xref: /linux/arch/powerpc/platforms/pseries/setup.c (revision 140eb5227767c6754742020a16d2691222b9c19b)
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/of_pci.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/xics.h>
60 #include <asm/xive.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 #include <asm/kexec.h>
70 #include <asm/isa-bridge.h>
71 
72 #include "pseries.h"
73 
74 int CMO_PrPSP = -1;
75 int CMO_SecPSP = -1;
76 unsigned long CMO_PageSize = (ASM_CONST(1) << IOMMU_PAGE_SHIFT_4K);
77 EXPORT_SYMBOL(CMO_PageSize);
78 
79 int fwnmi_active;  /* TRUE if an FWNMI handler is present */
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 	if (radix_enabled())
92 		seq_printf(m, "MMU\t\t: Radix\n");
93 	else
94 		seq_printf(m, "MMU\t\t: Hash\n");
95 }
96 
97 /* Initialize firmware assisted non-maskable interrupts if
98  * the firmware supports this feature.
99  */
100 static void __init fwnmi_init(void)
101 {
102 	unsigned long system_reset_addr, machine_check_addr;
103 
104 	int ibm_nmi_register = rtas_token("ibm,nmi-register");
105 	if (ibm_nmi_register == RTAS_UNKNOWN_SERVICE)
106 		return;
107 
108 	/* If the kernel's not linked at zero we point the firmware at low
109 	 * addresses anyway, and use a trampoline to get to the real code. */
110 	system_reset_addr  = __pa(system_reset_fwnmi) - PHYSICAL_START;
111 	machine_check_addr = __pa(machine_check_fwnmi) - PHYSICAL_START;
112 
113 	if (0 == rtas_call(ibm_nmi_register, 2, 1, NULL, system_reset_addr,
114 				machine_check_addr))
115 		fwnmi_active = 1;
116 }
117 
118 static void pseries_8259_cascade(struct irq_desc *desc)
119 {
120 	struct irq_chip *chip = irq_desc_get_chip(desc);
121 	unsigned int cascade_irq = i8259_irq();
122 
123 	if (cascade_irq)
124 		generic_handle_irq(cascade_irq);
125 
126 	chip->irq_eoi(&desc->irq_data);
127 }
128 
129 static void __init pseries_setup_i8259_cascade(void)
130 {
131 	struct device_node *np, *old, *found = NULL;
132 	unsigned int cascade;
133 	const u32 *addrp;
134 	unsigned long intack = 0;
135 	int naddr;
136 
137 	for_each_node_by_type(np, "interrupt-controller") {
138 		if (of_device_is_compatible(np, "chrp,iic")) {
139 			found = np;
140 			break;
141 		}
142 	}
143 
144 	if (found == NULL) {
145 		printk(KERN_DEBUG "pic: no ISA interrupt controller\n");
146 		return;
147 	}
148 
149 	cascade = irq_of_parse_and_map(found, 0);
150 	if (!cascade) {
151 		printk(KERN_ERR "pic: failed to map cascade interrupt");
152 		return;
153 	}
154 	pr_debug("pic: cascade mapped to irq %d\n", cascade);
155 
156 	for (old = of_node_get(found); old != NULL ; old = np) {
157 		np = of_get_parent(old);
158 		of_node_put(old);
159 		if (np == NULL)
160 			break;
161 		if (strcmp(np->name, "pci") != 0)
162 			continue;
163 		addrp = of_get_property(np, "8259-interrupt-acknowledge", NULL);
164 		if (addrp == NULL)
165 			continue;
166 		naddr = of_n_addr_cells(np);
167 		intack = addrp[naddr-1];
168 		if (naddr > 1)
169 			intack |= ((unsigned long)addrp[naddr-2]) << 32;
170 	}
171 	if (intack)
172 		printk(KERN_DEBUG "pic: PCI 8259 intack at 0x%016lx\n", intack);
173 	i8259_init(found, intack);
174 	of_node_put(found);
175 	irq_set_chained_handler(cascade, pseries_8259_cascade);
176 }
177 
