Pull acpi_device_handle_cleanup 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/cpu.h>
22 #include <linux/errno.h>
23 #include <linux/sched.h>
24 #include <linux/kernel.h>
25 #include <linux/mm.h>
26 #include <linux/stddef.h>
27 #include <linux/unistd.h>
28 #include <linux/slab.h>
29 #include <linux/user.h>
30 #include <linux/a.out.h>
31 #include <linux/tty.h>
32 #include <linux/major.h>
33 #include <linux/interrupt.h>
34 #include <linux/reboot.h>
35 #include <linux/init.h>
36 #include <linux/ioport.h>
37 #include <linux/console.h>
38 #include <linux/pci.h>
39 #include <linux/utsname.h>
40 #include <linux/adb.h>
41 #include <linux/module.h>
42 #include <linux/delay.h>
43 #include <linux/irq.h>
44 #include <linux/seq_file.h>
45 #include <linux/root_dev.h>
46
47 #include <asm/mmu.h>
48 #include <asm/processor.h>
49 #include <asm/io.h>
50 #include <asm/pgtable.h>
51 #include <asm/prom.h>
52 #include <asm/rtas.h>
53 #include <asm/pci-bridge.h>
54 #include <asm/iommu.h>
55 #include <asm/dma.h>
56 #include <asm/machdep.h>
57 #include <asm/irq.h>
58 #include <asm/kexec.h>
59 #include <asm/time.h>
60 #include <asm/nvram.h>
61 #include "xics.h"
62 #include <asm/pmc.h>
63 #include <asm/mpic.h>
64 #include <asm/ppc-pci.h>
65 #include <asm/i8259.h>
66 #include <asm/udbg.h>
67 #include <asm/smp.h>
68
69 #include "plpar_wrappers.h"
70 #include "ras.h"
71 #include "firmware.h"
72
73 #ifdef DEBUG
74 #define DBG(fmt...) udbg_printf(fmt)
75 #else
76 #define DBG(fmt...)
77 #endif
78
79 extern void find_udbg_vterm(void);
80
81 int fwnmi_active;  /* TRUE if an FWNMI handler is present */
82
83 static void pseries_shared_idle_sleep(void);
84 static void pseries_dedicated_idle_sleep(void);
85
86 struct mpic *pSeries_mpic;
87
88 static void pSeries_show_cpuinfo(struct seq_file *m)
89 {
90         struct device_node *root;
91         const char *model = "";
92
93         root = of_find_node_by_path("/");
94         if (root)
95                 model = get_property(root, "model", NULL);
96         seq_printf(m, "machine\t\t: CHRP %s\n", model);
97         of_node_put(root);
98 }
99
100 /* Initialize firmware assisted non-maskable interrupts if
101  * the firmware supports this feature.
102  */
103 static void __init fwnmi_init(void)
104 {
105         unsigned long system_reset_addr, machine_check_addr;
106
107         int ibm_nmi_register = rtas_token("ibm,nmi-register");
108         if (ibm_nmi_register == RTAS_UNKNOWN_SERVICE)
109                 return;
110
111         /* If the kernel's not linked at zero we point the firmware at low
112          * addresses anyway, and use a trampoline to get to the real code. */
113         system_reset_addr  = __pa(system_reset_fwnmi) - PHYSICAL_START;
114         machine_check_addr = __pa(machine_check_fwnmi) - PHYSICAL_START;
115
116         if (0 == rtas_call(ibm_nmi_register, 2, 1, NULL, system_reset_addr,
117                                 machine_check_addr))
118                 fwnmi_active = 1;
119 }
120
121 static void __init pSeries_init_mpic(void)
122 {
123         unsigned int *addrp;
124         struct device_node *np;
125         unsigned long intack = 0;
126
127         /* All ISUs are setup, complete initialization */
128         mpic_init(pSeries_mpic);
129
130         /* Check what kind of cascade ACK we have */
131         if (!(np = of_find_node_by_name(NULL, "pci"))
132             || !(addrp = (unsigned int *)
133                  get_property(np, "8259-interrupt-acknowledge", NULL)))
134                 printk(KERN_ERR "Cannot find pci to get ack address\n");
135         else
136                 intack = addrp[prom_n_addr_cells(np)-1];
137         of_node_put(np);
138
139         /* Setup the legacy interrupts & controller */
140         i8259_init(intack, 0);
141
142         /* Hook cascade to mpic */
143         mpic_setup_cascade(NUM_ISA_INTERRUPTS, i8259_irq_cascade, NULL);
