xref: /linux/arch/arm64/kernel/setup.c (revision 1a75e81baf4f1b322f3498ffd373eaada8e60589)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Based on arch/arm/kernel/setup.c
4  *
5  * Copyright (C) 1995-2001 Russell King
6  * Copyright (C) 2012 ARM Ltd.
7  */
8 
9 #include <linux/acpi.h>
10 #include <linux/export.h>
11 #include <linux/kernel.h>
12 #include <linux/stddef.h>
13 #include <linux/ioport.h>
14 #include <linux/delay.h>
15 #include <linux/initrd.h>
16 #include <linux/console.h>
17 #include <linux/cache.h>
18 #include <linux/screen_info.h>
19 #include <linux/init.h>
20 #include <linux/kexec.h>
21 #include <linux/root_dev.h>
22 #include <linux/cpu.h>
23 #include <linux/interrupt.h>
24 #include <linux/smp.h>
25 #include <linux/fs.h>
26 #include <linux/panic_notifier.h>
27 #include <linux/proc_fs.h>
28 #include <linux/memblock.h>
29 #include <linux/of_fdt.h>
30 #include <linux/efi.h>
31 #include <linux/psci.h>
32 #include <linux/sched/task.h>
33 #include <linux/scs.h>
34 #include <linux/mm.h>
35 
36 #include <asm/acpi.h>
37 #include <asm/fixmap.h>
38 #include <asm/cpu.h>
39 #include <asm/cputype.h>
40 #include <asm/daifflags.h>
41 #include <asm/elf.h>
42 #include <asm/cpufeature.h>
43 #include <asm/cpu_ops.h>
44 #include <asm/kasan.h>
45 #include <asm/numa.h>
46 #include <asm/scs.h>
47 #include <asm/sections.h>
48 #include <asm/setup.h>
49 #include <asm/smp_plat.h>
50 #include <asm/cacheflush.h>
51 #include <asm/tlbflush.h>
52 #include <asm/traps.h>
53 #include <asm/efi.h>
54 #include <asm/xen/hypervisor.h>
55 #include <asm/mmu_context.h>
56 
57 static int num_standard_resources;
58 static struct resource *standard_resources;
59 
60 phys_addr_t __fdt_pointer __initdata;
61 u64 mmu_enabled_at_boot __initdata;
62 
63 /*
64  * Standard memory resources
65  */
66 static struct resource mem_res[] = {
67 	{
68 		.name = "Kernel code",
69 		.start = 0,
70 		.end = 0,
71 		.flags = IORESOURCE_SYSTEM_RAM
72 	},
73 	{
74 		.name = "Kernel data",
75 		.start = 0,
76 		.end = 0,
77 		.flags = IORESOURCE_SYSTEM_RAM
78 	}
79 };
80 
81 #define kernel_code mem_res[0]
82 #define kernel_data mem_res[1]
83 
84 /*
85  * The recorded values of x0 .. x3 upon kernel entry.
86  */
87 u64 __cacheline_aligned boot_args[4];
88 
89 void __init smp_setup_processor_id(void)
90 {
91 	u64 mpidr = read_cpuid_mpidr() & MPIDR_HWID_BITMASK;
92 	set_cpu_logical_map(0, mpidr);
93 
94 	pr_info("Booting Linux on physical CPU 0x%010lx [0x%08x]\n",
95 		(unsigned long)mpidr, read_cpuid_id());
96 }
97 
98 bool arch_match_cpu_phys_id(int cpu, u64 phys_id)
99 {
100 	return phys_id == cpu_logical_map(cpu);
101 }
102 
103 struct mpidr_hash mpidr_hash;
104 /**
105  * smp_build_mpidr_hash - Pre-compute shifts required at each affinity
106  *			  level in order to build a linear index from an
107  *			  MPIDR value. Resulting algorithm is a collision
108  *			  free hash carried out through shifting and ORing
109  */
110 static void __init smp_build_mpidr_hash(void)
111 {
112 	u32 i, affinity, fs[4], bits[4], ls;
113 	u64 mask = 0;
114 	/*
115 	 * Pre-scan the list of MPIDRS and filter out bits that do
116 	 * not contribute to affinity levels, ie they never toggle.
