powerpc: Move boot_paca into early_setup
[linux-2.6-block.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/rtas.h>
56 #include <asm/iommu.h>
57 #include <asm/serial.h>
58 #include <asm/cache.h>
59 #include <asm/page.h>
60 #include <asm/mmu.h>
61 #include <asm/firmware.h>
62 #include <asm/xmon.h>
63 #include <asm/udbg.h>
64 #include <asm/kexec.h>
65 #include <asm/mmu_context.h>
66 #include <asm/code-patching.h>
67 #include <asm/kvm_ppc.h>
68 #include <asm/hugetlb.h>
69
70 #include "setup.h"
71
72 #ifdef DEBUG
73 #define DBG(fmt...) udbg_printf(fmt)
74 #else
75 #define DBG(fmt...)
76 #endif
77
78 int boot_cpuid = 0;
79 int __initdata spinning_secondaries;
80 u64 ppc64_pft_size;
81
82 /* Pick defaults since we might want to patch instructions
83  * before we've read this from the device tree.
84  */
85 struct ppc64_caches ppc64_caches = {
86         .dline_size = 0x40,
87         .log_dline_size = 6,
88         .iline_size = 0x40,
89         .log_iline_size = 6
90 };
91 EXPORT_SYMBOL_GPL(ppc64_caches);
92
93 /*
94  * These are used in binfmt_elf.c to put aux entries on the stack
95  * for each elf executable being started.
96  */
97 int dcache_bsize;
98 int icache_bsize;
99 int ucache_bsize;
100
101 #ifdef CONFIG_SMP
102
103 static char *smt_enabled_cmdline;
104
105 /* Look for ibm,smt-enabled OF option */
106 static void check_smt_enabled(void)
107 {
108         struct device_node *dn;
109         const char *smt_option;
110
111         /* Default to enabling all threads */
112         smt_enabled_at_boot = threads_per_core;
113
114         /* Allow the command line to overrule the OF option */
115         if (smt_enabled_cmdline) {
116                 if (!strcmp(smt_enabled_cmdline, "on"))
117                         smt_enabled_at_boot = threads_per_core;
118                 else if (!strcmp(smt_enabled_cmdline, "off"))
119                         smt_enabled_at_boot = 0;
120                 else {
121                         long smt;
122                         int rc;
123
124                         rc = strict_strtol(smt_enabled_cmdline, 10, &smt);
125                         if (!rc)
126                                 smt_enabled_at_boot =
127                                         min(threads_per_core, (int)smt);
128                 }
129         } else {
130                 dn = of_find_node_by_path("/options");
131                 if (dn) {
132                         smt_option = of_get_property(dn, "ibm,smt-enabled",
133                                                      NULL);
134
135                         if (smt_option) {
136                                 if (!strcmp(smt_option, "on"))
137                                         smt_enabled_at_boot = threads_per_core;
138                                 else if (!strcmp(smt_option, "off"))
139                                         smt_enabled_at_boot = 0;
140                         }
141
142                         of_node_put(dn);
143                 }
144         }
145 }
146
147 /* Look for smt-enabled= cmdline option */
148 static int __init early_smt_enabled(char *p)
149 {
150         smt_enabled_cmdline = p;
151         return 0;
152 }
153 early_param("smt-enabled", early_smt_enabled);
154
155 #else
156 #define check_smt_enabled()
157 #endif /* CONFIG_SMP */
158
159 /*
160  * Early initialization entry point. This is called by head.S
161  * with MMU translation disabled. We rely on the "feature" of
162  * the CPU that ignores the top 2 bits of the address in real
163  * mode so we can access kernel globals normally provided we
164  * only toy with things in the RMO region. From here, we do
165  * some early parsing of the device-tree to setup out MEMBLOCK
166  * data structures, and allocate & initialize the hash table
167  * and segment tables so we can start running with translation
168  * enabled.
169  *
170  * It is this function which will call the probe() callback of
171  * the various platform types and copy the matching one to the
172  * global ppc_md structure. Your platform can eventually do
173  * some very early initializations from the probe() routine, but
174  * this is not recommended, be very careful as, for example, the
175  * device-tree is not accessible via normal means at this point.
