powerpc: Move FW feature probing out of pseries probe()
[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>
5d31a96e 71#include <asm/livepatch.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
24d96495
BH
258 /* -------- printk is now safe to use ------- */
259
f2fd2513
BH
260 /* Enable early debugging if any specified (see udbg.h) */
261 udbg_early_init();
262
e8222502 263 DBG(" -> early_setup(), dt_ptr: 0x%lx\n", dt_ptr);
40ef8cbc 264
40ef8cbc 265 /*
3c607ce2
LV
266 * Do early initialization using the flattened device
267 * tree, such as retrieving the physical memory map or
268 * calculating/retrieving the hash table size.
40ef8cbc
PM
269 */
270 early_init_devtree(__va(dt_ptr));
271
4df20460 272 /* Now we know the logical id of our boot cpu, setup the paca. */
1426d5a3 273 setup_paca(&paca[boot_cpuid]);
25e13814 274 fixup_boot_paca();
4df20460 275
e8222502
BH
276 /* Probe the machine type */
277 probe_machine();
40ef8cbc 278
63c254a5
BH
279 /*
280 * Setup the trampolines from the lowmem exception vectors
281 * to the kdump kernel when not using a relocatable kernel.
282 */
47310413 283 setup_kdump_trampoline();
0cc4746c 284
757c74d2
BH
285 /* Initialize the hash table or TLB handling */
286 early_init_mmu();
40ef8cbc 287
c4bd6cb8
BH
288 /* Apply all the dynamic patching */
289 apply_feature_fixups();
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 297
40ef8cbc 298 DBG(" <- early_setup()\n");
7191b615
BH
299
300#ifdef CONFIG_PPC_EARLY_DEBUG_BOOTX
301 /*
302 * This needs to be done *last* (after the above DBG() even)
303 *
304 * Right after we return from this function, we turn on the MMU
305 * which means the real-mode access trick that btext does will
306 * no longer work, it needs to switch to using a real MMU
307 * mapping. This call will ensure that it does
308 */
309 btext_map();
310#endif /* CONFIG_PPC_EARLY_DEBUG_BOOTX */
40ef8cbc
PM
311}
312
799d6046
PM
313#ifdef CONFIG_SMP
314void early_setup_secondary(void)
315{
103b7827 316 /* Mark interrupts disabled in PACA */
757c74d2 317 get_paca()->soft_enabled = 0;
799d6046 318
757c74d2
BH
319 /* Initialize the hash table or TLB handling */
320 early_init_mmu_secondary();
a944a9c4
BH
321
322 /*
323 * At this point, we can let interrupts switch to virtual mode
324 * (the MMU has been setup), so adjust the MSR in the PACA to
325 * have IR and DR set.
326 */
8f619b54 327 cpu_ready_for_interrupts();
799d6046
PM
328}
329
330#endif /* CONFIG_SMP */
40ef8cbc 331
b8f51021 332#if defined(CONFIG_SMP) || defined(CONFIG_KEXEC)
567cf94d
SW
333static bool use_spinloop(void)
334{
335 if (!IS_ENABLED(CONFIG_PPC_BOOK3E))
336 return true;
337
338 /*
339 * When book3e boots from kexec, the ePAPR spin table does
340 * not get used.
341 */
342 return of_property_read_bool(of_chosen, "linux,booted-from-kexec");
343}
344
b8f51021
ME
345void smp_release_cpus(void)
346{
758438a7 347 unsigned long *ptr;
9d07bc84 348 int i;
b8f51021 349
567cf94d
SW
350 if (!use_spinloop())
351 return;
352
b8f51021
ME
353 DBG(" -> smp_release_cpus()\n");
354
355 /* All secondary cpus are spinning on a common spinloop, release them
356 * all now so they can start to spin on their individual paca
357 * spinloops. For non SMP kernels, the secondary cpus never get out
358 * of the common spinloop.
