Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / arch / powerpc / platforms / pseries / setup.c
1 /*
2  *  64-bit pSeries and RS/6000 setup code.
3  *
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
8  *
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.
13  */
14
15 /*
16  * bootup setup stuff..
17  */
18
19 #include <linux/cpu.h>
20 #include <linux/errno.h>
21 #include <linux/sched.h>
22 #include <linux/kernel.h>
23 #include <linux/mm.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>
42 #include <linux/of.h>
43 #include <linux/kexec.h>
44
45 #include <asm/mmu.h>
46 #include <asm/processor.h>
47 #include <asm/io.h>
48 #include <asm/pgtable.h>
49 #include <asm/prom.h>
50 #include <asm/rtas.h>
51 #include <asm/pci-bridge.h>
52 #include <asm/iommu.h>
53 #include <asm/dma.h>
54 #include <asm/machdep.h>
55 #include <asm/irq.h>
56 #include <asm/time.h>
57 #include <asm/nvram.h>
58 #include <asm/pmc.h>
59 #include <asm/mpic.h>
60 #include <asm/xics.h>
61 #include <asm/ppc-pci.h>
62 #include <asm/i8259.h>
63 #include <asm/udbg.h>
64 #include <asm/smp.h>
65 #include <asm/firmware.h>
66 #include <asm/eeh.h>
67 #include <asm/reg.h>
68 #include <asm/plpar_wrappers.h>
69
70 #include "pseries.h"
71
72 int CMO_PrPSP = -1;
73 int CMO_SecPSP = -1;
74 unsigned long CMO_PageSize = (ASM_CONST(1) << IOMMU_PAGE_SHIFT_4K);
75 EXPORT_SYMBOL(CMO_PageSize);
76
77 int fwnmi_active;  /* TRUE if an FWNMI handler is present */
78
79 static struct device_node *pSeries_mpic_node;
80
81 static void pSeries_show_cpuinfo(struct seq_file *m)
82 {
83         struct device_node *root;
84         const char *model = "";
85
86         root = of_find_node_by_path("/");
87         if (root)
88                 model = of_get_property(root, "model", NULL);
89         seq_printf(m, "machine\t\t: CHRP %s\n", model);
90         of_node_put(root);
91 }
92
93 /* Initialize firmware assisted non-maskable interrupts if
94  * the firmware supports this feature.
95  */
96 static void __init fwnmi_init(void)
97 {
98         unsigned long system_reset_addr, machine_check_addr;
99
100         int ibm_nmi_register = rtas_token("ibm,nmi-register");
101         if (ibm_nmi_register == RTAS_UNKNOWN_SERVICE)
102                 return;
103
104         /* If the kernel's not linked at zero we point the firmware at low
105          * addresses anyway, and use a trampoline to get to the real code. */
106         system_reset_addr  = __pa(system_reset_fwnmi) - PHYSICAL_START;
107         machine_check_addr = __pa(machine_check_fwnmi) - PHYSICAL_START;
108
109         if (0 == rtas_call(ibm_nmi_register, 2, 1, NULL, system_reset_addr,
110                                 machine_check_addr))
111                 fwnmi_active = 1;
112 }
113
114 static void pseries_8259_cascade(unsigned int irq, struct irq_desc *desc)
115 {
116         struct irq_chip *chip = irq_desc_get_chip(desc);
117         unsigned int cascade_irq = i8259_irq();
118
119         if (cascade_irq != NO_IRQ)
120                 generic_handle_irq(cascade_irq);
121
122         chip->irq_eoi(&desc->irq_data);
123 }
124
125 static void __init pseries_setup_i8259_cascade(void)
126 {
127         struct device_node *np, *old, *found = NULL;
128         unsigned int cascade;
129         const u32 *addrp;
130         unsigned long intack = 0;
131         int naddr;
132
133         for_each_node_by_type(np, "interrupt-controller") {
134                 if (of_device_is_compatible(np, "chrp,iic")) {
135                         found = np;
136                         break;
137                 }
138         }
139
140         if (found == NULL) {
141                 printk(KERN_DEBUG "pic: no ISA interrupt controller\n");
142                 return;
143         }
144
145         cascade = irq_of_parse_and_map(found, 0);
146         if (cascade == NO_IRQ) {
147                 printk(KERN_ERR "pic: failed to map cascade interrupt");
148                 return;
149         }
150         pr_debug("pic: cascade mapped to irq %d\n", cascade);
151
152         for (old = of_node_get(found); old != NULL ; old = np) {
153                 np = of_get_parent(old);
154                 of_node_put(old);
155                 if (np == NULL)
156                         break;
157                 if (strcmp(np->name, "pci") != 0)
158                         continue;
159                 addrp = of_get_property(np, "8259-interrupt-acknowledge", NULL);
160                 if (addrp == NULL)
161                         continue;
