xref: /openbmc/linux/arch/arm64/kernel/setup.c (revision aa74c44b)
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/mm.h>
34 
35 #include <asm/acpi.h>
36 #include <asm/fixmap.h>
37 #include <asm/cpu.h>
38 #include <asm/cputype.h>
39 #include <asm/daifflags.h>
40 #include <asm/elf.h>
41 #include <asm/cpufeature.h>
42 #include <asm/cpu_ops.h>
43 #include <asm/kasan.h>
44 #include <asm/numa.h>
45 #include <asm/sections.h>
46 #include <asm/setup.h>
47 #include <asm/smp_plat.h>
48 #include <asm/cacheflush.h>
49 #include <asm/tlbflush.h>
50 #include <asm/traps.h>
51 #include <asm/efi.h>
52 #include <asm/xen/hypervisor.h>
53 #include <asm/mmu_context.h>
54 
55 static int num_standard_resources;
56 static struct resource *standard_resources;
57 
58 phys_addr_t __fdt_pointer __initdata;
59 
60 /*
61  * Standard memory resources
62  */
63 static struct resource mem_res[] = {
64 	{
65 		.name = "Kernel code",
66 		.start = 0,
67 		.end = 0,
68 		.flags = IORESOURCE_SYSTEM_RAM
69 	},
70 	{
71 		.name = "Kernel data",
72 		.start = 0,
73 		.end = 0,
74 		.flags = IORESOURCE_SYSTEM_RAM
75 	}
76 };
77 
78 #define kernel_code mem_res[0]
79 #define kernel_data mem_res[1]
80 
81 /*
82  * The recorded values of x0 .. x3 upon kernel entry.
83  */
84 u64 __cacheline_aligned boot_args[4];
85 
86 void __init smp_setup_processor_id(void)
87 {
88 	u64 mpidr = read_cpuid_mpidr() & MPIDR_HWID_BITMASK;
89 	set_cpu_logical_map(0, mpidr);
90 
91 	pr_info("Booting Linux on physical CPU 0x%010lx [0x%08x]\n",
92 		(unsigned long)mpidr, read_cpuid_id());
93 }
94 
95 bool arch_match_cpu_phys_id(int cpu, u64 phys_id)
96 {
97 	return phys_id == cpu_logical_map(cpu);
98 }
99 
100 struct mpidr_hash mpidr_hash;
101 /**
102  * smp_build_mpidr_hash - Pre-compute shifts required at each affinity
103  *			  level in order to build a linear index from an
104  *			  MPIDR value. Resulting algorithm is a collision
105  *			  free hash carried out through shifting and ORing
106  */
107 static void __init smp_build_mpidr_hash(void)
108 {
109 	u32 i, affinity, fs[4], bits[4], ls;
110 	u64 mask = 0;
111 	/*
112 	 * Pre-scan the list of MPIDRS and filter out bits that do
113 	 * not contribute to affinity levels, ie they never toggle.
114 	 */
115 	for_each_possible_cpu(i)
116 		mask |= (cpu_logical_map(i) ^ cpu_logical_map(0));
117 	pr_debug("mask of set bits %#llx\n", mask);
118 	/*
119 	 * Find and stash the last and first bit set at all affinity levels to
120 	 * check how many bits are required to represent them.
121 	 */
122 	for (i = 0; i < 4; i++) {
123 		affinity = MPIDR_AFFINITY_LEVEL(mask, i);
124 		/*
125 		 * Find the MSB bit and LSB bits position
126 		 * to determine how many bits are required
127 		 * to express the affinity level.
128 		 */
129 		ls = fls(affinity);
130 		fs[i] = affinity ? ffs(affinity) - 1 : 0;
131 		bits[i] = ls - fs[i];
132 	}
133 	/*
134 	 * An index can be created from the MPIDR_EL1 by isolating the
135 	 * significant bits at each affinity level and by shifting
136 	 * them in order to compress the 32 bits values space to a
137 	 * compressed set of values. This is equivalent to hashing
138 	 * the MPIDR_EL1 through shifting and ORing. It is a collision free
139 	 * hash though not minimal since some levels might contain a number
140 	 * of CPUs that is not an exact power of 2 and their bit
141 	 * representation might contain holes, eg MPIDR_EL1[7:0] = {0x2, 0x80}.
