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>
40 #include <asm/setup.h>
41 #include <asm/security_features.h>
46 static bool fw_feature_is(const char *state, const char *name,
47 struct device_node *fw_features)
49 struct device_node *np;
52 np = of_get_child_by_name(fw_features, name);
54 rc = of_property_read_bool(np, state);
61 static void init_fw_feat_flags(struct device_node *np)
63 if (fw_feature_is("enabled", "inst-spec-barrier-ori31,31,0", np))
64 security_ftr_set(SEC_FTR_SPEC_BAR_ORI31);
66 if (fw_feature_is("enabled", "fw-bcctrl-serialized", np))
67 security_ftr_set(SEC_FTR_BCCTRL_SERIALISED);
69 if (fw_feature_is("enabled", "inst-l1d-flush-ori30,30,0", np))
70 security_ftr_set(SEC_FTR_L1D_FLUSH_ORI30);
72 if (fw_feature_is("enabled", "inst-l1d-flush-trig2", np))
73 security_ftr_set(SEC_FTR_L1D_FLUSH_TRIG2);
75 if (fw_feature_is("enabled", "fw-l1d-thread-split", np))
76 security_ftr_set(SEC_FTR_L1D_THREAD_PRIV);
78 if (fw_feature_is("enabled", "fw-count-cache-disabled", np))
79 security_ftr_set(SEC_FTR_COUNT_CACHE_DISABLED);
82 * The features below are enabled by default, so we instead look to see
83 * if firmware has *disabled* them, and clear them if so.
85 if (fw_feature_is("disabled", "speculation-policy-favor-security", np))
86 security_ftr_clear(SEC_FTR_FAVOUR_SECURITY);
88 if (fw_feature_is("disabled", "needs-l1d-flush-msr-pr-0-to-1", np))
89 security_ftr_clear(SEC_FTR_L1D_FLUSH_PR);
91 if (fw_feature_is("disabled", "needs-l1d-flush-msr-hv-1-to-0", np))
92 security_ftr_clear(SEC_FTR_L1D_FLUSH_HV);
94 if (fw_feature_is("disabled", "needs-spec-barrier-for-bound-checks", np))
95 security_ftr_clear(SEC_FTR_BNDS_CHK_SPEC_BAR);
98 static void pnv_setup_rfi_flush(void)
100 struct device_node *np, *fw_features;
101 enum l1d_flush_type type;
104 /* Default to fallback in case fw-features are not available */
105 type = L1D_FLUSH_FALLBACK;
107 np = of_find_node_by_name(NULL, "ibm,opal");
108 fw_features = of_get_child_by_name(np, "fw-features");
112 init_fw_feat_flags(fw_features);
113 of_node_put(fw_features);
115 if (security_ftr_enabled(SEC_FTR_L1D_FLUSH_TRIG2))
116 type = L1D_FLUSH_MTTRIG;
118 if (security_ftr_enabled(SEC_FTR_L1D_FLUSH_ORI30))
119 type = L1D_FLUSH_ORI;
122 enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) && \
123 (security_ftr_enabled(SEC_FTR_L1D_FLUSH_PR) || \
124 security_ftr_enabled(SEC_FTR_L1D_FLUSH_HV));
126 setup_rfi_flush(type, enable);
129 static void __init pnv_setup_arch(void)
131 set_arch_panic_timeout(10, ARCH_PANIC_TIMEOUT);
133 pnv_setup_rfi_flush();
141 /* Setup RTC and NVRAM callbacks */
142 if (firmware_has_feature(FW_FEATURE_OPAL))
145 /* Enable NAP mode */
151 static void __init pnv_init(void)
154 * Initialize the LPC bus now so that legacy serial
155 * ports can be found on it
159 #ifdef CONFIG_HVC_OPAL
160 if (firmware_has_feature(FW_FEATURE_OPAL))
161 hvc_opal_init_early();
164 add_preferred_console("hvc", 0, NULL);
167 static void __init pnv_init_IRQ(void)
169 /* Try using a XIVE if available, otherwise use a XICS */
170 if (!xive_native_init())
173 WARN_ON(!ppc_md.get_irq);
176 static void pnv_show_cpuinfo(struct seq_file *m)
178 struct device_node *root;
179 const char *model = "";
181 root = of_find_node_by_path("/");
