[PATCH] sched: resched and cpu_idle rework
[safe/jmp/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/time.h>
60 #include <asm/nvram.h>
61 #include "xics.h"
62 #include <asm/firmware.h>
63 #include <asm/pmc.h>
64 #include <asm/mpic.h>
65 #include <asm/ppc-pci.h>
66 #include <asm/i8259.h>
67 #include <asm/udbg.h>
68 #include <asm/smp.h>
69
70 #include "plpar_wrappers.h"
71
72 #ifdef DEBUG
73 #define DBG(fmt...) udbg_printf(fmt)
74 #else
75 #define DBG(fmt...)
76 #endif
77
78 extern void find_udbg_vterm(void);
79 extern void system_reset_fwnmi(void);   /* from head.S */
80 extern void machine_check_fwnmi(void);  /* from head.S */
81 extern void generic_find_legacy_serial_ports(u64 *physport,
82                 unsigned int *default_speed);
83
84 int fwnmi_active;  /* TRUE if an FWNMI handler is present */
85
86 extern void pSeries_system_reset_exception(struct pt_regs *regs);
87 extern int pSeries_machine_check_exception(struct pt_regs *regs);
88
89 static void pseries_shared_idle(void);
90 static void pseries_dedicated_idle(void);
91
92 struct mpic *pSeries_mpic;
93
94 void pSeries_show_cpuinfo(struct seq_file *m)
95 {
96         struct device_node *root;
97         const char *model = "";
98
99         root = of_find_node_by_path("/");
100         if (root)
101                 model = get_property(root, "model", NULL);
102         seq_printf(m, "machine\t\t: CHRP %s\n", model);
103         of_node_put(root);
104 }
105
106 /* Initialize firmware assisted non-maskable interrupts if
107  * the firmware supports this feature.
108  *
109  */
110 static void __init fwnmi_init(void)
111 {
112         int ret;
113         int ibm_nmi_register = rtas_token("ibm,nmi-register");
114         if (ibm_nmi_register == RTAS_UNKNOWN_SERVICE)
115                 return;
116         ret = rtas_call(ibm_nmi_register, 2, 1, NULL,
117                         __pa((unsigned long)system_reset_fwnmi),
118                         __pa((unsigned long)machine_check_fwnmi));
119         if (ret == 0)
120                 fwnmi_active = 1;
121 }
122
123 static void __init pSeries_init_mpic(void)
124 {
125         unsigned int *addrp;
126         struct device_node *np;
127         unsigned long intack = 0;
128
129         /* All ISUs are setup, complete initialization */
130         mpic_init(pSeries_mpic);
131
132         /* Check what kind of cascade ACK we have */
133         if (!(np = of_find_node_by_name(NULL, "pci"))
134             || !(addrp = (unsigned int *)
135                  get_property(np, "8259-interrupt-acknowledge", NULL)))
136                 printk(KERN_ERR "Cannot find pci to get ack address\n");
137         else
138                 intack = addrp[prom_n_addr_cells(np)-1];
139         of_node_put(np);
140
141         /* Setup the legacy interrupts & controller */
142         i8259_init(intack, 0);
143
144         /* Hook cascade to mpic */
145         mpic_setup_cascade(NUM_ISA_INTERRUPTS, i8259_irq_cascade, NULL);
146 }
147
148 static void __init pSeries_setup_mpic(void)
149 {
150         unsigned int *opprop;
151         unsigned long openpic_addr = 0;
152         unsigned char senses[NR_IRQS - NUM_ISA_INTERRUPTS];
153         struct device_node *root;
154         int irq_count;
155
156         /* Find the Open PIC if present */
157         root = of_find_node_by_path("/");
158         opprop = (unsigned int *) get_property(root, "platform-open-pic", NULL);
159         if (opprop != 0) {
160                 int n = prom_n_addr_cells(root);
161
162                 for (openpic_addr = 0; n > 0; --n)
163                         openpic_addr = (openpic_addr << 32) + *opprop++;
164                 printk(KERN_DEBUG "OpenPIC addr: %lx\n", openpic_addr);
165         }
166         of_node_put(root);
167
168         BUG_ON(openpic_addr == 0);
169
170         /* Get the sense values from OF */
