xref: /openbmc/linux/arch/arm64/kernel/setup.c (revision 2f61c664)
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/proc_fs.h>
27 #include <linux/memblock.h>
28 #include <linux/of_fdt.h>
29 #include <linux/efi.h>
30 #include <linux/psci.h>
31 #include <linux/sched/task.h>
32 #include <linux/mm.h>
33 
34 #include <asm/acpi.h>
35 #include <asm/fixmap.h>
36 #include <asm/cpu.h>
37 #include <asm/cputype.h>
38 #include <asm/daifflags.h>
39 #include <asm/elf.h>
40 #include <asm/cpufeature.h>
41 #include <asm/cpu_ops.h>
42 #include <asm/kasan.h>
43 #include <asm/numa.h>
44 #include <asm/sections.h>
45 #include <asm/setup.h>
46 #include <asm/smp_plat.h>
47 #include <asm/cacheflush.h>
48 #include <asm/tlbflush.h>
49 #include <asm/traps.h>
50 #include <asm/efi.h>
51 #include <asm/xen/hypervisor.h>
52 #include <asm/mmu_context.h>
53 
54 static int num_standard_resources;
55 static struct resource *standard_resources;
56 
57 phys_addr_t __fdt_pointer __initdata;
58 
59 /*
60  * Standard memory resources
61  */
62 static struct resource mem_res[] = {
63 	{
64 		.name = "Kernel code",
65 		.start = 0,
66 		.end = 0,
67 		.flags = IORESOURCE_SYSTEM_RAM
68 	},
69 	{
70 		.name = "Kernel data",
71 		.start = 0,
72 		.end = 0,
73 		.flags = IORESOURCE_SYSTEM_RAM
74 	}
75 };
76 
77 #define kernel_code mem_res[0]
78 #define kernel_data mem_res[1]
79 
80 /*
81  * The recorded values of x0 .. x3 upon kernel entry.
82  */
83 u64 __cacheline_aligned boot_args[4];
84 
85 void __init smp_setup_processor_id(void)
86 {
87 	u64 mpidr = read_cpuid_mpidr() & MPIDR_HWID_BITMASK;
88 	set_cpu_logical_map(0, mpidr);
89 
90 	/*
91 	 * clear __my_cpu_offset on boot CPU to avoid hang caused by
92 	 * using percpu variable early, for example, lockdep will
93 	 * access percpu variable inside lock_release
94 	 */
95 	set_my_cpu_offset(0);
96 	pr_info("Booting Linux on physical CPU 0x%010lx [0x%08x]\n",
97 		(unsigned long)mpidr, read_cpuid_id());
98 }
99 
100 bool arch_match_cpu_phys_id(int cpu, u64 phys_id)
101 {
102 	return phys_id == cpu_logical_map(cpu);
103 }
104 
105 struct mpidr_hash mpidr_hash;
106 /**
107  * smp_build_mpidr_hash - Pre-compute shifts required at each affinity
108  *			  level in order to build a linear index from an
109  *			  MPIDR value. Resulting algorithm is a collision
110  *			  free hash carried out through shifting and ORing
111  */
112 static void __init smp_build_mpidr_hash(void)
113 {
114 	u32 i, affinity, fs[4], bits[4], ls;
115 	u64 mask = 0;
116 	/*
117 	 * Pre-scan the list of MPIDRS and filter out bits that do
118 	 * not contribute to affinity levels, ie they never toggle.
119 	 */
120 	for_each_possible_cpu(i)
121 		mask |= (cpu_logical_map(i) ^ cpu_logical_map(0));
122 	pr_debug("mask of set bits %#llx\n", mask);
123 	/*
124 	 * Find and stash the last and first bit set at all affinity levels to
125 	 * check how many bits are required to represent them.
126 	 */
127 	for (i = 0; i < 4; i++) {
128 		affinity = MPIDR_AFFINITY_LEVEL(mask, i);
129 		/*
130 		 * Find the MSB bit and LSB bits position
131 		 * to determine how many bits are required
132 		 * to express the affinity level.
133 		 */
134 		ls = fls(affinity);
135 		fs[i] = affinity ? ffs(affinity) - 1 : 0;
136 		bits[i] = ls - fs[i];
137 	}
138 	/*
139 	 * An index can be created from the MPIDR_EL1 by isolating the
140 	 * significant bits at each affinity level and by shifting
141 	 * them in order to compress the 32 bits values space to a
142 	 * compressed set of values. This is equivalent to hashing
143 	 * the MPIDR_EL1 through shifting and ORing. It is a collision free
144 	 * hash though not minimal since some levels might contain a number
145 	 * of CPUs that is not an exact power of 2 and their bit
146 	 * representation might contain holes, eg MPIDR_EL1[7:0] = {0x2, 0x80}.