1f6a93e4 359 */
b8f51021 360
758438a7
ME
361 ptr = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
362 - PHYSICAL_START);
2751b628 363 *ptr = ppc_function_entry(generic_secondary_smp_init);
9d07bc84
BH
364
365 /* And wait a bit for them to catch up */
366 for (i = 0; i < 100000; i++) {
367 mb();
368 HMT_low();
7ac87abb 369 if (spinning_secondaries == 0)
9d07bc84
BH
370 break;
371 udelay(1);
372 }
7ac87abb 373 DBG("spinning_secondaries = %d\n", spinning_secondaries);
b8f51021
ME
374
375 DBG(" <- smp_release_cpus()\n");
376}
377#endif /* CONFIG_SMP || CONFIG_KEXEC */
378
40ef8cbc 379/*
799d6046
PM
380 * Initialize some remaining members of the ppc64_caches and systemcfg
381 * structures
40ef8cbc
PM
382 * (at least until we get rid of them completely). This is mostly some
383 * cache informations about the CPU that will be used by cache flush
384 * routines and/or provided to userland
385 */
386static void __init initialize_cache_info(void)
387{
388 struct device_node *np;
389 unsigned long num_cpus = 0;
390
391 DBG(" -> initialize_cache_info()\n");
392
94db7c5e 393 for_each_node_by_type(np, "cpu") {
40ef8cbc
PM
394 num_cpus += 1;
395
dfbe93a2
AB
396 /*
397 * We're assuming *all* of the CPUs have the same
40ef8cbc
PM
398 * d-cache and i-cache sizes... -Peter
399 */
dfbe93a2 400 if (num_cpus == 1) {
7946d5a5 401 const __be32 *sizep, *lsizep;
40ef8cbc 402 u32 size, lsize;
40ef8cbc
PM
403
404 size = 0;
405 lsize = cur_cpu_spec->dcache_bsize;
e2eb6392 406 sizep = of_get_property(np, "d-cache-size", NULL);
40ef8cbc 407 if (sizep != NULL)
7946d5a5 408 size = be32_to_cpu(*sizep);
dfbe93a2
AB
409 lsizep = of_get_property(np, "d-cache-block-size",
410 NULL);
20474abd
BH
411 /* fallback if block size missing */
412 if (lsizep == NULL)
dfbe93a2
AB
413 lsizep = of_get_property(np,
414 "d-cache-line-size",
415 NULL);
40ef8cbc 416 if (lsizep != NULL)
7946d5a5 417 lsize = be32_to_cpu(*lsizep);
b0d436c7 418 if (sizep == NULL || lsizep == NULL)
40ef8cbc
PM
419 DBG("Argh, can't find dcache properties ! "
420 "sizep: %p, lsizep: %p\n", sizep, lsizep);
421
a7f290da
BH
422 ppc64_caches.dsize = size;
423 ppc64_caches.dline_size = lsize;
40ef8cbc
PM
424 ppc64_caches.log_dline_size = __ilog2(lsize);
425 ppc64_caches.dlines_per_page = PAGE_SIZE / lsize;
426
427 size = 0;
428 lsize = cur_cpu_spec->icache_bsize;
e2eb6392 429 sizep = of_get_property(np, "i-cache-size", NULL);
40ef8cbc 430 if (sizep != NULL)
7946d5a5 431 size = be32_to_cpu(*sizep);
dfbe93a2
AB
432 lsizep = of_get_property(np, "i-cache-block-size",
433 NULL);
20474abd 434 if (lsizep == NULL)
dfbe93a2
AB
435 lsizep = of_get_property(np,
436 "i-cache-line-size",
437 NULL);
40ef8cbc 438 if (lsizep != NULL)
7946d5a5 439 lsize = be32_to_cpu(*lsizep);
b0d436c7 440 if (sizep == NULL || lsizep == NULL)
40ef8cbc
PM
441 DBG("Argh, can't find icache properties ! "
442 "sizep: %p, lsizep: %p\n", sizep, lsizep);
443
a7f290da
BH
444 ppc64_caches.isize = size;
445 ppc64_caches.iline_size = lsize;
40ef8cbc
PM
446 ppc64_caches.log_iline_size = __ilog2(lsize);
447 ppc64_caches.ilines_per_page = PAGE_SIZE / lsize;
448 }
449 }
450
40ef8cbc
PM
451 DBG(" <- initialize_cache_info()\n");
452}
453
40ef8cbc
PM
454
455/*
456 * Do some initial setup of the system. The parameters are those which
457 * were passed in from the bootloader.