171         prom_get_irq_senses(senses, NUM_ISA_INTERRUPTS, NR_IRQS);
172         
173         /* Setup the openpic driver */
174         irq_count = NR_IRQS - NUM_ISA_INTERRUPTS - 4; /* leave room for IPIs */
175         pSeries_mpic = mpic_alloc(openpic_addr, MPIC_PRIMARY,
176                                   16, 16, irq_count, /* isu size, irq offset, irq count */ 
177                                   NR_IRQS - 4, /* ipi offset */
178                                   senses, irq_count, /* sense & sense size */
179                                   " MPIC     ");
180 }
181
182 static void pseries_lpar_enable_pmcs(void)
183 {
184         unsigned long set, reset;
185
186         power4_enable_pmcs();
187
188         set = 1UL << 63;
189         reset = 0;
190         plpar_hcall_norets(H_PERFMON, set, reset);
191
192         /* instruct hypervisor to maintain PMCs */
193         if (firmware_has_feature(FW_FEATURE_SPLPAR))
194                 get_paca()->lppaca.pmcregs_in_use = 1;
195 }
196
197 static void __init pSeries_setup_arch(void)
198 {
199         /* Fixup ppc_md depending on the type of interrupt controller */
200         if (ppc64_interrupt_controller == IC_OPEN_PIC) {
201                 ppc_md.init_IRQ       = pSeries_init_mpic;
202                 ppc_md.get_irq        = mpic_get_irq;
203                 ppc_md.cpu_irq_down   = mpic_teardown_this_cpu;
204                 /* Allocate the mpic now, so that find_and_init_phbs() can
205                  * fill the ISUs */
206                 pSeries_setup_mpic();
207         } else {
208                 ppc_md.init_IRQ       = xics_init_IRQ;
209                 ppc_md.get_irq        = xics_get_irq;
210                 ppc_md.cpu_irq_down   = xics_teardown_cpu;
211         }
212
213 #ifdef CONFIG_SMP
214         smp_init_pSeries();
215 #endif
216         /* openpic global configuration register (64-bit format). */
217         /* openpic Interrupt Source Unit pointer (64-bit format). */
218         /* python0 facility area (mmio) (64-bit format) REAL address. */
219
220         /* init to some ~sane value until calibrate_delay() runs */
221         loops_per_jiffy = 50000000;
222
223         if (ROOT_DEV == 0) {
224                 printk("No ramdisk, default root is /dev/sda2\n");
225                 ROOT_DEV = Root_SDA2;
226         }
227
228         fwnmi_init();
229
230         /* Find and initialize PCI host bridges */
231         init_pci_config_tokens();
232         find_and_init_phbs();
233         eeh_init();
234
235         pSeries_nvram_init();
236
237         /* Choose an idle loop */
238         if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
239                 vpa_init(boot_cpuid);
240                 if (get_paca()->lppaca.shared_proc) {
241                         printk(KERN_INFO "Using shared processor idle loop\n");
242                         ppc_md.idle_loop = pseries_shared_idle;
243                 } else {
244                         printk(KERN_INFO "Using dedicated idle loop\n");
245                         ppc_md.idle_loop = pseries_dedicated_idle;
246                 }
247         } else {
248                 printk(KERN_INFO "Using default idle loop\n");
249                 ppc_md.idle_loop = default_idle;
250         }
251
252         if (systemcfg->platform & PLATFORM_LPAR)
253                 ppc_md.enable_pmcs = pseries_lpar_enable_pmcs;
254         else
255                 ppc_md.enable_pmcs = power4_enable_pmcs;
256 }
257
258 static int __init pSeries_init_panel(void)
259 {
260         /* Manually leave the kernel version on the panel. */
261         ppc_md.progress("Linux ppc64\n", 0);
262         ppc_md.progress(system_utsname.version, 0);
263
264         return 0;
265 }
266 arch_initcall(pSeries_init_panel);
267
268
269 /* Build up the ppc64_firmware_features bitmask field
270  * using contents of device-tree/ibm,hypertas-functions.
