xref: /freebsd/sys/arm64/arm64/machdep.c (revision 055229eda697445880edd0050d0230a3f1bc85b3)
1 /*-
2  * Copyright (c) 2014 Andrew Turner
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  *
26  */
27 
28 #include "opt_acpi.h"
29 #include "opt_kstack_pages.h"
30 #include "opt_platform.h"
31 #include "opt_ddb.h"
32 
33 #include <sys/param.h>
34 #include <sys/systm.h>
35 #include <sys/asan.h>
36 #include <sys/buf.h>
37 #include <sys/bus.h>
38 #include <sys/cons.h>
39 #include <sys/cpu.h>
40 #include <sys/csan.h>
41 #include <sys/efi.h>
42 #include <sys/efi_map.h>
43 #include <sys/exec.h>
44 #include <sys/imgact.h>
45 #include <sys/kdb.h>
46 #include <sys/kernel.h>
47 #include <sys/ktr.h>
48 #include <sys/limits.h>
49 #include <sys/linker.h>
50 #include <sys/msan.h>
51 #include <sys/msgbuf.h>
52 #include <sys/pcpu.h>
53 #include <sys/physmem.h>
54 #include <sys/proc.h>
55 #include <sys/ptrace.h>
56 #include <sys/reboot.h>
57 #include <sys/reg.h>
58 #include <sys/rwlock.h>
59 #include <sys/sched.h>
60 #include <sys/signalvar.h>
61 #include <sys/syscallsubr.h>
62 #include <sys/sysent.h>
63 #include <sys/sysproto.h>
64 #include <sys/ucontext.h>
65 #include <sys/vdso.h>
66 #include <sys/vmmeter.h>
67 
68 #include <vm/vm.h>
69 #include <vm/vm_param.h>
70 #include <vm/vm_kern.h>
71 #include <vm/vm_object.h>
72 #include <vm/vm_page.h>
73 #include <vm/vm_phys.h>
74 #include <vm/pmap.h>
75 #include <vm/vm_map.h>
76 #include <vm/vm_pager.h>
77 
78 #include <machine/armreg.h>
79 #include <machine/cpu.h>
80 #include <machine/cpu_feat.h>
81 #include <machine/debug_monitor.h>
82 #include <machine/hypervisor.h>
83 #include <machine/kdb.h>
84 #include <machine/machdep.h>
85 #include <machine/metadata.h>
86 #include <machine/md_var.h>
87 #include <machine/pcb.h>
88 #include <machine/undefined.h>
89 #include <machine/vmparam.h>
90 
91 #ifdef VFP
92 #include <machine/vfp.h>
93 #endif
94 
95 #ifdef DEV_ACPI
96 #include <contrib/dev/acpica/include/acpi.h>
97 #include <machine/acpica_machdep.h>
98 #endif
99 
100 #ifdef FDT
101 #include <dev/fdt/fdt_common.h>
102 #include <dev/ofw/openfirm.h>
103 #endif
104 
105 #include <dev/smbios/smbios.h>
106 
107 _Static_assert(sizeof(struct pcb) == 1248, "struct pcb is incorrect size");
108 _Static_assert(offsetof(struct pcb, pcb_fpusaved) == 136,
109     "pcb_fpusaved changed offset");
110 _Static_assert(offsetof(struct pcb, pcb_fpustate) == 192,
111     "pcb_fpustate changed offset");
112 
113 enum arm64_bus arm64_bus_method = ARM64_BUS_NONE;
114 
115 /*
116  * XXX: The .bss is assumed to be in the boot CPU NUMA domain. If not we
117  * could relocate this, but will need to keep the same virtual address as
118  * it's reverenced by the EARLY_COUNTER macro.
119  */
120 struct pcpu pcpu0;
121 
122 #if defined(PERTHREAD_SSP)
123 /*
124  * The boot SSP canary. Will be replaced with a per-thread canary when
125  * scheduling has started.
126  */
127 uintptr_t boot_canary = 0x49a2d892bc05a0b1ul;
128 #endif
129 
130 static struct trapframe proc0_tf;
131 
132 int early_boot = 1;
133 int cold = 1;
134 static int boot_el;
135 
136 struct kva_md_info kmi;
137 
138 int64_t dczva_line_size;	/* The size of cache line the dc zva zeroes */
139 int has_pan;
140 
141 #if defined(SOCDEV_PA)
142 /*
143  * This is the virtual address used to access SOCDEV_PA. As it's set before
144  * .bss is cleared we need to ensure it's preserved. To do this use
145  * __read_mostly as it's only ever set once but read in the putc functions.
146  */
147 uintptr_t socdev_va __read_mostly;
148 #endif
149 
150 /*
151  * Physical address of the EFI System Table. Stashed from the metadata hints
152  * passed into the kernel and used by the EFI code to call runtime services.
