powerpc: Free up some CPU feature bits by moving out MMU-related features
[pandora-kernel.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/module.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 <asm/io.h>
39 #include <asm/kdump.h>
40 #include <asm/prom.h>
41 #include <asm/processor.h>
42 #include <asm/pgtable.h>
43 #include <asm/smp.h>
44 #include <asm/elf.h>
45 #include <asm/machdep.h>
46 #include <asm/paca.h>
47 #include <asm/time.h>
48 #include <asm/cputable.h>
49 #include <asm/sections.h>
50 #include <asm/btext.h>
51 #include <asm/nvram.h>
52 #include <asm/setup.h>
53 #include <asm/system.h>
54 #include <asm/rtas.h>
55 #include <asm/iommu.h>
56 #include <asm/serial.h>
57 #include <asm/cache.h>
58 #include <asm/page.h>
59 #include <asm/mmu.h>
60 #include <asm/firmware.h>
61 #include <asm/xmon.h>
62 #include <asm/udbg.h>
63 #include <asm/kexec.h>
64 #include <asm/mmu_context.h>
65
66 #include "setup.h"
67
68 #ifdef DEBUG
69 #define DBG(fmt...) udbg_printf(fmt)
70 #else
71 #define DBG(fmt...)
72 #endif
73
74 int boot_cpuid = 0;
75 int __initdata boot_cpu_count;
76 u64 ppc64_pft_size;
77
78 /* Pick defaults since we might want to patch instructions
79  * before we've read this from the device tree.
80  */
81 struct ppc64_caches ppc64_caches = {
82         .dline_size = 0x40,
83         .log_dline_size = 6,
84         .iline_size = 0x40,
85         .log_iline_size = 6
86 };
87 EXPORT_SYMBOL_GPL(ppc64_caches);
88
89 /*
90  * These are used in binfmt_elf.c to put aux entries on the stack
91  * for each elf executable being started.
92  */
93 int dcache_bsize;
94 int icache_bsize;
95 int ucache_bsize;
96
97 #ifdef CONFIG_SMP
98
99 static char *smt_enabled_cmdline;
100
101 /* Look for ibm,smt-enabled OF option */
102 static void check_smt_enabled(void)
103 {
104         struct device_node *dn;
105         const char *smt_option;
106
107         /* Default to enabling all threads */
108         smt_enabled_at_boot = threads_per_core;
109
110         /* Allow the command line to overrule the OF option */
111         if (smt_enabled_cmdline) {
112                 if (!strcmp(smt_enabled_cmdline, "on"))
113                         smt_enabled_at_boot = threads_per_core;
114                 else if (!strcmp(smt_enabled_cmdline, "off"))
115                         smt_enabled_at_boot = 0;
116                 else {
117                         long smt;
118                         int rc;
119
120                         rc = strict_strtol(smt_enabled_cmdline, 10, &smt);
121                         if (!rc)
122                                 smt_enabled_at_boot =
123                                         min(threads_per_core, (int)smt);
124                 }
125         } else {
126                 dn = of_find_node_by_path("/options");
127                 if (dn) {
128                         smt_option = of_get_property(dn, "ibm,smt-enabled",
129                                                      NULL);
130
131                         if (smt_option) {
132                                 if (!strcmp(smt_option, "on"))
133                                         smt_enabled_at_boot = threads_per_core;
134                                 else if (!strcmp(smt_option, "off"))
135                                         smt_enabled_at_boot = 0;
136                         }
137
138                         of_node_put(dn);
139                 }
140         }
141 }
142
143 /* Look for smt-enabled= cmdline option */
144 static int __init early_smt_enabled(char *p)
145 {
146         smt_enabled_cmdline = p;
147         return 0;
148 }
149 early_param("smt-enabled", early_smt_enabled);
150
151 #else
152 #define check_smt_enabled()
153 #endif /* CONFIG_SMP */
154
155 /*
156  * Early initialization entry point. This is called by head.S
157  * with MMU translation disabled. We rely on the "feature" of
158  * the CPU that ignores the top 2 bits of the address in real
159  * mode so we can access kernel globals normally provided we
160  * only toy with things in the RMO region. From here, we do
161  * some early parsing of the device-tree to setup out MEMBLOCK
162  * data structures, and allocate & initialize the hash table
163  * and segment tables so we can start running with translation
164  * enabled.
