xref: /openbmc/linux/arch/arm64/kernel/acpi.c (revision 5c816641)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  ARM64 Specific Low-Level ACPI Boot Support
4  *
5  *  Copyright (C) 2013-2014, Linaro Ltd.
6  *	Author: Al Stone <al.stone@linaro.org>
7  *	Author: Graeme Gregory <graeme.gregory@linaro.org>
8  *	Author: Hanjun Guo <hanjun.guo@linaro.org>
9  *	Author: Tomasz Nowicki <tomasz.nowicki@linaro.org>
10  *	Author: Naresh Bhat <naresh.bhat@linaro.org>
11  */
12 
13 #define pr_fmt(fmt) "ACPI: " fmt
14 
15 #include <linux/acpi.h>
16 #include <linux/cpumask.h>
17 #include <linux/efi.h>
18 #include <linux/efi-bgrt.h>
19 #include <linux/init.h>
20 #include <linux/irq.h>
21 #include <linux/irqdomain.h>
22 #include <linux/irq_work.h>
23 #include <linux/memblock.h>
24 #include <linux/of_fdt.h>
25 #include <linux/libfdt.h>
26 #include <linux/smp.h>
27 #include <linux/serial_core.h>
28 #include <linux/pgtable.h>
29 
30 #include <acpi/ghes.h>
31 #include <asm/cputype.h>
32 #include <asm/cpu_ops.h>
33 #include <asm/daifflags.h>
34 #include <asm/smp_plat.h>
35 
36 int acpi_noirq = 1;		/* skip ACPI IRQ initialization */
37 int acpi_disabled = 1;
38 EXPORT_SYMBOL(acpi_disabled);
39 
40 int acpi_pci_disabled = 1;	/* skip ACPI PCI scan and IRQ initialization */
41 EXPORT_SYMBOL(acpi_pci_disabled);
42 
43 static bool param_acpi_off __initdata;
44 static bool param_acpi_on __initdata;
45 static bool param_acpi_force __initdata;
46 
47 static int __init parse_acpi(char *arg)
48 {
49 	if (!arg)
50 		return -EINVAL;
51 
52 	/* "acpi=off" disables both ACPI table parsing and interpreter */
53 	if (strcmp(arg, "off") == 0)
54 		param_acpi_off = true;
55 	else if (strcmp(arg, "on") == 0) /* prefer ACPI over DT */
56 		param_acpi_on = true;
57 	else if (strcmp(arg, "force") == 0) /* force ACPI to be enabled */
58 		param_acpi_force = true;
59 	else
60 		return -EINVAL;	/* Core will print when we return error */
61 
62 	return 0;
63 }
64 early_param("acpi", parse_acpi);
65 
66 static bool __init dt_is_stub(void)
67 {
68 	int node;
69 
70 	fdt_for_each_subnode(node, initial_boot_params, 0) {
71 		const char *name = fdt_get_name(initial_boot_params, node, NULL);
72 		if (strcmp(name, "chosen") == 0)
73 			continue;
74 		if (strcmp(name, "hypervisor") == 0 &&
75 		    of_flat_dt_is_compatible(node, "xen,xen"))
76 			continue;
77 
78 		return false;
79 	}
80 
81 	return true;
82 }
83 
84 /*
85  * __acpi_map_table() will be called before page_init(), so early_ioremap()
86  * or early_memremap() should be called here to for ACPI table mapping.