176  */
177
178 void __init early_setup(unsigned long dt_ptr)
179 {
180         static __initdata struct paca_struct boot_paca;
181
182         /* -------- printk is _NOT_ safe to use here ! ------- */
183
184         /* Identify CPU type */
185         identify_cpu(0, mfspr(SPRN_PVR));
186
187         /* Assume we're on cpu 0 for now. Don't write to the paca yet! */
188         initialise_paca(&boot_paca, 0);
189         setup_paca(&boot_paca);
190
191         /* Initialize lockdep early or else spinlocks will blow */
192         lockdep_init();
193
194         /* -------- printk is now safe to use ------- */
195
196         /* Enable early debugging if any specified (see udbg.h) */
197         udbg_early_init();
198
199         DBG(" -> early_setup(), dt_ptr: 0x%lx\n", dt_ptr);
200
201         /*
202          * Do early initialization using the flattened device
203          * tree, such as retrieving the physical memory map or
204          * calculating/retrieving the hash table size.
205          */
206         early_init_devtree(__va(dt_ptr));
207
208         /* Now we know the logical id of our boot cpu, setup the paca. */
209         setup_paca(&paca[boot_cpuid]);
210
211         /* Fix up paca fields required for the boot cpu */
212         get_paca()->cpu_start = 1;
213         /* Allow percpu accesses to "work" until we setup percpu data */
214         get_paca()->data_offset = 0;
215
216         /* Probe the machine type */
217         probe_machine();
218
219         setup_kdump_trampoline();
220
221         DBG("Found, Initializing memory management...\n");
222
223         /* Initialize the hash table or TLB handling */
224         early_init_mmu();
225
226         /*
227          * Reserve any gigantic pages requested on the command line.
228          * memblock needs to have been initialized by the time this is
229          * called since this will reserve memory.
230          */
231         reserve_hugetlb_gpages();
232
233         DBG(" <- early_setup()\n");
234 }
235
236 #ifdef CONFIG_SMP
237 void early_setup_secondary(void)
238 {
239         /* Mark interrupts enabled in PACA */
240         get_paca()->soft_enabled = 0;
241
242         /* Initialize the hash table or TLB handling */
243         early_init_mmu_secondary();
244 }
245
246 #endif /* CONFIG_SMP */
247
248 #if defined(CONFIG_SMP) || defined(CONFIG_KEXEC)
249 void smp_release_cpus(void)
250 {
251         unsigned long *ptr;
252         int i;
253
254         DBG(" -> smp_release_cpus()\n");
255
256         /* All secondary cpus are spinning on a common spinloop, release them
257          * all now so they can start to spin on their individual paca
258          * spinloops. For non SMP kernels, the secondary cpus never get out
259          * of the common spinloop.
260          */
261
262         ptr  = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
263                         - PHYSICAL_START);
264         *ptr = __pa(generic_secondary_smp_init);
265
266         /* And wait a bit for them to catch up */
267         for (i = 0; i < 100000; i++) {
268                 mb();
269                 HMT_low();
270                 if (spinning_secondaries == 0)
271                         break;
272                 udelay(1);
273         }
274         DBG("spinning_secondaries = %d\n", spinning_secondaries);
275
276         DBG(" <- smp_release_cpus()\n");
277 }
278 #endif /* CONFIG_SMP || CONFIG_KEXEC */
279
280 /*
281  * Initialize some remaining members of the ppc64_caches and systemcfg
282  * structures
283  * (at least until we get rid of them completely). This is mostly some
284  * cache informations about the CPU that will be used by cache flush
285  * routines and/or provided to userland
286  */
287 static void __init initialize_cache_info(void)
288 {
289         struct device_node *np;
290         unsigned long num_cpus = 0;
291
292         DBG(" -> initialize_cache_info()\n");
293
294         for_each_node_by_type(np, "cpu") {
295                 num_cpus += 1;
296
297                 /*
298                  * We're assuming *all* of the CPUs have the same
299                  * d-cache and i-cache sizes... -Peter
300                  */
301                 if (num_cpus == 1) {
302                         const u32 *sizep, *lsizep;
303                         u32 size, lsize;
304
305                         size = 0;
306                         lsize = cur_cpu_spec->dcache_bsize;
307                         sizep = of_get_property(np, "d-cache-size", NULL);
308                         if (sizep != NULL)
309                                 size = *sizep;
310                         lsizep = of_get_property(np, "d-cache-block-size",
311                                                  NULL);
312                         /* fallback if block size missing */
313                         if (lsizep == NULL)
314                                 lsizep = of_get_property(np,
315                                                          "d-cache-line-size",
316                                                          NULL);
317                         if (lsizep != NULL)
318                                 lsize = *lsizep;
319                         if (sizep == 0 || lsizep == 0)
320                                 DBG("Argh, can't find dcache properties ! "
321                                     "sizep: %p, lsizep: %p\n", sizep, lsizep);
322
323                         ppc64_caches.dsize = size;
324                         ppc64_caches.dline_size = lsize;
325                         ppc64_caches.log_dline_size = __ilog2(lsize);
326                         ppc64_caches.dlines_per_page = PAGE_SIZE / lsize;
327
328                         size = 0;
329                         lsize = cur_cpu_spec->icache_bsize;
330                         sizep = of_get_property(np, "i-cache-size", NULL);
331                         if (sizep != NULL)
332                                 size = *sizep;
333                         lsizep = of_get_property(np, "i-cache-block-size",
334                                                  NULL);
335                         if (lsizep == NULL)
336                                 lsizep = of_get_property(np,
337                                                          "i-cache-line-size",
338                                                          NULL);
339                         if (lsizep != NULL)
340                                 lsize = *lsizep;
341                         if (sizep == 0 || lsizep == 0)
342                                 DBG("Argh, can't find icache properties ! "
343                                     "sizep: %p, lsizep: %p\n", sizep, lsizep);
344
345                         ppc64_caches.isize = size;
346                         ppc64_caches.iline_size = lsize;
347                         ppc64_caches.log_iline_size = __ilog2(lsize);
348                         ppc64_caches.ilines_per_page = PAGE_SIZE / lsize;
349                 }
350         }
351
352         DBG(" <- initialize_cache_info()\n");
353 }
354
355
356 /*
357  * Do some initial setup of the system.  The parameters are those which 
358  * were passed in from the bootloader.
