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 }
128 
129 static void __init pnv_setup_arch(void)
130 {
131 	set_arch_panic_timeout(10, ARCH_PANIC_TIMEOUT);
132 
133 	pnv_setup_rfi_flush();
134 
135 	/* Initialize SMP */
136 	pnv_smp_init();
137 
138 	/* Setup PCI */
139 	pnv_pci_init();
140 
141 	/* Setup RTC and NVRAM callbacks */
142 	if (firmware_has_feature(FW_FEATURE_OPAL))
143 		opal_nvram_init();
144 
145 	/* Enable NAP mode */
146 	powersave_nap = 1;
147 
148 	/* XXX PMCS */
149 }
150 
151 static void __init pnv_init(void)
152 {
153 	/*
154 	 * Initialize the LPC bus now so that legacy serial
155 	 * ports can be found on it
156 	 */
157 	opal_lpc_init();
158 
159 #ifdef CONFIG_HVC_OPAL
160 	if (firmware_has_feature(FW_FEATURE_OPAL))
161 		hvc_opal_init_early();
162 	else
163 #endif
164 		add_preferred_console("hvc", 0, NULL);
165 }
166 
167 static void __init pnv_init_IRQ(void)
168 {
169 	/* Try using a XIVE if available, otherwise use a XICS */
170 	if (!xive_native_init())
171 		xics_init();
172 
173 	WARN_ON(!ppc_md.get_irq);
174 }
175 
176 static void pnv_show_cpuinfo(struct seq_file *m)
177 {
178 	struct device_node *root;
179 	const char *model = "";
180 
181 	root = of_find_node_by_path("/");
182 	if (root)
183 		model = of_get_property(root, "model", NULL);
184 	seq_printf(m, "machine\t\t: PowerNV %s\n", model);
185 	if (firmware_has_feature(FW_FEATURE_OPAL))
186 		seq_printf(m, "firmware\t: OPAL\n");
187 	else
188 		seq_printf(m, "firmware\t: BML\n");
189 	of_node_put(root);
190 	if (radix_enabled())
191 		seq_printf(m, "MMU\t\t: Radix\n");
192 	else
193 		seq_printf(m, "MMU\t\t: Hash\n");
194 }
195 
196 static void pnv_prepare_going_down(void)
197 {
198 	/*
199 	 * Disable all notifiers from OPAL, we can't
200 	 * service interrupts anymore anyway
201 	 */
202 	opal_event_shutdown();
203 
204 	/* Print flash update message if one is scheduled. */
205 	opal_flash_update_print_message();
206 
207 	smp_send_stop();
208 
209 	hard_irq_disable();
210 }
211 
212 static void  __noreturn pnv_restart(char *cmd)
213 {
214 	long rc = OPAL_BUSY;
215 
216 	pnv_prepare_going_down();
217 
218 	while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
219 		rc = opal_cec_reboot();
220 		if (rc == OPAL_BUSY_EVENT)
221 			opal_poll_events(NULL);
222 		else
223 			mdelay(10);
224 	}
225 	for (;;)
226 		opal_poll_events(NULL);
227 }
228 
229 static void __noreturn pnv_power_off(void)
230 {
231 	long rc = OPAL_BUSY;
232 
233 	pnv_prepare_going_down();
234 
235 	while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
236 		rc = opal_cec_power_down(0);
237 		if (rc == OPAL_BUSY_EVENT)
238 			opal_poll_events(NULL);
239 		else
240 			mdelay(10);
241 	}
242 	for (;;)
243 		opal_poll_events(NULL);
244 }
245 
246 static void __noreturn pnv_halt(void)
247 {
248 	pnv_power_off();
249 }
250 
251 static void pnv_progress(char *s, unsigned short hex)
252 {
253 }
254 
255 static void pnv_shutdown(void)
256 {
257 	/* Let the PCI code clear up IODA tables */
258 	pnv_pci_shutdown();
259 
260 	/*
261 	 * Stop OPAL activity: Unregister all OPAL interrupts so they
262 	 * don't fire up while we kexec and make sure all potentially
263 	 * DMA'ing ops are complete (such as dump retrieval).
264 	 */
265 	opal_shutdown();
266 }
267 
268 #ifdef CONFIG_KEXEC_CORE
269 static void pnv_kexec_wait_secondaries_down(void)
270 {
271 	int my_cpu, i, notified = -1;
272 
273 	my_cpu = get_cpu();
274 
275 	for_each_online_cpu(i) {
276 		uint8_t status;
277 		int64_t rc, timeout = 1000;
278 
279 		if (i == my_cpu)
280 			continue;
281 
282 		for (;;) {
283 			rc = opal_query_cpu_status(get_hard_smp_processor_id(i),
284 						   &status);
285 			if (rc != OPAL_SUCCESS || status != OPAL_THREAD_STARTED)
286 				break;
287 			barrier();
288 			if (i != notified) {
289 				printk(KERN_INFO "kexec: waiting for cpu %d "
290 				       "(physical %d) to enter OPAL\n",
291 				       i, paca_ptrs[i]->hw_cpu_id);
292 				notified = i;
293 			}
294 
295 			/*
296 			 * On crash secondaries might be unreachable or hung,
297 			 * so timeout if we've waited too long
298 			 * */
299 			mdelay(1);
300 			if (timeout-- == 0) {
301 				printk(KERN_ERR "kexec: timed out waiting for "
302 				       "cpu %d (physical %d) to enter OPAL\n",
303 				       i, paca_ptrs[i]->hw_cpu_id);
304 				break;
305 			}
306 		}
307 	}
308 }
309 
310 static void pnv_kexec_cpu_down(int crash_shutdown, int secondary)
311 {
312 	u64 reinit_flags;
313 
314 	if (xive_enabled())
315 		xive_kexec_teardown_cpu(secondary);
316 	else
317 		xics_kexec_teardown_cpu(secondary);
318 
319 	/* On OPAL, we return all CPUs to firmware */
320 	if (!firmware_has_feature(FW_FEATURE_OPAL))
321 		return;
322 
323 	if (secondary) {
324 		/* Return secondary CPUs to firmware on OPAL v3 */
325 		mb();
326 		get_paca()->kexec_state = KEXEC_STATE_REAL_MODE;
327 		mb();
328 
329 		/* Return the CPU to OPAL */
330 		opal_return_cpu();
331 	} else {
332 		/* Primary waits for the secondaries to have reached OPAL */
333 		pnv_kexec_wait_secondaries_down();
334 
335 		/* Switch XIVE back to emulation mode */
336 		if (xive_enabled())
337 			xive_shutdown();
338 
339 		/*
340 		 * We might be running as little-endian - now that interrupts
341 		 * are disabled, reset the HILE bit to big-endian so we don't
342 		 * take interrupts in the wrong endian later
343 		 *
344 		 * We reinit to enable both radix and hash on P9 to ensure
345 		 * the mode used by the next kernel is always supported.