178 static void __init pseries_init_irq(void)
179 {
180 	/* Try using a XIVE if available, otherwise use a XICS */
181 	if (!xive_spapr_init()) {
182 		xics_init();
183 		pseries_setup_i8259_cascade();
184 	}
185 }
186 
187 static void pseries_lpar_enable_pmcs(void)
188 {
189 	unsigned long set, reset;
190 
191 	set = 1UL << 63;
192 	reset = 0;
193 	plpar_hcall_norets(H_PERFMON, set, reset);
194 }
195 
196 static int pci_dn_reconfig_notifier(struct notifier_block *nb, unsigned long action, void *data)
197 {
198 	struct of_reconfig_data *rd = data;
199 	struct device_node *parent, *np = rd->dn;
200 	struct pci_dn *pdn;
201 	int err = NOTIFY_OK;
202 
203 	switch (action) {
204 	case OF_RECONFIG_ATTACH_NODE:
205 		parent = of_get_parent(np);
206 		pdn = parent ? PCI_DN(parent) : NULL;
207 		if (pdn)
208 			pci_add_device_node_info(pdn->phb, np);
209 
210 		of_node_put(parent);
211 		break;
212 	case OF_RECONFIG_DETACH_NODE:
213 		pdn = PCI_DN(np);
214 		if (pdn)
215 			list_del(&pdn->list);
216 		break;
217 	default:
218 		err = NOTIFY_DONE;
219 		break;
220 	}
221 	return err;
222 }
223 
224 static struct notifier_block pci_dn_reconfig_nb = {
225 	.notifier_call = pci_dn_reconfig_notifier,
226 };
227 
228 struct kmem_cache *dtl_cache;
229 
230 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
231 /*
232  * Allocate space for the dispatch trace log for all possible cpus
233  * and register the buffers with the hypervisor.  This is used for
234  * computing time stolen by the hypervisor.
235  */
236 static int alloc_dispatch_logs(void)
237 {
238 	int cpu, ret;
239 	struct paca_struct *pp;
240 	struct dtl_entry *dtl;
241 
242 	if (!firmware_has_feature(FW_FEATURE_SPLPAR))
243 		return 0;
244 
245 	if (!dtl_cache)
246 		return 0;
247 
248 	for_each_possible_cpu(cpu) {
249 		pp = &paca[cpu];
250 		dtl = kmem_cache_alloc(dtl_cache, GFP_KERNEL);
251 		if (!dtl) {
252 			pr_warn("Failed to allocate dispatch trace log for cpu %d\n",
253 				cpu);
254 			pr_warn("Stolen time statistics will be unreliable\n");
255 			break;
256 		}
257 
258 		pp->dtl_ridx = 0;
259 		pp->dispatch_log = dtl;
260 		pp->dispatch_log_end = dtl + N_DISPATCH_LOG;
261 		pp->dtl_curr = dtl;
262 	}
263 
264 	/* Register the DTL for the current (boot) cpu */
265 	dtl = get_paca()->dispatch_log;
266 	get_paca()->dtl_ridx = 0;
267 	get_paca()->dtl_curr = dtl;
268 	get_paca()->lppaca_ptr->dtl_idx = 0;
269 
270 	/* hypervisor reads buffer length from this field */
271 	dtl->enqueue_to_dispatch_time = cpu_to_be32(DISPATCH_LOG_BYTES);
272 	ret = register_dtl(hard_smp_processor_id(), __pa(dtl));
273 	if (ret)
274 		pr_err("WARNING: DTL registration of cpu %d (hw %d) failed "
275 		       "with %d\n", smp_processor_id(),
276 		       hard_smp_processor_id(), ret);
277 	get_paca()->lppaca_ptr->dtl_enable_mask = 2;
278 
279 	return 0;
280 }
281 #else /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
282 static inline int alloc_dispatch_logs(void)
283 {
284 	return 0;
285 }
286 #endif /* CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
287 
288 static int alloc_dispatch_log_kmem_cache(void)
289 {
290 	dtl_cache = kmem_cache_create("dtl", DISPATCH_LOG_BYTES,
291 						DISPATCH_LOG_BYTES, 0, NULL);
292 	if (!dtl_cache) {
