xref: /openbmc/linux/arch/arm64/kernel/setup.c (revision a266ef69)
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 
62 /*
63  * Standard memory resources
64  */
65 static struct resource mem_res[] = {
66 	{
67 		.name = "Kernel code",
68 		.start = 0,
69 		.end = 0,
70 		.flags = IORESOURCE_SYSTEM_RAM
71 	},
72 	{
73 		.name = "Kernel data",
74 		.start = 0,
75 		.end = 0,
76 		.flags = IORESOURCE_SYSTEM_RAM
77 	}
78 };
79 
80 #define kernel_code mem_res[0]
81 #define kernel_data mem_res[1]
82 
83 /*
84  * The recorded values of x0 .. x3 upon kernel entry.
85  */
86 u64 __cacheline_aligned boot_args[4];
87 
88 void __init smp_setup_processor_id(void)
89 {
90 	u64 mpidr = read_cpuid_mpidr() & MPIDR_HWID_BITMASK;
91 	set_cpu_logical_map(0, mpidr);
92 
93 	pr_info("Booting Linux on physical CPU 0x%010lx [0x%08x]\n",
94 		(unsigned long)mpidr, read_cpuid_id());
95 }
96 
97 bool arch_match_cpu_phys_id(int cpu, u64 phys_id)
98 {
99 	return phys_id == cpu_logical_map(cpu);
100 }
101 
102 struct mpidr_hash mpidr_hash;
103 /**
104  * smp_build_mpidr_hash - Pre-compute shifts required at each affinity
105  *			  level in order to build a linear index from an
106  *			  MPIDR value. Resulting algorithm is a collision
107  *			  free hash carried out through shifting and ORing
108  */
109 static void __init smp_build_mpidr_hash(void)
110 {
111 	u32 i, affinity, fs[4], bits[4], ls;
112 	u64 mask = 0;
113 	/*
114 	 * Pre-scan the list of MPIDRS and filter out bits that do
115 	 * not contribute to affinity levels, ie they never toggle.
116 	 */
117 	for_each_possible_cpu(i)
118 		mask |= (cpu_logical_map(i) ^ cpu_logical_map(0));
119 	pr_debug("mask of set bits %#llx\n", mask);
120 	/*
121 	 * Find and stash the last and first bit set at all affinity levels to
122 	 * check how many bits are required to represent them.
123 	 */
124 	for (i = 0; i < 4; i++) {
125 		affinity = MPIDR_AFFINITY_LEVEL(mask, i);
126 		/*
127 		 * Find the MSB bit and LSB bits position
128 		 * to determine how many bits are required
129 		 * to express the affinity level.
130 		 */
131 		ls = fls(affinity);
132 		fs[i] = affinity ? ffs(affinity) - 1 : 0;
133 		bits[i] = ls - fs[i];
134 	}
135 	/*
136 	 * An index can be created from the MPIDR_EL1 by isolating the
137 	 * significant bits at each affinity level and by shifting
138 	 * them in order to compress the 32 bits values space to a
139 	 * compressed set of values. This is equivalent to hashing
140 	 * the MPIDR_EL1 through shifting and ORing. It is a collision free
141 	 * hash though not minimal since some levels might contain a number
142 	 * of CPUs that is not an exact power of 2 and their bit
143 	 * representation might contain holes, eg MPIDR_EL1[7:0] = {0x2, 0x80}.
144 	 */
145 	mpidr_hash.shift_aff[0] = MPIDR_LEVEL_SHIFT(0) + fs[0];
146 	mpidr_hash.shift_aff[1] = MPIDR_LEVEL_SHIFT(1) + fs[1] - bits[0];
147 	mpidr_hash.shift_aff[2] = MPIDR_LEVEL_SHIFT(2) + fs[2] -
148 						(bits[1] + bits[0]);
149 	mpidr_hash.shift_aff[3] = MPIDR_LEVEL_SHIFT(3) +
150 				  fs[3] - (bits[2] + bits[1] + bits[0]);
151 	mpidr_hash.mask = mask;
152 	mpidr_hash.bits = bits[3] + bits[2] + bits[1] + bits[0];
153 	pr_debug("MPIDR hash: aff0[%u] aff1[%u] aff2[%u] aff3[%u] mask[%#llx] bits[%u]\n",
154 		mpidr_hash.shift_aff[0],
155 		mpidr_hash.shift_aff[1],
156 		mpidr_hash.shift_aff[2],
157 		mpidr_hash.shift_aff[3],
158 		mpidr_hash.mask,
159 		mpidr_hash.bits);
160 	/*
161 	 * 4x is an arbitrary value used to warn on a hash table much bigger
162 	 * than expected on most systems.
