Pull address_range into release branch
[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/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 #include "ras.h"
72 #include "firmware.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_sleep(void);
85 static void pseries_dedicated_idle_sleep(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.power_save = pseries_shared_idle_sleep;
240                 } else {
241                         printk(KERN_INFO "Using dedicated idle loop\n");
242                         ppc_md.power_save = pseries_dedicated_idle_sleep;
243                 }
244         } else {
245                 printk(KERN_INFO "Using default idle loop\n");
246         }
247
248         if (firmware_has_feature(FW_FEATURE_LPAR))
249                 ppc_md.enable_pmcs = pseries_lpar_enable_pmcs;
250         else
251                 ppc_md.enable_pmcs = power4_enable_pmcs;
252 }
253
254 static int __init pSeries_init_panel(void)
255 {
256         /* Manually leave the kernel version on the panel. */
257         ppc_md.progress("Linux ppc64\n", 0);
258         ppc_md.progress(system_utsname.release, 0);
259
260         return 0;
261 }
262 arch_initcall(pSeries_init_panel);
263
264 static  void __init pSeries_discover_pic(void)
265 {
266         struct device_node *np;
267         char *typep;
268
269         /*
270          * Setup interrupt mapping options that are needed for finish_device_tree
271          * to properly parse the OF interrupt tree & do the virtual irq mapping
272          */
273         __irq_offset_value = NUM_ISA_INTERRUPTS;
274         ppc64_interrupt_controller = IC_INVALID;
275         for (np = NULL; (np = of_find_node_by_name(np, "interrupt-controller"));) {
276                 typep = (char *)get_property(np, "compatible", NULL);
277                 if (strstr(typep, "open-pic")) {
278                         ppc64_interrupt_controller = IC_OPEN_PIC;
279                         break;
280                 } else if (strstr(typep, "ppc-xicp")) {
281                         ppc64_interrupt_controller = IC_PPC_XIC;
282                         break;
283                 }
284         }
285         if (ppc64_interrupt_controller == IC_INVALID)
286                 printk("pSeries_discover_pic: failed to recognize"
287                         " interrupt-controller\n");
288
289 }
290
291 static void pSeries_mach_cpu_die(void)
292 {
293         local_irq_disable();
294         idle_task_exit();
295         /* Some hardware requires clearing the CPPR, while other hardware does not
296          * it is safe either way
297          */
298         pSeriesLP_cppr_info(0, 0);
299         rtas_stop_self();
300         /* Should never get here... */
301         BUG();
302         for(;;);
303 }
304
305 static int pseries_set_dabr(unsigned long dabr)
306 {
307         return plpar_hcall_norets(H_SET_DABR, dabr);
308 }
309
310 static int pseries_set_xdabr(unsigned long dabr)
311 {
312         /* We want to catch accesses from kernel and userspace */
313         return plpar_hcall_norets(H_SET_XDABR, dabr,
314                         H_DABRX_KERNEL | H_DABRX_USER);
315 }
316
317 /*
318  * Early initialization.  Relocation is on but do not reference unbolted pages
319  */
320 static void __init pSeries_init_early(void)
321 {
322         DBG(" -> pSeries_init_early()\n");
323
324         fw_feature_init();
325         
326         if (firmware_has_feature(FW_FEATURE_LPAR))
327                 hpte_init_lpar();
328         else
329                 hpte_init_native();
330
331         if (firmware_has_feature(FW_FEATURE_LPAR))
332                 find_udbg_vterm();
333
334         if (firmware_has_feature(FW_FEATURE_DABR))
335                 ppc_md.set_dabr = pseries_set_dabr;
336         else if (firmware_has_feature(FW_FEATURE_XDABR))
337                 ppc_md.set_dabr = pseries_set_xdabr;
338
339         iommu_init_early_pSeries();
340
341         pSeries_discover_pic();
342
343         DBG(" <- pSeries_init_early()\n");
344 }
345
346
347 static int pSeries_check_legacy_ioport(unsigned int baseport)
348 {
349         struct device_node *np;
350
