xref: /linux/arch/powerpc/platforms/pseries/setup.c (revision b7019ac550eb3916f34d79db583e9b7ea2524afa)
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 
42 #include <asm/mmu.h>
43 #include <asm/processor.h>
44 #include <asm/io.h>
45 #include <asm/pgtable.h>
46 #include <asm/prom.h>
47 #include <asm/rtas.h>
48 #include <asm/pci-bridge.h>
49 #include <asm/iommu.h>
50 #include <asm/dma.h>
51 #include <asm/machdep.h>
52 #include <asm/irq.h>
53 #include <asm/time.h>
54 #include <asm/nvram.h>
55 #include <asm/pmc.h>
56 #include <asm/xics.h>
57 #include <asm/xive.h>
58 #include <asm/ppc-pci.h>
59 #include <asm/i8259.h>
60 #include <asm/udbg.h>
61 #include <asm/smp.h>
62 #include <asm/firmware.h>
63 #include <asm/eeh.h>
64 #include <asm/reg.h>
65 #include <asm/plpar_wrappers.h>
66 #include <asm/kexec.h>
67 #include <asm/isa-bridge.h>
68 #include <asm/security_features.h>
69 #include <asm/asm-const.h>
70 
71 #include "pseries.h"
72 #include "../../../../drivers/pci/pci.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 	u8 *mce_data_buf;
104 	unsigned int i;
105 	int nr_cpus = num_possible_cpus();
106 #ifdef CONFIG_PPC_BOOK3S_64
107 	struct slb_entry *slb_ptr;
108 	size_t size;
109 #endif
110 
111 	int ibm_nmi_register = rtas_token("ibm,nmi-register");
112 	if (ibm_nmi_register == RTAS_UNKNOWN_SERVICE)
113 		return;
114 
115 	/* If the kernel's not linked at zero we point the firmware at low
116 	 * addresses anyway, and use a trampoline to get to the real code. */
117 	system_reset_addr  = __pa(system_reset_fwnmi) - PHYSICAL_START;
118 	machine_check_addr = __pa(machine_check_fwnmi) - PHYSICAL_START;
119 
120 	if (0 == rtas_call(ibm_nmi_register, 2, 1, NULL, system_reset_addr,
121 				machine_check_addr))
122 		fwnmi_active = 1;
123 
124 	/*
125 	 * Allocate a chunk for per cpu buffer to hold rtas errorlog.
126 	 * It will be used in real mode mce handler, hence it needs to be
127 	 * below RMA.
128 	 */
129 	mce_data_buf = memblock_alloc_try_nid_raw(RTAS_ERROR_LOG_MAX * nr_cpus,
130 					RTAS_ERROR_LOG_MAX, MEMBLOCK_LOW_LIMIT,
131 					ppc64_rma_size, NUMA_NO_NODE);
132 	if (!mce_data_buf)
133 		panic("Failed to allocate %d bytes below %pa for MCE buffer\n",
134 		      RTAS_ERROR_LOG_MAX * nr_cpus, &ppc64_rma_size);
135 
136 	for_each_possible_cpu(i) {
137 		paca_ptrs[i]->mce_data_buf = mce_data_buf +
138 						(RTAS_ERROR_LOG_MAX * i);
139 	}
140 
141 #ifdef CONFIG_PPC_BOOK3S_64
142 	/* Allocate per cpu slb area to save old slb contents during MCE */
143 	size = sizeof(struct slb_entry) * mmu_slb_size * nr_cpus;
144 	slb_ptr = memblock_alloc_try_nid_raw(size, sizeof(struct slb_entry),
145 					MEMBLOCK_LOW_LIMIT, ppc64_rma_size,
146 					NUMA_NO_NODE);
147 	if (!slb_ptr)
148 		panic("Failed to allocate %zu bytes below %pa for slb area\n",
149 		      size, &ppc64_rma_size);
150 
151 	for_each_possible_cpu(i)
152 		paca_ptrs[i]->mce_faulty_slbs = slb_ptr + (mmu_slb_size * i);
153 #endif
154 }
155 
156 static void pseries_8259_cascade(struct irq_desc *desc)
157 {
158 	struct irq_chip *chip = irq_desc_get_chip(desc);
159 	unsigned int cascade_irq = i8259_irq();
160 
161 	if (cascade_irq)
162 		generic_handle_irq(cascade_irq);
163 
164 	chip->irq_eoi(&desc->irq_data);
165 }
166 
167 static void __init pseries_setup_i8259_cascade(void)
168 {
169 	struct device_node *np, *old, *found = NULL;
170 	unsigned int cascade;
171 	const u32 *addrp;
172 	unsigned long intack = 0;
173 	int naddr;
174 
175 	for_each_node_by_type(np, "interrupt-controller") {
176 		if (of_device_is_compatible(np, "chrp,iic")) {
177 			found = np;
178 			break;
179 		}
180 	}
181 
182 	if (found == NULL) {
183 		printk(KERN_DEBUG "pic: no ISA interrupt controller\n");
184 		return;
185 	}
186 
187 	cascade = irq_of_parse_and_map(found, 0);
188 	if (!cascade) {
189 		printk(KERN_ERR "pic: failed to map cascade interrupt");
190 		return;
191 	}
192 	pr_debug("pic: cascade mapped to irq %d\n", cascade);
193 
194 	for (old = of_node_get(found); old != NULL ; old = np) {
195 		np = of_get_parent(old);
196 		of_node_put(old);
197 		if (np == NULL)
198 			break;
199 		if (!of_node_name_eq(np, "pci"))
200 			continue;
201 		addrp = of_get_property(np, "8259-interrupt-acknowledge", NULL);
202 		if (addrp == NULL)
203 			continue;
204 		naddr = of_n_addr_cells(np);
205 		intack = addrp[naddr-1];
206 		if (naddr > 1)