458 */
459void __init setup_system(void)
460{
461 DBG(" -> setup_system()\n");
462
463 /*
464 * Unflatten the device-tree passed by prom_init or kexec
465 */
466 unflatten_device_tree();
467
468 /*
469 * Fill the ppc64_caches & systemcfg structures with informations
0ebfff14 470 * retrieved from the device-tree.
40ef8cbc
PM
471 */
472 initialize_cache_info();
473
474#ifdef CONFIG_PPC_RTAS
475 /*
476 * Initialize RTAS if available
477 */
478 rtas_initialize();
479#endif /* CONFIG_PPC_RTAS */
40ef8cbc
PM
480
481 /*
482 * Check if we have an initrd provided via the device-tree
483 */
484 check_for_initrd();
40ef8cbc
PM
485
486 /*
487 * Do some platform specific early initializations, that includes
488 * setting up the hash table pointers. It also sets up some interrupt-mapping
489 * related options that will be used by finish_device_tree()
490 */
57744ea9
GL
491 if (ppc_md.init_early)
492 ppc_md.init_early();
40ef8cbc 493
463ce0e1
BH
494 /*
495 * We can discover serial ports now since the above did setup the
496 * hash table management for us, thus ioremap works. We do that early
497 * so that further code can be debugged
498 */
463ce0e1 499 find_legacy_serial_ports();
463ce0e1 500
40ef8cbc
PM
501 /*
502 * Register early console
503 */
504 register_early_udbg_console();
40ef8cbc 505
47679283
ME
506 /*
507 * Initialize xmon
508 */
509 xmon_setup();
480f6f35 510
5ad57078 511 smp_setup_cpu_maps();
954e6da5 512 check_smt_enabled();
28efc35f 513 setup_tlb_core_data();
40ef8cbc 514
e16c8765
AF
515 /*
516 * Freescale Book3e parts spin in a loop provided by firmware,
517 * so smp_release_cpus() does nothing for them
518 */
567cf94d 519#if defined(CONFIG_SMP)
40ef8cbc
PM
520 /* Release secondary cpus out of their spinloops at 0x60 now that
521 * we can map physical -> logical CPU ids
522 */
523 smp_release_cpus();
f018b36f 524#endif
40ef8cbc 525
5c0aebf6
ME
526 pr_info("Starting Linux %s %s\n", init_utsname()->machine,
527 init_utsname()->version);
40ef8cbc 528
2c186e05
AB
529 pr_info("-----------------------------------------------------\n");
530 pr_info("ppc64_pft_size = 0x%llx\n", ppc64_pft_size);
531 pr_info("phys_mem_size = 0x%llx\n", memblock_phys_mem_size());
bdce97e9 532
9697add0 533 if (ppc64_caches.dline_size != 0x80)
2c186e05 534 pr_info("dcache_line_size = 0x%x\n", ppc64_caches.dline_size);
9697add0 535 if (ppc64_caches.iline_size != 0x80)
2c186e05 536 pr_info("icache_line_size = 0x%x\n", ppc64_caches.iline_size);
bdce97e9 537
2c186e05
AB
538 pr_info("cpu_features = 0x%016lx\n", cur_cpu_spec->cpu_features);
539 pr_info(" possible = 0x%016lx\n", CPU_FTRS_POSSIBLE);
540 pr_info(" always = 0x%016lx\n", CPU_FTRS_ALWAYS);
541 pr_info("cpu_user_features = 0x%08x 0x%08x\n", cur_cpu_spec->cpu_user_features,
87d99c0e 542 cur_cpu_spec->cpu_user_features2);
2c186e05
AB
543 pr_info("mmu_features = 0x%08x\n", cur_cpu_spec->mmu_features);
544 pr_info("firmware_features = 0x%016lx\n", powerpc_firmware_features);
87d99c0e 545
94491685 546#ifdef CONFIG_PPC_STD_MMU_64
9697add0 547 if (htab_address)
2c186e05 548 pr_info("htab_address = 0x%p\n", htab_address);
bdce97e9 549
2c186e05 550 pr_info("htab_hash_mask = 0x%lx\n", htab_hash_mask);
bdce97e9
ME
551#endif
552
b160544c 553 if (PHYSICAL_START > 0)
2c186e05 554 pr_info("physical_start = 0x%llx\n",
e468455e 555 (unsigned long long)PHYSICAL_START);
2c186e05 556 pr_info("-----------------------------------------------------\n");
40ef8cbc 557
40ef8cbc
PM
558 DBG(" <- setup_system()\n");
559}
560
40bd587a
BH
561/* This returns the limit below which memory accesses to the linear
562 * mapping are guarnateed not to cause a TLB or SLB miss. This is
563 * used to allocate interrupt or emergency stacks for which our
564 * exception entry path doesn't deal with being interrupted.