351 #define I8042_DATA_REG  0x60
352 #define FDC_BASE        0x3f0
353
354
355         switch(baseport) {
356         case I8042_DATA_REG:
357                 np = of_find_node_by_type(NULL, "8042");
358                 if (np == NULL)
359                         return -ENODEV;
360                 of_node_put(np);
361                 break;
362         case FDC_BASE:
363                 np = of_find_node_by_type(NULL, "fdc");
364                 if (np == NULL)
365                         return -ENODEV;
366                 of_node_put(np);
367                 break;
368         }
369         return 0;
370 }
371
372 /*
373  * Called very early, MMU is off, device-tree isn't unflattened
374  */
375
376 static int __init pSeries_probe_hypertas(unsigned long node,
377                                          const char *uname, int depth,
378                                          void *data)
379 {
380         if (depth != 1 ||
381             (strcmp(uname, "rtas") != 0 && strcmp(uname, "rtas@0") != 0))
382                 return 0;
383
384         if (of_get_flat_dt_prop(node, "ibm,hypertas-functions", NULL) != NULL)
385                 powerpc_firmware_features |= FW_FEATURE_LPAR;
386
387         return 1;
388 }
389
390 static int __init pSeries_probe(void)
391 {
392         unsigned long root = of_get_flat_dt_root();
393         char *dtype = of_get_flat_dt_prop(of_get_flat_dt_root(),
394                                           "device_type", NULL);
395         if (dtype == NULL)
396                 return 0;
397         if (strcmp(dtype, "chrp"))
398                 return 0;
399
400         /* Cell blades firmware claims to be chrp while it's not. Until this
401          * is fixed, we need to avoid those here.
402          */
403         if (of_flat_dt_is_compatible(root, "IBM,CPBW-1.0") ||
404             of_flat_dt_is_compatible(root, "IBM,CBEA"))
405                 return 0;
406
407         DBG("pSeries detected, looking for LPAR capability...\n");
408
409         /* Now try to figure out if we are running on LPAR */
410         of_scan_flat_dt(pSeries_probe_hypertas, NULL);
411
412         DBG("Machine is%s LPAR !\n",
413             (powerpc_firmware_features & FW_FEATURE_LPAR) ? "" : " not");
414
415         return 1;
416 }
417
418
419 DECLARE_PER_CPU(unsigned long, smt_snooze_delay);
420
421 static void pseries_dedicated_idle_sleep(void)
422
423         unsigned int cpu = smp_processor_id();
424         unsigned long start_snooze;
425         unsigned long *smt_snooze_delay = &__get_cpu_var(smt_snooze_delay);
426
427         /*
428          * Indicate to the HV that we are idle. Now would be
429          * a good time to find other work to dispatch.
430          */
431         get_lppaca()->idle = 1;
432
433         /*
434          * We come in with interrupts disabled, and need_resched()
435          * has been checked recently.  If we should poll for a little
436          * while, do so.
437          */
438         if (*smt_snooze_delay) {
439                 start_snooze = get_tb() +
440                         *smt_snooze_delay * tb_ticks_per_usec;
441                 local_irq_enable();
442                 set_thread_flag(TIF_POLLING_NRFLAG);
443
444                 while (get_tb() < start_snooze) {
445                         if (need_resched() || cpu_is_offline(cpu))
446                                 goto out;
447                         ppc64_runlatch_off();
448                         HMT_low();
449                         HMT_very_low();
450                 }
451
452                 HMT_medium();
453                 clear_thread_flag(TIF_POLLING_NRFLAG);
454                 smp_mb();
455                 local_irq_disable();
456                 if (need_resched() || cpu_is_offline(cpu))
457                         goto out;
458         }
459
460         /*
461          * Cede if the other thread is not idle, so that it can
462          * go single-threaded.  If the other thread is idle,
463          * we ask the hypervisor if it has pending work it
464          * wants to do and cede if it does.  Otherwise we keep
465          * polling in order to reduce interrupt latency.
