Merge branch 'master' of /home/src/linux-2.6/
[linux-2.6] / 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 #undef DEBUG
20
21 #include <linux/config.h>
22 #include <linux/cpu.h>
23 #include <linux/errno.h>
24 #include <linux/sched.h>
25 #include <linux/kernel.h>
26 #include <linux/mm.h>
27 #include <linux/stddef.h>
28 #include <linux/unistd.h>
29 #include <linux/slab.h>
30 #include <linux/user.h>
31 #include <linux/a.out.h>
32 #include <linux/tty.h>
33 #include <linux/major.h>
34 #include <linux/interrupt.h>
35 #include <linux/reboot.h>
36 #include <linux/init.h>
37 #include <linux/ioport.h>
38 #include <linux/console.h>
39 #include <linux/pci.h>
40 #include <linux/utsname.h>
41 #include <linux/adb.h>
42 #include <linux/module.h>
43 #include <linux/delay.h>
44 #include <linux/irq.h>
45 #include <linux/seq_file.h>
46 #include <linux/root_dev.h>
47
48 #include <asm/mmu.h>
49 #include <asm/processor.h>
50 #include <asm/io.h>
51 #include <asm/pgtable.h>
52 #include <asm/prom.h>
53 #include <asm/rtas.h>
54 #include <asm/pci-bridge.h>
55 #include <asm/iommu.h>
56 #include <asm/dma.h>
57 #include <asm/machdep.h>
58 #include <asm/irq.h>
59 #include <asm/kexec.h>
60 #include <asm/time.h>
61 #include <asm/nvram.h>
62 #include "xics.h"
63 #include <asm/firmware.h>
64 #include <asm/pmc.h>
65 #include <asm/mpic.h>
66 #include <asm/ppc-pci.h>
67 #include <asm/i8259.h>
68 #include <asm/udbg.h>
69 #include <asm/smp.h>
70
71 #include "plpar_wrappers.h"
72 #include "ras.h"
73
74 #ifdef DEBUG
75 #define DBG(fmt...) udbg_printf(fmt)
76 #else
77 #define DBG(fmt...)
78 #endif
79
80 extern void find_udbg_vterm(void);
81
82 int fwnmi_active;  /* TRUE if an FWNMI handler is present */
83
84 static void pseries_shared_idle(void);
85 static void pseries_dedicated_idle(void);
86
87 struct mpic *pSeries_mpic;
88
89 static void pSeries_show_cpuinfo(struct seq_file *m)
90 {
91         struct device_node *root;
92         const char *model = "";
93
94         root = of_find_node_by_path("/");
95         if (root)
96                 model = get_property(root, "model", NULL);
97         seq_printf(m, "machine\t\t: CHRP %s\n", model);
98         of_node_put(root);
99 }
100
101 /* Initialize firmware assisted non-maskable interrupts if
102  * the firmware supports this feature.
103  */
104 static void __init fwnmi_init(void)
105 {
106         unsigned long system_reset_addr, machine_check_addr;
107
108         int ibm_nmi_register = rtas_token("ibm,nmi-register");
109         if (ibm_nmi_register == RTAS_UNKNOWN_SERVICE)
110                 return;
111
112         /* If the kernel's not linked at zero we point the firmware at low
113          * addresses anyway, and use a trampoline to get to the real code. */
114         system_reset_addr  = __pa(system_reset_fwnmi) - PHYSICAL_START;
115         machine_check_addr = __pa(machine_check_fwnmi) - PHYSICAL_START;
116
117         if (0 == rtas_call(ibm_nmi_register, 2, 1, NULL, system_reset_addr,
118                                 machine_check_addr))
119                 fwnmi_active = 1;
120 }
121
122 static void __init pSeries_init_mpic(void)
123 {
124         unsigned int *addrp;
125         struct device_node *np;
126         unsigned long intack = 0;
127
128         /* All ISUs are setup, complete initialization */
129         mpic_init(pSeries_mpic);
130
131         /* Check what kind of cascade ACK we have */
132         if (!(np = of_find_node_by_name(NULL, "pci"))
133             || !(addrp = (unsigned int *)
134                  get_property(np, "8259-interrupt-acknowledge", NULL)))
135                 printk(KERN_ERR "Cannot find pci to get ack address\n");
136         else
137                 intack = addrp[prom_n_addr_cells(np)-1];
138         of_node_put(np);
139
140         /* Setup the legacy interrupts & controller */
141         i8259_init(intack, 0);
142
143         /* Hook cascade to mpic */
144         mpic_setup_cascade(NUM_ISA_INTERRUPTS, i8259_irq_cascade, NULL);
145 }
146
147 static void __init pSeries_setup_mpic(void)
148 {
149         unsigned int *opprop;
150         unsigned long openpic_addr = 0;
151         unsigned char senses[NR_IRQS - NUM_ISA_INTERRUPTS];
152         struct device_node *root;
153         int irq_count;
154
155         /* Find the Open PIC if present */
