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 #include "../../../../drivers/pci/pci.h"
78 
79 DEFINE_STATIC_KEY_FALSE(shared_processor);
80 EXPORT_SYMBOL(shared_processor);
81 
82 int CMO_PrPSP = -1;
83 int CMO_SecPSP = -1;
84 unsigned long CMO_PageSize = (ASM_CONST(1) << IOMMU_PAGE_SHIFT_4K);
85 EXPORT_SYMBOL(CMO_PageSize);
86 
87 int fwnmi_active;  /* TRUE if an FWNMI handler is present */
88 int ibm_nmi_interlock_token;
89 u32 pseries_security_flavor;
90 
91 static void pSeries_show_cpuinfo(struct seq_file *m)
92 {
93 	struct device_node *root;
94 	const char *model = "";
95 
96 	root = of_find_node_by_path("/");
97 	if (root)
98 		model = of_get_property(root, "model", NULL);
99 	seq_printf(m, "machine\t\t: CHRP %s\n", model);
100 	of_node_put(root);
101 	if (radix_enabled())
102 		seq_printf(m, "MMU\t\t: Radix\n");
103 	else
104 		seq_printf(m, "MMU\t\t: Hash\n");
105 }
106 
107 /* Initialize firmware assisted non-maskable interrupts if
108  * the firmware supports this feature.
109  */
110 static void __init fwnmi_init(void)
111 {
112 	unsigned long system_reset_addr, machine_check_addr;
113 	u8 *mce_data_buf;
114 	unsigned int i;
115 	int nr_cpus = num_possible_cpus();
116 #ifdef CONFIG_PPC_BOOK3S_64
117 	struct slb_entry *slb_ptr;
118 	size_t size;
119 #endif
120 	int ibm_nmi_register_token;
121 
122 	ibm_nmi_register_token = rtas_token("ibm,nmi-register");
123 	if (ibm_nmi_register_token == RTAS_UNKNOWN_SERVICE)
124 		return;
125 
126 	ibm_nmi_interlock_token = rtas_token("ibm,nmi-interlock");
127 	if (WARN_ON(ibm_nmi_interlock_token == RTAS_UNKNOWN_SERVICE))
128 		return;
129 
130 	/* If the kernel's not linked at zero we point the firmware at low
131 	 * addresses anyway, and use a trampoline to get to the real code. */
132 	system_reset_addr  = __pa(system_reset_fwnmi) - PHYSICAL_START;
133 	machine_check_addr = __pa(machine_check_fwnmi) - PHYSICAL_START;
134 
135 	if (0 == rtas_call(ibm_nmi_register_token, 2, 1, NULL,
136 			   system_reset_addr, machine_check_addr))
137 		fwnmi_active = 1;
138 
139 	/*
140 	 * Allocate a chunk for per cpu buffer to hold rtas errorlog.
141 	 * It will be used in real mode mce handler, hence it needs to be
142 	 * below RMA.
143 	 */
144 	mce_data_buf = memblock_alloc_try_nid_raw(RTAS_ERROR_LOG_MAX * nr_cpus,
145 					RTAS_ERROR_LOG_MAX, MEMBLOCK_LOW_LIMIT,
146 					ppc64_rma_size, NUMA_NO_NODE);
147 	if (!mce_data_buf)
148 		panic("Failed to allocate %d bytes below %pa for MCE buffer\n",
149 		      RTAS_ERROR_LOG_MAX * nr_cpus, &ppc64_rma_size);
150 
151 	for_each_possible_cpu(i) {
152 		paca_ptrs[i]->mce_data_buf = mce_data_buf +
153 						(RTAS_ERROR_LOG_MAX * i);
154 	}
155 
156 #ifdef CONFIG_PPC_BOOK3S_64
157 	if (!radix_enabled()) {
158 		/* Allocate per cpu area to save old slb contents during MCE */
159 		size = sizeof(struct slb_entry) * mmu_slb_size * nr_cpus;
160 		slb_ptr = memblock_alloc_try_nid_raw(size,
161 				sizeof(struct slb_entry), MEMBLOCK_LOW_LIMIT,
162 				ppc64_rma_size, NUMA_NO_NODE);
163 		if (!slb_ptr)
164 			panic("Failed to allocate %zu bytes below %pa for slb area\n",
165 			      size, &ppc64_rma_size);
166 
167 		for_each_possible_cpu(i)
168 			paca_ptrs[i]->mce_faulty_slbs = slb_ptr + (mmu_slb_size * i);
169 	}
170 #endif
171 }
172 
173 static void pseries_8259_cascade(struct irq_desc *desc)
174 {
175 	struct irq_chip *chip = irq_desc_get_chip(desc);
176 	unsigned int cascade_irq = i8259_irq();
177 
178 	if (cascade_irq)
179 		generic_handle_irq(cascade_irq);
180 
181 	chip->irq_eoi(&desc->irq_data);
182 }
183 
184 static void __init pseries_setup_i8259_cascade(void)
185 {
186 	struct device_node *np, *old, *found = NULL;
187 	unsigned int cascade;
188 	const u32 *addrp;
189 	unsigned long intack = 0;
190 	int naddr;
191 
192 	for_each_node_by_type(np, "interrupt-controller") {
193 		if (of_device_is_compatible(np, "chrp,iic")) {
194 			found = np;
195 			break;
196 		}
197 	}
198 
199 	if (found == NULL) {
200 		printk(KERN_DEBUG "pic: no ISA interrupt controller\n");
201 		return;
202 	}
203 
204 	cascade = irq_of_parse_and_map(found, 0);
205 	if (!cascade) {
