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/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#include <asm/idle.h>
71#include <asm/swiotlb.h>
72#include <asm/svm.h>
73#include <asm/dtl.h>
74
75#include "pseries.h"
76#include "../../../../drivers/pci/pci.h"
77
78DEFINE_STATIC_KEY_FALSE(shared_processor);
79EXPORT_SYMBOL(shared_processor);
80
81int CMO_PrPSP = -1;
82int CMO_SecPSP = -1;
83unsigned long CMO_PageSize = (ASM_CONST(1) << IOMMU_PAGE_SHIFT_4K);
84EXPORT_SYMBOL(CMO_PageSize);
85
86int fwnmi_active;  /* TRUE if an FWNMI handler is present */
87int ibm_nmi_interlock_token;
88
89static void pSeries_show_cpuinfo(struct seq_file *m)
90{
91	struct device_node *root;
92	const char *model = "";
93
94	root = of_find_node_by_path("/");
95	if (root)
96		model = of_get_property(root, "model", NULL);
97	seq_printf(m, "machine\t\t: CHRP %s\n", model);
98	of_node_put(root);
99	if (radix_enabled())
100		seq_printf(m, "MMU\t\t: Radix\n");
101	else
102		seq_printf(m, "MMU\t\t: Hash\n");
103}
104
105/* Initialize firmware assisted non-maskable interrupts if
106 * the firmware supports this feature.
107 */
108static void __init fwnmi_init(void)
109{
110	unsigned long system_reset_addr, machine_check_addr;
111	u8 *mce_data_buf;
112	unsigned int i;
113	int nr_cpus = num_possible_cpus();
114#ifdef CONFIG_PPC_BOOK3S_64
115	struct slb_entry *slb_ptr;
116	size_t size;
117#endif
118	int ibm_nmi_register_token;
119
120	ibm_nmi_register_token = rtas_token("ibm,nmi-register");
121	if (ibm_nmi_register_token == RTAS_UNKNOWN_SERVICE)
122		return;
123
124	ibm_nmi_interlock_token = rtas_token("ibm,nmi-interlock");
125	if (WARN_ON(ibm_nmi_interlock_token == RTAS_UNKNOWN_SERVICE))
126		return;
127
128	/* If the kernel's not linked at zero we point the firmware at low
129	 * addresses anyway, and use a trampoline to get to the real code. */
130	system_reset_addr  = __pa(system_reset_fwnmi) - PHYSICAL_START;
131	machine_check_addr = __pa(machine_check_fwnmi) - PHYSICAL_START;
132
133	if (0 == rtas_call(ibm_nmi_register_token, 2, 1, NULL,
134			   system_reset_addr, machine_check_addr))
135		fwnmi_active = 1;
136
137	/*
138	 * Allocate a chunk for per cpu buffer to hold rtas errorlog.
139	 * It will be used in real mode mce handler, hence it needs to be
140	 * below RMA.
141	 */
142	mce_data_buf = memblock_alloc_try_nid_raw(RTAS_ERROR_LOG_MAX * nr_cpus,
143					RTAS_ERROR_LOG_MAX, MEMBLOCK_LOW_LIMIT,
144					ppc64_rma_size, NUMA_NO_NODE);
145	if (!mce_data_buf)
146		panic("Failed to allocate %d bytes below %pa for MCE buffer\n",
147		      RTAS_ERROR_LOG_MAX * nr_cpus, &ppc64_rma_size);
148
149	for_each_possible_cpu(i) {
150		paca_ptrs[i]->mce_data_buf = mce_data_buf +
151						(RTAS_ERROR_LOG_MAX * i);
152	}
153
154#ifdef CONFIG_PPC_BOOK3S_64
155	if (!radix_enabled()) {
156		/* Allocate per cpu area to save old slb contents during MCE */
157		size = sizeof(struct slb_entry) * mmu_slb_size * nr_cpus;
158		slb_ptr = memblock_alloc_try_nid_raw(size,
159				sizeof(struct slb_entry), MEMBLOCK_LOW_LIMIT,
160				ppc64_rma_size, NUMA_NO_NODE);
161		if (!slb_ptr)
162			panic("Failed to allocate %zu bytes below %pa for slb area\n",
163			      size, &ppc64_rma_size);
164
165		for_each_possible_cpu(i)
166			paca_ptrs[i]->mce_faulty_slbs = slb_ptr + (mmu_slb_size * i);
167	}
168#endif
169}
170
171static void pseries_8259_cascade(struct irq_desc *desc)
172{
173	struct irq_chip *chip = irq_desc_get_chip(desc);
174	unsigned int cascade_irq = i8259_irq();
175
176	if (cascade_irq)
177		generic_handle_irq(cascade_irq);
178
179	chip->irq_eoi(&desc->irq_data);
180}
181
182static void __init pseries_setup_i8259_cascade(void)
183{
184	struct device_node *np, *old, *found = NULL;
185	unsigned int cascade;
186	const u32 *addrp;
187	unsigned long intack = 0;
188	int naddr;
189
190	for_each_node_by_type(np, "interrupt-controller") {
191		if (of_device_is_compatible(np, "chrp,iic")) {
192			found = np;
193			break;
194		}
195	}
196
197	if (found == NULL) {
198		printk(KERN_DEBUG "pic: no ISA interrupt controller\n");
199		return;
200	}
