- Update to 3.3-rc2.
[linux-flexiantxendom0-3.2.10.git] / arch / powerpc / kernel / setup_64.c
1 /*
2  * 
3  * Common boot and setup code.
4  *
5  * Copyright (C) 2001 PPC64 Team, IBM Corp
6  *
7  *      This program is free software; you can redistribute it and/or
8  *      modify it under the terms of the GNU General Public License
9  *      as published by the Free Software Foundation; either version
10  *      2 of the License, or (at your option) any later version.
11  */
12
13 #undef DEBUG
14
15 #include <linux/export.h>
16 #include <linux/string.h>
17 #include <linux/sched.h>
18 #include <linux/init.h>
19 #include <linux/kernel.h>
20 #include <linux/reboot.h>
21 #include <linux/delay.h>
22 #include <linux/initrd.h>
23 #include <linux/seq_file.h>
24 #include <linux/ioport.h>
25 #include <linux/console.h>
26 #include <linux/utsname.h>
27 #include <linux/tty.h>
28 #include <linux/root_dev.h>
29 #include <linux/notifier.h>
30 #include <linux/cpu.h>
31 #include <linux/unistd.h>
32 #include <linux/serial.h>
33 #include <linux/serial_8250.h>
34 #include <linux/bootmem.h>
35 #include <linux/pci.h>
36 #include <linux/lockdep.h>
37 #include <linux/memblock.h>
38 #include <linux/hugetlb.h>
39
40 #include <asm/io.h>
41 #include <asm/kdump.h>
42 #include <asm/prom.h>
43 #include <asm/processor.h>
44 #include <asm/pgtable.h>
45 #include <asm/smp.h>
46 #include <asm/elf.h>
47 #include <asm/machdep.h>
48 #include <asm/paca.h>
49 #include <asm/time.h>
50 #include <asm/cputable.h>
51 #include <asm/sections.h>
52 #include <asm/btext.h>
53 #include <asm/nvram.h>
54 #include <asm/setup.h>
55 #include <asm/system.h>
56 #include <asm/rtas.h>
57 #include <asm/iommu.h>
58 #include <asm/serial.h>
59 #include <asm/cache.h>
60 #include <asm/page.h>
61 #include <asm/mmu.h>
62 #include <asm/firmware.h>
63 #include <asm/xmon.h>
64 #include <asm/udbg.h>
65 #include <asm/kexec.h>
66 #include <asm/mmu_context.h>
67 #include <asm/code-patching.h>
68 #include <asm/kvm_ppc.h>
69 #include <asm/hugetlb.h>
70
71 #include "setup.h"
72
73 #ifdef DEBUG
74 #define DBG(fmt...) udbg_printf(fmt)
75 #else
76 #define DBG(fmt...)
77 #endif
78
79 int boot_cpuid = 0;
80 int __initdata spinning_secondaries;
81 u64 ppc64_pft_size;
82
83 /* Pick defaults since we might want to patch instructions
84  * before we've read this from the device tree.
85  */
86 struct ppc64_caches ppc64_caches = {
87         .dline_size = 0x40,
88         .log_dline_size = 6,
89         .iline_size = 0x40,
90         .log_iline_size = 6
91 };
92 EXPORT_SYMBOL_GPL(ppc64_caches);
93
94 /*
95  * These are used in binfmt_elf.c to put aux entries on the stack
96  * for each elf executable being started.
