4 * Copyright 2011 IBM Corp.
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.
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>
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>
32 #include <asm/machdep.h>
33 #include <asm/firmware.h>
37 #include <asm/kexec.h>
39 #include <asm/setup.h>
40 #include <asm/security_features.h>
45 static bool fw_feature_is(const char *state, const char *name,
46 struct device_node *fw_features)
48 struct device_node *np;
51 np = of_get_child_by_name(fw_features, name);
53 rc = of_property_read_bool(np, state);
60 static void init_fw_feat_flags(struct device_node *np)
62 if (fw_feature_is("enabled", "inst-spec-barrier-ori31,31,0", np))
63 security_ftr_set(SEC_FTR_SPEC_BAR_ORI31);
65 if (fw_feature_is("enabled", "fw-bcctrl-serialized", np))
66 security_ftr_set(SEC_FTR_BCCTRL_SERIALISED);
68 if (fw_feature_is("enabled", "inst-l1d-flush-ori30,30,0", np))
69 security_ftr_set(SEC_FTR_L1D_FLUSH_ORI30);
71 if (fw_feature_is("enabled", "inst-l1d-flush-trig2", np))
72 security_ftr_set(SEC_FTR_L1D_FLUSH_TRIG2);
74 if (fw_feature_is("enabled", "fw-l1d-thread-split", np))
75 security_ftr_set(SEC_FTR_L1D_THREAD_PRIV);
77 if (fw_feature_is("enabled", "fw-count-cache-disabled", np))
78 security_ftr_set(SEC_FTR_COUNT_CACHE_DISABLED);
80 if (fw_feature_is("enabled", "fw-count-cache-flush-bcctr2,0,0", np))
81 security_ftr_set(SEC_FTR_BCCTR_FLUSH_ASSIST);
83 if (fw_feature_is("enabled", "needs-count-cache-flush-on-context-switch", np))
84 security_ftr_set(SEC_FTR_FLUSH_COUNT_CACHE);
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.
90 if (fw_feature_is("disabled", "speculation-policy-favor-security", np))
91 security_ftr_clear(SEC_FTR_FAVOUR_SECURITY);
93 if (fw_feature_is("disabled", "needs-l1d-flush-msr-pr-0-to-1", np))
94 security_ftr_clear(SEC_FTR_L1D_FLUSH_PR);
96 if (fw_feature_is("disabled", "needs-l1d-flush-msr-hv-1-to-0", np))
97 security_ftr_clear(SEC_FTR_L1D_FLUSH_HV);
99 if (fw_feature_is("disabled", "needs-spec-barrier-for-bound-checks", np))
100 security_ftr_clear(SEC_FTR_BNDS_CHK_SPEC_BAR);
103 static void pnv_setup_rfi_flush(void)
105 struct device_node *np, *fw_features;
106 enum l1d_flush_type type;
109 /* Default to fallback in case fw-features are not available */
110 type = L1D_FLUSH_FALLBACK;
112 np = of_find_node_by_name(NULL, "ibm,opal");
113 fw_features = of_get_child_by_name(np, "fw-features");
117 init_fw_feat_flags(fw_features);
118 of_node_put(fw_features);
120 if (security_ftr_enabled(SEC_FTR_L1D_FLUSH_TRIG2))
121 type = L1D_FLUSH_MTTRIG;
123 if (security_ftr_enabled(SEC_FTR_L1D_FLUSH_ORI30))
124 type = L1D_FLUSH_ORI;
128 * If we are non-Power9 bare metal, we don't need to flush on kernel
129 * entry or after user access: they fix a P9 specific vulnerability.