466          *
467          * Doing the cede when the other thread is active will
468          * result in this thread going dormant, meaning the other
469          * thread gets to run in single-threaded (ST) mode, which
470          * is slightly faster than SMT mode with this thread at
471          * very low priority.  The cede enables interrupts, which
472          * doesn't matter here.
473          */
474         if (!lppaca[cpu ^ 1].idle || poll_pending() == H_PENDING)
475                 cede_processor();
476
477 out:
478         HMT_medium();
479         get_lppaca()->idle = 0;
480 }
481
482 static void pseries_shared_idle_sleep(void)
483 {
484         /*
485          * Indicate to the HV that we are idle. Now would be
486          * a good time to find other work to dispatch.
487          */
488         get_lppaca()->idle = 1;
489
490         /*
491          * Yield the processor to the hypervisor.  We return if
492          * an external interrupt occurs (which are driven prior
493          * to returning here) or if a prod occurs from another
494          * processor. When returning here, external interrupts
495          * are enabled.
496          */
497         cede_processor();
498
499         get_lppaca()->idle = 0;
500 }
501
502 static int pSeries_pci_probe_mode(struct pci_bus *bus)
503 {
504         if (firmware_has_feature(FW_FEATURE_LPAR))
505                 return PCI_PROBE_DEVTREE;
506         return PCI_PROBE_NORMAL;
507 }
508
509 #ifdef CONFIG_KEXEC
510 static void pseries_kexec_cpu_down(int crash_shutdown, int secondary)
511 {
512         /* Don't risk a hypervisor call if we're crashing */
513         if (firmware_has_feature(FW_FEATURE_SPLPAR) && !crash_shutdown) {
514                 unsigned long vpa = __pa(get_lppaca());
515
516                 if (unregister_vpa(hard_smp_processor_id(), vpa)) {
517                         printk("VPA deregistration of cpu %u (hw_cpu_id %d) "
518                                         "failed\n", smp_processor_id(),
519                                         hard_smp_processor_id());
520                 }
521         }
522
523         if (ppc64_interrupt_controller == IC_OPEN_PIC)
524                 mpic_teardown_this_cpu(secondary);
525         else
526                 xics_teardown_cpu(secondary);
527 }
528 #endif
529
530 define_machine(pseries) {
531         .name                   = "pSeries",
532         .probe                  = pSeries_probe,
533         .setup_arch             = pSeries_setup_arch,
534         .init_early             = pSeries_init_early,
535         .show_cpuinfo           = pSeries_show_cpuinfo,
536         .log_error              = pSeries_log_error,
537         .pcibios_fixup          = pSeries_final_fixup,
538         .pci_probe_mode         = pSeries_pci_probe_mode,
539         .irq_bus_setup          = pSeries_irq_bus_setup,
540         .restart                = rtas_restart,
541         .power_off              = rtas_power_off,
542         .halt                   = rtas_halt,
543         .panic                  = rtas_os_term,
544         .cpu_die                = pSeries_mach_cpu_die,
545         .get_boot_time          = rtas_get_boot_time,
546         .get_rtc_time           = rtas_get_rtc_time,
547         .set_rtc_time           = rtas_set_rtc_time,
548         .calibrate_decr         = generic_calibrate_decr,
549         .progress               = rtas_progress,
550         .check_legacy_ioport    = pSeries_check_legacy_ioport,
551         .system_reset_exception = pSeries_system_reset_exception,
552         .machine_check_exception = pSeries_machine_check_exception,
553 #ifdef CONFIG_KEXEC
554         .kexec_cpu_down         = pseries_kexec_cpu_down,
555         .machine_kexec          = default_machine_kexec,
556         .machine_kexec_prepare  = default_machine_kexec_prepare,
557         .machine_crash_shutdown = default_machine_crash_shutdown,
558 #endif
559 };