142 	 */
143 	mpidr_hash.shift_aff[0] = MPIDR_LEVEL_SHIFT(0) + fs[0];
144 	mpidr_hash.shift_aff[1] = MPIDR_LEVEL_SHIFT(1) + fs[1] - bits[0];
145 	mpidr_hash.shift_aff[2] = MPIDR_LEVEL_SHIFT(2) + fs[2] -
146 						(bits[1] + bits[0]);
147 	mpidr_hash.shift_aff[3] = MPIDR_LEVEL_SHIFT(3) +
148 				  fs[3] - (bits[2] + bits[1] + bits[0]);
149 	mpidr_hash.mask = mask;
150 	mpidr_hash.bits = bits[3] + bits[2] + bits[1] + bits[0];
151 	pr_debug("MPIDR hash: aff0[%u] aff1[%u] aff2[%u] aff3[%u] mask[%#llx] bits[%u]\n",
152 		mpidr_hash.shift_aff[0],
153 		mpidr_hash.shift_aff[1],
154 		mpidr_hash.shift_aff[2],
155 		mpidr_hash.shift_aff[3],
156 		mpidr_hash.mask,
157 		mpidr_hash.bits);
158 	/*
159 	 * 4x is an arbitrary value used to warn on a hash table much bigger
160 	 * than expected on most systems.
161 	 */
162 	if (mpidr_hash_size() > 4 * num_possible_cpus())
163 		pr_warn("Large number of MPIDR hash buckets detected\n");
164 }
165 
166 static void *early_fdt_ptr __initdata;
167 
168 void __init *get_early_fdt_ptr(void)
169 {
170 	return early_fdt_ptr;
171 }
172 
173 asmlinkage void __init early_fdt_map(u64 dt_phys)
174 {
175 	int fdt_size;
176 
177 	early_fixmap_init();
178 	early_fdt_ptr = fixmap_remap_fdt(dt_phys, &fdt_size, PAGE_KERNEL);
179 }
180 
181 static void __init setup_machine_fdt(phys_addr_t dt_phys)
182 {
183 	int size;
184 	void *dt_virt = fixmap_remap_fdt(dt_phys, &size, PAGE_KERNEL);
185 	const char *name;
186 
187 	if (dt_virt)
188 		memblock_reserve(dt_phys, size);
189 
190 	if (!dt_virt || !early_init_dt_scan(dt_virt)) {
191 		pr_crit("\n"
192 			"Error: invalid device tree blob at physical address %pa (virtual address 0x%px)\n"
193 			"The dtb must be 8-byte aligned and must not exceed 2 MB in size\n"
194 			"\nPlease check your bootloader.",
195 			&dt_phys, dt_virt);
196 
197 		/*
198 		 * Note that in this _really_ early stage we cannot even BUG()
199 		 * or oops, so the least terrible thing to do is cpu_relax(),
200 		 * or else we could end-up printing non-initialized data, etc.