565 */
566static u64 safe_stack_limit(void)
095c7965 567{
40bd587a
BH
568#ifdef CONFIG_PPC_BOOK3E
569 /* Freescale BookE bolts the entire linear mapping */
570 if (mmu_has_feature(MMU_FTR_TYPE_FSL_E))
571 return linear_map_top;
572 /* Other BookE, we assume the first GB is bolted */
573 return 1ul << 30;
574#else
575 /* BookS, the first segment is bolted */
576 if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
095c7965 577 return 1UL << SID_SHIFT_1T;
095c7965 578 return 1UL << SID_SHIFT;
40bd587a 579#endif
095c7965
AB
580}
581
40ef8cbc
PM
582static void __init irqstack_early_init(void)
583{
40bd587a 584 u64 limit = safe_stack_limit();
40ef8cbc
PM
585 unsigned int i;
586
587 /*
8f4da26e
AB
588 * Interrupt stacks must be in the first segment since we
589 * cannot afford to take SLB misses on them.
40ef8cbc 590 */
0e551954 591 for_each_possible_cpu(i) {
3c726f8d 592 softirq_ctx[i] = (struct thread_info *)
95f72d1e 593 __va(memblock_alloc_base(THREAD_SIZE,
095c7965 594 THREAD_SIZE, limit));
3c726f8d 595 hardirq_ctx[i] = (struct thread_info *)
95f72d1e 596 __va(memblock_alloc_base(THREAD_SIZE,
095c7965 597 THREAD_SIZE, limit));
40ef8cbc
PM
598 }
599}
40ef8cbc 600
2d27cfd3
BH
601#ifdef CONFIG_PPC_BOOK3E
602static void __init exc_lvl_early_init(void)
603{
604 unsigned int i;
160c7324 605 unsigned long sp;
2d27cfd3
BH
606
607 for_each_possible_cpu(i) {
160c7324
TC
608 sp = memblock_alloc(THREAD_SIZE, THREAD_SIZE);
609 critirq_ctx[i] = (struct thread_info *)__va(sp);
610 paca[i].crit_kstack = __va(sp + THREAD_SIZE);
611
612 sp = memblock_alloc(THREAD_SIZE, THREAD_SIZE);
613 dbgirq_ctx[i] = (struct thread_info *)__va(sp);
614 paca[i].dbg_kstack = __va(sp + THREAD_SIZE);
615
616 sp = memblock_alloc(THREAD_SIZE, THREAD_SIZE);
617 mcheckirq_ctx[i] = (struct thread_info *)__va(sp);
618 paca[i].mc_kstack = __va(sp + THREAD_SIZE);
2d27cfd3 619 }
d36b4c4f
KG
620
621 if (cpu_has_feature(CPU_FTR_DEBUG_LVL_EXC))
565c2f24 622 patch_exception(0x040, exc_debug_debug_book3e);
2d27cfd3
BH
623}
624#else
625#define exc_lvl_early_init()
626#endif
627
40ef8cbc
PM
628/*
629 * Stack space used when we detect a bad kernel stack pointer, and
729b0f71
MS
630 * early in SMP boots before relocation is enabled. Exclusive emergency
631 * stack for machine checks.