147 	 */
148 	mpidr_hash.shift_aff[0] = MPIDR_LEVEL_SHIFT(0) + fs[0];
149 	mpidr_hash.shift_aff[1] = MPIDR_LEVEL_SHIFT(1) + fs[1] - bits[0];
150 	mpidr_hash.shift_aff[2] = MPIDR_LEVEL_SHIFT(2) + fs[2] -
151 						(bits[1] + bits[0]);
152 	mpidr_hash.shift_aff[3] = MPIDR_LEVEL_SHIFT(3) +
153 				  fs[3] - (bits[2] + bits[1] + bits[0]);
154 	mpidr_hash.mask = mask;
155 	mpidr_hash.bits = bits[3] + bits[2] + bits[1] + bits[0];
156 	pr_debug("MPIDR hash: aff0[%u] aff1[%u] aff2[%u] aff3[%u] mask[%#llx] bits[%u]\n",
157 		mpidr_hash.shift_aff[0],
158 		mpidr_hash.shift_aff[1],
159 		mpidr_hash.shift_aff[2],
160 		mpidr_hash.shift_aff[3],
161 		mpidr_hash.mask,
162 		mpidr_hash.bits);
163 	/*
164 	 * 4x is an arbitrary value used to warn on a hash table much bigger
165 	 * than expected on most systems.
166 	 */
167 	if (mpidr_hash_size() > 4 * num_possible_cpus())
168 		pr_warn("Large number of MPIDR hash buckets detected\n");
169 }
170 
171 static void *early_fdt_ptr __initdata;
172 
173 void __init *get_early_fdt_ptr(void)
174 {
175 	return early_fdt_ptr;
176 }
177 
178 asmlinkage void __init early_fdt_map(u64 dt_phys)
179 {
180 	int fdt_size;
181 
182 	early_fixmap_init();
183 	early_fdt_ptr = fixmap_remap_fdt(dt_phys, &fdt_size, PAGE_KERNEL);
184 }
185 
186 static void __init setup_machine_fdt(phys_addr_t dt_phys)
187 {
188 	int size;
189 	void *dt_virt = fixmap_remap_fdt(dt_phys, &size, PAGE_KERNEL);
190 	const char *name;
191 
192 	if (dt_virt)
193 		memblock_reserve(dt_phys, size);
194 
195 	if (!dt_virt || !early_init_dt_scan(dt_virt)) {
196 		pr_crit("\n"
197 			"Error: invalid device tree blob at physical address %pa (virtual address 0x%p)\n"
198 			"The dtb must be 8-byte aligned and must not exceed 2 MB in size\n"
199 			"\nPlease check your bootloader.",
200 			&dt_phys, dt_virt);
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 		} else {
241 			res->name  = "System RAM";
242 			res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
243 		}
244 		res->start = __pfn_to_phys(memblock_region_memory_base_pfn(region));
245 		res->end = __pfn_to_phys(memblock_region_memory_end_pfn(region)) - 1;
246 
247 		request_resource(&iomem_resource, res);
248 
249 		if (kernel_code.start >= res->start &&
250 		    kernel_code.end <= res->end)
251 			request_resource(res, &kernel_code);
252 		if (kernel_data.start >= res->start &&
253 		    kernel_data.end <= res->end)
254 			request_resource(res, &kernel_data);
255 #ifdef CONFIG_KEXEC_CORE
256 		/* Userspace will find "Crash kernel" region in /proc/iomem. */
257 		if (crashk_res.end && crashk_res.start >= res->start &&
258 		    crashk_res.end <= res->end)
259 			request_resource(res, &crashk_res);
260 #endif
261 	}
262 }
263 
264 static int __init reserve_memblock_reserved_regions(void)
265 {
266 	u64 i, j;
267 
268 	for (i = 0; i < num_standard_resources; ++i) {
269 		struct resource *mem = &standard_resources[i];
270 		phys_addr_t r_start, r_end, mem_size = resource_size(mem);
271 
272 		if (!memblock_is_region_reserved(mem->start, mem_size))
273 			continue;
274 
275 		for_each_reserved_mem_range(j, &r_start, &r_end) {
276 			resource_size_t start, end;
277 
278 			start = max(PFN_PHYS(PFN_DOWN(r_start)), mem->start);
279 			end = min(PFN_PHYS(PFN_UP(r_end)) - 1, mem->end);
280 
281 			if (start > mem->end || end < mem->start)
282 				continue;
283 
284 			reserve_region_with_split(mem, start, end, "reserved");
285 		}
286 	}
287 
288 	return 0;
289 }
290 arch_initcall(reserve_memblock_reserved_regions);
291 
292 u64 __cpu_logical_map[NR_CPUS] = { [0 ... NR_CPUS-1] = INVALID_HWID };
293 
294 u64 cpu_logical_map(unsigned int cpu)
295 {
296 	return __cpu_logical_map[cpu];
297 }
298 
299 void __init __no_sanitize_address setup_arch(char **cmdline_p)
300 {
301 	init_mm.start_code = (unsigned long) _stext;
302 	init_mm.end_code   = (unsigned long) _etext;
303 	init_mm.end_data   = (unsigned long) _edata;
304 	init_mm.brk	   = (unsigned long) _end;
305 
306 	*cmdline_p = boot_command_line;
307 
308 	/*
309 	 * If know now we are going to need KPTI then use non-global
310 	 * mappings from the start, avoiding the cost of rewriting
311 	 * everything later.
