2 * 64-bit pSeries and RS/6000 setup code.
4 * Copyright (C) 1995 Linus Torvalds
5 * Adapted from 'alpha' version by Gary Thomas
6 * Modified by Cort Dougan (cort@cs.nmt.edu)
7 * Modified by PPC64 Team, IBM Corp
9 * This program is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU General Public License
11 * as published by the Free Software Foundation; either version
12 * 2 of the License, or (at your option) any later version.
16 * bootup setup stuff..
19 #include <linux/cpu.h>
20 #include <linux/errno.h>
21 #include <linux/sched.h>
22 #include <linux/kernel.h>
24 #include <linux/stddef.h>
25 #include <linux/unistd.h>
26 #include <linux/user.h>
27 #include <linux/tty.h>
28 #include <linux/major.h>
29 #include <linux/interrupt.h>
30 #include <linux/reboot.h>
31 #include <linux/init.h>
32 #include <linux/ioport.h>
33 #include <linux/console.h>
34 #include <linux/pci.h>
35 #include <linux/utsname.h>
36 #include <linux/adb.h>
37 #include <linux/export.h>
38 #include <linux/delay.h>
39 #include <linux/irq.h>
40 #include <linux/seq_file.h>
41 #include <linux/root_dev.h>
43 #include <linux/of_pci.h>
44 #include <linux/kexec.h>
47 #include <asm/processor.h>
49 #include <asm/pgtable.h>
52 #include <asm/pci-bridge.h>
53 #include <asm/iommu.h>
55 #include <asm/machdep.h>
58 #include <asm/nvram.h>
61 #include <asm/ppc-pci.h>
62 #include <asm/i8259.h>
65 #include <asm/firmware.h>
68 #include <asm/plpar_wrappers.h>
74 unsigned long CMO_PageSize = (ASM_CONST(1) << IOMMU_PAGE_SHIFT_4K);
75 EXPORT_SYMBOL(CMO_PageSize);
77 int fwnmi_active; /* TRUE if an FWNMI handler is present */
79 static void pSeries_show_cpuinfo(struct seq_file *m)
81 struct device_node *root;
82 const char *model = "";
84 root = of_find_node_by_path("/");
86 model = of_get_property(root, "model", NULL);
87 seq_printf(m, "machine\t\t: CHRP %s\n", model);
91 /* Initialize firmware assisted non-maskable interrupts if
92 * the firmware supports this feature.
94 static void __init fwnmi_init(void)
96 unsigned long system_reset_addr, machine_check_addr;
98 int ibm_nmi_register = rtas_token("ibm,nmi-register");
99 if (ibm_nmi_register == RTAS_UNKNOWN_SERVICE)
102 /* If the kernel's not linked at zero we point the firmware at low
103 * addresses anyway, and use a trampoline to get to the real code. */
104 system_reset_addr = __pa(system_reset_fwnmi) - PHYSICAL_START;
105 machine_check_addr = __pa(machine_check_fwnmi) - PHYSICAL_START;
107 if (0 == rtas_call(ibm_nmi_register, 2, 1, NULL, system_reset_addr,
112 static void pseries_8259_cascade(struct irq_desc *desc)
114 struct irq_chip *chip = irq_desc_get_chip(desc);
115 unsigned int cascade_irq = i8259_irq();
117 if (cascade_irq != NO_IRQ)
118 generic_handle_irq(cascade_irq);
120 chip->irq_eoi(&desc->irq_data);
123 static void __init pseries_setup_i8259_cascade(void)
125 struct device_node *np, *old, *found = NULL;
126 unsigned int cascade;
128 unsigned long intack = 0;
131 for_each_node_by_type(np, "interrupt-controller") {
132 if (of_device_is_compatible(np, "chrp,iic")) {
139 printk(KERN_DEBUG "pic: no ISA interrupt controller\n");
143 cascade = irq_of_parse_and_map(found, 0);
144 if (cascade == NO_IRQ) {
145 printk(KERN_ERR "pic: failed to map cascade interrupt");
148 pr_debug("pic: cascade mapped to irq %d\n", cascade);
150 for (old = of_node_get(found); old != NULL ; old = np) {
151 np = of_get_parent(old);
155 if (strcmp(np->name, "pci") != 0)
157 addrp = of_get_property(np, "8259-interrupt-acknowledge", NULL);
160 naddr = of_n_addr_cells(np);
161 intack = addrp[naddr-1];
163 intack |= ((unsigned long)addrp[naddr-2]) << 32;