131 if (!pvr_version_is(PVR_POWER9)) {
132 security_ftr_clear(SEC_FTR_L1D_FLUSH_ENTRY);
133 security_ftr_clear(SEC_FTR_L1D_FLUSH_UACCESS);
136 enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) && \
137 (security_ftr_enabled(SEC_FTR_L1D_FLUSH_PR) || \
138 security_ftr_enabled(SEC_FTR_L1D_FLUSH_HV));
140 setup_rfi_flush(type, enable);
141 setup_count_cache_flush();
143 enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) &&
144 security_ftr_enabled(SEC_FTR_L1D_FLUSH_ENTRY);
145 setup_entry_flush(enable);
147 enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) &&
148 security_ftr_enabled(SEC_FTR_L1D_FLUSH_UACCESS);
149 setup_uaccess_flush(enable);
152 static void __init pnv_setup_arch(void)
154 set_arch_panic_timeout(10, ARCH_PANIC_TIMEOUT);
156 pnv_setup_rfi_flush();
165 /* Setup RTC and NVRAM callbacks */
166 if (firmware_has_feature(FW_FEATURE_OPAL))
169 /* Enable NAP mode */
177 static void __init pnv_init(void)
180 * Initialize the LPC bus now so that legacy serial
181 * ports can be found on it
185 #ifdef CONFIG_HVC_OPAL
186 if (firmware_has_feature(FW_FEATURE_OPAL))
187 hvc_opal_init_early();
190 add_preferred_console("hvc", 0, NULL);
193 static void __init pnv_init_IRQ(void)
195 /* Try using a XIVE if available, otherwise use a XICS */
196 if (!xive_native_init())
199 WARN_ON(!ppc_md.get_irq);
202 static void pnv_show_cpuinfo(struct seq_file *m)
204 struct device_node *root;
205 const char *model = "";
207 root = of_find_node_by_path("/");
209 model = of_get_property(root, "model", NULL);
210 seq_printf(m, "machine\t\t: PowerNV %s\n", model);
211 if (firmware_has_feature(FW_FEATURE_OPAL))
212 seq_printf(m, "firmware\t: OPAL\n");
214 seq_printf(m, "firmware\t: BML\n");
217 seq_printf(m, "MMU\t\t: Radix\n");
219 seq_printf(m, "MMU\t\t: Hash\n");
222 static void pnv_prepare_going_down(void)
225 * Disable all notifiers from OPAL, we can't
226 * service interrupts anymore anyway
228 opal_event_shutdown();
230 /* Soft disable interrupts */
234 * Return secondary CPUs to firwmare if a flash update
235 * is pending otherwise we will get all sort of error
236 * messages about CPU being stuck etc.. This will also
237 * have the side effect of hard disabling interrupts so
238 * past this point, the kernel is effectively dead.
240 opal_flash_term_callback();
243 static void __noreturn pnv_restart(char *cmd)
247 pnv_prepare_going_down();
249 while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
250 rc = opal_cec_reboot();
251 if (rc == OPAL_BUSY_EVENT)
252 opal_poll_events(NULL);
257 opal_poll_events(NULL);
260 static void __noreturn pnv_power_off(void)
264 pnv_prepare_going_down();
266 while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
267 rc = opal_cec_power_down(0);
268 if (rc == OPAL_BUSY_EVENT)
269 opal_poll_events(NULL);
274 opal_poll_events(NULL);
277 static void __noreturn pnv_halt(void)
282 static void pnv_progress(char *s, unsigned short hex)
286 static void pnv_shutdown(void)
288 /* Let the PCI code clear up IODA tables */
292 * Stop OPAL activity: Unregister all OPAL interrupts so they
293 * don't fire up while we kexec and make sure all potentially
294 * DMA'ing ops are complete (such as dump retrieval).
