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 #include <asm/hvconsole.h>
75 
76 #include "pseries.h"
77 
78 DEFINE_STATIC_KEY_FALSE(shared_processor);
79 EXPORT_SYMBOL(shared_processor);
80 
81 int CMO_PrPSP = -1;
82 int CMO_SecPSP = -1;
83 unsigned long CMO_PageSize = (ASM_CONST(1) << IOMMU_PAGE_SHIFT_4K);
84 EXPORT_SYMBOL(CMO_PageSize);
85 
86 int fwnmi_active;  /* TRUE if an FWNMI handler is present */
87 int ibm_nmi_interlock_token;
88 u32 pseries_security_flavor;
89 
90 static void pSeries_show_cpuinfo(struct seq_file *m)
91 {
92 	struct device_node *root;
93 	const char *model = "";
94 
95 	root = of_find_node_by_path("/");
96 	if (root)
97 		model = of_get_property(root, "model", NULL);
98 	seq_printf(m, "machine\t\t: CHRP %s\n", model);
99 	of_node_put(root);
100 	if (radix_enabled())
101 		seq_printf(m, "MMU\t\t: Radix\n");
102 	else
103 		seq_printf(m, "MMU\t\t: Hash\n");
104 }
105 
106 /* Initialize firmware assisted non-maskable interrupts if
107  * the firmware supports this feature.
108  */
109 static void __init fwnmi_init(void)
110 {
111 	unsigned long system_reset_addr, machine_check_addr;
112 	u8 *mce_data_buf;
113 	unsigned int i;
114 	int nr_cpus = num_possible_cpus();
115 #ifdef CONFIG_PPC_64S_HASH_MMU
116 	struct slb_entry *slb_ptr;
117 	size_t size;
118 #endif
119 	int ibm_nmi_register_token;
120 
121 	ibm_nmi_register_token = rtas_token("ibm,nmi-register");
122 	if (ibm_nmi_register_token == RTAS_UNKNOWN_SERVICE)
123 		return;
124 
125 	ibm_nmi_interlock_token = rtas_token("ibm,nmi-interlock");
126 	if (WARN_ON(ibm_nmi_interlock_token == RTAS_UNKNOWN_SERVICE))
127 		return;
128 
129 	/* If the kernel's not linked at zero we point the firmware at low
130 	 * addresses anyway, and use a trampoline to get to the real code. */
131 	system_reset_addr  = __pa(system_reset_fwnmi) - PHYSICAL_START;
132 	machine_check_addr = __pa(machine_check_fwnmi) - PHYSICAL_START;
133 
134 	if (0 == rtas_call(ibm_nmi_register_token, 2, 1, NULL,
135 			   system_reset_addr, machine_check_addr))
136 		fwnmi_active = 1;
137 
138 	/*
139 	 * Allocate a chunk for per cpu buffer to hold rtas errorlog.
140 	 * It will be used in real mode mce handler, hence it needs to be
141 	 * below RMA.
142 	 */
143 	mce_data_buf = memblock_alloc_try_nid_raw(RTAS_ERROR_LOG_MAX * nr_cpus,
144 					RTAS_ERROR_LOG_MAX, MEMBLOCK_LOW_LIMIT,
145 					ppc64_rma_size, NUMA_NO_NODE);
146 	if (!mce_data_buf)
147 		panic("Failed to allocate %d bytes below %pa for MCE buffer\n",
148 		      RTAS_ERROR_LOG_MAX * nr_cpus, &ppc64_rma_size);
149 
150 	for_each_possible_cpu(i) {
151 		paca_ptrs[i]->mce_data_buf = mce_data_buf +
152 						(RTAS_ERROR_LOG_MAX * i);
153 	}
154 
155 #ifdef CONFIG_PPC_64S_HASH_MMU
156 	if (!radix_enabled()) {
157 		/* Allocate per cpu area to save old slb contents during MCE */
158 		size = sizeof(struct slb_entry) * mmu_slb_size * nr_cpus;
159 		slb_ptr = memblock_alloc_try_nid_raw(size,
160 				sizeof(struct slb_entry), MEMBLOCK_LOW_LIMIT,
161 				ppc64_rma_size, NUMA_NO_NODE);
162 		if (!slb_ptr)
163 			panic("Failed to allocate %zu bytes below %pa for slb area\n",
164 			      size, &ppc64_rma_size);
165 
166 		for_each_possible_cpu(i)
167 			paca_ptrs[i]->mce_faulty_slbs = slb_ptr + (mmu_slb_size * i);
168 	}
169 #endif
170 }
171 
172 static void pseries_8259_cascade(struct irq_desc *desc)
173 {
174 	struct irq_chip *chip = irq_desc_get_chip(desc);
175 	unsigned int cascade_irq = i8259_irq();
176 
177 	if (cascade_irq)
178 		generic_handle_irq(cascade_irq);
179 
180 	chip->irq_eoi(&desc->irq_data);
181 }
182 
183 static void __init pseries_setup_i8259_cascade(void)
184 {
185 	struct device_node *np, *old, *found = NULL;
186 	unsigned int cascade;
187 	const u32 *addrp;
188 	unsigned long intack = 0;
189 	int naddr;
190 
191 	for_each_node_by_type(np, "interrupt-controller") {
192 		if (of_device_is_compatible(np, "chrp,iic")) {
193 			found = np;
194 			break;
195 		}
196 	}
197 
198 	if (found == NULL) {
199 		printk(KERN_DEBUG "pic: no ISA interrupt controller\n");
200 		return;
201 	}
202 
203 	cascade = irq_of_parse_and_map(found, 0);
204 	if (!cascade) {