166 printk(KERN_DEBUG "pic: PCI 8259 intack at 0x%016lx\n", intack);
167 i8259_init(found, intack);
169 irq_set_chained_handler(cascade, pseries_8259_cascade);
172 static void __init pseries_init_irq(void)
175 pseries_setup_i8259_cascade();
178 static void pseries_lpar_enable_pmcs(void)
180 unsigned long set, reset;
184 plpar_hcall_norets(H_PERFMON, set, reset);
187 static int pci_dn_reconfig_notifier(struct notifier_block *nb, unsigned long action, void *data)
189 struct of_reconfig_data *rd = data;
190 struct device_node *parent, *np = rd->dn;
195 case OF_RECONFIG_ATTACH_NODE:
196 parent = of_get_parent(np);
197 pdn = parent ? PCI_DN(parent) : NULL;
199 pci_add_device_node_info(pdn->phb, np);
203 case OF_RECONFIG_DETACH_NODE:
206 list_del(&pdn->list);
215 static struct notifier_block pci_dn_reconfig_nb = {
216 .notifier_call = pci_dn_reconfig_notifier,
219 struct kmem_cache *dtl_cache;
221 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
223 * Allocate space for the dispatch trace log for all possible cpus
224 * and register the buffers with the hypervisor. This is used for
225 * computing time stolen by the hypervisor.
227 static int alloc_dispatch_logs(void)
230 struct paca_struct *pp;
231 struct dtl_entry *dtl;
233 if (!firmware_has_feature(FW_FEATURE_SPLPAR))
239 for_each_possible_cpu(cpu) {
241 dtl = kmem_cache_alloc(dtl_cache, GFP_KERNEL);
243 pr_warn("Failed to allocate dispatch trace log for cpu %d\n",
245 pr_warn("Stolen time statistics will be unreliable\n");
250 pp->dispatch_log = dtl;
251 pp->dispatch_log_end = dtl + N_DISPATCH_LOG;
255 /* Register the DTL for the current (boot) cpu */
256 dtl = get_paca()->dispatch_log;
257 get_paca()->dtl_ridx = 0;
258 get_paca()->dtl_curr = dtl;
259 get_paca()->lppaca_ptr->dtl_idx = 0;
261 /* hypervisor reads buffer length from this field */
262 dtl->enqueue_to_dispatch_time = cpu_to_be32(DISPATCH_LOG_BYTES);
263 ret = register_dtl(hard_smp_processor_id(), __pa(dtl));
265 pr_err("WARNING: DTL registration of cpu %d (hw %d) failed "
266 "with %d\n", smp_processor_id(),
267 hard_smp_processor_id(), ret);
268 get_paca()->lppaca_ptr->dtl_enable_mask = 2;
272 #else /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
273 static inline int alloc_dispatch_logs(void)
277 #endif /* CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
279 static int alloc_dispatch_log_kmem_cache(void)
281 dtl_cache = kmem_cache_create("dtl", DISPATCH_LOG_BYTES,
282 DISPATCH_LOG_BYTES, 0, NULL);
284 pr_warn("Failed to create dispatch trace log buffer cache\n");
285 pr_warn("Stolen time statistics will be unreliable\n");
289 return alloc_dispatch_logs();
291 machine_early_initcall(pseries, alloc_dispatch_log_kmem_cache);
293 static void pseries_lpar_idle(void)
296 * Default handler to go into low thread priority and possibly
297 * low power mode by ceding processor to hypervisor
300 /* Indicate to hypervisor that we are idle. */
301 get_lppaca()->idle = 1;
304 * Yield the processor to the hypervisor. We return if
305 * an external interrupt occurs (which are driven prior
306 * to returning here) or if a prod occurs from another
307 * processor. When returning here, external interrupts
312 get_lppaca()->idle = 0;
316 * Enable relocation on during exceptions. This has partition wide scope and
317 * may take a while to complete, if it takes longer than one second we will
318 * just give up rather than wasting any more time on this - if that turns out
319 * to ever be a problem in practice we can move this into a kernel thread to
320 * finish off the process later in boot.
