Update to 3.4-final.
[linux-flexiantxendom0-3.2.10.git] / arch / powerpc / platforms / pseries / setup.c
1 /*
2  *  64-bit pSeries and RS/6000 setup code.
3  *
4  *  Copyright (C) 1995  Linus Torvalds
5  *  Adapted from 'alpha' version by Gary Thomas
6  *  Modified by Cort Dougan (cort@cs.nmt.edu)
7  *  Modified by PPC64 Team, IBM Corp
8  *
9  * This program is free software; you can redistribute it and/or
10  * modify it under the terms of the GNU General Public License
11  * as published by the Free Software Foundation; either version
12  * 2 of the License, or (at your option) any later version.
13  */
14
15 /*
16  * bootup setup stuff..
17  */
18
19 #include <linux/cpu.h>
20 #include <linux/errno.h>
21 #include <linux/sched.h>
22 #include <linux/kernel.h>
23 #include <linux/mm.h>
24 #include <linux/stddef.h>
25 #include <linux/unistd.h>
26 #include <linux/user.h>
27 #include <linux/tty.h>
28 #include <linux/major.h>
29 #include <linux/interrupt.h>
30 #include <linux/reboot.h>
31 #include <linux/init.h>
32 #include <linux/ioport.h>
33 #include <linux/console.h>
34 #include <linux/pci.h>
35 #include <linux/utsname.h>
36 #include <linux/adb.h>
37 #include <linux/export.h>
38 #include <linux/delay.h>
39 #include <linux/irq.h>
40 #include <linux/seq_file.h>
41 #include <linux/root_dev.h>
42 #include <linux/cpuidle.h>
43
44 #include <asm/mmu.h>
45 #include <asm/processor.h>
46 #include <asm/io.h>
47 #include <asm/pgtable.h>
48 #include <asm/prom.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/mpic.h>
59 #include <asm/xics.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/pSeries_reconfig.h>
67
68 #include "plpar_wrappers.h"
69 #include "pseries.h"
70
71 int CMO_PrPSP = -1;
72 int CMO_SecPSP = -1;
73 unsigned long CMO_PageSize = (ASM_CONST(1) << IOMMU_PAGE_SHIFT);
74 EXPORT_SYMBOL(CMO_PageSize);
75
76 int fwnmi_active;  /* TRUE if an FWNMI handler is present */
77
78 static struct device_node *pSeries_mpic_node;
79
80 static void pSeries_show_cpuinfo(struct seq_file *m)
81 {
82         struct device_node *root;
83         const char *model = "";
84
85         root = of_find_node_by_path("/");
86         if (root)
87                 model = of_get_property(root, "model", NULL);
88         seq_printf(m, "machine\t\t: CHRP %s\n", model);
89         of_node_put(root);
90 }
91
92 /* Initialize firmware assisted non-maskable interrupts if
93  * the firmware supports this feature.
