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>
46 #include <asm/processor.h>
48 #include <asm/pgtable.h>
51 #include <asm/pci-bridge.h>
52 #include <asm/iommu.h>
54 #include <asm/machdep.h>
57 #include <asm/nvram.h>
60 #include <asm/ppc-pci.h>
61 #include <asm/i8259.h>
64 #include <asm/firmware.h>
67 #include <asm/plpar_wrappers.h>
68 #include <asm/kexec.h>
69 #include <asm/isa-bridge.h>
75 unsigned long CMO_PageSize
= (ASM_CONST(1) << IOMMU_PAGE_SHIFT_4K
);
76 EXPORT_SYMBOL(CMO_PageSize
);
78 int fwnmi_active
; /* TRUE if an FWNMI handler is present */
80 static void pSeries_show_cpuinfo(struct seq_file
*m
)
82 struct device_node
*root
;
83 const char *model
= "";
85 root
= of_find_node_by_path("/");
87 model
= of_get_property(root
, "model", NULL
);
88 seq_printf(m
, "machine\t\t: CHRP %s\n", model
);
91 seq_printf(m
, "MMU\t\t: Radix\n");
93 seq_printf(m
, "MMU\t\t: Hash\n");
96 /* Initialize firmware assisted non-maskable interrupts if
97 * the firmware supports this feature.
99 static void __init
fwnmi_init(void)
101 unsigned long system_reset_addr
, machine_check_addr
;
103 int ibm_nmi_register
= rtas_token("ibm,nmi-register");
104 if (ibm_nmi_register
== RTAS_UNKNOWN_SERVICE
)
107 /* If the kernel's not linked at zero we point the firmware at low
108 * addresses anyway, and use a trampoline to get to the real code. */
109 system_reset_addr
= __pa(system_reset_fwnmi
) - PHYSICAL_START
;
110 machine_check_addr
= __pa(machine_check_fwnmi
) - PHYSICAL_START
;
112 if (0 == rtas_call(ibm_nmi_register
, 2, 1, NULL
, system_reset_addr
,
117 static void pseries_8259_cascade(struct irq_desc
*desc
)
119 struct irq_chip
*chip
= irq_desc_get_chip(desc
);
120 unsigned int cascade_irq
= i8259_irq();
123 generic_handle_irq(cascade_irq
);
125 chip
->irq_eoi(&desc
->irq_data
);
128 static void __init
pseries_setup_i8259_cascade(void)
130 struct device_node
*np
, *old
, *found
= NULL
;
131 unsigned int cascade
;
133 unsigned long intack
= 0;
136 for_each_node_by_type(np
, "interrupt-controller") {
137 if (of_device_is_compatible(np
, "chrp,iic")) {
144 printk(KERN_DEBUG
"pic: no ISA interrupt controller\n");
148 cascade
= irq_of_parse_and_map(found
, 0);
150 printk(KERN_ERR
"pic: failed to map cascade interrupt");
153 pr_debug("pic: cascade mapped to irq %d\n", cascade
);
155 for (old
= of_node_get(found
); old
!= NULL
; old
= np
) {
156 np
= of_get_parent(old
);
160 if (strcmp(np
->name
, "pci") != 0)
162 addrp
= of_get_property(np
, "8259-interrupt-acknowledge", NULL
);
165 naddr
= of_n_addr_cells(np
);
166 intack
= addrp
[naddr
-1];
168 intack
|= ((unsigned long)addrp
[naddr
-2]) << 32;
171 printk(KERN_DEBUG
"pic: PCI 8259 intack at 0x%016lx\n", intack
);
172 i8259_init(found
, intack
);
174 irq_set_chained_handler(cascade
, pseries_8259_cascade
);
177 static void __init
pseries_init_irq(void)
180 pseries_setup_i8259_cascade();
183 static void pseries_lpar_enable_pmcs(void)
185 unsigned long set
, reset
;
189 plpar_hcall_norets(H_PERFMON
, set
, reset
);
192 static int pci_dn_reconfig_notifier(struct notifier_block
*nb
, unsigned long action
, void *data
)
194 struct of_reconfig_data
*rd
= data
;
195 struct device_node
*parent
, *np
= rd
->dn
;
200 case OF_RECONFIG_ATTACH_NODE
:
201 parent
= of_get_parent(np
);
202 pdn
= parent
? PCI_DN(parent
) : NULL
;
204 pci_add_device_node_info(pdn
->phb
, np
);
208 case OF_RECONFIG_DETACH_NODE
:
211 list_del(&pdn
->list
);
220 static struct notifier_block pci_dn_reconfig_nb
= {
221 .notifier_call
= pci_dn_reconfig_notifier
,
224 struct kmem_cache
*dtl_cache
;
226 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
228 * Allocate space for the dispatch trace log for all possible cpus
229 * and register the buffers with the hypervisor. This is used for
230 * computing time stolen by the hypervisor.