322 long pSeries_enable_reloc_on_exc(void)
325 unsigned int delay, total_delay = 0;
328 rc = enable_reloc_on_exceptions();
329 if (!H_IS_LONG_BUSY(rc))
332 delay = get_longbusy_msecs(rc);
333 total_delay += delay;
334 if (total_delay > 1000) {
335 pr_warn("Warning: Giving up waiting to enable "
336 "relocation on exceptions (%u msec)!\n",
344 EXPORT_SYMBOL(pSeries_enable_reloc_on_exc);
346 long pSeries_disable_reloc_on_exc(void)
351 rc = disable_reloc_on_exceptions();
352 if (!H_IS_LONG_BUSY(rc))
354 mdelay(get_longbusy_msecs(rc));
357 EXPORT_SYMBOL(pSeries_disable_reloc_on_exc);
360 static void pSeries_machine_kexec(struct kimage *image)
364 if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
365 rc = pSeries_disable_reloc_on_exc();
367 pr_warning("Warning: Failed to disable relocation on "
368 "exceptions: %ld\n", rc);
371 default_machine_kexec(image);
375 #ifdef __LITTLE_ENDIAN__
376 long pseries_big_endian_exceptions(void)
381 rc = enable_big_endian_exceptions();
382 if (!H_IS_LONG_BUSY(rc))
384 mdelay(get_longbusy_msecs(rc));
388 static long pseries_little_endian_exceptions(void)
393 rc = enable_little_endian_exceptions();
394 if (!H_IS_LONG_BUSY(rc))
396 mdelay(get_longbusy_msecs(rc));
401 static void __init find_and_init_phbs(void)
403 struct device_node *node;
404 struct pci_controller *phb;
405 struct device_node *root = of_find_node_by_path("/");
407 for_each_child_of_node(root, node) {
408 if (node->type == NULL || (strcmp(node->type, "pci") != 0 &&
409 strcmp(node->type, "pciex") != 0))
412 phb = pcibios_alloc_controller(node);
416 pci_process_bridge_OF_ranges(phb, node, 0);
417 isa_bridge_find_early(phb);
418 phb->controller_ops = pseries_pci_controller_ops;
424 * PCI_PROBE_ONLY and PCI_REASSIGN_ALL_BUS can be set via properties
427 of_pci_check_probe_only();
430 static void __init pSeries_setup_arch(void)
432 set_arch_panic_timeout(10, ARCH_PANIC_TIMEOUT);
434 /* Discover PIC type and setup ppc_md accordingly */
438 /* openpic global configuration register (64-bit format). */
439 /* openpic Interrupt Source Unit pointer (64-bit format). */
440 /* python0 facility area (mmio) (64-bit format) REAL address. */
442 /* init to some ~sane value until calibrate_delay() runs */
443 loops_per_jiffy = 50000000;
447 /* By default, only probe PCI (can be overridden by rtas_pci) */
448 pci_add_flags(PCI_PROBE_ONLY);
450 /* Find and initialize PCI host bridges */
451 init_pci_config_tokens();
452 find_and_init_phbs();
453 of_reconfig_notifier_register(&pci_dn_reconfig_nb);
455 pSeries_nvram_init();
457 if (firmware_has_feature(FW_FEATURE_LPAR)) {
458 vpa_init(boot_cpuid);
459 ppc_md.power_save = pseries_lpar_idle;
460 ppc_md.enable_pmcs = pseries_lpar_enable_pmcs;
462 /* No special idle routine */
463 ppc_md.enable_pmcs = power4_enable_pmcs;
466 ppc_md.pcibios_root_bridge_prepare = pseries_root_bridge_prepare;
468 if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
471 rc = pSeries_enable_reloc_on_exc();
473 pr_info("Relocation on exceptions not supported\n");
474 } else if (rc != H_SUCCESS) {
475 pr_warn("Unable to enable relocation on exceptions: "
481 static int __init pSeries_init_panel(void)
483 /* Manually leave the kernel version on the panel. */
484 #ifdef __BIG_ENDIAN__
485 ppc_md.progress("Linux ppc64\n", 0);
487 ppc_md.progress("Linux ppc64le\n", 0);
489 ppc_md.progress(init_utsname()->version, 0);
493 machine_arch_initcall(pseries, pSeries_init_panel);
495 static int pseries_set_dabr(unsigned long dabr, unsigned long dabrx)
497 return plpar_hcall_norets(H_SET_DABR, dabr);
