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/papr-sysparm.h>
61 #include <asm/ppc-pci.h>
62 #include <asm/i8259.h>
63 #include <asm/udbg.h>
64 #include <asm/smp.h>
65 #include <asm/firmware.h>
66 #include <asm/eeh.h>
67 #include <asm/reg.h>
68 #include <asm/plpar_wrappers.h>
69 #include <asm/kexec.h>
70 #include <asm/isa-bridge.h>
71 #include <asm/security_features.h>
72 #include <asm/asm-const.h>
73 #include <asm/idle.h>
74 #include <asm/swiotlb.h>
75 #include <asm/svm.h>
76 #include <asm/dtl.h>
77 #include <asm/hvconsole.h>
78 #include <asm/setup.h>
79
80 #include "pseries.h"
81
82 DEFINE_STATIC_KEY_FALSE(shared_processor);
83 EXPORT_SYMBOL(shared_processor);
84
85 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
86 struct static_key paravirt_steal_enabled;
87 struct static_key paravirt_steal_rq_enabled;
88
89 static bool steal_acc = true;
parse_no_stealacc(char * arg)90 static int __init parse_no_stealacc(char *arg)
91 {
92 steal_acc = false;
93 return 0;
94 }
95
96 early_param("no-steal-acc", parse_no_stealacc);
97 #endif
98
99 int CMO_PrPSP = -1;
100 int CMO_SecPSP = -1;
101 unsigned long CMO_PageSize = (ASM_CONST(1) << IOMMU_PAGE_SHIFT_4K);
102 EXPORT_SYMBOL(CMO_PageSize);
103
104 int fwnmi_active; /* TRUE if an FWNMI handler is present */
105 int ibm_nmi_interlock_token;
106 u32 pseries_security_flavor;
107
pSeries_show_cpuinfo(struct seq_file * m)108 static void pSeries_show_cpuinfo(struct seq_file *m)
109 {
110 struct device_node *root;
111 const char *model = "";
112
113 root = of_find_node_by_path("/");
114 if (root)
115 model = of_get_property(root, "model", NULL);
116 seq_printf(m, "machine\t\t: CHRP %s\n", model);
117 of_node_put(root);
118 if (radix_enabled())
119 seq_printf(m, "MMU\t\t: Radix\n");
120 else
121 seq_printf(m, "MMU\t\t: Hash\n");
122 }
123
124 /* Initialize firmware assisted non-maskable interrupts if
125 * the firmware supports this feature.
126 */
fwnmi_init(void)127 static void __init fwnmi_init(void)
128 {
129 unsigned long system_reset_addr, machine_check_addr;
130 u8 *mce_data_buf;
131 unsigned int i;
132 int nr_cpus = num_possible_cpus();
133 #ifdef CONFIG_PPC_64S_HASH_MMU
134 struct slb_entry *slb_ptr;
135 size_t size;
136 #endif
137 int ibm_nmi_register_token;
138
139 ibm_nmi_register_token = rtas_function_token(RTAS_FN_IBM_NMI_REGISTER);
140 if (ibm_nmi_register_token == RTAS_UNKNOWN_SERVICE)
141 return;
142
143 ibm_nmi_interlock_token = rtas_function_token(RTAS_FN_IBM_NMI_INTERLOCK);
144 if (WARN_ON(ibm_nmi_interlock_token == RTAS_UNKNOWN_SERVICE))
145 return;
146
147 /* If the kernel's not linked at zero we point the firmware at low
148 * addresses anyway, and use a trampoline to get to the real code. */
149 system_reset_addr = __pa(system_reset_fwnmi) - PHYSICAL_START;
150 machine_check_addr = __pa(machine_check_fwnmi) - PHYSICAL_START;
151
152 if (0 == rtas_call(ibm_nmi_register_token, 2, 1, NULL,
153 system_reset_addr, machine_check_addr))
154 fwnmi_active = 1;
155
156 /*
157 * Allocate a chunk for per cpu buffer to hold rtas errorlog.
158 * It will be used in real mode mce handler, hence it needs to be
159 * below RMA.
160 */
161 mce_data_buf = memblock_alloc_try_nid_raw(RTAS_ERROR_LOG_MAX * nr_cpus,
162 RTAS_ERROR_LOG_MAX, MEMBLOCK_LOW_LIMIT,
163 ppc64_rma_size, NUMA_NO_NODE);
164 if (!mce_data_buf)
165 panic("Failed to allocate %d bytes below %pa for MCE buffer\n",
166 RTAS_ERROR_LOG_MAX * nr_cpus, &ppc64_rma_size);
167
168 for_each_possible_cpu(i) {
169 paca_ptrs[i]->mce_data_buf = mce_data_buf +
170 (RTAS_ERROR_LOG_MAX * i);
171 }
172
173 #ifdef CONFIG_PPC_64S_HASH_MMU
174 if (!radix_enabled()) {
175 /* Allocate per cpu area to save old slb contents during MCE */
176 size = sizeof(struct slb_entry) * mmu_slb_size * nr_cpus;
177 slb_ptr = memblock_alloc_try_nid_raw(size,
178 sizeof(struct slb_entry), MEMBLOCK_LOW_LIMIT,
179 ppc64_rma_size, NUMA_NO_NODE);
180 if (!slb_ptr)
181 panic("Failed to allocate %zu bytes below %pa for slb area\n",
182 size, &ppc64_rma_size);
183
184 for_each_possible_cpu(i)
185 paca_ptrs[i]->mce_faulty_slbs = slb_ptr + (mmu_slb_size * i);
186 }
187 #endif
188 }
189
190 /*
191 * Affix a device for the first timer to the platform bus if
192 * we have firmware support for the H_WATCHDOG hypercall.