299 #ifdef CONFIG_KEXEC_CORE
300 static void pnv_kexec_wait_secondaries_down(void)
302 int my_cpu, i, notified = -1;
306 for_each_online_cpu(i) {
308 int64_t rc, timeout = 1000;
314 rc = opal_query_cpu_status(get_hard_smp_processor_id(i),
316 if (rc != OPAL_SUCCESS || status != OPAL_THREAD_STARTED)
320 printk(KERN_INFO "kexec: waiting for cpu %d "
321 "(physical %d) to enter OPAL\n",
322 i, paca[i].hw_cpu_id);
327 * On crash secondaries might be unreachable or hung,
328 * so timeout if we've waited too long
331 if (timeout-- == 0) {
332 printk(KERN_ERR "kexec: timed out waiting for "
333 "cpu %d (physical %d) to enter OPAL\n",
334 i, paca[i].hw_cpu_id);
341 static void pnv_kexec_cpu_down(int crash_shutdown, int secondary)
346 xive_kexec_teardown_cpu(secondary);
348 xics_kexec_teardown_cpu(secondary);
350 /* On OPAL, we return all CPUs to firmware */
351 if (!firmware_has_feature(FW_FEATURE_OPAL))
355 /* Return secondary CPUs to firmware on OPAL v3 */
357 get_paca()->kexec_state = KEXEC_STATE_REAL_MODE;
360 /* Return the CPU to OPAL */
363 /* Primary waits for the secondaries to have reached OPAL */
364 pnv_kexec_wait_secondaries_down();
366 /* Switch XIVE back to emulation mode */
371 * We might be running as little-endian - now that interrupts
372 * are disabled, reset the HILE bit to big-endian so we don't
373 * take interrupts in the wrong endian later
375 * We reinit to enable both radix and hash on P9 to ensure
376 * the mode used by the next kernel is always supported.
378 reinit_flags = OPAL_REINIT_CPUS_HILE_BE;
379 if (cpu_has_feature(CPU_FTR_ARCH_300))
380 reinit_flags |= OPAL_REINIT_CPUS_MMU_RADIX |
381 OPAL_REINIT_CPUS_MMU_HASH;
382 opal_reinit_cpus(reinit_flags);
385 #endif /* CONFIG_KEXEC_CORE */
387 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
388 static unsigned long pnv_memory_block_size(void)
391 * We map the kernel linear region with 1GB large pages on radix. For
392 * memory hot unplug to work our memory block size must be at least
396 return 1UL * 1024 * 1024 * 1024;
398 return 256UL * 1024 * 1024;
402 static void __init pnv_setup_machdep_opal(void)
404 ppc_md.get_boot_time = opal_get_boot_time;
405 ppc_md.restart = pnv_restart;
406 pm_power_off = pnv_power_off;
407 ppc_md.halt = pnv_halt;
408 ppc_md.machine_check_exception = opal_machine_check;
409 ppc_md.mce_check_early_recovery = opal_mce_check_early_recovery;
410 ppc_md.hmi_exception_early = opal_hmi_exception_early;
411 ppc_md.handle_hmi_exception = opal_handle_hmi_exception;
414 static int __init pnv_probe(void)
416 if (!of_machine_is_compatible("ibm,powernv"))
419 if (firmware_has_feature(FW_FEATURE_OPAL))
420 pnv_setup_machdep_opal();
422 pr_debug("PowerNV detected !\n");
430 * Returns the cpu frequency for 'cpu' in Hz. This is used by
433 static unsigned long pnv_get_proc_freq(unsigned int cpu)
435 unsigned long ret_freq;
437 ret_freq = cpufreq_get(cpu) * 1000ul;
440 * If the backend cpufreq driver does not exist,
441 * then fallback to old way of reporting the clockrate.
444 ret_freq = ppc_proc_freq;
448 define_machine(powernv) {
451 .setup_arch = pnv_setup_arch,
452 .init_IRQ = pnv_init_IRQ,
453 .show_cpuinfo = pnv_show_cpuinfo,
454 .get_proc_freq = pnv_get_proc_freq,
455 .progress = pnv_progress,
456 .machine_shutdown = pnv_shutdown,
458 .calibrate_decr = generic_calibrate_decr,
459 #ifdef CONFIG_KEXEC_CORE
460 .kexec_cpu_down = pnv_kexec_cpu_down,
462 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
463 .memory_block_size = pnv_memory_block_size,