1 // SPDX-License-Identifier: GPL-2.0-or-later
4 * Common boot and setup code.
6 * Copyright (C) 2001 PPC64 Team, IBM Corp
9 #include <linux/export.h>
10 #include <linux/string.h>
11 #include <linux/sched.h>
12 #include <linux/init.h>
13 #include <linux/kernel.h>
14 #include <linux/reboot.h>
15 #include <linux/delay.h>
16 #include <linux/initrd.h>
17 #include <linux/seq_file.h>
18 #include <linux/ioport.h>
19 #include <linux/console.h>
20 #include <linux/utsname.h>
21 #include <linux/tty.h>
22 #include <linux/root_dev.h>
23 #include <linux/notifier.h>
24 #include <linux/cpu.h>
25 #include <linux/unistd.h>
26 #include <linux/serial.h>
27 #include <linux/serial_8250.h>
28 #include <linux/memblock.h>
29 #include <linux/pci.h>
30 #include <linux/lockdep.h>
31 #include <linux/memory.h>
32 #include <linux/nmi.h>
33 #include <linux/pgtable.h>
35 #include <asm/debugfs.h>
37 #include <asm/kdump.h>
39 #include <asm/processor.h>
42 #include <asm/machdep.h>
45 #include <asm/cputable.h>
46 #include <asm/dt_cpu_ftrs.h>
47 #include <asm/sections.h>
48 #include <asm/btext.h>
49 #include <asm/nvram.h>
50 #include <asm/setup.h>
52 #include <asm/iommu.h>
53 #include <asm/serial.h>
54 #include <asm/cache.h>
57 #include <asm/firmware.h>
60 #include <asm/kexec.h>
61 #include <asm/code-patching.h>
62 #include <asm/livepatch.h>
64 #include <asm/cputhreads.h>
65 #include <asm/hw_irq.h>
66 #include <asm/feature-fixups.h>
68 #include <asm/early_ioremap.h>
69 #include <asm/pgalloc.h>
73 int spinning_secondaries
;
76 struct ppc64_caches ppc64_caches
= {
86 EXPORT_SYMBOL_GPL(ppc64_caches
);
88 #if defined(CONFIG_PPC_BOOK3E) && defined(CONFIG_SMP)
89 void __init
setup_tlb_core_data(void)
93 BUILD_BUG_ON(offsetof(struct tlb_core_data
, lock
) != 0);
95 for_each_possible_cpu(cpu
) {
96 int first
= cpu_first_thread_sibling(cpu
);
99 * If we boot via kdump on a non-primary thread,
100 * make sure we point at the thread that actually
103 if (cpu_first_thread_sibling(boot_cpuid
) == first
)
106 paca_ptrs
[cpu
]->tcd_ptr
= &paca_ptrs
[first
]->tcd
;
109 * If we have threads, we need either tlbsrx.
110 * or e6500 tablewalk mode, or else TLB handlers
111 * will be racy and could produce duplicate entries.
112 * Should we panic instead?
114 WARN_ONCE(smt_enabled_at_boot
>= 2 &&
115 !mmu_has_feature(MMU_FTR_USE_TLBRSRV
) &&
116 book3e_htw_mode
!= PPC_HTW_E6500
,
117 "%s: unsupported MMU configuration\n", __func__
);
124 static char *smt_enabled_cmdline
;
126 /* Look for ibm,smt-enabled OF option */
127 void __init
check_smt_enabled(void)
129 struct device_node
*dn
;
130 const char *smt_option
;
132 /* Default to enabling all threads */
133 smt_enabled_at_boot
= threads_per_core
;
135 /* Allow the command line to overrule the OF option */
136 if (smt_enabled_cmdline
) {
137 if (!strcmp(smt_enabled_cmdline
, "on"))
138 smt_enabled_at_boot
= threads_per_core
;
139 else if (!strcmp(smt_enabled_cmdline
, "off"))
140 smt_enabled_at_boot
= 0;
145 rc
= kstrtoint(smt_enabled_cmdline
, 10, &smt
);
147 smt_enabled_at_boot
=
148 min(threads_per_core
, smt
);
151 dn
= of_find_node_by_path("/options");
153 smt_option
= of_get_property(dn
, "ibm,smt-enabled",
157 if (!strcmp(smt_option
, "on"))
158 smt_enabled_at_boot
= threads_per_core
;
159 else if (!strcmp(smt_option
, "off"))
160 smt_enabled_at_boot
= 0;
168 /* Look for smt-enabled= cmdline option */
169 static int __init
early_smt_enabled(char *p
)
171 smt_enabled_cmdline
= p
;
174 early_param("smt-enabled", early_smt_enabled
);
176 #endif /* CONFIG_SMP */
178 /** Fix up paca fields required for the boot cpu */
179 static void __init
fixup_boot_paca(void)
181 /* The boot cpu is started */
182 get_paca()->cpu_start
= 1;
183 /* Allow percpu accesses to work until we setup percpu data */
184 get_paca()->data_offset
= 0;
185 /* Mark interrupts disabled in PACA */
186 irq_soft_mask_set(IRQS_DISABLED
);
189 static void __init
configure_exceptions(void)
192 * Setup the trampolines from the lowmem exception vectors
193 * to the kdump kernel when not using a relocatable kernel.
