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;
58 static struct task_struct *kopald_tsk;
60 void opal_configure_cores(void)
62 /* Do the actual re-init, This will clobber all FPRs, VRs, etc...
64 * It will preserve non volatile GPRs and HSPRG0/1. It will
65 * also restore HIDs and other SPRs to their original value
66 * but it might clobber a bunch.
69 opal_reinit_cpus(OPAL_REINIT_CPUS_HILE_BE);
71 opal_reinit_cpus(OPAL_REINIT_CPUS_HILE_LE);
74 /* Restore some bits */
75 if (cur_cpu_spec->cpu_restore)
76 cur_cpu_spec->cpu_restore();
79 int __init early_init_dt_scan_opal(unsigned long node,
80 const char *uname, int depth, void *data)
82 const void *basep, *entryp, *sizep;
83 int basesz, entrysz, runtimesz;
85 if (depth != 1 || strcmp(uname, "ibm,opal") != 0)
88 basep = of_get_flat_dt_prop(node, "opal-base-address", &basesz);
89 entryp = of_get_flat_dt_prop(node, "opal-entry-address", &entrysz);
90 sizep = of_get_flat_dt_prop(node, "opal-runtime-size", &runtimesz);
92 if (!basep || !entryp || !sizep)
95 opal.base = of_read_number(basep, basesz/4);
96 opal.entry = of_read_number(entryp, entrysz/4);
97 opal.size = of_read_number(sizep, runtimesz/4);
99 pr_debug("OPAL Base = 0x%llx (basep=%p basesz=%d)\n",
100 opal.base, basep, basesz);
101 pr_debug("OPAL Entry = 0x%llx (entryp=%p basesz=%d)\n",
102 opal.entry, entryp, entrysz);
103 pr_debug("OPAL Entry = 0x%llx (sizep=%p runtimesz=%d)\n",
104 opal.size, sizep, runtimesz);
106 if (of_flat_dt_is_compatible(node, "ibm,opal-v3")) {
107 powerpc_firmware_features |= FW_FEATURE_OPAL;
108 pr_info("OPAL detected !\n");
110 panic("OPAL != V3 detected, no longer supported.\n");
116 int __init early_init_dt_scan_recoverable_ranges(unsigned long node,
117 const char *uname, int depth, void *data)
122 if (depth != 1 || strcmp(uname, "ibm,opal") != 0)
125 prop = of_get_flat_dt_prop(node, "mcheck-recoverable-ranges", &psize);
130 pr_debug("Found machine check recoverable ranges.\n");
133 * Calculate number of available entries.
135 * Each recoverable address range entry is (start address, len,
136 * recovery address), 2 cells each for start and recovery address,
137 * 1 cell for len, totalling 5 cells per entry.
139 mc_recoverable_range_len = psize / (sizeof(*prop) * 5);
142 if (!mc_recoverable_range_len)
145 /* Size required to hold all the entries. */
146 size = mc_recoverable_range_len *
147 sizeof(struct mcheck_recoverable_range);
150 * Allocate a buffer to hold the MC recoverable ranges. We would be
151 * accessing them in real mode, hence it needs to be within
154 mc_recoverable_range =__va(memblock_alloc_base(size, __alignof__(u64),
156 memset(mc_recoverable_range, 0, size);
158 for (i = 0; i < mc_recoverable_range_len; i++) {
159 mc_recoverable_range[i].start_addr =
160 of_read_number(prop + (i * 5) + 0, 2);
161 mc_recoverable_range[i].end_addr =
162 mc_recoverable_range[i].start_addr +
163 of_read_number(prop + (i * 5) + 2, 1);
164 mc_recoverable_range[i].recover_addr =
165 of_read_number(prop + (i * 5) + 3, 2);
167 pr_debug("Machine check recoverable range: %llx..%llx: %llx\n",
168 mc_recoverable_range[i].start_addr,
169 mc_recoverable_range[i].end_addr,
170 mc_recoverable_range[i].recover_addr);
175 static int __init opal_register_exception_handlers(void)
177 #ifdef __BIG_ENDIAN__
180 if (!(powerpc_firmware_features & FW_FEATURE_OPAL))
183 /* Hookup some exception handlers except machine check. We use the
184 * fwnmi area at 0x7000 to provide the glue space to OPAL
189 * Check if we are running on newer firmware that exports
190 * OPAL_HANDLE_HMI token. If yes, then don't ask OPAL to patch
191 * the HMI interrupt and we catch it directly in Linux.