156         root = of_find_node_by_path("/");
157         opprop = (unsigned int *) get_property(root, "platform-open-pic", NULL);
158         if (opprop != 0) {
159                 int n = prom_n_addr_cells(root);
160
161                 for (openpic_addr = 0; n > 0; --n)
162                         openpic_addr = (openpic_addr << 32) + *opprop++;
163                 printk(KERN_DEBUG "OpenPIC addr: %lx\n", openpic_addr);
164         }
165         of_node_put(root);
166
167         BUG_ON(openpic_addr == 0);
168
169         /* Get the sense values from OF */
170         prom_get_irq_senses(senses, NUM_ISA_INTERRUPTS, NR_IRQS);
171         
172         /* Setup the openpic driver */
173         irq_count = NR_IRQS - NUM_ISA_INTERRUPTS - 4; /* leave room for IPIs */
174         pSeries_mpic = mpic_alloc(openpic_addr, MPIC_PRIMARY,
175                                   16, 16, irq_count, /* isu size, irq offset, irq count */ 
176                                   NR_IRQS - 4, /* ipi offset */
177                                   senses, irq_count, /* sense & sense size */
178                                   " MPIC     ");
179 }
180
181 static void pseries_lpar_enable_pmcs(void)
182 {
183         unsigned long set, reset;
184
185         power4_enable_pmcs();
186
187         set = 1UL << 63;
188         reset = 0;
189         plpar_hcall_norets(H_PERFMON, set, reset);
190
191         /* instruct hypervisor to maintain PMCs */
192         if (firmware_has_feature(FW_FEATURE_SPLPAR))
193                 get_lppaca()->pmcregs_in_use = 1;
194 }
195
196 static void __init pSeries_setup_arch(void)
197 {
198         /* Fixup ppc_md depending on the type of interrupt controller */
199         if (ppc64_interrupt_controller == IC_OPEN_PIC) {
200                 ppc_md.init_IRQ       = pSeries_init_mpic;
201                 ppc_md.get_irq        = mpic_get_irq;
202                 /* Allocate the mpic now, so that find_and_init_phbs() can
203                  * fill the ISUs */
204                 pSeries_setup_mpic();
205         } else {
206                 ppc_md.init_IRQ       = xics_init_IRQ;
207                 ppc_md.get_irq        = xics_get_irq;
208         }
209
210 #ifdef CONFIG_SMP
211         smp_init_pSeries();
212 #endif
213         /* openpic global configuration register (64-bit format). */
214         /* openpic Interrupt Source Unit pointer (64-bit format). */
215         /* python0 facility area (mmio) (64-bit format) REAL address. */
216
217         /* init to some ~sane value until calibrate_delay() runs */
218         loops_per_jiffy = 50000000;
219
220         if (ROOT_DEV == 0) {
221                 printk("No ramdisk, default root is /dev/sda2\n");
222                 ROOT_DEV = Root_SDA2;
223         }
224
225         fwnmi_init();
226
227         /* Find and initialize PCI host bridges */
228         init_pci_config_tokens();
229         find_and_init_phbs();
230         eeh_init();
231
232         pSeries_nvram_init();
233
234         /* Choose an idle loop */
235         if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
236                 vpa_init(boot_cpuid);
237                 if (get_lppaca()->shared_proc) {
238                         printk(KERN_INFO "Using shared processor idle loop\n");
239                         ppc_md.idle_loop = pseries_shared_idle;
240                 } else {
241                         printk(KERN_INFO "Using dedicated idle loop\n");
242                         ppc_md.idle_loop = pseries_dedicated_idle;
243                 }
244         } else {
245                 printk(KERN_INFO "Using default idle loop\n");
246                 ppc_md.idle_loop = default_idle;
247         }
248
249         if (platform_is_lpar())
250                 ppc_md.enable_pmcs = pseries_lpar_enable_pmcs;
251         else
252                 ppc_md.enable_pmcs = power4_enable_pmcs;
253 }
254
255 static int __init pSeries_init_panel(void)
256 {
257         /* Manually leave the kernel version on the panel. */
258         ppc_md.progress("Linux ppc64\n", 0);
259         ppc_md.progress(system_utsname.version, 0);
260
261         return 0;
262 }
263 arch_initcall(pSeries_init_panel);
264
265
266 /* Build up the ppc64_firmware_features bitmask field
267  * using contents of device-tree/ibm,hypertas-functions.