195 setup_kdump_trampoline();
197 /* Under a PAPR hypervisor, we need hypercalls */
198 if (firmware_has_feature(FW_FEATURE_SET_MODE
)) {
199 /* Enable AIL if possible */
200 if (!pseries_enable_reloc_on_exc()) {
201 init_task
.thread
.fscr
&= ~FSCR_SCV
;
202 cur_cpu_spec
->cpu_user_features2
&= ~PPC_FEATURE2_SCV
;
206 * Tell the hypervisor that we want our exceptions to
207 * be taken in little endian mode.
209 * We don't call this for big endian as our calling convention
210 * makes us always enter in BE, and the call may fail under
211 * some circumstances with kdump.
213 #ifdef __LITTLE_ENDIAN__
214 pseries_little_endian_exceptions();
217 /* Set endian mode using OPAL */
218 if (firmware_has_feature(FW_FEATURE_OPAL
))
219 opal_configure_cores();
221 /* AIL on native is done in cpu_ready_for_interrupts() */
225 static void cpu_ready_for_interrupts(void)
228 * Enable AIL if supported, and we are in hypervisor mode. This
229 * is called once for every processor.
231 * If we are not in hypervisor mode the job is done once for
232 * the whole partition in configure_exceptions().
234 if (cpu_has_feature(CPU_FTR_HVMODE
)) {
235 unsigned long lpcr
= mfspr(SPRN_LPCR
);
236 unsigned long new_lpcr
= lpcr
;
238 if (cpu_has_feature(CPU_FTR_ARCH_31
)) {
239 /* P10 DD1 does not have HAIL */
240 if (pvr_version_is(PVR_POWER10
) &&
241 (mfspr(SPRN_PVR
) & 0xf00) == 0x100)
242 new_lpcr
|= LPCR_AIL_3
;
244 new_lpcr
|= LPCR_HAIL
;
245 } else if (cpu_has_feature(CPU_FTR_ARCH_207S
)) {
246 new_lpcr
|= LPCR_AIL_3
;
249 if (new_lpcr
!= lpcr
)
250 mtspr(SPRN_LPCR
, new_lpcr
);
254 * Set HFSCR:TM based on CPU features:
255 * In the special case of TM no suspend (P9N DD2.1), Linux is
256 * told TM is off via the dt-ftrs but told to (partially) use
257 * it via OPAL_REINIT_CPUS_TM_SUSPEND_DISABLED. So HFSCR[TM]
258 * will be off from dt-ftrs but we need to turn it on for the
261 if (cpu_has_feature(CPU_FTR_HVMODE
)) {
262 if (cpu_has_feature(CPU_FTR_TM_COMP
))
263 mtspr(SPRN_HFSCR
, mfspr(SPRN_HFSCR
) | HFSCR_TM
);
265 mtspr(SPRN_HFSCR
, mfspr(SPRN_HFSCR
) & ~HFSCR_TM
);
268 /* Set IR and DR in PACA MSR */
269 get_paca()->kernel_msr
= MSR_KERNEL
;
272 unsigned long spr_default_dscr
= 0;
274 void __init
record_spr_defaults(void)
276 if (early_cpu_has_feature(CPU_FTR_DSCR
))
277 spr_default_dscr
= mfspr(SPRN_DSCR
);
281 * Early initialization entry point. This is called by head.S
282 * with MMU translation disabled. We rely on the "feature" of
283 * the CPU that ignores the top 2 bits of the address in real
284 * mode so we can access kernel globals normally provided we
285 * only toy with things in the RMO region. From here, we do
286 * some early parsing of the device-tree to setup out MEMBLOCK
287 * data structures, and allocate & initialize the hash table
288 * and segment tables so we can start running with translation
291 * It is this function which will call the probe() callback of
292 * the various platform types and copy the matching one to the
293 * global ppc_md structure. Your platform can eventually do
294 * some very early initializations from the probe() routine, but
295 * this is not recommended, be very careful as, for example, the
296 * device-tree is not accessible via normal means at this point.