162                 naddr = of_n_addr_cells(np);
163                 intack = addrp[naddr-1];
164                 if (naddr > 1)
165                         intack |= ((unsigned long)addrp[naddr-2]) << 32;
166         }
167         if (intack)
168                 printk(KERN_DEBUG "pic: PCI 8259 intack at 0x%016lx\n", intack);
169         i8259_init(found, intack);
170         of_node_put(found);
171         irq_set_chained_handler(cascade, pseries_8259_cascade);
172 }
173
174 static void __init pseries_mpic_init_IRQ(void)
175 {
176         struct device_node *np;
177         const unsigned int *opprop;
178         unsigned long openpic_addr = 0;
179         int naddr, n, i, opplen;
180         struct mpic *mpic;
181
182         np = of_find_node_by_path("/");
183         naddr = of_n_addr_cells(np);
184         opprop = of_get_property(np, "platform-open-pic", &opplen);
185         if (opprop != NULL) {
186                 openpic_addr = of_read_number(opprop, naddr);
187                 printk(KERN_DEBUG "OpenPIC addr: %lx\n", openpic_addr);
188         }
189         of_node_put(np);
190
191         BUG_ON(openpic_addr == 0);
192
193         /* Setup the openpic driver */
194         mpic = mpic_alloc(pSeries_mpic_node, openpic_addr,
195                         MPIC_NO_RESET, 16, 0, " MPIC     ");
196         BUG_ON(mpic == NULL);
197
198         /* Add ISUs */
199         opplen /= sizeof(u32);
200         for (n = 0, i = naddr; i < opplen; i += naddr, n++) {
201                 unsigned long isuaddr = of_read_number(opprop + i, naddr);
202                 mpic_assign_isu(mpic, n, isuaddr);
203         }
204
205         /* Setup top-level get_irq */
206         ppc_md.get_irq = mpic_get_irq;
207
208         /* All ISUs are setup, complete initialization */
209         mpic_init(mpic);
210
211         /* Look for cascade */
212         pseries_setup_i8259_cascade();
213 }
214
215 static void __init pseries_xics_init_IRQ(void)
216 {
217         xics_init();
218         pseries_setup_i8259_cascade();
219 }
220
221 static void pseries_lpar_enable_pmcs(void)
222 {
223         unsigned long set, reset;
224
225         set = 1UL << 63;
226         reset = 0;
227         plpar_hcall_norets(H_PERFMON, set, reset);
228 }
229
230 static void __init pseries_discover_pic(void)
231 {
232         struct device_node *np;
233         const char *typep;
234
235         for_each_node_by_name(np, "interrupt-controller") {
236                 typep = of_get_property(np, "compatible", NULL);
237                 if (strstr(typep, "open-pic")) {
238                         pSeries_mpic_node = of_node_get(np);
239                         ppc_md.init_IRQ       = pseries_mpic_init_IRQ;
240                         setup_kexec_cpu_down_mpic();
241                         smp_init_pseries_mpic();
242                         return;
243                 } else if (strstr(typep, "ppc-xicp")) {
244                         ppc_md.init_IRQ       = pseries_xics_init_IRQ;
245                         setup_kexec_cpu_down_xics();
246                         smp_init_pseries_xics();
247                         return;
248                 }
249         }
250         printk(KERN_ERR "pSeries_discover_pic: failed to recognize"
251                " interrupt-controller\n");
252 }
253
254 static int pci_dn_reconfig_notifier(struct notifier_block *nb, unsigned long action, void *data)
255 {
256         struct of_reconfig_data *rd = data;
257         struct device_node *np = rd->dn;
258         struct pci_dn *pci = NULL;
259         int err = NOTIFY_OK;
260
261         switch (action) {
262         case OF_RECONFIG_ATTACH_NODE:
263                 pci = np->parent->data;
264                 if (pci) {
265                         update_dn_pci_info(np, pci->phb);
266
267                         /* Create EEH device for the OF node */
268                         eeh_dev_init(PCI_DN(np), pci->phb);
269                 }
270                 break;
271         default:
272                 err = NOTIFY_DONE;
273                 break;
274         }
275         return err;
276 }
277
278 static struct notifier_block pci_dn_reconfig_nb = {
279         .notifier_call = pci_dn_reconfig_notifier,
280 };
281
282 struct kmem_cache *dtl_cache;
283
284 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
285 /*
286  * Allocate space for the dispatch trace log for all possible cpus
287  * and register the buffers with the hypervisor.  This is used for
288  * computing time stolen by the hypervisor.
