1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *  64-bit pSeries and RS/6000 setup code.
4  *
5  *  Copyright (C) 1995  Linus Torvalds
6  *  Adapted from 'alpha' version by Gary Thomas
7  *  Modified by Cort Dougan (cort@cs.nmt.edu)
8  *  Modified by PPC64 Team, IBM Corp
9  */
10 
11 /*
12  * bootup setup stuff..
13  */
14 
15 #include <linux/cpu.h>
16 #include <linux/errno.h>
17 #include <linux/sched.h>
18 #include <linux/kernel.h>
19 #include <linux/mm.h>
20 #include <linux/stddef.h>
21 #include <linux/unistd.h>
22 #include <linux/user.h>
23 #include <linux/tty.h>
24 #include <linux/major.h>
25 #include <linux/interrupt.h>
26 #include <linux/reboot.h>
27 #include <linux/init.h>
28 #include <linux/ioport.h>
29 #include <linux/console.h>
30 #include <linux/pci.h>
31 #include <linux/utsname.h>
32 #include <linux/adb.h>
33 #include <linux/export.h>
34 #include <linux/delay.h>
35 #include <linux/irq.h>
36 #include <linux/seq_file.h>
37 #include <linux/root_dev.h>
38 #include <linux/of.h>
39 #include <linux/of_pci.h>
40 #include <linux/memblock.h>
41 #include <linux/swiotlb.h>
42 
43 #include <asm/mmu.h>
44 #include <asm/processor.h>
45 #include <asm/io.h>
46 #include <asm/pgtable.h>
47 #include <asm/prom.h>
48 #include <asm/rtas.h>
49 #include <asm/pci-bridge.h>
50 #include <asm/iommu.h>
51 #include <asm/dma.h>
52 #include <asm/machdep.h>
53 #include <asm/irq.h>
54 #include <asm/time.h>
55 #include <asm/nvram.h>
56 #include <asm/pmc.h>
57 #include <asm/xics.h>
58 #include <asm/xive.h>
59 #include <asm/ppc-pci.h>
60 #include <asm/i8259.h>
61 #include <asm/udbg.h>
62 #include <asm/smp.h>
63 #include <asm/firmware.h>
64 #include <asm/eeh.h>
65 #include <asm/reg.h>
66 #include <asm/plpar_wrappers.h>
67 #include <asm/kexec.h>
68 #include <asm/isa-bridge.h>
69 #include <asm/security_features.h>
70 #include <asm/asm-const.h>
71 #include <asm/swiotlb.h>
72 #include <asm/svm.h>
73 
74 #include "pseries.h"
75 #include "../../../../drivers/pci/pci.h"
76 
77 int CMO_PrPSP = -1;
78 int CMO_SecPSP = -1;
79 unsigned long CMO_PageSize = (ASM_CONST(1) << IOMMU_PAGE_SHIFT_4K);
80 EXPORT_SYMBOL(CMO_PageSize);
81 
82 int fwnmi_active;  /* TRUE if an FWNMI handler is present */
83 
84 static void pSeries_show_cpuinfo(struct seq_file *m)
85 {
86 	struct device_node *root;
87 	const char *model = "";
88 
89 	root = of_find_node_by_path("/");
90 	if (root)
91 		model = of_get_property(root, "model", NULL);
92 	seq_printf(m, "machine\t\t: CHRP %s\n", model);
93 	of_node_put(root);
94 	if (radix_enabled())
95 		seq_printf(m, "MMU\t\t: Radix\n");
96 	else
97 		seq_printf(m, "MMU\t\t: Hash\n");
98 }
99 
100 /* Initialize firmware assisted non-maskable interrupts if
101  * the firmware supports this feature.
102  */
103 static void __init fwnmi_init(void)
104 {
105 	unsigned long system_reset_addr, machine_check_addr;
106 	u8 *mce_data_buf;
107 	unsigned int i;
108 	int nr_cpus = num_possible_cpus();
109 #ifdef CONFIG_PPC_BOOK3S_64
110 	struct slb_entry *slb_ptr;
111 	size_t size;
112 #endif
113 
114 	int ibm_nmi_register = rtas_token("ibm,nmi-register");
115 	if (ibm_nmi_register == RTAS_UNKNOWN_SERVICE)
116 		return;
117 
118 	/* If the kernel's not linked at zero we point the firmware at low
119 	 * addresses anyway, and use a trampoline to get to the real code. */
120 	system_reset_addr  = __pa(system_reset_fwnmi) - PHYSICAL_START;
121 	machine_check_addr = __pa(machine_check_fwnmi) - PHYSICAL_START;
122 
123 	if (0 == rtas_call(ibm_nmi_register, 2, 1, NULL, system_reset_addr,
124 				machine_check_addr))
125 		fwnmi_active = 1;
126 
127 	/*
128 	 * Allocate a chunk for per cpu buffer to hold rtas errorlog.
129 	 * It will be used in real mode mce handler, hence it needs to be
130 	 * below RMA.