206 		printk(KERN_ERR "pic: failed to map cascade interrupt");
207 		return;
208 	}
209 	pr_debug("pic: cascade mapped to irq %d\n", cascade);
210 
211 	for (old = of_node_get(found); old != NULL ; old = np) {
212 		np = of_get_parent(old);
213 		of_node_put(old);
214 		if (np == NULL)
215 			break;
216 		if (!of_node_name_eq(np, "pci"))
217 			continue;
218 		addrp = of_get_property(np, "8259-interrupt-acknowledge", NULL);
219 		if (addrp == NULL)
220 			continue;
221 		naddr = of_n_addr_cells(np);
222 		intack = addrp[naddr-1];
223 		if (naddr > 1)
224 			intack |= ((unsigned long)addrp[naddr-2]) << 32;
225 	}
226 	if (intack)
227 		printk(KERN_DEBUG "pic: PCI 8259 intack at 0x%016lx\n", intack);
228 	i8259_init(found, intack);
229 	of_node_put(found);
230 	irq_set_chained_handler(cascade, pseries_8259_cascade);
231 }
232 
233 static void __init pseries_init_irq(void)
234 {
235 	/* Try using a XIVE if available, otherwise use a XICS */
236 	if (!xive_spapr_init()) {
237 		xics_init();
238 		pseries_setup_i8259_cascade();
239 	}
240 }
241 
242 static void pseries_lpar_enable_pmcs(void)
243 {
244 	unsigned long set, reset;
245 
246 	set = 1UL << 63;
247 	reset = 0;
248 	plpar_hcall_norets(H_PERFMON, set, reset);
249 }
250 
251 static int pci_dn_reconfig_notifier(struct notifier_block *nb, unsigned long action, void *data)
252 {
253 	struct of_reconfig_data *rd = data;
254 	struct device_node *parent, *np = rd->dn;
255 	struct pci_dn *pdn;
256 	int err = NOTIFY_OK;
257 
258 	switch (action) {
259 	case OF_RECONFIG_ATTACH_NODE:
260 		parent = of_get_parent(np);
261 		pdn = parent ? PCI_DN(parent) : NULL;
262 		if (pdn)
263 			pci_add_device_node_info(pdn->phb, np);
264 
265 		of_node_put(parent);
266 		break;
267 	case OF_RECONFIG_DETACH_NODE:
268 		pdn = PCI_DN(np);
269 		if (pdn)
270 			list_del(&pdn->list);
271 		break;
272 	default:
273 		err = NOTIFY_DONE;
274 		break;
275 	}
276 	return err;
277 }
278 
279 static struct notifier_block pci_dn_reconfig_nb = {
280 	.notifier_call = pci_dn_reconfig_notifier,
281 };
282 
283 struct kmem_cache *dtl_cache;
284 
285 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
286 /*
287  * Allocate space for the dispatch trace log for all possible cpus
288  * and register the buffers with the hypervisor.  This is used for
289  * computing time stolen by the hypervisor.
290  */
291 static int alloc_dispatch_logs(void)
292 {
293 	if (!firmware_has_feature(FW_FEATURE_SPLPAR))
294 		return 0;
295 
296 	if (!dtl_cache)
297 		return 0;
298 
299 	alloc_dtl_buffers(0);
300 
301 	/* Register the DTL for the current (boot) cpu */
302 	register_dtl_buffer(smp_processor_id());
303 
304 	return 0;
305 }
306 #else /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
307 static inline int alloc_dispatch_logs(void)
308 {
309 	return 0;
310 }
311 #endif /* CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
312 
313 static int alloc_dispatch_log_kmem_cache(void)
314 {
315 	void (*ctor)(void *) = get_dtl_cache_ctor();
316 
317 	dtl_cache = kmem_cache_create("dtl", DISPATCH_LOG_BYTES,
318 						DISPATCH_LOG_BYTES, 0, ctor);
319 	if (!dtl_cache) {
320 		pr_warn("Failed to create dispatch trace log buffer cache\n");
321 		pr_warn("Stolen time statistics will be unreliable\n");
322 		return 0;
323 	}
324 
325 	return alloc_dispatch_logs();
326 }
327 machine_early_initcall(pseries, alloc_dispatch_log_kmem_cache);
328 
329 DEFINE_PER_CPU(u64, idle_spurr_cycles);
330 DEFINE_PER_CPU(u64, idle_entry_purr_snap);
331 DEFINE_PER_CPU(u64, idle_entry_spurr_snap);
332 static void pseries_lpar_idle(void)
333 {
334 	/*
335 	 * Default handler to go into low thread priority and possibly
336 	 * low power mode by ceding processor to hypervisor
337 	 */
338 
339 	if (!prep_irq_for_idle())
340 		return;
341 
342 	/* Indicate to hypervisor that we are idle. */
343 	pseries_idle_prolog();
344 
345 	/*
346 	 * Yield the processor to the hypervisor.  We return if
347 	 * an external interrupt occurs (which are driven prior
348 	 * to returning here) or if a prod occurs from another
349 	 * processor. When returning here, external interrupts
350 	 * are enabled.