289  */
290 static int alloc_dispatch_logs(void)
291 {
292         int cpu, ret;
293         struct paca_struct *pp;
294         struct dtl_entry *dtl;
295
296         if (!firmware_has_feature(FW_FEATURE_SPLPAR))
297                 return 0;
298
299         if (!dtl_cache)
300                 return 0;
301
302         for_each_possible_cpu(cpu) {
303                 pp = &paca[cpu];
304                 dtl = kmem_cache_alloc(dtl_cache, GFP_KERNEL);
305                 if (!dtl) {
306                         pr_warn("Failed to allocate dispatch trace log for cpu %d\n",
307                                 cpu);
308                         pr_warn("Stolen time statistics will be unreliable\n");
309                         break;
310                 }
311
312                 pp->dtl_ridx = 0;
313                 pp->dispatch_log = dtl;
314                 pp->dispatch_log_end = dtl + N_DISPATCH_LOG;
315                 pp->dtl_curr = dtl;
316         }
317
318         /* Register the DTL for the current (boot) cpu */
319         dtl = get_paca()->dispatch_log;
320         get_paca()->dtl_ridx = 0;
321         get_paca()->dtl_curr = dtl;
322         get_paca()->lppaca_ptr->dtl_idx = 0;
323
324         /* hypervisor reads buffer length from this field */
325         dtl->enqueue_to_dispatch_time = cpu_to_be32(DISPATCH_LOG_BYTES);
326         ret = register_dtl(hard_smp_processor_id(), __pa(dtl));
327         if (ret)
328                 pr_err("WARNING: DTL registration of cpu %d (hw %d) failed "
329                        "with %d\n", smp_processor_id(),
330                        hard_smp_processor_id(), ret);
331         get_paca()->lppaca_ptr->dtl_enable_mask = 2;
332
333         return 0;
334 }
335 #else /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
336 static inline int alloc_dispatch_logs(void)
337 {
338         return 0;
339 }
340 #endif /* CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
341
342 static int alloc_dispatch_log_kmem_cache(void)
343 {
344         dtl_cache = kmem_cache_create("dtl", DISPATCH_LOG_BYTES,
345                                                 DISPATCH_LOG_BYTES, 0, NULL);
346         if (!dtl_cache) {
347                 pr_warn("Failed to create dispatch trace log buffer cache\n");
348                 pr_warn("Stolen time statistics will be unreliable\n");
349                 return 0;
350         }
351
352         return alloc_dispatch_logs();
353 }
354 machine_early_initcall(pseries, alloc_dispatch_log_kmem_cache);
355
356 static void pseries_lpar_idle(void)
357 {
358         /*
359          * Default handler to go into low thread priority and possibly
360          * low power mode by cedeing processor to hypervisor
361          */
362
363         /* Indicate to hypervisor that we are idle. */
364         get_lppaca()->idle = 1;
365
366         /*
367          * Yield the processor to the hypervisor.  We return if
368          * an external interrupt occurs (which are driven prior
369          * to returning here) or if a prod occurs from another
370          * processor. When returning here, external interrupts
371          * are enabled.
372          */
373         cede_processor();
374
375         get_lppaca()->idle = 0;
376 }
377
378 /*
379  * Enable relocation on during exceptions. This has partition wide scope and
380  * may take a while to complete, if it takes longer than one second we will
381  * just give up rather than wasting any more time on this - if that turns out
382  * to ever be a problem in practice we can move this into a kernel thread to
383  * finish off the process later in boot.
