1 // SPDX-License-Identifier: GPL-2.0-or-later
5 * Copyright 2011 IBM Corp.
10 #include <linux/cpu.h>
11 #include <linux/errno.h>
12 #include <linux/sched.h>
13 #include <linux/kernel.h>
14 #include <linux/tty.h>
15 #include <linux/reboot.h>
16 #include <linux/init.h>
17 #include <linux/console.h>
18 #include <linux/delay.h>
19 #include <linux/irq.h>
20 #include <linux/seq_file.h>
22 #include <linux/of_fdt.h>
23 #include <linux/interrupt.h>
24 #include <linux/bug.h>
25 #include <linux/pci.h>
26 #include <linux/cpufreq.h>
27 #include <linux/memblock.h>
29 #include <asm/machdep.h>
30 #include <asm/firmware.h>
34 #include <asm/kexec.h>
37 #include <asm/setup.h>
38 #include <asm/security_features.h>
43 static bool fw_feature_is(const char *state, const char *name,
44 struct device_node *fw_features)
46 struct device_node *np;
49 np = of_get_child_by_name(fw_features, name);
51 rc = of_property_read_bool(np, state);
58 static void init_fw_feat_flags(struct device_node *np)
60 if (fw_feature_is("enabled", "inst-spec-barrier-ori31,31,0", np))
61 security_ftr_set(SEC_FTR_SPEC_BAR_ORI31);
63 if (fw_feature_is("enabled", "fw-bcctrl-serialized", np))
64 security_ftr_set(SEC_FTR_BCCTRL_SERIALISED);
66 if (fw_feature_is("enabled", "inst-l1d-flush-ori30,30,0", np))
67 security_ftr_set(SEC_FTR_L1D_FLUSH_ORI30);
69 if (fw_feature_is("enabled", "inst-l1d-flush-trig2", np))
70 security_ftr_set(SEC_FTR_L1D_FLUSH_TRIG2);
72 if (fw_feature_is("enabled", "fw-l1d-thread-split", np))
73 security_ftr_set(SEC_FTR_L1D_THREAD_PRIV);
75 if (fw_feature_is("enabled", "fw-count-cache-disabled", np))
76 security_ftr_set(SEC_FTR_COUNT_CACHE_DISABLED);
78 if (fw_feature_is("enabled", "fw-count-cache-flush-bcctr2,0,0", np))
79 security_ftr_set(SEC_FTR_BCCTR_FLUSH_ASSIST);
81 if (fw_feature_is("enabled", "needs-count-cache-flush-on-context-switch", np))
82 security_ftr_set(SEC_FTR_FLUSH_COUNT_CACHE);
85 * The features below are enabled by default, so we instead look to see
86 * if firmware has *disabled* them, and clear them if so.
88 if (fw_feature_is("disabled", "speculation-policy-favor-security", np))
89 security_ftr_clear(SEC_FTR_FAVOUR_SECURITY);
91 if (fw_feature_is("disabled", "needs-l1d-flush-msr-pr-0-to-1", np))
92 security_ftr_clear(SEC_FTR_L1D_FLUSH_PR);
94 if (fw_feature_is("disabled", "needs-l1d-flush-msr-hv-1-to-0", np))
95 security_ftr_clear(SEC_FTR_L1D_FLUSH_HV);
97 if (fw_feature_is("disabled", "needs-spec-barrier-for-bound-checks", np))
98 security_ftr_clear(SEC_FTR_BNDS_CHK_SPEC_BAR);
101 static void pnv_setup_rfi_flush(void)
103 struct device_node *np, *fw_features;
104 enum l1d_flush_type type;
107 /* Default to fallback in case fw-features are not available */
108 type = L1D_FLUSH_FALLBACK;
110 np = of_find_node_by_name(NULL, "ibm,opal");
111 fw_features = of_get_child_by_name(np, "fw-features");
115 init_fw_feat_flags(fw_features);
116 of_node_put(fw_features);
118 if (security_ftr_enabled(SEC_FTR_L1D_FLUSH_TRIG2))
119 type = L1D_FLUSH_MTTRIG;
121 if (security_ftr_enabled(SEC_FTR_L1D_FLUSH_ORI30))
122 type = L1D_FLUSH_ORI;
126 * If we are non-Power9 bare metal, we don't need to flush on kernel
127 * entry or after user access: they fix a P9 specific vulnerability.
