powerpc/book3e-64/kexec: create an identity TLB mapping
[linux-2.6-block.git] / arch / powerpc / kernel / setup_64.c
CommitLineData
40ef8cbc
PM
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
7191b615 13#define DEBUG
40ef8cbc 14
4b16f8e2 15#include <linux/export.h>
40ef8cbc
PM
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>
40ef8cbc
PM
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>
7a0268fa 34#include <linux/bootmem.h>
12d04eef 35#include <linux/pci.h>
945feb17 36#include <linux/lockdep.h>
95f72d1e 37#include <linux/memblock.h>
a6146888 38#include <linux/hugetlb.h>
a5d86257 39#include <linux/memory.h>
c54b2bf1 40#include <linux/nmi.h>
a6146888 41
40ef8cbc 42#include <asm/io.h>
0cc4746c 43#include <asm/kdump.h>
40ef8cbc
PM
44#include <asm/prom.h>
45#include <asm/processor.h>
46#include <asm/pgtable.h>
40ef8cbc
PM
47#include <asm/smp.h>
48#include <asm/elf.h>
49#include <asm/machdep.h>
50#include <asm/paca.h>
40ef8cbc
PM
51#include <asm/time.h>
52#include <asm/cputable.h>
53#include <asm/sections.h>
54#include <asm/btext.h>
55#include <asm/nvram.h>
56#include <asm/setup.h>
40ef8cbc
PM
57#include <asm/rtas.h>
58#include <asm/iommu.h>
59#include <asm/serial.h>
60#include <asm/cache.h>
61#include <asm/page.h>
62#include <asm/mmu.h>
40ef8cbc 63#include <asm/firmware.h>
f78541dc 64#include <asm/xmon.h>
dcad47fc 65#include <asm/udbg.h>
593e537b 66#include <asm/kexec.h>
25d21ad6 67#include <asm/mmu_context.h>
d36b4c4f 68#include <asm/code-patching.h>
aa04b4cc 69#include <asm/kvm_ppc.h>
a6146888 70#include <asm/hugetlb.h>
4e21b94c 71#include <asm/epapr_hcalls.h>
40ef8cbc
PM
72
73#ifdef DEBUG
74#define DBG(fmt...) udbg_printf(fmt)
75#else
76#define DBG(fmt...)
77#endif
78
8246aca7 79int spinning_secondaries;
40ef8cbc
PM
80u64 ppc64_pft_size;
81
dabcafd3
OJ
82/* Pick defaults since we might want to patch instructions
83 * before we've read this from the device tree.
84 */
85struct ppc64_caches ppc64_caches = {
5a2fe38d
OJ
86 .dline_size = 0x40,
87 .log_dline_size = 6,
88 .iline_size = 0x40,
89 .log_iline_size = 6
dabcafd3 90};
40ef8cbc
PM
91EXPORT_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 */
97int dcache_bsize;
98int icache_bsize;
99int ucache_bsize;
100
28efc35f
SW
101#if defined(CONFIG_PPC_BOOK3E) && defined(CONFIG_SMP)
102static void setup_tlb_core_data(void)
103{
104 int cpu;
105
82d86de2
SW
106 BUILD_BUG_ON(offsetof(struct tlb_core_data, lock) != 0);
107
28efc35f
SW
108 for_each_possible_cpu(cpu) {
109 int first = cpu_first_thread_sibling(cpu);
110
d9e1831a
SW
111 /*
112 * If we boot via kdump on a non-primary thread,
113 * make sure we point at the thread that actually
114 * set up this TLB.
115 */
116 if (cpu_first_thread_sibling(boot_cpuid) == first)
117 first = boot_cpuid;
118
28efc35f
SW
119 paca[cpu].tcd_ptr = &paca[first].tcd;
120
121 /*
122 * If we have threads, we need either tlbsrx.
123 * or e6500 tablewalk mode, or else TLB handlers
124 * will be racy and could produce duplicate entries.