351 	 */
352 	cede_processor();
353 
354 	pseries_idle_epilog();
355 }
356 
357 /*
358  * Enable relocation on during exceptions. This has partition wide scope and
359  * may take a while to complete, if it takes longer than one second we will
360  * just give up rather than wasting any more time on this - if that turns out
361  * to ever be a problem in practice we can move this into a kernel thread to
362  * finish off the process later in boot.
363  */
364 bool pseries_enable_reloc_on_exc(void)
365 {
366 	long rc;
367 	unsigned int delay, total_delay = 0;
368 
369 	while (1) {
370 		rc = enable_reloc_on_exceptions();
371 		if (!H_IS_LONG_BUSY(rc)) {
372 			if (rc == H_P2) {
373 				pr_info("Relocation on exceptions not"
374 					" supported\n");
375 				return false;
376 			} else if (rc != H_SUCCESS) {
377 				pr_warn("Unable to enable relocation"
378 					" on exceptions: %ld\n", rc);
379 				return false;
380 			}
381 			return true;
382 		}
383 
384 		delay = get_longbusy_msecs(rc);
385 		total_delay += delay;
386 		if (total_delay > 1000) {
387 			pr_warn("Warning: Giving up waiting to enable "
388 				"relocation on exceptions (%u msec)!\n",
389 				total_delay);
390 			return false;
391 		}
392 
393 		mdelay(delay);
394 	}
395 }
396 EXPORT_SYMBOL(pseries_enable_reloc_on_exc);
397 
398 void pseries_disable_reloc_on_exc(void)
399 {
400 	long rc;
401 
402 	while (1) {
403 		rc = disable_reloc_on_exceptions();
404 		if (!H_IS_LONG_BUSY(rc))
405 			break;
406 		mdelay(get_longbusy_msecs(rc));
407 	}
408 	if (rc != H_SUCCESS)
409 		pr_warn("Warning: Failed to disable relocation on exceptions: %ld\n",
410 			rc);
411 }
412 EXPORT_SYMBOL(pseries_disable_reloc_on_exc);
413 
414 #ifdef CONFIG_KEXEC_CORE
415 static void pSeries_machine_kexec(struct kimage *image)
416 {
417 	if (firmware_has_feature(FW_FEATURE_SET_MODE))
418 		pseries_disable_reloc_on_exc();
419 
420 	default_machine_kexec(image);
421 }
422 #endif
423 
424 #ifdef __LITTLE_ENDIAN__
425 void pseries_big_endian_exceptions(void)
426 {
427 	long rc;
428 
429 	while (1) {
430 		rc = enable_big_endian_exceptions();
431 		if (!H_IS_LONG_BUSY(rc))
432 			break;
433 		mdelay(get_longbusy_msecs(rc));
434 	}
435 
436 	/*
437 	 * At this point it is unlikely panic() will get anything
438 	 * out to the user, since this is called very late in kexec
439 	 * but at least this will stop us from continuing on further
440 	 * and creating an even more difficult to debug situation.
441 	 *
442 	 * There is a known problem when kdump'ing, if cpus are offline
443 	 * the above call will fail. Rather than panicking again, keep
444 	 * going and hope the kdump kernel is also little endian, which
445 	 * it usually is.
446 	 */
447 	if (rc && !kdump_in_progress())
448 		panic("Could not enable big endian exceptions");
449 }
450 
451 void pseries_little_endian_exceptions(void)
452 {
453 	long rc;
454 
455 	while (1) {
456 		rc = enable_little_endian_exceptions();
457 		if (!H_IS_LONG_BUSY(rc))
458 			break;
459 		mdelay(get_longbusy_msecs(rc));
460 	}
461 	if (rc) {
462 		ppc_md.progress("H_SET_MODE LE exception fail", 0);
463 		panic("Could not enable little endian exceptions");
464 	}
465 }
466 #endif
467 
468 static void __init pSeries_discover_phbs(void)
469 {
470 	struct device_node *node;
471 	struct pci_controller *phb;
472 	struct device_node *root = of_find_node_by_path("/");
473 
474 	for_each_child_of_node(root, node) {
475 		if (!of_node_is_type(node, "pci") &&
476 		    !of_node_is_type(node, "pciex"))
477 			continue;
478 
479 		phb = pcibios_alloc_controller(node);
480 		if (!phb)
481 			continue;
482 		rtas_setup_phb(phb);
483 		pci_process_bridge_OF_ranges(phb, node, 0);
484 		isa_bridge_find_early(phb);
485 		phb->controller_ops = pseries_pci_controller_ops;
486 
487 		/* create pci_dn's for DT nodes under this PHB */
488 		pci_devs_phb_init_dynamic(phb);
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->character & H_CPU_CHAR_BCCTR_LINK_FLUSH_ASSIST)
528 		security_ftr_set(SEC_FTR_BCCTR_LINK_FLUSH_ASSIST);
529 
530 	if (result->behaviour & H_CPU_BEHAV_FLUSH_COUNT_CACHE)
531 		security_ftr_set(SEC_FTR_FLUSH_COUNT_CACHE);
532 
533 	if (result->behaviour & H_CPU_BEHAV_FLUSH_LINK_STACK)
534 		security_ftr_set(SEC_FTR_FLUSH_LINK_STACK);
535 
536 	/*
537 	 * The features below are enabled by default, so we instead look to see
538 	 * if firmware has *disabled* them, and clear them if so.