205 		printk(KERN_ERR "pic: failed to map cascade interrupt");
206 		return;
207 	}
208 	pr_debug("pic: cascade mapped to irq %d\n", cascade);
209 
210 	for (old = of_node_get(found); old != NULL ; old = np) {
211 		np = of_get_parent(old);
212 		of_node_put(old);
213 		if (np == NULL)
214 			break;
215 		if (!of_node_name_eq(np, "pci"))
216 			continue;
217 		addrp = of_get_property(np, "8259-interrupt-acknowledge", NULL);
218 		if (addrp == NULL)
219 			continue;
220 		naddr = of_n_addr_cells(np);
221 		intack = addrp[naddr-1];
222 		if (naddr > 1)
223 			intack |= ((unsigned long)addrp[naddr-2]) << 32;
224 	}
225 	if (intack)
226 		printk(KERN_DEBUG "pic: PCI 8259 intack at 0x%016lx\n", intack);
227 	i8259_init(found, intack);
228 	of_node_put(found);
229 	irq_set_chained_handler(cascade, pseries_8259_cascade);
230 }
231 
232 static void __init pseries_init_irq(void)
233 {
234 	/* Try using a XIVE if available, otherwise use a XICS */
235 	if (!xive_spapr_init()) {
236 		xics_init();
237 		pseries_setup_i8259_cascade();
238 	}
239 }
240 
241 static void pseries_lpar_enable_pmcs(void)
242 {
243 	unsigned long set, reset;
244 
245 	set = 1UL << 63;
246 	reset = 0;
247 	plpar_hcall_norets(H_PERFMON, set, reset);
248 }
249 
250 static int pci_dn_reconfig_notifier(struct notifier_block *nb, unsigned long action, void *data)
251 {
252 	struct of_reconfig_data *rd = data;
253 	struct device_node *parent, *np = rd->dn;
254 	struct pci_dn *pdn;
255 	int err = NOTIFY_OK;
256 
257 	switch (action) {
258 	case OF_RECONFIG_ATTACH_NODE:
259 		parent = of_get_parent(np);
260 		pdn = parent ? PCI_DN(parent) : NULL;
261 		if (pdn)
262 			pci_add_device_node_info(pdn->phb, np);
263 
264 		of_node_put(parent);
265 		break;
266 	case OF_RECONFIG_DETACH_NODE:
267 		pdn = PCI_DN(np);
268 		if (pdn)
269 			list_del(&pdn->list);
270 		break;
271 	default:
272 		err = NOTIFY_DONE;
273 		break;
274 	}
275 	return err;
276 }
277 
278 static struct notifier_block pci_dn_reconfig_nb = {
279 	.notifier_call = pci_dn_reconfig_notifier,
280 };
281 
282 struct kmem_cache *dtl_cache;
283 
284 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
285 /*
286  * Allocate space for the dispatch trace log for all possible cpus
287  * and register the buffers with the hypervisor.  This is used for
288  * computing time stolen by the hypervisor.
289  */
290 static int alloc_dispatch_logs(void)
291 {
292 	if (!firmware_has_feature(FW_FEATURE_SPLPAR))
293 		return 0;
294 
295 	if (!dtl_cache)
296 		return 0;
297 
298 	alloc_dtl_buffers(0);
299 
300 	/* Register the DTL for the current (boot) cpu */
301 	register_dtl_buffer(smp_processor_id());
302 
303 	return 0;
304 }
305 #else /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
306 static inline int alloc_dispatch_logs(void)
307 {
308 	return 0;
309 }
310 #endif /* CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
311 
312 static int alloc_dispatch_log_kmem_cache(void)
313 {
314 	void (*ctor)(void *) = get_dtl_cache_ctor();
315 
316 	dtl_cache = kmem_cache_create("dtl", DISPATCH_LOG_BYTES,
317 						DISPATCH_LOG_BYTES, 0, ctor);
318 	if (!dtl_cache) {
319 		pr_warn("Failed to create dispatch trace log buffer cache\n");
320 		pr_warn("Stolen time statistics will be unreliable\n");
321 		return 0;
322 	}
323 
324 	return alloc_dispatch_logs();
325 }
326 machine_early_initcall(pseries, alloc_dispatch_log_kmem_cache);
327 
328 DEFINE_PER_CPU(u64, idle_spurr_cycles);
329 DEFINE_PER_CPU(u64, idle_entry_purr_snap);
330 DEFINE_PER_CPU(u64, idle_entry_spurr_snap);
331 static void pseries_lpar_idle(void)
332 {
333 	/*
334 	 * Default handler to go into low thread priority and possibly
335 	 * low power mode by ceding processor to hypervisor
336 	 */
337 
338 	if (!prep_irq_for_idle())
339 		return;
340 
341 	/* Indicate to hypervisor that we are idle. */
342 	pseries_idle_prolog();
343 
344 	/*
345 	 * Yield the processor to the hypervisor.  We return if
346 	 * an external interrupt occurs (which are driven prior
347 	 * to returning here) or if a prod occurs from another
348 	 * processor. When returning here, external interrupts
349 	 * are enabled.