125 */
126 if (smt_enabled_at_boot >= 2 &&
127 !mmu_has_feature(MMU_FTR_USE_TLBRSRV) &&
128 book3e_htw_mode != PPC_HTW_E6500) {
129 /* Should we panic instead? */
130 WARN_ONCE("%s: unsupported MMU configuration -- expect problems\n",
131 __func__);
132 }
133 }
134}
135#else
136static void setup_tlb_core_data(void)
137{
138}
139#endif
140
40ef8cbc
PM
141#ifdef CONFIG_SMP
142
954e6da5 143static char *smt_enabled_cmdline;
40ef8cbc
PM
144
145/* Look for ibm,smt-enabled OF option */
146static void check_smt_enabled(void)
147{
148 struct device_node *dn;
a7f67bdf 149 const char *smt_option;
40ef8cbc 150
954e6da5
NF
151 /* Default to enabling all threads */
152 smt_enabled_at_boot = threads_per_core;
40ef8cbc 153
954e6da5
NF
154 /* Allow the command line to overrule the OF option */
155 if (smt_enabled_cmdline) {
156 if (!strcmp(smt_enabled_cmdline, "on"))
157 smt_enabled_at_boot = threads_per_core;
158 else if (!strcmp(smt_enabled_cmdline, "off"))
159 smt_enabled_at_boot = 0;
160 else {
1618bd53 161 int smt;
954e6da5
NF
162 int rc;
163
1618bd53 164 rc = kstrtoint(smt_enabled_cmdline, 10, &smt);
954e6da5
NF
165 if (!rc)
166 smt_enabled_at_boot =
1618bd53 167 min(threads_per_core, smt);
954e6da5
NF
168 }
169 } else {
170 dn = of_find_node_by_path("/options");
171 if (dn) {
172 smt_option = of_get_property(dn, "ibm,smt-enabled",
173 NULL);
174
175 if (smt_option) {
176 if (!strcmp(smt_option, "on"))
177 smt_enabled_at_boot = threads_per_core;
178 else if (!strcmp(smt_option, "off"))
179 smt_enabled_at_boot = 0;
180 }
181
182 of_node_put(dn);
183 }
184 }
40ef8cbc
PM
185}
186
187/* Look for smt-enabled= cmdline option */
188static int __init early_smt_enabled(char *p)
189{
954e6da5 190 smt_enabled_cmdline = p;
40ef8cbc
PM
191 return 0;
192}
193early_param("smt-enabled", early_smt_enabled);
194
5ad57078
PM
195#else
196#define check_smt_enabled()
40ef8cbc
PM
197#endif /* CONFIG_SMP */
198
25e13814
ME
199/** Fix up paca fields required for the boot cpu */
200static void fixup_boot_paca(void)
201{
202 /* The boot cpu is started */
203 get_paca()->cpu_start = 1;
204 /* Allow percpu accesses to work until we setup percpu data */
205 get_paca()->data_offset = 0;
206}
207
8f619b54
BH
208static void cpu_ready_for_interrupts(void)
209{
210 /* Set IR and DR in PACA MSR */
211 get_paca()->kernel_msr = MSR_KERNEL;
212
633440f1
ME
213 /*
214 * Enable AIL if supported, and we are in hypervisor mode. If we are
215 * not in hypervisor mode, we enable relocation-on interrupts later
216 * in pSeries_setup_arch() using the H_SET_MODE hcall.
217 */
18aa0da3
PM
218 if (cpu_has_feature(CPU_FTR_HVMODE) &&
219 cpu_has_feature(CPU_FTR_ARCH_207S)) {
8f619b54
BH
220 unsigned long lpcr = mfspr(SPRN_LPCR);
221 mtspr(SPRN_LPCR, lpcr | LPCR_AIL_3);
222 }
223}
224
40ef8cbc
PM
225/*
226 * Early initialization entry point. This is called by head.S
227 * with MMU translation disabled. We rely on the "feature" of
228 * the CPU that ignores the top 2 bits of the address in real
229 * mode so we can access kernel globals normally provided we
230 * only toy with things in the RMO region. From here, we do
95f72d1e 231 * some early parsing of the device-tree to setup out MEMBLOCK
40ef8cbc
PM
232 * data structures, and allocate & initialize the hash table
233 * and segment tables so we can start running with translation
234 * enabled.
235 *
236 * It is this function which will call the probe() callback of
237 * the various platform types and copy the matching one to the
238 * global ppc_md structure. Your platform can eventually do
239 * some very early initializations from the probe() routine, but
240 * this is not recommended, be very careful as, for example, the
241 * device-tree is not accessible via normal means at this point.
