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/platform_device.h>
18 #include <linux/sched.h>
19 #include <linux/kernel.h>
20 #include <linux/mm.h>
21 #include <linux/stddef.h>
22 #include <linux/unistd.h>
23 #include <linux/user.h>
24 #include <linux/tty.h>
25 #include <linux/major.h>
26 #include <linux/interrupt.h>
27 #include <linux/reboot.h>
28 #include <linux/init.h>
29 #include <linux/ioport.h>
30 #include <linux/console.h>
31 #include <linux/pci.h>
32 #include <linux/utsname.h>
33 #include <linux/adb.h>
34 #include <linux/export.h>
35 #include <linux/delay.h>
36 #include <linux/irq.h>
37 #include <linux/seq_file.h>
38 #include <linux/root_dev.h>
39 #include <linux/of.h>
40 #include <linux/of_irq.h>
41 #include <linux/of_pci.h>
42 #include <linux/memblock.h>
43 #include <linux/swiotlb.h>
44 #include <linux/seq_buf.h>
45 
46 #include <asm/mmu.h>
47 #include <asm/processor.h>
48 #include <asm/io.h>
49 #include <asm/rtas.h>
50 #include <asm/pci-bridge.h>
51 #include <asm/iommu.h>
52 #include <asm/dma.h>
53 #include <asm/machdep.h>
54 #include <asm/irq.h>
55 #include <asm/time.h>
56 #include <asm/nvram.h>
57 #include <asm/pmc.h>
58 #include <asm/xics.h>
59 #include <asm/xive.h>
60 #include <asm/ppc-pci.h>
61 #include <asm/i8259.h>
62 #include <asm/udbg.h>
63 #include <asm/smp.h>
64 #include <asm/firmware.h>
65 #include <asm/eeh.h>
66 #include <asm/reg.h>
67 #include <asm/plpar_wrappers.h>
68 #include <asm/kexec.h>
69 #include <asm/isa-bridge.h>
70 #include <asm/security_features.h>
71 #include <asm/asm-const.h>
72 #include <asm/idle.h>
73 #include <asm/swiotlb.h>
74 #include <asm/svm.h>
75 #include <asm/dtl.h>
76 #include <asm/hvconsole.h>
77 #include <asm/setup.h>
78 
79 #include "pseries.h"
80 
81 DEFINE_STATIC_KEY_FALSE(shared_processor);
82 EXPORT_SYMBOL(shared_processor);
83 
84 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
85 struct static_key paravirt_steal_enabled;
86 struct static_key paravirt_steal_rq_enabled;
87 
88 static bool steal_acc = true;
89 static int __init parse_no_stealacc(char *arg)
90 {
91 	steal_acc = false;
92 	return 0;
93 }
94 
95 early_param("no-steal-acc", parse_no_stealacc);
96 #endif
97 
98 int CMO_PrPSP = -1;
99 int CMO_SecPSP = -1;
100 unsigned long CMO_PageSize = (ASM_CONST(1) << IOMMU_PAGE_SHIFT_4K);
101 EXPORT_SYMBOL(CMO_PageSize);
102 
103 int fwnmi_active;  /* TRUE if an FWNMI handler is present */
104 int ibm_nmi_interlock_token;
105 u32 pseries_security_flavor;
106 
107 static void pSeries_show_cpuinfo(struct seq_file *m)
108 {
109 	struct device_node *root;
110 	const char *model = "";
111 
112 	root = of_find_node_by_path("/");
113 	if (root)
114 		model = of_get_property(root, "model", NULL);
115 	seq_printf(m, "machine\t\t: CHRP %s\n", model);
116 	of_node_put(root);
117 	if (radix_enabled())
118 		seq_printf(m, "MMU\t\t: Radix\n");
119 	else
120 		seq_printf(m, "MMU\t\t: Hash\n");
121 }
122 
123 /* Initialize firmware assisted non-maskable interrupts if
124  * the firmware supports this feature.
125  */
126 static void __init fwnmi_init(void)
127 {
128 	unsigned long system_reset_addr, machine_check_addr;
129 	u8 *mce_data_buf;
130 	unsigned int i;
131 	int nr_cpus = num_possible_cpus();
132 #ifdef CONFIG_PPC_64S_HASH_MMU
133 	struct slb_entry *slb_ptr;
134 	size_t size;
135 #endif
136 	int ibm_nmi_register_token;
137 
138 	ibm_nmi_register_token = rtas_token("ibm,nmi-register");
139 	if (ibm_nmi_register_token == RTAS_UNKNOWN_SERVICE)
140 		return;
141 
142 	ibm_nmi_interlock_token = rtas_token("ibm,nmi-interlock");
143 	if (WARN_ON(ibm_nmi_interlock_token == RTAS_UNKNOWN_SERVICE))
144 		return;
145 
146 	/* If the kernel's not linked at zero we point the firmware at low
147 	 * addresses anyway, and use a trampoline to get to the real code. */
148 	system_reset_addr  = __pa(system_reset_fwnmi) - PHYSICAL_START;
149 	machine_check_addr = __pa(machine_check_fwnmi) - PHYSICAL_START;
150 
151 	if (0 == rtas_call(ibm_nmi_register_token, 2, 1, NULL,
152 			   system_reset_addr, machine_check_addr))
153 		fwnmi_active = 1;
154 
155 	/*
156 	 * Allocate a chunk for per cpu buffer to hold rtas errorlog.
157 	 * It will be used in real mode mce handler, hence it needs to be
158 	 * below RMA.