384  */
385 long pSeries_enable_reloc_on_exc(void)
386 {
387         long rc;
388         unsigned int delay, total_delay = 0;
389
390         while (1) {
391                 rc = enable_reloc_on_exceptions();
392                 if (!H_IS_LONG_BUSY(rc))
393                         return rc;
394
395                 delay = get_longbusy_msecs(rc);
396                 total_delay += delay;
397                 if (total_delay > 1000) {
398                         pr_warn("Warning: Giving up waiting to enable "
399                                 "relocation on exceptions (%u msec)!\n",
400                                 total_delay);
401                         return rc;
402                 }
403
404                 mdelay(delay);
405         }
406 }
407 EXPORT_SYMBOL(pSeries_enable_reloc_on_exc);
408
409 long pSeries_disable_reloc_on_exc(void)
410 {
411         long rc;
412
413         while (1) {
414                 rc = disable_reloc_on_exceptions();
415                 if (!H_IS_LONG_BUSY(rc))
416                         return rc;
417                 mdelay(get_longbusy_msecs(rc));
418         }
419 }
420 EXPORT_SYMBOL(pSeries_disable_reloc_on_exc);
421
422 #ifdef CONFIG_KEXEC
423 static void pSeries_machine_kexec(struct kimage *image)
424 {
425         long rc;
426
427         if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
428                 rc = pSeries_disable_reloc_on_exc();
429                 if (rc != H_SUCCESS)
430                         pr_warning("Warning: Failed to disable relocation on "
431                                    "exceptions: %ld\n", rc);
432         }
433
434         default_machine_kexec(image);
435 }
436 #endif
437
438 #ifdef __LITTLE_ENDIAN__
439 long pseries_big_endian_exceptions(void)
440 {
441         long rc;
442
443         while (1) {
444                 rc = enable_big_endian_exceptions();
445                 if (!H_IS_LONG_BUSY(rc))
446                         return rc;
447                 mdelay(get_longbusy_msecs(rc));
448         }
449 }
450
451 static long pseries_little_endian_exceptions(void)
452 {
453         long rc;
454
455         while (1) {
456                 rc = enable_little_endian_exceptions();
457                 if (!H_IS_LONG_BUSY(rc))
458                         return rc;
459                 mdelay(get_longbusy_msecs(rc));
460         }
461 }
462 #endif
463
464 static void __init find_and_init_phbs(void)
465 {
466         struct device_node *node;
467         struct pci_controller *phb;
468         struct device_node *root = of_find_node_by_path("/");
469
470         for_each_child_of_node(root, node) {
471                 if (node->type == NULL || (strcmp(node->type, "pci") != 0 &&
472                                            strcmp(node->type, "pciex") != 0))
473                         continue;
474
475                 phb = pcibios_alloc_controller(node);
476                 if (!phb)
477                         continue;
478                 rtas_setup_phb(phb);
479                 pci_process_bridge_OF_ranges(phb, node, 0);
480                 isa_bridge_find_early(phb);
481                 phb->controller_ops = pseries_pci_controller_ops;
482         }
483
484         of_node_put(root);
485         pci_devs_phb_init();
486
487         /*
488          * PCI_PROBE_ONLY and PCI_REASSIGN_ALL_BUS can be set via properties
489          * in chosen.
490          */
491         if (of_chosen) {
492                 const int *prop;
493
494                 prop = of_get_property(of_chosen,
495                                 "linux,pci-probe-only", NULL);
496                 if (prop) {
497                         if (*prop)
498                                 pci_add_flags(PCI_PROBE_ONLY);
499                         else
500                                 pci_clear_flags(PCI_PROBE_ONLY);
501                 }
502         }
503 }
504
505 static void __init pSeries_setup_arch(void)
506 {
507         set_arch_panic_timeout(10, ARCH_PANIC_TIMEOUT);
508
509         /* Discover PIC type and setup ppc_md accordingly */