153  */
154 vm_paddr_t efi_systbl_phys;
155 static struct efi_map_header *efihdr;
156 
157 /* pagezero_* implementations are provided in support.S */
158 void pagezero_simple(void *);
159 void pagezero_cache(void *);
160 
161 /* pagezero_simple is default pagezero */
162 void (*pagezero)(void *p) = pagezero_simple;
163 
164 int (*apei_nmi)(void);
165 
166 #if defined(PERTHREAD_SSP_WARNING)
167 static void
print_ssp_warning(void * data __unused)168 print_ssp_warning(void *data __unused)
169 {
170 	printf("WARNING: Per-thread SSP is enabled but the compiler is too old to support it\n");
171 }
172 SYSINIT(ssp_warn, SI_SUB_COPYRIGHT, SI_ORDER_ANY, print_ssp_warning, NULL);
173 SYSINIT(ssp_warn2, SI_SUB_LAST, SI_ORDER_ANY, print_ssp_warning, NULL);
174 #endif
175 
176 static cpu_feat_en
pan_check(const struct cpu_feat * feat __unused,u_int midr __unused)177 pan_check(const struct cpu_feat *feat __unused, u_int midr __unused)
178 {
179 	uint64_t id_aa64mfr1;
180 
181 	if (!get_kernel_reg(ID_AA64MMFR1_EL1, &id_aa64mfr1))
182 		return (FEAT_ALWAYS_DISABLE);
183 	if (ID_AA64MMFR1_PAN_VAL(id_aa64mfr1) == ID_AA64MMFR1_PAN_NONE)
184 		return (FEAT_ALWAYS_DISABLE);
185 
186 	return (FEAT_DEFAULT_ENABLE);
187 }
188 
189 static bool
pan_enable(const struct cpu_feat * feat __unused,cpu_feat_errata errata_status __unused,u_int * errata_list __unused,u_int errata_count __unused)190 pan_enable(const struct cpu_feat *feat __unused,
191     cpu_feat_errata errata_status __unused, u_int *errata_list __unused,
192     u_int errata_count __unused)
193 {
194 	has_pan = 1;
195 
196 	/*
197 	 * This sets the PAN bit, stopping the kernel from accessing
198 	 * memory when userspace can also access it unless the kernel
199 	 * uses the userspace load/store instructions.
200 	 */
201 	WRITE_SPECIALREG(sctlr_el1,
202 	    READ_SPECIALREG(sctlr_el1) & ~SCTLR_SPAN);
203 	__asm __volatile(
204 	    ".arch_extension pan	\n"
205 	    "msr pan, #1		\n"
206 	    ".arch_extension nopan	\n");
207 
208 	return (true);
209 }
210 
211 static void
pan_disabled(const struct cpu_feat * feat __unused)212 pan_disabled(const struct cpu_feat *feat __unused)
213 {
214 	if (PCPU_GET(cpuid) == 0)
215 		update_special_reg(ID_AA64MMFR1_EL1, ID_AA64MMFR1_PAN_MASK, 0);
216 }
217 
218 CPU_FEAT(feat_pan, "Privileged access never",
219     pan_check, NULL, pan_enable, pan_disabled,
220     CPU_FEAT_AFTER_DEV | CPU_FEAT_PER_CPU);
221 
222 bool
has_hyp(void)223 has_hyp(void)
224 {
225 	return (boot_el == CURRENTEL_EL_EL2);
226 }
227 
228 bool
in_vhe(void)229 in_vhe(void)
230 {
231 	/* If we are currently in EL2 then must be in VHE */
232 	return ((READ_SPECIALREG(CurrentEL) & CURRENTEL_EL_MASK) ==
233 	    CURRENTEL_EL_EL2);
234 }
235 
236 static void
cpu_startup(void * dummy)237 cpu_startup(void *dummy)
238 {
239 	vm_paddr_t size;
240 	int i;
241 
242 	printf("real memory  = %ju (%ju MB)\n", ptoa((uintmax_t)realmem),
243 	    ptoa((uintmax_t)realmem) / 1024 / 1024);
244 
245 	if (bootverbose) {
246 		printf("Physical memory chunk(s):\n");
247 		for (i = 0; phys_avail[i + 1] != 0; i += 2) {
248 			size = phys_avail[i + 1] - phys_avail[i];
249 			printf("%#016jx - %#016jx, %ju bytes (%ju pages)\n",
250 			    (uintmax_t)phys_avail[i],
251 			    (uintmax_t)phys_avail[i + 1] - 1,
252 			    (uintmax_t)size, (uintmax_t)size / PAGE_SIZE);
253 		}
254 	}
255 
256 	printf("avail memory = %ju (%ju MB)\n",
257 	    ptoa((uintmax_t)vm_free_count()),
258 	    ptoa((uintmax_t)vm_free_count()) / 1024 / 1024);
259 
260 	undef_init();
261 	install_cpu_errata();
262 
263 	vm_ksubmap_init(&kmi);
264 	bufinit();
265 	vm_pager_bufferinit();
266 }
267 
268 SYSINIT(cpu, SI_SUB_CPU, SI_ORDER_FIRST, cpu_startup, NULL);
269 
270 static void
late_ifunc_resolve(void * dummy __unused)271 late_ifunc_resolve(void *dummy __unused)
272 {
273 	link_elf_late_ireloc();
274 }
275 /* Late enough for cpu_feat to have completed */
276 SYSINIT(late_ifunc_resolve, SI_SUB_CONFIGURE, SI_ORDER_ANY,
277     late_ifunc_resolve, NULL);
278 
279 int
cpu_idle_wakeup(int cpu)280 cpu_idle_wakeup(int cpu)
281 {
282 
283 	return (0);
284 }
285 
286 void
cpu_idle(int busy)287 cpu_idle(int busy)
288 {
289 
290 	spinlock_enter();
291 	if (!busy)
292 		cpu_idleclock();
293 	if (!sched_runnable())
294 		__asm __volatile(
295 		    "dsb sy \n"
296 		    "wfi    \n");
297 	if (!busy)
298 		cpu_activeclock();
299 	spinlock_exit();
300 }
301 
302 void
cpu_halt(void)303 cpu_halt(void)
304 {
305 
306 	/* We should have shutdown by now, if not enter a low power sleep */
307 	intr_disable();
308 	while (1) {
309 		__asm __volatile("wfi");
310 	}
311 }
312 
313 /*
314  * Flush the D-cache for non-DMA I/O so that the I-cache can
315  * be made coherent later.