299 void __init
early_setup(unsigned long dt_ptr
)
301 static __initdata
struct paca_struct boot_paca
;
303 /* -------- printk is _NOT_ safe to use here ! ------- */
306 * Assume we're on cpu 0 for now.
308 * We need to load a PACA very early for a few reasons.
310 * The stack protector canary is stored in the paca, so as soon as we
311 * call any stack protected code we need r13 pointing somewhere valid.
313 * If we are using kcov it will call in_task() in its instrumentation,
314 * which relies on the current task from the PACA.
316 * dt_cpu_ftrs_init() calls into generic OF/fdt code, as well as
317 * printk(), which can trigger both stack protector and kcov.
319 * percpu variables and spin locks also use the paca.
321 * So set up a temporary paca. It will be replaced below once we know
322 * what CPU we are on.
324 initialise_paca(&boot_paca
, 0);
325 setup_paca(&boot_paca
);
328 /* -------- printk is now safe to use ------- */
330 /* Try new device tree based feature discovery ... */
331 if (!dt_cpu_ftrs_init(__va(dt_ptr
)))
332 /* Otherwise use the old style CPU table */
333 identify_cpu(0, mfspr(SPRN_PVR
));
335 /* Enable early debugging if any specified (see udbg.h) */
338 udbg_printf(" -> %s(), dt_ptr: 0x%lx\n", __func__
, dt_ptr
);
341 * Do early initialization using the flattened device
342 * tree, such as retrieving the physical memory map or
343 * calculating/retrieving the hash table size.
345 early_init_devtree(__va(dt_ptr
));
347 /* Now we know the logical id of our boot cpu, setup the paca. */
348 if (boot_cpuid
!= 0) {
349 /* Poison paca_ptrs[0] again if it's not the boot cpu */
350 memset(&paca_ptrs
[0], 0x88, sizeof(paca_ptrs
[0]));
352 setup_paca(paca_ptrs
[boot_cpuid
]);
356 * Configure exception handlers. This include setting up trampolines
357 * if needed, setting exception endian mode, etc...
359 configure_exceptions();
362 * Configure Kernel Userspace Protection. This needs to happen before
363 * feature fixups for platforms that implement this using features.
367 /* Apply all the dynamic patching */
368 apply_feature_fixups();
369 setup_feature_keys();
371 early_ioremap_setup();
373 /* Initialize the hash table or TLB handling */
377 * After firmware and early platform setup code has set things up,
378 * we note the SPR values for configurable control/performance
379 * registers, and use those as initial defaults.
381 record_spr_defaults();
384 * At this point, we can let interrupts switch to virtual mode
385 * (the MMU has been setup), so adjust the MSR in the PACA to
386 * have IR and DR set and enable AIL if it exists
388 cpu_ready_for_interrupts();
391 * We enable ftrace here, but since we only support DYNAMIC_FTRACE, it
392 * will only actually get enabled on the boot cpu much later once
393 * ftrace itself has been initialized.
395 this_cpu_enable_ftrace();
397 udbg_printf(" <- %s()\n", __func__
);
399 #ifdef CONFIG_PPC_EARLY_DEBUG_BOOTX
401 * This needs to be done *last* (after the above udbg_printf() even)
403 * Right after we return from this function, we turn on the MMU
404 * which means the real-mode access trick that btext does will
405 * no longer work, it needs to switch to using a real MMU
406 * mapping. This call will ensure that it does
409 #endif /* CONFIG_PPC_EARLY_DEBUG_BOOTX */
413 void early_setup_secondary(void)
415 /* Mark interrupts disabled in PACA */
416 irq_soft_mask_set(IRQS_DISABLED
);
418 /* Initialize the hash table or TLB handling */
419 early_init_mmu_secondary();
421 /* Perform any KUP setup that is per-cpu */
425 * At this point, we can let interrupts switch to virtual mode
426 * (the MMU has been setup), so adjust the MSR in the PACA to
427 * have IR and DR set.
429 cpu_ready_for_interrupts();
432 #endif /* CONFIG_SMP */
434 void panic_smp_self_stop(void)
442 #if defined(CONFIG_SMP) || defined(CONFIG_KEXEC_CORE)
443 static bool use_spinloop(void)
445 if (IS_ENABLED(CONFIG_PPC_BOOK3S
)) {
447 * See comments in head_64.S -- not all platforms insert
448 * secondaries at __secondary_hold and wait at the spin
451 if (firmware_has_feature(FW_FEATURE_OPAL
))
457 * When book3e boots from kexec, the ePAPR spin table does
460 return of_property_read_bool(of_chosen
, "linux,booted-from-kexec");
463 void smp_release_cpus(void)
471 /* All secondary cpus are spinning on a common spinloop, release them
472 * all now so they can start to spin on their individual paca
473 * spinloops. For non SMP kernels, the secondary cpus never get out
474 * of the common spinloop.