193 */
pseries_wdt_init(void)194 static __init int pseries_wdt_init(void)
195 {
196 if (firmware_has_feature(FW_FEATURE_WATCHDOG))
197 platform_device_register_simple("pseries-wdt", 0, NULL, 0);
198 return 0;
199 }
200 machine_subsys_initcall(pseries, pseries_wdt_init);
201
pseries_8259_cascade(struct irq_desc * desc)202 static void pseries_8259_cascade(struct irq_desc *desc)
203 {
204 struct irq_chip *chip = irq_desc_get_chip(desc);
205 unsigned int cascade_irq = i8259_irq();
206
207 if (cascade_irq)
208 generic_handle_irq(cascade_irq);
209
210 chip->irq_eoi(&desc->irq_data);
211 }
212
pseries_setup_i8259_cascade(void)213 static void __init pseries_setup_i8259_cascade(void)
214 {
215 struct device_node *np, *old, *found = NULL;
216 unsigned int cascade;
217 const u32 *addrp;
218 unsigned long intack = 0;
219 int naddr;
220
221 for_each_node_by_type(np, "interrupt-controller") {
222 if (of_device_is_compatible(np, "chrp,iic")) {
223 found = np;
224 break;
225 }
226 }
227
228 if (found == NULL) {
229 printk(KERN_DEBUG "pic: no ISA interrupt controller\n");
230 return;
231 }
232
233 cascade = irq_of_parse_and_map(found, 0);
234 if (!cascade) {
235 printk(KERN_ERR "pic: failed to map cascade interrupt");
236 return;
237 }
238 pr_debug("pic: cascade mapped to irq %d\n", cascade);
239
240 for (old = of_node_get(found); old != NULL ; old = np) {
241 np = of_get_parent(old);
242 of_node_put(old);
243 if (np == NULL)
244 break;
245 if (!of_node_name_eq(np, "pci"))
246 continue;
247 addrp = of_get_property(np, "8259-interrupt-acknowledge", NULL);
248 if (addrp == NULL)
249 continue;
250 naddr = of_n_addr_cells(np);
251 intack = addrp[naddr-1];
252 if (naddr > 1)
253 intack |= ((unsigned long)addrp[naddr-2]) << 32;
254 }
255 if (intack)
256 printk(KERN_DEBUG "pic: PCI 8259 intack at 0x%016lx\n", intack);
257 i8259_init(found, intack);
258 of_node_put(found);
259 irq_set_chained_handler(cascade, pseries_8259_cascade);
260 }
261
pseries_init_irq(void)262 static void __init pseries_init_irq(void)
263 {
264 /* Try using a XIVE if available, otherwise use a XICS */
265 if (!xive_spapr_init()) {
266 xics_init();
267 pseries_setup_i8259_cascade();
268 }
269 }
270
pseries_lpar_enable_pmcs(void)271 static void pseries_lpar_enable_pmcs(void)
272 {
273 unsigned long set, reset;
274
275 set = 1UL << 63;
276 reset = 0;
277 plpar_hcall_norets(H_PERFMON, set, reset);
278 }
279
pci_dn_reconfig_notifier(struct notifier_block * nb,unsigned long action,void * data)280 static int pci_dn_reconfig_notifier(struct notifier_block *nb, unsigned long action, void *data)
281 {
282 struct of_reconfig_data *rd = data;
283 struct device_node *parent, *np = rd->dn;
284 struct pci_dn *pdn;
285 int err = NOTIFY_OK;
286
287 switch (action) {
288 case OF_RECONFIG_ATTACH_NODE:
289 parent = of_get_parent(np);
290 pdn = parent ? PCI_DN(parent) : NULL;
291 if (pdn)
292 pci_add_device_node_info(pdn->phb, np);
293
294 of_node_put(parent);
295 break;
296 case OF_RECONFIG_DETACH_NODE:
297 pdn = PCI_DN(np);
298 if (pdn)
299 list_del(&pdn->list);
300 break;
301 default:
302 err = NOTIFY_DONE;
303 break;
304 }
305 return err;
306 }
307
308 static struct notifier_block pci_dn_reconfig_nb = {
309 .notifier_call = pci_dn_reconfig_notifier,
310 };
311
312 struct kmem_cache *dtl_cache;
313
314 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
315 /*
316 * Allocate space for the dispatch trace log for all possible cpus
317 * and register the buffers with the hypervisor. This is used for
318 * computing time stolen by the hypervisor.
319 */
alloc_dispatch_logs(void)320 static int alloc_dispatch_logs(void)
321 {
322 if (!firmware_has_feature(FW_FEATURE_SPLPAR))
323 return 0;
324
325 if (!dtl_cache)
326 return 0;
327
328 alloc_dtl_buffers(0);
329
330 /* Register the DTL for the current (boot) cpu */
331 register_dtl_buffer(smp_processor_id());
332
333 return 0;
334 }
335 #else /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
alloc_dispatch_logs(void)336 static inline int alloc_dispatch_logs(void)
337 {
338 return 0;
339 }
340 #endif /* CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
341
alloc_dispatch_log_kmem_cache(void)342 static int alloc_dispatch_log_kmem_cache(void)
343 {
344 void (*ctor)(void *) = get_dtl_cache_ctor();
345
346 dtl_cache = kmem_cache_create_usercopy("dtl", DISPATCH_LOG_BYTES,
347 DISPATCH_LOG_BYTES, 0, 0, DISPATCH_LOG_BYTES, ctor);
348 if (!dtl_cache) {
349 pr_warn("Failed to create dispatch trace log buffer cache\n");
350 pr_warn("Stolen time statistics will be unreliable\n");
351 return 0;
352 }
353
354 return alloc_dispatch_logs();
355 }
356 machine_early_initcall(pseries, alloc_dispatch_log_kmem_cache);
357
358 DEFINE_PER_CPU(u64, idle_spurr_cycles);
359 DEFINE_PER_CPU(u64, idle_entry_purr_snap);
360 DEFINE_PER_CPU(u64, idle_entry_spurr_snap);
pseries_lpar_idle(void)361 static void pseries_lpar_idle(void)
362 {
363 /*
364 * Default handler to go into low thread priority and possibly
365 * low power mode by ceding processor to hypervisor
366 */
367
368 if (!prep_irq_for_idle())
369 return;
370
371 /* Indicate to hypervisor that we are idle. */
372 pseries_idle_prolog();
373
374 /*
375 * Yield the processor to the hypervisor. We return if
376 * an external interrupt occurs (which are driven prior
377 * to returning here) or if a prod occurs from another
378 * processor. When returning here, external interrupts
379 * are enabled.