350 	 */
351 	cede_processor();
352 
353 	pseries_idle_epilog();
354 }
355 
356 /*
357  * Enable relocation on during exceptions. This has partition wide scope and
358  * may take a while to complete, if it takes longer than one second we will
359  * just give up rather than wasting any more time on this - if that turns out
360  * to ever be a problem in practice we can move this into a kernel thread to
361  * finish off the process later in boot.
362  */
363 bool pseries_enable_reloc_on_exc(void)
364 {
365 	long rc;
366 	unsigned int delay, total_delay = 0;
367 
368 	while (1) {
369 		rc = enable_reloc_on_exceptions();
370 		if (!H_IS_LONG_BUSY(rc)) {
371 			if (rc == H_P2) {
372 				pr_info("Relocation on exceptions not"
373 					" supported\n");
374 				return false;
375 			} else if (rc != H_SUCCESS) {
376 				pr_warn("Unable to enable relocation"
377 					" on exceptions: %ld\n", rc);
378 				return false;
379 			}
380 			return true;
381 		}
382 
383 		delay = get_longbusy_msecs(rc);
384 		total_delay += delay;
385 		if (total_delay > 1000) {
386 			pr_warn("Warning: Giving up waiting to enable "
387 				"relocation on exceptions (%u msec)!\n",
388 				total_delay);
389 			return false;
390 		}
391 
392 		mdelay(delay);
393 	}
394 }
395 EXPORT_SYMBOL(pseries_enable_reloc_on_exc);
396 
397 void pseries_disable_reloc_on_exc(void)
398 {
399 	long rc;
400 
401 	while (1) {
402 		rc = disable_reloc_on_exceptions();
403 		if (!H_IS_LONG_BUSY(rc))
404 			break;
405 		mdelay(get_longbusy_msecs(rc));
406 	}
407 	if (rc != H_SUCCESS)
408 		pr_warn("Warning: Failed to disable relocation on exceptions: %ld\n",
409 			rc);
410 }
411 EXPORT_SYMBOL(pseries_disable_reloc_on_exc);
412 
413 #ifdef CONFIG_KEXEC_CORE
414 static void pSeries_machine_kexec(struct kimage *image)
415 {
416 	if (firmware_has_feature(FW_FEATURE_SET_MODE))
417 		pseries_disable_reloc_on_exc();
418 
419 	default_machine_kexec(image);
420 }
421 #endif
422 
423 #ifdef __LITTLE_ENDIAN__
424 void pseries_big_endian_exceptions(void)
425 {
426 	long rc;
427 
428 	while (1) {
429 		rc = enable_big_endian_exceptions();
430 		if (!H_IS_LONG_BUSY(rc))
431 			break;
432 		mdelay(get_longbusy_msecs(rc));
433 	}
434 
435 	/*
436 	 * At this point it is unlikely panic() will get anything
437 	 * out to the user, since this is called very late in kexec
438 	 * but at least this will stop us from continuing on further
439 	 * and creating an even more difficult to debug situation.
440 	 *
441 	 * There is a known problem when kdump'ing, if cpus are offline
442 	 * the above call will fail. Rather than panicking again, keep
443 	 * going and hope the kdump kernel is also little endian, which
444 	 * it usually is.
445 	 */
446 	if (rc && !kdump_in_progress())
447 		panic("Could not enable big endian exceptions");
448 }
449 
450 void __init pseries_little_endian_exceptions(void)
451 {
452 	long rc;
453 
454 	while (1) {
455 		rc = enable_little_endian_exceptions();
456 		if (!H_IS_LONG_BUSY(rc))
457 			break;
458 		mdelay(get_longbusy_msecs(rc));
459 	}
460 	if (rc) {
461 		ppc_md.progress("H_SET_MODE LE exception fail", 0);
462 		panic("Could not enable little endian exceptions");
463 	}
464 }
465 #endif
466 
467 static void __init pSeries_discover_phbs(void)
468 {
469 	struct device_node *node;
470 	struct pci_controller *phb;
471 	struct device_node *root = of_find_node_by_path("/");
472 
473 	for_each_child_of_node(root, node) {
474 		if (!of_node_is_type(node, "pci") &&
475 		    !of_node_is_type(node, "pciex"))
476 			continue;
477 
478 		phb = pcibios_alloc_controller(node);
479 		if (!phb)
480 			continue;
481 		rtas_setup_phb(phb);
482 		pci_process_bridge_OF_ranges(phb, node, 0);
483 		isa_bridge_find_early(phb);
484 		phb->controller_ops = pseries_pci_controller_ops;
485 
486 		/* create pci_dn's for DT nodes under this PHB */
487 		pci_devs_phb_init_dynamic(phb);
488 
489 		pseries_msi_allocate_domains(phb);
490 	}
491 
492 	of_node_put(root);
493 
494 	/*
495 	 * PCI_PROBE_ONLY and PCI_REASSIGN_ALL_BUS can be set via properties
496 	 * in chosen.
497 	 */
498 	of_pci_check_probe_only();
499 }
500 
501 static void init_cpu_char_feature_flags(struct h_cpu_char_result *result)
502 {
503 	/*
504 	 * The features below are disabled by default, so we instead look to see
505 	 * if firmware has *enabled* them, and set them if so.