539 	 * H_CPU_BEHAV_FAVOUR_SECURITY_H could be set only if
540 	 * H_CPU_BEHAV_FAVOUR_SECURITY is.
541 	 */
542 	if (!(result->behaviour & H_CPU_BEHAV_FAVOUR_SECURITY)) {
543 		security_ftr_clear(SEC_FTR_FAVOUR_SECURITY);
544 		pseries_security_flavor = 0;
545 	} else if (result->behaviour & H_CPU_BEHAV_FAVOUR_SECURITY_H)
546 		pseries_security_flavor = 1;
547 	else
548 		pseries_security_flavor = 2;
549 
550 	if (!(result->behaviour & H_CPU_BEHAV_L1D_FLUSH_PR))
551 		security_ftr_clear(SEC_FTR_L1D_FLUSH_PR);
552 
553 	if (result->behaviour & H_CPU_BEHAV_NO_L1D_FLUSH_ENTRY)
554 		security_ftr_clear(SEC_FTR_L1D_FLUSH_ENTRY);
555 
556 	if (result->behaviour & H_CPU_BEHAV_NO_L1D_FLUSH_UACCESS)
557 		security_ftr_clear(SEC_FTR_L1D_FLUSH_UACCESS);
558 
559 	if (result->behaviour & H_CPU_BEHAV_NO_STF_BARRIER)
560 		security_ftr_clear(SEC_FTR_STF_BARRIER);
561 
562 	if (!(result->behaviour & H_CPU_BEHAV_BNDS_CHK_SPEC_BAR))
563 		security_ftr_clear(SEC_FTR_BNDS_CHK_SPEC_BAR);
564 }
565 
566 void pseries_setup_security_mitigations(void)
567 {
568 	struct h_cpu_char_result result;
569 	enum l1d_flush_type types;
570 	bool enable;
571 	long rc;
572 
573 	/*
574 	 * Set features to the defaults assumed by init_cpu_char_feature_flags()
575 	 * so it can set/clear again any features that might have changed after
576 	 * migration, and in case the hypercall fails and it is not even called.
577 	 */
578 	powerpc_security_features = SEC_FTR_DEFAULT;
579 
580 	rc = plpar_get_cpu_characteristics(&result);
581 	if (rc == H_SUCCESS)
582 		init_cpu_char_feature_flags(&result);
583 
584 	/*
585 	 * We're the guest so this doesn't apply to us, clear it to simplify
586 	 * handling of it elsewhere.
587 	 */
588 	security_ftr_clear(SEC_FTR_L1D_FLUSH_HV);
589 
590 	types = L1D_FLUSH_FALLBACK;
591 
592 	if (security_ftr_enabled(SEC_FTR_L1D_FLUSH_TRIG2))
593 		types |= L1D_FLUSH_MTTRIG;
594 
595 	if (security_ftr_enabled(SEC_FTR_L1D_FLUSH_ORI30))
596 		types |= L1D_FLUSH_ORI;
597 
598 	enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) && \
599 		 security_ftr_enabled(SEC_FTR_L1D_FLUSH_PR);
600 
601 	setup_rfi_flush(types, enable);
602 	setup_count_cache_flush();
603 
604 	enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) &&
605 		 security_ftr_enabled(SEC_FTR_L1D_FLUSH_ENTRY);
606 	setup_entry_flush(enable);
607 
608 	enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) &&
609 		 security_ftr_enabled(SEC_FTR_L1D_FLUSH_UACCESS);
610 	setup_uaccess_flush(enable);
611 
612 	setup_stf_barrier();
613 }
614 
615 #ifdef CONFIG_PCI_IOV
616 enum rtas_iov_fw_value_map {
617 	NUM_RES_PROPERTY  = 0, /* Number of Resources */
618 	LOW_INT           = 1, /* Lowest 32 bits of Address */
619 	START_OF_ENTRIES  = 2, /* Always start of entry */
620 	APERTURE_PROPERTY = 2, /* Start of entry+ to  Aperture Size */
621 	WDW_SIZE_PROPERTY = 4, /* Start of entry+ to Window Size */
622 	NEXT_ENTRY        = 7  /* Go to next entry on array */
623 };
624 
625 enum get_iov_fw_value_index {
626 	BAR_ADDRS     = 1,    /*  Get Bar Address */
627 	APERTURE_SIZE = 2,    /*  Get Aperture Size */
628 	WDW_SIZE      = 3     /*  Get Window Size */
629 };
630 
631 static resource_size_t pseries_get_iov_fw_value(struct pci_dev *dev, int resno,
632 						enum get_iov_fw_value_index value)
633 {
634 	const int *indexes;
635 	struct device_node *dn = pci_device_to_OF_node(dev);
636 	int i, num_res, ret = 0;
637 
638 	indexes = of_get_property(dn, "ibm,open-sriov-vf-bar-info", NULL);
639 	if (!indexes)
640 		return  0;
641 
642 	/*
643 	 * First element in the array is the number of Bars
644 	 * returned.  Search through the list to find the matching
645 	 * bar
646 	 */
647 	num_res = of_read_number(&indexes[NUM_RES_PROPERTY], 1);