477 ptr
= (unsigned long *)((unsigned long)&__secondary_hold_spinloop
479 *ptr
= ppc_function_entry(generic_secondary_smp_init
);
481 /* And wait a bit for them to catch up */
482 for (i
= 0; i
< 100000; i
++) {
485 if (spinning_secondaries
== 0)
489 pr_debug("spinning_secondaries = %d\n", spinning_secondaries
);
491 #endif /* CONFIG_SMP || CONFIG_KEXEC_CORE */
494 * Initialize some remaining members of the ppc64_caches and systemcfg
496 * (at least until we get rid of them completely). This is mostly some
497 * cache informations about the CPU that will be used by cache flush
498 * routines and/or provided to userland
501 static void init_cache_info(struct ppc_cache_info
*info
, u32 size
, u32 lsize
,
506 info
->line_size
= lsize
;
507 info
->block_size
= bsize
;
508 info
->log_block_size
= __ilog2(bsize
);
510 info
->blocks_per_page
= PAGE_SIZE
/ bsize
;
512 info
->blocks_per_page
= 0;
515 info
->assoc
= 0xffff;
517 info
->assoc
= size
/ (sets
* lsize
);
520 static bool __init
parse_cache_info(struct device_node
*np
,
522 struct ppc_cache_info
*info
)
524 static const char *ipropnames
[] __initdata
= {
527 "i-cache-block-size",
530 static const char *dpropnames
[] __initdata
= {
533 "d-cache-block-size",
536 const char **propnames
= icache
? ipropnames
: dpropnames
;
537 const __be32
*sizep
, *lsizep
, *bsizep
, *setsp
;
538 u32 size
, lsize
, bsize
, sets
;
543 lsize
= bsize
= cur_cpu_spec
->dcache_bsize
;
544 sizep
= of_get_property(np
, propnames
[0], NULL
);
546 size
= be32_to_cpu(*sizep
);
547 setsp
= of_get_property(np
, propnames
[1], NULL
);
549 sets
= be32_to_cpu(*setsp
);
550 bsizep
= of_get_property(np
, propnames
[2], NULL
);
551 lsizep
= of_get_property(np
, propnames
[3], NULL
);
557 lsize
= be32_to_cpu(*lsizep
);
559 bsize
= be32_to_cpu(*bsizep
);
560 if (sizep
== NULL
|| bsizep
== NULL
|| lsizep
== NULL
)
564 * OF is weird .. it represents fully associative caches
565 * as "1 way" which doesn't make much sense and doesn't
566 * leave room for direct mapped. We'll assume that 0
567 * in OF means direct mapped for that reason.
574 init_cache_info(info
, size
, lsize
, bsize
, sets
);
579 void __init
initialize_cache_info(void)
581 struct device_node
*cpu
= NULL
, *l2
, *l3
= NULL
;
585 * All shipping POWER8 machines have a firmware bug that
586 * puts incorrect information in the device-tree. This will
587 * be (hopefully) fixed for future chips but for now hard
588 * code the values if we are running on one of these
590 pvr
= PVR_VER(mfspr(SPRN_PVR
));
591 if (pvr
== PVR_POWER8
|| pvr
== PVR_POWER8E
||
592 pvr
== PVR_POWER8NVL
) {
593 /* size lsize blk sets */
594 init_cache_info(&ppc64_caches
.l1i
, 0x8000, 128, 128, 32);
595 init_cache_info(&ppc64_caches
.l1d
, 0x10000, 128, 128, 64);
596 init_cache_info(&ppc64_caches
.l2
, 0x80000, 128, 0, 512);
597 init_cache_info(&ppc64_caches
.l3
, 0x800000, 128, 0, 8192);
599 cpu
= of_find_node_by_type(NULL
, "cpu");
602 * We're assuming *all* of the CPUs have the same
603 * d-cache and i-cache sizes... -Peter
606 if (!parse_cache_info(cpu
, false, &ppc64_caches
.l1d
))
607 pr_warn("Argh, can't find dcache properties !\n");
609 if (!parse_cache_info(cpu
, true, &ppc64_caches
.l1i
))
610 pr_warn("Argh, can't find icache properties !\n");
613 * Try to find the L2 and L3 if any. Assume they are
614 * unified and use the D-side properties.