193 * For older firmware (i.e currently released POWER8 System Firmware
194 * as of today <= SV810_087), we fallback to old behavior and let OPAL
195 * patch the HMI vector and handle it inside OPAL firmware.
197 * For newer firmware (in development/yet to be released) we will
198 * start catching/handling HMI directly in Linux.
200 if (!opal_check_token(OPAL_HANDLE_HMI)) {
201 pr_info("Old firmware detected, OPAL handles HMIs.\n");
202 opal_register_exception_handler(
203 OPAL_HYPERVISOR_MAINTENANCE_HANDLER,
208 opal_register_exception_handler(OPAL_SOFTPATCH_HANDLER, 0, glue);
213 machine_early_initcall(powernv, opal_register_exception_handlers);
216 * Opal message notifier based on message type. Allow subscribers to get
217 * notified for specific messgae type.
219 int opal_message_notifier_register(enum opal_msg_type msg_type,
220 struct notifier_block *nb)
222 if (!nb || msg_type >= OPAL_MSG_TYPE_MAX) {
223 pr_warning("%s: Invalid arguments, msg_type:%d\n",
228 return atomic_notifier_chain_register(
229 &opal_msg_notifier_head[msg_type], nb);
231 EXPORT_SYMBOL_GPL(opal_message_notifier_register);
233 int opal_message_notifier_unregister(enum opal_msg_type msg_type,
234 struct notifier_block *nb)
236 return atomic_notifier_chain_unregister(
237 &opal_msg_notifier_head[msg_type], nb);
239 EXPORT_SYMBOL_GPL(opal_message_notifier_unregister);
241 static void opal_message_do_notify(uint32_t msg_type, void *msg)
243 /* notify subscribers */
244 atomic_notifier_call_chain(&opal_msg_notifier_head[msg_type],
248 static void opal_handle_message(void)
252 * TODO: pre-allocate a message buffer depending on opal-msg-size
253 * value in /proc/device-tree.
255 static struct opal_msg msg;
258 ret = opal_get_msg(__pa(&msg), sizeof(msg));
259 /* No opal message pending. */
260 if (ret == OPAL_RESOURCE)
263 /* check for errors. */
265 pr_warning("%s: Failed to retrieve opal message, err=%lld\n",
270 type = be32_to_cpu(msg.msg_type);
273 if (type >= OPAL_MSG_TYPE_MAX) {
274 pr_warn_once("%s: Unknown message type: %u\n", __func__, type);
277 opal_message_do_notify(type, (void *)&msg);
280 static irqreturn_t opal_message_notify(int irq, void *data)
282 opal_handle_message();
286 static int __init opal_message_init(void)
290 for (i = 0; i < OPAL_MSG_TYPE_MAX; i++)
291 ATOMIC_INIT_NOTIFIER_HEAD(&opal_msg_notifier_head[i]);
293 irq = opal_event_request(ilog2(OPAL_EVENT_MSG_PENDING));
295 pr_err("%s: Can't register OPAL event irq (%d)\n",
300 ret = request_irq(irq, opal_message_notify,
301 IRQ_TYPE_LEVEL_HIGH, "opal-msg", NULL);
303 pr_err("%s: Can't request OPAL event irq (%d)\n",
311 int opal_get_chars(uint32_t vtermno, char *buf, int count)
318 opal_poll_events(&evt);
319 if ((be64_to_cpu(evt) & OPAL_EVENT_CONSOLE_INPUT) == 0)
321 len = cpu_to_be64(count);
322 rc = opal_console_read(vtermno, &len, buf);
323 if (rc == OPAL_SUCCESS)
324 return be64_to_cpu(len);
328 int opal_put_chars(uint32_t vtermno, const char *data, int total_len)
339 /* We want put_chars to be atomic to avoid mangling of hvsi
340 * packets. To do that, we first test for room and return
341 * -EAGAIN if there isn't enough.