316  */
317 void
cpu_flush_dcache(void * ptr,size_t len)318 cpu_flush_dcache(void *ptr, size_t len)
319 {
320 
321 	/* ARM64TODO TBD */
322 }
323 
324 /* Get current clock frequency for the given CPU ID. */
325 int
cpu_est_clockrate(int cpu_id,uint64_t * rate)326 cpu_est_clockrate(int cpu_id, uint64_t *rate)
327 {
328 	struct pcpu *pc;
329 
330 	pc = pcpu_find(cpu_id);
331 	if (pc == NULL || rate == NULL)
332 		return (EINVAL);
333 
334 	if (pc->pc_clock == 0)
335 		return (EOPNOTSUPP);
336 
337 	*rate = pc->pc_clock;
338 	return (0);
339 }
340 
341 void
cpu_pcpu_init(struct pcpu * pcpu,int cpuid,size_t size)342 cpu_pcpu_init(struct pcpu *pcpu, int cpuid, size_t size)
343 {
344 
345 	pcpu->pc_acpi_id = 0xffffffff;
346 	pcpu->pc_mpidr = UINT64_MAX;
347 }
348 
349 void
spinlock_enter(void)350 spinlock_enter(void)
351 {
352 	struct thread *td;
353 	register_t daif;
354 
355 	td = curthread;
356 	if (td->td_md.md_spinlock_count == 0) {
357 		daif = intr_disable();
358 		td->td_md.md_spinlock_count = 1;
359 		td->td_md.md_saved_daif = daif;
360 		critical_enter();
361 	} else
362 		td->td_md.md_spinlock_count++;
363 }
364 
365 void
spinlock_exit(void)366 spinlock_exit(void)
367 {
368 	struct thread *td;
369 	register_t daif;
370 
371 	td = curthread;
372 	daif = td->td_md.md_saved_daif;
373 	td->td_md.md_spinlock_count--;
374 	if (td->td_md.md_spinlock_count == 0) {
375 		critical_exit();
376 		intr_restore(daif);
377 	}
378 }
379 
380 /*
381  * Construct a PCB from a trapframe. This is called from kdb_trap() where
382  * we want to start a backtrace from the function that caused us to enter
383  * the debugger. We have the context in the trapframe, but base the trace
384  * on the PCB. The PCB doesn't have to be perfect, as long as it contains
385  * enough for a backtrace.
386  */
387 void
makectx(struct trapframe * tf,struct pcb * pcb)388 makectx(struct trapframe *tf, struct pcb *pcb)
389 {
390 	int i;
391 
392 	/* NB: pcb_x[PCB_LR] is the PC, see PC_REGS() in db_machdep.h */
393 	for (i = 0; i < nitems(pcb->pcb_x); i++) {
394 		if (i == PCB_LR)
395 			pcb->pcb_x[i] = tf->tf_elr;
396 		else
397 			pcb->pcb_x[i] = tf->tf_x[i + PCB_X_START];
398 	}
399 
400 	pcb->pcb_sp = tf->tf_sp;
401 }
402 
403 static void
init_proc0(vm_offset_t kstack)404 init_proc0(vm_offset_t kstack)
405 {
406 	struct pcpu *pcpup;
407 
408 	pcpup = cpuid_to_pcpu[0];
409 	MPASS(pcpup != NULL);
410 
411 	proc_linkup0(&proc0, &thread0);
412 	thread0.td_kstack = kstack;
413 	thread0.td_kstack_pages = KSTACK_PAGES;
414 #if defined(PERTHREAD_SSP)
415 	thread0.td_md.md_canary = boot_canary;
416 #endif
417 	thread0.td_pcb = (struct pcb *)(thread0.td_kstack +
418 	    thread0.td_kstack_pages * PAGE_SIZE) - 1;
419 	thread0.td_pcb->pcb_flags = 0;
420 	thread0.td_pcb->pcb_fpflags = 0;
421 	thread0.td_pcb->pcb_fpusaved = &thread0.td_pcb->pcb_fpustate;
422 	thread0.td_pcb->pcb_vfpcpu = UINT_MAX;
423 	thread0.td_frame = &proc0_tf;
424 	ptrauth_thread0(&thread0);
425 	pcpup->pc_curpcb = thread0.td_pcb;
426 
427 	/*
428 	 * Unmask SError exceptions. They are used to signal a RAS failure,
429 	 * or other hardware error.