616 l2
= of_find_next_cache_node(cpu
);
619 parse_cache_info(l2
, false, &ppc64_caches
.l2
);
620 l3
= of_find_next_cache_node(l2
);
624 parse_cache_info(l3
, false, &ppc64_caches
.l3
);
629 /* For use by binfmt_elf */
630 dcache_bsize
= ppc64_caches
.l1d
.block_size
;
631 icache_bsize
= ppc64_caches
.l1i
.block_size
;
633 cur_cpu_spec
->dcache_bsize
= dcache_bsize
;
634 cur_cpu_spec
->icache_bsize
= icache_bsize
;
638 * This returns the limit below which memory accesses to the linear
639 * mapping are guarnateed not to cause an architectural exception (e.g.,
640 * TLB or SLB miss fault).
642 * This is used to allocate PACAs and various interrupt stacks that
643 * that are accessed early in interrupt handlers that must not cause
644 * re-entrant interrupts.
646 __init u64
ppc64_bolted_size(void)
648 #ifdef CONFIG_PPC_BOOK3E
649 /* Freescale BookE bolts the entire linear mapping */
650 /* XXX: BookE ppc64_rma_limit setup seems to disagree? */
651 if (early_mmu_has_feature(MMU_FTR_TYPE_FSL_E
))
652 return linear_map_top
;
653 /* Other BookE, we assume the first GB is bolted */
656 /* BookS radix, does not take faults on linear mapping */
657 if (early_radix_enabled())
660 /* BookS hash, the first segment is bolted */
661 if (early_mmu_has_feature(MMU_FTR_1T_SEGMENT
))
662 return 1UL << SID_SHIFT_1T
;
663 return 1UL << SID_SHIFT
;
667 static void *__init
alloc_stack(unsigned long limit
, int cpu
)
671 BUILD_BUG_ON(STACK_INT_FRAME_SIZE
% 16);
673 ptr
= memblock_alloc_try_nid(THREAD_SIZE
, THREAD_ALIGN
,
674 MEMBLOCK_LOW_LIMIT
, limit
,
675 early_cpu_to_node(cpu
));
677 panic("cannot allocate stacks");
682 void __init
irqstack_early_init(void)
684 u64 limit
= ppc64_bolted_size();
688 * Interrupt stacks must be in the first segment since we
689 * cannot afford to take SLB misses on them. They are not
690 * accessed in realmode.
692 for_each_possible_cpu(i
) {
693 softirq_ctx
[i
] = alloc_stack(limit
, i
);
694 hardirq_ctx
[i
] = alloc_stack(limit
, i
);
698 #ifdef CONFIG_PPC_BOOK3E
699 void __init
exc_lvl_early_init(void)
703 for_each_possible_cpu(i
) {
706 sp
= alloc_stack(ULONG_MAX
, i
);
708 paca_ptrs
[i
]->crit_kstack
= sp
+ THREAD_SIZE
;
710 sp
= alloc_stack(ULONG_MAX
, i
);
712 paca_ptrs
[i
]->dbg_kstack
= sp
+ THREAD_SIZE
;
714 sp
= alloc_stack(ULONG_MAX
, i
);
715 mcheckirq_ctx
[i
] = sp
;
716 paca_ptrs
[i
]->mc_kstack
= sp
+ THREAD_SIZE
;
719 if (cpu_has_feature(CPU_FTR_DEBUG_LVL_EXC
))
720 patch_exception(0x040, exc_debug_debug_book3e
);
725 * Stack space used when we detect a bad kernel stack pointer, and
726 * early in SMP boots before relocation is enabled. Exclusive emergency
727 * stack for machine checks.
729 void __init
emergency_stack_init(void)
731 u64 limit
, mce_limit
;
735 * Emergency stacks must be under 256MB, we cannot afford to take
736 * SLB misses on them. The ABI also requires them to be 128-byte
739 * Since we use these as temporary stacks during secondary CPU
740 * bringup, machine check, system reset, and HMI, we need to get
741 * at them in real mode. This means they must also be within the RMO
744 * The IRQ stacks allocated elsewhere in this file are zeroed and
745 * initialized in kernel/irq.c. These are initialized here in order
746 * to have emergency stacks available as early as possible.
748 limit
= mce_limit
= min(ppc64_bolted_size(), ppc64_rma_size
);
751 * Machine check on pseries calls rtas, but can't use the static
752 * rtas_args due to a machine check hitting while the lock is held.
753 * rtas args have to be under 4GB, so the machine check stack is
754 * limited to 4GB so args can be put on stack.