293 		pr_warn("Failed to create dispatch trace log buffer cache\n");
294 		pr_warn("Stolen time statistics will be unreliable\n");
295 		return 0;
296 	}
297 
298 	return alloc_dispatch_logs();
299 }
300 machine_early_initcall(pseries, alloc_dispatch_log_kmem_cache);
301 
302 static void pseries_lpar_idle(void)
303 {
304 	/*
305 	 * Default handler to go into low thread priority and possibly
306 	 * low power mode by ceding processor to hypervisor
307 	 */
308 
309 	/* Indicate to hypervisor that we are idle. */
310 	get_lppaca()->idle = 1;
311 
312 	/*
313 	 * Yield the processor to the hypervisor.  We return if
314 	 * an external interrupt occurs (which are driven prior
315 	 * to returning here) or if a prod occurs from another
316 	 * processor. When returning here, external interrupts
317 	 * are enabled.
318 	 */
319 	cede_processor();
320 
321 	get_lppaca()->idle = 0;
322 }
323 
324 /*
325  * Enable relocation on during exceptions. This has partition wide scope and
326  * may take a while to complete, if it takes longer than one second we will
327  * just give up rather than wasting any more time on this - if that turns out
328  * to ever be a problem in practice we can move this into a kernel thread to
329  * finish off the process later in boot.
330  */
331 void pseries_enable_reloc_on_exc(void)
332 {
333 	long rc;
334 	unsigned int delay, total_delay = 0;
335 
336 	while (1) {
337 		rc = enable_reloc_on_exceptions();
338 		if (!H_IS_LONG_BUSY(rc)) {
339 			if (rc == H_P2) {
340 				pr_info("Relocation on exceptions not"
341 					" supported\n");
342 			} else if (rc != H_SUCCESS) {
343 				pr_warn("Unable to enable relocation"
344 					" on exceptions: %ld\n", rc);
345 			}
346 			break;
347 		}
348 
349 		delay = get_longbusy_msecs(rc);
350 		total_delay += delay;
351 		if (total_delay > 1000) {
352 			pr_warn("Warning: Giving up waiting to enable "
353 				"relocation on exceptions (%u msec)!\n",
354 				total_delay);
355 			return;
356 		}
357 
358 		mdelay(delay);
359 	}
360 }
361 EXPORT_SYMBOL(pseries_enable_reloc_on_exc);
362 
363 void pseries_disable_reloc_on_exc(void)
364 {
365 	long rc;
366 
367 	while (1) {
368 		rc = disable_reloc_on_exceptions();
369 		if (!H_IS_LONG_BUSY(rc))
370 			break;
371 		mdelay(get_longbusy_msecs(rc));
372 	}
373 	if (rc != H_SUCCESS)
374 		pr_warning("Warning: Failed to disable relocation on "
375 			   "exceptions: %ld\n", rc);
376 }
377 EXPORT_SYMBOL(pseries_disable_reloc_on_exc);
378 
379 #ifdef CONFIG_KEXEC_CORE
380 static void pSeries_machine_kexec(struct kimage *image)
381 {
382 	if (firmware_has_feature(FW_FEATURE_SET_MODE))
383 		pseries_disable_reloc_on_exc();
384 
385 	default_machine_kexec(image);
386 }
387 #endif
388 
389 #ifdef __LITTLE_ENDIAN__
390 void pseries_big_endian_exceptions(void)
391 {
392 	long rc;
393 
394 	while (1) {
395 		rc = enable_big_endian_exceptions();
396 		if (!H_IS_LONG_BUSY(rc))
397 			break;
398 		mdelay(get_longbusy_msecs(rc));
399 	}
400 
401 	/*
402 	 * At this point it is unlikely panic() will get anything
403 	 * out to the user, since this is called very late in kexec
404 	 * but at least this will stop us from continuing on further
405 	 * and creating an even more difficult to debug situation.