129 if (!pvr_version_is(PVR_POWER9)) {
130 security_ftr_clear(SEC_FTR_L1D_FLUSH_ENTRY);
131 security_ftr_clear(SEC_FTR_L1D_FLUSH_UACCESS);
134 enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) && \
135 (security_ftr_enabled(SEC_FTR_L1D_FLUSH_PR) || \
136 security_ftr_enabled(SEC_FTR_L1D_FLUSH_HV));
138 setup_rfi_flush(type, enable);
139 setup_count_cache_flush();
141 enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) &&
142 security_ftr_enabled(SEC_FTR_L1D_FLUSH_ENTRY);
143 setup_entry_flush(enable);
145 enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) &&
146 security_ftr_enabled(SEC_FTR_L1D_FLUSH_UACCESS);
147 setup_uaccess_flush(enable);
150 static void __init pnv_setup_arch(void)
152 set_arch_panic_timeout(10, ARCH_PANIC_TIMEOUT);
154 pnv_setup_rfi_flush();
163 /* Setup RTC and NVRAM callbacks */
164 if (firmware_has_feature(FW_FEATURE_OPAL))
167 /* Enable NAP mode */
175 static void __init pnv_init(void)
178 * Initialize the LPC bus now so that legacy serial
179 * ports can be found on it
183 #ifdef CONFIG_HVC_OPAL
184 if (firmware_has_feature(FW_FEATURE_OPAL))
185 hvc_opal_init_early();
188 add_preferred_console("hvc", 0, NULL);
190 if (!radix_enabled()) {
191 size_t size = sizeof(struct slb_entry) * mmu_slb_size;
194 /* Allocate per cpu area to save old slb contents during MCE */
195 for_each_possible_cpu(i) {
196 paca_ptrs[i]->mce_faulty_slbs =
197 memblock_alloc_node(size,
198 __alignof__(struct slb_entry),
204 static void __init pnv_init_IRQ(void)
206 /* Try using a XIVE if available, otherwise use a XICS */
207 if (!xive_native_init())
210 WARN_ON(!ppc_md.get_irq);
213 static void pnv_show_cpuinfo(struct seq_file *m)
215 struct device_node *root;
216 const char *model = "";
218 root = of_find_node_by_path("/");
220 model = of_get_property(root, "model", NULL);
221 seq_printf(m, "machine\t\t: PowerNV %s\n", model);
222 if (firmware_has_feature(FW_FEATURE_OPAL))
223 seq_printf(m, "firmware\t: OPAL\n");
225 seq_printf(m, "firmware\t: BML\n");
228 seq_printf(m, "MMU\t\t: Radix\n");
230 seq_printf(m, "MMU\t\t: Hash\n");
233 static void pnv_prepare_going_down(void)
236 * Disable all notifiers from OPAL, we can't
237 * service interrupts anymore anyway
239 opal_event_shutdown();
241 /* Print flash update message if one is scheduled. */
242 opal_flash_update_print_message();
249 static void __noreturn pnv_restart(char *cmd)
253 pnv_prepare_going_down();
257 rc = opal_cec_reboot();
258 else if (strcmp(cmd, "full") == 0)
259 rc = opal_cec_reboot2(OPAL_REBOOT_FULL_IPL, NULL);
261 rc = OPAL_UNSUPPORTED;
263 if (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
264 /* Opal is busy wait for some time and retry */
265 opal_poll_events(NULL);
268 } else if (cmd && rc) {
269 /* Unknown error while issuing reboot */
270 if (rc == OPAL_UNSUPPORTED)
271 pr_err("Unsupported '%s' reboot.\n", cmd);
273 pr_err("Unable to issue '%s' reboot. Err=%ld\n",
275 pr_info("Forcing a cec-reboot\n");
279 } else if (rc != OPAL_SUCCESS) {
280 /* Unknown error while issuing cec-reboot */
281 pr_err("Unable to reboot. Err=%ld\n", rc);
284 } while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT);
287 opal_poll_events(NULL);
290 static void __noreturn pnv_power_off(void)
294 pnv_prepare_going_down();
296 while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
297 rc = opal_cec_power_down(0);
298 if (rc == OPAL_BUSY_EVENT)
299 opal_poll_events(NULL);
304 opal_poll_events(NULL);
307 static void __noreturn pnv_halt(void)
312 static void pnv_progress(char *s, unsigned short hex)
316 static void pnv_shutdown(void)
318 /* Let the PCI code clear up IODA tables */
322 * Stop OPAL activity: Unregister all OPAL interrupts so they
323 * don't fire up while we kexec and make sure all potentially
324 * DMA'ing ops are complete (such as dump retrieval).