242 */
243
244void __init early_setup(unsigned long dt_ptr)
245{
6a7e4064
GL
246 static __initdata struct paca_struct boot_paca;
247
24d96495
BH
248 /* -------- printk is _NOT_ safe to use here ! ------- */
249
42c4aaad 250 /* Identify CPU type */
974a76f5 251 identify_cpu(0, mfspr(SPRN_PVR));
42c4aaad 252
33dbcf72 253 /* Assume we're on cpu 0 for now. Don't write to the paca yet! */
1426d5a3
ME
254 initialise_paca(&boot_paca, 0);
255 setup_paca(&boot_paca);
25e13814 256 fixup_boot_paca();
33dbcf72 257
945feb17
BH
258 /* Initialize lockdep early or else spinlocks will blow */
259 lockdep_init();
260
24d96495
BH
261 /* -------- printk is now safe to use ------- */
262
f2fd2513
BH
263 /* Enable early debugging if any specified (see udbg.h) */
264 udbg_early_init();
265
e8222502 266 DBG(" -> early_setup(), dt_ptr: 0x%lx\n", dt_ptr);
40ef8cbc 267
40ef8cbc 268 /*
3c607ce2
LV
269 * Do early initialization using the flattened device
270 * tree, such as retrieving the physical memory map or
271 * calculating/retrieving the hash table size.
40ef8cbc
PM
272 */
273 early_init_devtree(__va(dt_ptr));
274
4e21b94c
LT
275 epapr_paravirt_early_init();
276
4df20460 277 /* Now we know the logical id of our boot cpu, setup the paca. */
1426d5a3 278 setup_paca(&paca[boot_cpuid]);
25e13814 279 fixup_boot_paca();
4df20460 280
e8222502
BH
281 /* Probe the machine type */
282 probe_machine();
40ef8cbc 283
47310413 284 setup_kdump_trampoline();
0cc4746c 285
40ef8cbc
PM
286 DBG("Found, Initializing memory management...\n");
287
757c74d2
BH
288 /* Initialize the hash table or TLB handling */
289 early_init_mmu();
40ef8cbc 290
a944a9c4
BH
291 /*
292 * At this point, we can let interrupts switch to virtual mode
293 * (the MMU has been setup), so adjust the MSR in the PACA to
8f619b54 294 * have IR and DR set and enable AIL if it exists
a944a9c4 295 */
8f619b54 296 cpu_ready_for_interrupts();
a944a9c4
BH
297
298 /* Reserve large chunks of memory for use by CMA for KVM */
fa61a4e3
AK
299 kvm_cma_reserve();
300
a6146888
BB
301 /*
302 * Reserve any gigantic pages requested on the command line.
303 * memblock needs to have been initialized by the time this is
304 * called since this will reserve memory.
305 */
306 reserve_hugetlb_gpages();
307
40ef8cbc 308 DBG(" <- early_setup()\n");
7191b615
BH
309
310#ifdef CONFIG_PPC_EARLY_DEBUG_BOOTX
311 /*
312 * This needs to be done *last* (after the above DBG() even)
313 *
314 * Right after we return from this function, we turn on the MMU
315 * which means the real-mode access trick that btext does will
316 * no longer work, it needs to switch to using a real MMU
317 * mapping. This call will ensure that it does
318 */
319 btext_map();
320#endif /* CONFIG_PPC_EARLY_DEBUG_BOOTX */
40ef8cbc
PM
321}
322
799d6046
PM
323#ifdef CONFIG_SMP
324void early_setup_secondary(void)
325{
d04c56f7 326 /* Mark interrupts enabled in PACA */
757c74d2 327 get_paca()->soft_enabled = 0;
799d6046 328
757c74d2
BH
329 /* Initialize the hash table or TLB handling */
330 early_init_mmu_secondary();
a944a9c4
BH
331
332 /*
333 * At this point, we can let interrupts switch to virtual mode
334 * (the MMU has been setup), so adjust the MSR in the PACA to
335 * have IR and DR set.