117 	 */
118 	for_each_possible_cpu(i)
119 		mask |= (cpu_logical_map(i) ^ cpu_logical_map(0));
120 	pr_debug("mask of set bits %#llx\n", mask);
121 	/*
122 	 * Find and stash the last and first bit set at all affinity levels to
123 	 * check how many bits are required to represent them.
124 	 */
125 	for (i = 0; i < 4; i++) {
126 		affinity = MPIDR_AFFINITY_LEVEL(mask, i);
127 		/*
128 		 * Find the MSB bit and LSB bits position
129 		 * to determine how many bits are required
130 		 * to express the affinity level.
131 		 */
132 		ls = fls(affinity);
133 		fs[i] = affinity ? ffs(affinity) - 1 : 0;
134 		bits[i] = ls - fs[i];
135 	}
136 	/*
137 	 * An index can be created from the MPIDR_EL1 by isolating the
138 	 * significant bits at each affinity level and by shifting
139 	 * them in order to compress the 32 bits values space to a
140 	 * compressed set of values. This is equivalent to hashing
141 	 * the MPIDR_EL1 through shifting and ORing. It is a collision free
142 	 * hash though not minimal since some levels might contain a number
143 	 * of CPUs that is not an exact power of 2 and their bit
144 	 * representation might contain holes, eg MPIDR_EL1[7:0] = {0x2, 0x80}.
145 	 */
146 	mpidr_hash.shift_aff[0] = MPIDR_LEVEL_SHIFT(0) + fs[0];
147 	mpidr_hash.shift_aff[1] = MPIDR_LEVEL_SHIFT(1) + fs[1] - bits[0];
148 	mpidr_hash.shift_aff[2] = MPIDR_LEVEL_SHIFT(2) + fs[2] -
149 						(bits[1] + bits[0]);
150 	mpidr_hash.shift_aff[3] = MPIDR_LEVEL_SHIFT(3) +
151 				  fs[3] - (bits[2] + bits[1] + bits[0]);
152 	mpidr_hash.mask = mask;
153 	mpidr_hash.bits = bits[3] + bits[2] + bits[1] + bits[0];
154 	pr_debug("MPIDR hash: aff0[%u] aff1[%u] aff2[%u] aff3[%u] mask[%#llx] bits[%u]\n",
155 		mpidr_hash.shift_aff[0],
156 		mpidr_hash.shift_aff[1],
157 		mpidr_hash.shift_aff[2],
158 		mpidr_hash.shift_aff[3],
159 		mpidr_hash.mask,
160 		mpidr_hash.bits);
161 	/*
162 	 * 4x is an arbitrary value used to warn on a hash table much bigger
163 	 * than expected on most systems.
164 	 */
165 	if (mpidr_hash_size() > 4 * num_possible_cpus())
166 		pr_warn("Large number of MPIDR hash buckets detected\n");
167 }
168 
169 static void __init setup_machine_fdt(phys_addr_t dt_phys)
170 {
171 	int size;
172 	void *dt_virt = fixmap_remap_fdt(dt_phys, &size, PAGE_KERNEL);
173 	const char *name;
174 
175 	if (dt_virt)
176 		memblock_reserve(dt_phys, size);
177 
178 	/*
179 	 * dt_virt is a fixmap address, hence __pa(dt_virt) can't be used.
180 	 * Pass dt_phys directly.
181 	 */
182 	if (!early_init_dt_scan(dt_virt, dt_phys)) {
183 		pr_crit("\n"
184 			"Error: invalid device tree blob at physical address %pa (virtual address 0x%px)\n"
185 			"The dtb must be 8-byte aligned and must not exceed 2 MB in size\n"
186 			"\nPlease check your bootloader.",
187 			&dt_phys, dt_virt);
188 
189 		/*
190 		 * Note that in this _really_ early stage we cannot even BUG()
191 		 * or oops, so the least terrible thing to do is cpu_relax(),
192 		 * or else we could end-up printing non-initialized data, etc.