430 	 */
431 	serror_enable();
432 }
433 
434 /*
435  * Get an address to be used to write to kernel data that may be mapped
436  * read-only, e.g. to patch kernel code.
437  */
438 bool
arm64_get_writable_addr(void * addr,void ** out)439 arm64_get_writable_addr(void *addr, void **out)
440 {
441 	vm_paddr_t pa;
442 
443 	/* Check if the page is writable */
444 	if (PAR_SUCCESS(arm64_address_translate_s1e1w((vm_offset_t)addr))) {
445 		*out = addr;
446 		return (true);
447 	}
448 
449 	/*
450 	 * Find the physical address of the given page.
451 	 */
452 	if (!pmap_klookup((vm_offset_t)addr, &pa)) {
453 		return (false);
454 	}
455 
456 	/*
457 	 * If it is within the DMAP region and is writable use that.
458 	 */
459 	if (PHYS_IN_DMAP_RANGE(pa)) {
460 		addr = (void *)PHYS_TO_DMAP(pa);
461 		if (PAR_SUCCESS(arm64_address_translate_s1e1w(
462 		    (vm_offset_t)addr))) {
463 			*out = addr;
464 			return (true);
465 		}
466 	}
467 
468 	return (false);
469 }
470 
471 /*
472  * Map the passed in VA in EFI space to a void * using the efi memory table to
473  * find the PA and return it in the DMAP, if it exists. We're used between the
474  * calls to pmap_bootstrap() and physmem_init_kernel_globals() to parse CFG
475  * tables We assume that either the entry you are mapping fits within its page,
476  * or if it spills to the next page, that's contiguous in PA and in the DMAP.
477  * All observed tables obey the first part of this precondition.
478  */
479 struct early_map_data
480 {
481 	vm_offset_t va;
482 	vm_offset_t pa;
483 };
484 
485 static void
efi_early_map_entry(struct efi_md * p,void * argp)486 efi_early_map_entry(struct efi_md *p, void *argp)
487 {
488 	struct early_map_data *emdp = argp;
489 	vm_offset_t s, e;
490 
491 	if (emdp->pa != 0)
492 		return;
493 	if ((p->md_attr & EFI_MD_ATTR_RT) == 0)
494 		return;
495 	s = p->md_virt;
496 	e = p->md_virt + p->md_pages * EFI_PAGE_SIZE;
497 	if (emdp->va < s  || emdp->va >= e)
498 		return;
499 	emdp->pa = p->md_phys + (emdp->va - p->md_virt);
500 }
501 
502 static void *
efi_early_map(vm_offset_t va)503 efi_early_map(vm_offset_t va)
504 {
505 	struct early_map_data emd = { .va = va };
506 
507 	efi_map_foreach_entry(efihdr, efi_early_map_entry, &emd);
508 	if (emd.pa == 0)
509 		return NULL;
510 	return (void *)PHYS_TO_DMAP(emd.pa);
511 }
512 
513 
514 /*
515  * When booted via kexec from Linux, the prior kernel will pass in reserved
516  * memory areas in an EFI config table. We need to find that table and walk
517  * through it excluding the memory ranges in it. btw, this is called too early
518  * for the printf to do anything (unless EARLY_PRINTF is defined) since msgbufp
519  * isn't initialized, let alone a console, but breakpoints in printf help
520  * diagnose rare failures.
521  */
522 static void
exclude_efi_memreserve(vm_paddr_t efi_systbl_phys)523 exclude_efi_memreserve(vm_paddr_t efi_systbl_phys)
524 {
525 	struct efi_systbl *systbl;
526 	efi_guid_t efi_memreserve = LINUX_EFI_MEMRESERVE_TABLE;
527 
528 	systbl = (struct efi_systbl *)PHYS_TO_DMAP(efi_systbl_phys);
529 	if (systbl == NULL) {
530 		printf("can't map systbl\n");
531 		return;
532 	}
533 	if (systbl->st_hdr.th_sig != EFI_SYSTBL_SIG) {
534 		printf("Bad signature for systbl %#lx\n", systbl->st_hdr.th_sig);
535 		return;
536 	}
537 
538 	/*
539 	 * We don't yet have the pmap system booted enough to create a pmap for
540 	 * the efi firmware's preferred address space from the GetMemoryMap()
541 	 * table. The st_cfgtbl is a VA in this space, so we need to do the
542 	 * mapping ourselves to a kernel VA with efi_early_map. We assume that
543 	 * the cfgtbl entries don't span a page. Other pointers are PAs, as
544 	 * noted below.
