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
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 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
cmma_test_essa(void)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" (S390_lowcore.program_new_psw),
82 "=Q" (old)
83 : [psw_old] "a" (&old),
84 [psw_pgm] "a" (&S390_lowcore.program_new_psw),
85 [cmd] "i" (ESSA_GET_STATE)
86 : "cc", "memory");
87 return rc;
88 }
89
cmma_init(void)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
setup_lpp(void)102 static void setup_lpp(void)
103 {
104 S390_lowcore.current_pid = 0;
105 S390_lowcore.lpp = LPP_MAGIC;
106 if (test_facility(40))
107 lpp(&S390_lowcore.lpp);
108 }
109
110 #ifdef CONFIG_KERNEL_UNCOMPRESSED
mem_safe_offset(void)111 static unsigned long mem_safe_offset(void)
112 {
113 return (unsigned long)_compressed_start;
114 }
115
deploy_kernel(void * output)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
rescue_initrd(unsigned long min,unsigned long max)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
copy_bootdata(void)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
kaslr_adjust_relocs(unsigned long min_addr,unsigned long max_addr,unsigned long offset,unsigned long phys_offset)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
kaslr_adjust_got(unsigned long offset)168 static void kaslr_adjust_got(unsigned long offset)
169 {
170 u64 *entry;
171
172 /*
173 * Even without -fPIE, Clang still uses a global offset table for some
174 * reason. Adjust the GOT entries.
175 */
176 for (entry = (u64 *)vmlinux.got_start; entry < (u64 *)vmlinux.got_end; entry++)
177 *entry += offset - __START_KERNEL;
178 }
179
180 /*
181 * Merge information from several sources into a single ident_map_size value.
182 * "ident_map_size" represents the upper limit of physical memory we may ever
183 * reach. It might not be all online memory, but also include standby (offline)
184 * memory. "ident_map_size" could be lower then actual standby or even online
185 * memory present, due to limiting factors. We should never go above this limit.
186 * It is the size of our identity mapping.
187 *
188 * Consider the following factors:
189 * 1. max_physmem_end - end of physical memory online or standby.
190 * Always >= end of the last online memory range (get_physmem_online_end()).
191 * 2. CONFIG_MAX_PHYSMEM_BITS - the maximum size of physical memory the
192 * kernel is able to support.
193 * 3. "mem=" kernel command line option which limits physical memory usage.
194 * 4. OLDMEM_BASE which is a kdump memory limit when the kernel is executed as
195 * crash kernel.
196 * 5. "hsa" size which is a memory limit when the kernel is executed during
197 * zfcp/nvme dump.
198 */
setup_ident_map_size(unsigned long max_physmem_end)199 static void setup_ident_map_size(unsigned long max_physmem_end)
200 {
201 unsigned long hsa_size;
202
203 ident_map_size = max_physmem_end;
204 if (memory_limit)
205 ident_map_size = min(ident_map_size, memory_limit);
206 ident_map_size = min(ident_map_size, 1UL << MAX_PHYSMEM_BITS);
207
208 #ifdef CONFIG_CRASH_DUMP
209 if (oldmem_data.start) {
210 __kaslr_enabled = 0;
211 ident_map_size = min(ident_map_size, oldmem_data.size);
212 } else if (ipl_block_valid && is_ipl_block_dump()) {
213 __kaslr_enabled = 0;
214 if (!sclp_early_get_hsa_size(&hsa_size) && hsa_size)
215 ident_map_size = min(ident_map_size, hsa_size);
216 }
217 #endif
218 }
219
220 #define FIXMAP_SIZE round_up(MEMCPY_REAL_SIZE + ABS_LOWCORE_MAP_SIZE, sizeof(struct lowcore))
221
get_vmem_size(unsigned long identity_size,unsigned long vmemmap_size,unsigned long vmalloc_size,unsigned long rte_size)222 static unsigned long get_vmem_size(unsigned long identity_size,
223 unsigned long vmemmap_size,
224 unsigned long vmalloc_size,
225 unsigned long rte_size)
226 {
227 unsigned long max_mappable, vsize;
228
229 max_mappable = max(identity_size, MAX_DCSS_ADDR);
230 vsize = round_up(SZ_2G + max_mappable, rte_size) +
231 round_up(vmemmap_size, rte_size) +
232 FIXMAP_SIZE + MODULES_LEN + KASLR_LEN;
233 return size_add(vsize, vmalloc_size);
234 }
235
setup_kernel_memory_layout(unsigned long kernel_size)236 static unsigned long setup_kernel_memory_layout(unsigned long kernel_size)
237 {
238 unsigned long vmemmap_start;
239 unsigned long kernel_start;
240 unsigned long asce_limit;