648 	if (resno >= num_res)
649 		return 0; /* or an errror */
650 
651 	i = START_OF_ENTRIES + NEXT_ENTRY * resno;
652 	switch (value) {
653 	case BAR_ADDRS:
654 		ret = of_read_number(&indexes[i], 2);
655 		break;
656 	case APERTURE_SIZE:
657 		ret = of_read_number(&indexes[i + APERTURE_PROPERTY], 2);
658 		break;
659 	case WDW_SIZE:
660 		ret = of_read_number(&indexes[i + WDW_SIZE_PROPERTY], 2);
661 		break;
662 	}
663 
664 	return ret;
665 }
666 
667 static void of_pci_set_vf_bar_size(struct pci_dev *dev, const int *indexes)
668 {
669 	struct resource *res;
670 	resource_size_t base, size;
671 	int i, r, num_res;
672 
673 	num_res = of_read_number(&indexes[NUM_RES_PROPERTY], 1);
674 	num_res = min_t(int, num_res, PCI_SRIOV_NUM_BARS);
675 	for (i = START_OF_ENTRIES, r = 0; r < num_res && r < PCI_SRIOV_NUM_BARS;
676 	     i += NEXT_ENTRY, r++) {
677 		res = &dev->resource[r + PCI_IOV_RESOURCES];
678 		base = of_read_number(&indexes[i], 2);
679 		size = of_read_number(&indexes[i + APERTURE_PROPERTY], 2);
680 		res->flags = pci_parse_of_flags(of_read_number
681 						(&indexes[i + LOW_INT], 1), 0);
682 		res->flags |= (IORESOURCE_MEM_64 | IORESOURCE_PCI_FIXED);
683 		res->name = pci_name(dev);
684 		res->start = base;
685 		res->end = base + size - 1;
686 	}
687 }
688 
689 static void of_pci_parse_iov_addrs(struct pci_dev *dev, const int *indexes)
690 {
691 	struct resource *res, *root, *conflict;
692 	resource_size_t base, size;
693 	int i, r, num_res;
694 
695 	/*
696 	 * First element in the array is the number of Bars
697 	 * returned.  Search through the list to find the matching
698 	 * bars assign them from firmware into resources structure.
699 	 */
700 	num_res = of_read_number(&indexes[NUM_RES_PROPERTY], 1);
701 	for (i = START_OF_ENTRIES, r = 0; r < num_res && r < PCI_SRIOV_NUM_BARS;
702 	     i += NEXT_ENTRY, r++) {
703 		res = &dev->resource[r + PCI_IOV_RESOURCES];
704 		base = of_read_number(&indexes[i], 2);
705 		size = of_read_number(&indexes[i + WDW_SIZE_PROPERTY], 2);
706 		res->name = pci_name(dev);
707 		res->start = base;
708 		res->end = base + size - 1;
709 		root = &iomem_resource;
710 		dev_dbg(&dev->dev,
711 			"pSeries IOV BAR %d: trying firmware assignment %pR\n",
712 			 r + PCI_IOV_RESOURCES, res);
713 		conflict = request_resource_conflict(root, res);
714 		if (conflict) {
715 			dev_info(&dev->dev,
716 				 "BAR %d: %pR conflicts with %s %pR\n",
717 				 r + PCI_IOV_RESOURCES, res,
718 				 conflict->name, conflict);
719 			res->flags |= IORESOURCE_UNSET;
720 		}
721 	}
722 }
723 
724 static void pseries_disable_sriov_resources(struct pci_dev *pdev)
725 {
726 	int i;
727 
728 	pci_warn(pdev, "No hypervisor support for SR-IOV on this device, IOV BARs disabled.\n");
729 	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++)
730 		pdev->resource[i + PCI_IOV_RESOURCES].flags = 0;
731 }
732 
733 static void pseries_pci_fixup_resources(struct pci_dev *pdev)
734 {
735 	const int *indexes;
736 	struct device_node *dn = pci_device_to_OF_node(pdev);
737 
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_set_vf_bar_size(pdev, indexes);
742 	else
743 		pseries_disable_sriov_resources(pdev);
744 }
745 
746 static void pseries_pci_fixup_iov_resources(struct pci_dev *pdev)
747 {
748 	const int *indexes;
749 	struct device_node *dn = pci_device_to_OF_node(pdev);
750 
751 	if (!pdev->is_physfn || pci_dev_is_added(pdev))
752 		return;
753 	/*Firmware must support open sriov otherwise dont configure*/
754 	indexes = of_get_property(dn, "ibm,open-sriov-vf-bar-info", NULL);
755 	if (indexes)
756 		of_pci_parse_iov_addrs(pdev, indexes);
757 	else
758 		pseries_disable_sriov_resources(pdev);
759 }
760 
761 static resource_size_t pseries_pci_iov_resource_alignment(struct pci_dev *pdev,