406 	 *
407 	 * There is a known problem when kdump'ing, if cpus are offline
408 	 * the above call will fail. Rather than panicking again, keep
409 	 * going and hope the kdump kernel is also little endian, which
410 	 * it usually is.
411 	 */
412 	if (rc && !kdump_in_progress())
413 		panic("Could not enable big endian exceptions");
414 }
415 
416 void pseries_little_endian_exceptions(void)
417 {
418 	long rc;
419 
420 	while (1) {
421 		rc = enable_little_endian_exceptions();
422 		if (!H_IS_LONG_BUSY(rc))
423 			break;
424 		mdelay(get_longbusy_msecs(rc));
425 	}
426 	if (rc) {
427 		ppc_md.progress("H_SET_MODE LE exception fail", 0);
428 		panic("Could not enable little endian exceptions");
429 	}
430 }
431 #endif
432 
433 static void __init find_and_init_phbs(void)
434 {
435 	struct device_node *node;
436 	struct pci_controller *phb;
437 	struct device_node *root = of_find_node_by_path("/");
438 
439 	for_each_child_of_node(root, node) {
440 		if (node->type == NULL || (strcmp(node->type, "pci") != 0 &&
441 					   strcmp(node->type, "pciex") != 0))
442 			continue;
443 
444 		phb = pcibios_alloc_controller(node);
445 		if (!phb)
446 			continue;
447 		rtas_setup_phb(phb);
448 		pci_process_bridge_OF_ranges(phb, node, 0);
449 		isa_bridge_find_early(phb);
450 		phb->controller_ops = pseries_pci_controller_ops;
451 	}
452 
453 	of_node_put(root);
454 
455 	/*
456 	 * PCI_PROBE_ONLY and PCI_REASSIGN_ALL_BUS can be set via properties
457 	 * in chosen.
458 	 */
459 	of_pci_check_probe_only();
460 }
461 
462 static void pseries_setup_rfi_flush(void)
463 {
464 	struct h_cpu_char_result result;
465 	enum l1d_flush_type types;
466 	bool enable;
467 	long rc;
468 
469 	/* Enable by default */
470 	enable = true;
471 
472 	rc = plpar_get_cpu_characteristics(&result);
473 	if (rc == H_SUCCESS) {
474 		types = L1D_FLUSH_NONE;
475 
476 		if (result.character & H_CPU_CHAR_L1D_FLUSH_TRIG2)
477 			types |= L1D_FLUSH_MTTRIG;
478 		if (result.character & H_CPU_CHAR_L1D_FLUSH_ORI30)
479 			types |= L1D_FLUSH_ORI;
480 
481 		/* Use fallback if nothing set in hcall */
482 		if (types == L1D_FLUSH_NONE)
483 			types = L1D_FLUSH_FALLBACK;
484 
485 		if (!(result.behaviour & H_CPU_BEHAV_L1D_FLUSH_PR))
486 			enable = false;
487 	} else {
488 		/* Default to fallback if case hcall is not available */
489 		types = L1D_FLUSH_FALLBACK;
490 	}
491 
492 	setup_rfi_flush(types, enable);
493 }
494 
495 static void __init pSeries_setup_arch(void)
496 {
497 	set_arch_panic_timeout(10, ARCH_PANIC_TIMEOUT);
498 
499 	/* Discover PIC type and setup ppc_md accordingly */
500 	smp_init_pseries();
501 
502 
503 	/* openpic global configuration register (64-bit format). */
504 	/* openpic Interrupt Source Unit pointer (64-bit format). */
505 	/* python0 facility area (mmio) (64-bit format) REAL address. */
506 
