3 * Common boot and setup code.
5 * Copyright (C) 2001 PPC64 Team, IBM Corp
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.
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 #include <linux/memory.h>
40 #include <linux/nmi.h>
43 #include <asm/kdump.h>
45 #include <asm/processor.h>
46 #include <asm/pgtable.h>
49 #include <asm/machdep.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>
58 #include <asm/iommu.h>
59 #include <asm/serial.h>
60 #include <asm/cache.h>
63 #include <asm/firmware.h>
66 #include <asm/kexec.h>
67 #include <asm/mmu_context.h>
68 #include <asm/code-patching.h>
69 #include <asm/kvm_ppc.h>
70 #include <asm/hugetlb.h>
71 #include <asm/epapr_hcalls.h>
74 #define DBG(fmt...) udbg_printf(fmt)
79 int spinning_secondaries
;
82 /* Pick defaults since we might want to patch instructions
83 * before we've read this from the device tree.
85 struct ppc64_caches ppc64_caches
= {
91 EXPORT_SYMBOL_GPL(ppc64_caches
);
94 * These are used in binfmt_elf.c to put aux entries on the stack
95 * for each elf executable being started.
101 #if defined(CONFIG_PPC_BOOK3E) && defined(CONFIG_SMP)
102 static void setup_tlb_core_data(void)
106 BUILD_BUG_ON(offsetof(struct tlb_core_data
, lock
) != 0);
108 for_each_possible_cpu(cpu
) {
109 int first
= cpu_first_thread_sibling(cpu
);
111 paca
[cpu
].tcd_ptr
= &paca
[first
].tcd
;
114 * If we have threads, we need either tlbsrx.
115 * or e6500 tablewalk mode, or else TLB handlers
116 * will be racy and could produce duplicate entries.
118 if (smt_enabled_at_boot
>= 2 &&
119 !mmu_has_feature(MMU_FTR_USE_TLBRSRV
) &&
120 book3e_htw_mode
!= PPC_HTW_E6500
) {
121 /* Should we panic instead? */
122 WARN_ONCE("%s: unsupported MMU configuration -- expect problems\n",
128 static void setup_tlb_core_data(void)
135 static char *smt_enabled_cmdline
;
137 /* Look for ibm,smt-enabled OF option */
138 static void check_smt_enabled(void)
140 struct device_node
*dn
;
141 const char *smt_option
;
143 /* Default to enabling all threads */
144 smt_enabled_at_boot
= threads_per_core
;
146 /* Allow the command line to overrule the OF option */
147 if (smt_enabled_cmdline
) {
148 if (!strcmp(smt_enabled_cmdline
, "on"))
149 smt_enabled_at_boot
= threads_per_core
;
150 else if (!strcmp(smt_enabled_cmdline
, "off"))
151 smt_enabled_at_boot
= 0;
156 rc
= kstrtoint(smt_enabled_cmdline
, 10, &smt
);
158 smt_enabled_at_boot
=
159 min(threads_per_core
, smt
);
162 dn
= of_find_node_by_path("/options");
164 smt_option
= of_get_property(dn
, "ibm,smt-enabled",
168 if (!strcmp(smt_option
, "on"))
169 smt_enabled_at_boot
= threads_per_core
;
170 else if (!strcmp(smt_option
, "off"))
171 smt_enabled_at_boot
= 0;
179 /* Look for smt-enabled= cmdline option */
180 static int __init
early_smt_enabled(char *p
)
182 smt_enabled_cmdline
= p
;
185 early_param("smt-enabled", early_smt_enabled
);
188 #define check_smt_enabled()
189 #endif /* CONFIG_SMP */
191 /** Fix up paca fields required for the boot cpu */
192 static void fixup_boot_paca(void)
194 /* The boot cpu is started */
195 get_paca()->cpu_start
= 1;
196 /* Allow percpu accesses to work until we setup percpu data */
197 get_paca()->data_offset
= 0;
200 static void cpu_ready_for_interrupts(void)
202 /* Set IR and DR in PACA MSR */
203 get_paca()->kernel_msr
= MSR_KERNEL
;
206 * Enable AIL if supported, and we are in hypervisor mode. If we are
207 * not in hypervisor mode, we enable relocation-on interrupts later
208 * in pSeries_setup_arch() using the H_SET_MODE hcall.