271  * Ultimately this functionality may be moved into prom.c prom_init().
272  */
273 static void __init fw_feature_init(void)
274 {
275         struct device_node * dn;
276         char * hypertas;
277         unsigned int len;
278
279         DBG(" -> fw_feature_init()\n");
280
281         ppc64_firmware_features = 0;
282         dn = of_find_node_by_path("/rtas");
283         if (dn == NULL) {
284                 printk(KERN_ERR "WARNING ! Cannot find RTAS in device-tree !\n");
285                 goto no_rtas;
286         }
287
288         hypertas = get_property(dn, "ibm,hypertas-functions", &len);
289         if (hypertas) {
290                 while (len > 0){
291                         int i, hypertas_len;
292                         /* check value against table of strings */
293                         for(i=0; i < FIRMWARE_MAX_FEATURES ;i++) {
294                                 if ((firmware_features_table[i].name) &&
295                                     (strcmp(firmware_features_table[i].name,hypertas))==0) {
296                                         /* we have a match */
297                                         ppc64_firmware_features |= 
298                                                 (firmware_features_table[i].val);
299                                         break;
300                                 } 
301                         }
302                         hypertas_len = strlen(hypertas);
303                         len -= hypertas_len +1;
304                         hypertas+= hypertas_len +1;
305                 }
306         }
307
308         of_node_put(dn);
309  no_rtas:
310         printk(KERN_INFO "firmware_features = 0x%lx\n", 
311                ppc64_firmware_features);
312
313         DBG(" <- fw_feature_init()\n");
314 }
315
316
317 static  void __init pSeries_discover_pic(void)
318 {
319         struct device_node *np;
320         char *typep;
321
322         /*
323          * Setup interrupt mapping options that are needed for finish_device_tree
324          * to properly parse the OF interrupt tree & do the virtual irq mapping
325          */
326         __irq_offset_value = NUM_ISA_INTERRUPTS;
327         ppc64_interrupt_controller = IC_INVALID;
328         for (np = NULL; (np = of_find_node_by_name(np, "interrupt-controller"));) {
329                 typep = (char *)get_property(np, "compatible", NULL);
330                 if (strstr(typep, "open-pic"))
331                         ppc64_interrupt_controller = IC_OPEN_PIC;
332                 else if (strstr(typep, "ppc-xicp"))
333                         ppc64_interrupt_controller = IC_PPC_XIC;
334                 else
335                         printk("pSeries_discover_pic: failed to recognize"
336                                " interrupt-controller\n");
337                 break;
338         }
339 }
340
341 static void pSeries_mach_cpu_die(void)
342 {
343         local_irq_disable();
344         idle_task_exit();
345         /* Some hardware requires clearing the CPPR, while other hardware does not
346          * it is safe either way
347          */
348         pSeriesLP_cppr_info(0, 0);
349         rtas_stop_self();
350         /* Should never get here... */
351         BUG();
352         for(;;);
353 }
354
355 static int pseries_set_dabr(unsigned long dabr)
356 {
357         return plpar_hcall_norets(H_SET_DABR, dabr);
358 }
359
360 static int pseries_set_xdabr(unsigned long dabr)
361 {
362         /* We want to catch accesses from kernel and userspace */
363         return plpar_hcall_norets(H_SET_XDABR, dabr,
364                         H_DABRX_KERNEL | H_DABRX_USER);
365 }
366
367 /*
368  * Early initialization.  Relocation is on but do not reference unbolted pages
369  */
370 static void __init pSeries_init_early(void)
371 {
372         void *comport;
373         int iommu_off = 0;
374         unsigned int default_speed;
375         u64 physport;
376
377         DBG(" -> pSeries_init_early()\n");
378
379         fw_feature_init();
380         
381         if (systemcfg->platform & PLATFORM_LPAR)
382                 hpte_init_lpar();
383         else {
384                 hpte_init_native();