207 			intack |= ((unsigned long)addrp[naddr-2]) << 32;
208 	}
209 	if (intack)
210 		printk(KERN_DEBUG "pic: PCI 8259 intack at 0x%016lx\n", intack);
211 	i8259_init(found, intack);
212 	of_node_put(found);
213 	irq_set_chained_handler(cascade, pseries_8259_cascade);
214 }
215 
216 static void __init pseries_init_irq(void)
217 {
218 	/* Try using a XIVE if available, otherwise use a XICS */
219 	if (!xive_spapr_init()) {
220 		xics_init();
221 		pseries_setup_i8259_cascade();
222 	}
223 }
224 
225 static void pseries_lpar_enable_pmcs(void)
226 {
227 	unsigned long set, reset;
228 
229 	set = 1UL << 63;
230 	reset = 0;
231 	plpar_hcall_norets(H_PERFMON, set, reset);
232 }
233 
234 static int pci_dn_reconfig_notifier(struct notifier_block *nb, unsigned long action, void *data)
235 {
236 	struct of_reconfig_data *rd = data;
237 	struct device_node *parent, *np = rd->dn;
238 	struct pci_dn *pdn;
239 	int err = NOTIFY_OK;
240 
241 	switch (action) {
242 	case OF_RECONFIG_ATTACH_NODE:
243 		parent = of_get_parent(np);
244 		pdn = parent ? PCI_DN(parent) : NULL;
245 		if (pdn)
246 			pci_add_device_node_info(pdn->phb, np);
247 
248 		of_node_put(parent);
249 		break;
250 	case OF_RECONFIG_DETACH_NODE:
251 		pdn = PCI_DN(np);
252 		if (pdn)
253 			list_del(&pdn->list);
254 		break;
255 	default:
256 		err = NOTIFY_DONE;
257 		break;
258 	}
259 	return err;
260 }
261 
262 static struct notifier_block pci_dn_reconfig_nb = {
263 	.notifier_call = pci_dn_reconfig_notifier,
264 };
265 
266 struct kmem_cache *dtl_cache;
267 
268 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
269 /*
270  * Allocate space for the dispatch trace log for all possible cpus
271  * and register the buffers with the hypervisor.  This is used for
272  * computing time stolen by the hypervisor.
273  */
274 static int alloc_dispatch_logs(void)
275 {
276 	int cpu, ret;
277 	struct paca_struct *pp;
278 	struct dtl_entry *dtl;
279 
280 	if (!firmware_has_feature(FW_FEATURE_SPLPAR))
281 		return 0;
282 
283 	if (!dtl_cache)
284 		return 0;
285 
286 	for_each_possible_cpu(cpu) {
287 		pp = paca_ptrs[cpu];
288 		dtl = kmem_cache_alloc(dtl_cache, GFP_KERNEL);
289 		if (!dtl) {
290 			pr_warn("Failed to allocate dispatch trace log for cpu %d\n",
291 				cpu);
292 			pr_warn("Stolen time statistics will be unreliable\n");
293 			break;
294 		}
295 
296 		pp->dtl_ridx = 0;
297 		pp->dispatch_log = dtl;
298 		pp->dispatch_log_end = dtl + N_DISPATCH_LOG;
299 		pp->dtl_curr = dtl;
300 	}
301 
302 	/* Register the DTL for the current (boot) cpu */
303 	dtl = get_paca()->dispatch_log;
304 	get_paca()->dtl_ridx = 0;
305 	get_paca()->dtl_curr = dtl;
306 	get_paca()->lppaca_ptr->dtl_idx = 0;
307 
308 	/* hypervisor reads buffer length from this field */
309 	dtl->enqueue_to_dispatch_time = cpu_to_be32(DISPATCH_LOG_BYTES);
310 	ret = register_dtl(hard_smp_processor_id(), __pa(dtl));
311 	if (ret)
312 		pr_err("WARNING: DTL registration of cpu %d (hw %d) failed "
313 		       "with %d\n", smp_processor_id(),
314 		       hard_smp_processor_id(), ret);
315 	get_paca()->lppaca_ptr->dtl_enable_mask = 2;
316 
317 	return 0;
318 }
319 #else /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
320 static inline int alloc_dispatch_logs(void)
321 {
322 	return 0;
323 }
324 #endif /* CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
325 
326 static int alloc_dispatch_log_kmem_cache(void)
327 {
328 	dtl_cache = kmem_cache_create("dtl", DISPATCH_LOG_BYTES,
329 						DISPATCH_LOG_BYTES, 0, NULL);
330 	if (!dtl_cache) {
331 		pr_warn("Failed to create dispatch trace log buffer cache\n");
332 		pr_warn("Stolen time statistics will be unreliable\n");
333 		return 0;
334 	}
335 
336 	return alloc_dispatch_logs();
337 }
338 machine_early_initcall(pseries, alloc_dispatch_log_kmem_cache);
339 
340 static void pseries_lpar_idle(void)
341 {
342 	/*
343 	 * Default handler to go into low thread priority and possibly
344 	 * low power mode by ceding processor to hypervisor
345 	 */
346 
347 	/* Indicate to hypervisor that we are idle. */
348 	get_lppaca()->idle = 1;
349 
350 	/*
351 	 * Yield the processor to the hypervisor.  We return if
352 	 * an external interrupt occurs (which are driven prior
353 	 * to returning here) or if a prod occurs from another
354 	 * processor. When returning here, external interrupts
355 	 * are enabled.