87  */
88 void __init __iomem *__acpi_map_table(unsigned long phys, unsigned long size)
89 {
90 	if (!size)
91 		return NULL;
92 
93 	return early_memremap(phys, size);
94 }
95 
96 void __init __acpi_unmap_table(void __iomem *map, unsigned long size)
97 {
98 	if (!map || !size)
99 		return;
100 
101 	early_memunmap(map, size);
102 }
103 
104 bool __init acpi_psci_present(void)
105 {
106 	return acpi_gbl_FADT.arm_boot_flags & ACPI_FADT_PSCI_COMPLIANT;
107 }
108 
109 /* Whether HVC must be used instead of SMC as the PSCI conduit */
110 bool acpi_psci_use_hvc(void)
111 {
112 	return acpi_gbl_FADT.arm_boot_flags & ACPI_FADT_PSCI_USE_HVC;
113 }
114 
115 /*
116  * acpi_fadt_sanity_check() - Check FADT presence and carry out sanity
117  *			      checks on it
118  *
119  * Return 0 on success,  <0 on failure
120  */
121 static int __init acpi_fadt_sanity_check(void)
122 {
123 	struct acpi_table_header *table;
124 	struct acpi_table_fadt *fadt;
125 	acpi_status status;
126 	int ret = 0;
127 
128 	/*
129 	 * FADT is required on arm64; retrieve it to check its presence
130 	 * and carry out revision and ACPI HW reduced compliancy tests
131 	 */
132 	status = acpi_get_table(ACPI_SIG_FADT, 0, &table);
133 	if (ACPI_FAILURE(status)) {
134 		const char *msg = acpi_format_exception(status);
135 
136 		pr_err("Failed to get FADT table, %s\n", msg);
137 		return -ENODEV;
138 	}
139 
140 	fadt = (struct acpi_table_fadt *)table;
141 
142 	/*
143 	 * Revision in table header is the FADT Major revision, and there
144 	 * is a minor revision of FADT which was introduced by ACPI 5.1,
145 	 * we only deal with ACPI 5.1 or newer revision to get GIC and SMP
146 	 * boot protocol configuration data.
147 	 */
148 	if (table->revision < 5 ||
149 	   (table->revision == 5 && fadt->minor_revision < 1)) {
150 		pr_err(FW_BUG "Unsupported FADT revision %d.%d, should be 5.1+\n",
151 		       table->revision, fadt->minor_revision);
152 
153 		if (!fadt->arm_boot_flags) {
154 			ret = -EINVAL;
155 			goto out;
156 		}
157 		pr_err("FADT has ARM boot flags set, assuming 5.1\n");
158 	}
159 
160 	if (!(fadt->flags & ACPI_FADT_HW_REDUCED)) {
161 		pr_err("FADT not ACPI hardware reduced compliant\n");
162 		ret = -EINVAL;
163 	}
164 
165 out:
166 	/*
167 	 * acpi_get_table() creates FADT table mapping that
168 	 * should be released after parsing and before resuming boot
169 	 */
170 	acpi_put_table(table);
171 	return ret;
172 }
173 
174 /*
175  * acpi_boot_table_init() called from setup_arch(), always.
176  *	1. find RSDP and get its address, and then find XSDT
177  *	2. extract all tables and checksums them all
178  *	3. check ACPI FADT revision
179  *	4. check ACPI FADT HW reduced flag
180  *
181  * We can parse ACPI boot-time tables such as MADT after
182  * this function is called.
183  *
184  * On return ACPI is enabled if either:
185  *
186  * - ACPI tables are initialized and sanity checks passed
187  * - acpi=force was passed in the command line and ACPI was not disabled
188  *   explicitly through acpi=off command line parameter
189  *
190  * ACPI is disabled on function return otherwise
191  */
192 void __init acpi_boot_table_init(void)
193 {
194 	/*
195 	 * Enable ACPI instead of device tree unless
196 	 * - ACPI has been disabled explicitly (acpi=off), or
197 	 * - the device tree is not empty (it has more than just a /chosen node,
198 	 *   and a /hypervisor node when running on Xen)
199 	 *   and ACPI has not been [force] enabled (acpi=on|force)
200 	 */
201 	if (param_acpi_off ||
202 	    (!param_acpi_on && !param_acpi_force && !dt_is_stub()))
203 		goto done;
204 
205 	/*
206 	 * ACPI is disabled at this point. Enable it in order to parse
207 	 * the ACPI tables and carry out sanity checks
208 	 */
209 	enable_acpi();
210 
211 	/*
212 	 * If ACPI tables are initialized and FADT sanity checks passed,
213 	 * leave ACPI enabled and carry on booting; otherwise disable ACPI
214 	 * on initialization error.