201
202	cascade = irq_of_parse_and_map(found, 0);
203	if (!cascade) {
204		printk(KERN_ERR "pic: failed to map cascade interrupt");
205		return;
206	}
207	pr_debug("pic: cascade mapped to irq %d\n", cascade);
208
209	for (old = of_node_get(found); old != NULL ; old = np) {
210		np = of_get_parent(old);
211		of_node_put(old);
212		if (np == NULL)
213			break;
214		if (!of_node_name_eq(np, "pci"))
215			continue;
216		addrp = of_get_property(np, "8259-interrupt-acknowledge", NULL);
217		if (addrp == NULL)
218			continue;
219		naddr = of_n_addr_cells(np);
220		intack = addrp[naddr-1];
221		if (naddr > 1)
222			intack |= ((unsigned long)addrp[naddr-2]) << 32;
223	}
224	if (intack)
225		printk(KERN_DEBUG "pic: PCI 8259 intack at 0x%016lx\n", intack);
226	i8259_init(found, intack);
227	of_node_put(found);
228	irq_set_chained_handler(cascade, pseries_8259_cascade);
229}
230
231static void __init pseries_init_irq(void)
232{
233	/* Try using a XIVE if available, otherwise use a XICS */
234	if (!xive_spapr_init()) {
235		xics_init();
236		pseries_setup_i8259_cascade();
237	}
238}
239
240static void pseries_lpar_enable_pmcs(void)
241{
242	unsigned long set, reset;
243
244	set = 1UL << 63;
245	reset = 0;
246	plpar_hcall_norets(H_PERFMON, set, reset);
247}
248
249static int pci_dn_reconfig_notifier(struct notifier_block *nb, unsigned long action, void *data)
250{
251	struct of_reconfig_data *rd = data;
252	struct device_node *parent, *np = rd->dn;
253	struct pci_dn *pdn;
254	int err = NOTIFY_OK;
255
256	switch (action) {
257	case OF_RECONFIG_ATTACH_NODE:
258		parent = of_get_parent(np);
259		pdn = parent ? PCI_DN(parent) : NULL;
260		if (pdn)
261			pci_add_device_node_info(pdn->phb, np);
262
263		of_node_put(parent);
264		break;
265	case OF_RECONFIG_DETACH_NODE:
266		pdn = PCI_DN(np);
267		if (pdn)
268			list_del(&pdn->list);
269		break;
270	default:
271		err = NOTIFY_DONE;
272		break;
273	}
274	return err;
275}
276
277static struct notifier_block pci_dn_reconfig_nb = {
278	.notifier_call = pci_dn_reconfig_notifier,
279};
280
281struct kmem_cache *dtl_cache;
282
283#ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
284/*
285 * Allocate space for the dispatch trace log for all possible cpus
286 * and register the buffers with the hypervisor.  This is used for
287 * computing time stolen by the hypervisor.
288 */
289static int alloc_dispatch_logs(void)
290{
291	if (!firmware_has_feature(FW_FEATURE_SPLPAR))
292		return 0;
293
294	if (!dtl_cache)
295		return 0;
296
297	alloc_dtl_buffers(0);
298
299	/* Register the DTL for the current (boot) cpu */
300	register_dtl_buffer(smp_processor_id());
301
302	return 0;
303}
304#else /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
305static inline int alloc_dispatch_logs(void)
306{
307	return 0;
308}
309#endif /* CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
310
311static int alloc_dispatch_log_kmem_cache(void)
312{
313	void (*ctor)(void *) = get_dtl_cache_ctor();
314
315	dtl_cache = kmem_cache_create("dtl", DISPATCH_LOG_BYTES,
316						DISPATCH_LOG_BYTES, 0, ctor);
317	if (!dtl_cache) {
318		pr_warn("Failed to create dispatch trace log buffer cache\n");
319		pr_warn("Stolen time statistics will be unreliable\n");
320		return 0;
321	}
322
323	return alloc_dispatch_logs();
324}
325machine_early_initcall(pseries, alloc_dispatch_log_kmem_cache);
326
327DEFINE_PER_CPU(u64, idle_spurr_cycles);
328DEFINE_PER_CPU(u64, idle_entry_purr_snap);
329DEFINE_PER_CPU(u64, idle_entry_spurr_snap);
330static void pseries_lpar_idle(void)
331{
332	/*
333	 * Default handler to go into low thread priority and possibly
334	 * low power mode by ceding processor to hypervisor
335	 */
336
337	if (!prep_irq_for_idle())
338		return;
339
340	/* Indicate to hypervisor that we are idle. */
341	pseries_idle_prolog();
342
343	/*
344	 * Yield the processor to the hypervisor.  We return if
345	 * an external interrupt occurs (which are driven prior
346	 * to returning here) or if a prod occurs from another
347	 * processor. When returning here, external interrupts
348	 * are enabled.