359  */
360 void __init setup_system(void)
361 {
362         DBG(" -> setup_system()\n");
363
364         /* Apply the CPUs-specific and firmware specific fixups to kernel
365          * text (nop out sections not relevant to this CPU or this firmware)
366          */
367         do_feature_fixups(cur_cpu_spec->cpu_features,
368                           &__start___ftr_fixup, &__stop___ftr_fixup);
369         do_feature_fixups(cur_cpu_spec->mmu_features,
370                           &__start___mmu_ftr_fixup, &__stop___mmu_ftr_fixup);
371         do_feature_fixups(powerpc_firmware_features,
372                           &__start___fw_ftr_fixup, &__stop___fw_ftr_fixup);
373         do_lwsync_fixups(cur_cpu_spec->cpu_features,
374                          &__start___lwsync_fixup, &__stop___lwsync_fixup);
375         do_final_fixups();
376
377         /*
378          * Unflatten the device-tree passed by prom_init or kexec
379          */
380         unflatten_device_tree();
381
382         /*
383          * Fill the ppc64_caches & systemcfg structures with informations
384          * retrieved from the device-tree.
385          */
386         initialize_cache_info();
387
388 #ifdef CONFIG_PPC_RTAS
389         /*
390          * Initialize RTAS if available
391          */
392         rtas_initialize();
393 #endif /* CONFIG_PPC_RTAS */
394
395         /*
396          * Check if we have an initrd provided via the device-tree
397          */
398         check_for_initrd();
399
400         /*
401          * Do some platform specific early initializations, that includes
402          * setting up the hash table pointers. It also sets up some interrupt-mapping
403          * related options that will be used by finish_device_tree()
404          */
405         if (ppc_md.init_early)
406                 ppc_md.init_early();
407
408         /*
409          * We can discover serial ports now since the above did setup the
410          * hash table management for us, thus ioremap works. We do that early
411          * so that further code can be debugged
412          */
413         find_legacy_serial_ports();
414
415         /*
416          * Register early console
417          */
418         register_early_udbg_console();
419
420         /*
421          * Initialize xmon
422          */
423         xmon_setup();
424
425         smp_setup_cpu_maps();
426         check_smt_enabled();
427
428 #ifdef CONFIG_SMP
429         /* Release secondary cpus out of their spinloops at 0x60 now that
430          * we can map physical -> logical CPU ids
431          */
432         smp_release_cpus();
433 #endif
434
435         printk("Starting Linux PPC64 %s\n", init_utsname()->version);
436
437         printk("-----------------------------------------------------\n");
438         printk("ppc64_pft_size                = 0x%llx\n", ppc64_pft_size);
439         printk("physicalMemorySize            = 0x%llx\n", memblock_phys_mem_size());
440         if (ppc64_caches.dline_size != 0x80)
441                 printk("ppc64_caches.dcache_line_size = 0x%x\n",
442                        ppc64_caches.dline_size);
443         if (ppc64_caches.iline_size != 0x80)
444                 printk("ppc64_caches.icache_line_size = 0x%x\n",
445                        ppc64_caches.iline_size);
446 #ifdef CONFIG_PPC_STD_MMU_64
447         if (htab_address)
448                 printk("htab_address                  = 0x%p\n", htab_address);
449         printk("htab_hash_mask                = 0x%lx\n", htab_hash_mask);
450 #endif /* CONFIG_PPC_STD_MMU_64 */
451         if (PHYSICAL_START > 0)
452                 printk("physical_start                = 0x%llx\n",
453                        (unsigned long long)PHYSICAL_START);
454         printk("-----------------------------------------------------\n");
455
456         DBG(" <- setup_system()\n");
457 }
458
459 /* This returns the limit below which memory accesses to the linear
460  * mapping are guarnateed not to cause a TLB or SLB miss. This is
461  * used to allocate interrupt or emergency stacks for which our
462  * exception entry path doesn't deal with being interrupted.