215 	 * If acpi=force was passed on the command line it forces ACPI
216 	 * to be enabled even if its initialization failed.
217 	 */
218 	if (acpi_table_init() || acpi_fadt_sanity_check()) {
219 		pr_err("Failed to init ACPI tables\n");
220 		if (!param_acpi_force)
221 			disable_acpi();
222 	}
223 
224 done:
225 	if (acpi_disabled) {
226 		if (earlycon_acpi_spcr_enable)
227 			early_init_dt_scan_chosen_stdout();
228 	} else {
229 		acpi_parse_spcr(earlycon_acpi_spcr_enable, true);
230 		if (IS_ENABLED(CONFIG_ACPI_BGRT))
231 			acpi_table_parse(ACPI_SIG_BGRT, acpi_parse_bgrt);
232 	}
233 }
234 
235 static pgprot_t __acpi_get_writethrough_mem_attribute(void)
236 {
237 	/*
238 	 * Although UEFI specifies the use of Normal Write-through for
239 	 * EFI_MEMORY_WT, it is seldom used in practice and not implemented
240 	 * by most (all?) CPUs. Rather than allocate a MAIR just for this
241 	 * purpose, emit a warning and use Normal Non-cacheable instead.
242 	 */
243 	pr_warn_once("No MAIR allocation for EFI_MEMORY_WT; treating as Normal Non-cacheable\n");
244 	return __pgprot(PROT_NORMAL_NC);
245 }
246 
247 pgprot_t __acpi_get_mem_attribute(phys_addr_t addr)
248 {
249 	/*
250 	 * According to "Table 8 Map: EFI memory types to AArch64 memory
251 	 * types" of UEFI 2.5 section 2.3.6.1, each EFI memory type is
252 	 * mapped to a corresponding MAIR attribute encoding.
253 	 * The EFI memory attribute advises all possible capabilities
254 	 * of a memory region.
255 	 */
256 
257 	u64 attr;
258 
259 	attr = efi_mem_attributes(addr);
260 	if (attr & EFI_MEMORY_WB)
261 		return PAGE_KERNEL;
262 	if (attr & EFI_MEMORY_WC)
263 		return __pgprot(PROT_NORMAL_NC);
264 	if (attr & EFI_MEMORY_WT)
265 		return __acpi_get_writethrough_mem_attribute();
266 	return __pgprot(PROT_DEVICE_nGnRnE);
267 }
268 
269 void __iomem *acpi_os_ioremap(acpi_physical_address phys, acpi_size size)
270 {
271 	efi_memory_desc_t *md, *region = NULL;
272 	pgprot_t prot;
273 
274 	if (WARN_ON_ONCE(!efi_enabled(EFI_MEMMAP)))
275 		return NULL;
276 
277 	for_each_efi_memory_desc(md) {
278 		u64 end = md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT);
279 
280 		if (phys < md->phys_addr || phys >= end)
281 			continue;
282 
283 		if (phys + size > end) {
284 			pr_warn(FW_BUG "requested region covers multiple EFI memory regions\n");
285 			return NULL;
286 		}
287 		region = md;
288 		break;
289 	}
290 
291 	/*
292 	 * It is fine for AML to remap regions that are not represented in the
293 	 * EFI memory map at all, as it only describes normal memory, and MMIO
294 	 * regions that require a virtual mapping to make them accessible to
295 	 * the EFI runtime services.