385                 iommu_off = (of_chosen &&
386                              get_property(of_chosen, "linux,iommu-off", NULL));
387         }
388
389         generic_find_legacy_serial_ports(&physport, &default_speed);
390
391         if (systemcfg->platform & PLATFORM_LPAR)
392                 find_udbg_vterm();
393         else if (physport) {
394                 /* Map the uart for udbg. */
395                 comport = (void *)ioremap(physport, 16);
396                 udbg_init_uart(comport, default_speed);
397
398                 DBG("Hello World !\n");
399         }
400
401         if (firmware_has_feature(FW_FEATURE_DABR))
402                 ppc_md.set_dabr = pseries_set_dabr;
403         else if (firmware_has_feature(FW_FEATURE_XDABR))
404                 ppc_md.set_dabr = pseries_set_xdabr;
405
406         iommu_init_early_pSeries();
407
408         pSeries_discover_pic();
409
410         DBG(" <- pSeries_init_early()\n");
411 }
412
413
414 static int pSeries_check_legacy_ioport(unsigned int baseport)
415 {
416         struct device_node *np;
417
418 #define I8042_DATA_REG  0x60
419 #define FDC_BASE        0x3f0
420
421
422         switch(baseport) {
423         case I8042_DATA_REG:
424                 np = of_find_node_by_type(NULL, "8042");
425                 if (np == NULL)
426                         return -ENODEV;
427                 of_node_put(np);
428                 break;
429         case FDC_BASE:
430                 np = of_find_node_by_type(NULL, "fdc");
431                 if (np == NULL)
432                         return -ENODEV;
433                 of_node_put(np);
434                 break;
435         }
436         return 0;
437 }
438
439 /*
440  * Called very early, MMU is off, device-tree isn't unflattened
441  */
442 extern struct machdep_calls pSeries_md;
443
444 static int __init pSeries_probe(int platform)
445 {
446         if (platform != PLATFORM_PSERIES &&
447             platform != PLATFORM_PSERIES_LPAR)
448                 return 0;
449
450         /* if we have some ppc_md fixups for LPAR to do, do
451          * it here ...
452          */
453
454         return 1;
455 }
456
457 DECLARE_PER_CPU(unsigned long, smt_snooze_delay);
458
459 static inline void dedicated_idle_sleep(unsigned int cpu)
460 {
461         struct paca_struct *ppaca = &paca[cpu ^ 1];
462
463         /* Only sleep if the other thread is not idle */
464         if (!(ppaca->lppaca.idle)) {
465                 local_irq_disable();
466
467                 /*
468                  * We are about to sleep the thread and so wont be polling any
469                  * more.
470                  */
471                 clear_thread_flag(TIF_POLLING_NRFLAG);
472                 smp_mb__after_clear_bit();
473
474                 /*
475                  * SMT dynamic mode. Cede will result in this thread going
476                  * dormant, if the partner thread is still doing work.  Thread
477                  * wakes up if partner goes idle, an interrupt is presented, or
478                  * a prod occurs.  Returning from the cede enables external
479                  * interrupts.
480                  */
481                 if (!need_resched())
482                         cede_processor();
483                 else
484                         local_irq_enable();
485                 set_thread_flag(TIF_POLLING_NRFLAG);
486         } else {
487                 /*
488                  * Give the HV an opportunity at the processor, since we are
489                  * not doing any work.
490                  */
491                 poll_pending();
492         }
493 }
494
495 static void pseries_dedicated_idle(void)
496
497         struct paca_struct *lpaca = get_paca();
498         unsigned int cpu = smp_processor_id();
499         unsigned long start_snooze;
500         unsigned long *smt_snooze_delay = &__get_cpu_var(smt_snooze_delay);
501         set_thread_flag(TIF_POLLING_NRFLAG);
502
503         while (1) {
504                 /*
505                  * Indicate to the HV that we are idle. Now would be
506                  * a good time to find other work to dispatch.