507 	/* init to some ~sane value until calibrate_delay() runs */
508 	loops_per_jiffy = 50000000;
509 
510 	fwnmi_init();
511 
512 	pseries_setup_rfi_flush();
513 
514 	/* By default, only probe PCI (can be overridden by rtas_pci) */
515 	pci_add_flags(PCI_PROBE_ONLY);
516 
517 	/* Find and initialize PCI host bridges */
518 	init_pci_config_tokens();
519 	find_and_init_phbs();
520 	of_reconfig_notifier_register(&pci_dn_reconfig_nb);
521 
522 	pSeries_nvram_init();
523 
524 	if (firmware_has_feature(FW_FEATURE_LPAR)) {
525 		vpa_init(boot_cpuid);
526 		ppc_md.power_save = pseries_lpar_idle;
527 		ppc_md.enable_pmcs = pseries_lpar_enable_pmcs;
528 	} else {
529 		/* No special idle routine */
530 		ppc_md.enable_pmcs = power4_enable_pmcs;
531 	}
532 
533 	ppc_md.pcibios_root_bridge_prepare = pseries_root_bridge_prepare;
534 }
535 
536 static int __init pSeries_init_panel(void)
537 {
538 	/* Manually leave the kernel version on the panel. */
539 #ifdef __BIG_ENDIAN__
540 	ppc_md.progress("Linux ppc64\n", 0);
541 #else
542 	ppc_md.progress("Linux ppc64le\n", 0);
543 #endif
544 	ppc_md.progress(init_utsname()->version, 0);
545 
546 	return 0;
547 }
548 machine_arch_initcall(pseries, pSeries_init_panel);
549 
550 static int pseries_set_dabr(unsigned long dabr, unsigned long dabrx)
551 {
552 	return plpar_hcall_norets(H_SET_DABR, dabr);
553 }
554 
555 static int pseries_set_xdabr(unsigned long dabr, unsigned long dabrx)
556 {
557 	/* Have to set at least one bit in the DABRX according to PAPR */
558 	if (dabrx == 0 && dabr == 0)
559 		dabrx = DABRX_USER;
560 	/* PAPR says we can only set kernel and user bits */
561 	dabrx &= DABRX_KERNEL | DABRX_USER;
562 
563 	return plpar_hcall_norets(H_SET_XDABR, dabr, dabrx);
564 }
565 
566 static int pseries_set_dawr(unsigned long dawr, unsigned long dawrx)
567 {
568 	/* PAPR says we can't set HYP */
569 	dawrx &= ~DAWRX_HYP;
570 
571 	return  plapr_set_watchpoint0(dawr, dawrx);
572 }
573 
574 #define CMO_CHARACTERISTICS_TOKEN 44
575 #define CMO_MAXLENGTH 1026
576 
577 void pSeries_coalesce_init(void)
578 {
579 	struct hvcall_mpp_x_data mpp_x_data;
580 
581 	if (firmware_has_feature(FW_FEATURE_CMO) && !h_get_mpp_x(&mpp_x_data))
582 		powerpc_firmware_features |= FW_FEATURE_XCMO;
583 	else
584 		powerpc_firmware_features &= ~FW_FEATURE_XCMO;
585 }
586 
587 /**
588  * fw_cmo_feature_init - FW_FEATURE_CMO is not stored in ibm,hypertas-functions,
589  * handle that here. (Stolen from parse_system_parameter_string)
590  */
591 static void pSeries_cmo_feature_init(void)
592 {
593 	char *ptr, *key, *value, *end;
594 	int call_status;
595 	int page_order = IOMMU_PAGE_SHIFT_4K;
596 
597 	pr_debug(" -> fw_cmo_feature_init()\n");
598 	spin_lock(&rtas_data_buf_lock);
599 	memset(rtas_data_buf, 0, RTAS_DATA_BUF_SIZE);
600 	call_status = rtas_call(rtas_token("ibm,get-system-parameter"), 3, 1,