144 }
145
146 static void __init pSeries_setup_mpic(void)
147 {
148         unsigned int *opprop;
149         unsigned long openpic_addr = 0;
150         unsigned char senses[NR_IRQS - NUM_ISA_INTERRUPTS];
151         struct device_node *root;
152         int irq_count;
153
154         /* Find the Open PIC if present */
155         root = of_find_node_by_path("/");
156         opprop = (unsigned int *) get_property(root, "platform-open-pic", NULL);
157         if (opprop != 0) {
158                 int n = prom_n_addr_cells(root);
159
160                 for (openpic_addr = 0; n > 0; --n)
161                         openpic_addr = (openpic_addr << 32) + *opprop++;
162                 printk(KERN_DEBUG "OpenPIC addr: %lx\n", openpic_addr);
163         }
164         of_node_put(root);
165
166         BUG_ON(openpic_addr == 0);
167
168         /* Get the sense values from OF */
169         prom_get_irq_senses(senses, NUM_ISA_INTERRUPTS, NR_IRQS);
170         
171         /* Setup the openpic driver */
172         irq_count = NR_IRQS - NUM_ISA_INTERRUPTS - 4; /* leave room for IPIs */
173         pSeries_mpic = mpic_alloc(openpic_addr, MPIC_PRIMARY,
174                                   16, 16, irq_count, /* isu size, irq offset, irq count */ 
175                                   NR_IRQS - 4, /* ipi offset */
176                                   senses, irq_count, /* sense & sense size */
177                                   " MPIC     ");
178 }
179
180 static void pseries_lpar_enable_pmcs(void)
181 {
182         unsigned long set, reset;
183
184         power4_enable_pmcs();
185
186         set = 1UL << 63;
187         reset = 0;
188         plpar_hcall_norets(H_PERFMON, set, reset);
189
190         /* instruct hypervisor to maintain PMCs */
191         if (firmware_has_feature(FW_FEATURE_SPLPAR))
192                 get_lppaca()->pmcregs_in_use = 1;
193 }
194
195 static void __init pSeries_setup_arch(void)
196 {
197         /* Fixup ppc_md depending on the type of interrupt controller */
198         if (ppc64_interrupt_controller == IC_OPEN_PIC) {
199                 ppc_md.init_IRQ       = pSeries_init_mpic;
200                 ppc_md.get_irq        = mpic_get_irq;
201                 /* Allocate the mpic now, so that find_and_init_phbs() can
202                  * fill the ISUs */
203                 pSeries_setup_mpic();
204         } else {
205                 ppc_md.init_IRQ       = xics_init_IRQ;
206                 ppc_md.get_irq        = xics_get_irq;
207         }
208
209 #ifdef CONFIG_SMP
210         smp_init_pSeries();
211 #endif
212         /* openpic global configuration register (64-bit format). */
213         /* openpic Interrupt Source Unit pointer (64-bit format). */
214         /* python0 facility area (mmio) (64-bit format) REAL address. */
215
216         /* init to some ~sane value until calibrate_delay() runs */
217         loops_per_jiffy = 50000000;
218
219         if (ROOT_DEV == 0) {
220                 printk("No ramdisk, default root is /dev/sda2\n");
221                 ROOT_DEV = Root_SDA2;
222         }
223
224         fwnmi_init();
225
226         /* Find and initialize PCI host bridges */
227         init_pci_config_tokens();
228         find_and_init_phbs();
229         eeh_init();
230
231         pSeries_nvram_init();
232
233         /* Choose an idle loop */
234         if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
235                 vpa_init(boot_cpuid);
236                 if (get_lppaca()->shared_proc) {
237                         printk(KERN_DEBUG "Using shared processor idle loop\n");
238                         ppc_md.power_save = pseries_shared_idle_sleep;
239                 } else {
240                         printk(KERN_DEBUG "Using dedicated idle loop\n");
241                         ppc_md.power_save = pseries_dedicated_idle_sleep;