336 */
8f619b54 337 cpu_ready_for_interrupts();
799d6046
PM
338}
339
340#endif /* CONFIG_SMP */
40ef8cbc 341
b8f51021
ME
342#if defined(CONFIG_SMP) || defined(CONFIG_KEXEC)
343void smp_release_cpus(void)
344{
758438a7 345 unsigned long *ptr;
9d07bc84 346 int i;
b8f51021
ME
347
348 DBG(" -> smp_release_cpus()\n");
349
350 /* All secondary cpus are spinning on a common spinloop, release them
351 * all now so they can start to spin on their individual paca
352 * spinloops. For non SMP kernels, the secondary cpus never get out
353 * of the common spinloop.
1f6a93e4 354 */
b8f51021 355
758438a7
ME
356 ptr = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
357 - PHYSICAL_START);
2751b628 358 *ptr = ppc_function_entry(generic_secondary_smp_init);
9d07bc84
BH
359
360 /* And wait a bit for them to catch up */
361 for (i = 0; i < 100000; i++) {
362 mb();
363 HMT_low();
7ac87abb 364 if (spinning_secondaries == 0)
9d07bc84
BH
365 break;
366 udelay(1);
367 }
7ac87abb 368 DBG("spinning_secondaries = %d\n", spinning_secondaries);
b8f51021
ME
369
370 DBG(" <- smp_release_cpus()\n");
371}
372#endif /* CONFIG_SMP || CONFIG_KEXEC */
373
40ef8cbc 374/*
799d6046
PM
375 * Initialize some remaining members of the ppc64_caches and systemcfg
376 * structures
40ef8cbc
PM
377 * (at least until we get rid of them completely). This is mostly some
378 * cache informations about the CPU that will be used by cache flush
379 * routines and/or provided to userland
380 */
381static void __init initialize_cache_info(void)
382{
383 struct device_node *np;
384 unsigned long num_cpus = 0;
385
386 DBG(" -> initialize_cache_info()\n");
387
94db7c5e 388 for_each_node_by_type(np, "cpu") {
40ef8cbc
PM
389 num_cpus += 1;
390
dfbe93a2
AB
391 /*
392 * We're assuming *all* of the CPUs have the same
40ef8cbc
PM
393 * d-cache and i-cache sizes... -Peter
394 */
dfbe93a2 395 if (num_cpus == 1) {
7946d5a5 396 const __be32 *sizep, *lsizep;
40ef8cbc 397 u32 size, lsize;
40ef8cbc
PM
398
399 size = 0;
400 lsize = cur_cpu_spec->dcache_bsize;
e2eb6392 401 sizep = of_get_property(np, "d-cache-size", NULL);
40ef8cbc 402 if (sizep != NULL)
7946d5a5 403 size = be32_to_cpu(*sizep);
dfbe93a2
AB
404 lsizep = of_get_property(np, "d-cache-block-size",
405 NULL);
20474abd
BH
406 /* fallback if block size missing */
407 if (lsizep == NULL)
dfbe93a2
AB
408 lsizep = of_get_property(np,
409 "d-cache-line-size",
410 NULL);
40ef8cbc 411 if (lsizep != NULL)
7946d5a5 412 lsize = be32_to_cpu(*lsizep);
b0d436c7 413 if (sizep == NULL || lsizep == NULL)
40ef8cbc
PM
414 DBG("Argh, can't find dcache properties ! "
415 "sizep: %p, lsizep: %p\n", sizep, lsizep);
416
a7f290da
BH
417 ppc64_caches.dsize = size;
418 ppc64_caches.dline_size = lsize;
40ef8cbc
PM
419 ppc64_caches.log_dline_size = __ilog2(lsize);
420 ppc64_caches.dlines_per_page = PAGE_SIZE / lsize;
421
422 size = 0;
423 lsize = cur_cpu_spec->icache_bsize;
e2eb6392 424 sizep = of_get_property(np, "i-cache-size", NULL);
40ef8cbc 425 if (sizep != NULL)
7946d5a5 426 size = be32_to_cpu(*sizep);
dfbe93a2
AB
427 lsizep = of_get_property(np, "i-cache-block-size",
428 NULL);
20474abd 429 if (lsizep == NULL)
dfbe93a2
AB
430 lsizep = of_get_property(np,
431 "i-cache-line-size",
432 NULL);
40ef8cbc 433 if (lsizep != NULL)
7946d5a5 434 lsize = be32_to_cpu(*lsizep);
b0d436c7 435 if (sizep == NULL || lsizep == NULL)
40ef8cbc
PM
436 DBG("Argh, can't find icache properties ! "
437 "sizep: %p, lsizep: %p\n", sizep, lsizep);
438
a7f290da
BH
439 ppc64_caches.isize = size;
440 ppc64_caches.iline_size = lsize;
40ef8cbc
PM
441 ppc64_caches.log_iline_size = __ilog2(lsize);
442 ppc64_caches.ilines_per_page = PAGE_SIZE / lsize;
443 }
444 }
445
40ef8cbc
PM
446 DBG(" <- initialize_cache_info()\n");
447}
448
40ef8cbc
PM
449
450/*
451 * Do some initial setup of the system. The parameters are those which
452 * were passed in from the bootloader.