241 unsigned long rte_size;
242 unsigned long pages;
243 unsigned long vsize;
244 unsigned long vmax;
245
246 pages = ident_map_size / PAGE_SIZE;
247 /* vmemmap contains a multiple of PAGES_PER_SECTION struct pages */
248 vmemmap_size = SECTION_ALIGN_UP(pages) * sizeof(struct page);
249
250 /* choose kernel address space layout: 4 or 3 levels. */
251 BUILD_BUG_ON(!IS_ALIGNED(__START_KERNEL, THREAD_SIZE));
252 BUILD_BUG_ON(!IS_ALIGNED(__NO_KASLR_START_KERNEL, THREAD_SIZE));
253 BUILD_BUG_ON(__NO_KASLR_END_KERNEL > _REGION1_SIZE);
254 vsize = get_vmem_size(ident_map_size, vmemmap_size, vmalloc_size, _REGION3_SIZE);
255 if (IS_ENABLED(CONFIG_KASAN) || __NO_KASLR_END_KERNEL > _REGION2_SIZE ||
256 (vsize > _REGION2_SIZE && kaslr_enabled())) {
257 asce_limit = _REGION1_SIZE;
258 if (__NO_KASLR_END_KERNEL > _REGION2_SIZE) {
259 rte_size = _REGION2_SIZE;
260 vsize = get_vmem_size(ident_map_size, vmemmap_size, vmalloc_size, _REGION2_SIZE);
261 } else {
262 rte_size = _REGION3_SIZE;
263 }
264 } else {
265 asce_limit = _REGION2_SIZE;
266 rte_size = _REGION3_SIZE;
267 }
268
269 /*
270 * Forcing modules and vmalloc area under the ultravisor
271 * secure storage limit, so that any vmalloc allocation
272 * we do could be used to back secure guest storage.
273 *
274 * Assume the secure storage limit always exceeds _REGION2_SIZE,
275 * otherwise asce_limit and rte_size would have been adjusted.
276 */
277 vmax = adjust_to_uv_max(asce_limit);
278 #ifdef CONFIG_KASAN
279 BUILD_BUG_ON(__NO_KASLR_END_KERNEL > KASAN_SHADOW_START);
280 /* force vmalloc and modules below kasan shadow */
281 vmax = min(vmax, KASAN_SHADOW_START);
282 #endif
283 vsize = min(vsize, vmax);
284 if (kaslr_enabled()) {
285 unsigned long kernel_end, kaslr_len, slots, pos;
286
287 kaslr_len = max(KASLR_LEN, vmax - vsize);
288 slots = DIV_ROUND_UP(kaslr_len - kernel_size, THREAD_SIZE);
289 if (get_random(slots, &pos))
290 pos = 0;
291 kernel_end = vmax - pos * THREAD_SIZE;
292 kernel_start = round_down(kernel_end - kernel_size, THREAD_SIZE);
293 } else if (vmax < __NO_KASLR_END_KERNEL || vsize > __NO_KASLR_END_KERNEL) {
294 kernel_start = round_down(vmax - kernel_size, THREAD_SIZE);
295 decompressor_printk("The kernel base address is forced to %lx\n", kernel_start);
296 } else {
297 kernel_start = __NO_KASLR_START_KERNEL;
298 }
299 __kaslr_offset = kernel_start;
300
301 MODULES_END = round_down(kernel_start, _SEGMENT_SIZE);
302 MODULES_VADDR = MODULES_END - MODULES_LEN;
303 VMALLOC_END = MODULES_VADDR;
304
305 /* allow vmalloc area to occupy up to about 1/2 of the rest virtual space left */
306 vsize = (VMALLOC_END - FIXMAP_SIZE) / 2;
307 vsize = round_down(vsize, _SEGMENT_SIZE);
308 vmalloc_size = min(vmalloc_size, vsize);
309 VMALLOC_START = VMALLOC_END - vmalloc_size;
310
311 __memcpy_real_area = round_down(VMALLOC_START - MEMCPY_REAL_SIZE, PAGE_SIZE);
312 __abs_lowcore = round_down(__memcpy_real_area - ABS_LOWCORE_MAP_SIZE,
313 sizeof(struct lowcore));
314
315 /* split remaining virtual space between 1:1 mapping & vmemmap array */
316 pages = __abs_lowcore / (PAGE_SIZE + sizeof(struct page));
317 pages = SECTION_ALIGN_UP(pages);
318 /* keep vmemmap_start aligned to a top level region table entry */
319 vmemmap_start = round_down(__abs_lowcore - pages * sizeof(struct page), rte_size);
320 /* make sure identity map doesn't overlay with vmemmap */
321 ident_map_size = min(ident_map_size, vmemmap_start);
322 vmemmap_size = SECTION_ALIGN_UP(ident_map_size / PAGE_SIZE) * sizeof(struct page);
323 /* make sure vmemmap doesn't overlay with absolute lowcore area */
324 if (vmemmap_start + vmemmap_size > __abs_lowcore) {
325 vmemmap_size = SECTION_ALIGN_DOWN(ident_map_size / PAGE_SIZE) * sizeof(struct page);
326 ident_map_size = vmemmap_size / sizeof(struct page) * PAGE_SIZE;
327 }
328 vmemmap = (struct page *)vmemmap_start;
329 /* maximum address for which linear mapping could be created (DCSS, memory) */
330 BUILD_BUG_ON(MAX_DCSS_ADDR > (1UL << MAX_PHYSMEM_BITS));
331 max_mappable = max(ident_map_size, MAX_DCSS_ADDR);
332 max_mappable = min(max_mappable, vmemmap_start);
333 __identity_base = round_down(vmemmap_start - max_mappable, rte_size);
334
335 return asce_limit;
336 }
337
338 /*
339 * This function clears the BSS section of the decompressed Linux kernel and NOT the decompressor's.