756 if (firmware_has_feature(FW_FEATURE_LPAR
) && mce_limit
> SZ_4G
)
759 for_each_possible_cpu(i
) {
760 paca_ptrs
[i
]->emergency_sp
= alloc_stack(limit
, i
) + THREAD_SIZE
;
762 #ifdef CONFIG_PPC_BOOK3S_64
763 /* emergency stack for NMI exception handling. */
764 paca_ptrs
[i
]->nmi_emergency_sp
= alloc_stack(limit
, i
) + THREAD_SIZE
;
766 /* emergency stack for machine check exception handling. */
767 paca_ptrs
[i
]->mc_emergency_sp
= alloc_stack(mce_limit
, i
) + THREAD_SIZE
;
774 * pcpu_alloc_bootmem - NUMA friendly alloc_bootmem wrapper for percpu
775 * @cpu: cpu to allocate for
776 * @size: size allocation in bytes
779 * Allocate @size bytes aligned at @align for cpu @cpu. This wrapper
780 * does the right thing for NUMA regardless of the current
784 * Pointer to the allocated area on success, NULL on failure.
786 static void * __init
pcpu_alloc_bootmem(unsigned int cpu
, size_t size
,
789 const unsigned long goal
= __pa(MAX_DMA_ADDRESS
);
790 #ifdef CONFIG_NEED_MULTIPLE_NODES
791 int node
= early_cpu_to_node(cpu
);
794 if (!node_online(node
) || !NODE_DATA(node
)) {
795 ptr
= memblock_alloc_from(size
, align
, goal
);
796 pr_info("cpu %d has no node %d or node-local memory\n",
798 pr_debug("per cpu data for cpu%d %lu bytes at %016lx\n",
799 cpu
, size
, __pa(ptr
));
801 ptr
= memblock_alloc_try_nid(size
, align
, goal
,
802 MEMBLOCK_ALLOC_ACCESSIBLE
, node
);
803 pr_debug("per cpu data for cpu%d %lu bytes on node%d at "
804 "%016lx\n", cpu
, size
, node
, __pa(ptr
));
808 return memblock_alloc_from(size
, align
, goal
);
812 static void __init
pcpu_free_bootmem(void *ptr
, size_t size
)
814 memblock_free(__pa(ptr
), size
);
817 static int pcpu_cpu_distance(unsigned int from
, unsigned int to
)
819 if (early_cpu_to_node(from
) == early_cpu_to_node(to
))
820 return LOCAL_DISTANCE
;
822 return REMOTE_DISTANCE
;
825 unsigned long __per_cpu_offset
[NR_CPUS
] __read_mostly
;
826 EXPORT_SYMBOL(__per_cpu_offset
);
828 static void __init
pcpu_populate_pte(unsigned long addr
)
830 pgd_t
*pgd
= pgd_offset_k(addr
);
835 p4d
= p4d_offset(pgd
, addr
);
836 if (p4d_none(*p4d
)) {
839 new = memblock_alloc(PUD_TABLE_SIZE
, PUD_TABLE_SIZE
);
842 p4d_populate(&init_mm
, p4d
, new);
845 pud
= pud_offset(p4d
, addr
);
846 if (pud_none(*pud
)) {
849 new = memblock_alloc(PMD_TABLE_SIZE
, PMD_TABLE_SIZE
);
852 pud_populate(&init_mm
, pud
, new);
855 pmd
= pmd_offset(pud
, addr
);
856 if (!pmd_present(*pmd
)) {
859 new = memblock_alloc(PTE_TABLE_SIZE
, PTE_TABLE_SIZE
);
862 pmd_populate_kernel(&init_mm
, pmd
, new);
868 panic("%s: Failed to allocate %lu bytes align=%lx from=%lx\n",
869 __func__
, PAGE_SIZE
, PAGE_SIZE
, PAGE_SIZE
);
873 void __init
setup_per_cpu_areas(void)
875 const size_t dyn_size
= PERCPU_MODULE_RESERVE
+ PERCPU_DYNAMIC_RESERVE
;
882 * Linear mapping is one of 4K, 1M and 16M. For 4K, no need
883 * to group units. For larger mappings, use 1M atom which
884 * should be large enough to contain a number of units.