242                 }
243         } else {
244                 printk(KERN_DEBUG "Using default idle loop\n");
245         }
246
247         if (firmware_has_feature(FW_FEATURE_LPAR))
248                 ppc_md.enable_pmcs = pseries_lpar_enable_pmcs;
249         else
250                 ppc_md.enable_pmcs = power4_enable_pmcs;
251 }
252
253 static int __init pSeries_init_panel(void)
254 {
255         /* Manually leave the kernel version on the panel. */
256         ppc_md.progress("Linux ppc64\n", 0);
257         ppc_md.progress(system_utsname.release, 0);
258
259         return 0;
260 }
261 arch_initcall(pSeries_init_panel);
262
263 static  void __init pSeries_discover_pic(void)
264 {
265         struct device_node *np;
266         char *typep;
267
268         /*
269          * Setup interrupt mapping options that are needed for finish_device_tree
270          * to properly parse the OF interrupt tree & do the virtual irq mapping
271          */
272         __irq_offset_value = NUM_ISA_INTERRUPTS;
273         ppc64_interrupt_controller = IC_INVALID;
274         for (np = NULL; (np = of_find_node_by_name(np, "interrupt-controller"));) {
275                 typep = (char *)get_property(np, "compatible", NULL);
276                 if (strstr(typep, "open-pic")) {
277                         ppc64_interrupt_controller = IC_OPEN_PIC;
278                         break;
279                 } else if (strstr(typep, "ppc-xicp")) {
280                         ppc64_interrupt_controller = IC_PPC_XIC;
281                         break;
282                 }
283         }
284         if (ppc64_interrupt_controller == IC_INVALID)
285                 printk("pSeries_discover_pic: failed to recognize"
286                         " interrupt-controller\n");
287
288 }
289
290 static void pSeries_mach_cpu_die(void)
291 {
292         local_irq_disable();
293         idle_task_exit();
294         /* Some hardware requires clearing the CPPR, while other hardware does not
295          * it is safe either way
296          */
297         pSeriesLP_cppr_info(0, 0);
298         rtas_stop_self();
299         /* Should never get here... */
300         BUG();
301         for(;;);
302 }
303
304 static int pseries_set_dabr(unsigned long dabr)
305 {
306         return plpar_hcall_norets(H_SET_DABR, dabr);
307 }
308
309 static int pseries_set_xdabr(unsigned long dabr)
310 {
311         /* We want to catch accesses from kernel and userspace */
312         return plpar_hcall_norets(H_SET_XDABR, dabr,
313                         H_DABRX_KERNEL | H_DABRX_USER);
314 }
315
316 /*
317  * Early initialization.  Relocation is on but do not reference unbolted pages
318  */
319 static void __init pSeries_init_early(void)
320 {
321         DBG(" -> pSeries_init_early()\n");
322
323         fw_feature_init();
324
325         if (firmware_has_feature(FW_FEATURE_LPAR))
326                 find_udbg_vterm();
327
328         if (firmware_has_feature(FW_FEATURE_DABR))
329                 ppc_md.set_dabr = pseries_set_dabr;
330         else if (firmware_has_feature(FW_FEATURE_XDABR))
331                 ppc_md.set_dabr = pseries_set_xdabr;
332
333         iommu_init_early_pSeries();
334
335         pSeries_discover_pic();
336
337         DBG(" <- pSeries_init_early()\n");
338 }
339
340
341 static int pSeries_check_legacy_ioport(unsigned int baseport)
342 {
343         struct device_node *np;
344
345 #define I8042_DATA_REG  0x60
346 #define FDC_BASE        0x3f0
347
348
349         switch(baseport) {
350         case I8042_DATA_REG:
351                 np = of_find_node_by_type(NULL, "8042");
352                 if (np == NULL)
353                         return -ENODEV;
354                 of_node_put(np);
355                 break;
356         case FDC_BASE:
357                 np = of_find_node_by_type(NULL, "fdc");
358                 if (np == NULL)