131 	 */
132 	mce_data_buf = memblock_alloc_try_nid_raw(RTAS_ERROR_LOG_MAX * nr_cpus,
133 					RTAS_ERROR_LOG_MAX, MEMBLOCK_LOW_LIMIT,
134 					ppc64_rma_size, NUMA_NO_NODE);
135 	if (!mce_data_buf)
136 		panic("Failed to allocate %d bytes below %pa for MCE buffer\n",
137 		      RTAS_ERROR_LOG_MAX * nr_cpus, &ppc64_rma_size);
138 
139 	for_each_possible_cpu(i) {
140 		paca_ptrs[i]->mce_data_buf = mce_data_buf +
141 						(RTAS_ERROR_LOG_MAX * i);
142 	}
143 
144 #ifdef CONFIG_PPC_BOOK3S_64
145 	if (!radix_enabled()) {
146 		/* Allocate per cpu area to save old slb contents during MCE */
147 		size = sizeof(struct slb_entry) * mmu_slb_size * nr_cpus;
148 		slb_ptr = memblock_alloc_try_nid_raw(size,
149 				sizeof(struct slb_entry), MEMBLOCK_LOW_LIMIT,
150 				ppc64_rma_size, NUMA_NO_NODE);
151 		if (!slb_ptr)
152 			panic("Failed to allocate %zu bytes below %pa for slb area\n",
153 			      size, &ppc64_rma_size);
154 
155 		for_each_possible_cpu(i)
156 			paca_ptrs[i]->mce_faulty_slbs = slb_ptr + (mmu_slb_size * i);
157 	}
158 #endif
159 }
160 
161 static void pseries_8259_cascade(struct irq_desc *desc)
162 {
163 	struct irq_chip *chip = irq_desc_get_chip(desc);
164 	unsigned int cascade_irq = i8259_irq();
165 
166 	if (cascade_irq)
167 		generic_handle_irq(cascade_irq);
168 
169 	chip->irq_eoi(&desc->irq_data);
170 }
171 
172 static void __init pseries_setup_i8259_cascade(void)
173 {
174 	struct device_node *np, *old, *found = NULL;
175 	unsigned int cascade;
176 	const u32 *addrp;
177 	unsigned long intack = 0;
178 	int naddr;
179 
180 	for_each_node_by_type(np, "interrupt-controller") {
181 		if (of_device_is_compatible(np, "chrp,iic")) {
182 			found = np;
183 			break;
184 		}
185 	}
186 
187 	if (found == NULL) {
188 		printk(KERN_DEBUG "pic: no ISA interrupt controller\n");
189 		return;
190 	}
191 
192 	cascade = irq_of_parse_and_map(found, 0);
193 	if (!cascade) {
194 		printk(KERN_ERR "pic: failed to map cascade interrupt");
195 		return;
196 	}
197 	pr_debug("pic: cascade mapped to irq %d\n", cascade);
198 
199 	for (old = of_node_get(found); old != NULL ; old = np) {
200 		np = of_get_parent(old);
201 		of_node_put(old);
202 		if (np == NULL)
203 			break;
204 		if (!of_node_name_eq(np, "pci"))
205 			continue;
206 		addrp = of_get_property(np, "8259-interrupt-acknowledge", NULL);
207 		if (addrp == NULL)
208 			continue;
209 		naddr = of_n_addr_cells(np);
210 		intack = addrp[naddr-1];
211 		if (naddr > 1)
212 			intack |= ((unsigned long)addrp[naddr-2]) << 32;
213 	}
214 	if (intack)
215 		printk(KERN_DEBUG "pic: PCI 8259 intack at 0x%016lx\n", intack);
216 	i8259_init(found, intack);
217 	of_node_put(found);
218 	irq_set_chained_handler(cascade, pseries_8259_cascade);
219 }
220 
221 static void __init pseries_init_irq(void)
222 {
223 	/* Try using a XIVE if available, otherwise use a XICS */
224 	if (!xive_spapr_init()) {
225 		xics_init();
226 		pseries_setup_i8259_cascade();
227 	}
228 }
229 
230 static void pseries_lpar_enable_pmcs(void)
231 {
232 	unsigned long set, reset;
233 
234 	set = 1UL << 63;
235 	reset = 0;
236 	plpar_hcall_norets(H_PERFMON, set, reset);
237 }
238 
239 static int pci_dn_reconfig_notifier(struct notifier_block *nb, unsigned long action, void *data)
240 {
241 	struct of_reconfig_data *rd = data;
242 	struct device_node *parent, *np = rd->dn;
243 	struct pci_dn *pdn;
244 	int err = NOTIFY_OK;
245 
246 	switch (action) {
247 	case OF_RECONFIG_ATTACH_NODE:
248 		parent = of_get_parent(np);
249 		pdn = parent ? PCI_DN(parent) : NULL;
250 		if (pdn)
251 			pci_add_device_node_info(pdn->phb, np);
252 
253 		of_node_put(parent);
254 		break;
255 	case OF_RECONFIG_DETACH_NODE:
256 		pdn = PCI_DN(np);
257 		if (pdn)
258 			list_del(&pdn->list);
259 		break;
260 	default:
261 		err = NOTIFY_DONE;
262 		break;
263 	}
264 	return err;
265 }
266 
267 static struct notifier_block pci_dn_reconfig_nb = {
268 	.notifier_call = pci_dn_reconfig_notifier,
269 };
270 
271 struct kmem_cache *dtl_cache;
272 
273 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
274 /*
275  * Allocate space for the dispatch trace log for all possible cpus
276  * and register the buffers with the hypervisor.  This is used for
277  * computing time stolen by the hypervisor.