507                  */
508                 lpaca->lppaca.idle = 1;
509
510                 if (!need_resched()) {
511                         start_snooze = __get_tb() +
512                                 *smt_snooze_delay * tb_ticks_per_usec;
513
514                         while (!need_resched() && !cpu_is_offline(cpu)) {
515                                 ppc64_runlatch_off();
516
517                                 /*
518                                  * Go into low thread priority and possibly
519                                  * low power mode.
520                                  */
521                                 HMT_low();
522                                 HMT_very_low();
523
524                                 if (*smt_snooze_delay != 0 &&
525                                     __get_tb() > start_snooze) {
526                                         HMT_medium();
527                                         dedicated_idle_sleep(cpu);
528                                 }
529
530                         }
531
532                         HMT_medium();
533                 }
534
535                 lpaca->lppaca.idle = 0;
536                 ppc64_runlatch_on();
537
538                 preempt_enable_no_resched();
539                 schedule();
540                 preempt_disable();
541
542                 if (cpu_is_offline(cpu) && system_state == SYSTEM_RUNNING)
543                         cpu_die();
544         }
545 }
546
547 static void pseries_shared_idle(void)
548 {
549         struct paca_struct *lpaca = get_paca();
550         unsigned int cpu = smp_processor_id();
551
552         while (1) {
553                 /*
554                  * Indicate to the HV that we are idle. Now would be
555                  * a good time to find other work to dispatch.
556                  */
557                 lpaca->lppaca.idle = 1;
558
559                 while (!need_resched() && !cpu_is_offline(cpu)) {
560                         local_irq_disable();
561                         ppc64_runlatch_off();
562
563                         /*
564                          * Yield the processor to the hypervisor.  We return if
565                          * an external interrupt occurs (which are driven prior
566                          * to returning here) or if a prod occurs from another
567                          * processor. When returning here, external interrupts
568                          * are enabled.
569                          *
570                          * Check need_resched() again with interrupts disabled
571                          * to avoid a race.
572                          */
573                         if (!need_resched())
574                                 cede_processor();
575                         else
576                                 local_irq_enable();
577
578                         HMT_medium();
579                 }
580
581                 lpaca->lppaca.idle = 0;
582                 ppc64_runlatch_on();
583
584                 preempt_enable_no_resched();
585                 schedule();
586                 preempt_disable();
587
588                 if (cpu_is_offline(cpu) && system_state == SYSTEM_RUNNING)
589                         cpu_die();
590         }
591 }
592
593 static int pSeries_pci_probe_mode(struct pci_bus *bus)
594 {
595         if (systemcfg->platform & PLATFORM_LPAR)
596                 return PCI_PROBE_DEVTREE;
597         return PCI_PROBE_NORMAL;
598 }
599
600 struct machdep_calls __initdata pSeries_md = {
601         .probe                  = pSeries_probe,
602         .setup_arch             = pSeries_setup_arch,
603         .init_early             = pSeries_init_early,
604         .show_cpuinfo           = pSeries_show_cpuinfo,
605         .log_error              = pSeries_log_error,
606         .pcibios_fixup          = pSeries_final_fixup,
607         .pci_probe_mode         = pSeries_pci_probe_mode,
608         .irq_bus_setup          = pSeries_irq_bus_setup,
609         .restart                = rtas_restart,
610         .power_off              = rtas_power_off,
611         .halt                   = rtas_halt,
612         .panic                  = rtas_os_term,
613         .cpu_die                = pSeries_mach_cpu_die,
614         .get_boot_time          = rtas_get_boot_time,
615         .get_rtc_time           = rtas_get_rtc_time,
616         .set_rtc_time           = rtas_set_rtc_time,
617         .calibrate_decr         = generic_calibrate_decr,
618         .progress               = rtas_progress,
619         .check_legacy_ioport    = pSeries_check_legacy_ioport,
620         .system_reset_exception = pSeries_system_reset_exception,
621         .machine_check_exception = pSeries_machine_check_exception,
622 };