312 	 */
313 	arm64_use_ng_mappings = kaslr_requires_kpti();
314 
315 	early_fixmap_init();
316 	early_ioremap_init();
317 
318 	setup_machine_fdt(__fdt_pointer);
319 
320 	/*
321 	 * Initialise the static keys early as they may be enabled by the
322 	 * cpufeature code and early parameters.
323 	 */
324 	jump_label_init();
325 	parse_early_param();
326 
327 	/*
328 	 * Unmask asynchronous aborts and fiq after bringing up possible
329 	 * earlycon. (Report possible System Errors once we can report this
330 	 * occurred).
331 	 */
332 	local_daif_restore(DAIF_PROCCTX_NOIRQ);
333 
334 	/*
335 	 * TTBR0 is only used for the identity mapping at this stage. Make it
336 	 * point to zero page to avoid speculatively fetching new entries.
337 	 */
338 	cpu_uninstall_idmap();
339 
340 	xen_early_init();
341 	efi_init();
342 
343 	if (!efi_enabled(EFI_BOOT) && ((u64)_text % MIN_KIMG_ALIGN) != 0)
344 	     pr_warn(FW_BUG "Kernel image misaligned at boot, please fix your bootloader!");
345 
346 	arm64_memblock_init();
347 
348 	paging_init();
349 
350 	acpi_table_upgrade();
351 
352 	/* Parse the ACPI tables for possible boot-time configuration */
353 	acpi_boot_table_init();
354 
355 	if (acpi_disabled)
356 		unflatten_device_tree();
357 
358 	bootmem_init();
359 
360 	kasan_init();
361 
362 	request_standard_resources();
363 
364 	early_ioremap_reset();
365 
366 	if (acpi_disabled)
367 		psci_dt_init();
368 	else
369 		psci_acpi_init();
370 
371 	init_bootcpu_ops();
372 	smp_init_cpus();
373 	smp_build_mpidr_hash();
374 
375 	/* Init percpu seeds for random tags after cpus are set up. */
376 	kasan_init_sw_tags();
377 
378 #ifdef CONFIG_ARM64_SW_TTBR0_PAN
379 	/*
380 	 * Make sure init_thread_info.ttbr0 always generates translation
381 	 * faults in case uaccess_enable() is inadvertently called by the init
382 	 * thread.
383 	 */
384 	init_task.thread_info.ttbr0 = __pa_symbol(reserved_pg_dir);
385 #endif
386 
387 	if (boot_args[1] || boot_args[2] || boot_args[3]) {
388 		pr_err("WARNING: x1-x3 nonzero in violation of boot protocol:\n"
389 			"\tx1: %016llx\n\tx2: %016llx\n\tx3: %016llx\n"
390 			"This indicates a broken bootloader or old kernel\n",
391 			boot_args[1], boot_args[2], boot_args[3]);
392 	}
393 }
394 
395 static inline bool cpu_can_disable(unsigned int cpu)
396 {
397 #ifdef CONFIG_HOTPLUG_CPU
398 	const struct cpu_operations *ops = get_cpu_ops(cpu);
399 
400 	if (ops && ops->cpu_can_disable)
401 		return ops->cpu_can_disable(cpu);
402 #endif
403 	return false;
404 }
405 
406 static int __init topology_init(void)
407 {
408 	int i;
409 
410 	for_each_online_node(i)
411 		register_one_node(i);
412 
413 	for_each_possible_cpu(i) {
414 		struct cpu *cpu = &per_cpu(cpu_data.cpu, i);
415 		cpu->hotpluggable = cpu_can_disable(i);
416 		register_cpu(cpu, i);
417 	}
418 
419 	return 0;
420 }
421 subsys_initcall(topology_init);
422 
423 static void dump_kernel_offset(void)
424 {
425 	const unsigned long offset = kaslr_offset();
426 
427 	if (IS_ENABLED(CONFIG_RANDOMIZE_BASE) && offset > 0) {
428 		pr_emerg("Kernel Offset: 0x%lx from 0x%lx\n",
429 			 offset, KIMAGE_VADDR);
430 		pr_emerg("PHYS_OFFSET: 0x%llx\n", PHYS_OFFSET);
431 	} else {
432 		pr_emerg("Kernel Offset: disabled\n");
433 	}
434 }
435 
436 static int arm64_panic_block_dump(struct notifier_block *self,
437 				  unsigned long v, void *p)
438 {
439 	dump_kernel_offset();
440 	dump_cpu_features();
441 	dump_mem_limit();
442 	return 0;
443 }
444 
445 static struct notifier_block arm64_panic_block = {
446 	.notifier_call = arm64_panic_block_dump
447 };
448 
449 static int __init register_arm64_panic_block(void)
450 {
451 	atomic_notifier_chain_register(&panic_notifier_list,
452 				       &arm64_panic_block);
453 	return 0;
454 }
455 device_initcall(register_arm64_panic_block);
456