165  *
166  * It is this function which will call the probe() callback of
167  * the various platform types and copy the matching one to the
168  * global ppc_md structure. Your platform can eventually do
169  * some very early initializations from the probe() routine, but
170  * this is not recommended, be very careful as, for example, the
171  * device-tree is not accessible via normal means at this point.
172  */
173
174 void __init early_setup(unsigned long dt_ptr)
175 {
176         /* -------- printk is _NOT_ safe to use here ! ------- */
177
178         /* Identify CPU type */
179         identify_cpu(0, mfspr(SPRN_PVR));
180
181         /* Assume we're on cpu 0 for now. Don't write to the paca yet! */
182         initialise_paca(&boot_paca, 0);
183         setup_paca(&boot_paca);
184
185         /* Initialize lockdep early or else spinlocks will blow */
186         lockdep_init();
187
188         /* -------- printk is now safe to use ------- */
189
190         /* Enable early debugging if any specified (see udbg.h) */
191         udbg_early_init();
192
193         DBG(" -> early_setup(), dt_ptr: 0x%lx\n", dt_ptr);
194
195         /*
196          * Do early initialization using the flattened device
197          * tree, such as retrieving the physical memory map or
198          * calculating/retrieving the hash table size.
199          */
200         early_init_devtree(__va(dt_ptr));
201
202         /* Now we know the logical id of our boot cpu, setup the paca. */
203         setup_paca(&paca[boot_cpuid]);
204
205         /* Fix up paca fields required for the boot cpu */
206         get_paca()->cpu_start = 1;
207
208         /* Probe the machine type */
209         probe_machine();
210
211         setup_kdump_trampoline();
212
213         DBG("Found, Initializing memory management...\n");
214
215         /* Initialize the hash table or TLB handling */
216         early_init_mmu();
217
218         DBG(" <- early_setup()\n");
219 }
220
221 #ifdef CONFIG_SMP
222 void early_setup_secondary(void)
223 {
224         /* Mark interrupts enabled in PACA */
225         get_paca()->soft_enabled = 0;
226
227         /* Initialize the hash table or TLB handling */
228         early_init_mmu_secondary();
229 }
230
231 #endif /* CONFIG_SMP */
232
233 #if defined(CONFIG_SMP) || defined(CONFIG_KEXEC)
234 void smp_release_cpus(void)
235 {
236         unsigned long *ptr;
237         int i;
238
239         DBG(" -> smp_release_cpus()\n");
240
241         /* All secondary cpus are spinning on a common spinloop, release them
242          * all now so they can start to spin on their individual paca
243          * spinloops. For non SMP kernels, the secondary cpus never get out
244          * of the common spinloop.
245          */
246
247         ptr  = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
248                         - PHYSICAL_START);
249         *ptr = __pa(generic_secondary_smp_init);
250
251         /* And wait a bit for them to catch up */
252         for (i = 0; i < 100000; i++) {
253                 mb();
254                 HMT_low();
255                 if (boot_cpu_count == 0)
256                         break;
257                 udelay(1);
258         }
259         DBG("boot_cpu_count = %d\n", boot_cpu_count);
260
261         DBG(" <- smp_release_cpus()\n");
262 }
263 #endif /* CONFIG_SMP || CONFIG_KEXEC */
264
265 /*
266  * Initialize some remaining members of the ppc64_caches and systemcfg
267  * structures
268  * (at least until we get rid of them completely). This is mostly some
269  * cache informations about the CPU that will be used by cache flush
270  * routines and/or provided to userland
271  */
272 static void __init initialize_cache_info(void)
273 {
274         struct device_node *np;
275         unsigned long num_cpus = 0;
276
277         DBG(" -> initialize_cache_info()\n");
278
279         for (np = NULL; (np = of_find_node_by_type(np, "cpu"));) {
280                 num_cpus += 1;
281
282                 /* We're assuming *all* of the CPUs have the same
283                  * d-cache and i-cache sizes... -Peter
284                  */
285
286                 if ( num_cpus == 1 ) {
287                         const u32 *sizep, *lsizep;
288                         u32 size, lsize;
289
290                         size = 0;
291                         lsize = cur_cpu_spec->dcache_bsize;
292                         sizep = of_get_property(np, "d-cache-size", NULL);
293                         if (sizep != NULL)
294                                 size = *sizep;