159 	 */
160 	mce_data_buf = memblock_alloc_try_nid_raw(RTAS_ERROR_LOG_MAX * nr_cpus,
161 					RTAS_ERROR_LOG_MAX, MEMBLOCK_LOW_LIMIT,
162 					ppc64_rma_size, NUMA_NO_NODE);
163 	if (!mce_data_buf)
164 		panic("Failed to allocate %d bytes below %pa for MCE buffer\n",
165 		      RTAS_ERROR_LOG_MAX * nr_cpus, &ppc64_rma_size);
166 
167 	for_each_possible_cpu(i) {
168 		paca_ptrs[i]->mce_data_buf = mce_data_buf +
169 						(RTAS_ERROR_LOG_MAX * i);
170 	}
171 
172 #ifdef CONFIG_PPC_64S_HASH_MMU
173 	if (!radix_enabled()) {
174 		/* Allocate per cpu area to save old slb contents during MCE */
175 		size = sizeof(struct slb_entry) * mmu_slb_size * nr_cpus;
176 		slb_ptr = memblock_alloc_try_nid_raw(size,
177 				sizeof(struct slb_entry), MEMBLOCK_LOW_LIMIT,
178 				ppc64_rma_size, NUMA_NO_NODE);
179 		if (!slb_ptr)
180 			panic("Failed to allocate %zu bytes below %pa for slb area\n",
181 			      size, &ppc64_rma_size);
182 
183 		for_each_possible_cpu(i)
184 			paca_ptrs[i]->mce_faulty_slbs = slb_ptr + (mmu_slb_size * i);
185 	}
186 #endif
187 }
188 
189 /*
190  * Affix a device for the first timer to the platform bus if
191  * we have firmware support for the H_WATCHDOG hypercall.
192  */
193 static __init int pseries_wdt_init(void)
194 {
195 	if (firmware_has_feature(FW_FEATURE_WATCHDOG))
196 		platform_device_register_simple("pseries-wdt", 0, NULL, 0);
197 	return 0;
198 }
199 machine_subsys_initcall(pseries, pseries_wdt_init);
200 
201 static void pseries_8259_cascade(struct irq_desc *desc)
202 {
203 	struct irq_chip *chip = irq_desc_get_chip(desc);
204 	unsigned int cascade_irq = i8259_irq();
205 
206 	if (cascade_irq)
207 		generic_handle_irq(cascade_irq);
208 
209 	chip->irq_eoi(&desc->irq_data);
210 }
211 
212 static void __init pseries_setup_i8259_cascade(void)
213 {
214 	struct device_node *np, *old, *found = NULL;
215 	unsigned int cascade;
216 	const u32 *addrp;
217 	unsigned long intack = 0;
218 	int naddr;
219 
220 	for_each_node_by_type(np, "interrupt-controller") {
221 		if (of_device_is_compatible(np, "chrp,iic")) {
222 			found = np;
223 			break;
224 		}
225 	}
226 
227 	if (found == NULL) {
228 		printk(KERN_DEBUG "pic: no ISA interrupt controller\n");
229 		return;
230 	}
231 
232 	cascade = irq_of_parse_and_map(found, 0);
233 	if (!cascade) {
234 		printk(KERN_ERR "pic: failed to map cascade interrupt");
235 		return;
236 	}
237 	pr_debug("pic: cascade mapped to irq %d\n", cascade);
238 
239 	for (old = of_node_get(found); old != NULL ; old = np) {
240 		np = of_get_parent(old);
241 		of_node_put(old);
242 		if (np == NULL)
243 			break;
244 		if (!of_node_name_eq(np, "pci"))
245 			continue;
246 		addrp = of_get_property(np, "8259-interrupt-acknowledge", NULL);
247 		if (addrp == NULL)
248 			continue;
249 		naddr = of_n_addr_cells(np);
250 		intack = addrp[naddr-1];
251 		if (naddr > 1)
252 			intack |= ((unsigned long)addrp[naddr-2]) << 32;
253 	}
254 	if (intack)
255 		printk(KERN_DEBUG "pic: PCI 8259 intack at 0x%016lx\n", intack);
256 	i8259_init(found, intack);
257 	of_node_put(found);
258 	irq_set_chained_handler(cascade, pseries_8259_cascade);
259 }
260 
261 static void __init pseries_init_irq(void)
262 {
263 	/* Try using a XIVE if available, otherwise use a XICS */
264 	if (!xive_spapr_init()) {
265 		xics_init();
266 		pseries_setup_i8259_cascade();
267 	}
268 }
269 
270 static void pseries_lpar_enable_pmcs(void)
271 {
272 	unsigned long set, reset;
273 
274 	set = 1UL << 63;
275 	reset = 0;
276 	plpar_hcall_norets(H_PERFMON, set, reset);
277 }
278 
279 static int pci_dn_reconfig_notifier(struct notifier_block *nb, unsigned long action, void *data)
280 {
281 	struct of_reconfig_data *rd = data;
282 	struct device_node *parent, *np = rd->dn;
283 	struct pci_dn *pdn;
284 	int err = NOTIFY_OK;
285 
286 	switch (action) {
287 	case OF_RECONFIG_ATTACH_NODE:
288 		parent = of_get_parent(np);
289 		pdn = parent ? PCI_DN(parent) : NULL;
290 		if (pdn)
291 			pci_add_device_node_info(pdn->phb, np);
292 
293 		of_node_put(parent);
294 		break;
295 	case OF_RECONFIG_DETACH_NODE:
296 		pdn = PCI_DN(np);
297 		if (pdn)
298 			list_del(&pdn->list);
299 		break;
300 	default:
301 		err = NOTIFY_DONE;
302 		break;
303 	}
304 	return err;
305 }
306 
307 static struct notifier_block pci_dn_reconfig_nb = {
308 	.notifier_call = pci_dn_reconfig_notifier,
309 };
310 
311 struct kmem_cache *dtl_cache;
312 
313 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
314 /*
315  * Allocate space for the dispatch trace log for all possible cpus
316  * and register the buffers with the hypervisor.  This is used for
317  * computing time stolen by the hypervisor.
