1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * PowerNV setup code.
4  *
5  * Copyright 2011 IBM Corp.
6  */
7 
8 #undef DEBUG
9 
10 #include <linux/cpu.h>
11 #include <linux/errno.h>
12 #include <linux/sched.h>
13 #include <linux/kernel.h>
14 #include <linux/tty.h>
15 #include <linux/reboot.h>
16 #include <linux/init.h>
17 #include <linux/console.h>
18 #include <linux/delay.h>
19 #include <linux/irq.h>
20 #include <linux/seq_file.h>
21 #include <linux/of.h>
22 #include <linux/of_fdt.h>
23 #include <linux/interrupt.h>
24 #include <linux/bug.h>
25 #include <linux/pci.h>
26 #include <linux/cpufreq.h>
27 #include <linux/memblock.h>
28 
29 #include <asm/machdep.h>
30 #include <asm/firmware.h>
31 #include <asm/xics.h>
32 #include <asm/xive.h>
33 #include <asm/opal.h>
34 #include <asm/kexec.h>
35 #include <asm/smp.h>
36 #include <asm/tm.h>
37 #include <asm/setup.h>
38 #include <asm/security_features.h>
39 
40 #include "powernv.h"
41 
42 
43 static bool __init fw_feature_is(const char *state, const char *name,
44 			  struct device_node *fw_features)
45 {
46 	struct device_node *np;
47 	bool rc = false;
48 
49 	np = of_get_child_by_name(fw_features, name);
50 	if (np) {
51 		rc = of_property_read_bool(np, state);
52 		of_node_put(np);
53 	}
54 
55 	return rc;
56 }
57 
58 static void __init init_fw_feat_flags(struct device_node *np)
59 {
60 	if (fw_feature_is("enabled", "inst-spec-barrier-ori31,31,0", np))
61 		security_ftr_set(SEC_FTR_SPEC_BAR_ORI31);
62 
63 	if (fw_feature_is("enabled", "fw-bcctrl-serialized", np))
64 		security_ftr_set(SEC_FTR_BCCTRL_SERIALISED);
65 
66 	if (fw_feature_is("enabled", "inst-l1d-flush-ori30,30,0", np))
67 		security_ftr_set(SEC_FTR_L1D_FLUSH_ORI30);
68 
69 	if (fw_feature_is("enabled", "inst-l1d-flush-trig2", np))
70 		security_ftr_set(SEC_FTR_L1D_FLUSH_TRIG2);
71 
72 	if (fw_feature_is("enabled", "fw-l1d-thread-split", np))
73 		security_ftr_set(SEC_FTR_L1D_THREAD_PRIV);
74 
75 	if (fw_feature_is("enabled", "fw-count-cache-disabled", np))
76 		security_ftr_set(SEC_FTR_COUNT_CACHE_DISABLED);
77 
78 	if (fw_feature_is("enabled", "fw-count-cache-flush-bcctr2,0,0", np))
79 		security_ftr_set(SEC_FTR_BCCTR_FLUSH_ASSIST);
80 
81 	if (fw_feature_is("enabled", "needs-count-cache-flush-on-context-switch", np))
82 		security_ftr_set(SEC_FTR_FLUSH_COUNT_CACHE);
83 
84 	/*
85 	 * The features below are enabled by default, so we instead look to see
86 	 * if firmware has *disabled* them, and clear them if so.