601 				NULL,
602 				CMO_CHARACTERISTICS_TOKEN,
603 				__pa(rtas_data_buf),
604 				RTAS_DATA_BUF_SIZE);
605 
606 	if (call_status != 0) {
607 		spin_unlock(&rtas_data_buf_lock);
608 		pr_debug("CMO not available\n");
609 		pr_debug(" <- fw_cmo_feature_init()\n");
610 		return;
611 	}
612 
613 	end = rtas_data_buf + CMO_MAXLENGTH - 2;
614 	ptr = rtas_data_buf + 2;	/* step over strlen value */
615 	key = value = ptr;
616 
617 	while (*ptr && (ptr <= end)) {
618 		/* Separate the key and value by replacing '=' with '\0' and
619 		 * point the value at the string after the '='
620 		 */
621 		if (ptr[0] == '=') {
622 			ptr[0] = '\0';
623 			value = ptr + 1;
624 		} else if (ptr[0] == '\0' || ptr[0] == ',') {
625 			/* Terminate the string containing the key/value pair */
626 			ptr[0] = '\0';
627 
628 			if (key == value) {
629 				pr_debug("Malformed key/value pair\n");
630 				/* Never found a '=', end processing */
631 				break;
632 			}
633 
634 			if (0 == strcmp(key, "CMOPageSize"))
635 				page_order = simple_strtol(value, NULL, 10);
636 			else if (0 == strcmp(key, "PrPSP"))
637 				CMO_PrPSP = simple_strtol(value, NULL, 10);
638 			else if (0 == strcmp(key, "SecPSP"))
639 				CMO_SecPSP = simple_strtol(value, NULL, 10);
640 			value = key = ptr + 1;
641 		}
642 		ptr++;
643 	}
644 
645 	/* Page size is returned as the power of 2 of the page size,
646 	 * convert to the page size in bytes before returning
647 	 */
648 	CMO_PageSize = 1 << page_order;
649 	pr_debug("CMO_PageSize = %lu\n", CMO_PageSize);
650 
651 	if (CMO_PrPSP != -1 || CMO_SecPSP != -1) {
652 		pr_info("CMO enabled\n");
653 		pr_debug("CMO enabled, PrPSP=%d, SecPSP=%d\n", CMO_PrPSP,
654 		         CMO_SecPSP);
655 		powerpc_firmware_features |= FW_FEATURE_CMO;
656 		pSeries_coalesce_init();
657 	} else
658 		pr_debug("CMO not enabled, PrPSP=%d, SecPSP=%d\n", CMO_PrPSP,
659 		         CMO_SecPSP);
660 	spin_unlock(&rtas_data_buf_lock);
661 	pr_debug(" <- fw_cmo_feature_init()\n");
662 }
663 
664 /*
665  * Early initialization.  Relocation is on but do not reference unbolted pages
666  */
667 static void __init pseries_init(void)
668 {
669 	pr_debug(" -> pseries_init()\n");
670 
671 #ifdef CONFIG_HVC_CONSOLE
672 	if (firmware_has_feature(FW_FEATURE_LPAR))
673 		hvc_vio_init_early();
674 #endif
675 	if (firmware_has_feature(FW_FEATURE_XDABR))
676 		ppc_md.set_dabr = pseries_set_xdabr;
677 	else if (firmware_has_feature(FW_FEATURE_DABR))
678 		ppc_md.set_dabr = pseries_set_dabr;
679 
680 	if (firmware_has_feature(FW_FEATURE_SET_MODE))
681 		ppc_md.set_dawr = pseries_set_dawr;
682 
683 	pSeries_cmo_feature_init();
684 	iommu_init_early_pSeries();
685 
686 	pr_debug(" <- pseries_init()\n");
687 }
688 
689 /**
690  * pseries_power_off - tell firmware about how to power off the system.