318  */
319 static int alloc_dispatch_logs(void)
320 {
321 	if (!firmware_has_feature(FW_FEATURE_SPLPAR))
322 		return 0;
323 
324 	if (!dtl_cache)
325 		return 0;
326 
327 	alloc_dtl_buffers(0);
328 
329 	/* Register the DTL for the current (boot) cpu */
330 	register_dtl_buffer(smp_processor_id());
331 
332 	return 0;
333 }
334 #else /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
335 static inline int alloc_dispatch_logs(void)
336 {
337 	return 0;
338 }
339 #endif /* CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
340 
341 static int alloc_dispatch_log_kmem_cache(void)
342 {
343 	void (*ctor)(void *) = get_dtl_cache_ctor();
344 
345 	dtl_cache = kmem_cache_create("dtl", DISPATCH_LOG_BYTES,
346 						DISPATCH_LOG_BYTES, 0, ctor);
347 	if (!dtl_cache) {
348 		pr_warn("Failed to create dispatch trace log buffer cache\n");
349 		pr_warn("Stolen time statistics will be unreliable\n");
350 		return 0;
351 	}
352 
353 	return alloc_dispatch_logs();
354 }
355 machine_early_initcall(pseries, alloc_dispatch_log_kmem_cache);
356 
357 DEFINE_PER_CPU(u64, idle_spurr_cycles);
358 DEFINE_PER_CPU(u64, idle_entry_purr_snap);
359 DEFINE_PER_CPU(u64, idle_entry_spurr_snap);
360 static void pseries_lpar_idle(void)
361 {
362 	/*
363 	 * Default handler to go into low thread priority and possibly
364 	 * low power mode by ceding processor to hypervisor
365 	 */
366 
367 	if (!prep_irq_for_idle())
368 		return;
369 
370 	/* Indicate to hypervisor that we are idle. */
371 	pseries_idle_prolog();
372 
373 	/*
374 	 * Yield the processor to the hypervisor.  We return if
375 	 * an external interrupt occurs (which are driven prior
376 	 * to returning here) or if a prod occurs from another
377 	 * processor. When returning here, external interrupts
378 	 * are enabled.
379 	 */
380 	cede_processor();
381 
382 	pseries_idle_epilog();
383 }
384 
385 static bool pseries_reloc_on_exception_enabled;
386 
387 bool pseries_reloc_on_exception(void)
388 {
389 	return pseries_reloc_on_exception_enabled;
390 }
391 EXPORT_SYMBOL_GPL(pseries_reloc_on_exception);
392 
393 /*
394  * Enable relocation on during exceptions. This has partition wide scope and
395  * may take a while to complete, if it takes longer than one second we will
396  * just give up rather than wasting any more time on this - if that turns out
397  * to ever be a problem in practice we can move this into a kernel thread to
398  * finish off the process later in boot.
399  */
400 bool pseries_enable_reloc_on_exc(void)
401 {
402 	long rc;
403 	unsigned int delay, total_delay = 0;
404 
405 	while (1) {
406 		rc = enable_reloc_on_exceptions();
407 		if (!H_IS_LONG_BUSY(rc)) {
408 			if (rc == H_P2) {
409 				pr_info("Relocation on exceptions not"
410 					" supported\n");
411 				return false;
412 			} else if (rc != H_SUCCESS) {
413 				pr_warn("Unable to enable relocation"
414 					" on exceptions: %ld\n", rc);
415 				return false;
416 			}
417 			pseries_reloc_on_exception_enabled = true;
418 			return true;
419 		}
420 
421 		delay = get_longbusy_msecs(rc);
422 		total_delay += delay;
423 		if (total_delay > 1000) {
424 			pr_warn("Warning: Giving up waiting to enable "
425 				"relocation on exceptions (%u msec)!\n",
426 				total_delay);
427 			return false;
428 		}
429 
430 		mdelay(delay);
431 	}
432 }
433 EXPORT_SYMBOL(pseries_enable_reloc_on_exc);
434 
435 void pseries_disable_reloc_on_exc(void)
436 {
437 	long rc;
438 
439 	while (1) {
440 		rc = disable_reloc_on_exceptions();
441 		if (!H_IS_LONG_BUSY(rc))
442 			break;
443 		mdelay(get_longbusy_msecs(rc));
444 	}
445 	if (rc == H_SUCCESS)
446 		pseries_reloc_on_exception_enabled = false;
447 	else
448 		pr_warn("Warning: Failed to disable relocation on exceptions: %ld\n",
449 			rc);
450 }
451 EXPORT_SYMBOL(pseries_disable_reloc_on_exc);
452 
453 #ifdef __LITTLE_ENDIAN__
454 void pseries_big_endian_exceptions(void)
455 {
456 	long rc;
457 
458 	while (1) {
459 		rc = enable_big_endian_exceptions();
460 		if (!H_IS_LONG_BUSY(rc))
461 			break;
462 		mdelay(get_longbusy_msecs(rc));
463 	}
464 
465 	/*
466 	 * At this point it is unlikely panic() will get anything
467 	 * out to the user, since this is called very late in kexec
468 	 * but at least this will stop us from continuing on further
469 	 * and creating an even more difficult to debug situation.
470 	 *
471 	 * There is a known problem when kdump'ing, if cpus are offline
472 	 * the above call will fail. Rather than panicking again, keep
473 	 * going and hope the kdump kernel is also little endian, which
474 	 * it usually is.