94  */
95 static void __init fwnmi_init(void)
96 {
97         unsigned long system_reset_addr, machine_check_addr;
98
99         int ibm_nmi_register = rtas_token("ibm,nmi-register");
100         if (ibm_nmi_register == RTAS_UNKNOWN_SERVICE)
101                 return;
102
103         /* If the kernel's not linked at zero we point the firmware at low
104          * addresses anyway, and use a trampoline to get to the real code. */
105         system_reset_addr  = __pa(system_reset_fwnmi) - PHYSICAL_START;
106         machine_check_addr = __pa(machine_check_fwnmi) - PHYSICAL_START;
107
108         if (0 == rtas_call(ibm_nmi_register, 2, 1, NULL, system_reset_addr,
109                                 machine_check_addr))
110                 fwnmi_active = 1;
111 }
112
113 static void pseries_8259_cascade(unsigned int irq, struct irq_desc *desc)
114 {
115         struct irq_chip *chip = irq_desc_get_chip(desc);
116         unsigned int cascade_irq = i8259_irq();
117
118         if (cascade_irq != NO_IRQ)
119                 generic_handle_irq(cascade_irq);
120
121         chip->irq_eoi(&desc->irq_data);
122 }
123
124 static void __init pseries_setup_i8259_cascade(void)
125 {
126         struct device_node *np, *old, *found = NULL;
127         unsigned int cascade;
128         const u32 *addrp;
129         unsigned long intack = 0;
130         int naddr;
131
132         for_each_node_by_type(np, "interrupt-controller") {
133                 if (of_device_is_compatible(np, "chrp,iic")) {
134                         found = np;
135                         break;
136                 }
137         }
138
139         if (found == NULL) {
140                 printk(KERN_DEBUG "pic: no ISA interrupt controller\n");
141                 return;
142         }
143
144         cascade = irq_of_parse_and_map(found, 0);
145         if (cascade == NO_IRQ) {
146                 printk(KERN_ERR "pic: failed to map cascade interrupt");
147                 return;
148         }
149         pr_debug("pic: cascade mapped to irq %d\n", cascade);
150
151         for (old = of_node_get(found); old != NULL ; old = np) {
152                 np = of_get_parent(old);
153                 of_node_put(old);
154                 if (np == NULL)
155                         break;
156                 if (strcmp(np->name, "pci") != 0)
157                         continue;
158                 addrp = of_get_property(np, "8259-interrupt-acknowledge", NULL);
159                 if (addrp == NULL)
160                         continue;
161                 naddr = of_n_addr_cells(np);
162                 intack = addrp[naddr-1];
163                 if (naddr > 1)
164                         intack |= ((unsigned long)addrp[naddr-2]) << 32;
165         }
166         if (intack)
167                 printk(KERN_DEBUG "pic: PCI 8259 intack at 0x%016lx\n", intack);
168         i8259_init(found, intack);
169         of_node_put(found);
170         irq_set_chained_handler(cascade, pseries_8259_cascade);
171 }
172
173 static void __init pseries_mpic_init_IRQ(void)
174 {
175         struct device_node *np;
176         const unsigned int *opprop;
177         unsigned long openpic_addr = 0;
178         int naddr, n, i, opplen;
179         struct mpic *mpic;
180
181         np = of_find_node_by_path("/");
182         naddr = of_n_addr_cells(np);
183         opprop = of_get_property(np, "platform-open-pic", &opplen);
184         if (opprop != 0) {
185                 openpic_addr = of_read_number(opprop, naddr);
186                 printk(KERN_DEBUG "OpenPIC addr: %lx\n", openpic_addr);
187         }
188         of_node_put(np);
189
190         BUG_ON(openpic_addr == 0);
191
192         /* Setup the openpic driver */
193         mpic = mpic_alloc(pSeries_mpic_node, openpic_addr,
194                         MPIC_NO_RESET, 16, 0, " MPIC     ");
195         BUG_ON(mpic == NULL);
196
197         /* Add ISUs */
198         opplen /= sizeof(u32);
199         for (n = 0, i = naddr; i < opplen; i += naddr, n++) {
200                 unsigned long isuaddr = of_read_number(opprop + i, naddr);
201                 mpic_assign_isu(mpic, n, isuaddr);
202         }
203
204         /* Setup top-level get_irq */
205         ppc_md.get_irq = mpic_get_irq;
206
207         /* All ISUs are setup, complete initialization */
208         mpic_init(mpic);
209
210         /* Look for cascade */
211         pseries_setup_i8259_cascade();
212 }
213
214 static void __init pseries_xics_init_IRQ(void)
215 {
216         xics_init();
217         pseries_setup_i8259_cascade();
218 }
219
220 static void pseries_lpar_enable_pmcs(void)
221 {
222         unsigned long set, reset;
223
224         set = 1UL << 63;
225         reset = 0;
226         plpar_hcall_norets(H_PERFMON, set, reset);
227 }
228
229 static void __init pseries_discover_pic(void)
230 {
231         struct device_node *np;
232         const char *typep;
233