87 	 */
88 	if (fw_feature_is("disabled", "speculation-policy-favor-security", np))
89 		security_ftr_clear(SEC_FTR_FAVOUR_SECURITY);
90 
91 	if (fw_feature_is("disabled", "needs-l1d-flush-msr-pr-0-to-1", np))
92 		security_ftr_clear(SEC_FTR_L1D_FLUSH_PR);
93 
94 	if (fw_feature_is("disabled", "needs-l1d-flush-msr-hv-1-to-0", np))
95 		security_ftr_clear(SEC_FTR_L1D_FLUSH_HV);
96 
97 	if (fw_feature_is("disabled", "needs-spec-barrier-for-bound-checks", np))
98 		security_ftr_clear(SEC_FTR_BNDS_CHK_SPEC_BAR);
99 }
100 
101 static void __init pnv_setup_security_mitigations(void)
102 {
103 	struct device_node *np, *fw_features;
104 	enum l1d_flush_type type;
105 	bool enable;
106 
107 	/* Default to fallback in case fw-features are not available */
108 	type = L1D_FLUSH_FALLBACK;
109 
110 	np = of_find_node_by_name(NULL, "ibm,opal");
111 	fw_features = of_get_child_by_name(np, "fw-features");
112 	of_node_put(np);
113 
114 	if (fw_features) {
115 		init_fw_feat_flags(fw_features);
116 		of_node_put(fw_features);
117 
118 		if (security_ftr_enabled(SEC_FTR_L1D_FLUSH_TRIG2))
119 			type = L1D_FLUSH_MTTRIG;
120 
121 		if (security_ftr_enabled(SEC_FTR_L1D_FLUSH_ORI30))
122 			type = L1D_FLUSH_ORI;
123 	}
124 
125 	/*
126 	 * The issues addressed by the entry and uaccess flush don't affect P7
127 	 * or P8, so on bare metal disable them explicitly in case firmware does
128 	 * not include the features to disable them. POWER9 and newer processors
129 	 * should have the appropriate firmware flags.
130 	 */
131 	if (pvr_version_is(PVR_POWER7) || pvr_version_is(PVR_POWER7p) ||
132 	    pvr_version_is(PVR_POWER8E) || pvr_version_is(PVR_POWER8NVL) ||
133 	    pvr_version_is(PVR_POWER8)) {
134 		security_ftr_clear(SEC_FTR_L1D_FLUSH_ENTRY);
135 		security_ftr_clear(SEC_FTR_L1D_FLUSH_UACCESS);
136 	}
137 
138 	enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) && \
139 		 (security_ftr_enabled(SEC_FTR_L1D_FLUSH_PR)   || \
140 		  security_ftr_enabled(SEC_FTR_L1D_FLUSH_HV));
141 
142 	setup_rfi_flush(type, enable);
143 	setup_count_cache_flush();
144 
145 	enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) &&
146 		 security_ftr_enabled(SEC_FTR_L1D_FLUSH_ENTRY);
147 	setup_entry_flush(enable);
148 
149 	enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) &&
150 		 security_ftr_enabled(SEC_FTR_L1D_FLUSH_UACCESS);
151 	setup_uaccess_flush(enable);
152 
153 	setup_stf_barrier();
154 }
155 
156 static void __init pnv_check_guarded_cores(void)
157 {
158 	struct device_node *dn;
159 	int bad_count = 0;
160 
161 	for_each_node_by_type(dn, "cpu") {
162 		if (of_property_match_string(dn, "status", "bad") >= 0)
163 			bad_count++;
164 	}
165 
166 	if (bad_count) {
167 		printk("  _     _______________\n");
168 		pr_cont(" | |   /               \\\n");
169 		pr_cont(" | |   |    WARNING!   |\n");
170 		pr_cont(" | |   |               |\n");
171 		pr_cont(" | |   | It looks like |\n");
172 		pr_cont(" |_|   |  you have %*d |\n", 3, bad_count);
173 		pr_cont("  _    | guarded cores |\n");
174 		pr_cont(" (_)   \\_______________/\n");
175 	}
176 }
177 
178 static void __init pnv_setup_arch(void)
179 {
180 	set_arch_panic_timeout(10, ARCH_PANIC_TIMEOUT);
181 
182 	pnv_setup_security_mitigations();
183 
184 	/* Initialize SMP */
185 	pnv_smp_init();
186 
187 	/* Setup RTC and NVRAM callbacks */
188 	if (firmware_has_feature(FW_FEATURE_OPAL))
189 		opal_nvram_init();
190 
191 	/* Enable NAP mode */
192 	powersave_nap = 1;
193 
194 	pnv_check_guarded_cores();
195 
196 	/* XXX PMCS */
197 }
198 
199 static void __init pnv_init(void)
200 {
201 	/*
202 	 * Initialize the LPC bus now so that legacy serial
203 	 * ports can be found on it
204 	 */
205 	opal_lpc_init();
206 
207 #ifdef CONFIG_HVC_OPAL
208 	if (firmware_has_feature(FW_FEATURE_OPAL))
209 		hvc_opal_init_early();
210 	else
211 #endif
212 		add_preferred_console("hvc", 0, NULL);