97  */
98 int dcache_bsize;
99 int icache_bsize;
100 int ucache_bsize;
101
102 #ifdef CONFIG_SMP
103
104 static char *smt_enabled_cmdline;
105
106 /* Look for ibm,smt-enabled OF option */
107 static void check_smt_enabled(void)
108 {
109         struct device_node *dn;
110         const char *smt_option;
111
112         /* Default to enabling all threads */
113         smt_enabled_at_boot = threads_per_core;
114
115         /* Allow the command line to overrule the OF option */
116         if (smt_enabled_cmdline) {
117                 if (!strcmp(smt_enabled_cmdline, "on"))
118                         smt_enabled_at_boot = threads_per_core;
119                 else if (!strcmp(smt_enabled_cmdline, "off"))
120                         smt_enabled_at_boot = 0;
121                 else {
122                         long smt;
123                         int rc;
124
125                         rc = strict_strtol(smt_enabled_cmdline, 10, &smt);
126                         if (!rc)
127                                 smt_enabled_at_boot =
128                                         min(threads_per_core, (int)smt);
129                 }
130         } else {
131                 dn = of_find_node_by_path("/options");
132                 if (dn) {
133                         smt_option = of_get_property(dn, "ibm,smt-enabled",
134                                                      NULL);
135
136                         if (smt_option) {
137                                 if (!strcmp(smt_option, "on"))
138                                         smt_enabled_at_boot = threads_per_core;
139                                 else if (!strcmp(smt_option, "off"))
140                                         smt_enabled_at_boot = 0;
141                         }
142
143                         of_node_put(dn);
144                 }
145         }
146 }
147
148 /* Look for smt-enabled= cmdline option */
149 static int __init early_smt_enabled(char *p)
150 {
151         smt_enabled_cmdline = p;
152         return 0;
153 }
154 early_param("smt-enabled", early_smt_enabled);
155
156 #else
157 #define check_smt_enabled()
158 #endif /* CONFIG_SMP */
159
160 /*
161  * Early initialization entry point. This is called by head.S
162  * with MMU translation disabled. We rely on the "feature" of
163  * the CPU that ignores the top 2 bits of the address in real
164  * mode so we can access kernel globals normally provided we
165  * only toy with things in the RMO region. From here, we do
166  * some early parsing of the device-tree to setup out MEMBLOCK
167  * data structures, and allocate & initialize the hash table
168  * and segment tables so we can start running with translation
169  * enabled.
170  *
171  * It is this function which will call the probe() callback of
172  * the various platform types and copy the matching one to the
173  * global ppc_md structure. Your platform can eventually do
174  * some very early initializations from the probe() routine, but
175  * this is not recommended, be very careful as, for example, the
176  * device-tree is not accessible via normal means at this point.
177  */
178
179 void __init early_setup(unsigned long dt_ptr)
180 {
181         /* -------- printk is _NOT_ safe to use here ! ------- */
182
183         /* Identify CPU type */
184         identify_cpu(0, mfspr(SPRN_PVR));
185
186         /* Assume we're on cpu 0 for now. Don't write to the paca yet! */
187         initialise_paca(&boot_paca, 0);
188         setup_paca(&boot_paca);
189
190         /* Initialize lockdep early or else spinlocks will blow */
191         lockdep_init();
192
193         /* -------- printk is now safe to use ------- */
194
195         /* Enable early debugging if any specified (see udbg.h) */
196         udbg_early_init();
197
198         DBG(" -> early_setup(), dt_ptr: 0x%lx\n", dt_ptr);
199
200         /*
201          * Do early initialization using the flattened device
202          * tree, such as retrieving the physical memory map or
203          * calculating/retrieving the hash table size.
204          */
205         early_init_devtree(__va(dt_ptr));
206
207         /* Now we know the logical id of our boot cpu, setup the paca. */
208         setup_paca(&paca[boot_cpuid]);
209
210         /* Fix up paca fields required for the boot cpu */
211         get_paca()->cpu_start = 1;
212
213         /* Probe the machine type */
214         probe_machine();
215
216         setup_kdump_trampoline();
217
218         DBG("Found, Initializing memory management...\n");
219
220         /* Initialize the hash table or TLB handling */
221         early_init_mmu();
222
223         /*
224          * Reserve any gigantic pages requested on the command line.
225          * memblock needs to have been initialized by the time this is
226          * called since this will reserve memory.
227          */
228         reserve_hugetlb_gpages();
229
230         DBG(" <- early_setup()\n");
231 }
232
233 #ifdef CONFIG_SMP
234 void early_setup_secondary(void)
235 {
236         /* Mark interrupts enabled in PACA */
237         get_paca()->soft_enabled = 0;
238
239         /* Initialize the hash table or TLB handling */
240         early_init_mmu_secondary();
241 }
242
243 #endif /* CONFIG_SMP */
244
245 #if defined(CONFIG_SMP) || defined(CONFIG_KEXEC)
246 void smp_release_cpus(void)
247 {
248         unsigned long *ptr;
249         int i;
250
251         DBG(" -> smp_release_cpus()\n");
252
253         /* All secondary cpus are spinning on a common spinloop, release them
254          * all now so they can start to spin on their individual paca
255          * spinloops. For non SMP kernels, the secondary cpus never get out
256          * of the common spinloop.