349	 */
350	cede_processor();
351
352	pseries_idle_epilog();
353}
354
355/*
356 * Enable relocation on during exceptions. This has partition wide scope and
357 * may take a while to complete, if it takes longer than one second we will
358 * just give up rather than wasting any more time on this - if that turns out
359 * to ever be a problem in practice we can move this into a kernel thread to
360 * finish off the process later in boot.
361 */
362bool pseries_enable_reloc_on_exc(void)
363{
364	long rc;
365	unsigned int delay, total_delay = 0;
366
367	while (1) {
368		rc = enable_reloc_on_exceptions();
369		if (!H_IS_LONG_BUSY(rc)) {
370			if (rc == H_P2) {
371				pr_info("Relocation on exceptions not"
372					" supported\n");
373				return false;
374			} else if (rc != H_SUCCESS) {
375				pr_warn("Unable to enable relocation"
376					" on exceptions: %ld\n", rc);
377				return false;
378			}
379			return true;
380		}
381
382		delay = get_longbusy_msecs(rc);
383		total_delay += delay;
384		if (total_delay > 1000) {
385			pr_warn("Warning: Giving up waiting to enable "
386				"relocation on exceptions (%u msec)!\n",
387				total_delay);
388			return false;
389		}
390
391		mdelay(delay);
392	}
393}
394EXPORT_SYMBOL(pseries_enable_reloc_on_exc);
395
396void pseries_disable_reloc_on_exc(void)
397{
398	long rc;
399
400	while (1) {
401		rc = disable_reloc_on_exceptions();
402		if (!H_IS_LONG_BUSY(rc))
403			break;
404		mdelay(get_longbusy_msecs(rc));
405	}
406	if (rc != H_SUCCESS)
407		pr_warn("Warning: Failed to disable relocation on exceptions: %ld\n",
408			rc);
409}
410EXPORT_SYMBOL(pseries_disable_reloc_on_exc);
411
412#ifdef CONFIG_KEXEC_CORE
413static void pSeries_machine_kexec(struct kimage *image)
414{
415	if (firmware_has_feature(FW_FEATURE_SET_MODE))
416		pseries_disable_reloc_on_exc();
417
418	default_machine_kexec(image);
419}
420#endif
421
422#ifdef __LITTLE_ENDIAN__
423void pseries_big_endian_exceptions(void)
424{
425	long rc;
426
427	while (1) {
428		rc = enable_big_endian_exceptions();
429		if (!H_IS_LONG_BUSY(rc))
430			break;
431		mdelay(get_longbusy_msecs(rc));
432	}
433
434	/*
435	 * At this point it is unlikely panic() will get anything
436	 * out to the user, since this is called very late in kexec
437	 * but at least this will stop us from continuing on further
438	 * and creating an even more difficult to debug situation.
439	 *
440	 * There is a known problem when kdump'ing, if cpus are offline
441	 * the above call will fail. Rather than panicking again, keep
442	 * going and hope the kdump kernel is also little endian, which
443	 * it usually is.
444	 */
445	if (rc && !kdump_in_progress())
446		panic("Could not enable big endian exceptions");
447}
448
449void pseries_little_endian_exceptions(void)
450{
451	long rc;
452
453	while (1) {
454		rc = enable_little_endian_exceptions();
455		if (!H_IS_LONG_BUSY(rc))
456			break;
457		mdelay(get_longbusy_msecs(rc));
458	}
459	if (rc) {
460		ppc_md.progress("H_SET_MODE LE exception fail", 0);
461		panic("Could not enable little endian exceptions");
462	}
463}
464#endif
465
466static void __init find_and_init_phbs(void)
467{
468	struct device_node *node;
469	struct pci_controller *phb;
470	struct device_node *root = of_find_node_by_path("/");
471
472	for_each_child_of_node(root, node) {
473		if (!of_node_is_type(node, "pci") &&
474		    !of_node_is_type(node, "pciex"))
475			continue;
476
477		phb = pcibios_alloc_controller(node);
478		if (!phb)
479			continue;
480		rtas_setup_phb(phb);
481		pci_process_bridge_OF_ranges(phb, node, 0);
482		isa_bridge_find_early(phb);
483		phb->controller_ops = pseries_pci_controller_ops;
484	}
485
486	of_node_put(root);
487
488	/*
489	 * PCI_PROBE_ONLY and PCI_REASSIGN_ALL_BUS can be set via properties
490	 * in chosen.