40ef8cbc
PM
632 */
633static void __init emergency_stack_init(void)
634{
095c7965 635 u64 limit;
40ef8cbc
PM
636 unsigned int i;
637
638 /*
639 * Emergency stacks must be under 256MB, we cannot afford to take
640 * SLB misses on them. The ABI also requires them to be 128-byte
641 * aligned.
642 *
643 * Since we use these as temporary stacks during secondary CPU
644 * bringup, we need to get at them in real mode. This means they
645 * must also be within the RMO region.
646 */
40bd587a 647 limit = min(safe_stack_limit(), ppc64_rma_size);
40ef8cbc 648
3243d874 649 for_each_possible_cpu(i) {
5d31a96e
ME
650 struct thread_info *ti;
651 ti = __va(memblock_alloc_base(THREAD_SIZE, THREAD_SIZE, limit));
652 klp_init_thread_info(ti);
653 paca[i].emergency_sp = (void *)ti + THREAD_SIZE;
729b0f71
MS
654
655#ifdef CONFIG_PPC_BOOK3S_64
656 /* emergency stack for machine check exception handling. */
5d31a96e
ME
657 ti = __va(memblock_alloc_base(THREAD_SIZE, THREAD_SIZE, limit));
658 klp_init_thread_info(ti);
659 paca[i].mc_emergency_sp = (void *)ti + THREAD_SIZE;
729b0f71 660#endif
3243d874 661 }
40ef8cbc
PM
662}
663
40ef8cbc 664/*
e39f223f 665 * Called into from start_kernel this initializes memblock, which is used
0f6b77ca 666 * to manage page allocation until mem_init is called.
40ef8cbc
PM
667 */
668void __init setup_arch(char **cmdline_p)
669{
3e47d147 670 *cmdline_p = boot_command_line;
40ef8cbc
PM
671
672 /*
673 * Set cache line size based on type of cpu as a default.
674 * Systems with OF can look in the properties on the cpu node(s)
675 * for a possibly more accurate value.
676 */
677 dcache_bsize = ppc64_caches.dline_size;
678 icache_bsize = ppc64_caches.iline_size;
679
de4cf3de
BH
680
681 /* Reserve large chunks of memory for use by CMA for KVM */
682 kvm_cma_reserve();
683
684 /*
685 * Reserve any gigantic pages requested on the command line.
686 * memblock needs to have been initialized by the time this is
687 * called since this will reserve memory.
688 */
689 reserve_hugetlb_gpages();
690
40ef8cbc 691 if (ppc_md.panic)
7e990266 692 setup_panic();
40ef8cbc 693
5d31a96e
ME
694 klp_init_thread_info(&init_thread_info);
695
4846c5de 696 init_mm.start_code = (unsigned long)_stext;
40ef8cbc
PM
697 init_mm.end_code = (unsigned long) _etext;
698 init_mm.end_data = (unsigned long) _edata;
699 init_mm.brk = klimit;
5c1f6ee9
AK
700#ifdef CONFIG_PPC_64K_PAGES
701 init_mm.context.pte_frag = NULL;
15b244a8
AK
702#endif
703#ifdef CONFIG_SPAPR_TCE_IOMMU
704 mm_iommu_init(&init_mm.context);
5c1f6ee9 705#endif
40ef8cbc 706 irqstack_early_init();
2d27cfd3 707 exc_lvl_early_init();
40ef8cbc
PM
708 emergency_stack_init();
709
10239733 710 initmem_init();
40ef8cbc 711
0458060c
PM
712#ifdef CONFIG_DUMMY_CONSOLE
713 conswitchp = &dummy_con;
714#endif
38db7e74
GL
715 if (ppc_md.setup_arch)
716 ppc_md.setup_arch();
40ef8cbc 717
40ef8cbc 718 paging_init();
6f0ef0f5
BH
719
720 /* Initialize the MMU context management stuff */