356 	 */
357 	cede_processor();
358 
359 	get_lppaca()->idle = 0;
360 }
361 
362 /*
363  * Enable relocation on during exceptions. This has partition wide scope and
364  * may take a while to complete, if it takes longer than one second we will
365  * just give up rather than wasting any more time on this - if that turns out
366  * to ever be a problem in practice we can move this into a kernel thread to
367  * finish off the process later in boot.
368  */
369 void pseries_enable_reloc_on_exc(void)
370 {
371 	long rc;
372 	unsigned int delay, total_delay = 0;
373 
374 	while (1) {
375 		rc = enable_reloc_on_exceptions();
376 		if (!H_IS_LONG_BUSY(rc)) {
377 			if (rc == H_P2) {
378 				pr_info("Relocation on exceptions not"
379 					" supported\n");
380 			} else if (rc != H_SUCCESS) {
381 				pr_warn("Unable to enable relocation"
382 					" on exceptions: %ld\n", rc);
383 			}
384 			break;
385 		}
386 
387 		delay = get_longbusy_msecs(rc);
388 		total_delay += delay;
389 		if (total_delay > 1000) {
390 			pr_warn("Warning: Giving up waiting to enable "
391 				"relocation on exceptions (%u msec)!\n",
392 				total_delay);
393 			return;
394 		}
395 
396 		mdelay(delay);
397 	}
398 }
399 EXPORT_SYMBOL(pseries_enable_reloc_on_exc);
400 
401 void pseries_disable_reloc_on_exc(void)
402 {
403 	long rc;
404 
405 	while (1) {
406 		rc = disable_reloc_on_exceptions();
407 		if (!H_IS_LONG_BUSY(rc))
408 			break;
409 		mdelay(get_longbusy_msecs(rc));
410 	}
411 	if (rc != H_SUCCESS)
412 		pr_warn("Warning: Failed to disable relocation on exceptions: %ld\n",
413 			rc);
414 }
415 EXPORT_SYMBOL(pseries_disable_reloc_on_exc);
416 
417 #ifdef CONFIG_KEXEC_CORE
418 static void pSeries_machine_kexec(struct kimage *image)
419 {
420 	if (firmware_has_feature(FW_FEATURE_SET_MODE))
421 		pseries_disable_reloc_on_exc();
422 
423 	default_machine_kexec(image);
424 }
425 #endif
426 
427 #ifdef __LITTLE_ENDIAN__
428 void pseries_big_endian_exceptions(void)
429 {
430 	long rc;
431 
432 	while (1) {
433 		rc = enable_big_endian_exceptions();
434 		if (!H_IS_LONG_BUSY(rc))
435 			break;
436 		mdelay(get_longbusy_msecs(rc));
437 	}
438 
439 	/*
440 	 * At this point it is unlikely panic() will get anything
441 	 * out to the user, since this is called very late in kexec
442 	 * but at least this will stop us from continuing on further
443 	 * and creating an even more difficult to debug situation.
444 	 *
445 	 * There is a known problem when kdump'ing, if cpus are offline
446 	 * the above call will fail. Rather than panicking again, keep
447 	 * going and hope the kdump kernel is also little endian, which
448 	 * it usually is.