278  */
279 static int alloc_dispatch_logs(void)
280 {
281 	if (!firmware_has_feature(FW_FEATURE_SPLPAR))
282 		return 0;
283 
284 	if (!dtl_cache)
285 		return 0;
286 
287 	alloc_dtl_buffers(0);
288 
289 	/* Register the DTL for the current (boot) cpu */
290 	register_dtl_buffer(smp_processor_id());
291 
292 	return 0;
293 }
294 #else /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
295 static inline int alloc_dispatch_logs(void)
296 {
297 	return 0;
298 }
299 #endif /* CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
300 
301 static int alloc_dispatch_log_kmem_cache(void)
302 {
303 	void (*ctor)(void *) = get_dtl_cache_ctor();
304 
305 	dtl_cache = kmem_cache_create("dtl", DISPATCH_LOG_BYTES,
306 						DISPATCH_LOG_BYTES, 0, ctor);
307 	if (!dtl_cache) {
308 		pr_warn("Failed to create dispatch trace log buffer cache\n");
309 		pr_warn("Stolen time statistics will be unreliable\n");
310 		return 0;
311 	}
312 
313 	return alloc_dispatch_logs();
314 }
315 machine_early_initcall(pseries, alloc_dispatch_log_kmem_cache);
316 
317 static void pseries_lpar_idle(void)
318 {
319 	/*
320 	 * Default handler to go into low thread priority and possibly
321 	 * low power mode by ceding processor to hypervisor
322 	 */
323 
324 	if (!prep_irq_for_idle())
325 		return;
326 
327 	/* Indicate to hypervisor that we are idle. */
328 	get_lppaca()->idle = 1;
329 
330 	/*
331 	 * Yield the processor to the hypervisor.  We return if
332 	 * an external interrupt occurs (which are driven prior
333 	 * to returning here) or if a prod occurs from another
334 	 * processor. When returning here, external interrupts
335 	 * are enabled.
336 	 */
337 	cede_processor();
338 
339 	get_lppaca()->idle = 0;
340 }
341 
342 /*
343  * Enable relocation on during exceptions. This has partition wide scope and
344  * may take a while to complete, if it takes longer than one second we will
345  * just give up rather than wasting any more time on this - if that turns out
346  * to ever be a problem in practice we can move this into a kernel thread to
347  * finish off the process later in boot.
348  */
349 void pseries_enable_reloc_on_exc(void)
350 {
351 	long rc;
352 	unsigned int delay, total_delay = 0;
353 
354 	while (1) {
355 		rc = enable_reloc_on_exceptions();
356 		if (!H_IS_LONG_BUSY(rc)) {
357 			if (rc == H_P2) {
358 				pr_info("Relocation on exceptions not"
359 					" supported\n");
360 			} else if (rc != H_SUCCESS) {
361 				pr_warn("Unable to enable relocation"
362 					" on exceptions: %ld\n", rc);
363 			}
364 			break;
365 		}
366 
367 		delay = get_longbusy_msecs(rc);
368 		total_delay += delay;
369 		if (total_delay > 1000) {
370 			pr_warn("Warning: Giving up waiting to enable "
371 				"relocation on exceptions (%u msec)!\n",
372 				total_delay);
373 			return;
374 		}
375 
376 		mdelay(delay);
377 	}
378 }
379 EXPORT_SYMBOL(pseries_enable_reloc_on_exc);
380 
381 void pseries_disable_reloc_on_exc(void)
382 {
383 	long rc;
384 
385 	while (1) {
386 		rc = disable_reloc_on_exceptions();
387 		if (!H_IS_LONG_BUSY(rc))
388 			break;
389 		mdelay(get_longbusy_msecs(rc));
390 	}
391 	if (rc != H_SUCCESS)
392 		pr_warn("Warning: Failed to disable relocation on exceptions: %ld\n",
393 			rc);
394 }
395 EXPORT_SYMBOL(pseries_disable_reloc_on_exc);
396 
397 #ifdef CONFIG_KEXEC_CORE
398 static void pSeries_machine_kexec(struct kimage *image)
399 {
400 	if (firmware_has_feature(FW_FEATURE_SET_MODE))
401 		pseries_disable_reloc_on_exc();
402 
403 	default_machine_kexec(image);
404 }
405 #endif
406 
407 #ifdef __LITTLE_ENDIAN__
408 void pseries_big_endian_exceptions(void)
409 {
410 	long rc;
411 
412 	while (1) {
413 		rc = enable_big_endian_exceptions();
414 		if (!H_IS_LONG_BUSY(rc))
415 			break;
416 		mdelay(get_longbusy_msecs(rc));
417 	}
418 
419 	/*
420 	 * At this point it is unlikely panic() will get anything
421 	 * out to the user, since this is called very late in kexec
422 	 * but at least this will stop us from continuing on further
423 	 * and creating an even more difficult to debug situation.