268  * Ultimately this functionality may be moved into prom.c prom_init().
269  */
270 static void __init fw_feature_init(void)
271 {
272         struct device_node * dn;
273         char * hypertas;
274         unsigned int len;
275
276         DBG(" -> fw_feature_init()\n");
277
278         ppc64_firmware_features = 0;
279         dn = of_find_node_by_path("/rtas");
280         if (dn == NULL) {
281                 printk(KERN_ERR "WARNING ! Cannot find RTAS in device-tree !\n");
282                 goto no_rtas;
283         }
284
285         hypertas = get_property(dn, "ibm,hypertas-functions", &len);
286         if (hypertas) {
287                 while (len > 0){
288                         int i, hypertas_len;
289                         /* check value against table of strings */
290                         for(i=0; i < FIRMWARE_MAX_FEATURES ;i++) {
291                                 if ((firmware_features_table[i].name) &&
292                                     (strcmp(firmware_features_table[i].name,hypertas))==0) {
293                                         /* we have a match */
294                                         ppc64_firmware_features |= 
295                                                 (firmware_features_table[i].val);
296                                         break;
297                                 } 
298                         }
299                         hypertas_len = strlen(hypertas);
300                         len -= hypertas_len +1;
301                         hypertas+= hypertas_len +1;
302                 }
303         }
304
305         of_node_put(dn);
306 no_rtas:
307
308         DBG(" <- fw_feature_init()\n");
309 }
310
311
312 static  void __init pSeries_discover_pic(void)
313 {
314         struct device_node *np;
315         char *typep;
316
317         /*
318          * Setup interrupt mapping options that are needed for finish_device_tree
319          * to properly parse the OF interrupt tree & do the virtual irq mapping
320          */
321         __irq_offset_value = NUM_ISA_INTERRUPTS;
322         ppc64_interrupt_controller = IC_INVALID;
323         for (np = NULL; (np = of_find_node_by_name(np, "interrupt-controller"));) {
324                 typep = (char *)get_property(np, "compatible", NULL);
325                 if (strstr(typep, "open-pic")) {
326                         ppc64_interrupt_controller = IC_OPEN_PIC;
327                         break;
328                 } else if (strstr(typep, "ppc-xicp")) {
329                         ppc64_interrupt_controller = IC_PPC_XIC;
330                         break;
331                 }
332         }
333         if (ppc64_interrupt_controller == IC_INVALID)
334                 printk("pSeries_discover_pic: failed to recognize"
335                         " interrupt-controller\n");
336
337 }
338
339 static void pSeries_mach_cpu_die(void)
340 {
341         local_irq_disable();
342         idle_task_exit();
343         /* Some hardware requires clearing the CPPR, while other hardware does not
344          * it is safe either way
345          */
346         pSeriesLP_cppr_info(0, 0);
347         rtas_stop_self();
348         /* Should never get here... */
349         BUG();
350         for(;;);
351 }
352
353 static int pseries_set_dabr(unsigned long dabr)
354 {
355         return plpar_hcall_norets(H_SET_DABR, dabr);
356 }
357
358 static int pseries_set_xdabr(unsigned long dabr)
359 {
360         /* We want to catch accesses from kernel and userspace */
361         return plpar_hcall_norets(H_SET_XDABR, dabr,
362                         H_DABRX_KERNEL | H_DABRX_USER);
363 }
364
365 /*
366  * Early initialization.  Relocation is on but do not reference unbolted pages
367  */
368 static void __init pSeries_init_early(void)
369 {
370         int iommu_off = 0;
371
372         DBG(" -> pSeries_init_early()\n");
373
374         fw_feature_init();
375         
376         if (platform_is_lpar())
377                 hpte_init_lpar();
378         else {
379                 hpte_init_native();
380                 iommu_off = (of_chosen &&
381                              get_property(of_chosen, "linux,iommu-off", NULL));