449 	 */
450 	if (rc && !kdump_in_progress())
451 		panic("Could not enable big endian exceptions");
452 }
453 
454 void pseries_little_endian_exceptions(void)
455 {
456 	long rc;
457 
458 	while (1) {
459 		rc = enable_little_endian_exceptions();
460 		if (!H_IS_LONG_BUSY(rc))
461 			break;
462 		mdelay(get_longbusy_msecs(rc));
463 	}
464 	if (rc) {
465 		ppc_md.progress("H_SET_MODE LE exception fail", 0);
466 		panic("Could not enable little endian exceptions");
467 	}
468 }
469 #endif
470 
471 static void __init find_and_init_phbs(void)
472 {
473 	struct device_node *node;
474 	struct pci_controller *phb;
475 	struct device_node *root = of_find_node_by_path("/");
476 
477 	for_each_child_of_node(root, node) {
478 		if (!of_node_is_type(node, "pci") &&
479 		    !of_node_is_type(node, "pciex"))
480 			continue;
481 
482 		phb = pcibios_alloc_controller(node);
483 		if (!phb)
484 			continue;
485 		rtas_setup_phb(phb);
486 		pci_process_bridge_OF_ranges(phb, node, 0);
487 		isa_bridge_find_early(phb);
488 		phb->controller_ops = pseries_pci_controller_ops;
489 	}
490 
491 	of_node_put(root);
492 
493 	/*
494 	 * PCI_PROBE_ONLY and PCI_REASSIGN_ALL_BUS can be set via properties
495 	 * in chosen.
496 	 */
497 	of_pci_check_probe_only();
498 }
499 
500 static void init_cpu_char_feature_flags(struct h_cpu_char_result *result)
501 {
502 	/*
503 	 * The features below are disabled by default, so we instead look to see
504 	 * if firmware has *enabled* them, and set them if so.
505 	 */
506 	if (result->character & H_CPU_CHAR_SPEC_BAR_ORI31)
507 		security_ftr_set(SEC_FTR_SPEC_BAR_ORI31);
508 
509 	if (result->character & H_CPU_CHAR_BCCTRL_SERIALISED)
510 		security_ftr_set(SEC_FTR_BCCTRL_SERIALISED);
511 
512 	if (result->character & H_CPU_CHAR_L1D_FLUSH_ORI30)
513 		security_ftr_set(SEC_FTR_L1D_FLUSH_ORI30);
514 
515 	if (result->character & H_CPU_CHAR_L1D_FLUSH_TRIG2)
516 		security_ftr_set(SEC_FTR_L1D_FLUSH_TRIG2);
517 
518 	if (result->character & H_CPU_CHAR_L1D_THREAD_PRIV)
519 		security_ftr_set(SEC_FTR_L1D_THREAD_PRIV);
520 
521 	if (result->character & H_CPU_CHAR_COUNT_CACHE_DISABLED)
522 		security_ftr_set(SEC_FTR_COUNT_CACHE_DISABLED);
523 
524 	if (result->character & H_CPU_CHAR_BCCTR_FLUSH_ASSIST)
525 		security_ftr_set(SEC_FTR_BCCTR_FLUSH_ASSIST);
526 
527 	if (result->behaviour & H_CPU_BEHAV_FLUSH_COUNT_CACHE)
528 		security_ftr_set(SEC_FTR_FLUSH_COUNT_CACHE);
529 
530 	/*
531 	 * The features below are enabled by default, so we instead look to see
532 	 * if firmware has *disabled* them, and clear them if so.
533 	 */
534 	if (!(result->behaviour & H_CPU_BEHAV_FAVOUR_SECURITY))
535 		security_ftr_clear(SEC_FTR_FAVOUR_SECURITY);
536 
537 	if (!(result->behaviour & H_CPU_BEHAV_L1D_FLUSH_PR))
538 		security_ftr_clear(SEC_FTR_L1D_FLUSH_PR);
539 
540 	if (!(result->behaviour & H_CPU_BEHAV_BNDS_CHK_SPEC_BAR))
541 		security_ftr_clear(SEC_FTR_BNDS_CHK_SPEC_BAR);
542 }
543 
544 void pseries_setup_rfi_flush(void)
545 {
546 	struct h_cpu_char_result result;
547 	enum l1d_flush_type types;
548 	bool enable;
549 	long rc;
550 
551 	/*
552 	 * Set features to the defaults assumed by init_cpu_char_feature_flags()
553 	 * so it can set/clear again any features that might have changed after
554 	 * migration, and in case the hypercall fails and it is not even called.
555 	 */
556 	powerpc_security_features = SEC_FTR_DEFAULT;
557 
558 	rc = plpar_get_cpu_characteristics(&result);
559 	if (rc == H_SUCCESS)
560 		init_cpu_char_feature_flags(&result);
561 
562 	/*
563 	 * We're the guest so this doesn't apply to us, clear it to simplify
564 	 * handling of it elsewhere.