424 	 *
425 	 * There is a known problem when kdump'ing, if cpus are offline
426 	 * the above call will fail. Rather than panicking again, keep
427 	 * going and hope the kdump kernel is also little endian, which
428 	 * it usually is.
429 	 */
430 	if (rc && !kdump_in_progress())
431 		panic("Could not enable big endian exceptions");
432 }
433 
434 void pseries_little_endian_exceptions(void)
435 {
436 	long rc;
437 
438 	while (1) {
439 		rc = enable_little_endian_exceptions();
440 		if (!H_IS_LONG_BUSY(rc))
441 			break;
442 		mdelay(get_longbusy_msecs(rc));
443 	}
444 	if (rc) {
445 		ppc_md.progress("H_SET_MODE LE exception fail", 0);
446 		panic("Could not enable little endian exceptions");
447 	}
448 }
449 #endif
450 
451 static void __init find_and_init_phbs(void)
452 {
453 	struct device_node *node;
454 	struct pci_controller *phb;
455 	struct device_node *root = of_find_node_by_path("/");
456 
457 	for_each_child_of_node(root, node) {
458 		if (!of_node_is_type(node, "pci") &&
459 		    !of_node_is_type(node, "pciex"))
460 			continue;
461 
462 		phb = pcibios_alloc_controller(node);
463 		if (!phb)
464 			continue;
465 		rtas_setup_phb(phb);
466 		pci_process_bridge_OF_ranges(phb, node, 0);
467 		isa_bridge_find_early(phb);
468 		phb->controller_ops = pseries_pci_controller_ops;
469 	}
470 
471 	of_node_put(root);
472 
473 	/*
474 	 * PCI_PROBE_ONLY and PCI_REASSIGN_ALL_BUS can be set via properties
475 	 * in chosen.
476 	 */
477 	of_pci_check_probe_only();
478 }
479 
480 static void init_cpu_char_feature_flags(struct h_cpu_char_result *result)
481 {
482 	/*
483 	 * The features below are disabled by default, so we instead look to see
484 	 * if firmware has *enabled* them, and set them if so.
485 	 */
486 	if (result->character & H_CPU_CHAR_SPEC_BAR_ORI31)
487 		security_ftr_set(SEC_FTR_SPEC_BAR_ORI31);
488 
489 	if (result->character & H_CPU_CHAR_BCCTRL_SERIALISED)
490 		security_ftr_set(SEC_FTR_BCCTRL_SERIALISED);
491 
492 	if (result->character & H_CPU_CHAR_L1D_FLUSH_ORI30)
493 		security_ftr_set(SEC_FTR_L1D_FLUSH_ORI30);
494 
495 	if (result->character & H_CPU_CHAR_L1D_FLUSH_TRIG2)
496 		security_ftr_set(SEC_FTR_L1D_FLUSH_TRIG2);
497 
498 	if (result->character & H_CPU_CHAR_L1D_THREAD_PRIV)
499 		security_ftr_set(SEC_FTR_L1D_THREAD_PRIV);
500 
501 	if (result->character & H_CPU_CHAR_COUNT_CACHE_DISABLED)
502 		security_ftr_set(SEC_FTR_COUNT_CACHE_DISABLED);
503 
504 	if (result->character & H_CPU_CHAR_BCCTR_FLUSH_ASSIST)
505 		security_ftr_set(SEC_FTR_BCCTR_FLUSH_ASSIST);
506 
507 	if (result->behaviour & H_CPU_BEHAV_FLUSH_COUNT_CACHE)
508 		security_ftr_set(SEC_FTR_FLUSH_COUNT_CACHE);
509 
510 	/*
511 	 * The features below are enabled by default, so we instead look to see
512 	 * if firmware has *disabled* them, and clear them if so.
513 	 */
514 	if (!(result->behaviour & H_CPU_BEHAV_FAVOUR_SECURITY))
515 		security_ftr_clear(SEC_FTR_FAVOUR_SECURITY);
516 
517 	if (!(result->behaviour & H_CPU_BEHAV_L1D_FLUSH_PR))
518 		security_ftr_clear(SEC_FTR_L1D_FLUSH_PR);
519 
520 	if (!(result->behaviour & H_CPU_BEHAV_BNDS_CHK_SPEC_BAR))
521 		security_ftr_clear(SEC_FTR_BNDS_CHK_SPEC_BAR);
522 }
523 
524 void pseries_setup_rfi_flush(void)
525 {
526 	struct h_cpu_char_result result;
527 	enum l1d_flush_type types;
528 	bool enable;
529 	long rc;
530 
531 	/*
532 	 * Set features to the defaults assumed by init_cpu_char_feature_flags()
533 	 * so it can set/clear again any features that might have changed after
534 	 * migration, and in case the hypercall fails and it is not even called.
535 	 */
536 	powerpc_security_features = SEC_FTR_DEFAULT;
537 
538 	rc = plpar_get_cpu_characteristics(&result);
539 	if (rc == H_SUCCESS)
540 		init_cpu_char_feature_flags(&result);
541 
542 	/*
543 	 * We're the guest so this doesn't apply to us, clear it to simplify
544 	 * handling of it elsewhere.