382         }
383
384         if (platform_is_lpar())
385                 find_udbg_vterm();
386
387         if (firmware_has_feature(FW_FEATURE_DABR))
388                 ppc_md.set_dabr = pseries_set_dabr;
389         else if (firmware_has_feature(FW_FEATURE_XDABR))
390                 ppc_md.set_dabr = pseries_set_xdabr;
391
392         iommu_init_early_pSeries();
393
394         pSeries_discover_pic();
395
396         DBG(" <- pSeries_init_early()\n");
397 }
398
399
400 static int pSeries_check_legacy_ioport(unsigned int baseport)
401 {
402         struct device_node *np;
403
404 #define I8042_DATA_REG  0x60
405 #define FDC_BASE        0x3f0
406
407
408         switch(baseport) {
409         case I8042_DATA_REG:
410                 np = of_find_node_by_type(NULL, "8042");
411                 if (np == NULL)
412                         return -ENODEV;
413                 of_node_put(np);
414                 break;
415         case FDC_BASE:
416                 np = of_find_node_by_type(NULL, "fdc");
417                 if (np == NULL)
418                         return -ENODEV;
419                 of_node_put(np);
420                 break;
421         }
422         return 0;
423 }
424
425 /*
426  * Called very early, MMU is off, device-tree isn't unflattened
427  */
428 extern struct machdep_calls pSeries_md;
429
430 static int __init pSeries_probe(int platform)
431 {
432         if (platform != PLATFORM_PSERIES &&
433             platform != PLATFORM_PSERIES_LPAR)
434                 return 0;
435
436         /* if we have some ppc_md fixups for LPAR to do, do
437          * it here ...
438          */
439
440         return 1;
441 }
442
443 DECLARE_PER_CPU(unsigned long, smt_snooze_delay);
444
445 static inline void dedicated_idle_sleep(unsigned int cpu)
446 {
447         struct lppaca *plppaca = &lppaca[cpu ^ 1];
448
449         /* Only sleep if the other thread is not idle */
450         if (!(plppaca->idle)) {
451                 local_irq_disable();
452
453                 /*
454                  * We are about to sleep the thread and so wont be polling any
455                  * more.
456                  */
457                 clear_thread_flag(TIF_POLLING_NRFLAG);
458                 smp_mb__after_clear_bit();
459
460                 /*
461                  * SMT dynamic mode. Cede will result in this thread going
462                  * dormant, if the partner thread is still doing work.  Thread
463                  * wakes up if partner goes idle, an interrupt is presented, or
464                  * a prod occurs.  Returning from the cede enables external
465                  * interrupts.
466                  */
467                 if (!need_resched())
468                         cede_processor();
469                 else
470                         local_irq_enable();
471                 set_thread_flag(TIF_POLLING_NRFLAG);
472         } else {
473                 /*
474                  * Give the HV an opportunity at the processor, since we are
475                  * not doing any work.
476                  */
477                 poll_pending();
478         }
479 }
480
481 static void pseries_dedicated_idle(void)
482
483         unsigned int cpu = smp_processor_id();
484         unsigned long start_snooze;
485         unsigned long *smt_snooze_delay = &__get_cpu_var(smt_snooze_delay);
486         set_thread_flag(TIF_POLLING_NRFLAG);
487
488         while (1) {
489                 /*
490                  * Indicate to the HV that we are idle. Now would be
491                  * a good time to find other work to dispatch.
492                  */
493                 get_lppaca()->idle = 1;
494
495                 if (!need_resched()) {
496                         start_snooze = get_tb() +
497                                 *smt_snooze_delay * tb_ticks_per_usec;
498
499                         while (!need_resched() && !cpu_is_offline(cpu)) {
500                                 ppc64_runlatch_off();
501
502                                 /*
503                                  * Go into low thread priority and possibly
504                                  * low power mode.