762 							  int resno)
763 {
764 	const __be32 *reg;
765 	struct device_node *dn = pci_device_to_OF_node(pdev);
766 
767 	/*Firmware must support open sriov otherwise report regular alignment*/
768 	reg = of_get_property(dn, "ibm,is-open-sriov-pf", NULL);
769 	if (!reg)
770 		return pci_iov_resource_size(pdev, resno);
771 
772 	if (!pdev->is_physfn)
773 		return 0;
774 	return pseries_get_iov_fw_value(pdev,
775 					resno - PCI_IOV_RESOURCES,
776 					APERTURE_SIZE);
777 }
778 #endif
779 
780 static void __init pSeries_setup_arch(void)
781 {
782 	set_arch_panic_timeout(10, ARCH_PANIC_TIMEOUT);
783 
784 	/* Discover PIC type and setup ppc_md accordingly */
785 	smp_init_pseries();
786 
787 
788 	if (radix_enabled() && !mmu_has_feature(MMU_FTR_GTSE))
789 		if (!firmware_has_feature(FW_FEATURE_RPT_INVALIDATE))
790 			panic("BUG: Radix support requires either GTSE or RPT_INVALIDATE\n");
791 
792 
793 	/* openpic global configuration register (64-bit format). */
794 	/* openpic Interrupt Source Unit pointer (64-bit format). */
795 	/* python0 facility area (mmio) (64-bit format) REAL address. */
796 
797 	/* init to some ~sane value until calibrate_delay() runs */
798 	loops_per_jiffy = 50000000;
799 
800 	fwnmi_init();
801 
802 	pseries_setup_security_mitigations();
803 	pseries_lpar_read_hblkrm_characteristics();
804 
805 	/* By default, only probe PCI (can be overridden by rtas_pci) */
806 	pci_add_flags(PCI_PROBE_ONLY);
807 
808 	/* Find and initialize PCI host bridges */
809 	init_pci_config_tokens();
810 	of_reconfig_notifier_register(&pci_dn_reconfig_nb);
811 
812 	pSeries_nvram_init();
813 
814 	if (firmware_has_feature(FW_FEATURE_LPAR)) {
815 		vpa_init(boot_cpuid);
816 
817 		if (lppaca_shared_proc(get_lppaca())) {
818 			static_branch_enable(&shared_processor);
819 			pv_spinlocks_init();
820 		}
821 
822 		ppc_md.power_save = pseries_lpar_idle;
823 		ppc_md.enable_pmcs = pseries_lpar_enable_pmcs;
824 #ifdef CONFIG_PCI_IOV
825 		ppc_md.pcibios_fixup_resources =
826 			pseries_pci_fixup_resources;
827 		ppc_md.pcibios_fixup_sriov =
828 			pseries_pci_fixup_iov_resources;
829 		ppc_md.pcibios_iov_resource_alignment =
830 			pseries_pci_iov_resource_alignment;
831 #endif
832 	} else {
833 		/* No special idle routine */
834 		ppc_md.enable_pmcs = power4_enable_pmcs;
835 	}
836 
837 	ppc_md.pcibios_root_bridge_prepare = pseries_root_bridge_prepare;
838 
839 	if (swiotlb_force == SWIOTLB_FORCE)
840 		ppc_swiotlb_enable = 1;
841 }
842 
843 static void pseries_panic(char *str)
844 {
845 	panic_flush_kmsg_end();
846 	rtas_os_term(str);
847 }
848 
849 static int __init pSeries_init_panel(void)
850 {
851 	/* Manually leave the kernel version on the panel. */
852 #ifdef __BIG_ENDIAN__
853 	ppc_md.progress("Linux ppc64\n", 0);
854 #else
855 	ppc_md.progress("Linux ppc64le\n", 0);
856 #endif
857 	ppc_md.progress(init_utsname()->version, 0);
858 
859 	return 0;
860 }
861 machine_arch_initcall(pseries, pSeries_init_panel);
862 
863 static int pseries_set_dabr(unsigned long dabr, unsigned long dabrx)
864 {
865 	return plpar_hcall_norets(H_SET_DABR, dabr);
866 }
867 
868 static int pseries_set_xdabr(unsigned long dabr, unsigned long dabrx)
869 {
870 	/* Have to set at least one bit in the DABRX according to PAPR */
871 	if (dabrx == 0 && dabr == 0)
872 		dabrx = DABRX_USER;
873 	/* PAPR says we can only set kernel and user bits */
874 	dabrx &= DABRX_KERNEL | DABRX_USER;
875 
876 	return plpar_hcall_norets(H_SET_XDABR, dabr, dabrx);
877 }
878 
879 static int pseries_set_dawr(int nr, unsigned long dawr, unsigned long dawrx)
880 {
881 	/* PAPR says we can't set HYP */
882 	dawrx &= ~DAWRX_HYP;
883 
884 	if (nr == 0)
885 		return plpar_set_watchpoint0(dawr, dawrx);
886 	else