886 if (mmu_linear_psize
== MMU_PAGE_4K
)
887 atom_size
= PAGE_SIZE
;
891 if (pcpu_chosen_fc
!= PCPU_FC_PAGE
) {
892 rc
= pcpu_embed_first_chunk(0, dyn_size
, atom_size
, pcpu_cpu_distance
,
893 pcpu_alloc_bootmem
, pcpu_free_bootmem
);
895 pr_warn("PERCPU: %s allocator failed (%d), "
896 "falling back to page size\n",
897 pcpu_fc_names
[pcpu_chosen_fc
], rc
);
901 rc
= pcpu_page_first_chunk(0, pcpu_alloc_bootmem
, pcpu_free_bootmem
,
904 panic("cannot initialize percpu area (err=%d)", rc
);
906 delta
= (unsigned long)pcpu_base_addr
- (unsigned long)__per_cpu_start
;
907 for_each_possible_cpu(cpu
) {
908 __per_cpu_offset
[cpu
] = delta
+ pcpu_unit_offsets
[cpu
];
909 paca_ptrs
[cpu
]->data_offset
= __per_cpu_offset
[cpu
];
914 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
915 unsigned long memory_block_size_bytes(void)
917 if (ppc_md
.memory_block_size
)
918 return ppc_md
.memory_block_size();
920 return MIN_MEMORY_BLOCK_SIZE
;
924 #if defined(CONFIG_PPC_INDIRECT_PIO) || defined(CONFIG_PPC_INDIRECT_MMIO)
925 struct ppc_pci_io ppc_pci_io
;
926 EXPORT_SYMBOL(ppc_pci_io
);
929 #ifdef CONFIG_HARDLOCKUP_DETECTOR_PERF
930 u64
hw_nmi_get_sample_period(int watchdog_thresh
)
932 return ppc_proc_freq
* watchdog_thresh
;
937 * The perf based hardlockup detector breaks PMU event based branches, so
938 * disable it by default. Book3S has a soft-nmi hardlockup detector based
939 * on the decrementer interrupt, so it does not suffer from this problem.
941 * It is likely to get false positives in VM guests, so disable it there
944 static int __init
disable_hardlockup_detector(void)
946 #ifdef CONFIG_HARDLOCKUP_DETECTOR_PERF
947 hardlockup_detector_disable();
949 if (firmware_has_feature(FW_FEATURE_LPAR
))
950 hardlockup_detector_disable();
955 early_initcall(disable_hardlockup_detector
);
957 #ifdef CONFIG_PPC_BOOK3S_64
958 static enum l1d_flush_type enabled_flush_types
;
959 static void *l1d_flush_fallback_area
;
960 static bool no_rfi_flush
;
961 static bool no_entry_flush
;
962 static bool no_uaccess_flush
;
966 DEFINE_STATIC_KEY_FALSE(uaccess_flush_key
);
967 EXPORT_SYMBOL(uaccess_flush_key
);
969 static int __init
handle_no_rfi_flush(char *p
)
971 pr_info("rfi-flush: disabled on command line.");
975 early_param("no_rfi_flush", handle_no_rfi_flush
);
977 static int __init
handle_no_entry_flush(char *p
)
979 pr_info("entry-flush: disabled on command line.");
980 no_entry_flush
= true;
983 early_param("no_entry_flush", handle_no_entry_flush
);
985 static int __init
handle_no_uaccess_flush(char *p
)
987 pr_info("uaccess-flush: disabled on command line.");
988 no_uaccess_flush
= true;
991 early_param("no_uaccess_flush", handle_no_uaccess_flush
);
994 * The RFI flush is not KPTI, but because users will see doco that says to use
995 * nopti we hijack that option here to also disable the RFI flush.
997 static int __init
handle_no_pti(char *p
)
999 pr_info("rfi-flush: disabling due to 'nopti' on command line.\n");
1000 handle_no_rfi_flush(NULL
);
1003 early_param("nopti", handle_no_pti
);
1005 static void do_nothing(void *unused
)
1008 * We don't need to do the flush explicitly, just enter+exit kernel is
1009 * sufficient, the RFI exit handlers will do the right thing.
1013 void rfi_flush_enable(bool enable
)
1016 do_rfi_flush_fixups(enabled_flush_types
);
1017 on_each_cpu(do_nothing
, NULL
, 1);
1019 do_rfi_flush_fixups(L1D_FLUSH_NONE
);
1024 void entry_flush_enable(bool enable
)
1027 do_entry_flush_fixups(enabled_flush_types
);
1028 on_each_cpu(do_nothing
, NULL
, 1);
1030 do_entry_flush_fixups(L1D_FLUSH_NONE
);
1033 entry_flush
= enable
;
1036 void uaccess_flush_enable(bool enable
)
1039 do_uaccess_flush_fixups(enabled_flush_types
);
1040 static_branch_enable(&uaccess_flush_key
);
1041 on_each_cpu(do_nothing
, NULL
, 1);
1043 static_branch_disable(&uaccess_flush_key
);
1044 do_uaccess_flush_fixups(L1D_FLUSH_NONE
);
1047 uaccess_flush
= enable
;
1050 static void __ref
init_fallback_flush(void)
1052 u64 l1d_size
, limit
;
1055 /* Only allocate the fallback flush area once (at boot time). */
1056 if (l1d_flush_fallback_area
)
1059 l1d_size
= ppc64_caches
.l1d
.size
;
1062 * If there is no d-cache-size property in the device tree, l1d_size
1063 * could be zero. That leads to the loop in the asm wrapping around to
1064 * 2^64-1, and then walking off the end of the fallback area and
1065 * eventually causing a page fault which is fatal. Just default to
1066 * something vaguely sane.