565 	 */
566 	security_ftr_clear(SEC_FTR_L1D_FLUSH_HV);
567 
568 	types = L1D_FLUSH_FALLBACK;
569 
570 	if (security_ftr_enabled(SEC_FTR_L1D_FLUSH_TRIG2))
571 		types |= L1D_FLUSH_MTTRIG;
572 
573 	if (security_ftr_enabled(SEC_FTR_L1D_FLUSH_ORI30))
574 		types |= L1D_FLUSH_ORI;
575 
576 	enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) && \
577 		 security_ftr_enabled(SEC_FTR_L1D_FLUSH_PR);
578 
579 	setup_rfi_flush(types, enable);
580 	setup_count_cache_flush();
581 }
582 
583 #ifdef CONFIG_PCI_IOV
584 enum rtas_iov_fw_value_map {
585 	NUM_RES_PROPERTY  = 0, /* Number of Resources */
586 	LOW_INT           = 1, /* Lowest 32 bits of Address */
587 	START_OF_ENTRIES  = 2, /* Always start of entry */
588 	APERTURE_PROPERTY = 2, /* Start of entry+ to  Aperture Size */
589 	WDW_SIZE_PROPERTY = 4, /* Start of entry+ to Window Size */
590 	NEXT_ENTRY        = 7  /* Go to next entry on array */
591 };
592 
593 enum get_iov_fw_value_index {
594 	BAR_ADDRS     = 1,    /*  Get Bar Address */
595 	APERTURE_SIZE = 2,    /*  Get Aperture Size */
596 	WDW_SIZE      = 3     /*  Get Window Size */
597 };
598 
599 resource_size_t pseries_get_iov_fw_value(struct pci_dev *dev, int resno,
600 					 enum get_iov_fw_value_index value)
601 {
602 	const int *indexes;
603 	struct device_node *dn = pci_device_to_OF_node(dev);
604 	int i, num_res, ret = 0;
605 
606 	indexes = of_get_property(dn, "ibm,open-sriov-vf-bar-info", NULL);
607 	if (!indexes)
608 		return  0;
609 
610 	/*
611 	 * First element in the array is the number of Bars
612 	 * returned.  Search through the list to find the matching
613 	 * bar
614 	 */
615 	num_res = of_read_number(&indexes[NUM_RES_PROPERTY], 1);
616 	if (resno >= num_res)
617 		return 0; /* or an errror */
618 
619 	i = START_OF_ENTRIES + NEXT_ENTRY * resno;
620 	switch (value) {
621 	case BAR_ADDRS:
622 		ret = of_read_number(&indexes[i], 2);
623 		break;
624 	case APERTURE_SIZE:
625 		ret = of_read_number(&indexes[i + APERTURE_PROPERTY], 2);
626 		break;
627 	case WDW_SIZE:
628 		ret = of_read_number(&indexes[i + WDW_SIZE_PROPERTY], 2);
629 		break;
630 	}
631 
632 	return ret;
633 }
634 
635 void of_pci_set_vf_bar_size(struct pci_dev *dev, const int *indexes)
636 {
637 	struct resource *res;
638 	resource_size_t base, size;
639 	int i, r, num_res;
640 
641 	num_res = of_read_number(&indexes[NUM_RES_PROPERTY], 1);
642 	num_res = min_t(int, num_res, PCI_SRIOV_NUM_BARS);
643 	for (i = START_OF_ENTRIES, r = 0; r < num_res && r < PCI_SRIOV_NUM_BARS;
644 	     i += NEXT_ENTRY, r++) {
645 		res = &dev->resource[r + PCI_IOV_RESOURCES];
646 		base = of_read_number(&indexes[i], 2);
647 		size = of_read_number(&indexes[i + APERTURE_PROPERTY], 2);
648 		res->flags = pci_parse_of_flags(of_read_number
649 						(&indexes[i + LOW_INT], 1), 0);
650 		res->flags |= (IORESOURCE_MEM_64 | IORESOURCE_PCI_FIXED);
651 		res->name = pci_name(dev);
652 		res->start = base;
653 		res->end = base + size - 1;
654 	}
655 }
656 
657 void of_pci_parse_iov_addrs(struct pci_dev *dev, const int *indexes)
658 {
659 	struct resource *res, *root, *conflict;
660 	resource_size_t base, size;
661 	int i, r, num_res;
662 
663 	/*
664 	 * First element in the array is the number of Bars
665 	 * returned.  Search through the list to find the matching
666 	 * bars assign them from firmware into resources structure.