545 	 */
546 	security_ftr_clear(SEC_FTR_L1D_FLUSH_HV);
547 
548 	types = L1D_FLUSH_FALLBACK;
549 
550 	if (security_ftr_enabled(SEC_FTR_L1D_FLUSH_TRIG2))
551 		types |= L1D_FLUSH_MTTRIG;
552 
553 	if (security_ftr_enabled(SEC_FTR_L1D_FLUSH_ORI30))
554 		types |= L1D_FLUSH_ORI;
555 
556 	enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) && \
557 		 security_ftr_enabled(SEC_FTR_L1D_FLUSH_PR);
558 
559 	setup_rfi_flush(types, enable);
560 	setup_count_cache_flush();
561 }
562 
563 #ifdef CONFIG_PCI_IOV
564 enum rtas_iov_fw_value_map {
565 	NUM_RES_PROPERTY  = 0, /* Number of Resources */
566 	LOW_INT           = 1, /* Lowest 32 bits of Address */
567 	START_OF_ENTRIES  = 2, /* Always start of entry */
568 	APERTURE_PROPERTY = 2, /* Start of entry+ to  Aperture Size */
569 	WDW_SIZE_PROPERTY = 4, /* Start of entry+ to Window Size */
570 	NEXT_ENTRY        = 7  /* Go to next entry on array */
571 };
572 
573 enum get_iov_fw_value_index {
574 	BAR_ADDRS     = 1,    /*  Get Bar Address */
575 	APERTURE_SIZE = 2,    /*  Get Aperture Size */
576 	WDW_SIZE      = 3     /*  Get Window Size */
577 };
578 
579 resource_size_t pseries_get_iov_fw_value(struct pci_dev *dev, int resno,
580 					 enum get_iov_fw_value_index value)
581 {
582 	const int *indexes;
583 	struct device_node *dn = pci_device_to_OF_node(dev);
584 	int i, num_res, ret = 0;
585 
586 	indexes = of_get_property(dn, "ibm,open-sriov-vf-bar-info", NULL);
587 	if (!indexes)
588 		return  0;
589 
590 	/*
591 	 * First element in the array is the number of Bars
592 	 * returned.  Search through the list to find the matching
593 	 * bar
594 	 */
595 	num_res = of_read_number(&indexes[NUM_RES_PROPERTY], 1);
596 	if (resno >= num_res)
597 		return 0; /* or an errror */
598 
599 	i = START_OF_ENTRIES + NEXT_ENTRY * resno;
600 	switch (value) {
601 	case BAR_ADDRS:
602 		ret = of_read_number(&indexes[i], 2);
603 		break;
604 	case APERTURE_SIZE:
605 		ret = of_read_number(&indexes[i + APERTURE_PROPERTY], 2);
606 		break;
607 	case WDW_SIZE:
608 		ret = of_read_number(&indexes[i + WDW_SIZE_PROPERTY], 2);
609 		break;
610 	}
611 
612 	return ret;
613 }
614 
615 void of_pci_set_vf_bar_size(struct pci_dev *dev, const int *indexes)
616 {
617 	struct resource *res;
618 	resource_size_t base, size;
619 	int i, r, num_res;
620 
621 	num_res = of_read_number(&indexes[NUM_RES_PROPERTY], 1);
622 	num_res = min_t(int, num_res, PCI_SRIOV_NUM_BARS);
623 	for (i = START_OF_ENTRIES, r = 0; r < num_res && r < PCI_SRIOV_NUM_BARS;
624 	     i += NEXT_ENTRY, r++) {
625 		res = &dev->resource[r + PCI_IOV_RESOURCES];
626 		base = of_read_number(&indexes[i], 2);
627 		size = of_read_number(&indexes[i + APERTURE_PROPERTY], 2);
628 		res->flags = pci_parse_of_flags(of_read_number
629 						(&indexes[i + LOW_INT], 1), 0);
630 		res->flags |= (IORESOURCE_MEM_64 | IORESOURCE_PCI_FIXED);
631 		res->name = pci_name(dev);
632 		res->start = base;
633 		res->end = base + size - 1;
634 	}
635 }
636 
637 void of_pci_parse_iov_addrs(struct pci_dev *dev, const int *indexes)
638 {
639 	struct resource *res, *root, *conflict;
640 	resource_size_t base, size;
641 	int i, r, num_res;
642 
643 	/*
644 	 * First element in the array is the number of Bars
645 	 * returned.  Search through the list to find the matching
646 	 * bars assign them from firmware into resources structure.
