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/setup.h>
40 #include <asm/security_features.h>
41
42 #include "powernv.h"
43
44
fw_feature_is(const char * state,const char * name,struct device_node * fw_features)45 static bool fw_feature_is(const char *state, const char *name,
46 struct device_node *fw_features)
47 {
48 struct device_node *np;
49 bool rc = false;
50
51 np = of_get_child_by_name(fw_features, name);
52 if (np) {
53 rc = of_property_read_bool(np, state);
54 of_node_put(np);
55 }
56
57 return rc;
58 }
59
init_fw_feat_flags(struct device_node * np)60 static void init_fw_feat_flags(struct device_node *np)
61 {
62 if (fw_feature_is("enabled", "inst-spec-barrier-ori31,31,0", np))
63 security_ftr_set(SEC_FTR_SPEC_BAR_ORI31);
64
65 if (fw_feature_is("enabled", "fw-bcctrl-serialized", np))
66 security_ftr_set(SEC_FTR_BCCTRL_SERIALISED);
67
68 if (fw_feature_is("enabled", "inst-l1d-flush-ori30,30,0", np))
69 security_ftr_set(SEC_FTR_L1D_FLUSH_ORI30);
70
71 if (fw_feature_is("enabled", "inst-l1d-flush-trig2", np))
72 security_ftr_set(SEC_FTR_L1D_FLUSH_TRIG2);
73
74 if (fw_feature_is("enabled", "fw-l1d-thread-split", np))
75 security_ftr_set(SEC_FTR_L1D_THREAD_PRIV);
76
77 if (fw_feature_is("enabled", "fw-count-cache-disabled", np))
78 security_ftr_set(SEC_FTR_COUNT_CACHE_DISABLED);
79
80 if (fw_feature_is("enabled", "fw-count-cache-flush-bcctr2,0,0", np))
81 security_ftr_set(SEC_FTR_BCCTR_FLUSH_ASSIST);
82
83 if (fw_feature_is("enabled", "needs-count-cache-flush-on-context-switch", np))
84 security_ftr_set(SEC_FTR_FLUSH_COUNT_CACHE);
85
86 /*
87 * The features below are enabled by default, so we instead look to see
88 * if firmware has *disabled* them, and clear them if so.
89 */
90 if (fw_feature_is("disabled", "speculation-policy-favor-security", np))
91 security_ftr_clear(SEC_FTR_FAVOUR_SECURITY);
92
93 if (fw_feature_is("disabled", "needs-l1d-flush-msr-pr-0-to-1", np))
94 security_ftr_clear(SEC_FTR_L1D_FLUSH_PR);
95
96 if (fw_feature_is("disabled", "needs-l1d-flush-msr-hv-1-to-0", np))
97 security_ftr_clear(SEC_FTR_L1D_FLUSH_HV);
98
99 if (fw_feature_is("disabled", "needs-spec-barrier-for-bound-checks", np))
100 security_ftr_clear(SEC_FTR_BNDS_CHK_SPEC_BAR);
101 }
102
pnv_setup_rfi_flush(void)103 static void pnv_setup_rfi_flush(void)
104 {
105 struct device_node *np, *fw_features;
106 enum l1d_flush_type type;
107 bool enable;
108
109 /* Default to fallback in case fw-features are not available */
110 type = L1D_FLUSH_FALLBACK;
111
112 np = of_find_node_by_name(NULL, "ibm,opal");
113 fw_features = of_get_child_by_name(np, "fw-features");
114 of_node_put(np);
115
116 if (fw_features) {
117 init_fw_feat_flags(fw_features);
118 of_node_put(fw_features);
119
120 if (security_ftr_enabled(SEC_FTR_L1D_FLUSH_TRIG2))
121 type = L1D_FLUSH_MTTRIG;
122
123 if (security_ftr_enabled(SEC_FTR_L1D_FLUSH_ORI30))
124 type = L1D_FLUSH_ORI;
125 }
126
127 enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) && \
128 (security_ftr_enabled(SEC_FTR_L1D_FLUSH_PR) || \
129 security_ftr_enabled(SEC_FTR_L1D_FLUSH_HV));
130
131 setup_rfi_flush(type, enable);
132 setup_count_cache_flush();
133 }
134
pnv_setup_arch(void)135 static void __init pnv_setup_arch(void)
136 {
137 set_arch_panic_timeout(10, ARCH_PANIC_TIMEOUT);
138
139 pnv_setup_rfi_flush();
140 setup_stf_barrier();
141
142 /* Initialize SMP */