505                                  */
506                                 HMT_low();
507                                 HMT_very_low();
508
509                                 if (*smt_snooze_delay != 0 &&
510                                     get_tb() > start_snooze) {
511                                         HMT_medium();
512                                         dedicated_idle_sleep(cpu);
513                                 }
514
515                         }
516
517                         HMT_medium();
518                 }
519
520                 get_lppaca()->idle = 0;
521                 ppc64_runlatch_on();
522
523                 preempt_enable_no_resched();
524                 schedule();
525                 preempt_disable();
526
527                 if (cpu_is_offline(cpu) && system_state == SYSTEM_RUNNING)
528                         cpu_die();
529         }
530 }
531
532 static void pseries_shared_idle(void)
533 {
534         unsigned int cpu = smp_processor_id();
535
536         while (1) {
537                 /*
538                  * Indicate to the HV that we are idle. Now would be
539                  * a good time to find other work to dispatch.
540                  */
541                 get_lppaca()->idle = 1;
542
543                 while (!need_resched() && !cpu_is_offline(cpu)) {
544                         local_irq_disable();
545                         ppc64_runlatch_off();
546
547                         /*
548                          * Yield the processor to the hypervisor.  We return if
549                          * an external interrupt occurs (which are driven prior
550                          * to returning here) or if a prod occurs from another
551                          * processor. When returning here, external interrupts
552                          * are enabled.
553                          *
554                          * Check need_resched() again with interrupts disabled
555                          * to avoid a race.
556                          */
557                         if (!need_resched())
558                                 cede_processor();
559                         else
560                                 local_irq_enable();
561
562                         HMT_medium();
563                 }
564
565                 get_lppaca()->idle = 0;
566                 ppc64_runlatch_on();
567
568                 preempt_enable_no_resched();
569                 schedule();
570                 preempt_disable();
571
572                 if (cpu_is_offline(cpu) && system_state == SYSTEM_RUNNING)
573                         cpu_die();
574         }
575 }
576
577 static int pSeries_pci_probe_mode(struct pci_bus *bus)
578 {
579         if (platform_is_lpar())
580                 return PCI_PROBE_DEVTREE;
581         return PCI_PROBE_NORMAL;
582 }
583
584 #ifdef CONFIG_KEXEC
585 static void pseries_kexec_cpu_down(int crash_shutdown, int secondary)
586 {
587         /* Don't risk a hypervisor call if we're crashing */
588         if (firmware_has_feature(FW_FEATURE_SPLPAR) && !crash_shutdown) {
589                 unsigned long vpa = __pa(get_lppaca());
590
591                 if (unregister_vpa(hard_smp_processor_id(), vpa)) {
592                         printk("VPA deregistration of cpu %u (hw_cpu_id %d) "
593                                         "failed\n", smp_processor_id(),
594                                         hard_smp_processor_id());
595                 }
596         }
597
598         if (ppc64_interrupt_controller == IC_OPEN_PIC)
599                 mpic_teardown_this_cpu(secondary);
600         else
601                 xics_teardown_cpu(secondary);
602 }
603 #endif
604
605 struct machdep_calls __initdata pSeries_md = {
606         .probe                  = pSeries_probe,
607         .setup_arch             = pSeries_setup_arch,
608         .init_early             = pSeries_init_early,
609         .show_cpuinfo           = pSeries_show_cpuinfo,
610         .log_error              = pSeries_log_error,
611         .pcibios_fixup          = pSeries_final_fixup,
612         .pci_probe_mode         = pSeries_pci_probe_mode,
613         .irq_bus_setup          = pSeries_irq_bus_setup,
614         .restart                = rtas_restart,
615         .power_off              = rtas_power_off,
616         .halt                   = rtas_halt,
617         .panic                  = rtas_os_term,
618         .cpu_die                = pSeries_mach_cpu_die,
619         .get_boot_time          = rtas_get_boot_time,
620         .get_rtc_time           = rtas_get_rtc_time,
621         .set_rtc_time           = rtas_set_rtc_time,
622         .calibrate_decr         = generic_calibrate_decr,
623         .progress               = rtas_progress,
624         .check_legacy_ioport    = pSeries_check_legacy_ioport,
625         .system_reset_exception = pSeries_system_reset_exception,
626         .machine_check_exception = pSeries_machine_check_exception,
627 #ifdef CONFIG_KEXEC
628         .kexec_cpu_down         = pseries_kexec_cpu_down,
629         .machine_kexec          = default_machine_kexec,
630         .machine_kexec_prepare  = default_machine_kexec_prepare,
631         .machine_crash_shutdown = default_machine_crash_shutdown,
632 #endif
633 };