295                         lsizep = of_get_property(np, "d-cache-block-size", NULL);
296                         /* fallback if block size missing */
297                         if (lsizep == NULL)
298                                 lsizep = of_get_property(np, "d-cache-line-size", NULL);
299                         if (lsizep != NULL)
300                                 lsize = *lsizep;
301                         if (sizep == 0 || lsizep == 0)
302                                 DBG("Argh, can't find dcache properties ! "
303                                     "sizep: %p, lsizep: %p\n", sizep, lsizep);
304
305                         ppc64_caches.dsize = size;
306                         ppc64_caches.dline_size = lsize;
307                         ppc64_caches.log_dline_size = __ilog2(lsize);
308                         ppc64_caches.dlines_per_page = PAGE_SIZE / lsize;
309
310                         size = 0;
311                         lsize = cur_cpu_spec->icache_bsize;
312                         sizep = of_get_property(np, "i-cache-size", NULL);
313                         if (sizep != NULL)
314                                 size = *sizep;
315                         lsizep = of_get_property(np, "i-cache-block-size", NULL);
316                         if (lsizep == NULL)
317                                 lsizep = of_get_property(np, "i-cache-line-size", NULL);
318                         if (lsizep != NULL)
319                                 lsize = *lsizep;
320                         if (sizep == 0 || lsizep == 0)
321                                 DBG("Argh, can't find icache properties ! "
322                                     "sizep: %p, lsizep: %p\n", sizep, lsizep);
323
324                         ppc64_caches.isize = size;
325                         ppc64_caches.iline_size = lsize;
326                         ppc64_caches.log_iline_size = __ilog2(lsize);
327                         ppc64_caches.ilines_per_page = PAGE_SIZE / lsize;
328                 }
329         }
330
331         DBG(" <- initialize_cache_info()\n");
332 }
333
334
335 /*
336  * Do some initial setup of the system.  The parameters are those which 
337  * were passed in from the bootloader.
338  */
339 void __init setup_system(void)
340 {
341         DBG(" -> setup_system()\n");
342
343         /* Apply the CPUs-specific and firmware specific fixups to kernel
344          * text (nop out sections not relevant to this CPU or this firmware)
345          */
346         do_feature_fixups(cur_cpu_spec->cpu_features,
347                           &__start___ftr_fixup, &__stop___ftr_fixup);
348         do_feature_fixups(cur_cpu_spec->mmu_features,
349                           &__start___mmu_ftr_fixup, &__stop___mmu_ftr_fixup);
350         do_feature_fixups(powerpc_firmware_features,
351                           &__start___fw_ftr_fixup, &__stop___fw_ftr_fixup);
352         do_lwsync_fixups(cur_cpu_spec->cpu_features,
353                          &__start___lwsync_fixup, &__stop___lwsync_fixup);
354
355         /*
356          * Unflatten the device-tree passed by prom_init or kexec
357          */
358         unflatten_device_tree();
359
360         /*
361          * Fill the ppc64_caches & systemcfg structures with informations
362          * retrieved from the device-tree.
363          */
364         initialize_cache_info();
365
366 #ifdef CONFIG_PPC_RTAS
367         /*
368          * Initialize RTAS if available
369          */
370         rtas_initialize();
371 #endif /* CONFIG_PPC_RTAS */
372
373         /*
374          * Check if we have an initrd provided via the device-tree
375          */
376         check_for_initrd();
377
378         /*
379          * Do some platform specific early initializations, that includes
380          * setting up the hash table pointers. It also sets up some interrupt-mapping
381          * related options that will be used by finish_device_tree()
382          */
383         if (ppc_md.init_early)
384                 ppc_md.init_early();
385
386         /*
387          * We can discover serial ports now since the above did setup the
388          * hash table management for us, thus ioremap works. We do that early
389          * so that further code can be debugged
390          */
391         find_legacy_serial_ports();
392
393         /*
394          * Register early console
395          */
396         register_early_udbg_console();
397
398         /*
399          * Initialize xmon
400          */
401         xmon_setup();
402
403         smp_setup_cpu_maps();
404         check_smt_enabled();