201 		 */
202 		while (true)
203 			cpu_relax();
204 	}
205 
206 	/* Early fixups are done, map the FDT as read-only now */
207 	fixmap_remap_fdt(dt_phys, &size, PAGE_KERNEL_RO);
208 
209 	name = of_flat_dt_get_machine_name();
210 	if (!name)
211 		return;
212 
213 	pr_info("Machine model: %s\n", name);
214 	dump_stack_set_arch_desc("%s (DT)", name);
215 }
216 
217 static void __init request_standard_resources(void)
218 {
219 	struct memblock_region *region;
220 	struct resource *res;
221 	unsigned long i = 0;
222 	size_t res_size;
223 
224 	kernel_code.start   = __pa_symbol(_stext);
225 	kernel_code.end     = __pa_symbol(__init_begin - 1);
226 	kernel_data.start   = __pa_symbol(_sdata);
227 	kernel_data.end     = __pa_symbol(_end - 1);
228 
229 	num_standard_resources = memblock.memory.cnt;
230 	res_size = num_standard_resources * sizeof(*standard_resources);
231 	standard_resources = memblock_alloc(res_size, SMP_CACHE_BYTES);
232 	if (!standard_resources)
233 		panic("%s: Failed to allocate %zu bytes\n", __func__, res_size);
234 
235 	for_each_mem_region(region) {
236 		res = &standard_resources[i++];
237 		if (memblock_is_nomap(region)) {
238 			res->name  = "reserved";
239 			res->flags = IORESOURCE_MEM;
240 			res->start = __pfn_to_phys(memblock_region_reserved_base_pfn(region));
241 			res->end = __pfn_to_phys(memblock_region_reserved_end_pfn(region)) - 1;
242 		} else {
243 			res->name  = "System RAM";
244 			res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
245 			res->start = __pfn_to_phys(memblock_region_memory_base_pfn(region));
246 			res->end = __pfn_to_phys(memblock_region_memory_end_pfn(region)) - 1;
247 		}
248 
249 		request_resource(&iomem_resource, res);
250 
251 		if (kernel_code.start >= res->start &&
252 		    kernel_code.end <= res->end)
253 			request_resource(res, &kernel_code);
254 		if (kernel_data.start >= res->start &&
255 		    kernel_data.end <= res->end)
256 			request_resource(res, &kernel_data);
257 #ifdef CONFIG_KEXEC_CORE
258 		/* Userspace will find "Crash kernel" region in /proc/iomem. */
259 		if (crashk_res.end && crashk_res.start >= res->start &&
260 		    crashk_res.end <= res->end)
261 			request_resource(res, &crashk_res);
262 #endif
263 	}
264 }
265 
266 static int __init reserve_memblock_reserved_regions(void)
267 {
268 	u64 i, j;
269 
270 	for (i = 0; i < num_standard_resources; ++i) {
271 		struct resource *mem = &standard_resources[i];
272 		phys_addr_t r_start, r_end, mem_size = resource_size(mem);
273 
274 		if (!memblock_is_region_reserved(mem->start, mem_size))
275 			continue;
276 
277 		for_each_reserved_mem_range(j, &r_start, &r_end) {
278 			resource_size_t start, end;
279 
280 			start = max(PFN_PHYS(PFN_DOWN(r_start)), mem->start);
281 			end = min(PFN_PHYS(PFN_UP(r_end)) - 1, mem->end);
282 
283 			if (start > mem->end || end < mem->start)
284 				continue;
285 
286 			reserve_region_with_split(mem, start, end, "reserved");
287 		}
288 	}
289 
290 	return 0;
291 }
292 arch_initcall(reserve_memblock_reserved_regions);
293 
294 u64 __cpu_logical_map[NR_CPUS] = { [0 ... NR_CPUS-1] = INVALID_HWID };
295 
296 u64 cpu_logical_map(unsigned int cpu)
297 {
298 	return __cpu_logical_map[cpu];
299 }
300 
301 void __init __no_sanitize_address setup_arch(char **cmdline_p)
302 {
303 	setup_initial_init_mm(_stext, _etext, _edata, _end);
304 
305 	*cmdline_p = boot_command_line;
306 
307 	/*
308 	 * If know now we are going to need KPTI then use non-global
309 	 * mappings from the start, avoiding the cost of rewriting
310 	 * everything later.
311 	 */
312 	arm64_use_ng_mappings = kaslr_requires_kpti();
313 
314 	early_fixmap_init();
315 	early_ioremap_init();
316 
317 	setup_machine_fdt(__fdt_pointer);
318 
319 	/*
320 	 * Initialise the static keys early as they may be enabled by the
321 	 * cpufeature code and early parameters.
322 	 */
323 	jump_label_init();
324 	parse_early_param();
325 
326 	/*
327 	 * Unmask asynchronous aborts and fiq after bringing up possible
328 	 * earlycon. (Report possible System Errors once we can report this
329 	 * occurred).
330 	 */
331 	local_daif_restore(DAIF_PROCCTX_NOIRQ);
332 
333 	/*
334 	 * TTBR0 is only used for the identity mapping at this stage. Make it
335 	 * point to zero page to avoid speculatively fetching new entries.