359                         return -ENODEV;
360                 of_node_put(np);
361                 break;
362         }
363         return 0;
364 }
365
366 /*
367  * Called very early, MMU is off, device-tree isn't unflattened
368  */
369
370 static int __init pSeries_probe_hypertas(unsigned long node,
371                                          const char *uname, int depth,
372                                          void *data)
373 {
374         if (depth != 1 ||
375             (strcmp(uname, "rtas") != 0 && strcmp(uname, "rtas@0") != 0))
376                 return 0;
377
378         if (of_get_flat_dt_prop(node, "ibm,hypertas-functions", NULL) != NULL)
379                 powerpc_firmware_features |= FW_FEATURE_LPAR;
380
381         if (firmware_has_feature(FW_FEATURE_LPAR))
382                 hpte_init_lpar();
383         else
384                 hpte_init_native();
385
386         return 1;
387 }
388
389 static int __init pSeries_probe(void)
390 {
391         unsigned long root = of_get_flat_dt_root();
392         char *dtype = of_get_flat_dt_prop(of_get_flat_dt_root(),
393                                           "device_type", NULL);
394         if (dtype == NULL)
395                 return 0;
396         if (strcmp(dtype, "chrp"))
397                 return 0;
398
399         /* Cell blades firmware claims to be chrp while it's not. Until this
400          * is fixed, we need to avoid those here.
401          */
402         if (of_flat_dt_is_compatible(root, "IBM,CPBW-1.0") ||
403             of_flat_dt_is_compatible(root, "IBM,CBEA"))
404                 return 0;
405
406         DBG("pSeries detected, looking for LPAR capability...\n");
407
408         /* Now try to figure out if we are running on LPAR */
409         of_scan_flat_dt(pSeries_probe_hypertas, NULL);
410
411         DBG("Machine is%s LPAR !\n",
412             (powerpc_firmware_features & FW_FEATURE_LPAR) ? "" : " not");
413
414         return 1;
415 }
416
417
418 DECLARE_PER_CPU(unsigned long, smt_snooze_delay);
419
420 static void pseries_dedicated_idle_sleep(void)
421
422         unsigned int cpu = smp_processor_id();
423         unsigned long start_snooze;
424         unsigned long *smt_snooze_delay = &__get_cpu_var(smt_snooze_delay);
425
426         /*
427          * Indicate to the HV that we are idle. Now would be
428          * a good time to find other work to dispatch.
429          */
430         get_lppaca()->idle = 1;
431
432         /*
433          * We come in with interrupts disabled, and need_resched()
434          * has been checked recently.  If we should poll for a little
435          * while, do so.
436          */
437         if (*smt_snooze_delay) {
438                 start_snooze = get_tb() +
439                         *smt_snooze_delay * tb_ticks_per_usec;
440                 local_irq_enable();
441                 set_thread_flag(TIF_POLLING_NRFLAG);
442
443                 while (get_tb() < start_snooze) {
444                         if (need_resched() || cpu_is_offline(cpu))
445                                 goto out;
446                         ppc64_runlatch_off();
447                         HMT_low();
448                         HMT_very_low();
449                 }
450
451                 HMT_medium();
452                 clear_thread_flag(TIF_POLLING_NRFLAG);
453                 smp_mb();
454                 local_irq_disable();
455                 if (need_resched() || cpu_is_offline(cpu))
456                         goto out;
457         }
458
459         /*
460          * Cede if the other thread is not idle, so that it can
461          * go single-threaded.  If the other thread is idle,
462          * we ask the hypervisor if it has pending work it
463          * wants to do and cede if it does.  Otherwise we keep
464          * polling in order to reduce interrupt latency.