491	 */
492	of_pci_check_probe_only();
493}
494
495static void init_cpu_char_feature_flags(struct h_cpu_char_result *result)
496{
497	/*
498	 * The features below are disabled by default, so we instead look to see
499	 * if firmware has *enabled* them, and set them if so.
500	 */
501	if (result->character & H_CPU_CHAR_SPEC_BAR_ORI31)
502		security_ftr_set(SEC_FTR_SPEC_BAR_ORI31);
503
504	if (result->character & H_CPU_CHAR_BCCTRL_SERIALISED)
505		security_ftr_set(SEC_FTR_BCCTRL_SERIALISED);
506
507	if (result->character & H_CPU_CHAR_L1D_FLUSH_ORI30)
508		security_ftr_set(SEC_FTR_L1D_FLUSH_ORI30);
509
510	if (result->character & H_CPU_CHAR_L1D_FLUSH_TRIG2)
511		security_ftr_set(SEC_FTR_L1D_FLUSH_TRIG2);
512
513	if (result->character & H_CPU_CHAR_L1D_THREAD_PRIV)
514		security_ftr_set(SEC_FTR_L1D_THREAD_PRIV);
515
516	if (result->character & H_CPU_CHAR_COUNT_CACHE_DISABLED)
517		security_ftr_set(SEC_FTR_COUNT_CACHE_DISABLED);
518
519	if (result->character & H_CPU_CHAR_BCCTR_FLUSH_ASSIST)
520		security_ftr_set(SEC_FTR_BCCTR_FLUSH_ASSIST);
521
522	if (result->character & H_CPU_CHAR_BCCTR_LINK_FLUSH_ASSIST)
523		security_ftr_set(SEC_FTR_BCCTR_LINK_FLUSH_ASSIST);
524
525	if (result->behaviour & H_CPU_BEHAV_FLUSH_COUNT_CACHE)
526		security_ftr_set(SEC_FTR_FLUSH_COUNT_CACHE);
527
528	if (result->behaviour & H_CPU_BEHAV_FLUSH_LINK_STACK)
529		security_ftr_set(SEC_FTR_FLUSH_LINK_STACK);
530
531	/*
532	 * The features below are enabled by default, so we instead look to see
533	 * if firmware has *disabled* them, and clear them if so.
534	 */
535	if (!(result->behaviour & H_CPU_BEHAV_FAVOUR_SECURITY))
536		security_ftr_clear(SEC_FTR_FAVOUR_SECURITY);
537
538	if (!(result->behaviour & H_CPU_BEHAV_L1D_FLUSH_PR))
539		security_ftr_clear(SEC_FTR_L1D_FLUSH_PR);
540
541	if (result->behaviour & H_CPU_BEHAV_NO_L1D_FLUSH_ENTRY)
542		security_ftr_clear(SEC_FTR_L1D_FLUSH_ENTRY);
543
544	if (result->behaviour & H_CPU_BEHAV_NO_L1D_FLUSH_UACCESS)
545		security_ftr_clear(SEC_FTR_L1D_FLUSH_UACCESS);
546
547	if (!(result->behaviour & H_CPU_BEHAV_BNDS_CHK_SPEC_BAR))
548		security_ftr_clear(SEC_FTR_BNDS_CHK_SPEC_BAR);
549}
550
551void pseries_setup_security_mitigations(void)
552{
553	struct h_cpu_char_result result;
554	enum l1d_flush_type types;
555	bool enable;
556	long rc;
557
558	/*
559	 * Set features to the defaults assumed by init_cpu_char_feature_flags()
560	 * so it can set/clear again any features that might have changed after
561	 * migration, and in case the hypercall fails and it is not even called.
562	 */
563	powerpc_security_features = SEC_FTR_DEFAULT;
564
565	rc = plpar_get_cpu_characteristics(&result);
566	if (rc == H_SUCCESS)
567		init_cpu_char_feature_flags(&result);
568
569	/*
570	 * We're the guest so this doesn't apply to us, clear it to simplify
571	 * handling of it elsewhere.