453 */
454void __init setup_system(void)
455{
456 DBG(" -> setup_system()\n");
457
826ea8f2
TB
458 /* Apply the CPUs-specific and firmware specific fixups to kernel
459 * text (nop out sections not relevant to this CPU or this firmware)
42c4aaad 460 */
0909c8c2 461 do_feature_fixups(cur_cpu_spec->cpu_features,
42c4aaad 462 &__start___ftr_fixup, &__stop___ftr_fixup);
7c03d653
BH
463 do_feature_fixups(cur_cpu_spec->mmu_features,
464 &__start___mmu_ftr_fixup, &__stop___mmu_ftr_fixup);
826ea8f2
TB
465 do_feature_fixups(powerpc_firmware_features,
466 &__start___fw_ftr_fixup, &__stop___fw_ftr_fixup);
2d1b2027
KG
467 do_lwsync_fixups(cur_cpu_spec->cpu_features,
468 &__start___lwsync_fixup, &__stop___lwsync_fixup);
d715e433 469 do_final_fixups();
42c4aaad 470
40ef8cbc
PM
471 /*
472 * Unflatten the device-tree passed by prom_init or kexec
473 */
474 unflatten_device_tree();
475
476 /*
477 * Fill the ppc64_caches & systemcfg structures with informations
0ebfff14 478 * retrieved from the device-tree.
40ef8cbc
PM
479 */
480 initialize_cache_info();
481
482#ifdef CONFIG_PPC_RTAS
483 /*
484 * Initialize RTAS if available
485 */
486 rtas_initialize();
487#endif /* CONFIG_PPC_RTAS */
40ef8cbc
PM
488
489 /*
490 * Check if we have an initrd provided via the device-tree
491 */
492 check_for_initrd();
40ef8cbc
PM
493
494 /*
495 * Do some platform specific early initializations, that includes
496 * setting up the hash table pointers. It also sets up some interrupt-mapping
497 * related options that will be used by finish_device_tree()
498 */
57744ea9
GL
499 if (ppc_md.init_early)
500 ppc_md.init_early();
40ef8cbc 501
463ce0e1
BH
502 /*
503 * We can discover serial ports now since the above did setup the
504 * hash table management for us, thus ioremap works. We do that early
505 * so that further code can be debugged
506 */
463ce0e1 507 find_legacy_serial_ports();
463ce0e1 508
40ef8cbc
PM
509 /*
510 * Register early console
511 */
512 register_early_udbg_console();
40ef8cbc 513
47679283
ME
514 /*
515 * Initialize xmon
516 */
517 xmon_setup();
480f6f35 518
5ad57078 519 smp_setup_cpu_maps();
954e6da5 520 check_smt_enabled();
28efc35f 521 setup_tlb_core_data();
40ef8cbc 522
e16c8765
AF
523 /*
524 * Freescale Book3e parts spin in a loop provided by firmware,
525 * so smp_release_cpus() does nothing for them
526 */
527#if defined(CONFIG_SMP) && !defined(CONFIG_PPC_FSL_BOOK3E)
40ef8cbc
PM
528 /* Release secondary cpus out of their spinloops at 0x60 now that
529 * we can map physical -> logical CPU ids
530 */
531 smp_release_cpus();
f018b36f 532#endif
40ef8cbc 533
5c0aebf6
ME
534 pr_info("Starting Linux %s %s\n", init_utsname()->machine,