210 if (cpu_has_feature(CPU_FTR_HVMODE
) &&
211 cpu_has_feature(CPU_FTR_ARCH_207S
)) {
212 unsigned long lpcr
= mfspr(SPRN_LPCR
);
213 mtspr(SPRN_LPCR
, lpcr
| LPCR_AIL_3
);
218 * Early initialization entry point. This is called by head.S
219 * with MMU translation disabled. We rely on the "feature" of
220 * the CPU that ignores the top 2 bits of the address in real
221 * mode so we can access kernel globals normally provided we
222 * only toy with things in the RMO region. From here, we do
223 * some early parsing of the device-tree to setup out MEMBLOCK
224 * data structures, and allocate & initialize the hash table
225 * and segment tables so we can start running with translation
228 * It is this function which will call the probe() callback of
229 * the various platform types and copy the matching one to the
230 * global ppc_md structure. Your platform can eventually do
231 * some very early initializations from the probe() routine, but
232 * this is not recommended, be very careful as, for example, the
233 * device-tree is not accessible via normal means at this point.
236 void __init
early_setup(unsigned long dt_ptr
)
238 static __initdata
struct paca_struct boot_paca
;
240 /* -------- printk is _NOT_ safe to use here ! ------- */
242 /* Identify CPU type */
243 identify_cpu(0, mfspr(SPRN_PVR
));
245 /* Assume we're on cpu 0 for now. Don't write to the paca yet! */
246 initialise_paca(&boot_paca
, 0);
247 setup_paca(&boot_paca
);
250 /* Initialize lockdep early or else spinlocks will blow */
253 /* -------- printk is now safe to use ------- */
255 /* Enable early debugging if any specified (see udbg.h) */
258 DBG(" -> early_setup(), dt_ptr: 0x%lx\n", dt_ptr
);
261 * Do early initialization using the flattened device
262 * tree, such as retrieving the physical memory map or
263 * calculating/retrieving the hash table size.
265 early_init_devtree(__va(dt_ptr
));
267 epapr_paravirt_early_init();
269 /* Now we know the logical id of our boot cpu, setup the paca. */
270 setup_paca(&paca
[boot_cpuid
]);
273 /* Probe the machine type */
276 setup_kdump_trampoline();
278 DBG("Found, Initializing memory management...\n");
280 /* Initialize the hash table or TLB handling */
284 * At this point, we can let interrupts switch to virtual mode
285 * (the MMU has been setup), so adjust the MSR in the PACA to
286 * have IR and DR set and enable AIL if it exists
288 cpu_ready_for_interrupts();
290 /* Reserve large chunks of memory for use by CMA for KVM */
294 * Reserve any gigantic pages requested on the command line.
295 * memblock needs to have been initialized by the time this is
296 * called since this will reserve memory.
298 reserve_hugetlb_gpages();
300 DBG(" <- early_setup()\n");
302 #ifdef CONFIG_PPC_EARLY_DEBUG_BOOTX
304 * This needs to be done *last* (after the above DBG() even)
306 * Right after we return from this function, we turn on the MMU
307 * which means the real-mode access trick that btext does will
308 * no longer work, it needs to switch to using a real MMU
309 * mapping. This call will ensure that it does
312 #endif /* CONFIG_PPC_EARLY_DEBUG_BOOTX */
316 void early_setup_secondary(void)
318 /* Mark interrupts enabled in PACA */
319 get_paca()->soft_enabled
= 0;
321 /* Initialize the hash table or TLB handling */
322 early_init_mmu_secondary();
325 * At this point, we can let interrupts switch to virtual mode
326 * (the MMU has been setup), so adjust the MSR in the PACA to
327 * have IR and DR set.
329 cpu_ready_for_interrupts();
332 #endif /* CONFIG_SMP */
334 #if defined(CONFIG_SMP) || defined(CONFIG_KEXEC)
335 void smp_release_cpus(void)
340 DBG(" -> smp_release_cpus()\n");
342 /* All secondary cpus are spinning on a common spinloop, release them
343 * all now so they can start to spin on their individual paca
344 * spinloops. For non SMP kernels, the secondary cpus never get out
345 * of the common spinloop.