572	 */
573	security_ftr_clear(SEC_FTR_L1D_FLUSH_HV);
574
575	types = L1D_FLUSH_FALLBACK;
576
577	if (security_ftr_enabled(SEC_FTR_L1D_FLUSH_TRIG2))
578		types |= L1D_FLUSH_MTTRIG;
579
580	if (security_ftr_enabled(SEC_FTR_L1D_FLUSH_ORI30))
581		types |= L1D_FLUSH_ORI;
582
583	enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) && \
584		 security_ftr_enabled(SEC_FTR_L1D_FLUSH_PR);
585
586	setup_rfi_flush(types, enable);
587	setup_count_cache_flush();
588
589	enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) &&
590		 security_ftr_enabled(SEC_FTR_L1D_FLUSH_ENTRY);
591	setup_entry_flush(enable);
592
593	enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) &&
594		 security_ftr_enabled(SEC_FTR_L1D_FLUSH_UACCESS);
595	setup_uaccess_flush(enable);
596
597	setup_stf_barrier();
598}
599
600#ifdef CONFIG_PCI_IOV
601enum rtas_iov_fw_value_map {
602	NUM_RES_PROPERTY  = 0, /* Number of Resources */
603	LOW_INT           = 1, /* Lowest 32 bits of Address */
604	START_OF_ENTRIES  = 2, /* Always start of entry */
605	APERTURE_PROPERTY = 2, /* Start of entry+ to  Aperture Size */
606	WDW_SIZE_PROPERTY = 4, /* Start of entry+ to Window Size */
607	NEXT_ENTRY        = 7  /* Go to next entry on array */
608};
609
610enum get_iov_fw_value_index {
611	BAR_ADDRS     = 1,    /*  Get Bar Address */
612	APERTURE_SIZE = 2,    /*  Get Aperture Size */
613	WDW_SIZE      = 3     /*  Get Window Size */
614};
615
616resource_size_t pseries_get_iov_fw_value(struct pci_dev *dev, int resno,
617					 enum get_iov_fw_value_index value)
618{
619	const int *indexes;
620	struct device_node *dn = pci_device_to_OF_node(dev);
621	int i, num_res, ret = 0;
622
623	indexes = of_get_property(dn, "ibm,open-sriov-vf-bar-info", NULL);
624	if (!indexes)
625		return  0;
626
627	/*
628	 * First element in the array is the number of Bars
629	 * returned.  Search through the list to find the matching
630	 * bar
631	 */
632	num_res = of_read_number(&indexes[NUM_RES_PROPERTY], 1);
633	if (resno >= num_res)
634		return 0; /* or an errror */
635
636	i = START_OF_ENTRIES + NEXT_ENTRY * resno;
637	switch (value) {
638	case BAR_ADDRS:
639		ret = of_read_number(&indexes[i], 2);
640		break;
641	case APERTURE_SIZE:
642		ret = of_read_number(&indexes[i + APERTURE_PROPERTY], 2);
643		break;
644	case WDW_SIZE:
645		ret = of_read_number(&indexes[i + WDW_SIZE_PROPERTY], 2);
646		break;
647	}
648
649	return ret;
650}
651
652void of_pci_set_vf_bar_size(struct pci_dev *dev, const int *indexes)
653{
654	struct resource *res;
655	resource_size_t base, size;
656	int i, r, num_res;
657
658	num_res = of_read_number(&indexes[NUM_RES_PROPERTY], 1);
659	num_res = min_t(int, num_res, PCI_SRIOV_NUM_BARS);
660	for (i = START_OF_ENTRIES, r = 0; r < num_res && r < PCI_SRIOV_NUM_BARS;
661	     i += NEXT_ENTRY, r++) {
662		res = &dev->resource[r + PCI_IOV_RESOURCES];
663		base = of_read_number(&indexes[i], 2);
664		size = of_read_number(&indexes[i + APERTURE_PROPERTY], 2);
665		res->flags = pci_parse_of_flags(of_read_number
666						(&indexes[i + LOW_INT], 1), 0);
667		res->flags |= (IORESOURCE_MEM_64 | IORESOURCE_PCI_FIXED);
668		res->name = pci_name(dev);
669		res->start = base;
670		res->end = base + size - 1;
671	}
672}
673
674void of_pci_parse_iov_addrs(struct pci_dev *dev, const int *indexes)
675{
676	struct resource *res, *root, *conflict;
677	resource_size_t base, size;
678	int i, r, num_res;
679
680	/*
681	 * First element in the array is the number of Bars
682	 * returned.  Search through the list to find the matching
683	 * bars assign them from firmware into resources structure.