257          */
258
259         ptr  = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
260                         - PHYSICAL_START);
261         *ptr = __pa(generic_secondary_smp_init);
262
263         /* And wait a bit for them to catch up */
264         for (i = 0; i < 100000; i++) {
265                 mb();
266                 HMT_low();
267                 if (spinning_secondaries == 0)
268                         break;
269                 udelay(1);
270         }
271         DBG("spinning_secondaries = %d\n", spinning_secondaries);
272
273         DBG(" <- smp_release_cpus()\n");
274 }
275 #endif /* CONFIG_SMP || CONFIG_KEXEC */
276
277 /*
278  * Initialize some remaining members of the ppc64_caches and systemcfg
279  * structures
280  * (at least until we get rid of them completely). This is mostly some
281  * cache informations about the CPU that will be used by cache flush
282  * routines and/or provided to userland
283  */
284 static void __init initialize_cache_info(void)
285 {
286         struct device_node *np;
287         unsigned long num_cpus = 0;
288
289         DBG(" -> initialize_cache_info()\n");
290
291         for_each_node_by_type(np, "cpu") {
292                 num_cpus += 1;
293
294                 /*
295                  * We're assuming *all* of the CPUs have the same
296                  * d-cache and i-cache sizes... -Peter
297                  */
298                 if (num_cpus == 1) {
299                         const u32 *sizep, *lsizep;
300                         u32 size, lsize;
301
302                         size = 0;
303                         lsize = cur_cpu_spec->dcache_bsize;
304                         sizep = of_get_property(np, "d-cache-size", NULL);
305                         if (sizep != NULL)
306                                 size = *sizep;
307                         lsizep = of_get_property(np, "d-cache-block-size",
308                                                  NULL);
309                         /* fallback if block size missing */
310                         if (lsizep == NULL)
311                                 lsizep = of_get_property(np,
312                                                          "d-cache-line-size",
313                                                          NULL);
314                         if (lsizep != NULL)
315                                 lsize = *lsizep;
316                         if (sizep == 0 || lsizep == 0)
317                                 DBG("Argh, can't find dcache properties ! "
318                                     "sizep: %p, lsizep: %p\n", sizep, lsizep);
319
320                         ppc64_caches.dsize = size;
321                         ppc64_caches.dline_size = lsize;
322                         ppc64_caches.log_dline_size = __ilog2(lsize);
323                         ppc64_caches.dlines_per_page = PAGE_SIZE / lsize;
324
325                         size = 0;
326                         lsize = cur_cpu_spec->icache_bsize;
327                         sizep = of_get_property(np, "i-cache-size", NULL);
328                         if (sizep != NULL)
329                                 size = *sizep;
330                         lsizep = of_get_property(np, "i-cache-block-size",
331                                                  NULL);
332                         if (lsizep == NULL)
333                                 lsizep = of_get_property(np,
334                                                          "i-cache-line-size",
335                                                          NULL);
336                         if (lsizep != NULL)
337                                 lsize = *lsizep;
338                         if (sizep == 0 || lsizep == 0)
339                                 DBG("Argh, can't find icache properties ! "
340                                     "sizep: %p, lsizep: %p\n", sizep, lsizep);
341
342                         ppc64_caches.isize = size;
343                         ppc64_caches.iline_size = lsize;
344                         ppc64_caches.log_iline_size = __ilog2(lsize);
345                         ppc64_caches.ilines_per_page = PAGE_SIZE / lsize;
346                 }
347         }
348
349         DBG(" <- initialize_cache_info()\n");
350 }
351
352
353 /*
354  * Do some initial setup of the system.  The parameters are those which 
355  * were passed in from the bootloader.