535 init_utsname()->version);
40ef8cbc 536
2c186e05
AB
537 pr_info("-----------------------------------------------------\n");
538 pr_info("ppc64_pft_size = 0x%llx\n", ppc64_pft_size);
539 pr_info("phys_mem_size = 0x%llx\n", memblock_phys_mem_size());
bdce97e9 540
9697add0 541 if (ppc64_caches.dline_size != 0x80)
2c186e05 542 pr_info("dcache_line_size = 0x%x\n", ppc64_caches.dline_size);
9697add0 543 if (ppc64_caches.iline_size != 0x80)
2c186e05 544 pr_info("icache_line_size = 0x%x\n", ppc64_caches.iline_size);
bdce97e9 545
2c186e05
AB
546 pr_info("cpu_features = 0x%016lx\n", cur_cpu_spec->cpu_features);
547 pr_info(" possible = 0x%016lx\n", CPU_FTRS_POSSIBLE);
548 pr_info(" always = 0x%016lx\n", CPU_FTRS_ALWAYS);
549 pr_info("cpu_user_features = 0x%08x 0x%08x\n", cur_cpu_spec->cpu_user_features,
87d99c0e 550 cur_cpu_spec->cpu_user_features2);
2c186e05
AB
551 pr_info("mmu_features = 0x%08x\n", cur_cpu_spec->mmu_features);
552 pr_info("firmware_features = 0x%016lx\n", powerpc_firmware_features);
87d99c0e 553
94491685 554#ifdef CONFIG_PPC_STD_MMU_64
9697add0 555 if (htab_address)
2c186e05 556 pr_info("htab_address = 0x%p\n", htab_address);
bdce97e9 557
2c186e05 558 pr_info("htab_hash_mask = 0x%lx\n", htab_hash_mask);
bdce97e9
ME
559#endif
560
b160544c 561 if (PHYSICAL_START > 0)
2c186e05 562 pr_info("physical_start = 0x%llx\n",
e468455e 563 (unsigned long long)PHYSICAL_START);
2c186e05 564 pr_info("-----------------------------------------------------\n");
40ef8cbc 565
40ef8cbc
PM
566 DBG(" <- setup_system()\n");
567}
568
40bd587a
BH
569/* This returns the limit below which memory accesses to the linear
570 * mapping are guarnateed not to cause a TLB or SLB miss. This is
571 * used to allocate interrupt or emergency stacks for which our
572 * exception entry path doesn't deal with being interrupted.
573 */
574static u64 safe_stack_limit(void)
095c7965 575{
40bd587a
BH
576#ifdef CONFIG_PPC_BOOK3E
577 /* Freescale BookE bolts the entire linear mapping */
578 if (mmu_has_feature(MMU_FTR_TYPE_FSL_E))
579 return linear_map_top;
580 /* Other BookE, we assume the first GB is bolted */
581 return 1ul << 30;
582#else
583 /* BookS, the first segment is bolted */
584 if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
095c7965 585 return 1UL << SID_SHIFT_1T;
095c7965 586 return 1UL << SID_SHIFT;
40bd587a 587#endif
095c7965
AB
588}
589
40ef8cbc
PM
590static void __init irqstack_early_init(void)
591{
40bd587a 592 u64 limit = safe_stack_limit();
40ef8cbc
PM
593 unsigned int i;
594
595 /*
8f4da26e
AB
596 * Interrupt stacks must be in the first segment since we
597 * cannot afford to take SLB misses on them.