234         for (np = NULL; (np = of_find_node_by_name(np,
235                                                    "interrupt-controller"));) {
236                 typep = of_get_property(np, "compatible", NULL);
237                 if (strstr(typep, "open-pic")) {
238                         pSeries_mpic_node = of_node_get(np);
239                         ppc_md.init_IRQ       = pseries_mpic_init_IRQ;
240                         setup_kexec_cpu_down_mpic();
241                         smp_init_pseries_mpic();
242                         return;
243                 } else if (strstr(typep, "ppc-xicp")) {
244                         ppc_md.init_IRQ       = pseries_xics_init_IRQ;
245                         setup_kexec_cpu_down_xics();
246                         smp_init_pseries_xics();
247                         return;
248                 }
249         }
250         printk(KERN_ERR "pSeries_discover_pic: failed to recognize"
251                " interrupt-controller\n");
252 }
253
254 static int pci_dn_reconfig_notifier(struct notifier_block *nb, unsigned long action, void *node)
255 {
256         struct device_node *np = node;
257         struct pci_dn *pci = NULL;
258         int err = NOTIFY_OK;
259
260         switch (action) {
261         case PSERIES_RECONFIG_ADD:
262                 pci = np->parent->data;
263                 if (pci) {
264                         update_dn_pci_info(np, pci->phb);
265
266                         /* Create EEH device for the OF node */
267                         eeh_dev_init(np, pci->phb);
268                 }
269                 break;
270         default:
271                 err = NOTIFY_DONE;
272                 break;
273         }
274         return err;
275 }
276
277 static struct notifier_block pci_dn_reconfig_nb = {
278         .notifier_call = pci_dn_reconfig_notifier,
279 };
280
281 struct kmem_cache *dtl_cache;
282
283 #ifdef CONFIG_VIRT_CPU_ACCOUNTING
284 /*
285  * Allocate space for the dispatch trace log for all possible cpus
286  * and register the buffers with the hypervisor.  This is used for
287  * computing time stolen by the hypervisor.
288  */
289 static int alloc_dispatch_logs(void)
290 {
291         int cpu, ret;
292         struct paca_struct *pp;
293         struct dtl_entry *dtl;
294
295         if (!firmware_has_feature(FW_FEATURE_SPLPAR))
296                 return 0;
297
298         if (!dtl_cache)
299                 return 0;
300
301         for_each_possible_cpu(cpu) {
302                 pp = &paca[cpu];
303                 dtl = kmem_cache_alloc(dtl_cache, GFP_KERNEL);
304                 if (!dtl) {
305                         pr_warn("Failed to allocate dispatch trace log for cpu %d\n",
306                                 cpu);
307                         pr_warn("Stolen time statistics will be unreliable\n");
308                         break;
309                 }
310
311                 pp->dtl_ridx = 0;
312                 pp->dispatch_log = dtl;
313                 pp->dispatch_log_end = dtl + N_DISPATCH_LOG;
314                 pp->dtl_curr = dtl;
315         }
316
317         /* Register the DTL for the current (boot) cpu */
318         dtl = get_paca()->dispatch_log;
319         get_paca()->dtl_ridx = 0;
320         get_paca()->dtl_curr = dtl;
321         get_paca()->lppaca_ptr->dtl_idx = 0;
322
323         /* hypervisor reads buffer length from this field */
324         dtl->enqueue_to_dispatch_time = DISPATCH_LOG_BYTES;
325         ret = register_dtl(hard_smp_processor_id(), __pa(dtl));
326         if (ret)
327                 pr_err("WARNING: DTL registration of cpu %d (hw %d) failed "
328                        "with %d\n", smp_processor_id(),
329                        hard_smp_processor_id(), ret);
330         get_paca()->lppaca_ptr->dtl_enable_mask = 2;
331
332         return 0;
333 }
334 #else /* !CONFIG_VIRT_CPU_ACCOUNTING */
335 static inline int alloc_dispatch_logs(void)
336 {
337         return 0;
338 }
339 #endif /* CONFIG_VIRT_CPU_ACCOUNTING */
340
341 static int alloc_dispatch_log_kmem_cache(void)
342 {
343         dtl_cache = kmem_cache_create("dtl", DISPATCH_LOG_BYTES,
344                                                 DISPATCH_LOG_BYTES, 0, NULL);
345         if (!dtl_cache) {
346                 pr_warn("Failed to create dispatch trace log buffer cache\n");
347                 pr_warn("Stolen time statistics will be unreliable\n");
348                 return 0;
349         }
350
351         return alloc_dispatch_logs();
352 }
353 early_initcall(alloc_dispatch_log_kmem_cache);
354
355 static void pSeries_idle(void)
356 {
357         /* This would call on the cpuidle framework, and the back-end pseries
358          * driver to  go to idle states
359          */
360         if (cpuidle_idle_call()) {
361                 /* On error, execute default handler
362                  * to go into low thread priority and possibly
363                  * low power mode.