647 	 */
648 	num_res = of_read_number(&indexes[NUM_RES_PROPERTY], 1);
649 	for (i = START_OF_ENTRIES, r = 0; r < num_res && r < PCI_SRIOV_NUM_BARS;
650 	     i += NEXT_ENTRY, r++) {
651 		res = &dev->resource[r + PCI_IOV_RESOURCES];
652 		base = of_read_number(&indexes[i], 2);
653 		size = of_read_number(&indexes[i + WDW_SIZE_PROPERTY], 2);
654 		res->name = pci_name(dev);
655 		res->start = base;
656 		res->end = base + size - 1;
657 		root = &iomem_resource;
658 		dev_dbg(&dev->dev,
659 			"pSeries IOV BAR %d: trying firmware assignment %pR\n",
660 			 r + PCI_IOV_RESOURCES, res);
661 		conflict = request_resource_conflict(root, res);
662 		if (conflict) {
663 			dev_info(&dev->dev,
664 				 "BAR %d: %pR conflicts with %s %pR\n",
665 				 r + PCI_IOV_RESOURCES, res,
666 				 conflict->name, conflict);
667 			res->flags |= IORESOURCE_UNSET;
668 		}
669 	}
670 }
671 
672 static void pseries_disable_sriov_resources(struct pci_dev *pdev)
673 {
674 	int i;
675 
676 	pci_warn(pdev, "No hypervisor support for SR-IOV on this device, IOV BARs disabled.\n");
677 	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++)
678 		pdev->resource[i + PCI_IOV_RESOURCES].flags = 0;
679 }
680 
681 static void pseries_pci_fixup_resources(struct pci_dev *pdev)
682 {
683 	const int *indexes;
684 	struct device_node *dn = pci_device_to_OF_node(pdev);
685 
686 	/*Firmware must support open sriov otherwise dont configure*/
687 	indexes = of_get_property(dn, "ibm,open-sriov-vf-bar-info", NULL);
688 	if (indexes)
689 		of_pci_set_vf_bar_size(pdev, indexes);
690 	else
691 		pseries_disable_sriov_resources(pdev);
692 }
693 
694 static void pseries_pci_fixup_iov_resources(struct pci_dev *pdev)
695 {
696 	const int *indexes;
697 	struct device_node *dn = pci_device_to_OF_node(pdev);
698 
699 	if (!pdev->is_physfn || pci_dev_is_added(pdev))
700 		return;
701 	/*Firmware must support open sriov otherwise dont configure*/
702 	indexes = of_get_property(dn, "ibm,open-sriov-vf-bar-info", NULL);
703 	if (indexes)
704 		of_pci_parse_iov_addrs(pdev, indexes);
705 	else
706 		pseries_disable_sriov_resources(pdev);
707 }
708 
709 static resource_size_t pseries_pci_iov_resource_alignment(struct pci_dev *pdev,
710 							  int resno)
711 {
712 	const __be32 *reg;
713 	struct device_node *dn = pci_device_to_OF_node(pdev);
714 
715 	/*Firmware must support open sriov otherwise report regular alignment*/
716 	reg = of_get_property(dn, "ibm,is-open-sriov-pf", NULL);
717 	if (!reg)
718 		return pci_iov_resource_size(pdev, resno);
719 
720 	if (!pdev->is_physfn)
721 		return 0;
722 	return pseries_get_iov_fw_value(pdev,
723 					resno - PCI_IOV_RESOURCES,
724 					APERTURE_SIZE);
725 }
726 #endif
727 
728 static void __init pSeries_setup_arch(void)
729 {
730 	set_arch_panic_timeout(10, ARCH_PANIC_TIMEOUT);
731 
732 	/* Discover PIC type and setup ppc_md accordingly */
733 	smp_init_pseries();
734 
735 
736 	/* openpic global configuration register (64-bit format). */
737 	/* openpic Interrupt Source Unit pointer (64-bit format). */
738 	/* python0 facility area (mmio) (64-bit format) REAL address. */
739 
740 	/* init to some ~sane value until calibrate_delay() runs */
741 	loops_per_jiffy = 50000000;
742 
743 	fwnmi_init();
744 
745 	pseries_setup_rfi_flush();
746 	setup_stf_barrier();
747 
748 	/* By default, only probe PCI (can be overridden by rtas_pci) */
749 	pci_add_flags(PCI_PROBE_ONLY);
750 
751 	/* Find and initialize PCI host bridges */
752 	init_pci_config_tokens();
753 	find_and_init_phbs();
754 	of_reconfig_notifier_register(&pci_dn_reconfig_nb);
755 
756 	pSeries_nvram_init();
757 
758 	if (firmware_has_feature(FW_FEATURE_LPAR)) {
759 		vpa_init(boot_cpuid);
760 		ppc_md.power_save = pseries_lpar_idle;
761 		ppc_md.enable_pmcs = pseries_lpar_enable_pmcs;
762 #ifdef CONFIG_PCI_IOV
763 		ppc_md.pcibios_fixup_resources =
764 			pseries_pci_fixup_resources;
765 		ppc_md.pcibios_fixup_sriov =
766 			pseries_pci_fixup_iov_resources;
767 		ppc_md.pcibios_iov_resource_alignment =
768 			pseries_pci_iov_resource_alignment;
769 #endif
770 	} else {
771 		/* No special idle routine */
772 		ppc_md.enable_pmcs = power4_enable_pmcs;
773 	}
774 
775 	ppc_md.pcibios_root_bridge_prepare = pseries_root_bridge_prepare;
776 
777 	if (swiotlb_force == SWIOTLB_FORCE)
778 		ppc_swiotlb_enable = 1;
779 }
780 
781 static void pseries_panic(char *str)
782 {
783 	panic_flush_kmsg_end();
784 	rtas_os_term(str);
785 }
786 
787 static int __init pSeries_init_panel(void)
788 {
789 	/* Manually leave the kernel version on the panel. */
790 #ifdef __BIG_ENDIAN__
791 	ppc_md.progress("Linux ppc64\n", 0);
792 #else