343 * Unfortunately, opal_console_write_buffer_space() doesn't
344 * appear to work on opal v1, so we just assume there is
345 * enough room and be done with it
347 spin_lock_irqsave(&opal_write_lock, flags);
348 rc = opal_console_write_buffer_space(vtermno, &olen);
349 len = be64_to_cpu(olen);
350 if (rc || len < total_len) {
351 spin_unlock_irqrestore(&opal_write_lock, flags);
352 /* Closed -> drop characters */
355 opal_poll_events(NULL);
359 /* We still try to handle partial completions, though they
360 * should no longer happen.
363 while(total_len > 0 && (rc == OPAL_BUSY ||
364 rc == OPAL_BUSY_EVENT || rc == OPAL_SUCCESS)) {
365 olen = cpu_to_be64(total_len);
366 rc = opal_console_write(vtermno, &olen, data);
367 len = be64_to_cpu(olen);
369 /* Closed or other error drop */
370 if (rc != OPAL_SUCCESS && rc != OPAL_BUSY &&
371 rc != OPAL_BUSY_EVENT) {
375 if (rc == OPAL_SUCCESS) {
380 /* This is a bit nasty but we need that for the console to
381 * flush when there aren't any interrupts. We will clean
382 * things a bit later to limit that to synchronous path
383 * such as the kernel console and xmon/udbg
386 opal_poll_events(&evt);
387 while(rc == OPAL_SUCCESS &&
388 (be64_to_cpu(evt) & OPAL_EVENT_CONSOLE_OUTPUT));
390 spin_unlock_irqrestore(&opal_write_lock, flags);
394 static int opal_recover_mce(struct pt_regs *regs,
395 struct machine_check_event *evt)
399 if (!(regs->msr & MSR_RI)) {
400 /* If MSR_RI isn't set, we cannot recover */
401 pr_err("Machine check interrupt unrecoverable: MSR(RI=0)\n");
403 } else if (evt->disposition == MCE_DISPOSITION_RECOVERED) {
404 /* Platform corrected itself */
406 } else if (evt->severity == MCE_SEV_FATAL) {
407 /* Fatal machine check */
408 pr_err("Machine check interrupt is fatal\n");
410 } else if ((evt->severity == MCE_SEV_ERROR_SYNC) &&
411 (user_mode(regs) && !is_global_init(current))) {
413 * For now, kill the task if we have received exception when
416 * TODO: Queue up this address for hwpoisioning later.
418 _exception(SIGBUS, regs, BUS_MCEERR_AR, regs->nip);
424 int opal_machine_check(struct pt_regs *regs)
426 struct machine_check_event evt;
429 if (!get_mce_event(&evt, MCE_EVENT_RELEASE))
432 /* Print things out */
433 if (evt.version != MCE_V1) {
434 pr_err("Machine Check Exception, Unknown event version %d !\n",
438 machine_check_print_event_info(&evt);
440 if (opal_recover_mce(regs, &evt))
444 * Unrecovered machine check, we are heading to panic path.
446 * We may have hit this MCE in very early stage of kernel
447 * initialization even before opal-prd has started running. If
448 * this is the case then this MCE error may go un-noticed or
449 * un-analyzed if we go down panic path. We need to inform
450 * BMC/OCC about this error so that they can collect relevant
451 * data for error analysis before rebooting.
452 * Use opal_cec_reboot2(OPAL_REBOOT_PLATFORM_ERROR) to do so.
453 * This function may not return on BMC based system.