405
406 #ifdef CONFIG_SMP
407         /* Release secondary cpus out of their spinloops at 0x60 now that
408          * we can map physical -> logical CPU ids
409          */
410         smp_release_cpus();
411 #endif
412
413         printk("Starting Linux PPC64 %s\n", init_utsname()->version);
414
415         printk("-----------------------------------------------------\n");
416         printk("ppc64_pft_size                = 0x%llx\n", ppc64_pft_size);
417         printk("physicalMemorySize            = 0x%llx\n", memblock_phys_mem_size());
418         if (ppc64_caches.dline_size != 0x80)
419                 printk("ppc64_caches.dcache_line_size = 0x%x\n",
420                        ppc64_caches.dline_size);
421         if (ppc64_caches.iline_size != 0x80)
422                 printk("ppc64_caches.icache_line_size = 0x%x\n",
423                        ppc64_caches.iline_size);
424 #ifdef CONFIG_PPC_STD_MMU_64
425         if (htab_address)
426                 printk("htab_address                  = 0x%p\n", htab_address);
427         printk("htab_hash_mask                = 0x%lx\n", htab_hash_mask);
428 #endif /* CONFIG_PPC_STD_MMU_64 */
429         if (PHYSICAL_START > 0)
430                 printk("physical_start                = 0x%llx\n",
431                        (unsigned long long)PHYSICAL_START);
432         printk("-----------------------------------------------------\n");
433
434         DBG(" <- setup_system()\n");
435 }
436
437 static u64 slb0_limit(void)
438 {
439         if (mmu_has_feature(MMU_FTR_1T_SEGMENT)) {
440                 return 1UL << SID_SHIFT_1T;
441         }
442         return 1UL << SID_SHIFT;
443 }
444
445 static void __init irqstack_early_init(void)
446 {
447         u64 limit = slb0_limit();
448         unsigned int i;
449
450         /*
451          * Interrupt stacks must be in the first segment since we
452          * cannot afford to take SLB misses on them.
453          */
454         for_each_possible_cpu(i) {
455                 softirq_ctx[i] = (struct thread_info *)
456                         __va(memblock_alloc_base(THREAD_SIZE,
457                                             THREAD_SIZE, limit));
458                 hardirq_ctx[i] = (struct thread_info *)
459                         __va(memblock_alloc_base(THREAD_SIZE,
460                                             THREAD_SIZE, limit));
461         }
462 }
463
464 #ifdef CONFIG_PPC_BOOK3E
465 static void __init exc_lvl_early_init(void)
466 {
467         unsigned int i;
468
469         for_each_possible_cpu(i) {
470                 critirq_ctx[i] = (struct thread_info *)
471                         __va(memblock_alloc(THREAD_SIZE, THREAD_SIZE));
472                 dbgirq_ctx[i] = (struct thread_info *)
473                         __va(memblock_alloc(THREAD_SIZE, THREAD_SIZE));
474                 mcheckirq_ctx[i] = (struct thread_info *)
475                         __va(memblock_alloc(THREAD_SIZE, THREAD_SIZE));
476         }
477 }
478 #else
479 #define exc_lvl_early_init()
480 #endif
481
482 /*
483  * Stack space used when we detect a bad kernel stack pointer, and
484  * early in SMP boots before relocation is enabled.
485  */
486 static void __init emergency_stack_init(void)
487 {
488         u64 limit;
489         unsigned int i;
490
491         /*
492          * Emergency stacks must be under 256MB, we cannot afford to take
493          * SLB misses on them. The ABI also requires them to be 128-byte
494          * aligned.
495          *
496          * Since we use these as temporary stacks during secondary CPU
497          * bringup, we need to get at them in real mode. This means they
498          * must also be within the RMO region.
499          */
500         limit = min(slb0_limit(), ppc64_rma_size);
501
502         for_each_possible_cpu(i) {
503                 unsigned long sp;
504                 sp  = memblock_alloc_base(THREAD_SIZE, THREAD_SIZE, limit);
505                 sp += THREAD_SIZE;
506                 paca[i].emergency_sp = __va(sp);
507         }
508 }
509
510 /*
511  * Called into from start_kernel this initializes bootmem, which is used
512  * to manage page allocation until mem_init is called.
513  */
514 void __init setup_arch(char **cmdline_p)
515 {
516         ppc64_boot_msg(0x12, "Setup Arch");
517
518         *cmdline_p = cmd_line;
519
520         /*
521          * Set cache line size based on type of cpu as a default.
522          * Systems with OF can look in the properties on the cpu node(s)
523          * for a possibly more accurate value.
