2 * PowerNV OPAL high level interfaces
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.
12 #define pr_fmt(fmt) "opal: " fmt
14 #include <linux/printk.h>
15 #include <linux/types.h>
17 #include <linux/of_fdt.h>
18 #include <linux/of_platform.h>
19 #include <linux/interrupt.h>
20 #include <linux/notifier.h>
21 #include <linux/slab.h>
22 #include <linux/sched.h>
23 #include <linux/kobject.h>
24 #include <linux/delay.h>
25 #include <linux/memblock.h>
26 #include <linux/kthread.h>
27 #include <linux/freezer.h>
29 #include <asm/machdep.h>
31 #include <asm/firmware.h>
36 /* /sys/firmware/opal */
37 struct kobject *opal_kobj;
45 struct mcheck_recoverable_range {
51 static struct mcheck_recoverable_range *mc_recoverable_range;
52 static int mc_recoverable_range_len;
54 struct device_node *opal_node;
55 static DEFINE_SPINLOCK(opal_write_lock);
56 static struct atomic_notifier_head opal_msg_notifier_head[OPAL_MSG_TYPE_MAX];
57 static uint32_t opal_heartbeat;
59 static void opal_reinit_cores(void)
61 /* Do the actual re-init, This will clobber all FPRs, VRs, etc...
63 * It will preserve non volatile GPRs and HSPRG0/1. It will
64 * also restore HIDs and other SPRs to their original value
65 * but it might clobber a bunch.
68 opal_reinit_cpus(OPAL_REINIT_CPUS_HILE_BE);
70 opal_reinit_cpus(OPAL_REINIT_CPUS_HILE_LE);
74 int __init early_init_dt_scan_opal(unsigned long node,
75 const char *uname, int depth, void *data)
77 const void *basep, *entryp, *sizep;
78 int basesz, entrysz, runtimesz;
80 if (depth != 1 || strcmp(uname, "ibm,opal") != 0)
83 basep = of_get_flat_dt_prop(node, "opal-base-address", &basesz);
84 entryp = of_get_flat_dt_prop(node, "opal-entry-address", &entrysz);
85 sizep = of_get_flat_dt_prop(node, "opal-runtime-size", &runtimesz);
87 if (!basep || !entryp || !sizep)
90 opal.base = of_read_number(basep, basesz/4);
91 opal.entry = of_read_number(entryp, entrysz/4);
92 opal.size = of_read_number(sizep, runtimesz/4);
94 pr_debug("OPAL Base = 0x%llx (basep=%p basesz=%d)\n",
95 opal.base, basep, basesz);
96 pr_debug("OPAL Entry = 0x%llx (entryp=%p basesz=%d)\n",
97 opal.entry, entryp, entrysz);
98 pr_debug("OPAL Entry = 0x%llx (sizep=%p runtimesz=%d)\n",
99 opal.size, sizep, runtimesz);
101 if (of_flat_dt_is_compatible(node, "ibm,opal-v3")) {
102 powerpc_firmware_features |= FW_FEATURE_OPAL;
103 pr_info("OPAL detected !\n");
105 panic("OPAL != V3 detected, no longer supported.\n");
108 /* Reinit all cores with the right endian */
111 /* Restore some bits */
112 if (cur_cpu_spec->cpu_restore)
113 cur_cpu_spec->cpu_restore();
118 int __init early_init_dt_scan_recoverable_ranges(unsigned long node,
119 const char *uname, int depth, void *data)
124 if (depth != 1 || strcmp(uname, "ibm,opal") != 0)
127 prop = of_get_flat_dt_prop(node, "mcheck-recoverable-ranges", &psize);
132 pr_debug("Found machine check recoverable ranges.\n");
135 * Calculate number of available entries.
137 * Each recoverable address range entry is (start address, len,
138 * recovery address), 2 cells each for start and recovery address,
139 * 1 cell for len, totalling 5 cells per entry.