455 ret = opal_cec_reboot2(OPAL_REBOOT_PLATFORM_ERROR,
456 "Unrecoverable Machine Check exception");
457 if (ret == OPAL_UNSUPPORTED) {
458 pr_emerg("Reboot type %d not supported\n",
459 OPAL_REBOOT_PLATFORM_ERROR);
463 * We reached here. There can be three possibilities:
464 * 1. We are running on a firmware level that do not support
466 * 2. We are running on a firmware level that do not support
467 * OPAL_REBOOT_PLATFORM_ERROR reboot type.
468 * 3. We are running on FSP based system that does not need opal
469 * to trigger checkstop explicitly for error analysis. The FSP
470 * PRD component would have already got notified about this
471 * error through other channels.
473 * If hardware marked this as an unrecoverable MCE, we are
474 * going to panic anyway. Even if it didn't, it's not safe to
475 * continue at this point, so we should explicitly panic.
478 panic("PowerNV Unrecovered Machine Check");
482 /* Early hmi handler called in real mode. */
483 int opal_hmi_exception_early(struct pt_regs *regs)
488 * call opal hmi handler. Pass paca address as token.
489 * The return value OPAL_SUCCESS is an indication that there is
490 * an HMI event generated waiting to pull by Linux.
492 rc = opal_handle_hmi();
493 if (rc == OPAL_SUCCESS) {
494 local_paca->hmi_event_available = 1;
500 /* HMI exception handler called in virtual mode during check_irq_replay. */
501 int opal_handle_hmi_exception(struct pt_regs *regs)
507 * Check if HMI event is available.
508 * if Yes, then call opal_poll_events to pull opal messages and
511 if (!local_paca->hmi_event_available)
514 local_paca->hmi_event_available = 0;
515 rc = opal_poll_events(&evt);
516 if (rc == OPAL_SUCCESS && evt)
517 opal_handle_events(be64_to_cpu(evt));
522 static uint64_t find_recovery_address(uint64_t nip)
526 for (i = 0; i < mc_recoverable_range_len; i++)
527 if ((nip >= mc_recoverable_range[i].start_addr) &&
528 (nip < mc_recoverable_range[i].end_addr))
529 return mc_recoverable_range[i].recover_addr;
533 bool opal_mce_check_early_recovery(struct pt_regs *regs)
535 uint64_t recover_addr = 0;
537 if (!opal.base || !opal.size)
540 if ((regs->nip >= opal.base) &&
541 (regs->nip < (opal.base + opal.size)))
542 recover_addr = find_recovery_address(regs->nip);
545 * Setup regs->nip to rfi into fixup address.
548 regs->nip = recover_addr;
551 return !!recover_addr;
554 static int opal_sysfs_init(void)
556 opal_kobj = kobject_create_and_add("opal", firmware_kobj);
558 pr_warn("kobject_create_and_add opal failed\n");
565 static ssize_t symbol_map_read(struct file *fp, struct kobject *kobj,
566 struct bin_attribute *bin_attr,
567 char *buf, loff_t off, size_t count)
569 return memory_read_from_buffer(buf, count, &off, bin_attr->private,
573 static BIN_ATTR_RO(symbol_map, 0);
575 static void opal_export_symmap(void)
579 struct device_node *fw;
582 fw = of_find_node_by_path("/ibm,opal/firmware");
585 syms = of_get_property(fw, "symbol-map", &size);
586 if (!syms || size != 2 * sizeof(__be64))
589 /* Setup attributes */
590 bin_attr_symbol_map.private = __va(be64_to_cpu(syms[0]));
591 bin_attr_symbol_map.size = be64_to_cpu(syms[1]);
593 rc = sysfs_create_bin_file(opal_kobj, &bin_attr_symbol_map);
595 pr_warn("Error %d creating OPAL symbols file\n", rc);