143 pnv_smp_init();
144
145 /* Setup PCI */
146 pnv_pci_init();
147
148 /* Setup RTC and NVRAM callbacks */
149 if (firmware_has_feature(FW_FEATURE_OPAL))
150 opal_nvram_init();
151
152 /* Enable NAP mode */
153 powersave_nap = 1;
154
155 /* XXX PMCS */
156 }
157
pnv_init(void)158 static void __init pnv_init(void)
159 {
160 /*
161 * Initialize the LPC bus now so that legacy serial
162 * ports can be found on it
163 */
164 opal_lpc_init();
165
166 #ifdef CONFIG_HVC_OPAL
167 if (firmware_has_feature(FW_FEATURE_OPAL))
168 hvc_opal_init_early();
169 else
170 #endif
171 add_preferred_console("hvc", 0, NULL);
172 }
173
pnv_init_IRQ(void)174 static void __init pnv_init_IRQ(void)
175 {
176 /* Try using a XIVE if available, otherwise use a XICS */
177 if (!xive_native_init())
178 xics_init();
179
180 WARN_ON(!ppc_md.get_irq);
181 }
182
pnv_show_cpuinfo(struct seq_file * m)183 static void pnv_show_cpuinfo(struct seq_file *m)
184 {
185 struct device_node *root;
186 const char *model = "";
187
188 root = of_find_node_by_path("/");
189 if (root)
190 model = of_get_property(root, "model", NULL);
191 seq_printf(m, "machine\t\t: PowerNV %s\n", model);
192 if (firmware_has_feature(FW_FEATURE_OPAL))
193 seq_printf(m, "firmware\t: OPAL\n");
194 else
195 seq_printf(m, "firmware\t: BML\n");
196 of_node_put(root);
197 if (radix_enabled())
198 seq_printf(m, "MMU\t\t: Radix\n");
199 else
200 seq_printf(m, "MMU\t\t: Hash\n");
201 }
202
pnv_prepare_going_down(void)203 static void pnv_prepare_going_down(void)
204 {
205 /*
206 * Disable all notifiers from OPAL, we can't
207 * service interrupts anymore anyway
208 */
209 opal_event_shutdown();
210
211 /* Soft disable interrupts */
212 local_irq_disable();
213
214 /*
215 * Return secondary CPUs to firwmare if a flash update
216 * is pending otherwise we will get all sort of error
217 * messages about CPU being stuck etc.. This will also
218 * have the side effect of hard disabling interrupts so
219 * past this point, the kernel is effectively dead.
220 */
221 opal_flash_term_callback();
222 }
223
pnv_restart(char * cmd)224 static void __noreturn pnv_restart(char *cmd)
225 {
226 long rc = OPAL_BUSY;
227
228 pnv_prepare_going_down();
229
230 while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
231 rc = opal_cec_reboot();
232 if (rc == OPAL_BUSY_EVENT)
233 opal_poll_events(NULL);
234 else
235 mdelay(10);
236 }
237 for (;;)
238 opal_poll_events(NULL);
239 }
240
pnv_power_off(void)241 static void __noreturn pnv_power_off(void)
242 {
243 long rc = OPAL_BUSY;
244
245 pnv_prepare_going_down();
246
247 while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
248 rc = opal_cec_power_down(0);
249 if (rc == OPAL_BUSY_EVENT)
250 opal_poll_events(NULL);
251 else
252 mdelay(10);
253 }
254 for (;;)
255 opal_poll_events(NULL);
256 }
257
pnv_halt(void)258 static void __noreturn pnv_halt(void)
259 {
260 pnv_power_off();
261 }
262
pnv_progress(char * s,unsigned short hex)263 static void pnv_progress(char *s, unsigned short hex)
264 {
265 }
266
pnv_shutdown(void)267 static void pnv_shutdown(void)
268 {
269 /* Let the PCI code clear up IODA tables */
270 pnv_pci_shutdown();
271
272 /*
273 * Stop OPAL activity: Unregister all OPAL interrupts so they
274 * don't fire up while we kexec and make sure all potentially
275 * DMA'ing ops are complete (such as dump retrieval).