475 	 */
476 	if (rc && !kdump_in_progress())
477 		panic("Could not enable big endian exceptions");
478 }
479 
480 void __init pseries_little_endian_exceptions(void)
481 {
482 	long rc;
483 
484 	while (1) {
485 		rc = enable_little_endian_exceptions();
486 		if (!H_IS_LONG_BUSY(rc))
487 			break;
488 		mdelay(get_longbusy_msecs(rc));
489 	}
490 	if (rc) {
491 		ppc_md.progress("H_SET_MODE LE exception fail", 0);
492 		panic("Could not enable little endian exceptions");
493 	}
494 }
495 #endif
496 
497 static void __init pSeries_discover_phbs(void)
498 {
499 	struct device_node *node;
500 	struct pci_controller *phb;
501 	struct device_node *root = of_find_node_by_path("/");
502 
503 	for_each_child_of_node(root, node) {
504 		if (!of_node_is_type(node, "pci") &&
505 		    !of_node_is_type(node, "pciex"))
506 			continue;
507 
508 		phb = pcibios_alloc_controller(node);
509 		if (!phb)
510 			continue;
511 		rtas_setup_phb(phb);
512 		pci_process_bridge_OF_ranges(phb, node, 0);
513 		isa_bridge_find_early(phb);
514 		phb->controller_ops = pseries_pci_controller_ops;
515 
516 		/* create pci_dn's for DT nodes under this PHB */
517 		pci_devs_phb_init_dynamic(phb);
518 
519 		pseries_msi_allocate_domains(phb);
520 	}
521 
522 	of_node_put(root);
523 
524 	/*
525 	 * PCI_PROBE_ONLY and PCI_REASSIGN_ALL_BUS can be set via properties
526 	 * in chosen.
527 	 */
528 	of_pci_check_probe_only();
529 }
530 
531 static void init_cpu_char_feature_flags(struct h_cpu_char_result *result)
532 {
533 	/*
534 	 * The features below are disabled by default, so we instead look to see
535 	 * if firmware has *enabled* them, and set them if so.
536 	 */
537 	if (result->character & H_CPU_CHAR_SPEC_BAR_ORI31)
538 		security_ftr_set(SEC_FTR_SPEC_BAR_ORI31);
539 
540 	if (result->character & H_CPU_CHAR_BCCTRL_SERIALISED)
541 		security_ftr_set(SEC_FTR_BCCTRL_SERIALISED);
542 
543 	if (result->character & H_CPU_CHAR_L1D_FLUSH_ORI30)
544 		security_ftr_set(SEC_FTR_L1D_FLUSH_ORI30);
545 
546 	if (result->character & H_CPU_CHAR_L1D_FLUSH_TRIG2)
547 		security_ftr_set(SEC_FTR_L1D_FLUSH_TRIG2);
548 
549 	if (result->character & H_CPU_CHAR_L1D_THREAD_PRIV)
550 		security_ftr_set(SEC_FTR_L1D_THREAD_PRIV);
551 
552 	if (result->character & H_CPU_CHAR_COUNT_CACHE_DISABLED)
553 		security_ftr_set(SEC_FTR_COUNT_CACHE_DISABLED);
554 
555 	if (result->character & H_CPU_CHAR_BCCTR_FLUSH_ASSIST)
556 		security_ftr_set(SEC_FTR_BCCTR_FLUSH_ASSIST);
557 
558 	if (result->character & H_CPU_CHAR_BCCTR_LINK_FLUSH_ASSIST)
559 		security_ftr_set(SEC_FTR_BCCTR_LINK_FLUSH_ASSIST);
560 
561 	if (result->behaviour & H_CPU_BEHAV_FLUSH_COUNT_CACHE)
562 		security_ftr_set(SEC_FTR_FLUSH_COUNT_CACHE);
563 
564 	if (result->behaviour & H_CPU_BEHAV_FLUSH_LINK_STACK)
565 		security_ftr_set(SEC_FTR_FLUSH_LINK_STACK);
566 
567 	/*
568 	 * The features below are enabled by default, so we instead look to see
569 	 * if firmware has *disabled* them, and clear them if so.
570 	 * H_CPU_BEHAV_FAVOUR_SECURITY_H could be set only if
571 	 * H_CPU_BEHAV_FAVOUR_SECURITY is.
572 	 */
573 	if (!(result->behaviour & H_CPU_BEHAV_FAVOUR_SECURITY)) {
574 		security_ftr_clear(SEC_FTR_FAVOUR_SECURITY);
575 		pseries_security_flavor = 0;
576 	} else if (result->behaviour & H_CPU_BEHAV_FAVOUR_SECURITY_H)
577 		pseries_security_flavor = 1;
578 	else
579 		pseries_security_flavor = 2;
580 
581 	if (!(result->behaviour & H_CPU_BEHAV_L1D_FLUSH_PR))
582 		security_ftr_clear(SEC_FTR_L1D_FLUSH_PR);
583 
584 	if (result->behaviour & H_CPU_BEHAV_NO_L1D_FLUSH_ENTRY)
585 		security_ftr_clear(SEC_FTR_L1D_FLUSH_ENTRY);
586 
587 	if (result->behaviour & H_CPU_BEHAV_NO_L1D_FLUSH_UACCESS)
588 		security_ftr_clear(SEC_FTR_L1D_FLUSH_UACCESS);
589 
590 	if (result->behaviour & H_CPU_BEHAV_NO_STF_BARRIER)
591 		security_ftr_clear(SEC_FTR_STF_BARRIER);
592 
593 	if (!(result->behaviour & H_CPU_BEHAV_BNDS_CHK_SPEC_BAR))
594 		security_ftr_clear(SEC_FTR_BNDS_CHK_SPEC_BAR);
595 }
596 
597 void pseries_setup_security_mitigations(void)
598 {
599 	struct h_cpu_char_result result;
600 	enum l1d_flush_type types;
601 	bool enable;
602 	long rc;
603 
604 	/*
605 	 * Set features to the defaults assumed by init_cpu_char_feature_flags()
606 	 * so it can set/clear again any features that might have changed after
607 	 * migration, and in case the hypercall fails and it is not even called.
608 	 */
609 	powerpc_security_features = SEC_FTR_DEFAULT;
610 
611 	rc = plpar_get_cpu_characteristics(&result);
612 	if (rc == H_SUCCESS)
613 		init_cpu_char_feature_flags(&result);
614 
615 	/*
616 	 * We're the guest so this doesn't apply to us, clear it to simplify
617 	 * handling of it elsewhere.