506 	 */
507 	if (result->character & H_CPU_CHAR_SPEC_BAR_ORI31)
508 		security_ftr_set(SEC_FTR_SPEC_BAR_ORI31);
509 
510 	if (result->character & H_CPU_CHAR_BCCTRL_SERIALISED)
511 		security_ftr_set(SEC_FTR_BCCTRL_SERIALISED);
512 
513 	if (result->character & H_CPU_CHAR_L1D_FLUSH_ORI30)
514 		security_ftr_set(SEC_FTR_L1D_FLUSH_ORI30);
515 
516 	if (result->character & H_CPU_CHAR_L1D_FLUSH_TRIG2)
517 		security_ftr_set(SEC_FTR_L1D_FLUSH_TRIG2);
518 
519 	if (result->character & H_CPU_CHAR_L1D_THREAD_PRIV)
520 		security_ftr_set(SEC_FTR_L1D_THREAD_PRIV);
521 
522 	if (result->character & H_CPU_CHAR_COUNT_CACHE_DISABLED)
523 		security_ftr_set(SEC_FTR_COUNT_CACHE_DISABLED);
524 
525 	if (result->character & H_CPU_CHAR_BCCTR_FLUSH_ASSIST)
526 		security_ftr_set(SEC_FTR_BCCTR_FLUSH_ASSIST);
527 
528 	if (result->character & H_CPU_CHAR_BCCTR_LINK_FLUSH_ASSIST)
529 		security_ftr_set(SEC_FTR_BCCTR_LINK_FLUSH_ASSIST);
530 
531 	if (result->behaviour & H_CPU_BEHAV_FLUSH_COUNT_CACHE)
532 		security_ftr_set(SEC_FTR_FLUSH_COUNT_CACHE);
533 
534 	if (result->behaviour & H_CPU_BEHAV_FLUSH_LINK_STACK)
535 		security_ftr_set(SEC_FTR_FLUSH_LINK_STACK);
536 
537 	/*
538 	 * The features below are enabled by default, so we instead look to see
539 	 * if firmware has *disabled* them, and clear them if so.
540 	 * H_CPU_BEHAV_FAVOUR_SECURITY_H could be set only if
541 	 * H_CPU_BEHAV_FAVOUR_SECURITY is.
542 	 */
543 	if (!(result->behaviour & H_CPU_BEHAV_FAVOUR_SECURITY)) {
544 		security_ftr_clear(SEC_FTR_FAVOUR_SECURITY);
545 		pseries_security_flavor = 0;
546 	} else if (result->behaviour & H_CPU_BEHAV_FAVOUR_SECURITY_H)
547 		pseries_security_flavor = 1;
548 	else
549 		pseries_security_flavor = 2;
550 
551 	if (!(result->behaviour & H_CPU_BEHAV_L1D_FLUSH_PR))
552 		security_ftr_clear(SEC_FTR_L1D_FLUSH_PR);
553 
554 	if (result->behaviour & H_CPU_BEHAV_NO_L1D_FLUSH_ENTRY)
555 		security_ftr_clear(SEC_FTR_L1D_FLUSH_ENTRY);
556 
557 	if (result->behaviour & H_CPU_BEHAV_NO_L1D_FLUSH_UACCESS)
558 		security_ftr_clear(SEC_FTR_L1D_FLUSH_UACCESS);
559 
560 	if (result->behaviour & H_CPU_BEHAV_NO_STF_BARRIER)
561 		security_ftr_clear(SEC_FTR_STF_BARRIER);
562 
563 	if (!(result->behaviour & H_CPU_BEHAV_BNDS_CHK_SPEC_BAR))
564 		security_ftr_clear(SEC_FTR_BNDS_CHK_SPEC_BAR);
565 }
566 
567 void pseries_setup_security_mitigations(void)
568 {
569 	struct h_cpu_char_result result;
570 	enum l1d_flush_type types;
571 	bool enable;
572 	long rc;
573 
574 	/*
575 	 * Set features to the defaults assumed by init_cpu_char_feature_flags()
576 	 * so it can set/clear again any features that might have changed after
577 	 * migration, and in case the hypercall fails and it is not even called.
578 	 */
579 	powerpc_security_features = SEC_FTR_DEFAULT;
580 
581 	rc = plpar_get_cpu_characteristics(&result);
582 	if (rc == H_SUCCESS)
583 		init_cpu_char_feature_flags(&result);
584 
585 	/*
586 	 * We're the guest so this doesn't apply to us, clear it to simplify
587 	 * handling of it elsewhere.