684	 */
685	num_res = of_read_number(&indexes[NUM_RES_PROPERTY], 1);
686	for (i = START_OF_ENTRIES, r = 0; r < num_res && r < PCI_SRIOV_NUM_BARS;
687	     i += NEXT_ENTRY, r++) {
688		res = &dev->resource[r + PCI_IOV_RESOURCES];
689		base = of_read_number(&indexes[i], 2);
690		size = of_read_number(&indexes[i + WDW_SIZE_PROPERTY], 2);
691		res->name = pci_name(dev);
692		res->start = base;
693		res->end = base + size - 1;
694		root = &iomem_resource;
695		dev_dbg(&dev->dev,
696			"pSeries IOV BAR %d: trying firmware assignment %pR\n",
697			 r + PCI_IOV_RESOURCES, res);
698		conflict = request_resource_conflict(root, res);
699		if (conflict) {
700			dev_info(&dev->dev,
701				 "BAR %d: %pR conflicts with %s %pR\n",
702				 r + PCI_IOV_RESOURCES, res,
703				 conflict->name, conflict);
704			res->flags |= IORESOURCE_UNSET;
705		}
706	}
707}
708
709static void pseries_disable_sriov_resources(struct pci_dev *pdev)
710{
711	int i;
712
713	pci_warn(pdev, "No hypervisor support for SR-IOV on this device, IOV BARs disabled.\n");
714	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++)
715		pdev->resource[i + PCI_IOV_RESOURCES].flags = 0;
716}
717
718static void pseries_pci_fixup_resources(struct pci_dev *pdev)
719{
720	const int *indexes;
721	struct device_node *dn = pci_device_to_OF_node(pdev);
722
723	/*Firmware must support open sriov otherwise dont configure*/
724	indexes = of_get_property(dn, "ibm,open-sriov-vf-bar-info", NULL);
725	if (indexes)
726		of_pci_set_vf_bar_size(pdev, indexes);
727	else
728		pseries_disable_sriov_resources(pdev);
729}
730
731static void pseries_pci_fixup_iov_resources(struct pci_dev *pdev)
732{
733	const int *indexes;
734	struct device_node *dn = pci_device_to_OF_node(pdev);
735
736	if (!pdev->is_physfn || pci_dev_is_added(pdev))
737		return;
738	/*Firmware must support open sriov otherwise dont configure*/
739	indexes = of_get_property(dn, "ibm,open-sriov-vf-bar-info", NULL);
740	if (indexes)
741		of_pci_parse_iov_addrs(pdev, indexes);
742	else
743		pseries_disable_sriov_resources(pdev);
744}
745
746static resource_size_t pseries_pci_iov_resource_alignment(struct pci_dev *pdev,
747							  int resno)
748{
749	const __be32 *reg;
750	struct device_node *dn = pci_device_to_OF_node(pdev);
751
752	/*Firmware must support open sriov otherwise report regular alignment*/
753	reg = of_get_property(dn, "ibm,is-open-sriov-pf", NULL);
754	if (!reg)
755		return pci_iov_resource_size(pdev, resno);
756
757	if (!pdev->is_physfn)
758		return 0;
759	return pseries_get_iov_fw_value(pdev,
760					resno - PCI_IOV_RESOURCES,
761					APERTURE_SIZE);
762}
763#endif
764
765static void __init pSeries_setup_arch(void)
766{
767	set_arch_panic_timeout(10, ARCH_PANIC_TIMEOUT);
768
769	/* Discover PIC type and setup ppc_md accordingly */
770	smp_init_pseries();
771
772
773	if (radix_enabled() && !mmu_has_feature(MMU_FTR_GTSE))
774		if (!firmware_has_feature(FW_FEATURE_RPT_INVALIDATE))
775			panic("BUG: Radix support requires either GTSE or RPT_INVALIDATE\n");
776
777
778	/* openpic global configuration register (64-bit format). */
779	/* openpic Interrupt Source Unit pointer (64-bit format). */
780	/* python0 facility area (mmio) (64-bit format) REAL address. */
781
782	/* init to some ~sane value until calibrate_delay() runs */
783	loops_per_jiffy = 50000000;
784
785	fwnmi_init();
786
787	pseries_setup_security_mitigations();
788	pseries_lpar_read_hblkrm_characteristics();
789
790	/* By default, only probe PCI (can be overridden by rtas_pci) */
791	pci_add_flags(PCI_PROBE_ONLY);
792
793	/* Find and initialize PCI host bridges */
794	init_pci_config_tokens();
795	find_and_init_phbs();
796	of_reconfig_notifier_register(&pci_dn_reconfig_nb);
797
798	pSeries_nvram_init();
799
800	if (firmware_has_feature(FW_FEATURE_LPAR)) {
801		vpa_init(boot_cpuid);
802
803		if (lppaca_shared_proc()) {
804			static_branch_enable(&shared_processor);
805			pv_spinlocks_init();
806		}
807
808		ppc_md.power_save = pseries_lpar_idle;
809		ppc_md.enable_pmcs = pseries_lpar_enable_pmcs;
810#ifdef CONFIG_PCI_IOV
811		ppc_md.pcibios_fixup_resources =
812			pseries_pci_fixup_resources;
813		ppc_md.pcibios_fixup_sriov =
814			pseries_pci_fixup_iov_resources;
815		ppc_md.pcibios_iov_resource_alignment =
816			pseries_pci_iov_resource_alignment;
817#endif
818	} else {
819		/* No special idle routine */
820		ppc_md.enable_pmcs = power4_enable_pmcs;
821	}
822
823	ppc_md.pcibios_root_bridge_prepare = pseries_root_bridge_prepare;
824
825	if (swiotlb_force == SWIOTLB_FORCE)
826		ppc_swiotlb_enable = 1;
827
828	pseries_rng_init();
829}
830
831static void pseries_panic(char *str)
832{
833	panic_flush_kmsg_end();
834	rtas_os_term(str);
835}
836
837static int __init pSeries_init_panel(void)