618 	 */
619 	security_ftr_clear(SEC_FTR_L1D_FLUSH_HV);
620 
621 	types = L1D_FLUSH_FALLBACK;
622 
623 	if (security_ftr_enabled(SEC_FTR_L1D_FLUSH_TRIG2))
624 		types |= L1D_FLUSH_MTTRIG;
625 
626 	if (security_ftr_enabled(SEC_FTR_L1D_FLUSH_ORI30))
627 		types |= L1D_FLUSH_ORI;
628 
629 	enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) && \
630 		 security_ftr_enabled(SEC_FTR_L1D_FLUSH_PR);
631 
632 	setup_rfi_flush(types, enable);
633 	setup_count_cache_flush();
634 
635 	enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) &&
636 		 security_ftr_enabled(SEC_FTR_L1D_FLUSH_ENTRY);
637 	setup_entry_flush(enable);
638 
639 	enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) &&
640 		 security_ftr_enabled(SEC_FTR_L1D_FLUSH_UACCESS);
641 	setup_uaccess_flush(enable);
642 
643 	setup_stf_barrier();
644 }
645 
646 #ifdef CONFIG_PCI_IOV
647 enum rtas_iov_fw_value_map {
648 	NUM_RES_PROPERTY  = 0, /* Number of Resources */
649 	LOW_INT           = 1, /* Lowest 32 bits of Address */
650 	START_OF_ENTRIES  = 2, /* Always start of entry */
651 	APERTURE_PROPERTY = 2, /* Start of entry+ to  Aperture Size */
652 	WDW_SIZE_PROPERTY = 4, /* Start of entry+ to Window Size */
653 	NEXT_ENTRY        = 7  /* Go to next entry on array */
654 };
655 
656 enum get_iov_fw_value_index {
657 	BAR_ADDRS     = 1,    /*  Get Bar Address */
658 	APERTURE_SIZE = 2,    /*  Get Aperture Size */
659 	WDW_SIZE      = 3     /*  Get Window Size */
660 };
661 
662 static resource_size_t pseries_get_iov_fw_value(struct pci_dev *dev, int resno,
663 						enum get_iov_fw_value_index value)
664 {
665 	const int *indexes;
666 	struct device_node *dn = pci_device_to_OF_node(dev);
667 	int i, num_res, ret = 0;
668 
669 	indexes = of_get_property(dn, "ibm,open-sriov-vf-bar-info", NULL);
670 	if (!indexes)
671 		return  0;
672 
673 	/*
674 	 * First element in the array is the number of Bars
675 	 * returned.  Search through the list to find the matching
676 	 * bar
677 	 */
678 	num_res = of_read_number(&indexes[NUM_RES_PROPERTY], 1);
679 	if (resno >= num_res)
680 		return 0; /* or an error */
681 
682 	i = START_OF_ENTRIES + NEXT_ENTRY * resno;
683 	switch (value) {
684 	case BAR_ADDRS:
685 		ret = of_read_number(&indexes[i], 2);
686 		break;
687 	case APERTURE_SIZE:
688 		ret = of_read_number(&indexes[i + APERTURE_PROPERTY], 2);
689 		break;
690 	case WDW_SIZE:
691 		ret = of_read_number(&indexes[i + WDW_SIZE_PROPERTY], 2);
692 		break;
693 	}
694 
695 	return ret;
696 }
697 
698 static void of_pci_set_vf_bar_size(struct pci_dev *dev, const int *indexes)
699 {
700 	struct resource *res;
701 	resource_size_t base, size;
702 	int i, r, num_res;
703 
704 	num_res = of_read_number(&indexes[NUM_RES_PROPERTY], 1);
705 	num_res = min_t(int, num_res, PCI_SRIOV_NUM_BARS);
706 	for (i = START_OF_ENTRIES, r = 0; r < num_res && r < PCI_SRIOV_NUM_BARS;
707 	     i += NEXT_ENTRY, r++) {
708 		res = &dev->resource[r + PCI_IOV_RESOURCES];
709 		base = of_read_number(&indexes[i], 2);
710 		size = of_read_number(&indexes[i + APERTURE_PROPERTY], 2);
711 		res->flags = pci_parse_of_flags(of_read_number
712 						(&indexes[i + LOW_INT], 1), 0);
713 		res->flags |= (IORESOURCE_MEM_64 | IORESOURCE_PCI_FIXED);
714 		res->name = pci_name(dev);
715 		res->start = base;
716 		res->end = base + size - 1;
717 	}
718 }
719 
720 static void of_pci_parse_iov_addrs(struct pci_dev *dev, const int *indexes)
721 {
722 	struct resource *res, *root, *conflict;
723 	resource_size_t base, size;
724 	int i, r, num_res;
725 
726 	/*
727 	 * First element in the array is the number of Bars
728 	 * returned.  Search through the list to find the matching
729 	 * bars assign them from firmware into resources structure.