500 static int pseries_set_xdabr(unsigned long dabr, unsigned long dabrx)
502 /* Have to set at least one bit in the DABRX according to PAPR */
503 if (dabrx == 0 && dabr == 0)
505 /* PAPR says we can only set kernel and user bits */
506 dabrx &= DABRX_KERNEL | DABRX_USER;
508 return plpar_hcall_norets(H_SET_XDABR, dabr, dabrx);
511 static int pseries_set_dawr(unsigned long dawr, unsigned long dawrx)
513 /* PAPR says we can't set HYP */
516 return plapr_set_watchpoint0(dawr, dawrx);
519 #define CMO_CHARACTERISTICS_TOKEN 44
520 #define CMO_MAXLENGTH 1026
522 void pSeries_coalesce_init(void)
524 struct hvcall_mpp_x_data mpp_x_data;
526 if (firmware_has_feature(FW_FEATURE_CMO) && !h_get_mpp_x(&mpp_x_data))
527 powerpc_firmware_features |= FW_FEATURE_XCMO;
529 powerpc_firmware_features &= ~FW_FEATURE_XCMO;
533 * fw_cmo_feature_init - FW_FEATURE_CMO is not stored in ibm,hypertas-functions,
534 * handle that here. (Stolen from parse_system_parameter_string)
536 static void pSeries_cmo_feature_init(void)
538 char *ptr, *key, *value, *end;
540 int page_order = IOMMU_PAGE_SHIFT_4K;
542 pr_debug(" -> fw_cmo_feature_init()\n");
543 spin_lock(&rtas_data_buf_lock);
544 memset(rtas_data_buf, 0, RTAS_DATA_BUF_SIZE);
545 call_status = rtas_call(rtas_token("ibm,get-system-parameter"), 3, 1,
547 CMO_CHARACTERISTICS_TOKEN,
551 if (call_status != 0) {
552 spin_unlock(&rtas_data_buf_lock);
553 pr_debug("CMO not available\n");
554 pr_debug(" <- fw_cmo_feature_init()\n");
558 end = rtas_data_buf + CMO_MAXLENGTH - 2;
559 ptr = rtas_data_buf + 2; /* step over strlen value */
562 while (*ptr && (ptr <= end)) {
563 /* Separate the key and value by replacing '=' with '\0' and
564 * point the value at the string after the '='
569 } else if (ptr[0] == '\0' || ptr[0] == ',') {
570 /* Terminate the string containing the key/value pair */
574 pr_debug("Malformed key/value pair\n");
575 /* Never found a '=', end processing */
579 if (0 == strcmp(key, "CMOPageSize"))
580 page_order = simple_strtol(value, NULL, 10);
581 else if (0 == strcmp(key, "PrPSP"))
582 CMO_PrPSP = simple_strtol(value, NULL, 10);
583 else if (0 == strcmp(key, "SecPSP"))
584 CMO_SecPSP = simple_strtol(value, NULL, 10);
585 value = key = ptr + 1;
590 /* Page size is returned as the power of 2 of the page size,
591 * convert to the page size in bytes before returning
593 CMO_PageSize = 1 << page_order;
594 pr_debug("CMO_PageSize = %lu\n", CMO_PageSize);
596 if (CMO_PrPSP != -1 || CMO_SecPSP != -1) {
597 pr_info("CMO enabled\n");
598 pr_debug("CMO enabled, PrPSP=%d, SecPSP=%d\n", CMO_PrPSP,
600 powerpc_firmware_features |= FW_FEATURE_CMO;
601 pSeries_coalesce_init();
603 pr_debug("CMO not enabled, PrPSP=%d, SecPSP=%d\n", CMO_PrPSP,
605 spin_unlock(&rtas_data_buf_lock);
606 pr_debug(" <- fw_cmo_feature_init()\n");
610 * Early initialization. Relocation is on but do not reference unbolted pages
612 static void __init pSeries_init_early(void)
614 pr_debug(" -> pSeries_init_early()\n");
616 #ifdef CONFIG_HVC_CONSOLE
617 if (firmware_has_feature(FW_FEATURE_LPAR))
618 hvc_vio_init_early();
620 if (firmware_has_feature(FW_FEATURE_XDABR))
621 ppc_md.set_dabr = pseries_set_xdabr;
622 else if (firmware_has_feature(FW_FEATURE_DABR))
623 ppc_md.set_dabr = pseries_set_dabr;
625 if (firmware_has_feature(FW_FEATURE_SET_MODE))
626 ppc_md.set_dawr = pseries_set_dawr;
628 pSeries_cmo_feature_init();
629 iommu_init_early_pSeries();
631 pr_debug(" <- pSeries_init_early()\n");
635 * pseries_power_off - tell firmware about how to power off the system.