380 */
381 cede_processor();
382
383 pseries_idle_epilog();
384 }
385
386 static bool pseries_reloc_on_exception_enabled;
387
pseries_reloc_on_exception(void)388 bool pseries_reloc_on_exception(void)
389 {
390 return pseries_reloc_on_exception_enabled;
391 }
392 EXPORT_SYMBOL_GPL(pseries_reloc_on_exception);
393
394 /*
395 * Enable relocation on during exceptions. This has partition wide scope and
396 * may take a while to complete, if it takes longer than one second we will
397 * just give up rather than wasting any more time on this - if that turns out
398 * to ever be a problem in practice we can move this into a kernel thread to
399 * finish off the process later in boot.
400 */
pseries_enable_reloc_on_exc(void)401 bool pseries_enable_reloc_on_exc(void)
402 {
403 long rc;
404 unsigned int delay, total_delay = 0;
405
406 while (1) {
407 rc = enable_reloc_on_exceptions();
408 if (!H_IS_LONG_BUSY(rc)) {
409 if (rc == H_P2) {
410 pr_info("Relocation on exceptions not"
411 " supported\n");
412 return false;
413 } else if (rc != H_SUCCESS) {
414 pr_warn("Unable to enable relocation"
415 " on exceptions: %ld\n", rc);
416 return false;
417 }
418 pseries_reloc_on_exception_enabled = true;
419 return true;
420 }
421
422 delay = get_longbusy_msecs(rc);
423 total_delay += delay;
424 if (total_delay > 1000) {
425 pr_warn("Warning: Giving up waiting to enable "
426 "relocation on exceptions (%u msec)!\n",
427 total_delay);
428 return false;
429 }
430
431 mdelay(delay);
432 }
433 }
434 EXPORT_SYMBOL(pseries_enable_reloc_on_exc);
435
pseries_disable_reloc_on_exc(void)436 void pseries_disable_reloc_on_exc(void)
437 {
438 long rc;
439
440 while (1) {
441 rc = disable_reloc_on_exceptions();
442 if (!H_IS_LONG_BUSY(rc))
443 break;
444 mdelay(get_longbusy_msecs(rc));
445 }
446 if (rc == H_SUCCESS)
447 pseries_reloc_on_exception_enabled = false;
448 else
449 pr_warn("Warning: Failed to disable relocation on exceptions: %ld\n",
450 rc);
451 }
452 EXPORT_SYMBOL(pseries_disable_reloc_on_exc);
453
454 #ifdef __LITTLE_ENDIAN__
pseries_big_endian_exceptions(void)455 void pseries_big_endian_exceptions(void)
456 {
457 long rc;
458
459 while (1) {
460 rc = enable_big_endian_exceptions();
461 if (!H_IS_LONG_BUSY(rc))
462 break;
463 mdelay(get_longbusy_msecs(rc));
464 }
465
466 /*
467 * At this point it is unlikely panic() will get anything
468 * out to the user, since this is called very late in kexec
469 * but at least this will stop us from continuing on further
470 * and creating an even more difficult to debug situation.
471 *
472 * There is a known problem when kdump'ing, if cpus are offline
473 * the above call will fail. Rather than panicking again, keep
474 * going and hope the kdump kernel is also little endian, which
475 * it usually is.
476 */
477 if (rc && !kdump_in_progress())
478 panic("Could not enable big endian exceptions");
479 }
480
pseries_little_endian_exceptions(void)481 void __init pseries_little_endian_exceptions(void)
482 {
483 long rc;
484
485 while (1) {
486 rc = enable_little_endian_exceptions();
487 if (!H_IS_LONG_BUSY(rc))
488 break;
489 mdelay(get_longbusy_msecs(rc));
490 }
491 if (rc) {
492 ppc_md.progress("H_SET_MODE LE exception fail", 0);
493 panic("Could not enable little endian exceptions");
494 }
495 }
496 #endif
497
pSeries_discover_phbs(void)498 static void __init pSeries_discover_phbs(void)
499 {
500 struct device_node *node;
501 struct pci_controller *phb;
502 struct device_node *root = of_find_node_by_path("/");
503
504 for_each_child_of_node(root, node) {
505 if (!of_node_is_type(node, "pci") &&
506 !of_node_is_type(node, "pciex"))
507 continue;
508
509 phb = pcibios_alloc_controller(node);
510 if (!phb)
511 continue;
512 rtas_setup_phb(phb);
513 pci_process_bridge_OF_ranges(phb, node, 0);
514 isa_bridge_find_early(phb);
515 phb->controller_ops = pseries_pci_controller_ops;
516
517 /* create pci_dn's for DT nodes under this PHB */
518 pci_devs_phb_init_dynamic(phb);
519
520 pseries_msi_allocate_domains(phb);
521 }
522
523 of_node_put(root);
524
525 /*
526 * PCI_PROBE_ONLY and PCI_REASSIGN_ALL_BUS can be set via properties
527 * in chosen.