730 	 */
731 	num_res = of_read_number(&indexes[NUM_RES_PROPERTY], 1);
732 	for (i = START_OF_ENTRIES, r = 0; r < num_res && r < PCI_SRIOV_NUM_BARS;
733 	     i += NEXT_ENTRY, r++) {
734 		res = &dev->resource[r + PCI_IOV_RESOURCES];
735 		base = of_read_number(&indexes[i], 2);
736 		size = of_read_number(&indexes[i + WDW_SIZE_PROPERTY], 2);
737 		res->name = pci_name(dev);
738 		res->start = base;
739 		res->end = base + size - 1;
740 		root = &iomem_resource;
741 		dev_dbg(&dev->dev,
742 			"pSeries IOV BAR %d: trying firmware assignment %pR\n",
743 			 r + PCI_IOV_RESOURCES, res);
744 		conflict = request_resource_conflict(root, res);
745 		if (conflict) {
746 			dev_info(&dev->dev,
747 				 "BAR %d: %pR conflicts with %s %pR\n",
748 				 r + PCI_IOV_RESOURCES, res,
749 				 conflict->name, conflict);
750 			res->flags |= IORESOURCE_UNSET;
751 		}
752 	}
753 }
754 
755 static void pseries_disable_sriov_resources(struct pci_dev *pdev)
756 {
757 	int i;
758 
759 	pci_warn(pdev, "No hypervisor support for SR-IOV on this device, IOV BARs disabled.\n");
760 	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++)
761 		pdev->resource[i + PCI_IOV_RESOURCES].flags = 0;
762 }
763 
764 static void pseries_pci_fixup_resources(struct pci_dev *pdev)
765 {
766 	const int *indexes;
767 	struct device_node *dn = pci_device_to_OF_node(pdev);
768 
769 	/*Firmware must support open sriov otherwise dont configure*/
770 	indexes = of_get_property(dn, "ibm,open-sriov-vf-bar-info", NULL);
771 	if (indexes)
772 		of_pci_set_vf_bar_size(pdev, indexes);
773 	else
774 		pseries_disable_sriov_resources(pdev);
775 }
776 
777 static void pseries_pci_fixup_iov_resources(struct pci_dev *pdev)
778 {
779 	const int *indexes;
780 	struct device_node *dn = pci_device_to_OF_node(pdev);
781 
782 	if (!pdev->is_physfn)
783 		return;
784 	/*Firmware must support open sriov otherwise don't configure*/
785 	indexes = of_get_property(dn, "ibm,open-sriov-vf-bar-info", NULL);
786 	if (indexes)
787 		of_pci_parse_iov_addrs(pdev, indexes);
788 	else
789 		pseries_disable_sriov_resources(pdev);
790 }
791 
792 static resource_size_t pseries_pci_iov_resource_alignment(struct pci_dev *pdev,
793 							  int resno)
794 {
795 	const __be32 *reg;
796 	struct device_node *dn = pci_device_to_OF_node(pdev);
797 
798 	/*Firmware must support open sriov otherwise report regular alignment*/
799 	reg = of_get_property(dn, "ibm,is-open-sriov-pf", NULL);
800 	if (!reg)
801 		return pci_iov_resource_size(pdev, resno);
802 
803 	if (!pdev->is_physfn)
804 		return 0;
805 	return pseries_get_iov_fw_value(pdev,
806 					resno - PCI_IOV_RESOURCES,
807 					APERTURE_SIZE);
808 }
809 #endif
810 
811 static void __init pSeries_setup_arch(void)
812 {
813 	set_arch_panic_timeout(10, ARCH_PANIC_TIMEOUT);
814 
815 	/* Discover PIC type and setup ppc_md accordingly */
816 	smp_init_pseries();
817 
818 
819 	if (radix_enabled() && !mmu_has_feature(MMU_FTR_GTSE))
820 		if (!firmware_has_feature(FW_FEATURE_RPT_INVALIDATE))
821 			panic("BUG: Radix support requires either GTSE or RPT_INVALIDATE\n");
822 
823 
824 	/* openpic global configuration register (64-bit format). */
825 	/* openpic Interrupt Source Unit pointer (64-bit format). */
826 	/* python0 facility area (mmio) (64-bit format) REAL address. */
827 
828 	/* init to some ~sane value until calibrate_delay() runs */
829 	loops_per_jiffy = 50000000;
830 
831 	fwnmi_init();
832 
833 	pseries_setup_security_mitigations();
834 	if (!radix_enabled())
835 		pseries_lpar_read_hblkrm_characteristics();
836 
837 	/* By default, only probe PCI (can be overridden by rtas_pci) */
838 	pci_add_flags(PCI_PROBE_ONLY);
839 
840 	/* Find and initialize PCI host bridges */
841 	init_pci_config_tokens();
842 	of_reconfig_notifier_register(&pci_dn_reconfig_nb);
843 
844 	pSeries_nvram_init();
845 
846 	if (firmware_has_feature(FW_FEATURE_LPAR)) {
847 		vpa_init(boot_cpuid);
848 
849 		if (lppaca_shared_proc(get_lppaca())) {
850 			static_branch_enable(&shared_processor);
851 			pv_spinlocks_init();
852 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
853 			static_key_slow_inc(&paravirt_steal_enabled);
854 			if (steal_acc)
855 				static_key_slow_inc(&paravirt_steal_rq_enabled);
856 #endif
857 		}
858 
859 		ppc_md.power_save = pseries_lpar_idle;
860 		ppc_md.enable_pmcs = pseries_lpar_enable_pmcs;
861 #ifdef CONFIG_PCI_IOV
862 		ppc_md.pcibios_fixup_resources =
863 			pseries_pci_fixup_resources;
864 		ppc_md.pcibios_fixup_sriov =
865 			pseries_pci_fixup_iov_resources;
866 		ppc_md.pcibios_iov_resource_alignment =
867 			pseries_pci_iov_resource_alignment;