141 mc_recoverable_range_len = psize / (sizeof(*prop) * 5);
144 if (!mc_recoverable_range_len)
147 /* Size required to hold all the entries. */
148 size = mc_recoverable_range_len *
149 sizeof(struct mcheck_recoverable_range);
152 * Allocate a buffer to hold the MC recoverable ranges. We would be
153 * accessing them in real mode, hence it needs to be within
156 mc_recoverable_range =__va(memblock_alloc_base(size, __alignof__(u64),
158 memset(mc_recoverable_range, 0, size);
160 for (i = 0; i < mc_recoverable_range_len; i++) {
161 mc_recoverable_range[i].start_addr =
162 of_read_number(prop + (i * 5) + 0, 2);
163 mc_recoverable_range[i].end_addr =
164 mc_recoverable_range[i].start_addr +
165 of_read_number(prop + (i * 5) + 2, 1);
166 mc_recoverable_range[i].recover_addr =
167 of_read_number(prop + (i * 5) + 3, 2);
169 pr_debug("Machine check recoverable range: %llx..%llx: %llx\n",
170 mc_recoverable_range[i].start_addr,
171 mc_recoverable_range[i].end_addr,
172 mc_recoverable_range[i].recover_addr);
177 static int __init opal_register_exception_handlers(void)
179 #ifdef __BIG_ENDIAN__
182 if (!(powerpc_firmware_features & FW_FEATURE_OPAL))
185 /* Hookup some exception handlers except machine check. We use the
186 * fwnmi area at 0x7000 to provide the glue space to OPAL
191 * Check if we are running on newer firmware that exports
192 * OPAL_HANDLE_HMI token. If yes, then don't ask OPAL to patch
193 * the HMI interrupt and we catch it directly in Linux.
195 * For older firmware (i.e currently released POWER8 System Firmware
196 * as of today <= SV810_087), we fallback to old behavior and let OPAL
197 * patch the HMI vector and handle it inside OPAL firmware.
199 * For newer firmware (in development/yet to be released) we will
200 * start catching/handling HMI directly in Linux.
202 if (!opal_check_token(OPAL_HANDLE_HMI)) {
203 pr_info("Old firmware detected, OPAL handles HMIs.\n");
204 opal_register_exception_handler(
205 OPAL_HYPERVISOR_MAINTENANCE_HANDLER,
210 opal_register_exception_handler(OPAL_SOFTPATCH_HANDLER, 0, glue);
215 machine_early_initcall(powernv, opal_register_exception_handlers);
218 * Opal message notifier based on message type. Allow subscribers to get
219 * notified for specific messgae type.
221 int opal_message_notifier_register(enum opal_msg_type msg_type,
222 struct notifier_block *nb)
224 if (!nb || msg_type >= OPAL_MSG_TYPE_MAX) {
225 pr_warning("%s: Invalid arguments, msg_type:%d\n",
230 return atomic_notifier_chain_register(
231 &opal_msg_notifier_head[msg_type], nb);
233 EXPORT_SYMBOL_GPL(opal_message_notifier_register);
235 int opal_message_notifier_unregister(enum opal_msg_type msg_type,
236 struct notifier_block *nb)
238 return atomic_notifier_chain_unregister(
239 &opal_msg_notifier_head[msg_type], nb);
241 EXPORT_SYMBOL_GPL(opal_message_notifier_unregister);
243 static void opal_message_do_notify(uint32_t msg_type, void *msg)
245 /* notify subscribers */
246 atomic_notifier_call_chain(&opal_msg_notifier_head[msg_type],
250 static void opal_handle_message(void)
254 * TODO: pre-allocate a message buffer depending on opal-msg-size
255 * value in /proc/device-tree.