510         pseries_discover_pic();
511
512         /* openpic global configuration register (64-bit format). */
513         /* openpic Interrupt Source Unit pointer (64-bit format). */
514         /* python0 facility area (mmio) (64-bit format) REAL address. */
515
516         /* init to some ~sane value until calibrate_delay() runs */
517         loops_per_jiffy = 50000000;
518
519         fwnmi_init();
520
521         /* By default, only probe PCI (can be overriden by rtas_pci) */
522         pci_add_flags(PCI_PROBE_ONLY);
523
524         /* Find and initialize PCI host bridges */
525         init_pci_config_tokens();
526         find_and_init_phbs();
527         of_reconfig_notifier_register(&pci_dn_reconfig_nb);
528
529         pSeries_nvram_init();
530
531         if (firmware_has_feature(FW_FEATURE_LPAR)) {
532                 vpa_init(boot_cpuid);
533                 ppc_md.power_save = pseries_lpar_idle;
534                 ppc_md.enable_pmcs = pseries_lpar_enable_pmcs;
535         } else {
536                 /* No special idle routine */
537                 ppc_md.enable_pmcs = power4_enable_pmcs;
538         }
539
540         ppc_md.pcibios_root_bridge_prepare = pseries_root_bridge_prepare;
541
542         if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
543                 long rc;
544
545                 rc = pSeries_enable_reloc_on_exc();
546                 if (rc == H_P2) {
547                         pr_info("Relocation on exceptions not supported\n");
548                 } else if (rc != H_SUCCESS) {
549                         pr_warn("Unable to enable relocation on exceptions: "
550                                 "%ld\n", rc);
551                 }
552         }
553 }
554
555 static int __init pSeries_init_panel(void)
556 {
557         /* Manually leave the kernel version on the panel. */
558 #ifdef __BIG_ENDIAN__
559         ppc_md.progress("Linux ppc64\n", 0);
560 #else
561         ppc_md.progress("Linux ppc64le\n", 0);
562 #endif
563         ppc_md.progress(init_utsname()->version, 0);
564
565         return 0;
566 }
567 machine_arch_initcall(pseries, pSeries_init_panel);
568
569 static int pseries_set_dabr(unsigned long dabr, unsigned long dabrx)
570 {
571         return plpar_hcall_norets(H_SET_DABR, dabr);
572 }
573
574 static int pseries_set_xdabr(unsigned long dabr, unsigned long dabrx)
575 {
576         /* Have to set at least one bit in the DABRX according to PAPR */
577         if (dabrx == 0 && dabr == 0)
578                 dabrx = DABRX_USER;
579         /* PAPR says we can only set kernel and user bits */
580         dabrx &= DABRX_KERNEL | DABRX_USER;
581
582         return plpar_hcall_norets(H_SET_XDABR, dabr, dabrx);
583 }
584
585 static int pseries_set_dawr(unsigned long dawr, unsigned long dawrx)
586 {
587         /* PAPR says we can't set HYP */
588         dawrx &= ~DAWRX_HYP;
589
590         return  plapr_set_watchpoint0(dawr, dawrx);
591 }
592
593 #define CMO_CHARACTERISTICS_TOKEN 44
594 #define CMO_MAXLENGTH 1026
595
596 void pSeries_coalesce_init(void)
597 {
598         struct hvcall_mpp_x_data mpp_x_data;
599
600         if (firmware_has_feature(FW_FEATURE_CMO) && !h_get_mpp_x(&mpp_x_data))
601                 powerpc_firmware_features |= FW_FEATURE_XCMO;
602         else
603                 powerpc_firmware_features &= ~FW_FEATURE_XCMO;
604 }
605
606 /**
607  * fw_cmo_feature_init - FW_FEATURE_CMO is not stored in ibm,hypertas-functions,
608  * handle that here. (Stolen from parse_system_parameter_string)
609  */
610 static void pSeries_cmo_feature_init(void)
611 {
612         char *ptr, *key, *value, *end;
613         int call_status;
614         int page_order = IOMMU_PAGE_SHIFT_4K;
615
616         pr_debug(" -> fw_cmo_feature_init()\n");
617         spin_lock(&rtas_data_buf_lock);
618         memset(rtas_data_buf, 0, RTAS_DATA_BUF_SIZE);
619         call_status = rtas_call(rtas_token("ibm,get-system-parameter"), 3, 1,
620                                 NULL,
621                                 CMO_CHARACTERISTICS_TOKEN,