348 ptr
= (unsigned long *)((unsigned long)&__secondary_hold_spinloop
350 *ptr
= ppc_function_entry(generic_secondary_smp_init
);
352 /* And wait a bit for them to catch up */
353 for (i
= 0; i
< 100000; i
++) {
356 if (spinning_secondaries
== 0)
360 DBG("spinning_secondaries = %d\n", spinning_secondaries
);
362 DBG(" <- smp_release_cpus()\n");
364 #endif /* CONFIG_SMP || CONFIG_KEXEC */
367 * Initialize some remaining members of the ppc64_caches and systemcfg
369 * (at least until we get rid of them completely). This is mostly some
370 * cache informations about the CPU that will be used by cache flush
371 * routines and/or provided to userland
373 static void __init
initialize_cache_info(void)
375 struct device_node
*np
;
376 unsigned long num_cpus
= 0;
378 DBG(" -> initialize_cache_info()\n");
380 for_each_node_by_type(np
, "cpu") {
384 * We're assuming *all* of the CPUs have the same
385 * d-cache and i-cache sizes... -Peter
388 const __be32
*sizep
, *lsizep
;
392 lsize
= cur_cpu_spec
->dcache_bsize
;
393 sizep
= of_get_property(np
, "d-cache-size", NULL
);
395 size
= be32_to_cpu(*sizep
);
396 lsizep
= of_get_property(np
, "d-cache-block-size",
398 /* fallback if block size missing */
400 lsizep
= of_get_property(np
,
404 lsize
= be32_to_cpu(*lsizep
);
405 if (sizep
== NULL
|| lsizep
== NULL
)
406 DBG("Argh, can't find dcache properties ! "
407 "sizep: %p, lsizep: %p\n", sizep
, lsizep
);
409 ppc64_caches
.dsize
= size
;
410 ppc64_caches
.dline_size
= lsize
;
411 ppc64_caches
.log_dline_size
= __ilog2(lsize
);
412 ppc64_caches
.dlines_per_page
= PAGE_SIZE
/ lsize
;
415 lsize
= cur_cpu_spec
->icache_bsize
;
416 sizep
= of_get_property(np
, "i-cache-size", NULL
);
418 size
= be32_to_cpu(*sizep
);
419 lsizep
= of_get_property(np
, "i-cache-block-size",
422 lsizep
= of_get_property(np
,
426 lsize
= be32_to_cpu(*lsizep
);
427 if (sizep
== NULL
|| lsizep
== NULL
)
428 DBG("Argh, can't find icache properties ! "
429 "sizep: %p, lsizep: %p\n", sizep
, lsizep
);
431 ppc64_caches
.isize
= size
;
432 ppc64_caches
.iline_size
= lsize
;
433 ppc64_caches
.log_iline_size
= __ilog2(lsize
);
434 ppc64_caches
.ilines_per_page
= PAGE_SIZE
/ lsize
;
438 DBG(" <- initialize_cache_info()\n");
443 * Do some initial setup of the system. The parameters are those which
444 * were passed in from the bootloader.
446 void __init
setup_system(void)
448 DBG(" -> setup_system()\n");
450 /* Apply the CPUs-specific and firmware specific fixups to kernel
451 * text (nop out sections not relevant to this CPU or this firmware)
453 do_feature_fixups(cur_cpu_spec
->cpu_features
,
454 &__start___ftr_fixup
, &__stop___ftr_fixup
);
455 do_feature_fixups(cur_cpu_spec
->mmu_features
,
456 &__start___mmu_ftr_fixup
, &__stop___mmu_ftr_fixup
);
457 do_feature_fixups(powerpc_firmware_features
,
458 &__start___fw_ftr_fixup
, &__stop___fw_ftr_fixup
);
459 do_lwsync_fixups(cur_cpu_spec
->cpu_features
,
460 &__start___lwsync_fixup
, &__stop___lwsync_fixup
);
464 * Unflatten the device-tree passed by prom_init or kexec
466 unflatten_device_tree();
469 * Fill the ppc64_caches & systemcfg structures with informations
470 * retrieved from the device-tree.