1069 l1d_size
= (64 * 1024);
1071 limit
= min(ppc64_bolted_size(), ppc64_rma_size
);
1074 * Align to L1d size, and size it at 2x L1d size, to catch possible
1075 * hardware prefetch runoff. We don't have a recipe for load patterns to
1076 * reliably avoid the prefetcher.
1078 l1d_flush_fallback_area
= memblock_alloc_try_nid(l1d_size
* 2,
1079 l1d_size
, MEMBLOCK_LOW_LIMIT
,
1080 limit
, NUMA_NO_NODE
);
1081 if (!l1d_flush_fallback_area
)
1082 panic("%s: Failed to allocate %llu bytes align=0x%llx max_addr=%pa\n",
1083 __func__
, l1d_size
* 2, l1d_size
, &limit
);
1086 for_each_possible_cpu(cpu
) {
1087 struct paca_struct
*paca
= paca_ptrs
[cpu
];
1088 paca
->rfi_flush_fallback_area
= l1d_flush_fallback_area
;
1089 paca
->l1d_flush_size
= l1d_size
;
1093 void setup_rfi_flush(enum l1d_flush_type types
, bool enable
)
1095 if (types
& L1D_FLUSH_FALLBACK
) {
1096 pr_info("rfi-flush: fallback displacement flush available\n");
1097 init_fallback_flush();
1100 if (types
& L1D_FLUSH_ORI
)
1101 pr_info("rfi-flush: ori type flush available\n");
1103 if (types
& L1D_FLUSH_MTTRIG
)
1104 pr_info("rfi-flush: mttrig type flush available\n");
1106 enabled_flush_types
= types
;
1108 if (!cpu_mitigations_off() && !no_rfi_flush
)
1109 rfi_flush_enable(enable
);
1112 void setup_entry_flush(bool enable
)
1114 if (cpu_mitigations_off())
1117 if (!no_entry_flush
)
1118 entry_flush_enable(enable
);
1121 void setup_uaccess_flush(bool enable
)
1123 if (cpu_mitigations_off())
1126 if (!no_uaccess_flush
)
1127 uaccess_flush_enable(enable
);
1130 #ifdef CONFIG_DEBUG_FS
1131 static int rfi_flush_set(void *data
, u64 val
)
1142 /* Only do anything if we're changing state */
1143 if (enable
!= rfi_flush
)
1144 rfi_flush_enable(enable
);
1149 static int rfi_flush_get(void *data
, u64
*val
)
1151 *val
= rfi_flush
? 1 : 0;
1155 DEFINE_SIMPLE_ATTRIBUTE(fops_rfi_flush
, rfi_flush_get
, rfi_flush_set
, "%llu\n");
1157 static int entry_flush_set(void *data
, u64 val
)
1168 /* Only do anything if we're changing state */
1169 if (enable
!= entry_flush
)
1170 entry_flush_enable(enable
);
1175 static int entry_flush_get(void *data
, u64
*val
)
1177 *val
= entry_flush
? 1 : 0;
1181 DEFINE_SIMPLE_ATTRIBUTE(fops_entry_flush
, entry_flush_get
, entry_flush_set
, "%llu\n");
1183 static int uaccess_flush_set(void *data
, u64 val
)
1194 /* Only do anything if we're changing state */
1195 if (enable
!= uaccess_flush
)
1196 uaccess_flush_enable(enable
);
1201 static int uaccess_flush_get(void *data
, u64
*val
)
1203 *val
= uaccess_flush
? 1 : 0;
1207 DEFINE_SIMPLE_ATTRIBUTE(fops_uaccess_flush
, uaccess_flush_get
, uaccess_flush_set
, "%llu\n");
1209 static __init
int rfi_flush_debugfs_init(void)
1211 debugfs_create_file("rfi_flush", 0600, powerpc_debugfs_root
, NULL
, &fops_rfi_flush
);
1212 debugfs_create_file("entry_flush", 0600, powerpc_debugfs_root
, NULL
, &fops_entry_flush
);
1213 debugfs_create_file("uaccess_flush", 0600, powerpc_debugfs_root
, NULL
, &fops_uaccess_flush
);
1216 device_initcall(rfi_flush_debugfs_init
);
1218 #endif /* CONFIG_PPC_BOOK3S_64 */