232 static int alloc_dispatch_logs(void)
235 struct paca_struct
*pp
;
236 struct dtl_entry
*dtl
;
238 if (!firmware_has_feature(FW_FEATURE_SPLPAR
))
244 for_each_possible_cpu(cpu
) {
246 dtl
= kmem_cache_alloc(dtl_cache
, GFP_KERNEL
);
248 pr_warn("Failed to allocate dispatch trace log for cpu %d\n",
250 pr_warn("Stolen time statistics will be unreliable\n");
255 pp
->dispatch_log
= dtl
;
256 pp
->dispatch_log_end
= dtl
+ N_DISPATCH_LOG
;
260 /* Register the DTL for the current (boot) cpu */
261 dtl
= get_paca()->dispatch_log
;
262 get_paca()->dtl_ridx
= 0;
263 get_paca()->dtl_curr
= dtl
;
264 get_paca()->lppaca_ptr
->dtl_idx
= 0;
266 /* hypervisor reads buffer length from this field */
267 dtl
->enqueue_to_dispatch_time
= cpu_to_be32(DISPATCH_LOG_BYTES
);
268 ret
= register_dtl(hard_smp_processor_id(), __pa(dtl
));
270 pr_err("WARNING: DTL registration of cpu %d (hw %d) failed "
271 "with %d\n", smp_processor_id(),
272 hard_smp_processor_id(), ret
);
273 get_paca()->lppaca_ptr
->dtl_enable_mask
= 2;
277 #else /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
278 static inline int alloc_dispatch_logs(void)
282 #endif /* CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
284 static int alloc_dispatch_log_kmem_cache(void)
286 dtl_cache
= kmem_cache_create("dtl", DISPATCH_LOG_BYTES
,
287 DISPATCH_LOG_BYTES
, 0, NULL
);
289 pr_warn("Failed to create dispatch trace log buffer cache\n");
290 pr_warn("Stolen time statistics will be unreliable\n");
294 return alloc_dispatch_logs();
296 machine_early_initcall(pseries
, alloc_dispatch_log_kmem_cache
);
298 static void pseries_lpar_idle(void)
301 * Default handler to go into low thread priority and possibly
302 * low power mode by ceding processor to hypervisor
305 /* Indicate to hypervisor that we are idle. */
306 get_lppaca()->idle
= 1;
309 * Yield the processor to the hypervisor. We return if
310 * an external interrupt occurs (which are driven prior
311 * to returning here) or if a prod occurs from another
312 * processor. When returning here, external interrupts
317 get_lppaca()->idle
= 0;
321 * Enable relocation on during exceptions. This has partition wide scope and
322 * may take a while to complete, if it takes longer than one second we will
323 * just give up rather than wasting any more time on this - if that turns out
324 * to ever be a problem in practice we can move this into a kernel thread to
325 * finish off the process later in boot.
327 void pseries_enable_reloc_on_exc(void)
330 unsigned int delay
, total_delay
= 0;
333 rc
= enable_reloc_on_exceptions();
334 if (!H_IS_LONG_BUSY(rc
)) {
336 pr_info("Relocation on exceptions not"
338 } else if (rc
!= H_SUCCESS
) {
339 pr_warn("Unable to enable relocation"
340 " on exceptions: %ld\n", rc
);
345 delay
= get_longbusy_msecs(rc
);
346 total_delay
+= delay
;
347 if (total_delay
> 1000) {
348 pr_warn("Warning: Giving up waiting to enable "
349 "relocation on exceptions (%u msec)!\n",
357 EXPORT_SYMBOL(pseries_enable_reloc_on_exc
);
359 void pseries_disable_reloc_on_exc(void)
364 rc
= disable_reloc_on_exceptions();
365 if (!H_IS_LONG_BUSY(rc
))
367 mdelay(get_longbusy_msecs(rc
));
370 pr_warning("Warning: Failed to disable relocation on "
371 "exceptions: %ld\n", rc
);
373 EXPORT_SYMBOL(pseries_disable_reloc_on_exc
);
375 #ifdef CONFIG_KEXEC_CORE
376 static void pSeries_machine_kexec(struct kimage
*image
)
378 if (firmware_has_feature(FW_FEATURE_SET_MODE
))
379 pseries_disable_reloc_on_exc();
381 default_machine_kexec(image
);
385 #ifdef __LITTLE_ENDIAN__
386 void pseries_big_endian_exceptions(void)
391 rc
= enable_big_endian_exceptions();
392 if (!H_IS_LONG_BUSY(rc
))
394 mdelay(get_longbusy_msecs(rc
));
398 * At this point it is unlikely panic() will get anything
399 * out to the user, since this is called very late in kexec
400 * but at least this will stop us from continuing on further
401 * and creating an even more difficult to debug situation.