356  */
357 void __init setup_system(void)
358 {
359         DBG(" -> setup_system()\n");
360
361         /* Apply the CPUs-specific and firmware specific fixups to kernel
362          * text (nop out sections not relevant to this CPU or this firmware)
363          */
364         do_feature_fixups(cur_cpu_spec->cpu_features,
365                           &__start___ftr_fixup, &__stop___ftr_fixup);
366         do_feature_fixups(cur_cpu_spec->mmu_features,
367                           &__start___mmu_ftr_fixup, &__stop___mmu_ftr_fixup);
368         do_feature_fixups(powerpc_firmware_features,
369                           &__start___fw_ftr_fixup, &__stop___fw_ftr_fixup);
370         do_lwsync_fixups(cur_cpu_spec->cpu_features,
371                          &__start___lwsync_fixup, &__stop___lwsync_fixup);
372         do_final_fixups();
373
374         /*
375          * Unflatten the device-tree passed by prom_init or kexec
376          */
377         unflatten_device_tree();
378
379         /*
380          * Fill the ppc64_caches & systemcfg structures with informations
381          * retrieved from the device-tree.
382          */
383         initialize_cache_info();
384
385 #ifdef CONFIG_PPC_RTAS
386         /*
387          * Initialize RTAS if available
388          */
389         rtas_initialize();
390 #endif /* CONFIG_PPC_RTAS */
391
392         /*
393          * Check if we have an initrd provided via the device-tree
394          */
395         check_for_initrd();
396
397         /*
398          * Do some platform specific early initializations, that includes
399          * setting up the hash table pointers. It also sets up some interrupt-mapping
400          * related options that will be used by finish_device_tree()
401          */
402         if (ppc_md.init_early)
403                 ppc_md.init_early();
404
405         /*
406          * We can discover serial ports now since the above did setup the
407          * hash table management for us, thus ioremap works. We do that early
408          * so that further code can be debugged
409          */
410         find_legacy_serial_ports();
411
412         /*
413          * Register early console
414          */
415         register_early_udbg_console();
416
417         /*
418          * Initialize xmon
419          */
420         xmon_setup();
421
422         smp_setup_cpu_maps();
423         check_smt_enabled();
424
425 #ifdef CONFIG_SMP
426         /* Release secondary cpus out of their spinloops at 0x60 now that
427          * we can map physical -> logical CPU ids
428          */
429         smp_release_cpus();
430 #endif
431
432         printk("Starting Linux PPC64 %s\n", init_utsname()->version);
433
434         printk("-----------------------------------------------------\n");
435         printk("ppc64_pft_size                = 0x%llx\n", ppc64_pft_size);
436         printk("physicalMemorySize            = 0x%llx\n", memblock_phys_mem_size());
437         if (ppc64_caches.dline_size != 0x80)
438                 printk("ppc64_caches.dcache_line_size = 0x%x\n",
439                        ppc64_caches.dline_size);
440         if (ppc64_caches.iline_size != 0x80)
441                 printk("ppc64_caches.icache_line_size = 0x%x\n",
442                        ppc64_caches.iline_size);
443 #ifdef CONFIG_PPC_STD_MMU_64
444         if (htab_address)
445                 printk("htab_address                  = 0x%p\n", htab_address);
446         printk("htab_hash_mask                = 0x%lx\n", htab_hash_mask);
447 #endif /* CONFIG_PPC_STD_MMU_64 */
448         if (PHYSICAL_START > 0)
449                 printk("physical_start                = 0x%llx\n",
450                        (unsigned long long)PHYSICAL_START);
451         printk("-----------------------------------------------------\n");
452
453         DBG(" <- setup_system()\n");
454 }
455
456 /* This returns the limit below which memory accesses to the linear
457  * mapping are guarnateed not to cause a TLB or SLB miss. This is
458  * used to allocate interrupt or emergency stacks for which our
459  * exception entry path doesn't deal with being interrupted.