667 	 */
668 	num_res = of_read_number(&indexes[NUM_RES_PROPERTY], 1);
669 	for (i = START_OF_ENTRIES, r = 0; r < num_res && r < PCI_SRIOV_NUM_BARS;
670 	     i += NEXT_ENTRY, r++) {
671 		res = &dev->resource[r + PCI_IOV_RESOURCES];
672 		base = of_read_number(&indexes[i], 2);
673 		size = of_read_number(&indexes[i + WDW_SIZE_PROPERTY], 2);
674 		res->name = pci_name(dev);
675 		res->start = base;
676 		res->end = base + size - 1;
677 		root = &iomem_resource;
678 		dev_dbg(&dev->dev,
679 			"pSeries IOV BAR %d: trying firmware assignment %pR\n",
680 			 r + PCI_IOV_RESOURCES, res);
681 		conflict = request_resource_conflict(root, res);
682 		if (conflict) {
683 			dev_info(&dev->dev,
684 				 "BAR %d: %pR conflicts with %s %pR\n",
685 				 r + PCI_IOV_RESOURCES, res,
686 				 conflict->name, conflict);
687 			res->flags |= IORESOURCE_UNSET;
688 		}
689 	}
690 }
691 
692 static void pseries_disable_sriov_resources(struct pci_dev *pdev)
693 {
694 	int i;
695 
696 	pci_warn(pdev, "No hypervisor support for SR-IOV on this device, IOV BARs disabled.\n");
697 	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++)
698 		pdev->resource[i + PCI_IOV_RESOURCES].flags = 0;
699 }
700 
701 static void pseries_pci_fixup_resources(struct pci_dev *pdev)
702 {
703 	const int *indexes;
704 	struct device_node *dn = pci_device_to_OF_node(pdev);
705 
706 	/*Firmware must support open sriov otherwise dont configure*/
707 	indexes = of_get_property(dn, "ibm,open-sriov-vf-bar-info", NULL);
708 	if (indexes)
709 		of_pci_set_vf_bar_size(pdev, indexes);
710 	else
711 		pseries_disable_sriov_resources(pdev);
712 }
713 
714 static void pseries_pci_fixup_iov_resources(struct pci_dev *pdev)
715 {
716 	const int *indexes;
717 	struct device_node *dn = pci_device_to_OF_node(pdev);
718 
719 	if (!pdev->is_physfn || pci_dev_is_added(pdev))
720 		return;
721 	/*Firmware must support open sriov otherwise dont configure*/
722 	indexes = of_get_property(dn, "ibm,open-sriov-vf-bar-info", NULL);
723 	if (indexes)
724 		of_pci_parse_iov_addrs(pdev, indexes);
725 	else
726 		pseries_disable_sriov_resources(pdev);
727 }
728 
729 static resource_size_t pseries_pci_iov_resource_alignment(struct pci_dev *pdev,
730 							  int resno)
731 {
732 	const __be32 *reg;
733 	struct device_node *dn = pci_device_to_OF_node(pdev);
734 
735 	/*Firmware must support open sriov otherwise report regular alignment*/
736 	reg = of_get_property(dn, "ibm,is-open-sriov-pf", NULL);
737 	if (!reg)
738 		return pci_iov_resource_size(pdev, resno);
739 
740 	if (!pdev->is_physfn)
741 		return 0;
742 	return pseries_get_iov_fw_value(pdev,
743 					resno - PCI_IOV_RESOURCES,
744 					APERTURE_SIZE);
745 }
746 #endif
747 
748 static void __init pSeries_setup_arch(void)
749 {
750 	set_arch_panic_timeout(10, ARCH_PANIC_TIMEOUT);
751 
752 	/* Discover PIC type and setup ppc_md accordingly */
753 	smp_init_pseries();
754 
755 
756 	/* openpic global configuration register (64-bit format). */
757 	/* openpic Interrupt Source Unit pointer (64-bit format). */
758 	/* python0 facility area (mmio) (64-bit format) REAL address. */
759 
760 	/* init to some ~sane value until calibrate_delay() runs */
761 	loops_per_jiffy = 50000000;
762 
763 	fwnmi_init();
764 
765 	pseries_setup_rfi_flush();
766 	setup_stf_barrier();
767 
768 	/* By default, only probe PCI (can be overridden by rtas_pci) */
769 	pci_add_flags(PCI_PROBE_ONLY);
770 
771 	/* Find and initialize PCI host bridges */
772 	init_pci_config_tokens();
773 	find_and_init_phbs();
774 	of_reconfig_notifier_register(&pci_dn_reconfig_nb);
775 
776 	pSeries_nvram_init();
777 
778 	if (firmware_has_feature(FW_FEATURE_LPAR)) {
779 		vpa_init(boot_cpuid);
780 		ppc_md.power_save = pseries_lpar_idle;
781 		ppc_md.enable_pmcs = pseries_lpar_enable_pmcs;
782 #ifdef CONFIG_PCI_IOV
783 		ppc_md.pcibios_fixup_resources =
784 			pseries_pci_fixup_resources;
785 		ppc_md.pcibios_fixup_sriov =
786 			pseries_pci_fixup_iov_resources;
787 		ppc_md.pcibios_iov_resource_alignment =
788 			pseries_pci_iov_resource_alignment;
789 #endif
790 	} else {
791 		/* No special idle routine */
792 		ppc_md.enable_pmcs = power4_enable_pmcs;
793 	}
794 
795 	ppc_md.pcibios_root_bridge_prepare = pseries_root_bridge_prepare;
796 }
797 
798 static void pseries_panic(char *str)
799 {
800 	panic_flush_kmsg_end();
801 	rtas_os_term(str);
802 }
803 
804 static int __init pSeries_init_panel(void)
805 {
806 	/* Manually leave the kernel version on the panel. */
807 #ifdef __BIG_ENDIAN__
808 	ppc_md.progress("Linux ppc64\n", 0);
809 #else
810 	ppc_md.progress("Linux ppc64le\n", 0);