257 static struct opal_msg msg;
260 ret = opal_get_msg(__pa(&msg), sizeof(msg));
261 /* No opal message pending. */
262 if (ret == OPAL_RESOURCE)
265 /* check for errors. */
267 pr_warning("%s: Failed to retrieve opal message, err=%lld\n",
272 type = be32_to_cpu(msg.msg_type);
275 if (type >= OPAL_MSG_TYPE_MAX) {
276 pr_warn_once("%s: Unknown message type: %u\n", __func__, type);
279 opal_message_do_notify(type, (void *)&msg);
282 static irqreturn_t opal_message_notify(int irq, void *data)
284 opal_handle_message();
288 static int __init opal_message_init(void)
292 for (i = 0; i < OPAL_MSG_TYPE_MAX; i++)
293 ATOMIC_INIT_NOTIFIER_HEAD(&opal_msg_notifier_head[i]);
295 irq = opal_event_request(ilog2(OPAL_EVENT_MSG_PENDING));
297 pr_err("%s: Can't register OPAL event irq (%d)\n",
302 ret = request_irq(irq, opal_message_notify,
303 IRQ_TYPE_LEVEL_HIGH, "opal-msg", NULL);
305 pr_err("%s: Can't request OPAL event irq (%d)\n",
313 int opal_get_chars(uint32_t vtermno, char *buf, int count)
320 opal_poll_events(&evt);
321 if ((be64_to_cpu(evt) & OPAL_EVENT_CONSOLE_INPUT) == 0)
323 len = cpu_to_be64(count);
324 rc = opal_console_read(vtermno, &len, buf);
325 if (rc == OPAL_SUCCESS)
326 return be64_to_cpu(len);
330 int opal_put_chars(uint32_t vtermno, const char *data, int total_len)
341 /* We want put_chars to be atomic to avoid mangling of hvsi
342 * packets. To do that, we first test for room and return
343 * -EAGAIN if there isn't enough.
345 * Unfortunately, opal_console_write_buffer_space() doesn't
346 * appear to work on opal v1, so we just assume there is
347 * enough room and be done with it
349 spin_lock_irqsave(&opal_write_lock, flags);
350 rc = opal_console_write_buffer_space(vtermno, &olen);
351 len = be64_to_cpu(olen);
352 if (rc || len < total_len) {
353 spin_unlock_irqrestore(&opal_write_lock, flags);
354 /* Closed -> drop characters */
357 opal_poll_events(NULL);
361 /* We still try to handle partial completions, though they
362 * should no longer happen.
365 while(total_len > 0 && (rc == OPAL_BUSY ||
366 rc == OPAL_BUSY_EVENT || rc == OPAL_SUCCESS)) {
367 olen = cpu_to_be64(total_len);
368 rc = opal_console_write(vtermno, &olen, data);
369 len = be64_to_cpu(olen);
371 /* Closed or other error drop */
372 if (rc != OPAL_SUCCESS && rc != OPAL_BUSY &&
373 rc != OPAL_BUSY_EVENT) {
377 if (rc == OPAL_SUCCESS) {
382 /* This is a bit nasty but we need that for the console to
383 * flush when there aren't any interrupts. We will clean
384 * things a bit later to limit that to synchronous path
385 * such as the kernel console and xmon/udbg
388 opal_poll_events(&evt);
389 while(rc == OPAL_SUCCESS &&
390 (be64_to_cpu(evt) & OPAL_EVENT_CONSOLE_OUTPUT));
392 spin_unlock_irqrestore(&opal_write_lock, flags);
396 static int opal_recover_mce(struct pt_regs *regs,
397 struct machine_check_event *evt)
400 uint64_t ea = get_mce_fault_addr(evt);
402 if (!(regs->msr & MSR_RI)) {
403 /* If MSR_RI isn't set, we cannot recover */
405 } else if (evt->disposition == MCE_DISPOSITION_RECOVERED) {
406 /* Platform corrected itself */
408 } else if (ea && !is_kernel_addr(ea)) {
410 * Faulting address is not in kernel text. We should be fine.
411 * We need to find which process uses this address.
412 * For now, kill the task if we have received exception when
415 * TODO: Queue up this address for hwpoisioning later.
417 if (user_mode(regs) && !is_global_init(current)) {
418 _exception(SIGBUS, regs, BUS_MCEERR_AR, regs->nip);
422 } else if (user_mode(regs) && !is_global_init(current) &&
423 evt->severity == MCE_SEV_ERROR_SYNC) {
425 * If we have received a synchronous error when in userspace
428 _exception(SIGBUS, regs, BUS_MCEERR_AR, regs->nip);
434 int opal_machine_check(struct pt_regs *regs)
436 struct machine_check_event evt;
439 if (!get_mce_event(&evt, MCE_EVENT_RELEASE))
442 /* Print things out */
443 if (evt.version != MCE_V1) {
444 pr_err("Machine Check Exception, Unknown event version %d !\n",
448 machine_check_print_event_info(&evt);
450 if (opal_recover_mce(regs, &evt))
454 * Unrecovered machine check, we are heading to panic path.