545 	 */
546 	if (systbl->st_cfgtbl == 0)	/* Failsafe st_entries should == 0 in this case */
547 		return;
548 	for (int i = 0; i < systbl->st_entries; i++) {
549 		struct efi_cfgtbl *cfgtbl;
550 		struct linux_efi_memreserve *mr;
551 
552 		cfgtbl = efi_early_map(systbl->st_cfgtbl + i * sizeof(*cfgtbl));
553 		if (cfgtbl == NULL)
554 			panic("Can't map the config table entry %d\n", i);
555 		if (memcmp(&cfgtbl->ct_guid, &efi_memreserve, sizeof(efi_guid_t)) != 0)
556 			continue;
557 
558 		/*
559 		 * cfgtbl points are either VA or PA, depending on the GUID of
560 		 * the table. memreserve GUID pointers are PA and not converted
561 		 * after a SetVirtualAddressMap(). The list's mr_next pointer
562 		 * is also a PA.
563 		 */
564 		mr = (struct linux_efi_memreserve *)PHYS_TO_DMAP(
565 			(vm_offset_t)cfgtbl->ct_data);
566 		while (true) {
567 			for (int j = 0; j < mr->mr_count; j++) {
568 				struct linux_efi_memreserve_entry *mre;
569 
570 				mre = &mr->mr_entry[j];
571 				physmem_exclude_region(mre->mre_base, mre->mre_size,
572 				    EXFLAG_NODUMP | EXFLAG_NOALLOC);
573 			}
574 			if (mr->mr_next == 0)
575 				break;
576 			mr = (struct linux_efi_memreserve *)PHYS_TO_DMAP(mr->mr_next);
577 		};
578 	}
579 
580 }
581 
582 #ifdef FDT
583 static void
try_load_dtb(void)584 try_load_dtb(void)
585 {
586 	vm_offset_t dtbp;
587 
588 	dtbp = MD_FETCH(preload_kmdp, MODINFOMD_DTBP, vm_offset_t);
589 #if defined(FDT_DTB_STATIC)
590 	/*
591 	 * In case the device tree blob was not retrieved (from metadata) try
592 	 * to use the statically embedded one.
593 	 */
594 	if (dtbp == 0)
595 		dtbp = (vm_offset_t)&fdt_static_dtb;
596 #endif
597 
598 	if (dtbp == (vm_offset_t)NULL) {
599 #ifndef TSLOG
600 		printf("ERROR loading DTB\n");
601 #endif
602 		return;
603 	}
604 
605 	if (!OF_install(OFW_FDT, 0))
606 		panic("Cannot install FDT");
607 
608 	if (OF_init((void *)dtbp) != 0)
609 		panic("OF_init failed with the found device tree");
610 
611 	parse_fdt_bootargs();
612 }
613 #endif
614 
615 static bool
bus_probe(void)616 bus_probe(void)
617 {
618 	bool has_acpi, has_fdt;
619 	char *order, *env;
620 
621 	has_acpi = has_fdt = false;
622 
623 #ifdef FDT
624 	has_fdt = (OF_peer(0) != 0);
625 #endif
626 #ifdef DEV_ACPI
627 	has_acpi = (AcpiOsGetRootPointer() != 0);
628 #endif
629 
630 	env = kern_getenv("kern.cfg.order");
631 	if (env != NULL) {
632 		order = env;
633 		while (order != NULL) {
634 			if (has_acpi &&
635 			    strncmp(order, "acpi", 4) == 0 &&
636 			    (order[4] == ',' || order[4] == '\0')) {
637 				arm64_bus_method = ARM64_BUS_ACPI;
638 				break;
639 			}
640 			if (has_fdt &&
641 			    strncmp(order, "fdt", 3) == 0 &&
642 			    (order[3] == ',' || order[3] == '\0')) {
643 				arm64_bus_method = ARM64_BUS_FDT;
644 				break;
645 			}
646 			order = strchr(order, ',');
647 			if (order != NULL)
648 				order++;	/* Skip comma */
649 		}
650 		freeenv(env);
651 
652 		/* If we set the bus method it is valid */
653 		if (arm64_bus_method != ARM64_BUS_NONE)
654 			return (true);
655 	}
656 	/* If no order or an invalid order was set use the default */
657 	if (arm64_bus_method == ARM64_BUS_NONE) {
658 		if (has_acpi)
659 			arm64_bus_method = ARM64_BUS_ACPI;
660 		else if (has_fdt)
661 			arm64_bus_method = ARM64_BUS_FDT;
662 	}
663 
664 	/*
665 	 * If no option was set the default is valid, otherwise we are
666 	 * setting one to get cninit() working, then calling panic to tell
667 	 * the user about the invalid bus setup.
668 	 */
669 	return (env == NULL);
670 }
671 
672 static void
cache_setup(void)673 cache_setup(void)
674 {
675 	int dczva_line_shift;
676 	uint32_t dczid_el0;
677 
678 	identify_cache(READ_SPECIALREG(ctr_el0));
679 
680 	dczid_el0 = READ_SPECIALREG(dczid_el0);
681 
682 	/* Check if dc zva is not prohibited */
683 	if (dczid_el0 & DCZID_DZP)
684 		dczva_line_size = 0;
685 	else {
686 		/* Same as with above calculations */
687 		dczva_line_shift = DCZID_BS_SIZE(dczid_el0);
688 		dczva_line_size = sizeof(int) << dczva_line_shift;
689 
690 		/* Change pagezero function */
691 		pagezero = pagezero_cache;
692 	}
693 }
694 
695 int
memory_mapping_mode(vm_paddr_t pa)696 memory_mapping_mode(vm_paddr_t pa)
697 {
698 	struct efi_md *map, *p;
699 	size_t efisz;
700 	int ndesc, i;
701 
702 	if (efihdr == NULL)
703 		return (VM_MEMATTR_WRITE_BACK);
704 
705 	/*
706 	 * Memory map data provided by UEFI via the GetMemoryMap
707 	 * Boot Services API.