811 #endif
812 	ppc_md.progress(init_utsname()->version, 0);
813 
814 	return 0;
815 }
816 machine_arch_initcall(pseries, pSeries_init_panel);
817 
818 static int pseries_set_dabr(unsigned long dabr, unsigned long dabrx)
819 {
820 	return plpar_hcall_norets(H_SET_DABR, dabr);
821 }
822 
823 static int pseries_set_xdabr(unsigned long dabr, unsigned long dabrx)
824 {
825 	/* Have to set at least one bit in the DABRX according to PAPR */
826 	if (dabrx == 0 && dabr == 0)
827 		dabrx = DABRX_USER;
828 	/* PAPR says we can only set kernel and user bits */
829 	dabrx &= DABRX_KERNEL | DABRX_USER;
830 
831 	return plpar_hcall_norets(H_SET_XDABR, dabr, dabrx);
832 }
833 
834 static int pseries_set_dawr(unsigned long dawr, unsigned long dawrx)
835 {
836 	/* PAPR says we can't set HYP */
837 	dawrx &= ~DAWRX_HYP;
838 
839 	return  plpar_set_watchpoint0(dawr, dawrx);
840 }
841 
842 #define CMO_CHARACTERISTICS_TOKEN 44
843 #define CMO_MAXLENGTH 1026
844 
845 void pSeries_coalesce_init(void)
846 {
847 	struct hvcall_mpp_x_data mpp_x_data;
848 
849 	if (firmware_has_feature(FW_FEATURE_CMO) && !h_get_mpp_x(&mpp_x_data))
850 		powerpc_firmware_features |= FW_FEATURE_XCMO;
851 	else
852 		powerpc_firmware_features &= ~FW_FEATURE_XCMO;
853 }
854 
855 /**
856  * fw_cmo_feature_init - FW_FEATURE_CMO is not stored in ibm,hypertas-functions,
857  * handle that here. (Stolen from parse_system_parameter_string)
858  */
859 static void pSeries_cmo_feature_init(void)
860 {
861 	char *ptr, *key, *value, *end;
862 	int call_status;
863 	int page_order = IOMMU_PAGE_SHIFT_4K;
864 
865 	pr_debug(" -> fw_cmo_feature_init()\n");
866 	spin_lock(&rtas_data_buf_lock);
867 	memset(rtas_data_buf, 0, RTAS_DATA_BUF_SIZE);
868 	call_status = rtas_call(rtas_token("ibm,get-system-parameter"), 3, 1,
869 				NULL,
870 				CMO_CHARACTERISTICS_TOKEN,
871 				__pa(rtas_data_buf),
872 				RTAS_DATA_BUF_SIZE);
873 
874 	if (call_status != 0) {
875 		spin_unlock(&rtas_data_buf_lock);
876 		pr_debug("CMO not available\n");
877 		pr_debug(" <- fw_cmo_feature_init()\n");
878 		return;
879 	}
880 
881 	end = rtas_data_buf + CMO_MAXLENGTH - 2;
882 	ptr = rtas_data_buf + 2;	/* step over strlen value */
883 	key = value = ptr;
884 
885 	while (*ptr && (ptr <= end)) {
886 		/* Separate the key and value by replacing '=' with '\0' and
887 		 * point the value at the string after the '='
888 		 */
889 		if (ptr[0] == '=') {
890 			ptr[0] = '\0';
891 			value = ptr + 1;
892 		} else if (ptr[0] == '\0' || ptr[0] == ',') {
893 			/* Terminate the string containing the key/value pair */
894 			ptr[0] = '\0';
895 
896 			if (key == value) {
897 				pr_debug("Malformed key/value pair\n");
898 				/* Never found a '=', end processing */
899 				break;
900 			}
901 
902 			if (0 == strcmp(key, "CMOPageSize"))
903 				page_order = simple_strtol(value, NULL, 10);
904 			else if (0 == strcmp(key, "PrPSP"))
905 				CMO_PrPSP = simple_strtol(value, NULL, 10);
906 			else if (0 == strcmp(key, "SecPSP"))
907 				CMO_SecPSP = simple_strtol(value, NULL, 10);
908 			value = key = ptr + 1;
909 		}
910 		ptr++;
911 	}
912 
913 	/* Page size is returned as the power of 2 of the page size,
914 	 * convert to the page size in bytes before returning
915 	 */
916 	CMO_PageSize = 1 << page_order;
917 	pr_debug("CMO_PageSize = %lu\n", CMO_PageSize);
918 
919 	if (CMO_PrPSP != -1 || CMO_SecPSP != -1) {
920 		pr_info("CMO enabled\n");
921 		pr_debug("CMO enabled, PrPSP=%d, SecPSP=%d\n", CMO_PrPSP,
922 		         CMO_SecPSP);
923 		powerpc_firmware_features |= FW_FEATURE_CMO;
924 		pSeries_coalesce_init();
925 	} else
926 		pr_debug("CMO not enabled, PrPSP=%d, SecPSP=%d\n", CMO_PrPSP,
927 		         CMO_SecPSP);
928 	spin_unlock(&rtas_data_buf_lock);
929 	pr_debug(" <- fw_cmo_feature_init()\n");
930 }
931 
932 /*
933  * Early initialization.  Relocation is on but do not reference unbolted pages
934  */
935 static void __init pseries_init(void)
936 {
937 	pr_debug(" -> pseries_init()\n");
938 
939 #ifdef CONFIG_HVC_CONSOLE
940 	if (firmware_has_feature(FW_FEATURE_LPAR))
941 		hvc_vio_init_early();
942 #endif
943 	if (firmware_has_feature(FW_FEATURE_XDABR))
944 		ppc_md.set_dabr = pseries_set_xdabr;
945 	else if (firmware_has_feature(FW_FEATURE_DABR))
946 		ppc_md.set_dabr = pseries_set_dabr;
947 
948 	if (firmware_has_feature(FW_FEATURE_SET_MODE))
949 		ppc_md.set_dawr = pseries_set_dawr;
950 
951 	pSeries_cmo_feature_init();
952 	iommu_init_early_pSeries();
953 
954 	pr_debug(" <- pseries_init()\n");
955 }
956 
957 /**
958  * pseries_power_off - tell firmware about how to power off the system.