528 */
529 of_pci_check_probe_only();
530 }
531
init_cpu_char_feature_flags(struct h_cpu_char_result * result)532 static void init_cpu_char_feature_flags(struct h_cpu_char_result *result)
533 {
534 /*
535 * The features below are disabled by default, so we instead look to see
536 * if firmware has *enabled* them, and set them if so.
537 */
538 if (result->character & H_CPU_CHAR_SPEC_BAR_ORI31)
539 security_ftr_set(SEC_FTR_SPEC_BAR_ORI31);
540
541 if (result->character & H_CPU_CHAR_BCCTRL_SERIALISED)
542 security_ftr_set(SEC_FTR_BCCTRL_SERIALISED);
543
544 if (result->character & H_CPU_CHAR_L1D_FLUSH_ORI30)
545 security_ftr_set(SEC_FTR_L1D_FLUSH_ORI30);
546
547 if (result->character & H_CPU_CHAR_L1D_FLUSH_TRIG2)
548 security_ftr_set(SEC_FTR_L1D_FLUSH_TRIG2);
549
550 if (result->character & H_CPU_CHAR_L1D_THREAD_PRIV)
551 security_ftr_set(SEC_FTR_L1D_THREAD_PRIV);
552
553 if (result->character & H_CPU_CHAR_COUNT_CACHE_DISABLED)
554 security_ftr_set(SEC_FTR_COUNT_CACHE_DISABLED);
555
556 if (result->character & H_CPU_CHAR_BCCTR_FLUSH_ASSIST)
557 security_ftr_set(SEC_FTR_BCCTR_FLUSH_ASSIST);
558
559 if (result->character & H_CPU_CHAR_BCCTR_LINK_FLUSH_ASSIST)
560 security_ftr_set(SEC_FTR_BCCTR_LINK_FLUSH_ASSIST);
561
562 if (result->behaviour & H_CPU_BEHAV_FLUSH_COUNT_CACHE)
563 security_ftr_set(SEC_FTR_FLUSH_COUNT_CACHE);
564
565 if (result->behaviour & H_CPU_BEHAV_FLUSH_LINK_STACK)
566 security_ftr_set(SEC_FTR_FLUSH_LINK_STACK);
567
568 /*
569 * The features below are enabled by default, so we instead look to see
570 * if firmware has *disabled* them, and clear them if so.
571 * H_CPU_BEHAV_FAVOUR_SECURITY_H could be set only if
572 * H_CPU_BEHAV_FAVOUR_SECURITY is.
573 */
574 if (!(result->behaviour & H_CPU_BEHAV_FAVOUR_SECURITY)) {
575 security_ftr_clear(SEC_FTR_FAVOUR_SECURITY);
576 pseries_security_flavor = 0;
577 } else if (result->behaviour & H_CPU_BEHAV_FAVOUR_SECURITY_H)
578 pseries_security_flavor = 1;
579 else
580 pseries_security_flavor = 2;
581
582 if (!(result->behaviour & H_CPU_BEHAV_L1D_FLUSH_PR))
583 security_ftr_clear(SEC_FTR_L1D_FLUSH_PR);
584
585 if (result->behaviour & H_CPU_BEHAV_NO_L1D_FLUSH_ENTRY)
586 security_ftr_clear(SEC_FTR_L1D_FLUSH_ENTRY);
587
588 if (result->behaviour & H_CPU_BEHAV_NO_L1D_FLUSH_UACCESS)
589 security_ftr_clear(SEC_FTR_L1D_FLUSH_UACCESS);
590
591 if (result->behaviour & H_CPU_BEHAV_NO_STF_BARRIER)
592 security_ftr_clear(SEC_FTR_STF_BARRIER);
593
594 if (!(result->behaviour & H_CPU_BEHAV_BNDS_CHK_SPEC_BAR))
595 security_ftr_clear(SEC_FTR_BNDS_CHK_SPEC_BAR);
596 }
597
pseries_setup_security_mitigations(void)598 void pseries_setup_security_mitigations(void)
599 {
600 struct h_cpu_char_result result;
601 enum l1d_flush_type types;
602 bool enable;
603 long rc;
604
605 /*
606 * Set features to the defaults assumed by init_cpu_char_feature_flags()
607 * so it can set/clear again any features that might have changed after
608 * migration, and in case the hypercall fails and it is not even called.
609 */
610 powerpc_security_features = SEC_FTR_DEFAULT;
611
612 rc = plpar_get_cpu_characteristics(&result);
613 if (rc == H_SUCCESS)
614 init_cpu_char_feature_flags(&result);
615
616 /*
617 * We're the guest so this doesn't apply to us, clear it to simplify
618 * handling of it elsewhere.