472 initialize_cache_info();
474 #ifdef CONFIG_PPC_RTAS
476 * Initialize RTAS if available
479 #endif /* CONFIG_PPC_RTAS */
482 * Check if we have an initrd provided via the device-tree
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()
491 if (ppc_md
.init_early
)
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
499 find_legacy_serial_ports();
502 * Register early console
504 register_early_udbg_console();
511 smp_setup_cpu_maps();
513 setup_tlb_core_data();
516 * Freescale Book3e parts spin in a loop provided by firmware,
517 * so smp_release_cpus() does nothing for them
519 #if defined(CONFIG_SMP) && !defined(CONFIG_PPC_FSL_BOOK3E)
520 /* Release secondary cpus out of their spinloops at 0x60 now that
521 * we can map physical -> logical CPU ids
526 pr_info("Starting Linux PPC64 %s\n", init_utsname()->version
);
528 pr_info("-----------------------------------------------------\n");
529 pr_info("ppc64_pft_size = 0x%llx\n", ppc64_pft_size
);
530 pr_info("phys_mem_size = 0x%llx\n", memblock_phys_mem_size());
532 if (ppc64_caches
.dline_size
!= 0x80)
533 pr_info("dcache_line_size = 0x%x\n", ppc64_caches
.dline_size
);
534 if (ppc64_caches
.iline_size
!= 0x80)
535 pr_info("icache_line_size = 0x%x\n", ppc64_caches
.iline_size
);
537 pr_info("cpu_features = 0x%016lx\n", cur_cpu_spec
->cpu_features
);
538 pr_info(" possible = 0x%016lx\n", CPU_FTRS_POSSIBLE
);
539 pr_info(" always = 0x%016lx\n", CPU_FTRS_ALWAYS
);
540 pr_info("cpu_user_features = 0x%08x 0x%08x\n", cur_cpu_spec
->cpu_user_features
,
541 cur_cpu_spec
->cpu_user_features2
);
542 pr_info("mmu_features = 0x%08x\n", cur_cpu_spec
->mmu_features
);
543 pr_info("firmware_features = 0x%016lx\n", powerpc_firmware_features
);
545 #ifdef CONFIG_PPC_STD_MMU_64
547 pr_info("htab_address = 0x%p\n", htab_address
);
549 pr_info("htab_hash_mask = 0x%lx\n", htab_hash_mask
);
552 if (PHYSICAL_START
> 0)
553 pr_info("physical_start = 0x%llx\n",
554 (unsigned long long)PHYSICAL_START
);
555 pr_info("-----------------------------------------------------\n");
557 DBG(" <- setup_system()\n");
560 /* This returns the limit below which memory accesses to the linear
561 * mapping are guarnateed not to cause a TLB or SLB miss. This is
562 * used to allocate interrupt or emergency stacks for which our
563 * exception entry path doesn't deal with being interrupted.
565 static u64
safe_stack_limit(void)
567 #ifdef CONFIG_PPC_BOOK3E
568 /* Freescale BookE bolts the entire linear mapping */
569 if (mmu_has_feature(MMU_FTR_TYPE_FSL_E
))
570 return linear_map_top
;
571 /* Other BookE, we assume the first GB is bolted */
574 /* BookS, the first segment is bolted */
575 if (mmu_has_feature(MMU_FTR_1T_SEGMENT
))
576 return 1UL << SID_SHIFT_1T
;
577 return 1UL << SID_SHIFT
;
581 static void __init
irqstack_early_init(void)
583 u64 limit
= safe_stack_limit();
587 * Interrupt stacks must be in the first segment since we
588 * cannot afford to take SLB misses on them.
590 for_each_possible_cpu(i
) {
591 softirq_ctx
[i
] = (struct thread_info
*)
592 __va(memblock_alloc_base(THREAD_SIZE
,
593 THREAD_SIZE
, limit
));
594 hardirq_ctx
[i
] = (struct thread_info
*)
595 __va(memblock_alloc_base(THREAD_SIZE
,
596 THREAD_SIZE
, limit
));
600 #ifdef CONFIG_PPC_BOOK3E
601 static void __init
exc_lvl_early_init(void)
606 for_each_possible_cpu(i
) {
607 sp
= memblock_alloc(THREAD_SIZE
, THREAD_SIZE
);
608 critirq_ctx
[i
] = (struct thread_info
*)__va(sp
);
609 paca
[i
].crit_kstack
= __va(sp
+ THREAD_SIZE
);
611 sp
= memblock_alloc(THREAD_SIZE
, THREAD_SIZE
);
612 dbgirq_ctx
[i
] = (struct thread_info
*)__va(sp
);
613 paca
[i
].dbg_kstack
= __va(sp
+ THREAD_SIZE
);
615 sp
= memblock_alloc(THREAD_SIZE
, THREAD_SIZE
);
616 mcheckirq_ctx
[i
] = (struct thread_info
*)__va(sp
);
617 paca
[i
].mc_kstack
= __va(sp
+ THREAD_SIZE
);
620 if (cpu_has_feature(CPU_FTR_DEBUG_LVL_EXC
))
621 patch_exception(0x040, exc_debug_debug_book3e
);
624 #define exc_lvl_early_init()
628 * Stack space used when we detect a bad kernel stack pointer, and
629 * early in SMP boots before relocation is enabled. Exclusive emergency
630 * stack for machine checks.
632 static void __init
emergency_stack_init(void)
638 * Emergency stacks must be under 256MB, we cannot afford to take
639 * SLB misses on them. The ABI also requires them to be 128-byte
642 * Since we use these as temporary stacks during secondary CPU
643 * bringup, we need to get at them in real mode. This means they
644 * must also be within the RMO region.