193 		 */
194 		while (true)
195 			cpu_relax();
196 	}
197 
198 	/* Early fixups are done, map the FDT as read-only now */
199 	fixmap_remap_fdt(dt_phys, &size, PAGE_KERNEL_RO);
200 
201 	name = of_flat_dt_get_machine_name();
202 	if (!name)
203 		return;
204 
205 	pr_info("Machine model: %s\n", name);
206 	dump_stack_set_arch_desc("%s (DT)", name);
207 }
208 
209 static void __init request_standard_resources(void)
210 {
211 	struct memblock_region *region;
212 	struct resource *res;
213 	unsigned long i = 0;
214 	size_t res_size;
215 
216 	kernel_code.start   = __pa_symbol(_stext);
217 	kernel_code.end     = __pa_symbol(__init_begin - 1);
218 	kernel_data.start   = __pa_symbol(_sdata);
219 	kernel_data.end     = __pa_symbol(_end - 1);
220 	insert_resource(&iomem_resource, &kernel_code);
221 	insert_resource(&iomem_resource, &kernel_data);
222 
223 	num_standard_resources = memblock.memory.cnt;
224 	res_size = num_standard_resources * sizeof(*standard_resources);
225 	standard_resources = memblock_alloc(res_size, SMP_CACHE_BYTES);
226 	if (!standard_resources)
227 		panic("%s: Failed to allocate %zu bytes\n", __func__, res_size);
228 
229 	for_each_mem_region(region) {
230 		res = &standard_resources[i++];
231 		if (memblock_is_nomap(region)) {
232 			res->name  = "reserved";
233 			res->flags = IORESOURCE_MEM;
234 			res->start = __pfn_to_phys(memblock_region_reserved_base_pfn(region));
235 			res->end = __pfn_to_phys(memblock_region_reserved_end_pfn(region)) - 1;
236 		} else {
237 			res->name  = "System RAM";
238 			res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
239 			res->start = __pfn_to_phys(memblock_region_memory_base_pfn(region));
240 			res->end = __pfn_to_phys(memblock_region_memory_end_pfn(region)) - 1;
241 		}
242 
243 		insert_resource(&iomem_resource, res);
244 	}
245 }
246 
247 static int __init reserve_memblock_reserved_regions(void)
248 {
249 	u64 i, j;
250 
251 	for (i = 0; i < num_standard_resources; ++i) {
252 		struct resource *mem = &standard_resources[i];
253 		phys_addr_t r_start, r_end, mem_size = resource_size(mem);
254 
255 		if (!memblock_is_region_reserved(mem->start, mem_size))
256 			continue;
257 
258 		for_each_reserved_mem_range(j, &r_start, &r_end) {
259 			resource_size_t start, end;
260 
261 			start = max(PFN_PHYS(PFN_DOWN(r_start)), mem->start);
262 			end = min(PFN_PHYS(PFN_UP(r_end)) - 1, mem->end);
263 
264 			if (start > mem->end || end < mem->start)
265 				continue;
266 
267 			reserve_region_with_split(mem, start, end, "reserved");
268 		}
269 	}
270 
271 	return 0;
272 }
273 arch_initcall(reserve_memblock_reserved_regions);
274 
275 u64 __cpu_logical_map[NR_CPUS] = { [0 ... NR_CPUS-1] = INVALID_HWID };
276 
277 u64 cpu_logical_map(unsigned int cpu)
278 {
279 	return __cpu_logical_map[cpu];
280 }
281 
282 void __init __no_sanitize_address setup_arch(char **cmdline_p)
283 {
284 	setup_initial_init_mm(_stext, _etext, _edata, _end);
285 
286 	*cmdline_p = boot_command_line;
287 
288 	kaslr_init();
289 
290 	early_fixmap_init();
291 	early_ioremap_init();
292 
293 	setup_machine_fdt(__fdt_pointer);
294 
295 	/*
296 	 * Initialise the static keys early as they may be enabled by the
297 	 * cpufeature code and early parameters.
298 	 */
299 	jump_label_init();
300 	parse_early_param();
301 
302 	dynamic_scs_init();
303 
304 	/*
305 	 * The primary CPU enters the kernel with all DAIF exceptions masked.
306 	 *
307 	 * We must unmask Debug and SError before preemption or scheduling is
308 	 * possible to ensure that these are consistently unmasked across
309 	 * threads, and we want to unmask SError as soon as possible after
310 	 * initializing earlycon so that we can report any SErrors immediately.
311 	 *
312 	 * IRQ and FIQ will be unmasked after the root irqchip has been
313 	 * detected and initialized.