691  *
692  * This function calls either the power-off rtas token in normal cases
693  * or the ibm,power-off-ups token (if present & requested) in case of
694  * a power failure. If power-off token is used, power on will only be
695  * possible with power button press. If ibm,power-off-ups token is used
696  * it will allow auto poweron after power is restored.
697  */
698 static void pseries_power_off(void)
699 {
700 	int rc;
701 	int rtas_poweroff_ups_token = rtas_token("ibm,power-off-ups");
702 
703 	if (rtas_flash_term_hook)
704 		rtas_flash_term_hook(SYS_POWER_OFF);
705 
706 	if (rtas_poweron_auto == 0 ||
707 		rtas_poweroff_ups_token == RTAS_UNKNOWN_SERVICE) {
708 		rc = rtas_call(rtas_token("power-off"), 2, 1, NULL, -1, -1);
709 		printk(KERN_INFO "RTAS power-off returned %d\n", rc);
710 	} else {
711 		rc = rtas_call(rtas_poweroff_ups_token, 0, 1, NULL);
712 		printk(KERN_INFO "RTAS ibm,power-off-ups returned %d\n", rc);
713 	}
714 	for (;;);
715 }
716 
717 static int __init pSeries_probe(void)
718 {
719 	const char *dtype = of_get_property(of_root, "device_type", NULL);
720 
721  	if (dtype == NULL)
722  		return 0;
723  	if (strcmp(dtype, "chrp"))
724 		return 0;
725 
726 	/* Cell blades firmware claims to be chrp while it's not. Until this
727 	 * is fixed, we need to avoid those here.
728 	 */
729 	if (of_machine_is_compatible("IBM,CPBW-1.0") ||
730 	    of_machine_is_compatible("IBM,CBEA"))
731 		return 0;
732 
733 	pm_power_off = pseries_power_off;
734 
735 	pr_debug("Machine is%s LPAR !\n",
736 	         (powerpc_firmware_features & FW_FEATURE_LPAR) ? "" : " not");
737 
738 	pseries_init();
739 
740 	return 1;
741 }
742 
743 static int pSeries_pci_probe_mode(struct pci_bus *bus)
744 {
745 	if (firmware_has_feature(FW_FEATURE_LPAR))
746 		return PCI_PROBE_DEVTREE;
747 	return PCI_PROBE_NORMAL;
748 }
749 
750 struct pci_controller_ops pseries_pci_controller_ops = {
751 	.probe_mode		= pSeries_pci_probe_mode,
752 };
753 
754 define_machine(pseries) {
755 	.name			= "pSeries",
756 	.probe			= pSeries_probe,
757 	.setup_arch		= pSeries_setup_arch,
758 	.init_IRQ		= pseries_init_irq,
759 	.show_cpuinfo		= pSeries_show_cpuinfo,
760 	.log_error		= pSeries_log_error,
761 	.pcibios_fixup		= pSeries_final_fixup,
762 	.restart		= rtas_restart,
763 	.halt			= rtas_halt,
764 	.panic			= rtas_os_term,
765 	.get_boot_time		= rtas_get_boot_time,
766 	.get_rtc_time		= rtas_get_rtc_time,
767 	.set_rtc_time		= rtas_set_rtc_time,
768 	.calibrate_decr		= generic_calibrate_decr,
769 	.progress		= rtas_progress,
770 	.system_reset_exception = pSeries_system_reset_exception,
771 	.machine_check_exception = pSeries_machine_check_exception,
772 #ifdef CONFIG_KEXEC_CORE
773 	.machine_kexec          = pSeries_machine_kexec,
774 	.kexec_cpu_down         = pseries_kexec_cpu_down,
775 #endif
776 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
777 	.memory_block_size	= pseries_memory_block_size,
778 #endif
779 };
780