868 #endif
869 	} else {
870 		/* No special idle routine */
871 		ppc_md.enable_pmcs = power4_enable_pmcs;
872 	}
873 
874 	ppc_md.pcibios_root_bridge_prepare = pseries_root_bridge_prepare;
875 	pseries_rng_init();
876 }
877 
878 static void pseries_panic(char *str)
879 {
880 	panic_flush_kmsg_end();
881 	rtas_os_term(str);
882 }
883 
884 static int __init pSeries_init_panel(void)
885 {
886 	/* Manually leave the kernel version on the panel. */
887 #ifdef __BIG_ENDIAN__
888 	ppc_md.progress("Linux ppc64\n", 0);
889 #else
890 	ppc_md.progress("Linux ppc64le\n", 0);
891 #endif
892 	ppc_md.progress(init_utsname()->version, 0);
893 
894 	return 0;
895 }
896 machine_arch_initcall(pseries, pSeries_init_panel);
897 
898 static int pseries_set_dabr(unsigned long dabr, unsigned long dabrx)
899 {
900 	return plpar_hcall_norets(H_SET_DABR, dabr);
901 }
902 
903 static int pseries_set_xdabr(unsigned long dabr, unsigned long dabrx)
904 {
905 	/* Have to set at least one bit in the DABRX according to PAPR */
906 	if (dabrx == 0 && dabr == 0)
907 		dabrx = DABRX_USER;
908 	/* PAPR says we can only set kernel and user bits */
909 	dabrx &= DABRX_KERNEL | DABRX_USER;
910 
911 	return plpar_hcall_norets(H_SET_XDABR, dabr, dabrx);
912 }
913 
914 static int pseries_set_dawr(int nr, unsigned long dawr, unsigned long dawrx)
915 {
916 	/* PAPR says we can't set HYP */
917 	dawrx &= ~DAWRX_HYP;
918 
919 	if (nr == 0)
920 		return plpar_set_watchpoint0(dawr, dawrx);
921 	else
922 		return plpar_set_watchpoint1(dawr, dawrx);
923 }
924 
925 #define CMO_CHARACTERISTICS_TOKEN 44
926 #define CMO_MAXLENGTH 1026
927 
928 void pSeries_coalesce_init(void)
929 {
930 	struct hvcall_mpp_x_data mpp_x_data;
931 
932 	if (firmware_has_feature(FW_FEATURE_CMO) && !h_get_mpp_x(&mpp_x_data))
933 		powerpc_firmware_features |= FW_FEATURE_XCMO;
934 	else
935 		powerpc_firmware_features &= ~FW_FEATURE_XCMO;
936 }
937 
938 /**
939  * fw_cmo_feature_init - FW_FEATURE_CMO is not stored in ibm,hypertas-functions,
940  * handle that here. (Stolen from parse_system_parameter_string)
941  */
942 static void __init pSeries_cmo_feature_init(void)
943 {
944 	char *ptr, *key, *value, *end;
945 	int call_status;
946 	int page_order = IOMMU_PAGE_SHIFT_4K;
947 
948 	pr_debug(" -> fw_cmo_feature_init()\n");
949 	spin_lock(&rtas_data_buf_lock);
950 	memset(rtas_data_buf, 0, RTAS_DATA_BUF_SIZE);
951 	call_status = rtas_call(rtas_token("ibm,get-system-parameter"), 3, 1,
952 				NULL,
953 				CMO_CHARACTERISTICS_TOKEN,
954 				__pa(rtas_data_buf),
955 				RTAS_DATA_BUF_SIZE);
956 
957 	if (call_status != 0) {
958 		spin_unlock(&rtas_data_buf_lock);
959 		pr_debug("CMO not available\n");
960 		pr_debug(" <- fw_cmo_feature_init()\n");
961 		return;
962 	}
963 
964 	end = rtas_data_buf + CMO_MAXLENGTH - 2;
965 	ptr = rtas_data_buf + 2;	/* step over strlen value */
966 	key = value = ptr;
967 
968 	while (*ptr && (ptr <= end)) {
969 		/* Separate the key and value by replacing '=' with '\0' and
970 		 * point the value at the string after the '='
971 		 */
972 		if (ptr[0] == '=') {
973 			ptr[0] = '\0';
974 			value = ptr + 1;
975 		} else if (ptr[0] == '\0' || ptr[0] == ',') {
976 			/* Terminate the string containing the key/value pair */
977 			ptr[0] = '\0';
978 
979 			if (key == value) {
980 				pr_debug("Malformed key/value pair\n");
981 				/* Never found a '=', end processing */
982 				break;
983 			}
984 
985 			if (0 == strcmp(key, "CMOPageSize"))
986 				page_order = simple_strtol(value, NULL, 10);
987 			else if (0 == strcmp(key, "PrPSP"))
988 				CMO_PrPSP = simple_strtol(value, NULL, 10);
989 			else if (0 == strcmp(key, "SecPSP"))
990 				CMO_SecPSP = simple_strtol(value, NULL, 10);
991 			value = key = ptr + 1;
992 		}
993 		ptr++;
994 	}
995 
996 	/* Page size is returned as the power of 2 of the page size,
997 	 * convert to the page size in bytes before returning
998 	 */
999 	CMO_PageSize = 1 << page_order;
1000 	pr_debug("CMO_PageSize = %lu\n", CMO_PageSize);
1001 
1002 	if (CMO_PrPSP != -1 || CMO_SecPSP != -1) {
1003 		pr_info("CMO enabled\n");
1004 		pr_debug("CMO enabled, PrPSP=%d, SecPSP=%d\n", CMO_PrPSP,
1005 		         CMO_SecPSP);
1006 		powerpc_firmware_features |= FW_FEATURE_CMO;
1007 		pSeries_coalesce_init();
1008 	} else
1009 		pr_debug("CMO not enabled, PrPSP=%d, SecPSP=%d\n", CMO_PrPSP,
1010 		         CMO_SecPSP);
1011 	spin_unlock(&rtas_data_buf_lock);
1012 	pr_debug(" <- fw_cmo_feature_init()\n");
1013 }
1014 
1015 static void __init pseries_add_hw_description(void)
1016 {
1017 	struct device_node *dn;
1018 	const char *s;
1019 