838{
839	/* Manually leave the kernel version on the panel. */
840#ifdef __BIG_ENDIAN__
841	ppc_md.progress("Linux ppc64\n", 0);
842#else
843	ppc_md.progress("Linux ppc64le\n", 0);
844#endif
845	ppc_md.progress(init_utsname()->version, 0);
846
847	return 0;
848}
849machine_arch_initcall(pseries, pSeries_init_panel);
850
851static int pseries_set_dabr(unsigned long dabr, unsigned long dabrx)
852{
853	return plpar_hcall_norets(H_SET_DABR, dabr);
854}
855
856static int pseries_set_xdabr(unsigned long dabr, unsigned long dabrx)
857{
858	/* Have to set at least one bit in the DABRX according to PAPR */
859	if (dabrx == 0 && dabr == 0)
860		dabrx = DABRX_USER;
861	/* PAPR says we can only set kernel and user bits */
862	dabrx &= DABRX_KERNEL | DABRX_USER;
863
864	return plpar_hcall_norets(H_SET_XDABR, dabr, dabrx);
865}
866
867static int pseries_set_dawr(int nr, unsigned long dawr, unsigned long dawrx)
868{
869	/* PAPR says we can't set HYP */
870	dawrx &= ~DAWRX_HYP;
871
872	if (nr == 0)
873		return plpar_set_watchpoint0(dawr, dawrx);
874	else
875		return plpar_set_watchpoint1(dawr, dawrx);
876}
877
878#define CMO_CHARACTERISTICS_TOKEN 44
879#define CMO_MAXLENGTH 1026
880
881void pSeries_coalesce_init(void)
882{
883	struct hvcall_mpp_x_data mpp_x_data;
884
885	if (firmware_has_feature(FW_FEATURE_CMO) && !h_get_mpp_x(&mpp_x_data))
886		powerpc_firmware_features |= FW_FEATURE_XCMO;
887	else
888		powerpc_firmware_features &= ~FW_FEATURE_XCMO;
889}
890
891/**
892 * fw_cmo_feature_init - FW_FEATURE_CMO is not stored in ibm,hypertas-functions,
893 * handle that here. (Stolen from parse_system_parameter_string)
894 */
895static void pSeries_cmo_feature_init(void)
896{
897	char *ptr, *key, *value, *end;
898	int call_status;
899	int page_order = IOMMU_PAGE_SHIFT_4K;
900
901	pr_debug(" -> fw_cmo_feature_init()\n");
902	spin_lock(&rtas_data_buf_lock);
903	memset(rtas_data_buf, 0, RTAS_DATA_BUF_SIZE);
904	call_status = rtas_call(rtas_token("ibm,get-system-parameter"), 3, 1,
905				NULL,
906				CMO_CHARACTERISTICS_TOKEN,
907				__pa(rtas_data_buf),
908				RTAS_DATA_BUF_SIZE);
909
910	if (call_status != 0) {
911		spin_unlock(&rtas_data_buf_lock);
912		pr_debug("CMO not available\n");
913		pr_debug(" <- fw_cmo_feature_init()\n");
914		return;
915	}
916
917	end = rtas_data_buf + CMO_MAXLENGTH - 2;
918	ptr = rtas_data_buf + 2;	/* step over strlen value */
919	key = value = ptr;
920
921	while (*ptr && (ptr <= end)) {
922		/* Separate the key and value by replacing '=' with '\0' and
923		 * point the value at the string after the '='
924		 */
925		if (ptr[0] == '=') {
926			ptr[0] = '\0';
927			value = ptr + 1;
928		} else if (ptr[0] == '\0' || ptr[0] == ',') {
929			/* Terminate the string containing the key/value pair */
930			ptr[0] = '\0';
931
932			if (key == value) {
933				pr_debug("Malformed key/value pair\n");
934				/* Never found a '=', end processing */
935				break;
936			}
937
938			if (0 == strcmp(key, "CMOPageSize"))
939				page_order = simple_strtol(value, NULL, 10);
940			else if (0 == strcmp(key, "PrPSP"))
941				CMO_PrPSP = simple_strtol(value, NULL, 10);
942			else if (0 == strcmp(key, "SecPSP"))
943				CMO_SecPSP = simple_strtol(value, NULL, 10);
944			value = key = ptr + 1;
945		}
946		ptr++;
947	}
948
949	/* Page size is returned as the power of 2 of the page size,
950	 * convert to the page size in bytes before returning
951	 */
952	CMO_PageSize = 1 << page_order;
953	pr_debug("CMO_PageSize = %lu\n", CMO_PageSize);
954
955	if (CMO_PrPSP != -1 || CMO_SecPSP != -1) {
956		pr_info("CMO enabled\n");
957		pr_debug("CMO enabled, PrPSP=%d, SecPSP=%d\n", CMO_PrPSP,
958		         CMO_SecPSP);
959		powerpc_firmware_features |= FW_FEATURE_CMO;
960		pSeries_coalesce_init();
961	} else
962		pr_debug("CMO not enabled, PrPSP=%d, SecPSP=%d\n", CMO_PrPSP,
963		         CMO_SecPSP);
964	spin_unlock(&rtas_data_buf_lock);
965	pr_debug(" <- fw_cmo_feature_init()\n");
966}
967
968/*
969 * Early initialization.  Relocation is on but do not reference unbolted pages
970 */
971static void __init pseries_init(void)
972{
973	pr_debug(" -> pseries_init()\n");
974
975#ifdef CONFIG_HVC_CONSOLE
976	if (firmware_has_feature(FW_FEATURE_LPAR))
977		hvc_vio_init_early();
978#endif
979	if (firmware_has_feature(FW_FEATURE_XDABR))
980		ppc_md.set_dabr = pseries_set_xdabr;
981	else if (firmware_has_feature(FW_FEATURE_DABR))
982		ppc_md.set_dabr = pseries_set_dabr;
983
984	if (firmware_has_feature(FW_FEATURE_SET_MODE))
985		ppc_md.set_dawr = pseries_set_dawr;
986
987	pSeries_cmo_feature_init();
988	iommu_init_early_pSeries();
989
990	pr_debug(" <- pseries_init()\n");
991}
992
993/**
994 * pseries_power_off - tell firmware about how to power off the system.