340 */
clear_bss_section(unsigned long kernel_start)341 static void clear_bss_section(unsigned long kernel_start)
342 {
343 memset((void *)kernel_start + vmlinux.image_size, 0, vmlinux.bss_size);
344 }
345
346 /*
347 * Set vmalloc area size to an 8th of (potential) physical memory
348 * size, unless size has been set by kernel command line parameter.
349 */
setup_vmalloc_size(void)350 static void setup_vmalloc_size(void)
351 {
352 unsigned long size;
353
354 if (vmalloc_size_set)
355 return;
356 size = round_up(ident_map_size / 8, _SEGMENT_SIZE);
357 vmalloc_size = max(size, vmalloc_size);
358 }
359
kaslr_adjust_vmlinux_info(long offset)360 static void kaslr_adjust_vmlinux_info(long offset)
361 {
362 vmlinux.bootdata_off += offset;
363 vmlinux.bootdata_preserved_off += offset;
364 vmlinux.got_start += offset;
365 vmlinux.got_end += offset;
366 vmlinux.init_mm_off += offset;
367 vmlinux.swapper_pg_dir_off += offset;
368 vmlinux.invalid_pg_dir_off += offset;
369 #ifdef CONFIG_KASAN
370 vmlinux.kasan_early_shadow_page_off += offset;
371 vmlinux.kasan_early_shadow_pte_off += offset;
372 vmlinux.kasan_early_shadow_pmd_off += offset;
373 vmlinux.kasan_early_shadow_pud_off += offset;
374 vmlinux.kasan_early_shadow_p4d_off += offset;
375 #endif
376 }
377
fixup_vmlinux_info(void)378 static void fixup_vmlinux_info(void)
379 {
380 vmlinux.entry -= __START_KERNEL;
381 kaslr_adjust_vmlinux_info(-__START_KERNEL);
382 }
383
startup_kernel(void)384 void startup_kernel(void)
385 {
386 unsigned long kernel_size = vmlinux.image_size + vmlinux.bss_size;
387 unsigned long nokaslr_offset_phys, kaslr_large_page_offset;
388 unsigned long amode31_lma = 0;
389 unsigned long max_physmem_end;
390 unsigned long asce_limit;
391 unsigned long safe_addr;
392 psw_t psw;
393
394 fixup_vmlinux_info();
395 setup_lpp();
396
397 /*
398 * Non-randomized kernel physical start address must be _SEGMENT_SIZE
399 * aligned (see blow).
400 */
401 nokaslr_offset_phys = ALIGN(mem_safe_offset(), _SEGMENT_SIZE);
402 safe_addr = PAGE_ALIGN(nokaslr_offset_phys + kernel_size);
403
404 /*
405 * Reserve decompressor memory together with decompression heap,
406 * buffer and memory which might be occupied by uncompressed kernel
407 * (if KASLR is off or failed).