276 */
277 opal_shutdown();
278 }
279
280 #ifdef CONFIG_KEXEC_CORE
pnv_kexec_wait_secondaries_down(void)281 static void pnv_kexec_wait_secondaries_down(void)
282 {
283 int my_cpu, i, notified = -1;
284
285 my_cpu = get_cpu();
286
287 for_each_online_cpu(i) {
288 uint8_t status;
289 int64_t rc, timeout = 1000;
290
291 if (i == my_cpu)
292 continue;
293
294 for (;;) {
295 rc = opal_query_cpu_status(get_hard_smp_processor_id(i),
296 &status);
297 if (rc != OPAL_SUCCESS || status != OPAL_THREAD_STARTED)
298 break;
299 barrier();
300 if (i != notified) {
301 printk(KERN_INFO "kexec: waiting for cpu %d "
302 "(physical %d) to enter OPAL\n",
303 i, paca[i].hw_cpu_id);
304 notified = i;
305 }
306
307 /*
308 * On crash secondaries might be unreachable or hung,
309 * so timeout if we've waited too long
310 * */
311 mdelay(1);
312 if (timeout-- == 0) {
313 printk(KERN_ERR "kexec: timed out waiting for "
314 "cpu %d (physical %d) to enter OPAL\n",
315 i, paca[i].hw_cpu_id);
316 break;
317 }
318 }
319 }
320 }
321
pnv_kexec_cpu_down(int crash_shutdown,int secondary)322 static void pnv_kexec_cpu_down(int crash_shutdown, int secondary)
323 {
324 u64 reinit_flags;
325
326 if (xive_enabled())
327 xive_kexec_teardown_cpu(secondary);
328 else
329 xics_kexec_teardown_cpu(secondary);
330
331 /* On OPAL, we return all CPUs to firmware */
332 if (!firmware_has_feature(FW_FEATURE_OPAL))
333 return;
334
335 if (secondary) {
336 /* Return secondary CPUs to firmware on OPAL v3 */
337 mb();
338 get_paca()->kexec_state = KEXEC_STATE_REAL_MODE;
339 mb();
340
341 /* Return the CPU to OPAL */
342 opal_return_cpu();
343 } else {
344 /* Primary waits for the secondaries to have reached OPAL */
345 pnv_kexec_wait_secondaries_down();
346
347 /* Switch XIVE back to emulation mode */
348 if (xive_enabled())
349 xive_shutdown();
350
351 /*
352 * We might be running as little-endian - now that interrupts
353 * are disabled, reset the HILE bit to big-endian so we don't
354 * take interrupts in the wrong endian later
355 *
356 * We reinit to enable both radix and hash on P9 to ensure
357 * the mode used by the next kernel is always supported.
358 */
359 reinit_flags = OPAL_REINIT_CPUS_HILE_BE;
360 if (cpu_has_feature(CPU_FTR_ARCH_300))
361 reinit_flags |= OPAL_REINIT_CPUS_MMU_RADIX |
362 OPAL_REINIT_CPUS_MMU_HASH;
363 opal_reinit_cpus(reinit_flags);
364 }
365 }
366 #endif /* CONFIG_KEXEC_CORE */
367
368 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
pnv_memory_block_size(void)369 static unsigned long pnv_memory_block_size(void)
370 {
371 /*
372 * We map the kernel linear region with 1GB large pages on radix. For
373 * memory hot unplug to work our memory block size must be at least
374 * this size.
375 */
376 if (radix_enabled())
377 return 1UL * 1024 * 1024 * 1024;
378 else
379 return 256UL * 1024 * 1024;
380 }
381 #endif
382
pnv_setup_machdep_opal(void)383 static void __init pnv_setup_machdep_opal(void)
384 {
385 ppc_md.get_boot_time = opal_get_boot_time;
386 ppc_md.restart = pnv_restart;
387 pm_power_off = pnv_power_off;
388 ppc_md.halt = pnv_halt;
389 ppc_md.machine_check_exception = opal_machine_check;
390 ppc_md.mce_check_early_recovery = opal_mce_check_early_recovery;
391 ppc_md.hmi_exception_early = opal_hmi_exception_early;
392 ppc_md.handle_hmi_exception = opal_handle_hmi_exception;
393 }
394
pnv_probe(void)395 static int __init pnv_probe(void)
396 {
397 if (!of_machine_is_compatible("ibm,powernv"))
398 return 0;
399
400 if (firmware_has_feature(FW_FEATURE_OPAL))
401 pnv_setup_machdep_opal();
402
403 pr_debug("PowerNV detected !\n");
404
405 pnv_init();
406
407 return 1;
408 }
409
410 /*
411 * Returns the cpu frequency for 'cpu' in Hz. This is used by
412 * /proc/cpuinfo
413 */
pnv_get_proc_freq(unsigned int cpu)414 static unsigned long pnv_get_proc_freq(unsigned int cpu)
415 {
416 unsigned long ret_freq;
417
418 ret_freq = cpufreq_get(cpu) * 1000ul;
419
420 /*
421 * If the backend cpufreq driver does not exist,
422 * then fallback to old way of reporting the clockrate.
423 */
424 if (!ret_freq)
425 ret_freq = ppc_proc_freq;
426 return ret_freq;
427 }
428
define_machine(powernv)429 define_machine(powernv) {
430 .name = "PowerNV",
431 .probe = pnv_probe,
432 .setup_arch = pnv_setup_arch,
433 .init_IRQ = pnv_init_IRQ,
434 .show_cpuinfo = pnv_show_cpuinfo,
435 .get_proc_freq = pnv_get_proc_freq,
436 .progress = pnv_progress,
437 .machine_shutdown = pnv_shutdown,
438 .power_save = NULL,
439 .calibrate_decr = generic_calibrate_decr,
440 #ifdef CONFIG_KEXEC_CORE
441 .kexec_cpu_down = pnv_kexec_cpu_down,
442 #endif
443 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
444 .memory_block_size = pnv_memory_block_size,
445 #endif
446 };
447