637 * This function calls either the power-off rtas token in normal cases
638 * or the ibm,power-off-ups token (if present & requested) in case of
639 * a power failure. If power-off token is used, power on will only be
640 * possible with power button press. If ibm,power-off-ups token is used
641 * it will allow auto poweron after power is restored.
643 static void pseries_power_off(void)
646 int rtas_poweroff_ups_token = rtas_token("ibm,power-off-ups");
648 if (rtas_flash_term_hook)
649 rtas_flash_term_hook(SYS_POWER_OFF);
651 if (rtas_poweron_auto == 0 ||
652 rtas_poweroff_ups_token == RTAS_UNKNOWN_SERVICE) {
653 rc = rtas_call(rtas_token("power-off"), 2, 1, NULL, -1, -1);
654 printk(KERN_INFO "RTAS power-off returned %d\n", rc);
656 rc = rtas_call(rtas_poweroff_ups_token, 0, 1, NULL);
657 printk(KERN_INFO "RTAS ibm,power-off-ups returned %d\n", rc);
663 * Called very early, MMU is off, device-tree isn't unflattened
666 static int __init pseries_probe_fw_features(unsigned long node,
667 const char *uname, int depth,
672 static int hypertas_found;
673 static int vec5_found;
678 if (!strcmp(uname, "rtas") || !strcmp(uname, "rtas@0")) {
679 prop = of_get_flat_dt_prop(node, "ibm,hypertas-functions",
682 powerpc_firmware_features |= FW_FEATURE_LPAR;
683 fw_hypertas_feature_init(prop, len);
689 if (!strcmp(uname, "chosen")) {
690 prop = of_get_flat_dt_prop(node, "ibm,architecture-vec-5",
693 fw_vec5_feature_init(prop, len);
698 return hypertas_found && vec5_found;
701 static int __init pSeries_probe(void)
703 unsigned long root = of_get_flat_dt_root();
704 const char *dtype = of_get_flat_dt_prop(root, "device_type", NULL);
708 if (strcmp(dtype, "chrp"))
711 /* Cell blades firmware claims to be chrp while it's not. Until this
712 * is fixed, we need to avoid those here.
714 if (of_flat_dt_is_compatible(root, "IBM,CPBW-1.0") ||
715 of_flat_dt_is_compatible(root, "IBM,CBEA"))
718 pr_debug("pSeries detected, looking for LPAR capability...\n");
720 /* Now try to figure out if we are running on LPAR */
721 of_scan_flat_dt(pseries_probe_fw_features, NULL);
723 #ifdef __LITTLE_ENDIAN__
724 if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
727 * Tell the hypervisor that we want our exceptions to
728 * be taken in little endian mode. If this fails we don't
729 * want to use BUG() because it will trigger an exception.
731 rc = pseries_little_endian_exceptions();
733 ppc_md.progress("H_SET_MODE LE exception fail", 0);
734 panic("Could not enable little endian exceptions");
739 if (firmware_has_feature(FW_FEATURE_LPAR))
744 pm_power_off = pseries_power_off;
746 pr_debug("Machine is%s LPAR !\n",
747 (powerpc_firmware_features & FW_FEATURE_LPAR) ? "" : " not");
752 static int pSeries_pci_probe_mode(struct pci_bus *bus)
754 if (firmware_has_feature(FW_FEATURE_LPAR))
755 return PCI_PROBE_DEVTREE;
756 return PCI_PROBE_NORMAL;
759 struct pci_controller_ops pseries_pci_controller_ops = {
760 .probe_mode = pSeries_pci_probe_mode,
763 define_machine(pseries) {
765 .probe = pSeries_probe,
766 .setup_arch = pSeries_setup_arch,
767 .init_early = pSeries_init_early,
768 .init_IRQ = pseries_init_irq,
769 .show_cpuinfo = pSeries_show_cpuinfo,
770 .log_error = pSeries_log_error,
771 .pcibios_fixup = pSeries_final_fixup,
772 .restart = rtas_restart,
774 .panic = rtas_os_term,
775 .get_boot_time = rtas_get_boot_time,
776 .get_rtc_time = rtas_get_rtc_time,
777 .set_rtc_time = rtas_set_rtc_time,
778 .calibrate_decr = generic_calibrate_decr,
779 .progress = rtas_progress,
780 .system_reset_exception = pSeries_system_reset_exception,
781 .machine_check_exception = pSeries_machine_check_exception,
783 .machine_kexec = pSeries_machine_kexec,
784 .kexec_cpu_down = pseries_kexec_cpu_down,
786 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
787 .memory_block_size = pseries_memory_block_size,