1 /*
2  * PowerNV setup code.
3  *
4  * Copyright 2011 IBM Corp.
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License
8  * as published by the Free Software Foundation; either version
9  * 2 of the License, or (at your option) any later version.
10  */
11 
12 #undef DEBUG
13 
14 #include <linux/cpu.h>
15 #include <linux/errno.h>
16 #include <linux/sched.h>
17 #include <linux/kernel.h>
18 #include <linux/tty.h>
19 #include <linux/reboot.h>
20 #include <linux/init.h>
21 #include <linux/console.h>
22 #include <linux/delay.h>
23 #include <linux/irq.h>
24 #include <linux/seq_file.h>
25 #include <linux/of.h>
26 #include <linux/of_fdt.h>
27 #include <linux/interrupt.h>
28 #include <linux/bug.h>
29 #include <linux/pci.h>
30 #include <linux/cpufreq.h>
31 
32 #include <asm/machdep.h>
33 #include <asm/firmware.h>
34 #include <asm/xics.h>
35 #include <asm/xive.h>
36 #include <asm/opal.h>
37 #include <asm/kexec.h>
38 #include <asm/smp.h>
39 #include <asm/tm.h>
40 #include <asm/setup.h>
41 #include <asm/security_features.h>
42 
43 #include "powernv.h"
44 
45 
46 static bool fw_feature_is(const char *state, const char *name,
47 			  struct device_node *fw_features)
48 {
49 	struct device_node *np;
50 	bool rc = false;
51 
52 	np = of_get_child_by_name(fw_features, name);
53 	if (np) {
54 		rc = of_property_read_bool(np, state);
55 		of_node_put(np);
56 	}
57 
58 	return rc;
59 }
60 
61 static void init_fw_feat_flags(struct device_node *np)
62 {
63 	if (fw_feature_is("enabled", "inst-spec-barrier-ori31,31,0", np))
64 		security_ftr_set(SEC_FTR_SPEC_BAR_ORI31);
65 
66 	if (fw_feature_is("enabled", "fw-bcctrl-serialized", np))
67 		security_ftr_set(SEC_FTR_BCCTRL_SERIALISED);
68 
69 	if (fw_feature_is("enabled", "inst-l1d-flush-ori30,30,0", np))
70 		security_ftr_set(SEC_FTR_L1D_FLUSH_ORI30);
71 
72 	if (fw_feature_is("enabled", "inst-l1d-flush-trig2", np))
73 		security_ftr_set(SEC_FTR_L1D_FLUSH_TRIG2);
74 
75 	if (fw_feature_is("enabled", "fw-l1d-thread-split", np))
76 		security_ftr_set(SEC_FTR_L1D_THREAD_PRIV);
77 
78 	if (fw_feature_is("enabled", "fw-count-cache-disabled", np))
79 		security_ftr_set(SEC_FTR_COUNT_CACHE_DISABLED);
80 
81 	/*
82 	 * The features below are enabled by default, so we instead look to see
83 	 * if firmware has *disabled* them, and clear them if so.