619 */
620 security_ftr_clear(SEC_FTR_L1D_FLUSH_HV);
621
622 types = L1D_FLUSH_FALLBACK;
623
624 if (security_ftr_enabled(SEC_FTR_L1D_FLUSH_TRIG2))
625 types |= L1D_FLUSH_MTTRIG;
626
627 if (security_ftr_enabled(SEC_FTR_L1D_FLUSH_ORI30))
628 types |= L1D_FLUSH_ORI;
629
630 enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) && \
631 security_ftr_enabled(SEC_FTR_L1D_FLUSH_PR);
632
633 setup_rfi_flush(types, enable);
634 setup_count_cache_flush();
635
636 enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) &&
637 security_ftr_enabled(SEC_FTR_L1D_FLUSH_ENTRY);
638 setup_entry_flush(enable);
639
640 enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) &&
641 security_ftr_enabled(SEC_FTR_L1D_FLUSH_UACCESS);
642 setup_uaccess_flush(enable);
643
644 setup_stf_barrier();
645 }
646
647 #ifdef CONFIG_PCI_IOV
648 enum rtas_iov_fw_value_map {
649 NUM_RES_PROPERTY = 0, /* Number of Resources */
650 LOW_INT = 1, /* Lowest 32 bits of Address */
651 START_OF_ENTRIES = 2, /* Always start of entry */
652 APERTURE_PROPERTY = 2, /* Start of entry+ to Aperture Size */
653 WDW_SIZE_PROPERTY = 4, /* Start of entry+ to Window Size */
654 NEXT_ENTRY = 7 /* Go to next entry on array */
655 };
656
657 enum get_iov_fw_value_index {
658 BAR_ADDRS = 1, /* Get Bar Address */
659 APERTURE_SIZE = 2, /* Get Aperture Size */
660 WDW_SIZE = 3 /* Get Window Size */
661 };
662
pseries_get_iov_fw_value(struct pci_dev * dev,int resno,enum get_iov_fw_value_index value)663 static resource_size_t pseries_get_iov_fw_value(struct pci_dev *dev, int resno,
664 enum get_iov_fw_value_index value)
665 {
666 const int *indexes;
667 struct device_node *dn = pci_device_to_OF_node(dev);
668 int i, num_res, ret = 0;
669
670 indexes = of_get_property(dn, "ibm,open-sriov-vf-bar-info", NULL);
671 if (!indexes)
672 return 0;
673
674 /*
675 * First element in the array is the number of Bars
676 * returned. Search through the list to find the matching
677 * bar
678 */
679 num_res = of_read_number(&indexes[NUM_RES_PROPERTY], 1);
680 if (resno >= num_res)
681 return 0; /* or an error */
682
683 i = START_OF_ENTRIES + NEXT_ENTRY * resno;
684 switch (value) {
685 case BAR_ADDRS:
686 ret = of_read_number(&indexes[i], 2);
687 break;
688 case APERTURE_SIZE:
689 ret = of_read_number(&indexes[i + APERTURE_PROPERTY], 2);
690 break;
691 case WDW_SIZE:
692 ret = of_read_number(&indexes[i + WDW_SIZE_PROPERTY], 2);
693 break;
694 }
695
696 return ret;
697 }
698
of_pci_set_vf_bar_size(struct pci_dev * dev,const int * indexes)699 static void of_pci_set_vf_bar_size(struct pci_dev *dev, const int *indexes)
700 {
701 struct resource *res;
702 resource_size_t base, size;
703 int i, r, num_res;
704
705 num_res = of_read_number(&indexes[NUM_RES_PROPERTY], 1);
706 num_res = min_t(int, num_res, PCI_SRIOV_NUM_BARS);
707 for (i = START_OF_ENTRIES, r = 0; r < num_res && r < PCI_SRIOV_NUM_BARS;
708 i += NEXT_ENTRY, r++) {
709 res = &dev->resource[r + PCI_IOV_RESOURCES];
710 base = of_read_number(&indexes[i], 2);
711 size = of_read_number(&indexes[i + APERTURE_PROPERTY], 2);
712 res->flags = pci_parse_of_flags(of_read_number
713 (&indexes[i + LOW_INT], 1), 0);
714 res->flags |= (IORESOURCE_MEM_64 | IORESOURCE_PCI_FIXED);
715 res->name = pci_name(dev);
716 res->start = base;
717 res->end = base + size - 1;
718 }
719 }
720
of_pci_parse_iov_addrs(struct pci_dev * dev,const int * indexes)721 static void of_pci_parse_iov_addrs(struct pci_dev *dev, const int *indexes)
722 {
723 struct resource *res, *root, *conflict;
724 resource_size_t base, size;
725 int i, r, num_res;
726
727 /*
728 * First element in the array is the number of Bars
729 * returned. Search through the list to find the matching
730 * bars assign them from firmware into resources structure.