588 	 */
589 	security_ftr_clear(SEC_FTR_L1D_FLUSH_HV);
590 
591 	types = L1D_FLUSH_FALLBACK;
592 
593 	if (security_ftr_enabled(SEC_FTR_L1D_FLUSH_TRIG2))
594 		types |= L1D_FLUSH_MTTRIG;
595 
596 	if (security_ftr_enabled(SEC_FTR_L1D_FLUSH_ORI30))
597 		types |= L1D_FLUSH_ORI;
598 
599 	enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) && \
600 		 security_ftr_enabled(SEC_FTR_L1D_FLUSH_PR);
601 
602 	setup_rfi_flush(types, enable);
603 	setup_count_cache_flush();
604 
605 	enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) &&
606 		 security_ftr_enabled(SEC_FTR_L1D_FLUSH_ENTRY);
607 	setup_entry_flush(enable);
608 
609 	enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) &&
610 		 security_ftr_enabled(SEC_FTR_L1D_FLUSH_UACCESS);
611 	setup_uaccess_flush(enable);
612 
613 	setup_stf_barrier();
614 }
615 
616 #ifdef CONFIG_PCI_IOV
617 enum rtas_iov_fw_value_map {
618 	NUM_RES_PROPERTY  = 0, /* Number of Resources */
619 	LOW_INT           = 1, /* Lowest 32 bits of Address */
620 	START_OF_ENTRIES  = 2, /* Always start of entry */
621 	APERTURE_PROPERTY = 2, /* Start of entry+ to  Aperture Size */
622 	WDW_SIZE_PROPERTY = 4, /* Start of entry+ to Window Size */
623 	NEXT_ENTRY        = 7  /* Go to next entry on array */
624 };
625 
626 enum get_iov_fw_value_index {
627 	BAR_ADDRS     = 1,    /*  Get Bar Address */
628 	APERTURE_SIZE = 2,    /*  Get Aperture Size */
629 	WDW_SIZE      = 3     /*  Get Window Size */
630 };
631 
632 static resource_size_t pseries_get_iov_fw_value(struct pci_dev *dev, int resno,
633 						enum get_iov_fw_value_index value)
634 {
635 	const int *indexes;
636 	struct device_node *dn = pci_device_to_OF_node(dev);
637 	int i, num_res, ret = 0;
638 
639 	indexes = of_get_property(dn, "ibm,open-sriov-vf-bar-info", NULL);
640 	if (!indexes)
641 		return  0;
642 
643 	/*
644 	 * First element in the array is the number of Bars
645 	 * returned.  Search through the list to find the matching
646 	 * bar
647 	 */
648 	num_res = of_read_number(&indexes[NUM_RES_PROPERTY], 1);
649 	if (resno >= num_res)
650 		return 0; /* or an errror */
651 
652 	i = START_OF_ENTRIES + NEXT_ENTRY * resno;
653 	switch (value) {
654 	case BAR_ADDRS:
655 		ret = of_read_number(&indexes[i], 2);
656 		break;
657 	case APERTURE_SIZE:
658 		ret = of_read_number(&indexes[i + APERTURE_PROPERTY], 2);
659 		break;
660 	case WDW_SIZE:
661 		ret = of_read_number(&indexes[i + WDW_SIZE_PROPERTY], 2);
662 		break;
663 	}
664 
665 	return ret;
666 }
667 
668 static void of_pci_set_vf_bar_size(struct pci_dev *dev, const int *indexes)
669 {
670 	struct resource *res;
671 	resource_size_t base, size;
672 	int i, r, num_res;
673 
674 	num_res = of_read_number(&indexes[NUM_RES_PROPERTY], 1);
675 	num_res = min_t(int, num_res, PCI_SRIOV_NUM_BARS);
676 	for (i = START_OF_ENTRIES, r = 0; r < num_res && r < PCI_SRIOV_NUM_BARS;
677 	     i += NEXT_ENTRY, r++) {
678 		res = &dev->resource[r + PCI_IOV_RESOURCES];
679 		base = of_read_number(&indexes[i], 2);
680 		size = of_read_number(&indexes[i + APERTURE_PROPERTY], 2);
681 		res->flags = pci_parse_of_flags(of_read_number
682 						(&indexes[i + LOW_INT], 1), 0);
683 		res->flags |= (IORESOURCE_MEM_64 | IORESOURCE_PCI_FIXED);
684 		res->name = pci_name(dev);
685 		res->start = base;
686 		res->end = base + size - 1;
687 	}
688 }
689 
690 static void of_pci_parse_iov_addrs(struct pci_dev *dev, const int *indexes)
691 {
692 	struct resource *res, *root, *conflict;
693 	resource_size_t base, size;
694 	int i, r, num_res;
695 
696 	/*
697 	 * First element in the array is the number of Bars
698 	 * returned.  Search through the list to find the matching
699 	 * bars assign them from firmware into resources structure.