84 	 */
85 	if (fw_feature_is("disabled", "speculation-policy-favor-security", np))
86 		security_ftr_clear(SEC_FTR_FAVOUR_SECURITY);
87 
88 	if (fw_feature_is("disabled", "needs-l1d-flush-msr-pr-0-to-1", np))
89 		security_ftr_clear(SEC_FTR_L1D_FLUSH_PR);
90 
91 	if (fw_feature_is("disabled", "needs-l1d-flush-msr-hv-1-to-0", np))
92 		security_ftr_clear(SEC_FTR_L1D_FLUSH_HV);
93 
94 	if (fw_feature_is("disabled", "needs-spec-barrier-for-bound-checks", np))
95 		security_ftr_clear(SEC_FTR_BNDS_CHK_SPEC_BAR);
96 }
97 
98 static void pnv_setup_rfi_flush(void)
99 {
100 	struct device_node *np, *fw_features;
101 	enum l1d_flush_type type;
102 	bool enable;
103 
104 	/* Default to fallback in case fw-features are not available */
105 	type = L1D_FLUSH_FALLBACK;
106 
107 	np = of_find_node_by_name(NULL, "ibm,opal");
108 	fw_features = of_get_child_by_name(np, "fw-features");
109 	of_node_put(np);
110 
111 	if (fw_features) {
112 		init_fw_feat_flags(fw_features);
113 		of_node_put(fw_features);
114 
115 		if (security_ftr_enabled(SEC_FTR_L1D_FLUSH_TRIG2))
116 			type = L1D_FLUSH_MTTRIG;
117 
118 		if (security_ftr_enabled(SEC_FTR_L1D_FLUSH_ORI30))
119 			type = L1D_FLUSH_ORI;
120 	}
121 
122 	enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) && \
123 		 (security_ftr_enabled(SEC_FTR_L1D_FLUSH_PR)   || \
124 		  security_ftr_enabled(SEC_FTR_L1D_FLUSH_HV));
125 
126 	setup_rfi_flush(type, enable);
127 	setup_barrier_nospec();
128 }
129 
130 static void __init pnv_setup_arch(void)
131 {
132 	set_arch_panic_timeout(10, ARCH_PANIC_TIMEOUT);
133 
134 	pnv_setup_rfi_flush();
135 	setup_stf_barrier();
136 
137 	/* Initialize SMP */
138 	pnv_smp_init();
139 
140 	/* Setup PCI */
141 	pnv_pci_init();
142 
143 	/* Setup RTC and NVRAM callbacks */
144 	if (firmware_has_feature(FW_FEATURE_OPAL))
145 		opal_nvram_init();
146 
147 	/* Enable NAP mode */
148 	powersave_nap = 1;
149 
150 	/* XXX PMCS */
151 }
152 
153 static void __init pnv_init(void)
154 {
155 	/*
156 	 * Initialize the LPC bus now so that legacy serial
157 	 * ports can be found on it
158 	 */
159 	opal_lpc_init();
160 
161 #ifdef CONFIG_HVC_OPAL
162 	if (firmware_has_feature(FW_FEATURE_OPAL))
163 		hvc_opal_init_early();
164 	else
165 #endif
166 		add_preferred_console("hvc", 0, NULL);
167 }
168 
169 static void __init pnv_init_IRQ(void)
170 {
171 	/* Try using a XIVE if available, otherwise use a XICS */
172 	if (!xive_native_init())
173 		xics_init();
174 
175 	WARN_ON(!ppc_md.get_irq);
176 }
177 
178 static void pnv_show_cpuinfo(struct seq_file *m)
179 {
180 	struct device_node *root;
181 	const char *model = "";
182 
183 	root = of_find_node_by_path("/");
184 	if (root)
185 		model = of_get_property(root, "model", NULL);
186 	seq_printf(m, "machine\t\t: PowerNV %s\n", model);
187 	if (firmware_has_feature(FW_FEATURE_OPAL))
188 		seq_printf(m, "firmware\t: OPAL\n");
189 	else
190 		seq_printf(m, "firmware\t: BML\n");
191 	of_node_put(root);
192 	if (radix_enabled())
193 		seq_printf(m, "MMU\t\t: Radix\n");
194 	else
195 		seq_printf(m, "MMU\t\t: Hash\n");
196 }
197 
198 static void pnv_prepare_going_down(void)
199 {
200 	/*
201 	 * Disable all notifiers from OPAL, we can't
202 	 * service interrupts anymore anyway
203 	 */
204 	opal_event_shutdown();
205 
206 	/* Print flash update message if one is scheduled. */
207 	opal_flash_update_print_message();
208 
209 	smp_send_stop();
210 
211 	hard_irq_disable();
212 }
213 
214 static void  __noreturn pnv_restart(char *cmd)
215 {
216 	long rc = OPAL_BUSY;
217 
218 	pnv_prepare_going_down();
219 
220 	while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
221 		rc = opal_cec_reboot();
222 		if (rc == OPAL_BUSY_EVENT)
223 			opal_poll_events(NULL);
224 		else
225 			mdelay(10);
226 	}
227 	for (;;)
228 		opal_poll_events(NULL);
229 }
230 
231 static void __noreturn pnv_power_off(void)
232 {
233 	long rc = OPAL_BUSY;
234 
235 	pnv_prepare_going_down();
236 
237 	while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
238 		rc = opal_cec_power_down(0);
239 		if (rc == OPAL_BUSY_EVENT)
240 			opal_poll_events(NULL);
241 		else
242 			mdelay(10);
243 	}
244 	for (;;)
245 		opal_poll_events(NULL);
246 }
247 
248 static void __noreturn pnv_halt(void)
249 {
250 	pnv_power_off();
251 }
252 
253 static void pnv_progress(char *s, unsigned short hex)
254 {
255 }
256 
257 static void pnv_shutdown(void)
258 {
259 	/* Let the PCI code clear up IODA tables */
260 	pnv_pci_shutdown();
261 
262 	/*
263 	 * Stop OPAL activity: Unregister all OPAL interrupts so they
264 	 * don't fire up while we kexec and make sure all potentially
265 	 * DMA'ing ops are complete (such as dump retrieval).