408 */
409 physmem_reserve(RR_DECOMPRESSOR, 0, safe_addr);
410 if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && parmarea.initrd_size)
411 physmem_reserve(RR_INITRD, parmarea.initrd_start, parmarea.initrd_size);
412 oldmem_data.start = parmarea.oldmem_base;
413 oldmem_data.size = parmarea.oldmem_size;
414
415 store_ipl_parmblock();
416 read_ipl_report();
417 uv_query_info();
418 sclp_early_read_info();
419 setup_boot_command_line();
420 parse_boot_command_line();
421 detect_facilities();
422 cmma_init();
423 sanitize_prot_virt_host();
424 max_physmem_end = detect_max_physmem_end();
425 setup_ident_map_size(max_physmem_end);
426 setup_vmalloc_size();
427 asce_limit = setup_kernel_memory_layout(kernel_size);
428 /* got final ident_map_size, physmem allocations could be performed now */
429 physmem_set_usable_limit(ident_map_size);
430 detect_physmem_online_ranges(max_physmem_end);
431 save_ipl_cert_comp_list();
432 rescue_initrd(safe_addr, ident_map_size);
433
434 /*
435 * __kaslr_offset_phys must be _SEGMENT_SIZE aligned, so the lower
436 * 20 bits (the offset within a large page) are zero. Copy the last
437 * 20 bits of __kaslr_offset, which is THREAD_SIZE aligned, to
438 * __kaslr_offset_phys.
439 *
440 * With this the last 20 bits of __kaslr_offset_phys and __kaslr_offset
441 * are identical, which is required to allow for large mappings of the
442 * kernel image.
443 */
444 kaslr_large_page_offset = __kaslr_offset & ~_SEGMENT_MASK;
445 if (kaslr_enabled()) {
446 unsigned long end = ident_map_size - kaslr_large_page_offset;
447
448 __kaslr_offset_phys = randomize_within_range(kernel_size, _SEGMENT_SIZE, 0, end);
449 }
450 if (!__kaslr_offset_phys)
451 __kaslr_offset_phys = nokaslr_offset_phys;
452 __kaslr_offset_phys |= kaslr_large_page_offset;
453 kaslr_adjust_vmlinux_info(__kaslr_offset_phys);
454 physmem_reserve(RR_VMLINUX, __kaslr_offset_phys, kernel_size);
455 deploy_kernel((void *)__kaslr_offset_phys);
456
457 /* vmlinux decompression is done, shrink reserved low memory */
458 physmem_reserve(RR_DECOMPRESSOR, 0, (unsigned long)_decompressor_end);
459
460 /*
461 * In case KASLR is enabled the randomized location of .amode31
462 * section might overlap with .vmlinux.relocs section. To avoid that
463 * the below randomize_within_range() could have been called with
464 * __vmlinux_relocs_64_end as the lower range address. However,
465 * .amode31 section is written to by the decompressed kernel - at
466 * that time the contents of .vmlinux.relocs is not needed anymore.
467 * Conversly, .vmlinux.relocs is read only by the decompressor, even
468 * before the kernel started. Therefore, in case the two sections
469 * overlap there is no risk of corrupting any data.
470 */
471 if (kaslr_enabled())
472 amode31_lma = randomize_within_range(vmlinux.amode31_size, PAGE_SIZE, 0, SZ_2G);
473 if (!amode31_lma)
474 amode31_lma = __kaslr_offset_phys - vmlinux.amode31_size;
475 physmem_reserve(RR_AMODE31, amode31_lma, vmlinux.amode31_size);
476
477 /*
478 * The order of the following operations is important:
479 *
480 * - kaslr_adjust_relocs() must follow clear_bss_section() to establish
481 * static memory references to data in .bss to be used by setup_vmem()
482 * (i.e init_mm.pgd)
483 *
484 * - setup_vmem() must follow kaslr_adjust_relocs() to be able using
485 * static memory references to data in .bss (i.e init_mm.pgd)
486 *
487 * - copy_bootdata() must follow setup_vmem() to propagate changes
488 * to bootdata made by setup_vmem()
489 */
490 clear_bss_section(__kaslr_offset_phys);
491 kaslr_adjust_relocs(__kaslr_offset_phys, __kaslr_offset_phys + vmlinux.image_size,
492 __kaslr_offset, __kaslr_offset_phys);
493 kaslr_adjust_got(__kaslr_offset);
494 setup_vmem(__kaslr_offset, __kaslr_offset + kernel_size, asce_limit);
495 copy_bootdata();
496
497 /*
498 * Save KASLR offset for early dumps, before vmcore_info is set.
499 * Mark as uneven to distinguish from real vmcore_info pointer.
500 */
501 S390_lowcore.vmcore_info = __kaslr_offset_phys ? __kaslr_offset_phys | 0x1UL : 0;
502
503 /*
504 * Jump to the decompressed kernel entry point and switch DAT mode on.
505 */
506 psw.addr = __kaslr_offset + vmlinux.entry;
507 psw.mask = PSW_KERNEL_BITS;
508 __load_psw(psw);
509 }
510