403 * There is a known problem when kdump'ing, if cpus are offline
404 * the above call will fail. Rather than panicking again, keep
405 * going and hope the kdump kernel is also little endian, which
408 if (rc
&& !kdump_in_progress())
409 panic("Could not enable big endian exceptions");
412 void pseries_little_endian_exceptions(void)
417 rc
= enable_little_endian_exceptions();
418 if (!H_IS_LONG_BUSY(rc
))
420 mdelay(get_longbusy_msecs(rc
));
423 ppc_md
.progress("H_SET_MODE LE exception fail", 0);
424 panic("Could not enable little endian exceptions");
429 static void __init
find_and_init_phbs(void)
431 struct device_node
*node
;
432 struct pci_controller
*phb
;
433 struct device_node
*root
= of_find_node_by_path("/");
435 for_each_child_of_node(root
, node
) {
436 if (node
->type
== NULL
|| (strcmp(node
->type
, "pci") != 0 &&
437 strcmp(node
->type
, "pciex") != 0))
440 phb
= pcibios_alloc_controller(node
);
444 pci_process_bridge_OF_ranges(phb
, node
, 0);
445 isa_bridge_find_early(phb
);
446 phb
->controller_ops
= pseries_pci_controller_ops
;
452 * PCI_PROBE_ONLY and PCI_REASSIGN_ALL_BUS can be set via properties
455 of_pci_check_probe_only();
458 static void pseries_setup_rfi_flush(void)
460 struct h_cpu_char_result result
;
461 enum l1d_flush_type types
;
465 /* Enable by default */
467 types
= L1D_FLUSH_FALLBACK
;
469 rc
= plpar_get_cpu_characteristics(&result
);
470 if (rc
== H_SUCCESS
) {
471 if (result
.character
& H_CPU_CHAR_L1D_FLUSH_TRIG2
)
472 types
|= L1D_FLUSH_MTTRIG
;
473 if (result
.character
& H_CPU_CHAR_L1D_FLUSH_ORI30
)
474 types
|= L1D_FLUSH_ORI
;
476 if ((!(result
.behaviour
& H_CPU_BEHAV_L1D_FLUSH_PR
)) ||
477 (!(result
.behaviour
& H_CPU_BEHAV_FAVOUR_SECURITY
)))
481 setup_rfi_flush(types
, enable
);
484 static void __init
pSeries_setup_arch(void)
486 set_arch_panic_timeout(10, ARCH_PANIC_TIMEOUT
);
488 /* Discover PIC type and setup ppc_md accordingly */
492 /* openpic global configuration register (64-bit format). */
493 /* openpic Interrupt Source Unit pointer (64-bit format). */
494 /* python0 facility area (mmio) (64-bit format) REAL address. */
496 /* init to some ~sane value until calibrate_delay() runs */
497 loops_per_jiffy
= 50000000;
501 pseries_setup_rfi_flush();
503 /* By default, only probe PCI (can be overridden by rtas_pci) */
504 pci_add_flags(PCI_PROBE_ONLY
);
506 /* Find and initialize PCI host bridges */
507 init_pci_config_tokens();
508 find_and_init_phbs();
509 of_reconfig_notifier_register(&pci_dn_reconfig_nb
);
511 pSeries_nvram_init();
513 if (firmware_has_feature(FW_FEATURE_LPAR
)) {
514 vpa_init(boot_cpuid
);
515 ppc_md
.power_save
= pseries_lpar_idle
;
516 ppc_md
.enable_pmcs
= pseries_lpar_enable_pmcs
;
518 /* No special idle routine */
519 ppc_md
.enable_pmcs
= power4_enable_pmcs
;
522 ppc_md
.pcibios_root_bridge_prepare
= pseries_root_bridge_prepare
;
525 static int __init
pSeries_init_panel(void)
527 /* Manually leave the kernel version on the panel. */
528 #ifdef __BIG_ENDIAN__
529 ppc_md
.progress("Linux ppc64\n", 0);
531 ppc_md
.progress("Linux ppc64le\n", 0);
533 ppc_md
.progress(init_utsname()->version
, 0);
537 machine_arch_initcall(pseries
, pSeries_init_panel
);
539 static int pseries_set_dabr(unsigned long dabr
, unsigned long dabrx
)
541 return plpar_hcall_norets(H_SET_DABR
, dabr
);
544 static int pseries_set_xdabr(unsigned long dabr
, unsigned long dabrx
)
546 /* Have to set at least one bit in the DABRX according to PAPR */