524          */
525         dcache_bsize = ppc64_caches.dline_size;
526         icache_bsize = ppc64_caches.iline_size;
527
528         /* reboot on panic */
529         panic_timeout = 180;
530
531         if (ppc_md.panic)
532                 setup_panic();
533
534         init_mm.start_code = (unsigned long)_stext;
535         init_mm.end_code = (unsigned long) _etext;
536         init_mm.end_data = (unsigned long) _edata;
537         init_mm.brk = klimit;
538         
539         irqstack_early_init();
540         exc_lvl_early_init();
541         emergency_stack_init();
542
543 #ifdef CONFIG_PPC_STD_MMU_64
544         stabs_alloc();
545 #endif
546         /* set up the bootmem stuff with available memory */
547         do_init_bootmem();
548         sparse_init();
549
550 #ifdef CONFIG_DUMMY_CONSOLE
551         conswitchp = &dummy_con;
552 #endif
553
554         if (ppc_md.setup_arch)
555                 ppc_md.setup_arch();
556
557         paging_init();
558
559         /* Initialize the MMU context management stuff */
560         mmu_context_init();
561
562         ppc64_boot_msg(0x15, "Setup Done");
563 }
564
565
566 /* ToDo: do something useful if ppc_md is not yet setup. */
567 #define PPC64_LINUX_FUNCTION 0x0f000000
568 #define PPC64_IPL_MESSAGE 0xc0000000
569 #define PPC64_TERM_MESSAGE 0xb0000000
570
571 static void ppc64_do_msg(unsigned int src, const char *msg)
572 {
573         if (ppc_md.progress) {
574                 char buf[128];
575
576                 sprintf(buf, "%08X\n", src);
577                 ppc_md.progress(buf, 0);
578                 snprintf(buf, 128, "%s", msg);
579                 ppc_md.progress(buf, 0);
580         }
581 }
582
583 /* Print a boot progress message. */
584 void ppc64_boot_msg(unsigned int src, const char *msg)
585 {
586         ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_IPL_MESSAGE|src, msg);
587         printk("[boot]%04x %s\n", src, msg);
588 }
589
590 #ifdef CONFIG_SMP
591 #define PCPU_DYN_SIZE           ()
592
593 static void * __init pcpu_fc_alloc(unsigned int cpu, size_t size, size_t align)
594 {
595         return __alloc_bootmem_node(NODE_DATA(cpu_to_node(cpu)), size, align,
596                                     __pa(MAX_DMA_ADDRESS));
597 }
598
599 static void __init pcpu_fc_free(void *ptr, size_t size)
600 {
601         free_bootmem(__pa(ptr), size);
602 }
603
604 static int pcpu_cpu_distance(unsigned int from, unsigned int to)
605 {
606         if (cpu_to_node(from) == cpu_to_node(to))
607                 return LOCAL_DISTANCE;
608         else
609                 return REMOTE_DISTANCE;
610 }
611
612 unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
613 EXPORT_SYMBOL(__per_cpu_offset);
614
615 void __init setup_per_cpu_areas(void)
616 {
617         const size_t dyn_size = PERCPU_MODULE_RESERVE + PERCPU_DYNAMIC_RESERVE;
618         size_t atom_size;
619         unsigned long delta;
620         unsigned int cpu;
621         int rc;
622
623         /*
624          * Linear mapping is one of 4K, 1M and 16M.  For 4K, no need
625          * to group units.  For larger mappings, use 1M atom which
626          * should be large enough to contain a number of units.
627          */
628         if (mmu_linear_psize == MMU_PAGE_4K)
629                 atom_size = PAGE_SIZE;
630         else
631                 atom_size = 1 << 20;
632
633         rc = pcpu_embed_first_chunk(0, dyn_size, atom_size, pcpu_cpu_distance,
634                                     pcpu_fc_alloc, pcpu_fc_free);
635         if (rc < 0)
636                 panic("cannot initialize percpu area (err=%d)", rc);
637
638         delta = (unsigned long)pcpu_base_addr - (unsigned long)__per_cpu_start;
639         for_each_possible_cpu(cpu) {
640                 __per_cpu_offset[cpu] = delta + pcpu_unit_offsets[cpu];
641                 paca[cpu].data_offset = __per_cpu_offset[cpu];
642         }
643 }
644 #endif
645
646
647 #ifdef CONFIG_PPC_INDIRECT_IO
648 struct ppc_pci_io ppc_pci_io;
649 EXPORT_SYMBOL(ppc_pci_io);
650 #endif /* CONFIG_PPC_INDIRECT_IO */
651