266 	 */
267 	opal_shutdown();
268 }
269 
270 #ifdef CONFIG_KEXEC_CORE
271 static void pnv_kexec_wait_secondaries_down(void)
272 {
273 	int my_cpu, i, notified = -1;
274 
275 	my_cpu = get_cpu();
276 
277 	for_each_online_cpu(i) {
278 		uint8_t status;
279 		int64_t rc, timeout = 1000;
280 
281 		if (i == my_cpu)
282 			continue;
283 
284 		for (;;) {
285 			rc = opal_query_cpu_status(get_hard_smp_processor_id(i),
286 						   &status);
287 			if (rc != OPAL_SUCCESS || status != OPAL_THREAD_STARTED)
288 				break;
289 			barrier();
290 			if (i != notified) {
291 				printk(KERN_INFO "kexec: waiting for cpu %d "
292 				       "(physical %d) to enter OPAL\n",
293 				       i, paca_ptrs[i]->hw_cpu_id);
294 				notified = i;
295 			}
296 
297 			/*
298 			 * On crash secondaries might be unreachable or hung,
299 			 * so timeout if we've waited too long
300 			 * */
301 			mdelay(1);
302 			if (timeout-- == 0) {
303 				printk(KERN_ERR "kexec: timed out waiting for "
304 				       "cpu %d (physical %d) to enter OPAL\n",
305 				       i, paca_ptrs[i]->hw_cpu_id);
306 				break;
307 			}
308 		}
309 	}
310 }
311 
312 static void pnv_kexec_cpu_down(int crash_shutdown, int secondary)
313 {
314 	u64 reinit_flags;
315 
316 	if (xive_enabled())
317 		xive_kexec_teardown_cpu(secondary);
318 	else
319 		xics_kexec_teardown_cpu(secondary);
320 
321 	/* On OPAL, we return all CPUs to firmware */
322 	if (!firmware_has_feature(FW_FEATURE_OPAL))
323 		return;
324 
325 	if (secondary) {
326 		/* Return secondary CPUs to firmware on OPAL v3 */
327 		mb();
328 		get_paca()->kexec_state = KEXEC_STATE_REAL_MODE;
329 		mb();
330 
331 		/* Return the CPU to OPAL */
332 		opal_return_cpu();
333 	} else {
334 		/* Primary waits for the secondaries to have reached OPAL */
335 		pnv_kexec_wait_secondaries_down();
336 
337 		/* Switch XIVE back to emulation mode */
338 		if (xive_enabled())
339 			xive_shutdown();
340 
341 		/*
342 		 * We might be running as little-endian - now that interrupts
343 		 * are disabled, reset the HILE bit to big-endian so we don't
344 		 * take interrupts in the wrong endian later
345 		 *
346 		 * We reinit to enable both radix and hash on P9 to ensure
347 		 * the mode used by the next kernel is always supported.