887 		return plpar_set_watchpoint1(dawr, dawrx);
888 }
889 
890 #define CMO_CHARACTERISTICS_TOKEN 44
891 #define CMO_MAXLENGTH 1026
892 
893 void pSeries_coalesce_init(void)
894 {
895 	struct hvcall_mpp_x_data mpp_x_data;
896 
897 	if (firmware_has_feature(FW_FEATURE_CMO) && !h_get_mpp_x(&mpp_x_data))
898 		powerpc_firmware_features |= FW_FEATURE_XCMO;
899 	else
900 		powerpc_firmware_features &= ~FW_FEATURE_XCMO;
901 }
902 
903 /**
904  * fw_cmo_feature_init - FW_FEATURE_CMO is not stored in ibm,hypertas-functions,
905  * handle that here. (Stolen from parse_system_parameter_string)
906  */
907 static void pSeries_cmo_feature_init(void)
908 {
909 	char *ptr, *key, *value, *end;
910 	int call_status;
911 	int page_order = IOMMU_PAGE_SHIFT_4K;
912 
913 	pr_debug(" -> fw_cmo_feature_init()\n");
914 	spin_lock(&rtas_data_buf_lock);
915 	memset(rtas_data_buf, 0, RTAS_DATA_BUF_SIZE);
916 	call_status = rtas_call(rtas_token("ibm,get-system-parameter"), 3, 1,
917 				NULL,
918 				CMO_CHARACTERISTICS_TOKEN,
919 				__pa(rtas_data_buf),
920 				RTAS_DATA_BUF_SIZE);
921 
922 	if (call_status != 0) {
923 		spin_unlock(&rtas_data_buf_lock);
924 		pr_debug("CMO not available\n");
925 		pr_debug(" <- fw_cmo_feature_init()\n");
926 		return;
927 	}
928 
929 	end = rtas_data_buf + CMO_MAXLENGTH - 2;
930 	ptr = rtas_data_buf + 2;	/* step over strlen value */
931 	key = value = ptr;
932 
933 	while (*ptr && (ptr <= end)) {
934 		/* Separate the key and value by replacing '=' with '\0' and
935 		 * point the value at the string after the '='
936 		 */
937 		if (ptr[0] == '=') {
938 			ptr[0] = '\0';
939 			value = ptr + 1;
940 		} else if (ptr[0] == '\0' || ptr[0] == ',') {
941 			/* Terminate the string containing the key/value pair */
942 			ptr[0] = '\0';
943 
944 			if (key == value) {
945 				pr_debug("Malformed key/value pair\n");
946 				/* Never found a '=', end processing */
947 				break;
948 			}
949 
950 			if (0 == strcmp(key, "CMOPageSize"))
951 				page_order = simple_strtol(value, NULL, 10);
952 			else if (0 == strcmp(key, "PrPSP"))
953 				CMO_PrPSP = simple_strtol(value, NULL, 10);
954 			else if (0 == strcmp(key, "SecPSP"))
955 				CMO_SecPSP = simple_strtol(value, NULL, 10);
956 			value = key = ptr + 1;
957 		}
958 		ptr++;
959 	}
960 
961 	/* Page size is returned as the power of 2 of the page size,
962 	 * convert to the page size in bytes before returning
963 	 */
964 	CMO_PageSize = 1 << page_order;
965 	pr_debug("CMO_PageSize = %lu\n", CMO_PageSize);
966 
967 	if (CMO_PrPSP != -1 || CMO_SecPSP != -1) {
968 		pr_info("CMO enabled\n");
969 		pr_debug("CMO enabled, PrPSP=%d, SecPSP=%d\n", CMO_PrPSP,
970 		         CMO_SecPSP);
971 		powerpc_firmware_features |= FW_FEATURE_CMO;
972 		pSeries_coalesce_init();
973 	} else
974 		pr_debug("CMO not enabled, PrPSP=%d, SecPSP=%d\n", CMO_PrPSP,
975 		         CMO_SecPSP);
976 	spin_unlock(&rtas_data_buf_lock);
977 	pr_debug(" <- fw_cmo_feature_init()\n");
978 }
979 
980 /*
981  * Early initialization.  Relocation is on but do not reference unbolted pages
982  */
983 static void __init pseries_init(void)
984 {
985 	pr_debug(" -> pseries_init()\n");
986 
987 #ifdef CONFIG_HVC_CONSOLE
988 	if (firmware_has_feature(FW_FEATURE_LPAR))
989 		hvc_vio_init_early();
990 #endif
991 	if (firmware_has_feature(FW_FEATURE_XDABR))
992 		ppc_md.set_dabr = pseries_set_xdabr;
993 	else if (firmware_has_feature(FW_FEATURE_DABR))
994 		ppc_md.set_dabr = pseries_set_dabr;
995 
996 	if (firmware_has_feature(FW_FEATURE_SET_MODE))
997 		ppc_md.set_dawr = pseries_set_dawr;
998 
999 	pSeries_cmo_feature_init();
1000 	iommu_init_early_pSeries();
1001 
1002 	pr_debug(" <- pseries_init()\n");
1003 }
1004 
1005 /**
1006  * pseries_power_off - tell firmware about how to power off the system.
