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/opal.h>
36 #include <asm/kexec.h>
37 #include <asm/smp.h>
38 
39 #include "powernv.h"
40 
41 static void __init pnv_setup_arch(void)
42 {
43 	set_arch_panic_timeout(10, ARCH_PANIC_TIMEOUT);
44 
45 	/* Initialize SMP */
46 	pnv_smp_init();
47 
48 	/* Setup PCI */
49 	pnv_pci_init();
50 
51 	/* Setup RTC and NVRAM callbacks */
52 	if (firmware_has_feature(FW_FEATURE_OPAL))
53 		opal_nvram_init();
54 
55 	/* Enable NAP mode */
56 	powersave_nap = 1;
57 
58 	/* XXX PMCS */
59 }
60 
61 static void __init pnv_init(void)
62 {
63 	/*
64 	 * Initialize the LPC bus now so that legacy serial
65 	 * ports can be found on it
66 	 */
67 	opal_lpc_init();
68 
69 #ifdef CONFIG_HVC_OPAL
70 	if (firmware_has_feature(FW_FEATURE_OPAL))
71 		hvc_opal_init_early();
72 	else
73 #endif
74 		add_preferred_console("hvc", 0, NULL);
75 }
76 
77 static void __init pnv_init_IRQ(void)
78 {
79 	xics_init();
80 
81 	WARN_ON(!ppc_md.get_irq);
82 }
83 
84 static void pnv_show_cpuinfo(struct seq_file *m)
85 {
86 	struct device_node *root;
87 	const char *model = "";
88 
89 	root = of_find_node_by_path("/");
90 	if (root)
91 		model = of_get_property(root, "model", NULL);
92 	seq_printf(m, "machine\t\t: PowerNV %s\n", model);
93 	if (firmware_has_feature(FW_FEATURE_OPAL))
94 		seq_printf(m, "firmware\t: OPAL\n");
95 	else
96 		seq_printf(m, "firmware\t: BML\n");
97 	of_node_put(root);
98 }
99 
100 static void pnv_prepare_going_down(void)
101 {
102 	/*
103 	 * Disable all notifiers from OPAL, we can't
104 	 * service interrupts anymore anyway
105 	 */
106 	opal_event_shutdown();
107 
108 	/* Soft disable interrupts */
109 	local_irq_disable();
110 
111 	/*
112 	 * Return secondary CPUs to firwmare if a flash update
113 	 * is pending otherwise we will get all sort of error
114 	 * messages about CPU being stuck etc.. This will also
115 	 * have the side effect of hard disabling interrupts so
116 	 * past this point, the kernel is effectively dead.
117 	 */
118 	opal_flash_term_callback();
119 }
120 
121 static void  __noreturn pnv_restart(char *cmd)
122 {
123 	long rc = OPAL_BUSY;
124 
125 	pnv_prepare_going_down();
126 
127 	while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
128 		rc = opal_cec_reboot();
129 		if (rc == OPAL_BUSY_EVENT)
130 			opal_poll_events(NULL);
131 		else
132 			mdelay(10);
133 	}
134 	for (;;)
135 		opal_poll_events(NULL);
136 }
137 
138 static void __noreturn pnv_power_off(void)
139 {
140 	long rc = OPAL_BUSY;
141 
142 	pnv_prepare_going_down();
143 
144 	while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
145 		rc = opal_cec_power_down(0);
146 		if (rc == OPAL_BUSY_EVENT)
147 			opal_poll_events(NULL);
148 		else
149 			mdelay(10);
150 	}
151 	for (;;)
152 		opal_poll_events(NULL);
153 }
154 
155 static void __noreturn pnv_halt(void)
156 {
157 	pnv_power_off();
158 }
159 
160 static void pnv_progress(char *s, unsigned short hex)
161 {
162 }
163 
164 static void pnv_shutdown(void)
165 {
166 	/* Let the PCI code clear up IODA tables */
167 	pnv_pci_shutdown();
168 
169 	/*
170 	 * Stop OPAL activity: Unregister all OPAL interrupts so they
171 	 * don't fire up while we kexec and make sure all potentially
172 	 * DMA'ing ops are complete (such as dump retrieval).