793 	ppc_md.progress("Linux ppc64le\n", 0);
794 #endif
795 	ppc_md.progress(init_utsname()->version, 0);
796 
797 	return 0;
798 }
799 machine_arch_initcall(pseries, pSeries_init_panel);
800 
801 static int pseries_set_dabr(unsigned long dabr, unsigned long dabrx)
802 {
803 	return plpar_hcall_norets(H_SET_DABR, dabr);
804 }
805 
806 static int pseries_set_xdabr(unsigned long dabr, unsigned long dabrx)
807 {
808 	/* Have to set at least one bit in the DABRX according to PAPR */
809 	if (dabrx == 0 && dabr == 0)
810 		dabrx = DABRX_USER;
811 	/* PAPR says we can only set kernel and user bits */
812 	dabrx &= DABRX_KERNEL | DABRX_USER;
813 
814 	return plpar_hcall_norets(H_SET_XDABR, dabr, dabrx);
815 }
816 
817 static int pseries_set_dawr(unsigned long dawr, unsigned long dawrx)
818 {
819 	/* PAPR says we can't set HYP */
820 	dawrx &= ~DAWRX_HYP;
821 
822 	return  plpar_set_watchpoint0(dawr, dawrx);
823 }
824 
825 #define CMO_CHARACTERISTICS_TOKEN 44
826 #define CMO_MAXLENGTH 1026
827 
828 void pSeries_coalesce_init(void)
829 {
830 	struct hvcall_mpp_x_data mpp_x_data;
831 
832 	if (firmware_has_feature(FW_FEATURE_CMO) && !h_get_mpp_x(&mpp_x_data))
833 		powerpc_firmware_features |= FW_FEATURE_XCMO;
834 	else
835 		powerpc_firmware_features &= ~FW_FEATURE_XCMO;
836 }
837 
838 /**
839  * fw_cmo_feature_init - FW_FEATURE_CMO is not stored in ibm,hypertas-functions,
840  * handle that here. (Stolen from parse_system_parameter_string)
841  */
842 static void pSeries_cmo_feature_init(void)
843 {
844 	char *ptr, *key, *value, *end;
845 	int call_status;
846 	int page_order = IOMMU_PAGE_SHIFT_4K;
847 
848 	pr_debug(" -> fw_cmo_feature_init()\n");
849 	spin_lock(&rtas_data_buf_lock);
850 	memset(rtas_data_buf, 0, RTAS_DATA_BUF_SIZE);
851 	call_status = rtas_call(rtas_token("ibm,get-system-parameter"), 3, 1,
852 				NULL,
853 				CMO_CHARACTERISTICS_TOKEN,
854 				__pa(rtas_data_buf),
855 				RTAS_DATA_BUF_SIZE);
856 
857 	if (call_status != 0) {
858 		spin_unlock(&rtas_data_buf_lock);
859 		pr_debug("CMO not available\n");
860 		pr_debug(" <- fw_cmo_feature_init()\n");
861 		return;
862 	}
863 
864 	end = rtas_data_buf + CMO_MAXLENGTH - 2;
865 	ptr = rtas_data_buf + 2;	/* step over strlen value */
866 	key = value = ptr;
867 
868 	while (*ptr && (ptr <= end)) {
869 		/* Separate the key and value by replacing '=' with '\0' and
870 		 * point the value at the string after the '='
871 		 */
872 		if (ptr[0] == '=') {
873 			ptr[0] = '\0';
874 			value = ptr + 1;
875 		} else if (ptr[0] == '\0' || ptr[0] == ',') {
876 			/* Terminate the string containing the key/value pair */
877 			ptr[0] = '\0';
878 
879 			if (key == value) {
880 				pr_debug("Malformed key/value pair\n");
881 				/* Never found a '=', end processing */
882 				break;
883 			}
884 
885 			if (0 == strcmp(key, "CMOPageSize"))
886 				page_order = simple_strtol(value, NULL, 10);
887 			else if (0 == strcmp(key, "PrPSP"))
888 				CMO_PrPSP = simple_strtol(value, NULL, 10);
889 			else if (0 == strcmp(key, "SecPSP"))
890 				CMO_SecPSP = simple_strtol(value, NULL, 10);
891 			value = key = ptr + 1;
892 		}
893 		ptr++;
894 	}
895 
896 	/* Page size is returned as the power of 2 of the page size,
897 	 * convert to the page size in bytes before returning
898 	 */
899 	CMO_PageSize = 1 << page_order;
900 	pr_debug("CMO_PageSize = %lu\n", CMO_PageSize);
901 
902 	if (CMO_PrPSP != -1 || CMO_SecPSP != -1) {
903 		pr_info("CMO enabled\n");
904 		pr_debug("CMO enabled, PrPSP=%d, SecPSP=%d\n", CMO_PrPSP,
905 		         CMO_SecPSP);
906 		powerpc_firmware_features |= FW_FEATURE_CMO;
907 		pSeries_coalesce_init();
908 	} else
909 		pr_debug("CMO not enabled, PrPSP=%d, SecPSP=%d\n", CMO_PrPSP,
910 		         CMO_SecPSP);
911 	spin_unlock(&rtas_data_buf_lock);
912 	pr_debug(" <- fw_cmo_feature_init()\n");
913 }
914 
915 /*
916  * Early initialization.  Relocation is on but do not reference unbolted pages
917  */
918 static void __init pseries_init(void)
919 {
920 	pr_debug(" -> pseries_init()\n");
921 
922 #ifdef CONFIG_HVC_CONSOLE
923 	if (firmware_has_feature(FW_FEATURE_LPAR))
924 		hvc_vio_init_early();
925 #endif
926 	if (firmware_has_feature(FW_FEATURE_XDABR))
927 		ppc_md.set_dabr = pseries_set_xdabr;
928 	else if (firmware_has_feature(FW_FEATURE_DABR))
929 		ppc_md.set_dabr = pseries_set_dabr;
930 
931 	if (firmware_has_feature(FW_FEATURE_SET_MODE))
932 		ppc_md.set_dawr = pseries_set_dawr;
933 
934 	pSeries_cmo_feature_init();
935 	iommu_init_early_pSeries();
936 
937 	pr_debug(" <- pseries_init()\n");
938 }
939 
940 /**
941  * pseries_power_off - tell firmware about how to power off the system.