598 static void __init opal_dump_region_init(void)
604 if (!opal_check_token(OPAL_REGISTER_DUMP_REGION))
607 /* Register kernel log buffer */
608 addr = log_buf_addr_get();
612 size = log_buf_len_get();
616 rc = opal_register_dump_region(OPAL_DUMP_REGION_LOG_BUF,
618 /* Don't warn if this is just an older OPAL that doesn't
619 * know about that call
621 if (rc && rc != OPAL_UNSUPPORTED)
622 pr_warn("DUMP: Failed to register kernel log buffer. "
626 static void opal_pdev_init(const char *compatible)
628 struct device_node *np;
630 for_each_compatible_node(np, NULL, compatible)
631 of_platform_device_create(np, NULL, NULL);
634 static int kopald(void *unused)
636 unsigned long timeout = msecs_to_jiffies(opal_heartbeat) + 1;
642 opal_poll_events(&events);
643 opal_handle_events(be64_to_cpu(events));
644 schedule_timeout_interruptible(timeout);
645 } while (!kthread_should_stop());
650 void opal_wake_poller(void)
653 wake_up_process(kopald_tsk);
656 static void opal_init_heartbeat(void)
658 /* Old firwmware, we assume the HVC heartbeat is sufficient */
659 if (of_property_read_u32(opal_node, "ibm,heartbeat-ms",
660 &opal_heartbeat) != 0)
664 kopald_tsk = kthread_run(kopald, NULL, "kopald");
667 static int __init opal_init(void)
669 struct device_node *np, *consoles, *leds;
672 opal_node = of_find_node_by_path("/ibm,opal");
674 pr_warn("Device node not found\n");
678 /* Register OPAL consoles if any ports */
679 consoles = of_find_node_by_path("/ibm,opal/consoles");
681 for_each_child_of_node(consoles, np) {
682 if (strcmp(np->name, "serial"))
684 of_platform_device_create(np, NULL, NULL);
686 of_node_put(consoles);
689 /* Initialise OPAL messaging system */
692 /* Initialise OPAL asynchronous completion interface */
693 opal_async_comp_init();
695 /* Initialise OPAL sensor interface */
698 /* Initialise OPAL hypervisor maintainence interrupt handling */
699 opal_hmi_handler_init();
701 /* Create i2c platform devices */
702 opal_pdev_init("ibm,opal-i2c");
704 /* Setup a heatbeat thread if requested by OPAL */
705 opal_init_heartbeat();
707 /* Create leds platform devices */
708 leds = of_find_node_by_path("/ibm,opal/leds");
710 of_platform_device_create(leds, "opal_leds", NULL);
714 /* Initialise OPAL message log interface */
717 /* Create "opal" kobject under /sys/firmware */
718 rc = opal_sysfs_init();
720 /* Export symbol map to userspace */
721 opal_export_symmap();
722 /* Setup dump region interface */
723 opal_dump_region_init();
724 /* Setup error log interface */
725 rc = opal_elog_init();
726 /* Setup code update interface */
727 opal_flash_update_init();
728 /* Setup platform dump extract interface */
729 opal_platform_dump_init();
730 /* Setup system parameters interface */
731 opal_sys_param_init();
732 /* Setup message log sysfs interface. */
733 opal_msglog_sysfs_init();
736 /* Initialize platform devices: IPMI backend, PRD & flash interface */
737 opal_pdev_init("ibm,opal-ipmi");
738 opal_pdev_init("ibm,opal-flash");
739 opal_pdev_init("ibm,opal-prd");
741 /* Initialise platform device: oppanel interface */
742 opal_pdev_init("ibm,opal-oppanel");
744 /* Initialise OPAL kmsg dumper for flushing console on panic */