213 
214 #ifdef CONFIG_PPC_64S_HASH_MMU
215 	if (!radix_enabled()) {
216 		size_t size = sizeof(struct slb_entry) * mmu_slb_size;
217 		int i;
218 
219 		/* Allocate per cpu area to save old slb contents during MCE */
220 		for_each_possible_cpu(i) {
221 			paca_ptrs[i]->mce_faulty_slbs =
222 					memblock_alloc_node(size,
223 						__alignof__(struct slb_entry),
224 						cpu_to_node(i));
225 		}
226 	}
227 #endif
228 }
229 
230 static void __init pnv_init_IRQ(void)
231 {
232 	/* Try using a XIVE if available, otherwise use a XICS */
233 	if (!xive_native_init())
234 		xics_init();
235 
236 	WARN_ON(!ppc_md.get_irq);
237 }
238 
239 static void pnv_show_cpuinfo(struct seq_file *m)
240 {
241 	struct device_node *root;
242 	const char *model = "";
243 
244 	root = of_find_node_by_path("/");
245 	if (root)
246 		model = of_get_property(root, "model", NULL);
247 	seq_printf(m, "machine\t\t: PowerNV %s\n", model);
248 	if (firmware_has_feature(FW_FEATURE_OPAL))
249 		seq_printf(m, "firmware\t: OPAL\n");
250 	else
251 		seq_printf(m, "firmware\t: BML\n");
252 	of_node_put(root);
253 	if (radix_enabled())
254 		seq_printf(m, "MMU\t\t: Radix\n");
255 	else
256 		seq_printf(m, "MMU\t\t: Hash\n");
257 }
258 
259 static void pnv_prepare_going_down(void)
260 {
261 	/*
262 	 * Disable all notifiers from OPAL, we can't
263 	 * service interrupts anymore anyway
264 	 */
265 	opal_event_shutdown();
266 
267 	/* Print flash update message if one is scheduled. */
268 	opal_flash_update_print_message();
269 
270 	smp_send_stop();
271 
272 	hard_irq_disable();
273 }
274 
275 static void  __noreturn pnv_restart(char *cmd)
276 {
277 	long rc;
278 
279 	pnv_prepare_going_down();
280 
281 	do {
282 		if (!cmd || !strlen(cmd))
283 			rc = opal_cec_reboot();
284 		else if (strcmp(cmd, "full") == 0)
285 			rc = opal_cec_reboot2(OPAL_REBOOT_FULL_IPL, NULL);
286 		else if (strcmp(cmd, "mpipl") == 0)
287 			rc = opal_cec_reboot2(OPAL_REBOOT_MPIPL, NULL);
288 		else if (strcmp(cmd, "error") == 0)
289 			rc = opal_cec_reboot2(OPAL_REBOOT_PLATFORM_ERROR, NULL);
290 		else if (strcmp(cmd, "fast") == 0)
291 			rc = opal_cec_reboot2(OPAL_REBOOT_FAST, NULL);
292 		else
293 			rc = OPAL_UNSUPPORTED;
294 
295 		if (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
296 			/* Opal is busy wait for some time and retry */
297 			opal_poll_events(NULL);
298 			mdelay(10);
299 
300 		} else	if (cmd && rc) {
301 			/* Unknown error while issuing reboot */
302 			if (rc == OPAL_UNSUPPORTED)
303 				pr_err("Unsupported '%s' reboot.\n", cmd);
304 			else
305 				pr_err("Unable to issue '%s' reboot. Err=%ld\n",
306 				       cmd, rc);
307 			pr_info("Forcing a cec-reboot\n");
308 			cmd = NULL;
309 			rc = OPAL_BUSY;
310 
311 		} else if (rc != OPAL_SUCCESS) {
312 			/* Unknown error while issuing cec-reboot */
313 			pr_err("Unable to reboot. Err=%ld\n", rc);
314 		}
315 
316 	} while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT);
317 
318 	for (;;)
319 		opal_poll_events(NULL);
320 }
321 
322 static void __noreturn pnv_power_off(void)
323 {
324 	long rc = OPAL_BUSY;
325 
326 	pnv_prepare_going_down();
327 
328 	while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
329 		rc = opal_cec_power_down(0);
330 		if (rc == OPAL_BUSY_EVENT)
331 			opal_poll_events(NULL);
332 		else
333 			mdelay(10);
334 	}
335 	for (;;)
336 		opal_poll_events(NULL);
337 }
338 
339 static void __noreturn pnv_halt(void)
340 {
341 	pnv_power_off();
342 }
343 
344 static void pnv_progress(char *s, unsigned short hex)
345 {
346 }
347 
348 static void pnv_shutdown(void)
349 {
350 	/* Let the PCI code clear up IODA tables */
351 	pnv_pci_shutdown();
352 
353 	/*
354 	 * Stop OPAL activity: Unregister all OPAL interrupts so they
355 	 * don't fire up while we kexec and make sure all potentially
356 	 * DMA'ing ops are complete (such as dump retrieval).