700 	 */
701 	num_res = of_read_number(&indexes[NUM_RES_PROPERTY], 1);
702 	for (i = START_OF_ENTRIES, r = 0; r < num_res && r < PCI_SRIOV_NUM_BARS;
703 	     i += NEXT_ENTRY, r++) {
704 		res = &dev->resource[r + PCI_IOV_RESOURCES];
705 		base = of_read_number(&indexes[i], 2);
706 		size = of_read_number(&indexes[i + WDW_SIZE_PROPERTY], 2);
707 		res->name = pci_name(dev);
708 		res->start = base;
709 		res->end = base + size - 1;
710 		root = &iomem_resource;
711 		dev_dbg(&dev->dev,
712 			"pSeries IOV BAR %d: trying firmware assignment %pR\n",
713 			 r + PCI_IOV_RESOURCES, res);
714 		conflict = request_resource_conflict(root, res);
715 		if (conflict) {
716 			dev_info(&dev->dev,
717 				 "BAR %d: %pR conflicts with %s %pR\n",
718 				 r + PCI_IOV_RESOURCES, res,
719 				 conflict->name, conflict);
720 			res->flags |= IORESOURCE_UNSET;
721 		}
722 	}
723 }
724 
725 static void pseries_disable_sriov_resources(struct pci_dev *pdev)
726 {
727 	int i;
728 
729 	pci_warn(pdev, "No hypervisor support for SR-IOV on this device, IOV BARs disabled.\n");
730 	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++)
731 		pdev->resource[i + PCI_IOV_RESOURCES].flags = 0;
732 }
733 
734 static void pseries_pci_fixup_resources(struct pci_dev *pdev)
735 {
736 	const int *indexes;
737 	struct device_node *dn = pci_device_to_OF_node(pdev);
738 
739 	/*Firmware must support open sriov otherwise dont configure*/
740 	indexes = of_get_property(dn, "ibm,open-sriov-vf-bar-info", NULL);
741 	if (indexes)
742 		of_pci_set_vf_bar_size(pdev, indexes);
743 	else
744 		pseries_disable_sriov_resources(pdev);
745 }
746 
747 static void pseries_pci_fixup_iov_resources(struct pci_dev *pdev)
748 {
749 	const int *indexes;
750 	struct device_node *dn = pci_device_to_OF_node(pdev);
751 
752 	if (!pdev->is_physfn)
753 		return;
754 	/*Firmware must support open sriov otherwise dont configure*/
755 	indexes = of_get_property(dn, "ibm,open-sriov-vf-bar-info", NULL);
756 	if (indexes)
757 		of_pci_parse_iov_addrs(pdev, indexes);
758 	else
759 		pseries_disable_sriov_resources(pdev);
760 }
761 
762 static resource_size_t pseries_pci_iov_resource_alignment(struct pci_dev *pdev,
763 							  int resno)
764 {
765 	const __be32 *reg;
766 	struct device_node *dn = pci_device_to_OF_node(pdev);
767 
768 	/*Firmware must support open sriov otherwise report regular alignment*/
769 	reg = of_get_property(dn, "ibm,is-open-sriov-pf", NULL);
770 	if (!reg)
771 		return pci_iov_resource_size(pdev, resno);
772 
773 	if (!pdev->is_physfn)
774 		return 0;
775 	return pseries_get_iov_fw_value(pdev,
776 					resno - PCI_IOV_RESOURCES,
777 					APERTURE_SIZE);
778 }
779 #endif
780 
781 static void __init pSeries_setup_arch(void)
782 {
783 	set_arch_panic_timeout(10, ARCH_PANIC_TIMEOUT);
784 
785 	/* Discover PIC type and setup ppc_md accordingly */
786 	smp_init_pseries();
787 
788 
789 	if (radix_enabled() && !mmu_has_feature(MMU_FTR_GTSE))
790 		if (!firmware_has_feature(FW_FEATURE_RPT_INVALIDATE))
791 			panic("BUG: Radix support requires either GTSE or RPT_INVALIDATE\n");
792 
793 
794 	/* openpic global configuration register (64-bit format). */
795 	/* openpic Interrupt Source Unit pointer (64-bit format). */
796 	/* python0 facility area (mmio) (64-bit format) REAL address. */
797 
798 	/* init to some ~sane value until calibrate_delay() runs */
799 	loops_per_jiffy = 50000000;
800 
801 	fwnmi_init();
802 
803 	pseries_setup_security_mitigations();
804 #ifdef CONFIG_PPC_64S_HASH_MMU
805 	pseries_lpar_read_hblkrm_characteristics();
806 #endif
807 
808 	/* By default, only probe PCI (can be overridden by rtas_pci) */
809 	pci_add_flags(PCI_PROBE_ONLY);
810 
811 	/* Find and initialize PCI host bridges */
812 	init_pci_config_tokens();
813 	of_reconfig_notifier_register(&pci_dn_reconfig_nb);
814 
815 	pSeries_nvram_init();
816 
817 	if (firmware_has_feature(FW_FEATURE_LPAR)) {
818 		vpa_init(boot_cpuid);
819 
820 		if (lppaca_shared_proc(get_lppaca())) {
821 			static_branch_enable(&shared_processor);
822 			pv_spinlocks_init();
823 		}
824 
825 		ppc_md.power_save = pseries_lpar_idle;
826 		ppc_md.enable_pmcs = pseries_lpar_enable_pmcs;
827 #ifdef CONFIG_PCI_IOV
828 		ppc_md.pcibios_fixup_resources =
829 			pseries_pci_fixup_resources;
830 		ppc_md.pcibios_fixup_sriov =
831 			pseries_pci_fixup_iov_resources;
832 		ppc_md.pcibios_iov_resource_alignment =
833 			pseries_pci_iov_resource_alignment;
834 #endif
835 	} else {
836 		/* No special idle routine */
837 		ppc_md.enable_pmcs = power4_enable_pmcs;
838 	}
839 
840 	ppc_md.pcibios_root_bridge_prepare = pseries_root_bridge_prepare;
841 
842 	if (swiotlb_force == SWIOTLB_FORCE)
843 		ppc_swiotlb_enable = 1;
844 }
845 
846 static void pseries_panic(char *str)
847 {
848 	panic_flush_kmsg_end();
849 	rtas_os_term(str);
850 }
851 
852 static int __init pSeries_init_panel(void)
853 {