731 */
732 num_res = of_read_number(&indexes[NUM_RES_PROPERTY], 1);
733 for (i = START_OF_ENTRIES, r = 0; r < num_res && r < PCI_SRIOV_NUM_BARS;
734 i += NEXT_ENTRY, r++) {
735 res = &dev->resource[r + PCI_IOV_RESOURCES];
736 base = of_read_number(&indexes[i], 2);
737 size = of_read_number(&indexes[i + WDW_SIZE_PROPERTY], 2);
738 res->name = pci_name(dev);
739 res->start = base;
740 res->end = base + size - 1;
741 root = &iomem_resource;
742 dev_dbg(&dev->dev,
743 "pSeries IOV BAR %d: trying firmware assignment %pR\n",
744 r + PCI_IOV_RESOURCES, res);
745 conflict = request_resource_conflict(root, res);
746 if (conflict) {
747 dev_info(&dev->dev,
748 "BAR %d: %pR conflicts with %s %pR\n",
749 r + PCI_IOV_RESOURCES, res,
750 conflict->name, conflict);
751 res->flags |= IORESOURCE_UNSET;
752 }
753 }
754 }
755
pseries_disable_sriov_resources(struct pci_dev * pdev)756 static void pseries_disable_sriov_resources(struct pci_dev *pdev)
757 {
758 int i;
759
760 pci_warn(pdev, "No hypervisor support for SR-IOV on this device, IOV BARs disabled.\n");
761 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++)
762 pdev->resource[i + PCI_IOV_RESOURCES].flags = 0;
763 }
764
pseries_pci_fixup_resources(struct pci_dev * pdev)765 static void pseries_pci_fixup_resources(struct pci_dev *pdev)
766 {
767 const int *indexes;
768 struct device_node *dn = pci_device_to_OF_node(pdev);
769
770 /*Firmware must support open sriov otherwise dont configure*/
771 indexes = of_get_property(dn, "ibm,open-sriov-vf-bar-info", NULL);
772 if (indexes)
773 of_pci_set_vf_bar_size(pdev, indexes);
774 else
775 pseries_disable_sriov_resources(pdev);
776 }
777
pseries_pci_fixup_iov_resources(struct pci_dev * pdev)778 static void pseries_pci_fixup_iov_resources(struct pci_dev *pdev)
779 {
780 const int *indexes;
781 struct device_node *dn = pci_device_to_OF_node(pdev);
782
783 if (!pdev->is_physfn)
784 return;
785 /*Firmware must support open sriov otherwise don't configure*/
786 indexes = of_get_property(dn, "ibm,open-sriov-vf-bar-info", NULL);
787 if (indexes)
788 of_pci_parse_iov_addrs(pdev, indexes);
789 else
790 pseries_disable_sriov_resources(pdev);
791 }
792
pseries_pci_iov_resource_alignment(struct pci_dev * pdev,int resno)793 static resource_size_t pseries_pci_iov_resource_alignment(struct pci_dev *pdev,
794 int resno)
795 {
796 const __be32 *reg;
797 struct device_node *dn = pci_device_to_OF_node(pdev);
798
799 /*Firmware must support open sriov otherwise report regular alignment*/
800 reg = of_get_property(dn, "ibm,is-open-sriov-pf", NULL);
801 if (!reg)
802 return pci_iov_resource_size(pdev, resno);
803
804 if (!pdev->is_physfn)
805 return 0;
806 return pseries_get_iov_fw_value(pdev,
807 resno - PCI_IOV_RESOURCES,
808 APERTURE_SIZE);
809 }
810 #endif
811
pSeries_setup_arch(void)812 static void __init pSeries_setup_arch(void)
813 {
814 set_arch_panic_timeout(10, ARCH_PANIC_TIMEOUT);
815
816 /* Discover PIC type and setup ppc_md accordingly */
817 smp_init_pseries();
818
819 // Setup CPU hotplug callbacks
820 pseries_cpu_hotplug_init();
821
822 if (radix_enabled() && !mmu_has_feature(MMU_FTR_GTSE))
823 if (!firmware_has_feature(FW_FEATURE_RPT_INVALIDATE))
824 panic("BUG: Radix support requires either GTSE or RPT_INVALIDATE\n");
825
826
827 /* openpic global configuration register (64-bit format). */
828 /* openpic Interrupt Source Unit pointer (64-bit format). */
829 /* python0 facility area (mmio) (64-bit format) REAL address. */
830
831 /* init to some ~sane value until calibrate_delay() runs */
832 loops_per_jiffy = 50000000;
833
834 fwnmi_init();
835
836 pseries_setup_security_mitigations();
837 if (!radix_enabled())
838 pseries_lpar_read_hblkrm_characteristics();
839
840 /* By default, only probe PCI (can be overridden by rtas_pci) */
841 pci_add_flags(PCI_PROBE_ONLY);
842
843 /* Find and initialize PCI host bridges */
844 init_pci_config_tokens();
845 of_reconfig_notifier_register(&pci_dn_reconfig_nb);
846
847 pSeries_nvram_init();
848
849 if (firmware_has_feature(FW_FEATURE_LPAR)) {
850 vpa_init(boot_cpuid);
851
852 if (lppaca_shared_proc()) {
853 static_branch_enable(&shared_processor);
854 pv_spinlocks_init();
855 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
856 static_key_slow_inc(¶virt_steal_enabled);
857 if (steal_acc)
858 static_key_slow_inc(¶virt_steal_rq_enabled);
859 #endif
860 }
861
862 ppc_md.power_save = pseries_lpar_idle;
863 ppc_md.enable_pmcs = pseries_lpar_enable_pmcs;
864 #ifdef CONFIG_PCI_IOV
865 ppc_md.pcibios_fixup_resources =
866 pseries_pci_fixup_resources;
867 ppc_md.pcibios_fixup_sriov =
868 pseries_pci_fixup_iov_resources;
869 ppc_md.pcibios_iov_resource_alignment =
870 pseries_pci_iov_resource_alignment;
871 #endif
872 } else {
873 /* No special idle routine */
874 ppc_md.enable_pmcs = power4_enable_pmcs;
875 }
876
877 ppc_md.pcibios_root_bridge_prepare = pseries_root_bridge_prepare;
878 pseries_rng_init();
879 }
880
pseries_panic(char * str)881 static void pseries_panic(char *str)
882 {
883 panic_flush_kmsg_end();
884 rtas_os_term(str);
885 }
886
pSeries_init_panel(void)887 static int __init pSeries_init_panel(void)
888 {
889 /* Manually leave the kernel version on the panel. */
890 #ifdef __BIG_ENDIAN__
891 ppc_md.progress("Linux ppc64\n", 0);
892 #else
893 ppc_md.progress("Linux ppc64le\n", 0);