1007  *
1008  * This function calls either the power-off rtas token in normal cases
1009  * or the ibm,power-off-ups token (if present & requested) in case of
1010  * a power failure. If power-off token is used, power on will only be
1011  * possible with power button press. If ibm,power-off-ups token is used
1012  * it will allow auto poweron after power is restored.
1013  */
1014 static void pseries_power_off(void)
1015 {
1016 	int rc;
1017 	int rtas_poweroff_ups_token = rtas_token("ibm,power-off-ups");
1018 
1019 	if (rtas_flash_term_hook)
1020 		rtas_flash_term_hook(SYS_POWER_OFF);
1021 
1022 	if (rtas_poweron_auto == 0 ||
1023 		rtas_poweroff_ups_token == RTAS_UNKNOWN_SERVICE) {
1024 		rc = rtas_call(rtas_token("power-off"), 2, 1, NULL, -1, -1);
1025 		printk(KERN_INFO "RTAS power-off returned %d\n", rc);
1026 	} else {
1027 		rc = rtas_call(rtas_poweroff_ups_token, 0, 1, NULL);
1028 		printk(KERN_INFO "RTAS ibm,power-off-ups returned %d\n", rc);
1029 	}
1030 	for (;;);
1031 }
1032 
1033 static int __init pSeries_probe(void)
1034 {
1035 	if (!of_node_is_type(of_root, "chrp"))
1036 		return 0;
1037 
1038 	/* Cell blades firmware claims to be chrp while it's not. Until this
1039 	 * is fixed, we need to avoid those here.
1040 	 */
1041 	if (of_machine_is_compatible("IBM,CPBW-1.0") ||
1042 	    of_machine_is_compatible("IBM,CBEA"))
1043 		return 0;
1044 
1045 	pm_power_off = pseries_power_off;
1046 
1047 	pr_debug("Machine is%s LPAR !\n",
1048 	         (powerpc_firmware_features & FW_FEATURE_LPAR) ? "" : " not");
1049 
1050 	pseries_init();
1051 
1052 	return 1;
1053 }
1054 
1055 static int pSeries_pci_probe_mode(struct pci_bus *bus)
1056 {
1057 	if (firmware_has_feature(FW_FEATURE_LPAR))
1058 		return PCI_PROBE_DEVTREE;
1059 	return PCI_PROBE_NORMAL;
1060 }
1061 
1062 struct pci_controller_ops pseries_pci_controller_ops = {
1063 	.probe_mode		= pSeries_pci_probe_mode,
1064 };
1065 
1066 define_machine(pseries) {
1067 	.name			= "pSeries",
1068 	.probe			= pSeries_probe,
1069 	.setup_arch		= pSeries_setup_arch,
1070 	.init_IRQ		= pseries_init_irq,
1071 	.show_cpuinfo		= pSeries_show_cpuinfo,
1072 	.log_error		= pSeries_log_error,
1073 	.discover_phbs		= pSeries_discover_phbs,
1074 	.pcibios_fixup		= pSeries_final_fixup,
1075 	.restart		= rtas_restart,
1076 	.halt			= rtas_halt,
1077 	.panic			= pseries_panic,
1078 	.get_boot_time		= rtas_get_boot_time,
1079 	.get_rtc_time		= rtas_get_rtc_time,
1080 	.set_rtc_time		= rtas_set_rtc_time,
1081 	.calibrate_decr		= generic_calibrate_decr,
1082 	.progress		= rtas_progress,
1083 	.system_reset_exception = pSeries_system_reset_exception,
1084 	.machine_check_early	= pseries_machine_check_realmode,
1085 	.machine_check_exception = pSeries_machine_check_exception,
1086 #ifdef CONFIG_KEXEC_CORE
1087 	.machine_kexec          = pSeries_machine_kexec,
1088 	.kexec_cpu_down         = pseries_kexec_cpu_down,
1089 #endif
1090 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
1091 	.memory_block_size	= pseries_memory_block_size,
1092 #endif
1093 };
1094