357 	 */
358 	opal_shutdown();
359 }
360 
361 #ifdef CONFIG_KEXEC_CORE
362 static void pnv_kexec_wait_secondaries_down(void)
363 {
364 	int my_cpu, i, notified = -1;
365 
366 	my_cpu = get_cpu();
367 
368 	for_each_online_cpu(i) {
369 		uint8_t status;
370 		int64_t rc, timeout = 1000;
371 
372 		if (i == my_cpu)
373 			continue;
374 
375 		for (;;) {
376 			rc = opal_query_cpu_status(get_hard_smp_processor_id(i),
377 						   &status);
378 			if (rc != OPAL_SUCCESS || status != OPAL_THREAD_STARTED)
379 				break;
380 			barrier();
381 			if (i != notified) {
382 				printk(KERN_INFO "kexec: waiting for cpu %d "
383 				       "(physical %d) to enter OPAL\n",
384 				       i, paca_ptrs[i]->hw_cpu_id);
385 				notified = i;
386 			}
387 
388 			/*
389 			 * On crash secondaries might be unreachable or hung,
390 			 * so timeout if we've waited too long
391 			 * */
392 			mdelay(1);
393 			if (timeout-- == 0) {
394 				printk(KERN_ERR "kexec: timed out waiting for "
395 				       "cpu %d (physical %d) to enter OPAL\n",
396 				       i, paca_ptrs[i]->hw_cpu_id);
397 				break;
398 			}
399 		}
400 	}
401 }
402 
403 static void pnv_kexec_cpu_down(int crash_shutdown, int secondary)
404 {
405 	u64 reinit_flags;
406 
407 	if (xive_enabled())
408 		xive_teardown_cpu();
409 	else
410 		xics_kexec_teardown_cpu(secondary);
411 
412 	/* On OPAL, we return all CPUs to firmware */
413 	if (!firmware_has_feature(FW_FEATURE_OPAL))
414 		return;
415 
416 	if (secondary) {
417 		/* Return secondary CPUs to firmware on OPAL v3 */
418 		mb();
419 		get_paca()->kexec_state = KEXEC_STATE_REAL_MODE;
420 		mb();
421 
422 		/* Return the CPU to OPAL */
423 		opal_return_cpu();
424 	} else {
425 		/* Primary waits for the secondaries to have reached OPAL */
426 		pnv_kexec_wait_secondaries_down();
427 
428 		/* Switch XIVE back to emulation mode */
429 		if (xive_enabled())
430 			xive_shutdown();
431 
432 		/*
433 		 * We might be running as little-endian - now that interrupts
434 		 * are disabled, reset the HILE bit to big-endian so we don't
435 		 * take interrupts in the wrong endian later
436 		 *
437 		 * We reinit to enable both radix and hash on P9 to ensure
438 		 * the mode used by the next kernel is always supported.
439 		 */
440 		reinit_flags = OPAL_REINIT_CPUS_HILE_BE;
441 		if (cpu_has_feature(CPU_FTR_ARCH_300))
442 			reinit_flags |= OPAL_REINIT_CPUS_MMU_RADIX |
443 				OPAL_REINIT_CPUS_MMU_HASH;
444 		opal_reinit_cpus(reinit_flags);
445 	}
446 }
447 #endif /* CONFIG_KEXEC_CORE */
448 
449 #ifdef CONFIG_MEMORY_HOTPLUG
450 static unsigned long pnv_memory_block_size(void)
451 {
452 	/*
453 	 * We map the kernel linear region with 1GB large pages on radix. For
454 	 * memory hot unplug to work our memory block size must be at least
455 	 * this size.