364                  */
365                 HMT_low();
366                 HMT_very_low();
367         }
368 }
369
370 static void __init pSeries_setup_arch(void)
371 {
372         panic_timeout = 10;
373
374         /* Discover PIC type and setup ppc_md accordingly */
375         pseries_discover_pic();
376
377         /* openpic global configuration register (64-bit format). */
378         /* openpic Interrupt Source Unit pointer (64-bit format). */
379         /* python0 facility area (mmio) (64-bit format) REAL address. */
380
381         /* init to some ~sane value until calibrate_delay() runs */
382         loops_per_jiffy = 50000000;
383
384         fwnmi_init();
385
386         /* By default, only probe PCI (can be overriden by rtas_pci) */
387         pci_add_flags(PCI_PROBE_ONLY);
388
389         /* Find and initialize PCI host bridges */
390         init_pci_config_tokens();
391         eeh_pseries_init();
392         find_and_init_phbs();
393         pSeries_reconfig_notifier_register(&pci_dn_reconfig_nb);
394         eeh_init();
395
396         pSeries_nvram_init();
397
398         if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
399                 vpa_init(boot_cpuid);
400                 ppc_md.power_save = pSeries_idle;
401         }
402
403         if (firmware_has_feature(FW_FEATURE_LPAR))
404                 ppc_md.enable_pmcs = pseries_lpar_enable_pmcs;
405         else
406                 ppc_md.enable_pmcs = power4_enable_pmcs;
407 }
408
409 static int __init pSeries_init_panel(void)
410 {
411         /* Manually leave the kernel version on the panel. */
412         ppc_md.progress("SUSE Linux\n", 0);
413         ppc_md.progress(init_utsname()->version, 0);
414
415         return 0;
416 }
417 machine_arch_initcall(pseries, pSeries_init_panel);
418
419 static int pseries_set_dabr(unsigned long dabr)
420 {
421         return plpar_hcall_norets(H_SET_DABR, dabr);
422 }
423
424 static int pseries_set_xdabr(unsigned long dabr)
425 {
426         /* We want to catch accesses from kernel and userspace */
427         return plpar_hcall_norets(H_SET_XDABR, dabr,
428                         H_DABRX_KERNEL | H_DABRX_USER);
429 }
430
431 #define CMO_CHARACTERISTICS_TOKEN 44
432 #define CMO_MAXLENGTH 1026
433
434 void pSeries_coalesce_init(void)
435 {
436         struct hvcall_mpp_x_data mpp_x_data;
437
438         if (firmware_has_feature(FW_FEATURE_CMO) && !h_get_mpp_x(&mpp_x_data))
439                 powerpc_firmware_features |= FW_FEATURE_XCMO;
440         else
441                 powerpc_firmware_features &= ~FW_FEATURE_XCMO;
442 }
443
444 /**
445  * fw_cmo_feature_init - FW_FEATURE_CMO is not stored in ibm,hypertas-functions,
446  * handle that here. (Stolen from parse_system_parameter_string)
447  */
448 void pSeries_cmo_feature_init(void)
449 {
450         char *ptr, *key, *value, *end;
451         int call_status;
452         int page_order = IOMMU_PAGE_SHIFT;
453
454         pr_debug(" -> fw_cmo_feature_init()\n");
455         spin_lock(&rtas_data_buf_lock);
456         memset(rtas_data_buf, 0, RTAS_DATA_BUF_SIZE);
457         call_status = rtas_call(rtas_token("ibm,get-system-parameter"), 3, 1,
458                                 NULL,
459                                 CMO_CHARACTERISTICS_TOKEN,
460                                 __pa(rtas_data_buf),
461                                 RTAS_DATA_BUF_SIZE);
462
463         if (call_status != 0) {
464                 spin_unlock(&rtas_data_buf_lock);
465                 pr_debug("CMO not available\n");
466                 pr_debug(" <- fw_cmo_feature_init()\n");
467                 return;
468         }
469
470         end = rtas_data_buf + CMO_MAXLENGTH - 2;
471         ptr = rtas_data_buf + 2;        /* step over strlen value */
472         key = value = ptr;
473
474         while (*ptr && (ptr <= end)) {
475                 /* Separate the key and value by replacing '=' with '\0' and
476                  * point the value at the string after the '='