547 if (dabrx
== 0 && dabr
== 0)
549 /* PAPR says we can only set kernel and user bits */
550 dabrx
&= DABRX_KERNEL
| DABRX_USER
;
552 return plpar_hcall_norets(H_SET_XDABR
, dabr
, dabrx
);
555 static int pseries_set_dawr(unsigned long dawr
, unsigned long dawrx
)
557 /* PAPR says we can't set HYP */
560 return plapr_set_watchpoint0(dawr
, dawrx
);
563 #define CMO_CHARACTERISTICS_TOKEN 44
564 #define CMO_MAXLENGTH 1026
566 void pSeries_coalesce_init(void)
568 struct hvcall_mpp_x_data mpp_x_data
;
570 if (firmware_has_feature(FW_FEATURE_CMO
) && !h_get_mpp_x(&mpp_x_data
))
571 powerpc_firmware_features
|= FW_FEATURE_XCMO
;
573 powerpc_firmware_features
&= ~FW_FEATURE_XCMO
;
577 * fw_cmo_feature_init - FW_FEATURE_CMO is not stored in ibm,hypertas-functions,
578 * handle that here. (Stolen from parse_system_parameter_string)
580 static void pSeries_cmo_feature_init(void)
582 char *ptr
, *key
, *value
, *end
;
584 int page_order
= IOMMU_PAGE_SHIFT_4K
;
586 pr_debug(" -> fw_cmo_feature_init()\n");
587 spin_lock(&rtas_data_buf_lock
);
588 memset(rtas_data_buf
, 0, RTAS_DATA_BUF_SIZE
);
589 call_status
= rtas_call(rtas_token("ibm,get-system-parameter"), 3, 1,
591 CMO_CHARACTERISTICS_TOKEN
,
595 if (call_status
!= 0) {
596 spin_unlock(&rtas_data_buf_lock
);
597 pr_debug("CMO not available\n");
598 pr_debug(" <- fw_cmo_feature_init()\n");
602 end
= rtas_data_buf
+ CMO_MAXLENGTH
- 2;
603 ptr
= rtas_data_buf
+ 2; /* step over strlen value */
606 while (*ptr
&& (ptr
<= end
)) {
607 /* Separate the key and value by replacing '=' with '\0' and
608 * point the value at the string after the '='
613 } else if (ptr
[0] == '\0' || ptr
[0] == ',') {
614 /* Terminate the string containing the key/value pair */
618 pr_debug("Malformed key/value pair\n");
619 /* Never found a '=', end processing */
623 if (0 == strcmp(key
, "CMOPageSize"))
624 page_order
= simple_strtol(value
, NULL
, 10);
625 else if (0 == strcmp(key
, "PrPSP"))
626 CMO_PrPSP
= simple_strtol(value
, NULL
, 10);
627 else if (0 == strcmp(key
, "SecPSP"))
628 CMO_SecPSP
= simple_strtol(value
, NULL
, 10);
629 value
= key
= ptr
+ 1;
634 /* Page size is returned as the power of 2 of the page size,
635 * convert to the page size in bytes before returning
637 CMO_PageSize
= 1 << page_order
;
638 pr_debug("CMO_PageSize = %lu\n", CMO_PageSize
);
640 if (CMO_PrPSP
!= -1 || CMO_SecPSP
!= -1) {
641 pr_info("CMO enabled\n");
642 pr_debug("CMO enabled, PrPSP=%d, SecPSP=%d\n", CMO_PrPSP
,
644 powerpc_firmware_features
|= FW_FEATURE_CMO
;
645 pSeries_coalesce_init();
647 pr_debug("CMO not enabled, PrPSP=%d, SecPSP=%d\n", CMO_PrPSP
,
649 spin_unlock(&rtas_data_buf_lock
);
650 pr_debug(" <- fw_cmo_feature_init()\n");
654 * Early initialization. Relocation is on but do not reference unbolted pages
656 static void __init
pseries_init(void)
658 pr_debug(" -> pseries_init()\n");
660 #ifdef CONFIG_HVC_CONSOLE
661 if (firmware_has_feature(FW_FEATURE_LPAR
))
662 hvc_vio_init_early();
664 if (firmware_has_feature(FW_FEATURE_XDABR
))
665 ppc_md
.set_dabr
= pseries_set_xdabr
;
666 else if (firmware_has_feature(FW_FEATURE_DABR
))
667 ppc_md
.set_dabr
= pseries_set_dabr
;
669 if (firmware_has_feature(FW_FEATURE_SET_MODE
))
670 ppc_md
.set_dawr
= pseries_set_dawr
;
672 pSeries_cmo_feature_init();
673 iommu_init_early_pSeries();
675 pr_debug(" <- pseries_init()\n");
679 * pseries_power_off - tell firmware about how to power off the system.