296 	 */
297 	prot = __pgprot(PROT_DEVICE_nGnRnE);
298 	if (region) {
299 		switch (region->type) {
300 		case EFI_LOADER_CODE:
301 		case EFI_LOADER_DATA:
302 		case EFI_BOOT_SERVICES_CODE:
303 		case EFI_BOOT_SERVICES_DATA:
304 		case EFI_CONVENTIONAL_MEMORY:
305 		case EFI_PERSISTENT_MEMORY:
306 			if (memblock_is_map_memory(phys) ||
307 			    !memblock_is_region_memory(phys, size)) {
308 				pr_warn(FW_BUG "requested region covers kernel memory @ %pa\n", &phys);
309 				return NULL;
310 			}
311 			/*
312 			 * Mapping kernel memory is permitted if the region in
313 			 * question is covered by a single memblock with the
314 			 * NOMAP attribute set: this enables the use of ACPI
315 			 * table overrides passed via initramfs, which are
316 			 * reserved in memory using arch_reserve_mem_area()
317 			 * below. As this particular use case only requires
318 			 * read access, fall through to the R/O mapping case.
319 			 */
320 			fallthrough;
321 
322 		case EFI_RUNTIME_SERVICES_CODE:
323 			/*
324 			 * This would be unusual, but not problematic per se,
325 			 * as long as we take care not to create a writable
326 			 * mapping for executable code.
327 			 */
328 			prot = PAGE_KERNEL_RO;
329 			break;
330 
331 		case EFI_ACPI_RECLAIM_MEMORY:
332 			/*
333 			 * ACPI reclaim memory is used to pass firmware tables
334 			 * and other data that is intended for consumption by
335 			 * the OS only, which may decide it wants to reclaim
336 			 * that memory and use it for something else. We never
337 			 * do that, but we usually add it to the linear map
338 			 * anyway, in which case we should use the existing
339 			 * mapping.
340 			 */
341 			if (memblock_is_map_memory(phys))
342 				return (void __iomem *)__phys_to_virt(phys);
343 			fallthrough;
344 
345 		default:
346 			if (region->attribute & EFI_MEMORY_WB)
347 				prot = PAGE_KERNEL;
348 			else if (region->attribute & EFI_MEMORY_WC)
349 				prot = __pgprot(PROT_NORMAL_NC);
350 			else if (region->attribute & EFI_MEMORY_WT)
351 				prot = __acpi_get_writethrough_mem_attribute();
352 		}
353 	}
354 	return __ioremap(phys, size, prot);
355 }
356 
357 /*
358  * Claim Synchronous External Aborts as a firmware first notification.
359  *
360  * Used by KVM and the arch do_sea handler.
361  * @regs may be NULL when called from process context.
362  */
363 int apei_claim_sea(struct pt_regs *regs)
364 {
365 	int err = -ENOENT;
366 	bool return_to_irqs_enabled;
367 	unsigned long current_flags;
368 
369 	if (!IS_ENABLED(CONFIG_ACPI_APEI_GHES))
370 		return err;
371 
372 	current_flags = local_daif_save_flags();
373 
374 	/* current_flags isn't useful here as daif doesn't tell us about pNMI */
375 	return_to_irqs_enabled = !irqs_disabled_flags(arch_local_save_flags());
376 
377 	if (regs)
378 		return_to_irqs_enabled = interrupts_enabled(regs);
379 
380 	/*
381 	 * SEA can interrupt SError, mask it and describe this as an NMI so
382 	 * that APEI defers the handling.
383 	 */
384 	local_daif_restore(DAIF_ERRCTX);
385 	nmi_enter();
386 	err = ghes_notify_sea();
387 	nmi_exit();
388 
389 	/*
390 	 * APEI NMI-like notifications are deferred to irq_work. Unless
391 	 * we interrupted irqs-masked code, we can do that now.
392 	 */
393 	if (!err) {
394 		if (return_to_irqs_enabled) {
395 			local_daif_restore(DAIF_PROCCTX_NOIRQ);
396 			__irq_enter();
397 			irq_work_run();
398 			__irq_exit();
399 		} else {
400 			pr_warn_ratelimited("APEI work queued but not completed");
401 			err = -EINPROGRESS;
402 		}
403 	}
404 
405 	local_daif_restore(current_flags);
406 
407 	return err;
408 }
409 
410 void arch_reserve_mem_area(acpi_physical_address addr, size_t size)
411 {
412 	memblock_mark_nomap(addr, size);
413 }
414