm.pgd)
358 *
359 * - copy_bootdata() must follow setup_vmem() to propagate changes to
360 * bootdata made by setup_vmem()
361 */
362 clear_bss_section(vmlinux_lma);
363 handle_relocs(__kaslr_offset);
364 setup_vmem(asce_limit);
365 copy_bootdata();
366
367 /*
368 * Save KASLR offset for early dumps, before vmcore_info is set.
369 * Mark as uneven to distinguish from real vmcore_info pointer.
370 */
371 S390_lowcore.vmcore_info = __kaslr_offset ? __kaslr_offset | 0x1UL : 0;
372
373 /*
374 * Jump to the decompressed kernel entry point and switch DAT mode on.
375 */
376 psw.addr = vmlinux.entry;
377 psw.mask = PSW_KERNEL_BITS;
378 __load_psw(psw);
379 }
380