622                                 __pa(rtas_data_buf),
623                                 RTAS_DATA_BUF_SIZE);
624
625         if (call_status != 0) {
626                 spin_unlock(&rtas_data_buf_lock);
627                 pr_debug("CMO not available\n");
628                 pr_debug(" <- fw_cmo_feature_init()\n");
629                 return;
630         }
631
632         end = rtas_data_buf + CMO_MAXLENGTH - 2;
633         ptr = rtas_data_buf + 2;        /* step over strlen value */
634         key = value = ptr;
635
636         while (*ptr && (ptr <= end)) {
637                 /* Separate the key and value by replacing '=' with '\0' and
638                  * point the value at the string after the '='
639                  */
640                 if (ptr[0] == '=') {
641                         ptr[0] = '\0';
642                         value = ptr + 1;
643                 } else if (ptr[0] == '\0' || ptr[0] == ',') {
644                         /* Terminate the string containing the key/value pair */
645                         ptr[0] = '\0';
646
647                         if (key == value) {
648                                 pr_debug("Malformed key/value pair\n");
649                                 /* Never found a '=', end processing */
650                                 break;
651                         }
652
653                         if (0 == strcmp(key, "CMOPageSize"))
654                                 page_order = simple_strtol(value, NULL, 10);
655                         else if (0 == strcmp(key, "PrPSP"))
656                                 CMO_PrPSP = simple_strtol(value, NULL, 10);
657                         else if (0 == strcmp(key, "SecPSP"))
658                                 CMO_SecPSP = simple_strtol(value, NULL, 10);
659                         value = key = ptr + 1;
660                 }
661                 ptr++;
662         }
663
664         /* Page size is returned as the power of 2 of the page size,
665          * convert to the page size in bytes before returning
666          */
667         CMO_PageSize = 1 << page_order;
668         pr_debug("CMO_PageSize = %lu\n", CMO_PageSize);
669
670         if (CMO_PrPSP != -1 || CMO_SecPSP != -1) {
671                 pr_info("CMO enabled\n");
672                 pr_debug("CMO enabled, PrPSP=%d, SecPSP=%d\n", CMO_PrPSP,
673                          CMO_SecPSP);
674                 powerpc_firmware_features |= FW_FEATURE_CMO;
675                 pSeries_coalesce_init();
676         } else
677                 pr_debug("CMO not enabled, PrPSP=%d, SecPSP=%d\n", CMO_PrPSP,
678                          CMO_SecPSP);
679         spin_unlock(&rtas_data_buf_lock);
680         pr_debug(" <- fw_cmo_feature_init()\n");
681 }
682
683 /*
684  * Early initialization.  Relocation is on but do not reference unbolted pages
685  */
686 static void __init pSeries_init_early(void)
687 {
688         pr_debug(" -> pSeries_init_early()\n");
689
690 #ifdef CONFIG_HVC_CONSOLE
691         if (firmware_has_feature(FW_FEATURE_LPAR))
692                 hvc_vio_init_early();
693 #endif
694         if (firmware_has_feature(FW_FEATURE_XDABR))
695                 ppc_md.set_dabr = pseries_set_xdabr;
696         else if (firmware_has_feature(FW_FEATURE_DABR))
697                 ppc_md.set_dabr = pseries_set_dabr;
698
699         if (firmware_has_feature(FW_FEATURE_SET_MODE))
700                 ppc_md.set_dawr = pseries_set_dawr;
701
702         pSeries_cmo_feature_init();
703         iommu_init_early_pSeries();
704
705         pr_debug(" <- pSeries_init_early()\n");
706 }
707
708 /**
709  * pseries_power_off - tell firmware about how to power off the system.
710  *
711  * This function calls either the power-off rtas token in normal cases
712  * or the ibm,power-off-ups token (if present & requested) in case of
713  * a power failure. If power-off token is used, power on will only be
714  * possible with power button press. If ibm,power-off-ups token is used
715  * it will allow auto poweron after power is restored.