995 *
996 * This function calls either the power-off rtas token in normal cases
997 * or the ibm,power-off-ups token (if present & requested) in case of
998 * a power failure. If power-off token is used, power on will only be
999 * possible with power button press. If ibm,power-off-ups token is used
1000 * it will allow auto poweron after power is restored.
1001 */
1002static void pseries_power_off(void)
1003{
1004	int rc;
1005	int rtas_poweroff_ups_token = rtas_token("ibm,power-off-ups");
1006
1007	if (rtas_flash_term_hook)
1008		rtas_flash_term_hook(SYS_POWER_OFF);
1009
1010	if (rtas_poweron_auto == 0 ||
1011		rtas_poweroff_ups_token == RTAS_UNKNOWN_SERVICE) {
1012		rc = rtas_call(rtas_token("power-off"), 2, 1, NULL, -1, -1);
1013		printk(KERN_INFO "RTAS power-off returned %d\n", rc);
1014	} else {
1015		rc = rtas_call(rtas_poweroff_ups_token, 0, 1, NULL);
1016		printk(KERN_INFO "RTAS ibm,power-off-ups returned %d\n", rc);
1017	}
1018	for (;;);
1019}
1020
1021static int __init pSeries_probe(void)
1022{
1023	if (!of_node_is_type(of_root, "chrp"))
1024		return 0;
1025
1026	/* Cell blades firmware claims to be chrp while it's not. Until this
1027	 * is fixed, we need to avoid those here.
1028	 */
1029	if (of_machine_is_compatible("IBM,CPBW-1.0") ||
1030	    of_machine_is_compatible("IBM,CBEA"))
1031		return 0;
1032
1033	pm_power_off = pseries_power_off;
1034
1035	pr_debug("Machine is%s LPAR !\n",
1036	         (powerpc_firmware_features & FW_FEATURE_LPAR) ? "" : " not");
1037
1038	pseries_init();
1039
1040	return 1;
1041}
1042
1043static int pSeries_pci_probe_mode(struct pci_bus *bus)
1044{
1045	if (firmware_has_feature(FW_FEATURE_LPAR))
1046		return PCI_PROBE_DEVTREE;
1047	return PCI_PROBE_NORMAL;
1048}
1049
1050struct pci_controller_ops pseries_pci_controller_ops = {
1051	.probe_mode		= pSeries_pci_probe_mode,
1052};
1053
1054define_machine(pseries) {
1055	.name			= "pSeries",
1056	.probe			= pSeries_probe,
1057	.setup_arch		= pSeries_setup_arch,
1058	.init_IRQ		= pseries_init_irq,
1059	.show_cpuinfo		= pSeries_show_cpuinfo,
1060	.log_error		= pSeries_log_error,
1061	.pcibios_fixup		= pSeries_final_fixup,
1062	.restart		= rtas_restart,
1063	.halt			= rtas_halt,
1064	.panic			= pseries_panic,
1065	.get_boot_time		= rtas_get_boot_time,
1066	.get_rtc_time		= rtas_get_rtc_time,
1067	.set_rtc_time		= rtas_set_rtc_time,
1068	.calibrate_decr		= generic_calibrate_decr,
1069	.progress		= rtas_progress,
1070	.system_reset_exception = pSeries_system_reset_exception,
1071	.machine_check_early	= pseries_machine_check_realmode,
1072	.machine_check_exception = pSeries_machine_check_exception,
1073#ifdef CONFIG_KEXEC_CORE
1074	.machine_kexec          = pSeries_machine_kexec,
1075	.kexec_cpu_down         = pseries_kexec_cpu_down,
1076#endif
1077#ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
1078	.memory_block_size	= pseries_memory_block_size,
1079#endif
1080};
1081