456 	 */
457 	if (radix_enabled())
458 		return radix_mem_block_size;
459 	else
460 		return 256UL * 1024 * 1024;
461 }
462 #endif
463 
464 static void __init pnv_setup_machdep_opal(void)
465 {
466 	ppc_md.get_boot_time = opal_get_boot_time;
467 	ppc_md.restart = pnv_restart;
468 	pm_power_off = pnv_power_off;
469 	ppc_md.halt = pnv_halt;
470 	/* ppc_md.system_reset_exception gets filled in by pnv_smp_init() */
471 	ppc_md.machine_check_exception = opal_machine_check;
472 	ppc_md.mce_check_early_recovery = opal_mce_check_early_recovery;
473 	if (opal_check_token(OPAL_HANDLE_HMI2))
474 		ppc_md.hmi_exception_early = opal_hmi_exception_early2;
475 	else
476 		ppc_md.hmi_exception_early = opal_hmi_exception_early;
477 	ppc_md.handle_hmi_exception = opal_handle_hmi_exception;
478 }
479 
480 static int __init pnv_probe(void)
481 {
482 	if (!of_machine_is_compatible("ibm,powernv"))
483 		return 0;
484 
485 	if (firmware_has_feature(FW_FEATURE_OPAL))
486 		pnv_setup_machdep_opal();
487 
488 	pr_debug("PowerNV detected !\n");
489 
490 	pnv_init();
491 
492 	return 1;
493 }
494 
495 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
496 void __init pnv_tm_init(void)
497 {
498 	if (!firmware_has_feature(FW_FEATURE_OPAL) ||
499 	    !pvr_version_is(PVR_POWER9) ||
500 	    early_cpu_has_feature(CPU_FTR_TM))
501 		return;
502 
503 	if (opal_reinit_cpus(OPAL_REINIT_CPUS_TM_SUSPEND_DISABLED) != OPAL_SUCCESS)
504 		return;
505 
506 	pr_info("Enabling TM (Transactional Memory) with Suspend Disabled\n");
507 	cur_cpu_spec->cpu_features |= CPU_FTR_TM;
508 	/* Make sure "normal" HTM is off (it should be) */
509 	cur_cpu_spec->cpu_user_features2 &= ~PPC_FEATURE2_HTM;
510 	/* Turn on no suspend mode, and HTM no SC */
511 	cur_cpu_spec->cpu_user_features2 |= PPC_FEATURE2_HTM_NO_SUSPEND | \
512 					    PPC_FEATURE2_HTM_NOSC;
513 	tm_suspend_disabled = true;
514 }
515 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
516 
517 /*
518  * Returns the cpu frequency for 'cpu' in Hz. This is used by
519  * /proc/cpuinfo
520  */
521 static unsigned long pnv_get_proc_freq(unsigned int cpu)
522 {
523 	unsigned long ret_freq;
524 
525 	ret_freq = cpufreq_get(cpu) * 1000ul;
526 
527 	/*
528 	 * If the backend cpufreq driver does not exist,
529          * then fallback to old way of reporting the clockrate.
530 	 */
531 	if (!ret_freq)
532 		ret_freq = ppc_proc_freq;
533 	return ret_freq;
534 }
535 
536 static long pnv_machine_check_early(struct pt_regs *regs)
537 {
538 	long handled = 0;
539 
540 	if (cur_cpu_spec && cur_cpu_spec->machine_check_early)
541 		handled = cur_cpu_spec->machine_check_early(regs);
542 
543 	return handled;
544 }
545 
546 define_machine(powernv) {
547 	.name			= "PowerNV",
548 	.probe			= pnv_probe,
549 	.setup_arch		= pnv_setup_arch,
550 	.init_IRQ		= pnv_init_IRQ,
551 	.show_cpuinfo		= pnv_show_cpuinfo,
552 	.get_proc_freq          = pnv_get_proc_freq,
553 	.discover_phbs		= pnv_pci_init,
554 	.progress		= pnv_progress,
555 	.machine_shutdown	= pnv_shutdown,
556 	.power_save             = NULL,
557 	.calibrate_decr		= generic_calibrate_decr,
558 	.machine_check_early	= pnv_machine_check_early,
559 #ifdef CONFIG_KEXEC_CORE
560 	.kexec_cpu_down		= pnv_kexec_cpu_down,
561 #endif
562 #ifdef CONFIG_MEMORY_HOTPLUG
563 	.memory_block_size	= pnv_memory_block_size,
564 #endif
565 };
566