708 	 */
709 	efisz = (sizeof(struct efi_map_header) + 0xf) & ~0xf;
710 	map = (struct efi_md *)((uint8_t *)efihdr + efisz);
711 
712 	if (efihdr->descriptor_size == 0)
713 		return (VM_MEMATTR_WRITE_BACK);
714 	ndesc = efihdr->memory_size / efihdr->descriptor_size;
715 
716 	for (i = 0, p = map; i < ndesc; i++,
717 	    p = efi_next_descriptor(p, efihdr->descriptor_size)) {
718 		if (pa < p->md_phys ||
719 		    pa >= p->md_phys + p->md_pages * EFI_PAGE_SIZE)
720 			continue;
721 		if (p->md_type == EFI_MD_TYPE_IOMEM ||
722 		    p->md_type == EFI_MD_TYPE_IOPORT)
723 			return (VM_MEMATTR_DEVICE);
724 		else if ((p->md_attr & EFI_MD_ATTR_WB) != 0 ||
725 		    p->md_type == EFI_MD_TYPE_RECLAIM)
726 			return (VM_MEMATTR_WRITE_BACK);
727 		else if ((p->md_attr & EFI_MD_ATTR_WT) != 0)
728 			return (VM_MEMATTR_WRITE_THROUGH);
729 		else if ((p->md_attr & EFI_MD_ATTR_WC) != 0)
730 			return (VM_MEMATTR_WRITE_COMBINING);
731 		break;
732 	}
733 
734 	return (VM_MEMATTR_DEVICE);
735 }
736 
737 #ifdef FDT
738 static void
fdt_physmem_hardware_region_cb(const struct mem_region * mr,void * arg __unused)739 fdt_physmem_hardware_region_cb(const struct mem_region *mr, void *arg __unused)
740 {
741 	physmem_hardware_region(mr->mr_start, mr->mr_size);
742 }
743 
744 static void
fdt_physmem_exclude_region_cb(const struct mem_region * mr,void * arg __unused)745 fdt_physmem_exclude_region_cb(const struct mem_region *mr, void *arg __unused)
746 {
747 	physmem_exclude_region(mr->mr_start, mr->mr_size,
748 	    EXFLAG_NODUMP | EXFLAG_NOALLOC);
749 }
750 #endif
751 
752 void
initarm(struct arm64_bootparams * abp)753 initarm(struct arm64_bootparams *abp)
754 {
755 	struct efi_fb *efifb;
756 	struct pcpu *pcpup;
757 	char *env;
758 #ifdef FDT
759 	phandle_t root;
760 	char dts_version[255];
761 #endif
762 	vm_offset_t lastaddr;
763 	bool valid;
764 
765 	TSRAW(&thread0, TS_ENTER, __func__, NULL);
766 
767 	boot_el = abp->boot_el;
768 
769 	/* Parse loader or FDT boot parameters. Determine last used address. */
770 	lastaddr = parse_boot_param(abp);
771 
772 	identify_cpu(0);
773 	identify_hypervisor_smbios();
774 
775 	update_special_regs(0);
776 
777 	/* Set the pcpu data, this is needed by pmap_bootstrap */
778 	pcpup = &pcpu0;
779 	pcpu_init(pcpup, 0, sizeof(struct pcpu));
780 
781 	/*
782 	 * Set the pcpu pointer with a backup in tpidr_el1 to be
783 	 * loaded when entering the kernel from userland.
784 	 */
785 	__asm __volatile(
786 	    "mov x18, %0 \n"
787 	    "msr tpidr_el1, %0" :: "r"(pcpup));
788 
789 	/* locore.S sets sp_el0 to &thread0 so no need to set it here. */
790 	PCPU_SET(curthread, &thread0);
791 	PCPU_SET(midr, get_midr());
792 
793 	link_elf_ireloc();
794 #ifdef FDT
795 	try_load_dtb();
796 #endif
797 
798 	efi_systbl_phys = MD_FETCH(preload_kmdp, MODINFOMD_FW_HANDLE,
799 	    vm_paddr_t);
800 
801 	/* Load the physical memory ranges */
802 	efihdr = (struct efi_map_header *)preload_search_info(preload_kmdp,
803 	    MODINFO_METADATA | MODINFOMD_EFI_MAP);
804 	if (efihdr != NULL)
805 		efi_map_add_entries(efihdr);
806 #ifdef FDT
807 	else {
808 		/* Grab physical memory regions information from device tree. */
809 		if (fdt_foreach_mem_region(fdt_physmem_hardware_region_cb,
810 		    NULL) != 0)
811 			panic("Cannot get physical memory regions");
812 	}
813 	fdt_foreach_reserved_mem(fdt_physmem_exclude_region_cb, NULL);
814 #endif
815 
816 	/* Exclude the EFI framebuffer from our view of physical memory. */
817 	efifb = (struct efi_fb *)preload_search_info(preload_kmdp,
818 	    MODINFO_METADATA | MODINFOMD_EFI_FB);
819 	if (efifb != NULL)
820 		physmem_exclude_region(efifb->fb_addr, efifb->fb_size,
821 		    EXFLAG_NOALLOC);
822 
823 	/* Do basic tuning, hz etc */
824 	init_param1();
825 
826 	cache_setup();
827 
828 	/*
829 	 * Perform a staged bootstrap of virtual memory.