721 mmu_context_init();
722
61e2390e
MN
723 /* Interrupt code needs to be 64K-aligned */
724 if ((unsigned long)_stext & 0xffff)
725 panic("Kernelbase not 64K-aligned (0x%lx)!\n",
726 (unsigned long)_stext);
40ef8cbc
PM
727}
728
7a0268fa 729#ifdef CONFIG_SMP
c2a7e818
TH
730#define PCPU_DYN_SIZE ()
731
732static void * __init pcpu_fc_alloc(unsigned int cpu, size_t size, size_t align)
7a0268fa 733{
c2a7e818
TH
734 return __alloc_bootmem_node(NODE_DATA(cpu_to_node(cpu)), size, align,
735 __pa(MAX_DMA_ADDRESS));
736}
7a0268fa 737
c2a7e818
TH
738static void __init pcpu_fc_free(void *ptr, size_t size)
739{
740 free_bootmem(__pa(ptr), size);
741}
7a0268fa 742
c2a7e818
TH
743static int pcpu_cpu_distance(unsigned int from, unsigned int to)
744{
745 if (cpu_to_node(from) == cpu_to_node(to))
746 return LOCAL_DISTANCE;
747 else
748 return REMOTE_DISTANCE;
749}
750
ae01f84b
AB
751unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
752EXPORT_SYMBOL(__per_cpu_offset);
753
c2a7e818
TH
754void __init setup_per_cpu_areas(void)
755{
756 const size_t dyn_size = PERCPU_MODULE_RESERVE + PERCPU_DYNAMIC_RESERVE;
757 size_t atom_size;
758 unsigned long delta;
759 unsigned int cpu;
760 int rc;
761
762 /*
763 * Linear mapping is one of 4K, 1M and 16M. For 4K, no need
764 * to group units. For larger mappings, use 1M atom which
765 * should be large enough to contain a number of units.
766 */
767 if (mmu_linear_psize == MMU_PAGE_4K)
768 atom_size = PAGE_SIZE;
769 else
770 atom_size = 1 << 20;
771
772 rc = pcpu_embed_first_chunk(0, dyn_size, atom_size, pcpu_cpu_distance,
773 pcpu_fc_alloc, pcpu_fc_free);
774 if (rc < 0)
775 panic("cannot initialize percpu area (err=%d)", rc);
776
777 delta = (unsigned long)pcpu_base_addr - (unsigned long)__per_cpu_start;
ae01f84b
AB
778 for_each_possible_cpu(cpu) {
779 __per_cpu_offset[cpu] = delta + pcpu_unit_offsets[cpu];
780 paca[cpu].data_offset = __per_cpu_offset[cpu];
781 }
7a0268fa
AB
782}
783#endif
4cb3cee0 784
a5d86257
AB
785#ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
786unsigned long memory_block_size_bytes(void)
787{
788 if (ppc_md.memory_block_size)
789 return ppc_md.memory_block_size();
790
791 return MIN_MEMORY_BLOCK_SIZE;
792}
793#endif
4cb3cee0 794
ecd73cc5 795#if defined(CONFIG_PPC_INDIRECT_PIO) || defined(CONFIG_PPC_INDIRECT_MMIO)
4cb3cee0
BH
796struct ppc_pci_io ppc_pci_io;
797EXPORT_SYMBOL(ppc_pci_io);
ecd73cc5 798#endif
c54b2bf1
AB
799
800#ifdef CONFIG_HARDLOCKUP_DETECTOR
801u64 hw_nmi_get_sample_period(int watchdog_thresh)
802{
803 return ppc_proc_freq * watchdog_thresh;
804}
805
806/*
807 * The hardlockup detector breaks PMU event based branches and is likely
808 * to get false positives in KVM guests, so disable it by default.
809 */
810static int __init disable_hardlockup_detector(void)
811{
d19d5efd 812 hardlockup_detector_disable();
c54b2bf1
AB
813
814 return 0;
815}
816early_initcall(disable_hardlockup_detector);
817#endif