463  */
464 static u64 safe_stack_limit(void)
465 {
466 #ifdef CONFIG_PPC_BOOK3E
467         /* Freescale BookE bolts the entire linear mapping */
468         if (mmu_has_feature(MMU_FTR_TYPE_FSL_E))
469                 return linear_map_top;
470         /* Other BookE, we assume the first GB is bolted */
471         return 1ul << 30;
472 #else
473         /* BookS, the first segment is bolted */
474         if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
475                 return 1UL << SID_SHIFT_1T;
476         return 1UL << SID_SHIFT;
477 #endif
478 }
479
480 static void __init irqstack_early_init(void)
481 {
482         u64 limit = safe_stack_limit();
483         unsigned int i;
484
485         /*
486          * Interrupt stacks must be in the first segment since we
487          * cannot afford to take SLB misses on them.
488          */
489         for_each_possible_cpu(i) {
490                 softirq_ctx[i] = (struct thread_info *)
491                         __va(memblock_alloc_base(THREAD_SIZE,
492                                             THREAD_SIZE, limit));
493                 hardirq_ctx[i] = (struct thread_info *)
494                         __va(memblock_alloc_base(THREAD_SIZE,
495                                             THREAD_SIZE, limit));
496         }
497 }
498
499 #ifdef CONFIG_PPC_BOOK3E
500 static void __init exc_lvl_early_init(void)
501 {
502         extern unsigned int interrupt_base_book3e;
503         extern unsigned int exc_debug_debug_book3e;
504
505         unsigned int i;
506
507         for_each_possible_cpu(i) {
508                 critirq_ctx[i] = (struct thread_info *)
509                         __va(memblock_alloc(THREAD_SIZE, THREAD_SIZE));
510                 dbgirq_ctx[i] = (struct thread_info *)
511                         __va(memblock_alloc(THREAD_SIZE, THREAD_SIZE));
512                 mcheckirq_ctx[i] = (struct thread_info *)
513                         __va(memblock_alloc(THREAD_SIZE, THREAD_SIZE));
514         }
515
516         if (cpu_has_feature(CPU_FTR_DEBUG_LVL_EXC))
517                 patch_branch(&interrupt_base_book3e + (0x040 / 4) + 1,
518                              (unsigned long)&exc_debug_debug_book3e, 0);
519 }
520 #else
521 #define exc_lvl_early_init()
522 #endif
523
524 /*
525  * Stack space used when we detect a bad kernel stack pointer, and
526  * early in SMP boots before relocation is enabled.
527  */
528 static void __init emergency_stack_init(void)
529 {
530         u64 limit;
531         unsigned int i;
532
533         /*
534          * Emergency stacks must be under 256MB, we cannot afford to take
535          * SLB misses on them. The ABI also requires them to be 128-byte
536          * aligned.
537          *
538          * Since we use these as temporary stacks during secondary CPU
539          * bringup, we need to get at them in real mode. This means they
540          * must also be within the RMO region.
541          */
542         limit = min(safe_stack_limit(), ppc64_rma_size);
543
544         for_each_possible_cpu(i) {
545                 unsigned long sp;
546                 sp  = memblock_alloc_base(THREAD_SIZE, THREAD_SIZE, limit);
547                 sp += THREAD_SIZE;
548                 paca[i].emergency_sp = __va(sp);
549         }
550 }
551
552 /*
553  * Called into from start_kernel this initializes bootmem, which is used
554  * to manage page allocation until mem_init is called.
555  */
556 void __init setup_arch(char **cmdline_p)
557 {
558         ppc64_boot_msg(0x12, "Setup Arch");
559
560         *cmdline_p = cmd_line;
561
562         /*
563          * Set cache line size based on type of cpu as a default.
564          * Systems with OF can look in the properties on the cpu node(s)
565          * for a possibly more accurate value.