329 #ifdef CONFIG_KEXEC_CORE
330 static void pnv_kexec_wait_secondaries_down(void)
332 int my_cpu, i, notified = -1;
336 for_each_online_cpu(i) {
338 int64_t rc, timeout = 1000;
344 rc = opal_query_cpu_status(get_hard_smp_processor_id(i),
346 if (rc != OPAL_SUCCESS || status != OPAL_THREAD_STARTED)
350 printk(KERN_INFO "kexec: waiting for cpu %d "
351 "(physical %d) to enter OPAL\n",
352 i, paca_ptrs[i]->hw_cpu_id);
357 * On crash secondaries might be unreachable or hung,
358 * so timeout if we've waited too long
361 if (timeout-- == 0) {
362 printk(KERN_ERR "kexec: timed out waiting for "
363 "cpu %d (physical %d) to enter OPAL\n",
364 i, paca_ptrs[i]->hw_cpu_id);
371 static void pnv_kexec_cpu_down(int crash_shutdown, int secondary)
378 xics_kexec_teardown_cpu(secondary);
380 /* On OPAL, we return all CPUs to firmware */
381 if (!firmware_has_feature(FW_FEATURE_OPAL))
385 /* Return secondary CPUs to firmware on OPAL v3 */
387 get_paca()->kexec_state = KEXEC_STATE_REAL_MODE;
390 /* Return the CPU to OPAL */
393 /* Primary waits for the secondaries to have reached OPAL */
394 pnv_kexec_wait_secondaries_down();
396 /* Switch XIVE back to emulation mode */
401 * We might be running as little-endian - now that interrupts
402 * are disabled, reset the HILE bit to big-endian so we don't
403 * take interrupts in the wrong endian later
405 * We reinit to enable both radix and hash on P9 to ensure
406 * the mode used by the next kernel is always supported.
408 reinit_flags = OPAL_REINIT_CPUS_HILE_BE;
409 if (cpu_has_feature(CPU_FTR_ARCH_300))
410 reinit_flags |= OPAL_REINIT_CPUS_MMU_RADIX |
411 OPAL_REINIT_CPUS_MMU_HASH;
412 opal_reinit_cpus(reinit_flags);
415 #endif /* CONFIG_KEXEC_CORE */
417 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
418 static unsigned long pnv_memory_block_size(void)
420 return 256UL * 1024 * 1024;
424 static void __init pnv_setup_machdep_opal(void)
426 ppc_md.get_boot_time = opal_get_boot_time;
427 ppc_md.restart = pnv_restart;
428 pm_power_off = pnv_power_off;
429 ppc_md.halt = pnv_halt;
430 /* ppc_md.system_reset_exception gets filled in by pnv_smp_init() */
431 ppc_md.machine_check_exception = opal_machine_check;
432 ppc_md.mce_check_early_recovery = opal_mce_check_early_recovery;
433 if (opal_check_token(OPAL_HANDLE_HMI2))
434 ppc_md.hmi_exception_early = opal_hmi_exception_early2;
436 ppc_md.hmi_exception_early = opal_hmi_exception_early;
437 ppc_md.handle_hmi_exception = opal_handle_hmi_exception;
440 static int __init pnv_probe(void)
442 if (!of_machine_is_compatible("ibm,powernv"))
445 if (firmware_has_feature(FW_FEATURE_OPAL))
446 pnv_setup_machdep_opal();
448 pr_debug("PowerNV detected !\n");
455 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
456 void __init pnv_tm_init(void)
458 if (!firmware_has_feature(FW_FEATURE_OPAL) ||
459 !pvr_version_is(PVR_POWER9) ||
460 early_cpu_has_feature(CPU_FTR_TM))
463 if (opal_reinit_cpus(OPAL_REINIT_CPUS_TM_SUSPEND_DISABLED) != OPAL_SUCCESS)
466 pr_info("Enabling TM (Transactional Memory) with Suspend Disabled\n");
467 cur_cpu_spec->cpu_features |= CPU_FTR_TM;
468 /* Make sure "normal" HTM is off (it should be) */
469 cur_cpu_spec->cpu_user_features2 &= ~PPC_FEATURE2_HTM;
470 /* Turn on no suspend mode, and HTM no SC */
471 cur_cpu_spec->cpu_user_features2 |= PPC_FEATURE2_HTM_NO_SUSPEND | \
472 PPC_FEATURE2_HTM_NOSC;
473 tm_suspend_disabled = true;
475 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
478 * Returns the cpu frequency for 'cpu' in Hz. This is used by
481 static unsigned long pnv_get_proc_freq(unsigned int cpu)
483 unsigned long ret_freq;
485 ret_freq = cpufreq_get(cpu) * 1000ul;
488 * If the backend cpufreq driver does not exist,
489 * then fallback to old way of reporting the clockrate.
492 ret_freq = ppc_proc_freq;
496 static long pnv_machine_check_early(struct pt_regs *regs)
500 if (cur_cpu_spec && cur_cpu_spec->machine_check_early)
501 handled = cur_cpu_spec->machine_check_early(regs);
506 define_machine(powernv) {
509 .setup_arch = pnv_setup_arch,
510 .init_IRQ = pnv_init_IRQ,
511 .show_cpuinfo = pnv_show_cpuinfo,
512 .get_proc_freq = pnv_get_proc_freq,
513 .progress = pnv_progress,
514 .machine_shutdown = pnv_shutdown,
516 .calibrate_decr = generic_calibrate_decr,
517 .machine_check_early = pnv_machine_check_early,
518 #ifdef CONFIG_KEXEC_CORE
519 .kexec_cpu_down = pnv_kexec_cpu_down,
521 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
522 .memory_block_size = pnv_memory_block_size,