749 machine_subsys_initcall(powernv, opal_init);
751 void opal_shutdown(void)
755 opal_event_shutdown();
758 * Then sync with OPAL which ensure anything that can
759 * potentially write to our memory has completed such
760 * as an ongoing dump retrieval
762 while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
763 rc = opal_sync_host_reboot();
765 opal_poll_events(NULL);
770 /* Unregister memory dump region */
771 if (opal_check_token(OPAL_UNREGISTER_DUMP_REGION))
772 opal_unregister_dump_region(OPAL_DUMP_REGION_LOG_BUF);
775 /* Export this so that test modules can use it */
776 EXPORT_SYMBOL_GPL(opal_invalid_call);
777 EXPORT_SYMBOL_GPL(opal_xscom_read);
778 EXPORT_SYMBOL_GPL(opal_xscom_write);
779 EXPORT_SYMBOL_GPL(opal_ipmi_send);
780 EXPORT_SYMBOL_GPL(opal_ipmi_recv);
781 EXPORT_SYMBOL_GPL(opal_flash_read);
782 EXPORT_SYMBOL_GPL(opal_flash_write);
783 EXPORT_SYMBOL_GPL(opal_flash_erase);
784 EXPORT_SYMBOL_GPL(opal_prd_msg);
786 /* Convert a region of vmalloc memory to an opal sg list */
787 struct opal_sg_list *opal_vmalloc_to_sg_list(void *vmalloc_addr,
788 unsigned long vmalloc_size)
790 struct opal_sg_list *sg, *first = NULL;
793 sg = kzalloc(PAGE_SIZE, GFP_KERNEL);
799 while (vmalloc_size > 0) {
800 uint64_t data = vmalloc_to_pfn(vmalloc_addr) << PAGE_SHIFT;
801 uint64_t length = min(vmalloc_size, PAGE_SIZE);
803 sg->entry[i].data = cpu_to_be64(data);
804 sg->entry[i].length = cpu_to_be64(length);
807 if (i >= SG_ENTRIES_PER_NODE) {
808 struct opal_sg_list *next;
810 next = kzalloc(PAGE_SIZE, GFP_KERNEL);
814 sg->length = cpu_to_be64(
815 i * sizeof(struct opal_sg_entry) + 16);
817 sg->next = cpu_to_be64(__pa(next));
821 vmalloc_addr += length;
822 vmalloc_size -= length;
825 sg->length = cpu_to_be64(i * sizeof(struct opal_sg_entry) + 16);
830 pr_err("%s : Failed to allocate memory\n", __func__);
831 opal_free_sg_list(first);
835 void opal_free_sg_list(struct opal_sg_list *sg)
838 uint64_t next = be64_to_cpu(sg->next);
849 int opal_error_code(int rc)
852 case OPAL_SUCCESS: return 0;
854 case OPAL_PARAMETER: return -EINVAL;
855 case OPAL_ASYNC_COMPLETION: return -EINPROGRESS;
856 case OPAL_BUSY_EVENT: return -EBUSY;
857 case OPAL_NO_MEM: return -ENOMEM;
858 case OPAL_PERMISSION: return -EPERM;
860 case OPAL_UNSUPPORTED: return -EIO;
861 case OPAL_HARDWARE: return -EIO;
862 case OPAL_INTERNAL_ERROR: return -EIO;
864 pr_err("%s: unexpected OPAL error %d\n", __func__, rc);
869 void powernv_set_nmmu_ptcr(unsigned long ptcr)
873 if (firmware_has_feature(FW_FEATURE_OPAL)) {
874 rc = opal_nmmu_set_ptcr(-1UL, ptcr);
875 if (rc != OPAL_SUCCESS && rc != OPAL_UNSUPPORTED)
876 pr_warn("%s: Unable to set nest mmu ptcr\n", __func__);
880 EXPORT_SYMBOL_GPL(opal_poll_events);
881 EXPORT_SYMBOL_GPL(opal_rtc_read);
882 EXPORT_SYMBOL_GPL(opal_rtc_write);
883 EXPORT_SYMBOL_GPL(opal_tpo_read);
884 EXPORT_SYMBOL_GPL(opal_tpo_write);
885 EXPORT_SYMBOL_GPL(opal_i2c_request);
886 /* Export these symbols for PowerNV LED class driver */
887 EXPORT_SYMBOL_GPL(opal_leds_get_ind);
888 EXPORT_SYMBOL_GPL(opal_leds_set_ind);
889 /* Export this symbol for PowerNV Operator Panel class driver */
890 EXPORT_SYMBOL_GPL(opal_write_oppanel_async);
891 /* Export this for KVM */
892 EXPORT_SYMBOL_GPL(opal_int_set_mfrr);