183 model = of_get_property(root, "model", NULL);
184 seq_printf(m, "machine\t\t: PowerNV %s\n", model);
185 if (firmware_has_feature(FW_FEATURE_OPAL))
186 seq_printf(m, "firmware\t: OPAL\n");
188 seq_printf(m, "firmware\t: BML\n");
191 seq_printf(m, "MMU\t\t: Radix\n");
193 seq_printf(m, "MMU\t\t: Hash\n");
196 static void pnv_prepare_going_down(void)
199 * Disable all notifiers from OPAL, we can't
200 * service interrupts anymore anyway
202 opal_event_shutdown();
204 /* Soft disable interrupts */
208 * Return secondary CPUs to firwmare if a flash update
209 * is pending otherwise we will get all sort of error
210 * messages about CPU being stuck etc.. This will also
211 * have the side effect of hard disabling interrupts so
212 * past this point, the kernel is effectively dead.
214 opal_flash_term_callback();
217 static void __noreturn pnv_restart(char *cmd)
221 pnv_prepare_going_down();
223 while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
224 rc = opal_cec_reboot();
225 if (rc == OPAL_BUSY_EVENT)
226 opal_poll_events(NULL);
231 opal_poll_events(NULL);
234 static void __noreturn pnv_power_off(void)
238 pnv_prepare_going_down();
240 while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
241 rc = opal_cec_power_down(0);
242 if (rc == OPAL_BUSY_EVENT)
243 opal_poll_events(NULL);
248 opal_poll_events(NULL);
251 static void __noreturn pnv_halt(void)
256 static void pnv_progress(char *s, unsigned short hex)
260 static void pnv_shutdown(void)
262 /* Let the PCI code clear up IODA tables */
266 * Stop OPAL activity: Unregister all OPAL interrupts so they
267 * don't fire up while we kexec and make sure all potentially
268 * DMA'ing ops are complete (such as dump retrieval).
273 #ifdef CONFIG_KEXEC_CORE
274 static void pnv_kexec_wait_secondaries_down(void)
276 int my_cpu, i, notified = -1;
280 for_each_online_cpu(i) {
282 int64_t rc, timeout = 1000;
288 rc = opal_query_cpu_status(get_hard_smp_processor_id(i),
290 if (rc != OPAL_SUCCESS || status != OPAL_THREAD_STARTED)
294 printk(KERN_INFO "kexec: waiting for cpu %d "
295 "(physical %d) to enter OPAL\n",
296 i, paca_ptrs[i]->hw_cpu_id);
301 * On crash secondaries might be unreachable or hung,
302 * so timeout if we've waited too long
305 if (timeout-- == 0) {
306 printk(KERN_ERR "kexec: timed out waiting for "
307 "cpu %d (physical %d) to enter OPAL\n",
308 i, paca_ptrs[i]->hw_cpu_id);
315 static void pnv_kexec_cpu_down(int crash_shutdown, int secondary)
320 xive_kexec_teardown_cpu(secondary);
322 xics_kexec_teardown_cpu(secondary);
324 /* On OPAL, we return all CPUs to firmware */
325 if (!firmware_has_feature(FW_FEATURE_OPAL))
329 /* Return secondary CPUs to firmware on OPAL v3 */
331 get_paca()->kexec_state = KEXEC_STATE_REAL_MODE;
334 /* Return the CPU to OPAL */
337 /* Primary waits for the secondaries to have reached OPAL */
338 pnv_kexec_wait_secondaries_down();
340 /* Switch XIVE back to emulation mode */
345 * We might be running as little-endian - now that interrupts
346 * are disabled, reset the HILE bit to big-endian so we don't
347 * take interrupts in the wrong endian later
349 * We reinit to enable both radix and hash on P9 to ensure
350 * the mode used by the next kernel is always supported.