163 	 */
164 	if (mpidr_hash_size() > 4 * num_possible_cpus())
165 		pr_warn("Large number of MPIDR hash buckets detected\n");
166 }
167 
168 static void *early_fdt_ptr __initdata;
169 
170 void __init *get_early_fdt_ptr(void)
171 {
172 	return early_fdt_ptr;
173 }
174 
175 asmlinkage void __init early_fdt_map(u64 dt_phys)
176 {
177 	int fdt_size;
178 
179 	early_fixmap_init();
180 	early_fdt_ptr = fixmap_remap_fdt(dt_phys, &fdt_size, PAGE_KERNEL);
181 }
182 
183 static void __init setup_machine_fdt(phys_addr_t dt_phys)
184 {
185 	int size;
186 	void *dt_virt = fixmap_remap_fdt(dt_phys, &size, PAGE_KERNEL);
187 	const char *name;
188 
189 	if (dt_virt)
190 		memblock_reserve(dt_phys, size);
191 
192 	if (!dt_virt || !early_init_dt_scan(dt_virt)) {
193 		pr_crit("\n"
194 			"Error: invalid device tree blob at physical address %pa (virtual address 0x%px)\n"
195 			"The dtb must be 8-byte aligned and must not exceed 2 MB in size\n"
196 			"\nPlease check your bootloader.",
197 			&dt_phys, dt_virt);
198 
199 		/*
200 		 * Note that in this _really_ early stage we cannot even BUG()
201 		 * or oops, so the least terrible thing to do is cpu_relax(),
202 		 * or else we could end-up printing non-initialized data, etc.
203 		 */
204 		while (true)
205 			cpu_relax();
206 	}
207 
208 	/* Early fixups are done, map the FDT as read-only now */
209 	fixmap_remap_fdt(dt_phys, &size, PAGE_KERNEL_RO);
210 
211 	name = of_flat_dt_get_machine_name();
212 	if (!name)
213 		return;
214 
215 	pr_info("Machine model: %s\n", name);
216 	dump_stack_set_arch_desc("%s (DT)", name);
217 }
218 
219 static void __init request_standard_resources(void)
220 {
221 	struct memblock_region *region;
222 	struct resource *res;
223 	unsigned long i = 0;
224 	size_t res_size;
225 
226 	kernel_code.start   = __pa_symbol(_stext);
227 	kernel_code.end     = __pa_symbol(__init_begin - 1);
228 	kernel_data.start   = __pa_symbol(_sdata);
229 	kernel_data.end     = __pa_symbol(_end - 1);
230 	insert_resource(&iomem_resource, &kernel_code);
231 	insert_resource(&iomem_resource, &kernel_data);
232 
233 	num_standard_resources = memblock.memory.cnt;
234 	res_size = num_standard_resources * sizeof(*standard_resources);
235 	standard_resources = memblock_alloc(res_size, SMP_CACHE_BYTES);
236 	if (!standard_resources)
237 		panic("%s: Failed to allocate %zu bytes\n", __func__, res_size);
238 
239 	for_each_mem_region(region) {
240 		res = &standard_resources[i++];
241 		if (memblock_is_nomap(region)) {
242 			res->name  = "reserved";
243 			res->flags = IORESOURCE_MEM;
244 			res->start = __pfn_to_phys(memblock_region_reserved_base_pfn(region));
245 			res->end = __pfn_to_phys(memblock_region_reserved_end_pfn(region)) - 1;
246 		} else {
247 			res->name  = "System RAM";
248 			res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
249 			res->start = __pfn_to_phys(memblock_region_memory_base_pfn(region));
250 			res->end = __pfn_to_phys(memblock_region_memory_end_pfn(region)) - 1;
251 		}
252 
253 		insert_resource(&iomem_resource, res);
254 	}
255 }
256 
257 static int __init reserve_memblock_reserved_regions(void)
258 {
259 	u64 i, j;
260 
261 	for (i = 0; i < num_standard_resources; ++i) {
262 		struct resource *mem = &standard_resources[i];
263 		phys_addr_t r_start, r_end, mem_size = resource_size(mem);
264 
265 		if (!memblock_is_region_reserved(mem->start, mem_size))
266 			continue;
267 
268 		for_each_reserved_mem_range(j, &r_start, &r_end) {
269 			resource_size_t start, end;
270 
271 			start = max(PFN_PHYS(PFN_DOWN(r_start)), mem->start);
272 			end = min(PFN_PHYS(PFN_UP(r_end)) - 1, mem->end);
273 
274 			if (start > mem->end || end < mem->start)
275 				continue;
276 
277 			reserve_region_with_split(mem, start, end, "reserved");
278 		}
279 	}
280 
281 	return 0;
282 }
283 arch_initcall(reserve_memblock_reserved_regions);
284 
285 u64 __cpu_logical_map[NR_CPUS] = { [0 ... NR_CPUS-1] = INVALID_HWID };
286 
287 u64 cpu_logical_map(unsigned int cpu)
288 {
289 	return __cpu_logical_map[cpu];
290 }
291 
292 void __init __no_sanitize_address setup_arch(char **cmdline_p)
293 {
294 	setup_initial_init_mm(_stext, _etext, _edata, _end);
295 
296 	*cmdline_p = boot_command_line;
297 
298 	/*
299 	 * If know now we are going to need KPTI then use non-global
300 	 * mappings from the start, avoiding the cost of rewriting
301 	 * everything later.