456 * We may have hit this MCE in very early stage of kernel
457 * initialization even before opal-prd has started running. If
458 * this is the case then this MCE error may go un-noticed or
459 * un-analyzed if we go down panic path. We need to inform
460 * BMC/OCC about this error so that they can collect relevant
461 * data for error analysis before rebooting.
462 * Use opal_cec_reboot2(OPAL_REBOOT_PLATFORM_ERROR) to do so.
463 * This function may not return on BMC based system.
465 ret = opal_cec_reboot2(OPAL_REBOOT_PLATFORM_ERROR,
466 "Unrecoverable Machine Check exception");
467 if (ret == OPAL_UNSUPPORTED) {
468 pr_emerg("Reboot type %d not supported\n",
469 OPAL_REBOOT_PLATFORM_ERROR);
473 * We reached here. There can be three possibilities:
474 * 1. We are running on a firmware level that do not support
476 * 2. We are running on a firmware level that do not support
477 * OPAL_REBOOT_PLATFORM_ERROR reboot type.
478 * 3. We are running on FSP based system that does not need opal
479 * to trigger checkstop explicitly for error analysis. The FSP
480 * PRD component would have already got notified about this
481 * error through other channels.
483 * If hardware marked this as an unrecoverable MCE, we are
484 * going to panic anyway. Even if it didn't, it's not safe to
485 * continue at this point, so we should explicitly panic.
488 panic("PowerNV Unrecovered Machine Check");
492 /* Early hmi handler called in real mode. */
493 int opal_hmi_exception_early(struct pt_regs *regs)
498 * call opal hmi handler. Pass paca address as token.
499 * The return value OPAL_SUCCESS is an indication that there is
500 * an HMI event generated waiting to pull by Linux.
502 rc = opal_handle_hmi();
503 if (rc == OPAL_SUCCESS) {
504 local_paca->hmi_event_available = 1;
510 /* HMI exception handler called in virtual mode during check_irq_replay. */
511 int opal_handle_hmi_exception(struct pt_regs *regs)
517 * Check if HMI event is available.
518 * if Yes, then call opal_poll_events to pull opal messages and
521 if (!local_paca->hmi_event_available)
524 local_paca->hmi_event_available = 0;
525 rc = opal_poll_events(&evt);
526 if (rc == OPAL_SUCCESS && evt)
527 opal_handle_events(be64_to_cpu(evt));
532 static uint64_t find_recovery_address(uint64_t nip)
536 for (i = 0; i < mc_recoverable_range_len; i++)
537 if ((nip >= mc_recoverable_range[i].start_addr) &&
538 (nip < mc_recoverable_range[i].end_addr))
539 return mc_recoverable_range[i].recover_addr;
543 bool opal_mce_check_early_recovery(struct pt_regs *regs)
545 uint64_t recover_addr = 0;
547 if (!opal.base || !opal.size)
550 if ((regs->nip >= opal.base) &&
551 (regs->nip < (opal.base + opal.size)))
552 recover_addr = find_recovery_address(regs->nip);
555 * Setup regs->nip to rfi into fixup address.