830 	 *
831 	 * - First we create the DMAP region. This allows it to be used in
832 	 *   later bootstrapping.
833 	 * - Next exclude memory that is needed in the DMAP region, but must
834 	 *   not be used by FreeBSD.
835 	 * - Lastly complete the bootstrapping. It may use the physical
836 	 *   memory map so any excluded memory must be marked as such before
837 	 *   pmap_bootstrap() is called.
838 	 */
839 	pmap_bootstrap_dmap(lastaddr - KERNBASE);
840 	/*
841 	 * Exclude EFI entries needed in the DMAP, e.g. EFI_MD_TYPE_RECLAIM
842 	 * may contain the ACPI tables but shouldn't be used by the kernel
843 	 */
844 	if (efihdr != NULL)
845 		efi_map_exclude_entries(efihdr);
846 	/*  Do the same for reserve entries in the EFI MEMRESERVE table */
847 	if (efi_systbl_phys != 0)
848 		exclude_efi_memreserve(efi_systbl_phys);
849 	/* Continue bootstrapping pmap */
850 	pmap_bootstrap();
851 
852 	/*
853 	 * We carefully bootstrap the sanitizer map after we've excluded
854 	 * absolutely everything else that could impact phys_avail.  There's not
855 	 * always enough room for the initial shadow map after the kernel, so
856 	 * we'll end up searching for segments that we can safely use.  Those
857 	 * segments also get excluded from phys_avail.
858 	 */
859 #if defined(KASAN) || defined(KMSAN)
860 	pmap_bootstrap_san();
861 #endif
862 
863 	physmem_init_kernel_globals();
864 
865 	valid = bus_probe();
866 
867 	cninit();
868 	set_ttbr0(abp->kern_ttbr0);
869 	pmap_s1_invalidate_all_kernel();
870 
871 	if (!valid)
872 		panic("Invalid bus configuration: %s",
873 		    kern_getenv("kern.cfg.order"));
874 
875 	/* Detect early CPU feature support */
876 	enable_cpu_feat(CPU_FEAT_EARLY_BOOT);
877 
878 	/*
879 	 * Dump the boot metadata. We have to wait for cninit() since console
880 	 * output is required. If it's grossly incorrect the kernel will never
881 	 * make it this far.
882 	 */
883 	if (getenv_is_true("debug.dump_modinfo_at_boot"))
884 		preload_dump();
885 
886 	init_proc0(abp->kern_stack);
887 	msgbufinit(msgbufp, msgbufsize);
888 	mutex_init();
889 	init_param2(physmem);
890 
891 	dbg_init();
892 	kdb_init();
893 #ifdef KDB
894 	if ((boothowto & RB_KDB) != 0)
895 		kdb_enter(KDB_WHY_BOOTFLAGS, "Boot flags requested debugger");
896 #endif
897 
898 	kcsan_cpu_init(0);
899 	kasan_init();
900 	kmsan_init();
901 
902 	env = kern_getenv("kernelname");
903 	if (env != NULL)
904 		strlcpy(kernelname, env, sizeof(kernelname));
905 
906 #ifdef FDT
907 	if (arm64_bus_method == ARM64_BUS_FDT) {
908 		root = OF_finddevice("/");
909 		if (OF_getprop(root, "freebsd,dts-version", dts_version, sizeof(dts_version)) > 0) {
910 			if (strcmp(LINUX_DTS_VERSION, dts_version) != 0)
911 				printf("WARNING: DTB version is %s while kernel expects %s, "
912 				    "please update the DTB in the ESP\n",
913 				    dts_version,
914 				    LINUX_DTS_VERSION);
915 		} else {
916 			printf("WARNING: Cannot find freebsd,dts-version property, "
917 			    "cannot check DTB compliance\n");
918 		}
919 	}
920 #endif
921 
922 	if (boothowto & RB_VERBOSE) {
923 		if (efihdr != NULL)
924 			efi_map_print_entries(efihdr);
925 		physmem_print_tables();
926 	}
927 
928 	early_boot = 0;
929 
930 	if (bootverbose && kstack_pages != KSTACK_PAGES)