854 	/* Manually leave the kernel version on the panel. */
855 #ifdef __BIG_ENDIAN__
856 	ppc_md.progress("Linux ppc64\n", 0);
857 #else
858 	ppc_md.progress("Linux ppc64le\n", 0);
859 #endif
860 	ppc_md.progress(init_utsname()->version, 0);
861 
862 	return 0;
863 }
864 machine_arch_initcall(pseries, pSeries_init_panel);
865 
866 static int pseries_set_dabr(unsigned long dabr, unsigned long dabrx)
867 {
868 	return plpar_hcall_norets(H_SET_DABR, dabr);
869 }
870 
871 static int pseries_set_xdabr(unsigned long dabr, unsigned long dabrx)
872 {
873 	/* Have to set at least one bit in the DABRX according to PAPR */
874 	if (dabrx == 0 && dabr == 0)
875 		dabrx = DABRX_USER;
876 	/* PAPR says we can only set kernel and user bits */
877 	dabrx &= DABRX_KERNEL | DABRX_USER;
878 
879 	return plpar_hcall_norets(H_SET_XDABR, dabr, dabrx);
880 }
881 
882 static int pseries_set_dawr(int nr, unsigned long dawr, unsigned long dawrx)
883 {
884 	/* PAPR says we can't set HYP */
885 	dawrx &= ~DAWRX_HYP;
886 
887 	if (nr == 0)
888 		return plpar_set_watchpoint0(dawr, dawrx);
889 	else
890 		return plpar_set_watchpoint1(dawr, dawrx);
891 }
892 
893 #define CMO_CHARACTERISTICS_TOKEN 44
894 #define CMO_MAXLENGTH 1026
895 
896 void pSeries_coalesce_init(void)
897 {
898 	struct hvcall_mpp_x_data mpp_x_data;
899 
900 	if (firmware_has_feature(FW_FEATURE_CMO) && !h_get_mpp_x(&mpp_x_data))
901 		powerpc_firmware_features |= FW_FEATURE_XCMO;
902 	else
903 		powerpc_firmware_features &= ~FW_FEATURE_XCMO;
904 }
905 
906 /**
907  * fw_cmo_feature_init - FW_FEATURE_CMO is not stored in ibm,hypertas-functions,
908  * handle that here. (Stolen from parse_system_parameter_string)
909  */
910 static void __init pSeries_cmo_feature_init(void)
911 {
912 	char *ptr, *key, *value, *end;
913 	int call_status;
914 	int page_order = IOMMU_PAGE_SHIFT_4K;
915 
916 	pr_debug(" -> fw_cmo_feature_init()\n");
917 	spin_lock(&rtas_data_buf_lock);
918 	memset(rtas_data_buf, 0, RTAS_DATA_BUF_SIZE);
919 	call_status = rtas_call(rtas_token("ibm,get-system-parameter"), 3, 1,
920 				NULL,
921 				CMO_CHARACTERISTICS_TOKEN,
922 				__pa(rtas_data_buf),
923 				RTAS_DATA_BUF_SIZE);
924 
925 	if (call_status != 0) {
926 		spin_unlock(&rtas_data_buf_lock);
927 		pr_debug("CMO not available\n");
928 		pr_debug(" <- fw_cmo_feature_init()\n");
929 		return;
930 	}
931 
932 	end = rtas_data_buf + CMO_MAXLENGTH - 2;
933 	ptr = rtas_data_buf + 2;	/* step over strlen value */
934 	key = value = ptr;
935 
936 	while (*ptr && (ptr <= end)) {
937 		/* Separate the key and value by replacing '=' with '\0' and
938 		 * point the value at the string after the '='
939 		 */
940 		if (ptr[0] == '=') {
941 			ptr[0] = '\0';
942 			value = ptr + 1;
943 		} else if (ptr[0] == '\0' || ptr[0] == ',') {
944 			/* Terminate the string containing the key/value pair */
945 			ptr[0] = '\0';
946 
947 			if (key == value) {
948 				pr_debug("Malformed key/value pair\n");
949 				/* Never found a '=', end processing */
950 				break;
951 			}
952 
953 			if (0 == strcmp(key, "CMOPageSize"))
954 				page_order = simple_strtol(value, NULL, 10);
955 			else if (0 == strcmp(key, "PrPSP"))
956 				CMO_PrPSP = simple_strtol(value, NULL, 10);
957 			else if (0 == strcmp(key, "SecPSP"))
958 				CMO_SecPSP = simple_strtol(value, NULL, 10);
959 			value = key = ptr + 1;
960 		}
961 		ptr++;
962 	}
963 
964 	/* Page size is returned as the power of 2 of the page size,
965 	 * convert to the page size in bytes before returning
966 	 */
967 	CMO_PageSize = 1 << page_order;
968 	pr_debug("CMO_PageSize = %lu\n", CMO_PageSize);
969 
970 	if (CMO_PrPSP != -1 || CMO_SecPSP != -1) {
971 		pr_info("CMO enabled\n");
972 		pr_debug("CMO enabled, PrPSP=%d, SecPSP=%d\n", CMO_PrPSP,
973 		         CMO_SecPSP);
974 		powerpc_firmware_features |= FW_FEATURE_CMO;
975 		pSeries_coalesce_init();
976 	} else
977 		pr_debug("CMO not enabled, PrPSP=%d, SecPSP=%d\n", CMO_PrPSP,
978 		         CMO_SecPSP);
979 	spin_unlock(&rtas_data_buf_lock);
980 	pr_debug(" <- fw_cmo_feature_init()\n");
981 }
982 
983 /*
984  * Early initialization.  Relocation is on but do not reference unbolted pages
985  */
986 static void __init pseries_init(void)
987 {
988 	pr_debug(" -> pseries_init()\n");
989 
990 #ifdef CONFIG_HVC_CONSOLE
991 	if (firmware_has_feature(FW_FEATURE_LPAR))
992 		hvc_vio_init_early();
993 #endif
994 	if (firmware_has_feature(FW_FEATURE_XDABR))
995 		ppc_md.set_dabr = pseries_set_xdabr;
996 	else if (firmware_has_feature(FW_FEATURE_DABR))
997 		ppc_md.set_dabr = pseries_set_dabr;
998 
999 	if (firmware_has_feature(FW_FEATURE_SET_MODE))
1000 		ppc_md.set_dawr = pseries_set_dawr;
1001 
1002 	pSeries_cmo_feature_init();
1003 	iommu_init_early_pSeries();
1004 
1005 	pr_debug(" <- pseries_init()\n");
1006 }
1007 
1008 /**
1009  * pseries_power_off - tell firmware about how to power off the system.