336 	 */
337 	cpu_uninstall_idmap();
338 
339 	xen_early_init();
340 	efi_init();
341 
342 	if (!efi_enabled(EFI_BOOT) && ((u64)_text % MIN_KIMG_ALIGN) != 0)
343 	     pr_warn(FW_BUG "Kernel image misaligned at boot, please fix your bootloader!");
344 
345 	arm64_memblock_init();
346 
347 	paging_init();
348 
349 	acpi_table_upgrade();
350 
351 	/* Parse the ACPI tables for possible boot-time configuration */
352 	acpi_boot_table_init();
353 
354 	if (acpi_disabled)
355 		unflatten_device_tree();
356 
357 	bootmem_init();
358 
359 	kasan_init();
360 
361 	request_standard_resources();
362 
363 	early_ioremap_reset();
364 
365 	if (acpi_disabled)
366 		psci_dt_init();
367 	else
368 		psci_acpi_init();
369 
370 	init_bootcpu_ops();
371 	smp_init_cpus();
372 	smp_build_mpidr_hash();
373 
374 	/* Init percpu seeds for random tags after cpus are set up. */
375 	kasan_init_sw_tags();
376 
377 #ifdef CONFIG_ARM64_SW_TTBR0_PAN
378 	/*
379 	 * Make sure init_thread_info.ttbr0 always generates translation
380 	 * faults in case uaccess_enable() is inadvertently called by the init
381 	 * thread.
382 	 */
383 	init_task.thread_info.ttbr0 = phys_to_ttbr(__pa_symbol(reserved_pg_dir));
384 #endif
385 
386 	if (boot_args[1] || boot_args[2] || boot_args[3]) {
387 		pr_err("WARNING: x1-x3 nonzero in violation of boot protocol:\n"
388 			"\tx1: %016llx\n\tx2: %016llx\n\tx3: %016llx\n"
389 			"This indicates a broken bootloader or old kernel\n",
390 			boot_args[1], boot_args[2], boot_args[3]);
391 	}
392 }
393 
394 static inline bool cpu_can_disable(unsigned int cpu)
395 {
396 #ifdef CONFIG_HOTPLUG_CPU
397 	const struct cpu_operations *ops = get_cpu_ops(cpu);
398 
399 	if (ops && ops->cpu_can_disable)
400 		return ops->cpu_can_disable(cpu);
401 #endif
402 	return false;
403 }
404 
405 static int __init topology_init(void)
406 {
407 	int i;
408 
409 	for_each_online_node(i)
410 		register_one_node(i);
411 
412 	for_each_possible_cpu(i) {
413 		struct cpu *cpu = &per_cpu(cpu_data.cpu, i);
414 		cpu->hotpluggable = cpu_can_disable(i);
415 		register_cpu(cpu, i);
416 	}
417 
418 	return 0;
419 }
420 subsys_initcall(topology_init);
421 
422 static void dump_kernel_offset(void)
423 {
424 	const unsigned long offset = kaslr_offset();
425 
426 	if (IS_ENABLED(CONFIG_RANDOMIZE_BASE) && offset > 0) {
427 		pr_emerg("Kernel Offset: 0x%lx from 0x%lx\n",
428 			 offset, KIMAGE_VADDR);
429 		pr_emerg("PHYS_OFFSET: 0x%llx\n", PHYS_OFFSET);
430 	} else {
431 		pr_emerg("Kernel Offset: disabled\n");
432 	}
433 }
434 
435 static int arm64_panic_block_dump(struct notifier_block *self,
436 				  unsigned long v, void *p)
437 {
438 	dump_kernel_offset();
439 	dump_cpu_features();
440 	dump_mem_limit();
441 	return 0;
442 }
443 
444 static struct notifier_block arm64_panic_block = {
445 	.notifier_call = arm64_panic_block_dump
446 };
447 
448 static int __init register_arm64_panic_block(void)
449 {
450 	atomic_notifier_chain_register(&panic_notifier_list,
451 				       &arm64_panic_block);
452 	return 0;
453 }
454 device_initcall(register_arm64_panic_block);
455