40ef8cbc 598 */
0e551954 599 for_each_possible_cpu(i) {
3c726f8d 600 softirq_ctx[i] = (struct thread_info *)
95f72d1e 601 __va(memblock_alloc_base(THREAD_SIZE,
095c7965 602 THREAD_SIZE, limit));
3c726f8d 603 hardirq_ctx[i] = (struct thread_info *)
95f72d1e 604 __va(memblock_alloc_base(THREAD_SIZE,
095c7965 605 THREAD_SIZE, limit));
40ef8cbc
PM
606 }
607}
40ef8cbc 608
2d27cfd3
BH
609#ifdef CONFIG_PPC_BOOK3E
610static void __init exc_lvl_early_init(void)
611{
612 unsigned int i;
160c7324 613 unsigned long sp;
2d27cfd3
BH
614
615 for_each_possible_cpu(i) {
160c7324
TC
616 sp = memblock_alloc(THREAD_SIZE, THREAD_SIZE);
617 critirq_ctx[i] = (struct thread_info *)__va(sp);
618 paca[i].crit_kstack = __va(sp + THREAD_SIZE);
619
620 sp = memblock_alloc(THREAD_SIZE, THREAD_SIZE);
621 dbgirq_ctx[i] = (struct thread_info *)__va(sp);
622 paca[i].dbg_kstack = __va(sp + THREAD_SIZE);
623
624 sp = memblock_alloc(THREAD_SIZE, THREAD_SIZE);
625 mcheckirq_ctx[i] = (struct thread_info *)__va(sp);
626 paca[i].mc_kstack = __va(sp + THREAD_SIZE);
2d27cfd3 627 }
d36b4c4f
KG
628
629 if (cpu_has_feature(CPU_FTR_DEBUG_LVL_EXC))
565c2f24 630 patch_exception(0x040, exc_debug_debug_book3e);
2d27cfd3
BH
631}
632#else
633#define exc_lvl_early_init()
634#endif
635
40ef8cbc
PM
636/*
637 * Stack space used when we detect a bad kernel stack pointer, and
729b0f71
MS
638 * early in SMP boots before relocation is enabled. Exclusive emergency
639 * stack for machine checks.
40ef8cbc
PM
640 */
641static void __init emergency_stack_init(void)
642{
095c7965 643 u64 limit;
40ef8cbc
PM
644 unsigned int i;
645
646 /*
647 * Emergency stacks must be under 256MB, we cannot afford to take
648 * SLB misses on them. The ABI also requires them to be 128-byte
649 * aligned.
650 *
651 * Since we use these as temporary stacks during secondary CPU
652 * bringup, we need to get at them in real mode. This means they
653 * must also be within the RMO region.
654 */
40bd587a 655 limit = min(safe_stack_limit(), ppc64_rma_size);
40ef8cbc 656
3243d874
ME
657 for_each_possible_cpu(i) {
658 unsigned long sp;
95f72d1e 659 sp = memblock_alloc_base(THREAD_SIZE, THREAD_SIZE, limit);
3243d874
ME
660 sp += THREAD_SIZE;
661 paca[i].emergency_sp = __va(sp);
729b0f71
MS
662
663#ifdef CONFIG_PPC_BOOK3S_64
664 /* emergency stack for machine check exception handling. */
665 sp = memblock_alloc_base(THREAD_SIZE, THREAD_SIZE, limit);
666 sp += THREAD_SIZE;
667 paca[i].mc_emergency_sp = __va(sp);
668#endif
3243d874 669 }
40ef8cbc
PM
670}
671
40ef8cbc 672/*
e39f223f 673 * Called into from start_kernel this initializes memblock, which is used
0f6b77ca 674 * to manage page allocation until mem_init is called.
40ef8cbc
PM
675 */
676void __init setup_arch(char **cmdline_p)
677{
3e47d147 678 *cmdline_p = boot_command_line;
40ef8cbc
PM
679
680 /*
681 * Set cache line size based on type of cpu as a default.
682 * Systems with OF can look in the properties on the cpu node(s)
683 * for a possibly more accurate value.