346 		 */
347 		reinit_flags = OPAL_REINIT_CPUS_HILE_BE;
348 		if (cpu_has_feature(CPU_FTR_ARCH_300))
349 			reinit_flags |= OPAL_REINIT_CPUS_MMU_RADIX |
350 				OPAL_REINIT_CPUS_MMU_HASH;
351 		opal_reinit_cpus(reinit_flags);
352 	}
353 }
354 #endif /* CONFIG_KEXEC_CORE */
355 
356 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
357 static unsigned long pnv_memory_block_size(void)
358 {
359 	/*
360 	 * We map the kernel linear region with 1GB large pages on radix. For
361 	 * memory hot unplug to work our memory block size must be at least
362 	 * this size.
363 	 */
364 	if (radix_enabled())
365 		return 1UL * 1024 * 1024 * 1024;
366 	else
367 		return 256UL * 1024 * 1024;
368 }
369 #endif
370 
371 static void __init pnv_setup_machdep_opal(void)
372 {
373 	ppc_md.get_boot_time = opal_get_boot_time;
374 	ppc_md.restart = pnv_restart;
375 	pm_power_off = pnv_power_off;
376 	ppc_md.halt = pnv_halt;
377 	/* ppc_md.system_reset_exception gets filled in by pnv_smp_init() */
378 	ppc_md.machine_check_exception = opal_machine_check;
379 	ppc_md.mce_check_early_recovery = opal_mce_check_early_recovery;
380 	ppc_md.hmi_exception_early = opal_hmi_exception_early;
381 	ppc_md.handle_hmi_exception = opal_handle_hmi_exception;
382 }
383 
384 static int __init pnv_probe(void)
385 {
386 	if (!of_machine_is_compatible("ibm,powernv"))
387 		return 0;
388 
389 	if (firmware_has_feature(FW_FEATURE_OPAL))
390 		pnv_setup_machdep_opal();
391 
392 	pr_debug("PowerNV detected !\n");
393 
394 	pnv_init();
395 
396 	return 1;
397 }
398 
399 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
400 void __init pnv_tm_init(void)
401 {
402 	if (!firmware_has_feature(FW_FEATURE_OPAL) ||
403 	    !pvr_version_is(PVR_POWER9) ||
404 	    early_cpu_has_feature(CPU_FTR_TM))
405 		return;
406 
407 	if (opal_reinit_cpus(OPAL_REINIT_CPUS_TM_SUSPEND_DISABLED) != OPAL_SUCCESS)
408 		return;
409 
410 	pr_info("Enabling TM (Transactional Memory) with Suspend Disabled\n");
411 	cur_cpu_spec->cpu_features |= CPU_FTR_TM;
412 	/* Make sure "normal" HTM is off (it should be) */
413 	cur_cpu_spec->cpu_user_features2 &= ~PPC_FEATURE2_HTM;
414 	/* Turn on no suspend mode, and HTM no SC */
415 	cur_cpu_spec->cpu_user_features2 |= PPC_FEATURE2_HTM_NO_SUSPEND | \
416 					    PPC_FEATURE2_HTM_NOSC;
417 	tm_suspend_disabled = true;
418 }
419 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
420 
421 /*
422  * Returns the cpu frequency for 'cpu' in Hz. This is used by
423  * /proc/cpuinfo
424  */
425 static unsigned long pnv_get_proc_freq(unsigned int cpu)
426 {
427 	unsigned long ret_freq;
428 
429 	ret_freq = cpufreq_get(cpu) * 1000ul;
430 
431 	/*
432 	 * If the backend cpufreq driver does not exist,
433          * then fallback to old way of reporting the clockrate.
434 	 */
435 	if (!ret_freq)
436 		ret_freq = ppc_proc_freq;
437 	return ret_freq;
438 }
439 
440 define_machine(powernv) {
441 	.name			= "PowerNV",
442 	.probe			= pnv_probe,
443 	.setup_arch		= pnv_setup_arch,
444 	.init_IRQ		= pnv_init_IRQ,
445 	.show_cpuinfo		= pnv_show_cpuinfo,
446 	.get_proc_freq          = pnv_get_proc_freq,
447 	.progress		= pnv_progress,
448 	.machine_shutdown	= pnv_shutdown,
449 	.power_save             = NULL,
450 	.calibrate_decr		= generic_calibrate_decr,
451 #ifdef CONFIG_KEXEC_CORE
452 	.kexec_cpu_down		= pnv_kexec_cpu_down,
453 #endif
454 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
455 	.memory_block_size	= pnv_memory_block_size,
456 #endif
457 };
458