558 regs->nip = recover_addr;
561 return !!recover_addr;
564 static int opal_sysfs_init(void)
566 opal_kobj = kobject_create_and_add("opal", firmware_kobj);
568 pr_warn("kobject_create_and_add opal failed\n");
575 static ssize_t symbol_map_read(struct file *fp, struct kobject *kobj,
576 struct bin_attribute *bin_attr,
577 char *buf, loff_t off, size_t count)
579 return memory_read_from_buffer(buf, count, &off, bin_attr->private,
583 static BIN_ATTR_RO(symbol_map, 0);
585 static void opal_export_symmap(void)
589 struct device_node *fw;
592 fw = of_find_node_by_path("/ibm,opal/firmware");
595 syms = of_get_property(fw, "symbol-map", &size);
596 if (!syms || size != 2 * sizeof(__be64))
599 /* Setup attributes */
600 bin_attr_symbol_map.private = __va(be64_to_cpu(syms[0]));
601 bin_attr_symbol_map.size = be64_to_cpu(syms[1]);
603 rc = sysfs_create_bin_file(opal_kobj, &bin_attr_symbol_map);
605 pr_warn("Error %d creating OPAL symbols file\n", rc);
608 static void __init opal_dump_region_init(void)
614 if (!opal_check_token(OPAL_REGISTER_DUMP_REGION))
617 /* Register kernel log buffer */
618 addr = log_buf_addr_get();
622 size = log_buf_len_get();
626 rc = opal_register_dump_region(OPAL_DUMP_REGION_LOG_BUF,
628 /* Don't warn if this is just an older OPAL that doesn't
629 * know about that call
631 if (rc && rc != OPAL_UNSUPPORTED)
632 pr_warn("DUMP: Failed to register kernel log buffer. "
636 static void opal_pdev_init(struct device_node *opal_node,
637 const char *compatible)
639 struct device_node *np;
641 for_each_child_of_node(opal_node, np)
642 if (of_device_is_compatible(np, compatible))
643 of_platform_device_create(np, NULL, NULL);
646 static void opal_i2c_create_devs(void)
648 struct device_node *np;
650 for_each_compatible_node(np, NULL, "ibm,opal-i2c")
651 of_platform_device_create(np, NULL, NULL);
654 static int kopald(void *unused)
661 opal_poll_events(&events);
662 opal_handle_events(be64_to_cpu(events));
663 msleep_interruptible(opal_heartbeat);
664 } while (!kthread_should_stop());
669 static void opal_init_heartbeat(void)
671 /* Old firwmware, we assume the HVC heartbeat is sufficient */
672 if (of_property_read_u32(opal_node, "ibm,heartbeat-ms",
673 &opal_heartbeat) != 0)
677 kthread_run(kopald, NULL, "kopald");
680 static int __init opal_init(void)
682 struct device_node *np, *consoles, *leds;
685 opal_node = of_find_node_by_path("/ibm,opal");
687 pr_warn("Device node not found\n");
691 /* Register OPAL consoles if any ports */
692 consoles = of_find_node_by_path("/ibm,opal/consoles");
694 for_each_child_of_node(consoles, np) {
695 if (strcmp(np->name, "serial"))
697 of_platform_device_create(np, NULL, NULL);
699 of_node_put(consoles);
702 /* Initialise OPAL messaging system */
705 /* Initialise OPAL asynchronous completion interface */
706 opal_async_comp_init();
708 /* Initialise OPAL sensor interface */
711 /* Initialise OPAL hypervisor maintainence interrupt handling */
712 opal_hmi_handler_init();
714 /* Create i2c platform devices */
715 opal_i2c_create_devs();
717 /* Setup a heatbeat thread if requested by OPAL */
718 opal_init_heartbeat();
720 /* Create leds platform devices */
721 leds = of_find_node_by_path("/ibm,opal/leds");
723 of_platform_device_create(leds, "opal_leds", NULL);
727 /* Initialise OPAL message log interface */
730 /* Create "opal" kobject under /sys/firmware */
731 rc = opal_sysfs_init();
733 /* Export symbol map to userspace */
734 opal_export_symmap();
735 /* Setup dump region interface */
736 opal_dump_region_init();
737 /* Setup error log interface */
738 rc = opal_elog_init();
