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_early(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_OPALv3))
94 		seq_printf(m, "firmware\t: OPAL v3\n");
95 	else if (firmware_has_feature(FW_FEATURE_OPALv2))
96 		seq_printf(m, "firmware\t: OPAL v2\n");
97 	else if (firmware_has_feature(FW_FEATURE_OPAL))
98 		seq_printf(m, "firmware\t: OPAL v1\n");
99 	else
100 		seq_printf(m, "firmware\t: BML\n");
101 	of_node_put(root);
102 }
103 
104 static void pnv_prepare_going_down(void)
105 {
106 	/*
107 	 * Disable all notifiers from OPAL, we can't
108 	 * service interrupts anymore anyway
109 	 */
110 	opal_event_shutdown();
111 
112 	/* Soft disable interrupts */
113 	local_irq_disable();
114 
115 	/*
116 	 * Return secondary CPUs to firwmare if a flash update
117 	 * is pending otherwise we will get all sort of error
118 	 * messages about CPU being stuck etc.. This will also
119 	 * have the side effect of hard disabling interrupts so
120 	 * past this point, the kernel is effectively dead.
121 	 */
122 	opal_flash_term_callback();
123 }
124 
125 static void  __noreturn pnv_restart(char *cmd)
126 {
127 	long rc = OPAL_BUSY;
128 
129 	pnv_prepare_going_down();
130 
131 	while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
132 		rc = opal_cec_reboot();
133 		if (rc == OPAL_BUSY_EVENT)
134 			opal_poll_events(NULL);
135 		else
136 			mdelay(10);
137 	}
138 	for (;;)
139 		opal_poll_events(NULL);
140 }
141 
142 static void __noreturn pnv_power_off(void)
143 {
144 	long rc = OPAL_BUSY;
145 
146 	pnv_prepare_going_down();
147 
148 	while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
149 		rc = opal_cec_power_down(0);
150 		if (rc == OPAL_BUSY_EVENT)
151 			opal_poll_events(NULL);
152 		else
153 			mdelay(10);
154 	}
155 	for (;;)
156 		opal_poll_events(NULL);
157 }
158 
159 static void __noreturn pnv_halt(void)
160 {
161 	pnv_power_off();
162 }
163 
164 static void pnv_progress(char *s, unsigned short hex)
165 {
166 }
167 
168 static u64 pnv_dma_get_required_mask(struct device *dev)
169 {
170 	if (dev_is_pci(dev))
171 		return pnv_pci_dma_get_required_mask(to_pci_dev(dev));
172 
173 	return __dma_get_required_mask(dev);
174 }
175 
176 static void pnv_shutdown(void)
177 {
178 	/* Let the PCI code clear up IODA tables */
179 	pnv_pci_shutdown();
180 
181 	/*
182 	 * Stop OPAL activity: Unregister all OPAL interrupts so they
183 	 * don't fire up while we kexec and make sure all potentially
184 	 * DMA'ing ops are complete (such as dump retrieval).
185 	 */
186 	opal_shutdown();
187 }
188 
189 #ifdef CONFIG_KEXEC
190 static void pnv_kexec_wait_secondaries_down(void)
191 {
192 	int my_cpu, i, notified = -1;
193 
194 	my_cpu = get_cpu();
195 
196 	for_each_online_cpu(i) {
197 		uint8_t status;
198 		int64_t rc;
199 
200 		if (i == my_cpu)
201 			continue;
202 
203 		for (;;) {
204 			rc = opal_query_cpu_status(get_hard_smp_processor_id(i),
205 						   &status);
206 			if (rc != OPAL_SUCCESS || status != OPAL_THREAD_STARTED)
207 				break;
208 			barrier();
209 			if (i != notified) {
210 				printk(KERN_INFO "kexec: waiting for cpu %d "
211 				       "(physical %d) to enter OPAL\n",
212 				       i, paca[i].hw_cpu_id);
213 				notified = i;
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 v3, we return all CPUs to firmware */
224 
225 	if (!firmware_has_feature(FW_FEATURE_OPALv3))
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 if (crash_shutdown) {
237 		/*
238 		 * On crash, we don't wait for secondaries to go
239 		 * down as they might be unreachable or hung, so
240 		 * instead we just wait a bit and move on.
241 		 */
242 		mdelay(1);
243 	} else {
244 		/* Primary waits for the secondaries to have reached OPAL */
245 		pnv_kexec_wait_secondaries_down();
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 	unsigned long root = of_get_flat_dt_root();
272 
273 	if (!of_flat_dt_is_compatible(root, "ibm,powernv"))
274 		return 0;
275 
276 	hpte_init_native();
277 
278 	if (firmware_has_feature(FW_FEATURE_OPAL))
279 		pnv_setup_machdep_opal();
280 
281 	pr_debug("PowerNV detected !\n");
282 
283 	return 1;
284 }
285 
286 /*
287  * Returns the cpu frequency for 'cpu' in Hz. This is used by
288  * /proc/cpuinfo
289  */
290 static unsigned long pnv_get_proc_freq(unsigned int cpu)
291 {
292 	unsigned long ret_freq;
293 
294 	ret_freq = cpufreq_quick_get(cpu) * 1000ul;
295 
296 	/*
297 	 * If the backend cpufreq driver does not exist,
298          * then fallback to old way of reporting the clockrate.
299 	 */
300 	if (!ret_freq)
301 		ret_freq = ppc_proc_freq;
302 	return ret_freq;
303 }
304 
305 define_machine(powernv) {
306 	.name			= "PowerNV",
307 	.probe			= pnv_probe,
308 	.init_early		= pnv_init_early,
309 	.setup_arch		= pnv_setup_arch,
310 	.init_IRQ		= pnv_init_IRQ,
311 	.show_cpuinfo		= pnv_show_cpuinfo,
312 	.get_proc_freq          = pnv_get_proc_freq,
313 	.progress		= pnv_progress,
314 	.machine_shutdown	= pnv_shutdown,
315 	.power_save             = power7_idle,
316 	.calibrate_decr		= generic_calibrate_decr,
317 	.dma_get_required_mask	= pnv_dma_get_required_mask,
318 #ifdef CONFIG_KEXEC
319 	.kexec_cpu_down		= pnv_kexec_cpu_down,
320 #endif
321 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
322 	.memory_block_size	= pnv_memory_block_size,
323 #endif
324 };
325