894 #endif
895 ppc_md.progress(init_utsname()->version, 0);
896
897 return 0;
898 }
899 machine_arch_initcall(pseries, pSeries_init_panel);
900
pseries_set_dabr(unsigned long dabr,unsigned long dabrx)901 static int pseries_set_dabr(unsigned long dabr, unsigned long dabrx)
902 {
903 return plpar_hcall_norets(H_SET_DABR, dabr);
904 }
905
pseries_set_xdabr(unsigned long dabr,unsigned long dabrx)906 static int pseries_set_xdabr(unsigned long dabr, unsigned long dabrx)
907 {
908 /* Have to set at least one bit in the DABRX according to PAPR */
909 if (dabrx == 0 && dabr == 0)
910 dabrx = DABRX_USER;
911 /* PAPR says we can only set kernel and user bits */
912 dabrx &= DABRX_KERNEL | DABRX_USER;
913
914 return plpar_hcall_norets(H_SET_XDABR, dabr, dabrx);
915 }
916
pseries_set_dawr(int nr,unsigned long dawr,unsigned long dawrx)917 static int pseries_set_dawr(int nr, unsigned long dawr, unsigned long dawrx)
918 {
919 /* PAPR says we can't set HYP */
920 dawrx &= ~DAWRX_HYP;
921
922 if (nr == 0)
923 return plpar_set_watchpoint0(dawr, dawrx);
924 else
925 return plpar_set_watchpoint1(dawr, dawrx);
926 }
927
928 #define CMO_CHARACTERISTICS_TOKEN 44
929 #define CMO_MAXLENGTH 1026
930
pSeries_coalesce_init(void)931 void pSeries_coalesce_init(void)
932 {
933 struct hvcall_mpp_x_data mpp_x_data;
934
935 if (firmware_has_feature(FW_FEATURE_CMO) && !h_get_mpp_x(&mpp_x_data))
936 powerpc_firmware_features |= FW_FEATURE_XCMO;
937 else
938 powerpc_firmware_features &= ~FW_FEATURE_XCMO;
939 }
940
941 /**
942 * fw_cmo_feature_init - FW_FEATURE_CMO is not stored in ibm,hypertas-functions,
943 * handle that here. (Stolen from parse_system_parameter_string)
944 */
pSeries_cmo_feature_init(void)945 static void __init pSeries_cmo_feature_init(void)
946 {
947 static struct papr_sysparm_buf buf __initdata;
948 static_assert(sizeof(buf.val) >= CMO_MAXLENGTH);
949 char *ptr, *key, *value, *end;
950 int page_order = IOMMU_PAGE_SHIFT_4K;
951
952 pr_debug(" -> fw_cmo_feature_init()\n");
953
954 if (papr_sysparm_get(PAPR_SYSPARM_COOP_MEM_OVERCOMMIT_ATTRS, &buf)) {
955 pr_debug("CMO not available\n");
956 pr_debug(" <- fw_cmo_feature_init()\n");
957 return;
958 }
959
960 end = &buf.val[CMO_MAXLENGTH];
961 ptr = &buf.val[0];
962 key = value = ptr;
963
964 while (*ptr && (ptr <= end)) {
965 /* Separate the key and value by replacing '=' with '\0' and
966 * point the value at the string after the '='
967 */
968 if (ptr[0] == '=') {
969 ptr[0] = '\0';
970 value = ptr + 1;
971 } else if (ptr[0] == '\0' || ptr[0] == ',') {
972 /* Terminate the string containing the key/value pair */
973 ptr[0] = '\0';
974
975 if (key == value) {
976 pr_debug("Malformed key/value pair\n");
977 /* Never found a '=', end processing */
978 break;
979 }
980
981 if (0 == strcmp(key, "CMOPageSize"))
982 page_order = simple_strtol(value, NULL, 10);
983 else if (0 == strcmp(key, "PrPSP"))
984 CMO_PrPSP = simple_strtol(value, NULL, 10);
985 else if (0 == strcmp(key, "SecPSP"))
986 CMO_SecPSP = simple_strtol(value, NULL, 10);
987 value = key = ptr + 1;
988 }
989 ptr++;
990 }
991
992 /* Page size is returned as the power of 2 of the page size,
993 * convert to the page size in bytes before returning
994 */
995 CMO_PageSize = 1 << page_order;
996 pr_debug("CMO_PageSize = %lu\n", CMO_PageSize);
997
998 if (CMO_PrPSP != -1 || CMO_SecPSP != -1) {
999 pr_info("CMO enabled\n");
1000 pr_debug("CMO enabled, PrPSP=%d, SecPSP=%d\n", CMO_PrPSP,
1001 CMO_SecPSP);
1002 powerpc_firmware_features |= FW_FEATURE_CMO;
1003 pSeries_coalesce_init();
1004 } else
1005 pr_debug("CMO not enabled, PrPSP=%d, SecPSP=%d\n", CMO_PrPSP,
1006 CMO_SecPSP);
1007 pr_debug(" <- fw_cmo_feature_init()\n");
1008 }
1009
pseries_add_hw_description(void)1010 static void __init pseries_add_hw_description(void)
1011 {
1012 struct device_node *dn;
1013 const char *s;
1014
1015 dn = of_find_node_by_path("/openprom");
1016 if (dn) {
1017 if (of_property_read_string(dn, "model", &s) == 0)
1018 seq_buf_printf(&ppc_hw_desc, "of:%s ", s);
1019
1020 of_node_put(dn);
1021 }
1022
1023 dn = of_find_node_by_path("/hypervisor");
1024 if (dn) {
1025 if (of_property_read_string(dn, "compatible", &s) == 0)
1026 seq_buf_printf(&ppc_hw_desc, "hv:%s ", s);
1027
1028 of_node_put(dn);
1029 return;
1030 }
1031
1032 if (of_property_read_bool(of_root, "ibm,powervm-partition") ||
1033 of_property_read_bool(of_root, "ibm,fw-net-version"))
1034 seq_buf_printf(&ppc_hw_desc, "hv:phyp ");
1035 }
1036
1037 /*
1038 * Early initialization. Relocation is on but do not reference unbolted pages
1039 */
pseries_init(void)1040 static void __init pseries_init(void)
1041 {
1042 pr_debug(" -> pseries_init()\n");
1043
1044 pseries_add_hw_description();
1045
1046 #ifdef CONFIG_HVC_CONSOLE
1047 if (firmware_has_feature(FW_FEATURE_LPAR))
1048 hvc_vio_init_early();
1049 #endif
1050 if (firmware_has_feature(FW_FEATURE_XDABR))
1051 ppc_md.set_dabr = pseries_set_xdabr;
1052 else if (firmware_has_feature(FW_FEATURE_DABR))
1053 ppc_md.set_dabr = pseries_set_dabr;
1054
1055 if (firmware_has_feature(FW_FEATURE_SET_MODE))
1056 ppc_md.set_dawr = pseries_set_dawr;
1057
1058 pSeries_cmo_feature_init();
1059 iommu_init_early_pSeries();
1060
1061 pr_debug(" <- pseries_init()\n");
1062 }
1063
1064 /**
1065 * pseries_power_off - tell firmware about how to power off the system.