460  */
461 static u64 safe_stack_limit(void)
462 {
463 #ifdef CONFIG_PPC_BOOK3E
464         /* Freescale BookE bolts the entire linear mapping */
465         if (mmu_has_feature(MMU_FTR_TYPE_FSL_E))
466                 return linear_map_top;
467         /* Other BookE, we assume the first GB is bolted */
468         return 1ul << 30;
469 #else
470         /* BookS, the first segment is bolted */
471         if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
472                 return 1UL << SID_SHIFT_1T;
473         return 1UL << SID_SHIFT;
474 #endif
475 }
476
477 static void __init irqstack_early_init(void)
478 {
479         u64 limit = safe_stack_limit();
480         unsigned int i;
481
482         /*
483          * Interrupt stacks must be in the first segment since we
484          * cannot afford to take SLB misses on them.
485          */
486         for_each_possible_cpu(i) {
487                 softirq_ctx[i] = (struct thread_info *)
488                         __va(memblock_alloc_base(THREAD_SIZE,
489                                             THREAD_SIZE, limit));
490                 hardirq_ctx[i] = (struct thread_info *)
491                         __va(memblock_alloc_base(THREAD_SIZE,
492                                             THREAD_SIZE, limit));
493         }
494 }
495
496 #ifdef CONFIG_PPC_BOOK3E
497 static void __init exc_lvl_early_init(void)
498 {
499         extern unsigned int interrupt_base_book3e;
500         extern unsigned int exc_debug_debug_book3e;
501
502         unsigned int i;
503
504         for_each_possible_cpu(i) {
505                 critirq_ctx[i] = (struct thread_info *)
506                         __va(memblock_alloc(THREAD_SIZE, THREAD_SIZE));
507                 dbgirq_ctx[i] = (struct thread_info *)
508                         __va(memblock_alloc(THREAD_SIZE, THREAD_SIZE));
509                 mcheckirq_ctx[i] = (struct thread_info *)
510                         __va(memblock_alloc(THREAD_SIZE, THREAD_SIZE));
511         }
512
513         if (cpu_has_feature(CPU_FTR_DEBUG_LVL_EXC))
514                 patch_branch(&interrupt_base_book3e + (0x040 / 4) + 1,
515                              (unsigned long)&exc_debug_debug_book3e, 0);
516 }
517 #else
518 #define exc_lvl_early_init()
519 #endif
520
521 /*
522  * Stack space used when we detect a bad kernel stack pointer, and
523  * early in SMP boots before relocation is enabled.
524  */
525 static void __init emergency_stack_init(void)
526 {
527         u64 limit;
528         unsigned int i;
529
530         /*
531          * Emergency stacks must be under 256MB, we cannot afford to take
532          * SLB misses on them. The ABI also requires them to be 128-byte
533          * aligned.
534          *
535          * Since we use these as temporary stacks during secondary CPU
536          * bringup, we need to get at them in real mode. This means they
537          * must also be within the RMO region.
538          */
539         limit = min(safe_stack_limit(), ppc64_rma_size);
540
541         for_each_possible_cpu(i) {
542                 unsigned long sp;
543                 sp  = memblock_alloc_base(THREAD_SIZE, THREAD_SIZE, limit);
544                 sp += THREAD_SIZE;
545                 paca[i].emergency_sp = __va(sp);
546         }
547 }
548
549 /*
550  * Called into from start_kernel this initializes bootmem, which is used
551  * to manage page allocation until mem_init is called.
552  */
553 void __init setup_arch(char **cmdline_p)
554 {
555         ppc64_boot_msg(0x12, "Setup Arch");
556
557         *cmdline_p = cmd_line;
558
559         /*
560          * Set cache line size based on type of cpu as a default.
561          * Systems with OF can look in the properties on the cpu node(s)
562          * for a possibly more accurate value.