314 	 */
315 	local_daif_restore(DAIF_PROCCTX_NOIRQ);
316 
317 	/*
318 	 * TTBR0 is only used for the identity mapping at this stage. Make it
319 	 * point to zero page to avoid speculatively fetching new entries.
320 	 */
321 	cpu_uninstall_idmap();
322 
323 	xen_early_init();
324 	efi_init();
325 
326 	if (!efi_enabled(EFI_BOOT)) {
327 		if ((u64)_text % MIN_KIMG_ALIGN)
328 			pr_warn(FW_BUG "Kernel image misaligned at boot, please fix your bootloader!");
329 		WARN_TAINT(mmu_enabled_at_boot, TAINT_FIRMWARE_WORKAROUND,
330 			   FW_BUG "Booted with MMU enabled!");
331 	}
332 
333 	arm64_memblock_init();
334 
335 	paging_init();
336 
337 	acpi_table_upgrade();
338 
339 	/* Parse the ACPI tables for possible boot-time configuration */
340 	acpi_boot_table_init();
341 
342 	if (acpi_disabled)
343 		unflatten_device_tree();
344 
345 	bootmem_init();
346 
347 	kasan_init();
348 
349 	request_standard_resources();
350 
351 	early_ioremap_reset();
352 
353 	if (acpi_disabled)
354 		psci_dt_init();
355 	else
356 		psci_acpi_init();
357 
358 	init_bootcpu_ops();
359 	smp_init_cpus();
360 	smp_build_mpidr_hash();
361 
362 #ifdef CONFIG_ARM64_SW_TTBR0_PAN
363 	/*
364 	 * Make sure init_thread_info.ttbr0 always generates translation
365 	 * faults in case uaccess_enable() is inadvertently called by the init
366 	 * thread.
367 	 */
368 	init_task.thread_info.ttbr0 = phys_to_ttbr(__pa_symbol(reserved_pg_dir));
369 #endif
370 
371 	if (boot_args[1] || boot_args[2] || boot_args[3]) {
372 		pr_err("WARNING: x1-x3 nonzero in violation of boot protocol:\n"
373 			"\tx1: %016llx\n\tx2: %016llx\n\tx3: %016llx\n"
374 			"This indicates a broken bootloader or old kernel\n",
375 			boot_args[1], boot_args[2], boot_args[3]);
376 	}
377 }
378 
379 static inline bool cpu_can_disable(unsigned int cpu)
380 {
381 #ifdef CONFIG_HOTPLUG_CPU
382 	const struct cpu_operations *ops = get_cpu_ops(cpu);
383 
384 	if (ops && ops->cpu_can_disable)
385 		return ops->cpu_can_disable(cpu);
386 #endif
387 	return false;
388 }
389 
390 bool arch_cpu_is_hotpluggable(int num)
391 {
392 	return cpu_can_disable(num);
393 }
394 
395 static void dump_kernel_offset(void)
396 {
397 	const unsigned long offset = kaslr_offset();
398 
399 	if (IS_ENABLED(CONFIG_RANDOMIZE_BASE) && offset > 0) {
400 		pr_emerg("Kernel Offset: 0x%lx from 0x%lx\n",
401 			 offset, KIMAGE_VADDR);
402 		pr_emerg("PHYS_OFFSET: 0x%llx\n", PHYS_OFFSET);
403 	} else {
404 		pr_emerg("Kernel Offset: disabled\n");
405 	}
406 }
407 
408 static int arm64_panic_block_dump(struct notifier_block *self,
409 				  unsigned long v, void *p)
410 {
411 	dump_kernel_offset();
412 	dump_cpu_features();
413 	dump_mem_limit();
414 	return 0;
415 }
416 
417 static struct notifier_block arm64_panic_block = {
418 	.notifier_call = arm64_panic_block_dump
419 };
420 
421 static int __init register_arm64_panic_block(void)
422 {
423 	atomic_notifier_chain_register(&panic_notifier_list,
424 				       &arm64_panic_block);
425 	return 0;
426 }
427 device_initcall(register_arm64_panic_block);
428 
429 static int __init check_mmu_enabled_at_boot(void)
430 {
431 	if (!efi_enabled(EFI_BOOT) && mmu_enabled_at_boot)
432 		panic("Non-EFI boot detected with MMU and caches enabled");
433 	return 0;
434 }
435 device_initcall_sync(check_mmu_enabled_at_boot);
436