959  *
960  * This function calls either the power-off rtas token in normal cases
961  * or the ibm,power-off-ups token (if present & requested) in case of
962  * a power failure. If power-off token is used, power on will only be
963  * possible with power button press. If ibm,power-off-ups token is used
964  * it will allow auto poweron after power is restored.
965  */
966 static void pseries_power_off(void)
967 {
968 	int rc;
969 	int rtas_poweroff_ups_token = rtas_token("ibm,power-off-ups");
970 
971 	if (rtas_flash_term_hook)
972 		rtas_flash_term_hook(SYS_POWER_OFF);
973 
974 	if (rtas_poweron_auto == 0 ||
975 		rtas_poweroff_ups_token == RTAS_UNKNOWN_SERVICE) {
976 		rc = rtas_call(rtas_token("power-off"), 2, 1, NULL, -1, -1);
977 		printk(KERN_INFO "RTAS power-off returned %d\n", rc);
978 	} else {
979 		rc = rtas_call(rtas_poweroff_ups_token, 0, 1, NULL);
980 		printk(KERN_INFO "RTAS ibm,power-off-ups returned %d\n", rc);
981 	}
982 	for (;;);
983 }
984 
985 static int __init pSeries_probe(void)
986 {
987 	if (!of_node_is_type(of_root, "chrp"))
988 		return 0;
989 
990 	/* Cell blades firmware claims to be chrp while it's not. Until this
991 	 * is fixed, we need to avoid those here.
992 	 */
993 	if (of_machine_is_compatible("IBM,CPBW-1.0") ||
994 	    of_machine_is_compatible("IBM,CBEA"))
995 		return 0;
996 
997 	pm_power_off = pseries_power_off;
998 
999 	pr_debug("Machine is%s LPAR !\n",
1000 	         (powerpc_firmware_features & FW_FEATURE_LPAR) ? "" : " not");
1001 
1002 	pseries_init();
1003 
1004 	return 1;
1005 }
1006 
1007 static int pSeries_pci_probe_mode(struct pci_bus *bus)
1008 {
1009 	if (firmware_has_feature(FW_FEATURE_LPAR))
1010 		return PCI_PROBE_DEVTREE;
1011 	return PCI_PROBE_NORMAL;
1012 }
1013 
1014 struct pci_controller_ops pseries_pci_controller_ops = {
1015 	.probe_mode		= pSeries_pci_probe_mode,
1016 };
1017 
1018 define_machine(pseries) {
1019 	.name			= "pSeries",
1020 	.probe			= pSeries_probe,
1021 	.setup_arch		= pSeries_setup_arch,
1022 	.init_IRQ		= pseries_init_irq,
1023 	.show_cpuinfo		= pSeries_show_cpuinfo,
1024 	.log_error		= pSeries_log_error,
1025 	.pcibios_fixup		= pSeries_final_fixup,
1026 	.restart		= rtas_restart,
1027 	.halt			= rtas_halt,
1028 	.panic			= pseries_panic,
1029 	.get_boot_time		= rtas_get_boot_time,
1030 	.get_rtc_time		= rtas_get_rtc_time,
1031 	.set_rtc_time		= rtas_set_rtc_time,
1032 	.calibrate_decr		= generic_calibrate_decr,
1033 	.progress		= rtas_progress,
1034 	.system_reset_exception = pSeries_system_reset_exception,
1035 	.machine_check_early	= pseries_machine_check_realmode,
1036 	.machine_check_exception = pSeries_machine_check_exception,
1037 #ifdef CONFIG_KEXEC_CORE
1038 	.machine_kexec          = pSeries_machine_kexec,
1039 	.kexec_cpu_down         = pseries_kexec_cpu_down,
1040 #endif
1041 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
1042 	.memory_block_size	= pseries_memory_block_size,
1043 #endif
1044 };
1045