716  */
717 static void pseries_power_off(void)
718 {
719         int rc;
720         int rtas_poweroff_ups_token = rtas_token("ibm,power-off-ups");
721
722         if (rtas_flash_term_hook)
723                 rtas_flash_term_hook(SYS_POWER_OFF);
724
725         if (rtas_poweron_auto == 0 ||
726                 rtas_poweroff_ups_token == RTAS_UNKNOWN_SERVICE) {
727                 rc = rtas_call(rtas_token("power-off"), 2, 1, NULL, -1, -1);
728                 printk(KERN_INFO "RTAS power-off returned %d\n", rc);
729         } else {
730                 rc = rtas_call(rtas_poweroff_ups_token, 0, 1, NULL);
731                 printk(KERN_INFO "RTAS ibm,power-off-ups returned %d\n", rc);
732         }
733         for (;;);
734 }
735
736 /*
737  * Called very early, MMU is off, device-tree isn't unflattened
738  */
739
740 static int __init pseries_probe_fw_features(unsigned long node,
741                                             const char *uname, int depth,
742                                             void *data)
743 {
744         const char *prop;
745         int len;
746         static int hypertas_found;
747         static int vec5_found;
748
749         if (depth != 1)
750                 return 0;
751
752         if (!strcmp(uname, "rtas") || !strcmp(uname, "rtas@0")) {
753                 prop = of_get_flat_dt_prop(node, "ibm,hypertas-functions",
754                                            &len);
755                 if (prop) {
756                         powerpc_firmware_features |= FW_FEATURE_LPAR;
757                         fw_hypertas_feature_init(prop, len);
758                 }
759
760                 hypertas_found = 1;
761         }
762
763         if (!strcmp(uname, "chosen")) {
764                 prop = of_get_flat_dt_prop(node, "ibm,architecture-vec-5",
765                                            &len);
766                 if (prop)
767                         fw_vec5_feature_init(prop, len);
768
769                 vec5_found = 1;
770         }
771
772         return hypertas_found && vec5_found;
773 }
774
775 static int __init pSeries_probe(void)
776 {
777         unsigned long root = of_get_flat_dt_root();
778         const char *dtype = of_get_flat_dt_prop(root, "device_type", NULL);
779
780         if (dtype == NULL)
781                 return 0;
782         if (strcmp(dtype, "chrp"))
783                 return 0;
784
785         /* Cell blades firmware claims to be chrp while it's not. Until this
786          * is fixed, we need to avoid those here.
787          */
788         if (of_flat_dt_is_compatible(root, "IBM,CPBW-1.0") ||
789             of_flat_dt_is_compatible(root, "IBM,CBEA"))
790                 return 0;
791
792         pr_debug("pSeries detected, looking for LPAR capability...\n");
793
794         /* Now try to figure out if we are running on LPAR */
795         of_scan_flat_dt(pseries_probe_fw_features, NULL);
796
797 #ifdef __LITTLE_ENDIAN__
798         if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
799                 long rc;
800                 /*
801                  * Tell the hypervisor that we want our exceptions to
802                  * be taken in little endian mode. If this fails we don't
803                  * want to use BUG() because it will trigger an exception.
804                  */
805                 rc = pseries_little_endian_exceptions();
806                 if (rc) {
807                         ppc_md.progress("H_SET_MODE LE exception fail", 0);
808                         panic("Could not enable little endian exceptions");
809                 }
810         }
811 #endif
812
813         if (firmware_has_feature(FW_FEATURE_LPAR))
814                 hpte_init_lpar();
815         else
816                 hpte_init_native();
817
818         pm_power_off = pseries_power_off;
819
820         pr_debug("Machine is%s LPAR !\n",
821                  (powerpc_firmware_features & FW_FEATURE_LPAR) ? "" : " not");
822
823         return 1;
824 }
825
826 static int pSeries_pci_probe_mode(struct pci_bus *bus)
827 {
828         if (firmware_has_feature(FW_FEATURE_LPAR))
829                 return PCI_PROBE_DEVTREE;
830         return PCI_PROBE_NORMAL;
831 }
832
833 #ifndef CONFIG_PCI
834 void pSeries_final_fixup(void) { }
835 #endif
836
837 struct pci_controller_ops pseries_pci_controller_ops = {
838         .probe_mode             = pSeries_pci_probe_mode,
839 };
840
841 define_machine(pseries) {
842         .name                   = "pSeries",
843         .probe                  = pSeries_probe,
844         .setup_arch             = pSeries_setup_arch,
845         .init_early             = pSeries_init_early,
846         .show_cpuinfo           = pSeries_show_cpuinfo,
847         .log_error              = pSeries_log_error,
848         .pcibios_fixup          = pSeries_final_fixup,
849         .restart                = rtas_restart,
850         .halt                   = rtas_halt,
851         .panic                  = rtas_os_term,
852         .get_boot_time          = rtas_get_boot_time,
853         .get_rtc_time           = rtas_get_rtc_time,
854         .set_rtc_time           = rtas_set_rtc_time,
855         .calibrate_decr         = generic_calibrate_decr,
856         .progress               = rtas_progress,
857         .system_reset_exception = pSeries_system_reset_exception,
858         .machine_check_exception = pSeries_machine_check_exception,
859 #ifdef CONFIG_KEXEC
860         .machine_kexec          = pSeries_machine_kexec,
861 #endif
862 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
863         .memory_block_size      = pseries_memory_block_size,
864 #endif
865 };