563          */
564         dcache_bsize = ppc64_caches.dline_size;
565         icache_bsize = ppc64_caches.iline_size;
566
567         /* reboot on panic */
568         panic_timeout = 180;
569
570         if (ppc_md.panic)
571                 setup_panic();
572
573         init_mm.start_code = (unsigned long)_stext;
574         init_mm.end_code = (unsigned long) _etext;
575         init_mm.end_data = (unsigned long) _edata;
576         init_mm.brk = klimit;
577         
578         irqstack_early_init();
579         exc_lvl_early_init();
580         emergency_stack_init();
581
582 #ifdef CONFIG_PPC_STD_MMU_64
583         stabs_alloc();
584 #endif
585         /* set up the bootmem stuff with available memory */
586         do_init_bootmem();
587         sparse_init();
588
589 #ifdef CONFIG_DUMMY_CONSOLE
590         conswitchp = &dummy_con;
591 #endif
592
593         if (ppc_md.setup_arch)
594                 ppc_md.setup_arch();
595
596         paging_init();
597
598         /* Initialize the MMU context management stuff */
599         mmu_context_init();
600
601         kvm_linear_init();
602
603         ppc64_boot_msg(0x15, "Setup Done");
604 }
605
606
607 /* ToDo: do something useful if ppc_md is not yet setup. */
608 #define PPC64_LINUX_FUNCTION 0x0f000000
609 #define PPC64_IPL_MESSAGE 0xc0000000
610 #define PPC64_TERM_MESSAGE 0xb0000000
611
612 static void ppc64_do_msg(unsigned int src, const char *msg)
613 {
614         if (ppc_md.progress) {
615                 char buf[128];
616
617                 sprintf(buf, "%08X\n", src);
618                 ppc_md.progress(buf, 0);
619                 snprintf(buf, 128, "%s", msg);
620                 ppc_md.progress(buf, 0);
621         }
622 }
623
624 /* Print a boot progress message. */
625 void ppc64_boot_msg(unsigned int src, const char *msg)
626 {
627         ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_IPL_MESSAGE|src, msg);
628         printk("[boot]%04x %s\n", src, msg);
629 }
630
631 #ifdef CONFIG_SMP
632 #define PCPU_DYN_SIZE           ()
633
634 static void * __init pcpu_fc_alloc(unsigned int cpu, size_t size, size_t align)
635 {
636         return __alloc_bootmem_node(NODE_DATA(cpu_to_node(cpu)), size, align,
637                                     __pa(MAX_DMA_ADDRESS));
638 }
639
640 static void __init pcpu_fc_free(void *ptr, size_t size)
641 {
642         free_bootmem(__pa(ptr), size);
643 }
644
645 static int pcpu_cpu_distance(unsigned int from, unsigned int to)
646 {
647         if (cpu_to_node(from) == cpu_to_node(to))
648                 return LOCAL_DISTANCE;
649         else
650                 return REMOTE_DISTANCE;
651 }
652
653 unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
654 EXPORT_SYMBOL(__per_cpu_offset);
655
656 void __init setup_per_cpu_areas(void)
657 {
658         const size_t dyn_size = PERCPU_MODULE_RESERVE + PERCPU_DYNAMIC_RESERVE;
659         size_t atom_size;
660         unsigned long delta;
661         unsigned int cpu;
662         int rc;
663
664         /*
665          * Linear mapping is one of 4K, 1M and 16M.  For 4K, no need
666          * to group units.  For larger mappings, use 1M atom which
667          * should be large enough to contain a number of units.
668          */
669         if (mmu_linear_psize == MMU_PAGE_4K)
670                 atom_size = PAGE_SIZE;
671         else
672                 atom_size = 1 << 20;
673
674         rc = pcpu_embed_first_chunk(0, dyn_size, atom_size, pcpu_cpu_distance,
675                                     pcpu_fc_alloc, pcpu_fc_free);
676         if (rc < 0)
677                 panic("cannot initialize percpu area (err=%d)", rc);
678
679         delta = (unsigned long)pcpu_base_addr - (unsigned long)__per_cpu_start;
680         for_each_possible_cpu(cpu) {
681                 __per_cpu_offset[cpu] = delta + pcpu_unit_offsets[cpu];
682                 paca[cpu].data_offset = __per_cpu_offset[cpu];
683         }
684 }
685 #endif
686
687
688 #ifdef CONFIG_PPC_INDIRECT_IO
689 struct ppc_pci_io ppc_pci_io;
690 EXPORT_SYMBOL(ppc_pci_io);
691 #endif /* CONFIG_PPC_INDIRECT_IO */
692