302 	 */
303 	arm64_use_ng_mappings = kaslr_requires_kpti();
304 
305 	early_fixmap_init();
306 	early_ioremap_init();
307 
308 	setup_machine_fdt(__fdt_pointer);
309 
310 	/*
311 	 * Initialise the static keys early as they may be enabled by the
312 	 * cpufeature code and early parameters.
313 	 */
314 	jump_label_init();
315 	parse_early_param();
316 
317 	dynamic_scs_init();
318 
319 	/*
320 	 * Unmask asynchronous aborts and fiq after bringing up possible
321 	 * earlycon. (Report possible System Errors once we can report this
322 	 * occurred).
323 	 */
324 	local_daif_restore(DAIF_PROCCTX_NOIRQ);
325 
326 	/*
327 	 * TTBR0 is only used for the identity mapping at this stage. Make it
328 	 * point to zero page to avoid speculatively fetching new entries.
329 	 */
330 	cpu_uninstall_idmap();
331 
332 	xen_early_init();
333 	efi_init();
334 
335 	if (!efi_enabled(EFI_BOOT) && ((u64)_text % MIN_KIMG_ALIGN) != 0)
336 	     pr_warn(FW_BUG "Kernel image misaligned at boot, please fix your bootloader!");
337 
338 	arm64_memblock_init();
339 
340 	paging_init();
341 
342 	acpi_table_upgrade();
343 
344 	/* Parse the ACPI tables for possible boot-time configuration */
345 	acpi_boot_table_init();
346 
347 	if (acpi_disabled)
348 		unflatten_device_tree();
349 
350 	bootmem_init();
351 
352 	kasan_init();
353 
354 	request_standard_resources();
355 
356 	early_ioremap_reset();
357 
358 	if (acpi_disabled)
359 		psci_dt_init();
360 	else
361 		psci_acpi_init();
362 
363 	init_bootcpu_ops();
364 	smp_init_cpus();
365 	smp_build_mpidr_hash();
366 
367 	/* Init percpu seeds for random tags after cpus are set up. */
368 	kasan_init_sw_tags();
369 
370 #ifdef CONFIG_ARM64_SW_TTBR0_PAN
371 	/*
372 	 * Make sure init_thread_info.ttbr0 always generates translation
373 	 * faults in case uaccess_enable() is inadvertently called by the init
374 	 * thread.
375 	 */
376 	init_task.thread_info.ttbr0 = phys_to_ttbr(__pa_symbol(reserved_pg_dir));
377 #endif
378 
379 	if (boot_args[1] || boot_args[2] || boot_args[3]) {
380 		pr_err("WARNING: x1-x3 nonzero in violation of boot protocol:\n"
381 			"\tx1: %016llx\n\tx2: %016llx\n\tx3: %016llx\n"
382 			"This indicates a broken bootloader or old kernel\n",
383 			boot_args[1], boot_args[2], boot_args[3]);
384 	}
385 }
386 
387 static inline bool cpu_can_disable(unsigned int cpu)
388 {
389 #ifdef CONFIG_HOTPLUG_CPU
390 	const struct cpu_operations *ops = get_cpu_ops(cpu);
391 
392 	if (ops && ops->cpu_can_disable)
393 		return ops->cpu_can_disable(cpu);
394 #endif
395 	return false;
396 }
397 
398 static int __init topology_init(void)
399 {
400 	int i;
401 
402 	for_each_possible_cpu(i) {
403 		struct cpu *cpu = &per_cpu(cpu_data.cpu, i);
404 		cpu->hotpluggable = cpu_can_disable(i);
405 		register_cpu(cpu, i);
406 	}
407 
408 	return 0;
409 }
410 subsys_initcall(topology_init);
411 
412 static void dump_kernel_offset(void)
413 {
414 	const unsigned long offset = kaslr_offset();
415 
416 	if (IS_ENABLED(CONFIG_RANDOMIZE_BASE) && offset > 0) {
417 		pr_emerg("Kernel Offset: 0x%lx from 0x%lx\n",
418 			 offset, KIMAGE_VADDR);
419 		pr_emerg("PHYS_OFFSET: 0x%llx\n", PHYS_OFFSET);
420 	} else {
421 		pr_emerg("Kernel Offset: disabled\n");
422 	}
423 }
424 
425 static int arm64_panic_block_dump(struct notifier_block *self,
426 				  unsigned long v, void *p)
427 {
428 	dump_kernel_offset();
429 	dump_cpu_features();
430 	dump_mem_limit();
431 	return 0;
432 }
433 
434 static struct notifier_block arm64_panic_block = {
435 	.notifier_call = arm64_panic_block_dump
436 };
437 
438 static int __init register_arm64_panic_block(void)
439 {
440 	atomic_notifier_chain_register(&panic_notifier_list,
441 				       &arm64_panic_block);
442 	return 0;
443 }
444 device_initcall(register_arm64_panic_block);
445