477                  */
478                 if (ptr[0] == '=') {
479                         ptr[0] = '\0';
480                         value = ptr + 1;
481                 } else if (ptr[0] == '\0' || ptr[0] == ',') {
482                         /* Terminate the string containing the key/value pair */
483                         ptr[0] = '\0';
484
485                         if (key == value) {
486                                 pr_debug("Malformed key/value pair\n");
487                                 /* Never found a '=', end processing */
488                                 break;
489                         }
490
491                         if (0 == strcmp(key, "CMOPageSize"))
492                                 page_order = simple_strtol(value, NULL, 10);
493                         else if (0 == strcmp(key, "PrPSP"))
494                                 CMO_PrPSP = simple_strtol(value, NULL, 10);
495                         else if (0 == strcmp(key, "SecPSP"))
496                                 CMO_SecPSP = simple_strtol(value, NULL, 10);
497                         value = key = ptr + 1;
498                 }
499                 ptr++;
500         }
501
502         /* Page size is returned as the power of 2 of the page size,
503          * convert to the page size in bytes before returning
504          */
505         CMO_PageSize = 1 << page_order;
506         pr_debug("CMO_PageSize = %lu\n", CMO_PageSize);
507
508         if (CMO_PrPSP != -1 || CMO_SecPSP != -1) {
509                 pr_info("CMO enabled\n");
510                 pr_debug("CMO enabled, PrPSP=%d, SecPSP=%d\n", CMO_PrPSP,
511                          CMO_SecPSP);
512                 powerpc_firmware_features |= FW_FEATURE_CMO;
513                 pSeries_coalesce_init();
514         } else
515                 pr_debug("CMO not enabled, PrPSP=%d, SecPSP=%d\n", CMO_PrPSP,
516                          CMO_SecPSP);
517         spin_unlock(&rtas_data_buf_lock);
518         pr_debug(" <- fw_cmo_feature_init()\n");
519 }
520
521 /*
522  * Early initialization.  Relocation is on but do not reference unbolted pages
523  */
524 static void __init pSeries_init_early(void)
525 {
526         pr_debug(" -> pSeries_init_early()\n");
527
528 #ifdef CONFIG_HVC_CONSOLE
529         if (firmware_has_feature(FW_FEATURE_LPAR))
530                 hvc_vio_init_early();
531 #endif
532         if (firmware_has_feature(FW_FEATURE_DABR))
533                 ppc_md.set_dabr = pseries_set_dabr;
534         else if (firmware_has_feature(FW_FEATURE_XDABR))
535                 ppc_md.set_dabr = pseries_set_xdabr;
536
537         pSeries_cmo_feature_init();
538         iommu_init_early_pSeries();
539
540         pr_debug(" <- pSeries_init_early()\n");
541 }
542
543 /*
544  * Called very early, MMU is off, device-tree isn't unflattened
545  */
546
547 static int __init pSeries_probe_hypertas(unsigned long node,
548                                          const char *uname, int depth,
549                                          void *data)
550 {
551         const char *hypertas;
552         unsigned long len;
553
554         if (depth != 1 ||
555             (strcmp(uname, "rtas") != 0 && strcmp(uname, "rtas@0") != 0))
556                 return 0;
557
558         hypertas = of_get_flat_dt_prop(node, "ibm,hypertas-functions", &len);
559         if (!hypertas)
560                 return 1;
561
562         powerpc_firmware_features |= FW_FEATURE_LPAR;
563         fw_feature_init(hypertas, len);
564
565         return 1;
566 }
567
568 static int __init pSeries_probe(void)
569 {
570         unsigned long root = of_get_flat_dt_root();
571         char *dtype = of_get_flat_dt_prop(root, "device_type", NULL);
572
573         if (dtype == NULL)
574                 return 0;
575         if (strcmp(dtype, "chrp"))
576                 return 0;
577
578         /* Cell blades firmware claims to be chrp while it's not. Until this
579          * is fixed, we need to avoid those here.