681 * This function calls either the power-off rtas token in normal cases
682 * or the ibm,power-off-ups token (if present & requested) in case of
683 * a power failure. If power-off token is used, power on will only be
684 * possible with power button press. If ibm,power-off-ups token is used
685 * it will allow auto poweron after power is restored.
687 static void pseries_power_off(void)
690 int rtas_poweroff_ups_token
= rtas_token("ibm,power-off-ups");
692 if (rtas_flash_term_hook
)
693 rtas_flash_term_hook(SYS_POWER_OFF
);
695 if (rtas_poweron_auto
== 0 ||
696 rtas_poweroff_ups_token
== RTAS_UNKNOWN_SERVICE
) {
697 rc
= rtas_call(rtas_token("power-off"), 2, 1, NULL
, -1, -1);
698 printk(KERN_INFO
"RTAS power-off returned %d\n", rc
);
700 rc
= rtas_call(rtas_poweroff_ups_token
, 0, 1, NULL
);
701 printk(KERN_INFO
"RTAS ibm,power-off-ups returned %d\n", rc
);
706 static int __init
pSeries_probe(void)
708 const char *dtype
= of_get_property(of_root
, "device_type", NULL
);
712 if (strcmp(dtype
, "chrp"))
715 /* Cell blades firmware claims to be chrp while it's not. Until this
716 * is fixed, we need to avoid those here.
718 if (of_machine_is_compatible("IBM,CPBW-1.0") ||
719 of_machine_is_compatible("IBM,CBEA"))
722 pm_power_off
= pseries_power_off
;
724 pr_debug("Machine is%s LPAR !\n",
725 (powerpc_firmware_features
& FW_FEATURE_LPAR
) ? "" : " not");
732 static int pSeries_pci_probe_mode(struct pci_bus
*bus
)
734 if (firmware_has_feature(FW_FEATURE_LPAR
))
735 return PCI_PROBE_DEVTREE
;
736 return PCI_PROBE_NORMAL
;
739 struct pci_controller_ops pseries_pci_controller_ops
= {
740 .probe_mode
= pSeries_pci_probe_mode
,
743 define_machine(pseries
) {
745 .probe
= pSeries_probe
,
746 .setup_arch
= pSeries_setup_arch
,
747 .init_IRQ
= pseries_init_irq
,
748 .show_cpuinfo
= pSeries_show_cpuinfo
,
749 .log_error
= pSeries_log_error
,
750 .pcibios_fixup
= pSeries_final_fixup
,
751 .restart
= rtas_restart
,
753 .get_boot_time
= rtas_get_boot_time
,
754 .get_rtc_time
= rtas_get_rtc_time
,
755 .set_rtc_time
= rtas_set_rtc_time
,
756 .calibrate_decr
= generic_calibrate_decr
,
757 .progress
= rtas_progress
,
758 .system_reset_exception
= pSeries_system_reset_exception
,
759 .machine_check_exception
= pSeries_machine_check_exception
,
760 #ifdef CONFIG_KEXEC_CORE
761 .machine_kexec
= pSeries_machine_kexec
,
762 .kexec_cpu_down
= pseries_kexec_cpu_down
,
764 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
765 .memory_block_size
= pseries_memory_block_size
,