942  *
943  * This function calls either the power-off rtas token in normal cases
944  * or the ibm,power-off-ups token (if present & requested) in case of
945  * a power failure. If power-off token is used, power on will only be
946  * possible with power button press. If ibm,power-off-ups token is used
947  * it will allow auto poweron after power is restored.
948  */
949 static void pseries_power_off(void)
950 {
951 	int rc;
952 	int rtas_poweroff_ups_token = rtas_token("ibm,power-off-ups");
953 
954 	if (rtas_flash_term_hook)
955 		rtas_flash_term_hook(SYS_POWER_OFF);
956 
957 	if (rtas_poweron_auto == 0 ||
958 		rtas_poweroff_ups_token == RTAS_UNKNOWN_SERVICE) {
959 		rc = rtas_call(rtas_token("power-off"), 2, 1, NULL, -1, -1);
960 		printk(KERN_INFO "RTAS power-off returned %d\n", rc);
961 	} else {
962 		rc = rtas_call(rtas_poweroff_ups_token, 0, 1, NULL);
963 		printk(KERN_INFO "RTAS ibm,power-off-ups returned %d\n", rc);
964 	}
965 	for (;;);
966 }
967 
968 static int __init pSeries_probe(void)
969 {
970 	if (!of_node_is_type(of_root, "chrp"))
971 		return 0;
972 
973 	/* Cell blades firmware claims to be chrp while it's not. Until this
974 	 * is fixed, we need to avoid those here.
975 	 */
976 	if (of_machine_is_compatible("IBM,CPBW-1.0") ||
977 	    of_machine_is_compatible("IBM,CBEA"))
978 		return 0;
979 
980 	pm_power_off = pseries_power_off;
981 
982 	pr_debug("Machine is%s LPAR !\n",
983 	         (powerpc_firmware_features & FW_FEATURE_LPAR) ? "" : " not");
984 
985 	pseries_init();
986 
987 	return 1;
988 }
989 
990 static int pSeries_pci_probe_mode(struct pci_bus *bus)
991 {
992 	if (firmware_has_feature(FW_FEATURE_LPAR))
993 		return PCI_PROBE_DEVTREE;
994 	return PCI_PROBE_NORMAL;
995 }
996 
997 struct pci_controller_ops pseries_pci_controller_ops = {
998 	.probe_mode		= pSeries_pci_probe_mode,
999 };
1000 
1001 define_machine(pseries) {
1002 	.name			= "pSeries",
1003 	.probe			= pSeries_probe,
1004 	.setup_arch		= pSeries_setup_arch,
1005 	.init_IRQ		= pseries_init_irq,
1006 	.show_cpuinfo		= pSeries_show_cpuinfo,
1007 	.log_error		= pSeries_log_error,
1008 	.pcibios_fixup		= pSeries_final_fixup,
1009 	.restart		= rtas_restart,
1010 	.halt			= rtas_halt,
1011 	.panic			= pseries_panic,
1012 	.get_boot_time		= rtas_get_boot_time,
1013 	.get_rtc_time		= rtas_get_rtc_time,
1014 	.set_rtc_time		= rtas_set_rtc_time,
1015 	.calibrate_decr		= generic_calibrate_decr,
1016 	.progress		= rtas_progress,
1017 	.system_reset_exception = pSeries_system_reset_exception,
1018 	.machine_check_early	= pseries_machine_check_realmode,
1019 	.machine_check_exception = pSeries_machine_check_exception,
1020 #ifdef CONFIG_KEXEC_CORE
1021 	.machine_kexec          = pSeries_machine_kexec,
1022 	.kexec_cpu_down         = pseries_kexec_cpu_down,
1023 #endif
1024 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
1025 	.memory_block_size	= pseries_memory_block_size,
1026 #endif
1027 };
1028