352 reinit_flags = OPAL_REINIT_CPUS_HILE_BE;
353 if (cpu_has_feature(CPU_FTR_ARCH_300))
354 reinit_flags |= OPAL_REINIT_CPUS_MMU_RADIX |
355 OPAL_REINIT_CPUS_MMU_HASH;
356 opal_reinit_cpus(reinit_flags);
359 #endif /* CONFIG_KEXEC_CORE */
361 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
362 static unsigned long pnv_memory_block_size(void)
365 * We map the kernel linear region with 1GB large pages on radix. For
366 * memory hot unplug to work our memory block size must be at least
370 return 1UL * 1024 * 1024 * 1024;
372 return 256UL * 1024 * 1024;
376 static void __init pnv_setup_machdep_opal(void)
378 ppc_md.get_boot_time = opal_get_boot_time;
379 ppc_md.restart = pnv_restart;
380 pm_power_off = pnv_power_off;
381 ppc_md.halt = pnv_halt;
382 /* ppc_md.system_reset_exception gets filled in by pnv_smp_init() */
383 ppc_md.machine_check_exception = opal_machine_check;
384 ppc_md.mce_check_early_recovery = opal_mce_check_early_recovery;
385 ppc_md.hmi_exception_early = opal_hmi_exception_early;
386 ppc_md.handle_hmi_exception = opal_handle_hmi_exception;
389 static int __init pnv_probe(void)
391 if (!of_machine_is_compatible("ibm,powernv"))
394 if (firmware_has_feature(FW_FEATURE_OPAL))
395 pnv_setup_machdep_opal();
397 pr_debug("PowerNV detected !\n");
404 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
405 void __init pnv_tm_init(void)
407 if (!firmware_has_feature(FW_FEATURE_OPAL) ||
408 !pvr_version_is(PVR_POWER9) ||
409 early_cpu_has_feature(CPU_FTR_TM))
412 if (opal_reinit_cpus(OPAL_REINIT_CPUS_TM_SUSPEND_DISABLED) != OPAL_SUCCESS)
415 pr_info("Enabling TM (Transactional Memory) with Suspend Disabled\n");
416 cur_cpu_spec->cpu_features |= CPU_FTR_TM;
417 /* Make sure "normal" HTM is off (it should be) */
418 cur_cpu_spec->cpu_user_features2 &= ~PPC_FEATURE2_HTM;
419 /* Turn on no suspend mode, and HTM no SC */
420 cur_cpu_spec->cpu_user_features2 |= PPC_FEATURE2_HTM_NO_SUSPEND | \
421 PPC_FEATURE2_HTM_NOSC;
422 tm_suspend_disabled = true;
424 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
427 * Returns the cpu frequency for 'cpu' in Hz. This is used by
430 static unsigned long pnv_get_proc_freq(unsigned int cpu)
432 unsigned long ret_freq;
434 ret_freq = cpufreq_get(cpu) * 1000ul;
437 * If the backend cpufreq driver does not exist,
438 * then fallback to old way of reporting the clockrate.
441 ret_freq = ppc_proc_freq;
445 define_machine(powernv) {
448 .setup_arch = pnv_setup_arch,
449 .init_IRQ = pnv_init_IRQ,
450 .show_cpuinfo = pnv_show_cpuinfo,
451 .get_proc_freq = pnv_get_proc_freq,
452 .progress = pnv_progress,
453 .machine_shutdown = pnv_shutdown,
455 .calibrate_decr = generic_calibrate_decr,
456 #ifdef CONFIG_KEXEC_CORE
457 .kexec_cpu_down = pnv_kexec_cpu_down,
459 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
460 .memory_block_size = pnv_memory_block_size,