646 limit
= min(safe_stack_limit(), ppc64_rma_size
);
648 for_each_possible_cpu(i
) {
650 sp
= memblock_alloc_base(THREAD_SIZE
, THREAD_SIZE
, limit
);
652 paca
[i
].emergency_sp
= __va(sp
);
654 #ifdef CONFIG_PPC_BOOK3S_64
655 /* emergency stack for machine check exception handling. */
656 sp
= memblock_alloc_base(THREAD_SIZE
, THREAD_SIZE
, limit
);
658 paca
[i
].mc_emergency_sp
= __va(sp
);
664 * Called into from start_kernel this initializes memblock, which is used
665 * to manage page allocation until mem_init is called.
667 void __init
setup_arch(char **cmdline_p
)
669 *cmdline_p
= boot_command_line
;
672 * Set cache line size based on type of cpu as a default.
673 * Systems with OF can look in the properties on the cpu node(s)
674 * for a possibly more accurate value.
676 dcache_bsize
= ppc64_caches
.dline_size
;
677 icache_bsize
= ppc64_caches
.iline_size
;
682 init_mm
.start_code
= (unsigned long)_stext
;
683 init_mm
.end_code
= (unsigned long) _etext
;
684 init_mm
.end_data
= (unsigned long) _edata
;
685 init_mm
.brk
= klimit
;
686 #ifdef CONFIG_PPC_64K_PAGES
687 init_mm
.context
.pte_frag
= NULL
;
689 irqstack_early_init();
690 exc_lvl_early_init();
691 emergency_stack_init();
695 #ifdef CONFIG_DUMMY_CONSOLE
696 conswitchp
= &dummy_con
;
699 if (ppc_md
.setup_arch
)
704 /* Initialize the MMU context management stuff */
707 /* Interrupt code needs to be 64K-aligned */
708 if ((unsigned long)_stext
& 0xffff)
709 panic("Kernelbase not 64K-aligned (0x%lx)!\n",
710 (unsigned long)_stext
);
714 #define PCPU_DYN_SIZE ()
716 static void * __init
pcpu_fc_alloc(unsigned int cpu
, size_t size
, size_t align
)
718 return __alloc_bootmem_node(NODE_DATA(cpu_to_node(cpu
)), size
, align
,
719 __pa(MAX_DMA_ADDRESS
));
722 static void __init
pcpu_fc_free(void *ptr
, size_t size
)
724 free_bootmem(__pa(ptr
), size
);
727 static int pcpu_cpu_distance(unsigned int from
, unsigned int to
)
729 if (cpu_to_node(from
) == cpu_to_node(to
))
730 return LOCAL_DISTANCE
;
732 return REMOTE_DISTANCE
;
735 unsigned long __per_cpu_offset
[NR_CPUS
] __read_mostly
;
736 EXPORT_SYMBOL(__per_cpu_offset
);
738 void __init
setup_per_cpu_areas(void)
740 const size_t dyn_size
= PERCPU_MODULE_RESERVE
+ PERCPU_DYNAMIC_RESERVE
;
747 * Linear mapping is one of 4K, 1M and 16M. For 4K, no need
748 * to group units. For larger mappings, use 1M atom which
749 * should be large enough to contain a number of units.
751 if (mmu_linear_psize
== MMU_PAGE_4K
)
752 atom_size
= PAGE_SIZE
;
756 rc
= pcpu_embed_first_chunk(0, dyn_size
, atom_size
, pcpu_cpu_distance
,
757 pcpu_fc_alloc
, pcpu_fc_free
);
759 panic("cannot initialize percpu area (err=%d)", rc
);
761 delta
= (unsigned long)pcpu_base_addr
- (unsigned long)__per_cpu_start
;
762 for_each_possible_cpu(cpu
) {
763 __per_cpu_offset
[cpu
] = delta
+ pcpu_unit_offsets
[cpu
];
764 paca
[cpu
].data_offset
= __per_cpu_offset
[cpu
];
769 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
770 unsigned long memory_block_size_bytes(void)
772 if (ppc_md
.memory_block_size
)
773 return ppc_md
.memory_block_size();
775 return MIN_MEMORY_BLOCK_SIZE
;
779 #if defined(CONFIG_PPC_INDIRECT_PIO) || defined(CONFIG_PPC_INDIRECT_MMIO)
780 struct ppc_pci_io ppc_pci_io
;
781 EXPORT_SYMBOL(ppc_pci_io
);
784 #ifdef CONFIG_HARDLOCKUP_DETECTOR
785 u64
hw_nmi_get_sample_period(int watchdog_thresh
)
787 return ppc_proc_freq
* watchdog_thresh
;
791 * The hardlockup detector breaks PMU event based branches and is likely
792 * to get false positives in KVM guests, so disable it by default.
794 static int __init
disable_hardlockup_detector(void)
796 hardlockup_detector_disable();
800 early_initcall(disable_hardlockup_detector
);