931 		printf("kern.kstack_pages = %d ignored for thread0\n",
932 		    kstack_pages);
933 
934 	TSEXIT();
935 }
936 
937 void
dbg_init(void)938 dbg_init(void)
939 {
940 
941 	/* Clear OS lock */
942 	WRITE_SPECIALREG(oslar_el1, 0);
943 
944 	/* This permits DDB to use debug registers for watchpoints. */
945 	dbg_monitor_init();
946 
947 	/* TODO: Eventually will need to initialize debug registers here. */
948 }
949 
950 #ifdef DDB
951 #include <ddb/ddb.h>
952 
DB_SHOW_COMMAND(specialregs,db_show_spregs)953 DB_SHOW_COMMAND(specialregs, db_show_spregs)
954 {
955 #define	PRINT_REG(reg)	\
956     db_printf(__STRING(reg) " = %#016lx\n", READ_SPECIALREG(reg))
957 
958 	PRINT_REG(actlr_el1);
959 	PRINT_REG(afsr0_el1);
960 	PRINT_REG(afsr1_el1);
961 	PRINT_REG(aidr_el1);
962 	PRINT_REG(amair_el1);
963 	PRINT_REG(ccsidr_el1);
964 	PRINT_REG(clidr_el1);
965 	PRINT_REG(contextidr_el1);
966 	PRINT_REG(cpacr_el1);
967 	PRINT_REG(csselr_el1);
968 	PRINT_REG(ctr_el0);
969 	PRINT_REG(currentel);
970 	PRINT_REG(daif);
971 	PRINT_REG(dczid_el0);
972 	PRINT_REG(elr_el1);
973 	PRINT_REG(esr_el1);
974 	PRINT_REG(far_el1);
975 #if 0
976 	/* ARM64TODO: Enable VFP before reading floating-point registers */
977 	PRINT_REG(fpcr);
978 	PRINT_REG(fpsr);
979 #endif
980 	PRINT_REG(id_aa64afr0_el1);
981 	PRINT_REG(id_aa64afr1_el1);
982 	PRINT_REG(id_aa64dfr0_el1);
983 	PRINT_REG(id_aa64dfr1_el1);
984 	PRINT_REG(id_aa64isar0_el1);
985 	PRINT_REG(id_aa64isar1_el1);
986 	PRINT_REG(id_aa64pfr0_el1);
987 	PRINT_REG(id_aa64pfr1_el1);
988 	PRINT_REG(id_afr0_el1);
989 	PRINT_REG(id_dfr0_el1);
990 	PRINT_REG(id_isar0_el1);
991 	PRINT_REG(id_isar1_el1);
992 	PRINT_REG(id_isar2_el1);
993 	PRINT_REG(id_isar3_el1);
994 	PRINT_REG(id_isar4_el1);
995 	PRINT_REG(id_isar5_el1);
996 	PRINT_REG(id_mmfr0_el1);
997 	PRINT_REG(id_mmfr1_el1);
998 	PRINT_REG(id_mmfr2_el1);
999 	PRINT_REG(id_mmfr3_el1);
1000 #if 0
1001 	/* Missing from llvm */
1002 	PRINT_REG(id_mmfr4_el1);
1003 #endif
1004 	PRINT_REG(id_pfr0_el1);
1005 	PRINT_REG(id_pfr1_el1);
1006 	PRINT_REG(isr_el1);
1007 	PRINT_REG(mair_el1);
1008 	PRINT_REG(midr_el1);
1009 	PRINT_REG(mpidr_el1);
1010 	PRINT_REG(mvfr0_el1);
1011 	PRINT_REG(mvfr1_el1);
1012 	PRINT_REG(mvfr2_el1);
1013 	PRINT_REG(revidr_el1);
1014 	PRINT_REG(sctlr_el1);
1015 	PRINT_REG(sp_el0);
1016 	PRINT_REG(spsel);
1017 	PRINT_REG(spsr_el1);
1018 	PRINT_REG(tcr_el1);
1019 	PRINT_REG(tpidr_el0);
1020 	PRINT_REG(tpidr_el1);
1021 	PRINT_REG(tpidrro_el0);
1022 	PRINT_REG(ttbr0_el1);
1023 	PRINT_REG(ttbr1_el1);
1024 	PRINT_REG(vbar_el1);
1025 #undef PRINT_REG
1026 }
1027 
DB_SHOW_COMMAND(vtop,db_show_vtop)1028 DB_SHOW_COMMAND(vtop, db_show_vtop)
1029 {
1030 	uint64_t phys;
1031 
1032 	if (have_addr) {
1033 		phys = arm64_address_translate_s1e1r(addr);
1034 		db_printf("EL1 physical address reg (read):  0x%016lx\n", phys);
1035 		phys = arm64_address_translate_s1e1w(addr);
1036 		db_printf("EL1 physical address reg (write): 0x%016lx\n", phys);
1037 		phys = arm64_address_translate_s1e0r(addr);
1038 		db_printf("EL0 physical address reg (read):  0x%016lx\n", phys);
1039 		phys = arm64_address_translate_s1e0w(addr);
1040 		db_printf("EL0 physical address reg (write): 0x%016lx\n", phys);
1041 	} else
1042 		db_printf("show vtop <virt_addr>\n");
1043 }
1044 #endif
1045