1066 *
1067 * This function calls either the power-off rtas token in normal cases
1068 * or the ibm,power-off-ups token (if present & requested) in case of
1069 * a power failure. If power-off token is used, power on will only be
1070 * possible with power button press. If ibm,power-off-ups token is used
1071 * it will allow auto poweron after power is restored.
1072 */
pseries_power_off(void)1073 static void pseries_power_off(void)
1074 {
1075 int rc;
1076 int rtas_poweroff_ups_token = rtas_function_token(RTAS_FN_IBM_POWER_OFF_UPS);
1077
1078 if (rtas_flash_term_hook)
1079 rtas_flash_term_hook(SYS_POWER_OFF);
1080
1081 if (rtas_poweron_auto == 0 ||
1082 rtas_poweroff_ups_token == RTAS_UNKNOWN_SERVICE) {
1083 rc = rtas_call(rtas_function_token(RTAS_FN_POWER_OFF), 2, 1, NULL, -1, -1);
1084 printk(KERN_INFO "RTAS power-off returned %d\n", rc);
1085 } else {
1086 rc = rtas_call(rtas_poweroff_ups_token, 0, 1, NULL);
1087 printk(KERN_INFO "RTAS ibm,power-off-ups returned %d\n", rc);
1088 }
1089 for (;;);
1090 }
1091
pSeries_probe(void)1092 static int __init pSeries_probe(void)
1093 {
1094 if (!of_node_is_type(of_root, "chrp"))
1095 return 0;
1096
1097 /* Cell blades firmware claims to be chrp while it's not. Until this
1098 * is fixed, we need to avoid those here.
1099 */
1100 if (of_machine_is_compatible("IBM,CPBW-1.0") ||
1101 of_machine_is_compatible("IBM,CBEA"))
1102 return 0;
1103
1104 pm_power_off = pseries_power_off;
1105
1106 pr_debug("Machine is%s LPAR !\n",
1107 (powerpc_firmware_features & FW_FEATURE_LPAR) ? "" : " not");
1108
1109 pseries_init();
1110
1111 return 1;
1112 }
1113
pSeries_pci_probe_mode(struct pci_bus * bus)1114 static int pSeries_pci_probe_mode(struct pci_bus *bus)
1115 {
1116 if (firmware_has_feature(FW_FEATURE_LPAR))
1117 return PCI_PROBE_DEVTREE;
1118 return PCI_PROBE_NORMAL;
1119 }
1120
1121 #ifdef CONFIG_MEMORY_HOTPLUG
pseries_memory_block_size(void)1122 static unsigned long pseries_memory_block_size(void)
1123 {
1124 return memory_block_size;
1125 }
1126 #endif
1127
1128 struct pci_controller_ops pseries_pci_controller_ops = {
1129 .probe_mode = pSeries_pci_probe_mode,
1130 #ifdef CONFIG_SPAPR_TCE_IOMMU
1131 .device_group = pSeries_pci_device_group,
1132 #endif
1133 };
1134
define_machine(pseries)1135 define_machine(pseries) {
1136 .name = "pSeries",
1137 .probe = pSeries_probe,
1138 .setup_arch = pSeries_setup_arch,
1139 .init_IRQ = pseries_init_irq,
1140 .show_cpuinfo = pSeries_show_cpuinfo,
1141 .log_error = pSeries_log_error,
1142 .discover_phbs = pSeries_discover_phbs,
1143 .pcibios_fixup = pSeries_final_fixup,
1144 .restart = rtas_restart,
1145 .halt = rtas_halt,
1146 .panic = pseries_panic,
1147 .get_boot_time = rtas_get_boot_time,
1148 .get_rtc_time = rtas_get_rtc_time,
1149 .set_rtc_time = rtas_set_rtc_time,
1150 .progress = rtas_progress,
1151 .system_reset_exception = pSeries_system_reset_exception,
1152 .machine_check_early = pseries_machine_check_realmode,
1153 .machine_check_exception = pSeries_machine_check_exception,
1154 .machine_check_log_err = pSeries_machine_check_log_err,
1155 #ifdef CONFIG_KEXEC_CORE
1156 .kexec_cpu_down = pseries_kexec_cpu_down,
1157 #endif
1158 #ifdef CONFIG_MEMORY_HOTPLUG
1159 .memory_block_size = pseries_memory_block_size,
1160 #endif
1161 };
1162