566          */
567         dcache_bsize = ppc64_caches.dline_size;
568         icache_bsize = ppc64_caches.iline_size;
569
570         /* reboot on panic */
571         panic_timeout = 180;
572
573         if (ppc_md.panic)
574                 setup_panic();
575
576         init_mm.start_code = (unsigned long)_stext;
577         init_mm.end_code = (unsigned long) _etext;
578         init_mm.end_data = (unsigned long) _edata;
579         init_mm.brk = klimit;
580         
581         irqstack_early_init();
582         exc_lvl_early_init();
583         emergency_stack_init();
584
585 #ifdef CONFIG_PPC_STD_MMU_64
586         stabs_alloc();
587 #endif
588         /* set up the bootmem stuff with available memory */
589         do_init_bootmem();
590         sparse_init();
591
592 #ifdef CONFIG_DUMMY_CONSOLE
593         conswitchp = &dummy_con;
594 #endif
595
596         if (ppc_md.setup_arch)
597                 ppc_md.setup_arch();
598
599         paging_init();
600
601         /* Initialize the MMU context management stuff */
602         mmu_context_init();
603
604         kvm_linear_init();
605
606         /* Interrupt code needs to be 64K-aligned */
607         if ((unsigned long)_stext & 0xffff)
608                 panic("Kernelbase not 64K-aligned (0x%lx)!\n",
609                       (unsigned long)_stext);
610
611         ppc64_boot_msg(0x15, "Setup Done");
612 }
613
614
615 /* ToDo: do something useful if ppc_md is not yet setup. */
616 #define PPC64_LINUX_FUNCTION 0x0f000000
617 #define PPC64_IPL_MESSAGE 0xc0000000
618 #define PPC64_TERM_MESSAGE 0xb0000000
619
620 static void ppc64_do_msg(unsigned int src, const char *msg)
621 {
622         if (ppc_md.progress) {
623                 char buf[128];
624
625                 sprintf(buf, "%08X\n", src);
626                 ppc_md.progress(buf, 0);
627                 snprintf(buf, 128, "%s", msg);
628                 ppc_md.progress(buf, 0);
629         }
630 }
631
632 /* Print a boot progress message. */
633 void ppc64_boot_msg(unsigned int src, const char *msg)
634 {
635         ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_IPL_MESSAGE|src, msg);
636         printk("[boot]%04x %s\n", src, msg);
637 }
638
639 #ifdef CONFIG_SMP
640 #define PCPU_DYN_SIZE           ()
641
642 static void * __init pcpu_fc_alloc(unsigned int cpu, size_t size, size_t align)
643 {
644         return __alloc_bootmem_node(NODE_DATA(cpu_to_node(cpu)), size, align,
645                                     __pa(MAX_DMA_ADDRESS));
646 }
647
648 static void __init pcpu_fc_free(void *ptr, size_t size)
649 {
650         free_bootmem(__pa(ptr), size);
651 }
652
653 static int pcpu_cpu_distance(unsigned int from, unsigned int to)
654 {
655         if (cpu_to_node(from) == cpu_to_node(to))
656                 return LOCAL_DISTANCE;
657         else
658                 return REMOTE_DISTANCE;
659 }
660
661 unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
662 EXPORT_SYMBOL(__per_cpu_offset);
663
664 void __init setup_per_cpu_areas(void)
665 {
666         const size_t dyn_size = PERCPU_MODULE_RESERVE + PERCPU_DYNAMIC_RESERVE;
667         size_t atom_size;
668         unsigned long delta;
669         unsigned int cpu;
670         int rc;
671
672         /*
673          * Linear mapping is one of 4K, 1M and 16M.  For 4K, no need
674          * to group units.  For larger mappings, use 1M atom which
675          * should be large enough to contain a number of units.
676          */
677         if (mmu_linear_psize == MMU_PAGE_4K)
678                 atom_size = PAGE_SIZE;
679         else
680                 atom_size = 1 << 20;
681
682         rc = pcpu_embed_first_chunk(0, dyn_size, atom_size, pcpu_cpu_distance,
683                                     pcpu_fc_alloc, pcpu_fc_free);
684         if (rc < 0)
685                 panic("cannot initialize percpu area (err=%d)", rc);
686
687         delta = (unsigned long)pcpu_base_addr - (unsigned long)__per_cpu_start;
688         for_each_possible_cpu(cpu) {
689                 __per_cpu_offset[cpu] = delta + pcpu_unit_offsets[cpu];
690                 paca[cpu].data_offset = __per_cpu_offset[cpu];
691         }
692 }
693 #endif
694
695
696 #ifdef CONFIG_PPC_INDIRECT_IO
697 struct ppc_pci_io ppc_pci_io;
698 EXPORT_SYMBOL(ppc_pci_io);
699 #endif /* CONFIG_PPC_INDIRECT_IO */
700