348 		 */
349 		reinit_flags = OPAL_REINIT_CPUS_HILE_BE;
350 		if (cpu_has_feature(CPU_FTR_ARCH_300))
351 			reinit_flags |= OPAL_REINIT_CPUS_MMU_RADIX |
352 				OPAL_REINIT_CPUS_MMU_HASH;
353 		opal_reinit_cpus(reinit_flags);
354 	}
355 }
356 #endif /* CONFIG_KEXEC_CORE */
357 
358 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
359 static unsigned long pnv_memory_block_size(void)
360 {
361 	return 256UL * 1024 * 1024;
362 }
363 #endif
364 
365 static void __init pnv_setup_machdep_opal(void)
366 {
367 	ppc_md.get_boot_time = opal_get_boot_time;
368 	ppc_md.restart = pnv_restart;
369 	pm_power_off = pnv_power_off;
370 	ppc_md.halt = pnv_halt;
371 	/* ppc_md.system_reset_exception gets filled in by pnv_smp_init() */
372 	ppc_md.machine_check_exception = opal_machine_check;
373 	ppc_md.mce_check_early_recovery = opal_mce_check_early_recovery;
374 	ppc_md.hmi_exception_early = opal_hmi_exception_early;
375 	ppc_md.handle_hmi_exception = opal_handle_hmi_exception;
376 }
377 
378 static int __init pnv_probe(void)
379 {
380 	if (!of_machine_is_compatible("ibm,powernv"))
381 		return 0;
382 
383 	if (firmware_has_feature(FW_FEATURE_OPAL))
384 		pnv_setup_machdep_opal();
385 
386 	pr_debug("PowerNV detected !\n");
387 
388 	pnv_init();
389 
390 	return 1;
391 }
392 
393 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
394 void __init pnv_tm_init(void)
395 {
396 	if (!firmware_has_feature(FW_FEATURE_OPAL) ||
397 	    !pvr_version_is(PVR_POWER9) ||
398 	    early_cpu_has_feature(CPU_FTR_TM))
399 		return;
400 
401 	if (opal_reinit_cpus(OPAL_REINIT_CPUS_TM_SUSPEND_DISABLED) != OPAL_SUCCESS)
402 		return;
403 
404 	pr_info("Enabling TM (Transactional Memory) with Suspend Disabled\n");
405 	cur_cpu_spec->cpu_features |= CPU_FTR_TM;
406 	/* Make sure "normal" HTM is off (it should be) */
407 	cur_cpu_spec->cpu_user_features2 &= ~PPC_FEATURE2_HTM;
408 	/* Turn on no suspend mode, and HTM no SC */
409 	cur_cpu_spec->cpu_user_features2 |= PPC_FEATURE2_HTM_NO_SUSPEND | \
410 					    PPC_FEATURE2_HTM_NOSC;
411 	tm_suspend_disabled = true;
412 }
413 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
414 
415 /*
416  * Returns the cpu frequency for 'cpu' in Hz. This is used by
417  * /proc/cpuinfo
418  */
419 static unsigned long pnv_get_proc_freq(unsigned int cpu)
420 {
421 	unsigned long ret_freq;
422 
423 	ret_freq = cpufreq_get(cpu) * 1000ul;
424 
425 	/*
426 	 * If the backend cpufreq driver does not exist,
427          * then fallback to old way of reporting the clockrate.
428 	 */
429 	if (!ret_freq)
430 		ret_freq = ppc_proc_freq;
431 	return ret_freq;
432 }
433 
434 define_machine(powernv) {
435 	.name			= "PowerNV",
436 	.probe			= pnv_probe,
437 	.setup_arch		= pnv_setup_arch,
438 	.init_IRQ		= pnv_init_IRQ,
439 	.show_cpuinfo		= pnv_show_cpuinfo,
440 	.get_proc_freq          = pnv_get_proc_freq,
441 	.progress		= pnv_progress,
442 	.machine_shutdown	= pnv_shutdown,
443 	.power_save             = NULL,
444 	.calibrate_decr		= generic_calibrate_decr,
445 #ifdef CONFIG_KEXEC_CORE
446 	.kexec_cpu_down		= pnv_kexec_cpu_down,
447 #endif
448 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
449 	.memory_block_size	= pnv_memory_block_size,
450 #endif
451 };
452