465          *
466          * Doing the cede when the other thread is active will
467          * result in this thread going dormant, meaning the other
468          * thread gets to run in single-threaded (ST) mode, which
469          * is slightly faster than SMT mode with this thread at
470          * very low priority.  The cede enables interrupts, which
471          * doesn't matter here.
472          */
473         if (!lppaca[cpu ^ 1].idle || poll_pending() == H_PENDING)
474                 cede_processor();
475
476 out:
477         HMT_medium();
478         get_lppaca()->idle = 0;
479 }
480
481 static void pseries_shared_idle_sleep(void)
482 {
483         /*
484          * Indicate to the HV that we are idle. Now would be
485          * a good time to find other work to dispatch.
486          */
487         get_lppaca()->idle = 1;
488
489         /*
490          * Yield the processor to the hypervisor.  We return if
491          * an external interrupt occurs (which are driven prior
492          * to returning here) or if a prod occurs from another
493          * processor. When returning here, external interrupts
494          * are enabled.
495          */
496         cede_processor();
497
498         get_lppaca()->idle = 0;
499 }
500
501 static int pSeries_pci_probe_mode(struct pci_bus *bus)
502 {
503         if (firmware_has_feature(FW_FEATURE_LPAR))
504                 return PCI_PROBE_DEVTREE;
505         return PCI_PROBE_NORMAL;
506 }
507
508 #ifdef CONFIG_KEXEC
509 static void pseries_kexec_cpu_down(int crash_shutdown, int secondary)
510 {
511         /* Don't risk a hypervisor call if we're crashing */
512         if (firmware_has_feature(FW_FEATURE_SPLPAR) && !crash_shutdown) {
513                 unsigned long vpa = __pa(get_lppaca());
514
515                 if (unregister_vpa(hard_smp_processor_id(), vpa)) {
516                         printk("VPA deregistration of cpu %u (hw_cpu_id %d) "
517                                         "failed\n", smp_processor_id(),
518                                         hard_smp_processor_id());
519                 }
520         }
521
522         if (ppc64_interrupt_controller == IC_OPEN_PIC)
523                 mpic_teardown_this_cpu(secondary);
524         else
525                 xics_teardown_cpu(secondary);
526 }
527 #endif
528
529 define_machine(pseries) {
530         .name                   = "pSeries",
531         .probe                  = pSeries_probe,
532         .setup_arch             = pSeries_setup_arch,
533         .init_early             = pSeries_init_early,
534         .show_cpuinfo           = pSeries_show_cpuinfo,
535         .log_error              = pSeries_log_error,
536         .pcibios_fixup          = pSeries_final_fixup,
537         .pci_probe_mode         = pSeries_pci_probe_mode,
538         .irq_bus_setup          = pSeries_irq_bus_setup,
539         .restart                = rtas_restart,
540         .power_off              = rtas_power_off,
541         .halt                   = rtas_halt,
542         .panic                  = rtas_os_term,
543         .cpu_die                = pSeries_mach_cpu_die,
544         .get_boot_time          = rtas_get_boot_time,
545         .get_rtc_time           = rtas_get_rtc_time,
546         .set_rtc_time           = rtas_set_rtc_time,
547         .calibrate_decr         = generic_calibrate_decr,
548         .progress               = rtas_progress,
549         .check_legacy_ioport    = pSeries_check_legacy_ioport,
550         .system_reset_exception = pSeries_system_reset_exception,
551         .machine_check_exception = pSeries_machine_check_exception,
552 #ifdef CONFIG_KEXEC
553         .kexec_cpu_down         = pseries_kexec_cpu_down,
554         .machine_kexec          = default_machine_kexec,
555         .machine_kexec_prepare  = default_machine_kexec_prepare,
556         .machine_crash_shutdown = default_machine_crash_shutdown,
557 #endif
558 };