1010  *
1011  * This function calls either the power-off rtas token in normal cases
1012  * or the ibm,power-off-ups token (if present & requested) in case of
1013  * a power failure. If power-off token is used, power on will only be
1014  * possible with power button press. If ibm,power-off-ups token is used
1015  * it will allow auto poweron after power is restored.
1016  */
1017 static void pseries_power_off(void)
1018 {
1019 	int rc;
1020 	int rtas_poweroff_ups_token = rtas_token("ibm,power-off-ups");
1021 
1022 	if (rtas_flash_term_hook)
1023 		rtas_flash_term_hook(SYS_POWER_OFF);
1024 
1025 	if (rtas_poweron_auto == 0 ||
1026 		rtas_poweroff_ups_token == RTAS_UNKNOWN_SERVICE) {
1027 		rc = rtas_call(rtas_token("power-off"), 2, 1, NULL, -1, -1);
1028 		printk(KERN_INFO "RTAS power-off returned %d\n", rc);
1029 	} else {
1030 		rc = rtas_call(rtas_poweroff_ups_token, 0, 1, NULL);
1031 		printk(KERN_INFO "RTAS ibm,power-off-ups returned %d\n", rc);
1032 	}
1033 	for (;;);
1034 }
1035 
1036 static int __init pSeries_probe(void)
1037 {
1038 	if (!of_node_is_type(of_root, "chrp"))
1039 		return 0;
1040 
1041 	/* Cell blades firmware claims to be chrp while it's not. Until this
1042 	 * is fixed, we need to avoid those here.
1043 	 */
1044 	if (of_machine_is_compatible("IBM,CPBW-1.0") ||
1045 	    of_machine_is_compatible("IBM,CBEA"))
1046 		return 0;
1047 
1048 	pm_power_off = pseries_power_off;
1049 
1050 	pr_debug("Machine is%s LPAR !\n",
1051 	         (powerpc_firmware_features & FW_FEATURE_LPAR) ? "" : " not");
1052 
1053 	pseries_init();
1054 
1055 	return 1;
1056 }
1057 
1058 static int pSeries_pci_probe_mode(struct pci_bus *bus)
1059 {
1060 	if (firmware_has_feature(FW_FEATURE_LPAR))
1061 		return PCI_PROBE_DEVTREE;
1062 	return PCI_PROBE_NORMAL;
1063 }
1064 
1065 struct pci_controller_ops pseries_pci_controller_ops = {
1066 	.probe_mode		= pSeries_pci_probe_mode,
1067 };
1068 
1069 define_machine(pseries) {
1070 	.name			= "pSeries",
1071 	.probe			= pSeries_probe,
1072 	.setup_arch		= pSeries_setup_arch,
1073 	.init_IRQ		= pseries_init_irq,
1074 	.show_cpuinfo		= pSeries_show_cpuinfo,
1075 	.log_error		= pSeries_log_error,
1076 	.discover_phbs		= pSeries_discover_phbs,
1077 	.pcibios_fixup		= pSeries_final_fixup,
1078 	.restart		= rtas_restart,
1079 	.halt			= rtas_halt,
1080 	.panic			= pseries_panic,
1081 	.get_boot_time		= rtas_get_boot_time,
1082 	.get_rtc_time		= rtas_get_rtc_time,
1083 	.set_rtc_time		= rtas_set_rtc_time,
1084 	.calibrate_decr		= generic_calibrate_decr,
1085 	.progress		= rtas_progress,
1086 	.system_reset_exception = pSeries_system_reset_exception,
1087 	.machine_check_early	= pseries_machine_check_realmode,
1088 	.machine_check_exception = pSeries_machine_check_exception,
1089 #ifdef CONFIG_KEXEC_CORE
1090 	.machine_kexec          = pSeries_machine_kexec,
1091 	.kexec_cpu_down         = pseries_kexec_cpu_down,
1092 #endif
1093 #ifdef CONFIG_MEMORY_HOTPLUG
1094 	.memory_block_size	= pseries_memory_block_size,
1095 #endif
1096 };
1097