1020 	dn = of_find_node_by_path("/openprom");
1021 	if (dn) {
1022 		if (of_property_read_string(dn, "model", &s) == 0)
1023 			seq_buf_printf(&ppc_hw_desc, "of:%s ", s);
1024 
1025 		of_node_put(dn);
1026 	}
1027 
1028 	dn = of_find_node_by_path("/hypervisor");
1029 	if (dn) {
1030 		if (of_property_read_string(dn, "compatible", &s) == 0)
1031 			seq_buf_printf(&ppc_hw_desc, "hv:%s ", s);
1032 
1033 		of_node_put(dn);
1034 		return;
1035 	}
1036 
1037 	if (of_property_read_bool(of_root, "ibm,powervm-partition") ||
1038 	    of_property_read_bool(of_root, "ibm,fw-net-version"))
1039 		seq_buf_printf(&ppc_hw_desc, "hv:phyp ");
1040 }
1041 
1042 /*
1043  * Early initialization.  Relocation is on but do not reference unbolted pages
1044  */
1045 static void __init pseries_init(void)
1046 {
1047 	pr_debug(" -> pseries_init()\n");
1048 
1049 	pseries_add_hw_description();
1050 
1051 #ifdef CONFIG_HVC_CONSOLE
1052 	if (firmware_has_feature(FW_FEATURE_LPAR))
1053 		hvc_vio_init_early();
1054 #endif
1055 	if (firmware_has_feature(FW_FEATURE_XDABR))
1056 		ppc_md.set_dabr = pseries_set_xdabr;
1057 	else if (firmware_has_feature(FW_FEATURE_DABR))
1058 		ppc_md.set_dabr = pseries_set_dabr;
1059 
1060 	if (firmware_has_feature(FW_FEATURE_SET_MODE))
1061 		ppc_md.set_dawr = pseries_set_dawr;
1062 
1063 	pSeries_cmo_feature_init();
1064 	iommu_init_early_pSeries();
1065 
1066 	pr_debug(" <- pseries_init()\n");
1067 }
1068 
1069 /**
1070  * pseries_power_off - tell firmware about how to power off the system.
1071  *
1072  * This function calls either the power-off rtas token in normal cases
1073  * or the ibm,power-off-ups token (if present & requested) in case of
1074  * a power failure. If power-off token is used, power on will only be
1075  * possible with power button press. If ibm,power-off-ups token is used
1076  * it will allow auto poweron after power is restored.
1077  */
1078 static void pseries_power_off(void)
1079 {
1080 	int rc;
1081 	int rtas_poweroff_ups_token = rtas_token("ibm,power-off-ups");
1082 
1083 	if (rtas_flash_term_hook)
1084 		rtas_flash_term_hook(SYS_POWER_OFF);
1085 
1086 	if (rtas_poweron_auto == 0 ||
1087 		rtas_poweroff_ups_token == RTAS_UNKNOWN_SERVICE) {
1088 		rc = rtas_call(rtas_token("power-off"), 2, 1, NULL, -1, -1);
1089 		printk(KERN_INFO "RTAS power-off returned %d\n", rc);
1090 	} else {
1091 		rc = rtas_call(rtas_poweroff_ups_token, 0, 1, NULL);
1092 		printk(KERN_INFO "RTAS ibm,power-off-ups returned %d\n", rc);
1093 	}
1094 	for (;;);
1095 }
1096 
1097 static int __init pSeries_probe(void)
1098 {
1099 	if (!of_node_is_type(of_root, "chrp"))
1100 		return 0;
1101 
1102 	/* Cell blades firmware claims to be chrp while it's not. Until this
1103 	 * is fixed, we need to avoid those here.
1104 	 */
1105 	if (of_machine_is_compatible("IBM,CPBW-1.0") ||
1106 	    of_machine_is_compatible("IBM,CBEA"))
1107 		return 0;
1108 
1109 	pm_power_off = pseries_power_off;
1110 
1111 	pr_debug("Machine is%s LPAR !\n",
1112 	         (powerpc_firmware_features & FW_FEATURE_LPAR) ? "" : " not");
1113 
1114 	pseries_init();
1115 
1116 	return 1;
1117 }
1118 
1119 static int pSeries_pci_probe_mode(struct pci_bus *bus)
1120 {
1121 	if (firmware_has_feature(FW_FEATURE_LPAR))
1122 		return PCI_PROBE_DEVTREE;
1123 	return PCI_PROBE_NORMAL;
1124 }
1125 
1126 struct pci_controller_ops pseries_pci_controller_ops = {
1127 	.probe_mode		= pSeries_pci_probe_mode,
1128 };
1129 
1130 define_machine(pseries) {
1131 	.name			= "pSeries",
1132 	.probe			= pSeries_probe,
1133 	.setup_arch		= pSeries_setup_arch,
1134 	.init_IRQ		= pseries_init_irq,
1135 	.show_cpuinfo		= pSeries_show_cpuinfo,
1136 	.log_error		= pSeries_log_error,
1137 	.discover_phbs		= pSeries_discover_phbs,
1138 	.pcibios_fixup		= pSeries_final_fixup,
1139 	.restart		= rtas_restart,
1140 	.halt			= rtas_halt,
1141 	.panic			= pseries_panic,
1142 	.get_boot_time		= rtas_get_boot_time,
1143 	.get_rtc_time		= rtas_get_rtc_time,
1144 	.set_rtc_time		= rtas_set_rtc_time,
1145 	.calibrate_decr		= generic_calibrate_decr,
1146 	.progress		= rtas_progress,
1147 	.system_reset_exception = pSeries_system_reset_exception,
1148 	.machine_check_early	= pseries_machine_check_realmode,
1149 	.machine_check_exception = pSeries_machine_check_exception,
1150 	.machine_check_log_err	= pSeries_machine_check_log_err,
1151 #ifdef CONFIG_KEXEC_CORE
1152 	.machine_kexec          = pseries_machine_kexec,
1153 	.kexec_cpu_down         = pseries_kexec_cpu_down,
1154 #endif
1155 #ifdef CONFIG_MEMORY_HOTPLUG
1156 	.memory_block_size	= pseries_memory_block_size,
1157 #endif
1158 };
1159