684 */
685 dcache_bsize = ppc64_caches.dline_size;
686 icache_bsize = ppc64_caches.iline_size;
687
40ef8cbc 688 if (ppc_md.panic)
7e990266 689 setup_panic();
40ef8cbc 690
4846c5de 691 init_mm.start_code = (unsigned long)_stext;
40ef8cbc
PM
692 init_mm.end_code = (unsigned long) _etext;
693 init_mm.end_data = (unsigned long) _edata;
694 init_mm.brk = klimit;
5c1f6ee9
AK
695#ifdef CONFIG_PPC_64K_PAGES
696 init_mm.context.pte_frag = NULL;
15b244a8
AK
697#endif
698#ifdef CONFIG_SPAPR_TCE_IOMMU
699 mm_iommu_init(&init_mm.context);
5c1f6ee9 700#endif
40ef8cbc 701 irqstack_early_init();
2d27cfd3 702 exc_lvl_early_init();
40ef8cbc
PM
703 emergency_stack_init();
704
10239733 705 initmem_init();
40ef8cbc 706
0458060c
PM
707#ifdef CONFIG_DUMMY_CONSOLE
708 conswitchp = &dummy_con;
709#endif
710
38db7e74
GL
711 if (ppc_md.setup_arch)
712 ppc_md.setup_arch();
40ef8cbc 713
40ef8cbc 714 paging_init();
6f0ef0f5
BH
715
716 /* Initialize the MMU context management stuff */
717 mmu_context_init();
718
61e2390e
MN
719 /* Interrupt code needs to be 64K-aligned */
720 if ((unsigned long)_stext & 0xffff)
721 panic("Kernelbase not 64K-aligned (0x%lx)!\n",
722 (unsigned long)_stext);
40ef8cbc
PM
723}
724
7a0268fa 725#ifdef CONFIG_SMP
c2a7e818
TH
726#define PCPU_DYN_SIZE ()
727
728static void * __init pcpu_fc_alloc(unsigned int cpu, size_t size, size_t align)
7a0268fa 729{
c2a7e818
TH
730 return __alloc_bootmem_node(NODE_DATA(cpu_to_node(cpu)), size, align,
731 __pa(MAX_DMA_ADDRESS));
732}
7a0268fa 733
c2a7e818
TH
734static void __init pcpu_fc_free(void *ptr, size_t size)
735{
736 free_bootmem(__pa(ptr), size);
737}
7a0268fa 738
c2a7e818
TH
739static int pcpu_cpu_distance(unsigned int from, unsigned int to)
740{
741 if (cpu_to_node(from) == cpu_to_node(to))
742 return LOCAL_DISTANCE;
743 else
744 return REMOTE_DISTANCE;
745}
746
ae01f84b
AB
747unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
748EXPORT_SYMBOL(__per_cpu_offset);
749
c2a7e818
TH
750void __init setup_per_cpu_areas(void)
751{
752 const size_t dyn_size = PERCPU_MODULE_RESERVE + PERCPU_DYNAMIC_RESERVE;
753 size_t atom_size;
754 unsigned long delta;
755 unsigned int cpu;
756 int rc;
757
758 /*
759 * Linear mapping is one of 4K, 1M and 16M. For 4K, no need
760 * to group units. For larger mappings, use 1M atom which
761 * should be large enough to contain a number of units.
762 */
763 if (mmu_linear_psize == MMU_PAGE_4K)
764 atom_size = PAGE_SIZE;
765 else
766 atom_size = 1 << 20;
767
768 rc = pcpu_embed_first_chunk(0, dyn_size, atom_size, pcpu_cpu_distance,
769 pcpu_fc_alloc, pcpu_fc_free);
770 if (rc < 0)
771 panic("cannot initialize percpu area (err=%d)", rc);
772
773 delta = (unsigned long)pcpu_base_addr - (unsigned long)__per_cpu_start;
ae01f84b
AB
774 for_each_possible_cpu(cpu) {
775 __per_cpu_offset[cpu] = delta + pcpu_unit_offsets[cpu];
776 paca[cpu].data_offset = __per_cpu_offset[cpu];
777 }
7a0268fa
AB
778}
779#endif
4cb3cee0 780
a5d86257
AB
781#ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
782unsigned long memory_block_size_bytes(void)
783{
784 if (ppc_md.memory_block_size)
785 return ppc_md.memory_block_size();
786
787 return MIN_MEMORY_BLOCK_SIZE;
788}
789#endif
4cb3cee0 790
ecd73cc5 791#if defined(CONFIG_PPC_INDIRECT_PIO) || defined(CONFIG_PPC_INDIRECT_MMIO)
4cb3cee0
BH
792struct ppc_pci_io ppc_pci_io;
793EXPORT_SYMBOL(ppc_pci_io);
ecd73cc5 794#endif
c54b2bf1
AB
795
796#ifdef CONFIG_HARDLOCKUP_DETECTOR
797u64 hw_nmi_get_sample_period(int watchdog_thresh)
798{
799 return ppc_proc_freq * watchdog_thresh;
800}
801
802/*
803 * The hardlockup detector breaks PMU event based branches and is likely
804 * to get false positives in KVM guests, so disable it by default.
805 */
806static int __init disable_hardlockup_detector(void)
807{
d19d5efd 808 hardlockup_detector_disable();
c54b2bf1
AB
809
810 return 0;
811}
812early_initcall(disable_hardlockup_detector);
813#endif