739 /* Setup code update interface */
740 opal_flash_update_init();
741 /* Setup platform dump extract interface */
742 opal_platform_dump_init();
743 /* Setup system parameters interface */
744 opal_sys_param_init();
745 /* Setup message log sysfs interface. */
746 opal_msglog_sysfs_init();
749 /* Initialize platform devices: IPMI backend, PRD & flash interface */
750 opal_pdev_init(opal_node, "ibm,opal-ipmi");
751 opal_pdev_init(opal_node, "ibm,opal-flash");
752 opal_pdev_init(opal_node, "ibm,opal-prd");
754 /* Initialise platform device: oppanel interface */
755 opal_pdev_init(opal_node, "ibm,opal-oppanel");
757 /* Initialise OPAL kmsg dumper for flushing console on panic */
762 machine_subsys_initcall(powernv, opal_init);
764 void opal_shutdown(void)
768 opal_event_shutdown();
771 * Then sync with OPAL which ensure anything that can
772 * potentially write to our memory has completed such
773 * as an ongoing dump retrieval
775 while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
776 rc = opal_sync_host_reboot();
778 opal_poll_events(NULL);
783 /* Unregister memory dump region */
784 if (opal_check_token(OPAL_UNREGISTER_DUMP_REGION))
785 opal_unregister_dump_region(OPAL_DUMP_REGION_LOG_BUF);
788 /* Export this so that test modules can use it */
789 EXPORT_SYMBOL_GPL(opal_invalid_call);
790 EXPORT_SYMBOL_GPL(opal_xscom_read);
791 EXPORT_SYMBOL_GPL(opal_xscom_write);
792 EXPORT_SYMBOL_GPL(opal_ipmi_send);
793 EXPORT_SYMBOL_GPL(opal_ipmi_recv);
794 EXPORT_SYMBOL_GPL(opal_flash_read);
795 EXPORT_SYMBOL_GPL(opal_flash_write);
796 EXPORT_SYMBOL_GPL(opal_flash_erase);
797 EXPORT_SYMBOL_GPL(opal_prd_msg);
799 /* Convert a region of vmalloc memory to an opal sg list */
800 struct opal_sg_list *opal_vmalloc_to_sg_list(void *vmalloc_addr,
801 unsigned long vmalloc_size)
803 struct opal_sg_list *sg, *first = NULL;
806 sg = kzalloc(PAGE_SIZE, GFP_KERNEL);
812 while (vmalloc_size > 0) {
813 uint64_t data = vmalloc_to_pfn(vmalloc_addr) << PAGE_SHIFT;
814 uint64_t length = min(vmalloc_size, PAGE_SIZE);
816 sg->entry[i].data = cpu_to_be64(data);
817 sg->entry[i].length = cpu_to_be64(length);
820 if (i >= SG_ENTRIES_PER_NODE) {
821 struct opal_sg_list *next;
823 next = kzalloc(PAGE_SIZE, GFP_KERNEL);
827 sg->length = cpu_to_be64(
828 i * sizeof(struct opal_sg_entry) + 16);
830 sg->next = cpu_to_be64(__pa(next));
834 vmalloc_addr += length;
835 vmalloc_size -= length;
838 sg->length = cpu_to_be64(i * sizeof(struct opal_sg_entry) + 16);
843 pr_err("%s : Failed to allocate memory\n", __func__);
844 opal_free_sg_list(first);
848 void opal_free_sg_list(struct opal_sg_list *sg)
851 uint64_t next = be64_to_cpu(sg->next);
862 int opal_error_code(int rc)
865 case OPAL_SUCCESS: return 0;
867 case OPAL_PARAMETER: return -EINVAL;
868 case OPAL_ASYNC_COMPLETION: return -EINPROGRESS;
869 case OPAL_BUSY_EVENT: return -EBUSY;
870 case OPAL_NO_MEM: return -ENOMEM;
871 case OPAL_PERMISSION: return -EPERM;
873 case OPAL_UNSUPPORTED: return -EIO;
874 case OPAL_HARDWARE: return -EIO;
875 case OPAL_INTERNAL_ERROR: return -EIO;
877 pr_err("%s: unexpected OPAL error %d\n", __func__, rc);
882 EXPORT_SYMBOL_GPL(opal_poll_events);
883 EXPORT_SYMBOL_GPL(opal_rtc_read);
884 EXPORT_SYMBOL_GPL(opal_rtc_write);
885 EXPORT_SYMBOL_GPL(opal_tpo_read);
886 EXPORT_SYMBOL_GPL(opal_tpo_write);
887 EXPORT_SYMBOL_GPL(opal_i2c_request);
888 /* Export these symbols for PowerNV LED class driver */
889 EXPORT_SYMBOL_GPL(opal_leds_get_ind);
890 EXPORT_SYMBOL_GPL(opal_leds_set_ind);
891 /* Export this symbol for PowerNV Operator Panel class driver */
892 EXPORT_SYMBOL_GPL(opal_write_oppanel_async);