173 	 */
174 	opal_shutdown();
175 }
176 
177 #ifdef CONFIG_KEXEC
178 static void pnv_kexec_wait_secondaries_down(void)
179 {
180 	int my_cpu, i, notified = -1;
181 
182 	my_cpu = get_cpu();
183 
184 	for_each_online_cpu(i) {
185 		uint8_t status;
186 		int64_t rc, timeout = 1000;
187 
188 		if (i == my_cpu)
189 			continue;
190 
191 		for (;;) {
192 			rc = opal_query_cpu_status(get_hard_smp_processor_id(i),
193 						   &status);
194 			if (rc != OPAL_SUCCESS || status != OPAL_THREAD_STARTED)
195 				break;
196 			barrier();
197 			if (i != notified) {
198 				printk(KERN_INFO "kexec: waiting for cpu %d "
199 				       "(physical %d) to enter OPAL\n",
200 				       i, paca[i].hw_cpu_id);
201 				notified = i;
202 			}
203 
204 			/*
205 			 * On crash secondaries might be unreachable or hung,
206 			 * so timeout if we've waited too long
207 			 * */
208 			mdelay(1);
209 			if (timeout-- == 0) {
210 				printk(KERN_ERR "kexec: timed out waiting for "
211 				       "cpu %d (physical %d) to enter OPAL\n",
212 				       i, paca[i].hw_cpu_id);
213 				break;
214 			}
215 		}
216 	}
217 }
218 
219 static void pnv_kexec_cpu_down(int crash_shutdown, int secondary)
220 {
221 	xics_kexec_teardown_cpu(secondary);
222 
223 	/* On OPAL, we return all CPUs to firmware */
224 
225 	if (!firmware_has_feature(FW_FEATURE_OPAL))
226 		return;
227 
228 	if (secondary) {
229 		/* Return secondary CPUs to firmware on OPAL v3 */
230 		mb();
231 		get_paca()->kexec_state = KEXEC_STATE_REAL_MODE;
232 		mb();
233 
234 		/* Return the CPU to OPAL */
235 		opal_return_cpu();
236 	} else {
237 		/* Primary waits for the secondaries to have reached OPAL */
238 		pnv_kexec_wait_secondaries_down();
239 
240 		/*
241 		 * We might be running as little-endian - now that interrupts
242 		 * are disabled, reset the HILE bit to big-endian so we don't
243 		 * take interrupts in the wrong endian later
244 		 */
245 		opal_reinit_cpus(OPAL_REINIT_CPUS_HILE_BE);
246 	}
247 }
248 #endif /* CONFIG_KEXEC */
249 
250 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
251 static unsigned long pnv_memory_block_size(void)
252 {
253 	return 256UL * 1024 * 1024;
254 }
255 #endif
256 
257 static void __init pnv_setup_machdep_opal(void)
258 {
259 	ppc_md.get_boot_time = opal_get_boot_time;
260 	ppc_md.restart = pnv_restart;
261 	pm_power_off = pnv_power_off;
262 	ppc_md.halt = pnv_halt;
263 	ppc_md.machine_check_exception = opal_machine_check;
264 	ppc_md.mce_check_early_recovery = opal_mce_check_early_recovery;
265 	ppc_md.hmi_exception_early = opal_hmi_exception_early;
266 	ppc_md.handle_hmi_exception = opal_handle_hmi_exception;
267 }
268 
269 static int __init pnv_probe(void)
270 {
271 	if (!of_machine_is_compatible("ibm,powernv"))
272 		return 0;
273 
274 	if (firmware_has_feature(FW_FEATURE_OPAL))
275 		pnv_setup_machdep_opal();
276 
277 	pr_debug("PowerNV detected !\n");
278 
279 	pnv_init();
280 
281 	return 1;
282 }
283 
284 /*
285  * Returns the cpu frequency for 'cpu' in Hz. This is used by
286  * /proc/cpuinfo
287  */
288 static unsigned long pnv_get_proc_freq(unsigned int cpu)
289 {
290 	unsigned long ret_freq;
291 
292 	ret_freq = cpufreq_quick_get(cpu) * 1000ul;
293 
294 	/*
295 	 * If the backend cpufreq driver does not exist,
296          * then fallback to old way of reporting the clockrate.
297 	 */
298 	if (!ret_freq)
299 		ret_freq = ppc_proc_freq;
300 	return ret_freq;
301 }
302 
303 define_machine(powernv) {
304 	.name			= "PowerNV",
305 	.probe			= pnv_probe,
306 	.setup_arch		= pnv_setup_arch,
307 	.init_IRQ		= pnv_init_IRQ,
308 	.show_cpuinfo		= pnv_show_cpuinfo,
309 	.get_proc_freq          = pnv_get_proc_freq,
310 	.progress		= pnv_progress,
311 	.machine_shutdown	= pnv_shutdown,
312 	.power_save             = NULL,
313 	.calibrate_decr		= generic_calibrate_decr,
314 #ifdef CONFIG_KEXEC
315 	.kexec_cpu_down		= pnv_kexec_cpu_down,
316 #endif
317 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
318 	.memory_block_size	= pnv_memory_block_size,
319 #endif
320 };
321