580          */
581         if (of_flat_dt_is_compatible(root, "IBM,CPBW-1.0") ||
582             of_flat_dt_is_compatible(root, "IBM,CBEA"))
583                 return 0;
584
585         pr_debug("pSeries detected, looking for LPAR capability...\n");
586
587         /* Now try to figure out if we are running on LPAR */
588         of_scan_flat_dt(pSeries_probe_hypertas, NULL);
589
590         if (firmware_has_feature(FW_FEATURE_LPAR))
591                 hpte_init_lpar();
592         else
593                 hpte_init_native();
594
595         pr_debug("Machine is%s LPAR !\n",
596                  (powerpc_firmware_features & FW_FEATURE_LPAR) ? "" : " not");
597
598         return 1;
599 }
600
601 static int pSeries_pci_probe_mode(struct pci_bus *bus)
602 {
603         if (firmware_has_feature(FW_FEATURE_LPAR))
604                 return PCI_PROBE_DEVTREE;
605         return PCI_PROBE_NORMAL;
606 }
607
608 /**
609  * pSeries_power_off - tell firmware about how to power off the system.
610  *
611  * This function calls either the power-off rtas token in normal cases
612  * or the ibm,power-off-ups token (if present & requested) in case of
613  * a power failure. If power-off token is used, power on will only be
614  * possible with power button press. If ibm,power-off-ups token is used
615  * it will allow auto poweron after power is restored.
616  */
617 static void pSeries_power_off(void)
618 {
619         int rc;
620         int rtas_poweroff_ups_token = rtas_token("ibm,power-off-ups");
621
622         if (rtas_flash_term_hook)
623                 rtas_flash_term_hook(SYS_POWER_OFF);
624
625         if (rtas_poweron_auto == 0 ||
626                 rtas_poweroff_ups_token == RTAS_UNKNOWN_SERVICE) {
627                 rc = rtas_call(rtas_token("power-off"), 2, 1, NULL, -1, -1);
628                 printk(KERN_INFO "RTAS power-off returned %d\n", rc);
629         } else {
630                 rc = rtas_call(rtas_poweroff_ups_token, 0, 1, NULL);
631                 printk(KERN_INFO "RTAS ibm,power-off-ups returned %d\n", rc);
632         }
633         for (;;);
634 }
635
636 #ifndef CONFIG_PCI
637 void pSeries_final_fixup(void) { }
638 #endif
639
640 define_machine(pseries) {
641         .name                   = "pSeries",
642         .probe                  = pSeries_probe,
643         .setup_arch             = pSeries_setup_arch,
644         .init_early             = pSeries_init_early,
645         .show_cpuinfo           = pSeries_show_cpuinfo,
646         .log_error              = pSeries_log_error,
647         .pcibios_fixup          = pSeries_final_fixup,
648         .pci_probe_mode         = pSeries_pci_probe_mode,
649         .restart                = rtas_restart,
650         .power_off              = pSeries_power_off,
651         .halt                   = rtas_halt,
652         .panic                  = rtas_os_term,
653         .get_boot_time          = rtas_get_boot_time,
654         .get_rtc_time           = rtas_get_rtc_time,
655         .set_rtc_time           = rtas_set_rtc_time,
656         .calibrate_decr         = generic_calibrate_decr,
657         .progress               = rtas_progress,
658         .system_reset_exception = pSeries_system_reset_exception,
659         .machine_check_exception = pSeries_machine_check_exception,
660 };